1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 *
21 * $FreeBSD: stable/10/sys/cddl/contrib/opensolaris/uts/common/dtrace/dtrace.c 266102 2014-05-15 00:52:17Z markj $
22 */
23
24/*
25 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
26 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
27 * Copyright (c) 2012 by Delphix. All rights reserved.
28 */
29
30#pragma ident "%Z%%M% %I% %E% SMI"
31
32/*
33 * DTrace - Dynamic Tracing for Solaris
34 *
35 * This is the implementation of the Solaris Dynamic Tracing framework
36 * (DTrace). The user-visible interface to DTrace is described at length in
37 * the "Solaris Dynamic Tracing Guide". The interfaces between the libdtrace
38 * library, the in-kernel DTrace framework, and the DTrace providers are
39 * described in the block comments in the <sys/dtrace.h> header file. The
40 * internal architecture of DTrace is described in the block comments in the
41 * <sys/dtrace_impl.h> header file. The comments contained within the DTrace
42 * implementation very much assume mastery of all of these sources; if one has
43 * an unanswered question about the implementation, one should consult them
44 * first.
45 *
46 * The functions here are ordered roughly as follows:
47 *
48 * - Probe context functions
49 * - Probe hashing functions
50 * - Non-probe context utility functions
51 * - Matching functions
52 * - Provider-to-Framework API functions
53 * - Probe management functions
54 * - DIF object functions
55 * - Format functions
56 * - Predicate functions
57 * - ECB functions
58 * - Buffer functions
59 * - Enabling functions
60 * - DOF functions
61 * - Anonymous enabling functions
62 * - Consumer state functions
63 * - Helper functions
64 * - Hook functions
65 * - Driver cookbook functions
66 *
67 * Each group of functions begins with a block comment labelled the "DTrace
68 * [Group] Functions", allowing one to find each block by searching forward
69 * on capital-f functions.
70 */
71#include <sys/errno.h>
72#if !defined(sun)
73#include <sys/time.h>
74#endif
75#include <sys/stat.h>
76#include <sys/modctl.h>
77#include <sys/conf.h>
78#include <sys/systm.h>
79#if defined(sun)
80#include <sys/ddi.h>
81#include <sys/sunddi.h>
82#endif
83#include <sys/cpuvar.h>
84#include <sys/kmem.h>
85#if defined(sun)
86#include <sys/strsubr.h>
87#endif
88#include <sys/sysmacros.h>
89#include <sys/dtrace_impl.h>
90#include <sys/atomic.h>
91#include <sys/cmn_err.h>
92#if defined(sun)
93#include <sys/mutex_impl.h>
94#include <sys/rwlock_impl.h>
95#endif
96#include <sys/ctf_api.h>
97#if defined(sun)
98#include <sys/panic.h>
99#include <sys/priv_impl.h>
100#endif
101#include <sys/policy.h>
102#if defined(sun)
103#include <sys/cred_impl.h>
104#include <sys/procfs_isa.h>
105#endif
106#include <sys/taskq.h>
107#if defined(sun)
108#include <sys/mkdev.h>
109#include <sys/kdi.h>
110#endif
111#include <sys/zone.h>
112#include <sys/socket.h>
113#include <netinet/in.h>
114
115/* FreeBSD includes: */
116#if !defined(sun)
117#include <sys/callout.h>
118#include <sys/ctype.h>
119#include <sys/eventhandler.h>
120#include <sys/limits.h>
121#include <sys/kdb.h>
122#include <sys/kernel.h>
123#include <sys/malloc.h>
124#include <sys/sysctl.h>
125#include <sys/lock.h>
126#include <sys/mutex.h>
127#include <sys/rwlock.h>
128#include <sys/sx.h>
129#include <sys/dtrace_bsd.h>
130#include <netinet/in.h>
131#include "dtrace_cddl.h"
132#include "dtrace_debug.c"
133#endif
134
135/*
136 * DTrace Tunable Variables
137 *
138 * The following variables may be tuned by adding a line to /etc/system that
139 * includes both the name of the DTrace module ("dtrace") and the name of the
140 * variable. For example:
141 *
142 * set dtrace:dtrace_destructive_disallow = 1
143 *
144 * In general, the only variables that one should be tuning this way are those
145 * that affect system-wide DTrace behavior, and for which the default behavior
146 * is undesirable. Most of these variables are tunable on a per-consumer
147 * basis using DTrace options, and need not be tuned on a system-wide basis.
148 * When tuning these variables, avoid pathological values; while some attempt
149 * is made to verify the integrity of these variables, they are not considered
150 * part of the supported interface to DTrace, and they are therefore not
151 * checked comprehensively. Further, these variables should not be tuned
152 * dynamically via "mdb -kw" or other means; they should only be tuned via
153 * /etc/system.
154 */
155int dtrace_destructive_disallow = 0;
156dtrace_optval_t dtrace_nonroot_maxsize = (16 * 1024 * 1024);
157size_t dtrace_difo_maxsize = (256 * 1024);
158dtrace_optval_t dtrace_dof_maxsize = (8 * 1024 * 1024);
159size_t dtrace_global_maxsize = (16 * 1024);
160size_t dtrace_actions_max = (16 * 1024);
161size_t dtrace_retain_max = 1024;
162dtrace_optval_t dtrace_helper_actions_max = 128;
163dtrace_optval_t dtrace_helper_providers_max = 32;
164dtrace_optval_t dtrace_dstate_defsize = (1 * 1024 * 1024);
165size_t dtrace_strsize_default = 256;
166dtrace_optval_t dtrace_cleanrate_default = 9900990; /* 101 hz */
167dtrace_optval_t dtrace_cleanrate_min = 200000; /* 5000 hz */
168dtrace_optval_t dtrace_cleanrate_max = (uint64_t)60 * NANOSEC; /* 1/minute */
169dtrace_optval_t dtrace_aggrate_default = NANOSEC; /* 1 hz */
170dtrace_optval_t dtrace_statusrate_default = NANOSEC; /* 1 hz */
171dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC; /* 6/minute */
172dtrace_optval_t dtrace_switchrate_default = NANOSEC; /* 1 hz */
173dtrace_optval_t dtrace_nspec_default = 1;
174dtrace_optval_t dtrace_specsize_default = 32 * 1024;
175dtrace_optval_t dtrace_stackframes_default = 20;
176dtrace_optval_t dtrace_ustackframes_default = 20;
177dtrace_optval_t dtrace_jstackframes_default = 50;
178dtrace_optval_t dtrace_jstackstrsize_default = 512;
179int dtrace_msgdsize_max = 128;
180hrtime_t dtrace_chill_max = 500 * (NANOSEC / MILLISEC); /* 500 ms */
181hrtime_t dtrace_chill_interval = NANOSEC; /* 1000 ms */
182int dtrace_devdepth_max = 32;
183int dtrace_err_verbose;
184hrtime_t dtrace_deadman_interval = NANOSEC;
185hrtime_t dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC;
186hrtime_t dtrace_deadman_user = (hrtime_t)30 * NANOSEC;
187hrtime_t dtrace_unregister_defunct_reap = (hrtime_t)60 * NANOSEC;
188
189/*
190 * DTrace External Variables
191 *
192 * As dtrace(7D) is a kernel module, any DTrace variables are obviously
193 * available to DTrace consumers via the backtick (`) syntax. One of these,
194 * dtrace_zero, is made deliberately so: it is provided as a source of
195 * well-known, zero-filled memory. While this variable is not documented,
196 * it is used by some translators as an implementation detail.
197 */
198const char dtrace_zero[256] = { 0 }; /* zero-filled memory */
199
200/*
201 * DTrace Internal Variables
202 */
203#if defined(sun)
204static dev_info_t *dtrace_devi; /* device info */
205#endif
206#if defined(sun)
207static vmem_t *dtrace_arena; /* probe ID arena */
208static vmem_t *dtrace_minor; /* minor number arena */
209#else
210static taskq_t *dtrace_taskq; /* task queue */
211static struct unrhdr *dtrace_arena; /* Probe ID number. */
212#endif
213static dtrace_probe_t **dtrace_probes; /* array of all probes */
214static int dtrace_nprobes; /* number of probes */
215static dtrace_provider_t *dtrace_provider; /* provider list */
216static dtrace_meta_t *dtrace_meta_pid; /* user-land meta provider */
217static int dtrace_opens; /* number of opens */
218static int dtrace_helpers; /* number of helpers */
219#if defined(sun)
220static void *dtrace_softstate; /* softstate pointer */
221#endif
222static dtrace_hash_t *dtrace_bymod; /* probes hashed by module */
223static dtrace_hash_t *dtrace_byfunc; /* probes hashed by function */
224static dtrace_hash_t *dtrace_byname; /* probes hashed by name */
225static dtrace_toxrange_t *dtrace_toxrange; /* toxic range array */
226static int dtrace_toxranges; /* number of toxic ranges */
227static int dtrace_toxranges_max; /* size of toxic range array */
228static dtrace_anon_t dtrace_anon; /* anonymous enabling */
229static kmem_cache_t *dtrace_state_cache; /* cache for dynamic state */
230static uint64_t dtrace_vtime_references; /* number of vtimestamp refs */
231static kthread_t *dtrace_panicked; /* panicking thread */
232static dtrace_ecb_t *dtrace_ecb_create_cache; /* cached created ECB */
233static dtrace_genid_t dtrace_probegen; /* current probe generation */
234static dtrace_helpers_t *dtrace_deferred_pid; /* deferred helper list */
235static dtrace_enabling_t *dtrace_retained; /* list of retained enablings */
236static dtrace_dynvar_t dtrace_dynhash_sink; /* end of dynamic hash chains */
237#if !defined(sun)
238static struct mtx dtrace_unr_mtx;
239MTX_SYSINIT(dtrace_unr_mtx, &dtrace_unr_mtx, "Unique resource identifier", MTX_DEF);
240int dtrace_in_probe; /* non-zero if executing a probe */
241#if defined(__i386__) || defined(__amd64__) || defined(__mips__) || defined(__powerpc__)
242uintptr_t dtrace_in_probe_addr; /* Address of invop when already in probe */
243#endif
244static eventhandler_tag dtrace_kld_load_tag;
245static eventhandler_tag dtrace_kld_unload_try_tag;
246#endif
247
248/*
249 * DTrace Locking
250 * DTrace is protected by three (relatively coarse-grained) locks:
251 *
252 * (1) dtrace_lock is required to manipulate essentially any DTrace state,
253 * including enabling state, probes, ECBs, consumer state, helper state,
254 * etc. Importantly, dtrace_lock is _not_ required when in probe context;
255 * probe context is lock-free -- synchronization is handled via the
256 * dtrace_sync() cross call mechanism.
257 *
258 * (2) dtrace_provider_lock is required when manipulating provider state, or
259 * when provider state must be held constant.
260 *
261 * (3) dtrace_meta_lock is required when manipulating meta provider state, or
262 * when meta provider state must be held constant.
263 *
264 * The lock ordering between these three locks is dtrace_meta_lock before
265 * dtrace_provider_lock before dtrace_lock. (In particular, there are
266 * several places where dtrace_provider_lock is held by the framework as it
267 * calls into the providers -- which then call back into the framework,
268 * grabbing dtrace_lock.)
269 *
270 * There are two other locks in the mix: mod_lock and cpu_lock. With respect
271 * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical
272 * role as a coarse-grained lock; it is acquired before both of these locks.
273 * With respect to dtrace_meta_lock, its behavior is stranger: cpu_lock must
274 * be acquired _between_ dtrace_meta_lock and any other DTrace locks.
275 * mod_lock is similar with respect to dtrace_provider_lock in that it must be
276 * acquired _between_ dtrace_provider_lock and dtrace_lock.
277 */
278static kmutex_t dtrace_lock; /* probe state lock */
279static kmutex_t dtrace_provider_lock; /* provider state lock */
280static kmutex_t dtrace_meta_lock; /* meta-provider state lock */
281
282#if !defined(sun)
283/* XXX FreeBSD hacks. */
284#define cr_suid cr_svuid
285#define cr_sgid cr_svgid
286#define ipaddr_t in_addr_t
287#define mod_modname pathname
288#define vuprintf vprintf
289#define ttoproc(_a) ((_a)->td_proc)
290#define crgetzoneid(_a) 0
291#define NCPU MAXCPU
292#define SNOCD 0
293#define CPU_ON_INTR(_a) 0
294
295#define PRIV_EFFECTIVE (1 << 0)
296#define PRIV_DTRACE_KERNEL (1 << 1)
297#define PRIV_DTRACE_PROC (1 << 2)
298#define PRIV_DTRACE_USER (1 << 3)
299#define PRIV_PROC_OWNER (1 << 4)
300#define PRIV_PROC_ZONE (1 << 5)
301#define PRIV_ALL ~0
302
303SYSCTL_DECL(_debug_dtrace);
304SYSCTL_DECL(_kern_dtrace);
305#endif
306
307#if defined(sun)
308#define curcpu CPU->cpu_id
309#endif
310
311
312/*
313 * DTrace Provider Variables
314 *
315 * These are the variables relating to DTrace as a provider (that is, the
316 * provider of the BEGIN, END, and ERROR probes).
317 */
318static dtrace_pattr_t dtrace_provider_attr = {
319{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
320{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
321{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
322{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
323{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
324};
325
326static void
327dtrace_nullop(void)
328{}
329
330static dtrace_pops_t dtrace_provider_ops = {
331 (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop,
332 (void (*)(void *, modctl_t *))dtrace_nullop,
333 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
334 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
335 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
336 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
337 NULL,
338 NULL,
339 NULL,
340 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop
341};
342
343static dtrace_id_t dtrace_probeid_begin; /* special BEGIN probe */
344static dtrace_id_t dtrace_probeid_end; /* special END probe */
345dtrace_id_t dtrace_probeid_error; /* special ERROR probe */
346
347/*
348 * DTrace Helper Tracing Variables
349 */
350uint32_t dtrace_helptrace_next = 0;
351uint32_t dtrace_helptrace_nlocals;
352char *dtrace_helptrace_buffer;
353int dtrace_helptrace_bufsize = 512 * 1024;
354
355#ifdef DEBUG
356int dtrace_helptrace_enabled = 1;
357#else
358int dtrace_helptrace_enabled = 0;
359#endif
360
361/*
362 * DTrace Error Hashing
363 *
364 * On DEBUG kernels, DTrace will track the errors that has seen in a hash
365 * table. This is very useful for checking coverage of tests that are
366 * expected to induce DIF or DOF processing errors, and may be useful for
367 * debugging problems in the DIF code generator or in DOF generation . The
368 * error hash may be examined with the ::dtrace_errhash MDB dcmd.
369 */
370#ifdef DEBUG
371static dtrace_errhash_t dtrace_errhash[DTRACE_ERRHASHSZ];
372static const char *dtrace_errlast;
373static kthread_t *dtrace_errthread;
374static kmutex_t dtrace_errlock;
375#endif
376
377/*
378 * DTrace Macros and Constants
379 *
380 * These are various macros that are useful in various spots in the
381 * implementation, along with a few random constants that have no meaning
382 * outside of the implementation. There is no real structure to this cpp
383 * mishmash -- but is there ever?
384 */
385#define DTRACE_HASHSTR(hash, probe) \
386 dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs)))
387
388#define DTRACE_HASHNEXT(hash, probe) \
389 (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs)
390
391#define DTRACE_HASHPREV(hash, probe) \
392 (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs)
393
394#define DTRACE_HASHEQ(hash, lhs, rhs) \
395 (strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \
396 *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0)
397
398#define DTRACE_AGGHASHSIZE_SLEW 17
399
400#define DTRACE_V4MAPPED_OFFSET (sizeof (uint32_t) * 3)
401
402/*
403 * The key for a thread-local variable consists of the lower 61 bits of the
404 * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL.
405 * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never
406 * equal to a variable identifier. This is necessary (but not sufficient) to
407 * assure that global associative arrays never collide with thread-local
408 * variables. To guarantee that they cannot collide, we must also define the
409 * order for keying dynamic variables. That order is:
410 *
411 * [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ]
412 *
413 * Because the variable-key and the tls-key are in orthogonal spaces, there is
414 * no way for a global variable key signature to match a thread-local key
415 * signature.
416 */
417#if defined(sun)
418#define DTRACE_TLS_THRKEY(where) { \
419 uint_t intr = 0; \
420 uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \
421 for (; actv; actv >>= 1) \
422 intr++; \
423 ASSERT(intr < (1 << 3)); \
424 (where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \
425 (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \
426}
427#else
428#define DTRACE_TLS_THRKEY(where) { \
429 solaris_cpu_t *_c = &solaris_cpu[curcpu]; \
430 uint_t intr = 0; \
431 uint_t actv = _c->cpu_intr_actv; \
432 for (; actv; actv >>= 1) \
433 intr++; \
434 ASSERT(intr < (1 << 3)); \
435 (where) = ((curthread->td_tid + DIF_VARIABLE_MAX) & \
436 (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \
437}
438#endif
439
440#define DT_BSWAP_8(x) ((x) & 0xff)
441#define DT_BSWAP_16(x) ((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8))
442#define DT_BSWAP_32(x) ((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16))
443#define DT_BSWAP_64(x) ((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32))
444
445#define DT_MASK_LO 0x00000000FFFFFFFFULL
446
447#define DTRACE_STORE(type, tomax, offset, what) \
448 *((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what);
449
450#ifndef __x86
451#define DTRACE_ALIGNCHECK(addr, size, flags) \
452 if (addr & (size - 1)) { \
453 *flags |= CPU_DTRACE_BADALIGN; \
454 cpu_core[curcpu].cpuc_dtrace_illval = addr; \
455 return (0); \
456 }
457#else
458#define DTRACE_ALIGNCHECK(addr, size, flags)
459#endif
460
461/*
462 * Test whether a range of memory starting at testaddr of size testsz falls
463 * within the range of memory described by addr, sz. We take care to avoid
464 * problems with overflow and underflow of the unsigned quantities, and
465 * disallow all negative sizes. Ranges of size 0 are allowed.
466 */
467#define DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \
468 ((testaddr) - (baseaddr) < (basesz) && \
469 (testaddr) + (testsz) - (baseaddr) <= (basesz) && \
470 (testaddr) + (testsz) >= (testaddr))
471
472/*
473 * Test whether alloc_sz bytes will fit in the scratch region. We isolate
474 * alloc_sz on the righthand side of the comparison in order to avoid overflow
475 * or underflow in the comparison with it. This is simpler than the INRANGE
476 * check above, because we know that the dtms_scratch_ptr is valid in the
477 * range. Allocations of size zero are allowed.
478 */
479#define DTRACE_INSCRATCH(mstate, alloc_sz) \
480 ((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \
481 (mstate)->dtms_scratch_ptr >= (alloc_sz))
482
483#define DTRACE_LOADFUNC(bits) \
484/*CSTYLED*/ \
485uint##bits##_t \
486dtrace_load##bits(uintptr_t addr) \
487{ \
488 size_t size = bits / NBBY; \
489 /*CSTYLED*/ \
490 uint##bits##_t rval; \
491 int i; \
492 volatile uint16_t *flags = (volatile uint16_t *) \
493 &cpu_core[curcpu].cpuc_dtrace_flags; \
494 \
495 DTRACE_ALIGNCHECK(addr, size, flags); \
496 \
497 for (i = 0; i < dtrace_toxranges; i++) { \
498 if (addr >= dtrace_toxrange[i].dtt_limit) \
499 continue; \
500 \
501 if (addr + size <= dtrace_toxrange[i].dtt_base) \
502 continue; \
503 \
504 /* \
505 * This address falls within a toxic region; return 0. \
506 */ \
507 *flags |= CPU_DTRACE_BADADDR; \
508 cpu_core[curcpu].cpuc_dtrace_illval = addr; \
509 return (0); \
510 } \
511 \
512 *flags |= CPU_DTRACE_NOFAULT; \
513 /*CSTYLED*/ \
514 rval = *((volatile uint##bits##_t *)addr); \
515 *flags &= ~CPU_DTRACE_NOFAULT; \
516 \
517 return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0); \
518}
519
520#ifdef _LP64
521#define dtrace_loadptr dtrace_load64
522#else
523#define dtrace_loadptr dtrace_load32
524#endif
525
526#define DTRACE_DYNHASH_FREE 0
527#define DTRACE_DYNHASH_SINK 1
528#define DTRACE_DYNHASH_VALID 2
529
530#define DTRACE_MATCH_NEXT 0
531#define DTRACE_MATCH_DONE 1
532#define DTRACE_ANCHORED(probe) ((probe)->dtpr_func[0] != '\0')
533#define DTRACE_STATE_ALIGN 64
534
535#define DTRACE_FLAGS2FLT(flags) \
536 (((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR : \
537 ((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP : \
538 ((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO : \
539 ((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV : \
540 ((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV : \
541 ((flags) & CPU_DTRACE_TUPOFLOW) ? DTRACEFLT_TUPOFLOW : \
542 ((flags) & CPU_DTRACE_BADALIGN) ? DTRACEFLT_BADALIGN : \
543 ((flags) & CPU_DTRACE_NOSCRATCH) ? DTRACEFLT_NOSCRATCH : \
544 ((flags) & CPU_DTRACE_BADSTACK) ? DTRACEFLT_BADSTACK : \
545 DTRACEFLT_UNKNOWN)
546
547#define DTRACEACT_ISSTRING(act) \
548 ((act)->dta_kind == DTRACEACT_DIFEXPR && \
549 (act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING)
550
551/* Function prototype definitions: */
552static size_t dtrace_strlen(const char *, size_t);
553static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id);
554static void dtrace_enabling_provide(dtrace_provider_t *);
555static int dtrace_enabling_match(dtrace_enabling_t *, int *);
556static void dtrace_enabling_matchall(void);
557static void dtrace_enabling_reap(void);
558static dtrace_state_t *dtrace_anon_grab(void);
559static uint64_t dtrace_helper(int, dtrace_mstate_t *,
560 dtrace_state_t *, uint64_t, uint64_t);
561static dtrace_helpers_t *dtrace_helpers_create(proc_t *);
562static void dtrace_buffer_drop(dtrace_buffer_t *);
563static int dtrace_buffer_consumed(dtrace_buffer_t *, hrtime_t when);
564static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t,
565 dtrace_state_t *, dtrace_mstate_t *);
566static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t,
567 dtrace_optval_t);
568static int dtrace_ecb_create_enable(dtrace_probe_t *, void *);
569static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *);
570uint16_t dtrace_load16(uintptr_t);
571uint32_t dtrace_load32(uintptr_t);
572uint64_t dtrace_load64(uintptr_t);
573uint8_t dtrace_load8(uintptr_t);
574void dtrace_dynvar_clean(dtrace_dstate_t *);
575dtrace_dynvar_t *dtrace_dynvar(dtrace_dstate_t *, uint_t, dtrace_key_t *,
576 size_t, dtrace_dynvar_op_t, dtrace_mstate_t *, dtrace_vstate_t *);
577uintptr_t dtrace_dif_varstr(uintptr_t, dtrace_state_t *, dtrace_mstate_t *);
578
579/*
580 * DTrace Probe Context Functions
581 *
582 * These functions are called from probe context. Because probe context is
583 * any context in which C may be called, arbitrarily locks may be held,
584 * interrupts may be disabled, we may be in arbitrary dispatched state, etc.
585 * As a result, functions called from probe context may only call other DTrace
586 * support functions -- they may not interact at all with the system at large.
587 * (Note that the ASSERT macro is made probe-context safe by redefining it in
588 * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary
589 * loads are to be performed from probe context, they _must_ be in terms of
590 * the safe dtrace_load*() variants.
591 *
592 * Some functions in this block are not actually called from probe context;
593 * for these functions, there will be a comment above the function reading
594 * "Note: not called from probe context."
595 */
596void
597dtrace_panic(const char *format, ...)
598{
599 va_list alist;
600
601 va_start(alist, format);
602 dtrace_vpanic(format, alist);
603 va_end(alist);
604}
605
606int
607dtrace_assfail(const char *a, const char *f, int l)
608{
609 dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l);
610
611 /*
612 * We just need something here that even the most clever compiler
613 * cannot optimize away.
614 */
615 return (a[(uintptr_t)f]);
616}
617
618/*
619 * Atomically increment a specified error counter from probe context.
620 */
621static void
622dtrace_error(uint32_t *counter)
623{
624 /*
625 * Most counters stored to in probe context are per-CPU counters.
626 * However, there are some error conditions that are sufficiently
627 * arcane that they don't merit per-CPU storage. If these counters
628 * are incremented concurrently on different CPUs, scalability will be
629 * adversely affected -- but we don't expect them to be white-hot in a
630 * correctly constructed enabling...
631 */
632 uint32_t oval, nval;
633
634 do {
635 oval = *counter;
636
637 if ((nval = oval + 1) == 0) {
638 /*
639 * If the counter would wrap, set it to 1 -- assuring
640 * that the counter is never zero when we have seen
641 * errors. (The counter must be 32-bits because we
642 * aren't guaranteed a 64-bit compare&swap operation.)
643 * To save this code both the infamy of being fingered
644 * by a priggish news story and the indignity of being
645 * the target of a neo-puritan witch trial, we're
646 * carefully avoiding any colorful description of the
647 * likelihood of this condition -- but suffice it to
648 * say that it is only slightly more likely than the
649 * overflow of predicate cache IDs, as discussed in
650 * dtrace_predicate_create().
651 */
652 nval = 1;
653 }
654 } while (dtrace_cas32(counter, oval, nval) != oval);
655}
656
657/*
658 * Use the DTRACE_LOADFUNC macro to define functions for each of loading a
659 * uint8_t, a uint16_t, a uint32_t and a uint64_t.
660 */
661DTRACE_LOADFUNC(8)
662DTRACE_LOADFUNC(16)
663DTRACE_LOADFUNC(32)
664DTRACE_LOADFUNC(64)
665
666static int
667dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate)
668{
669 if (dest < mstate->dtms_scratch_base)
670 return (0);
671
672 if (dest + size < dest)
673 return (0);
674
675 if (dest + size > mstate->dtms_scratch_ptr)
676 return (0);
677
678 return (1);
679}
680
681static int
682dtrace_canstore_statvar(uint64_t addr, size_t sz,
683 dtrace_statvar_t **svars, int nsvars)
684{
685 int i;
686
687 for (i = 0; i < nsvars; i++) {
688 dtrace_statvar_t *svar = svars[i];
689
690 if (svar == NULL || svar->dtsv_size == 0)
691 continue;
692
693 if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size))
694 return (1);
695 }
696
697 return (0);
698}
699
700/*
701 * Check to see if the address is within a memory region to which a store may
702 * be issued. This includes the DTrace scratch areas, and any DTrace variable
703 * region. The caller of dtrace_canstore() is responsible for performing any
704 * alignment checks that are needed before stores are actually executed.
705 */
706static int
707dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
708 dtrace_vstate_t *vstate)
709{
710 /*
711 * First, check to see if the address is in scratch space...
712 */
713 if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base,
714 mstate->dtms_scratch_size))
715 return (1);
716
717 /*
718 * Now check to see if it's a dynamic variable. This check will pick
719 * up both thread-local variables and any global dynamically-allocated
720 * variables.
721 */
722 if (DTRACE_INRANGE(addr, sz, (uintptr_t)vstate->dtvs_dynvars.dtds_base,
723 vstate->dtvs_dynvars.dtds_size)) {
724 dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
725 uintptr_t base = (uintptr_t)dstate->dtds_base +
726 (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t));
727 uintptr_t chunkoffs;
728
729 /*
730 * Before we assume that we can store here, we need to make
731 * sure that it isn't in our metadata -- storing to our
732 * dynamic variable metadata would corrupt our state. For
733 * the range to not include any dynamic variable metadata,
734 * it must:
735 *
736 * (1) Start above the hash table that is at the base of
737 * the dynamic variable space
738 *
739 * (2) Have a starting chunk offset that is beyond the
740 * dtrace_dynvar_t that is at the base of every chunk
741 *
742 * (3) Not span a chunk boundary
743 *
744 */
745 if (addr < base)
746 return (0);
747
748 chunkoffs = (addr - base) % dstate->dtds_chunksize;
749
750 if (chunkoffs < sizeof (dtrace_dynvar_t))
751 return (0);
752
753 if (chunkoffs + sz > dstate->dtds_chunksize)
754 return (0);
755
756 return (1);
757 }
758
759 /*
760 * Finally, check the static local and global variables. These checks
761 * take the longest, so we perform them last.
762 */
763 if (dtrace_canstore_statvar(addr, sz,
764 vstate->dtvs_locals, vstate->dtvs_nlocals))
765 return (1);
766
767 if (dtrace_canstore_statvar(addr, sz,
768 vstate->dtvs_globals, vstate->dtvs_nglobals))
769 return (1);
770
771 return (0);
772}
773
774
775/*
776 * Convenience routine to check to see if the address is within a memory
777 * region in which a load may be issued given the user's privilege level;
778 * if not, it sets the appropriate error flags and loads 'addr' into the
779 * illegal value slot.
780 *
781 * DTrace subroutines (DIF_SUBR_*) should use this helper to implement
782 * appropriate memory access protection.
783 */
784static int
785dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
786 dtrace_vstate_t *vstate)
787{
788 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
789
790 /*
791 * If we hold the privilege to read from kernel memory, then
792 * everything is readable.
793 */
794 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
795 return (1);
796
797 /*
798 * You can obviously read that which you can store.
799 */
800 if (dtrace_canstore(addr, sz, mstate, vstate))
801 return (1);
802
803 /*
804 * We're allowed to read from our own string table.
805 */
806 if (DTRACE_INRANGE(addr, sz, (uintptr_t)mstate->dtms_difo->dtdo_strtab,
807 mstate->dtms_difo->dtdo_strlen))
808 return (1);
809
810 DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV);
811 *illval = addr;
812 return (0);
813}
814
815/*
816 * Convenience routine to check to see if a given string is within a memory
817 * region in which a load may be issued given the user's privilege level;
818 * this exists so that we don't need to issue unnecessary dtrace_strlen()
819 * calls in the event that the user has all privileges.
820 */
821static int
822dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
823 dtrace_vstate_t *vstate)
824{
825 size_t strsz;
826
827 /*
828 * If we hold the privilege to read from kernel memory, then
829 * everything is readable.
830 */
831 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
832 return (1);
833
834 strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz);
835 if (dtrace_canload(addr, strsz, mstate, vstate))
836 return (1);
837
838 return (0);
839}
840
841/*
842 * Convenience routine to check to see if a given variable is within a memory
843 * region in which a load may be issued given the user's privilege level.
844 */
845static int
846dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate,
847 dtrace_vstate_t *vstate)
848{
849 size_t sz;
850 ASSERT(type->dtdt_flags & DIF_TF_BYREF);
851
852 /*
853 * If we hold the privilege to read from kernel memory, then
854 * everything is readable.
855 */
856 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
857 return (1);
858
859 if (type->dtdt_kind == DIF_TYPE_STRING)
860 sz = dtrace_strlen(src,
861 vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1;
862 else
863 sz = type->dtdt_size;
864
865 return (dtrace_canload((uintptr_t)src, sz, mstate, vstate));
866}
867
868/*
869 * Compare two strings using safe loads.
870 */
871static int
872dtrace_strncmp(char *s1, char *s2, size_t limit)
873{
874 uint8_t c1, c2;
875 volatile uint16_t *flags;
876
877 if (s1 == s2 || limit == 0)
878 return (0);
879
880 flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
881
882 do {
883 if (s1 == NULL) {
884 c1 = '\0';
885 } else {
886 c1 = dtrace_load8((uintptr_t)s1++);
887 }
888
889 if (s2 == NULL) {
890 c2 = '\0';
891 } else {
892 c2 = dtrace_load8((uintptr_t)s2++);
893 }
894
895 if (c1 != c2)
896 return (c1 - c2);
897 } while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT));
898
899 return (0);
900}
901
902/*
903 * Compute strlen(s) for a string using safe memory accesses. The additional
904 * len parameter is used to specify a maximum length to ensure completion.
905 */
906static size_t
907dtrace_strlen(const char *s, size_t lim)
908{
909 uint_t len;
910
911 for (len = 0; len != lim; len++) {
912 if (dtrace_load8((uintptr_t)s++) == '\0')
913 break;
914 }
915
916 return (len);
917}
918
919/*
920 * Check if an address falls within a toxic region.
921 */
922static int
923dtrace_istoxic(uintptr_t kaddr, size_t size)
924{
925 uintptr_t taddr, tsize;
926 int i;
927
928 for (i = 0; i < dtrace_toxranges; i++) {
929 taddr = dtrace_toxrange[i].dtt_base;
930 tsize = dtrace_toxrange[i].dtt_limit - taddr;
931
932 if (kaddr - taddr < tsize) {
933 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
934 cpu_core[curcpu].cpuc_dtrace_illval = kaddr;
935 return (1);
936 }
937
938 if (taddr - kaddr < size) {
939 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
940 cpu_core[curcpu].cpuc_dtrace_illval = taddr;
941 return (1);
942 }
943 }
944
945 return (0);
946}
947
948/*
949 * Copy src to dst using safe memory accesses. The src is assumed to be unsafe
950 * memory specified by the DIF program. The dst is assumed to be safe memory
951 * that we can store to directly because it is managed by DTrace. As with
952 * standard bcopy, overlapping copies are handled properly.
953 */
954static void
955dtrace_bcopy(const void *src, void *dst, size_t len)
956{
957 if (len != 0) {
958 uint8_t *s1 = dst;
959 const uint8_t *s2 = src;
960
961 if (s1 <= s2) {
962 do {
963 *s1++ = dtrace_load8((uintptr_t)s2++);
964 } while (--len != 0);
965 } else {
966 s2 += len;
967 s1 += len;
968
969 do {
970 *--s1 = dtrace_load8((uintptr_t)--s2);
971 } while (--len != 0);
972 }
973 }
974}
975
976/*
977 * Copy src to dst using safe memory accesses, up to either the specified
978 * length, or the point that a nul byte is encountered. The src is assumed to
979 * be unsafe memory specified by the DIF program. The dst is assumed to be
980 * safe memory that we can store to directly because it is managed by DTrace.
981 * Unlike dtrace_bcopy(), overlapping regions are not handled.
982 */
983static void
984dtrace_strcpy(const void *src, void *dst, size_t len)
985{
986 if (len != 0) {
987 uint8_t *s1 = dst, c;
988 const uint8_t *s2 = src;
989
990 do {
991 *s1++ = c = dtrace_load8((uintptr_t)s2++);
992 } while (--len != 0 && c != '\0');
993 }
994}
995
996/*
997 * Copy src to dst, deriving the size and type from the specified (BYREF)
998 * variable type. The src is assumed to be unsafe memory specified by the DIF
999 * program. The dst is assumed to be DTrace variable memory that is of the
1000 * specified type; we assume that we can store to directly.
1001 */
1002static void
1003dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type)
1004{
1005 ASSERT(type->dtdt_flags & DIF_TF_BYREF);
1006
1007 if (type->dtdt_kind == DIF_TYPE_STRING) {
1008 dtrace_strcpy(src, dst, type->dtdt_size);
1009 } else {
1010 dtrace_bcopy(src, dst, type->dtdt_size);
1011 }
1012}
1013
1014/*
1015 * Compare s1 to s2 using safe memory accesses. The s1 data is assumed to be
1016 * unsafe memory specified by the DIF program. The s2 data is assumed to be
1017 * safe memory that we can access directly because it is managed by DTrace.
1018 */
1019static int
1020dtrace_bcmp(const void *s1, const void *s2, size_t len)
1021{
1022 volatile uint16_t *flags;
1023
1024 flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
1025
1026 if (s1 == s2)
1027 return (0);
1028
1029 if (s1 == NULL || s2 == NULL)
1030 return (1);
1031
1032 if (s1 != s2 && len != 0) {
1033 const uint8_t *ps1 = s1;
1034 const uint8_t *ps2 = s2;
1035
1036 do {
1037 if (dtrace_load8((uintptr_t)ps1++) != *ps2++)
1038 return (1);
1039 } while (--len != 0 && !(*flags & CPU_DTRACE_FAULT));
1040 }
1041 return (0);
1042}
1043
1044/*
1045 * Zero the specified region using a simple byte-by-byte loop. Note that this
1046 * is for safe DTrace-managed memory only.
1047 */
1048static void
1049dtrace_bzero(void *dst, size_t len)
1050{
1051 uchar_t *cp;
1052
1053 for (cp = dst; len != 0; len--)
1054 *cp++ = 0;
1055}
1056
1057static void
1058dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum)
1059{
1060 uint64_t result[2];
1061
1062 result[0] = addend1[0] + addend2[0];
1063 result[1] = addend1[1] + addend2[1] +
1064 (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0);
1065
1066 sum[0] = result[0];
1067 sum[1] = result[1];
1068}
1069
1070/*
1071 * Shift the 128-bit value in a by b. If b is positive, shift left.
1072 * If b is negative, shift right.
1073 */
1074static void
1075dtrace_shift_128(uint64_t *a, int b)
1076{
1077 uint64_t mask;
1078
1079 if (b == 0)
1080 return;
1081
1082 if (b < 0) {
1083 b = -b;
1084 if (b >= 64) {
1085 a[0] = a[1] >> (b - 64);
1086 a[1] = 0;
1087 } else {
1088 a[0] >>= b;
1089 mask = 1LL << (64 - b);
1090 mask -= 1;
1091 a[0] |= ((a[1] & mask) << (64 - b));
1092 a[1] >>= b;
1093 }
1094 } else {
1095 if (b >= 64) {
1096 a[1] = a[0] << (b - 64);
1097 a[0] = 0;
1098 } else {
1099 a[1] <<= b;
1100 mask = a[0] >> (64 - b);
1101 a[1] |= mask;
1102 a[0] <<= b;
1103 }
1104 }
1105}
1106
1107/*
1108 * The basic idea is to break the 2 64-bit values into 4 32-bit values,
1109 * use native multiplication on those, and then re-combine into the
1110 * resulting 128-bit value.
1111 *
1112 * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) =
1113 * hi1 * hi2 << 64 +
1114 * hi1 * lo2 << 32 +
1115 * hi2 * lo1 << 32 +
1116 * lo1 * lo2
1117 */
1118static void
1119dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product)
1120{
1121 uint64_t hi1, hi2, lo1, lo2;
1122 uint64_t tmp[2];
1123
1124 hi1 = factor1 >> 32;
1125 hi2 = factor2 >> 32;
1126
1127 lo1 = factor1 & DT_MASK_LO;
1128 lo2 = factor2 & DT_MASK_LO;
1129
1130 product[0] = lo1 * lo2;
1131 product[1] = hi1 * hi2;
1132
1133 tmp[0] = hi1 * lo2;
1134 tmp[1] = 0;
1135 dtrace_shift_128(tmp, 32);
1136 dtrace_add_128(product, tmp, product);
1137
1138 tmp[0] = hi2 * lo1;
1139 tmp[1] = 0;
1140 dtrace_shift_128(tmp, 32);
1141 dtrace_add_128(product, tmp, product);
1142}
1143
1144/*
1145 * This privilege check should be used by actions and subroutines to
1146 * verify that the user credentials of the process that enabled the
1147 * invoking ECB match the target credentials
1148 */
1149static int
1150dtrace_priv_proc_common_user(dtrace_state_t *state)
1151{
1152 cred_t *cr, *s_cr = state->dts_cred.dcr_cred;
1153
1154 /*
1155 * We should always have a non-NULL state cred here, since if cred
1156 * is null (anonymous tracing), we fast-path bypass this routine.
1157 */
1158 ASSERT(s_cr != NULL);
1159
1160 if ((cr = CRED()) != NULL &&
1161 s_cr->cr_uid == cr->cr_uid &&
1162 s_cr->cr_uid == cr->cr_ruid &&
1163 s_cr->cr_uid == cr->cr_suid &&
1164 s_cr->cr_gid == cr->cr_gid &&
1165 s_cr->cr_gid == cr->cr_rgid &&
1166 s_cr->cr_gid == cr->cr_sgid)
1167 return (1);
1168
1169 return (0);
1170}
1171
1172/*
1173 * This privilege check should be used by actions and subroutines to
1174 * verify that the zone of the process that enabled the invoking ECB
1175 * matches the target credentials
1176 */
1177static int
1178dtrace_priv_proc_common_zone(dtrace_state_t *state)
1179{
1180#if defined(sun)
1181 cred_t *cr, *s_cr = state->dts_cred.dcr_cred;
1182
1183 /*
1184 * We should always have a non-NULL state cred here, since if cred
1185 * is null (anonymous tracing), we fast-path bypass this routine.
1186 */
1187 ASSERT(s_cr != NULL);
1188
1189 if ((cr = CRED()) != NULL &&
1190 s_cr->cr_zone == cr->cr_zone)
1191 return (1);
1192
1193 return (0);
1194#else
1195 return (1);
1196#endif
1197}
1198
1199/*
1200 * This privilege check should be used by actions and subroutines to
1201 * verify that the process has not setuid or changed credentials.
1202 */
1203static int
1204dtrace_priv_proc_common_nocd(void)
1205{
1206 proc_t *proc;
1207
1208 if ((proc = ttoproc(curthread)) != NULL &&
1209 !(proc->p_flag & SNOCD))
1210 return (1);
1211
1212 return (0);
1213}
1214
1215static int
1216dtrace_priv_proc_destructive(dtrace_state_t *state)
1217{
1218 int action = state->dts_cred.dcr_action;
1219
1220 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) &&
1221 dtrace_priv_proc_common_zone(state) == 0)
1222 goto bad;
1223
1224 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) &&
1225 dtrace_priv_proc_common_user(state) == 0)
1226 goto bad;
1227
1228 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) &&
1229 dtrace_priv_proc_common_nocd() == 0)
1230 goto bad;
1231
1232 return (1);
1233
1234bad:
1235 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1236
1237 return (0);
1238}
1239
1240static int
1241dtrace_priv_proc_control(dtrace_state_t *state)
1242{
1243 if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL)
1244 return (1);
1245
1246 if (dtrace_priv_proc_common_zone(state) &&
1247 dtrace_priv_proc_common_user(state) &&
1248 dtrace_priv_proc_common_nocd())
1249 return (1);
1250
1251 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1252
1253 return (0);
1254}
1255
1256static int
1257dtrace_priv_proc(dtrace_state_t *state)
1258{
1259 if (state->dts_cred.dcr_action & DTRACE_CRA_PROC)
1260 return (1);
1261
1262 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1263
1264 return (0);
1265}
1266
1267static int
1268dtrace_priv_kernel(dtrace_state_t *state)
1269{
1270 if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL)
1271 return (1);
1272
1273 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;
1274
1275 return (0);
1276}
1277
1278static int
1279dtrace_priv_kernel_destructive(dtrace_state_t *state)
1280{
1281 if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE)
1282 return (1);
1283
1284 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;
1285
1286 return (0);
1287}
1288
1289/*
1290 * Note: not called from probe context. This function is called
1291 * asynchronously (and at a regular interval) from outside of probe context to
1292 * clean the dirty dynamic variable lists on all CPUs. Dynamic variable
1293 * cleaning is explained in detail in <sys/dtrace_impl.h>.
1294 */
1295void
1296dtrace_dynvar_clean(dtrace_dstate_t *dstate)
1297{
1298 dtrace_dynvar_t *dirty;
1299 dtrace_dstate_percpu_t *dcpu;
1300 int i, work = 0;
1301
1302 for (i = 0; i < NCPU; i++) {
1303 dcpu = &dstate->dtds_percpu[i];
1304
1305 ASSERT(dcpu->dtdsc_rinsing == NULL);
1306
1307 /*
1308 * If the dirty list is NULL, there is no dirty work to do.
1309 */
1310 if (dcpu->dtdsc_dirty == NULL)
1311 continue;
1312
1313 /*
1314 * If the clean list is non-NULL, then we're not going to do
1315 * any work for this CPU -- it means that there has not been
1316 * a dtrace_dynvar() allocation on this CPU (or from this CPU)
1317 * since the last time we cleaned house.
1318 */
1319 if (dcpu->dtdsc_clean != NULL)
1320 continue;
1321
1322 work = 1;
1323
1324 /*
1325 * Atomically move the dirty list aside.
1326 */
1327 do {
1328 dirty = dcpu->dtdsc_dirty;
1329
1330 /*
1331 * Before we zap the dirty list, set the rinsing list.
1332 * (This allows for a potential assertion in
1333 * dtrace_dynvar(): if a free dynamic variable appears
1334 * on a hash chain, either the dirty list or the
1335 * rinsing list for some CPU must be non-NULL.)
1336 */
1337 dcpu->dtdsc_rinsing = dirty;
1338 dtrace_membar_producer();
1339 } while (dtrace_casptr(&dcpu->dtdsc_dirty,
1340 dirty, NULL) != dirty);
1341 }
1342
1343 if (!work) {
1344 /*
1345 * We have no work to do; we can simply return.
1346 */
1347 return;
1348 }
1349
1350 dtrace_sync();
1351
1352 for (i = 0; i < NCPU; i++) {
1353 dcpu = &dstate->dtds_percpu[i];
1354
1355 if (dcpu->dtdsc_rinsing == NULL)
1356 continue;
1357
1358 /*
1359 * We are now guaranteed that no hash chain contains a pointer
1360 * into this dirty list; we can make it clean.
1361 */
1362 ASSERT(dcpu->dtdsc_clean == NULL);
1363 dcpu->dtdsc_clean = dcpu->dtdsc_rinsing;
1364 dcpu->dtdsc_rinsing = NULL;
1365 }
1366
1367 /*
1368 * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make
1369 * sure that all CPUs have seen all of the dtdsc_clean pointers.
1370 * This prevents a race whereby a CPU incorrectly decides that
1371 * the state should be something other than DTRACE_DSTATE_CLEAN
1372 * after dtrace_dynvar_clean() has completed.
1373 */
1374 dtrace_sync();
1375
1376 dstate->dtds_state = DTRACE_DSTATE_CLEAN;
1377}
1378
1379/*
1380 * Depending on the value of the op parameter, this function looks-up,
1381 * allocates or deallocates an arbitrarily-keyed dynamic variable. If an
1382 * allocation is requested, this function will return a pointer to a
1383 * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no
1384 * variable can be allocated. If NULL is returned, the appropriate counter
1385 * will be incremented.
1386 */
1387dtrace_dynvar_t *
1388dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys,
1389 dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op,
1390 dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
1391{
1392 uint64_t hashval = DTRACE_DYNHASH_VALID;
1393 dtrace_dynhash_t *hash = dstate->dtds_hash;
1394 dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL;
1395 processorid_t me = curcpu, cpu = me;
1396 dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me];
1397 size_t bucket, ksize;
1398 size_t chunksize = dstate->dtds_chunksize;
1399 uintptr_t kdata, lock, nstate;
1400 uint_t i;
1401
1402 ASSERT(nkeys != 0);
1403
1404 /*
1405 * Hash the key. As with aggregations, we use Jenkins' "One-at-a-time"
1406 * algorithm. For the by-value portions, we perform the algorithm in
1407 * 16-bit chunks (as opposed to 8-bit chunks). This speeds things up a
1408 * bit, and seems to have only a minute effect on distribution. For
1409 * the by-reference data, we perform "One-at-a-time" iterating (safely)
1410 * over each referenced byte. It's painful to do this, but it's much
1411 * better than pathological hash distribution. The efficacy of the
1412 * hashing algorithm (and a comparison with other algorithms) may be
1413 * found by running the ::dtrace_dynstat MDB dcmd.
1414 */
1415 for (i = 0; i < nkeys; i++) {
1416 if (key[i].dttk_size == 0) {
1417 uint64_t val = key[i].dttk_value;
1418
1419 hashval += (val >> 48) & 0xffff;
1420 hashval += (hashval << 10);
1421 hashval ^= (hashval >> 6);
1422
1423 hashval += (val >> 32) & 0xffff;
1424 hashval += (hashval << 10);
1425 hashval ^= (hashval >> 6);
1426
1427 hashval += (val >> 16) & 0xffff;
1428 hashval += (hashval << 10);
1429 hashval ^= (hashval >> 6);
1430
1431 hashval += val & 0xffff;
1432 hashval += (hashval << 10);
1433 hashval ^= (hashval >> 6);
1434 } else {
1435 /*
1436 * This is incredibly painful, but it beats the hell
1437 * out of the alternative.
1438 */
1439 uint64_t j, size = key[i].dttk_size;
1440 uintptr_t base = (uintptr_t)key[i].dttk_value;
1441
1442 if (!dtrace_canload(base, size, mstate, vstate))
1443 break;
1444
1445 for (j = 0; j < size; j++) {
1446 hashval += dtrace_load8(base + j);
1447 hashval += (hashval << 10);
1448 hashval ^= (hashval >> 6);
1449 }
1450 }
1451 }
1452
1453 if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT))
1454 return (NULL);
1455
1456 hashval += (hashval << 3);
1457 hashval ^= (hashval >> 11);
1458 hashval += (hashval << 15);
1459
1460 /*
1461 * There is a remote chance (ideally, 1 in 2^31) that our hashval
1462 * comes out to be one of our two sentinel hash values. If this
1463 * actually happens, we set the hashval to be a value known to be a
1464 * non-sentinel value.
1465 */
1466 if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK)
1467 hashval = DTRACE_DYNHASH_VALID;
1468
1469 /*
1470 * Yes, it's painful to do a divide here. If the cycle count becomes
1471 * important here, tricks can be pulled to reduce it. (However, it's
1472 * critical that hash collisions be kept to an absolute minimum;
1473 * they're much more painful than a divide.) It's better to have a
1474 * solution that generates few collisions and still keeps things
1475 * relatively simple.
1476 */
1477 bucket = hashval % dstate->dtds_hashsize;
1478
1479 if (op == DTRACE_DYNVAR_DEALLOC) {
1480 volatile uintptr_t *lockp = &hash[bucket].dtdh_lock;
1481
1482 for (;;) {
1483 while ((lock = *lockp) & 1)
1484 continue;
1485
1486 if (dtrace_casptr((volatile void *)lockp,
1487 (volatile void *)lock, (volatile void *)(lock + 1)) == (void *)lock)
1488 break;
1489 }
1490
1491 dtrace_membar_producer();
1492 }
1493
1494top:
1495 prev = NULL;
1496 lock = hash[bucket].dtdh_lock;
1497
1498 dtrace_membar_consumer();
1499
1500 start = hash[bucket].dtdh_chain;
1501 ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK ||
1502 start->dtdv_hashval != DTRACE_DYNHASH_FREE ||
1503 op != DTRACE_DYNVAR_DEALLOC));
1504
1505 for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) {
1506 dtrace_tuple_t *dtuple = &dvar->dtdv_tuple;
1507 dtrace_key_t *dkey = &dtuple->dtt_key[0];
1508
1509 if (dvar->dtdv_hashval != hashval) {
1510 if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) {
1511 /*
1512 * We've reached the sink, and therefore the
1513 * end of the hash chain; we can kick out of
1514 * the loop knowing that we have seen a valid
1515 * snapshot of state.
1516 */
1517 ASSERT(dvar->dtdv_next == NULL);
1518 ASSERT(dvar == &dtrace_dynhash_sink);
1519 break;
1520 }
1521
1522 if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) {
1523 /*
1524 * We've gone off the rails: somewhere along
1525 * the line, one of the members of this hash
1526 * chain was deleted. Note that we could also
1527 * detect this by simply letting this loop run
1528 * to completion, as we would eventually hit
1529 * the end of the dirty list. However, we
1530 * want to avoid running the length of the
1531 * dirty list unnecessarily (it might be quite
1532 * long), so we catch this as early as
1533 * possible by detecting the hash marker. In
1534 * this case, we simply set dvar to NULL and
1535 * break; the conditional after the loop will
1536 * send us back to top.
1537 */
1538 dvar = NULL;
1539 break;
1540 }
1541
1542 goto next;
1543 }
1544
1545 if (dtuple->dtt_nkeys != nkeys)
1546 goto next;
1547
1548 for (i = 0; i < nkeys; i++, dkey++) {
1549 if (dkey->dttk_size != key[i].dttk_size)
1550 goto next; /* size or type mismatch */
1551
1552 if (dkey->dttk_size != 0) {
1553 if (dtrace_bcmp(
1554 (void *)(uintptr_t)key[i].dttk_value,
1555 (void *)(uintptr_t)dkey->dttk_value,
1556 dkey->dttk_size))
1557 goto next;
1558 } else {
1559 if (dkey->dttk_value != key[i].dttk_value)
1560 goto next;
1561 }
1562 }
1563
1564 if (op != DTRACE_DYNVAR_DEALLOC)
1565 return (dvar);
1566
1567 ASSERT(dvar->dtdv_next == NULL ||
1568 dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE);
1569
1570 if (prev != NULL) {
1571 ASSERT(hash[bucket].dtdh_chain != dvar);
1572 ASSERT(start != dvar);
1573 ASSERT(prev->dtdv_next == dvar);
1574 prev->dtdv_next = dvar->dtdv_next;
1575 } else {
1576 if (dtrace_casptr(&hash[bucket].dtdh_chain,
1577 start, dvar->dtdv_next) != start) {
1578 /*
1579 * We have failed to atomically swing the
1580 * hash table head pointer, presumably because
1581 * of a conflicting allocation on another CPU.
1582 * We need to reread the hash chain and try
1583 * again.
1584 */
1585 goto top;
1586 }
1587 }
1588
1589 dtrace_membar_producer();
1590
1591 /*
1592 * Now set the hash value to indicate that it's free.
1593 */
1594 ASSERT(hash[bucket].dtdh_chain != dvar);
1595 dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;
1596
1597 dtrace_membar_producer();
1598
1599 /*
1600 * Set the next pointer to point at the dirty list, and
1601 * atomically swing the dirty pointer to the newly freed dvar.
1602 */
1603 do {
1604 next = dcpu->dtdsc_dirty;
1605 dvar->dtdv_next = next;
1606 } while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next);
1607
1608 /*
1609 * Finally, unlock this hash bucket.
1610 */
1611 ASSERT(hash[bucket].dtdh_lock == lock);
1612 ASSERT(lock & 1);
1613 hash[bucket].dtdh_lock++;
1614
1615 return (NULL);
1616next:
1617 prev = dvar;
1618 continue;
1619 }
1620
1621 if (dvar == NULL) {
1622 /*
1623 * If dvar is NULL, it is because we went off the rails:
1624 * one of the elements that we traversed in the hash chain
1625 * was deleted while we were traversing it. In this case,
1626 * we assert that we aren't doing a dealloc (deallocs lock
1627 * the hash bucket to prevent themselves from racing with
1628 * one another), and retry the hash chain traversal.
1629 */
1630 ASSERT(op != DTRACE_DYNVAR_DEALLOC);
1631 goto top;
1632 }
1633
1634 if (op != DTRACE_DYNVAR_ALLOC) {
1635 /*
1636 * If we are not to allocate a new variable, we want to
1637 * return NULL now. Before we return, check that the value
1638 * of the lock word hasn't changed. If it has, we may have
1639 * seen an inconsistent snapshot.
1640 */
1641 if (op == DTRACE_DYNVAR_NOALLOC) {
1642 if (hash[bucket].dtdh_lock != lock)
1643 goto top;
1644 } else {
1645 ASSERT(op == DTRACE_DYNVAR_DEALLOC);
1646 ASSERT(hash[bucket].dtdh_lock == lock);
1647 ASSERT(lock & 1);
1648 hash[bucket].dtdh_lock++;
1649 }
1650
1651 return (NULL);
1652 }
1653
1654 /*
1655 * We need to allocate a new dynamic variable. The size we need is the
1656 * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the
1657 * size of any auxiliary key data (rounded up to 8-byte alignment) plus
1658 * the size of any referred-to data (dsize). We then round the final
1659 * size up to the chunksize for allocation.
1660 */
1661 for (ksize = 0, i = 0; i < nkeys; i++)
1662 ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));
1663
1664 /*
1665 * This should be pretty much impossible, but could happen if, say,
1666 * strange DIF specified the tuple. Ideally, this should be an
1667 * assertion and not an error condition -- but that requires that the
1668 * chunksize calculation in dtrace_difo_chunksize() be absolutely
1669 * bullet-proof. (That is, it must not be able to be fooled by
1670 * malicious DIF.) Given the lack of backwards branches in DIF,
1671 * solving this would presumably not amount to solving the Halting
1672 * Problem -- but it still seems awfully hard.
1673 */
1674 if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) +
1675 ksize + dsize > chunksize) {
1676 dcpu->dtdsc_drops++;
1677 return (NULL);
1678 }
1679
1680 nstate = DTRACE_DSTATE_EMPTY;
1681
1682 do {
1683retry:
1684 free = dcpu->dtdsc_free;
1685
1686 if (free == NULL) {
1687 dtrace_dynvar_t *clean = dcpu->dtdsc_clean;
1688 void *rval;
1689
1690 if (clean == NULL) {
1691 /*
1692 * We're out of dynamic variable space on
1693 * this CPU. Unless we have tried all CPUs,
1694 * we'll try to allocate from a different
1695 * CPU.
1696 */
1697 switch (dstate->dtds_state) {
1698 case DTRACE_DSTATE_CLEAN: {
1699 void *sp = &dstate->dtds_state;
1700
1701 if (++cpu >= NCPU)
1702 cpu = 0;
1703
1704 if (dcpu->dtdsc_dirty != NULL &&
1705 nstate == DTRACE_DSTATE_EMPTY)
1706 nstate = DTRACE_DSTATE_DIRTY;
1707
1708 if (dcpu->dtdsc_rinsing != NULL)
1709 nstate = DTRACE_DSTATE_RINSING;
1710
1711 dcpu = &dstate->dtds_percpu[cpu];
1712
1713 if (cpu != me)
1714 goto retry;
1715
1716 (void) dtrace_cas32(sp,
1717 DTRACE_DSTATE_CLEAN, nstate);
1718
1719 /*
1720 * To increment the correct bean
1721 * counter, take another lap.
1722 */
1723 goto retry;
1724 }
1725
1726 case DTRACE_DSTATE_DIRTY:
1727 dcpu->dtdsc_dirty_drops++;
1728 break;
1729
1730 case DTRACE_DSTATE_RINSING:
1731 dcpu->dtdsc_rinsing_drops++;
1732 break;
1733
1734 case DTRACE_DSTATE_EMPTY:
1735 dcpu->dtdsc_drops++;
1736 break;
1737 }
1738
1739 DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP);
1740 return (NULL);
1741 }
1742
1743 /*
1744 * The clean list appears to be non-empty. We want to
1745 * move the clean list to the free list; we start by
1746 * moving the clean pointer aside.
1747 */
1748 if (dtrace_casptr(&dcpu->dtdsc_clean,
1749 clean, NULL) != clean) {
1750 /*
1751 * We are in one of two situations:
1752 *
1753 * (a) The clean list was switched to the
1754 * free list by another CPU.
1755 *
1756 * (b) The clean list was added to by the
1757 * cleansing cyclic.
1758 *
1759 * In either of these situations, we can
1760 * just reattempt the free list allocation.
1761 */
1762 goto retry;
1763 }
1764
1765 ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE);
1766
1767 /*
1768 * Now we'll move the clean list to the free list.
1769 * It's impossible for this to fail: the only way
1770 * the free list can be updated is through this
1771 * code path, and only one CPU can own the clean list.
1772 * Thus, it would only be possible for this to fail if
1773 * this code were racing with dtrace_dynvar_clean().
1774 * (That is, if dtrace_dynvar_clean() updated the clean
1775 * list, and we ended up racing to update the free
1776 * list.) This race is prevented by the dtrace_sync()
1777 * in dtrace_dynvar_clean() -- which flushes the
1778 * owners of the clean lists out before resetting
1779 * the clean lists.
1780 */
1781 rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean);
1782 ASSERT(rval == NULL);
1783 goto retry;
1784 }
1785
1786 dvar = free;
1787 new_free = dvar->dtdv_next;
1788 } while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free);
1789
1790 /*
1791 * We have now allocated a new chunk. We copy the tuple keys into the
1792 * tuple array and copy any referenced key data into the data space
1793 * following the tuple array. As we do this, we relocate dttk_value
1794 * in the final tuple to point to the key data address in the chunk.
1795 */
1796 kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys];
1797 dvar->dtdv_data = (void *)(kdata + ksize);
1798 dvar->dtdv_tuple.dtt_nkeys = nkeys;
1799
1800 for (i = 0; i < nkeys; i++) {
1801 dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i];
1802 size_t kesize = key[i].dttk_size;
1803
1804 if (kesize != 0) {
1805 dtrace_bcopy(
1806 (const void *)(uintptr_t)key[i].dttk_value,
1807 (void *)kdata, kesize);
1808 dkey->dttk_value = kdata;
1809 kdata += P2ROUNDUP(kesize, sizeof (uint64_t));
1810 } else {
1811 dkey->dttk_value = key[i].dttk_value;
1812 }
1813
1814 dkey->dttk_size = kesize;
1815 }
1816
1817 ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE);
1818 dvar->dtdv_hashval = hashval;
1819 dvar->dtdv_next = start;
1820
1821 if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start)
1822 return (dvar);
1823
1824 /*
1825 * The cas has failed. Either another CPU is adding an element to
1826 * this hash chain, or another CPU is deleting an element from this
1827 * hash chain. The simplest way to deal with both of these cases
1828 * (though not necessarily the most efficient) is to free our
1829 * allocated block and tail-call ourselves. Note that the free is
1830 * to the dirty list and _not_ to the free list. This is to prevent
1831 * races with allocators, above.
1832 */
1833 dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;
1834
1835 dtrace_membar_producer();
1836
1837 do {
1838 free = dcpu->dtdsc_dirty;
1839 dvar->dtdv_next = free;
1840 } while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free);
1841
1842 return (dtrace_dynvar(dstate, nkeys, key, dsize, op, mstate, vstate));
1843}
1844
1845/*ARGSUSED*/
1846static void
1847dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg)
1848{
1849 if ((int64_t)nval < (int64_t)*oval)
1850 *oval = nval;
1851}
1852
1853/*ARGSUSED*/
1854static void
1855dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg)
1856{
1857 if ((int64_t)nval > (int64_t)*oval)
1858 *oval = nval;
1859}
1860
1861static void
1862dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr)
1863{
1864 int i, zero = DTRACE_QUANTIZE_ZEROBUCKET;
1865 int64_t val = (int64_t)nval;
1866
1867 if (val < 0) {
1868 for (i = 0; i < zero; i++) {
1869 if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) {
1870 quanta[i] += incr;
1871 return;
1872 }
1873 }
1874 } else {
1875 for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) {
1876 if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) {
1877 quanta[i - 1] += incr;
1878 return;
1879 }
1880 }
1881
1882 quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr;
1883 return;
1884 }
1885
1886 ASSERT(0);
1887}
1888
1889static void
1890dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr)
1891{
1892 uint64_t arg = *lquanta++;
1893 int32_t base = DTRACE_LQUANTIZE_BASE(arg);
1894 uint16_t step = DTRACE_LQUANTIZE_STEP(arg);
1895 uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg);
1896 int32_t val = (int32_t)nval, level;
1897
1898 ASSERT(step != 0);
1899 ASSERT(levels != 0);
1900
1901 if (val < base) {
1902 /*
1903 * This is an underflow.
1904 */
1905 lquanta[0] += incr;
1906 return;
1907 }
1908
1909 level = (val - base) / step;
1910
1911 if (level < levels) {
1912 lquanta[level + 1] += incr;
1913 return;
1914 }
1915
1916 /*
1917 * This is an overflow.
1918 */
1919 lquanta[levels + 1] += incr;
1920}
1921
1922static int
1923dtrace_aggregate_llquantize_bucket(uint16_t factor, uint16_t low,
1924 uint16_t high, uint16_t nsteps, int64_t value)
1925{
1926 int64_t this = 1, last, next;
1927 int base = 1, order;
1928
1929 ASSERT(factor <= nsteps);
1930 ASSERT(nsteps % factor == 0);
1931
1932 for (order = 0; order < low; order++)
1933 this *= factor;
1934
1935 /*
1936 * If our value is less than our factor taken to the power of the
1937 * low order of magnitude, it goes into the zeroth bucket.
1938 */
1939 if (value < (last = this))
1940 return (0);
1941
1942 for (this *= factor; order <= high; order++) {
1943 int nbuckets = this > nsteps ? nsteps : this;
1944
1945 if ((next = this * factor) < this) {
1946 /*
1947 * We should not generally get log/linear quantizations
1948 * with a high magnitude that allows 64-bits to
1949 * overflow, but we nonetheless protect against this
1950 * by explicitly checking for overflow, and clamping
1951 * our value accordingly.
1952 */
1953 value = this - 1;
1954 }
1955
1956 if (value < this) {
1957 /*
1958 * If our value lies within this order of magnitude,
1959 * determine its position by taking the offset within
1960 * the order of magnitude, dividing by the bucket
1961 * width, and adding to our (accumulated) base.
1962 */
1963 return (base + (value - last) / (this / nbuckets));
1964 }
1965
1966 base += nbuckets - (nbuckets / factor);
1967 last = this;
1968 this = next;
1969 }
1970
1971 /*
1972 * Our value is greater than or equal to our factor taken to the
1973 * power of one plus the high magnitude -- return the top bucket.
1974 */
1975 return (base);
1976}
1977
1978static void
1979dtrace_aggregate_llquantize(uint64_t *llquanta, uint64_t nval, uint64_t incr)
1980{
1981 uint64_t arg = *llquanta++;
1982 uint16_t factor = DTRACE_LLQUANTIZE_FACTOR(arg);
1983 uint16_t low = DTRACE_LLQUANTIZE_LOW(arg);
1984 uint16_t high = DTRACE_LLQUANTIZE_HIGH(arg);
1985 uint16_t nsteps = DTRACE_LLQUANTIZE_NSTEP(arg);
1986
1987 llquanta[dtrace_aggregate_llquantize_bucket(factor,
1988 low, high, nsteps, nval)] += incr;
1989}
1990
1991/*ARGSUSED*/
1992static void
1993dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg)
1994{
1995 data[0]++;
1996 data[1] += nval;
1997}
1998
1999/*ARGSUSED*/
2000static void
2001dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg)
2002{
2003 int64_t snval = (int64_t)nval;
2004 uint64_t tmp[2];
2005
2006 data[0]++;
2007 data[1] += nval;
2008
2009 /*
2010 * What we want to say here is:
2011 *
2012 * data[2] += nval * nval;
2013 *
2014 * But given that nval is 64-bit, we could easily overflow, so
2015 * we do this as 128-bit arithmetic.
2016 */
2017 if (snval < 0)
2018 snval = -snval;
2019
2020 dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp);
2021 dtrace_add_128(data + 2, tmp, data + 2);
2022}
2023
2024/*ARGSUSED*/
2025static void
2026dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg)
2027{
2028 *oval = *oval + 1;
2029}
2030
2031/*ARGSUSED*/
2032static void
2033dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg)
2034{
2035 *oval += nval;
2036}
2037
2038/*
2039 * Aggregate given the tuple in the principal data buffer, and the aggregating
2040 * action denoted by the specified dtrace_aggregation_t. The aggregation
2041 * buffer is specified as the buf parameter. This routine does not return
2042 * failure; if there is no space in the aggregation buffer, the data will be
2043 * dropped, and a corresponding counter incremented.
2044 */
2045static void
2046dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf,
2047 intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg)
2048{
2049 dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec;
2050 uint32_t i, ndx, size, fsize;
2051 uint32_t align = sizeof (uint64_t) - 1;
2052 dtrace_aggbuffer_t *agb;
2053 dtrace_aggkey_t *key;
2054 uint32_t hashval = 0, limit, isstr;
2055 caddr_t tomax, data, kdata;
2056 dtrace_actkind_t action;
2057 dtrace_action_t *act;
2058 uintptr_t offs;
2059
2060 if (buf == NULL)
2061 return;
2062
2063 if (!agg->dtag_hasarg) {
2064 /*
2065 * Currently, only quantize() and lquantize() take additional
2066 * arguments, and they have the same semantics: an increment
2067 * value that defaults to 1 when not present. If additional
2068 * aggregating actions take arguments, the setting of the
2069 * default argument value will presumably have to become more
2070 * sophisticated...
2071 */
2072 arg = 1;
2073 }
2074
2075 action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION;
2076 size = rec->dtrd_offset - agg->dtag_base;
2077 fsize = size + rec->dtrd_size;
2078
2079 ASSERT(dbuf->dtb_tomax != NULL);
2080 data = dbuf->dtb_tomax + offset + agg->dtag_base;
2081
2082 if ((tomax = buf->dtb_tomax) == NULL) {
2083 dtrace_buffer_drop(buf);
2084 return;
2085 }
2086
2087 /*
2088 * The metastructure is always at the bottom of the buffer.
2089 */
2090 agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size -
2091 sizeof (dtrace_aggbuffer_t));
2092
2093 if (buf->dtb_offset == 0) {
2094 /*
2095 * We just kludge up approximately 1/8th of the size to be
2096 * buckets. If this guess ends up being routinely
2097 * off-the-mark, we may need to dynamically readjust this
2098 * based on past performance.
2099 */
2100 uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t);
2101
2102 if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) <
2103 (uintptr_t)tomax || hashsize == 0) {
2104 /*
2105 * We've been given a ludicrously small buffer;
2106 * increment our drop count and leave.
2107 */
2108 dtrace_buffer_drop(buf);
2109 return;
2110 }
2111
2112 /*
2113 * And now, a pathetic attempt to try to get a an odd (or
2114 * perchance, a prime) hash size for better hash distribution.
2115 */
2116 if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3))
2117 hashsize -= DTRACE_AGGHASHSIZE_SLEW;
2118
2119 agb->dtagb_hashsize = hashsize;
2120 agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb -
2121 agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *));
2122 agb->dtagb_free = (uintptr_t)agb->dtagb_hash;
2123
2124 for (i = 0; i < agb->dtagb_hashsize; i++)
2125 agb->dtagb_hash[i] = NULL;
2126 }
2127
2128 ASSERT(agg->dtag_first != NULL);
2129 ASSERT(agg->dtag_first->dta_intuple);
2130
2131 /*
2132 * Calculate the hash value based on the key. Note that we _don't_
2133 * include the aggid in the hashing (but we will store it as part of
2134 * the key). The hashing algorithm is Bob Jenkins' "One-at-a-time"
2135 * algorithm: a simple, quick algorithm that has no known funnels, and
2136 * gets good distribution in practice. The efficacy of the hashing
2137 * algorithm (and a comparison with other algorithms) may be found by
2138 * running the ::dtrace_aggstat MDB dcmd.
2139 */
2140 for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
2141 i = act->dta_rec.dtrd_offset - agg->dtag_base;
2142 limit = i + act->dta_rec.dtrd_size;
2143 ASSERT(limit <= size);
2144 isstr = DTRACEACT_ISSTRING(act);
2145
2146 for (; i < limit; i++) {
2147 hashval += data[i];
2148 hashval += (hashval << 10);
2149 hashval ^= (hashval >> 6);
2150
2151 if (isstr && data[i] == '\0')
2152 break;
2153 }
2154 }
2155
2156 hashval += (hashval << 3);
2157 hashval ^= (hashval >> 11);
2158 hashval += (hashval << 15);
2159
2160 /*
2161 * Yes, the divide here is expensive -- but it's generally the least
2162 * of the performance issues given the amount of data that we iterate
2163 * over to compute hash values, compare data, etc.
2164 */
2165 ndx = hashval % agb->dtagb_hashsize;
2166
2167 for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) {
2168 ASSERT((caddr_t)key >= tomax);
2169 ASSERT((caddr_t)key < tomax + buf->dtb_size);
2170
2171 if (hashval != key->dtak_hashval || key->dtak_size != size)
2172 continue;
2173
2174 kdata = key->dtak_data;
2175 ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size);
2176
2177 for (act = agg->dtag_first; act->dta_intuple;
2178 act = act->dta_next) {
2179 i = act->dta_rec.dtrd_offset - agg->dtag_base;
2180 limit = i + act->dta_rec.dtrd_size;
2181 ASSERT(limit <= size);
2182 isstr = DTRACEACT_ISSTRING(act);
2183
2184 for (; i < limit; i++) {
2185 if (kdata[i] != data[i])
2186 goto next;
2187
2188 if (isstr && data[i] == '\0')
2189 break;
2190 }
2191 }
2192
2193 if (action != key->dtak_action) {
2194 /*
2195 * We are aggregating on the same value in the same
2196 * aggregation with two different aggregating actions.
2197 * (This should have been picked up in the compiler,
2198 * so we may be dealing with errant or devious DIF.)
2199 * This is an error condition; we indicate as much,
2200 * and return.
2201 */
2202 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
2203 return;
2204 }
2205
2206 /*
2207 * This is a hit: we need to apply the aggregator to
2208 * the value at this key.
2209 */
2210 agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg);
2211 return;
2212next:
2213 continue;
2214 }
2215
2216 /*
2217 * We didn't find it. We need to allocate some zero-filled space,
2218 * link it into the hash table appropriately, and apply the aggregator
2219 * to the (zero-filled) value.
2220 */
2221 offs = buf->dtb_offset;
2222 while (offs & (align - 1))
2223 offs += sizeof (uint32_t);
2224
2225 /*
2226 * If we don't have enough room to both allocate a new key _and_
2227 * its associated data, increment the drop count and return.
2228 */
2229 if ((uintptr_t)tomax + offs + fsize >
2230 agb->dtagb_free - sizeof (dtrace_aggkey_t)) {
2231 dtrace_buffer_drop(buf);
2232 return;
2233 }
2234
2235 /*CONSTCOND*/
2236 ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1)));
2237 key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t));
2238 agb->dtagb_free -= sizeof (dtrace_aggkey_t);
2239
2240 key->dtak_data = kdata = tomax + offs;
2241 buf->dtb_offset = offs + fsize;
2242
2243 /*
2244 * Now copy the data across.
2245 */
2246 *((dtrace_aggid_t *)kdata) = agg->dtag_id;
2247
2248 for (i = sizeof (dtrace_aggid_t); i < size; i++)
2249 kdata[i] = data[i];
2250
2251 /*
2252 * Because strings are not zeroed out by default, we need to iterate
2253 * looking for actions that store strings, and we need to explicitly
2254 * pad these strings out with zeroes.
2255 */
2256 for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
2257 int nul;
2258
2259 if (!DTRACEACT_ISSTRING(act))
2260 continue;
2261
2262 i = act->dta_rec.dtrd_offset - agg->dtag_base;
2263 limit = i + act->dta_rec.dtrd_size;
2264 ASSERT(limit <= size);
2265
2266 for (nul = 0; i < limit; i++) {
2267 if (nul) {
2268 kdata[i] = '\0';
2269 continue;
2270 }
2271
2272 if (data[i] != '\0')
2273 continue;
2274
2275 nul = 1;
2276 }
2277 }
2278
2279 for (i = size; i < fsize; i++)
2280 kdata[i] = 0;
2281
2282 key->dtak_hashval = hashval;
2283 key->dtak_size = size;
2284 key->dtak_action = action;
2285 key->dtak_next = agb->dtagb_hash[ndx];
2286 agb->dtagb_hash[ndx] = key;
2287
2288 /*
2289 * Finally, apply the aggregator.
2290 */
2291 *((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial;
2292 agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg);
2293}
2294
2295/*
2296 * Given consumer state, this routine finds a speculation in the INACTIVE
2297 * state and transitions it into the ACTIVE state. If there is no speculation
2298 * in the INACTIVE state, 0 is returned. In this case, no error counter is
2299 * incremented -- it is up to the caller to take appropriate action.
2300 */
2301static int
2302dtrace_speculation(dtrace_state_t *state)
2303{
2304 int i = 0;
2305 dtrace_speculation_state_t current;
2306 uint32_t *stat = &state->dts_speculations_unavail, count;
2307
2308 while (i < state->dts_nspeculations) {
2309 dtrace_speculation_t *spec = &state->dts_speculations[i];
2310
2311 current = spec->dtsp_state;
2312
2313 if (current != DTRACESPEC_INACTIVE) {
2314 if (current == DTRACESPEC_COMMITTINGMANY ||
2315 current == DTRACESPEC_COMMITTING ||
2316 current == DTRACESPEC_DISCARDING)
2317 stat = &state->dts_speculations_busy;
2318 i++;
2319 continue;
2320 }
2321
2322 if (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2323 current, DTRACESPEC_ACTIVE) == current)
2324 return (i + 1);
2325 }
2326
2327 /*
2328 * We couldn't find a speculation. If we found as much as a single
2329 * busy speculation buffer, we'll attribute this failure as "busy"
2330 * instead of "unavail".
2331 */
2332 do {
2333 count = *stat;
2334 } while (dtrace_cas32(stat, count, count + 1) != count);
2335
2336 return (0);
2337}
2338
2339/*
2340 * This routine commits an active speculation. If the specified speculation
2341 * is not in a valid state to perform a commit(), this routine will silently do
2342 * nothing. The state of the specified speculation is transitioned according
2343 * to the state transition diagram outlined in <sys/dtrace_impl.h>
2344 */
2345static void
2346dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu,
2347 dtrace_specid_t which)
2348{
2349 dtrace_speculation_t *spec;
2350 dtrace_buffer_t *src, *dest;
2351 uintptr_t daddr, saddr, dlimit, slimit;
2352 dtrace_speculation_state_t current, new = 0;
2353 intptr_t offs;
2354 uint64_t timestamp;
2355
2356 if (which == 0)
2357 return;
2358
2359 if (which > state->dts_nspeculations) {
2360 cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2361 return;
2362 }
2363
2364 spec = &state->dts_speculations[which - 1];
2365 src = &spec->dtsp_buffer[cpu];
2366 dest = &state->dts_buffer[cpu];
2367
2368 do {
2369 current = spec->dtsp_state;
2370
2371 if (current == DTRACESPEC_COMMITTINGMANY)
2372 break;
2373
2374 switch (current) {
2375 case DTRACESPEC_INACTIVE:
2376 case DTRACESPEC_DISCARDING:
2377 return;
2378
2379 case DTRACESPEC_COMMITTING:
2380 /*
2381 * This is only possible if we are (a) commit()'ing
2382 * without having done a prior speculate() on this CPU
2383 * and (b) racing with another commit() on a different
2384 * CPU. There's nothing to do -- we just assert that
2385 * our offset is 0.
2386 */
2387 ASSERT(src->dtb_offset == 0);
2388 return;
2389
2390 case DTRACESPEC_ACTIVE:
2391 new = DTRACESPEC_COMMITTING;
2392 break;
2393
2394 case DTRACESPEC_ACTIVEONE:
2395 /*
2396 * This speculation is active on one CPU. If our
2397 * buffer offset is non-zero, we know that the one CPU
2398 * must be us. Otherwise, we are committing on a
2399 * different CPU from the speculate(), and we must
2400 * rely on being asynchronously cleaned.
2401 */
2402 if (src->dtb_offset != 0) {
2403 new = DTRACESPEC_COMMITTING;
2404 break;
2405 }
2406 /*FALLTHROUGH*/
2407
2408 case DTRACESPEC_ACTIVEMANY:
2409 new = DTRACESPEC_COMMITTINGMANY;
2410 break;
2411
2412 default:
2413 ASSERT(0);
2414 }
2415 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2416 current, new) != current);
2417
2418 /*
2419 * We have set the state to indicate that we are committing this
2420 * speculation. Now reserve the necessary space in the destination
2421 * buffer.
2422 */
2423 if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset,
2424 sizeof (uint64_t), state, NULL)) < 0) {
2425 dtrace_buffer_drop(dest);
2426 goto out;
2427 }
2428
2429 /*
2430 * We have sufficient space to copy the speculative buffer into the
2431 * primary buffer. First, modify the speculative buffer, filling
2432 * in the timestamp of all entries with the current time. The data
2433 * must have the commit() time rather than the time it was traced,
2434 * so that all entries in the primary buffer are in timestamp order.
2435 */
2436 timestamp = dtrace_gethrtime();
2437 saddr = (uintptr_t)src->dtb_tomax;
2438 slimit = saddr + src->dtb_offset;
2439 while (saddr < slimit) {
2440 size_t size;
2441 dtrace_rechdr_t *dtrh = (dtrace_rechdr_t *)saddr;
2442
2443 if (dtrh->dtrh_epid == DTRACE_EPIDNONE) {
2444 saddr += sizeof (dtrace_epid_t);
2445 continue;
2446 }
2447 ASSERT3U(dtrh->dtrh_epid, <=, state->dts_necbs);
2448 size = state->dts_ecbs[dtrh->dtrh_epid - 1]->dte_size;
2449
2450 ASSERT3U(saddr + size, <=, slimit);
2451 ASSERT3U(size, >=, sizeof (dtrace_rechdr_t));
2452 ASSERT3U(DTRACE_RECORD_LOAD_TIMESTAMP(dtrh), ==, UINT64_MAX);
2453
2454 DTRACE_RECORD_STORE_TIMESTAMP(dtrh, timestamp);
2455
2456 saddr += size;
2457 }
2458
2459 /*
2460 * Copy the buffer across. (Note that this is a
2461 * highly subobtimal bcopy(); in the unlikely event that this becomes
2462 * a serious performance issue, a high-performance DTrace-specific
2463 * bcopy() should obviously be invented.)
2464 */
2465 daddr = (uintptr_t)dest->dtb_tomax + offs;
2466 dlimit = daddr + src->dtb_offset;
2467 saddr = (uintptr_t)src->dtb_tomax;
2468
2469 /*
2470 * First, the aligned portion.
2471 */
2472 while (dlimit - daddr >= sizeof (uint64_t)) {
2473 *((uint64_t *)daddr) = *((uint64_t *)saddr);
2474
2475 daddr += sizeof (uint64_t);
2476 saddr += sizeof (uint64_t);
2477 }
2478
2479 /*
2480 * Now any left-over bit...
2481 */
2482 while (dlimit - daddr)
2483 *((uint8_t *)daddr++) = *((uint8_t *)saddr++);
2484
2485 /*
2486 * Finally, commit the reserved space in the destination buffer.
2487 */
2488 dest->dtb_offset = offs + src->dtb_offset;
2489
2490out:
2491 /*
2492 * If we're lucky enough to be the only active CPU on this speculation
2493 * buffer, we can just set the state back to DTRACESPEC_INACTIVE.
2494 */
2495 if (current == DTRACESPEC_ACTIVE ||
2496 (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) {
2497 uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state,
2498 DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE);
2499
2500 ASSERT(rval == DTRACESPEC_COMMITTING);
2501 }
2502
2503 src->dtb_offset = 0;
2504 src->dtb_xamot_drops += src->dtb_drops;
2505 src->dtb_drops = 0;
2506}
2507
2508/*
2509 * This routine discards an active speculation. If the specified speculation
2510 * is not in a valid state to perform a discard(), this routine will silently
2511 * do nothing. The state of the specified speculation is transitioned
2512 * according to the state transition diagram outlined in <sys/dtrace_impl.h>
2513 */
2514static void
2515dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu,
2516 dtrace_specid_t which)
2517{
2518 dtrace_speculation_t *spec;
2519 dtrace_speculation_state_t current, new = 0;
2520 dtrace_buffer_t *buf;
2521
2522 if (which == 0)
2523 return;
2524
2525 if (which > state->dts_nspeculations) {
2526 cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2527 return;
2528 }
2529
2530 spec = &state->dts_speculations[which - 1];
2531 buf = &spec->dtsp_buffer[cpu];
2532
2533 do {
2534 current = spec->dtsp_state;
2535
2536 switch (current) {
2537 case DTRACESPEC_INACTIVE:
2538 case DTRACESPEC_COMMITTINGMANY:
2539 case DTRACESPEC_COMMITTING:
2540 case DTRACESPEC_DISCARDING:
2541 return;
2542
2543 case DTRACESPEC_ACTIVE:
2544 case DTRACESPEC_ACTIVEMANY:
2545 new = DTRACESPEC_DISCARDING;
2546 break;
2547
2548 case DTRACESPEC_ACTIVEONE:
2549 if (buf->dtb_offset != 0) {
2550 new = DTRACESPEC_INACTIVE;
2551 } else {
2552 new = DTRACESPEC_DISCARDING;
2553 }
2554 break;
2555
2556 default:
2557 ASSERT(0);
2558 }
2559 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2560 current, new) != current);
2561
2562 buf->dtb_offset = 0;
2563 buf->dtb_drops = 0;
2564}
2565
2566/*
2567 * Note: not called from probe context. This function is called
2568 * asynchronously from cross call context to clean any speculations that are
2569 * in the COMMITTINGMANY or DISCARDING states. These speculations may not be
2570 * transitioned back to the INACTIVE state until all CPUs have cleaned the
2571 * speculation.
2572 */
2573static void
2574dtrace_speculation_clean_here(dtrace_state_t *state)
2575{
2576 dtrace_icookie_t cookie;
2577 processorid_t cpu = curcpu;
2578 dtrace_buffer_t *dest = &state->dts_buffer[cpu];
2579 dtrace_specid_t i;
2580
2581 cookie = dtrace_interrupt_disable();
2582
2583 if (dest->dtb_tomax == NULL) {
2584 dtrace_interrupt_enable(cookie);
2585 return;
2586 }
2587
2588 for (i = 0; i < state->dts_nspeculations; i++) {
2589 dtrace_speculation_t *spec = &state->dts_speculations[i];
2590 dtrace_buffer_t *src = &spec->dtsp_buffer[cpu];
2591
2592 if (src->dtb_tomax == NULL)
2593 continue;
2594
2595 if (spec->dtsp_state == DTRACESPEC_DISCARDING) {
2596 src->dtb_offset = 0;
2597 continue;
2598 }
2599
2600 if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
2601 continue;
2602
2603 if (src->dtb_offset == 0)
2604 continue;
2605
2606 dtrace_speculation_commit(state, cpu, i + 1);
2607 }
2608
2609 dtrace_interrupt_enable(cookie);
2610}
2611
2612/*
2613 * Note: not called from probe context. This function is called
2614 * asynchronously (and at a regular interval) to clean any speculations that
2615 * are in the COMMITTINGMANY or DISCARDING states. If it discovers that there
2616 * is work to be done, it cross calls all CPUs to perform that work;
2617 * COMMITMANY and DISCARDING speculations may not be transitioned back to the
2618 * INACTIVE state until they have been cleaned by all CPUs.
2619 */
2620static void
2621dtrace_speculation_clean(dtrace_state_t *state)
2622{
2623 int work = 0, rv;
2624 dtrace_specid_t i;
2625
2626 for (i = 0; i < state->dts_nspeculations; i++) {
2627 dtrace_speculation_t *spec = &state->dts_speculations[i];
2628
2629 ASSERT(!spec->dtsp_cleaning);
2630
2631 if (spec->dtsp_state != DTRACESPEC_DISCARDING &&
2632 spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
2633 continue;
2634
2635 work++;
2636 spec->dtsp_cleaning = 1;
2637 }
2638
2639 if (!work)
2640 return;
2641
2642 dtrace_xcall(DTRACE_CPUALL,
2643 (dtrace_xcall_t)dtrace_speculation_clean_here, state);
2644
2645 /*
2646 * We now know that all CPUs have committed or discarded their
2647 * speculation buffers, as appropriate. We can now set the state
2648 * to inactive.
2649 */
2650 for (i = 0; i < state->dts_nspeculations; i++) {
2651 dtrace_speculation_t *spec = &state->dts_speculations[i];
2652 dtrace_speculation_state_t current, new;
2653
2654 if (!spec->dtsp_cleaning)
2655 continue;
2656
2657 current = spec->dtsp_state;
2658 ASSERT(current == DTRACESPEC_DISCARDING ||
2659 current == DTRACESPEC_COMMITTINGMANY);
2660
2661 new = DTRACESPEC_INACTIVE;
2662
2663 rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new);
2664 ASSERT(rv == current);
2665 spec->dtsp_cleaning = 0;
2666 }
2667}
2668
2669/*
2670 * Called as part of a speculate() to get the speculative buffer associated
2671 * with a given speculation. Returns NULL if the specified speculation is not
2672 * in an ACTIVE state. If the speculation is in the ACTIVEONE state -- and
2673 * the active CPU is not the specified CPU -- the speculation will be
2674 * atomically transitioned into the ACTIVEMANY state.
2675 */
2676static dtrace_buffer_t *
2677dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid,
2678 dtrace_specid_t which)
2679{
2680 dtrace_speculation_t *spec;
2681 dtrace_speculation_state_t current, new = 0;
2682 dtrace_buffer_t *buf;
2683
2684 if (which == 0)
2685 return (NULL);
2686
2687 if (which > state->dts_nspeculations) {
2688 cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2689 return (NULL);
2690 }
2691
2692 spec = &state->dts_speculations[which - 1];
2693 buf = &spec->dtsp_buffer[cpuid];
2694
2695 do {
2696 current = spec->dtsp_state;
2697
2698 switch (current) {
2699 case DTRACESPEC_INACTIVE:
2700 case DTRACESPEC_COMMITTINGMANY:
2701 case DTRACESPEC_DISCARDING:
2702 return (NULL);
2703
2704 case DTRACESPEC_COMMITTING:
2705 ASSERT(buf->dtb_offset == 0);
2706 return (NULL);
2707
2708 case DTRACESPEC_ACTIVEONE:
2709 /*
2710 * This speculation is currently active on one CPU.
2711 * Check the offset in the buffer; if it's non-zero,
2712 * that CPU must be us (and we leave the state alone).
2713 * If it's zero, assume that we're starting on a new
2714 * CPU -- and change the state to indicate that the
2715 * speculation is active on more than one CPU.
2716 */
2717 if (buf->dtb_offset != 0)
2718 return (buf);
2719
2720 new = DTRACESPEC_ACTIVEMANY;
2721 break;
2722
2723 case DTRACESPEC_ACTIVEMANY:
2724 return (buf);
2725
2726 case DTRACESPEC_ACTIVE:
2727 new = DTRACESPEC_ACTIVEONE;
2728 break;
2729
2730 default:
2731 ASSERT(0);
2732 }
2733 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2734 current, new) != current);
2735
2736 ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY);
2737 return (buf);
2738}
2739
2740/*
2741 * Return a string. In the event that the user lacks the privilege to access
2742 * arbitrary kernel memory, we copy the string out to scratch memory so that we
2743 * don't fail access checking.
2744 *
2745 * dtrace_dif_variable() uses this routine as a helper for various
2746 * builtin values such as 'execname' and 'probefunc.'
2747 */
2748uintptr_t
2749dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state,
2750 dtrace_mstate_t *mstate)
2751{
2752 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
2753 uintptr_t ret;
2754 size_t strsz;
2755
2756 /*
2757 * The easy case: this probe is allowed to read all of memory, so
2758 * we can just return this as a vanilla pointer.
2759 */
2760 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
2761 return (addr);
2762
2763 /*
2764 * This is the tougher case: we copy the string in question from
2765 * kernel memory into scratch memory and return it that way: this
2766 * ensures that we won't trip up when access checking tests the
2767 * BYREF return value.
2768 */
2769 strsz = dtrace_strlen((char *)addr, size) + 1;
2770
2771 if (mstate->dtms_scratch_ptr + strsz >
2772 mstate->dtms_scratch_base + mstate->dtms_scratch_size) {
2773 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
2774 return (0);
2775 }
2776
2777 dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr,
2778 strsz);
2779 ret = mstate->dtms_scratch_ptr;
2780 mstate->dtms_scratch_ptr += strsz;
2781 return (ret);
2782}
2783
2784/*
2785 * Return a string from a memoy address which is known to have one or
2786 * more concatenated, individually zero terminated, sub-strings.
2787 * In the event that the user lacks the privilege to access
2788 * arbitrary kernel memory, we copy the string out to scratch memory so that we
2789 * don't fail access checking.
2790 *
2791 * dtrace_dif_variable() uses this routine as a helper for various
2792 * builtin values such as 'execargs'.
2793 */
2794static uintptr_t
2795dtrace_dif_varstrz(uintptr_t addr, size_t strsz, dtrace_state_t *state,
2796 dtrace_mstate_t *mstate)
2797{
2798 char *p;
2799 size_t i;
2800 uintptr_t ret;
2801
2802 if (mstate->dtms_scratch_ptr + strsz >
2803 mstate->dtms_scratch_base + mstate->dtms_scratch_size) {
2804 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
2805 return (0);
2806 }
2807
2808 dtrace_bcopy((const void *)addr, (void *)mstate->dtms_scratch_ptr,
2809 strsz);
2810
2811 /* Replace sub-string termination characters with a space. */
2812 for (p = (char *) mstate->dtms_scratch_ptr, i = 0; i < strsz - 1;
2813 p++, i++)
2814 if (*p == '\0')
2815 *p = ' ';
2816
2817 ret = mstate->dtms_scratch_ptr;
2818 mstate->dtms_scratch_ptr += strsz;
2819 return (ret);
2820}
2821
2822/*
2823 * This function implements the DIF emulator's variable lookups. The emulator
2824 * passes a reserved variable identifier and optional built-in array index.
2825 */
2826static uint64_t
2827dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v,
2828 uint64_t ndx)
2829{
2830 /*
2831 * If we're accessing one of the uncached arguments, we'll turn this
2832 * into a reference in the args array.
2833 */
2834 if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) {
2835 ndx = v - DIF_VAR_ARG0;
2836 v = DIF_VAR_ARGS;
2837 }
2838
2839 switch (v) {
2840 case DIF_VAR_ARGS:
2841 ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS);
2842 if (ndx >= sizeof (mstate->dtms_arg) /
2843 sizeof (mstate->dtms_arg[0])) {
2844 int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2845 dtrace_provider_t *pv;
2846 uint64_t val;
2847
2848 pv = mstate->dtms_probe->dtpr_provider;
2849 if (pv->dtpv_pops.dtps_getargval != NULL)
2850 val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg,
2851 mstate->dtms_probe->dtpr_id,
2852 mstate->dtms_probe->dtpr_arg, ndx, aframes);
2853 else
2854 val = dtrace_getarg(ndx, aframes);
2855
2856 /*
2857 * This is regrettably required to keep the compiler
2858 * from tail-optimizing the call to dtrace_getarg().
2859 * The condition always evaluates to true, but the
2860 * compiler has no way of figuring that out a priori.
2861 * (None of this would be necessary if the compiler
2862 * could be relied upon to _always_ tail-optimize
2863 * the call to dtrace_getarg() -- but it can't.)
2864 */
2865 if (mstate->dtms_probe != NULL)
2866 return (val);
2867
2868 ASSERT(0);
2869 }
2870
2871 return (mstate->dtms_arg[ndx]);
2872
2873#if defined(sun)
2874 case DIF_VAR_UREGS: {
2875 klwp_t *lwp;
2876
2877 if (!dtrace_priv_proc(state))
2878 return (0);
2879
2880 if ((lwp = curthread->t_lwp) == NULL) {
2881 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
2882 cpu_core[curcpu].cpuc_dtrace_illval = NULL;
2883 return (0);
2884 }
2885
2886 return (dtrace_getreg(lwp->lwp_regs, ndx));
2887 return (0);
2888 }
2889#else
2890 case DIF_VAR_UREGS: {
2891 struct trapframe *tframe;
2892
2893 if (!dtrace_priv_proc(state))
2894 return (0);
2895
2896 if ((tframe = curthread->td_frame) == NULL) {
2897 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
2898 cpu_core[curcpu].cpuc_dtrace_illval = 0;
2899 return (0);
2900 }
2901
2902 return (dtrace_getreg(tframe, ndx));
2903 }
2904#endif
2905
2906 case DIF_VAR_CURTHREAD:
2907 if (!dtrace_priv_kernel(state))
2908 return (0);
2909 return ((uint64_t)(uintptr_t)curthread);
2910
2911 case DIF_VAR_TIMESTAMP:
2912 if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) {
2913 mstate->dtms_timestamp = dtrace_gethrtime();
2914 mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP;
2915 }
2916 return (mstate->dtms_timestamp);
2917
2918 case DIF_VAR_VTIMESTAMP:
2919 ASSERT(dtrace_vtime_references != 0);
2920 return (curthread->t_dtrace_vtime);
2921
2922 case DIF_VAR_WALLTIMESTAMP:
2923 if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) {
2924 mstate->dtms_walltimestamp = dtrace_gethrestime();
2925 mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP;
2926 }
2927 return (mstate->dtms_walltimestamp);
2928
2929#if defined(sun)
2930 case DIF_VAR_IPL:
2931 if (!dtrace_priv_kernel(state))
2932 return (0);
2933 if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) {
2934 mstate->dtms_ipl = dtrace_getipl();
2935 mstate->dtms_present |= DTRACE_MSTATE_IPL;
2936 }
2937 return (mstate->dtms_ipl);
2938#endif
2939
2940 case DIF_VAR_EPID:
2941 ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID);
2942 return (mstate->dtms_epid);
2943
2944 case DIF_VAR_ID:
2945 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2946 return (mstate->dtms_probe->dtpr_id);
2947
2948 case DIF_VAR_STACKDEPTH:
2949 if (!dtrace_priv_kernel(state))
2950 return (0);
2951 if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) {
2952 int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2953
2954 mstate->dtms_stackdepth = dtrace_getstackdepth(aframes);
2955 mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH;
2956 }
2957 return (mstate->dtms_stackdepth);
2958
2959 case DIF_VAR_USTACKDEPTH:
2960 if (!dtrace_priv_proc(state))
2961 return (0);
2962 if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) {
2963 /*
2964 * See comment in DIF_VAR_PID.
2965 */
2966 if (DTRACE_ANCHORED(mstate->dtms_probe) &&
2967 CPU_ON_INTR(CPU)) {
2968 mstate->dtms_ustackdepth = 0;
2969 } else {
2970 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
2971 mstate->dtms_ustackdepth =
2972 dtrace_getustackdepth();
2973 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
2974 }
2975 mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH;
2976 }
2977 return (mstate->dtms_ustackdepth);
2978
2979 case DIF_VAR_CALLER:
2980 if (!dtrace_priv_kernel(state))
2981 return (0);
2982 if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) {
2983 int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2984
2985 if (!DTRACE_ANCHORED(mstate->dtms_probe)) {
2986 /*
2987 * If this is an unanchored probe, we are
2988 * required to go through the slow path:
2989 * dtrace_caller() only guarantees correct
2990 * results for anchored probes.
2991 */
2992 pc_t caller[2] = {0, 0};
2993
2994 dtrace_getpcstack(caller, 2, aframes,
2995 (uint32_t *)(uintptr_t)mstate->dtms_arg[0]);
2996 mstate->dtms_caller = caller[1];
2997 } else if ((mstate->dtms_caller =
2998 dtrace_caller(aframes)) == -1) {
2999 /*
3000 * We have failed to do this the quick way;
3001 * we must resort to the slower approach of
3002 * calling dtrace_getpcstack().
3003 */
3004 pc_t caller = 0;
3005
3006 dtrace_getpcstack(&caller, 1, aframes, NULL);
3007 mstate->dtms_caller = caller;
3008 }
3009
3010 mstate->dtms_present |= DTRACE_MSTATE_CALLER;
3011 }
3012 return (mstate->dtms_caller);
3013
3014 case DIF_VAR_UCALLER:
3015 if (!dtrace_priv_proc(state))
3016 return (0);
3017
3018 if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) {
3019 uint64_t ustack[3];
3020
3021 /*
3022 * dtrace_getupcstack() fills in the first uint64_t
3023 * with the current PID. The second uint64_t will
3024 * be the program counter at user-level. The third
3025 * uint64_t will contain the caller, which is what
3026 * we're after.
3027 */
3028 ustack[2] = 0;
3029 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3030 dtrace_getupcstack(ustack, 3);
3031 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3032 mstate->dtms_ucaller = ustack[2];
3033 mstate->dtms_present |= DTRACE_MSTATE_UCALLER;
3034 }
3035
3036 return (mstate->dtms_ucaller);
3037
3038 case DIF_VAR_PROBEPROV:
3039 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3040 return (dtrace_dif_varstr(
3041 (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name,
3042 state, mstate));
3043
3044 case DIF_VAR_PROBEMOD:
3045 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3046 return (dtrace_dif_varstr(
3047 (uintptr_t)mstate->dtms_probe->dtpr_mod,
3048 state, mstate));
3049
3050 case DIF_VAR_PROBEFUNC:
3051 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3052 return (dtrace_dif_varstr(
3053 (uintptr_t)mstate->dtms_probe->dtpr_func,
3054 state, mstate));
3055
3056 case DIF_VAR_PROBENAME:
3057 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
3058 return (dtrace_dif_varstr(
3059 (uintptr_t)mstate->dtms_probe->dtpr_name,
3060 state, mstate));
3061
3062 case DIF_VAR_PID:
3063 if (!dtrace_priv_proc(state))
3064 return (0);
3065
3066#if defined(sun)
3067 /*
3068 * Note that we are assuming that an unanchored probe is
3069 * always due to a high-level interrupt. (And we're assuming
3070 * that there is only a single high level interrupt.)
3071 */
3072 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3073 return (pid0.pid_id);
3074
3075 /*
3076 * It is always safe to dereference one's own t_procp pointer:
3077 * it always points to a valid, allocated proc structure.
3078 * Further, it is always safe to dereference the p_pidp member
3079 * of one's own proc structure. (These are truisms becuase
3080 * threads and processes don't clean up their own state --
3081 * they leave that task to whomever reaps them.)
3082 */
3083 return ((uint64_t)curthread->t_procp->p_pidp->pid_id);
3084#else
3085 return ((uint64_t)curproc->p_pid);
3086#endif
3087
3088 case DIF_VAR_PPID:
3089 if (!dtrace_priv_proc(state))
3090 return (0);
3091
3092#if defined(sun)
3093 /*
3094 * See comment in DIF_VAR_PID.
3095 */
3096 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3097 return (pid0.pid_id);
3098
3099 /*
3100 * It is always safe to dereference one's own t_procp pointer:
3101 * it always points to a valid, allocated proc structure.
3102 * (This is true because threads don't clean up their own
3103 * state -- they leave that task to whomever reaps them.)
3104 */
3105 return ((uint64_t)curthread->t_procp->p_ppid);
3106#else
3107 return ((uint64_t)curproc->p_pptr->p_pid);
3108#endif
3109
3110 case DIF_VAR_TID:
3111#if defined(sun)
3112 /*
3113 * See comment in DIF_VAR_PID.
3114 */
3115 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3116 return (0);
3117#endif
3118
3119 return ((uint64_t)curthread->t_tid);
3120
3121 case DIF_VAR_EXECARGS: {
3122 struct pargs *p_args = curthread->td_proc->p_args;
3123
3124 if (p_args == NULL)
3125 return(0);
3126
3127 return (dtrace_dif_varstrz(
3128 (uintptr_t) p_args->ar_args, p_args->ar_length, state, mstate));
3129 }
3130
3131 case DIF_VAR_EXECNAME:
3132#if defined(sun)
3133 if (!dtrace_priv_proc(state))
3134 return (0);
3135
3136 /*
3137 * See comment in DIF_VAR_PID.
3138 */
3139 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3140 return ((uint64_t)(uintptr_t)p0.p_user.u_comm);
3141
3142 /*
3143 * It is always safe to dereference one's own t_procp pointer:
3144 * it always points to a valid, allocated proc structure.
3145 * (This is true because threads don't clean up their own
3146 * state -- they leave that task to whomever reaps them.)
3147 */
3148 return (dtrace_dif_varstr(
3149 (uintptr_t)curthread->t_procp->p_user.u_comm,
3150 state, mstate));
3151#else
3152 return (dtrace_dif_varstr(
3153 (uintptr_t) curthread->td_proc->p_comm, state, mstate));
3154#endif
3155
3156 case DIF_VAR_ZONENAME:
3157#if defined(sun)
3158 if (!dtrace_priv_proc(state))
3159 return (0);
3160
3161 /*
3162 * See comment in DIF_VAR_PID.
3163 */
3164 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3165 return ((uint64_t)(uintptr_t)p0.p_zone->zone_name);
3166
3167 /*
3168 * It is always safe to dereference one's own t_procp pointer:
3169 * it always points to a valid, allocated proc structure.
3170 * (This is true because threads don't clean up their own
3171 * state -- they leave that task to whomever reaps them.)
3172 */
3173 return (dtrace_dif_varstr(
3174 (uintptr_t)curthread->t_procp->p_zone->zone_name,
3175 state, mstate));
3176#else
3177 return (0);
3178#endif
3179
3180 case DIF_VAR_UID:
3181 if (!dtrace_priv_proc(state))
3182 return (0);
3183
3184#if defined(sun)
3185 /*
3186 * See comment in DIF_VAR_PID.
3187 */
3188 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3189 return ((uint64_t)p0.p_cred->cr_uid);
3190#endif
3191
3192 /*
3193 * It is always safe to dereference one's own t_procp pointer:
3194 * it always points to a valid, allocated proc structure.
3195 * (This is true because threads don't clean up their own
3196 * state -- they leave that task to whomever reaps them.)
3197 *
3198 * Additionally, it is safe to dereference one's own process
3199 * credential, since this is never NULL after process birth.
3200 */
3201 return ((uint64_t)curthread->t_procp->p_cred->cr_uid);
3202
3203 case DIF_VAR_GID:
3204 if (!dtrace_priv_proc(state))
3205 return (0);
3206
3207#if defined(sun)
3208 /*
3209 * See comment in DIF_VAR_PID.
3210 */
3211 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3212 return ((uint64_t)p0.p_cred->cr_gid);
3213#endif
3214
3215 /*
3216 * It is always safe to dereference one's own t_procp pointer:
3217 * it always points to a valid, allocated proc structure.
3218 * (This is true because threads don't clean up their own
3219 * state -- they leave that task to whomever reaps them.)
3220 *
3221 * Additionally, it is safe to dereference one's own process
3222 * credential, since this is never NULL after process birth.
3223 */
3224 return ((uint64_t)curthread->t_procp->p_cred->cr_gid);
3225
3226 case DIF_VAR_ERRNO: {
3227#if defined(sun)
3228 klwp_t *lwp;
3229 if (!dtrace_priv_proc(state))
3230 return (0);
3231
3232 /*
3233 * See comment in DIF_VAR_PID.
3234 */
3235 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3236 return (0);
3237
3238 /*
3239 * It is always safe to dereference one's own t_lwp pointer in
3240 * the event that this pointer is non-NULL. (This is true
3241 * because threads and lwps don't clean up their own state --
3242 * they leave that task to whomever reaps them.)
3243 */
3244 if ((lwp = curthread->t_lwp) == NULL)
3245 return (0);
3246
3247 return ((uint64_t)lwp->lwp_errno);
3248#else
3249 return (curthread->td_errno);
3250#endif
3251 }
3252#if !defined(sun)
3253 case DIF_VAR_CPU: {
3254 return curcpu;
3255 }
3256#endif
3257 default:
3258 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
3259 return (0);
3260 }
3261}
3262
3263/*
3264 * Emulate the execution of DTrace ID subroutines invoked by the call opcode.
3265 * Notice that we don't bother validating the proper number of arguments or
3266 * their types in the tuple stack. This isn't needed because all argument
3267 * interpretation is safe because of our load safety -- the worst that can
3268 * happen is that a bogus program can obtain bogus results.
3269 */
3270static void
3271dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs,
3272 dtrace_key_t *tupregs, int nargs,
3273 dtrace_mstate_t *mstate, dtrace_state_t *state)
3274{
3275 volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
3276 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
3277 dtrace_vstate_t *vstate = &state->dts_vstate;
3278
3279#if defined(sun)
3280 union {
3281 mutex_impl_t mi;
3282 uint64_t mx;
3283 } m;
3284
3285 union {
3286 krwlock_t ri;
3287 uintptr_t rw;
3288 } r;
3289#else
3290 struct thread *lowner;
3291 union {
3292 struct lock_object *li;
3293 uintptr_t lx;
3294 } l;
3295#endif
3296
3297 switch (subr) {
3298 case DIF_SUBR_RAND:
3299 regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875;
3300 break;
3301
3302#if defined(sun)
3303 case DIF_SUBR_MUTEX_OWNED:
3304 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3305 mstate, vstate)) {
3306 regs[rd] = 0;
3307 break;
3308 }
3309
3310 m.mx = dtrace_load64(tupregs[0].dttk_value);
3311 if (MUTEX_TYPE_ADAPTIVE(&m.mi))
3312 regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER;
3313 else
3314 regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock);
3315 break;
3316
3317 case DIF_SUBR_MUTEX_OWNER:
3318 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3319 mstate, vstate)) {
3320 regs[rd] = 0;
3321 break;
3322 }
3323
3324 m.mx = dtrace_load64(tupregs[0].dttk_value);
3325 if (MUTEX_TYPE_ADAPTIVE(&m.mi) &&
3326 MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER)
3327 regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi);
3328 else
3329 regs[rd] = 0;
3330 break;
3331
3332 case DIF_SUBR_MUTEX_TYPE_ADAPTIVE:
3333 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3334 mstate, vstate)) {
3335 regs[rd] = 0;
3336 break;
3337 }
3338
3339 m.mx = dtrace_load64(tupregs[0].dttk_value);
3340 regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi);
3341 break;
3342
3343 case DIF_SUBR_MUTEX_TYPE_SPIN:
3344 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3345 mstate, vstate)) {
3346 regs[rd] = 0;
3347 break;
3348 }
3349
3350 m.mx = dtrace_load64(tupregs[0].dttk_value);
3351 regs[rd] = MUTEX_TYPE_SPIN(&m.mi);
3352 break;
3353
3354 case DIF_SUBR_RW_READ_HELD: {
3355 uintptr_t tmp;
3356
3357 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3358 mstate, vstate)) {
3359 regs[rd] = 0;
3360 break;
3361 }
3362
3363 r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3364 regs[rd] = _RW_READ_HELD(&r.ri, tmp);
3365 break;
3366 }
3367
3368 case DIF_SUBR_RW_WRITE_HELD:
3369 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t),
3370 mstate, vstate)) {
3371 regs[rd] = 0;
3372 break;
3373 }
3374
3375 r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3376 regs[rd] = _RW_WRITE_HELD(&r.ri);
3377 break;
3378
3379 case DIF_SUBR_RW_ISWRITER:
3380 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t),
3381 mstate, vstate)) {
3382 regs[rd] = 0;
3383 break;
3384 }
3385
3386 r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3387 regs[rd] = _RW_ISWRITER(&r.ri);
3388 break;
3389
3390#else
3391 case DIF_SUBR_MUTEX_OWNED:
3392 if (!dtrace_canload(tupregs[0].dttk_value,
3393 sizeof (struct lock_object), mstate, vstate)) {
3394 regs[rd] = 0;
3395 break;
3396 }
3397 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3398 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3399 break;
3400
3401 case DIF_SUBR_MUTEX_OWNER:
3402 if (!dtrace_canload(tupregs[0].dttk_value,
3403 sizeof (struct lock_object), mstate, vstate)) {
3404 regs[rd] = 0;
3405 break;
3406 }
3407 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3408 LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3409 regs[rd] = (uintptr_t)lowner;
3410 break;
3411
3412 case DIF_SUBR_MUTEX_TYPE_ADAPTIVE:
3413 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (struct mtx),
3414 mstate, vstate)) {
3415 regs[rd] = 0;
3416 break;
3417 }
3418 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3419 /* XXX - should be only LC_SLEEPABLE? */
3420 regs[rd] = (LOCK_CLASS(l.li)->lc_flags &
3421 (LC_SLEEPLOCK | LC_SLEEPABLE)) != 0;
3422 break;
3423
3424 case DIF_SUBR_MUTEX_TYPE_SPIN:
3425 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (struct mtx),
3426 mstate, vstate)) {
3427 regs[rd] = 0;
3428 break;
3429 }
3430 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3431 regs[rd] = (LOCK_CLASS(l.li)->lc_flags & LC_SPINLOCK) != 0;
3432 break;
3433
3434 case DIF_SUBR_RW_READ_HELD:
3435 case DIF_SUBR_SX_SHARED_HELD:
3436 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3437 mstate, vstate)) {
3438 regs[rd] = 0;
3439 break;
3440 }
3441 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3442 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner) &&
3443 lowner == NULL;
3444 break;
3445
3446 case DIF_SUBR_RW_WRITE_HELD:
3447 case DIF_SUBR_SX_EXCLUSIVE_HELD:
3448 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3449 mstate, vstate)) {
3450 regs[rd] = 0;
3451 break;
3452 }
3453 l.lx = dtrace_loadptr(tupregs[0].dttk_value);
3454 LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3455 regs[rd] = (lowner == curthread);
3456 break;
3457
3458 case DIF_SUBR_RW_ISWRITER:
3459 case DIF_SUBR_SX_ISEXCLUSIVE:
3460 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3461 mstate, vstate)) {
3462 regs[rd] = 0;
3463 break;
3464 }
3465 l.lx = dtrace_loadptr(tupregs[0].dttk_value);
3466 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner) &&
3467 lowner != NULL;
3468 break;
3469#endif /* ! defined(sun) */
3470
3471 case DIF_SUBR_BCOPY: {
3472 /*
3473 * We need to be sure that the destination is in the scratch
3474 * region -- no other region is allowed.
3475 */
3476 uintptr_t src = tupregs[0].dttk_value;
3477 uintptr_t dest = tupregs[1].dttk_value;
3478 size_t size = tupregs[2].dttk_value;
3479
3480 if (!dtrace_inscratch(dest, size, mstate)) {
3481 *flags |= CPU_DTRACE_BADADDR;
3482 *illval = regs[rd];
3483 break;
3484 }
3485
3486 if (!dtrace_canload(src, size, mstate, vstate)) {
3487 regs[rd] = 0;
3488 break;
3489 }
3490
3491 dtrace_bcopy((void *)src, (void *)dest, size);
3492 break;
3493 }
3494
3495 case DIF_SUBR_ALLOCA:
3496 case DIF_SUBR_COPYIN: {
3497 uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
3498 uint64_t size =
3499 tupregs[subr == DIF_SUBR_ALLOCA ? 0 : 1].dttk_value;
3500 size_t scratch_size = (dest - mstate->dtms_scratch_ptr) + size;
3501
3502 /*
3503 * This action doesn't require any credential checks since
3504 * probes will not activate in user contexts to which the
3505 * enabling user does not have permissions.
3506 */
3507
3508 /*
3509 * Rounding up the user allocation size could have overflowed
3510 * a large, bogus allocation (like -1ULL) to 0.
3511 */
3512 if (scratch_size < size ||
3513 !DTRACE_INSCRATCH(mstate, scratch_size)) {
3514 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3515 regs[rd] = 0;
3516 break;
3517 }
3518
3519 if (subr == DIF_SUBR_COPYIN) {
3520 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3521 dtrace_copyin(tupregs[0].dttk_value, dest, size, flags);
3522 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3523 }
3524
3525 mstate->dtms_scratch_ptr += scratch_size;
3526 regs[rd] = dest;
3527 break;
3528 }
3529
3530 case DIF_SUBR_COPYINTO: {
3531 uint64_t size = tupregs[1].dttk_value;
3532 uintptr_t dest = tupregs[2].dttk_value;
3533
3534 /*
3535 * This action doesn't require any credential checks since
3536 * probes will not activate in user contexts to which the
3537 * enabling user does not have permissions.
3538 */
3539 if (!dtrace_inscratch(dest, size, mstate)) {
3540 *flags |= CPU_DTRACE_BADADDR;
3541 *illval = regs[rd];
3542 break;
3543 }
3544
3545 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3546 dtrace_copyin(tupregs[0].dttk_value, dest, size, flags);
3547 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3548 break;
3549 }
3550
3551 case DIF_SUBR_COPYINSTR: {
3552 uintptr_t dest = mstate->dtms_scratch_ptr;
3553 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3554
3555 if (nargs > 1 && tupregs[1].dttk_value < size)
3556 size = tupregs[1].dttk_value + 1;
3557
3558 /*
3559 * This action doesn't require any credential checks since
3560 * probes will not activate in user contexts to which the
3561 * enabling user does not have permissions.
3562 */
3563 if (!DTRACE_INSCRATCH(mstate, size)) {
3564 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3565 regs[rd] = 0;
3566 break;
3567 }
3568
3569 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3570 dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags);
3571 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3572
3573 ((char *)dest)[size - 1] = '\0';
3574 mstate->dtms_scratch_ptr += size;
3575 regs[rd] = dest;
3576 break;
3577 }
3578
3579#if defined(sun)
3580 case DIF_SUBR_MSGSIZE:
3581 case DIF_SUBR_MSGDSIZE: {
3582 uintptr_t baddr = tupregs[0].dttk_value, daddr;
3583 uintptr_t wptr, rptr;
3584 size_t count = 0;
3585 int cont = 0;
3586
3587 while (baddr != 0 && !(*flags & CPU_DTRACE_FAULT)) {
3588
3589 if (!dtrace_canload(baddr, sizeof (mblk_t), mstate,
3590 vstate)) {
3591 regs[rd] = 0;
3592 break;
3593 }
3594
3595 wptr = dtrace_loadptr(baddr +
3596 offsetof(mblk_t, b_wptr));
3597
3598 rptr = dtrace_loadptr(baddr +
3599 offsetof(mblk_t, b_rptr));
3600
3601 if (wptr < rptr) {
3602 *flags |= CPU_DTRACE_BADADDR;
3603 *illval = tupregs[0].dttk_value;
3604 break;
3605 }
3606
3607 daddr = dtrace_loadptr(baddr +
3608 offsetof(mblk_t, b_datap));
3609
3610 baddr = dtrace_loadptr(baddr +
3611 offsetof(mblk_t, b_cont));
3612
3613 /*
3614 * We want to prevent against denial-of-service here,
3615 * so we're only going to search the list for
3616 * dtrace_msgdsize_max mblks.
3617 */
3618 if (cont++ > dtrace_msgdsize_max) {
3619 *flags |= CPU_DTRACE_ILLOP;
3620 break;
3621 }
3622
3623 if (subr == DIF_SUBR_MSGDSIZE) {
3624 if (dtrace_load8(daddr +
3625 offsetof(dblk_t, db_type)) != M_DATA)
3626 continue;
3627 }
3628
3629 count += wptr - rptr;
3630 }
3631
3632 if (!(*flags & CPU_DTRACE_FAULT))
3633 regs[rd] = count;
3634
3635 break;
3636 }
3637#endif
3638
3639 case DIF_SUBR_PROGENYOF: {
3640 pid_t pid = tupregs[0].dttk_value;
3641 proc_t *p;
3642 int rval = 0;
3643
3644 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3645
3646 for (p = curthread->t_procp; p != NULL; p = p->p_parent) {
3647#if defined(sun)
3648 if (p->p_pidp->pid_id == pid) {
3649#else
3650 if (p->p_pid == pid) {
3651#endif
3652 rval = 1;
3653 break;
3654 }
3655 }
3656
3657 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3658
3659 regs[rd] = rval;
3660 break;
3661 }
3662
3663 case DIF_SUBR_SPECULATION:
3664 regs[rd] = dtrace_speculation(state);
3665 break;
3666
3667 case DIF_SUBR_COPYOUT: {
3668 uintptr_t kaddr = tupregs[0].dttk_value;
3669 uintptr_t uaddr = tupregs[1].dttk_value;
3670 uint64_t size = tupregs[2].dttk_value;
3671
3672 if (!dtrace_destructive_disallow &&
3673 dtrace_priv_proc_control(state) &&
3674 !dtrace_istoxic(kaddr, size)) {
3675 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3676 dtrace_copyout(kaddr, uaddr, size, flags);
3677 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3678 }
3679 break;
3680 }
3681
3682 case DIF_SUBR_COPYOUTSTR: {
3683 uintptr_t kaddr = tupregs[0].dttk_value;
3684 uintptr_t uaddr = tupregs[1].dttk_value;
3685 uint64_t size = tupregs[2].dttk_value;
3686
3687 if (!dtrace_destructive_disallow &&
3688 dtrace_priv_proc_control(state) &&
3689 !dtrace_istoxic(kaddr, size)) {
3690 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3691 dtrace_copyoutstr(kaddr, uaddr, size, flags);
3692 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3693 }
3694 break;
3695 }
3696
3697 case DIF_SUBR_STRLEN: {
3698 size_t sz;
3699 uintptr_t addr = (uintptr_t)tupregs[0].dttk_value;
3700 sz = dtrace_strlen((char *)addr,
3701 state->dts_options[DTRACEOPT_STRSIZE]);
3702
3703 if (!dtrace_canload(addr, sz + 1, mstate, vstate)) {
3704 regs[rd] = 0;
3705 break;
3706 }
3707
3708 regs[rd] = sz;
3709
3710 break;
3711 }
3712
3713 case DIF_SUBR_STRCHR:
3714 case DIF_SUBR_STRRCHR: {
3715 /*
3716 * We're going to iterate over the string looking for the
3717 * specified character. We will iterate until we have reached
3718 * the string length or we have found the character. If this
3719 * is DIF_SUBR_STRRCHR, we will look for the last occurrence
3720 * of the specified character instead of the first.
3721 */
3722 uintptr_t saddr = tupregs[0].dttk_value;
3723 uintptr_t addr = tupregs[0].dttk_value;
3724 uintptr_t limit = addr + state->dts_options[DTRACEOPT_STRSIZE];
3725 char c, target = (char)tupregs[1].dttk_value;
3726
3727 for (regs[rd] = 0; addr < limit; addr++) {
3728 if ((c = dtrace_load8(addr)) == target) {
3729 regs[rd] = addr;
3730
3731 if (subr == DIF_SUBR_STRCHR)
3732 break;
3733 }
3734
3735 if (c == '\0')
3736 break;
3737 }
3738
3739 if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) {
3740 regs[rd] = 0;
3741 break;
3742 }
3743
3744 break;
3745 }
3746
3747 case DIF_SUBR_STRSTR:
3748 case DIF_SUBR_INDEX:
3749 case DIF_SUBR_RINDEX: {
3750 /*
3751 * We're going to iterate over the string looking for the
3752 * specified string. We will iterate until we have reached
3753 * the string length or we have found the string. (Yes, this
3754 * is done in the most naive way possible -- but considering
3755 * that the string we're searching for is likely to be
3756 * relatively short, the complexity of Rabin-Karp or similar
3757 * hardly seems merited.)
3758 */
3759 char *addr = (char *)(uintptr_t)tupregs[0].dttk_value;
3760 char *substr = (char *)(uintptr_t)tupregs[1].dttk_value;
3761 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3762 size_t len = dtrace_strlen(addr, size);
3763 size_t sublen = dtrace_strlen(substr, size);
3764 char *limit = addr + len, *orig = addr;
3765 int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1;
3766 int inc = 1;
3767
3768 regs[rd] = notfound;
3769
3770 if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) {
3771 regs[rd] = 0;
3772 break;
3773 }
3774
3775 if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate,
3776 vstate)) {
3777 regs[rd] = 0;
3778 break;
3779 }
3780
3781 /*
3782 * strstr() and index()/rindex() have similar semantics if
3783 * both strings are the empty string: strstr() returns a
3784 * pointer to the (empty) string, and index() and rindex()
3785 * both return index 0 (regardless of any position argument).
3786 */
3787 if (sublen == 0 && len == 0) {
3788 if (subr == DIF_SUBR_STRSTR)
3789 regs[rd] = (uintptr_t)addr;
3790 else
3791 regs[rd] = 0;
3792 break;
3793 }
3794
3795 if (subr != DIF_SUBR_STRSTR) {
3796 if (subr == DIF_SUBR_RINDEX) {
3797 limit = orig - 1;
3798 addr += len;
3799 inc = -1;
3800 }
3801
3802 /*
3803 * Both index() and rindex() take an optional position
3804 * argument that denotes the starting position.
3805 */
3806 if (nargs == 3) {
3807 int64_t pos = (int64_t)tupregs[2].dttk_value;
3808
3809 /*
3810 * If the position argument to index() is
3811 * negative, Perl implicitly clamps it at
3812 * zero. This semantic is a little surprising
3813 * given the special meaning of negative
3814 * positions to similar Perl functions like
3815 * substr(), but it appears to reflect a
3816 * notion that index() can start from a
3817 * negative index and increment its way up to
3818 * the string. Given this notion, Perl's
3819 * rindex() is at least self-consistent in
3820 * that it implicitly clamps positions greater
3821 * than the string length to be the string
3822 * length. Where Perl completely loses
3823 * coherence, however, is when the specified
3824 * substring is the empty string (""). In
3825 * this case, even if the position is
3826 * negative, rindex() returns 0 -- and even if
3827 * the position is greater than the length,
3828 * index() returns the string length. These
3829 * semantics violate the notion that index()
3830 * should never return a value less than the
3831 * specified position and that rindex() should
3832 * never return a value greater than the
3833 * specified position. (One assumes that
3834 * these semantics are artifacts of Perl's
3835 * implementation and not the results of
3836 * deliberate design -- it beggars belief that
3837 * even Larry Wall could desire such oddness.)
3838 * While in the abstract one would wish for
3839 * consistent position semantics across
3840 * substr(), index() and rindex() -- or at the
3841 * very least self-consistent position
3842 * semantics for index() and rindex() -- we
3843 * instead opt to keep with the extant Perl
3844 * semantics, in all their broken glory. (Do
3845 * we have more desire to maintain Perl's
3846 * semantics than Perl does? Probably.)
3847 */
3848 if (subr == DIF_SUBR_RINDEX) {
3849 if (pos < 0) {
3850 if (sublen == 0)
3851 regs[rd] = 0;
3852 break;
3853 }
3854
3855 if (pos > len)
3856 pos = len;
3857 } else {
3858 if (pos < 0)
3859 pos = 0;
3860
3861 if (pos >= len) {
3862 if (sublen == 0)
3863 regs[rd] = len;
3864 break;
3865 }
3866 }
3867
3868 addr = orig + pos;
3869 }
3870 }
3871
3872 for (regs[rd] = notfound; addr != limit; addr += inc) {
3873 if (dtrace_strncmp(addr, substr, sublen) == 0) {
3874 if (subr != DIF_SUBR_STRSTR) {
3875 /*
3876 * As D index() and rindex() are
3877 * modeled on Perl (and not on awk),
3878 * we return a zero-based (and not a
3879 * one-based) index. (For you Perl
3880 * weenies: no, we're not going to add
3881 * $[ -- and shouldn't you be at a con
3882 * or something?)
3883 */
3884 regs[rd] = (uintptr_t)(addr - orig);
3885 break;
3886 }
3887
3888 ASSERT(subr == DIF_SUBR_STRSTR);
3889 regs[rd] = (uintptr_t)addr;
3890 break;
3891 }
3892 }
3893
3894 break;
3895 }
3896
3897 case DIF_SUBR_STRTOK: {
3898 uintptr_t addr = tupregs[0].dttk_value;
3899 uintptr_t tokaddr = tupregs[1].dttk_value;
3900 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3901 uintptr_t limit, toklimit = tokaddr + size;
3902 uint8_t c = 0, tokmap[32]; /* 256 / 8 */
3903 char *dest = (char *)mstate->dtms_scratch_ptr;
3904 int i;
3905
3906 /*
3907 * Check both the token buffer and (later) the input buffer,
3908 * since both could be non-scratch addresses.
3909 */
3910 if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) {
3911 regs[rd] = 0;
3912 break;
3913 }
3914
3915 if (!DTRACE_INSCRATCH(mstate, size)) {
3916 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3917 regs[rd] = 0;
3918 break;
3919 }
3920
3921 if (addr == 0) {
3922 /*
3923 * If the address specified is NULL, we use our saved
3924 * strtok pointer from the mstate. Note that this
3925 * means that the saved strtok pointer is _only_
3926 * valid within multiple enablings of the same probe --
3927 * it behaves like an implicit clause-local variable.
3928 */
3929 addr = mstate->dtms_strtok;
3930 } else {
3931 /*
3932 * If the user-specified address is non-NULL we must
3933 * access check it. This is the only time we have
3934 * a chance to do so, since this address may reside
3935 * in the string table of this clause-- future calls
3936 * (when we fetch addr from mstate->dtms_strtok)
3937 * would fail this access check.
3938 */
3939 if (!dtrace_strcanload(addr, size, mstate, vstate)) {
3940 regs[rd] = 0;
3941 break;
3942 }
3943 }
3944
3945 /*
3946 * First, zero the token map, and then process the token
3947 * string -- setting a bit in the map for every character
3948 * found in the token string.
3949 */
3950 for (i = 0; i < sizeof (tokmap); i++)
3951 tokmap[i] = 0;
3952
3953 for (; tokaddr < toklimit; tokaddr++) {
3954 if ((c = dtrace_load8(tokaddr)) == '\0')
3955 break;
3956
3957 ASSERT((c >> 3) < sizeof (tokmap));
3958 tokmap[c >> 3] |= (1 << (c & 0x7));
3959 }
3960
3961 for (limit = addr + size; addr < limit; addr++) {
3962 /*
3963 * We're looking for a character that is _not_ contained
3964 * in the token string.
3965 */
3966 if ((c = dtrace_load8(addr)) == '\0')
3967 break;
3968
3969 if (!(tokmap[c >> 3] & (1 << (c & 0x7))))
3970 break;
3971 }
3972
3973 if (c == '\0') {
3974 /*
3975 * We reached the end of the string without finding
3976 * any character that was not in the token string.
3977 * We return NULL in this case, and we set the saved
3978 * address to NULL as well.
3979 */
3980 regs[rd] = 0;
3981 mstate->dtms_strtok = 0;
3982 break;
3983 }
3984
3985 /*
3986 * From here on, we're copying into the destination string.
3987 */
3988 for (i = 0; addr < limit && i < size - 1; addr++) {
3989 if ((c = dtrace_load8(addr)) == '\0')
3990 break;
3991
3992 if (tokmap[c >> 3] & (1 << (c & 0x7)))
3993 break;
3994
3995 ASSERT(i < size);
3996 dest[i++] = c;
3997 }
3998
3999 ASSERT(i < size);
4000 dest[i] = '\0';
4001 regs[rd] = (uintptr_t)dest;
4002 mstate->dtms_scratch_ptr += size;
4003 mstate->dtms_strtok = addr;
4004 break;
4005 }
4006
4007 case DIF_SUBR_SUBSTR: {
4008 uintptr_t s = tupregs[0].dttk_value;
4009 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4010 char *d = (char *)mstate->dtms_scratch_ptr;
4011 int64_t index = (int64_t)tupregs[1].dttk_value;
4012 int64_t remaining = (int64_t)tupregs[2].dttk_value;
4013 size_t len = dtrace_strlen((char *)s, size);
4014 int64_t i = 0;
4015
4016 if (!dtrace_canload(s, len + 1, mstate, vstate)) {
4017 regs[rd] = 0;
4018 break;
4019 }
4020
4021 if (!DTRACE_INSCRATCH(mstate, size)) {
4022 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4023 regs[rd] = 0;
4024 break;
4025 }
4026
4027 if (nargs <= 2)
4028 remaining = (int64_t)size;
4029
4030 if (index < 0) {
4031 index += len;
4032
4033 if (index < 0 && index + remaining > 0) {
4034 remaining += index;
4035 index = 0;
4036 }
4037 }
4038
4039 if (index >= len || index < 0) {
4040 remaining = 0;
4041 } else if (remaining < 0) {
4042 remaining += len - index;
4043 } else if (index + remaining > size) {
4044 remaining = size - index;
4045 }
4046
4047 for (i = 0; i < remaining; i++) {
4048 if ((d[i] = dtrace_load8(s + index + i)) == '\0')
4049 break;
4050 }
4051
4052 d[i] = '\0';
4053
4054 mstate->dtms_scratch_ptr += size;
4055 regs[rd] = (uintptr_t)d;
4056 break;
4057 }
4058
4059 case DIF_SUBR_TOUPPER:
4060 case DIF_SUBR_TOLOWER: {
4061 uintptr_t s = tupregs[0].dttk_value;
4062 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4063 char *dest = (char *)mstate->dtms_scratch_ptr, c;
4064 size_t len = dtrace_strlen((char *)s, size);
4065 char lower, upper, convert;
4066 int64_t i;
4067
4068 if (subr == DIF_SUBR_TOUPPER) {
4069 lower = 'a';
4070 upper = 'z';
4071 convert = 'A';
4072 } else {
4073 lower = 'A';
4074 upper = 'Z';
4075 convert = 'a';
4076 }
4077
4078 if (!dtrace_canload(s, len + 1, mstate, vstate)) {
4079 regs[rd] = 0;
4080 break;
4081 }
4082
4083 if (!DTRACE_INSCRATCH(mstate, size)) {
4084 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4085 regs[rd] = 0;
4086 break;
4087 }
4088
4089 for (i = 0; i < size - 1; i++) {
4090 if ((c = dtrace_load8(s + i)) == '\0')
4091 break;
4092
4093 if (c >= lower && c <= upper)
4094 c = convert + (c - lower);
4095
4096 dest[i] = c;
4097 }
4098
4099 ASSERT(i < size);
4100 dest[i] = '\0';
4101 regs[rd] = (uintptr_t)dest;
4102 mstate->dtms_scratch_ptr += size;
4103 break;
4104 }
4105
4106#if defined(sun)
4107 case DIF_SUBR_GETMAJOR:
4108#ifdef _LP64
4109 regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR64) & MAXMAJ64;
4110#else
4111 regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR) & MAXMAJ;
4112#endif
4113 break;
4114
4115 case DIF_SUBR_GETMINOR:
4116#ifdef _LP64
4117 regs[rd] = tupregs[0].dttk_value & MAXMIN64;
4118#else
4119 regs[rd] = tupregs[0].dttk_value & MAXMIN;
4120#endif
4121 break;
4122
4123 case DIF_SUBR_DDI_PATHNAME: {
4124 /*
4125 * This one is a galactic mess. We are going to roughly
4126 * emulate ddi_pathname(), but it's made more complicated
4127 * by the fact that we (a) want to include the minor name and
4128 * (b) must proceed iteratively instead of recursively.
4129 */
4130 uintptr_t dest = mstate->dtms_scratch_ptr;
4131 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4132 char *start = (char *)dest, *end = start + size - 1;
4133 uintptr_t daddr = tupregs[0].dttk_value;
4134 int64_t minor = (int64_t)tupregs[1].dttk_value;
4135 char *s;
4136 int i, len, depth = 0;
4137
4138 /*
4139 * Due to all the pointer jumping we do and context we must
4140 * rely upon, we just mandate that the user must have kernel
4141 * read privileges to use this routine.
4142 */
4143 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) {
4144 *flags |= CPU_DTRACE_KPRIV;
4145 *illval = daddr;
4146 regs[rd] = 0;
4147 }
4148
4149 if (!DTRACE_INSCRATCH(mstate, size)) {
4150 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4151 regs[rd] = 0;
4152 break;
4153 }
4154
4155 *end = '\0';
4156
4157 /*
4158 * We want to have a name for the minor. In order to do this,
4159 * we need to walk the minor list from the devinfo. We want
4160 * to be sure that we don't infinitely walk a circular list,
4161 * so we check for circularity by sending a scout pointer
4162 * ahead two elements for every element that we iterate over;
4163 * if the list is circular, these will ultimately point to the
4164 * same element. You may recognize this little trick as the
4165 * answer to a stupid interview question -- one that always
4166 * seems to be asked by those who had to have it laboriously
4167 * explained to them, and who can't even concisely describe
4168 * the conditions under which one would be forced to resort to
4169 * this technique. Needless to say, those conditions are
4170 * found here -- and probably only here. Is this the only use
4171 * of this infamous trick in shipping, production code? If it
4172 * isn't, it probably should be...
4173 */
4174 if (minor != -1) {
4175 uintptr_t maddr = dtrace_loadptr(daddr +
4176 offsetof(struct dev_info, devi_minor));
4177
4178 uintptr_t next = offsetof(struct ddi_minor_data, next);
4179 uintptr_t name = offsetof(struct ddi_minor_data,
4180 d_minor) + offsetof(struct ddi_minor, name);
4181 uintptr_t dev = offsetof(struct ddi_minor_data,
4182 d_minor) + offsetof(struct ddi_minor, dev);
4183 uintptr_t scout;
4184
4185 if (maddr != NULL)
4186 scout = dtrace_loadptr(maddr + next);
4187
4188 while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4189 uint64_t m;
4190#ifdef _LP64
4191 m = dtrace_load64(maddr + dev) & MAXMIN64;
4192#else
4193 m = dtrace_load32(maddr + dev) & MAXMIN;
4194#endif
4195 if (m != minor) {
4196 maddr = dtrace_loadptr(maddr + next);
4197
4198 if (scout == NULL)
4199 continue;
4200
4201 scout = dtrace_loadptr(scout + next);
4202
4203 if (scout == NULL)
4204 continue;
4205
4206 scout = dtrace_loadptr(scout + next);
4207
4208 if (scout == NULL)
4209 continue;
4210
4211 if (scout == maddr) {
4212 *flags |= CPU_DTRACE_ILLOP;
4213 break;
4214 }
4215
4216 continue;
4217 }
4218
4219 /*
4220 * We have the minor data. Now we need to
4221 * copy the minor's name into the end of the
4222 * pathname.
4223 */
4224 s = (char *)dtrace_loadptr(maddr + name);
4225 len = dtrace_strlen(s, size);
4226
4227 if (*flags & CPU_DTRACE_FAULT)
4228 break;
4229
4230 if (len != 0) {
4231 if ((end -= (len + 1)) < start)
4232 break;
4233
4234 *end = ':';
4235 }
4236
4237 for (i = 1; i <= len; i++)
4238 end[i] = dtrace_load8((uintptr_t)s++);
4239 break;
4240 }
4241 }
4242
4243 while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4244 ddi_node_state_t devi_state;
4245
4246 devi_state = dtrace_load32(daddr +
4247 offsetof(struct dev_info, devi_node_state));
4248
4249 if (*flags & CPU_DTRACE_FAULT)
4250 break;
4251
4252 if (devi_state >= DS_INITIALIZED) {
4253 s = (char *)dtrace_loadptr(daddr +
4254 offsetof(struct dev_info, devi_addr));
4255 len = dtrace_strlen(s, size);
4256
4257 if (*flags & CPU_DTRACE_FAULT)
4258 break;
4259
4260 if (len != 0) {
4261 if ((end -= (len + 1)) < start)
4262 break;
4263
4264 *end = '@';
4265 }
4266
4267 for (i = 1; i <= len; i++)
4268 end[i] = dtrace_load8((uintptr_t)s++);
4269 }
4270
4271 /*
4272 * Now for the node name...
4273 */
4274 s = (char *)dtrace_loadptr(daddr +
4275 offsetof(struct dev_info, devi_node_name));
4276
4277 daddr = dtrace_loadptr(daddr +
4278 offsetof(struct dev_info, devi_parent));
4279
4280 /*
4281 * If our parent is NULL (that is, if we're the root
4282 * node), we're going to use the special path
4283 * "devices".
4284 */
4285 if (daddr == 0)
4286 s = "devices";
4287
4288 len = dtrace_strlen(s, size);
4289 if (*flags & CPU_DTRACE_FAULT)
4290 break;
4291
4292 if ((end -= (len + 1)) < start)
4293 break;
4294
4295 for (i = 1; i <= len; i++)
4296 end[i] = dtrace_load8((uintptr_t)s++);
4297 *end = '/';
4298
4299 if (depth++ > dtrace_devdepth_max) {
4300 *flags |= CPU_DTRACE_ILLOP;
4301 break;
4302 }
4303 }
4304
4305 if (end < start)
4306 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4307
4308 if (daddr == 0) {
4309 regs[rd] = (uintptr_t)end;
4310 mstate->dtms_scratch_ptr += size;
4311 }
4312
4313 break;
4314 }
4315#endif
4316
4317 case DIF_SUBR_STRJOIN: {
4318 char *d = (char *)mstate->dtms_scratch_ptr;
4319 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4320 uintptr_t s1 = tupregs[0].dttk_value;
4321 uintptr_t s2 = tupregs[1].dttk_value;
4322 int i = 0;
4323
4324 if (!dtrace_strcanload(s1, size, mstate, vstate) ||
4325 !dtrace_strcanload(s2, size, mstate, vstate)) {
4326 regs[rd] = 0;
4327 break;
4328 }
4329
4330 if (!DTRACE_INSCRATCH(mstate, size)) {
4331 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4332 regs[rd] = 0;
4333 break;
4334 }
4335
4336 for (;;) {
4337 if (i >= size) {
4338 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4339 regs[rd] = 0;
4340 break;
4341 }
4342
4343 if ((d[i++] = dtrace_load8(s1++)) == '\0') {
4344 i--;
4345 break;
4346 }
4347 }
4348
4349 for (;;) {
4350 if (i >= size) {
4351 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4352 regs[rd] = 0;
4353 break;
4354 }
4355
4356 if ((d[i++] = dtrace_load8(s2++)) == '\0')
4357 break;
4358 }
4359
4360 if (i < size) {
4361 mstate->dtms_scratch_ptr += i;
4362 regs[rd] = (uintptr_t)d;
4363 }
4364
4365 break;
4366 }
4367
4368 case DIF_SUBR_LLTOSTR: {
4369 int64_t i = (int64_t)tupregs[0].dttk_value;
4370 uint64_t val, digit;
4371 uint64_t size = 65; /* enough room for 2^64 in binary */
4372 char *end = (char *)mstate->dtms_scratch_ptr + size - 1;
4373 int base = 10;
4374
4375 if (nargs > 1) {
4376 if ((base = tupregs[1].dttk_value) <= 1 ||
4377 base > ('z' - 'a' + 1) + ('9' - '0' + 1)) {
4378 *flags |= CPU_DTRACE_ILLOP;
4379 break;
4380 }
4381 }
4382
4383 val = (base == 10 && i < 0) ? i * -1 : i;
4384
4385 if (!DTRACE_INSCRATCH(mstate, size)) {
4386 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4387 regs[rd] = 0;
4388 break;
4389 }
4390
4391 for (*end-- = '\0'; val; val /= base) {
4392 if ((digit = val % base) <= '9' - '0') {
4393 *end-- = '0' + digit;
4394 } else {
4395 *end-- = 'a' + (digit - ('9' - '0') - 1);
4396 }
4397 }
4398
4399 if (i == 0 && base == 16)
4400 *end-- = '0';
4401
4402 if (base == 16)
4403 *end-- = 'x';
4404
4405 if (i == 0 || base == 8 || base == 16)
4406 *end-- = '0';
4407
4408 if (i < 0 && base == 10)
4409 *end-- = '-';
4410
4411 regs[rd] = (uintptr_t)end + 1;
4412 mstate->dtms_scratch_ptr += size;
4413 break;
4414 }
4415
4416 case DIF_SUBR_HTONS:
4417 case DIF_SUBR_NTOHS:
4418#if BYTE_ORDER == BIG_ENDIAN
4419 regs[rd] = (uint16_t)tupregs[0].dttk_value;
4420#else
4421 regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value);
4422#endif
4423 break;
4424
4425
4426 case DIF_SUBR_HTONL:
4427 case DIF_SUBR_NTOHL:
4428#if BYTE_ORDER == BIG_ENDIAN
4429 regs[rd] = (uint32_t)tupregs[0].dttk_value;
4430#else
4431 regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value);
4432#endif
4433 break;
4434
4435
4436 case DIF_SUBR_HTONLL:
4437 case DIF_SUBR_NTOHLL:
4438#if BYTE_ORDER == BIG_ENDIAN
4439 regs[rd] = (uint64_t)tupregs[0].dttk_value;
4440#else
4441 regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value);
4442#endif
4443 break;
4444
4445
4446 case DIF_SUBR_DIRNAME:
4447 case DIF_SUBR_BASENAME: {
4448 char *dest = (char *)mstate->dtms_scratch_ptr;
4449 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4450 uintptr_t src = tupregs[0].dttk_value;
4451 int i, j, len = dtrace_strlen((char *)src, size);
4452 int lastbase = -1, firstbase = -1, lastdir = -1;
4453 int start, end;
4454
4455 if (!dtrace_canload(src, len + 1, mstate, vstate)) {
4456 regs[rd] = 0;
4457 break;
4458 }
4459
4460 if (!DTRACE_INSCRATCH(mstate, size)) {
4461 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4462 regs[rd] = 0;
4463 break;
4464 }
4465
4466 /*
4467 * The basename and dirname for a zero-length string is
4468 * defined to be "."
4469 */
4470 if (len == 0) {
4471 len = 1;
4472 src = (uintptr_t)".";
4473 }
4474
4475 /*
4476 * Start from the back of the string, moving back toward the
4477 * front until we see a character that isn't a slash. That
4478 * character is the last character in the basename.
4479 */
4480 for (i = len - 1; i >= 0; i--) {
4481 if (dtrace_load8(src + i) != '/')
4482 break;
4483 }
4484
4485 if (i >= 0)
4486 lastbase = i;
4487
4488 /*
4489 * Starting from the last character in the basename, move
4490 * towards the front until we find a slash. The character
4491 * that we processed immediately before that is the first
4492 * character in the basename.
4493 */
4494 for (; i >= 0; i--) {
4495 if (dtrace_load8(src + i) == '/')
4496 break;
4497 }
4498
4499 if (i >= 0)
4500 firstbase = i + 1;
4501
4502 /*
4503 * Now keep going until we find a non-slash character. That
4504 * character is the last character in the dirname.
4505 */
4506 for (; i >= 0; i--) {
4507 if (dtrace_load8(src + i) != '/')
4508 break;
4509 }
4510
4511 if (i >= 0)
4512 lastdir = i;
4513
4514 ASSERT(!(lastbase == -1 && firstbase != -1));
4515 ASSERT(!(firstbase == -1 && lastdir != -1));
4516
4517 if (lastbase == -1) {
4518 /*
4519 * We didn't find a non-slash character. We know that
4520 * the length is non-zero, so the whole string must be
4521 * slashes. In either the dirname or the basename
4522 * case, we return '/'.
4523 */
4524 ASSERT(firstbase == -1);
4525 firstbase = lastbase = lastdir = 0;
4526 }
4527
4528 if (firstbase == -1) {
4529 /*
4530 * The entire string consists only of a basename
4531 * component. If we're looking for dirname, we need
4532 * to change our string to be just "."; if we're
4533 * looking for a basename, we'll just set the first
4534 * character of the basename to be 0.
4535 */
4536 if (subr == DIF_SUBR_DIRNAME) {
4537 ASSERT(lastdir == -1);
4538 src = (uintptr_t)".";
4539 lastdir = 0;
4540 } else {
4541 firstbase = 0;
4542 }
4543 }
4544
4545 if (subr == DIF_SUBR_DIRNAME) {
4546 if (lastdir == -1) {
4547 /*
4548 * We know that we have a slash in the name --
4549 * or lastdir would be set to 0, above. And
4550 * because lastdir is -1, we know that this
4551 * slash must be the first character. (That
4552 * is, the full string must be of the form
4553 * "/basename".) In this case, the last
4554 * character of the directory name is 0.
4555 */
4556 lastdir = 0;
4557 }
4558
4559 start = 0;
4560 end = lastdir;
4561 } else {
4562 ASSERT(subr == DIF_SUBR_BASENAME);
4563 ASSERT(firstbase != -1 && lastbase != -1);
4564 start = firstbase;
4565 end = lastbase;
4566 }
4567
4568 for (i = start, j = 0; i <= end && j < size - 1; i++, j++)
4569 dest[j] = dtrace_load8(src + i);
4570
4571 dest[j] = '\0';
4572 regs[rd] = (uintptr_t)dest;
4573 mstate->dtms_scratch_ptr += size;
4574 break;
4575 }
4576
4577 case DIF_SUBR_CLEANPATH: {
4578 char *dest = (char *)mstate->dtms_scratch_ptr, c;
4579 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4580 uintptr_t src = tupregs[0].dttk_value;
4581 int i = 0, j = 0;
4582
4583 if (!dtrace_strcanload(src, size, mstate, vstate)) {
4584 regs[rd] = 0;
4585 break;
4586 }
4587
4588 if (!DTRACE_INSCRATCH(mstate, size)) {
4589 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4590 regs[rd] = 0;
4591 break;
4592 }
4593
4594 /*
4595 * Move forward, loading each character.
4596 */
4597 do {
4598 c = dtrace_load8(src + i++);
4599next:
4600 if (j + 5 >= size) /* 5 = strlen("/..c\0") */
4601 break;
4602
4603 if (c != '/') {
4604 dest[j++] = c;
4605 continue;
4606 }
4607
4608 c = dtrace_load8(src + i++);
4609
4610 if (c == '/') {
4611 /*
4612 * We have two slashes -- we can just advance
4613 * to the next character.
4614 */
4615 goto next;
4616 }
4617
4618 if (c != '.') {
4619 /*
4620 * This is not "." and it's not ".." -- we can
4621 * just store the "/" and this character and
4622 * drive on.
4623 */
4624 dest[j++] = '/';
4625 dest[j++] = c;
4626 continue;
4627 }
4628
4629 c = dtrace_load8(src + i++);
4630
4631 if (c == '/') {
4632 /*
4633 * This is a "/./" component. We're not going
4634 * to store anything in the destination buffer;
4635 * we're just going to go to the next component.
4636 */
4637 goto next;
4638 }
4639
4640 if (c != '.') {
4641 /*
4642 * This is not ".." -- we can just store the
4643 * "/." and this character and continue
4644 * processing.
4645 */
4646 dest[j++] = '/';
4647 dest[j++] = '.';
4648 dest[j++] = c;
4649 continue;
4650 }
4651
4652 c = dtrace_load8(src + i++);
4653
4654 if (c != '/' && c != '\0') {
4655 /*
4656 * This is not ".." -- it's "..[mumble]".
4657 * We'll store the "/.." and this character
4658 * and continue processing.
4659 */
4660 dest[j++] = '/';
4661 dest[j++] = '.';
4662 dest[j++] = '.';
4663 dest[j++] = c;
4664 continue;
4665 }
4666
4667 /*
4668 * This is "/../" or "/..\0". We need to back up
4669 * our destination pointer until we find a "/".
4670 */
4671 i--;
4672 while (j != 0 && dest[--j] != '/')
4673 continue;
4674
4675 if (c == '\0')
4676 dest[++j] = '/';
4677 } while (c != '\0');
4678
4679 dest[j] = '\0';
4680 regs[rd] = (uintptr_t)dest;
4681 mstate->dtms_scratch_ptr += size;
4682 break;
4683 }
4684
4685 case DIF_SUBR_INET_NTOA:
4686 case DIF_SUBR_INET_NTOA6:
4687 case DIF_SUBR_INET_NTOP: {
4688 size_t size;
4689 int af, argi, i;
4690 char *base, *end;
4691
4692 if (subr == DIF_SUBR_INET_NTOP) {
4693 af = (int)tupregs[0].dttk_value;
4694 argi = 1;
4695 } else {
4696 af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6;
4697 argi = 0;
4698 }
4699
4700 if (af == AF_INET) {
4701 ipaddr_t ip4;
4702 uint8_t *ptr8, val;
4703
4704 /*
4705 * Safely load the IPv4 address.
4706 */
4707 ip4 = dtrace_load32(tupregs[argi].dttk_value);
4708
4709 /*
4710 * Check an IPv4 string will fit in scratch.
4711 */
4712 size = INET_ADDRSTRLEN;
4713 if (!DTRACE_INSCRATCH(mstate, size)) {
4714 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4715 regs[rd] = 0;
4716 break;
4717 }
4718 base = (char *)mstate->dtms_scratch_ptr;
4719 end = (char *)mstate->dtms_scratch_ptr + size - 1;
4720
4721 /*
4722 * Stringify as a dotted decimal quad.
4723 */
4724 *end-- = '\0';
4725 ptr8 = (uint8_t *)&ip4;
4726 for (i = 3; i >= 0; i--) {
4727 val = ptr8[i];
4728
4729 if (val == 0) {
4730 *end-- = '0';
4731 } else {
4732 for (; val; val /= 10) {
4733 *end-- = '0' + (val % 10);
4734 }
4735 }
4736
4737 if (i > 0)
4738 *end-- = '.';
4739 }
4740 ASSERT(end + 1 >= base);
4741
4742 } else if (af == AF_INET6) {
4743 struct in6_addr ip6;
4744 int firstzero, tryzero, numzero, v6end;
4745 uint16_t val;
4746 const char digits[] = "0123456789abcdef";
4747
4748 /*
4749 * Stringify using RFC 1884 convention 2 - 16 bit
4750 * hexadecimal values with a zero-run compression.
4751 * Lower case hexadecimal digits are used.
4752 * eg, fe80::214:4fff:fe0b:76c8.
4753 * The IPv4 embedded form is returned for inet_ntop,
4754 * just the IPv4 string is returned for inet_ntoa6.
4755 */
4756
4757 /*
4758 * Safely load the IPv6 address.
4759 */
4760 dtrace_bcopy(
4761 (void *)(uintptr_t)tupregs[argi].dttk_value,
4762 (void *)(uintptr_t)&ip6, sizeof (struct in6_addr));
4763
4764 /*
4765 * Check an IPv6 string will fit in scratch.
4766 */
4767 size = INET6_ADDRSTRLEN;
4768 if (!DTRACE_INSCRATCH(mstate, size)) {
4769 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4770 regs[rd] = 0;
4771 break;
4772 }
4773 base = (char *)mstate->dtms_scratch_ptr;
4774 end = (char *)mstate->dtms_scratch_ptr + size - 1;
4775 *end-- = '\0';
4776
4777 /*
4778 * Find the longest run of 16 bit zero values
4779 * for the single allowed zero compression - "::".
4780 */
4781 firstzero = -1;
4782 tryzero = -1;
4783 numzero = 1;
4784 for (i = 0; i < sizeof (struct in6_addr); i++) {
4785#if defined(sun)
4786 if (ip6._S6_un._S6_u8[i] == 0 &&
4787#else
4788 if (ip6.__u6_addr.__u6_addr8[i] == 0 &&
4789#endif
4790 tryzero == -1 && i % 2 == 0) {
4791 tryzero = i;
4792 continue;
4793 }
4794
4795 if (tryzero != -1 &&
4796#if defined(sun)
4797 (ip6._S6_un._S6_u8[i] != 0 ||
4798#else
4799 (ip6.__u6_addr.__u6_addr8[i] != 0 ||
4800#endif
4801 i == sizeof (struct in6_addr) - 1)) {
4802
4803 if (i - tryzero <= numzero) {
4804 tryzero = -1;
4805 continue;
4806 }
4807
4808 firstzero = tryzero;
4809 numzero = i - i % 2 - tryzero;
4810 tryzero = -1;
4811
4812#if defined(sun)
4813 if (ip6._S6_un._S6_u8[i] == 0 &&
4814#else
4815 if (ip6.__u6_addr.__u6_addr8[i] == 0 &&
4816#endif
4817 i == sizeof (struct in6_addr) - 1)
4818 numzero += 2;
4819 }
4820 }
4821 ASSERT(firstzero + numzero <= sizeof (struct in6_addr));
4822
4823 /*
4824 * Check for an IPv4 embedded address.
4825 */
4826 v6end = sizeof (struct in6_addr) - 2;
4827 if (IN6_IS_ADDR_V4MAPPED(&ip6) ||
4828 IN6_IS_ADDR_V4COMPAT(&ip6)) {
4829 for (i = sizeof (struct in6_addr) - 1;
4830 i >= DTRACE_V4MAPPED_OFFSET; i--) {
4831 ASSERT(end >= base);
4832
4833#if defined(sun)
4834 val = ip6._S6_un._S6_u8[i];
4835#else
4836 val = ip6.__u6_addr.__u6_addr8[i];
4837#endif
4838
4839 if (val == 0) {
4840 *end-- = '0';
4841 } else {
4842 for (; val; val /= 10) {
4843 *end-- = '0' + val % 10;
4844 }
4845 }
4846
4847 if (i > DTRACE_V4MAPPED_OFFSET)
4848 *end-- = '.';
4849 }
4850
4851 if (subr == DIF_SUBR_INET_NTOA6)
4852 goto inetout;
4853
4854 /*
4855 * Set v6end to skip the IPv4 address that
4856 * we have already stringified.
4857 */
4858 v6end = 10;
4859 }
4860
4861 /*
4862 * Build the IPv6 string by working through the
4863 * address in reverse.
4864 */
4865 for (i = v6end; i >= 0; i -= 2) {
4866 ASSERT(end >= base);
4867
4868 if (i == firstzero + numzero - 2) {
4869 *end-- = ':';
4870 *end-- = ':';
4871 i -= numzero - 2;
4872 continue;
4873 }
4874
4875 if (i < 14 && i != firstzero - 2)
4876 *end-- = ':';
4877
4878#if defined(sun)
4879 val = (ip6._S6_un._S6_u8[i] << 8) +
4880 ip6._S6_un._S6_u8[i + 1];
4881#else
4882 val = (ip6.__u6_addr.__u6_addr8[i] << 8) +
4883 ip6.__u6_addr.__u6_addr8[i + 1];
4884#endif
4885
4886 if (val == 0) {
4887 *end-- = '0';
4888 } else {
4889 for (; val; val /= 16) {
4890 *end-- = digits[val % 16];
4891 }
4892 }
4893 }
4894 ASSERT(end + 1 >= base);
4895
4896 } else {
4897 /*
4898 * The user didn't use AH_INET or AH_INET6.
4899 */
4900 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
4901 regs[rd] = 0;
4902 break;
4903 }
4904
4905inetout: regs[rd] = (uintptr_t)end + 1;
4906 mstate->dtms_scratch_ptr += size;
4907 break;
4908 }
4909
4910 case DIF_SUBR_MEMREF: {
4911 uintptr_t size = 2 * sizeof(uintptr_t);
4912 uintptr_t *memref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t));
4913 size_t scratch_size = ((uintptr_t) memref - mstate->dtms_scratch_ptr) + size;
4914
4915 /* address and length */
4916 memref[0] = tupregs[0].dttk_value;
4917 memref[1] = tupregs[1].dttk_value;
4918
4919 regs[rd] = (uintptr_t) memref;
4920 mstate->dtms_scratch_ptr += scratch_size;
4921 break;
4922 }
4923
4924 case DIF_SUBR_TYPEREF: {
4925 uintptr_t size = 4 * sizeof(uintptr_t);
4926 uintptr_t *typeref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t));
4927 size_t scratch_size = ((uintptr_t) typeref - mstate->dtms_scratch_ptr) + size;
4928
4929 /* address, num_elements, type_str, type_len */
4930 typeref[0] = tupregs[0].dttk_value;
4931 typeref[1] = tupregs[1].dttk_value;
4932 typeref[2] = tupregs[2].dttk_value;
4933 typeref[3] = tupregs[3].dttk_value;
4934
4935 regs[rd] = (uintptr_t) typeref;
4936 mstate->dtms_scratch_ptr += scratch_size;
4937 break;
4938 }
4939 }
4940}
4941
4942/*
4943 * Emulate the execution of DTrace IR instructions specified by the given
4944 * DIF object. This function is deliberately void of assertions as all of
4945 * the necessary checks are handled by a call to dtrace_difo_validate().
4946 */
4947static uint64_t
4948dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate,
4949 dtrace_vstate_t *vstate, dtrace_state_t *state)
4950{
4951 const dif_instr_t *text = difo->dtdo_buf;
4952 const uint_t textlen = difo->dtdo_len;
4953 const char *strtab = difo->dtdo_strtab;
4954 const uint64_t *inttab = difo->dtdo_inttab;
4955
4956 uint64_t rval = 0;
4957 dtrace_statvar_t *svar;
4958 dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
4959 dtrace_difv_t *v;
4960 volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
4961 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
4962
4963 dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */
4964 uint64_t regs[DIF_DIR_NREGS];
4965 uint64_t *tmp;
4966
4967 uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0;
4968 int64_t cc_r;
4969 uint_t pc = 0, id, opc = 0;
4970 uint8_t ttop = 0;
4971 dif_instr_t instr;
4972 uint_t r1, r2, rd;
4973
4974 /*
4975 * We stash the current DIF object into the machine state: we need it
4976 * for subsequent access checking.
4977 */
4978 mstate->dtms_difo = difo;
4979
4980 regs[DIF_REG_R0] = 0; /* %r0 is fixed at zero */
4981
4982 while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) {
4983 opc = pc;
4984
4985 instr = text[pc++];
4986 r1 = DIF_INSTR_R1(instr);
4987 r2 = DIF_INSTR_R2(instr);
4988 rd = DIF_INSTR_RD(instr);
4989
4990 switch (DIF_INSTR_OP(instr)) {
4991 case DIF_OP_OR:
4992 regs[rd] = regs[r1] | regs[r2];
4993 break;
4994 case DIF_OP_XOR:
4995 regs[rd] = regs[r1] ^ regs[r2];
4996 break;
4997 case DIF_OP_AND:
4998 regs[rd] = regs[r1] & regs[r2];
4999 break;
5000 case DIF_OP_SLL:
5001 regs[rd] = regs[r1] << regs[r2];
5002 break;
5003 case DIF_OP_SRL:
5004 regs[rd] = regs[r1] >> regs[r2];
5005 break;
5006 case DIF_OP_SUB:
5007 regs[rd] = regs[r1] - regs[r2];
5008 break;
5009 case DIF_OP_ADD:
5010 regs[rd] = regs[r1] + regs[r2];
5011 break;
5012 case DIF_OP_MUL:
5013 regs[rd] = regs[r1] * regs[r2];
5014 break;
5015 case DIF_OP_SDIV:
5016 if (regs[r2] == 0) {
5017 regs[rd] = 0;
5018 *flags |= CPU_DTRACE_DIVZERO;
5019 } else {
5020 regs[rd] = (int64_t)regs[r1] /
5021 (int64_t)regs[r2];
5022 }
5023 break;
5024
5025 case DIF_OP_UDIV:
5026 if (regs[r2] == 0) {
5027 regs[rd] = 0;
5028 *flags |= CPU_DTRACE_DIVZERO;
5029 } else {
5030 regs[rd] = regs[r1] / regs[r2];
5031 }
5032 break;
5033
5034 case DIF_OP_SREM:
5035 if (regs[r2] == 0) {
5036 regs[rd] = 0;
5037 *flags |= CPU_DTRACE_DIVZERO;
5038 } else {
5039 regs[rd] = (int64_t)regs[r1] %
5040 (int64_t)regs[r2];
5041 }
5042 break;
5043
5044 case DIF_OP_UREM:
5045 if (regs[r2] == 0) {
5046 regs[rd] = 0;
5047 *flags |= CPU_DTRACE_DIVZERO;
5048 } else {
5049 regs[rd] = regs[r1] % regs[r2];
5050 }
5051 break;
5052
5053 case DIF_OP_NOT:
5054 regs[rd] = ~regs[r1];
5055 break;
5056 case DIF_OP_MOV:
5057 regs[rd] = regs[r1];
5058 break;
5059 case DIF_OP_CMP:
5060 cc_r = regs[r1] - regs[r2];
5061 cc_n = cc_r < 0;
5062 cc_z = cc_r == 0;
5063 cc_v = 0;
5064 cc_c = regs[r1] < regs[r2];
5065 break;
5066 case DIF_OP_TST:
5067 cc_n = cc_v = cc_c = 0;
5068 cc_z = regs[r1] == 0;
5069 break;
5070 case DIF_OP_BA:
5071 pc = DIF_INSTR_LABEL(instr);
5072 break;
5073 case DIF_OP_BE:
5074 if (cc_z)
5075 pc = DIF_INSTR_LABEL(instr);
5076 break;
5077 case DIF_OP_BNE:
5078 if (cc_z == 0)
5079 pc = DIF_INSTR_LABEL(instr);
5080 break;
5081 case DIF_OP_BG:
5082 if ((cc_z | (cc_n ^ cc_v)) == 0)
5083 pc = DIF_INSTR_LABEL(instr);
5084 break;
5085 case DIF_OP_BGU:
5086 if ((cc_c | cc_z) == 0)
5087 pc = DIF_INSTR_LABEL(instr);
5088 break;
5089 case DIF_OP_BGE:
5090 if ((cc_n ^ cc_v) == 0)
5091 pc = DIF_INSTR_LABEL(instr);
5092 break;
5093 case DIF_OP_BGEU:
5094 if (cc_c == 0)
5095 pc = DIF_INSTR_LABEL(instr);
5096 break;
5097 case DIF_OP_BL:
5098 if (cc_n ^ cc_v)
5099 pc = DIF_INSTR_LABEL(instr);
5100 break;
5101 case DIF_OP_BLU:
5102 if (cc_c)
5103 pc = DIF_INSTR_LABEL(instr);
5104 break;
5105 case DIF_OP_BLE:
5106 if (cc_z | (cc_n ^ cc_v))
5107 pc = DIF_INSTR_LABEL(instr);
5108 break;
5109 case DIF_OP_BLEU:
5110 if (cc_c | cc_z)
5111 pc = DIF_INSTR_LABEL(instr);
5112 break;
5113 case DIF_OP_RLDSB:
5114 if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) {
5115 *flags |= CPU_DTRACE_KPRIV;
5116 *illval = regs[r1];
5117 break;
5118 }
5119 /*FALLTHROUGH*/
5120 case DIF_OP_LDSB:
5121 regs[rd] = (int8_t)dtrace_load8(regs[r1]);
5122 break;
5123 case DIF_OP_RLDSH:
5124 if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) {
5125 *flags |= CPU_DTRACE_KPRIV;
5126 *illval = regs[r1];
5127 break;
5128 }
5129 /*FALLTHROUGH*/
5130 case DIF_OP_LDSH:
5131 regs[rd] = (int16_t)dtrace_load16(regs[r1]);
5132 break;
5133 case DIF_OP_RLDSW:
5134 if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) {
5135 *flags |= CPU_DTRACE_KPRIV;
5136 *illval = regs[r1];
5137 break;
5138 }
5139 /*FALLTHROUGH*/
5140 case DIF_OP_LDSW:
5141 regs[rd] = (int32_t)dtrace_load32(regs[r1]);
5142 break;
5143 case DIF_OP_RLDUB:
5144 if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) {
5145 *flags |= CPU_DTRACE_KPRIV;
5146 *illval = regs[r1];
5147 break;
5148 }
5149 /*FALLTHROUGH*/
5150 case DIF_OP_LDUB:
5151 regs[rd] = dtrace_load8(regs[r1]);
5152 break;
5153 case DIF_OP_RLDUH:
5154 if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) {
5155 *flags |= CPU_DTRACE_KPRIV;
5156 *illval = regs[r1];
5157 break;
5158 }
5159 /*FALLTHROUGH*/
5160 case DIF_OP_LDUH:
5161 regs[rd] = dtrace_load16(regs[r1]);
5162 break;
5163 case DIF_OP_RLDUW:
5164 if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) {
5165 *flags |= CPU_DTRACE_KPRIV;
5166 *illval = regs[r1];
5167 break;
5168 }
5169 /*FALLTHROUGH*/
5170 case DIF_OP_LDUW:
5171 regs[rd] = dtrace_load32(regs[r1]);
5172 break;
5173 case DIF_OP_RLDX:
5174 if (!dtrace_canstore(regs[r1], 8, mstate, vstate)) {
5175 *flags |= CPU_DTRACE_KPRIV;
5176 *illval = regs[r1];
5177 break;
5178 }
5179 /*FALLTHROUGH*/
5180 case DIF_OP_LDX:
5181 regs[rd] = dtrace_load64(regs[r1]);
5182 break;
5183 case DIF_OP_ULDSB:
5184 regs[rd] = (int8_t)
5185 dtrace_fuword8((void *)(uintptr_t)regs[r1]);
5186 break;
5187 case DIF_OP_ULDSH:
5188 regs[rd] = (int16_t)
5189 dtrace_fuword16((void *)(uintptr_t)regs[r1]);
5190 break;
5191 case DIF_OP_ULDSW:
5192 regs[rd] = (int32_t)
5193 dtrace_fuword32((void *)(uintptr_t)regs[r1]);
5194 break;
5195 case DIF_OP_ULDUB:
5196 regs[rd] =
5197 dtrace_fuword8((void *)(uintptr_t)regs[r1]);
5198 break;
5199 case DIF_OP_ULDUH:
5200 regs[rd] =
5201 dtrace_fuword16((void *)(uintptr_t)regs[r1]);
5202 break;
5203 case DIF_OP_ULDUW:
5204 regs[rd] =
5205 dtrace_fuword32((void *)(uintptr_t)regs[r1]);
5206 break;
5207 case DIF_OP_ULDX:
5208 regs[rd] =
5209 dtrace_fuword64((void *)(uintptr_t)regs[r1]);
5210 break;
5211 case DIF_OP_RET:
5212 rval = regs[rd];
5213 pc = textlen;
5214 break;
5215 case DIF_OP_NOP:
5216 break;
5217 case DIF_OP_SETX:
5218 regs[rd] = inttab[DIF_INSTR_INTEGER(instr)];
5219 break;
5220 case DIF_OP_SETS:
5221 regs[rd] = (uint64_t)(uintptr_t)
5222 (strtab + DIF_INSTR_STRING(instr));
5223 break;
5224 case DIF_OP_SCMP: {
5225 size_t sz = state->dts_options[DTRACEOPT_STRSIZE];
5226 uintptr_t s1 = regs[r1];
5227 uintptr_t s2 = regs[r2];
5228
5229 if (s1 != 0 &&
5230 !dtrace_strcanload(s1, sz, mstate, vstate))
5231 break;
5232 if (s2 != 0 &&
5233 !dtrace_strcanload(s2, sz, mstate, vstate))
5234 break;
5235
5236 cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz);
5237
5238 cc_n = cc_r < 0;
5239 cc_z = cc_r == 0;
5240 cc_v = cc_c = 0;
5241 break;
5242 }
5243 case DIF_OP_LDGA:
5244 regs[rd] = dtrace_dif_variable(mstate, state,
5245 r1, regs[r2]);
5246 break;
5247 case DIF_OP_LDGS:
5248 id = DIF_INSTR_VAR(instr);
5249
5250 if (id >= DIF_VAR_OTHER_UBASE) {
5251 uintptr_t a;
5252
5253 id -= DIF_VAR_OTHER_UBASE;
5254 svar = vstate->dtvs_globals[id];
5255 ASSERT(svar != NULL);
5256 v = &svar->dtsv_var;
5257
5258 if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) {
5259 regs[rd] = svar->dtsv_data;
5260 break;
5261 }
5262
5263 a = (uintptr_t)svar->dtsv_data;
5264
5265 if (*(uint8_t *)a == UINT8_MAX) {
5266 /*
5267 * If the 0th byte is set to UINT8_MAX
5268 * then this is to be treated as a
5269 * reference to a NULL variable.
5270 */
5271 regs[rd] = 0;
5272 } else {
5273 regs[rd] = a + sizeof (uint64_t);
5274 }
5275
5276 break;
5277 }
5278
5279 regs[rd] = dtrace_dif_variable(mstate, state, id, 0);
5280 break;
5281
5282 case DIF_OP_STGS:
5283 id = DIF_INSTR_VAR(instr);
5284
5285 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5286 id -= DIF_VAR_OTHER_UBASE;
5287
5288 svar = vstate->dtvs_globals[id];
5289 ASSERT(svar != NULL);
5290 v = &svar->dtsv_var;
5291
5292 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5293 uintptr_t a = (uintptr_t)svar->dtsv_data;
5294
5295 ASSERT(a != 0);
5296 ASSERT(svar->dtsv_size != 0);
5297
5298 if (regs[rd] == 0) {
5299 *(uint8_t *)a = UINT8_MAX;
5300 break;
5301 } else {
5302 *(uint8_t *)a = 0;
5303 a += sizeof (uint64_t);
5304 }
5305 if (!dtrace_vcanload(
5306 (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5307 mstate, vstate))
5308 break;
5309
5310 dtrace_vcopy((void *)(uintptr_t)regs[rd],
5311 (void *)a, &v->dtdv_type);
5312 break;
5313 }
5314
5315 svar->dtsv_data = regs[rd];
5316 break;
5317
5318 case DIF_OP_LDTA:
5319 /*
5320 * There are no DTrace built-in thread-local arrays at
5321 * present. This opcode is saved for future work.
5322 */
5323 *flags |= CPU_DTRACE_ILLOP;
5324 regs[rd] = 0;
5325 break;
5326
5327 case DIF_OP_LDLS:
5328 id = DIF_INSTR_VAR(instr);
5329
5330 if (id < DIF_VAR_OTHER_UBASE) {
5331 /*
5332 * For now, this has no meaning.
5333 */
5334 regs[rd] = 0;
5335 break;
5336 }
5337
5338 id -= DIF_VAR_OTHER_UBASE;
5339
5340 ASSERT(id < vstate->dtvs_nlocals);
5341 ASSERT(vstate->dtvs_locals != NULL);
5342
5343 svar = vstate->dtvs_locals[id];
5344 ASSERT(svar != NULL);
5345 v = &svar->dtsv_var;
5346
5347 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5348 uintptr_t a = (uintptr_t)svar->dtsv_data;
5349 size_t sz = v->dtdv_type.dtdt_size;
5350
5351 sz += sizeof (uint64_t);
5352 ASSERT(svar->dtsv_size == NCPU * sz);
5353 a += curcpu * sz;
5354
5355 if (*(uint8_t *)a == UINT8_MAX) {
5356 /*
5357 * If the 0th byte is set to UINT8_MAX
5358 * then this is to be treated as a
5359 * reference to a NULL variable.
5360 */
5361 regs[rd] = 0;
5362 } else {
5363 regs[rd] = a + sizeof (uint64_t);
5364 }
5365
5366 break;
5367 }
5368
5369 ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t));
5370 tmp = (uint64_t *)(uintptr_t)svar->dtsv_data;
5371 regs[rd] = tmp[curcpu];
5372 break;
5373
5374 case DIF_OP_STLS:
5375 id = DIF_INSTR_VAR(instr);
5376
5377 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5378 id -= DIF_VAR_OTHER_UBASE;
5379 ASSERT(id < vstate->dtvs_nlocals);
5380
5381 ASSERT(vstate->dtvs_locals != NULL);
5382 svar = vstate->dtvs_locals[id];
5383 ASSERT(svar != NULL);
5384 v = &svar->dtsv_var;
5385
5386 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5387 uintptr_t a = (uintptr_t)svar->dtsv_data;
5388 size_t sz = v->dtdv_type.dtdt_size;
5389
5390 sz += sizeof (uint64_t);
5391 ASSERT(svar->dtsv_size == NCPU * sz);
5392 a += curcpu * sz;
5393
5394 if (regs[rd] == 0) {
5395 *(uint8_t *)a = UINT8_MAX;
5396 break;
5397 } else {
5398 *(uint8_t *)a = 0;
5399 a += sizeof (uint64_t);
5400 }
5401
5402 if (!dtrace_vcanload(
5403 (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5404 mstate, vstate))
5405 break;
5406
5407 dtrace_vcopy((void *)(uintptr_t)regs[rd],
5408 (void *)a, &v->dtdv_type);
5409 break;
5410 }
5411
5412 ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t));
5413 tmp = (uint64_t *)(uintptr_t)svar->dtsv_data;
5414 tmp[curcpu] = regs[rd];
5415 break;
5416
5417 case DIF_OP_LDTS: {
5418 dtrace_dynvar_t *dvar;
5419 dtrace_key_t *key;
5420
5421 id = DIF_INSTR_VAR(instr);
5422 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5423 id -= DIF_VAR_OTHER_UBASE;
5424 v = &vstate->dtvs_tlocals[id];
5425
5426 key = &tupregs[DIF_DTR_NREGS];
5427 key[0].dttk_value = (uint64_t)id;
5428 key[0].dttk_size = 0;
5429 DTRACE_TLS_THRKEY(key[1].dttk_value);
5430 key[1].dttk_size = 0;
5431
5432 dvar = dtrace_dynvar(dstate, 2, key,
5433 sizeof (uint64_t), DTRACE_DYNVAR_NOALLOC,
5434 mstate, vstate);
5435
5436 if (dvar == NULL) {
5437 regs[rd] = 0;
5438 break;
5439 }
5440
5441 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5442 regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data;
5443 } else {
5444 regs[rd] = *((uint64_t *)dvar->dtdv_data);
5445 }
5446
5447 break;
5448 }
5449
5450 case DIF_OP_STTS: {
5451 dtrace_dynvar_t *dvar;
5452 dtrace_key_t *key;
5453
5454 id = DIF_INSTR_VAR(instr);
5455 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5456 id -= DIF_VAR_OTHER_UBASE;
5457
5458 key = &tupregs[DIF_DTR_NREGS];
5459 key[0].dttk_value = (uint64_t)id;
5460 key[0].dttk_size = 0;
5461 DTRACE_TLS_THRKEY(key[1].dttk_value);
5462 key[1].dttk_size = 0;
5463 v = &vstate->dtvs_tlocals[id];
5464
5465 dvar = dtrace_dynvar(dstate, 2, key,
5466 v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5467 v->dtdv_type.dtdt_size : sizeof (uint64_t),
5468 regs[rd] ? DTRACE_DYNVAR_ALLOC :
5469 DTRACE_DYNVAR_DEALLOC, mstate, vstate);
5470
5471 /*
5472 * Given that we're storing to thread-local data,
5473 * we need to flush our predicate cache.
5474 */
5475 curthread->t_predcache = 0;
5476
5477 if (dvar == NULL)
5478 break;
5479
5480 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5481 if (!dtrace_vcanload(
5482 (void *)(uintptr_t)regs[rd],
5483 &v->dtdv_type, mstate, vstate))
5484 break;
5485
5486 dtrace_vcopy((void *)(uintptr_t)regs[rd],
5487 dvar->dtdv_data, &v->dtdv_type);
5488 } else {
5489 *((uint64_t *)dvar->dtdv_data) = regs[rd];
5490 }
5491
5492 break;
5493 }
5494
5495 case DIF_OP_SRA:
5496 regs[rd] = (int64_t)regs[r1] >> regs[r2];
5497 break;
5498
5499 case DIF_OP_CALL:
5500 dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd,
5501 regs, tupregs, ttop, mstate, state);
5502 break;
5503
5504 case DIF_OP_PUSHTR:
5505 if (ttop == DIF_DTR_NREGS) {
5506 *flags |= CPU_DTRACE_TUPOFLOW;
5507 break;
5508 }
5509
5510 if (r1 == DIF_TYPE_STRING) {
5511 /*
5512 * If this is a string type and the size is 0,
5513 * we'll use the system-wide default string
5514 * size. Note that we are _not_ looking at
5515 * the value of the DTRACEOPT_STRSIZE option;
5516 * had this been set, we would expect to have
5517 * a non-zero size value in the "pushtr".
5518 */
5519 tupregs[ttop].dttk_size =
5520 dtrace_strlen((char *)(uintptr_t)regs[rd],
5521 regs[r2] ? regs[r2] :
5522 dtrace_strsize_default) + 1;
5523 } else {
5524 tupregs[ttop].dttk_size = regs[r2];
5525 }
5526
5527 tupregs[ttop++].dttk_value = regs[rd];
5528 break;
5529
5530 case DIF_OP_PUSHTV:
5531 if (ttop == DIF_DTR_NREGS) {
5532 *flags |= CPU_DTRACE_TUPOFLOW;
5533 break;
5534 }
5535
5536 tupregs[ttop].dttk_value = regs[rd];
5537 tupregs[ttop++].dttk_size = 0;
5538 break;
5539
5540 case DIF_OP_POPTS:
5541 if (ttop != 0)
5542 ttop--;
5543 break;
5544
5545 case DIF_OP_FLUSHTS:
5546 ttop = 0;
5547 break;
5548
5549 case DIF_OP_LDGAA:
5550 case DIF_OP_LDTAA: {
5551 dtrace_dynvar_t *dvar;
5552 dtrace_key_t *key = tupregs;
5553 uint_t nkeys = ttop;
5554
5555 id = DIF_INSTR_VAR(instr);
5556 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5557 id -= DIF_VAR_OTHER_UBASE;
5558
5559 key[nkeys].dttk_value = (uint64_t)id;
5560 key[nkeys++].dttk_size = 0;
5561
5562 if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) {
5563 DTRACE_TLS_THRKEY(key[nkeys].dttk_value);
5564 key[nkeys++].dttk_size = 0;
5565 v = &vstate->dtvs_tlocals[id];
5566 } else {
5567 v = &vstate->dtvs_globals[id]->dtsv_var;
5568 }
5569
5570 dvar = dtrace_dynvar(dstate, nkeys, key,
5571 v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5572 v->dtdv_type.dtdt_size : sizeof (uint64_t),
5573 DTRACE_DYNVAR_NOALLOC, mstate, vstate);
5574
5575 if (dvar == NULL) {
5576 regs[rd] = 0;
5577 break;
5578 }
5579
5580 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5581 regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data;
5582 } else {
5583 regs[rd] = *((uint64_t *)dvar->dtdv_data);
5584 }
5585
5586 break;
5587 }
5588
5589 case DIF_OP_STGAA:
5590 case DIF_OP_STTAA: {
5591 dtrace_dynvar_t *dvar;
5592 dtrace_key_t *key = tupregs;
5593 uint_t nkeys = ttop;
5594
5595 id = DIF_INSTR_VAR(instr);
5596 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5597 id -= DIF_VAR_OTHER_UBASE;
5598
5599 key[nkeys].dttk_value = (uint64_t)id;
5600 key[nkeys++].dttk_size = 0;
5601
5602 if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) {
5603 DTRACE_TLS_THRKEY(key[nkeys].dttk_value);
5604 key[nkeys++].dttk_size = 0;
5605 v = &vstate->dtvs_tlocals[id];
5606 } else {
5607 v = &vstate->dtvs_globals[id]->dtsv_var;
5608 }
5609
5610 dvar = dtrace_dynvar(dstate, nkeys, key,
5611 v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5612 v->dtdv_type.dtdt_size : sizeof (uint64_t),
5613 regs[rd] ? DTRACE_DYNVAR_ALLOC :
5614 DTRACE_DYNVAR_DEALLOC, mstate, vstate);
5615
5616 if (dvar == NULL)
5617 break;
5618
5619 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5620 if (!dtrace_vcanload(
5621 (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5622 mstate, vstate))
5623 break;
5624
5625 dtrace_vcopy((void *)(uintptr_t)regs[rd],
5626 dvar->dtdv_data, &v->dtdv_type);
5627 } else {
5628 *((uint64_t *)dvar->dtdv_data) = regs[rd];
5629 }
5630
5631 break;
5632 }
5633
5634 case DIF_OP_ALLOCS: {
5635 uintptr_t ptr = P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
5636 size_t size = ptr - mstate->dtms_scratch_ptr + regs[r1];
5637
5638 /*
5639 * Rounding up the user allocation size could have
5640 * overflowed large, bogus allocations (like -1ULL) to
5641 * 0.
5642 */
5643 if (size < regs[r1] ||
5644 !DTRACE_INSCRATCH(mstate, size)) {
5645 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5646 regs[rd] = 0;
5647 break;
5648 }
5649
5650 dtrace_bzero((void *) mstate->dtms_scratch_ptr, size);
5651 mstate->dtms_scratch_ptr += size;
5652 regs[rd] = ptr;
5653 break;
5654 }
5655
5656 case DIF_OP_COPYS:
5657 if (!dtrace_canstore(regs[rd], regs[r2],
5658 mstate, vstate)) {
5659 *flags |= CPU_DTRACE_BADADDR;
5660 *illval = regs[rd];
5661 break;
5662 }
5663
5664 if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate))
5665 break;
5666
5667 dtrace_bcopy((void *)(uintptr_t)regs[r1],
5668 (void *)(uintptr_t)regs[rd], (size_t)regs[r2]);
5669 break;
5670
5671 case DIF_OP_STB:
5672 if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) {
5673 *flags |= CPU_DTRACE_BADADDR;
5674 *illval = regs[rd];
5675 break;
5676 }
5677 *((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1];
5678 break;
5679
5680 case DIF_OP_STH:
5681 if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) {
5682 *flags |= CPU_DTRACE_BADADDR;
5683 *illval = regs[rd];
5684 break;
5685 }
5686 if (regs[rd] & 1) {
5687 *flags |= CPU_DTRACE_BADALIGN;
5688 *illval = regs[rd];
5689 break;
5690 }
5691 *((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1];
5692 break;
5693
5694 case DIF_OP_STW:
5695 if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) {
5696 *flags |= CPU_DTRACE_BADADDR;
5697 *illval = regs[rd];
5698 break;
5699 }
5700 if (regs[rd] & 3) {
5701 *flags |= CPU_DTRACE_BADALIGN;
5702 *illval = regs[rd];
5703 break;
5704 }
5705 *((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1];
5706 break;
5707
5708 case DIF_OP_STX:
5709 if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) {
5710 *flags |= CPU_DTRACE_BADADDR;
5711 *illval = regs[rd];
5712 break;
5713 }
5714 if (regs[rd] & 7) {
5715 *flags |= CPU_DTRACE_BADALIGN;
5716 *illval = regs[rd];
5717 break;
5718 }
5719 *((uint64_t *)(uintptr_t)regs[rd]) = regs[r1];
5720 break;
5721 }
5722 }
5723
5724 if (!(*flags & CPU_DTRACE_FAULT))
5725 return (rval);
5726
5727 mstate->dtms_fltoffs = opc * sizeof (dif_instr_t);
5728 mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS;
5729
5730 return (0);
5731}
5732
5733static void
5734dtrace_action_breakpoint(dtrace_ecb_t *ecb)
5735{
5736 dtrace_probe_t *probe = ecb->dte_probe;
5737 dtrace_provider_t *prov = probe->dtpr_provider;
5738 char c[DTRACE_FULLNAMELEN + 80], *str;
5739 char *msg = "dtrace: breakpoint action at probe ";
5740 char *ecbmsg = " (ecb ";
5741 uintptr_t mask = (0xf << (sizeof (uintptr_t) * NBBY / 4));
5742 uintptr_t val = (uintptr_t)ecb;
5743 int shift = (sizeof (uintptr_t) * NBBY) - 4, i = 0;
5744
5745 if (dtrace_destructive_disallow)
5746 return;
5747
5748 /*
5749 * It's impossible to be taking action on the NULL probe.
5750 */
5751 ASSERT(probe != NULL);
5752
5753 /*
5754 * This is a poor man's (destitute man's?) sprintf(): we want to
5755 * print the provider name, module name, function name and name of
5756 * the probe, along with the hex address of the ECB with the breakpoint
5757 * action -- all of which we must place in the character buffer by
5758 * hand.
5759 */
5760 while (*msg != '\0')
5761 c[i++] = *msg++;
5762
5763 for (str = prov->dtpv_name; *str != '\0'; str++)
5764 c[i++] = *str;
5765 c[i++] = ':';
5766
5767 for (str = probe->dtpr_mod; *str != '\0'; str++)
5768 c[i++] = *str;
5769 c[i++] = ':';
5770
5771 for (str = probe->dtpr_func; *str != '\0'; str++)
5772 c[i++] = *str;
5773 c[i++] = ':';
5774
5775 for (str = probe->dtpr_name; *str != '\0'; str++)
5776 c[i++] = *str;
5777
5778 while (*ecbmsg != '\0')
5779 c[i++] = *ecbmsg++;
5780
5781 while (shift >= 0) {
5782 mask = (uintptr_t)0xf << shift;
5783
5784 if (val >= ((uintptr_t)1 << shift))
5785 c[i++] = "0123456789abcdef"[(val & mask) >> shift];
5786 shift -= 4;
5787 }
5788
5789 c[i++] = ')';
5790 c[i] = '\0';
5791
5792#if defined(sun)
5793 debug_enter(c);
5794#else
5795 kdb_enter(KDB_WHY_DTRACE, "breakpoint action");
5796#endif
5797}
5798
5799static void
5800dtrace_action_panic(dtrace_ecb_t *ecb)
5801{
5802 dtrace_probe_t *probe = ecb->dte_probe;
5803
5804 /*
5805 * It's impossible to be taking action on the NULL probe.
5806 */
5807 ASSERT(probe != NULL);
5808
5809 if (dtrace_destructive_disallow)
5810 return;
5811
5812 if (dtrace_panicked != NULL)
5813 return;
5814
5815 if (dtrace_casptr(&dtrace_panicked, NULL, curthread) != NULL)
5816 return;
5817
5818 /*
5819 * We won the right to panic. (We want to be sure that only one
5820 * thread calls panic() from dtrace_probe(), and that panic() is
5821 * called exactly once.)
5822 */
5823 dtrace_panic("dtrace: panic action at probe %s:%s:%s:%s (ecb %p)",
5824 probe->dtpr_provider->dtpv_name, probe->dtpr_mod,
5825 probe->dtpr_func, probe->dtpr_name, (void *)ecb);
5826}
5827
5828static void
5829dtrace_action_raise(uint64_t sig)
5830{
5831 if (dtrace_destructive_disallow)
5832 return;
5833
5834 if (sig >= NSIG) {
5835 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
5836 return;
5837 }
5838
5839#if defined(sun)
5840 /*
5841 * raise() has a queue depth of 1 -- we ignore all subsequent
5842 * invocations of the raise() action.
5843 */
5844 if (curthread->t_dtrace_sig == 0)
5845 curthread->t_dtrace_sig = (uint8_t)sig;
5846
5847 curthread->t_sig_check = 1;
5848 aston(curthread);
5849#else
5850 struct proc *p = curproc;
5851 PROC_LOCK(p);
5852 kern_psignal(p, sig);
5853 PROC_UNLOCK(p);
5854#endif
5855}
5856
5857static void
5858dtrace_action_stop(void)
5859{
5860 if (dtrace_destructive_disallow)
5861 return;
5862
5863#if defined(sun)
5864 if (!curthread->t_dtrace_stop) {
5865 curthread->t_dtrace_stop = 1;
5866 curthread->t_sig_check = 1;
5867 aston(curthread);
5868 }
5869#else
5870 struct proc *p = curproc;
5871 PROC_LOCK(p);
5872 kern_psignal(p, SIGSTOP);
5873 PROC_UNLOCK(p);
5874#endif
5875}
5876
5877static void
5878dtrace_action_chill(dtrace_mstate_t *mstate, hrtime_t val)
5879{
5880 hrtime_t now;
5881 volatile uint16_t *flags;
5882#if defined(sun)
5883 cpu_t *cpu = CPU;
5884#else
5885 cpu_t *cpu = &solaris_cpu[curcpu];
5886#endif
5887
5888 if (dtrace_destructive_disallow)
5889 return;
5890
5891 flags = (volatile uint16_t *)&cpu_core[cpu->cpu_id].cpuc_dtrace_flags;
5892
5893 now = dtrace_gethrtime();
5894
5895 if (now - cpu->cpu_dtrace_chillmark > dtrace_chill_interval) {
5896 /*
5897 * We need to advance the mark to the current time.
5898 */
5899 cpu->cpu_dtrace_chillmark = now;
5900 cpu->cpu_dtrace_chilled = 0;
5901 }
5902
5903 /*
5904 * Now check to see if the requested chill time would take us over
5905 * the maximum amount of time allowed in the chill interval. (Or
5906 * worse, if the calculation itself induces overflow.)
5907 */
5908 if (cpu->cpu_dtrace_chilled + val > dtrace_chill_max ||
5909 cpu->cpu_dtrace_chilled + val < cpu->cpu_dtrace_chilled) {
5910 *flags |= CPU_DTRACE_ILLOP;
5911 return;
5912 }
5913
5914 while (dtrace_gethrtime() - now < val)
5915 continue;
5916
5917 /*
5918 * Normally, we assure that the value of the variable "timestamp" does
5919 * not change within an ECB. The presence of chill() represents an
5920 * exception to this rule, however.
5921 */
5922 mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP;
5923 cpu->cpu_dtrace_chilled += val;
5924}
5925
5926static void
5927dtrace_action_ustack(dtrace_mstate_t *mstate, dtrace_state_t *state,
5928 uint64_t *buf, uint64_t arg)
5929{
5930 int nframes = DTRACE_USTACK_NFRAMES(arg);
5931 int strsize = DTRACE_USTACK_STRSIZE(arg);
5932 uint64_t *pcs = &buf[1], *fps;
5933 char *str = (char *)&pcs[nframes];
5934 int size, offs = 0, i, j;
5935 uintptr_t old = mstate->dtms_scratch_ptr, saved;
5936 uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
5937 char *sym;
5938
5939 /*
5940 * Should be taking a faster path if string space has not been
5941 * allocated.
5942 */
5943 ASSERT(strsize != 0);
5944
5945 /*
5946 * We will first allocate some temporary space for the frame pointers.
5947 */
5948 fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
5949 size = (uintptr_t)fps - mstate->dtms_scratch_ptr +
5950 (nframes * sizeof (uint64_t));
5951
5952 if (!DTRACE_INSCRATCH(mstate, size)) {
5953 /*
5954 * Not enough room for our frame pointers -- need to indicate
5955 * that we ran out of scratch space.
5956 */
5957 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5958 return;
5959 }
5960
5961 mstate->dtms_scratch_ptr += size;
5962 saved = mstate->dtms_scratch_ptr;
5963
5964 /*
5965 * Now get a stack with both program counters and frame pointers.
5966 */
5967 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5968 dtrace_getufpstack(buf, fps, nframes + 1);
5969 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5970
5971 /*
5972 * If that faulted, we're cooked.
5973 */
5974 if (*flags & CPU_DTRACE_FAULT)
5975 goto out;
5976
5977 /*
5978 * Now we want to walk up the stack, calling the USTACK helper. For
5979 * each iteration, we restore the scratch pointer.
5980 */
5981 for (i = 0; i < nframes; i++) {
5982 mstate->dtms_scratch_ptr = saved;
5983
5984 if (offs >= strsize)
5985 break;
5986
5987 sym = (char *)(uintptr_t)dtrace_helper(
5988 DTRACE_HELPER_ACTION_USTACK,
5989 mstate, state, pcs[i], fps[i]);
5990
5991 /*
5992 * If we faulted while running the helper, we're going to
5993 * clear the fault and null out the corresponding string.
5994 */
5995 if (*flags & CPU_DTRACE_FAULT) {
5996 *flags &= ~CPU_DTRACE_FAULT;
5997 str[offs++] = '\0';
5998 continue;
5999 }
6000
6001 if (sym == NULL) {
6002 str[offs++] = '\0';
6003 continue;
6004 }
6005
6006 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6007
6008 /*
6009 * Now copy in the string that the helper returned to us.
6010 */
6011 for (j = 0; offs + j < strsize; j++) {
6012 if ((str[offs + j] = sym[j]) == '\0')
6013 break;
6014 }
6015
6016 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6017
6018 offs += j + 1;
6019 }
6020
6021 if (offs >= strsize) {
6022 /*
6023 * If we didn't have room for all of the strings, we don't
6024 * abort processing -- this needn't be a fatal error -- but we
6025 * still want to increment a counter (dts_stkstroverflows) to
6026 * allow this condition to be warned about. (If this is from
6027 * a jstack() action, it is easily tuned via jstackstrsize.)
6028 */
6029 dtrace_error(&state->dts_stkstroverflows);
6030 }
6031
6032 while (offs < strsize)
6033 str[offs++] = '\0';
6034
6035out:
6036 mstate->dtms_scratch_ptr = old;
6037}
6038
6039/*
6040 * If you're looking for the epicenter of DTrace, you just found it. This
6041 * is the function called by the provider to fire a probe -- from which all
6042 * subsequent probe-context DTrace activity emanates.
6043 */
6044void
6045dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1,
6046 uintptr_t arg2, uintptr_t arg3, uintptr_t arg4)
6047{
6048 processorid_t cpuid;
6049 dtrace_icookie_t cookie;
6050 dtrace_probe_t *probe;
6051 dtrace_mstate_t mstate;
6052 dtrace_ecb_t *ecb;
6053 dtrace_action_t *act;
6054 intptr_t offs;
6055 size_t size;
6056 int vtime, onintr;
6057 volatile uint16_t *flags;
6058 hrtime_t now;
6059
6060 if (panicstr != NULL)
6061 return;
6062
6063#if defined(sun)
6064 /*
6065 * Kick out immediately if this CPU is still being born (in which case
6066 * curthread will be set to -1) or the current thread can't allow
6067 * probes in its current context.
6068 */
6069 if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE))
6070 return;
6071#endif
6072
6073 cookie = dtrace_interrupt_disable();
6074 probe = dtrace_probes[id - 1];
6075 cpuid = curcpu;
6076 onintr = CPU_ON_INTR(CPU);
6077
6078 if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE &&
6079 probe->dtpr_predcache == curthread->t_predcache) {
6080 /*
6081 * We have hit in the predicate cache; we know that
6082 * this predicate would evaluate to be false.
6083 */
6084 dtrace_interrupt_enable(cookie);
6085 return;
6086 }
6087
6088#if defined(sun)
6089 if (panic_quiesce) {
6090#else
6091 if (panicstr != NULL) {
6092#endif
6093 /*
6094 * We don't trace anything if we're panicking.
6095 */
6096 dtrace_interrupt_enable(cookie);
6097 return;
6098 }
6099
6100 now = dtrace_gethrtime();
6101 vtime = dtrace_vtime_references != 0;
6102
6103 if (vtime && curthread->t_dtrace_start)
6104 curthread->t_dtrace_vtime += now - curthread->t_dtrace_start;
6105
6106 mstate.dtms_difo = NULL;
6107 mstate.dtms_probe = probe;
6108 mstate.dtms_strtok = 0;
6109 mstate.dtms_arg[0] = arg0;
6110 mstate.dtms_arg[1] = arg1;
6111 mstate.dtms_arg[2] = arg2;
6112 mstate.dtms_arg[3] = arg3;
6113 mstate.dtms_arg[4] = arg4;
6114
6115 flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags;
6116
6117 for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) {
6118 dtrace_predicate_t *pred = ecb->dte_predicate;
6119 dtrace_state_t *state = ecb->dte_state;
6120 dtrace_buffer_t *buf = &state->dts_buffer[cpuid];
6121 dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid];
6122 dtrace_vstate_t *vstate = &state->dts_vstate;
6123 dtrace_provider_t *prov = probe->dtpr_provider;
6124 uint64_t tracememsize = 0;
6125 int committed = 0;
6126 caddr_t tomax;
6127
6128 /*
6129 * A little subtlety with the following (seemingly innocuous)
6130 * declaration of the automatic 'val': by looking at the
6131 * code, you might think that it could be declared in the
6132 * action processing loop, below. (That is, it's only used in
6133 * the action processing loop.) However, it must be declared
6134 * out of that scope because in the case of DIF expression
6135 * arguments to aggregating actions, one iteration of the
6136 * action loop will use the last iteration's value.
6137 */
6138 uint64_t val = 0;
6139
6140 mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE;
6141 *flags &= ~CPU_DTRACE_ERROR;
6142
6143 if (prov == dtrace_provider) {
6144 /*
6145 * If dtrace itself is the provider of this probe,
6146 * we're only going to continue processing the ECB if
6147 * arg0 (the dtrace_state_t) is equal to the ECB's
6148 * creating state. (This prevents disjoint consumers
6149 * from seeing one another's metaprobes.)
6150 */
6151 if (arg0 != (uint64_t)(uintptr_t)state)
6152 continue;
6153 }
6154
6155 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) {
6156 /*
6157 * We're not currently active. If our provider isn't
6158 * the dtrace pseudo provider, we're not interested.
6159 */
6160 if (prov != dtrace_provider)
6161 continue;
6162
6163 /*
6164 * Now we must further check if we are in the BEGIN
6165 * probe. If we are, we will only continue processing
6166 * if we're still in WARMUP -- if one BEGIN enabling
6167 * has invoked the exit() action, we don't want to
6168 * evaluate subsequent BEGIN enablings.
6169 */
6170 if (probe->dtpr_id == dtrace_probeid_begin &&
6171 state->dts_activity != DTRACE_ACTIVITY_WARMUP) {
6172 ASSERT(state->dts_activity ==
6173 DTRACE_ACTIVITY_DRAINING);
6174 continue;
6175 }
6176 }
6177
6178 if (ecb->dte_cond) {
6179 /*
6180 * If the dte_cond bits indicate that this
6181 * consumer is only allowed to see user-mode firings
6182 * of this probe, call the provider's dtps_usermode()
6183 * entry point to check that the probe was fired
6184 * while in a user context. Skip this ECB if that's
6185 * not the case.
6186 */
6187 if ((ecb->dte_cond & DTRACE_COND_USERMODE) &&
6188 prov->dtpv_pops.dtps_usermode(prov->dtpv_arg,
6189 probe->dtpr_id, probe->dtpr_arg) == 0)
6190 continue;
6191
6192#if defined(sun)
6193 /*
6194 * This is more subtle than it looks. We have to be
6195 * absolutely certain that CRED() isn't going to
6196 * change out from under us so it's only legit to
6197 * examine that structure if we're in constrained
6198 * situations. Currently, the only times we'll this
6199 * check is if a non-super-user has enabled the
6200 * profile or syscall providers -- providers that
6201 * allow visibility of all processes. For the
6202 * profile case, the check above will ensure that
6203 * we're examining a user context.
6204 */
6205 if (ecb->dte_cond & DTRACE_COND_OWNER) {
6206 cred_t *cr;
6207 cred_t *s_cr =
6208 ecb->dte_state->dts_cred.dcr_cred;
6209 proc_t *proc;
6210
6211 ASSERT(s_cr != NULL);
6212
6213 if ((cr = CRED()) == NULL ||
6214 s_cr->cr_uid != cr->cr_uid ||
6215 s_cr->cr_uid != cr->cr_ruid ||
6216 s_cr->cr_uid != cr->cr_suid ||
6217 s_cr->cr_gid != cr->cr_gid ||
6218 s_cr->cr_gid != cr->cr_rgid ||
6219 s_cr->cr_gid != cr->cr_sgid ||
6220 (proc = ttoproc(curthread)) == NULL ||
6221 (proc->p_flag & SNOCD))
6222 continue;
6223 }
6224
6225 if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) {
6226 cred_t *cr;
6227 cred_t *s_cr =
6228 ecb->dte_state->dts_cred.dcr_cred;
6229
6230 ASSERT(s_cr != NULL);
6231
6232 if ((cr = CRED()) == NULL ||
6233 s_cr->cr_zone->zone_id !=
6234 cr->cr_zone->zone_id)
6235 continue;
6236 }
6237#endif
6238 }
6239
6240 if (now - state->dts_alive > dtrace_deadman_timeout) {
6241 /*
6242 * We seem to be dead. Unless we (a) have kernel
6243 * destructive permissions (b) have explicitly enabled
6244 * destructive actions and (c) destructive actions have
6245 * not been disabled, we're going to transition into
6246 * the KILLED state, from which no further processing
6247 * on this state will be performed.
6248 */
6249 if (!dtrace_priv_kernel_destructive(state) ||
6250 !state->dts_cred.dcr_destructive ||
6251 dtrace_destructive_disallow) {
6252 void *activity = &state->dts_activity;
6253 dtrace_activity_t current;
6254
6255 do {
6256 current = state->dts_activity;
6257 } while (dtrace_cas32(activity, current,
6258 DTRACE_ACTIVITY_KILLED) != current);
6259
6260 continue;
6261 }
6262 }
6263
6264 if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed,
6265 ecb->dte_alignment, state, &mstate)) < 0)
6266 continue;
6267
6268 tomax = buf->dtb_tomax;
6269 ASSERT(tomax != NULL);
6270
6271 if (ecb->dte_size != 0) {
6272 dtrace_rechdr_t dtrh;
6273 if (!(mstate.dtms_present & DTRACE_MSTATE_TIMESTAMP)) {
6274 mstate.dtms_timestamp = dtrace_gethrtime();
6275 mstate.dtms_present |= DTRACE_MSTATE_TIMESTAMP;
6276 }
6277 ASSERT3U(ecb->dte_size, >=, sizeof (dtrace_rechdr_t));
6278 dtrh.dtrh_epid = ecb->dte_epid;
6279 DTRACE_RECORD_STORE_TIMESTAMP(&dtrh,
6280 mstate.dtms_timestamp);
6281 *((dtrace_rechdr_t *)(tomax + offs)) = dtrh;
6282 }
6283
6284 mstate.dtms_epid = ecb->dte_epid;
6285 mstate.dtms_present |= DTRACE_MSTATE_EPID;
6286
6287 if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL)
6288 mstate.dtms_access = DTRACE_ACCESS_KERNEL;
6289 else
6290 mstate.dtms_access = 0;
6291
6292 if (pred != NULL) {
6293 dtrace_difo_t *dp = pred->dtp_difo;
6294 int rval;
6295
6296 rval = dtrace_dif_emulate(dp, &mstate, vstate, state);
6297
6298 if (!(*flags & CPU_DTRACE_ERROR) && !rval) {
6299 dtrace_cacheid_t cid = probe->dtpr_predcache;
6300
6301 if (cid != DTRACE_CACHEIDNONE && !onintr) {
6302 /*
6303 * Update the predicate cache...
6304 */
6305 ASSERT(cid == pred->dtp_cacheid);
6306 curthread->t_predcache = cid;
6307 }
6308
6309 continue;
6310 }
6311 }
6312
6313 for (act = ecb->dte_action; !(*flags & CPU_DTRACE_ERROR) &&
6314 act != NULL; act = act->dta_next) {
6315 size_t valoffs;
6316 dtrace_difo_t *dp;
6317 dtrace_recdesc_t *rec = &act->dta_rec;
6318
6319 size = rec->dtrd_size;
6320 valoffs = offs + rec->dtrd_offset;
6321
6322 if (DTRACEACT_ISAGG(act->dta_kind)) {
6323 uint64_t v = 0xbad;
6324 dtrace_aggregation_t *agg;
6325
6326 agg = (dtrace_aggregation_t *)act;
6327
6328 if ((dp = act->dta_difo) != NULL)
6329 v = dtrace_dif_emulate(dp,
6330 &mstate, vstate, state);
6331
6332 if (*flags & CPU_DTRACE_ERROR)
6333 continue;
6334
6335 /*
6336 * Note that we always pass the expression
6337 * value from the previous iteration of the
6338 * action loop. This value will only be used
6339 * if there is an expression argument to the
6340 * aggregating action, denoted by the
6341 * dtag_hasarg field.
6342 */
6343 dtrace_aggregate(agg, buf,
6344 offs, aggbuf, v, val);
6345 continue;
6346 }
6347
6348 switch (act->dta_kind) {
6349 case DTRACEACT_STOP:
6350 if (dtrace_priv_proc_destructive(state))
6351 dtrace_action_stop();
6352 continue;
6353
6354 case DTRACEACT_BREAKPOINT:
6355 if (dtrace_priv_kernel_destructive(state))
6356 dtrace_action_breakpoint(ecb);
6357 continue;
6358
6359 case DTRACEACT_PANIC:
6360 if (dtrace_priv_kernel_destructive(state))
6361 dtrace_action_panic(ecb);
6362 continue;
6363
6364 case DTRACEACT_STACK:
6365 if (!dtrace_priv_kernel(state))
6366 continue;
6367
6368 dtrace_getpcstack((pc_t *)(tomax + valoffs),
6369 size / sizeof (pc_t), probe->dtpr_aframes,
6370 DTRACE_ANCHORED(probe) ? NULL :
6371 (uint32_t *)arg0);
6372 continue;
6373
6374 case DTRACEACT_JSTACK:
6375 case DTRACEACT_USTACK:
6376 if (!dtrace_priv_proc(state))
6377 continue;
6378
6379 /*
6380 * See comment in DIF_VAR_PID.
6381 */
6382 if (DTRACE_ANCHORED(mstate.dtms_probe) &&
6383 CPU_ON_INTR(CPU)) {
6384 int depth = DTRACE_USTACK_NFRAMES(
6385 rec->dtrd_arg) + 1;
6386
6387 dtrace_bzero((void *)(tomax + valoffs),
6388 DTRACE_USTACK_STRSIZE(rec->dtrd_arg)
6389 + depth * sizeof (uint64_t));
6390
6391 continue;
6392 }
6393
6394 if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 &&
6395 curproc->p_dtrace_helpers != NULL) {
6396 /*
6397 * This is the slow path -- we have
6398 * allocated string space, and we're
6399 * getting the stack of a process that
6400 * has helpers. Call into a separate
6401 * routine to perform this processing.
6402 */
6403 dtrace_action_ustack(&mstate, state,
6404 (uint64_t *)(tomax + valoffs),
6405 rec->dtrd_arg);
6406 continue;
6407 }
6408
6409 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6410 dtrace_getupcstack((uint64_t *)
6411 (tomax + valoffs),
6412 DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + 1);
6413 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6414 continue;
6415
6416 default:
6417 break;
6418 }
6419
6420 dp = act->dta_difo;
6421 ASSERT(dp != NULL);
6422
6423 val = dtrace_dif_emulate(dp, &mstate, vstate, state);
6424
6425 if (*flags & CPU_DTRACE_ERROR)
6426 continue;
6427
6428 switch (act->dta_kind) {
6429 case DTRACEACT_SPECULATE: {
6430 dtrace_rechdr_t *dtrh;
6431
6432 ASSERT(buf == &state->dts_buffer[cpuid]);
6433 buf = dtrace_speculation_buffer(state,
6434 cpuid, val);
6435
6436 if (buf == NULL) {
6437 *flags |= CPU_DTRACE_DROP;
6438 continue;
6439 }
6440
6441 offs = dtrace_buffer_reserve(buf,
6442 ecb->dte_needed, ecb->dte_alignment,
6443 state, NULL);
6444
6445 if (offs < 0) {
6446 *flags |= CPU_DTRACE_DROP;
6447 continue;
6448 }
6449
6450 tomax = buf->dtb_tomax;
6451 ASSERT(tomax != NULL);
6452
6453 if (ecb->dte_size == 0)
6454 continue;
6455
6456 ASSERT3U(ecb->dte_size, >=,
6457 sizeof (dtrace_rechdr_t));
6458 dtrh = ((void *)(tomax + offs));
6459 dtrh->dtrh_epid = ecb->dte_epid;
6460 /*
6461 * When the speculation is committed, all of
6462 * the records in the speculative buffer will
6463 * have their timestamps set to the commit
6464 * time. Until then, it is set to a sentinel
6465 * value, for debugability.
6466 */
6467 DTRACE_RECORD_STORE_TIMESTAMP(dtrh, UINT64_MAX);
6468 continue;
6469 }
6470
6471 case DTRACEACT_PRINTM: {
6472 /* The DIF returns a 'memref'. */
6473 uintptr_t *memref = (uintptr_t *)(uintptr_t) val;
6474
6475 /* Get the size from the memref. */
6476 size = memref[1];
6477
6478 /*
6479 * Check if the size exceeds the allocated
6480 * buffer size.
6481 */
6482 if (size + sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) {
6483 /* Flag a drop! */
6484 *flags |= CPU_DTRACE_DROP;
6485 continue;
6486 }
6487
6488 /* Store the size in the buffer first. */
6489 DTRACE_STORE(uintptr_t, tomax,
6490 valoffs, size);
6491
6492 /*
6493 * Offset the buffer address to the start
6494 * of the data.
6495 */
6496 valoffs += sizeof(uintptr_t);
6497
6498 /*
6499 * Reset to the memory address rather than
6500 * the memref array, then let the BYREF
6501 * code below do the work to store the
6502 * memory data in the buffer.
6503 */
6504 val = memref[0];
6505 break;
6506 }
6507
6508 case DTRACEACT_PRINTT: {
6509 /* The DIF returns a 'typeref'. */
6510 uintptr_t *typeref = (uintptr_t *)(uintptr_t) val;
6511 char c = '\0' + 1;
6512 size_t s;
6513
6514 /*
6515 * Get the type string length and round it
6516 * up so that the data that follows is
6517 * aligned for easy access.
6518 */
6519 size_t typs = strlen((char *) typeref[2]) + 1;
6520 typs = roundup(typs, sizeof(uintptr_t));
6521
6522 /*
6523 *Get the size from the typeref using the
6524 * number of elements and the type size.
6525 */
6526 size = typeref[1] * typeref[3];
6527
6528 /*
6529 * Check if the size exceeds the allocated
6530 * buffer size.
6531 */
6532 if (size + typs + 2 * sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) {
6533 /* Flag a drop! */
6534 *flags |= CPU_DTRACE_DROP;
6535
6536 }
6537
6538 /* Store the size in the buffer first. */
6539 DTRACE_STORE(uintptr_t, tomax,
6540 valoffs, size);
6541 valoffs += sizeof(uintptr_t);
6542
6543 /* Store the type size in the buffer. */
6544 DTRACE_STORE(uintptr_t, tomax,
6545 valoffs, typeref[3]);
6546 valoffs += sizeof(uintptr_t);
6547
6548 val = typeref[2];
6549
6550 for (s = 0; s < typs; s++) {
6551 if (c != '\0')
6552 c = dtrace_load8(val++);
6553
6554 DTRACE_STORE(uint8_t, tomax,
6555 valoffs++, c);
6556 }
6557
6558 /*
6559 * Reset to the memory address rather than
6560 * the typeref array, then let the BYREF
6561 * code below do the work to store the
6562 * memory data in the buffer.
6563 */
6564 val = typeref[0];
6565 break;
6566 }
6567
6568 case DTRACEACT_CHILL:
6569 if (dtrace_priv_kernel_destructive(state))
6570 dtrace_action_chill(&mstate, val);
6571 continue;
6572
6573 case DTRACEACT_RAISE:
6574 if (dtrace_priv_proc_destructive(state))
6575 dtrace_action_raise(val);
6576 continue;
6577
6578 case DTRACEACT_COMMIT:
6579 ASSERT(!committed);
6580
6581 /*
6582 * We need to commit our buffer state.
6583 */
6584 if (ecb->dte_size)
6585 buf->dtb_offset = offs + ecb->dte_size;
6586 buf = &state->dts_buffer[cpuid];
6587 dtrace_speculation_commit(state, cpuid, val);
6588 committed = 1;
6589 continue;
6590
6591 case DTRACEACT_DISCARD:
6592 dtrace_speculation_discard(state, cpuid, val);
6593 continue;
6594
6595 case DTRACEACT_DIFEXPR:
6596 case DTRACEACT_LIBACT:
6597 case DTRACEACT_PRINTF:
6598 case DTRACEACT_PRINTA:
6599 case DTRACEACT_SYSTEM:
6600 case DTRACEACT_FREOPEN:
6601 case DTRACEACT_TRACEMEM:
6602 break;
6603
6604 case DTRACEACT_TRACEMEM_DYNSIZE:
6605 tracememsize = val;
6606 break;
6607
6608 case DTRACEACT_SYM:
6609 case DTRACEACT_MOD:
6610 if (!dtrace_priv_kernel(state))
6611 continue;
6612 break;
6613
6614 case DTRACEACT_USYM:
6615 case DTRACEACT_UMOD:
6616 case DTRACEACT_UADDR: {
6617#if defined(sun)
6618 struct pid *pid = curthread->t_procp->p_pidp;
6619#endif
6620
6621 if (!dtrace_priv_proc(state))
6622 continue;
6623
6624 DTRACE_STORE(uint64_t, tomax,
6625#if defined(sun)
6626 valoffs, (uint64_t)pid->pid_id);
6627#else
6628 valoffs, (uint64_t) curproc->p_pid);
6629#endif
6630 DTRACE_STORE(uint64_t, tomax,
6631 valoffs + sizeof (uint64_t), val);
6632
6633 continue;
6634 }
6635
6636 case DTRACEACT_EXIT: {
6637 /*
6638 * For the exit action, we are going to attempt
6639 * to atomically set our activity to be
6640 * draining. If this fails (either because
6641 * another CPU has beat us to the exit action,
6642 * or because our current activity is something
6643 * other than ACTIVE or WARMUP), we will
6644 * continue. This assures that the exit action
6645 * can be successfully recorded at most once
6646 * when we're in the ACTIVE state. If we're
6647 * encountering the exit() action while in
6648 * COOLDOWN, however, we want to honor the new
6649 * status code. (We know that we're the only
6650 * thread in COOLDOWN, so there is no race.)
6651 */
6652 void *activity = &state->dts_activity;
6653 dtrace_activity_t current = state->dts_activity;
6654
6655 if (current == DTRACE_ACTIVITY_COOLDOWN)
6656 break;
6657
6658 if (current != DTRACE_ACTIVITY_WARMUP)
6659 current = DTRACE_ACTIVITY_ACTIVE;
6660
6661 if (dtrace_cas32(activity, current,
6662 DTRACE_ACTIVITY_DRAINING) != current) {
6663 *flags |= CPU_DTRACE_DROP;
6664 continue;
6665 }
6666
6667 break;
6668 }
6669
6670 default:
6671 ASSERT(0);
6672 }
6673
6674 if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF) {
6675 uintptr_t end = valoffs + size;
6676
6677 if (tracememsize != 0 &&
6678 valoffs + tracememsize < end) {
6679 end = valoffs + tracememsize;
6680 tracememsize = 0;
6681 }
6682
6683 if (!dtrace_vcanload((void *)(uintptr_t)val,
6684 &dp->dtdo_rtype, &mstate, vstate))
6685 continue;
6686
6687 /*
6688 * If this is a string, we're going to only
6689 * load until we find the zero byte -- after
6690 * which we'll store zero bytes.
6691 */
6692 if (dp->dtdo_rtype.dtdt_kind ==
6693 DIF_TYPE_STRING) {
6694 char c = '\0' + 1;
6695 int intuple = act->dta_intuple;
6696 size_t s;
6697
6698 for (s = 0; s < size; s++) {
6699 if (c != '\0')
6700 c = dtrace_load8(val++);
6701
6702 DTRACE_STORE(uint8_t, tomax,
6703 valoffs++, c);
6704
6705 if (c == '\0' && intuple)
6706 break;
6707 }
6708
6709 continue;
6710 }
6711
6712 while (valoffs < end) {
6713 DTRACE_STORE(uint8_t, tomax, valoffs++,
6714 dtrace_load8(val++));
6715 }
6716
6717 continue;
6718 }
6719
6720 switch (size) {
6721 case 0:
6722 break;
6723
6724 case sizeof (uint8_t):
6725 DTRACE_STORE(uint8_t, tomax, valoffs, val);
6726 break;
6727 case sizeof (uint16_t):
6728 DTRACE_STORE(uint16_t, tomax, valoffs, val);
6729 break;
6730 case sizeof (uint32_t):
6731 DTRACE_STORE(uint32_t, tomax, valoffs, val);
6732 break;
6733 case sizeof (uint64_t):
6734 DTRACE_STORE(uint64_t, tomax, valoffs, val);
6735 break;
6736 default:
6737 /*
6738 * Any other size should have been returned by
6739 * reference, not by value.
6740 */
6741 ASSERT(0);
6742 break;
6743 }
6744 }
6745
6746 if (*flags & CPU_DTRACE_DROP)
6747 continue;
6748
6749 if (*flags & CPU_DTRACE_FAULT) {
6750 int ndx;
6751 dtrace_action_t *err;
6752
6753 buf->dtb_errors++;
6754
6755 if (probe->dtpr_id == dtrace_probeid_error) {
6756 /*
6757 * There's nothing we can do -- we had an
6758 * error on the error probe. We bump an
6759 * error counter to at least indicate that
6760 * this condition happened.
6761 */
6762 dtrace_error(&state->dts_dblerrors);
6763 continue;
6764 }
6765
6766 if (vtime) {
6767 /*
6768 * Before recursing on dtrace_probe(), we
6769 * need to explicitly clear out our start
6770 * time to prevent it from being accumulated
6771 * into t_dtrace_vtime.
6772 */
6773 curthread->t_dtrace_start = 0;
6774 }
6775
6776 /*
6777 * Iterate over the actions to figure out which action
6778 * we were processing when we experienced the error.
6779 * Note that act points _past_ the faulting action; if
6780 * act is ecb->dte_action, the fault was in the
6781 * predicate, if it's ecb->dte_action->dta_next it's
6782 * in action #1, and so on.
6783 */
6784 for (err = ecb->dte_action, ndx = 0;
6785 err != act; err = err->dta_next, ndx++)
6786 continue;
6787
6788 dtrace_probe_error(state, ecb->dte_epid, ndx,
6789 (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) ?
6790 mstate.dtms_fltoffs : -1, DTRACE_FLAGS2FLT(*flags),
6791 cpu_core[cpuid].cpuc_dtrace_illval);
6792
6793 continue;
6794 }
6795
6796 if (!committed)
6797 buf->dtb_offset = offs + ecb->dte_size;
6798 }
6799
6800 if (vtime)
6801 curthread->t_dtrace_start = dtrace_gethrtime();
6802
6803 dtrace_interrupt_enable(cookie);
6804}
6805
6806/*
6807 * DTrace Probe Hashing Functions
6808 *
6809 * The functions in this section (and indeed, the functions in remaining
6810 * sections) are not _called_ from probe context. (Any exceptions to this are
6811 * marked with a "Note:".) Rather, they are called from elsewhere in the
6812 * DTrace framework to look-up probes in, add probes to and remove probes from
6813 * the DTrace probe hashes. (Each probe is hashed by each element of the
6814 * probe tuple -- allowing for fast lookups, regardless of what was
6815 * specified.)
6816 */
6817static uint_t
6818dtrace_hash_str(const char *p)
6819{
6820 unsigned int g;
6821 uint_t hval = 0;
6822
6823 while (*p) {
6824 hval = (hval << 4) + *p++;
6825 if ((g = (hval & 0xf0000000)) != 0)
6826 hval ^= g >> 24;
6827 hval &= ~g;
6828 }
6829 return (hval);
6830}
6831
6832static dtrace_hash_t *
6833dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, uintptr_t prevoffs)
6834{
6835 dtrace_hash_t *hash = kmem_zalloc(sizeof (dtrace_hash_t), KM_SLEEP);
6836
6837 hash->dth_stroffs = stroffs;
6838 hash->dth_nextoffs = nextoffs;
6839 hash->dth_prevoffs = prevoffs;
6840
6841 hash->dth_size = 1;
6842 hash->dth_mask = hash->dth_size - 1;
6843
6844 hash->dth_tab = kmem_zalloc(hash->dth_size *
6845 sizeof (dtrace_hashbucket_t *), KM_SLEEP);
6846
6847 return (hash);
6848}
6849
6850static void
6851dtrace_hash_destroy(dtrace_hash_t *hash)
6852{
6853#ifdef DEBUG
6854 int i;
6855
6856 for (i = 0; i < hash->dth_size; i++)
6857 ASSERT(hash->dth_tab[i] == NULL);
6858#endif
6859
6860 kmem_free(hash->dth_tab,
6861 hash->dth_size * sizeof (dtrace_hashbucket_t *));
6862 kmem_free(hash, sizeof (dtrace_hash_t));
6863}
6864
6865static void
6866dtrace_hash_resize(dtrace_hash_t *hash)
6867{
6868 int size = hash->dth_size, i, ndx;
6869 int new_size = hash->dth_size << 1;
6870 int new_mask = new_size - 1;
6871 dtrace_hashbucket_t **new_tab, *bucket, *next;
6872
6873 ASSERT((new_size & new_mask) == 0);
6874
6875 new_tab = kmem_zalloc(new_size * sizeof (void *), KM_SLEEP);
6876
6877 for (i = 0; i < size; i++) {
6878 for (bucket = hash->dth_tab[i]; bucket != NULL; bucket = next) {
6879 dtrace_probe_t *probe = bucket->dthb_chain;
6880
6881 ASSERT(probe != NULL);
6882 ndx = DTRACE_HASHSTR(hash, probe) & new_mask;
6883
6884 next = bucket->dthb_next;
6885 bucket->dthb_next = new_tab[ndx];
6886 new_tab[ndx] = bucket;
6887 }
6888 }
6889
6890 kmem_free(hash->dth_tab, hash->dth_size * sizeof (void *));
6891 hash->dth_tab = new_tab;
6892 hash->dth_size = new_size;
6893 hash->dth_mask = new_mask;
6894}
6895
6896static void
6897dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new)
6898{
6899 int hashval = DTRACE_HASHSTR(hash, new);
6900 int ndx = hashval & hash->dth_mask;
6901 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6902 dtrace_probe_t **nextp, **prevp;
6903
6904 for (; bucket != NULL; bucket = bucket->dthb_next) {
6905 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new))
6906 goto add;
6907 }
6908
6909 if ((hash->dth_nbuckets >> 1) > hash->dth_size) {
6910 dtrace_hash_resize(hash);
6911 dtrace_hash_add(hash, new);
6912 return;
6913 }
6914
6915 bucket = kmem_zalloc(sizeof (dtrace_hashbucket_t), KM_SLEEP);
6916 bucket->dthb_next = hash->dth_tab[ndx];
6917 hash->dth_tab[ndx] = bucket;
6918 hash->dth_nbuckets++;
6919
6920add:
6921 nextp = DTRACE_HASHNEXT(hash, new);
6922 ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL);
6923 *nextp = bucket->dthb_chain;
6924
6925 if (bucket->dthb_chain != NULL) {
6926 prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain);
6927 ASSERT(*prevp == NULL);
6928 *prevp = new;
6929 }
6930
6931 bucket->dthb_chain = new;
6932 bucket->dthb_len++;
6933}
6934
6935static dtrace_probe_t *
6936dtrace_hash_lookup(dtrace_hash_t *hash, dtrace_probe_t *template)
6937{
6938 int hashval = DTRACE_HASHSTR(hash, template);
6939 int ndx = hashval & hash->dth_mask;
6940 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6941
6942 for (; bucket != NULL; bucket = bucket->dthb_next) {
6943 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template))
6944 return (bucket->dthb_chain);
6945 }
6946
6947 return (NULL);
6948}
6949
6950static int
6951dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template)
6952{
6953 int hashval = DTRACE_HASHSTR(hash, template);
6954 int ndx = hashval & hash->dth_mask;
6955 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6956
6957 for (; bucket != NULL; bucket = bucket->dthb_next) {
6958 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template))
6959 return (bucket->dthb_len);
6960 }
6961
6962 return (0);
6963}
6964
6965static void
6966dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe)
6967{
6968 int ndx = DTRACE_HASHSTR(hash, probe) & hash->dth_mask;
6969 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6970
6971 dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe);
6972 dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe);
6973
6974 /*
6975 * Find the bucket that we're removing this probe from.
6976 */
6977 for (; bucket != NULL; bucket = bucket->dthb_next) {
6978 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe))
6979 break;
6980 }
6981
6982 ASSERT(bucket != NULL);
6983
6984 if (*prevp == NULL) {
6985 if (*nextp == NULL) {
6986 /*
6987 * The removed probe was the only probe on this
6988 * bucket; we need to remove the bucket.
6989 */
6990 dtrace_hashbucket_t *b = hash->dth_tab[ndx];
6991
6992 ASSERT(bucket->dthb_chain == probe);
6993 ASSERT(b != NULL);
6994
6995 if (b == bucket) {
6996 hash->dth_tab[ndx] = bucket->dthb_next;
6997 } else {
6998 while (b->dthb_next != bucket)
6999 b = b->dthb_next;
7000 b->dthb_next = bucket->dthb_next;
7001 }
7002
7003 ASSERT(hash->dth_nbuckets > 0);
7004 hash->dth_nbuckets--;
7005 kmem_free(bucket, sizeof (dtrace_hashbucket_t));
7006 return;
7007 }
7008
7009 bucket->dthb_chain = *nextp;
7010 } else {
7011 *(DTRACE_HASHNEXT(hash, *prevp)) = *nextp;
7012 }
7013
7014 if (*nextp != NULL)
7015 *(DTRACE_HASHPREV(hash, *nextp)) = *prevp;
7016}
7017
7018/*
7019 * DTrace Utility Functions
7020 *
7021 * These are random utility functions that are _not_ called from probe context.
7022 */
7023static int
7024dtrace_badattr(const dtrace_attribute_t *a)
7025{
7026 return (a->dtat_name > DTRACE_STABILITY_MAX ||
7027 a->dtat_data > DTRACE_STABILITY_MAX ||
7028 a->dtat_class > DTRACE_CLASS_MAX);
7029}
7030
7031/*
7032 * Return a duplicate copy of a string. If the specified string is NULL,
7033 * this function returns a zero-length string.
7034 */
7035static char *
7036dtrace_strdup(const char *str)
7037{
7038 char *new = kmem_zalloc((str != NULL ? strlen(str) : 0) + 1, KM_SLEEP);
7039
7040 if (str != NULL)
7041 (void) strcpy(new, str);
7042
7043 return (new);
7044}
7045
7046#define DTRACE_ISALPHA(c) \
7047 (((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z'))
7048
7049static int
7050dtrace_badname(const char *s)
7051{
7052 char c;
7053
7054 if (s == NULL || (c = *s++) == '\0')
7055 return (0);
7056
7057 if (!DTRACE_ISALPHA(c) && c != '-' && c != '_' && c != '.')
7058 return (1);
7059
7060 while ((c = *s++) != '\0') {
7061 if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') &&
7062 c != '-' && c != '_' && c != '.' && c != '`')
7063 return (1);
7064 }
7065
7066 return (0);
7067}
7068
7069static void
7070dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp)
7071{
7072 uint32_t priv;
7073
7074#if defined(sun)
7075 if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) {
7076 /*
7077 * For DTRACE_PRIV_ALL, the uid and zoneid don't matter.
7078 */
7079 priv = DTRACE_PRIV_ALL;
7080 } else {
7081 *uidp = crgetuid(cr);
7082 *zoneidp = crgetzoneid(cr);
7083
7084 priv = 0;
7085 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE))
7086 priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER;
7087 else if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE))
7088 priv |= DTRACE_PRIV_USER;
7089 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE))
7090 priv |= DTRACE_PRIV_PROC;
7091 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
7092 priv |= DTRACE_PRIV_OWNER;
7093 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
7094 priv |= DTRACE_PRIV_ZONEOWNER;
7095 }
7096#else
7097 priv = DTRACE_PRIV_ALL;
7098#endif
7099
7100 *privp = priv;
7101}
7102
7103#ifdef DTRACE_ERRDEBUG
7104static void
7105dtrace_errdebug(const char *str)
7106{
7107 int hval = dtrace_hash_str(str) % DTRACE_ERRHASHSZ;
7108 int occupied = 0;
7109
7110 mutex_enter(&dtrace_errlock);
7111 dtrace_errlast = str;
7112 dtrace_errthread = curthread;
7113
7114 while (occupied++ < DTRACE_ERRHASHSZ) {
7115 if (dtrace_errhash[hval].dter_msg == str) {
7116 dtrace_errhash[hval].dter_count++;
7117 goto out;
7118 }
7119
7120 if (dtrace_errhash[hval].dter_msg != NULL) {
7121 hval = (hval + 1) % DTRACE_ERRHASHSZ;
7122 continue;
7123 }
7124
7125 dtrace_errhash[hval].dter_msg = str;
7126 dtrace_errhash[hval].dter_count = 1;
7127 goto out;
7128 }
7129
7130 panic("dtrace: undersized error hash");
7131out:
7132 mutex_exit(&dtrace_errlock);
7133}
7134#endif
7135
7136/*
7137 * DTrace Matching Functions
7138 *
7139 * These functions are used to match groups of probes, given some elements of
7140 * a probe tuple, or some globbed expressions for elements of a probe tuple.
7141 */
7142static int
7143dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid,
7144 zoneid_t zoneid)
7145{
7146 if (priv != DTRACE_PRIV_ALL) {
7147 uint32_t ppriv = prp->dtpr_provider->dtpv_priv.dtpp_flags;
7148 uint32_t match = priv & ppriv;
7149
7150 /*
7151 * No PRIV_DTRACE_* privileges...
7152 */
7153 if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER |
7154 DTRACE_PRIV_KERNEL)) == 0)
7155 return (0);
7156
7157 /*
7158 * No matching bits, but there were bits to match...
7159 */
7160 if (match == 0 && ppriv != 0)
7161 return (0);
7162
7163 /*
7164 * Need to have permissions to the process, but don't...
7165 */
7166 if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 &&
7167 uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) {
7168 return (0);
7169 }
7170
7171 /*
7172 * Need to be in the same zone unless we possess the
7173 * privilege to examine all zones.
7174 */
7175 if (((ppriv & ~match) & DTRACE_PRIV_ZONEOWNER) != 0 &&
7176 zoneid != prp->dtpr_provider->dtpv_priv.dtpp_zoneid) {
7177 return (0);
7178 }
7179 }
7180
7181 return (1);
7182}
7183
7184/*
7185 * dtrace_match_probe compares a dtrace_probe_t to a pre-compiled key, which
7186 * consists of input pattern strings and an ops-vector to evaluate them.
7187 * This function returns >0 for match, 0 for no match, and <0 for error.
7188 */
7189static int
7190dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp,
7191 uint32_t priv, uid_t uid, zoneid_t zoneid)
7192{
7193 dtrace_provider_t *pvp = prp->dtpr_provider;
7194 int rv;
7195
7196 if (pvp->dtpv_defunct)
7197 return (0);
7198
7199 if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0)
7200 return (rv);
7201
7202 if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0)
7203 return (rv);
7204
7205 if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0)
7206 return (rv);
7207
7208 if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0)
7209 return (rv);
7210
7211 if (dtrace_match_priv(prp, priv, uid, zoneid) == 0)
7212 return (0);
7213
7214 return (rv);
7215}
7216
7217/*
7218 * dtrace_match_glob() is a safe kernel implementation of the gmatch(3GEN)
7219 * interface for matching a glob pattern 'p' to an input string 's'. Unlike
7220 * libc's version, the kernel version only applies to 8-bit ASCII strings.
7221 * In addition, all of the recursion cases except for '*' matching have been
7222 * unwound. For '*', we still implement recursive evaluation, but a depth
7223 * counter is maintained and matching is aborted if we recurse too deep.
7224 * The function returns 0 if no match, >0 if match, and <0 if recursion error.
7225 */
7226static int
7227dtrace_match_glob(const char *s, const char *p, int depth)
7228{
7229 const char *olds;
7230 char s1, c;
7231 int gs;
7232
7233 if (depth > DTRACE_PROBEKEY_MAXDEPTH)
7234 return (-1);
7235
7236 if (s == NULL)
7237 s = ""; /* treat NULL as empty string */
7238
7239top:
7240 olds = s;
7241 s1 = *s++;
7242
7243 if (p == NULL)
7244 return (0);
7245
7246 if ((c = *p++) == '\0')
7247 return (s1 == '\0');
7248
7249 switch (c) {
7250 case '[': {
7251 int ok = 0, notflag = 0;
7252 char lc = '\0';
7253
7254 if (s1 == '\0')
7255 return (0);
7256
7257 if (*p == '!') {
7258 notflag = 1;
7259 p++;
7260 }
7261
7262 if ((c = *p++) == '\0')
7263 return (0);
7264
7265 do {
7266 if (c == '-' && lc != '\0' && *p != ']') {
7267 if ((c = *p++) == '\0')
7268 return (0);
7269 if (c == '\\' && (c = *p++) == '\0')
7270 return (0);
7271
7272 if (notflag) {
7273 if (s1 < lc || s1 > c)
7274 ok++;
7275 else
7276 return (0);
7277 } else if (lc <= s1 && s1 <= c)
7278 ok++;
7279
7280 } else if (c == '\\' && (c = *p++) == '\0')
7281 return (0);
7282
7283 lc = c; /* save left-hand 'c' for next iteration */
7284
7285 if (notflag) {
7286 if (s1 != c)
7287 ok++;
7288 else
7289 return (0);
7290 } else if (s1 == c)
7291 ok++;
7292
7293 if ((c = *p++) == '\0')
7294 return (0);
7295
7296 } while (c != ']');
7297
7298 if (ok)
7299 goto top;
7300
7301 return (0);
7302 }
7303
7304 case '\\':
7305 if ((c = *p++) == '\0')
7306 return (0);
7307 /*FALLTHRU*/
7308
7309 default:
7310 if (c != s1)
7311 return (0);
7312 /*FALLTHRU*/
7313
7314 case '?':
7315 if (s1 != '\0')
7316 goto top;
7317 return (0);
7318
7319 case '*':
7320 while (*p == '*')
7321 p++; /* consecutive *'s are identical to a single one */
7322
7323 if (*p == '\0')
7324 return (1);
7325
7326 for (s = olds; *s != '\0'; s++) {
7327 if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0)
7328 return (gs);
7329 }
7330
7331 return (0);
7332 }
7333}
7334
7335/*ARGSUSED*/
7336static int
7337dtrace_match_string(const char *s, const char *p, int depth)
7338{
7339 return (s != NULL && strcmp(s, p) == 0);
7340}
7341
7342/*ARGSUSED*/
7343static int
7344dtrace_match_nul(const char *s, const char *p, int depth)
7345{
7346 return (1); /* always match the empty pattern */
7347}
7348
7349/*ARGSUSED*/
7350static int
7351dtrace_match_nonzero(const char *s, const char *p, int depth)
7352{
7353 return (s != NULL && s[0] != '\0');
7354}
7355
7356static int
7357dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid,
7358 zoneid_t zoneid, int (*matched)(dtrace_probe_t *, void *), void *arg)
7359{
7360 dtrace_probe_t template, *probe;
7361 dtrace_hash_t *hash = NULL;
7362 int len, best = INT_MAX, nmatched = 0;
7363 dtrace_id_t i;
7364
7365 ASSERT(MUTEX_HELD(&dtrace_lock));
7366
7367 /*
7368 * If the probe ID is specified in the key, just lookup by ID and
7369 * invoke the match callback once if a matching probe is found.
7370 */
7371 if (pkp->dtpk_id != DTRACE_IDNONE) {
7372 if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL &&
7373 dtrace_match_probe(probe, pkp, priv, uid, zoneid) > 0) {
7374 (void) (*matched)(probe, arg);
7375 nmatched++;
7376 }
7377 return (nmatched);
7378 }
7379
7380 template.dtpr_mod = (char *)pkp->dtpk_mod;
7381 template.dtpr_func = (char *)pkp->dtpk_func;
7382 template.dtpr_name = (char *)pkp->dtpk_name;
7383
7384 /*
7385 * We want to find the most distinct of the module name, function
7386 * name, and name. So for each one that is not a glob pattern or
7387 * empty string, we perform a lookup in the corresponding hash and
7388 * use the hash table with the fewest collisions to do our search.
7389 */
7390 if (pkp->dtpk_mmatch == &dtrace_match_string &&
7391 (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) {
7392 best = len;
7393 hash = dtrace_bymod;
7394 }
7395
7396 if (pkp->dtpk_fmatch == &dtrace_match_string &&
7397 (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) {
7398 best = len;
7399 hash = dtrace_byfunc;
7400 }
7401
7402 if (pkp->dtpk_nmatch == &dtrace_match_string &&
7403 (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) {
7404 best = len;
7405 hash = dtrace_byname;
7406 }
7407
7408 /*
7409 * If we did not select a hash table, iterate over every probe and
7410 * invoke our callback for each one that matches our input probe key.
7411 */
7412 if (hash == NULL) {
7413 for (i = 0; i < dtrace_nprobes; i++) {
7414 if ((probe = dtrace_probes[i]) == NULL ||
7415 dtrace_match_probe(probe, pkp, priv, uid,
7416 zoneid) <= 0)
7417 continue;
7418
7419 nmatched++;
7420
7421 if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT)
7422 break;
7423 }
7424
7425 return (nmatched);
7426 }
7427
7428 /*
7429 * If we selected a hash table, iterate over each probe of the same key
7430 * name and invoke the callback for every probe that matches the other
7431 * attributes of our input probe key.
7432 */
7433 for (probe = dtrace_hash_lookup(hash, &template); probe != NULL;
7434 probe = *(DTRACE_HASHNEXT(hash, probe))) {
7435
7436 if (dtrace_match_probe(probe, pkp, priv, uid, zoneid) <= 0)
7437 continue;
7438
7439 nmatched++;
7440
7441 if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT)
7442 break;
7443 }
7444
7445 return (nmatched);
7446}
7447
7448/*
7449 * Return the function pointer dtrace_probecmp() should use to compare the
7450 * specified pattern with a string. For NULL or empty patterns, we select
7451 * dtrace_match_nul(). For glob pattern strings, we use dtrace_match_glob().
7452 * For non-empty non-glob strings, we use dtrace_match_string().
7453 */
7454static dtrace_probekey_f *
7455dtrace_probekey_func(const char *p)
7456{
7457 char c;
7458
7459 if (p == NULL || *p == '\0')
7460 return (&dtrace_match_nul);
7461
7462 while ((c = *p++) != '\0') {
7463 if (c == '[' || c == '?' || c == '*' || c == '\\')
7464 return (&dtrace_match_glob);
7465 }
7466
7467 return (&dtrace_match_string);
7468}
7469
7470/*
7471 * Build a probe comparison key for use with dtrace_match_probe() from the
7472 * given probe description. By convention, a null key only matches anchored
7473 * probes: if each field is the empty string, reset dtpk_fmatch to
7474 * dtrace_match_nonzero().
7475 */
7476static void
7477dtrace_probekey(dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp)
7478{
7479 pkp->dtpk_prov = pdp->dtpd_provider;
7480 pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider);
7481
7482 pkp->dtpk_mod = pdp->dtpd_mod;
7483 pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod);
7484
7485 pkp->dtpk_func = pdp->dtpd_func;
7486 pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func);
7487
7488 pkp->dtpk_name = pdp->dtpd_name;
7489 pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name);
7490
7491 pkp->dtpk_id = pdp->dtpd_id;
7492
7493 if (pkp->dtpk_id == DTRACE_IDNONE &&
7494 pkp->dtpk_pmatch == &dtrace_match_nul &&
7495 pkp->dtpk_mmatch == &dtrace_match_nul &&
7496 pkp->dtpk_fmatch == &dtrace_match_nul &&
7497 pkp->dtpk_nmatch == &dtrace_match_nul)
7498 pkp->dtpk_fmatch = &dtrace_match_nonzero;
7499}
7500
7501/*
7502 * DTrace Provider-to-Framework API Functions
7503 *
7504 * These functions implement much of the Provider-to-Framework API, as
7505 * described in <sys/dtrace.h>. The parts of the API not in this section are
7506 * the functions in the API for probe management (found below), and
7507 * dtrace_probe() itself (found above).
7508 */
7509
7510/*
7511 * Register the calling provider with the DTrace framework. This should
7512 * generally be called by DTrace providers in their attach(9E) entry point.
7513 */
7514int
7515dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv,
7516 cred_t *cr, const dtrace_pops_t *pops, void *arg, dtrace_provider_id_t *idp)
7517{
7518 dtrace_provider_t *provider;
7519
7520 if (name == NULL || pap == NULL || pops == NULL || idp == NULL) {
7521 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7522 "arguments", name ? name : "<NULL>");
7523 return (EINVAL);
7524 }
7525
7526 if (name[0] == '\0' || dtrace_badname(name)) {
7527 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7528 "provider name", name);
7529 return (EINVAL);
7530 }
7531
7532 if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) ||
7533 pops->dtps_enable == NULL || pops->dtps_disable == NULL ||
7534 pops->dtps_destroy == NULL ||
7535 ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) {
7536 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7537 "provider ops", name);
7538 return (EINVAL);
7539 }
7540
7541 if (dtrace_badattr(&pap->dtpa_provider) ||
7542 dtrace_badattr(&pap->dtpa_mod) ||
7543 dtrace_badattr(&pap->dtpa_func) ||
7544 dtrace_badattr(&pap->dtpa_name) ||
7545 dtrace_badattr(&pap->dtpa_args)) {
7546 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7547 "provider attributes", name);
7548 return (EINVAL);
7549 }
7550
7551 if (priv & ~DTRACE_PRIV_ALL) {
7552 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7553 "privilege attributes", name);
7554 return (EINVAL);
7555 }
7556
7557 if ((priv & DTRACE_PRIV_KERNEL) &&
7558 (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) &&
7559 pops->dtps_usermode == NULL) {
7560 cmn_err(CE_WARN, "failed to register provider '%s': need "
7561 "dtps_usermode() op for given privilege attributes", name);
7562 return (EINVAL);
7563 }
7564
7565 provider = kmem_zalloc(sizeof (dtrace_provider_t), KM_SLEEP);
7566 provider->dtpv_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
7567 (void) strcpy(provider->dtpv_name, name);
7568
7569 provider->dtpv_attr = *pap;
7570 provider->dtpv_priv.dtpp_flags = priv;
7571 if (cr != NULL) {
7572 provider->dtpv_priv.dtpp_uid = crgetuid(cr);
7573 provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr);
7574 }
7575 provider->dtpv_pops = *pops;
7576
7577 if (pops->dtps_provide == NULL) {
7578 ASSERT(pops->dtps_provide_module != NULL);
7579 provider->dtpv_pops.dtps_provide =
7580 (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop;
7581 }
7582
7583 if (pops->dtps_provide_module == NULL) {
7584 ASSERT(pops->dtps_provide != NULL);
7585 provider->dtpv_pops.dtps_provide_module =
7586 (void (*)(void *, modctl_t *))dtrace_nullop;
7587 }
7588
7589 if (pops->dtps_suspend == NULL) {
7590 ASSERT(pops->dtps_resume == NULL);
7591 provider->dtpv_pops.dtps_suspend =
7592 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop;
7593 provider->dtpv_pops.dtps_resume =
7594 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop;
7595 }
7596
7597 provider->dtpv_arg = arg;
7598 *idp = (dtrace_provider_id_t)provider;
7599
7600 if (pops == &dtrace_provider_ops) {
7601 ASSERT(MUTEX_HELD(&dtrace_provider_lock));
7602 ASSERT(MUTEX_HELD(&dtrace_lock));
7603 ASSERT(dtrace_anon.dta_enabling == NULL);
7604
7605 /*
7606 * We make sure that the DTrace provider is at the head of
7607 * the provider chain.
7608 */
7609 provider->dtpv_next = dtrace_provider;
7610 dtrace_provider = provider;
7611 return (0);
7612 }
7613
7614 mutex_enter(&dtrace_provider_lock);
7615 mutex_enter(&dtrace_lock);
7616
7617 /*
7618 * If there is at least one provider registered, we'll add this
7619 * provider after the first provider.
7620 */
7621 if (dtrace_provider != NULL) {
7622 provider->dtpv_next = dtrace_provider->dtpv_next;
7623 dtrace_provider->dtpv_next = provider;
7624 } else {
7625 dtrace_provider = provider;
7626 }
7627
7628 if (dtrace_retained != NULL) {
7629 dtrace_enabling_provide(provider);
7630
7631 /*
7632 * Now we need to call dtrace_enabling_matchall() -- which
7633 * will acquire cpu_lock and dtrace_lock. We therefore need
7634 * to drop all of our locks before calling into it...
7635 */
7636 mutex_exit(&dtrace_lock);
7637 mutex_exit(&dtrace_provider_lock);
7638 dtrace_enabling_matchall();
7639
7640 return (0);
7641 }
7642
7643 mutex_exit(&dtrace_lock);
7644 mutex_exit(&dtrace_provider_lock);
7645
7646 return (0);
7647}
7648
7649/*
7650 * Unregister the specified provider from the DTrace framework. This should
7651 * generally be called by DTrace providers in their detach(9E) entry point.
7652 */
7653int
7654dtrace_unregister(dtrace_provider_id_t id)
7655{
7656 dtrace_provider_t *old = (dtrace_provider_t *)id;
7657 dtrace_provider_t *prev = NULL;
7658 int i, self = 0, noreap = 0;
7659 dtrace_probe_t *probe, *first = NULL;
7660
7661 if (old->dtpv_pops.dtps_enable ==
7662 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop) {
7663 /*
7664 * If DTrace itself is the provider, we're called with locks
7665 * already held.
7666 */
7667 ASSERT(old == dtrace_provider);
7668#if defined(sun)
7669 ASSERT(dtrace_devi != NULL);
7670#endif
7671 ASSERT(MUTEX_HELD(&dtrace_provider_lock));
7672 ASSERT(MUTEX_HELD(&dtrace_lock));
7673 self = 1;
7674
7675 if (dtrace_provider->dtpv_next != NULL) {
7676 /*
7677 * There's another provider here; return failure.
7678 */
7679 return (EBUSY);
7680 }
7681 } else {
7682 mutex_enter(&dtrace_provider_lock);
7683#if defined(sun)
7684 mutex_enter(&mod_lock);
7685#endif
7686 mutex_enter(&dtrace_lock);
7687 }
7688
7689 /*
7690 * If anyone has /dev/dtrace open, or if there are anonymous enabled
7691 * probes, we refuse to let providers slither away, unless this
7692 * provider has already been explicitly invalidated.
7693 */
7694 if (!old->dtpv_defunct &&
7695 (dtrace_opens || (dtrace_anon.dta_state != NULL &&
7696 dtrace_anon.dta_state->dts_necbs > 0))) {
7697 if (!self) {
7698 mutex_exit(&dtrace_lock);
7699#if defined(sun)
7700 mutex_exit(&mod_lock);
7701#endif
7702 mutex_exit(&dtrace_provider_lock);
7703 }
7704 return (EBUSY);
7705 }
7706
7707 /*
7708 * Attempt to destroy the probes associated with this provider.
7709 */
7710 for (i = 0; i < dtrace_nprobes; i++) {
7711 if ((probe = dtrace_probes[i]) == NULL)
7712 continue;
7713
7714 if (probe->dtpr_provider != old)
7715 continue;
7716
7717 if (probe->dtpr_ecb == NULL)
7718 continue;
7719
7720 /*
7721 * If we are trying to unregister a defunct provider, and the
7722 * provider was made defunct within the interval dictated by
7723 * dtrace_unregister_defunct_reap, we'll (asynchronously)
7724 * attempt to reap our enablings. To denote that the provider
7725 * should reattempt to unregister itself at some point in the
7726 * future, we will return a differentiable error code (EAGAIN
7727 * instead of EBUSY) in this case.
7728 */
7729 if (dtrace_gethrtime() - old->dtpv_defunct >
7730 dtrace_unregister_defunct_reap)
7731 noreap = 1;
7732
7733 if (!self) {
7734 mutex_exit(&dtrace_lock);
7735#if defined(sun)
7736 mutex_exit(&mod_lock);
7737#endif
7738 mutex_exit(&dtrace_provider_lock);
7739 }
7740
7741 if (noreap)
7742 return (EBUSY);
7743
7744 (void) taskq_dispatch(dtrace_taskq,
7745 (task_func_t *)dtrace_enabling_reap, NULL, TQ_SLEEP);
7746
7747 return (EAGAIN);
7748 }
7749
7750 /*
7751 * All of the probes for this provider are disabled; we can safely
7752 * remove all of them from their hash chains and from the probe array.
7753 */
7754 for (i = 0; i < dtrace_nprobes; i++) {
7755 if ((probe = dtrace_probes[i]) == NULL)
7756 continue;
7757
7758 if (probe->dtpr_provider != old)
7759 continue;
7760
7761 dtrace_probes[i] = NULL;
7762
7763 dtrace_hash_remove(dtrace_bymod, probe);
7764 dtrace_hash_remove(dtrace_byfunc, probe);
7765 dtrace_hash_remove(dtrace_byname, probe);
7766
7767 if (first == NULL) {
7768 first = probe;
7769 probe->dtpr_nextmod = NULL;
7770 } else {
7771 probe->dtpr_nextmod = first;
7772 first = probe;
7773 }
7774 }
7775
7776 /*
7777 * The provider's probes have been removed from the hash chains and
7778 * from the probe array. Now issue a dtrace_sync() to be sure that
7779 * everyone has cleared out from any probe array processing.
7780 */
7781 dtrace_sync();
7782
7783 for (probe = first; probe != NULL; probe = first) {
7784 first = probe->dtpr_nextmod;
7785
7786 old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe->dtpr_id,
7787 probe->dtpr_arg);
7788 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
7789 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
7790 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
7791#if defined(sun)
7792 vmem_free(dtrace_arena, (void *)(uintptr_t)(probe->dtpr_id), 1);
7793#else
7794 free_unr(dtrace_arena, probe->dtpr_id);
7795#endif
7796 kmem_free(probe, sizeof (dtrace_probe_t));
7797 }
7798
7799 if ((prev = dtrace_provider) == old) {
7800#if defined(sun)
7801 ASSERT(self || dtrace_devi == NULL);
7802 ASSERT(old->dtpv_next == NULL || dtrace_devi == NULL);
7803#endif
7804 dtrace_provider = old->dtpv_next;
7805 } else {
7806 while (prev != NULL && prev->dtpv_next != old)
7807 prev = prev->dtpv_next;
7808
7809 if (prev == NULL) {
7810 panic("attempt to unregister non-existent "
7811 "dtrace provider %p\n", (void *)id);
7812 }
7813
7814 prev->dtpv_next = old->dtpv_next;
7815 }
7816
7817 if (!self) {
7818 mutex_exit(&dtrace_lock);
7819#if defined(sun)
7820 mutex_exit(&mod_lock);
7821#endif
7822 mutex_exit(&dtrace_provider_lock);
7823 }
7824
7825 kmem_free(old->dtpv_name, strlen(old->dtpv_name) + 1);
7826 kmem_free(old, sizeof (dtrace_provider_t));
7827
7828 return (0);
7829}
7830
7831/*
7832 * Invalidate the specified provider. All subsequent probe lookups for the
7833 * specified provider will fail, but its probes will not be removed.
7834 */
7835void
7836dtrace_invalidate(dtrace_provider_id_t id)
7837{
7838 dtrace_provider_t *pvp = (dtrace_provider_t *)id;
7839
7840 ASSERT(pvp->dtpv_pops.dtps_enable !=
7841 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop);
7842
7843 mutex_enter(&dtrace_provider_lock);
7844 mutex_enter(&dtrace_lock);
7845
7846 pvp->dtpv_defunct = dtrace_gethrtime();
7847
7848 mutex_exit(&dtrace_lock);
7849 mutex_exit(&dtrace_provider_lock);
7850}
7851
7852/*
7853 * Indicate whether or not DTrace has attached.
7854 */
7855int
7856dtrace_attached(void)
7857{
7858 /*
7859 * dtrace_provider will be non-NULL iff the DTrace driver has
7860 * attached. (It's non-NULL because DTrace is always itself a
7861 * provider.)
7862 */
7863 return (dtrace_provider != NULL);
7864}
7865
7866/*
7867 * Remove all the unenabled probes for the given provider. This function is
7868 * not unlike dtrace_unregister(), except that it doesn't remove the provider
7869 * -- just as many of its associated probes as it can.
7870 */
7871int
7872dtrace_condense(dtrace_provider_id_t id)
7873{
7874 dtrace_provider_t *prov = (dtrace_provider_t *)id;
7875 int i;
7876 dtrace_probe_t *probe;
7877
7878 /*
7879 * Make sure this isn't the dtrace provider itself.
7880 */
7881 ASSERT(prov->dtpv_pops.dtps_enable !=
7882 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop);
7883
7884 mutex_enter(&dtrace_provider_lock);
7885 mutex_enter(&dtrace_lock);
7886
7887 /*
7888 * Attempt to destroy the probes associated with this provider.
7889 */
7890 for (i = 0; i < dtrace_nprobes; i++) {
7891 if ((probe = dtrace_probes[i]) == NULL)
7892 continue;
7893
7894 if (probe->dtpr_provider != prov)
7895 continue;
7896
7897 if (probe->dtpr_ecb != NULL)
7898 continue;
7899
7900 dtrace_probes[i] = NULL;
7901
7902 dtrace_hash_remove(dtrace_bymod, probe);
7903 dtrace_hash_remove(dtrace_byfunc, probe);
7904 dtrace_hash_remove(dtrace_byname, probe);
7905
7906 prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, i + 1,
7907 probe->dtpr_arg);
7908 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
7909 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
7910 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
7911 kmem_free(probe, sizeof (dtrace_probe_t));
7912#if defined(sun)
7913 vmem_free(dtrace_arena, (void *)((uintptr_t)i + 1), 1);
7914#else
7915 free_unr(dtrace_arena, i + 1);
7916#endif
7917 }
7918
7919 mutex_exit(&dtrace_lock);
7920 mutex_exit(&dtrace_provider_lock);
7921
7922 return (0);
7923}
7924
7925/*
7926 * DTrace Probe Management Functions
7927 *
7928 * The functions in this section perform the DTrace probe management,
7929 * including functions to create probes, look-up probes, and call into the
7930 * providers to request that probes be provided. Some of these functions are
7931 * in the Provider-to-Framework API; these functions can be identified by the
7932 * fact that they are not declared "static".
7933 */
7934
7935/*
7936 * Create a probe with the specified module name, function name, and name.
7937 */
7938dtrace_id_t
7939dtrace_probe_create(dtrace_provider_id_t prov, const char *mod,
7940 const char *func, const char *name, int aframes, void *arg)
7941{
7942 dtrace_probe_t *probe, **probes;
7943 dtrace_provider_t *provider = (dtrace_provider_t *)prov;
7944 dtrace_id_t id;
7945
7946 if (provider == dtrace_provider) {
7947 ASSERT(MUTEX_HELD(&dtrace_lock));
7948 } else {
7949 mutex_enter(&dtrace_lock);
7950 }
7951
7952#if defined(sun)
7953 id = (dtrace_id_t)(uintptr_t)vmem_alloc(dtrace_arena, 1,
7954 VM_BESTFIT | VM_SLEEP);
7955#else
7956 id = alloc_unr(dtrace_arena);
7957#endif
7958 probe = kmem_zalloc(sizeof (dtrace_probe_t), KM_SLEEP);
7959
7960 probe->dtpr_id = id;
7961 probe->dtpr_gen = dtrace_probegen++;
7962 probe->dtpr_mod = dtrace_strdup(mod);
7963 probe->dtpr_func = dtrace_strdup(func);
7964 probe->dtpr_name = dtrace_strdup(name);
7965 probe->dtpr_arg = arg;
7966 probe->dtpr_aframes = aframes;
7967 probe->dtpr_provider = provider;
7968
7969 dtrace_hash_add(dtrace_bymod, probe);
7970 dtrace_hash_add(dtrace_byfunc, probe);
7971 dtrace_hash_add(dtrace_byname, probe);
7972
7973 if (id - 1 >= dtrace_nprobes) {
7974 size_t osize = dtrace_nprobes * sizeof (dtrace_probe_t *);
7975 size_t nsize = osize << 1;
7976
7977 if (nsize == 0) {
7978 ASSERT(osize == 0);
7979 ASSERT(dtrace_probes == NULL);
7980 nsize = sizeof (dtrace_probe_t *);
7981 }
7982
7983 probes = kmem_zalloc(nsize, KM_SLEEP);
7984
7985 if (dtrace_probes == NULL) {
7986 ASSERT(osize == 0);
7987 dtrace_probes = probes;
7988 dtrace_nprobes = 1;
7989 } else {
7990 dtrace_probe_t **oprobes = dtrace_probes;
7991
7992 bcopy(oprobes, probes, osize);
7993 dtrace_membar_producer();
7994 dtrace_probes = probes;
7995
7996 dtrace_sync();
7997
7998 /*
7999 * All CPUs are now seeing the new probes array; we can
8000 * safely free the old array.
8001 */
8002 kmem_free(oprobes, osize);
8003 dtrace_nprobes <<= 1;
8004 }
8005
8006 ASSERT(id - 1 < dtrace_nprobes);
8007 }
8008
8009 ASSERT(dtrace_probes[id - 1] == NULL);
8010 dtrace_probes[id - 1] = probe;
8011
8012 if (provider != dtrace_provider)
8013 mutex_exit(&dtrace_lock);
8014
8015 return (id);
8016}
8017
8018static dtrace_probe_t *
8019dtrace_probe_lookup_id(dtrace_id_t id)
8020{
8021 ASSERT(MUTEX_HELD(&dtrace_lock));
8022
8023 if (id == 0 || id > dtrace_nprobes)
8024 return (NULL);
8025
8026 return (dtrace_probes[id - 1]);
8027}
8028
8029static int
8030dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg)
8031{
8032 *((dtrace_id_t *)arg) = probe->dtpr_id;
8033
8034 return (DTRACE_MATCH_DONE);
8035}
8036
8037/*
8038 * Look up a probe based on provider and one or more of module name, function
8039 * name and probe name.
8040 */
8041dtrace_id_t
8042dtrace_probe_lookup(dtrace_provider_id_t prid, char *mod,
8043 char *func, char *name)
8044{
8045 dtrace_probekey_t pkey;
8046 dtrace_id_t id;
8047 int match;
8048
8049 pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name;
8050 pkey.dtpk_pmatch = &dtrace_match_string;
8051 pkey.dtpk_mod = mod;
8052 pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul;
8053 pkey.dtpk_func = func;
8054 pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul;
8055 pkey.dtpk_name = name;
8056 pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul;
8057 pkey.dtpk_id = DTRACE_IDNONE;
8058
8059 mutex_enter(&dtrace_lock);
8060 match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, 0,
8061 dtrace_probe_lookup_match, &id);
8062 mutex_exit(&dtrace_lock);
8063
8064 ASSERT(match == 1 || match == 0);
8065 return (match ? id : 0);
8066}
8067
8068/*
8069 * Returns the probe argument associated with the specified probe.
8070 */
8071void *
8072dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid)
8073{
8074 dtrace_probe_t *probe;
8075 void *rval = NULL;
8076
8077 mutex_enter(&dtrace_lock);
8078
8079 if ((probe = dtrace_probe_lookup_id(pid)) != NULL &&
8080 probe->dtpr_provider == (dtrace_provider_t *)id)
8081 rval = probe->dtpr_arg;
8082
8083 mutex_exit(&dtrace_lock);
8084
8085 return (rval);
8086}
8087
8088/*
8089 * Copy a probe into a probe description.
8090 */
8091static void
8092dtrace_probe_description(const dtrace_probe_t *prp, dtrace_probedesc_t *pdp)
8093{
8094 bzero(pdp, sizeof (dtrace_probedesc_t));
8095 pdp->dtpd_id = prp->dtpr_id;
8096
8097 (void) strncpy(pdp->dtpd_provider,
8098 prp->dtpr_provider->dtpv_name, DTRACE_PROVNAMELEN - 1);
8099
8100 (void) strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1);
8101 (void) strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1);
8102 (void) strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1);
8103}
8104
8105/*
8106 * Called to indicate that a probe -- or probes -- should be provided by a
8107 * specfied provider. If the specified description is NULL, the provider will
8108 * be told to provide all of its probes. (This is done whenever a new
8109 * consumer comes along, or whenever a retained enabling is to be matched.) If
8110 * the specified description is non-NULL, the provider is given the
8111 * opportunity to dynamically provide the specified probe, allowing providers
8112 * to support the creation of probes on-the-fly. (So-called _autocreated_
8113 * probes.) If the provider is NULL, the operations will be applied to all
8114 * providers; if the provider is non-NULL the operations will only be applied
8115 * to the specified provider. The dtrace_provider_lock must be held, and the
8116 * dtrace_lock must _not_ be held -- the provider's dtps_provide() operation
8117 * will need to grab the dtrace_lock when it reenters the framework through
8118 * dtrace_probe_lookup(), dtrace_probe_create(), etc.
8119 */
8120static void
8121dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv)
8122{
8123#if defined(sun)
8124 modctl_t *ctl;
8125#endif
8126 int all = 0;
8127
8128 ASSERT(MUTEX_HELD(&dtrace_provider_lock));
8129
8130 if (prv == NULL) {
8131 all = 1;
8132 prv = dtrace_provider;
8133 }
8134
8135 do {
8136 /*
8137 * First, call the blanket provide operation.
8138 */
8139 prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc);
8140
8141#if defined(sun)
8142 /*
8143 * Now call the per-module provide operation. We will grab
8144 * mod_lock to prevent the list from being modified. Note
8145 * that this also prevents the mod_busy bits from changing.
8146 * (mod_busy can only be changed with mod_lock held.)
8147 */
8148 mutex_enter(&mod_lock);
8149
8150 ctl = &modules;
8151 do {
8152 if (ctl->mod_busy || ctl->mod_mp == NULL)
8153 continue;
8154
8155 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl);
8156
8157 } while ((ctl = ctl->mod_next) != &modules);
8158
8159 mutex_exit(&mod_lock);
8160#endif
8161 } while (all && (prv = prv->dtpv_next) != NULL);
8162}
8163
8164#if defined(sun)
8165/*
8166 * Iterate over each probe, and call the Framework-to-Provider API function
8167 * denoted by offs.
8168 */
8169static void
8170dtrace_probe_foreach(uintptr_t offs)
8171{
8172 dtrace_provider_t *prov;
8173 void (*func)(void *, dtrace_id_t, void *);
8174 dtrace_probe_t *probe;
8175 dtrace_icookie_t cookie;
8176 int i;
8177
8178 /*
8179 * We disable interrupts to walk through the probe array. This is
8180 * safe -- the dtrace_sync() in dtrace_unregister() assures that we
8181 * won't see stale data.
8182 */
8183 cookie = dtrace_interrupt_disable();
8184
8185 for (i = 0; i < dtrace_nprobes; i++) {
8186 if ((probe = dtrace_probes[i]) == NULL)
8187 continue;
8188
8189 if (probe->dtpr_ecb == NULL) {
8190 /*
8191 * This probe isn't enabled -- don't call the function.
8192 */
8193 continue;
8194 }
8195
8196 prov = probe->dtpr_provider;
8197 func = *((void(**)(void *, dtrace_id_t, void *))
8198 ((uintptr_t)&prov->dtpv_pops + offs));
8199
8200 func(prov->dtpv_arg, i + 1, probe->dtpr_arg);
8201 }
8202
8203 dtrace_interrupt_enable(cookie);
8204}
8205#endif
8206
8207static int
8208dtrace_probe_enable(dtrace_probedesc_t *desc, dtrace_enabling_t *enab)
8209{
8210 dtrace_probekey_t pkey;
8211 uint32_t priv;
8212 uid_t uid;
8213 zoneid_t zoneid;
8214
8215 ASSERT(MUTEX_HELD(&dtrace_lock));
8216 dtrace_ecb_create_cache = NULL;
8217
8218 if (desc == NULL) {
8219 /*
8220 * If we're passed a NULL description, we're being asked to
8221 * create an ECB with a NULL probe.
8222 */
8223 (void) dtrace_ecb_create_enable(NULL, enab);
8224 return (0);
8225 }
8226
8227 dtrace_probekey(desc, &pkey);
8228 dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred,
8229 &priv, &uid, &zoneid);
8230
8231 return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable,
8232 enab));
8233}
8234
8235/*
8236 * DTrace Helper Provider Functions
8237 */
8238static void
8239dtrace_dofattr2attr(dtrace_attribute_t *attr, const dof_attr_t dofattr)
8240{
8241 attr->dtat_name = DOF_ATTR_NAME(dofattr);
8242 attr->dtat_data = DOF_ATTR_DATA(dofattr);
8243 attr->dtat_class = DOF_ATTR_CLASS(dofattr);
8244}
8245
8246static void
8247dtrace_dofprov2hprov(dtrace_helper_provdesc_t *hprov,
8248 const dof_provider_t *dofprov, char *strtab)
8249{
8250 hprov->dthpv_provname = strtab + dofprov->dofpv_name;
8251 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_provider,
8252 dofprov->dofpv_provattr);
8253 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_mod,
8254 dofprov->dofpv_modattr);
8255 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_func,
8256 dofprov->dofpv_funcattr);
8257 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_name,
8258 dofprov->dofpv_nameattr);
8259 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_args,
8260 dofprov->dofpv_argsattr);
8261}
8262
8263static void
8264dtrace_helper_provide_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid)
8265{
8266 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8267 dof_hdr_t *dof = (dof_hdr_t *)daddr;
8268 dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec;
8269 dof_provider_t *provider;
8270 dof_probe_t *probe;
8271 uint32_t *off, *enoff;
8272 uint8_t *arg;
8273 char *strtab;
8274 uint_t i, nprobes;
8275 dtrace_helper_provdesc_t dhpv;
8276 dtrace_helper_probedesc_t dhpb;
8277 dtrace_meta_t *meta = dtrace_meta_pid;
8278 dtrace_mops_t *mops = &meta->dtm_mops;
8279 void *parg;
8280
8281 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
8282 str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8283 provider->dofpv_strtab * dof->dofh_secsize);
8284 prb_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8285 provider->dofpv_probes * dof->dofh_secsize);
8286 arg_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8287 provider->dofpv_prargs * dof->dofh_secsize);
8288 off_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8289 provider->dofpv_proffs * dof->dofh_secsize);
8290
8291 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
8292 off = (uint32_t *)(uintptr_t)(daddr + off_sec->dofs_offset);
8293 arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset);
8294 enoff = NULL;
8295
8296 /*
8297 * See dtrace_helper_provider_validate().
8298 */
8299 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
8300 provider->dofpv_prenoffs != DOF_SECT_NONE) {
8301 enoff_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8302 provider->dofpv_prenoffs * dof->dofh_secsize);
8303 enoff = (uint32_t *)(uintptr_t)(daddr + enoff_sec->dofs_offset);
8304 }
8305
8306 nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize;
8307
8308 /*
8309 * Create the provider.
8310 */
8311 dtrace_dofprov2hprov(&dhpv, provider, strtab);
8312
8313 if ((parg = mops->dtms_provide_pid(meta->dtm_arg, &dhpv, pid)) == NULL)
8314 return;
8315
8316 meta->dtm_count++;
8317
8318 /*
8319 * Create the probes.
8320 */
8321 for (i = 0; i < nprobes; i++) {
8322 probe = (dof_probe_t *)(uintptr_t)(daddr +
8323 prb_sec->dofs_offset + i * prb_sec->dofs_entsize);
8324
8325 dhpb.dthpb_mod = dhp->dofhp_mod;
8326 dhpb.dthpb_func = strtab + probe->dofpr_func;
8327 dhpb.dthpb_name = strtab + probe->dofpr_name;
8328 dhpb.dthpb_base = probe->dofpr_addr;
8329 dhpb.dthpb_offs = off + probe->dofpr_offidx;
8330 dhpb.dthpb_noffs = probe->dofpr_noffs;
8331 if (enoff != NULL) {
8332 dhpb.dthpb_enoffs = enoff + probe->dofpr_enoffidx;
8333 dhpb.dthpb_nenoffs = probe->dofpr_nenoffs;
8334 } else {
8335 dhpb.dthpb_enoffs = NULL;
8336 dhpb.dthpb_nenoffs = 0;
8337 }
8338 dhpb.dthpb_args = arg + probe->dofpr_argidx;
8339 dhpb.dthpb_nargc = probe->dofpr_nargc;
8340 dhpb.dthpb_xargc = probe->dofpr_xargc;
8341 dhpb.dthpb_ntypes = strtab + probe->dofpr_nargv;
8342 dhpb.dthpb_xtypes = strtab + probe->dofpr_xargv;
8343
8344 mops->dtms_create_probe(meta->dtm_arg, parg, &dhpb);
8345 }
8346}
8347
8348static void
8349dtrace_helper_provide(dof_helper_t *dhp, pid_t pid)
8350{
8351 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8352 dof_hdr_t *dof = (dof_hdr_t *)daddr;
8353 int i;
8354
8355 ASSERT(MUTEX_HELD(&dtrace_meta_lock));
8356
8357 for (i = 0; i < dof->dofh_secnum; i++) {
8358 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
8359 dof->dofh_secoff + i * dof->dofh_secsize);
8360
8361 if (sec->dofs_type != DOF_SECT_PROVIDER)
8362 continue;
8363
8364 dtrace_helper_provide_one(dhp, sec, pid);
8365 }
8366
8367 /*
8368 * We may have just created probes, so we must now rematch against
8369 * any retained enablings. Note that this call will acquire both
8370 * cpu_lock and dtrace_lock; the fact that we are holding
8371 * dtrace_meta_lock now is what defines the ordering with respect to
8372 * these three locks.
8373 */
8374 dtrace_enabling_matchall();
8375}
8376
8377static void
8378dtrace_helper_provider_remove_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid)
8379{
8380 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8381 dof_hdr_t *dof = (dof_hdr_t *)daddr;
8382 dof_sec_t *str_sec;
8383 dof_provider_t *provider;
8384 char *strtab;
8385 dtrace_helper_provdesc_t dhpv;
8386 dtrace_meta_t *meta = dtrace_meta_pid;
8387 dtrace_mops_t *mops = &meta->dtm_mops;
8388
8389 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
8390 str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8391 provider->dofpv_strtab * dof->dofh_secsize);
8392
8393 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
8394
8395 /*
8396 * Create the provider.
8397 */
8398 dtrace_dofprov2hprov(&dhpv, provider, strtab);
8399
8400 mops->dtms_remove_pid(meta->dtm_arg, &dhpv, pid);
8401
8402 meta->dtm_count--;
8403}
8404
8405static void
8406dtrace_helper_provider_remove(dof_helper_t *dhp, pid_t pid)
8407{
8408 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8409 dof_hdr_t *dof = (dof_hdr_t *)daddr;
8410 int i;
8411
8412 ASSERT(MUTEX_HELD(&dtrace_meta_lock));
8413
8414 for (i = 0; i < dof->dofh_secnum; i++) {
8415 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
8416 dof->dofh_secoff + i * dof->dofh_secsize);
8417
8418 if (sec->dofs_type != DOF_SECT_PROVIDER)
8419 continue;
8420
8421 dtrace_helper_provider_remove_one(dhp, sec, pid);
8422 }
8423}
8424
8425/*
8426 * DTrace Meta Provider-to-Framework API Functions
8427 *
8428 * These functions implement the Meta Provider-to-Framework API, as described
8429 * in <sys/dtrace.h>.
8430 */
8431int
8432dtrace_meta_register(const char *name, const dtrace_mops_t *mops, void *arg,
8433 dtrace_meta_provider_id_t *idp)
8434{
8435 dtrace_meta_t *meta;
8436 dtrace_helpers_t *help, *next;
8437 int i;
8438
8439 *idp = DTRACE_METAPROVNONE;
8440
8441 /*
8442 * We strictly don't need the name, but we hold onto it for
8443 * debuggability. All hail error queues!
8444 */
8445 if (name == NULL) {
8446 cmn_err(CE_WARN, "failed to register meta-provider: "
8447 "invalid name");
8448 return (EINVAL);
8449 }
8450
8451 if (mops == NULL ||
8452 mops->dtms_create_probe == NULL ||
8453 mops->dtms_provide_pid == NULL ||
8454 mops->dtms_remove_pid == NULL) {
8455 cmn_err(CE_WARN, "failed to register meta-register %s: "
8456 "invalid ops", name);
8457 return (EINVAL);
8458 }
8459
8460 meta = kmem_zalloc(sizeof (dtrace_meta_t), KM_SLEEP);
8461 meta->dtm_mops = *mops;
8462 meta->dtm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
8463 (void) strcpy(meta->dtm_name, name);
8464 meta->dtm_arg = arg;
8465
8466 mutex_enter(&dtrace_meta_lock);
8467 mutex_enter(&dtrace_lock);
8468
8469 if (dtrace_meta_pid != NULL) {
8470 mutex_exit(&dtrace_lock);
8471 mutex_exit(&dtrace_meta_lock);
8472 cmn_err(CE_WARN, "failed to register meta-register %s: "
8473 "user-land meta-provider exists", name);
8474 kmem_free(meta->dtm_name, strlen(meta->dtm_name) + 1);
8475 kmem_free(meta, sizeof (dtrace_meta_t));
8476 return (EINVAL);
8477 }
8478
8479 dtrace_meta_pid = meta;
8480 *idp = (dtrace_meta_provider_id_t)meta;
8481
8482 /*
8483 * If there are providers and probes ready to go, pass them
8484 * off to the new meta provider now.
8485 */
8486
8487 help = dtrace_deferred_pid;
8488 dtrace_deferred_pid = NULL;
8489
8490 mutex_exit(&dtrace_lock);
8491
8492 while (help != NULL) {
8493 for (i = 0; i < help->dthps_nprovs; i++) {
8494 dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov,
8495 help->dthps_pid);
8496 }
8497
8498 next = help->dthps_next;
8499 help->dthps_next = NULL;
8500 help->dthps_prev = NULL;
8501 help->dthps_deferred = 0;
8502 help = next;
8503 }
8504
8505 mutex_exit(&dtrace_meta_lock);
8506
8507 return (0);
8508}
8509
8510int
8511dtrace_meta_unregister(dtrace_meta_provider_id_t id)
8512{
8513 dtrace_meta_t **pp, *old = (dtrace_meta_t *)id;
8514
8515 mutex_enter(&dtrace_meta_lock);
8516 mutex_enter(&dtrace_lock);
8517
8518 if (old == dtrace_meta_pid) {
8519 pp = &dtrace_meta_pid;
8520 } else {
8521 panic("attempt to unregister non-existent "
8522 "dtrace meta-provider %p\n", (void *)old);
8523 }
8524
8525 if (old->dtm_count != 0) {
8526 mutex_exit(&dtrace_lock);
8527 mutex_exit(&dtrace_meta_lock);
8528 return (EBUSY);
8529 }
8530
8531 *pp = NULL;
8532
8533 mutex_exit(&dtrace_lock);
8534 mutex_exit(&dtrace_meta_lock);
8535
8536 kmem_free(old->dtm_name, strlen(old->dtm_name) + 1);
8537 kmem_free(old, sizeof (dtrace_meta_t));
8538
8539 return (0);
8540}
8541
8542
8543/*
8544 * DTrace DIF Object Functions
8545 */
8546static int
8547dtrace_difo_err(uint_t pc, const char *format, ...)
8548{
8549 if (dtrace_err_verbose) {
8550 va_list alist;
8551
8552 (void) uprintf("dtrace DIF object error: [%u]: ", pc);
8553 va_start(alist, format);
8554 (void) vuprintf(format, alist);
8555 va_end(alist);
8556 }
8557
8558#ifdef DTRACE_ERRDEBUG
8559 dtrace_errdebug(format);
8560#endif
8561 return (1);
8562}
8563
8564/*
8565 * Validate a DTrace DIF object by checking the IR instructions. The following
8566 * rules are currently enforced by dtrace_difo_validate():
8567 *
8568 * 1. Each instruction must have a valid opcode
8569 * 2. Each register, string, variable, or subroutine reference must be valid
8570 * 3. No instruction can modify register %r0 (must be zero)
8571 * 4. All instruction reserved bits must be set to zero
8572 * 5. The last instruction must be a "ret" instruction
8573 * 6. All branch targets must reference a valid instruction _after_ the branch
8574 */
8575static int
8576dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, uint_t nregs,
8577 cred_t *cr)
8578{
8579 int err = 0, i;
8580 int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err;
8581 int kcheckload;
8582 uint_t pc;
8583
8584 kcheckload = cr == NULL ||
8585 (vstate->dtvs_state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) == 0;
8586
8587 dp->dtdo_destructive = 0;
8588
8589 for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) {
8590 dif_instr_t instr = dp->dtdo_buf[pc];
8591
8592 uint_t r1 = DIF_INSTR_R1(instr);
8593 uint_t r2 = DIF_INSTR_R2(instr);
8594 uint_t rd = DIF_INSTR_RD(instr);
8595 uint_t rs = DIF_INSTR_RS(instr);
8596 uint_t label = DIF_INSTR_LABEL(instr);
8597 uint_t v = DIF_INSTR_VAR(instr);
8598 uint_t subr = DIF_INSTR_SUBR(instr);
8599 uint_t type = DIF_INSTR_TYPE(instr);
8600 uint_t op = DIF_INSTR_OP(instr);
8601
8602 switch (op) {
8603 case DIF_OP_OR:
8604 case DIF_OP_XOR:
8605 case DIF_OP_AND:
8606 case DIF_OP_SLL:
8607 case DIF_OP_SRL:
8608 case DIF_OP_SRA:
8609 case DIF_OP_SUB:
8610 case DIF_OP_ADD:
8611 case DIF_OP_MUL:
8612 case DIF_OP_SDIV:
8613 case DIF_OP_UDIV:
8614 case DIF_OP_SREM:
8615 case DIF_OP_UREM:
8616 case DIF_OP_COPYS:
8617 if (r1 >= nregs)
8618 err += efunc(pc, "invalid register %u\n", r1);
8619 if (r2 >= nregs)
8620 err += efunc(pc, "invalid register %u\n", r2);
8621 if (rd >= nregs)
8622 err += efunc(pc, "invalid register %u\n", rd);
8623 if (rd == 0)
8624 err += efunc(pc, "cannot write to %r0\n");
8625 break;
8626 case DIF_OP_NOT:
8627 case DIF_OP_MOV:
8628 case DIF_OP_ALLOCS:
8629 if (r1 >= nregs)
8630 err += efunc(pc, "invalid register %u\n", r1);
8631 if (r2 != 0)
8632 err += efunc(pc, "non-zero reserved bits\n");
8633 if (rd >= nregs)
8634 err += efunc(pc, "invalid register %u\n", rd);
8635 if (rd == 0)
8636 err += efunc(pc, "cannot write to %r0\n");
8637 break;
8638 case DIF_OP_LDSB:
8639 case DIF_OP_LDSH:
8640 case DIF_OP_LDSW:
8641 case DIF_OP_LDUB:
8642 case DIF_OP_LDUH:
8643 case DIF_OP_LDUW:
8644 case DIF_OP_LDX:
8645 if (r1 >= nregs)
8646 err += efunc(pc, "invalid register %u\n", r1);
8647 if (r2 != 0)
8648 err += efunc(pc, "non-zero reserved bits\n");
8649 if (rd >= nregs)
8650 err += efunc(pc, "invalid register %u\n", rd);
8651 if (rd == 0)
8652 err += efunc(pc, "cannot write to %r0\n");
8653 if (kcheckload)
8654 dp->dtdo_buf[pc] = DIF_INSTR_LOAD(op +
8655 DIF_OP_RLDSB - DIF_OP_LDSB, r1, rd);
8656 break;
8657 case DIF_OP_RLDSB:
8658 case DIF_OP_RLDSH:
8659 case DIF_OP_RLDSW:
8660 case DIF_OP_RLDUB:
8661 case DIF_OP_RLDUH:
8662 case DIF_OP_RLDUW:
8663 case DIF_OP_RLDX:
8664 if (r1 >= nregs)
8665 err += efunc(pc, "invalid register %u\n", r1);
8666 if (r2 != 0)
8667 err += efunc(pc, "non-zero reserved bits\n");
8668 if (rd >= nregs)
8669 err += efunc(pc, "invalid register %u\n", rd);
8670 if (rd == 0)
8671 err += efunc(pc, "cannot write to %r0\n");
8672 break;
8673 case DIF_OP_ULDSB:
8674 case DIF_OP_ULDSH:
8675 case DIF_OP_ULDSW:
8676 case DIF_OP_ULDUB:
8677 case DIF_OP_ULDUH:
8678 case DIF_OP_ULDUW:
8679 case DIF_OP_ULDX:
8680 if (r1 >= nregs)
8681 err += efunc(pc, "invalid register %u\n", r1);
8682 if (r2 != 0)
8683 err += efunc(pc, "non-zero reserved bits\n");
8684 if (rd >= nregs)
8685 err += efunc(pc, "invalid register %u\n", rd);
8686 if (rd == 0)
8687 err += efunc(pc, "cannot write to %r0\n");
8688 break;
8689 case DIF_OP_STB:
8690 case DIF_OP_STH:
8691 case DIF_OP_STW:
8692 case DIF_OP_STX:
8693 if (r1 >= nregs)
8694 err += efunc(pc, "invalid register %u\n", r1);
8695 if (r2 != 0)
8696 err += efunc(pc, "non-zero reserved bits\n");
8697 if (rd >= nregs)
8698 err += efunc(pc, "invalid register %u\n", rd);
8699 if (rd == 0)
8700 err += efunc(pc, "cannot write to 0 address\n");
8701 break;
8702 case DIF_OP_CMP:
8703 case DIF_OP_SCMP:
8704 if (r1 >= nregs)
8705 err += efunc(pc, "invalid register %u\n", r1);
8706 if (r2 >= nregs)
8707 err += efunc(pc, "invalid register %u\n", r2);
8708 if (rd != 0)
8709 err += efunc(pc, "non-zero reserved bits\n");
8710 break;
8711 case DIF_OP_TST:
8712 if (r1 >= nregs)
8713 err += efunc(pc, "invalid register %u\n", r1);
8714 if (r2 != 0 || rd != 0)
8715 err += efunc(pc, "non-zero reserved bits\n");
8716 break;
8717 case DIF_OP_BA:
8718 case DIF_OP_BE:
8719 case DIF_OP_BNE:
8720 case DIF_OP_BG:
8721 case DIF_OP_BGU:
8722 case DIF_OP_BGE:
8723 case DIF_OP_BGEU:
8724 case DIF_OP_BL:
8725 case DIF_OP_BLU:
8726 case DIF_OP_BLE:
8727 case DIF_OP_BLEU:
8728 if (label >= dp->dtdo_len) {
8729 err += efunc(pc, "invalid branch target %u\n",
8730 label);
8731 }
8732 if (label <= pc) {
8733 err += efunc(pc, "backward branch to %u\n",
8734 label);
8735 }
8736 break;
8737 case DIF_OP_RET:
8738 if (r1 != 0 || r2 != 0)
8739 err += efunc(pc, "non-zero reserved bits\n");
8740 if (rd >= nregs)
8741 err += efunc(pc, "invalid register %u\n", rd);
8742 break;
8743 case DIF_OP_NOP:
8744 case DIF_OP_POPTS:
8745 case DIF_OP_FLUSHTS:
8746 if (r1 != 0 || r2 != 0 || rd != 0)
8747 err += efunc(pc, "non-zero reserved bits\n");
8748 break;
8749 case DIF_OP_SETX:
8750 if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) {
8751 err += efunc(pc, "invalid integer ref %u\n",
8752 DIF_INSTR_INTEGER(instr));
8753 }
8754 if (rd >= nregs)
8755 err += efunc(pc, "invalid register %u\n", rd);
8756 if (rd == 0)
8757 err += efunc(pc, "cannot write to %r0\n");
8758 break;
8759 case DIF_OP_SETS:
8760 if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) {
8761 err += efunc(pc, "invalid string ref %u\n",
8762 DIF_INSTR_STRING(instr));
8763 }
8764 if (rd >= nregs)
8765 err += efunc(pc, "invalid register %u\n", rd);
8766 if (rd == 0)
8767 err += efunc(pc, "cannot write to %r0\n");
8768 break;
8769 case DIF_OP_LDGA:
8770 case DIF_OP_LDTA:
8771 if (r1 > DIF_VAR_ARRAY_MAX)
8772 err += efunc(pc, "invalid array %u\n", r1);
8773 if (r2 >= nregs)
8774 err += efunc(pc, "invalid register %u\n", r2);
8775 if (rd >= nregs)
8776 err += efunc(pc, "invalid register %u\n", rd);
8777 if (rd == 0)
8778 err += efunc(pc, "cannot write to %r0\n");
8779 break;
8780 case DIF_OP_LDGS:
8781 case DIF_OP_LDTS:
8782 case DIF_OP_LDLS:
8783 case DIF_OP_LDGAA:
8784 case DIF_OP_LDTAA:
8785 if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX)
8786 err += efunc(pc, "invalid variable %u\n", v);
8787 if (rd >= nregs)
8788 err += efunc(pc, "invalid register %u\n", rd);
8789 if (rd == 0)
8790 err += efunc(pc, "cannot write to %r0\n");
8791 break;
8792 case DIF_OP_STGS:
8793 case DIF_OP_STTS:
8794 case DIF_OP_STLS:
8795 case DIF_OP_STGAA:
8796 case DIF_OP_STTAA:
8797 if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX)
8798 err += efunc(pc, "invalid variable %u\n", v);
8799 if (rs >= nregs)
8800 err += efunc(pc, "invalid register %u\n", rd);
8801 break;
8802 case DIF_OP_CALL:
8803 if (subr > DIF_SUBR_MAX)
8804 err += efunc(pc, "invalid subr %u\n", subr);
8805 if (rd >= nregs)
8806 err += efunc(pc, "invalid register %u\n", rd);
8807 if (rd == 0)
8808 err += efunc(pc, "cannot write to %r0\n");
8809
8810 if (subr == DIF_SUBR_COPYOUT ||
8811 subr == DIF_SUBR_COPYOUTSTR) {
8812 dp->dtdo_destructive = 1;
8813 }
8814 break;
8815 case DIF_OP_PUSHTR:
8816 if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF)
8817 err += efunc(pc, "invalid ref type %u\n", type);
8818 if (r2 >= nregs)
8819 err += efunc(pc, "invalid register %u\n", r2);
8820 if (rs >= nregs)
8821 err += efunc(pc, "invalid register %u\n", rs);
8822 break;
8823 case DIF_OP_PUSHTV:
8824 if (type != DIF_TYPE_CTF)
8825 err += efunc(pc, "invalid val type %u\n", type);
8826 if (r2 >= nregs)
8827 err += efunc(pc, "invalid register %u\n", r2);
8828 if (rs >= nregs)
8829 err += efunc(pc, "invalid register %u\n", rs);
8830 break;
8831 default:
8832 err += efunc(pc, "invalid opcode %u\n",
8833 DIF_INSTR_OP(instr));
8834 }
8835 }
8836
8837 if (dp->dtdo_len != 0 &&
8838 DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) {
8839 err += efunc(dp->dtdo_len - 1,
8840 "expected 'ret' as last DIF instruction\n");
8841 }
8842
8843 if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) {
8844 /*
8845 * If we're not returning by reference, the size must be either
8846 * 0 or the size of one of the base types.
8847 */
8848 switch (dp->dtdo_rtype.dtdt_size) {
8849 case 0:
8850 case sizeof (uint8_t):
8851 case sizeof (uint16_t):
8852 case sizeof (uint32_t):
8853 case sizeof (uint64_t):
8854 break;
8855
8856 default:
8857 err += efunc(dp->dtdo_len - 1, "bad return size");
8858 }
8859 }
8860
8861 for (i = 0; i < dp->dtdo_varlen && err == 0; i++) {
8862 dtrace_difv_t *v = &dp->dtdo_vartab[i], *existing = NULL;
8863 dtrace_diftype_t *vt, *et;
8864 uint_t id, ndx;
8865
8866 if (v->dtdv_scope != DIFV_SCOPE_GLOBAL &&
8867 v->dtdv_scope != DIFV_SCOPE_THREAD &&
8868 v->dtdv_scope != DIFV_SCOPE_LOCAL) {
8869 err += efunc(i, "unrecognized variable scope %d\n",
8870 v->dtdv_scope);
8871 break;
8872 }
8873
8874 if (v->dtdv_kind != DIFV_KIND_ARRAY &&
8875 v->dtdv_kind != DIFV_KIND_SCALAR) {
8876 err += efunc(i, "unrecognized variable type %d\n",
8877 v->dtdv_kind);
8878 break;
8879 }
8880
8881 if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) {
8882 err += efunc(i, "%d exceeds variable id limit\n", id);
8883 break;
8884 }
8885
8886 if (id < DIF_VAR_OTHER_UBASE)
8887 continue;
8888
8889 /*
8890 * For user-defined variables, we need to check that this
8891 * definition is identical to any previous definition that we
8892 * encountered.
8893 */
8894 ndx = id - DIF_VAR_OTHER_UBASE;
8895
8896 switch (v->dtdv_scope) {
8897 case DIFV_SCOPE_GLOBAL:
8898 if (ndx < vstate->dtvs_nglobals) {
8899 dtrace_statvar_t *svar;
8900
8901 if ((svar = vstate->dtvs_globals[ndx]) != NULL)
8902 existing = &svar->dtsv_var;
8903 }
8904
8905 break;
8906
8907 case DIFV_SCOPE_THREAD:
8908 if (ndx < vstate->dtvs_ntlocals)
8909 existing = &vstate->dtvs_tlocals[ndx];
8910 break;
8911
8912 case DIFV_SCOPE_LOCAL:
8913 if (ndx < vstate->dtvs_nlocals) {
8914 dtrace_statvar_t *svar;
8915
8916 if ((svar = vstate->dtvs_locals[ndx]) != NULL)
8917 existing = &svar->dtsv_var;
8918 }
8919
8920 break;
8921 }
8922
8923 vt = &v->dtdv_type;
8924
8925 if (vt->dtdt_flags & DIF_TF_BYREF) {
8926 if (vt->dtdt_size == 0) {
8927 err += efunc(i, "zero-sized variable\n");
8928 break;
8929 }
8930
8931 if (v->dtdv_scope == DIFV_SCOPE_GLOBAL &&
8932 vt->dtdt_size > dtrace_global_maxsize) {
8933 err += efunc(i, "oversized by-ref global\n");
8934 break;
8935 }
8936 }
8937
8938 if (existing == NULL || existing->dtdv_id == 0)
8939 continue;
8940
8941 ASSERT(existing->dtdv_id == v->dtdv_id);
8942 ASSERT(existing->dtdv_scope == v->dtdv_scope);
8943
8944 if (existing->dtdv_kind != v->dtdv_kind)
8945 err += efunc(i, "%d changed variable kind\n", id);
8946
8947 et = &existing->dtdv_type;
8948
8949 if (vt->dtdt_flags != et->dtdt_flags) {
8950 err += efunc(i, "%d changed variable type flags\n", id);
8951 break;
8952 }
8953
8954 if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) {
8955 err += efunc(i, "%d changed variable type size\n", id);
8956 break;
8957 }
8958 }
8959
8960 return (err);
8961}
8962
8963/*
8964 * Validate a DTrace DIF object that it is to be used as a helper. Helpers
8965 * are much more constrained than normal DIFOs. Specifically, they may
8966 * not:
8967 *
8968 * 1. Make calls to subroutines other than copyin(), copyinstr() or
8969 * miscellaneous string routines
8970 * 2. Access DTrace variables other than the args[] array, and the
8971 * curthread, pid, ppid, tid, execname, zonename, uid and gid variables.
8972 * 3. Have thread-local variables.
8973 * 4. Have dynamic variables.
8974 */
8975static int
8976dtrace_difo_validate_helper(dtrace_difo_t *dp)
8977{
8978 int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err;
8979 int err = 0;
8980 uint_t pc;
8981
8982 for (pc = 0; pc < dp->dtdo_len; pc++) {
8983 dif_instr_t instr = dp->dtdo_buf[pc];
8984
8985 uint_t v = DIF_INSTR_VAR(instr);
8986 uint_t subr = DIF_INSTR_SUBR(instr);
8987 uint_t op = DIF_INSTR_OP(instr);
8988
8989 switch (op) {
8990 case DIF_OP_OR:
8991 case DIF_OP_XOR:
8992 case DIF_OP_AND:
8993 case DIF_OP_SLL:
8994 case DIF_OP_SRL:
8995 case DIF_OP_SRA:
8996 case DIF_OP_SUB:
8997 case DIF_OP_ADD:
8998 case DIF_OP_MUL:
8999 case DIF_OP_SDIV:
9000 case DIF_OP_UDIV:
9001 case DIF_OP_SREM:
9002 case DIF_OP_UREM:
9003 case DIF_OP_COPYS:
9004 case DIF_OP_NOT:
9005 case DIF_OP_MOV:
9006 case DIF_OP_RLDSB:
9007 case DIF_OP_RLDSH:
9008 case DIF_OP_RLDSW:
9009 case DIF_OP_RLDUB:
9010 case DIF_OP_RLDUH:
9011 case DIF_OP_RLDUW:
9012 case DIF_OP_RLDX:
9013 case DIF_OP_ULDSB:
9014 case DIF_OP_ULDSH:
9015 case DIF_OP_ULDSW:
9016 case DIF_OP_ULDUB:
9017 case DIF_OP_ULDUH:
9018 case DIF_OP_ULDUW:
9019 case DIF_OP_ULDX:
9020 case DIF_OP_STB:
9021 case DIF_OP_STH:
9022 case DIF_OP_STW:
9023 case DIF_OP_STX:
9024 case DIF_OP_ALLOCS:
9025 case DIF_OP_CMP:
9026 case DIF_OP_SCMP:
9027 case DIF_OP_TST:
9028 case DIF_OP_BA:
9029 case DIF_OP_BE:
9030 case DIF_OP_BNE:
9031 case DIF_OP_BG:
9032 case DIF_OP_BGU:
9033 case DIF_OP_BGE:
9034 case DIF_OP_BGEU:
9035 case DIF_OP_BL:
9036 case DIF_OP_BLU:
9037 case DIF_OP_BLE:
9038 case DIF_OP_BLEU:
9039 case DIF_OP_RET:
9040 case DIF_OP_NOP:
9041 case DIF_OP_POPTS:
9042 case DIF_OP_FLUSHTS:
9043 case DIF_OP_SETX:
9044 case DIF_OP_SETS:
9045 case DIF_OP_LDGA:
9046 case DIF_OP_LDLS:
9047 case DIF_OP_STGS:
9048 case DIF_OP_STLS:
9049 case DIF_OP_PUSHTR:
9050 case DIF_OP_PUSHTV:
9051 break;
9052
9053 case DIF_OP_LDGS:
9054 if (v >= DIF_VAR_OTHER_UBASE)
9055 break;
9056
9057 if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9)
9058 break;
9059
9060 if (v == DIF_VAR_CURTHREAD || v == DIF_VAR_PID ||
9061 v == DIF_VAR_PPID || v == DIF_VAR_TID ||
9062 v == DIF_VAR_EXECARGS ||
9063 v == DIF_VAR_EXECNAME || v == DIF_VAR_ZONENAME ||
9064 v == DIF_VAR_UID || v == DIF_VAR_GID)
9065 break;
9066
9067 err += efunc(pc, "illegal variable %u\n", v);
9068 break;
9069
9070 case DIF_OP_LDTA:
9071 case DIF_OP_LDTS:
9072 case DIF_OP_LDGAA:
9073 case DIF_OP_LDTAA:
9074 err += efunc(pc, "illegal dynamic variable load\n");
9075 break;
9076
9077 case DIF_OP_STTS:
9078 case DIF_OP_STGAA:
9079 case DIF_OP_STTAA:
9080 err += efunc(pc, "illegal dynamic variable store\n");
9081 break;
9082
9083 case DIF_OP_CALL:
9084 if (subr == DIF_SUBR_ALLOCA ||
9085 subr == DIF_SUBR_BCOPY ||
9086 subr == DIF_SUBR_COPYIN ||
9087 subr == DIF_SUBR_COPYINTO ||
9088 subr == DIF_SUBR_COPYINSTR ||
9089 subr == DIF_SUBR_INDEX ||
9090 subr == DIF_SUBR_INET_NTOA ||
9091 subr == DIF_SUBR_INET_NTOA6 ||
9092 subr == DIF_SUBR_INET_NTOP ||
9093 subr == DIF_SUBR_LLTOSTR ||
9094 subr == DIF_SUBR_RINDEX ||
9095 subr == DIF_SUBR_STRCHR ||
9096 subr == DIF_SUBR_STRJOIN ||
9097 subr == DIF_SUBR_STRRCHR ||
9098 subr == DIF_SUBR_STRSTR ||
9099 subr == DIF_SUBR_HTONS ||
9100 subr == DIF_SUBR_HTONL ||
9101 subr == DIF_SUBR_HTONLL ||
9102 subr == DIF_SUBR_NTOHS ||
9103 subr == DIF_SUBR_NTOHL ||
9104 subr == DIF_SUBR_NTOHLL ||
9105 subr == DIF_SUBR_MEMREF ||
9106 subr == DIF_SUBR_TYPEREF)
9107 break;
9108
9109 err += efunc(pc, "invalid subr %u\n", subr);
9110 break;
9111
9112 default:
9113 err += efunc(pc, "invalid opcode %u\n",
9114 DIF_INSTR_OP(instr));
9115 }
9116 }
9117
9118 return (err);
9119}
9120
9121/*
9122 * Returns 1 if the expression in the DIF object can be cached on a per-thread
9123 * basis; 0 if not.
9124 */
9125static int
9126dtrace_difo_cacheable(dtrace_difo_t *dp)
9127{
9128 int i;
9129
9130 if (dp == NULL)
9131 return (0);
9132
9133 for (i = 0; i < dp->dtdo_varlen; i++) {
9134 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9135
9136 if (v->dtdv_scope != DIFV_SCOPE_GLOBAL)
9137 continue;
9138
9139 switch (v->dtdv_id) {
9140 case DIF_VAR_CURTHREAD:
9141 case DIF_VAR_PID:
9142 case DIF_VAR_TID:
9143 case DIF_VAR_EXECARGS:
9144 case DIF_VAR_EXECNAME:
9145 case DIF_VAR_ZONENAME:
9146 break;
9147
9148 default:
9149 return (0);
9150 }
9151 }
9152
9153 /*
9154 * This DIF object may be cacheable. Now we need to look for any
9155 * array loading instructions, any memory loading instructions, or
9156 * any stores to thread-local variables.
9157 */
9158 for (i = 0; i < dp->dtdo_len; i++) {
9159 uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]);
9160
9161 if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) ||
9162 (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) ||
9163 (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) ||
9164 op == DIF_OP_LDGA || op == DIF_OP_STTS)
9165 return (0);
9166 }
9167
9168 return (1);
9169}
9170
9171static void
9172dtrace_difo_hold(dtrace_difo_t *dp)
9173{
9174 int i;
9175
9176 ASSERT(MUTEX_HELD(&dtrace_lock));
9177
9178 dp->dtdo_refcnt++;
9179 ASSERT(dp->dtdo_refcnt != 0);
9180
9181 /*
9182 * We need to check this DIF object for references to the variable
9183 * DIF_VAR_VTIMESTAMP.
9184 */
9185 for (i = 0; i < dp->dtdo_varlen; i++) {
9186 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9187
9188 if (v->dtdv_id != DIF_VAR_VTIMESTAMP)
9189 continue;
9190
9191 if (dtrace_vtime_references++ == 0)
9192 dtrace_vtime_enable();
9193 }
9194}
9195
9196/*
9197 * This routine calculates the dynamic variable chunksize for a given DIF
9198 * object. The calculation is not fool-proof, and can probably be tricked by
9199 * malicious DIF -- but it works for all compiler-generated DIF. Because this
9200 * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail
9201 * if a dynamic variable size exceeds the chunksize.
9202 */
9203static void
9204dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9205{
9206 uint64_t sval = 0;
9207 dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */
9208 const dif_instr_t *text = dp->dtdo_buf;
9209 uint_t pc, srd = 0;
9210 uint_t ttop = 0;
9211 size_t size, ksize;
9212 uint_t id, i;
9213
9214 for (pc = 0; pc < dp->dtdo_len; pc++) {
9215 dif_instr_t instr = text[pc];
9216 uint_t op = DIF_INSTR_OP(instr);
9217 uint_t rd = DIF_INSTR_RD(instr);
9218 uint_t r1 = DIF_INSTR_R1(instr);
9219 uint_t nkeys = 0;
9220 uchar_t scope = 0;
9221
9222 dtrace_key_t *key = tupregs;
9223
9224 switch (op) {
9225 case DIF_OP_SETX:
9226 sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)];
9227 srd = rd;
9228 continue;
9229
9230 case DIF_OP_STTS:
9231 key = &tupregs[DIF_DTR_NREGS];
9232 key[0].dttk_size = 0;
9233 key[1].dttk_size = 0;
9234 nkeys = 2;
9235 scope = DIFV_SCOPE_THREAD;
9236 break;
9237
9238 case DIF_OP_STGAA:
9239 case DIF_OP_STTAA:
9240 nkeys = ttop;
9241
9242 if (DIF_INSTR_OP(instr) == DIF_OP_STTAA)
9243 key[nkeys++].dttk_size = 0;
9244
9245 key[nkeys++].dttk_size = 0;
9246
9247 if (op == DIF_OP_STTAA) {
9248 scope = DIFV_SCOPE_THREAD;
9249 } else {
9250 scope = DIFV_SCOPE_GLOBAL;
9251 }
9252
9253 break;
9254
9255 case DIF_OP_PUSHTR:
9256 if (ttop == DIF_DTR_NREGS)
9257 return;
9258
9259 if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) {
9260 /*
9261 * If the register for the size of the "pushtr"
9262 * is %r0 (or the value is 0) and the type is
9263 * a string, we'll use the system-wide default
9264 * string size.
9265 */
9266 tupregs[ttop++].dttk_size =
9267 dtrace_strsize_default;
9268 } else {
9269 if (srd == 0)
9270 return;
9271
9272 tupregs[ttop++].dttk_size = sval;
9273 }
9274
9275 break;
9276
9277 case DIF_OP_PUSHTV:
9278 if (ttop == DIF_DTR_NREGS)
9279 return;
9280
9281 tupregs[ttop++].dttk_size = 0;
9282 break;
9283
9284 case DIF_OP_FLUSHTS:
9285 ttop = 0;
9286 break;
9287
9288 case DIF_OP_POPTS:
9289 if (ttop != 0)
9290 ttop--;
9291 break;
9292 }
9293
9294 sval = 0;
9295 srd = 0;
9296
9297 if (nkeys == 0)
9298 continue;
9299
9300 /*
9301 * We have a dynamic variable allocation; calculate its size.
9302 */
9303 for (ksize = 0, i = 0; i < nkeys; i++)
9304 ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));
9305
9306 size = sizeof (dtrace_dynvar_t);
9307 size += sizeof (dtrace_key_t) * (nkeys - 1);
9308 size += ksize;
9309
9310 /*
9311 * Now we need to determine the size of the stored data.
9312 */
9313 id = DIF_INSTR_VAR(instr);
9314
9315 for (i = 0; i < dp->dtdo_varlen; i++) {
9316 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9317
9318 if (v->dtdv_id == id && v->dtdv_scope == scope) {
9319 size += v->dtdv_type.dtdt_size;
9320 break;
9321 }
9322 }
9323
9324 if (i == dp->dtdo_varlen)
9325 return;
9326
9327 /*
9328 * We have the size. If this is larger than the chunk size
9329 * for our dynamic variable state, reset the chunk size.
9330 */
9331 size = P2ROUNDUP(size, sizeof (uint64_t));
9332
9333 if (size > vstate->dtvs_dynvars.dtds_chunksize)
9334 vstate->dtvs_dynvars.dtds_chunksize = size;
9335 }
9336}
9337
9338static void
9339dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9340{
9341 int i, oldsvars, osz, nsz, otlocals, ntlocals;
9342 uint_t id;
9343
9344 ASSERT(MUTEX_HELD(&dtrace_lock));
9345 ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0);
9346
9347 for (i = 0; i < dp->dtdo_varlen; i++) {
9348 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9349 dtrace_statvar_t *svar, ***svarp = NULL;
9350 size_t dsize = 0;
9351 uint8_t scope = v->dtdv_scope;
9352 int *np = NULL;
9353
9354 if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE)
9355 continue;
9356
9357 id -= DIF_VAR_OTHER_UBASE;
9358
9359 switch (scope) {
9360 case DIFV_SCOPE_THREAD:
9361 while (id >= (otlocals = vstate->dtvs_ntlocals)) {
9362 dtrace_difv_t *tlocals;
9363
9364 if ((ntlocals = (otlocals << 1)) == 0)
9365 ntlocals = 1;
9366
9367 osz = otlocals * sizeof (dtrace_difv_t);
9368 nsz = ntlocals * sizeof (dtrace_difv_t);
9369
9370 tlocals = kmem_zalloc(nsz, KM_SLEEP);
9371
9372 if (osz != 0) {
9373 bcopy(vstate->dtvs_tlocals,
9374 tlocals, osz);
9375 kmem_free(vstate->dtvs_tlocals, osz);
9376 }
9377
9378 vstate->dtvs_tlocals = tlocals;
9379 vstate->dtvs_ntlocals = ntlocals;
9380 }
9381
9382 vstate->dtvs_tlocals[id] = *v;
9383 continue;
9384
9385 case DIFV_SCOPE_LOCAL:
9386 np = &vstate->dtvs_nlocals;
9387 svarp = &vstate->dtvs_locals;
9388
9389 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF)
9390 dsize = NCPU * (v->dtdv_type.dtdt_size +
9391 sizeof (uint64_t));
9392 else
9393 dsize = NCPU * sizeof (uint64_t);
9394
9395 break;
9396
9397 case DIFV_SCOPE_GLOBAL:
9398 np = &vstate->dtvs_nglobals;
9399 svarp = &vstate->dtvs_globals;
9400
9401 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF)
9402 dsize = v->dtdv_type.dtdt_size +
9403 sizeof (uint64_t);
9404
9405 break;
9406
9407 default:
9408 ASSERT(0);
9409 }
9410
9411 while (id >= (oldsvars = *np)) {
9412 dtrace_statvar_t **statics;
9413 int newsvars, oldsize, newsize;
9414
9415 if ((newsvars = (oldsvars << 1)) == 0)
9416 newsvars = 1;
9417
9418 oldsize = oldsvars * sizeof (dtrace_statvar_t *);
9419 newsize = newsvars * sizeof (dtrace_statvar_t *);
9420
9421 statics = kmem_zalloc(newsize, KM_SLEEP);
9422
9423 if (oldsize != 0) {
9424 bcopy(*svarp, statics, oldsize);
9425 kmem_free(*svarp, oldsize);
9426 }
9427
9428 *svarp = statics;
9429 *np = newsvars;
9430 }
9431
9432 if ((svar = (*svarp)[id]) == NULL) {
9433 svar = kmem_zalloc(sizeof (dtrace_statvar_t), KM_SLEEP);
9434 svar->dtsv_var = *v;
9435
9436 if ((svar->dtsv_size = dsize) != 0) {
9437 svar->dtsv_data = (uint64_t)(uintptr_t)
9438 kmem_zalloc(dsize, KM_SLEEP);
9439 }
9440
9441 (*svarp)[id] = svar;
9442 }
9443
9444 svar->dtsv_refcnt++;
9445 }
9446
9447 dtrace_difo_chunksize(dp, vstate);
9448 dtrace_difo_hold(dp);
9449}
9450
9451static dtrace_difo_t *
9452dtrace_difo_duplicate(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9453{
9454 dtrace_difo_t *new;
9455 size_t sz;
9456
9457 ASSERT(dp->dtdo_buf != NULL);
9458 ASSERT(dp->dtdo_refcnt != 0);
9459
9460 new = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP);
9461
9462 ASSERT(dp->dtdo_buf != NULL);
9463 sz = dp->dtdo_len * sizeof (dif_instr_t);
9464 new->dtdo_buf = kmem_alloc(sz, KM_SLEEP);
9465 bcopy(dp->dtdo_buf, new->dtdo_buf, sz);
9466 new->dtdo_len = dp->dtdo_len;
9467
9468 if (dp->dtdo_strtab != NULL) {
9469 ASSERT(dp->dtdo_strlen != 0);
9470 new->dtdo_strtab = kmem_alloc(dp->dtdo_strlen, KM_SLEEP);
9471 bcopy(dp->dtdo_strtab, new->dtdo_strtab, dp->dtdo_strlen);
9472 new->dtdo_strlen = dp->dtdo_strlen;
9473 }
9474
9475 if (dp->dtdo_inttab != NULL) {
9476 ASSERT(dp->dtdo_intlen != 0);
9477 sz = dp->dtdo_intlen * sizeof (uint64_t);
9478 new->dtdo_inttab = kmem_alloc(sz, KM_SLEEP);
9479 bcopy(dp->dtdo_inttab, new->dtdo_inttab, sz);
9480 new->dtdo_intlen = dp->dtdo_intlen;
9481 }
9482
9483 if (dp->dtdo_vartab != NULL) {
9484 ASSERT(dp->dtdo_varlen != 0);
9485 sz = dp->dtdo_varlen * sizeof (dtrace_difv_t);
9486 new->dtdo_vartab = kmem_alloc(sz, KM_SLEEP);
9487 bcopy(dp->dtdo_vartab, new->dtdo_vartab, sz);
9488 new->dtdo_varlen = dp->dtdo_varlen;
9489 }
9490
9491 dtrace_difo_init(new, vstate);
9492 return (new);
9493}
9494
9495static void
9496dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9497{
9498 int i;
9499
9500 ASSERT(dp->dtdo_refcnt == 0);
9501
9502 for (i = 0; i < dp->dtdo_varlen; i++) {
9503 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9504 dtrace_statvar_t *svar, **svarp = NULL;
9505 uint_t id;
9506 uint8_t scope = v->dtdv_scope;
9507 int *np = NULL;
9508
9509 switch (scope) {
9510 case DIFV_SCOPE_THREAD:
9511 continue;
9512
9513 case DIFV_SCOPE_LOCAL:
9514 np = &vstate->dtvs_nlocals;
9515 svarp = vstate->dtvs_locals;
9516 break;
9517
9518 case DIFV_SCOPE_GLOBAL:
9519 np = &vstate->dtvs_nglobals;
9520 svarp = vstate->dtvs_globals;
9521 break;
9522
9523 default:
9524 ASSERT(0);
9525 }
9526
9527 if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE)
9528 continue;
9529
9530 id -= DIF_VAR_OTHER_UBASE;
9531 ASSERT(id < *np);
9532
9533 svar = svarp[id];
9534 ASSERT(svar != NULL);
9535 ASSERT(svar->dtsv_refcnt > 0);
9536
9537 if (--svar->dtsv_refcnt > 0)
9538 continue;
9539
9540 if (svar->dtsv_size != 0) {
9541 ASSERT(svar->dtsv_data != 0);
9542 kmem_free((void *)(uintptr_t)svar->dtsv_data,
9543 svar->dtsv_size);
9544 }
9545
9546 kmem_free(svar, sizeof (dtrace_statvar_t));
9547 svarp[id] = NULL;
9548 }
9549
9550 if (dp->dtdo_buf != NULL)
9551 kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t));
9552 if (dp->dtdo_inttab != NULL)
9553 kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t));
9554 if (dp->dtdo_strtab != NULL)
9555 kmem_free(dp->dtdo_strtab, dp->dtdo_strlen);
9556 if (dp->dtdo_vartab != NULL)
9557 kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t));
9558
9559 kmem_free(dp, sizeof (dtrace_difo_t));
9560}
9561
9562static void
9563dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9564{
9565 int i;
9566
9567 ASSERT(MUTEX_HELD(&dtrace_lock));
9568 ASSERT(dp->dtdo_refcnt != 0);
9569
9570 for (i = 0; i < dp->dtdo_varlen; i++) {
9571 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9572
9573 if (v->dtdv_id != DIF_VAR_VTIMESTAMP)
9574 continue;
9575
9576 ASSERT(dtrace_vtime_references > 0);
9577 if (--dtrace_vtime_references == 0)
9578 dtrace_vtime_disable();
9579 }
9580
9581 if (--dp->dtdo_refcnt == 0)
9582 dtrace_difo_destroy(dp, vstate);
9583}
9584
9585/*
9586 * DTrace Format Functions
9587 */
9588static uint16_t
9589dtrace_format_add(dtrace_state_t *state, char *str)
9590{
9591 char *fmt, **new;
9592 uint16_t ndx, len = strlen(str) + 1;
9593
9594 fmt = kmem_zalloc(len, KM_SLEEP);
9595 bcopy(str, fmt, len);
9596
9597 for (ndx = 0; ndx < state->dts_nformats; ndx++) {
9598 if (state->dts_formats[ndx] == NULL) {
9599 state->dts_formats[ndx] = fmt;
9600 return (ndx + 1);
9601 }
9602 }
9603
9604 if (state->dts_nformats == USHRT_MAX) {
9605 /*
9606 * This is only likely if a denial-of-service attack is being
9607 * attempted. As such, it's okay to fail silently here.
9608 */
9609 kmem_free(fmt, len);
9610 return (0);
9611 }
9612
9613 /*
9614 * For simplicity, we always resize the formats array to be exactly the
9615 * number of formats.
9616 */
9617 ndx = state->dts_nformats++;
9618 new = kmem_alloc((ndx + 1) * sizeof (char *), KM_SLEEP);
9619
9620 if (state->dts_formats != NULL) {
9621 ASSERT(ndx != 0);
9622 bcopy(state->dts_formats, new, ndx * sizeof (char *));
9623 kmem_free(state->dts_formats, ndx * sizeof (char *));
9624 }
9625
9626 state->dts_formats = new;
9627 state->dts_formats[ndx] = fmt;
9628
9629 return (ndx + 1);
9630}
9631
9632static void
9633dtrace_format_remove(dtrace_state_t *state, uint16_t format)
9634{
9635 char *fmt;
9636
9637 ASSERT(state->dts_formats != NULL);
9638 ASSERT(format <= state->dts_nformats);
9639 ASSERT(state->dts_formats[format - 1] != NULL);
9640
9641 fmt = state->dts_formats[format - 1];
9642 kmem_free(fmt, strlen(fmt) + 1);
9643 state->dts_formats[format - 1] = NULL;
9644}
9645
9646static void
9647dtrace_format_destroy(dtrace_state_t *state)
9648{
9649 int i;
9650
9651 if (state->dts_nformats == 0) {
9652 ASSERT(state->dts_formats == NULL);
9653 return;
9654 }
9655
9656 ASSERT(state->dts_formats != NULL);
9657
9658 for (i = 0; i < state->dts_nformats; i++) {
9659 char *fmt = state->dts_formats[i];
9660
9661 if (fmt == NULL)
9662 continue;
9663
9664 kmem_free(fmt, strlen(fmt) + 1);
9665 }
9666
9667 kmem_free(state->dts_formats, state->dts_nformats * sizeof (char *));
9668 state->dts_nformats = 0;
9669 state->dts_formats = NULL;
9670}
9671
9672/*
9673 * DTrace Predicate Functions
9674 */
9675static dtrace_predicate_t *
9676dtrace_predicate_create(dtrace_difo_t *dp)
9677{
9678 dtrace_predicate_t *pred;
9679
9680 ASSERT(MUTEX_HELD(&dtrace_lock));
9681 ASSERT(dp->dtdo_refcnt != 0);
9682
9683 pred = kmem_zalloc(sizeof (dtrace_predicate_t), KM_SLEEP);
9684 pred->dtp_difo = dp;
9685 pred->dtp_refcnt = 1;
9686
9687 if (!dtrace_difo_cacheable(dp))
9688 return (pred);
9689
9690 if (dtrace_predcache_id == DTRACE_CACHEIDNONE) {
9691 /*
9692 * This is only theoretically possible -- we have had 2^32
9693 * cacheable predicates on this machine. We cannot allow any
9694 * more predicates to become cacheable: as unlikely as it is,
9695 * there may be a thread caching a (now stale) predicate cache
9696 * ID. (N.B.: the temptation is being successfully resisted to
9697 * have this cmn_err() "Holy shit -- we executed this code!")
9698 */
9699 return (pred);
9700 }
9701
9702 pred->dtp_cacheid = dtrace_predcache_id++;
9703
9704 return (pred);
9705}
9706
9707static void
9708dtrace_predicate_hold(dtrace_predicate_t *pred)
9709{
9710 ASSERT(MUTEX_HELD(&dtrace_lock));
9711 ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0);
9712 ASSERT(pred->dtp_refcnt > 0);
9713
9714 pred->dtp_refcnt++;
9715}
9716
9717static void
9718dtrace_predicate_release(dtrace_predicate_t *pred, dtrace_vstate_t *vstate)
9719{
9720 dtrace_difo_t *dp = pred->dtp_difo;
9721
9722 ASSERT(MUTEX_HELD(&dtrace_lock));
9723 ASSERT(dp != NULL && dp->dtdo_refcnt != 0);
9724 ASSERT(pred->dtp_refcnt > 0);
9725
9726 if (--pred->dtp_refcnt == 0) {
9727 dtrace_difo_release(pred->dtp_difo, vstate);
9728 kmem_free(pred, sizeof (dtrace_predicate_t));
9729 }
9730}
9731
9732/*
9733 * DTrace Action Description Functions
9734 */
9735static dtrace_actdesc_t *
9736dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple,
9737 uint64_t uarg, uint64_t arg)
9738{
9739 dtrace_actdesc_t *act;
9740
9741#if defined(sun)
9742 ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || (arg != NULL &&
9743 arg >= KERNELBASE) || (arg == NULL && kind == DTRACEACT_PRINTA));
9744#endif
9745
9746 act = kmem_zalloc(sizeof (dtrace_actdesc_t), KM_SLEEP);
9747 act->dtad_kind = kind;
9748 act->dtad_ntuple = ntuple;
9749 act->dtad_uarg = uarg;
9750 act->dtad_arg = arg;
9751 act->dtad_refcnt = 1;
9752
9753 return (act);
9754}
9755
9756static void
9757dtrace_actdesc_hold(dtrace_actdesc_t *act)
9758{
9759 ASSERT(act->dtad_refcnt >= 1);
9760 act->dtad_refcnt++;
9761}
9762
9763static void
9764dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate)
9765{
9766 dtrace_actkind_t kind = act->dtad_kind;
9767 dtrace_difo_t *dp;
9768
9769 ASSERT(act->dtad_refcnt >= 1);
9770
9771 if (--act->dtad_refcnt != 0)
9772 return;
9773
9774 if ((dp = act->dtad_difo) != NULL)
9775 dtrace_difo_release(dp, vstate);
9776
9777 if (DTRACEACT_ISPRINTFLIKE(kind)) {
9778 char *str = (char *)(uintptr_t)act->dtad_arg;
9779
9780#if defined(sun)
9781 ASSERT((str != NULL && (uintptr_t)str >= KERNELBASE) ||
9782 (str == NULL && act->dtad_kind == DTRACEACT_PRINTA));
9783#endif
9784
9785 if (str != NULL)
9786 kmem_free(str, strlen(str) + 1);
9787 }
9788
9789 kmem_free(act, sizeof (dtrace_actdesc_t));
9790}
9791
9792/*
9793 * DTrace ECB Functions
9794 */
9795static dtrace_ecb_t *
9796dtrace_ecb_add(dtrace_state_t *state, dtrace_probe_t *probe)
9797{
9798 dtrace_ecb_t *ecb;
9799 dtrace_epid_t epid;
9800
9801 ASSERT(MUTEX_HELD(&dtrace_lock));
9802
9803 ecb = kmem_zalloc(sizeof (dtrace_ecb_t), KM_SLEEP);
9804 ecb->dte_predicate = NULL;
9805 ecb->dte_probe = probe;
9806
9807 /*
9808 * The default size is the size of the default action: recording
9809 * the header.
9810 */
9811 ecb->dte_size = ecb->dte_needed = sizeof (dtrace_rechdr_t);
9812 ecb->dte_alignment = sizeof (dtrace_epid_t);
9813
9814 epid = state->dts_epid++;
9815
9816 if (epid - 1 >= state->dts_necbs) {
9817 dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs;
9818 int necbs = state->dts_necbs << 1;
9819
9820 ASSERT(epid == state->dts_necbs + 1);
9821
9822 if (necbs == 0) {
9823 ASSERT(oecbs == NULL);
9824 necbs = 1;
9825 }
9826
9827 ecbs = kmem_zalloc(necbs * sizeof (*ecbs), KM_SLEEP);
9828
9829 if (oecbs != NULL)
9830 bcopy(oecbs, ecbs, state->dts_necbs * sizeof (*ecbs));
9831
9832 dtrace_membar_producer();
9833 state->dts_ecbs = ecbs;
9834
9835 if (oecbs != NULL) {
9836 /*
9837 * If this state is active, we must dtrace_sync()
9838 * before we can free the old dts_ecbs array: we're
9839 * coming in hot, and there may be active ring
9840 * buffer processing (which indexes into the dts_ecbs
9841 * array) on another CPU.
9842 */
9843 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
9844 dtrace_sync();
9845
9846 kmem_free(oecbs, state->dts_necbs * sizeof (*ecbs));
9847 }
9848
9849 dtrace_membar_producer();
9850 state->dts_necbs = necbs;
9851 }
9852
9853 ecb->dte_state = state;
9854
9855 ASSERT(state->dts_ecbs[epid - 1] == NULL);
9856 dtrace_membar_producer();
9857 state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb;
9858
9859 return (ecb);
9860}
9861
9862static void
9863dtrace_ecb_enable(dtrace_ecb_t *ecb)
9864{
9865 dtrace_probe_t *probe = ecb->dte_probe;
9866
9867 ASSERT(MUTEX_HELD(&cpu_lock));
9868 ASSERT(MUTEX_HELD(&dtrace_lock));
9869 ASSERT(ecb->dte_next == NULL);
9870
9871 if (probe == NULL) {
9872 /*
9873 * This is the NULL probe -- there's nothing to do.
9874 */
9875 return;
9876 }
9877
9878 if (probe->dtpr_ecb == NULL) {
9879 dtrace_provider_t *prov = probe->dtpr_provider;
9880
9881 /*
9882 * We're the first ECB on this probe.
9883 */
9884 probe->dtpr_ecb = probe->dtpr_ecb_last = ecb;
9885
9886 if (ecb->dte_predicate != NULL)
9887 probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid;
9888
9889 prov->dtpv_pops.dtps_enable(prov->dtpv_arg,
9890 probe->dtpr_id, probe->dtpr_arg);
9891 } else {
9892 /*
9893 * This probe is already active. Swing the last pointer to
9894 * point to the new ECB, and issue a dtrace_sync() to assure
9895 * that all CPUs have seen the change.
9896 */
9897 ASSERT(probe->dtpr_ecb_last != NULL);
9898 probe->dtpr_ecb_last->dte_next = ecb;
9899 probe->dtpr_ecb_last = ecb;
9900 probe->dtpr_predcache = 0;
9901
9902 dtrace_sync();
9903 }
9904}
9905
9906static void
9907dtrace_ecb_resize(dtrace_ecb_t *ecb)
9908{
9909 dtrace_action_t *act;
9910 uint32_t curneeded = UINT32_MAX;
9911 uint32_t aggbase = UINT32_MAX;
9912
9913 /*
9914 * If we record anything, we always record the dtrace_rechdr_t. (And
9915 * we always record it first.)
9916 */
9917 ecb->dte_size = sizeof (dtrace_rechdr_t);
9918 ecb->dte_alignment = sizeof (dtrace_epid_t);
9919
9920 for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
9921 dtrace_recdesc_t *rec = &act->dta_rec;
9922 ASSERT(rec->dtrd_size > 0 || rec->dtrd_alignment == 1);
9923
9924 ecb->dte_alignment = MAX(ecb->dte_alignment,
9925 rec->dtrd_alignment);
9926
9927 if (DTRACEACT_ISAGG(act->dta_kind)) {
9928 dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act;
9929
9930 ASSERT(rec->dtrd_size != 0);
9931 ASSERT(agg->dtag_first != NULL);
9932 ASSERT(act->dta_prev->dta_intuple);
9933 ASSERT(aggbase != UINT32_MAX);
9934 ASSERT(curneeded != UINT32_MAX);
9935
9936 agg->dtag_base = aggbase;
9937
9938 curneeded = P2ROUNDUP(curneeded, rec->dtrd_alignment);
9939 rec->dtrd_offset = curneeded;
9940 curneeded += rec->dtrd_size;
9941 ecb->dte_needed = MAX(ecb->dte_needed, curneeded);
9942
9943 aggbase = UINT32_MAX;
9944 curneeded = UINT32_MAX;
9945 } else if (act->dta_intuple) {
9946 if (curneeded == UINT32_MAX) {
9947 /*
9948 * This is the first record in a tuple. Align
9949 * curneeded to be at offset 4 in an 8-byte
9950 * aligned block.
9951 */
9952 ASSERT(act->dta_prev == NULL ||
9953 !act->dta_prev->dta_intuple);
9954 ASSERT3U(aggbase, ==, UINT32_MAX);
9955 curneeded = P2PHASEUP(ecb->dte_size,
9956 sizeof (uint64_t), sizeof (dtrace_aggid_t));
9957
9958 aggbase = curneeded - sizeof (dtrace_aggid_t);
9959 ASSERT(IS_P2ALIGNED(aggbase,
9960 sizeof (uint64_t)));
9961 }
9962 curneeded = P2ROUNDUP(curneeded, rec->dtrd_alignment);
9963 rec->dtrd_offset = curneeded;
9964 curneeded += rec->dtrd_size;
9965 } else {
9966 /* tuples must be followed by an aggregation */
9967 ASSERT(act->dta_prev == NULL ||
9968 !act->dta_prev->dta_intuple);
9969
9970 ecb->dte_size = P2ROUNDUP(ecb->dte_size,
9971 rec->dtrd_alignment);
9972 rec->dtrd_offset = ecb->dte_size;
9973 ecb->dte_size += rec->dtrd_size;
9974 ecb->dte_needed = MAX(ecb->dte_needed, ecb->dte_size);
9975 }
9976 }
9977
9978 if ((act = ecb->dte_action) != NULL &&
9979 !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) &&
9980 ecb->dte_size == sizeof (dtrace_rechdr_t)) {
9981 /*
9982 * If the size is still sizeof (dtrace_rechdr_t), then all
9983 * actions store no data; set the size to 0.
9984 */
9985 ecb->dte_size = 0;
9986 }
9987
9988 ecb->dte_size = P2ROUNDUP(ecb->dte_size, sizeof (dtrace_epid_t));
9989 ecb->dte_needed = P2ROUNDUP(ecb->dte_needed, (sizeof (dtrace_epid_t)));
9990 ecb->dte_state->dts_needed = MAX(ecb->dte_state->dts_needed,
9991 ecb->dte_needed);
9992}
9993
9994static dtrace_action_t *
9995dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc)
9996{
9997 dtrace_aggregation_t *agg;
9998 size_t size = sizeof (uint64_t);
9999 int ntuple = desc->dtad_ntuple;
10000 dtrace_action_t *act;
10001 dtrace_recdesc_t *frec;
10002 dtrace_aggid_t aggid;
10003 dtrace_state_t *state = ecb->dte_state;
10004
10005 agg = kmem_zalloc(sizeof (dtrace_aggregation_t), KM_SLEEP);
10006 agg->dtag_ecb = ecb;
10007
10008 ASSERT(DTRACEACT_ISAGG(desc->dtad_kind));
10009
10010 switch (desc->dtad_kind) {
10011 case DTRACEAGG_MIN:
10012 agg->dtag_initial = INT64_MAX;
10013 agg->dtag_aggregate = dtrace_aggregate_min;
10014 break;
10015
10016 case DTRACEAGG_MAX:
10017 agg->dtag_initial = INT64_MIN;
10018 agg->dtag_aggregate = dtrace_aggregate_max;
10019 break;
10020
10021 case DTRACEAGG_COUNT:
10022 agg->dtag_aggregate = dtrace_aggregate_count;
10023 break;
10024
10025 case DTRACEAGG_QUANTIZE:
10026 agg->dtag_aggregate = dtrace_aggregate_quantize;
10027 size = (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) *
10028 sizeof (uint64_t);
10029 break;
10030
10031 case DTRACEAGG_LQUANTIZE: {
10032 uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg);
10033 uint16_t levels = DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg);
10034
10035 agg->dtag_initial = desc->dtad_arg;
10036 agg->dtag_aggregate = dtrace_aggregate_lquantize;
10037
10038 if (step == 0 || levels == 0)
10039 goto err;
10040
10041 size = levels * sizeof (uint64_t) + 3 * sizeof (uint64_t);
10042 break;
10043 }
10044
10045 case DTRACEAGG_LLQUANTIZE: {
10046 uint16_t factor = DTRACE_LLQUANTIZE_FACTOR(desc->dtad_arg);
10047 uint16_t low = DTRACE_LLQUANTIZE_LOW(desc->dtad_arg);
10048 uint16_t high = DTRACE_LLQUANTIZE_HIGH(desc->dtad_arg);
10049 uint16_t nsteps = DTRACE_LLQUANTIZE_NSTEP(desc->dtad_arg);
10050 int64_t v;
10051
10052 agg->dtag_initial = desc->dtad_arg;
10053 agg->dtag_aggregate = dtrace_aggregate_llquantize;
10054
10055 if (factor < 2 || low >= high || nsteps < factor)
10056 goto err;
10057
10058 /*
10059 * Now check that the number of steps evenly divides a power
10060 * of the factor. (This assures both integer bucket size and
10061 * linearity within each magnitude.)
10062 */
10063 for (v = factor; v < nsteps; v *= factor)
10064 continue;
10065
10066 if ((v % nsteps) || (nsteps % factor))
10067 goto err;
10068
10069 size = (dtrace_aggregate_llquantize_bucket(factor,
10070 low, high, nsteps, INT64_MAX) + 2) * sizeof (uint64_t);
10071 break;
10072 }
10073
10074 case DTRACEAGG_AVG:
10075 agg->dtag_aggregate = dtrace_aggregate_avg;
10076 size = sizeof (uint64_t) * 2;
10077 break;
10078
10079 case DTRACEAGG_STDDEV:
10080 agg->dtag_aggregate = dtrace_aggregate_stddev;
10081 size = sizeof (uint64_t) * 4;
10082 break;
10083
10084 case DTRACEAGG_SUM:
10085 agg->dtag_aggregate = dtrace_aggregate_sum;
10086 break;
10087
10088 default:
10089 goto err;
10090 }
10091
10092 agg->dtag_action.dta_rec.dtrd_size = size;
10093
10094 if (ntuple == 0)
10095 goto err;
10096
10097 /*
10098 * We must make sure that we have enough actions for the n-tuple.
10099 */
10100 for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) {
10101 if (DTRACEACT_ISAGG(act->dta_kind))
10102 break;
10103
10104 if (--ntuple == 0) {
10105 /*
10106 * This is the action with which our n-tuple begins.
10107 */
10108 agg->dtag_first = act;
10109 goto success;
10110 }
10111 }
10112
10113 /*
10114 * This n-tuple is short by ntuple elements. Return failure.
10115 */
10116 ASSERT(ntuple != 0);
10117err:
10118 kmem_free(agg, sizeof (dtrace_aggregation_t));
10119 return (NULL);
10120
10121success:
10122 /*
10123 * If the last action in the tuple has a size of zero, it's actually
10124 * an expression argument for the aggregating action.
10125 */
10126 ASSERT(ecb->dte_action_last != NULL);
10127 act = ecb->dte_action_last;
10128
10129 if (act->dta_kind == DTRACEACT_DIFEXPR) {
10130 ASSERT(act->dta_difo != NULL);
10131
10132 if (act->dta_difo->dtdo_rtype.dtdt_size == 0)
10133 agg->dtag_hasarg = 1;
10134 }
10135
10136 /*
10137 * We need to allocate an id for this aggregation.
10138 */
10139#if defined(sun)
10140 aggid = (dtrace_aggid_t)(uintptr_t)vmem_alloc(state->dts_aggid_arena, 1,
10141 VM_BESTFIT | VM_SLEEP);
10142#else
10143 aggid = alloc_unr(state->dts_aggid_arena);
10144#endif
10145
10146 if (aggid - 1 >= state->dts_naggregations) {
10147 dtrace_aggregation_t **oaggs = state->dts_aggregations;
10148 dtrace_aggregation_t **aggs;
10149 int naggs = state->dts_naggregations << 1;
10150 int onaggs = state->dts_naggregations;
10151
10152 ASSERT(aggid == state->dts_naggregations + 1);
10153
10154 if (naggs == 0) {
10155 ASSERT(oaggs == NULL);
10156 naggs = 1;
10157 }
10158
10159 aggs = kmem_zalloc(naggs * sizeof (*aggs), KM_SLEEP);
10160
10161 if (oaggs != NULL) {
10162 bcopy(oaggs, aggs, onaggs * sizeof (*aggs));
10163 kmem_free(oaggs, onaggs * sizeof (*aggs));
10164 }
10165
10166 state->dts_aggregations = aggs;
10167 state->dts_naggregations = naggs;
10168 }
10169
10170 ASSERT(state->dts_aggregations[aggid - 1] == NULL);
10171 state->dts_aggregations[(agg->dtag_id = aggid) - 1] = agg;
10172
10173 frec = &agg->dtag_first->dta_rec;
10174 if (frec->dtrd_alignment < sizeof (dtrace_aggid_t))
10175 frec->dtrd_alignment = sizeof (dtrace_aggid_t);
10176
10177 for (act = agg->dtag_first; act != NULL; act = act->dta_next) {
10178 ASSERT(!act->dta_intuple);
10179 act->dta_intuple = 1;
10180 }
10181
10182 return (&agg->dtag_action);
10183}
10184
10185static void
10186dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act)
10187{
10188 dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act;
10189 dtrace_state_t *state = ecb->dte_state;
10190 dtrace_aggid_t aggid = agg->dtag_id;
10191
10192 ASSERT(DTRACEACT_ISAGG(act->dta_kind));
10193#if defined(sun)
10194 vmem_free(state->dts_aggid_arena, (void *)(uintptr_t)aggid, 1);
10195#else
10196 free_unr(state->dts_aggid_arena, aggid);
10197#endif
10198
10199 ASSERT(state->dts_aggregations[aggid - 1] == agg);
10200 state->dts_aggregations[aggid - 1] = NULL;
10201
10202 kmem_free(agg, sizeof (dtrace_aggregation_t));
10203}
10204
10205static int
10206dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc)
10207{
10208 dtrace_action_t *action, *last;
10209 dtrace_difo_t *dp = desc->dtad_difo;
10210 uint32_t size = 0, align = sizeof (uint8_t), mask;
10211 uint16_t format = 0;
10212 dtrace_recdesc_t *rec;
10213 dtrace_state_t *state = ecb->dte_state;
10214 dtrace_optval_t *opt = state->dts_options, nframes = 0, strsize;
10215 uint64_t arg = desc->dtad_arg;
10216
10217 ASSERT(MUTEX_HELD(&dtrace_lock));
10218 ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1);
10219
10220 if (DTRACEACT_ISAGG(desc->dtad_kind)) {
10221 /*
10222 * If this is an aggregating action, there must be neither
10223 * a speculate nor a commit on the action chain.
10224 */
10225 dtrace_action_t *act;
10226
10227 for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
10228 if (act->dta_kind == DTRACEACT_COMMIT)
10229 return (EINVAL);
10230
10231 if (act->dta_kind == DTRACEACT_SPECULATE)
10232 return (EINVAL);
10233 }
10234
10235 action = dtrace_ecb_aggregation_create(ecb, desc);
10236
10237 if (action == NULL)
10238 return (EINVAL);
10239 } else {
10240 if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) ||
10241 (desc->dtad_kind == DTRACEACT_DIFEXPR &&
10242 dp != NULL && dp->dtdo_destructive)) {
10243 state->dts_destructive = 1;
10244 }
10245
10246 switch (desc->dtad_kind) {
10247 case DTRACEACT_PRINTF:
10248 case DTRACEACT_PRINTA:
10249 case DTRACEACT_SYSTEM:
10250 case DTRACEACT_FREOPEN:
10251 case DTRACEACT_DIFEXPR:
10252 /*
10253 * We know that our arg is a string -- turn it into a
10254 * format.
10255 */
10256 if (arg == 0) {
10257 ASSERT(desc->dtad_kind == DTRACEACT_PRINTA ||
10258 desc->dtad_kind == DTRACEACT_DIFEXPR);
10259 format = 0;
10260 } else {
10261 ASSERT(arg != 0);
10262#if defined(sun)
10263 ASSERT(arg > KERNELBASE);
10264#endif
10265 format = dtrace_format_add(state,
10266 (char *)(uintptr_t)arg);
10267 }
10268
10269 /*FALLTHROUGH*/
10270 case DTRACEACT_LIBACT:
10271 case DTRACEACT_TRACEMEM:
10272 case DTRACEACT_TRACEMEM_DYNSIZE:
10273 if (dp == NULL)
10274 return (EINVAL);
10275
10276 if ((size = dp->dtdo_rtype.dtdt_size) != 0)
10277 break;
10278
10279 if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) {
10280 if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10281 return (EINVAL);
10282
10283 size = opt[DTRACEOPT_STRSIZE];
10284 }
10285
10286 break;
10287
10288 case DTRACEACT_STACK:
10289 if ((nframes = arg) == 0) {
10290 nframes = opt[DTRACEOPT_STACKFRAMES];
10291 ASSERT(nframes > 0);
10292 arg = nframes;
10293 }
10294
10295 size = nframes * sizeof (pc_t);
10296 break;
10297
10298 case DTRACEACT_JSTACK:
10299 if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0)
10300 strsize = opt[DTRACEOPT_JSTACKSTRSIZE];
10301
10302 if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0)
10303 nframes = opt[DTRACEOPT_JSTACKFRAMES];
10304
10305 arg = DTRACE_USTACK_ARG(nframes, strsize);
10306
10307 /*FALLTHROUGH*/
10308 case DTRACEACT_USTACK:
10309 if (desc->dtad_kind != DTRACEACT_JSTACK &&
10310 (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) {
10311 strsize = DTRACE_USTACK_STRSIZE(arg);
10312 nframes = opt[DTRACEOPT_USTACKFRAMES];
10313 ASSERT(nframes > 0);
10314 arg = DTRACE_USTACK_ARG(nframes, strsize);
10315 }
10316
10317 /*
10318 * Save a slot for the pid.
10319 */
10320 size = (nframes + 1) * sizeof (uint64_t);
10321 size += DTRACE_USTACK_STRSIZE(arg);
10322 size = P2ROUNDUP(size, (uint32_t)(sizeof (uintptr_t)));
10323
10324 break;
10325
10326 case DTRACEACT_SYM:
10327 case DTRACEACT_MOD:
10328 if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) !=
10329 sizeof (uint64_t)) ||
10330 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10331 return (EINVAL);
10332 break;
10333
10334 case DTRACEACT_USYM:
10335 case DTRACEACT_UMOD:
10336 case DTRACEACT_UADDR:
10337 if (dp == NULL ||
10338 (dp->dtdo_rtype.dtdt_size != sizeof (uint64_t)) ||
10339 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10340 return (EINVAL);
10341
10342 /*
10343 * We have a slot for the pid, plus a slot for the
10344 * argument. To keep things simple (aligned with
10345 * bitness-neutral sizing), we store each as a 64-bit
10346 * quantity.
10347 */
10348 size = 2 * sizeof (uint64_t);
10349 break;
10350
10351 case DTRACEACT_STOP:
10352 case DTRACEACT_BREAKPOINT:
10353 case DTRACEACT_PANIC:
10354 break;
10355
10356 case DTRACEACT_CHILL:
10357 case DTRACEACT_DISCARD:
10358 case DTRACEACT_RAISE:
10359 if (dp == NULL)
10360 return (EINVAL);
10361 break;
10362
10363 case DTRACEACT_EXIT:
10364 if (dp == NULL ||
10365 (size = dp->dtdo_rtype.dtdt_size) != sizeof (int) ||
10366 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10367 return (EINVAL);
10368 break;
10369
10370 case DTRACEACT_SPECULATE:
10371 if (ecb->dte_size > sizeof (dtrace_rechdr_t))
10372 return (EINVAL);
10373
10374 if (dp == NULL)
10375 return (EINVAL);
10376
10377 state->dts_speculates = 1;
10378 break;
10379
10380 case DTRACEACT_PRINTM:
10381 size = dp->dtdo_rtype.dtdt_size;
10382 break;
10383
10384 case DTRACEACT_PRINTT:
10385 size = dp->dtdo_rtype.dtdt_size;
10386 break;
10387
10388 case DTRACEACT_COMMIT: {
10389 dtrace_action_t *act = ecb->dte_action;
10390
10391 for (; act != NULL; act = act->dta_next) {
10392 if (act->dta_kind == DTRACEACT_COMMIT)
10393 return (EINVAL);
10394 }
10395
10396 if (dp == NULL)
10397 return (EINVAL);
10398 break;
10399 }
10400
10401 default:
10402 return (EINVAL);
10403 }
10404
10405 if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) {
10406 /*
10407 * If this is a data-storing action or a speculate,
10408 * we must be sure that there isn't a commit on the
10409 * action chain.
10410 */
10411 dtrace_action_t *act = ecb->dte_action;
10412
10413 for (; act != NULL; act = act->dta_next) {
10414 if (act->dta_kind == DTRACEACT_COMMIT)
10415 return (EINVAL);
10416 }
10417 }
10418
10419 action = kmem_zalloc(sizeof (dtrace_action_t), KM_SLEEP);
10420 action->dta_rec.dtrd_size = size;
10421 }
10422
10423 action->dta_refcnt = 1;
10424 rec = &action->dta_rec;
10425 size = rec->dtrd_size;
10426
10427 for (mask = sizeof (uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) {
10428 if (!(size & mask)) {
10429 align = mask + 1;
10430 break;
10431 }
10432 }
10433
10434 action->dta_kind = desc->dtad_kind;
10435
10436 if ((action->dta_difo = dp) != NULL)
10437 dtrace_difo_hold(dp);
10438
10439 rec->dtrd_action = action->dta_kind;
10440 rec->dtrd_arg = arg;
10441 rec->dtrd_uarg = desc->dtad_uarg;
10442 rec->dtrd_alignment = (uint16_t)align;
10443 rec->dtrd_format = format;
10444
10445 if ((last = ecb->dte_action_last) != NULL) {
10446 ASSERT(ecb->dte_action != NULL);
10447 action->dta_prev = last;
10448 last->dta_next = action;
10449 } else {
10450 ASSERT(ecb->dte_action == NULL);
10451 ecb->dte_action = action;
10452 }
10453
10454 ecb->dte_action_last = action;
10455
10456 return (0);
10457}
10458
10459static void
10460dtrace_ecb_action_remove(dtrace_ecb_t *ecb)
10461{
10462 dtrace_action_t *act = ecb->dte_action, *next;
10463 dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate;
10464 dtrace_difo_t *dp;
10465 uint16_t format;
10466
10467 if (act != NULL && act->dta_refcnt > 1) {
10468 ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1);
10469 act->dta_refcnt--;
10470 } else {
10471 for (; act != NULL; act = next) {
10472 next = act->dta_next;
10473 ASSERT(next != NULL || act == ecb->dte_action_last);
10474 ASSERT(act->dta_refcnt == 1);
10475
10476 if ((format = act->dta_rec.dtrd_format) != 0)
10477 dtrace_format_remove(ecb->dte_state, format);
10478
10479 if ((dp = act->dta_difo) != NULL)
10480 dtrace_difo_release(dp, vstate);
10481
10482 if (DTRACEACT_ISAGG(act->dta_kind)) {
10483 dtrace_ecb_aggregation_destroy(ecb, act);
10484 } else {
10485 kmem_free(act, sizeof (dtrace_action_t));
10486 }
10487 }
10488 }
10489
10490 ecb->dte_action = NULL;
10491 ecb->dte_action_last = NULL;
10492 ecb->dte_size = 0;
10493}
10494
10495static void
10496dtrace_ecb_disable(dtrace_ecb_t *ecb)
10497{
10498 /*
10499 * We disable the ECB by removing it from its probe.
10500 */
10501 dtrace_ecb_t *pecb, *prev = NULL;
10502 dtrace_probe_t *probe = ecb->dte_probe;
10503
10504 ASSERT(MUTEX_HELD(&dtrace_lock));
10505
10506 if (probe == NULL) {
10507 /*
10508 * This is the NULL probe; there is nothing to disable.
10509 */
10510 return;
10511 }
10512
10513 for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) {
10514 if (pecb == ecb)
10515 break;
10516 prev = pecb;
10517 }
10518
10519 ASSERT(pecb != NULL);
10520
10521 if (prev == NULL) {
10522 probe->dtpr_ecb = ecb->dte_next;
10523 } else {
10524 prev->dte_next = ecb->dte_next;
10525 }
10526
10527 if (ecb == probe->dtpr_ecb_last) {
10528 ASSERT(ecb->dte_next == NULL);
10529 probe->dtpr_ecb_last = prev;
10530 }
10531
10532 /*
10533 * The ECB has been disconnected from the probe; now sync to assure
10534 * that all CPUs have seen the change before returning.
10535 */
10536 dtrace_sync();
10537
10538 if (probe->dtpr_ecb == NULL) {
10539 /*
10540 * That was the last ECB on the probe; clear the predicate
10541 * cache ID for the probe, disable it and sync one more time
10542 * to assure that we'll never hit it again.
10543 */
10544 dtrace_provider_t *prov = probe->dtpr_provider;
10545
10546 ASSERT(ecb->dte_next == NULL);
10547 ASSERT(probe->dtpr_ecb_last == NULL);
10548 probe->dtpr_predcache = DTRACE_CACHEIDNONE;
10549 prov->dtpv_pops.dtps_disable(prov->dtpv_arg,
10550 probe->dtpr_id, probe->dtpr_arg);
10551 dtrace_sync();
10552 } else {
10553 /*
10554 * There is at least one ECB remaining on the probe. If there
10555 * is _exactly_ one, set the probe's predicate cache ID to be
10556 * the predicate cache ID of the remaining ECB.
10557 */
10558 ASSERT(probe->dtpr_ecb_last != NULL);
10559 ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE);
10560
10561 if (probe->dtpr_ecb == probe->dtpr_ecb_last) {
10562 dtrace_predicate_t *p = probe->dtpr_ecb->dte_predicate;
10563
10564 ASSERT(probe->dtpr_ecb->dte_next == NULL);
10565
10566 if (p != NULL)
10567 probe->dtpr_predcache = p->dtp_cacheid;
10568 }
10569
10570 ecb->dte_next = NULL;
10571 }
10572}
10573
10574static void
10575dtrace_ecb_destroy(dtrace_ecb_t *ecb)
10576{
10577 dtrace_state_t *state = ecb->dte_state;
10578 dtrace_vstate_t *vstate = &state->dts_vstate;
10579 dtrace_predicate_t *pred;
10580 dtrace_epid_t epid = ecb->dte_epid;
10581
10582 ASSERT(MUTEX_HELD(&dtrace_lock));
10583 ASSERT(ecb->dte_next == NULL);
10584 ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb);
10585
10586 if ((pred = ecb->dte_predicate) != NULL)
10587 dtrace_predicate_release(pred, vstate);
10588
10589 dtrace_ecb_action_remove(ecb);
10590
10591 ASSERT(state->dts_ecbs[epid - 1] == ecb);
10592 state->dts_ecbs[epid - 1] = NULL;
10593
10594 kmem_free(ecb, sizeof (dtrace_ecb_t));
10595}
10596
10597static dtrace_ecb_t *
10598dtrace_ecb_create(dtrace_state_t *state, dtrace_probe_t *probe,
10599 dtrace_enabling_t *enab)
10600{
10601 dtrace_ecb_t *ecb;
10602 dtrace_predicate_t *pred;
10603 dtrace_actdesc_t *act;
10604 dtrace_provider_t *prov;
10605 dtrace_ecbdesc_t *desc = enab->dten_current;
10606
10607 ASSERT(MUTEX_HELD(&dtrace_lock));
10608 ASSERT(state != NULL);
10609
10610 ecb = dtrace_ecb_add(state, probe);
10611 ecb->dte_uarg = desc->dted_uarg;
10612
10613 if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) {
10614 dtrace_predicate_hold(pred);
10615 ecb->dte_predicate = pred;
10616 }
10617
10618 if (probe != NULL) {
10619 /*
10620 * If the provider shows more leg than the consumer is old
10621 * enough to see, we need to enable the appropriate implicit
10622 * predicate bits to prevent the ecb from activating at
10623 * revealing times.
10624 *
10625 * Providers specifying DTRACE_PRIV_USER at register time
10626 * are stating that they need the /proc-style privilege
10627 * model to be enforced, and this is what DTRACE_COND_OWNER
10628 * and DTRACE_COND_ZONEOWNER will then do at probe time.
10629 */
10630 prov = probe->dtpr_provider;
10631 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) &&
10632 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER))
10633 ecb->dte_cond |= DTRACE_COND_OWNER;
10634
10635 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLZONE) &&
10636 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER))
10637 ecb->dte_cond |= DTRACE_COND_ZONEOWNER;
10638
10639 /*
10640 * If the provider shows us kernel innards and the user
10641 * is lacking sufficient privilege, enable the
10642 * DTRACE_COND_USERMODE implicit predicate.
10643 */
10644 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) &&
10645 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL))
10646 ecb->dte_cond |= DTRACE_COND_USERMODE;
10647 }
10648
10649 if (dtrace_ecb_create_cache != NULL) {
10650 /*
10651 * If we have a cached ecb, we'll use its action list instead
10652 * of creating our own (saving both time and space).
10653 */
10654 dtrace_ecb_t *cached = dtrace_ecb_create_cache;
10655 dtrace_action_t *act = cached->dte_action;
10656
10657 if (act != NULL) {
10658 ASSERT(act->dta_refcnt > 0);
10659 act->dta_refcnt++;
10660 ecb->dte_action = act;
10661 ecb->dte_action_last = cached->dte_action_last;
10662 ecb->dte_needed = cached->dte_needed;
10663 ecb->dte_size = cached->dte_size;
10664 ecb->dte_alignment = cached->dte_alignment;
10665 }
10666
10667 return (ecb);
10668 }
10669
10670 for (act = desc->dted_action; act != NULL; act = act->dtad_next) {
10671 if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) {
10672 dtrace_ecb_destroy(ecb);
10673 return (NULL);
10674 }
10675 }
10676
10677 dtrace_ecb_resize(ecb);
10678
10679 return (dtrace_ecb_create_cache = ecb);
10680}
10681
10682static int
10683dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg)
10684{
10685 dtrace_ecb_t *ecb;
10686 dtrace_enabling_t *enab = arg;
10687 dtrace_state_t *state = enab->dten_vstate->dtvs_state;
10688
10689 ASSERT(state != NULL);
10690
10691 if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) {
10692 /*
10693 * This probe was created in a generation for which this
10694 * enabling has previously created ECBs; we don't want to
10695 * enable it again, so just kick out.
10696 */
10697 return (DTRACE_MATCH_NEXT);
10698 }
10699
10700 if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL)
10701 return (DTRACE_MATCH_DONE);
10702
10703 dtrace_ecb_enable(ecb);
10704 return (DTRACE_MATCH_NEXT);
10705}
10706
10707static dtrace_ecb_t *
10708dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id)
10709{
10710 dtrace_ecb_t *ecb;
10711
10712 ASSERT(MUTEX_HELD(&dtrace_lock));
10713
10714 if (id == 0 || id > state->dts_necbs)
10715 return (NULL);
10716
10717 ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL);
10718 ASSERT((ecb = state->dts_ecbs[id - 1]) == NULL || ecb->dte_epid == id);
10719
10720 return (state->dts_ecbs[id - 1]);
10721}
10722
10723static dtrace_aggregation_t *
10724dtrace_aggid2agg(dtrace_state_t *state, dtrace_aggid_t id)
10725{
10726 dtrace_aggregation_t *agg;
10727
10728 ASSERT(MUTEX_HELD(&dtrace_lock));
10729
10730 if (id == 0 || id > state->dts_naggregations)
10731 return (NULL);
10732
10733 ASSERT(state->dts_naggregations > 0 && state->dts_aggregations != NULL);
10734 ASSERT((agg = state->dts_aggregations[id - 1]) == NULL ||
10735 agg->dtag_id == id);
10736
10737 return (state->dts_aggregations[id - 1]);
10738}
10739
10740/*
10741 * DTrace Buffer Functions
10742 *
10743 * The following functions manipulate DTrace buffers. Most of these functions
10744 * are called in the context of establishing or processing consumer state;
10745 * exceptions are explicitly noted.
10746 */
10747
10748/*
10749 * Note: called from cross call context. This function switches the two
10750 * buffers on a given CPU. The atomicity of this operation is assured by
10751 * disabling interrupts while the actual switch takes place; the disabling of
10752 * interrupts serializes the execution with any execution of dtrace_probe() on
10753 * the same CPU.
10754 */
10755static void
10756dtrace_buffer_switch(dtrace_buffer_t *buf)
10757{
10758 caddr_t tomax = buf->dtb_tomax;
10759 caddr_t xamot = buf->dtb_xamot;
10760 dtrace_icookie_t cookie;
10761 hrtime_t now;
10762
10763 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
10764 ASSERT(!(buf->dtb_flags & DTRACEBUF_RING));
10765
10766 cookie = dtrace_interrupt_disable();
10767 now = dtrace_gethrtime();
10768 buf->dtb_tomax = xamot;
10769 buf->dtb_xamot = tomax;
10770 buf->dtb_xamot_drops = buf->dtb_drops;
10771 buf->dtb_xamot_offset = buf->dtb_offset;
10772 buf->dtb_xamot_errors = buf->dtb_errors;
10773 buf->dtb_xamot_flags = buf->dtb_flags;
10774 buf->dtb_offset = 0;
10775 buf->dtb_drops = 0;
10776 buf->dtb_errors = 0;
10777 buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED);
10778 buf->dtb_interval = now - buf->dtb_switched;
10779 buf->dtb_switched = now;
10780 dtrace_interrupt_enable(cookie);
10781}
10782
10783/*
10784 * Note: called from cross call context. This function activates a buffer
10785 * on a CPU. As with dtrace_buffer_switch(), the atomicity of the operation
10786 * is guaranteed by the disabling of interrupts.
10787 */
10788static void
10789dtrace_buffer_activate(dtrace_state_t *state)
10790{
10791 dtrace_buffer_t *buf;
10792 dtrace_icookie_t cookie = dtrace_interrupt_disable();
10793
10794 buf = &state->dts_buffer[curcpu];
10795
10796 if (buf->dtb_tomax != NULL) {
10797 /*
10798 * We might like to assert that the buffer is marked inactive,
10799 * but this isn't necessarily true: the buffer for the CPU
10800 * that processes the BEGIN probe has its buffer activated
10801 * manually. In this case, we take the (harmless) action
10802 * re-clearing the bit INACTIVE bit.
10803 */
10804 buf->dtb_flags &= ~DTRACEBUF_INACTIVE;
10805 }
10806
10807 dtrace_interrupt_enable(cookie);
10808}
10809
10810static int
10811dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags,
10812 processorid_t cpu)
10813{
10814#if defined(sun)
10815 cpu_t *cp;
10816#endif
10817 dtrace_buffer_t *buf;
10818
10819#if defined(sun)
10820 ASSERT(MUTEX_HELD(&cpu_lock));
10821 ASSERT(MUTEX_HELD(&dtrace_lock));
10822
10823 if (size > dtrace_nonroot_maxsize &&
10824 !PRIV_POLICY_CHOICE(CRED(), PRIV_ALL, B_FALSE))
10825 return (EFBIG);
10826
10827 cp = cpu_list;
10828
10829 do {
10830 if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id)
10831 continue;
10832
10833 buf = &bufs[cp->cpu_id];
10834
10835 /*
10836 * If there is already a buffer allocated for this CPU, it
10837 * is only possible that this is a DR event. In this case,
10838 */
10839 if (buf->dtb_tomax != NULL) {
10840 ASSERT(buf->dtb_size == size);
10841 continue;
10842 }
10843
10844 ASSERT(buf->dtb_xamot == NULL);
10845
10846 if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
10847 goto err;
10848
10849 buf->dtb_size = size;
10850 buf->dtb_flags = flags;
10851 buf->dtb_offset = 0;
10852 buf->dtb_drops = 0;
10853
10854 if (flags & DTRACEBUF_NOSWITCH)
10855 continue;
10856
10857 if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
10858 goto err;
10859 } while ((cp = cp->cpu_next) != cpu_list);
10860
10861 return (0);
10862
10863err:
10864 cp = cpu_list;
10865
10866 do {
10867 if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id)
10868 continue;
10869
10870 buf = &bufs[cp->cpu_id];
10871
10872 if (buf->dtb_xamot != NULL) {
10873 ASSERT(buf->dtb_tomax != NULL);
10874 ASSERT(buf->dtb_size == size);
10875 kmem_free(buf->dtb_xamot, size);
10876 }
10877
10878 if (buf->dtb_tomax != NULL) {
10879 ASSERT(buf->dtb_size == size);
10880 kmem_free(buf->dtb_tomax, size);
10881 }
10882
10883 buf->dtb_tomax = NULL;
10884 buf->dtb_xamot = NULL;
10885 buf->dtb_size = 0;
10886 } while ((cp = cp->cpu_next) != cpu_list);
10887
10888 return (ENOMEM);
10889#else
10890 int i;
10891
10892#if defined(__amd64__) || defined(__mips__) || defined(__powerpc__)
10893 /*
10894 * FreeBSD isn't good at limiting the amount of memory we
10895 * ask to malloc, so let's place a limit here before trying
10896 * to do something that might well end in tears at bedtime.
10897 */
10898 if (size > physmem * PAGE_SIZE / (128 * (mp_maxid + 1)))
10899 return(ENOMEM);
10900#endif
10901
10902 ASSERT(MUTEX_HELD(&dtrace_lock));
10903 CPU_FOREACH(i) {
10904 if (cpu != DTRACE_CPUALL && cpu != i)
10905 continue;
10906
10907 buf = &bufs[i];
10908
10909 /*
10910 * If there is already a buffer allocated for this CPU, it
10911 * is only possible that this is a DR event. In this case,
10912 * the buffer size must match our specified size.
10913 */
10914 if (buf->dtb_tomax != NULL) {
10915 ASSERT(buf->dtb_size == size);
10916 continue;
10917 }
10918
10919 ASSERT(buf->dtb_xamot == NULL);
10920
10921 if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
10922 goto err;
10923
10924 buf->dtb_size = size;
10925 buf->dtb_flags = flags;
10926 buf->dtb_offset = 0;
10927 buf->dtb_drops = 0;
10928
10929 if (flags & DTRACEBUF_NOSWITCH)
10930 continue;
10931
10932 if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
10933 goto err;
10934 }
10935
10936 return (0);
10937
10938err:
10939 /*
10940 * Error allocating memory, so free the buffers that were
10941 * allocated before the failed allocation.
10942 */
10943 CPU_FOREACH(i) {
10944 if (cpu != DTRACE_CPUALL && cpu != i)
10945 continue;
10946
10947 buf = &bufs[i];
10948
10949 if (buf->dtb_xamot != NULL) {
10950 ASSERT(buf->dtb_tomax != NULL);
10951 ASSERT(buf->dtb_size == size);
10952 kmem_free(buf->dtb_xamot, size);
10953 }
10954
10955 if (buf->dtb_tomax != NULL) {
10956 ASSERT(buf->dtb_size == size);
10957 kmem_free(buf->dtb_tomax, size);
10958 }
10959
10960 buf->dtb_tomax = NULL;
10961 buf->dtb_xamot = NULL;
10962 buf->dtb_size = 0;
10963
10964 }
10965
10966 return (ENOMEM);
10967#endif
10968}
10969
10970/*
10971 * Note: called from probe context. This function just increments the drop
10972 * count on a buffer. It has been made a function to allow for the
10973 * possibility of understanding the source of mysterious drop counts. (A
10974 * problem for which one may be particularly disappointed that DTrace cannot
10975 * be used to understand DTrace.)
10976 */
10977static void
10978dtrace_buffer_drop(dtrace_buffer_t *buf)
10979{
10980 buf->dtb_drops++;
10981}
10982
10983/*
10984 * Note: called from probe context. This function is called to reserve space
10985 * in a buffer. If mstate is non-NULL, sets the scratch base and size in the
10986 * mstate. Returns the new offset in the buffer, or a negative value if an
10987 * error has occurred.
10988 */
10989static intptr_t
10990dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, size_t align,
10991 dtrace_state_t *state, dtrace_mstate_t *mstate)
10992{
10993 intptr_t offs = buf->dtb_offset, soffs;
10994 intptr_t woffs;
10995 caddr_t tomax;
10996 size_t total;
10997
10998 if (buf->dtb_flags & DTRACEBUF_INACTIVE)
10999 return (-1);
11000
11001 if ((tomax = buf->dtb_tomax) == NULL) {
11002 dtrace_buffer_drop(buf);
11003 return (-1);
11004 }
11005
11006 if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) {
11007 while (offs & (align - 1)) {
11008 /*
11009 * Assert that our alignment is off by a number which
11010 * is itself sizeof (uint32_t) aligned.
11011 */
11012 ASSERT(!((align - (offs & (align - 1))) &
11013 (sizeof (uint32_t) - 1)));
11014 DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE);
11015 offs += sizeof (uint32_t);
11016 }
11017
11018 if ((soffs = offs + needed) > buf->dtb_size) {
11019 dtrace_buffer_drop(buf);
11020 return (-1);
11021 }
11022
11023 if (mstate == NULL)
11024 return (offs);
11025
11026 mstate->dtms_scratch_base = (uintptr_t)tomax + soffs;
11027 mstate->dtms_scratch_size = buf->dtb_size - soffs;
11028 mstate->dtms_scratch_ptr = mstate->dtms_scratch_base;
11029
11030 return (offs);
11031 }
11032
11033 if (buf->dtb_flags & DTRACEBUF_FILL) {
11034 if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN &&
11035 (buf->dtb_flags & DTRACEBUF_FULL))
11036 return (-1);
11037 goto out;
11038 }
11039
11040 total = needed + (offs & (align - 1));
11041
11042 /*
11043 * For a ring buffer, life is quite a bit more complicated. Before
11044 * we can store any padding, we need to adjust our wrapping offset.
11045 * (If we've never before wrapped or we're not about to, no adjustment
11046 * is required.)
11047 */
11048 if ((buf->dtb_flags & DTRACEBUF_WRAPPED) ||
11049 offs + total > buf->dtb_size) {
11050 woffs = buf->dtb_xamot_offset;
11051
11052 if (offs + total > buf->dtb_size) {
11053 /*
11054 * We can't fit in the end of the buffer. First, a
11055 * sanity check that we can fit in the buffer at all.
11056 */
11057 if (total > buf->dtb_size) {
11058 dtrace_buffer_drop(buf);
11059 return (-1);
11060 }
11061
11062 /*
11063 * We're going to be storing at the top of the buffer,
11064 * so now we need to deal with the wrapped offset. We
11065 * only reset our wrapped offset to 0 if it is
11066 * currently greater than the current offset. If it
11067 * is less than the current offset, it is because a
11068 * previous allocation induced a wrap -- but the
11069 * allocation didn't subsequently take the space due
11070 * to an error or false predicate evaluation. In this
11071 * case, we'll just leave the wrapped offset alone: if
11072 * the wrapped offset hasn't been advanced far enough
11073 * for this allocation, it will be adjusted in the
11074 * lower loop.
11075 */
11076 if (buf->dtb_flags & DTRACEBUF_WRAPPED) {
11077 if (woffs >= offs)
11078 woffs = 0;
11079 } else {
11080 woffs = 0;
11081 }
11082
11083 /*
11084 * Now we know that we're going to be storing to the
11085 * top of the buffer and that there is room for us
11086 * there. We need to clear the buffer from the current
11087 * offset to the end (there may be old gunk there).
11088 */
11089 while (offs < buf->dtb_size)
11090 tomax[offs++] = 0;
11091
11092 /*
11093 * We need to set our offset to zero. And because we
11094 * are wrapping, we need to set the bit indicating as
11095 * much. We can also adjust our needed space back
11096 * down to the space required by the ECB -- we know
11097 * that the top of the buffer is aligned.
11098 */
11099 offs = 0;
11100 total = needed;
11101 buf->dtb_flags |= DTRACEBUF_WRAPPED;
11102 } else {
11103 /*
11104 * There is room for us in the buffer, so we simply
11105 * need to check the wrapped offset.
11106 */
11107 if (woffs < offs) {
11108 /*
11109 * The wrapped offset is less than the offset.
11110 * This can happen if we allocated buffer space
11111 * that induced a wrap, but then we didn't
11112 * subsequently take the space due to an error
11113 * or false predicate evaluation. This is
11114 * okay; we know that _this_ allocation isn't
11115 * going to induce a wrap. We still can't
11116 * reset the wrapped offset to be zero,
11117 * however: the space may have been trashed in
11118 * the previous failed probe attempt. But at
11119 * least the wrapped offset doesn't need to
11120 * be adjusted at all...
11121 */
11122 goto out;
11123 }
11124 }
11125
11126 while (offs + total > woffs) {
11127 dtrace_epid_t epid = *(uint32_t *)(tomax + woffs);
11128 size_t size;
11129
11130 if (epid == DTRACE_EPIDNONE) {
11131 size = sizeof (uint32_t);
11132 } else {
11133 ASSERT3U(epid, <=, state->dts_necbs);
11134 ASSERT(state->dts_ecbs[epid - 1] != NULL);
11135
11136 size = state->dts_ecbs[epid - 1]->dte_size;
11137 }
11138
11139 ASSERT(woffs + size <= buf->dtb_size);
11140 ASSERT(size != 0);
11141
11142 if (woffs + size == buf->dtb_size) {
11143 /*
11144 * We've reached the end of the buffer; we want
11145 * to set the wrapped offset to 0 and break
11146 * out. However, if the offs is 0, then we're
11147 * in a strange edge-condition: the amount of
11148 * space that we want to reserve plus the size
11149 * of the record that we're overwriting is
11150 * greater than the size of the buffer. This
11151 * is problematic because if we reserve the
11152 * space but subsequently don't consume it (due
11153 * to a failed predicate or error) the wrapped
11154 * offset will be 0 -- yet the EPID at offset 0
11155 * will not be committed. This situation is
11156 * relatively easy to deal with: if we're in
11157 * this case, the buffer is indistinguishable
11158 * from one that hasn't wrapped; we need only
11159 * finish the job by clearing the wrapped bit,
11160 * explicitly setting the offset to be 0, and
11161 * zero'ing out the old data in the buffer.
11162 */
11163 if (offs == 0) {
11164 buf->dtb_flags &= ~DTRACEBUF_WRAPPED;
11165 buf->dtb_offset = 0;
11166 woffs = total;
11167
11168 while (woffs < buf->dtb_size)
11169 tomax[woffs++] = 0;
11170 }
11171
11172 woffs = 0;
11173 break;
11174 }
11175
11176 woffs += size;
11177 }
11178
11179 /*
11180 * We have a wrapped offset. It may be that the wrapped offset
11181 * has become zero -- that's okay.
11182 */
11183 buf->dtb_xamot_offset = woffs;
11184 }
11185
11186out:
11187 /*
11188 * Now we can plow the buffer with any necessary padding.
11189 */
11190 while (offs & (align - 1)) {
11191 /*
11192 * Assert that our alignment is off by a number which
11193 * is itself sizeof (uint32_t) aligned.
11194 */
11195 ASSERT(!((align - (offs & (align - 1))) &
11196 (sizeof (uint32_t) - 1)));
11197 DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE);
11198 offs += sizeof (uint32_t);
11199 }
11200
11201 if (buf->dtb_flags & DTRACEBUF_FILL) {
11202 if (offs + needed > buf->dtb_size - state->dts_reserve) {
11203 buf->dtb_flags |= DTRACEBUF_FULL;
11204 return (-1);
11205 }
11206 }
11207
11208 if (mstate == NULL)
11209 return (offs);
11210
11211 /*
11212 * For ring buffers and fill buffers, the scratch space is always
11213 * the inactive buffer.
11214 */
11215 mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot;
11216 mstate->dtms_scratch_size = buf->dtb_size;
11217 mstate->dtms_scratch_ptr = mstate->dtms_scratch_base;
11218
11219 return (offs);
11220}
11221
11222static void
11223dtrace_buffer_polish(dtrace_buffer_t *buf)
11224{
11225 ASSERT(buf->dtb_flags & DTRACEBUF_RING);
11226 ASSERT(MUTEX_HELD(&dtrace_lock));
11227
11228 if (!(buf->dtb_flags & DTRACEBUF_WRAPPED))
11229 return;
11230
11231 /*
11232 * We need to polish the ring buffer. There are three cases:
11233 *
11234 * - The first (and presumably most common) is that there is no gap
11235 * between the buffer offset and the wrapped offset. In this case,
11236 * there is nothing in the buffer that isn't valid data; we can
11237 * mark the buffer as polished and return.
11238 *
11239 * - The second (less common than the first but still more common
11240 * than the third) is that there is a gap between the buffer offset
11241 * and the wrapped offset, and the wrapped offset is larger than the
11242 * buffer offset. This can happen because of an alignment issue, or
11243 * can happen because of a call to dtrace_buffer_reserve() that
11244 * didn't subsequently consume the buffer space. In this case,
11245 * we need to zero the data from the buffer offset to the wrapped
11246 * offset.
11247 *
11248 * - The third (and least common) is that there is a gap between the
11249 * buffer offset and the wrapped offset, but the wrapped offset is
11250 * _less_ than the buffer offset. This can only happen because a
11251 * call to dtrace_buffer_reserve() induced a wrap, but the space
11252 * was not subsequently consumed. In this case, we need to zero the
11253 * space from the offset to the end of the buffer _and_ from the
11254 * top of the buffer to the wrapped offset.
11255 */
11256 if (buf->dtb_offset < buf->dtb_xamot_offset) {
11257 bzero(buf->dtb_tomax + buf->dtb_offset,
11258 buf->dtb_xamot_offset - buf->dtb_offset);
11259 }
11260
11261 if (buf->dtb_offset > buf->dtb_xamot_offset) {
11262 bzero(buf->dtb_tomax + buf->dtb_offset,
11263 buf->dtb_size - buf->dtb_offset);
11264 bzero(buf->dtb_tomax, buf->dtb_xamot_offset);
11265 }
11266}
11267
11268/*
11269 * This routine determines if data generated at the specified time has likely
11270 * been entirely consumed at user-level. This routine is called to determine
11271 * if an ECB on a defunct probe (but for an active enabling) can be safely
11272 * disabled and destroyed.
11273 */
11274static int
11275dtrace_buffer_consumed(dtrace_buffer_t *bufs, hrtime_t when)
11276{
11277 int i;
11278
11279 for (i = 0; i < NCPU; i++) {
11280 dtrace_buffer_t *buf = &bufs[i];
11281
11282 if (buf->dtb_size == 0)
11283 continue;
11284
11285 if (buf->dtb_flags & DTRACEBUF_RING)
11286 return (0);
11287
11288 if (!buf->dtb_switched && buf->dtb_offset != 0)
11289 return (0);
11290
11291 if (buf->dtb_switched - buf->dtb_interval < when)
11292 return (0);
11293 }
11294
11295 return (1);
11296}
11297
11298static void
11299dtrace_buffer_free(dtrace_buffer_t *bufs)
11300{
11301 int i;
11302
11303 for (i = 0; i < NCPU; i++) {
11304 dtrace_buffer_t *buf = &bufs[i];
11305
11306 if (buf->dtb_tomax == NULL) {
11307 ASSERT(buf->dtb_xamot == NULL);
11308 ASSERT(buf->dtb_size == 0);
11309 continue;
11310 }
11311
11312 if (buf->dtb_xamot != NULL) {
11313 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
11314 kmem_free(buf->dtb_xamot, buf->dtb_size);
11315 }
11316
11317 kmem_free(buf->dtb_tomax, buf->dtb_size);
11318 buf->dtb_size = 0;
11319 buf->dtb_tomax = NULL;
11320 buf->dtb_xamot = NULL;
11321 }
11322}
11323
11324/*
11325 * DTrace Enabling Functions
11326 */
11327static dtrace_enabling_t *
11328dtrace_enabling_create(dtrace_vstate_t *vstate)
11329{
11330 dtrace_enabling_t *enab;
11331
11332 enab = kmem_zalloc(sizeof (dtrace_enabling_t), KM_SLEEP);
11333 enab->dten_vstate = vstate;
11334
11335 return (enab);
11336}
11337
11338static void
11339dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb)
11340{
11341 dtrace_ecbdesc_t **ndesc;
11342 size_t osize, nsize;
11343
11344 /*
11345 * We can't add to enablings after we've enabled them, or after we've
11346 * retained them.
11347 */
11348 ASSERT(enab->dten_probegen == 0);
11349 ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL);
11350
11351 if (enab->dten_ndesc < enab->dten_maxdesc) {
11352 enab->dten_desc[enab->dten_ndesc++] = ecb;
11353 return;
11354 }
11355
11356 osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *);
11357
11358 if (enab->dten_maxdesc == 0) {
11359 enab->dten_maxdesc = 1;
11360 } else {
11361 enab->dten_maxdesc <<= 1;
11362 }
11363
11364 ASSERT(enab->dten_ndesc < enab->dten_maxdesc);
11365
11366 nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *);
11367 ndesc = kmem_zalloc(nsize, KM_SLEEP);
11368 bcopy(enab->dten_desc, ndesc, osize);
11369 if (enab->dten_desc != NULL)
11370 kmem_free(enab->dten_desc, osize);
11371
11372 enab->dten_desc = ndesc;
11373 enab->dten_desc[enab->dten_ndesc++] = ecb;
11374}
11375
11376static void
11377dtrace_enabling_addlike(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb,
11378 dtrace_probedesc_t *pd)
11379{
11380 dtrace_ecbdesc_t *new;
11381 dtrace_predicate_t *pred;
11382 dtrace_actdesc_t *act;
11383
11384 /*
11385 * We're going to create a new ECB description that matches the
11386 * specified ECB in every way, but has the specified probe description.
11387 */
11388 new = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP);
11389
11390 if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL)
11391 dtrace_predicate_hold(pred);
11392
11393 for (act = ecb->dted_action; act != NULL; act = act->dtad_next)
11394 dtrace_actdesc_hold(act);
11395
11396 new->dted_action = ecb->dted_action;
11397 new->dted_pred = ecb->dted_pred;
11398 new->dted_probe = *pd;
11399 new->dted_uarg = ecb->dted_uarg;
11400
11401 dtrace_enabling_add(enab, new);
11402}
11403
11404static void
11405dtrace_enabling_dump(dtrace_enabling_t *enab)
11406{
11407 int i;
11408
11409 for (i = 0; i < enab->dten_ndesc; i++) {
11410 dtrace_probedesc_t *desc = &enab->dten_desc[i]->dted_probe;
11411
11412 cmn_err(CE_NOTE, "enabling probe %d (%s:%s:%s:%s)", i,
11413 desc->dtpd_provider, desc->dtpd_mod,
11414 desc->dtpd_func, desc->dtpd_name);
11415 }
11416}
11417
11418static void
11419dtrace_enabling_destroy(dtrace_enabling_t *enab)
11420{
11421 int i;
11422 dtrace_ecbdesc_t *ep;
11423 dtrace_vstate_t *vstate = enab->dten_vstate;
11424
11425 ASSERT(MUTEX_HELD(&dtrace_lock));
11426
11427 for (i = 0; i < enab->dten_ndesc; i++) {
11428 dtrace_actdesc_t *act, *next;
11429 dtrace_predicate_t *pred;
11430
11431 ep = enab->dten_desc[i];
11432
11433 if ((pred = ep->dted_pred.dtpdd_predicate) != NULL)
11434 dtrace_predicate_release(pred, vstate);
11435
11436 for (act = ep->dted_action; act != NULL; act = next) {
11437 next = act->dtad_next;
11438 dtrace_actdesc_release(act, vstate);
11439 }
11440
11441 kmem_free(ep, sizeof (dtrace_ecbdesc_t));
11442 }
11443
11444 if (enab->dten_desc != NULL)
11445 kmem_free(enab->dten_desc,
11446 enab->dten_maxdesc * sizeof (dtrace_enabling_t *));
11447
11448 /*
11449 * If this was a retained enabling, decrement the dts_nretained count
11450 * and take it off of the dtrace_retained list.
11451 */
11452 if (enab->dten_prev != NULL || enab->dten_next != NULL ||
11453 dtrace_retained == enab) {
11454 ASSERT(enab->dten_vstate->dtvs_state != NULL);
11455 ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0);
11456 enab->dten_vstate->dtvs_state->dts_nretained--;
11457 }
11458
11459 if (enab->dten_prev == NULL) {
11460 if (dtrace_retained == enab) {
11461 dtrace_retained = enab->dten_next;
11462
11463 if (dtrace_retained != NULL)
11464 dtrace_retained->dten_prev = NULL;
11465 }
11466 } else {
11467 ASSERT(enab != dtrace_retained);
11468 ASSERT(dtrace_retained != NULL);
11469 enab->dten_prev->dten_next = enab->dten_next;
11470 }
11471
11472 if (enab->dten_next != NULL) {
11473 ASSERT(dtrace_retained != NULL);
11474 enab->dten_next->dten_prev = enab->dten_prev;
11475 }
11476
11477 kmem_free(enab, sizeof (dtrace_enabling_t));
11478}
11479
11480static int
11481dtrace_enabling_retain(dtrace_enabling_t *enab)
11482{
11483 dtrace_state_t *state;
11484
11485 ASSERT(MUTEX_HELD(&dtrace_lock));
11486 ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL);
11487 ASSERT(enab->dten_vstate != NULL);
11488
11489 state = enab->dten_vstate->dtvs_state;
11490 ASSERT(state != NULL);
11491
11492 /*
11493 * We only allow each state to retain dtrace_retain_max enablings.
11494 */
11495 if (state->dts_nretained >= dtrace_retain_max)
11496 return (ENOSPC);
11497
11498 state->dts_nretained++;
11499
11500 if (dtrace_retained == NULL) {
11501 dtrace_retained = enab;
11502 return (0);
11503 }
11504
11505 enab->dten_next = dtrace_retained;
11506 dtrace_retained->dten_prev = enab;
11507 dtrace_retained = enab;
11508
11509 return (0);
11510}
11511
11512static int
11513dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match,
11514 dtrace_probedesc_t *create)
11515{
11516 dtrace_enabling_t *new, *enab;
11517 int found = 0, err = ENOENT;
11518
11519 ASSERT(MUTEX_HELD(&dtrace_lock));
11520 ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN);
11521 ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN);
11522 ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN);
11523 ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN);
11524
11525 new = dtrace_enabling_create(&state->dts_vstate);
11526
11527 /*
11528 * Iterate over all retained enablings, looking for enablings that
11529 * match the specified state.
11530 */
11531 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11532 int i;
11533
11534 /*
11535 * dtvs_state can only be NULL for helper enablings -- and
11536 * helper enablings can't be retained.
11537 */
11538 ASSERT(enab->dten_vstate->dtvs_state != NULL);
11539
11540 if (enab->dten_vstate->dtvs_state != state)
11541 continue;
11542
11543 /*
11544 * Now iterate over each probe description; we're looking for
11545 * an exact match to the specified probe description.
11546 */
11547 for (i = 0; i < enab->dten_ndesc; i++) {
11548 dtrace_ecbdesc_t *ep = enab->dten_desc[i];
11549 dtrace_probedesc_t *pd = &ep->dted_probe;
11550
11551 if (strcmp(pd->dtpd_provider, match->dtpd_provider))
11552 continue;
11553
11554 if (strcmp(pd->dtpd_mod, match->dtpd_mod))
11555 continue;
11556
11557 if (strcmp(pd->dtpd_func, match->dtpd_func))
11558 continue;
11559
11560 if (strcmp(pd->dtpd_name, match->dtpd_name))
11561 continue;
11562
11563 /*
11564 * We have a winning probe! Add it to our growing
11565 * enabling.
11566 */
11567 found = 1;
11568 dtrace_enabling_addlike(new, ep, create);
11569 }
11570 }
11571
11572 if (!found || (err = dtrace_enabling_retain(new)) != 0) {
11573 dtrace_enabling_destroy(new);
11574 return (err);
11575 }
11576
11577 return (0);
11578}
11579
11580static void
11581dtrace_enabling_retract(dtrace_state_t *state)
11582{
11583 dtrace_enabling_t *enab, *next;
11584
11585 ASSERT(MUTEX_HELD(&dtrace_lock));
11586
11587 /*
11588 * Iterate over all retained enablings, destroy the enablings retained
11589 * for the specified state.
11590 */
11591 for (enab = dtrace_retained; enab != NULL; enab = next) {
11592 next = enab->dten_next;
11593
11594 /*
11595 * dtvs_state can only be NULL for helper enablings -- and
11596 * helper enablings can't be retained.
11597 */
11598 ASSERT(enab->dten_vstate->dtvs_state != NULL);
11599
11600 if (enab->dten_vstate->dtvs_state == state) {
11601 ASSERT(state->dts_nretained > 0);
11602 dtrace_enabling_destroy(enab);
11603 }
11604 }
11605
11606 ASSERT(state->dts_nretained == 0);
11607}
11608
11609static int
11610dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched)
11611{
11612 int i = 0;
11613 int matched = 0;
11614
11615 ASSERT(MUTEX_HELD(&cpu_lock));
11616 ASSERT(MUTEX_HELD(&dtrace_lock));
11617
11618 for (i = 0; i < enab->dten_ndesc; i++) {
11619 dtrace_ecbdesc_t *ep = enab->dten_desc[i];
11620
11621 enab->dten_current = ep;
11622 enab->dten_error = 0;
11623
11624 matched += dtrace_probe_enable(&ep->dted_probe, enab);
11625
11626 if (enab->dten_error != 0) {
11627 /*
11628 * If we get an error half-way through enabling the
11629 * probes, we kick out -- perhaps with some number of
11630 * them enabled. Leaving enabled probes enabled may
11631 * be slightly confusing for user-level, but we expect
11632 * that no one will attempt to actually drive on in
11633 * the face of such errors. If this is an anonymous
11634 * enabling (indicated with a NULL nmatched pointer),
11635 * we cmn_err() a message. We aren't expecting to
11636 * get such an error -- such as it can exist at all,
11637 * it would be a result of corrupted DOF in the driver
11638 * properties.
11639 */
11640 if (nmatched == NULL) {
11641 cmn_err(CE_WARN, "dtrace_enabling_match() "
11642 "error on %p: %d", (void *)ep,
11643 enab->dten_error);
11644 }
11645
11646 return (enab->dten_error);
11647 }
11648 }
11649
11650 enab->dten_probegen = dtrace_probegen;
11651 if (nmatched != NULL)
11652 *nmatched = matched;
11653
11654 return (0);
11655}
11656
11657static void
11658dtrace_enabling_matchall(void)
11659{
11660 dtrace_enabling_t *enab;
11661
11662 mutex_enter(&cpu_lock);
11663 mutex_enter(&dtrace_lock);
11664
11665 /*
11666 * Iterate over all retained enablings to see if any probes match
11667 * against them. We only perform this operation on enablings for which
11668 * we have sufficient permissions by virtue of being in the global zone
11669 * or in the same zone as the DTrace client. Because we can be called
11670 * after dtrace_detach() has been called, we cannot assert that there
11671 * are retained enablings. We can safely load from dtrace_retained,
11672 * however: the taskq_destroy() at the end of dtrace_detach() will
11673 * block pending our completion.
11674 */
11675 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11676#if defined(sun)
11677 cred_t *cr = enab->dten_vstate->dtvs_state->dts_cred.dcr_cred;
11678
11679 if (INGLOBALZONE(curproc) || getzoneid() == crgetzoneid(cr))
11680#endif
11681 (void) dtrace_enabling_match(enab, NULL);
11682 }
11683
11684 mutex_exit(&dtrace_lock);
11685 mutex_exit(&cpu_lock);
11686}
11687
11688/*
11689 * If an enabling is to be enabled without having matched probes (that is, if
11690 * dtrace_state_go() is to be called on the underlying dtrace_state_t), the
11691 * enabling must be _primed_ by creating an ECB for every ECB description.
11692 * This must be done to assure that we know the number of speculations, the
11693 * number of aggregations, the minimum buffer size needed, etc. before we
11694 * transition out of DTRACE_ACTIVITY_INACTIVE. To do this without actually
11695 * enabling any probes, we create ECBs for every ECB decription, but with a
11696 * NULL probe -- which is exactly what this function does.
11697 */
11698static void
11699dtrace_enabling_prime(dtrace_state_t *state)
11700{
11701 dtrace_enabling_t *enab;
11702 int i;
11703
11704 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11705 ASSERT(enab->dten_vstate->dtvs_state != NULL);
11706
11707 if (enab->dten_vstate->dtvs_state != state)
11708 continue;
11709
11710 /*
11711 * We don't want to prime an enabling more than once, lest
11712 * we allow a malicious user to induce resource exhaustion.
11713 * (The ECBs that result from priming an enabling aren't
11714 * leaked -- but they also aren't deallocated until the
11715 * consumer state is destroyed.)
11716 */
11717 if (enab->dten_primed)
11718 continue;
11719
11720 for (i = 0; i < enab->dten_ndesc; i++) {
11721 enab->dten_current = enab->dten_desc[i];
11722 (void) dtrace_probe_enable(NULL, enab);
11723 }
11724
11725 enab->dten_primed = 1;
11726 }
11727}
11728
11729/*
11730 * Called to indicate that probes should be provided due to retained
11731 * enablings. This is implemented in terms of dtrace_probe_provide(), but it
11732 * must take an initial lap through the enabling calling the dtps_provide()
11733 * entry point explicitly to allow for autocreated probes.
11734 */
11735static void
11736dtrace_enabling_provide(dtrace_provider_t *prv)
11737{
11738 int i, all = 0;
11739 dtrace_probedesc_t desc;
11740
11741 ASSERT(MUTEX_HELD(&dtrace_lock));
11742 ASSERT(MUTEX_HELD(&dtrace_provider_lock));
11743
11744 if (prv == NULL) {
11745 all = 1;
11746 prv = dtrace_provider;
11747 }
11748
11749 do {
11750 dtrace_enabling_t *enab = dtrace_retained;
11751 void *parg = prv->dtpv_arg;
11752
11753 for (; enab != NULL; enab = enab->dten_next) {
11754 for (i = 0; i < enab->dten_ndesc; i++) {
11755 desc = enab->dten_desc[i]->dted_probe;
11756 mutex_exit(&dtrace_lock);
11757 prv->dtpv_pops.dtps_provide(parg, &desc);
11758 mutex_enter(&dtrace_lock);
11759 }
11760 }
11761 } while (all && (prv = prv->dtpv_next) != NULL);
11762
11763 mutex_exit(&dtrace_lock);
11764 dtrace_probe_provide(NULL, all ? NULL : prv);
11765 mutex_enter(&dtrace_lock);
11766}
11767
11768/*
11769 * Called to reap ECBs that are attached to probes from defunct providers.
11770 */
11771static void
11772dtrace_enabling_reap(void)
11773{
11774 dtrace_provider_t *prov;
11775 dtrace_probe_t *probe;
11776 dtrace_ecb_t *ecb;
11777 hrtime_t when;
11778 int i;
11779
11780 mutex_enter(&cpu_lock);
11781 mutex_enter(&dtrace_lock);
11782
11783 for (i = 0; i < dtrace_nprobes; i++) {
11784 if ((probe = dtrace_probes[i]) == NULL)
11785 continue;
11786
11787 if (probe->dtpr_ecb == NULL)
11788 continue;
11789
11790 prov = probe->dtpr_provider;
11791
11792 if ((when = prov->dtpv_defunct) == 0)
11793 continue;
11794
11795 /*
11796 * We have ECBs on a defunct provider: we want to reap these
11797 * ECBs to allow the provider to unregister. The destruction
11798 * of these ECBs must be done carefully: if we destroy the ECB
11799 * and the consumer later wishes to consume an EPID that
11800 * corresponds to the destroyed ECB (and if the EPID metadata
11801 * has not been previously consumed), the consumer will abort
11802 * processing on the unknown EPID. To reduce (but not, sadly,
11803 * eliminate) the possibility of this, we will only destroy an
11804 * ECB for a defunct provider if, for the state that
11805 * corresponds to the ECB:
11806 *
11807 * (a) There is no speculative tracing (which can effectively
11808 * cache an EPID for an arbitrary amount of time).
11809 *
11810 * (b) The principal buffers have been switched twice since the
11811 * provider became defunct.
11812 *
11813 * (c) The aggregation buffers are of zero size or have been
11814 * switched twice since the provider became defunct.
11815 *
11816 * We use dts_speculates to determine (a) and call a function
11817 * (dtrace_buffer_consumed()) to determine (b) and (c). Note
11818 * that as soon as we've been unable to destroy one of the ECBs
11819 * associated with the probe, we quit trying -- reaping is only
11820 * fruitful in as much as we can destroy all ECBs associated
11821 * with the defunct provider's probes.
11822 */
11823 while ((ecb = probe->dtpr_ecb) != NULL) {
11824 dtrace_state_t *state = ecb->dte_state;
11825 dtrace_buffer_t *buf = state->dts_buffer;
11826 dtrace_buffer_t *aggbuf = state->dts_aggbuffer;
11827
11828 if (state->dts_speculates)
11829 break;
11830
11831 if (!dtrace_buffer_consumed(buf, when))
11832 break;
11833
11834 if (!dtrace_buffer_consumed(aggbuf, when))
11835 break;
11836
11837 dtrace_ecb_disable(ecb);
11838 ASSERT(probe->dtpr_ecb != ecb);
11839 dtrace_ecb_destroy(ecb);
11840 }
11841 }
11842
11843 mutex_exit(&dtrace_lock);
11844 mutex_exit(&cpu_lock);
11845}
11846
11847/*
11848 * DTrace DOF Functions
11849 */
11850/*ARGSUSED*/
11851static void
11852dtrace_dof_error(dof_hdr_t *dof, const char *str)
11853{
11854 if (dtrace_err_verbose)
11855 cmn_err(CE_WARN, "failed to process DOF: %s", str);
11856
11857#ifdef DTRACE_ERRDEBUG
11858 dtrace_errdebug(str);
11859#endif
11860}
11861
11862/*
11863 * Create DOF out of a currently enabled state. Right now, we only create
11864 * DOF containing the run-time options -- but this could be expanded to create
11865 * complete DOF representing the enabled state.
11866 */
11867static dof_hdr_t *
11868dtrace_dof_create(dtrace_state_t *state)
11869{
11870 dof_hdr_t *dof;
11871 dof_sec_t *sec;
11872 dof_optdesc_t *opt;
11873 int i, len = sizeof (dof_hdr_t) +
11874 roundup(sizeof (dof_sec_t), sizeof (uint64_t)) +
11875 sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
11876
11877 ASSERT(MUTEX_HELD(&dtrace_lock));
11878
11879 dof = kmem_zalloc(len, KM_SLEEP);
11880 dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0;
11881 dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1;
11882 dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2;
11883 dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3;
11884
11885 dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE;
11886 dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE;
11887 dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION;
11888 dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION;
11889 dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS;
11890 dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS;
11891
11892 dof->dofh_flags = 0;
11893 dof->dofh_hdrsize = sizeof (dof_hdr_t);
11894 dof->dofh_secsize = sizeof (dof_sec_t);
11895 dof->dofh_secnum = 1; /* only DOF_SECT_OPTDESC */
11896 dof->dofh_secoff = sizeof (dof_hdr_t);
11897 dof->dofh_loadsz = len;
11898 dof->dofh_filesz = len;
11899 dof->dofh_pad = 0;
11900
11901 /*
11902 * Fill in the option section header...
11903 */
11904 sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t));
11905 sec->dofs_type = DOF_SECT_OPTDESC;
11906 sec->dofs_align = sizeof (uint64_t);
11907 sec->dofs_flags = DOF_SECF_LOAD;
11908 sec->dofs_entsize = sizeof (dof_optdesc_t);
11909
11910 opt = (dof_optdesc_t *)((uintptr_t)sec +
11911 roundup(sizeof (dof_sec_t), sizeof (uint64_t)));
11912
11913 sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof;
11914 sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
11915
11916 for (i = 0; i < DTRACEOPT_MAX; i++) {
11917 opt[i].dofo_option = i;
11918 opt[i].dofo_strtab = DOF_SECIDX_NONE;
11919 opt[i].dofo_value = state->dts_options[i];
11920 }
11921
11922 return (dof);
11923}
11924
11925static dof_hdr_t *
11926dtrace_dof_copyin(uintptr_t uarg, int *errp)
11927{
11928 dof_hdr_t hdr, *dof;
11929
11930 ASSERT(!MUTEX_HELD(&dtrace_lock));
11931
11932 /*
11933 * First, we're going to copyin() the sizeof (dof_hdr_t).
11934 */
11935 if (copyin((void *)uarg, &hdr, sizeof (hdr)) != 0) {
11936 dtrace_dof_error(NULL, "failed to copyin DOF header");
11937 *errp = EFAULT;
11938 return (NULL);
11939 }
11940
11941 /*
11942 * Now we'll allocate the entire DOF and copy it in -- provided
11943 * that the length isn't outrageous.
11944 */
11945 if (hdr.dofh_loadsz >= dtrace_dof_maxsize) {
11946 dtrace_dof_error(&hdr, "load size exceeds maximum");
11947 *errp = E2BIG;
11948 return (NULL);
11949 }
11950
11951 if (hdr.dofh_loadsz < sizeof (hdr)) {
11952 dtrace_dof_error(&hdr, "invalid load size");
11953 *errp = EINVAL;
11954 return (NULL);
11955 }
11956
11957 dof = kmem_alloc(hdr.dofh_loadsz, KM_SLEEP);
11958
11959 if (copyin((void *)uarg, dof, hdr.dofh_loadsz) != 0) {
11960 kmem_free(dof, hdr.dofh_loadsz);
11961 *errp = EFAULT;
11962 return (NULL);
11963 }
11964
11965 return (dof);
11966}
11967
11968#if !defined(sun)
11969static __inline uchar_t
11970dtrace_dof_char(char c) {
11971 switch (c) {
11972 case '0':
11973 case '1':
11974 case '2':
11975 case '3':
11976 case '4':
11977 case '5':
11978 case '6':
11979 case '7':
11980 case '8':
11981 case '9':
11982 return (c - '0');
11983 case 'A':
11984 case 'B':
11985 case 'C':
11986 case 'D':
11987 case 'E':
11988 case 'F':
11989 return (c - 'A' + 10);
11990 case 'a':
11991 case 'b':
11992 case 'c':
11993 case 'd':
11994 case 'e':
11995 case 'f':
11996 return (c - 'a' + 10);
11997 }
11998 /* Should not reach here. */
11999 return (0);
12000}
12001#endif
12002
12003static dof_hdr_t *
12004dtrace_dof_property(const char *name)
12005{
12006 uchar_t *buf;
12007 uint64_t loadsz;
12008 unsigned int len, i;
12009 dof_hdr_t *dof;
12010
12011#if defined(sun)
12012 /*
12013 * Unfortunately, array of values in .conf files are always (and
12014 * only) interpreted to be integer arrays. We must read our DOF
12015 * as an integer array, and then squeeze it into a byte array.
12016 */
12017 if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0,
12018 (char *)name, (int **)&buf, &len) != DDI_PROP_SUCCESS)
12019 return (NULL);
12020
12021 for (i = 0; i < len; i++)
12022 buf[i] = (uchar_t)(((int *)buf)[i]);
12023
12024 if (len < sizeof (dof_hdr_t)) {
12025 ddi_prop_free(buf);
12026 dtrace_dof_error(NULL, "truncated header");
12027 return (NULL);
12028 }
12029
12030 if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) {
12031 ddi_prop_free(buf);
12032 dtrace_dof_error(NULL, "truncated DOF");
12033 return (NULL);
12034 }
12035
12036 if (loadsz >= dtrace_dof_maxsize) {
12037 ddi_prop_free(buf);
12038 dtrace_dof_error(NULL, "oversized DOF");
12039 return (NULL);
12040 }
12041
12042 dof = kmem_alloc(loadsz, KM_SLEEP);
12043 bcopy(buf, dof, loadsz);
12044 ddi_prop_free(buf);
12045#else
12046 char *p;
12047 char *p_env;
12048
12049 if ((p_env = getenv(name)) == NULL)
12050 return (NULL);
12051
12052 len = strlen(p_env) / 2;
12053
12054 buf = kmem_alloc(len, KM_SLEEP);
12055
12056 dof = (dof_hdr_t *) buf;
12057
12058 p = p_env;
12059
12060 for (i = 0; i < len; i++) {
12061 buf[i] = (dtrace_dof_char(p[0]) << 4) |
12062 dtrace_dof_char(p[1]);
12063 p += 2;
12064 }
12065
12066 freeenv(p_env);
12067
12068 if (len < sizeof (dof_hdr_t)) {
12069 kmem_free(buf, 0);
12070 dtrace_dof_error(NULL, "truncated header");
12071 return (NULL);
12072 }
12073
12074 if (len < (loadsz = dof->dofh_loadsz)) {
12075 kmem_free(buf, 0);
12076 dtrace_dof_error(NULL, "truncated DOF");
12077 return (NULL);
12078 }
12079
12080 if (loadsz >= dtrace_dof_maxsize) {
12081 kmem_free(buf, 0);
12082 dtrace_dof_error(NULL, "oversized DOF");
12083 return (NULL);
12084 }
12085#endif
12086
12087 return (dof);
12088}
12089
12090static void
12091dtrace_dof_destroy(dof_hdr_t *dof)
12092{
12093 kmem_free(dof, dof->dofh_loadsz);
12094}
12095
12096/*
12097 * Return the dof_sec_t pointer corresponding to a given section index. If the
12098 * index is not valid, dtrace_dof_error() is called and NULL is returned. If
12099 * a type other than DOF_SECT_NONE is specified, the header is checked against
12100 * this type and NULL is returned if the types do not match.
12101 */
12102static dof_sec_t *
12103dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, dof_secidx_t i)
12104{
12105 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)
12106 ((uintptr_t)dof + dof->dofh_secoff + i * dof->dofh_secsize);
12107
12108 if (i >= dof->dofh_secnum) {
12109 dtrace_dof_error(dof, "referenced section index is invalid");
12110 return (NULL);
12111 }
12112
12113 if (!(sec->dofs_flags & DOF_SECF_LOAD)) {
12114 dtrace_dof_error(dof, "referenced section is not loadable");
12115 return (NULL);
12116 }
12117
12118 if (type != DOF_SECT_NONE && type != sec->dofs_type) {
12119 dtrace_dof_error(dof, "referenced section is the wrong type");
12120 return (NULL);
12121 }
12122
12123 return (sec);
12124}
12125
12126static dtrace_probedesc_t *
12127dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_probedesc_t *desc)
12128{
12129 dof_probedesc_t *probe;
12130 dof_sec_t *strtab;
12131 uintptr_t daddr = (uintptr_t)dof;
12132 uintptr_t str;
12133 size_t size;
12134
12135 if (sec->dofs_type != DOF_SECT_PROBEDESC) {
12136 dtrace_dof_error(dof, "invalid probe section");
12137 return (NULL);
12138 }
12139
12140 if (sec->dofs_align != sizeof (dof_secidx_t)) {
12141 dtrace_dof_error(dof, "bad alignment in probe description");
12142 return (NULL);
12143 }
12144
12145 if (sec->dofs_offset + sizeof (dof_probedesc_t) > dof->dofh_loadsz) {
12146 dtrace_dof_error(dof, "truncated probe description");
12147 return (NULL);
12148 }
12149
12150 probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset);
12151 strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab);
12152
12153 if (strtab == NULL)
12154 return (NULL);
12155
12156 str = daddr + strtab->dofs_offset;
12157 size = strtab->dofs_size;
12158
12159 if (probe->dofp_provider >= strtab->dofs_size) {
12160 dtrace_dof_error(dof, "corrupt probe provider");
12161 return (NULL);
12162 }
12163
12164 (void) strncpy(desc->dtpd_provider,
12165 (char *)(str + probe->dofp_provider),
12166 MIN(DTRACE_PROVNAMELEN - 1, size - probe->dofp_provider));
12167
12168 if (probe->dofp_mod >= strtab->dofs_size) {
12169 dtrace_dof_error(dof, "corrupt probe module");
12170 return (NULL);
12171 }
12172
12173 (void) strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod),
12174 MIN(DTRACE_MODNAMELEN - 1, size - probe->dofp_mod));
12175
12176 if (probe->dofp_func >= strtab->dofs_size) {
12177 dtrace_dof_error(dof, "corrupt probe function");
12178 return (NULL);
12179 }
12180
12181 (void) strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func),
12182 MIN(DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func));
12183
12184 if (probe->dofp_name >= strtab->dofs_size) {
12185 dtrace_dof_error(dof, "corrupt probe name");
12186 return (NULL);
12187 }
12188
12189 (void) strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name),
12190 MIN(DTRACE_NAMELEN - 1, size - probe->dofp_name));
12191
12192 return (desc);
12193}
12194
12195static dtrace_difo_t *
12196dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12197 cred_t *cr)
12198{
12199 dtrace_difo_t *dp;
12200 size_t ttl = 0;
12201 dof_difohdr_t *dofd;
12202 uintptr_t daddr = (uintptr_t)dof;
12203 size_t max = dtrace_difo_maxsize;
12204 int i, l, n;
12205
12206 static const struct {
12207 int section;
12208 int bufoffs;
12209 int lenoffs;
12210 int entsize;
12211 int align;
12212 const char *msg;
12213 } difo[] = {
12214 { DOF_SECT_DIF, offsetof(dtrace_difo_t, dtdo_buf),
12215 offsetof(dtrace_difo_t, dtdo_len), sizeof (dif_instr_t),
12216 sizeof (dif_instr_t), "multiple DIF sections" },
12217
12218 { DOF_SECT_INTTAB, offsetof(dtrace_difo_t, dtdo_inttab),
12219 offsetof(dtrace_difo_t, dtdo_intlen), sizeof (uint64_t),
12220 sizeof (uint64_t), "multiple integer tables" },
12221
12222 { DOF_SECT_STRTAB, offsetof(dtrace_difo_t, dtdo_strtab),
12223 offsetof(dtrace_difo_t, dtdo_strlen), 0,
12224 sizeof (char), "multiple string tables" },
12225
12226 { DOF_SECT_VARTAB, offsetof(dtrace_difo_t, dtdo_vartab),
12227 offsetof(dtrace_difo_t, dtdo_varlen), sizeof (dtrace_difv_t),
12228 sizeof (uint_t), "multiple variable tables" },
12229
12230 { DOF_SECT_NONE, 0, 0, 0, 0, NULL }
12231 };
12232
12233 if (sec->dofs_type != DOF_SECT_DIFOHDR) {
12234 dtrace_dof_error(dof, "invalid DIFO header section");
12235 return (NULL);
12236 }
12237
12238 if (sec->dofs_align != sizeof (dof_secidx_t)) {
12239 dtrace_dof_error(dof, "bad alignment in DIFO header");
12240 return (NULL);
12241 }
12242
12243 if (sec->dofs_size < sizeof (dof_difohdr_t) ||
12244 sec->dofs_size % sizeof (dof_secidx_t)) {
12245 dtrace_dof_error(dof, "bad size in DIFO header");
12246 return (NULL);
12247 }
12248
12249 dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset);
12250 n = (sec->dofs_size - sizeof (*dofd)) / sizeof (dof_secidx_t) + 1;
12251
12252 dp = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP);
12253 dp->dtdo_rtype = dofd->dofd_rtype;
12254
12255 for (l = 0; l < n; l++) {
12256 dof_sec_t *subsec;
12257 void **bufp;
12258 uint32_t *lenp;
12259
12260 if ((subsec = dtrace_dof_sect(dof, DOF_SECT_NONE,
12261 dofd->dofd_links[l])) == NULL)
12262 goto err; /* invalid section link */
12263
12264 if (ttl + subsec->dofs_size > max) {
12265 dtrace_dof_error(dof, "exceeds maximum size");
12266 goto err;
12267 }
12268
12269 ttl += subsec->dofs_size;
12270
12271 for (i = 0; difo[i].section != DOF_SECT_NONE; i++) {
12272 if (subsec->dofs_type != difo[i].section)
12273 continue;
12274
12275 if (!(subsec->dofs_flags & DOF_SECF_LOAD)) {
12276 dtrace_dof_error(dof, "section not loaded");
12277 goto err;
12278 }
12279
12280 if (subsec->dofs_align != difo[i].align) {
12281 dtrace_dof_error(dof, "bad alignment");
12282 goto err;
12283 }
12284
12285 bufp = (void **)((uintptr_t)dp + difo[i].bufoffs);
12286 lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs);
12287
12288 if (*bufp != NULL) {
12289 dtrace_dof_error(dof, difo[i].msg);
12290 goto err;
12291 }
12292
12293 if (difo[i].entsize != subsec->dofs_entsize) {
12294 dtrace_dof_error(dof, "entry size mismatch");
12295 goto err;
12296 }
12297
12298 if (subsec->dofs_entsize != 0 &&
12299 (subsec->dofs_size % subsec->dofs_entsize) != 0) {
12300 dtrace_dof_error(dof, "corrupt entry size");
12301 goto err;
12302 }
12303
12304 *lenp = subsec->dofs_size;
12305 *bufp = kmem_alloc(subsec->dofs_size, KM_SLEEP);
12306 bcopy((char *)(uintptr_t)(daddr + subsec->dofs_offset),
12307 *bufp, subsec->dofs_size);
12308
12309 if (subsec->dofs_entsize != 0)
12310 *lenp /= subsec->dofs_entsize;
12311
12312 break;
12313 }
12314
12315 /*
12316 * If we encounter a loadable DIFO sub-section that is not
12317 * known to us, assume this is a broken program and fail.
12318 */
12319 if (difo[i].section == DOF_SECT_NONE &&
12320 (subsec->dofs_flags & DOF_SECF_LOAD)) {
12321 dtrace_dof_error(dof, "unrecognized DIFO subsection");
12322 goto err;
12323 }
12324 }
12325
12326 if (dp->dtdo_buf == NULL) {
12327 /*
12328 * We can't have a DIF object without DIF text.
12329 */
12330 dtrace_dof_error(dof, "missing DIF text");
12331 goto err;
12332 }
12333
12334 /*
12335 * Before we validate the DIF object, run through the variable table
12336 * looking for the strings -- if any of their size are under, we'll set
12337 * their size to be the system-wide default string size. Note that
12338 * this should _not_ happen if the "strsize" option has been set --
12339 * in this case, the compiler should have set the size to reflect the
12340 * setting of the option.
12341 */
12342 for (i = 0; i < dp->dtdo_varlen; i++) {
12343 dtrace_difv_t *v = &dp->dtdo_vartab[i];
12344 dtrace_diftype_t *t = &v->dtdv_type;
12345
12346 if (v->dtdv_id < DIF_VAR_OTHER_UBASE)
12347 continue;
12348
12349 if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0)
12350 t->dtdt_size = dtrace_strsize_default;
12351 }
12352
12353 if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0)
12354 goto err;
12355
12356 dtrace_difo_init(dp, vstate);
12357 return (dp);
12358
12359err:
12360 kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t));
12361 kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t));
12362 kmem_free(dp->dtdo_strtab, dp->dtdo_strlen);
12363 kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t));
12364
12365 kmem_free(dp, sizeof (dtrace_difo_t));
12366 return (NULL);
12367}
12368
12369static dtrace_predicate_t *
12370dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12371 cred_t *cr)
12372{
12373 dtrace_difo_t *dp;
12374
12375 if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL)
12376 return (NULL);
12377
12378 return (dtrace_predicate_create(dp));
12379}
12380
12381static dtrace_actdesc_t *
12382dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12383 cred_t *cr)
12384{
12385 dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next;
12386 dof_actdesc_t *desc;
12387 dof_sec_t *difosec;
12388 size_t offs;
12389 uintptr_t daddr = (uintptr_t)dof;
12390 uint64_t arg;
12391 dtrace_actkind_t kind;
12392
12393 if (sec->dofs_type != DOF_SECT_ACTDESC) {
12394 dtrace_dof_error(dof, "invalid action section");
12395 return (NULL);
12396 }
12397
12398 if (sec->dofs_offset + sizeof (dof_actdesc_t) > dof->dofh_loadsz) {
12399 dtrace_dof_error(dof, "truncated action description");
12400 return (NULL);
12401 }
12402
12403 if (sec->dofs_align != sizeof (uint64_t)) {
12404 dtrace_dof_error(dof, "bad alignment in action description");
12405 return (NULL);
12406 }
12407
12408 if (sec->dofs_size < sec->dofs_entsize) {
12409 dtrace_dof_error(dof, "section entry size exceeds total size");
12410 return (NULL);
12411 }
12412
12413 if (sec->dofs_entsize != sizeof (dof_actdesc_t)) {
12414 dtrace_dof_error(dof, "bad entry size in action description");
12415 return (NULL);
12416 }
12417
12418 if (sec->dofs_size / sec->dofs_entsize > dtrace_actions_max) {
12419 dtrace_dof_error(dof, "actions exceed dtrace_actions_max");
12420 return (NULL);
12421 }
12422
12423 for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) {
12424 desc = (dof_actdesc_t *)(daddr +
12425 (uintptr_t)sec->dofs_offset + offs);
12426 kind = (dtrace_actkind_t)desc->dofa_kind;
12427
12428 if ((DTRACEACT_ISPRINTFLIKE(kind) &&
12429 (kind != DTRACEACT_PRINTA ||
12430 desc->dofa_strtab != DOF_SECIDX_NONE)) ||
12431 (kind == DTRACEACT_DIFEXPR &&
12432 desc->dofa_strtab != DOF_SECIDX_NONE)) {
12433 dof_sec_t *strtab;
12434 char *str, *fmt;
12435 uint64_t i;
12436
12437 /*
12438 * The argument to these actions is an index into the
12439 * DOF string table. For printf()-like actions, this
12440 * is the format string. For print(), this is the
12441 * CTF type of the expression result.
12442 */
12443 if ((strtab = dtrace_dof_sect(dof,
12444 DOF_SECT_STRTAB, desc->dofa_strtab)) == NULL)
12445 goto err;
12446
12447 str = (char *)((uintptr_t)dof +
12448 (uintptr_t)strtab->dofs_offset);
12449
12450 for (i = desc->dofa_arg; i < strtab->dofs_size; i++) {
12451 if (str[i] == '\0')
12452 break;
12453 }
12454
12455 if (i >= strtab->dofs_size) {
12456 dtrace_dof_error(dof, "bogus format string");
12457 goto err;
12458 }
12459
12460 if (i == desc->dofa_arg) {
12461 dtrace_dof_error(dof, "empty format string");
12462 goto err;
12463 }
12464
12465 i -= desc->dofa_arg;
12466 fmt = kmem_alloc(i + 1, KM_SLEEP);
12467 bcopy(&str[desc->dofa_arg], fmt, i + 1);
12468 arg = (uint64_t)(uintptr_t)fmt;
12469 } else {
12470 if (kind == DTRACEACT_PRINTA) {
12471 ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE);
12472 arg = 0;
12473 } else {
12474 arg = desc->dofa_arg;
12475 }
12476 }
12477
12478 act = dtrace_actdesc_create(kind, desc->dofa_ntuple,
12479 desc->dofa_uarg, arg);
12480
12481 if (last != NULL) {
12482 last->dtad_next = act;
12483 } else {
12484 first = act;
12485 }
12486
12487 last = act;
12488
12489 if (desc->dofa_difo == DOF_SECIDX_NONE)
12490 continue;
12491
12492 if ((difosec = dtrace_dof_sect(dof,
12493 DOF_SECT_DIFOHDR, desc->dofa_difo)) == NULL)
12494 goto err;
12495
12496 act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr);
12497
12498 if (act->dtad_difo == NULL)
12499 goto err;
12500 }
12501
12502 ASSERT(first != NULL);
12503 return (first);
12504
12505err:
12506 for (act = first; act != NULL; act = next) {
12507 next = act->dtad_next;
12508 dtrace_actdesc_release(act, vstate);
12509 }
12510
12511 return (NULL);
12512}
12513
12514static dtrace_ecbdesc_t *
12515dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12516 cred_t *cr)
12517{
12518 dtrace_ecbdesc_t *ep;
12519 dof_ecbdesc_t *ecb;
12520 dtrace_probedesc_t *desc;
12521 dtrace_predicate_t *pred = NULL;
12522
12523 if (sec->dofs_size < sizeof (dof_ecbdesc_t)) {
12524 dtrace_dof_error(dof, "truncated ECB description");
12525 return (NULL);
12526 }
12527
12528 if (sec->dofs_align != sizeof (uint64_t)) {
12529 dtrace_dof_error(dof, "bad alignment in ECB description");
12530 return (NULL);
12531 }
12532
12533 ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset);
12534 sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes);
12535
12536 if (sec == NULL)
12537 return (NULL);
12538
12539 ep = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP);
12540 ep->dted_uarg = ecb->dofe_uarg;
12541 desc = &ep->dted_probe;
12542
12543 if (dtrace_dof_probedesc(dof, sec, desc) == NULL)
12544 goto err;
12545
12546 if (ecb->dofe_pred != DOF_SECIDX_NONE) {
12547 if ((sec = dtrace_dof_sect(dof,
12548 DOF_SECT_DIFOHDR, ecb->dofe_pred)) == NULL)
12549 goto err;
12550
12551 if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL)
12552 goto err;
12553
12554 ep->dted_pred.dtpdd_predicate = pred;
12555 }
12556
12557 if (ecb->dofe_actions != DOF_SECIDX_NONE) {
12558 if ((sec = dtrace_dof_sect(dof,
12559 DOF_SECT_ACTDESC, ecb->dofe_actions)) == NULL)
12560 goto err;
12561
12562 ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr);
12563
12564 if (ep->dted_action == NULL)
12565 goto err;
12566 }
12567
12568 return (ep);
12569
12570err:
12571 if (pred != NULL)
12572 dtrace_predicate_release(pred, vstate);
12573 kmem_free(ep, sizeof (dtrace_ecbdesc_t));
12574 return (NULL);
12575}
12576
12577/*
12578 * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the
12579 * specified DOF. At present, this amounts to simply adding 'ubase' to the
12580 * site of any user SETX relocations to account for load object base address.
12581 * In the future, if we need other relocations, this function can be extended.
12582 */
12583static int
12584dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase)
12585{
12586 uintptr_t daddr = (uintptr_t)dof;
12587 dof_relohdr_t *dofr =
12588 (dof_relohdr_t *)(uintptr_t)(daddr + sec->dofs_offset);
12589 dof_sec_t *ss, *rs, *ts;
12590 dof_relodesc_t *r;
12591 uint_t i, n;
12592
12593 if (sec->dofs_size < sizeof (dof_relohdr_t) ||
12594 sec->dofs_align != sizeof (dof_secidx_t)) {
12595 dtrace_dof_error(dof, "invalid relocation header");
12596 return (-1);
12597 }
12598
12599 ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab);
12600 rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec);
12601 ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec);
12602
12603 if (ss == NULL || rs == NULL || ts == NULL)
12604 return (-1); /* dtrace_dof_error() has been called already */
12605
12606 if (rs->dofs_entsize < sizeof (dof_relodesc_t) ||
12607 rs->dofs_align != sizeof (uint64_t)) {
12608 dtrace_dof_error(dof, "invalid relocation section");
12609 return (-1);
12610 }
12611
12612 r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset);
12613 n = rs->dofs_size / rs->dofs_entsize;
12614
12615 for (i = 0; i < n; i++) {
12616 uintptr_t taddr = daddr + ts->dofs_offset + r->dofr_offset;
12617
12618 switch (r->dofr_type) {
12619 case DOF_RELO_NONE:
12620 break;
12621 case DOF_RELO_SETX:
12622 if (r->dofr_offset >= ts->dofs_size || r->dofr_offset +
12623 sizeof (uint64_t) > ts->dofs_size) {
12624 dtrace_dof_error(dof, "bad relocation offset");
12625 return (-1);
12626 }
12627
12628 if (!IS_P2ALIGNED(taddr, sizeof (uint64_t))) {
12629 dtrace_dof_error(dof, "misaligned setx relo");
12630 return (-1);
12631 }
12632
12633 *(uint64_t *)taddr += ubase;
12634 break;
12635 default:
12636 dtrace_dof_error(dof, "invalid relocation type");
12637 return (-1);
12638 }
12639
12640 r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize);
12641 }
12642
12643 return (0);
12644}
12645
12646/*
12647 * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated
12648 * header: it should be at the front of a memory region that is at least
12649 * sizeof (dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in
12650 * size. It need not be validated in any other way.
12651 */
12652static int
12653dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr,
12654 dtrace_enabling_t **enabp, uint64_t ubase, int noprobes)
12655{
12656 uint64_t len = dof->dofh_loadsz, seclen;
12657 uintptr_t daddr = (uintptr_t)dof;
12658 dtrace_ecbdesc_t *ep;
12659 dtrace_enabling_t *enab;
12660 uint_t i;
12661
12662 ASSERT(MUTEX_HELD(&dtrace_lock));
12663 ASSERT(dof->dofh_loadsz >= sizeof (dof_hdr_t));
12664
12665 /*
12666 * Check the DOF header identification bytes. In addition to checking
12667 * valid settings, we also verify that unused bits/bytes are zeroed so
12668 * we can use them later without fear of regressing existing binaries.
12669 */
12670 if (bcmp(&dof->dofh_ident[DOF_ID_MAG0],
12671 DOF_MAG_STRING, DOF_MAG_STRLEN) != 0) {
12672 dtrace_dof_error(dof, "DOF magic string mismatch");
12673 return (-1);
12674 }
12675
12676 if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 &&
12677 dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) {
12678 dtrace_dof_error(dof, "DOF has invalid data model");
12679 return (-1);
12680 }
12681
12682 if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) {
12683 dtrace_dof_error(dof, "DOF encoding mismatch");
12684 return (-1);
12685 }
12686
12687 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
12688 dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) {
12689 dtrace_dof_error(dof, "DOF version mismatch");
12690 return (-1);
12691 }
12692
12693 if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) {
12694 dtrace_dof_error(dof, "DOF uses unsupported instruction set");
12695 return (-1);
12696 }
12697
12698 if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) {
12699 dtrace_dof_error(dof, "DOF uses too many integer registers");
12700 return (-1);
12701 }
12702
12703 if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) {
12704 dtrace_dof_error(dof, "DOF uses too many tuple registers");
12705 return (-1);
12706 }
12707
12708 for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) {
12709 if (dof->dofh_ident[i] != 0) {
12710 dtrace_dof_error(dof, "DOF has invalid ident byte set");
12711 return (-1);
12712 }
12713 }
12714
12715 if (dof->dofh_flags & ~DOF_FL_VALID) {
12716 dtrace_dof_error(dof, "DOF has invalid flag bits set");
12717 return (-1);
12718 }
12719
12720 if (dof->dofh_secsize == 0) {
12721 dtrace_dof_error(dof, "zero section header size");
12722 return (-1);
12723 }
12724
12725 /*
12726 * Check that the section headers don't exceed the amount of DOF
12727 * data. Note that we cast the section size and number of sections
12728 * to uint64_t's to prevent possible overflow in the multiplication.
12729 */
12730 seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize;
12731
12732 if (dof->dofh_secoff > len || seclen > len ||
12733 dof->dofh_secoff + seclen > len) {
12734 dtrace_dof_error(dof, "truncated section headers");
12735 return (-1);
12736 }
12737
12738 if (!IS_P2ALIGNED(dof->dofh_secoff, sizeof (uint64_t))) {
12739 dtrace_dof_error(dof, "misaligned section headers");
12740 return (-1);
12741 }
12742
12743 if (!IS_P2ALIGNED(dof->dofh_secsize, sizeof (uint64_t))) {
12744 dtrace_dof_error(dof, "misaligned section size");
12745 return (-1);
12746 }
12747
12748 /*
12749 * Take an initial pass through the section headers to be sure that
12750 * the headers don't have stray offsets. If the 'noprobes' flag is
12751 * set, do not permit sections relating to providers, probes, or args.
12752 */
12753 for (i = 0; i < dof->dofh_secnum; i++) {
12754 dof_sec_t *sec = (dof_sec_t *)(daddr +
12755 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12756
12757 if (noprobes) {
12758 switch (sec->dofs_type) {
12759 case DOF_SECT_PROVIDER:
12760 case DOF_SECT_PROBES:
12761 case DOF_SECT_PRARGS:
12762 case DOF_SECT_PROFFS:
12763 dtrace_dof_error(dof, "illegal sections "
12764 "for enabling");
12765 return (-1);
12766 }
12767 }
12768
12769 if (!(sec->dofs_flags & DOF_SECF_LOAD))
12770 continue; /* just ignore non-loadable sections */
12771
12772 if (sec->dofs_align & (sec->dofs_align - 1)) {
12773 dtrace_dof_error(dof, "bad section alignment");
12774 return (-1);
12775 }
12776
12777 if (sec->dofs_offset & (sec->dofs_align - 1)) {
12778 dtrace_dof_error(dof, "misaligned section");
12779 return (-1);
12780 }
12781
12782 if (sec->dofs_offset > len || sec->dofs_size > len ||
12783 sec->dofs_offset + sec->dofs_size > len) {
12784 dtrace_dof_error(dof, "corrupt section header");
12785 return (-1);
12786 }
12787
12788 if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr +
12789 sec->dofs_offset + sec->dofs_size - 1) != '\0') {
12790 dtrace_dof_error(dof, "non-terminating string table");
12791 return (-1);
12792 }
12793 }
12794
12795 /*
12796 * Take a second pass through the sections and locate and perform any
12797 * relocations that are present. We do this after the first pass to
12798 * be sure that all sections have had their headers validated.
12799 */
12800 for (i = 0; i < dof->dofh_secnum; i++) {
12801 dof_sec_t *sec = (dof_sec_t *)(daddr +
12802 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12803
12804 if (!(sec->dofs_flags & DOF_SECF_LOAD))
12805 continue; /* skip sections that are not loadable */
12806
12807 switch (sec->dofs_type) {
12808 case DOF_SECT_URELHDR:
12809 if (dtrace_dof_relocate(dof, sec, ubase) != 0)
12810 return (-1);
12811 break;
12812 }
12813 }
12814
12815 if ((enab = *enabp) == NULL)
12816 enab = *enabp = dtrace_enabling_create(vstate);
12817
12818 for (i = 0; i < dof->dofh_secnum; i++) {
12819 dof_sec_t *sec = (dof_sec_t *)(daddr +
12820 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12821
12822 if (sec->dofs_type != DOF_SECT_ECBDESC)
12823 continue;
12824
12825 if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) {
12826 dtrace_enabling_destroy(enab);
12827 *enabp = NULL;
12828 return (-1);
12829 }
12830
12831 dtrace_enabling_add(enab, ep);
12832 }
12833
12834 return (0);
12835}
12836
12837/*
12838 * Process DOF for any options. This routine assumes that the DOF has been
12839 * at least processed by dtrace_dof_slurp().
12840 */
12841static int
12842dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state)
12843{
12844 int i, rval;
12845 uint32_t entsize;
12846 size_t offs;
12847 dof_optdesc_t *desc;
12848
12849 for (i = 0; i < dof->dofh_secnum; i++) {
12850 dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof +
12851 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12852
12853 if (sec->dofs_type != DOF_SECT_OPTDESC)
12854 continue;
12855
12856 if (sec->dofs_align != sizeof (uint64_t)) {
12857 dtrace_dof_error(dof, "bad alignment in "
12858 "option description");
12859 return (EINVAL);
12860 }
12861
12862 if ((entsize = sec->dofs_entsize) == 0) {
12863 dtrace_dof_error(dof, "zeroed option entry size");
12864 return (EINVAL);
12865 }
12866
12867 if (entsize < sizeof (dof_optdesc_t)) {
12868 dtrace_dof_error(dof, "bad option entry size");
12869 return (EINVAL);
12870 }
12871
12872 for (offs = 0; offs < sec->dofs_size; offs += entsize) {
12873 desc = (dof_optdesc_t *)((uintptr_t)dof +
12874 (uintptr_t)sec->dofs_offset + offs);
12875
12876 if (desc->dofo_strtab != DOF_SECIDX_NONE) {
12877 dtrace_dof_error(dof, "non-zero option string");
12878 return (EINVAL);
12879 }
12880
12881 if (desc->dofo_value == DTRACEOPT_UNSET) {
12882 dtrace_dof_error(dof, "unset option");
12883 return (EINVAL);
12884 }
12885
12886 if ((rval = dtrace_state_option(state,
12887 desc->dofo_option, desc->dofo_value)) != 0) {
12888 dtrace_dof_error(dof, "rejected option");
12889 return (rval);
12890 }
12891 }
12892 }
12893
12894 return (0);
12895}
12896
12897/*
12898 * DTrace Consumer State Functions
12899 */
12900static int
12901dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size)
12902{
12903 size_t hashsize, maxper, min, chunksize = dstate->dtds_chunksize;
12904 void *base;
12905 uintptr_t limit;
12906 dtrace_dynvar_t *dvar, *next, *start;
12907 int i;
12908
12909 ASSERT(MUTEX_HELD(&dtrace_lock));
12910 ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL);
12911
12912 bzero(dstate, sizeof (dtrace_dstate_t));
12913
12914 if ((dstate->dtds_chunksize = chunksize) == 0)
12915 dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE;
12916
12917 if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)))
12918 size = min;
12919
12920 if ((base = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
12921 return (ENOMEM);
12922
12923 dstate->dtds_size = size;
12924 dstate->dtds_base = base;
12925 dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, KM_SLEEP);
12926 bzero(dstate->dtds_percpu, NCPU * sizeof (dtrace_dstate_percpu_t));
12927
12928 hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t));
12929
12930 if (hashsize != 1 && (hashsize & 1))
12931 hashsize--;
12932
12933 dstate->dtds_hashsize = hashsize;
12934 dstate->dtds_hash = dstate->dtds_base;
12935
12936 /*
12937 * Set all of our hash buckets to point to the single sink, and (if
12938 * it hasn't already been set), set the sink's hash value to be the
12939 * sink sentinel value. The sink is needed for dynamic variable
12940 * lookups to know that they have iterated over an entire, valid hash
12941 * chain.
12942 */
12943 for (i = 0; i < hashsize; i++)
12944 dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink;
12945
12946 if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK)
12947 dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK;
12948
12949 /*
12950 * Determine number of active CPUs. Divide free list evenly among
12951 * active CPUs.
12952 */
12953 start = (dtrace_dynvar_t *)
12954 ((uintptr_t)base + hashsize * sizeof (dtrace_dynhash_t));
12955 limit = (uintptr_t)base + size;
12956
12957 maxper = (limit - (uintptr_t)start) / NCPU;
12958 maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize;
12959
12960#if !defined(sun)
12961 CPU_FOREACH(i) {
12962#else
12963 for (i = 0; i < NCPU; i++) {
12964#endif
12965 dstate->dtds_percpu[i].dtdsc_free = dvar = start;
12966
12967 /*
12968 * If we don't even have enough chunks to make it once through
12969 * NCPUs, we're just going to allocate everything to the first
12970 * CPU. And if we're on the last CPU, we're going to allocate
12971 * whatever is left over. In either case, we set the limit to
12972 * be the limit of the dynamic variable space.
12973 */
12974 if (maxper == 0 || i == NCPU - 1) {
12975 limit = (uintptr_t)base + size;
12976 start = NULL;
12977 } else {
12978 limit = (uintptr_t)start + maxper;
12979 start = (dtrace_dynvar_t *)limit;
12980 }
12981
12982 ASSERT(limit <= (uintptr_t)base + size);
12983
12984 for (;;) {
12985 next = (dtrace_dynvar_t *)((uintptr_t)dvar +
12986 dstate->dtds_chunksize);
12987
12988 if ((uintptr_t)next + dstate->dtds_chunksize >= limit)
12989 break;
12990
12991 dvar->dtdv_next = next;
12992 dvar = next;
12993 }
12994
12995 if (maxper == 0)
12996 break;
12997 }
12998
12999 return (0);
13000}
13001
13002static void
13003dtrace_dstate_fini(dtrace_dstate_t *dstate)
13004{
13005 ASSERT(MUTEX_HELD(&cpu_lock));
13006
13007 if (dstate->dtds_base == NULL)
13008 return;
13009
13010 kmem_free(dstate->dtds_base, dstate->dtds_size);
13011 kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu);
13012}
13013
13014static void
13015dtrace_vstate_fini(dtrace_vstate_t *vstate)
13016{
13017 /*
13018 * Logical XOR, where are you?
13019 */
13020 ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL));
13021
13022 if (vstate->dtvs_nglobals > 0) {
13023 kmem_free(vstate->dtvs_globals, vstate->dtvs_nglobals *
13024 sizeof (dtrace_statvar_t *));
13025 }
13026
13027 if (vstate->dtvs_ntlocals > 0) {
13028 kmem_free(vstate->dtvs_tlocals, vstate->dtvs_ntlocals *
13029 sizeof (dtrace_difv_t));
13030 }
13031
13032 ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL));
13033
13034 if (vstate->dtvs_nlocals > 0) {
13035 kmem_free(vstate->dtvs_locals, vstate->dtvs_nlocals *
13036 sizeof (dtrace_statvar_t *));
13037 }
13038}
13039
13040#if defined(sun)
13041static void
13042dtrace_state_clean(dtrace_state_t *state)
13043{
13044 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE)
13045 return;
13046
13047 dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars);
13048 dtrace_speculation_clean(state);
13049}
13050
13051static void
13052dtrace_state_deadman(dtrace_state_t *state)
13053{
13054 hrtime_t now;
13055
13056 dtrace_sync();
13057
13058 now = dtrace_gethrtime();
13059
13060 if (state != dtrace_anon.dta_state &&
13061 now - state->dts_laststatus >= dtrace_deadman_user)
13062 return;
13063
13064 /*
13065 * We must be sure that dts_alive never appears to be less than the
13066 * value upon entry to dtrace_state_deadman(), and because we lack a
13067 * dtrace_cas64(), we cannot store to it atomically. We thus instead
13068 * store INT64_MAX to it, followed by a memory barrier, followed by
13069 * the new value. This assures that dts_alive never appears to be
13070 * less than its true value, regardless of the order in which the
13071 * stores to the underlying storage are issued.
13072 */
13073 state->dts_alive = INT64_MAX;
13074 dtrace_membar_producer();
13075 state->dts_alive = now;
13076}
13077#else
13078static void
13079dtrace_state_clean(void *arg)
13080{
13081 dtrace_state_t *state = arg;
13082 dtrace_optval_t *opt = state->dts_options;
13083
13084 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE)
13085 return;
13086
13087 dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars);
13088 dtrace_speculation_clean(state);
13089
13090 callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC,
13091 dtrace_state_clean, state);
13092}
13093
13094static void
13095dtrace_state_deadman(void *arg)
13096{
13097 dtrace_state_t *state = arg;
13098 hrtime_t now;
13099
13100 dtrace_sync();
13101
13102 dtrace_debug_output();
13103
13104 now = dtrace_gethrtime();
13105
13106 if (state != dtrace_anon.dta_state &&
13107 now - state->dts_laststatus >= dtrace_deadman_user)
13108 return;
13109
13110 /*
13111 * We must be sure that dts_alive never appears to be less than the
13112 * value upon entry to dtrace_state_deadman(), and because we lack a
13113 * dtrace_cas64(), we cannot store to it atomically. We thus instead
13114 * store INT64_MAX to it, followed by a memory barrier, followed by
13115 * the new value. This assures that dts_alive never appears to be
13116 * less than its true value, regardless of the order in which the
13117 * stores to the underlying storage are issued.
13118 */
13119 state->dts_alive = INT64_MAX;
13120 dtrace_membar_producer();
13121 state->dts_alive = now;
13122
13123 callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC,
13124 dtrace_state_deadman, state);
13125}
13126#endif
13127
13128static dtrace_state_t *
13129#if defined(sun)
13130dtrace_state_create(dev_t *devp, cred_t *cr)
13131#else
13132dtrace_state_create(struct cdev *dev)
13133#endif
13134{
13135#if defined(sun)
13136 minor_t minor;
13137 major_t major;
13138#else
13139 cred_t *cr = NULL;
13140 int m = 0;
13141#endif
13142 char c[30];
13143 dtrace_state_t *state;
13144 dtrace_optval_t *opt;
13145 int bufsize = NCPU * sizeof (dtrace_buffer_t), i;
13146
13147 ASSERT(MUTEX_HELD(&dtrace_lock));
13148 ASSERT(MUTEX_HELD(&cpu_lock));
13149
13150#if defined(sun)
13151 minor = (minor_t)(uintptr_t)vmem_alloc(dtrace_minor, 1,
13152 VM_BESTFIT | VM_SLEEP);
13153
13154 if (ddi_soft_state_zalloc(dtrace_softstate, minor) != DDI_SUCCESS) {
13155 vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1);
13156 return (NULL);
13157 }
13158
13159 state = ddi_get_soft_state(dtrace_softstate, minor);
13160#else
13161 if (dev != NULL) {
13162 cr = dev->si_cred;
13163 m = dev2unit(dev);
13164 }
13165
13166 /* Allocate memory for the state. */
13167 state = kmem_zalloc(sizeof(dtrace_state_t), KM_SLEEP);
13168#endif
13169
13170 state->dts_epid = DTRACE_EPIDNONE + 1;
13171
13172 (void) snprintf(c, sizeof (c), "dtrace_aggid_%d", m);
13173#if defined(sun)
13174 state->dts_aggid_arena = vmem_create(c, (void *)1, UINT32_MAX, 1,
13175 NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
13176
13177 if (devp != NULL) {
13178 major = getemajor(*devp);
13179 } else {
13180 major = ddi_driver_major(dtrace_devi);
13181 }
13182
13183 state->dts_dev = makedevice(major, minor);
13184
13185 if (devp != NULL)
13186 *devp = state->dts_dev;
13187#else
13188 state->dts_aggid_arena = new_unrhdr(1, INT_MAX, &dtrace_unr_mtx);
13189 state->dts_dev = dev;
13190#endif
13191
13192 /*
13193 * We allocate NCPU buffers. On the one hand, this can be quite
13194 * a bit of memory per instance (nearly 36K on a Starcat). On the
13195 * other hand, it saves an additional memory reference in the probe
13196 * path.
13197 */
13198 state->dts_buffer = kmem_zalloc(bufsize, KM_SLEEP);
13199 state->dts_aggbuffer = kmem_zalloc(bufsize, KM_SLEEP);
13200
13201#if defined(sun)
13202 state->dts_cleaner = CYCLIC_NONE;
13203 state->dts_deadman = CYCLIC_NONE;
13204#else
13205 callout_init(&state->dts_cleaner, CALLOUT_MPSAFE);
13206 callout_init(&state->dts_deadman, CALLOUT_MPSAFE);
13207#endif
13208 state->dts_vstate.dtvs_state = state;
13209
13210 for (i = 0; i < DTRACEOPT_MAX; i++)
13211 state->dts_options[i] = DTRACEOPT_UNSET;
13212
13213 /*
13214 * Set the default options.
13215 */
13216 opt = state->dts_options;
13217 opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH;
13218 opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO;
13219 opt[DTRACEOPT_NSPEC] = dtrace_nspec_default;
13220 opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default;
13221 opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL;
13222 opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default;
13223 opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default;
13224 opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default;
13225 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default;
13226 opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default;
13227 opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default;
13228 opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default;
13229 opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default;
13230 opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default;
13231
13232 state->dts_activity = DTRACE_ACTIVITY_INACTIVE;
13233
13234 /*
13235 * Depending on the user credentials, we set flag bits which alter probe
13236 * visibility or the amount of destructiveness allowed. In the case of
13237 * actual anonymous tracing, or the possession of all privileges, all of
13238 * the normal checks are bypassed.
13239 */
13240 if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) {
13241 state->dts_cred.dcr_visible = DTRACE_CRV_ALL;
13242 state->dts_cred.dcr_action = DTRACE_CRA_ALL;
13243 } else {
13244 /*
13245 * Set up the credentials for this instantiation. We take a
13246 * hold on the credential to prevent it from disappearing on
13247 * us; this in turn prevents the zone_t referenced by this
13248 * credential from disappearing. This means that we can
13249 * examine the credential and the zone from probe context.
13250 */
13251 crhold(cr);
13252 state->dts_cred.dcr_cred = cr;
13253
13254 /*
13255 * CRA_PROC means "we have *some* privilege for dtrace" and
13256 * unlocks the use of variables like pid, zonename, etc.
13257 */
13258 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE) ||
13259 PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) {
13260 state->dts_cred.dcr_action |= DTRACE_CRA_PROC;
13261 }
13262
13263 /*
13264 * dtrace_user allows use of syscall and profile providers.
13265 * If the user also has proc_owner and/or proc_zone, we
13266 * extend the scope to include additional visibility and
13267 * destructive power.
13268 */
13269 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) {
13270 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) {
13271 state->dts_cred.dcr_visible |=
13272 DTRACE_CRV_ALLPROC;
13273
13274 state->dts_cred.dcr_action |=
13275 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
13276 }
13277
13278 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) {
13279 state->dts_cred.dcr_visible |=
13280 DTRACE_CRV_ALLZONE;
13281
13282 state->dts_cred.dcr_action |=
13283 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
13284 }
13285
13286 /*
13287 * If we have all privs in whatever zone this is,
13288 * we can do destructive things to processes which
13289 * have altered credentials.
13290 */
13291#if defined(sun)
13292 if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE),
13293 cr->cr_zone->zone_privset)) {
13294 state->dts_cred.dcr_action |=
13295 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG;
13296 }
13297#endif
13298 }
13299
13300 /*
13301 * Holding the dtrace_kernel privilege also implies that
13302 * the user has the dtrace_user privilege from a visibility
13303 * perspective. But without further privileges, some
13304 * destructive actions are not available.
13305 */
13306 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) {
13307 /*
13308 * Make all probes in all zones visible. However,
13309 * this doesn't mean that all actions become available
13310 * to all zones.
13311 */
13312 state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL |
13313 DTRACE_CRV_ALLPROC | DTRACE_CRV_ALLZONE;
13314
13315 state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL |
13316 DTRACE_CRA_PROC;
13317 /*
13318 * Holding proc_owner means that destructive actions
13319 * for *this* zone are allowed.
13320 */
13321 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
13322 state->dts_cred.dcr_action |=
13323 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
13324
13325 /*
13326 * Holding proc_zone means that destructive actions
13327 * for this user/group ID in all zones is allowed.
13328 */
13329 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
13330 state->dts_cred.dcr_action |=
13331 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
13332
13333#if defined(sun)
13334 /*
13335 * If we have all privs in whatever zone this is,
13336 * we can do destructive things to processes which
13337 * have altered credentials.
13338 */
13339 if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE),
13340 cr->cr_zone->zone_privset)) {
13341 state->dts_cred.dcr_action |=
13342 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG;
13343 }
13344#endif
13345 }
13346
13347 /*
13348 * Holding the dtrace_proc privilege gives control over fasttrap
13349 * and pid providers. We need to grant wider destructive
13350 * privileges in the event that the user has proc_owner and/or
13351 * proc_zone.
13352 */
13353 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) {
13354 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
13355 state->dts_cred.dcr_action |=
13356 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
13357
13358 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
13359 state->dts_cred.dcr_action |=
13360 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
13361 }
13362 }
13363
13364 return (state);
13365}
13366
13367static int
13368dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, int which)
13369{
13370 dtrace_optval_t *opt = state->dts_options, size;
13371 processorid_t cpu = 0;;
13372 int flags = 0, rval;
13373
13374 ASSERT(MUTEX_HELD(&dtrace_lock));
13375 ASSERT(MUTEX_HELD(&cpu_lock));
13376 ASSERT(which < DTRACEOPT_MAX);
13377 ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE ||
13378 (state == dtrace_anon.dta_state &&
13379 state->dts_activity == DTRACE_ACTIVITY_ACTIVE));
13380
13381 if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0)
13382 return (0);
13383
13384 if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET)
13385 cpu = opt[DTRACEOPT_CPU];
13386
13387 if (which == DTRACEOPT_SPECSIZE)
13388 flags |= DTRACEBUF_NOSWITCH;
13389
13390 if (which == DTRACEOPT_BUFSIZE) {
13391 if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING)
13392 flags |= DTRACEBUF_RING;
13393
13394 if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL)
13395 flags |= DTRACEBUF_FILL;
13396
13397 if (state != dtrace_anon.dta_state ||
13398 state->dts_activity != DTRACE_ACTIVITY_ACTIVE)
13399 flags |= DTRACEBUF_INACTIVE;
13400 }
13401
13402 for (size = opt[which]; size >= sizeof (uint64_t); size >>= 1) {
13403 /*
13404 * The size must be 8-byte aligned. If the size is not 8-byte
13405 * aligned, drop it down by the difference.
13406 */
13407 if (size & (sizeof (uint64_t) - 1))
13408 size -= size & (sizeof (uint64_t) - 1);
13409
13410 if (size < state->dts_reserve) {
13411 /*
13412 * Buffers always must be large enough to accommodate
13413 * their prereserved space. We return E2BIG instead
13414 * of ENOMEM in this case to allow for user-level
13415 * software to differentiate the cases.
13416 */
13417 return (E2BIG);
13418 }
13419
13420 rval = dtrace_buffer_alloc(buf, size, flags, cpu);
13421
13422 if (rval != ENOMEM) {
13423 opt[which] = size;
13424 return (rval);
13425 }
13426
13427 if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL)
13428 return (rval);
13429 }
13430
13431 return (ENOMEM);
13432}
13433
13434static int
13435dtrace_state_buffers(dtrace_state_t *state)
13436{
13437 dtrace_speculation_t *spec = state->dts_speculations;
13438 int rval, i;
13439
13440 if ((rval = dtrace_state_buffer(state, state->dts_buffer,
13441 DTRACEOPT_BUFSIZE)) != 0)
13442 return (rval);
13443
13444 if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer,
13445 DTRACEOPT_AGGSIZE)) != 0)
13446 return (rval);
13447
13448 for (i = 0; i < state->dts_nspeculations; i++) {
13449 if ((rval = dtrace_state_buffer(state,
13450 spec[i].dtsp_buffer, DTRACEOPT_SPECSIZE)) != 0)
13451 return (rval);
13452 }
13453
13454 return (0);
13455}
13456
13457static void
13458dtrace_state_prereserve(dtrace_state_t *state)
13459{
13460 dtrace_ecb_t *ecb;
13461 dtrace_probe_t *probe;
13462
13463 state->dts_reserve = 0;
13464
13465 if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL)
13466 return;
13467
13468 /*
13469 * If our buffer policy is a "fill" buffer policy, we need to set the
13470 * prereserved space to be the space required by the END probes.
13471 */
13472 probe = dtrace_probes[dtrace_probeid_end - 1];
13473 ASSERT(probe != NULL);
13474
13475 for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) {
13476 if (ecb->dte_state != state)
13477 continue;
13478
13479 state->dts_reserve += ecb->dte_needed + ecb->dte_alignment;
13480 }
13481}
13482
13483static int
13484dtrace_state_go(dtrace_state_t *state, processorid_t *cpu)
13485{
13486 dtrace_optval_t *opt = state->dts_options, sz, nspec;
13487 dtrace_speculation_t *spec;
13488 dtrace_buffer_t *buf;
13489#if defined(sun)
13490 cyc_handler_t hdlr;
13491 cyc_time_t when;
13492#endif
13493 int rval = 0, i, bufsize = NCPU * sizeof (dtrace_buffer_t);
13494 dtrace_icookie_t cookie;
13495
13496 mutex_enter(&cpu_lock);
13497 mutex_enter(&dtrace_lock);
13498
13499 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) {
13500 rval = EBUSY;
13501 goto out;
13502 }
13503
13504 /*
13505 * Before we can perform any checks, we must prime all of the
13506 * retained enablings that correspond to this state.
13507 */
13508 dtrace_enabling_prime(state);
13509
13510 if (state->dts_destructive && !state->dts_cred.dcr_destructive) {
13511 rval = EACCES;
13512 goto out;
13513 }
13514
13515 dtrace_state_prereserve(state);
13516
13517 /*
13518 * Now we want to do is try to allocate our speculations.
13519 * We do not automatically resize the number of speculations; if
13520 * this fails, we will fail the operation.
13521 */
13522 nspec = opt[DTRACEOPT_NSPEC];
13523 ASSERT(nspec != DTRACEOPT_UNSET);
13524
13525 if (nspec > INT_MAX) {
13526 rval = ENOMEM;
13527 goto out;
13528 }
13529
13530 spec = kmem_zalloc(nspec * sizeof (dtrace_speculation_t), KM_NOSLEEP);
13531
13532 if (spec == NULL) {
13533 rval = ENOMEM;
13534 goto out;
13535 }
13536
13537 state->dts_speculations = spec;
13538 state->dts_nspeculations = (int)nspec;
13539
13540 for (i = 0; i < nspec; i++) {
13541 if ((buf = kmem_zalloc(bufsize, KM_NOSLEEP)) == NULL) {
13542 rval = ENOMEM;
13543 goto err;
13544 }
13545
13546 spec[i].dtsp_buffer = buf;
13547 }
13548
13549 if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) {
13550 if (dtrace_anon.dta_state == NULL) {
13551 rval = ENOENT;
13552 goto out;
13553 }
13554
13555 if (state->dts_necbs != 0) {
13556 rval = EALREADY;
13557 goto out;
13558 }
13559
13560 state->dts_anon = dtrace_anon_grab();
13561 ASSERT(state->dts_anon != NULL);
13562 state = state->dts_anon;
13563
13564 /*
13565 * We want "grabanon" to be set in the grabbed state, so we'll
13566 * copy that option value from the grabbing state into the
13567 * grabbed state.
13568 */
13569 state->dts_options[DTRACEOPT_GRABANON] =
13570 opt[DTRACEOPT_GRABANON];
13571
13572 *cpu = dtrace_anon.dta_beganon;
13573
13574 /*
13575 * If the anonymous state is active (as it almost certainly
13576 * is if the anonymous enabling ultimately matched anything),
13577 * we don't allow any further option processing -- but we
13578 * don't return failure.
13579 */
13580 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
13581 goto out;
13582 }
13583
13584 if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET &&
13585 opt[DTRACEOPT_AGGSIZE] != 0) {
13586 if (state->dts_aggregations == NULL) {
13587 /*
13588 * We're not going to create an aggregation buffer
13589 * because we don't have any ECBs that contain
13590 * aggregations -- set this option to 0.
13591 */
13592 opt[DTRACEOPT_AGGSIZE] = 0;
13593 } else {
13594 /*
13595 * If we have an aggregation buffer, we must also have
13596 * a buffer to use as scratch.
13597 */
13598 if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET ||
13599 opt[DTRACEOPT_BUFSIZE] < state->dts_needed) {
13600 opt[DTRACEOPT_BUFSIZE] = state->dts_needed;
13601 }
13602 }
13603 }
13604
13605 if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET &&
13606 opt[DTRACEOPT_SPECSIZE] != 0) {
13607 if (!state->dts_speculates) {
13608 /*
13609 * We're not going to create speculation buffers
13610 * because we don't have any ECBs that actually
13611 * speculate -- set the speculation size to 0.
13612 */
13613 opt[DTRACEOPT_SPECSIZE] = 0;
13614 }
13615 }
13616
13617 /*
13618 * The bare minimum size for any buffer that we're actually going to
13619 * do anything to is sizeof (uint64_t).
13620 */
13621 sz = sizeof (uint64_t);
13622
13623 if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) ||
13624 (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) ||
13625 (state->dts_aggregations != NULL && opt[DTRACEOPT_AGGSIZE] < sz)) {
13626 /*
13627 * A buffer size has been explicitly set to 0 (or to a size
13628 * that will be adjusted to 0) and we need the space -- we
13629 * need to return failure. We return ENOSPC to differentiate
13630 * it from failing to allocate a buffer due to failure to meet
13631 * the reserve (for which we return E2BIG).
13632 */
13633 rval = ENOSPC;
13634 goto out;
13635 }
13636
13637 if ((rval = dtrace_state_buffers(state)) != 0)
13638 goto err;
13639
13640 if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET)
13641 sz = dtrace_dstate_defsize;
13642
13643 do {
13644 rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz);
13645
13646 if (rval == 0)
13647 break;
13648
13649 if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL)
13650 goto err;
13651 } while (sz >>= 1);
13652
13653 opt[DTRACEOPT_DYNVARSIZE] = sz;
13654
13655 if (rval != 0)
13656 goto err;
13657
13658 if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max)
13659 opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max;
13660
13661 if (opt[DTRACEOPT_CLEANRATE] == 0)
13662 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max;
13663
13664 if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min)
13665 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min;
13666
13667 if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max)
13668 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max;
13669
13670 state->dts_alive = state->dts_laststatus = dtrace_gethrtime();
13671#if defined(sun)
13672 hdlr.cyh_func = (cyc_func_t)dtrace_state_clean;
13673 hdlr.cyh_arg = state;
13674 hdlr.cyh_level = CY_LOW_LEVEL;
13675
13676 when.cyt_when = 0;
13677 when.cyt_interval = opt[DTRACEOPT_CLEANRATE];
13678
13679 state->dts_cleaner = cyclic_add(&hdlr, &when);
13680
13681 hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman;
13682 hdlr.cyh_arg = state;
13683 hdlr.cyh_level = CY_LOW_LEVEL;
13684
13685 when.cyt_when = 0;
13686 when.cyt_interval = dtrace_deadman_interval;
13687
13688 state->dts_deadman = cyclic_add(&hdlr, &when);
13689#else
13690 callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC,
13691 dtrace_state_clean, state);
13692 callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC,
13693 dtrace_state_deadman, state);
13694#endif
13695
13696 state->dts_activity = DTRACE_ACTIVITY_WARMUP;
13697
13698 /*
13699 * Now it's time to actually fire the BEGIN probe. We need to disable
13700 * interrupts here both to record the CPU on which we fired the BEGIN
13701 * probe (the data from this CPU will be processed first at user
13702 * level) and to manually activate the buffer for this CPU.
13703 */
13704 cookie = dtrace_interrupt_disable();
13705 *cpu = curcpu;
13706 ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE);
13707 state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE;
13708
13709 dtrace_probe(dtrace_probeid_begin,
13710 (uint64_t)(uintptr_t)state, 0, 0, 0, 0);
13711 dtrace_interrupt_enable(cookie);
13712 /*
13713 * We may have had an exit action from a BEGIN probe; only change our
13714 * state to ACTIVE if we're still in WARMUP.
13715 */
13716 ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP ||
13717 state->dts_activity == DTRACE_ACTIVITY_DRAINING);
13718
13719 if (state->dts_activity == DTRACE_ACTIVITY_WARMUP)
13720 state->dts_activity = DTRACE_ACTIVITY_ACTIVE;
13721
13722 /*
13723 * Regardless of whether or not now we're in ACTIVE or DRAINING, we
13724 * want each CPU to transition its principal buffer out of the
13725 * INACTIVE state. Doing this assures that no CPU will suddenly begin
13726 * processing an ECB halfway down a probe's ECB chain; all CPUs will
13727 * atomically transition from processing none of a state's ECBs to
13728 * processing all of them.
13729 */
13730 dtrace_xcall(DTRACE_CPUALL,
13731 (dtrace_xcall_t)dtrace_buffer_activate, state);
13732 goto out;
13733
13734err:
13735 dtrace_buffer_free(state->dts_buffer);
13736 dtrace_buffer_free(state->dts_aggbuffer);
13737
13738 if ((nspec = state->dts_nspeculations) == 0) {
13739 ASSERT(state->dts_speculations == NULL);
13740 goto out;
13741 }
13742
13743 spec = state->dts_speculations;
13744 ASSERT(spec != NULL);
13745
13746 for (i = 0; i < state->dts_nspeculations; i++) {
13747 if ((buf = spec[i].dtsp_buffer) == NULL)
13748 break;
13749
13750 dtrace_buffer_free(buf);
13751 kmem_free(buf, bufsize);
13752 }
13753
13754 kmem_free(spec, nspec * sizeof (dtrace_speculation_t));
13755 state->dts_nspeculations = 0;
13756 state->dts_speculations = NULL;
13757
13758out:
13759 mutex_exit(&dtrace_lock);
13760 mutex_exit(&cpu_lock);
13761
13762 return (rval);
13763}
13764
13765static int
13766dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu)
13767{
13768 dtrace_icookie_t cookie;
13769
13770 ASSERT(MUTEX_HELD(&dtrace_lock));
13771
13772 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE &&
13773 state->dts_activity != DTRACE_ACTIVITY_DRAINING)
13774 return (EINVAL);
13775
13776 /*
13777 * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync
13778 * to be sure that every CPU has seen it. See below for the details
13779 * on why this is done.
13780 */
13781 state->dts_activity = DTRACE_ACTIVITY_DRAINING;
13782 dtrace_sync();
13783
13784 /*
13785 * By this point, it is impossible for any CPU to be still processing
13786 * with DTRACE_ACTIVITY_ACTIVE. We can thus set our activity to
13787 * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any
13788 * other CPU in dtrace_buffer_reserve(). This allows dtrace_probe()
13789 * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN
13790 * iff we're in the END probe.
13791 */
13792 state->dts_activity = DTRACE_ACTIVITY_COOLDOWN;
13793 dtrace_sync();
13794 ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN);
13795
13796 /*
13797 * Finally, we can release the reserve and call the END probe. We
13798 * disable interrupts across calling the END probe to allow us to
13799 * return the CPU on which we actually called the END probe. This
13800 * allows user-land to be sure that this CPU's principal buffer is
13801 * processed last.
13802 */
13803 state->dts_reserve = 0;
13804
13805 cookie = dtrace_interrupt_disable();
13806 *cpu = curcpu;
13807 dtrace_probe(dtrace_probeid_end,
13808 (uint64_t)(uintptr_t)state, 0, 0, 0, 0);
13809 dtrace_interrupt_enable(cookie);
13810
13811 state->dts_activity = DTRACE_ACTIVITY_STOPPED;
13812 dtrace_sync();
13813
13814 return (0);
13815}
13816
13817static int
13818dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option,
13819 dtrace_optval_t val)
13820{
13821 ASSERT(MUTEX_HELD(&dtrace_lock));
13822
13823 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
13824 return (EBUSY);
13825
13826 if (option >= DTRACEOPT_MAX)
13827 return (EINVAL);
13828
13829 if (option != DTRACEOPT_CPU && val < 0)
13830 return (EINVAL);
13831
13832 switch (option) {
13833 case DTRACEOPT_DESTRUCTIVE:
13834 if (dtrace_destructive_disallow)
13835 return (EACCES);
13836
13837 state->dts_cred.dcr_destructive = 1;
13838 break;
13839
13840 case DTRACEOPT_BUFSIZE:
13841 case DTRACEOPT_DYNVARSIZE:
13842 case DTRACEOPT_AGGSIZE:
13843 case DTRACEOPT_SPECSIZE:
13844 case DTRACEOPT_STRSIZE:
13845 if (val < 0)
13846 return (EINVAL);
13847
13848 if (val >= LONG_MAX) {
13849 /*
13850 * If this is an otherwise negative value, set it to
13851 * the highest multiple of 128m less than LONG_MAX.
13852 * Technically, we're adjusting the size without
13853 * regard to the buffer resizing policy, but in fact,
13854 * this has no effect -- if we set the buffer size to
13855 * ~LONG_MAX and the buffer policy is ultimately set to
13856 * be "manual", the buffer allocation is guaranteed to
13857 * fail, if only because the allocation requires two
13858 * buffers. (We set the the size to the highest
13859 * multiple of 128m because it ensures that the size
13860 * will remain a multiple of a megabyte when
13861 * repeatedly halved -- all the way down to 15m.)
13862 */
13863 val = LONG_MAX - (1 << 27) + 1;
13864 }
13865 }
13866
13867 state->dts_options[option] = val;
13868
13869 return (0);
13870}
13871
13872static void
13873dtrace_state_destroy(dtrace_state_t *state)
13874{
13875 dtrace_ecb_t *ecb;
13876 dtrace_vstate_t *vstate = &state->dts_vstate;
13877#if defined(sun)
13878 minor_t minor = getminor(state->dts_dev);
13879#endif
13880 int i, bufsize = NCPU * sizeof (dtrace_buffer_t);
13881 dtrace_speculation_t *spec = state->dts_speculations;
13882 int nspec = state->dts_nspeculations;
13883 uint32_t match;
13884
13885 ASSERT(MUTEX_HELD(&dtrace_lock));
13886 ASSERT(MUTEX_HELD(&cpu_lock));
13887
13888 /*
13889 * First, retract any retained enablings for this state.
13890 */
13891 dtrace_enabling_retract(state);
13892 ASSERT(state->dts_nretained == 0);
13893
13894 if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE ||
13895 state->dts_activity == DTRACE_ACTIVITY_DRAINING) {
13896 /*
13897 * We have managed to come into dtrace_state_destroy() on a
13898 * hot enabling -- almost certainly because of a disorderly
13899 * shutdown of a consumer. (That is, a consumer that is
13900 * exiting without having called dtrace_stop().) In this case,
13901 * we're going to set our activity to be KILLED, and then
13902 * issue a sync to be sure that everyone is out of probe
13903 * context before we start blowing away ECBs.
13904 */
13905 state->dts_activity = DTRACE_ACTIVITY_KILLED;
13906 dtrace_sync();
13907 }
13908
13909 /*
13910 * Release the credential hold we took in dtrace_state_create().
13911 */
13912 if (state->dts_cred.dcr_cred != NULL)
13913 crfree(state->dts_cred.dcr_cred);
13914
13915 /*
13916 * Now we can safely disable and destroy any enabled probes. Because
13917 * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress
13918 * (especially if they're all enabled), we take two passes through the
13919 * ECBs: in the first, we disable just DTRACE_PRIV_KERNEL probes, and
13920 * in the second we disable whatever is left over.
13921 */
13922 for (match = DTRACE_PRIV_KERNEL; ; match = 0) {
13923 for (i = 0; i < state->dts_necbs; i++) {
13924 if ((ecb = state->dts_ecbs[i]) == NULL)
13925 continue;
13926
13927 if (match && ecb->dte_probe != NULL) {
13928 dtrace_probe_t *probe = ecb->dte_probe;
13929 dtrace_provider_t *prov = probe->dtpr_provider;
13930
13931 if (!(prov->dtpv_priv.dtpp_flags & match))
13932 continue;
13933 }
13934
13935 dtrace_ecb_disable(ecb);
13936 dtrace_ecb_destroy(ecb);
13937 }
13938
13939 if (!match)
13940 break;
13941 }
13942
13943 /*
13944 * Before we free the buffers, perform one more sync to assure that
13945 * every CPU is out of probe context.
13946 */
13947 dtrace_sync();
13948
13949 dtrace_buffer_free(state->dts_buffer);
13950 dtrace_buffer_free(state->dts_aggbuffer);
13951
13952 for (i = 0; i < nspec; i++)
13953 dtrace_buffer_free(spec[i].dtsp_buffer);
13954
13955#if defined(sun)
13956 if (state->dts_cleaner != CYCLIC_NONE)
13957 cyclic_remove(state->dts_cleaner);
13958
13959 if (state->dts_deadman != CYCLIC_NONE)
13960 cyclic_remove(state->dts_deadman);
13961#else
13962 callout_stop(&state->dts_cleaner);
13963 callout_drain(&state->dts_cleaner);
13964 callout_stop(&state->dts_deadman);
13965 callout_drain(&state->dts_deadman);
13966#endif
13967
13968 dtrace_dstate_fini(&vstate->dtvs_dynvars);
13969 dtrace_vstate_fini(vstate);
13970 if (state->dts_ecbs != NULL)
13971 kmem_free(state->dts_ecbs, state->dts_necbs * sizeof (dtrace_ecb_t *));
13972
13973 if (state->dts_aggregations != NULL) {
13974#ifdef DEBUG
13975 for (i = 0; i < state->dts_naggregations; i++)
13976 ASSERT(state->dts_aggregations[i] == NULL);
13977#endif
13978 ASSERT(state->dts_naggregations > 0);
13979 kmem_free(state->dts_aggregations,
13980 state->dts_naggregations * sizeof (dtrace_aggregation_t *));
13981 }
13982
13983 kmem_free(state->dts_buffer, bufsize);
13984 kmem_free(state->dts_aggbuffer, bufsize);
13985
13986 for (i = 0; i < nspec; i++)
13987 kmem_free(spec[i].dtsp_buffer, bufsize);
13988
13989 if (spec != NULL)
13990 kmem_free(spec, nspec * sizeof (dtrace_speculation_t));
13991
13992 dtrace_format_destroy(state);
13993
13994 if (state->dts_aggid_arena != NULL) {
13995#if defined(sun)
13996 vmem_destroy(state->dts_aggid_arena);
13997#else
13998 delete_unrhdr(state->dts_aggid_arena);
13999#endif
14000 state->dts_aggid_arena = NULL;
14001 }
14002#if defined(sun)
14003 ddi_soft_state_free(dtrace_softstate, minor);
14004 vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1);
14005#endif
14006}
14007
14008/*
14009 * DTrace Anonymous Enabling Functions
14010 */
14011static dtrace_state_t *
14012dtrace_anon_grab(void)
14013{
14014 dtrace_state_t *state;
14015
14016 ASSERT(MUTEX_HELD(&dtrace_lock));
14017
14018 if ((state = dtrace_anon.dta_state) == NULL) {
14019 ASSERT(dtrace_anon.dta_enabling == NULL);
14020 return (NULL);
14021 }
14022
14023 ASSERT(dtrace_anon.dta_enabling != NULL);
14024 ASSERT(dtrace_retained != NULL);
14025
14026 dtrace_enabling_destroy(dtrace_anon.dta_enabling);
14027 dtrace_anon.dta_enabling = NULL;
14028 dtrace_anon.dta_state = NULL;
14029
14030 return (state);
14031}
14032
14033static void
14034dtrace_anon_property(void)
14035{
14036 int i, rv;
14037 dtrace_state_t *state;
14038 dof_hdr_t *dof;
14039 char c[32]; /* enough for "dof-data-" + digits */
14040
14041 ASSERT(MUTEX_HELD(&dtrace_lock));
14042 ASSERT(MUTEX_HELD(&cpu_lock));
14043
14044 for (i = 0; ; i++) {
14045 (void) snprintf(c, sizeof (c), "dof-data-%d", i);
14046
14047 dtrace_err_verbose = 1;
14048
14049 if ((dof = dtrace_dof_property(c)) == NULL) {
14050 dtrace_err_verbose = 0;
14051 break;
14052 }
14053
14054#if defined(sun)
14055 /*
14056 * We want to create anonymous state, so we need to transition
14057 * the kernel debugger to indicate that DTrace is active. If
14058 * this fails (e.g. because the debugger has modified text in
14059 * some way), we won't continue with the processing.
14060 */
14061 if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) {
14062 cmn_err(CE_NOTE, "kernel debugger active; anonymous "
14063 "enabling ignored.");
14064 dtrace_dof_destroy(dof);
14065 break;
14066 }
14067#endif
14068
14069 /*
14070 * If we haven't allocated an anonymous state, we'll do so now.
14071 */
14072 if ((state = dtrace_anon.dta_state) == NULL) {
14073#if defined(sun)
14074 state = dtrace_state_create(NULL, NULL);
14075#else
14076 state = dtrace_state_create(NULL);
14077#endif
14078 dtrace_anon.dta_state = state;
14079
14080 if (state == NULL) {
14081 /*
14082 * This basically shouldn't happen: the only
14083 * failure mode from dtrace_state_create() is a
14084 * failure of ddi_soft_state_zalloc() that
14085 * itself should never happen. Still, the
14086 * interface allows for a failure mode, and
14087 * we want to fail as gracefully as possible:
14088 * we'll emit an error message and cease
14089 * processing anonymous state in this case.
14090 */
14091 cmn_err(CE_WARN, "failed to create "
14092 "anonymous state");
14093 dtrace_dof_destroy(dof);
14094 break;
14095 }
14096 }
14097
14098 rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(),
14099 &dtrace_anon.dta_enabling, 0, B_TRUE);
14100
14101 if (rv == 0)
14102 rv = dtrace_dof_options(dof, state);
14103
14104 dtrace_err_verbose = 0;
14105 dtrace_dof_destroy(dof);
14106
14107 if (rv != 0) {
14108 /*
14109 * This is malformed DOF; chuck any anonymous state
14110 * that we created.
14111 */
14112 ASSERT(dtrace_anon.dta_enabling == NULL);
14113 dtrace_state_destroy(state);
14114 dtrace_anon.dta_state = NULL;
14115 break;
14116 }
14117
14118 ASSERT(dtrace_anon.dta_enabling != NULL);
14119 }
14120
14121 if (dtrace_anon.dta_enabling != NULL) {
14122 int rval;
14123
14124 /*
14125 * dtrace_enabling_retain() can only fail because we are
14126 * trying to retain more enablings than are allowed -- but
14127 * we only have one anonymous enabling, and we are guaranteed
14128 * to be allowed at least one retained enabling; we assert
14129 * that dtrace_enabling_retain() returns success.
14130 */
14131 rval = dtrace_enabling_retain(dtrace_anon.dta_enabling);
14132 ASSERT(rval == 0);
14133
14134 dtrace_enabling_dump(dtrace_anon.dta_enabling);
14135 }
14136}
14137
14138/*
14139 * DTrace Helper Functions
14140 */
14141static void
14142dtrace_helper_trace(dtrace_helper_action_t *helper,
14143 dtrace_mstate_t *mstate, dtrace_vstate_t *vstate, int where)
14144{
14145 uint32_t size, next, nnext, i;
14146 dtrace_helptrace_t *ent;
14147 uint16_t flags = cpu_core[curcpu].cpuc_dtrace_flags;
14148
14149 if (!dtrace_helptrace_enabled)
14150 return;
14151
14152 ASSERT(vstate->dtvs_nlocals <= dtrace_helptrace_nlocals);
14153
14154 /*
14155 * What would a tracing framework be without its own tracing
14156 * framework? (Well, a hell of a lot simpler, for starters...)
14157 */
14158 size = sizeof (dtrace_helptrace_t) + dtrace_helptrace_nlocals *
14159 sizeof (uint64_t) - sizeof (uint64_t);
14160
14161 /*
14162 * Iterate until we can allocate a slot in the trace buffer.
14163 */
14164 do {
14165 next = dtrace_helptrace_next;
14166
14167 if (next + size < dtrace_helptrace_bufsize) {
14168 nnext = next + size;
14169 } else {
14170 nnext = size;
14171 }
14172 } while (dtrace_cas32(&dtrace_helptrace_next, next, nnext) != next);
14173
14174 /*
14175 * We have our slot; fill it in.
14176 */
14177 if (nnext == size)
14178 next = 0;
14179
14180 ent = (dtrace_helptrace_t *)&dtrace_helptrace_buffer[next];
14181 ent->dtht_helper = helper;
14182 ent->dtht_where = where;
14183 ent->dtht_nlocals = vstate->dtvs_nlocals;
14184
14185 ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) ?
14186 mstate->dtms_fltoffs : -1;
14187 ent->dtht_fault = DTRACE_FLAGS2FLT(flags);
14188 ent->dtht_illval = cpu_core[curcpu].cpuc_dtrace_illval;
14189
14190 for (i = 0; i < vstate->dtvs_nlocals; i++) {
14191 dtrace_statvar_t *svar;
14192
14193 if ((svar = vstate->dtvs_locals[i]) == NULL)
14194 continue;
14195
14196 ASSERT(svar->dtsv_size >= NCPU * sizeof (uint64_t));
14197 ent->dtht_locals[i] =
14198 ((uint64_t *)(uintptr_t)svar->dtsv_data)[curcpu];
14199 }
14200}
14201
14202static uint64_t
14203dtrace_helper(int which, dtrace_mstate_t *mstate,
14204 dtrace_state_t *state, uint64_t arg0, uint64_t arg1)
14205{
14206 uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
14207 uint64_t sarg0 = mstate->dtms_arg[0];
14208 uint64_t sarg1 = mstate->dtms_arg[1];
14209 uint64_t rval = 0;
14210 dtrace_helpers_t *helpers = curproc->p_dtrace_helpers;
14211 dtrace_helper_action_t *helper;
14212 dtrace_vstate_t *vstate;
14213 dtrace_difo_t *pred;
14214 int i, trace = dtrace_helptrace_enabled;
14215
14216 ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS);
14217
14218 if (helpers == NULL)
14219 return (0);
14220
14221 if ((helper = helpers->dthps_actions[which]) == NULL)
14222 return (0);
14223
14224 vstate = &helpers->dthps_vstate;
14225 mstate->dtms_arg[0] = arg0;
14226 mstate->dtms_arg[1] = arg1;
14227
14228 /*
14229 * Now iterate over each helper. If its predicate evaluates to 'true',
14230 * we'll call the corresponding actions. Note that the below calls
14231 * to dtrace_dif_emulate() may set faults in machine state. This is
14232 * okay: our caller (the outer dtrace_dif_emulate()) will simply plow
14233 * the stored DIF offset with its own (which is the desired behavior).
14234 * Also, note the calls to dtrace_dif_emulate() may allocate scratch
14235 * from machine state; this is okay, too.
14236 */
14237 for (; helper != NULL; helper = helper->dtha_next) {
14238 if ((pred = helper->dtha_predicate) != NULL) {
14239 if (trace)
14240 dtrace_helper_trace(helper, mstate, vstate, 0);
14241
14242 if (!dtrace_dif_emulate(pred, mstate, vstate, state))
14243 goto next;
14244
14245 if (*flags & CPU_DTRACE_FAULT)
14246 goto err;
14247 }
14248
14249 for (i = 0; i < helper->dtha_nactions; i++) {
14250 if (trace)
14251 dtrace_helper_trace(helper,
14252 mstate, vstate, i + 1);
14253
14254 rval = dtrace_dif_emulate(helper->dtha_actions[i],
14255 mstate, vstate, state);
14256
14257 if (*flags & CPU_DTRACE_FAULT)
14258 goto err;
14259 }
14260
14261next:
14262 if (trace)
14263 dtrace_helper_trace(helper, mstate, vstate,
14264 DTRACE_HELPTRACE_NEXT);
14265 }
14266
14267 if (trace)
14268 dtrace_helper_trace(helper, mstate, vstate,
14269 DTRACE_HELPTRACE_DONE);
14270
14271 /*
14272 * Restore the arg0 that we saved upon entry.
14273 */
14274 mstate->dtms_arg[0] = sarg0;
14275 mstate->dtms_arg[1] = sarg1;
14276
14277 return (rval);
14278
14279err:
14280 if (trace)
14281 dtrace_helper_trace(helper, mstate, vstate,
14282 DTRACE_HELPTRACE_ERR);
14283
14284 /*
14285 * Restore the arg0 that we saved upon entry.
14286 */
14287 mstate->dtms_arg[0] = sarg0;
14288 mstate->dtms_arg[1] = sarg1;
14289
14290 return (0);
14291}
14292
14293static void
14294dtrace_helper_action_destroy(dtrace_helper_action_t *helper,
14295 dtrace_vstate_t *vstate)
14296{
14297 int i;
14298
14299 if (helper->dtha_predicate != NULL)
14300 dtrace_difo_release(helper->dtha_predicate, vstate);
14301
14302 for (i = 0; i < helper->dtha_nactions; i++) {
14303 ASSERT(helper->dtha_actions[i] != NULL);
14304 dtrace_difo_release(helper->dtha_actions[i], vstate);
14305 }
14306
14307 kmem_free(helper->dtha_actions,
14308 helper->dtha_nactions * sizeof (dtrace_difo_t *));
14309 kmem_free(helper, sizeof (dtrace_helper_action_t));
14310}
14311
14312static int
14313dtrace_helper_destroygen(int gen)
14314{
14315 proc_t *p = curproc;
14316 dtrace_helpers_t *help = p->p_dtrace_helpers;
14317 dtrace_vstate_t *vstate;
14318 int i;
14319
14320 ASSERT(MUTEX_HELD(&dtrace_lock));
14321
14322 if (help == NULL || gen > help->dthps_generation)
14323 return (EINVAL);
14324
14325 vstate = &help->dthps_vstate;
14326
14327 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
14328 dtrace_helper_action_t *last = NULL, *h, *next;
14329
14330 for (h = help->dthps_actions[i]; h != NULL; h = next) {
14331 next = h->dtha_next;
14332
14333 if (h->dtha_generation == gen) {
14334 if (last != NULL) {
14335 last->dtha_next = next;
14336 } else {
14337 help->dthps_actions[i] = next;
14338 }
14339
14340 dtrace_helper_action_destroy(h, vstate);
14341 } else {
14342 last = h;
14343 }
14344 }
14345 }
14346
14347 /*
14348 * Interate until we've cleared out all helper providers with the
14349 * given generation number.
14350 */
14351 for (;;) {
14352 dtrace_helper_provider_t *prov;
14353
14354 /*
14355 * Look for a helper provider with the right generation. We
14356 * have to start back at the beginning of the list each time
14357 * because we drop dtrace_lock. It's unlikely that we'll make
14358 * more than two passes.
14359 */
14360 for (i = 0; i < help->dthps_nprovs; i++) {
14361 prov = help->dthps_provs[i];
14362
14363 if (prov->dthp_generation == gen)
14364 break;
14365 }
14366
14367 /*
14368 * If there were no matches, we're done.
14369 */
14370 if (i == help->dthps_nprovs)
14371 break;
14372
14373 /*
14374 * Move the last helper provider into this slot.
14375 */
14376 help->dthps_nprovs--;
14377 help->dthps_provs[i] = help->dthps_provs[help->dthps_nprovs];
14378 help->dthps_provs[help->dthps_nprovs] = NULL;
14379
14380 mutex_exit(&dtrace_lock);
14381
14382 /*
14383 * If we have a meta provider, remove this helper provider.
14384 */
14385 mutex_enter(&dtrace_meta_lock);
14386 if (dtrace_meta_pid != NULL) {
14387 ASSERT(dtrace_deferred_pid == NULL);
14388 dtrace_helper_provider_remove(&prov->dthp_prov,
14389 p->p_pid);
14390 }
14391 mutex_exit(&dtrace_meta_lock);
14392
14393 dtrace_helper_provider_destroy(prov);
14394
14395 mutex_enter(&dtrace_lock);
14396 }
14397
14398 return (0);
14399}
14400
14401static int
14402dtrace_helper_validate(dtrace_helper_action_t *helper)
14403{
14404 int err = 0, i;
14405 dtrace_difo_t *dp;
14406
14407 if ((dp = helper->dtha_predicate) != NULL)
14408 err += dtrace_difo_validate_helper(dp);
14409
14410 for (i = 0; i < helper->dtha_nactions; i++)
14411 err += dtrace_difo_validate_helper(helper->dtha_actions[i]);
14412
14413 return (err == 0);
14414}
14415
14416static int
14417dtrace_helper_action_add(int which, dtrace_ecbdesc_t *ep)
14418{
14419 dtrace_helpers_t *help;
14420 dtrace_helper_action_t *helper, *last;
14421 dtrace_actdesc_t *act;
14422 dtrace_vstate_t *vstate;
14423 dtrace_predicate_t *pred;
14424 int count = 0, nactions = 0, i;
14425
14426 if (which < 0 || which >= DTRACE_NHELPER_ACTIONS)
14427 return (EINVAL);
14428
14429 help = curproc->p_dtrace_helpers;
14430 last = help->dthps_actions[which];
14431 vstate = &help->dthps_vstate;
14432
14433 for (count = 0; last != NULL; last = last->dtha_next) {
14434 count++;
14435 if (last->dtha_next == NULL)
14436 break;
14437 }
14438
14439 /*
14440 * If we already have dtrace_helper_actions_max helper actions for this
14441 * helper action type, we'll refuse to add a new one.
14442 */
14443 if (count >= dtrace_helper_actions_max)
14444 return (ENOSPC);
14445
14446 helper = kmem_zalloc(sizeof (dtrace_helper_action_t), KM_SLEEP);
14447 helper->dtha_generation = help->dthps_generation;
14448
14449 if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) {
14450 ASSERT(pred->dtp_difo != NULL);
14451 dtrace_difo_hold(pred->dtp_difo);
14452 helper->dtha_predicate = pred->dtp_difo;
14453 }
14454
14455 for (act = ep->dted_action; act != NULL; act = act->dtad_next) {
14456 if (act->dtad_kind != DTRACEACT_DIFEXPR)
14457 goto err;
14458
14459 if (act->dtad_difo == NULL)
14460 goto err;
14461
14462 nactions++;
14463 }
14464
14465 helper->dtha_actions = kmem_zalloc(sizeof (dtrace_difo_t *) *
14466 (helper->dtha_nactions = nactions), KM_SLEEP);
14467
14468 for (act = ep->dted_action, i = 0; act != NULL; act = act->dtad_next) {
14469 dtrace_difo_hold(act->dtad_difo);
14470 helper->dtha_actions[i++] = act->dtad_difo;
14471 }
14472
14473 if (!dtrace_helper_validate(helper))
14474 goto err;
14475
14476 if (last == NULL) {
14477 help->dthps_actions[which] = helper;
14478 } else {
14479 last->dtha_next = helper;
14480 }
14481
14482 if (vstate->dtvs_nlocals > dtrace_helptrace_nlocals) {
14483 dtrace_helptrace_nlocals = vstate->dtvs_nlocals;
14484 dtrace_helptrace_next = 0;
14485 }
14486
14487 return (0);
14488err:
14489 dtrace_helper_action_destroy(helper, vstate);
14490 return (EINVAL);
14491}
14492
14493static void
14494dtrace_helper_provider_register(proc_t *p, dtrace_helpers_t *help,
14495 dof_helper_t *dofhp)
14496{
14497 ASSERT(MUTEX_NOT_HELD(&dtrace_lock));
14498
14499 mutex_enter(&dtrace_meta_lock);
14500 mutex_enter(&dtrace_lock);
14501
14502 if (!dtrace_attached() || dtrace_meta_pid == NULL) {
14503 /*
14504 * If the dtrace module is loaded but not attached, or if
14505 * there aren't isn't a meta provider registered to deal with
14506 * these provider descriptions, we need to postpone creating
14507 * the actual providers until later.
14508 */
14509
14510 if (help->dthps_next == NULL && help->dthps_prev == NULL &&
14511 dtrace_deferred_pid != help) {
14512 help->dthps_deferred = 1;
14513 help->dthps_pid = p->p_pid;
14514 help->dthps_next = dtrace_deferred_pid;
14515 help->dthps_prev = NULL;
14516 if (dtrace_deferred_pid != NULL)
14517 dtrace_deferred_pid->dthps_prev = help;
14518 dtrace_deferred_pid = help;
14519 }
14520
14521 mutex_exit(&dtrace_lock);
14522
14523 } else if (dofhp != NULL) {
14524 /*
14525 * If the dtrace module is loaded and we have a particular
14526 * helper provider description, pass that off to the
14527 * meta provider.
14528 */
14529
14530 mutex_exit(&dtrace_lock);
14531
14532 dtrace_helper_provide(dofhp, p->p_pid);
14533
14534 } else {
14535 /*
14536 * Otherwise, just pass all the helper provider descriptions
14537 * off to the meta provider.
14538 */
14539
14540 int i;
14541 mutex_exit(&dtrace_lock);
14542
14543 for (i = 0; i < help->dthps_nprovs; i++) {
14544 dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov,
14545 p->p_pid);
14546 }
14547 }
14548
14549 mutex_exit(&dtrace_meta_lock);
14550}
14551
14552static int
14553dtrace_helper_provider_add(dof_helper_t *dofhp, int gen)
14554{
14555 dtrace_helpers_t *help;
14556 dtrace_helper_provider_t *hprov, **tmp_provs;
14557 uint_t tmp_maxprovs, i;
14558
14559 ASSERT(MUTEX_HELD(&dtrace_lock));
14560
14561 help = curproc->p_dtrace_helpers;
14562 ASSERT(help != NULL);
14563
14564 /*
14565 * If we already have dtrace_helper_providers_max helper providers,
14566 * we're refuse to add a new one.
14567 */
14568 if (help->dthps_nprovs >= dtrace_helper_providers_max)
14569 return (ENOSPC);
14570
14571 /*
14572 * Check to make sure this isn't a duplicate.
14573 */
14574 for (i = 0; i < help->dthps_nprovs; i++) {
14575 if (dofhp->dofhp_dof ==
14576 help->dthps_provs[i]->dthp_prov.dofhp_dof)
14577 return (EALREADY);
14578 }
14579
14580 hprov = kmem_zalloc(sizeof (dtrace_helper_provider_t), KM_SLEEP);
14581 hprov->dthp_prov = *dofhp;
14582 hprov->dthp_ref = 1;
14583 hprov->dthp_generation = gen;
14584
14585 /*
14586 * Allocate a bigger table for helper providers if it's already full.
14587 */
14588 if (help->dthps_maxprovs == help->dthps_nprovs) {
14589 tmp_maxprovs = help->dthps_maxprovs;
14590 tmp_provs = help->dthps_provs;
14591
14592 if (help->dthps_maxprovs == 0)
14593 help->dthps_maxprovs = 2;
14594 else
14595 help->dthps_maxprovs *= 2;
14596 if (help->dthps_maxprovs > dtrace_helper_providers_max)
14597 help->dthps_maxprovs = dtrace_helper_providers_max;
14598
14599 ASSERT(tmp_maxprovs < help->dthps_maxprovs);
14600
14601 help->dthps_provs = kmem_zalloc(help->dthps_maxprovs *
14602 sizeof (dtrace_helper_provider_t *), KM_SLEEP);
14603
14604 if (tmp_provs != NULL) {
14605 bcopy(tmp_provs, help->dthps_provs, tmp_maxprovs *
14606 sizeof (dtrace_helper_provider_t *));
14607 kmem_free(tmp_provs, tmp_maxprovs *
14608 sizeof (dtrace_helper_provider_t *));
14609 }
14610 }
14611
14612 help->dthps_provs[help->dthps_nprovs] = hprov;
14613 help->dthps_nprovs++;
14614
14615 return (0);
14616}
14617
14618static void
14619dtrace_helper_provider_destroy(dtrace_helper_provider_t *hprov)
14620{
14621 mutex_enter(&dtrace_lock);
14622
14623 if (--hprov->dthp_ref == 0) {
14624 dof_hdr_t *dof;
14625 mutex_exit(&dtrace_lock);
14626 dof = (dof_hdr_t *)(uintptr_t)hprov->dthp_prov.dofhp_dof;
14627 dtrace_dof_destroy(dof);
14628 kmem_free(hprov, sizeof (dtrace_helper_provider_t));
14629 } else {
14630 mutex_exit(&dtrace_lock);
14631 }
14632}
14633
14634static int
14635dtrace_helper_provider_validate(dof_hdr_t *dof, dof_sec_t *sec)
14636{
14637 uintptr_t daddr = (uintptr_t)dof;
14638 dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec;
14639 dof_provider_t *provider;
14640 dof_probe_t *probe;
14641 uint8_t *arg;
14642 char *strtab, *typestr;
14643 dof_stridx_t typeidx;
14644 size_t typesz;
14645 uint_t nprobes, j, k;
14646
14647 ASSERT(sec->dofs_type == DOF_SECT_PROVIDER);
14648
14649 if (sec->dofs_offset & (sizeof (uint_t) - 1)) {
14650 dtrace_dof_error(dof, "misaligned section offset");
14651 return (-1);
14652 }
14653
14654 /*
14655 * The section needs to be large enough to contain the DOF provider
14656 * structure appropriate for the given version.
14657 */
14658 if (sec->dofs_size <
14659 ((dof->dofh_ident[DOF_ID_VERSION] == DOF_VERSION_1) ?
14660 offsetof(dof_provider_t, dofpv_prenoffs) :
14661 sizeof (dof_provider_t))) {
14662 dtrace_dof_error(dof, "provider section too small");
14663 return (-1);
14664 }
14665
14666 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
14667 str_sec = dtrace_dof_sect(dof, DOF_SECT_STRTAB, provider->dofpv_strtab);
14668 prb_sec = dtrace_dof_sect(dof, DOF_SECT_PROBES, provider->dofpv_probes);
14669 arg_sec = dtrace_dof_sect(dof, DOF_SECT_PRARGS, provider->dofpv_prargs);
14670 off_sec = dtrace_dof_sect(dof, DOF_SECT_PROFFS, provider->dofpv_proffs);
14671
14672 if (str_sec == NULL || prb_sec == NULL ||
14673 arg_sec == NULL || off_sec == NULL)
14674 return (-1);
14675
14676 enoff_sec = NULL;
14677
14678 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
14679 provider->dofpv_prenoffs != DOF_SECT_NONE &&
14680 (enoff_sec = dtrace_dof_sect(dof, DOF_SECT_PRENOFFS,
14681 provider->dofpv_prenoffs)) == NULL)
14682 return (-1);
14683
14684 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
14685
14686 if (provider->dofpv_name >= str_sec->dofs_size ||
14687 strlen(strtab + provider->dofpv_name) >= DTRACE_PROVNAMELEN) {
14688 dtrace_dof_error(dof, "invalid provider name");
14689 return (-1);
14690 }
14691
14692 if (prb_sec->dofs_entsize == 0 ||
14693 prb_sec->dofs_entsize > prb_sec->dofs_size) {
14694 dtrace_dof_error(dof, "invalid entry size");
14695 return (-1);
14696 }
14697
14698 if (prb_sec->dofs_entsize & (sizeof (uintptr_t) - 1)) {
14699 dtrace_dof_error(dof, "misaligned entry size");
14700 return (-1);
14701 }
14702
14703 if (off_sec->dofs_entsize != sizeof (uint32_t)) {
14704 dtrace_dof_error(dof, "invalid entry size");
14705 return (-1);
14706 }
14707
14708 if (off_sec->dofs_offset & (sizeof (uint32_t) - 1)) {
14709 dtrace_dof_error(dof, "misaligned section offset");
14710 return (-1);
14711 }
14712
14713 if (arg_sec->dofs_entsize != sizeof (uint8_t)) {
14714 dtrace_dof_error(dof, "invalid entry size");
14715 return (-1);
14716 }
14717
14718 arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset);
14719
14720 nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize;
14721
14722 /*
14723 * Take a pass through the probes to check for errors.
14724 */
14725 for (j = 0; j < nprobes; j++) {
14726 probe = (dof_probe_t *)(uintptr_t)(daddr +
14727 prb_sec->dofs_offset + j * prb_sec->dofs_entsize);
14728
14729 if (probe->dofpr_func >= str_sec->dofs_size) {
14730 dtrace_dof_error(dof, "invalid function name");
14731 return (-1);
14732 }
14733
14734 if (strlen(strtab + probe->dofpr_func) >= DTRACE_FUNCNAMELEN) {
14735 dtrace_dof_error(dof, "function name too long");
14736 return (-1);
14737 }
14738
14739 if (probe->dofpr_name >= str_sec->dofs_size ||
14740 strlen(strtab + probe->dofpr_name) >= DTRACE_NAMELEN) {
14741 dtrace_dof_error(dof, "invalid probe name");
14742 return (-1);
14743 }
14744
14745 /*
14746 * The offset count must not wrap the index, and the offsets
14747 * must also not overflow the section's data.
14748 */
14749 if (probe->dofpr_offidx + probe->dofpr_noffs <
14750 probe->dofpr_offidx ||
14751 (probe->dofpr_offidx + probe->dofpr_noffs) *
14752 off_sec->dofs_entsize > off_sec->dofs_size) {
14753 dtrace_dof_error(dof, "invalid probe offset");
14754 return (-1);
14755 }
14756
14757 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1) {
14758 /*
14759 * If there's no is-enabled offset section, make sure
14760 * there aren't any is-enabled offsets. Otherwise
14761 * perform the same checks as for probe offsets
14762 * (immediately above).
14763 */
14764 if (enoff_sec == NULL) {
14765 if (probe->dofpr_enoffidx != 0 ||
14766 probe->dofpr_nenoffs != 0) {
14767 dtrace_dof_error(dof, "is-enabled "
14768 "offsets with null section");
14769 return (-1);
14770 }
14771 } else if (probe->dofpr_enoffidx +
14772 probe->dofpr_nenoffs < probe->dofpr_enoffidx ||
14773 (probe->dofpr_enoffidx + probe->dofpr_nenoffs) *
14774 enoff_sec->dofs_entsize > enoff_sec->dofs_size) {
14775 dtrace_dof_error(dof, "invalid is-enabled "
14776 "offset");
14777 return (-1);
14778 }
14779
14780 if (probe->dofpr_noffs + probe->dofpr_nenoffs == 0) {
14781 dtrace_dof_error(dof, "zero probe and "
14782 "is-enabled offsets");
14783 return (-1);
14784 }
14785 } else if (probe->dofpr_noffs == 0) {
14786 dtrace_dof_error(dof, "zero probe offsets");
14787 return (-1);
14788 }
14789
14790 if (probe->dofpr_argidx + probe->dofpr_xargc <
14791 probe->dofpr_argidx ||
14792 (probe->dofpr_argidx + probe->dofpr_xargc) *
14793 arg_sec->dofs_entsize > arg_sec->dofs_size) {
14794 dtrace_dof_error(dof, "invalid args");
14795 return (-1);
14796 }
14797
14798 typeidx = probe->dofpr_nargv;
14799 typestr = strtab + probe->dofpr_nargv;
14800 for (k = 0; k < probe->dofpr_nargc; k++) {
14801 if (typeidx >= str_sec->dofs_size) {
14802 dtrace_dof_error(dof, "bad "
14803 "native argument type");
14804 return (-1);
14805 }
14806
14807 typesz = strlen(typestr) + 1;
14808 if (typesz > DTRACE_ARGTYPELEN) {
14809 dtrace_dof_error(dof, "native "
14810 "argument type too long");
14811 return (-1);
14812 }
14813 typeidx += typesz;
14814 typestr += typesz;
14815 }
14816
14817 typeidx = probe->dofpr_xargv;
14818 typestr = strtab + probe->dofpr_xargv;
14819 for (k = 0; k < probe->dofpr_xargc; k++) {
14820 if (arg[probe->dofpr_argidx + k] > probe->dofpr_nargc) {
14821 dtrace_dof_error(dof, "bad "
14822 "native argument index");
14823 return (-1);
14824 }
14825
14826 if (typeidx >= str_sec->dofs_size) {
14827 dtrace_dof_error(dof, "bad "
14828 "translated argument type");
14829 return (-1);
14830 }
14831
14832 typesz = strlen(typestr) + 1;
14833 if (typesz > DTRACE_ARGTYPELEN) {
14834 dtrace_dof_error(dof, "translated argument "
14835 "type too long");
14836 return (-1);
14837 }
14838
14839 typeidx += typesz;
14840 typestr += typesz;
14841 }
14842 }
14843
14844 return (0);
14845}
14846
14847static int
14848dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp)
14849{
14850 dtrace_helpers_t *help;
14851 dtrace_vstate_t *vstate;
14852 dtrace_enabling_t *enab = NULL;
14853 int i, gen, rv, nhelpers = 0, nprovs = 0, destroy = 1;
14854 uintptr_t daddr = (uintptr_t)dof;
14855
14856 ASSERT(MUTEX_HELD(&dtrace_lock));
14857
14858 if ((help = curproc->p_dtrace_helpers) == NULL)
14859 help = dtrace_helpers_create(curproc);
14860
14861 vstate = &help->dthps_vstate;
14862
14863 if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab,
14864 dhp != NULL ? dhp->dofhp_addr : 0, B_FALSE)) != 0) {
14865 dtrace_dof_destroy(dof);
14866 return (rv);
14867 }
14868
14869 /*
14870 * Look for helper providers and validate their descriptions.
14871 */
14872 if (dhp != NULL) {
14873 for (i = 0; i < dof->dofh_secnum; i++) {
14874 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
14875 dof->dofh_secoff + i * dof->dofh_secsize);
14876
14877 if (sec->dofs_type != DOF_SECT_PROVIDER)
14878 continue;
14879
14880 if (dtrace_helper_provider_validate(dof, sec) != 0) {
14881 dtrace_enabling_destroy(enab);
14882 dtrace_dof_destroy(dof);
14883 return (-1);
14884 }
14885
14886 nprovs++;
14887 }
14888 }
14889
14890 /*
14891 * Now we need to walk through the ECB descriptions in the enabling.
14892 */
14893 for (i = 0; i < enab->dten_ndesc; i++) {
14894 dtrace_ecbdesc_t *ep = enab->dten_desc[i];
14895 dtrace_probedesc_t *desc = &ep->dted_probe;
14896
14897 if (strcmp(desc->dtpd_provider, "dtrace") != 0)
14898 continue;
14899
14900 if (strcmp(desc->dtpd_mod, "helper") != 0)
14901 continue;
14902
14903 if (strcmp(desc->dtpd_func, "ustack") != 0)
14904 continue;
14905
14906 if ((rv = dtrace_helper_action_add(DTRACE_HELPER_ACTION_USTACK,
14907 ep)) != 0) {
14908 /*
14909 * Adding this helper action failed -- we are now going
14910 * to rip out the entire generation and return failure.
14911 */
14912 (void) dtrace_helper_destroygen(help->dthps_generation);
14913 dtrace_enabling_destroy(enab);
14914 dtrace_dof_destroy(dof);
14915 return (-1);
14916 }
14917
14918 nhelpers++;
14919 }
14920
14921 if (nhelpers < enab->dten_ndesc)
14922 dtrace_dof_error(dof, "unmatched helpers");
14923
14924 gen = help->dthps_generation++;
14925 dtrace_enabling_destroy(enab);
14926
14927 if (dhp != NULL && nprovs > 0) {
14928 dhp->dofhp_dof = (uint64_t)(uintptr_t)dof;
14929 if (dtrace_helper_provider_add(dhp, gen) == 0) {
14930 mutex_exit(&dtrace_lock);
14931 dtrace_helper_provider_register(curproc, help, dhp);
14932 mutex_enter(&dtrace_lock);
14933
14934 destroy = 0;
14935 }
14936 }
14937
14938 if (destroy)
14939 dtrace_dof_destroy(dof);
14940
14941 return (gen);
14942}
14943
14944static dtrace_helpers_t *
14945dtrace_helpers_create(proc_t *p)
14946{
14947 dtrace_helpers_t *help;
14948
14949 ASSERT(MUTEX_HELD(&dtrace_lock));
14950 ASSERT(p->p_dtrace_helpers == NULL);
14951
14952 help = kmem_zalloc(sizeof (dtrace_helpers_t), KM_SLEEP);
14953 help->dthps_actions = kmem_zalloc(sizeof (dtrace_helper_action_t *) *
14954 DTRACE_NHELPER_ACTIONS, KM_SLEEP);
14955
14956 p->p_dtrace_helpers = help;
14957 dtrace_helpers++;
14958
14959 return (help);
14960}
14961
14962#if defined(sun)
14963static
14964#endif
14965void
14966dtrace_helpers_destroy(proc_t *p)
14967{
14968 dtrace_helpers_t *help;
14969 dtrace_vstate_t *vstate;
14970#if defined(sun)
14971 proc_t *p = curproc;
14972#endif
14973 int i;
14974
14975 mutex_enter(&dtrace_lock);
14976
14977 ASSERT(p->p_dtrace_helpers != NULL);
14978 ASSERT(dtrace_helpers > 0);
14979
14980 help = p->p_dtrace_helpers;
14981 vstate = &help->dthps_vstate;
14982
14983 /*
14984 * We're now going to lose the help from this process.
14985 */
14986 p->p_dtrace_helpers = NULL;
14987 dtrace_sync();
14988
14989 /*
14990 * Destory the helper actions.
14991 */
14992 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
14993 dtrace_helper_action_t *h, *next;
14994
14995 for (h = help->dthps_actions[i]; h != NULL; h = next) {
14996 next = h->dtha_next;
14997 dtrace_helper_action_destroy(h, vstate);
14998 h = next;
14999 }
15000 }
15001
15002 mutex_exit(&dtrace_lock);
15003
15004 /*
15005 * Destroy the helper providers.
15006 */
15007 if (help->dthps_maxprovs > 0) {
15008 mutex_enter(&dtrace_meta_lock);
15009 if (dtrace_meta_pid != NULL) {
15010 ASSERT(dtrace_deferred_pid == NULL);
15011
15012 for (i = 0; i < help->dthps_nprovs; i++) {
15013 dtrace_helper_provider_remove(
15014 &help->dthps_provs[i]->dthp_prov, p->p_pid);
15015 }
15016 } else {
15017 mutex_enter(&dtrace_lock);
15018 ASSERT(help->dthps_deferred == 0 ||
15019 help->dthps_next != NULL ||
15020 help->dthps_prev != NULL ||
15021 help == dtrace_deferred_pid);
15022
15023 /*
15024 * Remove the helper from the deferred list.
15025 */
15026 if (help->dthps_next != NULL)
15027 help->dthps_next->dthps_prev = help->dthps_prev;
15028 if (help->dthps_prev != NULL)
15029 help->dthps_prev->dthps_next = help->dthps_next;
15030 if (dtrace_deferred_pid == help) {
15031 dtrace_deferred_pid = help->dthps_next;
15032 ASSERT(help->dthps_prev == NULL);
15033 }
15034
15035 mutex_exit(&dtrace_lock);
15036 }
15037
15038 mutex_exit(&dtrace_meta_lock);
15039
15040 for (i = 0; i < help->dthps_nprovs; i++) {
15041 dtrace_helper_provider_destroy(help->dthps_provs[i]);
15042 }
15043
15044 kmem_free(help->dthps_provs, help->dthps_maxprovs *
15045 sizeof (dtrace_helper_provider_t *));
15046 }
15047
15048 mutex_enter(&dtrace_lock);
15049
15050 dtrace_vstate_fini(&help->dthps_vstate);
15051 kmem_free(help->dthps_actions,
15052 sizeof (dtrace_helper_action_t *) * DTRACE_NHELPER_ACTIONS);
15053 kmem_free(help, sizeof (dtrace_helpers_t));
15054
15055 --dtrace_helpers;
15056 mutex_exit(&dtrace_lock);
15057}
15058
15059#if defined(sun)
15060static
15061#endif
15062void
15063dtrace_helpers_duplicate(proc_t *from, proc_t *to)
15064{
15065 dtrace_helpers_t *help, *newhelp;
15066 dtrace_helper_action_t *helper, *new, *last;
15067 dtrace_difo_t *dp;
15068 dtrace_vstate_t *vstate;
15069 int i, j, sz, hasprovs = 0;
15070
15071 mutex_enter(&dtrace_lock);
15072 ASSERT(from->p_dtrace_helpers != NULL);
15073 ASSERT(dtrace_helpers > 0);
15074
15075 help = from->p_dtrace_helpers;
15076 newhelp = dtrace_helpers_create(to);
15077 ASSERT(to->p_dtrace_helpers != NULL);
15078
15079 newhelp->dthps_generation = help->dthps_generation;
15080 vstate = &newhelp->dthps_vstate;
15081
15082 /*
15083 * Duplicate the helper actions.
15084 */
15085 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
15086 if ((helper = help->dthps_actions[i]) == NULL)
15087 continue;
15088
15089 for (last = NULL; helper != NULL; helper = helper->dtha_next) {
15090 new = kmem_zalloc(sizeof (dtrace_helper_action_t),
15091 KM_SLEEP);
15092 new->dtha_generation = helper->dtha_generation;
15093
15094 if ((dp = helper->dtha_predicate) != NULL) {
15095 dp = dtrace_difo_duplicate(dp, vstate);
15096 new->dtha_predicate = dp;
15097 }
15098
15099 new->dtha_nactions = helper->dtha_nactions;
15100 sz = sizeof (dtrace_difo_t *) * new->dtha_nactions;
15101 new->dtha_actions = kmem_alloc(sz, KM_SLEEP);
15102
15103 for (j = 0; j < new->dtha_nactions; j++) {
15104 dtrace_difo_t *dp = helper->dtha_actions[j];
15105
15106 ASSERT(dp != NULL);
15107 dp = dtrace_difo_duplicate(dp, vstate);
15108 new->dtha_actions[j] = dp;
15109 }
15110
15111 if (last != NULL) {
15112 last->dtha_next = new;
15113 } else {
15114 newhelp->dthps_actions[i] = new;
15115 }
15116
15117 last = new;
15118 }
15119 }
15120
15121 /*
15122 * Duplicate the helper providers and register them with the
15123 * DTrace framework.
15124 */
15125 if (help->dthps_nprovs > 0) {
15126 newhelp->dthps_nprovs = help->dthps_nprovs;
15127 newhelp->dthps_maxprovs = help->dthps_nprovs;
15128 newhelp->dthps_provs = kmem_alloc(newhelp->dthps_nprovs *
15129 sizeof (dtrace_helper_provider_t *), KM_SLEEP);
15130 for (i = 0; i < newhelp->dthps_nprovs; i++) {
15131 newhelp->dthps_provs[i] = help->dthps_provs[i];
15132 newhelp->dthps_provs[i]->dthp_ref++;
15133 }
15134
15135 hasprovs = 1;
15136 }
15137
15138 mutex_exit(&dtrace_lock);
15139
15140 if (hasprovs)
15141 dtrace_helper_provider_register(to, newhelp, NULL);
15142}
15143
15144/*
15145 * DTrace Hook Functions
15146 */
15147static void
15148dtrace_module_loaded(modctl_t *ctl)
15149{
15150 dtrace_provider_t *prv;
15151
15152 mutex_enter(&dtrace_provider_lock);
15153#if defined(sun)
15154 mutex_enter(&mod_lock);
15155#endif
15156
15157#if defined(sun)
15158 ASSERT(ctl->mod_busy);
15159#endif
15160
15161 /*
15162 * We're going to call each providers per-module provide operation
15163 * specifying only this module.
15164 */
15165 for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next)
15166 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl);
15167
15168#if defined(sun)
15169 mutex_exit(&mod_lock);
15170#endif
15171 mutex_exit(&dtrace_provider_lock);
15172
15173 /*
15174 * If we have any retained enablings, we need to match against them.
15175 * Enabling probes requires that cpu_lock be held, and we cannot hold
15176 * cpu_lock here -- it is legal for cpu_lock to be held when loading a
15177 * module. (In particular, this happens when loading scheduling
15178 * classes.) So if we have any retained enablings, we need to dispatch
15179 * our task queue to do the match for us.
15180 */
15181 mutex_enter(&dtrace_lock);
15182
15183 if (dtrace_retained == NULL) {
15184 mutex_exit(&dtrace_lock);
15185 return;
15186 }
15187
15188 (void) taskq_dispatch(dtrace_taskq,
15189 (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP);
15190
15191 mutex_exit(&dtrace_lock);
15192
15193 /*
15194 * And now, for a little heuristic sleaze: in general, we want to
15195 * match modules as soon as they load. However, we cannot guarantee
15196 * this, because it would lead us to the lock ordering violation
15197 * outlined above. The common case, of course, is that cpu_lock is
15198 * _not_ held -- so we delay here for a clock tick, hoping that that's
15199 * long enough for the task queue to do its work. If it's not, it's
15200 * not a serious problem -- it just means that the module that we
15201 * just loaded may not be immediately instrumentable.
15202 */
15203 delay(1);
15204}
15205
15206static void
15207#if defined(sun)
15208dtrace_module_unloaded(modctl_t *ctl)
15209#else
15210dtrace_module_unloaded(modctl_t *ctl, int *error)
15211#endif
15212{
15213 dtrace_probe_t template, *probe, *first, *next;
15214 dtrace_provider_t *prov;
15215#if !defined(sun)
15216 char modname[DTRACE_MODNAMELEN];
15217 size_t len;
15218#endif
15219
15220#if defined(sun)
15221 template.dtpr_mod = ctl->mod_modname;
15222#else
15223 /* Handle the fact that ctl->filename may end in ".ko". */
15224 strlcpy(modname, ctl->filename, sizeof(modname));
15225 len = strlen(ctl->filename);
15226 if (len > 3 && strcmp(modname + len - 3, ".ko") == 0)
15227 modname[len - 3] = '\0';
15228 template.dtpr_mod = modname;
15229#endif
15230
15231 mutex_enter(&dtrace_provider_lock);
15232#if defined(sun)
15233 mutex_enter(&mod_lock);
15234#endif
15235 mutex_enter(&dtrace_lock);
15236
15237#if !defined(sun)
15238 if (ctl->nenabled > 0) {
15239 /* Don't allow unloads if a probe is enabled. */
15240 mutex_exit(&dtrace_provider_lock);
15241 mutex_exit(&dtrace_lock);
15242 *error = -1;
15243 printf(
15244 "kldunload: attempt to unload module that has DTrace probes enabled\n");
15245 return;
15246 }
15247#endif
15248
15249 if (dtrace_bymod == NULL) {
15250 /*
15251 * The DTrace module is loaded (obviously) but not attached;
15252 * we don't have any work to do.
15253 */
15254 mutex_exit(&dtrace_provider_lock);
15255#if defined(sun)
15256 mutex_exit(&mod_lock);
15257#endif
15258 mutex_exit(&dtrace_lock);
15259 return;
15260 }
15261
15262 for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template);
15263 probe != NULL; probe = probe->dtpr_nextmod) {
15264 if (probe->dtpr_ecb != NULL) {
15265 mutex_exit(&dtrace_provider_lock);
15266#if defined(sun)
15267 mutex_exit(&mod_lock);
15268#endif
15269 mutex_exit(&dtrace_lock);
15270
15271 /*
15272 * This shouldn't _actually_ be possible -- we're
15273 * unloading a module that has an enabled probe in it.
15274 * (It's normally up to the provider to make sure that
15275 * this can't happen.) However, because dtps_enable()
15276 * doesn't have a failure mode, there can be an
15277 * enable/unload race. Upshot: we don't want to
15278 * assert, but we're not going to disable the
15279 * probe, either.
15280 */
15281 if (dtrace_err_verbose) {
15282#if defined(sun)
15283 cmn_err(CE_WARN, "unloaded module '%s' had "
15284 "enabled probes", ctl->mod_modname);
15285#else
15286 cmn_err(CE_WARN, "unloaded module '%s' had "
15287 "enabled probes", modname);
15288#endif
15289 }
15290
15291 return;
15292 }
15293 }
15294
15295 probe = first;
15296
15297 for (first = NULL; probe != NULL; probe = next) {
15298 ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe);
15299
15300 dtrace_probes[probe->dtpr_id - 1] = NULL;
15301
15302 next = probe->dtpr_nextmod;
15303 dtrace_hash_remove(dtrace_bymod, probe);
15304 dtrace_hash_remove(dtrace_byfunc, probe);
15305 dtrace_hash_remove(dtrace_byname, probe);
15306
15307 if (first == NULL) {
15308 first = probe;
15309 probe->dtpr_nextmod = NULL;
15310 } else {
15311 probe->dtpr_nextmod = first;
15312 first = probe;
15313 }
15314 }
15315
15316 /*
15317 * We've removed all of the module's probes from the hash chains and
15318 * from the probe array. Now issue a dtrace_sync() to be sure that
15319 * everyone has cleared out from any probe array processing.
15320 */
15321 dtrace_sync();
15322
15323 for (probe = first; probe != NULL; probe = first) {
15324 first = probe->dtpr_nextmod;
15325 prov = probe->dtpr_provider;
15326 prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id,
15327 probe->dtpr_arg);
15328 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
15329 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
15330 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
15331#if defined(sun)
15332 vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1);
15333#else
15334 free_unr(dtrace_arena, probe->dtpr_id);
15335#endif
15336 kmem_free(probe, sizeof (dtrace_probe_t));
15337 }
15338
15339 mutex_exit(&dtrace_lock);
15340#if defined(sun)
15341 mutex_exit(&mod_lock);
15342#endif
15343 mutex_exit(&dtrace_provider_lock);
15344}
15345
15346#if !defined(sun)
15347static void
15348dtrace_kld_load(void *arg __unused, linker_file_t lf)
15349{
15350
15351 dtrace_module_loaded(lf);
15352}
15353
15354static void
15355dtrace_kld_unload_try(void *arg __unused, linker_file_t lf, int *error)
15356{
15357
15358 if (*error != 0)
15359 /* We already have an error, so don't do anything. */
15360 return;
15361 dtrace_module_unloaded(lf, error);
15362}
15363#endif
15364
15365#if defined(sun)
15366static void
15367dtrace_suspend(void)
15368{
15369 dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_suspend));
15370}
15371
15372static void
15373dtrace_resume(void)
15374{
15375 dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_resume));
15376}
15377#endif
15378
15379static int
15380dtrace_cpu_setup(cpu_setup_t what, processorid_t cpu)
15381{
15382 ASSERT(MUTEX_HELD(&cpu_lock));
15383 mutex_enter(&dtrace_lock);
15384
15385 switch (what) {
15386 case CPU_CONFIG: {
15387 dtrace_state_t *state;
15388 dtrace_optval_t *opt, rs, c;
15389
15390 /*
15391 * For now, we only allocate a new buffer for anonymous state.
15392 */
15393 if ((state = dtrace_anon.dta_state) == NULL)
15394 break;
15395
15396 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE)
15397 break;
15398
15399 opt = state->dts_options;
15400 c = opt[DTRACEOPT_CPU];
15401
15402 if (c != DTRACE_CPUALL && c != DTRACEOPT_UNSET && c != cpu)
15403 break;
15404
15405 /*
15406 * Regardless of what the actual policy is, we're going to
15407 * temporarily set our resize policy to be manual. We're
15408 * also going to temporarily set our CPU option to denote
15409 * the newly configured CPU.
15410 */
15411 rs = opt[DTRACEOPT_BUFRESIZE];
15412 opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_MANUAL;
15413 opt[DTRACEOPT_CPU] = (dtrace_optval_t)cpu;
15414
15415 (void) dtrace_state_buffers(state);
15416
15417 opt[DTRACEOPT_BUFRESIZE] = rs;
15418 opt[DTRACEOPT_CPU] = c;
15419
15420 break;
15421 }
15422
15423 case CPU_UNCONFIG:
15424 /*
15425 * We don't free the buffer in the CPU_UNCONFIG case. (The
15426 * buffer will be freed when the consumer exits.)
15427 */
15428 break;
15429
15430 default:
15431 break;
15432 }
15433
15434 mutex_exit(&dtrace_lock);
15435 return (0);
15436}
15437
15438#if defined(sun)
15439static void
15440dtrace_cpu_setup_initial(processorid_t cpu)
15441{
15442 (void) dtrace_cpu_setup(CPU_CONFIG, cpu);
15443}
15444#endif
15445
15446static void
15447dtrace_toxrange_add(uintptr_t base, uintptr_t limit)
15448{
15449 if (dtrace_toxranges >= dtrace_toxranges_max) {
15450 int osize, nsize;
15451 dtrace_toxrange_t *range;
15452
15453 osize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t);
15454
15455 if (osize == 0) {
15456 ASSERT(dtrace_toxrange == NULL);
15457 ASSERT(dtrace_toxranges_max == 0);
15458 dtrace_toxranges_max = 1;
15459 } else {
15460 dtrace_toxranges_max <<= 1;
15461 }
15462
15463 nsize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t);
15464 range = kmem_zalloc(nsize, KM_SLEEP);
15465
15466 if (dtrace_toxrange != NULL) {
15467 ASSERT(osize != 0);
15468 bcopy(dtrace_toxrange, range, osize);
15469 kmem_free(dtrace_toxrange, osize);
15470 }
15471
15472 dtrace_toxrange = range;
15473 }
15474
15475 ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == 0);
15476 ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == 0);
15477
15478 dtrace_toxrange[dtrace_toxranges].dtt_base = base;
15479 dtrace_toxrange[dtrace_toxranges].dtt_limit = limit;
15480 dtrace_toxranges++;
15481}
15482
15483/*
15484 * DTrace Driver Cookbook Functions
15485 */
15486#if defined(sun)
15487/*ARGSUSED*/
15488static int
15489dtrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
15490{
15491 dtrace_provider_id_t id;
15492 dtrace_state_t *state = NULL;
15493 dtrace_enabling_t *enab;
15494
15495 mutex_enter(&cpu_lock);
15496 mutex_enter(&dtrace_provider_lock);
15497 mutex_enter(&dtrace_lock);
15498
15499 if (ddi_soft_state_init(&dtrace_softstate,
15500 sizeof (dtrace_state_t), 0) != 0) {
15501 cmn_err(CE_NOTE, "/dev/dtrace failed to initialize soft state");
15502 mutex_exit(&cpu_lock);
15503 mutex_exit(&dtrace_provider_lock);
15504 mutex_exit(&dtrace_lock);
15505 return (DDI_FAILURE);
15506 }
15507
15508 if (ddi_create_minor_node(devi, DTRACEMNR_DTRACE, S_IFCHR,
15509 DTRACEMNRN_DTRACE, DDI_PSEUDO, NULL) == DDI_FAILURE ||
15510 ddi_create_minor_node(devi, DTRACEMNR_HELPER, S_IFCHR,
15511 DTRACEMNRN_HELPER, DDI_PSEUDO, NULL) == DDI_FAILURE) {
15512 cmn_err(CE_NOTE, "/dev/dtrace couldn't create minor nodes");
15513 ddi_remove_minor_node(devi, NULL);
15514 ddi_soft_state_fini(&dtrace_softstate);
15515 mutex_exit(&cpu_lock);
15516 mutex_exit(&dtrace_provider_lock);
15517 mutex_exit(&dtrace_lock);
15518 return (DDI_FAILURE);
15519 }
15520
15521 ddi_report_dev(devi);
15522 dtrace_devi = devi;
15523
15524 dtrace_modload = dtrace_module_loaded;
15525 dtrace_modunload = dtrace_module_unloaded;
15526 dtrace_cpu_init = dtrace_cpu_setup_initial;
15527 dtrace_helpers_cleanup = dtrace_helpers_destroy;
15528 dtrace_helpers_fork = dtrace_helpers_duplicate;
15529 dtrace_cpustart_init = dtrace_suspend;
15530 dtrace_cpustart_fini = dtrace_resume;
15531 dtrace_debugger_init = dtrace_suspend;
15532 dtrace_debugger_fini = dtrace_resume;
15533
15534 register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL);
15535
15536 ASSERT(MUTEX_HELD(&cpu_lock));
15537
15538 dtrace_arena = vmem_create("dtrace", (void *)1, UINT32_MAX, 1,
15539 NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
15540 dtrace_minor = vmem_create("dtrace_minor", (void *)DTRACEMNRN_CLONE,
15541 UINT32_MAX - DTRACEMNRN_CLONE, 1, NULL, NULL, NULL, 0,
15542 VM_SLEEP | VMC_IDENTIFIER);
15543 dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri,
15544 1, INT_MAX, 0);
15545
15546 dtrace_state_cache = kmem_cache_create("dtrace_state_cache",
15547 sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN,
15548 NULL, NULL, NULL, NULL, NULL, 0);
15549
15550 ASSERT(MUTEX_HELD(&cpu_lock));
15551 dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod),
15552 offsetof(dtrace_probe_t, dtpr_nextmod),
15553 offsetof(dtrace_probe_t, dtpr_prevmod));
15554
15555 dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func),
15556 offsetof(dtrace_probe_t, dtpr_nextfunc),
15557 offsetof(dtrace_probe_t, dtpr_prevfunc));
15558
15559 dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name),
15560 offsetof(dtrace_probe_t, dtpr_nextname),
15561 offsetof(dtrace_probe_t, dtpr_prevname));
15562
15563 if (dtrace_retain_max < 1) {
15564 cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; "
15565 "setting to 1", dtrace_retain_max);
15566 dtrace_retain_max = 1;
15567 }
15568
15569 /*
15570 * Now discover our toxic ranges.
15571 */
15572 dtrace_toxic_ranges(dtrace_toxrange_add);
15573
15574 /*
15575 * Before we register ourselves as a provider to our own framework,
15576 * we would like to assert that dtrace_provider is NULL -- but that's
15577 * not true if we were loaded as a dependency of a DTrace provider.
15578 * Once we've registered, we can assert that dtrace_provider is our
15579 * pseudo provider.
15580 */
15581 (void) dtrace_register("dtrace", &dtrace_provider_attr,
15582 DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id);
15583
15584 ASSERT(dtrace_provider != NULL);
15585 ASSERT((dtrace_provider_id_t)dtrace_provider == id);
15586
15587 dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t)
15588 dtrace_provider, NULL, NULL, "BEGIN", 0, NULL);
15589 dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t)
15590 dtrace_provider, NULL, NULL, "END", 0, NULL);
15591 dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t)
15592 dtrace_provider, NULL, NULL, "ERROR", 1, NULL);
15593
15594 dtrace_anon_property();
15595 mutex_exit(&cpu_lock);
15596
15597 /*
15598 * If DTrace helper tracing is enabled, we need to allocate the
15599 * trace buffer and initialize the values.
15600 */
15601 if (dtrace_helptrace_enabled) {
15602 ASSERT(dtrace_helptrace_buffer == NULL);
15603 dtrace_helptrace_buffer =
15604 kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP);
15605 dtrace_helptrace_next = 0;
15606 }
15607
15608 /*
15609 * If there are already providers, we must ask them to provide their
15610 * probes, and then match any anonymous enabling against them. Note
15611 * that there should be no other retained enablings at this time:
15612 * the only retained enablings at this time should be the anonymous
15613 * enabling.
15614 */
15615 if (dtrace_anon.dta_enabling != NULL) {
15616 ASSERT(dtrace_retained == dtrace_anon.dta_enabling);
15617
15618 dtrace_enabling_provide(NULL);
15619 state = dtrace_anon.dta_state;
15620
15621 /*
15622 * We couldn't hold cpu_lock across the above call to
15623 * dtrace_enabling_provide(), but we must hold it to actually
15624 * enable the probes. We have to drop all of our locks, pick
15625 * up cpu_lock, and regain our locks before matching the
15626 * retained anonymous enabling.
15627 */
15628 mutex_exit(&dtrace_lock);
15629 mutex_exit(&dtrace_provider_lock);
15630
15631 mutex_enter(&cpu_lock);
15632 mutex_enter(&dtrace_provider_lock);
15633 mutex_enter(&dtrace_lock);
15634
15635 if ((enab = dtrace_anon.dta_enabling) != NULL)
15636 (void) dtrace_enabling_match(enab, NULL);
15637
15638 mutex_exit(&cpu_lock);
15639 }
15640
15641 mutex_exit(&dtrace_lock);
15642 mutex_exit(&dtrace_provider_lock);
15643
15644 if (state != NULL) {
15645 /*
15646 * If we created any anonymous state, set it going now.
15647 */
15648 (void) dtrace_state_go(state, &dtrace_anon.dta_beganon);
15649 }
15650
15651 return (DDI_SUCCESS);
15652}
15653#endif
15654
15655#if !defined(sun)
15656#if __FreeBSD_version >= 800039
15657static void dtrace_dtr(void *);
15658#endif
15659#endif
15660
15661/*ARGSUSED*/
15662static int
15663#if defined(sun)
15664dtrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
15665#else
15666dtrace_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
15667#endif
15668{
15669 dtrace_state_t *state;
15670 uint32_t priv;
15671 uid_t uid;
15672 zoneid_t zoneid;
15673
15674#if defined(sun)
15675 if (getminor(*devp) == DTRACEMNRN_HELPER)
15676 return (0);
15677
15678 /*
15679 * If this wasn't an open with the "helper" minor, then it must be
15680 * the "dtrace" minor.
15681 */
15682 ASSERT(getminor(*devp) == DTRACEMNRN_DTRACE);
15683#else
15684 cred_t *cred_p = NULL;
15685
15686#if __FreeBSD_version < 800039
15687 /*
15688 * The first minor device is the one that is cloned so there is
15689 * nothing more to do here.
15690 */
15691 if (dev2unit(dev) == 0)
15692 return 0;
15693
15694 /*
15695 * Devices are cloned, so if the DTrace state has already
15696 * been allocated, that means this device belongs to a
15697 * different client. Each client should open '/dev/dtrace'
15698 * to get a cloned device.
15699 */
15700 if (dev->si_drv1 != NULL)
15701 return (EBUSY);
15702#endif
15703
15704 cred_p = dev->si_cred;
15705#endif
15706
15707 /*
15708 * If no DTRACE_PRIV_* bits are set in the credential, then the
15709 * caller lacks sufficient permission to do anything with DTrace.
15710 */
15711 dtrace_cred2priv(cred_p, &priv, &uid, &zoneid);
15712 if (priv == DTRACE_PRIV_NONE) {
15713#if !defined(sun)
15714#if __FreeBSD_version < 800039
15715 /* Destroy the cloned device. */
15716 destroy_dev(dev);
15717#endif
15718#endif
15719
15720 return (EACCES);
15721 }
15722
15723 /*
15724 * Ask all providers to provide all their probes.
15725 */
15726 mutex_enter(&dtrace_provider_lock);
15727 dtrace_probe_provide(NULL, NULL);
15728 mutex_exit(&dtrace_provider_lock);
15729
15730 mutex_enter(&cpu_lock);
15731 mutex_enter(&dtrace_lock);
15732 dtrace_opens++;
15733 dtrace_membar_producer();
15734
15735#if defined(sun)
15736 /*
15737 * If the kernel debugger is active (that is, if the kernel debugger
15738 * modified text in some way), we won't allow the open.
15739 */
15740 if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) {
15741 dtrace_opens--;
15742 mutex_exit(&cpu_lock);
15743 mutex_exit(&dtrace_lock);
15744 return (EBUSY);
15745 }
15746
15747 state = dtrace_state_create(devp, cred_p);
15748#else
15749 state = dtrace_state_create(dev);
15750#if __FreeBSD_version < 800039
15751 dev->si_drv1 = state;
15752#else
15753 devfs_set_cdevpriv(state, dtrace_dtr);
15754#endif
15755#endif
15756
15757 mutex_exit(&cpu_lock);
15758
15759 if (state == NULL) {
15760#if defined(sun)
15761 if (--dtrace_opens == 0)
15762 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
15763#else
15764 --dtrace_opens;
15765#endif
15766 mutex_exit(&dtrace_lock);
15767#if !defined(sun)
15768#if __FreeBSD_version < 800039
15769 /* Destroy the cloned device. */
15770 destroy_dev(dev);
15771#endif
15772#endif
15773 return (EAGAIN);
15774 }
15775
15776 mutex_exit(&dtrace_lock);
15777
15778 return (0);
15779}
15780
15781/*ARGSUSED*/
15782#if defined(sun)
15783static int
15784dtrace_close(dev_t dev, int flag, int otyp, cred_t *cred_p)
15785#elif __FreeBSD_version < 800039
15786static int
15787dtrace_close(struct cdev *dev, int flags, int fmt __unused, struct thread *td)
15788#else
15789static void
15790dtrace_dtr(void *data)
15791#endif
15792{
15793#if defined(sun)
15794 minor_t minor = getminor(dev);
15795 dtrace_state_t *state;
15796
15797 if (minor == DTRACEMNRN_HELPER)
15798 return (0);
15799
15800 state = ddi_get_soft_state(dtrace_softstate, minor);
15801#else
15802#if __FreeBSD_version < 800039
15803 dtrace_state_t *state = dev->si_drv1;
15804
15805 /* Check if this is not a cloned device. */
15806 if (dev2unit(dev) == 0)
15807 return (0);
15808#else
15809 dtrace_state_t *state = data;
15810#endif
15811
15812#endif
15813
15814 mutex_enter(&cpu_lock);
15815 mutex_enter(&dtrace_lock);
15816
15817 if (state != NULL) {
15818 if (state->dts_anon) {
15819 /*
15820 * There is anonymous state. Destroy that first.
15821 */
15822 ASSERT(dtrace_anon.dta_state == NULL);
15823 dtrace_state_destroy(state->dts_anon);
15824 }
15825
15826 dtrace_state_destroy(state);
15827
15828#if !defined(sun)
15829 kmem_free(state, 0);
15830#if __FreeBSD_version < 800039
15831 dev->si_drv1 = NULL;
15832#endif
15833#endif
15834 }
15835
15836 ASSERT(dtrace_opens > 0);
15837#if defined(sun)
15838 if (--dtrace_opens == 0)
15839 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
15840#else
15841 --dtrace_opens;
15842#endif
15843
15844 mutex_exit(&dtrace_lock);
15845 mutex_exit(&cpu_lock);
15846
15847#if __FreeBSD_version < 800039
15848 /* Schedule this cloned device to be destroyed. */
15849 destroy_dev_sched(dev);
15850#endif
15851
15852#if defined(sun) || __FreeBSD_version < 800039
15853 return (0);
15854#endif
15855}
15856
15857#if defined(sun)
15858/*ARGSUSED*/
15859static int
15860dtrace_ioctl_helper(int cmd, intptr_t arg, int *rv)
15861{
15862 int rval;
15863 dof_helper_t help, *dhp = NULL;
15864
15865 switch (cmd) {
15866 case DTRACEHIOC_ADDDOF:
15867 if (copyin((void *)arg, &help, sizeof (help)) != 0) {
15868 dtrace_dof_error(NULL, "failed to copyin DOF helper");
15869 return (EFAULT);
15870 }
15871
15872 dhp = &help;
15873 arg = (intptr_t)help.dofhp_dof;
15874 /*FALLTHROUGH*/
15875
15876 case DTRACEHIOC_ADD: {
15877 dof_hdr_t *dof = dtrace_dof_copyin(arg, &rval);
15878
15879 if (dof == NULL)
15880 return (rval);
15881
15882 mutex_enter(&dtrace_lock);
15883
15884 /*
15885 * dtrace_helper_slurp() takes responsibility for the dof --
15886 * it may free it now or it may save it and free it later.
15887 */
15888 if ((rval = dtrace_helper_slurp(dof, dhp)) != -1) {
15889 *rv = rval;
15890 rval = 0;
15891 } else {
15892 rval = EINVAL;
15893 }
15894
15895 mutex_exit(&dtrace_lock);
15896 return (rval);
15897 }
15898
15899 case DTRACEHIOC_REMOVE: {
15900 mutex_enter(&dtrace_lock);
15901 rval = dtrace_helper_destroygen(arg);
15902 mutex_exit(&dtrace_lock);
15903
15904 return (rval);
15905 }
15906
15907 default:
15908 break;
15909 }
15910
15911 return (ENOTTY);
15912}
15913
15914/*ARGSUSED*/
15915static int
15916dtrace_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv)
15917{
15918 minor_t minor = getminor(dev);
15919 dtrace_state_t *state;
15920 int rval;
15921
15922 if (minor == DTRACEMNRN_HELPER)
15923 return (dtrace_ioctl_helper(cmd, arg, rv));
15924
15925 state = ddi_get_soft_state(dtrace_softstate, minor);
15926
15927 if (state->dts_anon) {
15928 ASSERT(dtrace_anon.dta_state == NULL);
15929 state = state->dts_anon;
15930 }
15931
15932 switch (cmd) {
15933 case DTRACEIOC_PROVIDER: {
15934 dtrace_providerdesc_t pvd;
15935 dtrace_provider_t *pvp;
15936
15937 if (copyin((void *)arg, &pvd, sizeof (pvd)) != 0)
15938 return (EFAULT);
15939
15940 pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0';
15941 mutex_enter(&dtrace_provider_lock);
15942
15943 for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) {
15944 if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0)
15945 break;
15946 }
15947
15948 mutex_exit(&dtrace_provider_lock);
15949
15950 if (pvp == NULL)
15951 return (ESRCH);
15952
15953 bcopy(&pvp->dtpv_priv, &pvd.dtvd_priv, sizeof (dtrace_ppriv_t));
15954 bcopy(&pvp->dtpv_attr, &pvd.dtvd_attr, sizeof (dtrace_pattr_t));
15955
15956 if (copyout(&pvd, (void *)arg, sizeof (pvd)) != 0)
15957 return (EFAULT);
15958
15959 return (0);
15960 }
15961
15962 case DTRACEIOC_EPROBE: {
15963 dtrace_eprobedesc_t epdesc;
15964 dtrace_ecb_t *ecb;
15965 dtrace_action_t *act;
15966 void *buf;
15967 size_t size;
15968 uintptr_t dest;
15969 int nrecs;
15970
15971 if (copyin((void *)arg, &epdesc, sizeof (epdesc)) != 0)
15972 return (EFAULT);
15973
15974 mutex_enter(&dtrace_lock);
15975
15976 if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) {
15977 mutex_exit(&dtrace_lock);
15978 return (EINVAL);
15979 }
15980
15981 if (ecb->dte_probe == NULL) {
15982 mutex_exit(&dtrace_lock);
15983 return (EINVAL);
15984 }
15985
15986 epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id;
15987 epdesc.dtepd_uarg = ecb->dte_uarg;
15988 epdesc.dtepd_size = ecb->dte_size;
15989
15990 nrecs = epdesc.dtepd_nrecs;
15991 epdesc.dtepd_nrecs = 0;
15992 for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
15993 if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple)
15994 continue;
15995
15996 epdesc.dtepd_nrecs++;
15997 }
15998
15999 /*
16000 * Now that we have the size, we need to allocate a temporary
16001 * buffer in which to store the complete description. We need
16002 * the temporary buffer to be able to drop dtrace_lock()
16003 * across the copyout(), below.
16004 */
16005 size = sizeof (dtrace_eprobedesc_t) +
16006 (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t));
16007
16008 buf = kmem_alloc(size, KM_SLEEP);
16009 dest = (uintptr_t)buf;
16010
16011 bcopy(&epdesc, (void *)dest, sizeof (epdesc));
16012 dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]);
16013
16014 for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
16015 if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple)
16016 continue;
16017
16018 if (nrecs-- == 0)
16019 break;
16020
16021 bcopy(&act->dta_rec, (void *)dest,
16022 sizeof (dtrace_recdesc_t));
16023 dest += sizeof (dtrace_recdesc_t);
16024 }
16025
16026 mutex_exit(&dtrace_lock);
16027
16028 if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) {
16029 kmem_free(buf, size);
16030 return (EFAULT);
16031 }
16032
16033 kmem_free(buf, size);
16034 return (0);
16035 }
16036
16037 case DTRACEIOC_AGGDESC: {
16038 dtrace_aggdesc_t aggdesc;
16039 dtrace_action_t *act;
16040 dtrace_aggregation_t *agg;
16041 int nrecs;
16042 uint32_t offs;
16043 dtrace_recdesc_t *lrec;
16044 void *buf;
16045 size_t size;
16046 uintptr_t dest;
16047
16048 if (copyin((void *)arg, &aggdesc, sizeof (aggdesc)) != 0)
16049 return (EFAULT);
16050
16051 mutex_enter(&dtrace_lock);
16052
16053 if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) {
16054 mutex_exit(&dtrace_lock);
16055 return (EINVAL);
16056 }
16057
16058 aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid;
16059
16060 nrecs = aggdesc.dtagd_nrecs;
16061 aggdesc.dtagd_nrecs = 0;
16062
16063 offs = agg->dtag_base;
16064 lrec = &agg->dtag_action.dta_rec;
16065 aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs;
16066
16067 for (act = agg->dtag_first; ; act = act->dta_next) {
16068 ASSERT(act->dta_intuple ||
16069 DTRACEACT_ISAGG(act->dta_kind));
16070
16071 /*
16072 * If this action has a record size of zero, it
16073 * denotes an argument to the aggregating action.
16074 * Because the presence of this record doesn't (or
16075 * shouldn't) affect the way the data is interpreted,
16076 * we don't copy it out to save user-level the
16077 * confusion of dealing with a zero-length record.
16078 */
16079 if (act->dta_rec.dtrd_size == 0) {
16080 ASSERT(agg->dtag_hasarg);
16081 continue;
16082 }
16083
16084 aggdesc.dtagd_nrecs++;
16085
16086 if (act == &agg->dtag_action)
16087 break;
16088 }
16089
16090 /*
16091 * Now that we have the size, we need to allocate a temporary
16092 * buffer in which to store the complete description. We need
16093 * the temporary buffer to be able to drop dtrace_lock()
16094 * across the copyout(), below.
16095 */
16096 size = sizeof (dtrace_aggdesc_t) +
16097 (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t));
16098
16099 buf = kmem_alloc(size, KM_SLEEP);
16100 dest = (uintptr_t)buf;
16101
16102 bcopy(&aggdesc, (void *)dest, sizeof (aggdesc));
16103 dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]);
16104
16105 for (act = agg->dtag_first; ; act = act->dta_next) {
16106 dtrace_recdesc_t rec = act->dta_rec;
16107
16108 /*
16109 * See the comment in the above loop for why we pass
16110 * over zero-length records.
16111 */
16112 if (rec.dtrd_size == 0) {
16113 ASSERT(agg->dtag_hasarg);
16114 continue;
16115 }
16116
16117 if (nrecs-- == 0)
16118 break;
16119
16120 rec.dtrd_offset -= offs;
16121 bcopy(&rec, (void *)dest, sizeof (rec));
16122 dest += sizeof (dtrace_recdesc_t);
16123
16124 if (act == &agg->dtag_action)
16125 break;
16126 }
16127
16128 mutex_exit(&dtrace_lock);
16129
16130 if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) {
16131 kmem_free(buf, size);
16132 return (EFAULT);
16133 }
16134
16135 kmem_free(buf, size);
16136 return (0);
16137 }
16138
16139 case DTRACEIOC_ENABLE: {
16140 dof_hdr_t *dof;
16141 dtrace_enabling_t *enab = NULL;
16142 dtrace_vstate_t *vstate;
16143 int err = 0;
16144
16145 *rv = 0;
16146
16147 /*
16148 * If a NULL argument has been passed, we take this as our
16149 * cue to reevaluate our enablings.
16150 */
16151 if (arg == NULL) {
16152 dtrace_enabling_matchall();
16153
16154 return (0);
16155 }
16156
16157 if ((dof = dtrace_dof_copyin(arg, &rval)) == NULL)
16158 return (rval);
16159
16160 mutex_enter(&cpu_lock);
16161 mutex_enter(&dtrace_lock);
16162 vstate = &state->dts_vstate;
16163
16164 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) {
16165 mutex_exit(&dtrace_lock);
16166 mutex_exit(&cpu_lock);
16167 dtrace_dof_destroy(dof);
16168 return (EBUSY);
16169 }
16170
16171 if (dtrace_dof_slurp(dof, vstate, cr, &enab, 0, B_TRUE) != 0) {
16172 mutex_exit(&dtrace_lock);
16173 mutex_exit(&cpu_lock);
16174 dtrace_dof_destroy(dof);
16175 return (EINVAL);
16176 }
16177
16178 if ((rval = dtrace_dof_options(dof, state)) != 0) {
16179 dtrace_enabling_destroy(enab);
16180 mutex_exit(&dtrace_lock);
16181 mutex_exit(&cpu_lock);
16182 dtrace_dof_destroy(dof);
16183 return (rval);
16184 }
16185
16186 if ((err = dtrace_enabling_match(enab, rv)) == 0) {
16187 err = dtrace_enabling_retain(enab);
16188 } else {
16189 dtrace_enabling_destroy(enab);
16190 }
16191
16192 mutex_exit(&cpu_lock);
16193 mutex_exit(&dtrace_lock);
16194 dtrace_dof_destroy(dof);
16195
16196 return (err);
16197 }
16198
16199 case DTRACEIOC_REPLICATE: {
16200 dtrace_repldesc_t desc;
16201 dtrace_probedesc_t *match = &desc.dtrpd_match;
16202 dtrace_probedesc_t *create = &desc.dtrpd_create;
16203 int err;
16204
16205 if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
16206 return (EFAULT);
16207
16208 match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
16209 match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
16210 match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
16211 match->dtpd_name[DTRACE_NAMELEN - 1] = '\0';
16212
16213 create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
16214 create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
16215 create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
16216 create->dtpd_name[DTRACE_NAMELEN - 1] = '\0';
16217
16218 mutex_enter(&dtrace_lock);
16219 err = dtrace_enabling_replicate(state, match, create);
16220 mutex_exit(&dtrace_lock);
16221
16222 return (err);
16223 }
16224
16225 case DTRACEIOC_PROBEMATCH:
16226 case DTRACEIOC_PROBES: {
16227 dtrace_probe_t *probe = NULL;
16228 dtrace_probedesc_t desc;
16229 dtrace_probekey_t pkey;
16230 dtrace_id_t i;
16231 int m = 0;
16232 uint32_t priv;
16233 uid_t uid;
16234 zoneid_t zoneid;
16235
16236 if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
16237 return (EFAULT);
16238
16239 desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
16240 desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
16241 desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
16242 desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0';
16243
16244 /*
16245 * Before we attempt to match this probe, we want to give
16246 * all providers the opportunity to provide it.
16247 */
16248 if (desc.dtpd_id == DTRACE_IDNONE) {
16249 mutex_enter(&dtrace_provider_lock);
16250 dtrace_probe_provide(&desc, NULL);
16251 mutex_exit(&dtrace_provider_lock);
16252 desc.dtpd_id++;
16253 }
16254
16255 if (cmd == DTRACEIOC_PROBEMATCH) {
16256 dtrace_probekey(&desc, &pkey);
16257 pkey.dtpk_id = DTRACE_IDNONE;
16258 }
16259
16260 dtrace_cred2priv(cr, &priv, &uid, &zoneid);
16261
16262 mutex_enter(&dtrace_lock);
16263
16264 if (cmd == DTRACEIOC_PROBEMATCH) {
16265 for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) {
16266 if ((probe = dtrace_probes[i - 1]) != NULL &&
16267 (m = dtrace_match_probe(probe, &pkey,
16268 priv, uid, zoneid)) != 0)
16269 break;
16270 }
16271
16272 if (m < 0) {
16273 mutex_exit(&dtrace_lock);
16274 return (EINVAL);
16275 }
16276
16277 } else {
16278 for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) {
16279 if ((probe = dtrace_probes[i - 1]) != NULL &&
16280 dtrace_match_priv(probe, priv, uid, zoneid))
16281 break;
16282 }
16283 }
16284
16285 if (probe == NULL) {
16286 mutex_exit(&dtrace_lock);
16287 return (ESRCH);
16288 }
16289
16290 dtrace_probe_description(probe, &desc);
16291 mutex_exit(&dtrace_lock);
16292
16293 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16294 return (EFAULT);
16295
16296 return (0);
16297 }
16298
16299 case DTRACEIOC_PROBEARG: {
16300 dtrace_argdesc_t desc;
16301 dtrace_probe_t *probe;
16302 dtrace_provider_t *prov;
16303
16304 if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
16305 return (EFAULT);
16306
16307 if (desc.dtargd_id == DTRACE_IDNONE)
16308 return (EINVAL);
16309
16310 if (desc.dtargd_ndx == DTRACE_ARGNONE)
16311 return (EINVAL);
16312
16313 mutex_enter(&dtrace_provider_lock);
16314 mutex_enter(&mod_lock);
16315 mutex_enter(&dtrace_lock);
16316
16317 if (desc.dtargd_id > dtrace_nprobes) {
16318 mutex_exit(&dtrace_lock);
16319 mutex_exit(&mod_lock);
16320 mutex_exit(&dtrace_provider_lock);
16321 return (EINVAL);
16322 }
16323
16324 if ((probe = dtrace_probes[desc.dtargd_id - 1]) == NULL) {
16325 mutex_exit(&dtrace_lock);
16326 mutex_exit(&mod_lock);
16327 mutex_exit(&dtrace_provider_lock);
16328 return (EINVAL);
16329 }
16330
16331 mutex_exit(&dtrace_lock);
16332
16333 prov = probe->dtpr_provider;
16334
16335 if (prov->dtpv_pops.dtps_getargdesc == NULL) {
16336 /*
16337 * There isn't any typed information for this probe.
16338 * Set the argument number to DTRACE_ARGNONE.
16339 */
16340 desc.dtargd_ndx = DTRACE_ARGNONE;
16341 } else {
16342 desc.dtargd_native[0] = '\0';
16343 desc.dtargd_xlate[0] = '\0';
16344 desc.dtargd_mapping = desc.dtargd_ndx;
16345
16346 prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg,
16347 probe->dtpr_id, probe->dtpr_arg, &desc);
16348 }
16349
16350 mutex_exit(&mod_lock);
16351 mutex_exit(&dtrace_provider_lock);
16352
16353 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16354 return (EFAULT);
16355
16356 return (0);
16357 }
16358
16359 case DTRACEIOC_GO: {
16360 processorid_t cpuid;
16361 rval = dtrace_state_go(state, &cpuid);
16362
16363 if (rval != 0)
16364 return (rval);
16365
16366 if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0)
16367 return (EFAULT);
16368
16369 return (0);
16370 }
16371
16372 case DTRACEIOC_STOP: {
16373 processorid_t cpuid;
16374
16375 mutex_enter(&dtrace_lock);
16376 rval = dtrace_state_stop(state, &cpuid);
16377 mutex_exit(&dtrace_lock);
16378
16379 if (rval != 0)
16380 return (rval);
16381
16382 if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0)
16383 return (EFAULT);
16384
16385 return (0);
16386 }
16387
16388 case DTRACEIOC_DOFGET: {
16389 dof_hdr_t hdr, *dof;
16390 uint64_t len;
16391
16392 if (copyin((void *)arg, &hdr, sizeof (hdr)) != 0)
16393 return (EFAULT);
16394
16395 mutex_enter(&dtrace_lock);
16396 dof = dtrace_dof_create(state);
16397 mutex_exit(&dtrace_lock);
16398
16399 len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz);
16400 rval = copyout(dof, (void *)arg, len);
16401 dtrace_dof_destroy(dof);
16402
16403 return (rval == 0 ? 0 : EFAULT);
16404 }
16405
16406 case DTRACEIOC_AGGSNAP:
16407 case DTRACEIOC_BUFSNAP: {
16408 dtrace_bufdesc_t desc;
16409 caddr_t cached;
16410 dtrace_buffer_t *buf;
16411
16412 if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
16413 return (EFAULT);
16414
16415 if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NCPU)
16416 return (EINVAL);
16417
16418 mutex_enter(&dtrace_lock);
16419
16420 if (cmd == DTRACEIOC_BUFSNAP) {
16421 buf = &state->dts_buffer[desc.dtbd_cpu];
16422 } else {
16423 buf = &state->dts_aggbuffer[desc.dtbd_cpu];
16424 }
16425
16426 if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) {
16427 size_t sz = buf->dtb_offset;
16428
16429 if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) {
16430 mutex_exit(&dtrace_lock);
16431 return (EBUSY);
16432 }
16433
16434 /*
16435 * If this buffer has already been consumed, we're
16436 * going to indicate that there's nothing left here
16437 * to consume.
16438 */
16439 if (buf->dtb_flags & DTRACEBUF_CONSUMED) {
16440 mutex_exit(&dtrace_lock);
16441
16442 desc.dtbd_size = 0;
16443 desc.dtbd_drops = 0;
16444 desc.dtbd_errors = 0;
16445 desc.dtbd_oldest = 0;
16446 sz = sizeof (desc);
16447
16448 if (copyout(&desc, (void *)arg, sz) != 0)
16449 return (EFAULT);
16450
16451 return (0);
16452 }
16453
16454 /*
16455 * If this is a ring buffer that has wrapped, we want
16456 * to copy the whole thing out.
16457 */
16458 if (buf->dtb_flags & DTRACEBUF_WRAPPED) {
16459 dtrace_buffer_polish(buf);
16460 sz = buf->dtb_size;
16461 }
16462
16463 if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) {
16464 mutex_exit(&dtrace_lock);
16465 return (EFAULT);
16466 }
16467
16468 desc.dtbd_size = sz;
16469 desc.dtbd_drops = buf->dtb_drops;
16470 desc.dtbd_errors = buf->dtb_errors;
16471 desc.dtbd_oldest = buf->dtb_xamot_offset;
16472 desc.dtbd_timestamp = dtrace_gethrtime();
16473
16474 mutex_exit(&dtrace_lock);
16475
16476 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16477 return (EFAULT);
16478
16479 buf->dtb_flags |= DTRACEBUF_CONSUMED;
16480
16481 return (0);
16482 }
16483
16484 if (buf->dtb_tomax == NULL) {
16485 ASSERT(buf->dtb_xamot == NULL);
16486 mutex_exit(&dtrace_lock);
16487 return (ENOENT);
16488 }
16489
16490 cached = buf->dtb_tomax;
16491 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
16492
16493 dtrace_xcall(desc.dtbd_cpu,
16494 (dtrace_xcall_t)dtrace_buffer_switch, buf);
16495
16496 state->dts_errors += buf->dtb_xamot_errors;
16497
16498 /*
16499 * If the buffers did not actually switch, then the cross call
16500 * did not take place -- presumably because the given CPU is
16501 * not in the ready set. If this is the case, we'll return
16502 * ENOENT.
16503 */
16504 if (buf->dtb_tomax == cached) {
16505 ASSERT(buf->dtb_xamot != cached);
16506 mutex_exit(&dtrace_lock);
16507 return (ENOENT);
16508 }
16509
16510 ASSERT(cached == buf->dtb_xamot);
16511
16512 /*
16513 * We have our snapshot; now copy it out.
16514 */
16515 if (copyout(buf->dtb_xamot, desc.dtbd_data,
16516 buf->dtb_xamot_offset) != 0) {
16517 mutex_exit(&dtrace_lock);
16518 return (EFAULT);
16519 }
16520
16521 desc.dtbd_size = buf->dtb_xamot_offset;
16522 desc.dtbd_drops = buf->dtb_xamot_drops;
16523 desc.dtbd_errors = buf->dtb_xamot_errors;
16524 desc.dtbd_oldest = 0;
16525 desc.dtbd_timestamp = buf->dtb_switched;
16526
16527 mutex_exit(&dtrace_lock);
16528
16529 /*
16530 * Finally, copy out the buffer description.
16531 */
16532 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16533 return (EFAULT);
16534
16535 return (0);
16536 }
16537
16538 case DTRACEIOC_CONF: {
16539 dtrace_conf_t conf;
16540
16541 bzero(&conf, sizeof (conf));
16542 conf.dtc_difversion = DIF_VERSION;
16543 conf.dtc_difintregs = DIF_DIR_NREGS;
16544 conf.dtc_diftupregs = DIF_DTR_NREGS;
16545 conf.dtc_ctfmodel = CTF_MODEL_NATIVE;
16546
16547 if (copyout(&conf, (void *)arg, sizeof (conf)) != 0)
16548 return (EFAULT);
16549
16550 return (0);
16551 }
16552
16553 case DTRACEIOC_STATUS: {
16554 dtrace_status_t stat;
16555 dtrace_dstate_t *dstate;
16556 int i, j;
16557 uint64_t nerrs;
16558
16559 /*
16560 * See the comment in dtrace_state_deadman() for the reason
16561 * for setting dts_laststatus to INT64_MAX before setting
16562 * it to the correct value.
16563 */
16564 state->dts_laststatus = INT64_MAX;
16565 dtrace_membar_producer();
16566 state->dts_laststatus = dtrace_gethrtime();
16567
16568 bzero(&stat, sizeof (stat));
16569
16570 mutex_enter(&dtrace_lock);
16571
16572 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) {
16573 mutex_exit(&dtrace_lock);
16574 return (ENOENT);
16575 }
16576
16577 if (state->dts_activity == DTRACE_ACTIVITY_DRAINING)
16578 stat.dtst_exiting = 1;
16579
16580 nerrs = state->dts_errors;
16581 dstate = &state->dts_vstate.dtvs_dynvars;
16582
16583 for (i = 0; i < NCPU; i++) {
16584 dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i];
16585
16586 stat.dtst_dyndrops += dcpu->dtdsc_drops;
16587 stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops;
16588 stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops;
16589
16590 if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL)
16591 stat.dtst_filled++;
16592
16593 nerrs += state->dts_buffer[i].dtb_errors;
16594
16595 for (j = 0; j < state->dts_nspeculations; j++) {
16596 dtrace_speculation_t *spec;
16597 dtrace_buffer_t *buf;
16598
16599 spec = &state->dts_speculations[j];
16600 buf = &spec->dtsp_buffer[i];
16601 stat.dtst_specdrops += buf->dtb_xamot_drops;
16602 }
16603 }
16604
16605 stat.dtst_specdrops_busy = state->dts_speculations_busy;
16606 stat.dtst_specdrops_unavail = state->dts_speculations_unavail;
16607 stat.dtst_stkstroverflows = state->dts_stkstroverflows;
16608 stat.dtst_dblerrors = state->dts_dblerrors;
16609 stat.dtst_killed =
16610 (state->dts_activity == DTRACE_ACTIVITY_KILLED);
16611 stat.dtst_errors = nerrs;
16612
16613 mutex_exit(&dtrace_lock);
16614
16615 if (copyout(&stat, (void *)arg, sizeof (stat)) != 0)
16616 return (EFAULT);
16617
16618 return (0);
16619 }
16620
16621 case DTRACEIOC_FORMAT: {
16622 dtrace_fmtdesc_t fmt;
16623 char *str;
16624 int len;
16625
16626 if (copyin((void *)arg, &fmt, sizeof (fmt)) != 0)
16627 return (EFAULT);
16628
16629 mutex_enter(&dtrace_lock);
16630
16631 if (fmt.dtfd_format == 0 ||
16632 fmt.dtfd_format > state->dts_nformats) {
16633 mutex_exit(&dtrace_lock);
16634 return (EINVAL);
16635 }
16636
16637 /*
16638 * Format strings are allocated contiguously and they are
16639 * never freed; if a format index is less than the number
16640 * of formats, we can assert that the format map is non-NULL
16641 * and that the format for the specified index is non-NULL.
16642 */
16643 ASSERT(state->dts_formats != NULL);
16644 str = state->dts_formats[fmt.dtfd_format - 1];
16645 ASSERT(str != NULL);
16646
16647 len = strlen(str) + 1;
16648
16649 if (len > fmt.dtfd_length) {
16650 fmt.dtfd_length = len;
16651
16652 if (copyout(&fmt, (void *)arg, sizeof (fmt)) != 0) {
16653 mutex_exit(&dtrace_lock);
16654 return (EINVAL);
16655 }
16656 } else {
16657 if (copyout(str, fmt.dtfd_string, len) != 0) {
16658 mutex_exit(&dtrace_lock);
16659 return (EINVAL);
16660 }
16661 }
16662
16663 mutex_exit(&dtrace_lock);
16664 return (0);
16665 }
16666
16667 default:
16668 break;
16669 }
16670
16671 return (ENOTTY);
16672}
16673
16674/*ARGSUSED*/
16675static int
16676dtrace_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
16677{
16678 dtrace_state_t *state;
16679
16680 switch (cmd) {
16681 case DDI_DETACH:
16682 break;
16683
16684 case DDI_SUSPEND:
16685 return (DDI_SUCCESS);
16686
16687 default:
16688 return (DDI_FAILURE);
16689 }
16690
16691 mutex_enter(&cpu_lock);
16692 mutex_enter(&dtrace_provider_lock);
16693 mutex_enter(&dtrace_lock);
16694
16695 ASSERT(dtrace_opens == 0);
16696
16697 if (dtrace_helpers > 0) {
16698 mutex_exit(&dtrace_provider_lock);
16699 mutex_exit(&dtrace_lock);
16700 mutex_exit(&cpu_lock);
16701 return (DDI_FAILURE);
16702 }
16703
16704 if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) {
16705 mutex_exit(&dtrace_provider_lock);
16706 mutex_exit(&dtrace_lock);
16707 mutex_exit(&cpu_lock);
16708 return (DDI_FAILURE);
16709 }
16710
16711 dtrace_provider = NULL;
16712
16713 if ((state = dtrace_anon_grab()) != NULL) {
16714 /*
16715 * If there were ECBs on this state, the provider should
16716 * have not been allowed to detach; assert that there is
16717 * none.
16718 */
16719 ASSERT(state->dts_necbs == 0);
16720 dtrace_state_destroy(state);
16721
16722 /*
16723 * If we're being detached with anonymous state, we need to
16724 * indicate to the kernel debugger that DTrace is now inactive.
16725 */
16726 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
16727 }
16728
16729 bzero(&dtrace_anon, sizeof (dtrace_anon_t));
16730 unregister_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL);
16731 dtrace_cpu_init = NULL;
16732 dtrace_helpers_cleanup = NULL;
16733 dtrace_helpers_fork = NULL;
16734 dtrace_cpustart_init = NULL;
16735 dtrace_cpustart_fini = NULL;
16736 dtrace_debugger_init = NULL;
16737 dtrace_debugger_fini = NULL;
16738 dtrace_modload = NULL;
16739 dtrace_modunload = NULL;
16740
16741 mutex_exit(&cpu_lock);
16742
16743 if (dtrace_helptrace_enabled) {
16744 kmem_free(dtrace_helptrace_buffer, dtrace_helptrace_bufsize);
16745 dtrace_helptrace_buffer = NULL;
16746 }
16747
16748 kmem_free(dtrace_probes, dtrace_nprobes * sizeof (dtrace_probe_t *));
16749 dtrace_probes = NULL;
16750 dtrace_nprobes = 0;
16751
16752 dtrace_hash_destroy(dtrace_bymod);
16753 dtrace_hash_destroy(dtrace_byfunc);
16754 dtrace_hash_destroy(dtrace_byname);
16755 dtrace_bymod = NULL;
16756 dtrace_byfunc = NULL;
16757 dtrace_byname = NULL;
16758
16759 kmem_cache_destroy(dtrace_state_cache);
16760 vmem_destroy(dtrace_minor);
16761 vmem_destroy(dtrace_arena);
16762
16763 if (dtrace_toxrange != NULL) {
16764 kmem_free(dtrace_toxrange,
16765 dtrace_toxranges_max * sizeof (dtrace_toxrange_t));
16766 dtrace_toxrange = NULL;
16767 dtrace_toxranges = 0;
16768 dtrace_toxranges_max = 0;
16769 }
16770
16771 ddi_remove_minor_node(dtrace_devi, NULL);
16772 dtrace_devi = NULL;
16773
16774 ddi_soft_state_fini(&dtrace_softstate);
16775
16776 ASSERT(dtrace_vtime_references == 0);
16777 ASSERT(dtrace_opens == 0);
16778 ASSERT(dtrace_retained == NULL);
16779
16780 mutex_exit(&dtrace_lock);
16781 mutex_exit(&dtrace_provider_lock);
16782
16783 /*
16784 * We don't destroy the task queue until after we have dropped our
16785 * locks (taskq_destroy() may block on running tasks). To prevent
16786 * attempting to do work after we have effectively detached but before
16787 * the task queue has been destroyed, all tasks dispatched via the
16788 * task queue must check that DTrace is still attached before
16789 * performing any operation.
16790 */
16791 taskq_destroy(dtrace_taskq);
16792 dtrace_taskq = NULL;
16793
16794 return (DDI_SUCCESS);
16795}
16796#endif
16797
16798#if defined(sun)
16799/*ARGSUSED*/
16800static int
16801dtrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
16802{
16803 int error;
16804
16805 switch (infocmd) {
16806 case DDI_INFO_DEVT2DEVINFO:
16807 *result = (void *)dtrace_devi;
16808 error = DDI_SUCCESS;
16809 break;
16810 case DDI_INFO_DEVT2INSTANCE:
16811 *result = (void *)0;
16812 error = DDI_SUCCESS;
16813 break;
16814 default:
16815 error = DDI_FAILURE;
16816 }
16817 return (error);
16818}
16819#endif
16820
16821#if defined(sun)
16822static struct cb_ops dtrace_cb_ops = {
16823 dtrace_open, /* open */
16824 dtrace_close, /* close */
16825 nulldev, /* strategy */
16826 nulldev, /* print */
16827 nodev, /* dump */
16828 nodev, /* read */
16829 nodev, /* write */
16830 dtrace_ioctl, /* ioctl */
16831 nodev, /* devmap */
16832 nodev, /* mmap */
16833 nodev, /* segmap */
16834 nochpoll, /* poll */
16835 ddi_prop_op, /* cb_prop_op */
16836 0, /* streamtab */
16837 D_NEW | D_MP /* Driver compatibility flag */
16838};
16839
16840static struct dev_ops dtrace_ops = {
16841 DEVO_REV, /* devo_rev */
16842 0, /* refcnt */
16843 dtrace_info, /* get_dev_info */
16844 nulldev, /* identify */
16845 nulldev, /* probe */
16846 dtrace_attach, /* attach */
16847 dtrace_detach, /* detach */
16848 nodev, /* reset */
16849 &dtrace_cb_ops, /* driver operations */
16850 NULL, /* bus operations */
16851 nodev /* dev power */
16852};
16853
16854static struct modldrv modldrv = {
16855 &mod_driverops, /* module type (this is a pseudo driver) */
16856 "Dynamic Tracing", /* name of module */
16857 &dtrace_ops, /* driver ops */
16858};
16859
16860static struct modlinkage modlinkage = {
16861 MODREV_1,
16862 (void *)&modldrv,
16863 NULL
16864};
16865
16866int
16867_init(void)
16868{
16869 return (mod_install(&modlinkage));
16870}
16871
16872int
16873_info(struct modinfo *modinfop)
16874{
16875 return (mod_info(&modlinkage, modinfop));
16876}
16877
16878int
16879_fini(void)
16880{
16881 return (mod_remove(&modlinkage));
16882}
16883#else
16884
16885static d_ioctl_t dtrace_ioctl;
16886static d_ioctl_t dtrace_ioctl_helper;
16887static void dtrace_load(void *);
16888static int dtrace_unload(void);
16889#if __FreeBSD_version < 800039
16890static void dtrace_clone(void *, struct ucred *, char *, int , struct cdev **);
16891static struct clonedevs *dtrace_clones; /* Ptr to the array of cloned devices. */
16892static eventhandler_tag eh_tag; /* Event handler tag. */
16893#else
16894static struct cdev *dtrace_dev;
16895static struct cdev *helper_dev;
16896#endif
16897
16898void dtrace_invop_init(void);
16899void dtrace_invop_uninit(void);
16900
16901static struct cdevsw dtrace_cdevsw = {
16902 .d_version = D_VERSION,
16903#if __FreeBSD_version < 800039
16904 .d_flags = D_TRACKCLOSE | D_NEEDMINOR,
16905 .d_close = dtrace_close,
16906#endif
16907 .d_ioctl = dtrace_ioctl,
16908 .d_open = dtrace_open,
16909 .d_name = "dtrace",
16910};
16911
16912static struct cdevsw helper_cdevsw = {
16913 .d_version = D_VERSION,
16914 .d_ioctl = dtrace_ioctl_helper,
16915 .d_name = "helper",
16916};
16917
16918#include <dtrace_anon.c>
16919#if __FreeBSD_version < 800039
16920#include <dtrace_clone.c>
16921#endif
16922#include <dtrace_ioctl.c>
16923#include <dtrace_load.c>
16924#include <dtrace_modevent.c>
16925#include <dtrace_sysctl.c>
16926#include <dtrace_unload.c>
16927#include <dtrace_vtime.c>
16928#include <dtrace_hacks.c>
16929#include <dtrace_isa.c>
16930
16931SYSINIT(dtrace_load, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_load, NULL);
16932SYSUNINIT(dtrace_unload, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_unload, NULL);
16933SYSINIT(dtrace_anon_init, SI_SUB_DTRACE_ANON, SI_ORDER_FIRST, dtrace_anon_init, NULL);
16934
16935DEV_MODULE(dtrace, dtrace_modevent, NULL);
16936MODULE_VERSION(dtrace, 1);
16937MODULE_DEPEND(dtrace, cyclic, 1, 1, 1);
16938MODULE_DEPEND(dtrace, opensolaris, 1, 1, 1);
16939#endif