22 */
23
24/*
25 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
26 * Use is subject to license terms.
27 */
28
29#pragma ident "%Z%%M% %I% %E% SMI"
30
31/*
32 * DTrace - Dynamic Tracing for Solaris
33 *
34 * This is the implementation of the Solaris Dynamic Tracing framework
35 * (DTrace). The user-visible interface to DTrace is described at length in
36 * the "Solaris Dynamic Tracing Guide". The interfaces between the libdtrace
37 * library, the in-kernel DTrace framework, and the DTrace providers are
38 * described in the block comments in the <sys/dtrace.h> header file. The
39 * internal architecture of DTrace is described in the block comments in the
40 * <sys/dtrace_impl.h> header file. The comments contained within the DTrace
41 * implementation very much assume mastery of all of these sources; if one has
42 * an unanswered question about the implementation, one should consult them
43 * first.
44 *
45 * The functions here are ordered roughly as follows:
46 *
47 * - Probe context functions
48 * - Probe hashing functions
49 * - Non-probe context utility functions
50 * - Matching functions
51 * - Provider-to-Framework API functions
52 * - Probe management functions
53 * - DIF object functions
54 * - Format functions
55 * - Predicate functions
56 * - ECB functions
57 * - Buffer functions
58 * - Enabling functions
59 * - DOF functions
60 * - Anonymous enabling functions
61 * - Consumer state functions
62 * - Helper functions
63 * - Hook functions
64 * - Driver cookbook functions
65 *
66 * Each group of functions begins with a block comment labelled the "DTrace
67 * [Group] Functions", allowing one to find each block by searching forward
68 * on capital-f functions.
69 */
70#include <sys/errno.h>
71#if !defined(sun)
72#include <sys/time.h>
73#endif
74#include <sys/stat.h>
75#include <sys/modctl.h>
76#include <sys/conf.h>
77#include <sys/systm.h>
78#if defined(sun)
79#include <sys/ddi.h>
80#include <sys/sunddi.h>
81#endif
82#include <sys/cpuvar.h>
83#include <sys/kmem.h>
84#if defined(sun)
85#include <sys/strsubr.h>
86#endif
87#include <sys/sysmacros.h>
88#include <sys/dtrace_impl.h>
89#include <sys/atomic.h>
90#include <sys/cmn_err.h>
91#if defined(sun)
92#include <sys/mutex_impl.h>
93#include <sys/rwlock_impl.h>
94#endif
95#include <sys/ctf_api.h>
96#if defined(sun)
97#include <sys/panic.h>
98#include <sys/priv_impl.h>
99#endif
100#include <sys/policy.h>
101#if defined(sun)
102#include <sys/cred_impl.h>
103#include <sys/procfs_isa.h>
104#endif
105#include <sys/taskq.h>
106#if defined(sun)
107#include <sys/mkdev.h>
108#include <sys/kdi.h>
109#endif
110#include <sys/zone.h>
111#include <sys/socket.h>
112#include <netinet/in.h>
113
114/* FreeBSD includes: */
115#if !defined(sun)
116#include <sys/callout.h>
117#include <sys/ctype.h>
118#include <sys/limits.h>
119#include <sys/kdb.h>
120#include <sys/kernel.h>
121#include <sys/malloc.h>
122#include <sys/sysctl.h>
123#include <sys/lock.h>
124#include <sys/mutex.h>
125#include <sys/rwlock.h>
126#include <sys/sx.h>
127#include <sys/dtrace_bsd.h>
128#include <netinet/in.h>
129#include "dtrace_cddl.h"
130#include "dtrace_debug.c"
131#endif
132
133/*
134 * DTrace Tunable Variables
135 *
136 * The following variables may be tuned by adding a line to /etc/system that
137 * includes both the name of the DTrace module ("dtrace") and the name of the
138 * variable. For example:
139 *
140 * set dtrace:dtrace_destructive_disallow = 1
141 *
142 * In general, the only variables that one should be tuning this way are those
143 * that affect system-wide DTrace behavior, and for which the default behavior
144 * is undesirable. Most of these variables are tunable on a per-consumer
145 * basis using DTrace options, and need not be tuned on a system-wide basis.
146 * When tuning these variables, avoid pathological values; while some attempt
147 * is made to verify the integrity of these variables, they are not considered
148 * part of the supported interface to DTrace, and they are therefore not
149 * checked comprehensively. Further, these variables should not be tuned
150 * dynamically via "mdb -kw" or other means; they should only be tuned via
151 * /etc/system.
152 */
153int dtrace_destructive_disallow = 0;
154dtrace_optval_t dtrace_nonroot_maxsize = (16 * 1024 * 1024);
155size_t dtrace_difo_maxsize = (256 * 1024);
156dtrace_optval_t dtrace_dof_maxsize = (256 * 1024);
157size_t dtrace_global_maxsize = (16 * 1024);
158size_t dtrace_actions_max = (16 * 1024);
159size_t dtrace_retain_max = 1024;
160dtrace_optval_t dtrace_helper_actions_max = 32;
161dtrace_optval_t dtrace_helper_providers_max = 32;
162dtrace_optval_t dtrace_dstate_defsize = (1 * 1024 * 1024);
163size_t dtrace_strsize_default = 256;
164dtrace_optval_t dtrace_cleanrate_default = 9900990; /* 101 hz */
165dtrace_optval_t dtrace_cleanrate_min = 200000; /* 5000 hz */
166dtrace_optval_t dtrace_cleanrate_max = (uint64_t)60 * NANOSEC; /* 1/minute */
167dtrace_optval_t dtrace_aggrate_default = NANOSEC; /* 1 hz */
168dtrace_optval_t dtrace_statusrate_default = NANOSEC; /* 1 hz */
169dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC; /* 6/minute */
170dtrace_optval_t dtrace_switchrate_default = NANOSEC; /* 1 hz */
171dtrace_optval_t dtrace_nspec_default = 1;
172dtrace_optval_t dtrace_specsize_default = 32 * 1024;
173dtrace_optval_t dtrace_stackframes_default = 20;
174dtrace_optval_t dtrace_ustackframes_default = 20;
175dtrace_optval_t dtrace_jstackframes_default = 50;
176dtrace_optval_t dtrace_jstackstrsize_default = 512;
177int dtrace_msgdsize_max = 128;
178hrtime_t dtrace_chill_max = 500 * (NANOSEC / MILLISEC); /* 500 ms */
179hrtime_t dtrace_chill_interval = NANOSEC; /* 1000 ms */
180int dtrace_devdepth_max = 32;
181int dtrace_err_verbose;
182hrtime_t dtrace_deadman_interval = NANOSEC;
183hrtime_t dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC;
184hrtime_t dtrace_deadman_user = (hrtime_t)30 * NANOSEC;
185
186/*
187 * DTrace External Variables
188 *
189 * As dtrace(7D) is a kernel module, any DTrace variables are obviously
190 * available to DTrace consumers via the backtick (`) syntax. One of these,
191 * dtrace_zero, is made deliberately so: it is provided as a source of
192 * well-known, zero-filled memory. While this variable is not documented,
193 * it is used by some translators as an implementation detail.
194 */
195const char dtrace_zero[256] = { 0 }; /* zero-filled memory */
196
197/*
198 * DTrace Internal Variables
199 */
200#if defined(sun)
201static dev_info_t *dtrace_devi; /* device info */
202#endif
203#if defined(sun)
204static vmem_t *dtrace_arena; /* probe ID arena */
205static vmem_t *dtrace_minor; /* minor number arena */
206static taskq_t *dtrace_taskq; /* task queue */
207#else
208static struct unrhdr *dtrace_arena; /* Probe ID number. */
209#endif
210static dtrace_probe_t **dtrace_probes; /* array of all probes */
211static int dtrace_nprobes; /* number of probes */
212static dtrace_provider_t *dtrace_provider; /* provider list */
213static dtrace_meta_t *dtrace_meta_pid; /* user-land meta provider */
214static int dtrace_opens; /* number of opens */
215static int dtrace_helpers; /* number of helpers */
216#if defined(sun)
217static void *dtrace_softstate; /* softstate pointer */
218#endif
219static dtrace_hash_t *dtrace_bymod; /* probes hashed by module */
220static dtrace_hash_t *dtrace_byfunc; /* probes hashed by function */
221static dtrace_hash_t *dtrace_byname; /* probes hashed by name */
222static dtrace_toxrange_t *dtrace_toxrange; /* toxic range array */
223static int dtrace_toxranges; /* number of toxic ranges */
224static int dtrace_toxranges_max; /* size of toxic range array */
225static dtrace_anon_t dtrace_anon; /* anonymous enabling */
226static kmem_cache_t *dtrace_state_cache; /* cache for dynamic state */
227static uint64_t dtrace_vtime_references; /* number of vtimestamp refs */
228static kthread_t *dtrace_panicked; /* panicking thread */
229static dtrace_ecb_t *dtrace_ecb_create_cache; /* cached created ECB */
230static dtrace_genid_t dtrace_probegen; /* current probe generation */
231static dtrace_helpers_t *dtrace_deferred_pid; /* deferred helper list */
232static dtrace_enabling_t *dtrace_retained; /* list of retained enablings */
233static dtrace_dynvar_t dtrace_dynhash_sink; /* end of dynamic hash chains */
234#if !defined(sun)
235static struct mtx dtrace_unr_mtx;
236MTX_SYSINIT(dtrace_unr_mtx, &dtrace_unr_mtx, "Unique resource identifier", MTX_DEF);
237int dtrace_in_probe; /* non-zero if executing a probe */
238#if defined(__i386__) || defined(__amd64__)
239uintptr_t dtrace_in_probe_addr; /* Address of invop when already in probe */
240#endif
241#endif
242
243/*
244 * DTrace Locking
245 * DTrace is protected by three (relatively coarse-grained) locks:
246 *
247 * (1) dtrace_lock is required to manipulate essentially any DTrace state,
248 * including enabling state, probes, ECBs, consumer state, helper state,
249 * etc. Importantly, dtrace_lock is _not_ required when in probe context;
250 * probe context is lock-free -- synchronization is handled via the
251 * dtrace_sync() cross call mechanism.
252 *
253 * (2) dtrace_provider_lock is required when manipulating provider state, or
254 * when provider state must be held constant.
255 *
256 * (3) dtrace_meta_lock is required when manipulating meta provider state, or
257 * when meta provider state must be held constant.
258 *
259 * The lock ordering between these three locks is dtrace_meta_lock before
260 * dtrace_provider_lock before dtrace_lock. (In particular, there are
261 * several places where dtrace_provider_lock is held by the framework as it
262 * calls into the providers -- which then call back into the framework,
263 * grabbing dtrace_lock.)
264 *
265 * There are two other locks in the mix: mod_lock and cpu_lock. With respect
266 * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical
267 * role as a coarse-grained lock; it is acquired before both of these locks.
268 * With respect to dtrace_meta_lock, its behavior is stranger: cpu_lock must
269 * be acquired _between_ dtrace_meta_lock and any other DTrace locks.
270 * mod_lock is similar with respect to dtrace_provider_lock in that it must be
271 * acquired _between_ dtrace_provider_lock and dtrace_lock.
272 */
273static kmutex_t dtrace_lock; /* probe state lock */
274static kmutex_t dtrace_provider_lock; /* provider state lock */
275static kmutex_t dtrace_meta_lock; /* meta-provider state lock */
276
277#if !defined(sun)
278/* XXX FreeBSD hacks. */
279static kmutex_t mod_lock;
280
281#define cr_suid cr_svuid
282#define cr_sgid cr_svgid
283#define ipaddr_t in_addr_t
284#define mod_modname pathname
285#define vuprintf vprintf
286#define ttoproc(_a) ((_a)->td_proc)
287#define crgetzoneid(_a) 0
288#define NCPU MAXCPU
289#define SNOCD 0
290#define CPU_ON_INTR(_a) 0
291
292#define PRIV_EFFECTIVE (1 << 0)
293#define PRIV_DTRACE_KERNEL (1 << 1)
294#define PRIV_DTRACE_PROC (1 << 2)
295#define PRIV_DTRACE_USER (1 << 3)
296#define PRIV_PROC_OWNER (1 << 4)
297#define PRIV_PROC_ZONE (1 << 5)
298#define PRIV_ALL ~0
299
300SYSCTL_NODE(_debug, OID_AUTO, dtrace, CTLFLAG_RD, 0, "DTrace Information");
301#endif
302
303#if defined(sun)
304#define curcpu CPU->cpu_id
305#endif
306
307
308/*
309 * DTrace Provider Variables
310 *
311 * These are the variables relating to DTrace as a provider (that is, the
312 * provider of the BEGIN, END, and ERROR probes).
313 */
314static dtrace_pattr_t dtrace_provider_attr = {
315{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
316{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
317{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
318{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
319{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON },
320};
321
322static void
323dtrace_nullop(void)
324{}
325
326static dtrace_pops_t dtrace_provider_ops = {
327 (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop,
328 (void (*)(void *, modctl_t *))dtrace_nullop,
329 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
330 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
331 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
332 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop,
333 NULL,
334 NULL,
335 NULL,
336 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop
337};
338
339static dtrace_id_t dtrace_probeid_begin; /* special BEGIN probe */
340static dtrace_id_t dtrace_probeid_end; /* special END probe */
341dtrace_id_t dtrace_probeid_error; /* special ERROR probe */
342
343/*
344 * DTrace Helper Tracing Variables
345 */
346uint32_t dtrace_helptrace_next = 0;
347uint32_t dtrace_helptrace_nlocals;
348char *dtrace_helptrace_buffer;
349int dtrace_helptrace_bufsize = 512 * 1024;
350
351#ifdef DEBUG
352int dtrace_helptrace_enabled = 1;
353#else
354int dtrace_helptrace_enabled = 0;
355#endif
356
357/*
358 * DTrace Error Hashing
359 *
360 * On DEBUG kernels, DTrace will track the errors that has seen in a hash
361 * table. This is very useful for checking coverage of tests that are
362 * expected to induce DIF or DOF processing errors, and may be useful for
363 * debugging problems in the DIF code generator or in DOF generation . The
364 * error hash may be examined with the ::dtrace_errhash MDB dcmd.
365 */
366#ifdef DEBUG
367static dtrace_errhash_t dtrace_errhash[DTRACE_ERRHASHSZ];
368static const char *dtrace_errlast;
369static kthread_t *dtrace_errthread;
370static kmutex_t dtrace_errlock;
371#endif
372
373/*
374 * DTrace Macros and Constants
375 *
376 * These are various macros that are useful in various spots in the
377 * implementation, along with a few random constants that have no meaning
378 * outside of the implementation. There is no real structure to this cpp
379 * mishmash -- but is there ever?
380 */
381#define DTRACE_HASHSTR(hash, probe) \
382 dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs)))
383
384#define DTRACE_HASHNEXT(hash, probe) \
385 (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs)
386
387#define DTRACE_HASHPREV(hash, probe) \
388 (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs)
389
390#define DTRACE_HASHEQ(hash, lhs, rhs) \
391 (strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \
392 *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0)
393
394#define DTRACE_AGGHASHSIZE_SLEW 17
395
396#define DTRACE_V4MAPPED_OFFSET (sizeof (uint32_t) * 3)
397
398/*
399 * The key for a thread-local variable consists of the lower 61 bits of the
400 * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL.
401 * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never
402 * equal to a variable identifier. This is necessary (but not sufficient) to
403 * assure that global associative arrays never collide with thread-local
404 * variables. To guarantee that they cannot collide, we must also define the
405 * order for keying dynamic variables. That order is:
406 *
407 * [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ]
408 *
409 * Because the variable-key and the tls-key are in orthogonal spaces, there is
410 * no way for a global variable key signature to match a thread-local key
411 * signature.
412 */
413#if defined(sun)
414#define DTRACE_TLS_THRKEY(where) { \
415 uint_t intr = 0; \
416 uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \
417 for (; actv; actv >>= 1) \
418 intr++; \
419 ASSERT(intr < (1 << 3)); \
420 (where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \
421 (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \
422}
423#else
424#define DTRACE_TLS_THRKEY(where) { \
425 solaris_cpu_t *_c = &solaris_cpu[curcpu]; \
426 uint_t intr = 0; \
427 uint_t actv = _c->cpu_intr_actv; \
428 for (; actv; actv >>= 1) \
429 intr++; \
430 ASSERT(intr < (1 << 3)); \
431 (where) = ((curthread->td_tid + DIF_VARIABLE_MAX) & \
432 (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \
433}
434#endif
435
436#define DT_BSWAP_8(x) ((x) & 0xff)
437#define DT_BSWAP_16(x) ((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8))
438#define DT_BSWAP_32(x) ((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16))
439#define DT_BSWAP_64(x) ((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32))
440
441#define DT_MASK_LO 0x00000000FFFFFFFFULL
442
443#define DTRACE_STORE(type, tomax, offset, what) \
444 *((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what);
445
446#ifndef __i386
447#define DTRACE_ALIGNCHECK(addr, size, flags) \
448 if (addr & (size - 1)) { \
449 *flags |= CPU_DTRACE_BADALIGN; \
450 cpu_core[curcpu].cpuc_dtrace_illval = addr; \
451 return (0); \
452 }
453#else
454#define DTRACE_ALIGNCHECK(addr, size, flags)
455#endif
456
457/*
458 * Test whether a range of memory starting at testaddr of size testsz falls
459 * within the range of memory described by addr, sz. We take care to avoid
460 * problems with overflow and underflow of the unsigned quantities, and
461 * disallow all negative sizes. Ranges of size 0 are allowed.
462 */
463#define DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \
464 ((testaddr) - (baseaddr) < (basesz) && \
465 (testaddr) + (testsz) - (baseaddr) <= (basesz) && \
466 (testaddr) + (testsz) >= (testaddr))
467
468/*
469 * Test whether alloc_sz bytes will fit in the scratch region. We isolate
470 * alloc_sz on the righthand side of the comparison in order to avoid overflow
471 * or underflow in the comparison with it. This is simpler than the INRANGE
472 * check above, because we know that the dtms_scratch_ptr is valid in the
473 * range. Allocations of size zero are allowed.
474 */
475#define DTRACE_INSCRATCH(mstate, alloc_sz) \
476 ((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \
477 (mstate)->dtms_scratch_ptr >= (alloc_sz))
478
479#define DTRACE_LOADFUNC(bits) \
480/*CSTYLED*/ \
481uint##bits##_t \
482dtrace_load##bits(uintptr_t addr) \
483{ \
484 size_t size = bits / NBBY; \
485 /*CSTYLED*/ \
486 uint##bits##_t rval; \
487 int i; \
488 volatile uint16_t *flags = (volatile uint16_t *) \
489 &cpu_core[curcpu].cpuc_dtrace_flags; \
490 \
491 DTRACE_ALIGNCHECK(addr, size, flags); \
492 \
493 for (i = 0; i < dtrace_toxranges; i++) { \
494 if (addr >= dtrace_toxrange[i].dtt_limit) \
495 continue; \
496 \
497 if (addr + size <= dtrace_toxrange[i].dtt_base) \
498 continue; \
499 \
500 /* \
501 * This address falls within a toxic region; return 0. \
502 */ \
503 *flags |= CPU_DTRACE_BADADDR; \
504 cpu_core[curcpu].cpuc_dtrace_illval = addr; \
505 return (0); \
506 } \
507 \
508 *flags |= CPU_DTRACE_NOFAULT; \
509 /*CSTYLED*/ \
510 rval = *((volatile uint##bits##_t *)addr); \
511 *flags &= ~CPU_DTRACE_NOFAULT; \
512 \
513 return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0); \
514}
515
516#ifdef _LP64
517#define dtrace_loadptr dtrace_load64
518#else
519#define dtrace_loadptr dtrace_load32
520#endif
521
522#define DTRACE_DYNHASH_FREE 0
523#define DTRACE_DYNHASH_SINK 1
524#define DTRACE_DYNHASH_VALID 2
525
526#define DTRACE_MATCH_NEXT 0
527#define DTRACE_MATCH_DONE 1
528#define DTRACE_ANCHORED(probe) ((probe)->dtpr_func[0] != '\0')
529#define DTRACE_STATE_ALIGN 64
530
531#define DTRACE_FLAGS2FLT(flags) \
532 (((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR : \
533 ((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP : \
534 ((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO : \
535 ((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV : \
536 ((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV : \
537 ((flags) & CPU_DTRACE_TUPOFLOW) ? DTRACEFLT_TUPOFLOW : \
538 ((flags) & CPU_DTRACE_BADALIGN) ? DTRACEFLT_BADALIGN : \
539 ((flags) & CPU_DTRACE_NOSCRATCH) ? DTRACEFLT_NOSCRATCH : \
540 ((flags) & CPU_DTRACE_BADSTACK) ? DTRACEFLT_BADSTACK : \
541 DTRACEFLT_UNKNOWN)
542
543#define DTRACEACT_ISSTRING(act) \
544 ((act)->dta_kind == DTRACEACT_DIFEXPR && \
545 (act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING)
546
547/* Function prototype definitions: */
548static size_t dtrace_strlen(const char *, size_t);
549static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id);
550static void dtrace_enabling_provide(dtrace_provider_t *);
551static int dtrace_enabling_match(dtrace_enabling_t *, int *);
552static void dtrace_enabling_matchall(void);
553static dtrace_state_t *dtrace_anon_grab(void);
554#if defined(sun)
555static uint64_t dtrace_helper(int, dtrace_mstate_t *,
556 dtrace_state_t *, uint64_t, uint64_t);
557static dtrace_helpers_t *dtrace_helpers_create(proc_t *);
558#endif
559static void dtrace_buffer_drop(dtrace_buffer_t *);
560static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t,
561 dtrace_state_t *, dtrace_mstate_t *);
562static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t,
563 dtrace_optval_t);
564static int dtrace_ecb_create_enable(dtrace_probe_t *, void *);
565#if defined(sun)
566static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *);
567#endif
568uint16_t dtrace_load16(uintptr_t);
569uint32_t dtrace_load32(uintptr_t);
570uint64_t dtrace_load64(uintptr_t);
571uint8_t dtrace_load8(uintptr_t);
572void dtrace_dynvar_clean(dtrace_dstate_t *);
573dtrace_dynvar_t *dtrace_dynvar(dtrace_dstate_t *, uint_t, dtrace_key_t *,
574 size_t, dtrace_dynvar_op_t, dtrace_mstate_t *, dtrace_vstate_t *);
575uintptr_t dtrace_dif_varstr(uintptr_t, dtrace_state_t *, dtrace_mstate_t *);
576
577/*
578 * DTrace Probe Context Functions
579 *
580 * These functions are called from probe context. Because probe context is
581 * any context in which C may be called, arbitrarily locks may be held,
582 * interrupts may be disabled, we may be in arbitrary dispatched state, etc.
583 * As a result, functions called from probe context may only call other DTrace
584 * support functions -- they may not interact at all with the system at large.
585 * (Note that the ASSERT macro is made probe-context safe by redefining it in
586 * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary
587 * loads are to be performed from probe context, they _must_ be in terms of
588 * the safe dtrace_load*() variants.
589 *
590 * Some functions in this block are not actually called from probe context;
591 * for these functions, there will be a comment above the function reading
592 * "Note: not called from probe context."
593 */
594void
595dtrace_panic(const char *format, ...)
596{
597 va_list alist;
598
599 va_start(alist, format);
600 dtrace_vpanic(format, alist);
601 va_end(alist);
602}
603
604int
605dtrace_assfail(const char *a, const char *f, int l)
606{
607 dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l);
608
609 /*
610 * We just need something here that even the most clever compiler
611 * cannot optimize away.
612 */
613 return (a[(uintptr_t)f]);
614}
615
616/*
617 * Atomically increment a specified error counter from probe context.
618 */
619static void
620dtrace_error(uint32_t *counter)
621{
622 /*
623 * Most counters stored to in probe context are per-CPU counters.
624 * However, there are some error conditions that are sufficiently
625 * arcane that they don't merit per-CPU storage. If these counters
626 * are incremented concurrently on different CPUs, scalability will be
627 * adversely affected -- but we don't expect them to be white-hot in a
628 * correctly constructed enabling...
629 */
630 uint32_t oval, nval;
631
632 do {
633 oval = *counter;
634
635 if ((nval = oval + 1) == 0) {
636 /*
637 * If the counter would wrap, set it to 1 -- assuring
638 * that the counter is never zero when we have seen
639 * errors. (The counter must be 32-bits because we
640 * aren't guaranteed a 64-bit compare&swap operation.)
641 * To save this code both the infamy of being fingered
642 * by a priggish news story and the indignity of being
643 * the target of a neo-puritan witch trial, we're
644 * carefully avoiding any colorful description of the
645 * likelihood of this condition -- but suffice it to
646 * say that it is only slightly more likely than the
647 * overflow of predicate cache IDs, as discussed in
648 * dtrace_predicate_create().
649 */
650 nval = 1;
651 }
652 } while (dtrace_cas32(counter, oval, nval) != oval);
653}
654
655/*
656 * Use the DTRACE_LOADFUNC macro to define functions for each of loading a
657 * uint8_t, a uint16_t, a uint32_t and a uint64_t.
658 */
659DTRACE_LOADFUNC(8)
660DTRACE_LOADFUNC(16)
661DTRACE_LOADFUNC(32)
662DTRACE_LOADFUNC(64)
663
664static int
665dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate)
666{
667 if (dest < mstate->dtms_scratch_base)
668 return (0);
669
670 if (dest + size < dest)
671 return (0);
672
673 if (dest + size > mstate->dtms_scratch_ptr)
674 return (0);
675
676 return (1);
677}
678
679static int
680dtrace_canstore_statvar(uint64_t addr, size_t sz,
681 dtrace_statvar_t **svars, int nsvars)
682{
683 int i;
684
685 for (i = 0; i < nsvars; i++) {
686 dtrace_statvar_t *svar = svars[i];
687
688 if (svar == NULL || svar->dtsv_size == 0)
689 continue;
690
691 if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size))
692 return (1);
693 }
694
695 return (0);
696}
697
698/*
699 * Check to see if the address is within a memory region to which a store may
700 * be issued. This includes the DTrace scratch areas, and any DTrace variable
701 * region. The caller of dtrace_canstore() is responsible for performing any
702 * alignment checks that are needed before stores are actually executed.
703 */
704static int
705dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
706 dtrace_vstate_t *vstate)
707{
708 /*
709 * First, check to see if the address is in scratch space...
710 */
711 if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base,
712 mstate->dtms_scratch_size))
713 return (1);
714
715 /*
716 * Now check to see if it's a dynamic variable. This check will pick
717 * up both thread-local variables and any global dynamically-allocated
718 * variables.
719 */
720 if (DTRACE_INRANGE(addr, sz, (uintptr_t)vstate->dtvs_dynvars.dtds_base,
721 vstate->dtvs_dynvars.dtds_size)) {
722 dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
723 uintptr_t base = (uintptr_t)dstate->dtds_base +
724 (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t));
725 uintptr_t chunkoffs;
726
727 /*
728 * Before we assume that we can store here, we need to make
729 * sure that it isn't in our metadata -- storing to our
730 * dynamic variable metadata would corrupt our state. For
731 * the range to not include any dynamic variable metadata,
732 * it must:
733 *
734 * (1) Start above the hash table that is at the base of
735 * the dynamic variable space
736 *
737 * (2) Have a starting chunk offset that is beyond the
738 * dtrace_dynvar_t that is at the base of every chunk
739 *
740 * (3) Not span a chunk boundary
741 *
742 */
743 if (addr < base)
744 return (0);
745
746 chunkoffs = (addr - base) % dstate->dtds_chunksize;
747
748 if (chunkoffs < sizeof (dtrace_dynvar_t))
749 return (0);
750
751 if (chunkoffs + sz > dstate->dtds_chunksize)
752 return (0);
753
754 return (1);
755 }
756
757 /*
758 * Finally, check the static local and global variables. These checks
759 * take the longest, so we perform them last.
760 */
761 if (dtrace_canstore_statvar(addr, sz,
762 vstate->dtvs_locals, vstate->dtvs_nlocals))
763 return (1);
764
765 if (dtrace_canstore_statvar(addr, sz,
766 vstate->dtvs_globals, vstate->dtvs_nglobals))
767 return (1);
768
769 return (0);
770}
771
772
773/*
774 * Convenience routine to check to see if the address is within a memory
775 * region in which a load may be issued given the user's privilege level;
776 * if not, it sets the appropriate error flags and loads 'addr' into the
777 * illegal value slot.
778 *
779 * DTrace subroutines (DIF_SUBR_*) should use this helper to implement
780 * appropriate memory access protection.
781 */
782static int
783dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
784 dtrace_vstate_t *vstate)
785{
786 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
787
788 /*
789 * If we hold the privilege to read from kernel memory, then
790 * everything is readable.
791 */
792 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
793 return (1);
794
795 /*
796 * You can obviously read that which you can store.
797 */
798 if (dtrace_canstore(addr, sz, mstate, vstate))
799 return (1);
800
801 /*
802 * We're allowed to read from our own string table.
803 */
804 if (DTRACE_INRANGE(addr, sz, (uintptr_t)mstate->dtms_difo->dtdo_strtab,
805 mstate->dtms_difo->dtdo_strlen))
806 return (1);
807
808 DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV);
809 *illval = addr;
810 return (0);
811}
812
813/*
814 * Convenience routine to check to see if a given string is within a memory
815 * region in which a load may be issued given the user's privilege level;
816 * this exists so that we don't need to issue unnecessary dtrace_strlen()
817 * calls in the event that the user has all privileges.
818 */
819static int
820dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate,
821 dtrace_vstate_t *vstate)
822{
823 size_t strsz;
824
825 /*
826 * If we hold the privilege to read from kernel memory, then
827 * everything is readable.
828 */
829 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
830 return (1);
831
832 strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz);
833 if (dtrace_canload(addr, strsz, mstate, vstate))
834 return (1);
835
836 return (0);
837}
838
839/*
840 * Convenience routine to check to see if a given variable is within a memory
841 * region in which a load may be issued given the user's privilege level.
842 */
843static int
844dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate,
845 dtrace_vstate_t *vstate)
846{
847 size_t sz;
848 ASSERT(type->dtdt_flags & DIF_TF_BYREF);
849
850 /*
851 * If we hold the privilege to read from kernel memory, then
852 * everything is readable.
853 */
854 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
855 return (1);
856
857 if (type->dtdt_kind == DIF_TYPE_STRING)
858 sz = dtrace_strlen(src,
859 vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1;
860 else
861 sz = type->dtdt_size;
862
863 return (dtrace_canload((uintptr_t)src, sz, mstate, vstate));
864}
865
866/*
867 * Compare two strings using safe loads.
868 */
869static int
870dtrace_strncmp(char *s1, char *s2, size_t limit)
871{
872 uint8_t c1, c2;
873 volatile uint16_t *flags;
874
875 if (s1 == s2 || limit == 0)
876 return (0);
877
878 flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
879
880 do {
881 if (s1 == NULL) {
882 c1 = '\0';
883 } else {
884 c1 = dtrace_load8((uintptr_t)s1++);
885 }
886
887 if (s2 == NULL) {
888 c2 = '\0';
889 } else {
890 c2 = dtrace_load8((uintptr_t)s2++);
891 }
892
893 if (c1 != c2)
894 return (c1 - c2);
895 } while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT));
896
897 return (0);
898}
899
900/*
901 * Compute strlen(s) for a string using safe memory accesses. The additional
902 * len parameter is used to specify a maximum length to ensure completion.
903 */
904static size_t
905dtrace_strlen(const char *s, size_t lim)
906{
907 uint_t len;
908
909 for (len = 0; len != lim; len++) {
910 if (dtrace_load8((uintptr_t)s++) == '\0')
911 break;
912 }
913
914 return (len);
915}
916
917/*
918 * Check if an address falls within a toxic region.
919 */
920static int
921dtrace_istoxic(uintptr_t kaddr, size_t size)
922{
923 uintptr_t taddr, tsize;
924 int i;
925
926 for (i = 0; i < dtrace_toxranges; i++) {
927 taddr = dtrace_toxrange[i].dtt_base;
928 tsize = dtrace_toxrange[i].dtt_limit - taddr;
929
930 if (kaddr - taddr < tsize) {
931 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
932 cpu_core[curcpu].cpuc_dtrace_illval = kaddr;
933 return (1);
934 }
935
936 if (taddr - kaddr < size) {
937 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
938 cpu_core[curcpu].cpuc_dtrace_illval = taddr;
939 return (1);
940 }
941 }
942
943 return (0);
944}
945
946/*
947 * Copy src to dst using safe memory accesses. The src is assumed to be unsafe
948 * memory specified by the DIF program. The dst is assumed to be safe memory
949 * that we can store to directly because it is managed by DTrace. As with
950 * standard bcopy, overlapping copies are handled properly.
951 */
952static void
953dtrace_bcopy(const void *src, void *dst, size_t len)
954{
955 if (len != 0) {
956 uint8_t *s1 = dst;
957 const uint8_t *s2 = src;
958
959 if (s1 <= s2) {
960 do {
961 *s1++ = dtrace_load8((uintptr_t)s2++);
962 } while (--len != 0);
963 } else {
964 s2 += len;
965 s1 += len;
966
967 do {
968 *--s1 = dtrace_load8((uintptr_t)--s2);
969 } while (--len != 0);
970 }
971 }
972}
973
974/*
975 * Copy src to dst using safe memory accesses, up to either the specified
976 * length, or the point that a nul byte is encountered. The src is assumed to
977 * be unsafe memory specified by the DIF program. The dst is assumed to be
978 * safe memory that we can store to directly because it is managed by DTrace.
979 * Unlike dtrace_bcopy(), overlapping regions are not handled.
980 */
981static void
982dtrace_strcpy(const void *src, void *dst, size_t len)
983{
984 if (len != 0) {
985 uint8_t *s1 = dst, c;
986 const uint8_t *s2 = src;
987
988 do {
989 *s1++ = c = dtrace_load8((uintptr_t)s2++);
990 } while (--len != 0 && c != '\0');
991 }
992}
993
994/*
995 * Copy src to dst, deriving the size and type from the specified (BYREF)
996 * variable type. The src is assumed to be unsafe memory specified by the DIF
997 * program. The dst is assumed to be DTrace variable memory that is of the
998 * specified type; we assume that we can store to directly.
999 */
1000static void
1001dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type)
1002{
1003 ASSERT(type->dtdt_flags & DIF_TF_BYREF);
1004
1005 if (type->dtdt_kind == DIF_TYPE_STRING) {
1006 dtrace_strcpy(src, dst, type->dtdt_size);
1007 } else {
1008 dtrace_bcopy(src, dst, type->dtdt_size);
1009 }
1010}
1011
1012/*
1013 * Compare s1 to s2 using safe memory accesses. The s1 data is assumed to be
1014 * unsafe memory specified by the DIF program. The s2 data is assumed to be
1015 * safe memory that we can access directly because it is managed by DTrace.
1016 */
1017static int
1018dtrace_bcmp(const void *s1, const void *s2, size_t len)
1019{
1020 volatile uint16_t *flags;
1021
1022 flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags;
1023
1024 if (s1 == s2)
1025 return (0);
1026
1027 if (s1 == NULL || s2 == NULL)
1028 return (1);
1029
1030 if (s1 != s2 && len != 0) {
1031 const uint8_t *ps1 = s1;
1032 const uint8_t *ps2 = s2;
1033
1034 do {
1035 if (dtrace_load8((uintptr_t)ps1++) != *ps2++)
1036 return (1);
1037 } while (--len != 0 && !(*flags & CPU_DTRACE_FAULT));
1038 }
1039 return (0);
1040}
1041
1042/*
1043 * Zero the specified region using a simple byte-by-byte loop. Note that this
1044 * is for safe DTrace-managed memory only.
1045 */
1046static void
1047dtrace_bzero(void *dst, size_t len)
1048{
1049 uchar_t *cp;
1050
1051 for (cp = dst; len != 0; len--)
1052 *cp++ = 0;
1053}
1054
1055static void
1056dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum)
1057{
1058 uint64_t result[2];
1059
1060 result[0] = addend1[0] + addend2[0];
1061 result[1] = addend1[1] + addend2[1] +
1062 (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0);
1063
1064 sum[0] = result[0];
1065 sum[1] = result[1];
1066}
1067
1068/*
1069 * Shift the 128-bit value in a by b. If b is positive, shift left.
1070 * If b is negative, shift right.
1071 */
1072static void
1073dtrace_shift_128(uint64_t *a, int b)
1074{
1075 uint64_t mask;
1076
1077 if (b == 0)
1078 return;
1079
1080 if (b < 0) {
1081 b = -b;
1082 if (b >= 64) {
1083 a[0] = a[1] >> (b - 64);
1084 a[1] = 0;
1085 } else {
1086 a[0] >>= b;
1087 mask = 1LL << (64 - b);
1088 mask -= 1;
1089 a[0] |= ((a[1] & mask) << (64 - b));
1090 a[1] >>= b;
1091 }
1092 } else {
1093 if (b >= 64) {
1094 a[1] = a[0] << (b - 64);
1095 a[0] = 0;
1096 } else {
1097 a[1] <<= b;
1098 mask = a[0] >> (64 - b);
1099 a[1] |= mask;
1100 a[0] <<= b;
1101 }
1102 }
1103}
1104
1105/*
1106 * The basic idea is to break the 2 64-bit values into 4 32-bit values,
1107 * use native multiplication on those, and then re-combine into the
1108 * resulting 128-bit value.
1109 *
1110 * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) =
1111 * hi1 * hi2 << 64 +
1112 * hi1 * lo2 << 32 +
1113 * hi2 * lo1 << 32 +
1114 * lo1 * lo2
1115 */
1116static void
1117dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product)
1118{
1119 uint64_t hi1, hi2, lo1, lo2;
1120 uint64_t tmp[2];
1121
1122 hi1 = factor1 >> 32;
1123 hi2 = factor2 >> 32;
1124
1125 lo1 = factor1 & DT_MASK_LO;
1126 lo2 = factor2 & DT_MASK_LO;
1127
1128 product[0] = lo1 * lo2;
1129 product[1] = hi1 * hi2;
1130
1131 tmp[0] = hi1 * lo2;
1132 tmp[1] = 0;
1133 dtrace_shift_128(tmp, 32);
1134 dtrace_add_128(product, tmp, product);
1135
1136 tmp[0] = hi2 * lo1;
1137 tmp[1] = 0;
1138 dtrace_shift_128(tmp, 32);
1139 dtrace_add_128(product, tmp, product);
1140}
1141
1142/*
1143 * This privilege check should be used by actions and subroutines to
1144 * verify that the user credentials of the process that enabled the
1145 * invoking ECB match the target credentials
1146 */
1147static int
1148dtrace_priv_proc_common_user(dtrace_state_t *state)
1149{
1150 cred_t *cr, *s_cr = state->dts_cred.dcr_cred;
1151
1152 /*
1153 * We should always have a non-NULL state cred here, since if cred
1154 * is null (anonymous tracing), we fast-path bypass this routine.
1155 */
1156 ASSERT(s_cr != NULL);
1157
1158 if ((cr = CRED()) != NULL &&
1159 s_cr->cr_uid == cr->cr_uid &&
1160 s_cr->cr_uid == cr->cr_ruid &&
1161 s_cr->cr_uid == cr->cr_suid &&
1162 s_cr->cr_gid == cr->cr_gid &&
1163 s_cr->cr_gid == cr->cr_rgid &&
1164 s_cr->cr_gid == cr->cr_sgid)
1165 return (1);
1166
1167 return (0);
1168}
1169
1170/*
1171 * This privilege check should be used by actions and subroutines to
1172 * verify that the zone of the process that enabled the invoking ECB
1173 * matches the target credentials
1174 */
1175static int
1176dtrace_priv_proc_common_zone(dtrace_state_t *state)
1177{
1178#if defined(sun)
1179 cred_t *cr, *s_cr = state->dts_cred.dcr_cred;
1180
1181 /*
1182 * We should always have a non-NULL state cred here, since if cred
1183 * is null (anonymous tracing), we fast-path bypass this routine.
1184 */
1185 ASSERT(s_cr != NULL);
1186
1187 if ((cr = CRED()) != NULL &&
1188 s_cr->cr_zone == cr->cr_zone)
1189 return (1);
1190
1191 return (0);
1192#else
1193 return (1);
1194#endif
1195}
1196
1197/*
1198 * This privilege check should be used by actions and subroutines to
1199 * verify that the process has not setuid or changed credentials.
1200 */
1201static int
1202dtrace_priv_proc_common_nocd(void)
1203{
1204 proc_t *proc;
1205
1206 if ((proc = ttoproc(curthread)) != NULL &&
1207 !(proc->p_flag & SNOCD))
1208 return (1);
1209
1210 return (0);
1211}
1212
1213static int
1214dtrace_priv_proc_destructive(dtrace_state_t *state)
1215{
1216 int action = state->dts_cred.dcr_action;
1217
1218 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) &&
1219 dtrace_priv_proc_common_zone(state) == 0)
1220 goto bad;
1221
1222 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) &&
1223 dtrace_priv_proc_common_user(state) == 0)
1224 goto bad;
1225
1226 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) &&
1227 dtrace_priv_proc_common_nocd() == 0)
1228 goto bad;
1229
1230 return (1);
1231
1232bad:
1233 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1234
1235 return (0);
1236}
1237
1238static int
1239dtrace_priv_proc_control(dtrace_state_t *state)
1240{
1241 if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL)
1242 return (1);
1243
1244 if (dtrace_priv_proc_common_zone(state) &&
1245 dtrace_priv_proc_common_user(state) &&
1246 dtrace_priv_proc_common_nocd())
1247 return (1);
1248
1249 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1250
1251 return (0);
1252}
1253
1254static int
1255dtrace_priv_proc(dtrace_state_t *state)
1256{
1257 if (state->dts_cred.dcr_action & DTRACE_CRA_PROC)
1258 return (1);
1259
1260 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV;
1261
1262 return (0);
1263}
1264
1265static int
1266dtrace_priv_kernel(dtrace_state_t *state)
1267{
1268 if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL)
1269 return (1);
1270
1271 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;
1272
1273 return (0);
1274}
1275
1276static int
1277dtrace_priv_kernel_destructive(dtrace_state_t *state)
1278{
1279 if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE)
1280 return (1);
1281
1282 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV;
1283
1284 return (0);
1285}
1286
1287/*
1288 * Note: not called from probe context. This function is called
1289 * asynchronously (and at a regular interval) from outside of probe context to
1290 * clean the dirty dynamic variable lists on all CPUs. Dynamic variable
1291 * cleaning is explained in detail in <sys/dtrace_impl.h>.
1292 */
1293void
1294dtrace_dynvar_clean(dtrace_dstate_t *dstate)
1295{
1296 dtrace_dynvar_t *dirty;
1297 dtrace_dstate_percpu_t *dcpu;
1298 int i, work = 0;
1299
1300 for (i = 0; i < NCPU; i++) {
1301 dcpu = &dstate->dtds_percpu[i];
1302
1303 ASSERT(dcpu->dtdsc_rinsing == NULL);
1304
1305 /*
1306 * If the dirty list is NULL, there is no dirty work to do.
1307 */
1308 if (dcpu->dtdsc_dirty == NULL)
1309 continue;
1310
1311 /*
1312 * If the clean list is non-NULL, then we're not going to do
1313 * any work for this CPU -- it means that there has not been
1314 * a dtrace_dynvar() allocation on this CPU (or from this CPU)
1315 * since the last time we cleaned house.
1316 */
1317 if (dcpu->dtdsc_clean != NULL)
1318 continue;
1319
1320 work = 1;
1321
1322 /*
1323 * Atomically move the dirty list aside.
1324 */
1325 do {
1326 dirty = dcpu->dtdsc_dirty;
1327
1328 /*
1329 * Before we zap the dirty list, set the rinsing list.
1330 * (This allows for a potential assertion in
1331 * dtrace_dynvar(): if a free dynamic variable appears
1332 * on a hash chain, either the dirty list or the
1333 * rinsing list for some CPU must be non-NULL.)
1334 */
1335 dcpu->dtdsc_rinsing = dirty;
1336 dtrace_membar_producer();
1337 } while (dtrace_casptr(&dcpu->dtdsc_dirty,
1338 dirty, NULL) != dirty);
1339 }
1340
1341 if (!work) {
1342 /*
1343 * We have no work to do; we can simply return.
1344 */
1345 return;
1346 }
1347
1348 dtrace_sync();
1349
1350 for (i = 0; i < NCPU; i++) {
1351 dcpu = &dstate->dtds_percpu[i];
1352
1353 if (dcpu->dtdsc_rinsing == NULL)
1354 continue;
1355
1356 /*
1357 * We are now guaranteed that no hash chain contains a pointer
1358 * into this dirty list; we can make it clean.
1359 */
1360 ASSERT(dcpu->dtdsc_clean == NULL);
1361 dcpu->dtdsc_clean = dcpu->dtdsc_rinsing;
1362 dcpu->dtdsc_rinsing = NULL;
1363 }
1364
1365 /*
1366 * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make
1367 * sure that all CPUs have seen all of the dtdsc_clean pointers.
1368 * This prevents a race whereby a CPU incorrectly decides that
1369 * the state should be something other than DTRACE_DSTATE_CLEAN
1370 * after dtrace_dynvar_clean() has completed.
1371 */
1372 dtrace_sync();
1373
1374 dstate->dtds_state = DTRACE_DSTATE_CLEAN;
1375}
1376
1377/*
1378 * Depending on the value of the op parameter, this function looks-up,
1379 * allocates or deallocates an arbitrarily-keyed dynamic variable. If an
1380 * allocation is requested, this function will return a pointer to a
1381 * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no
1382 * variable can be allocated. If NULL is returned, the appropriate counter
1383 * will be incremented.
1384 */
1385dtrace_dynvar_t *
1386dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys,
1387 dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op,
1388 dtrace_mstate_t *mstate, dtrace_vstate_t *vstate)
1389{
1390 uint64_t hashval = DTRACE_DYNHASH_VALID;
1391 dtrace_dynhash_t *hash = dstate->dtds_hash;
1392 dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL;
1393 processorid_t me = curcpu, cpu = me;
1394 dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me];
1395 size_t bucket, ksize;
1396 size_t chunksize = dstate->dtds_chunksize;
1397 uintptr_t kdata, lock, nstate;
1398 uint_t i;
1399
1400 ASSERT(nkeys != 0);
1401
1402 /*
1403 * Hash the key. As with aggregations, we use Jenkins' "One-at-a-time"
1404 * algorithm. For the by-value portions, we perform the algorithm in
1405 * 16-bit chunks (as opposed to 8-bit chunks). This speeds things up a
1406 * bit, and seems to have only a minute effect on distribution. For
1407 * the by-reference data, we perform "One-at-a-time" iterating (safely)
1408 * over each referenced byte. It's painful to do this, but it's much
1409 * better than pathological hash distribution. The efficacy of the
1410 * hashing algorithm (and a comparison with other algorithms) may be
1411 * found by running the ::dtrace_dynstat MDB dcmd.
1412 */
1413 for (i = 0; i < nkeys; i++) {
1414 if (key[i].dttk_size == 0) {
1415 uint64_t val = key[i].dttk_value;
1416
1417 hashval += (val >> 48) & 0xffff;
1418 hashval += (hashval << 10);
1419 hashval ^= (hashval >> 6);
1420
1421 hashval += (val >> 32) & 0xffff;
1422 hashval += (hashval << 10);
1423 hashval ^= (hashval >> 6);
1424
1425 hashval += (val >> 16) & 0xffff;
1426 hashval += (hashval << 10);
1427 hashval ^= (hashval >> 6);
1428
1429 hashval += val & 0xffff;
1430 hashval += (hashval << 10);
1431 hashval ^= (hashval >> 6);
1432 } else {
1433 /*
1434 * This is incredibly painful, but it beats the hell
1435 * out of the alternative.
1436 */
1437 uint64_t j, size = key[i].dttk_size;
1438 uintptr_t base = (uintptr_t)key[i].dttk_value;
1439
1440 if (!dtrace_canload(base, size, mstate, vstate))
1441 break;
1442
1443 for (j = 0; j < size; j++) {
1444 hashval += dtrace_load8(base + j);
1445 hashval += (hashval << 10);
1446 hashval ^= (hashval >> 6);
1447 }
1448 }
1449 }
1450
1451 if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT))
1452 return (NULL);
1453
1454 hashval += (hashval << 3);
1455 hashval ^= (hashval >> 11);
1456 hashval += (hashval << 15);
1457
1458 /*
1459 * There is a remote chance (ideally, 1 in 2^31) that our hashval
1460 * comes out to be one of our two sentinel hash values. If this
1461 * actually happens, we set the hashval to be a value known to be a
1462 * non-sentinel value.
1463 */
1464 if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK)
1465 hashval = DTRACE_DYNHASH_VALID;
1466
1467 /*
1468 * Yes, it's painful to do a divide here. If the cycle count becomes
1469 * important here, tricks can be pulled to reduce it. (However, it's
1470 * critical that hash collisions be kept to an absolute minimum;
1471 * they're much more painful than a divide.) It's better to have a
1472 * solution that generates few collisions and still keeps things
1473 * relatively simple.
1474 */
1475 bucket = hashval % dstate->dtds_hashsize;
1476
1477 if (op == DTRACE_DYNVAR_DEALLOC) {
1478 volatile uintptr_t *lockp = &hash[bucket].dtdh_lock;
1479
1480 for (;;) {
1481 while ((lock = *lockp) & 1)
1482 continue;
1483
1484 if (dtrace_casptr((volatile void *)lockp,
1485 (volatile void *)lock, (volatile void *)(lock + 1)) == (void *)lock)
1486 break;
1487 }
1488
1489 dtrace_membar_producer();
1490 }
1491
1492top:
1493 prev = NULL;
1494 lock = hash[bucket].dtdh_lock;
1495
1496 dtrace_membar_consumer();
1497
1498 start = hash[bucket].dtdh_chain;
1499 ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK ||
1500 start->dtdv_hashval != DTRACE_DYNHASH_FREE ||
1501 op != DTRACE_DYNVAR_DEALLOC));
1502
1503 for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) {
1504 dtrace_tuple_t *dtuple = &dvar->dtdv_tuple;
1505 dtrace_key_t *dkey = &dtuple->dtt_key[0];
1506
1507 if (dvar->dtdv_hashval != hashval) {
1508 if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) {
1509 /*
1510 * We've reached the sink, and therefore the
1511 * end of the hash chain; we can kick out of
1512 * the loop knowing that we have seen a valid
1513 * snapshot of state.
1514 */
1515 ASSERT(dvar->dtdv_next == NULL);
1516 ASSERT(dvar == &dtrace_dynhash_sink);
1517 break;
1518 }
1519
1520 if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) {
1521 /*
1522 * We've gone off the rails: somewhere along
1523 * the line, one of the members of this hash
1524 * chain was deleted. Note that we could also
1525 * detect this by simply letting this loop run
1526 * to completion, as we would eventually hit
1527 * the end of the dirty list. However, we
1528 * want to avoid running the length of the
1529 * dirty list unnecessarily (it might be quite
1530 * long), so we catch this as early as
1531 * possible by detecting the hash marker. In
1532 * this case, we simply set dvar to NULL and
1533 * break; the conditional after the loop will
1534 * send us back to top.
1535 */
1536 dvar = NULL;
1537 break;
1538 }
1539
1540 goto next;
1541 }
1542
1543 if (dtuple->dtt_nkeys != nkeys)
1544 goto next;
1545
1546 for (i = 0; i < nkeys; i++, dkey++) {
1547 if (dkey->dttk_size != key[i].dttk_size)
1548 goto next; /* size or type mismatch */
1549
1550 if (dkey->dttk_size != 0) {
1551 if (dtrace_bcmp(
1552 (void *)(uintptr_t)key[i].dttk_value,
1553 (void *)(uintptr_t)dkey->dttk_value,
1554 dkey->dttk_size))
1555 goto next;
1556 } else {
1557 if (dkey->dttk_value != key[i].dttk_value)
1558 goto next;
1559 }
1560 }
1561
1562 if (op != DTRACE_DYNVAR_DEALLOC)
1563 return (dvar);
1564
1565 ASSERT(dvar->dtdv_next == NULL ||
1566 dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE);
1567
1568 if (prev != NULL) {
1569 ASSERT(hash[bucket].dtdh_chain != dvar);
1570 ASSERT(start != dvar);
1571 ASSERT(prev->dtdv_next == dvar);
1572 prev->dtdv_next = dvar->dtdv_next;
1573 } else {
1574 if (dtrace_casptr(&hash[bucket].dtdh_chain,
1575 start, dvar->dtdv_next) != start) {
1576 /*
1577 * We have failed to atomically swing the
1578 * hash table head pointer, presumably because
1579 * of a conflicting allocation on another CPU.
1580 * We need to reread the hash chain and try
1581 * again.
1582 */
1583 goto top;
1584 }
1585 }
1586
1587 dtrace_membar_producer();
1588
1589 /*
1590 * Now set the hash value to indicate that it's free.
1591 */
1592 ASSERT(hash[bucket].dtdh_chain != dvar);
1593 dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;
1594
1595 dtrace_membar_producer();
1596
1597 /*
1598 * Set the next pointer to point at the dirty list, and
1599 * atomically swing the dirty pointer to the newly freed dvar.
1600 */
1601 do {
1602 next = dcpu->dtdsc_dirty;
1603 dvar->dtdv_next = next;
1604 } while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next);
1605
1606 /*
1607 * Finally, unlock this hash bucket.
1608 */
1609 ASSERT(hash[bucket].dtdh_lock == lock);
1610 ASSERT(lock & 1);
1611 hash[bucket].dtdh_lock++;
1612
1613 return (NULL);
1614next:
1615 prev = dvar;
1616 continue;
1617 }
1618
1619 if (dvar == NULL) {
1620 /*
1621 * If dvar is NULL, it is because we went off the rails:
1622 * one of the elements that we traversed in the hash chain
1623 * was deleted while we were traversing it. In this case,
1624 * we assert that we aren't doing a dealloc (deallocs lock
1625 * the hash bucket to prevent themselves from racing with
1626 * one another), and retry the hash chain traversal.
1627 */
1628 ASSERT(op != DTRACE_DYNVAR_DEALLOC);
1629 goto top;
1630 }
1631
1632 if (op != DTRACE_DYNVAR_ALLOC) {
1633 /*
1634 * If we are not to allocate a new variable, we want to
1635 * return NULL now. Before we return, check that the value
1636 * of the lock word hasn't changed. If it has, we may have
1637 * seen an inconsistent snapshot.
1638 */
1639 if (op == DTRACE_DYNVAR_NOALLOC) {
1640 if (hash[bucket].dtdh_lock != lock)
1641 goto top;
1642 } else {
1643 ASSERT(op == DTRACE_DYNVAR_DEALLOC);
1644 ASSERT(hash[bucket].dtdh_lock == lock);
1645 ASSERT(lock & 1);
1646 hash[bucket].dtdh_lock++;
1647 }
1648
1649 return (NULL);
1650 }
1651
1652 /*
1653 * We need to allocate a new dynamic variable. The size we need is the
1654 * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the
1655 * size of any auxiliary key data (rounded up to 8-byte alignment) plus
1656 * the size of any referred-to data (dsize). We then round the final
1657 * size up to the chunksize for allocation.
1658 */
1659 for (ksize = 0, i = 0; i < nkeys; i++)
1660 ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));
1661
1662 /*
1663 * This should be pretty much impossible, but could happen if, say,
1664 * strange DIF specified the tuple. Ideally, this should be an
1665 * assertion and not an error condition -- but that requires that the
1666 * chunksize calculation in dtrace_difo_chunksize() be absolutely
1667 * bullet-proof. (That is, it must not be able to be fooled by
1668 * malicious DIF.) Given the lack of backwards branches in DIF,
1669 * solving this would presumably not amount to solving the Halting
1670 * Problem -- but it still seems awfully hard.
1671 */
1672 if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) +
1673 ksize + dsize > chunksize) {
1674 dcpu->dtdsc_drops++;
1675 return (NULL);
1676 }
1677
1678 nstate = DTRACE_DSTATE_EMPTY;
1679
1680 do {
1681retry:
1682 free = dcpu->dtdsc_free;
1683
1684 if (free == NULL) {
1685 dtrace_dynvar_t *clean = dcpu->dtdsc_clean;
1686 void *rval;
1687
1688 if (clean == NULL) {
1689 /*
1690 * We're out of dynamic variable space on
1691 * this CPU. Unless we have tried all CPUs,
1692 * we'll try to allocate from a different
1693 * CPU.
1694 */
1695 switch (dstate->dtds_state) {
1696 case DTRACE_DSTATE_CLEAN: {
1697 void *sp = &dstate->dtds_state;
1698
1699 if (++cpu >= NCPU)
1700 cpu = 0;
1701
1702 if (dcpu->dtdsc_dirty != NULL &&
1703 nstate == DTRACE_DSTATE_EMPTY)
1704 nstate = DTRACE_DSTATE_DIRTY;
1705
1706 if (dcpu->dtdsc_rinsing != NULL)
1707 nstate = DTRACE_DSTATE_RINSING;
1708
1709 dcpu = &dstate->dtds_percpu[cpu];
1710
1711 if (cpu != me)
1712 goto retry;
1713
1714 (void) dtrace_cas32(sp,
1715 DTRACE_DSTATE_CLEAN, nstate);
1716
1717 /*
1718 * To increment the correct bean
1719 * counter, take another lap.
1720 */
1721 goto retry;
1722 }
1723
1724 case DTRACE_DSTATE_DIRTY:
1725 dcpu->dtdsc_dirty_drops++;
1726 break;
1727
1728 case DTRACE_DSTATE_RINSING:
1729 dcpu->dtdsc_rinsing_drops++;
1730 break;
1731
1732 case DTRACE_DSTATE_EMPTY:
1733 dcpu->dtdsc_drops++;
1734 break;
1735 }
1736
1737 DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP);
1738 return (NULL);
1739 }
1740
1741 /*
1742 * The clean list appears to be non-empty. We want to
1743 * move the clean list to the free list; we start by
1744 * moving the clean pointer aside.
1745 */
1746 if (dtrace_casptr(&dcpu->dtdsc_clean,
1747 clean, NULL) != clean) {
1748 /*
1749 * We are in one of two situations:
1750 *
1751 * (a) The clean list was switched to the
1752 * free list by another CPU.
1753 *
1754 * (b) The clean list was added to by the
1755 * cleansing cyclic.
1756 *
1757 * In either of these situations, we can
1758 * just reattempt the free list allocation.
1759 */
1760 goto retry;
1761 }
1762
1763 ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE);
1764
1765 /*
1766 * Now we'll move the clean list to the free list.
1767 * It's impossible for this to fail: the only way
1768 * the free list can be updated is through this
1769 * code path, and only one CPU can own the clean list.
1770 * Thus, it would only be possible for this to fail if
1771 * this code were racing with dtrace_dynvar_clean().
1772 * (That is, if dtrace_dynvar_clean() updated the clean
1773 * list, and we ended up racing to update the free
1774 * list.) This race is prevented by the dtrace_sync()
1775 * in dtrace_dynvar_clean() -- which flushes the
1776 * owners of the clean lists out before resetting
1777 * the clean lists.
1778 */
1779 rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean);
1780 ASSERT(rval == NULL);
1781 goto retry;
1782 }
1783
1784 dvar = free;
1785 new_free = dvar->dtdv_next;
1786 } while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free);
1787
1788 /*
1789 * We have now allocated a new chunk. We copy the tuple keys into the
1790 * tuple array and copy any referenced key data into the data space
1791 * following the tuple array. As we do this, we relocate dttk_value
1792 * in the final tuple to point to the key data address in the chunk.
1793 */
1794 kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys];
1795 dvar->dtdv_data = (void *)(kdata + ksize);
1796 dvar->dtdv_tuple.dtt_nkeys = nkeys;
1797
1798 for (i = 0; i < nkeys; i++) {
1799 dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i];
1800 size_t kesize = key[i].dttk_size;
1801
1802 if (kesize != 0) {
1803 dtrace_bcopy(
1804 (const void *)(uintptr_t)key[i].dttk_value,
1805 (void *)kdata, kesize);
1806 dkey->dttk_value = kdata;
1807 kdata += P2ROUNDUP(kesize, sizeof (uint64_t));
1808 } else {
1809 dkey->dttk_value = key[i].dttk_value;
1810 }
1811
1812 dkey->dttk_size = kesize;
1813 }
1814
1815 ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE);
1816 dvar->dtdv_hashval = hashval;
1817 dvar->dtdv_next = start;
1818
1819 if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start)
1820 return (dvar);
1821
1822 /*
1823 * The cas has failed. Either another CPU is adding an element to
1824 * this hash chain, or another CPU is deleting an element from this
1825 * hash chain. The simplest way to deal with both of these cases
1826 * (though not necessarily the most efficient) is to free our
1827 * allocated block and tail-call ourselves. Note that the free is
1828 * to the dirty list and _not_ to the free list. This is to prevent
1829 * races with allocators, above.
1830 */
1831 dvar->dtdv_hashval = DTRACE_DYNHASH_FREE;
1832
1833 dtrace_membar_producer();
1834
1835 do {
1836 free = dcpu->dtdsc_dirty;
1837 dvar->dtdv_next = free;
1838 } while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free);
1839
1840 return (dtrace_dynvar(dstate, nkeys, key, dsize, op, mstate, vstate));
1841}
1842
1843/*ARGSUSED*/
1844static void
1845dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg)
1846{
1847 if ((int64_t)nval < (int64_t)*oval)
1848 *oval = nval;
1849}
1850
1851/*ARGSUSED*/
1852static void
1853dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg)
1854{
1855 if ((int64_t)nval > (int64_t)*oval)
1856 *oval = nval;
1857}
1858
1859static void
1860dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr)
1861{
1862 int i, zero = DTRACE_QUANTIZE_ZEROBUCKET;
1863 int64_t val = (int64_t)nval;
1864
1865 if (val < 0) {
1866 for (i = 0; i < zero; i++) {
1867 if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) {
1868 quanta[i] += incr;
1869 return;
1870 }
1871 }
1872 } else {
1873 for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) {
1874 if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) {
1875 quanta[i - 1] += incr;
1876 return;
1877 }
1878 }
1879
1880 quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr;
1881 return;
1882 }
1883
1884 ASSERT(0);
1885}
1886
1887static void
1888dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr)
1889{
1890 uint64_t arg = *lquanta++;
1891 int32_t base = DTRACE_LQUANTIZE_BASE(arg);
1892 uint16_t step = DTRACE_LQUANTIZE_STEP(arg);
1893 uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg);
1894 int32_t val = (int32_t)nval, level;
1895
1896 ASSERT(step != 0);
1897 ASSERT(levels != 0);
1898
1899 if (val < base) {
1900 /*
1901 * This is an underflow.
1902 */
1903 lquanta[0] += incr;
1904 return;
1905 }
1906
1907 level = (val - base) / step;
1908
1909 if (level < levels) {
1910 lquanta[level + 1] += incr;
1911 return;
1912 }
1913
1914 /*
1915 * This is an overflow.
1916 */
1917 lquanta[levels + 1] += incr;
1918}
1919
1920/*ARGSUSED*/
1921static void
1922dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg)
1923{
1924 data[0]++;
1925 data[1] += nval;
1926}
1927
1928/*ARGSUSED*/
1929static void
1930dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg)
1931{
1932 int64_t snval = (int64_t)nval;
1933 uint64_t tmp[2];
1934
1935 data[0]++;
1936 data[1] += nval;
1937
1938 /*
1939 * What we want to say here is:
1940 *
1941 * data[2] += nval * nval;
1942 *
1943 * But given that nval is 64-bit, we could easily overflow, so
1944 * we do this as 128-bit arithmetic.
1945 */
1946 if (snval < 0)
1947 snval = -snval;
1948
1949 dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp);
1950 dtrace_add_128(data + 2, tmp, data + 2);
1951}
1952
1953/*ARGSUSED*/
1954static void
1955dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg)
1956{
1957 *oval = *oval + 1;
1958}
1959
1960/*ARGSUSED*/
1961static void
1962dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg)
1963{
1964 *oval += nval;
1965}
1966
1967/*
1968 * Aggregate given the tuple in the principal data buffer, and the aggregating
1969 * action denoted by the specified dtrace_aggregation_t. The aggregation
1970 * buffer is specified as the buf parameter. This routine does not return
1971 * failure; if there is no space in the aggregation buffer, the data will be
1972 * dropped, and a corresponding counter incremented.
1973 */
1974static void
1975dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf,
1976 intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg)
1977{
1978 dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec;
1979 uint32_t i, ndx, size, fsize;
1980 uint32_t align = sizeof (uint64_t) - 1;
1981 dtrace_aggbuffer_t *agb;
1982 dtrace_aggkey_t *key;
1983 uint32_t hashval = 0, limit, isstr;
1984 caddr_t tomax, data, kdata;
1985 dtrace_actkind_t action;
1986 dtrace_action_t *act;
1987 uintptr_t offs;
1988
1989 if (buf == NULL)
1990 return;
1991
1992 if (!agg->dtag_hasarg) {
1993 /*
1994 * Currently, only quantize() and lquantize() take additional
1995 * arguments, and they have the same semantics: an increment
1996 * value that defaults to 1 when not present. If additional
1997 * aggregating actions take arguments, the setting of the
1998 * default argument value will presumably have to become more
1999 * sophisticated...
2000 */
2001 arg = 1;
2002 }
2003
2004 action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION;
2005 size = rec->dtrd_offset - agg->dtag_base;
2006 fsize = size + rec->dtrd_size;
2007
2008 ASSERT(dbuf->dtb_tomax != NULL);
2009 data = dbuf->dtb_tomax + offset + agg->dtag_base;
2010
2011 if ((tomax = buf->dtb_tomax) == NULL) {
2012 dtrace_buffer_drop(buf);
2013 return;
2014 }
2015
2016 /*
2017 * The metastructure is always at the bottom of the buffer.
2018 */
2019 agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size -
2020 sizeof (dtrace_aggbuffer_t));
2021
2022 if (buf->dtb_offset == 0) {
2023 /*
2024 * We just kludge up approximately 1/8th of the size to be
2025 * buckets. If this guess ends up being routinely
2026 * off-the-mark, we may need to dynamically readjust this
2027 * based on past performance.
2028 */
2029 uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t);
2030
2031 if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) <
2032 (uintptr_t)tomax || hashsize == 0) {
2033 /*
2034 * We've been given a ludicrously small buffer;
2035 * increment our drop count and leave.
2036 */
2037 dtrace_buffer_drop(buf);
2038 return;
2039 }
2040
2041 /*
2042 * And now, a pathetic attempt to try to get a an odd (or
2043 * perchance, a prime) hash size for better hash distribution.
2044 */
2045 if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3))
2046 hashsize -= DTRACE_AGGHASHSIZE_SLEW;
2047
2048 agb->dtagb_hashsize = hashsize;
2049 agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb -
2050 agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *));
2051 agb->dtagb_free = (uintptr_t)agb->dtagb_hash;
2052
2053 for (i = 0; i < agb->dtagb_hashsize; i++)
2054 agb->dtagb_hash[i] = NULL;
2055 }
2056
2057 ASSERT(agg->dtag_first != NULL);
2058 ASSERT(agg->dtag_first->dta_intuple);
2059
2060 /*
2061 * Calculate the hash value based on the key. Note that we _don't_
2062 * include the aggid in the hashing (but we will store it as part of
2063 * the key). The hashing algorithm is Bob Jenkins' "One-at-a-time"
2064 * algorithm: a simple, quick algorithm that has no known funnels, and
2065 * gets good distribution in practice. The efficacy of the hashing
2066 * algorithm (and a comparison with other algorithms) may be found by
2067 * running the ::dtrace_aggstat MDB dcmd.
2068 */
2069 for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
2070 i = act->dta_rec.dtrd_offset - agg->dtag_base;
2071 limit = i + act->dta_rec.dtrd_size;
2072 ASSERT(limit <= size);
2073 isstr = DTRACEACT_ISSTRING(act);
2074
2075 for (; i < limit; i++) {
2076 hashval += data[i];
2077 hashval += (hashval << 10);
2078 hashval ^= (hashval >> 6);
2079
2080 if (isstr && data[i] == '\0')
2081 break;
2082 }
2083 }
2084
2085 hashval += (hashval << 3);
2086 hashval ^= (hashval >> 11);
2087 hashval += (hashval << 15);
2088
2089 /*
2090 * Yes, the divide here is expensive -- but it's generally the least
2091 * of the performance issues given the amount of data that we iterate
2092 * over to compute hash values, compare data, etc.
2093 */
2094 ndx = hashval % agb->dtagb_hashsize;
2095
2096 for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) {
2097 ASSERT((caddr_t)key >= tomax);
2098 ASSERT((caddr_t)key < tomax + buf->dtb_size);
2099
2100 if (hashval != key->dtak_hashval || key->dtak_size != size)
2101 continue;
2102
2103 kdata = key->dtak_data;
2104 ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size);
2105
2106 for (act = agg->dtag_first; act->dta_intuple;
2107 act = act->dta_next) {
2108 i = act->dta_rec.dtrd_offset - agg->dtag_base;
2109 limit = i + act->dta_rec.dtrd_size;
2110 ASSERT(limit <= size);
2111 isstr = DTRACEACT_ISSTRING(act);
2112
2113 for (; i < limit; i++) {
2114 if (kdata[i] != data[i])
2115 goto next;
2116
2117 if (isstr && data[i] == '\0')
2118 break;
2119 }
2120 }
2121
2122 if (action != key->dtak_action) {
2123 /*
2124 * We are aggregating on the same value in the same
2125 * aggregation with two different aggregating actions.
2126 * (This should have been picked up in the compiler,
2127 * so we may be dealing with errant or devious DIF.)
2128 * This is an error condition; we indicate as much,
2129 * and return.
2130 */
2131 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
2132 return;
2133 }
2134
2135 /*
2136 * This is a hit: we need to apply the aggregator to
2137 * the value at this key.
2138 */
2139 agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg);
2140 return;
2141next:
2142 continue;
2143 }
2144
2145 /*
2146 * We didn't find it. We need to allocate some zero-filled space,
2147 * link it into the hash table appropriately, and apply the aggregator
2148 * to the (zero-filled) value.
2149 */
2150 offs = buf->dtb_offset;
2151 while (offs & (align - 1))
2152 offs += sizeof (uint32_t);
2153
2154 /*
2155 * If we don't have enough room to both allocate a new key _and_
2156 * its associated data, increment the drop count and return.
2157 */
2158 if ((uintptr_t)tomax + offs + fsize >
2159 agb->dtagb_free - sizeof (dtrace_aggkey_t)) {
2160 dtrace_buffer_drop(buf);
2161 return;
2162 }
2163
2164 /*CONSTCOND*/
2165 ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1)));
2166 key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t));
2167 agb->dtagb_free -= sizeof (dtrace_aggkey_t);
2168
2169 key->dtak_data = kdata = tomax + offs;
2170 buf->dtb_offset = offs + fsize;
2171
2172 /*
2173 * Now copy the data across.
2174 */
2175 *((dtrace_aggid_t *)kdata) = agg->dtag_id;
2176
2177 for (i = sizeof (dtrace_aggid_t); i < size; i++)
2178 kdata[i] = data[i];
2179
2180 /*
2181 * Because strings are not zeroed out by default, we need to iterate
2182 * looking for actions that store strings, and we need to explicitly
2183 * pad these strings out with zeroes.
2184 */
2185 for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) {
2186 int nul;
2187
2188 if (!DTRACEACT_ISSTRING(act))
2189 continue;
2190
2191 i = act->dta_rec.dtrd_offset - agg->dtag_base;
2192 limit = i + act->dta_rec.dtrd_size;
2193 ASSERT(limit <= size);
2194
2195 for (nul = 0; i < limit; i++) {
2196 if (nul) {
2197 kdata[i] = '\0';
2198 continue;
2199 }
2200
2201 if (data[i] != '\0')
2202 continue;
2203
2204 nul = 1;
2205 }
2206 }
2207
2208 for (i = size; i < fsize; i++)
2209 kdata[i] = 0;
2210
2211 key->dtak_hashval = hashval;
2212 key->dtak_size = size;
2213 key->dtak_action = action;
2214 key->dtak_next = agb->dtagb_hash[ndx];
2215 agb->dtagb_hash[ndx] = key;
2216
2217 /*
2218 * Finally, apply the aggregator.
2219 */
2220 *((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial;
2221 agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg);
2222}
2223
2224/*
2225 * Given consumer state, this routine finds a speculation in the INACTIVE
2226 * state and transitions it into the ACTIVE state. If there is no speculation
2227 * in the INACTIVE state, 0 is returned. In this case, no error counter is
2228 * incremented -- it is up to the caller to take appropriate action.
2229 */
2230static int
2231dtrace_speculation(dtrace_state_t *state)
2232{
2233 int i = 0;
2234 dtrace_speculation_state_t current;
2235 uint32_t *stat = &state->dts_speculations_unavail, count;
2236
2237 while (i < state->dts_nspeculations) {
2238 dtrace_speculation_t *spec = &state->dts_speculations[i];
2239
2240 current = spec->dtsp_state;
2241
2242 if (current != DTRACESPEC_INACTIVE) {
2243 if (current == DTRACESPEC_COMMITTINGMANY ||
2244 current == DTRACESPEC_COMMITTING ||
2245 current == DTRACESPEC_DISCARDING)
2246 stat = &state->dts_speculations_busy;
2247 i++;
2248 continue;
2249 }
2250
2251 if (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2252 current, DTRACESPEC_ACTIVE) == current)
2253 return (i + 1);
2254 }
2255
2256 /*
2257 * We couldn't find a speculation. If we found as much as a single
2258 * busy speculation buffer, we'll attribute this failure as "busy"
2259 * instead of "unavail".
2260 */
2261 do {
2262 count = *stat;
2263 } while (dtrace_cas32(stat, count, count + 1) != count);
2264
2265 return (0);
2266}
2267
2268/*
2269 * This routine commits an active speculation. If the specified speculation
2270 * is not in a valid state to perform a commit(), this routine will silently do
2271 * nothing. The state of the specified speculation is transitioned according
2272 * to the state transition diagram outlined in <sys/dtrace_impl.h>
2273 */
2274static void
2275dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu,
2276 dtrace_specid_t which)
2277{
2278 dtrace_speculation_t *spec;
2279 dtrace_buffer_t *src, *dest;
2280 uintptr_t daddr, saddr, dlimit;
2281 dtrace_speculation_state_t current, new = 0;
2282 intptr_t offs;
2283
2284 if (which == 0)
2285 return;
2286
2287 if (which > state->dts_nspeculations) {
2288 cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2289 return;
2290 }
2291
2292 spec = &state->dts_speculations[which - 1];
2293 src = &spec->dtsp_buffer[cpu];
2294 dest = &state->dts_buffer[cpu];
2295
2296 do {
2297 current = spec->dtsp_state;
2298
2299 if (current == DTRACESPEC_COMMITTINGMANY)
2300 break;
2301
2302 switch (current) {
2303 case DTRACESPEC_INACTIVE:
2304 case DTRACESPEC_DISCARDING:
2305 return;
2306
2307 case DTRACESPEC_COMMITTING:
2308 /*
2309 * This is only possible if we are (a) commit()'ing
2310 * without having done a prior speculate() on this CPU
2311 * and (b) racing with another commit() on a different
2312 * CPU. There's nothing to do -- we just assert that
2313 * our offset is 0.
2314 */
2315 ASSERT(src->dtb_offset == 0);
2316 return;
2317
2318 case DTRACESPEC_ACTIVE:
2319 new = DTRACESPEC_COMMITTING;
2320 break;
2321
2322 case DTRACESPEC_ACTIVEONE:
2323 /*
2324 * This speculation is active on one CPU. If our
2325 * buffer offset is non-zero, we know that the one CPU
2326 * must be us. Otherwise, we are committing on a
2327 * different CPU from the speculate(), and we must
2328 * rely on being asynchronously cleaned.
2329 */
2330 if (src->dtb_offset != 0) {
2331 new = DTRACESPEC_COMMITTING;
2332 break;
2333 }
2334 /*FALLTHROUGH*/
2335
2336 case DTRACESPEC_ACTIVEMANY:
2337 new = DTRACESPEC_COMMITTINGMANY;
2338 break;
2339
2340 default:
2341 ASSERT(0);
2342 }
2343 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2344 current, new) != current);
2345
2346 /*
2347 * We have set the state to indicate that we are committing this
2348 * speculation. Now reserve the necessary space in the destination
2349 * buffer.
2350 */
2351 if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset,
2352 sizeof (uint64_t), state, NULL)) < 0) {
2353 dtrace_buffer_drop(dest);
2354 goto out;
2355 }
2356
2357 /*
2358 * We have the space; copy the buffer across. (Note that this is a
2359 * highly subobtimal bcopy(); in the unlikely event that this becomes
2360 * a serious performance issue, a high-performance DTrace-specific
2361 * bcopy() should obviously be invented.)
2362 */
2363 daddr = (uintptr_t)dest->dtb_tomax + offs;
2364 dlimit = daddr + src->dtb_offset;
2365 saddr = (uintptr_t)src->dtb_tomax;
2366
2367 /*
2368 * First, the aligned portion.
2369 */
2370 while (dlimit - daddr >= sizeof (uint64_t)) {
2371 *((uint64_t *)daddr) = *((uint64_t *)saddr);
2372
2373 daddr += sizeof (uint64_t);
2374 saddr += sizeof (uint64_t);
2375 }
2376
2377 /*
2378 * Now any left-over bit...
2379 */
2380 while (dlimit - daddr)
2381 *((uint8_t *)daddr++) = *((uint8_t *)saddr++);
2382
2383 /*
2384 * Finally, commit the reserved space in the destination buffer.
2385 */
2386 dest->dtb_offset = offs + src->dtb_offset;
2387
2388out:
2389 /*
2390 * If we're lucky enough to be the only active CPU on this speculation
2391 * buffer, we can just set the state back to DTRACESPEC_INACTIVE.
2392 */
2393 if (current == DTRACESPEC_ACTIVE ||
2394 (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) {
2395 uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state,
2396 DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE);
2397
2398 ASSERT(rval == DTRACESPEC_COMMITTING);
2399 }
2400
2401 src->dtb_offset = 0;
2402 src->dtb_xamot_drops += src->dtb_drops;
2403 src->dtb_drops = 0;
2404}
2405
2406/*
2407 * This routine discards an active speculation. If the specified speculation
2408 * is not in a valid state to perform a discard(), this routine will silently
2409 * do nothing. The state of the specified speculation is transitioned
2410 * according to the state transition diagram outlined in <sys/dtrace_impl.h>
2411 */
2412static void
2413dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu,
2414 dtrace_specid_t which)
2415{
2416 dtrace_speculation_t *spec;
2417 dtrace_speculation_state_t current, new = 0;
2418 dtrace_buffer_t *buf;
2419
2420 if (which == 0)
2421 return;
2422
2423 if (which > state->dts_nspeculations) {
2424 cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2425 return;
2426 }
2427
2428 spec = &state->dts_speculations[which - 1];
2429 buf = &spec->dtsp_buffer[cpu];
2430
2431 do {
2432 current = spec->dtsp_state;
2433
2434 switch (current) {
2435 case DTRACESPEC_INACTIVE:
2436 case DTRACESPEC_COMMITTINGMANY:
2437 case DTRACESPEC_COMMITTING:
2438 case DTRACESPEC_DISCARDING:
2439 return;
2440
2441 case DTRACESPEC_ACTIVE:
2442 case DTRACESPEC_ACTIVEMANY:
2443 new = DTRACESPEC_DISCARDING;
2444 break;
2445
2446 case DTRACESPEC_ACTIVEONE:
2447 if (buf->dtb_offset != 0) {
2448 new = DTRACESPEC_INACTIVE;
2449 } else {
2450 new = DTRACESPEC_DISCARDING;
2451 }
2452 break;
2453
2454 default:
2455 ASSERT(0);
2456 }
2457 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2458 current, new) != current);
2459
2460 buf->dtb_offset = 0;
2461 buf->dtb_drops = 0;
2462}
2463
2464/*
2465 * Note: not called from probe context. This function is called
2466 * asynchronously from cross call context to clean any speculations that are
2467 * in the COMMITTINGMANY or DISCARDING states. These speculations may not be
2468 * transitioned back to the INACTIVE state until all CPUs have cleaned the
2469 * speculation.
2470 */
2471static void
2472dtrace_speculation_clean_here(dtrace_state_t *state)
2473{
2474 dtrace_icookie_t cookie;
2475 processorid_t cpu = curcpu;
2476 dtrace_buffer_t *dest = &state->dts_buffer[cpu];
2477 dtrace_specid_t i;
2478
2479 cookie = dtrace_interrupt_disable();
2480
2481 if (dest->dtb_tomax == NULL) {
2482 dtrace_interrupt_enable(cookie);
2483 return;
2484 }
2485
2486 for (i = 0; i < state->dts_nspeculations; i++) {
2487 dtrace_speculation_t *spec = &state->dts_speculations[i];
2488 dtrace_buffer_t *src = &spec->dtsp_buffer[cpu];
2489
2490 if (src->dtb_tomax == NULL)
2491 continue;
2492
2493 if (spec->dtsp_state == DTRACESPEC_DISCARDING) {
2494 src->dtb_offset = 0;
2495 continue;
2496 }
2497
2498 if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
2499 continue;
2500
2501 if (src->dtb_offset == 0)
2502 continue;
2503
2504 dtrace_speculation_commit(state, cpu, i + 1);
2505 }
2506
2507 dtrace_interrupt_enable(cookie);
2508}
2509
2510/*
2511 * Note: not called from probe context. This function is called
2512 * asynchronously (and at a regular interval) to clean any speculations that
2513 * are in the COMMITTINGMANY or DISCARDING states. If it discovers that there
2514 * is work to be done, it cross calls all CPUs to perform that work;
2515 * COMMITMANY and DISCARDING speculations may not be transitioned back to the
2516 * INACTIVE state until they have been cleaned by all CPUs.
2517 */
2518static void
2519dtrace_speculation_clean(dtrace_state_t *state)
2520{
2521 int work = 0, rv;
2522 dtrace_specid_t i;
2523
2524 for (i = 0; i < state->dts_nspeculations; i++) {
2525 dtrace_speculation_t *spec = &state->dts_speculations[i];
2526
2527 ASSERT(!spec->dtsp_cleaning);
2528
2529 if (spec->dtsp_state != DTRACESPEC_DISCARDING &&
2530 spec->dtsp_state != DTRACESPEC_COMMITTINGMANY)
2531 continue;
2532
2533 work++;
2534 spec->dtsp_cleaning = 1;
2535 }
2536
2537 if (!work)
2538 return;
2539
2540 dtrace_xcall(DTRACE_CPUALL,
2541 (dtrace_xcall_t)dtrace_speculation_clean_here, state);
2542
2543 /*
2544 * We now know that all CPUs have committed or discarded their
2545 * speculation buffers, as appropriate. We can now set the state
2546 * to inactive.
2547 */
2548 for (i = 0; i < state->dts_nspeculations; i++) {
2549 dtrace_speculation_t *spec = &state->dts_speculations[i];
2550 dtrace_speculation_state_t current, new;
2551
2552 if (!spec->dtsp_cleaning)
2553 continue;
2554
2555 current = spec->dtsp_state;
2556 ASSERT(current == DTRACESPEC_DISCARDING ||
2557 current == DTRACESPEC_COMMITTINGMANY);
2558
2559 new = DTRACESPEC_INACTIVE;
2560
2561 rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new);
2562 ASSERT(rv == current);
2563 spec->dtsp_cleaning = 0;
2564 }
2565}
2566
2567/*
2568 * Called as part of a speculate() to get the speculative buffer associated
2569 * with a given speculation. Returns NULL if the specified speculation is not
2570 * in an ACTIVE state. If the speculation is in the ACTIVEONE state -- and
2571 * the active CPU is not the specified CPU -- the speculation will be
2572 * atomically transitioned into the ACTIVEMANY state.
2573 */
2574static dtrace_buffer_t *
2575dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid,
2576 dtrace_specid_t which)
2577{
2578 dtrace_speculation_t *spec;
2579 dtrace_speculation_state_t current, new = 0;
2580 dtrace_buffer_t *buf;
2581
2582 if (which == 0)
2583 return (NULL);
2584
2585 if (which > state->dts_nspeculations) {
2586 cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
2587 return (NULL);
2588 }
2589
2590 spec = &state->dts_speculations[which - 1];
2591 buf = &spec->dtsp_buffer[cpuid];
2592
2593 do {
2594 current = spec->dtsp_state;
2595
2596 switch (current) {
2597 case DTRACESPEC_INACTIVE:
2598 case DTRACESPEC_COMMITTINGMANY:
2599 case DTRACESPEC_DISCARDING:
2600 return (NULL);
2601
2602 case DTRACESPEC_COMMITTING:
2603 ASSERT(buf->dtb_offset == 0);
2604 return (NULL);
2605
2606 case DTRACESPEC_ACTIVEONE:
2607 /*
2608 * This speculation is currently active on one CPU.
2609 * Check the offset in the buffer; if it's non-zero,
2610 * that CPU must be us (and we leave the state alone).
2611 * If it's zero, assume that we're starting on a new
2612 * CPU -- and change the state to indicate that the
2613 * speculation is active on more than one CPU.
2614 */
2615 if (buf->dtb_offset != 0)
2616 return (buf);
2617
2618 new = DTRACESPEC_ACTIVEMANY;
2619 break;
2620
2621 case DTRACESPEC_ACTIVEMANY:
2622 return (buf);
2623
2624 case DTRACESPEC_ACTIVE:
2625 new = DTRACESPEC_ACTIVEONE;
2626 break;
2627
2628 default:
2629 ASSERT(0);
2630 }
2631 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state,
2632 current, new) != current);
2633
2634 ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY);
2635 return (buf);
2636}
2637
2638/*
2639 * Return a string. In the event that the user lacks the privilege to access
2640 * arbitrary kernel memory, we copy the string out to scratch memory so that we
2641 * don't fail access checking.
2642 *
2643 * dtrace_dif_variable() uses this routine as a helper for various
2644 * builtin values such as 'execname' and 'probefunc.'
2645 */
2646uintptr_t
2647dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state,
2648 dtrace_mstate_t *mstate)
2649{
2650 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
2651 uintptr_t ret;
2652 size_t strsz;
2653
2654 /*
2655 * The easy case: this probe is allowed to read all of memory, so
2656 * we can just return this as a vanilla pointer.
2657 */
2658 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0)
2659 return (addr);
2660
2661 /*
2662 * This is the tougher case: we copy the string in question from
2663 * kernel memory into scratch memory and return it that way: this
2664 * ensures that we won't trip up when access checking tests the
2665 * BYREF return value.
2666 */
2667 strsz = dtrace_strlen((char *)addr, size) + 1;
2668
2669 if (mstate->dtms_scratch_ptr + strsz >
2670 mstate->dtms_scratch_base + mstate->dtms_scratch_size) {
2671 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
2672 return (0);
2673 }
2674
2675 dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr,
2676 strsz);
2677 ret = mstate->dtms_scratch_ptr;
2678 mstate->dtms_scratch_ptr += strsz;
2679 return (ret);
2680}
2681
2682/*
2683 * Return a string from a memoy address which is known to have one or
2684 * more concatenated, individually zero terminated, sub-strings.
2685 * In the event that the user lacks the privilege to access
2686 * arbitrary kernel memory, we copy the string out to scratch memory so that we
2687 * don't fail access checking.
2688 *
2689 * dtrace_dif_variable() uses this routine as a helper for various
2690 * builtin values such as 'execargs'.
2691 */
2692static uintptr_t
2693dtrace_dif_varstrz(uintptr_t addr, size_t strsz, dtrace_state_t *state,
2694 dtrace_mstate_t *mstate)
2695{
2696 char *p;
2697 size_t i;
2698 uintptr_t ret;
2699
2700 if (mstate->dtms_scratch_ptr + strsz >
2701 mstate->dtms_scratch_base + mstate->dtms_scratch_size) {
2702 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
2703 return (0);
2704 }
2705
2706 dtrace_bcopy((const void *)addr, (void *)mstate->dtms_scratch_ptr,
2707 strsz);
2708
2709 /* Replace sub-string termination characters with a space. */
2710 for (p = (char *) mstate->dtms_scratch_ptr, i = 0; i < strsz - 1;
2711 p++, i++)
2712 if (*p == '\0')
2713 *p = ' ';
2714
2715 ret = mstate->dtms_scratch_ptr;
2716 mstate->dtms_scratch_ptr += strsz;
2717 return (ret);
2718}
2719
2720/*
2721 * This function implements the DIF emulator's variable lookups. The emulator
2722 * passes a reserved variable identifier and optional built-in array index.
2723 */
2724static uint64_t
2725dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v,
2726 uint64_t ndx)
2727{
2728 /*
2729 * If we're accessing one of the uncached arguments, we'll turn this
2730 * into a reference in the args array.
2731 */
2732 if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) {
2733 ndx = v - DIF_VAR_ARG0;
2734 v = DIF_VAR_ARGS;
2735 }
2736
2737 switch (v) {
2738 case DIF_VAR_ARGS:
2739 ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS);
2740 if (ndx >= sizeof (mstate->dtms_arg) /
2741 sizeof (mstate->dtms_arg[0])) {
2742 int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2743 dtrace_provider_t *pv;
2744 uint64_t val;
2745
2746 pv = mstate->dtms_probe->dtpr_provider;
2747 if (pv->dtpv_pops.dtps_getargval != NULL)
2748 val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg,
2749 mstate->dtms_probe->dtpr_id,
2750 mstate->dtms_probe->dtpr_arg, ndx, aframes);
2751 else
2752 val = dtrace_getarg(ndx, aframes);
2753
2754 /*
2755 * This is regrettably required to keep the compiler
2756 * from tail-optimizing the call to dtrace_getarg().
2757 * The condition always evaluates to true, but the
2758 * compiler has no way of figuring that out a priori.
2759 * (None of this would be necessary if the compiler
2760 * could be relied upon to _always_ tail-optimize
2761 * the call to dtrace_getarg() -- but it can't.)
2762 */
2763 if (mstate->dtms_probe != NULL)
2764 return (val);
2765
2766 ASSERT(0);
2767 }
2768
2769 return (mstate->dtms_arg[ndx]);
2770
2771#if defined(sun)
2772 case DIF_VAR_UREGS: {
2773 klwp_t *lwp;
2774
2775 if (!dtrace_priv_proc(state))
2776 return (0);
2777
2778 if ((lwp = curthread->t_lwp) == NULL) {
2779 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR);
2780 cpu_core[curcpu].cpuc_dtrace_illval = NULL;
2781 return (0);
2782 }
2783
2784 return (dtrace_getreg(lwp->lwp_regs, ndx));
2785 return (0);
2786 }
2787#endif
2788
2789 case DIF_VAR_CURTHREAD:
2790 if (!dtrace_priv_kernel(state))
2791 return (0);
2792 return ((uint64_t)(uintptr_t)curthread);
2793
2794 case DIF_VAR_TIMESTAMP:
2795 if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) {
2796 mstate->dtms_timestamp = dtrace_gethrtime();
2797 mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP;
2798 }
2799 return (mstate->dtms_timestamp);
2800
2801 case DIF_VAR_VTIMESTAMP:
2802 ASSERT(dtrace_vtime_references != 0);
2803 return (curthread->t_dtrace_vtime);
2804
2805 case DIF_VAR_WALLTIMESTAMP:
2806 if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) {
2807 mstate->dtms_walltimestamp = dtrace_gethrestime();
2808 mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP;
2809 }
2810 return (mstate->dtms_walltimestamp);
2811
2812#if defined(sun)
2813 case DIF_VAR_IPL:
2814 if (!dtrace_priv_kernel(state))
2815 return (0);
2816 if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) {
2817 mstate->dtms_ipl = dtrace_getipl();
2818 mstate->dtms_present |= DTRACE_MSTATE_IPL;
2819 }
2820 return (mstate->dtms_ipl);
2821#endif
2822
2823 case DIF_VAR_EPID:
2824 ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID);
2825 return (mstate->dtms_epid);
2826
2827 case DIF_VAR_ID:
2828 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2829 return (mstate->dtms_probe->dtpr_id);
2830
2831 case DIF_VAR_STACKDEPTH:
2832 if (!dtrace_priv_kernel(state))
2833 return (0);
2834 if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) {
2835 int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2836
2837 mstate->dtms_stackdepth = dtrace_getstackdepth(aframes);
2838 mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH;
2839 }
2840 return (mstate->dtms_stackdepth);
2841
2842#if defined(sun)
2843 case DIF_VAR_USTACKDEPTH:
2844 if (!dtrace_priv_proc(state))
2845 return (0);
2846 if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) {
2847 /*
2848 * See comment in DIF_VAR_PID.
2849 */
2850 if (DTRACE_ANCHORED(mstate->dtms_probe) &&
2851 CPU_ON_INTR(CPU)) {
2852 mstate->dtms_ustackdepth = 0;
2853 } else {
2854 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
2855 mstate->dtms_ustackdepth =
2856 dtrace_getustackdepth();
2857 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
2858 }
2859 mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH;
2860 }
2861 return (mstate->dtms_ustackdepth);
2862#endif
2863
2864 case DIF_VAR_CALLER:
2865 if (!dtrace_priv_kernel(state))
2866 return (0);
2867 if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) {
2868 int aframes = mstate->dtms_probe->dtpr_aframes + 2;
2869
2870 if (!DTRACE_ANCHORED(mstate->dtms_probe)) {
2871 /*
2872 * If this is an unanchored probe, we are
2873 * required to go through the slow path:
2874 * dtrace_caller() only guarantees correct
2875 * results for anchored probes.
2876 */
2877 pc_t caller[2] = {0, 0};
2878
2879 dtrace_getpcstack(caller, 2, aframes,
2880 (uint32_t *)(uintptr_t)mstate->dtms_arg[0]);
2881 mstate->dtms_caller = caller[1];
2882 } else if ((mstate->dtms_caller =
2883 dtrace_caller(aframes)) == -1) {
2884 /*
2885 * We have failed to do this the quick way;
2886 * we must resort to the slower approach of
2887 * calling dtrace_getpcstack().
2888 */
2889 pc_t caller = 0;
2890
2891 dtrace_getpcstack(&caller, 1, aframes, NULL);
2892 mstate->dtms_caller = caller;
2893 }
2894
2895 mstate->dtms_present |= DTRACE_MSTATE_CALLER;
2896 }
2897 return (mstate->dtms_caller);
2898
2899#if defined(sun)
2900 case DIF_VAR_UCALLER:
2901 if (!dtrace_priv_proc(state))
2902 return (0);
2903
2904 if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) {
2905 uint64_t ustack[3];
2906
2907 /*
2908 * dtrace_getupcstack() fills in the first uint64_t
2909 * with the current PID. The second uint64_t will
2910 * be the program counter at user-level. The third
2911 * uint64_t will contain the caller, which is what
2912 * we're after.
2913 */
2914 ustack[2] = 0;
2915 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
2916 dtrace_getupcstack(ustack, 3);
2917 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
2918 mstate->dtms_ucaller = ustack[2];
2919 mstate->dtms_present |= DTRACE_MSTATE_UCALLER;
2920 }
2921
2922 return (mstate->dtms_ucaller);
2923#endif
2924
2925 case DIF_VAR_PROBEPROV:
2926 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2927 return (dtrace_dif_varstr(
2928 (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name,
2929 state, mstate));
2930
2931 case DIF_VAR_PROBEMOD:
2932 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2933 return (dtrace_dif_varstr(
2934 (uintptr_t)mstate->dtms_probe->dtpr_mod,
2935 state, mstate));
2936
2937 case DIF_VAR_PROBEFUNC:
2938 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2939 return (dtrace_dif_varstr(
2940 (uintptr_t)mstate->dtms_probe->dtpr_func,
2941 state, mstate));
2942
2943 case DIF_VAR_PROBENAME:
2944 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE);
2945 return (dtrace_dif_varstr(
2946 (uintptr_t)mstate->dtms_probe->dtpr_name,
2947 state, mstate));
2948
2949 case DIF_VAR_PID:
2950 if (!dtrace_priv_proc(state))
2951 return (0);
2952
2953#if defined(sun)
2954 /*
2955 * Note that we are assuming that an unanchored probe is
2956 * always due to a high-level interrupt. (And we're assuming
2957 * that there is only a single high level interrupt.)
2958 */
2959 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
2960 return (pid0.pid_id);
2961
2962 /*
2963 * It is always safe to dereference one's own t_procp pointer:
2964 * it always points to a valid, allocated proc structure.
2965 * Further, it is always safe to dereference the p_pidp member
2966 * of one's own proc structure. (These are truisms becuase
2967 * threads and processes don't clean up their own state --
2968 * they leave that task to whomever reaps them.)
2969 */
2970 return ((uint64_t)curthread->t_procp->p_pidp->pid_id);
2971#else
2972 return ((uint64_t)curproc->p_pid);
2973#endif
2974
2975 case DIF_VAR_PPID:
2976 if (!dtrace_priv_proc(state))
2977 return (0);
2978
2979#if defined(sun)
2980 /*
2981 * See comment in DIF_VAR_PID.
2982 */
2983 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
2984 return (pid0.pid_id);
2985
2986 /*
2987 * It is always safe to dereference one's own t_procp pointer:
2988 * it always points to a valid, allocated proc structure.
2989 * (This is true because threads don't clean up their own
2990 * state -- they leave that task to whomever reaps them.)
2991 */
2992 return ((uint64_t)curthread->t_procp->p_ppid);
2993#else
2994 return ((uint64_t)curproc->p_pptr->p_pid);
2995#endif
2996
2997 case DIF_VAR_TID:
2998#if defined(sun)
2999 /*
3000 * See comment in DIF_VAR_PID.
3001 */
3002 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3003 return (0);
3004#endif
3005
3006 return ((uint64_t)curthread->t_tid);
3007
3008 case DIF_VAR_EXECARGS: {
3009 struct pargs *p_args = curthread->td_proc->p_args;
3010
3011 if (p_args == NULL)
3012 return(0);
3013
3014 return (dtrace_dif_varstrz(
3015 (uintptr_t) p_args->ar_args, p_args->ar_length, state, mstate));
3016 }
3017
3018 case DIF_VAR_EXECNAME:
3019#if defined(sun)
3020 if (!dtrace_priv_proc(state))
3021 return (0);
3022
3023 /*
3024 * See comment in DIF_VAR_PID.
3025 */
3026 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3027 return ((uint64_t)(uintptr_t)p0.p_user.u_comm);
3028
3029 /*
3030 * It is always safe to dereference one's own t_procp pointer:
3031 * it always points to a valid, allocated proc structure.
3032 * (This is true because threads don't clean up their own
3033 * state -- they leave that task to whomever reaps them.)
3034 */
3035 return (dtrace_dif_varstr(
3036 (uintptr_t)curthread->t_procp->p_user.u_comm,
3037 state, mstate));
3038#else
3039 return (dtrace_dif_varstr(
3040 (uintptr_t) curthread->td_proc->p_comm, state, mstate));
3041#endif
3042
3043 case DIF_VAR_ZONENAME:
3044#if defined(sun)
3045 if (!dtrace_priv_proc(state))
3046 return (0);
3047
3048 /*
3049 * See comment in DIF_VAR_PID.
3050 */
3051 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3052 return ((uint64_t)(uintptr_t)p0.p_zone->zone_name);
3053
3054 /*
3055 * It is always safe to dereference one's own t_procp pointer:
3056 * it always points to a valid, allocated proc structure.
3057 * (This is true because threads don't clean up their own
3058 * state -- they leave that task to whomever reaps them.)
3059 */
3060 return (dtrace_dif_varstr(
3061 (uintptr_t)curthread->t_procp->p_zone->zone_name,
3062 state, mstate));
3063#else
3064 return (0);
3065#endif
3066
3067 case DIF_VAR_UID:
3068 if (!dtrace_priv_proc(state))
3069 return (0);
3070
3071#if defined(sun)
3072 /*
3073 * See comment in DIF_VAR_PID.
3074 */
3075 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3076 return ((uint64_t)p0.p_cred->cr_uid);
3077#endif
3078
3079 /*
3080 * It is always safe to dereference one's own t_procp pointer:
3081 * it always points to a valid, allocated proc structure.
3082 * (This is true because threads don't clean up their own
3083 * state -- they leave that task to whomever reaps them.)
3084 *
3085 * Additionally, it is safe to dereference one's own process
3086 * credential, since this is never NULL after process birth.
3087 */
3088 return ((uint64_t)curthread->t_procp->p_cred->cr_uid);
3089
3090 case DIF_VAR_GID:
3091 if (!dtrace_priv_proc(state))
3092 return (0);
3093
3094#if defined(sun)
3095 /*
3096 * See comment in DIF_VAR_PID.
3097 */
3098 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3099 return ((uint64_t)p0.p_cred->cr_gid);
3100#endif
3101
3102 /*
3103 * It is always safe to dereference one's own t_procp pointer:
3104 * it always points to a valid, allocated proc structure.
3105 * (This is true because threads don't clean up their own
3106 * state -- they leave that task to whomever reaps them.)
3107 *
3108 * Additionally, it is safe to dereference one's own process
3109 * credential, since this is never NULL after process birth.
3110 */
3111 return ((uint64_t)curthread->t_procp->p_cred->cr_gid);
3112
3113 case DIF_VAR_ERRNO: {
3114#if defined(sun)
3115 klwp_t *lwp;
3116 if (!dtrace_priv_proc(state))
3117 return (0);
3118
3119 /*
3120 * See comment in DIF_VAR_PID.
3121 */
3122 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU))
3123 return (0);
3124
3125 /*
3126 * It is always safe to dereference one's own t_lwp pointer in
3127 * the event that this pointer is non-NULL. (This is true
3128 * because threads and lwps don't clean up their own state --
3129 * they leave that task to whomever reaps them.)
3130 */
3131 if ((lwp = curthread->t_lwp) == NULL)
3132 return (0);
3133
3134 return ((uint64_t)lwp->lwp_errno);
3135#else
3136 return (curthread->td_errno);
3137#endif
3138 }
3139 default:
3140 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
3141 return (0);
3142 }
3143}
3144
3145/*
3146 * Emulate the execution of DTrace ID subroutines invoked by the call opcode.
3147 * Notice that we don't bother validating the proper number of arguments or
3148 * their types in the tuple stack. This isn't needed because all argument
3149 * interpretation is safe because of our load safety -- the worst that can
3150 * happen is that a bogus program can obtain bogus results.
3151 */
3152static void
3153dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs,
3154 dtrace_key_t *tupregs, int nargs,
3155 dtrace_mstate_t *mstate, dtrace_state_t *state)
3156{
3157 volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
3158 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
3159 dtrace_vstate_t *vstate = &state->dts_vstate;
3160
3161#if defined(sun)
3162 union {
3163 mutex_impl_t mi;
3164 uint64_t mx;
3165 } m;
3166
3167 union {
3168 krwlock_t ri;
3169 uintptr_t rw;
3170 } r;
3171#else
3172 struct thread *lowner;
3173 union {
3174 struct lock_object *li;
3175 uintptr_t lx;
3176 } l;
3177#endif
3178
3179 switch (subr) {
3180 case DIF_SUBR_RAND:
3181 regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875;
3182 break;
3183
3184#if defined(sun)
3185 case DIF_SUBR_MUTEX_OWNED:
3186 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3187 mstate, vstate)) {
3188 regs[rd] = 0;
3189 break;
3190 }
3191
3192 m.mx = dtrace_load64(tupregs[0].dttk_value);
3193 if (MUTEX_TYPE_ADAPTIVE(&m.mi))
3194 regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER;
3195 else
3196 regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock);
3197 break;
3198
3199 case DIF_SUBR_MUTEX_OWNER:
3200 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3201 mstate, vstate)) {
3202 regs[rd] = 0;
3203 break;
3204 }
3205
3206 m.mx = dtrace_load64(tupregs[0].dttk_value);
3207 if (MUTEX_TYPE_ADAPTIVE(&m.mi) &&
3208 MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER)
3209 regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi);
3210 else
3211 regs[rd] = 0;
3212 break;
3213
3214 case DIF_SUBR_MUTEX_TYPE_ADAPTIVE:
3215 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3216 mstate, vstate)) {
3217 regs[rd] = 0;
3218 break;
3219 }
3220
3221 m.mx = dtrace_load64(tupregs[0].dttk_value);
3222 regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi);
3223 break;
3224
3225 case DIF_SUBR_MUTEX_TYPE_SPIN:
3226 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t),
3227 mstate, vstate)) {
3228 regs[rd] = 0;
3229 break;
3230 }
3231
3232 m.mx = dtrace_load64(tupregs[0].dttk_value);
3233 regs[rd] = MUTEX_TYPE_SPIN(&m.mi);
3234 break;
3235
3236 case DIF_SUBR_RW_READ_HELD: {
3237 uintptr_t tmp;
3238
3239 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3240 mstate, vstate)) {
3241 regs[rd] = 0;
3242 break;
3243 }
3244
3245 r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3246 regs[rd] = _RW_READ_HELD(&r.ri, tmp);
3247 break;
3248 }
3249
3250 case DIF_SUBR_RW_WRITE_HELD:
3251 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t),
3252 mstate, vstate)) {
3253 regs[rd] = 0;
3254 break;
3255 }
3256
3257 r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3258 regs[rd] = _RW_WRITE_HELD(&r.ri);
3259 break;
3260
3261 case DIF_SUBR_RW_ISWRITER:
3262 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t),
3263 mstate, vstate)) {
3264 regs[rd] = 0;
3265 break;
3266 }
3267
3268 r.rw = dtrace_loadptr(tupregs[0].dttk_value);
3269 regs[rd] = _RW_ISWRITER(&r.ri);
3270 break;
3271
3272#else
3273 case DIF_SUBR_MUTEX_OWNED:
3274 if (!dtrace_canload(tupregs[0].dttk_value,
3275 sizeof (struct lock_object), mstate, vstate)) {
3276 regs[rd] = 0;
3277 break;
3278 }
3279 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3280 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3281 break;
3282
3283 case DIF_SUBR_MUTEX_OWNER:
3284 if (!dtrace_canload(tupregs[0].dttk_value,
3285 sizeof (struct lock_object), mstate, vstate)) {
3286 regs[rd] = 0;
3287 break;
3288 }
3289 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3290 LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3291 regs[rd] = (uintptr_t)lowner;
3292 break;
3293
3294 case DIF_SUBR_MUTEX_TYPE_ADAPTIVE:
3295 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (struct mtx),
3296 mstate, vstate)) {
3297 regs[rd] = 0;
3298 break;
3299 }
3300 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3301 /* XXX - should be only LC_SLEEPABLE? */
3302 regs[rd] = (LOCK_CLASS(l.li)->lc_flags &
3303 (LC_SLEEPLOCK | LC_SLEEPABLE)) != 0;
3304 break;
3305
3306 case DIF_SUBR_MUTEX_TYPE_SPIN:
3307 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (struct mtx),
3308 mstate, vstate)) {
3309 regs[rd] = 0;
3310 break;
3311 }
3312 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3313 regs[rd] = (LOCK_CLASS(l.li)->lc_flags & LC_SPINLOCK) != 0;
3314 break;
3315
3316 case DIF_SUBR_RW_READ_HELD:
3317 case DIF_SUBR_SX_SHARED_HELD:
3318 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3319 mstate, vstate)) {
3320 regs[rd] = 0;
3321 break;
3322 }
3323 l.lx = dtrace_loadptr((uintptr_t)&tupregs[0].dttk_value);
3324 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner) &&
3325 lowner == NULL;
3326 break;
3327
3328 case DIF_SUBR_RW_WRITE_HELD:
3329 case DIF_SUBR_SX_EXCLUSIVE_HELD:
3330 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3331 mstate, vstate)) {
3332 regs[rd] = 0;
3333 break;
3334 }
3335 l.lx = dtrace_loadptr(tupregs[0].dttk_value);
3336 LOCK_CLASS(l.li)->lc_owner(l.li, &lowner);
3337 regs[rd] = (lowner == curthread);
3338 break;
3339
3340 case DIF_SUBR_RW_ISWRITER:
3341 case DIF_SUBR_SX_ISEXCLUSIVE:
3342 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t),
3343 mstate, vstate)) {
3344 regs[rd] = 0;
3345 break;
3346 }
3347 l.lx = dtrace_loadptr(tupregs[0].dttk_value);
3348 regs[rd] = LOCK_CLASS(l.li)->lc_owner(l.li, &lowner) &&
3349 lowner != NULL;
3350 break;
3351#endif /* ! defined(sun) */
3352
3353 case DIF_SUBR_BCOPY: {
3354 /*
3355 * We need to be sure that the destination is in the scratch
3356 * region -- no other region is allowed.
3357 */
3358 uintptr_t src = tupregs[0].dttk_value;
3359 uintptr_t dest = tupregs[1].dttk_value;
3360 size_t size = tupregs[2].dttk_value;
3361
3362 if (!dtrace_inscratch(dest, size, mstate)) {
3363 *flags |= CPU_DTRACE_BADADDR;
3364 *illval = regs[rd];
3365 break;
3366 }
3367
3368 if (!dtrace_canload(src, size, mstate, vstate)) {
3369 regs[rd] = 0;
3370 break;
3371 }
3372
3373 dtrace_bcopy((void *)src, (void *)dest, size);
3374 break;
3375 }
3376
3377 case DIF_SUBR_ALLOCA:
3378 case DIF_SUBR_COPYIN: {
3379 uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
3380 uint64_t size =
3381 tupregs[subr == DIF_SUBR_ALLOCA ? 0 : 1].dttk_value;
3382 size_t scratch_size = (dest - mstate->dtms_scratch_ptr) + size;
3383
3384 /*
3385 * This action doesn't require any credential checks since
3386 * probes will not activate in user contexts to which the
3387 * enabling user does not have permissions.
3388 */
3389
3390 /*
3391 * Rounding up the user allocation size could have overflowed
3392 * a large, bogus allocation (like -1ULL) to 0.
3393 */
3394 if (scratch_size < size ||
3395 !DTRACE_INSCRATCH(mstate, scratch_size)) {
3396 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3397 regs[rd] = 0;
3398 break;
3399 }
3400
3401 if (subr == DIF_SUBR_COPYIN) {
3402 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3403 dtrace_copyin(tupregs[0].dttk_value, dest, size, flags);
3404 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3405 }
3406
3407 mstate->dtms_scratch_ptr += scratch_size;
3408 regs[rd] = dest;
3409 break;
3410 }
3411
3412 case DIF_SUBR_COPYINTO: {
3413 uint64_t size = tupregs[1].dttk_value;
3414 uintptr_t dest = tupregs[2].dttk_value;
3415
3416 /*
3417 * This action doesn't require any credential checks since
3418 * probes will not activate in user contexts to which the
3419 * enabling user does not have permissions.
3420 */
3421 if (!dtrace_inscratch(dest, size, mstate)) {
3422 *flags |= CPU_DTRACE_BADADDR;
3423 *illval = regs[rd];
3424 break;
3425 }
3426
3427 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3428 dtrace_copyin(tupregs[0].dttk_value, dest, size, flags);
3429 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3430 break;
3431 }
3432
3433 case DIF_SUBR_COPYINSTR: {
3434 uintptr_t dest = mstate->dtms_scratch_ptr;
3435 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3436
3437 if (nargs > 1 && tupregs[1].dttk_value < size)
3438 size = tupregs[1].dttk_value + 1;
3439
3440 /*
3441 * This action doesn't require any credential checks since
3442 * probes will not activate in user contexts to which the
3443 * enabling user does not have permissions.
3444 */
3445 if (!DTRACE_INSCRATCH(mstate, size)) {
3446 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3447 regs[rd] = 0;
3448 break;
3449 }
3450
3451 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3452 dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags);
3453 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3454
3455 ((char *)dest)[size - 1] = '\0';
3456 mstate->dtms_scratch_ptr += size;
3457 regs[rd] = dest;
3458 break;
3459 }
3460
3461#if defined(sun)
3462 case DIF_SUBR_MSGSIZE:
3463 case DIF_SUBR_MSGDSIZE: {
3464 uintptr_t baddr = tupregs[0].dttk_value, daddr;
3465 uintptr_t wptr, rptr;
3466 size_t count = 0;
3467 int cont = 0;
3468
3469 while (baddr != 0 && !(*flags & CPU_DTRACE_FAULT)) {
3470
3471 if (!dtrace_canload(baddr, sizeof (mblk_t), mstate,
3472 vstate)) {
3473 regs[rd] = 0;
3474 break;
3475 }
3476
3477 wptr = dtrace_loadptr(baddr +
3478 offsetof(mblk_t, b_wptr));
3479
3480 rptr = dtrace_loadptr(baddr +
3481 offsetof(mblk_t, b_rptr));
3482
3483 if (wptr < rptr) {
3484 *flags |= CPU_DTRACE_BADADDR;
3485 *illval = tupregs[0].dttk_value;
3486 break;
3487 }
3488
3489 daddr = dtrace_loadptr(baddr +
3490 offsetof(mblk_t, b_datap));
3491
3492 baddr = dtrace_loadptr(baddr +
3493 offsetof(mblk_t, b_cont));
3494
3495 /*
3496 * We want to prevent against denial-of-service here,
3497 * so we're only going to search the list for
3498 * dtrace_msgdsize_max mblks.
3499 */
3500 if (cont++ > dtrace_msgdsize_max) {
3501 *flags |= CPU_DTRACE_ILLOP;
3502 break;
3503 }
3504
3505 if (subr == DIF_SUBR_MSGDSIZE) {
3506 if (dtrace_load8(daddr +
3507 offsetof(dblk_t, db_type)) != M_DATA)
3508 continue;
3509 }
3510
3511 count += wptr - rptr;
3512 }
3513
3514 if (!(*flags & CPU_DTRACE_FAULT))
3515 regs[rd] = count;
3516
3517 break;
3518 }
3519#endif
3520
3521 case DIF_SUBR_PROGENYOF: {
3522 pid_t pid = tupregs[0].dttk_value;
3523 proc_t *p;
3524 int rval = 0;
3525
3526 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3527
3528 for (p = curthread->t_procp; p != NULL; p = p->p_parent) {
3529#if defined(sun)
3530 if (p->p_pidp->pid_id == pid) {
3531#else
3532 if (p->p_pid == pid) {
3533#endif
3534 rval = 1;
3535 break;
3536 }
3537 }
3538
3539 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3540
3541 regs[rd] = rval;
3542 break;
3543 }
3544
3545 case DIF_SUBR_SPECULATION:
3546 regs[rd] = dtrace_speculation(state);
3547 break;
3548
3549 case DIF_SUBR_COPYOUT: {
3550 uintptr_t kaddr = tupregs[0].dttk_value;
3551 uintptr_t uaddr = tupregs[1].dttk_value;
3552 uint64_t size = tupregs[2].dttk_value;
3553
3554 if (!dtrace_destructive_disallow &&
3555 dtrace_priv_proc_control(state) &&
3556 !dtrace_istoxic(kaddr, size)) {
3557 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3558 dtrace_copyout(kaddr, uaddr, size, flags);
3559 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3560 }
3561 break;
3562 }
3563
3564 case DIF_SUBR_COPYOUTSTR: {
3565 uintptr_t kaddr = tupregs[0].dttk_value;
3566 uintptr_t uaddr = tupregs[1].dttk_value;
3567 uint64_t size = tupregs[2].dttk_value;
3568
3569 if (!dtrace_destructive_disallow &&
3570 dtrace_priv_proc_control(state) &&
3571 !dtrace_istoxic(kaddr, size)) {
3572 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
3573 dtrace_copyoutstr(kaddr, uaddr, size, flags);
3574 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
3575 }
3576 break;
3577 }
3578
3579 case DIF_SUBR_STRLEN: {
3580 size_t sz;
3581 uintptr_t addr = (uintptr_t)tupregs[0].dttk_value;
3582 sz = dtrace_strlen((char *)addr,
3583 state->dts_options[DTRACEOPT_STRSIZE]);
3584
3585 if (!dtrace_canload(addr, sz + 1, mstate, vstate)) {
3586 regs[rd] = 0;
3587 break;
3588 }
3589
3590 regs[rd] = sz;
3591
3592 break;
3593 }
3594
3595 case DIF_SUBR_STRCHR:
3596 case DIF_SUBR_STRRCHR: {
3597 /*
3598 * We're going to iterate over the string looking for the
3599 * specified character. We will iterate until we have reached
3600 * the string length or we have found the character. If this
3601 * is DIF_SUBR_STRRCHR, we will look for the last occurrence
3602 * of the specified character instead of the first.
3603 */
3604 uintptr_t saddr = tupregs[0].dttk_value;
3605 uintptr_t addr = tupregs[0].dttk_value;
3606 uintptr_t limit = addr + state->dts_options[DTRACEOPT_STRSIZE];
3607 char c, target = (char)tupregs[1].dttk_value;
3608
3609 for (regs[rd] = 0; addr < limit; addr++) {
3610 if ((c = dtrace_load8(addr)) == target) {
3611 regs[rd] = addr;
3612
3613 if (subr == DIF_SUBR_STRCHR)
3614 break;
3615 }
3616
3617 if (c == '\0')
3618 break;
3619 }
3620
3621 if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) {
3622 regs[rd] = 0;
3623 break;
3624 }
3625
3626 break;
3627 }
3628
3629 case DIF_SUBR_STRSTR:
3630 case DIF_SUBR_INDEX:
3631 case DIF_SUBR_RINDEX: {
3632 /*
3633 * We're going to iterate over the string looking for the
3634 * specified string. We will iterate until we have reached
3635 * the string length or we have found the string. (Yes, this
3636 * is done in the most naive way possible -- but considering
3637 * that the string we're searching for is likely to be
3638 * relatively short, the complexity of Rabin-Karp or similar
3639 * hardly seems merited.)
3640 */
3641 char *addr = (char *)(uintptr_t)tupregs[0].dttk_value;
3642 char *substr = (char *)(uintptr_t)tupregs[1].dttk_value;
3643 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3644 size_t len = dtrace_strlen(addr, size);
3645 size_t sublen = dtrace_strlen(substr, size);
3646 char *limit = addr + len, *orig = addr;
3647 int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1;
3648 int inc = 1;
3649
3650 regs[rd] = notfound;
3651
3652 if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) {
3653 regs[rd] = 0;
3654 break;
3655 }
3656
3657 if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate,
3658 vstate)) {
3659 regs[rd] = 0;
3660 break;
3661 }
3662
3663 /*
3664 * strstr() and index()/rindex() have similar semantics if
3665 * both strings are the empty string: strstr() returns a
3666 * pointer to the (empty) string, and index() and rindex()
3667 * both return index 0 (regardless of any position argument).
3668 */
3669 if (sublen == 0 && len == 0) {
3670 if (subr == DIF_SUBR_STRSTR)
3671 regs[rd] = (uintptr_t)addr;
3672 else
3673 regs[rd] = 0;
3674 break;
3675 }
3676
3677 if (subr != DIF_SUBR_STRSTR) {
3678 if (subr == DIF_SUBR_RINDEX) {
3679 limit = orig - 1;
3680 addr += len;
3681 inc = -1;
3682 }
3683
3684 /*
3685 * Both index() and rindex() take an optional position
3686 * argument that denotes the starting position.
3687 */
3688 if (nargs == 3) {
3689 int64_t pos = (int64_t)tupregs[2].dttk_value;
3690
3691 /*
3692 * If the position argument to index() is
3693 * negative, Perl implicitly clamps it at
3694 * zero. This semantic is a little surprising
3695 * given the special meaning of negative
3696 * positions to similar Perl functions like
3697 * substr(), but it appears to reflect a
3698 * notion that index() can start from a
3699 * negative index and increment its way up to
3700 * the string. Given this notion, Perl's
3701 * rindex() is at least self-consistent in
3702 * that it implicitly clamps positions greater
3703 * than the string length to be the string
3704 * length. Where Perl completely loses
3705 * coherence, however, is when the specified
3706 * substring is the empty string (""). In
3707 * this case, even if the position is
3708 * negative, rindex() returns 0 -- and even if
3709 * the position is greater than the length,
3710 * index() returns the string length. These
3711 * semantics violate the notion that index()
3712 * should never return a value less than the
3713 * specified position and that rindex() should
3714 * never return a value greater than the
3715 * specified position. (One assumes that
3716 * these semantics are artifacts of Perl's
3717 * implementation and not the results of
3718 * deliberate design -- it beggars belief that
3719 * even Larry Wall could desire such oddness.)
3720 * While in the abstract one would wish for
3721 * consistent position semantics across
3722 * substr(), index() and rindex() -- or at the
3723 * very least self-consistent position
3724 * semantics for index() and rindex() -- we
3725 * instead opt to keep with the extant Perl
3726 * semantics, in all their broken glory. (Do
3727 * we have more desire to maintain Perl's
3728 * semantics than Perl does? Probably.)
3729 */
3730 if (subr == DIF_SUBR_RINDEX) {
3731 if (pos < 0) {
3732 if (sublen == 0)
3733 regs[rd] = 0;
3734 break;
3735 }
3736
3737 if (pos > len)
3738 pos = len;
3739 } else {
3740 if (pos < 0)
3741 pos = 0;
3742
3743 if (pos >= len) {
3744 if (sublen == 0)
3745 regs[rd] = len;
3746 break;
3747 }
3748 }
3749
3750 addr = orig + pos;
3751 }
3752 }
3753
3754 for (regs[rd] = notfound; addr != limit; addr += inc) {
3755 if (dtrace_strncmp(addr, substr, sublen) == 0) {
3756 if (subr != DIF_SUBR_STRSTR) {
3757 /*
3758 * As D index() and rindex() are
3759 * modeled on Perl (and not on awk),
3760 * we return a zero-based (and not a
3761 * one-based) index. (For you Perl
3762 * weenies: no, we're not going to add
3763 * $[ -- and shouldn't you be at a con
3764 * or something?)
3765 */
3766 regs[rd] = (uintptr_t)(addr - orig);
3767 break;
3768 }
3769
3770 ASSERT(subr == DIF_SUBR_STRSTR);
3771 regs[rd] = (uintptr_t)addr;
3772 break;
3773 }
3774 }
3775
3776 break;
3777 }
3778
3779 case DIF_SUBR_STRTOK: {
3780 uintptr_t addr = tupregs[0].dttk_value;
3781 uintptr_t tokaddr = tupregs[1].dttk_value;
3782 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3783 uintptr_t limit, toklimit = tokaddr + size;
3784 uint8_t c = 0, tokmap[32]; /* 256 / 8 */
3785 char *dest = (char *)mstate->dtms_scratch_ptr;
3786 int i;
3787
3788 /*
3789 * Check both the token buffer and (later) the input buffer,
3790 * since both could be non-scratch addresses.
3791 */
3792 if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) {
3793 regs[rd] = 0;
3794 break;
3795 }
3796
3797 if (!DTRACE_INSCRATCH(mstate, size)) {
3798 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3799 regs[rd] = 0;
3800 break;
3801 }
3802
3803 if (addr == 0) {
3804 /*
3805 * If the address specified is NULL, we use our saved
3806 * strtok pointer from the mstate. Note that this
3807 * means that the saved strtok pointer is _only_
3808 * valid within multiple enablings of the same probe --
3809 * it behaves like an implicit clause-local variable.
3810 */
3811 addr = mstate->dtms_strtok;
3812 } else {
3813 /*
3814 * If the user-specified address is non-NULL we must
3815 * access check it. This is the only time we have
3816 * a chance to do so, since this address may reside
3817 * in the string table of this clause-- future calls
3818 * (when we fetch addr from mstate->dtms_strtok)
3819 * would fail this access check.
3820 */
3821 if (!dtrace_strcanload(addr, size, mstate, vstate)) {
3822 regs[rd] = 0;
3823 break;
3824 }
3825 }
3826
3827 /*
3828 * First, zero the token map, and then process the token
3829 * string -- setting a bit in the map for every character
3830 * found in the token string.
3831 */
3832 for (i = 0; i < sizeof (tokmap); i++)
3833 tokmap[i] = 0;
3834
3835 for (; tokaddr < toklimit; tokaddr++) {
3836 if ((c = dtrace_load8(tokaddr)) == '\0')
3837 break;
3838
3839 ASSERT((c >> 3) < sizeof (tokmap));
3840 tokmap[c >> 3] |= (1 << (c & 0x7));
3841 }
3842
3843 for (limit = addr + size; addr < limit; addr++) {
3844 /*
3845 * We're looking for a character that is _not_ contained
3846 * in the token string.
3847 */
3848 if ((c = dtrace_load8(addr)) == '\0')
3849 break;
3850
3851 if (!(tokmap[c >> 3] & (1 << (c & 0x7))))
3852 break;
3853 }
3854
3855 if (c == '\0') {
3856 /*
3857 * We reached the end of the string without finding
3858 * any character that was not in the token string.
3859 * We return NULL in this case, and we set the saved
3860 * address to NULL as well.
3861 */
3862 regs[rd] = 0;
3863 mstate->dtms_strtok = 0;
3864 break;
3865 }
3866
3867 /*
3868 * From here on, we're copying into the destination string.
3869 */
3870 for (i = 0; addr < limit && i < size - 1; addr++) {
3871 if ((c = dtrace_load8(addr)) == '\0')
3872 break;
3873
3874 if (tokmap[c >> 3] & (1 << (c & 0x7)))
3875 break;
3876
3877 ASSERT(i < size);
3878 dest[i++] = c;
3879 }
3880
3881 ASSERT(i < size);
3882 dest[i] = '\0';
3883 regs[rd] = (uintptr_t)dest;
3884 mstate->dtms_scratch_ptr += size;
3885 mstate->dtms_strtok = addr;
3886 break;
3887 }
3888
3889 case DIF_SUBR_SUBSTR: {
3890 uintptr_t s = tupregs[0].dttk_value;
3891 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3892 char *d = (char *)mstate->dtms_scratch_ptr;
3893 int64_t index = (int64_t)tupregs[1].dttk_value;
3894 int64_t remaining = (int64_t)tupregs[2].dttk_value;
3895 size_t len = dtrace_strlen((char *)s, size);
3896 int64_t i = 0;
3897
3898 if (!dtrace_canload(s, len + 1, mstate, vstate)) {
3899 regs[rd] = 0;
3900 break;
3901 }
3902
3903 if (!DTRACE_INSCRATCH(mstate, size)) {
3904 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3905 regs[rd] = 0;
3906 break;
3907 }
3908
3909 if (nargs <= 2)
3910 remaining = (int64_t)size;
3911
3912 if (index < 0) {
3913 index += len;
3914
3915 if (index < 0 && index + remaining > 0) {
3916 remaining += index;
3917 index = 0;
3918 }
3919 }
3920
3921 if (index >= len || index < 0) {
3922 remaining = 0;
3923 } else if (remaining < 0) {
3924 remaining += len - index;
3925 } else if (index + remaining > size) {
3926 remaining = size - index;
3927 }
3928
3929 for (i = 0; i < remaining; i++) {
3930 if ((d[i] = dtrace_load8(s + index + i)) == '\0')
3931 break;
3932 }
3933
3934 d[i] = '\0';
3935
3936 mstate->dtms_scratch_ptr += size;
3937 regs[rd] = (uintptr_t)d;
3938 break;
3939 }
3940
3941#if defined(sun)
3942 case DIF_SUBR_GETMAJOR:
3943#ifdef _LP64
3944 regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR64) & MAXMAJ64;
3945#else
3946 regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR) & MAXMAJ;
3947#endif
3948 break;
3949
3950 case DIF_SUBR_GETMINOR:
3951#ifdef _LP64
3952 regs[rd] = tupregs[0].dttk_value & MAXMIN64;
3953#else
3954 regs[rd] = tupregs[0].dttk_value & MAXMIN;
3955#endif
3956 break;
3957
3958 case DIF_SUBR_DDI_PATHNAME: {
3959 /*
3960 * This one is a galactic mess. We are going to roughly
3961 * emulate ddi_pathname(), but it's made more complicated
3962 * by the fact that we (a) want to include the minor name and
3963 * (b) must proceed iteratively instead of recursively.
3964 */
3965 uintptr_t dest = mstate->dtms_scratch_ptr;
3966 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
3967 char *start = (char *)dest, *end = start + size - 1;
3968 uintptr_t daddr = tupregs[0].dttk_value;
3969 int64_t minor = (int64_t)tupregs[1].dttk_value;
3970 char *s;
3971 int i, len, depth = 0;
3972
3973 /*
3974 * Due to all the pointer jumping we do and context we must
3975 * rely upon, we just mandate that the user must have kernel
3976 * read privileges to use this routine.
3977 */
3978 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) {
3979 *flags |= CPU_DTRACE_KPRIV;
3980 *illval = daddr;
3981 regs[rd] = 0;
3982 }
3983
3984 if (!DTRACE_INSCRATCH(mstate, size)) {
3985 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
3986 regs[rd] = 0;
3987 break;
3988 }
3989
3990 *end = '\0';
3991
3992 /*
3993 * We want to have a name for the minor. In order to do this,
3994 * we need to walk the minor list from the devinfo. We want
3995 * to be sure that we don't infinitely walk a circular list,
3996 * so we check for circularity by sending a scout pointer
3997 * ahead two elements for every element that we iterate over;
3998 * if the list is circular, these will ultimately point to the
3999 * same element. You may recognize this little trick as the
4000 * answer to a stupid interview question -- one that always
4001 * seems to be asked by those who had to have it laboriously
4002 * explained to them, and who can't even concisely describe
4003 * the conditions under which one would be forced to resort to
4004 * this technique. Needless to say, those conditions are
4005 * found here -- and probably only here. Is this the only use
4006 * of this infamous trick in shipping, production code? If it
4007 * isn't, it probably should be...
4008 */
4009 if (minor != -1) {
4010 uintptr_t maddr = dtrace_loadptr(daddr +
4011 offsetof(struct dev_info, devi_minor));
4012
4013 uintptr_t next = offsetof(struct ddi_minor_data, next);
4014 uintptr_t name = offsetof(struct ddi_minor_data,
4015 d_minor) + offsetof(struct ddi_minor, name);
4016 uintptr_t dev = offsetof(struct ddi_minor_data,
4017 d_minor) + offsetof(struct ddi_minor, dev);
4018 uintptr_t scout;
4019
4020 if (maddr != NULL)
4021 scout = dtrace_loadptr(maddr + next);
4022
4023 while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4024 uint64_t m;
4025#ifdef _LP64
4026 m = dtrace_load64(maddr + dev) & MAXMIN64;
4027#else
4028 m = dtrace_load32(maddr + dev) & MAXMIN;
4029#endif
4030 if (m != minor) {
4031 maddr = dtrace_loadptr(maddr + next);
4032
4033 if (scout == NULL)
4034 continue;
4035
4036 scout = dtrace_loadptr(scout + next);
4037
4038 if (scout == NULL)
4039 continue;
4040
4041 scout = dtrace_loadptr(scout + next);
4042
4043 if (scout == NULL)
4044 continue;
4045
4046 if (scout == maddr) {
4047 *flags |= CPU_DTRACE_ILLOP;
4048 break;
4049 }
4050
4051 continue;
4052 }
4053
4054 /*
4055 * We have the minor data. Now we need to
4056 * copy the minor's name into the end of the
4057 * pathname.
4058 */
4059 s = (char *)dtrace_loadptr(maddr + name);
4060 len = dtrace_strlen(s, size);
4061
4062 if (*flags & CPU_DTRACE_FAULT)
4063 break;
4064
4065 if (len != 0) {
4066 if ((end -= (len + 1)) < start)
4067 break;
4068
4069 *end = ':';
4070 }
4071
4072 for (i = 1; i <= len; i++)
4073 end[i] = dtrace_load8((uintptr_t)s++);
4074 break;
4075 }
4076 }
4077
4078 while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) {
4079 ddi_node_state_t devi_state;
4080
4081 devi_state = dtrace_load32(daddr +
4082 offsetof(struct dev_info, devi_node_state));
4083
4084 if (*flags & CPU_DTRACE_FAULT)
4085 break;
4086
4087 if (devi_state >= DS_INITIALIZED) {
4088 s = (char *)dtrace_loadptr(daddr +
4089 offsetof(struct dev_info, devi_addr));
4090 len = dtrace_strlen(s, size);
4091
4092 if (*flags & CPU_DTRACE_FAULT)
4093 break;
4094
4095 if (len != 0) {
4096 if ((end -= (len + 1)) < start)
4097 break;
4098
4099 *end = '@';
4100 }
4101
4102 for (i = 1; i <= len; i++)
4103 end[i] = dtrace_load8((uintptr_t)s++);
4104 }
4105
4106 /*
4107 * Now for the node name...
4108 */
4109 s = (char *)dtrace_loadptr(daddr +
4110 offsetof(struct dev_info, devi_node_name));
4111
4112 daddr = dtrace_loadptr(daddr +
4113 offsetof(struct dev_info, devi_parent));
4114
4115 /*
4116 * If our parent is NULL (that is, if we're the root
4117 * node), we're going to use the special path
4118 * "devices".
4119 */
4120 if (daddr == 0)
4121 s = "devices";
4122
4123 len = dtrace_strlen(s, size);
4124 if (*flags & CPU_DTRACE_FAULT)
4125 break;
4126
4127 if ((end -= (len + 1)) < start)
4128 break;
4129
4130 for (i = 1; i <= len; i++)
4131 end[i] = dtrace_load8((uintptr_t)s++);
4132 *end = '/';
4133
4134 if (depth++ > dtrace_devdepth_max) {
4135 *flags |= CPU_DTRACE_ILLOP;
4136 break;
4137 }
4138 }
4139
4140 if (end < start)
4141 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4142
4143 if (daddr == 0) {
4144 regs[rd] = (uintptr_t)end;
4145 mstate->dtms_scratch_ptr += size;
4146 }
4147
4148 break;
4149 }
4150#endif
4151
4152 case DIF_SUBR_STRJOIN: {
4153 char *d = (char *)mstate->dtms_scratch_ptr;
4154 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4155 uintptr_t s1 = tupregs[0].dttk_value;
4156 uintptr_t s2 = tupregs[1].dttk_value;
4157 int i = 0;
4158
4159 if (!dtrace_strcanload(s1, size, mstate, vstate) ||
4160 !dtrace_strcanload(s2, size, mstate, vstate)) {
4161 regs[rd] = 0;
4162 break;
4163 }
4164
4165 if (!DTRACE_INSCRATCH(mstate, size)) {
4166 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4167 regs[rd] = 0;
4168 break;
4169 }
4170
4171 for (;;) {
4172 if (i >= size) {
4173 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4174 regs[rd] = 0;
4175 break;
4176 }
4177
4178 if ((d[i++] = dtrace_load8(s1++)) == '\0') {
4179 i--;
4180 break;
4181 }
4182 }
4183
4184 for (;;) {
4185 if (i >= size) {
4186 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4187 regs[rd] = 0;
4188 break;
4189 }
4190
4191 if ((d[i++] = dtrace_load8(s2++)) == '\0')
4192 break;
4193 }
4194
4195 if (i < size) {
4196 mstate->dtms_scratch_ptr += i;
4197 regs[rd] = (uintptr_t)d;
4198 }
4199
4200 break;
4201 }
4202
4203 case DIF_SUBR_LLTOSTR: {
4204 int64_t i = (int64_t)tupregs[0].dttk_value;
4205 int64_t val = i < 0 ? i * -1 : i;
4206 uint64_t size = 22; /* enough room for 2^64 in decimal */
4207 char *end = (char *)mstate->dtms_scratch_ptr + size - 1;
4208
4209 if (!DTRACE_INSCRATCH(mstate, size)) {
4210 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4211 regs[rd] = 0;
4212 break;
4213 }
4214
4215 for (*end-- = '\0'; val; val /= 10)
4216 *end-- = '0' + (val % 10);
4217
4218 if (i == 0)
4219 *end-- = '0';
4220
4221 if (i < 0)
4222 *end-- = '-';
4223
4224 regs[rd] = (uintptr_t)end + 1;
4225 mstate->dtms_scratch_ptr += size;
4226 break;
4227 }
4228
4229 case DIF_SUBR_HTONS:
4230 case DIF_SUBR_NTOHS:
4231#if BYTE_ORDER == BIG_ENDIAN
4232 regs[rd] = (uint16_t)tupregs[0].dttk_value;
4233#else
4234 regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value);
4235#endif
4236 break;
4237
4238
4239 case DIF_SUBR_HTONL:
4240 case DIF_SUBR_NTOHL:
4241#if BYTE_ORDER == BIG_ENDIAN
4242 regs[rd] = (uint32_t)tupregs[0].dttk_value;
4243#else
4244 regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value);
4245#endif
4246 break;
4247
4248
4249 case DIF_SUBR_HTONLL:
4250 case DIF_SUBR_NTOHLL:
4251#if BYTE_ORDER == BIG_ENDIAN
4252 regs[rd] = (uint64_t)tupregs[0].dttk_value;
4253#else
4254 regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value);
4255#endif
4256 break;
4257
4258
4259 case DIF_SUBR_DIRNAME:
4260 case DIF_SUBR_BASENAME: {
4261 char *dest = (char *)mstate->dtms_scratch_ptr;
4262 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4263 uintptr_t src = tupregs[0].dttk_value;
4264 int i, j, len = dtrace_strlen((char *)src, size);
4265 int lastbase = -1, firstbase = -1, lastdir = -1;
4266 int start, end;
4267
4268 if (!dtrace_canload(src, len + 1, mstate, vstate)) {
4269 regs[rd] = 0;
4270 break;
4271 }
4272
4273 if (!DTRACE_INSCRATCH(mstate, size)) {
4274 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4275 regs[rd] = 0;
4276 break;
4277 }
4278
4279 /*
4280 * The basename and dirname for a zero-length string is
4281 * defined to be "."
4282 */
4283 if (len == 0) {
4284 len = 1;
4285 src = (uintptr_t)".";
4286 }
4287
4288 /*
4289 * Start from the back of the string, moving back toward the
4290 * front until we see a character that isn't a slash. That
4291 * character is the last character in the basename.
4292 */
4293 for (i = len - 1; i >= 0; i--) {
4294 if (dtrace_load8(src + i) != '/')
4295 break;
4296 }
4297
4298 if (i >= 0)
4299 lastbase = i;
4300
4301 /*
4302 * Starting from the last character in the basename, move
4303 * towards the front until we find a slash. The character
4304 * that we processed immediately before that is the first
4305 * character in the basename.
4306 */
4307 for (; i >= 0; i--) {
4308 if (dtrace_load8(src + i) == '/')
4309 break;
4310 }
4311
4312 if (i >= 0)
4313 firstbase = i + 1;
4314
4315 /*
4316 * Now keep going until we find a non-slash character. That
4317 * character is the last character in the dirname.
4318 */
4319 for (; i >= 0; i--) {
4320 if (dtrace_load8(src + i) != '/')
4321 break;
4322 }
4323
4324 if (i >= 0)
4325 lastdir = i;
4326
4327 ASSERT(!(lastbase == -1 && firstbase != -1));
4328 ASSERT(!(firstbase == -1 && lastdir != -1));
4329
4330 if (lastbase == -1) {
4331 /*
4332 * We didn't find a non-slash character. We know that
4333 * the length is non-zero, so the whole string must be
4334 * slashes. In either the dirname or the basename
4335 * case, we return '/'.
4336 */
4337 ASSERT(firstbase == -1);
4338 firstbase = lastbase = lastdir = 0;
4339 }
4340
4341 if (firstbase == -1) {
4342 /*
4343 * The entire string consists only of a basename
4344 * component. If we're looking for dirname, we need
4345 * to change our string to be just "."; if we're
4346 * looking for a basename, we'll just set the first
4347 * character of the basename to be 0.
4348 */
4349 if (subr == DIF_SUBR_DIRNAME) {
4350 ASSERT(lastdir == -1);
4351 src = (uintptr_t)".";
4352 lastdir = 0;
4353 } else {
4354 firstbase = 0;
4355 }
4356 }
4357
4358 if (subr == DIF_SUBR_DIRNAME) {
4359 if (lastdir == -1) {
4360 /*
4361 * We know that we have a slash in the name --
4362 * or lastdir would be set to 0, above. And
4363 * because lastdir is -1, we know that this
4364 * slash must be the first character. (That
4365 * is, the full string must be of the form
4366 * "/basename".) In this case, the last
4367 * character of the directory name is 0.
4368 */
4369 lastdir = 0;
4370 }
4371
4372 start = 0;
4373 end = lastdir;
4374 } else {
4375 ASSERT(subr == DIF_SUBR_BASENAME);
4376 ASSERT(firstbase != -1 && lastbase != -1);
4377 start = firstbase;
4378 end = lastbase;
4379 }
4380
4381 for (i = start, j = 0; i <= end && j < size - 1; i++, j++)
4382 dest[j] = dtrace_load8(src + i);
4383
4384 dest[j] = '\0';
4385 regs[rd] = (uintptr_t)dest;
4386 mstate->dtms_scratch_ptr += size;
4387 break;
4388 }
4389
4390 case DIF_SUBR_CLEANPATH: {
4391 char *dest = (char *)mstate->dtms_scratch_ptr, c;
4392 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE];
4393 uintptr_t src = tupregs[0].dttk_value;
4394 int i = 0, j = 0;
4395
4396 if (!dtrace_strcanload(src, size, mstate, vstate)) {
4397 regs[rd] = 0;
4398 break;
4399 }
4400
4401 if (!DTRACE_INSCRATCH(mstate, size)) {
4402 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4403 regs[rd] = 0;
4404 break;
4405 }
4406
4407 /*
4408 * Move forward, loading each character.
4409 */
4410 do {
4411 c = dtrace_load8(src + i++);
4412next:
4413 if (j + 5 >= size) /* 5 = strlen("/..c\0") */
4414 break;
4415
4416 if (c != '/') {
4417 dest[j++] = c;
4418 continue;
4419 }
4420
4421 c = dtrace_load8(src + i++);
4422
4423 if (c == '/') {
4424 /*
4425 * We have two slashes -- we can just advance
4426 * to the next character.
4427 */
4428 goto next;
4429 }
4430
4431 if (c != '.') {
4432 /*
4433 * This is not "." and it's not ".." -- we can
4434 * just store the "/" and this character and
4435 * drive on.
4436 */
4437 dest[j++] = '/';
4438 dest[j++] = c;
4439 continue;
4440 }
4441
4442 c = dtrace_load8(src + i++);
4443
4444 if (c == '/') {
4445 /*
4446 * This is a "/./" component. We're not going
4447 * to store anything in the destination buffer;
4448 * we're just going to go to the next component.
4449 */
4450 goto next;
4451 }
4452
4453 if (c != '.') {
4454 /*
4455 * This is not ".." -- we can just store the
4456 * "/." and this character and continue
4457 * processing.
4458 */
4459 dest[j++] = '/';
4460 dest[j++] = '.';
4461 dest[j++] = c;
4462 continue;
4463 }
4464
4465 c = dtrace_load8(src + i++);
4466
4467 if (c != '/' && c != '\0') {
4468 /*
4469 * This is not ".." -- it's "..[mumble]".
4470 * We'll store the "/.." and this character
4471 * and continue processing.
4472 */
4473 dest[j++] = '/';
4474 dest[j++] = '.';
4475 dest[j++] = '.';
4476 dest[j++] = c;
4477 continue;
4478 }
4479
4480 /*
4481 * This is "/../" or "/..\0". We need to back up
4482 * our destination pointer until we find a "/".
4483 */
4484 i--;
4485 while (j != 0 && dest[--j] != '/')
4486 continue;
4487
4488 if (c == '\0')
4489 dest[++j] = '/';
4490 } while (c != '\0');
4491
4492 dest[j] = '\0';
4493 regs[rd] = (uintptr_t)dest;
4494 mstate->dtms_scratch_ptr += size;
4495 break;
4496 }
4497
4498 case DIF_SUBR_INET_NTOA:
4499 case DIF_SUBR_INET_NTOA6:
4500 case DIF_SUBR_INET_NTOP: {
4501 size_t size;
4502 int af, argi, i;
4503 char *base, *end;
4504
4505 if (subr == DIF_SUBR_INET_NTOP) {
4506 af = (int)tupregs[0].dttk_value;
4507 argi = 1;
4508 } else {
4509 af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6;
4510 argi = 0;
4511 }
4512
4513 if (af == AF_INET) {
4514 ipaddr_t ip4;
4515 uint8_t *ptr8, val;
4516
4517 /*
4518 * Safely load the IPv4 address.
4519 */
4520 ip4 = dtrace_load32(tupregs[argi].dttk_value);
4521
4522 /*
4523 * Check an IPv4 string will fit in scratch.
4524 */
4525 size = INET_ADDRSTRLEN;
4526 if (!DTRACE_INSCRATCH(mstate, size)) {
4527 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4528 regs[rd] = 0;
4529 break;
4530 }
4531 base = (char *)mstate->dtms_scratch_ptr;
4532 end = (char *)mstate->dtms_scratch_ptr + size - 1;
4533
4534 /*
4535 * Stringify as a dotted decimal quad.
4536 */
4537 *end-- = '\0';
4538 ptr8 = (uint8_t *)&ip4;
4539 for (i = 3; i >= 0; i--) {
4540 val = ptr8[i];
4541
4542 if (val == 0) {
4543 *end-- = '0';
4544 } else {
4545 for (; val; val /= 10) {
4546 *end-- = '0' + (val % 10);
4547 }
4548 }
4549
4550 if (i > 0)
4551 *end-- = '.';
4552 }
4553 ASSERT(end + 1 >= base);
4554
4555 } else if (af == AF_INET6) {
4556 struct in6_addr ip6;
4557 int firstzero, tryzero, numzero, v6end;
4558 uint16_t val;
4559 const char digits[] = "0123456789abcdef";
4560
4561 /*
4562 * Stringify using RFC 1884 convention 2 - 16 bit
4563 * hexadecimal values with a zero-run compression.
4564 * Lower case hexadecimal digits are used.
4565 * eg, fe80::214:4fff:fe0b:76c8.
4566 * The IPv4 embedded form is returned for inet_ntop,
4567 * just the IPv4 string is returned for inet_ntoa6.
4568 */
4569
4570 /*
4571 * Safely load the IPv6 address.
4572 */
4573 dtrace_bcopy(
4574 (void *)(uintptr_t)tupregs[argi].dttk_value,
4575 (void *)(uintptr_t)&ip6, sizeof (struct in6_addr));
4576
4577 /*
4578 * Check an IPv6 string will fit in scratch.
4579 */
4580 size = INET6_ADDRSTRLEN;
4581 if (!DTRACE_INSCRATCH(mstate, size)) {
4582 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
4583 regs[rd] = 0;
4584 break;
4585 }
4586 base = (char *)mstate->dtms_scratch_ptr;
4587 end = (char *)mstate->dtms_scratch_ptr + size - 1;
4588 *end-- = '\0';
4589
4590 /*
4591 * Find the longest run of 16 bit zero values
4592 * for the single allowed zero compression - "::".
4593 */
4594 firstzero = -1;
4595 tryzero = -1;
4596 numzero = 1;
4597 for (i = 0; i < sizeof (struct in6_addr); i++) {
4598#if defined(sun)
4599 if (ip6._S6_un._S6_u8[i] == 0 &&
4600#else
4601 if (ip6.__u6_addr.__u6_addr8[i] == 0 &&
4602#endif
4603 tryzero == -1 && i % 2 == 0) {
4604 tryzero = i;
4605 continue;
4606 }
4607
4608 if (tryzero != -1 &&
4609#if defined(sun)
4610 (ip6._S6_un._S6_u8[i] != 0 ||
4611#else
4612 (ip6.__u6_addr.__u6_addr8[i] != 0 ||
4613#endif
4614 i == sizeof (struct in6_addr) - 1)) {
4615
4616 if (i - tryzero <= numzero) {
4617 tryzero = -1;
4618 continue;
4619 }
4620
4621 firstzero = tryzero;
4622 numzero = i - i % 2 - tryzero;
4623 tryzero = -1;
4624
4625#if defined(sun)
4626 if (ip6._S6_un._S6_u8[i] == 0 &&
4627#else
4628 if (ip6.__u6_addr.__u6_addr8[i] == 0 &&
4629#endif
4630 i == sizeof (struct in6_addr) - 1)
4631 numzero += 2;
4632 }
4633 }
4634 ASSERT(firstzero + numzero <= sizeof (struct in6_addr));
4635
4636 /*
4637 * Check for an IPv4 embedded address.
4638 */
4639 v6end = sizeof (struct in6_addr) - 2;
4640 if (IN6_IS_ADDR_V4MAPPED(&ip6) ||
4641 IN6_IS_ADDR_V4COMPAT(&ip6)) {
4642 for (i = sizeof (struct in6_addr) - 1;
4643 i >= DTRACE_V4MAPPED_OFFSET; i--) {
4644 ASSERT(end >= base);
4645
4646#if defined(sun)
4647 val = ip6._S6_un._S6_u8[i];
4648#else
4649 val = ip6.__u6_addr.__u6_addr8[i];
4650#endif
4651
4652 if (val == 0) {
4653 *end-- = '0';
4654 } else {
4655 for (; val; val /= 10) {
4656 *end-- = '0' + val % 10;
4657 }
4658 }
4659
4660 if (i > DTRACE_V4MAPPED_OFFSET)
4661 *end-- = '.';
4662 }
4663
4664 if (subr == DIF_SUBR_INET_NTOA6)
4665 goto inetout;
4666
4667 /*
4668 * Set v6end to skip the IPv4 address that
4669 * we have already stringified.
4670 */
4671 v6end = 10;
4672 }
4673
4674 /*
4675 * Build the IPv6 string by working through the
4676 * address in reverse.
4677 */
4678 for (i = v6end; i >= 0; i -= 2) {
4679 ASSERT(end >= base);
4680
4681 if (i == firstzero + numzero - 2) {
4682 *end-- = ':';
4683 *end-- = ':';
4684 i -= numzero - 2;
4685 continue;
4686 }
4687
4688 if (i < 14 && i != firstzero - 2)
4689 *end-- = ':';
4690
4691#if defined(sun)
4692 val = (ip6._S6_un._S6_u8[i] << 8) +
4693 ip6._S6_un._S6_u8[i + 1];
4694#else
4695 val = (ip6.__u6_addr.__u6_addr8[i] << 8) +
4696 ip6.__u6_addr.__u6_addr8[i + 1];
4697#endif
4698
4699 if (val == 0) {
4700 *end-- = '0';
4701 } else {
4702 for (; val; val /= 16) {
4703 *end-- = digits[val % 16];
4704 }
4705 }
4706 }
4707 ASSERT(end + 1 >= base);
4708
4709 } else {
4710 /*
4711 * The user didn't use AH_INET or AH_INET6.
4712 */
4713 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
4714 regs[rd] = 0;
4715 break;
4716 }
4717
4718inetout: regs[rd] = (uintptr_t)end + 1;
4719 mstate->dtms_scratch_ptr += size;
4720 break;
4721 }
4722
4723 case DIF_SUBR_MEMREF: {
4724 uintptr_t size = 2 * sizeof(uintptr_t);
4725 uintptr_t *memref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t));
4726 size_t scratch_size = ((uintptr_t) memref - mstate->dtms_scratch_ptr) + size;
4727
4728 /* address and length */
4729 memref[0] = tupregs[0].dttk_value;
4730 memref[1] = tupregs[1].dttk_value;
4731
4732 regs[rd] = (uintptr_t) memref;
4733 mstate->dtms_scratch_ptr += scratch_size;
4734 break;
4735 }
4736
4737 case DIF_SUBR_TYPEREF: {
4738 uintptr_t size = 4 * sizeof(uintptr_t);
4739 uintptr_t *typeref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t));
4740 size_t scratch_size = ((uintptr_t) typeref - mstate->dtms_scratch_ptr) + size;
4741
4742 /* address, num_elements, type_str, type_len */
4743 typeref[0] = tupregs[0].dttk_value;
4744 typeref[1] = tupregs[1].dttk_value;
4745 typeref[2] = tupregs[2].dttk_value;
4746 typeref[3] = tupregs[3].dttk_value;
4747
4748 regs[rd] = (uintptr_t) typeref;
4749 mstate->dtms_scratch_ptr += scratch_size;
4750 break;
4751 }
4752 }
4753}
4754
4755/*
4756 * Emulate the execution of DTrace IR instructions specified by the given
4757 * DIF object. This function is deliberately void of assertions as all of
4758 * the necessary checks are handled by a call to dtrace_difo_validate().
4759 */
4760static uint64_t
4761dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate,
4762 dtrace_vstate_t *vstate, dtrace_state_t *state)
4763{
4764 const dif_instr_t *text = difo->dtdo_buf;
4765 const uint_t textlen = difo->dtdo_len;
4766 const char *strtab = difo->dtdo_strtab;
4767 const uint64_t *inttab = difo->dtdo_inttab;
4768
4769 uint64_t rval = 0;
4770 dtrace_statvar_t *svar;
4771 dtrace_dstate_t *dstate = &vstate->dtvs_dynvars;
4772 dtrace_difv_t *v;
4773 volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
4774 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval;
4775
4776 dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */
4777 uint64_t regs[DIF_DIR_NREGS];
4778 uint64_t *tmp;
4779
4780 uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0;
4781 int64_t cc_r;
4782 uint_t pc = 0, id, opc = 0;
4783 uint8_t ttop = 0;
4784 dif_instr_t instr;
4785 uint_t r1, r2, rd;
4786
4787 /*
4788 * We stash the current DIF object into the machine state: we need it
4789 * for subsequent access checking.
4790 */
4791 mstate->dtms_difo = difo;
4792
4793 regs[DIF_REG_R0] = 0; /* %r0 is fixed at zero */
4794
4795 while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) {
4796 opc = pc;
4797
4798 instr = text[pc++];
4799 r1 = DIF_INSTR_R1(instr);
4800 r2 = DIF_INSTR_R2(instr);
4801 rd = DIF_INSTR_RD(instr);
4802
4803 switch (DIF_INSTR_OP(instr)) {
4804 case DIF_OP_OR:
4805 regs[rd] = regs[r1] | regs[r2];
4806 break;
4807 case DIF_OP_XOR:
4808 regs[rd] = regs[r1] ^ regs[r2];
4809 break;
4810 case DIF_OP_AND:
4811 regs[rd] = regs[r1] & regs[r2];
4812 break;
4813 case DIF_OP_SLL:
4814 regs[rd] = regs[r1] << regs[r2];
4815 break;
4816 case DIF_OP_SRL:
4817 regs[rd] = regs[r1] >> regs[r2];
4818 break;
4819 case DIF_OP_SUB:
4820 regs[rd] = regs[r1] - regs[r2];
4821 break;
4822 case DIF_OP_ADD:
4823 regs[rd] = regs[r1] + regs[r2];
4824 break;
4825 case DIF_OP_MUL:
4826 regs[rd] = regs[r1] * regs[r2];
4827 break;
4828 case DIF_OP_SDIV:
4829 if (regs[r2] == 0) {
4830 regs[rd] = 0;
4831 *flags |= CPU_DTRACE_DIVZERO;
4832 } else {
4833 regs[rd] = (int64_t)regs[r1] /
4834 (int64_t)regs[r2];
4835 }
4836 break;
4837
4838 case DIF_OP_UDIV:
4839 if (regs[r2] == 0) {
4840 regs[rd] = 0;
4841 *flags |= CPU_DTRACE_DIVZERO;
4842 } else {
4843 regs[rd] = regs[r1] / regs[r2];
4844 }
4845 break;
4846
4847 case DIF_OP_SREM:
4848 if (regs[r2] == 0) {
4849 regs[rd] = 0;
4850 *flags |= CPU_DTRACE_DIVZERO;
4851 } else {
4852 regs[rd] = (int64_t)regs[r1] %
4853 (int64_t)regs[r2];
4854 }
4855 break;
4856
4857 case DIF_OP_UREM:
4858 if (regs[r2] == 0) {
4859 regs[rd] = 0;
4860 *flags |= CPU_DTRACE_DIVZERO;
4861 } else {
4862 regs[rd] = regs[r1] % regs[r2];
4863 }
4864 break;
4865
4866 case DIF_OP_NOT:
4867 regs[rd] = ~regs[r1];
4868 break;
4869 case DIF_OP_MOV:
4870 regs[rd] = regs[r1];
4871 break;
4872 case DIF_OP_CMP:
4873 cc_r = regs[r1] - regs[r2];
4874 cc_n = cc_r < 0;
4875 cc_z = cc_r == 0;
4876 cc_v = 0;
4877 cc_c = regs[r1] < regs[r2];
4878 break;
4879 case DIF_OP_TST:
4880 cc_n = cc_v = cc_c = 0;
4881 cc_z = regs[r1] == 0;
4882 break;
4883 case DIF_OP_BA:
4884 pc = DIF_INSTR_LABEL(instr);
4885 break;
4886 case DIF_OP_BE:
4887 if (cc_z)
4888 pc = DIF_INSTR_LABEL(instr);
4889 break;
4890 case DIF_OP_BNE:
4891 if (cc_z == 0)
4892 pc = DIF_INSTR_LABEL(instr);
4893 break;
4894 case DIF_OP_BG:
4895 if ((cc_z | (cc_n ^ cc_v)) == 0)
4896 pc = DIF_INSTR_LABEL(instr);
4897 break;
4898 case DIF_OP_BGU:
4899 if ((cc_c | cc_z) == 0)
4900 pc = DIF_INSTR_LABEL(instr);
4901 break;
4902 case DIF_OP_BGE:
4903 if ((cc_n ^ cc_v) == 0)
4904 pc = DIF_INSTR_LABEL(instr);
4905 break;
4906 case DIF_OP_BGEU:
4907 if (cc_c == 0)
4908 pc = DIF_INSTR_LABEL(instr);
4909 break;
4910 case DIF_OP_BL:
4911 if (cc_n ^ cc_v)
4912 pc = DIF_INSTR_LABEL(instr);
4913 break;
4914 case DIF_OP_BLU:
4915 if (cc_c)
4916 pc = DIF_INSTR_LABEL(instr);
4917 break;
4918 case DIF_OP_BLE:
4919 if (cc_z | (cc_n ^ cc_v))
4920 pc = DIF_INSTR_LABEL(instr);
4921 break;
4922 case DIF_OP_BLEU:
4923 if (cc_c | cc_z)
4924 pc = DIF_INSTR_LABEL(instr);
4925 break;
4926 case DIF_OP_RLDSB:
4927 if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) {
4928 *flags |= CPU_DTRACE_KPRIV;
4929 *illval = regs[r1];
4930 break;
4931 }
4932 /*FALLTHROUGH*/
4933 case DIF_OP_LDSB:
4934 regs[rd] = (int8_t)dtrace_load8(regs[r1]);
4935 break;
4936 case DIF_OP_RLDSH:
4937 if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) {
4938 *flags |= CPU_DTRACE_KPRIV;
4939 *illval = regs[r1];
4940 break;
4941 }
4942 /*FALLTHROUGH*/
4943 case DIF_OP_LDSH:
4944 regs[rd] = (int16_t)dtrace_load16(regs[r1]);
4945 break;
4946 case DIF_OP_RLDSW:
4947 if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) {
4948 *flags |= CPU_DTRACE_KPRIV;
4949 *illval = regs[r1];
4950 break;
4951 }
4952 /*FALLTHROUGH*/
4953 case DIF_OP_LDSW:
4954 regs[rd] = (int32_t)dtrace_load32(regs[r1]);
4955 break;
4956 case DIF_OP_RLDUB:
4957 if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) {
4958 *flags |= CPU_DTRACE_KPRIV;
4959 *illval = regs[r1];
4960 break;
4961 }
4962 /*FALLTHROUGH*/
4963 case DIF_OP_LDUB:
4964 regs[rd] = dtrace_load8(regs[r1]);
4965 break;
4966 case DIF_OP_RLDUH:
4967 if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) {
4968 *flags |= CPU_DTRACE_KPRIV;
4969 *illval = regs[r1];
4970 break;
4971 }
4972 /*FALLTHROUGH*/
4973 case DIF_OP_LDUH:
4974 regs[rd] = dtrace_load16(regs[r1]);
4975 break;
4976 case DIF_OP_RLDUW:
4977 if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) {
4978 *flags |= CPU_DTRACE_KPRIV;
4979 *illval = regs[r1];
4980 break;
4981 }
4982 /*FALLTHROUGH*/
4983 case DIF_OP_LDUW:
4984 regs[rd] = dtrace_load32(regs[r1]);
4985 break;
4986 case DIF_OP_RLDX:
4987 if (!dtrace_canstore(regs[r1], 8, mstate, vstate)) {
4988 *flags |= CPU_DTRACE_KPRIV;
4989 *illval = regs[r1];
4990 break;
4991 }
4992 /*FALLTHROUGH*/
4993 case DIF_OP_LDX:
4994 regs[rd] = dtrace_load64(regs[r1]);
4995 break;
4996 case DIF_OP_ULDSB:
4997 regs[rd] = (int8_t)
4998 dtrace_fuword8((void *)(uintptr_t)regs[r1]);
4999 break;
5000 case DIF_OP_ULDSH:
5001 regs[rd] = (int16_t)
5002 dtrace_fuword16((void *)(uintptr_t)regs[r1]);
5003 break;
5004 case DIF_OP_ULDSW:
5005 regs[rd] = (int32_t)
5006 dtrace_fuword32((void *)(uintptr_t)regs[r1]);
5007 break;
5008 case DIF_OP_ULDUB:
5009 regs[rd] =
5010 dtrace_fuword8((void *)(uintptr_t)regs[r1]);
5011 break;
5012 case DIF_OP_ULDUH:
5013 regs[rd] =
5014 dtrace_fuword16((void *)(uintptr_t)regs[r1]);
5015 break;
5016 case DIF_OP_ULDUW:
5017 regs[rd] =
5018 dtrace_fuword32((void *)(uintptr_t)regs[r1]);
5019 break;
5020 case DIF_OP_ULDX:
5021 regs[rd] =
5022 dtrace_fuword64((void *)(uintptr_t)regs[r1]);
5023 break;
5024 case DIF_OP_RET:
5025 rval = regs[rd];
5026 pc = textlen;
5027 break;
5028 case DIF_OP_NOP:
5029 break;
5030 case DIF_OP_SETX:
5031 regs[rd] = inttab[DIF_INSTR_INTEGER(instr)];
5032 break;
5033 case DIF_OP_SETS:
5034 regs[rd] = (uint64_t)(uintptr_t)
5035 (strtab + DIF_INSTR_STRING(instr));
5036 break;
5037 case DIF_OP_SCMP: {
5038 size_t sz = state->dts_options[DTRACEOPT_STRSIZE];
5039 uintptr_t s1 = regs[r1];
5040 uintptr_t s2 = regs[r2];
5041
5042 if (s1 != 0 &&
5043 !dtrace_strcanload(s1, sz, mstate, vstate))
5044 break;
5045 if (s2 != 0 &&
5046 !dtrace_strcanload(s2, sz, mstate, vstate))
5047 break;
5048
5049 cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz);
5050
5051 cc_n = cc_r < 0;
5052 cc_z = cc_r == 0;
5053 cc_v = cc_c = 0;
5054 break;
5055 }
5056 case DIF_OP_LDGA:
5057 regs[rd] = dtrace_dif_variable(mstate, state,
5058 r1, regs[r2]);
5059 break;
5060 case DIF_OP_LDGS:
5061 id = DIF_INSTR_VAR(instr);
5062
5063 if (id >= DIF_VAR_OTHER_UBASE) {
5064 uintptr_t a;
5065
5066 id -= DIF_VAR_OTHER_UBASE;
5067 svar = vstate->dtvs_globals[id];
5068 ASSERT(svar != NULL);
5069 v = &svar->dtsv_var;
5070
5071 if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) {
5072 regs[rd] = svar->dtsv_data;
5073 break;
5074 }
5075
5076 a = (uintptr_t)svar->dtsv_data;
5077
5078 if (*(uint8_t *)a == UINT8_MAX) {
5079 /*
5080 * If the 0th byte is set to UINT8_MAX
5081 * then this is to be treated as a
5082 * reference to a NULL variable.
5083 */
5084 regs[rd] = 0;
5085 } else {
5086 regs[rd] = a + sizeof (uint64_t);
5087 }
5088
5089 break;
5090 }
5091
5092 regs[rd] = dtrace_dif_variable(mstate, state, id, 0);
5093 break;
5094
5095 case DIF_OP_STGS:
5096 id = DIF_INSTR_VAR(instr);
5097
5098 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5099 id -= DIF_VAR_OTHER_UBASE;
5100
5101 svar = vstate->dtvs_globals[id];
5102 ASSERT(svar != NULL);
5103 v = &svar->dtsv_var;
5104
5105 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5106 uintptr_t a = (uintptr_t)svar->dtsv_data;
5107
5108 ASSERT(a != 0);
5109 ASSERT(svar->dtsv_size != 0);
5110
5111 if (regs[rd] == 0) {
5112 *(uint8_t *)a = UINT8_MAX;
5113 break;
5114 } else {
5115 *(uint8_t *)a = 0;
5116 a += sizeof (uint64_t);
5117 }
5118 if (!dtrace_vcanload(
5119 (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5120 mstate, vstate))
5121 break;
5122
5123 dtrace_vcopy((void *)(uintptr_t)regs[rd],
5124 (void *)a, &v->dtdv_type);
5125 break;
5126 }
5127
5128 svar->dtsv_data = regs[rd];
5129 break;
5130
5131 case DIF_OP_LDTA:
5132 /*
5133 * There are no DTrace built-in thread-local arrays at
5134 * present. This opcode is saved for future work.
5135 */
5136 *flags |= CPU_DTRACE_ILLOP;
5137 regs[rd] = 0;
5138 break;
5139
5140 case DIF_OP_LDLS:
5141 id = DIF_INSTR_VAR(instr);
5142
5143 if (id < DIF_VAR_OTHER_UBASE) {
5144 /*
5145 * For now, this has no meaning.
5146 */
5147 regs[rd] = 0;
5148 break;
5149 }
5150
5151 id -= DIF_VAR_OTHER_UBASE;
5152
5153 ASSERT(id < vstate->dtvs_nlocals);
5154 ASSERT(vstate->dtvs_locals != NULL);
5155
5156 svar = vstate->dtvs_locals[id];
5157 ASSERT(svar != NULL);
5158 v = &svar->dtsv_var;
5159
5160 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5161 uintptr_t a = (uintptr_t)svar->dtsv_data;
5162 size_t sz = v->dtdv_type.dtdt_size;
5163
5164 sz += sizeof (uint64_t);
5165 ASSERT(svar->dtsv_size == NCPU * sz);
5166 a += curcpu * sz;
5167
5168 if (*(uint8_t *)a == UINT8_MAX) {
5169 /*
5170 * If the 0th byte is set to UINT8_MAX
5171 * then this is to be treated as a
5172 * reference to a NULL variable.
5173 */
5174 regs[rd] = 0;
5175 } else {
5176 regs[rd] = a + sizeof (uint64_t);
5177 }
5178
5179 break;
5180 }
5181
5182 ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t));
5183 tmp = (uint64_t *)(uintptr_t)svar->dtsv_data;
5184 regs[rd] = tmp[curcpu];
5185 break;
5186
5187 case DIF_OP_STLS:
5188 id = DIF_INSTR_VAR(instr);
5189
5190 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5191 id -= DIF_VAR_OTHER_UBASE;
5192 ASSERT(id < vstate->dtvs_nlocals);
5193
5194 ASSERT(vstate->dtvs_locals != NULL);
5195 svar = vstate->dtvs_locals[id];
5196 ASSERT(svar != NULL);
5197 v = &svar->dtsv_var;
5198
5199 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5200 uintptr_t a = (uintptr_t)svar->dtsv_data;
5201 size_t sz = v->dtdv_type.dtdt_size;
5202
5203 sz += sizeof (uint64_t);
5204 ASSERT(svar->dtsv_size == NCPU * sz);
5205 a += curcpu * sz;
5206
5207 if (regs[rd] == 0) {
5208 *(uint8_t *)a = UINT8_MAX;
5209 break;
5210 } else {
5211 *(uint8_t *)a = 0;
5212 a += sizeof (uint64_t);
5213 }
5214
5215 if (!dtrace_vcanload(
5216 (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5217 mstate, vstate))
5218 break;
5219
5220 dtrace_vcopy((void *)(uintptr_t)regs[rd],
5221 (void *)a, &v->dtdv_type);
5222 break;
5223 }
5224
5225 ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t));
5226 tmp = (uint64_t *)(uintptr_t)svar->dtsv_data;
5227 tmp[curcpu] = regs[rd];
5228 break;
5229
5230 case DIF_OP_LDTS: {
5231 dtrace_dynvar_t *dvar;
5232 dtrace_key_t *key;
5233
5234 id = DIF_INSTR_VAR(instr);
5235 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5236 id -= DIF_VAR_OTHER_UBASE;
5237 v = &vstate->dtvs_tlocals[id];
5238
5239 key = &tupregs[DIF_DTR_NREGS];
5240 key[0].dttk_value = (uint64_t)id;
5241 key[0].dttk_size = 0;
5242 DTRACE_TLS_THRKEY(key[1].dttk_value);
5243 key[1].dttk_size = 0;
5244
5245 dvar = dtrace_dynvar(dstate, 2, key,
5246 sizeof (uint64_t), DTRACE_DYNVAR_NOALLOC,
5247 mstate, vstate);
5248
5249 if (dvar == NULL) {
5250 regs[rd] = 0;
5251 break;
5252 }
5253
5254 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5255 regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data;
5256 } else {
5257 regs[rd] = *((uint64_t *)dvar->dtdv_data);
5258 }
5259
5260 break;
5261 }
5262
5263 case DIF_OP_STTS: {
5264 dtrace_dynvar_t *dvar;
5265 dtrace_key_t *key;
5266
5267 id = DIF_INSTR_VAR(instr);
5268 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5269 id -= DIF_VAR_OTHER_UBASE;
5270
5271 key = &tupregs[DIF_DTR_NREGS];
5272 key[0].dttk_value = (uint64_t)id;
5273 key[0].dttk_size = 0;
5274 DTRACE_TLS_THRKEY(key[1].dttk_value);
5275 key[1].dttk_size = 0;
5276 v = &vstate->dtvs_tlocals[id];
5277
5278 dvar = dtrace_dynvar(dstate, 2, key,
5279 v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5280 v->dtdv_type.dtdt_size : sizeof (uint64_t),
5281 regs[rd] ? DTRACE_DYNVAR_ALLOC :
5282 DTRACE_DYNVAR_DEALLOC, mstate, vstate);
5283
5284 /*
5285 * Given that we're storing to thread-local data,
5286 * we need to flush our predicate cache.
5287 */
5288 curthread->t_predcache = 0;
5289
5290 if (dvar == NULL)
5291 break;
5292
5293 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5294 if (!dtrace_vcanload(
5295 (void *)(uintptr_t)regs[rd],
5296 &v->dtdv_type, mstate, vstate))
5297 break;
5298
5299 dtrace_vcopy((void *)(uintptr_t)regs[rd],
5300 dvar->dtdv_data, &v->dtdv_type);
5301 } else {
5302 *((uint64_t *)dvar->dtdv_data) = regs[rd];
5303 }
5304
5305 break;
5306 }
5307
5308 case DIF_OP_SRA:
5309 regs[rd] = (int64_t)regs[r1] >> regs[r2];
5310 break;
5311
5312 case DIF_OP_CALL:
5313 dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd,
5314 regs, tupregs, ttop, mstate, state);
5315 break;
5316
5317 case DIF_OP_PUSHTR:
5318 if (ttop == DIF_DTR_NREGS) {
5319 *flags |= CPU_DTRACE_TUPOFLOW;
5320 break;
5321 }
5322
5323 if (r1 == DIF_TYPE_STRING) {
5324 /*
5325 * If this is a string type and the size is 0,
5326 * we'll use the system-wide default string
5327 * size. Note that we are _not_ looking at
5328 * the value of the DTRACEOPT_STRSIZE option;
5329 * had this been set, we would expect to have
5330 * a non-zero size value in the "pushtr".
5331 */
5332 tupregs[ttop].dttk_size =
5333 dtrace_strlen((char *)(uintptr_t)regs[rd],
5334 regs[r2] ? regs[r2] :
5335 dtrace_strsize_default) + 1;
5336 } else {
5337 tupregs[ttop].dttk_size = regs[r2];
5338 }
5339
5340 tupregs[ttop++].dttk_value = regs[rd];
5341 break;
5342
5343 case DIF_OP_PUSHTV:
5344 if (ttop == DIF_DTR_NREGS) {
5345 *flags |= CPU_DTRACE_TUPOFLOW;
5346 break;
5347 }
5348
5349 tupregs[ttop].dttk_value = regs[rd];
5350 tupregs[ttop++].dttk_size = 0;
5351 break;
5352
5353 case DIF_OP_POPTS:
5354 if (ttop != 0)
5355 ttop--;
5356 break;
5357
5358 case DIF_OP_FLUSHTS:
5359 ttop = 0;
5360 break;
5361
5362 case DIF_OP_LDGAA:
5363 case DIF_OP_LDTAA: {
5364 dtrace_dynvar_t *dvar;
5365 dtrace_key_t *key = tupregs;
5366 uint_t nkeys = ttop;
5367
5368 id = DIF_INSTR_VAR(instr);
5369 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5370 id -= DIF_VAR_OTHER_UBASE;
5371
5372 key[nkeys].dttk_value = (uint64_t)id;
5373 key[nkeys++].dttk_size = 0;
5374
5375 if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) {
5376 DTRACE_TLS_THRKEY(key[nkeys].dttk_value);
5377 key[nkeys++].dttk_size = 0;
5378 v = &vstate->dtvs_tlocals[id];
5379 } else {
5380 v = &vstate->dtvs_globals[id]->dtsv_var;
5381 }
5382
5383 dvar = dtrace_dynvar(dstate, nkeys, key,
5384 v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5385 v->dtdv_type.dtdt_size : sizeof (uint64_t),
5386 DTRACE_DYNVAR_NOALLOC, mstate, vstate);
5387
5388 if (dvar == NULL) {
5389 regs[rd] = 0;
5390 break;
5391 }
5392
5393 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5394 regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data;
5395 } else {
5396 regs[rd] = *((uint64_t *)dvar->dtdv_data);
5397 }
5398
5399 break;
5400 }
5401
5402 case DIF_OP_STGAA:
5403 case DIF_OP_STTAA: {
5404 dtrace_dynvar_t *dvar;
5405 dtrace_key_t *key = tupregs;
5406 uint_t nkeys = ttop;
5407
5408 id = DIF_INSTR_VAR(instr);
5409 ASSERT(id >= DIF_VAR_OTHER_UBASE);
5410 id -= DIF_VAR_OTHER_UBASE;
5411
5412 key[nkeys].dttk_value = (uint64_t)id;
5413 key[nkeys++].dttk_size = 0;
5414
5415 if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) {
5416 DTRACE_TLS_THRKEY(key[nkeys].dttk_value);
5417 key[nkeys++].dttk_size = 0;
5418 v = &vstate->dtvs_tlocals[id];
5419 } else {
5420 v = &vstate->dtvs_globals[id]->dtsv_var;
5421 }
5422
5423 dvar = dtrace_dynvar(dstate, nkeys, key,
5424 v->dtdv_type.dtdt_size > sizeof (uint64_t) ?
5425 v->dtdv_type.dtdt_size : sizeof (uint64_t),
5426 regs[rd] ? DTRACE_DYNVAR_ALLOC :
5427 DTRACE_DYNVAR_DEALLOC, mstate, vstate);
5428
5429 if (dvar == NULL)
5430 break;
5431
5432 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) {
5433 if (!dtrace_vcanload(
5434 (void *)(uintptr_t)regs[rd], &v->dtdv_type,
5435 mstate, vstate))
5436 break;
5437
5438 dtrace_vcopy((void *)(uintptr_t)regs[rd],
5439 dvar->dtdv_data, &v->dtdv_type);
5440 } else {
5441 *((uint64_t *)dvar->dtdv_data) = regs[rd];
5442 }
5443
5444 break;
5445 }
5446
5447 case DIF_OP_ALLOCS: {
5448 uintptr_t ptr = P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
5449 size_t size = ptr - mstate->dtms_scratch_ptr + regs[r1];
5450
5451 /*
5452 * Rounding up the user allocation size could have
5453 * overflowed large, bogus allocations (like -1ULL) to
5454 * 0.
5455 */
5456 if (size < regs[r1] ||
5457 !DTRACE_INSCRATCH(mstate, size)) {
5458 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5459 regs[rd] = 0;
5460 break;
5461 }
5462
5463 dtrace_bzero((void *) mstate->dtms_scratch_ptr, size);
5464 mstate->dtms_scratch_ptr += size;
5465 regs[rd] = ptr;
5466 break;
5467 }
5468
5469 case DIF_OP_COPYS:
5470 if (!dtrace_canstore(regs[rd], regs[r2],
5471 mstate, vstate)) {
5472 *flags |= CPU_DTRACE_BADADDR;
5473 *illval = regs[rd];
5474 break;
5475 }
5476
5477 if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate))
5478 break;
5479
5480 dtrace_bcopy((void *)(uintptr_t)regs[r1],
5481 (void *)(uintptr_t)regs[rd], (size_t)regs[r2]);
5482 break;
5483
5484 case DIF_OP_STB:
5485 if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) {
5486 *flags |= CPU_DTRACE_BADADDR;
5487 *illval = regs[rd];
5488 break;
5489 }
5490 *((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1];
5491 break;
5492
5493 case DIF_OP_STH:
5494 if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) {
5495 *flags |= CPU_DTRACE_BADADDR;
5496 *illval = regs[rd];
5497 break;
5498 }
5499 if (regs[rd] & 1) {
5500 *flags |= CPU_DTRACE_BADALIGN;
5501 *illval = regs[rd];
5502 break;
5503 }
5504 *((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1];
5505 break;
5506
5507 case DIF_OP_STW:
5508 if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) {
5509 *flags |= CPU_DTRACE_BADADDR;
5510 *illval = regs[rd];
5511 break;
5512 }
5513 if (regs[rd] & 3) {
5514 *flags |= CPU_DTRACE_BADALIGN;
5515 *illval = regs[rd];
5516 break;
5517 }
5518 *((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1];
5519 break;
5520
5521 case DIF_OP_STX:
5522 if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) {
5523 *flags |= CPU_DTRACE_BADADDR;
5524 *illval = regs[rd];
5525 break;
5526 }
5527 if (regs[rd] & 7) {
5528 *flags |= CPU_DTRACE_BADALIGN;
5529 *illval = regs[rd];
5530 break;
5531 }
5532 *((uint64_t *)(uintptr_t)regs[rd]) = regs[r1];
5533 break;
5534 }
5535 }
5536
5537 if (!(*flags & CPU_DTRACE_FAULT))
5538 return (rval);
5539
5540 mstate->dtms_fltoffs = opc * sizeof (dif_instr_t);
5541 mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS;
5542
5543 return (0);
5544}
5545
5546static void
5547dtrace_action_breakpoint(dtrace_ecb_t *ecb)
5548{
5549 dtrace_probe_t *probe = ecb->dte_probe;
5550 dtrace_provider_t *prov = probe->dtpr_provider;
5551 char c[DTRACE_FULLNAMELEN + 80], *str;
5552 char *msg = "dtrace: breakpoint action at probe ";
5553 char *ecbmsg = " (ecb ";
5554 uintptr_t mask = (0xf << (sizeof (uintptr_t) * NBBY / 4));
5555 uintptr_t val = (uintptr_t)ecb;
5556 int shift = (sizeof (uintptr_t) * NBBY) - 4, i = 0;
5557
5558 if (dtrace_destructive_disallow)
5559 return;
5560
5561 /*
5562 * It's impossible to be taking action on the NULL probe.
5563 */
5564 ASSERT(probe != NULL);
5565
5566 /*
5567 * This is a poor man's (destitute man's?) sprintf(): we want to
5568 * print the provider name, module name, function name and name of
5569 * the probe, along with the hex address of the ECB with the breakpoint
5570 * action -- all of which we must place in the character buffer by
5571 * hand.
5572 */
5573 while (*msg != '\0')
5574 c[i++] = *msg++;
5575
5576 for (str = prov->dtpv_name; *str != '\0'; str++)
5577 c[i++] = *str;
5578 c[i++] = ':';
5579
5580 for (str = probe->dtpr_mod; *str != '\0'; str++)
5581 c[i++] = *str;
5582 c[i++] = ':';
5583
5584 for (str = probe->dtpr_func; *str != '\0'; str++)
5585 c[i++] = *str;
5586 c[i++] = ':';
5587
5588 for (str = probe->dtpr_name; *str != '\0'; str++)
5589 c[i++] = *str;
5590
5591 while (*ecbmsg != '\0')
5592 c[i++] = *ecbmsg++;
5593
5594 while (shift >= 0) {
5595 mask = (uintptr_t)0xf << shift;
5596
5597 if (val >= ((uintptr_t)1 << shift))
5598 c[i++] = "0123456789abcdef"[(val & mask) >> shift];
5599 shift -= 4;
5600 }
5601
5602 c[i++] = ')';
5603 c[i] = '\0';
5604
5605#if defined(sun)
5606 debug_enter(c);
5607#else
5608 kdb_enter(KDB_WHY_DTRACE, "breakpoint action");
5609#endif
5610}
5611
5612static void
5613dtrace_action_panic(dtrace_ecb_t *ecb)
5614{
5615 dtrace_probe_t *probe = ecb->dte_probe;
5616
5617 /*
5618 * It's impossible to be taking action on the NULL probe.
5619 */
5620 ASSERT(probe != NULL);
5621
5622 if (dtrace_destructive_disallow)
5623 return;
5624
5625 if (dtrace_panicked != NULL)
5626 return;
5627
5628 if (dtrace_casptr(&dtrace_panicked, NULL, curthread) != NULL)
5629 return;
5630
5631 /*
5632 * We won the right to panic. (We want to be sure that only one
5633 * thread calls panic() from dtrace_probe(), and that panic() is
5634 * called exactly once.)
5635 */
5636 dtrace_panic("dtrace: panic action at probe %s:%s:%s:%s (ecb %p)",
5637 probe->dtpr_provider->dtpv_name, probe->dtpr_mod,
5638 probe->dtpr_func, probe->dtpr_name, (void *)ecb);
5639}
5640
5641static void
5642dtrace_action_raise(uint64_t sig)
5643{
5644 if (dtrace_destructive_disallow)
5645 return;
5646
5647 if (sig >= NSIG) {
5648 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP);
5649 return;
5650 }
5651
5652#if defined(sun)
5653 /*
5654 * raise() has a queue depth of 1 -- we ignore all subsequent
5655 * invocations of the raise() action.
5656 */
5657 if (curthread->t_dtrace_sig == 0)
5658 curthread->t_dtrace_sig = (uint8_t)sig;
5659
5660 curthread->t_sig_check = 1;
5661 aston(curthread);
5662#else
5663 struct proc *p = curproc;
5664 PROC_LOCK(p);
5665 psignal(p, sig);
5666 PROC_UNLOCK(p);
5667#endif
5668}
5669
5670static void
5671dtrace_action_stop(void)
5672{
5673 if (dtrace_destructive_disallow)
5674 return;
5675
5676#if defined(sun)
5677 if (!curthread->t_dtrace_stop) {
5678 curthread->t_dtrace_stop = 1;
5679 curthread->t_sig_check = 1;
5680 aston(curthread);
5681 }
5682#else
5683 struct proc *p = curproc;
5684 PROC_LOCK(p);
5685 psignal(p, SIGSTOP);
5686 PROC_UNLOCK(p);
5687#endif
5688}
5689
5690static void
5691dtrace_action_chill(dtrace_mstate_t *mstate, hrtime_t val)
5692{
5693 hrtime_t now;
5694 volatile uint16_t *flags;
5695#if defined(sun)
5696 cpu_t *cpu = CPU;
5697#else
5698 cpu_t *cpu = &solaris_cpu[curcpu];
5699#endif
5700
5701 if (dtrace_destructive_disallow)
5702 return;
5703
5704 flags = (volatile uint16_t *)&cpu_core[cpu->cpu_id].cpuc_dtrace_flags;
5705
5706 now = dtrace_gethrtime();
5707
5708 if (now - cpu->cpu_dtrace_chillmark > dtrace_chill_interval) {
5709 /*
5710 * We need to advance the mark to the current time.
5711 */
5712 cpu->cpu_dtrace_chillmark = now;
5713 cpu->cpu_dtrace_chilled = 0;
5714 }
5715
5716 /*
5717 * Now check to see if the requested chill time would take us over
5718 * the maximum amount of time allowed in the chill interval. (Or
5719 * worse, if the calculation itself induces overflow.)
5720 */
5721 if (cpu->cpu_dtrace_chilled + val > dtrace_chill_max ||
5722 cpu->cpu_dtrace_chilled + val < cpu->cpu_dtrace_chilled) {
5723 *flags |= CPU_DTRACE_ILLOP;
5724 return;
5725 }
5726
5727 while (dtrace_gethrtime() - now < val)
5728 continue;
5729
5730 /*
5731 * Normally, we assure that the value of the variable "timestamp" does
5732 * not change within an ECB. The presence of chill() represents an
5733 * exception to this rule, however.
5734 */
5735 mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP;
5736 cpu->cpu_dtrace_chilled += val;
5737}
5738
5739#if defined(sun)
5740static void
5741dtrace_action_ustack(dtrace_mstate_t *mstate, dtrace_state_t *state,
5742 uint64_t *buf, uint64_t arg)
5743{
5744 int nframes = DTRACE_USTACK_NFRAMES(arg);
5745 int strsize = DTRACE_USTACK_STRSIZE(arg);
5746 uint64_t *pcs = &buf[1], *fps;
5747 char *str = (char *)&pcs[nframes];
5748 int size, offs = 0, i, j;
5749 uintptr_t old = mstate->dtms_scratch_ptr, saved;
5750 uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
5751 char *sym;
5752
5753 /*
5754 * Should be taking a faster path if string space has not been
5755 * allocated.
5756 */
5757 ASSERT(strsize != 0);
5758
5759 /*
5760 * We will first allocate some temporary space for the frame pointers.
5761 */
5762 fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8);
5763 size = (uintptr_t)fps - mstate->dtms_scratch_ptr +
5764 (nframes * sizeof (uint64_t));
5765
5766 if (!DTRACE_INSCRATCH(mstate, size)) {
5767 /*
5768 * Not enough room for our frame pointers -- need to indicate
5769 * that we ran out of scratch space.
5770 */
5771 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH);
5772 return;
5773 }
5774
5775 mstate->dtms_scratch_ptr += size;
5776 saved = mstate->dtms_scratch_ptr;
5777
5778 /*
5779 * Now get a stack with both program counters and frame pointers.
5780 */
5781 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5782 dtrace_getufpstack(buf, fps, nframes + 1);
5783 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5784
5785 /*
5786 * If that faulted, we're cooked.
5787 */
5788 if (*flags & CPU_DTRACE_FAULT)
5789 goto out;
5790
5791 /*
5792 * Now we want to walk up the stack, calling the USTACK helper. For
5793 * each iteration, we restore the scratch pointer.
5794 */
5795 for (i = 0; i < nframes; i++) {
5796 mstate->dtms_scratch_ptr = saved;
5797
5798 if (offs >= strsize)
5799 break;
5800
5801 sym = (char *)(uintptr_t)dtrace_helper(
5802 DTRACE_HELPER_ACTION_USTACK,
5803 mstate, state, pcs[i], fps[i]);
5804
5805 /*
5806 * If we faulted while running the helper, we're going to
5807 * clear the fault and null out the corresponding string.
5808 */
5809 if (*flags & CPU_DTRACE_FAULT) {
5810 *flags &= ~CPU_DTRACE_FAULT;
5811 str[offs++] = '\0';
5812 continue;
5813 }
5814
5815 if (sym == NULL) {
5816 str[offs++] = '\0';
5817 continue;
5818 }
5819
5820 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
5821
5822 /*
5823 * Now copy in the string that the helper returned to us.
5824 */
5825 for (j = 0; offs + j < strsize; j++) {
5826 if ((str[offs + j] = sym[j]) == '\0')
5827 break;
5828 }
5829
5830 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
5831
5832 offs += j + 1;
5833 }
5834
5835 if (offs >= strsize) {
5836 /*
5837 * If we didn't have room for all of the strings, we don't
5838 * abort processing -- this needn't be a fatal error -- but we
5839 * still want to increment a counter (dts_stkstroverflows) to
5840 * allow this condition to be warned about. (If this is from
5841 * a jstack() action, it is easily tuned via jstackstrsize.)
5842 */
5843 dtrace_error(&state->dts_stkstroverflows);
5844 }
5845
5846 while (offs < strsize)
5847 str[offs++] = '\0';
5848
5849out:
5850 mstate->dtms_scratch_ptr = old;
5851}
5852#endif
5853
5854/*
5855 * If you're looking for the epicenter of DTrace, you just found it. This
5856 * is the function called by the provider to fire a probe -- from which all
5857 * subsequent probe-context DTrace activity emanates.
5858 */
5859void
5860dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1,
5861 uintptr_t arg2, uintptr_t arg3, uintptr_t arg4)
5862{
5863 processorid_t cpuid;
5864 dtrace_icookie_t cookie;
5865 dtrace_probe_t *probe;
5866 dtrace_mstate_t mstate;
5867 dtrace_ecb_t *ecb;
5868 dtrace_action_t *act;
5869 intptr_t offs;
5870 size_t size;
5871 int vtime, onintr;
5872 volatile uint16_t *flags;
5873 hrtime_t now;
5874
5875#if defined(sun)
5876 /*
5877 * Kick out immediately if this CPU is still being born (in which case
5878 * curthread will be set to -1) or the current thread can't allow
5879 * probes in its current context.
5880 */
5881 if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE))
5882 return;
5883#endif
5884
5885 cookie = dtrace_interrupt_disable();
5886 probe = dtrace_probes[id - 1];
5887 cpuid = curcpu;
5888 onintr = CPU_ON_INTR(CPU);
5889
5890 if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE &&
5891 probe->dtpr_predcache == curthread->t_predcache) {
5892 /*
5893 * We have hit in the predicate cache; we know that
5894 * this predicate would evaluate to be false.
5895 */
5896 dtrace_interrupt_enable(cookie);
5897 return;
5898 }
5899
5900#if defined(sun)
5901 if (panic_quiesce) {
5902#else
5903 if (panicstr != NULL) {
5904#endif
5905 /*
5906 * We don't trace anything if we're panicking.
5907 */
5908 dtrace_interrupt_enable(cookie);
5909 return;
5910 }
5911
5912 now = dtrace_gethrtime();
5913 vtime = dtrace_vtime_references != 0;
5914
5915 if (vtime && curthread->t_dtrace_start)
5916 curthread->t_dtrace_vtime += now - curthread->t_dtrace_start;
5917
5918 mstate.dtms_difo = NULL;
5919 mstate.dtms_probe = probe;
5920 mstate.dtms_strtok = 0;
5921 mstate.dtms_arg[0] = arg0;
5922 mstate.dtms_arg[1] = arg1;
5923 mstate.dtms_arg[2] = arg2;
5924 mstate.dtms_arg[3] = arg3;
5925 mstate.dtms_arg[4] = arg4;
5926
5927 flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags;
5928
5929 for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) {
5930 dtrace_predicate_t *pred = ecb->dte_predicate;
5931 dtrace_state_t *state = ecb->dte_state;
5932 dtrace_buffer_t *buf = &state->dts_buffer[cpuid];
5933 dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid];
5934 dtrace_vstate_t *vstate = &state->dts_vstate;
5935 dtrace_provider_t *prov = probe->dtpr_provider;
5936 int committed = 0;
5937 caddr_t tomax;
5938
5939 /*
5940 * A little subtlety with the following (seemingly innocuous)
5941 * declaration of the automatic 'val': by looking at the
5942 * code, you might think that it could be declared in the
5943 * action processing loop, below. (That is, it's only used in
5944 * the action processing loop.) However, it must be declared
5945 * out of that scope because in the case of DIF expression
5946 * arguments to aggregating actions, one iteration of the
5947 * action loop will use the last iteration's value.
5948 */
5949 uint64_t val = 0;
5950
5951 mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE;
5952 *flags &= ~CPU_DTRACE_ERROR;
5953
5954 if (prov == dtrace_provider) {
5955 /*
5956 * If dtrace itself is the provider of this probe,
5957 * we're only going to continue processing the ECB if
5958 * arg0 (the dtrace_state_t) is equal to the ECB's
5959 * creating state. (This prevents disjoint consumers
5960 * from seeing one another's metaprobes.)
5961 */
5962 if (arg0 != (uint64_t)(uintptr_t)state)
5963 continue;
5964 }
5965
5966 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) {
5967 /*
5968 * We're not currently active. If our provider isn't
5969 * the dtrace pseudo provider, we're not interested.
5970 */
5971 if (prov != dtrace_provider)
5972 continue;
5973
5974 /*
5975 * Now we must further check if we are in the BEGIN
5976 * probe. If we are, we will only continue processing
5977 * if we're still in WARMUP -- if one BEGIN enabling
5978 * has invoked the exit() action, we don't want to
5979 * evaluate subsequent BEGIN enablings.
5980 */
5981 if (probe->dtpr_id == dtrace_probeid_begin &&
5982 state->dts_activity != DTRACE_ACTIVITY_WARMUP) {
5983 ASSERT(state->dts_activity ==
5984 DTRACE_ACTIVITY_DRAINING);
5985 continue;
5986 }
5987 }
5988
5989 if (ecb->dte_cond) {
5990 /*
5991 * If the dte_cond bits indicate that this
5992 * consumer is only allowed to see user-mode firings
5993 * of this probe, call the provider's dtps_usermode()
5994 * entry point to check that the probe was fired
5995 * while in a user context. Skip this ECB if that's
5996 * not the case.
5997 */
5998 if ((ecb->dte_cond & DTRACE_COND_USERMODE) &&
5999 prov->dtpv_pops.dtps_usermode(prov->dtpv_arg,
6000 probe->dtpr_id, probe->dtpr_arg) == 0)
6001 continue;
6002
6003#if defined(sun)
6004 /*
6005 * This is more subtle than it looks. We have to be
6006 * absolutely certain that CRED() isn't going to
6007 * change out from under us so it's only legit to
6008 * examine that structure if we're in constrained
6009 * situations. Currently, the only times we'll this
6010 * check is if a non-super-user has enabled the
6011 * profile or syscall providers -- providers that
6012 * allow visibility of all processes. For the
6013 * profile case, the check above will ensure that
6014 * we're examining a user context.
6015 */
6016 if (ecb->dte_cond & DTRACE_COND_OWNER) {
6017 cred_t *cr;
6018 cred_t *s_cr =
6019 ecb->dte_state->dts_cred.dcr_cred;
6020 proc_t *proc;
6021
6022 ASSERT(s_cr != NULL);
6023
6024 if ((cr = CRED()) == NULL ||
6025 s_cr->cr_uid != cr->cr_uid ||
6026 s_cr->cr_uid != cr->cr_ruid ||
6027 s_cr->cr_uid != cr->cr_suid ||
6028 s_cr->cr_gid != cr->cr_gid ||
6029 s_cr->cr_gid != cr->cr_rgid ||
6030 s_cr->cr_gid != cr->cr_sgid ||
6031 (proc = ttoproc(curthread)) == NULL ||
6032 (proc->p_flag & SNOCD))
6033 continue;
6034 }
6035
6036 if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) {
6037 cred_t *cr;
6038 cred_t *s_cr =
6039 ecb->dte_state->dts_cred.dcr_cred;
6040
6041 ASSERT(s_cr != NULL);
6042
6043 if ((cr = CRED()) == NULL ||
6044 s_cr->cr_zone->zone_id !=
6045 cr->cr_zone->zone_id)
6046 continue;
6047 }
6048#endif
6049 }
6050
6051 if (now - state->dts_alive > dtrace_deadman_timeout) {
6052 /*
6053 * We seem to be dead. Unless we (a) have kernel
6054 * destructive permissions (b) have expicitly enabled
6055 * destructive actions and (c) destructive actions have
6056 * not been disabled, we're going to transition into
6057 * the KILLED state, from which no further processing
6058 * on this state will be performed.
6059 */
6060 if (!dtrace_priv_kernel_destructive(state) ||
6061 !state->dts_cred.dcr_destructive ||
6062 dtrace_destructive_disallow) {
6063 void *activity = &state->dts_activity;
6064 dtrace_activity_t current;
6065
6066 do {
6067 current = state->dts_activity;
6068 } while (dtrace_cas32(activity, current,
6069 DTRACE_ACTIVITY_KILLED) != current);
6070
6071 continue;
6072 }
6073 }
6074
6075 if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed,
6076 ecb->dte_alignment, state, &mstate)) < 0)
6077 continue;
6078
6079 tomax = buf->dtb_tomax;
6080 ASSERT(tomax != NULL);
6081
6082 if (ecb->dte_size != 0)
6083 DTRACE_STORE(uint32_t, tomax, offs, ecb->dte_epid);
6084
6085 mstate.dtms_epid = ecb->dte_epid;
6086 mstate.dtms_present |= DTRACE_MSTATE_EPID;
6087
6088 if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL)
6089 mstate.dtms_access = DTRACE_ACCESS_KERNEL;
6090 else
6091 mstate.dtms_access = 0;
6092
6093 if (pred != NULL) {
6094 dtrace_difo_t *dp = pred->dtp_difo;
6095 int rval;
6096
6097 rval = dtrace_dif_emulate(dp, &mstate, vstate, state);
6098
6099 if (!(*flags & CPU_DTRACE_ERROR) && !rval) {
6100 dtrace_cacheid_t cid = probe->dtpr_predcache;
6101
6102 if (cid != DTRACE_CACHEIDNONE && !onintr) {
6103 /*
6104 * Update the predicate cache...
6105 */
6106 ASSERT(cid == pred->dtp_cacheid);
6107 curthread->t_predcache = cid;
6108 }
6109
6110 continue;
6111 }
6112 }
6113
6114 for (act = ecb->dte_action; !(*flags & CPU_DTRACE_ERROR) &&
6115 act != NULL; act = act->dta_next) {
6116 size_t valoffs;
6117 dtrace_difo_t *dp;
6118 dtrace_recdesc_t *rec = &act->dta_rec;
6119
6120 size = rec->dtrd_size;
6121 valoffs = offs + rec->dtrd_offset;
6122
6123 if (DTRACEACT_ISAGG(act->dta_kind)) {
6124 uint64_t v = 0xbad;
6125 dtrace_aggregation_t *agg;
6126
6127 agg = (dtrace_aggregation_t *)act;
6128
6129 if ((dp = act->dta_difo) != NULL)
6130 v = dtrace_dif_emulate(dp,
6131 &mstate, vstate, state);
6132
6133 if (*flags & CPU_DTRACE_ERROR)
6134 continue;
6135
6136 /*
6137 * Note that we always pass the expression
6138 * value from the previous iteration of the
6139 * action loop. This value will only be used
6140 * if there is an expression argument to the
6141 * aggregating action, denoted by the
6142 * dtag_hasarg field.
6143 */
6144 dtrace_aggregate(agg, buf,
6145 offs, aggbuf, v, val);
6146 continue;
6147 }
6148
6149 switch (act->dta_kind) {
6150 case DTRACEACT_STOP:
6151 if (dtrace_priv_proc_destructive(state))
6152 dtrace_action_stop();
6153 continue;
6154
6155 case DTRACEACT_BREAKPOINT:
6156 if (dtrace_priv_kernel_destructive(state))
6157 dtrace_action_breakpoint(ecb);
6158 continue;
6159
6160 case DTRACEACT_PANIC:
6161 if (dtrace_priv_kernel_destructive(state))
6162 dtrace_action_panic(ecb);
6163 continue;
6164
6165 case DTRACEACT_STACK:
6166 if (!dtrace_priv_kernel(state))
6167 continue;
6168
6169 dtrace_getpcstack((pc_t *)(tomax + valoffs),
6170 size / sizeof (pc_t), probe->dtpr_aframes,
6171 DTRACE_ANCHORED(probe) ? NULL :
6172 (uint32_t *)arg0);
6173 continue;
6174
6175#if defined(sun)
6176 case DTRACEACT_JSTACK:
6177 case DTRACEACT_USTACK:
6178 if (!dtrace_priv_proc(state))
6179 continue;
6180
6181 /*
6182 * See comment in DIF_VAR_PID.
6183 */
6184 if (DTRACE_ANCHORED(mstate.dtms_probe) &&
6185 CPU_ON_INTR(CPU)) {
6186 int depth = DTRACE_USTACK_NFRAMES(
6187 rec->dtrd_arg) + 1;
6188
6189 dtrace_bzero((void *)(tomax + valoffs),
6190 DTRACE_USTACK_STRSIZE(rec->dtrd_arg)
6191 + depth * sizeof (uint64_t));
6192
6193 continue;
6194 }
6195
6196 if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 &&
6197 curproc->p_dtrace_helpers != NULL) {
6198 /*
6199 * This is the slow path -- we have
6200 * allocated string space, and we're
6201 * getting the stack of a process that
6202 * has helpers. Call into a separate
6203 * routine to perform this processing.
6204 */
6205 dtrace_action_ustack(&mstate, state,
6206 (uint64_t *)(tomax + valoffs),
6207 rec->dtrd_arg);
6208 continue;
6209 }
6210
6211 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT);
6212 dtrace_getupcstack((uint64_t *)
6213 (tomax + valoffs),
6214 DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + 1);
6215 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT);
6216 continue;
6217#endif
6218
6219 default:
6220 break;
6221 }
6222
6223 dp = act->dta_difo;
6224 ASSERT(dp != NULL);
6225
6226 val = dtrace_dif_emulate(dp, &mstate, vstate, state);
6227
6228 if (*flags & CPU_DTRACE_ERROR)
6229 continue;
6230
6231 switch (act->dta_kind) {
6232 case DTRACEACT_SPECULATE:
6233 ASSERT(buf == &state->dts_buffer[cpuid]);
6234 buf = dtrace_speculation_buffer(state,
6235 cpuid, val);
6236
6237 if (buf == NULL) {
6238 *flags |= CPU_DTRACE_DROP;
6239 continue;
6240 }
6241
6242 offs = dtrace_buffer_reserve(buf,
6243 ecb->dte_needed, ecb->dte_alignment,
6244 state, NULL);
6245
6246 if (offs < 0) {
6247 *flags |= CPU_DTRACE_DROP;
6248 continue;
6249 }
6250
6251 tomax = buf->dtb_tomax;
6252 ASSERT(tomax != NULL);
6253
6254 if (ecb->dte_size != 0)
6255 DTRACE_STORE(uint32_t, tomax, offs,
6256 ecb->dte_epid);
6257 continue;
6258
6259 case DTRACEACT_PRINTM: {
6260 /* The DIF returns a 'memref'. */
6261 uintptr_t *memref = (uintptr_t *)(uintptr_t) val;
6262
6263 /* Get the size from the memref. */
6264 size = memref[1];
6265
6266 /*
6267 * Check if the size exceeds the allocated
6268 * buffer size.
6269 */
6270 if (size + sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) {
6271 /* Flag a drop! */
6272 *flags |= CPU_DTRACE_DROP;
6273 continue;
6274 }
6275
6276 /* Store the size in the buffer first. */
6277 DTRACE_STORE(uintptr_t, tomax,
6278 valoffs, size);
6279
6280 /*
6281 * Offset the buffer address to the start
6282 * of the data.
6283 */
6284 valoffs += sizeof(uintptr_t);
6285
6286 /*
6287 * Reset to the memory address rather than
6288 * the memref array, then let the BYREF
6289 * code below do the work to store the
6290 * memory data in the buffer.
6291 */
6292 val = memref[0];
6293 break;
6294 }
6295
6296 case DTRACEACT_PRINTT: {
6297 /* The DIF returns a 'typeref'. */
6298 uintptr_t *typeref = (uintptr_t *)(uintptr_t) val;
6299 char c = '\0' + 1;
6300 size_t s;
6301
6302 /*
6303 * Get the type string length and round it
6304 * up so that the data that follows is
6305 * aligned for easy access.
6306 */
6307 size_t typs = strlen((char *) typeref[2]) + 1;
6308 typs = roundup(typs, sizeof(uintptr_t));
6309
6310 /*
6311 *Get the size from the typeref using the
6312 * number of elements and the type size.
6313 */
6314 size = typeref[1] * typeref[3];
6315
6316 /*
6317 * Check if the size exceeds the allocated
6318 * buffer size.
6319 */
6320 if (size + typs + 2 * sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) {
6321 /* Flag a drop! */
6322 *flags |= CPU_DTRACE_DROP;
6323
6324 }
6325
6326 /* Store the size in the buffer first. */
6327 DTRACE_STORE(uintptr_t, tomax,
6328 valoffs, size);
6329 valoffs += sizeof(uintptr_t);
6330
6331 /* Store the type size in the buffer. */
6332 DTRACE_STORE(uintptr_t, tomax,
6333 valoffs, typeref[3]);
6334 valoffs += sizeof(uintptr_t);
6335
6336 val = typeref[2];
6337
6338 for (s = 0; s < typs; s++) {
6339 if (c != '\0')
6340 c = dtrace_load8(val++);
6341
6342 DTRACE_STORE(uint8_t, tomax,
6343 valoffs++, c);
6344 }
6345
6346 /*
6347 * Reset to the memory address rather than
6348 * the typeref array, then let the BYREF
6349 * code below do the work to store the
6350 * memory data in the buffer.
6351 */
6352 val = typeref[0];
6353 break;
6354 }
6355
6356 case DTRACEACT_CHILL:
6357 if (dtrace_priv_kernel_destructive(state))
6358 dtrace_action_chill(&mstate, val);
6359 continue;
6360
6361 case DTRACEACT_RAISE:
6362 if (dtrace_priv_proc_destructive(state))
6363 dtrace_action_raise(val);
6364 continue;
6365
6366 case DTRACEACT_COMMIT:
6367 ASSERT(!committed);
6368
6369 /*
6370 * We need to commit our buffer state.
6371 */
6372 if (ecb->dte_size)
6373 buf->dtb_offset = offs + ecb->dte_size;
6374 buf = &state->dts_buffer[cpuid];
6375 dtrace_speculation_commit(state, cpuid, val);
6376 committed = 1;
6377 continue;
6378
6379 case DTRACEACT_DISCARD:
6380 dtrace_speculation_discard(state, cpuid, val);
6381 continue;
6382
6383 case DTRACEACT_DIFEXPR:
6384 case DTRACEACT_LIBACT:
6385 case DTRACEACT_PRINTF:
6386 case DTRACEACT_PRINTA:
6387 case DTRACEACT_SYSTEM:
6388 case DTRACEACT_FREOPEN:
6389 break;
6390
6391 case DTRACEACT_SYM:
6392 case DTRACEACT_MOD:
6393 if (!dtrace_priv_kernel(state))
6394 continue;
6395 break;
6396
6397 case DTRACEACT_USYM:
6398 case DTRACEACT_UMOD:
6399 case DTRACEACT_UADDR: {
6400#if defined(sun)
6401 struct pid *pid = curthread->t_procp->p_pidp;
6402#endif
6403
6404 if (!dtrace_priv_proc(state))
6405 continue;
6406
6407 DTRACE_STORE(uint64_t, tomax,
6408#if defined(sun)
6409 valoffs, (uint64_t)pid->pid_id);
6410#else
6411 valoffs, (uint64_t) curproc->p_pid);
6412#endif
6413 DTRACE_STORE(uint64_t, tomax,
6414 valoffs + sizeof (uint64_t), val);
6415
6416 continue;
6417 }
6418
6419 case DTRACEACT_EXIT: {
6420 /*
6421 * For the exit action, we are going to attempt
6422 * to atomically set our activity to be
6423 * draining. If this fails (either because
6424 * another CPU has beat us to the exit action,
6425 * or because our current activity is something
6426 * other than ACTIVE or WARMUP), we will
6427 * continue. This assures that the exit action
6428 * can be successfully recorded at most once
6429 * when we're in the ACTIVE state. If we're
6430 * encountering the exit() action while in
6431 * COOLDOWN, however, we want to honor the new
6432 * status code. (We know that we're the only
6433 * thread in COOLDOWN, so there is no race.)
6434 */
6435 void *activity = &state->dts_activity;
6436 dtrace_activity_t current = state->dts_activity;
6437
6438 if (current == DTRACE_ACTIVITY_COOLDOWN)
6439 break;
6440
6441 if (current != DTRACE_ACTIVITY_WARMUP)
6442 current = DTRACE_ACTIVITY_ACTIVE;
6443
6444 if (dtrace_cas32(activity, current,
6445 DTRACE_ACTIVITY_DRAINING) != current) {
6446 *flags |= CPU_DTRACE_DROP;
6447 continue;
6448 }
6449
6450 break;
6451 }
6452
6453 default:
6454 ASSERT(0);
6455 }
6456
6457 if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF) {
6458 uintptr_t end = valoffs + size;
6459
6460 if (!dtrace_vcanload((void *)(uintptr_t)val,
6461 &dp->dtdo_rtype, &mstate, vstate))
6462 continue;
6463
6464 /*
6465 * If this is a string, we're going to only
6466 * load until we find the zero byte -- after
6467 * which we'll store zero bytes.
6468 */
6469 if (dp->dtdo_rtype.dtdt_kind ==
6470 DIF_TYPE_STRING) {
6471 char c = '\0' + 1;
6472 int intuple = act->dta_intuple;
6473 size_t s;
6474
6475 for (s = 0; s < size; s++) {
6476 if (c != '\0')
6477 c = dtrace_load8(val++);
6478
6479 DTRACE_STORE(uint8_t, tomax,
6480 valoffs++, c);
6481
6482 if (c == '\0' && intuple)
6483 break;
6484 }
6485
6486 continue;
6487 }
6488
6489 while (valoffs < end) {
6490 DTRACE_STORE(uint8_t, tomax, valoffs++,
6491 dtrace_load8(val++));
6492 }
6493
6494 continue;
6495 }
6496
6497 switch (size) {
6498 case 0:
6499 break;
6500
6501 case sizeof (uint8_t):
6502 DTRACE_STORE(uint8_t, tomax, valoffs, val);
6503 break;
6504 case sizeof (uint16_t):
6505 DTRACE_STORE(uint16_t, tomax, valoffs, val);
6506 break;
6507 case sizeof (uint32_t):
6508 DTRACE_STORE(uint32_t, tomax, valoffs, val);
6509 break;
6510 case sizeof (uint64_t):
6511 DTRACE_STORE(uint64_t, tomax, valoffs, val);
6512 break;
6513 default:
6514 /*
6515 * Any other size should have been returned by
6516 * reference, not by value.
6517 */
6518 ASSERT(0);
6519 break;
6520 }
6521 }
6522
6523 if (*flags & CPU_DTRACE_DROP)
6524 continue;
6525
6526 if (*flags & CPU_DTRACE_FAULT) {
6527 int ndx;
6528 dtrace_action_t *err;
6529
6530 buf->dtb_errors++;
6531
6532 if (probe->dtpr_id == dtrace_probeid_error) {
6533 /*
6534 * There's nothing we can do -- we had an
6535 * error on the error probe. We bump an
6536 * error counter to at least indicate that
6537 * this condition happened.
6538 */
6539 dtrace_error(&state->dts_dblerrors);
6540 continue;
6541 }
6542
6543 if (vtime) {
6544 /*
6545 * Before recursing on dtrace_probe(), we
6546 * need to explicitly clear out our start
6547 * time to prevent it from being accumulated
6548 * into t_dtrace_vtime.
6549 */
6550 curthread->t_dtrace_start = 0;
6551 }
6552
6553 /*
6554 * Iterate over the actions to figure out which action
6555 * we were processing when we experienced the error.
6556 * Note that act points _past_ the faulting action; if
6557 * act is ecb->dte_action, the fault was in the
6558 * predicate, if it's ecb->dte_action->dta_next it's
6559 * in action #1, and so on.
6560 */
6561 for (err = ecb->dte_action, ndx = 0;
6562 err != act; err = err->dta_next, ndx++)
6563 continue;
6564
6565 dtrace_probe_error(state, ecb->dte_epid, ndx,
6566 (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) ?
6567 mstate.dtms_fltoffs : -1, DTRACE_FLAGS2FLT(*flags),
6568 cpu_core[cpuid].cpuc_dtrace_illval);
6569
6570 continue;
6571 }
6572
6573 if (!committed)
6574 buf->dtb_offset = offs + ecb->dte_size;
6575 }
6576
6577 if (vtime)
6578 curthread->t_dtrace_start = dtrace_gethrtime();
6579
6580 dtrace_interrupt_enable(cookie);
6581}
6582
6583/*
6584 * DTrace Probe Hashing Functions
6585 *
6586 * The functions in this section (and indeed, the functions in remaining
6587 * sections) are not _called_ from probe context. (Any exceptions to this are
6588 * marked with a "Note:".) Rather, they are called from elsewhere in the
6589 * DTrace framework to look-up probes in, add probes to and remove probes from
6590 * the DTrace probe hashes. (Each probe is hashed by each element of the
6591 * probe tuple -- allowing for fast lookups, regardless of what was
6592 * specified.)
6593 */
6594static uint_t
6595dtrace_hash_str(const char *p)
6596{
6597 unsigned int g;
6598 uint_t hval = 0;
6599
6600 while (*p) {
6601 hval = (hval << 4) + *p++;
6602 if ((g = (hval & 0xf0000000)) != 0)
6603 hval ^= g >> 24;
6604 hval &= ~g;
6605 }
6606 return (hval);
6607}
6608
6609static dtrace_hash_t *
6610dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, uintptr_t prevoffs)
6611{
6612 dtrace_hash_t *hash = kmem_zalloc(sizeof (dtrace_hash_t), KM_SLEEP);
6613
6614 hash->dth_stroffs = stroffs;
6615 hash->dth_nextoffs = nextoffs;
6616 hash->dth_prevoffs = prevoffs;
6617
6618 hash->dth_size = 1;
6619 hash->dth_mask = hash->dth_size - 1;
6620
6621 hash->dth_tab = kmem_zalloc(hash->dth_size *
6622 sizeof (dtrace_hashbucket_t *), KM_SLEEP);
6623
6624 return (hash);
6625}
6626
6627static void
6628dtrace_hash_destroy(dtrace_hash_t *hash)
6629{
6630#ifdef DEBUG
6631 int i;
6632
6633 for (i = 0; i < hash->dth_size; i++)
6634 ASSERT(hash->dth_tab[i] == NULL);
6635#endif
6636
6637 kmem_free(hash->dth_tab,
6638 hash->dth_size * sizeof (dtrace_hashbucket_t *));
6639 kmem_free(hash, sizeof (dtrace_hash_t));
6640}
6641
6642static void
6643dtrace_hash_resize(dtrace_hash_t *hash)
6644{
6645 int size = hash->dth_size, i, ndx;
6646 int new_size = hash->dth_size << 1;
6647 int new_mask = new_size - 1;
6648 dtrace_hashbucket_t **new_tab, *bucket, *next;
6649
6650 ASSERT((new_size & new_mask) == 0);
6651
6652 new_tab = kmem_zalloc(new_size * sizeof (void *), KM_SLEEP);
6653
6654 for (i = 0; i < size; i++) {
6655 for (bucket = hash->dth_tab[i]; bucket != NULL; bucket = next) {
6656 dtrace_probe_t *probe = bucket->dthb_chain;
6657
6658 ASSERT(probe != NULL);
6659 ndx = DTRACE_HASHSTR(hash, probe) & new_mask;
6660
6661 next = bucket->dthb_next;
6662 bucket->dthb_next = new_tab[ndx];
6663 new_tab[ndx] = bucket;
6664 }
6665 }
6666
6667 kmem_free(hash->dth_tab, hash->dth_size * sizeof (void *));
6668 hash->dth_tab = new_tab;
6669 hash->dth_size = new_size;
6670 hash->dth_mask = new_mask;
6671}
6672
6673static void
6674dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new)
6675{
6676 int hashval = DTRACE_HASHSTR(hash, new);
6677 int ndx = hashval & hash->dth_mask;
6678 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6679 dtrace_probe_t **nextp, **prevp;
6680
6681 for (; bucket != NULL; bucket = bucket->dthb_next) {
6682 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new))
6683 goto add;
6684 }
6685
6686 if ((hash->dth_nbuckets >> 1) > hash->dth_size) {
6687 dtrace_hash_resize(hash);
6688 dtrace_hash_add(hash, new);
6689 return;
6690 }
6691
6692 bucket = kmem_zalloc(sizeof (dtrace_hashbucket_t), KM_SLEEP);
6693 bucket->dthb_next = hash->dth_tab[ndx];
6694 hash->dth_tab[ndx] = bucket;
6695 hash->dth_nbuckets++;
6696
6697add:
6698 nextp = DTRACE_HASHNEXT(hash, new);
6699 ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL);
6700 *nextp = bucket->dthb_chain;
6701
6702 if (bucket->dthb_chain != NULL) {
6703 prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain);
6704 ASSERT(*prevp == NULL);
6705 *prevp = new;
6706 }
6707
6708 bucket->dthb_chain = new;
6709 bucket->dthb_len++;
6710}
6711
6712static dtrace_probe_t *
6713dtrace_hash_lookup(dtrace_hash_t *hash, dtrace_probe_t *template)
6714{
6715 int hashval = DTRACE_HASHSTR(hash, template);
6716 int ndx = hashval & hash->dth_mask;
6717 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6718
6719 for (; bucket != NULL; bucket = bucket->dthb_next) {
6720 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template))
6721 return (bucket->dthb_chain);
6722 }
6723
6724 return (NULL);
6725}
6726
6727static int
6728dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template)
6729{
6730 int hashval = DTRACE_HASHSTR(hash, template);
6731 int ndx = hashval & hash->dth_mask;
6732 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6733
6734 for (; bucket != NULL; bucket = bucket->dthb_next) {
6735 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template))
6736 return (bucket->dthb_len);
6737 }
6738
6739 return (0);
6740}
6741
6742static void
6743dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe)
6744{
6745 int ndx = DTRACE_HASHSTR(hash, probe) & hash->dth_mask;
6746 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx];
6747
6748 dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe);
6749 dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe);
6750
6751 /*
6752 * Find the bucket that we're removing this probe from.
6753 */
6754 for (; bucket != NULL; bucket = bucket->dthb_next) {
6755 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe))
6756 break;
6757 }
6758
6759 ASSERT(bucket != NULL);
6760
6761 if (*prevp == NULL) {
6762 if (*nextp == NULL) {
6763 /*
6764 * The removed probe was the only probe on this
6765 * bucket; we need to remove the bucket.
6766 */
6767 dtrace_hashbucket_t *b = hash->dth_tab[ndx];
6768
6769 ASSERT(bucket->dthb_chain == probe);
6770 ASSERT(b != NULL);
6771
6772 if (b == bucket) {
6773 hash->dth_tab[ndx] = bucket->dthb_next;
6774 } else {
6775 while (b->dthb_next != bucket)
6776 b = b->dthb_next;
6777 b->dthb_next = bucket->dthb_next;
6778 }
6779
6780 ASSERT(hash->dth_nbuckets > 0);
6781 hash->dth_nbuckets--;
6782 kmem_free(bucket, sizeof (dtrace_hashbucket_t));
6783 return;
6784 }
6785
6786 bucket->dthb_chain = *nextp;
6787 } else {
6788 *(DTRACE_HASHNEXT(hash, *prevp)) = *nextp;
6789 }
6790
6791 if (*nextp != NULL)
6792 *(DTRACE_HASHPREV(hash, *nextp)) = *prevp;
6793}
6794
6795/*
6796 * DTrace Utility Functions
6797 *
6798 * These are random utility functions that are _not_ called from probe context.
6799 */
6800static int
6801dtrace_badattr(const dtrace_attribute_t *a)
6802{
6803 return (a->dtat_name > DTRACE_STABILITY_MAX ||
6804 a->dtat_data > DTRACE_STABILITY_MAX ||
6805 a->dtat_class > DTRACE_CLASS_MAX);
6806}
6807
6808/*
6809 * Return a duplicate copy of a string. If the specified string is NULL,
6810 * this function returns a zero-length string.
6811 */
6812static char *
6813dtrace_strdup(const char *str)
6814{
6815 char *new = kmem_zalloc((str != NULL ? strlen(str) : 0) + 1, KM_SLEEP);
6816
6817 if (str != NULL)
6818 (void) strcpy(new, str);
6819
6820 return (new);
6821}
6822
6823#define DTRACE_ISALPHA(c) \
6824 (((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z'))
6825
6826static int
6827dtrace_badname(const char *s)
6828{
6829 char c;
6830
6831 if (s == NULL || (c = *s++) == '\0')
6832 return (0);
6833
6834 if (!DTRACE_ISALPHA(c) && c != '-' && c != '_' && c != '.')
6835 return (1);
6836
6837 while ((c = *s++) != '\0') {
6838 if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') &&
6839 c != '-' && c != '_' && c != '.' && c != '`')
6840 return (1);
6841 }
6842
6843 return (0);
6844}
6845
6846static void
6847dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp)
6848{
6849 uint32_t priv;
6850
6851#if defined(sun)
6852 if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) {
6853 /*
6854 * For DTRACE_PRIV_ALL, the uid and zoneid don't matter.
6855 */
6856 priv = DTRACE_PRIV_ALL;
6857 } else {
6858 *uidp = crgetuid(cr);
6859 *zoneidp = crgetzoneid(cr);
6860
6861 priv = 0;
6862 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE))
6863 priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER;
6864 else if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE))
6865 priv |= DTRACE_PRIV_USER;
6866 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE))
6867 priv |= DTRACE_PRIV_PROC;
6868 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
6869 priv |= DTRACE_PRIV_OWNER;
6870 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
6871 priv |= DTRACE_PRIV_ZONEOWNER;
6872 }
6873#else
6874 priv = DTRACE_PRIV_ALL;
6875#endif
6876
6877 *privp = priv;
6878}
6879
6880#ifdef DTRACE_ERRDEBUG
6881static void
6882dtrace_errdebug(const char *str)
6883{
6884 int hval = dtrace_hash_str(str) % DTRACE_ERRHASHSZ;
6885 int occupied = 0;
6886
6887 mutex_enter(&dtrace_errlock);
6888 dtrace_errlast = str;
6889 dtrace_errthread = curthread;
6890
6891 while (occupied++ < DTRACE_ERRHASHSZ) {
6892 if (dtrace_errhash[hval].dter_msg == str) {
6893 dtrace_errhash[hval].dter_count++;
6894 goto out;
6895 }
6896
6897 if (dtrace_errhash[hval].dter_msg != NULL) {
6898 hval = (hval + 1) % DTRACE_ERRHASHSZ;
6899 continue;
6900 }
6901
6902 dtrace_errhash[hval].dter_msg = str;
6903 dtrace_errhash[hval].dter_count = 1;
6904 goto out;
6905 }
6906
6907 panic("dtrace: undersized error hash");
6908out:
6909 mutex_exit(&dtrace_errlock);
6910}
6911#endif
6912
6913/*
6914 * DTrace Matching Functions
6915 *
6916 * These functions are used to match groups of probes, given some elements of
6917 * a probe tuple, or some globbed expressions for elements of a probe tuple.
6918 */
6919static int
6920dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid,
6921 zoneid_t zoneid)
6922{
6923 if (priv != DTRACE_PRIV_ALL) {
6924 uint32_t ppriv = prp->dtpr_provider->dtpv_priv.dtpp_flags;
6925 uint32_t match = priv & ppriv;
6926
6927 /*
6928 * No PRIV_DTRACE_* privileges...
6929 */
6930 if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER |
6931 DTRACE_PRIV_KERNEL)) == 0)
6932 return (0);
6933
6934 /*
6935 * No matching bits, but there were bits to match...
6936 */
6937 if (match == 0 && ppriv != 0)
6938 return (0);
6939
6940 /*
6941 * Need to have permissions to the process, but don't...
6942 */
6943 if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 &&
6944 uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) {
6945 return (0);
6946 }
6947
6948 /*
6949 * Need to be in the same zone unless we possess the
6950 * privilege to examine all zones.
6951 */
6952 if (((ppriv & ~match) & DTRACE_PRIV_ZONEOWNER) != 0 &&
6953 zoneid != prp->dtpr_provider->dtpv_priv.dtpp_zoneid) {
6954 return (0);
6955 }
6956 }
6957
6958 return (1);
6959}
6960
6961/*
6962 * dtrace_match_probe compares a dtrace_probe_t to a pre-compiled key, which
6963 * consists of input pattern strings and an ops-vector to evaluate them.
6964 * This function returns >0 for match, 0 for no match, and <0 for error.
6965 */
6966static int
6967dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp,
6968 uint32_t priv, uid_t uid, zoneid_t zoneid)
6969{
6970 dtrace_provider_t *pvp = prp->dtpr_provider;
6971 int rv;
6972
6973 if (pvp->dtpv_defunct)
6974 return (0);
6975
6976 if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0)
6977 return (rv);
6978
6979 if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0)
6980 return (rv);
6981
6982 if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0)
6983 return (rv);
6984
6985 if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0)
6986 return (rv);
6987
6988 if (dtrace_match_priv(prp, priv, uid, zoneid) == 0)
6989 return (0);
6990
6991 return (rv);
6992}
6993
6994/*
6995 * dtrace_match_glob() is a safe kernel implementation of the gmatch(3GEN)
6996 * interface for matching a glob pattern 'p' to an input string 's'. Unlike
6997 * libc's version, the kernel version only applies to 8-bit ASCII strings.
6998 * In addition, all of the recursion cases except for '*' matching have been
6999 * unwound. For '*', we still implement recursive evaluation, but a depth
7000 * counter is maintained and matching is aborted if we recurse too deep.
7001 * The function returns 0 if no match, >0 if match, and <0 if recursion error.
7002 */
7003static int
7004dtrace_match_glob(const char *s, const char *p, int depth)
7005{
7006 const char *olds;
7007 char s1, c;
7008 int gs;
7009
7010 if (depth > DTRACE_PROBEKEY_MAXDEPTH)
7011 return (-1);
7012
7013 if (s == NULL)
7014 s = ""; /* treat NULL as empty string */
7015
7016top:
7017 olds = s;
7018 s1 = *s++;
7019
7020 if (p == NULL)
7021 return (0);
7022
7023 if ((c = *p++) == '\0')
7024 return (s1 == '\0');
7025
7026 switch (c) {
7027 case '[': {
7028 int ok = 0, notflag = 0;
7029 char lc = '\0';
7030
7031 if (s1 == '\0')
7032 return (0);
7033
7034 if (*p == '!') {
7035 notflag = 1;
7036 p++;
7037 }
7038
7039 if ((c = *p++) == '\0')
7040 return (0);
7041
7042 do {
7043 if (c == '-' && lc != '\0' && *p != ']') {
7044 if ((c = *p++) == '\0')
7045 return (0);
7046 if (c == '\\' && (c = *p++) == '\0')
7047 return (0);
7048
7049 if (notflag) {
7050 if (s1 < lc || s1 > c)
7051 ok++;
7052 else
7053 return (0);
7054 } else if (lc <= s1 && s1 <= c)
7055 ok++;
7056
7057 } else if (c == '\\' && (c = *p++) == '\0')
7058 return (0);
7059
7060 lc = c; /* save left-hand 'c' for next iteration */
7061
7062 if (notflag) {
7063 if (s1 != c)
7064 ok++;
7065 else
7066 return (0);
7067 } else if (s1 == c)
7068 ok++;
7069
7070 if ((c = *p++) == '\0')
7071 return (0);
7072
7073 } while (c != ']');
7074
7075 if (ok)
7076 goto top;
7077
7078 return (0);
7079 }
7080
7081 case '\\':
7082 if ((c = *p++) == '\0')
7083 return (0);
7084 /*FALLTHRU*/
7085
7086 default:
7087 if (c != s1)
7088 return (0);
7089 /*FALLTHRU*/
7090
7091 case '?':
7092 if (s1 != '\0')
7093 goto top;
7094 return (0);
7095
7096 case '*':
7097 while (*p == '*')
7098 p++; /* consecutive *'s are identical to a single one */
7099
7100 if (*p == '\0')
7101 return (1);
7102
7103 for (s = olds; *s != '\0'; s++) {
7104 if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0)
7105 return (gs);
7106 }
7107
7108 return (0);
7109 }
7110}
7111
7112/*ARGSUSED*/
7113static int
7114dtrace_match_string(const char *s, const char *p, int depth)
7115{
7116 return (s != NULL && strcmp(s, p) == 0);
7117}
7118
7119/*ARGSUSED*/
7120static int
7121dtrace_match_nul(const char *s, const char *p, int depth)
7122{
7123 return (1); /* always match the empty pattern */
7124}
7125
7126/*ARGSUSED*/
7127static int
7128dtrace_match_nonzero(const char *s, const char *p, int depth)
7129{
7130 return (s != NULL && s[0] != '\0');
7131}
7132
7133static int
7134dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid,
7135 zoneid_t zoneid, int (*matched)(dtrace_probe_t *, void *), void *arg)
7136{
7137 dtrace_probe_t template, *probe;
7138 dtrace_hash_t *hash = NULL;
7139 int len, best = INT_MAX, nmatched = 0;
7140 dtrace_id_t i;
7141
7142 ASSERT(MUTEX_HELD(&dtrace_lock));
7143
7144 /*
7145 * If the probe ID is specified in the key, just lookup by ID and
7146 * invoke the match callback once if a matching probe is found.
7147 */
7148 if (pkp->dtpk_id != DTRACE_IDNONE) {
7149 if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL &&
7150 dtrace_match_probe(probe, pkp, priv, uid, zoneid) > 0) {
7151 (void) (*matched)(probe, arg);
7152 nmatched++;
7153 }
7154 return (nmatched);
7155 }
7156
7157 template.dtpr_mod = (char *)pkp->dtpk_mod;
7158 template.dtpr_func = (char *)pkp->dtpk_func;
7159 template.dtpr_name = (char *)pkp->dtpk_name;
7160
7161 /*
7162 * We want to find the most distinct of the module name, function
7163 * name, and name. So for each one that is not a glob pattern or
7164 * empty string, we perform a lookup in the corresponding hash and
7165 * use the hash table with the fewest collisions to do our search.
7166 */
7167 if (pkp->dtpk_mmatch == &dtrace_match_string &&
7168 (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) {
7169 best = len;
7170 hash = dtrace_bymod;
7171 }
7172
7173 if (pkp->dtpk_fmatch == &dtrace_match_string &&
7174 (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) {
7175 best = len;
7176 hash = dtrace_byfunc;
7177 }
7178
7179 if (pkp->dtpk_nmatch == &dtrace_match_string &&
7180 (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) {
7181 best = len;
7182 hash = dtrace_byname;
7183 }
7184
7185 /*
7186 * If we did not select a hash table, iterate over every probe and
7187 * invoke our callback for each one that matches our input probe key.
7188 */
7189 if (hash == NULL) {
7190 for (i = 0; i < dtrace_nprobes; i++) {
7191 if ((probe = dtrace_probes[i]) == NULL ||
7192 dtrace_match_probe(probe, pkp, priv, uid,
7193 zoneid) <= 0)
7194 continue;
7195
7196 nmatched++;
7197
7198 if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT)
7199 break;
7200 }
7201
7202 return (nmatched);
7203 }
7204
7205 /*
7206 * If we selected a hash table, iterate over each probe of the same key
7207 * name and invoke the callback for every probe that matches the other
7208 * attributes of our input probe key.
7209 */
7210 for (probe = dtrace_hash_lookup(hash, &template); probe != NULL;
7211 probe = *(DTRACE_HASHNEXT(hash, probe))) {
7212
7213 if (dtrace_match_probe(probe, pkp, priv, uid, zoneid) <= 0)
7214 continue;
7215
7216 nmatched++;
7217
7218 if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT)
7219 break;
7220 }
7221
7222 return (nmatched);
7223}
7224
7225/*
7226 * Return the function pointer dtrace_probecmp() should use to compare the
7227 * specified pattern with a string. For NULL or empty patterns, we select
7228 * dtrace_match_nul(). For glob pattern strings, we use dtrace_match_glob().
7229 * For non-empty non-glob strings, we use dtrace_match_string().
7230 */
7231static dtrace_probekey_f *
7232dtrace_probekey_func(const char *p)
7233{
7234 char c;
7235
7236 if (p == NULL || *p == '\0')
7237 return (&dtrace_match_nul);
7238
7239 while ((c = *p++) != '\0') {
7240 if (c == '[' || c == '?' || c == '*' || c == '\\')
7241 return (&dtrace_match_glob);
7242 }
7243
7244 return (&dtrace_match_string);
7245}
7246
7247/*
7248 * Build a probe comparison key for use with dtrace_match_probe() from the
7249 * given probe description. By convention, a null key only matches anchored
7250 * probes: if each field is the empty string, reset dtpk_fmatch to
7251 * dtrace_match_nonzero().
7252 */
7253static void
7254dtrace_probekey(dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp)
7255{
7256 pkp->dtpk_prov = pdp->dtpd_provider;
7257 pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider);
7258
7259 pkp->dtpk_mod = pdp->dtpd_mod;
7260 pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod);
7261
7262 pkp->dtpk_func = pdp->dtpd_func;
7263 pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func);
7264
7265 pkp->dtpk_name = pdp->dtpd_name;
7266 pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name);
7267
7268 pkp->dtpk_id = pdp->dtpd_id;
7269
7270 if (pkp->dtpk_id == DTRACE_IDNONE &&
7271 pkp->dtpk_pmatch == &dtrace_match_nul &&
7272 pkp->dtpk_mmatch == &dtrace_match_nul &&
7273 pkp->dtpk_fmatch == &dtrace_match_nul &&
7274 pkp->dtpk_nmatch == &dtrace_match_nul)
7275 pkp->dtpk_fmatch = &dtrace_match_nonzero;
7276}
7277
7278/*
7279 * DTrace Provider-to-Framework API Functions
7280 *
7281 * These functions implement much of the Provider-to-Framework API, as
7282 * described in <sys/dtrace.h>. The parts of the API not in this section are
7283 * the functions in the API for probe management (found below), and
7284 * dtrace_probe() itself (found above).
7285 */
7286
7287/*
7288 * Register the calling provider with the DTrace framework. This should
7289 * generally be called by DTrace providers in their attach(9E) entry point.
7290 */
7291int
7292dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv,
7293 cred_t *cr, const dtrace_pops_t *pops, void *arg, dtrace_provider_id_t *idp)
7294{
7295 dtrace_provider_t *provider;
7296
7297 if (name == NULL || pap == NULL || pops == NULL || idp == NULL) {
7298 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7299 "arguments", name ? name : "<NULL>");
7300 return (EINVAL);
7301 }
7302
7303 if (name[0] == '\0' || dtrace_badname(name)) {
7304 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7305 "provider name", name);
7306 return (EINVAL);
7307 }
7308
7309 if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) ||
7310 pops->dtps_enable == NULL || pops->dtps_disable == NULL ||
7311 pops->dtps_destroy == NULL ||
7312 ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) {
7313 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7314 "provider ops", name);
7315 return (EINVAL);
7316 }
7317
7318 if (dtrace_badattr(&pap->dtpa_provider) ||
7319 dtrace_badattr(&pap->dtpa_mod) ||
7320 dtrace_badattr(&pap->dtpa_func) ||
7321 dtrace_badattr(&pap->dtpa_name) ||
7322 dtrace_badattr(&pap->dtpa_args)) {
7323 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7324 "provider attributes", name);
7325 return (EINVAL);
7326 }
7327
7328 if (priv & ~DTRACE_PRIV_ALL) {
7329 cmn_err(CE_WARN, "failed to register provider '%s': invalid "
7330 "privilege attributes", name);
7331 return (EINVAL);
7332 }
7333
7334 if ((priv & DTRACE_PRIV_KERNEL) &&
7335 (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) &&
7336 pops->dtps_usermode == NULL) {
7337 cmn_err(CE_WARN, "failed to register provider '%s': need "
7338 "dtps_usermode() op for given privilege attributes", name);
7339 return (EINVAL);
7340 }
7341
7342 provider = kmem_zalloc(sizeof (dtrace_provider_t), KM_SLEEP);
7343 provider->dtpv_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
7344 (void) strcpy(provider->dtpv_name, name);
7345
7346 provider->dtpv_attr = *pap;
7347 provider->dtpv_priv.dtpp_flags = priv;
7348 if (cr != NULL) {
7349 provider->dtpv_priv.dtpp_uid = crgetuid(cr);
7350 provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr);
7351 }
7352 provider->dtpv_pops = *pops;
7353
7354 if (pops->dtps_provide == NULL) {
7355 ASSERT(pops->dtps_provide_module != NULL);
7356 provider->dtpv_pops.dtps_provide =
7357 (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop;
7358 }
7359
7360 if (pops->dtps_provide_module == NULL) {
7361 ASSERT(pops->dtps_provide != NULL);
7362 provider->dtpv_pops.dtps_provide_module =
7363 (void (*)(void *, modctl_t *))dtrace_nullop;
7364 }
7365
7366 if (pops->dtps_suspend == NULL) {
7367 ASSERT(pops->dtps_resume == NULL);
7368 provider->dtpv_pops.dtps_suspend =
7369 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop;
7370 provider->dtpv_pops.dtps_resume =
7371 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop;
7372 }
7373
7374 provider->dtpv_arg = arg;
7375 *idp = (dtrace_provider_id_t)provider;
7376
7377 if (pops == &dtrace_provider_ops) {
7378 ASSERT(MUTEX_HELD(&dtrace_provider_lock));
7379 ASSERT(MUTEX_HELD(&dtrace_lock));
7380 ASSERT(dtrace_anon.dta_enabling == NULL);
7381
7382 /*
7383 * We make sure that the DTrace provider is at the head of
7384 * the provider chain.
7385 */
7386 provider->dtpv_next = dtrace_provider;
7387 dtrace_provider = provider;
7388 return (0);
7389 }
7390
7391 mutex_enter(&dtrace_provider_lock);
7392 mutex_enter(&dtrace_lock);
7393
7394 /*
7395 * If there is at least one provider registered, we'll add this
7396 * provider after the first provider.
7397 */
7398 if (dtrace_provider != NULL) {
7399 provider->dtpv_next = dtrace_provider->dtpv_next;
7400 dtrace_provider->dtpv_next = provider;
7401 } else {
7402 dtrace_provider = provider;
7403 }
7404
7405 if (dtrace_retained != NULL) {
7406 dtrace_enabling_provide(provider);
7407
7408 /*
7409 * Now we need to call dtrace_enabling_matchall() -- which
7410 * will acquire cpu_lock and dtrace_lock. We therefore need
7411 * to drop all of our locks before calling into it...
7412 */
7413 mutex_exit(&dtrace_lock);
7414 mutex_exit(&dtrace_provider_lock);
7415 dtrace_enabling_matchall();
7416
7417 return (0);
7418 }
7419
7420 mutex_exit(&dtrace_lock);
7421 mutex_exit(&dtrace_provider_lock);
7422
7423 return (0);
7424}
7425
7426/*
7427 * Unregister the specified provider from the DTrace framework. This should
7428 * generally be called by DTrace providers in their detach(9E) entry point.
7429 */
7430int
7431dtrace_unregister(dtrace_provider_id_t id)
7432{
7433 dtrace_provider_t *old = (dtrace_provider_t *)id;
7434 dtrace_provider_t *prev = NULL;
7435 int i, self = 0;
7436 dtrace_probe_t *probe, *first = NULL;
7437
7438 if (old->dtpv_pops.dtps_enable ==
7439 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop) {
7440 /*
7441 * If DTrace itself is the provider, we're called with locks
7442 * already held.
7443 */
7444 ASSERT(old == dtrace_provider);
7445#if defined(sun)
7446 ASSERT(dtrace_devi != NULL);
7447#endif
7448 ASSERT(MUTEX_HELD(&dtrace_provider_lock));
7449 ASSERT(MUTEX_HELD(&dtrace_lock));
7450 self = 1;
7451
7452 if (dtrace_provider->dtpv_next != NULL) {
7453 /*
7454 * There's another provider here; return failure.
7455 */
7456 return (EBUSY);
7457 }
7458 } else {
7459 mutex_enter(&dtrace_provider_lock);
7460 mutex_enter(&mod_lock);
7461 mutex_enter(&dtrace_lock);
7462 }
7463
7464 /*
7465 * If anyone has /dev/dtrace open, or if there are anonymous enabled
7466 * probes, we refuse to let providers slither away, unless this
7467 * provider has already been explicitly invalidated.
7468 */
7469 if (!old->dtpv_defunct &&
7470 (dtrace_opens || (dtrace_anon.dta_state != NULL &&
7471 dtrace_anon.dta_state->dts_necbs > 0))) {
7472 if (!self) {
7473 mutex_exit(&dtrace_lock);
7474 mutex_exit(&mod_lock);
7475 mutex_exit(&dtrace_provider_lock);
7476 }
7477 return (EBUSY);
7478 }
7479
7480 /*
7481 * Attempt to destroy the probes associated with this provider.
7482 */
7483 for (i = 0; i < dtrace_nprobes; i++) {
7484 if ((probe = dtrace_probes[i]) == NULL)
7485 continue;
7486
7487 if (probe->dtpr_provider != old)
7488 continue;
7489
7490 if (probe->dtpr_ecb == NULL)
7491 continue;
7492
7493 /*
7494 * We have at least one ECB; we can't remove this provider.
7495 */
7496 if (!self) {
7497 mutex_exit(&dtrace_lock);
7498 mutex_exit(&mod_lock);
7499 mutex_exit(&dtrace_provider_lock);
7500 }
7501 return (EBUSY);
7502 }
7503
7504 /*
7505 * All of the probes for this provider are disabled; we can safely
7506 * remove all of them from their hash chains and from the probe array.
7507 */
7508 for (i = 0; i < dtrace_nprobes; i++) {
7509 if ((probe = dtrace_probes[i]) == NULL)
7510 continue;
7511
7512 if (probe->dtpr_provider != old)
7513 continue;
7514
7515 dtrace_probes[i] = NULL;
7516
7517 dtrace_hash_remove(dtrace_bymod, probe);
7518 dtrace_hash_remove(dtrace_byfunc, probe);
7519 dtrace_hash_remove(dtrace_byname, probe);
7520
7521 if (first == NULL) {
7522 first = probe;
7523 probe->dtpr_nextmod = NULL;
7524 } else {
7525 probe->dtpr_nextmod = first;
7526 first = probe;
7527 }
7528 }
7529
7530 /*
7531 * The provider's probes have been removed from the hash chains and
7532 * from the probe array. Now issue a dtrace_sync() to be sure that
7533 * everyone has cleared out from any probe array processing.
7534 */
7535 dtrace_sync();
7536
7537 for (probe = first; probe != NULL; probe = first) {
7538 first = probe->dtpr_nextmod;
7539
7540 old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe->dtpr_id,
7541 probe->dtpr_arg);
7542 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
7543 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
7544 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
7545#if defined(sun)
7546 vmem_free(dtrace_arena, (void *)(uintptr_t)(probe->dtpr_id), 1);
7547#else
7548 free_unr(dtrace_arena, probe->dtpr_id);
7549#endif
7550 kmem_free(probe, sizeof (dtrace_probe_t));
7551 }
7552
7553 if ((prev = dtrace_provider) == old) {
7554#if defined(sun)
7555 ASSERT(self || dtrace_devi == NULL);
7556 ASSERT(old->dtpv_next == NULL || dtrace_devi == NULL);
7557#endif
7558 dtrace_provider = old->dtpv_next;
7559 } else {
7560 while (prev != NULL && prev->dtpv_next != old)
7561 prev = prev->dtpv_next;
7562
7563 if (prev == NULL) {
7564 panic("attempt to unregister non-existent "
7565 "dtrace provider %p\n", (void *)id);
7566 }
7567
7568 prev->dtpv_next = old->dtpv_next;
7569 }
7570
7571 if (!self) {
7572 mutex_exit(&dtrace_lock);
7573 mutex_exit(&mod_lock);
7574 mutex_exit(&dtrace_provider_lock);
7575 }
7576
7577 kmem_free(old->dtpv_name, strlen(old->dtpv_name) + 1);
7578 kmem_free(old, sizeof (dtrace_provider_t));
7579
7580 return (0);
7581}
7582
7583/*
7584 * Invalidate the specified provider. All subsequent probe lookups for the
7585 * specified provider will fail, but its probes will not be removed.
7586 */
7587void
7588dtrace_invalidate(dtrace_provider_id_t id)
7589{
7590 dtrace_provider_t *pvp = (dtrace_provider_t *)id;
7591
7592 ASSERT(pvp->dtpv_pops.dtps_enable !=
7593 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop);
7594
7595 mutex_enter(&dtrace_provider_lock);
7596 mutex_enter(&dtrace_lock);
7597
7598 pvp->dtpv_defunct = 1;
7599
7600 mutex_exit(&dtrace_lock);
7601 mutex_exit(&dtrace_provider_lock);
7602}
7603
7604/*
7605 * Indicate whether or not DTrace has attached.
7606 */
7607int
7608dtrace_attached(void)
7609{
7610 /*
7611 * dtrace_provider will be non-NULL iff the DTrace driver has
7612 * attached. (It's non-NULL because DTrace is always itself a
7613 * provider.)
7614 */
7615 return (dtrace_provider != NULL);
7616}
7617
7618/*
7619 * Remove all the unenabled probes for the given provider. This function is
7620 * not unlike dtrace_unregister(), except that it doesn't remove the provider
7621 * -- just as many of its associated probes as it can.
7622 */
7623int
7624dtrace_condense(dtrace_provider_id_t id)
7625{
7626 dtrace_provider_t *prov = (dtrace_provider_t *)id;
7627 int i;
7628 dtrace_probe_t *probe;
7629
7630 /*
7631 * Make sure this isn't the dtrace provider itself.
7632 */
7633 ASSERT(prov->dtpv_pops.dtps_enable !=
7634 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop);
7635
7636 mutex_enter(&dtrace_provider_lock);
7637 mutex_enter(&dtrace_lock);
7638
7639 /*
7640 * Attempt to destroy the probes associated with this provider.
7641 */
7642 for (i = 0; i < dtrace_nprobes; i++) {
7643 if ((probe = dtrace_probes[i]) == NULL)
7644 continue;
7645
7646 if (probe->dtpr_provider != prov)
7647 continue;
7648
7649 if (probe->dtpr_ecb != NULL)
7650 continue;
7651
7652 dtrace_probes[i] = NULL;
7653
7654 dtrace_hash_remove(dtrace_bymod, probe);
7655 dtrace_hash_remove(dtrace_byfunc, probe);
7656 dtrace_hash_remove(dtrace_byname, probe);
7657
7658 prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, i + 1,
7659 probe->dtpr_arg);
7660 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
7661 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
7662 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
7663 kmem_free(probe, sizeof (dtrace_probe_t));
7664#if defined(sun)
7665 vmem_free(dtrace_arena, (void *)((uintptr_t)i + 1), 1);
7666#else
7667 free_unr(dtrace_arena, i + 1);
7668#endif
7669 }
7670
7671 mutex_exit(&dtrace_lock);
7672 mutex_exit(&dtrace_provider_lock);
7673
7674 return (0);
7675}
7676
7677/*
7678 * DTrace Probe Management Functions
7679 *
7680 * The functions in this section perform the DTrace probe management,
7681 * including functions to create probes, look-up probes, and call into the
7682 * providers to request that probes be provided. Some of these functions are
7683 * in the Provider-to-Framework API; these functions can be identified by the
7684 * fact that they are not declared "static".
7685 */
7686
7687/*
7688 * Create a probe with the specified module name, function name, and name.
7689 */
7690dtrace_id_t
7691dtrace_probe_create(dtrace_provider_id_t prov, const char *mod,
7692 const char *func, const char *name, int aframes, void *arg)
7693{
7694 dtrace_probe_t *probe, **probes;
7695 dtrace_provider_t *provider = (dtrace_provider_t *)prov;
7696 dtrace_id_t id;
7697
7698 if (provider == dtrace_provider) {
7699 ASSERT(MUTEX_HELD(&dtrace_lock));
7700 } else {
7701 mutex_enter(&dtrace_lock);
7702 }
7703
7704#if defined(sun)
7705 id = (dtrace_id_t)(uintptr_t)vmem_alloc(dtrace_arena, 1,
7706 VM_BESTFIT | VM_SLEEP);
7707#else
7708 id = alloc_unr(dtrace_arena);
7709#endif
7710 probe = kmem_zalloc(sizeof (dtrace_probe_t), KM_SLEEP);
7711
7712 probe->dtpr_id = id;
7713 probe->dtpr_gen = dtrace_probegen++;
7714 probe->dtpr_mod = dtrace_strdup(mod);
7715 probe->dtpr_func = dtrace_strdup(func);
7716 probe->dtpr_name = dtrace_strdup(name);
7717 probe->dtpr_arg = arg;
7718 probe->dtpr_aframes = aframes;
7719 probe->dtpr_provider = provider;
7720
7721 dtrace_hash_add(dtrace_bymod, probe);
7722 dtrace_hash_add(dtrace_byfunc, probe);
7723 dtrace_hash_add(dtrace_byname, probe);
7724
7725 if (id - 1 >= dtrace_nprobes) {
7726 size_t osize = dtrace_nprobes * sizeof (dtrace_probe_t *);
7727 size_t nsize = osize << 1;
7728
7729 if (nsize == 0) {
7730 ASSERT(osize == 0);
7731 ASSERT(dtrace_probes == NULL);
7732 nsize = sizeof (dtrace_probe_t *);
7733 }
7734
7735 probes = kmem_zalloc(nsize, KM_SLEEP);
7736
7737 if (dtrace_probes == NULL) {
7738 ASSERT(osize == 0);
7739 dtrace_probes = probes;
7740 dtrace_nprobes = 1;
7741 } else {
7742 dtrace_probe_t **oprobes = dtrace_probes;
7743
7744 bcopy(oprobes, probes, osize);
7745 dtrace_membar_producer();
7746 dtrace_probes = probes;
7747
7748 dtrace_sync();
7749
7750 /*
7751 * All CPUs are now seeing the new probes array; we can
7752 * safely free the old array.
7753 */
7754 kmem_free(oprobes, osize);
7755 dtrace_nprobes <<= 1;
7756 }
7757
7758 ASSERT(id - 1 < dtrace_nprobes);
7759 }
7760
7761 ASSERT(dtrace_probes[id - 1] == NULL);
7762 dtrace_probes[id - 1] = probe;
7763
7764 if (provider != dtrace_provider)
7765 mutex_exit(&dtrace_lock);
7766
7767 return (id);
7768}
7769
7770static dtrace_probe_t *
7771dtrace_probe_lookup_id(dtrace_id_t id)
7772{
7773 ASSERT(MUTEX_HELD(&dtrace_lock));
7774
7775 if (id == 0 || id > dtrace_nprobes)
7776 return (NULL);
7777
7778 return (dtrace_probes[id - 1]);
7779}
7780
7781static int
7782dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg)
7783{
7784 *((dtrace_id_t *)arg) = probe->dtpr_id;
7785
7786 return (DTRACE_MATCH_DONE);
7787}
7788
7789/*
7790 * Look up a probe based on provider and one or more of module name, function
7791 * name and probe name.
7792 */
7793dtrace_id_t
7794dtrace_probe_lookup(dtrace_provider_id_t prid, char *mod,
7795 char *func, char *name)
7796{
7797 dtrace_probekey_t pkey;
7798 dtrace_id_t id;
7799 int match;
7800
7801 pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name;
7802 pkey.dtpk_pmatch = &dtrace_match_string;
7803 pkey.dtpk_mod = mod;
7804 pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul;
7805 pkey.dtpk_func = func;
7806 pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul;
7807 pkey.dtpk_name = name;
7808 pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul;
7809 pkey.dtpk_id = DTRACE_IDNONE;
7810
7811 mutex_enter(&dtrace_lock);
7812 match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, 0,
7813 dtrace_probe_lookup_match, &id);
7814 mutex_exit(&dtrace_lock);
7815
7816 ASSERT(match == 1 || match == 0);
7817 return (match ? id : 0);
7818}
7819
7820/*
7821 * Returns the probe argument associated with the specified probe.
7822 */
7823void *
7824dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid)
7825{
7826 dtrace_probe_t *probe;
7827 void *rval = NULL;
7828
7829 mutex_enter(&dtrace_lock);
7830
7831 if ((probe = dtrace_probe_lookup_id(pid)) != NULL &&
7832 probe->dtpr_provider == (dtrace_provider_t *)id)
7833 rval = probe->dtpr_arg;
7834
7835 mutex_exit(&dtrace_lock);
7836
7837 return (rval);
7838}
7839
7840/*
7841 * Copy a probe into a probe description.
7842 */
7843static void
7844dtrace_probe_description(const dtrace_probe_t *prp, dtrace_probedesc_t *pdp)
7845{
7846 bzero(pdp, sizeof (dtrace_probedesc_t));
7847 pdp->dtpd_id = prp->dtpr_id;
7848
7849 (void) strncpy(pdp->dtpd_provider,
7850 prp->dtpr_provider->dtpv_name, DTRACE_PROVNAMELEN - 1);
7851
7852 (void) strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1);
7853 (void) strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1);
7854 (void) strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1);
7855}
7856
7857#if !defined(sun)
7858static int
7859dtrace_probe_provide_cb(linker_file_t lf, void *arg)
7860{
7861 dtrace_provider_t *prv = (dtrace_provider_t *) arg;
7862
7863 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, lf);
7864
7865 return(0);
7866}
7867#endif
7868
7869
7870/*
7871 * Called to indicate that a probe -- or probes -- should be provided by a
7872 * specfied provider. If the specified description is NULL, the provider will
7873 * be told to provide all of its probes. (This is done whenever a new
7874 * consumer comes along, or whenever a retained enabling is to be matched.) If
7875 * the specified description is non-NULL, the provider is given the
7876 * opportunity to dynamically provide the specified probe, allowing providers
7877 * to support the creation of probes on-the-fly. (So-called _autocreated_
7878 * probes.) If the provider is NULL, the operations will be applied to all
7879 * providers; if the provider is non-NULL the operations will only be applied
7880 * to the specified provider. The dtrace_provider_lock must be held, and the
7881 * dtrace_lock must _not_ be held -- the provider's dtps_provide() operation
7882 * will need to grab the dtrace_lock when it reenters the framework through
7883 * dtrace_probe_lookup(), dtrace_probe_create(), etc.
7884 */
7885static void
7886dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv)
7887{
7888#if defined(sun)
7889 modctl_t *ctl;
7890#endif
7891 int all = 0;
7892
7893 ASSERT(MUTEX_HELD(&dtrace_provider_lock));
7894
7895 if (prv == NULL) {
7896 all = 1;
7897 prv = dtrace_provider;
7898 }
7899
7900 do {
7901 /*
7902 * First, call the blanket provide operation.
7903 */
7904 prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc);
7905
7906 /*
7907 * Now call the per-module provide operation. We will grab
7908 * mod_lock to prevent the list from being modified. Note
7909 * that this also prevents the mod_busy bits from changing.
7910 * (mod_busy can only be changed with mod_lock held.)
7911 */
7912 mutex_enter(&mod_lock);
7913
7914#if defined(sun)
7915 ctl = &modules;
7916 do {
7917 if (ctl->mod_busy || ctl->mod_mp == NULL)
7918 continue;
7919
7920 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl);
7921
7922 } while ((ctl = ctl->mod_next) != &modules);
7923#else
7924 (void) linker_file_foreach(dtrace_probe_provide_cb, prv);
7925#endif
7926
7927 mutex_exit(&mod_lock);
7928 } while (all && (prv = prv->dtpv_next) != NULL);
7929}
7930
7931#if defined(sun)
7932/*
7933 * Iterate over each probe, and call the Framework-to-Provider API function
7934 * denoted by offs.
7935 */
7936static void
7937dtrace_probe_foreach(uintptr_t offs)
7938{
7939 dtrace_provider_t *prov;
7940 void (*func)(void *, dtrace_id_t, void *);
7941 dtrace_probe_t *probe;
7942 dtrace_icookie_t cookie;
7943 int i;
7944
7945 /*
7946 * We disable interrupts to walk through the probe array. This is
7947 * safe -- the dtrace_sync() in dtrace_unregister() assures that we
7948 * won't see stale data.
7949 */
7950 cookie = dtrace_interrupt_disable();
7951
7952 for (i = 0; i < dtrace_nprobes; i++) {
7953 if ((probe = dtrace_probes[i]) == NULL)
7954 continue;
7955
7956 if (probe->dtpr_ecb == NULL) {
7957 /*
7958 * This probe isn't enabled -- don't call the function.
7959 */
7960 continue;
7961 }
7962
7963 prov = probe->dtpr_provider;
7964 func = *((void(**)(void *, dtrace_id_t, void *))
7965 ((uintptr_t)&prov->dtpv_pops + offs));
7966
7967 func(prov->dtpv_arg, i + 1, probe->dtpr_arg);
7968 }
7969
7970 dtrace_interrupt_enable(cookie);
7971}
7972#endif
7973
7974static int
7975dtrace_probe_enable(dtrace_probedesc_t *desc, dtrace_enabling_t *enab)
7976{
7977 dtrace_probekey_t pkey;
7978 uint32_t priv;
7979 uid_t uid;
7980 zoneid_t zoneid;
7981
7982 ASSERT(MUTEX_HELD(&dtrace_lock));
7983 dtrace_ecb_create_cache = NULL;
7984
7985 if (desc == NULL) {
7986 /*
7987 * If we're passed a NULL description, we're being asked to
7988 * create an ECB with a NULL probe.
7989 */
7990 (void) dtrace_ecb_create_enable(NULL, enab);
7991 return (0);
7992 }
7993
7994 dtrace_probekey(desc, &pkey);
7995 dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred,
7996 &priv, &uid, &zoneid);
7997
7998 return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable,
7999 enab));
8000}
8001
8002/*
8003 * DTrace Helper Provider Functions
8004 */
8005static void
8006dtrace_dofattr2attr(dtrace_attribute_t *attr, const dof_attr_t dofattr)
8007{
8008 attr->dtat_name = DOF_ATTR_NAME(dofattr);
8009 attr->dtat_data = DOF_ATTR_DATA(dofattr);
8010 attr->dtat_class = DOF_ATTR_CLASS(dofattr);
8011}
8012
8013static void
8014dtrace_dofprov2hprov(dtrace_helper_provdesc_t *hprov,
8015 const dof_provider_t *dofprov, char *strtab)
8016{
8017 hprov->dthpv_provname = strtab + dofprov->dofpv_name;
8018 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_provider,
8019 dofprov->dofpv_provattr);
8020 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_mod,
8021 dofprov->dofpv_modattr);
8022 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_func,
8023 dofprov->dofpv_funcattr);
8024 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_name,
8025 dofprov->dofpv_nameattr);
8026 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_args,
8027 dofprov->dofpv_argsattr);
8028}
8029
8030static void
8031dtrace_helper_provide_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid)
8032{
8033 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8034 dof_hdr_t *dof = (dof_hdr_t *)daddr;
8035 dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec;
8036 dof_provider_t *provider;
8037 dof_probe_t *probe;
8038 uint32_t *off, *enoff;
8039 uint8_t *arg;
8040 char *strtab;
8041 uint_t i, nprobes;
8042 dtrace_helper_provdesc_t dhpv;
8043 dtrace_helper_probedesc_t dhpb;
8044 dtrace_meta_t *meta = dtrace_meta_pid;
8045 dtrace_mops_t *mops = &meta->dtm_mops;
8046 void *parg;
8047
8048 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
8049 str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8050 provider->dofpv_strtab * dof->dofh_secsize);
8051 prb_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8052 provider->dofpv_probes * dof->dofh_secsize);
8053 arg_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8054 provider->dofpv_prargs * dof->dofh_secsize);
8055 off_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8056 provider->dofpv_proffs * dof->dofh_secsize);
8057
8058 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
8059 off = (uint32_t *)(uintptr_t)(daddr + off_sec->dofs_offset);
8060 arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset);
8061 enoff = NULL;
8062
8063 /*
8064 * See dtrace_helper_provider_validate().
8065 */
8066 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
8067 provider->dofpv_prenoffs != DOF_SECT_NONE) {
8068 enoff_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8069 provider->dofpv_prenoffs * dof->dofh_secsize);
8070 enoff = (uint32_t *)(uintptr_t)(daddr + enoff_sec->dofs_offset);
8071 }
8072
8073 nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize;
8074
8075 /*
8076 * Create the provider.
8077 */
8078 dtrace_dofprov2hprov(&dhpv, provider, strtab);
8079
8080 if ((parg = mops->dtms_provide_pid(meta->dtm_arg, &dhpv, pid)) == NULL)
8081 return;
8082
8083 meta->dtm_count++;
8084
8085 /*
8086 * Create the probes.
8087 */
8088 for (i = 0; i < nprobes; i++) {
8089 probe = (dof_probe_t *)(uintptr_t)(daddr +
8090 prb_sec->dofs_offset + i * prb_sec->dofs_entsize);
8091
8092 dhpb.dthpb_mod = dhp->dofhp_mod;
8093 dhpb.dthpb_func = strtab + probe->dofpr_func;
8094 dhpb.dthpb_name = strtab + probe->dofpr_name;
8095 dhpb.dthpb_base = probe->dofpr_addr;
8096 dhpb.dthpb_offs = off + probe->dofpr_offidx;
8097 dhpb.dthpb_noffs = probe->dofpr_noffs;
8098 if (enoff != NULL) {
8099 dhpb.dthpb_enoffs = enoff + probe->dofpr_enoffidx;
8100 dhpb.dthpb_nenoffs = probe->dofpr_nenoffs;
8101 } else {
8102 dhpb.dthpb_enoffs = NULL;
8103 dhpb.dthpb_nenoffs = 0;
8104 }
8105 dhpb.dthpb_args = arg + probe->dofpr_argidx;
8106 dhpb.dthpb_nargc = probe->dofpr_nargc;
8107 dhpb.dthpb_xargc = probe->dofpr_xargc;
8108 dhpb.dthpb_ntypes = strtab + probe->dofpr_nargv;
8109 dhpb.dthpb_xtypes = strtab + probe->dofpr_xargv;
8110
8111 mops->dtms_create_probe(meta->dtm_arg, parg, &dhpb);
8112 }
8113}
8114
8115static void
8116dtrace_helper_provide(dof_helper_t *dhp, pid_t pid)
8117{
8118 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8119 dof_hdr_t *dof = (dof_hdr_t *)daddr;
8120 int i;
8121
8122 ASSERT(MUTEX_HELD(&dtrace_meta_lock));
8123
8124 for (i = 0; i < dof->dofh_secnum; i++) {
8125 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
8126 dof->dofh_secoff + i * dof->dofh_secsize);
8127
8128 if (sec->dofs_type != DOF_SECT_PROVIDER)
8129 continue;
8130
8131 dtrace_helper_provide_one(dhp, sec, pid);
8132 }
8133
8134 /*
8135 * We may have just created probes, so we must now rematch against
8136 * any retained enablings. Note that this call will acquire both
8137 * cpu_lock and dtrace_lock; the fact that we are holding
8138 * dtrace_meta_lock now is what defines the ordering with respect to
8139 * these three locks.
8140 */
8141 dtrace_enabling_matchall();
8142}
8143
8144#if defined(sun)
8145static void
8146dtrace_helper_provider_remove_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid)
8147{
8148 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8149 dof_hdr_t *dof = (dof_hdr_t *)daddr;
8150 dof_sec_t *str_sec;
8151 dof_provider_t *provider;
8152 char *strtab;
8153 dtrace_helper_provdesc_t dhpv;
8154 dtrace_meta_t *meta = dtrace_meta_pid;
8155 dtrace_mops_t *mops = &meta->dtm_mops;
8156
8157 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
8158 str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff +
8159 provider->dofpv_strtab * dof->dofh_secsize);
8160
8161 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
8162
8163 /*
8164 * Create the provider.
8165 */
8166 dtrace_dofprov2hprov(&dhpv, provider, strtab);
8167
8168 mops->dtms_remove_pid(meta->dtm_arg, &dhpv, pid);
8169
8170 meta->dtm_count--;
8171}
8172
8173static void
8174dtrace_helper_provider_remove(dof_helper_t *dhp, pid_t pid)
8175{
8176 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof;
8177 dof_hdr_t *dof = (dof_hdr_t *)daddr;
8178 int i;
8179
8180 ASSERT(MUTEX_HELD(&dtrace_meta_lock));
8181
8182 for (i = 0; i < dof->dofh_secnum; i++) {
8183 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
8184 dof->dofh_secoff + i * dof->dofh_secsize);
8185
8186 if (sec->dofs_type != DOF_SECT_PROVIDER)
8187 continue;
8188
8189 dtrace_helper_provider_remove_one(dhp, sec, pid);
8190 }
8191}
8192#endif
8193
8194/*
8195 * DTrace Meta Provider-to-Framework API Functions
8196 *
8197 * These functions implement the Meta Provider-to-Framework API, as described
8198 * in <sys/dtrace.h>.
8199 */
8200int
8201dtrace_meta_register(const char *name, const dtrace_mops_t *mops, void *arg,
8202 dtrace_meta_provider_id_t *idp)
8203{
8204 dtrace_meta_t *meta;
8205 dtrace_helpers_t *help, *next;
8206 int i;
8207
8208 *idp = DTRACE_METAPROVNONE;
8209
8210 /*
8211 * We strictly don't need the name, but we hold onto it for
8212 * debuggability. All hail error queues!
8213 */
8214 if (name == NULL) {
8215 cmn_err(CE_WARN, "failed to register meta-provider: "
8216 "invalid name");
8217 return (EINVAL);
8218 }
8219
8220 if (mops == NULL ||
8221 mops->dtms_create_probe == NULL ||
8222 mops->dtms_provide_pid == NULL ||
8223 mops->dtms_remove_pid == NULL) {
8224 cmn_err(CE_WARN, "failed to register meta-register %s: "
8225 "invalid ops", name);
8226 return (EINVAL);
8227 }
8228
8229 meta = kmem_zalloc(sizeof (dtrace_meta_t), KM_SLEEP);
8230 meta->dtm_mops = *mops;
8231 meta->dtm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP);
8232 (void) strcpy(meta->dtm_name, name);
8233 meta->dtm_arg = arg;
8234
8235 mutex_enter(&dtrace_meta_lock);
8236 mutex_enter(&dtrace_lock);
8237
8238 if (dtrace_meta_pid != NULL) {
8239 mutex_exit(&dtrace_lock);
8240 mutex_exit(&dtrace_meta_lock);
8241 cmn_err(CE_WARN, "failed to register meta-register %s: "
8242 "user-land meta-provider exists", name);
8243 kmem_free(meta->dtm_name, strlen(meta->dtm_name) + 1);
8244 kmem_free(meta, sizeof (dtrace_meta_t));
8245 return (EINVAL);
8246 }
8247
8248 dtrace_meta_pid = meta;
8249 *idp = (dtrace_meta_provider_id_t)meta;
8250
8251 /*
8252 * If there are providers and probes ready to go, pass them
8253 * off to the new meta provider now.
8254 */
8255
8256 help = dtrace_deferred_pid;
8257 dtrace_deferred_pid = NULL;
8258
8259 mutex_exit(&dtrace_lock);
8260
8261 while (help != NULL) {
8262 for (i = 0; i < help->dthps_nprovs; i++) {
8263 dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov,
8264 help->dthps_pid);
8265 }
8266
8267 next = help->dthps_next;
8268 help->dthps_next = NULL;
8269 help->dthps_prev = NULL;
8270 help->dthps_deferred = 0;
8271 help = next;
8272 }
8273
8274 mutex_exit(&dtrace_meta_lock);
8275
8276 return (0);
8277}
8278
8279int
8280dtrace_meta_unregister(dtrace_meta_provider_id_t id)
8281{
8282 dtrace_meta_t **pp, *old = (dtrace_meta_t *)id;
8283
8284 mutex_enter(&dtrace_meta_lock);
8285 mutex_enter(&dtrace_lock);
8286
8287 if (old == dtrace_meta_pid) {
8288 pp = &dtrace_meta_pid;
8289 } else {
8290 panic("attempt to unregister non-existent "
8291 "dtrace meta-provider %p\n", (void *)old);
8292 }
8293
8294 if (old->dtm_count != 0) {
8295 mutex_exit(&dtrace_lock);
8296 mutex_exit(&dtrace_meta_lock);
8297 return (EBUSY);
8298 }
8299
8300 *pp = NULL;
8301
8302 mutex_exit(&dtrace_lock);
8303 mutex_exit(&dtrace_meta_lock);
8304
8305 kmem_free(old->dtm_name, strlen(old->dtm_name) + 1);
8306 kmem_free(old, sizeof (dtrace_meta_t));
8307
8308 return (0);
8309}
8310
8311
8312/*
8313 * DTrace DIF Object Functions
8314 */
8315static int
8316dtrace_difo_err(uint_t pc, const char *format, ...)
8317{
8318 if (dtrace_err_verbose) {
8319 va_list alist;
8320
8321 (void) uprintf("dtrace DIF object error: [%u]: ", pc);
8322 va_start(alist, format);
8323 (void) vuprintf(format, alist);
8324 va_end(alist);
8325 }
8326
8327#ifdef DTRACE_ERRDEBUG
8328 dtrace_errdebug(format);
8329#endif
8330 return (1);
8331}
8332
8333/*
8334 * Validate a DTrace DIF object by checking the IR instructions. The following
8335 * rules are currently enforced by dtrace_difo_validate():
8336 *
8337 * 1. Each instruction must have a valid opcode
8338 * 2. Each register, string, variable, or subroutine reference must be valid
8339 * 3. No instruction can modify register %r0 (must be zero)
8340 * 4. All instruction reserved bits must be set to zero
8341 * 5. The last instruction must be a "ret" instruction
8342 * 6. All branch targets must reference a valid instruction _after_ the branch
8343 */
8344static int
8345dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, uint_t nregs,
8346 cred_t *cr)
8347{
8348 int err = 0, i;
8349 int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err;
8350 int kcheckload;
8351 uint_t pc;
8352
8353 kcheckload = cr == NULL ||
8354 (vstate->dtvs_state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) == 0;
8355
8356 dp->dtdo_destructive = 0;
8357
8358 for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) {
8359 dif_instr_t instr = dp->dtdo_buf[pc];
8360
8361 uint_t r1 = DIF_INSTR_R1(instr);
8362 uint_t r2 = DIF_INSTR_R2(instr);
8363 uint_t rd = DIF_INSTR_RD(instr);
8364 uint_t rs = DIF_INSTR_RS(instr);
8365 uint_t label = DIF_INSTR_LABEL(instr);
8366 uint_t v = DIF_INSTR_VAR(instr);
8367 uint_t subr = DIF_INSTR_SUBR(instr);
8368 uint_t type = DIF_INSTR_TYPE(instr);
8369 uint_t op = DIF_INSTR_OP(instr);
8370
8371 switch (op) {
8372 case DIF_OP_OR:
8373 case DIF_OP_XOR:
8374 case DIF_OP_AND:
8375 case DIF_OP_SLL:
8376 case DIF_OP_SRL:
8377 case DIF_OP_SRA:
8378 case DIF_OP_SUB:
8379 case DIF_OP_ADD:
8380 case DIF_OP_MUL:
8381 case DIF_OP_SDIV:
8382 case DIF_OP_UDIV:
8383 case DIF_OP_SREM:
8384 case DIF_OP_UREM:
8385 case DIF_OP_COPYS:
8386 if (r1 >= nregs)
8387 err += efunc(pc, "invalid register %u\n", r1);
8388 if (r2 >= nregs)
8389 err += efunc(pc, "invalid register %u\n", r2);
8390 if (rd >= nregs)
8391 err += efunc(pc, "invalid register %u\n", rd);
8392 if (rd == 0)
8393 err += efunc(pc, "cannot write to %r0\n");
8394 break;
8395 case DIF_OP_NOT:
8396 case DIF_OP_MOV:
8397 case DIF_OP_ALLOCS:
8398 if (r1 >= nregs)
8399 err += efunc(pc, "invalid register %u\n", r1);
8400 if (r2 != 0)
8401 err += efunc(pc, "non-zero reserved bits\n");
8402 if (rd >= nregs)
8403 err += efunc(pc, "invalid register %u\n", rd);
8404 if (rd == 0)
8405 err += efunc(pc, "cannot write to %r0\n");
8406 break;
8407 case DIF_OP_LDSB:
8408 case DIF_OP_LDSH:
8409 case DIF_OP_LDSW:
8410 case DIF_OP_LDUB:
8411 case DIF_OP_LDUH:
8412 case DIF_OP_LDUW:
8413 case DIF_OP_LDX:
8414 if (r1 >= nregs)
8415 err += efunc(pc, "invalid register %u\n", r1);
8416 if (r2 != 0)
8417 err += efunc(pc, "non-zero reserved bits\n");
8418 if (rd >= nregs)
8419 err += efunc(pc, "invalid register %u\n", rd);
8420 if (rd == 0)
8421 err += efunc(pc, "cannot write to %r0\n");
8422 if (kcheckload)
8423 dp->dtdo_buf[pc] = DIF_INSTR_LOAD(op +
8424 DIF_OP_RLDSB - DIF_OP_LDSB, r1, rd);
8425 break;
8426 case DIF_OP_RLDSB:
8427 case DIF_OP_RLDSH:
8428 case DIF_OP_RLDSW:
8429 case DIF_OP_RLDUB:
8430 case DIF_OP_RLDUH:
8431 case DIF_OP_RLDUW:
8432 case DIF_OP_RLDX:
8433 if (r1 >= nregs)
8434 err += efunc(pc, "invalid register %u\n", r1);
8435 if (r2 != 0)
8436 err += efunc(pc, "non-zero reserved bits\n");
8437 if (rd >= nregs)
8438 err += efunc(pc, "invalid register %u\n", rd);
8439 if (rd == 0)
8440 err += efunc(pc, "cannot write to %r0\n");
8441 break;
8442 case DIF_OP_ULDSB:
8443 case DIF_OP_ULDSH:
8444 case DIF_OP_ULDSW:
8445 case DIF_OP_ULDUB:
8446 case DIF_OP_ULDUH:
8447 case DIF_OP_ULDUW:
8448 case DIF_OP_ULDX:
8449 if (r1 >= nregs)
8450 err += efunc(pc, "invalid register %u\n", r1);
8451 if (r2 != 0)
8452 err += efunc(pc, "non-zero reserved bits\n");
8453 if (rd >= nregs)
8454 err += efunc(pc, "invalid register %u\n", rd);
8455 if (rd == 0)
8456 err += efunc(pc, "cannot write to %r0\n");
8457 break;
8458 case DIF_OP_STB:
8459 case DIF_OP_STH:
8460 case DIF_OP_STW:
8461 case DIF_OP_STX:
8462 if (r1 >= nregs)
8463 err += efunc(pc, "invalid register %u\n", r1);
8464 if (r2 != 0)
8465 err += efunc(pc, "non-zero reserved bits\n");
8466 if (rd >= nregs)
8467 err += efunc(pc, "invalid register %u\n", rd);
8468 if (rd == 0)
8469 err += efunc(pc, "cannot write to 0 address\n");
8470 break;
8471 case DIF_OP_CMP:
8472 case DIF_OP_SCMP:
8473 if (r1 >= nregs)
8474 err += efunc(pc, "invalid register %u\n", r1);
8475 if (r2 >= nregs)
8476 err += efunc(pc, "invalid register %u\n", r2);
8477 if (rd != 0)
8478 err += efunc(pc, "non-zero reserved bits\n");
8479 break;
8480 case DIF_OP_TST:
8481 if (r1 >= nregs)
8482 err += efunc(pc, "invalid register %u\n", r1);
8483 if (r2 != 0 || rd != 0)
8484 err += efunc(pc, "non-zero reserved bits\n");
8485 break;
8486 case DIF_OP_BA:
8487 case DIF_OP_BE:
8488 case DIF_OP_BNE:
8489 case DIF_OP_BG:
8490 case DIF_OP_BGU:
8491 case DIF_OP_BGE:
8492 case DIF_OP_BGEU:
8493 case DIF_OP_BL:
8494 case DIF_OP_BLU:
8495 case DIF_OP_BLE:
8496 case DIF_OP_BLEU:
8497 if (label >= dp->dtdo_len) {
8498 err += efunc(pc, "invalid branch target %u\n",
8499 label);
8500 }
8501 if (label <= pc) {
8502 err += efunc(pc, "backward branch to %u\n",
8503 label);
8504 }
8505 break;
8506 case DIF_OP_RET:
8507 if (r1 != 0 || r2 != 0)
8508 err += efunc(pc, "non-zero reserved bits\n");
8509 if (rd >= nregs)
8510 err += efunc(pc, "invalid register %u\n", rd);
8511 break;
8512 case DIF_OP_NOP:
8513 case DIF_OP_POPTS:
8514 case DIF_OP_FLUSHTS:
8515 if (r1 != 0 || r2 != 0 || rd != 0)
8516 err += efunc(pc, "non-zero reserved bits\n");
8517 break;
8518 case DIF_OP_SETX:
8519 if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) {
8520 err += efunc(pc, "invalid integer ref %u\n",
8521 DIF_INSTR_INTEGER(instr));
8522 }
8523 if (rd >= nregs)
8524 err += efunc(pc, "invalid register %u\n", rd);
8525 if (rd == 0)
8526 err += efunc(pc, "cannot write to %r0\n");
8527 break;
8528 case DIF_OP_SETS:
8529 if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) {
8530 err += efunc(pc, "invalid string ref %u\n",
8531 DIF_INSTR_STRING(instr));
8532 }
8533 if (rd >= nregs)
8534 err += efunc(pc, "invalid register %u\n", rd);
8535 if (rd == 0)
8536 err += efunc(pc, "cannot write to %r0\n");
8537 break;
8538 case DIF_OP_LDGA:
8539 case DIF_OP_LDTA:
8540 if (r1 > DIF_VAR_ARRAY_MAX)
8541 err += efunc(pc, "invalid array %u\n", r1);
8542 if (r2 >= nregs)
8543 err += efunc(pc, "invalid register %u\n", r2);
8544 if (rd >= nregs)
8545 err += efunc(pc, "invalid register %u\n", rd);
8546 if (rd == 0)
8547 err += efunc(pc, "cannot write to %r0\n");
8548 break;
8549 case DIF_OP_LDGS:
8550 case DIF_OP_LDTS:
8551 case DIF_OP_LDLS:
8552 case DIF_OP_LDGAA:
8553 case DIF_OP_LDTAA:
8554 if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX)
8555 err += efunc(pc, "invalid variable %u\n", v);
8556 if (rd >= nregs)
8557 err += efunc(pc, "invalid register %u\n", rd);
8558 if (rd == 0)
8559 err += efunc(pc, "cannot write to %r0\n");
8560 break;
8561 case DIF_OP_STGS:
8562 case DIF_OP_STTS:
8563 case DIF_OP_STLS:
8564 case DIF_OP_STGAA:
8565 case DIF_OP_STTAA:
8566 if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX)
8567 err += efunc(pc, "invalid variable %u\n", v);
8568 if (rs >= nregs)
8569 err += efunc(pc, "invalid register %u\n", rd);
8570 break;
8571 case DIF_OP_CALL:
8572 if (subr > DIF_SUBR_MAX)
8573 err += efunc(pc, "invalid subr %u\n", subr);
8574 if (rd >= nregs)
8575 err += efunc(pc, "invalid register %u\n", rd);
8576 if (rd == 0)
8577 err += efunc(pc, "cannot write to %r0\n");
8578
8579 if (subr == DIF_SUBR_COPYOUT ||
8580 subr == DIF_SUBR_COPYOUTSTR) {
8581 dp->dtdo_destructive = 1;
8582 }
8583 break;
8584 case DIF_OP_PUSHTR:
8585 if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF)
8586 err += efunc(pc, "invalid ref type %u\n", type);
8587 if (r2 >= nregs)
8588 err += efunc(pc, "invalid register %u\n", r2);
8589 if (rs >= nregs)
8590 err += efunc(pc, "invalid register %u\n", rs);
8591 break;
8592 case DIF_OP_PUSHTV:
8593 if (type != DIF_TYPE_CTF)
8594 err += efunc(pc, "invalid val type %u\n", type);
8595 if (r2 >= nregs)
8596 err += efunc(pc, "invalid register %u\n", r2);
8597 if (rs >= nregs)
8598 err += efunc(pc, "invalid register %u\n", rs);
8599 break;
8600 default:
8601 err += efunc(pc, "invalid opcode %u\n",
8602 DIF_INSTR_OP(instr));
8603 }
8604 }
8605
8606 if (dp->dtdo_len != 0 &&
8607 DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) {
8608 err += efunc(dp->dtdo_len - 1,
8609 "expected 'ret' as last DIF instruction\n");
8610 }
8611
8612 if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) {
8613 /*
8614 * If we're not returning by reference, the size must be either
8615 * 0 or the size of one of the base types.
8616 */
8617 switch (dp->dtdo_rtype.dtdt_size) {
8618 case 0:
8619 case sizeof (uint8_t):
8620 case sizeof (uint16_t):
8621 case sizeof (uint32_t):
8622 case sizeof (uint64_t):
8623 break;
8624
8625 default:
8626 err += efunc(dp->dtdo_len - 1, "bad return size");
8627 }
8628 }
8629
8630 for (i = 0; i < dp->dtdo_varlen && err == 0; i++) {
8631 dtrace_difv_t *v = &dp->dtdo_vartab[i], *existing = NULL;
8632 dtrace_diftype_t *vt, *et;
8633 uint_t id, ndx;
8634
8635 if (v->dtdv_scope != DIFV_SCOPE_GLOBAL &&
8636 v->dtdv_scope != DIFV_SCOPE_THREAD &&
8637 v->dtdv_scope != DIFV_SCOPE_LOCAL) {
8638 err += efunc(i, "unrecognized variable scope %d\n",
8639 v->dtdv_scope);
8640 break;
8641 }
8642
8643 if (v->dtdv_kind != DIFV_KIND_ARRAY &&
8644 v->dtdv_kind != DIFV_KIND_SCALAR) {
8645 err += efunc(i, "unrecognized variable type %d\n",
8646 v->dtdv_kind);
8647 break;
8648 }
8649
8650 if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) {
8651 err += efunc(i, "%d exceeds variable id limit\n", id);
8652 break;
8653 }
8654
8655 if (id < DIF_VAR_OTHER_UBASE)
8656 continue;
8657
8658 /*
8659 * For user-defined variables, we need to check that this
8660 * definition is identical to any previous definition that we
8661 * encountered.
8662 */
8663 ndx = id - DIF_VAR_OTHER_UBASE;
8664
8665 switch (v->dtdv_scope) {
8666 case DIFV_SCOPE_GLOBAL:
8667 if (ndx < vstate->dtvs_nglobals) {
8668 dtrace_statvar_t *svar;
8669
8670 if ((svar = vstate->dtvs_globals[ndx]) != NULL)
8671 existing = &svar->dtsv_var;
8672 }
8673
8674 break;
8675
8676 case DIFV_SCOPE_THREAD:
8677 if (ndx < vstate->dtvs_ntlocals)
8678 existing = &vstate->dtvs_tlocals[ndx];
8679 break;
8680
8681 case DIFV_SCOPE_LOCAL:
8682 if (ndx < vstate->dtvs_nlocals) {
8683 dtrace_statvar_t *svar;
8684
8685 if ((svar = vstate->dtvs_locals[ndx]) != NULL)
8686 existing = &svar->dtsv_var;
8687 }
8688
8689 break;
8690 }
8691
8692 vt = &v->dtdv_type;
8693
8694 if (vt->dtdt_flags & DIF_TF_BYREF) {
8695 if (vt->dtdt_size == 0) {
8696 err += efunc(i, "zero-sized variable\n");
8697 break;
8698 }
8699
8700 if (v->dtdv_scope == DIFV_SCOPE_GLOBAL &&
8701 vt->dtdt_size > dtrace_global_maxsize) {
8702 err += efunc(i, "oversized by-ref global\n");
8703 break;
8704 }
8705 }
8706
8707 if (existing == NULL || existing->dtdv_id == 0)
8708 continue;
8709
8710 ASSERT(existing->dtdv_id == v->dtdv_id);
8711 ASSERT(existing->dtdv_scope == v->dtdv_scope);
8712
8713 if (existing->dtdv_kind != v->dtdv_kind)
8714 err += efunc(i, "%d changed variable kind\n", id);
8715
8716 et = &existing->dtdv_type;
8717
8718 if (vt->dtdt_flags != et->dtdt_flags) {
8719 err += efunc(i, "%d changed variable type flags\n", id);
8720 break;
8721 }
8722
8723 if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) {
8724 err += efunc(i, "%d changed variable type size\n", id);
8725 break;
8726 }
8727 }
8728
8729 return (err);
8730}
8731
8732#if defined(sun)
8733/*
8734 * Validate a DTrace DIF object that it is to be used as a helper. Helpers
8735 * are much more constrained than normal DIFOs. Specifically, they may
8736 * not:
8737 *
8738 * 1. Make calls to subroutines other than copyin(), copyinstr() or
8739 * miscellaneous string routines
8740 * 2. Access DTrace variables other than the args[] array, and the
8741 * curthread, pid, ppid, tid, execname, zonename, uid and gid variables.
8742 * 3. Have thread-local variables.
8743 * 4. Have dynamic variables.
8744 */
8745static int
8746dtrace_difo_validate_helper(dtrace_difo_t *dp)
8747{
8748 int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err;
8749 int err = 0;
8750 uint_t pc;
8751
8752 for (pc = 0; pc < dp->dtdo_len; pc++) {
8753 dif_instr_t instr = dp->dtdo_buf[pc];
8754
8755 uint_t v = DIF_INSTR_VAR(instr);
8756 uint_t subr = DIF_INSTR_SUBR(instr);
8757 uint_t op = DIF_INSTR_OP(instr);
8758
8759 switch (op) {
8760 case DIF_OP_OR:
8761 case DIF_OP_XOR:
8762 case DIF_OP_AND:
8763 case DIF_OP_SLL:
8764 case DIF_OP_SRL:
8765 case DIF_OP_SRA:
8766 case DIF_OP_SUB:
8767 case DIF_OP_ADD:
8768 case DIF_OP_MUL:
8769 case DIF_OP_SDIV:
8770 case DIF_OP_UDIV:
8771 case DIF_OP_SREM:
8772 case DIF_OP_UREM:
8773 case DIF_OP_COPYS:
8774 case DIF_OP_NOT:
8775 case DIF_OP_MOV:
8776 case DIF_OP_RLDSB:
8777 case DIF_OP_RLDSH:
8778 case DIF_OP_RLDSW:
8779 case DIF_OP_RLDUB:
8780 case DIF_OP_RLDUH:
8781 case DIF_OP_RLDUW:
8782 case DIF_OP_RLDX:
8783 case DIF_OP_ULDSB:
8784 case DIF_OP_ULDSH:
8785 case DIF_OP_ULDSW:
8786 case DIF_OP_ULDUB:
8787 case DIF_OP_ULDUH:
8788 case DIF_OP_ULDUW:
8789 case DIF_OP_ULDX:
8790 case DIF_OP_STB:
8791 case DIF_OP_STH:
8792 case DIF_OP_STW:
8793 case DIF_OP_STX:
8794 case DIF_OP_ALLOCS:
8795 case DIF_OP_CMP:
8796 case DIF_OP_SCMP:
8797 case DIF_OP_TST:
8798 case DIF_OP_BA:
8799 case DIF_OP_BE:
8800 case DIF_OP_BNE:
8801 case DIF_OP_BG:
8802 case DIF_OP_BGU:
8803 case DIF_OP_BGE:
8804 case DIF_OP_BGEU:
8805 case DIF_OP_BL:
8806 case DIF_OP_BLU:
8807 case DIF_OP_BLE:
8808 case DIF_OP_BLEU:
8809 case DIF_OP_RET:
8810 case DIF_OP_NOP:
8811 case DIF_OP_POPTS:
8812 case DIF_OP_FLUSHTS:
8813 case DIF_OP_SETX:
8814 case DIF_OP_SETS:
8815 case DIF_OP_LDGA:
8816 case DIF_OP_LDLS:
8817 case DIF_OP_STGS:
8818 case DIF_OP_STLS:
8819 case DIF_OP_PUSHTR:
8820 case DIF_OP_PUSHTV:
8821 break;
8822
8823 case DIF_OP_LDGS:
8824 if (v >= DIF_VAR_OTHER_UBASE)
8825 break;
8826
8827 if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9)
8828 break;
8829
8830 if (v == DIF_VAR_CURTHREAD || v == DIF_VAR_PID ||
8831 v == DIF_VAR_PPID || v == DIF_VAR_TID ||
8832 v == DIF_VAR_EXECARGS ||
8833 v == DIF_VAR_EXECNAME || v == DIF_VAR_ZONENAME ||
8834 v == DIF_VAR_UID || v == DIF_VAR_GID)
8835 break;
8836
8837 err += efunc(pc, "illegal variable %u\n", v);
8838 break;
8839
8840 case DIF_OP_LDTA:
8841 case DIF_OP_LDTS:
8842 case DIF_OP_LDGAA:
8843 case DIF_OP_LDTAA:
8844 err += efunc(pc, "illegal dynamic variable load\n");
8845 break;
8846
8847 case DIF_OP_STTS:
8848 case DIF_OP_STGAA:
8849 case DIF_OP_STTAA:
8850 err += efunc(pc, "illegal dynamic variable store\n");
8851 break;
8852
8853 case DIF_OP_CALL:
8854 if (subr == DIF_SUBR_ALLOCA ||
8855 subr == DIF_SUBR_BCOPY ||
8856 subr == DIF_SUBR_COPYIN ||
8857 subr == DIF_SUBR_COPYINTO ||
8858 subr == DIF_SUBR_COPYINSTR ||
8859 subr == DIF_SUBR_INDEX ||
8860 subr == DIF_SUBR_INET_NTOA ||
8861 subr == DIF_SUBR_INET_NTOA6 ||
8862 subr == DIF_SUBR_INET_NTOP ||
8863 subr == DIF_SUBR_LLTOSTR ||
8864 subr == DIF_SUBR_RINDEX ||
8865 subr == DIF_SUBR_STRCHR ||
8866 subr == DIF_SUBR_STRJOIN ||
8867 subr == DIF_SUBR_STRRCHR ||
8868 subr == DIF_SUBR_STRSTR ||
8869 subr == DIF_SUBR_HTONS ||
8870 subr == DIF_SUBR_HTONL ||
8871 subr == DIF_SUBR_HTONLL ||
8872 subr == DIF_SUBR_NTOHS ||
8873 subr == DIF_SUBR_NTOHL ||
8874 subr == DIF_SUBR_NTOHLL ||
8875 subr == DIF_SUBR_MEMREF ||
8876 subr == DIF_SUBR_TYPEREF)
8877 break;
8878
8879 err += efunc(pc, "invalid subr %u\n", subr);
8880 break;
8881
8882 default:
8883 err += efunc(pc, "invalid opcode %u\n",
8884 DIF_INSTR_OP(instr));
8885 }
8886 }
8887
8888 return (err);
8889}
8890#endif
8891
8892/*
8893 * Returns 1 if the expression in the DIF object can be cached on a per-thread
8894 * basis; 0 if not.
8895 */
8896static int
8897dtrace_difo_cacheable(dtrace_difo_t *dp)
8898{
8899 int i;
8900
8901 if (dp == NULL)
8902 return (0);
8903
8904 for (i = 0; i < dp->dtdo_varlen; i++) {
8905 dtrace_difv_t *v = &dp->dtdo_vartab[i];
8906
8907 if (v->dtdv_scope != DIFV_SCOPE_GLOBAL)
8908 continue;
8909
8910 switch (v->dtdv_id) {
8911 case DIF_VAR_CURTHREAD:
8912 case DIF_VAR_PID:
8913 case DIF_VAR_TID:
8914 case DIF_VAR_EXECARGS:
8915 case DIF_VAR_EXECNAME:
8916 case DIF_VAR_ZONENAME:
8917 break;
8918
8919 default:
8920 return (0);
8921 }
8922 }
8923
8924 /*
8925 * This DIF object may be cacheable. Now we need to look for any
8926 * array loading instructions, any memory loading instructions, or
8927 * any stores to thread-local variables.
8928 */
8929 for (i = 0; i < dp->dtdo_len; i++) {
8930 uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]);
8931
8932 if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) ||
8933 (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) ||
8934 (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) ||
8935 op == DIF_OP_LDGA || op == DIF_OP_STTS)
8936 return (0);
8937 }
8938
8939 return (1);
8940}
8941
8942static void
8943dtrace_difo_hold(dtrace_difo_t *dp)
8944{
8945 int i;
8946
8947 ASSERT(MUTEX_HELD(&dtrace_lock));
8948
8949 dp->dtdo_refcnt++;
8950 ASSERT(dp->dtdo_refcnt != 0);
8951
8952 /*
8953 * We need to check this DIF object for references to the variable
8954 * DIF_VAR_VTIMESTAMP.
8955 */
8956 for (i = 0; i < dp->dtdo_varlen; i++) {
8957 dtrace_difv_t *v = &dp->dtdo_vartab[i];
8958
8959 if (v->dtdv_id != DIF_VAR_VTIMESTAMP)
8960 continue;
8961
8962 if (dtrace_vtime_references++ == 0)
8963 dtrace_vtime_enable();
8964 }
8965}
8966
8967/*
8968 * This routine calculates the dynamic variable chunksize for a given DIF
8969 * object. The calculation is not fool-proof, and can probably be tricked by
8970 * malicious DIF -- but it works for all compiler-generated DIF. Because this
8971 * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail
8972 * if a dynamic variable size exceeds the chunksize.
8973 */
8974static void
8975dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
8976{
8977 uint64_t sval = 0;
8978 dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */
8979 const dif_instr_t *text = dp->dtdo_buf;
8980 uint_t pc, srd = 0;
8981 uint_t ttop = 0;
8982 size_t size, ksize;
8983 uint_t id, i;
8984
8985 for (pc = 0; pc < dp->dtdo_len; pc++) {
8986 dif_instr_t instr = text[pc];
8987 uint_t op = DIF_INSTR_OP(instr);
8988 uint_t rd = DIF_INSTR_RD(instr);
8989 uint_t r1 = DIF_INSTR_R1(instr);
8990 uint_t nkeys = 0;
8991 uchar_t scope = 0;
8992
8993 dtrace_key_t *key = tupregs;
8994
8995 switch (op) {
8996 case DIF_OP_SETX:
8997 sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)];
8998 srd = rd;
8999 continue;
9000
9001 case DIF_OP_STTS:
9002 key = &tupregs[DIF_DTR_NREGS];
9003 key[0].dttk_size = 0;
9004 key[1].dttk_size = 0;
9005 nkeys = 2;
9006 scope = DIFV_SCOPE_THREAD;
9007 break;
9008
9009 case DIF_OP_STGAA:
9010 case DIF_OP_STTAA:
9011 nkeys = ttop;
9012
9013 if (DIF_INSTR_OP(instr) == DIF_OP_STTAA)
9014 key[nkeys++].dttk_size = 0;
9015
9016 key[nkeys++].dttk_size = 0;
9017
9018 if (op == DIF_OP_STTAA) {
9019 scope = DIFV_SCOPE_THREAD;
9020 } else {
9021 scope = DIFV_SCOPE_GLOBAL;
9022 }
9023
9024 break;
9025
9026 case DIF_OP_PUSHTR:
9027 if (ttop == DIF_DTR_NREGS)
9028 return;
9029
9030 if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) {
9031 /*
9032 * If the register for the size of the "pushtr"
9033 * is %r0 (or the value is 0) and the type is
9034 * a string, we'll use the system-wide default
9035 * string size.
9036 */
9037 tupregs[ttop++].dttk_size =
9038 dtrace_strsize_default;
9039 } else {
9040 if (srd == 0)
9041 return;
9042
9043 tupregs[ttop++].dttk_size = sval;
9044 }
9045
9046 break;
9047
9048 case DIF_OP_PUSHTV:
9049 if (ttop == DIF_DTR_NREGS)
9050 return;
9051
9052 tupregs[ttop++].dttk_size = 0;
9053 break;
9054
9055 case DIF_OP_FLUSHTS:
9056 ttop = 0;
9057 break;
9058
9059 case DIF_OP_POPTS:
9060 if (ttop != 0)
9061 ttop--;
9062 break;
9063 }
9064
9065 sval = 0;
9066 srd = 0;
9067
9068 if (nkeys == 0)
9069 continue;
9070
9071 /*
9072 * We have a dynamic variable allocation; calculate its size.
9073 */
9074 for (ksize = 0, i = 0; i < nkeys; i++)
9075 ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t));
9076
9077 size = sizeof (dtrace_dynvar_t);
9078 size += sizeof (dtrace_key_t) * (nkeys - 1);
9079 size += ksize;
9080
9081 /*
9082 * Now we need to determine the size of the stored data.
9083 */
9084 id = DIF_INSTR_VAR(instr);
9085
9086 for (i = 0; i < dp->dtdo_varlen; i++) {
9087 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9088
9089 if (v->dtdv_id == id && v->dtdv_scope == scope) {
9090 size += v->dtdv_type.dtdt_size;
9091 break;
9092 }
9093 }
9094
9095 if (i == dp->dtdo_varlen)
9096 return;
9097
9098 /*
9099 * We have the size. If this is larger than the chunk size
9100 * for our dynamic variable state, reset the chunk size.
9101 */
9102 size = P2ROUNDUP(size, sizeof (uint64_t));
9103
9104 if (size > vstate->dtvs_dynvars.dtds_chunksize)
9105 vstate->dtvs_dynvars.dtds_chunksize = size;
9106 }
9107}
9108
9109static void
9110dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9111{
9112 int i, oldsvars, osz, nsz, otlocals, ntlocals;
9113 uint_t id;
9114
9115 ASSERT(MUTEX_HELD(&dtrace_lock));
9116 ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0);
9117
9118 for (i = 0; i < dp->dtdo_varlen; i++) {
9119 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9120 dtrace_statvar_t *svar, ***svarp = NULL;
9121 size_t dsize = 0;
9122 uint8_t scope = v->dtdv_scope;
9123 int *np = NULL;
9124
9125 if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE)
9126 continue;
9127
9128 id -= DIF_VAR_OTHER_UBASE;
9129
9130 switch (scope) {
9131 case DIFV_SCOPE_THREAD:
9132 while (id >= (otlocals = vstate->dtvs_ntlocals)) {
9133 dtrace_difv_t *tlocals;
9134
9135 if ((ntlocals = (otlocals << 1)) == 0)
9136 ntlocals = 1;
9137
9138 osz = otlocals * sizeof (dtrace_difv_t);
9139 nsz = ntlocals * sizeof (dtrace_difv_t);
9140
9141 tlocals = kmem_zalloc(nsz, KM_SLEEP);
9142
9143 if (osz != 0) {
9144 bcopy(vstate->dtvs_tlocals,
9145 tlocals, osz);
9146 kmem_free(vstate->dtvs_tlocals, osz);
9147 }
9148
9149 vstate->dtvs_tlocals = tlocals;
9150 vstate->dtvs_ntlocals = ntlocals;
9151 }
9152
9153 vstate->dtvs_tlocals[id] = *v;
9154 continue;
9155
9156 case DIFV_SCOPE_LOCAL:
9157 np = &vstate->dtvs_nlocals;
9158 svarp = &vstate->dtvs_locals;
9159
9160 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF)
9161 dsize = NCPU * (v->dtdv_type.dtdt_size +
9162 sizeof (uint64_t));
9163 else
9164 dsize = NCPU * sizeof (uint64_t);
9165
9166 break;
9167
9168 case DIFV_SCOPE_GLOBAL:
9169 np = &vstate->dtvs_nglobals;
9170 svarp = &vstate->dtvs_globals;
9171
9172 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF)
9173 dsize = v->dtdv_type.dtdt_size +
9174 sizeof (uint64_t);
9175
9176 break;
9177
9178 default:
9179 ASSERT(0);
9180 }
9181
9182 while (id >= (oldsvars = *np)) {
9183 dtrace_statvar_t **statics;
9184 int newsvars, oldsize, newsize;
9185
9186 if ((newsvars = (oldsvars << 1)) == 0)
9187 newsvars = 1;
9188
9189 oldsize = oldsvars * sizeof (dtrace_statvar_t *);
9190 newsize = newsvars * sizeof (dtrace_statvar_t *);
9191
9192 statics = kmem_zalloc(newsize, KM_SLEEP);
9193
9194 if (oldsize != 0) {
9195 bcopy(*svarp, statics, oldsize);
9196 kmem_free(*svarp, oldsize);
9197 }
9198
9199 *svarp = statics;
9200 *np = newsvars;
9201 }
9202
9203 if ((svar = (*svarp)[id]) == NULL) {
9204 svar = kmem_zalloc(sizeof (dtrace_statvar_t), KM_SLEEP);
9205 svar->dtsv_var = *v;
9206
9207 if ((svar->dtsv_size = dsize) != 0) {
9208 svar->dtsv_data = (uint64_t)(uintptr_t)
9209 kmem_zalloc(dsize, KM_SLEEP);
9210 }
9211
9212 (*svarp)[id] = svar;
9213 }
9214
9215 svar->dtsv_refcnt++;
9216 }
9217
9218 dtrace_difo_chunksize(dp, vstate);
9219 dtrace_difo_hold(dp);
9220}
9221
9222#if defined(sun)
9223static dtrace_difo_t *
9224dtrace_difo_duplicate(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9225{
9226 dtrace_difo_t *new;
9227 size_t sz;
9228
9229 ASSERT(dp->dtdo_buf != NULL);
9230 ASSERT(dp->dtdo_refcnt != 0);
9231
9232 new = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP);
9233
9234 ASSERT(dp->dtdo_buf != NULL);
9235 sz = dp->dtdo_len * sizeof (dif_instr_t);
9236 new->dtdo_buf = kmem_alloc(sz, KM_SLEEP);
9237 bcopy(dp->dtdo_buf, new->dtdo_buf, sz);
9238 new->dtdo_len = dp->dtdo_len;
9239
9240 if (dp->dtdo_strtab != NULL) {
9241 ASSERT(dp->dtdo_strlen != 0);
9242 new->dtdo_strtab = kmem_alloc(dp->dtdo_strlen, KM_SLEEP);
9243 bcopy(dp->dtdo_strtab, new->dtdo_strtab, dp->dtdo_strlen);
9244 new->dtdo_strlen = dp->dtdo_strlen;
9245 }
9246
9247 if (dp->dtdo_inttab != NULL) {
9248 ASSERT(dp->dtdo_intlen != 0);
9249 sz = dp->dtdo_intlen * sizeof (uint64_t);
9250 new->dtdo_inttab = kmem_alloc(sz, KM_SLEEP);
9251 bcopy(dp->dtdo_inttab, new->dtdo_inttab, sz);
9252 new->dtdo_intlen = dp->dtdo_intlen;
9253 }
9254
9255 if (dp->dtdo_vartab != NULL) {
9256 ASSERT(dp->dtdo_varlen != 0);
9257 sz = dp->dtdo_varlen * sizeof (dtrace_difv_t);
9258 new->dtdo_vartab = kmem_alloc(sz, KM_SLEEP);
9259 bcopy(dp->dtdo_vartab, new->dtdo_vartab, sz);
9260 new->dtdo_varlen = dp->dtdo_varlen;
9261 }
9262
9263 dtrace_difo_init(new, vstate);
9264 return (new);
9265}
9266#endif
9267
9268static void
9269dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9270{
9271 int i;
9272
9273 ASSERT(dp->dtdo_refcnt == 0);
9274
9275 for (i = 0; i < dp->dtdo_varlen; i++) {
9276 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9277 dtrace_statvar_t *svar, **svarp = NULL;
9278 uint_t id;
9279 uint8_t scope = v->dtdv_scope;
9280 int *np = NULL;
9281
9282 switch (scope) {
9283 case DIFV_SCOPE_THREAD:
9284 continue;
9285
9286 case DIFV_SCOPE_LOCAL:
9287 np = &vstate->dtvs_nlocals;
9288 svarp = vstate->dtvs_locals;
9289 break;
9290
9291 case DIFV_SCOPE_GLOBAL:
9292 np = &vstate->dtvs_nglobals;
9293 svarp = vstate->dtvs_globals;
9294 break;
9295
9296 default:
9297 ASSERT(0);
9298 }
9299
9300 if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE)
9301 continue;
9302
9303 id -= DIF_VAR_OTHER_UBASE;
9304 ASSERT(id < *np);
9305
9306 svar = svarp[id];
9307 ASSERT(svar != NULL);
9308 ASSERT(svar->dtsv_refcnt > 0);
9309
9310 if (--svar->dtsv_refcnt > 0)
9311 continue;
9312
9313 if (svar->dtsv_size != 0) {
9314 ASSERT(svar->dtsv_data != 0);
9315 kmem_free((void *)(uintptr_t)svar->dtsv_data,
9316 svar->dtsv_size);
9317 }
9318
9319 kmem_free(svar, sizeof (dtrace_statvar_t));
9320 svarp[id] = NULL;
9321 }
9322
9323 if (dp->dtdo_buf != NULL)
9324 kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t));
9325 if (dp->dtdo_inttab != NULL)
9326 kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t));
9327 if (dp->dtdo_strtab != NULL)
9328 kmem_free(dp->dtdo_strtab, dp->dtdo_strlen);
9329 if (dp->dtdo_vartab != NULL)
9330 kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t));
9331
9332 kmem_free(dp, sizeof (dtrace_difo_t));
9333}
9334
9335static void
9336dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate)
9337{
9338 int i;
9339
9340 ASSERT(MUTEX_HELD(&dtrace_lock));
9341 ASSERT(dp->dtdo_refcnt != 0);
9342
9343 for (i = 0; i < dp->dtdo_varlen; i++) {
9344 dtrace_difv_t *v = &dp->dtdo_vartab[i];
9345
9346 if (v->dtdv_id != DIF_VAR_VTIMESTAMP)
9347 continue;
9348
9349 ASSERT(dtrace_vtime_references > 0);
9350 if (--dtrace_vtime_references == 0)
9351 dtrace_vtime_disable();
9352 }
9353
9354 if (--dp->dtdo_refcnt == 0)
9355 dtrace_difo_destroy(dp, vstate);
9356}
9357
9358/*
9359 * DTrace Format Functions
9360 */
9361static uint16_t
9362dtrace_format_add(dtrace_state_t *state, char *str)
9363{
9364 char *fmt, **new;
9365 uint16_t ndx, len = strlen(str) + 1;
9366
9367 fmt = kmem_zalloc(len, KM_SLEEP);
9368 bcopy(str, fmt, len);
9369
9370 for (ndx = 0; ndx < state->dts_nformats; ndx++) {
9371 if (state->dts_formats[ndx] == NULL) {
9372 state->dts_formats[ndx] = fmt;
9373 return (ndx + 1);
9374 }
9375 }
9376
9377 if (state->dts_nformats == USHRT_MAX) {
9378 /*
9379 * This is only likely if a denial-of-service attack is being
9380 * attempted. As such, it's okay to fail silently here.
9381 */
9382 kmem_free(fmt, len);
9383 return (0);
9384 }
9385
9386 /*
9387 * For simplicity, we always resize the formats array to be exactly the
9388 * number of formats.
9389 */
9390 ndx = state->dts_nformats++;
9391 new = kmem_alloc((ndx + 1) * sizeof (char *), KM_SLEEP);
9392
9393 if (state->dts_formats != NULL) {
9394 ASSERT(ndx != 0);
9395 bcopy(state->dts_formats, new, ndx * sizeof (char *));
9396 kmem_free(state->dts_formats, ndx * sizeof (char *));
9397 }
9398
9399 state->dts_formats = new;
9400 state->dts_formats[ndx] = fmt;
9401
9402 return (ndx + 1);
9403}
9404
9405static void
9406dtrace_format_remove(dtrace_state_t *state, uint16_t format)
9407{
9408 char *fmt;
9409
9410 ASSERT(state->dts_formats != NULL);
9411 ASSERT(format <= state->dts_nformats);
9412 ASSERT(state->dts_formats[format - 1] != NULL);
9413
9414 fmt = state->dts_formats[format - 1];
9415 kmem_free(fmt, strlen(fmt) + 1);
9416 state->dts_formats[format - 1] = NULL;
9417}
9418
9419static void
9420dtrace_format_destroy(dtrace_state_t *state)
9421{
9422 int i;
9423
9424 if (state->dts_nformats == 0) {
9425 ASSERT(state->dts_formats == NULL);
9426 return;
9427 }
9428
9429 ASSERT(state->dts_formats != NULL);
9430
9431 for (i = 0; i < state->dts_nformats; i++) {
9432 char *fmt = state->dts_formats[i];
9433
9434 if (fmt == NULL)
9435 continue;
9436
9437 kmem_free(fmt, strlen(fmt) + 1);
9438 }
9439
9440 kmem_free(state->dts_formats, state->dts_nformats * sizeof (char *));
9441 state->dts_nformats = 0;
9442 state->dts_formats = NULL;
9443}
9444
9445/*
9446 * DTrace Predicate Functions
9447 */
9448static dtrace_predicate_t *
9449dtrace_predicate_create(dtrace_difo_t *dp)
9450{
9451 dtrace_predicate_t *pred;
9452
9453 ASSERT(MUTEX_HELD(&dtrace_lock));
9454 ASSERT(dp->dtdo_refcnt != 0);
9455
9456 pred = kmem_zalloc(sizeof (dtrace_predicate_t), KM_SLEEP);
9457 pred->dtp_difo = dp;
9458 pred->dtp_refcnt = 1;
9459
9460 if (!dtrace_difo_cacheable(dp))
9461 return (pred);
9462
9463 if (dtrace_predcache_id == DTRACE_CACHEIDNONE) {
9464 /*
9465 * This is only theoretically possible -- we have had 2^32
9466 * cacheable predicates on this machine. We cannot allow any
9467 * more predicates to become cacheable: as unlikely as it is,
9468 * there may be a thread caching a (now stale) predicate cache
9469 * ID. (N.B.: the temptation is being successfully resisted to
9470 * have this cmn_err() "Holy shit -- we executed this code!")
9471 */
9472 return (pred);
9473 }
9474
9475 pred->dtp_cacheid = dtrace_predcache_id++;
9476
9477 return (pred);
9478}
9479
9480static void
9481dtrace_predicate_hold(dtrace_predicate_t *pred)
9482{
9483 ASSERT(MUTEX_HELD(&dtrace_lock));
9484 ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0);
9485 ASSERT(pred->dtp_refcnt > 0);
9486
9487 pred->dtp_refcnt++;
9488}
9489
9490static void
9491dtrace_predicate_release(dtrace_predicate_t *pred, dtrace_vstate_t *vstate)
9492{
9493 dtrace_difo_t *dp = pred->dtp_difo;
9494
9495 ASSERT(MUTEX_HELD(&dtrace_lock));
9496 ASSERT(dp != NULL && dp->dtdo_refcnt != 0);
9497 ASSERT(pred->dtp_refcnt > 0);
9498
9499 if (--pred->dtp_refcnt == 0) {
9500 dtrace_difo_release(pred->dtp_difo, vstate);
9501 kmem_free(pred, sizeof (dtrace_predicate_t));
9502 }
9503}
9504
9505/*
9506 * DTrace Action Description Functions
9507 */
9508static dtrace_actdesc_t *
9509dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple,
9510 uint64_t uarg, uint64_t arg)
9511{
9512 dtrace_actdesc_t *act;
9513
9514#if defined(sun)
9515 ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || (arg != NULL &&
9516 arg >= KERNELBASE) || (arg == NULL && kind == DTRACEACT_PRINTA));
9517#endif
9518
9519 act = kmem_zalloc(sizeof (dtrace_actdesc_t), KM_SLEEP);
9520 act->dtad_kind = kind;
9521 act->dtad_ntuple = ntuple;
9522 act->dtad_uarg = uarg;
9523 act->dtad_arg = arg;
9524 act->dtad_refcnt = 1;
9525
9526 return (act);
9527}
9528
9529static void
9530dtrace_actdesc_hold(dtrace_actdesc_t *act)
9531{
9532 ASSERT(act->dtad_refcnt >= 1);
9533 act->dtad_refcnt++;
9534}
9535
9536static void
9537dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate)
9538{
9539 dtrace_actkind_t kind = act->dtad_kind;
9540 dtrace_difo_t *dp;
9541
9542 ASSERT(act->dtad_refcnt >= 1);
9543
9544 if (--act->dtad_refcnt != 0)
9545 return;
9546
9547 if ((dp = act->dtad_difo) != NULL)
9548 dtrace_difo_release(dp, vstate);
9549
9550 if (DTRACEACT_ISPRINTFLIKE(kind)) {
9551 char *str = (char *)(uintptr_t)act->dtad_arg;
9552
9553#if defined(sun)
9554 ASSERT((str != NULL && (uintptr_t)str >= KERNELBASE) ||
9555 (str == NULL && act->dtad_kind == DTRACEACT_PRINTA));
9556#endif
9557
9558 if (str != NULL)
9559 kmem_free(str, strlen(str) + 1);
9560 }
9561
9562 kmem_free(act, sizeof (dtrace_actdesc_t));
9563}
9564
9565/*
9566 * DTrace ECB Functions
9567 */
9568static dtrace_ecb_t *
9569dtrace_ecb_add(dtrace_state_t *state, dtrace_probe_t *probe)
9570{
9571 dtrace_ecb_t *ecb;
9572 dtrace_epid_t epid;
9573
9574 ASSERT(MUTEX_HELD(&dtrace_lock));
9575
9576 ecb = kmem_zalloc(sizeof (dtrace_ecb_t), KM_SLEEP);
9577 ecb->dte_predicate = NULL;
9578 ecb->dte_probe = probe;
9579
9580 /*
9581 * The default size is the size of the default action: recording
9582 * the epid.
9583 */
9584 ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t);
9585 ecb->dte_alignment = sizeof (dtrace_epid_t);
9586
9587 epid = state->dts_epid++;
9588
9589 if (epid - 1 >= state->dts_necbs) {
9590 dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs;
9591 int necbs = state->dts_necbs << 1;
9592
9593 ASSERT(epid == state->dts_necbs + 1);
9594
9595 if (necbs == 0) {
9596 ASSERT(oecbs == NULL);
9597 necbs = 1;
9598 }
9599
9600 ecbs = kmem_zalloc(necbs * sizeof (*ecbs), KM_SLEEP);
9601
9602 if (oecbs != NULL)
9603 bcopy(oecbs, ecbs, state->dts_necbs * sizeof (*ecbs));
9604
9605 dtrace_membar_producer();
9606 state->dts_ecbs = ecbs;
9607
9608 if (oecbs != NULL) {
9609 /*
9610 * If this state is active, we must dtrace_sync()
9611 * before we can free the old dts_ecbs array: we're
9612 * coming in hot, and there may be active ring
9613 * buffer processing (which indexes into the dts_ecbs
9614 * array) on another CPU.
9615 */
9616 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
9617 dtrace_sync();
9618
9619 kmem_free(oecbs, state->dts_necbs * sizeof (*ecbs));
9620 }
9621
9622 dtrace_membar_producer();
9623 state->dts_necbs = necbs;
9624 }
9625
9626 ecb->dte_state = state;
9627
9628 ASSERT(state->dts_ecbs[epid - 1] == NULL);
9629 dtrace_membar_producer();
9630 state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb;
9631
9632 return (ecb);
9633}
9634
9635static void
9636dtrace_ecb_enable(dtrace_ecb_t *ecb)
9637{
9638 dtrace_probe_t *probe = ecb->dte_probe;
9639
9640 ASSERT(MUTEX_HELD(&cpu_lock));
9641 ASSERT(MUTEX_HELD(&dtrace_lock));
9642 ASSERT(ecb->dte_next == NULL);
9643
9644 if (probe == NULL) {
9645 /*
9646 * This is the NULL probe -- there's nothing to do.
9647 */
9648 return;
9649 }
9650
9651 if (probe->dtpr_ecb == NULL) {
9652 dtrace_provider_t *prov = probe->dtpr_provider;
9653
9654 /*
9655 * We're the first ECB on this probe.
9656 */
9657 probe->dtpr_ecb = probe->dtpr_ecb_last = ecb;
9658
9659 if (ecb->dte_predicate != NULL)
9660 probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid;
9661
9662 prov->dtpv_pops.dtps_enable(prov->dtpv_arg,
9663 probe->dtpr_id, probe->dtpr_arg);
9664 } else {
9665 /*
9666 * This probe is already active. Swing the last pointer to
9667 * point to the new ECB, and issue a dtrace_sync() to assure
9668 * that all CPUs have seen the change.
9669 */
9670 ASSERT(probe->dtpr_ecb_last != NULL);
9671 probe->dtpr_ecb_last->dte_next = ecb;
9672 probe->dtpr_ecb_last = ecb;
9673 probe->dtpr_predcache = 0;
9674
9675 dtrace_sync();
9676 }
9677}
9678
9679static void
9680dtrace_ecb_resize(dtrace_ecb_t *ecb)
9681{
9682 uint32_t maxalign = sizeof (dtrace_epid_t);
9683 uint32_t align = sizeof (uint8_t), offs, diff;
9684 dtrace_action_t *act;
9685 int wastuple = 0;
9686 uint32_t aggbase = UINT32_MAX;
9687 dtrace_state_t *state = ecb->dte_state;
9688
9689 /*
9690 * If we record anything, we always record the epid. (And we always
9691 * record it first.)
9692 */
9693 offs = sizeof (dtrace_epid_t);
9694 ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t);
9695
9696 for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
9697 dtrace_recdesc_t *rec = &act->dta_rec;
9698
9699 if ((align = rec->dtrd_alignment) > maxalign)
9700 maxalign = align;
9701
9702 if (!wastuple && act->dta_intuple) {
9703 /*
9704 * This is the first record in a tuple. Align the
9705 * offset to be at offset 4 in an 8-byte aligned
9706 * block.
9707 */
9708 diff = offs + sizeof (dtrace_aggid_t);
9709
9710 if ((diff = (diff & (sizeof (uint64_t) - 1))))
9711 offs += sizeof (uint64_t) - diff;
9712
9713 aggbase = offs - sizeof (dtrace_aggid_t);
9714 ASSERT(!(aggbase & (sizeof (uint64_t) - 1)));
9715 }
9716
9717 /*LINTED*/
9718 if (rec->dtrd_size != 0 && (diff = (offs & (align - 1)))) {
9719 /*
9720 * The current offset is not properly aligned; align it.
9721 */
9722 offs += align - diff;
9723 }
9724
9725 rec->dtrd_offset = offs;
9726
9727 if (offs + rec->dtrd_size > ecb->dte_needed) {
9728 ecb->dte_needed = offs + rec->dtrd_size;
9729
9730 if (ecb->dte_needed > state->dts_needed)
9731 state->dts_needed = ecb->dte_needed;
9732 }
9733
9734 if (DTRACEACT_ISAGG(act->dta_kind)) {
9735 dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act;
9736 dtrace_action_t *first = agg->dtag_first, *prev;
9737
9738 ASSERT(rec->dtrd_size != 0 && first != NULL);
9739 ASSERT(wastuple);
9740 ASSERT(aggbase != UINT32_MAX);
9741
9742 agg->dtag_base = aggbase;
9743
9744 while ((prev = first->dta_prev) != NULL &&
9745 DTRACEACT_ISAGG(prev->dta_kind)) {
9746 agg = (dtrace_aggregation_t *)prev;
9747 first = agg->dtag_first;
9748 }
9749
9750 if (prev != NULL) {
9751 offs = prev->dta_rec.dtrd_offset +
9752 prev->dta_rec.dtrd_size;
9753 } else {
9754 offs = sizeof (dtrace_epid_t);
9755 }
9756 wastuple = 0;
9757 } else {
9758 if (!act->dta_intuple)
9759 ecb->dte_size = offs + rec->dtrd_size;
9760
9761 offs += rec->dtrd_size;
9762 }
9763
9764 wastuple = act->dta_intuple;
9765 }
9766
9767 if ((act = ecb->dte_action) != NULL &&
9768 !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) &&
9769 ecb->dte_size == sizeof (dtrace_epid_t)) {
9770 /*
9771 * If the size is still sizeof (dtrace_epid_t), then all
9772 * actions store no data; set the size to 0.
9773 */
9774 ecb->dte_alignment = maxalign;
9775 ecb->dte_size = 0;
9776
9777 /*
9778 * If the needed space is still sizeof (dtrace_epid_t), then
9779 * all actions need no additional space; set the needed
9780 * size to 0.
9781 */
9782 if (ecb->dte_needed == sizeof (dtrace_epid_t))
9783 ecb->dte_needed = 0;
9784
9785 return;
9786 }
9787
9788 /*
9789 * Set our alignment, and make sure that the dte_size and dte_needed
9790 * are aligned to the size of an EPID.
9791 */
9792 ecb->dte_alignment = maxalign;
9793 ecb->dte_size = (ecb->dte_size + (sizeof (dtrace_epid_t) - 1)) &
9794 ~(sizeof (dtrace_epid_t) - 1);
9795 ecb->dte_needed = (ecb->dte_needed + (sizeof (dtrace_epid_t) - 1)) &
9796 ~(sizeof (dtrace_epid_t) - 1);
9797 ASSERT(ecb->dte_size <= ecb->dte_needed);
9798}
9799
9800static dtrace_action_t *
9801dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc)
9802{
9803 dtrace_aggregation_t *agg;
9804 size_t size = sizeof (uint64_t);
9805 int ntuple = desc->dtad_ntuple;
9806 dtrace_action_t *act;
9807 dtrace_recdesc_t *frec;
9808 dtrace_aggid_t aggid;
9809 dtrace_state_t *state = ecb->dte_state;
9810
9811 agg = kmem_zalloc(sizeof (dtrace_aggregation_t), KM_SLEEP);
9812 agg->dtag_ecb = ecb;
9813
9814 ASSERT(DTRACEACT_ISAGG(desc->dtad_kind));
9815
9816 switch (desc->dtad_kind) {
9817 case DTRACEAGG_MIN:
9818 agg->dtag_initial = INT64_MAX;
9819 agg->dtag_aggregate = dtrace_aggregate_min;
9820 break;
9821
9822 case DTRACEAGG_MAX:
9823 agg->dtag_initial = INT64_MIN;
9824 agg->dtag_aggregate = dtrace_aggregate_max;
9825 break;
9826
9827 case DTRACEAGG_COUNT:
9828 agg->dtag_aggregate = dtrace_aggregate_count;
9829 break;
9830
9831 case DTRACEAGG_QUANTIZE:
9832 agg->dtag_aggregate = dtrace_aggregate_quantize;
9833 size = (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) *
9834 sizeof (uint64_t);
9835 break;
9836
9837 case DTRACEAGG_LQUANTIZE: {
9838 uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg);
9839 uint16_t levels = DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg);
9840
9841 agg->dtag_initial = desc->dtad_arg;
9842 agg->dtag_aggregate = dtrace_aggregate_lquantize;
9843
9844 if (step == 0 || levels == 0)
9845 goto err;
9846
9847 size = levels * sizeof (uint64_t) + 3 * sizeof (uint64_t);
9848 break;
9849 }
9850
9851 case DTRACEAGG_AVG:
9852 agg->dtag_aggregate = dtrace_aggregate_avg;
9853 size = sizeof (uint64_t) * 2;
9854 break;
9855
9856 case DTRACEAGG_STDDEV:
9857 agg->dtag_aggregate = dtrace_aggregate_stddev;
9858 size = sizeof (uint64_t) * 4;
9859 break;
9860
9861 case DTRACEAGG_SUM:
9862 agg->dtag_aggregate = dtrace_aggregate_sum;
9863 break;
9864
9865 default:
9866 goto err;
9867 }
9868
9869 agg->dtag_action.dta_rec.dtrd_size = size;
9870
9871 if (ntuple == 0)
9872 goto err;
9873
9874 /*
9875 * We must make sure that we have enough actions for the n-tuple.
9876 */
9877 for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) {
9878 if (DTRACEACT_ISAGG(act->dta_kind))
9879 break;
9880
9881 if (--ntuple == 0) {
9882 /*
9883 * This is the action with which our n-tuple begins.
9884 */
9885 agg->dtag_first = act;
9886 goto success;
9887 }
9888 }
9889
9890 /*
9891 * This n-tuple is short by ntuple elements. Return failure.
9892 */
9893 ASSERT(ntuple != 0);
9894err:
9895 kmem_free(agg, sizeof (dtrace_aggregation_t));
9896 return (NULL);
9897
9898success:
9899 /*
9900 * If the last action in the tuple has a size of zero, it's actually
9901 * an expression argument for the aggregating action.
9902 */
9903 ASSERT(ecb->dte_action_last != NULL);
9904 act = ecb->dte_action_last;
9905
9906 if (act->dta_kind == DTRACEACT_DIFEXPR) {
9907 ASSERT(act->dta_difo != NULL);
9908
9909 if (act->dta_difo->dtdo_rtype.dtdt_size == 0)
9910 agg->dtag_hasarg = 1;
9911 }
9912
9913 /*
9914 * We need to allocate an id for this aggregation.
9915 */
9916#if defined(sun)
9917 aggid = (dtrace_aggid_t)(uintptr_t)vmem_alloc(state->dts_aggid_arena, 1,
9918 VM_BESTFIT | VM_SLEEP);
9919#else
9920 aggid = alloc_unr(state->dts_aggid_arena);
9921#endif
9922
9923 if (aggid - 1 >= state->dts_naggregations) {
9924 dtrace_aggregation_t **oaggs = state->dts_aggregations;
9925 dtrace_aggregation_t **aggs;
9926 int naggs = state->dts_naggregations << 1;
9927 int onaggs = state->dts_naggregations;
9928
9929 ASSERT(aggid == state->dts_naggregations + 1);
9930
9931 if (naggs == 0) {
9932 ASSERT(oaggs == NULL);
9933 naggs = 1;
9934 }
9935
9936 aggs = kmem_zalloc(naggs * sizeof (*aggs), KM_SLEEP);
9937
9938 if (oaggs != NULL) {
9939 bcopy(oaggs, aggs, onaggs * sizeof (*aggs));
9940 kmem_free(oaggs, onaggs * sizeof (*aggs));
9941 }
9942
9943 state->dts_aggregations = aggs;
9944 state->dts_naggregations = naggs;
9945 }
9946
9947 ASSERT(state->dts_aggregations[aggid - 1] == NULL);
9948 state->dts_aggregations[(agg->dtag_id = aggid) - 1] = agg;
9949
9950 frec = &agg->dtag_first->dta_rec;
9951 if (frec->dtrd_alignment < sizeof (dtrace_aggid_t))
9952 frec->dtrd_alignment = sizeof (dtrace_aggid_t);
9953
9954 for (act = agg->dtag_first; act != NULL; act = act->dta_next) {
9955 ASSERT(!act->dta_intuple);
9956 act->dta_intuple = 1;
9957 }
9958
9959 return (&agg->dtag_action);
9960}
9961
9962static void
9963dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act)
9964{
9965 dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act;
9966 dtrace_state_t *state = ecb->dte_state;
9967 dtrace_aggid_t aggid = agg->dtag_id;
9968
9969 ASSERT(DTRACEACT_ISAGG(act->dta_kind));
9970#if defined(sun)
9971 vmem_free(state->dts_aggid_arena, (void *)(uintptr_t)aggid, 1);
9972#else
9973 free_unr(state->dts_aggid_arena, aggid);
9974#endif
9975
9976 ASSERT(state->dts_aggregations[aggid - 1] == agg);
9977 state->dts_aggregations[aggid - 1] = NULL;
9978
9979 kmem_free(agg, sizeof (dtrace_aggregation_t));
9980}
9981
9982static int
9983dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc)
9984{
9985 dtrace_action_t *action, *last;
9986 dtrace_difo_t *dp = desc->dtad_difo;
9987 uint32_t size = 0, align = sizeof (uint8_t), mask;
9988 uint16_t format = 0;
9989 dtrace_recdesc_t *rec;
9990 dtrace_state_t *state = ecb->dte_state;
9991 dtrace_optval_t *opt = state->dts_options, nframes = 0, strsize;
9992 uint64_t arg = desc->dtad_arg;
9993
9994 ASSERT(MUTEX_HELD(&dtrace_lock));
9995 ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1);
9996
9997 if (DTRACEACT_ISAGG(desc->dtad_kind)) {
9998 /*
9999 * If this is an aggregating action, there must be neither
10000 * a speculate nor a commit on the action chain.
10001 */
10002 dtrace_action_t *act;
10003
10004 for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
10005 if (act->dta_kind == DTRACEACT_COMMIT)
10006 return (EINVAL);
10007
10008 if (act->dta_kind == DTRACEACT_SPECULATE)
10009 return (EINVAL);
10010 }
10011
10012 action = dtrace_ecb_aggregation_create(ecb, desc);
10013
10014 if (action == NULL)
10015 return (EINVAL);
10016 } else {
10017 if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) ||
10018 (desc->dtad_kind == DTRACEACT_DIFEXPR &&
10019 dp != NULL && dp->dtdo_destructive)) {
10020 state->dts_destructive = 1;
10021 }
10022
10023 switch (desc->dtad_kind) {
10024 case DTRACEACT_PRINTF:
10025 case DTRACEACT_PRINTA:
10026 case DTRACEACT_SYSTEM:
10027 case DTRACEACT_FREOPEN:
10028 /*
10029 * We know that our arg is a string -- turn it into a
10030 * format.
10031 */
10032 if (arg == 0) {
10033 ASSERT(desc->dtad_kind == DTRACEACT_PRINTA);
10034 format = 0;
10035 } else {
10036 ASSERT(arg != 0);
10037#if defined(sun)
10038 ASSERT(arg > KERNELBASE);
10039#endif
10040 format = dtrace_format_add(state,
10041 (char *)(uintptr_t)arg);
10042 }
10043
10044 /*FALLTHROUGH*/
10045 case DTRACEACT_LIBACT:
10046 case DTRACEACT_DIFEXPR:
10047 if (dp == NULL)
10048 return (EINVAL);
10049
10050 if ((size = dp->dtdo_rtype.dtdt_size) != 0)
10051 break;
10052
10053 if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) {
10054 if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10055 return (EINVAL);
10056
10057 size = opt[DTRACEOPT_STRSIZE];
10058 }
10059
10060 break;
10061
10062 case DTRACEACT_STACK:
10063 if ((nframes = arg) == 0) {
10064 nframes = opt[DTRACEOPT_STACKFRAMES];
10065 ASSERT(nframes > 0);
10066 arg = nframes;
10067 }
10068
10069 size = nframes * sizeof (pc_t);
10070 break;
10071
10072 case DTRACEACT_JSTACK:
10073 if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0)
10074 strsize = opt[DTRACEOPT_JSTACKSTRSIZE];
10075
10076 if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0)
10077 nframes = opt[DTRACEOPT_JSTACKFRAMES];
10078
10079 arg = DTRACE_USTACK_ARG(nframes, strsize);
10080
10081 /*FALLTHROUGH*/
10082 case DTRACEACT_USTACK:
10083 if (desc->dtad_kind != DTRACEACT_JSTACK &&
10084 (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) {
10085 strsize = DTRACE_USTACK_STRSIZE(arg);
10086 nframes = opt[DTRACEOPT_USTACKFRAMES];
10087 ASSERT(nframes > 0);
10088 arg = DTRACE_USTACK_ARG(nframes, strsize);
10089 }
10090
10091 /*
10092 * Save a slot for the pid.
10093 */
10094 size = (nframes + 1) * sizeof (uint64_t);
10095 size += DTRACE_USTACK_STRSIZE(arg);
10096 size = P2ROUNDUP(size, (uint32_t)(sizeof (uintptr_t)));
10097
10098 break;
10099
10100 case DTRACEACT_SYM:
10101 case DTRACEACT_MOD:
10102 if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) !=
10103 sizeof (uint64_t)) ||
10104 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10105 return (EINVAL);
10106 break;
10107
10108 case DTRACEACT_USYM:
10109 case DTRACEACT_UMOD:
10110 case DTRACEACT_UADDR:
10111 if (dp == NULL ||
10112 (dp->dtdo_rtype.dtdt_size != sizeof (uint64_t)) ||
10113 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10114 return (EINVAL);
10115
10116 /*
10117 * We have a slot for the pid, plus a slot for the
10118 * argument. To keep things simple (aligned with
10119 * bitness-neutral sizing), we store each as a 64-bit
10120 * quantity.
10121 */
10122 size = 2 * sizeof (uint64_t);
10123 break;
10124
10125 case DTRACEACT_STOP:
10126 case DTRACEACT_BREAKPOINT:
10127 case DTRACEACT_PANIC:
10128 break;
10129
10130 case DTRACEACT_CHILL:
10131 case DTRACEACT_DISCARD:
10132 case DTRACEACT_RAISE:
10133 if (dp == NULL)
10134 return (EINVAL);
10135 break;
10136
10137 case DTRACEACT_EXIT:
10138 if (dp == NULL ||
10139 (size = dp->dtdo_rtype.dtdt_size) != sizeof (int) ||
10140 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF))
10141 return (EINVAL);
10142 break;
10143
10144 case DTRACEACT_SPECULATE:
10145 if (ecb->dte_size > sizeof (dtrace_epid_t))
10146 return (EINVAL);
10147
10148 if (dp == NULL)
10149 return (EINVAL);
10150
10151 state->dts_speculates = 1;
10152 break;
10153
10154 case DTRACEACT_PRINTM:
10155 size = dp->dtdo_rtype.dtdt_size;
10156 break;
10157
10158 case DTRACEACT_PRINTT:
10159 size = dp->dtdo_rtype.dtdt_size;
10160 break;
10161
10162 case DTRACEACT_COMMIT: {
10163 dtrace_action_t *act = ecb->dte_action;
10164
10165 for (; act != NULL; act = act->dta_next) {
10166 if (act->dta_kind == DTRACEACT_COMMIT)
10167 return (EINVAL);
10168 }
10169
10170 if (dp == NULL)
10171 return (EINVAL);
10172 break;
10173 }
10174
10175 default:
10176 return (EINVAL);
10177 }
10178
10179 if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) {
10180 /*
10181 * If this is a data-storing action or a speculate,
10182 * we must be sure that there isn't a commit on the
10183 * action chain.
10184 */
10185 dtrace_action_t *act = ecb->dte_action;
10186
10187 for (; act != NULL; act = act->dta_next) {
10188 if (act->dta_kind == DTRACEACT_COMMIT)
10189 return (EINVAL);
10190 }
10191 }
10192
10193 action = kmem_zalloc(sizeof (dtrace_action_t), KM_SLEEP);
10194 action->dta_rec.dtrd_size = size;
10195 }
10196
10197 action->dta_refcnt = 1;
10198 rec = &action->dta_rec;
10199 size = rec->dtrd_size;
10200
10201 for (mask = sizeof (uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) {
10202 if (!(size & mask)) {
10203 align = mask + 1;
10204 break;
10205 }
10206 }
10207
10208 action->dta_kind = desc->dtad_kind;
10209
10210 if ((action->dta_difo = dp) != NULL)
10211 dtrace_difo_hold(dp);
10212
10213 rec->dtrd_action = action->dta_kind;
10214 rec->dtrd_arg = arg;
10215 rec->dtrd_uarg = desc->dtad_uarg;
10216 rec->dtrd_alignment = (uint16_t)align;
10217 rec->dtrd_format = format;
10218
10219 if ((last = ecb->dte_action_last) != NULL) {
10220 ASSERT(ecb->dte_action != NULL);
10221 action->dta_prev = last;
10222 last->dta_next = action;
10223 } else {
10224 ASSERT(ecb->dte_action == NULL);
10225 ecb->dte_action = action;
10226 }
10227
10228 ecb->dte_action_last = action;
10229
10230 return (0);
10231}
10232
10233static void
10234dtrace_ecb_action_remove(dtrace_ecb_t *ecb)
10235{
10236 dtrace_action_t *act = ecb->dte_action, *next;
10237 dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate;
10238 dtrace_difo_t *dp;
10239 uint16_t format;
10240
10241 if (act != NULL && act->dta_refcnt > 1) {
10242 ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1);
10243 act->dta_refcnt--;
10244 } else {
10245 for (; act != NULL; act = next) {
10246 next = act->dta_next;
10247 ASSERT(next != NULL || act == ecb->dte_action_last);
10248 ASSERT(act->dta_refcnt == 1);
10249
10250 if ((format = act->dta_rec.dtrd_format) != 0)
10251 dtrace_format_remove(ecb->dte_state, format);
10252
10253 if ((dp = act->dta_difo) != NULL)
10254 dtrace_difo_release(dp, vstate);
10255
10256 if (DTRACEACT_ISAGG(act->dta_kind)) {
10257 dtrace_ecb_aggregation_destroy(ecb, act);
10258 } else {
10259 kmem_free(act, sizeof (dtrace_action_t));
10260 }
10261 }
10262 }
10263
10264 ecb->dte_action = NULL;
10265 ecb->dte_action_last = NULL;
10266 ecb->dte_size = sizeof (dtrace_epid_t);
10267}
10268
10269static void
10270dtrace_ecb_disable(dtrace_ecb_t *ecb)
10271{
10272 /*
10273 * We disable the ECB by removing it from its probe.
10274 */
10275 dtrace_ecb_t *pecb, *prev = NULL;
10276 dtrace_probe_t *probe = ecb->dte_probe;
10277
10278 ASSERT(MUTEX_HELD(&dtrace_lock));
10279
10280 if (probe == NULL) {
10281 /*
10282 * This is the NULL probe; there is nothing to disable.
10283 */
10284 return;
10285 }
10286
10287 for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) {
10288 if (pecb == ecb)
10289 break;
10290 prev = pecb;
10291 }
10292
10293 ASSERT(pecb != NULL);
10294
10295 if (prev == NULL) {
10296 probe->dtpr_ecb = ecb->dte_next;
10297 } else {
10298 prev->dte_next = ecb->dte_next;
10299 }
10300
10301 if (ecb == probe->dtpr_ecb_last) {
10302 ASSERT(ecb->dte_next == NULL);
10303 probe->dtpr_ecb_last = prev;
10304 }
10305
10306 /*
10307 * The ECB has been disconnected from the probe; now sync to assure
10308 * that all CPUs have seen the change before returning.
10309 */
10310 dtrace_sync();
10311
10312 if (probe->dtpr_ecb == NULL) {
10313 /*
10314 * That was the last ECB on the probe; clear the predicate
10315 * cache ID for the probe, disable it and sync one more time
10316 * to assure that we'll never hit it again.
10317 */
10318 dtrace_provider_t *prov = probe->dtpr_provider;
10319
10320 ASSERT(ecb->dte_next == NULL);
10321 ASSERT(probe->dtpr_ecb_last == NULL);
10322 probe->dtpr_predcache = DTRACE_CACHEIDNONE;
10323 prov->dtpv_pops.dtps_disable(prov->dtpv_arg,
10324 probe->dtpr_id, probe->dtpr_arg);
10325 dtrace_sync();
10326 } else {
10327 /*
10328 * There is at least one ECB remaining on the probe. If there
10329 * is _exactly_ one, set the probe's predicate cache ID to be
10330 * the predicate cache ID of the remaining ECB.
10331 */
10332 ASSERT(probe->dtpr_ecb_last != NULL);
10333 ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE);
10334
10335 if (probe->dtpr_ecb == probe->dtpr_ecb_last) {
10336 dtrace_predicate_t *p = probe->dtpr_ecb->dte_predicate;
10337
10338 ASSERT(probe->dtpr_ecb->dte_next == NULL);
10339
10340 if (p != NULL)
10341 probe->dtpr_predcache = p->dtp_cacheid;
10342 }
10343
10344 ecb->dte_next = NULL;
10345 }
10346}
10347
10348static void
10349dtrace_ecb_destroy(dtrace_ecb_t *ecb)
10350{
10351 dtrace_state_t *state = ecb->dte_state;
10352 dtrace_vstate_t *vstate = &state->dts_vstate;
10353 dtrace_predicate_t *pred;
10354 dtrace_epid_t epid = ecb->dte_epid;
10355
10356 ASSERT(MUTEX_HELD(&dtrace_lock));
10357 ASSERT(ecb->dte_next == NULL);
10358 ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb);
10359
10360 if ((pred = ecb->dte_predicate) != NULL)
10361 dtrace_predicate_release(pred, vstate);
10362
10363 dtrace_ecb_action_remove(ecb);
10364
10365 ASSERT(state->dts_ecbs[epid - 1] == ecb);
10366 state->dts_ecbs[epid - 1] = NULL;
10367
10368 kmem_free(ecb, sizeof (dtrace_ecb_t));
10369}
10370
10371static dtrace_ecb_t *
10372dtrace_ecb_create(dtrace_state_t *state, dtrace_probe_t *probe,
10373 dtrace_enabling_t *enab)
10374{
10375 dtrace_ecb_t *ecb;
10376 dtrace_predicate_t *pred;
10377 dtrace_actdesc_t *act;
10378 dtrace_provider_t *prov;
10379 dtrace_ecbdesc_t *desc = enab->dten_current;
10380
10381 ASSERT(MUTEX_HELD(&dtrace_lock));
10382 ASSERT(state != NULL);
10383
10384 ecb = dtrace_ecb_add(state, probe);
10385 ecb->dte_uarg = desc->dted_uarg;
10386
10387 if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) {
10388 dtrace_predicate_hold(pred);
10389 ecb->dte_predicate = pred;
10390 }
10391
10392 if (probe != NULL) {
10393 /*
10394 * If the provider shows more leg than the consumer is old
10395 * enough to see, we need to enable the appropriate implicit
10396 * predicate bits to prevent the ecb from activating at
10397 * revealing times.
10398 *
10399 * Providers specifying DTRACE_PRIV_USER at register time
10400 * are stating that they need the /proc-style privilege
10401 * model to be enforced, and this is what DTRACE_COND_OWNER
10402 * and DTRACE_COND_ZONEOWNER will then do at probe time.
10403 */
10404 prov = probe->dtpr_provider;
10405 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) &&
10406 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER))
10407 ecb->dte_cond |= DTRACE_COND_OWNER;
10408
10409 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLZONE) &&
10410 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER))
10411 ecb->dte_cond |= DTRACE_COND_ZONEOWNER;
10412
10413 /*
10414 * If the provider shows us kernel innards and the user
10415 * is lacking sufficient privilege, enable the
10416 * DTRACE_COND_USERMODE implicit predicate.
10417 */
10418 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) &&
10419 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL))
10420 ecb->dte_cond |= DTRACE_COND_USERMODE;
10421 }
10422
10423 if (dtrace_ecb_create_cache != NULL) {
10424 /*
10425 * If we have a cached ecb, we'll use its action list instead
10426 * of creating our own (saving both time and space).
10427 */
10428 dtrace_ecb_t *cached = dtrace_ecb_create_cache;
10429 dtrace_action_t *act = cached->dte_action;
10430
10431 if (act != NULL) {
10432 ASSERT(act->dta_refcnt > 0);
10433 act->dta_refcnt++;
10434 ecb->dte_action = act;
10435 ecb->dte_action_last = cached->dte_action_last;
10436 ecb->dte_needed = cached->dte_needed;
10437 ecb->dte_size = cached->dte_size;
10438 ecb->dte_alignment = cached->dte_alignment;
10439 }
10440
10441 return (ecb);
10442 }
10443
10444 for (act = desc->dted_action; act != NULL; act = act->dtad_next) {
10445 if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) {
10446 dtrace_ecb_destroy(ecb);
10447 return (NULL);
10448 }
10449 }
10450
10451 dtrace_ecb_resize(ecb);
10452
10453 return (dtrace_ecb_create_cache = ecb);
10454}
10455
10456static int
10457dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg)
10458{
10459 dtrace_ecb_t *ecb;
10460 dtrace_enabling_t *enab = arg;
10461 dtrace_state_t *state = enab->dten_vstate->dtvs_state;
10462
10463 ASSERT(state != NULL);
10464
10465 if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) {
10466 /*
10467 * This probe was created in a generation for which this
10468 * enabling has previously created ECBs; we don't want to
10469 * enable it again, so just kick out.
10470 */
10471 return (DTRACE_MATCH_NEXT);
10472 }
10473
10474 if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL)
10475 return (DTRACE_MATCH_DONE);
10476
10477 dtrace_ecb_enable(ecb);
10478 return (DTRACE_MATCH_NEXT);
10479}
10480
10481static dtrace_ecb_t *
10482dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id)
10483{
10484 dtrace_ecb_t *ecb;
10485
10486 ASSERT(MUTEX_HELD(&dtrace_lock));
10487
10488 if (id == 0 || id > state->dts_necbs)
10489 return (NULL);
10490
10491 ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL);
10492 ASSERT((ecb = state->dts_ecbs[id - 1]) == NULL || ecb->dte_epid == id);
10493
10494 return (state->dts_ecbs[id - 1]);
10495}
10496
10497static dtrace_aggregation_t *
10498dtrace_aggid2agg(dtrace_state_t *state, dtrace_aggid_t id)
10499{
10500 dtrace_aggregation_t *agg;
10501
10502 ASSERT(MUTEX_HELD(&dtrace_lock));
10503
10504 if (id == 0 || id > state->dts_naggregations)
10505 return (NULL);
10506
10507 ASSERT(state->dts_naggregations > 0 && state->dts_aggregations != NULL);
10508 ASSERT((agg = state->dts_aggregations[id - 1]) == NULL ||
10509 agg->dtag_id == id);
10510
10511 return (state->dts_aggregations[id - 1]);
10512}
10513
10514/*
10515 * DTrace Buffer Functions
10516 *
10517 * The following functions manipulate DTrace buffers. Most of these functions
10518 * are called in the context of establishing or processing consumer state;
10519 * exceptions are explicitly noted.
10520 */
10521
10522/*
10523 * Note: called from cross call context. This function switches the two
10524 * buffers on a given CPU. The atomicity of this operation is assured by
10525 * disabling interrupts while the actual switch takes place; the disabling of
10526 * interrupts serializes the execution with any execution of dtrace_probe() on
10527 * the same CPU.
10528 */
10529static void
10530dtrace_buffer_switch(dtrace_buffer_t *buf)
10531{
10532 caddr_t tomax = buf->dtb_tomax;
10533 caddr_t xamot = buf->dtb_xamot;
10534 dtrace_icookie_t cookie;
10535
10536 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
10537 ASSERT(!(buf->dtb_flags & DTRACEBUF_RING));
10538
10539 cookie = dtrace_interrupt_disable();
10540 buf->dtb_tomax = xamot;
10541 buf->dtb_xamot = tomax;
10542 buf->dtb_xamot_drops = buf->dtb_drops;
10543 buf->dtb_xamot_offset = buf->dtb_offset;
10544 buf->dtb_xamot_errors = buf->dtb_errors;
10545 buf->dtb_xamot_flags = buf->dtb_flags;
10546 buf->dtb_offset = 0;
10547 buf->dtb_drops = 0;
10548 buf->dtb_errors = 0;
10549 buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED);
10550 dtrace_interrupt_enable(cookie);
10551}
10552
10553/*
10554 * Note: called from cross call context. This function activates a buffer
10555 * on a CPU. As with dtrace_buffer_switch(), the atomicity of the operation
10556 * is guaranteed by the disabling of interrupts.
10557 */
10558static void
10559dtrace_buffer_activate(dtrace_state_t *state)
10560{
10561 dtrace_buffer_t *buf;
10562 dtrace_icookie_t cookie = dtrace_interrupt_disable();
10563
10564 buf = &state->dts_buffer[curcpu];
10565
10566 if (buf->dtb_tomax != NULL) {
10567 /*
10568 * We might like to assert that the buffer is marked inactive,
10569 * but this isn't necessarily true: the buffer for the CPU
10570 * that processes the BEGIN probe has its buffer activated
10571 * manually. In this case, we take the (harmless) action
10572 * re-clearing the bit INACTIVE bit.
10573 */
10574 buf->dtb_flags &= ~DTRACEBUF_INACTIVE;
10575 }
10576
10577 dtrace_interrupt_enable(cookie);
10578}
10579
10580static int
10581dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags,
10582 processorid_t cpu)
10583{
10584#if defined(sun)
10585 cpu_t *cp;
|
10714 if (cpu != DTRACE_CPUALL && cpu != i)
10715 continue;
10716
10717 buf = &bufs[i];
10718
10719 if (buf->dtb_xamot != NULL) {
10720 ASSERT(buf->dtb_tomax != NULL);
10721 ASSERT(buf->dtb_size == size);
10722 kmem_free(buf->dtb_xamot, size);
10723 }
10724
10725 if (buf->dtb_tomax != NULL) {
10726 ASSERT(buf->dtb_size == size);
10727 kmem_free(buf->dtb_tomax, size);
10728 }
10729
10730 buf->dtb_tomax = NULL;
10731 buf->dtb_xamot = NULL;
10732 buf->dtb_size = 0;
10733
10734 }
10735
10736 return (ENOMEM);
10737#endif
10738}
10739
10740/*
10741 * Note: called from probe context. This function just increments the drop
10742 * count on a buffer. It has been made a function to allow for the
10743 * possibility of understanding the source of mysterious drop counts. (A
10744 * problem for which one may be particularly disappointed that DTrace cannot
10745 * be used to understand DTrace.)
10746 */
10747static void
10748dtrace_buffer_drop(dtrace_buffer_t *buf)
10749{
10750 buf->dtb_drops++;
10751}
10752
10753/*
10754 * Note: called from probe context. This function is called to reserve space
10755 * in a buffer. If mstate is non-NULL, sets the scratch base and size in the
10756 * mstate. Returns the new offset in the buffer, or a negative value if an
10757 * error has occurred.
10758 */
10759static intptr_t
10760dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, size_t align,
10761 dtrace_state_t *state, dtrace_mstate_t *mstate)
10762{
10763 intptr_t offs = buf->dtb_offset, soffs;
10764 intptr_t woffs;
10765 caddr_t tomax;
10766 size_t total;
10767
10768 if (buf->dtb_flags & DTRACEBUF_INACTIVE)
10769 return (-1);
10770
10771 if ((tomax = buf->dtb_tomax) == NULL) {
10772 dtrace_buffer_drop(buf);
10773 return (-1);
10774 }
10775
10776 if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) {
10777 while (offs & (align - 1)) {
10778 /*
10779 * Assert that our alignment is off by a number which
10780 * is itself sizeof (uint32_t) aligned.
10781 */
10782 ASSERT(!((align - (offs & (align - 1))) &
10783 (sizeof (uint32_t) - 1)));
10784 DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE);
10785 offs += sizeof (uint32_t);
10786 }
10787
10788 if ((soffs = offs + needed) > buf->dtb_size) {
10789 dtrace_buffer_drop(buf);
10790 return (-1);
10791 }
10792
10793 if (mstate == NULL)
10794 return (offs);
10795
10796 mstate->dtms_scratch_base = (uintptr_t)tomax + soffs;
10797 mstate->dtms_scratch_size = buf->dtb_size - soffs;
10798 mstate->dtms_scratch_ptr = mstate->dtms_scratch_base;
10799
10800 return (offs);
10801 }
10802
10803 if (buf->dtb_flags & DTRACEBUF_FILL) {
10804 if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN &&
10805 (buf->dtb_flags & DTRACEBUF_FULL))
10806 return (-1);
10807 goto out;
10808 }
10809
10810 total = needed + (offs & (align - 1));
10811
10812 /*
10813 * For a ring buffer, life is quite a bit more complicated. Before
10814 * we can store any padding, we need to adjust our wrapping offset.
10815 * (If we've never before wrapped or we're not about to, no adjustment
10816 * is required.)
10817 */
10818 if ((buf->dtb_flags & DTRACEBUF_WRAPPED) ||
10819 offs + total > buf->dtb_size) {
10820 woffs = buf->dtb_xamot_offset;
10821
10822 if (offs + total > buf->dtb_size) {
10823 /*
10824 * We can't fit in the end of the buffer. First, a
10825 * sanity check that we can fit in the buffer at all.
10826 */
10827 if (total > buf->dtb_size) {
10828 dtrace_buffer_drop(buf);
10829 return (-1);
10830 }
10831
10832 /*
10833 * We're going to be storing at the top of the buffer,
10834 * so now we need to deal with the wrapped offset. We
10835 * only reset our wrapped offset to 0 if it is
10836 * currently greater than the current offset. If it
10837 * is less than the current offset, it is because a
10838 * previous allocation induced a wrap -- but the
10839 * allocation didn't subsequently take the space due
10840 * to an error or false predicate evaluation. In this
10841 * case, we'll just leave the wrapped offset alone: if
10842 * the wrapped offset hasn't been advanced far enough
10843 * for this allocation, it will be adjusted in the
10844 * lower loop.
10845 */
10846 if (buf->dtb_flags & DTRACEBUF_WRAPPED) {
10847 if (woffs >= offs)
10848 woffs = 0;
10849 } else {
10850 woffs = 0;
10851 }
10852
10853 /*
10854 * Now we know that we're going to be storing to the
10855 * top of the buffer and that there is room for us
10856 * there. We need to clear the buffer from the current
10857 * offset to the end (there may be old gunk there).
10858 */
10859 while (offs < buf->dtb_size)
10860 tomax[offs++] = 0;
10861
10862 /*
10863 * We need to set our offset to zero. And because we
10864 * are wrapping, we need to set the bit indicating as
10865 * much. We can also adjust our needed space back
10866 * down to the space required by the ECB -- we know
10867 * that the top of the buffer is aligned.
10868 */
10869 offs = 0;
10870 total = needed;
10871 buf->dtb_flags |= DTRACEBUF_WRAPPED;
10872 } else {
10873 /*
10874 * There is room for us in the buffer, so we simply
10875 * need to check the wrapped offset.
10876 */
10877 if (woffs < offs) {
10878 /*
10879 * The wrapped offset is less than the offset.
10880 * This can happen if we allocated buffer space
10881 * that induced a wrap, but then we didn't
10882 * subsequently take the space due to an error
10883 * or false predicate evaluation. This is
10884 * okay; we know that _this_ allocation isn't
10885 * going to induce a wrap. We still can't
10886 * reset the wrapped offset to be zero,
10887 * however: the space may have been trashed in
10888 * the previous failed probe attempt. But at
10889 * least the wrapped offset doesn't need to
10890 * be adjusted at all...
10891 */
10892 goto out;
10893 }
10894 }
10895
10896 while (offs + total > woffs) {
10897 dtrace_epid_t epid = *(uint32_t *)(tomax + woffs);
10898 size_t size;
10899
10900 if (epid == DTRACE_EPIDNONE) {
10901 size = sizeof (uint32_t);
10902 } else {
10903 ASSERT(epid <= state->dts_necbs);
10904 ASSERT(state->dts_ecbs[epid - 1] != NULL);
10905
10906 size = state->dts_ecbs[epid - 1]->dte_size;
10907 }
10908
10909 ASSERT(woffs + size <= buf->dtb_size);
10910 ASSERT(size != 0);
10911
10912 if (woffs + size == buf->dtb_size) {
10913 /*
10914 * We've reached the end of the buffer; we want
10915 * to set the wrapped offset to 0 and break
10916 * out. However, if the offs is 0, then we're
10917 * in a strange edge-condition: the amount of
10918 * space that we want to reserve plus the size
10919 * of the record that we're overwriting is
10920 * greater than the size of the buffer. This
10921 * is problematic because if we reserve the
10922 * space but subsequently don't consume it (due
10923 * to a failed predicate or error) the wrapped
10924 * offset will be 0 -- yet the EPID at offset 0
10925 * will not be committed. This situation is
10926 * relatively easy to deal with: if we're in
10927 * this case, the buffer is indistinguishable
10928 * from one that hasn't wrapped; we need only
10929 * finish the job by clearing the wrapped bit,
10930 * explicitly setting the offset to be 0, and
10931 * zero'ing out the old data in the buffer.
10932 */
10933 if (offs == 0) {
10934 buf->dtb_flags &= ~DTRACEBUF_WRAPPED;
10935 buf->dtb_offset = 0;
10936 woffs = total;
10937
10938 while (woffs < buf->dtb_size)
10939 tomax[woffs++] = 0;
10940 }
10941
10942 woffs = 0;
10943 break;
10944 }
10945
10946 woffs += size;
10947 }
10948
10949 /*
10950 * We have a wrapped offset. It may be that the wrapped offset
10951 * has become zero -- that's okay.
10952 */
10953 buf->dtb_xamot_offset = woffs;
10954 }
10955
10956out:
10957 /*
10958 * Now we can plow the buffer with any necessary padding.
10959 */
10960 while (offs & (align - 1)) {
10961 /*
10962 * Assert that our alignment is off by a number which
10963 * is itself sizeof (uint32_t) aligned.
10964 */
10965 ASSERT(!((align - (offs & (align - 1))) &
10966 (sizeof (uint32_t) - 1)));
10967 DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE);
10968 offs += sizeof (uint32_t);
10969 }
10970
10971 if (buf->dtb_flags & DTRACEBUF_FILL) {
10972 if (offs + needed > buf->dtb_size - state->dts_reserve) {
10973 buf->dtb_flags |= DTRACEBUF_FULL;
10974 return (-1);
10975 }
10976 }
10977
10978 if (mstate == NULL)
10979 return (offs);
10980
10981 /*
10982 * For ring buffers and fill buffers, the scratch space is always
10983 * the inactive buffer.
10984 */
10985 mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot;
10986 mstate->dtms_scratch_size = buf->dtb_size;
10987 mstate->dtms_scratch_ptr = mstate->dtms_scratch_base;
10988
10989 return (offs);
10990}
10991
10992static void
10993dtrace_buffer_polish(dtrace_buffer_t *buf)
10994{
10995 ASSERT(buf->dtb_flags & DTRACEBUF_RING);
10996 ASSERT(MUTEX_HELD(&dtrace_lock));
10997
10998 if (!(buf->dtb_flags & DTRACEBUF_WRAPPED))
10999 return;
11000
11001 /*
11002 * We need to polish the ring buffer. There are three cases:
11003 *
11004 * - The first (and presumably most common) is that there is no gap
11005 * between the buffer offset and the wrapped offset. In this case,
11006 * there is nothing in the buffer that isn't valid data; we can
11007 * mark the buffer as polished and return.
11008 *
11009 * - The second (less common than the first but still more common
11010 * than the third) is that there is a gap between the buffer offset
11011 * and the wrapped offset, and the wrapped offset is larger than the
11012 * buffer offset. This can happen because of an alignment issue, or
11013 * can happen because of a call to dtrace_buffer_reserve() that
11014 * didn't subsequently consume the buffer space. In this case,
11015 * we need to zero the data from the buffer offset to the wrapped
11016 * offset.
11017 *
11018 * - The third (and least common) is that there is a gap between the
11019 * buffer offset and the wrapped offset, but the wrapped offset is
11020 * _less_ than the buffer offset. This can only happen because a
11021 * call to dtrace_buffer_reserve() induced a wrap, but the space
11022 * was not subsequently consumed. In this case, we need to zero the
11023 * space from the offset to the end of the buffer _and_ from the
11024 * top of the buffer to the wrapped offset.
11025 */
11026 if (buf->dtb_offset < buf->dtb_xamot_offset) {
11027 bzero(buf->dtb_tomax + buf->dtb_offset,
11028 buf->dtb_xamot_offset - buf->dtb_offset);
11029 }
11030
11031 if (buf->dtb_offset > buf->dtb_xamot_offset) {
11032 bzero(buf->dtb_tomax + buf->dtb_offset,
11033 buf->dtb_size - buf->dtb_offset);
11034 bzero(buf->dtb_tomax, buf->dtb_xamot_offset);
11035 }
11036}
11037
11038static void
11039dtrace_buffer_free(dtrace_buffer_t *bufs)
11040{
11041 int i;
11042
11043 for (i = 0; i < NCPU; i++) {
11044 dtrace_buffer_t *buf = &bufs[i];
11045
11046 if (buf->dtb_tomax == NULL) {
11047 ASSERT(buf->dtb_xamot == NULL);
11048 ASSERT(buf->dtb_size == 0);
11049 continue;
11050 }
11051
11052 if (buf->dtb_xamot != NULL) {
11053 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
11054 kmem_free(buf->dtb_xamot, buf->dtb_size);
11055 }
11056
11057 kmem_free(buf->dtb_tomax, buf->dtb_size);
11058 buf->dtb_size = 0;
11059 buf->dtb_tomax = NULL;
11060 buf->dtb_xamot = NULL;
11061 }
11062}
11063
11064/*
11065 * DTrace Enabling Functions
11066 */
11067static dtrace_enabling_t *
11068dtrace_enabling_create(dtrace_vstate_t *vstate)
11069{
11070 dtrace_enabling_t *enab;
11071
11072 enab = kmem_zalloc(sizeof (dtrace_enabling_t), KM_SLEEP);
11073 enab->dten_vstate = vstate;
11074
11075 return (enab);
11076}
11077
11078static void
11079dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb)
11080{
11081 dtrace_ecbdesc_t **ndesc;
11082 size_t osize, nsize;
11083
11084 /*
11085 * We can't add to enablings after we've enabled them, or after we've
11086 * retained them.
11087 */
11088 ASSERT(enab->dten_probegen == 0);
11089 ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL);
11090
11091 if (enab->dten_ndesc < enab->dten_maxdesc) {
11092 enab->dten_desc[enab->dten_ndesc++] = ecb;
11093 return;
11094 }
11095
11096 osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *);
11097
11098 if (enab->dten_maxdesc == 0) {
11099 enab->dten_maxdesc = 1;
11100 } else {
11101 enab->dten_maxdesc <<= 1;
11102 }
11103
11104 ASSERT(enab->dten_ndesc < enab->dten_maxdesc);
11105
11106 nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *);
11107 ndesc = kmem_zalloc(nsize, KM_SLEEP);
11108 bcopy(enab->dten_desc, ndesc, osize);
11109 if (enab->dten_desc != NULL)
11110 kmem_free(enab->dten_desc, osize);
11111
11112 enab->dten_desc = ndesc;
11113 enab->dten_desc[enab->dten_ndesc++] = ecb;
11114}
11115
11116static void
11117dtrace_enabling_addlike(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb,
11118 dtrace_probedesc_t *pd)
11119{
11120 dtrace_ecbdesc_t *new;
11121 dtrace_predicate_t *pred;
11122 dtrace_actdesc_t *act;
11123
11124 /*
11125 * We're going to create a new ECB description that matches the
11126 * specified ECB in every way, but has the specified probe description.
11127 */
11128 new = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP);
11129
11130 if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL)
11131 dtrace_predicate_hold(pred);
11132
11133 for (act = ecb->dted_action; act != NULL; act = act->dtad_next)
11134 dtrace_actdesc_hold(act);
11135
11136 new->dted_action = ecb->dted_action;
11137 new->dted_pred = ecb->dted_pred;
11138 new->dted_probe = *pd;
11139 new->dted_uarg = ecb->dted_uarg;
11140
11141 dtrace_enabling_add(enab, new);
11142}
11143
11144static void
11145dtrace_enabling_dump(dtrace_enabling_t *enab)
11146{
11147 int i;
11148
11149 for (i = 0; i < enab->dten_ndesc; i++) {
11150 dtrace_probedesc_t *desc = &enab->dten_desc[i]->dted_probe;
11151
11152 cmn_err(CE_NOTE, "enabling probe %d (%s:%s:%s:%s)", i,
11153 desc->dtpd_provider, desc->dtpd_mod,
11154 desc->dtpd_func, desc->dtpd_name);
11155 }
11156}
11157
11158static void
11159dtrace_enabling_destroy(dtrace_enabling_t *enab)
11160{
11161 int i;
11162 dtrace_ecbdesc_t *ep;
11163 dtrace_vstate_t *vstate = enab->dten_vstate;
11164
11165 ASSERT(MUTEX_HELD(&dtrace_lock));
11166
11167 for (i = 0; i < enab->dten_ndesc; i++) {
11168 dtrace_actdesc_t *act, *next;
11169 dtrace_predicate_t *pred;
11170
11171 ep = enab->dten_desc[i];
11172
11173 if ((pred = ep->dted_pred.dtpdd_predicate) != NULL)
11174 dtrace_predicate_release(pred, vstate);
11175
11176 for (act = ep->dted_action; act != NULL; act = next) {
11177 next = act->dtad_next;
11178 dtrace_actdesc_release(act, vstate);
11179 }
11180
11181 kmem_free(ep, sizeof (dtrace_ecbdesc_t));
11182 }
11183
11184 if (enab->dten_desc != NULL)
11185 kmem_free(enab->dten_desc,
11186 enab->dten_maxdesc * sizeof (dtrace_enabling_t *));
11187
11188 /*
11189 * If this was a retained enabling, decrement the dts_nretained count
11190 * and take it off of the dtrace_retained list.
11191 */
11192 if (enab->dten_prev != NULL || enab->dten_next != NULL ||
11193 dtrace_retained == enab) {
11194 ASSERT(enab->dten_vstate->dtvs_state != NULL);
11195 ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0);
11196 enab->dten_vstate->dtvs_state->dts_nretained--;
11197 }
11198
11199 if (enab->dten_prev == NULL) {
11200 if (dtrace_retained == enab) {
11201 dtrace_retained = enab->dten_next;
11202
11203 if (dtrace_retained != NULL)
11204 dtrace_retained->dten_prev = NULL;
11205 }
11206 } else {
11207 ASSERT(enab != dtrace_retained);
11208 ASSERT(dtrace_retained != NULL);
11209 enab->dten_prev->dten_next = enab->dten_next;
11210 }
11211
11212 if (enab->dten_next != NULL) {
11213 ASSERT(dtrace_retained != NULL);
11214 enab->dten_next->dten_prev = enab->dten_prev;
11215 }
11216
11217 kmem_free(enab, sizeof (dtrace_enabling_t));
11218}
11219
11220static int
11221dtrace_enabling_retain(dtrace_enabling_t *enab)
11222{
11223 dtrace_state_t *state;
11224
11225 ASSERT(MUTEX_HELD(&dtrace_lock));
11226 ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL);
11227 ASSERT(enab->dten_vstate != NULL);
11228
11229 state = enab->dten_vstate->dtvs_state;
11230 ASSERT(state != NULL);
11231
11232 /*
11233 * We only allow each state to retain dtrace_retain_max enablings.
11234 */
11235 if (state->dts_nretained >= dtrace_retain_max)
11236 return (ENOSPC);
11237
11238 state->dts_nretained++;
11239
11240 if (dtrace_retained == NULL) {
11241 dtrace_retained = enab;
11242 return (0);
11243 }
11244
11245 enab->dten_next = dtrace_retained;
11246 dtrace_retained->dten_prev = enab;
11247 dtrace_retained = enab;
11248
11249 return (0);
11250}
11251
11252static int
11253dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match,
11254 dtrace_probedesc_t *create)
11255{
11256 dtrace_enabling_t *new, *enab;
11257 int found = 0, err = ENOENT;
11258
11259 ASSERT(MUTEX_HELD(&dtrace_lock));
11260 ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN);
11261 ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN);
11262 ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN);
11263 ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN);
11264
11265 new = dtrace_enabling_create(&state->dts_vstate);
11266
11267 /*
11268 * Iterate over all retained enablings, looking for enablings that
11269 * match the specified state.
11270 */
11271 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11272 int i;
11273
11274 /*
11275 * dtvs_state can only be NULL for helper enablings -- and
11276 * helper enablings can't be retained.
11277 */
11278 ASSERT(enab->dten_vstate->dtvs_state != NULL);
11279
11280 if (enab->dten_vstate->dtvs_state != state)
11281 continue;
11282
11283 /*
11284 * Now iterate over each probe description; we're looking for
11285 * an exact match to the specified probe description.
11286 */
11287 for (i = 0; i < enab->dten_ndesc; i++) {
11288 dtrace_ecbdesc_t *ep = enab->dten_desc[i];
11289 dtrace_probedesc_t *pd = &ep->dted_probe;
11290
11291 if (strcmp(pd->dtpd_provider, match->dtpd_provider))
11292 continue;
11293
11294 if (strcmp(pd->dtpd_mod, match->dtpd_mod))
11295 continue;
11296
11297 if (strcmp(pd->dtpd_func, match->dtpd_func))
11298 continue;
11299
11300 if (strcmp(pd->dtpd_name, match->dtpd_name))
11301 continue;
11302
11303 /*
11304 * We have a winning probe! Add it to our growing
11305 * enabling.
11306 */
11307 found = 1;
11308 dtrace_enabling_addlike(new, ep, create);
11309 }
11310 }
11311
11312 if (!found || (err = dtrace_enabling_retain(new)) != 0) {
11313 dtrace_enabling_destroy(new);
11314 return (err);
11315 }
11316
11317 return (0);
11318}
11319
11320static void
11321dtrace_enabling_retract(dtrace_state_t *state)
11322{
11323 dtrace_enabling_t *enab, *next;
11324
11325 ASSERT(MUTEX_HELD(&dtrace_lock));
11326
11327 /*
11328 * Iterate over all retained enablings, destroy the enablings retained
11329 * for the specified state.
11330 */
11331 for (enab = dtrace_retained; enab != NULL; enab = next) {
11332 next = enab->dten_next;
11333
11334 /*
11335 * dtvs_state can only be NULL for helper enablings -- and
11336 * helper enablings can't be retained.
11337 */
11338 ASSERT(enab->dten_vstate->dtvs_state != NULL);
11339
11340 if (enab->dten_vstate->dtvs_state == state) {
11341 ASSERT(state->dts_nretained > 0);
11342 dtrace_enabling_destroy(enab);
11343 }
11344 }
11345
11346 ASSERT(state->dts_nretained == 0);
11347}
11348
11349static int
11350dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched)
11351{
11352 int i = 0;
11353 int matched = 0;
11354
11355 ASSERT(MUTEX_HELD(&cpu_lock));
11356 ASSERT(MUTEX_HELD(&dtrace_lock));
11357
11358 for (i = 0; i < enab->dten_ndesc; i++) {
11359 dtrace_ecbdesc_t *ep = enab->dten_desc[i];
11360
11361 enab->dten_current = ep;
11362 enab->dten_error = 0;
11363
11364 matched += dtrace_probe_enable(&ep->dted_probe, enab);
11365
11366 if (enab->dten_error != 0) {
11367 /*
11368 * If we get an error half-way through enabling the
11369 * probes, we kick out -- perhaps with some number of
11370 * them enabled. Leaving enabled probes enabled may
11371 * be slightly confusing for user-level, but we expect
11372 * that no one will attempt to actually drive on in
11373 * the face of such errors. If this is an anonymous
11374 * enabling (indicated with a NULL nmatched pointer),
11375 * we cmn_err() a message. We aren't expecting to
11376 * get such an error -- such as it can exist at all,
11377 * it would be a result of corrupted DOF in the driver
11378 * properties.
11379 */
11380 if (nmatched == NULL) {
11381 cmn_err(CE_WARN, "dtrace_enabling_match() "
11382 "error on %p: %d", (void *)ep,
11383 enab->dten_error);
11384 }
11385
11386 return (enab->dten_error);
11387 }
11388 }
11389
11390 enab->dten_probegen = dtrace_probegen;
11391 if (nmatched != NULL)
11392 *nmatched = matched;
11393
11394 return (0);
11395}
11396
11397static void
11398dtrace_enabling_matchall(void)
11399{
11400 dtrace_enabling_t *enab;
11401
11402 mutex_enter(&cpu_lock);
11403 mutex_enter(&dtrace_lock);
11404
11405 /*
11406 * Iterate over all retained enablings to see if any probes match
11407 * against them. We only perform this operation on enablings for which
11408 * we have sufficient permissions by virtue of being in the global zone
11409 * or in the same zone as the DTrace client. Because we can be called
11410 * after dtrace_detach() has been called, we cannot assert that there
11411 * are retained enablings. We can safely load from dtrace_retained,
11412 * however: the taskq_destroy() at the end of dtrace_detach() will
11413 * block pending our completion.
11414 */
11415 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11416#if defined(sun)
11417 cred_t *cr = enab->dten_vstate->dtvs_state->dts_cred.dcr_cred;
11418
11419 if (INGLOBALZONE(curproc) || getzoneid() == crgetzoneid(cr))
11420#endif
11421 (void) dtrace_enabling_match(enab, NULL);
11422 }
11423
11424 mutex_exit(&dtrace_lock);
11425 mutex_exit(&cpu_lock);
11426}
11427
11428/*
11429 * If an enabling is to be enabled without having matched probes (that is, if
11430 * dtrace_state_go() is to be called on the underlying dtrace_state_t), the
11431 * enabling must be _primed_ by creating an ECB for every ECB description.
11432 * This must be done to assure that we know the number of speculations, the
11433 * number of aggregations, the minimum buffer size needed, etc. before we
11434 * transition out of DTRACE_ACTIVITY_INACTIVE. To do this without actually
11435 * enabling any probes, we create ECBs for every ECB decription, but with a
11436 * NULL probe -- which is exactly what this function does.
11437 */
11438static void
11439dtrace_enabling_prime(dtrace_state_t *state)
11440{
11441 dtrace_enabling_t *enab;
11442 int i;
11443
11444 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) {
11445 ASSERT(enab->dten_vstate->dtvs_state != NULL);
11446
11447 if (enab->dten_vstate->dtvs_state != state)
11448 continue;
11449
11450 /*
11451 * We don't want to prime an enabling more than once, lest
11452 * we allow a malicious user to induce resource exhaustion.
11453 * (The ECBs that result from priming an enabling aren't
11454 * leaked -- but they also aren't deallocated until the
11455 * consumer state is destroyed.)
11456 */
11457 if (enab->dten_primed)
11458 continue;
11459
11460 for (i = 0; i < enab->dten_ndesc; i++) {
11461 enab->dten_current = enab->dten_desc[i];
11462 (void) dtrace_probe_enable(NULL, enab);
11463 }
11464
11465 enab->dten_primed = 1;
11466 }
11467}
11468
11469/*
11470 * Called to indicate that probes should be provided due to retained
11471 * enablings. This is implemented in terms of dtrace_probe_provide(), but it
11472 * must take an initial lap through the enabling calling the dtps_provide()
11473 * entry point explicitly to allow for autocreated probes.
11474 */
11475static void
11476dtrace_enabling_provide(dtrace_provider_t *prv)
11477{
11478 int i, all = 0;
11479 dtrace_probedesc_t desc;
11480
11481 ASSERT(MUTEX_HELD(&dtrace_lock));
11482 ASSERT(MUTEX_HELD(&dtrace_provider_lock));
11483
11484 if (prv == NULL) {
11485 all = 1;
11486 prv = dtrace_provider;
11487 }
11488
11489 do {
11490 dtrace_enabling_t *enab = dtrace_retained;
11491 void *parg = prv->dtpv_arg;
11492
11493 for (; enab != NULL; enab = enab->dten_next) {
11494 for (i = 0; i < enab->dten_ndesc; i++) {
11495 desc = enab->dten_desc[i]->dted_probe;
11496 mutex_exit(&dtrace_lock);
11497 prv->dtpv_pops.dtps_provide(parg, &desc);
11498 mutex_enter(&dtrace_lock);
11499 }
11500 }
11501 } while (all && (prv = prv->dtpv_next) != NULL);
11502
11503 mutex_exit(&dtrace_lock);
11504 dtrace_probe_provide(NULL, all ? NULL : prv);
11505 mutex_enter(&dtrace_lock);
11506}
11507
11508/*
11509 * DTrace DOF Functions
11510 */
11511/*ARGSUSED*/
11512static void
11513dtrace_dof_error(dof_hdr_t *dof, const char *str)
11514{
11515 if (dtrace_err_verbose)
11516 cmn_err(CE_WARN, "failed to process DOF: %s", str);
11517
11518#ifdef DTRACE_ERRDEBUG
11519 dtrace_errdebug(str);
11520#endif
11521}
11522
11523/*
11524 * Create DOF out of a currently enabled state. Right now, we only create
11525 * DOF containing the run-time options -- but this could be expanded to create
11526 * complete DOF representing the enabled state.
11527 */
11528static dof_hdr_t *
11529dtrace_dof_create(dtrace_state_t *state)
11530{
11531 dof_hdr_t *dof;
11532 dof_sec_t *sec;
11533 dof_optdesc_t *opt;
11534 int i, len = sizeof (dof_hdr_t) +
11535 roundup(sizeof (dof_sec_t), sizeof (uint64_t)) +
11536 sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
11537
11538 ASSERT(MUTEX_HELD(&dtrace_lock));
11539
11540 dof = kmem_zalloc(len, KM_SLEEP);
11541 dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0;
11542 dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1;
11543 dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2;
11544 dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3;
11545
11546 dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE;
11547 dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE;
11548 dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION;
11549 dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION;
11550 dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS;
11551 dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS;
11552
11553 dof->dofh_flags = 0;
11554 dof->dofh_hdrsize = sizeof (dof_hdr_t);
11555 dof->dofh_secsize = sizeof (dof_sec_t);
11556 dof->dofh_secnum = 1; /* only DOF_SECT_OPTDESC */
11557 dof->dofh_secoff = sizeof (dof_hdr_t);
11558 dof->dofh_loadsz = len;
11559 dof->dofh_filesz = len;
11560 dof->dofh_pad = 0;
11561
11562 /*
11563 * Fill in the option section header...
11564 */
11565 sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t));
11566 sec->dofs_type = DOF_SECT_OPTDESC;
11567 sec->dofs_align = sizeof (uint64_t);
11568 sec->dofs_flags = DOF_SECF_LOAD;
11569 sec->dofs_entsize = sizeof (dof_optdesc_t);
11570
11571 opt = (dof_optdesc_t *)((uintptr_t)sec +
11572 roundup(sizeof (dof_sec_t), sizeof (uint64_t)));
11573
11574 sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof;
11575 sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
11576
11577 for (i = 0; i < DTRACEOPT_MAX; i++) {
11578 opt[i].dofo_option = i;
11579 opt[i].dofo_strtab = DOF_SECIDX_NONE;
11580 opt[i].dofo_value = state->dts_options[i];
11581 }
11582
11583 return (dof);
11584}
11585
11586static dof_hdr_t *
11587dtrace_dof_copyin(uintptr_t uarg, int *errp)
11588{
11589 dof_hdr_t hdr, *dof;
11590
11591 ASSERT(!MUTEX_HELD(&dtrace_lock));
11592
11593 /*
11594 * First, we're going to copyin() the sizeof (dof_hdr_t).
11595 */
11596 if (copyin((void *)uarg, &hdr, sizeof (hdr)) != 0) {
11597 dtrace_dof_error(NULL, "failed to copyin DOF header");
11598 *errp = EFAULT;
11599 return (NULL);
11600 }
11601
11602 /*
11603 * Now we'll allocate the entire DOF and copy it in -- provided
11604 * that the length isn't outrageous.
11605 */
11606 if (hdr.dofh_loadsz >= dtrace_dof_maxsize) {
11607 dtrace_dof_error(&hdr, "load size exceeds maximum");
11608 *errp = E2BIG;
11609 return (NULL);
11610 }
11611
11612 if (hdr.dofh_loadsz < sizeof (hdr)) {
11613 dtrace_dof_error(&hdr, "invalid load size");
11614 *errp = EINVAL;
11615 return (NULL);
11616 }
11617
11618 dof = kmem_alloc(hdr.dofh_loadsz, KM_SLEEP);
11619
11620 if (copyin((void *)uarg, dof, hdr.dofh_loadsz) != 0) {
11621 kmem_free(dof, hdr.dofh_loadsz);
11622 *errp = EFAULT;
11623 return (NULL);
11624 }
11625
11626 return (dof);
11627}
11628
11629#if !defined(sun)
11630static __inline uchar_t
11631dtrace_dof_char(char c) {
11632 switch (c) {
11633 case '0':
11634 case '1':
11635 case '2':
11636 case '3':
11637 case '4':
11638 case '5':
11639 case '6':
11640 case '7':
11641 case '8':
11642 case '9':
11643 return (c - '0');
11644 case 'A':
11645 case 'B':
11646 case 'C':
11647 case 'D':
11648 case 'E':
11649 case 'F':
11650 return (c - 'A' + 10);
11651 case 'a':
11652 case 'b':
11653 case 'c':
11654 case 'd':
11655 case 'e':
11656 case 'f':
11657 return (c - 'a' + 10);
11658 }
11659 /* Should not reach here. */
11660 return (0);
11661}
11662#endif
11663
11664static dof_hdr_t *
11665dtrace_dof_property(const char *name)
11666{
11667 uchar_t *buf;
11668 uint64_t loadsz;
11669 unsigned int len, i;
11670 dof_hdr_t *dof;
11671
11672#if defined(sun)
11673 /*
11674 * Unfortunately, array of values in .conf files are always (and
11675 * only) interpreted to be integer arrays. We must read our DOF
11676 * as an integer array, and then squeeze it into a byte array.
11677 */
11678 if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0,
11679 (char *)name, (int **)&buf, &len) != DDI_PROP_SUCCESS)
11680 return (NULL);
11681
11682 for (i = 0; i < len; i++)
11683 buf[i] = (uchar_t)(((int *)buf)[i]);
11684
11685 if (len < sizeof (dof_hdr_t)) {
11686 ddi_prop_free(buf);
11687 dtrace_dof_error(NULL, "truncated header");
11688 return (NULL);
11689 }
11690
11691 if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) {
11692 ddi_prop_free(buf);
11693 dtrace_dof_error(NULL, "truncated DOF");
11694 return (NULL);
11695 }
11696
11697 if (loadsz >= dtrace_dof_maxsize) {
11698 ddi_prop_free(buf);
11699 dtrace_dof_error(NULL, "oversized DOF");
11700 return (NULL);
11701 }
11702
11703 dof = kmem_alloc(loadsz, KM_SLEEP);
11704 bcopy(buf, dof, loadsz);
11705 ddi_prop_free(buf);
11706#else
11707 char *p;
11708 char *p_env;
11709
11710 if ((p_env = getenv(name)) == NULL)
11711 return (NULL);
11712
11713 len = strlen(p_env) / 2;
11714
11715 buf = kmem_alloc(len, KM_SLEEP);
11716
11717 dof = (dof_hdr_t *) buf;
11718
11719 p = p_env;
11720
11721 for (i = 0; i < len; i++) {
11722 buf[i] = (dtrace_dof_char(p[0]) << 4) |
11723 dtrace_dof_char(p[1]);
11724 p += 2;
11725 }
11726
11727 freeenv(p_env);
11728
11729 if (len < sizeof (dof_hdr_t)) {
11730 kmem_free(buf, 0);
11731 dtrace_dof_error(NULL, "truncated header");
11732 return (NULL);
11733 }
11734
11735 if (len < (loadsz = dof->dofh_loadsz)) {
11736 kmem_free(buf, 0);
11737 dtrace_dof_error(NULL, "truncated DOF");
11738 return (NULL);
11739 }
11740
11741 if (loadsz >= dtrace_dof_maxsize) {
11742 kmem_free(buf, 0);
11743 dtrace_dof_error(NULL, "oversized DOF");
11744 return (NULL);
11745 }
11746#endif
11747
11748 return (dof);
11749}
11750
11751static void
11752dtrace_dof_destroy(dof_hdr_t *dof)
11753{
11754 kmem_free(dof, dof->dofh_loadsz);
11755}
11756
11757/*
11758 * Return the dof_sec_t pointer corresponding to a given section index. If the
11759 * index is not valid, dtrace_dof_error() is called and NULL is returned. If
11760 * a type other than DOF_SECT_NONE is specified, the header is checked against
11761 * this type and NULL is returned if the types do not match.
11762 */
11763static dof_sec_t *
11764dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, dof_secidx_t i)
11765{
11766 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)
11767 ((uintptr_t)dof + dof->dofh_secoff + i * dof->dofh_secsize);
11768
11769 if (i >= dof->dofh_secnum) {
11770 dtrace_dof_error(dof, "referenced section index is invalid");
11771 return (NULL);
11772 }
11773
11774 if (!(sec->dofs_flags & DOF_SECF_LOAD)) {
11775 dtrace_dof_error(dof, "referenced section is not loadable");
11776 return (NULL);
11777 }
11778
11779 if (type != DOF_SECT_NONE && type != sec->dofs_type) {
11780 dtrace_dof_error(dof, "referenced section is the wrong type");
11781 return (NULL);
11782 }
11783
11784 return (sec);
11785}
11786
11787static dtrace_probedesc_t *
11788dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_probedesc_t *desc)
11789{
11790 dof_probedesc_t *probe;
11791 dof_sec_t *strtab;
11792 uintptr_t daddr = (uintptr_t)dof;
11793 uintptr_t str;
11794 size_t size;
11795
11796 if (sec->dofs_type != DOF_SECT_PROBEDESC) {
11797 dtrace_dof_error(dof, "invalid probe section");
11798 return (NULL);
11799 }
11800
11801 if (sec->dofs_align != sizeof (dof_secidx_t)) {
11802 dtrace_dof_error(dof, "bad alignment in probe description");
11803 return (NULL);
11804 }
11805
11806 if (sec->dofs_offset + sizeof (dof_probedesc_t) > dof->dofh_loadsz) {
11807 dtrace_dof_error(dof, "truncated probe description");
11808 return (NULL);
11809 }
11810
11811 probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset);
11812 strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab);
11813
11814 if (strtab == NULL)
11815 return (NULL);
11816
11817 str = daddr + strtab->dofs_offset;
11818 size = strtab->dofs_size;
11819
11820 if (probe->dofp_provider >= strtab->dofs_size) {
11821 dtrace_dof_error(dof, "corrupt probe provider");
11822 return (NULL);
11823 }
11824
11825 (void) strncpy(desc->dtpd_provider,
11826 (char *)(str + probe->dofp_provider),
11827 MIN(DTRACE_PROVNAMELEN - 1, size - probe->dofp_provider));
11828
11829 if (probe->dofp_mod >= strtab->dofs_size) {
11830 dtrace_dof_error(dof, "corrupt probe module");
11831 return (NULL);
11832 }
11833
11834 (void) strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod),
11835 MIN(DTRACE_MODNAMELEN - 1, size - probe->dofp_mod));
11836
11837 if (probe->dofp_func >= strtab->dofs_size) {
11838 dtrace_dof_error(dof, "corrupt probe function");
11839 return (NULL);
11840 }
11841
11842 (void) strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func),
11843 MIN(DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func));
11844
11845 if (probe->dofp_name >= strtab->dofs_size) {
11846 dtrace_dof_error(dof, "corrupt probe name");
11847 return (NULL);
11848 }
11849
11850 (void) strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name),
11851 MIN(DTRACE_NAMELEN - 1, size - probe->dofp_name));
11852
11853 return (desc);
11854}
11855
11856static dtrace_difo_t *
11857dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
11858 cred_t *cr)
11859{
11860 dtrace_difo_t *dp;
11861 size_t ttl = 0;
11862 dof_difohdr_t *dofd;
11863 uintptr_t daddr = (uintptr_t)dof;
11864 size_t max = dtrace_difo_maxsize;
11865 int i, l, n;
11866
11867 static const struct {
11868 int section;
11869 int bufoffs;
11870 int lenoffs;
11871 int entsize;
11872 int align;
11873 const char *msg;
11874 } difo[] = {
11875 { DOF_SECT_DIF, offsetof(dtrace_difo_t, dtdo_buf),
11876 offsetof(dtrace_difo_t, dtdo_len), sizeof (dif_instr_t),
11877 sizeof (dif_instr_t), "multiple DIF sections" },
11878
11879 { DOF_SECT_INTTAB, offsetof(dtrace_difo_t, dtdo_inttab),
11880 offsetof(dtrace_difo_t, dtdo_intlen), sizeof (uint64_t),
11881 sizeof (uint64_t), "multiple integer tables" },
11882
11883 { DOF_SECT_STRTAB, offsetof(dtrace_difo_t, dtdo_strtab),
11884 offsetof(dtrace_difo_t, dtdo_strlen), 0,
11885 sizeof (char), "multiple string tables" },
11886
11887 { DOF_SECT_VARTAB, offsetof(dtrace_difo_t, dtdo_vartab),
11888 offsetof(dtrace_difo_t, dtdo_varlen), sizeof (dtrace_difv_t),
11889 sizeof (uint_t), "multiple variable tables" },
11890
11891 { DOF_SECT_NONE, 0, 0, 0, 0, NULL }
11892 };
11893
11894 if (sec->dofs_type != DOF_SECT_DIFOHDR) {
11895 dtrace_dof_error(dof, "invalid DIFO header section");
11896 return (NULL);
11897 }
11898
11899 if (sec->dofs_align != sizeof (dof_secidx_t)) {
11900 dtrace_dof_error(dof, "bad alignment in DIFO header");
11901 return (NULL);
11902 }
11903
11904 if (sec->dofs_size < sizeof (dof_difohdr_t) ||
11905 sec->dofs_size % sizeof (dof_secidx_t)) {
11906 dtrace_dof_error(dof, "bad size in DIFO header");
11907 return (NULL);
11908 }
11909
11910 dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset);
11911 n = (sec->dofs_size - sizeof (*dofd)) / sizeof (dof_secidx_t) + 1;
11912
11913 dp = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP);
11914 dp->dtdo_rtype = dofd->dofd_rtype;
11915
11916 for (l = 0; l < n; l++) {
11917 dof_sec_t *subsec;
11918 void **bufp;
11919 uint32_t *lenp;
11920
11921 if ((subsec = dtrace_dof_sect(dof, DOF_SECT_NONE,
11922 dofd->dofd_links[l])) == NULL)
11923 goto err; /* invalid section link */
11924
11925 if (ttl + subsec->dofs_size > max) {
11926 dtrace_dof_error(dof, "exceeds maximum size");
11927 goto err;
11928 }
11929
11930 ttl += subsec->dofs_size;
11931
11932 for (i = 0; difo[i].section != DOF_SECT_NONE; i++) {
11933 if (subsec->dofs_type != difo[i].section)
11934 continue;
11935
11936 if (!(subsec->dofs_flags & DOF_SECF_LOAD)) {
11937 dtrace_dof_error(dof, "section not loaded");
11938 goto err;
11939 }
11940
11941 if (subsec->dofs_align != difo[i].align) {
11942 dtrace_dof_error(dof, "bad alignment");
11943 goto err;
11944 }
11945
11946 bufp = (void **)((uintptr_t)dp + difo[i].bufoffs);
11947 lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs);
11948
11949 if (*bufp != NULL) {
11950 dtrace_dof_error(dof, difo[i].msg);
11951 goto err;
11952 }
11953
11954 if (difo[i].entsize != subsec->dofs_entsize) {
11955 dtrace_dof_error(dof, "entry size mismatch");
11956 goto err;
11957 }
11958
11959 if (subsec->dofs_entsize != 0 &&
11960 (subsec->dofs_size % subsec->dofs_entsize) != 0) {
11961 dtrace_dof_error(dof, "corrupt entry size");
11962 goto err;
11963 }
11964
11965 *lenp = subsec->dofs_size;
11966 *bufp = kmem_alloc(subsec->dofs_size, KM_SLEEP);
11967 bcopy((char *)(uintptr_t)(daddr + subsec->dofs_offset),
11968 *bufp, subsec->dofs_size);
11969
11970 if (subsec->dofs_entsize != 0)
11971 *lenp /= subsec->dofs_entsize;
11972
11973 break;
11974 }
11975
11976 /*
11977 * If we encounter a loadable DIFO sub-section that is not
11978 * known to us, assume this is a broken program and fail.
11979 */
11980 if (difo[i].section == DOF_SECT_NONE &&
11981 (subsec->dofs_flags & DOF_SECF_LOAD)) {
11982 dtrace_dof_error(dof, "unrecognized DIFO subsection");
11983 goto err;
11984 }
11985 }
11986
11987 if (dp->dtdo_buf == NULL) {
11988 /*
11989 * We can't have a DIF object without DIF text.
11990 */
11991 dtrace_dof_error(dof, "missing DIF text");
11992 goto err;
11993 }
11994
11995 /*
11996 * Before we validate the DIF object, run through the variable table
11997 * looking for the strings -- if any of their size are under, we'll set
11998 * their size to be the system-wide default string size. Note that
11999 * this should _not_ happen if the "strsize" option has been set --
12000 * in this case, the compiler should have set the size to reflect the
12001 * setting of the option.
12002 */
12003 for (i = 0; i < dp->dtdo_varlen; i++) {
12004 dtrace_difv_t *v = &dp->dtdo_vartab[i];
12005 dtrace_diftype_t *t = &v->dtdv_type;
12006
12007 if (v->dtdv_id < DIF_VAR_OTHER_UBASE)
12008 continue;
12009
12010 if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0)
12011 t->dtdt_size = dtrace_strsize_default;
12012 }
12013
12014 if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0)
12015 goto err;
12016
12017 dtrace_difo_init(dp, vstate);
12018 return (dp);
12019
12020err:
12021 kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t));
12022 kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t));
12023 kmem_free(dp->dtdo_strtab, dp->dtdo_strlen);
12024 kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t));
12025
12026 kmem_free(dp, sizeof (dtrace_difo_t));
12027 return (NULL);
12028}
12029
12030static dtrace_predicate_t *
12031dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12032 cred_t *cr)
12033{
12034 dtrace_difo_t *dp;
12035
12036 if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL)
12037 return (NULL);
12038
12039 return (dtrace_predicate_create(dp));
12040}
12041
12042static dtrace_actdesc_t *
12043dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12044 cred_t *cr)
12045{
12046 dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next;
12047 dof_actdesc_t *desc;
12048 dof_sec_t *difosec;
12049 size_t offs;
12050 uintptr_t daddr = (uintptr_t)dof;
12051 uint64_t arg;
12052 dtrace_actkind_t kind;
12053
12054 if (sec->dofs_type != DOF_SECT_ACTDESC) {
12055 dtrace_dof_error(dof, "invalid action section");
12056 return (NULL);
12057 }
12058
12059 if (sec->dofs_offset + sizeof (dof_actdesc_t) > dof->dofh_loadsz) {
12060 dtrace_dof_error(dof, "truncated action description");
12061 return (NULL);
12062 }
12063
12064 if (sec->dofs_align != sizeof (uint64_t)) {
12065 dtrace_dof_error(dof, "bad alignment in action description");
12066 return (NULL);
12067 }
12068
12069 if (sec->dofs_size < sec->dofs_entsize) {
12070 dtrace_dof_error(dof, "section entry size exceeds total size");
12071 return (NULL);
12072 }
12073
12074 if (sec->dofs_entsize != sizeof (dof_actdesc_t)) {
12075 dtrace_dof_error(dof, "bad entry size in action description");
12076 return (NULL);
12077 }
12078
12079 if (sec->dofs_size / sec->dofs_entsize > dtrace_actions_max) {
12080 dtrace_dof_error(dof, "actions exceed dtrace_actions_max");
12081 return (NULL);
12082 }
12083
12084 for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) {
12085 desc = (dof_actdesc_t *)(daddr +
12086 (uintptr_t)sec->dofs_offset + offs);
12087 kind = (dtrace_actkind_t)desc->dofa_kind;
12088
12089 if (DTRACEACT_ISPRINTFLIKE(kind) &&
12090 (kind != DTRACEACT_PRINTA ||
12091 desc->dofa_strtab != DOF_SECIDX_NONE)) {
12092 dof_sec_t *strtab;
12093 char *str, *fmt;
12094 uint64_t i;
12095
12096 /*
12097 * printf()-like actions must have a format string.
12098 */
12099 if ((strtab = dtrace_dof_sect(dof,
12100 DOF_SECT_STRTAB, desc->dofa_strtab)) == NULL)
12101 goto err;
12102
12103 str = (char *)((uintptr_t)dof +
12104 (uintptr_t)strtab->dofs_offset);
12105
12106 for (i = desc->dofa_arg; i < strtab->dofs_size; i++) {
12107 if (str[i] == '\0')
12108 break;
12109 }
12110
12111 if (i >= strtab->dofs_size) {
12112 dtrace_dof_error(dof, "bogus format string");
12113 goto err;
12114 }
12115
12116 if (i == desc->dofa_arg) {
12117 dtrace_dof_error(dof, "empty format string");
12118 goto err;
12119 }
12120
12121 i -= desc->dofa_arg;
12122 fmt = kmem_alloc(i + 1, KM_SLEEP);
12123 bcopy(&str[desc->dofa_arg], fmt, i + 1);
12124 arg = (uint64_t)(uintptr_t)fmt;
12125 } else {
12126 if (kind == DTRACEACT_PRINTA) {
12127 ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE);
12128 arg = 0;
12129 } else {
12130 arg = desc->dofa_arg;
12131 }
12132 }
12133
12134 act = dtrace_actdesc_create(kind, desc->dofa_ntuple,
12135 desc->dofa_uarg, arg);
12136
12137 if (last != NULL) {
12138 last->dtad_next = act;
12139 } else {
12140 first = act;
12141 }
12142
12143 last = act;
12144
12145 if (desc->dofa_difo == DOF_SECIDX_NONE)
12146 continue;
12147
12148 if ((difosec = dtrace_dof_sect(dof,
12149 DOF_SECT_DIFOHDR, desc->dofa_difo)) == NULL)
12150 goto err;
12151
12152 act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr);
12153
12154 if (act->dtad_difo == NULL)
12155 goto err;
12156 }
12157
12158 ASSERT(first != NULL);
12159 return (first);
12160
12161err:
12162 for (act = first; act != NULL; act = next) {
12163 next = act->dtad_next;
12164 dtrace_actdesc_release(act, vstate);
12165 }
12166
12167 return (NULL);
12168}
12169
12170static dtrace_ecbdesc_t *
12171dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate,
12172 cred_t *cr)
12173{
12174 dtrace_ecbdesc_t *ep;
12175 dof_ecbdesc_t *ecb;
12176 dtrace_probedesc_t *desc;
12177 dtrace_predicate_t *pred = NULL;
12178
12179 if (sec->dofs_size < sizeof (dof_ecbdesc_t)) {
12180 dtrace_dof_error(dof, "truncated ECB description");
12181 return (NULL);
12182 }
12183
12184 if (sec->dofs_align != sizeof (uint64_t)) {
12185 dtrace_dof_error(dof, "bad alignment in ECB description");
12186 return (NULL);
12187 }
12188
12189 ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset);
12190 sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes);
12191
12192 if (sec == NULL)
12193 return (NULL);
12194
12195 ep = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP);
12196 ep->dted_uarg = ecb->dofe_uarg;
12197 desc = &ep->dted_probe;
12198
12199 if (dtrace_dof_probedesc(dof, sec, desc) == NULL)
12200 goto err;
12201
12202 if (ecb->dofe_pred != DOF_SECIDX_NONE) {
12203 if ((sec = dtrace_dof_sect(dof,
12204 DOF_SECT_DIFOHDR, ecb->dofe_pred)) == NULL)
12205 goto err;
12206
12207 if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL)
12208 goto err;
12209
12210 ep->dted_pred.dtpdd_predicate = pred;
12211 }
12212
12213 if (ecb->dofe_actions != DOF_SECIDX_NONE) {
12214 if ((sec = dtrace_dof_sect(dof,
12215 DOF_SECT_ACTDESC, ecb->dofe_actions)) == NULL)
12216 goto err;
12217
12218 ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr);
12219
12220 if (ep->dted_action == NULL)
12221 goto err;
12222 }
12223
12224 return (ep);
12225
12226err:
12227 if (pred != NULL)
12228 dtrace_predicate_release(pred, vstate);
12229 kmem_free(ep, sizeof (dtrace_ecbdesc_t));
12230 return (NULL);
12231}
12232
12233/*
12234 * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the
12235 * specified DOF. At present, this amounts to simply adding 'ubase' to the
12236 * site of any user SETX relocations to account for load object base address.
12237 * In the future, if we need other relocations, this function can be extended.
12238 */
12239static int
12240dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase)
12241{
12242 uintptr_t daddr = (uintptr_t)dof;
12243 dof_relohdr_t *dofr =
12244 (dof_relohdr_t *)(uintptr_t)(daddr + sec->dofs_offset);
12245 dof_sec_t *ss, *rs, *ts;
12246 dof_relodesc_t *r;
12247 uint_t i, n;
12248
12249 if (sec->dofs_size < sizeof (dof_relohdr_t) ||
12250 sec->dofs_align != sizeof (dof_secidx_t)) {
12251 dtrace_dof_error(dof, "invalid relocation header");
12252 return (-1);
12253 }
12254
12255 ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab);
12256 rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec);
12257 ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec);
12258
12259 if (ss == NULL || rs == NULL || ts == NULL)
12260 return (-1); /* dtrace_dof_error() has been called already */
12261
12262 if (rs->dofs_entsize < sizeof (dof_relodesc_t) ||
12263 rs->dofs_align != sizeof (uint64_t)) {
12264 dtrace_dof_error(dof, "invalid relocation section");
12265 return (-1);
12266 }
12267
12268 r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset);
12269 n = rs->dofs_size / rs->dofs_entsize;
12270
12271 for (i = 0; i < n; i++) {
12272 uintptr_t taddr = daddr + ts->dofs_offset + r->dofr_offset;
12273
12274 switch (r->dofr_type) {
12275 case DOF_RELO_NONE:
12276 break;
12277 case DOF_RELO_SETX:
12278 if (r->dofr_offset >= ts->dofs_size || r->dofr_offset +
12279 sizeof (uint64_t) > ts->dofs_size) {
12280 dtrace_dof_error(dof, "bad relocation offset");
12281 return (-1);
12282 }
12283
12284 if (!IS_P2ALIGNED(taddr, sizeof (uint64_t))) {
12285 dtrace_dof_error(dof, "misaligned setx relo");
12286 return (-1);
12287 }
12288
12289 *(uint64_t *)taddr += ubase;
12290 break;
12291 default:
12292 dtrace_dof_error(dof, "invalid relocation type");
12293 return (-1);
12294 }
12295
12296 r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize);
12297 }
12298
12299 return (0);
12300}
12301
12302/*
12303 * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated
12304 * header: it should be at the front of a memory region that is at least
12305 * sizeof (dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in
12306 * size. It need not be validated in any other way.
12307 */
12308static int
12309dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr,
12310 dtrace_enabling_t **enabp, uint64_t ubase, int noprobes)
12311{
12312 uint64_t len = dof->dofh_loadsz, seclen;
12313 uintptr_t daddr = (uintptr_t)dof;
12314 dtrace_ecbdesc_t *ep;
12315 dtrace_enabling_t *enab;
12316 uint_t i;
12317
12318 ASSERT(MUTEX_HELD(&dtrace_lock));
12319 ASSERT(dof->dofh_loadsz >= sizeof (dof_hdr_t));
12320
12321 /*
12322 * Check the DOF header identification bytes. In addition to checking
12323 * valid settings, we also verify that unused bits/bytes are zeroed so
12324 * we can use them later without fear of regressing existing binaries.
12325 */
12326 if (bcmp(&dof->dofh_ident[DOF_ID_MAG0],
12327 DOF_MAG_STRING, DOF_MAG_STRLEN) != 0) {
12328 dtrace_dof_error(dof, "DOF magic string mismatch");
12329 return (-1);
12330 }
12331
12332 if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 &&
12333 dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) {
12334 dtrace_dof_error(dof, "DOF has invalid data model");
12335 return (-1);
12336 }
12337
12338 if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) {
12339 dtrace_dof_error(dof, "DOF encoding mismatch");
12340 return (-1);
12341 }
12342
12343 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
12344 dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) {
12345 dtrace_dof_error(dof, "DOF version mismatch");
12346 return (-1);
12347 }
12348
12349 if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) {
12350 dtrace_dof_error(dof, "DOF uses unsupported instruction set");
12351 return (-1);
12352 }
12353
12354 if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) {
12355 dtrace_dof_error(dof, "DOF uses too many integer registers");
12356 return (-1);
12357 }
12358
12359 if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) {
12360 dtrace_dof_error(dof, "DOF uses too many tuple registers");
12361 return (-1);
12362 }
12363
12364 for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) {
12365 if (dof->dofh_ident[i] != 0) {
12366 dtrace_dof_error(dof, "DOF has invalid ident byte set");
12367 return (-1);
12368 }
12369 }
12370
12371 if (dof->dofh_flags & ~DOF_FL_VALID) {
12372 dtrace_dof_error(dof, "DOF has invalid flag bits set");
12373 return (-1);
12374 }
12375
12376 if (dof->dofh_secsize == 0) {
12377 dtrace_dof_error(dof, "zero section header size");
12378 return (-1);
12379 }
12380
12381 /*
12382 * Check that the section headers don't exceed the amount of DOF
12383 * data. Note that we cast the section size and number of sections
12384 * to uint64_t's to prevent possible overflow in the multiplication.
12385 */
12386 seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize;
12387
12388 if (dof->dofh_secoff > len || seclen > len ||
12389 dof->dofh_secoff + seclen > len) {
12390 dtrace_dof_error(dof, "truncated section headers");
12391 return (-1);
12392 }
12393
12394 if (!IS_P2ALIGNED(dof->dofh_secoff, sizeof (uint64_t))) {
12395 dtrace_dof_error(dof, "misaligned section headers");
12396 return (-1);
12397 }
12398
12399 if (!IS_P2ALIGNED(dof->dofh_secsize, sizeof (uint64_t))) {
12400 dtrace_dof_error(dof, "misaligned section size");
12401 return (-1);
12402 }
12403
12404 /*
12405 * Take an initial pass through the section headers to be sure that
12406 * the headers don't have stray offsets. If the 'noprobes' flag is
12407 * set, do not permit sections relating to providers, probes, or args.
12408 */
12409 for (i = 0; i < dof->dofh_secnum; i++) {
12410 dof_sec_t *sec = (dof_sec_t *)(daddr +
12411 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12412
12413 if (noprobes) {
12414 switch (sec->dofs_type) {
12415 case DOF_SECT_PROVIDER:
12416 case DOF_SECT_PROBES:
12417 case DOF_SECT_PRARGS:
12418 case DOF_SECT_PROFFS:
12419 dtrace_dof_error(dof, "illegal sections "
12420 "for enabling");
12421 return (-1);
12422 }
12423 }
12424
12425 if (!(sec->dofs_flags & DOF_SECF_LOAD))
12426 continue; /* just ignore non-loadable sections */
12427
12428 if (sec->dofs_align & (sec->dofs_align - 1)) {
12429 dtrace_dof_error(dof, "bad section alignment");
12430 return (-1);
12431 }
12432
12433 if (sec->dofs_offset & (sec->dofs_align - 1)) {
12434 dtrace_dof_error(dof, "misaligned section");
12435 return (-1);
12436 }
12437
12438 if (sec->dofs_offset > len || sec->dofs_size > len ||
12439 sec->dofs_offset + sec->dofs_size > len) {
12440 dtrace_dof_error(dof, "corrupt section header");
12441 return (-1);
12442 }
12443
12444 if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr +
12445 sec->dofs_offset + sec->dofs_size - 1) != '\0') {
12446 dtrace_dof_error(dof, "non-terminating string table");
12447 return (-1);
12448 }
12449 }
12450
12451 /*
12452 * Take a second pass through the sections and locate and perform any
12453 * relocations that are present. We do this after the first pass to
12454 * be sure that all sections have had their headers validated.
12455 */
12456 for (i = 0; i < dof->dofh_secnum; i++) {
12457 dof_sec_t *sec = (dof_sec_t *)(daddr +
12458 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12459
12460 if (!(sec->dofs_flags & DOF_SECF_LOAD))
12461 continue; /* skip sections that are not loadable */
12462
12463 switch (sec->dofs_type) {
12464 case DOF_SECT_URELHDR:
12465 if (dtrace_dof_relocate(dof, sec, ubase) != 0)
12466 return (-1);
12467 break;
12468 }
12469 }
12470
12471 if ((enab = *enabp) == NULL)
12472 enab = *enabp = dtrace_enabling_create(vstate);
12473
12474 for (i = 0; i < dof->dofh_secnum; i++) {
12475 dof_sec_t *sec = (dof_sec_t *)(daddr +
12476 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12477
12478 if (sec->dofs_type != DOF_SECT_ECBDESC)
12479 continue;
12480
12481 if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) {
12482 dtrace_enabling_destroy(enab);
12483 *enabp = NULL;
12484 return (-1);
12485 }
12486
12487 dtrace_enabling_add(enab, ep);
12488 }
12489
12490 return (0);
12491}
12492
12493/*
12494 * Process DOF for any options. This routine assumes that the DOF has been
12495 * at least processed by dtrace_dof_slurp().
12496 */
12497static int
12498dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state)
12499{
12500 int i, rval;
12501 uint32_t entsize;
12502 size_t offs;
12503 dof_optdesc_t *desc;
12504
12505 for (i = 0; i < dof->dofh_secnum; i++) {
12506 dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof +
12507 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize);
12508
12509 if (sec->dofs_type != DOF_SECT_OPTDESC)
12510 continue;
12511
12512 if (sec->dofs_align != sizeof (uint64_t)) {
12513 dtrace_dof_error(dof, "bad alignment in "
12514 "option description");
12515 return (EINVAL);
12516 }
12517
12518 if ((entsize = sec->dofs_entsize) == 0) {
12519 dtrace_dof_error(dof, "zeroed option entry size");
12520 return (EINVAL);
12521 }
12522
12523 if (entsize < sizeof (dof_optdesc_t)) {
12524 dtrace_dof_error(dof, "bad option entry size");
12525 return (EINVAL);
12526 }
12527
12528 for (offs = 0; offs < sec->dofs_size; offs += entsize) {
12529 desc = (dof_optdesc_t *)((uintptr_t)dof +
12530 (uintptr_t)sec->dofs_offset + offs);
12531
12532 if (desc->dofo_strtab != DOF_SECIDX_NONE) {
12533 dtrace_dof_error(dof, "non-zero option string");
12534 return (EINVAL);
12535 }
12536
12537 if (desc->dofo_value == DTRACEOPT_UNSET) {
12538 dtrace_dof_error(dof, "unset option");
12539 return (EINVAL);
12540 }
12541
12542 if ((rval = dtrace_state_option(state,
12543 desc->dofo_option, desc->dofo_value)) != 0) {
12544 dtrace_dof_error(dof, "rejected option");
12545 return (rval);
12546 }
12547 }
12548 }
12549
12550 return (0);
12551}
12552
12553/*
12554 * DTrace Consumer State Functions
12555 */
12556static int
12557dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size)
12558{
12559 size_t hashsize, maxper, min, chunksize = dstate->dtds_chunksize;
12560 void *base;
12561 uintptr_t limit;
12562 dtrace_dynvar_t *dvar, *next, *start;
12563 int i;
12564
12565 ASSERT(MUTEX_HELD(&dtrace_lock));
12566 ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL);
12567
12568 bzero(dstate, sizeof (dtrace_dstate_t));
12569
12570 if ((dstate->dtds_chunksize = chunksize) == 0)
12571 dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE;
12572
12573 if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)))
12574 size = min;
12575
12576 if ((base = kmem_zalloc(size, KM_NOSLEEP)) == NULL)
12577 return (ENOMEM);
12578
12579 dstate->dtds_size = size;
12580 dstate->dtds_base = base;
12581 dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, KM_SLEEP);
12582 bzero(dstate->dtds_percpu, NCPU * sizeof (dtrace_dstate_percpu_t));
12583
12584 hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t));
12585
12586 if (hashsize != 1 && (hashsize & 1))
12587 hashsize--;
12588
12589 dstate->dtds_hashsize = hashsize;
12590 dstate->dtds_hash = dstate->dtds_base;
12591
12592 /*
12593 * Set all of our hash buckets to point to the single sink, and (if
12594 * it hasn't already been set), set the sink's hash value to be the
12595 * sink sentinel value. The sink is needed for dynamic variable
12596 * lookups to know that they have iterated over an entire, valid hash
12597 * chain.
12598 */
12599 for (i = 0; i < hashsize; i++)
12600 dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink;
12601
12602 if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK)
12603 dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK;
12604
12605 /*
12606 * Determine number of active CPUs. Divide free list evenly among
12607 * active CPUs.
12608 */
12609 start = (dtrace_dynvar_t *)
12610 ((uintptr_t)base + hashsize * sizeof (dtrace_dynhash_t));
12611 limit = (uintptr_t)base + size;
12612
12613 maxper = (limit - (uintptr_t)start) / NCPU;
12614 maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize;
12615
|
12620#endif
12621 dstate->dtds_percpu[i].dtdsc_free = dvar = start;
12622
12623 /*
12624 * If we don't even have enough chunks to make it once through
12625 * NCPUs, we're just going to allocate everything to the first
12626 * CPU. And if we're on the last CPU, we're going to allocate
12627 * whatever is left over. In either case, we set the limit to
12628 * be the limit of the dynamic variable space.
12629 */
12630 if (maxper == 0 || i == NCPU - 1) {
12631 limit = (uintptr_t)base + size;
12632 start = NULL;
12633 } else {
12634 limit = (uintptr_t)start + maxper;
12635 start = (dtrace_dynvar_t *)limit;
12636 }
12637
12638 ASSERT(limit <= (uintptr_t)base + size);
12639
12640 for (;;) {
12641 next = (dtrace_dynvar_t *)((uintptr_t)dvar +
12642 dstate->dtds_chunksize);
12643
12644 if ((uintptr_t)next + dstate->dtds_chunksize >= limit)
12645 break;
12646
12647 dvar->dtdv_next = next;
12648 dvar = next;
12649 }
12650
12651 if (maxper == 0)
12652 break;
12653 }
12654
12655 return (0);
12656}
12657
12658static void
12659dtrace_dstate_fini(dtrace_dstate_t *dstate)
12660{
12661 ASSERT(MUTEX_HELD(&cpu_lock));
12662
12663 if (dstate->dtds_base == NULL)
12664 return;
12665
12666 kmem_free(dstate->dtds_base, dstate->dtds_size);
12667 kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu);
12668}
12669
12670static void
12671dtrace_vstate_fini(dtrace_vstate_t *vstate)
12672{
12673 /*
12674 * Logical XOR, where are you?
12675 */
12676 ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL));
12677
12678 if (vstate->dtvs_nglobals > 0) {
12679 kmem_free(vstate->dtvs_globals, vstate->dtvs_nglobals *
12680 sizeof (dtrace_statvar_t *));
12681 }
12682
12683 if (vstate->dtvs_ntlocals > 0) {
12684 kmem_free(vstate->dtvs_tlocals, vstate->dtvs_ntlocals *
12685 sizeof (dtrace_difv_t));
12686 }
12687
12688 ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL));
12689
12690 if (vstate->dtvs_nlocals > 0) {
12691 kmem_free(vstate->dtvs_locals, vstate->dtvs_nlocals *
12692 sizeof (dtrace_statvar_t *));
12693 }
12694}
12695
12696#if defined(sun)
12697static void
12698dtrace_state_clean(dtrace_state_t *state)
12699{
12700 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE)
12701 return;
12702
12703 dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars);
12704 dtrace_speculation_clean(state);
12705}
12706
12707static void
12708dtrace_state_deadman(dtrace_state_t *state)
12709{
12710 hrtime_t now;
12711
12712 dtrace_sync();
12713
12714 now = dtrace_gethrtime();
12715
12716 if (state != dtrace_anon.dta_state &&
12717 now - state->dts_laststatus >= dtrace_deadman_user)
12718 return;
12719
12720 /*
12721 * We must be sure that dts_alive never appears to be less than the
12722 * value upon entry to dtrace_state_deadman(), and because we lack a
12723 * dtrace_cas64(), we cannot store to it atomically. We thus instead
12724 * store INT64_MAX to it, followed by a memory barrier, followed by
12725 * the new value. This assures that dts_alive never appears to be
12726 * less than its true value, regardless of the order in which the
12727 * stores to the underlying storage are issued.
12728 */
12729 state->dts_alive = INT64_MAX;
12730 dtrace_membar_producer();
12731 state->dts_alive = now;
12732}
12733#else
12734static void
12735dtrace_state_clean(void *arg)
12736{
12737 dtrace_state_t *state = arg;
12738 dtrace_optval_t *opt = state->dts_options;
12739
12740 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE)
12741 return;
12742
12743 dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars);
12744 dtrace_speculation_clean(state);
12745
12746 callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC,
12747 dtrace_state_clean, state);
12748}
12749
12750static void
12751dtrace_state_deadman(void *arg)
12752{
12753 dtrace_state_t *state = arg;
12754 hrtime_t now;
12755
12756 dtrace_sync();
12757
12758 dtrace_debug_output();
12759
12760 now = dtrace_gethrtime();
12761
12762 if (state != dtrace_anon.dta_state &&
12763 now - state->dts_laststatus >= dtrace_deadman_user)
12764 return;
12765
12766 /*
12767 * We must be sure that dts_alive never appears to be less than the
12768 * value upon entry to dtrace_state_deadman(), and because we lack a
12769 * dtrace_cas64(), we cannot store to it atomically. We thus instead
12770 * store INT64_MAX to it, followed by a memory barrier, followed by
12771 * the new value. This assures that dts_alive never appears to be
12772 * less than its true value, regardless of the order in which the
12773 * stores to the underlying storage are issued.
12774 */
12775 state->dts_alive = INT64_MAX;
12776 dtrace_membar_producer();
12777 state->dts_alive = now;
12778
12779 callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC,
12780 dtrace_state_deadman, state);
12781}
12782#endif
12783
12784static dtrace_state_t *
12785#if defined(sun)
12786dtrace_state_create(dev_t *devp, cred_t *cr)
12787#else
12788dtrace_state_create(struct cdev *dev)
12789#endif
12790{
12791#if defined(sun)
12792 minor_t minor;
12793 major_t major;
12794#else
12795 cred_t *cr = NULL;
12796 int m = 0;
12797#endif
12798 char c[30];
12799 dtrace_state_t *state;
12800 dtrace_optval_t *opt;
12801 int bufsize = NCPU * sizeof (dtrace_buffer_t), i;
12802
12803 ASSERT(MUTEX_HELD(&dtrace_lock));
12804 ASSERT(MUTEX_HELD(&cpu_lock));
12805
12806#if defined(sun)
12807 minor = (minor_t)(uintptr_t)vmem_alloc(dtrace_minor, 1,
12808 VM_BESTFIT | VM_SLEEP);
12809
12810 if (ddi_soft_state_zalloc(dtrace_softstate, minor) != DDI_SUCCESS) {
12811 vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1);
12812 return (NULL);
12813 }
12814
12815 state = ddi_get_soft_state(dtrace_softstate, minor);
12816#else
12817 if (dev != NULL) {
12818 cr = dev->si_cred;
12819 m = dev2unit(dev);
12820 }
12821
12822 /* Allocate memory for the state. */
12823 state = kmem_zalloc(sizeof(dtrace_state_t), KM_SLEEP);
12824#endif
12825
12826 state->dts_epid = DTRACE_EPIDNONE + 1;
12827
12828 (void) snprintf(c, sizeof (c), "dtrace_aggid_%d", m);
12829#if defined(sun)
12830 state->dts_aggid_arena = vmem_create(c, (void *)1, UINT32_MAX, 1,
12831 NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
12832
12833 if (devp != NULL) {
12834 major = getemajor(*devp);
12835 } else {
12836 major = ddi_driver_major(dtrace_devi);
12837 }
12838
12839 state->dts_dev = makedevice(major, minor);
12840
12841 if (devp != NULL)
12842 *devp = state->dts_dev;
12843#else
12844 state->dts_aggid_arena = new_unrhdr(1, INT_MAX, &dtrace_unr_mtx);
12845 state->dts_dev = dev;
12846#endif
12847
12848 /*
12849 * We allocate NCPU buffers. On the one hand, this can be quite
12850 * a bit of memory per instance (nearly 36K on a Starcat). On the
12851 * other hand, it saves an additional memory reference in the probe
12852 * path.
12853 */
12854 state->dts_buffer = kmem_zalloc(bufsize, KM_SLEEP);
12855 state->dts_aggbuffer = kmem_zalloc(bufsize, KM_SLEEP);
12856
12857#if defined(sun)
12858 state->dts_cleaner = CYCLIC_NONE;
12859 state->dts_deadman = CYCLIC_NONE;
12860#else
12861 callout_init(&state->dts_cleaner, CALLOUT_MPSAFE);
12862 callout_init(&state->dts_deadman, CALLOUT_MPSAFE);
12863#endif
12864 state->dts_vstate.dtvs_state = state;
12865
12866 for (i = 0; i < DTRACEOPT_MAX; i++)
12867 state->dts_options[i] = DTRACEOPT_UNSET;
12868
12869 /*
12870 * Set the default options.
12871 */
12872 opt = state->dts_options;
12873 opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH;
12874 opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO;
12875 opt[DTRACEOPT_NSPEC] = dtrace_nspec_default;
12876 opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default;
12877 opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL;
12878 opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default;
12879 opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default;
12880 opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default;
12881 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default;
12882 opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default;
12883 opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default;
12884 opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default;
12885 opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default;
12886 opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default;
12887
12888 state->dts_activity = DTRACE_ACTIVITY_INACTIVE;
12889
12890 /*
12891 * Depending on the user credentials, we set flag bits which alter probe
12892 * visibility or the amount of destructiveness allowed. In the case of
12893 * actual anonymous tracing, or the possession of all privileges, all of
12894 * the normal checks are bypassed.
12895 */
12896 if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) {
12897 state->dts_cred.dcr_visible = DTRACE_CRV_ALL;
12898 state->dts_cred.dcr_action = DTRACE_CRA_ALL;
12899 } else {
12900 /*
12901 * Set up the credentials for this instantiation. We take a
12902 * hold on the credential to prevent it from disappearing on
12903 * us; this in turn prevents the zone_t referenced by this
12904 * credential from disappearing. This means that we can
12905 * examine the credential and the zone from probe context.
12906 */
12907 crhold(cr);
12908 state->dts_cred.dcr_cred = cr;
12909
12910 /*
12911 * CRA_PROC means "we have *some* privilege for dtrace" and
12912 * unlocks the use of variables like pid, zonename, etc.
12913 */
12914 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE) ||
12915 PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) {
12916 state->dts_cred.dcr_action |= DTRACE_CRA_PROC;
12917 }
12918
12919 /*
12920 * dtrace_user allows use of syscall and profile providers.
12921 * If the user also has proc_owner and/or proc_zone, we
12922 * extend the scope to include additional visibility and
12923 * destructive power.
12924 */
12925 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) {
12926 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) {
12927 state->dts_cred.dcr_visible |=
12928 DTRACE_CRV_ALLPROC;
12929
12930 state->dts_cred.dcr_action |=
12931 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
12932 }
12933
12934 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) {
12935 state->dts_cred.dcr_visible |=
12936 DTRACE_CRV_ALLZONE;
12937
12938 state->dts_cred.dcr_action |=
12939 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
12940 }
12941
12942 /*
12943 * If we have all privs in whatever zone this is,
12944 * we can do destructive things to processes which
12945 * have altered credentials.
12946 */
12947#if defined(sun)
12948 if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE),
12949 cr->cr_zone->zone_privset)) {
12950 state->dts_cred.dcr_action |=
12951 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG;
12952 }
12953#endif
12954 }
12955
12956 /*
12957 * Holding the dtrace_kernel privilege also implies that
12958 * the user has the dtrace_user privilege from a visibility
12959 * perspective. But without further privileges, some
12960 * destructive actions are not available.
12961 */
12962 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) {
12963 /*
12964 * Make all probes in all zones visible. However,
12965 * this doesn't mean that all actions become available
12966 * to all zones.
12967 */
12968 state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL |
12969 DTRACE_CRV_ALLPROC | DTRACE_CRV_ALLZONE;
12970
12971 state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL |
12972 DTRACE_CRA_PROC;
12973 /*
12974 * Holding proc_owner means that destructive actions
12975 * for *this* zone are allowed.
12976 */
12977 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
12978 state->dts_cred.dcr_action |=
12979 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
12980
12981 /*
12982 * Holding proc_zone means that destructive actions
12983 * for this user/group ID in all zones is allowed.
12984 */
12985 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
12986 state->dts_cred.dcr_action |=
12987 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
12988
12989#if defined(sun)
12990 /*
12991 * If we have all privs in whatever zone this is,
12992 * we can do destructive things to processes which
12993 * have altered credentials.
12994 */
12995 if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE),
12996 cr->cr_zone->zone_privset)) {
12997 state->dts_cred.dcr_action |=
12998 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG;
12999 }
13000#endif
13001 }
13002
13003 /*
13004 * Holding the dtrace_proc privilege gives control over fasttrap
13005 * and pid providers. We need to grant wider destructive
13006 * privileges in the event that the user has proc_owner and/or
13007 * proc_zone.
13008 */
13009 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) {
13010 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE))
13011 state->dts_cred.dcr_action |=
13012 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER;
13013
13014 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE))
13015 state->dts_cred.dcr_action |=
13016 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE;
13017 }
13018 }
13019
13020 return (state);
13021}
13022
13023static int
13024dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, int which)
13025{
13026 dtrace_optval_t *opt = state->dts_options, size;
13027 processorid_t cpu = 0;;
13028 int flags = 0, rval;
13029
13030 ASSERT(MUTEX_HELD(&dtrace_lock));
13031 ASSERT(MUTEX_HELD(&cpu_lock));
13032 ASSERT(which < DTRACEOPT_MAX);
13033 ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE ||
13034 (state == dtrace_anon.dta_state &&
13035 state->dts_activity == DTRACE_ACTIVITY_ACTIVE));
13036
13037 if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0)
13038 return (0);
13039
13040 if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET)
13041 cpu = opt[DTRACEOPT_CPU];
13042
13043 if (which == DTRACEOPT_SPECSIZE)
13044 flags |= DTRACEBUF_NOSWITCH;
13045
13046 if (which == DTRACEOPT_BUFSIZE) {
13047 if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING)
13048 flags |= DTRACEBUF_RING;
13049
13050 if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL)
13051 flags |= DTRACEBUF_FILL;
13052
13053 if (state != dtrace_anon.dta_state ||
13054 state->dts_activity != DTRACE_ACTIVITY_ACTIVE)
13055 flags |= DTRACEBUF_INACTIVE;
13056 }
13057
13058 for (size = opt[which]; size >= sizeof (uint64_t); size >>= 1) {
13059 /*
13060 * The size must be 8-byte aligned. If the size is not 8-byte
13061 * aligned, drop it down by the difference.
13062 */
13063 if (size & (sizeof (uint64_t) - 1))
13064 size -= size & (sizeof (uint64_t) - 1);
13065
13066 if (size < state->dts_reserve) {
13067 /*
13068 * Buffers always must be large enough to accommodate
13069 * their prereserved space. We return E2BIG instead
13070 * of ENOMEM in this case to allow for user-level
13071 * software to differentiate the cases.
13072 */
13073 return (E2BIG);
13074 }
13075
13076 rval = dtrace_buffer_alloc(buf, size, flags, cpu);
13077
13078 if (rval != ENOMEM) {
13079 opt[which] = size;
13080 return (rval);
13081 }
13082
13083 if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL)
13084 return (rval);
13085 }
13086
13087 return (ENOMEM);
13088}
13089
13090static int
13091dtrace_state_buffers(dtrace_state_t *state)
13092{
13093 dtrace_speculation_t *spec = state->dts_speculations;
13094 int rval, i;
13095
13096 if ((rval = dtrace_state_buffer(state, state->dts_buffer,
13097 DTRACEOPT_BUFSIZE)) != 0)
13098 return (rval);
13099
13100 if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer,
13101 DTRACEOPT_AGGSIZE)) != 0)
13102 return (rval);
13103
13104 for (i = 0; i < state->dts_nspeculations; i++) {
13105 if ((rval = dtrace_state_buffer(state,
13106 spec[i].dtsp_buffer, DTRACEOPT_SPECSIZE)) != 0)
13107 return (rval);
13108 }
13109
13110 return (0);
13111}
13112
13113static void
13114dtrace_state_prereserve(dtrace_state_t *state)
13115{
13116 dtrace_ecb_t *ecb;
13117 dtrace_probe_t *probe;
13118
13119 state->dts_reserve = 0;
13120
13121 if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL)
13122 return;
13123
13124 /*
13125 * If our buffer policy is a "fill" buffer policy, we need to set the
13126 * prereserved space to be the space required by the END probes.
13127 */
13128 probe = dtrace_probes[dtrace_probeid_end - 1];
13129 ASSERT(probe != NULL);
13130
13131 for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) {
13132 if (ecb->dte_state != state)
13133 continue;
13134
13135 state->dts_reserve += ecb->dte_needed + ecb->dte_alignment;
13136 }
13137}
13138
13139static int
13140dtrace_state_go(dtrace_state_t *state, processorid_t *cpu)
13141{
13142 dtrace_optval_t *opt = state->dts_options, sz, nspec;
13143 dtrace_speculation_t *spec;
13144 dtrace_buffer_t *buf;
13145#if defined(sun)
13146 cyc_handler_t hdlr;
13147 cyc_time_t when;
13148#endif
13149 int rval = 0, i, bufsize = NCPU * sizeof (dtrace_buffer_t);
13150 dtrace_icookie_t cookie;
13151
13152 mutex_enter(&cpu_lock);
13153 mutex_enter(&dtrace_lock);
13154
13155 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) {
13156 rval = EBUSY;
13157 goto out;
13158 }
13159
13160 /*
13161 * Before we can perform any checks, we must prime all of the
13162 * retained enablings that correspond to this state.
13163 */
13164 dtrace_enabling_prime(state);
13165
13166 if (state->dts_destructive && !state->dts_cred.dcr_destructive) {
13167 rval = EACCES;
13168 goto out;
13169 }
13170
13171 dtrace_state_prereserve(state);
13172
13173 /*
13174 * Now we want to do is try to allocate our speculations.
13175 * We do not automatically resize the number of speculations; if
13176 * this fails, we will fail the operation.
13177 */
13178 nspec = opt[DTRACEOPT_NSPEC];
13179 ASSERT(nspec != DTRACEOPT_UNSET);
13180
13181 if (nspec > INT_MAX) {
13182 rval = ENOMEM;
13183 goto out;
13184 }
13185
13186 spec = kmem_zalloc(nspec * sizeof (dtrace_speculation_t), KM_NOSLEEP);
13187
13188 if (spec == NULL) {
13189 rval = ENOMEM;
13190 goto out;
13191 }
13192
13193 state->dts_speculations = spec;
13194 state->dts_nspeculations = (int)nspec;
13195
13196 for (i = 0; i < nspec; i++) {
13197 if ((buf = kmem_zalloc(bufsize, KM_NOSLEEP)) == NULL) {
13198 rval = ENOMEM;
13199 goto err;
13200 }
13201
13202 spec[i].dtsp_buffer = buf;
13203 }
13204
13205 if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) {
13206 if (dtrace_anon.dta_state == NULL) {
13207 rval = ENOENT;
13208 goto out;
13209 }
13210
13211 if (state->dts_necbs != 0) {
13212 rval = EALREADY;
13213 goto out;
13214 }
13215
13216 state->dts_anon = dtrace_anon_grab();
13217 ASSERT(state->dts_anon != NULL);
13218 state = state->dts_anon;
13219
13220 /*
13221 * We want "grabanon" to be set in the grabbed state, so we'll
13222 * copy that option value from the grabbing state into the
13223 * grabbed state.
13224 */
13225 state->dts_options[DTRACEOPT_GRABANON] =
13226 opt[DTRACEOPT_GRABANON];
13227
13228 *cpu = dtrace_anon.dta_beganon;
13229
13230 /*
13231 * If the anonymous state is active (as it almost certainly
13232 * is if the anonymous enabling ultimately matched anything),
13233 * we don't allow any further option processing -- but we
13234 * don't return failure.
13235 */
13236 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
13237 goto out;
13238 }
13239
13240 if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET &&
13241 opt[DTRACEOPT_AGGSIZE] != 0) {
13242 if (state->dts_aggregations == NULL) {
13243 /*
13244 * We're not going to create an aggregation buffer
13245 * because we don't have any ECBs that contain
13246 * aggregations -- set this option to 0.
13247 */
13248 opt[DTRACEOPT_AGGSIZE] = 0;
13249 } else {
13250 /*
13251 * If we have an aggregation buffer, we must also have
13252 * a buffer to use as scratch.
13253 */
13254 if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET ||
13255 opt[DTRACEOPT_BUFSIZE] < state->dts_needed) {
13256 opt[DTRACEOPT_BUFSIZE] = state->dts_needed;
13257 }
13258 }
13259 }
13260
13261 if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET &&
13262 opt[DTRACEOPT_SPECSIZE] != 0) {
13263 if (!state->dts_speculates) {
13264 /*
13265 * We're not going to create speculation buffers
13266 * because we don't have any ECBs that actually
13267 * speculate -- set the speculation size to 0.
13268 */
13269 opt[DTRACEOPT_SPECSIZE] = 0;
13270 }
13271 }
13272
13273 /*
13274 * The bare minimum size for any buffer that we're actually going to
13275 * do anything to is sizeof (uint64_t).
13276 */
13277 sz = sizeof (uint64_t);
13278
13279 if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) ||
13280 (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) ||
13281 (state->dts_aggregations != NULL && opt[DTRACEOPT_AGGSIZE] < sz)) {
13282 /*
13283 * A buffer size has been explicitly set to 0 (or to a size
13284 * that will be adjusted to 0) and we need the space -- we
13285 * need to return failure. We return ENOSPC to differentiate
13286 * it from failing to allocate a buffer due to failure to meet
13287 * the reserve (for which we return E2BIG).
13288 */
13289 rval = ENOSPC;
13290 goto out;
13291 }
13292
13293 if ((rval = dtrace_state_buffers(state)) != 0)
13294 goto err;
13295
13296 if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET)
13297 sz = dtrace_dstate_defsize;
13298
13299 do {
13300 rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz);
13301
13302 if (rval == 0)
13303 break;
13304
13305 if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL)
13306 goto err;
13307 } while (sz >>= 1);
13308
13309 opt[DTRACEOPT_DYNVARSIZE] = sz;
13310
13311 if (rval != 0)
13312 goto err;
13313
13314 if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max)
13315 opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max;
13316
13317 if (opt[DTRACEOPT_CLEANRATE] == 0)
13318 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max;
13319
13320 if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min)
13321 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min;
13322
13323 if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max)
13324 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max;
13325
13326 state->dts_alive = state->dts_laststatus = dtrace_gethrtime();
13327#if defined(sun)
13328 hdlr.cyh_func = (cyc_func_t)dtrace_state_clean;
13329 hdlr.cyh_arg = state;
13330 hdlr.cyh_level = CY_LOW_LEVEL;
13331
13332 when.cyt_when = 0;
13333 when.cyt_interval = opt[DTRACEOPT_CLEANRATE];
13334
13335 state->dts_cleaner = cyclic_add(&hdlr, &when);
13336
13337 hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman;
13338 hdlr.cyh_arg = state;
13339 hdlr.cyh_level = CY_LOW_LEVEL;
13340
13341 when.cyt_when = 0;
13342 when.cyt_interval = dtrace_deadman_interval;
13343
13344 state->dts_deadman = cyclic_add(&hdlr, &when);
13345#else
13346 callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC,
13347 dtrace_state_clean, state);
13348 callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC,
13349 dtrace_state_deadman, state);
13350#endif
13351
13352 state->dts_activity = DTRACE_ACTIVITY_WARMUP;
13353
13354 /*
13355 * Now it's time to actually fire the BEGIN probe. We need to disable
13356 * interrupts here both to record the CPU on which we fired the BEGIN
13357 * probe (the data from this CPU will be processed first at user
13358 * level) and to manually activate the buffer for this CPU.
13359 */
13360 cookie = dtrace_interrupt_disable();
13361 *cpu = curcpu;
13362 ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE);
13363 state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE;
13364
13365 dtrace_probe(dtrace_probeid_begin,
13366 (uint64_t)(uintptr_t)state, 0, 0, 0, 0);
13367 dtrace_interrupt_enable(cookie);
13368 /*
13369 * We may have had an exit action from a BEGIN probe; only change our
13370 * state to ACTIVE if we're still in WARMUP.
13371 */
13372 ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP ||
13373 state->dts_activity == DTRACE_ACTIVITY_DRAINING);
13374
13375 if (state->dts_activity == DTRACE_ACTIVITY_WARMUP)
13376 state->dts_activity = DTRACE_ACTIVITY_ACTIVE;
13377
13378 /*
13379 * Regardless of whether or not now we're in ACTIVE or DRAINING, we
13380 * want each CPU to transition its principal buffer out of the
13381 * INACTIVE state. Doing this assures that no CPU will suddenly begin
13382 * processing an ECB halfway down a probe's ECB chain; all CPUs will
13383 * atomically transition from processing none of a state's ECBs to
13384 * processing all of them.
13385 */
13386 dtrace_xcall(DTRACE_CPUALL,
13387 (dtrace_xcall_t)dtrace_buffer_activate, state);
13388 goto out;
13389
13390err:
13391 dtrace_buffer_free(state->dts_buffer);
13392 dtrace_buffer_free(state->dts_aggbuffer);
13393
13394 if ((nspec = state->dts_nspeculations) == 0) {
13395 ASSERT(state->dts_speculations == NULL);
13396 goto out;
13397 }
13398
13399 spec = state->dts_speculations;
13400 ASSERT(spec != NULL);
13401
13402 for (i = 0; i < state->dts_nspeculations; i++) {
13403 if ((buf = spec[i].dtsp_buffer) == NULL)
13404 break;
13405
13406 dtrace_buffer_free(buf);
13407 kmem_free(buf, bufsize);
13408 }
13409
13410 kmem_free(spec, nspec * sizeof (dtrace_speculation_t));
13411 state->dts_nspeculations = 0;
13412 state->dts_speculations = NULL;
13413
13414out:
13415 mutex_exit(&dtrace_lock);
13416 mutex_exit(&cpu_lock);
13417
13418 return (rval);
13419}
13420
13421static int
13422dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu)
13423{
13424 dtrace_icookie_t cookie;
13425
13426 ASSERT(MUTEX_HELD(&dtrace_lock));
13427
13428 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE &&
13429 state->dts_activity != DTRACE_ACTIVITY_DRAINING)
13430 return (EINVAL);
13431
13432 /*
13433 * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync
13434 * to be sure that every CPU has seen it. See below for the details
13435 * on why this is done.
13436 */
13437 state->dts_activity = DTRACE_ACTIVITY_DRAINING;
13438 dtrace_sync();
13439
13440 /*
13441 * By this point, it is impossible for any CPU to be still processing
13442 * with DTRACE_ACTIVITY_ACTIVE. We can thus set our activity to
13443 * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any
13444 * other CPU in dtrace_buffer_reserve(). This allows dtrace_probe()
13445 * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN
13446 * iff we're in the END probe.
13447 */
13448 state->dts_activity = DTRACE_ACTIVITY_COOLDOWN;
13449 dtrace_sync();
13450 ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN);
13451
13452 /*
13453 * Finally, we can release the reserve and call the END probe. We
13454 * disable interrupts across calling the END probe to allow us to
13455 * return the CPU on which we actually called the END probe. This
13456 * allows user-land to be sure that this CPU's principal buffer is
13457 * processed last.
13458 */
13459 state->dts_reserve = 0;
13460
13461 cookie = dtrace_interrupt_disable();
13462 *cpu = curcpu;
13463 dtrace_probe(dtrace_probeid_end,
13464 (uint64_t)(uintptr_t)state, 0, 0, 0, 0);
13465 dtrace_interrupt_enable(cookie);
13466
13467 state->dts_activity = DTRACE_ACTIVITY_STOPPED;
13468 dtrace_sync();
13469
13470 return (0);
13471}
13472
13473static int
13474dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option,
13475 dtrace_optval_t val)
13476{
13477 ASSERT(MUTEX_HELD(&dtrace_lock));
13478
13479 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE)
13480 return (EBUSY);
13481
13482 if (option >= DTRACEOPT_MAX)
13483 return (EINVAL);
13484
13485 if (option != DTRACEOPT_CPU && val < 0)
13486 return (EINVAL);
13487
13488 switch (option) {
13489 case DTRACEOPT_DESTRUCTIVE:
13490 if (dtrace_destructive_disallow)
13491 return (EACCES);
13492
13493 state->dts_cred.dcr_destructive = 1;
13494 break;
13495
13496 case DTRACEOPT_BUFSIZE:
13497 case DTRACEOPT_DYNVARSIZE:
13498 case DTRACEOPT_AGGSIZE:
13499 case DTRACEOPT_SPECSIZE:
13500 case DTRACEOPT_STRSIZE:
13501 if (val < 0)
13502 return (EINVAL);
13503
13504 if (val >= LONG_MAX) {
13505 /*
13506 * If this is an otherwise negative value, set it to
13507 * the highest multiple of 128m less than LONG_MAX.
13508 * Technically, we're adjusting the size without
13509 * regard to the buffer resizing policy, but in fact,
13510 * this has no effect -- if we set the buffer size to
13511 * ~LONG_MAX and the buffer policy is ultimately set to
13512 * be "manual", the buffer allocation is guaranteed to
13513 * fail, if only because the allocation requires two
13514 * buffers. (We set the the size to the highest
13515 * multiple of 128m because it ensures that the size
13516 * will remain a multiple of a megabyte when
13517 * repeatedly halved -- all the way down to 15m.)
13518 */
13519 val = LONG_MAX - (1 << 27) + 1;
13520 }
13521 }
13522
13523 state->dts_options[option] = val;
13524
13525 return (0);
13526}
13527
13528static void
13529dtrace_state_destroy(dtrace_state_t *state)
13530{
13531 dtrace_ecb_t *ecb;
13532 dtrace_vstate_t *vstate = &state->dts_vstate;
13533#if defined(sun)
13534 minor_t minor = getminor(state->dts_dev);
13535#endif
13536 int i, bufsize = NCPU * sizeof (dtrace_buffer_t);
13537 dtrace_speculation_t *spec = state->dts_speculations;
13538 int nspec = state->dts_nspeculations;
13539 uint32_t match;
13540
13541 ASSERT(MUTEX_HELD(&dtrace_lock));
13542 ASSERT(MUTEX_HELD(&cpu_lock));
13543
13544 /*
13545 * First, retract any retained enablings for this state.
13546 */
13547 dtrace_enabling_retract(state);
13548 ASSERT(state->dts_nretained == 0);
13549
13550 if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE ||
13551 state->dts_activity == DTRACE_ACTIVITY_DRAINING) {
13552 /*
13553 * We have managed to come into dtrace_state_destroy() on a
13554 * hot enabling -- almost certainly because of a disorderly
13555 * shutdown of a consumer. (That is, a consumer that is
13556 * exiting without having called dtrace_stop().) In this case,
13557 * we're going to set our activity to be KILLED, and then
13558 * issue a sync to be sure that everyone is out of probe
13559 * context before we start blowing away ECBs.
13560 */
13561 state->dts_activity = DTRACE_ACTIVITY_KILLED;
13562 dtrace_sync();
13563 }
13564
13565 /*
13566 * Release the credential hold we took in dtrace_state_create().
13567 */
13568 if (state->dts_cred.dcr_cred != NULL)
13569 crfree(state->dts_cred.dcr_cred);
13570
13571 /*
13572 * Now we can safely disable and destroy any enabled probes. Because
13573 * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress
13574 * (especially if they're all enabled), we take two passes through the
13575 * ECBs: in the first, we disable just DTRACE_PRIV_KERNEL probes, and
13576 * in the second we disable whatever is left over.
13577 */
13578 for (match = DTRACE_PRIV_KERNEL; ; match = 0) {
13579 for (i = 0; i < state->dts_necbs; i++) {
13580 if ((ecb = state->dts_ecbs[i]) == NULL)
13581 continue;
13582
13583 if (match && ecb->dte_probe != NULL) {
13584 dtrace_probe_t *probe = ecb->dte_probe;
13585 dtrace_provider_t *prov = probe->dtpr_provider;
13586
13587 if (!(prov->dtpv_priv.dtpp_flags & match))
13588 continue;
13589 }
13590
13591 dtrace_ecb_disable(ecb);
13592 dtrace_ecb_destroy(ecb);
13593 }
13594
13595 if (!match)
13596 break;
13597 }
13598
13599 /*
13600 * Before we free the buffers, perform one more sync to assure that
13601 * every CPU is out of probe context.
13602 */
13603 dtrace_sync();
13604
13605 dtrace_buffer_free(state->dts_buffer);
13606 dtrace_buffer_free(state->dts_aggbuffer);
13607
13608 for (i = 0; i < nspec; i++)
13609 dtrace_buffer_free(spec[i].dtsp_buffer);
13610
13611#if defined(sun)
13612 if (state->dts_cleaner != CYCLIC_NONE)
13613 cyclic_remove(state->dts_cleaner);
13614
13615 if (state->dts_deadman != CYCLIC_NONE)
13616 cyclic_remove(state->dts_deadman);
13617#else
13618 callout_stop(&state->dts_cleaner);
13619 callout_drain(&state->dts_cleaner);
13620 callout_stop(&state->dts_deadman);
13621 callout_drain(&state->dts_deadman);
13622#endif
13623
13624 dtrace_dstate_fini(&vstate->dtvs_dynvars);
13625 dtrace_vstate_fini(vstate);
13626 if (state->dts_ecbs != NULL)
13627 kmem_free(state->dts_ecbs, state->dts_necbs * sizeof (dtrace_ecb_t *));
13628
13629 if (state->dts_aggregations != NULL) {
13630#ifdef DEBUG
13631 for (i = 0; i < state->dts_naggregations; i++)
13632 ASSERT(state->dts_aggregations[i] == NULL);
13633#endif
13634 ASSERT(state->dts_naggregations > 0);
13635 kmem_free(state->dts_aggregations,
13636 state->dts_naggregations * sizeof (dtrace_aggregation_t *));
13637 }
13638
13639 kmem_free(state->dts_buffer, bufsize);
13640 kmem_free(state->dts_aggbuffer, bufsize);
13641
13642 for (i = 0; i < nspec; i++)
13643 kmem_free(spec[i].dtsp_buffer, bufsize);
13644
13645 if (spec != NULL)
13646 kmem_free(spec, nspec * sizeof (dtrace_speculation_t));
13647
13648 dtrace_format_destroy(state);
13649
13650 if (state->dts_aggid_arena != NULL) {
13651#if defined(sun)
13652 vmem_destroy(state->dts_aggid_arena);
13653#else
13654 delete_unrhdr(state->dts_aggid_arena);
13655#endif
13656 state->dts_aggid_arena = NULL;
13657 }
13658#if defined(sun)
13659 ddi_soft_state_free(dtrace_softstate, minor);
13660 vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1);
13661#endif
13662}
13663
13664/*
13665 * DTrace Anonymous Enabling Functions
13666 */
13667static dtrace_state_t *
13668dtrace_anon_grab(void)
13669{
13670 dtrace_state_t *state;
13671
13672 ASSERT(MUTEX_HELD(&dtrace_lock));
13673
13674 if ((state = dtrace_anon.dta_state) == NULL) {
13675 ASSERT(dtrace_anon.dta_enabling == NULL);
13676 return (NULL);
13677 }
13678
13679 ASSERT(dtrace_anon.dta_enabling != NULL);
13680 ASSERT(dtrace_retained != NULL);
13681
13682 dtrace_enabling_destroy(dtrace_anon.dta_enabling);
13683 dtrace_anon.dta_enabling = NULL;
13684 dtrace_anon.dta_state = NULL;
13685
13686 return (state);
13687}
13688
13689static void
13690dtrace_anon_property(void)
13691{
13692 int i, rv;
13693 dtrace_state_t *state;
13694 dof_hdr_t *dof;
13695 char c[32]; /* enough for "dof-data-" + digits */
13696
13697 ASSERT(MUTEX_HELD(&dtrace_lock));
13698 ASSERT(MUTEX_HELD(&cpu_lock));
13699
13700 for (i = 0; ; i++) {
13701 (void) snprintf(c, sizeof (c), "dof-data-%d", i);
13702
13703 dtrace_err_verbose = 1;
13704
13705 if ((dof = dtrace_dof_property(c)) == NULL) {
13706 dtrace_err_verbose = 0;
13707 break;
13708 }
13709
13710#if defined(sun)
13711 /*
13712 * We want to create anonymous state, so we need to transition
13713 * the kernel debugger to indicate that DTrace is active. If
13714 * this fails (e.g. because the debugger has modified text in
13715 * some way), we won't continue with the processing.
13716 */
13717 if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) {
13718 cmn_err(CE_NOTE, "kernel debugger active; anonymous "
13719 "enabling ignored.");
13720 dtrace_dof_destroy(dof);
13721 break;
13722 }
13723#endif
13724
13725 /*
13726 * If we haven't allocated an anonymous state, we'll do so now.
13727 */
13728 if ((state = dtrace_anon.dta_state) == NULL) {
13729#if defined(sun)
13730 state = dtrace_state_create(NULL, NULL);
13731#else
13732 state = dtrace_state_create(NULL);
13733#endif
13734 dtrace_anon.dta_state = state;
13735
13736 if (state == NULL) {
13737 /*
13738 * This basically shouldn't happen: the only
13739 * failure mode from dtrace_state_create() is a
13740 * failure of ddi_soft_state_zalloc() that
13741 * itself should never happen. Still, the
13742 * interface allows for a failure mode, and
13743 * we want to fail as gracefully as possible:
13744 * we'll emit an error message and cease
13745 * processing anonymous state in this case.
13746 */
13747 cmn_err(CE_WARN, "failed to create "
13748 "anonymous state");
13749 dtrace_dof_destroy(dof);
13750 break;
13751 }
13752 }
13753
13754 rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(),
13755 &dtrace_anon.dta_enabling, 0, B_TRUE);
13756
13757 if (rv == 0)
13758 rv = dtrace_dof_options(dof, state);
13759
13760 dtrace_err_verbose = 0;
13761 dtrace_dof_destroy(dof);
13762
13763 if (rv != 0) {
13764 /*
13765 * This is malformed DOF; chuck any anonymous state
13766 * that we created.
13767 */
13768 ASSERT(dtrace_anon.dta_enabling == NULL);
13769 dtrace_state_destroy(state);
13770 dtrace_anon.dta_state = NULL;
13771 break;
13772 }
13773
13774 ASSERT(dtrace_anon.dta_enabling != NULL);
13775 }
13776
13777 if (dtrace_anon.dta_enabling != NULL) {
13778 int rval;
13779
13780 /*
13781 * dtrace_enabling_retain() can only fail because we are
13782 * trying to retain more enablings than are allowed -- but
13783 * we only have one anonymous enabling, and we are guaranteed
13784 * to be allowed at least one retained enabling; we assert
13785 * that dtrace_enabling_retain() returns success.
13786 */
13787 rval = dtrace_enabling_retain(dtrace_anon.dta_enabling);
13788 ASSERT(rval == 0);
13789
13790 dtrace_enabling_dump(dtrace_anon.dta_enabling);
13791 }
13792}
13793
13794#if defined(sun)
13795/*
13796 * DTrace Helper Functions
13797 */
13798static void
13799dtrace_helper_trace(dtrace_helper_action_t *helper,
13800 dtrace_mstate_t *mstate, dtrace_vstate_t *vstate, int where)
13801{
13802 uint32_t size, next, nnext, i;
13803 dtrace_helptrace_t *ent;
13804 uint16_t flags = cpu_core[curcpu].cpuc_dtrace_flags;
13805
13806 if (!dtrace_helptrace_enabled)
13807 return;
13808
13809 ASSERT(vstate->dtvs_nlocals <= dtrace_helptrace_nlocals);
13810
13811 /*
13812 * What would a tracing framework be without its own tracing
13813 * framework? (Well, a hell of a lot simpler, for starters...)
13814 */
13815 size = sizeof (dtrace_helptrace_t) + dtrace_helptrace_nlocals *
13816 sizeof (uint64_t) - sizeof (uint64_t);
13817
13818 /*
13819 * Iterate until we can allocate a slot in the trace buffer.
13820 */
13821 do {
13822 next = dtrace_helptrace_next;
13823
13824 if (next + size < dtrace_helptrace_bufsize) {
13825 nnext = next + size;
13826 } else {
13827 nnext = size;
13828 }
13829 } while (dtrace_cas32(&dtrace_helptrace_next, next, nnext) != next);
13830
13831 /*
13832 * We have our slot; fill it in.
13833 */
13834 if (nnext == size)
13835 next = 0;
13836
13837 ent = (dtrace_helptrace_t *)&dtrace_helptrace_buffer[next];
13838 ent->dtht_helper = helper;
13839 ent->dtht_where = where;
13840 ent->dtht_nlocals = vstate->dtvs_nlocals;
13841
13842 ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) ?
13843 mstate->dtms_fltoffs : -1;
13844 ent->dtht_fault = DTRACE_FLAGS2FLT(flags);
13845 ent->dtht_illval = cpu_core[curcpu].cpuc_dtrace_illval;
13846
13847 for (i = 0; i < vstate->dtvs_nlocals; i++) {
13848 dtrace_statvar_t *svar;
13849
13850 if ((svar = vstate->dtvs_locals[i]) == NULL)
13851 continue;
13852
13853 ASSERT(svar->dtsv_size >= NCPU * sizeof (uint64_t));
13854 ent->dtht_locals[i] =
13855 ((uint64_t *)(uintptr_t)svar->dtsv_data)[curcpu];
13856 }
13857}
13858#endif
13859
13860#if defined(sun)
13861static uint64_t
13862dtrace_helper(int which, dtrace_mstate_t *mstate,
13863 dtrace_state_t *state, uint64_t arg0, uint64_t arg1)
13864{
13865 uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags;
13866 uint64_t sarg0 = mstate->dtms_arg[0];
13867 uint64_t sarg1 = mstate->dtms_arg[1];
13868 uint64_t rval;
13869 dtrace_helpers_t *helpers = curproc->p_dtrace_helpers;
13870 dtrace_helper_action_t *helper;
13871 dtrace_vstate_t *vstate;
13872 dtrace_difo_t *pred;
13873 int i, trace = dtrace_helptrace_enabled;
13874
13875 ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS);
13876
13877 if (helpers == NULL)
13878 return (0);
13879
13880 if ((helper = helpers->dthps_actions[which]) == NULL)
13881 return (0);
13882
13883 vstate = &helpers->dthps_vstate;
13884 mstate->dtms_arg[0] = arg0;
13885 mstate->dtms_arg[1] = arg1;
13886
13887 /*
13888 * Now iterate over each helper. If its predicate evaluates to 'true',
13889 * we'll call the corresponding actions. Note that the below calls
13890 * to dtrace_dif_emulate() may set faults in machine state. This is
13891 * okay: our caller (the outer dtrace_dif_emulate()) will simply plow
13892 * the stored DIF offset with its own (which is the desired behavior).
13893 * Also, note the calls to dtrace_dif_emulate() may allocate scratch
13894 * from machine state; this is okay, too.
13895 */
13896 for (; helper != NULL; helper = helper->dtha_next) {
13897 if ((pred = helper->dtha_predicate) != NULL) {
13898 if (trace)
13899 dtrace_helper_trace(helper, mstate, vstate, 0);
13900
13901 if (!dtrace_dif_emulate(pred, mstate, vstate, state))
13902 goto next;
13903
13904 if (*flags & CPU_DTRACE_FAULT)
13905 goto err;
13906 }
13907
13908 for (i = 0; i < helper->dtha_nactions; i++) {
13909 if (trace)
13910 dtrace_helper_trace(helper,
13911 mstate, vstate, i + 1);
13912
13913 rval = dtrace_dif_emulate(helper->dtha_actions[i],
13914 mstate, vstate, state);
13915
13916 if (*flags & CPU_DTRACE_FAULT)
13917 goto err;
13918 }
13919
13920next:
13921 if (trace)
13922 dtrace_helper_trace(helper, mstate, vstate,
13923 DTRACE_HELPTRACE_NEXT);
13924 }
13925
13926 if (trace)
13927 dtrace_helper_trace(helper, mstate, vstate,
13928 DTRACE_HELPTRACE_DONE);
13929
13930 /*
13931 * Restore the arg0 that we saved upon entry.
13932 */
13933 mstate->dtms_arg[0] = sarg0;
13934 mstate->dtms_arg[1] = sarg1;
13935
13936 return (rval);
13937
13938err:
13939 if (trace)
13940 dtrace_helper_trace(helper, mstate, vstate,
13941 DTRACE_HELPTRACE_ERR);
13942
13943 /*
13944 * Restore the arg0 that we saved upon entry.
13945 */
13946 mstate->dtms_arg[0] = sarg0;
13947 mstate->dtms_arg[1] = sarg1;
13948
13949 return (0);
13950}
13951
13952static void
13953dtrace_helper_action_destroy(dtrace_helper_action_t *helper,
13954 dtrace_vstate_t *vstate)
13955{
13956 int i;
13957
13958 if (helper->dtha_predicate != NULL)
13959 dtrace_difo_release(helper->dtha_predicate, vstate);
13960
13961 for (i = 0; i < helper->dtha_nactions; i++) {
13962 ASSERT(helper->dtha_actions[i] != NULL);
13963 dtrace_difo_release(helper->dtha_actions[i], vstate);
13964 }
13965
13966 kmem_free(helper->dtha_actions,
13967 helper->dtha_nactions * sizeof (dtrace_difo_t *));
13968 kmem_free(helper, sizeof (dtrace_helper_action_t));
13969}
13970
13971static int
13972dtrace_helper_destroygen(int gen)
13973{
13974 proc_t *p = curproc;
13975 dtrace_helpers_t *help = p->p_dtrace_helpers;
13976 dtrace_vstate_t *vstate;
13977 int i;
13978
13979 ASSERT(MUTEX_HELD(&dtrace_lock));
13980
13981 if (help == NULL || gen > help->dthps_generation)
13982 return (EINVAL);
13983
13984 vstate = &help->dthps_vstate;
13985
13986 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
13987 dtrace_helper_action_t *last = NULL, *h, *next;
13988
13989 for (h = help->dthps_actions[i]; h != NULL; h = next) {
13990 next = h->dtha_next;
13991
13992 if (h->dtha_generation == gen) {
13993 if (last != NULL) {
13994 last->dtha_next = next;
13995 } else {
13996 help->dthps_actions[i] = next;
13997 }
13998
13999 dtrace_helper_action_destroy(h, vstate);
14000 } else {
14001 last = h;
14002 }
14003 }
14004 }
14005
14006 /*
14007 * Interate until we've cleared out all helper providers with the
14008 * given generation number.
14009 */
14010 for (;;) {
14011 dtrace_helper_provider_t *prov;
14012
14013 /*
14014 * Look for a helper provider with the right generation. We
14015 * have to start back at the beginning of the list each time
14016 * because we drop dtrace_lock. It's unlikely that we'll make
14017 * more than two passes.
14018 */
14019 for (i = 0; i < help->dthps_nprovs; i++) {
14020 prov = help->dthps_provs[i];
14021
14022 if (prov->dthp_generation == gen)
14023 break;
14024 }
14025
14026 /*
14027 * If there were no matches, we're done.
14028 */
14029 if (i == help->dthps_nprovs)
14030 break;
14031
14032 /*
14033 * Move the last helper provider into this slot.
14034 */
14035 help->dthps_nprovs--;
14036 help->dthps_provs[i] = help->dthps_provs[help->dthps_nprovs];
14037 help->dthps_provs[help->dthps_nprovs] = NULL;
14038
14039 mutex_exit(&dtrace_lock);
14040
14041 /*
14042 * If we have a meta provider, remove this helper provider.
14043 */
14044 mutex_enter(&dtrace_meta_lock);
14045 if (dtrace_meta_pid != NULL) {
14046 ASSERT(dtrace_deferred_pid == NULL);
14047 dtrace_helper_provider_remove(&prov->dthp_prov,
14048 p->p_pid);
14049 }
14050 mutex_exit(&dtrace_meta_lock);
14051
14052 dtrace_helper_provider_destroy(prov);
14053
14054 mutex_enter(&dtrace_lock);
14055 }
14056
14057 return (0);
14058}
14059#endif
14060
14061#if defined(sun)
14062static int
14063dtrace_helper_validate(dtrace_helper_action_t *helper)
14064{
14065 int err = 0, i;
14066 dtrace_difo_t *dp;
14067
14068 if ((dp = helper->dtha_predicate) != NULL)
14069 err += dtrace_difo_validate_helper(dp);
14070
14071 for (i = 0; i < helper->dtha_nactions; i++)
14072 err += dtrace_difo_validate_helper(helper->dtha_actions[i]);
14073
14074 return (err == 0);
14075}
14076#endif
14077
14078#if defined(sun)
14079static int
14080dtrace_helper_action_add(int which, dtrace_ecbdesc_t *ep)
14081{
14082 dtrace_helpers_t *help;
14083 dtrace_helper_action_t *helper, *last;
14084 dtrace_actdesc_t *act;
14085 dtrace_vstate_t *vstate;
14086 dtrace_predicate_t *pred;
14087 int count = 0, nactions = 0, i;
14088
14089 if (which < 0 || which >= DTRACE_NHELPER_ACTIONS)
14090 return (EINVAL);
14091
14092 help = curproc->p_dtrace_helpers;
14093 last = help->dthps_actions[which];
14094 vstate = &help->dthps_vstate;
14095
14096 for (count = 0; last != NULL; last = last->dtha_next) {
14097 count++;
14098 if (last->dtha_next == NULL)
14099 break;
14100 }
14101
14102 /*
14103 * If we already have dtrace_helper_actions_max helper actions for this
14104 * helper action type, we'll refuse to add a new one.
14105 */
14106 if (count >= dtrace_helper_actions_max)
14107 return (ENOSPC);
14108
14109 helper = kmem_zalloc(sizeof (dtrace_helper_action_t), KM_SLEEP);
14110 helper->dtha_generation = help->dthps_generation;
14111
14112 if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) {
14113 ASSERT(pred->dtp_difo != NULL);
14114 dtrace_difo_hold(pred->dtp_difo);
14115 helper->dtha_predicate = pred->dtp_difo;
14116 }
14117
14118 for (act = ep->dted_action; act != NULL; act = act->dtad_next) {
14119 if (act->dtad_kind != DTRACEACT_DIFEXPR)
14120 goto err;
14121
14122 if (act->dtad_difo == NULL)
14123 goto err;
14124
14125 nactions++;
14126 }
14127
14128 helper->dtha_actions = kmem_zalloc(sizeof (dtrace_difo_t *) *
14129 (helper->dtha_nactions = nactions), KM_SLEEP);
14130
14131 for (act = ep->dted_action, i = 0; act != NULL; act = act->dtad_next) {
14132 dtrace_difo_hold(act->dtad_difo);
14133 helper->dtha_actions[i++] = act->dtad_difo;
14134 }
14135
14136 if (!dtrace_helper_validate(helper))
14137 goto err;
14138
14139 if (last == NULL) {
14140 help->dthps_actions[which] = helper;
14141 } else {
14142 last->dtha_next = helper;
14143 }
14144
14145 if (vstate->dtvs_nlocals > dtrace_helptrace_nlocals) {
14146 dtrace_helptrace_nlocals = vstate->dtvs_nlocals;
14147 dtrace_helptrace_next = 0;
14148 }
14149
14150 return (0);
14151err:
14152 dtrace_helper_action_destroy(helper, vstate);
14153 return (EINVAL);
14154}
14155
14156static void
14157dtrace_helper_provider_register(proc_t *p, dtrace_helpers_t *help,
14158 dof_helper_t *dofhp)
14159{
14160 ASSERT(MUTEX_NOT_HELD(&dtrace_lock));
14161
14162 mutex_enter(&dtrace_meta_lock);
14163 mutex_enter(&dtrace_lock);
14164
14165 if (!dtrace_attached() || dtrace_meta_pid == NULL) {
14166 /*
14167 * If the dtrace module is loaded but not attached, or if
14168 * there aren't isn't a meta provider registered to deal with
14169 * these provider descriptions, we need to postpone creating
14170 * the actual providers until later.
14171 */
14172
14173 if (help->dthps_next == NULL && help->dthps_prev == NULL &&
14174 dtrace_deferred_pid != help) {
14175 help->dthps_deferred = 1;
14176 help->dthps_pid = p->p_pid;
14177 help->dthps_next = dtrace_deferred_pid;
14178 help->dthps_prev = NULL;
14179 if (dtrace_deferred_pid != NULL)
14180 dtrace_deferred_pid->dthps_prev = help;
14181 dtrace_deferred_pid = help;
14182 }
14183
14184 mutex_exit(&dtrace_lock);
14185
14186 } else if (dofhp != NULL) {
14187 /*
14188 * If the dtrace module is loaded and we have a particular
14189 * helper provider description, pass that off to the
14190 * meta provider.
14191 */
14192
14193 mutex_exit(&dtrace_lock);
14194
14195 dtrace_helper_provide(dofhp, p->p_pid);
14196
14197 } else {
14198 /*
14199 * Otherwise, just pass all the helper provider descriptions
14200 * off to the meta provider.
14201 */
14202
14203 int i;
14204 mutex_exit(&dtrace_lock);
14205
14206 for (i = 0; i < help->dthps_nprovs; i++) {
14207 dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov,
14208 p->p_pid);
14209 }
14210 }
14211
14212 mutex_exit(&dtrace_meta_lock);
14213}
14214
14215static int
14216dtrace_helper_provider_add(dof_helper_t *dofhp, int gen)
14217{
14218 dtrace_helpers_t *help;
14219 dtrace_helper_provider_t *hprov, **tmp_provs;
14220 uint_t tmp_maxprovs, i;
14221
14222 ASSERT(MUTEX_HELD(&dtrace_lock));
14223
14224 help = curproc->p_dtrace_helpers;
14225 ASSERT(help != NULL);
14226
14227 /*
14228 * If we already have dtrace_helper_providers_max helper providers,
14229 * we're refuse to add a new one.
14230 */
14231 if (help->dthps_nprovs >= dtrace_helper_providers_max)
14232 return (ENOSPC);
14233
14234 /*
14235 * Check to make sure this isn't a duplicate.
14236 */
14237 for (i = 0; i < help->dthps_nprovs; i++) {
14238 if (dofhp->dofhp_addr ==
14239 help->dthps_provs[i]->dthp_prov.dofhp_addr)
14240 return (EALREADY);
14241 }
14242
14243 hprov = kmem_zalloc(sizeof (dtrace_helper_provider_t), KM_SLEEP);
14244 hprov->dthp_prov = *dofhp;
14245 hprov->dthp_ref = 1;
14246 hprov->dthp_generation = gen;
14247
14248 /*
14249 * Allocate a bigger table for helper providers if it's already full.
14250 */
14251 if (help->dthps_maxprovs == help->dthps_nprovs) {
14252 tmp_maxprovs = help->dthps_maxprovs;
14253 tmp_provs = help->dthps_provs;
14254
14255 if (help->dthps_maxprovs == 0)
14256 help->dthps_maxprovs = 2;
14257 else
14258 help->dthps_maxprovs *= 2;
14259 if (help->dthps_maxprovs > dtrace_helper_providers_max)
14260 help->dthps_maxprovs = dtrace_helper_providers_max;
14261
14262 ASSERT(tmp_maxprovs < help->dthps_maxprovs);
14263
14264 help->dthps_provs = kmem_zalloc(help->dthps_maxprovs *
14265 sizeof (dtrace_helper_provider_t *), KM_SLEEP);
14266
14267 if (tmp_provs != NULL) {
14268 bcopy(tmp_provs, help->dthps_provs, tmp_maxprovs *
14269 sizeof (dtrace_helper_provider_t *));
14270 kmem_free(tmp_provs, tmp_maxprovs *
14271 sizeof (dtrace_helper_provider_t *));
14272 }
14273 }
14274
14275 help->dthps_provs[help->dthps_nprovs] = hprov;
14276 help->dthps_nprovs++;
14277
14278 return (0);
14279}
14280
14281static void
14282dtrace_helper_provider_destroy(dtrace_helper_provider_t *hprov)
14283{
14284 mutex_enter(&dtrace_lock);
14285
14286 if (--hprov->dthp_ref == 0) {
14287 dof_hdr_t *dof;
14288 mutex_exit(&dtrace_lock);
14289 dof = (dof_hdr_t *)(uintptr_t)hprov->dthp_prov.dofhp_dof;
14290 dtrace_dof_destroy(dof);
14291 kmem_free(hprov, sizeof (dtrace_helper_provider_t));
14292 } else {
14293 mutex_exit(&dtrace_lock);
14294 }
14295}
14296
14297static int
14298dtrace_helper_provider_validate(dof_hdr_t *dof, dof_sec_t *sec)
14299{
14300 uintptr_t daddr = (uintptr_t)dof;
14301 dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec;
14302 dof_provider_t *provider;
14303 dof_probe_t *probe;
14304 uint8_t *arg;
14305 char *strtab, *typestr;
14306 dof_stridx_t typeidx;
14307 size_t typesz;
14308 uint_t nprobes, j, k;
14309
14310 ASSERT(sec->dofs_type == DOF_SECT_PROVIDER);
14311
14312 if (sec->dofs_offset & (sizeof (uint_t) - 1)) {
14313 dtrace_dof_error(dof, "misaligned section offset");
14314 return (-1);
14315 }
14316
14317 /*
14318 * The section needs to be large enough to contain the DOF provider
14319 * structure appropriate for the given version.
14320 */
14321 if (sec->dofs_size <
14322 ((dof->dofh_ident[DOF_ID_VERSION] == DOF_VERSION_1) ?
14323 offsetof(dof_provider_t, dofpv_prenoffs) :
14324 sizeof (dof_provider_t))) {
14325 dtrace_dof_error(dof, "provider section too small");
14326 return (-1);
14327 }
14328
14329 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset);
14330 str_sec = dtrace_dof_sect(dof, DOF_SECT_STRTAB, provider->dofpv_strtab);
14331 prb_sec = dtrace_dof_sect(dof, DOF_SECT_PROBES, provider->dofpv_probes);
14332 arg_sec = dtrace_dof_sect(dof, DOF_SECT_PRARGS, provider->dofpv_prargs);
14333 off_sec = dtrace_dof_sect(dof, DOF_SECT_PROFFS, provider->dofpv_proffs);
14334
14335 if (str_sec == NULL || prb_sec == NULL ||
14336 arg_sec == NULL || off_sec == NULL)
14337 return (-1);
14338
14339 enoff_sec = NULL;
14340
14341 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 &&
14342 provider->dofpv_prenoffs != DOF_SECT_NONE &&
14343 (enoff_sec = dtrace_dof_sect(dof, DOF_SECT_PRENOFFS,
14344 provider->dofpv_prenoffs)) == NULL)
14345 return (-1);
14346
14347 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset);
14348
14349 if (provider->dofpv_name >= str_sec->dofs_size ||
14350 strlen(strtab + provider->dofpv_name) >= DTRACE_PROVNAMELEN) {
14351 dtrace_dof_error(dof, "invalid provider name");
14352 return (-1);
14353 }
14354
14355 if (prb_sec->dofs_entsize == 0 ||
14356 prb_sec->dofs_entsize > prb_sec->dofs_size) {
14357 dtrace_dof_error(dof, "invalid entry size");
14358 return (-1);
14359 }
14360
14361 if (prb_sec->dofs_entsize & (sizeof (uintptr_t) - 1)) {
14362 dtrace_dof_error(dof, "misaligned entry size");
14363 return (-1);
14364 }
14365
14366 if (off_sec->dofs_entsize != sizeof (uint32_t)) {
14367 dtrace_dof_error(dof, "invalid entry size");
14368 return (-1);
14369 }
14370
14371 if (off_sec->dofs_offset & (sizeof (uint32_t) - 1)) {
14372 dtrace_dof_error(dof, "misaligned section offset");
14373 return (-1);
14374 }
14375
14376 if (arg_sec->dofs_entsize != sizeof (uint8_t)) {
14377 dtrace_dof_error(dof, "invalid entry size");
14378 return (-1);
14379 }
14380
14381 arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset);
14382
14383 nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize;
14384
14385 /*
14386 * Take a pass through the probes to check for errors.
14387 */
14388 for (j = 0; j < nprobes; j++) {
14389 probe = (dof_probe_t *)(uintptr_t)(daddr +
14390 prb_sec->dofs_offset + j * prb_sec->dofs_entsize);
14391
14392 if (probe->dofpr_func >= str_sec->dofs_size) {
14393 dtrace_dof_error(dof, "invalid function name");
14394 return (-1);
14395 }
14396
14397 if (strlen(strtab + probe->dofpr_func) >= DTRACE_FUNCNAMELEN) {
14398 dtrace_dof_error(dof, "function name too long");
14399 return (-1);
14400 }
14401
14402 if (probe->dofpr_name >= str_sec->dofs_size ||
14403 strlen(strtab + probe->dofpr_name) >= DTRACE_NAMELEN) {
14404 dtrace_dof_error(dof, "invalid probe name");
14405 return (-1);
14406 }
14407
14408 /*
14409 * The offset count must not wrap the index, and the offsets
14410 * must also not overflow the section's data.
14411 */
14412 if (probe->dofpr_offidx + probe->dofpr_noffs <
14413 probe->dofpr_offidx ||
14414 (probe->dofpr_offidx + probe->dofpr_noffs) *
14415 off_sec->dofs_entsize > off_sec->dofs_size) {
14416 dtrace_dof_error(dof, "invalid probe offset");
14417 return (-1);
14418 }
14419
14420 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1) {
14421 /*
14422 * If there's no is-enabled offset section, make sure
14423 * there aren't any is-enabled offsets. Otherwise
14424 * perform the same checks as for probe offsets
14425 * (immediately above).
14426 */
14427 if (enoff_sec == NULL) {
14428 if (probe->dofpr_enoffidx != 0 ||
14429 probe->dofpr_nenoffs != 0) {
14430 dtrace_dof_error(dof, "is-enabled "
14431 "offsets with null section");
14432 return (-1);
14433 }
14434 } else if (probe->dofpr_enoffidx +
14435 probe->dofpr_nenoffs < probe->dofpr_enoffidx ||
14436 (probe->dofpr_enoffidx + probe->dofpr_nenoffs) *
14437 enoff_sec->dofs_entsize > enoff_sec->dofs_size) {
14438 dtrace_dof_error(dof, "invalid is-enabled "
14439 "offset");
14440 return (-1);
14441 }
14442
14443 if (probe->dofpr_noffs + probe->dofpr_nenoffs == 0) {
14444 dtrace_dof_error(dof, "zero probe and "
14445 "is-enabled offsets");
14446 return (-1);
14447 }
14448 } else if (probe->dofpr_noffs == 0) {
14449 dtrace_dof_error(dof, "zero probe offsets");
14450 return (-1);
14451 }
14452
14453 if (probe->dofpr_argidx + probe->dofpr_xargc <
14454 probe->dofpr_argidx ||
14455 (probe->dofpr_argidx + probe->dofpr_xargc) *
14456 arg_sec->dofs_entsize > arg_sec->dofs_size) {
14457 dtrace_dof_error(dof, "invalid args");
14458 return (-1);
14459 }
14460
14461 typeidx = probe->dofpr_nargv;
14462 typestr = strtab + probe->dofpr_nargv;
14463 for (k = 0; k < probe->dofpr_nargc; k++) {
14464 if (typeidx >= str_sec->dofs_size) {
14465 dtrace_dof_error(dof, "bad "
14466 "native argument type");
14467 return (-1);
14468 }
14469
14470 typesz = strlen(typestr) + 1;
14471 if (typesz > DTRACE_ARGTYPELEN) {
14472 dtrace_dof_error(dof, "native "
14473 "argument type too long");
14474 return (-1);
14475 }
14476 typeidx += typesz;
14477 typestr += typesz;
14478 }
14479
14480 typeidx = probe->dofpr_xargv;
14481 typestr = strtab + probe->dofpr_xargv;
14482 for (k = 0; k < probe->dofpr_xargc; k++) {
14483 if (arg[probe->dofpr_argidx + k] > probe->dofpr_nargc) {
14484 dtrace_dof_error(dof, "bad "
14485 "native argument index");
14486 return (-1);
14487 }
14488
14489 if (typeidx >= str_sec->dofs_size) {
14490 dtrace_dof_error(dof, "bad "
14491 "translated argument type");
14492 return (-1);
14493 }
14494
14495 typesz = strlen(typestr) + 1;
14496 if (typesz > DTRACE_ARGTYPELEN) {
14497 dtrace_dof_error(dof, "translated argument "
14498 "type too long");
14499 return (-1);
14500 }
14501
14502 typeidx += typesz;
14503 typestr += typesz;
14504 }
14505 }
14506
14507 return (0);
14508}
14509
14510static int
14511dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp)
14512{
14513 dtrace_helpers_t *help;
14514 dtrace_vstate_t *vstate;
14515 dtrace_enabling_t *enab = NULL;
14516 int i, gen, rv, nhelpers = 0, nprovs = 0, destroy = 1;
14517 uintptr_t daddr = (uintptr_t)dof;
14518
14519 ASSERT(MUTEX_HELD(&dtrace_lock));
14520
14521 if ((help = curproc->p_dtrace_helpers) == NULL)
14522 help = dtrace_helpers_create(curproc);
14523
14524 vstate = &help->dthps_vstate;
14525
14526 if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab,
14527 dhp != NULL ? dhp->dofhp_addr : 0, B_FALSE)) != 0) {
14528 dtrace_dof_destroy(dof);
14529 return (rv);
14530 }
14531
14532 /*
14533 * Look for helper providers and validate their descriptions.
14534 */
14535 if (dhp != NULL) {
14536 for (i = 0; i < dof->dofh_secnum; i++) {
14537 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr +
14538 dof->dofh_secoff + i * dof->dofh_secsize);
14539
14540 if (sec->dofs_type != DOF_SECT_PROVIDER)
14541 continue;
14542
14543 if (dtrace_helper_provider_validate(dof, sec) != 0) {
14544 dtrace_enabling_destroy(enab);
14545 dtrace_dof_destroy(dof);
14546 return (-1);
14547 }
14548
14549 nprovs++;
14550 }
14551 }
14552
14553 /*
14554 * Now we need to walk through the ECB descriptions in the enabling.
14555 */
14556 for (i = 0; i < enab->dten_ndesc; i++) {
14557 dtrace_ecbdesc_t *ep = enab->dten_desc[i];
14558 dtrace_probedesc_t *desc = &ep->dted_probe;
14559
14560 if (strcmp(desc->dtpd_provider, "dtrace") != 0)
14561 continue;
14562
14563 if (strcmp(desc->dtpd_mod, "helper") != 0)
14564 continue;
14565
14566 if (strcmp(desc->dtpd_func, "ustack") != 0)
14567 continue;
14568
14569 if ((rv = dtrace_helper_action_add(DTRACE_HELPER_ACTION_USTACK,
14570 ep)) != 0) {
14571 /*
14572 * Adding this helper action failed -- we are now going
14573 * to rip out the entire generation and return failure.
14574 */
14575 (void) dtrace_helper_destroygen(help->dthps_generation);
14576 dtrace_enabling_destroy(enab);
14577 dtrace_dof_destroy(dof);
14578 return (-1);
14579 }
14580
14581 nhelpers++;
14582 }
14583
14584 if (nhelpers < enab->dten_ndesc)
14585 dtrace_dof_error(dof, "unmatched helpers");
14586
14587 gen = help->dthps_generation++;
14588 dtrace_enabling_destroy(enab);
14589
14590 if (dhp != NULL && nprovs > 0) {
14591 dhp->dofhp_dof = (uint64_t)(uintptr_t)dof;
14592 if (dtrace_helper_provider_add(dhp, gen) == 0) {
14593 mutex_exit(&dtrace_lock);
14594 dtrace_helper_provider_register(curproc, help, dhp);
14595 mutex_enter(&dtrace_lock);
14596
14597 destroy = 0;
14598 }
14599 }
14600
14601 if (destroy)
14602 dtrace_dof_destroy(dof);
14603
14604 return (gen);
14605}
14606
14607static dtrace_helpers_t *
14608dtrace_helpers_create(proc_t *p)
14609{
14610 dtrace_helpers_t *help;
14611
14612 ASSERT(MUTEX_HELD(&dtrace_lock));
14613 ASSERT(p->p_dtrace_helpers == NULL);
14614
14615 help = kmem_zalloc(sizeof (dtrace_helpers_t), KM_SLEEP);
14616 help->dthps_actions = kmem_zalloc(sizeof (dtrace_helper_action_t *) *
14617 DTRACE_NHELPER_ACTIONS, KM_SLEEP);
14618
14619 p->p_dtrace_helpers = help;
14620 dtrace_helpers++;
14621
14622 return (help);
14623}
14624
14625static void
14626dtrace_helpers_destroy(void)
14627{
14628 dtrace_helpers_t *help;
14629 dtrace_vstate_t *vstate;
14630 proc_t *p = curproc;
14631 int i;
14632
14633 mutex_enter(&dtrace_lock);
14634
14635 ASSERT(p->p_dtrace_helpers != NULL);
14636 ASSERT(dtrace_helpers > 0);
14637
14638 help = p->p_dtrace_helpers;
14639 vstate = &help->dthps_vstate;
14640
14641 /*
14642 * We're now going to lose the help from this process.
14643 */
14644 p->p_dtrace_helpers = NULL;
14645 dtrace_sync();
14646
14647 /*
14648 * Destory the helper actions.
14649 */
14650 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
14651 dtrace_helper_action_t *h, *next;
14652
14653 for (h = help->dthps_actions[i]; h != NULL; h = next) {
14654 next = h->dtha_next;
14655 dtrace_helper_action_destroy(h, vstate);
14656 h = next;
14657 }
14658 }
14659
14660 mutex_exit(&dtrace_lock);
14661
14662 /*
14663 * Destroy the helper providers.
14664 */
14665 if (help->dthps_maxprovs > 0) {
14666 mutex_enter(&dtrace_meta_lock);
14667 if (dtrace_meta_pid != NULL) {
14668 ASSERT(dtrace_deferred_pid == NULL);
14669
14670 for (i = 0; i < help->dthps_nprovs; i++) {
14671 dtrace_helper_provider_remove(
14672 &help->dthps_provs[i]->dthp_prov, p->p_pid);
14673 }
14674 } else {
14675 mutex_enter(&dtrace_lock);
14676 ASSERT(help->dthps_deferred == 0 ||
14677 help->dthps_next != NULL ||
14678 help->dthps_prev != NULL ||
14679 help == dtrace_deferred_pid);
14680
14681 /*
14682 * Remove the helper from the deferred list.
14683 */
14684 if (help->dthps_next != NULL)
14685 help->dthps_next->dthps_prev = help->dthps_prev;
14686 if (help->dthps_prev != NULL)
14687 help->dthps_prev->dthps_next = help->dthps_next;
14688 if (dtrace_deferred_pid == help) {
14689 dtrace_deferred_pid = help->dthps_next;
14690 ASSERT(help->dthps_prev == NULL);
14691 }
14692
14693 mutex_exit(&dtrace_lock);
14694 }
14695
14696 mutex_exit(&dtrace_meta_lock);
14697
14698 for (i = 0; i < help->dthps_nprovs; i++) {
14699 dtrace_helper_provider_destroy(help->dthps_provs[i]);
14700 }
14701
14702 kmem_free(help->dthps_provs, help->dthps_maxprovs *
14703 sizeof (dtrace_helper_provider_t *));
14704 }
14705
14706 mutex_enter(&dtrace_lock);
14707
14708 dtrace_vstate_fini(&help->dthps_vstate);
14709 kmem_free(help->dthps_actions,
14710 sizeof (dtrace_helper_action_t *) * DTRACE_NHELPER_ACTIONS);
14711 kmem_free(help, sizeof (dtrace_helpers_t));
14712
14713 --dtrace_helpers;
14714 mutex_exit(&dtrace_lock);
14715}
14716
14717static void
14718dtrace_helpers_duplicate(proc_t *from, proc_t *to)
14719{
14720 dtrace_helpers_t *help, *newhelp;
14721 dtrace_helper_action_t *helper, *new, *last;
14722 dtrace_difo_t *dp;
14723 dtrace_vstate_t *vstate;
14724 int i, j, sz, hasprovs = 0;
14725
14726 mutex_enter(&dtrace_lock);
14727 ASSERT(from->p_dtrace_helpers != NULL);
14728 ASSERT(dtrace_helpers > 0);
14729
14730 help = from->p_dtrace_helpers;
14731 newhelp = dtrace_helpers_create(to);
14732 ASSERT(to->p_dtrace_helpers != NULL);
14733
14734 newhelp->dthps_generation = help->dthps_generation;
14735 vstate = &newhelp->dthps_vstate;
14736
14737 /*
14738 * Duplicate the helper actions.
14739 */
14740 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) {
14741 if ((helper = help->dthps_actions[i]) == NULL)
14742 continue;
14743
14744 for (last = NULL; helper != NULL; helper = helper->dtha_next) {
14745 new = kmem_zalloc(sizeof (dtrace_helper_action_t),
14746 KM_SLEEP);
14747 new->dtha_generation = helper->dtha_generation;
14748
14749 if ((dp = helper->dtha_predicate) != NULL) {
14750 dp = dtrace_difo_duplicate(dp, vstate);
14751 new->dtha_predicate = dp;
14752 }
14753
14754 new->dtha_nactions = helper->dtha_nactions;
14755 sz = sizeof (dtrace_difo_t *) * new->dtha_nactions;
14756 new->dtha_actions = kmem_alloc(sz, KM_SLEEP);
14757
14758 for (j = 0; j < new->dtha_nactions; j++) {
14759 dtrace_difo_t *dp = helper->dtha_actions[j];
14760
14761 ASSERT(dp != NULL);
14762 dp = dtrace_difo_duplicate(dp, vstate);
14763 new->dtha_actions[j] = dp;
14764 }
14765
14766 if (last != NULL) {
14767 last->dtha_next = new;
14768 } else {
14769 newhelp->dthps_actions[i] = new;
14770 }
14771
14772 last = new;
14773 }
14774 }
14775
14776 /*
14777 * Duplicate the helper providers and register them with the
14778 * DTrace framework.
14779 */
14780 if (help->dthps_nprovs > 0) {
14781 newhelp->dthps_nprovs = help->dthps_nprovs;
14782 newhelp->dthps_maxprovs = help->dthps_nprovs;
14783 newhelp->dthps_provs = kmem_alloc(newhelp->dthps_nprovs *
14784 sizeof (dtrace_helper_provider_t *), KM_SLEEP);
14785 for (i = 0; i < newhelp->dthps_nprovs; i++) {
14786 newhelp->dthps_provs[i] = help->dthps_provs[i];
14787 newhelp->dthps_provs[i]->dthp_ref++;
14788 }
14789
14790 hasprovs = 1;
14791 }
14792
14793 mutex_exit(&dtrace_lock);
14794
14795 if (hasprovs)
14796 dtrace_helper_provider_register(to, newhelp, NULL);
14797}
14798#endif
14799
14800#if defined(sun)
14801/*
14802 * DTrace Hook Functions
14803 */
14804static void
14805dtrace_module_loaded(modctl_t *ctl)
14806{
14807 dtrace_provider_t *prv;
14808
14809 mutex_enter(&dtrace_provider_lock);
14810 mutex_enter(&mod_lock);
14811
14812 ASSERT(ctl->mod_busy);
14813
14814 /*
14815 * We're going to call each providers per-module provide operation
14816 * specifying only this module.
14817 */
14818 for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next)
14819 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl);
14820
14821 mutex_exit(&mod_lock);
14822 mutex_exit(&dtrace_provider_lock);
14823
14824 /*
14825 * If we have any retained enablings, we need to match against them.
14826 * Enabling probes requires that cpu_lock be held, and we cannot hold
14827 * cpu_lock here -- it is legal for cpu_lock to be held when loading a
14828 * module. (In particular, this happens when loading scheduling
14829 * classes.) So if we have any retained enablings, we need to dispatch
14830 * our task queue to do the match for us.
14831 */
14832 mutex_enter(&dtrace_lock);
14833
14834 if (dtrace_retained == NULL) {
14835 mutex_exit(&dtrace_lock);
14836 return;
14837 }
14838
14839 (void) taskq_dispatch(dtrace_taskq,
14840 (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP);
14841
14842 mutex_exit(&dtrace_lock);
14843
14844 /*
14845 * And now, for a little heuristic sleaze: in general, we want to
14846 * match modules as soon as they load. However, we cannot guarantee
14847 * this, because it would lead us to the lock ordering violation
14848 * outlined above. The common case, of course, is that cpu_lock is
14849 * _not_ held -- so we delay here for a clock tick, hoping that that's
14850 * long enough for the task queue to do its work. If it's not, it's
14851 * not a serious problem -- it just means that the module that we
14852 * just loaded may not be immediately instrumentable.
14853 */
14854 delay(1);
14855}
14856
14857static void
14858dtrace_module_unloaded(modctl_t *ctl)
14859{
14860 dtrace_probe_t template, *probe, *first, *next;
14861 dtrace_provider_t *prov;
14862
14863 template.dtpr_mod = ctl->mod_modname;
14864
14865 mutex_enter(&dtrace_provider_lock);
14866 mutex_enter(&mod_lock);
14867 mutex_enter(&dtrace_lock);
14868
14869 if (dtrace_bymod == NULL) {
14870 /*
14871 * The DTrace module is loaded (obviously) but not attached;
14872 * we don't have any work to do.
14873 */
14874 mutex_exit(&dtrace_provider_lock);
14875 mutex_exit(&mod_lock);
14876 mutex_exit(&dtrace_lock);
14877 return;
14878 }
14879
14880 for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template);
14881 probe != NULL; probe = probe->dtpr_nextmod) {
14882 if (probe->dtpr_ecb != NULL) {
14883 mutex_exit(&dtrace_provider_lock);
14884 mutex_exit(&mod_lock);
14885 mutex_exit(&dtrace_lock);
14886
14887 /*
14888 * This shouldn't _actually_ be possible -- we're
14889 * unloading a module that has an enabled probe in it.
14890 * (It's normally up to the provider to make sure that
14891 * this can't happen.) However, because dtps_enable()
14892 * doesn't have a failure mode, there can be an
14893 * enable/unload race. Upshot: we don't want to
14894 * assert, but we're not going to disable the
14895 * probe, either.
14896 */
14897 if (dtrace_err_verbose) {
14898 cmn_err(CE_WARN, "unloaded module '%s' had "
14899 "enabled probes", ctl->mod_modname);
14900 }
14901
14902 return;
14903 }
14904 }
14905
14906 probe = first;
14907
14908 for (first = NULL; probe != NULL; probe = next) {
14909 ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe);
14910
14911 dtrace_probes[probe->dtpr_id - 1] = NULL;
14912
14913 next = probe->dtpr_nextmod;
14914 dtrace_hash_remove(dtrace_bymod, probe);
14915 dtrace_hash_remove(dtrace_byfunc, probe);
14916 dtrace_hash_remove(dtrace_byname, probe);
14917
14918 if (first == NULL) {
14919 first = probe;
14920 probe->dtpr_nextmod = NULL;
14921 } else {
14922 probe->dtpr_nextmod = first;
14923 first = probe;
14924 }
14925 }
14926
14927 /*
14928 * We've removed all of the module's probes from the hash chains and
14929 * from the probe array. Now issue a dtrace_sync() to be sure that
14930 * everyone has cleared out from any probe array processing.
14931 */
14932 dtrace_sync();
14933
14934 for (probe = first; probe != NULL; probe = first) {
14935 first = probe->dtpr_nextmod;
14936 prov = probe->dtpr_provider;
14937 prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id,
14938 probe->dtpr_arg);
14939 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1);
14940 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1);
14941 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1);
14942 vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1);
14943 kmem_free(probe, sizeof (dtrace_probe_t));
14944 }
14945
14946 mutex_exit(&dtrace_lock);
14947 mutex_exit(&mod_lock);
14948 mutex_exit(&dtrace_provider_lock);
14949}
14950
14951static void
14952dtrace_suspend(void)
14953{
14954 dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_suspend));
14955}
14956
14957static void
14958dtrace_resume(void)
14959{
14960 dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_resume));
14961}
14962#endif
14963
14964static int
14965dtrace_cpu_setup(cpu_setup_t what, processorid_t cpu)
14966{
14967 ASSERT(MUTEX_HELD(&cpu_lock));
14968 mutex_enter(&dtrace_lock);
14969
14970 switch (what) {
14971 case CPU_CONFIG: {
14972 dtrace_state_t *state;
14973 dtrace_optval_t *opt, rs, c;
14974
14975 /*
14976 * For now, we only allocate a new buffer for anonymous state.
14977 */
14978 if ((state = dtrace_anon.dta_state) == NULL)
14979 break;
14980
14981 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE)
14982 break;
14983
14984 opt = state->dts_options;
14985 c = opt[DTRACEOPT_CPU];
14986
14987 if (c != DTRACE_CPUALL && c != DTRACEOPT_UNSET && c != cpu)
14988 break;
14989
14990 /*
14991 * Regardless of what the actual policy is, we're going to
14992 * temporarily set our resize policy to be manual. We're
14993 * also going to temporarily set our CPU option to denote
14994 * the newly configured CPU.
14995 */
14996 rs = opt[DTRACEOPT_BUFRESIZE];
14997 opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_MANUAL;
14998 opt[DTRACEOPT_CPU] = (dtrace_optval_t)cpu;
14999
15000 (void) dtrace_state_buffers(state);
15001
15002 opt[DTRACEOPT_BUFRESIZE] = rs;
15003 opt[DTRACEOPT_CPU] = c;
15004
15005 break;
15006 }
15007
15008 case CPU_UNCONFIG:
15009 /*
15010 * We don't free the buffer in the CPU_UNCONFIG case. (The
15011 * buffer will be freed when the consumer exits.)
15012 */
15013 break;
15014
15015 default:
15016 break;
15017 }
15018
15019 mutex_exit(&dtrace_lock);
15020 return (0);
15021}
15022
15023#if defined(sun)
15024static void
15025dtrace_cpu_setup_initial(processorid_t cpu)
15026{
15027 (void) dtrace_cpu_setup(CPU_CONFIG, cpu);
15028}
15029#endif
15030
15031static void
15032dtrace_toxrange_add(uintptr_t base, uintptr_t limit)
15033{
15034 if (dtrace_toxranges >= dtrace_toxranges_max) {
15035 int osize, nsize;
15036 dtrace_toxrange_t *range;
15037
15038 osize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t);
15039
15040 if (osize == 0) {
15041 ASSERT(dtrace_toxrange == NULL);
15042 ASSERT(dtrace_toxranges_max == 0);
15043 dtrace_toxranges_max = 1;
15044 } else {
15045 dtrace_toxranges_max <<= 1;
15046 }
15047
15048 nsize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t);
15049 range = kmem_zalloc(nsize, KM_SLEEP);
15050
15051 if (dtrace_toxrange != NULL) {
15052 ASSERT(osize != 0);
15053 bcopy(dtrace_toxrange, range, osize);
15054 kmem_free(dtrace_toxrange, osize);
15055 }
15056
15057 dtrace_toxrange = range;
15058 }
15059
15060 ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == 0);
15061 ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == 0);
15062
15063 dtrace_toxrange[dtrace_toxranges].dtt_base = base;
15064 dtrace_toxrange[dtrace_toxranges].dtt_limit = limit;
15065 dtrace_toxranges++;
15066}
15067
15068/*
15069 * DTrace Driver Cookbook Functions
15070 */
15071#if defined(sun)
15072/*ARGSUSED*/
15073static int
15074dtrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
15075{
15076 dtrace_provider_id_t id;
15077 dtrace_state_t *state = NULL;
15078 dtrace_enabling_t *enab;
15079
15080 mutex_enter(&cpu_lock);
15081 mutex_enter(&dtrace_provider_lock);
15082 mutex_enter(&dtrace_lock);
15083
15084 if (ddi_soft_state_init(&dtrace_softstate,
15085 sizeof (dtrace_state_t), 0) != 0) {
15086 cmn_err(CE_NOTE, "/dev/dtrace failed to initialize soft state");
15087 mutex_exit(&cpu_lock);
15088 mutex_exit(&dtrace_provider_lock);
15089 mutex_exit(&dtrace_lock);
15090 return (DDI_FAILURE);
15091 }
15092
15093 if (ddi_create_minor_node(devi, DTRACEMNR_DTRACE, S_IFCHR,
15094 DTRACEMNRN_DTRACE, DDI_PSEUDO, NULL) == DDI_FAILURE ||
15095 ddi_create_minor_node(devi, DTRACEMNR_HELPER, S_IFCHR,
15096 DTRACEMNRN_HELPER, DDI_PSEUDO, NULL) == DDI_FAILURE) {
15097 cmn_err(CE_NOTE, "/dev/dtrace couldn't create minor nodes");
15098 ddi_remove_minor_node(devi, NULL);
15099 ddi_soft_state_fini(&dtrace_softstate);
15100 mutex_exit(&cpu_lock);
15101 mutex_exit(&dtrace_provider_lock);
15102 mutex_exit(&dtrace_lock);
15103 return (DDI_FAILURE);
15104 }
15105
15106 ddi_report_dev(devi);
15107 dtrace_devi = devi;
15108
15109 dtrace_modload = dtrace_module_loaded;
15110 dtrace_modunload = dtrace_module_unloaded;
15111 dtrace_cpu_init = dtrace_cpu_setup_initial;
15112 dtrace_helpers_cleanup = dtrace_helpers_destroy;
15113 dtrace_helpers_fork = dtrace_helpers_duplicate;
15114 dtrace_cpustart_init = dtrace_suspend;
15115 dtrace_cpustart_fini = dtrace_resume;
15116 dtrace_debugger_init = dtrace_suspend;
15117 dtrace_debugger_fini = dtrace_resume;
15118
15119 register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL);
15120
15121 ASSERT(MUTEX_HELD(&cpu_lock));
15122
15123 dtrace_arena = vmem_create("dtrace", (void *)1, UINT32_MAX, 1,
15124 NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
15125 dtrace_minor = vmem_create("dtrace_minor", (void *)DTRACEMNRN_CLONE,
15126 UINT32_MAX - DTRACEMNRN_CLONE, 1, NULL, NULL, NULL, 0,
15127 VM_SLEEP | VMC_IDENTIFIER);
15128 dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri,
15129 1, INT_MAX, 0);
15130
15131 dtrace_state_cache = kmem_cache_create("dtrace_state_cache",
15132 sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN,
15133 NULL, NULL, NULL, NULL, NULL, 0);
15134
15135 ASSERT(MUTEX_HELD(&cpu_lock));
15136 dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod),
15137 offsetof(dtrace_probe_t, dtpr_nextmod),
15138 offsetof(dtrace_probe_t, dtpr_prevmod));
15139
15140 dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func),
15141 offsetof(dtrace_probe_t, dtpr_nextfunc),
15142 offsetof(dtrace_probe_t, dtpr_prevfunc));
15143
15144 dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name),
15145 offsetof(dtrace_probe_t, dtpr_nextname),
15146 offsetof(dtrace_probe_t, dtpr_prevname));
15147
15148 if (dtrace_retain_max < 1) {
15149 cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; "
15150 "setting to 1", dtrace_retain_max);
15151 dtrace_retain_max = 1;
15152 }
15153
15154 /*
15155 * Now discover our toxic ranges.
15156 */
15157 dtrace_toxic_ranges(dtrace_toxrange_add);
15158
15159 /*
15160 * Before we register ourselves as a provider to our own framework,
15161 * we would like to assert that dtrace_provider is NULL -- but that's
15162 * not true if we were loaded as a dependency of a DTrace provider.
15163 * Once we've registered, we can assert that dtrace_provider is our
15164 * pseudo provider.
15165 */
15166 (void) dtrace_register("dtrace", &dtrace_provider_attr,
15167 DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id);
15168
15169 ASSERT(dtrace_provider != NULL);
15170 ASSERT((dtrace_provider_id_t)dtrace_provider == id);
15171
15172 dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t)
15173 dtrace_provider, NULL, NULL, "BEGIN", 0, NULL);
15174 dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t)
15175 dtrace_provider, NULL, NULL, "END", 0, NULL);
15176 dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t)
15177 dtrace_provider, NULL, NULL, "ERROR", 1, NULL);
15178
15179 dtrace_anon_property();
15180 mutex_exit(&cpu_lock);
15181
15182 /*
15183 * If DTrace helper tracing is enabled, we need to allocate the
15184 * trace buffer and initialize the values.
15185 */
15186 if (dtrace_helptrace_enabled) {
15187 ASSERT(dtrace_helptrace_buffer == NULL);
15188 dtrace_helptrace_buffer =
15189 kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP);
15190 dtrace_helptrace_next = 0;
15191 }
15192
15193 /*
15194 * If there are already providers, we must ask them to provide their
15195 * probes, and then match any anonymous enabling against them. Note
15196 * that there should be no other retained enablings at this time:
15197 * the only retained enablings at this time should be the anonymous
15198 * enabling.
15199 */
15200 if (dtrace_anon.dta_enabling != NULL) {
15201 ASSERT(dtrace_retained == dtrace_anon.dta_enabling);
15202
15203 dtrace_enabling_provide(NULL);
15204 state = dtrace_anon.dta_state;
15205
15206 /*
15207 * We couldn't hold cpu_lock across the above call to
15208 * dtrace_enabling_provide(), but we must hold it to actually
15209 * enable the probes. We have to drop all of our locks, pick
15210 * up cpu_lock, and regain our locks before matching the
15211 * retained anonymous enabling.
15212 */
15213 mutex_exit(&dtrace_lock);
15214 mutex_exit(&dtrace_provider_lock);
15215
15216 mutex_enter(&cpu_lock);
15217 mutex_enter(&dtrace_provider_lock);
15218 mutex_enter(&dtrace_lock);
15219
15220 if ((enab = dtrace_anon.dta_enabling) != NULL)
15221 (void) dtrace_enabling_match(enab, NULL);
15222
15223 mutex_exit(&cpu_lock);
15224 }
15225
15226 mutex_exit(&dtrace_lock);
15227 mutex_exit(&dtrace_provider_lock);
15228
15229 if (state != NULL) {
15230 /*
15231 * If we created any anonymous state, set it going now.
15232 */
15233 (void) dtrace_state_go(state, &dtrace_anon.dta_beganon);
15234 }
15235
15236 return (DDI_SUCCESS);
15237}
15238#endif
15239
15240#if !defined(sun)
15241#if __FreeBSD_version >= 800039
15242static void
15243dtrace_dtr(void *data __unused)
15244{
15245}
15246#endif
15247#endif
15248
15249/*ARGSUSED*/
15250static int
15251#if defined(sun)
15252dtrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
15253#else
15254dtrace_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
15255#endif
15256{
15257 dtrace_state_t *state;
15258 uint32_t priv;
15259 uid_t uid;
15260 zoneid_t zoneid;
15261
15262#if defined(sun)
15263 if (getminor(*devp) == DTRACEMNRN_HELPER)
15264 return (0);
15265
15266 /*
15267 * If this wasn't an open with the "helper" minor, then it must be
15268 * the "dtrace" minor.
15269 */
15270 ASSERT(getminor(*devp) == DTRACEMNRN_DTRACE);
15271#else
15272 cred_t *cred_p = NULL;
15273
15274#if __FreeBSD_version < 800039
15275 /*
15276 * The first minor device is the one that is cloned so there is
15277 * nothing more to do here.
15278 */
15279 if (dev2unit(dev) == 0)
15280 return 0;
15281
15282 /*
15283 * Devices are cloned, so if the DTrace state has already
15284 * been allocated, that means this device belongs to a
15285 * different client. Each client should open '/dev/dtrace'
15286 * to get a cloned device.
15287 */
15288 if (dev->si_drv1 != NULL)
15289 return (EBUSY);
15290#endif
15291
15292 cred_p = dev->si_cred;
15293#endif
15294
15295 /*
15296 * If no DTRACE_PRIV_* bits are set in the credential, then the
15297 * caller lacks sufficient permission to do anything with DTrace.
15298 */
15299 dtrace_cred2priv(cred_p, &priv, &uid, &zoneid);
15300 if (priv == DTRACE_PRIV_NONE) {
15301#if !defined(sun)
15302#if __FreeBSD_version < 800039
15303 /* Destroy the cloned device. */
15304 destroy_dev(dev);
15305#endif
15306#endif
15307
15308 return (EACCES);
15309 }
15310
15311 /*
15312 * Ask all providers to provide all their probes.
15313 */
15314 mutex_enter(&dtrace_provider_lock);
15315 dtrace_probe_provide(NULL, NULL);
15316 mutex_exit(&dtrace_provider_lock);
15317
15318 mutex_enter(&cpu_lock);
15319 mutex_enter(&dtrace_lock);
15320 dtrace_opens++;
15321 dtrace_membar_producer();
15322
15323#if defined(sun)
15324 /*
15325 * If the kernel debugger is active (that is, if the kernel debugger
15326 * modified text in some way), we won't allow the open.
15327 */
15328 if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) {
15329 dtrace_opens--;
15330 mutex_exit(&cpu_lock);
15331 mutex_exit(&dtrace_lock);
15332 return (EBUSY);
15333 }
15334
15335 state = dtrace_state_create(devp, cred_p);
15336#else
15337 state = dtrace_state_create(dev);
15338#if __FreeBSD_version < 800039
15339 dev->si_drv1 = state;
15340#else
15341 devfs_set_cdevpriv(state, dtrace_dtr);
15342#endif
15343#endif
15344
15345 mutex_exit(&cpu_lock);
15346
15347 if (state == NULL) {
15348#if defined(sun)
15349 if (--dtrace_opens == 0)
15350 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
15351#else
15352 --dtrace_opens;
15353#endif
15354 mutex_exit(&dtrace_lock);
15355#if !defined(sun)
15356#if __FreeBSD_version < 800039
15357 /* Destroy the cloned device. */
15358 destroy_dev(dev);
15359#endif
15360#endif
15361 return (EAGAIN);
15362 }
15363
15364 mutex_exit(&dtrace_lock);
15365
15366 return (0);
15367}
15368
15369/*ARGSUSED*/
15370static int
15371#if defined(sun)
15372dtrace_close(dev_t dev, int flag, int otyp, cred_t *cred_p)
15373#else
15374dtrace_close(struct cdev *dev, int flags, int fmt __unused, struct thread *td)
15375#endif
15376{
15377#if defined(sun)
15378 minor_t minor = getminor(dev);
15379 dtrace_state_t *state;
15380
15381 if (minor == DTRACEMNRN_HELPER)
15382 return (0);
15383
15384 state = ddi_get_soft_state(dtrace_softstate, minor);
15385#else
15386#if __FreeBSD_version < 800039
15387 dtrace_state_t *state = dev->si_drv1;
15388
15389 /* Check if this is not a cloned device. */
15390 if (dev2unit(dev) == 0)
15391 return (0);
15392#else
15393 dtrace_state_t *state;
15394 devfs_get_cdevpriv((void **) &state);
15395#endif
15396
15397#endif
15398
15399 mutex_enter(&cpu_lock);
15400 mutex_enter(&dtrace_lock);
15401
15402 if (state != NULL) {
15403 if (state->dts_anon) {
15404 /*
15405 * There is anonymous state. Destroy that first.
15406 */
15407 ASSERT(dtrace_anon.dta_state == NULL);
15408 dtrace_state_destroy(state->dts_anon);
15409 }
15410
15411 dtrace_state_destroy(state);
15412
15413#if !defined(sun)
15414 kmem_free(state, 0);
15415#if __FreeBSD_version < 800039
15416 dev->si_drv1 = NULL;
15417#else
15418 devfs_clear_cdevpriv();
15419#endif
15420#endif
15421 }
15422
15423 ASSERT(dtrace_opens > 0);
15424#if defined(sun)
15425 if (--dtrace_opens == 0)
15426 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
15427#else
15428 --dtrace_opens;
15429#endif
15430
15431 mutex_exit(&dtrace_lock);
15432 mutex_exit(&cpu_lock);
15433
15434#if __FreeBSD_version < 800039
15435 /* Schedule this cloned device to be destroyed. */
15436 destroy_dev_sched(dev);
15437#endif
15438
15439 return (0);
15440}
15441
15442#if defined(sun)
15443/*ARGSUSED*/
15444static int
15445dtrace_ioctl_helper(int cmd, intptr_t arg, int *rv)
15446{
15447 int rval;
15448 dof_helper_t help, *dhp = NULL;
15449
15450 switch (cmd) {
15451 case DTRACEHIOC_ADDDOF:
15452 if (copyin((void *)arg, &help, sizeof (help)) != 0) {
15453 dtrace_dof_error(NULL, "failed to copyin DOF helper");
15454 return (EFAULT);
15455 }
15456
15457 dhp = &help;
15458 arg = (intptr_t)help.dofhp_dof;
15459 /*FALLTHROUGH*/
15460
15461 case DTRACEHIOC_ADD: {
15462 dof_hdr_t *dof = dtrace_dof_copyin(arg, &rval);
15463
15464 if (dof == NULL)
15465 return (rval);
15466
15467 mutex_enter(&dtrace_lock);
15468
15469 /*
15470 * dtrace_helper_slurp() takes responsibility for the dof --
15471 * it may free it now or it may save it and free it later.
15472 */
15473 if ((rval = dtrace_helper_slurp(dof, dhp)) != -1) {
15474 *rv = rval;
15475 rval = 0;
15476 } else {
15477 rval = EINVAL;
15478 }
15479
15480 mutex_exit(&dtrace_lock);
15481 return (rval);
15482 }
15483
15484 case DTRACEHIOC_REMOVE: {
15485 mutex_enter(&dtrace_lock);
15486 rval = dtrace_helper_destroygen(arg);
15487 mutex_exit(&dtrace_lock);
15488
15489 return (rval);
15490 }
15491
15492 default:
15493 break;
15494 }
15495
15496 return (ENOTTY);
15497}
15498
15499/*ARGSUSED*/
15500static int
15501dtrace_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv)
15502{
15503 minor_t minor = getminor(dev);
15504 dtrace_state_t *state;
15505 int rval;
15506
15507 if (minor == DTRACEMNRN_HELPER)
15508 return (dtrace_ioctl_helper(cmd, arg, rv));
15509
15510 state = ddi_get_soft_state(dtrace_softstate, minor);
15511
15512 if (state->dts_anon) {
15513 ASSERT(dtrace_anon.dta_state == NULL);
15514 state = state->dts_anon;
15515 }
15516
15517 switch (cmd) {
15518 case DTRACEIOC_PROVIDER: {
15519 dtrace_providerdesc_t pvd;
15520 dtrace_provider_t *pvp;
15521
15522 if (copyin((void *)arg, &pvd, sizeof (pvd)) != 0)
15523 return (EFAULT);
15524
15525 pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0';
15526 mutex_enter(&dtrace_provider_lock);
15527
15528 for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) {
15529 if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0)
15530 break;
15531 }
15532
15533 mutex_exit(&dtrace_provider_lock);
15534
15535 if (pvp == NULL)
15536 return (ESRCH);
15537
15538 bcopy(&pvp->dtpv_priv, &pvd.dtvd_priv, sizeof (dtrace_ppriv_t));
15539 bcopy(&pvp->dtpv_attr, &pvd.dtvd_attr, sizeof (dtrace_pattr_t));
15540
15541 if (copyout(&pvd, (void *)arg, sizeof (pvd)) != 0)
15542 return (EFAULT);
15543
15544 return (0);
15545 }
15546
15547 case DTRACEIOC_EPROBE: {
15548 dtrace_eprobedesc_t epdesc;
15549 dtrace_ecb_t *ecb;
15550 dtrace_action_t *act;
15551 void *buf;
15552 size_t size;
15553 uintptr_t dest;
15554 int nrecs;
15555
15556 if (copyin((void *)arg, &epdesc, sizeof (epdesc)) != 0)
15557 return (EFAULT);
15558
15559 mutex_enter(&dtrace_lock);
15560
15561 if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) {
15562 mutex_exit(&dtrace_lock);
15563 return (EINVAL);
15564 }
15565
15566 if (ecb->dte_probe == NULL) {
15567 mutex_exit(&dtrace_lock);
15568 return (EINVAL);
15569 }
15570
15571 epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id;
15572 epdesc.dtepd_uarg = ecb->dte_uarg;
15573 epdesc.dtepd_size = ecb->dte_size;
15574
15575 nrecs = epdesc.dtepd_nrecs;
15576 epdesc.dtepd_nrecs = 0;
15577 for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
15578 if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple)
15579 continue;
15580
15581 epdesc.dtepd_nrecs++;
15582 }
15583
15584 /*
15585 * Now that we have the size, we need to allocate a temporary
15586 * buffer in which to store the complete description. We need
15587 * the temporary buffer to be able to drop dtrace_lock()
15588 * across the copyout(), below.
15589 */
15590 size = sizeof (dtrace_eprobedesc_t) +
15591 (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t));
15592
15593 buf = kmem_alloc(size, KM_SLEEP);
15594 dest = (uintptr_t)buf;
15595
15596 bcopy(&epdesc, (void *)dest, sizeof (epdesc));
15597 dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]);
15598
15599 for (act = ecb->dte_action; act != NULL; act = act->dta_next) {
15600 if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple)
15601 continue;
15602
15603 if (nrecs-- == 0)
15604 break;
15605
15606 bcopy(&act->dta_rec, (void *)dest,
15607 sizeof (dtrace_recdesc_t));
15608 dest += sizeof (dtrace_recdesc_t);
15609 }
15610
15611 mutex_exit(&dtrace_lock);
15612
15613 if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) {
15614 kmem_free(buf, size);
15615 return (EFAULT);
15616 }
15617
15618 kmem_free(buf, size);
15619 return (0);
15620 }
15621
15622 case DTRACEIOC_AGGDESC: {
15623 dtrace_aggdesc_t aggdesc;
15624 dtrace_action_t *act;
15625 dtrace_aggregation_t *agg;
15626 int nrecs;
15627 uint32_t offs;
15628 dtrace_recdesc_t *lrec;
15629 void *buf;
15630 size_t size;
15631 uintptr_t dest;
15632
15633 if (copyin((void *)arg, &aggdesc, sizeof (aggdesc)) != 0)
15634 return (EFAULT);
15635
15636 mutex_enter(&dtrace_lock);
15637
15638 if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) {
15639 mutex_exit(&dtrace_lock);
15640 return (EINVAL);
15641 }
15642
15643 aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid;
15644
15645 nrecs = aggdesc.dtagd_nrecs;
15646 aggdesc.dtagd_nrecs = 0;
15647
15648 offs = agg->dtag_base;
15649 lrec = &agg->dtag_action.dta_rec;
15650 aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs;
15651
15652 for (act = agg->dtag_first; ; act = act->dta_next) {
15653 ASSERT(act->dta_intuple ||
15654 DTRACEACT_ISAGG(act->dta_kind));
15655
15656 /*
15657 * If this action has a record size of zero, it
15658 * denotes an argument to the aggregating action.
15659 * Because the presence of this record doesn't (or
15660 * shouldn't) affect the way the data is interpreted,
15661 * we don't copy it out to save user-level the
15662 * confusion of dealing with a zero-length record.
15663 */
15664 if (act->dta_rec.dtrd_size == 0) {
15665 ASSERT(agg->dtag_hasarg);
15666 continue;
15667 }
15668
15669 aggdesc.dtagd_nrecs++;
15670
15671 if (act == &agg->dtag_action)
15672 break;
15673 }
15674
15675 /*
15676 * Now that we have the size, we need to allocate a temporary
15677 * buffer in which to store the complete description. We need
15678 * the temporary buffer to be able to drop dtrace_lock()
15679 * across the copyout(), below.
15680 */
15681 size = sizeof (dtrace_aggdesc_t) +
15682 (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t));
15683
15684 buf = kmem_alloc(size, KM_SLEEP);
15685 dest = (uintptr_t)buf;
15686
15687 bcopy(&aggdesc, (void *)dest, sizeof (aggdesc));
15688 dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]);
15689
15690 for (act = agg->dtag_first; ; act = act->dta_next) {
15691 dtrace_recdesc_t rec = act->dta_rec;
15692
15693 /*
15694 * See the comment in the above loop for why we pass
15695 * over zero-length records.
15696 */
15697 if (rec.dtrd_size == 0) {
15698 ASSERT(agg->dtag_hasarg);
15699 continue;
15700 }
15701
15702 if (nrecs-- == 0)
15703 break;
15704
15705 rec.dtrd_offset -= offs;
15706 bcopy(&rec, (void *)dest, sizeof (rec));
15707 dest += sizeof (dtrace_recdesc_t);
15708
15709 if (act == &agg->dtag_action)
15710 break;
15711 }
15712
15713 mutex_exit(&dtrace_lock);
15714
15715 if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) {
15716 kmem_free(buf, size);
15717 return (EFAULT);
15718 }
15719
15720 kmem_free(buf, size);
15721 return (0);
15722 }
15723
15724 case DTRACEIOC_ENABLE: {
15725 dof_hdr_t *dof;
15726 dtrace_enabling_t *enab = NULL;
15727 dtrace_vstate_t *vstate;
15728 int err = 0;
15729
15730 *rv = 0;
15731
15732 /*
15733 * If a NULL argument has been passed, we take this as our
15734 * cue to reevaluate our enablings.
15735 */
15736 if (arg == NULL) {
15737 dtrace_enabling_matchall();
15738
15739 return (0);
15740 }
15741
15742 if ((dof = dtrace_dof_copyin(arg, &rval)) == NULL)
15743 return (rval);
15744
15745 mutex_enter(&cpu_lock);
15746 mutex_enter(&dtrace_lock);
15747 vstate = &state->dts_vstate;
15748
15749 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) {
15750 mutex_exit(&dtrace_lock);
15751 mutex_exit(&cpu_lock);
15752 dtrace_dof_destroy(dof);
15753 return (EBUSY);
15754 }
15755
15756 if (dtrace_dof_slurp(dof, vstate, cr, &enab, 0, B_TRUE) != 0) {
15757 mutex_exit(&dtrace_lock);
15758 mutex_exit(&cpu_lock);
15759 dtrace_dof_destroy(dof);
15760 return (EINVAL);
15761 }
15762
15763 if ((rval = dtrace_dof_options(dof, state)) != 0) {
15764 dtrace_enabling_destroy(enab);
15765 mutex_exit(&dtrace_lock);
15766 mutex_exit(&cpu_lock);
15767 dtrace_dof_destroy(dof);
15768 return (rval);
15769 }
15770
15771 if ((err = dtrace_enabling_match(enab, rv)) == 0) {
15772 err = dtrace_enabling_retain(enab);
15773 } else {
15774 dtrace_enabling_destroy(enab);
15775 }
15776
15777 mutex_exit(&cpu_lock);
15778 mutex_exit(&dtrace_lock);
15779 dtrace_dof_destroy(dof);
15780
15781 return (err);
15782 }
15783
15784 case DTRACEIOC_REPLICATE: {
15785 dtrace_repldesc_t desc;
15786 dtrace_probedesc_t *match = &desc.dtrpd_match;
15787 dtrace_probedesc_t *create = &desc.dtrpd_create;
15788 int err;
15789
15790 if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
15791 return (EFAULT);
15792
15793 match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
15794 match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
15795 match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
15796 match->dtpd_name[DTRACE_NAMELEN - 1] = '\0';
15797
15798 create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
15799 create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
15800 create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
15801 create->dtpd_name[DTRACE_NAMELEN - 1] = '\0';
15802
15803 mutex_enter(&dtrace_lock);
15804 err = dtrace_enabling_replicate(state, match, create);
15805 mutex_exit(&dtrace_lock);
15806
15807 return (err);
15808 }
15809
15810 case DTRACEIOC_PROBEMATCH:
15811 case DTRACEIOC_PROBES: {
15812 dtrace_probe_t *probe = NULL;
15813 dtrace_probedesc_t desc;
15814 dtrace_probekey_t pkey;
15815 dtrace_id_t i;
15816 int m = 0;
15817 uint32_t priv;
15818 uid_t uid;
15819 zoneid_t zoneid;
15820
15821 if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
15822 return (EFAULT);
15823
15824 desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0';
15825 desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0';
15826 desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0';
15827 desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0';
15828
15829 /*
15830 * Before we attempt to match this probe, we want to give
15831 * all providers the opportunity to provide it.
15832 */
15833 if (desc.dtpd_id == DTRACE_IDNONE) {
15834 mutex_enter(&dtrace_provider_lock);
15835 dtrace_probe_provide(&desc, NULL);
15836 mutex_exit(&dtrace_provider_lock);
15837 desc.dtpd_id++;
15838 }
15839
15840 if (cmd == DTRACEIOC_PROBEMATCH) {
15841 dtrace_probekey(&desc, &pkey);
15842 pkey.dtpk_id = DTRACE_IDNONE;
15843 }
15844
15845 dtrace_cred2priv(cr, &priv, &uid, &zoneid);
15846
15847 mutex_enter(&dtrace_lock);
15848
15849 if (cmd == DTRACEIOC_PROBEMATCH) {
15850 for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) {
15851 if ((probe = dtrace_probes[i - 1]) != NULL &&
15852 (m = dtrace_match_probe(probe, &pkey,
15853 priv, uid, zoneid)) != 0)
15854 break;
15855 }
15856
15857 if (m < 0) {
15858 mutex_exit(&dtrace_lock);
15859 return (EINVAL);
15860 }
15861
15862 } else {
15863 for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) {
15864 if ((probe = dtrace_probes[i - 1]) != NULL &&
15865 dtrace_match_priv(probe, priv, uid, zoneid))
15866 break;
15867 }
15868 }
15869
15870 if (probe == NULL) {
15871 mutex_exit(&dtrace_lock);
15872 return (ESRCH);
15873 }
15874
15875 dtrace_probe_description(probe, &desc);
15876 mutex_exit(&dtrace_lock);
15877
15878 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
15879 return (EFAULT);
15880
15881 return (0);
15882 }
15883
15884 case DTRACEIOC_PROBEARG: {
15885 dtrace_argdesc_t desc;
15886 dtrace_probe_t *probe;
15887 dtrace_provider_t *prov;
15888
15889 if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
15890 return (EFAULT);
15891
15892 if (desc.dtargd_id == DTRACE_IDNONE)
15893 return (EINVAL);
15894
15895 if (desc.dtargd_ndx == DTRACE_ARGNONE)
15896 return (EINVAL);
15897
15898 mutex_enter(&dtrace_provider_lock);
15899 mutex_enter(&mod_lock);
15900 mutex_enter(&dtrace_lock);
15901
15902 if (desc.dtargd_id > dtrace_nprobes) {
15903 mutex_exit(&dtrace_lock);
15904 mutex_exit(&mod_lock);
15905 mutex_exit(&dtrace_provider_lock);
15906 return (EINVAL);
15907 }
15908
15909 if ((probe = dtrace_probes[desc.dtargd_id - 1]) == NULL) {
15910 mutex_exit(&dtrace_lock);
15911 mutex_exit(&mod_lock);
15912 mutex_exit(&dtrace_provider_lock);
15913 return (EINVAL);
15914 }
15915
15916 mutex_exit(&dtrace_lock);
15917
15918 prov = probe->dtpr_provider;
15919
15920 if (prov->dtpv_pops.dtps_getargdesc == NULL) {
15921 /*
15922 * There isn't any typed information for this probe.
15923 * Set the argument number to DTRACE_ARGNONE.
15924 */
15925 desc.dtargd_ndx = DTRACE_ARGNONE;
15926 } else {
15927 desc.dtargd_native[0] = '\0';
15928 desc.dtargd_xlate[0] = '\0';
15929 desc.dtargd_mapping = desc.dtargd_ndx;
15930
15931 prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg,
15932 probe->dtpr_id, probe->dtpr_arg, &desc);
15933 }
15934
15935 mutex_exit(&mod_lock);
15936 mutex_exit(&dtrace_provider_lock);
15937
15938 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
15939 return (EFAULT);
15940
15941 return (0);
15942 }
15943
15944 case DTRACEIOC_GO: {
15945 processorid_t cpuid;
15946 rval = dtrace_state_go(state, &cpuid);
15947
15948 if (rval != 0)
15949 return (rval);
15950
15951 if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0)
15952 return (EFAULT);
15953
15954 return (0);
15955 }
15956
15957 case DTRACEIOC_STOP: {
15958 processorid_t cpuid;
15959
15960 mutex_enter(&dtrace_lock);
15961 rval = dtrace_state_stop(state, &cpuid);
15962 mutex_exit(&dtrace_lock);
15963
15964 if (rval != 0)
15965 return (rval);
15966
15967 if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0)
15968 return (EFAULT);
15969
15970 return (0);
15971 }
15972
15973 case DTRACEIOC_DOFGET: {
15974 dof_hdr_t hdr, *dof;
15975 uint64_t len;
15976
15977 if (copyin((void *)arg, &hdr, sizeof (hdr)) != 0)
15978 return (EFAULT);
15979
15980 mutex_enter(&dtrace_lock);
15981 dof = dtrace_dof_create(state);
15982 mutex_exit(&dtrace_lock);
15983
15984 len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz);
15985 rval = copyout(dof, (void *)arg, len);
15986 dtrace_dof_destroy(dof);
15987
15988 return (rval == 0 ? 0 : EFAULT);
15989 }
15990
15991 case DTRACEIOC_AGGSNAP:
15992 case DTRACEIOC_BUFSNAP: {
15993 dtrace_bufdesc_t desc;
15994 caddr_t cached;
15995 dtrace_buffer_t *buf;
15996
15997 if (copyin((void *)arg, &desc, sizeof (desc)) != 0)
15998 return (EFAULT);
15999
16000 if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NCPU)
16001 return (EINVAL);
16002
16003 mutex_enter(&dtrace_lock);
16004
16005 if (cmd == DTRACEIOC_BUFSNAP) {
16006 buf = &state->dts_buffer[desc.dtbd_cpu];
16007 } else {
16008 buf = &state->dts_aggbuffer[desc.dtbd_cpu];
16009 }
16010
16011 if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) {
16012 size_t sz = buf->dtb_offset;
16013
16014 if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) {
16015 mutex_exit(&dtrace_lock);
16016 return (EBUSY);
16017 }
16018
16019 /*
16020 * If this buffer has already been consumed, we're
16021 * going to indicate that there's nothing left here
16022 * to consume.
16023 */
16024 if (buf->dtb_flags & DTRACEBUF_CONSUMED) {
16025 mutex_exit(&dtrace_lock);
16026
16027 desc.dtbd_size = 0;
16028 desc.dtbd_drops = 0;
16029 desc.dtbd_errors = 0;
16030 desc.dtbd_oldest = 0;
16031 sz = sizeof (desc);
16032
16033 if (copyout(&desc, (void *)arg, sz) != 0)
16034 return (EFAULT);
16035
16036 return (0);
16037 }
16038
16039 /*
16040 * If this is a ring buffer that has wrapped, we want
16041 * to copy the whole thing out.
16042 */
16043 if (buf->dtb_flags & DTRACEBUF_WRAPPED) {
16044 dtrace_buffer_polish(buf);
16045 sz = buf->dtb_size;
16046 }
16047
16048 if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) {
16049 mutex_exit(&dtrace_lock);
16050 return (EFAULT);
16051 }
16052
16053 desc.dtbd_size = sz;
16054 desc.dtbd_drops = buf->dtb_drops;
16055 desc.dtbd_errors = buf->dtb_errors;
16056 desc.dtbd_oldest = buf->dtb_xamot_offset;
16057
16058 mutex_exit(&dtrace_lock);
16059
16060 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16061 return (EFAULT);
16062
16063 buf->dtb_flags |= DTRACEBUF_CONSUMED;
16064
16065 return (0);
16066 }
16067
16068 if (buf->dtb_tomax == NULL) {
16069 ASSERT(buf->dtb_xamot == NULL);
16070 mutex_exit(&dtrace_lock);
16071 return (ENOENT);
16072 }
16073
16074 cached = buf->dtb_tomax;
16075 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH));
16076
16077 dtrace_xcall(desc.dtbd_cpu,
16078 (dtrace_xcall_t)dtrace_buffer_switch, buf);
16079
16080 state->dts_errors += buf->dtb_xamot_errors;
16081
16082 /*
16083 * If the buffers did not actually switch, then the cross call
16084 * did not take place -- presumably because the given CPU is
16085 * not in the ready set. If this is the case, we'll return
16086 * ENOENT.
16087 */
16088 if (buf->dtb_tomax == cached) {
16089 ASSERT(buf->dtb_xamot != cached);
16090 mutex_exit(&dtrace_lock);
16091 return (ENOENT);
16092 }
16093
16094 ASSERT(cached == buf->dtb_xamot);
16095
16096 /*
16097 * We have our snapshot; now copy it out.
16098 */
16099 if (copyout(buf->dtb_xamot, desc.dtbd_data,
16100 buf->dtb_xamot_offset) != 0) {
16101 mutex_exit(&dtrace_lock);
16102 return (EFAULT);
16103 }
16104
16105 desc.dtbd_size = buf->dtb_xamot_offset;
16106 desc.dtbd_drops = buf->dtb_xamot_drops;
16107 desc.dtbd_errors = buf->dtb_xamot_errors;
16108 desc.dtbd_oldest = 0;
16109
16110 mutex_exit(&dtrace_lock);
16111
16112 /*
16113 * Finally, copy out the buffer description.
16114 */
16115 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0)
16116 return (EFAULT);
16117
16118 return (0);
16119 }
16120
16121 case DTRACEIOC_CONF: {
16122 dtrace_conf_t conf;
16123
16124 bzero(&conf, sizeof (conf));
16125 conf.dtc_difversion = DIF_VERSION;
16126 conf.dtc_difintregs = DIF_DIR_NREGS;
16127 conf.dtc_diftupregs = DIF_DTR_NREGS;
16128 conf.dtc_ctfmodel = CTF_MODEL_NATIVE;
16129
16130 if (copyout(&conf, (void *)arg, sizeof (conf)) != 0)
16131 return (EFAULT);
16132
16133 return (0);
16134 }
16135
16136 case DTRACEIOC_STATUS: {
16137 dtrace_status_t stat;
16138 dtrace_dstate_t *dstate;
16139 int i, j;
16140 uint64_t nerrs;
16141
16142 /*
16143 * See the comment in dtrace_state_deadman() for the reason
16144 * for setting dts_laststatus to INT64_MAX before setting
16145 * it to the correct value.
16146 */
16147 state->dts_laststatus = INT64_MAX;
16148 dtrace_membar_producer();
16149 state->dts_laststatus = dtrace_gethrtime();
16150
16151 bzero(&stat, sizeof (stat));
16152
16153 mutex_enter(&dtrace_lock);
16154
16155 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) {
16156 mutex_exit(&dtrace_lock);
16157 return (ENOENT);
16158 }
16159
16160 if (state->dts_activity == DTRACE_ACTIVITY_DRAINING)
16161 stat.dtst_exiting = 1;
16162
16163 nerrs = state->dts_errors;
16164 dstate = &state->dts_vstate.dtvs_dynvars;
16165
16166 for (i = 0; i < NCPU; i++) {
16167 dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i];
16168
16169 stat.dtst_dyndrops += dcpu->dtdsc_drops;
16170 stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops;
16171 stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops;
16172
16173 if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL)
16174 stat.dtst_filled++;
16175
16176 nerrs += state->dts_buffer[i].dtb_errors;
16177
16178 for (j = 0; j < state->dts_nspeculations; j++) {
16179 dtrace_speculation_t *spec;
16180 dtrace_buffer_t *buf;
16181
16182 spec = &state->dts_speculations[j];
16183 buf = &spec->dtsp_buffer[i];
16184 stat.dtst_specdrops += buf->dtb_xamot_drops;
16185 }
16186 }
16187
16188 stat.dtst_specdrops_busy = state->dts_speculations_busy;
16189 stat.dtst_specdrops_unavail = state->dts_speculations_unavail;
16190 stat.dtst_stkstroverflows = state->dts_stkstroverflows;
16191 stat.dtst_dblerrors = state->dts_dblerrors;
16192 stat.dtst_killed =
16193 (state->dts_activity == DTRACE_ACTIVITY_KILLED);
16194 stat.dtst_errors = nerrs;
16195
16196 mutex_exit(&dtrace_lock);
16197
16198 if (copyout(&stat, (void *)arg, sizeof (stat)) != 0)
16199 return (EFAULT);
16200
16201 return (0);
16202 }
16203
16204 case DTRACEIOC_FORMAT: {
16205 dtrace_fmtdesc_t fmt;
16206 char *str;
16207 int len;
16208
16209 if (copyin((void *)arg, &fmt, sizeof (fmt)) != 0)
16210 return (EFAULT);
16211
16212 mutex_enter(&dtrace_lock);
16213
16214 if (fmt.dtfd_format == 0 ||
16215 fmt.dtfd_format > state->dts_nformats) {
16216 mutex_exit(&dtrace_lock);
16217 return (EINVAL);
16218 }
16219
16220 /*
16221 * Format strings are allocated contiguously and they are
16222 * never freed; if a format index is less than the number
16223 * of formats, we can assert that the format map is non-NULL
16224 * and that the format for the specified index is non-NULL.
16225 */
16226 ASSERT(state->dts_formats != NULL);
16227 str = state->dts_formats[fmt.dtfd_format - 1];
16228 ASSERT(str != NULL);
16229
16230 len = strlen(str) + 1;
16231
16232 if (len > fmt.dtfd_length) {
16233 fmt.dtfd_length = len;
16234
16235 if (copyout(&fmt, (void *)arg, sizeof (fmt)) != 0) {
16236 mutex_exit(&dtrace_lock);
16237 return (EINVAL);
16238 }
16239 } else {
16240 if (copyout(str, fmt.dtfd_string, len) != 0) {
16241 mutex_exit(&dtrace_lock);
16242 return (EINVAL);
16243 }
16244 }
16245
16246 mutex_exit(&dtrace_lock);
16247 return (0);
16248 }
16249
16250 default:
16251 break;
16252 }
16253
16254 return (ENOTTY);
16255}
16256
16257/*ARGSUSED*/
16258static int
16259dtrace_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
16260{
16261 dtrace_state_t *state;
16262
16263 switch (cmd) {
16264 case DDI_DETACH:
16265 break;
16266
16267 case DDI_SUSPEND:
16268 return (DDI_SUCCESS);
16269
16270 default:
16271 return (DDI_FAILURE);
16272 }
16273
16274 mutex_enter(&cpu_lock);
16275 mutex_enter(&dtrace_provider_lock);
16276 mutex_enter(&dtrace_lock);
16277
16278 ASSERT(dtrace_opens == 0);
16279
16280 if (dtrace_helpers > 0) {
16281 mutex_exit(&dtrace_provider_lock);
16282 mutex_exit(&dtrace_lock);
16283 mutex_exit(&cpu_lock);
16284 return (DDI_FAILURE);
16285 }
16286
16287 if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) {
16288 mutex_exit(&dtrace_provider_lock);
16289 mutex_exit(&dtrace_lock);
16290 mutex_exit(&cpu_lock);
16291 return (DDI_FAILURE);
16292 }
16293
16294 dtrace_provider = NULL;
16295
16296 if ((state = dtrace_anon_grab()) != NULL) {
16297 /*
16298 * If there were ECBs on this state, the provider should
16299 * have not been allowed to detach; assert that there is
16300 * none.
16301 */
16302 ASSERT(state->dts_necbs == 0);
16303 dtrace_state_destroy(state);
16304
16305 /*
16306 * If we're being detached with anonymous state, we need to
16307 * indicate to the kernel debugger that DTrace is now inactive.
16308 */
16309 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE);
16310 }
16311
16312 bzero(&dtrace_anon, sizeof (dtrace_anon_t));
16313 unregister_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL);
16314 dtrace_cpu_init = NULL;
16315 dtrace_helpers_cleanup = NULL;
16316 dtrace_helpers_fork = NULL;
16317 dtrace_cpustart_init = NULL;
16318 dtrace_cpustart_fini = NULL;
16319 dtrace_debugger_init = NULL;
16320 dtrace_debugger_fini = NULL;
16321 dtrace_modload = NULL;
16322 dtrace_modunload = NULL;
16323
16324 mutex_exit(&cpu_lock);
16325
16326 if (dtrace_helptrace_enabled) {
16327 kmem_free(dtrace_helptrace_buffer, dtrace_helptrace_bufsize);
16328 dtrace_helptrace_buffer = NULL;
16329 }
16330
16331 kmem_free(dtrace_probes, dtrace_nprobes * sizeof (dtrace_probe_t *));
16332 dtrace_probes = NULL;
16333 dtrace_nprobes = 0;
16334
16335 dtrace_hash_destroy(dtrace_bymod);
16336 dtrace_hash_destroy(dtrace_byfunc);
16337 dtrace_hash_destroy(dtrace_byname);
16338 dtrace_bymod = NULL;
16339 dtrace_byfunc = NULL;
16340 dtrace_byname = NULL;
16341
16342 kmem_cache_destroy(dtrace_state_cache);
16343 vmem_destroy(dtrace_minor);
16344 vmem_destroy(dtrace_arena);
16345
16346 if (dtrace_toxrange != NULL) {
16347 kmem_free(dtrace_toxrange,
16348 dtrace_toxranges_max * sizeof (dtrace_toxrange_t));
16349 dtrace_toxrange = NULL;
16350 dtrace_toxranges = 0;
16351 dtrace_toxranges_max = 0;
16352 }
16353
16354 ddi_remove_minor_node(dtrace_devi, NULL);
16355 dtrace_devi = NULL;
16356
16357 ddi_soft_state_fini(&dtrace_softstate);
16358
16359 ASSERT(dtrace_vtime_references == 0);
16360 ASSERT(dtrace_opens == 0);
16361 ASSERT(dtrace_retained == NULL);
16362
16363 mutex_exit(&dtrace_lock);
16364 mutex_exit(&dtrace_provider_lock);
16365
16366 /*
16367 * We don't destroy the task queue until after we have dropped our
16368 * locks (taskq_destroy() may block on running tasks). To prevent
16369 * attempting to do work after we have effectively detached but before
16370 * the task queue has been destroyed, all tasks dispatched via the
16371 * task queue must check that DTrace is still attached before
16372 * performing any operation.
16373 */
16374 taskq_destroy(dtrace_taskq);
16375 dtrace_taskq = NULL;
16376
16377 return (DDI_SUCCESS);
16378}
16379#endif
16380
16381#if defined(sun)
16382/*ARGSUSED*/
16383static int
16384dtrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
16385{
16386 int error;
16387
16388 switch (infocmd) {
16389 case DDI_INFO_DEVT2DEVINFO:
16390 *result = (void *)dtrace_devi;
16391 error = DDI_SUCCESS;
16392 break;
16393 case DDI_INFO_DEVT2INSTANCE:
16394 *result = (void *)0;
16395 error = DDI_SUCCESS;
16396 break;
16397 default:
16398 error = DDI_FAILURE;
16399 }
16400 return (error);
16401}
16402#endif
16403
16404#if defined(sun)
16405static struct cb_ops dtrace_cb_ops = {
16406 dtrace_open, /* open */
16407 dtrace_close, /* close */
16408 nulldev, /* strategy */
16409 nulldev, /* print */
16410 nodev, /* dump */
16411 nodev, /* read */
16412 nodev, /* write */
16413 dtrace_ioctl, /* ioctl */
16414 nodev, /* devmap */
16415 nodev, /* mmap */
16416 nodev, /* segmap */
16417 nochpoll, /* poll */
16418 ddi_prop_op, /* cb_prop_op */
16419 0, /* streamtab */
16420 D_NEW | D_MP /* Driver compatibility flag */
16421};
16422
16423static struct dev_ops dtrace_ops = {
16424 DEVO_REV, /* devo_rev */
16425 0, /* refcnt */
16426 dtrace_info, /* get_dev_info */
16427 nulldev, /* identify */
16428 nulldev, /* probe */
16429 dtrace_attach, /* attach */
16430 dtrace_detach, /* detach */
16431 nodev, /* reset */
16432 &dtrace_cb_ops, /* driver operations */
16433 NULL, /* bus operations */
16434 nodev /* dev power */
16435};
16436
16437static struct modldrv modldrv = {
16438 &mod_driverops, /* module type (this is a pseudo driver) */
16439 "Dynamic Tracing", /* name of module */
16440 &dtrace_ops, /* driver ops */
16441};
16442
16443static struct modlinkage modlinkage = {
16444 MODREV_1,
16445 (void *)&modldrv,
16446 NULL
16447};
16448
16449int
16450_init(void)
16451{
16452 return (mod_install(&modlinkage));
16453}
16454
16455int
16456_info(struct modinfo *modinfop)
16457{
16458 return (mod_info(&modlinkage, modinfop));
16459}
16460
16461int
16462_fini(void)
16463{
16464 return (mod_remove(&modlinkage));
16465}
16466#else
16467
16468static d_ioctl_t dtrace_ioctl;
16469static void dtrace_load(void *);
16470static int dtrace_unload(void);
16471#if __FreeBSD_version < 800039
16472static void dtrace_clone(void *, struct ucred *, char *, int , struct cdev **);
16473static struct clonedevs *dtrace_clones; /* Ptr to the array of cloned devices. */
16474static eventhandler_tag eh_tag; /* Event handler tag. */
16475#else
16476static struct cdev *dtrace_dev;
16477#endif
16478
16479void dtrace_invop_init(void);
16480void dtrace_invop_uninit(void);
16481
16482static struct cdevsw dtrace_cdevsw = {
16483 .d_version = D_VERSION,
16484 .d_flags = D_TRACKCLOSE | D_NEEDMINOR,
16485 .d_close = dtrace_close,
16486 .d_ioctl = dtrace_ioctl,
16487 .d_open = dtrace_open,
16488 .d_name = "dtrace",
16489};
16490
16491#include <dtrace_anon.c>
16492#if __FreeBSD_version < 800039
16493#include <dtrace_clone.c>
16494#endif
16495#include <dtrace_ioctl.c>
16496#include <dtrace_load.c>
16497#include <dtrace_modevent.c>
16498#include <dtrace_sysctl.c>
16499#include <dtrace_unload.c>
16500#include <dtrace_vtime.c>
16501#include <dtrace_hacks.c>
16502#include <dtrace_isa.c>
16503
16504SYSINIT(dtrace_load, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_load, NULL);
16505SYSUNINIT(dtrace_unload, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_unload, NULL);
16506SYSINIT(dtrace_anon_init, SI_SUB_DTRACE_ANON, SI_ORDER_FIRST, dtrace_anon_init, NULL);
16507
16508DEV_MODULE(dtrace, dtrace_modevent, NULL);
16509MODULE_VERSION(dtrace, 1);
16510MODULE_DEPEND(dtrace, cyclic, 1, 1, 1);
16511MODULE_DEPEND(dtrace, opensolaris, 1, 1, 1);
16512#endif
|