dtrace.c revision 184698
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 * 21 * $FreeBSD: head/sys/cddl/contrib/opensolaris/uts/common/dtrace/dtrace.c 184698 2008-11-05 19:39:11Z rodrigc $ 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/sx.h> 126#include <sys/dtrace_bsd.h> 127#include <netinet/in.h> 128#include "dtrace_cddl.h" 129#include "dtrace_debug.c" 130#endif 131 132/* 133 * DTrace Tunable Variables 134 * 135 * The following variables may be tuned by adding a line to /etc/system that 136 * includes both the name of the DTrace module ("dtrace") and the name of the 137 * variable. For example: 138 * 139 * set dtrace:dtrace_destructive_disallow = 1 140 * 141 * In general, the only variables that one should be tuning this way are those 142 * that affect system-wide DTrace behavior, and for which the default behavior 143 * is undesirable. Most of these variables are tunable on a per-consumer 144 * basis using DTrace options, and need not be tuned on a system-wide basis. 145 * When tuning these variables, avoid pathological values; while some attempt 146 * is made to verify the integrity of these variables, they are not considered 147 * part of the supported interface to DTrace, and they are therefore not 148 * checked comprehensively. Further, these variables should not be tuned 149 * dynamically via "mdb -kw" or other means; they should only be tuned via 150 * /etc/system. 151 */ 152int dtrace_destructive_disallow = 0; 153dtrace_optval_t dtrace_nonroot_maxsize = (16 * 1024 * 1024); 154size_t dtrace_difo_maxsize = (256 * 1024); 155dtrace_optval_t dtrace_dof_maxsize = (256 * 1024); 156size_t dtrace_global_maxsize = (16 * 1024); 157size_t dtrace_actions_max = (16 * 1024); 158size_t dtrace_retain_max = 1024; 159dtrace_optval_t dtrace_helper_actions_max = 32; 160dtrace_optval_t dtrace_helper_providers_max = 32; 161dtrace_optval_t dtrace_dstate_defsize = (1 * 1024 * 1024); 162size_t dtrace_strsize_default = 256; 163dtrace_optval_t dtrace_cleanrate_default = 9900990; /* 101 hz */ 164dtrace_optval_t dtrace_cleanrate_min = 200000; /* 5000 hz */ 165dtrace_optval_t dtrace_cleanrate_max = (uint64_t)60 * NANOSEC; /* 1/minute */ 166dtrace_optval_t dtrace_aggrate_default = NANOSEC; /* 1 hz */ 167dtrace_optval_t dtrace_statusrate_default = NANOSEC; /* 1 hz */ 168dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC; /* 6/minute */ 169dtrace_optval_t dtrace_switchrate_default = NANOSEC; /* 1 hz */ 170dtrace_optval_t dtrace_nspec_default = 1; 171dtrace_optval_t dtrace_specsize_default = 32 * 1024; 172dtrace_optval_t dtrace_stackframes_default = 20; 173dtrace_optval_t dtrace_ustackframes_default = 20; 174dtrace_optval_t dtrace_jstackframes_default = 50; 175dtrace_optval_t dtrace_jstackstrsize_default = 512; 176int dtrace_msgdsize_max = 128; 177hrtime_t dtrace_chill_max = 500 * (NANOSEC / MILLISEC); /* 500 ms */ 178hrtime_t dtrace_chill_interval = NANOSEC; /* 1000 ms */ 179int dtrace_devdepth_max = 32; 180int dtrace_err_verbose; 181hrtime_t dtrace_deadman_interval = NANOSEC; 182hrtime_t dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC; 183hrtime_t dtrace_deadman_user = (hrtime_t)30 * NANOSEC; 184 185/* 186 * DTrace External Variables 187 * 188 * As dtrace(7D) is a kernel module, any DTrace variables are obviously 189 * available to DTrace consumers via the backtick (`) syntax. One of these, 190 * dtrace_zero, is made deliberately so: it is provided as a source of 191 * well-known, zero-filled memory. While this variable is not documented, 192 * it is used by some translators as an implementation detail. 193 */ 194const char dtrace_zero[256] = { 0 }; /* zero-filled memory */ 195 196/* 197 * DTrace Internal Variables 198 */ 199#if defined(sun) 200static dev_info_t *dtrace_devi; /* device info */ 201#endif 202#if defined(sun) 203static vmem_t *dtrace_arena; /* probe ID arena */ 204static vmem_t *dtrace_minor; /* minor number arena */ 205static taskq_t *dtrace_taskq; /* task queue */ 206#else 207static struct unrhdr *dtrace_arena; /* Probe ID number. */ 208#endif 209static dtrace_probe_t **dtrace_probes; /* array of all probes */ 210static int dtrace_nprobes; /* number of probes */ 211static dtrace_provider_t *dtrace_provider; /* provider list */ 212static dtrace_meta_t *dtrace_meta_pid; /* user-land meta provider */ 213static int dtrace_opens; /* number of opens */ 214static int dtrace_helpers; /* number of helpers */ 215#if defined(sun) 216static void *dtrace_softstate; /* softstate pointer */ 217#endif 218static dtrace_hash_t *dtrace_bymod; /* probes hashed by module */ 219static dtrace_hash_t *dtrace_byfunc; /* probes hashed by function */ 220static dtrace_hash_t *dtrace_byname; /* probes hashed by name */ 221static dtrace_toxrange_t *dtrace_toxrange; /* toxic range array */ 222static int dtrace_toxranges; /* number of toxic ranges */ 223static int dtrace_toxranges_max; /* size of toxic range array */ 224static dtrace_anon_t dtrace_anon; /* anonymous enabling */ 225static kmem_cache_t *dtrace_state_cache; /* cache for dynamic state */ 226static uint64_t dtrace_vtime_references; /* number of vtimestamp refs */ 227static kthread_t *dtrace_panicked; /* panicking thread */ 228static dtrace_ecb_t *dtrace_ecb_create_cache; /* cached created ECB */ 229static dtrace_genid_t dtrace_probegen; /* current probe generation */ 230static dtrace_helpers_t *dtrace_deferred_pid; /* deferred helper list */ 231static dtrace_enabling_t *dtrace_retained; /* list of retained enablings */ 232static dtrace_dynvar_t dtrace_dynhash_sink; /* end of dynamic hash chains */ 233#if !defined(sun) 234static struct mtx dtrace_unr_mtx; 235MTX_SYSINIT(dtrace_unr_mtx, &dtrace_unr_mtx, "Unique resource identifier", MTX_DEF); 236int dtrace_in_probe; /* non-zero if executing a probe */ 237#if defined(__i386__) || defined(__amd64__) 238uintptr_t dtrace_in_probe_addr; /* Address of invop when already in probe */ 239#endif 240#endif 241 242/* 243 * DTrace Locking 244 * DTrace is protected by three (relatively coarse-grained) locks: 245 * 246 * (1) dtrace_lock is required to manipulate essentially any DTrace state, 247 * including enabling state, probes, ECBs, consumer state, helper state, 248 * etc. Importantly, dtrace_lock is _not_ required when in probe context; 249 * probe context is lock-free -- synchronization is handled via the 250 * dtrace_sync() cross call mechanism. 251 * 252 * (2) dtrace_provider_lock is required when manipulating provider state, or 253 * when provider state must be held constant. 254 * 255 * (3) dtrace_meta_lock is required when manipulating meta provider state, or 256 * when meta provider state must be held constant. 257 * 258 * The lock ordering between these three locks is dtrace_meta_lock before 259 * dtrace_provider_lock before dtrace_lock. (In particular, there are 260 * several places where dtrace_provider_lock is held by the framework as it 261 * calls into the providers -- which then call back into the framework, 262 * grabbing dtrace_lock.) 263 * 264 * There are two other locks in the mix: mod_lock and cpu_lock. With respect 265 * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical 266 * role as a coarse-grained lock; it is acquired before both of these locks. 267 * With respect to dtrace_meta_lock, its behavior is stranger: cpu_lock must 268 * be acquired _between_ dtrace_meta_lock and any other DTrace locks. 269 * mod_lock is similar with respect to dtrace_provider_lock in that it must be 270 * acquired _between_ dtrace_provider_lock and dtrace_lock. 271 */ 272static kmutex_t dtrace_lock; /* probe state lock */ 273static kmutex_t dtrace_provider_lock; /* provider state lock */ 274static kmutex_t dtrace_meta_lock; /* meta-provider state lock */ 275 276#if !defined(sun) 277/* XXX FreeBSD hacks. */ 278static kmutex_t mod_lock; 279 280#define cr_suid cr_svuid 281#define cr_sgid cr_svgid 282#define ipaddr_t in_addr_t 283#define mod_modname pathname 284#define vuprintf vprintf 285#define ttoproc(_a) ((_a)->td_proc) 286#define crgetzoneid(_a) 0 287#define NCPU MAXCPU 288#define SNOCD 0 289#define CPU_ON_INTR(_a) 0 290 291#define PRIV_EFFECTIVE (1 << 0) 292#define PRIV_DTRACE_KERNEL (1 << 1) 293#define PRIV_DTRACE_PROC (1 << 2) 294#define PRIV_DTRACE_USER (1 << 3) 295#define PRIV_PROC_OWNER (1 << 4) 296#define PRIV_PROC_ZONE (1 << 5) 297#define PRIV_ALL ~0 298 299SYSCTL_NODE(_debug, OID_AUTO, dtrace, CTLFLAG_RD, 0, "DTrace Information"); 300#endif 301 302#if defined(sun) 303#define curcpu CPU->cpu_id 304#endif 305 306 307/* 308 * DTrace Provider Variables 309 * 310 * These are the variables relating to DTrace as a provider (that is, the 311 * provider of the BEGIN, END, and ERROR probes). 312 */ 313static dtrace_pattr_t dtrace_provider_attr = { 314{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, 315{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 316{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 317{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, 318{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, 319}; 320 321static void 322dtrace_nullop(void) 323{} 324 325static dtrace_pops_t dtrace_provider_ops = { 326 (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop, 327 (void (*)(void *, modctl_t *))dtrace_nullop, 328 (void (*)(void *, dtrace_id_t, void *))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 NULL, 333 NULL, 334 NULL, 335 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop 336}; 337 338static dtrace_id_t dtrace_probeid_begin; /* special BEGIN probe */ 339static dtrace_id_t dtrace_probeid_end; /* special END probe */ 340dtrace_id_t dtrace_probeid_error; /* special ERROR probe */ 341 342/* 343 * DTrace Helper Tracing Variables 344 */ 345uint32_t dtrace_helptrace_next = 0; 346uint32_t dtrace_helptrace_nlocals; 347char *dtrace_helptrace_buffer; 348int dtrace_helptrace_bufsize = 512 * 1024; 349 350#ifdef DEBUG 351int dtrace_helptrace_enabled = 1; 352#else 353int dtrace_helptrace_enabled = 0; 354#endif 355 356/* 357 * DTrace Error Hashing 358 * 359 * On DEBUG kernels, DTrace will track the errors that has seen in a hash 360 * table. This is very useful for checking coverage of tests that are 361 * expected to induce DIF or DOF processing errors, and may be useful for 362 * debugging problems in the DIF code generator or in DOF generation . The 363 * error hash may be examined with the ::dtrace_errhash MDB dcmd. 364 */ 365#ifdef DEBUG 366static dtrace_errhash_t dtrace_errhash[DTRACE_ERRHASHSZ]; 367static const char *dtrace_errlast; 368static kthread_t *dtrace_errthread; 369static kmutex_t dtrace_errlock; 370#endif 371 372/* 373 * DTrace Macros and Constants 374 * 375 * These are various macros that are useful in various spots in the 376 * implementation, along with a few random constants that have no meaning 377 * outside of the implementation. There is no real structure to this cpp 378 * mishmash -- but is there ever? 379 */ 380#define DTRACE_HASHSTR(hash, probe) \ 381 dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs))) 382 383#define DTRACE_HASHNEXT(hash, probe) \ 384 (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs) 385 386#define DTRACE_HASHPREV(hash, probe) \ 387 (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs) 388 389#define DTRACE_HASHEQ(hash, lhs, rhs) \ 390 (strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \ 391 *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0) 392 393#define DTRACE_AGGHASHSIZE_SLEW 17 394 395#define DTRACE_V4MAPPED_OFFSET (sizeof (uint32_t) * 3) 396 397/* 398 * The key for a thread-local variable consists of the lower 61 bits of the 399 * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL. 400 * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never 401 * equal to a variable identifier. This is necessary (but not sufficient) to 402 * assure that global associative arrays never collide with thread-local 403 * variables. To guarantee that they cannot collide, we must also define the 404 * order for keying dynamic variables. That order is: 405 * 406 * [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ] 407 * 408 * Because the variable-key and the tls-key are in orthogonal spaces, there is 409 * no way for a global variable key signature to match a thread-local key 410 * signature. 411 */ 412#if defined(sun) 413#define DTRACE_TLS_THRKEY(where) { \ 414 uint_t intr = 0; \ 415 uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \ 416 for (; actv; actv >>= 1) \ 417 intr++; \ 418 ASSERT(intr < (1 << 3)); \ 419 (where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \ 420 (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \ 421} 422#else 423#define DTRACE_TLS_THRKEY(where) { \ 424 solaris_cpu_t *_c = &solaris_cpu[curcpu]; \ 425 uint_t intr = 0; \ 426 uint_t actv = _c->cpu_intr_actv; \ 427 for (; actv; actv >>= 1) \ 428 intr++; \ 429 ASSERT(intr < (1 << 3)); \ 430 (where) = ((curthread->td_tid + DIF_VARIABLE_MAX) & \ 431 (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \ 432} 433#endif 434 435#define DT_BSWAP_8(x) ((x) & 0xff) 436#define DT_BSWAP_16(x) ((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8)) 437#define DT_BSWAP_32(x) ((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16)) 438#define DT_BSWAP_64(x) ((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32)) 439 440#define DT_MASK_LO 0x00000000FFFFFFFFULL 441 442#define DTRACE_STORE(type, tomax, offset, what) \ 443 *((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what); 444 445#ifndef __i386 446#define DTRACE_ALIGNCHECK(addr, size, flags) \ 447 if (addr & (size - 1)) { \ 448 *flags |= CPU_DTRACE_BADALIGN; \ 449 cpu_core[curcpu].cpuc_dtrace_illval = addr; \ 450 return (0); \ 451 } 452#else 453#define DTRACE_ALIGNCHECK(addr, size, flags) 454#endif 455 456/* 457 * Test whether a range of memory starting at testaddr of size testsz falls 458 * within the range of memory described by addr, sz. We take care to avoid 459 * problems with overflow and underflow of the unsigned quantities, and 460 * disallow all negative sizes. Ranges of size 0 are allowed. 461 */ 462#define DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \ 463 ((testaddr) - (baseaddr) < (basesz) && \ 464 (testaddr) + (testsz) - (baseaddr) <= (basesz) && \ 465 (testaddr) + (testsz) >= (testaddr)) 466 467/* 468 * Test whether alloc_sz bytes will fit in the scratch region. We isolate 469 * alloc_sz on the righthand side of the comparison in order to avoid overflow 470 * or underflow in the comparison with it. This is simpler than the INRANGE 471 * check above, because we know that the dtms_scratch_ptr is valid in the 472 * range. Allocations of size zero are allowed. 473 */ 474#define DTRACE_INSCRATCH(mstate, alloc_sz) \ 475 ((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \ 476 (mstate)->dtms_scratch_ptr >= (alloc_sz)) 477 478#define DTRACE_LOADFUNC(bits) \ 479/*CSTYLED*/ \ 480uint##bits##_t \ 481dtrace_load##bits(uintptr_t addr) \ 482{ \ 483 size_t size = bits / NBBY; \ 484 /*CSTYLED*/ \ 485 uint##bits##_t rval; \ 486 int i; \ 487 volatile uint16_t *flags = (volatile uint16_t *) \ 488 &cpu_core[curcpu].cpuc_dtrace_flags; \ 489 \ 490 DTRACE_ALIGNCHECK(addr, size, flags); \ 491 \ 492 for (i = 0; i < dtrace_toxranges; i++) { \ 493 if (addr >= dtrace_toxrange[i].dtt_limit) \ 494 continue; \ 495 \ 496 if (addr + size <= dtrace_toxrange[i].dtt_base) \ 497 continue; \ 498 \ 499 /* \ 500 * This address falls within a toxic region; return 0. \ 501 */ \ 502 *flags |= CPU_DTRACE_BADADDR; \ 503 cpu_core[curcpu].cpuc_dtrace_illval = addr; \ 504 return (0); \ 505 } \ 506 \ 507 *flags |= CPU_DTRACE_NOFAULT; \ 508 /*CSTYLED*/ \ 509 rval = *((volatile uint##bits##_t *)addr); \ 510 *flags &= ~CPU_DTRACE_NOFAULT; \ 511 \ 512 return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0); \ 513} 514 515#ifdef _LP64 516#define dtrace_loadptr dtrace_load64 517#else 518#define dtrace_loadptr dtrace_load32 519#endif 520 521#define DTRACE_DYNHASH_FREE 0 522#define DTRACE_DYNHASH_SINK 1 523#define DTRACE_DYNHASH_VALID 2 524 525#define DTRACE_MATCH_NEXT 0 526#define DTRACE_MATCH_DONE 1 527#define DTRACE_ANCHORED(probe) ((probe)->dtpr_func[0] != '\0') 528#define DTRACE_STATE_ALIGN 64 529 530#define DTRACE_FLAGS2FLT(flags) \ 531 (((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR : \ 532 ((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP : \ 533 ((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO : \ 534 ((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV : \ 535 ((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV : \ 536 ((flags) & CPU_DTRACE_TUPOFLOW) ? DTRACEFLT_TUPOFLOW : \ 537 ((flags) & CPU_DTRACE_BADALIGN) ? DTRACEFLT_BADALIGN : \ 538 ((flags) & CPU_DTRACE_NOSCRATCH) ? DTRACEFLT_NOSCRATCH : \ 539 ((flags) & CPU_DTRACE_BADSTACK) ? DTRACEFLT_BADSTACK : \ 540 DTRACEFLT_UNKNOWN) 541 542#define DTRACEACT_ISSTRING(act) \ 543 ((act)->dta_kind == DTRACEACT_DIFEXPR && \ 544 (act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) 545 546/* Function prototype definitions: */ 547static size_t dtrace_strlen(const char *, size_t); 548static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id); 549static void dtrace_enabling_provide(dtrace_provider_t *); 550static int dtrace_enabling_match(dtrace_enabling_t *, int *); 551static void dtrace_enabling_matchall(void); 552static dtrace_state_t *dtrace_anon_grab(void); 553#if defined(sun) 554static uint64_t dtrace_helper(int, dtrace_mstate_t *, 555 dtrace_state_t *, uint64_t, uint64_t); 556static dtrace_helpers_t *dtrace_helpers_create(proc_t *); 557#endif 558static void dtrace_buffer_drop(dtrace_buffer_t *); 559static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t, 560 dtrace_state_t *, dtrace_mstate_t *); 561static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t, 562 dtrace_optval_t); 563static int dtrace_ecb_create_enable(dtrace_probe_t *, void *); 564#if defined(sun) 565static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *); 566#endif 567uint16_t dtrace_load16(uintptr_t); 568uint32_t dtrace_load32(uintptr_t); 569uint64_t dtrace_load64(uintptr_t); 570uint8_t dtrace_load8(uintptr_t); 571void dtrace_dynvar_clean(dtrace_dstate_t *); 572dtrace_dynvar_t *dtrace_dynvar(dtrace_dstate_t *, uint_t, dtrace_key_t *, 573 size_t, dtrace_dynvar_op_t, dtrace_mstate_t *, dtrace_vstate_t *); 574uintptr_t dtrace_dif_varstr(uintptr_t, dtrace_state_t *, dtrace_mstate_t *); 575 576/* 577 * DTrace Probe Context Functions 578 * 579 * These functions are called from probe context. Because probe context is 580 * any context in which C may be called, arbitrarily locks may be held, 581 * interrupts may be disabled, we may be in arbitrary dispatched state, etc. 582 * As a result, functions called from probe context may only call other DTrace 583 * support functions -- they may not interact at all with the system at large. 584 * (Note that the ASSERT macro is made probe-context safe by redefining it in 585 * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary 586 * loads are to be performed from probe context, they _must_ be in terms of 587 * the safe dtrace_load*() variants. 588 * 589 * Some functions in this block are not actually called from probe context; 590 * for these functions, there will be a comment above the function reading 591 * "Note: not called from probe context." 592 */ 593void 594dtrace_panic(const char *format, ...) 595{ 596 va_list alist; 597 598 va_start(alist, format); 599 dtrace_vpanic(format, alist); 600 va_end(alist); 601} 602 603int 604dtrace_assfail(const char *a, const char *f, int l) 605{ 606 dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l); 607 608 /* 609 * We just need something here that even the most clever compiler 610 * cannot optimize away. 611 */ 612 return (a[(uintptr_t)f]); 613} 614 615/* 616 * Atomically increment a specified error counter from probe context. 617 */ 618static void 619dtrace_error(uint32_t *counter) 620{ 621 /* 622 * Most counters stored to in probe context are per-CPU counters. 623 * However, there are some error conditions that are sufficiently 624 * arcane that they don't merit per-CPU storage. If these counters 625 * are incremented concurrently on different CPUs, scalability will be 626 * adversely affected -- but we don't expect them to be white-hot in a 627 * correctly constructed enabling... 628 */ 629 uint32_t oval, nval; 630 631 do { 632 oval = *counter; 633 634 if ((nval = oval + 1) == 0) { 635 /* 636 * If the counter would wrap, set it to 1 -- assuring 637 * that the counter is never zero when we have seen 638 * errors. (The counter must be 32-bits because we 639 * aren't guaranteed a 64-bit compare&swap operation.) 640 * To save this code both the infamy of being fingered 641 * by a priggish news story and the indignity of being 642 * the target of a neo-puritan witch trial, we're 643 * carefully avoiding any colorful description of the 644 * likelihood of this condition -- but suffice it to 645 * say that it is only slightly more likely than the 646 * overflow of predicate cache IDs, as discussed in 647 * dtrace_predicate_create(). 648 */ 649 nval = 1; 650 } 651 } while (dtrace_cas32(counter, oval, nval) != oval); 652} 653 654/* 655 * Use the DTRACE_LOADFUNC macro to define functions for each of loading a 656 * uint8_t, a uint16_t, a uint32_t and a uint64_t. 657 */ 658DTRACE_LOADFUNC(8) 659DTRACE_LOADFUNC(16) 660DTRACE_LOADFUNC(32) 661DTRACE_LOADFUNC(64) 662 663static int 664dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate) 665{ 666 if (dest < mstate->dtms_scratch_base) 667 return (0); 668 669 if (dest + size < dest) 670 return (0); 671 672 if (dest + size > mstate->dtms_scratch_ptr) 673 return (0); 674 675 return (1); 676} 677 678static int 679dtrace_canstore_statvar(uint64_t addr, size_t sz, 680 dtrace_statvar_t **svars, int nsvars) 681{ 682 int i; 683 684 for (i = 0; i < nsvars; i++) { 685 dtrace_statvar_t *svar = svars[i]; 686 687 if (svar == NULL || svar->dtsv_size == 0) 688 continue; 689 690 if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size)) 691 return (1); 692 } 693 694 return (0); 695} 696 697/* 698 * Check to see if the address is within a memory region to which a store may 699 * be issued. This includes the DTrace scratch areas, and any DTrace variable 700 * region. The caller of dtrace_canstore() is responsible for performing any 701 * alignment checks that are needed before stores are actually executed. 702 */ 703static int 704dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, 705 dtrace_vstate_t *vstate) 706{ 707 /* 708 * First, check to see if the address is in scratch space... 709 */ 710 if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base, 711 mstate->dtms_scratch_size)) 712 return (1); 713 714 /* 715 * Now check to see if it's a dynamic variable. This check will pick 716 * up both thread-local variables and any global dynamically-allocated 717 * variables. 718 */ 719 if (DTRACE_INRANGE(addr, sz, (uintptr_t)vstate->dtvs_dynvars.dtds_base, 720 vstate->dtvs_dynvars.dtds_size)) { 721 dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; 722 uintptr_t base = (uintptr_t)dstate->dtds_base + 723 (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t)); 724 uintptr_t chunkoffs; 725 726 /* 727 * Before we assume that we can store here, we need to make 728 * sure that it isn't in our metadata -- storing to our 729 * dynamic variable metadata would corrupt our state. For 730 * the range to not include any dynamic variable metadata, 731 * it must: 732 * 733 * (1) Start above the hash table that is at the base of 734 * the dynamic variable space 735 * 736 * (2) Have a starting chunk offset that is beyond the 737 * dtrace_dynvar_t that is at the base of every chunk 738 * 739 * (3) Not span a chunk boundary 740 * 741 */ 742 if (addr < base) 743 return (0); 744 745 chunkoffs = (addr - base) % dstate->dtds_chunksize; 746 747 if (chunkoffs < sizeof (dtrace_dynvar_t)) 748 return (0); 749 750 if (chunkoffs + sz > dstate->dtds_chunksize) 751 return (0); 752 753 return (1); 754 } 755 756 /* 757 * Finally, check the static local and global variables. These checks 758 * take the longest, so we perform them last. 759 */ 760 if (dtrace_canstore_statvar(addr, sz, 761 vstate->dtvs_locals, vstate->dtvs_nlocals)) 762 return (1); 763 764 if (dtrace_canstore_statvar(addr, sz, 765 vstate->dtvs_globals, vstate->dtvs_nglobals)) 766 return (1); 767 768 return (0); 769} 770 771 772/* 773 * Convenience routine to check to see if the address is within a memory 774 * region in which a load may be issued given the user's privilege level; 775 * if not, it sets the appropriate error flags and loads 'addr' into the 776 * illegal value slot. 777 * 778 * DTrace subroutines (DIF_SUBR_*) should use this helper to implement 779 * appropriate memory access protection. 780 */ 781static int 782dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, 783 dtrace_vstate_t *vstate) 784{ 785 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; 786 787 /* 788 * If we hold the privilege to read from kernel memory, then 789 * everything is readable. 790 */ 791 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) 792 return (1); 793 794 /* 795 * You can obviously read that which you can store. 796 */ 797 if (dtrace_canstore(addr, sz, mstate, vstate)) 798 return (1); 799 800 /* 801 * We're allowed to read from our own string table. 802 */ 803 if (DTRACE_INRANGE(addr, sz, (uintptr_t)mstate->dtms_difo->dtdo_strtab, 804 mstate->dtms_difo->dtdo_strlen)) 805 return (1); 806 807 DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV); 808 *illval = addr; 809 return (0); 810} 811 812/* 813 * Convenience routine to check to see if a given string is within a memory 814 * region in which a load may be issued given the user's privilege level; 815 * this exists so that we don't need to issue unnecessary dtrace_strlen() 816 * calls in the event that the user has all privileges. 817 */ 818static int 819dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, 820 dtrace_vstate_t *vstate) 821{ 822 size_t strsz; 823 824 /* 825 * If we hold the privilege to read from kernel memory, then 826 * everything is readable. 827 */ 828 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) 829 return (1); 830 831 strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz); 832 if (dtrace_canload(addr, strsz, mstate, vstate)) 833 return (1); 834 835 return (0); 836} 837 838/* 839 * Convenience routine to check to see if a given variable is within a memory 840 * region in which a load may be issued given the user's privilege level. 841 */ 842static int 843dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate, 844 dtrace_vstate_t *vstate) 845{ 846 size_t sz; 847 ASSERT(type->dtdt_flags & DIF_TF_BYREF); 848 849 /* 850 * If we hold the privilege to read from kernel memory, then 851 * everything is readable. 852 */ 853 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) 854 return (1); 855 856 if (type->dtdt_kind == DIF_TYPE_STRING) 857 sz = dtrace_strlen(src, 858 vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1; 859 else 860 sz = type->dtdt_size; 861 862 return (dtrace_canload((uintptr_t)src, sz, mstate, vstate)); 863} 864 865/* 866 * Compare two strings using safe loads. 867 */ 868static int 869dtrace_strncmp(char *s1, char *s2, size_t limit) 870{ 871 uint8_t c1, c2; 872 volatile uint16_t *flags; 873 874 if (s1 == s2 || limit == 0) 875 return (0); 876 877 flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; 878 879 do { 880 if (s1 == NULL) { 881 c1 = '\0'; 882 } else { 883 c1 = dtrace_load8((uintptr_t)s1++); 884 } 885 886 if (s2 == NULL) { 887 c2 = '\0'; 888 } else { 889 c2 = dtrace_load8((uintptr_t)s2++); 890 } 891 892 if (c1 != c2) 893 return (c1 - c2); 894 } while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT)); 895 896 return (0); 897} 898 899/* 900 * Compute strlen(s) for a string using safe memory accesses. The additional 901 * len parameter is used to specify a maximum length to ensure completion. 902 */ 903static size_t 904dtrace_strlen(const char *s, size_t lim) 905{ 906 uint_t len; 907 908 for (len = 0; len != lim; len++) { 909 if (dtrace_load8((uintptr_t)s++) == '\0') 910 break; 911 } 912 913 return (len); 914} 915 916/* 917 * Check if an address falls within a toxic region. 918 */ 919static int 920dtrace_istoxic(uintptr_t kaddr, size_t size) 921{ 922 uintptr_t taddr, tsize; 923 int i; 924 925 for (i = 0; i < dtrace_toxranges; i++) { 926 taddr = dtrace_toxrange[i].dtt_base; 927 tsize = dtrace_toxrange[i].dtt_limit - taddr; 928 929 if (kaddr - taddr < tsize) { 930 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); 931 cpu_core[curcpu].cpuc_dtrace_illval = kaddr; 932 return (1); 933 } 934 935 if (taddr - kaddr < size) { 936 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); 937 cpu_core[curcpu].cpuc_dtrace_illval = taddr; 938 return (1); 939 } 940 } 941 942 return (0); 943} 944 945/* 946 * Copy src to dst using safe memory accesses. The src is assumed to be unsafe 947 * memory specified by the DIF program. The dst is assumed to be safe memory 948 * that we can store to directly because it is managed by DTrace. As with 949 * standard bcopy, overlapping copies are handled properly. 950 */ 951static void 952dtrace_bcopy(const void *src, void *dst, size_t len) 953{ 954 if (len != 0) { 955 uint8_t *s1 = dst; 956 const uint8_t *s2 = src; 957 958 if (s1 <= s2) { 959 do { 960 *s1++ = dtrace_load8((uintptr_t)s2++); 961 } while (--len != 0); 962 } else { 963 s2 += len; 964 s1 += len; 965 966 do { 967 *--s1 = dtrace_load8((uintptr_t)--s2); 968 } while (--len != 0); 969 } 970 } 971} 972 973/* 974 * Copy src to dst using safe memory accesses, up to either the specified 975 * length, or the point that a nul byte is encountered. The src is assumed to 976 * be unsafe memory specified by the DIF program. The dst is assumed to be 977 * safe memory that we can store to directly because it is managed by DTrace. 978 * Unlike dtrace_bcopy(), overlapping regions are not handled. 979 */ 980static void 981dtrace_strcpy(const void *src, void *dst, size_t len) 982{ 983 if (len != 0) { 984 uint8_t *s1 = dst, c; 985 const uint8_t *s2 = src; 986 987 do { 988 *s1++ = c = dtrace_load8((uintptr_t)s2++); 989 } while (--len != 0 && c != '\0'); 990 } 991} 992 993/* 994 * Copy src to dst, deriving the size and type from the specified (BYREF) 995 * variable type. The src is assumed to be unsafe memory specified by the DIF 996 * program. The dst is assumed to be DTrace variable memory that is of the 997 * specified type; we assume that we can store to directly. 998 */ 999static void 1000dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type) 1001{ 1002 ASSERT(type->dtdt_flags & DIF_TF_BYREF); 1003 1004 if (type->dtdt_kind == DIF_TYPE_STRING) { 1005 dtrace_strcpy(src, dst, type->dtdt_size); 1006 } else { 1007 dtrace_bcopy(src, dst, type->dtdt_size); 1008 } 1009} 1010 1011/* 1012 * Compare s1 to s2 using safe memory accesses. The s1 data is assumed to be 1013 * unsafe memory specified by the DIF program. The s2 data is assumed to be 1014 * safe memory that we can access directly because it is managed by DTrace. 1015 */ 1016static int 1017dtrace_bcmp(const void *s1, const void *s2, size_t len) 1018{ 1019 volatile uint16_t *flags; 1020 1021 flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; 1022 1023 if (s1 == s2) 1024 return (0); 1025 1026 if (s1 == NULL || s2 == NULL) 1027 return (1); 1028 1029 if (s1 != s2 && len != 0) { 1030 const uint8_t *ps1 = s1; 1031 const uint8_t *ps2 = s2; 1032 1033 do { 1034 if (dtrace_load8((uintptr_t)ps1++) != *ps2++) 1035 return (1); 1036 } while (--len != 0 && !(*flags & CPU_DTRACE_FAULT)); 1037 } 1038 return (0); 1039} 1040 1041/* 1042 * Zero the specified region using a simple byte-by-byte loop. Note that this 1043 * is for safe DTrace-managed memory only. 1044 */ 1045static void 1046dtrace_bzero(void *dst, size_t len) 1047{ 1048 uchar_t *cp; 1049 1050 for (cp = dst; len != 0; len--) 1051 *cp++ = 0; 1052} 1053 1054static void 1055dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum) 1056{ 1057 uint64_t result[2]; 1058 1059 result[0] = addend1[0] + addend2[0]; 1060 result[1] = addend1[1] + addend2[1] + 1061 (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0); 1062 1063 sum[0] = result[0]; 1064 sum[1] = result[1]; 1065} 1066 1067/* 1068 * Shift the 128-bit value in a by b. If b is positive, shift left. 1069 * If b is negative, shift right. 1070 */ 1071static void 1072dtrace_shift_128(uint64_t *a, int b) 1073{ 1074 uint64_t mask; 1075 1076 if (b == 0) 1077 return; 1078 1079 if (b < 0) { 1080 b = -b; 1081 if (b >= 64) { 1082 a[0] = a[1] >> (b - 64); 1083 a[1] = 0; 1084 } else { 1085 a[0] >>= b; 1086 mask = 1LL << (64 - b); 1087 mask -= 1; 1088 a[0] |= ((a[1] & mask) << (64 - b)); 1089 a[1] >>= b; 1090 } 1091 } else { 1092 if (b >= 64) { 1093 a[1] = a[0] << (b - 64); 1094 a[0] = 0; 1095 } else { 1096 a[1] <<= b; 1097 mask = a[0] >> (64 - b); 1098 a[1] |= mask; 1099 a[0] <<= b; 1100 } 1101 } 1102} 1103 1104/* 1105 * The basic idea is to break the 2 64-bit values into 4 32-bit values, 1106 * use native multiplication on those, and then re-combine into the 1107 * resulting 128-bit value. 1108 * 1109 * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) = 1110 * hi1 * hi2 << 64 + 1111 * hi1 * lo2 << 32 + 1112 * hi2 * lo1 << 32 + 1113 * lo1 * lo2 1114 */ 1115static void 1116dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product) 1117{ 1118 uint64_t hi1, hi2, lo1, lo2; 1119 uint64_t tmp[2]; 1120 1121 hi1 = factor1 >> 32; 1122 hi2 = factor2 >> 32; 1123 1124 lo1 = factor1 & DT_MASK_LO; 1125 lo2 = factor2 & DT_MASK_LO; 1126 1127 product[0] = lo1 * lo2; 1128 product[1] = hi1 * hi2; 1129 1130 tmp[0] = hi1 * lo2; 1131 tmp[1] = 0; 1132 dtrace_shift_128(tmp, 32); 1133 dtrace_add_128(product, tmp, product); 1134 1135 tmp[0] = hi2 * lo1; 1136 tmp[1] = 0; 1137 dtrace_shift_128(tmp, 32); 1138 dtrace_add_128(product, tmp, product); 1139} 1140 1141/* 1142 * This privilege check should be used by actions and subroutines to 1143 * verify that the user credentials of the process that enabled the 1144 * invoking ECB match the target credentials 1145 */ 1146static int 1147dtrace_priv_proc_common_user(dtrace_state_t *state) 1148{ 1149 cred_t *cr, *s_cr = state->dts_cred.dcr_cred; 1150 1151 /* 1152 * We should always have a non-NULL state cred here, since if cred 1153 * is null (anonymous tracing), we fast-path bypass this routine. 1154 */ 1155 ASSERT(s_cr != NULL); 1156 1157 if ((cr = CRED()) != NULL && 1158 s_cr->cr_uid == cr->cr_uid && 1159 s_cr->cr_uid == cr->cr_ruid && 1160 s_cr->cr_uid == cr->cr_suid && 1161 s_cr->cr_gid == cr->cr_gid && 1162 s_cr->cr_gid == cr->cr_rgid && 1163 s_cr->cr_gid == cr->cr_sgid) 1164 return (1); 1165 1166 return (0); 1167} 1168 1169/* 1170 * This privilege check should be used by actions and subroutines to 1171 * verify that the zone of the process that enabled the invoking ECB 1172 * matches the target credentials 1173 */ 1174static int 1175dtrace_priv_proc_common_zone(dtrace_state_t *state) 1176{ 1177#if defined(sun) 1178 cred_t *cr, *s_cr = state->dts_cred.dcr_cred; 1179 1180 /* 1181 * We should always have a non-NULL state cred here, since if cred 1182 * is null (anonymous tracing), we fast-path bypass this routine. 1183 */ 1184 ASSERT(s_cr != NULL); 1185 1186 if ((cr = CRED()) != NULL && 1187 s_cr->cr_zone == cr->cr_zone) 1188 return (1); 1189 1190 return (0); 1191#else 1192 return (1); 1193#endif 1194} 1195 1196/* 1197 * This privilege check should be used by actions and subroutines to 1198 * verify that the process has not setuid or changed credentials. 1199 */ 1200static int 1201dtrace_priv_proc_common_nocd(void) 1202{ 1203 proc_t *proc; 1204 1205 if ((proc = ttoproc(curthread)) != NULL && 1206 !(proc->p_flag & SNOCD)) 1207 return (1); 1208 1209 return (0); 1210} 1211 1212static int 1213dtrace_priv_proc_destructive(dtrace_state_t *state) 1214{ 1215 int action = state->dts_cred.dcr_action; 1216 1217 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) && 1218 dtrace_priv_proc_common_zone(state) == 0) 1219 goto bad; 1220 1221 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) && 1222 dtrace_priv_proc_common_user(state) == 0) 1223 goto bad; 1224 1225 if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) && 1226 dtrace_priv_proc_common_nocd() == 0) 1227 goto bad; 1228 1229 return (1); 1230 1231bad: 1232 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; 1233 1234 return (0); 1235} 1236 1237static int 1238dtrace_priv_proc_control(dtrace_state_t *state) 1239{ 1240 if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL) 1241 return (1); 1242 1243 if (dtrace_priv_proc_common_zone(state) && 1244 dtrace_priv_proc_common_user(state) && 1245 dtrace_priv_proc_common_nocd()) 1246 return (1); 1247 1248 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; 1249 1250 return (0); 1251} 1252 1253static int 1254dtrace_priv_proc(dtrace_state_t *state) 1255{ 1256 if (state->dts_cred.dcr_action & DTRACE_CRA_PROC) 1257 return (1); 1258 1259 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; 1260 1261 return (0); 1262} 1263 1264static int 1265dtrace_priv_kernel(dtrace_state_t *state) 1266{ 1267 if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL) 1268 return (1); 1269 1270 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV; 1271 1272 return (0); 1273} 1274 1275static int 1276dtrace_priv_kernel_destructive(dtrace_state_t *state) 1277{ 1278 if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE) 1279 return (1); 1280 1281 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV; 1282 1283 return (0); 1284} 1285 1286/* 1287 * Note: not called from probe context. This function is called 1288 * asynchronously (and at a regular interval) from outside of probe context to 1289 * clean the dirty dynamic variable lists on all CPUs. Dynamic variable 1290 * cleaning is explained in detail in <sys/dtrace_impl.h>. 1291 */ 1292void 1293dtrace_dynvar_clean(dtrace_dstate_t *dstate) 1294{ 1295 dtrace_dynvar_t *dirty; 1296 dtrace_dstate_percpu_t *dcpu; 1297 int i, work = 0; 1298 1299 for (i = 0; i < NCPU; i++) { 1300 dcpu = &dstate->dtds_percpu[i]; 1301 1302 ASSERT(dcpu->dtdsc_rinsing == NULL); 1303 1304 /* 1305 * If the dirty list is NULL, there is no dirty work to do. 1306 */ 1307 if (dcpu->dtdsc_dirty == NULL) 1308 continue; 1309 1310 /* 1311 * If the clean list is non-NULL, then we're not going to do 1312 * any work for this CPU -- it means that there has not been 1313 * a dtrace_dynvar() allocation on this CPU (or from this CPU) 1314 * since the last time we cleaned house. 1315 */ 1316 if (dcpu->dtdsc_clean != NULL) 1317 continue; 1318 1319 work = 1; 1320 1321 /* 1322 * Atomically move the dirty list aside. 1323 */ 1324 do { 1325 dirty = dcpu->dtdsc_dirty; 1326 1327 /* 1328 * Before we zap the dirty list, set the rinsing list. 1329 * (This allows for a potential assertion in 1330 * dtrace_dynvar(): if a free dynamic variable appears 1331 * on a hash chain, either the dirty list or the 1332 * rinsing list for some CPU must be non-NULL.) 1333 */ 1334 dcpu->dtdsc_rinsing = dirty; 1335 dtrace_membar_producer(); 1336 } while (dtrace_casptr(&dcpu->dtdsc_dirty, 1337 dirty, NULL) != dirty); 1338 } 1339 1340 if (!work) { 1341 /* 1342 * We have no work to do; we can simply return. 1343 */ 1344 return; 1345 } 1346 1347 dtrace_sync(); 1348 1349 for (i = 0; i < NCPU; i++) { 1350 dcpu = &dstate->dtds_percpu[i]; 1351 1352 if (dcpu->dtdsc_rinsing == NULL) 1353 continue; 1354 1355 /* 1356 * We are now guaranteed that no hash chain contains a pointer 1357 * into this dirty list; we can make it clean. 1358 */ 1359 ASSERT(dcpu->dtdsc_clean == NULL); 1360 dcpu->dtdsc_clean = dcpu->dtdsc_rinsing; 1361 dcpu->dtdsc_rinsing = NULL; 1362 } 1363 1364 /* 1365 * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make 1366 * sure that all CPUs have seen all of the dtdsc_clean pointers. 1367 * This prevents a race whereby a CPU incorrectly decides that 1368 * the state should be something other than DTRACE_DSTATE_CLEAN 1369 * after dtrace_dynvar_clean() has completed. 1370 */ 1371 dtrace_sync(); 1372 1373 dstate->dtds_state = DTRACE_DSTATE_CLEAN; 1374} 1375 1376/* 1377 * Depending on the value of the op parameter, this function looks-up, 1378 * allocates or deallocates an arbitrarily-keyed dynamic variable. If an 1379 * allocation is requested, this function will return a pointer to a 1380 * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no 1381 * variable can be allocated. If NULL is returned, the appropriate counter 1382 * will be incremented. 1383 */ 1384dtrace_dynvar_t * 1385dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys, 1386 dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op, 1387 dtrace_mstate_t *mstate, dtrace_vstate_t *vstate) 1388{ 1389 uint64_t hashval = DTRACE_DYNHASH_VALID; 1390 dtrace_dynhash_t *hash = dstate->dtds_hash; 1391 dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL; 1392 processorid_t me = curcpu, cpu = me; 1393 dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me]; 1394 size_t bucket, ksize; 1395 size_t chunksize = dstate->dtds_chunksize; 1396 uintptr_t kdata, lock, nstate; 1397 uint_t i; 1398 1399 ASSERT(nkeys != 0); 1400 1401 /* 1402 * Hash the key. As with aggregations, we use Jenkins' "One-at-a-time" 1403 * algorithm. For the by-value portions, we perform the algorithm in 1404 * 16-bit chunks (as opposed to 8-bit chunks). This speeds things up a 1405 * bit, and seems to have only a minute effect on distribution. For 1406 * the by-reference data, we perform "One-at-a-time" iterating (safely) 1407 * over each referenced byte. It's painful to do this, but it's much 1408 * better than pathological hash distribution. The efficacy of the 1409 * hashing algorithm (and a comparison with other algorithms) may be 1410 * found by running the ::dtrace_dynstat MDB dcmd. 1411 */ 1412 for (i = 0; i < nkeys; i++) { 1413 if (key[i].dttk_size == 0) { 1414 uint64_t val = key[i].dttk_value; 1415 1416 hashval += (val >> 48) & 0xffff; 1417 hashval += (hashval << 10); 1418 hashval ^= (hashval >> 6); 1419 1420 hashval += (val >> 32) & 0xffff; 1421 hashval += (hashval << 10); 1422 hashval ^= (hashval >> 6); 1423 1424 hashval += (val >> 16) & 0xffff; 1425 hashval += (hashval << 10); 1426 hashval ^= (hashval >> 6); 1427 1428 hashval += val & 0xffff; 1429 hashval += (hashval << 10); 1430 hashval ^= (hashval >> 6); 1431 } else { 1432 /* 1433 * This is incredibly painful, but it beats the hell 1434 * out of the alternative. 1435 */ 1436 uint64_t j, size = key[i].dttk_size; 1437 uintptr_t base = (uintptr_t)key[i].dttk_value; 1438 1439 if (!dtrace_canload(base, size, mstate, vstate)) 1440 break; 1441 1442 for (j = 0; j < size; j++) { 1443 hashval += dtrace_load8(base + j); 1444 hashval += (hashval << 10); 1445 hashval ^= (hashval >> 6); 1446 } 1447 } 1448 } 1449 1450 if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) 1451 return (NULL); 1452 1453 hashval += (hashval << 3); 1454 hashval ^= (hashval >> 11); 1455 hashval += (hashval << 15); 1456 1457 /* 1458 * There is a remote chance (ideally, 1 in 2^31) that our hashval 1459 * comes out to be one of our two sentinel hash values. If this 1460 * actually happens, we set the hashval to be a value known to be a 1461 * non-sentinel value. 1462 */ 1463 if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK) 1464 hashval = DTRACE_DYNHASH_VALID; 1465 1466 /* 1467 * Yes, it's painful to do a divide here. If the cycle count becomes 1468 * important here, tricks can be pulled to reduce it. (However, it's 1469 * critical that hash collisions be kept to an absolute minimum; 1470 * they're much more painful than a divide.) It's better to have a 1471 * solution that generates few collisions and still keeps things 1472 * relatively simple. 1473 */ 1474 bucket = hashval % dstate->dtds_hashsize; 1475 1476 if (op == DTRACE_DYNVAR_DEALLOC) { 1477 volatile uintptr_t *lockp = &hash[bucket].dtdh_lock; 1478 1479 for (;;) { 1480 while ((lock = *lockp) & 1) 1481 continue; 1482 1483 if (dtrace_casptr((volatile void *)lockp, 1484 (volatile void *)lock, (volatile void *)(lock + 1)) == (void *)lock) 1485 break; 1486 } 1487 1488 dtrace_membar_producer(); 1489 } 1490 1491top: 1492 prev = NULL; 1493 lock = hash[bucket].dtdh_lock; 1494 1495 dtrace_membar_consumer(); 1496 1497 start = hash[bucket].dtdh_chain; 1498 ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK || 1499 start->dtdv_hashval != DTRACE_DYNHASH_FREE || 1500 op != DTRACE_DYNVAR_DEALLOC)); 1501 1502 for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) { 1503 dtrace_tuple_t *dtuple = &dvar->dtdv_tuple; 1504 dtrace_key_t *dkey = &dtuple->dtt_key[0]; 1505 1506 if (dvar->dtdv_hashval != hashval) { 1507 if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) { 1508 /* 1509 * We've reached the sink, and therefore the 1510 * end of the hash chain; we can kick out of 1511 * the loop knowing that we have seen a valid 1512 * snapshot of state. 1513 */ 1514 ASSERT(dvar->dtdv_next == NULL); 1515 ASSERT(dvar == &dtrace_dynhash_sink); 1516 break; 1517 } 1518 1519 if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) { 1520 /* 1521 * We've gone off the rails: somewhere along 1522 * the line, one of the members of this hash 1523 * chain was deleted. Note that we could also 1524 * detect this by simply letting this loop run 1525 * to completion, as we would eventually hit 1526 * the end of the dirty list. However, we 1527 * want to avoid running the length of the 1528 * dirty list unnecessarily (it might be quite 1529 * long), so we catch this as early as 1530 * possible by detecting the hash marker. In 1531 * this case, we simply set dvar to NULL and 1532 * break; the conditional after the loop will 1533 * send us back to top. 1534 */ 1535 dvar = NULL; 1536 break; 1537 } 1538 1539 goto next; 1540 } 1541 1542 if (dtuple->dtt_nkeys != nkeys) 1543 goto next; 1544 1545 for (i = 0; i < nkeys; i++, dkey++) { 1546 if (dkey->dttk_size != key[i].dttk_size) 1547 goto next; /* size or type mismatch */ 1548 1549 if (dkey->dttk_size != 0) { 1550 if (dtrace_bcmp( 1551 (void *)(uintptr_t)key[i].dttk_value, 1552 (void *)(uintptr_t)dkey->dttk_value, 1553 dkey->dttk_size)) 1554 goto next; 1555 } else { 1556 if (dkey->dttk_value != key[i].dttk_value) 1557 goto next; 1558 } 1559 } 1560 1561 if (op != DTRACE_DYNVAR_DEALLOC) 1562 return (dvar); 1563 1564 ASSERT(dvar->dtdv_next == NULL || 1565 dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE); 1566 1567 if (prev != NULL) { 1568 ASSERT(hash[bucket].dtdh_chain != dvar); 1569 ASSERT(start != dvar); 1570 ASSERT(prev->dtdv_next == dvar); 1571 prev->dtdv_next = dvar->dtdv_next; 1572 } else { 1573 if (dtrace_casptr(&hash[bucket].dtdh_chain, 1574 start, dvar->dtdv_next) != start) { 1575 /* 1576 * We have failed to atomically swing the 1577 * hash table head pointer, presumably because 1578 * of a conflicting allocation on another CPU. 1579 * We need to reread the hash chain and try 1580 * again. 1581 */ 1582 goto top; 1583 } 1584 } 1585 1586 dtrace_membar_producer(); 1587 1588 /* 1589 * Now set the hash value to indicate that it's free. 1590 */ 1591 ASSERT(hash[bucket].dtdh_chain != dvar); 1592 dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; 1593 1594 dtrace_membar_producer(); 1595 1596 /* 1597 * Set the next pointer to point at the dirty list, and 1598 * atomically swing the dirty pointer to the newly freed dvar. 1599 */ 1600 do { 1601 next = dcpu->dtdsc_dirty; 1602 dvar->dtdv_next = next; 1603 } while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next); 1604 1605 /* 1606 * Finally, unlock this hash bucket. 1607 */ 1608 ASSERT(hash[bucket].dtdh_lock == lock); 1609 ASSERT(lock & 1); 1610 hash[bucket].dtdh_lock++; 1611 1612 return (NULL); 1613next: 1614 prev = dvar; 1615 continue; 1616 } 1617 1618 if (dvar == NULL) { 1619 /* 1620 * If dvar is NULL, it is because we went off the rails: 1621 * one of the elements that we traversed in the hash chain 1622 * was deleted while we were traversing it. In this case, 1623 * we assert that we aren't doing a dealloc (deallocs lock 1624 * the hash bucket to prevent themselves from racing with 1625 * one another), and retry the hash chain traversal. 1626 */ 1627 ASSERT(op != DTRACE_DYNVAR_DEALLOC); 1628 goto top; 1629 } 1630 1631 if (op != DTRACE_DYNVAR_ALLOC) { 1632 /* 1633 * If we are not to allocate a new variable, we want to 1634 * return NULL now. Before we return, check that the value 1635 * of the lock word hasn't changed. If it has, we may have 1636 * seen an inconsistent snapshot. 1637 */ 1638 if (op == DTRACE_DYNVAR_NOALLOC) { 1639 if (hash[bucket].dtdh_lock != lock) 1640 goto top; 1641 } else { 1642 ASSERT(op == DTRACE_DYNVAR_DEALLOC); 1643 ASSERT(hash[bucket].dtdh_lock == lock); 1644 ASSERT(lock & 1); 1645 hash[bucket].dtdh_lock++; 1646 } 1647 1648 return (NULL); 1649 } 1650 1651 /* 1652 * We need to allocate a new dynamic variable. The size we need is the 1653 * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the 1654 * size of any auxiliary key data (rounded up to 8-byte alignment) plus 1655 * the size of any referred-to data (dsize). We then round the final 1656 * size up to the chunksize for allocation. 1657 */ 1658 for (ksize = 0, i = 0; i < nkeys; i++) 1659 ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t)); 1660 1661 /* 1662 * This should be pretty much impossible, but could happen if, say, 1663 * strange DIF specified the tuple. Ideally, this should be an 1664 * assertion and not an error condition -- but that requires that the 1665 * chunksize calculation in dtrace_difo_chunksize() be absolutely 1666 * bullet-proof. (That is, it must not be able to be fooled by 1667 * malicious DIF.) Given the lack of backwards branches in DIF, 1668 * solving this would presumably not amount to solving the Halting 1669 * Problem -- but it still seems awfully hard. 1670 */ 1671 if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) + 1672 ksize + dsize > chunksize) { 1673 dcpu->dtdsc_drops++; 1674 return (NULL); 1675 } 1676 1677 nstate = DTRACE_DSTATE_EMPTY; 1678 1679 do { 1680retry: 1681 free = dcpu->dtdsc_free; 1682 1683 if (free == NULL) { 1684 dtrace_dynvar_t *clean = dcpu->dtdsc_clean; 1685 void *rval; 1686 1687 if (clean == NULL) { 1688 /* 1689 * We're out of dynamic variable space on 1690 * this CPU. Unless we have tried all CPUs, 1691 * we'll try to allocate from a different 1692 * CPU. 1693 */ 1694 switch (dstate->dtds_state) { 1695 case DTRACE_DSTATE_CLEAN: { 1696 void *sp = &dstate->dtds_state; 1697 1698 if (++cpu >= NCPU) 1699 cpu = 0; 1700 1701 if (dcpu->dtdsc_dirty != NULL && 1702 nstate == DTRACE_DSTATE_EMPTY) 1703 nstate = DTRACE_DSTATE_DIRTY; 1704 1705 if (dcpu->dtdsc_rinsing != NULL) 1706 nstate = DTRACE_DSTATE_RINSING; 1707 1708 dcpu = &dstate->dtds_percpu[cpu]; 1709 1710 if (cpu != me) 1711 goto retry; 1712 1713 (void) dtrace_cas32(sp, 1714 DTRACE_DSTATE_CLEAN, nstate); 1715 1716 /* 1717 * To increment the correct bean 1718 * counter, take another lap. 1719 */ 1720 goto retry; 1721 } 1722 1723 case DTRACE_DSTATE_DIRTY: 1724 dcpu->dtdsc_dirty_drops++; 1725 break; 1726 1727 case DTRACE_DSTATE_RINSING: 1728 dcpu->dtdsc_rinsing_drops++; 1729 break; 1730 1731 case DTRACE_DSTATE_EMPTY: 1732 dcpu->dtdsc_drops++; 1733 break; 1734 } 1735 1736 DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP); 1737 return (NULL); 1738 } 1739 1740 /* 1741 * The clean list appears to be non-empty. We want to 1742 * move the clean list to the free list; we start by 1743 * moving the clean pointer aside. 1744 */ 1745 if (dtrace_casptr(&dcpu->dtdsc_clean, 1746 clean, NULL) != clean) { 1747 /* 1748 * We are in one of two situations: 1749 * 1750 * (a) The clean list was switched to the 1751 * free list by another CPU. 1752 * 1753 * (b) The clean list was added to by the 1754 * cleansing cyclic. 1755 * 1756 * In either of these situations, we can 1757 * just reattempt the free list allocation. 1758 */ 1759 goto retry; 1760 } 1761 1762 ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE); 1763 1764 /* 1765 * Now we'll move the clean list to the free list. 1766 * It's impossible for this to fail: the only way 1767 * the free list can be updated is through this 1768 * code path, and only one CPU can own the clean list. 1769 * Thus, it would only be possible for this to fail if 1770 * this code were racing with dtrace_dynvar_clean(). 1771 * (That is, if dtrace_dynvar_clean() updated the clean 1772 * list, and we ended up racing to update the free 1773 * list.) This race is prevented by the dtrace_sync() 1774 * in dtrace_dynvar_clean() -- which flushes the 1775 * owners of the clean lists out before resetting 1776 * the clean lists. 1777 */ 1778 rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean); 1779 ASSERT(rval == NULL); 1780 goto retry; 1781 } 1782 1783 dvar = free; 1784 new_free = dvar->dtdv_next; 1785 } while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free); 1786 1787 /* 1788 * We have now allocated a new chunk. We copy the tuple keys into the 1789 * tuple array and copy any referenced key data into the data space 1790 * following the tuple array. As we do this, we relocate dttk_value 1791 * in the final tuple to point to the key data address in the chunk. 1792 */ 1793 kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys]; 1794 dvar->dtdv_data = (void *)(kdata + ksize); 1795 dvar->dtdv_tuple.dtt_nkeys = nkeys; 1796 1797 for (i = 0; i < nkeys; i++) { 1798 dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i]; 1799 size_t kesize = key[i].dttk_size; 1800 1801 if (kesize != 0) { 1802 dtrace_bcopy( 1803 (const void *)(uintptr_t)key[i].dttk_value, 1804 (void *)kdata, kesize); 1805 dkey->dttk_value = kdata; 1806 kdata += P2ROUNDUP(kesize, sizeof (uint64_t)); 1807 } else { 1808 dkey->dttk_value = key[i].dttk_value; 1809 } 1810 1811 dkey->dttk_size = kesize; 1812 } 1813 1814 ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE); 1815 dvar->dtdv_hashval = hashval; 1816 dvar->dtdv_next = start; 1817 1818 if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start) 1819 return (dvar); 1820 1821 /* 1822 * The cas has failed. Either another CPU is adding an element to 1823 * this hash chain, or another CPU is deleting an element from this 1824 * hash chain. The simplest way to deal with both of these cases 1825 * (though not necessarily the most efficient) is to free our 1826 * allocated block and tail-call ourselves. Note that the free is 1827 * to the dirty list and _not_ to the free list. This is to prevent 1828 * races with allocators, above. 1829 */ 1830 dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; 1831 1832 dtrace_membar_producer(); 1833 1834 do { 1835 free = dcpu->dtdsc_dirty; 1836 dvar->dtdv_next = free; 1837 } while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free); 1838 1839 return (dtrace_dynvar(dstate, nkeys, key, dsize, op, mstate, vstate)); 1840} 1841 1842/*ARGSUSED*/ 1843static void 1844dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg) 1845{ 1846 if ((int64_t)nval < (int64_t)*oval) 1847 *oval = nval; 1848} 1849 1850/*ARGSUSED*/ 1851static void 1852dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg) 1853{ 1854 if ((int64_t)nval > (int64_t)*oval) 1855 *oval = nval; 1856} 1857 1858static void 1859dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr) 1860{ 1861 int i, zero = DTRACE_QUANTIZE_ZEROBUCKET; 1862 int64_t val = (int64_t)nval; 1863 1864 if (val < 0) { 1865 for (i = 0; i < zero; i++) { 1866 if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) { 1867 quanta[i] += incr; 1868 return; 1869 } 1870 } 1871 } else { 1872 for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) { 1873 if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) { 1874 quanta[i - 1] += incr; 1875 return; 1876 } 1877 } 1878 1879 quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr; 1880 return; 1881 } 1882 1883 ASSERT(0); 1884} 1885 1886static void 1887dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr) 1888{ 1889 uint64_t arg = *lquanta++; 1890 int32_t base = DTRACE_LQUANTIZE_BASE(arg); 1891 uint16_t step = DTRACE_LQUANTIZE_STEP(arg); 1892 uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg); 1893 int32_t val = (int32_t)nval, level; 1894 1895 ASSERT(step != 0); 1896 ASSERT(levels != 0); 1897 1898 if (val < base) { 1899 /* 1900 * This is an underflow. 1901 */ 1902 lquanta[0] += incr; 1903 return; 1904 } 1905 1906 level = (val - base) / step; 1907 1908 if (level < levels) { 1909 lquanta[level + 1] += incr; 1910 return; 1911 } 1912 1913 /* 1914 * This is an overflow. 1915 */ 1916 lquanta[levels + 1] += incr; 1917} 1918 1919/*ARGSUSED*/ 1920static void 1921dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg) 1922{ 1923 data[0]++; 1924 data[1] += nval; 1925} 1926 1927/*ARGSUSED*/ 1928static void 1929dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg) 1930{ 1931 int64_t snval = (int64_t)nval; 1932 uint64_t tmp[2]; 1933 1934 data[0]++; 1935 data[1] += nval; 1936 1937 /* 1938 * What we want to say here is: 1939 * 1940 * data[2] += nval * nval; 1941 * 1942 * But given that nval is 64-bit, we could easily overflow, so 1943 * we do this as 128-bit arithmetic. 1944 */ 1945 if (snval < 0) 1946 snval = -snval; 1947 1948 dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp); 1949 dtrace_add_128(data + 2, tmp, data + 2); 1950} 1951 1952/*ARGSUSED*/ 1953static void 1954dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg) 1955{ 1956 *oval = *oval + 1; 1957} 1958 1959/*ARGSUSED*/ 1960static void 1961dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg) 1962{ 1963 *oval += nval; 1964} 1965 1966/* 1967 * Aggregate given the tuple in the principal data buffer, and the aggregating 1968 * action denoted by the specified dtrace_aggregation_t. The aggregation 1969 * buffer is specified as the buf parameter. This routine does not return 1970 * failure; if there is no space in the aggregation buffer, the data will be 1971 * dropped, and a corresponding counter incremented. 1972 */ 1973static void 1974dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf, 1975 intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg) 1976{ 1977 dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec; 1978 uint32_t i, ndx, size, fsize; 1979 uint32_t align = sizeof (uint64_t) - 1; 1980 dtrace_aggbuffer_t *agb; 1981 dtrace_aggkey_t *key; 1982 uint32_t hashval = 0, limit, isstr; 1983 caddr_t tomax, data, kdata; 1984 dtrace_actkind_t action; 1985 dtrace_action_t *act; 1986 uintptr_t offs; 1987 1988 if (buf == NULL) 1989 return; 1990 1991 if (!agg->dtag_hasarg) { 1992 /* 1993 * Currently, only quantize() and lquantize() take additional 1994 * arguments, and they have the same semantics: an increment 1995 * value that defaults to 1 when not present. If additional 1996 * aggregating actions take arguments, the setting of the 1997 * default argument value will presumably have to become more 1998 * sophisticated... 1999 */ 2000 arg = 1; 2001 } 2002 2003 action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION; 2004 size = rec->dtrd_offset - agg->dtag_base; 2005 fsize = size + rec->dtrd_size; 2006 2007 ASSERT(dbuf->dtb_tomax != NULL); 2008 data = dbuf->dtb_tomax + offset + agg->dtag_base; 2009 2010 if ((tomax = buf->dtb_tomax) == NULL) { 2011 dtrace_buffer_drop(buf); 2012 return; 2013 } 2014 2015 /* 2016 * The metastructure is always at the bottom of the buffer. 2017 */ 2018 agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size - 2019 sizeof (dtrace_aggbuffer_t)); 2020 2021 if (buf->dtb_offset == 0) { 2022 /* 2023 * We just kludge up approximately 1/8th of the size to be 2024 * buckets. If this guess ends up being routinely 2025 * off-the-mark, we may need to dynamically readjust this 2026 * based on past performance. 2027 */ 2028 uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t); 2029 2030 if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) < 2031 (uintptr_t)tomax || hashsize == 0) { 2032 /* 2033 * We've been given a ludicrously small buffer; 2034 * increment our drop count and leave. 2035 */ 2036 dtrace_buffer_drop(buf); 2037 return; 2038 } 2039 2040 /* 2041 * And now, a pathetic attempt to try to get a an odd (or 2042 * perchance, a prime) hash size for better hash distribution. 2043 */ 2044 if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3)) 2045 hashsize -= DTRACE_AGGHASHSIZE_SLEW; 2046 2047 agb->dtagb_hashsize = hashsize; 2048 agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb - 2049 agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *)); 2050 agb->dtagb_free = (uintptr_t)agb->dtagb_hash; 2051 2052 for (i = 0; i < agb->dtagb_hashsize; i++) 2053 agb->dtagb_hash[i] = NULL; 2054 } 2055 2056 ASSERT(agg->dtag_first != NULL); 2057 ASSERT(agg->dtag_first->dta_intuple); 2058 2059 /* 2060 * Calculate the hash value based on the key. Note that we _don't_ 2061 * include the aggid in the hashing (but we will store it as part of 2062 * the key). The hashing algorithm is Bob Jenkins' "One-at-a-time" 2063 * algorithm: a simple, quick algorithm that has no known funnels, and 2064 * gets good distribution in practice. The efficacy of the hashing 2065 * algorithm (and a comparison with other algorithms) may be found by 2066 * running the ::dtrace_aggstat MDB dcmd. 2067 */ 2068 for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { 2069 i = act->dta_rec.dtrd_offset - agg->dtag_base; 2070 limit = i + act->dta_rec.dtrd_size; 2071 ASSERT(limit <= size); 2072 isstr = DTRACEACT_ISSTRING(act); 2073 2074 for (; i < limit; i++) { 2075 hashval += data[i]; 2076 hashval += (hashval << 10); 2077 hashval ^= (hashval >> 6); 2078 2079 if (isstr && data[i] == '\0') 2080 break; 2081 } 2082 } 2083 2084 hashval += (hashval << 3); 2085 hashval ^= (hashval >> 11); 2086 hashval += (hashval << 15); 2087 2088 /* 2089 * Yes, the divide here is expensive -- but it's generally the least 2090 * of the performance issues given the amount of data that we iterate 2091 * over to compute hash values, compare data, etc. 2092 */ 2093 ndx = hashval % agb->dtagb_hashsize; 2094 2095 for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) { 2096 ASSERT((caddr_t)key >= tomax); 2097 ASSERT((caddr_t)key < tomax + buf->dtb_size); 2098 2099 if (hashval != key->dtak_hashval || key->dtak_size != size) 2100 continue; 2101 2102 kdata = key->dtak_data; 2103 ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size); 2104 2105 for (act = agg->dtag_first; act->dta_intuple; 2106 act = act->dta_next) { 2107 i = act->dta_rec.dtrd_offset - agg->dtag_base; 2108 limit = i + act->dta_rec.dtrd_size; 2109 ASSERT(limit <= size); 2110 isstr = DTRACEACT_ISSTRING(act); 2111 2112 for (; i < limit; i++) { 2113 if (kdata[i] != data[i]) 2114 goto next; 2115 2116 if (isstr && data[i] == '\0') 2117 break; 2118 } 2119 } 2120 2121 if (action != key->dtak_action) { 2122 /* 2123 * We are aggregating on the same value in the same 2124 * aggregation with two different aggregating actions. 2125 * (This should have been picked up in the compiler, 2126 * so we may be dealing with errant or devious DIF.) 2127 * This is an error condition; we indicate as much, 2128 * and return. 2129 */ 2130 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); 2131 return; 2132 } 2133 2134 /* 2135 * This is a hit: we need to apply the aggregator to 2136 * the value at this key. 2137 */ 2138 agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg); 2139 return; 2140next: 2141 continue; 2142 } 2143 2144 /* 2145 * We didn't find it. We need to allocate some zero-filled space, 2146 * link it into the hash table appropriately, and apply the aggregator 2147 * to the (zero-filled) value. 2148 */ 2149 offs = buf->dtb_offset; 2150 while (offs & (align - 1)) 2151 offs += sizeof (uint32_t); 2152 2153 /* 2154 * If we don't have enough room to both allocate a new key _and_ 2155 * its associated data, increment the drop count and return. 2156 */ 2157 if ((uintptr_t)tomax + offs + fsize > 2158 agb->dtagb_free - sizeof (dtrace_aggkey_t)) { 2159 dtrace_buffer_drop(buf); 2160 return; 2161 } 2162 2163 /*CONSTCOND*/ 2164 ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1))); 2165 key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t)); 2166 agb->dtagb_free -= sizeof (dtrace_aggkey_t); 2167 2168 key->dtak_data = kdata = tomax + offs; 2169 buf->dtb_offset = offs + fsize; 2170 2171 /* 2172 * Now copy the data across. 2173 */ 2174 *((dtrace_aggid_t *)kdata) = agg->dtag_id; 2175 2176 for (i = sizeof (dtrace_aggid_t); i < size; i++) 2177 kdata[i] = data[i]; 2178 2179 /* 2180 * Because strings are not zeroed out by default, we need to iterate 2181 * looking for actions that store strings, and we need to explicitly 2182 * pad these strings out with zeroes. 2183 */ 2184 for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { 2185 int nul; 2186 2187 if (!DTRACEACT_ISSTRING(act)) 2188 continue; 2189 2190 i = act->dta_rec.dtrd_offset - agg->dtag_base; 2191 limit = i + act->dta_rec.dtrd_size; 2192 ASSERT(limit <= size); 2193 2194 for (nul = 0; i < limit; i++) { 2195 if (nul) { 2196 kdata[i] = '\0'; 2197 continue; 2198 } 2199 2200 if (data[i] != '\0') 2201 continue; 2202 2203 nul = 1; 2204 } 2205 } 2206 2207 for (i = size; i < fsize; i++) 2208 kdata[i] = 0; 2209 2210 key->dtak_hashval = hashval; 2211 key->dtak_size = size; 2212 key->dtak_action = action; 2213 key->dtak_next = agb->dtagb_hash[ndx]; 2214 agb->dtagb_hash[ndx] = key; 2215 2216 /* 2217 * Finally, apply the aggregator. 2218 */ 2219 *((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial; 2220 agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg); 2221} 2222 2223/* 2224 * Given consumer state, this routine finds a speculation in the INACTIVE 2225 * state and transitions it into the ACTIVE state. If there is no speculation 2226 * in the INACTIVE state, 0 is returned. In this case, no error counter is 2227 * incremented -- it is up to the caller to take appropriate action. 2228 */ 2229static int 2230dtrace_speculation(dtrace_state_t *state) 2231{ 2232 int i = 0; 2233 dtrace_speculation_state_t current; 2234 uint32_t *stat = &state->dts_speculations_unavail, count; 2235 2236 while (i < state->dts_nspeculations) { 2237 dtrace_speculation_t *spec = &state->dts_speculations[i]; 2238 2239 current = spec->dtsp_state; 2240 2241 if (current != DTRACESPEC_INACTIVE) { 2242 if (current == DTRACESPEC_COMMITTINGMANY || 2243 current == DTRACESPEC_COMMITTING || 2244 current == DTRACESPEC_DISCARDING) 2245 stat = &state->dts_speculations_busy; 2246 i++; 2247 continue; 2248 } 2249 2250 if (dtrace_cas32((uint32_t *)&spec->dtsp_state, 2251 current, DTRACESPEC_ACTIVE) == current) 2252 return (i + 1); 2253 } 2254 2255 /* 2256 * We couldn't find a speculation. If we found as much as a single 2257 * busy speculation buffer, we'll attribute this failure as "busy" 2258 * instead of "unavail". 2259 */ 2260 do { 2261 count = *stat; 2262 } while (dtrace_cas32(stat, count, count + 1) != count); 2263 2264 return (0); 2265} 2266 2267/* 2268 * This routine commits an active speculation. If the specified speculation 2269 * is not in a valid state to perform a commit(), this routine will silently do 2270 * nothing. The state of the specified speculation is transitioned according 2271 * to the state transition diagram outlined in <sys/dtrace_impl.h> 2272 */ 2273static void 2274dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu, 2275 dtrace_specid_t which) 2276{ 2277 dtrace_speculation_t *spec; 2278 dtrace_buffer_t *src, *dest; 2279 uintptr_t daddr, saddr, dlimit; 2280 dtrace_speculation_state_t current, new = 0; 2281 intptr_t offs; 2282 2283 if (which == 0) 2284 return; 2285 2286 if (which > state->dts_nspeculations) { 2287 cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; 2288 return; 2289 } 2290 2291 spec = &state->dts_speculations[which - 1]; 2292 src = &spec->dtsp_buffer[cpu]; 2293 dest = &state->dts_buffer[cpu]; 2294 2295 do { 2296 current = spec->dtsp_state; 2297 2298 if (current == DTRACESPEC_COMMITTINGMANY) 2299 break; 2300 2301 switch (current) { 2302 case DTRACESPEC_INACTIVE: 2303 case DTRACESPEC_DISCARDING: 2304 return; 2305 2306 case DTRACESPEC_COMMITTING: 2307 /* 2308 * This is only possible if we are (a) commit()'ing 2309 * without having done a prior speculate() on this CPU 2310 * and (b) racing with another commit() on a different 2311 * CPU. There's nothing to do -- we just assert that 2312 * our offset is 0. 2313 */ 2314 ASSERT(src->dtb_offset == 0); 2315 return; 2316 2317 case DTRACESPEC_ACTIVE: 2318 new = DTRACESPEC_COMMITTING; 2319 break; 2320 2321 case DTRACESPEC_ACTIVEONE: 2322 /* 2323 * This speculation is active on one CPU. If our 2324 * buffer offset is non-zero, we know that the one CPU 2325 * must be us. Otherwise, we are committing on a 2326 * different CPU from the speculate(), and we must 2327 * rely on being asynchronously cleaned. 2328 */ 2329 if (src->dtb_offset != 0) { 2330 new = DTRACESPEC_COMMITTING; 2331 break; 2332 } 2333 /*FALLTHROUGH*/ 2334 2335 case DTRACESPEC_ACTIVEMANY: 2336 new = DTRACESPEC_COMMITTINGMANY; 2337 break; 2338 2339 default: 2340 ASSERT(0); 2341 } 2342 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, 2343 current, new) != current); 2344 2345 /* 2346 * We have set the state to indicate that we are committing this 2347 * speculation. Now reserve the necessary space in the destination 2348 * buffer. 2349 */ 2350 if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset, 2351 sizeof (uint64_t), state, NULL)) < 0) { 2352 dtrace_buffer_drop(dest); 2353 goto out; 2354 } 2355 2356 /* 2357 * We have the space; copy the buffer across. (Note that this is a 2358 * highly subobtimal bcopy(); in the unlikely event that this becomes 2359 * a serious performance issue, a high-performance DTrace-specific 2360 * bcopy() should obviously be invented.) 2361 */ 2362 daddr = (uintptr_t)dest->dtb_tomax + offs; 2363 dlimit = daddr + src->dtb_offset; 2364 saddr = (uintptr_t)src->dtb_tomax; 2365 2366 /* 2367 * First, the aligned portion. 2368 */ 2369 while (dlimit - daddr >= sizeof (uint64_t)) { 2370 *((uint64_t *)daddr) = *((uint64_t *)saddr); 2371 2372 daddr += sizeof (uint64_t); 2373 saddr += sizeof (uint64_t); 2374 } 2375 2376 /* 2377 * Now any left-over bit... 2378 */ 2379 while (dlimit - daddr) 2380 *((uint8_t *)daddr++) = *((uint8_t *)saddr++); 2381 2382 /* 2383 * Finally, commit the reserved space in the destination buffer. 2384 */ 2385 dest->dtb_offset = offs + src->dtb_offset; 2386 2387out: 2388 /* 2389 * If we're lucky enough to be the only active CPU on this speculation 2390 * buffer, we can just set the state back to DTRACESPEC_INACTIVE. 2391 */ 2392 if (current == DTRACESPEC_ACTIVE || 2393 (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) { 2394 uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state, 2395 DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE); 2396 2397 ASSERT(rval == DTRACESPEC_COMMITTING); 2398 } 2399 2400 src->dtb_offset = 0; 2401 src->dtb_xamot_drops += src->dtb_drops; 2402 src->dtb_drops = 0; 2403} 2404 2405/* 2406 * This routine discards an active speculation. If the specified speculation 2407 * is not in a valid state to perform a discard(), this routine will silently 2408 * do nothing. The state of the specified speculation is transitioned 2409 * according to the state transition diagram outlined in <sys/dtrace_impl.h> 2410 */ 2411static void 2412dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu, 2413 dtrace_specid_t which) 2414{ 2415 dtrace_speculation_t *spec; 2416 dtrace_speculation_state_t current, new = 0; 2417 dtrace_buffer_t *buf; 2418 2419 if (which == 0) 2420 return; 2421 2422 if (which > state->dts_nspeculations) { 2423 cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; 2424 return; 2425 } 2426 2427 spec = &state->dts_speculations[which - 1]; 2428 buf = &spec->dtsp_buffer[cpu]; 2429 2430 do { 2431 current = spec->dtsp_state; 2432 2433 switch (current) { 2434 case DTRACESPEC_INACTIVE: 2435 case DTRACESPEC_COMMITTINGMANY: 2436 case DTRACESPEC_COMMITTING: 2437 case DTRACESPEC_DISCARDING: 2438 return; 2439 2440 case DTRACESPEC_ACTIVE: 2441 case DTRACESPEC_ACTIVEMANY: 2442 new = DTRACESPEC_DISCARDING; 2443 break; 2444 2445 case DTRACESPEC_ACTIVEONE: 2446 if (buf->dtb_offset != 0) { 2447 new = DTRACESPEC_INACTIVE; 2448 } else { 2449 new = DTRACESPEC_DISCARDING; 2450 } 2451 break; 2452 2453 default: 2454 ASSERT(0); 2455 } 2456 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, 2457 current, new) != current); 2458 2459 buf->dtb_offset = 0; 2460 buf->dtb_drops = 0; 2461} 2462 2463/* 2464 * Note: not called from probe context. This function is called 2465 * asynchronously from cross call context to clean any speculations that are 2466 * in the COMMITTINGMANY or DISCARDING states. These speculations may not be 2467 * transitioned back to the INACTIVE state until all CPUs have cleaned the 2468 * speculation. 2469 */ 2470static void 2471dtrace_speculation_clean_here(dtrace_state_t *state) 2472{ 2473 dtrace_icookie_t cookie; 2474 processorid_t cpu = curcpu; 2475 dtrace_buffer_t *dest = &state->dts_buffer[cpu]; 2476 dtrace_specid_t i; 2477 2478 cookie = dtrace_interrupt_disable(); 2479 2480 if (dest->dtb_tomax == NULL) { 2481 dtrace_interrupt_enable(cookie); 2482 return; 2483 } 2484 2485 for (i = 0; i < state->dts_nspeculations; i++) { 2486 dtrace_speculation_t *spec = &state->dts_speculations[i]; 2487 dtrace_buffer_t *src = &spec->dtsp_buffer[cpu]; 2488 2489 if (src->dtb_tomax == NULL) 2490 continue; 2491 2492 if (spec->dtsp_state == DTRACESPEC_DISCARDING) { 2493 src->dtb_offset = 0; 2494 continue; 2495 } 2496 2497 if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) 2498 continue; 2499 2500 if (src->dtb_offset == 0) 2501 continue; 2502 2503 dtrace_speculation_commit(state, cpu, i + 1); 2504 } 2505 2506 dtrace_interrupt_enable(cookie); 2507} 2508 2509/* 2510 * Note: not called from probe context. This function is called 2511 * asynchronously (and at a regular interval) to clean any speculations that 2512 * are in the COMMITTINGMANY or DISCARDING states. If it discovers that there 2513 * is work to be done, it cross calls all CPUs to perform that work; 2514 * COMMITMANY and DISCARDING speculations may not be transitioned back to the 2515 * INACTIVE state until they have been cleaned by all CPUs. 2516 */ 2517static void 2518dtrace_speculation_clean(dtrace_state_t *state) 2519{ 2520 int work = 0, rv; 2521 dtrace_specid_t i; 2522 2523 for (i = 0; i < state->dts_nspeculations; i++) { 2524 dtrace_speculation_t *spec = &state->dts_speculations[i]; 2525 2526 ASSERT(!spec->dtsp_cleaning); 2527 2528 if (spec->dtsp_state != DTRACESPEC_DISCARDING && 2529 spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) 2530 continue; 2531 2532 work++; 2533 spec->dtsp_cleaning = 1; 2534 } 2535 2536 if (!work) 2537 return; 2538 2539 dtrace_xcall(DTRACE_CPUALL, 2540 (dtrace_xcall_t)dtrace_speculation_clean_here, state); 2541 2542 /* 2543 * We now know that all CPUs have committed or discarded their 2544 * speculation buffers, as appropriate. We can now set the state 2545 * to inactive. 2546 */ 2547 for (i = 0; i < state->dts_nspeculations; i++) { 2548 dtrace_speculation_t *spec = &state->dts_speculations[i]; 2549 dtrace_speculation_state_t current, new; 2550 2551 if (!spec->dtsp_cleaning) 2552 continue; 2553 2554 current = spec->dtsp_state; 2555 ASSERT(current == DTRACESPEC_DISCARDING || 2556 current == DTRACESPEC_COMMITTINGMANY); 2557 2558 new = DTRACESPEC_INACTIVE; 2559 2560 rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new); 2561 ASSERT(rv == current); 2562 spec->dtsp_cleaning = 0; 2563 } 2564} 2565 2566/* 2567 * Called as part of a speculate() to get the speculative buffer associated 2568 * with a given speculation. Returns NULL if the specified speculation is not 2569 * in an ACTIVE state. If the speculation is in the ACTIVEONE state -- and 2570 * the active CPU is not the specified CPU -- the speculation will be 2571 * atomically transitioned into the ACTIVEMANY state. 2572 */ 2573static dtrace_buffer_t * 2574dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid, 2575 dtrace_specid_t which) 2576{ 2577 dtrace_speculation_t *spec; 2578 dtrace_speculation_state_t current, new = 0; 2579 dtrace_buffer_t *buf; 2580 2581 if (which == 0) 2582 return (NULL); 2583 2584 if (which > state->dts_nspeculations) { 2585 cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; 2586 return (NULL); 2587 } 2588 2589 spec = &state->dts_speculations[which - 1]; 2590 buf = &spec->dtsp_buffer[cpuid]; 2591 2592 do { 2593 current = spec->dtsp_state; 2594 2595 switch (current) { 2596 case DTRACESPEC_INACTIVE: 2597 case DTRACESPEC_COMMITTINGMANY: 2598 case DTRACESPEC_DISCARDING: 2599 return (NULL); 2600 2601 case DTRACESPEC_COMMITTING: 2602 ASSERT(buf->dtb_offset == 0); 2603 return (NULL); 2604 2605 case DTRACESPEC_ACTIVEONE: 2606 /* 2607 * This speculation is currently active on one CPU. 2608 * Check the offset in the buffer; if it's non-zero, 2609 * that CPU must be us (and we leave the state alone). 2610 * If it's zero, assume that we're starting on a new 2611 * CPU -- and change the state to indicate that the 2612 * speculation is active on more than one CPU. 2613 */ 2614 if (buf->dtb_offset != 0) 2615 return (buf); 2616 2617 new = DTRACESPEC_ACTIVEMANY; 2618 break; 2619 2620 case DTRACESPEC_ACTIVEMANY: 2621 return (buf); 2622 2623 case DTRACESPEC_ACTIVE: 2624 new = DTRACESPEC_ACTIVEONE; 2625 break; 2626 2627 default: 2628 ASSERT(0); 2629 } 2630 } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, 2631 current, new) != current); 2632 2633 ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY); 2634 return (buf); 2635} 2636 2637/* 2638 * Return a string. In the event that the user lacks the privilege to access 2639 * arbitrary kernel memory, we copy the string out to scratch memory so that we 2640 * don't fail access checking. 2641 * 2642 * dtrace_dif_variable() uses this routine as a helper for various 2643 * builtin values such as 'execname' and 'probefunc.' 2644 */ 2645uintptr_t 2646dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state, 2647 dtrace_mstate_t *mstate) 2648{ 2649 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 2650 uintptr_t ret; 2651 size_t strsz; 2652 2653 /* 2654 * The easy case: this probe is allowed to read all of memory, so 2655 * we can just return this as a vanilla pointer. 2656 */ 2657 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) 2658 return (addr); 2659 2660 /* 2661 * This is the tougher case: we copy the string in question from 2662 * kernel memory into scratch memory and return it that way: this 2663 * ensures that we won't trip up when access checking tests the 2664 * BYREF return value. 2665 */ 2666 strsz = dtrace_strlen((char *)addr, size) + 1; 2667 2668 if (mstate->dtms_scratch_ptr + strsz > 2669 mstate->dtms_scratch_base + mstate->dtms_scratch_size) { 2670 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 2671 return (0); 2672 } 2673 2674 dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr, 2675 strsz); 2676 ret = mstate->dtms_scratch_ptr; 2677 mstate->dtms_scratch_ptr += strsz; 2678 return (ret); 2679} 2680 2681/* 2682 * Return a string from a memoy address which is known to have one or 2683 * more concatenated, individually zero terminated, sub-strings. 2684 * In the event that the user lacks the privilege to access 2685 * arbitrary kernel memory, we copy the string out to scratch memory so that we 2686 * don't fail access checking. 2687 * 2688 * dtrace_dif_variable() uses this routine as a helper for various 2689 * builtin values such as 'execargs'. 2690 */ 2691static uintptr_t 2692dtrace_dif_varstrz(uintptr_t addr, size_t strsz, dtrace_state_t *state, 2693 dtrace_mstate_t *mstate) 2694{ 2695 char *p; 2696 size_t i; 2697 uintptr_t ret; 2698 2699 if (mstate->dtms_scratch_ptr + strsz > 2700 mstate->dtms_scratch_base + mstate->dtms_scratch_size) { 2701 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 2702 return (0); 2703 } 2704 2705 dtrace_bcopy((const void *)addr, (void *)mstate->dtms_scratch_ptr, 2706 strsz); 2707 2708 /* Replace sub-string termination characters with a space. */ 2709 for (p = (char *) mstate->dtms_scratch_ptr, i = 0; i < strsz - 1; 2710 p++, i++) 2711 if (*p == '\0') 2712 *p = ' '; 2713 2714 ret = mstate->dtms_scratch_ptr; 2715 mstate->dtms_scratch_ptr += strsz; 2716 return (ret); 2717} 2718 2719/* 2720 * This function implements the DIF emulator's variable lookups. The emulator 2721 * passes a reserved variable identifier and optional built-in array index. 2722 */ 2723static uint64_t 2724dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v, 2725 uint64_t ndx) 2726{ 2727 /* 2728 * If we're accessing one of the uncached arguments, we'll turn this 2729 * into a reference in the args array. 2730 */ 2731 if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) { 2732 ndx = v - DIF_VAR_ARG0; 2733 v = DIF_VAR_ARGS; 2734 } 2735 2736 switch (v) { 2737 case DIF_VAR_ARGS: 2738 ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS); 2739 if (ndx >= sizeof (mstate->dtms_arg) / 2740 sizeof (mstate->dtms_arg[0])) { 2741 int aframes = mstate->dtms_probe->dtpr_aframes + 2; 2742 dtrace_provider_t *pv; 2743 uint64_t val; 2744 2745 pv = mstate->dtms_probe->dtpr_provider; 2746 if (pv->dtpv_pops.dtps_getargval != NULL) 2747 val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg, 2748 mstate->dtms_probe->dtpr_id, 2749 mstate->dtms_probe->dtpr_arg, ndx, aframes); 2750 else 2751 val = dtrace_getarg(ndx, aframes); 2752 2753 /* 2754 * This is regrettably required to keep the compiler 2755 * from tail-optimizing the call to dtrace_getarg(). 2756 * The condition always evaluates to true, but the 2757 * compiler has no way of figuring that out a priori. 2758 * (None of this would be necessary if the compiler 2759 * could be relied upon to _always_ tail-optimize 2760 * the call to dtrace_getarg() -- but it can't.) 2761 */ 2762 if (mstate->dtms_probe != NULL) 2763 return (val); 2764 2765 ASSERT(0); 2766 } 2767 2768 return (mstate->dtms_arg[ndx]); 2769 2770#if defined(sun) 2771 case DIF_VAR_UREGS: { 2772 klwp_t *lwp; 2773 2774 if (!dtrace_priv_proc(state)) 2775 return (0); 2776 2777 if ((lwp = curthread->t_lwp) == NULL) { 2778 DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); 2779 cpu_core[curcpu].cpuc_dtrace_illval = NULL; 2780 return (0); 2781 } 2782 2783 return (dtrace_getreg(lwp->lwp_regs, ndx)); 2784 return (0); 2785 } 2786#endif 2787 2788 case DIF_VAR_CURTHREAD: 2789 if (!dtrace_priv_kernel(state)) 2790 return (0); 2791 return ((uint64_t)(uintptr_t)curthread); 2792 2793 case DIF_VAR_TIMESTAMP: 2794 if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) { 2795 mstate->dtms_timestamp = dtrace_gethrtime(); 2796 mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP; 2797 } 2798 return (mstate->dtms_timestamp); 2799 2800 case DIF_VAR_VTIMESTAMP: 2801 ASSERT(dtrace_vtime_references != 0); 2802 return (curthread->t_dtrace_vtime); 2803 2804 case DIF_VAR_WALLTIMESTAMP: 2805 if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) { 2806 mstate->dtms_walltimestamp = dtrace_gethrestime(); 2807 mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP; 2808 } 2809 return (mstate->dtms_walltimestamp); 2810 2811#if defined(sun) 2812 case DIF_VAR_IPL: 2813 if (!dtrace_priv_kernel(state)) 2814 return (0); 2815 if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) { 2816 mstate->dtms_ipl = dtrace_getipl(); 2817 mstate->dtms_present |= DTRACE_MSTATE_IPL; 2818 } 2819 return (mstate->dtms_ipl); 2820#endif 2821 2822 case DIF_VAR_EPID: 2823 ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID); 2824 return (mstate->dtms_epid); 2825 2826 case DIF_VAR_ID: 2827 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 2828 return (mstate->dtms_probe->dtpr_id); 2829 2830 case DIF_VAR_STACKDEPTH: 2831 if (!dtrace_priv_kernel(state)) 2832 return (0); 2833 if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) { 2834 int aframes = mstate->dtms_probe->dtpr_aframes + 2; 2835 2836 mstate->dtms_stackdepth = dtrace_getstackdepth(aframes); 2837 mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH; 2838 } 2839 return (mstate->dtms_stackdepth); 2840 2841#if defined(sun) 2842 case DIF_VAR_USTACKDEPTH: 2843 if (!dtrace_priv_proc(state)) 2844 return (0); 2845 if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) { 2846 /* 2847 * See comment in DIF_VAR_PID. 2848 */ 2849 if (DTRACE_ANCHORED(mstate->dtms_probe) && 2850 CPU_ON_INTR(CPU)) { 2851 mstate->dtms_ustackdepth = 0; 2852 } else { 2853 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 2854 mstate->dtms_ustackdepth = 2855 dtrace_getustackdepth(); 2856 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 2857 } 2858 mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH; 2859 } 2860 return (mstate->dtms_ustackdepth); 2861#endif 2862 2863 case DIF_VAR_CALLER: 2864 if (!dtrace_priv_kernel(state)) 2865 return (0); 2866 if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) { 2867 int aframes = mstate->dtms_probe->dtpr_aframes + 2; 2868 2869 if (!DTRACE_ANCHORED(mstate->dtms_probe)) { 2870 /* 2871 * If this is an unanchored probe, we are 2872 * required to go through the slow path: 2873 * dtrace_caller() only guarantees correct 2874 * results for anchored probes. 2875 */ 2876 pc_t caller[2] = {0, 0}; 2877 2878 dtrace_getpcstack(caller, 2, aframes, 2879 (uint32_t *)(uintptr_t)mstate->dtms_arg[0]); 2880 mstate->dtms_caller = caller[1]; 2881 } else if ((mstate->dtms_caller = 2882 dtrace_caller(aframes)) == -1) { 2883 /* 2884 * We have failed to do this the quick way; 2885 * we must resort to the slower approach of 2886 * calling dtrace_getpcstack(). 2887 */ 2888 pc_t caller = 0; 2889 2890 dtrace_getpcstack(&caller, 1, aframes, NULL); 2891 mstate->dtms_caller = caller; 2892 } 2893 2894 mstate->dtms_present |= DTRACE_MSTATE_CALLER; 2895 } 2896 return (mstate->dtms_caller); 2897 2898#if defined(sun) 2899 case DIF_VAR_UCALLER: 2900 if (!dtrace_priv_proc(state)) 2901 return (0); 2902 2903 if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) { 2904 uint64_t ustack[3]; 2905 2906 /* 2907 * dtrace_getupcstack() fills in the first uint64_t 2908 * with the current PID. The second uint64_t will 2909 * be the program counter at user-level. The third 2910 * uint64_t will contain the caller, which is what 2911 * we're after. 2912 */ 2913 ustack[2] = 0; 2914 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 2915 dtrace_getupcstack(ustack, 3); 2916 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 2917 mstate->dtms_ucaller = ustack[2]; 2918 mstate->dtms_present |= DTRACE_MSTATE_UCALLER; 2919 } 2920 2921 return (mstate->dtms_ucaller); 2922#endif 2923 2924 case DIF_VAR_PROBEPROV: 2925 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 2926 return (dtrace_dif_varstr( 2927 (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name, 2928 state, mstate)); 2929 2930 case DIF_VAR_PROBEMOD: 2931 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 2932 return (dtrace_dif_varstr( 2933 (uintptr_t)mstate->dtms_probe->dtpr_mod, 2934 state, mstate)); 2935 2936 case DIF_VAR_PROBEFUNC: 2937 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 2938 return (dtrace_dif_varstr( 2939 (uintptr_t)mstate->dtms_probe->dtpr_func, 2940 state, mstate)); 2941 2942 case DIF_VAR_PROBENAME: 2943 ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); 2944 return (dtrace_dif_varstr( 2945 (uintptr_t)mstate->dtms_probe->dtpr_name, 2946 state, mstate)); 2947 2948 case DIF_VAR_PID: 2949 if (!dtrace_priv_proc(state)) 2950 return (0); 2951 2952#if defined(sun) 2953 /* 2954 * Note that we are assuming that an unanchored probe is 2955 * always due to a high-level interrupt. (And we're assuming 2956 * that there is only a single high level interrupt.) 2957 */ 2958 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 2959 return (pid0.pid_id); 2960 2961 /* 2962 * It is always safe to dereference one's own t_procp pointer: 2963 * it always points to a valid, allocated proc structure. 2964 * Further, it is always safe to dereference the p_pidp member 2965 * of one's own proc structure. (These are truisms becuase 2966 * threads and processes don't clean up their own state -- 2967 * they leave that task to whomever reaps them.) 2968 */ 2969 return ((uint64_t)curthread->t_procp->p_pidp->pid_id); 2970#else 2971 return ((uint64_t)curproc->p_pid); 2972#endif 2973 2974 case DIF_VAR_PPID: 2975 if (!dtrace_priv_proc(state)) 2976 return (0); 2977 2978#if defined(sun) 2979 /* 2980 * See comment in DIF_VAR_PID. 2981 */ 2982 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 2983 return (pid0.pid_id); 2984 2985 /* 2986 * It is always safe to dereference one's own t_procp pointer: 2987 * it always points to a valid, allocated proc structure. 2988 * (This is true because threads don't clean up their own 2989 * state -- they leave that task to whomever reaps them.) 2990 */ 2991 return ((uint64_t)curthread->t_procp->p_ppid); 2992#else 2993 return ((uint64_t)curproc->p_pptr->p_pid); 2994#endif 2995 2996 case DIF_VAR_TID: 2997#if defined(sun) 2998 /* 2999 * See comment in DIF_VAR_PID. 3000 */ 3001 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3002 return (0); 3003#endif 3004 3005 return ((uint64_t)curthread->t_tid); 3006 3007 case DIF_VAR_EXECARGS: { 3008 struct pargs *p_args = curthread->td_proc->p_args; 3009 3010 if (p_args == NULL) 3011 return(0); 3012 3013 return (dtrace_dif_varstrz( 3014 (uintptr_t) p_args->ar_args, p_args->ar_length, state, mstate)); 3015 } 3016 3017 case DIF_VAR_EXECNAME: 3018#if defined(sun) 3019 if (!dtrace_priv_proc(state)) 3020 return (0); 3021 3022 /* 3023 * See comment in DIF_VAR_PID. 3024 */ 3025 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3026 return ((uint64_t)(uintptr_t)p0.p_user.u_comm); 3027 3028 /* 3029 * It is always safe to dereference one's own t_procp pointer: 3030 * it always points to a valid, allocated proc structure. 3031 * (This is true because threads don't clean up their own 3032 * state -- they leave that task to whomever reaps them.) 3033 */ 3034 return (dtrace_dif_varstr( 3035 (uintptr_t)curthread->t_procp->p_user.u_comm, 3036 state, mstate)); 3037#else 3038 return (dtrace_dif_varstr( 3039 (uintptr_t) curthread->td_proc->p_comm, state, mstate)); 3040#endif 3041 3042 case DIF_VAR_ZONENAME: 3043#if defined(sun) 3044 if (!dtrace_priv_proc(state)) 3045 return (0); 3046 3047 /* 3048 * See comment in DIF_VAR_PID. 3049 */ 3050 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3051 return ((uint64_t)(uintptr_t)p0.p_zone->zone_name); 3052 3053 /* 3054 * It is always safe to dereference one's own t_procp pointer: 3055 * it always points to a valid, allocated proc structure. 3056 * (This is true because threads don't clean up their own 3057 * state -- they leave that task to whomever reaps them.) 3058 */ 3059 return (dtrace_dif_varstr( 3060 (uintptr_t)curthread->t_procp->p_zone->zone_name, 3061 state, mstate)); 3062#else 3063 return (0); 3064#endif 3065 3066 case DIF_VAR_UID: 3067 if (!dtrace_priv_proc(state)) 3068 return (0); 3069 3070#if defined(sun) 3071 /* 3072 * See comment in DIF_VAR_PID. 3073 */ 3074 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3075 return ((uint64_t)p0.p_cred->cr_uid); 3076#endif 3077 3078 /* 3079 * It is always safe to dereference one's own t_procp pointer: 3080 * it always points to a valid, allocated proc structure. 3081 * (This is true because threads don't clean up their own 3082 * state -- they leave that task to whomever reaps them.) 3083 * 3084 * Additionally, it is safe to dereference one's own process 3085 * credential, since this is never NULL after process birth. 3086 */ 3087 return ((uint64_t)curthread->t_procp->p_cred->cr_uid); 3088 3089 case DIF_VAR_GID: 3090 if (!dtrace_priv_proc(state)) 3091 return (0); 3092 3093#if defined(sun) 3094 /* 3095 * See comment in DIF_VAR_PID. 3096 */ 3097 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3098 return ((uint64_t)p0.p_cred->cr_gid); 3099#endif 3100 3101 /* 3102 * It is always safe to dereference one's own t_procp pointer: 3103 * it always points to a valid, allocated proc structure. 3104 * (This is true because threads don't clean up their own 3105 * state -- they leave that task to whomever reaps them.) 3106 * 3107 * Additionally, it is safe to dereference one's own process 3108 * credential, since this is never NULL after process birth. 3109 */ 3110 return ((uint64_t)curthread->t_procp->p_cred->cr_gid); 3111 3112 case DIF_VAR_ERRNO: { 3113#if defined(sun) 3114 klwp_t *lwp; 3115 if (!dtrace_priv_proc(state)) 3116 return (0); 3117 3118 /* 3119 * See comment in DIF_VAR_PID. 3120 */ 3121 if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) 3122 return (0); 3123 3124 /* 3125 * It is always safe to dereference one's own t_lwp pointer in 3126 * the event that this pointer is non-NULL. (This is true 3127 * because threads and lwps don't clean up their own state -- 3128 * they leave that task to whomever reaps them.) 3129 */ 3130 if ((lwp = curthread->t_lwp) == NULL) 3131 return (0); 3132 3133 return ((uint64_t)lwp->lwp_errno); 3134#else 3135 return (curthread->td_errno); 3136#endif 3137 } 3138 default: 3139 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); 3140 return (0); 3141 } 3142} 3143 3144/* 3145 * Emulate the execution of DTrace ID subroutines invoked by the call opcode. 3146 * Notice that we don't bother validating the proper number of arguments or 3147 * their types in the tuple stack. This isn't needed because all argument 3148 * interpretation is safe because of our load safety -- the worst that can 3149 * happen is that a bogus program can obtain bogus results. 3150 */ 3151static void 3152dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs, 3153 dtrace_key_t *tupregs, int nargs, 3154 dtrace_mstate_t *mstate, dtrace_state_t *state) 3155{ 3156 volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; 3157 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; 3158 dtrace_vstate_t *vstate = &state->dts_vstate; 3159 3160#if defined(sun) 3161 union { 3162 mutex_impl_t mi; 3163 uint64_t mx; 3164 } m; 3165 3166 union { 3167 krwlock_t ri; 3168 uintptr_t rw; 3169 } r; 3170#else 3171 union { 3172 struct mtx *mi; 3173 uintptr_t mx; 3174 } m; 3175 union { 3176 struct sx *si; 3177 uintptr_t sx; 3178 } s; 3179#endif 3180 3181 switch (subr) { 3182 case DIF_SUBR_RAND: 3183 regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875; 3184 break; 3185 3186#if defined(sun) 3187 case DIF_SUBR_MUTEX_OWNED: 3188 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), 3189 mstate, vstate)) { 3190 regs[rd] = 0; 3191 break; 3192 } 3193 3194 m.mx = dtrace_load64(tupregs[0].dttk_value); 3195 if (MUTEX_TYPE_ADAPTIVE(&m.mi)) 3196 regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER; 3197 else 3198 regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock); 3199 break; 3200 3201 case DIF_SUBR_MUTEX_OWNER: 3202 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), 3203 mstate, vstate)) { 3204 regs[rd] = 0; 3205 break; 3206 } 3207 3208 m.mx = dtrace_load64(tupregs[0].dttk_value); 3209 if (MUTEX_TYPE_ADAPTIVE(&m.mi) && 3210 MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER) 3211 regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi); 3212 else 3213 regs[rd] = 0; 3214 break; 3215 3216 case DIF_SUBR_MUTEX_TYPE_ADAPTIVE: 3217 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), 3218 mstate, vstate)) { 3219 regs[rd] = 0; 3220 break; 3221 } 3222 3223 m.mx = dtrace_load64(tupregs[0].dttk_value); 3224 regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi); 3225 break; 3226 3227 case DIF_SUBR_MUTEX_TYPE_SPIN: 3228 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), 3229 mstate, vstate)) { 3230 regs[rd] = 0; 3231 break; 3232 } 3233 3234 m.mx = dtrace_load64(tupregs[0].dttk_value); 3235 regs[rd] = MUTEX_TYPE_SPIN(&m.mi); 3236 break; 3237 3238 case DIF_SUBR_RW_READ_HELD: { 3239 uintptr_t tmp; 3240 3241 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t), 3242 mstate, vstate)) { 3243 regs[rd] = 0; 3244 break; 3245 } 3246 3247 r.rw = dtrace_loadptr(tupregs[0].dttk_value); 3248 regs[rd] = _RW_READ_HELD(&r.ri, tmp); 3249 break; 3250 } 3251 3252 case DIF_SUBR_RW_WRITE_HELD: 3253 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t), 3254 mstate, vstate)) { 3255 regs[rd] = 0; 3256 break; 3257 } 3258 3259 r.rw = dtrace_loadptr(tupregs[0].dttk_value); 3260 regs[rd] = _RW_WRITE_HELD(&r.ri); 3261 break; 3262 3263 case DIF_SUBR_RW_ISWRITER: 3264 if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t), 3265 mstate, vstate)) { 3266 regs[rd] = 0; 3267 break; 3268 } 3269 3270 r.rw = dtrace_loadptr(tupregs[0].dttk_value); 3271 regs[rd] = _RW_ISWRITER(&r.ri); 3272 break; 3273 3274#else 3275 /* 3276 * XXX - The following code works because mutex, rwlocks, & sxlocks 3277 * all have similar data structures in FreeBSD. This may not be 3278 * good if someone changes one of the lock data structures. 3279 * Ideally, it would be nice if all these shared a common lock 3280 * object. 3281 */ 3282 case DIF_SUBR_MUTEX_OWNED: 3283 /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ 3284 m.mx = tupregs[0].dttk_value; 3285 3286#ifdef DOODAD 3287 if (LO_CLASSINDEX(&(m.mi->lock_object)) < 2) { 3288 regs[rd] = !(m.mi->mtx_lock & MTX_UNOWNED); 3289 } else { 3290 regs[rd] = !(m.mi->mtx_lock & SX_UNLOCKED); 3291 } 3292#endif 3293 break; 3294 3295 case DIF_SUBR_MUTEX_OWNER: 3296 /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ 3297 m.mx = tupregs[0].dttk_value; 3298 3299 if (LO_CLASSINDEX(&(m.mi->lock_object)) < 2) { 3300 regs[rd] = m.mi->mtx_lock & ~MTX_FLAGMASK; 3301 } else { 3302 if (!(m.mi->mtx_lock & SX_LOCK_SHARED)) 3303 regs[rd] = SX_OWNER(m.mi->mtx_lock); 3304 else 3305 regs[rd] = 0; 3306 } 3307 break; 3308 3309 case DIF_SUBR_MUTEX_TYPE_ADAPTIVE: 3310 /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ 3311 m.mx = tupregs[0].dttk_value; 3312 3313 regs[rd] = (LO_CLASSINDEX(&(m.mi->lock_object)) != 0); 3314 break; 3315 3316 case DIF_SUBR_MUTEX_TYPE_SPIN: 3317 /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ 3318 m.mx = tupregs[0].dttk_value; 3319 3320 regs[rd] = (LO_CLASSINDEX(&(m.mi->lock_object)) == 0); 3321 break; 3322 3323 case DIF_SUBR_RW_READ_HELD: 3324 case DIF_SUBR_SX_SHARED_HELD: 3325 /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ 3326 s.sx = tupregs[0].dttk_value; 3327 regs[rd] = ((s.si->sx_lock & SX_LOCK_SHARED) && 3328 (SX_OWNER(s.si->sx_lock) >> SX_SHARERS_SHIFT) != 0); 3329 break; 3330 3331 case DIF_SUBR_RW_WRITE_HELD: 3332 case DIF_SUBR_SX_EXCLUSIVE_HELD: 3333 /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ 3334 s.sx = tupregs[0].dttk_value; 3335 regs[rd] = (SX_OWNER(s.si->sx_lock) == (uintptr_t) curthread); 3336 break; 3337 3338 case DIF_SUBR_RW_ISWRITER: 3339 case DIF_SUBR_SX_ISEXCLUSIVE: 3340 /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ 3341 s.sx = tupregs[0].dttk_value; 3342 regs[rd] = ((s.si->sx_lock & SX_LOCK_EXCLUSIVE_WAITERS) || 3343 !(s.si->sx_lock & SX_LOCK_SHARED)); 3344 break; 3345#endif /* ! defined(sun) */ 3346 3347 case DIF_SUBR_BCOPY: { 3348 /* 3349 * We need to be sure that the destination is in the scratch 3350 * region -- no other region is allowed. 3351 */ 3352 uintptr_t src = tupregs[0].dttk_value; 3353 uintptr_t dest = tupregs[1].dttk_value; 3354 size_t size = tupregs[2].dttk_value; 3355 3356 if (!dtrace_inscratch(dest, size, mstate)) { 3357 *flags |= CPU_DTRACE_BADADDR; 3358 *illval = regs[rd]; 3359 break; 3360 } 3361 3362 if (!dtrace_canload(src, size, mstate, vstate)) { 3363 regs[rd] = 0; 3364 break; 3365 } 3366 3367 dtrace_bcopy((void *)src, (void *)dest, size); 3368 break; 3369 } 3370 3371 case DIF_SUBR_ALLOCA: 3372 case DIF_SUBR_COPYIN: { 3373 uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); 3374 uint64_t size = 3375 tupregs[subr == DIF_SUBR_ALLOCA ? 0 : 1].dttk_value; 3376 size_t scratch_size = (dest - mstate->dtms_scratch_ptr) + size; 3377 3378 /* 3379 * This action doesn't require any credential checks since 3380 * probes will not activate in user contexts to which the 3381 * enabling user does not have permissions. 3382 */ 3383 3384 /* 3385 * Rounding up the user allocation size could have overflowed 3386 * a large, bogus allocation (like -1ULL) to 0. 3387 */ 3388 if (scratch_size < size || 3389 !DTRACE_INSCRATCH(mstate, scratch_size)) { 3390 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 3391 regs[rd] = 0; 3392 break; 3393 } 3394 3395 if (subr == DIF_SUBR_COPYIN) { 3396 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3397 dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); 3398 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3399 } 3400 3401 mstate->dtms_scratch_ptr += scratch_size; 3402 regs[rd] = dest; 3403 break; 3404 } 3405 3406 case DIF_SUBR_COPYINTO: { 3407 uint64_t size = tupregs[1].dttk_value; 3408 uintptr_t dest = tupregs[2].dttk_value; 3409 3410 /* 3411 * This action doesn't require any credential checks since 3412 * probes will not activate in user contexts to which the 3413 * enabling user does not have permissions. 3414 */ 3415 if (!dtrace_inscratch(dest, size, mstate)) { 3416 *flags |= CPU_DTRACE_BADADDR; 3417 *illval = regs[rd]; 3418 break; 3419 } 3420 3421 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3422 dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); 3423 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3424 break; 3425 } 3426 3427 case DIF_SUBR_COPYINSTR: { 3428 uintptr_t dest = mstate->dtms_scratch_ptr; 3429 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 3430 3431 if (nargs > 1 && tupregs[1].dttk_value < size) 3432 size = tupregs[1].dttk_value + 1; 3433 3434 /* 3435 * This action doesn't require any credential checks since 3436 * probes will not activate in user contexts to which the 3437 * enabling user does not have permissions. 3438 */ 3439 if (!DTRACE_INSCRATCH(mstate, size)) { 3440 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 3441 regs[rd] = 0; 3442 break; 3443 } 3444 3445 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3446 dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags); 3447 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3448 3449 ((char *)dest)[size - 1] = '\0'; 3450 mstate->dtms_scratch_ptr += size; 3451 regs[rd] = dest; 3452 break; 3453 } 3454 3455#if defined(sun) 3456 case DIF_SUBR_MSGSIZE: 3457 case DIF_SUBR_MSGDSIZE: { 3458 uintptr_t baddr = tupregs[0].dttk_value, daddr; 3459 uintptr_t wptr, rptr; 3460 size_t count = 0; 3461 int cont = 0; 3462 3463 while (baddr != 0 && !(*flags & CPU_DTRACE_FAULT)) { 3464 3465 if (!dtrace_canload(baddr, sizeof (mblk_t), mstate, 3466 vstate)) { 3467 regs[rd] = 0; 3468 break; 3469 } 3470 3471 wptr = dtrace_loadptr(baddr + 3472 offsetof(mblk_t, b_wptr)); 3473 3474 rptr = dtrace_loadptr(baddr + 3475 offsetof(mblk_t, b_rptr)); 3476 3477 if (wptr < rptr) { 3478 *flags |= CPU_DTRACE_BADADDR; 3479 *illval = tupregs[0].dttk_value; 3480 break; 3481 } 3482 3483 daddr = dtrace_loadptr(baddr + 3484 offsetof(mblk_t, b_datap)); 3485 3486 baddr = dtrace_loadptr(baddr + 3487 offsetof(mblk_t, b_cont)); 3488 3489 /* 3490 * We want to prevent against denial-of-service here, 3491 * so we're only going to search the list for 3492 * dtrace_msgdsize_max mblks. 3493 */ 3494 if (cont++ > dtrace_msgdsize_max) { 3495 *flags |= CPU_DTRACE_ILLOP; 3496 break; 3497 } 3498 3499 if (subr == DIF_SUBR_MSGDSIZE) { 3500 if (dtrace_load8(daddr + 3501 offsetof(dblk_t, db_type)) != M_DATA) 3502 continue; 3503 } 3504 3505 count += wptr - rptr; 3506 } 3507 3508 if (!(*flags & CPU_DTRACE_FAULT)) 3509 regs[rd] = count; 3510 3511 break; 3512 } 3513#endif 3514 3515 case DIF_SUBR_PROGENYOF: { 3516 pid_t pid = tupregs[0].dttk_value; 3517 proc_t *p; 3518 int rval = 0; 3519 3520 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3521 3522 for (p = curthread->t_procp; p != NULL; p = p->p_parent) { 3523#if defined(sun) 3524 if (p->p_pidp->pid_id == pid) { 3525#else 3526 if (p->p_pid == pid) { 3527#endif 3528 rval = 1; 3529 break; 3530 } 3531 } 3532 3533 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3534 3535 regs[rd] = rval; 3536 break; 3537 } 3538 3539 case DIF_SUBR_SPECULATION: 3540 regs[rd] = dtrace_speculation(state); 3541 break; 3542 3543 case DIF_SUBR_COPYOUT: { 3544 uintptr_t kaddr = tupregs[0].dttk_value; 3545 uintptr_t uaddr = tupregs[1].dttk_value; 3546 uint64_t size = tupregs[2].dttk_value; 3547 3548 if (!dtrace_destructive_disallow && 3549 dtrace_priv_proc_control(state) && 3550 !dtrace_istoxic(kaddr, size)) { 3551 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3552 dtrace_copyout(kaddr, uaddr, size, flags); 3553 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3554 } 3555 break; 3556 } 3557 3558 case DIF_SUBR_COPYOUTSTR: { 3559 uintptr_t kaddr = tupregs[0].dttk_value; 3560 uintptr_t uaddr = tupregs[1].dttk_value; 3561 uint64_t size = tupregs[2].dttk_value; 3562 3563 if (!dtrace_destructive_disallow && 3564 dtrace_priv_proc_control(state) && 3565 !dtrace_istoxic(kaddr, size)) { 3566 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 3567 dtrace_copyoutstr(kaddr, uaddr, size, flags); 3568 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 3569 } 3570 break; 3571 } 3572 3573 case DIF_SUBR_STRLEN: { 3574 size_t sz; 3575 uintptr_t addr = (uintptr_t)tupregs[0].dttk_value; 3576 sz = dtrace_strlen((char *)addr, 3577 state->dts_options[DTRACEOPT_STRSIZE]); 3578 3579 if (!dtrace_canload(addr, sz + 1, mstate, vstate)) { 3580 regs[rd] = 0; 3581 break; 3582 } 3583 3584 regs[rd] = sz; 3585 3586 break; 3587 } 3588 3589 case DIF_SUBR_STRCHR: 3590 case DIF_SUBR_STRRCHR: { 3591 /* 3592 * We're going to iterate over the string looking for the 3593 * specified character. We will iterate until we have reached 3594 * the string length or we have found the character. If this 3595 * is DIF_SUBR_STRRCHR, we will look for the last occurrence 3596 * of the specified character instead of the first. 3597 */ 3598 uintptr_t saddr = tupregs[0].dttk_value; 3599 uintptr_t addr = tupregs[0].dttk_value; 3600 uintptr_t limit = addr + state->dts_options[DTRACEOPT_STRSIZE]; 3601 char c, target = (char)tupregs[1].dttk_value; 3602 3603 for (regs[rd] = 0; addr < limit; addr++) { 3604 if ((c = dtrace_load8(addr)) == target) { 3605 regs[rd] = addr; 3606 3607 if (subr == DIF_SUBR_STRCHR) 3608 break; 3609 } 3610 3611 if (c == '\0') 3612 break; 3613 } 3614 3615 if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) { 3616 regs[rd] = 0; 3617 break; 3618 } 3619 3620 break; 3621 } 3622 3623 case DIF_SUBR_STRSTR: 3624 case DIF_SUBR_INDEX: 3625 case DIF_SUBR_RINDEX: { 3626 /* 3627 * We're going to iterate over the string looking for the 3628 * specified string. We will iterate until we have reached 3629 * the string length or we have found the string. (Yes, this 3630 * is done in the most naive way possible -- but considering 3631 * that the string we're searching for is likely to be 3632 * relatively short, the complexity of Rabin-Karp or similar 3633 * hardly seems merited.) 3634 */ 3635 char *addr = (char *)(uintptr_t)tupregs[0].dttk_value; 3636 char *substr = (char *)(uintptr_t)tupregs[1].dttk_value; 3637 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 3638 size_t len = dtrace_strlen(addr, size); 3639 size_t sublen = dtrace_strlen(substr, size); 3640 char *limit = addr + len, *orig = addr; 3641 int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1; 3642 int inc = 1; 3643 3644 regs[rd] = notfound; 3645 3646 if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) { 3647 regs[rd] = 0; 3648 break; 3649 } 3650 3651 if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate, 3652 vstate)) { 3653 regs[rd] = 0; 3654 break; 3655 } 3656 3657 /* 3658 * strstr() and index()/rindex() have similar semantics if 3659 * both strings are the empty string: strstr() returns a 3660 * pointer to the (empty) string, and index() and rindex() 3661 * both return index 0 (regardless of any position argument). 3662 */ 3663 if (sublen == 0 && len == 0) { 3664 if (subr == DIF_SUBR_STRSTR) 3665 regs[rd] = (uintptr_t)addr; 3666 else 3667 regs[rd] = 0; 3668 break; 3669 } 3670 3671 if (subr != DIF_SUBR_STRSTR) { 3672 if (subr == DIF_SUBR_RINDEX) { 3673 limit = orig - 1; 3674 addr += len; 3675 inc = -1; 3676 } 3677 3678 /* 3679 * Both index() and rindex() take an optional position 3680 * argument that denotes the starting position. 3681 */ 3682 if (nargs == 3) { 3683 int64_t pos = (int64_t)tupregs[2].dttk_value; 3684 3685 /* 3686 * If the position argument to index() is 3687 * negative, Perl implicitly clamps it at 3688 * zero. This semantic is a little surprising 3689 * given the special meaning of negative 3690 * positions to similar Perl functions like 3691 * substr(), but it appears to reflect a 3692 * notion that index() can start from a 3693 * negative index and increment its way up to 3694 * the string. Given this notion, Perl's 3695 * rindex() is at least self-consistent in 3696 * that it implicitly clamps positions greater 3697 * than the string length to be the string 3698 * length. Where Perl completely loses 3699 * coherence, however, is when the specified 3700 * substring is the empty string (""). In 3701 * this case, even if the position is 3702 * negative, rindex() returns 0 -- and even if 3703 * the position is greater than the length, 3704 * index() returns the string length. These 3705 * semantics violate the notion that index() 3706 * should never return a value less than the 3707 * specified position and that rindex() should 3708 * never return a value greater than the 3709 * specified position. (One assumes that 3710 * these semantics are artifacts of Perl's 3711 * implementation and not the results of 3712 * deliberate design -- it beggars belief that 3713 * even Larry Wall could desire such oddness.) 3714 * While in the abstract one would wish for 3715 * consistent position semantics across 3716 * substr(), index() and rindex() -- or at the 3717 * very least self-consistent position 3718 * semantics for index() and rindex() -- we 3719 * instead opt to keep with the extant Perl 3720 * semantics, in all their broken glory. (Do 3721 * we have more desire to maintain Perl's 3722 * semantics than Perl does? Probably.) 3723 */ 3724 if (subr == DIF_SUBR_RINDEX) { 3725 if (pos < 0) { 3726 if (sublen == 0) 3727 regs[rd] = 0; 3728 break; 3729 } 3730 3731 if (pos > len) 3732 pos = len; 3733 } else { 3734 if (pos < 0) 3735 pos = 0; 3736 3737 if (pos >= len) { 3738 if (sublen == 0) 3739 regs[rd] = len; 3740 break; 3741 } 3742 } 3743 3744 addr = orig + pos; 3745 } 3746 } 3747 3748 for (regs[rd] = notfound; addr != limit; addr += inc) { 3749 if (dtrace_strncmp(addr, substr, sublen) == 0) { 3750 if (subr != DIF_SUBR_STRSTR) { 3751 /* 3752 * As D index() and rindex() are 3753 * modeled on Perl (and not on awk), 3754 * we return a zero-based (and not a 3755 * one-based) index. (For you Perl 3756 * weenies: no, we're not going to add 3757 * $[ -- and shouldn't you be at a con 3758 * or something?) 3759 */ 3760 regs[rd] = (uintptr_t)(addr - orig); 3761 break; 3762 } 3763 3764 ASSERT(subr == DIF_SUBR_STRSTR); 3765 regs[rd] = (uintptr_t)addr; 3766 break; 3767 } 3768 } 3769 3770 break; 3771 } 3772 3773 case DIF_SUBR_STRTOK: { 3774 uintptr_t addr = tupregs[0].dttk_value; 3775 uintptr_t tokaddr = tupregs[1].dttk_value; 3776 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 3777 uintptr_t limit, toklimit = tokaddr + size; 3778 uint8_t c = 0, tokmap[32]; /* 256 / 8 */ 3779 char *dest = (char *)mstate->dtms_scratch_ptr; 3780 int i; 3781 3782 /* 3783 * Check both the token buffer and (later) the input buffer, 3784 * since both could be non-scratch addresses. 3785 */ 3786 if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) { 3787 regs[rd] = 0; 3788 break; 3789 } 3790 3791 if (!DTRACE_INSCRATCH(mstate, size)) { 3792 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 3793 regs[rd] = 0; 3794 break; 3795 } 3796 3797 if (addr == 0) { 3798 /* 3799 * If the address specified is NULL, we use our saved 3800 * strtok pointer from the mstate. Note that this 3801 * means that the saved strtok pointer is _only_ 3802 * valid within multiple enablings of the same probe -- 3803 * it behaves like an implicit clause-local variable. 3804 */ 3805 addr = mstate->dtms_strtok; 3806 } else { 3807 /* 3808 * If the user-specified address is non-NULL we must 3809 * access check it. This is the only time we have 3810 * a chance to do so, since this address may reside 3811 * in the string table of this clause-- future calls 3812 * (when we fetch addr from mstate->dtms_strtok) 3813 * would fail this access check. 3814 */ 3815 if (!dtrace_strcanload(addr, size, mstate, vstate)) { 3816 regs[rd] = 0; 3817 break; 3818 } 3819 } 3820 3821 /* 3822 * First, zero the token map, and then process the token 3823 * string -- setting a bit in the map for every character 3824 * found in the token string. 3825 */ 3826 for (i = 0; i < sizeof (tokmap); i++) 3827 tokmap[i] = 0; 3828 3829 for (; tokaddr < toklimit; tokaddr++) { 3830 if ((c = dtrace_load8(tokaddr)) == '\0') 3831 break; 3832 3833 ASSERT((c >> 3) < sizeof (tokmap)); 3834 tokmap[c >> 3] |= (1 << (c & 0x7)); 3835 } 3836 3837 for (limit = addr + size; addr < limit; addr++) { 3838 /* 3839 * We're looking for a character that is _not_ contained 3840 * in the token string. 3841 */ 3842 if ((c = dtrace_load8(addr)) == '\0') 3843 break; 3844 3845 if (!(tokmap[c >> 3] & (1 << (c & 0x7)))) 3846 break; 3847 } 3848 3849 if (c == '\0') { 3850 /* 3851 * We reached the end of the string without finding 3852 * any character that was not in the token string. 3853 * We return NULL in this case, and we set the saved 3854 * address to NULL as well. 3855 */ 3856 regs[rd] = 0; 3857 mstate->dtms_strtok = 0; 3858 break; 3859 } 3860 3861 /* 3862 * From here on, we're copying into the destination string. 3863 */ 3864 for (i = 0; addr < limit && i < size - 1; addr++) { 3865 if ((c = dtrace_load8(addr)) == '\0') 3866 break; 3867 3868 if (tokmap[c >> 3] & (1 << (c & 0x7))) 3869 break; 3870 3871 ASSERT(i < size); 3872 dest[i++] = c; 3873 } 3874 3875 ASSERT(i < size); 3876 dest[i] = '\0'; 3877 regs[rd] = (uintptr_t)dest; 3878 mstate->dtms_scratch_ptr += size; 3879 mstate->dtms_strtok = addr; 3880 break; 3881 } 3882 3883 case DIF_SUBR_SUBSTR: { 3884 uintptr_t s = tupregs[0].dttk_value; 3885 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 3886 char *d = (char *)mstate->dtms_scratch_ptr; 3887 int64_t index = (int64_t)tupregs[1].dttk_value; 3888 int64_t remaining = (int64_t)tupregs[2].dttk_value; 3889 size_t len = dtrace_strlen((char *)s, size); 3890 int64_t i = 0; 3891 3892 if (!dtrace_canload(s, len + 1, mstate, vstate)) { 3893 regs[rd] = 0; 3894 break; 3895 } 3896 3897 if (!DTRACE_INSCRATCH(mstate, size)) { 3898 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 3899 regs[rd] = 0; 3900 break; 3901 } 3902 3903 if (nargs <= 2) 3904 remaining = (int64_t)size; 3905 3906 if (index < 0) { 3907 index += len; 3908 3909 if (index < 0 && index + remaining > 0) { 3910 remaining += index; 3911 index = 0; 3912 } 3913 } 3914 3915 if (index >= len || index < 0) { 3916 remaining = 0; 3917 } else if (remaining < 0) { 3918 remaining += len - index; 3919 } else if (index + remaining > size) { 3920 remaining = size - index; 3921 } 3922 3923 for (i = 0; i < remaining; i++) { 3924 if ((d[i] = dtrace_load8(s + index + i)) == '\0') 3925 break; 3926 } 3927 3928 d[i] = '\0'; 3929 3930 mstate->dtms_scratch_ptr += size; 3931 regs[rd] = (uintptr_t)d; 3932 break; 3933 } 3934 3935#if defined(sun) 3936 case DIF_SUBR_GETMAJOR: 3937#ifdef _LP64 3938 regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR64) & MAXMAJ64; 3939#else 3940 regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR) & MAXMAJ; 3941#endif 3942 break; 3943 3944 case DIF_SUBR_GETMINOR: 3945#ifdef _LP64 3946 regs[rd] = tupregs[0].dttk_value & MAXMIN64; 3947#else 3948 regs[rd] = tupregs[0].dttk_value & MAXMIN; 3949#endif 3950 break; 3951 3952 case DIF_SUBR_DDI_PATHNAME: { 3953 /* 3954 * This one is a galactic mess. We are going to roughly 3955 * emulate ddi_pathname(), but it's made more complicated 3956 * by the fact that we (a) want to include the minor name and 3957 * (b) must proceed iteratively instead of recursively. 3958 */ 3959 uintptr_t dest = mstate->dtms_scratch_ptr; 3960 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 3961 char *start = (char *)dest, *end = start + size - 1; 3962 uintptr_t daddr = tupregs[0].dttk_value; 3963 int64_t minor = (int64_t)tupregs[1].dttk_value; 3964 char *s; 3965 int i, len, depth = 0; 3966 3967 /* 3968 * Due to all the pointer jumping we do and context we must 3969 * rely upon, we just mandate that the user must have kernel 3970 * read privileges to use this routine. 3971 */ 3972 if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) { 3973 *flags |= CPU_DTRACE_KPRIV; 3974 *illval = daddr; 3975 regs[rd] = 0; 3976 } 3977 3978 if (!DTRACE_INSCRATCH(mstate, size)) { 3979 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 3980 regs[rd] = 0; 3981 break; 3982 } 3983 3984 *end = '\0'; 3985 3986 /* 3987 * We want to have a name for the minor. In order to do this, 3988 * we need to walk the minor list from the devinfo. We want 3989 * to be sure that we don't infinitely walk a circular list, 3990 * so we check for circularity by sending a scout pointer 3991 * ahead two elements for every element that we iterate over; 3992 * if the list is circular, these will ultimately point to the 3993 * same element. You may recognize this little trick as the 3994 * answer to a stupid interview question -- one that always 3995 * seems to be asked by those who had to have it laboriously 3996 * explained to them, and who can't even concisely describe 3997 * the conditions under which one would be forced to resort to 3998 * this technique. Needless to say, those conditions are 3999 * found here -- and probably only here. Is this the only use 4000 * of this infamous trick in shipping, production code? If it 4001 * isn't, it probably should be... 4002 */ 4003 if (minor != -1) { 4004 uintptr_t maddr = dtrace_loadptr(daddr + 4005 offsetof(struct dev_info, devi_minor)); 4006 4007 uintptr_t next = offsetof(struct ddi_minor_data, next); 4008 uintptr_t name = offsetof(struct ddi_minor_data, 4009 d_minor) + offsetof(struct ddi_minor, name); 4010 uintptr_t dev = offsetof(struct ddi_minor_data, 4011 d_minor) + offsetof(struct ddi_minor, dev); 4012 uintptr_t scout; 4013 4014 if (maddr != NULL) 4015 scout = dtrace_loadptr(maddr + next); 4016 4017 while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { 4018 uint64_t m; 4019#ifdef _LP64 4020 m = dtrace_load64(maddr + dev) & MAXMIN64; 4021#else 4022 m = dtrace_load32(maddr + dev) & MAXMIN; 4023#endif 4024 if (m != minor) { 4025 maddr = dtrace_loadptr(maddr + next); 4026 4027 if (scout == NULL) 4028 continue; 4029 4030 scout = dtrace_loadptr(scout + next); 4031 4032 if (scout == NULL) 4033 continue; 4034 4035 scout = dtrace_loadptr(scout + next); 4036 4037 if (scout == NULL) 4038 continue; 4039 4040 if (scout == maddr) { 4041 *flags |= CPU_DTRACE_ILLOP; 4042 break; 4043 } 4044 4045 continue; 4046 } 4047 4048 /* 4049 * We have the minor data. Now we need to 4050 * copy the minor's name into the end of the 4051 * pathname. 4052 */ 4053 s = (char *)dtrace_loadptr(maddr + name); 4054 len = dtrace_strlen(s, size); 4055 4056 if (*flags & CPU_DTRACE_FAULT) 4057 break; 4058 4059 if (len != 0) { 4060 if ((end -= (len + 1)) < start) 4061 break; 4062 4063 *end = ':'; 4064 } 4065 4066 for (i = 1; i <= len; i++) 4067 end[i] = dtrace_load8((uintptr_t)s++); 4068 break; 4069 } 4070 } 4071 4072 while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { 4073 ddi_node_state_t devi_state; 4074 4075 devi_state = dtrace_load32(daddr + 4076 offsetof(struct dev_info, devi_node_state)); 4077 4078 if (*flags & CPU_DTRACE_FAULT) 4079 break; 4080 4081 if (devi_state >= DS_INITIALIZED) { 4082 s = (char *)dtrace_loadptr(daddr + 4083 offsetof(struct dev_info, devi_addr)); 4084 len = dtrace_strlen(s, size); 4085 4086 if (*flags & CPU_DTRACE_FAULT) 4087 break; 4088 4089 if (len != 0) { 4090 if ((end -= (len + 1)) < start) 4091 break; 4092 4093 *end = '@'; 4094 } 4095 4096 for (i = 1; i <= len; i++) 4097 end[i] = dtrace_load8((uintptr_t)s++); 4098 } 4099 4100 /* 4101 * Now for the node name... 4102 */ 4103 s = (char *)dtrace_loadptr(daddr + 4104 offsetof(struct dev_info, devi_node_name)); 4105 4106 daddr = dtrace_loadptr(daddr + 4107 offsetof(struct dev_info, devi_parent)); 4108 4109 /* 4110 * If our parent is NULL (that is, if we're the root 4111 * node), we're going to use the special path 4112 * "devices". 4113 */ 4114 if (daddr == 0) 4115 s = "devices"; 4116 4117 len = dtrace_strlen(s, size); 4118 if (*flags & CPU_DTRACE_FAULT) 4119 break; 4120 4121 if ((end -= (len + 1)) < start) 4122 break; 4123 4124 for (i = 1; i <= len; i++) 4125 end[i] = dtrace_load8((uintptr_t)s++); 4126 *end = '/'; 4127 4128 if (depth++ > dtrace_devdepth_max) { 4129 *flags |= CPU_DTRACE_ILLOP; 4130 break; 4131 } 4132 } 4133 4134 if (end < start) 4135 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4136 4137 if (daddr == 0) { 4138 regs[rd] = (uintptr_t)end; 4139 mstate->dtms_scratch_ptr += size; 4140 } 4141 4142 break; 4143 } 4144#endif 4145 4146 case DIF_SUBR_STRJOIN: { 4147 char *d = (char *)mstate->dtms_scratch_ptr; 4148 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4149 uintptr_t s1 = tupregs[0].dttk_value; 4150 uintptr_t s2 = tupregs[1].dttk_value; 4151 int i = 0; 4152 4153 if (!dtrace_strcanload(s1, size, mstate, vstate) || 4154 !dtrace_strcanload(s2, size, mstate, vstate)) { 4155 regs[rd] = 0; 4156 break; 4157 } 4158 4159 if (!DTRACE_INSCRATCH(mstate, size)) { 4160 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4161 regs[rd] = 0; 4162 break; 4163 } 4164 4165 for (;;) { 4166 if (i >= size) { 4167 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4168 regs[rd] = 0; 4169 break; 4170 } 4171 4172 if ((d[i++] = dtrace_load8(s1++)) == '\0') { 4173 i--; 4174 break; 4175 } 4176 } 4177 4178 for (;;) { 4179 if (i >= size) { 4180 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4181 regs[rd] = 0; 4182 break; 4183 } 4184 4185 if ((d[i++] = dtrace_load8(s2++)) == '\0') 4186 break; 4187 } 4188 4189 if (i < size) { 4190 mstate->dtms_scratch_ptr += i; 4191 regs[rd] = (uintptr_t)d; 4192 } 4193 4194 break; 4195 } 4196 4197 case DIF_SUBR_LLTOSTR: { 4198 int64_t i = (int64_t)tupregs[0].dttk_value; 4199 int64_t val = i < 0 ? i * -1 : i; 4200 uint64_t size = 22; /* enough room for 2^64 in decimal */ 4201 char *end = (char *)mstate->dtms_scratch_ptr + size - 1; 4202 4203 if (!DTRACE_INSCRATCH(mstate, size)) { 4204 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4205 regs[rd] = 0; 4206 break; 4207 } 4208 4209 for (*end-- = '\0'; val; val /= 10) 4210 *end-- = '0' + (val % 10); 4211 4212 if (i == 0) 4213 *end-- = '0'; 4214 4215 if (i < 0) 4216 *end-- = '-'; 4217 4218 regs[rd] = (uintptr_t)end + 1; 4219 mstate->dtms_scratch_ptr += size; 4220 break; 4221 } 4222 4223 case DIF_SUBR_HTONS: 4224 case DIF_SUBR_NTOHS: 4225#if BYTE_ORDER == BIG_ENDIAN 4226 regs[rd] = (uint16_t)tupregs[0].dttk_value; 4227#else 4228 regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value); 4229#endif 4230 break; 4231 4232 4233 case DIF_SUBR_HTONL: 4234 case DIF_SUBR_NTOHL: 4235#if BYTE_ORDER == BIG_ENDIAN 4236 regs[rd] = (uint32_t)tupregs[0].dttk_value; 4237#else 4238 regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value); 4239#endif 4240 break; 4241 4242 4243 case DIF_SUBR_HTONLL: 4244 case DIF_SUBR_NTOHLL: 4245#if BYTE_ORDER == BIG_ENDIAN 4246 regs[rd] = (uint64_t)tupregs[0].dttk_value; 4247#else 4248 regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value); 4249#endif 4250 break; 4251 4252 4253 case DIF_SUBR_DIRNAME: 4254 case DIF_SUBR_BASENAME: { 4255 char *dest = (char *)mstate->dtms_scratch_ptr; 4256 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4257 uintptr_t src = tupregs[0].dttk_value; 4258 int i, j, len = dtrace_strlen((char *)src, size); 4259 int lastbase = -1, firstbase = -1, lastdir = -1; 4260 int start, end; 4261 4262 if (!dtrace_canload(src, len + 1, mstate, vstate)) { 4263 regs[rd] = 0; 4264 break; 4265 } 4266 4267 if (!DTRACE_INSCRATCH(mstate, size)) { 4268 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4269 regs[rd] = 0; 4270 break; 4271 } 4272 4273 /* 4274 * The basename and dirname for a zero-length string is 4275 * defined to be "." 4276 */ 4277 if (len == 0) { 4278 len = 1; 4279 src = (uintptr_t)"."; 4280 } 4281 4282 /* 4283 * Start from the back of the string, moving back toward the 4284 * front until we see a character that isn't a slash. That 4285 * character is the last character in the basename. 4286 */ 4287 for (i = len - 1; i >= 0; i--) { 4288 if (dtrace_load8(src + i) != '/') 4289 break; 4290 } 4291 4292 if (i >= 0) 4293 lastbase = i; 4294 4295 /* 4296 * Starting from the last character in the basename, move 4297 * towards the front until we find a slash. The character 4298 * that we processed immediately before that is the first 4299 * character in the basename. 4300 */ 4301 for (; i >= 0; i--) { 4302 if (dtrace_load8(src + i) == '/') 4303 break; 4304 } 4305 4306 if (i >= 0) 4307 firstbase = i + 1; 4308 4309 /* 4310 * Now keep going until we find a non-slash character. That 4311 * character is the last character in the dirname. 4312 */ 4313 for (; i >= 0; i--) { 4314 if (dtrace_load8(src + i) != '/') 4315 break; 4316 } 4317 4318 if (i >= 0) 4319 lastdir = i; 4320 4321 ASSERT(!(lastbase == -1 && firstbase != -1)); 4322 ASSERT(!(firstbase == -1 && lastdir != -1)); 4323 4324 if (lastbase == -1) { 4325 /* 4326 * We didn't find a non-slash character. We know that 4327 * the length is non-zero, so the whole string must be 4328 * slashes. In either the dirname or the basename 4329 * case, we return '/'. 4330 */ 4331 ASSERT(firstbase == -1); 4332 firstbase = lastbase = lastdir = 0; 4333 } 4334 4335 if (firstbase == -1) { 4336 /* 4337 * The entire string consists only of a basename 4338 * component. If we're looking for dirname, we need 4339 * to change our string to be just "."; if we're 4340 * looking for a basename, we'll just set the first 4341 * character of the basename to be 0. 4342 */ 4343 if (subr == DIF_SUBR_DIRNAME) { 4344 ASSERT(lastdir == -1); 4345 src = (uintptr_t)"."; 4346 lastdir = 0; 4347 } else { 4348 firstbase = 0; 4349 } 4350 } 4351 4352 if (subr == DIF_SUBR_DIRNAME) { 4353 if (lastdir == -1) { 4354 /* 4355 * We know that we have a slash in the name -- 4356 * or lastdir would be set to 0, above. And 4357 * because lastdir is -1, we know that this 4358 * slash must be the first character. (That 4359 * is, the full string must be of the form 4360 * "/basename".) In this case, the last 4361 * character of the directory name is 0. 4362 */ 4363 lastdir = 0; 4364 } 4365 4366 start = 0; 4367 end = lastdir; 4368 } else { 4369 ASSERT(subr == DIF_SUBR_BASENAME); 4370 ASSERT(firstbase != -1 && lastbase != -1); 4371 start = firstbase; 4372 end = lastbase; 4373 } 4374 4375 for (i = start, j = 0; i <= end && j < size - 1; i++, j++) 4376 dest[j] = dtrace_load8(src + i); 4377 4378 dest[j] = '\0'; 4379 regs[rd] = (uintptr_t)dest; 4380 mstate->dtms_scratch_ptr += size; 4381 break; 4382 } 4383 4384 case DIF_SUBR_CLEANPATH: { 4385 char *dest = (char *)mstate->dtms_scratch_ptr, c; 4386 uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; 4387 uintptr_t src = tupregs[0].dttk_value; 4388 int i = 0, j = 0; 4389 4390 if (!dtrace_strcanload(src, size, mstate, vstate)) { 4391 regs[rd] = 0; 4392 break; 4393 } 4394 4395 if (!DTRACE_INSCRATCH(mstate, size)) { 4396 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4397 regs[rd] = 0; 4398 break; 4399 } 4400 4401 /* 4402 * Move forward, loading each character. 4403 */ 4404 do { 4405 c = dtrace_load8(src + i++); 4406next: 4407 if (j + 5 >= size) /* 5 = strlen("/..c\0") */ 4408 break; 4409 4410 if (c != '/') { 4411 dest[j++] = c; 4412 continue; 4413 } 4414 4415 c = dtrace_load8(src + i++); 4416 4417 if (c == '/') { 4418 /* 4419 * We have two slashes -- we can just advance 4420 * to the next character. 4421 */ 4422 goto next; 4423 } 4424 4425 if (c != '.') { 4426 /* 4427 * This is not "." and it's not ".." -- we can 4428 * just store the "/" and this character and 4429 * drive on. 4430 */ 4431 dest[j++] = '/'; 4432 dest[j++] = c; 4433 continue; 4434 } 4435 4436 c = dtrace_load8(src + i++); 4437 4438 if (c == '/') { 4439 /* 4440 * This is a "/./" component. We're not going 4441 * to store anything in the destination buffer; 4442 * we're just going to go to the next component. 4443 */ 4444 goto next; 4445 } 4446 4447 if (c != '.') { 4448 /* 4449 * This is not ".." -- we can just store the 4450 * "/." and this character and continue 4451 * processing. 4452 */ 4453 dest[j++] = '/'; 4454 dest[j++] = '.'; 4455 dest[j++] = c; 4456 continue; 4457 } 4458 4459 c = dtrace_load8(src + i++); 4460 4461 if (c != '/' && c != '\0') { 4462 /* 4463 * This is not ".." -- it's "..[mumble]". 4464 * We'll store the "/.." and this character 4465 * and continue processing. 4466 */ 4467 dest[j++] = '/'; 4468 dest[j++] = '.'; 4469 dest[j++] = '.'; 4470 dest[j++] = c; 4471 continue; 4472 } 4473 4474 /* 4475 * This is "/../" or "/..\0". We need to back up 4476 * our destination pointer until we find a "/". 4477 */ 4478 i--; 4479 while (j != 0 && dest[--j] != '/') 4480 continue; 4481 4482 if (c == '\0') 4483 dest[++j] = '/'; 4484 } while (c != '\0'); 4485 4486 dest[j] = '\0'; 4487 regs[rd] = (uintptr_t)dest; 4488 mstate->dtms_scratch_ptr += size; 4489 break; 4490 } 4491 4492 case DIF_SUBR_INET_NTOA: 4493 case DIF_SUBR_INET_NTOA6: 4494 case DIF_SUBR_INET_NTOP: { 4495 size_t size; 4496 int af, argi, i; 4497 char *base, *end; 4498 4499 if (subr == DIF_SUBR_INET_NTOP) { 4500 af = (int)tupregs[0].dttk_value; 4501 argi = 1; 4502 } else { 4503 af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6; 4504 argi = 0; 4505 } 4506 4507 if (af == AF_INET) { 4508 ipaddr_t ip4; 4509 uint8_t *ptr8, val; 4510 4511 /* 4512 * Safely load the IPv4 address. 4513 */ 4514 ip4 = dtrace_load32(tupregs[argi].dttk_value); 4515 4516 /* 4517 * Check an IPv4 string will fit in scratch. 4518 */ 4519 size = INET_ADDRSTRLEN; 4520 if (!DTRACE_INSCRATCH(mstate, size)) { 4521 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4522 regs[rd] = 0; 4523 break; 4524 } 4525 base = (char *)mstate->dtms_scratch_ptr; 4526 end = (char *)mstate->dtms_scratch_ptr + size - 1; 4527 4528 /* 4529 * Stringify as a dotted decimal quad. 4530 */ 4531 *end-- = '\0'; 4532 ptr8 = (uint8_t *)&ip4; 4533 for (i = 3; i >= 0; i--) { 4534 val = ptr8[i]; 4535 4536 if (val == 0) { 4537 *end-- = '0'; 4538 } else { 4539 for (; val; val /= 10) { 4540 *end-- = '0' + (val % 10); 4541 } 4542 } 4543 4544 if (i > 0) 4545 *end-- = '.'; 4546 } 4547 ASSERT(end + 1 >= base); 4548 4549 } else if (af == AF_INET6) { 4550 struct in6_addr ip6; 4551 int firstzero, tryzero, numzero, v6end; 4552 uint16_t val; 4553 const char digits[] = "0123456789abcdef"; 4554 4555 /* 4556 * Stringify using RFC 1884 convention 2 - 16 bit 4557 * hexadecimal values with a zero-run compression. 4558 * Lower case hexadecimal digits are used. 4559 * eg, fe80::214:4fff:fe0b:76c8. 4560 * The IPv4 embedded form is returned for inet_ntop, 4561 * just the IPv4 string is returned for inet_ntoa6. 4562 */ 4563 4564 /* 4565 * Safely load the IPv6 address. 4566 */ 4567 dtrace_bcopy( 4568 (void *)(uintptr_t)tupregs[argi].dttk_value, 4569 (void *)(uintptr_t)&ip6, sizeof (struct in6_addr)); 4570 4571 /* 4572 * Check an IPv6 string will fit in scratch. 4573 */ 4574 size = INET6_ADDRSTRLEN; 4575 if (!DTRACE_INSCRATCH(mstate, size)) { 4576 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 4577 regs[rd] = 0; 4578 break; 4579 } 4580 base = (char *)mstate->dtms_scratch_ptr; 4581 end = (char *)mstate->dtms_scratch_ptr + size - 1; 4582 *end-- = '\0'; 4583 4584 /* 4585 * Find the longest run of 16 bit zero values 4586 * for the single allowed zero compression - "::". 4587 */ 4588 firstzero = -1; 4589 tryzero = -1; 4590 numzero = 1; 4591 for (i = 0; i < sizeof (struct in6_addr); i++) { 4592#if defined(sun) 4593 if (ip6._S6_un._S6_u8[i] == 0 && 4594#else 4595 if (ip6.__u6_addr.__u6_addr8[i] == 0 && 4596#endif 4597 tryzero == -1 && i % 2 == 0) { 4598 tryzero = i; 4599 continue; 4600 } 4601 4602 if (tryzero != -1 && 4603#if defined(sun) 4604 (ip6._S6_un._S6_u8[i] != 0 || 4605#else 4606 (ip6.__u6_addr.__u6_addr8[i] != 0 || 4607#endif 4608 i == sizeof (struct in6_addr) - 1)) { 4609 4610 if (i - tryzero <= numzero) { 4611 tryzero = -1; 4612 continue; 4613 } 4614 4615 firstzero = tryzero; 4616 numzero = i - i % 2 - tryzero; 4617 tryzero = -1; 4618 4619#if defined(sun) 4620 if (ip6._S6_un._S6_u8[i] == 0 && 4621#else 4622 if (ip6.__u6_addr.__u6_addr8[i] == 0 && 4623#endif 4624 i == sizeof (struct in6_addr) - 1) 4625 numzero += 2; 4626 } 4627 } 4628 ASSERT(firstzero + numzero <= sizeof (struct in6_addr)); 4629 4630 /* 4631 * Check for an IPv4 embedded address. 4632 */ 4633 v6end = sizeof (struct in6_addr) - 2; 4634 if (IN6_IS_ADDR_V4MAPPED(&ip6) || 4635 IN6_IS_ADDR_V4COMPAT(&ip6)) { 4636 for (i = sizeof (struct in6_addr) - 1; 4637 i >= DTRACE_V4MAPPED_OFFSET; i--) { 4638 ASSERT(end >= base); 4639 4640#if defined(sun) 4641 val = ip6._S6_un._S6_u8[i]; 4642#else 4643 val = ip6.__u6_addr.__u6_addr8[i]; 4644#endif 4645 4646 if (val == 0) { 4647 *end-- = '0'; 4648 } else { 4649 for (; val; val /= 10) { 4650 *end-- = '0' + val % 10; 4651 } 4652 } 4653 4654 if (i > DTRACE_V4MAPPED_OFFSET) 4655 *end-- = '.'; 4656 } 4657 4658 if (subr == DIF_SUBR_INET_NTOA6) 4659 goto inetout; 4660 4661 /* 4662 * Set v6end to skip the IPv4 address that 4663 * we have already stringified. 4664 */ 4665 v6end = 10; 4666 } 4667 4668 /* 4669 * Build the IPv6 string by working through the 4670 * address in reverse. 4671 */ 4672 for (i = v6end; i >= 0; i -= 2) { 4673 ASSERT(end >= base); 4674 4675 if (i == firstzero + numzero - 2) { 4676 *end-- = ':'; 4677 *end-- = ':'; 4678 i -= numzero - 2; 4679 continue; 4680 } 4681 4682 if (i < 14 && i != firstzero - 2) 4683 *end-- = ':'; 4684 4685#if defined(sun) 4686 val = (ip6._S6_un._S6_u8[i] << 8) + 4687 ip6._S6_un._S6_u8[i + 1]; 4688#else 4689 val = (ip6.__u6_addr.__u6_addr8[i] << 8) + 4690 ip6.__u6_addr.__u6_addr8[i + 1]; 4691#endif 4692 4693 if (val == 0) { 4694 *end-- = '0'; 4695 } else { 4696 for (; val; val /= 16) { 4697 *end-- = digits[val % 16]; 4698 } 4699 } 4700 } 4701 ASSERT(end + 1 >= base); 4702 4703 } else { 4704 /* 4705 * The user didn't use AH_INET or AH_INET6. 4706 */ 4707 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); 4708 regs[rd] = 0; 4709 break; 4710 } 4711 4712inetout: regs[rd] = (uintptr_t)end + 1; 4713 mstate->dtms_scratch_ptr += size; 4714 break; 4715 } 4716 4717 case DIF_SUBR_MEMREF: { 4718 uintptr_t size = 2 * sizeof(uintptr_t); 4719 uintptr_t *memref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t)); 4720 size_t scratch_size = ((uintptr_t) memref - mstate->dtms_scratch_ptr) + size; 4721 4722 /* address and length */ 4723 memref[0] = tupregs[0].dttk_value; 4724 memref[1] = tupregs[1].dttk_value; 4725 4726 regs[rd] = (uintptr_t) memref; 4727 mstate->dtms_scratch_ptr += scratch_size; 4728 break; 4729 } 4730 4731 case DIF_SUBR_TYPEREF: { 4732 uintptr_t size = 4 * sizeof(uintptr_t); 4733 uintptr_t *typeref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t)); 4734 size_t scratch_size = ((uintptr_t) typeref - mstate->dtms_scratch_ptr) + size; 4735 4736 /* address, num_elements, type_str, type_len */ 4737 typeref[0] = tupregs[0].dttk_value; 4738 typeref[1] = tupregs[1].dttk_value; 4739 typeref[2] = tupregs[2].dttk_value; 4740 typeref[3] = tupregs[3].dttk_value; 4741 4742 regs[rd] = (uintptr_t) typeref; 4743 mstate->dtms_scratch_ptr += scratch_size; 4744 break; 4745 } 4746 } 4747} 4748 4749/* 4750 * Emulate the execution of DTrace IR instructions specified by the given 4751 * DIF object. This function is deliberately void of assertions as all of 4752 * the necessary checks are handled by a call to dtrace_difo_validate(). 4753 */ 4754static uint64_t 4755dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate, 4756 dtrace_vstate_t *vstate, dtrace_state_t *state) 4757{ 4758 const dif_instr_t *text = difo->dtdo_buf; 4759 const uint_t textlen = difo->dtdo_len; 4760 const char *strtab = difo->dtdo_strtab; 4761 const uint64_t *inttab = difo->dtdo_inttab; 4762 4763 uint64_t rval = 0; 4764 dtrace_statvar_t *svar; 4765 dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; 4766 dtrace_difv_t *v; 4767 volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; 4768 volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; 4769 4770 dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */ 4771 uint64_t regs[DIF_DIR_NREGS]; 4772 uint64_t *tmp; 4773 4774 uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0; 4775 int64_t cc_r; 4776 uint_t pc = 0, id, opc = 0; 4777 uint8_t ttop = 0; 4778 dif_instr_t instr; 4779 uint_t r1, r2, rd; 4780 4781 /* 4782 * We stash the current DIF object into the machine state: we need it 4783 * for subsequent access checking. 4784 */ 4785 mstate->dtms_difo = difo; 4786 4787 regs[DIF_REG_R0] = 0; /* %r0 is fixed at zero */ 4788 4789 while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) { 4790 opc = pc; 4791 4792 instr = text[pc++]; 4793 r1 = DIF_INSTR_R1(instr); 4794 r2 = DIF_INSTR_R2(instr); 4795 rd = DIF_INSTR_RD(instr); 4796 4797 switch (DIF_INSTR_OP(instr)) { 4798 case DIF_OP_OR: 4799 regs[rd] = regs[r1] | regs[r2]; 4800 break; 4801 case DIF_OP_XOR: 4802 regs[rd] = regs[r1] ^ regs[r2]; 4803 break; 4804 case DIF_OP_AND: 4805 regs[rd] = regs[r1] & regs[r2]; 4806 break; 4807 case DIF_OP_SLL: 4808 regs[rd] = regs[r1] << regs[r2]; 4809 break; 4810 case DIF_OP_SRL: 4811 regs[rd] = regs[r1] >> regs[r2]; 4812 break; 4813 case DIF_OP_SUB: 4814 regs[rd] = regs[r1] - regs[r2]; 4815 break; 4816 case DIF_OP_ADD: 4817 regs[rd] = regs[r1] + regs[r2]; 4818 break; 4819 case DIF_OP_MUL: 4820 regs[rd] = regs[r1] * regs[r2]; 4821 break; 4822 case DIF_OP_SDIV: 4823 if (regs[r2] == 0) { 4824 regs[rd] = 0; 4825 *flags |= CPU_DTRACE_DIVZERO; 4826 } else { 4827 regs[rd] = (int64_t)regs[r1] / 4828 (int64_t)regs[r2]; 4829 } 4830 break; 4831 4832 case DIF_OP_UDIV: 4833 if (regs[r2] == 0) { 4834 regs[rd] = 0; 4835 *flags |= CPU_DTRACE_DIVZERO; 4836 } else { 4837 regs[rd] = regs[r1] / regs[r2]; 4838 } 4839 break; 4840 4841 case DIF_OP_SREM: 4842 if (regs[r2] == 0) { 4843 regs[rd] = 0; 4844 *flags |= CPU_DTRACE_DIVZERO; 4845 } else { 4846 regs[rd] = (int64_t)regs[r1] % 4847 (int64_t)regs[r2]; 4848 } 4849 break; 4850 4851 case DIF_OP_UREM: 4852 if (regs[r2] == 0) { 4853 regs[rd] = 0; 4854 *flags |= CPU_DTRACE_DIVZERO; 4855 } else { 4856 regs[rd] = regs[r1] % regs[r2]; 4857 } 4858 break; 4859 4860 case DIF_OP_NOT: 4861 regs[rd] = ~regs[r1]; 4862 break; 4863 case DIF_OP_MOV: 4864 regs[rd] = regs[r1]; 4865 break; 4866 case DIF_OP_CMP: 4867 cc_r = regs[r1] - regs[r2]; 4868 cc_n = cc_r < 0; 4869 cc_z = cc_r == 0; 4870 cc_v = 0; 4871 cc_c = regs[r1] < regs[r2]; 4872 break; 4873 case DIF_OP_TST: 4874 cc_n = cc_v = cc_c = 0; 4875 cc_z = regs[r1] == 0; 4876 break; 4877 case DIF_OP_BA: 4878 pc = DIF_INSTR_LABEL(instr); 4879 break; 4880 case DIF_OP_BE: 4881 if (cc_z) 4882 pc = DIF_INSTR_LABEL(instr); 4883 break; 4884 case DIF_OP_BNE: 4885 if (cc_z == 0) 4886 pc = DIF_INSTR_LABEL(instr); 4887 break; 4888 case DIF_OP_BG: 4889 if ((cc_z | (cc_n ^ cc_v)) == 0) 4890 pc = DIF_INSTR_LABEL(instr); 4891 break; 4892 case DIF_OP_BGU: 4893 if ((cc_c | cc_z) == 0) 4894 pc = DIF_INSTR_LABEL(instr); 4895 break; 4896 case DIF_OP_BGE: 4897 if ((cc_n ^ cc_v) == 0) 4898 pc = DIF_INSTR_LABEL(instr); 4899 break; 4900 case DIF_OP_BGEU: 4901 if (cc_c == 0) 4902 pc = DIF_INSTR_LABEL(instr); 4903 break; 4904 case DIF_OP_BL: 4905 if (cc_n ^ cc_v) 4906 pc = DIF_INSTR_LABEL(instr); 4907 break; 4908 case DIF_OP_BLU: 4909 if (cc_c) 4910 pc = DIF_INSTR_LABEL(instr); 4911 break; 4912 case DIF_OP_BLE: 4913 if (cc_z | (cc_n ^ cc_v)) 4914 pc = DIF_INSTR_LABEL(instr); 4915 break; 4916 case DIF_OP_BLEU: 4917 if (cc_c | cc_z) 4918 pc = DIF_INSTR_LABEL(instr); 4919 break; 4920 case DIF_OP_RLDSB: 4921 if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { 4922 *flags |= CPU_DTRACE_KPRIV; 4923 *illval = regs[r1]; 4924 break; 4925 } 4926 /*FALLTHROUGH*/ 4927 case DIF_OP_LDSB: 4928 regs[rd] = (int8_t)dtrace_load8(regs[r1]); 4929 break; 4930 case DIF_OP_RLDSH: 4931 if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { 4932 *flags |= CPU_DTRACE_KPRIV; 4933 *illval = regs[r1]; 4934 break; 4935 } 4936 /*FALLTHROUGH*/ 4937 case DIF_OP_LDSH: 4938 regs[rd] = (int16_t)dtrace_load16(regs[r1]); 4939 break; 4940 case DIF_OP_RLDSW: 4941 if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { 4942 *flags |= CPU_DTRACE_KPRIV; 4943 *illval = regs[r1]; 4944 break; 4945 } 4946 /*FALLTHROUGH*/ 4947 case DIF_OP_LDSW: 4948 regs[rd] = (int32_t)dtrace_load32(regs[r1]); 4949 break; 4950 case DIF_OP_RLDUB: 4951 if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { 4952 *flags |= CPU_DTRACE_KPRIV; 4953 *illval = regs[r1]; 4954 break; 4955 } 4956 /*FALLTHROUGH*/ 4957 case DIF_OP_LDUB: 4958 regs[rd] = dtrace_load8(regs[r1]); 4959 break; 4960 case DIF_OP_RLDUH: 4961 if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { 4962 *flags |= CPU_DTRACE_KPRIV; 4963 *illval = regs[r1]; 4964 break; 4965 } 4966 /*FALLTHROUGH*/ 4967 case DIF_OP_LDUH: 4968 regs[rd] = dtrace_load16(regs[r1]); 4969 break; 4970 case DIF_OP_RLDUW: 4971 if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { 4972 *flags |= CPU_DTRACE_KPRIV; 4973 *illval = regs[r1]; 4974 break; 4975 } 4976 /*FALLTHROUGH*/ 4977 case DIF_OP_LDUW: 4978 regs[rd] = dtrace_load32(regs[r1]); 4979 break; 4980 case DIF_OP_RLDX: 4981 if (!dtrace_canstore(regs[r1], 8, mstate, vstate)) { 4982 *flags |= CPU_DTRACE_KPRIV; 4983 *illval = regs[r1]; 4984 break; 4985 } 4986 /*FALLTHROUGH*/ 4987 case DIF_OP_LDX: 4988 regs[rd] = dtrace_load64(regs[r1]); 4989 break; 4990 case DIF_OP_ULDSB: 4991 regs[rd] = (int8_t) 4992 dtrace_fuword8((void *)(uintptr_t)regs[r1]); 4993 break; 4994 case DIF_OP_ULDSH: 4995 regs[rd] = (int16_t) 4996 dtrace_fuword16((void *)(uintptr_t)regs[r1]); 4997 break; 4998 case DIF_OP_ULDSW: 4999 regs[rd] = (int32_t) 5000 dtrace_fuword32((void *)(uintptr_t)regs[r1]); 5001 break; 5002 case DIF_OP_ULDUB: 5003 regs[rd] = 5004 dtrace_fuword8((void *)(uintptr_t)regs[r1]); 5005 break; 5006 case DIF_OP_ULDUH: 5007 regs[rd] = 5008 dtrace_fuword16((void *)(uintptr_t)regs[r1]); 5009 break; 5010 case DIF_OP_ULDUW: 5011 regs[rd] = 5012 dtrace_fuword32((void *)(uintptr_t)regs[r1]); 5013 break; 5014 case DIF_OP_ULDX: 5015 regs[rd] = 5016 dtrace_fuword64((void *)(uintptr_t)regs[r1]); 5017 break; 5018 case DIF_OP_RET: 5019 rval = regs[rd]; 5020 pc = textlen; 5021 break; 5022 case DIF_OP_NOP: 5023 break; 5024 case DIF_OP_SETX: 5025 regs[rd] = inttab[DIF_INSTR_INTEGER(instr)]; 5026 break; 5027 case DIF_OP_SETS: 5028 regs[rd] = (uint64_t)(uintptr_t) 5029 (strtab + DIF_INSTR_STRING(instr)); 5030 break; 5031 case DIF_OP_SCMP: { 5032 size_t sz = state->dts_options[DTRACEOPT_STRSIZE]; 5033 uintptr_t s1 = regs[r1]; 5034 uintptr_t s2 = regs[r2]; 5035 5036 if (s1 != 0 && 5037 !dtrace_strcanload(s1, sz, mstate, vstate)) 5038 break; 5039 if (s2 != 0 && 5040 !dtrace_strcanload(s2, sz, mstate, vstate)) 5041 break; 5042 5043 cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz); 5044 5045 cc_n = cc_r < 0; 5046 cc_z = cc_r == 0; 5047 cc_v = cc_c = 0; 5048 break; 5049 } 5050 case DIF_OP_LDGA: 5051 regs[rd] = dtrace_dif_variable(mstate, state, 5052 r1, regs[r2]); 5053 break; 5054 case DIF_OP_LDGS: 5055 id = DIF_INSTR_VAR(instr); 5056 5057 if (id >= DIF_VAR_OTHER_UBASE) { 5058 uintptr_t a; 5059 5060 id -= DIF_VAR_OTHER_UBASE; 5061 svar = vstate->dtvs_globals[id]; 5062 ASSERT(svar != NULL); 5063 v = &svar->dtsv_var; 5064 5065 if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) { 5066 regs[rd] = svar->dtsv_data; 5067 break; 5068 } 5069 5070 a = (uintptr_t)svar->dtsv_data; 5071 5072 if (*(uint8_t *)a == UINT8_MAX) { 5073 /* 5074 * If the 0th byte is set to UINT8_MAX 5075 * then this is to be treated as a 5076 * reference to a NULL variable. 5077 */ 5078 regs[rd] = 0; 5079 } else { 5080 regs[rd] = a + sizeof (uint64_t); 5081 } 5082 5083 break; 5084 } 5085 5086 regs[rd] = dtrace_dif_variable(mstate, state, id, 0); 5087 break; 5088 5089 case DIF_OP_STGS: 5090 id = DIF_INSTR_VAR(instr); 5091 5092 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5093 id -= DIF_VAR_OTHER_UBASE; 5094 5095 svar = vstate->dtvs_globals[id]; 5096 ASSERT(svar != NULL); 5097 v = &svar->dtsv_var; 5098 5099 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5100 uintptr_t a = (uintptr_t)svar->dtsv_data; 5101 5102 ASSERT(a != 0); 5103 ASSERT(svar->dtsv_size != 0); 5104 5105 if (regs[rd] == 0) { 5106 *(uint8_t *)a = UINT8_MAX; 5107 break; 5108 } else { 5109 *(uint8_t *)a = 0; 5110 a += sizeof (uint64_t); 5111 } 5112 if (!dtrace_vcanload( 5113 (void *)(uintptr_t)regs[rd], &v->dtdv_type, 5114 mstate, vstate)) 5115 break; 5116 5117 dtrace_vcopy((void *)(uintptr_t)regs[rd], 5118 (void *)a, &v->dtdv_type); 5119 break; 5120 } 5121 5122 svar->dtsv_data = regs[rd]; 5123 break; 5124 5125 case DIF_OP_LDTA: 5126 /* 5127 * There are no DTrace built-in thread-local arrays at 5128 * present. This opcode is saved for future work. 5129 */ 5130 *flags |= CPU_DTRACE_ILLOP; 5131 regs[rd] = 0; 5132 break; 5133 5134 case DIF_OP_LDLS: 5135 id = DIF_INSTR_VAR(instr); 5136 5137 if (id < DIF_VAR_OTHER_UBASE) { 5138 /* 5139 * For now, this has no meaning. 5140 */ 5141 regs[rd] = 0; 5142 break; 5143 } 5144 5145 id -= DIF_VAR_OTHER_UBASE; 5146 5147 ASSERT(id < vstate->dtvs_nlocals); 5148 ASSERT(vstate->dtvs_locals != NULL); 5149 5150 svar = vstate->dtvs_locals[id]; 5151 ASSERT(svar != NULL); 5152 v = &svar->dtsv_var; 5153 5154 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5155 uintptr_t a = (uintptr_t)svar->dtsv_data; 5156 size_t sz = v->dtdv_type.dtdt_size; 5157 5158 sz += sizeof (uint64_t); 5159 ASSERT(svar->dtsv_size == NCPU * sz); 5160 a += curcpu * sz; 5161 5162 if (*(uint8_t *)a == UINT8_MAX) { 5163 /* 5164 * If the 0th byte is set to UINT8_MAX 5165 * then this is to be treated as a 5166 * reference to a NULL variable. 5167 */ 5168 regs[rd] = 0; 5169 } else { 5170 regs[rd] = a + sizeof (uint64_t); 5171 } 5172 5173 break; 5174 } 5175 5176 ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t)); 5177 tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; 5178 regs[rd] = tmp[curcpu]; 5179 break; 5180 5181 case DIF_OP_STLS: 5182 id = DIF_INSTR_VAR(instr); 5183 5184 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5185 id -= DIF_VAR_OTHER_UBASE; 5186 ASSERT(id < vstate->dtvs_nlocals); 5187 5188 ASSERT(vstate->dtvs_locals != NULL); 5189 svar = vstate->dtvs_locals[id]; 5190 ASSERT(svar != NULL); 5191 v = &svar->dtsv_var; 5192 5193 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5194 uintptr_t a = (uintptr_t)svar->dtsv_data; 5195 size_t sz = v->dtdv_type.dtdt_size; 5196 5197 sz += sizeof (uint64_t); 5198 ASSERT(svar->dtsv_size == NCPU * sz); 5199 a += curcpu * sz; 5200 5201 if (regs[rd] == 0) { 5202 *(uint8_t *)a = UINT8_MAX; 5203 break; 5204 } else { 5205 *(uint8_t *)a = 0; 5206 a += sizeof (uint64_t); 5207 } 5208 5209 if (!dtrace_vcanload( 5210 (void *)(uintptr_t)regs[rd], &v->dtdv_type, 5211 mstate, vstate)) 5212 break; 5213 5214 dtrace_vcopy((void *)(uintptr_t)regs[rd], 5215 (void *)a, &v->dtdv_type); 5216 break; 5217 } 5218 5219 ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t)); 5220 tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; 5221 tmp[curcpu] = regs[rd]; 5222 break; 5223 5224 case DIF_OP_LDTS: { 5225 dtrace_dynvar_t *dvar; 5226 dtrace_key_t *key; 5227 5228 id = DIF_INSTR_VAR(instr); 5229 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5230 id -= DIF_VAR_OTHER_UBASE; 5231 v = &vstate->dtvs_tlocals[id]; 5232 5233 key = &tupregs[DIF_DTR_NREGS]; 5234 key[0].dttk_value = (uint64_t)id; 5235 key[0].dttk_size = 0; 5236 DTRACE_TLS_THRKEY(key[1].dttk_value); 5237 key[1].dttk_size = 0; 5238 5239 dvar = dtrace_dynvar(dstate, 2, key, 5240 sizeof (uint64_t), DTRACE_DYNVAR_NOALLOC, 5241 mstate, vstate); 5242 5243 if (dvar == NULL) { 5244 regs[rd] = 0; 5245 break; 5246 } 5247 5248 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5249 regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; 5250 } else { 5251 regs[rd] = *((uint64_t *)dvar->dtdv_data); 5252 } 5253 5254 break; 5255 } 5256 5257 case DIF_OP_STTS: { 5258 dtrace_dynvar_t *dvar; 5259 dtrace_key_t *key; 5260 5261 id = DIF_INSTR_VAR(instr); 5262 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5263 id -= DIF_VAR_OTHER_UBASE; 5264 5265 key = &tupregs[DIF_DTR_NREGS]; 5266 key[0].dttk_value = (uint64_t)id; 5267 key[0].dttk_size = 0; 5268 DTRACE_TLS_THRKEY(key[1].dttk_value); 5269 key[1].dttk_size = 0; 5270 v = &vstate->dtvs_tlocals[id]; 5271 5272 dvar = dtrace_dynvar(dstate, 2, key, 5273 v->dtdv_type.dtdt_size > sizeof (uint64_t) ? 5274 v->dtdv_type.dtdt_size : sizeof (uint64_t), 5275 regs[rd] ? DTRACE_DYNVAR_ALLOC : 5276 DTRACE_DYNVAR_DEALLOC, mstate, vstate); 5277 5278 /* 5279 * Given that we're storing to thread-local data, 5280 * we need to flush our predicate cache. 5281 */ 5282 curthread->t_predcache = 0; 5283 5284 if (dvar == NULL) 5285 break; 5286 5287 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5288 if (!dtrace_vcanload( 5289 (void *)(uintptr_t)regs[rd], 5290 &v->dtdv_type, mstate, vstate)) 5291 break; 5292 5293 dtrace_vcopy((void *)(uintptr_t)regs[rd], 5294 dvar->dtdv_data, &v->dtdv_type); 5295 } else { 5296 *((uint64_t *)dvar->dtdv_data) = regs[rd]; 5297 } 5298 5299 break; 5300 } 5301 5302 case DIF_OP_SRA: 5303 regs[rd] = (int64_t)regs[r1] >> regs[r2]; 5304 break; 5305 5306 case DIF_OP_CALL: 5307 dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd, 5308 regs, tupregs, ttop, mstate, state); 5309 break; 5310 5311 case DIF_OP_PUSHTR: 5312 if (ttop == DIF_DTR_NREGS) { 5313 *flags |= CPU_DTRACE_TUPOFLOW; 5314 break; 5315 } 5316 5317 if (r1 == DIF_TYPE_STRING) { 5318 /* 5319 * If this is a string type and the size is 0, 5320 * we'll use the system-wide default string 5321 * size. Note that we are _not_ looking at 5322 * the value of the DTRACEOPT_STRSIZE option; 5323 * had this been set, we would expect to have 5324 * a non-zero size value in the "pushtr". 5325 */ 5326 tupregs[ttop].dttk_size = 5327 dtrace_strlen((char *)(uintptr_t)regs[rd], 5328 regs[r2] ? regs[r2] : 5329 dtrace_strsize_default) + 1; 5330 } else { 5331 tupregs[ttop].dttk_size = regs[r2]; 5332 } 5333 5334 tupregs[ttop++].dttk_value = regs[rd]; 5335 break; 5336 5337 case DIF_OP_PUSHTV: 5338 if (ttop == DIF_DTR_NREGS) { 5339 *flags |= CPU_DTRACE_TUPOFLOW; 5340 break; 5341 } 5342 5343 tupregs[ttop].dttk_value = regs[rd]; 5344 tupregs[ttop++].dttk_size = 0; 5345 break; 5346 5347 case DIF_OP_POPTS: 5348 if (ttop != 0) 5349 ttop--; 5350 break; 5351 5352 case DIF_OP_FLUSHTS: 5353 ttop = 0; 5354 break; 5355 5356 case DIF_OP_LDGAA: 5357 case DIF_OP_LDTAA: { 5358 dtrace_dynvar_t *dvar; 5359 dtrace_key_t *key = tupregs; 5360 uint_t nkeys = ttop; 5361 5362 id = DIF_INSTR_VAR(instr); 5363 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5364 id -= DIF_VAR_OTHER_UBASE; 5365 5366 key[nkeys].dttk_value = (uint64_t)id; 5367 key[nkeys++].dttk_size = 0; 5368 5369 if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) { 5370 DTRACE_TLS_THRKEY(key[nkeys].dttk_value); 5371 key[nkeys++].dttk_size = 0; 5372 v = &vstate->dtvs_tlocals[id]; 5373 } else { 5374 v = &vstate->dtvs_globals[id]->dtsv_var; 5375 } 5376 5377 dvar = dtrace_dynvar(dstate, nkeys, key, 5378 v->dtdv_type.dtdt_size > sizeof (uint64_t) ? 5379 v->dtdv_type.dtdt_size : sizeof (uint64_t), 5380 DTRACE_DYNVAR_NOALLOC, mstate, vstate); 5381 5382 if (dvar == NULL) { 5383 regs[rd] = 0; 5384 break; 5385 } 5386 5387 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5388 regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; 5389 } else { 5390 regs[rd] = *((uint64_t *)dvar->dtdv_data); 5391 } 5392 5393 break; 5394 } 5395 5396 case DIF_OP_STGAA: 5397 case DIF_OP_STTAA: { 5398 dtrace_dynvar_t *dvar; 5399 dtrace_key_t *key = tupregs; 5400 uint_t nkeys = ttop; 5401 5402 id = DIF_INSTR_VAR(instr); 5403 ASSERT(id >= DIF_VAR_OTHER_UBASE); 5404 id -= DIF_VAR_OTHER_UBASE; 5405 5406 key[nkeys].dttk_value = (uint64_t)id; 5407 key[nkeys++].dttk_size = 0; 5408 5409 if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) { 5410 DTRACE_TLS_THRKEY(key[nkeys].dttk_value); 5411 key[nkeys++].dttk_size = 0; 5412 v = &vstate->dtvs_tlocals[id]; 5413 } else { 5414 v = &vstate->dtvs_globals[id]->dtsv_var; 5415 } 5416 5417 dvar = dtrace_dynvar(dstate, nkeys, key, 5418 v->dtdv_type.dtdt_size > sizeof (uint64_t) ? 5419 v->dtdv_type.dtdt_size : sizeof (uint64_t), 5420 regs[rd] ? DTRACE_DYNVAR_ALLOC : 5421 DTRACE_DYNVAR_DEALLOC, mstate, vstate); 5422 5423 if (dvar == NULL) 5424 break; 5425 5426 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { 5427 if (!dtrace_vcanload( 5428 (void *)(uintptr_t)regs[rd], &v->dtdv_type, 5429 mstate, vstate)) 5430 break; 5431 5432 dtrace_vcopy((void *)(uintptr_t)regs[rd], 5433 dvar->dtdv_data, &v->dtdv_type); 5434 } else { 5435 *((uint64_t *)dvar->dtdv_data) = regs[rd]; 5436 } 5437 5438 break; 5439 } 5440 5441 case DIF_OP_ALLOCS: { 5442 uintptr_t ptr = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); 5443 size_t size = ptr - mstate->dtms_scratch_ptr + regs[r1]; 5444 5445 /* 5446 * Rounding up the user allocation size could have 5447 * overflowed large, bogus allocations (like -1ULL) to 5448 * 0. 5449 */ 5450 if (size < regs[r1] || 5451 !DTRACE_INSCRATCH(mstate, size)) { 5452 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 5453 regs[rd] = 0; 5454 break; 5455 } 5456 5457 dtrace_bzero((void *) mstate->dtms_scratch_ptr, size); 5458 mstate->dtms_scratch_ptr += size; 5459 regs[rd] = ptr; 5460 break; 5461 } 5462 5463 case DIF_OP_COPYS: 5464 if (!dtrace_canstore(regs[rd], regs[r2], 5465 mstate, vstate)) { 5466 *flags |= CPU_DTRACE_BADADDR; 5467 *illval = regs[rd]; 5468 break; 5469 } 5470 5471 if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate)) 5472 break; 5473 5474 dtrace_bcopy((void *)(uintptr_t)regs[r1], 5475 (void *)(uintptr_t)regs[rd], (size_t)regs[r2]); 5476 break; 5477 5478 case DIF_OP_STB: 5479 if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) { 5480 *flags |= CPU_DTRACE_BADADDR; 5481 *illval = regs[rd]; 5482 break; 5483 } 5484 *((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1]; 5485 break; 5486 5487 case DIF_OP_STH: 5488 if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) { 5489 *flags |= CPU_DTRACE_BADADDR; 5490 *illval = regs[rd]; 5491 break; 5492 } 5493 if (regs[rd] & 1) { 5494 *flags |= CPU_DTRACE_BADALIGN; 5495 *illval = regs[rd]; 5496 break; 5497 } 5498 *((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1]; 5499 break; 5500 5501 case DIF_OP_STW: 5502 if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) { 5503 *flags |= CPU_DTRACE_BADADDR; 5504 *illval = regs[rd]; 5505 break; 5506 } 5507 if (regs[rd] & 3) { 5508 *flags |= CPU_DTRACE_BADALIGN; 5509 *illval = regs[rd]; 5510 break; 5511 } 5512 *((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1]; 5513 break; 5514 5515 case DIF_OP_STX: 5516 if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) { 5517 *flags |= CPU_DTRACE_BADADDR; 5518 *illval = regs[rd]; 5519 break; 5520 } 5521 if (regs[rd] & 7) { 5522 *flags |= CPU_DTRACE_BADALIGN; 5523 *illval = regs[rd]; 5524 break; 5525 } 5526 *((uint64_t *)(uintptr_t)regs[rd]) = regs[r1]; 5527 break; 5528 } 5529 } 5530 5531 if (!(*flags & CPU_DTRACE_FAULT)) 5532 return (rval); 5533 5534 mstate->dtms_fltoffs = opc * sizeof (dif_instr_t); 5535 mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS; 5536 5537 return (0); 5538} 5539 5540static void 5541dtrace_action_breakpoint(dtrace_ecb_t *ecb) 5542{ 5543 dtrace_probe_t *probe = ecb->dte_probe; 5544 dtrace_provider_t *prov = probe->dtpr_provider; 5545 char c[DTRACE_FULLNAMELEN + 80], *str; 5546 char *msg = "dtrace: breakpoint action at probe "; 5547 char *ecbmsg = " (ecb "; 5548 uintptr_t mask = (0xf << (sizeof (uintptr_t) * NBBY / 4)); 5549 uintptr_t val = (uintptr_t)ecb; 5550 int shift = (sizeof (uintptr_t) * NBBY) - 4, i = 0; 5551 5552 if (dtrace_destructive_disallow) 5553 return; 5554 5555 /* 5556 * It's impossible to be taking action on the NULL probe. 5557 */ 5558 ASSERT(probe != NULL); 5559 5560 /* 5561 * This is a poor man's (destitute man's?) sprintf(): we want to 5562 * print the provider name, module name, function name and name of 5563 * the probe, along with the hex address of the ECB with the breakpoint 5564 * action -- all of which we must place in the character buffer by 5565 * hand. 5566 */ 5567 while (*msg != '\0') 5568 c[i++] = *msg++; 5569 5570 for (str = prov->dtpv_name; *str != '\0'; str++) 5571 c[i++] = *str; 5572 c[i++] = ':'; 5573 5574 for (str = probe->dtpr_mod; *str != '\0'; str++) 5575 c[i++] = *str; 5576 c[i++] = ':'; 5577 5578 for (str = probe->dtpr_func; *str != '\0'; str++) 5579 c[i++] = *str; 5580 c[i++] = ':'; 5581 5582 for (str = probe->dtpr_name; *str != '\0'; str++) 5583 c[i++] = *str; 5584 5585 while (*ecbmsg != '\0') 5586 c[i++] = *ecbmsg++; 5587 5588 while (shift >= 0) { 5589 mask = (uintptr_t)0xf << shift; 5590 5591 if (val >= ((uintptr_t)1 << shift)) 5592 c[i++] = "0123456789abcdef"[(val & mask) >> shift]; 5593 shift -= 4; 5594 } 5595 5596 c[i++] = ')'; 5597 c[i] = '\0'; 5598 5599#if defined(sun) 5600 debug_enter(c); 5601#else 5602 kdb_enter(KDB_WHY_DTRACE, "breakpoint action"); 5603#endif 5604} 5605 5606static void 5607dtrace_action_panic(dtrace_ecb_t *ecb) 5608{ 5609 dtrace_probe_t *probe = ecb->dte_probe; 5610 5611 /* 5612 * It's impossible to be taking action on the NULL probe. 5613 */ 5614 ASSERT(probe != NULL); 5615 5616 if (dtrace_destructive_disallow) 5617 return; 5618 5619 if (dtrace_panicked != NULL) 5620 return; 5621 5622 if (dtrace_casptr(&dtrace_panicked, NULL, curthread) != NULL) 5623 return; 5624 5625 /* 5626 * We won the right to panic. (We want to be sure that only one 5627 * thread calls panic() from dtrace_probe(), and that panic() is 5628 * called exactly once.) 5629 */ 5630 dtrace_panic("dtrace: panic action at probe %s:%s:%s:%s (ecb %p)", 5631 probe->dtpr_provider->dtpv_name, probe->dtpr_mod, 5632 probe->dtpr_func, probe->dtpr_name, (void *)ecb); 5633} 5634 5635static void 5636dtrace_action_raise(uint64_t sig) 5637{ 5638 if (dtrace_destructive_disallow) 5639 return; 5640 5641 if (sig >= NSIG) { 5642 DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); 5643 return; 5644 } 5645 5646#if defined(sun) 5647 /* 5648 * raise() has a queue depth of 1 -- we ignore all subsequent 5649 * invocations of the raise() action. 5650 */ 5651 if (curthread->t_dtrace_sig == 0) 5652 curthread->t_dtrace_sig = (uint8_t)sig; 5653 5654 curthread->t_sig_check = 1; 5655 aston(curthread); 5656#else 5657 struct proc *p = curproc; 5658 PROC_LOCK(p); 5659 psignal(p, sig); 5660 PROC_UNLOCK(p); 5661#endif 5662} 5663 5664static void 5665dtrace_action_stop(void) 5666{ 5667 if (dtrace_destructive_disallow) 5668 return; 5669 5670#if defined(sun) 5671 if (!curthread->t_dtrace_stop) { 5672 curthread->t_dtrace_stop = 1; 5673 curthread->t_sig_check = 1; 5674 aston(curthread); 5675 } 5676#else 5677 struct proc *p = curproc; 5678 PROC_LOCK(p); 5679 psignal(p, SIGSTOP); 5680 PROC_UNLOCK(p); 5681#endif 5682} 5683 5684static void 5685dtrace_action_chill(dtrace_mstate_t *mstate, hrtime_t val) 5686{ 5687 hrtime_t now; 5688 volatile uint16_t *flags; 5689#if defined(sun) 5690 cpu_t *cpu = CPU; 5691#else 5692 cpu_t *cpu = &solaris_cpu[curcpu]; 5693#endif 5694 5695 if (dtrace_destructive_disallow) 5696 return; 5697 5698 flags = (volatile uint16_t *)&cpu_core[cpu->cpu_id].cpuc_dtrace_flags; 5699 5700 now = dtrace_gethrtime(); 5701 5702 if (now - cpu->cpu_dtrace_chillmark > dtrace_chill_interval) { 5703 /* 5704 * We need to advance the mark to the current time. 5705 */ 5706 cpu->cpu_dtrace_chillmark = now; 5707 cpu->cpu_dtrace_chilled = 0; 5708 } 5709 5710 /* 5711 * Now check to see if the requested chill time would take us over 5712 * the maximum amount of time allowed in the chill interval. (Or 5713 * worse, if the calculation itself induces overflow.) 5714 */ 5715 if (cpu->cpu_dtrace_chilled + val > dtrace_chill_max || 5716 cpu->cpu_dtrace_chilled + val < cpu->cpu_dtrace_chilled) { 5717 *flags |= CPU_DTRACE_ILLOP; 5718 return; 5719 } 5720 5721 while (dtrace_gethrtime() - now < val) 5722 continue; 5723 5724 /* 5725 * Normally, we assure that the value of the variable "timestamp" does 5726 * not change within an ECB. The presence of chill() represents an 5727 * exception to this rule, however. 5728 */ 5729 mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP; 5730 cpu->cpu_dtrace_chilled += val; 5731} 5732 5733#if defined(sun) 5734static void 5735dtrace_action_ustack(dtrace_mstate_t *mstate, dtrace_state_t *state, 5736 uint64_t *buf, uint64_t arg) 5737{ 5738 int nframes = DTRACE_USTACK_NFRAMES(arg); 5739 int strsize = DTRACE_USTACK_STRSIZE(arg); 5740 uint64_t *pcs = &buf[1], *fps; 5741 char *str = (char *)&pcs[nframes]; 5742 int size, offs = 0, i, j; 5743 uintptr_t old = mstate->dtms_scratch_ptr, saved; 5744 uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; 5745 char *sym; 5746 5747 /* 5748 * Should be taking a faster path if string space has not been 5749 * allocated. 5750 */ 5751 ASSERT(strsize != 0); 5752 5753 /* 5754 * We will first allocate some temporary space for the frame pointers. 5755 */ 5756 fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8); 5757 size = (uintptr_t)fps - mstate->dtms_scratch_ptr + 5758 (nframes * sizeof (uint64_t)); 5759 5760 if (!DTRACE_INSCRATCH(mstate, size)) { 5761 /* 5762 * Not enough room for our frame pointers -- need to indicate 5763 * that we ran out of scratch space. 5764 */ 5765 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); 5766 return; 5767 } 5768 5769 mstate->dtms_scratch_ptr += size; 5770 saved = mstate->dtms_scratch_ptr; 5771 5772 /* 5773 * Now get a stack with both program counters and frame pointers. 5774 */ 5775 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 5776 dtrace_getufpstack(buf, fps, nframes + 1); 5777 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 5778 5779 /* 5780 * If that faulted, we're cooked. 5781 */ 5782 if (*flags & CPU_DTRACE_FAULT) 5783 goto out; 5784 5785 /* 5786 * Now we want to walk up the stack, calling the USTACK helper. For 5787 * each iteration, we restore the scratch pointer. 5788 */ 5789 for (i = 0; i < nframes; i++) { 5790 mstate->dtms_scratch_ptr = saved; 5791 5792 if (offs >= strsize) 5793 break; 5794 5795 sym = (char *)(uintptr_t)dtrace_helper( 5796 DTRACE_HELPER_ACTION_USTACK, 5797 mstate, state, pcs[i], fps[i]); 5798 5799 /* 5800 * If we faulted while running the helper, we're going to 5801 * clear the fault and null out the corresponding string. 5802 */ 5803 if (*flags & CPU_DTRACE_FAULT) { 5804 *flags &= ~CPU_DTRACE_FAULT; 5805 str[offs++] = '\0'; 5806 continue; 5807 } 5808 5809 if (sym == NULL) { 5810 str[offs++] = '\0'; 5811 continue; 5812 } 5813 5814 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 5815 5816 /* 5817 * Now copy in the string that the helper returned to us. 5818 */ 5819 for (j = 0; offs + j < strsize; j++) { 5820 if ((str[offs + j] = sym[j]) == '\0') 5821 break; 5822 } 5823 5824 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 5825 5826 offs += j + 1; 5827 } 5828 5829 if (offs >= strsize) { 5830 /* 5831 * If we didn't have room for all of the strings, we don't 5832 * abort processing -- this needn't be a fatal error -- but we 5833 * still want to increment a counter (dts_stkstroverflows) to 5834 * allow this condition to be warned about. (If this is from 5835 * a jstack() action, it is easily tuned via jstackstrsize.) 5836 */ 5837 dtrace_error(&state->dts_stkstroverflows); 5838 } 5839 5840 while (offs < strsize) 5841 str[offs++] = '\0'; 5842 5843out: 5844 mstate->dtms_scratch_ptr = old; 5845} 5846#endif 5847 5848/* 5849 * If you're looking for the epicenter of DTrace, you just found it. This 5850 * is the function called by the provider to fire a probe -- from which all 5851 * subsequent probe-context DTrace activity emanates. 5852 */ 5853void 5854dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1, 5855 uintptr_t arg2, uintptr_t arg3, uintptr_t arg4) 5856{ 5857 processorid_t cpuid; 5858 dtrace_icookie_t cookie; 5859 dtrace_probe_t *probe; 5860 dtrace_mstate_t mstate; 5861 dtrace_ecb_t *ecb; 5862 dtrace_action_t *act; 5863 intptr_t offs; 5864 size_t size; 5865 int vtime, onintr; 5866 volatile uint16_t *flags; 5867 hrtime_t now; 5868 5869#if defined(sun) 5870 /* 5871 * Kick out immediately if this CPU is still being born (in which case 5872 * curthread will be set to -1) or the current thread can't allow 5873 * probes in its current context. 5874 */ 5875 if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE)) 5876 return; 5877#endif 5878 5879 cookie = dtrace_interrupt_disable(); 5880 probe = dtrace_probes[id - 1]; 5881 cpuid = curcpu; 5882 onintr = CPU_ON_INTR(CPU); 5883 5884 if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE && 5885 probe->dtpr_predcache == curthread->t_predcache) { 5886 /* 5887 * We have hit in the predicate cache; we know that 5888 * this predicate would evaluate to be false. 5889 */ 5890 dtrace_interrupt_enable(cookie); 5891 return; 5892 } 5893 5894#if defined(sun) 5895 if (panic_quiesce) { 5896#else 5897 if (panicstr != NULL) { 5898#endif 5899 /* 5900 * We don't trace anything if we're panicking. 5901 */ 5902 dtrace_interrupt_enable(cookie); 5903 return; 5904 } 5905 5906 now = dtrace_gethrtime(); 5907 vtime = dtrace_vtime_references != 0; 5908 5909 if (vtime && curthread->t_dtrace_start) 5910 curthread->t_dtrace_vtime += now - curthread->t_dtrace_start; 5911 5912 mstate.dtms_difo = NULL; 5913 mstate.dtms_probe = probe; 5914 mstate.dtms_strtok = 0; 5915 mstate.dtms_arg[0] = arg0; 5916 mstate.dtms_arg[1] = arg1; 5917 mstate.dtms_arg[2] = arg2; 5918 mstate.dtms_arg[3] = arg3; 5919 mstate.dtms_arg[4] = arg4; 5920 5921 flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags; 5922 5923 for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { 5924 dtrace_predicate_t *pred = ecb->dte_predicate; 5925 dtrace_state_t *state = ecb->dte_state; 5926 dtrace_buffer_t *buf = &state->dts_buffer[cpuid]; 5927 dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid]; 5928 dtrace_vstate_t *vstate = &state->dts_vstate; 5929 dtrace_provider_t *prov = probe->dtpr_provider; 5930 int committed = 0; 5931 caddr_t tomax; 5932 5933 /* 5934 * A little subtlety with the following (seemingly innocuous) 5935 * declaration of the automatic 'val': by looking at the 5936 * code, you might think that it could be declared in the 5937 * action processing loop, below. (That is, it's only used in 5938 * the action processing loop.) However, it must be declared 5939 * out of that scope because in the case of DIF expression 5940 * arguments to aggregating actions, one iteration of the 5941 * action loop will use the last iteration's value. 5942 */ 5943 uint64_t val = 0; 5944 5945 mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE; 5946 *flags &= ~CPU_DTRACE_ERROR; 5947 5948 if (prov == dtrace_provider) { 5949 /* 5950 * If dtrace itself is the provider of this probe, 5951 * we're only going to continue processing the ECB if 5952 * arg0 (the dtrace_state_t) is equal to the ECB's 5953 * creating state. (This prevents disjoint consumers 5954 * from seeing one another's metaprobes.) 5955 */ 5956 if (arg0 != (uint64_t)(uintptr_t)state) 5957 continue; 5958 } 5959 5960 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) { 5961 /* 5962 * We're not currently active. If our provider isn't 5963 * the dtrace pseudo provider, we're not interested. 5964 */ 5965 if (prov != dtrace_provider) 5966 continue; 5967 5968 /* 5969 * Now we must further check if we are in the BEGIN 5970 * probe. If we are, we will only continue processing 5971 * if we're still in WARMUP -- if one BEGIN enabling 5972 * has invoked the exit() action, we don't want to 5973 * evaluate subsequent BEGIN enablings. 5974 */ 5975 if (probe->dtpr_id == dtrace_probeid_begin && 5976 state->dts_activity != DTRACE_ACTIVITY_WARMUP) { 5977 ASSERT(state->dts_activity == 5978 DTRACE_ACTIVITY_DRAINING); 5979 continue; 5980 } 5981 } 5982 5983 if (ecb->dte_cond) { 5984 /* 5985 * If the dte_cond bits indicate that this 5986 * consumer is only allowed to see user-mode firings 5987 * of this probe, call the provider's dtps_usermode() 5988 * entry point to check that the probe was fired 5989 * while in a user context. Skip this ECB if that's 5990 * not the case. 5991 */ 5992 if ((ecb->dte_cond & DTRACE_COND_USERMODE) && 5993 prov->dtpv_pops.dtps_usermode(prov->dtpv_arg, 5994 probe->dtpr_id, probe->dtpr_arg) == 0) 5995 continue; 5996 5997#if defined(sun) 5998 /* 5999 * This is more subtle than it looks. We have to be 6000 * absolutely certain that CRED() isn't going to 6001 * change out from under us so it's only legit to 6002 * examine that structure if we're in constrained 6003 * situations. Currently, the only times we'll this 6004 * check is if a non-super-user has enabled the 6005 * profile or syscall providers -- providers that 6006 * allow visibility of all processes. For the 6007 * profile case, the check above will ensure that 6008 * we're examining a user context. 6009 */ 6010 if (ecb->dte_cond & DTRACE_COND_OWNER) { 6011 cred_t *cr; 6012 cred_t *s_cr = 6013 ecb->dte_state->dts_cred.dcr_cred; 6014 proc_t *proc; 6015 6016 ASSERT(s_cr != NULL); 6017 6018 if ((cr = CRED()) == NULL || 6019 s_cr->cr_uid != cr->cr_uid || 6020 s_cr->cr_uid != cr->cr_ruid || 6021 s_cr->cr_uid != cr->cr_suid || 6022 s_cr->cr_gid != cr->cr_gid || 6023 s_cr->cr_gid != cr->cr_rgid || 6024 s_cr->cr_gid != cr->cr_sgid || 6025 (proc = ttoproc(curthread)) == NULL || 6026 (proc->p_flag & SNOCD)) 6027 continue; 6028 } 6029 6030 if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) { 6031 cred_t *cr; 6032 cred_t *s_cr = 6033 ecb->dte_state->dts_cred.dcr_cred; 6034 6035 ASSERT(s_cr != NULL); 6036 6037 if ((cr = CRED()) == NULL || 6038 s_cr->cr_zone->zone_id != 6039 cr->cr_zone->zone_id) 6040 continue; 6041 } 6042#endif 6043 } 6044 6045 if (now - state->dts_alive > dtrace_deadman_timeout) { 6046 /* 6047 * We seem to be dead. Unless we (a) have kernel 6048 * destructive permissions (b) have expicitly enabled 6049 * destructive actions and (c) destructive actions have 6050 * not been disabled, we're going to transition into 6051 * the KILLED state, from which no further processing 6052 * on this state will be performed. 6053 */ 6054 if (!dtrace_priv_kernel_destructive(state) || 6055 !state->dts_cred.dcr_destructive || 6056 dtrace_destructive_disallow) { 6057 void *activity = &state->dts_activity; 6058 dtrace_activity_t current; 6059 6060 do { 6061 current = state->dts_activity; 6062 } while (dtrace_cas32(activity, current, 6063 DTRACE_ACTIVITY_KILLED) != current); 6064 6065 continue; 6066 } 6067 } 6068 6069 if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed, 6070 ecb->dte_alignment, state, &mstate)) < 0) 6071 continue; 6072 6073 tomax = buf->dtb_tomax; 6074 ASSERT(tomax != NULL); 6075 6076 if (ecb->dte_size != 0) 6077 DTRACE_STORE(uint32_t, tomax, offs, ecb->dte_epid); 6078 6079 mstate.dtms_epid = ecb->dte_epid; 6080 mstate.dtms_present |= DTRACE_MSTATE_EPID; 6081 6082 if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) 6083 mstate.dtms_access = DTRACE_ACCESS_KERNEL; 6084 else 6085 mstate.dtms_access = 0; 6086 6087 if (pred != NULL) { 6088 dtrace_difo_t *dp = pred->dtp_difo; 6089 int rval; 6090 6091 rval = dtrace_dif_emulate(dp, &mstate, vstate, state); 6092 6093 if (!(*flags & CPU_DTRACE_ERROR) && !rval) { 6094 dtrace_cacheid_t cid = probe->dtpr_predcache; 6095 6096 if (cid != DTRACE_CACHEIDNONE && !onintr) { 6097 /* 6098 * Update the predicate cache... 6099 */ 6100 ASSERT(cid == pred->dtp_cacheid); 6101 curthread->t_predcache = cid; 6102 } 6103 6104 continue; 6105 } 6106 } 6107 6108 for (act = ecb->dte_action; !(*flags & CPU_DTRACE_ERROR) && 6109 act != NULL; act = act->dta_next) { 6110 size_t valoffs; 6111 dtrace_difo_t *dp; 6112 dtrace_recdesc_t *rec = &act->dta_rec; 6113 6114 size = rec->dtrd_size; 6115 valoffs = offs + rec->dtrd_offset; 6116 6117 if (DTRACEACT_ISAGG(act->dta_kind)) { 6118 uint64_t v = 0xbad; 6119 dtrace_aggregation_t *agg; 6120 6121 agg = (dtrace_aggregation_t *)act; 6122 6123 if ((dp = act->dta_difo) != NULL) 6124 v = dtrace_dif_emulate(dp, 6125 &mstate, vstate, state); 6126 6127 if (*flags & CPU_DTRACE_ERROR) 6128 continue; 6129 6130 /* 6131 * Note that we always pass the expression 6132 * value from the previous iteration of the 6133 * action loop. This value will only be used 6134 * if there is an expression argument to the 6135 * aggregating action, denoted by the 6136 * dtag_hasarg field. 6137 */ 6138 dtrace_aggregate(agg, buf, 6139 offs, aggbuf, v, val); 6140 continue; 6141 } 6142 6143 switch (act->dta_kind) { 6144 case DTRACEACT_STOP: 6145 if (dtrace_priv_proc_destructive(state)) 6146 dtrace_action_stop(); 6147 continue; 6148 6149 case DTRACEACT_BREAKPOINT: 6150 if (dtrace_priv_kernel_destructive(state)) 6151 dtrace_action_breakpoint(ecb); 6152 continue; 6153 6154 case DTRACEACT_PANIC: 6155 if (dtrace_priv_kernel_destructive(state)) 6156 dtrace_action_panic(ecb); 6157 continue; 6158 6159 case DTRACEACT_STACK: 6160 if (!dtrace_priv_kernel(state)) 6161 continue; 6162 6163 dtrace_getpcstack((pc_t *)(tomax + valoffs), 6164 size / sizeof (pc_t), probe->dtpr_aframes, 6165 DTRACE_ANCHORED(probe) ? NULL : 6166 (uint32_t *)arg0); 6167 continue; 6168 6169#if defined(sun) 6170 case DTRACEACT_JSTACK: 6171 case DTRACEACT_USTACK: 6172 if (!dtrace_priv_proc(state)) 6173 continue; 6174 6175 /* 6176 * See comment in DIF_VAR_PID. 6177 */ 6178 if (DTRACE_ANCHORED(mstate.dtms_probe) && 6179 CPU_ON_INTR(CPU)) { 6180 int depth = DTRACE_USTACK_NFRAMES( 6181 rec->dtrd_arg) + 1; 6182 6183 dtrace_bzero((void *)(tomax + valoffs), 6184 DTRACE_USTACK_STRSIZE(rec->dtrd_arg) 6185 + depth * sizeof (uint64_t)); 6186 6187 continue; 6188 } 6189 6190 if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 && 6191 curproc->p_dtrace_helpers != NULL) { 6192 /* 6193 * This is the slow path -- we have 6194 * allocated string space, and we're 6195 * getting the stack of a process that 6196 * has helpers. Call into a separate 6197 * routine to perform this processing. 6198 */ 6199 dtrace_action_ustack(&mstate, state, 6200 (uint64_t *)(tomax + valoffs), 6201 rec->dtrd_arg); 6202 continue; 6203 } 6204 6205 DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); 6206 dtrace_getupcstack((uint64_t *) 6207 (tomax + valoffs), 6208 DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + 1); 6209 DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); 6210 continue; 6211#endif 6212 6213 default: 6214 break; 6215 } 6216 6217 dp = act->dta_difo; 6218 ASSERT(dp != NULL); 6219 6220 val = dtrace_dif_emulate(dp, &mstate, vstate, state); 6221 6222 if (*flags & CPU_DTRACE_ERROR) 6223 continue; 6224 6225 switch (act->dta_kind) { 6226 case DTRACEACT_SPECULATE: 6227 ASSERT(buf == &state->dts_buffer[cpuid]); 6228 buf = dtrace_speculation_buffer(state, 6229 cpuid, val); 6230 6231 if (buf == NULL) { 6232 *flags |= CPU_DTRACE_DROP; 6233 continue; 6234 } 6235 6236 offs = dtrace_buffer_reserve(buf, 6237 ecb->dte_needed, ecb->dte_alignment, 6238 state, NULL); 6239 6240 if (offs < 0) { 6241 *flags |= CPU_DTRACE_DROP; 6242 continue; 6243 } 6244 6245 tomax = buf->dtb_tomax; 6246 ASSERT(tomax != NULL); 6247 6248 if (ecb->dte_size != 0) 6249 DTRACE_STORE(uint32_t, tomax, offs, 6250 ecb->dte_epid); 6251 continue; 6252 6253 case DTRACEACT_PRINTM: { 6254 /* The DIF returns a 'memref'. */ 6255 uintptr_t *memref = (uintptr_t *)(uintptr_t) val; 6256 6257 /* Get the size from the memref. */ 6258 size = memref[1]; 6259 6260 /* 6261 * Check if the size exceeds the allocated 6262 * buffer size. 6263 */ 6264 if (size + sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) { 6265 /* Flag a drop! */ 6266 *flags |= CPU_DTRACE_DROP; 6267 continue; 6268 } 6269 6270 /* Store the size in the buffer first. */ 6271 DTRACE_STORE(uintptr_t, tomax, 6272 valoffs, size); 6273 6274 /* 6275 * Offset the buffer address to the start 6276 * of the data. 6277 */ 6278 valoffs += sizeof(uintptr_t); 6279 6280 /* 6281 * Reset to the memory address rather than 6282 * the memref array, then let the BYREF 6283 * code below do the work to store the 6284 * memory data in the buffer. 6285 */ 6286 val = memref[0]; 6287 break; 6288 } 6289 6290 case DTRACEACT_PRINTT: { 6291 /* The DIF returns a 'typeref'. */ 6292 uintptr_t *typeref = (uintptr_t *)(uintptr_t) val; 6293 char c = '\0' + 1; 6294 size_t s; 6295 6296 /* 6297 * Get the type string length and round it 6298 * up so that the data that follows is 6299 * aligned for easy access. 6300 */ 6301 size_t typs = strlen((char *) typeref[2]) + 1; 6302 typs = roundup(typs, sizeof(uintptr_t)); 6303 6304 /* 6305 *Get the size from the typeref using the 6306 * number of elements and the type size. 6307 */ 6308 size = typeref[1] * typeref[3]; 6309 6310 /* 6311 * Check if the size exceeds the allocated 6312 * buffer size. 6313 */ 6314 if (size + typs + 2 * sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) { 6315 /* Flag a drop! */ 6316 *flags |= CPU_DTRACE_DROP; 6317 6318 } 6319 6320 /* Store the size in the buffer first. */ 6321 DTRACE_STORE(uintptr_t, tomax, 6322 valoffs, size); 6323 valoffs += sizeof(uintptr_t); 6324 6325 /* Store the type size in the buffer. */ 6326 DTRACE_STORE(uintptr_t, tomax, 6327 valoffs, typeref[3]); 6328 valoffs += sizeof(uintptr_t); 6329 6330 val = typeref[2]; 6331 6332 for (s = 0; s < typs; s++) { 6333 if (c != '\0') 6334 c = dtrace_load8(val++); 6335 6336 DTRACE_STORE(uint8_t, tomax, 6337 valoffs++, c); 6338 } 6339 6340 /* 6341 * Reset to the memory address rather than 6342 * the typeref array, then let the BYREF 6343 * code below do the work to store the 6344 * memory data in the buffer. 6345 */ 6346 val = typeref[0]; 6347 break; 6348 } 6349 6350 case DTRACEACT_CHILL: 6351 if (dtrace_priv_kernel_destructive(state)) 6352 dtrace_action_chill(&mstate, val); 6353 continue; 6354 6355 case DTRACEACT_RAISE: 6356 if (dtrace_priv_proc_destructive(state)) 6357 dtrace_action_raise(val); 6358 continue; 6359 6360 case DTRACEACT_COMMIT: 6361 ASSERT(!committed); 6362 6363 /* 6364 * We need to commit our buffer state. 6365 */ 6366 if (ecb->dte_size) 6367 buf->dtb_offset = offs + ecb->dte_size; 6368 buf = &state->dts_buffer[cpuid]; 6369 dtrace_speculation_commit(state, cpuid, val); 6370 committed = 1; 6371 continue; 6372 6373 case DTRACEACT_DISCARD: 6374 dtrace_speculation_discard(state, cpuid, val); 6375 continue; 6376 6377 case DTRACEACT_DIFEXPR: 6378 case DTRACEACT_LIBACT: 6379 case DTRACEACT_PRINTF: 6380 case DTRACEACT_PRINTA: 6381 case DTRACEACT_SYSTEM: 6382 case DTRACEACT_FREOPEN: 6383 break; 6384 6385 case DTRACEACT_SYM: 6386 case DTRACEACT_MOD: 6387 if (!dtrace_priv_kernel(state)) 6388 continue; 6389 break; 6390 6391 case DTRACEACT_USYM: 6392 case DTRACEACT_UMOD: 6393 case DTRACEACT_UADDR: { 6394#if defined(sun) 6395 struct pid *pid = curthread->t_procp->p_pidp; 6396#endif 6397 6398 if (!dtrace_priv_proc(state)) 6399 continue; 6400 6401 DTRACE_STORE(uint64_t, tomax, 6402#if defined(sun) 6403 valoffs, (uint64_t)pid->pid_id); 6404#else 6405 valoffs, (uint64_t) curproc->p_pid); 6406#endif 6407 DTRACE_STORE(uint64_t, tomax, 6408 valoffs + sizeof (uint64_t), val); 6409 6410 continue; 6411 } 6412 6413 case DTRACEACT_EXIT: { 6414 /* 6415 * For the exit action, we are going to attempt 6416 * to atomically set our activity to be 6417 * draining. If this fails (either because 6418 * another CPU has beat us to the exit action, 6419 * or because our current activity is something 6420 * other than ACTIVE or WARMUP), we will 6421 * continue. This assures that the exit action 6422 * can be successfully recorded at most once 6423 * when we're in the ACTIVE state. If we're 6424 * encountering the exit() action while in 6425 * COOLDOWN, however, we want to honor the new 6426 * status code. (We know that we're the only 6427 * thread in COOLDOWN, so there is no race.) 6428 */ 6429 void *activity = &state->dts_activity; 6430 dtrace_activity_t current = state->dts_activity; 6431 6432 if (current == DTRACE_ACTIVITY_COOLDOWN) 6433 break; 6434 6435 if (current != DTRACE_ACTIVITY_WARMUP) 6436 current = DTRACE_ACTIVITY_ACTIVE; 6437 6438 if (dtrace_cas32(activity, current, 6439 DTRACE_ACTIVITY_DRAINING) != current) { 6440 *flags |= CPU_DTRACE_DROP; 6441 continue; 6442 } 6443 6444 break; 6445 } 6446 6447 default: 6448 ASSERT(0); 6449 } 6450 6451 if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF) { 6452 uintptr_t end = valoffs + size; 6453 6454 if (!dtrace_vcanload((void *)(uintptr_t)val, 6455 &dp->dtdo_rtype, &mstate, vstate)) 6456 continue; 6457 6458 /* 6459 * If this is a string, we're going to only 6460 * load until we find the zero byte -- after 6461 * which we'll store zero bytes. 6462 */ 6463 if (dp->dtdo_rtype.dtdt_kind == 6464 DIF_TYPE_STRING) { 6465 char c = '\0' + 1; 6466 int intuple = act->dta_intuple; 6467 size_t s; 6468 6469 for (s = 0; s < size; s++) { 6470 if (c != '\0') 6471 c = dtrace_load8(val++); 6472 6473 DTRACE_STORE(uint8_t, tomax, 6474 valoffs++, c); 6475 6476 if (c == '\0' && intuple) 6477 break; 6478 } 6479 6480 continue; 6481 } 6482 6483 while (valoffs < end) { 6484 DTRACE_STORE(uint8_t, tomax, valoffs++, 6485 dtrace_load8(val++)); 6486 } 6487 6488 continue; 6489 } 6490 6491 switch (size) { 6492 case 0: 6493 break; 6494 6495 case sizeof (uint8_t): 6496 DTRACE_STORE(uint8_t, tomax, valoffs, val); 6497 break; 6498 case sizeof (uint16_t): 6499 DTRACE_STORE(uint16_t, tomax, valoffs, val); 6500 break; 6501 case sizeof (uint32_t): 6502 DTRACE_STORE(uint32_t, tomax, valoffs, val); 6503 break; 6504 case sizeof (uint64_t): 6505 DTRACE_STORE(uint64_t, tomax, valoffs, val); 6506 break; 6507 default: 6508 /* 6509 * Any other size should have been returned by 6510 * reference, not by value. 6511 */ 6512 ASSERT(0); 6513 break; 6514 } 6515 } 6516 6517 if (*flags & CPU_DTRACE_DROP) 6518 continue; 6519 6520 if (*flags & CPU_DTRACE_FAULT) { 6521 int ndx; 6522 dtrace_action_t *err; 6523 6524 buf->dtb_errors++; 6525 6526 if (probe->dtpr_id == dtrace_probeid_error) { 6527 /* 6528 * There's nothing we can do -- we had an 6529 * error on the error probe. We bump an 6530 * error counter to at least indicate that 6531 * this condition happened. 6532 */ 6533 dtrace_error(&state->dts_dblerrors); 6534 continue; 6535 } 6536 6537 if (vtime) { 6538 /* 6539 * Before recursing on dtrace_probe(), we 6540 * need to explicitly clear out our start 6541 * time to prevent it from being accumulated 6542 * into t_dtrace_vtime. 6543 */ 6544 curthread->t_dtrace_start = 0; 6545 } 6546 6547 /* 6548 * Iterate over the actions to figure out which action 6549 * we were processing when we experienced the error. 6550 * Note that act points _past_ the faulting action; if 6551 * act is ecb->dte_action, the fault was in the 6552 * predicate, if it's ecb->dte_action->dta_next it's 6553 * in action #1, and so on. 6554 */ 6555 for (err = ecb->dte_action, ndx = 0; 6556 err != act; err = err->dta_next, ndx++) 6557 continue; 6558 6559 dtrace_probe_error(state, ecb->dte_epid, ndx, 6560 (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) ? 6561 mstate.dtms_fltoffs : -1, DTRACE_FLAGS2FLT(*flags), 6562 cpu_core[cpuid].cpuc_dtrace_illval); 6563 6564 continue; 6565 } 6566 6567 if (!committed) 6568 buf->dtb_offset = offs + ecb->dte_size; 6569 } 6570 6571 if (vtime) 6572 curthread->t_dtrace_start = dtrace_gethrtime(); 6573 6574 dtrace_interrupt_enable(cookie); 6575} 6576 6577/* 6578 * DTrace Probe Hashing Functions 6579 * 6580 * The functions in this section (and indeed, the functions in remaining 6581 * sections) are not _called_ from probe context. (Any exceptions to this are 6582 * marked with a "Note:".) Rather, they are called from elsewhere in the 6583 * DTrace framework to look-up probes in, add probes to and remove probes from 6584 * the DTrace probe hashes. (Each probe is hashed by each element of the 6585 * probe tuple -- allowing for fast lookups, regardless of what was 6586 * specified.) 6587 */ 6588static uint_t 6589dtrace_hash_str(const char *p) 6590{ 6591 unsigned int g; 6592 uint_t hval = 0; 6593 6594 while (*p) { 6595 hval = (hval << 4) + *p++; 6596 if ((g = (hval & 0xf0000000)) != 0) 6597 hval ^= g >> 24; 6598 hval &= ~g; 6599 } 6600 return (hval); 6601} 6602 6603static dtrace_hash_t * 6604dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, uintptr_t prevoffs) 6605{ 6606 dtrace_hash_t *hash = kmem_zalloc(sizeof (dtrace_hash_t), KM_SLEEP); 6607 6608 hash->dth_stroffs = stroffs; 6609 hash->dth_nextoffs = nextoffs; 6610 hash->dth_prevoffs = prevoffs; 6611 6612 hash->dth_size = 1; 6613 hash->dth_mask = hash->dth_size - 1; 6614 6615 hash->dth_tab = kmem_zalloc(hash->dth_size * 6616 sizeof (dtrace_hashbucket_t *), KM_SLEEP); 6617 6618 return (hash); 6619} 6620 6621static void 6622dtrace_hash_destroy(dtrace_hash_t *hash) 6623{ 6624#ifdef DEBUG 6625 int i; 6626 6627 for (i = 0; i < hash->dth_size; i++) 6628 ASSERT(hash->dth_tab[i] == NULL); 6629#endif 6630 6631 kmem_free(hash->dth_tab, 6632 hash->dth_size * sizeof (dtrace_hashbucket_t *)); 6633 kmem_free(hash, sizeof (dtrace_hash_t)); 6634} 6635 6636static void 6637dtrace_hash_resize(dtrace_hash_t *hash) 6638{ 6639 int size = hash->dth_size, i, ndx; 6640 int new_size = hash->dth_size << 1; 6641 int new_mask = new_size - 1; 6642 dtrace_hashbucket_t **new_tab, *bucket, *next; 6643 6644 ASSERT((new_size & new_mask) == 0); 6645 6646 new_tab = kmem_zalloc(new_size * sizeof (void *), KM_SLEEP); 6647 6648 for (i = 0; i < size; i++) { 6649 for (bucket = hash->dth_tab[i]; bucket != NULL; bucket = next) { 6650 dtrace_probe_t *probe = bucket->dthb_chain; 6651 6652 ASSERT(probe != NULL); 6653 ndx = DTRACE_HASHSTR(hash, probe) & new_mask; 6654 6655 next = bucket->dthb_next; 6656 bucket->dthb_next = new_tab[ndx]; 6657 new_tab[ndx] = bucket; 6658 } 6659 } 6660 6661 kmem_free(hash->dth_tab, hash->dth_size * sizeof (void *)); 6662 hash->dth_tab = new_tab; 6663 hash->dth_size = new_size; 6664 hash->dth_mask = new_mask; 6665} 6666 6667static void 6668dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new) 6669{ 6670 int hashval = DTRACE_HASHSTR(hash, new); 6671 int ndx = hashval & hash->dth_mask; 6672 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; 6673 dtrace_probe_t **nextp, **prevp; 6674 6675 for (; bucket != NULL; bucket = bucket->dthb_next) { 6676 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new)) 6677 goto add; 6678 } 6679 6680 if ((hash->dth_nbuckets >> 1) > hash->dth_size) { 6681 dtrace_hash_resize(hash); 6682 dtrace_hash_add(hash, new); 6683 return; 6684 } 6685 6686 bucket = kmem_zalloc(sizeof (dtrace_hashbucket_t), KM_SLEEP); 6687 bucket->dthb_next = hash->dth_tab[ndx]; 6688 hash->dth_tab[ndx] = bucket; 6689 hash->dth_nbuckets++; 6690 6691add: 6692 nextp = DTRACE_HASHNEXT(hash, new); 6693 ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL); 6694 *nextp = bucket->dthb_chain; 6695 6696 if (bucket->dthb_chain != NULL) { 6697 prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain); 6698 ASSERT(*prevp == NULL); 6699 *prevp = new; 6700 } 6701 6702 bucket->dthb_chain = new; 6703 bucket->dthb_len++; 6704} 6705 6706static dtrace_probe_t * 6707dtrace_hash_lookup(dtrace_hash_t *hash, dtrace_probe_t *template) 6708{ 6709 int hashval = DTRACE_HASHSTR(hash, template); 6710 int ndx = hashval & hash->dth_mask; 6711 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; 6712 6713 for (; bucket != NULL; bucket = bucket->dthb_next) { 6714 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) 6715 return (bucket->dthb_chain); 6716 } 6717 6718 return (NULL); 6719} 6720 6721static int 6722dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template) 6723{ 6724 int hashval = DTRACE_HASHSTR(hash, template); 6725 int ndx = hashval & hash->dth_mask; 6726 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; 6727 6728 for (; bucket != NULL; bucket = bucket->dthb_next) { 6729 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) 6730 return (bucket->dthb_len); 6731 } 6732 6733 return (0); 6734} 6735 6736static void 6737dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe) 6738{ 6739 int ndx = DTRACE_HASHSTR(hash, probe) & hash->dth_mask; 6740 dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; 6741 6742 dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe); 6743 dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe); 6744 6745 /* 6746 * Find the bucket that we're removing this probe from. 6747 */ 6748 for (; bucket != NULL; bucket = bucket->dthb_next) { 6749 if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe)) 6750 break; 6751 } 6752 6753 ASSERT(bucket != NULL); 6754 6755 if (*prevp == NULL) { 6756 if (*nextp == NULL) { 6757 /* 6758 * The removed probe was the only probe on this 6759 * bucket; we need to remove the bucket. 6760 */ 6761 dtrace_hashbucket_t *b = hash->dth_tab[ndx]; 6762 6763 ASSERT(bucket->dthb_chain == probe); 6764 ASSERT(b != NULL); 6765 6766 if (b == bucket) { 6767 hash->dth_tab[ndx] = bucket->dthb_next; 6768 } else { 6769 while (b->dthb_next != bucket) 6770 b = b->dthb_next; 6771 b->dthb_next = bucket->dthb_next; 6772 } 6773 6774 ASSERT(hash->dth_nbuckets > 0); 6775 hash->dth_nbuckets--; 6776 kmem_free(bucket, sizeof (dtrace_hashbucket_t)); 6777 return; 6778 } 6779 6780 bucket->dthb_chain = *nextp; 6781 } else { 6782 *(DTRACE_HASHNEXT(hash, *prevp)) = *nextp; 6783 } 6784 6785 if (*nextp != NULL) 6786 *(DTRACE_HASHPREV(hash, *nextp)) = *prevp; 6787} 6788 6789/* 6790 * DTrace Utility Functions 6791 * 6792 * These are random utility functions that are _not_ called from probe context. 6793 */ 6794static int 6795dtrace_badattr(const dtrace_attribute_t *a) 6796{ 6797 return (a->dtat_name > DTRACE_STABILITY_MAX || 6798 a->dtat_data > DTRACE_STABILITY_MAX || 6799 a->dtat_class > DTRACE_CLASS_MAX); 6800} 6801 6802/* 6803 * Return a duplicate copy of a string. If the specified string is NULL, 6804 * this function returns a zero-length string. 6805 */ 6806static char * 6807dtrace_strdup(const char *str) 6808{ 6809 char *new = kmem_zalloc((str != NULL ? strlen(str) : 0) + 1, KM_SLEEP); 6810 6811 if (str != NULL) 6812 (void) strcpy(new, str); 6813 6814 return (new); 6815} 6816 6817#define DTRACE_ISALPHA(c) \ 6818 (((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z')) 6819 6820static int 6821dtrace_badname(const char *s) 6822{ 6823 char c; 6824 6825 if (s == NULL || (c = *s++) == '\0') 6826 return (0); 6827 6828 if (!DTRACE_ISALPHA(c) && c != '-' && c != '_' && c != '.') 6829 return (1); 6830 6831 while ((c = *s++) != '\0') { 6832 if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') && 6833 c != '-' && c != '_' && c != '.' && c != '`') 6834 return (1); 6835 } 6836 6837 return (0); 6838} 6839 6840static void 6841dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp) 6842{ 6843 uint32_t priv; 6844 6845#if defined(sun) 6846 if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) { 6847 /* 6848 * For DTRACE_PRIV_ALL, the uid and zoneid don't matter. 6849 */ 6850 priv = DTRACE_PRIV_ALL; 6851 } else { 6852 *uidp = crgetuid(cr); 6853 *zoneidp = crgetzoneid(cr); 6854 6855 priv = 0; 6856 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) 6857 priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER; 6858 else if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) 6859 priv |= DTRACE_PRIV_USER; 6860 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) 6861 priv |= DTRACE_PRIV_PROC; 6862 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) 6863 priv |= DTRACE_PRIV_OWNER; 6864 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) 6865 priv |= DTRACE_PRIV_ZONEOWNER; 6866 } 6867#else 6868 priv = DTRACE_PRIV_ALL; 6869#endif 6870 6871 *privp = priv; 6872} 6873 6874#ifdef DTRACE_ERRDEBUG 6875static void 6876dtrace_errdebug(const char *str) 6877{ 6878 int hval = dtrace_hash_str(str) % DTRACE_ERRHASHSZ; 6879 int occupied = 0; 6880 6881 mutex_enter(&dtrace_errlock); 6882 dtrace_errlast = str; 6883 dtrace_errthread = curthread; 6884 6885 while (occupied++ < DTRACE_ERRHASHSZ) { 6886 if (dtrace_errhash[hval].dter_msg == str) { 6887 dtrace_errhash[hval].dter_count++; 6888 goto out; 6889 } 6890 6891 if (dtrace_errhash[hval].dter_msg != NULL) { 6892 hval = (hval + 1) % DTRACE_ERRHASHSZ; 6893 continue; 6894 } 6895 6896 dtrace_errhash[hval].dter_msg = str; 6897 dtrace_errhash[hval].dter_count = 1; 6898 goto out; 6899 } 6900 6901 panic("dtrace: undersized error hash"); 6902out: 6903 mutex_exit(&dtrace_errlock); 6904} 6905#endif 6906 6907/* 6908 * DTrace Matching Functions 6909 * 6910 * These functions are used to match groups of probes, given some elements of 6911 * a probe tuple, or some globbed expressions for elements of a probe tuple. 6912 */ 6913static int 6914dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid, 6915 zoneid_t zoneid) 6916{ 6917 if (priv != DTRACE_PRIV_ALL) { 6918 uint32_t ppriv = prp->dtpr_provider->dtpv_priv.dtpp_flags; 6919 uint32_t match = priv & ppriv; 6920 6921 /* 6922 * No PRIV_DTRACE_* privileges... 6923 */ 6924 if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER | 6925 DTRACE_PRIV_KERNEL)) == 0) 6926 return (0); 6927 6928 /* 6929 * No matching bits, but there were bits to match... 6930 */ 6931 if (match == 0 && ppriv != 0) 6932 return (0); 6933 6934 /* 6935 * Need to have permissions to the process, but don't... 6936 */ 6937 if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 && 6938 uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) { 6939 return (0); 6940 } 6941 6942 /* 6943 * Need to be in the same zone unless we possess the 6944 * privilege to examine all zones. 6945 */ 6946 if (((ppriv & ~match) & DTRACE_PRIV_ZONEOWNER) != 0 && 6947 zoneid != prp->dtpr_provider->dtpv_priv.dtpp_zoneid) { 6948 return (0); 6949 } 6950 } 6951 6952 return (1); 6953} 6954 6955/* 6956 * dtrace_match_probe compares a dtrace_probe_t to a pre-compiled key, which 6957 * consists of input pattern strings and an ops-vector to evaluate them. 6958 * This function returns >0 for match, 0 for no match, and <0 for error. 6959 */ 6960static int 6961dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp, 6962 uint32_t priv, uid_t uid, zoneid_t zoneid) 6963{ 6964 dtrace_provider_t *pvp = prp->dtpr_provider; 6965 int rv; 6966 6967 if (pvp->dtpv_defunct) 6968 return (0); 6969 6970 if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0) 6971 return (rv); 6972 6973 if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0) 6974 return (rv); 6975 6976 if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0) 6977 return (rv); 6978 6979 if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0) 6980 return (rv); 6981 6982 if (dtrace_match_priv(prp, priv, uid, zoneid) == 0) 6983 return (0); 6984 6985 return (rv); 6986} 6987 6988/* 6989 * dtrace_match_glob() is a safe kernel implementation of the gmatch(3GEN) 6990 * interface for matching a glob pattern 'p' to an input string 's'. Unlike 6991 * libc's version, the kernel version only applies to 8-bit ASCII strings. 6992 * In addition, all of the recursion cases except for '*' matching have been 6993 * unwound. For '*', we still implement recursive evaluation, but a depth 6994 * counter is maintained and matching is aborted if we recurse too deep. 6995 * The function returns 0 if no match, >0 if match, and <0 if recursion error. 6996 */ 6997static int 6998dtrace_match_glob(const char *s, const char *p, int depth) 6999{ 7000 const char *olds; 7001 char s1, c; 7002 int gs; 7003 7004 if (depth > DTRACE_PROBEKEY_MAXDEPTH) 7005 return (-1); 7006 7007 if (s == NULL) 7008 s = ""; /* treat NULL as empty string */ 7009 7010top: 7011 olds = s; 7012 s1 = *s++; 7013 7014 if (p == NULL) 7015 return (0); 7016 7017 if ((c = *p++) == '\0') 7018 return (s1 == '\0'); 7019 7020 switch (c) { 7021 case '[': { 7022 int ok = 0, notflag = 0; 7023 char lc = '\0'; 7024 7025 if (s1 == '\0') 7026 return (0); 7027 7028 if (*p == '!') { 7029 notflag = 1; 7030 p++; 7031 } 7032 7033 if ((c = *p++) == '\0') 7034 return (0); 7035 7036 do { 7037 if (c == '-' && lc != '\0' && *p != ']') { 7038 if ((c = *p++) == '\0') 7039 return (0); 7040 if (c == '\\' && (c = *p++) == '\0') 7041 return (0); 7042 7043 if (notflag) { 7044 if (s1 < lc || s1 > c) 7045 ok++; 7046 else 7047 return (0); 7048 } else if (lc <= s1 && s1 <= c) 7049 ok++; 7050 7051 } else if (c == '\\' && (c = *p++) == '\0') 7052 return (0); 7053 7054 lc = c; /* save left-hand 'c' for next iteration */ 7055 7056 if (notflag) { 7057 if (s1 != c) 7058 ok++; 7059 else 7060 return (0); 7061 } else if (s1 == c) 7062 ok++; 7063 7064 if ((c = *p++) == '\0') 7065 return (0); 7066 7067 } while (c != ']'); 7068 7069 if (ok) 7070 goto top; 7071 7072 return (0); 7073 } 7074 7075 case '\\': 7076 if ((c = *p++) == '\0') 7077 return (0); 7078 /*FALLTHRU*/ 7079 7080 default: 7081 if (c != s1) 7082 return (0); 7083 /*FALLTHRU*/ 7084 7085 case '?': 7086 if (s1 != '\0') 7087 goto top; 7088 return (0); 7089 7090 case '*': 7091 while (*p == '*') 7092 p++; /* consecutive *'s are identical to a single one */ 7093 7094 if (*p == '\0') 7095 return (1); 7096 7097 for (s = olds; *s != '\0'; s++) { 7098 if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0) 7099 return (gs); 7100 } 7101 7102 return (0); 7103 } 7104} 7105 7106/*ARGSUSED*/ 7107static int 7108dtrace_match_string(const char *s, const char *p, int depth) 7109{ 7110 return (s != NULL && strcmp(s, p) == 0); 7111} 7112 7113/*ARGSUSED*/ 7114static int 7115dtrace_match_nul(const char *s, const char *p, int depth) 7116{ 7117 return (1); /* always match the empty pattern */ 7118} 7119 7120/*ARGSUSED*/ 7121static int 7122dtrace_match_nonzero(const char *s, const char *p, int depth) 7123{ 7124 return (s != NULL && s[0] != '\0'); 7125} 7126 7127static int 7128dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid, 7129 zoneid_t zoneid, int (*matched)(dtrace_probe_t *, void *), void *arg) 7130{ 7131 dtrace_probe_t template, *probe; 7132 dtrace_hash_t *hash = NULL; 7133 int len, best = INT_MAX, nmatched = 0; 7134 dtrace_id_t i; 7135 7136 ASSERT(MUTEX_HELD(&dtrace_lock)); 7137 7138 /* 7139 * If the probe ID is specified in the key, just lookup by ID and 7140 * invoke the match callback once if a matching probe is found. 7141 */ 7142 if (pkp->dtpk_id != DTRACE_IDNONE) { 7143 if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL && 7144 dtrace_match_probe(probe, pkp, priv, uid, zoneid) > 0) { 7145 (void) (*matched)(probe, arg); 7146 nmatched++; 7147 } 7148 return (nmatched); 7149 } 7150 7151 template.dtpr_mod = (char *)pkp->dtpk_mod; 7152 template.dtpr_func = (char *)pkp->dtpk_func; 7153 template.dtpr_name = (char *)pkp->dtpk_name; 7154 7155 /* 7156 * We want to find the most distinct of the module name, function 7157 * name, and name. So for each one that is not a glob pattern or 7158 * empty string, we perform a lookup in the corresponding hash and 7159 * use the hash table with the fewest collisions to do our search. 7160 */ 7161 if (pkp->dtpk_mmatch == &dtrace_match_string && 7162 (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) { 7163 best = len; 7164 hash = dtrace_bymod; 7165 } 7166 7167 if (pkp->dtpk_fmatch == &dtrace_match_string && 7168 (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) { 7169 best = len; 7170 hash = dtrace_byfunc; 7171 } 7172 7173 if (pkp->dtpk_nmatch == &dtrace_match_string && 7174 (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) { 7175 best = len; 7176 hash = dtrace_byname; 7177 } 7178 7179 /* 7180 * If we did not select a hash table, iterate over every probe and 7181 * invoke our callback for each one that matches our input probe key. 7182 */ 7183 if (hash == NULL) { 7184 for (i = 0; i < dtrace_nprobes; i++) { 7185 if ((probe = dtrace_probes[i]) == NULL || 7186 dtrace_match_probe(probe, pkp, priv, uid, 7187 zoneid) <= 0) 7188 continue; 7189 7190 nmatched++; 7191 7192 if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT) 7193 break; 7194 } 7195 7196 return (nmatched); 7197 } 7198 7199 /* 7200 * If we selected a hash table, iterate over each probe of the same key 7201 * name and invoke the callback for every probe that matches the other 7202 * attributes of our input probe key. 7203 */ 7204 for (probe = dtrace_hash_lookup(hash, &template); probe != NULL; 7205 probe = *(DTRACE_HASHNEXT(hash, probe))) { 7206 7207 if (dtrace_match_probe(probe, pkp, priv, uid, zoneid) <= 0) 7208 continue; 7209 7210 nmatched++; 7211 7212 if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT) 7213 break; 7214 } 7215 7216 return (nmatched); 7217} 7218 7219/* 7220 * Return the function pointer dtrace_probecmp() should use to compare the 7221 * specified pattern with a string. For NULL or empty patterns, we select 7222 * dtrace_match_nul(). For glob pattern strings, we use dtrace_match_glob(). 7223 * For non-empty non-glob strings, we use dtrace_match_string(). 7224 */ 7225static dtrace_probekey_f * 7226dtrace_probekey_func(const char *p) 7227{ 7228 char c; 7229 7230 if (p == NULL || *p == '\0') 7231 return (&dtrace_match_nul); 7232 7233 while ((c = *p++) != '\0') { 7234 if (c == '[' || c == '?' || c == '*' || c == '\\') 7235 return (&dtrace_match_glob); 7236 } 7237 7238 return (&dtrace_match_string); 7239} 7240 7241/* 7242 * Build a probe comparison key for use with dtrace_match_probe() from the 7243 * given probe description. By convention, a null key only matches anchored 7244 * probes: if each field is the empty string, reset dtpk_fmatch to 7245 * dtrace_match_nonzero(). 7246 */ 7247static void 7248dtrace_probekey(dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp) 7249{ 7250 pkp->dtpk_prov = pdp->dtpd_provider; 7251 pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider); 7252 7253 pkp->dtpk_mod = pdp->dtpd_mod; 7254 pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod); 7255 7256 pkp->dtpk_func = pdp->dtpd_func; 7257 pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func); 7258 7259 pkp->dtpk_name = pdp->dtpd_name; 7260 pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name); 7261 7262 pkp->dtpk_id = pdp->dtpd_id; 7263 7264 if (pkp->dtpk_id == DTRACE_IDNONE && 7265 pkp->dtpk_pmatch == &dtrace_match_nul && 7266 pkp->dtpk_mmatch == &dtrace_match_nul && 7267 pkp->dtpk_fmatch == &dtrace_match_nul && 7268 pkp->dtpk_nmatch == &dtrace_match_nul) 7269 pkp->dtpk_fmatch = &dtrace_match_nonzero; 7270} 7271 7272/* 7273 * DTrace Provider-to-Framework API Functions 7274 * 7275 * These functions implement much of the Provider-to-Framework API, as 7276 * described in <sys/dtrace.h>. The parts of the API not in this section are 7277 * the functions in the API for probe management (found below), and 7278 * dtrace_probe() itself (found above). 7279 */ 7280 7281/* 7282 * Register the calling provider with the DTrace framework. This should 7283 * generally be called by DTrace providers in their attach(9E) entry point. 7284 */ 7285int 7286dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv, 7287 cred_t *cr, const dtrace_pops_t *pops, void *arg, dtrace_provider_id_t *idp) 7288{ 7289 dtrace_provider_t *provider; 7290 7291 if (name == NULL || pap == NULL || pops == NULL || idp == NULL) { 7292 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7293 "arguments", name ? name : "<NULL>"); 7294 return (EINVAL); 7295 } 7296 7297 if (name[0] == '\0' || dtrace_badname(name)) { 7298 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7299 "provider name", name); 7300 return (EINVAL); 7301 } 7302 7303 if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) || 7304 pops->dtps_enable == NULL || pops->dtps_disable == NULL || 7305 pops->dtps_destroy == NULL || 7306 ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) { 7307 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7308 "provider ops", name); 7309 return (EINVAL); 7310 } 7311 7312 if (dtrace_badattr(&pap->dtpa_provider) || 7313 dtrace_badattr(&pap->dtpa_mod) || 7314 dtrace_badattr(&pap->dtpa_func) || 7315 dtrace_badattr(&pap->dtpa_name) || 7316 dtrace_badattr(&pap->dtpa_args)) { 7317 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7318 "provider attributes", name); 7319 return (EINVAL); 7320 } 7321 7322 if (priv & ~DTRACE_PRIV_ALL) { 7323 cmn_err(CE_WARN, "failed to register provider '%s': invalid " 7324 "privilege attributes", name); 7325 return (EINVAL); 7326 } 7327 7328 if ((priv & DTRACE_PRIV_KERNEL) && 7329 (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) && 7330 pops->dtps_usermode == NULL) { 7331 cmn_err(CE_WARN, "failed to register provider '%s': need " 7332 "dtps_usermode() op for given privilege attributes", name); 7333 return (EINVAL); 7334 } 7335 7336 provider = kmem_zalloc(sizeof (dtrace_provider_t), KM_SLEEP); 7337 provider->dtpv_name = kmem_alloc(strlen(name) + 1, KM_SLEEP); 7338 (void) strcpy(provider->dtpv_name, name); 7339 7340 provider->dtpv_attr = *pap; 7341 provider->dtpv_priv.dtpp_flags = priv; 7342 if (cr != NULL) { 7343 provider->dtpv_priv.dtpp_uid = crgetuid(cr); 7344 provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr); 7345 } 7346 provider->dtpv_pops = *pops; 7347 7348 if (pops->dtps_provide == NULL) { 7349 ASSERT(pops->dtps_provide_module != NULL); 7350 provider->dtpv_pops.dtps_provide = 7351 (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop; 7352 } 7353 7354 if (pops->dtps_provide_module == NULL) { 7355 ASSERT(pops->dtps_provide != NULL); 7356 provider->dtpv_pops.dtps_provide_module = 7357 (void (*)(void *, modctl_t *))dtrace_nullop; 7358 } 7359 7360 if (pops->dtps_suspend == NULL) { 7361 ASSERT(pops->dtps_resume == NULL); 7362 provider->dtpv_pops.dtps_suspend = 7363 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; 7364 provider->dtpv_pops.dtps_resume = 7365 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; 7366 } 7367 7368 provider->dtpv_arg = arg; 7369 *idp = (dtrace_provider_id_t)provider; 7370 7371 if (pops == &dtrace_provider_ops) { 7372 ASSERT(MUTEX_HELD(&dtrace_provider_lock)); 7373 ASSERT(MUTEX_HELD(&dtrace_lock)); 7374 ASSERT(dtrace_anon.dta_enabling == NULL); 7375 7376 /* 7377 * We make sure that the DTrace provider is at the head of 7378 * the provider chain. 7379 */ 7380 provider->dtpv_next = dtrace_provider; 7381 dtrace_provider = provider; 7382 return (0); 7383 } 7384 7385 mutex_enter(&dtrace_provider_lock); 7386 mutex_enter(&dtrace_lock); 7387 7388 /* 7389 * If there is at least one provider registered, we'll add this 7390 * provider after the first provider. 7391 */ 7392 if (dtrace_provider != NULL) { 7393 provider->dtpv_next = dtrace_provider->dtpv_next; 7394 dtrace_provider->dtpv_next = provider; 7395 } else { 7396 dtrace_provider = provider; 7397 } 7398 7399 if (dtrace_retained != NULL) { 7400 dtrace_enabling_provide(provider); 7401 7402 /* 7403 * Now we need to call dtrace_enabling_matchall() -- which 7404 * will acquire cpu_lock and dtrace_lock. We therefore need 7405 * to drop all of our locks before calling into it... 7406 */ 7407 mutex_exit(&dtrace_lock); 7408 mutex_exit(&dtrace_provider_lock); 7409 dtrace_enabling_matchall(); 7410 7411 return (0); 7412 } 7413 7414 mutex_exit(&dtrace_lock); 7415 mutex_exit(&dtrace_provider_lock); 7416 7417 return (0); 7418} 7419 7420/* 7421 * Unregister the specified provider from the DTrace framework. This should 7422 * generally be called by DTrace providers in their detach(9E) entry point. 7423 */ 7424int 7425dtrace_unregister(dtrace_provider_id_t id) 7426{ 7427 dtrace_provider_t *old = (dtrace_provider_t *)id; 7428 dtrace_provider_t *prev = NULL; 7429 int i, self = 0; 7430 dtrace_probe_t *probe, *first = NULL; 7431 7432 if (old->dtpv_pops.dtps_enable == 7433 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop) { 7434 /* 7435 * If DTrace itself is the provider, we're called with locks 7436 * already held. 7437 */ 7438 ASSERT(old == dtrace_provider); 7439#if defined(sun) 7440 ASSERT(dtrace_devi != NULL); 7441#endif 7442 ASSERT(MUTEX_HELD(&dtrace_provider_lock)); 7443 ASSERT(MUTEX_HELD(&dtrace_lock)); 7444 self = 1; 7445 7446 if (dtrace_provider->dtpv_next != NULL) { 7447 /* 7448 * There's another provider here; return failure. 7449 */ 7450 return (EBUSY); 7451 } 7452 } else { 7453 mutex_enter(&dtrace_provider_lock); 7454 mutex_enter(&mod_lock); 7455 mutex_enter(&dtrace_lock); 7456 } 7457 7458 /* 7459 * If anyone has /dev/dtrace open, or if there are anonymous enabled 7460 * probes, we refuse to let providers slither away, unless this 7461 * provider has already been explicitly invalidated. 7462 */ 7463 if (!old->dtpv_defunct && 7464 (dtrace_opens || (dtrace_anon.dta_state != NULL && 7465 dtrace_anon.dta_state->dts_necbs > 0))) { 7466 if (!self) { 7467 mutex_exit(&dtrace_lock); 7468 mutex_exit(&mod_lock); 7469 mutex_exit(&dtrace_provider_lock); 7470 } 7471 return (EBUSY); 7472 } 7473 7474 /* 7475 * Attempt to destroy the probes associated with this provider. 7476 */ 7477 for (i = 0; i < dtrace_nprobes; i++) { 7478 if ((probe = dtrace_probes[i]) == NULL) 7479 continue; 7480 7481 if (probe->dtpr_provider != old) 7482 continue; 7483 7484 if (probe->dtpr_ecb == NULL) 7485 continue; 7486 7487 /* 7488 * We have at least one ECB; we can't remove this provider. 7489 */ 7490 if (!self) { 7491 mutex_exit(&dtrace_lock); 7492 mutex_exit(&mod_lock); 7493 mutex_exit(&dtrace_provider_lock); 7494 } 7495 return (EBUSY); 7496 } 7497 7498 /* 7499 * All of the probes for this provider are disabled; we can safely 7500 * remove all of them from their hash chains and from the probe array. 7501 */ 7502 for (i = 0; i < dtrace_nprobes; i++) { 7503 if ((probe = dtrace_probes[i]) == NULL) 7504 continue; 7505 7506 if (probe->dtpr_provider != old) 7507 continue; 7508 7509 dtrace_probes[i] = NULL; 7510 7511 dtrace_hash_remove(dtrace_bymod, probe); 7512 dtrace_hash_remove(dtrace_byfunc, probe); 7513 dtrace_hash_remove(dtrace_byname, probe); 7514 7515 if (first == NULL) { 7516 first = probe; 7517 probe->dtpr_nextmod = NULL; 7518 } else { 7519 probe->dtpr_nextmod = first; 7520 first = probe; 7521 } 7522 } 7523 7524 /* 7525 * The provider's probes have been removed from the hash chains and 7526 * from the probe array. Now issue a dtrace_sync() to be sure that 7527 * everyone has cleared out from any probe array processing. 7528 */ 7529 dtrace_sync(); 7530 7531 for (probe = first; probe != NULL; probe = first) { 7532 first = probe->dtpr_nextmod; 7533 7534 old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe->dtpr_id, 7535 probe->dtpr_arg); 7536 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); 7537 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); 7538 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); 7539#if defined(sun) 7540 vmem_free(dtrace_arena, (void *)(uintptr_t)(probe->dtpr_id), 1); 7541#else 7542 free_unr(dtrace_arena, probe->dtpr_id); 7543#endif 7544 kmem_free(probe, sizeof (dtrace_probe_t)); 7545 } 7546 7547 if ((prev = dtrace_provider) == old) { 7548#if defined(sun) 7549 ASSERT(self || dtrace_devi == NULL); 7550 ASSERT(old->dtpv_next == NULL || dtrace_devi == NULL); 7551#endif 7552 dtrace_provider = old->dtpv_next; 7553 } else { 7554 while (prev != NULL && prev->dtpv_next != old) 7555 prev = prev->dtpv_next; 7556 7557 if (prev == NULL) { 7558 panic("attempt to unregister non-existent " 7559 "dtrace provider %p\n", (void *)id); 7560 } 7561 7562 prev->dtpv_next = old->dtpv_next; 7563 } 7564 7565 if (!self) { 7566 mutex_exit(&dtrace_lock); 7567 mutex_exit(&mod_lock); 7568 mutex_exit(&dtrace_provider_lock); 7569 } 7570 7571 kmem_free(old->dtpv_name, strlen(old->dtpv_name) + 1); 7572 kmem_free(old, sizeof (dtrace_provider_t)); 7573 7574 return (0); 7575} 7576 7577/* 7578 * Invalidate the specified provider. All subsequent probe lookups for the 7579 * specified provider will fail, but its probes will not be removed. 7580 */ 7581void 7582dtrace_invalidate(dtrace_provider_id_t id) 7583{ 7584 dtrace_provider_t *pvp = (dtrace_provider_t *)id; 7585 7586 ASSERT(pvp->dtpv_pops.dtps_enable != 7587 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop); 7588 7589 mutex_enter(&dtrace_provider_lock); 7590 mutex_enter(&dtrace_lock); 7591 7592 pvp->dtpv_defunct = 1; 7593 7594 mutex_exit(&dtrace_lock); 7595 mutex_exit(&dtrace_provider_lock); 7596} 7597 7598/* 7599 * Indicate whether or not DTrace has attached. 7600 */ 7601int 7602dtrace_attached(void) 7603{ 7604 /* 7605 * dtrace_provider will be non-NULL iff the DTrace driver has 7606 * attached. (It's non-NULL because DTrace is always itself a 7607 * provider.) 7608 */ 7609 return (dtrace_provider != NULL); 7610} 7611 7612/* 7613 * Remove all the unenabled probes for the given provider. This function is 7614 * not unlike dtrace_unregister(), except that it doesn't remove the provider 7615 * -- just as many of its associated probes as it can. 7616 */ 7617int 7618dtrace_condense(dtrace_provider_id_t id) 7619{ 7620 dtrace_provider_t *prov = (dtrace_provider_t *)id; 7621 int i; 7622 dtrace_probe_t *probe; 7623 7624 /* 7625 * Make sure this isn't the dtrace provider itself. 7626 */ 7627 ASSERT(prov->dtpv_pops.dtps_enable != 7628 (void (*)(void *, dtrace_id_t, void *))dtrace_nullop); 7629 7630 mutex_enter(&dtrace_provider_lock); 7631 mutex_enter(&dtrace_lock); 7632 7633 /* 7634 * Attempt to destroy the probes associated with this provider. 7635 */ 7636 for (i = 0; i < dtrace_nprobes; i++) { 7637 if ((probe = dtrace_probes[i]) == NULL) 7638 continue; 7639 7640 if (probe->dtpr_provider != prov) 7641 continue; 7642 7643 if (probe->dtpr_ecb != NULL) 7644 continue; 7645 7646 dtrace_probes[i] = NULL; 7647 7648 dtrace_hash_remove(dtrace_bymod, probe); 7649 dtrace_hash_remove(dtrace_byfunc, probe); 7650 dtrace_hash_remove(dtrace_byname, probe); 7651 7652 prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, i + 1, 7653 probe->dtpr_arg); 7654 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); 7655 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); 7656 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); 7657 kmem_free(probe, sizeof (dtrace_probe_t)); 7658#if defined(sun) 7659 vmem_free(dtrace_arena, (void *)((uintptr_t)i + 1), 1); 7660#else 7661 free_unr(dtrace_arena, i + 1); 7662#endif 7663 } 7664 7665 mutex_exit(&dtrace_lock); 7666 mutex_exit(&dtrace_provider_lock); 7667 7668 return (0); 7669} 7670 7671/* 7672 * DTrace Probe Management Functions 7673 * 7674 * The functions in this section perform the DTrace probe management, 7675 * including functions to create probes, look-up probes, and call into the 7676 * providers to request that probes be provided. Some of these functions are 7677 * in the Provider-to-Framework API; these functions can be identified by the 7678 * fact that they are not declared "static". 7679 */ 7680 7681/* 7682 * Create a probe with the specified module name, function name, and name. 7683 */ 7684dtrace_id_t 7685dtrace_probe_create(dtrace_provider_id_t prov, const char *mod, 7686 const char *func, const char *name, int aframes, void *arg) 7687{ 7688 dtrace_probe_t *probe, **probes; 7689 dtrace_provider_t *provider = (dtrace_provider_t *)prov; 7690 dtrace_id_t id; 7691 7692 if (provider == dtrace_provider) { 7693 ASSERT(MUTEX_HELD(&dtrace_lock)); 7694 } else { 7695 mutex_enter(&dtrace_lock); 7696 } 7697 7698#if defined(sun) 7699 id = (dtrace_id_t)(uintptr_t)vmem_alloc(dtrace_arena, 1, 7700 VM_BESTFIT | VM_SLEEP); 7701#else 7702 id = alloc_unr(dtrace_arena); 7703#endif 7704 probe = kmem_zalloc(sizeof (dtrace_probe_t), KM_SLEEP); 7705 7706 probe->dtpr_id = id; 7707 probe->dtpr_gen = dtrace_probegen++; 7708 probe->dtpr_mod = dtrace_strdup(mod); 7709 probe->dtpr_func = dtrace_strdup(func); 7710 probe->dtpr_name = dtrace_strdup(name); 7711 probe->dtpr_arg = arg; 7712 probe->dtpr_aframes = aframes; 7713 probe->dtpr_provider = provider; 7714 7715 dtrace_hash_add(dtrace_bymod, probe); 7716 dtrace_hash_add(dtrace_byfunc, probe); 7717 dtrace_hash_add(dtrace_byname, probe); 7718 7719 if (id - 1 >= dtrace_nprobes) { 7720 size_t osize = dtrace_nprobes * sizeof (dtrace_probe_t *); 7721 size_t nsize = osize << 1; 7722 7723 if (nsize == 0) { 7724 ASSERT(osize == 0); 7725 ASSERT(dtrace_probes == NULL); 7726 nsize = sizeof (dtrace_probe_t *); 7727 } 7728 7729 probes = kmem_zalloc(nsize, KM_SLEEP); 7730 7731 if (dtrace_probes == NULL) { 7732 ASSERT(osize == 0); 7733 dtrace_probes = probes; 7734 dtrace_nprobes = 1; 7735 } else { 7736 dtrace_probe_t **oprobes = dtrace_probes; 7737 7738 bcopy(oprobes, probes, osize); 7739 dtrace_membar_producer(); 7740 dtrace_probes = probes; 7741 7742 dtrace_sync(); 7743 7744 /* 7745 * All CPUs are now seeing the new probes array; we can 7746 * safely free the old array. 7747 */ 7748 kmem_free(oprobes, osize); 7749 dtrace_nprobes <<= 1; 7750 } 7751 7752 ASSERT(id - 1 < dtrace_nprobes); 7753 } 7754 7755 ASSERT(dtrace_probes[id - 1] == NULL); 7756 dtrace_probes[id - 1] = probe; 7757 7758 if (provider != dtrace_provider) 7759 mutex_exit(&dtrace_lock); 7760 7761 return (id); 7762} 7763 7764static dtrace_probe_t * 7765dtrace_probe_lookup_id(dtrace_id_t id) 7766{ 7767 ASSERT(MUTEX_HELD(&dtrace_lock)); 7768 7769 if (id == 0 || id > dtrace_nprobes) 7770 return (NULL); 7771 7772 return (dtrace_probes[id - 1]); 7773} 7774 7775static int 7776dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg) 7777{ 7778 *((dtrace_id_t *)arg) = probe->dtpr_id; 7779 7780 return (DTRACE_MATCH_DONE); 7781} 7782 7783/* 7784 * Look up a probe based on provider and one or more of module name, function 7785 * name and probe name. 7786 */ 7787dtrace_id_t 7788dtrace_probe_lookup(dtrace_provider_id_t prid, char *mod, 7789 char *func, char *name) 7790{ 7791 dtrace_probekey_t pkey; 7792 dtrace_id_t id; 7793 int match; 7794 7795 pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name; 7796 pkey.dtpk_pmatch = &dtrace_match_string; 7797 pkey.dtpk_mod = mod; 7798 pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul; 7799 pkey.dtpk_func = func; 7800 pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul; 7801 pkey.dtpk_name = name; 7802 pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul; 7803 pkey.dtpk_id = DTRACE_IDNONE; 7804 7805 mutex_enter(&dtrace_lock); 7806 match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, 0, 7807 dtrace_probe_lookup_match, &id); 7808 mutex_exit(&dtrace_lock); 7809 7810 ASSERT(match == 1 || match == 0); 7811 return (match ? id : 0); 7812} 7813 7814/* 7815 * Returns the probe argument associated with the specified probe. 7816 */ 7817void * 7818dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid) 7819{ 7820 dtrace_probe_t *probe; 7821 void *rval = NULL; 7822 7823 mutex_enter(&dtrace_lock); 7824 7825 if ((probe = dtrace_probe_lookup_id(pid)) != NULL && 7826 probe->dtpr_provider == (dtrace_provider_t *)id) 7827 rval = probe->dtpr_arg; 7828 7829 mutex_exit(&dtrace_lock); 7830 7831 return (rval); 7832} 7833 7834/* 7835 * Copy a probe into a probe description. 7836 */ 7837static void 7838dtrace_probe_description(const dtrace_probe_t *prp, dtrace_probedesc_t *pdp) 7839{ 7840 bzero(pdp, sizeof (dtrace_probedesc_t)); 7841 pdp->dtpd_id = prp->dtpr_id; 7842 7843 (void) strncpy(pdp->dtpd_provider, 7844 prp->dtpr_provider->dtpv_name, DTRACE_PROVNAMELEN - 1); 7845 7846 (void) strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1); 7847 (void) strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1); 7848 (void) strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1); 7849} 7850 7851#if !defined(sun) 7852static int 7853dtrace_probe_provide_cb(linker_file_t lf, void *arg) 7854{ 7855 dtrace_provider_t *prv = (dtrace_provider_t *) arg; 7856 7857 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, lf); 7858 7859 return(0); 7860} 7861#endif 7862 7863 7864/* 7865 * Called to indicate that a probe -- or probes -- should be provided by a 7866 * specfied provider. If the specified description is NULL, the provider will 7867 * be told to provide all of its probes. (This is done whenever a new 7868 * consumer comes along, or whenever a retained enabling is to be matched.) If 7869 * the specified description is non-NULL, the provider is given the 7870 * opportunity to dynamically provide the specified probe, allowing providers 7871 * to support the creation of probes on-the-fly. (So-called _autocreated_ 7872 * probes.) If the provider is NULL, the operations will be applied to all 7873 * providers; if the provider is non-NULL the operations will only be applied 7874 * to the specified provider. The dtrace_provider_lock must be held, and the 7875 * dtrace_lock must _not_ be held -- the provider's dtps_provide() operation 7876 * will need to grab the dtrace_lock when it reenters the framework through 7877 * dtrace_probe_lookup(), dtrace_probe_create(), etc. 7878 */ 7879static void 7880dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv) 7881{ 7882#if defined(sun) 7883 modctl_t *ctl; 7884#endif 7885 int all = 0; 7886 7887 ASSERT(MUTEX_HELD(&dtrace_provider_lock)); 7888 7889 if (prv == NULL) { 7890 all = 1; 7891 prv = dtrace_provider; 7892 } 7893 7894 do { 7895 /* 7896 * First, call the blanket provide operation. 7897 */ 7898 prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc); 7899 7900 /* 7901 * Now call the per-module provide operation. We will grab 7902 * mod_lock to prevent the list from being modified. Note 7903 * that this also prevents the mod_busy bits from changing. 7904 * (mod_busy can only be changed with mod_lock held.) 7905 */ 7906 mutex_enter(&mod_lock); 7907 7908#if defined(sun) 7909 ctl = &modules; 7910 do { 7911 if (ctl->mod_busy || ctl->mod_mp == NULL) 7912 continue; 7913 7914 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl); 7915 7916 } while ((ctl = ctl->mod_next) != &modules); 7917#else 7918 (void) linker_file_foreach(dtrace_probe_provide_cb, prv); 7919#endif 7920 7921 mutex_exit(&mod_lock); 7922 } while (all && (prv = prv->dtpv_next) != NULL); 7923} 7924 7925#if defined(sun) 7926/* 7927 * Iterate over each probe, and call the Framework-to-Provider API function 7928 * denoted by offs. 7929 */ 7930static void 7931dtrace_probe_foreach(uintptr_t offs) 7932{ 7933 dtrace_provider_t *prov; 7934 void (*func)(void *, dtrace_id_t, void *); 7935 dtrace_probe_t *probe; 7936 dtrace_icookie_t cookie; 7937 int i; 7938 7939 /* 7940 * We disable interrupts to walk through the probe array. This is 7941 * safe -- the dtrace_sync() in dtrace_unregister() assures that we 7942 * won't see stale data. 7943 */ 7944 cookie = dtrace_interrupt_disable(); 7945 7946 for (i = 0; i < dtrace_nprobes; i++) { 7947 if ((probe = dtrace_probes[i]) == NULL) 7948 continue; 7949 7950 if (probe->dtpr_ecb == NULL) { 7951 /* 7952 * This probe isn't enabled -- don't call the function. 7953 */ 7954 continue; 7955 } 7956 7957 prov = probe->dtpr_provider; 7958 func = *((void(**)(void *, dtrace_id_t, void *)) 7959 ((uintptr_t)&prov->dtpv_pops + offs)); 7960 7961 func(prov->dtpv_arg, i + 1, probe->dtpr_arg); 7962 } 7963 7964 dtrace_interrupt_enable(cookie); 7965} 7966#endif 7967 7968static int 7969dtrace_probe_enable(dtrace_probedesc_t *desc, dtrace_enabling_t *enab) 7970{ 7971 dtrace_probekey_t pkey; 7972 uint32_t priv; 7973 uid_t uid; 7974 zoneid_t zoneid; 7975 7976 ASSERT(MUTEX_HELD(&dtrace_lock)); 7977 dtrace_ecb_create_cache = NULL; 7978 7979 if (desc == NULL) { 7980 /* 7981 * If we're passed a NULL description, we're being asked to 7982 * create an ECB with a NULL probe. 7983 */ 7984 (void) dtrace_ecb_create_enable(NULL, enab); 7985 return (0); 7986 } 7987 7988 dtrace_probekey(desc, &pkey); 7989 dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred, 7990 &priv, &uid, &zoneid); 7991 7992 return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable, 7993 enab)); 7994} 7995 7996/* 7997 * DTrace Helper Provider Functions 7998 */ 7999static void 8000dtrace_dofattr2attr(dtrace_attribute_t *attr, const dof_attr_t dofattr) 8001{ 8002 attr->dtat_name = DOF_ATTR_NAME(dofattr); 8003 attr->dtat_data = DOF_ATTR_DATA(dofattr); 8004 attr->dtat_class = DOF_ATTR_CLASS(dofattr); 8005} 8006 8007static void 8008dtrace_dofprov2hprov(dtrace_helper_provdesc_t *hprov, 8009 const dof_provider_t *dofprov, char *strtab) 8010{ 8011 hprov->dthpv_provname = strtab + dofprov->dofpv_name; 8012 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_provider, 8013 dofprov->dofpv_provattr); 8014 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_mod, 8015 dofprov->dofpv_modattr); 8016 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_func, 8017 dofprov->dofpv_funcattr); 8018 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_name, 8019 dofprov->dofpv_nameattr); 8020 dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_args, 8021 dofprov->dofpv_argsattr); 8022} 8023 8024static void 8025dtrace_helper_provide_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid) 8026{ 8027 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; 8028 dof_hdr_t *dof = (dof_hdr_t *)daddr; 8029 dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec; 8030 dof_provider_t *provider; 8031 dof_probe_t *probe; 8032 uint32_t *off, *enoff; 8033 uint8_t *arg; 8034 char *strtab; 8035 uint_t i, nprobes; 8036 dtrace_helper_provdesc_t dhpv; 8037 dtrace_helper_probedesc_t dhpb; 8038 dtrace_meta_t *meta = dtrace_meta_pid; 8039 dtrace_mops_t *mops = &meta->dtm_mops; 8040 void *parg; 8041 8042 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); 8043 str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8044 provider->dofpv_strtab * dof->dofh_secsize); 8045 prb_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8046 provider->dofpv_probes * dof->dofh_secsize); 8047 arg_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8048 provider->dofpv_prargs * dof->dofh_secsize); 8049 off_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8050 provider->dofpv_proffs * dof->dofh_secsize); 8051 8052 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); 8053 off = (uint32_t *)(uintptr_t)(daddr + off_sec->dofs_offset); 8054 arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset); 8055 enoff = NULL; 8056 8057 /* 8058 * See dtrace_helper_provider_validate(). 8059 */ 8060 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && 8061 provider->dofpv_prenoffs != DOF_SECT_NONE) { 8062 enoff_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8063 provider->dofpv_prenoffs * dof->dofh_secsize); 8064 enoff = (uint32_t *)(uintptr_t)(daddr + enoff_sec->dofs_offset); 8065 } 8066 8067 nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize; 8068 8069 /* 8070 * Create the provider. 8071 */ 8072 dtrace_dofprov2hprov(&dhpv, provider, strtab); 8073 8074 if ((parg = mops->dtms_provide_pid(meta->dtm_arg, &dhpv, pid)) == NULL) 8075 return; 8076 8077 meta->dtm_count++; 8078 8079 /* 8080 * Create the probes. 8081 */ 8082 for (i = 0; i < nprobes; i++) { 8083 probe = (dof_probe_t *)(uintptr_t)(daddr + 8084 prb_sec->dofs_offset + i * prb_sec->dofs_entsize); 8085 8086 dhpb.dthpb_mod = dhp->dofhp_mod; 8087 dhpb.dthpb_func = strtab + probe->dofpr_func; 8088 dhpb.dthpb_name = strtab + probe->dofpr_name; 8089 dhpb.dthpb_base = probe->dofpr_addr; 8090 dhpb.dthpb_offs = off + probe->dofpr_offidx; 8091 dhpb.dthpb_noffs = probe->dofpr_noffs; 8092 if (enoff != NULL) { 8093 dhpb.dthpb_enoffs = enoff + probe->dofpr_enoffidx; 8094 dhpb.dthpb_nenoffs = probe->dofpr_nenoffs; 8095 } else { 8096 dhpb.dthpb_enoffs = NULL; 8097 dhpb.dthpb_nenoffs = 0; 8098 } 8099 dhpb.dthpb_args = arg + probe->dofpr_argidx; 8100 dhpb.dthpb_nargc = probe->dofpr_nargc; 8101 dhpb.dthpb_xargc = probe->dofpr_xargc; 8102 dhpb.dthpb_ntypes = strtab + probe->dofpr_nargv; 8103 dhpb.dthpb_xtypes = strtab + probe->dofpr_xargv; 8104 8105 mops->dtms_create_probe(meta->dtm_arg, parg, &dhpb); 8106 } 8107} 8108 8109static void 8110dtrace_helper_provide(dof_helper_t *dhp, pid_t pid) 8111{ 8112 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; 8113 dof_hdr_t *dof = (dof_hdr_t *)daddr; 8114 int i; 8115 8116 ASSERT(MUTEX_HELD(&dtrace_meta_lock)); 8117 8118 for (i = 0; i < dof->dofh_secnum; i++) { 8119 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + 8120 dof->dofh_secoff + i * dof->dofh_secsize); 8121 8122 if (sec->dofs_type != DOF_SECT_PROVIDER) 8123 continue; 8124 8125 dtrace_helper_provide_one(dhp, sec, pid); 8126 } 8127 8128 /* 8129 * We may have just created probes, so we must now rematch against 8130 * any retained enablings. Note that this call will acquire both 8131 * cpu_lock and dtrace_lock; the fact that we are holding 8132 * dtrace_meta_lock now is what defines the ordering with respect to 8133 * these three locks. 8134 */ 8135 dtrace_enabling_matchall(); 8136} 8137 8138#if defined(sun) 8139static void 8140dtrace_helper_provider_remove_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid) 8141{ 8142 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; 8143 dof_hdr_t *dof = (dof_hdr_t *)daddr; 8144 dof_sec_t *str_sec; 8145 dof_provider_t *provider; 8146 char *strtab; 8147 dtrace_helper_provdesc_t dhpv; 8148 dtrace_meta_t *meta = dtrace_meta_pid; 8149 dtrace_mops_t *mops = &meta->dtm_mops; 8150 8151 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); 8152 str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + 8153 provider->dofpv_strtab * dof->dofh_secsize); 8154 8155 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); 8156 8157 /* 8158 * Create the provider. 8159 */ 8160 dtrace_dofprov2hprov(&dhpv, provider, strtab); 8161 8162 mops->dtms_remove_pid(meta->dtm_arg, &dhpv, pid); 8163 8164 meta->dtm_count--; 8165} 8166 8167static void 8168dtrace_helper_provider_remove(dof_helper_t *dhp, pid_t pid) 8169{ 8170 uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; 8171 dof_hdr_t *dof = (dof_hdr_t *)daddr; 8172 int i; 8173 8174 ASSERT(MUTEX_HELD(&dtrace_meta_lock)); 8175 8176 for (i = 0; i < dof->dofh_secnum; i++) { 8177 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + 8178 dof->dofh_secoff + i * dof->dofh_secsize); 8179 8180 if (sec->dofs_type != DOF_SECT_PROVIDER) 8181 continue; 8182 8183 dtrace_helper_provider_remove_one(dhp, sec, pid); 8184 } 8185} 8186#endif 8187 8188/* 8189 * DTrace Meta Provider-to-Framework API Functions 8190 * 8191 * These functions implement the Meta Provider-to-Framework API, as described 8192 * in <sys/dtrace.h>. 8193 */ 8194int 8195dtrace_meta_register(const char *name, const dtrace_mops_t *mops, void *arg, 8196 dtrace_meta_provider_id_t *idp) 8197{ 8198 dtrace_meta_t *meta; 8199 dtrace_helpers_t *help, *next; 8200 int i; 8201 8202 *idp = DTRACE_METAPROVNONE; 8203 8204 /* 8205 * We strictly don't need the name, but we hold onto it for 8206 * debuggability. All hail error queues! 8207 */ 8208 if (name == NULL) { 8209 cmn_err(CE_WARN, "failed to register meta-provider: " 8210 "invalid name"); 8211 return (EINVAL); 8212 } 8213 8214 if (mops == NULL || 8215 mops->dtms_create_probe == NULL || 8216 mops->dtms_provide_pid == NULL || 8217 mops->dtms_remove_pid == NULL) { 8218 cmn_err(CE_WARN, "failed to register meta-register %s: " 8219 "invalid ops", name); 8220 return (EINVAL); 8221 } 8222 8223 meta = kmem_zalloc(sizeof (dtrace_meta_t), KM_SLEEP); 8224 meta->dtm_mops = *mops; 8225 meta->dtm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP); 8226 (void) strcpy(meta->dtm_name, name); 8227 meta->dtm_arg = arg; 8228 8229 mutex_enter(&dtrace_meta_lock); 8230 mutex_enter(&dtrace_lock); 8231 8232 if (dtrace_meta_pid != NULL) { 8233 mutex_exit(&dtrace_lock); 8234 mutex_exit(&dtrace_meta_lock); 8235 cmn_err(CE_WARN, "failed to register meta-register %s: " 8236 "user-land meta-provider exists", name); 8237 kmem_free(meta->dtm_name, strlen(meta->dtm_name) + 1); 8238 kmem_free(meta, sizeof (dtrace_meta_t)); 8239 return (EINVAL); 8240 } 8241 8242 dtrace_meta_pid = meta; 8243 *idp = (dtrace_meta_provider_id_t)meta; 8244 8245 /* 8246 * If there are providers and probes ready to go, pass them 8247 * off to the new meta provider now. 8248 */ 8249 8250 help = dtrace_deferred_pid; 8251 dtrace_deferred_pid = NULL; 8252 8253 mutex_exit(&dtrace_lock); 8254 8255 while (help != NULL) { 8256 for (i = 0; i < help->dthps_nprovs; i++) { 8257 dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov, 8258 help->dthps_pid); 8259 } 8260 8261 next = help->dthps_next; 8262 help->dthps_next = NULL; 8263 help->dthps_prev = NULL; 8264 help->dthps_deferred = 0; 8265 help = next; 8266 } 8267 8268 mutex_exit(&dtrace_meta_lock); 8269 8270 return (0); 8271} 8272 8273int 8274dtrace_meta_unregister(dtrace_meta_provider_id_t id) 8275{ 8276 dtrace_meta_t **pp, *old = (dtrace_meta_t *)id; 8277 8278 mutex_enter(&dtrace_meta_lock); 8279 mutex_enter(&dtrace_lock); 8280 8281 if (old == dtrace_meta_pid) { 8282 pp = &dtrace_meta_pid; 8283 } else { 8284 panic("attempt to unregister non-existent " 8285 "dtrace meta-provider %p\n", (void *)old); 8286 } 8287 8288 if (old->dtm_count != 0) { 8289 mutex_exit(&dtrace_lock); 8290 mutex_exit(&dtrace_meta_lock); 8291 return (EBUSY); 8292 } 8293 8294 *pp = NULL; 8295 8296 mutex_exit(&dtrace_lock); 8297 mutex_exit(&dtrace_meta_lock); 8298 8299 kmem_free(old->dtm_name, strlen(old->dtm_name) + 1); 8300 kmem_free(old, sizeof (dtrace_meta_t)); 8301 8302 return (0); 8303} 8304 8305 8306/* 8307 * DTrace DIF Object Functions 8308 */ 8309static int 8310dtrace_difo_err(uint_t pc, const char *format, ...) 8311{ 8312 if (dtrace_err_verbose) { 8313 va_list alist; 8314 8315 (void) uprintf("dtrace DIF object error: [%u]: ", pc); 8316 va_start(alist, format); 8317 (void) vuprintf(format, alist); 8318 va_end(alist); 8319 } 8320 8321#ifdef DTRACE_ERRDEBUG 8322 dtrace_errdebug(format); 8323#endif 8324 return (1); 8325} 8326 8327/* 8328 * Validate a DTrace DIF object by checking the IR instructions. The following 8329 * rules are currently enforced by dtrace_difo_validate(): 8330 * 8331 * 1. Each instruction must have a valid opcode 8332 * 2. Each register, string, variable, or subroutine reference must be valid 8333 * 3. No instruction can modify register %r0 (must be zero) 8334 * 4. All instruction reserved bits must be set to zero 8335 * 5. The last instruction must be a "ret" instruction 8336 * 6. All branch targets must reference a valid instruction _after_ the branch 8337 */ 8338static int 8339dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, uint_t nregs, 8340 cred_t *cr) 8341{ 8342 int err = 0, i; 8343 int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; 8344 int kcheckload; 8345 uint_t pc; 8346 8347 kcheckload = cr == NULL || 8348 (vstate->dtvs_state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) == 0; 8349 8350 dp->dtdo_destructive = 0; 8351 8352 for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) { 8353 dif_instr_t instr = dp->dtdo_buf[pc]; 8354 8355 uint_t r1 = DIF_INSTR_R1(instr); 8356 uint_t r2 = DIF_INSTR_R2(instr); 8357 uint_t rd = DIF_INSTR_RD(instr); 8358 uint_t rs = DIF_INSTR_RS(instr); 8359 uint_t label = DIF_INSTR_LABEL(instr); 8360 uint_t v = DIF_INSTR_VAR(instr); 8361 uint_t subr = DIF_INSTR_SUBR(instr); 8362 uint_t type = DIF_INSTR_TYPE(instr); 8363 uint_t op = DIF_INSTR_OP(instr); 8364 8365 switch (op) { 8366 case DIF_OP_OR: 8367 case DIF_OP_XOR: 8368 case DIF_OP_AND: 8369 case DIF_OP_SLL: 8370 case DIF_OP_SRL: 8371 case DIF_OP_SRA: 8372 case DIF_OP_SUB: 8373 case DIF_OP_ADD: 8374 case DIF_OP_MUL: 8375 case DIF_OP_SDIV: 8376 case DIF_OP_UDIV: 8377 case DIF_OP_SREM: 8378 case DIF_OP_UREM: 8379 case DIF_OP_COPYS: 8380 if (r1 >= nregs) 8381 err += efunc(pc, "invalid register %u\n", r1); 8382 if (r2 >= nregs) 8383 err += efunc(pc, "invalid register %u\n", r2); 8384 if (rd >= nregs) 8385 err += efunc(pc, "invalid register %u\n", rd); 8386 if (rd == 0) 8387 err += efunc(pc, "cannot write to %r0\n"); 8388 break; 8389 case DIF_OP_NOT: 8390 case DIF_OP_MOV: 8391 case DIF_OP_ALLOCS: 8392 if (r1 >= nregs) 8393 err += efunc(pc, "invalid register %u\n", r1); 8394 if (r2 != 0) 8395 err += efunc(pc, "non-zero reserved bits\n"); 8396 if (rd >= nregs) 8397 err += efunc(pc, "invalid register %u\n", rd); 8398 if (rd == 0) 8399 err += efunc(pc, "cannot write to %r0\n"); 8400 break; 8401 case DIF_OP_LDSB: 8402 case DIF_OP_LDSH: 8403 case DIF_OP_LDSW: 8404 case DIF_OP_LDUB: 8405 case DIF_OP_LDUH: 8406 case DIF_OP_LDUW: 8407 case DIF_OP_LDX: 8408 if (r1 >= nregs) 8409 err += efunc(pc, "invalid register %u\n", r1); 8410 if (r2 != 0) 8411 err += efunc(pc, "non-zero reserved bits\n"); 8412 if (rd >= nregs) 8413 err += efunc(pc, "invalid register %u\n", rd); 8414 if (rd == 0) 8415 err += efunc(pc, "cannot write to %r0\n"); 8416 if (kcheckload) 8417 dp->dtdo_buf[pc] = DIF_INSTR_LOAD(op + 8418 DIF_OP_RLDSB - DIF_OP_LDSB, r1, rd); 8419 break; 8420 case DIF_OP_RLDSB: 8421 case DIF_OP_RLDSH: 8422 case DIF_OP_RLDSW: 8423 case DIF_OP_RLDUB: 8424 case DIF_OP_RLDUH: 8425 case DIF_OP_RLDUW: 8426 case DIF_OP_RLDX: 8427 if (r1 >= nregs) 8428 err += efunc(pc, "invalid register %u\n", r1); 8429 if (r2 != 0) 8430 err += efunc(pc, "non-zero reserved bits\n"); 8431 if (rd >= nregs) 8432 err += efunc(pc, "invalid register %u\n", rd); 8433 if (rd == 0) 8434 err += efunc(pc, "cannot write to %r0\n"); 8435 break; 8436 case DIF_OP_ULDSB: 8437 case DIF_OP_ULDSH: 8438 case DIF_OP_ULDSW: 8439 case DIF_OP_ULDUB: 8440 case DIF_OP_ULDUH: 8441 case DIF_OP_ULDUW: 8442 case DIF_OP_ULDX: 8443 if (r1 >= nregs) 8444 err += efunc(pc, "invalid register %u\n", r1); 8445 if (r2 != 0) 8446 err += efunc(pc, "non-zero reserved bits\n"); 8447 if (rd >= nregs) 8448 err += efunc(pc, "invalid register %u\n", rd); 8449 if (rd == 0) 8450 err += efunc(pc, "cannot write to %r0\n"); 8451 break; 8452 case DIF_OP_STB: 8453 case DIF_OP_STH: 8454 case DIF_OP_STW: 8455 case DIF_OP_STX: 8456 if (r1 >= nregs) 8457 err += efunc(pc, "invalid register %u\n", r1); 8458 if (r2 != 0) 8459 err += efunc(pc, "non-zero reserved bits\n"); 8460 if (rd >= nregs) 8461 err += efunc(pc, "invalid register %u\n", rd); 8462 if (rd == 0) 8463 err += efunc(pc, "cannot write to 0 address\n"); 8464 break; 8465 case DIF_OP_CMP: 8466 case DIF_OP_SCMP: 8467 if (r1 >= nregs) 8468 err += efunc(pc, "invalid register %u\n", r1); 8469 if (r2 >= nregs) 8470 err += efunc(pc, "invalid register %u\n", r2); 8471 if (rd != 0) 8472 err += efunc(pc, "non-zero reserved bits\n"); 8473 break; 8474 case DIF_OP_TST: 8475 if (r1 >= nregs) 8476 err += efunc(pc, "invalid register %u\n", r1); 8477 if (r2 != 0 || rd != 0) 8478 err += efunc(pc, "non-zero reserved bits\n"); 8479 break; 8480 case DIF_OP_BA: 8481 case DIF_OP_BE: 8482 case DIF_OP_BNE: 8483 case DIF_OP_BG: 8484 case DIF_OP_BGU: 8485 case DIF_OP_BGE: 8486 case DIF_OP_BGEU: 8487 case DIF_OP_BL: 8488 case DIF_OP_BLU: 8489 case DIF_OP_BLE: 8490 case DIF_OP_BLEU: 8491 if (label >= dp->dtdo_len) { 8492 err += efunc(pc, "invalid branch target %u\n", 8493 label); 8494 } 8495 if (label <= pc) { 8496 err += efunc(pc, "backward branch to %u\n", 8497 label); 8498 } 8499 break; 8500 case DIF_OP_RET: 8501 if (r1 != 0 || r2 != 0) 8502 err += efunc(pc, "non-zero reserved bits\n"); 8503 if (rd >= nregs) 8504 err += efunc(pc, "invalid register %u\n", rd); 8505 break; 8506 case DIF_OP_NOP: 8507 case DIF_OP_POPTS: 8508 case DIF_OP_FLUSHTS: 8509 if (r1 != 0 || r2 != 0 || rd != 0) 8510 err += efunc(pc, "non-zero reserved bits\n"); 8511 break; 8512 case DIF_OP_SETX: 8513 if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) { 8514 err += efunc(pc, "invalid integer ref %u\n", 8515 DIF_INSTR_INTEGER(instr)); 8516 } 8517 if (rd >= nregs) 8518 err += efunc(pc, "invalid register %u\n", rd); 8519 if (rd == 0) 8520 err += efunc(pc, "cannot write to %r0\n"); 8521 break; 8522 case DIF_OP_SETS: 8523 if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) { 8524 err += efunc(pc, "invalid string ref %u\n", 8525 DIF_INSTR_STRING(instr)); 8526 } 8527 if (rd >= nregs) 8528 err += efunc(pc, "invalid register %u\n", rd); 8529 if (rd == 0) 8530 err += efunc(pc, "cannot write to %r0\n"); 8531 break; 8532 case DIF_OP_LDGA: 8533 case DIF_OP_LDTA: 8534 if (r1 > DIF_VAR_ARRAY_MAX) 8535 err += efunc(pc, "invalid array %u\n", r1); 8536 if (r2 >= nregs) 8537 err += efunc(pc, "invalid register %u\n", r2); 8538 if (rd >= nregs) 8539 err += efunc(pc, "invalid register %u\n", rd); 8540 if (rd == 0) 8541 err += efunc(pc, "cannot write to %r0\n"); 8542 break; 8543 case DIF_OP_LDGS: 8544 case DIF_OP_LDTS: 8545 case DIF_OP_LDLS: 8546 case DIF_OP_LDGAA: 8547 case DIF_OP_LDTAA: 8548 if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX) 8549 err += efunc(pc, "invalid variable %u\n", v); 8550 if (rd >= nregs) 8551 err += efunc(pc, "invalid register %u\n", rd); 8552 if (rd == 0) 8553 err += efunc(pc, "cannot write to %r0\n"); 8554 break; 8555 case DIF_OP_STGS: 8556 case DIF_OP_STTS: 8557 case DIF_OP_STLS: 8558 case DIF_OP_STGAA: 8559 case DIF_OP_STTAA: 8560 if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX) 8561 err += efunc(pc, "invalid variable %u\n", v); 8562 if (rs >= nregs) 8563 err += efunc(pc, "invalid register %u\n", rd); 8564 break; 8565 case DIF_OP_CALL: 8566 if (subr > DIF_SUBR_MAX) 8567 err += efunc(pc, "invalid subr %u\n", subr); 8568 if (rd >= nregs) 8569 err += efunc(pc, "invalid register %u\n", rd); 8570 if (rd == 0) 8571 err += efunc(pc, "cannot write to %r0\n"); 8572 8573 if (subr == DIF_SUBR_COPYOUT || 8574 subr == DIF_SUBR_COPYOUTSTR) { 8575 dp->dtdo_destructive = 1; 8576 } 8577 break; 8578 case DIF_OP_PUSHTR: 8579 if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF) 8580 err += efunc(pc, "invalid ref type %u\n", type); 8581 if (r2 >= nregs) 8582 err += efunc(pc, "invalid register %u\n", r2); 8583 if (rs >= nregs) 8584 err += efunc(pc, "invalid register %u\n", rs); 8585 break; 8586 case DIF_OP_PUSHTV: 8587 if (type != DIF_TYPE_CTF) 8588 err += efunc(pc, "invalid val type %u\n", type); 8589 if (r2 >= nregs) 8590 err += efunc(pc, "invalid register %u\n", r2); 8591 if (rs >= nregs) 8592 err += efunc(pc, "invalid register %u\n", rs); 8593 break; 8594 default: 8595 err += efunc(pc, "invalid opcode %u\n", 8596 DIF_INSTR_OP(instr)); 8597 } 8598 } 8599 8600 if (dp->dtdo_len != 0 && 8601 DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) { 8602 err += efunc(dp->dtdo_len - 1, 8603 "expected 'ret' as last DIF instruction\n"); 8604 } 8605 8606 if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) { 8607 /* 8608 * If we're not returning by reference, the size must be either 8609 * 0 or the size of one of the base types. 8610 */ 8611 switch (dp->dtdo_rtype.dtdt_size) { 8612 case 0: 8613 case sizeof (uint8_t): 8614 case sizeof (uint16_t): 8615 case sizeof (uint32_t): 8616 case sizeof (uint64_t): 8617 break; 8618 8619 default: 8620 err += efunc(dp->dtdo_len - 1, "bad return size"); 8621 } 8622 } 8623 8624 for (i = 0; i < dp->dtdo_varlen && err == 0; i++) { 8625 dtrace_difv_t *v = &dp->dtdo_vartab[i], *existing = NULL; 8626 dtrace_diftype_t *vt, *et; 8627 uint_t id, ndx; 8628 8629 if (v->dtdv_scope != DIFV_SCOPE_GLOBAL && 8630 v->dtdv_scope != DIFV_SCOPE_THREAD && 8631 v->dtdv_scope != DIFV_SCOPE_LOCAL) { 8632 err += efunc(i, "unrecognized variable scope %d\n", 8633 v->dtdv_scope); 8634 break; 8635 } 8636 8637 if (v->dtdv_kind != DIFV_KIND_ARRAY && 8638 v->dtdv_kind != DIFV_KIND_SCALAR) { 8639 err += efunc(i, "unrecognized variable type %d\n", 8640 v->dtdv_kind); 8641 break; 8642 } 8643 8644 if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) { 8645 err += efunc(i, "%d exceeds variable id limit\n", id); 8646 break; 8647 } 8648 8649 if (id < DIF_VAR_OTHER_UBASE) 8650 continue; 8651 8652 /* 8653 * For user-defined variables, we need to check that this 8654 * definition is identical to any previous definition that we 8655 * encountered. 8656 */ 8657 ndx = id - DIF_VAR_OTHER_UBASE; 8658 8659 switch (v->dtdv_scope) { 8660 case DIFV_SCOPE_GLOBAL: 8661 if (ndx < vstate->dtvs_nglobals) { 8662 dtrace_statvar_t *svar; 8663 8664 if ((svar = vstate->dtvs_globals[ndx]) != NULL) 8665 existing = &svar->dtsv_var; 8666 } 8667 8668 break; 8669 8670 case DIFV_SCOPE_THREAD: 8671 if (ndx < vstate->dtvs_ntlocals) 8672 existing = &vstate->dtvs_tlocals[ndx]; 8673 break; 8674 8675 case DIFV_SCOPE_LOCAL: 8676 if (ndx < vstate->dtvs_nlocals) { 8677 dtrace_statvar_t *svar; 8678 8679 if ((svar = vstate->dtvs_locals[ndx]) != NULL) 8680 existing = &svar->dtsv_var; 8681 } 8682 8683 break; 8684 } 8685 8686 vt = &v->dtdv_type; 8687 8688 if (vt->dtdt_flags & DIF_TF_BYREF) { 8689 if (vt->dtdt_size == 0) { 8690 err += efunc(i, "zero-sized variable\n"); 8691 break; 8692 } 8693 8694 if (v->dtdv_scope == DIFV_SCOPE_GLOBAL && 8695 vt->dtdt_size > dtrace_global_maxsize) { 8696 err += efunc(i, "oversized by-ref global\n"); 8697 break; 8698 } 8699 } 8700 8701 if (existing == NULL || existing->dtdv_id == 0) 8702 continue; 8703 8704 ASSERT(existing->dtdv_id == v->dtdv_id); 8705 ASSERT(existing->dtdv_scope == v->dtdv_scope); 8706 8707 if (existing->dtdv_kind != v->dtdv_kind) 8708 err += efunc(i, "%d changed variable kind\n", id); 8709 8710 et = &existing->dtdv_type; 8711 8712 if (vt->dtdt_flags != et->dtdt_flags) { 8713 err += efunc(i, "%d changed variable type flags\n", id); 8714 break; 8715 } 8716 8717 if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) { 8718 err += efunc(i, "%d changed variable type size\n", id); 8719 break; 8720 } 8721 } 8722 8723 return (err); 8724} 8725 8726#if defined(sun) 8727/* 8728 * Validate a DTrace DIF object that it is to be used as a helper. Helpers 8729 * are much more constrained than normal DIFOs. Specifically, they may 8730 * not: 8731 * 8732 * 1. Make calls to subroutines other than copyin(), copyinstr() or 8733 * miscellaneous string routines 8734 * 2. Access DTrace variables other than the args[] array, and the 8735 * curthread, pid, ppid, tid, execname, zonename, uid and gid variables. 8736 * 3. Have thread-local variables. 8737 * 4. Have dynamic variables. 8738 */ 8739static int 8740dtrace_difo_validate_helper(dtrace_difo_t *dp) 8741{ 8742 int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; 8743 int err = 0; 8744 uint_t pc; 8745 8746 for (pc = 0; pc < dp->dtdo_len; pc++) { 8747 dif_instr_t instr = dp->dtdo_buf[pc]; 8748 8749 uint_t v = DIF_INSTR_VAR(instr); 8750 uint_t subr = DIF_INSTR_SUBR(instr); 8751 uint_t op = DIF_INSTR_OP(instr); 8752 8753 switch (op) { 8754 case DIF_OP_OR: 8755 case DIF_OP_XOR: 8756 case DIF_OP_AND: 8757 case DIF_OP_SLL: 8758 case DIF_OP_SRL: 8759 case DIF_OP_SRA: 8760 case DIF_OP_SUB: 8761 case DIF_OP_ADD: 8762 case DIF_OP_MUL: 8763 case DIF_OP_SDIV: 8764 case DIF_OP_UDIV: 8765 case DIF_OP_SREM: 8766 case DIF_OP_UREM: 8767 case DIF_OP_COPYS: 8768 case DIF_OP_NOT: 8769 case DIF_OP_MOV: 8770 case DIF_OP_RLDSB: 8771 case DIF_OP_RLDSH: 8772 case DIF_OP_RLDSW: 8773 case DIF_OP_RLDUB: 8774 case DIF_OP_RLDUH: 8775 case DIF_OP_RLDUW: 8776 case DIF_OP_RLDX: 8777 case DIF_OP_ULDSB: 8778 case DIF_OP_ULDSH: 8779 case DIF_OP_ULDSW: 8780 case DIF_OP_ULDUB: 8781 case DIF_OP_ULDUH: 8782 case DIF_OP_ULDUW: 8783 case DIF_OP_ULDX: 8784 case DIF_OP_STB: 8785 case DIF_OP_STH: 8786 case DIF_OP_STW: 8787 case DIF_OP_STX: 8788 case DIF_OP_ALLOCS: 8789 case DIF_OP_CMP: 8790 case DIF_OP_SCMP: 8791 case DIF_OP_TST: 8792 case DIF_OP_BA: 8793 case DIF_OP_BE: 8794 case DIF_OP_BNE: 8795 case DIF_OP_BG: 8796 case DIF_OP_BGU: 8797 case DIF_OP_BGE: 8798 case DIF_OP_BGEU: 8799 case DIF_OP_BL: 8800 case DIF_OP_BLU: 8801 case DIF_OP_BLE: 8802 case DIF_OP_BLEU: 8803 case DIF_OP_RET: 8804 case DIF_OP_NOP: 8805 case DIF_OP_POPTS: 8806 case DIF_OP_FLUSHTS: 8807 case DIF_OP_SETX: 8808 case DIF_OP_SETS: 8809 case DIF_OP_LDGA: 8810 case DIF_OP_LDLS: 8811 case DIF_OP_STGS: 8812 case DIF_OP_STLS: 8813 case DIF_OP_PUSHTR: 8814 case DIF_OP_PUSHTV: 8815 break; 8816 8817 case DIF_OP_LDGS: 8818 if (v >= DIF_VAR_OTHER_UBASE) 8819 break; 8820 8821 if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) 8822 break; 8823 8824 if (v == DIF_VAR_CURTHREAD || v == DIF_VAR_PID || 8825 v == DIF_VAR_PPID || v == DIF_VAR_TID || 8826 v == DIF_VAR_EXECARGS || 8827 v == DIF_VAR_EXECNAME || v == DIF_VAR_ZONENAME || 8828 v == DIF_VAR_UID || v == DIF_VAR_GID) 8829 break; 8830 8831 err += efunc(pc, "illegal variable %u\n", v); 8832 break; 8833 8834 case DIF_OP_LDTA: 8835 case DIF_OP_LDTS: 8836 case DIF_OP_LDGAA: 8837 case DIF_OP_LDTAA: 8838 err += efunc(pc, "illegal dynamic variable load\n"); 8839 break; 8840 8841 case DIF_OP_STTS: 8842 case DIF_OP_STGAA: 8843 case DIF_OP_STTAA: 8844 err += efunc(pc, "illegal dynamic variable store\n"); 8845 break; 8846 8847 case DIF_OP_CALL: 8848 if (subr == DIF_SUBR_ALLOCA || 8849 subr == DIF_SUBR_BCOPY || 8850 subr == DIF_SUBR_COPYIN || 8851 subr == DIF_SUBR_COPYINTO || 8852 subr == DIF_SUBR_COPYINSTR || 8853 subr == DIF_SUBR_INDEX || 8854 subr == DIF_SUBR_INET_NTOA || 8855 subr == DIF_SUBR_INET_NTOA6 || 8856 subr == DIF_SUBR_INET_NTOP || 8857 subr == DIF_SUBR_LLTOSTR || 8858 subr == DIF_SUBR_RINDEX || 8859 subr == DIF_SUBR_STRCHR || 8860 subr == DIF_SUBR_STRJOIN || 8861 subr == DIF_SUBR_STRRCHR || 8862 subr == DIF_SUBR_STRSTR || 8863 subr == DIF_SUBR_HTONS || 8864 subr == DIF_SUBR_HTONL || 8865 subr == DIF_SUBR_HTONLL || 8866 subr == DIF_SUBR_NTOHS || 8867 subr == DIF_SUBR_NTOHL || 8868 subr == DIF_SUBR_NTOHLL || 8869 subr == DIF_SUBR_MEMREF || 8870 subr == DIF_SUBR_TYPEREF) 8871 break; 8872 8873 err += efunc(pc, "invalid subr %u\n", subr); 8874 break; 8875 8876 default: 8877 err += efunc(pc, "invalid opcode %u\n", 8878 DIF_INSTR_OP(instr)); 8879 } 8880 } 8881 8882 return (err); 8883} 8884#endif 8885 8886/* 8887 * Returns 1 if the expression in the DIF object can be cached on a per-thread 8888 * basis; 0 if not. 8889 */ 8890static int 8891dtrace_difo_cacheable(dtrace_difo_t *dp) 8892{ 8893 int i; 8894 8895 if (dp == NULL) 8896 return (0); 8897 8898 for (i = 0; i < dp->dtdo_varlen; i++) { 8899 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 8900 8901 if (v->dtdv_scope != DIFV_SCOPE_GLOBAL) 8902 continue; 8903 8904 switch (v->dtdv_id) { 8905 case DIF_VAR_CURTHREAD: 8906 case DIF_VAR_PID: 8907 case DIF_VAR_TID: 8908 case DIF_VAR_EXECARGS: 8909 case DIF_VAR_EXECNAME: 8910 case DIF_VAR_ZONENAME: 8911 break; 8912 8913 default: 8914 return (0); 8915 } 8916 } 8917 8918 /* 8919 * This DIF object may be cacheable. Now we need to look for any 8920 * array loading instructions, any memory loading instructions, or 8921 * any stores to thread-local variables. 8922 */ 8923 for (i = 0; i < dp->dtdo_len; i++) { 8924 uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]); 8925 8926 if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) || 8927 (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) || 8928 (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) || 8929 op == DIF_OP_LDGA || op == DIF_OP_STTS) 8930 return (0); 8931 } 8932 8933 return (1); 8934} 8935 8936static void 8937dtrace_difo_hold(dtrace_difo_t *dp) 8938{ 8939 int i; 8940 8941 ASSERT(MUTEX_HELD(&dtrace_lock)); 8942 8943 dp->dtdo_refcnt++; 8944 ASSERT(dp->dtdo_refcnt != 0); 8945 8946 /* 8947 * We need to check this DIF object for references to the variable 8948 * DIF_VAR_VTIMESTAMP. 8949 */ 8950 for (i = 0; i < dp->dtdo_varlen; i++) { 8951 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 8952 8953 if (v->dtdv_id != DIF_VAR_VTIMESTAMP) 8954 continue; 8955 8956 if (dtrace_vtime_references++ == 0) 8957 dtrace_vtime_enable(); 8958 } 8959} 8960 8961/* 8962 * This routine calculates the dynamic variable chunksize for a given DIF 8963 * object. The calculation is not fool-proof, and can probably be tricked by 8964 * malicious DIF -- but it works for all compiler-generated DIF. Because this 8965 * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail 8966 * if a dynamic variable size exceeds the chunksize. 8967 */ 8968static void 8969dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 8970{ 8971 uint64_t sval = 0; 8972 dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */ 8973 const dif_instr_t *text = dp->dtdo_buf; 8974 uint_t pc, srd = 0; 8975 uint_t ttop = 0; 8976 size_t size, ksize; 8977 uint_t id, i; 8978 8979 for (pc = 0; pc < dp->dtdo_len; pc++) { 8980 dif_instr_t instr = text[pc]; 8981 uint_t op = DIF_INSTR_OP(instr); 8982 uint_t rd = DIF_INSTR_RD(instr); 8983 uint_t r1 = DIF_INSTR_R1(instr); 8984 uint_t nkeys = 0; 8985 uchar_t scope = 0; 8986 8987 dtrace_key_t *key = tupregs; 8988 8989 switch (op) { 8990 case DIF_OP_SETX: 8991 sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)]; 8992 srd = rd; 8993 continue; 8994 8995 case DIF_OP_STTS: 8996 key = &tupregs[DIF_DTR_NREGS]; 8997 key[0].dttk_size = 0; 8998 key[1].dttk_size = 0; 8999 nkeys = 2; 9000 scope = DIFV_SCOPE_THREAD; 9001 break; 9002 9003 case DIF_OP_STGAA: 9004 case DIF_OP_STTAA: 9005 nkeys = ttop; 9006 9007 if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) 9008 key[nkeys++].dttk_size = 0; 9009 9010 key[nkeys++].dttk_size = 0; 9011 9012 if (op == DIF_OP_STTAA) { 9013 scope = DIFV_SCOPE_THREAD; 9014 } else { 9015 scope = DIFV_SCOPE_GLOBAL; 9016 } 9017 9018 break; 9019 9020 case DIF_OP_PUSHTR: 9021 if (ttop == DIF_DTR_NREGS) 9022 return; 9023 9024 if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) { 9025 /* 9026 * If the register for the size of the "pushtr" 9027 * is %r0 (or the value is 0) and the type is 9028 * a string, we'll use the system-wide default 9029 * string size. 9030 */ 9031 tupregs[ttop++].dttk_size = 9032 dtrace_strsize_default; 9033 } else { 9034 if (srd == 0) 9035 return; 9036 9037 tupregs[ttop++].dttk_size = sval; 9038 } 9039 9040 break; 9041 9042 case DIF_OP_PUSHTV: 9043 if (ttop == DIF_DTR_NREGS) 9044 return; 9045 9046 tupregs[ttop++].dttk_size = 0; 9047 break; 9048 9049 case DIF_OP_FLUSHTS: 9050 ttop = 0; 9051 break; 9052 9053 case DIF_OP_POPTS: 9054 if (ttop != 0) 9055 ttop--; 9056 break; 9057 } 9058 9059 sval = 0; 9060 srd = 0; 9061 9062 if (nkeys == 0) 9063 continue; 9064 9065 /* 9066 * We have a dynamic variable allocation; calculate its size. 9067 */ 9068 for (ksize = 0, i = 0; i < nkeys; i++) 9069 ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t)); 9070 9071 size = sizeof (dtrace_dynvar_t); 9072 size += sizeof (dtrace_key_t) * (nkeys - 1); 9073 size += ksize; 9074 9075 /* 9076 * Now we need to determine the size of the stored data. 9077 */ 9078 id = DIF_INSTR_VAR(instr); 9079 9080 for (i = 0; i < dp->dtdo_varlen; i++) { 9081 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9082 9083 if (v->dtdv_id == id && v->dtdv_scope == scope) { 9084 size += v->dtdv_type.dtdt_size; 9085 break; 9086 } 9087 } 9088 9089 if (i == dp->dtdo_varlen) 9090 return; 9091 9092 /* 9093 * We have the size. If this is larger than the chunk size 9094 * for our dynamic variable state, reset the chunk size. 9095 */ 9096 size = P2ROUNDUP(size, sizeof (uint64_t)); 9097 9098 if (size > vstate->dtvs_dynvars.dtds_chunksize) 9099 vstate->dtvs_dynvars.dtds_chunksize = size; 9100 } 9101} 9102 9103static void 9104dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9105{ 9106 int i, oldsvars, osz, nsz, otlocals, ntlocals; 9107 uint_t id; 9108 9109 ASSERT(MUTEX_HELD(&dtrace_lock)); 9110 ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0); 9111 9112 for (i = 0; i < dp->dtdo_varlen; i++) { 9113 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9114 dtrace_statvar_t *svar, ***svarp = NULL; 9115 size_t dsize = 0; 9116 uint8_t scope = v->dtdv_scope; 9117 int *np = NULL; 9118 9119 if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) 9120 continue; 9121 9122 id -= DIF_VAR_OTHER_UBASE; 9123 9124 switch (scope) { 9125 case DIFV_SCOPE_THREAD: 9126 while (id >= (otlocals = vstate->dtvs_ntlocals)) { 9127 dtrace_difv_t *tlocals; 9128 9129 if ((ntlocals = (otlocals << 1)) == 0) 9130 ntlocals = 1; 9131 9132 osz = otlocals * sizeof (dtrace_difv_t); 9133 nsz = ntlocals * sizeof (dtrace_difv_t); 9134 9135 tlocals = kmem_zalloc(nsz, KM_SLEEP); 9136 9137 if (osz != 0) { 9138 bcopy(vstate->dtvs_tlocals, 9139 tlocals, osz); 9140 kmem_free(vstate->dtvs_tlocals, osz); 9141 } 9142 9143 vstate->dtvs_tlocals = tlocals; 9144 vstate->dtvs_ntlocals = ntlocals; 9145 } 9146 9147 vstate->dtvs_tlocals[id] = *v; 9148 continue; 9149 9150 case DIFV_SCOPE_LOCAL: 9151 np = &vstate->dtvs_nlocals; 9152 svarp = &vstate->dtvs_locals; 9153 9154 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) 9155 dsize = NCPU * (v->dtdv_type.dtdt_size + 9156 sizeof (uint64_t)); 9157 else 9158 dsize = NCPU * sizeof (uint64_t); 9159 9160 break; 9161 9162 case DIFV_SCOPE_GLOBAL: 9163 np = &vstate->dtvs_nglobals; 9164 svarp = &vstate->dtvs_globals; 9165 9166 if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) 9167 dsize = v->dtdv_type.dtdt_size + 9168 sizeof (uint64_t); 9169 9170 break; 9171 9172 default: 9173 ASSERT(0); 9174 } 9175 9176 while (id >= (oldsvars = *np)) { 9177 dtrace_statvar_t **statics; 9178 int newsvars, oldsize, newsize; 9179 9180 if ((newsvars = (oldsvars << 1)) == 0) 9181 newsvars = 1; 9182 9183 oldsize = oldsvars * sizeof (dtrace_statvar_t *); 9184 newsize = newsvars * sizeof (dtrace_statvar_t *); 9185 9186 statics = kmem_zalloc(newsize, KM_SLEEP); 9187 9188 if (oldsize != 0) { 9189 bcopy(*svarp, statics, oldsize); 9190 kmem_free(*svarp, oldsize); 9191 } 9192 9193 *svarp = statics; 9194 *np = newsvars; 9195 } 9196 9197 if ((svar = (*svarp)[id]) == NULL) { 9198 svar = kmem_zalloc(sizeof (dtrace_statvar_t), KM_SLEEP); 9199 svar->dtsv_var = *v; 9200 9201 if ((svar->dtsv_size = dsize) != 0) { 9202 svar->dtsv_data = (uint64_t)(uintptr_t) 9203 kmem_zalloc(dsize, KM_SLEEP); 9204 } 9205 9206 (*svarp)[id] = svar; 9207 } 9208 9209 svar->dtsv_refcnt++; 9210 } 9211 9212 dtrace_difo_chunksize(dp, vstate); 9213 dtrace_difo_hold(dp); 9214} 9215 9216#if defined(sun) 9217static dtrace_difo_t * 9218dtrace_difo_duplicate(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9219{ 9220 dtrace_difo_t *new; 9221 size_t sz; 9222 9223 ASSERT(dp->dtdo_buf != NULL); 9224 ASSERT(dp->dtdo_refcnt != 0); 9225 9226 new = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP); 9227 9228 ASSERT(dp->dtdo_buf != NULL); 9229 sz = dp->dtdo_len * sizeof (dif_instr_t); 9230 new->dtdo_buf = kmem_alloc(sz, KM_SLEEP); 9231 bcopy(dp->dtdo_buf, new->dtdo_buf, sz); 9232 new->dtdo_len = dp->dtdo_len; 9233 9234 if (dp->dtdo_strtab != NULL) { 9235 ASSERT(dp->dtdo_strlen != 0); 9236 new->dtdo_strtab = kmem_alloc(dp->dtdo_strlen, KM_SLEEP); 9237 bcopy(dp->dtdo_strtab, new->dtdo_strtab, dp->dtdo_strlen); 9238 new->dtdo_strlen = dp->dtdo_strlen; 9239 } 9240 9241 if (dp->dtdo_inttab != NULL) { 9242 ASSERT(dp->dtdo_intlen != 0); 9243 sz = dp->dtdo_intlen * sizeof (uint64_t); 9244 new->dtdo_inttab = kmem_alloc(sz, KM_SLEEP); 9245 bcopy(dp->dtdo_inttab, new->dtdo_inttab, sz); 9246 new->dtdo_intlen = dp->dtdo_intlen; 9247 } 9248 9249 if (dp->dtdo_vartab != NULL) { 9250 ASSERT(dp->dtdo_varlen != 0); 9251 sz = dp->dtdo_varlen * sizeof (dtrace_difv_t); 9252 new->dtdo_vartab = kmem_alloc(sz, KM_SLEEP); 9253 bcopy(dp->dtdo_vartab, new->dtdo_vartab, sz); 9254 new->dtdo_varlen = dp->dtdo_varlen; 9255 } 9256 9257 dtrace_difo_init(new, vstate); 9258 return (new); 9259} 9260#endif 9261 9262static void 9263dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9264{ 9265 int i; 9266 9267 ASSERT(dp->dtdo_refcnt == 0); 9268 9269 for (i = 0; i < dp->dtdo_varlen; i++) { 9270 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9271 dtrace_statvar_t *svar, **svarp = NULL; 9272 uint_t id; 9273 uint8_t scope = v->dtdv_scope; 9274 int *np = NULL; 9275 9276 switch (scope) { 9277 case DIFV_SCOPE_THREAD: 9278 continue; 9279 9280 case DIFV_SCOPE_LOCAL: 9281 np = &vstate->dtvs_nlocals; 9282 svarp = vstate->dtvs_locals; 9283 break; 9284 9285 case DIFV_SCOPE_GLOBAL: 9286 np = &vstate->dtvs_nglobals; 9287 svarp = vstate->dtvs_globals; 9288 break; 9289 9290 default: 9291 ASSERT(0); 9292 } 9293 9294 if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) 9295 continue; 9296 9297 id -= DIF_VAR_OTHER_UBASE; 9298 ASSERT(id < *np); 9299 9300 svar = svarp[id]; 9301 ASSERT(svar != NULL); 9302 ASSERT(svar->dtsv_refcnt > 0); 9303 9304 if (--svar->dtsv_refcnt > 0) 9305 continue; 9306 9307 if (svar->dtsv_size != 0) { 9308 ASSERT(svar->dtsv_data != 0); 9309 kmem_free((void *)(uintptr_t)svar->dtsv_data, 9310 svar->dtsv_size); 9311 } 9312 9313 kmem_free(svar, sizeof (dtrace_statvar_t)); 9314 svarp[id] = NULL; 9315 } 9316 9317 if (dp->dtdo_buf != NULL) 9318 kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t)); 9319 if (dp->dtdo_inttab != NULL) 9320 kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t)); 9321 if (dp->dtdo_strtab != NULL) 9322 kmem_free(dp->dtdo_strtab, dp->dtdo_strlen); 9323 if (dp->dtdo_vartab != NULL) 9324 kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t)); 9325 9326 kmem_free(dp, sizeof (dtrace_difo_t)); 9327} 9328 9329static void 9330dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate) 9331{ 9332 int i; 9333 9334 ASSERT(MUTEX_HELD(&dtrace_lock)); 9335 ASSERT(dp->dtdo_refcnt != 0); 9336 9337 for (i = 0; i < dp->dtdo_varlen; i++) { 9338 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 9339 9340 if (v->dtdv_id != DIF_VAR_VTIMESTAMP) 9341 continue; 9342 9343 ASSERT(dtrace_vtime_references > 0); 9344 if (--dtrace_vtime_references == 0) 9345 dtrace_vtime_disable(); 9346 } 9347 9348 if (--dp->dtdo_refcnt == 0) 9349 dtrace_difo_destroy(dp, vstate); 9350} 9351 9352/* 9353 * DTrace Format Functions 9354 */ 9355static uint16_t 9356dtrace_format_add(dtrace_state_t *state, char *str) 9357{ 9358 char *fmt, **new; 9359 uint16_t ndx, len = strlen(str) + 1; 9360 9361 fmt = kmem_zalloc(len, KM_SLEEP); 9362 bcopy(str, fmt, len); 9363 9364 for (ndx = 0; ndx < state->dts_nformats; ndx++) { 9365 if (state->dts_formats[ndx] == NULL) { 9366 state->dts_formats[ndx] = fmt; 9367 return (ndx + 1); 9368 } 9369 } 9370 9371 if (state->dts_nformats == USHRT_MAX) { 9372 /* 9373 * This is only likely if a denial-of-service attack is being 9374 * attempted. As such, it's okay to fail silently here. 9375 */ 9376 kmem_free(fmt, len); 9377 return (0); 9378 } 9379 9380 /* 9381 * For simplicity, we always resize the formats array to be exactly the 9382 * number of formats. 9383 */ 9384 ndx = state->dts_nformats++; 9385 new = kmem_alloc((ndx + 1) * sizeof (char *), KM_SLEEP); 9386 9387 if (state->dts_formats != NULL) { 9388 ASSERT(ndx != 0); 9389 bcopy(state->dts_formats, new, ndx * sizeof (char *)); 9390 kmem_free(state->dts_formats, ndx * sizeof (char *)); 9391 } 9392 9393 state->dts_formats = new; 9394 state->dts_formats[ndx] = fmt; 9395 9396 return (ndx + 1); 9397} 9398 9399static void 9400dtrace_format_remove(dtrace_state_t *state, uint16_t format) 9401{ 9402 char *fmt; 9403 9404 ASSERT(state->dts_formats != NULL); 9405 ASSERT(format <= state->dts_nformats); 9406 ASSERT(state->dts_formats[format - 1] != NULL); 9407 9408 fmt = state->dts_formats[format - 1]; 9409 kmem_free(fmt, strlen(fmt) + 1); 9410 state->dts_formats[format - 1] = NULL; 9411} 9412 9413static void 9414dtrace_format_destroy(dtrace_state_t *state) 9415{ 9416 int i; 9417 9418 if (state->dts_nformats == 0) { 9419 ASSERT(state->dts_formats == NULL); 9420 return; 9421 } 9422 9423 ASSERT(state->dts_formats != NULL); 9424 9425 for (i = 0; i < state->dts_nformats; i++) { 9426 char *fmt = state->dts_formats[i]; 9427 9428 if (fmt == NULL) 9429 continue; 9430 9431 kmem_free(fmt, strlen(fmt) + 1); 9432 } 9433 9434 kmem_free(state->dts_formats, state->dts_nformats * sizeof (char *)); 9435 state->dts_nformats = 0; 9436 state->dts_formats = NULL; 9437} 9438 9439/* 9440 * DTrace Predicate Functions 9441 */ 9442static dtrace_predicate_t * 9443dtrace_predicate_create(dtrace_difo_t *dp) 9444{ 9445 dtrace_predicate_t *pred; 9446 9447 ASSERT(MUTEX_HELD(&dtrace_lock)); 9448 ASSERT(dp->dtdo_refcnt != 0); 9449 9450 pred = kmem_zalloc(sizeof (dtrace_predicate_t), KM_SLEEP); 9451 pred->dtp_difo = dp; 9452 pred->dtp_refcnt = 1; 9453 9454 if (!dtrace_difo_cacheable(dp)) 9455 return (pred); 9456 9457 if (dtrace_predcache_id == DTRACE_CACHEIDNONE) { 9458 /* 9459 * This is only theoretically possible -- we have had 2^32 9460 * cacheable predicates on this machine. We cannot allow any 9461 * more predicates to become cacheable: as unlikely as it is, 9462 * there may be a thread caching a (now stale) predicate cache 9463 * ID. (N.B.: the temptation is being successfully resisted to 9464 * have this cmn_err() "Holy shit -- we executed this code!") 9465 */ 9466 return (pred); 9467 } 9468 9469 pred->dtp_cacheid = dtrace_predcache_id++; 9470 9471 return (pred); 9472} 9473 9474static void 9475dtrace_predicate_hold(dtrace_predicate_t *pred) 9476{ 9477 ASSERT(MUTEX_HELD(&dtrace_lock)); 9478 ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0); 9479 ASSERT(pred->dtp_refcnt > 0); 9480 9481 pred->dtp_refcnt++; 9482} 9483 9484static void 9485dtrace_predicate_release(dtrace_predicate_t *pred, dtrace_vstate_t *vstate) 9486{ 9487 dtrace_difo_t *dp = pred->dtp_difo; 9488 9489 ASSERT(MUTEX_HELD(&dtrace_lock)); 9490 ASSERT(dp != NULL && dp->dtdo_refcnt != 0); 9491 ASSERT(pred->dtp_refcnt > 0); 9492 9493 if (--pred->dtp_refcnt == 0) { 9494 dtrace_difo_release(pred->dtp_difo, vstate); 9495 kmem_free(pred, sizeof (dtrace_predicate_t)); 9496 } 9497} 9498 9499/* 9500 * DTrace Action Description Functions 9501 */ 9502static dtrace_actdesc_t * 9503dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple, 9504 uint64_t uarg, uint64_t arg) 9505{ 9506 dtrace_actdesc_t *act; 9507 9508#if defined(sun) 9509 ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || (arg != NULL && 9510 arg >= KERNELBASE) || (arg == NULL && kind == DTRACEACT_PRINTA)); 9511#endif 9512 9513 act = kmem_zalloc(sizeof (dtrace_actdesc_t), KM_SLEEP); 9514 act->dtad_kind = kind; 9515 act->dtad_ntuple = ntuple; 9516 act->dtad_uarg = uarg; 9517 act->dtad_arg = arg; 9518 act->dtad_refcnt = 1; 9519 9520 return (act); 9521} 9522 9523static void 9524dtrace_actdesc_hold(dtrace_actdesc_t *act) 9525{ 9526 ASSERT(act->dtad_refcnt >= 1); 9527 act->dtad_refcnt++; 9528} 9529 9530static void 9531dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate) 9532{ 9533 dtrace_actkind_t kind = act->dtad_kind; 9534 dtrace_difo_t *dp; 9535 9536 ASSERT(act->dtad_refcnt >= 1); 9537 9538 if (--act->dtad_refcnt != 0) 9539 return; 9540 9541 if ((dp = act->dtad_difo) != NULL) 9542 dtrace_difo_release(dp, vstate); 9543 9544 if (DTRACEACT_ISPRINTFLIKE(kind)) { 9545 char *str = (char *)(uintptr_t)act->dtad_arg; 9546 9547#if defined(sun) 9548 ASSERT((str != NULL && (uintptr_t)str >= KERNELBASE) || 9549 (str == NULL && act->dtad_kind == DTRACEACT_PRINTA)); 9550#endif 9551 9552 if (str != NULL) 9553 kmem_free(str, strlen(str) + 1); 9554 } 9555 9556 kmem_free(act, sizeof (dtrace_actdesc_t)); 9557} 9558 9559/* 9560 * DTrace ECB Functions 9561 */ 9562static dtrace_ecb_t * 9563dtrace_ecb_add(dtrace_state_t *state, dtrace_probe_t *probe) 9564{ 9565 dtrace_ecb_t *ecb; 9566 dtrace_epid_t epid; 9567 9568 ASSERT(MUTEX_HELD(&dtrace_lock)); 9569 9570 ecb = kmem_zalloc(sizeof (dtrace_ecb_t), KM_SLEEP); 9571 ecb->dte_predicate = NULL; 9572 ecb->dte_probe = probe; 9573 9574 /* 9575 * The default size is the size of the default action: recording 9576 * the epid. 9577 */ 9578 ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t); 9579 ecb->dte_alignment = sizeof (dtrace_epid_t); 9580 9581 epid = state->dts_epid++; 9582 9583 if (epid - 1 >= state->dts_necbs) { 9584 dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs; 9585 int necbs = state->dts_necbs << 1; 9586 9587 ASSERT(epid == state->dts_necbs + 1); 9588 9589 if (necbs == 0) { 9590 ASSERT(oecbs == NULL); 9591 necbs = 1; 9592 } 9593 9594 ecbs = kmem_zalloc(necbs * sizeof (*ecbs), KM_SLEEP); 9595 9596 if (oecbs != NULL) 9597 bcopy(oecbs, ecbs, state->dts_necbs * sizeof (*ecbs)); 9598 9599 dtrace_membar_producer(); 9600 state->dts_ecbs = ecbs; 9601 9602 if (oecbs != NULL) { 9603 /* 9604 * If this state is active, we must dtrace_sync() 9605 * before we can free the old dts_ecbs array: we're 9606 * coming in hot, and there may be active ring 9607 * buffer processing (which indexes into the dts_ecbs 9608 * array) on another CPU. 9609 */ 9610 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) 9611 dtrace_sync(); 9612 9613 kmem_free(oecbs, state->dts_necbs * sizeof (*ecbs)); 9614 } 9615 9616 dtrace_membar_producer(); 9617 state->dts_necbs = necbs; 9618 } 9619 9620 ecb->dte_state = state; 9621 9622 ASSERT(state->dts_ecbs[epid - 1] == NULL); 9623 dtrace_membar_producer(); 9624 state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb; 9625 9626 return (ecb); 9627} 9628 9629static void 9630dtrace_ecb_enable(dtrace_ecb_t *ecb) 9631{ 9632 dtrace_probe_t *probe = ecb->dte_probe; 9633 9634 ASSERT(MUTEX_HELD(&cpu_lock)); 9635 ASSERT(MUTEX_HELD(&dtrace_lock)); 9636 ASSERT(ecb->dte_next == NULL); 9637 9638 if (probe == NULL) { 9639 /* 9640 * This is the NULL probe -- there's nothing to do. 9641 */ 9642 return; 9643 } 9644 9645 if (probe->dtpr_ecb == NULL) { 9646 dtrace_provider_t *prov = probe->dtpr_provider; 9647 9648 /* 9649 * We're the first ECB on this probe. 9650 */ 9651 probe->dtpr_ecb = probe->dtpr_ecb_last = ecb; 9652 9653 if (ecb->dte_predicate != NULL) 9654 probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid; 9655 9656 prov->dtpv_pops.dtps_enable(prov->dtpv_arg, 9657 probe->dtpr_id, probe->dtpr_arg); 9658 } else { 9659 /* 9660 * This probe is already active. Swing the last pointer to 9661 * point to the new ECB, and issue a dtrace_sync() to assure 9662 * that all CPUs have seen the change. 9663 */ 9664 ASSERT(probe->dtpr_ecb_last != NULL); 9665 probe->dtpr_ecb_last->dte_next = ecb; 9666 probe->dtpr_ecb_last = ecb; 9667 probe->dtpr_predcache = 0; 9668 9669 dtrace_sync(); 9670 } 9671} 9672 9673static void 9674dtrace_ecb_resize(dtrace_ecb_t *ecb) 9675{ 9676 uint32_t maxalign = sizeof (dtrace_epid_t); 9677 uint32_t align = sizeof (uint8_t), offs, diff; 9678 dtrace_action_t *act; 9679 int wastuple = 0; 9680 uint32_t aggbase = UINT32_MAX; 9681 dtrace_state_t *state = ecb->dte_state; 9682 9683 /* 9684 * If we record anything, we always record the epid. (And we always 9685 * record it first.) 9686 */ 9687 offs = sizeof (dtrace_epid_t); 9688 ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t); 9689 9690 for (act = ecb->dte_action; act != NULL; act = act->dta_next) { 9691 dtrace_recdesc_t *rec = &act->dta_rec; 9692 9693 if ((align = rec->dtrd_alignment) > maxalign) 9694 maxalign = align; 9695 9696 if (!wastuple && act->dta_intuple) { 9697 /* 9698 * This is the first record in a tuple. Align the 9699 * offset to be at offset 4 in an 8-byte aligned 9700 * block. 9701 */ 9702 diff = offs + sizeof (dtrace_aggid_t); 9703 9704 if ((diff = (diff & (sizeof (uint64_t) - 1)))) 9705 offs += sizeof (uint64_t) - diff; 9706 9707 aggbase = offs - sizeof (dtrace_aggid_t); 9708 ASSERT(!(aggbase & (sizeof (uint64_t) - 1))); 9709 } 9710 9711 /*LINTED*/ 9712 if (rec->dtrd_size != 0 && (diff = (offs & (align - 1)))) { 9713 /* 9714 * The current offset is not properly aligned; align it. 9715 */ 9716 offs += align - diff; 9717 } 9718 9719 rec->dtrd_offset = offs; 9720 9721 if (offs + rec->dtrd_size > ecb->dte_needed) { 9722 ecb->dte_needed = offs + rec->dtrd_size; 9723 9724 if (ecb->dte_needed > state->dts_needed) 9725 state->dts_needed = ecb->dte_needed; 9726 } 9727 9728 if (DTRACEACT_ISAGG(act->dta_kind)) { 9729 dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; 9730 dtrace_action_t *first = agg->dtag_first, *prev; 9731 9732 ASSERT(rec->dtrd_size != 0 && first != NULL); 9733 ASSERT(wastuple); 9734 ASSERT(aggbase != UINT32_MAX); 9735 9736 agg->dtag_base = aggbase; 9737 9738 while ((prev = first->dta_prev) != NULL && 9739 DTRACEACT_ISAGG(prev->dta_kind)) { 9740 agg = (dtrace_aggregation_t *)prev; 9741 first = agg->dtag_first; 9742 } 9743 9744 if (prev != NULL) { 9745 offs = prev->dta_rec.dtrd_offset + 9746 prev->dta_rec.dtrd_size; 9747 } else { 9748 offs = sizeof (dtrace_epid_t); 9749 } 9750 wastuple = 0; 9751 } else { 9752 if (!act->dta_intuple) 9753 ecb->dte_size = offs + rec->dtrd_size; 9754 9755 offs += rec->dtrd_size; 9756 } 9757 9758 wastuple = act->dta_intuple; 9759 } 9760 9761 if ((act = ecb->dte_action) != NULL && 9762 !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) && 9763 ecb->dte_size == sizeof (dtrace_epid_t)) { 9764 /* 9765 * If the size is still sizeof (dtrace_epid_t), then all 9766 * actions store no data; set the size to 0. 9767 */ 9768 ecb->dte_alignment = maxalign; 9769 ecb->dte_size = 0; 9770 9771 /* 9772 * If the needed space is still sizeof (dtrace_epid_t), then 9773 * all actions need no additional space; set the needed 9774 * size to 0. 9775 */ 9776 if (ecb->dte_needed == sizeof (dtrace_epid_t)) 9777 ecb->dte_needed = 0; 9778 9779 return; 9780 } 9781 9782 /* 9783 * Set our alignment, and make sure that the dte_size and dte_needed 9784 * are aligned to the size of an EPID. 9785 */ 9786 ecb->dte_alignment = maxalign; 9787 ecb->dte_size = (ecb->dte_size + (sizeof (dtrace_epid_t) - 1)) & 9788 ~(sizeof (dtrace_epid_t) - 1); 9789 ecb->dte_needed = (ecb->dte_needed + (sizeof (dtrace_epid_t) - 1)) & 9790 ~(sizeof (dtrace_epid_t) - 1); 9791 ASSERT(ecb->dte_size <= ecb->dte_needed); 9792} 9793 9794static dtrace_action_t * 9795dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) 9796{ 9797 dtrace_aggregation_t *agg; 9798 size_t size = sizeof (uint64_t); 9799 int ntuple = desc->dtad_ntuple; 9800 dtrace_action_t *act; 9801 dtrace_recdesc_t *frec; 9802 dtrace_aggid_t aggid; 9803 dtrace_state_t *state = ecb->dte_state; 9804 9805 agg = kmem_zalloc(sizeof (dtrace_aggregation_t), KM_SLEEP); 9806 agg->dtag_ecb = ecb; 9807 9808 ASSERT(DTRACEACT_ISAGG(desc->dtad_kind)); 9809 9810 switch (desc->dtad_kind) { 9811 case DTRACEAGG_MIN: 9812 agg->dtag_initial = INT64_MAX; 9813 agg->dtag_aggregate = dtrace_aggregate_min; 9814 break; 9815 9816 case DTRACEAGG_MAX: 9817 agg->dtag_initial = INT64_MIN; 9818 agg->dtag_aggregate = dtrace_aggregate_max; 9819 break; 9820 9821 case DTRACEAGG_COUNT: 9822 agg->dtag_aggregate = dtrace_aggregate_count; 9823 break; 9824 9825 case DTRACEAGG_QUANTIZE: 9826 agg->dtag_aggregate = dtrace_aggregate_quantize; 9827 size = (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) * 9828 sizeof (uint64_t); 9829 break; 9830 9831 case DTRACEAGG_LQUANTIZE: { 9832 uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg); 9833 uint16_t levels = DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg); 9834 9835 agg->dtag_initial = desc->dtad_arg; 9836 agg->dtag_aggregate = dtrace_aggregate_lquantize; 9837 9838 if (step == 0 || levels == 0) 9839 goto err; 9840 9841 size = levels * sizeof (uint64_t) + 3 * sizeof (uint64_t); 9842 break; 9843 } 9844 9845 case DTRACEAGG_AVG: 9846 agg->dtag_aggregate = dtrace_aggregate_avg; 9847 size = sizeof (uint64_t) * 2; 9848 break; 9849 9850 case DTRACEAGG_STDDEV: 9851 agg->dtag_aggregate = dtrace_aggregate_stddev; 9852 size = sizeof (uint64_t) * 4; 9853 break; 9854 9855 case DTRACEAGG_SUM: 9856 agg->dtag_aggregate = dtrace_aggregate_sum; 9857 break; 9858 9859 default: 9860 goto err; 9861 } 9862 9863 agg->dtag_action.dta_rec.dtrd_size = size; 9864 9865 if (ntuple == 0) 9866 goto err; 9867 9868 /* 9869 * We must make sure that we have enough actions for the n-tuple. 9870 */ 9871 for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) { 9872 if (DTRACEACT_ISAGG(act->dta_kind)) 9873 break; 9874 9875 if (--ntuple == 0) { 9876 /* 9877 * This is the action with which our n-tuple begins. 9878 */ 9879 agg->dtag_first = act; 9880 goto success; 9881 } 9882 } 9883 9884 /* 9885 * This n-tuple is short by ntuple elements. Return failure. 9886 */ 9887 ASSERT(ntuple != 0); 9888err: 9889 kmem_free(agg, sizeof (dtrace_aggregation_t)); 9890 return (NULL); 9891 9892success: 9893 /* 9894 * If the last action in the tuple has a size of zero, it's actually 9895 * an expression argument for the aggregating action. 9896 */ 9897 ASSERT(ecb->dte_action_last != NULL); 9898 act = ecb->dte_action_last; 9899 9900 if (act->dta_kind == DTRACEACT_DIFEXPR) { 9901 ASSERT(act->dta_difo != NULL); 9902 9903 if (act->dta_difo->dtdo_rtype.dtdt_size == 0) 9904 agg->dtag_hasarg = 1; 9905 } 9906 9907 /* 9908 * We need to allocate an id for this aggregation. 9909 */ 9910#if defined(sun) 9911 aggid = (dtrace_aggid_t)(uintptr_t)vmem_alloc(state->dts_aggid_arena, 1, 9912 VM_BESTFIT | VM_SLEEP); 9913#else 9914 aggid = alloc_unr(state->dts_aggid_arena); 9915#endif 9916 9917 if (aggid - 1 >= state->dts_naggregations) { 9918 dtrace_aggregation_t **oaggs = state->dts_aggregations; 9919 dtrace_aggregation_t **aggs; 9920 int naggs = state->dts_naggregations << 1; 9921 int onaggs = state->dts_naggregations; 9922 9923 ASSERT(aggid == state->dts_naggregations + 1); 9924 9925 if (naggs == 0) { 9926 ASSERT(oaggs == NULL); 9927 naggs = 1; 9928 } 9929 9930 aggs = kmem_zalloc(naggs * sizeof (*aggs), KM_SLEEP); 9931 9932 if (oaggs != NULL) { 9933 bcopy(oaggs, aggs, onaggs * sizeof (*aggs)); 9934 kmem_free(oaggs, onaggs * sizeof (*aggs)); 9935 } 9936 9937 state->dts_aggregations = aggs; 9938 state->dts_naggregations = naggs; 9939 } 9940 9941 ASSERT(state->dts_aggregations[aggid - 1] == NULL); 9942 state->dts_aggregations[(agg->dtag_id = aggid) - 1] = agg; 9943 9944 frec = &agg->dtag_first->dta_rec; 9945 if (frec->dtrd_alignment < sizeof (dtrace_aggid_t)) 9946 frec->dtrd_alignment = sizeof (dtrace_aggid_t); 9947 9948 for (act = agg->dtag_first; act != NULL; act = act->dta_next) { 9949 ASSERT(!act->dta_intuple); 9950 act->dta_intuple = 1; 9951 } 9952 9953 return (&agg->dtag_action); 9954} 9955 9956static void 9957dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act) 9958{ 9959 dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; 9960 dtrace_state_t *state = ecb->dte_state; 9961 dtrace_aggid_t aggid = agg->dtag_id; 9962 9963 ASSERT(DTRACEACT_ISAGG(act->dta_kind)); 9964#if defined(sun) 9965 vmem_free(state->dts_aggid_arena, (void *)(uintptr_t)aggid, 1); 9966#else 9967 free_unr(state->dts_aggid_arena, aggid); 9968#endif 9969 9970 ASSERT(state->dts_aggregations[aggid - 1] == agg); 9971 state->dts_aggregations[aggid - 1] = NULL; 9972 9973 kmem_free(agg, sizeof (dtrace_aggregation_t)); 9974} 9975 9976static int 9977dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) 9978{ 9979 dtrace_action_t *action, *last; 9980 dtrace_difo_t *dp = desc->dtad_difo; 9981 uint32_t size = 0, align = sizeof (uint8_t), mask; 9982 uint16_t format = 0; 9983 dtrace_recdesc_t *rec; 9984 dtrace_state_t *state = ecb->dte_state; 9985 dtrace_optval_t *opt = state->dts_options, nframes = 0, strsize; 9986 uint64_t arg = desc->dtad_arg; 9987 9988 ASSERT(MUTEX_HELD(&dtrace_lock)); 9989 ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1); 9990 9991 if (DTRACEACT_ISAGG(desc->dtad_kind)) { 9992 /* 9993 * If this is an aggregating action, there must be neither 9994 * a speculate nor a commit on the action chain. 9995 */ 9996 dtrace_action_t *act; 9997 9998 for (act = ecb->dte_action; act != NULL; act = act->dta_next) { 9999 if (act->dta_kind == DTRACEACT_COMMIT) 10000 return (EINVAL); 10001 10002 if (act->dta_kind == DTRACEACT_SPECULATE) 10003 return (EINVAL); 10004 } 10005 10006 action = dtrace_ecb_aggregation_create(ecb, desc); 10007 10008 if (action == NULL) 10009 return (EINVAL); 10010 } else { 10011 if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) || 10012 (desc->dtad_kind == DTRACEACT_DIFEXPR && 10013 dp != NULL && dp->dtdo_destructive)) { 10014 state->dts_destructive = 1; 10015 } 10016 10017 switch (desc->dtad_kind) { 10018 case DTRACEACT_PRINTF: 10019 case DTRACEACT_PRINTA: 10020 case DTRACEACT_SYSTEM: 10021 case DTRACEACT_FREOPEN: 10022 /* 10023 * We know that our arg is a string -- turn it into a 10024 * format. 10025 */ 10026 if (arg == 0) { 10027 ASSERT(desc->dtad_kind == DTRACEACT_PRINTA); 10028 format = 0; 10029 } else { 10030 ASSERT(arg != 0); 10031#if defined(sun) 10032 ASSERT(arg > KERNELBASE); 10033#endif 10034 format = dtrace_format_add(state, 10035 (char *)(uintptr_t)arg); 10036 } 10037 10038 /*FALLTHROUGH*/ 10039 case DTRACEACT_LIBACT: 10040 case DTRACEACT_DIFEXPR: 10041 if (dp == NULL) 10042 return (EINVAL); 10043 10044 if ((size = dp->dtdo_rtype.dtdt_size) != 0) 10045 break; 10046 10047 if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) { 10048 if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) 10049 return (EINVAL); 10050 10051 size = opt[DTRACEOPT_STRSIZE]; 10052 } 10053 10054 break; 10055 10056 case DTRACEACT_STACK: 10057 if ((nframes = arg) == 0) { 10058 nframes = opt[DTRACEOPT_STACKFRAMES]; 10059 ASSERT(nframes > 0); 10060 arg = nframes; 10061 } 10062 10063 size = nframes * sizeof (pc_t); 10064 break; 10065 10066 case DTRACEACT_JSTACK: 10067 if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0) 10068 strsize = opt[DTRACEOPT_JSTACKSTRSIZE]; 10069 10070 if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) 10071 nframes = opt[DTRACEOPT_JSTACKFRAMES]; 10072 10073 arg = DTRACE_USTACK_ARG(nframes, strsize); 10074 10075 /*FALLTHROUGH*/ 10076 case DTRACEACT_USTACK: 10077 if (desc->dtad_kind != DTRACEACT_JSTACK && 10078 (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) { 10079 strsize = DTRACE_USTACK_STRSIZE(arg); 10080 nframes = opt[DTRACEOPT_USTACKFRAMES]; 10081 ASSERT(nframes > 0); 10082 arg = DTRACE_USTACK_ARG(nframes, strsize); 10083 } 10084 10085 /* 10086 * Save a slot for the pid. 10087 */ 10088 size = (nframes + 1) * sizeof (uint64_t); 10089 size += DTRACE_USTACK_STRSIZE(arg); 10090 size = P2ROUNDUP(size, (uint32_t)(sizeof (uintptr_t))); 10091 10092 break; 10093 10094 case DTRACEACT_SYM: 10095 case DTRACEACT_MOD: 10096 if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) != 10097 sizeof (uint64_t)) || 10098 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) 10099 return (EINVAL); 10100 break; 10101 10102 case DTRACEACT_USYM: 10103 case DTRACEACT_UMOD: 10104 case DTRACEACT_UADDR: 10105 if (dp == NULL || 10106 (dp->dtdo_rtype.dtdt_size != sizeof (uint64_t)) || 10107 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) 10108 return (EINVAL); 10109 10110 /* 10111 * We have a slot for the pid, plus a slot for the 10112 * argument. To keep things simple (aligned with 10113 * bitness-neutral sizing), we store each as a 64-bit 10114 * quantity. 10115 */ 10116 size = 2 * sizeof (uint64_t); 10117 break; 10118 10119 case DTRACEACT_STOP: 10120 case DTRACEACT_BREAKPOINT: 10121 case DTRACEACT_PANIC: 10122 break; 10123 10124 case DTRACEACT_CHILL: 10125 case DTRACEACT_DISCARD: 10126 case DTRACEACT_RAISE: 10127 if (dp == NULL) 10128 return (EINVAL); 10129 break; 10130 10131 case DTRACEACT_EXIT: 10132 if (dp == NULL || 10133 (size = dp->dtdo_rtype.dtdt_size) != sizeof (int) || 10134 (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) 10135 return (EINVAL); 10136 break; 10137 10138 case DTRACEACT_SPECULATE: 10139 if (ecb->dte_size > sizeof (dtrace_epid_t)) 10140 return (EINVAL); 10141 10142 if (dp == NULL) 10143 return (EINVAL); 10144 10145 state->dts_speculates = 1; 10146 break; 10147 10148 case DTRACEACT_PRINTM: 10149 size = dp->dtdo_rtype.dtdt_size; 10150 break; 10151 10152 case DTRACEACT_PRINTT: 10153 size = dp->dtdo_rtype.dtdt_size; 10154 break; 10155 10156 case DTRACEACT_COMMIT: { 10157 dtrace_action_t *act = ecb->dte_action; 10158 10159 for (; act != NULL; act = act->dta_next) { 10160 if (act->dta_kind == DTRACEACT_COMMIT) 10161 return (EINVAL); 10162 } 10163 10164 if (dp == NULL) 10165 return (EINVAL); 10166 break; 10167 } 10168 10169 default: 10170 return (EINVAL); 10171 } 10172 10173 if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) { 10174 /* 10175 * If this is a data-storing action or a speculate, 10176 * we must be sure that there isn't a commit on the 10177 * action chain. 10178 */ 10179 dtrace_action_t *act = ecb->dte_action; 10180 10181 for (; act != NULL; act = act->dta_next) { 10182 if (act->dta_kind == DTRACEACT_COMMIT) 10183 return (EINVAL); 10184 } 10185 } 10186 10187 action = kmem_zalloc(sizeof (dtrace_action_t), KM_SLEEP); 10188 action->dta_rec.dtrd_size = size; 10189 } 10190 10191 action->dta_refcnt = 1; 10192 rec = &action->dta_rec; 10193 size = rec->dtrd_size; 10194 10195 for (mask = sizeof (uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) { 10196 if (!(size & mask)) { 10197 align = mask + 1; 10198 break; 10199 } 10200 } 10201 10202 action->dta_kind = desc->dtad_kind; 10203 10204 if ((action->dta_difo = dp) != NULL) 10205 dtrace_difo_hold(dp); 10206 10207 rec->dtrd_action = action->dta_kind; 10208 rec->dtrd_arg = arg; 10209 rec->dtrd_uarg = desc->dtad_uarg; 10210 rec->dtrd_alignment = (uint16_t)align; 10211 rec->dtrd_format = format; 10212 10213 if ((last = ecb->dte_action_last) != NULL) { 10214 ASSERT(ecb->dte_action != NULL); 10215 action->dta_prev = last; 10216 last->dta_next = action; 10217 } else { 10218 ASSERT(ecb->dte_action == NULL); 10219 ecb->dte_action = action; 10220 } 10221 10222 ecb->dte_action_last = action; 10223 10224 return (0); 10225} 10226 10227static void 10228dtrace_ecb_action_remove(dtrace_ecb_t *ecb) 10229{ 10230 dtrace_action_t *act = ecb->dte_action, *next; 10231 dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate; 10232 dtrace_difo_t *dp; 10233 uint16_t format; 10234 10235 if (act != NULL && act->dta_refcnt > 1) { 10236 ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1); 10237 act->dta_refcnt--; 10238 } else { 10239 for (; act != NULL; act = next) { 10240 next = act->dta_next; 10241 ASSERT(next != NULL || act == ecb->dte_action_last); 10242 ASSERT(act->dta_refcnt == 1); 10243 10244 if ((format = act->dta_rec.dtrd_format) != 0) 10245 dtrace_format_remove(ecb->dte_state, format); 10246 10247 if ((dp = act->dta_difo) != NULL) 10248 dtrace_difo_release(dp, vstate); 10249 10250 if (DTRACEACT_ISAGG(act->dta_kind)) { 10251 dtrace_ecb_aggregation_destroy(ecb, act); 10252 } else { 10253 kmem_free(act, sizeof (dtrace_action_t)); 10254 } 10255 } 10256 } 10257 10258 ecb->dte_action = NULL; 10259 ecb->dte_action_last = NULL; 10260 ecb->dte_size = sizeof (dtrace_epid_t); 10261} 10262 10263static void 10264dtrace_ecb_disable(dtrace_ecb_t *ecb) 10265{ 10266 /* 10267 * We disable the ECB by removing it from its probe. 10268 */ 10269 dtrace_ecb_t *pecb, *prev = NULL; 10270 dtrace_probe_t *probe = ecb->dte_probe; 10271 10272 ASSERT(MUTEX_HELD(&dtrace_lock)); 10273 10274 if (probe == NULL) { 10275 /* 10276 * This is the NULL probe; there is nothing to disable. 10277 */ 10278 return; 10279 } 10280 10281 for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) { 10282 if (pecb == ecb) 10283 break; 10284 prev = pecb; 10285 } 10286 10287 ASSERT(pecb != NULL); 10288 10289 if (prev == NULL) { 10290 probe->dtpr_ecb = ecb->dte_next; 10291 } else { 10292 prev->dte_next = ecb->dte_next; 10293 } 10294 10295 if (ecb == probe->dtpr_ecb_last) { 10296 ASSERT(ecb->dte_next == NULL); 10297 probe->dtpr_ecb_last = prev; 10298 } 10299 10300 /* 10301 * The ECB has been disconnected from the probe; now sync to assure 10302 * that all CPUs have seen the change before returning. 10303 */ 10304 dtrace_sync(); 10305 10306 if (probe->dtpr_ecb == NULL) { 10307 /* 10308 * That was the last ECB on the probe; clear the predicate 10309 * cache ID for the probe, disable it and sync one more time 10310 * to assure that we'll never hit it again. 10311 */ 10312 dtrace_provider_t *prov = probe->dtpr_provider; 10313 10314 ASSERT(ecb->dte_next == NULL); 10315 ASSERT(probe->dtpr_ecb_last == NULL); 10316 probe->dtpr_predcache = DTRACE_CACHEIDNONE; 10317 prov->dtpv_pops.dtps_disable(prov->dtpv_arg, 10318 probe->dtpr_id, probe->dtpr_arg); 10319 dtrace_sync(); 10320 } else { 10321 /* 10322 * There is at least one ECB remaining on the probe. If there 10323 * is _exactly_ one, set the probe's predicate cache ID to be 10324 * the predicate cache ID of the remaining ECB. 10325 */ 10326 ASSERT(probe->dtpr_ecb_last != NULL); 10327 ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE); 10328 10329 if (probe->dtpr_ecb == probe->dtpr_ecb_last) { 10330 dtrace_predicate_t *p = probe->dtpr_ecb->dte_predicate; 10331 10332 ASSERT(probe->dtpr_ecb->dte_next == NULL); 10333 10334 if (p != NULL) 10335 probe->dtpr_predcache = p->dtp_cacheid; 10336 } 10337 10338 ecb->dte_next = NULL; 10339 } 10340} 10341 10342static void 10343dtrace_ecb_destroy(dtrace_ecb_t *ecb) 10344{ 10345 dtrace_state_t *state = ecb->dte_state; 10346 dtrace_vstate_t *vstate = &state->dts_vstate; 10347 dtrace_predicate_t *pred; 10348 dtrace_epid_t epid = ecb->dte_epid; 10349 10350 ASSERT(MUTEX_HELD(&dtrace_lock)); 10351 ASSERT(ecb->dte_next == NULL); 10352 ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb); 10353 10354 if ((pred = ecb->dte_predicate) != NULL) 10355 dtrace_predicate_release(pred, vstate); 10356 10357 dtrace_ecb_action_remove(ecb); 10358 10359 ASSERT(state->dts_ecbs[epid - 1] == ecb); 10360 state->dts_ecbs[epid - 1] = NULL; 10361 10362 kmem_free(ecb, sizeof (dtrace_ecb_t)); 10363} 10364 10365static dtrace_ecb_t * 10366dtrace_ecb_create(dtrace_state_t *state, dtrace_probe_t *probe, 10367 dtrace_enabling_t *enab) 10368{ 10369 dtrace_ecb_t *ecb; 10370 dtrace_predicate_t *pred; 10371 dtrace_actdesc_t *act; 10372 dtrace_provider_t *prov; 10373 dtrace_ecbdesc_t *desc = enab->dten_current; 10374 10375 ASSERT(MUTEX_HELD(&dtrace_lock)); 10376 ASSERT(state != NULL); 10377 10378 ecb = dtrace_ecb_add(state, probe); 10379 ecb->dte_uarg = desc->dted_uarg; 10380 10381 if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) { 10382 dtrace_predicate_hold(pred); 10383 ecb->dte_predicate = pred; 10384 } 10385 10386 if (probe != NULL) { 10387 /* 10388 * If the provider shows more leg than the consumer is old 10389 * enough to see, we need to enable the appropriate implicit 10390 * predicate bits to prevent the ecb from activating at 10391 * revealing times. 10392 * 10393 * Providers specifying DTRACE_PRIV_USER at register time 10394 * are stating that they need the /proc-style privilege 10395 * model to be enforced, and this is what DTRACE_COND_OWNER 10396 * and DTRACE_COND_ZONEOWNER will then do at probe time. 10397 */ 10398 prov = probe->dtpr_provider; 10399 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) && 10400 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) 10401 ecb->dte_cond |= DTRACE_COND_OWNER; 10402 10403 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLZONE) && 10404 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) 10405 ecb->dte_cond |= DTRACE_COND_ZONEOWNER; 10406 10407 /* 10408 * If the provider shows us kernel innards and the user 10409 * is lacking sufficient privilege, enable the 10410 * DTRACE_COND_USERMODE implicit predicate. 10411 */ 10412 if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) && 10413 (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL)) 10414 ecb->dte_cond |= DTRACE_COND_USERMODE; 10415 } 10416 10417 if (dtrace_ecb_create_cache != NULL) { 10418 /* 10419 * If we have a cached ecb, we'll use its action list instead 10420 * of creating our own (saving both time and space). 10421 */ 10422 dtrace_ecb_t *cached = dtrace_ecb_create_cache; 10423 dtrace_action_t *act = cached->dte_action; 10424 10425 if (act != NULL) { 10426 ASSERT(act->dta_refcnt > 0); 10427 act->dta_refcnt++; 10428 ecb->dte_action = act; 10429 ecb->dte_action_last = cached->dte_action_last; 10430 ecb->dte_needed = cached->dte_needed; 10431 ecb->dte_size = cached->dte_size; 10432 ecb->dte_alignment = cached->dte_alignment; 10433 } 10434 10435 return (ecb); 10436 } 10437 10438 for (act = desc->dted_action; act != NULL; act = act->dtad_next) { 10439 if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) { 10440 dtrace_ecb_destroy(ecb); 10441 return (NULL); 10442 } 10443 } 10444 10445 dtrace_ecb_resize(ecb); 10446 10447 return (dtrace_ecb_create_cache = ecb); 10448} 10449 10450static int 10451dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg) 10452{ 10453 dtrace_ecb_t *ecb; 10454 dtrace_enabling_t *enab = arg; 10455 dtrace_state_t *state = enab->dten_vstate->dtvs_state; 10456 10457 ASSERT(state != NULL); 10458 10459 if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) { 10460 /* 10461 * This probe was created in a generation for which this 10462 * enabling has previously created ECBs; we don't want to 10463 * enable it again, so just kick out. 10464 */ 10465 return (DTRACE_MATCH_NEXT); 10466 } 10467 10468 if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL) 10469 return (DTRACE_MATCH_DONE); 10470 10471 dtrace_ecb_enable(ecb); 10472 return (DTRACE_MATCH_NEXT); 10473} 10474 10475static dtrace_ecb_t * 10476dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id) 10477{ 10478 dtrace_ecb_t *ecb; 10479 10480 ASSERT(MUTEX_HELD(&dtrace_lock)); 10481 10482 if (id == 0 || id > state->dts_necbs) 10483 return (NULL); 10484 10485 ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL); 10486 ASSERT((ecb = state->dts_ecbs[id - 1]) == NULL || ecb->dte_epid == id); 10487 10488 return (state->dts_ecbs[id - 1]); 10489} 10490 10491static dtrace_aggregation_t * 10492dtrace_aggid2agg(dtrace_state_t *state, dtrace_aggid_t id) 10493{ 10494 dtrace_aggregation_t *agg; 10495 10496 ASSERT(MUTEX_HELD(&dtrace_lock)); 10497 10498 if (id == 0 || id > state->dts_naggregations) 10499 return (NULL); 10500 10501 ASSERT(state->dts_naggregations > 0 && state->dts_aggregations != NULL); 10502 ASSERT((agg = state->dts_aggregations[id - 1]) == NULL || 10503 agg->dtag_id == id); 10504 10505 return (state->dts_aggregations[id - 1]); 10506} 10507 10508/* 10509 * DTrace Buffer Functions 10510 * 10511 * The following functions manipulate DTrace buffers. Most of these functions 10512 * are called in the context of establishing or processing consumer state; 10513 * exceptions are explicitly noted. 10514 */ 10515 10516/* 10517 * Note: called from cross call context. This function switches the two 10518 * buffers on a given CPU. The atomicity of this operation is assured by 10519 * disabling interrupts while the actual switch takes place; the disabling of 10520 * interrupts serializes the execution with any execution of dtrace_probe() on 10521 * the same CPU. 10522 */ 10523static void 10524dtrace_buffer_switch(dtrace_buffer_t *buf) 10525{ 10526 caddr_t tomax = buf->dtb_tomax; 10527 caddr_t xamot = buf->dtb_xamot; 10528 dtrace_icookie_t cookie; 10529 10530 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); 10531 ASSERT(!(buf->dtb_flags & DTRACEBUF_RING)); 10532 10533 cookie = dtrace_interrupt_disable(); 10534 buf->dtb_tomax = xamot; 10535 buf->dtb_xamot = tomax; 10536 buf->dtb_xamot_drops = buf->dtb_drops; 10537 buf->dtb_xamot_offset = buf->dtb_offset; 10538 buf->dtb_xamot_errors = buf->dtb_errors; 10539 buf->dtb_xamot_flags = buf->dtb_flags; 10540 buf->dtb_offset = 0; 10541 buf->dtb_drops = 0; 10542 buf->dtb_errors = 0; 10543 buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED); 10544 dtrace_interrupt_enable(cookie); 10545} 10546 10547/* 10548 * Note: called from cross call context. This function activates a buffer 10549 * on a CPU. As with dtrace_buffer_switch(), the atomicity of the operation 10550 * is guaranteed by the disabling of interrupts. 10551 */ 10552static void 10553dtrace_buffer_activate(dtrace_state_t *state) 10554{ 10555 dtrace_buffer_t *buf; 10556 dtrace_icookie_t cookie = dtrace_interrupt_disable(); 10557 10558 buf = &state->dts_buffer[curcpu]; 10559 10560 if (buf->dtb_tomax != NULL) { 10561 /* 10562 * We might like to assert that the buffer is marked inactive, 10563 * but this isn't necessarily true: the buffer for the CPU 10564 * that processes the BEGIN probe has its buffer activated 10565 * manually. In this case, we take the (harmless) action 10566 * re-clearing the bit INACTIVE bit. 10567 */ 10568 buf->dtb_flags &= ~DTRACEBUF_INACTIVE; 10569 } 10570 10571 dtrace_interrupt_enable(cookie); 10572} 10573 10574static int 10575dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags, 10576 processorid_t cpu) 10577{ 10578#if defined(sun) 10579 cpu_t *cp; 10580#else 10581 struct pcpu *cp; 10582#endif 10583 dtrace_buffer_t *buf; 10584 10585#if defined(sun) 10586 ASSERT(MUTEX_HELD(&cpu_lock)); 10587 ASSERT(MUTEX_HELD(&dtrace_lock)); 10588 10589 if (size > dtrace_nonroot_maxsize && 10590 !PRIV_POLICY_CHOICE(CRED(), PRIV_ALL, B_FALSE)) 10591 return (EFBIG); 10592 10593 cp = cpu_list; 10594 10595 do { 10596 if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id) 10597 continue; 10598 10599 buf = &bufs[cp->cpu_id]; 10600 10601 /* 10602 * If there is already a buffer allocated for this CPU, it 10603 * is only possible that this is a DR event. In this case, 10604 */ 10605 if (buf->dtb_tomax != NULL) { 10606 ASSERT(buf->dtb_size == size); 10607 continue; 10608 } 10609 10610 ASSERT(buf->dtb_xamot == NULL); 10611 10612 if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL) 10613 goto err; 10614 10615 buf->dtb_size = size; 10616 buf->dtb_flags = flags; 10617 buf->dtb_offset = 0; 10618 buf->dtb_drops = 0; 10619 10620 if (flags & DTRACEBUF_NOSWITCH) 10621 continue; 10622 10623 if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL) 10624 goto err; 10625 } while ((cp = cp->cpu_next) != cpu_list); 10626 10627 return (0); 10628 10629err: 10630 cp = cpu_list; 10631 10632 do { 10633 if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id) 10634 continue; 10635 10636 buf = &bufs[cp->cpu_id]; 10637 10638 if (buf->dtb_xamot != NULL) { 10639 ASSERT(buf->dtb_tomax != NULL); 10640 ASSERT(buf->dtb_size == size); 10641 kmem_free(buf->dtb_xamot, size); 10642 } 10643 10644 if (buf->dtb_tomax != NULL) { 10645 ASSERT(buf->dtb_size == size); 10646 kmem_free(buf->dtb_tomax, size); 10647 } 10648 10649 buf->dtb_tomax = NULL; 10650 buf->dtb_xamot = NULL; 10651 buf->dtb_size = 0; 10652 } while ((cp = cp->cpu_next) != cpu_list); 10653 10654 return (ENOMEM); 10655#else 10656 int i; 10657 10658#if defined(__amd64__) 10659 /* 10660 * FreeBSD isn't good at limiting the amount of memory we 10661 * ask to malloc, so let's place a limit here before trying 10662 * to do something that might well end in tears at bedtime. 10663 */ 10664 if (size > physmem * PAGE_SIZE / (128 * (mp_maxid + 1))) 10665 return(ENOMEM); 10666#endif 10667 10668 ASSERT(MUTEX_HELD(&dtrace_lock)); 10669 for (i = 0; i <= mp_maxid; i++) { 10670 if ((cp = pcpu_find(i)) == NULL) 10671 continue; 10672 10673 if (cpu != DTRACE_CPUALL && cpu != i) 10674 continue; 10675 10676 buf = &bufs[i]; 10677 10678 /* 10679 * If there is already a buffer allocated for this CPU, it 10680 * is only possible that this is a DR event. In this case, 10681 * the buffer size must match our specified size. 10682 */ 10683 if (buf->dtb_tomax != NULL) { 10684 ASSERT(buf->dtb_size == size); 10685 continue; 10686 } 10687 10688 ASSERT(buf->dtb_xamot == NULL); 10689 10690 if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL) 10691 goto err; 10692 10693 buf->dtb_size = size; 10694 buf->dtb_flags = flags; 10695 buf->dtb_offset = 0; 10696 buf->dtb_drops = 0; 10697 10698 if (flags & DTRACEBUF_NOSWITCH) 10699 continue; 10700 10701 if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL) 10702 goto err; 10703 } 10704 10705 return (0); 10706 10707err: 10708 /* 10709 * Error allocating memory, so free the buffers that were 10710 * allocated before the failed allocation. 10711 */ 10712 for (i = 0; i <= mp_maxid; i++) { 10713 if ((cp = pcpu_find(i)) == NULL) 10714 continue; 10715 10716 if (cpu != DTRACE_CPUALL && cpu != i) 10717 continue; 10718 10719 buf = &bufs[i]; 10720 10721 if (buf->dtb_xamot != NULL) { 10722 ASSERT(buf->dtb_tomax != NULL); 10723 ASSERT(buf->dtb_size == size); 10724 kmem_free(buf->dtb_xamot, size); 10725 } 10726 10727 if (buf->dtb_tomax != NULL) { 10728 ASSERT(buf->dtb_size == size); 10729 kmem_free(buf->dtb_tomax, size); 10730 } 10731 10732 buf->dtb_tomax = NULL; 10733 buf->dtb_xamot = NULL; 10734 buf->dtb_size = 0; 10735 10736 } 10737 10738 return (ENOMEM); 10739#endif 10740} 10741 10742/* 10743 * Note: called from probe context. This function just increments the drop 10744 * count on a buffer. It has been made a function to allow for the 10745 * possibility of understanding the source of mysterious drop counts. (A 10746 * problem for which one may be particularly disappointed that DTrace cannot 10747 * be used to understand DTrace.) 10748 */ 10749static void 10750dtrace_buffer_drop(dtrace_buffer_t *buf) 10751{ 10752 buf->dtb_drops++; 10753} 10754 10755/* 10756 * Note: called from probe context. This function is called to reserve space 10757 * in a buffer. If mstate is non-NULL, sets the scratch base and size in the 10758 * mstate. Returns the new offset in the buffer, or a negative value if an 10759 * error has occurred. 10760 */ 10761static intptr_t 10762dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, size_t align, 10763 dtrace_state_t *state, dtrace_mstate_t *mstate) 10764{ 10765 intptr_t offs = buf->dtb_offset, soffs; 10766 intptr_t woffs; 10767 caddr_t tomax; 10768 size_t total; 10769 10770 if (buf->dtb_flags & DTRACEBUF_INACTIVE) 10771 return (-1); 10772 10773 if ((tomax = buf->dtb_tomax) == NULL) { 10774 dtrace_buffer_drop(buf); 10775 return (-1); 10776 } 10777 10778 if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) { 10779 while (offs & (align - 1)) { 10780 /* 10781 * Assert that our alignment is off by a number which 10782 * is itself sizeof (uint32_t) aligned. 10783 */ 10784 ASSERT(!((align - (offs & (align - 1))) & 10785 (sizeof (uint32_t) - 1))); 10786 DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); 10787 offs += sizeof (uint32_t); 10788 } 10789 10790 if ((soffs = offs + needed) > buf->dtb_size) { 10791 dtrace_buffer_drop(buf); 10792 return (-1); 10793 } 10794 10795 if (mstate == NULL) 10796 return (offs); 10797 10798 mstate->dtms_scratch_base = (uintptr_t)tomax + soffs; 10799 mstate->dtms_scratch_size = buf->dtb_size - soffs; 10800 mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; 10801 10802 return (offs); 10803 } 10804 10805 if (buf->dtb_flags & DTRACEBUF_FILL) { 10806 if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN && 10807 (buf->dtb_flags & DTRACEBUF_FULL)) 10808 return (-1); 10809 goto out; 10810 } 10811 10812 total = needed + (offs & (align - 1)); 10813 10814 /* 10815 * For a ring buffer, life is quite a bit more complicated. Before 10816 * we can store any padding, we need to adjust our wrapping offset. 10817 * (If we've never before wrapped or we're not about to, no adjustment 10818 * is required.) 10819 */ 10820 if ((buf->dtb_flags & DTRACEBUF_WRAPPED) || 10821 offs + total > buf->dtb_size) { 10822 woffs = buf->dtb_xamot_offset; 10823 10824 if (offs + total > buf->dtb_size) { 10825 /* 10826 * We can't fit in the end of the buffer. First, a 10827 * sanity check that we can fit in the buffer at all. 10828 */ 10829 if (total > buf->dtb_size) { 10830 dtrace_buffer_drop(buf); 10831 return (-1); 10832 } 10833 10834 /* 10835 * We're going to be storing at the top of the buffer, 10836 * so now we need to deal with the wrapped offset. We 10837 * only reset our wrapped offset to 0 if it is 10838 * currently greater than the current offset. If it 10839 * is less than the current offset, it is because a 10840 * previous allocation induced a wrap -- but the 10841 * allocation didn't subsequently take the space due 10842 * to an error or false predicate evaluation. In this 10843 * case, we'll just leave the wrapped offset alone: if 10844 * the wrapped offset hasn't been advanced far enough 10845 * for this allocation, it will be adjusted in the 10846 * lower loop. 10847 */ 10848 if (buf->dtb_flags & DTRACEBUF_WRAPPED) { 10849 if (woffs >= offs) 10850 woffs = 0; 10851 } else { 10852 woffs = 0; 10853 } 10854 10855 /* 10856 * Now we know that we're going to be storing to the 10857 * top of the buffer and that there is room for us 10858 * there. We need to clear the buffer from the current 10859 * offset to the end (there may be old gunk there). 10860 */ 10861 while (offs < buf->dtb_size) 10862 tomax[offs++] = 0; 10863 10864 /* 10865 * We need to set our offset to zero. And because we 10866 * are wrapping, we need to set the bit indicating as 10867 * much. We can also adjust our needed space back 10868 * down to the space required by the ECB -- we know 10869 * that the top of the buffer is aligned. 10870 */ 10871 offs = 0; 10872 total = needed; 10873 buf->dtb_flags |= DTRACEBUF_WRAPPED; 10874 } else { 10875 /* 10876 * There is room for us in the buffer, so we simply 10877 * need to check the wrapped offset. 10878 */ 10879 if (woffs < offs) { 10880 /* 10881 * The wrapped offset is less than the offset. 10882 * This can happen if we allocated buffer space 10883 * that induced a wrap, but then we didn't 10884 * subsequently take the space due to an error 10885 * or false predicate evaluation. This is 10886 * okay; we know that _this_ allocation isn't 10887 * going to induce a wrap. We still can't 10888 * reset the wrapped offset to be zero, 10889 * however: the space may have been trashed in 10890 * the previous failed probe attempt. But at 10891 * least the wrapped offset doesn't need to 10892 * be adjusted at all... 10893 */ 10894 goto out; 10895 } 10896 } 10897 10898 while (offs + total > woffs) { 10899 dtrace_epid_t epid = *(uint32_t *)(tomax + woffs); 10900 size_t size; 10901 10902 if (epid == DTRACE_EPIDNONE) { 10903 size = sizeof (uint32_t); 10904 } else { 10905 ASSERT(epid <= state->dts_necbs); 10906 ASSERT(state->dts_ecbs[epid - 1] != NULL); 10907 10908 size = state->dts_ecbs[epid - 1]->dte_size; 10909 } 10910 10911 ASSERT(woffs + size <= buf->dtb_size); 10912 ASSERT(size != 0); 10913 10914 if (woffs + size == buf->dtb_size) { 10915 /* 10916 * We've reached the end of the buffer; we want 10917 * to set the wrapped offset to 0 and break 10918 * out. However, if the offs is 0, then we're 10919 * in a strange edge-condition: the amount of 10920 * space that we want to reserve plus the size 10921 * of the record that we're overwriting is 10922 * greater than the size of the buffer. This 10923 * is problematic because if we reserve the 10924 * space but subsequently don't consume it (due 10925 * to a failed predicate or error) the wrapped 10926 * offset will be 0 -- yet the EPID at offset 0 10927 * will not be committed. This situation is 10928 * relatively easy to deal with: if we're in 10929 * this case, the buffer is indistinguishable 10930 * from one that hasn't wrapped; we need only 10931 * finish the job by clearing the wrapped bit, 10932 * explicitly setting the offset to be 0, and 10933 * zero'ing out the old data in the buffer. 10934 */ 10935 if (offs == 0) { 10936 buf->dtb_flags &= ~DTRACEBUF_WRAPPED; 10937 buf->dtb_offset = 0; 10938 woffs = total; 10939 10940 while (woffs < buf->dtb_size) 10941 tomax[woffs++] = 0; 10942 } 10943 10944 woffs = 0; 10945 break; 10946 } 10947 10948 woffs += size; 10949 } 10950 10951 /* 10952 * We have a wrapped offset. It may be that the wrapped offset 10953 * has become zero -- that's okay. 10954 */ 10955 buf->dtb_xamot_offset = woffs; 10956 } 10957 10958out: 10959 /* 10960 * Now we can plow the buffer with any necessary padding. 10961 */ 10962 while (offs & (align - 1)) { 10963 /* 10964 * Assert that our alignment is off by a number which 10965 * is itself sizeof (uint32_t) aligned. 10966 */ 10967 ASSERT(!((align - (offs & (align - 1))) & 10968 (sizeof (uint32_t) - 1))); 10969 DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); 10970 offs += sizeof (uint32_t); 10971 } 10972 10973 if (buf->dtb_flags & DTRACEBUF_FILL) { 10974 if (offs + needed > buf->dtb_size - state->dts_reserve) { 10975 buf->dtb_flags |= DTRACEBUF_FULL; 10976 return (-1); 10977 } 10978 } 10979 10980 if (mstate == NULL) 10981 return (offs); 10982 10983 /* 10984 * For ring buffers and fill buffers, the scratch space is always 10985 * the inactive buffer. 10986 */ 10987 mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot; 10988 mstate->dtms_scratch_size = buf->dtb_size; 10989 mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; 10990 10991 return (offs); 10992} 10993 10994static void 10995dtrace_buffer_polish(dtrace_buffer_t *buf) 10996{ 10997 ASSERT(buf->dtb_flags & DTRACEBUF_RING); 10998 ASSERT(MUTEX_HELD(&dtrace_lock)); 10999 11000 if (!(buf->dtb_flags & DTRACEBUF_WRAPPED)) 11001 return; 11002 11003 /* 11004 * We need to polish the ring buffer. There are three cases: 11005 * 11006 * - The first (and presumably most common) is that there is no gap 11007 * between the buffer offset and the wrapped offset. In this case, 11008 * there is nothing in the buffer that isn't valid data; we can 11009 * mark the buffer as polished and return. 11010 * 11011 * - The second (less common than the first but still more common 11012 * than the third) is that there is a gap between the buffer offset 11013 * and the wrapped offset, and the wrapped offset is larger than the 11014 * buffer offset. This can happen because of an alignment issue, or 11015 * can happen because of a call to dtrace_buffer_reserve() that 11016 * didn't subsequently consume the buffer space. In this case, 11017 * we need to zero the data from the buffer offset to the wrapped 11018 * offset. 11019 * 11020 * - The third (and least common) is that there is a gap between the 11021 * buffer offset and the wrapped offset, but the wrapped offset is 11022 * _less_ than the buffer offset. This can only happen because a 11023 * call to dtrace_buffer_reserve() induced a wrap, but the space 11024 * was not subsequently consumed. In this case, we need to zero the 11025 * space from the offset to the end of the buffer _and_ from the 11026 * top of the buffer to the wrapped offset. 11027 */ 11028 if (buf->dtb_offset < buf->dtb_xamot_offset) { 11029 bzero(buf->dtb_tomax + buf->dtb_offset, 11030 buf->dtb_xamot_offset - buf->dtb_offset); 11031 } 11032 11033 if (buf->dtb_offset > buf->dtb_xamot_offset) { 11034 bzero(buf->dtb_tomax + buf->dtb_offset, 11035 buf->dtb_size - buf->dtb_offset); 11036 bzero(buf->dtb_tomax, buf->dtb_xamot_offset); 11037 } 11038} 11039 11040static void 11041dtrace_buffer_free(dtrace_buffer_t *bufs) 11042{ 11043 int i; 11044 11045 for (i = 0; i < NCPU; i++) { 11046 dtrace_buffer_t *buf = &bufs[i]; 11047 11048 if (buf->dtb_tomax == NULL) { 11049 ASSERT(buf->dtb_xamot == NULL); 11050 ASSERT(buf->dtb_size == 0); 11051 continue; 11052 } 11053 11054 if (buf->dtb_xamot != NULL) { 11055 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); 11056 kmem_free(buf->dtb_xamot, buf->dtb_size); 11057 } 11058 11059 kmem_free(buf->dtb_tomax, buf->dtb_size); 11060 buf->dtb_size = 0; 11061 buf->dtb_tomax = NULL; 11062 buf->dtb_xamot = NULL; 11063 } 11064} 11065 11066/* 11067 * DTrace Enabling Functions 11068 */ 11069static dtrace_enabling_t * 11070dtrace_enabling_create(dtrace_vstate_t *vstate) 11071{ 11072 dtrace_enabling_t *enab; 11073 11074 enab = kmem_zalloc(sizeof (dtrace_enabling_t), KM_SLEEP); 11075 enab->dten_vstate = vstate; 11076 11077 return (enab); 11078} 11079 11080static void 11081dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb) 11082{ 11083 dtrace_ecbdesc_t **ndesc; 11084 size_t osize, nsize; 11085 11086 /* 11087 * We can't add to enablings after we've enabled them, or after we've 11088 * retained them. 11089 */ 11090 ASSERT(enab->dten_probegen == 0); 11091 ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); 11092 11093 if (enab->dten_ndesc < enab->dten_maxdesc) { 11094 enab->dten_desc[enab->dten_ndesc++] = ecb; 11095 return; 11096 } 11097 11098 osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); 11099 11100 if (enab->dten_maxdesc == 0) { 11101 enab->dten_maxdesc = 1; 11102 } else { 11103 enab->dten_maxdesc <<= 1; 11104 } 11105 11106 ASSERT(enab->dten_ndesc < enab->dten_maxdesc); 11107 11108 nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); 11109 ndesc = kmem_zalloc(nsize, KM_SLEEP); 11110 bcopy(enab->dten_desc, ndesc, osize); 11111 if (enab->dten_desc != NULL) 11112 kmem_free(enab->dten_desc, osize); 11113 11114 enab->dten_desc = ndesc; 11115 enab->dten_desc[enab->dten_ndesc++] = ecb; 11116} 11117 11118static void 11119dtrace_enabling_addlike(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb, 11120 dtrace_probedesc_t *pd) 11121{ 11122 dtrace_ecbdesc_t *new; 11123 dtrace_predicate_t *pred; 11124 dtrace_actdesc_t *act; 11125 11126 /* 11127 * We're going to create a new ECB description that matches the 11128 * specified ECB in every way, but has the specified probe description. 11129 */ 11130 new = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP); 11131 11132 if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL) 11133 dtrace_predicate_hold(pred); 11134 11135 for (act = ecb->dted_action; act != NULL; act = act->dtad_next) 11136 dtrace_actdesc_hold(act); 11137 11138 new->dted_action = ecb->dted_action; 11139 new->dted_pred = ecb->dted_pred; 11140 new->dted_probe = *pd; 11141 new->dted_uarg = ecb->dted_uarg; 11142 11143 dtrace_enabling_add(enab, new); 11144} 11145 11146static void 11147dtrace_enabling_dump(dtrace_enabling_t *enab) 11148{ 11149 int i; 11150 11151 for (i = 0; i < enab->dten_ndesc; i++) { 11152 dtrace_probedesc_t *desc = &enab->dten_desc[i]->dted_probe; 11153 11154 cmn_err(CE_NOTE, "enabling probe %d (%s:%s:%s:%s)", i, 11155 desc->dtpd_provider, desc->dtpd_mod, 11156 desc->dtpd_func, desc->dtpd_name); 11157 } 11158} 11159 11160static void 11161dtrace_enabling_destroy(dtrace_enabling_t *enab) 11162{ 11163 int i; 11164 dtrace_ecbdesc_t *ep; 11165 dtrace_vstate_t *vstate = enab->dten_vstate; 11166 11167 ASSERT(MUTEX_HELD(&dtrace_lock)); 11168 11169 for (i = 0; i < enab->dten_ndesc; i++) { 11170 dtrace_actdesc_t *act, *next; 11171 dtrace_predicate_t *pred; 11172 11173 ep = enab->dten_desc[i]; 11174 11175 if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) 11176 dtrace_predicate_release(pred, vstate); 11177 11178 for (act = ep->dted_action; act != NULL; act = next) { 11179 next = act->dtad_next; 11180 dtrace_actdesc_release(act, vstate); 11181 } 11182 11183 kmem_free(ep, sizeof (dtrace_ecbdesc_t)); 11184 } 11185 11186 if (enab->dten_desc != NULL) 11187 kmem_free(enab->dten_desc, 11188 enab->dten_maxdesc * sizeof (dtrace_enabling_t *)); 11189 11190 /* 11191 * If this was a retained enabling, decrement the dts_nretained count 11192 * and take it off of the dtrace_retained list. 11193 */ 11194 if (enab->dten_prev != NULL || enab->dten_next != NULL || 11195 dtrace_retained == enab) { 11196 ASSERT(enab->dten_vstate->dtvs_state != NULL); 11197 ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0); 11198 enab->dten_vstate->dtvs_state->dts_nretained--; 11199 } 11200 11201 if (enab->dten_prev == NULL) { 11202 if (dtrace_retained == enab) { 11203 dtrace_retained = enab->dten_next; 11204 11205 if (dtrace_retained != NULL) 11206 dtrace_retained->dten_prev = NULL; 11207 } 11208 } else { 11209 ASSERT(enab != dtrace_retained); 11210 ASSERT(dtrace_retained != NULL); 11211 enab->dten_prev->dten_next = enab->dten_next; 11212 } 11213 11214 if (enab->dten_next != NULL) { 11215 ASSERT(dtrace_retained != NULL); 11216 enab->dten_next->dten_prev = enab->dten_prev; 11217 } 11218 11219 kmem_free(enab, sizeof (dtrace_enabling_t)); 11220} 11221 11222static int 11223dtrace_enabling_retain(dtrace_enabling_t *enab) 11224{ 11225 dtrace_state_t *state; 11226 11227 ASSERT(MUTEX_HELD(&dtrace_lock)); 11228 ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); 11229 ASSERT(enab->dten_vstate != NULL); 11230 11231 state = enab->dten_vstate->dtvs_state; 11232 ASSERT(state != NULL); 11233 11234 /* 11235 * We only allow each state to retain dtrace_retain_max enablings. 11236 */ 11237 if (state->dts_nretained >= dtrace_retain_max) 11238 return (ENOSPC); 11239 11240 state->dts_nretained++; 11241 11242 if (dtrace_retained == NULL) { 11243 dtrace_retained = enab; 11244 return (0); 11245 } 11246 11247 enab->dten_next = dtrace_retained; 11248 dtrace_retained->dten_prev = enab; 11249 dtrace_retained = enab; 11250 11251 return (0); 11252} 11253 11254static int 11255dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match, 11256 dtrace_probedesc_t *create) 11257{ 11258 dtrace_enabling_t *new, *enab; 11259 int found = 0, err = ENOENT; 11260 11261 ASSERT(MUTEX_HELD(&dtrace_lock)); 11262 ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN); 11263 ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN); 11264 ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN); 11265 ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN); 11266 11267 new = dtrace_enabling_create(&state->dts_vstate); 11268 11269 /* 11270 * Iterate over all retained enablings, looking for enablings that 11271 * match the specified state. 11272 */ 11273 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { 11274 int i; 11275 11276 /* 11277 * dtvs_state can only be NULL for helper enablings -- and 11278 * helper enablings can't be retained. 11279 */ 11280 ASSERT(enab->dten_vstate->dtvs_state != NULL); 11281 11282 if (enab->dten_vstate->dtvs_state != state) 11283 continue; 11284 11285 /* 11286 * Now iterate over each probe description; we're looking for 11287 * an exact match to the specified probe description. 11288 */ 11289 for (i = 0; i < enab->dten_ndesc; i++) { 11290 dtrace_ecbdesc_t *ep = enab->dten_desc[i]; 11291 dtrace_probedesc_t *pd = &ep->dted_probe; 11292 11293 if (strcmp(pd->dtpd_provider, match->dtpd_provider)) 11294 continue; 11295 11296 if (strcmp(pd->dtpd_mod, match->dtpd_mod)) 11297 continue; 11298 11299 if (strcmp(pd->dtpd_func, match->dtpd_func)) 11300 continue; 11301 11302 if (strcmp(pd->dtpd_name, match->dtpd_name)) 11303 continue; 11304 11305 /* 11306 * We have a winning probe! Add it to our growing 11307 * enabling. 11308 */ 11309 found = 1; 11310 dtrace_enabling_addlike(new, ep, create); 11311 } 11312 } 11313 11314 if (!found || (err = dtrace_enabling_retain(new)) != 0) { 11315 dtrace_enabling_destroy(new); 11316 return (err); 11317 } 11318 11319 return (0); 11320} 11321 11322static void 11323dtrace_enabling_retract(dtrace_state_t *state) 11324{ 11325 dtrace_enabling_t *enab, *next; 11326 11327 ASSERT(MUTEX_HELD(&dtrace_lock)); 11328 11329 /* 11330 * Iterate over all retained enablings, destroy the enablings retained 11331 * for the specified state. 11332 */ 11333 for (enab = dtrace_retained; enab != NULL; enab = next) { 11334 next = enab->dten_next; 11335 11336 /* 11337 * dtvs_state can only be NULL for helper enablings -- and 11338 * helper enablings can't be retained. 11339 */ 11340 ASSERT(enab->dten_vstate->dtvs_state != NULL); 11341 11342 if (enab->dten_vstate->dtvs_state == state) { 11343 ASSERT(state->dts_nretained > 0); 11344 dtrace_enabling_destroy(enab); 11345 } 11346 } 11347 11348 ASSERT(state->dts_nretained == 0); 11349} 11350 11351static int 11352dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched) 11353{ 11354 int i = 0; 11355 int matched = 0; 11356 11357 ASSERT(MUTEX_HELD(&cpu_lock)); 11358 ASSERT(MUTEX_HELD(&dtrace_lock)); 11359 11360 for (i = 0; i < enab->dten_ndesc; i++) { 11361 dtrace_ecbdesc_t *ep = enab->dten_desc[i]; 11362 11363 enab->dten_current = ep; 11364 enab->dten_error = 0; 11365 11366 matched += dtrace_probe_enable(&ep->dted_probe, enab); 11367 11368 if (enab->dten_error != 0) { 11369 /* 11370 * If we get an error half-way through enabling the 11371 * probes, we kick out -- perhaps with some number of 11372 * them enabled. Leaving enabled probes enabled may 11373 * be slightly confusing for user-level, but we expect 11374 * that no one will attempt to actually drive on in 11375 * the face of such errors. If this is an anonymous 11376 * enabling (indicated with a NULL nmatched pointer), 11377 * we cmn_err() a message. We aren't expecting to 11378 * get such an error -- such as it can exist at all, 11379 * it would be a result of corrupted DOF in the driver 11380 * properties. 11381 */ 11382 if (nmatched == NULL) { 11383 cmn_err(CE_WARN, "dtrace_enabling_match() " 11384 "error on %p: %d", (void *)ep, 11385 enab->dten_error); 11386 } 11387 11388 return (enab->dten_error); 11389 } 11390 } 11391 11392 enab->dten_probegen = dtrace_probegen; 11393 if (nmatched != NULL) 11394 *nmatched = matched; 11395 11396 return (0); 11397} 11398 11399static void 11400dtrace_enabling_matchall(void) 11401{ 11402 dtrace_enabling_t *enab; 11403 11404 mutex_enter(&cpu_lock); 11405 mutex_enter(&dtrace_lock); 11406 11407 /* 11408 * Iterate over all retained enablings to see if any probes match 11409 * against them. We only perform this operation on enablings for which 11410 * we have sufficient permissions by virtue of being in the global zone 11411 * or in the same zone as the DTrace client. Because we can be called 11412 * after dtrace_detach() has been called, we cannot assert that there 11413 * are retained enablings. We can safely load from dtrace_retained, 11414 * however: the taskq_destroy() at the end of dtrace_detach() will 11415 * block pending our completion. 11416 */ 11417 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { 11418#if defined(sun) 11419 cred_t *cr = enab->dten_vstate->dtvs_state->dts_cred.dcr_cred; 11420 11421 if (INGLOBALZONE(curproc) || getzoneid() == crgetzoneid(cr)) 11422#endif 11423 (void) dtrace_enabling_match(enab, NULL); 11424 } 11425 11426 mutex_exit(&dtrace_lock); 11427 mutex_exit(&cpu_lock); 11428} 11429 11430/* 11431 * If an enabling is to be enabled without having matched probes (that is, if 11432 * dtrace_state_go() is to be called on the underlying dtrace_state_t), the 11433 * enabling must be _primed_ by creating an ECB for every ECB description. 11434 * This must be done to assure that we know the number of speculations, the 11435 * number of aggregations, the minimum buffer size needed, etc. before we 11436 * transition out of DTRACE_ACTIVITY_INACTIVE. To do this without actually 11437 * enabling any probes, we create ECBs for every ECB decription, but with a 11438 * NULL probe -- which is exactly what this function does. 11439 */ 11440static void 11441dtrace_enabling_prime(dtrace_state_t *state) 11442{ 11443 dtrace_enabling_t *enab; 11444 int i; 11445 11446 for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { 11447 ASSERT(enab->dten_vstate->dtvs_state != NULL); 11448 11449 if (enab->dten_vstate->dtvs_state != state) 11450 continue; 11451 11452 /* 11453 * We don't want to prime an enabling more than once, lest 11454 * we allow a malicious user to induce resource exhaustion. 11455 * (The ECBs that result from priming an enabling aren't 11456 * leaked -- but they also aren't deallocated until the 11457 * consumer state is destroyed.) 11458 */ 11459 if (enab->dten_primed) 11460 continue; 11461 11462 for (i = 0; i < enab->dten_ndesc; i++) { 11463 enab->dten_current = enab->dten_desc[i]; 11464 (void) dtrace_probe_enable(NULL, enab); 11465 } 11466 11467 enab->dten_primed = 1; 11468 } 11469} 11470 11471/* 11472 * Called to indicate that probes should be provided due to retained 11473 * enablings. This is implemented in terms of dtrace_probe_provide(), but it 11474 * must take an initial lap through the enabling calling the dtps_provide() 11475 * entry point explicitly to allow for autocreated probes. 11476 */ 11477static void 11478dtrace_enabling_provide(dtrace_provider_t *prv) 11479{ 11480 int i, all = 0; 11481 dtrace_probedesc_t desc; 11482 11483 ASSERT(MUTEX_HELD(&dtrace_lock)); 11484 ASSERT(MUTEX_HELD(&dtrace_provider_lock)); 11485 11486 if (prv == NULL) { 11487 all = 1; 11488 prv = dtrace_provider; 11489 } 11490 11491 do { 11492 dtrace_enabling_t *enab = dtrace_retained; 11493 void *parg = prv->dtpv_arg; 11494 11495 for (; enab != NULL; enab = enab->dten_next) { 11496 for (i = 0; i < enab->dten_ndesc; i++) { 11497 desc = enab->dten_desc[i]->dted_probe; 11498 mutex_exit(&dtrace_lock); 11499 prv->dtpv_pops.dtps_provide(parg, &desc); 11500 mutex_enter(&dtrace_lock); 11501 } 11502 } 11503 } while (all && (prv = prv->dtpv_next) != NULL); 11504 11505 mutex_exit(&dtrace_lock); 11506 dtrace_probe_provide(NULL, all ? NULL : prv); 11507 mutex_enter(&dtrace_lock); 11508} 11509 11510/* 11511 * DTrace DOF Functions 11512 */ 11513/*ARGSUSED*/ 11514static void 11515dtrace_dof_error(dof_hdr_t *dof, const char *str) 11516{ 11517 if (dtrace_err_verbose) 11518 cmn_err(CE_WARN, "failed to process DOF: %s", str); 11519 11520#ifdef DTRACE_ERRDEBUG 11521 dtrace_errdebug(str); 11522#endif 11523} 11524 11525/* 11526 * Create DOF out of a currently enabled state. Right now, we only create 11527 * DOF containing the run-time options -- but this could be expanded to create 11528 * complete DOF representing the enabled state. 11529 */ 11530static dof_hdr_t * 11531dtrace_dof_create(dtrace_state_t *state) 11532{ 11533 dof_hdr_t *dof; 11534 dof_sec_t *sec; 11535 dof_optdesc_t *opt; 11536 int i, len = sizeof (dof_hdr_t) + 11537 roundup(sizeof (dof_sec_t), sizeof (uint64_t)) + 11538 sizeof (dof_optdesc_t) * DTRACEOPT_MAX; 11539 11540 ASSERT(MUTEX_HELD(&dtrace_lock)); 11541 11542 dof = kmem_zalloc(len, KM_SLEEP); 11543 dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0; 11544 dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1; 11545 dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2; 11546 dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3; 11547 11548 dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE; 11549 dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE; 11550 dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION; 11551 dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION; 11552 dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS; 11553 dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS; 11554 11555 dof->dofh_flags = 0; 11556 dof->dofh_hdrsize = sizeof (dof_hdr_t); 11557 dof->dofh_secsize = sizeof (dof_sec_t); 11558 dof->dofh_secnum = 1; /* only DOF_SECT_OPTDESC */ 11559 dof->dofh_secoff = sizeof (dof_hdr_t); 11560 dof->dofh_loadsz = len; 11561 dof->dofh_filesz = len; 11562 dof->dofh_pad = 0; 11563 11564 /* 11565 * Fill in the option section header... 11566 */ 11567 sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t)); 11568 sec->dofs_type = DOF_SECT_OPTDESC; 11569 sec->dofs_align = sizeof (uint64_t); 11570 sec->dofs_flags = DOF_SECF_LOAD; 11571 sec->dofs_entsize = sizeof (dof_optdesc_t); 11572 11573 opt = (dof_optdesc_t *)((uintptr_t)sec + 11574 roundup(sizeof (dof_sec_t), sizeof (uint64_t))); 11575 11576 sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof; 11577 sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX; 11578 11579 for (i = 0; i < DTRACEOPT_MAX; i++) { 11580 opt[i].dofo_option = i; 11581 opt[i].dofo_strtab = DOF_SECIDX_NONE; 11582 opt[i].dofo_value = state->dts_options[i]; 11583 } 11584 11585 return (dof); 11586} 11587 11588static dof_hdr_t * 11589dtrace_dof_copyin(uintptr_t uarg, int *errp) 11590{ 11591 dof_hdr_t hdr, *dof; 11592 11593 ASSERT(!MUTEX_HELD(&dtrace_lock)); 11594 11595 /* 11596 * First, we're going to copyin() the sizeof (dof_hdr_t). 11597 */ 11598 if (copyin((void *)uarg, &hdr, sizeof (hdr)) != 0) { 11599 dtrace_dof_error(NULL, "failed to copyin DOF header"); 11600 *errp = EFAULT; 11601 return (NULL); 11602 } 11603 11604 /* 11605 * Now we'll allocate the entire DOF and copy it in -- provided 11606 * that the length isn't outrageous. 11607 */ 11608 if (hdr.dofh_loadsz >= dtrace_dof_maxsize) { 11609 dtrace_dof_error(&hdr, "load size exceeds maximum"); 11610 *errp = E2BIG; 11611 return (NULL); 11612 } 11613 11614 if (hdr.dofh_loadsz < sizeof (hdr)) { 11615 dtrace_dof_error(&hdr, "invalid load size"); 11616 *errp = EINVAL; 11617 return (NULL); 11618 } 11619 11620 dof = kmem_alloc(hdr.dofh_loadsz, KM_SLEEP); 11621 11622 if (copyin((void *)uarg, dof, hdr.dofh_loadsz) != 0) { 11623 kmem_free(dof, hdr.dofh_loadsz); 11624 *errp = EFAULT; 11625 return (NULL); 11626 } 11627 11628 return (dof); 11629} 11630 11631#if !defined(sun) 11632static __inline uchar_t 11633dtrace_dof_char(char c) { 11634 switch (c) { 11635 case '0': 11636 case '1': 11637 case '2': 11638 case '3': 11639 case '4': 11640 case '5': 11641 case '6': 11642 case '7': 11643 case '8': 11644 case '9': 11645 return (c - '0'); 11646 case 'A': 11647 case 'B': 11648 case 'C': 11649 case 'D': 11650 case 'E': 11651 case 'F': 11652 return (c - 'A' + 10); 11653 case 'a': 11654 case 'b': 11655 case 'c': 11656 case 'd': 11657 case 'e': 11658 case 'f': 11659 return (c - 'a' + 10); 11660 } 11661 /* Should not reach here. */ 11662 return (0); 11663} 11664#endif 11665 11666static dof_hdr_t * 11667dtrace_dof_property(const char *name) 11668{ 11669 uchar_t *buf; 11670 uint64_t loadsz; 11671 unsigned int len, i; 11672 dof_hdr_t *dof; 11673 11674#if defined(sun) 11675 /* 11676 * Unfortunately, array of values in .conf files are always (and 11677 * only) interpreted to be integer arrays. We must read our DOF 11678 * as an integer array, and then squeeze it into a byte array. 11679 */ 11680 if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0, 11681 (char *)name, (int **)&buf, &len) != DDI_PROP_SUCCESS) 11682 return (NULL); 11683 11684 for (i = 0; i < len; i++) 11685 buf[i] = (uchar_t)(((int *)buf)[i]); 11686 11687 if (len < sizeof (dof_hdr_t)) { 11688 ddi_prop_free(buf); 11689 dtrace_dof_error(NULL, "truncated header"); 11690 return (NULL); 11691 } 11692 11693 if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) { 11694 ddi_prop_free(buf); 11695 dtrace_dof_error(NULL, "truncated DOF"); 11696 return (NULL); 11697 } 11698 11699 if (loadsz >= dtrace_dof_maxsize) { 11700 ddi_prop_free(buf); 11701 dtrace_dof_error(NULL, "oversized DOF"); 11702 return (NULL); 11703 } 11704 11705 dof = kmem_alloc(loadsz, KM_SLEEP); 11706 bcopy(buf, dof, loadsz); 11707 ddi_prop_free(buf); 11708#else 11709 char *p; 11710 char *p_env; 11711 11712 if ((p_env = getenv(name)) == NULL) 11713 return (NULL); 11714 11715 len = strlen(p_env) / 2; 11716 11717 buf = kmem_alloc(len, KM_SLEEP); 11718 11719 dof = (dof_hdr_t *) buf; 11720 11721 p = p_env; 11722 11723 for (i = 0; i < len; i++) { 11724 buf[i] = (dtrace_dof_char(p[0]) << 4) | 11725 dtrace_dof_char(p[1]); 11726 p += 2; 11727 } 11728 11729 freeenv(p_env); 11730 11731 if (len < sizeof (dof_hdr_t)) { 11732 kmem_free(buf, 0); 11733 dtrace_dof_error(NULL, "truncated header"); 11734 return (NULL); 11735 } 11736 11737 if (len < (loadsz = dof->dofh_loadsz)) { 11738 kmem_free(buf, 0); 11739 dtrace_dof_error(NULL, "truncated DOF"); 11740 return (NULL); 11741 } 11742 11743 if (loadsz >= dtrace_dof_maxsize) { 11744 kmem_free(buf, 0); 11745 dtrace_dof_error(NULL, "oversized DOF"); 11746 return (NULL); 11747 } 11748#endif 11749 11750 return (dof); 11751} 11752 11753static void 11754dtrace_dof_destroy(dof_hdr_t *dof) 11755{ 11756 kmem_free(dof, dof->dofh_loadsz); 11757} 11758 11759/* 11760 * Return the dof_sec_t pointer corresponding to a given section index. If the 11761 * index is not valid, dtrace_dof_error() is called and NULL is returned. If 11762 * a type other than DOF_SECT_NONE is specified, the header is checked against 11763 * this type and NULL is returned if the types do not match. 11764 */ 11765static dof_sec_t * 11766dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, dof_secidx_t i) 11767{ 11768 dof_sec_t *sec = (dof_sec_t *)(uintptr_t) 11769 ((uintptr_t)dof + dof->dofh_secoff + i * dof->dofh_secsize); 11770 11771 if (i >= dof->dofh_secnum) { 11772 dtrace_dof_error(dof, "referenced section index is invalid"); 11773 return (NULL); 11774 } 11775 11776 if (!(sec->dofs_flags & DOF_SECF_LOAD)) { 11777 dtrace_dof_error(dof, "referenced section is not loadable"); 11778 return (NULL); 11779 } 11780 11781 if (type != DOF_SECT_NONE && type != sec->dofs_type) { 11782 dtrace_dof_error(dof, "referenced section is the wrong type"); 11783 return (NULL); 11784 } 11785 11786 return (sec); 11787} 11788 11789static dtrace_probedesc_t * 11790dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_probedesc_t *desc) 11791{ 11792 dof_probedesc_t *probe; 11793 dof_sec_t *strtab; 11794 uintptr_t daddr = (uintptr_t)dof; 11795 uintptr_t str; 11796 size_t size; 11797 11798 if (sec->dofs_type != DOF_SECT_PROBEDESC) { 11799 dtrace_dof_error(dof, "invalid probe section"); 11800 return (NULL); 11801 } 11802 11803 if (sec->dofs_align != sizeof (dof_secidx_t)) { 11804 dtrace_dof_error(dof, "bad alignment in probe description"); 11805 return (NULL); 11806 } 11807 11808 if (sec->dofs_offset + sizeof (dof_probedesc_t) > dof->dofh_loadsz) { 11809 dtrace_dof_error(dof, "truncated probe description"); 11810 return (NULL); 11811 } 11812 11813 probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset); 11814 strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab); 11815 11816 if (strtab == NULL) 11817 return (NULL); 11818 11819 str = daddr + strtab->dofs_offset; 11820 size = strtab->dofs_size; 11821 11822 if (probe->dofp_provider >= strtab->dofs_size) { 11823 dtrace_dof_error(dof, "corrupt probe provider"); 11824 return (NULL); 11825 } 11826 11827 (void) strncpy(desc->dtpd_provider, 11828 (char *)(str + probe->dofp_provider), 11829 MIN(DTRACE_PROVNAMELEN - 1, size - probe->dofp_provider)); 11830 11831 if (probe->dofp_mod >= strtab->dofs_size) { 11832 dtrace_dof_error(dof, "corrupt probe module"); 11833 return (NULL); 11834 } 11835 11836 (void) strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod), 11837 MIN(DTRACE_MODNAMELEN - 1, size - probe->dofp_mod)); 11838 11839 if (probe->dofp_func >= strtab->dofs_size) { 11840 dtrace_dof_error(dof, "corrupt probe function"); 11841 return (NULL); 11842 } 11843 11844 (void) strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func), 11845 MIN(DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func)); 11846 11847 if (probe->dofp_name >= strtab->dofs_size) { 11848 dtrace_dof_error(dof, "corrupt probe name"); 11849 return (NULL); 11850 } 11851 11852 (void) strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name), 11853 MIN(DTRACE_NAMELEN - 1, size - probe->dofp_name)); 11854 11855 return (desc); 11856} 11857 11858static dtrace_difo_t * 11859dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, 11860 cred_t *cr) 11861{ 11862 dtrace_difo_t *dp; 11863 size_t ttl = 0; 11864 dof_difohdr_t *dofd; 11865 uintptr_t daddr = (uintptr_t)dof; 11866 size_t max = dtrace_difo_maxsize; 11867 int i, l, n; 11868 11869 static const struct { 11870 int section; 11871 int bufoffs; 11872 int lenoffs; 11873 int entsize; 11874 int align; 11875 const char *msg; 11876 } difo[] = { 11877 { DOF_SECT_DIF, offsetof(dtrace_difo_t, dtdo_buf), 11878 offsetof(dtrace_difo_t, dtdo_len), sizeof (dif_instr_t), 11879 sizeof (dif_instr_t), "multiple DIF sections" }, 11880 11881 { DOF_SECT_INTTAB, offsetof(dtrace_difo_t, dtdo_inttab), 11882 offsetof(dtrace_difo_t, dtdo_intlen), sizeof (uint64_t), 11883 sizeof (uint64_t), "multiple integer tables" }, 11884 11885 { DOF_SECT_STRTAB, offsetof(dtrace_difo_t, dtdo_strtab), 11886 offsetof(dtrace_difo_t, dtdo_strlen), 0, 11887 sizeof (char), "multiple string tables" }, 11888 11889 { DOF_SECT_VARTAB, offsetof(dtrace_difo_t, dtdo_vartab), 11890 offsetof(dtrace_difo_t, dtdo_varlen), sizeof (dtrace_difv_t), 11891 sizeof (uint_t), "multiple variable tables" }, 11892 11893 { DOF_SECT_NONE, 0, 0, 0, 0, NULL } 11894 }; 11895 11896 if (sec->dofs_type != DOF_SECT_DIFOHDR) { 11897 dtrace_dof_error(dof, "invalid DIFO header section"); 11898 return (NULL); 11899 } 11900 11901 if (sec->dofs_align != sizeof (dof_secidx_t)) { 11902 dtrace_dof_error(dof, "bad alignment in DIFO header"); 11903 return (NULL); 11904 } 11905 11906 if (sec->dofs_size < sizeof (dof_difohdr_t) || 11907 sec->dofs_size % sizeof (dof_secidx_t)) { 11908 dtrace_dof_error(dof, "bad size in DIFO header"); 11909 return (NULL); 11910 } 11911 11912 dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); 11913 n = (sec->dofs_size - sizeof (*dofd)) / sizeof (dof_secidx_t) + 1; 11914 11915 dp = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP); 11916 dp->dtdo_rtype = dofd->dofd_rtype; 11917 11918 for (l = 0; l < n; l++) { 11919 dof_sec_t *subsec; 11920 void **bufp; 11921 uint32_t *lenp; 11922 11923 if ((subsec = dtrace_dof_sect(dof, DOF_SECT_NONE, 11924 dofd->dofd_links[l])) == NULL) 11925 goto err; /* invalid section link */ 11926 11927 if (ttl + subsec->dofs_size > max) { 11928 dtrace_dof_error(dof, "exceeds maximum size"); 11929 goto err; 11930 } 11931 11932 ttl += subsec->dofs_size; 11933 11934 for (i = 0; difo[i].section != DOF_SECT_NONE; i++) { 11935 if (subsec->dofs_type != difo[i].section) 11936 continue; 11937 11938 if (!(subsec->dofs_flags & DOF_SECF_LOAD)) { 11939 dtrace_dof_error(dof, "section not loaded"); 11940 goto err; 11941 } 11942 11943 if (subsec->dofs_align != difo[i].align) { 11944 dtrace_dof_error(dof, "bad alignment"); 11945 goto err; 11946 } 11947 11948 bufp = (void **)((uintptr_t)dp + difo[i].bufoffs); 11949 lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs); 11950 11951 if (*bufp != NULL) { 11952 dtrace_dof_error(dof, difo[i].msg); 11953 goto err; 11954 } 11955 11956 if (difo[i].entsize != subsec->dofs_entsize) { 11957 dtrace_dof_error(dof, "entry size mismatch"); 11958 goto err; 11959 } 11960 11961 if (subsec->dofs_entsize != 0 && 11962 (subsec->dofs_size % subsec->dofs_entsize) != 0) { 11963 dtrace_dof_error(dof, "corrupt entry size"); 11964 goto err; 11965 } 11966 11967 *lenp = subsec->dofs_size; 11968 *bufp = kmem_alloc(subsec->dofs_size, KM_SLEEP); 11969 bcopy((char *)(uintptr_t)(daddr + subsec->dofs_offset), 11970 *bufp, subsec->dofs_size); 11971 11972 if (subsec->dofs_entsize != 0) 11973 *lenp /= subsec->dofs_entsize; 11974 11975 break; 11976 } 11977 11978 /* 11979 * If we encounter a loadable DIFO sub-section that is not 11980 * known to us, assume this is a broken program and fail. 11981 */ 11982 if (difo[i].section == DOF_SECT_NONE && 11983 (subsec->dofs_flags & DOF_SECF_LOAD)) { 11984 dtrace_dof_error(dof, "unrecognized DIFO subsection"); 11985 goto err; 11986 } 11987 } 11988 11989 if (dp->dtdo_buf == NULL) { 11990 /* 11991 * We can't have a DIF object without DIF text. 11992 */ 11993 dtrace_dof_error(dof, "missing DIF text"); 11994 goto err; 11995 } 11996 11997 /* 11998 * Before we validate the DIF object, run through the variable table 11999 * looking for the strings -- if any of their size are under, we'll set 12000 * their size to be the system-wide default string size. Note that 12001 * this should _not_ happen if the "strsize" option has been set -- 12002 * in this case, the compiler should have set the size to reflect the 12003 * setting of the option. 12004 */ 12005 for (i = 0; i < dp->dtdo_varlen; i++) { 12006 dtrace_difv_t *v = &dp->dtdo_vartab[i]; 12007 dtrace_diftype_t *t = &v->dtdv_type; 12008 12009 if (v->dtdv_id < DIF_VAR_OTHER_UBASE) 12010 continue; 12011 12012 if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0) 12013 t->dtdt_size = dtrace_strsize_default; 12014 } 12015 12016 if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0) 12017 goto err; 12018 12019 dtrace_difo_init(dp, vstate); 12020 return (dp); 12021 12022err: 12023 kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t)); 12024 kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t)); 12025 kmem_free(dp->dtdo_strtab, dp->dtdo_strlen); 12026 kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t)); 12027 12028 kmem_free(dp, sizeof (dtrace_difo_t)); 12029 return (NULL); 12030} 12031 12032static dtrace_predicate_t * 12033dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, 12034 cred_t *cr) 12035{ 12036 dtrace_difo_t *dp; 12037 12038 if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL) 12039 return (NULL); 12040 12041 return (dtrace_predicate_create(dp)); 12042} 12043 12044static dtrace_actdesc_t * 12045dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, 12046 cred_t *cr) 12047{ 12048 dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next; 12049 dof_actdesc_t *desc; 12050 dof_sec_t *difosec; 12051 size_t offs; 12052 uintptr_t daddr = (uintptr_t)dof; 12053 uint64_t arg; 12054 dtrace_actkind_t kind; 12055 12056 if (sec->dofs_type != DOF_SECT_ACTDESC) { 12057 dtrace_dof_error(dof, "invalid action section"); 12058 return (NULL); 12059 } 12060 12061 if (sec->dofs_offset + sizeof (dof_actdesc_t) > dof->dofh_loadsz) { 12062 dtrace_dof_error(dof, "truncated action description"); 12063 return (NULL); 12064 } 12065 12066 if (sec->dofs_align != sizeof (uint64_t)) { 12067 dtrace_dof_error(dof, "bad alignment in action description"); 12068 return (NULL); 12069 } 12070 12071 if (sec->dofs_size < sec->dofs_entsize) { 12072 dtrace_dof_error(dof, "section entry size exceeds total size"); 12073 return (NULL); 12074 } 12075 12076 if (sec->dofs_entsize != sizeof (dof_actdesc_t)) { 12077 dtrace_dof_error(dof, "bad entry size in action description"); 12078 return (NULL); 12079 } 12080 12081 if (sec->dofs_size / sec->dofs_entsize > dtrace_actions_max) { 12082 dtrace_dof_error(dof, "actions exceed dtrace_actions_max"); 12083 return (NULL); 12084 } 12085 12086 for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) { 12087 desc = (dof_actdesc_t *)(daddr + 12088 (uintptr_t)sec->dofs_offset + offs); 12089 kind = (dtrace_actkind_t)desc->dofa_kind; 12090 12091 if (DTRACEACT_ISPRINTFLIKE(kind) && 12092 (kind != DTRACEACT_PRINTA || 12093 desc->dofa_strtab != DOF_SECIDX_NONE)) { 12094 dof_sec_t *strtab; 12095 char *str, *fmt; 12096 uint64_t i; 12097 12098 /* 12099 * printf()-like actions must have a format string. 12100 */ 12101 if ((strtab = dtrace_dof_sect(dof, 12102 DOF_SECT_STRTAB, desc->dofa_strtab)) == NULL) 12103 goto err; 12104 12105 str = (char *)((uintptr_t)dof + 12106 (uintptr_t)strtab->dofs_offset); 12107 12108 for (i = desc->dofa_arg; i < strtab->dofs_size; i++) { 12109 if (str[i] == '\0') 12110 break; 12111 } 12112 12113 if (i >= strtab->dofs_size) { 12114 dtrace_dof_error(dof, "bogus format string"); 12115 goto err; 12116 } 12117 12118 if (i == desc->dofa_arg) { 12119 dtrace_dof_error(dof, "empty format string"); 12120 goto err; 12121 } 12122 12123 i -= desc->dofa_arg; 12124 fmt = kmem_alloc(i + 1, KM_SLEEP); 12125 bcopy(&str[desc->dofa_arg], fmt, i + 1); 12126 arg = (uint64_t)(uintptr_t)fmt; 12127 } else { 12128 if (kind == DTRACEACT_PRINTA) { 12129 ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE); 12130 arg = 0; 12131 } else { 12132 arg = desc->dofa_arg; 12133 } 12134 } 12135 12136 act = dtrace_actdesc_create(kind, desc->dofa_ntuple, 12137 desc->dofa_uarg, arg); 12138 12139 if (last != NULL) { 12140 last->dtad_next = act; 12141 } else { 12142 first = act; 12143 } 12144 12145 last = act; 12146 12147 if (desc->dofa_difo == DOF_SECIDX_NONE) 12148 continue; 12149 12150 if ((difosec = dtrace_dof_sect(dof, 12151 DOF_SECT_DIFOHDR, desc->dofa_difo)) == NULL) 12152 goto err; 12153 12154 act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr); 12155 12156 if (act->dtad_difo == NULL) 12157 goto err; 12158 } 12159 12160 ASSERT(first != NULL); 12161 return (first); 12162 12163err: 12164 for (act = first; act != NULL; act = next) { 12165 next = act->dtad_next; 12166 dtrace_actdesc_release(act, vstate); 12167 } 12168 12169 return (NULL); 12170} 12171 12172static dtrace_ecbdesc_t * 12173dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, 12174 cred_t *cr) 12175{ 12176 dtrace_ecbdesc_t *ep; 12177 dof_ecbdesc_t *ecb; 12178 dtrace_probedesc_t *desc; 12179 dtrace_predicate_t *pred = NULL; 12180 12181 if (sec->dofs_size < sizeof (dof_ecbdesc_t)) { 12182 dtrace_dof_error(dof, "truncated ECB description"); 12183 return (NULL); 12184 } 12185 12186 if (sec->dofs_align != sizeof (uint64_t)) { 12187 dtrace_dof_error(dof, "bad alignment in ECB description"); 12188 return (NULL); 12189 } 12190 12191 ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset); 12192 sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes); 12193 12194 if (sec == NULL) 12195 return (NULL); 12196 12197 ep = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP); 12198 ep->dted_uarg = ecb->dofe_uarg; 12199 desc = &ep->dted_probe; 12200 12201 if (dtrace_dof_probedesc(dof, sec, desc) == NULL) 12202 goto err; 12203 12204 if (ecb->dofe_pred != DOF_SECIDX_NONE) { 12205 if ((sec = dtrace_dof_sect(dof, 12206 DOF_SECT_DIFOHDR, ecb->dofe_pred)) == NULL) 12207 goto err; 12208 12209 if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL) 12210 goto err; 12211 12212 ep->dted_pred.dtpdd_predicate = pred; 12213 } 12214 12215 if (ecb->dofe_actions != DOF_SECIDX_NONE) { 12216 if ((sec = dtrace_dof_sect(dof, 12217 DOF_SECT_ACTDESC, ecb->dofe_actions)) == NULL) 12218 goto err; 12219 12220 ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr); 12221 12222 if (ep->dted_action == NULL) 12223 goto err; 12224 } 12225 12226 return (ep); 12227 12228err: 12229 if (pred != NULL) 12230 dtrace_predicate_release(pred, vstate); 12231 kmem_free(ep, sizeof (dtrace_ecbdesc_t)); 12232 return (NULL); 12233} 12234 12235/* 12236 * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the 12237 * specified DOF. At present, this amounts to simply adding 'ubase' to the 12238 * site of any user SETX relocations to account for load object base address. 12239 * In the future, if we need other relocations, this function can be extended. 12240 */ 12241static int 12242dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase) 12243{ 12244 uintptr_t daddr = (uintptr_t)dof; 12245 dof_relohdr_t *dofr = 12246 (dof_relohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); 12247 dof_sec_t *ss, *rs, *ts; 12248 dof_relodesc_t *r; 12249 uint_t i, n; 12250 12251 if (sec->dofs_size < sizeof (dof_relohdr_t) || 12252 sec->dofs_align != sizeof (dof_secidx_t)) { 12253 dtrace_dof_error(dof, "invalid relocation header"); 12254 return (-1); 12255 } 12256 12257 ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab); 12258 rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec); 12259 ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec); 12260 12261 if (ss == NULL || rs == NULL || ts == NULL) 12262 return (-1); /* dtrace_dof_error() has been called already */ 12263 12264 if (rs->dofs_entsize < sizeof (dof_relodesc_t) || 12265 rs->dofs_align != sizeof (uint64_t)) { 12266 dtrace_dof_error(dof, "invalid relocation section"); 12267 return (-1); 12268 } 12269 12270 r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset); 12271 n = rs->dofs_size / rs->dofs_entsize; 12272 12273 for (i = 0; i < n; i++) { 12274 uintptr_t taddr = daddr + ts->dofs_offset + r->dofr_offset; 12275 12276 switch (r->dofr_type) { 12277 case DOF_RELO_NONE: 12278 break; 12279 case DOF_RELO_SETX: 12280 if (r->dofr_offset >= ts->dofs_size || r->dofr_offset + 12281 sizeof (uint64_t) > ts->dofs_size) { 12282 dtrace_dof_error(dof, "bad relocation offset"); 12283 return (-1); 12284 } 12285 12286 if (!IS_P2ALIGNED(taddr, sizeof (uint64_t))) { 12287 dtrace_dof_error(dof, "misaligned setx relo"); 12288 return (-1); 12289 } 12290 12291 *(uint64_t *)taddr += ubase; 12292 break; 12293 default: 12294 dtrace_dof_error(dof, "invalid relocation type"); 12295 return (-1); 12296 } 12297 12298 r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize); 12299 } 12300 12301 return (0); 12302} 12303 12304/* 12305 * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated 12306 * header: it should be at the front of a memory region that is at least 12307 * sizeof (dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in 12308 * size. It need not be validated in any other way. 12309 */ 12310static int 12311dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr, 12312 dtrace_enabling_t **enabp, uint64_t ubase, int noprobes) 12313{ 12314 uint64_t len = dof->dofh_loadsz, seclen; 12315 uintptr_t daddr = (uintptr_t)dof; 12316 dtrace_ecbdesc_t *ep; 12317 dtrace_enabling_t *enab; 12318 uint_t i; 12319 12320 ASSERT(MUTEX_HELD(&dtrace_lock)); 12321 ASSERT(dof->dofh_loadsz >= sizeof (dof_hdr_t)); 12322 12323 /* 12324 * Check the DOF header identification bytes. In addition to checking 12325 * valid settings, we also verify that unused bits/bytes are zeroed so 12326 * we can use them later without fear of regressing existing binaries. 12327 */ 12328 if (bcmp(&dof->dofh_ident[DOF_ID_MAG0], 12329 DOF_MAG_STRING, DOF_MAG_STRLEN) != 0) { 12330 dtrace_dof_error(dof, "DOF magic string mismatch"); 12331 return (-1); 12332 } 12333 12334 if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 && 12335 dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) { 12336 dtrace_dof_error(dof, "DOF has invalid data model"); 12337 return (-1); 12338 } 12339 12340 if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) { 12341 dtrace_dof_error(dof, "DOF encoding mismatch"); 12342 return (-1); 12343 } 12344 12345 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && 12346 dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) { 12347 dtrace_dof_error(dof, "DOF version mismatch"); 12348 return (-1); 12349 } 12350 12351 if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) { 12352 dtrace_dof_error(dof, "DOF uses unsupported instruction set"); 12353 return (-1); 12354 } 12355 12356 if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) { 12357 dtrace_dof_error(dof, "DOF uses too many integer registers"); 12358 return (-1); 12359 } 12360 12361 if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) { 12362 dtrace_dof_error(dof, "DOF uses too many tuple registers"); 12363 return (-1); 12364 } 12365 12366 for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) { 12367 if (dof->dofh_ident[i] != 0) { 12368 dtrace_dof_error(dof, "DOF has invalid ident byte set"); 12369 return (-1); 12370 } 12371 } 12372 12373 if (dof->dofh_flags & ~DOF_FL_VALID) { 12374 dtrace_dof_error(dof, "DOF has invalid flag bits set"); 12375 return (-1); 12376 } 12377 12378 if (dof->dofh_secsize == 0) { 12379 dtrace_dof_error(dof, "zero section header size"); 12380 return (-1); 12381 } 12382 12383 /* 12384 * Check that the section headers don't exceed the amount of DOF 12385 * data. Note that we cast the section size and number of sections 12386 * to uint64_t's to prevent possible overflow in the multiplication. 12387 */ 12388 seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize; 12389 12390 if (dof->dofh_secoff > len || seclen > len || 12391 dof->dofh_secoff + seclen > len) { 12392 dtrace_dof_error(dof, "truncated section headers"); 12393 return (-1); 12394 } 12395 12396 if (!IS_P2ALIGNED(dof->dofh_secoff, sizeof (uint64_t))) { 12397 dtrace_dof_error(dof, "misaligned section headers"); 12398 return (-1); 12399 } 12400 12401 if (!IS_P2ALIGNED(dof->dofh_secsize, sizeof (uint64_t))) { 12402 dtrace_dof_error(dof, "misaligned section size"); 12403 return (-1); 12404 } 12405 12406 /* 12407 * Take an initial pass through the section headers to be sure that 12408 * the headers don't have stray offsets. If the 'noprobes' flag is 12409 * set, do not permit sections relating to providers, probes, or args. 12410 */ 12411 for (i = 0; i < dof->dofh_secnum; i++) { 12412 dof_sec_t *sec = (dof_sec_t *)(daddr + 12413 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); 12414 12415 if (noprobes) { 12416 switch (sec->dofs_type) { 12417 case DOF_SECT_PROVIDER: 12418 case DOF_SECT_PROBES: 12419 case DOF_SECT_PRARGS: 12420 case DOF_SECT_PROFFS: 12421 dtrace_dof_error(dof, "illegal sections " 12422 "for enabling"); 12423 return (-1); 12424 } 12425 } 12426 12427 if (!(sec->dofs_flags & DOF_SECF_LOAD)) 12428 continue; /* just ignore non-loadable sections */ 12429 12430 if (sec->dofs_align & (sec->dofs_align - 1)) { 12431 dtrace_dof_error(dof, "bad section alignment"); 12432 return (-1); 12433 } 12434 12435 if (sec->dofs_offset & (sec->dofs_align - 1)) { 12436 dtrace_dof_error(dof, "misaligned section"); 12437 return (-1); 12438 } 12439 12440 if (sec->dofs_offset > len || sec->dofs_size > len || 12441 sec->dofs_offset + sec->dofs_size > len) { 12442 dtrace_dof_error(dof, "corrupt section header"); 12443 return (-1); 12444 } 12445 12446 if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr + 12447 sec->dofs_offset + sec->dofs_size - 1) != '\0') { 12448 dtrace_dof_error(dof, "non-terminating string table"); 12449 return (-1); 12450 } 12451 } 12452 12453 /* 12454 * Take a second pass through the sections and locate and perform any 12455 * relocations that are present. We do this after the first pass to 12456 * be sure that all sections have had their headers validated. 12457 */ 12458 for (i = 0; i < dof->dofh_secnum; i++) { 12459 dof_sec_t *sec = (dof_sec_t *)(daddr + 12460 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); 12461 12462 if (!(sec->dofs_flags & DOF_SECF_LOAD)) 12463 continue; /* skip sections that are not loadable */ 12464 12465 switch (sec->dofs_type) { 12466 case DOF_SECT_URELHDR: 12467 if (dtrace_dof_relocate(dof, sec, ubase) != 0) 12468 return (-1); 12469 break; 12470 } 12471 } 12472 12473 if ((enab = *enabp) == NULL) 12474 enab = *enabp = dtrace_enabling_create(vstate); 12475 12476 for (i = 0; i < dof->dofh_secnum; i++) { 12477 dof_sec_t *sec = (dof_sec_t *)(daddr + 12478 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); 12479 12480 if (sec->dofs_type != DOF_SECT_ECBDESC) 12481 continue; 12482 12483 if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) { 12484 dtrace_enabling_destroy(enab); 12485 *enabp = NULL; 12486 return (-1); 12487 } 12488 12489 dtrace_enabling_add(enab, ep); 12490 } 12491 12492 return (0); 12493} 12494 12495/* 12496 * Process DOF for any options. This routine assumes that the DOF has been 12497 * at least processed by dtrace_dof_slurp(). 12498 */ 12499static int 12500dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state) 12501{ 12502 int i, rval; 12503 uint32_t entsize; 12504 size_t offs; 12505 dof_optdesc_t *desc; 12506 12507 for (i = 0; i < dof->dofh_secnum; i++) { 12508 dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof + 12509 (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); 12510 12511 if (sec->dofs_type != DOF_SECT_OPTDESC) 12512 continue; 12513 12514 if (sec->dofs_align != sizeof (uint64_t)) { 12515 dtrace_dof_error(dof, "bad alignment in " 12516 "option description"); 12517 return (EINVAL); 12518 } 12519 12520 if ((entsize = sec->dofs_entsize) == 0) { 12521 dtrace_dof_error(dof, "zeroed option entry size"); 12522 return (EINVAL); 12523 } 12524 12525 if (entsize < sizeof (dof_optdesc_t)) { 12526 dtrace_dof_error(dof, "bad option entry size"); 12527 return (EINVAL); 12528 } 12529 12530 for (offs = 0; offs < sec->dofs_size; offs += entsize) { 12531 desc = (dof_optdesc_t *)((uintptr_t)dof + 12532 (uintptr_t)sec->dofs_offset + offs); 12533 12534 if (desc->dofo_strtab != DOF_SECIDX_NONE) { 12535 dtrace_dof_error(dof, "non-zero option string"); 12536 return (EINVAL); 12537 } 12538 12539 if (desc->dofo_value == DTRACEOPT_UNSET) { 12540 dtrace_dof_error(dof, "unset option"); 12541 return (EINVAL); 12542 } 12543 12544 if ((rval = dtrace_state_option(state, 12545 desc->dofo_option, desc->dofo_value)) != 0) { 12546 dtrace_dof_error(dof, "rejected option"); 12547 return (rval); 12548 } 12549 } 12550 } 12551 12552 return (0); 12553} 12554 12555/* 12556 * DTrace Consumer State Functions 12557 */ 12558static int 12559dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size) 12560{ 12561 size_t hashsize, maxper, min, chunksize = dstate->dtds_chunksize; 12562 void *base; 12563 uintptr_t limit; 12564 dtrace_dynvar_t *dvar, *next, *start; 12565 int i; 12566 12567 ASSERT(MUTEX_HELD(&dtrace_lock)); 12568 ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL); 12569 12570 bzero(dstate, sizeof (dtrace_dstate_t)); 12571 12572 if ((dstate->dtds_chunksize = chunksize) == 0) 12573 dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE; 12574 12575 if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t))) 12576 size = min; 12577 12578 if ((base = kmem_zalloc(size, KM_NOSLEEP)) == NULL) 12579 return (ENOMEM); 12580 12581 dstate->dtds_size = size; 12582 dstate->dtds_base = base; 12583 dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, KM_SLEEP); 12584 bzero(dstate->dtds_percpu, NCPU * sizeof (dtrace_dstate_percpu_t)); 12585 12586 hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)); 12587 12588 if (hashsize != 1 && (hashsize & 1)) 12589 hashsize--; 12590 12591 dstate->dtds_hashsize = hashsize; 12592 dstate->dtds_hash = dstate->dtds_base; 12593 12594 /* 12595 * Set all of our hash buckets to point to the single sink, and (if 12596 * it hasn't already been set), set the sink's hash value to be the 12597 * sink sentinel value. The sink is needed for dynamic variable 12598 * lookups to know that they have iterated over an entire, valid hash 12599 * chain. 12600 */ 12601 for (i = 0; i < hashsize; i++) 12602 dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink; 12603 12604 if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK) 12605 dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK; 12606 12607 /* 12608 * Determine number of active CPUs. Divide free list evenly among 12609 * active CPUs. 12610 */ 12611 start = (dtrace_dynvar_t *) 12612 ((uintptr_t)base + hashsize * sizeof (dtrace_dynhash_t)); 12613 limit = (uintptr_t)base + size; 12614 12615 maxper = (limit - (uintptr_t)start) / NCPU; 12616 maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize; 12617 12618 for (i = 0; i < NCPU; i++) { 12619#if !defined(sun) 12620 if (CPU_ABSENT(i)) 12621 continue; 12622#endif 12623 dstate->dtds_percpu[i].dtdsc_free = dvar = start; 12624 12625 /* 12626 * If we don't even have enough chunks to make it once through 12627 * NCPUs, we're just going to allocate everything to the first 12628 * CPU. And if we're on the last CPU, we're going to allocate 12629 * whatever is left over. In either case, we set the limit to 12630 * be the limit of the dynamic variable space. 12631 */ 12632 if (maxper == 0 || i == NCPU - 1) { 12633 limit = (uintptr_t)base + size; 12634 start = NULL; 12635 } else { 12636 limit = (uintptr_t)start + maxper; 12637 start = (dtrace_dynvar_t *)limit; 12638 } 12639 12640 ASSERT(limit <= (uintptr_t)base + size); 12641 12642 for (;;) { 12643 next = (dtrace_dynvar_t *)((uintptr_t)dvar + 12644 dstate->dtds_chunksize); 12645 12646 if ((uintptr_t)next + dstate->dtds_chunksize >= limit) 12647 break; 12648 12649 dvar->dtdv_next = next; 12650 dvar = next; 12651 } 12652 12653 if (maxper == 0) 12654 break; 12655 } 12656 12657 return (0); 12658} 12659 12660static void 12661dtrace_dstate_fini(dtrace_dstate_t *dstate) 12662{ 12663 ASSERT(MUTEX_HELD(&cpu_lock)); 12664 12665 if (dstate->dtds_base == NULL) 12666 return; 12667 12668 kmem_free(dstate->dtds_base, dstate->dtds_size); 12669 kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu); 12670} 12671 12672static void 12673dtrace_vstate_fini(dtrace_vstate_t *vstate) 12674{ 12675 /* 12676 * Logical XOR, where are you? 12677 */ 12678 ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL)); 12679 12680 if (vstate->dtvs_nglobals > 0) { 12681 kmem_free(vstate->dtvs_globals, vstate->dtvs_nglobals * 12682 sizeof (dtrace_statvar_t *)); 12683 } 12684 12685 if (vstate->dtvs_ntlocals > 0) { 12686 kmem_free(vstate->dtvs_tlocals, vstate->dtvs_ntlocals * 12687 sizeof (dtrace_difv_t)); 12688 } 12689 12690 ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL)); 12691 12692 if (vstate->dtvs_nlocals > 0) { 12693 kmem_free(vstate->dtvs_locals, vstate->dtvs_nlocals * 12694 sizeof (dtrace_statvar_t *)); 12695 } 12696} 12697 12698#if defined(sun) 12699static void 12700dtrace_state_clean(dtrace_state_t *state) 12701{ 12702 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) 12703 return; 12704 12705 dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars); 12706 dtrace_speculation_clean(state); 12707} 12708 12709static void 12710dtrace_state_deadman(dtrace_state_t *state) 12711{ 12712 hrtime_t now; 12713 12714 dtrace_sync(); 12715 12716 now = dtrace_gethrtime(); 12717 12718 if (state != dtrace_anon.dta_state && 12719 now - state->dts_laststatus >= dtrace_deadman_user) 12720 return; 12721 12722 /* 12723 * We must be sure that dts_alive never appears to be less than the 12724 * value upon entry to dtrace_state_deadman(), and because we lack a 12725 * dtrace_cas64(), we cannot store to it atomically. We thus instead 12726 * store INT64_MAX to it, followed by a memory barrier, followed by 12727 * the new value. This assures that dts_alive never appears to be 12728 * less than its true value, regardless of the order in which the 12729 * stores to the underlying storage are issued. 12730 */ 12731 state->dts_alive = INT64_MAX; 12732 dtrace_membar_producer(); 12733 state->dts_alive = now; 12734} 12735#else 12736static void 12737dtrace_state_clean(void *arg) 12738{ 12739 dtrace_state_t *state = arg; 12740 dtrace_optval_t *opt = state->dts_options; 12741 12742 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) 12743 return; 12744 12745 dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars); 12746 dtrace_speculation_clean(state); 12747 12748 callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC, 12749 dtrace_state_clean, state); 12750} 12751 12752static void 12753dtrace_state_deadman(void *arg) 12754{ 12755 dtrace_state_t *state = arg; 12756 hrtime_t now; 12757 12758 dtrace_sync(); 12759 12760 dtrace_debug_output(); 12761 12762 now = dtrace_gethrtime(); 12763 12764 if (state != dtrace_anon.dta_state && 12765 now - state->dts_laststatus >= dtrace_deadman_user) 12766 return; 12767 12768 /* 12769 * We must be sure that dts_alive never appears to be less than the 12770 * value upon entry to dtrace_state_deadman(), and because we lack a 12771 * dtrace_cas64(), we cannot store to it atomically. We thus instead 12772 * store INT64_MAX to it, followed by a memory barrier, followed by 12773 * the new value. This assures that dts_alive never appears to be 12774 * less than its true value, regardless of the order in which the 12775 * stores to the underlying storage are issued. 12776 */ 12777 state->dts_alive = INT64_MAX; 12778 dtrace_membar_producer(); 12779 state->dts_alive = now; 12780 12781 callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC, 12782 dtrace_state_deadman, state); 12783} 12784#endif 12785 12786static dtrace_state_t * 12787#if defined(sun) 12788dtrace_state_create(dev_t *devp, cred_t *cr) 12789#else 12790dtrace_state_create(struct cdev *dev) 12791#endif 12792{ 12793#if defined(sun) 12794 minor_t minor; 12795 major_t major; 12796#else 12797 cred_t *cr = NULL; 12798 int m = 0; 12799#endif 12800 char c[30]; 12801 dtrace_state_t *state; 12802 dtrace_optval_t *opt; 12803 int bufsize = NCPU * sizeof (dtrace_buffer_t), i; 12804 12805 ASSERT(MUTEX_HELD(&dtrace_lock)); 12806 ASSERT(MUTEX_HELD(&cpu_lock)); 12807 12808#if defined(sun) 12809 minor = (minor_t)(uintptr_t)vmem_alloc(dtrace_minor, 1, 12810 VM_BESTFIT | VM_SLEEP); 12811 12812 if (ddi_soft_state_zalloc(dtrace_softstate, minor) != DDI_SUCCESS) { 12813 vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1); 12814 return (NULL); 12815 } 12816 12817 state = ddi_get_soft_state(dtrace_softstate, minor); 12818#else 12819 if (dev != NULL) { 12820 cr = dev->si_cred; 12821 m = dev2unit(dev); 12822 } 12823 12824 /* Allocate memory for the state. */ 12825 state = kmem_zalloc(sizeof(dtrace_state_t), KM_SLEEP); 12826#endif 12827 12828 state->dts_epid = DTRACE_EPIDNONE + 1; 12829 12830 (void) snprintf(c, sizeof (c), "dtrace_aggid_%d", m); 12831#if defined(sun) 12832 state->dts_aggid_arena = vmem_create(c, (void *)1, UINT32_MAX, 1, 12833 NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); 12834 12835 if (devp != NULL) { 12836 major = getemajor(*devp); 12837 } else { 12838 major = ddi_driver_major(dtrace_devi); 12839 } 12840 12841 state->dts_dev = makedevice(major, minor); 12842 12843 if (devp != NULL) 12844 *devp = state->dts_dev; 12845#else 12846 state->dts_aggid_arena = new_unrhdr(1, INT_MAX, &dtrace_unr_mtx); 12847 state->dts_dev = dev; 12848#endif 12849 12850 /* 12851 * We allocate NCPU buffers. On the one hand, this can be quite 12852 * a bit of memory per instance (nearly 36K on a Starcat). On the 12853 * other hand, it saves an additional memory reference in the probe 12854 * path. 12855 */ 12856 state->dts_buffer = kmem_zalloc(bufsize, KM_SLEEP); 12857 state->dts_aggbuffer = kmem_zalloc(bufsize, KM_SLEEP); 12858 12859#if defined(sun) 12860 state->dts_cleaner = CYCLIC_NONE; 12861 state->dts_deadman = CYCLIC_NONE; 12862#else 12863 callout_init(&state->dts_cleaner, CALLOUT_MPSAFE); 12864 callout_init(&state->dts_deadman, CALLOUT_MPSAFE); 12865#endif 12866 state->dts_vstate.dtvs_state = state; 12867 12868 for (i = 0; i < DTRACEOPT_MAX; i++) 12869 state->dts_options[i] = DTRACEOPT_UNSET; 12870 12871 /* 12872 * Set the default options. 12873 */ 12874 opt = state->dts_options; 12875 opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH; 12876 opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO; 12877 opt[DTRACEOPT_NSPEC] = dtrace_nspec_default; 12878 opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default; 12879 opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL; 12880 opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default; 12881 opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default; 12882 opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default; 12883 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default; 12884 opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default; 12885 opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default; 12886 opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default; 12887 opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default; 12888 opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default; 12889 12890 state->dts_activity = DTRACE_ACTIVITY_INACTIVE; 12891 12892 /* 12893 * Depending on the user credentials, we set flag bits which alter probe 12894 * visibility or the amount of destructiveness allowed. In the case of 12895 * actual anonymous tracing, or the possession of all privileges, all of 12896 * the normal checks are bypassed. 12897 */ 12898 if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) { 12899 state->dts_cred.dcr_visible = DTRACE_CRV_ALL; 12900 state->dts_cred.dcr_action = DTRACE_CRA_ALL; 12901 } else { 12902 /* 12903 * Set up the credentials for this instantiation. We take a 12904 * hold on the credential to prevent it from disappearing on 12905 * us; this in turn prevents the zone_t referenced by this 12906 * credential from disappearing. This means that we can 12907 * examine the credential and the zone from probe context. 12908 */ 12909 crhold(cr); 12910 state->dts_cred.dcr_cred = cr; 12911 12912 /* 12913 * CRA_PROC means "we have *some* privilege for dtrace" and 12914 * unlocks the use of variables like pid, zonename, etc. 12915 */ 12916 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE) || 12917 PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) { 12918 state->dts_cred.dcr_action |= DTRACE_CRA_PROC; 12919 } 12920 12921 /* 12922 * dtrace_user allows use of syscall and profile providers. 12923 * If the user also has proc_owner and/or proc_zone, we 12924 * extend the scope to include additional visibility and 12925 * destructive power. 12926 */ 12927 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) { 12928 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) { 12929 state->dts_cred.dcr_visible |= 12930 DTRACE_CRV_ALLPROC; 12931 12932 state->dts_cred.dcr_action |= 12933 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; 12934 } 12935 12936 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) { 12937 state->dts_cred.dcr_visible |= 12938 DTRACE_CRV_ALLZONE; 12939 12940 state->dts_cred.dcr_action |= 12941 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; 12942 } 12943 12944 /* 12945 * If we have all privs in whatever zone this is, 12946 * we can do destructive things to processes which 12947 * have altered credentials. 12948 */ 12949#if defined(sun) 12950 if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE), 12951 cr->cr_zone->zone_privset)) { 12952 state->dts_cred.dcr_action |= 12953 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG; 12954 } 12955#endif 12956 } 12957 12958 /* 12959 * Holding the dtrace_kernel privilege also implies that 12960 * the user has the dtrace_user privilege from a visibility 12961 * perspective. But without further privileges, some 12962 * destructive actions are not available. 12963 */ 12964 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) { 12965 /* 12966 * Make all probes in all zones visible. However, 12967 * this doesn't mean that all actions become available 12968 * to all zones. 12969 */ 12970 state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL | 12971 DTRACE_CRV_ALLPROC | DTRACE_CRV_ALLZONE; 12972 12973 state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL | 12974 DTRACE_CRA_PROC; 12975 /* 12976 * Holding proc_owner means that destructive actions 12977 * for *this* zone are allowed. 12978 */ 12979 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) 12980 state->dts_cred.dcr_action |= 12981 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; 12982 12983 /* 12984 * Holding proc_zone means that destructive actions 12985 * for this user/group ID in all zones is allowed. 12986 */ 12987 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) 12988 state->dts_cred.dcr_action |= 12989 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; 12990 12991#if defined(sun) 12992 /* 12993 * If we have all privs in whatever zone this is, 12994 * we can do destructive things to processes which 12995 * have altered credentials. 12996 */ 12997 if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE), 12998 cr->cr_zone->zone_privset)) { 12999 state->dts_cred.dcr_action |= 13000 DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG; 13001 } 13002#endif 13003 } 13004 13005 /* 13006 * Holding the dtrace_proc privilege gives control over fasttrap 13007 * and pid providers. We need to grant wider destructive 13008 * privileges in the event that the user has proc_owner and/or 13009 * proc_zone. 13010 */ 13011 if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) { 13012 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) 13013 state->dts_cred.dcr_action |= 13014 DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; 13015 13016 if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) 13017 state->dts_cred.dcr_action |= 13018 DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; 13019 } 13020 } 13021 13022 return (state); 13023} 13024 13025static int 13026dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, int which) 13027{ 13028 dtrace_optval_t *opt = state->dts_options, size; 13029 processorid_t cpu = 0;; 13030 int flags = 0, rval; 13031 13032 ASSERT(MUTEX_HELD(&dtrace_lock)); 13033 ASSERT(MUTEX_HELD(&cpu_lock)); 13034 ASSERT(which < DTRACEOPT_MAX); 13035 ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE || 13036 (state == dtrace_anon.dta_state && 13037 state->dts_activity == DTRACE_ACTIVITY_ACTIVE)); 13038 13039 if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0) 13040 return (0); 13041 13042 if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET) 13043 cpu = opt[DTRACEOPT_CPU]; 13044 13045 if (which == DTRACEOPT_SPECSIZE) 13046 flags |= DTRACEBUF_NOSWITCH; 13047 13048 if (which == DTRACEOPT_BUFSIZE) { 13049 if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING) 13050 flags |= DTRACEBUF_RING; 13051 13052 if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL) 13053 flags |= DTRACEBUF_FILL; 13054 13055 if (state != dtrace_anon.dta_state || 13056 state->dts_activity != DTRACE_ACTIVITY_ACTIVE) 13057 flags |= DTRACEBUF_INACTIVE; 13058 } 13059 13060 for (size = opt[which]; size >= sizeof (uint64_t); size >>= 1) { 13061 /* 13062 * The size must be 8-byte aligned. If the size is not 8-byte 13063 * aligned, drop it down by the difference. 13064 */ 13065 if (size & (sizeof (uint64_t) - 1)) 13066 size -= size & (sizeof (uint64_t) - 1); 13067 13068 if (size < state->dts_reserve) { 13069 /* 13070 * Buffers always must be large enough to accommodate 13071 * their prereserved space. We return E2BIG instead 13072 * of ENOMEM in this case to allow for user-level 13073 * software to differentiate the cases. 13074 */ 13075 return (E2BIG); 13076 } 13077 13078 rval = dtrace_buffer_alloc(buf, size, flags, cpu); 13079 13080 if (rval != ENOMEM) { 13081 opt[which] = size; 13082 return (rval); 13083 } 13084 13085 if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) 13086 return (rval); 13087 } 13088 13089 return (ENOMEM); 13090} 13091 13092static int 13093dtrace_state_buffers(dtrace_state_t *state) 13094{ 13095 dtrace_speculation_t *spec = state->dts_speculations; 13096 int rval, i; 13097 13098 if ((rval = dtrace_state_buffer(state, state->dts_buffer, 13099 DTRACEOPT_BUFSIZE)) != 0) 13100 return (rval); 13101 13102 if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer, 13103 DTRACEOPT_AGGSIZE)) != 0) 13104 return (rval); 13105 13106 for (i = 0; i < state->dts_nspeculations; i++) { 13107 if ((rval = dtrace_state_buffer(state, 13108 spec[i].dtsp_buffer, DTRACEOPT_SPECSIZE)) != 0) 13109 return (rval); 13110 } 13111 13112 return (0); 13113} 13114 13115static void 13116dtrace_state_prereserve(dtrace_state_t *state) 13117{ 13118 dtrace_ecb_t *ecb; 13119 dtrace_probe_t *probe; 13120 13121 state->dts_reserve = 0; 13122 13123 if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL) 13124 return; 13125 13126 /* 13127 * If our buffer policy is a "fill" buffer policy, we need to set the 13128 * prereserved space to be the space required by the END probes. 13129 */ 13130 probe = dtrace_probes[dtrace_probeid_end - 1]; 13131 ASSERT(probe != NULL); 13132 13133 for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { 13134 if (ecb->dte_state != state) 13135 continue; 13136 13137 state->dts_reserve += ecb->dte_needed + ecb->dte_alignment; 13138 } 13139} 13140 13141static int 13142dtrace_state_go(dtrace_state_t *state, processorid_t *cpu) 13143{ 13144 dtrace_optval_t *opt = state->dts_options, sz, nspec; 13145 dtrace_speculation_t *spec; 13146 dtrace_buffer_t *buf; 13147#if defined(sun) 13148 cyc_handler_t hdlr; 13149 cyc_time_t when; 13150#endif 13151 int rval = 0, i, bufsize = NCPU * sizeof (dtrace_buffer_t); 13152 dtrace_icookie_t cookie; 13153 13154 mutex_enter(&cpu_lock); 13155 mutex_enter(&dtrace_lock); 13156 13157 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { 13158 rval = EBUSY; 13159 goto out; 13160 } 13161 13162 /* 13163 * Before we can perform any checks, we must prime all of the 13164 * retained enablings that correspond to this state. 13165 */ 13166 dtrace_enabling_prime(state); 13167 13168 if (state->dts_destructive && !state->dts_cred.dcr_destructive) { 13169 rval = EACCES; 13170 goto out; 13171 } 13172 13173 dtrace_state_prereserve(state); 13174 13175 /* 13176 * Now we want to do is try to allocate our speculations. 13177 * We do not automatically resize the number of speculations; if 13178 * this fails, we will fail the operation. 13179 */ 13180 nspec = opt[DTRACEOPT_NSPEC]; 13181 ASSERT(nspec != DTRACEOPT_UNSET); 13182 13183 if (nspec > INT_MAX) { 13184 rval = ENOMEM; 13185 goto out; 13186 } 13187 13188 spec = kmem_zalloc(nspec * sizeof (dtrace_speculation_t), KM_NOSLEEP); 13189 13190 if (spec == NULL) { 13191 rval = ENOMEM; 13192 goto out; 13193 } 13194 13195 state->dts_speculations = spec; 13196 state->dts_nspeculations = (int)nspec; 13197 13198 for (i = 0; i < nspec; i++) { 13199 if ((buf = kmem_zalloc(bufsize, KM_NOSLEEP)) == NULL) { 13200 rval = ENOMEM; 13201 goto err; 13202 } 13203 13204 spec[i].dtsp_buffer = buf; 13205 } 13206 13207 if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) { 13208 if (dtrace_anon.dta_state == NULL) { 13209 rval = ENOENT; 13210 goto out; 13211 } 13212 13213 if (state->dts_necbs != 0) { 13214 rval = EALREADY; 13215 goto out; 13216 } 13217 13218 state->dts_anon = dtrace_anon_grab(); 13219 ASSERT(state->dts_anon != NULL); 13220 state = state->dts_anon; 13221 13222 /* 13223 * We want "grabanon" to be set in the grabbed state, so we'll 13224 * copy that option value from the grabbing state into the 13225 * grabbed state. 13226 */ 13227 state->dts_options[DTRACEOPT_GRABANON] = 13228 opt[DTRACEOPT_GRABANON]; 13229 13230 *cpu = dtrace_anon.dta_beganon; 13231 13232 /* 13233 * If the anonymous state is active (as it almost certainly 13234 * is if the anonymous enabling ultimately matched anything), 13235 * we don't allow any further option processing -- but we 13236 * don't return failure. 13237 */ 13238 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) 13239 goto out; 13240 } 13241 13242 if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET && 13243 opt[DTRACEOPT_AGGSIZE] != 0) { 13244 if (state->dts_aggregations == NULL) { 13245 /* 13246 * We're not going to create an aggregation buffer 13247 * because we don't have any ECBs that contain 13248 * aggregations -- set this option to 0. 13249 */ 13250 opt[DTRACEOPT_AGGSIZE] = 0; 13251 } else { 13252 /* 13253 * If we have an aggregation buffer, we must also have 13254 * a buffer to use as scratch. 13255 */ 13256 if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET || 13257 opt[DTRACEOPT_BUFSIZE] < state->dts_needed) { 13258 opt[DTRACEOPT_BUFSIZE] = state->dts_needed; 13259 } 13260 } 13261 } 13262 13263 if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET && 13264 opt[DTRACEOPT_SPECSIZE] != 0) { 13265 if (!state->dts_speculates) { 13266 /* 13267 * We're not going to create speculation buffers 13268 * because we don't have any ECBs that actually 13269 * speculate -- set the speculation size to 0. 13270 */ 13271 opt[DTRACEOPT_SPECSIZE] = 0; 13272 } 13273 } 13274 13275 /* 13276 * The bare minimum size for any buffer that we're actually going to 13277 * do anything to is sizeof (uint64_t). 13278 */ 13279 sz = sizeof (uint64_t); 13280 13281 if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) || 13282 (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) || 13283 (state->dts_aggregations != NULL && opt[DTRACEOPT_AGGSIZE] < sz)) { 13284 /* 13285 * A buffer size has been explicitly set to 0 (or to a size 13286 * that will be adjusted to 0) and we need the space -- we 13287 * need to return failure. We return ENOSPC to differentiate 13288 * it from failing to allocate a buffer due to failure to meet 13289 * the reserve (for which we return E2BIG). 13290 */ 13291 rval = ENOSPC; 13292 goto out; 13293 } 13294 13295 if ((rval = dtrace_state_buffers(state)) != 0) 13296 goto err; 13297 13298 if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET) 13299 sz = dtrace_dstate_defsize; 13300 13301 do { 13302 rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz); 13303 13304 if (rval == 0) 13305 break; 13306 13307 if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) 13308 goto err; 13309 } while (sz >>= 1); 13310 13311 opt[DTRACEOPT_DYNVARSIZE] = sz; 13312 13313 if (rval != 0) 13314 goto err; 13315 13316 if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max) 13317 opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max; 13318 13319 if (opt[DTRACEOPT_CLEANRATE] == 0) 13320 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; 13321 13322 if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min) 13323 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min; 13324 13325 if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max) 13326 opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; 13327 13328 state->dts_alive = state->dts_laststatus = dtrace_gethrtime(); 13329#if defined(sun) 13330 hdlr.cyh_func = (cyc_func_t)dtrace_state_clean; 13331 hdlr.cyh_arg = state; 13332 hdlr.cyh_level = CY_LOW_LEVEL; 13333 13334 when.cyt_when = 0; 13335 when.cyt_interval = opt[DTRACEOPT_CLEANRATE]; 13336 13337 state->dts_cleaner = cyclic_add(&hdlr, &when); 13338 13339 hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman; 13340 hdlr.cyh_arg = state; 13341 hdlr.cyh_level = CY_LOW_LEVEL; 13342 13343 when.cyt_when = 0; 13344 when.cyt_interval = dtrace_deadman_interval; 13345 13346 state->dts_deadman = cyclic_add(&hdlr, &when); 13347#else 13348 callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC, 13349 dtrace_state_clean, state); 13350 callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC, 13351 dtrace_state_deadman, state); 13352#endif 13353 13354 state->dts_activity = DTRACE_ACTIVITY_WARMUP; 13355 13356 /* 13357 * Now it's time to actually fire the BEGIN probe. We need to disable 13358 * interrupts here both to record the CPU on which we fired the BEGIN 13359 * probe (the data from this CPU will be processed first at user 13360 * level) and to manually activate the buffer for this CPU. 13361 */ 13362 cookie = dtrace_interrupt_disable(); 13363 *cpu = curcpu; 13364 ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE); 13365 state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE; 13366 13367 dtrace_probe(dtrace_probeid_begin, 13368 (uint64_t)(uintptr_t)state, 0, 0, 0, 0); 13369 dtrace_interrupt_enable(cookie); 13370 /* 13371 * We may have had an exit action from a BEGIN probe; only change our 13372 * state to ACTIVE if we're still in WARMUP. 13373 */ 13374 ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP || 13375 state->dts_activity == DTRACE_ACTIVITY_DRAINING); 13376 13377 if (state->dts_activity == DTRACE_ACTIVITY_WARMUP) 13378 state->dts_activity = DTRACE_ACTIVITY_ACTIVE; 13379 13380 /* 13381 * Regardless of whether or not now we're in ACTIVE or DRAINING, we 13382 * want each CPU to transition its principal buffer out of the 13383 * INACTIVE state. Doing this assures that no CPU will suddenly begin 13384 * processing an ECB halfway down a probe's ECB chain; all CPUs will 13385 * atomically transition from processing none of a state's ECBs to 13386 * processing all of them. 13387 */ 13388 dtrace_xcall(DTRACE_CPUALL, 13389 (dtrace_xcall_t)dtrace_buffer_activate, state); 13390 goto out; 13391 13392err: 13393 dtrace_buffer_free(state->dts_buffer); 13394 dtrace_buffer_free(state->dts_aggbuffer); 13395 13396 if ((nspec = state->dts_nspeculations) == 0) { 13397 ASSERT(state->dts_speculations == NULL); 13398 goto out; 13399 } 13400 13401 spec = state->dts_speculations; 13402 ASSERT(spec != NULL); 13403 13404 for (i = 0; i < state->dts_nspeculations; i++) { 13405 if ((buf = spec[i].dtsp_buffer) == NULL) 13406 break; 13407 13408 dtrace_buffer_free(buf); 13409 kmem_free(buf, bufsize); 13410 } 13411 13412 kmem_free(spec, nspec * sizeof (dtrace_speculation_t)); 13413 state->dts_nspeculations = 0; 13414 state->dts_speculations = NULL; 13415 13416out: 13417 mutex_exit(&dtrace_lock); 13418 mutex_exit(&cpu_lock); 13419 13420 return (rval); 13421} 13422 13423static int 13424dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu) 13425{ 13426 dtrace_icookie_t cookie; 13427 13428 ASSERT(MUTEX_HELD(&dtrace_lock)); 13429 13430 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE && 13431 state->dts_activity != DTRACE_ACTIVITY_DRAINING) 13432 return (EINVAL); 13433 13434 /* 13435 * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync 13436 * to be sure that every CPU has seen it. See below for the details 13437 * on why this is done. 13438 */ 13439 state->dts_activity = DTRACE_ACTIVITY_DRAINING; 13440 dtrace_sync(); 13441 13442 /* 13443 * By this point, it is impossible for any CPU to be still processing 13444 * with DTRACE_ACTIVITY_ACTIVE. We can thus set our activity to 13445 * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any 13446 * other CPU in dtrace_buffer_reserve(). This allows dtrace_probe() 13447 * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN 13448 * iff we're in the END probe. 13449 */ 13450 state->dts_activity = DTRACE_ACTIVITY_COOLDOWN; 13451 dtrace_sync(); 13452 ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN); 13453 13454 /* 13455 * Finally, we can release the reserve and call the END probe. We 13456 * disable interrupts across calling the END probe to allow us to 13457 * return the CPU on which we actually called the END probe. This 13458 * allows user-land to be sure that this CPU's principal buffer is 13459 * processed last. 13460 */ 13461 state->dts_reserve = 0; 13462 13463 cookie = dtrace_interrupt_disable(); 13464 *cpu = curcpu; 13465 dtrace_probe(dtrace_probeid_end, 13466 (uint64_t)(uintptr_t)state, 0, 0, 0, 0); 13467 dtrace_interrupt_enable(cookie); 13468 13469 state->dts_activity = DTRACE_ACTIVITY_STOPPED; 13470 dtrace_sync(); 13471 13472 return (0); 13473} 13474 13475static int 13476dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option, 13477 dtrace_optval_t val) 13478{ 13479 ASSERT(MUTEX_HELD(&dtrace_lock)); 13480 13481 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) 13482 return (EBUSY); 13483 13484 if (option >= DTRACEOPT_MAX) 13485 return (EINVAL); 13486 13487 if (option != DTRACEOPT_CPU && val < 0) 13488 return (EINVAL); 13489 13490 switch (option) { 13491 case DTRACEOPT_DESTRUCTIVE: 13492 if (dtrace_destructive_disallow) 13493 return (EACCES); 13494 13495 state->dts_cred.dcr_destructive = 1; 13496 break; 13497 13498 case DTRACEOPT_BUFSIZE: 13499 case DTRACEOPT_DYNVARSIZE: 13500 case DTRACEOPT_AGGSIZE: 13501 case DTRACEOPT_SPECSIZE: 13502 case DTRACEOPT_STRSIZE: 13503 if (val < 0) 13504 return (EINVAL); 13505 13506 if (val >= LONG_MAX) { 13507 /* 13508 * If this is an otherwise negative value, set it to 13509 * the highest multiple of 128m less than LONG_MAX. 13510 * Technically, we're adjusting the size without 13511 * regard to the buffer resizing policy, but in fact, 13512 * this has no effect -- if we set the buffer size to 13513 * ~LONG_MAX and the buffer policy is ultimately set to 13514 * be "manual", the buffer allocation is guaranteed to 13515 * fail, if only because the allocation requires two 13516 * buffers. (We set the the size to the highest 13517 * multiple of 128m because it ensures that the size 13518 * will remain a multiple of a megabyte when 13519 * repeatedly halved -- all the way down to 15m.) 13520 */ 13521 val = LONG_MAX - (1 << 27) + 1; 13522 } 13523 } 13524 13525 state->dts_options[option] = val; 13526 13527 return (0); 13528} 13529 13530static void 13531dtrace_state_destroy(dtrace_state_t *state) 13532{ 13533 dtrace_ecb_t *ecb; 13534 dtrace_vstate_t *vstate = &state->dts_vstate; 13535#if defined(sun) 13536 minor_t minor = getminor(state->dts_dev); 13537#endif 13538 int i, bufsize = NCPU * sizeof (dtrace_buffer_t); 13539 dtrace_speculation_t *spec = state->dts_speculations; 13540 int nspec = state->dts_nspeculations; 13541 uint32_t match; 13542 13543 ASSERT(MUTEX_HELD(&dtrace_lock)); 13544 ASSERT(MUTEX_HELD(&cpu_lock)); 13545 13546 /* 13547 * First, retract any retained enablings for this state. 13548 */ 13549 dtrace_enabling_retract(state); 13550 ASSERT(state->dts_nretained == 0); 13551 13552 if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE || 13553 state->dts_activity == DTRACE_ACTIVITY_DRAINING) { 13554 /* 13555 * We have managed to come into dtrace_state_destroy() on a 13556 * hot enabling -- almost certainly because of a disorderly 13557 * shutdown of a consumer. (That is, a consumer that is 13558 * exiting without having called dtrace_stop().) In this case, 13559 * we're going to set our activity to be KILLED, and then 13560 * issue a sync to be sure that everyone is out of probe 13561 * context before we start blowing away ECBs. 13562 */ 13563 state->dts_activity = DTRACE_ACTIVITY_KILLED; 13564 dtrace_sync(); 13565 } 13566 13567 /* 13568 * Release the credential hold we took in dtrace_state_create(). 13569 */ 13570 if (state->dts_cred.dcr_cred != NULL) 13571 crfree(state->dts_cred.dcr_cred); 13572 13573 /* 13574 * Now we can safely disable and destroy any enabled probes. Because 13575 * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress 13576 * (especially if they're all enabled), we take two passes through the 13577 * ECBs: in the first, we disable just DTRACE_PRIV_KERNEL probes, and 13578 * in the second we disable whatever is left over. 13579 */ 13580 for (match = DTRACE_PRIV_KERNEL; ; match = 0) { 13581 for (i = 0; i < state->dts_necbs; i++) { 13582 if ((ecb = state->dts_ecbs[i]) == NULL) 13583 continue; 13584 13585 if (match && ecb->dte_probe != NULL) { 13586 dtrace_probe_t *probe = ecb->dte_probe; 13587 dtrace_provider_t *prov = probe->dtpr_provider; 13588 13589 if (!(prov->dtpv_priv.dtpp_flags & match)) 13590 continue; 13591 } 13592 13593 dtrace_ecb_disable(ecb); 13594 dtrace_ecb_destroy(ecb); 13595 } 13596 13597 if (!match) 13598 break; 13599 } 13600 13601 /* 13602 * Before we free the buffers, perform one more sync to assure that 13603 * every CPU is out of probe context. 13604 */ 13605 dtrace_sync(); 13606 13607 dtrace_buffer_free(state->dts_buffer); 13608 dtrace_buffer_free(state->dts_aggbuffer); 13609 13610 for (i = 0; i < nspec; i++) 13611 dtrace_buffer_free(spec[i].dtsp_buffer); 13612 13613#if defined(sun) 13614 if (state->dts_cleaner != CYCLIC_NONE) 13615 cyclic_remove(state->dts_cleaner); 13616 13617 if (state->dts_deadman != CYCLIC_NONE) 13618 cyclic_remove(state->dts_deadman); 13619#else 13620 callout_stop(&state->dts_cleaner); 13621 callout_drain(&state->dts_cleaner); 13622 callout_stop(&state->dts_deadman); 13623 callout_drain(&state->dts_deadman); 13624#endif 13625 13626 dtrace_dstate_fini(&vstate->dtvs_dynvars); 13627 dtrace_vstate_fini(vstate); 13628 if (state->dts_ecbs != NULL) 13629 kmem_free(state->dts_ecbs, state->dts_necbs * sizeof (dtrace_ecb_t *)); 13630 13631 if (state->dts_aggregations != NULL) { 13632#ifdef DEBUG 13633 for (i = 0; i < state->dts_naggregations; i++) 13634 ASSERT(state->dts_aggregations[i] == NULL); 13635#endif 13636 ASSERT(state->dts_naggregations > 0); 13637 kmem_free(state->dts_aggregations, 13638 state->dts_naggregations * sizeof (dtrace_aggregation_t *)); 13639 } 13640 13641 kmem_free(state->dts_buffer, bufsize); 13642 kmem_free(state->dts_aggbuffer, bufsize); 13643 13644 for (i = 0; i < nspec; i++) 13645 kmem_free(spec[i].dtsp_buffer, bufsize); 13646 13647 if (spec != NULL) 13648 kmem_free(spec, nspec * sizeof (dtrace_speculation_t)); 13649 13650 dtrace_format_destroy(state); 13651 13652 if (state->dts_aggid_arena != NULL) { 13653#if defined(sun) 13654 vmem_destroy(state->dts_aggid_arena); 13655#else 13656 delete_unrhdr(state->dts_aggid_arena); 13657#endif 13658 state->dts_aggid_arena = NULL; 13659 } 13660#if defined(sun) 13661 ddi_soft_state_free(dtrace_softstate, minor); 13662 vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1); 13663#endif 13664} 13665 13666/* 13667 * DTrace Anonymous Enabling Functions 13668 */ 13669static dtrace_state_t * 13670dtrace_anon_grab(void) 13671{ 13672 dtrace_state_t *state; 13673 13674 ASSERT(MUTEX_HELD(&dtrace_lock)); 13675 13676 if ((state = dtrace_anon.dta_state) == NULL) { 13677 ASSERT(dtrace_anon.dta_enabling == NULL); 13678 return (NULL); 13679 } 13680 13681 ASSERT(dtrace_anon.dta_enabling != NULL); 13682 ASSERT(dtrace_retained != NULL); 13683 13684 dtrace_enabling_destroy(dtrace_anon.dta_enabling); 13685 dtrace_anon.dta_enabling = NULL; 13686 dtrace_anon.dta_state = NULL; 13687 13688 return (state); 13689} 13690 13691static void 13692dtrace_anon_property(void) 13693{ 13694 int i, rv; 13695 dtrace_state_t *state; 13696 dof_hdr_t *dof; 13697 char c[32]; /* enough for "dof-data-" + digits */ 13698 13699 ASSERT(MUTEX_HELD(&dtrace_lock)); 13700 ASSERT(MUTEX_HELD(&cpu_lock)); 13701 13702 for (i = 0; ; i++) { 13703 (void) snprintf(c, sizeof (c), "dof-data-%d", i); 13704 13705 dtrace_err_verbose = 1; 13706 13707 if ((dof = dtrace_dof_property(c)) == NULL) { 13708 dtrace_err_verbose = 0; 13709 break; 13710 } 13711 13712#if defined(sun) 13713 /* 13714 * We want to create anonymous state, so we need to transition 13715 * the kernel debugger to indicate that DTrace is active. If 13716 * this fails (e.g. because the debugger has modified text in 13717 * some way), we won't continue with the processing. 13718 */ 13719 if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { 13720 cmn_err(CE_NOTE, "kernel debugger active; anonymous " 13721 "enabling ignored."); 13722 dtrace_dof_destroy(dof); 13723 break; 13724 } 13725#endif 13726 13727 /* 13728 * If we haven't allocated an anonymous state, we'll do so now. 13729 */ 13730 if ((state = dtrace_anon.dta_state) == NULL) { 13731#if defined(sun) 13732 state = dtrace_state_create(NULL, NULL); 13733#else 13734 state = dtrace_state_create(NULL); 13735#endif 13736 dtrace_anon.dta_state = state; 13737 13738 if (state == NULL) { 13739 /* 13740 * This basically shouldn't happen: the only 13741 * failure mode from dtrace_state_create() is a 13742 * failure of ddi_soft_state_zalloc() that 13743 * itself should never happen. Still, the 13744 * interface allows for a failure mode, and 13745 * we want to fail as gracefully as possible: 13746 * we'll emit an error message and cease 13747 * processing anonymous state in this case. 13748 */ 13749 cmn_err(CE_WARN, "failed to create " 13750 "anonymous state"); 13751 dtrace_dof_destroy(dof); 13752 break; 13753 } 13754 } 13755 13756 rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(), 13757 &dtrace_anon.dta_enabling, 0, B_TRUE); 13758 13759 if (rv == 0) 13760 rv = dtrace_dof_options(dof, state); 13761 13762 dtrace_err_verbose = 0; 13763 dtrace_dof_destroy(dof); 13764 13765 if (rv != 0) { 13766 /* 13767 * This is malformed DOF; chuck any anonymous state 13768 * that we created. 13769 */ 13770 ASSERT(dtrace_anon.dta_enabling == NULL); 13771 dtrace_state_destroy(state); 13772 dtrace_anon.dta_state = NULL; 13773 break; 13774 } 13775 13776 ASSERT(dtrace_anon.dta_enabling != NULL); 13777 } 13778 13779 if (dtrace_anon.dta_enabling != NULL) { 13780 int rval; 13781 13782 /* 13783 * dtrace_enabling_retain() can only fail because we are 13784 * trying to retain more enablings than are allowed -- but 13785 * we only have one anonymous enabling, and we are guaranteed 13786 * to be allowed at least one retained enabling; we assert 13787 * that dtrace_enabling_retain() returns success. 13788 */ 13789 rval = dtrace_enabling_retain(dtrace_anon.dta_enabling); 13790 ASSERT(rval == 0); 13791 13792 dtrace_enabling_dump(dtrace_anon.dta_enabling); 13793 } 13794} 13795 13796#if defined(sun) 13797/* 13798 * DTrace Helper Functions 13799 */ 13800static void 13801dtrace_helper_trace(dtrace_helper_action_t *helper, 13802 dtrace_mstate_t *mstate, dtrace_vstate_t *vstate, int where) 13803{ 13804 uint32_t size, next, nnext, i; 13805 dtrace_helptrace_t *ent; 13806 uint16_t flags = cpu_core[curcpu].cpuc_dtrace_flags; 13807 13808 if (!dtrace_helptrace_enabled) 13809 return; 13810 13811 ASSERT(vstate->dtvs_nlocals <= dtrace_helptrace_nlocals); 13812 13813 /* 13814 * What would a tracing framework be without its own tracing 13815 * framework? (Well, a hell of a lot simpler, for starters...) 13816 */ 13817 size = sizeof (dtrace_helptrace_t) + dtrace_helptrace_nlocals * 13818 sizeof (uint64_t) - sizeof (uint64_t); 13819 13820 /* 13821 * Iterate until we can allocate a slot in the trace buffer. 13822 */ 13823 do { 13824 next = dtrace_helptrace_next; 13825 13826 if (next + size < dtrace_helptrace_bufsize) { 13827 nnext = next + size; 13828 } else { 13829 nnext = size; 13830 } 13831 } while (dtrace_cas32(&dtrace_helptrace_next, next, nnext) != next); 13832 13833 /* 13834 * We have our slot; fill it in. 13835 */ 13836 if (nnext == size) 13837 next = 0; 13838 13839 ent = (dtrace_helptrace_t *)&dtrace_helptrace_buffer[next]; 13840 ent->dtht_helper = helper; 13841 ent->dtht_where = where; 13842 ent->dtht_nlocals = vstate->dtvs_nlocals; 13843 13844 ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) ? 13845 mstate->dtms_fltoffs : -1; 13846 ent->dtht_fault = DTRACE_FLAGS2FLT(flags); 13847 ent->dtht_illval = cpu_core[curcpu].cpuc_dtrace_illval; 13848 13849 for (i = 0; i < vstate->dtvs_nlocals; i++) { 13850 dtrace_statvar_t *svar; 13851 13852 if ((svar = vstate->dtvs_locals[i]) == NULL) 13853 continue; 13854 13855 ASSERT(svar->dtsv_size >= NCPU * sizeof (uint64_t)); 13856 ent->dtht_locals[i] = 13857 ((uint64_t *)(uintptr_t)svar->dtsv_data)[curcpu]; 13858 } 13859} 13860#endif 13861 13862#if defined(sun) 13863static uint64_t 13864dtrace_helper(int which, dtrace_mstate_t *mstate, 13865 dtrace_state_t *state, uint64_t arg0, uint64_t arg1) 13866{ 13867 uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; 13868 uint64_t sarg0 = mstate->dtms_arg[0]; 13869 uint64_t sarg1 = mstate->dtms_arg[1]; 13870 uint64_t rval; 13871 dtrace_helpers_t *helpers = curproc->p_dtrace_helpers; 13872 dtrace_helper_action_t *helper; 13873 dtrace_vstate_t *vstate; 13874 dtrace_difo_t *pred; 13875 int i, trace = dtrace_helptrace_enabled; 13876 13877 ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS); 13878 13879 if (helpers == NULL) 13880 return (0); 13881 13882 if ((helper = helpers->dthps_actions[which]) == NULL) 13883 return (0); 13884 13885 vstate = &helpers->dthps_vstate; 13886 mstate->dtms_arg[0] = arg0; 13887 mstate->dtms_arg[1] = arg1; 13888 13889 /* 13890 * Now iterate over each helper. If its predicate evaluates to 'true', 13891 * we'll call the corresponding actions. Note that the below calls 13892 * to dtrace_dif_emulate() may set faults in machine state. This is 13893 * okay: our caller (the outer dtrace_dif_emulate()) will simply plow 13894 * the stored DIF offset with its own (which is the desired behavior). 13895 * Also, note the calls to dtrace_dif_emulate() may allocate scratch 13896 * from machine state; this is okay, too. 13897 */ 13898 for (; helper != NULL; helper = helper->dtha_next) { 13899 if ((pred = helper->dtha_predicate) != NULL) { 13900 if (trace) 13901 dtrace_helper_trace(helper, mstate, vstate, 0); 13902 13903 if (!dtrace_dif_emulate(pred, mstate, vstate, state)) 13904 goto next; 13905 13906 if (*flags & CPU_DTRACE_FAULT) 13907 goto err; 13908 } 13909 13910 for (i = 0; i < helper->dtha_nactions; i++) { 13911 if (trace) 13912 dtrace_helper_trace(helper, 13913 mstate, vstate, i + 1); 13914 13915 rval = dtrace_dif_emulate(helper->dtha_actions[i], 13916 mstate, vstate, state); 13917 13918 if (*flags & CPU_DTRACE_FAULT) 13919 goto err; 13920 } 13921 13922next: 13923 if (trace) 13924 dtrace_helper_trace(helper, mstate, vstate, 13925 DTRACE_HELPTRACE_NEXT); 13926 } 13927 13928 if (trace) 13929 dtrace_helper_trace(helper, mstate, vstate, 13930 DTRACE_HELPTRACE_DONE); 13931 13932 /* 13933 * Restore the arg0 that we saved upon entry. 13934 */ 13935 mstate->dtms_arg[0] = sarg0; 13936 mstate->dtms_arg[1] = sarg1; 13937 13938 return (rval); 13939 13940err: 13941 if (trace) 13942 dtrace_helper_trace(helper, mstate, vstate, 13943 DTRACE_HELPTRACE_ERR); 13944 13945 /* 13946 * Restore the arg0 that we saved upon entry. 13947 */ 13948 mstate->dtms_arg[0] = sarg0; 13949 mstate->dtms_arg[1] = sarg1; 13950 13951 return (0); 13952} 13953 13954static void 13955dtrace_helper_action_destroy(dtrace_helper_action_t *helper, 13956 dtrace_vstate_t *vstate) 13957{ 13958 int i; 13959 13960 if (helper->dtha_predicate != NULL) 13961 dtrace_difo_release(helper->dtha_predicate, vstate); 13962 13963 for (i = 0; i < helper->dtha_nactions; i++) { 13964 ASSERT(helper->dtha_actions[i] != NULL); 13965 dtrace_difo_release(helper->dtha_actions[i], vstate); 13966 } 13967 13968 kmem_free(helper->dtha_actions, 13969 helper->dtha_nactions * sizeof (dtrace_difo_t *)); 13970 kmem_free(helper, sizeof (dtrace_helper_action_t)); 13971} 13972 13973static int 13974dtrace_helper_destroygen(int gen) 13975{ 13976 proc_t *p = curproc; 13977 dtrace_helpers_t *help = p->p_dtrace_helpers; 13978 dtrace_vstate_t *vstate; 13979 int i; 13980 13981 ASSERT(MUTEX_HELD(&dtrace_lock)); 13982 13983 if (help == NULL || gen > help->dthps_generation) 13984 return (EINVAL); 13985 13986 vstate = &help->dthps_vstate; 13987 13988 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { 13989 dtrace_helper_action_t *last = NULL, *h, *next; 13990 13991 for (h = help->dthps_actions[i]; h != NULL; h = next) { 13992 next = h->dtha_next; 13993 13994 if (h->dtha_generation == gen) { 13995 if (last != NULL) { 13996 last->dtha_next = next; 13997 } else { 13998 help->dthps_actions[i] = next; 13999 } 14000 14001 dtrace_helper_action_destroy(h, vstate); 14002 } else { 14003 last = h; 14004 } 14005 } 14006 } 14007 14008 /* 14009 * Interate until we've cleared out all helper providers with the 14010 * given generation number. 14011 */ 14012 for (;;) { 14013 dtrace_helper_provider_t *prov; 14014 14015 /* 14016 * Look for a helper provider with the right generation. We 14017 * have to start back at the beginning of the list each time 14018 * because we drop dtrace_lock. It's unlikely that we'll make 14019 * more than two passes. 14020 */ 14021 for (i = 0; i < help->dthps_nprovs; i++) { 14022 prov = help->dthps_provs[i]; 14023 14024 if (prov->dthp_generation == gen) 14025 break; 14026 } 14027 14028 /* 14029 * If there were no matches, we're done. 14030 */ 14031 if (i == help->dthps_nprovs) 14032 break; 14033 14034 /* 14035 * Move the last helper provider into this slot. 14036 */ 14037 help->dthps_nprovs--; 14038 help->dthps_provs[i] = help->dthps_provs[help->dthps_nprovs]; 14039 help->dthps_provs[help->dthps_nprovs] = NULL; 14040 14041 mutex_exit(&dtrace_lock); 14042 14043 /* 14044 * If we have a meta provider, remove this helper provider. 14045 */ 14046 mutex_enter(&dtrace_meta_lock); 14047 if (dtrace_meta_pid != NULL) { 14048 ASSERT(dtrace_deferred_pid == NULL); 14049 dtrace_helper_provider_remove(&prov->dthp_prov, 14050 p->p_pid); 14051 } 14052 mutex_exit(&dtrace_meta_lock); 14053 14054 dtrace_helper_provider_destroy(prov); 14055 14056 mutex_enter(&dtrace_lock); 14057 } 14058 14059 return (0); 14060} 14061#endif 14062 14063#if defined(sun) 14064static int 14065dtrace_helper_validate(dtrace_helper_action_t *helper) 14066{ 14067 int err = 0, i; 14068 dtrace_difo_t *dp; 14069 14070 if ((dp = helper->dtha_predicate) != NULL) 14071 err += dtrace_difo_validate_helper(dp); 14072 14073 for (i = 0; i < helper->dtha_nactions; i++) 14074 err += dtrace_difo_validate_helper(helper->dtha_actions[i]); 14075 14076 return (err == 0); 14077} 14078#endif 14079 14080#if defined(sun) 14081static int 14082dtrace_helper_action_add(int which, dtrace_ecbdesc_t *ep) 14083{ 14084 dtrace_helpers_t *help; 14085 dtrace_helper_action_t *helper, *last; 14086 dtrace_actdesc_t *act; 14087 dtrace_vstate_t *vstate; 14088 dtrace_predicate_t *pred; 14089 int count = 0, nactions = 0, i; 14090 14091 if (which < 0 || which >= DTRACE_NHELPER_ACTIONS) 14092 return (EINVAL); 14093 14094 help = curproc->p_dtrace_helpers; 14095 last = help->dthps_actions[which]; 14096 vstate = &help->dthps_vstate; 14097 14098 for (count = 0; last != NULL; last = last->dtha_next) { 14099 count++; 14100 if (last->dtha_next == NULL) 14101 break; 14102 } 14103 14104 /* 14105 * If we already have dtrace_helper_actions_max helper actions for this 14106 * helper action type, we'll refuse to add a new one. 14107 */ 14108 if (count >= dtrace_helper_actions_max) 14109 return (ENOSPC); 14110 14111 helper = kmem_zalloc(sizeof (dtrace_helper_action_t), KM_SLEEP); 14112 helper->dtha_generation = help->dthps_generation; 14113 14114 if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) { 14115 ASSERT(pred->dtp_difo != NULL); 14116 dtrace_difo_hold(pred->dtp_difo); 14117 helper->dtha_predicate = pred->dtp_difo; 14118 } 14119 14120 for (act = ep->dted_action; act != NULL; act = act->dtad_next) { 14121 if (act->dtad_kind != DTRACEACT_DIFEXPR) 14122 goto err; 14123 14124 if (act->dtad_difo == NULL) 14125 goto err; 14126 14127 nactions++; 14128 } 14129 14130 helper->dtha_actions = kmem_zalloc(sizeof (dtrace_difo_t *) * 14131 (helper->dtha_nactions = nactions), KM_SLEEP); 14132 14133 for (act = ep->dted_action, i = 0; act != NULL; act = act->dtad_next) { 14134 dtrace_difo_hold(act->dtad_difo); 14135 helper->dtha_actions[i++] = act->dtad_difo; 14136 } 14137 14138 if (!dtrace_helper_validate(helper)) 14139 goto err; 14140 14141 if (last == NULL) { 14142 help->dthps_actions[which] = helper; 14143 } else { 14144 last->dtha_next = helper; 14145 } 14146 14147 if (vstate->dtvs_nlocals > dtrace_helptrace_nlocals) { 14148 dtrace_helptrace_nlocals = vstate->dtvs_nlocals; 14149 dtrace_helptrace_next = 0; 14150 } 14151 14152 return (0); 14153err: 14154 dtrace_helper_action_destroy(helper, vstate); 14155 return (EINVAL); 14156} 14157 14158static void 14159dtrace_helper_provider_register(proc_t *p, dtrace_helpers_t *help, 14160 dof_helper_t *dofhp) 14161{ 14162 ASSERT(MUTEX_NOT_HELD(&dtrace_lock)); 14163 14164 mutex_enter(&dtrace_meta_lock); 14165 mutex_enter(&dtrace_lock); 14166 14167 if (!dtrace_attached() || dtrace_meta_pid == NULL) { 14168 /* 14169 * If the dtrace module is loaded but not attached, or if 14170 * there aren't isn't a meta provider registered to deal with 14171 * these provider descriptions, we need to postpone creating 14172 * the actual providers until later. 14173 */ 14174 14175 if (help->dthps_next == NULL && help->dthps_prev == NULL && 14176 dtrace_deferred_pid != help) { 14177 help->dthps_deferred = 1; 14178 help->dthps_pid = p->p_pid; 14179 help->dthps_next = dtrace_deferred_pid; 14180 help->dthps_prev = NULL; 14181 if (dtrace_deferred_pid != NULL) 14182 dtrace_deferred_pid->dthps_prev = help; 14183 dtrace_deferred_pid = help; 14184 } 14185 14186 mutex_exit(&dtrace_lock); 14187 14188 } else if (dofhp != NULL) { 14189 /* 14190 * If the dtrace module is loaded and we have a particular 14191 * helper provider description, pass that off to the 14192 * meta provider. 14193 */ 14194 14195 mutex_exit(&dtrace_lock); 14196 14197 dtrace_helper_provide(dofhp, p->p_pid); 14198 14199 } else { 14200 /* 14201 * Otherwise, just pass all the helper provider descriptions 14202 * off to the meta provider. 14203 */ 14204 14205 int i; 14206 mutex_exit(&dtrace_lock); 14207 14208 for (i = 0; i < help->dthps_nprovs; i++) { 14209 dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov, 14210 p->p_pid); 14211 } 14212 } 14213 14214 mutex_exit(&dtrace_meta_lock); 14215} 14216 14217static int 14218dtrace_helper_provider_add(dof_helper_t *dofhp, int gen) 14219{ 14220 dtrace_helpers_t *help; 14221 dtrace_helper_provider_t *hprov, **tmp_provs; 14222 uint_t tmp_maxprovs, i; 14223 14224 ASSERT(MUTEX_HELD(&dtrace_lock)); 14225 14226 help = curproc->p_dtrace_helpers; 14227 ASSERT(help != NULL); 14228 14229 /* 14230 * If we already have dtrace_helper_providers_max helper providers, 14231 * we're refuse to add a new one. 14232 */ 14233 if (help->dthps_nprovs >= dtrace_helper_providers_max) 14234 return (ENOSPC); 14235 14236 /* 14237 * Check to make sure this isn't a duplicate. 14238 */ 14239 for (i = 0; i < help->dthps_nprovs; i++) { 14240 if (dofhp->dofhp_addr == 14241 help->dthps_provs[i]->dthp_prov.dofhp_addr) 14242 return (EALREADY); 14243 } 14244 14245 hprov = kmem_zalloc(sizeof (dtrace_helper_provider_t), KM_SLEEP); 14246 hprov->dthp_prov = *dofhp; 14247 hprov->dthp_ref = 1; 14248 hprov->dthp_generation = gen; 14249 14250 /* 14251 * Allocate a bigger table for helper providers if it's already full. 14252 */ 14253 if (help->dthps_maxprovs == help->dthps_nprovs) { 14254 tmp_maxprovs = help->dthps_maxprovs; 14255 tmp_provs = help->dthps_provs; 14256 14257 if (help->dthps_maxprovs == 0) 14258 help->dthps_maxprovs = 2; 14259 else 14260 help->dthps_maxprovs *= 2; 14261 if (help->dthps_maxprovs > dtrace_helper_providers_max) 14262 help->dthps_maxprovs = dtrace_helper_providers_max; 14263 14264 ASSERT(tmp_maxprovs < help->dthps_maxprovs); 14265 14266 help->dthps_provs = kmem_zalloc(help->dthps_maxprovs * 14267 sizeof (dtrace_helper_provider_t *), KM_SLEEP); 14268 14269 if (tmp_provs != NULL) { 14270 bcopy(tmp_provs, help->dthps_provs, tmp_maxprovs * 14271 sizeof (dtrace_helper_provider_t *)); 14272 kmem_free(tmp_provs, tmp_maxprovs * 14273 sizeof (dtrace_helper_provider_t *)); 14274 } 14275 } 14276 14277 help->dthps_provs[help->dthps_nprovs] = hprov; 14278 help->dthps_nprovs++; 14279 14280 return (0); 14281} 14282 14283static void 14284dtrace_helper_provider_destroy(dtrace_helper_provider_t *hprov) 14285{ 14286 mutex_enter(&dtrace_lock); 14287 14288 if (--hprov->dthp_ref == 0) { 14289 dof_hdr_t *dof; 14290 mutex_exit(&dtrace_lock); 14291 dof = (dof_hdr_t *)(uintptr_t)hprov->dthp_prov.dofhp_dof; 14292 dtrace_dof_destroy(dof); 14293 kmem_free(hprov, sizeof (dtrace_helper_provider_t)); 14294 } else { 14295 mutex_exit(&dtrace_lock); 14296 } 14297} 14298 14299static int 14300dtrace_helper_provider_validate(dof_hdr_t *dof, dof_sec_t *sec) 14301{ 14302 uintptr_t daddr = (uintptr_t)dof; 14303 dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec; 14304 dof_provider_t *provider; 14305 dof_probe_t *probe; 14306 uint8_t *arg; 14307 char *strtab, *typestr; 14308 dof_stridx_t typeidx; 14309 size_t typesz; 14310 uint_t nprobes, j, k; 14311 14312 ASSERT(sec->dofs_type == DOF_SECT_PROVIDER); 14313 14314 if (sec->dofs_offset & (sizeof (uint_t) - 1)) { 14315 dtrace_dof_error(dof, "misaligned section offset"); 14316 return (-1); 14317 } 14318 14319 /* 14320 * The section needs to be large enough to contain the DOF provider 14321 * structure appropriate for the given version. 14322 */ 14323 if (sec->dofs_size < 14324 ((dof->dofh_ident[DOF_ID_VERSION] == DOF_VERSION_1) ? 14325 offsetof(dof_provider_t, dofpv_prenoffs) : 14326 sizeof (dof_provider_t))) { 14327 dtrace_dof_error(dof, "provider section too small"); 14328 return (-1); 14329 } 14330 14331 provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); 14332 str_sec = dtrace_dof_sect(dof, DOF_SECT_STRTAB, provider->dofpv_strtab); 14333 prb_sec = dtrace_dof_sect(dof, DOF_SECT_PROBES, provider->dofpv_probes); 14334 arg_sec = dtrace_dof_sect(dof, DOF_SECT_PRARGS, provider->dofpv_prargs); 14335 off_sec = dtrace_dof_sect(dof, DOF_SECT_PROFFS, provider->dofpv_proffs); 14336 14337 if (str_sec == NULL || prb_sec == NULL || 14338 arg_sec == NULL || off_sec == NULL) 14339 return (-1); 14340 14341 enoff_sec = NULL; 14342 14343 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && 14344 provider->dofpv_prenoffs != DOF_SECT_NONE && 14345 (enoff_sec = dtrace_dof_sect(dof, DOF_SECT_PRENOFFS, 14346 provider->dofpv_prenoffs)) == NULL) 14347 return (-1); 14348 14349 strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); 14350 14351 if (provider->dofpv_name >= str_sec->dofs_size || 14352 strlen(strtab + provider->dofpv_name) >= DTRACE_PROVNAMELEN) { 14353 dtrace_dof_error(dof, "invalid provider name"); 14354 return (-1); 14355 } 14356 14357 if (prb_sec->dofs_entsize == 0 || 14358 prb_sec->dofs_entsize > prb_sec->dofs_size) { 14359 dtrace_dof_error(dof, "invalid entry size"); 14360 return (-1); 14361 } 14362 14363 if (prb_sec->dofs_entsize & (sizeof (uintptr_t) - 1)) { 14364 dtrace_dof_error(dof, "misaligned entry size"); 14365 return (-1); 14366 } 14367 14368 if (off_sec->dofs_entsize != sizeof (uint32_t)) { 14369 dtrace_dof_error(dof, "invalid entry size"); 14370 return (-1); 14371 } 14372 14373 if (off_sec->dofs_offset & (sizeof (uint32_t) - 1)) { 14374 dtrace_dof_error(dof, "misaligned section offset"); 14375 return (-1); 14376 } 14377 14378 if (arg_sec->dofs_entsize != sizeof (uint8_t)) { 14379 dtrace_dof_error(dof, "invalid entry size"); 14380 return (-1); 14381 } 14382 14383 arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset); 14384 14385 nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize; 14386 14387 /* 14388 * Take a pass through the probes to check for errors. 14389 */ 14390 for (j = 0; j < nprobes; j++) { 14391 probe = (dof_probe_t *)(uintptr_t)(daddr + 14392 prb_sec->dofs_offset + j * prb_sec->dofs_entsize); 14393 14394 if (probe->dofpr_func >= str_sec->dofs_size) { 14395 dtrace_dof_error(dof, "invalid function name"); 14396 return (-1); 14397 } 14398 14399 if (strlen(strtab + probe->dofpr_func) >= DTRACE_FUNCNAMELEN) { 14400 dtrace_dof_error(dof, "function name too long"); 14401 return (-1); 14402 } 14403 14404 if (probe->dofpr_name >= str_sec->dofs_size || 14405 strlen(strtab + probe->dofpr_name) >= DTRACE_NAMELEN) { 14406 dtrace_dof_error(dof, "invalid probe name"); 14407 return (-1); 14408 } 14409 14410 /* 14411 * The offset count must not wrap the index, and the offsets 14412 * must also not overflow the section's data. 14413 */ 14414 if (probe->dofpr_offidx + probe->dofpr_noffs < 14415 probe->dofpr_offidx || 14416 (probe->dofpr_offidx + probe->dofpr_noffs) * 14417 off_sec->dofs_entsize > off_sec->dofs_size) { 14418 dtrace_dof_error(dof, "invalid probe offset"); 14419 return (-1); 14420 } 14421 14422 if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1) { 14423 /* 14424 * If there's no is-enabled offset section, make sure 14425 * there aren't any is-enabled offsets. Otherwise 14426 * perform the same checks as for probe offsets 14427 * (immediately above). 14428 */ 14429 if (enoff_sec == NULL) { 14430 if (probe->dofpr_enoffidx != 0 || 14431 probe->dofpr_nenoffs != 0) { 14432 dtrace_dof_error(dof, "is-enabled " 14433 "offsets with null section"); 14434 return (-1); 14435 } 14436 } else if (probe->dofpr_enoffidx + 14437 probe->dofpr_nenoffs < probe->dofpr_enoffidx || 14438 (probe->dofpr_enoffidx + probe->dofpr_nenoffs) * 14439 enoff_sec->dofs_entsize > enoff_sec->dofs_size) { 14440 dtrace_dof_error(dof, "invalid is-enabled " 14441 "offset"); 14442 return (-1); 14443 } 14444 14445 if (probe->dofpr_noffs + probe->dofpr_nenoffs == 0) { 14446 dtrace_dof_error(dof, "zero probe and " 14447 "is-enabled offsets"); 14448 return (-1); 14449 } 14450 } else if (probe->dofpr_noffs == 0) { 14451 dtrace_dof_error(dof, "zero probe offsets"); 14452 return (-1); 14453 } 14454 14455 if (probe->dofpr_argidx + probe->dofpr_xargc < 14456 probe->dofpr_argidx || 14457 (probe->dofpr_argidx + probe->dofpr_xargc) * 14458 arg_sec->dofs_entsize > arg_sec->dofs_size) { 14459 dtrace_dof_error(dof, "invalid args"); 14460 return (-1); 14461 } 14462 14463 typeidx = probe->dofpr_nargv; 14464 typestr = strtab + probe->dofpr_nargv; 14465 for (k = 0; k < probe->dofpr_nargc; k++) { 14466 if (typeidx >= str_sec->dofs_size) { 14467 dtrace_dof_error(dof, "bad " 14468 "native argument type"); 14469 return (-1); 14470 } 14471 14472 typesz = strlen(typestr) + 1; 14473 if (typesz > DTRACE_ARGTYPELEN) { 14474 dtrace_dof_error(dof, "native " 14475 "argument type too long"); 14476 return (-1); 14477 } 14478 typeidx += typesz; 14479 typestr += typesz; 14480 } 14481 14482 typeidx = probe->dofpr_xargv; 14483 typestr = strtab + probe->dofpr_xargv; 14484 for (k = 0; k < probe->dofpr_xargc; k++) { 14485 if (arg[probe->dofpr_argidx + k] > probe->dofpr_nargc) { 14486 dtrace_dof_error(dof, "bad " 14487 "native argument index"); 14488 return (-1); 14489 } 14490 14491 if (typeidx >= str_sec->dofs_size) { 14492 dtrace_dof_error(dof, "bad " 14493 "translated argument type"); 14494 return (-1); 14495 } 14496 14497 typesz = strlen(typestr) + 1; 14498 if (typesz > DTRACE_ARGTYPELEN) { 14499 dtrace_dof_error(dof, "translated argument " 14500 "type too long"); 14501 return (-1); 14502 } 14503 14504 typeidx += typesz; 14505 typestr += typesz; 14506 } 14507 } 14508 14509 return (0); 14510} 14511 14512static int 14513dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp) 14514{ 14515 dtrace_helpers_t *help; 14516 dtrace_vstate_t *vstate; 14517 dtrace_enabling_t *enab = NULL; 14518 int i, gen, rv, nhelpers = 0, nprovs = 0, destroy = 1; 14519 uintptr_t daddr = (uintptr_t)dof; 14520 14521 ASSERT(MUTEX_HELD(&dtrace_lock)); 14522 14523 if ((help = curproc->p_dtrace_helpers) == NULL) 14524 help = dtrace_helpers_create(curproc); 14525 14526 vstate = &help->dthps_vstate; 14527 14528 if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab, 14529 dhp != NULL ? dhp->dofhp_addr : 0, B_FALSE)) != 0) { 14530 dtrace_dof_destroy(dof); 14531 return (rv); 14532 } 14533 14534 /* 14535 * Look for helper providers and validate their descriptions. 14536 */ 14537 if (dhp != NULL) { 14538 for (i = 0; i < dof->dofh_secnum; i++) { 14539 dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + 14540 dof->dofh_secoff + i * dof->dofh_secsize); 14541 14542 if (sec->dofs_type != DOF_SECT_PROVIDER) 14543 continue; 14544 14545 if (dtrace_helper_provider_validate(dof, sec) != 0) { 14546 dtrace_enabling_destroy(enab); 14547 dtrace_dof_destroy(dof); 14548 return (-1); 14549 } 14550 14551 nprovs++; 14552 } 14553 } 14554 14555 /* 14556 * Now we need to walk through the ECB descriptions in the enabling. 14557 */ 14558 for (i = 0; i < enab->dten_ndesc; i++) { 14559 dtrace_ecbdesc_t *ep = enab->dten_desc[i]; 14560 dtrace_probedesc_t *desc = &ep->dted_probe; 14561 14562 if (strcmp(desc->dtpd_provider, "dtrace") != 0) 14563 continue; 14564 14565 if (strcmp(desc->dtpd_mod, "helper") != 0) 14566 continue; 14567 14568 if (strcmp(desc->dtpd_func, "ustack") != 0) 14569 continue; 14570 14571 if ((rv = dtrace_helper_action_add(DTRACE_HELPER_ACTION_USTACK, 14572 ep)) != 0) { 14573 /* 14574 * Adding this helper action failed -- we are now going 14575 * to rip out the entire generation and return failure. 14576 */ 14577 (void) dtrace_helper_destroygen(help->dthps_generation); 14578 dtrace_enabling_destroy(enab); 14579 dtrace_dof_destroy(dof); 14580 return (-1); 14581 } 14582 14583 nhelpers++; 14584 } 14585 14586 if (nhelpers < enab->dten_ndesc) 14587 dtrace_dof_error(dof, "unmatched helpers"); 14588 14589 gen = help->dthps_generation++; 14590 dtrace_enabling_destroy(enab); 14591 14592 if (dhp != NULL && nprovs > 0) { 14593 dhp->dofhp_dof = (uint64_t)(uintptr_t)dof; 14594 if (dtrace_helper_provider_add(dhp, gen) == 0) { 14595 mutex_exit(&dtrace_lock); 14596 dtrace_helper_provider_register(curproc, help, dhp); 14597 mutex_enter(&dtrace_lock); 14598 14599 destroy = 0; 14600 } 14601 } 14602 14603 if (destroy) 14604 dtrace_dof_destroy(dof); 14605 14606 return (gen); 14607} 14608 14609static dtrace_helpers_t * 14610dtrace_helpers_create(proc_t *p) 14611{ 14612 dtrace_helpers_t *help; 14613 14614 ASSERT(MUTEX_HELD(&dtrace_lock)); 14615 ASSERT(p->p_dtrace_helpers == NULL); 14616 14617 help = kmem_zalloc(sizeof (dtrace_helpers_t), KM_SLEEP); 14618 help->dthps_actions = kmem_zalloc(sizeof (dtrace_helper_action_t *) * 14619 DTRACE_NHELPER_ACTIONS, KM_SLEEP); 14620 14621 p->p_dtrace_helpers = help; 14622 dtrace_helpers++; 14623 14624 return (help); 14625} 14626 14627static void 14628dtrace_helpers_destroy(void) 14629{ 14630 dtrace_helpers_t *help; 14631 dtrace_vstate_t *vstate; 14632 proc_t *p = curproc; 14633 int i; 14634 14635 mutex_enter(&dtrace_lock); 14636 14637 ASSERT(p->p_dtrace_helpers != NULL); 14638 ASSERT(dtrace_helpers > 0); 14639 14640 help = p->p_dtrace_helpers; 14641 vstate = &help->dthps_vstate; 14642 14643 /* 14644 * We're now going to lose the help from this process. 14645 */ 14646 p->p_dtrace_helpers = NULL; 14647 dtrace_sync(); 14648 14649 /* 14650 * Destory the helper actions. 14651 */ 14652 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { 14653 dtrace_helper_action_t *h, *next; 14654 14655 for (h = help->dthps_actions[i]; h != NULL; h = next) { 14656 next = h->dtha_next; 14657 dtrace_helper_action_destroy(h, vstate); 14658 h = next; 14659 } 14660 } 14661 14662 mutex_exit(&dtrace_lock); 14663 14664 /* 14665 * Destroy the helper providers. 14666 */ 14667 if (help->dthps_maxprovs > 0) { 14668 mutex_enter(&dtrace_meta_lock); 14669 if (dtrace_meta_pid != NULL) { 14670 ASSERT(dtrace_deferred_pid == NULL); 14671 14672 for (i = 0; i < help->dthps_nprovs; i++) { 14673 dtrace_helper_provider_remove( 14674 &help->dthps_provs[i]->dthp_prov, p->p_pid); 14675 } 14676 } else { 14677 mutex_enter(&dtrace_lock); 14678 ASSERT(help->dthps_deferred == 0 || 14679 help->dthps_next != NULL || 14680 help->dthps_prev != NULL || 14681 help == dtrace_deferred_pid); 14682 14683 /* 14684 * Remove the helper from the deferred list. 14685 */ 14686 if (help->dthps_next != NULL) 14687 help->dthps_next->dthps_prev = help->dthps_prev; 14688 if (help->dthps_prev != NULL) 14689 help->dthps_prev->dthps_next = help->dthps_next; 14690 if (dtrace_deferred_pid == help) { 14691 dtrace_deferred_pid = help->dthps_next; 14692 ASSERT(help->dthps_prev == NULL); 14693 } 14694 14695 mutex_exit(&dtrace_lock); 14696 } 14697 14698 mutex_exit(&dtrace_meta_lock); 14699 14700 for (i = 0; i < help->dthps_nprovs; i++) { 14701 dtrace_helper_provider_destroy(help->dthps_provs[i]); 14702 } 14703 14704 kmem_free(help->dthps_provs, help->dthps_maxprovs * 14705 sizeof (dtrace_helper_provider_t *)); 14706 } 14707 14708 mutex_enter(&dtrace_lock); 14709 14710 dtrace_vstate_fini(&help->dthps_vstate); 14711 kmem_free(help->dthps_actions, 14712 sizeof (dtrace_helper_action_t *) * DTRACE_NHELPER_ACTIONS); 14713 kmem_free(help, sizeof (dtrace_helpers_t)); 14714 14715 --dtrace_helpers; 14716 mutex_exit(&dtrace_lock); 14717} 14718 14719static void 14720dtrace_helpers_duplicate(proc_t *from, proc_t *to) 14721{ 14722 dtrace_helpers_t *help, *newhelp; 14723 dtrace_helper_action_t *helper, *new, *last; 14724 dtrace_difo_t *dp; 14725 dtrace_vstate_t *vstate; 14726 int i, j, sz, hasprovs = 0; 14727 14728 mutex_enter(&dtrace_lock); 14729 ASSERT(from->p_dtrace_helpers != NULL); 14730 ASSERT(dtrace_helpers > 0); 14731 14732 help = from->p_dtrace_helpers; 14733 newhelp = dtrace_helpers_create(to); 14734 ASSERT(to->p_dtrace_helpers != NULL); 14735 14736 newhelp->dthps_generation = help->dthps_generation; 14737 vstate = &newhelp->dthps_vstate; 14738 14739 /* 14740 * Duplicate the helper actions. 14741 */ 14742 for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { 14743 if ((helper = help->dthps_actions[i]) == NULL) 14744 continue; 14745 14746 for (last = NULL; helper != NULL; helper = helper->dtha_next) { 14747 new = kmem_zalloc(sizeof (dtrace_helper_action_t), 14748 KM_SLEEP); 14749 new->dtha_generation = helper->dtha_generation; 14750 14751 if ((dp = helper->dtha_predicate) != NULL) { 14752 dp = dtrace_difo_duplicate(dp, vstate); 14753 new->dtha_predicate = dp; 14754 } 14755 14756 new->dtha_nactions = helper->dtha_nactions; 14757 sz = sizeof (dtrace_difo_t *) * new->dtha_nactions; 14758 new->dtha_actions = kmem_alloc(sz, KM_SLEEP); 14759 14760 for (j = 0; j < new->dtha_nactions; j++) { 14761 dtrace_difo_t *dp = helper->dtha_actions[j]; 14762 14763 ASSERT(dp != NULL); 14764 dp = dtrace_difo_duplicate(dp, vstate); 14765 new->dtha_actions[j] = dp; 14766 } 14767 14768 if (last != NULL) { 14769 last->dtha_next = new; 14770 } else { 14771 newhelp->dthps_actions[i] = new; 14772 } 14773 14774 last = new; 14775 } 14776 } 14777 14778 /* 14779 * Duplicate the helper providers and register them with the 14780 * DTrace framework. 14781 */ 14782 if (help->dthps_nprovs > 0) { 14783 newhelp->dthps_nprovs = help->dthps_nprovs; 14784 newhelp->dthps_maxprovs = help->dthps_nprovs; 14785 newhelp->dthps_provs = kmem_alloc(newhelp->dthps_nprovs * 14786 sizeof (dtrace_helper_provider_t *), KM_SLEEP); 14787 for (i = 0; i < newhelp->dthps_nprovs; i++) { 14788 newhelp->dthps_provs[i] = help->dthps_provs[i]; 14789 newhelp->dthps_provs[i]->dthp_ref++; 14790 } 14791 14792 hasprovs = 1; 14793 } 14794 14795 mutex_exit(&dtrace_lock); 14796 14797 if (hasprovs) 14798 dtrace_helper_provider_register(to, newhelp, NULL); 14799} 14800#endif 14801 14802#if defined(sun) 14803/* 14804 * DTrace Hook Functions 14805 */ 14806static void 14807dtrace_module_loaded(modctl_t *ctl) 14808{ 14809 dtrace_provider_t *prv; 14810 14811 mutex_enter(&dtrace_provider_lock); 14812 mutex_enter(&mod_lock); 14813 14814 ASSERT(ctl->mod_busy); 14815 14816 /* 14817 * We're going to call each providers per-module provide operation 14818 * specifying only this module. 14819 */ 14820 for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next) 14821 prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl); 14822 14823 mutex_exit(&mod_lock); 14824 mutex_exit(&dtrace_provider_lock); 14825 14826 /* 14827 * If we have any retained enablings, we need to match against them. 14828 * Enabling probes requires that cpu_lock be held, and we cannot hold 14829 * cpu_lock here -- it is legal for cpu_lock to be held when loading a 14830 * module. (In particular, this happens when loading scheduling 14831 * classes.) So if we have any retained enablings, we need to dispatch 14832 * our task queue to do the match for us. 14833 */ 14834 mutex_enter(&dtrace_lock); 14835 14836 if (dtrace_retained == NULL) { 14837 mutex_exit(&dtrace_lock); 14838 return; 14839 } 14840 14841 (void) taskq_dispatch(dtrace_taskq, 14842 (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP); 14843 14844 mutex_exit(&dtrace_lock); 14845 14846 /* 14847 * And now, for a little heuristic sleaze: in general, we want to 14848 * match modules as soon as they load. However, we cannot guarantee 14849 * this, because it would lead us to the lock ordering violation 14850 * outlined above. The common case, of course, is that cpu_lock is 14851 * _not_ held -- so we delay here for a clock tick, hoping that that's 14852 * long enough for the task queue to do its work. If it's not, it's 14853 * not a serious problem -- it just means that the module that we 14854 * just loaded may not be immediately instrumentable. 14855 */ 14856 delay(1); 14857} 14858 14859static void 14860dtrace_module_unloaded(modctl_t *ctl) 14861{ 14862 dtrace_probe_t template, *probe, *first, *next; 14863 dtrace_provider_t *prov; 14864 14865 template.dtpr_mod = ctl->mod_modname; 14866 14867 mutex_enter(&dtrace_provider_lock); 14868 mutex_enter(&mod_lock); 14869 mutex_enter(&dtrace_lock); 14870 14871 if (dtrace_bymod == NULL) { 14872 /* 14873 * The DTrace module is loaded (obviously) but not attached; 14874 * we don't have any work to do. 14875 */ 14876 mutex_exit(&dtrace_provider_lock); 14877 mutex_exit(&mod_lock); 14878 mutex_exit(&dtrace_lock); 14879 return; 14880 } 14881 14882 for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template); 14883 probe != NULL; probe = probe->dtpr_nextmod) { 14884 if (probe->dtpr_ecb != NULL) { 14885 mutex_exit(&dtrace_provider_lock); 14886 mutex_exit(&mod_lock); 14887 mutex_exit(&dtrace_lock); 14888 14889 /* 14890 * This shouldn't _actually_ be possible -- we're 14891 * unloading a module that has an enabled probe in it. 14892 * (It's normally up to the provider to make sure that 14893 * this can't happen.) However, because dtps_enable() 14894 * doesn't have a failure mode, there can be an 14895 * enable/unload race. Upshot: we don't want to 14896 * assert, but we're not going to disable the 14897 * probe, either. 14898 */ 14899 if (dtrace_err_verbose) { 14900 cmn_err(CE_WARN, "unloaded module '%s' had " 14901 "enabled probes", ctl->mod_modname); 14902 } 14903 14904 return; 14905 } 14906 } 14907 14908 probe = first; 14909 14910 for (first = NULL; probe != NULL; probe = next) { 14911 ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe); 14912 14913 dtrace_probes[probe->dtpr_id - 1] = NULL; 14914 14915 next = probe->dtpr_nextmod; 14916 dtrace_hash_remove(dtrace_bymod, probe); 14917 dtrace_hash_remove(dtrace_byfunc, probe); 14918 dtrace_hash_remove(dtrace_byname, probe); 14919 14920 if (first == NULL) { 14921 first = probe; 14922 probe->dtpr_nextmod = NULL; 14923 } else { 14924 probe->dtpr_nextmod = first; 14925 first = probe; 14926 } 14927 } 14928 14929 /* 14930 * We've removed all of the module's probes from the hash chains and 14931 * from the probe array. Now issue a dtrace_sync() to be sure that 14932 * everyone has cleared out from any probe array processing. 14933 */ 14934 dtrace_sync(); 14935 14936 for (probe = first; probe != NULL; probe = first) { 14937 first = probe->dtpr_nextmod; 14938 prov = probe->dtpr_provider; 14939 prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id, 14940 probe->dtpr_arg); 14941 kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); 14942 kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); 14943 kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); 14944 vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1); 14945 kmem_free(probe, sizeof (dtrace_probe_t)); 14946 } 14947 14948 mutex_exit(&dtrace_lock); 14949 mutex_exit(&mod_lock); 14950 mutex_exit(&dtrace_provider_lock); 14951} 14952 14953static void 14954dtrace_suspend(void) 14955{ 14956 dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_suspend)); 14957} 14958 14959static void 14960dtrace_resume(void) 14961{ 14962 dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_resume)); 14963} 14964#endif 14965 14966static int 14967dtrace_cpu_setup(cpu_setup_t what, processorid_t cpu) 14968{ 14969 ASSERT(MUTEX_HELD(&cpu_lock)); 14970 mutex_enter(&dtrace_lock); 14971 14972 switch (what) { 14973 case CPU_CONFIG: { 14974 dtrace_state_t *state; 14975 dtrace_optval_t *opt, rs, c; 14976 14977 /* 14978 * For now, we only allocate a new buffer for anonymous state. 14979 */ 14980 if ((state = dtrace_anon.dta_state) == NULL) 14981 break; 14982 14983 if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) 14984 break; 14985 14986 opt = state->dts_options; 14987 c = opt[DTRACEOPT_CPU]; 14988 14989 if (c != DTRACE_CPUALL && c != DTRACEOPT_UNSET && c != cpu) 14990 break; 14991 14992 /* 14993 * Regardless of what the actual policy is, we're going to 14994 * temporarily set our resize policy to be manual. We're 14995 * also going to temporarily set our CPU option to denote 14996 * the newly configured CPU. 14997 */ 14998 rs = opt[DTRACEOPT_BUFRESIZE]; 14999 opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_MANUAL; 15000 opt[DTRACEOPT_CPU] = (dtrace_optval_t)cpu; 15001 15002 (void) dtrace_state_buffers(state); 15003 15004 opt[DTRACEOPT_BUFRESIZE] = rs; 15005 opt[DTRACEOPT_CPU] = c; 15006 15007 break; 15008 } 15009 15010 case CPU_UNCONFIG: 15011 /* 15012 * We don't free the buffer in the CPU_UNCONFIG case. (The 15013 * buffer will be freed when the consumer exits.) 15014 */ 15015 break; 15016 15017 default: 15018 break; 15019 } 15020 15021 mutex_exit(&dtrace_lock); 15022 return (0); 15023} 15024 15025#if defined(sun) 15026static void 15027dtrace_cpu_setup_initial(processorid_t cpu) 15028{ 15029 (void) dtrace_cpu_setup(CPU_CONFIG, cpu); 15030} 15031#endif 15032 15033static void 15034dtrace_toxrange_add(uintptr_t base, uintptr_t limit) 15035{ 15036 if (dtrace_toxranges >= dtrace_toxranges_max) { 15037 int osize, nsize; 15038 dtrace_toxrange_t *range; 15039 15040 osize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t); 15041 15042 if (osize == 0) { 15043 ASSERT(dtrace_toxrange == NULL); 15044 ASSERT(dtrace_toxranges_max == 0); 15045 dtrace_toxranges_max = 1; 15046 } else { 15047 dtrace_toxranges_max <<= 1; 15048 } 15049 15050 nsize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t); 15051 range = kmem_zalloc(nsize, KM_SLEEP); 15052 15053 if (dtrace_toxrange != NULL) { 15054 ASSERT(osize != 0); 15055 bcopy(dtrace_toxrange, range, osize); 15056 kmem_free(dtrace_toxrange, osize); 15057 } 15058 15059 dtrace_toxrange = range; 15060 } 15061 15062 ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == 0); 15063 ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == 0); 15064 15065 dtrace_toxrange[dtrace_toxranges].dtt_base = base; 15066 dtrace_toxrange[dtrace_toxranges].dtt_limit = limit; 15067 dtrace_toxranges++; 15068} 15069 15070/* 15071 * DTrace Driver Cookbook Functions 15072 */ 15073#if defined(sun) 15074/*ARGSUSED*/ 15075static int 15076dtrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) 15077{ 15078 dtrace_provider_id_t id; 15079 dtrace_state_t *state = NULL; 15080 dtrace_enabling_t *enab; 15081 15082 mutex_enter(&cpu_lock); 15083 mutex_enter(&dtrace_provider_lock); 15084 mutex_enter(&dtrace_lock); 15085 15086 if (ddi_soft_state_init(&dtrace_softstate, 15087 sizeof (dtrace_state_t), 0) != 0) { 15088 cmn_err(CE_NOTE, "/dev/dtrace failed to initialize soft state"); 15089 mutex_exit(&cpu_lock); 15090 mutex_exit(&dtrace_provider_lock); 15091 mutex_exit(&dtrace_lock); 15092 return (DDI_FAILURE); 15093 } 15094 15095 if (ddi_create_minor_node(devi, DTRACEMNR_DTRACE, S_IFCHR, 15096 DTRACEMNRN_DTRACE, DDI_PSEUDO, NULL) == DDI_FAILURE || 15097 ddi_create_minor_node(devi, DTRACEMNR_HELPER, S_IFCHR, 15098 DTRACEMNRN_HELPER, DDI_PSEUDO, NULL) == DDI_FAILURE) { 15099 cmn_err(CE_NOTE, "/dev/dtrace couldn't create minor nodes"); 15100 ddi_remove_minor_node(devi, NULL); 15101 ddi_soft_state_fini(&dtrace_softstate); 15102 mutex_exit(&cpu_lock); 15103 mutex_exit(&dtrace_provider_lock); 15104 mutex_exit(&dtrace_lock); 15105 return (DDI_FAILURE); 15106 } 15107 15108 ddi_report_dev(devi); 15109 dtrace_devi = devi; 15110 15111 dtrace_modload = dtrace_module_loaded; 15112 dtrace_modunload = dtrace_module_unloaded; 15113 dtrace_cpu_init = dtrace_cpu_setup_initial; 15114 dtrace_helpers_cleanup = dtrace_helpers_destroy; 15115 dtrace_helpers_fork = dtrace_helpers_duplicate; 15116 dtrace_cpustart_init = dtrace_suspend; 15117 dtrace_cpustart_fini = dtrace_resume; 15118 dtrace_debugger_init = dtrace_suspend; 15119 dtrace_debugger_fini = dtrace_resume; 15120 15121 register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); 15122 15123 ASSERT(MUTEX_HELD(&cpu_lock)); 15124 15125 dtrace_arena = vmem_create("dtrace", (void *)1, UINT32_MAX, 1, 15126 NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); 15127 dtrace_minor = vmem_create("dtrace_minor", (void *)DTRACEMNRN_CLONE, 15128 UINT32_MAX - DTRACEMNRN_CLONE, 1, NULL, NULL, NULL, 0, 15129 VM_SLEEP | VMC_IDENTIFIER); 15130 dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri, 15131 1, INT_MAX, 0); 15132 15133 dtrace_state_cache = kmem_cache_create("dtrace_state_cache", 15134 sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN, 15135 NULL, NULL, NULL, NULL, NULL, 0); 15136 15137 ASSERT(MUTEX_HELD(&cpu_lock)); 15138 dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod), 15139 offsetof(dtrace_probe_t, dtpr_nextmod), 15140 offsetof(dtrace_probe_t, dtpr_prevmod)); 15141 15142 dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func), 15143 offsetof(dtrace_probe_t, dtpr_nextfunc), 15144 offsetof(dtrace_probe_t, dtpr_prevfunc)); 15145 15146 dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name), 15147 offsetof(dtrace_probe_t, dtpr_nextname), 15148 offsetof(dtrace_probe_t, dtpr_prevname)); 15149 15150 if (dtrace_retain_max < 1) { 15151 cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; " 15152 "setting to 1", dtrace_retain_max); 15153 dtrace_retain_max = 1; 15154 } 15155 15156 /* 15157 * Now discover our toxic ranges. 15158 */ 15159 dtrace_toxic_ranges(dtrace_toxrange_add); 15160 15161 /* 15162 * Before we register ourselves as a provider to our own framework, 15163 * we would like to assert that dtrace_provider is NULL -- but that's 15164 * not true if we were loaded as a dependency of a DTrace provider. 15165 * Once we've registered, we can assert that dtrace_provider is our 15166 * pseudo provider. 15167 */ 15168 (void) dtrace_register("dtrace", &dtrace_provider_attr, 15169 DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id); 15170 15171 ASSERT(dtrace_provider != NULL); 15172 ASSERT((dtrace_provider_id_t)dtrace_provider == id); 15173 15174 dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t) 15175 dtrace_provider, NULL, NULL, "BEGIN", 0, NULL); 15176 dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t) 15177 dtrace_provider, NULL, NULL, "END", 0, NULL); 15178 dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t) 15179 dtrace_provider, NULL, NULL, "ERROR", 1, NULL); 15180 15181 dtrace_anon_property(); 15182 mutex_exit(&cpu_lock); 15183 15184 /* 15185 * If DTrace helper tracing is enabled, we need to allocate the 15186 * trace buffer and initialize the values. 15187 */ 15188 if (dtrace_helptrace_enabled) { 15189 ASSERT(dtrace_helptrace_buffer == NULL); 15190 dtrace_helptrace_buffer = 15191 kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP); 15192 dtrace_helptrace_next = 0; 15193 } 15194 15195 /* 15196 * If there are already providers, we must ask them to provide their 15197 * probes, and then match any anonymous enabling against them. Note 15198 * that there should be no other retained enablings at this time: 15199 * the only retained enablings at this time should be the anonymous 15200 * enabling. 15201 */ 15202 if (dtrace_anon.dta_enabling != NULL) { 15203 ASSERT(dtrace_retained == dtrace_anon.dta_enabling); 15204 15205 dtrace_enabling_provide(NULL); 15206 state = dtrace_anon.dta_state; 15207 15208 /* 15209 * We couldn't hold cpu_lock across the above call to 15210 * dtrace_enabling_provide(), but we must hold it to actually 15211 * enable the probes. We have to drop all of our locks, pick 15212 * up cpu_lock, and regain our locks before matching the 15213 * retained anonymous enabling. 15214 */ 15215 mutex_exit(&dtrace_lock); 15216 mutex_exit(&dtrace_provider_lock); 15217 15218 mutex_enter(&cpu_lock); 15219 mutex_enter(&dtrace_provider_lock); 15220 mutex_enter(&dtrace_lock); 15221 15222 if ((enab = dtrace_anon.dta_enabling) != NULL) 15223 (void) dtrace_enabling_match(enab, NULL); 15224 15225 mutex_exit(&cpu_lock); 15226 } 15227 15228 mutex_exit(&dtrace_lock); 15229 mutex_exit(&dtrace_provider_lock); 15230 15231 if (state != NULL) { 15232 /* 15233 * If we created any anonymous state, set it going now. 15234 */ 15235 (void) dtrace_state_go(state, &dtrace_anon.dta_beganon); 15236 } 15237 15238 return (DDI_SUCCESS); 15239} 15240#endif 15241 15242#if !defined(sun) 15243#if __FreeBSD_version >= 800039 15244static void 15245dtrace_dtr(void *data __unused) 15246{ 15247} 15248#endif 15249#endif 15250 15251/*ARGSUSED*/ 15252static int 15253#if defined(sun) 15254dtrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) 15255#else 15256dtrace_open(struct cdev *dev, int oflags, int devtype, struct thread *td) 15257#endif 15258{ 15259 dtrace_state_t *state; 15260 uint32_t priv; 15261 uid_t uid; 15262 zoneid_t zoneid; 15263 15264#if defined(sun) 15265 if (getminor(*devp) == DTRACEMNRN_HELPER) 15266 return (0); 15267 15268 /* 15269 * If this wasn't an open with the "helper" minor, then it must be 15270 * the "dtrace" minor. 15271 */ 15272 ASSERT(getminor(*devp) == DTRACEMNRN_DTRACE); 15273#else 15274 cred_t *cred_p = NULL; 15275 15276#if __FreeBSD_version < 800039 15277 /* 15278 * The first minor device is the one that is cloned so there is 15279 * nothing more to do here. 15280 */ 15281 if (dev2unit(dev) == 0) 15282 return 0; 15283 15284 /* 15285 * Devices are cloned, so if the DTrace state has already 15286 * been allocated, that means this device belongs to a 15287 * different client. Each client should open '/dev/dtrace' 15288 * to get a cloned device. 15289 */ 15290 if (dev->si_drv1 != NULL) 15291 return (EBUSY); 15292#endif 15293 15294 cred_p = dev->si_cred; 15295#endif 15296 15297 /* 15298 * If no DTRACE_PRIV_* bits are set in the credential, then the 15299 * caller lacks sufficient permission to do anything with DTrace. 15300 */ 15301 dtrace_cred2priv(cred_p, &priv, &uid, &zoneid); 15302 if (priv == DTRACE_PRIV_NONE) { 15303#if !defined(sun) 15304#if __FreeBSD_version < 800039 15305 /* Destroy the cloned device. */ 15306 destroy_dev(dev); 15307#endif 15308#endif 15309 15310 return (EACCES); 15311 } 15312 15313 /* 15314 * Ask all providers to provide all their probes. 15315 */ 15316 mutex_enter(&dtrace_provider_lock); 15317 dtrace_probe_provide(NULL, NULL); 15318 mutex_exit(&dtrace_provider_lock); 15319 15320 mutex_enter(&cpu_lock); 15321 mutex_enter(&dtrace_lock); 15322 dtrace_opens++; 15323 dtrace_membar_producer(); 15324 15325#if defined(sun) 15326 /* 15327 * If the kernel debugger is active (that is, if the kernel debugger 15328 * modified text in some way), we won't allow the open. 15329 */ 15330 if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { 15331 dtrace_opens--; 15332 mutex_exit(&cpu_lock); 15333 mutex_exit(&dtrace_lock); 15334 return (EBUSY); 15335 } 15336 15337 state = dtrace_state_create(devp, cred_p); 15338#else 15339 state = dtrace_state_create(dev); 15340#if __FreeBSD_version < 800039 15341 dev->si_drv1 = state; 15342#else 15343 devfs_set_cdevpriv(state, dtrace_dtr); 15344#endif 15345#endif 15346 15347 mutex_exit(&cpu_lock); 15348 15349 if (state == NULL) { 15350#if defined(sun) 15351 if (--dtrace_opens == 0) 15352 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); 15353#else 15354 --dtrace_opens; 15355#endif 15356 mutex_exit(&dtrace_lock); 15357#if !defined(sun) 15358#if __FreeBSD_version < 800039 15359 /* Destroy the cloned device. */ 15360 destroy_dev(dev); 15361#endif 15362#endif 15363 return (EAGAIN); 15364 } 15365 15366 mutex_exit(&dtrace_lock); 15367 15368 return (0); 15369} 15370 15371/*ARGSUSED*/ 15372static int 15373#if defined(sun) 15374dtrace_close(dev_t dev, int flag, int otyp, cred_t *cred_p) 15375#else 15376dtrace_close(struct cdev *dev, int flags, int fmt __unused, struct thread *td) 15377#endif 15378{ 15379#if defined(sun) 15380 minor_t minor = getminor(dev); 15381 dtrace_state_t *state; 15382 15383 if (minor == DTRACEMNRN_HELPER) 15384 return (0); 15385 15386 state = ddi_get_soft_state(dtrace_softstate, minor); 15387#else 15388#if __FreeBSD_version < 800039 15389 dtrace_state_t *state = dev->si_drv1; 15390 15391 /* Check if this is not a cloned device. */ 15392 if (dev2unit(dev) == 0) 15393 return (0); 15394#else 15395 dtrace_state_t *state; 15396 devfs_get_cdevpriv((void **) &state); 15397#endif 15398 15399#endif 15400 15401 mutex_enter(&cpu_lock); 15402 mutex_enter(&dtrace_lock); 15403 15404 if (state != NULL) { 15405 if (state->dts_anon) { 15406 /* 15407 * There is anonymous state. Destroy that first. 15408 */ 15409 ASSERT(dtrace_anon.dta_state == NULL); 15410 dtrace_state_destroy(state->dts_anon); 15411 } 15412 15413 dtrace_state_destroy(state); 15414 15415#if !defined(sun) 15416 kmem_free(state, 0); 15417#if __FreeBSD_version < 800039 15418 dev->si_drv1 = NULL; 15419#else 15420 devfs_clear_cdevpriv(); 15421#endif 15422#endif 15423 } 15424 15425 ASSERT(dtrace_opens > 0); 15426#if defined(sun) 15427 if (--dtrace_opens == 0) 15428 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); 15429#else 15430 --dtrace_opens; 15431#endif 15432 15433 mutex_exit(&dtrace_lock); 15434 mutex_exit(&cpu_lock); 15435 15436#if __FreeBSD_version < 800039 15437 /* Schedule this cloned device to be destroyed. */ 15438 destroy_dev_sched(dev); 15439#endif 15440 15441 return (0); 15442} 15443 15444#if defined(sun) 15445/*ARGSUSED*/ 15446static int 15447dtrace_ioctl_helper(int cmd, intptr_t arg, int *rv) 15448{ 15449 int rval; 15450 dof_helper_t help, *dhp = NULL; 15451 15452 switch (cmd) { 15453 case DTRACEHIOC_ADDDOF: 15454 if (copyin((void *)arg, &help, sizeof (help)) != 0) { 15455 dtrace_dof_error(NULL, "failed to copyin DOF helper"); 15456 return (EFAULT); 15457 } 15458 15459 dhp = &help; 15460 arg = (intptr_t)help.dofhp_dof; 15461 /*FALLTHROUGH*/ 15462 15463 case DTRACEHIOC_ADD: { 15464 dof_hdr_t *dof = dtrace_dof_copyin(arg, &rval); 15465 15466 if (dof == NULL) 15467 return (rval); 15468 15469 mutex_enter(&dtrace_lock); 15470 15471 /* 15472 * dtrace_helper_slurp() takes responsibility for the dof -- 15473 * it may free it now or it may save it and free it later. 15474 */ 15475 if ((rval = dtrace_helper_slurp(dof, dhp)) != -1) { 15476 *rv = rval; 15477 rval = 0; 15478 } else { 15479 rval = EINVAL; 15480 } 15481 15482 mutex_exit(&dtrace_lock); 15483 return (rval); 15484 } 15485 15486 case DTRACEHIOC_REMOVE: { 15487 mutex_enter(&dtrace_lock); 15488 rval = dtrace_helper_destroygen(arg); 15489 mutex_exit(&dtrace_lock); 15490 15491 return (rval); 15492 } 15493 15494 default: 15495 break; 15496 } 15497 15498 return (ENOTTY); 15499} 15500 15501/*ARGSUSED*/ 15502static int 15503dtrace_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv) 15504{ 15505 minor_t minor = getminor(dev); 15506 dtrace_state_t *state; 15507 int rval; 15508 15509 if (minor == DTRACEMNRN_HELPER) 15510 return (dtrace_ioctl_helper(cmd, arg, rv)); 15511 15512 state = ddi_get_soft_state(dtrace_softstate, minor); 15513 15514 if (state->dts_anon) { 15515 ASSERT(dtrace_anon.dta_state == NULL); 15516 state = state->dts_anon; 15517 } 15518 15519 switch (cmd) { 15520 case DTRACEIOC_PROVIDER: { 15521 dtrace_providerdesc_t pvd; 15522 dtrace_provider_t *pvp; 15523 15524 if (copyin((void *)arg, &pvd, sizeof (pvd)) != 0) 15525 return (EFAULT); 15526 15527 pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0'; 15528 mutex_enter(&dtrace_provider_lock); 15529 15530 for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) { 15531 if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0) 15532 break; 15533 } 15534 15535 mutex_exit(&dtrace_provider_lock); 15536 15537 if (pvp == NULL) 15538 return (ESRCH); 15539 15540 bcopy(&pvp->dtpv_priv, &pvd.dtvd_priv, sizeof (dtrace_ppriv_t)); 15541 bcopy(&pvp->dtpv_attr, &pvd.dtvd_attr, sizeof (dtrace_pattr_t)); 15542 15543 if (copyout(&pvd, (void *)arg, sizeof (pvd)) != 0) 15544 return (EFAULT); 15545 15546 return (0); 15547 } 15548 15549 case DTRACEIOC_EPROBE: { 15550 dtrace_eprobedesc_t epdesc; 15551 dtrace_ecb_t *ecb; 15552 dtrace_action_t *act; 15553 void *buf; 15554 size_t size; 15555 uintptr_t dest; 15556 int nrecs; 15557 15558 if (copyin((void *)arg, &epdesc, sizeof (epdesc)) != 0) 15559 return (EFAULT); 15560 15561 mutex_enter(&dtrace_lock); 15562 15563 if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) { 15564 mutex_exit(&dtrace_lock); 15565 return (EINVAL); 15566 } 15567 15568 if (ecb->dte_probe == NULL) { 15569 mutex_exit(&dtrace_lock); 15570 return (EINVAL); 15571 } 15572 15573 epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id; 15574 epdesc.dtepd_uarg = ecb->dte_uarg; 15575 epdesc.dtepd_size = ecb->dte_size; 15576 15577 nrecs = epdesc.dtepd_nrecs; 15578 epdesc.dtepd_nrecs = 0; 15579 for (act = ecb->dte_action; act != NULL; act = act->dta_next) { 15580 if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) 15581 continue; 15582 15583 epdesc.dtepd_nrecs++; 15584 } 15585 15586 /* 15587 * Now that we have the size, we need to allocate a temporary 15588 * buffer in which to store the complete description. We need 15589 * the temporary buffer to be able to drop dtrace_lock() 15590 * across the copyout(), below. 15591 */ 15592 size = sizeof (dtrace_eprobedesc_t) + 15593 (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t)); 15594 15595 buf = kmem_alloc(size, KM_SLEEP); 15596 dest = (uintptr_t)buf; 15597 15598 bcopy(&epdesc, (void *)dest, sizeof (epdesc)); 15599 dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]); 15600 15601 for (act = ecb->dte_action; act != NULL; act = act->dta_next) { 15602 if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) 15603 continue; 15604 15605 if (nrecs-- == 0) 15606 break; 15607 15608 bcopy(&act->dta_rec, (void *)dest, 15609 sizeof (dtrace_recdesc_t)); 15610 dest += sizeof (dtrace_recdesc_t); 15611 } 15612 15613 mutex_exit(&dtrace_lock); 15614 15615 if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) { 15616 kmem_free(buf, size); 15617 return (EFAULT); 15618 } 15619 15620 kmem_free(buf, size); 15621 return (0); 15622 } 15623 15624 case DTRACEIOC_AGGDESC: { 15625 dtrace_aggdesc_t aggdesc; 15626 dtrace_action_t *act; 15627 dtrace_aggregation_t *agg; 15628 int nrecs; 15629 uint32_t offs; 15630 dtrace_recdesc_t *lrec; 15631 void *buf; 15632 size_t size; 15633 uintptr_t dest; 15634 15635 if (copyin((void *)arg, &aggdesc, sizeof (aggdesc)) != 0) 15636 return (EFAULT); 15637 15638 mutex_enter(&dtrace_lock); 15639 15640 if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) { 15641 mutex_exit(&dtrace_lock); 15642 return (EINVAL); 15643 } 15644 15645 aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid; 15646 15647 nrecs = aggdesc.dtagd_nrecs; 15648 aggdesc.dtagd_nrecs = 0; 15649 15650 offs = agg->dtag_base; 15651 lrec = &agg->dtag_action.dta_rec; 15652 aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs; 15653 15654 for (act = agg->dtag_first; ; act = act->dta_next) { 15655 ASSERT(act->dta_intuple || 15656 DTRACEACT_ISAGG(act->dta_kind)); 15657 15658 /* 15659 * If this action has a record size of zero, it 15660 * denotes an argument to the aggregating action. 15661 * Because the presence of this record doesn't (or 15662 * shouldn't) affect the way the data is interpreted, 15663 * we don't copy it out to save user-level the 15664 * confusion of dealing with a zero-length record. 15665 */ 15666 if (act->dta_rec.dtrd_size == 0) { 15667 ASSERT(agg->dtag_hasarg); 15668 continue; 15669 } 15670 15671 aggdesc.dtagd_nrecs++; 15672 15673 if (act == &agg->dtag_action) 15674 break; 15675 } 15676 15677 /* 15678 * Now that we have the size, we need to allocate a temporary 15679 * buffer in which to store the complete description. We need 15680 * the temporary buffer to be able to drop dtrace_lock() 15681 * across the copyout(), below. 15682 */ 15683 size = sizeof (dtrace_aggdesc_t) + 15684 (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t)); 15685 15686 buf = kmem_alloc(size, KM_SLEEP); 15687 dest = (uintptr_t)buf; 15688 15689 bcopy(&aggdesc, (void *)dest, sizeof (aggdesc)); 15690 dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]); 15691 15692 for (act = agg->dtag_first; ; act = act->dta_next) { 15693 dtrace_recdesc_t rec = act->dta_rec; 15694 15695 /* 15696 * See the comment in the above loop for why we pass 15697 * over zero-length records. 15698 */ 15699 if (rec.dtrd_size == 0) { 15700 ASSERT(agg->dtag_hasarg); 15701 continue; 15702 } 15703 15704 if (nrecs-- == 0) 15705 break; 15706 15707 rec.dtrd_offset -= offs; 15708 bcopy(&rec, (void *)dest, sizeof (rec)); 15709 dest += sizeof (dtrace_recdesc_t); 15710 15711 if (act == &agg->dtag_action) 15712 break; 15713 } 15714 15715 mutex_exit(&dtrace_lock); 15716 15717 if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) { 15718 kmem_free(buf, size); 15719 return (EFAULT); 15720 } 15721 15722 kmem_free(buf, size); 15723 return (0); 15724 } 15725 15726 case DTRACEIOC_ENABLE: { 15727 dof_hdr_t *dof; 15728 dtrace_enabling_t *enab = NULL; 15729 dtrace_vstate_t *vstate; 15730 int err = 0; 15731 15732 *rv = 0; 15733 15734 /* 15735 * If a NULL argument has been passed, we take this as our 15736 * cue to reevaluate our enablings. 15737 */ 15738 if (arg == NULL) { 15739 dtrace_enabling_matchall(); 15740 15741 return (0); 15742 } 15743 15744 if ((dof = dtrace_dof_copyin(arg, &rval)) == NULL) 15745 return (rval); 15746 15747 mutex_enter(&cpu_lock); 15748 mutex_enter(&dtrace_lock); 15749 vstate = &state->dts_vstate; 15750 15751 if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { 15752 mutex_exit(&dtrace_lock); 15753 mutex_exit(&cpu_lock); 15754 dtrace_dof_destroy(dof); 15755 return (EBUSY); 15756 } 15757 15758 if (dtrace_dof_slurp(dof, vstate, cr, &enab, 0, B_TRUE) != 0) { 15759 mutex_exit(&dtrace_lock); 15760 mutex_exit(&cpu_lock); 15761 dtrace_dof_destroy(dof); 15762 return (EINVAL); 15763 } 15764 15765 if ((rval = dtrace_dof_options(dof, state)) != 0) { 15766 dtrace_enabling_destroy(enab); 15767 mutex_exit(&dtrace_lock); 15768 mutex_exit(&cpu_lock); 15769 dtrace_dof_destroy(dof); 15770 return (rval); 15771 } 15772 15773 if ((err = dtrace_enabling_match(enab, rv)) == 0) { 15774 err = dtrace_enabling_retain(enab); 15775 } else { 15776 dtrace_enabling_destroy(enab); 15777 } 15778 15779 mutex_exit(&cpu_lock); 15780 mutex_exit(&dtrace_lock); 15781 dtrace_dof_destroy(dof); 15782 15783 return (err); 15784 } 15785 15786 case DTRACEIOC_REPLICATE: { 15787 dtrace_repldesc_t desc; 15788 dtrace_probedesc_t *match = &desc.dtrpd_match; 15789 dtrace_probedesc_t *create = &desc.dtrpd_create; 15790 int err; 15791 15792 if (copyin((void *)arg, &desc, sizeof (desc)) != 0) 15793 return (EFAULT); 15794 15795 match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; 15796 match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; 15797 match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; 15798 match->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; 15799 15800 create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; 15801 create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; 15802 create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; 15803 create->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; 15804 15805 mutex_enter(&dtrace_lock); 15806 err = dtrace_enabling_replicate(state, match, create); 15807 mutex_exit(&dtrace_lock); 15808 15809 return (err); 15810 } 15811 15812 case DTRACEIOC_PROBEMATCH: 15813 case DTRACEIOC_PROBES: { 15814 dtrace_probe_t *probe = NULL; 15815 dtrace_probedesc_t desc; 15816 dtrace_probekey_t pkey; 15817 dtrace_id_t i; 15818 int m = 0; 15819 uint32_t priv; 15820 uid_t uid; 15821 zoneid_t zoneid; 15822 15823 if (copyin((void *)arg, &desc, sizeof (desc)) != 0) 15824 return (EFAULT); 15825 15826 desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; 15827 desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; 15828 desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; 15829 desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0'; 15830 15831 /* 15832 * Before we attempt to match this probe, we want to give 15833 * all providers the opportunity to provide it. 15834 */ 15835 if (desc.dtpd_id == DTRACE_IDNONE) { 15836 mutex_enter(&dtrace_provider_lock); 15837 dtrace_probe_provide(&desc, NULL); 15838 mutex_exit(&dtrace_provider_lock); 15839 desc.dtpd_id++; 15840 } 15841 15842 if (cmd == DTRACEIOC_PROBEMATCH) { 15843 dtrace_probekey(&desc, &pkey); 15844 pkey.dtpk_id = DTRACE_IDNONE; 15845 } 15846 15847 dtrace_cred2priv(cr, &priv, &uid, &zoneid); 15848 15849 mutex_enter(&dtrace_lock); 15850 15851 if (cmd == DTRACEIOC_PROBEMATCH) { 15852 for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) { 15853 if ((probe = dtrace_probes[i - 1]) != NULL && 15854 (m = dtrace_match_probe(probe, &pkey, 15855 priv, uid, zoneid)) != 0) 15856 break; 15857 } 15858 15859 if (m < 0) { 15860 mutex_exit(&dtrace_lock); 15861 return (EINVAL); 15862 } 15863 15864 } else { 15865 for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) { 15866 if ((probe = dtrace_probes[i - 1]) != NULL && 15867 dtrace_match_priv(probe, priv, uid, zoneid)) 15868 break; 15869 } 15870 } 15871 15872 if (probe == NULL) { 15873 mutex_exit(&dtrace_lock); 15874 return (ESRCH); 15875 } 15876 15877 dtrace_probe_description(probe, &desc); 15878 mutex_exit(&dtrace_lock); 15879 15880 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) 15881 return (EFAULT); 15882 15883 return (0); 15884 } 15885 15886 case DTRACEIOC_PROBEARG: { 15887 dtrace_argdesc_t desc; 15888 dtrace_probe_t *probe; 15889 dtrace_provider_t *prov; 15890 15891 if (copyin((void *)arg, &desc, sizeof (desc)) != 0) 15892 return (EFAULT); 15893 15894 if (desc.dtargd_id == DTRACE_IDNONE) 15895 return (EINVAL); 15896 15897 if (desc.dtargd_ndx == DTRACE_ARGNONE) 15898 return (EINVAL); 15899 15900 mutex_enter(&dtrace_provider_lock); 15901 mutex_enter(&mod_lock); 15902 mutex_enter(&dtrace_lock); 15903 15904 if (desc.dtargd_id > dtrace_nprobes) { 15905 mutex_exit(&dtrace_lock); 15906 mutex_exit(&mod_lock); 15907 mutex_exit(&dtrace_provider_lock); 15908 return (EINVAL); 15909 } 15910 15911 if ((probe = dtrace_probes[desc.dtargd_id - 1]) == NULL) { 15912 mutex_exit(&dtrace_lock); 15913 mutex_exit(&mod_lock); 15914 mutex_exit(&dtrace_provider_lock); 15915 return (EINVAL); 15916 } 15917 15918 mutex_exit(&dtrace_lock); 15919 15920 prov = probe->dtpr_provider; 15921 15922 if (prov->dtpv_pops.dtps_getargdesc == NULL) { 15923 /* 15924 * There isn't any typed information for this probe. 15925 * Set the argument number to DTRACE_ARGNONE. 15926 */ 15927 desc.dtargd_ndx = DTRACE_ARGNONE; 15928 } else { 15929 desc.dtargd_native[0] = '\0'; 15930 desc.dtargd_xlate[0] = '\0'; 15931 desc.dtargd_mapping = desc.dtargd_ndx; 15932 15933 prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg, 15934 probe->dtpr_id, probe->dtpr_arg, &desc); 15935 } 15936 15937 mutex_exit(&mod_lock); 15938 mutex_exit(&dtrace_provider_lock); 15939 15940 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) 15941 return (EFAULT); 15942 15943 return (0); 15944 } 15945 15946 case DTRACEIOC_GO: { 15947 processorid_t cpuid; 15948 rval = dtrace_state_go(state, &cpuid); 15949 15950 if (rval != 0) 15951 return (rval); 15952 15953 if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0) 15954 return (EFAULT); 15955 15956 return (0); 15957 } 15958 15959 case DTRACEIOC_STOP: { 15960 processorid_t cpuid; 15961 15962 mutex_enter(&dtrace_lock); 15963 rval = dtrace_state_stop(state, &cpuid); 15964 mutex_exit(&dtrace_lock); 15965 15966 if (rval != 0) 15967 return (rval); 15968 15969 if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0) 15970 return (EFAULT); 15971 15972 return (0); 15973 } 15974 15975 case DTRACEIOC_DOFGET: { 15976 dof_hdr_t hdr, *dof; 15977 uint64_t len; 15978 15979 if (copyin((void *)arg, &hdr, sizeof (hdr)) != 0) 15980 return (EFAULT); 15981 15982 mutex_enter(&dtrace_lock); 15983 dof = dtrace_dof_create(state); 15984 mutex_exit(&dtrace_lock); 15985 15986 len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz); 15987 rval = copyout(dof, (void *)arg, len); 15988 dtrace_dof_destroy(dof); 15989 15990 return (rval == 0 ? 0 : EFAULT); 15991 } 15992 15993 case DTRACEIOC_AGGSNAP: 15994 case DTRACEIOC_BUFSNAP: { 15995 dtrace_bufdesc_t desc; 15996 caddr_t cached; 15997 dtrace_buffer_t *buf; 15998 15999 if (copyin((void *)arg, &desc, sizeof (desc)) != 0) 16000 return (EFAULT); 16001 16002 if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NCPU) 16003 return (EINVAL); 16004 16005 mutex_enter(&dtrace_lock); 16006 16007 if (cmd == DTRACEIOC_BUFSNAP) { 16008 buf = &state->dts_buffer[desc.dtbd_cpu]; 16009 } else { 16010 buf = &state->dts_aggbuffer[desc.dtbd_cpu]; 16011 } 16012 16013 if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) { 16014 size_t sz = buf->dtb_offset; 16015 16016 if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) { 16017 mutex_exit(&dtrace_lock); 16018 return (EBUSY); 16019 } 16020 16021 /* 16022 * If this buffer has already been consumed, we're 16023 * going to indicate that there's nothing left here 16024 * to consume. 16025 */ 16026 if (buf->dtb_flags & DTRACEBUF_CONSUMED) { 16027 mutex_exit(&dtrace_lock); 16028 16029 desc.dtbd_size = 0; 16030 desc.dtbd_drops = 0; 16031 desc.dtbd_errors = 0; 16032 desc.dtbd_oldest = 0; 16033 sz = sizeof (desc); 16034 16035 if (copyout(&desc, (void *)arg, sz) != 0) 16036 return (EFAULT); 16037 16038 return (0); 16039 } 16040 16041 /* 16042 * If this is a ring buffer that has wrapped, we want 16043 * to copy the whole thing out. 16044 */ 16045 if (buf->dtb_flags & DTRACEBUF_WRAPPED) { 16046 dtrace_buffer_polish(buf); 16047 sz = buf->dtb_size; 16048 } 16049 16050 if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) { 16051 mutex_exit(&dtrace_lock); 16052 return (EFAULT); 16053 } 16054 16055 desc.dtbd_size = sz; 16056 desc.dtbd_drops = buf->dtb_drops; 16057 desc.dtbd_errors = buf->dtb_errors; 16058 desc.dtbd_oldest = buf->dtb_xamot_offset; 16059 16060 mutex_exit(&dtrace_lock); 16061 16062 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) 16063 return (EFAULT); 16064 16065 buf->dtb_flags |= DTRACEBUF_CONSUMED; 16066 16067 return (0); 16068 } 16069 16070 if (buf->dtb_tomax == NULL) { 16071 ASSERT(buf->dtb_xamot == NULL); 16072 mutex_exit(&dtrace_lock); 16073 return (ENOENT); 16074 } 16075 16076 cached = buf->dtb_tomax; 16077 ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); 16078 16079 dtrace_xcall(desc.dtbd_cpu, 16080 (dtrace_xcall_t)dtrace_buffer_switch, buf); 16081 16082 state->dts_errors += buf->dtb_xamot_errors; 16083 16084 /* 16085 * If the buffers did not actually switch, then the cross call 16086 * did not take place -- presumably because the given CPU is 16087 * not in the ready set. If this is the case, we'll return 16088 * ENOENT. 16089 */ 16090 if (buf->dtb_tomax == cached) { 16091 ASSERT(buf->dtb_xamot != cached); 16092 mutex_exit(&dtrace_lock); 16093 return (ENOENT); 16094 } 16095 16096 ASSERT(cached == buf->dtb_xamot); 16097 16098 /* 16099 * We have our snapshot; now copy it out. 16100 */ 16101 if (copyout(buf->dtb_xamot, desc.dtbd_data, 16102 buf->dtb_xamot_offset) != 0) { 16103 mutex_exit(&dtrace_lock); 16104 return (EFAULT); 16105 } 16106 16107 desc.dtbd_size = buf->dtb_xamot_offset; 16108 desc.dtbd_drops = buf->dtb_xamot_drops; 16109 desc.dtbd_errors = buf->dtb_xamot_errors; 16110 desc.dtbd_oldest = 0; 16111 16112 mutex_exit(&dtrace_lock); 16113 16114 /* 16115 * Finally, copy out the buffer description. 16116 */ 16117 if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) 16118 return (EFAULT); 16119 16120 return (0); 16121 } 16122 16123 case DTRACEIOC_CONF: { 16124 dtrace_conf_t conf; 16125 16126 bzero(&conf, sizeof (conf)); 16127 conf.dtc_difversion = DIF_VERSION; 16128 conf.dtc_difintregs = DIF_DIR_NREGS; 16129 conf.dtc_diftupregs = DIF_DTR_NREGS; 16130 conf.dtc_ctfmodel = CTF_MODEL_NATIVE; 16131 16132 if (copyout(&conf, (void *)arg, sizeof (conf)) != 0) 16133 return (EFAULT); 16134 16135 return (0); 16136 } 16137 16138 case DTRACEIOC_STATUS: { 16139 dtrace_status_t stat; 16140 dtrace_dstate_t *dstate; 16141 int i, j; 16142 uint64_t nerrs; 16143 16144 /* 16145 * See the comment in dtrace_state_deadman() for the reason 16146 * for setting dts_laststatus to INT64_MAX before setting 16147 * it to the correct value. 16148 */ 16149 state->dts_laststatus = INT64_MAX; 16150 dtrace_membar_producer(); 16151 state->dts_laststatus = dtrace_gethrtime(); 16152 16153 bzero(&stat, sizeof (stat)); 16154 16155 mutex_enter(&dtrace_lock); 16156 16157 if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) { 16158 mutex_exit(&dtrace_lock); 16159 return (ENOENT); 16160 } 16161 16162 if (state->dts_activity == DTRACE_ACTIVITY_DRAINING) 16163 stat.dtst_exiting = 1; 16164 16165 nerrs = state->dts_errors; 16166 dstate = &state->dts_vstate.dtvs_dynvars; 16167 16168 for (i = 0; i < NCPU; i++) { 16169 dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i]; 16170 16171 stat.dtst_dyndrops += dcpu->dtdsc_drops; 16172 stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops; 16173 stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops; 16174 16175 if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL) 16176 stat.dtst_filled++; 16177 16178 nerrs += state->dts_buffer[i].dtb_errors; 16179 16180 for (j = 0; j < state->dts_nspeculations; j++) { 16181 dtrace_speculation_t *spec; 16182 dtrace_buffer_t *buf; 16183 16184 spec = &state->dts_speculations[j]; 16185 buf = &spec->dtsp_buffer[i]; 16186 stat.dtst_specdrops += buf->dtb_xamot_drops; 16187 } 16188 } 16189 16190 stat.dtst_specdrops_busy = state->dts_speculations_busy; 16191 stat.dtst_specdrops_unavail = state->dts_speculations_unavail; 16192 stat.dtst_stkstroverflows = state->dts_stkstroverflows; 16193 stat.dtst_dblerrors = state->dts_dblerrors; 16194 stat.dtst_killed = 16195 (state->dts_activity == DTRACE_ACTIVITY_KILLED); 16196 stat.dtst_errors = nerrs; 16197 16198 mutex_exit(&dtrace_lock); 16199 16200 if (copyout(&stat, (void *)arg, sizeof (stat)) != 0) 16201 return (EFAULT); 16202 16203 return (0); 16204 } 16205 16206 case DTRACEIOC_FORMAT: { 16207 dtrace_fmtdesc_t fmt; 16208 char *str; 16209 int len; 16210 16211 if (copyin((void *)arg, &fmt, sizeof (fmt)) != 0) 16212 return (EFAULT); 16213 16214 mutex_enter(&dtrace_lock); 16215 16216 if (fmt.dtfd_format == 0 || 16217 fmt.dtfd_format > state->dts_nformats) { 16218 mutex_exit(&dtrace_lock); 16219 return (EINVAL); 16220 } 16221 16222 /* 16223 * Format strings are allocated contiguously and they are 16224 * never freed; if a format index is less than the number 16225 * of formats, we can assert that the format map is non-NULL 16226 * and that the format for the specified index is non-NULL. 16227 */ 16228 ASSERT(state->dts_formats != NULL); 16229 str = state->dts_formats[fmt.dtfd_format - 1]; 16230 ASSERT(str != NULL); 16231 16232 len = strlen(str) + 1; 16233 16234 if (len > fmt.dtfd_length) { 16235 fmt.dtfd_length = len; 16236 16237 if (copyout(&fmt, (void *)arg, sizeof (fmt)) != 0) { 16238 mutex_exit(&dtrace_lock); 16239 return (EINVAL); 16240 } 16241 } else { 16242 if (copyout(str, fmt.dtfd_string, len) != 0) { 16243 mutex_exit(&dtrace_lock); 16244 return (EINVAL); 16245 } 16246 } 16247 16248 mutex_exit(&dtrace_lock); 16249 return (0); 16250 } 16251 16252 default: 16253 break; 16254 } 16255 16256 return (ENOTTY); 16257} 16258 16259/*ARGSUSED*/ 16260static int 16261dtrace_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) 16262{ 16263 dtrace_state_t *state; 16264 16265 switch (cmd) { 16266 case DDI_DETACH: 16267 break; 16268 16269 case DDI_SUSPEND: 16270 return (DDI_SUCCESS); 16271 16272 default: 16273 return (DDI_FAILURE); 16274 } 16275 16276 mutex_enter(&cpu_lock); 16277 mutex_enter(&dtrace_provider_lock); 16278 mutex_enter(&dtrace_lock); 16279 16280 ASSERT(dtrace_opens == 0); 16281 16282 if (dtrace_helpers > 0) { 16283 mutex_exit(&dtrace_provider_lock); 16284 mutex_exit(&dtrace_lock); 16285 mutex_exit(&cpu_lock); 16286 return (DDI_FAILURE); 16287 } 16288 16289 if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) { 16290 mutex_exit(&dtrace_provider_lock); 16291 mutex_exit(&dtrace_lock); 16292 mutex_exit(&cpu_lock); 16293 return (DDI_FAILURE); 16294 } 16295 16296 dtrace_provider = NULL; 16297 16298 if ((state = dtrace_anon_grab()) != NULL) { 16299 /* 16300 * If there were ECBs on this state, the provider should 16301 * have not been allowed to detach; assert that there is 16302 * none. 16303 */ 16304 ASSERT(state->dts_necbs == 0); 16305 dtrace_state_destroy(state); 16306 16307 /* 16308 * If we're being detached with anonymous state, we need to 16309 * indicate to the kernel debugger that DTrace is now inactive. 16310 */ 16311 (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); 16312 } 16313 16314 bzero(&dtrace_anon, sizeof (dtrace_anon_t)); 16315 unregister_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); 16316 dtrace_cpu_init = NULL; 16317 dtrace_helpers_cleanup = NULL; 16318 dtrace_helpers_fork = NULL; 16319 dtrace_cpustart_init = NULL; 16320 dtrace_cpustart_fini = NULL; 16321 dtrace_debugger_init = NULL; 16322 dtrace_debugger_fini = NULL; 16323 dtrace_modload = NULL; 16324 dtrace_modunload = NULL; 16325 16326 mutex_exit(&cpu_lock); 16327 16328 if (dtrace_helptrace_enabled) { 16329 kmem_free(dtrace_helptrace_buffer, dtrace_helptrace_bufsize); 16330 dtrace_helptrace_buffer = NULL; 16331 } 16332 16333 kmem_free(dtrace_probes, dtrace_nprobes * sizeof (dtrace_probe_t *)); 16334 dtrace_probes = NULL; 16335 dtrace_nprobes = 0; 16336 16337 dtrace_hash_destroy(dtrace_bymod); 16338 dtrace_hash_destroy(dtrace_byfunc); 16339 dtrace_hash_destroy(dtrace_byname); 16340 dtrace_bymod = NULL; 16341 dtrace_byfunc = NULL; 16342 dtrace_byname = NULL; 16343 16344 kmem_cache_destroy(dtrace_state_cache); 16345 vmem_destroy(dtrace_minor); 16346 vmem_destroy(dtrace_arena); 16347 16348 if (dtrace_toxrange != NULL) { 16349 kmem_free(dtrace_toxrange, 16350 dtrace_toxranges_max * sizeof (dtrace_toxrange_t)); 16351 dtrace_toxrange = NULL; 16352 dtrace_toxranges = 0; 16353 dtrace_toxranges_max = 0; 16354 } 16355 16356 ddi_remove_minor_node(dtrace_devi, NULL); 16357 dtrace_devi = NULL; 16358 16359 ddi_soft_state_fini(&dtrace_softstate); 16360 16361 ASSERT(dtrace_vtime_references == 0); 16362 ASSERT(dtrace_opens == 0); 16363 ASSERT(dtrace_retained == NULL); 16364 16365 mutex_exit(&dtrace_lock); 16366 mutex_exit(&dtrace_provider_lock); 16367 16368 /* 16369 * We don't destroy the task queue until after we have dropped our 16370 * locks (taskq_destroy() may block on running tasks). To prevent 16371 * attempting to do work after we have effectively detached but before 16372 * the task queue has been destroyed, all tasks dispatched via the 16373 * task queue must check that DTrace is still attached before 16374 * performing any operation. 16375 */ 16376 taskq_destroy(dtrace_taskq); 16377 dtrace_taskq = NULL; 16378 16379 return (DDI_SUCCESS); 16380} 16381#endif 16382 16383#if defined(sun) 16384/*ARGSUSED*/ 16385static int 16386dtrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) 16387{ 16388 int error; 16389 16390 switch (infocmd) { 16391 case DDI_INFO_DEVT2DEVINFO: 16392 *result = (void *)dtrace_devi; 16393 error = DDI_SUCCESS; 16394 break; 16395 case DDI_INFO_DEVT2INSTANCE: 16396 *result = (void *)0; 16397 error = DDI_SUCCESS; 16398 break; 16399 default: 16400 error = DDI_FAILURE; 16401 } 16402 return (error); 16403} 16404#endif 16405 16406#if defined(sun) 16407static struct cb_ops dtrace_cb_ops = { 16408 dtrace_open, /* open */ 16409 dtrace_close, /* close */ 16410 nulldev, /* strategy */ 16411 nulldev, /* print */ 16412 nodev, /* dump */ 16413 nodev, /* read */ 16414 nodev, /* write */ 16415 dtrace_ioctl, /* ioctl */ 16416 nodev, /* devmap */ 16417 nodev, /* mmap */ 16418 nodev, /* segmap */ 16419 nochpoll, /* poll */ 16420 ddi_prop_op, /* cb_prop_op */ 16421 0, /* streamtab */ 16422 D_NEW | D_MP /* Driver compatibility flag */ 16423}; 16424 16425static struct dev_ops dtrace_ops = { 16426 DEVO_REV, /* devo_rev */ 16427 0, /* refcnt */ 16428 dtrace_info, /* get_dev_info */ 16429 nulldev, /* identify */ 16430 nulldev, /* probe */ 16431 dtrace_attach, /* attach */ 16432 dtrace_detach, /* detach */ 16433 nodev, /* reset */ 16434 &dtrace_cb_ops, /* driver operations */ 16435 NULL, /* bus operations */ 16436 nodev /* dev power */ 16437}; 16438 16439static struct modldrv modldrv = { 16440 &mod_driverops, /* module type (this is a pseudo driver) */ 16441 "Dynamic Tracing", /* name of module */ 16442 &dtrace_ops, /* driver ops */ 16443}; 16444 16445static struct modlinkage modlinkage = { 16446 MODREV_1, 16447 (void *)&modldrv, 16448 NULL 16449}; 16450 16451int 16452_init(void) 16453{ 16454 return (mod_install(&modlinkage)); 16455} 16456 16457int 16458_info(struct modinfo *modinfop) 16459{ 16460 return (mod_info(&modlinkage, modinfop)); 16461} 16462 16463int 16464_fini(void) 16465{ 16466 return (mod_remove(&modlinkage)); 16467} 16468#else 16469 16470static d_ioctl_t dtrace_ioctl; 16471static void dtrace_load(void *); 16472static int dtrace_unload(void); 16473#if __FreeBSD_version < 800039 16474static void dtrace_clone(void *, struct ucred *, char *, int , struct cdev **); 16475static struct clonedevs *dtrace_clones; /* Ptr to the array of cloned devices. */ 16476static eventhandler_tag eh_tag; /* Event handler tag. */ 16477#else 16478static struct cdev *dtrace_dev; 16479#endif 16480 16481void dtrace_invop_init(void); 16482void dtrace_invop_uninit(void); 16483 16484static struct cdevsw dtrace_cdevsw = { 16485 .d_version = D_VERSION, 16486 .d_flags = D_TRACKCLOSE | D_NEEDMINOR, 16487 .d_close = dtrace_close, 16488 .d_ioctl = dtrace_ioctl, 16489 .d_open = dtrace_open, 16490 .d_name = "dtrace", 16491}; 16492 16493#include <dtrace_anon.c> 16494#if __FreeBSD_version < 800039 16495#include <dtrace_clone.c> 16496#endif 16497#include <dtrace_ioctl.c> 16498#include <dtrace_load.c> 16499#include <dtrace_modevent.c> 16500#include <dtrace_sysctl.c> 16501#include <dtrace_unload.c> 16502#include <dtrace_vtime.c> 16503#include <dtrace_hacks.c> 16504#include <dtrace_isa.c> 16505 16506SYSINIT(dtrace_load, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_load, NULL); 16507SYSUNINIT(dtrace_unload, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_unload, NULL); 16508SYSINIT(dtrace_anon_init, SI_SUB_DTRACE_ANON, SI_ORDER_FIRST, dtrace_anon_init, NULL); 16509 16510DEV_MODULE(dtrace, dtrace_modevent, NULL); 16511MODULE_VERSION(dtrace, 1); 16512MODULE_DEPEND(dtrace, cyclic, 1, 1, 1); 16513MODULE_DEPEND(dtrace, opensolaris, 1, 1, 1); 16514#endif 16515