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/* 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26/* #pragma ident "@(#)sdt.c 1.9 08/07/01 SMI" */ 27 28#ifdef KERNEL 29#ifndef _KERNEL 30#define _KERNEL /* Solaris vs. Darwin */ 31#endif 32#endif 33 34#include <sys/param.h> 35#include <sys/systm.h> 36#include <sys/errno.h> 37#include <sys/stat.h> 38#include <sys/ioctl.h> 39#include <sys/conf.h> 40#include <sys/fcntl.h> 41#include <miscfs/devfs/devfs.h> 42 43 44#include <sys/dtrace.h> 45#include <sys/dtrace_impl.h> 46 47#include <sys/dtrace_glue.h> 48 49#include <sys/sdt_impl.h> 50extern int dtrace_kernel_symbol_mode; 51 52struct savearea_t; /* Used anonymously */ 53typedef kern_return_t (*perfCallback)(int, struct savearea_t *, uintptr_t *, int); 54 55#if defined(__x86_64__) 56extern perfCallback tempDTraceTrapHook; 57extern kern_return_t fbt_perfCallback(int, struct savearea_t *, int, int); 58#define SDT_PATCHVAL 0xf0 59#define SDT_AFRAMES 6 60#else 61#error Unknown architecture 62#endif 63 64#define SDT_PROBETAB_SIZE 0x1000 /* 4k entries -- 16K total */ 65 66#define DTRACE_PROBE_PREFIX "_dtrace_probe$" 67 68static dev_info_t *sdt_devi; 69static int sdt_verbose = 0; 70sdt_probe_t **sdt_probetab; 71int sdt_probetab_size; 72int sdt_probetab_mask; 73 74/*ARGSUSED*/ 75static void 76__sdt_provide_module(void *arg, struct modctl *ctl) 77{ 78#pragma unused(arg) 79 struct module *mp = (struct module *)ctl->mod_address; 80 char *modname = ctl->mod_modname; 81 sdt_probedesc_t *sdpd; 82 sdt_probe_t *sdp, *old; 83 sdt_provider_t *prov; 84 int len; 85 86 /* 87 * One for all, and all for one: if we haven't yet registered all of 88 * our providers, we'll refuse to provide anything. 89 */ 90 for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { 91 if (prov->sdtp_id == DTRACE_PROVNONE) 92 return; 93 } 94 95 if (!mp || mp->sdt_nprobes != 0 || (sdpd = mp->sdt_probes) == NULL) 96 return; 97 98 for (sdpd = mp->sdt_probes; sdpd != NULL; sdpd = sdpd->sdpd_next) { 99 const char *name = sdpd->sdpd_name, *func; 100 char *nname; 101 int i, j; 102 dtrace_id_t id; 103 104 for (prov = sdt_providers; prov->sdtp_prefix != NULL; prov++) { 105 const char *prefpart, *prefix = prov->sdtp_prefix; 106 107 if ((prefpart = strstr(name, prefix))) { 108 name = prefpart + strlen(prefix); 109 break; 110 } 111 } 112 113 nname = kmem_alloc(len = strlen(name) + 1, KM_SLEEP); 114 115 for (i = 0, j = 0; name[j] != '\0'; i++) { 116 if (name[j] == '_' && name[j + 1] == '_') { 117 nname[i] = '-'; 118 j += 2; 119 } else { 120 nname[i] = name[j++]; 121 } 122 } 123 124 nname[i] = '\0'; 125 126 sdp = kmem_zalloc(sizeof (sdt_probe_t), KM_SLEEP); 127 sdp->sdp_loadcnt = ctl->mod_loadcnt; 128 sdp->sdp_ctl = ctl; 129 sdp->sdp_name = nname; 130 sdp->sdp_namelen = len; 131 sdp->sdp_provider = prov; 132 133 func = sdpd->sdpd_func; 134 135 if (func == NULL) 136 func = "<unknown>"; 137 138 /* 139 * We have our provider. Now create the probe. 140 */ 141 if ((id = dtrace_probe_lookup(prov->sdtp_id, modname, 142 func, nname)) != DTRACE_IDNONE) { 143 old = dtrace_probe_arg(prov->sdtp_id, id); 144 ASSERT(old != NULL); 145 146 sdp->sdp_next = old->sdp_next; 147 sdp->sdp_id = id; 148 old->sdp_next = sdp; 149 } else { 150 sdp->sdp_id = dtrace_probe_create(prov->sdtp_id, 151 modname, func, nname, SDT_AFRAMES, sdp); 152 153 mp->sdt_nprobes++; 154 } 155 156#if 0 157 printf ("__sdt_provide_module: sdpd=0x%p sdp=0x%p name=%s, id=%d\n", sdpd, sdp, nname, sdp->sdp_id); 158#endif 159 160 sdp->sdp_hashnext = 161 sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)]; 162 sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)] = sdp; 163 164 sdp->sdp_patchval = SDT_PATCHVAL; 165 sdp->sdp_patchpoint = (sdt_instr_t *)sdpd->sdpd_offset; 166 sdp->sdp_savedval = *sdp->sdp_patchpoint; 167 } 168} 169 170/*ARGSUSED*/ 171static void 172sdt_destroy(void *arg, dtrace_id_t id, void *parg) 173{ 174#pragma unused(arg,id) 175 sdt_probe_t *sdp = parg, *old, *last, *hash; 176 int ndx; 177#if !defined(__APPLE__) 178 struct modctl *ctl = sdp->sdp_ctl; 179 180 if (ctl != NULL && ctl->mod_loadcnt == sdp->sdp_loadcnt) { 181 if ((ctl->mod_loadcnt == sdp->sdp_loadcnt && 182 ctl->mod_loaded)) { 183 ((struct module *)(ctl->mod_mp))->sdt_nprobes--; 184 } 185 } 186#endif /* __APPLE__ */ 187 188 while (sdp != NULL) { 189 old = sdp; 190 191 /* 192 * Now we need to remove this probe from the sdt_probetab. 193 */ 194 ndx = SDT_ADDR2NDX(sdp->sdp_patchpoint); 195 last = NULL; 196 hash = sdt_probetab[ndx]; 197 198 while (hash != sdp) { 199 ASSERT(hash != NULL); 200 last = hash; 201 hash = hash->sdp_hashnext; 202 } 203 204 if (last != NULL) { 205 last->sdp_hashnext = sdp->sdp_hashnext; 206 } else { 207 sdt_probetab[ndx] = sdp->sdp_hashnext; 208 } 209 210 kmem_free(sdp->sdp_name, sdp->sdp_namelen); 211 sdp = sdp->sdp_next; 212 kmem_free(old, sizeof (sdt_probe_t)); 213 } 214} 215 216/*ARGSUSED*/ 217static int 218sdt_enable(void *arg, dtrace_id_t id, void *parg) 219{ 220#pragma unused(arg,id) 221 sdt_probe_t *sdp = parg; 222 struct modctl *ctl = sdp->sdp_ctl; 223 224 ctl->mod_nenabled++; 225 226 /* 227 * If this module has disappeared since we discovered its probes, 228 * refuse to enable it. 229 */ 230 if (!ctl->mod_loaded) { 231 if (sdt_verbose) { 232 cmn_err(CE_NOTE, "sdt is failing for probe %s " 233 "(module %s unloaded)", 234 sdp->sdp_name, ctl->mod_modname); 235 } 236 goto err; 237 } 238 239 /* 240 * Now check that our modctl has the expected load count. If it 241 * doesn't, this module must have been unloaded and reloaded -- and 242 * we're not going to touch it. 243 */ 244 if (ctl->mod_loadcnt != sdp->sdp_loadcnt) { 245 if (sdt_verbose) { 246 cmn_err(CE_NOTE, "sdt is failing for probe %s " 247 "(module %s reloaded)", 248 sdp->sdp_name, ctl->mod_modname); 249 } 250 goto err; 251 } 252 253 dtrace_casptr(&tempDTraceTrapHook, NULL, fbt_perfCallback); 254 if (tempDTraceTrapHook != (perfCallback)fbt_perfCallback) { 255 if (sdt_verbose) { 256 cmn_err(CE_NOTE, "sdt_enable is failing for probe %s " 257 "in module %s: tempDTraceTrapHook already occupied.", 258 sdp->sdp_name, ctl->mod_modname); 259 } 260 return (0); 261 } 262 263 while (sdp != NULL) { 264 (void)ml_nofault_copy( (vm_offset_t)&sdp->sdp_patchval, (vm_offset_t)sdp->sdp_patchpoint, 265 (vm_size_t)sizeof(sdp->sdp_patchval)); 266 sdp = sdp->sdp_next; 267 } 268 269err: 270 return (0); 271} 272 273/*ARGSUSED*/ 274static void 275sdt_disable(void *arg, dtrace_id_t id, void *parg) 276{ 277#pragma unused(arg,id) 278 sdt_probe_t *sdp = parg; 279 struct modctl *ctl = sdp->sdp_ctl; 280 281 ctl->mod_nenabled--; 282 283 if (!ctl->mod_loaded || ctl->mod_loadcnt != sdp->sdp_loadcnt) 284 goto err; 285 286 while (sdp != NULL) { 287 (void)ml_nofault_copy( (vm_offset_t)&sdp->sdp_savedval, (vm_offset_t)sdp->sdp_patchpoint, 288 (vm_size_t)sizeof(sdp->sdp_savedval)); 289 sdp = sdp->sdp_next; 290 } 291 292err: 293 ; 294} 295 296static dtrace_pops_t sdt_pops = { 297 NULL, 298 sdt_provide_module, 299 sdt_enable, 300 sdt_disable, 301 NULL, 302 NULL, 303 sdt_getargdesc, 304 sdt_getarg, 305 NULL, 306 sdt_destroy 307}; 308 309/*ARGSUSED*/ 310static int 311sdt_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) 312{ 313#pragma unused(cmd) 314 sdt_provider_t *prov; 315 316 if (ddi_create_minor_node(devi, "sdt", S_IFCHR, 317 0, DDI_PSEUDO, 0) == DDI_FAILURE) { 318 cmn_err(CE_NOTE, "/dev/sdt couldn't create minor node"); 319 ddi_remove_minor_node(devi, NULL); 320 return (DDI_FAILURE); 321 } 322 323 ddi_report_dev(devi); 324 sdt_devi = devi; 325 326 if (sdt_probetab_size == 0) 327 sdt_probetab_size = SDT_PROBETAB_SIZE; 328 329 sdt_probetab_mask = sdt_probetab_size - 1; 330 sdt_probetab = 331 kmem_zalloc(sdt_probetab_size * sizeof (sdt_probe_t *), KM_SLEEP); 332 dtrace_invop_add(sdt_invop); 333 334 for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { 335 if (dtrace_register(prov->sdtp_name, prov->sdtp_attr, 336 DTRACE_PRIV_KERNEL, NULL, 337 &sdt_pops, prov, &prov->sdtp_id) != 0) { 338 cmn_err(CE_WARN, "failed to register sdt provider %s", 339 prov->sdtp_name); 340 } 341 } 342 343 return (DDI_SUCCESS); 344} 345 346#if !defined(__APPLE__) 347/*ARGSUSED*/ 348static int 349sdt_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) 350{ 351 sdt_provider_t *prov; 352 353 switch (cmd) { 354 case DDI_DETACH: 355 break; 356 357 case DDI_SUSPEND: 358 return (DDI_SUCCESS); 359 360 default: 361 return (DDI_FAILURE); 362 } 363 364 for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { 365 if (prov->sdtp_id != DTRACE_PROVNONE) { 366 if (dtrace_unregister(prov->sdtp_id) != 0) 367 return (DDI_FAILURE); 368 369 prov->sdtp_id = DTRACE_PROVNONE; 370 } 371 } 372 373 dtrace_invop_remove(sdt_invop); 374 kmem_free(sdt_probetab, sdt_probetab_size * sizeof (sdt_probe_t *)); 375 376 return (DDI_SUCCESS); 377} 378 379/*ARGSUSED*/ 380static int 381sdt_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) 382{ 383 int error; 384 385 switch (infocmd) { 386 case DDI_INFO_DEVT2DEVINFO: 387 *result = (void *)sdt_devi; 388 error = DDI_SUCCESS; 389 break; 390 case DDI_INFO_DEVT2INSTANCE: 391 *result = (void *)0; 392 error = DDI_SUCCESS; 393 break; 394 default: 395 error = DDI_FAILURE; 396 } 397 return (error); 398} 399 400/*ARGSUSED*/ 401static int 402sdt_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) 403{ 404 return (0); 405} 406 407static struct cb_ops sdt_cb_ops = { 408 sdt_open, /* open */ 409 nodev, /* close */ 410 nulldev, /* strategy */ 411 nulldev, /* print */ 412 nodev, /* dump */ 413 nodev, /* read */ 414 nodev, /* write */ 415 nodev, /* ioctl */ 416 nodev, /* devmap */ 417 nodev, /* mmap */ 418 nodev, /* segmap */ 419 nochpoll, /* poll */ 420 ddi_prop_op, /* cb_prop_op */ 421 0, /* streamtab */ 422 D_NEW | D_MP /* Driver compatibility flag */ 423}; 424 425static struct dev_ops sdt_ops = { 426 DEVO_REV, /* devo_rev, */ 427 0, /* refcnt */ 428 sdt_info, /* get_dev_info */ 429 nulldev, /* identify */ 430 nulldev, /* probe */ 431 sdt_attach, /* attach */ 432 sdt_detach, /* detach */ 433 nodev, /* reset */ 434 &sdt_cb_ops, /* driver operations */ 435 NULL, /* bus operations */ 436 nodev /* dev power */ 437}; 438 439/* 440 * Module linkage information for the kernel. 441 */ 442static struct modldrv modldrv = { 443 &mod_driverops, /* module type (this is a pseudo driver) */ 444 "Statically Defined Tracing", /* name of module */ 445 &sdt_ops, /* driver ops */ 446}; 447 448static struct modlinkage modlinkage = { 449 MODREV_1, 450 (void *)&modldrv, 451 NULL 452}; 453 454int 455_init(void) 456{ 457 return (mod_install(&modlinkage)); 458} 459 460int 461_info(struct modinfo *modinfop) 462{ 463 return (mod_info(&modlinkage, modinfop)); 464} 465 466int 467_fini(void) 468{ 469 return (mod_remove(&modlinkage)); 470} 471#else 472d_open_t _sdt_open; 473 474int _sdt_open(dev_t dev, int flags, int devtype, struct proc *p) 475{ 476#pragma unused(dev,flags,devtype,p) 477 return 0; 478} 479 480#define SDT_MAJOR -24 /* let the kernel pick the device number */ 481 482/* 483 * A struct describing which functions will get invoked for certain 484 * actions. 485 */ 486static struct cdevsw sdt_cdevsw = 487{ 488 _sdt_open, /* open */ 489 eno_opcl, /* close */ 490 eno_rdwrt, /* read */ 491 eno_rdwrt, /* write */ 492 eno_ioctl, /* ioctl */ 493 (stop_fcn_t *)nulldev, /* stop */ 494 (reset_fcn_t *)nulldev, /* reset */ 495 NULL, /* tty's */ 496 eno_select, /* select */ 497 eno_mmap, /* mmap */ 498 eno_strat, /* strategy */ 499 eno_getc, /* getc */ 500 eno_putc, /* putc */ 501 0 /* type */ 502}; 503 504static int gSDTInited = 0; 505static struct modctl g_sdt_kernctl; 506static struct module g_sdt_mach_module; 507 508#include <mach-o/nlist.h> 509#include <libkern/kernel_mach_header.h> 510 511void sdt_init( void ) 512{ 513 if (0 == gSDTInited) 514 { 515 int majdevno = cdevsw_add(SDT_MAJOR, &sdt_cdevsw); 516 517 if (majdevno < 0) { 518 printf("sdt_init: failed to allocate a major number!\n"); 519 gSDTInited = 0; 520 return; 521 } 522 523 if (MH_MAGIC_KERNEL != _mh_execute_header.magic) { 524 g_sdt_kernctl.mod_address = (vm_address_t)NULL; 525 g_sdt_kernctl.mod_size = 0; 526 } else { 527 kernel_mach_header_t *mh; 528 struct load_command *cmd; 529 kernel_segment_command_t *orig_ts = NULL, *orig_le = NULL; 530 struct symtab_command *orig_st = NULL; 531 kernel_nlist_t *sym = NULL; 532 char *strings; 533 unsigned int i; 534 535 g_sdt_mach_module.sdt_nprobes = 0; 536 g_sdt_mach_module.sdt_probes = NULL; 537 538 g_sdt_kernctl.mod_address = (vm_address_t)&g_sdt_mach_module; 539 g_sdt_kernctl.mod_size = 0; 540 strncpy((char *)&(g_sdt_kernctl.mod_modname), "mach_kernel", KMOD_MAX_NAME); 541 542 g_sdt_kernctl.mod_next = NULL; 543 g_sdt_kernctl.mod_stale = NULL; 544 g_sdt_kernctl.mod_id = 0; 545 g_sdt_kernctl.mod_loadcnt = 1; 546 g_sdt_kernctl.mod_loaded = 1; 547 g_sdt_kernctl.mod_flags = 0; 548 g_sdt_kernctl.mod_nenabled = 0; 549 550 mh = &_mh_execute_header; 551 cmd = (struct load_command*) &mh[1]; 552 for (i = 0; i < mh->ncmds; i++) { 553 if (cmd->cmd == LC_SEGMENT_KERNEL) { 554 kernel_segment_command_t *orig_sg = (kernel_segment_command_t *) cmd; 555 556 if (LIT_STRNEQL(orig_sg->segname, SEG_TEXT)) 557 orig_ts = orig_sg; 558 else if (LIT_STRNEQL(orig_sg->segname, SEG_LINKEDIT)) 559 orig_le = orig_sg; 560 else if (LIT_STRNEQL(orig_sg->segname, "")) 561 orig_ts = orig_sg; /* kexts have a single unnamed segment */ 562 } 563 else if (cmd->cmd == LC_SYMTAB) 564 orig_st = (struct symtab_command *) cmd; 565 566 cmd = (struct load_command *) ((uintptr_t) cmd + cmd->cmdsize); 567 } 568 569 if ((orig_ts == NULL) || (orig_st == NULL) || (orig_le == NULL)) 570 return; 571 572 sym = (kernel_nlist_t *)(orig_le->vmaddr + orig_st->symoff - orig_le->fileoff); 573 strings = (char *)(orig_le->vmaddr + orig_st->stroff - orig_le->fileoff); 574 575 for (i = 0; i < orig_st->nsyms; i++) { 576 uint8_t n_type = sym[i].n_type & (N_TYPE | N_EXT); 577 char *name = strings + sym[i].n_un.n_strx; 578 const char *prev_name; 579 unsigned long best; 580 unsigned int j; 581 582 /* Check that the symbol is a global and that it has a name. */ 583 if (((N_SECT | N_EXT) != n_type && (N_ABS | N_EXT) != n_type)) 584 continue; 585 586 if (0 == sym[i].n_un.n_strx) /* iff a null, "", name. */ 587 continue; 588 589 /* Lop off omnipresent leading underscore. */ 590 if (*name == '_') 591 name += 1; 592 593 if (strncmp(name, DTRACE_PROBE_PREFIX, sizeof(DTRACE_PROBE_PREFIX) - 1) == 0) { 594 sdt_probedesc_t *sdpd = kmem_alloc(sizeof(sdt_probedesc_t), KM_SLEEP); 595 int len = strlen(name) + 1; 596 597 sdpd->sdpd_name = kmem_alloc(len, KM_SLEEP); 598 strncpy(sdpd->sdpd_name, name, len); /* NUL termination is ensured. */ 599 600 prev_name = "<unknown>"; 601 best = 0; 602 603 /* 604 * Find the symbol immediately preceding the sdt probe site just discovered, 605 * that symbol names the function containing the sdt probe. 606 */ 607 for (j = 0; j < orig_st->nsyms; j++) { 608 uint8_t jn_type = sym[j].n_type & (N_TYPE | N_EXT); 609 char *jname = strings + sym[j].n_un.n_strx; 610 611 if (((N_SECT | N_EXT) != jn_type && (N_ABS | N_EXT) != jn_type)) 612 continue; 613 614 if (0 == sym[j].n_un.n_strx) /* iff a null, "", name. */ 615 continue; 616 617 if (*jname == '_') 618 jname += 1; 619 620 if (*(unsigned long *)sym[i].n_value <= (unsigned long)sym[j].n_value) 621 continue; 622 623 if ((unsigned long)sym[j].n_value > best) { 624 best = (unsigned long)sym[j].n_value; 625 prev_name = jname; 626 } 627 } 628 629 sdpd->sdpd_func = kmem_alloc((len = strlen(prev_name) + 1), KM_SLEEP); 630 strncpy(sdpd->sdpd_func, prev_name, len); /* NUL termination is ensured. */ 631 632 sdpd->sdpd_offset = *(unsigned long *)sym[i].n_value; 633 634#if 0 635 printf("sdt_init: sdpd_offset=0x%lx, n_value=0x%lx, name=%s\n", 636 sdpd->sdpd_offset, *(unsigned long *)sym[i].n_value, name); 637#endif 638 639 sdpd->sdpd_next = g_sdt_mach_module.sdt_probes; 640 g_sdt_mach_module.sdt_probes = sdpd; 641 } else { 642 prev_name = name; 643 } 644 } 645 } 646 647 sdt_attach( (dev_info_t *)(uintptr_t)majdevno, DDI_ATTACH ); 648 649 gSDTInited = 1; 650 } else 651 panic("sdt_init: called twice!\n"); 652} 653 654#undef SDT_MAJOR 655 656/*ARGSUSED*/ 657void 658sdt_provide_module(void *arg, struct modctl *ctl) 659{ 660#pragma unused(arg) 661 ASSERT(ctl != NULL); 662 ASSERT(dtrace_kernel_symbol_mode != DTRACE_KERNEL_SYMBOLS_NEVER); 663 lck_mtx_assert(&mod_lock, LCK_MTX_ASSERT_OWNED); 664 665 if (MOD_SDT_DONE(ctl)) 666 return; 667 668 if (MOD_IS_MACH_KERNEL(ctl)) { 669 __sdt_provide_module(arg, &g_sdt_kernctl); 670 671 sdt_probedesc_t *sdpd = g_sdt_mach_module.sdt_probes; 672 while (sdpd) { 673 sdt_probedesc_t *this_sdpd = sdpd; 674 kmem_free((void *)sdpd->sdpd_name, strlen(sdpd->sdpd_name) + 1); 675 kmem_free((void *)sdpd->sdpd_func, strlen(sdpd->sdpd_func) + 1); 676 sdpd = sdpd->sdpd_next; 677 kmem_free((void *)this_sdpd, sizeof(sdt_probedesc_t)); 678 } 679 g_sdt_mach_module.sdt_probes = NULL; 680 } else { 681 /* FIXME -- sdt in kext not yet supported */ 682 } 683 684 /* Need to mark this module as completed */ 685 ctl->mod_flags |= MODCTL_SDT_PROBES_PROVIDED; 686} 687 688#endif /* __APPLE__ */ 689