kern_ktrace.c revision 147183
1/*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)kern_ktrace.c 8.2 (Berkeley) 9/23/93 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: head/sys/kern/kern_ktrace.c 147183 2005-06-09 18:33:21Z pjd $"); 34 35#include "opt_ktrace.h" 36#include "opt_mac.h" 37 38#include <sys/param.h> 39#include <sys/systm.h> 40#include <sys/fcntl.h> 41#include <sys/kernel.h> 42#include <sys/kthread.h> 43#include <sys/lock.h> 44#include <sys/mutex.h> 45#include <sys/mac.h> 46#include <sys/malloc.h> 47#include <sys/namei.h> 48#include <sys/proc.h> 49#include <sys/unistd.h> 50#include <sys/vnode.h> 51#include <sys/ktrace.h> 52#include <sys/sx.h> 53#include <sys/sysctl.h> 54#include <sys/syslog.h> 55#include <sys/sysproto.h> 56 57static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE"); 58 59#ifdef KTRACE 60 61#ifndef KTRACE_REQUEST_POOL 62#define KTRACE_REQUEST_POOL 100 63#endif 64 65struct ktr_request { 66 struct ktr_header ktr_header; 67 struct ucred *ktr_cred; 68 struct vnode *ktr_vp; 69 union { 70 struct ktr_syscall ktr_syscall; 71 struct ktr_sysret ktr_sysret; 72 struct ktr_genio ktr_genio; 73 struct ktr_psig ktr_psig; 74 struct ktr_csw ktr_csw; 75 } ktr_data; 76 STAILQ_ENTRY(ktr_request) ktr_list; 77}; 78 79static int data_lengths[] = { 80 0, /* none */ 81 offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */ 82 sizeof(struct ktr_sysret), /* KTR_SYSRET */ 83 0, /* KTR_NAMEI */ 84 sizeof(struct ktr_genio), /* KTR_GENIO */ 85 sizeof(struct ktr_psig), /* KTR_PSIG */ 86 sizeof(struct ktr_csw), /* KTR_CSW */ 87 0 /* KTR_USER */ 88}; 89 90static STAILQ_HEAD(, ktr_request) ktr_todo; 91static STAILQ_HEAD(, ktr_request) ktr_free; 92 93static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options"); 94 95static u_int ktr_requestpool = KTRACE_REQUEST_POOL; 96TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool); 97 98static u_int ktr_geniosize = PAGE_SIZE; 99TUNABLE_INT("kern.ktrace.genio_size", &ktr_geniosize); 100SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RW, &ktr_geniosize, 101 0, "Maximum size of genio event payload"); 102 103static int print_message = 1; 104struct mtx ktrace_mtx; 105static struct cv ktrace_cv; 106 107static void ktrace_init(void *dummy); 108static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS); 109static u_int ktrace_resize_pool(u_int newsize); 110static struct ktr_request *ktr_getrequest(int type); 111static void ktr_submitrequest(struct ktr_request *req); 112static void ktr_freerequest(struct ktr_request *req); 113static void ktr_loop(void *dummy); 114static void ktr_writerequest(struct ktr_request *req); 115static int ktrcanset(struct thread *,struct proc *); 116static int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *); 117static int ktrops(struct thread *,struct proc *,int,int,struct vnode *); 118 119static void 120ktrace_init(void *dummy) 121{ 122 struct ktr_request *req; 123 int i; 124 125 mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET); 126 cv_init(&ktrace_cv, "ktrace"); 127 STAILQ_INIT(&ktr_todo); 128 STAILQ_INIT(&ktr_free); 129 for (i = 0; i < ktr_requestpool; i++) { 130 req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK); 131 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 132 } 133 kthread_create(ktr_loop, NULL, NULL, RFHIGHPID, 0, "ktrace"); 134} 135SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL); 136 137static int 138sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS) 139{ 140 struct thread *td; 141 u_int newsize, oldsize, wantsize; 142 int error; 143 144 /* Handle easy read-only case first to avoid warnings from GCC. */ 145 if (!req->newptr) { 146 mtx_lock(&ktrace_mtx); 147 oldsize = ktr_requestpool; 148 mtx_unlock(&ktrace_mtx); 149 return (SYSCTL_OUT(req, &oldsize, sizeof(u_int))); 150 } 151 152 error = SYSCTL_IN(req, &wantsize, sizeof(u_int)); 153 if (error) 154 return (error); 155 td = curthread; 156 td->td_pflags |= TDP_INKTRACE; 157 mtx_lock(&ktrace_mtx); 158 oldsize = ktr_requestpool; 159 newsize = ktrace_resize_pool(wantsize); 160 mtx_unlock(&ktrace_mtx); 161 td->td_pflags &= ~TDP_INKTRACE; 162 error = SYSCTL_OUT(req, &oldsize, sizeof(u_int)); 163 if (error) 164 return (error); 165 if (wantsize > oldsize && newsize < wantsize) 166 return (ENOSPC); 167 return (0); 168} 169SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW, 170 &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU", ""); 171 172static u_int 173ktrace_resize_pool(u_int newsize) 174{ 175 struct ktr_request *req; 176 int bound; 177 178 mtx_assert(&ktrace_mtx, MA_OWNED); 179 print_message = 1; 180 bound = newsize - ktr_requestpool; 181 if (bound == 0) 182 return (ktr_requestpool); 183 if (bound < 0) 184 /* Shrink pool down to newsize if possible. */ 185 while (bound++ < 0) { 186 req = STAILQ_FIRST(&ktr_free); 187 if (req == NULL) 188 return (ktr_requestpool); 189 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 190 ktr_requestpool--; 191 mtx_unlock(&ktrace_mtx); 192 free(req, M_KTRACE); 193 mtx_lock(&ktrace_mtx); 194 } 195 else 196 /* Grow pool up to newsize. */ 197 while (bound-- > 0) { 198 mtx_unlock(&ktrace_mtx); 199 req = malloc(sizeof(struct ktr_request), M_KTRACE, 200 M_WAITOK); 201 mtx_lock(&ktrace_mtx); 202 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 203 ktr_requestpool++; 204 } 205 return (ktr_requestpool); 206} 207 208static struct ktr_request * 209ktr_getrequest(int type) 210{ 211 struct ktr_request *req; 212 struct thread *td = curthread; 213 struct proc *p = td->td_proc; 214 int pm; 215 216 td->td_pflags |= TDP_INKTRACE; 217 mtx_lock(&ktrace_mtx); 218 if (!KTRCHECK(td, type)) { 219 mtx_unlock(&ktrace_mtx); 220 td->td_pflags &= ~TDP_INKTRACE; 221 return (NULL); 222 } 223 req = STAILQ_FIRST(&ktr_free); 224 if (req != NULL) { 225 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 226 req->ktr_header.ktr_type = type; 227 if (p->p_traceflag & KTRFAC_DROP) { 228 req->ktr_header.ktr_type |= KTR_DROP; 229 p->p_traceflag &= ~KTRFAC_DROP; 230 } 231 KASSERT(p->p_tracevp != NULL, ("ktrace: no trace vnode")); 232 KASSERT(p->p_tracecred != NULL, ("ktrace: no trace cred")); 233 req->ktr_vp = p->p_tracevp; 234 VREF(p->p_tracevp); 235 req->ktr_cred = crhold(p->p_tracecred); 236 mtx_unlock(&ktrace_mtx); 237 microtime(&req->ktr_header.ktr_time); 238 req->ktr_header.ktr_pid = p->p_pid; 239 bcopy(p->p_comm, req->ktr_header.ktr_comm, MAXCOMLEN + 1); 240 req->ktr_header.ktr_buffer = NULL; 241 req->ktr_header.ktr_len = 0; 242 } else { 243 p->p_traceflag |= KTRFAC_DROP; 244 pm = print_message; 245 print_message = 0; 246 mtx_unlock(&ktrace_mtx); 247 if (pm) 248 printf("Out of ktrace request objects.\n"); 249 td->td_pflags &= ~TDP_INKTRACE; 250 } 251 return (req); 252} 253 254static void 255ktr_submitrequest(struct ktr_request *req) 256{ 257 258 mtx_lock(&ktrace_mtx); 259 STAILQ_INSERT_TAIL(&ktr_todo, req, ktr_list); 260 cv_signal(&ktrace_cv); 261 mtx_unlock(&ktrace_mtx); 262 curthread->td_pflags &= ~TDP_INKTRACE; 263} 264 265static void 266ktr_freerequest(struct ktr_request *req) 267{ 268 269 crfree(req->ktr_cred); 270 if (req->ktr_vp != NULL) { 271 mtx_lock(&Giant); 272 vrele(req->ktr_vp); 273 mtx_unlock(&Giant); 274 } 275 if (req->ktr_header.ktr_buffer != NULL) 276 free(req->ktr_header.ktr_buffer, M_KTRACE); 277 mtx_lock(&ktrace_mtx); 278 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 279 mtx_unlock(&ktrace_mtx); 280} 281 282static void 283ktr_loop(void *dummy) 284{ 285 struct ktr_request *req; 286 struct thread *td; 287 struct ucred *cred; 288 289 /* Only cache these values once. */ 290 td = curthread; 291 cred = td->td_ucred; 292 for (;;) { 293 mtx_lock(&ktrace_mtx); 294 while (STAILQ_EMPTY(&ktr_todo)) 295 cv_wait(&ktrace_cv, &ktrace_mtx); 296 req = STAILQ_FIRST(&ktr_todo); 297 STAILQ_REMOVE_HEAD(&ktr_todo, ktr_list); 298 KASSERT(req != NULL, ("got a NULL request")); 299 mtx_unlock(&ktrace_mtx); 300 /* 301 * It is not enough just to pass the cached cred 302 * to the VOP's in ktr_writerequest(). Some VFS 303 * operations use curthread->td_ucred, so we need 304 * to modify our thread's credentials as well. 305 * Evil. 306 */ 307 td->td_ucred = req->ktr_cred; 308 ktr_writerequest(req); 309 td->td_ucred = cred; 310 ktr_freerequest(req); 311 } 312} 313 314/* 315 * MPSAFE 316 */ 317void 318ktrsyscall(code, narg, args) 319 int code, narg; 320 register_t args[]; 321{ 322 struct ktr_request *req; 323 struct ktr_syscall *ktp; 324 size_t buflen; 325 char *buf = NULL; 326 327 buflen = sizeof(register_t) * narg; 328 if (buflen > 0) { 329 buf = malloc(buflen, M_KTRACE, M_WAITOK); 330 bcopy(args, buf, buflen); 331 } 332 req = ktr_getrequest(KTR_SYSCALL); 333 if (req == NULL) { 334 if (buf != NULL) 335 free(buf, M_KTRACE); 336 return; 337 } 338 ktp = &req->ktr_data.ktr_syscall; 339 ktp->ktr_code = code; 340 ktp->ktr_narg = narg; 341 if (buflen > 0) { 342 req->ktr_header.ktr_len = buflen; 343 req->ktr_header.ktr_buffer = buf; 344 } 345 ktr_submitrequest(req); 346} 347 348/* 349 * MPSAFE 350 */ 351void 352ktrsysret(code, error, retval) 353 int code, error; 354 register_t retval; 355{ 356 struct ktr_request *req; 357 struct ktr_sysret *ktp; 358 359 req = ktr_getrequest(KTR_SYSRET); 360 if (req == NULL) 361 return; 362 ktp = &req->ktr_data.ktr_sysret; 363 ktp->ktr_code = code; 364 ktp->ktr_error = error; 365 ktp->ktr_retval = retval; /* what about val2 ? */ 366 ktr_submitrequest(req); 367} 368 369void 370ktrnamei(path) 371 char *path; 372{ 373 struct ktr_request *req; 374 int namelen; 375 char *buf = NULL; 376 377 namelen = strlen(path); 378 if (namelen > 0) { 379 buf = malloc(namelen, M_KTRACE, M_WAITOK); 380 bcopy(path, buf, namelen); 381 } 382 req = ktr_getrequest(KTR_NAMEI); 383 if (req == NULL) { 384 if (buf != NULL) 385 free(buf, M_KTRACE); 386 return; 387 } 388 if (namelen > 0) { 389 req->ktr_header.ktr_len = namelen; 390 req->ktr_header.ktr_buffer = buf; 391 } 392 ktr_submitrequest(req); 393} 394 395/* 396 * Since the uio may not stay valid, we can not hand off this request to 397 * the thread and need to process it synchronously. However, we wish to 398 * keep the relative order of records in a trace file correct, so we 399 * do put this request on the queue (if it isn't empty) and then block. 400 * The ktrace thread waks us back up when it is time for this event to 401 * be posted and blocks until we have completed writing out the event 402 * and woken it back up. 403 */ 404void 405ktrgenio(fd, rw, uio, error) 406 int fd; 407 enum uio_rw rw; 408 struct uio *uio; 409 int error; 410{ 411 struct ktr_request *req; 412 struct ktr_genio *ktg; 413 int datalen; 414 char *buf; 415 416 if (error) { 417 free(uio, M_IOV); 418 return; 419 } 420 uio->uio_offset = 0; 421 uio->uio_rw = UIO_WRITE; 422 datalen = imin(uio->uio_resid, ktr_geniosize); 423 buf = malloc(datalen, M_KTRACE, M_WAITOK); 424 error = uiomove(buf, datalen, uio); 425 free(uio, M_IOV); 426 if (error) { 427 free(buf, M_KTRACE); 428 return; 429 } 430 req = ktr_getrequest(KTR_GENIO); 431 if (req == NULL) { 432 free(buf, M_KTRACE); 433 return; 434 } 435 ktg = &req->ktr_data.ktr_genio; 436 ktg->ktr_fd = fd; 437 ktg->ktr_rw = rw; 438 req->ktr_header.ktr_len = datalen; 439 req->ktr_header.ktr_buffer = buf; 440 ktr_submitrequest(req); 441} 442 443void 444ktrpsig(sig, action, mask, code) 445 int sig; 446 sig_t action; 447 sigset_t *mask; 448 int code; 449{ 450 struct ktr_request *req; 451 struct ktr_psig *kp; 452 453 req = ktr_getrequest(KTR_PSIG); 454 if (req == NULL) 455 return; 456 kp = &req->ktr_data.ktr_psig; 457 kp->signo = (char)sig; 458 kp->action = action; 459 kp->mask = *mask; 460 kp->code = code; 461 ktr_submitrequest(req); 462} 463 464void 465ktrcsw(out, user) 466 int out, user; 467{ 468 struct ktr_request *req; 469 struct ktr_csw *kc; 470 471 req = ktr_getrequest(KTR_CSW); 472 if (req == NULL) 473 return; 474 kc = &req->ktr_data.ktr_csw; 475 kc->out = out; 476 kc->user = user; 477 ktr_submitrequest(req); 478} 479#endif /* KTRACE */ 480 481/* Interface and common routines */ 482 483/* 484 * ktrace system call 485 * 486 * MPSAFE 487 */ 488#ifndef _SYS_SYSPROTO_H_ 489struct ktrace_args { 490 char *fname; 491 int ops; 492 int facs; 493 int pid; 494}; 495#endif 496/* ARGSUSED */ 497int 498ktrace(td, uap) 499 struct thread *td; 500 register struct ktrace_args *uap; 501{ 502#ifdef KTRACE 503 register struct vnode *vp = NULL; 504 register struct proc *p; 505 struct pgrp *pg; 506 int facs = uap->facs & ~KTRFAC_ROOT; 507 int ops = KTROP(uap->ops); 508 int descend = uap->ops & KTRFLAG_DESCEND; 509 int ret = 0; 510 int flags, error = 0; 511 struct nameidata nd; 512 struct ucred *cred; 513 514 /* 515 * Need something to (un)trace. 516 */ 517 if (ops != KTROP_CLEARFILE && facs == 0) 518 return (EINVAL); 519 520 td->td_pflags |= TDP_INKTRACE; 521 if (ops != KTROP_CLEAR) { 522 /* 523 * an operation which requires a file argument. 524 */ 525 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname, td); 526 flags = FREAD | FWRITE | O_NOFOLLOW; 527 mtx_lock(&Giant); 528 error = vn_open(&nd, &flags, 0, -1); 529 if (error) { 530 mtx_unlock(&Giant); 531 td->td_pflags &= ~TDP_INKTRACE; 532 return (error); 533 } 534 NDFREE(&nd, NDF_ONLY_PNBUF); 535 vp = nd.ni_vp; 536 VOP_UNLOCK(vp, 0, td); 537 if (vp->v_type != VREG) { 538 (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td); 539 mtx_unlock(&Giant); 540 td->td_pflags &= ~TDP_INKTRACE; 541 return (EACCES); 542 } 543 mtx_unlock(&Giant); 544 } 545 /* 546 * Clear all uses of the tracefile. 547 */ 548 if (ops == KTROP_CLEARFILE) { 549 sx_slock(&allproc_lock); 550 LIST_FOREACH(p, &allproc, p_list) { 551 PROC_LOCK(p); 552 if (p->p_tracevp == vp) { 553 if (ktrcanset(td, p)) { 554 mtx_lock(&ktrace_mtx); 555 cred = p->p_tracecred; 556 p->p_tracecred = NULL; 557 p->p_tracevp = NULL; 558 p->p_traceflag = 0; 559 mtx_unlock(&ktrace_mtx); 560 PROC_UNLOCK(p); 561 mtx_lock(&Giant); 562 (void) vn_close(vp, FREAD|FWRITE, 563 cred, td); 564 mtx_unlock(&Giant); 565 crfree(cred); 566 } else { 567 PROC_UNLOCK(p); 568 error = EPERM; 569 } 570 } else 571 PROC_UNLOCK(p); 572 } 573 sx_sunlock(&allproc_lock); 574 goto done; 575 } 576 /* 577 * do it 578 */ 579 sx_slock(&proctree_lock); 580 if (uap->pid < 0) { 581 /* 582 * by process group 583 */ 584 pg = pgfind(-uap->pid); 585 if (pg == NULL) { 586 sx_sunlock(&proctree_lock); 587 error = ESRCH; 588 goto done; 589 } 590 /* 591 * ktrops() may call vrele(). Lock pg_members 592 * by the proctree_lock rather than pg_mtx. 593 */ 594 PGRP_UNLOCK(pg); 595 LIST_FOREACH(p, &pg->pg_members, p_pglist) 596 if (descend) 597 ret |= ktrsetchildren(td, p, ops, facs, vp); 598 else 599 ret |= ktrops(td, p, ops, facs, vp); 600 } else { 601 /* 602 * by pid 603 */ 604 p = pfind(uap->pid); 605 if (p == NULL) { 606 sx_sunlock(&proctree_lock); 607 error = ESRCH; 608 goto done; 609 } 610 error = p_cansee(td, p); 611 /* 612 * The slock of the proctree lock will keep this process 613 * from going away, so unlocking the proc here is ok. 614 */ 615 PROC_UNLOCK(p); 616 if (error) 617 goto done; 618 if (descend) 619 ret |= ktrsetchildren(td, p, ops, facs, vp); 620 else 621 ret |= ktrops(td, p, ops, facs, vp); 622 } 623 sx_sunlock(&proctree_lock); 624 if (!ret) 625 error = EPERM; 626done: 627 if (vp != NULL) { 628 mtx_lock(&Giant); 629 (void) vn_close(vp, FWRITE, td->td_ucred, td); 630 mtx_unlock(&Giant); 631 } 632 td->td_pflags &= ~TDP_INKTRACE; 633 return (error); 634#else /* !KTRACE */ 635 return (ENOSYS); 636#endif /* KTRACE */ 637} 638 639/* 640 * utrace system call 641 * 642 * MPSAFE 643 */ 644/* ARGSUSED */ 645int 646utrace(td, uap) 647 struct thread *td; 648 register struct utrace_args *uap; 649{ 650 651#ifdef KTRACE 652 struct ktr_request *req; 653 void *cp; 654 int error; 655 656 if (!KTRPOINT(td, KTR_USER)) 657 return (0); 658 if (uap->len > KTR_USER_MAXLEN) 659 return (EINVAL); 660 cp = malloc(uap->len, M_KTRACE, M_WAITOK); 661 error = copyin(uap->addr, cp, uap->len); 662 if (error) { 663 free(cp, M_KTRACE); 664 return (error); 665 } 666 req = ktr_getrequest(KTR_USER); 667 if (req == NULL) { 668 free(cp, M_KTRACE); 669 return (ENOMEM); 670 } 671 req->ktr_header.ktr_buffer = cp; 672 req->ktr_header.ktr_len = uap->len; 673 ktr_submitrequest(req); 674 return (0); 675#else /* !KTRACE */ 676 return (ENOSYS); 677#endif /* KTRACE */ 678} 679 680#ifdef KTRACE 681static int 682ktrops(td, p, ops, facs, vp) 683 struct thread *td; 684 struct proc *p; 685 int ops, facs; 686 struct vnode *vp; 687{ 688 struct vnode *tracevp = NULL; 689 struct ucred *tracecred = NULL; 690 691 PROC_LOCK(p); 692 if (!ktrcanset(td, p)) { 693 PROC_UNLOCK(p); 694 return (0); 695 } 696 mtx_lock(&ktrace_mtx); 697 if (ops == KTROP_SET) { 698 if (p->p_tracevp != vp) { 699 /* 700 * if trace file already in use, relinquish below 701 */ 702 tracevp = p->p_tracevp; 703 VREF(vp); 704 p->p_tracevp = vp; 705 } 706 if (p->p_tracecred != td->td_ucred) { 707 tracecred = p->p_tracecred; 708 p->p_tracecred = crhold(td->td_ucred); 709 } 710 p->p_traceflag |= facs; 711 if (td->td_ucred->cr_uid == 0) 712 p->p_traceflag |= KTRFAC_ROOT; 713 } else { 714 /* KTROP_CLEAR */ 715 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) { 716 /* no more tracing */ 717 p->p_traceflag = 0; 718 tracevp = p->p_tracevp; 719 p->p_tracevp = NULL; 720 tracecred = p->p_tracecred; 721 p->p_tracecred = NULL; 722 } 723 } 724 mtx_unlock(&ktrace_mtx); 725 PROC_UNLOCK(p); 726 if (tracevp != NULL) { 727 mtx_lock(&Giant); 728 vrele(tracevp); 729 mtx_unlock(&Giant); 730 } 731 if (tracecred != NULL) 732 crfree(tracecred); 733 734 return (1); 735} 736 737static int 738ktrsetchildren(td, top, ops, facs, vp) 739 struct thread *td; 740 struct proc *top; 741 int ops, facs; 742 struct vnode *vp; 743{ 744 register struct proc *p; 745 register int ret = 0; 746 747 p = top; 748 sx_assert(&proctree_lock, SX_LOCKED); 749 for (;;) { 750 ret |= ktrops(td, p, ops, facs, vp); 751 /* 752 * If this process has children, descend to them next, 753 * otherwise do any siblings, and if done with this level, 754 * follow back up the tree (but not past top). 755 */ 756 if (!LIST_EMPTY(&p->p_children)) 757 p = LIST_FIRST(&p->p_children); 758 else for (;;) { 759 if (p == top) 760 return (ret); 761 if (LIST_NEXT(p, p_sibling)) { 762 p = LIST_NEXT(p, p_sibling); 763 break; 764 } 765 p = p->p_pptr; 766 } 767 } 768 /*NOTREACHED*/ 769} 770 771static void 772ktr_writerequest(struct ktr_request *req) 773{ 774 struct ktr_header *kth; 775 struct vnode *vp; 776 struct proc *p; 777 struct thread *td; 778 struct ucred *cred; 779 struct uio auio; 780 struct iovec aiov[3]; 781 struct mount *mp; 782 int datalen, buflen, vrele_count; 783 int error; 784 785 vp = req->ktr_vp; 786 /* 787 * If vp is NULL, the vp has been cleared out from under this 788 * request, so just drop it. 789 */ 790 if (vp == NULL) 791 return; 792 kth = &req->ktr_header; 793 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP]; 794 buflen = kth->ktr_len; 795 cred = req->ktr_cred; 796 td = curthread; 797 auio.uio_iov = &aiov[0]; 798 auio.uio_offset = 0; 799 auio.uio_segflg = UIO_SYSSPACE; 800 auio.uio_rw = UIO_WRITE; 801 aiov[0].iov_base = (caddr_t)kth; 802 aiov[0].iov_len = sizeof(struct ktr_header); 803 auio.uio_resid = sizeof(struct ktr_header); 804 auio.uio_iovcnt = 1; 805 auio.uio_td = td; 806 if (datalen != 0) { 807 aiov[1].iov_base = (caddr_t)&req->ktr_data; 808 aiov[1].iov_len = datalen; 809 auio.uio_resid += datalen; 810 auio.uio_iovcnt++; 811 kth->ktr_len += datalen; 812 } 813 if (buflen != 0) { 814 KASSERT(kth->ktr_buffer != NULL, ("ktrace: nothing to write")); 815 aiov[auio.uio_iovcnt].iov_base = kth->ktr_buffer; 816 aiov[auio.uio_iovcnt].iov_len = buflen; 817 auio.uio_resid += buflen; 818 auio.uio_iovcnt++; 819 } 820 mtx_lock(&Giant); 821 vn_start_write(vp, &mp, V_WAIT); 822 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 823 (void)VOP_LEASE(vp, td, cred, LEASE_WRITE); 824#ifdef MAC 825 error = mac_check_vnode_write(cred, NOCRED, vp); 826 if (error == 0) 827#endif 828 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred); 829 VOP_UNLOCK(vp, 0, td); 830 vn_finished_write(mp); 831 mtx_unlock(&Giant); 832 if (!error) 833 return; 834 /* 835 * If error encountered, give up tracing on this vnode. We defer 836 * all the vrele()'s on the vnode until after we are finished walking 837 * the various lists to avoid needlessly holding locks. 838 */ 839 log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n", 840 error); 841 vrele_count = 0; 842 /* 843 * First, clear this vnode from being used by any processes in the 844 * system. 845 * XXX - If one process gets an EPERM writing to the vnode, should 846 * we really do this? Other processes might have suitable 847 * credentials for the operation. 848 */ 849 cred = NULL; 850 sx_slock(&allproc_lock); 851 LIST_FOREACH(p, &allproc, p_list) { 852 PROC_LOCK(p); 853 if (p->p_tracevp == vp) { 854 mtx_lock(&ktrace_mtx); 855 p->p_tracevp = NULL; 856 p->p_traceflag = 0; 857 cred = p->p_tracecred; 858 p->p_tracecred = NULL; 859 mtx_unlock(&ktrace_mtx); 860 vrele_count++; 861 } 862 PROC_UNLOCK(p); 863 if (cred != NULL) { 864 crfree(cred); 865 cred = NULL; 866 } 867 } 868 sx_sunlock(&allproc_lock); 869 /* 870 * Second, clear this vnode from any pending requests. 871 */ 872 mtx_lock(&ktrace_mtx); 873 STAILQ_FOREACH(req, &ktr_todo, ktr_list) { 874 if (req->ktr_vp == vp) { 875 req->ktr_vp = NULL; 876 vrele_count++; 877 } 878 } 879 mtx_unlock(&ktrace_mtx); 880 mtx_lock(&Giant); 881 while (vrele_count-- > 0) 882 vrele(vp); 883 mtx_unlock(&Giant); 884} 885 886/* 887 * Return true if caller has permission to set the ktracing state 888 * of target. Essentially, the target can't possess any 889 * more permissions than the caller. KTRFAC_ROOT signifies that 890 * root previously set the tracing status on the target process, and 891 * so, only root may further change it. 892 */ 893static int 894ktrcanset(td, targetp) 895 struct thread *td; 896 struct proc *targetp; 897{ 898 899 PROC_LOCK_ASSERT(targetp, MA_OWNED); 900 if (targetp->p_traceflag & KTRFAC_ROOT && 901 suser_cred(td->td_ucred, SUSER_ALLOWJAIL)) 902 return (0); 903 904 if (p_candebug(td, targetp) != 0) 905 return (0); 906 907 return (1); 908} 909 910#endif /* KTRACE */ 911