1 /*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2005 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)kern_ktrace.c 8.2 (Berkeley) 9/23/93 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_ktrace.h" 38 #include "opt_mac.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/fcntl.h> 43 #include <sys/kernel.h> 44 #include <sys/kthread.h> 45 #include <sys/lock.h> 46 #include <sys/mutex.h> 47 #include <sys/malloc.h> 48 #include <sys/mount.h> 49 #include <sys/namei.h> 50 #include <sys/priv.h> 51 #include <sys/proc.h> 52 #include <sys/unistd.h> 53 #include <sys/vnode.h> 54 #include <sys/socket.h> 55 #include <sys/stat.h> 56 #include <sys/ktrace.h> 57 #include <sys/sx.h> 58 #include <sys/sysctl.h> 59 #include <sys/syslog.h> 60 #include <sys/sysproto.h> 61 62 #include <security/mac/mac_framework.h> 63 64 /* 65 * The ktrace facility allows the tracing of certain key events in user space 66 * processes, such as system calls, signal delivery, context switches, and 67 * user generated events using utrace(2). It works by streaming event 68 * records and data to a vnode associated with the process using the 69 * ktrace(2) system call. In general, records can be written directly from 70 * the context that generates the event. One important exception to this is 71 * during a context switch, where sleeping is not permitted. To handle this 72 * case, trace events are generated using in-kernel ktr_request records, and 73 * then delivered to disk at a convenient moment -- either immediately, the 74 * next traceable event, at system call return, or at process exit. 75 * 76 * When dealing with multiple threads or processes writing to the same event 77 * log, ordering guarantees are weak: specifically, if an event has multiple 78 * records (i.e., system call enter and return), they may be interlaced with 79 * records from another event. Process and thread ID information is provided 80 * in the record, and user applications can de-interlace events if required. 81 */ 82 83 static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE"); 84 85 #ifdef KTRACE 86 87 #ifndef KTRACE_REQUEST_POOL 88 #define KTRACE_REQUEST_POOL 100 89 #endif 90 91 struct ktr_request { 92 struct ktr_header ktr_header; 93 void *ktr_buffer; 94 union { 95 struct ktr_syscall ktr_syscall; 96 struct ktr_sysret ktr_sysret; 97 struct ktr_genio ktr_genio; 98 struct ktr_psig ktr_psig; 99 struct ktr_csw ktr_csw; 100 } ktr_data; 101 STAILQ_ENTRY(ktr_request) ktr_list; 102 }; 103 104 static int data_lengths[] = { 105 0, /* none */ 106 offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */ 107 sizeof(struct ktr_sysret), /* KTR_SYSRET */ 108 0, /* KTR_NAMEI */ 109 sizeof(struct ktr_genio), /* KTR_GENIO */ 110 sizeof(struct ktr_psig), /* KTR_PSIG */ 111 sizeof(struct ktr_csw), /* KTR_CSW */ 112 0, /* KTR_USER */ 113 0, /* KTR_STRUCT */ 114 }; 115 116 static STAILQ_HEAD(, ktr_request) ktr_free; 117 118 static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options"); 119 120 static u_int ktr_requestpool = KTRACE_REQUEST_POOL; 121 TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool); 122 123 static u_int ktr_geniosize = PAGE_SIZE; 124 TUNABLE_INT("kern.ktrace.genio_size", &ktr_geniosize); 125 SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RW, &ktr_geniosize, 126 0, "Maximum size of genio event payload"); 127 128 static int print_message = 1; 129 struct mtx ktrace_mtx; 130 static struct sx ktrace_sx; 131 132 static void ktrace_init(void *dummy); 133 static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS); 134 static u_int ktrace_resize_pool(u_int newsize); 135 static struct ktr_request *ktr_getrequest(int type); 136 static void ktr_submitrequest(struct thread *td, struct ktr_request *req); 137 static void ktr_freerequest(struct ktr_request *req); 138 static void ktr_writerequest(struct thread *td, struct ktr_request *req); 139 static int ktrcanset(struct thread *,struct proc *); 140 static int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *); 141 static int ktrops(struct thread *,struct proc *,int,int,struct vnode *); 142 143 /* 144 * ktrace itself generates events, such as context switches, which we do not 145 * wish to trace. Maintain a flag, TDP_INKTRACE, on each thread to determine 146 * whether or not it is in a region where tracing of events should be 147 * suppressed. 148 */ 149 static void 150 ktrace_enter(struct thread *td) 151 { 152 153 KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set")); 154 td->td_pflags |= TDP_INKTRACE; 155 } 156 157 static void 158 ktrace_exit(struct thread *td) 159 { 160 161 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set")); 162 td->td_pflags &= ~TDP_INKTRACE; 163 } 164 165 static void 166 ktrace_assert(struct thread *td) 167 { 168 169 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set")); 170 } 171 172 static void 173 ktrace_init(void *dummy) 174 { 175 struct ktr_request *req; 176 int i; 177 178 mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET); 179 sx_init(&ktrace_sx, "ktrace_sx"); 180 STAILQ_INIT(&ktr_free); 181 for (i = 0; i < ktr_requestpool; i++) { 182 req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK); 183 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 184 } 185 } 186 SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL); 187 188 static int 189 sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS) 190 { 191 struct thread *td; 192 u_int newsize, oldsize, wantsize; 193 int error; 194 195 /* Handle easy read-only case first to avoid warnings from GCC. */ 196 if (!req->newptr) { 197 mtx_lock(&ktrace_mtx); 198 oldsize = ktr_requestpool; 199 mtx_unlock(&ktrace_mtx); 200 return (SYSCTL_OUT(req, &oldsize, sizeof(u_int))); 201 } 202 203 error = SYSCTL_IN(req, &wantsize, sizeof(u_int)); 204 if (error) 205 return (error); 206 td = curthread; 207 ktrace_enter(td); 208 mtx_lock(&ktrace_mtx); 209 oldsize = ktr_requestpool; 210 newsize = ktrace_resize_pool(wantsize); 211 mtx_unlock(&ktrace_mtx); 212 ktrace_exit(td); 213 error = SYSCTL_OUT(req, &oldsize, sizeof(u_int)); 214 if (error) 215 return (error); 216 if (wantsize > oldsize && newsize < wantsize) 217 return (ENOSPC); 218 return (0); 219 } 220 SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW, 221 &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU", ""); 222 223 static u_int 224 ktrace_resize_pool(u_int newsize) 225 { 226 struct ktr_request *req; 227 int bound; 228 229 mtx_assert(&ktrace_mtx, MA_OWNED); 230 print_message = 1; 231 bound = newsize - ktr_requestpool; 232 if (bound == 0) 233 return (ktr_requestpool); 234 if (bound < 0) 235 /* Shrink pool down to newsize if possible. */ 236 while (bound++ < 0) { 237 req = STAILQ_FIRST(&ktr_free); 238 if (req == NULL) 239 return (ktr_requestpool); 240 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 241 ktr_requestpool--; 242 mtx_unlock(&ktrace_mtx); 243 free(req, M_KTRACE); 244 mtx_lock(&ktrace_mtx); 245 } 246 else 247 /* Grow pool up to newsize. */ 248 while (bound-- > 0) { 249 mtx_unlock(&ktrace_mtx); 250 req = malloc(sizeof(struct ktr_request), M_KTRACE, 251 M_WAITOK); 252 mtx_lock(&ktrace_mtx); 253 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 254 ktr_requestpool++; 255 } 256 return (ktr_requestpool); 257 } 258 259 static struct ktr_request * 260 ktr_getrequest(int type) 261 { 262 struct ktr_request *req; 263 struct thread *td = curthread; 264 struct proc *p = td->td_proc; 265 int pm; 266 267 ktrace_enter(td); /* XXX: In caller instead? */ 268 mtx_lock(&ktrace_mtx); 269 if (!KTRCHECK(td, type)) { 270 mtx_unlock(&ktrace_mtx); 271 ktrace_exit(td); 272 return (NULL); 273 } 274 req = STAILQ_FIRST(&ktr_free); 275 if (req != NULL) { 276 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 277 req->ktr_header.ktr_type = type; 278 if (p->p_traceflag & KTRFAC_DROP) { 279 req->ktr_header.ktr_type |= KTR_DROP; 280 p->p_traceflag &= ~KTRFAC_DROP; 281 } 282 mtx_unlock(&ktrace_mtx); 283 microtime(&req->ktr_header.ktr_time); 284 req->ktr_header.ktr_pid = p->p_pid; 285 req->ktr_header.ktr_tid = td->td_tid; 286 bcopy(td->td_name, req->ktr_header.ktr_comm, MAXCOMLEN + 1); 287 req->ktr_buffer = NULL; 288 req->ktr_header.ktr_len = 0; 289 } else { 290 p->p_traceflag |= KTRFAC_DROP; 291 pm = print_message; 292 print_message = 0; 293 mtx_unlock(&ktrace_mtx); 294 if (pm) 295 printf("Out of ktrace request objects.\n"); 296 ktrace_exit(td); 297 } 298 return (req); 299 } 300 301 /* 302 * Some trace generation environments don't permit direct access to VFS, 303 * such as during a context switch where sleeping is not allowed. Under these 304 * circumstances, queue a request to the thread to be written asynchronously 305 * later. 306 */ 307 static void 308 ktr_enqueuerequest(struct thread *td, struct ktr_request *req) 309 { 310 311 mtx_lock(&ktrace_mtx); 312 STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list); 313 mtx_unlock(&ktrace_mtx); 314 ktrace_exit(td); 315 } 316 317 /* 318 * Drain any pending ktrace records from the per-thread queue to disk. This 319 * is used both internally before committing other records, and also on 320 * system call return. We drain all the ones we can find at the time when 321 * drain is requested, but don't keep draining after that as those events 322 * may me approximately "after" the current event. 323 */ 324 static void 325 ktr_drain(struct thread *td) 326 { 327 struct ktr_request *queued_req; 328 STAILQ_HEAD(, ktr_request) local_queue; 329 330 ktrace_assert(td); 331 sx_assert(&ktrace_sx, SX_XLOCKED); 332 333 STAILQ_INIT(&local_queue); /* XXXRW: needed? */ 334 335 if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) { 336 mtx_lock(&ktrace_mtx); 337 STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr); 338 mtx_unlock(&ktrace_mtx); 339 340 while ((queued_req = STAILQ_FIRST(&local_queue))) { 341 STAILQ_REMOVE_HEAD(&local_queue, ktr_list); 342 ktr_writerequest(td, queued_req); 343 ktr_freerequest(queued_req); 344 } 345 } 346 } 347 348 /* 349 * Submit a trace record for immediate commit to disk -- to be used only 350 * where entering VFS is OK. First drain any pending records that may have 351 * been cached in the thread. 352 */ 353 static void 354 ktr_submitrequest(struct thread *td, struct ktr_request *req) 355 { 356 357 ktrace_assert(td); 358 359 sx_xlock(&ktrace_sx); 360 ktr_drain(td); 361 ktr_writerequest(td, req); 362 ktr_freerequest(req); 363 sx_xunlock(&ktrace_sx); 364 365 ktrace_exit(td); 366 } 367 368 static void 369 ktr_freerequest(struct ktr_request *req) 370 { 371 372 if (req->ktr_buffer != NULL) 373 free(req->ktr_buffer, M_KTRACE); 374 mtx_lock(&ktrace_mtx); 375 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 376 mtx_unlock(&ktrace_mtx); 377 } 378 379 void 380 ktrsyscall(code, narg, args) 381 int code, narg; 382 register_t args[]; 383 { 384 struct ktr_request *req; 385 struct ktr_syscall *ktp; 386 size_t buflen; 387 char *buf = NULL; 388 389 buflen = sizeof(register_t) * narg; 390 if (buflen > 0) { 391 buf = malloc(buflen, M_KTRACE, M_WAITOK); 392 bcopy(args, buf, buflen); 393 } 394 req = ktr_getrequest(KTR_SYSCALL); 395 if (req == NULL) { 396 if (buf != NULL) 397 free(buf, M_KTRACE); 398 return; 399 } 400 ktp = &req->ktr_data.ktr_syscall; 401 ktp->ktr_code = code; 402 ktp->ktr_narg = narg; 403 if (buflen > 0) { 404 req->ktr_header.ktr_len = buflen; 405 req->ktr_buffer = buf; 406 } 407 ktr_submitrequest(curthread, req); 408 } 409 410 void 411 ktrsysret(code, error, retval) 412 int code, error; 413 register_t retval; 414 { 415 struct ktr_request *req; 416 struct ktr_sysret *ktp; 417 418 req = ktr_getrequest(KTR_SYSRET); 419 if (req == NULL) 420 return; 421 ktp = &req->ktr_data.ktr_sysret; 422 ktp->ktr_code = code; 423 ktp->ktr_error = error; 424 ktp->ktr_retval = retval; /* what about val2 ? */ 425 ktr_submitrequest(curthread, req); 426 } 427 428 /* 429 * When a process exits, drain per-process asynchronous trace records. 430 */ 431 void 432 ktrprocexit(struct thread *td) 433 { 434 435 ktrace_enter(td); 436 sx_xlock(&ktrace_sx); 437 ktr_drain(td); 438 sx_xunlock(&ktrace_sx); 439 ktrace_exit(td); 440 } 441 442 /* 443 * When a thread returns, drain any asynchronous records generated by the 444 * system call. 445 */ 446 void 447 ktruserret(struct thread *td) 448 { 449 450 ktrace_enter(td); 451 sx_xlock(&ktrace_sx); 452 ktr_drain(td); 453 sx_xunlock(&ktrace_sx); 454 ktrace_exit(td); 455 } 456 457 void 458 ktrnamei(path) 459 char *path; 460 { 461 struct ktr_request *req; 462 int namelen; 463 char *buf = NULL; 464 465 namelen = strlen(path); 466 if (namelen > 0) { 467 buf = malloc(namelen, M_KTRACE, M_WAITOK); 468 bcopy(path, buf, namelen); 469 } 470 req = ktr_getrequest(KTR_NAMEI); 471 if (req == NULL) { 472 if (buf != NULL) 473 free(buf, M_KTRACE); 474 return; 475 } 476 if (namelen > 0) { 477 req->ktr_header.ktr_len = namelen; 478 req->ktr_buffer = buf; 479 } 480 ktr_submitrequest(curthread, req); 481 } 482 483 void 484 ktrgenio(fd, rw, uio, error) 485 int fd; 486 enum uio_rw rw; 487 struct uio *uio; 488 int error; 489 { 490 struct ktr_request *req; 491 struct ktr_genio *ktg; 492 int datalen; 493 char *buf; 494 495 if (error) { 496 free(uio, M_IOV); 497 return; 498 } 499 uio->uio_offset = 0; 500 uio->uio_rw = UIO_WRITE; 501 datalen = imin(uio->uio_resid, ktr_geniosize); 502 buf = malloc(datalen, M_KTRACE, M_WAITOK); 503 error = uiomove(buf, datalen, uio); 504 free(uio, M_IOV); 505 if (error) { 506 free(buf, M_KTRACE); 507 return; 508 } 509 req = ktr_getrequest(KTR_GENIO); 510 if (req == NULL) { 511 free(buf, M_KTRACE); 512 return; 513 } 514 ktg = &req->ktr_data.ktr_genio; 515 ktg->ktr_fd = fd; 516 ktg->ktr_rw = rw; 517 req->ktr_header.ktr_len = datalen; 518 req->ktr_buffer = buf; 519 ktr_submitrequest(curthread, req); 520 } 521 522 void 523 ktrpsig(sig, action, mask, code) 524 int sig; 525 sig_t action; 526 sigset_t *mask; 527 int code; 528 { 529 struct ktr_request *req; 530 struct ktr_psig *kp; 531 532 req = ktr_getrequest(KTR_PSIG); 533 if (req == NULL) 534 return; 535 kp = &req->ktr_data.ktr_psig; 536 kp->signo = (char)sig; 537 kp->action = action; 538 kp->mask = *mask; 539 kp->code = code; 540 ktr_enqueuerequest(curthread, req); 541 } 542 543 void 544 ktrcsw(out, user) 545 int out, user; 546 { 547 struct ktr_request *req; 548 struct ktr_csw *kc; 549 550 req = ktr_getrequest(KTR_CSW); 551 if (req == NULL) 552 return; 553 kc = &req->ktr_data.ktr_csw; 554 kc->out = out; 555 kc->user = user; 556 ktr_enqueuerequest(curthread, req); 557 } 558 559 void 560 ktrstruct(name, namelen, data, datalen) 561 const char *name; 562 size_t namelen; 563 void *data; 564 size_t datalen; 565 { 566 struct ktr_request *req; 567 char *buf = NULL; 568 size_t buflen; 569 570 if (!data) 571 datalen = 0; 572 buflen = namelen + 1 + datalen; 573 buf = malloc(buflen, M_KTRACE, M_WAITOK); 574 bcopy(name, buf, namelen); 575 buf[namelen] = '\0'; 576 bcopy(data, buf + namelen + 1, datalen); 577 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) { 578 free(buf, M_KTRACE); 579 return; 580 } 581 req->ktr_buffer = buf; 582 req->ktr_header.ktr_len = buflen; 583 ktr_submitrequest(curthread, req); 584 } 585 #endif /* KTRACE */ 586 587 /* Interface and common routines */ 588 589 #ifndef _SYS_SYSPROTO_H_ 590 struct ktrace_args { 591 char *fname; 592 int ops; 593 int facs; 594 int pid; 595 }; 596 #endif 597 /* ARGSUSED */ 598 int 599 ktrace(td, uap) 600 struct thread *td; 601 register struct ktrace_args *uap; 602 { 603 #ifdef KTRACE 604 register struct vnode *vp = NULL; 605 register struct proc *p; 606 struct pgrp *pg; 607 int facs = uap->facs & ~KTRFAC_ROOT; 608 int ops = KTROP(uap->ops); 609 int descend = uap->ops & KTRFLAG_DESCEND; 610 int nfound, ret = 0; 611 int flags, error = 0, vfslocked; 612 struct nameidata nd; 613 struct ucred *cred; 614 615 /* 616 * Need something to (un)trace. 617 */ 618 if (ops != KTROP_CLEARFILE && facs == 0) 619 return (EINVAL); 620 621 ktrace_enter(td); 622 if (ops != KTROP_CLEAR) { 623 /* 624 * an operation which requires a file argument. 625 */ 626 NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_USERSPACE, 627 uap->fname, td); 628 flags = FREAD | FWRITE | O_NOFOLLOW; 629 error = vn_open(&nd, &flags, 0, NULL); 630 if (error) { 631 ktrace_exit(td); 632 return (error); 633 } 634 vfslocked = NDHASGIANT(&nd); 635 NDFREE(&nd, NDF_ONLY_PNBUF); 636 vp = nd.ni_vp; 637 VOP_UNLOCK(vp, 0); 638 if (vp->v_type != VREG) { 639 (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td); 640 VFS_UNLOCK_GIANT(vfslocked); 641 ktrace_exit(td); 642 return (EACCES); 643 } 644 VFS_UNLOCK_GIANT(vfslocked); 645 } 646 /* 647 * Clear all uses of the tracefile. 648 */ 649 if (ops == KTROP_CLEARFILE) { 650 int vrele_count; 651 652 vrele_count = 0; 653 sx_slock(&allproc_lock); 654 FOREACH_PROC_IN_SYSTEM(p) { 655 PROC_LOCK(p); 656 if (p->p_tracevp == vp) { 657 if (ktrcanset(td, p)) { 658 mtx_lock(&ktrace_mtx); 659 cred = p->p_tracecred; 660 p->p_tracecred = NULL; 661 p->p_tracevp = NULL; 662 p->p_traceflag = 0; 663 mtx_unlock(&ktrace_mtx); 664 vrele_count++; 665 crfree(cred); 666 } else 667 error = EPERM; 668 } 669 PROC_UNLOCK(p); 670 } 671 sx_sunlock(&allproc_lock); 672 if (vrele_count > 0) { 673 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 674 while (vrele_count-- > 0) 675 vrele(vp); 676 VFS_UNLOCK_GIANT(vfslocked); 677 } 678 goto done; 679 } 680 /* 681 * do it 682 */ 683 sx_slock(&proctree_lock); 684 if (uap->pid < 0) { 685 /* 686 * by process group 687 */ 688 pg = pgfind(-uap->pid); 689 if (pg == NULL) { 690 sx_sunlock(&proctree_lock); 691 error = ESRCH; 692 goto done; 693 } 694 /* 695 * ktrops() may call vrele(). Lock pg_members 696 * by the proctree_lock rather than pg_mtx. 697 */ 698 PGRP_UNLOCK(pg); 699 nfound = 0; 700 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 701 PROC_LOCK(p); 702 if (p_cansee(td, p) != 0) { 703 PROC_UNLOCK(p); 704 continue; 705 } 706 PROC_UNLOCK(p); 707 nfound++; 708 if (descend) 709 ret |= ktrsetchildren(td, p, ops, facs, vp); 710 else 711 ret |= ktrops(td, p, ops, facs, vp); 712 } 713 if (nfound == 0) { 714 sx_sunlock(&proctree_lock); 715 error = ESRCH; 716 goto done; 717 } 718 } else { 719 /* 720 * by pid 721 */ 722 p = pfind(uap->pid); 723 if (p == NULL) { 724 sx_sunlock(&proctree_lock); 725 error = ESRCH; 726 goto done; 727 } 728 error = p_cansee(td, p); 729 /* 730 * The slock of the proctree lock will keep this process 731 * from going away, so unlocking the proc here is ok. 732 */ 733 PROC_UNLOCK(p); 734 if (error) { 735 sx_sunlock(&proctree_lock); 736 goto done; 737 } 738 if (descend) 739 ret |= ktrsetchildren(td, p, ops, facs, vp); 740 else 741 ret |= ktrops(td, p, ops, facs, vp); 742 } 743 sx_sunlock(&proctree_lock); 744 if (!ret) 745 error = EPERM; 746 done: 747 if (vp != NULL) { 748 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 749 (void) vn_close(vp, FWRITE, td->td_ucred, td); 750 VFS_UNLOCK_GIANT(vfslocked); 751 } 752 ktrace_exit(td); 753 return (error); 754 #else /* !KTRACE */ 755 return (ENOSYS); 756 #endif /* KTRACE */ 757 } 758 759 /* ARGSUSED */ 760 int 761 utrace(td, uap) 762 struct thread *td; 763 register struct utrace_args *uap; 764 { 765 766 #ifdef KTRACE 767 struct ktr_request *req; 768 void *cp; 769 int error; 770 771 if (!KTRPOINT(td, KTR_USER)) 772 return (0); 773 if (uap->len > KTR_USER_MAXLEN) 774 return (EINVAL); 775 cp = malloc(uap->len, M_KTRACE, M_WAITOK); 776 error = copyin(uap->addr, cp, uap->len); 777 if (error) { 778 free(cp, M_KTRACE); 779 return (error); 780 } 781 req = ktr_getrequest(KTR_USER); 782 if (req == NULL) { 783 free(cp, M_KTRACE); 784 return (ENOMEM); 785 } 786 req->ktr_buffer = cp; 787 req->ktr_header.ktr_len = uap->len; 788 ktr_submitrequest(td, req); 789 return (0); 790 #else /* !KTRACE */ 791 return (ENOSYS); 792 #endif /* KTRACE */ 793 } 794 795 #ifdef KTRACE 796 static int 797 ktrops(td, p, ops, facs, vp) 798 struct thread *td; 799 struct proc *p; 800 int ops, facs; 801 struct vnode *vp; 802 { 803 struct vnode *tracevp = NULL; 804 struct ucred *tracecred = NULL; 805 806 PROC_LOCK(p); 807 if (!ktrcanset(td, p)) { 808 PROC_UNLOCK(p); 809 return (0); 810 } 811 mtx_lock(&ktrace_mtx); 812 if (ops == KTROP_SET) { 813 if (p->p_tracevp != vp) { 814 /* 815 * if trace file already in use, relinquish below 816 */ 817 tracevp = p->p_tracevp; 818 VREF(vp); 819 p->p_tracevp = vp; 820 } 821 if (p->p_tracecred != td->td_ucred) { 822 tracecred = p->p_tracecred; 823 p->p_tracecred = crhold(td->td_ucred); 824 } 825 p->p_traceflag |= facs; 826 if (priv_check(td, PRIV_KTRACE) == 0) 827 p->p_traceflag |= KTRFAC_ROOT; 828 } else { 829 /* KTROP_CLEAR */ 830 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) { 831 /* no more tracing */ 832 p->p_traceflag = 0; 833 tracevp = p->p_tracevp; 834 p->p_tracevp = NULL; 835 tracecred = p->p_tracecred; 836 p->p_tracecred = NULL; 837 } 838 } 839 mtx_unlock(&ktrace_mtx); 840 PROC_UNLOCK(p); 841 if (tracevp != NULL) { 842 int vfslocked; 843 844 vfslocked = VFS_LOCK_GIANT(tracevp->v_mount); 845 vrele(tracevp); 846 VFS_UNLOCK_GIANT(vfslocked); 847 } 848 if (tracecred != NULL) 849 crfree(tracecred); 850 851 return (1); 852 } 853 854 static int 855 ktrsetchildren(td, top, ops, facs, vp) 856 struct thread *td; 857 struct proc *top; 858 int ops, facs; 859 struct vnode *vp; 860 { 861 register struct proc *p; 862 register int ret = 0; 863 864 p = top; 865 sx_assert(&proctree_lock, SX_LOCKED); 866 for (;;) { 867 ret |= ktrops(td, p, ops, facs, vp); 868 /* 869 * If this process has children, descend to them next, 870 * otherwise do any siblings, and if done with this level, 871 * follow back up the tree (but not past top). 872 */ 873 if (!LIST_EMPTY(&p->p_children)) 874 p = LIST_FIRST(&p->p_children); 875 else for (;;) { 876 if (p == top) 877 return (ret); 878 if (LIST_NEXT(p, p_sibling)) { 879 p = LIST_NEXT(p, p_sibling); 880 break; 881 } 882 p = p->p_pptr; 883 } 884 } 885 /*NOTREACHED*/ 886 } 887 888 static void 889 ktr_writerequest(struct thread *td, struct ktr_request *req) 890 { 891 struct ktr_header *kth; 892 struct vnode *vp; 893 struct proc *p; 894 struct ucred *cred; 895 struct uio auio; 896 struct iovec aiov[3]; 897 struct mount *mp; 898 int datalen, buflen, vrele_count; 899 int error, vfslocked; 900 901 /* 902 * We hold the vnode and credential for use in I/O in case ktrace is 903 * disabled on the process as we write out the request. 904 * 905 * XXXRW: This is not ideal: we could end up performing a write after 906 * the vnode has been closed. 907 */ 908 mtx_lock(&ktrace_mtx); 909 vp = td->td_proc->p_tracevp; 910 if (vp != NULL) 911 VREF(vp); 912 cred = td->td_proc->p_tracecred; 913 if (cred != NULL) 914 crhold(cred); 915 mtx_unlock(&ktrace_mtx); 916 917 /* 918 * If vp is NULL, the vp has been cleared out from under this 919 * request, so just drop it. Make sure the credential and vnode are 920 * in sync: we should have both or neither. 921 */ 922 if (vp == NULL) { 923 KASSERT(cred == NULL, ("ktr_writerequest: cred != NULL")); 924 return; 925 } 926 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL")); 927 928 kth = &req->ktr_header; 929 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP]; 930 buflen = kth->ktr_len; 931 auio.uio_iov = &aiov[0]; 932 auio.uio_offset = 0; 933 auio.uio_segflg = UIO_SYSSPACE; 934 auio.uio_rw = UIO_WRITE; 935 aiov[0].iov_base = (caddr_t)kth; 936 aiov[0].iov_len = sizeof(struct ktr_header); 937 auio.uio_resid = sizeof(struct ktr_header); 938 auio.uio_iovcnt = 1; 939 auio.uio_td = td; 940 if (datalen != 0) { 941 aiov[1].iov_base = (caddr_t)&req->ktr_data; 942 aiov[1].iov_len = datalen; 943 auio.uio_resid += datalen; 944 auio.uio_iovcnt++; 945 kth->ktr_len += datalen; 946 } 947 if (buflen != 0) { 948 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write")); 949 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer; 950 aiov[auio.uio_iovcnt].iov_len = buflen; 951 auio.uio_resid += buflen; 952 auio.uio_iovcnt++; 953 } 954 955 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 956 vn_start_write(vp, &mp, V_WAIT); 957 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 958 (void)VOP_LEASE(vp, td, cred, LEASE_WRITE); 959 #ifdef MAC 960 error = mac_vnode_check_write(cred, NOCRED, vp); 961 if (error == 0) 962 #endif 963 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred); 964 VOP_UNLOCK(vp, 0); 965 vn_finished_write(mp); 966 vrele(vp); 967 VFS_UNLOCK_GIANT(vfslocked); 968 if (!error) 969 return; 970 /* 971 * If error encountered, give up tracing on this vnode. We defer 972 * all the vrele()'s on the vnode until after we are finished walking 973 * the various lists to avoid needlessly holding locks. 974 */ 975 log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n", 976 error); 977 vrele_count = 0; 978 /* 979 * First, clear this vnode from being used by any processes in the 980 * system. 981 * XXX - If one process gets an EPERM writing to the vnode, should 982 * we really do this? Other processes might have suitable 983 * credentials for the operation. 984 */ 985 cred = NULL; 986 sx_slock(&allproc_lock); 987 FOREACH_PROC_IN_SYSTEM(p) { 988 PROC_LOCK(p); 989 if (p->p_tracevp == vp) { 990 mtx_lock(&ktrace_mtx); 991 p->p_tracevp = NULL; 992 p->p_traceflag = 0; 993 cred = p->p_tracecred; 994 p->p_tracecred = NULL; 995 mtx_unlock(&ktrace_mtx); 996 vrele_count++; 997 } 998 PROC_UNLOCK(p); 999 if (cred != NULL) { 1000 crfree(cred); 1001 cred = NULL; 1002 } 1003 } 1004 sx_sunlock(&allproc_lock); 1005 1006 /* 1007 * We can't clear any pending requests in threads that have cached 1008 * them but not yet committed them, as those are per-thread. The 1009 * thread will have to clear it itself on system call return. 1010 */ 1011 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1012 while (vrele_count-- > 0) 1013 vrele(vp); 1014 VFS_UNLOCK_GIANT(vfslocked); 1015 } 1016 1017 /* 1018 * Return true if caller has permission to set the ktracing state 1019 * of target. Essentially, the target can't possess any 1020 * more permissions than the caller. KTRFAC_ROOT signifies that 1021 * root previously set the tracing status on the target process, and 1022 * so, only root may further change it. 1023 */ 1024 static int 1025 ktrcanset(td, targetp) 1026 struct thread *td; 1027 struct proc *targetp; 1028 { 1029 1030 PROC_LOCK_ASSERT(targetp, MA_OWNED); 1031 if (targetp->p_traceflag & KTRFAC_ROOT && 1032 priv_check(td, PRIV_KTRACE)) 1033 return (0); 1034 1035 if (p_candebug(td, targetp) != 0) 1036 return (0); 1037 1038 return (1); 1039 } 1040 1041 #endif /* KTRACE */ 1042