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 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/fcntl.h> 42 #include <sys/kernel.h> 43 #include <sys/kthread.h> 44 #include <sys/lock.h> 45 #include <sys/mutex.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/namei.h> 49 #include <sys/priv.h> 50 #include <sys/proc.h> 51 #include <sys/unistd.h> 52 #include <sys/vnode.h> 53 #include <sys/socket.h> 54 #include <sys/stat.h> 55 #include <sys/ktrace.h> 56 #include <sys/sx.h> 57 #include <sys/sysctl.h> 58 #include <sys/syslog.h> 59 #include <sys/sysproto.h> 60 61 #include <security/mac/mac_framework.h> 62 63 /* 64 * The ktrace facility allows the tracing of certain key events in user space 65 * processes, such as system calls, signal delivery, context switches, and 66 * user generated events using utrace(2). It works by streaming event 67 * records and data to a vnode associated with the process using the 68 * ktrace(2) system call. In general, records can be written directly from 69 * the context that generates the event. One important exception to this is 70 * during a context switch, where sleeping is not permitted. To handle this 71 * case, trace events are generated using in-kernel ktr_request records, and 72 * then delivered to disk at a convenient moment -- either immediately, the 73 * next traceable event, at system call return, or at process exit. 74 * 75 * When dealing with multiple threads or processes writing to the same event 76 * log, ordering guarantees are weak: specifically, if an event has multiple 77 * records (i.e., system call enter and return), they may be interlaced with 78 * records from another event. Process and thread ID information is provided 79 * in the record, and user applications can de-interlace events if required. 80 */ 81 82 static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE"); 83 84 #ifdef KTRACE 85 86 #ifndef KTRACE_REQUEST_POOL 87 #define KTRACE_REQUEST_POOL 100 88 #endif 89 90 struct ktr_request { 91 struct ktr_header ktr_header; 92 void *ktr_buffer; 93 union { 94 struct ktr_syscall ktr_syscall; 95 struct ktr_sysret ktr_sysret; 96 struct ktr_genio ktr_genio; 97 struct ktr_psig ktr_psig; 98 struct ktr_csw ktr_csw; 99 } ktr_data; 100 STAILQ_ENTRY(ktr_request) ktr_list; 101 }; 102 103 static int data_lengths[] = { 104 0, /* none */ 105 offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */ 106 sizeof(struct ktr_sysret), /* KTR_SYSRET */ 107 0, /* KTR_NAMEI */ 108 sizeof(struct ktr_genio), /* KTR_GENIO */ 109 sizeof(struct ktr_psig), /* KTR_PSIG */ 110 sizeof(struct ktr_csw), /* KTR_CSW */ 111 0, /* KTR_USER */ 112 0, /* KTR_STRUCT */ 113 0, /* KTR_SYSCTL */ 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 /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */ 260 CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) == 261 (sizeof((struct thread *)NULL)->td_name)); 262 263 static struct ktr_request * 264 ktr_getrequest(int type) 265 { 266 struct ktr_request *req; 267 struct thread *td = curthread; 268 struct proc *p = td->td_proc; 269 int pm; 270 271 ktrace_enter(td); /* XXX: In caller instead? */ 272 mtx_lock(&ktrace_mtx); 273 if (!KTRCHECK(td, type)) { 274 mtx_unlock(&ktrace_mtx); 275 ktrace_exit(td); 276 return (NULL); 277 } 278 req = STAILQ_FIRST(&ktr_free); 279 if (req != NULL) { 280 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 281 req->ktr_header.ktr_type = type; 282 if (p->p_traceflag & KTRFAC_DROP) { 283 req->ktr_header.ktr_type |= KTR_DROP; 284 p->p_traceflag &= ~KTRFAC_DROP; 285 } 286 mtx_unlock(&ktrace_mtx); 287 microtime(&req->ktr_header.ktr_time); 288 req->ktr_header.ktr_pid = p->p_pid; 289 req->ktr_header.ktr_tid = td->td_tid; 290 bcopy(td->td_name, req->ktr_header.ktr_comm, 291 sizeof(req->ktr_header.ktr_comm)); 292 req->ktr_buffer = NULL; 293 req->ktr_header.ktr_len = 0; 294 } else { 295 p->p_traceflag |= KTRFAC_DROP; 296 pm = print_message; 297 print_message = 0; 298 mtx_unlock(&ktrace_mtx); 299 if (pm) 300 printf("Out of ktrace request objects.\n"); 301 ktrace_exit(td); 302 } 303 return (req); 304 } 305 306 /* 307 * Some trace generation environments don't permit direct access to VFS, 308 * such as during a context switch where sleeping is not allowed. Under these 309 * circumstances, queue a request to the thread to be written asynchronously 310 * later. 311 */ 312 static void 313 ktr_enqueuerequest(struct thread *td, struct ktr_request *req) 314 { 315 316 mtx_lock(&ktrace_mtx); 317 STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list); 318 mtx_unlock(&ktrace_mtx); 319 ktrace_exit(td); 320 } 321 322 /* 323 * Drain any pending ktrace records from the per-thread queue to disk. This 324 * is used both internally before committing other records, and also on 325 * system call return. We drain all the ones we can find at the time when 326 * drain is requested, but don't keep draining after that as those events 327 * may be approximately "after" the current event. 328 */ 329 static void 330 ktr_drain(struct thread *td) 331 { 332 struct ktr_request *queued_req; 333 STAILQ_HEAD(, ktr_request) local_queue; 334 335 ktrace_assert(td); 336 sx_assert(&ktrace_sx, SX_XLOCKED); 337 338 STAILQ_INIT(&local_queue); /* XXXRW: needed? */ 339 340 if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) { 341 mtx_lock(&ktrace_mtx); 342 STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr); 343 mtx_unlock(&ktrace_mtx); 344 345 while ((queued_req = STAILQ_FIRST(&local_queue))) { 346 STAILQ_REMOVE_HEAD(&local_queue, ktr_list); 347 ktr_writerequest(td, queued_req); 348 ktr_freerequest(queued_req); 349 } 350 } 351 } 352 353 /* 354 * Submit a trace record for immediate commit to disk -- to be used only 355 * where entering VFS is OK. First drain any pending records that may have 356 * been cached in the thread. 357 */ 358 static void 359 ktr_submitrequest(struct thread *td, struct ktr_request *req) 360 { 361 362 ktrace_assert(td); 363 364 sx_xlock(&ktrace_sx); 365 ktr_drain(td); 366 ktr_writerequest(td, req); 367 ktr_freerequest(req); 368 sx_xunlock(&ktrace_sx); 369 370 ktrace_exit(td); 371 } 372 373 static void 374 ktr_freerequest(struct ktr_request *req) 375 { 376 377 if (req->ktr_buffer != NULL) 378 free(req->ktr_buffer, M_KTRACE); 379 mtx_lock(&ktrace_mtx); 380 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 381 mtx_unlock(&ktrace_mtx); 382 } 383 384 void 385 ktrsyscall(code, narg, args) 386 int code, narg; 387 register_t args[]; 388 { 389 struct ktr_request *req; 390 struct ktr_syscall *ktp; 391 size_t buflen; 392 char *buf = NULL; 393 394 buflen = sizeof(register_t) * narg; 395 if (buflen > 0) { 396 buf = malloc(buflen, M_KTRACE, M_WAITOK); 397 bcopy(args, buf, buflen); 398 } 399 req = ktr_getrequest(KTR_SYSCALL); 400 if (req == NULL) { 401 if (buf != NULL) 402 free(buf, M_KTRACE); 403 return; 404 } 405 ktp = &req->ktr_data.ktr_syscall; 406 ktp->ktr_code = code; 407 ktp->ktr_narg = narg; 408 if (buflen > 0) { 409 req->ktr_header.ktr_len = buflen; 410 req->ktr_buffer = buf; 411 } 412 ktr_submitrequest(curthread, req); 413 } 414 415 void 416 ktrsysret(code, error, retval) 417 int code, error; 418 register_t retval; 419 { 420 struct ktr_request *req; 421 struct ktr_sysret *ktp; 422 423 req = ktr_getrequest(KTR_SYSRET); 424 if (req == NULL) 425 return; 426 ktp = &req->ktr_data.ktr_sysret; 427 ktp->ktr_code = code; 428 ktp->ktr_error = error; 429 ktp->ktr_retval = retval; /* what about val2 ? */ 430 ktr_submitrequest(curthread, req); 431 } 432 433 /* 434 * When a process exits, drain per-process asynchronous trace records. 435 */ 436 void 437 ktrprocexit(struct thread *td) 438 { 439 440 ktrace_enter(td); 441 sx_xlock(&ktrace_sx); 442 ktr_drain(td); 443 sx_xunlock(&ktrace_sx); 444 ktrace_exit(td); 445 } 446 447 /* 448 * When a thread returns, drain any asynchronous records generated by the 449 * system call. 450 */ 451 void 452 ktruserret(struct thread *td) 453 { 454 455 ktrace_enter(td); 456 sx_xlock(&ktrace_sx); 457 ktr_drain(td); 458 sx_xunlock(&ktrace_sx); 459 ktrace_exit(td); 460 } 461 462 void 463 ktrnamei(path) 464 char *path; 465 { 466 struct ktr_request *req; 467 int namelen; 468 char *buf = NULL; 469 470 namelen = strlen(path); 471 if (namelen > 0) { 472 buf = malloc(namelen, M_KTRACE, M_WAITOK); 473 bcopy(path, buf, namelen); 474 } 475 req = ktr_getrequest(KTR_NAMEI); 476 if (req == NULL) { 477 if (buf != NULL) 478 free(buf, M_KTRACE); 479 return; 480 } 481 if (namelen > 0) { 482 req->ktr_header.ktr_len = namelen; 483 req->ktr_buffer = buf; 484 } 485 ktr_submitrequest(curthread, req); 486 } 487 488 void 489 ktrsysctl(name, namelen) 490 int *name; 491 u_int namelen; 492 { 493 struct ktr_request *req; 494 u_int mib[CTL_MAXNAME + 2]; 495 char *mibname; 496 size_t mibnamelen; 497 int error; 498 499 /* Lookup name of mib. */ 500 KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long")); 501 mib[0] = 0; 502 mib[1] = 1; 503 bcopy(name, mib + 2, namelen * sizeof(*name)); 504 mibnamelen = 128; 505 mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK); 506 error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen, 507 NULL, 0, &mibnamelen, 0); 508 if (error) { 509 free(mibname, M_KTRACE); 510 return; 511 } 512 req = ktr_getrequest(KTR_SYSCTL); 513 if (req == NULL) { 514 free(mibname, M_KTRACE); 515 return; 516 } 517 req->ktr_header.ktr_len = mibnamelen; 518 req->ktr_buffer = mibname; 519 ktr_submitrequest(curthread, req); 520 } 521 522 void 523 ktrgenio(fd, rw, uio, error) 524 int fd; 525 enum uio_rw rw; 526 struct uio *uio; 527 int error; 528 { 529 struct ktr_request *req; 530 struct ktr_genio *ktg; 531 int datalen; 532 char *buf; 533 534 if (error) { 535 free(uio, M_IOV); 536 return; 537 } 538 uio->uio_offset = 0; 539 uio->uio_rw = UIO_WRITE; 540 datalen = imin(uio->uio_resid, ktr_geniosize); 541 buf = malloc(datalen, M_KTRACE, M_WAITOK); 542 error = uiomove(buf, datalen, uio); 543 free(uio, M_IOV); 544 if (error) { 545 free(buf, M_KTRACE); 546 return; 547 } 548 req = ktr_getrequest(KTR_GENIO); 549 if (req == NULL) { 550 free(buf, M_KTRACE); 551 return; 552 } 553 ktg = &req->ktr_data.ktr_genio; 554 ktg->ktr_fd = fd; 555 ktg->ktr_rw = rw; 556 req->ktr_header.ktr_len = datalen; 557 req->ktr_buffer = buf; 558 ktr_submitrequest(curthread, req); 559 } 560 561 void 562 ktrpsig(sig, action, mask, code) 563 int sig; 564 sig_t action; 565 sigset_t *mask; 566 int code; 567 { 568 struct ktr_request *req; 569 struct ktr_psig *kp; 570 571 req = ktr_getrequest(KTR_PSIG); 572 if (req == NULL) 573 return; 574 kp = &req->ktr_data.ktr_psig; 575 kp->signo = (char)sig; 576 kp->action = action; 577 kp->mask = *mask; 578 kp->code = code; 579 ktr_enqueuerequest(curthread, req); 580 } 581 582 void 583 ktrcsw(out, user) 584 int out, user; 585 { 586 struct ktr_request *req; 587 struct ktr_csw *kc; 588 589 req = ktr_getrequest(KTR_CSW); 590 if (req == NULL) 591 return; 592 kc = &req->ktr_data.ktr_csw; 593 kc->out = out; 594 kc->user = user; 595 ktr_enqueuerequest(curthread, req); 596 } 597 598 void 599 ktrstruct(name, namelen, data, datalen) 600 const char *name; 601 size_t namelen; 602 void *data; 603 size_t datalen; 604 { 605 struct ktr_request *req; 606 char *buf = NULL; 607 size_t buflen; 608 609 if (!data) 610 datalen = 0; 611 buflen = namelen + 1 + datalen; 612 buf = malloc(buflen, M_KTRACE, M_WAITOK); 613 bcopy(name, buf, namelen); 614 buf[namelen] = '\0'; 615 bcopy(data, buf + namelen + 1, datalen); 616 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) { 617 free(buf, M_KTRACE); 618 return; 619 } 620 req->ktr_buffer = buf; 621 req->ktr_header.ktr_len = buflen; 622 ktr_submitrequest(curthread, req); 623 } 624 #endif /* KTRACE */ 625 626 /* Interface and common routines */ 627 628 #ifndef _SYS_SYSPROTO_H_ 629 struct ktrace_args { 630 char *fname; 631 int ops; 632 int facs; 633 int pid; 634 }; 635 #endif 636 /* ARGSUSED */ 637 int 638 ktrace(td, uap) 639 struct thread *td; 640 register struct ktrace_args *uap; 641 { 642 #ifdef KTRACE 643 register struct vnode *vp = NULL; 644 register struct proc *p; 645 struct pgrp *pg; 646 int facs = uap->facs & ~KTRFAC_ROOT; 647 int ops = KTROP(uap->ops); 648 int descend = uap->ops & KTRFLAG_DESCEND; 649 int nfound, ret = 0; 650 int flags, error = 0, vfslocked; 651 struct nameidata nd; 652 struct ucred *cred; 653 654 /* 655 * Need something to (un)trace. 656 */ 657 if (ops != KTROP_CLEARFILE && facs == 0) 658 return (EINVAL); 659 660 ktrace_enter(td); 661 if (ops != KTROP_CLEAR) { 662 /* 663 * an operation which requires a file argument. 664 */ 665 NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_USERSPACE, 666 uap->fname, td); 667 flags = FREAD | FWRITE | O_NOFOLLOW; 668 error = vn_open(&nd, &flags, 0, NULL); 669 if (error) { 670 ktrace_exit(td); 671 return (error); 672 } 673 vfslocked = NDHASGIANT(&nd); 674 NDFREE(&nd, NDF_ONLY_PNBUF); 675 vp = nd.ni_vp; 676 VOP_UNLOCK(vp, 0); 677 if (vp->v_type != VREG) { 678 (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td); 679 VFS_UNLOCK_GIANT(vfslocked); 680 ktrace_exit(td); 681 return (EACCES); 682 } 683 VFS_UNLOCK_GIANT(vfslocked); 684 } 685 /* 686 * Clear all uses of the tracefile. 687 */ 688 if (ops == KTROP_CLEARFILE) { 689 int vrele_count; 690 691 vrele_count = 0; 692 sx_slock(&allproc_lock); 693 FOREACH_PROC_IN_SYSTEM(p) { 694 PROC_LOCK(p); 695 if (p->p_tracevp == vp) { 696 if (ktrcanset(td, p)) { 697 mtx_lock(&ktrace_mtx); 698 cred = p->p_tracecred; 699 p->p_tracecred = NULL; 700 p->p_tracevp = NULL; 701 p->p_traceflag = 0; 702 mtx_unlock(&ktrace_mtx); 703 vrele_count++; 704 crfree(cred); 705 } else 706 error = EPERM; 707 } 708 PROC_UNLOCK(p); 709 } 710 sx_sunlock(&allproc_lock); 711 if (vrele_count > 0) { 712 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 713 while (vrele_count-- > 0) 714 vrele(vp); 715 VFS_UNLOCK_GIANT(vfslocked); 716 } 717 goto done; 718 } 719 /* 720 * do it 721 */ 722 sx_slock(&proctree_lock); 723 if (uap->pid < 0) { 724 /* 725 * by process group 726 */ 727 pg = pgfind(-uap->pid); 728 if (pg == NULL) { 729 sx_sunlock(&proctree_lock); 730 error = ESRCH; 731 goto done; 732 } 733 /* 734 * ktrops() may call vrele(). Lock pg_members 735 * by the proctree_lock rather than pg_mtx. 736 */ 737 PGRP_UNLOCK(pg); 738 nfound = 0; 739 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 740 PROC_LOCK(p); 741 if (p_cansee(td, p) != 0) { 742 PROC_UNLOCK(p); 743 continue; 744 } 745 PROC_UNLOCK(p); 746 nfound++; 747 if (descend) 748 ret |= ktrsetchildren(td, p, ops, facs, vp); 749 else 750 ret |= ktrops(td, p, ops, facs, vp); 751 } 752 if (nfound == 0) { 753 sx_sunlock(&proctree_lock); 754 error = ESRCH; 755 goto done; 756 } 757 } else { 758 /* 759 * by pid 760 */ 761 p = pfind(uap->pid); 762 if (p == NULL) { 763 sx_sunlock(&proctree_lock); 764 error = ESRCH; 765 goto done; 766 } 767 error = p_cansee(td, p); 768 /* 769 * The slock of the proctree lock will keep this process 770 * from going away, so unlocking the proc here is ok. 771 */ 772 PROC_UNLOCK(p); 773 if (error) { 774 sx_sunlock(&proctree_lock); 775 goto done; 776 } 777 if (descend) 778 ret |= ktrsetchildren(td, p, ops, facs, vp); 779 else 780 ret |= ktrops(td, p, ops, facs, vp); 781 } 782 sx_sunlock(&proctree_lock); 783 if (!ret) 784 error = EPERM; 785 done: 786 if (vp != NULL) { 787 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 788 (void) vn_close(vp, FWRITE, td->td_ucred, td); 789 VFS_UNLOCK_GIANT(vfslocked); 790 } 791 ktrace_exit(td); 792 return (error); 793 #else /* !KTRACE */ 794 return (ENOSYS); 795 #endif /* KTRACE */ 796 } 797 798 /* ARGSUSED */ 799 int 800 utrace(td, uap) 801 struct thread *td; 802 register struct utrace_args *uap; 803 { 804 805 #ifdef KTRACE 806 struct ktr_request *req; 807 void *cp; 808 int error; 809 810 if (!KTRPOINT(td, KTR_USER)) 811 return (0); 812 if (uap->len > KTR_USER_MAXLEN) 813 return (EINVAL); 814 cp = malloc(uap->len, M_KTRACE, M_WAITOK); 815 error = copyin(uap->addr, cp, uap->len); 816 if (error) { 817 free(cp, M_KTRACE); 818 return (error); 819 } 820 req = ktr_getrequest(KTR_USER); 821 if (req == NULL) { 822 free(cp, M_KTRACE); 823 return (ENOMEM); 824 } 825 req->ktr_buffer = cp; 826 req->ktr_header.ktr_len = uap->len; 827 ktr_submitrequest(td, req); 828 return (0); 829 #else /* !KTRACE */ 830 return (ENOSYS); 831 #endif /* KTRACE */ 832 } 833 834 #ifdef KTRACE 835 static int 836 ktrops(td, p, ops, facs, vp) 837 struct thread *td; 838 struct proc *p; 839 int ops, facs; 840 struct vnode *vp; 841 { 842 struct vnode *tracevp = NULL; 843 struct ucred *tracecred = NULL; 844 845 PROC_LOCK(p); 846 if (!ktrcanset(td, p)) { 847 PROC_UNLOCK(p); 848 return (0); 849 } 850 mtx_lock(&ktrace_mtx); 851 if (ops == KTROP_SET) { 852 if (p->p_tracevp != vp) { 853 /* 854 * if trace file already in use, relinquish below 855 */ 856 tracevp = p->p_tracevp; 857 VREF(vp); 858 p->p_tracevp = vp; 859 } 860 if (p->p_tracecred != td->td_ucred) { 861 tracecred = p->p_tracecred; 862 p->p_tracecred = crhold(td->td_ucred); 863 } 864 p->p_traceflag |= facs; 865 if (priv_check(td, PRIV_KTRACE) == 0) 866 p->p_traceflag |= KTRFAC_ROOT; 867 } else { 868 /* KTROP_CLEAR */ 869 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) { 870 /* no more tracing */ 871 p->p_traceflag = 0; 872 tracevp = p->p_tracevp; 873 p->p_tracevp = NULL; 874 tracecred = p->p_tracecred; 875 p->p_tracecred = NULL; 876 } 877 } 878 mtx_unlock(&ktrace_mtx); 879 PROC_UNLOCK(p); 880 if (tracevp != NULL) { 881 int vfslocked; 882 883 vfslocked = VFS_LOCK_GIANT(tracevp->v_mount); 884 vrele(tracevp); 885 VFS_UNLOCK_GIANT(vfslocked); 886 } 887 if (tracecred != NULL) 888 crfree(tracecred); 889 890 return (1); 891 } 892 893 static int 894 ktrsetchildren(td, top, ops, facs, vp) 895 struct thread *td; 896 struct proc *top; 897 int ops, facs; 898 struct vnode *vp; 899 { 900 register struct proc *p; 901 register int ret = 0; 902 903 p = top; 904 sx_assert(&proctree_lock, SX_LOCKED); 905 for (;;) { 906 ret |= ktrops(td, p, ops, facs, vp); 907 /* 908 * If this process has children, descend to them next, 909 * otherwise do any siblings, and if done with this level, 910 * follow back up the tree (but not past top). 911 */ 912 if (!LIST_EMPTY(&p->p_children)) 913 p = LIST_FIRST(&p->p_children); 914 else for (;;) { 915 if (p == top) 916 return (ret); 917 if (LIST_NEXT(p, p_sibling)) { 918 p = LIST_NEXT(p, p_sibling); 919 break; 920 } 921 p = p->p_pptr; 922 } 923 } 924 /*NOTREACHED*/ 925 } 926 927 static void 928 ktr_writerequest(struct thread *td, struct ktr_request *req) 929 { 930 struct ktr_header *kth; 931 struct vnode *vp; 932 struct proc *p; 933 struct ucred *cred; 934 struct uio auio; 935 struct iovec aiov[3]; 936 struct mount *mp; 937 int datalen, buflen, vrele_count; 938 int error, vfslocked; 939 940 /* 941 * We hold the vnode and credential for use in I/O in case ktrace is 942 * disabled on the process as we write out the request. 943 * 944 * XXXRW: This is not ideal: we could end up performing a write after 945 * the vnode has been closed. 946 */ 947 mtx_lock(&ktrace_mtx); 948 vp = td->td_proc->p_tracevp; 949 cred = td->td_proc->p_tracecred; 950 951 /* 952 * If vp is NULL, the vp has been cleared out from under this 953 * request, so just drop it. Make sure the credential and vnode are 954 * in sync: we should have both or neither. 955 */ 956 if (vp == NULL) { 957 KASSERT(cred == NULL, ("ktr_writerequest: cred != NULL")); 958 mtx_unlock(&ktrace_mtx); 959 return; 960 } 961 VREF(vp); 962 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL")); 963 crhold(cred); 964 mtx_unlock(&ktrace_mtx); 965 966 kth = &req->ktr_header; 967 KASSERT(((u_short)kth->ktr_type & ~KTR_DROP) < 968 sizeof(data_lengths) / sizeof(data_lengths[0]), 969 ("data_lengths array overflow")); 970 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP]; 971 buflen = kth->ktr_len; 972 auio.uio_iov = &aiov[0]; 973 auio.uio_offset = 0; 974 auio.uio_segflg = UIO_SYSSPACE; 975 auio.uio_rw = UIO_WRITE; 976 aiov[0].iov_base = (caddr_t)kth; 977 aiov[0].iov_len = sizeof(struct ktr_header); 978 auio.uio_resid = sizeof(struct ktr_header); 979 auio.uio_iovcnt = 1; 980 auio.uio_td = td; 981 if (datalen != 0) { 982 aiov[1].iov_base = (caddr_t)&req->ktr_data; 983 aiov[1].iov_len = datalen; 984 auio.uio_resid += datalen; 985 auio.uio_iovcnt++; 986 kth->ktr_len += datalen; 987 } 988 if (buflen != 0) { 989 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write")); 990 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer; 991 aiov[auio.uio_iovcnt].iov_len = buflen; 992 auio.uio_resid += buflen; 993 auio.uio_iovcnt++; 994 } 995 996 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 997 vn_start_write(vp, &mp, V_WAIT); 998 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 999 #ifdef MAC 1000 error = mac_vnode_check_write(cred, NOCRED, vp); 1001 if (error == 0) 1002 #endif 1003 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred); 1004 VOP_UNLOCK(vp, 0); 1005 vn_finished_write(mp); 1006 crfree(cred); 1007 if (!error) { 1008 vrele(vp); 1009 VFS_UNLOCK_GIANT(vfslocked); 1010 return; 1011 } 1012 VFS_UNLOCK_GIANT(vfslocked); 1013 1014 /* 1015 * If error encountered, give up tracing on this vnode. We defer 1016 * all the vrele()'s on the vnode until after we are finished walking 1017 * the various lists to avoid needlessly holding locks. 1018 * NB: at this point we still hold the vnode reference that must 1019 * not go away as we need the valid vnode to compare with. Thus let 1020 * vrele_count start at 1 and the reference will be freed 1021 * by the loop at the end after our last use of vp. 1022 */ 1023 log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n", 1024 error); 1025 vrele_count = 1; 1026 /* 1027 * First, clear this vnode from being used by any processes in the 1028 * system. 1029 * XXX - If one process gets an EPERM writing to the vnode, should 1030 * we really do this? Other processes might have suitable 1031 * credentials for the operation. 1032 */ 1033 cred = NULL; 1034 sx_slock(&allproc_lock); 1035 FOREACH_PROC_IN_SYSTEM(p) { 1036 PROC_LOCK(p); 1037 if (p->p_tracevp == vp) { 1038 mtx_lock(&ktrace_mtx); 1039 p->p_tracevp = NULL; 1040 p->p_traceflag = 0; 1041 cred = p->p_tracecred; 1042 p->p_tracecred = NULL; 1043 mtx_unlock(&ktrace_mtx); 1044 vrele_count++; 1045 } 1046 PROC_UNLOCK(p); 1047 if (cred != NULL) { 1048 crfree(cred); 1049 cred = NULL; 1050 } 1051 } 1052 sx_sunlock(&allproc_lock); 1053 1054 /* 1055 * We can't clear any pending requests in threads that have cached 1056 * them but not yet committed them, as those are per-thread. The 1057 * thread will have to clear it itself on system call return. 1058 */ 1059 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 1060 while (vrele_count-- > 0) 1061 vrele(vp); 1062 VFS_UNLOCK_GIANT(vfslocked); 1063 } 1064 1065 /* 1066 * Return true if caller has permission to set the ktracing state 1067 * of target. Essentially, the target can't possess any 1068 * more permissions than the caller. KTRFAC_ROOT signifies that 1069 * root previously set the tracing status on the target process, and 1070 * so, only root may further change it. 1071 */ 1072 static int 1073 ktrcanset(td, targetp) 1074 struct thread *td; 1075 struct proc *targetp; 1076 { 1077 1078 PROC_LOCK_ASSERT(targetp, MA_OWNED); 1079 if (targetp->p_traceflag & KTRFAC_ROOT && 1080 priv_check(td, PRIV_KTRACE)) 1081 return (0); 1082 1083 if (p_candebug(td, targetp) != 0) 1084 return (0); 1085 1086 return (1); 1087 } 1088 1089 #endif /* KTRACE */ 1090