2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2005 Robert N. M. Watson
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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.
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
31 * @(#)kern_ktrace.c 8.2 (Berkeley) 9/23/93
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_ktrace.h"
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>
45 #include <sys/mutex.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/namei.h>
51 #include <sys/unistd.h>
52 #include <sys/vnode.h>
53 #include <sys/socket.h>
55 #include <sys/ktrace.h>
57 #include <sys/sysctl.h>
58 #include <sys/syslog.h>
59 #include <sys/sysproto.h>
61 #include <security/mac/mac_framework.h>
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.
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.
82 static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE");
86 #ifndef KTRACE_REQUEST_POOL
87 #define KTRACE_REQUEST_POOL 100
91 struct ktr_header ktr_header;
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;
100 STAILQ_ENTRY(ktr_request) ktr_list;
103 static int data_lengths[] = {
105 offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */
106 sizeof(struct ktr_sysret), /* KTR_SYSRET */
108 sizeof(struct ktr_genio), /* KTR_GENIO */
109 sizeof(struct ktr_psig), /* KTR_PSIG */
110 sizeof(struct ktr_csw), /* KTR_CSW */
116 static STAILQ_HEAD(, ktr_request) ktr_free;
118 static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options");
120 static u_int ktr_requestpool = KTRACE_REQUEST_POOL;
121 TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool);
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");
128 static int print_message = 1;
129 static struct mtx ktrace_mtx;
130 static struct sx ktrace_sx;
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_freeproc(struct proc *p, struct ucred **uc,
139 static void ktr_freerequest(struct ktr_request *req);
140 static void ktr_freerequest_locked(struct ktr_request *req);
141 static void ktr_writerequest(struct thread *td, struct ktr_request *req);
142 static int ktrcanset(struct thread *,struct proc *);
143 static int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *);
144 static int ktrops(struct thread *,struct proc *,int,int,struct vnode *);
147 * ktrace itself generates events, such as context switches, which we do not
148 * wish to trace. Maintain a flag, TDP_INKTRACE, on each thread to determine
149 * whether or not it is in a region where tracing of events should be
153 ktrace_enter(struct thread *td)
156 KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set"));
157 td->td_pflags |= TDP_INKTRACE;
161 ktrace_exit(struct thread *td)
164 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set"));
165 td->td_pflags &= ~TDP_INKTRACE;
169 ktrace_assert(struct thread *td)
172 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set"));
176 ktrace_init(void *dummy)
178 struct ktr_request *req;
181 mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET);
182 sx_init(&ktrace_sx, "ktrace_sx");
183 STAILQ_INIT(&ktr_free);
184 for (i = 0; i < ktr_requestpool; i++) {
185 req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK);
186 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
189 SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL);
192 sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)
195 u_int newsize, oldsize, wantsize;
198 /* Handle easy read-only case first to avoid warnings from GCC. */
200 mtx_lock(&ktrace_mtx);
201 oldsize = ktr_requestpool;
202 mtx_unlock(&ktrace_mtx);
203 return (SYSCTL_OUT(req, &oldsize, sizeof(u_int)));
206 error = SYSCTL_IN(req, &wantsize, sizeof(u_int));
211 mtx_lock(&ktrace_mtx);
212 oldsize = ktr_requestpool;
213 newsize = ktrace_resize_pool(wantsize);
214 mtx_unlock(&ktrace_mtx);
216 error = SYSCTL_OUT(req, &oldsize, sizeof(u_int));
219 if (wantsize > oldsize && newsize < wantsize)
223 SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW,
224 &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU",
225 "Pool buffer size for ktrace(1)");
228 ktrace_resize_pool(u_int newsize)
230 struct ktr_request *req;
233 mtx_assert(&ktrace_mtx, MA_OWNED);
235 bound = newsize - ktr_requestpool;
237 return (ktr_requestpool);
239 /* Shrink pool down to newsize if possible. */
240 while (bound++ < 0) {
241 req = STAILQ_FIRST(&ktr_free);
243 return (ktr_requestpool);
244 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
246 mtx_unlock(&ktrace_mtx);
248 mtx_lock(&ktrace_mtx);
251 /* Grow pool up to newsize. */
252 while (bound-- > 0) {
253 mtx_unlock(&ktrace_mtx);
254 req = malloc(sizeof(struct ktr_request), M_KTRACE,
256 mtx_lock(&ktrace_mtx);
257 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
260 return (ktr_requestpool);
263 /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */
264 CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) ==
265 (sizeof((struct thread *)NULL)->td_name));
267 static struct ktr_request *
268 ktr_getrequest(int type)
270 struct ktr_request *req;
271 struct thread *td = curthread;
272 struct proc *p = td->td_proc;
275 ktrace_enter(td); /* XXX: In caller instead? */
276 mtx_lock(&ktrace_mtx);
277 if (!KTRCHECK(td, type)) {
278 mtx_unlock(&ktrace_mtx);
282 req = STAILQ_FIRST(&ktr_free);
284 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
285 req->ktr_header.ktr_type = type;
286 if (p->p_traceflag & KTRFAC_DROP) {
287 req->ktr_header.ktr_type |= KTR_DROP;
288 p->p_traceflag &= ~KTRFAC_DROP;
290 mtx_unlock(&ktrace_mtx);
291 microtime(&req->ktr_header.ktr_time);
292 req->ktr_header.ktr_pid = p->p_pid;
293 req->ktr_header.ktr_tid = td->td_tid;
294 bcopy(td->td_name, req->ktr_header.ktr_comm,
295 sizeof(req->ktr_header.ktr_comm));
296 req->ktr_buffer = NULL;
297 req->ktr_header.ktr_len = 0;
299 p->p_traceflag |= KTRFAC_DROP;
302 mtx_unlock(&ktrace_mtx);
304 printf("Out of ktrace request objects.\n");
311 * Some trace generation environments don't permit direct access to VFS,
312 * such as during a context switch where sleeping is not allowed. Under these
313 * circumstances, queue a request to the thread to be written asynchronously
317 ktr_enqueuerequest(struct thread *td, struct ktr_request *req)
320 mtx_lock(&ktrace_mtx);
321 STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list);
322 mtx_unlock(&ktrace_mtx);
327 * Drain any pending ktrace records from the per-thread queue to disk. This
328 * is used both internally before committing other records, and also on
329 * system call return. We drain all the ones we can find at the time when
330 * drain is requested, but don't keep draining after that as those events
331 * may be approximately "after" the current event.
334 ktr_drain(struct thread *td)
336 struct ktr_request *queued_req;
337 STAILQ_HEAD(, ktr_request) local_queue;
340 sx_assert(&ktrace_sx, SX_XLOCKED);
342 STAILQ_INIT(&local_queue);
344 if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) {
345 mtx_lock(&ktrace_mtx);
346 STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr);
347 mtx_unlock(&ktrace_mtx);
349 while ((queued_req = STAILQ_FIRST(&local_queue))) {
350 STAILQ_REMOVE_HEAD(&local_queue, ktr_list);
351 ktr_writerequest(td, queued_req);
352 ktr_freerequest(queued_req);
358 * Submit a trace record for immediate commit to disk -- to be used only
359 * where entering VFS is OK. First drain any pending records that may have
360 * been cached in the thread.
363 ktr_submitrequest(struct thread *td, struct ktr_request *req)
368 sx_xlock(&ktrace_sx);
370 ktr_writerequest(td, req);
371 ktr_freerequest(req);
372 sx_xunlock(&ktrace_sx);
378 ktr_freerequest(struct ktr_request *req)
381 mtx_lock(&ktrace_mtx);
382 ktr_freerequest_locked(req);
383 mtx_unlock(&ktrace_mtx);
387 ktr_freerequest_locked(struct ktr_request *req)
390 mtx_assert(&ktrace_mtx, MA_OWNED);
391 if (req->ktr_buffer != NULL)
392 free(req->ktr_buffer, M_KTRACE);
393 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
397 * Disable tracing for a process and release all associated resources.
398 * The caller is responsible for releasing a reference on the returned
399 * vnode and credentials.
402 ktr_freeproc(struct proc *p, struct ucred **uc, struct vnode **vp)
404 struct ktr_request *req;
406 PROC_LOCK_ASSERT(p, MA_OWNED);
407 mtx_assert(&ktrace_mtx, MA_OWNED);
408 *uc = p->p_tracecred;
409 p->p_tracecred = NULL;
414 while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) {
415 STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list);
416 ktr_freerequest_locked(req);
421 ktrsyscall(code, narg, args)
425 struct ktr_request *req;
426 struct ktr_syscall *ktp;
430 buflen = sizeof(register_t) * narg;
432 buf = malloc(buflen, M_KTRACE, M_WAITOK);
433 bcopy(args, buf, buflen);
435 req = ktr_getrequest(KTR_SYSCALL);
441 ktp = &req->ktr_data.ktr_syscall;
442 ktp->ktr_code = code;
443 ktp->ktr_narg = narg;
445 req->ktr_header.ktr_len = buflen;
446 req->ktr_buffer = buf;
448 ktr_submitrequest(curthread, req);
452 ktrsysret(code, error, retval)
456 struct ktr_request *req;
457 struct ktr_sysret *ktp;
459 req = ktr_getrequest(KTR_SYSRET);
462 ktp = &req->ktr_data.ktr_sysret;
463 ktp->ktr_code = code;
464 ktp->ktr_error = error;
465 ktp->ktr_retval = retval; /* what about val2 ? */
466 ktr_submitrequest(curthread, req);
470 * When a setuid process execs, disable tracing.
472 * XXX: We toss any pending asynchronous records.
475 ktrprocexec(struct proc *p, struct ucred **uc, struct vnode **vp)
478 PROC_LOCK_ASSERT(p, MA_OWNED);
479 mtx_lock(&ktrace_mtx);
480 ktr_freeproc(p, uc, vp);
481 mtx_unlock(&ktrace_mtx);
485 * When a process exits, drain per-process asynchronous trace records
486 * and disable tracing.
489 ktrprocexit(struct thread *td)
497 if (p->p_traceflag == 0)
501 sx_xlock(&ktrace_sx);
503 sx_xunlock(&ktrace_sx);
505 mtx_lock(&ktrace_mtx);
506 ktr_freeproc(p, &cred, &vp);
507 mtx_unlock(&ktrace_mtx);
510 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
512 VFS_UNLOCK_GIANT(vfslocked);
520 * When a process forks, enable tracing in the new process if needed.
523 ktrprocfork(struct proc *p1, struct proc *p2)
526 PROC_LOCK_ASSERT(p1, MA_OWNED);
527 PROC_LOCK_ASSERT(p2, MA_OWNED);
528 mtx_lock(&ktrace_mtx);
529 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
530 if (p1->p_traceflag & KTRFAC_INHERIT) {
531 p2->p_traceflag = p1->p_traceflag;
532 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
534 KASSERT(p1->p_tracecred != NULL,
535 ("ktrace vnode with no cred"));
536 p2->p_tracecred = crhold(p1->p_tracecred);
539 mtx_unlock(&ktrace_mtx);
543 * When a thread returns, drain any asynchronous records generated by the
547 ktruserret(struct thread *td)
551 sx_xlock(&ktrace_sx);
553 sx_xunlock(&ktrace_sx);
561 struct ktr_request *req;
565 namelen = strlen(path);
567 buf = malloc(namelen, M_KTRACE, M_WAITOK);
568 bcopy(path, buf, namelen);
570 req = ktr_getrequest(KTR_NAMEI);
577 req->ktr_header.ktr_len = namelen;
578 req->ktr_buffer = buf;
580 ktr_submitrequest(curthread, req);
584 ktrsysctl(name, namelen)
588 struct ktr_request *req;
589 u_int mib[CTL_MAXNAME + 2];
594 /* Lookup name of mib. */
595 KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long"));
598 bcopy(name, mib + 2, namelen * sizeof(*name));
600 mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK);
601 error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen,
602 NULL, 0, &mibnamelen, 0);
604 free(mibname, M_KTRACE);
607 req = ktr_getrequest(KTR_SYSCTL);
609 free(mibname, M_KTRACE);
612 req->ktr_header.ktr_len = mibnamelen;
613 req->ktr_buffer = mibname;
614 ktr_submitrequest(curthread, req);
618 ktrgenio(fd, rw, uio, error)
624 struct ktr_request *req;
625 struct ktr_genio *ktg;
634 uio->uio_rw = UIO_WRITE;
635 datalen = imin(uio->uio_resid, ktr_geniosize);
636 buf = malloc(datalen, M_KTRACE, M_WAITOK);
637 error = uiomove(buf, datalen, uio);
643 req = ktr_getrequest(KTR_GENIO);
648 ktg = &req->ktr_data.ktr_genio;
651 req->ktr_header.ktr_len = datalen;
652 req->ktr_buffer = buf;
653 ktr_submitrequest(curthread, req);
657 ktrpsig(sig, action, mask, code)
663 struct ktr_request *req;
666 req = ktr_getrequest(KTR_PSIG);
669 kp = &req->ktr_data.ktr_psig;
670 kp->signo = (char)sig;
674 ktr_enqueuerequest(curthread, req);
681 struct ktr_request *req;
684 req = ktr_getrequest(KTR_CSW);
687 kc = &req->ktr_data.ktr_csw;
690 ktr_enqueuerequest(curthread, req);
694 ktrstruct(name, data, datalen)
699 struct ktr_request *req;
705 buflen = strlen(name) + 1 + datalen;
706 buf = malloc(buflen, M_KTRACE, M_WAITOK);
708 bcopy(data, buf + strlen(name) + 1, datalen);
709 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) {
713 req->ktr_buffer = buf;
714 req->ktr_header.ktr_len = buflen;
715 ktr_submitrequest(curthread, req);
719 /* Interface and common routines */
721 #ifndef _SYS_SYSPROTO_H_
733 register struct ktrace_args *uap;
736 register struct vnode *vp = NULL;
737 register struct proc *p;
739 int facs = uap->facs & ~KTRFAC_ROOT;
740 int ops = KTROP(uap->ops);
741 int descend = uap->ops & KTRFLAG_DESCEND;
743 int flags, error = 0, vfslocked;
748 * Need something to (un)trace.
750 if (ops != KTROP_CLEARFILE && facs == 0)
754 if (ops != KTROP_CLEAR) {
756 * an operation which requires a file argument.
758 NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_USERSPACE,
760 flags = FREAD | FWRITE | O_NOFOLLOW;
761 error = vn_open(&nd, &flags, 0, NULL);
766 vfslocked = NDHASGIANT(&nd);
767 NDFREE(&nd, NDF_ONLY_PNBUF);
770 if (vp->v_type != VREG) {
771 (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
772 VFS_UNLOCK_GIANT(vfslocked);
776 VFS_UNLOCK_GIANT(vfslocked);
779 * Clear all uses of the tracefile.
781 if (ops == KTROP_CLEARFILE) {
785 sx_slock(&allproc_lock);
786 FOREACH_PROC_IN_SYSTEM(p) {
788 if (p->p_tracevp == vp) {
789 if (ktrcanset(td, p)) {
790 mtx_lock(&ktrace_mtx);
791 ktr_freeproc(p, &cred, NULL);
792 mtx_unlock(&ktrace_mtx);
800 sx_sunlock(&allproc_lock);
801 if (vrele_count > 0) {
802 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
803 while (vrele_count-- > 0)
805 VFS_UNLOCK_GIANT(vfslocked);
812 sx_slock(&proctree_lock);
817 pg = pgfind(-uap->pid);
819 sx_sunlock(&proctree_lock);
824 * ktrops() may call vrele(). Lock pg_members
825 * by the proctree_lock rather than pg_mtx.
829 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
831 if (p_cansee(td, p) != 0) {
837 ret |= ktrsetchildren(td, p, ops, facs, vp);
839 ret |= ktrops(td, p, ops, facs, vp);
842 sx_sunlock(&proctree_lock);
854 error = p_cansee(td, p);
858 sx_sunlock(&proctree_lock);
862 ret |= ktrsetchildren(td, p, ops, facs, vp);
864 ret |= ktrops(td, p, ops, facs, vp);
866 sx_sunlock(&proctree_lock);
871 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
872 (void) vn_close(vp, FWRITE, td->td_ucred, td);
873 VFS_UNLOCK_GIANT(vfslocked);
886 register struct utrace_args *uap;
890 struct ktr_request *req;
894 if (!KTRPOINT(td, KTR_USER))
896 if (uap->len > KTR_USER_MAXLEN)
898 cp = malloc(uap->len, M_KTRACE, M_WAITOK);
899 error = copyin(uap->addr, cp, uap->len);
904 req = ktr_getrequest(KTR_USER);
909 req->ktr_buffer = cp;
910 req->ktr_header.ktr_len = uap->len;
911 ktr_submitrequest(td, req);
920 ktrops(td, p, ops, facs, vp)
926 struct vnode *tracevp = NULL;
927 struct ucred *tracecred = NULL;
929 PROC_LOCK_ASSERT(p, MA_OWNED);
930 if (!ktrcanset(td, p)) {
934 if (p->p_flag & P_WEXIT) {
935 /* If the process is exiting, just ignore it. */
939 mtx_lock(&ktrace_mtx);
940 if (ops == KTROP_SET) {
941 if (p->p_tracevp != vp) {
943 * if trace file already in use, relinquish below
945 tracevp = p->p_tracevp;
949 if (p->p_tracecred != td->td_ucred) {
950 tracecred = p->p_tracecred;
951 p->p_tracecred = crhold(td->td_ucred);
953 p->p_traceflag |= facs;
954 if (priv_check(td, PRIV_KTRACE) == 0)
955 p->p_traceflag |= KTRFAC_ROOT;
958 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0)
959 /* no more tracing */
960 ktr_freeproc(p, &tracecred, &tracevp);
962 mtx_unlock(&ktrace_mtx);
964 if (tracevp != NULL) {
967 vfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
969 VFS_UNLOCK_GIANT(vfslocked);
971 if (tracecred != NULL)
978 ktrsetchildren(td, top, ops, facs, vp)
984 register struct proc *p;
985 register int ret = 0;
988 PROC_LOCK_ASSERT(p, MA_OWNED);
989 sx_assert(&proctree_lock, SX_LOCKED);
991 ret |= ktrops(td, p, ops, facs, vp);
993 * If this process has children, descend to them next,
994 * otherwise do any siblings, and if done with this level,
995 * follow back up the tree (but not past top).
997 if (!LIST_EMPTY(&p->p_children))
998 p = LIST_FIRST(&p->p_children);
1002 if (LIST_NEXT(p, p_sibling)) {
1003 p = LIST_NEXT(p, p_sibling);
1014 ktr_writerequest(struct thread *td, struct ktr_request *req)
1016 struct ktr_header *kth;
1021 struct iovec aiov[3];
1023 int datalen, buflen, vrele_count;
1024 int error, vfslocked;
1027 * We hold the vnode and credential for use in I/O in case ktrace is
1028 * disabled on the process as we write out the request.
1030 * XXXRW: This is not ideal: we could end up performing a write after
1031 * the vnode has been closed.
1033 mtx_lock(&ktrace_mtx);
1034 vp = td->td_proc->p_tracevp;
1035 cred = td->td_proc->p_tracecred;
1038 * If vp is NULL, the vp has been cleared out from under this
1039 * request, so just drop it. Make sure the credential and vnode are
1040 * in sync: we should have both or neither.
1043 KASSERT(cred == NULL, ("ktr_writerequest: cred != NULL"));
1044 mtx_unlock(&ktrace_mtx);
1048 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL"));
1050 mtx_unlock(&ktrace_mtx);
1052 kth = &req->ktr_header;
1053 KASSERT(((u_short)kth->ktr_type & ~KTR_DROP) <
1054 sizeof(data_lengths) / sizeof(data_lengths[0]),
1055 ("data_lengths array overflow"));
1056 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP];
1057 buflen = kth->ktr_len;
1058 auio.uio_iov = &aiov[0];
1059 auio.uio_offset = 0;
1060 auio.uio_segflg = UIO_SYSSPACE;
1061 auio.uio_rw = UIO_WRITE;
1062 aiov[0].iov_base = (caddr_t)kth;
1063 aiov[0].iov_len = sizeof(struct ktr_header);
1064 auio.uio_resid = sizeof(struct ktr_header);
1065 auio.uio_iovcnt = 1;
1068 aiov[1].iov_base = (caddr_t)&req->ktr_data;
1069 aiov[1].iov_len = datalen;
1070 auio.uio_resid += datalen;
1072 kth->ktr_len += datalen;
1075 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write"));
1076 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer;
1077 aiov[auio.uio_iovcnt].iov_len = buflen;
1078 auio.uio_resid += buflen;
1082 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1083 vn_start_write(vp, &mp, V_WAIT);
1084 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1086 error = mac_vnode_check_write(cred, NOCRED, vp);
1089 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred);
1091 vn_finished_write(mp);
1095 VFS_UNLOCK_GIANT(vfslocked);
1098 VFS_UNLOCK_GIANT(vfslocked);
1101 * If error encountered, give up tracing on this vnode. We defer
1102 * all the vrele()'s on the vnode until after we are finished walking
1103 * the various lists to avoid needlessly holding locks.
1104 * NB: at this point we still hold the vnode reference that must
1105 * not go away as we need the valid vnode to compare with. Thus let
1106 * vrele_count start at 1 and the reference will be freed
1107 * by the loop at the end after our last use of vp.
1109 log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n",
1113 * First, clear this vnode from being used by any processes in the
1115 * XXX - If one process gets an EPERM writing to the vnode, should
1116 * we really do this? Other processes might have suitable
1117 * credentials for the operation.
1120 sx_slock(&allproc_lock);
1121 FOREACH_PROC_IN_SYSTEM(p) {
1123 if (p->p_tracevp == vp) {
1124 mtx_lock(&ktrace_mtx);
1125 ktr_freeproc(p, &cred, NULL);
1126 mtx_unlock(&ktrace_mtx);
1135 sx_sunlock(&allproc_lock);
1137 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1138 while (vrele_count-- > 0)
1140 VFS_UNLOCK_GIANT(vfslocked);
1144 * Return true if caller has permission to set the ktracing state
1145 * of target. Essentially, the target can't possess any
1146 * more permissions than the caller. KTRFAC_ROOT signifies that
1147 * root previously set the tracing status on the target process, and
1148 * so, only root may further change it.
1151 ktrcanset(td, targetp)
1153 struct proc *targetp;
1156 PROC_LOCK_ASSERT(targetp, MA_OWNED);
1157 if (targetp->p_traceflag & KTRFAC_ROOT &&
1158 priv_check(td, PRIV_KTRACE))
1161 if (p_candebug(td, targetp) != 0)