2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2005 Robert N. M. Watson
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)kern_ktrace.c 8.2 (Berkeley) 9/23/93
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include "opt_ktrace.h"
41 #include <sys/param.h>
42 #include <sys/capsicum.h>
43 #include <sys/systm.h>
44 #include <sys/fcntl.h>
45 #include <sys/kernel.h>
46 #include <sys/kthread.h>
48 #include <sys/mutex.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/namei.h>
54 #include <sys/resourcevar.h>
55 #include <sys/unistd.h>
56 #include <sys/vnode.h>
57 #include <sys/socket.h>
59 #include <sys/ktrace.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysent.h>
63 #include <sys/syslog.h>
64 #include <sys/sysproto.h>
66 #include <security/mac/mac_framework.h>
69 * The ktrace facility allows the tracing of certain key events in user space
70 * processes, such as system calls, signal delivery, context switches, and
71 * user generated events using utrace(2). It works by streaming event
72 * records and data to a vnode associated with the process using the
73 * ktrace(2) system call. In general, records can be written directly from
74 * the context that generates the event. One important exception to this is
75 * during a context switch, where sleeping is not permitted. To handle this
76 * case, trace events are generated using in-kernel ktr_request records, and
77 * then delivered to disk at a convenient moment -- either immediately, the
78 * next traceable event, at system call return, or at process exit.
80 * When dealing with multiple threads or processes writing to the same event
81 * log, ordering guarantees are weak: specifically, if an event has multiple
82 * records (i.e., system call enter and return), they may be interlaced with
83 * records from another event. Process and thread ID information is provided
84 * in the record, and user applications can de-interlace events if required.
87 static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE");
91 FEATURE(ktrace, "Kernel support for system-call tracing");
93 #ifndef KTRACE_REQUEST_POOL
94 #define KTRACE_REQUEST_POOL 100
98 struct ktr_header ktr_header;
101 struct ktr_proc_ctor ktr_proc_ctor;
102 struct ktr_cap_fail ktr_cap_fail;
103 struct ktr_syscall ktr_syscall;
104 struct ktr_sysret ktr_sysret;
105 struct ktr_genio ktr_genio;
106 struct ktr_psig ktr_psig;
107 struct ktr_csw ktr_csw;
108 struct ktr_fault ktr_fault;
109 struct ktr_faultend ktr_faultend;
110 struct ktr_struct_array ktr_struct_array;
112 STAILQ_ENTRY(ktr_request) ktr_list;
115 static int data_lengths[] = {
116 [KTR_SYSCALL] = offsetof(struct ktr_syscall, ktr_args),
117 [KTR_SYSRET] = sizeof(struct ktr_sysret),
119 [KTR_GENIO] = sizeof(struct ktr_genio),
120 [KTR_PSIG] = sizeof(struct ktr_psig),
121 [KTR_CSW] = sizeof(struct ktr_csw),
125 [KTR_PROCCTOR] = sizeof(struct ktr_proc_ctor),
127 [KTR_CAPFAIL] = sizeof(struct ktr_cap_fail),
128 [KTR_FAULT] = sizeof(struct ktr_fault),
129 [KTR_FAULTEND] = sizeof(struct ktr_faultend),
130 [KTR_STRUCT_ARRAY] = sizeof(struct ktr_struct_array),
133 static STAILQ_HEAD(, ktr_request) ktr_free;
135 static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
138 static u_int ktr_requestpool = KTRACE_REQUEST_POOL;
139 TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool);
141 u_int ktr_geniosize = PAGE_SIZE;
142 SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RWTUN, &ktr_geniosize,
143 0, "Maximum size of genio event payload");
146 * Allow to not to send signal to traced process, in which context the
147 * ktr record is written. The limit is applied from the process that
148 * set up ktrace, so killing the traced process is not completely fair.
150 int ktr_filesize_limit_signal = 0;
151 SYSCTL_INT(_kern_ktrace, OID_AUTO, filesize_limit_signal, CTLFLAG_RWTUN,
152 &ktr_filesize_limit_signal, 0,
153 "Send SIGXFSZ to the traced process when the log size limit is exceeded");
155 static int print_message = 1;
156 static struct mtx ktrace_mtx;
157 static struct sx ktrace_sx;
159 struct ktr_io_params {
166 static void ktrace_init(void *dummy);
167 static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS);
168 static u_int ktrace_resize_pool(u_int oldsize, u_int newsize);
169 static struct ktr_request *ktr_getrequest_entered(struct thread *td, int type);
170 static struct ktr_request *ktr_getrequest(int type);
171 static void ktr_submitrequest(struct thread *td, struct ktr_request *req);
172 static struct ktr_io_params *ktr_freeproc(struct proc *p);
173 static void ktr_freerequest(struct ktr_request *req);
174 static void ktr_freerequest_locked(struct ktr_request *req);
175 static void ktr_writerequest(struct thread *td, struct ktr_request *req);
176 static int ktrcanset(struct thread *,struct proc *);
177 static int ktrsetchildren(struct thread *, struct proc *, int, int,
178 struct ktr_io_params *);
179 static int ktrops(struct thread *, struct proc *, int, int,
180 struct ktr_io_params *);
181 static void ktrprocctor_entered(struct thread *, struct proc *);
184 * ktrace itself generates events, such as context switches, which we do not
185 * wish to trace. Maintain a flag, TDP_INKTRACE, on each thread to determine
186 * whether or not it is in a region where tracing of events should be
190 ktrace_enter(struct thread *td)
193 KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set"));
194 td->td_pflags |= TDP_INKTRACE;
198 ktrace_exit(struct thread *td)
201 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set"));
202 td->td_pflags &= ~TDP_INKTRACE;
206 ktrace_assert(struct thread *td)
209 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set"));
213 ktrace_init(void *dummy)
215 struct ktr_request *req;
218 mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET);
219 sx_init(&ktrace_sx, "ktrace_sx");
220 STAILQ_INIT(&ktr_free);
221 for (i = 0; i < ktr_requestpool; i++) {
222 req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK |
224 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
227 SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL);
230 sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)
233 u_int newsize, oldsize, wantsize;
236 /* Handle easy read-only case first to avoid warnings from GCC. */
238 oldsize = ktr_requestpool;
239 return (SYSCTL_OUT(req, &oldsize, sizeof(u_int)));
242 error = SYSCTL_IN(req, &wantsize, sizeof(u_int));
247 oldsize = ktr_requestpool;
248 newsize = ktrace_resize_pool(oldsize, wantsize);
250 error = SYSCTL_OUT(req, &oldsize, sizeof(u_int));
253 if (wantsize > oldsize && newsize < wantsize)
257 SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool,
258 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &ktr_requestpool, 0,
259 sysctl_kern_ktrace_request_pool, "IU",
260 "Pool buffer size for ktrace(1)");
263 ktrace_resize_pool(u_int oldsize, u_int newsize)
265 STAILQ_HEAD(, ktr_request) ktr_new;
266 struct ktr_request *req;
270 bound = newsize - oldsize;
272 return (ktr_requestpool);
274 mtx_lock(&ktrace_mtx);
275 /* Shrink pool down to newsize if possible. */
276 while (bound++ < 0) {
277 req = STAILQ_FIRST(&ktr_free);
280 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
285 /* Grow pool up to newsize. */
286 STAILQ_INIT(&ktr_new);
287 while (bound-- > 0) {
288 req = malloc(sizeof(struct ktr_request), M_KTRACE,
290 STAILQ_INSERT_HEAD(&ktr_new, req, ktr_list);
292 mtx_lock(&ktrace_mtx);
293 STAILQ_CONCAT(&ktr_free, &ktr_new);
294 ktr_requestpool += (newsize - oldsize);
296 mtx_unlock(&ktrace_mtx);
297 return (ktr_requestpool);
300 /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */
301 CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) ==
302 (sizeof((struct thread *)NULL)->td_name));
304 static struct ktr_request *
305 ktr_getrequest_entered(struct thread *td, int type)
307 struct ktr_request *req;
308 struct proc *p = td->td_proc;
311 mtx_lock(&ktrace_mtx);
312 if (!KTRCHECK(td, type)) {
313 mtx_unlock(&ktrace_mtx);
316 req = STAILQ_FIRST(&ktr_free);
318 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
319 req->ktr_header.ktr_type = type;
320 if (p->p_traceflag & KTRFAC_DROP) {
321 req->ktr_header.ktr_type |= KTR_DROP;
322 p->p_traceflag &= ~KTRFAC_DROP;
324 mtx_unlock(&ktrace_mtx);
325 microtime(&req->ktr_header.ktr_time);
326 req->ktr_header.ktr_pid = p->p_pid;
327 req->ktr_header.ktr_tid = td->td_tid;
328 bcopy(td->td_name, req->ktr_header.ktr_comm,
329 sizeof(req->ktr_header.ktr_comm));
330 req->ktr_buffer = NULL;
331 req->ktr_header.ktr_len = 0;
333 p->p_traceflag |= KTRFAC_DROP;
336 mtx_unlock(&ktrace_mtx);
338 printf("Out of ktrace request objects.\n");
343 static struct ktr_request *
344 ktr_getrequest(int type)
346 struct thread *td = curthread;
347 struct ktr_request *req;
350 req = ktr_getrequest_entered(td, type);
358 * Some trace generation environments don't permit direct access to VFS,
359 * such as during a context switch where sleeping is not allowed. Under these
360 * circumstances, queue a request to the thread to be written asynchronously
364 ktr_enqueuerequest(struct thread *td, struct ktr_request *req)
367 mtx_lock(&ktrace_mtx);
368 STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list);
369 mtx_unlock(&ktrace_mtx);
371 td->td_flags |= TDF_ASTPENDING;
376 * Drain any pending ktrace records from the per-thread queue to disk. This
377 * is used both internally before committing other records, and also on
378 * system call return. We drain all the ones we can find at the time when
379 * drain is requested, but don't keep draining after that as those events
380 * may be approximately "after" the current event.
383 ktr_drain(struct thread *td)
385 struct ktr_request *queued_req;
386 STAILQ_HEAD(, ktr_request) local_queue;
389 sx_assert(&ktrace_sx, SX_XLOCKED);
391 STAILQ_INIT(&local_queue);
393 if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) {
394 mtx_lock(&ktrace_mtx);
395 STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr);
396 mtx_unlock(&ktrace_mtx);
398 while ((queued_req = STAILQ_FIRST(&local_queue))) {
399 STAILQ_REMOVE_HEAD(&local_queue, ktr_list);
400 ktr_writerequest(td, queued_req);
401 ktr_freerequest(queued_req);
407 * Submit a trace record for immediate commit to disk -- to be used only
408 * where entering VFS is OK. First drain any pending records that may have
409 * been cached in the thread.
412 ktr_submitrequest(struct thread *td, struct ktr_request *req)
417 sx_xlock(&ktrace_sx);
419 ktr_writerequest(td, req);
420 ktr_freerequest(req);
421 sx_xunlock(&ktrace_sx);
426 ktr_freerequest(struct ktr_request *req)
429 mtx_lock(&ktrace_mtx);
430 ktr_freerequest_locked(req);
431 mtx_unlock(&ktrace_mtx);
435 ktr_freerequest_locked(struct ktr_request *req)
438 mtx_assert(&ktrace_mtx, MA_OWNED);
439 if (req->ktr_buffer != NULL)
440 free(req->ktr_buffer, M_KTRACE);
441 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
445 ktr_io_params_ref(struct ktr_io_params *kiop)
447 mtx_assert(&ktrace_mtx, MA_OWNED);
451 static struct ktr_io_params *
452 ktr_io_params_rele(struct ktr_io_params *kiop)
454 mtx_assert(&ktrace_mtx, MA_OWNED);
457 KASSERT(kiop->refs > 0, ("kiop ref == 0 %p", kiop));
458 return (--(kiop->refs) == 0 ? kiop : NULL);
462 ktr_io_params_free(struct ktr_io_params *kiop)
467 MPASS(kiop->refs == 0);
468 vn_close(kiop->vp, FWRITE, kiop->cr, curthread);
470 free(kiop, M_KTRACE);
473 static struct ktr_io_params *
474 ktr_io_params_alloc(struct thread *td, struct vnode *vp)
476 struct ktr_io_params *res;
478 res = malloc(sizeof(struct ktr_io_params), M_KTRACE, M_WAITOK);
480 res->cr = crhold(td->td_ucred);
481 res->lim = lim_cur(td, RLIMIT_FSIZE);
487 * Disable tracing for a process and release all associated resources.
488 * The caller is responsible for releasing a reference on the returned
489 * vnode and credentials.
491 static struct ktr_io_params *
492 ktr_freeproc(struct proc *p)
494 struct ktr_io_params *kiop;
495 struct ktr_request *req;
497 PROC_LOCK_ASSERT(p, MA_OWNED);
498 mtx_assert(&ktrace_mtx, MA_OWNED);
499 kiop = ktr_io_params_rele(p->p_ktrioparms);
500 p->p_ktrioparms = NULL;
502 while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) {
503 STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list);
504 ktr_freerequest_locked(req);
510 ktr_get_tracevp(struct proc *p, bool ref)
514 PROC_LOCK_ASSERT(p, MA_OWNED);
516 if (p->p_ktrioparms != NULL) {
517 vp = p->p_ktrioparms->vp;
527 ktrsyscall(int code, int narg, register_t args[])
529 struct ktr_request *req;
530 struct ktr_syscall *ktp;
534 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
537 buflen = sizeof(register_t) * narg;
539 buf = malloc(buflen, M_KTRACE, M_WAITOK);
540 bcopy(args, buf, buflen);
542 req = ktr_getrequest(KTR_SYSCALL);
548 ktp = &req->ktr_data.ktr_syscall;
549 ktp->ktr_code = code;
550 ktp->ktr_narg = narg;
552 req->ktr_header.ktr_len = buflen;
553 req->ktr_buffer = buf;
555 ktr_submitrequest(curthread, req);
559 ktrsysret(int code, int error, register_t retval)
561 struct ktr_request *req;
562 struct ktr_sysret *ktp;
564 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
567 req = ktr_getrequest(KTR_SYSRET);
570 ktp = &req->ktr_data.ktr_sysret;
571 ktp->ktr_code = code;
572 ktp->ktr_error = error;
573 ktp->ktr_retval = ((error == 0) ? retval: 0); /* what about val2 ? */
574 ktr_submitrequest(curthread, req);
578 * When a setuid process execs, disable tracing.
580 * XXX: We toss any pending asynchronous records.
582 struct ktr_io_params *
583 ktrprocexec(struct proc *p)
585 struct ktr_io_params *kiop;
587 PROC_LOCK_ASSERT(p, MA_OWNED);
589 kiop = p->p_ktrioparms;
590 if (kiop == NULL || priv_check_cred(kiop->cr, PRIV_DEBUG_DIFFCRED))
593 mtx_lock(&ktrace_mtx);
594 kiop = ktr_freeproc(p);
595 mtx_unlock(&ktrace_mtx);
600 * When a process exits, drain per-process asynchronous trace records
601 * and disable tracing.
604 ktrprocexit(struct thread *td)
606 struct ktr_request *req;
608 struct ktr_io_params *kiop;
611 if (p->p_traceflag == 0)
615 req = ktr_getrequest_entered(td, KTR_PROCDTOR);
617 ktr_enqueuerequest(td, req);
618 sx_xlock(&ktrace_sx);
620 sx_xunlock(&ktrace_sx);
622 mtx_lock(&ktrace_mtx);
623 kiop = ktr_freeproc(p);
624 mtx_unlock(&ktrace_mtx);
626 ktr_io_params_free(kiop);
631 ktrprocctor_entered(struct thread *td, struct proc *p)
633 struct ktr_proc_ctor *ktp;
634 struct ktr_request *req;
638 td2 = FIRST_THREAD_IN_PROC(p);
639 req = ktr_getrequest_entered(td2, KTR_PROCCTOR);
642 ktp = &req->ktr_data.ktr_proc_ctor;
643 ktp->sv_flags = p->p_sysent->sv_flags;
644 ktr_enqueuerequest(td2, req);
648 ktrprocctor(struct proc *p)
650 struct thread *td = curthread;
652 if ((p->p_traceflag & KTRFAC_MASK) == 0)
656 ktrprocctor_entered(td, p);
661 * When a process forks, enable tracing in the new process if needed.
664 ktrprocfork(struct proc *p1, struct proc *p2)
667 MPASS(p2->p_ktrioparms == NULL);
668 MPASS(p2->p_traceflag == 0);
670 if (p1->p_traceflag == 0)
674 mtx_lock(&ktrace_mtx);
675 if (p1->p_traceflag & KTRFAC_INHERIT) {
676 p2->p_traceflag = p1->p_traceflag;
677 if ((p2->p_ktrioparms = p1->p_ktrioparms) != NULL)
678 p1->p_ktrioparms->refs++;
680 mtx_unlock(&ktrace_mtx);
687 * When a thread returns, drain any asynchronous records generated by the
691 ktruserret(struct thread *td)
695 sx_xlock(&ktrace_sx);
697 sx_xunlock(&ktrace_sx);
705 struct ktr_request *req;
709 namelen = strlen(path);
711 buf = malloc(namelen, M_KTRACE, M_WAITOK);
712 bcopy(path, buf, namelen);
714 req = ktr_getrequest(KTR_NAMEI);
721 req->ktr_header.ktr_len = namelen;
722 req->ktr_buffer = buf;
724 ktr_submitrequest(curthread, req);
728 ktrsysctl(int *name, u_int namelen)
730 struct ktr_request *req;
731 u_int mib[CTL_MAXNAME + 2];
736 /* Lookup name of mib. */
737 KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long"));
740 bcopy(name, mib + 2, namelen * sizeof(*name));
742 mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK);
743 error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen,
744 NULL, 0, &mibnamelen, 0);
746 free(mibname, M_KTRACE);
749 req = ktr_getrequest(KTR_SYSCTL);
751 free(mibname, M_KTRACE);
754 req->ktr_header.ktr_len = mibnamelen;
755 req->ktr_buffer = mibname;
756 ktr_submitrequest(curthread, req);
760 ktrgenio(int fd, enum uio_rw rw, struct uio *uio, int error)
762 struct ktr_request *req;
763 struct ktr_genio *ktg;
772 uio->uio_rw = UIO_WRITE;
773 datalen = MIN(uio->uio_resid, ktr_geniosize);
774 buf = malloc(datalen, M_KTRACE, M_WAITOK);
775 error = uiomove(buf, datalen, uio);
781 req = ktr_getrequest(KTR_GENIO);
786 ktg = &req->ktr_data.ktr_genio;
789 req->ktr_header.ktr_len = datalen;
790 req->ktr_buffer = buf;
791 ktr_submitrequest(curthread, req);
795 ktrpsig(int sig, sig_t action, sigset_t *mask, int code)
797 struct thread *td = curthread;
798 struct ktr_request *req;
801 req = ktr_getrequest(KTR_PSIG);
804 kp = &req->ktr_data.ktr_psig;
805 kp->signo = (char)sig;
809 ktr_enqueuerequest(td, req);
814 ktrcsw(int out, int user, const char *wmesg)
816 struct thread *td = curthread;
817 struct ktr_request *req;
820 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
823 req = ktr_getrequest(KTR_CSW);
826 kc = &req->ktr_data.ktr_csw;
830 strlcpy(kc->wmesg, wmesg, sizeof(kc->wmesg));
832 bzero(kc->wmesg, sizeof(kc->wmesg));
833 ktr_enqueuerequest(td, req);
838 ktrstruct(const char *name, const void *data, size_t datalen)
840 struct ktr_request *req;
842 size_t buflen, namelen;
844 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
849 namelen = strlen(name) + 1;
850 buflen = namelen + datalen;
851 buf = malloc(buflen, M_KTRACE, M_WAITOK);
853 bcopy(data, buf + namelen, datalen);
854 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) {
858 req->ktr_buffer = buf;
859 req->ktr_header.ktr_len = buflen;
860 ktr_submitrequest(curthread, req);
864 ktrstruct_error(const char *name, const void *data, size_t datalen, int error)
868 ktrstruct(name, data, datalen);
872 ktrstructarray(const char *name, enum uio_seg seg, const void *data,
873 int num_items, size_t struct_size)
875 struct ktr_request *req;
876 struct ktr_struct_array *ksa;
878 size_t buflen, datalen, namelen;
881 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
886 /* Trim array length to genio size. */
887 max_items = ktr_geniosize / struct_size;
888 if (num_items > max_items) {
892 num_items = max_items;
894 datalen = num_items * struct_size;
899 namelen = strlen(name) + 1;
900 buflen = namelen + datalen;
901 buf = malloc(buflen, M_KTRACE, M_WAITOK);
903 if (seg == UIO_SYSSPACE)
904 bcopy(data, buf + namelen, datalen);
906 if (copyin(data, buf + namelen, datalen) != 0) {
911 if ((req = ktr_getrequest(KTR_STRUCT_ARRAY)) == NULL) {
915 ksa = &req->ktr_data.ktr_struct_array;
916 ksa->struct_size = struct_size;
917 req->ktr_buffer = buf;
918 req->ktr_header.ktr_len = buflen;
919 ktr_submitrequest(curthread, req);
923 ktrcapfail(enum ktr_cap_fail_type type, const cap_rights_t *needed,
924 const cap_rights_t *held)
926 struct thread *td = curthread;
927 struct ktr_request *req;
928 struct ktr_cap_fail *kcf;
930 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
933 req = ktr_getrequest(KTR_CAPFAIL);
936 kcf = &req->ktr_data.ktr_cap_fail;
937 kcf->cap_type = type;
939 kcf->cap_needed = *needed;
941 cap_rights_init(&kcf->cap_needed);
943 kcf->cap_held = *held;
945 cap_rights_init(&kcf->cap_held);
946 ktr_enqueuerequest(td, req);
951 ktrfault(vm_offset_t vaddr, int type)
953 struct thread *td = curthread;
954 struct ktr_request *req;
955 struct ktr_fault *kf;
957 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
960 req = ktr_getrequest(KTR_FAULT);
963 kf = &req->ktr_data.ktr_fault;
966 ktr_enqueuerequest(td, req);
971 ktrfaultend(int result)
973 struct thread *td = curthread;
974 struct ktr_request *req;
975 struct ktr_faultend *kf;
977 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
980 req = ktr_getrequest(KTR_FAULTEND);
983 kf = &req->ktr_data.ktr_faultend;
985 ktr_enqueuerequest(td, req);
990 /* Interface and common routines */
992 #ifndef _SYS_SYSPROTO_H_
1002 sys_ktrace(struct thread *td, struct ktrace_args *uap)
1005 struct vnode *vp = NULL;
1008 int facs = uap->facs & ~KTRFAC_ROOT;
1009 int ops = KTROP(uap->ops);
1010 int descend = uap->ops & KTRFLAG_DESCEND;
1011 int nfound, ret = 0;
1012 int flags, error = 0;
1013 struct nameidata nd;
1014 struct ktr_io_params *kiop, *old_kiop;
1017 * Need something to (un)trace.
1019 if (ops != KTROP_CLEARFILE && facs == 0)
1024 if (ops != KTROP_CLEAR) {
1026 * an operation which requires a file argument.
1028 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname, td);
1029 flags = FREAD | FWRITE | O_NOFOLLOW;
1030 error = vn_open(&nd, &flags, 0, NULL);
1035 NDFREE(&nd, NDF_ONLY_PNBUF);
1038 if (vp->v_type != VREG) {
1039 (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
1043 kiop = ktr_io_params_alloc(td, vp);
1046 * Clear all uses of the tracefile.
1048 if (ops == KTROP_CLEARFILE) {
1050 sx_slock(&allproc_lock);
1051 FOREACH_PROC_IN_SYSTEM(p) {
1054 if (p->p_ktrioparms != NULL &&
1055 p->p_ktrioparms->vp == vp) {
1056 if (ktrcanset(td, p)) {
1057 mtx_lock(&ktrace_mtx);
1058 old_kiop = ktr_freeproc(p);
1059 mtx_unlock(&ktrace_mtx);
1064 if (old_kiop != NULL) {
1065 sx_sunlock(&allproc_lock);
1066 ktr_io_params_free(old_kiop);
1070 sx_sunlock(&allproc_lock);
1076 sx_slock(&proctree_lock);
1081 pg = pgfind(-uap->pid);
1083 sx_sunlock(&proctree_lock);
1088 * ktrops() may call vrele(). Lock pg_members
1089 * by the proctree_lock rather than pg_mtx.
1093 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1095 if (p->p_state == PRS_NEW ||
1096 p_cansee(td, p) != 0) {
1102 ret |= ktrsetchildren(td, p, ops, facs, kiop);
1104 ret |= ktrops(td, p, ops, facs, kiop);
1107 sx_sunlock(&proctree_lock);
1115 p = pfind(uap->pid);
1119 error = p_cansee(td, p);
1123 sx_sunlock(&proctree_lock);
1127 ret |= ktrsetchildren(td, p, ops, facs, kiop);
1129 ret |= ktrops(td, p, ops, facs, kiop);
1131 sx_sunlock(&proctree_lock);
1136 mtx_lock(&ktrace_mtx);
1137 kiop = ktr_io_params_rele(kiop);
1138 mtx_unlock(&ktrace_mtx);
1139 ktr_io_params_free(kiop);
1150 sys_utrace(struct thread *td, struct utrace_args *uap)
1154 struct ktr_request *req;
1158 if (!KTRPOINT(td, KTR_USER))
1160 if (uap->len > KTR_USER_MAXLEN)
1162 cp = malloc(uap->len, M_KTRACE, M_WAITOK);
1163 error = copyin(uap->addr, cp, uap->len);
1168 req = ktr_getrequest(KTR_USER);
1173 req->ktr_buffer = cp;
1174 req->ktr_header.ktr_len = uap->len;
1175 ktr_submitrequest(td, req);
1184 ktrops(struct thread *td, struct proc *p, int ops, int facs,
1185 struct ktr_io_params *new_kiop)
1187 struct ktr_io_params *old_kiop;
1189 PROC_LOCK_ASSERT(p, MA_OWNED);
1190 if (!ktrcanset(td, p)) {
1194 if (p->p_flag & P_WEXIT) {
1195 /* If the process is exiting, just ignore it. */
1200 mtx_lock(&ktrace_mtx);
1201 if (ops == KTROP_SET) {
1202 if (p->p_ktrioparms != NULL &&
1203 p->p_ktrioparms->vp != new_kiop->vp) {
1204 /* if trace file already in use, relinquish below */
1205 old_kiop = ktr_io_params_rele(p->p_ktrioparms);
1206 p->p_ktrioparms = NULL;
1208 if (p->p_ktrioparms == NULL) {
1209 p->p_ktrioparms = new_kiop;
1210 ktr_io_params_ref(new_kiop);
1212 p->p_traceflag |= facs;
1213 if (priv_check(td, PRIV_KTRACE) == 0)
1214 p->p_traceflag |= KTRFAC_ROOT;
1217 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0)
1218 /* no more tracing */
1219 old_kiop = ktr_freeproc(p);
1221 mtx_unlock(&ktrace_mtx);
1222 if ((p->p_traceflag & KTRFAC_MASK) != 0)
1223 ktrprocctor_entered(td, p);
1225 ktr_io_params_free(old_kiop);
1231 ktrsetchildren(struct thread *td, struct proc *top, int ops, int facs,
1232 struct ktr_io_params *new_kiop)
1238 PROC_LOCK_ASSERT(p, MA_OWNED);
1239 sx_assert(&proctree_lock, SX_LOCKED);
1241 ret |= ktrops(td, p, ops, facs, new_kiop);
1243 * If this process has children, descend to them next,
1244 * otherwise do any siblings, and if done with this level,
1245 * follow back up the tree (but not past top).
1247 if (!LIST_EMPTY(&p->p_children))
1248 p = LIST_FIRST(&p->p_children);
1252 if (LIST_NEXT(p, p_sibling)) {
1253 p = LIST_NEXT(p, p_sibling);
1264 ktr_writerequest(struct thread *td, struct ktr_request *req)
1266 struct ktr_io_params *kiop, *kiop1;
1267 struct ktr_header *kth;
1272 struct iovec aiov[3];
1275 int datalen, buflen;
1281 * We reference the kiop for use in I/O in case ktrace is
1282 * disabled on the process as we write out the request.
1284 mtx_lock(&ktrace_mtx);
1285 kiop = p->p_ktrioparms;
1288 * If kiop is NULL, it has been cleared out from under this
1289 * request, so just drop it.
1292 mtx_unlock(&ktrace_mtx);
1296 ktr_io_params_ref(kiop);
1301 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL"));
1302 mtx_unlock(&ktrace_mtx);
1304 kth = &req->ktr_header;
1305 KASSERT(((u_short)kth->ktr_type & ~KTR_DROP) < nitems(data_lengths),
1306 ("data_lengths array overflow"));
1307 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP];
1308 buflen = kth->ktr_len;
1309 auio.uio_iov = &aiov[0];
1310 auio.uio_offset = 0;
1311 auio.uio_segflg = UIO_SYSSPACE;
1312 auio.uio_rw = UIO_WRITE;
1313 aiov[0].iov_base = (caddr_t)kth;
1314 aiov[0].iov_len = sizeof(struct ktr_header);
1315 auio.uio_resid = sizeof(struct ktr_header);
1316 auio.uio_iovcnt = 1;
1319 aiov[1].iov_base = (caddr_t)&req->ktr_data;
1320 aiov[1].iov_len = datalen;
1321 auio.uio_resid += datalen;
1323 kth->ktr_len += datalen;
1326 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write"));
1327 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer;
1328 aiov[auio.uio_iovcnt].iov_len = buflen;
1329 auio.uio_resid += buflen;
1333 vn_start_write(vp, &mp, V_WAIT);
1334 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1335 td->td_ktr_io_lim = lim;
1337 error = mac_vnode_check_write(cred, NOCRED, vp);
1340 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred);
1342 vn_finished_write(mp);
1344 mtx_lock(&ktrace_mtx);
1345 kiop = ktr_io_params_rele(kiop);
1346 mtx_unlock(&ktrace_mtx);
1347 ktr_io_params_free(kiop);
1352 * If error encountered, give up tracing on this vnode on this
1353 * process. Other processes might still be suitable for
1354 * writes to this vnode.
1357 "ktrace write failed, errno %d, tracing stopped for pid %d\n",
1362 mtx_lock(&ktrace_mtx);
1363 if (p->p_ktrioparms != NULL && p->p_ktrioparms->vp == vp)
1364 kiop1 = ktr_freeproc(p);
1365 kiop = ktr_io_params_rele(kiop);
1366 mtx_unlock(&ktrace_mtx);
1368 ktr_io_params_free(kiop1);
1369 ktr_io_params_free(kiop);
1373 * Return true if caller has permission to set the ktracing state
1374 * of target. Essentially, the target can't possess any
1375 * more permissions than the caller. KTRFAC_ROOT signifies that
1376 * root previously set the tracing status on the target process, and
1377 * so, only root may further change it.
1380 ktrcanset(struct thread *td, struct proc *targetp)
1383 PROC_LOCK_ASSERT(targetp, MA_OWNED);
1384 if (targetp->p_traceflag & KTRFAC_ROOT &&
1385 priv_check(td, PRIV_KTRACE))
1388 if (p_candebug(td, targetp) != 0)