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 const 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 ast_ktrace(struct thread *td, int tda __unused)
219 ktrace_init(void *dummy)
221 struct ktr_request *req;
224 mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET);
225 sx_init(&ktrace_sx, "ktrace_sx");
226 STAILQ_INIT(&ktr_free);
227 for (i = 0; i < ktr_requestpool; i++) {
228 req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK |
230 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
232 ast_register(TDA_KTRACE, ASTR_ASTF_REQUIRED, 0, ast_ktrace);
234 SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL);
237 sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)
240 u_int newsize, oldsize, wantsize;
243 /* Handle easy read-only case first to avoid warnings from GCC. */
245 oldsize = ktr_requestpool;
246 return (SYSCTL_OUT(req, &oldsize, sizeof(u_int)));
249 error = SYSCTL_IN(req, &wantsize, sizeof(u_int));
254 oldsize = ktr_requestpool;
255 newsize = ktrace_resize_pool(oldsize, wantsize);
257 error = SYSCTL_OUT(req, &oldsize, sizeof(u_int));
260 if (wantsize > oldsize && newsize < wantsize)
264 SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool,
265 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &ktr_requestpool, 0,
266 sysctl_kern_ktrace_request_pool, "IU",
267 "Pool buffer size for ktrace(1)");
270 ktrace_resize_pool(u_int oldsize, u_int newsize)
272 STAILQ_HEAD(, ktr_request) ktr_new;
273 struct ktr_request *req;
277 bound = newsize - oldsize;
279 return (ktr_requestpool);
281 mtx_lock(&ktrace_mtx);
282 /* Shrink pool down to newsize if possible. */
283 while (bound++ < 0) {
284 req = STAILQ_FIRST(&ktr_free);
287 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
292 /* Grow pool up to newsize. */
293 STAILQ_INIT(&ktr_new);
294 while (bound-- > 0) {
295 req = malloc(sizeof(struct ktr_request), M_KTRACE,
297 STAILQ_INSERT_HEAD(&ktr_new, req, ktr_list);
299 mtx_lock(&ktrace_mtx);
300 STAILQ_CONCAT(&ktr_free, &ktr_new);
301 ktr_requestpool += (newsize - oldsize);
303 mtx_unlock(&ktrace_mtx);
304 return (ktr_requestpool);
307 /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */
308 CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) ==
309 (sizeof((struct thread *)NULL)->td_name));
311 static struct ktr_request *
312 ktr_getrequest_entered(struct thread *td, int type)
314 struct ktr_request *req;
315 struct proc *p = td->td_proc;
318 mtx_lock(&ktrace_mtx);
319 if (!KTRCHECK(td, type)) {
320 mtx_unlock(&ktrace_mtx);
323 req = STAILQ_FIRST(&ktr_free);
325 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
326 req->ktr_header.ktr_type = type;
327 if (p->p_traceflag & KTRFAC_DROP) {
328 req->ktr_header.ktr_type |= KTR_DROP;
329 p->p_traceflag &= ~KTRFAC_DROP;
331 mtx_unlock(&ktrace_mtx);
332 nanotime(&req->ktr_header.ktr_time);
333 req->ktr_header.ktr_type |= KTR_VERSIONED;
334 req->ktr_header.ktr_pid = p->p_pid;
335 req->ktr_header.ktr_tid = td->td_tid;
336 req->ktr_header.ktr_cpu = PCPU_GET(cpuid);
337 req->ktr_header.ktr_version = KTR_VERSION1;
338 bcopy(td->td_name, req->ktr_header.ktr_comm,
339 sizeof(req->ktr_header.ktr_comm));
340 req->ktr_buffer = NULL;
341 req->ktr_header.ktr_len = 0;
343 p->p_traceflag |= KTRFAC_DROP;
346 mtx_unlock(&ktrace_mtx);
348 printf("Out of ktrace request objects.\n");
353 static struct ktr_request *
354 ktr_getrequest(int type)
356 struct thread *td = curthread;
357 struct ktr_request *req;
360 req = ktr_getrequest_entered(td, type);
368 * Some trace generation environments don't permit direct access to VFS,
369 * such as during a context switch where sleeping is not allowed. Under these
370 * circumstances, queue a request to the thread to be written asynchronously
374 ktr_enqueuerequest(struct thread *td, struct ktr_request *req)
377 mtx_lock(&ktrace_mtx);
378 STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list);
379 mtx_unlock(&ktrace_mtx);
380 ast_sched(td, TDA_KTRACE);
384 * Drain any pending ktrace records from the per-thread queue to disk. This
385 * is used both internally before committing other records, and also on
386 * system call return. We drain all the ones we can find at the time when
387 * drain is requested, but don't keep draining after that as those events
388 * may be approximately "after" the current event.
391 ktr_drain(struct thread *td)
393 struct ktr_request *queued_req;
394 STAILQ_HEAD(, ktr_request) local_queue;
397 sx_assert(&ktrace_sx, SX_XLOCKED);
399 STAILQ_INIT(&local_queue);
401 if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) {
402 mtx_lock(&ktrace_mtx);
403 STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr);
404 mtx_unlock(&ktrace_mtx);
406 while ((queued_req = STAILQ_FIRST(&local_queue))) {
407 STAILQ_REMOVE_HEAD(&local_queue, ktr_list);
408 ktr_writerequest(td, queued_req);
409 ktr_freerequest(queued_req);
415 * Submit a trace record for immediate commit to disk -- to be used only
416 * where entering VFS is OK. First drain any pending records that may have
417 * been cached in the thread.
420 ktr_submitrequest(struct thread *td, struct ktr_request *req)
425 sx_xlock(&ktrace_sx);
427 ktr_writerequest(td, req);
428 ktr_freerequest(req);
429 sx_xunlock(&ktrace_sx);
434 ktr_freerequest(struct ktr_request *req)
437 mtx_lock(&ktrace_mtx);
438 ktr_freerequest_locked(req);
439 mtx_unlock(&ktrace_mtx);
443 ktr_freerequest_locked(struct ktr_request *req)
446 mtx_assert(&ktrace_mtx, MA_OWNED);
447 if (req->ktr_buffer != NULL)
448 free(req->ktr_buffer, M_KTRACE);
449 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
453 ktr_io_params_ref(struct ktr_io_params *kiop)
455 mtx_assert(&ktrace_mtx, MA_OWNED);
459 static struct ktr_io_params *
460 ktr_io_params_rele(struct ktr_io_params *kiop)
462 mtx_assert(&ktrace_mtx, MA_OWNED);
465 KASSERT(kiop->refs > 0, ("kiop ref == 0 %p", kiop));
466 return (--(kiop->refs) == 0 ? kiop : NULL);
470 ktr_io_params_free(struct ktr_io_params *kiop)
475 MPASS(kiop->refs == 0);
476 vn_close(kiop->vp, FWRITE, kiop->cr, curthread);
478 free(kiop, M_KTRACE);
481 static struct ktr_io_params *
482 ktr_io_params_alloc(struct thread *td, struct vnode *vp)
484 struct ktr_io_params *res;
486 res = malloc(sizeof(struct ktr_io_params), M_KTRACE, M_WAITOK);
488 res->cr = crhold(td->td_ucred);
489 res->lim = lim_cur(td, RLIMIT_FSIZE);
495 * Disable tracing for a process and release all associated resources.
496 * The caller is responsible for releasing a reference on the returned
497 * vnode and credentials.
499 static struct ktr_io_params *
500 ktr_freeproc(struct proc *p)
502 struct ktr_io_params *kiop;
503 struct ktr_request *req;
505 PROC_LOCK_ASSERT(p, MA_OWNED);
506 mtx_assert(&ktrace_mtx, MA_OWNED);
507 kiop = ktr_io_params_rele(p->p_ktrioparms);
508 p->p_ktrioparms = NULL;
510 while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) {
511 STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list);
512 ktr_freerequest_locked(req);
518 ktr_get_tracevp(struct proc *p, bool ref)
522 PROC_LOCK_ASSERT(p, MA_OWNED);
524 if (p->p_ktrioparms != NULL) {
525 vp = p->p_ktrioparms->vp;
535 ktrsyscall(int code, int narg, syscallarg_t args[])
537 struct ktr_request *req;
538 struct ktr_syscall *ktp;
542 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
545 buflen = sizeof(register_t) * narg;
547 buf = malloc(buflen, M_KTRACE, M_WAITOK);
548 bcopy(args, buf, buflen);
550 req = ktr_getrequest(KTR_SYSCALL);
556 ktp = &req->ktr_data.ktr_syscall;
557 ktp->ktr_code = code;
558 ktp->ktr_narg = narg;
560 req->ktr_header.ktr_len = buflen;
561 req->ktr_buffer = buf;
563 ktr_submitrequest(curthread, req);
567 ktrsysret(int code, int error, register_t retval)
569 struct ktr_request *req;
570 struct ktr_sysret *ktp;
572 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
575 req = ktr_getrequest(KTR_SYSRET);
578 ktp = &req->ktr_data.ktr_sysret;
579 ktp->ktr_code = code;
580 ktp->ktr_error = error;
581 ktp->ktr_retval = ((error == 0) ? retval: 0); /* what about val2 ? */
582 ktr_submitrequest(curthread, req);
586 * When a setuid process execs, disable tracing.
588 * XXX: We toss any pending asynchronous records.
590 struct ktr_io_params *
591 ktrprocexec(struct proc *p)
593 struct ktr_io_params *kiop;
595 PROC_LOCK_ASSERT(p, MA_OWNED);
597 kiop = p->p_ktrioparms;
598 if (kiop == NULL || priv_check_cred(kiop->cr, PRIV_DEBUG_DIFFCRED))
601 mtx_lock(&ktrace_mtx);
602 kiop = ktr_freeproc(p);
603 mtx_unlock(&ktrace_mtx);
608 * When a process exits, drain per-process asynchronous trace records
609 * and disable tracing.
612 ktrprocexit(struct thread *td)
614 struct ktr_request *req;
616 struct ktr_io_params *kiop;
619 if (p->p_traceflag == 0)
623 req = ktr_getrequest_entered(td, KTR_PROCDTOR);
625 ktr_enqueuerequest(td, req);
626 sx_xlock(&ktrace_sx);
628 sx_xunlock(&ktrace_sx);
630 mtx_lock(&ktrace_mtx);
631 kiop = ktr_freeproc(p);
632 mtx_unlock(&ktrace_mtx);
634 ktr_io_params_free(kiop);
639 ktrprocctor_entered(struct thread *td, struct proc *p)
641 struct ktr_proc_ctor *ktp;
642 struct ktr_request *req;
646 td2 = FIRST_THREAD_IN_PROC(p);
647 req = ktr_getrequest_entered(td2, KTR_PROCCTOR);
650 ktp = &req->ktr_data.ktr_proc_ctor;
651 ktp->sv_flags = p->p_sysent->sv_flags;
652 ktr_enqueuerequest(td2, req);
656 ktrprocctor(struct proc *p)
658 struct thread *td = curthread;
660 if ((p->p_traceflag & KTRFAC_MASK) == 0)
664 ktrprocctor_entered(td, p);
669 * When a process forks, enable tracing in the new process if needed.
672 ktrprocfork(struct proc *p1, struct proc *p2)
675 MPASS(p2->p_ktrioparms == NULL);
676 MPASS(p2->p_traceflag == 0);
678 if (p1->p_traceflag == 0)
682 mtx_lock(&ktrace_mtx);
683 if (p1->p_traceflag & KTRFAC_INHERIT) {
684 p2->p_traceflag = p1->p_traceflag;
685 if ((p2->p_ktrioparms = p1->p_ktrioparms) != NULL)
686 p1->p_ktrioparms->refs++;
688 mtx_unlock(&ktrace_mtx);
695 * When a thread returns, drain any asynchronous records generated by the
699 ktruserret(struct thread *td)
703 sx_xlock(&ktrace_sx);
705 sx_xunlock(&ktrace_sx);
710 ktrnamei(const char *path)
712 struct ktr_request *req;
716 namelen = strlen(path);
718 buf = malloc(namelen, M_KTRACE, M_WAITOK);
719 bcopy(path, buf, namelen);
721 req = ktr_getrequest(KTR_NAMEI);
728 req->ktr_header.ktr_len = namelen;
729 req->ktr_buffer = buf;
731 ktr_submitrequest(curthread, req);
735 ktrsysctl(int *name, u_int namelen)
737 struct ktr_request *req;
738 u_int mib[CTL_MAXNAME + 2];
743 /* Lookup name of mib. */
744 KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long"));
747 bcopy(name, mib + 2, namelen * sizeof(*name));
749 mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK);
750 error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen,
751 NULL, 0, &mibnamelen, 0);
753 free(mibname, M_KTRACE);
756 req = ktr_getrequest(KTR_SYSCTL);
758 free(mibname, M_KTRACE);
761 req->ktr_header.ktr_len = mibnamelen;
762 req->ktr_buffer = mibname;
763 ktr_submitrequest(curthread, req);
767 ktrgenio(int fd, enum uio_rw rw, struct uio *uio, int error)
769 struct ktr_request *req;
770 struct ktr_genio *ktg;
779 uio->uio_rw = UIO_WRITE;
780 datalen = MIN(uio->uio_resid, ktr_geniosize);
781 buf = malloc(datalen, M_KTRACE, M_WAITOK);
782 error = uiomove(buf, datalen, uio);
788 req = ktr_getrequest(KTR_GENIO);
793 ktg = &req->ktr_data.ktr_genio;
796 req->ktr_header.ktr_len = datalen;
797 req->ktr_buffer = buf;
798 ktr_submitrequest(curthread, req);
802 ktrpsig(int sig, sig_t action, sigset_t *mask, int code)
804 struct thread *td = curthread;
805 struct ktr_request *req;
808 req = ktr_getrequest(KTR_PSIG);
811 kp = &req->ktr_data.ktr_psig;
812 kp->signo = (char)sig;
816 ktr_enqueuerequest(td, req);
821 ktrcsw(int out, int user, const char *wmesg)
823 struct thread *td = curthread;
824 struct ktr_request *req;
827 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
830 req = ktr_getrequest(KTR_CSW);
833 kc = &req->ktr_data.ktr_csw;
837 strlcpy(kc->wmesg, wmesg, sizeof(kc->wmesg));
839 bzero(kc->wmesg, sizeof(kc->wmesg));
840 ktr_enqueuerequest(td, req);
845 ktrstruct(const char *name, const void *data, size_t datalen)
847 struct ktr_request *req;
849 size_t buflen, namelen;
851 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
856 namelen = strlen(name) + 1;
857 buflen = namelen + datalen;
858 buf = malloc(buflen, M_KTRACE, M_WAITOK);
860 bcopy(data, buf + namelen, datalen);
861 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) {
865 req->ktr_buffer = buf;
866 req->ktr_header.ktr_len = buflen;
867 ktr_submitrequest(curthread, req);
871 ktrstruct_error(const char *name, const void *data, size_t datalen, int error)
875 ktrstruct(name, data, datalen);
879 ktrstructarray(const char *name, enum uio_seg seg, const void *data,
880 int num_items, size_t struct_size)
882 struct ktr_request *req;
883 struct ktr_struct_array *ksa;
885 size_t buflen, datalen, namelen;
888 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
893 /* Trim array length to genio size. */
894 max_items = ktr_geniosize / struct_size;
895 if (num_items > max_items) {
899 num_items = max_items;
901 datalen = num_items * struct_size;
906 namelen = strlen(name) + 1;
907 buflen = namelen + datalen;
908 buf = malloc(buflen, M_KTRACE, M_WAITOK);
910 if (seg == UIO_SYSSPACE)
911 bcopy(data, buf + namelen, datalen);
913 if (copyin(data, buf + namelen, datalen) != 0) {
918 if ((req = ktr_getrequest(KTR_STRUCT_ARRAY)) == NULL) {
922 ksa = &req->ktr_data.ktr_struct_array;
923 ksa->struct_size = struct_size;
924 req->ktr_buffer = buf;
925 req->ktr_header.ktr_len = buflen;
926 ktr_submitrequest(curthread, req);
930 ktrcapfail(enum ktr_cap_fail_type type, const cap_rights_t *needed,
931 const cap_rights_t *held)
933 struct thread *td = curthread;
934 struct ktr_request *req;
935 struct ktr_cap_fail *kcf;
937 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
940 req = ktr_getrequest(KTR_CAPFAIL);
943 kcf = &req->ktr_data.ktr_cap_fail;
944 kcf->cap_type = type;
946 kcf->cap_needed = *needed;
948 cap_rights_init(&kcf->cap_needed);
950 kcf->cap_held = *held;
952 cap_rights_init(&kcf->cap_held);
953 ktr_enqueuerequest(td, req);
958 ktrfault(vm_offset_t vaddr, int type)
960 struct thread *td = curthread;
961 struct ktr_request *req;
962 struct ktr_fault *kf;
964 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
967 req = ktr_getrequest(KTR_FAULT);
970 kf = &req->ktr_data.ktr_fault;
973 ktr_enqueuerequest(td, req);
978 ktrfaultend(int result)
980 struct thread *td = curthread;
981 struct ktr_request *req;
982 struct ktr_faultend *kf;
984 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
987 req = ktr_getrequest(KTR_FAULTEND);
990 kf = &req->ktr_data.ktr_faultend;
992 ktr_enqueuerequest(td, req);
997 /* Interface and common routines */
999 #ifndef _SYS_SYSPROTO_H_
1000 struct ktrace_args {
1009 sys_ktrace(struct thread *td, struct ktrace_args *uap)
1012 struct vnode *vp = NULL;
1015 int facs = uap->facs & ~KTRFAC_ROOT;
1016 int ops = KTROP(uap->ops);
1017 int descend = uap->ops & KTRFLAG_DESCEND;
1019 int flags, error = 0;
1020 struct nameidata nd;
1021 struct ktr_io_params *kiop, *old_kiop;
1024 * Need something to (un)trace.
1026 if (ops != KTROP_CLEARFILE && facs == 0)
1030 if (ops != KTROP_CLEAR) {
1032 * an operation which requires a file argument.
1034 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname);
1035 flags = FREAD | FWRITE | O_NOFOLLOW;
1036 error = vn_open(&nd, &flags, 0, NULL);
1042 if (vp->v_type != VREG) {
1043 (void)vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
1046 kiop = ktr_io_params_alloc(td, vp);
1050 * Clear all uses of the tracefile.
1053 if (ops == KTROP_CLEARFILE) {
1055 sx_slock(&allproc_lock);
1056 FOREACH_PROC_IN_SYSTEM(p) {
1059 if (p->p_ktrioparms != NULL &&
1060 p->p_ktrioparms->vp == vp) {
1061 if (ktrcanset(td, p)) {
1062 mtx_lock(&ktrace_mtx);
1063 old_kiop = ktr_freeproc(p);
1064 mtx_unlock(&ktrace_mtx);
1069 if (old_kiop != NULL) {
1070 sx_sunlock(&allproc_lock);
1071 ktr_io_params_free(old_kiop);
1075 sx_sunlock(&allproc_lock);
1081 sx_slock(&proctree_lock);
1086 pg = pgfind(-uap->pid);
1088 sx_sunlock(&proctree_lock);
1094 * ktrops() may call vrele(). Lock pg_members
1095 * by the proctree_lock rather than pg_mtx.
1098 if (LIST_EMPTY(&pg->pg_members)) {
1099 sx_sunlock(&proctree_lock);
1103 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1106 ret |= ktrsetchildren(td, p, ops, facs, kiop);
1108 ret |= ktrops(td, p, ops, facs, kiop);
1114 p = pfind(uap->pid);
1117 sx_sunlock(&proctree_lock);
1121 ret |= ktrsetchildren(td, p, ops, facs, kiop);
1123 ret |= ktrops(td, p, ops, facs, kiop);
1125 sx_sunlock(&proctree_lock);
1130 mtx_lock(&ktrace_mtx);
1131 kiop = ktr_io_params_rele(kiop);
1132 mtx_unlock(&ktrace_mtx);
1133 ktr_io_params_free(kiop);
1144 sys_utrace(struct thread *td, struct utrace_args *uap)
1148 struct ktr_request *req;
1152 if (!KTRPOINT(td, KTR_USER))
1154 if (uap->len > KTR_USER_MAXLEN)
1156 cp = malloc(uap->len, M_KTRACE, M_WAITOK);
1157 error = copyin(uap->addr, cp, uap->len);
1162 req = ktr_getrequest(KTR_USER);
1167 req->ktr_buffer = cp;
1168 req->ktr_header.ktr_len = uap->len;
1169 ktr_submitrequest(td, req);
1178 ktrops(struct thread *td, struct proc *p, int ops, int facs,
1179 struct ktr_io_params *new_kiop)
1181 struct ktr_io_params *old_kiop;
1183 PROC_LOCK_ASSERT(p, MA_OWNED);
1184 if (!ktrcanset(td, p)) {
1188 if ((ops == KTROP_SET && p->p_state == PRS_NEW) ||
1189 p_cansee(td, p) != 0) {
1191 * Disallow setting trace points if the process is being born.
1192 * This avoids races with trace point inheritance in
1198 if ((p->p_flag & P_WEXIT) != 0) {
1200 * There's nothing to do if the process is exiting, but avoid
1201 * signaling an error.
1207 mtx_lock(&ktrace_mtx);
1208 if (ops == KTROP_SET) {
1209 if (p->p_ktrioparms != NULL &&
1210 p->p_ktrioparms->vp != new_kiop->vp) {
1211 /* if trace file already in use, relinquish below */
1212 old_kiop = ktr_io_params_rele(p->p_ktrioparms);
1213 p->p_ktrioparms = NULL;
1215 if (p->p_ktrioparms == NULL) {
1216 p->p_ktrioparms = new_kiop;
1217 ktr_io_params_ref(new_kiop);
1219 p->p_traceflag |= facs;
1220 if (priv_check(td, PRIV_KTRACE) == 0)
1221 p->p_traceflag |= KTRFAC_ROOT;
1224 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0)
1225 /* no more tracing */
1226 old_kiop = ktr_freeproc(p);
1228 mtx_unlock(&ktrace_mtx);
1229 if ((p->p_traceflag & KTRFAC_MASK) != 0)
1230 ktrprocctor_entered(td, p);
1232 ktr_io_params_free(old_kiop);
1238 ktrsetchildren(struct thread *td, struct proc *top, int ops, int facs,
1239 struct ktr_io_params *new_kiop)
1245 PROC_LOCK_ASSERT(p, MA_OWNED);
1246 sx_assert(&proctree_lock, SX_LOCKED);
1248 ret |= ktrops(td, p, ops, facs, new_kiop);
1250 * If this process has children, descend to them next,
1251 * otherwise do any siblings, and if done with this level,
1252 * follow back up the tree (but not past top).
1254 if (!LIST_EMPTY(&p->p_children))
1255 p = LIST_FIRST(&p->p_children);
1259 if (LIST_NEXT(p, p_sibling)) {
1260 p = LIST_NEXT(p, p_sibling);
1271 ktr_writerequest(struct thread *td, struct ktr_request *req)
1273 struct ktr_io_params *kiop, *kiop1;
1274 struct ktr_header *kth;
1279 struct iovec aiov[3];
1282 int datalen, buflen;
1288 * We reference the kiop for use in I/O in case ktrace is
1289 * disabled on the process as we write out the request.
1291 mtx_lock(&ktrace_mtx);
1292 kiop = p->p_ktrioparms;
1295 * If kiop is NULL, it has been cleared out from under this
1296 * request, so just drop it.
1299 mtx_unlock(&ktrace_mtx);
1303 ktr_io_params_ref(kiop);
1308 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL"));
1309 mtx_unlock(&ktrace_mtx);
1311 kth = &req->ktr_header;
1312 KASSERT(((u_short)kth->ktr_type & ~KTR_TYPE) < nitems(data_lengths),
1313 ("data_lengths array overflow"));
1314 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_TYPE];
1315 buflen = kth->ktr_len;
1316 auio.uio_iov = &aiov[0];
1317 auio.uio_offset = 0;
1318 auio.uio_segflg = UIO_SYSSPACE;
1319 auio.uio_rw = UIO_WRITE;
1320 aiov[0].iov_base = (caddr_t)kth;
1321 aiov[0].iov_len = sizeof(struct ktr_header);
1322 auio.uio_resid = sizeof(struct ktr_header);
1323 auio.uio_iovcnt = 1;
1326 aiov[1].iov_base = (caddr_t)&req->ktr_data;
1327 aiov[1].iov_len = datalen;
1328 auio.uio_resid += datalen;
1330 kth->ktr_len += datalen;
1333 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write"));
1334 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer;
1335 aiov[auio.uio_iovcnt].iov_len = buflen;
1336 auio.uio_resid += buflen;
1340 vn_start_write(vp, &mp, V_WAIT);
1341 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1342 td->td_ktr_io_lim = lim;
1344 error = mac_vnode_check_write(cred, NOCRED, vp);
1347 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred);
1349 vn_finished_write(mp);
1351 mtx_lock(&ktrace_mtx);
1352 kiop = ktr_io_params_rele(kiop);
1353 mtx_unlock(&ktrace_mtx);
1354 ktr_io_params_free(kiop);
1359 * If error encountered, give up tracing on this vnode on this
1360 * process. Other processes might still be suitable for
1361 * writes to this vnode.
1364 "ktrace write failed, errno %d, tracing stopped for pid %d\n",
1369 mtx_lock(&ktrace_mtx);
1370 if (p->p_ktrioparms != NULL && p->p_ktrioparms->vp == vp)
1371 kiop1 = ktr_freeproc(p);
1372 kiop = ktr_io_params_rele(kiop);
1373 mtx_unlock(&ktrace_mtx);
1375 ktr_io_params_free(kiop1);
1376 ktr_io_params_free(kiop);
1380 * Return true if caller has permission to set the ktracing state
1381 * of target. Essentially, the target can't possess any
1382 * more permissions than the caller. KTRFAC_ROOT signifies that
1383 * root previously set the tracing status on the target process, and
1384 * so, only root may further change it.
1387 ktrcanset(struct thread *td, struct proc *targetp)
1390 PROC_LOCK_ASSERT(targetp, MA_OWNED);
1391 if (targetp->p_traceflag & KTRFAC_ROOT &&
1392 priv_check(td, PRIV_KTRACE))
1395 if (p_candebug(td, targetp) != 0)