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/sysent.h>
59 #include <sys/syslog.h>
60 #include <sys/sysproto.h>
61
62 #include <security/mac/mac_framework.h>
63
64 /*
65 * The ktrace facility allows the tracing of certain key events in user space
66 * processes, such as system calls, signal delivery, context switches, and
67 * user generated events using utrace(2). It works by streaming event
68 * records and data to a vnode associated with the process using the
69 * ktrace(2) system call. In general, records can be written directly from
70 * the context that generates the event. One important exception to this is
71 * during a context switch, where sleeping is not permitted. To handle this
72 * case, trace events are generated using in-kernel ktr_request records, and
73 * then delivered to disk at a convenient moment -- either immediately, the
74 * next traceable event, at system call return, or at process exit.
75 *
76 * When dealing with multiple threads or processes writing to the same event
77 * log, ordering guarantees are weak: specifically, if an event has multiple
78 * records (i.e., system call enter and return), they may be interlaced with
79 * records from another event. Process and thread ID information is provided
80 * in the record, and user applications can de-interlace events if required.
81 */
82
83 static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE");
84
85 #ifdef KTRACE
86
87 #ifndef KTRACE_REQUEST_POOL
88 #define KTRACE_REQUEST_POOL 100
89 #endif
90
91 struct ktr_request {
92 struct ktr_header ktr_header;
93 void *ktr_buffer;
94 union {
95 struct ktr_proc_ctor ktr_proc_ctor;
96 struct ktr_syscall ktr_syscall;
97 struct ktr_sysret ktr_sysret;
98 struct ktr_genio ktr_genio;
99 struct ktr_psig ktr_psig;
100 struct ktr_csw ktr_csw;
101 struct ktr_fault ktr_fault;
102 struct ktr_faultend ktr_faultend;
103 } ktr_data;
104 STAILQ_ENTRY(ktr_request) ktr_list;
105 };
106
107 static int data_lengths[] = {
108 0, /* none */
109 offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */
110 sizeof(struct ktr_sysret), /* KTR_SYSRET */
111 0, /* KTR_NAMEI */
112 sizeof(struct ktr_genio), /* KTR_GENIO */
113 sizeof(struct ktr_psig), /* KTR_PSIG */
114 sizeof(struct ktr_csw), /* KTR_CSW */
115 0, /* KTR_USER */
116 0, /* KTR_STRUCT */
117 0, /* KTR_SYSCTL */
118 sizeof(struct ktr_proc_ctor), /* KTR_PROCCTOR */
119 0, /* KTR_PROCDTOR */
120 0, /* unused */
121 sizeof(struct ktr_fault), /* KTR_FAULT */
122 sizeof(struct ktr_faultend), /* KTR_FAULTEND */
123 };
124
125 static STAILQ_HEAD(, ktr_request) ktr_free;
126
127 static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options");
128
129 static u_int ktr_requestpool = KTRACE_REQUEST_POOL;
130 TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool);
131
132 static u_int ktr_geniosize = PAGE_SIZE;
133 TUNABLE_INT("kern.ktrace.genio_size", &ktr_geniosize);
134 SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RW, &ktr_geniosize,
135 0, "Maximum size of genio event payload");
136
137 static int print_message = 1;
138 static struct mtx ktrace_mtx;
139 static struct sx ktrace_sx;
140
141 static void ktrace_init(void *dummy);
142 static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS);
143 static u_int ktrace_resize_pool(u_int oldsize, u_int newsize);
144 static struct ktr_request *ktr_getrequest_entered(struct thread *td, int type);
145 static struct ktr_request *ktr_getrequest(int type);
146 static void ktr_submitrequest(struct thread *td, struct ktr_request *req);
147 static void ktr_freeproc(struct proc *p, struct ucred **uc,
148 struct vnode **vp);
149 static void ktr_freerequest(struct ktr_request *req);
150 static void ktr_freerequest_locked(struct ktr_request *req);
151 static void ktr_writerequest(struct thread *td, struct ktr_request *req);
152 static int ktrcanset(struct thread *,struct proc *);
153 static int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *);
154 static int ktrops(struct thread *,struct proc *,int,int,struct vnode *);
155 static void ktrprocctor_entered(struct thread *, struct proc *);
156
157 /*
158 * ktrace itself generates events, such as context switches, which we do not
159 * wish to trace. Maintain a flag, TDP_INKTRACE, on each thread to determine
160 * whether or not it is in a region where tracing of events should be
161 * suppressed.
162 */
163 static void
164 ktrace_enter(struct thread *td)
165 {
166
167 KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set"));
168 td->td_pflags |= TDP_INKTRACE;
169 }
170
171 static void
172 ktrace_exit(struct thread *td)
173 {
174
175 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set"));
176 td->td_pflags &= ~TDP_INKTRACE;
177 }
178
179 static void
180 ktrace_assert(struct thread *td)
181 {
182
183 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set"));
184 }
185
186 static void
187 ktrace_init(void *dummy)
188 {
189 struct ktr_request *req;
190 int i;
191
192 mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET);
193 sx_init(&ktrace_sx, "ktrace_sx");
194 STAILQ_INIT(&ktr_free);
195 for (i = 0; i < ktr_requestpool; i++) {
196 req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK);
197 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
198 }
199 }
200 SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL);
201
202 static int
203 sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)
204 {
205 struct thread *td;
206 u_int newsize, oldsize, wantsize;
207 int error;
208
209 /* Handle easy read-only case first to avoid warnings from GCC. */
210 if (!req->newptr) {
211 oldsize = ktr_requestpool;
212 return (SYSCTL_OUT(req, &oldsize, sizeof(u_int)));
213 }
214
215 error = SYSCTL_IN(req, &wantsize, sizeof(u_int));
216 if (error)
217 return (error);
218 td = curthread;
219 ktrace_enter(td);
220 oldsize = ktr_requestpool;
221 newsize = ktrace_resize_pool(oldsize, wantsize);
222 ktrace_exit(td);
223 error = SYSCTL_OUT(req, &oldsize, sizeof(u_int));
224 if (error)
225 return (error);
226 if (wantsize > oldsize && newsize < wantsize)
227 return (ENOSPC);
228 return (0);
229 }
230 SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW,
231 &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU", "");
232
233 static u_int
234 ktrace_resize_pool(u_int oldsize, u_int newsize)
235 {
236 STAILQ_HEAD(, ktr_request) ktr_new;
237 struct ktr_request *req;
238 int bound;
239
240 print_message = 1;
241 bound = newsize - oldsize;
242 if (bound == 0)
243 return (ktr_requestpool);
244 if (bound < 0) {
245 mtx_lock(&ktrace_mtx);
246 /* Shrink pool down to newsize if possible. */
247 while (bound++ < 0) {
248 req = STAILQ_FIRST(&ktr_free);
249 if (req == NULL)
250 break;
251 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
252 ktr_requestpool--;
253 free(req, M_KTRACE);
254 }
255 } else {
256 /* Grow pool up to newsize. */
257 STAILQ_INIT(&ktr_new);
258 while (bound-- > 0) {
259 req = malloc(sizeof(struct ktr_request), M_KTRACE,
260 M_WAITOK);
261 STAILQ_INSERT_HEAD(&ktr_new, req, ktr_list);
262 }
263 mtx_lock(&ktrace_mtx);
264 STAILQ_CONCAT(&ktr_free, &ktr_new);
265 ktr_requestpool += (newsize - oldsize);
266 }
267 mtx_unlock(&ktrace_mtx);
268 return (ktr_requestpool);
269 }
270
271 /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */
272 CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) ==
273 (sizeof((struct thread *)NULL)->td_name));
274
275 static struct ktr_request *
276 ktr_getrequest_entered(struct thread *td, int type)
277 {
278 struct ktr_request *req;
279 struct proc *p = td->td_proc;
280 int pm;
281
282 mtx_lock(&ktrace_mtx);
283 if (!KTRCHECK(td, type)) {
284 mtx_unlock(&ktrace_mtx);
285 return (NULL);
286 }
287 req = STAILQ_FIRST(&ktr_free);
288 if (req != NULL) {
289 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
290 req->ktr_header.ktr_type = type;
291 if (p->p_traceflag & KTRFAC_DROP) {
292 req->ktr_header.ktr_type |= KTR_DROP;
293 p->p_traceflag &= ~KTRFAC_DROP;
294 }
295 mtx_unlock(&ktrace_mtx);
296 microtime(&req->ktr_header.ktr_time);
297 req->ktr_header.ktr_pid = p->p_pid;
298 req->ktr_header.ktr_tid = td->td_tid;
299 bcopy(td->td_name, req->ktr_header.ktr_comm,
300 sizeof(req->ktr_header.ktr_comm));
301 req->ktr_buffer = NULL;
302 req->ktr_header.ktr_len = 0;
303 } else {
304 p->p_traceflag |= KTRFAC_DROP;
305 pm = print_message;
306 print_message = 0;
307 mtx_unlock(&ktrace_mtx);
308 if (pm)
309 printf("Out of ktrace request objects.\n");
310 }
311 return (req);
312 }
313
314 static struct ktr_request *
315 ktr_getrequest(int type)
316 {
317 struct thread *td = curthread;
318 struct ktr_request *req;
319
320 ktrace_enter(td);
321 req = ktr_getrequest_entered(td, type);
322 if (req == NULL)
323 ktrace_exit(td);
324
325 return (req);
326 }
327
328 /*
329 * Some trace generation environments don't permit direct access to VFS,
330 * such as during a context switch where sleeping is not allowed. Under these
331 * circumstances, queue a request to the thread to be written asynchronously
332 * later.
333 */
334 static void
335 ktr_enqueuerequest(struct thread *td, struct ktr_request *req)
336 {
337
338 mtx_lock(&ktrace_mtx);
339 STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list);
340 mtx_unlock(&ktrace_mtx);
341 }
342
343 /*
344 * Drain any pending ktrace records from the per-thread queue to disk. This
345 * is used both internally before committing other records, and also on
346 * system call return. We drain all the ones we can find at the time when
347 * drain is requested, but don't keep draining after that as those events
348 * may be approximately "after" the current event.
349 */
350 static void
351 ktr_drain(struct thread *td)
352 {
353 struct ktr_request *queued_req;
354 STAILQ_HEAD(, ktr_request) local_queue;
355
356 ktrace_assert(td);
357 sx_assert(&ktrace_sx, SX_XLOCKED);
358
359 STAILQ_INIT(&local_queue); /* XXXRW: needed? */
360
361 if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) {
362 mtx_lock(&ktrace_mtx);
363 STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr);
364 mtx_unlock(&ktrace_mtx);
365
366 while ((queued_req = STAILQ_FIRST(&local_queue))) {
367 STAILQ_REMOVE_HEAD(&local_queue, ktr_list);
368 ktr_writerequest(td, queued_req);
369 ktr_freerequest(queued_req);
370 }
371 }
372 }
373
374 /*
375 * Submit a trace record for immediate commit to disk -- to be used only
376 * where entering VFS is OK. First drain any pending records that may have
377 * been cached in the thread.
378 */
379 static void
380 ktr_submitrequest(struct thread *td, struct ktr_request *req)
381 {
382
383 ktrace_assert(td);
384
385 sx_xlock(&ktrace_sx);
386 ktr_drain(td);
387 ktr_writerequest(td, req);
388 ktr_freerequest(req);
389 sx_xunlock(&ktrace_sx);
390 ktrace_exit(td);
391 }
392
393 static void
394 ktr_freerequest(struct ktr_request *req)
395 {
396
397 mtx_lock(&ktrace_mtx);
398 ktr_freerequest_locked(req);
399 mtx_unlock(&ktrace_mtx);
400 }
401
402 static void
403 ktr_freerequest_locked(struct ktr_request *req)
404 {
405
406 mtx_assert(&ktrace_mtx, MA_OWNED);
407 if (req->ktr_buffer != NULL)
408 free(req->ktr_buffer, M_KTRACE);
409 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
410 }
411
412 /*
413 * Disable tracing for a process and release all associated resources.
414 * The caller is responsible for releasing a reference on the returned
415 * vnode and credentials.
416 */
417 static void
418 ktr_freeproc(struct proc *p, struct ucred **uc, struct vnode **vp)
419 {
420 struct ktr_request *req;
421
422 PROC_LOCK_ASSERT(p, MA_OWNED);
423 mtx_assert(&ktrace_mtx, MA_OWNED);
424 *uc = p->p_tracecred;
425 p->p_tracecred = NULL;
426 if (vp != NULL)
427 *vp = p->p_tracevp;
428 p->p_tracevp = NULL;
429 p->p_traceflag = 0;
430 while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) {
431 STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list);
432 ktr_freerequest_locked(req);
433 }
434 }
435
436 void
437 ktrsyscall(code, narg, args)
438 int code, narg;
439 register_t args[];
440 {
441 struct ktr_request *req;
442 struct ktr_syscall *ktp;
443 size_t buflen;
444 char *buf = NULL;
445
446 buflen = sizeof(register_t) * narg;
447 if (buflen > 0) {
448 buf = malloc(buflen, M_KTRACE, M_WAITOK);
449 bcopy(args, buf, buflen);
450 }
451 req = ktr_getrequest(KTR_SYSCALL);
452 if (req == NULL) {
453 if (buf != NULL)
454 free(buf, M_KTRACE);
455 return;
456 }
457 ktp = &req->ktr_data.ktr_syscall;
458 ktp->ktr_code = code;
459 ktp->ktr_narg = narg;
460 if (buflen > 0) {
461 req->ktr_header.ktr_len = buflen;
462 req->ktr_buffer = buf;
463 }
464 ktr_submitrequest(curthread, req);
465 }
466
467 void
468 ktrsysret(code, error, retval)
469 int code, error;
470 register_t retval;
471 {
472 struct ktr_request *req;
473 struct ktr_sysret *ktp;
474
475 req = ktr_getrequest(KTR_SYSRET);
476 if (req == NULL)
477 return;
478 ktp = &req->ktr_data.ktr_sysret;
479 ktp->ktr_code = code;
480 ktp->ktr_error = error;
481 ktp->ktr_retval = ((error == 0) ? retval: 0); /* what about val2 ? */
482 ktr_submitrequest(curthread, req);
483 }
484
485 /*
486 * When a setuid process execs, disable tracing.
487 *
488 * XXX: We toss any pending asynchronous records.
489 */
490 void
491 ktrprocexec(struct proc *p, struct ucred **uc, struct vnode **vp)
492 {
493
494 PROC_LOCK_ASSERT(p, MA_OWNED);
495 mtx_lock(&ktrace_mtx);
496 ktr_freeproc(p, uc, vp);
497 mtx_unlock(&ktrace_mtx);
498 }
499
500 /*
501 * When a process exits, drain per-process asynchronous trace records
502 * and disable tracing.
503 */
504 void
505 ktrprocexit(struct thread *td)
506 {
507 struct ktr_request *req;
508 struct proc *p;
509 struct ucred *cred;
510 struct vnode *vp;
511 int vfslocked;
512
513 p = td->td_proc;
514 if (p->p_traceflag == 0)
515 return;
516
517 ktrace_enter(td);
518 req = ktr_getrequest_entered(td, KTR_PROCDTOR);
519 if (req != NULL)
520 ktr_enqueuerequest(td, req);
521 sx_xlock(&ktrace_sx);
522 ktr_drain(td);
523 sx_xunlock(&ktrace_sx);
524 PROC_LOCK(p);
525 mtx_lock(&ktrace_mtx);
526 ktr_freeproc(p, &cred, &vp);
527 mtx_unlock(&ktrace_mtx);
528 PROC_UNLOCK(p);
529 if (vp != NULL) {
530 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
531 vrele(vp);
532 VFS_UNLOCK_GIANT(vfslocked);
533 }
534 if (cred != NULL)
535 crfree(cred);
536 ktrace_exit(td);
537 }
538
539 static void
540 ktrprocctor_entered(struct thread *td, struct proc *p)
541 {
542 struct ktr_proc_ctor *ktp;
543 struct ktr_request *req;
544 struct thread *td2;
545
546 ktrace_assert(td);
547 td2 = FIRST_THREAD_IN_PROC(p);
548 req = ktr_getrequest_entered(td2, KTR_PROCCTOR);
549 if (req == NULL)
550 return;
551 ktp = &req->ktr_data.ktr_proc_ctor;
552 ktp->sv_flags = p->p_sysent->sv_flags;
553 ktr_enqueuerequest(td2, req);
554 }
555
556 void
557 ktrprocctor(struct proc *p)
558 {
559 struct thread *td = curthread;
560
561 if ((p->p_traceflag & KTRFAC_MASK) == 0)
562 return;
563
564 ktrace_enter(td);
565 ktrprocctor_entered(td, p);
566 ktrace_exit(td);
567 }
568
569 /*
570 * When a process forks, enable tracing in the new process if needed.
571 */
572 void
573 ktrprocfork(struct proc *p1, struct proc *p2)
574 {
575
576 PROC_LOCK(p1);
577 mtx_lock(&ktrace_mtx);
578 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
579 if (p1->p_traceflag & KTRFAC_INHERIT) {
580 p2->p_traceflag = p1->p_traceflag;
581 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
582 VREF(p2->p_tracevp);
583 KASSERT(p1->p_tracecred != NULL,
584 ("ktrace vnode with no cred"));
585 p2->p_tracecred = crhold(p1->p_tracecred);
586 }
587 }
588 mtx_unlock(&ktrace_mtx);
589 PROC_UNLOCK(p1);
590
591 ktrprocctor(p2);
592 }
593
594 /*
595 * When a thread returns, drain any asynchronous records generated by the
596 * system call.
597 */
598 void
599 ktruserret(struct thread *td)
600 {
601
602 ktrace_enter(td);
603 sx_xlock(&ktrace_sx);
604 ktr_drain(td);
605 sx_xunlock(&ktrace_sx);
606 ktrace_exit(td);
607 }
608
609 void
610 ktrnamei(path)
611 char *path;
612 {
613 struct ktr_request *req;
614 int namelen;
615 char *buf = NULL;
616
617 namelen = strlen(path);
618 if (namelen > 0) {
619 buf = malloc(namelen, M_KTRACE, M_WAITOK);
620 bcopy(path, buf, namelen);
621 }
622 req = ktr_getrequest(KTR_NAMEI);
623 if (req == NULL) {
624 if (buf != NULL)
625 free(buf, M_KTRACE);
626 return;
627 }
628 if (namelen > 0) {
629 req->ktr_header.ktr_len = namelen;
630 req->ktr_buffer = buf;
631 }
632 ktr_submitrequest(curthread, req);
633 }
634
635 void
636 ktrsysctl(name, namelen)
637 int *name;
638 u_int namelen;
639 {
640 struct ktr_request *req;
641 u_int mib[CTL_MAXNAME + 2];
642 char *mibname;
643 size_t mibnamelen;
644 int error;
645
646 /* Lookup name of mib. */
647 KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long"));
648 mib[0] = 0;
649 mib[1] = 1;
650 bcopy(name, mib + 2, namelen * sizeof(*name));
651 mibnamelen = 128;
652 mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK);
653 error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen,
654 NULL, 0, &mibnamelen, 0);
655 if (error) {
656 free(mibname, M_KTRACE);
657 return;
658 }
659 req = ktr_getrequest(KTR_SYSCTL);
660 if (req == NULL) {
661 free(mibname, M_KTRACE);
662 return;
663 }
664 req->ktr_header.ktr_len = mibnamelen;
665 req->ktr_buffer = mibname;
666 ktr_submitrequest(curthread, req);
667 }
668
669 void
670 ktrgenio(fd, rw, uio, error)
671 int fd;
672 enum uio_rw rw;
673 struct uio *uio;
674 int error;
675 {
676 struct ktr_request *req;
677 struct ktr_genio *ktg;
678 int datalen;
679 char *buf;
680
681 if (error) {
682 free(uio, M_IOV);
683 return;
684 }
685 uio->uio_offset = 0;
686 uio->uio_rw = UIO_WRITE;
687 datalen = imin(uio->uio_resid, ktr_geniosize);
688 buf = malloc(datalen, M_KTRACE, M_WAITOK);
689 error = uiomove(buf, datalen, uio);
690 free(uio, M_IOV);
691 if (error) {
692 free(buf, M_KTRACE);
693 return;
694 }
695 req = ktr_getrequest(KTR_GENIO);
696 if (req == NULL) {
697 free(buf, M_KTRACE);
698 return;
699 }
700 ktg = &req->ktr_data.ktr_genio;
701 ktg->ktr_fd = fd;
702 ktg->ktr_rw = rw;
703 req->ktr_header.ktr_len = datalen;
704 req->ktr_buffer = buf;
705 ktr_submitrequest(curthread, req);
706 }
707
708 void
709 ktrpsig(sig, action, mask, code)
710 int sig;
711 sig_t action;
712 sigset_t *mask;
713 int code;
714 {
715 struct thread *td = curthread;
716 struct ktr_request *req;
717 struct ktr_psig *kp;
718
719 req = ktr_getrequest(KTR_PSIG);
720 if (req == NULL)
721 return;
722 kp = &req->ktr_data.ktr_psig;
723 kp->signo = (char)sig;
724 kp->action = action;
725 kp->mask = *mask;
726 kp->code = code;
727 ktr_enqueuerequest(td, req);
728 ktrace_exit(td);
729 }
730
731 void
732 ktrcsw(out, user, wmesg)
733 int out, user;
734 const char *wmesg;
735 {
736 struct thread *td = curthread;
737 struct ktr_request *req;
738 struct ktr_csw *kc;
739
740 req = ktr_getrequest(KTR_CSW);
741 if (req == NULL)
742 return;
743 kc = &req->ktr_data.ktr_csw;
744 kc->out = out;
745 kc->user = user;
746 if (wmesg != NULL)
747 strlcpy(kc->wmesg, wmesg, sizeof(kc->wmesg));
748 else
749 bzero(kc->wmesg, sizeof(kc->wmesg));
750 ktr_enqueuerequest(td, req);
751 ktrace_exit(td);
752 }
753
754 void
755 ktrstruct(name, namelen, data, datalen)
756 const char *name;
757 size_t namelen;
758 void *data;
759 size_t datalen;
760 {
761 struct ktr_request *req;
762 char *buf = NULL;
763 size_t buflen;
764
765 if (!data)
766 datalen = 0;
767 buflen = namelen + 1 + datalen;
768 buf = malloc(buflen, M_KTRACE, M_WAITOK);
769 bcopy(name, buf, namelen);
770 buf[namelen] = '\0';
771 bcopy(data, buf + namelen + 1, datalen);
772 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) {
773 free(buf, M_KTRACE);
774 return;
775 }
776 req->ktr_buffer = buf;
777 req->ktr_header.ktr_len = buflen;
778 ktr_submitrequest(curthread, req);
779 }
780
781 void
782 ktrfault(vaddr, type)
783 vm_offset_t vaddr;
784 int type;
785 {
786 struct thread *td = curthread;
787 struct ktr_request *req;
788 struct ktr_fault *kf;
789
790 req = ktr_getrequest(KTR_FAULT);
791 if (req == NULL)
792 return;
793 kf = &req->ktr_data.ktr_fault;
794 kf->vaddr = vaddr;
795 kf->type = type;
796 ktr_enqueuerequest(td, req);
797 ktrace_exit(td);
798 }
799
800 void
801 ktrfaultend(result)
802 int result;
803 {
804 struct thread *td = curthread;
805 struct ktr_request *req;
806 struct ktr_faultend *kf;
807
808 req = ktr_getrequest(KTR_FAULTEND);
809 if (req == NULL)
810 return;
811 kf = &req->ktr_data.ktr_faultend;
812 kf->result = result;
813 ktr_enqueuerequest(td, req);
814 ktrace_exit(td);
815 }
816 #endif /* KTRACE */
817
818 /* Interface and common routines */
819
820 #ifndef _SYS_SYSPROTO_H_
821 struct ktrace_args {
822 char *fname;
823 int ops;
824 int facs;
825 int pid;
826 };
827 #endif
828 /* ARGSUSED */
829 int
830 ktrace(td, uap)
831 struct thread *td;
832 register struct ktrace_args *uap;
833 {
834 #ifdef KTRACE
835 register struct vnode *vp = NULL;
836 register struct proc *p;
837 struct pgrp *pg;
838 int facs = uap->facs & ~KTRFAC_ROOT;
839 int ops = KTROP(uap->ops);
840 int descend = uap->ops & KTRFLAG_DESCEND;
841 int nfound, ret = 0;
842 int flags, error = 0, vfslocked;
843 struct nameidata nd;
844 struct ucred *cred;
845
846 /*
847 * Need something to (un)trace.
848 */
849 if (ops != KTROP_CLEARFILE && facs == 0)
850 return (EINVAL);
851
852 ktrace_enter(td);
853 if (ops != KTROP_CLEAR) {
854 /*
855 * an operation which requires a file argument.
856 */
857 NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_USERSPACE,
858 uap->fname, td);
859 flags = FREAD | FWRITE | O_NOFOLLOW;
860 error = vn_open(&nd, &flags, 0, NULL);
861 if (error) {
862 ktrace_exit(td);
863 return (error);
864 }
865 vfslocked = NDHASGIANT(&nd);
866 NDFREE(&nd, NDF_ONLY_PNBUF);
867 vp = nd.ni_vp;
868 VOP_UNLOCK(vp, 0);
869 if (vp->v_type != VREG) {
870 (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
871 VFS_UNLOCK_GIANT(vfslocked);
872 ktrace_exit(td);
873 return (EACCES);
874 }
875 VFS_UNLOCK_GIANT(vfslocked);
876 }
877 /*
878 * Clear all uses of the tracefile.
879 */
880 if (ops == KTROP_CLEARFILE) {
881 int vrele_count;
882
883 vrele_count = 0;
884 sx_slock(&allproc_lock);
885 FOREACH_PROC_IN_SYSTEM(p) {
886 PROC_LOCK(p);
887 if (p->p_tracevp == vp) {
888 if (ktrcanset(td, p)) {
889 mtx_lock(&ktrace_mtx);
890 ktr_freeproc(p, &cred, NULL);
891 mtx_unlock(&ktrace_mtx);
892 vrele_count++;
893 crfree(cred);
894 } else
895 error = EPERM;
896 }
897 PROC_UNLOCK(p);
898 }
899 sx_sunlock(&allproc_lock);
900 if (vrele_count > 0) {
901 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
902 while (vrele_count-- > 0)
903 vrele(vp);
904 VFS_UNLOCK_GIANT(vfslocked);
905 }
906 goto done;
907 }
908 /*
909 * do it
910 */
911 sx_slock(&proctree_lock);
912 if (uap->pid < 0) {
913 /*
914 * by process group
915 */
916 pg = pgfind(-uap->pid);
917 if (pg == NULL) {
918 sx_sunlock(&proctree_lock);
919 error = ESRCH;
920 goto done;
921 }
922 /*
923 * ktrops() may call vrele(). Lock pg_members
924 * by the proctree_lock rather than pg_mtx.
925 */
926 PGRP_UNLOCK(pg);
927 nfound = 0;
928 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
929 PROC_LOCK(p);
930 if (p->p_state == PRS_NEW ||
931 p_cansee(td, p) != 0) {
932 PROC_UNLOCK(p);
933 continue;
934 }
935 PROC_UNLOCK(p);
936 nfound++;
937 if (descend)
938 ret |= ktrsetchildren(td, p, ops, facs, vp);
939 else
940 ret |= ktrops(td, p, ops, facs, vp);
941 }
942 if (nfound == 0) {
943 sx_sunlock(&proctree_lock);
944 error = ESRCH;
945 goto done;
946 }
947 } else {
948 /*
949 * by pid
950 */
951 p = pfind(uap->pid);
952 if (p == NULL) {
953 sx_sunlock(&proctree_lock);
954 error = ESRCH;
955 goto done;
956 }
957 error = p_cansee(td, p);
958 /*
959 * The slock of the proctree lock will keep this process
960 * from going away, so unlocking the proc here is ok.
961 */
962 PROC_UNLOCK(p);
963 if (error) {
964 sx_sunlock(&proctree_lock);
965 goto done;
966 }
967 if (descend)
968 ret |= ktrsetchildren(td, p, ops, facs, vp);
969 else
970 ret |= ktrops(td, p, ops, facs, vp);
971 }
972 sx_sunlock(&proctree_lock);
973 if (!ret)
974 error = EPERM;
975 done:
976 if (vp != NULL) {
977 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
978 (void) vn_close(vp, FWRITE, td->td_ucred, td);
979 VFS_UNLOCK_GIANT(vfslocked);
980 }
981 ktrace_exit(td);
982 return (error);
983 #else /* !KTRACE */
984 return (ENOSYS);
985 #endif /* KTRACE */
986 }
987
988 /* ARGSUSED */
989 int
990 utrace(td, uap)
991 struct thread *td;
992 register struct utrace_args *uap;
993 {
994
995 #ifdef KTRACE
996 struct ktr_request *req;
997 void *cp;
998 int error;
999
1000 if (!KTRPOINT(td, KTR_USER))
1001 return (0);
1002 if (uap->len > KTR_USER_MAXLEN)
1003 return (EINVAL);
1004 cp = malloc(uap->len, M_KTRACE, M_WAITOK);
1005 error = copyin(uap->addr, cp, uap->len);
1006 if (error) {
1007 free(cp, M_KTRACE);
1008 return (error);
1009 }
1010 req = ktr_getrequest(KTR_USER);
1011 if (req == NULL) {
1012 free(cp, M_KTRACE);
1013 return (ENOMEM);
1014 }
1015 req->ktr_buffer = cp;
1016 req->ktr_header.ktr_len = uap->len;
1017 ktr_submitrequest(td, req);
1018 return (0);
1019 #else /* !KTRACE */
1020 return (ENOSYS);
1021 #endif /* KTRACE */
1022 }
1023
1024 #ifdef KTRACE
1025 static int
1026 ktrops(td, p, ops, facs, vp)
1027 struct thread *td;
1028 struct proc *p;
1029 int ops, facs;
1030 struct vnode *vp;
1031 {
1032 struct vnode *tracevp = NULL;
1033 struct ucred *tracecred = NULL;
1034
1035 PROC_LOCK(p);
1036 if (!ktrcanset(td, p)) {
1037 PROC_UNLOCK(p);
1038 return (0);
1039 }
1040 mtx_lock(&ktrace_mtx);
1041 if (ops == KTROP_SET) {
1042 if (p->p_tracevp != vp) {
1043 /*
1044 * if trace file already in use, relinquish below
1045 */
1046 tracevp = p->p_tracevp;
1047 VREF(vp);
1048 p->p_tracevp = vp;
1049 }
1050 if (p->p_tracecred != td->td_ucred) {
1051 tracecred = p->p_tracecred;
1052 p->p_tracecred = crhold(td->td_ucred);
1053 }
1054 p->p_traceflag |= facs;
1055 if (priv_check(td, PRIV_KTRACE) == 0)
1056 p->p_traceflag |= KTRFAC_ROOT;
1057 } else {
1058 /* KTROP_CLEAR */
1059 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0)
1060 /* no more tracing */
1061 ktr_freeproc(p, &tracecred, &tracevp);
1062 }
1063 mtx_unlock(&ktrace_mtx);
1064 if ((p->p_traceflag & KTRFAC_MASK) != 0)
1065 ktrprocctor_entered(td, p);
1066 PROC_UNLOCK(p);
1067 if (tracevp != NULL) {
1068 int vfslocked;
1069
1070 vfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
1071 vrele(tracevp);
1072 VFS_UNLOCK_GIANT(vfslocked);
1073 }
1074 if (tracecred != NULL)
1075 crfree(tracecred);
1076
1077 return (1);
1078 }
1079
1080 static int
1081 ktrsetchildren(td, top, ops, facs, vp)
1082 struct thread *td;
1083 struct proc *top;
1084 int ops, facs;
1085 struct vnode *vp;
1086 {
1087 register struct proc *p;
1088 register int ret = 0;
1089
1090 p = top;
1091 sx_assert(&proctree_lock, SX_LOCKED);
1092 for (;;) {
1093 ret |= ktrops(td, p, ops, facs, vp);
1094 /*
1095 * If this process has children, descend to them next,
1096 * otherwise do any siblings, and if done with this level,
1097 * follow back up the tree (but not past top).
1098 */
1099 if (!LIST_EMPTY(&p->p_children))
1100 p = LIST_FIRST(&p->p_children);
1101 else for (;;) {
1102 if (p == top)
1103 return (ret);
1104 if (LIST_NEXT(p, p_sibling)) {
1105 p = LIST_NEXT(p, p_sibling);
1106 break;
1107 }
1108 p = p->p_pptr;
1109 }
1110 }
1111 /*NOTREACHED*/
1112 }
1113
1114 static void
1115 ktr_writerequest(struct thread *td, struct ktr_request *req)
1116 {
1117 struct ktr_header *kth;
1118 struct vnode *vp;
1119 struct proc *p;
1120 struct ucred *cred;
1121 struct uio auio;
1122 struct iovec aiov[3];
1123 struct mount *mp;
1124 int datalen, buflen, vrele_count;
1125 int error, vfslocked;
1126
1127 /*
1128 * We hold the vnode and credential for use in I/O in case ktrace is
1129 * disabled on the process as we write out the request.
1130 *
1131 * XXXRW: This is not ideal: we could end up performing a write after
1132 * the vnode has been closed.
1133 */
1134 mtx_lock(&ktrace_mtx);
1135 vp = td->td_proc->p_tracevp;
1136 cred = td->td_proc->p_tracecred;
1137
1138 /*
1139 * If vp is NULL, the vp has been cleared out from under this
1140 * request, so just drop it. Make sure the credential and vnode are
1141 * in sync: we should have both or neither.
1142 */
1143 if (vp == NULL) {
1144 KASSERT(cred == NULL, ("ktr_writerequest: cred != NULL"));
1145 mtx_unlock(&ktrace_mtx);
1146 return;
1147 }
1148 VREF(vp);
1149 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL"));
1150 crhold(cred);
1151 mtx_unlock(&ktrace_mtx);
1152
1153 kth = &req->ktr_header;
1154 KASSERT(((u_short)kth->ktr_type & ~KTR_DROP) <
1155 sizeof(data_lengths) / sizeof(data_lengths[0]),
1156 ("data_lengths array overflow"));
1157 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP];
1158 buflen = kth->ktr_len;
1159 auio.uio_iov = &aiov[0];
1160 auio.uio_offset = 0;
1161 auio.uio_segflg = UIO_SYSSPACE;
1162 auio.uio_rw = UIO_WRITE;
1163 aiov[0].iov_base = (caddr_t)kth;
1164 aiov[0].iov_len = sizeof(struct ktr_header);
1165 auio.uio_resid = sizeof(struct ktr_header);
1166 auio.uio_iovcnt = 1;
1167 auio.uio_td = td;
1168 if (datalen != 0) {
1169 aiov[1].iov_base = (caddr_t)&req->ktr_data;
1170 aiov[1].iov_len = datalen;
1171 auio.uio_resid += datalen;
1172 auio.uio_iovcnt++;
1173 kth->ktr_len += datalen;
1174 }
1175 if (buflen != 0) {
1176 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write"));
1177 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer;
1178 aiov[auio.uio_iovcnt].iov_len = buflen;
1179 auio.uio_resid += buflen;
1180 auio.uio_iovcnt++;
1181 }
1182
1183 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1184 vn_start_write(vp, &mp, V_WAIT);
1185 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1186 #ifdef MAC
1187 error = mac_vnode_check_write(cred, NOCRED, vp);
1188 if (error == 0)
1189 #endif
1190 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred);
1191 VOP_UNLOCK(vp, 0);
1192 vn_finished_write(mp);
1193 crfree(cred);
1194 if (!error) {
1195 vrele(vp);
1196 VFS_UNLOCK_GIANT(vfslocked);
1197 return;
1198 }
1199 VFS_UNLOCK_GIANT(vfslocked);
1200
1201 /*
1202 * If error encountered, give up tracing on this vnode. We defer
1203 * all the vrele()'s on the vnode until after we are finished walking
1204 * the various lists to avoid needlessly holding locks.
1205 * NB: at this point we still hold the vnode reference that must
1206 * not go away as we need the valid vnode to compare with. Thus let
1207 * vrele_count start at 1 and the reference will be freed
1208 * by the loop at the end after our last use of vp.
1209 */
1210 log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n",
1211 error);
1212 vrele_count = 1;
1213 /*
1214 * First, clear this vnode from being used by any processes in the
1215 * system.
1216 * XXX - If one process gets an EPERM writing to the vnode, should
1217 * we really do this? Other processes might have suitable
1218 * credentials for the operation.
1219 */
1220 cred = NULL;
1221 sx_slock(&allproc_lock);
1222 FOREACH_PROC_IN_SYSTEM(p) {
1223 PROC_LOCK(p);
1224 if (p->p_tracevp == vp) {
1225 mtx_lock(&ktrace_mtx);
1226 ktr_freeproc(p, &cred, NULL);
1227 mtx_unlock(&ktrace_mtx);
1228 vrele_count++;
1229 }
1230 PROC_UNLOCK(p);
1231 if (cred != NULL) {
1232 crfree(cred);
1233 cred = NULL;
1234 }
1235 }
1236 sx_sunlock(&allproc_lock);
1237
1238 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1239 while (vrele_count-- > 0)
1240 vrele(vp);
1241 VFS_UNLOCK_GIANT(vfslocked);
1242 }
1243
1244 /*
1245 * Return true if caller has permission to set the ktracing state
1246 * of target. Essentially, the target can't possess any
1247 * more permissions than the caller. KTRFAC_ROOT signifies that
1248 * root previously set the tracing status on the target process, and
1249 * so, only root may further change it.
1250 */
1251 static int
1252 ktrcanset(td, targetp)
1253 struct thread *td;
1254 struct proc *targetp;
1255 {
1256
1257 PROC_LOCK_ASSERT(targetp, MA_OWNED);
1258 if (targetp->p_traceflag & KTRFAC_ROOT &&
1259 priv_check(td, PRIV_KTRACE))
1260 return (0);
1261
1262 if (p_candebug(td, targetp) != 0)
1263 return (0);
1264
1265 return (1);
1266 }
1267
1268 #endif /* KTRACE */
Cache object: 39b75bd5205e303f59132be35ff7165a
|