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: releng/9.2/sys/kern/kern_ktrace.c 267018 2014-06-03 19:03:11Z delphij $");
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 FEATURE(ktrace, "Kernel support for system-call tracing");
88
89 #ifndef KTRACE_REQUEST_POOL
90 #define KTRACE_REQUEST_POOL 100
91 #endif
92
93 struct ktr_request {
94 struct ktr_header ktr_header;
95 void *ktr_buffer;
96 union {
97 struct ktr_proc_ctor ktr_proc_ctor;
98 struct ktr_syscall ktr_syscall;
99 struct ktr_sysret ktr_sysret;
100 struct ktr_genio ktr_genio;
101 struct ktr_psig ktr_psig;
102 struct ktr_csw ktr_csw;
103 struct ktr_fault ktr_fault;
104 struct ktr_faultend ktr_faultend;
105 } ktr_data;
106 STAILQ_ENTRY(ktr_request) ktr_list;
107 };
108
109 static int data_lengths[] = {
110 0, /* none */
111 offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */
112 sizeof(struct ktr_sysret), /* KTR_SYSRET */
113 0, /* KTR_NAMEI */
114 sizeof(struct ktr_genio), /* KTR_GENIO */
115 sizeof(struct ktr_psig), /* KTR_PSIG */
116 sizeof(struct ktr_csw), /* KTR_CSW */
117 0, /* KTR_USER */
118 0, /* KTR_STRUCT */
119 0, /* KTR_SYSCTL */
120 sizeof(struct ktr_proc_ctor), /* KTR_PROCCTOR */
121 0, /* KTR_PROCDTOR */
122 0, /* unused */
123 sizeof(struct ktr_fault), /* KTR_FAULT */
124 sizeof(struct ktr_faultend), /* KTR_FAULTEND */
125 };
126
127 static STAILQ_HEAD(, ktr_request) ktr_free;
128
129 static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options");
130
131 static u_int ktr_requestpool = KTRACE_REQUEST_POOL;
132 TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool);
133
134 static u_int ktr_geniosize = PAGE_SIZE;
135 TUNABLE_INT("kern.ktrace.genio_size", &ktr_geniosize);
136 SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RW, &ktr_geniosize,
137 0, "Maximum size of genio event payload");
138
139 static int print_message = 1;
140 static struct mtx ktrace_mtx;
141 static struct sx ktrace_sx;
142
143 static void ktrace_init(void *dummy);
144 static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS);
145 static u_int ktrace_resize_pool(u_int oldsize, u_int newsize);
146 static struct ktr_request *ktr_getrequest_entered(struct thread *td, int type);
147 static struct ktr_request *ktr_getrequest(int type);
148 static void ktr_submitrequest(struct thread *td, struct ktr_request *req);
149 static void ktr_freeproc(struct proc *p, struct ucred **uc,
150 struct vnode **vp);
151 static void ktr_freerequest(struct ktr_request *req);
152 static void ktr_freerequest_locked(struct ktr_request *req);
153 static void ktr_writerequest(struct thread *td, struct ktr_request *req);
154 static int ktrcanset(struct thread *,struct proc *);
155 static int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *);
156 static int ktrops(struct thread *,struct proc *,int,int,struct vnode *);
157 static void ktrprocctor_entered(struct thread *, struct proc *);
158
159 /*
160 * ktrace itself generates events, such as context switches, which we do not
161 * wish to trace. Maintain a flag, TDP_INKTRACE, on each thread to determine
162 * whether or not it is in a region where tracing of events should be
163 * suppressed.
164 */
165 static void
166 ktrace_enter(struct thread *td)
167 {
168
169 KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set"));
170 td->td_pflags |= TDP_INKTRACE;
171 }
172
173 static void
174 ktrace_exit(struct thread *td)
175 {
176
177 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set"));
178 td->td_pflags &= ~TDP_INKTRACE;
179 }
180
181 static void
182 ktrace_assert(struct thread *td)
183 {
184
185 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set"));
186 }
187
188 static void
189 ktrace_init(void *dummy)
190 {
191 struct ktr_request *req;
192 int i;
193
194 mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET);
195 sx_init(&ktrace_sx, "ktrace_sx");
196 STAILQ_INIT(&ktr_free);
197 for (i = 0; i < ktr_requestpool; i++) {
198 req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK);
199 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
200 }
201 }
202 SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL);
203
204 static int
205 sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)
206 {
207 struct thread *td;
208 u_int newsize, oldsize, wantsize;
209 int error;
210
211 /* Handle easy read-only case first to avoid warnings from GCC. */
212 if (!req->newptr) {
213 oldsize = ktr_requestpool;
214 return (SYSCTL_OUT(req, &oldsize, sizeof(u_int)));
215 }
216
217 error = SYSCTL_IN(req, &wantsize, sizeof(u_int));
218 if (error)
219 return (error);
220 td = curthread;
221 ktrace_enter(td);
222 oldsize = ktr_requestpool;
223 newsize = ktrace_resize_pool(oldsize, wantsize);
224 ktrace_exit(td);
225 error = SYSCTL_OUT(req, &oldsize, sizeof(u_int));
226 if (error)
227 return (error);
228 if (wantsize > oldsize && newsize < wantsize)
229 return (ENOSPC);
230 return (0);
231 }
232 SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW,
233 &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU",
234 "Pool buffer size for ktrace(1)");
235
236 static u_int
237 ktrace_resize_pool(u_int oldsize, u_int newsize)
238 {
239 STAILQ_HEAD(, ktr_request) ktr_new;
240 struct ktr_request *req;
241 int bound;
242
243 print_message = 1;
244 bound = newsize - oldsize;
245 if (bound == 0)
246 return (ktr_requestpool);
247 if (bound < 0) {
248 mtx_lock(&ktrace_mtx);
249 /* Shrink pool down to newsize if possible. */
250 while (bound++ < 0) {
251 req = STAILQ_FIRST(&ktr_free);
252 if (req == NULL)
253 break;
254 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
255 ktr_requestpool--;
256 free(req, M_KTRACE);
257 }
258 } else {
259 /* Grow pool up to newsize. */
260 STAILQ_INIT(&ktr_new);
261 while (bound-- > 0) {
262 req = malloc(sizeof(struct ktr_request), M_KTRACE,
263 M_WAITOK);
264 STAILQ_INSERT_HEAD(&ktr_new, req, ktr_list);
265 }
266 mtx_lock(&ktrace_mtx);
267 STAILQ_CONCAT(&ktr_free, &ktr_new);
268 ktr_requestpool += (newsize - oldsize);
269 }
270 mtx_unlock(&ktrace_mtx);
271 return (ktr_requestpool);
272 }
273
274 /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */
275 CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) ==
276 (sizeof((struct thread *)NULL)->td_name));
277
278 static struct ktr_request *
279 ktr_getrequest_entered(struct thread *td, int type)
280 {
281 struct ktr_request *req;
282 struct proc *p = td->td_proc;
283 int pm;
284
285 mtx_lock(&ktrace_mtx);
286 if (!KTRCHECK(td, type)) {
287 mtx_unlock(&ktrace_mtx);
288 return (NULL);
289 }
290 req = STAILQ_FIRST(&ktr_free);
291 if (req != NULL) {
292 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
293 req->ktr_header.ktr_type = type;
294 if (p->p_traceflag & KTRFAC_DROP) {
295 req->ktr_header.ktr_type |= KTR_DROP;
296 p->p_traceflag &= ~KTRFAC_DROP;
297 }
298 mtx_unlock(&ktrace_mtx);
299 microtime(&req->ktr_header.ktr_time);
300 req->ktr_header.ktr_pid = p->p_pid;
301 req->ktr_header.ktr_tid = td->td_tid;
302 bcopy(td->td_name, req->ktr_header.ktr_comm,
303 sizeof(req->ktr_header.ktr_comm));
304 req->ktr_buffer = NULL;
305 req->ktr_header.ktr_len = 0;
306 } else {
307 p->p_traceflag |= KTRFAC_DROP;
308 pm = print_message;
309 print_message = 0;
310 mtx_unlock(&ktrace_mtx);
311 if (pm)
312 printf("Out of ktrace request objects.\n");
313 }
314 return (req);
315 }
316
317 static struct ktr_request *
318 ktr_getrequest(int type)
319 {
320 struct thread *td = curthread;
321 struct ktr_request *req;
322
323 ktrace_enter(td);
324 req = ktr_getrequest_entered(td, type);
325 if (req == NULL)
326 ktrace_exit(td);
327
328 return (req);
329 }
330
331 /*
332 * Some trace generation environments don't permit direct access to VFS,
333 * such as during a context switch where sleeping is not allowed. Under these
334 * circumstances, queue a request to the thread to be written asynchronously
335 * later.
336 */
337 static void
338 ktr_enqueuerequest(struct thread *td, struct ktr_request *req)
339 {
340
341 mtx_lock(&ktrace_mtx);
342 STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list);
343 mtx_unlock(&ktrace_mtx);
344 }
345
346 /*
347 * Drain any pending ktrace records from the per-thread queue to disk. This
348 * is used both internally before committing other records, and also on
349 * system call return. We drain all the ones we can find at the time when
350 * drain is requested, but don't keep draining after that as those events
351 * may be approximately "after" the current event.
352 */
353 static void
354 ktr_drain(struct thread *td)
355 {
356 struct ktr_request *queued_req;
357 STAILQ_HEAD(, ktr_request) local_queue;
358
359 ktrace_assert(td);
360 sx_assert(&ktrace_sx, SX_XLOCKED);
361
362 STAILQ_INIT(&local_queue);
363
364 if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) {
365 mtx_lock(&ktrace_mtx);
366 STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr);
367 mtx_unlock(&ktrace_mtx);
368
369 while ((queued_req = STAILQ_FIRST(&local_queue))) {
370 STAILQ_REMOVE_HEAD(&local_queue, ktr_list);
371 ktr_writerequest(td, queued_req);
372 ktr_freerequest(queued_req);
373 }
374 }
375 }
376
377 /*
378 * Submit a trace record for immediate commit to disk -- to be used only
379 * where entering VFS is OK. First drain any pending records that may have
380 * been cached in the thread.
381 */
382 static void
383 ktr_submitrequest(struct thread *td, struct ktr_request *req)
384 {
385
386 ktrace_assert(td);
387
388 sx_xlock(&ktrace_sx);
389 ktr_drain(td);
390 ktr_writerequest(td, req);
391 ktr_freerequest(req);
392 sx_xunlock(&ktrace_sx);
393 ktrace_exit(td);
394 }
395
396 static void
397 ktr_freerequest(struct ktr_request *req)
398 {
399
400 mtx_lock(&ktrace_mtx);
401 ktr_freerequest_locked(req);
402 mtx_unlock(&ktrace_mtx);
403 }
404
405 static void
406 ktr_freerequest_locked(struct ktr_request *req)
407 {
408
409 mtx_assert(&ktrace_mtx, MA_OWNED);
410 if (req->ktr_buffer != NULL)
411 free(req->ktr_buffer, M_KTRACE);
412 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
413 }
414
415 /*
416 * Disable tracing for a process and release all associated resources.
417 * The caller is responsible for releasing a reference on the returned
418 * vnode and credentials.
419 */
420 static void
421 ktr_freeproc(struct proc *p, struct ucred **uc, struct vnode **vp)
422 {
423 struct ktr_request *req;
424
425 PROC_LOCK_ASSERT(p, MA_OWNED);
426 mtx_assert(&ktrace_mtx, MA_OWNED);
427 *uc = p->p_tracecred;
428 p->p_tracecred = NULL;
429 if (vp != NULL)
430 *vp = p->p_tracevp;
431 p->p_tracevp = NULL;
432 p->p_traceflag = 0;
433 while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) {
434 STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list);
435 ktr_freerequest_locked(req);
436 }
437 }
438
439 void
440 ktrsyscall(code, narg, args)
441 int code, narg;
442 register_t args[];
443 {
444 struct ktr_request *req;
445 struct ktr_syscall *ktp;
446 size_t buflen;
447 char *buf = NULL;
448
449 buflen = sizeof(register_t) * narg;
450 if (buflen > 0) {
451 buf = malloc(buflen, M_KTRACE, M_WAITOK);
452 bcopy(args, buf, buflen);
453 }
454 req = ktr_getrequest(KTR_SYSCALL);
455 if (req == NULL) {
456 if (buf != NULL)
457 free(buf, M_KTRACE);
458 return;
459 }
460 ktp = &req->ktr_data.ktr_syscall;
461 ktp->ktr_code = code;
462 ktp->ktr_narg = narg;
463 if (buflen > 0) {
464 req->ktr_header.ktr_len = buflen;
465 req->ktr_buffer = buf;
466 }
467 ktr_submitrequest(curthread, req);
468 }
469
470 void
471 ktrsysret(code, error, retval)
472 int code, error;
473 register_t retval;
474 {
475 struct ktr_request *req;
476 struct ktr_sysret *ktp;
477
478 req = ktr_getrequest(KTR_SYSRET);
479 if (req == NULL)
480 return;
481 ktp = &req->ktr_data.ktr_sysret;
482 ktp->ktr_code = code;
483 ktp->ktr_error = error;
484 ktp->ktr_retval = ((error == 0) ? retval: 0); /* what about val2 ? */
485 ktr_submitrequest(curthread, req);
486 }
487
488 /*
489 * When a setuid process execs, disable tracing.
490 *
491 * XXX: We toss any pending asynchronous records.
492 */
493 void
494 ktrprocexec(struct proc *p, struct ucred **uc, struct vnode **vp)
495 {
496
497 PROC_LOCK_ASSERT(p, MA_OWNED);
498 mtx_lock(&ktrace_mtx);
499 ktr_freeproc(p, uc, vp);
500 mtx_unlock(&ktrace_mtx);
501 }
502
503 /*
504 * When a process exits, drain per-process asynchronous trace records
505 * and disable tracing.
506 */
507 void
508 ktrprocexit(struct thread *td)
509 {
510 struct ktr_request *req;
511 struct proc *p;
512 struct ucred *cred;
513 struct vnode *vp;
514 int vfslocked;
515
516 p = td->td_proc;
517 if (p->p_traceflag == 0)
518 return;
519
520 ktrace_enter(td);
521 req = ktr_getrequest_entered(td, KTR_PROCDTOR);
522 if (req != NULL)
523 ktr_enqueuerequest(td, req);
524 sx_xlock(&ktrace_sx);
525 ktr_drain(td);
526 sx_xunlock(&ktrace_sx);
527 PROC_LOCK(p);
528 mtx_lock(&ktrace_mtx);
529 ktr_freeproc(p, &cred, &vp);
530 mtx_unlock(&ktrace_mtx);
531 PROC_UNLOCK(p);
532 if (vp != NULL) {
533 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
534 vrele(vp);
535 VFS_UNLOCK_GIANT(vfslocked);
536 }
537 if (cred != NULL)
538 crfree(cred);
539 ktrace_exit(td);
540 }
541
542 static void
543 ktrprocctor_entered(struct thread *td, struct proc *p)
544 {
545 struct ktr_proc_ctor *ktp;
546 struct ktr_request *req;
547 struct thread *td2;
548
549 ktrace_assert(td);
550 td2 = FIRST_THREAD_IN_PROC(p);
551 req = ktr_getrequest_entered(td2, KTR_PROCCTOR);
552 if (req == NULL)
553 return;
554 ktp = &req->ktr_data.ktr_proc_ctor;
555 ktp->sv_flags = p->p_sysent->sv_flags;
556 ktr_enqueuerequest(td2, req);
557 }
558
559 void
560 ktrprocctor(struct proc *p)
561 {
562 struct thread *td = curthread;
563
564 if ((p->p_traceflag & KTRFAC_MASK) == 0)
565 return;
566
567 ktrace_enter(td);
568 ktrprocctor_entered(td, p);
569 ktrace_exit(td);
570 }
571
572 /*
573 * When a process forks, enable tracing in the new process if needed.
574 */
575 void
576 ktrprocfork(struct proc *p1, struct proc *p2)
577 {
578
579 PROC_LOCK(p1);
580 mtx_lock(&ktrace_mtx);
581 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
582 if (p1->p_traceflag & KTRFAC_INHERIT) {
583 p2->p_traceflag = p1->p_traceflag;
584 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
585 VREF(p2->p_tracevp);
586 KASSERT(p1->p_tracecred != NULL,
587 ("ktrace vnode with no cred"));
588 p2->p_tracecred = crhold(p1->p_tracecred);
589 }
590 }
591 mtx_unlock(&ktrace_mtx);
592 PROC_UNLOCK(p1);
593
594 ktrprocctor(p2);
595 }
596
597 /*
598 * When a thread returns, drain any asynchronous records generated by the
599 * system call.
600 */
601 void
602 ktruserret(struct thread *td)
603 {
604
605 ktrace_enter(td);
606 sx_xlock(&ktrace_sx);
607 ktr_drain(td);
608 sx_xunlock(&ktrace_sx);
609 ktrace_exit(td);
610 }
611
612 void
613 ktrnamei(path)
614 char *path;
615 {
616 struct ktr_request *req;
617 int namelen;
618 char *buf = NULL;
619
620 namelen = strlen(path);
621 if (namelen > 0) {
622 buf = malloc(namelen, M_KTRACE, M_WAITOK);
623 bcopy(path, buf, namelen);
624 }
625 req = ktr_getrequest(KTR_NAMEI);
626 if (req == NULL) {
627 if (buf != NULL)
628 free(buf, M_KTRACE);
629 return;
630 }
631 if (namelen > 0) {
632 req->ktr_header.ktr_len = namelen;
633 req->ktr_buffer = buf;
634 }
635 ktr_submitrequest(curthread, req);
636 }
637
638 void
639 ktrsysctl(name, namelen)
640 int *name;
641 u_int namelen;
642 {
643 struct ktr_request *req;
644 u_int mib[CTL_MAXNAME + 2];
645 char *mibname;
646 size_t mibnamelen;
647 int error;
648
649 /* Lookup name of mib. */
650 KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long"));
651 mib[0] = 0;
652 mib[1] = 1;
653 bcopy(name, mib + 2, namelen * sizeof(*name));
654 mibnamelen = 128;
655 mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK);
656 error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen,
657 NULL, 0, &mibnamelen, 0);
658 if (error) {
659 free(mibname, M_KTRACE);
660 return;
661 }
662 req = ktr_getrequest(KTR_SYSCTL);
663 if (req == NULL) {
664 free(mibname, M_KTRACE);
665 return;
666 }
667 req->ktr_header.ktr_len = mibnamelen;
668 req->ktr_buffer = mibname;
669 ktr_submitrequest(curthread, req);
670 }
671
672 void
673 ktrgenio(fd, rw, uio, error)
674 int fd;
675 enum uio_rw rw;
676 struct uio *uio;
677 int error;
678 {
679 struct ktr_request *req;
680 struct ktr_genio *ktg;
681 int datalen;
682 char *buf;
683
684 if (error) {
685 free(uio, M_IOV);
686 return;
687 }
688 uio->uio_offset = 0;
689 uio->uio_rw = UIO_WRITE;
690 datalen = MIN(uio->uio_resid, ktr_geniosize);
691 buf = malloc(datalen, M_KTRACE, M_WAITOK);
692 error = uiomove(buf, datalen, uio);
693 free(uio, M_IOV);
694 if (error) {
695 free(buf, M_KTRACE);
696 return;
697 }
698 req = ktr_getrequest(KTR_GENIO);
699 if (req == NULL) {
700 free(buf, M_KTRACE);
701 return;
702 }
703 ktg = &req->ktr_data.ktr_genio;
704 ktg->ktr_fd = fd;
705 ktg->ktr_rw = rw;
706 req->ktr_header.ktr_len = datalen;
707 req->ktr_buffer = buf;
708 ktr_submitrequest(curthread, req);
709 }
710
711 void
712 ktrpsig(sig, action, mask, code)
713 int sig;
714 sig_t action;
715 sigset_t *mask;
716 int code;
717 {
718 struct thread *td = curthread;
719 struct ktr_request *req;
720 struct ktr_psig *kp;
721
722 req = ktr_getrequest(KTR_PSIG);
723 if (req == NULL)
724 return;
725 kp = &req->ktr_data.ktr_psig;
726 kp->signo = (char)sig;
727 kp->action = action;
728 kp->mask = *mask;
729 kp->code = code;
730 ktr_enqueuerequest(td, req);
731 ktrace_exit(td);
732 }
733
734 void
735 ktrcsw(out, user, wmesg)
736 int out, user;
737 const char *wmesg;
738 {
739 struct thread *td = curthread;
740 struct ktr_request *req;
741 struct ktr_csw *kc;
742
743 req = ktr_getrequest(KTR_CSW);
744 if (req == NULL)
745 return;
746 kc = &req->ktr_data.ktr_csw;
747 kc->out = out;
748 kc->user = user;
749 if (wmesg != NULL)
750 strlcpy(kc->wmesg, wmesg, sizeof(kc->wmesg));
751 else
752 bzero(kc->wmesg, sizeof(kc->wmesg));
753 ktr_enqueuerequest(td, req);
754 ktrace_exit(td);
755 }
756
757 void
758 ktrstruct(name, data, datalen)
759 const char *name;
760 void *data;
761 size_t datalen;
762 {
763 struct ktr_request *req;
764 char *buf = NULL;
765 size_t buflen;
766
767 if (!data)
768 datalen = 0;
769 buflen = strlen(name) + 1 + datalen;
770 buf = malloc(buflen, M_KTRACE, M_WAITOK);
771 strcpy(buf, name);
772 bcopy(data, buf + strlen(name) + 1, datalen);
773 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) {
774 free(buf, M_KTRACE);
775 return;
776 }
777 req->ktr_buffer = buf;
778 req->ktr_header.ktr_len = buflen;
779 ktr_submitrequest(curthread, req);
780 }
781
782 void
783 ktrfault(vaddr, type)
784 vm_offset_t vaddr;
785 int type;
786 {
787 struct thread *td = curthread;
788 struct ktr_request *req;
789 struct ktr_fault *kf;
790
791 req = ktr_getrequest(KTR_FAULT);
792 if (req == NULL)
793 return;
794 kf = &req->ktr_data.ktr_fault;
795 kf->vaddr = vaddr;
796 kf->type = type;
797 ktr_enqueuerequest(td, req);
798 ktrace_exit(td);
799 }
800
801 void
802 ktrfaultend(result)
803 int result;
804 {
805 struct thread *td = curthread;
806 struct ktr_request *req;
807 struct ktr_faultend *kf;
808
809 req = ktr_getrequest(KTR_FAULTEND);
810 if (req == NULL)
811 return;
812 kf = &req->ktr_data.ktr_faultend;
813 kf->result = result;
814 ktr_enqueuerequest(td, req);
815 ktrace_exit(td);
816 }
817 #endif /* KTRACE */
818
819 /* Interface and common routines */
820
821 #ifndef _SYS_SYSPROTO_H_
822 struct ktrace_args {
823 char *fname;
824 int ops;
825 int facs;
826 int pid;
827 };
828 #endif
829 /* ARGSUSED */
830 int
831 sys_ktrace(td, uap)
832 struct thread *td;
833 register struct ktrace_args *uap;
834 {
835 #ifdef KTRACE
836 register struct vnode *vp = NULL;
837 register struct proc *p;
838 struct pgrp *pg;
839 int facs = uap->facs & ~KTRFAC_ROOT;
840 int ops = KTROP(uap->ops);
841 int descend = uap->ops & KTRFLAG_DESCEND;
842 int nfound, ret = 0;
843 int flags, error = 0, vfslocked;
844 struct nameidata nd;
845 struct ucred *cred;
846
847 /*
848 * Need something to (un)trace.
849 */
850 if (ops != KTROP_CLEARFILE && facs == 0)
851 return (EINVAL);
852
853 ktrace_enter(td);
854 if (ops != KTROP_CLEAR) {
855 /*
856 * an operation which requires a file argument.
857 */
858 NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_USERSPACE,
859 uap->fname, td);
860 flags = FREAD | FWRITE | O_NOFOLLOW;
861 error = vn_open(&nd, &flags, 0, NULL);
862 if (error) {
863 ktrace_exit(td);
864 return (error);
865 }
866 vfslocked = NDHASGIANT(&nd);
867 NDFREE(&nd, NDF_ONLY_PNBUF);
868 vp = nd.ni_vp;
869 VOP_UNLOCK(vp, 0);
870 if (vp->v_type != VREG) {
871 (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
872 VFS_UNLOCK_GIANT(vfslocked);
873 ktrace_exit(td);
874 return (EACCES);
875 }
876 VFS_UNLOCK_GIANT(vfslocked);
877 }
878 /*
879 * Clear all uses of the tracefile.
880 */
881 if (ops == KTROP_CLEARFILE) {
882 int vrele_count;
883
884 vrele_count = 0;
885 sx_slock(&allproc_lock);
886 FOREACH_PROC_IN_SYSTEM(p) {
887 PROC_LOCK(p);
888 if (p->p_tracevp == vp) {
889 if (ktrcanset(td, p)) {
890 mtx_lock(&ktrace_mtx);
891 ktr_freeproc(p, &cred, NULL);
892 mtx_unlock(&ktrace_mtx);
893 vrele_count++;
894 crfree(cred);
895 } else
896 error = EPERM;
897 }
898 PROC_UNLOCK(p);
899 }
900 sx_sunlock(&allproc_lock);
901 if (vrele_count > 0) {
902 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
903 while (vrele_count-- > 0)
904 vrele(vp);
905 VFS_UNLOCK_GIANT(vfslocked);
906 }
907 goto done;
908 }
909 /*
910 * do it
911 */
912 sx_slock(&proctree_lock);
913 if (uap->pid < 0) {
914 /*
915 * by process group
916 */
917 pg = pgfind(-uap->pid);
918 if (pg == NULL) {
919 sx_sunlock(&proctree_lock);
920 error = ESRCH;
921 goto done;
922 }
923 /*
924 * ktrops() may call vrele(). Lock pg_members
925 * by the proctree_lock rather than pg_mtx.
926 */
927 PGRP_UNLOCK(pg);
928 nfound = 0;
929 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
930 PROC_LOCK(p);
931 if (p->p_state == PRS_NEW ||
932 p_cansee(td, p) != 0) {
933 PROC_UNLOCK(p);
934 continue;
935 }
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 error = ESRCH;
954 else
955 error = p_cansee(td, p);
956 if (error) {
957 if (p != NULL)
958 PROC_UNLOCK(p);
959 sx_sunlock(&proctree_lock);
960 goto done;
961 }
962 if (descend)
963 ret |= ktrsetchildren(td, p, ops, facs, vp);
964 else
965 ret |= ktrops(td, p, ops, facs, vp);
966 }
967 sx_sunlock(&proctree_lock);
968 if (!ret)
969 error = EPERM;
970 done:
971 if (vp != NULL) {
972 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
973 (void) vn_close(vp, FWRITE, td->td_ucred, td);
974 VFS_UNLOCK_GIANT(vfslocked);
975 }
976 ktrace_exit(td);
977 return (error);
978 #else /* !KTRACE */
979 return (ENOSYS);
980 #endif /* KTRACE */
981 }
982
983 /* ARGSUSED */
984 int
985 sys_utrace(td, uap)
986 struct thread *td;
987 register struct utrace_args *uap;
988 {
989
990 #ifdef KTRACE
991 struct ktr_request *req;
992 void *cp;
993 int error;
994
995 if (!KTRPOINT(td, KTR_USER))
996 return (0);
997 if (uap->len > KTR_USER_MAXLEN)
998 return (EINVAL);
999 cp = malloc(uap->len, M_KTRACE, M_WAITOK);
1000 error = copyin(uap->addr, cp, uap->len);
1001 if (error) {
1002 free(cp, M_KTRACE);
1003 return (error);
1004 }
1005 req = ktr_getrequest(KTR_USER);
1006 if (req == NULL) {
1007 free(cp, M_KTRACE);
1008 return (ENOMEM);
1009 }
1010 req->ktr_buffer = cp;
1011 req->ktr_header.ktr_len = uap->len;
1012 ktr_submitrequest(td, req);
1013 return (0);
1014 #else /* !KTRACE */
1015 return (ENOSYS);
1016 #endif /* KTRACE */
1017 }
1018
1019 #ifdef KTRACE
1020 static int
1021 ktrops(td, p, ops, facs, vp)
1022 struct thread *td;
1023 struct proc *p;
1024 int ops, facs;
1025 struct vnode *vp;
1026 {
1027 struct vnode *tracevp = NULL;
1028 struct ucred *tracecred = NULL;
1029
1030 PROC_LOCK_ASSERT(p, MA_OWNED);
1031 if (!ktrcanset(td, p)) {
1032 PROC_UNLOCK(p);
1033 return (0);
1034 }
1035 if (p->p_flag & P_WEXIT) {
1036 /* If the process is exiting, just ignore it. */
1037 PROC_UNLOCK(p);
1038 return (1);
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 PROC_LOCK_ASSERT(p, MA_OWNED);
1092 sx_assert(&proctree_lock, SX_LOCKED);
1093 for (;;) {
1094 ret |= ktrops(td, p, ops, facs, vp);
1095 /*
1096 * If this process has children, descend to them next,
1097 * otherwise do any siblings, and if done with this level,
1098 * follow back up the tree (but not past top).
1099 */
1100 if (!LIST_EMPTY(&p->p_children))
1101 p = LIST_FIRST(&p->p_children);
1102 else for (;;) {
1103 if (p == top)
1104 return (ret);
1105 if (LIST_NEXT(p, p_sibling)) {
1106 p = LIST_NEXT(p, p_sibling);
1107 break;
1108 }
1109 p = p->p_pptr;
1110 }
1111 PROC_LOCK(p);
1112 }
1113 /*NOTREACHED*/
1114 }
1115
1116 static void
1117 ktr_writerequest(struct thread *td, struct ktr_request *req)
1118 {
1119 struct ktr_header *kth;
1120 struct vnode *vp;
1121 struct proc *p;
1122 struct ucred *cred;
1123 struct uio auio;
1124 struct iovec aiov[3];
1125 struct mount *mp;
1126 int datalen, buflen, vrele_count;
1127 int error, vfslocked;
1128
1129 /*
1130 * We hold the vnode and credential for use in I/O in case ktrace is
1131 * disabled on the process as we write out the request.
1132 *
1133 * XXXRW: This is not ideal: we could end up performing a write after
1134 * the vnode has been closed.
1135 */
1136 mtx_lock(&ktrace_mtx);
1137 vp = td->td_proc->p_tracevp;
1138 cred = td->td_proc->p_tracecred;
1139
1140 /*
1141 * If vp is NULL, the vp has been cleared out from under this
1142 * request, so just drop it. Make sure the credential and vnode are
1143 * in sync: we should have both or neither.
1144 */
1145 if (vp == NULL) {
1146 KASSERT(cred == NULL, ("ktr_writerequest: cred != NULL"));
1147 mtx_unlock(&ktrace_mtx);
1148 return;
1149 }
1150 VREF(vp);
1151 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL"));
1152 crhold(cred);
1153 mtx_unlock(&ktrace_mtx);
1154
1155 kth = &req->ktr_header;
1156 KASSERT(((u_short)kth->ktr_type & ~KTR_DROP) <
1157 sizeof(data_lengths) / sizeof(data_lengths[0]),
1158 ("data_lengths array overflow"));
1159 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP];
1160 buflen = kth->ktr_len;
1161 auio.uio_iov = &aiov[0];
1162 auio.uio_offset = 0;
1163 auio.uio_segflg = UIO_SYSSPACE;
1164 auio.uio_rw = UIO_WRITE;
1165 aiov[0].iov_base = (caddr_t)kth;
1166 aiov[0].iov_len = sizeof(struct ktr_header);
1167 auio.uio_resid = sizeof(struct ktr_header);
1168 auio.uio_iovcnt = 1;
1169 auio.uio_td = td;
1170 if (datalen != 0) {
1171 aiov[1].iov_base = (caddr_t)&req->ktr_data;
1172 aiov[1].iov_len = datalen;
1173 auio.uio_resid += datalen;
1174 auio.uio_iovcnt++;
1175 kth->ktr_len += datalen;
1176 }
1177 if (buflen != 0) {
1178 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write"));
1179 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer;
1180 aiov[auio.uio_iovcnt].iov_len = buflen;
1181 auio.uio_resid += buflen;
1182 auio.uio_iovcnt++;
1183 }
1184
1185 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1186 vn_start_write(vp, &mp, V_WAIT);
1187 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1188 #ifdef MAC
1189 error = mac_vnode_check_write(cred, NOCRED, vp);
1190 if (error == 0)
1191 #endif
1192 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred);
1193 VOP_UNLOCK(vp, 0);
1194 vn_finished_write(mp);
1195 crfree(cred);
1196 if (!error) {
1197 vrele(vp);
1198 VFS_UNLOCK_GIANT(vfslocked);
1199 return;
1200 }
1201 VFS_UNLOCK_GIANT(vfslocked);
1202
1203 /*
1204 * If error encountered, give up tracing on this vnode. We defer
1205 * all the vrele()'s on the vnode until after we are finished walking
1206 * the various lists to avoid needlessly holding locks.
1207 * NB: at this point we still hold the vnode reference that must
1208 * not go away as we need the valid vnode to compare with. Thus let
1209 * vrele_count start at 1 and the reference will be freed
1210 * by the loop at the end after our last use of vp.
1211 */
1212 log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n",
1213 error);
1214 vrele_count = 1;
1215 /*
1216 * First, clear this vnode from being used by any processes in the
1217 * system.
1218 * XXX - If one process gets an EPERM writing to the vnode, should
1219 * we really do this? Other processes might have suitable
1220 * credentials for the operation.
1221 */
1222 cred = NULL;
1223 sx_slock(&allproc_lock);
1224 FOREACH_PROC_IN_SYSTEM(p) {
1225 PROC_LOCK(p);
1226 if (p->p_tracevp == vp) {
1227 mtx_lock(&ktrace_mtx);
1228 ktr_freeproc(p, &cred, NULL);
1229 mtx_unlock(&ktrace_mtx);
1230 vrele_count++;
1231 }
1232 PROC_UNLOCK(p);
1233 if (cred != NULL) {
1234 crfree(cred);
1235 cred = NULL;
1236 }
1237 }
1238 sx_sunlock(&allproc_lock);
1239
1240 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1241 while (vrele_count-- > 0)
1242 vrele(vp);
1243 VFS_UNLOCK_GIANT(vfslocked);
1244 }
1245
1246 /*
1247 * Return true if caller has permission to set the ktracing state
1248 * of target. Essentially, the target can't possess any
1249 * more permissions than the caller. KTRFAC_ROOT signifies that
1250 * root previously set the tracing status on the target process, and
1251 * so, only root may further change it.
1252 */
1253 static int
1254 ktrcanset(td, targetp)
1255 struct thread *td;
1256 struct proc *targetp;
1257 {
1258
1259 PROC_LOCK_ASSERT(targetp, MA_OWNED);
1260 if (targetp->p_traceflag & KTRFAC_ROOT &&
1261 priv_check(td, PRIV_KTRACE))
1262 return (0);
1263
1264 if (p_candebug(td, targetp) != 0)
1265 return (0);
1266
1267 return (1);
1268 }
1269
1270 #endif /* KTRACE */
Cache object: 032abec00efca38fb5f5fbdb63a03672
|