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/capsicum.h>
41 #include <sys/systm.h>
42 #include <sys/fcntl.h>
43 #include <sys/kernel.h>
44 #include <sys/kthread.h>
45 #include <sys/lock.h>
46 #include <sys/mutex.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/namei.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/unistd.h>
53 #include <sys/vnode.h>
54 #include <sys/socket.h>
55 #include <sys/stat.h>
56 #include <sys/ktrace.h>
57 #include <sys/sx.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/syslog.h>
61 #include <sys/sysproto.h>
62
63 #include <security/mac/mac_framework.h>
64
65 /*
66 * The ktrace facility allows the tracing of certain key events in user space
67 * processes, such as system calls, signal delivery, context switches, and
68 * user generated events using utrace(2). It works by streaming event
69 * records and data to a vnode associated with the process using the
70 * ktrace(2) system call. In general, records can be written directly from
71 * the context that generates the event. One important exception to this is
72 * during a context switch, where sleeping is not permitted. To handle this
73 * case, trace events are generated using in-kernel ktr_request records, and
74 * then delivered to disk at a convenient moment -- either immediately, the
75 * next traceable event, at system call return, or at process exit.
76 *
77 * When dealing with multiple threads or processes writing to the same event
78 * log, ordering guarantees are weak: specifically, if an event has multiple
79 * records (i.e., system call enter and return), they may be interlaced with
80 * records from another event. Process and thread ID information is provided
81 * in the record, and user applications can de-interlace events if required.
82 */
83
84 static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE");
85
86 #ifdef KTRACE
87
88 FEATURE(ktrace, "Kernel support for system-call tracing");
89
90 #ifndef KTRACE_REQUEST_POOL
91 #define KTRACE_REQUEST_POOL 100
92 #endif
93
94 struct ktr_request {
95 struct ktr_header ktr_header;
96 void *ktr_buffer;
97 union {
98 struct ktr_proc_ctor ktr_proc_ctor;
99 struct ktr_cap_fail ktr_cap_fail;
100 struct ktr_syscall ktr_syscall;
101 struct ktr_sysret ktr_sysret;
102 struct ktr_genio ktr_genio;
103 struct ktr_psig ktr_psig;
104 struct ktr_csw ktr_csw;
105 struct ktr_fault ktr_fault;
106 struct ktr_faultend ktr_faultend;
107 } ktr_data;
108 STAILQ_ENTRY(ktr_request) ktr_list;
109 };
110
111 static int data_lengths[] = {
112 [KTR_SYSCALL] = offsetof(struct ktr_syscall, ktr_args),
113 [KTR_SYSRET] = sizeof(struct ktr_sysret),
114 [KTR_NAMEI] = 0,
115 [KTR_GENIO] = sizeof(struct ktr_genio),
116 [KTR_PSIG] = sizeof(struct ktr_psig),
117 [KTR_CSW] = sizeof(struct ktr_csw),
118 [KTR_USER] = 0,
119 [KTR_STRUCT] = 0,
120 [KTR_SYSCTL] = 0,
121 [KTR_PROCCTOR] = sizeof(struct ktr_proc_ctor),
122 [KTR_PROCDTOR] = 0,
123 [KTR_CAPFAIL] = sizeof(struct ktr_cap_fail),
124 [KTR_FAULT] = sizeof(struct ktr_fault),
125 [KTR_FAULTEND] = sizeof(struct ktr_faultend),
126 };
127
128 static STAILQ_HEAD(, ktr_request) ktr_free;
129
130 static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options");
131
132 static u_int ktr_requestpool = KTRACE_REQUEST_POOL;
133 TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool);
134
135 static u_int ktr_geniosize = PAGE_SIZE;
136 SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RWTUN, &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
ktrace_enter(struct thread * td)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
ktrace_exit(struct thread * td)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
ktrace_assert(struct thread * td)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
ktrace_init(void * dummy)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
sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)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
ktrace_resize_pool(u_int oldsize,u_int newsize)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 *
ktr_getrequest_entered(struct thread * td,int type)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 *
ktr_getrequest(int type)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
ktr_enqueuerequest(struct thread * td,struct ktr_request * req)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
ktr_drain(struct thread * td)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
ktr_submitrequest(struct thread * td,struct ktr_request * req)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
ktr_freerequest(struct ktr_request * req)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
ktr_freerequest_locked(struct ktr_request * req)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
ktr_freeproc(struct proc * p,struct ucred ** uc,struct vnode ** vp)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
ktrsyscall(code,narg,args)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
ktrsysret(code,error,retval)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
ktrprocexec(struct proc * p,struct ucred ** uc,struct vnode ** vp)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
ktrprocexit(struct thread * td)508 ktrprocexit(struct thread *td)
509 {
510 struct ktr_request *req;
511 struct proc *p;
512 struct ucred *cred;
513 struct vnode *vp;
514
515 p = td->td_proc;
516 if (p->p_traceflag == 0)
517 return;
518
519 ktrace_enter(td);
520 req = ktr_getrequest_entered(td, KTR_PROCDTOR);
521 if (req != NULL)
522 ktr_enqueuerequest(td, req);
523 sx_xlock(&ktrace_sx);
524 ktr_drain(td);
525 sx_xunlock(&ktrace_sx);
526 PROC_LOCK(p);
527 mtx_lock(&ktrace_mtx);
528 ktr_freeproc(p, &cred, &vp);
529 mtx_unlock(&ktrace_mtx);
530 PROC_UNLOCK(p);
531 if (vp != NULL)
532 vrele(vp);
533 if (cred != NULL)
534 crfree(cred);
535 ktrace_exit(td);
536 }
537
538 static void
ktrprocctor_entered(struct thread * td,struct proc * p)539 ktrprocctor_entered(struct thread *td, struct proc *p)
540 {
541 struct ktr_proc_ctor *ktp;
542 struct ktr_request *req;
543 struct thread *td2;
544
545 ktrace_assert(td);
546 td2 = FIRST_THREAD_IN_PROC(p);
547 req = ktr_getrequest_entered(td2, KTR_PROCCTOR);
548 if (req == NULL)
549 return;
550 ktp = &req->ktr_data.ktr_proc_ctor;
551 ktp->sv_flags = p->p_sysent->sv_flags;
552 ktr_enqueuerequest(td2, req);
553 }
554
555 void
ktrprocctor(struct proc * p)556 ktrprocctor(struct proc *p)
557 {
558 struct thread *td = curthread;
559
560 if ((p->p_traceflag & KTRFAC_MASK) == 0)
561 return;
562
563 ktrace_enter(td);
564 ktrprocctor_entered(td, p);
565 ktrace_exit(td);
566 }
567
568 /*
569 * When a process forks, enable tracing in the new process if needed.
570 */
571 void
ktrprocfork(struct proc * p1,struct proc * p2)572 ktrprocfork(struct proc *p1, struct proc *p2)
573 {
574
575 PROC_LOCK(p1);
576 mtx_lock(&ktrace_mtx);
577 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
578 if (p1->p_traceflag & KTRFAC_INHERIT) {
579 p2->p_traceflag = p1->p_traceflag;
580 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
581 VREF(p2->p_tracevp);
582 KASSERT(p1->p_tracecred != NULL,
583 ("ktrace vnode with no cred"));
584 p2->p_tracecred = crhold(p1->p_tracecred);
585 }
586 }
587 mtx_unlock(&ktrace_mtx);
588 PROC_UNLOCK(p1);
589
590 ktrprocctor(p2);
591 }
592
593 /*
594 * When a thread returns, drain any asynchronous records generated by the
595 * system call.
596 */
597 void
ktruserret(struct thread * td)598 ktruserret(struct thread *td)
599 {
600
601 ktrace_enter(td);
602 sx_xlock(&ktrace_sx);
603 ktr_drain(td);
604 sx_xunlock(&ktrace_sx);
605 ktrace_exit(td);
606 }
607
608 void
ktrnamei(path)609 ktrnamei(path)
610 char *path;
611 {
612 struct ktr_request *req;
613 int namelen;
614 char *buf = NULL;
615
616 namelen = strlen(path);
617 if (namelen > 0) {
618 buf = malloc(namelen, M_KTRACE, M_WAITOK);
619 bcopy(path, buf, namelen);
620 }
621 req = ktr_getrequest(KTR_NAMEI);
622 if (req == NULL) {
623 if (buf != NULL)
624 free(buf, M_KTRACE);
625 return;
626 }
627 if (namelen > 0) {
628 req->ktr_header.ktr_len = namelen;
629 req->ktr_buffer = buf;
630 }
631 ktr_submitrequest(curthread, req);
632 }
633
634 void
ktrsysctl(name,namelen)635 ktrsysctl(name, namelen)
636 int *name;
637 u_int namelen;
638 {
639 struct ktr_request *req;
640 u_int mib[CTL_MAXNAME + 2];
641 char *mibname;
642 size_t mibnamelen;
643 int error;
644
645 /* Lookup name of mib. */
646 KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long"));
647 mib[0] = 0;
648 mib[1] = 1;
649 bcopy(name, mib + 2, namelen * sizeof(*name));
650 mibnamelen = 128;
651 mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK);
652 error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen,
653 NULL, 0, &mibnamelen, 0);
654 if (error) {
655 free(mibname, M_KTRACE);
656 return;
657 }
658 req = ktr_getrequest(KTR_SYSCTL);
659 if (req == NULL) {
660 free(mibname, M_KTRACE);
661 return;
662 }
663 req->ktr_header.ktr_len = mibnamelen;
664 req->ktr_buffer = mibname;
665 ktr_submitrequest(curthread, req);
666 }
667
668 void
ktrgenio(fd,rw,uio,error)669 ktrgenio(fd, rw, uio, error)
670 int fd;
671 enum uio_rw rw;
672 struct uio *uio;
673 int error;
674 {
675 struct ktr_request *req;
676 struct ktr_genio *ktg;
677 int datalen;
678 char *buf;
679
680 if (error) {
681 free(uio, M_IOV);
682 return;
683 }
684 uio->uio_offset = 0;
685 uio->uio_rw = UIO_WRITE;
686 datalen = MIN(uio->uio_resid, ktr_geniosize);
687 buf = malloc(datalen, M_KTRACE, M_WAITOK);
688 error = uiomove(buf, datalen, uio);
689 free(uio, M_IOV);
690 if (error) {
691 free(buf, M_KTRACE);
692 return;
693 }
694 req = ktr_getrequest(KTR_GENIO);
695 if (req == NULL) {
696 free(buf, M_KTRACE);
697 return;
698 }
699 ktg = &req->ktr_data.ktr_genio;
700 ktg->ktr_fd = fd;
701 ktg->ktr_rw = rw;
702 req->ktr_header.ktr_len = datalen;
703 req->ktr_buffer = buf;
704 ktr_submitrequest(curthread, req);
705 }
706
707 void
ktrpsig(sig,action,mask,code)708 ktrpsig(sig, action, mask, code)
709 int sig;
710 sig_t action;
711 sigset_t *mask;
712 int code;
713 {
714 struct thread *td = curthread;
715 struct ktr_request *req;
716 struct ktr_psig *kp;
717
718 req = ktr_getrequest(KTR_PSIG);
719 if (req == NULL)
720 return;
721 kp = &req->ktr_data.ktr_psig;
722 kp->signo = (char)sig;
723 kp->action = action;
724 kp->mask = *mask;
725 kp->code = code;
726 ktr_enqueuerequest(td, req);
727 ktrace_exit(td);
728 }
729
730 void
ktrcsw(out,user,wmesg)731 ktrcsw(out, user, wmesg)
732 int out, user;
733 const char *wmesg;
734 {
735 struct thread *td = curthread;
736 struct ktr_request *req;
737 struct ktr_csw *kc;
738
739 req = ktr_getrequest(KTR_CSW);
740 if (req == NULL)
741 return;
742 kc = &req->ktr_data.ktr_csw;
743 kc->out = out;
744 kc->user = user;
745 if (wmesg != NULL)
746 strlcpy(kc->wmesg, wmesg, sizeof(kc->wmesg));
747 else
748 bzero(kc->wmesg, sizeof(kc->wmesg));
749 ktr_enqueuerequest(td, req);
750 ktrace_exit(td);
751 }
752
753 void
ktrstruct(name,data,datalen)754 ktrstruct(name, data, datalen)
755 const char *name;
756 void *data;
757 size_t datalen;
758 {
759 struct ktr_request *req;
760 char *buf = NULL;
761 size_t buflen;
762
763 if (!data)
764 datalen = 0;
765 buflen = strlen(name) + 1 + datalen;
766 buf = malloc(buflen, M_KTRACE, M_WAITOK);
767 strcpy(buf, name);
768 bcopy(data, buf + strlen(name) + 1, datalen);
769 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) {
770 free(buf, M_KTRACE);
771 return;
772 }
773 req->ktr_buffer = buf;
774 req->ktr_header.ktr_len = buflen;
775 ktr_submitrequest(curthread, req);
776 }
777
778 void
ktrcapfail(type,needed,held)779 ktrcapfail(type, needed, held)
780 enum ktr_cap_fail_type type;
781 const cap_rights_t *needed;
782 const cap_rights_t *held;
783 {
784 struct thread *td = curthread;
785 struct ktr_request *req;
786 struct ktr_cap_fail *kcf;
787
788 req = ktr_getrequest(KTR_CAPFAIL);
789 if (req == NULL)
790 return;
791 kcf = &req->ktr_data.ktr_cap_fail;
792 kcf->cap_type = type;
793 if (needed != NULL)
794 kcf->cap_needed = *needed;
795 else
796 cap_rights_init(&kcf->cap_needed);
797 if (held != NULL)
798 kcf->cap_held = *held;
799 else
800 cap_rights_init(&kcf->cap_held);
801 ktr_enqueuerequest(td, req);
802 ktrace_exit(td);
803 }
804
805 void
ktrfault(vaddr,type)806 ktrfault(vaddr, type)
807 vm_offset_t vaddr;
808 int type;
809 {
810 struct thread *td = curthread;
811 struct ktr_request *req;
812 struct ktr_fault *kf;
813
814 req = ktr_getrequest(KTR_FAULT);
815 if (req == NULL)
816 return;
817 kf = &req->ktr_data.ktr_fault;
818 kf->vaddr = vaddr;
819 kf->type = type;
820 ktr_enqueuerequest(td, req);
821 ktrace_exit(td);
822 }
823
824 void
ktrfaultend(result)825 ktrfaultend(result)
826 int result;
827 {
828 struct thread *td = curthread;
829 struct ktr_request *req;
830 struct ktr_faultend *kf;
831
832 req = ktr_getrequest(KTR_FAULTEND);
833 if (req == NULL)
834 return;
835 kf = &req->ktr_data.ktr_faultend;
836 kf->result = result;
837 ktr_enqueuerequest(td, req);
838 ktrace_exit(td);
839 }
840 #endif /* KTRACE */
841
842 /* Interface and common routines */
843
844 #ifndef _SYS_SYSPROTO_H_
845 struct ktrace_args {
846 char *fname;
847 int ops;
848 int facs;
849 int pid;
850 };
851 #endif
852 /* ARGSUSED */
853 int
sys_ktrace(td,uap)854 sys_ktrace(td, uap)
855 struct thread *td;
856 register struct ktrace_args *uap;
857 {
858 #ifdef KTRACE
859 register struct vnode *vp = NULL;
860 register struct proc *p;
861 struct pgrp *pg;
862 int facs = uap->facs & ~KTRFAC_ROOT;
863 int ops = KTROP(uap->ops);
864 int descend = uap->ops & KTRFLAG_DESCEND;
865 int nfound, ret = 0;
866 int flags, error = 0;
867 struct nameidata nd;
868 struct ucred *cred;
869
870 /*
871 * Need something to (un)trace.
872 */
873 if (ops != KTROP_CLEARFILE && facs == 0)
874 return (EINVAL);
875
876 ktrace_enter(td);
877 if (ops != KTROP_CLEAR) {
878 /*
879 * an operation which requires a file argument.
880 */
881 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname, td);
882 flags = FREAD | FWRITE | O_NOFOLLOW;
883 error = vn_open(&nd, &flags, 0, NULL);
884 if (error) {
885 ktrace_exit(td);
886 return (error);
887 }
888 NDFREE(&nd, NDF_ONLY_PNBUF);
889 vp = nd.ni_vp;
890 VOP_UNLOCK(vp, 0);
891 if (vp->v_type != VREG) {
892 (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
893 ktrace_exit(td);
894 return (EACCES);
895 }
896 }
897 /*
898 * Clear all uses of the tracefile.
899 */
900 if (ops == KTROP_CLEARFILE) {
901 int vrele_count;
902
903 vrele_count = 0;
904 sx_slock(&allproc_lock);
905 FOREACH_PROC_IN_SYSTEM(p) {
906 PROC_LOCK(p);
907 if (p->p_tracevp == vp) {
908 if (ktrcanset(td, p)) {
909 mtx_lock(&ktrace_mtx);
910 ktr_freeproc(p, &cred, NULL);
911 mtx_unlock(&ktrace_mtx);
912 vrele_count++;
913 crfree(cred);
914 } else
915 error = EPERM;
916 }
917 PROC_UNLOCK(p);
918 }
919 sx_sunlock(&allproc_lock);
920 if (vrele_count > 0) {
921 while (vrele_count-- > 0)
922 vrele(vp);
923 }
924 goto done;
925 }
926 /*
927 * do it
928 */
929 sx_slock(&proctree_lock);
930 if (uap->pid < 0) {
931 /*
932 * by process group
933 */
934 pg = pgfind(-uap->pid);
935 if (pg == NULL) {
936 sx_sunlock(&proctree_lock);
937 error = ESRCH;
938 goto done;
939 }
940 /*
941 * ktrops() may call vrele(). Lock pg_members
942 * by the proctree_lock rather than pg_mtx.
943 */
944 PGRP_UNLOCK(pg);
945 nfound = 0;
946 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
947 PROC_LOCK(p);
948 if (p->p_state == PRS_NEW ||
949 p_cansee(td, p) != 0) {
950 PROC_UNLOCK(p);
951 continue;
952 }
953 nfound++;
954 if (descend)
955 ret |= ktrsetchildren(td, p, ops, facs, vp);
956 else
957 ret |= ktrops(td, p, ops, facs, vp);
958 }
959 if (nfound == 0) {
960 sx_sunlock(&proctree_lock);
961 error = ESRCH;
962 goto done;
963 }
964 } else {
965 /*
966 * by pid
967 */
968 p = pfind(uap->pid);
969 if (p == NULL)
970 error = ESRCH;
971 else
972 error = p_cansee(td, p);
973 if (error) {
974 if (p != NULL)
975 PROC_UNLOCK(p);
976 sx_sunlock(&proctree_lock);
977 goto done;
978 }
979 if (descend)
980 ret |= ktrsetchildren(td, p, ops, facs, vp);
981 else
982 ret |= ktrops(td, p, ops, facs, vp);
983 }
984 sx_sunlock(&proctree_lock);
985 if (!ret)
986 error = EPERM;
987 done:
988 if (vp != NULL)
989 (void) vn_close(vp, FWRITE, td->td_ucred, td);
990 ktrace_exit(td);
991 return (error);
992 #else /* !KTRACE */
993 return (ENOSYS);
994 #endif /* KTRACE */
995 }
996
997 /* ARGSUSED */
998 int
sys_utrace(td,uap)999 sys_utrace(td, uap)
1000 struct thread *td;
1001 register struct utrace_args *uap;
1002 {
1003
1004 #ifdef KTRACE
1005 struct ktr_request *req;
1006 void *cp;
1007 int error;
1008
1009 if (!KTRPOINT(td, KTR_USER))
1010 return (0);
1011 if (uap->len > KTR_USER_MAXLEN)
1012 return (EINVAL);
1013 cp = malloc(uap->len, M_KTRACE, M_WAITOK);
1014 error = copyin(uap->addr, cp, uap->len);
1015 if (error) {
1016 free(cp, M_KTRACE);
1017 return (error);
1018 }
1019 req = ktr_getrequest(KTR_USER);
1020 if (req == NULL) {
1021 free(cp, M_KTRACE);
1022 return (ENOMEM);
1023 }
1024 req->ktr_buffer = cp;
1025 req->ktr_header.ktr_len = uap->len;
1026 ktr_submitrequest(td, req);
1027 return (0);
1028 #else /* !KTRACE */
1029 return (ENOSYS);
1030 #endif /* KTRACE */
1031 }
1032
1033 #ifdef KTRACE
1034 static int
ktrops(td,p,ops,facs,vp)1035 ktrops(td, p, ops, facs, vp)
1036 struct thread *td;
1037 struct proc *p;
1038 int ops, facs;
1039 struct vnode *vp;
1040 {
1041 struct vnode *tracevp = NULL;
1042 struct ucred *tracecred = NULL;
1043
1044 PROC_LOCK_ASSERT(p, MA_OWNED);
1045 if (!ktrcanset(td, p)) {
1046 PROC_UNLOCK(p);
1047 return (0);
1048 }
1049 if (p->p_flag & P_WEXIT) {
1050 /* If the process is exiting, just ignore it. */
1051 PROC_UNLOCK(p);
1052 return (1);
1053 }
1054 mtx_lock(&ktrace_mtx);
1055 if (ops == KTROP_SET) {
1056 if (p->p_tracevp != vp) {
1057 /*
1058 * if trace file already in use, relinquish below
1059 */
1060 tracevp = p->p_tracevp;
1061 VREF(vp);
1062 p->p_tracevp = vp;
1063 }
1064 if (p->p_tracecred != td->td_ucred) {
1065 tracecred = p->p_tracecred;
1066 p->p_tracecred = crhold(td->td_ucred);
1067 }
1068 p->p_traceflag |= facs;
1069 if (priv_check(td, PRIV_KTRACE) == 0)
1070 p->p_traceflag |= KTRFAC_ROOT;
1071 } else {
1072 /* KTROP_CLEAR */
1073 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0)
1074 /* no more tracing */
1075 ktr_freeproc(p, &tracecred, &tracevp);
1076 }
1077 mtx_unlock(&ktrace_mtx);
1078 if ((p->p_traceflag & KTRFAC_MASK) != 0)
1079 ktrprocctor_entered(td, p);
1080 PROC_UNLOCK(p);
1081 if (tracevp != NULL)
1082 vrele(tracevp);
1083 if (tracecred != NULL)
1084 crfree(tracecred);
1085
1086 return (1);
1087 }
1088
1089 static int
ktrsetchildren(td,top,ops,facs,vp)1090 ktrsetchildren(td, top, ops, facs, vp)
1091 struct thread *td;
1092 struct proc *top;
1093 int ops, facs;
1094 struct vnode *vp;
1095 {
1096 register struct proc *p;
1097 register int ret = 0;
1098
1099 p = top;
1100 PROC_LOCK_ASSERT(p, MA_OWNED);
1101 sx_assert(&proctree_lock, SX_LOCKED);
1102 for (;;) {
1103 ret |= ktrops(td, p, ops, facs, vp);
1104 /*
1105 * If this process has children, descend to them next,
1106 * otherwise do any siblings, and if done with this level,
1107 * follow back up the tree (but not past top).
1108 */
1109 if (!LIST_EMPTY(&p->p_children))
1110 p = LIST_FIRST(&p->p_children);
1111 else for (;;) {
1112 if (p == top)
1113 return (ret);
1114 if (LIST_NEXT(p, p_sibling)) {
1115 p = LIST_NEXT(p, p_sibling);
1116 break;
1117 }
1118 p = p->p_pptr;
1119 }
1120 PROC_LOCK(p);
1121 }
1122 /*NOTREACHED*/
1123 }
1124
1125 static void
ktr_writerequest(struct thread * td,struct ktr_request * req)1126 ktr_writerequest(struct thread *td, struct ktr_request *req)
1127 {
1128 struct ktr_header *kth;
1129 struct vnode *vp;
1130 struct proc *p;
1131 struct ucred *cred;
1132 struct uio auio;
1133 struct iovec aiov[3];
1134 struct mount *mp;
1135 int datalen, buflen, vrele_count;
1136 int error;
1137
1138 /*
1139 * We hold the vnode and credential for use in I/O in case ktrace is
1140 * disabled on the process as we write out the request.
1141 *
1142 * XXXRW: This is not ideal: we could end up performing a write after
1143 * the vnode has been closed.
1144 */
1145 mtx_lock(&ktrace_mtx);
1146 vp = td->td_proc->p_tracevp;
1147 cred = td->td_proc->p_tracecred;
1148
1149 /*
1150 * If vp is NULL, the vp has been cleared out from under this
1151 * request, so just drop it. Make sure the credential and vnode are
1152 * in sync: we should have both or neither.
1153 */
1154 if (vp == NULL) {
1155 KASSERT(cred == NULL, ("ktr_writerequest: cred != NULL"));
1156 mtx_unlock(&ktrace_mtx);
1157 return;
1158 }
1159 VREF(vp);
1160 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL"));
1161 crhold(cred);
1162 mtx_unlock(&ktrace_mtx);
1163
1164 kth = &req->ktr_header;
1165 KASSERT(((u_short)kth->ktr_type & ~KTR_DROP) <
1166 sizeof(data_lengths) / sizeof(data_lengths[0]),
1167 ("data_lengths array overflow"));
1168 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP];
1169 buflen = kth->ktr_len;
1170 auio.uio_iov = &aiov[0];
1171 auio.uio_offset = 0;
1172 auio.uio_segflg = UIO_SYSSPACE;
1173 auio.uio_rw = UIO_WRITE;
1174 aiov[0].iov_base = (caddr_t)kth;
1175 aiov[0].iov_len = sizeof(struct ktr_header);
1176 auio.uio_resid = sizeof(struct ktr_header);
1177 auio.uio_iovcnt = 1;
1178 auio.uio_td = td;
1179 if (datalen != 0) {
1180 aiov[1].iov_base = (caddr_t)&req->ktr_data;
1181 aiov[1].iov_len = datalen;
1182 auio.uio_resid += datalen;
1183 auio.uio_iovcnt++;
1184 kth->ktr_len += datalen;
1185 }
1186 if (buflen != 0) {
1187 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write"));
1188 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer;
1189 aiov[auio.uio_iovcnt].iov_len = buflen;
1190 auio.uio_resid += buflen;
1191 auio.uio_iovcnt++;
1192 }
1193
1194 vn_start_write(vp, &mp, V_WAIT);
1195 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1196 #ifdef MAC
1197 error = mac_vnode_check_write(cred, NOCRED, vp);
1198 if (error == 0)
1199 #endif
1200 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred);
1201 VOP_UNLOCK(vp, 0);
1202 vn_finished_write(mp);
1203 crfree(cred);
1204 if (!error) {
1205 vrele(vp);
1206 return;
1207 }
1208
1209 /*
1210 * If error encountered, give up tracing on this vnode. We defer
1211 * all the vrele()'s on the vnode until after we are finished walking
1212 * the various lists to avoid needlessly holding locks.
1213 * NB: at this point we still hold the vnode reference that must
1214 * not go away as we need the valid vnode to compare with. Thus let
1215 * vrele_count start at 1 and the reference will be freed
1216 * by the loop at the end after our last use of vp.
1217 */
1218 log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n",
1219 error);
1220 vrele_count = 1;
1221 /*
1222 * First, clear this vnode from being used by any processes in the
1223 * system.
1224 * XXX - If one process gets an EPERM writing to the vnode, should
1225 * we really do this? Other processes might have suitable
1226 * credentials for the operation.
1227 */
1228 cred = NULL;
1229 sx_slock(&allproc_lock);
1230 FOREACH_PROC_IN_SYSTEM(p) {
1231 PROC_LOCK(p);
1232 if (p->p_tracevp == vp) {
1233 mtx_lock(&ktrace_mtx);
1234 ktr_freeproc(p, &cred, NULL);
1235 mtx_unlock(&ktrace_mtx);
1236 vrele_count++;
1237 }
1238 PROC_UNLOCK(p);
1239 if (cred != NULL) {
1240 crfree(cred);
1241 cred = NULL;
1242 }
1243 }
1244 sx_sunlock(&allproc_lock);
1245
1246 while (vrele_count-- > 0)
1247 vrele(vp);
1248 }
1249
1250 /*
1251 * Return true if caller has permission to set the ktracing state
1252 * of target. Essentially, the target can't possess any
1253 * more permissions than the caller. KTRFAC_ROOT signifies that
1254 * root previously set the tracing status on the target process, and
1255 * so, only root may further change it.
1256 */
1257 static int
ktrcanset(td,targetp)1258 ktrcanset(td, targetp)
1259 struct thread *td;
1260 struct proc *targetp;
1261 {
1262
1263 PROC_LOCK_ASSERT(targetp, MA_OWNED);
1264 if (targetp->p_traceflag & KTRFAC_ROOT &&
1265 priv_check(td, PRIV_KTRACE))
1266 return (0);
1267
1268 if (p_candebug(td, targetp) != 0)
1269 return (0);
1270
1271 return (1);
1272 }
1273
1274 #endif /* KTRACE */
1275