1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2005 Robert N. M. Watson
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34 #include <sys/cdefs.h>
35 #include "opt_ktrace.h"
36
37 #include <sys/param.h>
38 #include <sys/capsicum.h>
39 #include <sys/systm.h>
40 #include <sys/fcntl.h>
41 #include <sys/kernel.h>
42 #include <sys/kthread.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/malloc.h>
46 #include <sys/mount.h>
47 #include <sys/namei.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/resourcevar.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_cap_fail ktr_cap_fail;
99 struct ktr_syscall ktr_syscall;
100 struct ktr_sysret ktr_sysret;
101 struct ktr_genio ktr_genio;
102 struct ktr_psig ktr_psig;
103 struct ktr_csw ktr_csw;
104 struct ktr_fault ktr_fault;
105 struct ktr_faultend ktr_faultend;
106 struct ktr_struct_array ktr_struct_array;
107 } ktr_data;
108 STAILQ_ENTRY(ktr_request) ktr_list;
109 };
110
111 static const 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 [KTR_STRUCT_ARRAY] = sizeof(struct ktr_struct_array),
127 };
128
129 static STAILQ_HEAD(, ktr_request) ktr_free;
130
131 static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
132 "KTRACE options");
133
134 static u_int ktr_requestpool = KTRACE_REQUEST_POOL;
135 TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool);
136
137 u_int ktr_geniosize = PAGE_SIZE;
138 SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RWTUN, &ktr_geniosize,
139 0, "Maximum size of genio event payload");
140
141 /*
142 * Allow to not to send signal to traced process, in which context the
143 * ktr record is written. The limit is applied from the process that
144 * set up ktrace, so killing the traced process is not completely fair.
145 */
146 int ktr_filesize_limit_signal = 0;
147 SYSCTL_INT(_kern_ktrace, OID_AUTO, filesize_limit_signal, CTLFLAG_RWTUN,
148 &ktr_filesize_limit_signal, 0,
149 "Send SIGXFSZ to the traced process when the log size limit is exceeded");
150
151 static int print_message = 1;
152 static struct mtx ktrace_mtx;
153 static struct sx ktrace_sx;
154
155 struct ktr_io_params {
156 struct vnode *vp;
157 struct ucred *cr;
158 off_t lim;
159 u_int refs;
160 };
161
162 static void ktrace_init(void *dummy);
163 static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS);
164 static u_int ktrace_resize_pool(u_int oldsize, u_int newsize);
165 static struct ktr_request *ktr_getrequest_entered(struct thread *td, int type);
166 static struct ktr_request *ktr_getrequest(int type);
167 static void ktr_submitrequest(struct thread *td, struct ktr_request *req);
168 static struct ktr_io_params *ktr_freeproc(struct proc *p);
169 static void ktr_freerequest(struct ktr_request *req);
170 static void ktr_freerequest_locked(struct ktr_request *req);
171 static void ktr_writerequest(struct thread *td, struct ktr_request *req);
172 static int ktrcanset(struct thread *,struct proc *);
173 static int ktrsetchildren(struct thread *, struct proc *, int, int,
174 struct ktr_io_params *);
175 static int ktrops(struct thread *, struct proc *, int, int,
176 struct ktr_io_params *);
177 static void ktrprocctor_entered(struct thread *, struct proc *);
178
179 /*
180 * ktrace itself generates events, such as context switches, which we do not
181 * wish to trace. Maintain a flag, TDP_INKTRACE, on each thread to determine
182 * whether or not it is in a region where tracing of events should be
183 * suppressed.
184 */
185 static void
ktrace_enter(struct thread * td)186 ktrace_enter(struct thread *td)
187 {
188
189 KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set"));
190 td->td_pflags |= TDP_INKTRACE;
191 }
192
193 static void
ktrace_exit(struct thread * td)194 ktrace_exit(struct thread *td)
195 {
196
197 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set"));
198 td->td_pflags &= ~TDP_INKTRACE;
199 }
200
201 static void
ktrace_assert(struct thread * td)202 ktrace_assert(struct thread *td)
203 {
204
205 KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set"));
206 }
207
208 static void
ast_ktrace(struct thread * td,int tda __unused)209 ast_ktrace(struct thread *td, int tda __unused)
210 {
211 KTRUSERRET(td);
212 }
213
214 static void
ktrace_init(void * dummy)215 ktrace_init(void *dummy)
216 {
217 struct ktr_request *req;
218 int i;
219
220 mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET);
221 sx_init(&ktrace_sx, "ktrace_sx");
222 STAILQ_INIT(&ktr_free);
223 for (i = 0; i < ktr_requestpool; i++) {
224 req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK |
225 M_ZERO);
226 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
227 }
228 ast_register(TDA_KTRACE, ASTR_ASTF_REQUIRED, 0, ast_ktrace);
229 }
230 SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL);
231
232 static int
sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)233 sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)
234 {
235 struct thread *td;
236 u_int newsize, oldsize, wantsize;
237 int error;
238
239 /* Handle easy read-only case first to avoid warnings from GCC. */
240 if (!req->newptr) {
241 oldsize = ktr_requestpool;
242 return (SYSCTL_OUT(req, &oldsize, sizeof(u_int)));
243 }
244
245 error = SYSCTL_IN(req, &wantsize, sizeof(u_int));
246 if (error)
247 return (error);
248 td = curthread;
249 ktrace_enter(td);
250 oldsize = ktr_requestpool;
251 newsize = ktrace_resize_pool(oldsize, wantsize);
252 ktrace_exit(td);
253 error = SYSCTL_OUT(req, &oldsize, sizeof(u_int));
254 if (error)
255 return (error);
256 if (wantsize > oldsize && newsize < wantsize)
257 return (ENOSPC);
258 return (0);
259 }
260 SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool,
261 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &ktr_requestpool, 0,
262 sysctl_kern_ktrace_request_pool, "IU",
263 "Pool buffer size for ktrace(1)");
264
265 static u_int
ktrace_resize_pool(u_int oldsize,u_int newsize)266 ktrace_resize_pool(u_int oldsize, u_int newsize)
267 {
268 STAILQ_HEAD(, ktr_request) ktr_new;
269 struct ktr_request *req;
270 int bound;
271
272 print_message = 1;
273 bound = newsize - oldsize;
274 if (bound == 0)
275 return (ktr_requestpool);
276 if (bound < 0) {
277 mtx_lock(&ktrace_mtx);
278 /* Shrink pool down to newsize if possible. */
279 while (bound++ < 0) {
280 req = STAILQ_FIRST(&ktr_free);
281 if (req == NULL)
282 break;
283 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
284 ktr_requestpool--;
285 free(req, M_KTRACE);
286 }
287 } else {
288 /* Grow pool up to newsize. */
289 STAILQ_INIT(&ktr_new);
290 while (bound-- > 0) {
291 req = malloc(sizeof(struct ktr_request), M_KTRACE,
292 M_WAITOK | M_ZERO);
293 STAILQ_INSERT_HEAD(&ktr_new, req, ktr_list);
294 }
295 mtx_lock(&ktrace_mtx);
296 STAILQ_CONCAT(&ktr_free, &ktr_new);
297 ktr_requestpool += (newsize - oldsize);
298 }
299 mtx_unlock(&ktrace_mtx);
300 return (ktr_requestpool);
301 }
302
303 /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */
304 CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) ==
305 (sizeof((struct thread *)NULL)->td_name));
306
307 static struct ktr_request *
ktr_getrequest_entered(struct thread * td,int type)308 ktr_getrequest_entered(struct thread *td, int type)
309 {
310 struct ktr_request *req;
311 struct proc *p = td->td_proc;
312 int pm;
313
314 mtx_lock(&ktrace_mtx);
315 if (!KTRCHECK(td, type)) {
316 mtx_unlock(&ktrace_mtx);
317 return (NULL);
318 }
319 req = STAILQ_FIRST(&ktr_free);
320 if (req != NULL) {
321 STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
322 req->ktr_header.ktr_type = type;
323 if (p->p_traceflag & KTRFAC_DROP) {
324 req->ktr_header.ktr_type |= KTR_DROP;
325 p->p_traceflag &= ~KTRFAC_DROP;
326 }
327 mtx_unlock(&ktrace_mtx);
328 nanotime(&req->ktr_header.ktr_time);
329 req->ktr_header.ktr_type |= KTR_VERSIONED;
330 req->ktr_header.ktr_pid = p->p_pid;
331 req->ktr_header.ktr_tid = td->td_tid;
332 req->ktr_header.ktr_cpu = PCPU_GET(cpuid);
333 req->ktr_header.ktr_version = KTR_VERSION1;
334 bcopy(td->td_name, req->ktr_header.ktr_comm,
335 sizeof(req->ktr_header.ktr_comm));
336 req->ktr_buffer = NULL;
337 req->ktr_header.ktr_len = 0;
338 } else {
339 p->p_traceflag |= KTRFAC_DROP;
340 pm = print_message;
341 print_message = 0;
342 mtx_unlock(&ktrace_mtx);
343 if (pm)
344 printf("Out of ktrace request objects.\n");
345 }
346 return (req);
347 }
348
349 static struct ktr_request *
ktr_getrequest(int type)350 ktr_getrequest(int type)
351 {
352 struct thread *td = curthread;
353 struct ktr_request *req;
354
355 ktrace_enter(td);
356 req = ktr_getrequest_entered(td, type);
357 if (req == NULL)
358 ktrace_exit(td);
359
360 return (req);
361 }
362
363 /*
364 * Some trace generation environments don't permit direct access to VFS,
365 * such as during a context switch where sleeping is not allowed. Under these
366 * circumstances, queue a request to the thread to be written asynchronously
367 * later.
368 */
369 static void
ktr_enqueuerequest(struct thread * td,struct ktr_request * req)370 ktr_enqueuerequest(struct thread *td, struct ktr_request *req)
371 {
372
373 mtx_lock(&ktrace_mtx);
374 STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list);
375 mtx_unlock(&ktrace_mtx);
376 ast_sched(td, TDA_KTRACE);
377 }
378
379 /*
380 * Drain any pending ktrace records from the per-thread queue to disk. This
381 * is used both internally before committing other records, and also on
382 * system call return. We drain all the ones we can find at the time when
383 * drain is requested, but don't keep draining after that as those events
384 * may be approximately "after" the current event.
385 */
386 static void
ktr_drain(struct thread * td)387 ktr_drain(struct thread *td)
388 {
389 struct ktr_request *queued_req;
390 STAILQ_HEAD(, ktr_request) local_queue;
391
392 ktrace_assert(td);
393 sx_assert(&ktrace_sx, SX_XLOCKED);
394
395 STAILQ_INIT(&local_queue);
396
397 if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) {
398 mtx_lock(&ktrace_mtx);
399 STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr);
400 mtx_unlock(&ktrace_mtx);
401
402 while ((queued_req = STAILQ_FIRST(&local_queue))) {
403 STAILQ_REMOVE_HEAD(&local_queue, ktr_list);
404 ktr_writerequest(td, queued_req);
405 ktr_freerequest(queued_req);
406 }
407 }
408 }
409
410 /*
411 * Submit a trace record for immediate commit to disk -- to be used only
412 * where entering VFS is OK. First drain any pending records that may have
413 * been cached in the thread.
414 */
415 static void
ktr_submitrequest(struct thread * td,struct ktr_request * req)416 ktr_submitrequest(struct thread *td, struct ktr_request *req)
417 {
418
419 ktrace_assert(td);
420
421 sx_xlock(&ktrace_sx);
422 ktr_drain(td);
423 ktr_writerequest(td, req);
424 ktr_freerequest(req);
425 sx_xunlock(&ktrace_sx);
426 ktrace_exit(td);
427 }
428
429 static void
ktr_freerequest(struct ktr_request * req)430 ktr_freerequest(struct ktr_request *req)
431 {
432
433 mtx_lock(&ktrace_mtx);
434 ktr_freerequest_locked(req);
435 mtx_unlock(&ktrace_mtx);
436 }
437
438 static void
ktr_freerequest_locked(struct ktr_request * req)439 ktr_freerequest_locked(struct ktr_request *req)
440 {
441
442 mtx_assert(&ktrace_mtx, MA_OWNED);
443 if (req->ktr_buffer != NULL)
444 free(req->ktr_buffer, M_KTRACE);
445 STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
446 }
447
448 static void
ktr_io_params_ref(struct ktr_io_params * kiop)449 ktr_io_params_ref(struct ktr_io_params *kiop)
450 {
451 mtx_assert(&ktrace_mtx, MA_OWNED);
452 kiop->refs++;
453 }
454
455 static struct ktr_io_params *
ktr_io_params_rele(struct ktr_io_params * kiop)456 ktr_io_params_rele(struct ktr_io_params *kiop)
457 {
458 mtx_assert(&ktrace_mtx, MA_OWNED);
459 if (kiop == NULL)
460 return (NULL);
461 KASSERT(kiop->refs > 0, ("kiop ref == 0 %p", kiop));
462 return (--(kiop->refs) == 0 ? kiop : NULL);
463 }
464
465 void
ktr_io_params_free(struct ktr_io_params * kiop)466 ktr_io_params_free(struct ktr_io_params *kiop)
467 {
468 if (kiop == NULL)
469 return;
470
471 MPASS(kiop->refs == 0);
472 vn_close(kiop->vp, FWRITE, kiop->cr, curthread);
473 crfree(kiop->cr);
474 free(kiop, M_KTRACE);
475 }
476
477 static struct ktr_io_params *
ktr_io_params_alloc(struct thread * td,struct vnode * vp)478 ktr_io_params_alloc(struct thread *td, struct vnode *vp)
479 {
480 struct ktr_io_params *res;
481
482 res = malloc(sizeof(struct ktr_io_params), M_KTRACE, M_WAITOK);
483 res->vp = vp;
484 res->cr = crhold(td->td_ucred);
485 res->lim = lim_cur(td, RLIMIT_FSIZE);
486 res->refs = 1;
487 return (res);
488 }
489
490 /*
491 * Disable tracing for a process and release all associated resources.
492 * The caller is responsible for releasing a reference on the returned
493 * vnode and credentials.
494 */
495 static struct ktr_io_params *
ktr_freeproc(struct proc * p)496 ktr_freeproc(struct proc *p)
497 {
498 struct ktr_io_params *kiop;
499 struct ktr_request *req;
500
501 PROC_LOCK_ASSERT(p, MA_OWNED);
502 mtx_assert(&ktrace_mtx, MA_OWNED);
503 kiop = ktr_io_params_rele(p->p_ktrioparms);
504 p->p_ktrioparms = NULL;
505 p->p_traceflag = 0;
506 while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) {
507 STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list);
508 ktr_freerequest_locked(req);
509 }
510 return (kiop);
511 }
512
513 struct vnode *
ktr_get_tracevp(struct proc * p,bool ref)514 ktr_get_tracevp(struct proc *p, bool ref)
515 {
516 struct vnode *vp;
517
518 PROC_LOCK_ASSERT(p, MA_OWNED);
519
520 if (p->p_ktrioparms != NULL) {
521 vp = p->p_ktrioparms->vp;
522 if (ref)
523 vrefact(vp);
524 } else {
525 vp = NULL;
526 }
527 return (vp);
528 }
529
530 void
ktrsyscall(int code,int narg,syscallarg_t args[])531 ktrsyscall(int code, int narg, syscallarg_t args[])
532 {
533 struct ktr_request *req;
534 struct ktr_syscall *ktp;
535 size_t buflen;
536 char *buf = NULL;
537
538 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
539 return;
540
541 buflen = sizeof(register_t) * narg;
542 if (buflen > 0) {
543 buf = malloc(buflen, M_KTRACE, M_WAITOK);
544 bcopy(args, buf, buflen);
545 }
546 req = ktr_getrequest(KTR_SYSCALL);
547 if (req == NULL) {
548 if (buf != NULL)
549 free(buf, M_KTRACE);
550 return;
551 }
552 ktp = &req->ktr_data.ktr_syscall;
553 ktp->ktr_code = code;
554 ktp->ktr_narg = narg;
555 if (buflen > 0) {
556 req->ktr_header.ktr_len = buflen;
557 req->ktr_buffer = buf;
558 }
559 ktr_submitrequest(curthread, req);
560 }
561
562 void
ktrsysret(int code,int error,register_t retval)563 ktrsysret(int code, int error, register_t retval)
564 {
565 struct ktr_request *req;
566 struct ktr_sysret *ktp;
567
568 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
569 return;
570
571 req = ktr_getrequest(KTR_SYSRET);
572 if (req == NULL)
573 return;
574 ktp = &req->ktr_data.ktr_sysret;
575 ktp->ktr_code = code;
576 ktp->ktr_error = error;
577 ktp->ktr_retval = ((error == 0) ? retval: 0); /* what about val2 ? */
578 ktr_submitrequest(curthread, req);
579 }
580
581 /*
582 * When a setuid process execs, disable tracing.
583 *
584 * XXX: We toss any pending asynchronous records.
585 */
586 struct ktr_io_params *
ktrprocexec(struct proc * p)587 ktrprocexec(struct proc *p)
588 {
589 struct ktr_io_params *kiop;
590
591 PROC_LOCK_ASSERT(p, MA_OWNED);
592
593 kiop = p->p_ktrioparms;
594 if (kiop == NULL || priv_check_cred(kiop->cr, PRIV_DEBUG_DIFFCRED))
595 return (NULL);
596
597 mtx_lock(&ktrace_mtx);
598 kiop = ktr_freeproc(p);
599 mtx_unlock(&ktrace_mtx);
600 return (kiop);
601 }
602
603 /*
604 * When a process exits, drain per-process asynchronous trace records
605 * and disable tracing.
606 */
607 void
ktrprocexit(struct thread * td)608 ktrprocexit(struct thread *td)
609 {
610 struct ktr_request *req;
611 struct proc *p;
612 struct ktr_io_params *kiop;
613
614 p = td->td_proc;
615 if (p->p_traceflag == 0)
616 return;
617
618 ktrace_enter(td);
619 req = ktr_getrequest_entered(td, KTR_PROCDTOR);
620 if (req != NULL)
621 ktr_enqueuerequest(td, req);
622 sx_xlock(&ktrace_sx);
623 ktr_drain(td);
624 sx_xunlock(&ktrace_sx);
625 PROC_LOCK(p);
626 mtx_lock(&ktrace_mtx);
627 kiop = ktr_freeproc(p);
628 mtx_unlock(&ktrace_mtx);
629 PROC_UNLOCK(p);
630 ktr_io_params_free(kiop);
631 ktrace_exit(td);
632 }
633
634 static void
ktrprocctor_entered(struct thread * td,struct proc * p)635 ktrprocctor_entered(struct thread *td, struct proc *p)
636 {
637 struct ktr_proc_ctor *ktp;
638 struct ktr_request *req;
639 struct thread *td2;
640
641 ktrace_assert(td);
642 td2 = FIRST_THREAD_IN_PROC(p);
643 req = ktr_getrequest_entered(td2, KTR_PROCCTOR);
644 if (req == NULL)
645 return;
646 ktp = &req->ktr_data.ktr_proc_ctor;
647 ktp->sv_flags = p->p_sysent->sv_flags;
648 ktr_enqueuerequest(td2, req);
649 }
650
651 void
ktrprocctor(struct proc * p)652 ktrprocctor(struct proc *p)
653 {
654 struct thread *td = curthread;
655
656 if ((p->p_traceflag & KTRFAC_MASK) == 0)
657 return;
658
659 ktrace_enter(td);
660 ktrprocctor_entered(td, p);
661 ktrace_exit(td);
662 }
663
664 /*
665 * When a process forks, enable tracing in the new process if needed.
666 */
667 void
ktrprocfork(struct proc * p1,struct proc * p2)668 ktrprocfork(struct proc *p1, struct proc *p2)
669 {
670
671 MPASS(p2->p_ktrioparms == NULL);
672 MPASS(p2->p_traceflag == 0);
673
674 if (p1->p_traceflag == 0)
675 return;
676
677 PROC_LOCK(p1);
678 mtx_lock(&ktrace_mtx);
679 if (p1->p_traceflag & KTRFAC_INHERIT) {
680 p2->p_traceflag = p1->p_traceflag;
681 if ((p2->p_ktrioparms = p1->p_ktrioparms) != NULL)
682 p1->p_ktrioparms->refs++;
683 }
684 mtx_unlock(&ktrace_mtx);
685 PROC_UNLOCK(p1);
686
687 ktrprocctor(p2);
688 }
689
690 /*
691 * When a thread returns, drain any asynchronous records generated by the
692 * system call.
693 */
694 void
ktruserret(struct thread * td)695 ktruserret(struct thread *td)
696 {
697
698 ktrace_enter(td);
699 sx_xlock(&ktrace_sx);
700 ktr_drain(td);
701 sx_xunlock(&ktrace_sx);
702 ktrace_exit(td);
703 }
704
705 void
ktrnamei(const char * path)706 ktrnamei(const char *path)
707 {
708 struct ktr_request *req;
709 int namelen;
710 char *buf = NULL;
711
712 namelen = strlen(path);
713 if (namelen > 0) {
714 buf = malloc(namelen, M_KTRACE, M_WAITOK);
715 bcopy(path, buf, namelen);
716 }
717 req = ktr_getrequest(KTR_NAMEI);
718 if (req == NULL) {
719 if (buf != NULL)
720 free(buf, M_KTRACE);
721 return;
722 }
723 if (namelen > 0) {
724 req->ktr_header.ktr_len = namelen;
725 req->ktr_buffer = buf;
726 }
727 ktr_submitrequest(curthread, req);
728 }
729
730 void
ktrsysctl(int * name,u_int namelen)731 ktrsysctl(int *name, u_int namelen)
732 {
733 struct ktr_request *req;
734 u_int mib[CTL_MAXNAME + 2];
735 char *mibname;
736 size_t mibnamelen;
737 int error;
738
739 /* Lookup name of mib. */
740 KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long"));
741 mib[0] = 0;
742 mib[1] = 1;
743 bcopy(name, mib + 2, namelen * sizeof(*name));
744 mibnamelen = 128;
745 mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK);
746 error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen,
747 NULL, 0, &mibnamelen, 0);
748 if (error) {
749 free(mibname, M_KTRACE);
750 return;
751 }
752 req = ktr_getrequest(KTR_SYSCTL);
753 if (req == NULL) {
754 free(mibname, M_KTRACE);
755 return;
756 }
757 req->ktr_header.ktr_len = mibnamelen;
758 req->ktr_buffer = mibname;
759 ktr_submitrequest(curthread, req);
760 }
761
762 void
ktrgenio(int fd,enum uio_rw rw,struct uio * uio,int error)763 ktrgenio(int fd, enum uio_rw rw, struct uio *uio, int error)
764 {
765 struct ktr_request *req;
766 struct ktr_genio *ktg;
767 int datalen;
768 char *buf;
769
770 if (error != 0 && (rw == UIO_READ || error == EFAULT)) {
771 freeuio(uio);
772 return;
773 }
774 uio->uio_offset = 0;
775 uio->uio_rw = UIO_WRITE;
776 datalen = MIN(uio->uio_resid, ktr_geniosize);
777 buf = malloc(datalen, M_KTRACE, M_WAITOK);
778 error = uiomove(buf, datalen, uio);
779 freeuio(uio);
780 if (error) {
781 free(buf, M_KTRACE);
782 return;
783 }
784 req = ktr_getrequest(KTR_GENIO);
785 if (req == NULL) {
786 free(buf, M_KTRACE);
787 return;
788 }
789 ktg = &req->ktr_data.ktr_genio;
790 ktg->ktr_fd = fd;
791 ktg->ktr_rw = rw;
792 req->ktr_header.ktr_len = datalen;
793 req->ktr_buffer = buf;
794 ktr_submitrequest(curthread, req);
795 }
796
797 void
ktrpsig(int sig,sig_t action,sigset_t * mask,int code)798 ktrpsig(int sig, sig_t action, sigset_t *mask, int code)
799 {
800 struct thread *td = curthread;
801 struct ktr_request *req;
802 struct ktr_psig *kp;
803
804 req = ktr_getrequest(KTR_PSIG);
805 if (req == NULL)
806 return;
807 kp = &req->ktr_data.ktr_psig;
808 kp->signo = (char)sig;
809 kp->action = action;
810 kp->mask = *mask;
811 kp->code = code;
812 ktr_enqueuerequest(td, req);
813 ktrace_exit(td);
814 }
815
816 void
ktrcsw(int out,int user,const char * wmesg)817 ktrcsw(int out, int user, const char *wmesg)
818 {
819 struct thread *td = curthread;
820 struct ktr_request *req;
821 struct ktr_csw *kc;
822
823 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
824 return;
825
826 req = ktr_getrequest(KTR_CSW);
827 if (req == NULL)
828 return;
829 kc = &req->ktr_data.ktr_csw;
830 kc->out = out;
831 kc->user = user;
832 if (wmesg != NULL)
833 strlcpy(kc->wmesg, wmesg, sizeof(kc->wmesg));
834 else
835 bzero(kc->wmesg, sizeof(kc->wmesg));
836 ktr_enqueuerequest(td, req);
837 ktrace_exit(td);
838 }
839
840 void
ktrstruct(const char * name,const void * data,size_t datalen)841 ktrstruct(const char *name, const void *data, size_t datalen)
842 {
843 struct ktr_request *req;
844 char *buf;
845 size_t buflen, namelen;
846
847 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
848 return;
849
850 if (data == NULL)
851 datalen = 0;
852 namelen = strlen(name) + 1;
853 buflen = namelen + datalen;
854 buf = malloc(buflen, M_KTRACE, M_WAITOK);
855 strcpy(buf, name);
856 bcopy(data, buf + namelen, datalen);
857 if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) {
858 free(buf, M_KTRACE);
859 return;
860 }
861 req->ktr_buffer = buf;
862 req->ktr_header.ktr_len = buflen;
863 ktr_submitrequest(curthread, req);
864 }
865
866 void
ktrstruct_error(const char * name,const void * data,size_t datalen,int error)867 ktrstruct_error(const char *name, const void *data, size_t datalen, int error)
868 {
869
870 if (error == 0)
871 ktrstruct(name, data, datalen);
872 }
873
874 void
ktrstructarray(const char * name,enum uio_seg seg,const void * data,int num_items,size_t struct_size)875 ktrstructarray(const char *name, enum uio_seg seg, const void *data,
876 int num_items, size_t struct_size)
877 {
878 struct ktr_request *req;
879 struct ktr_struct_array *ksa;
880 char *buf;
881 size_t buflen, datalen, namelen;
882 int max_items;
883
884 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
885 return;
886 if (num_items < 0)
887 return;
888
889 /* Trim array length to genio size. */
890 max_items = ktr_geniosize / struct_size;
891 if (num_items > max_items) {
892 if (max_items == 0)
893 num_items = 1;
894 else
895 num_items = max_items;
896 }
897 datalen = num_items * struct_size;
898
899 if (data == NULL)
900 datalen = 0;
901
902 namelen = strlen(name) + 1;
903 buflen = namelen + datalen;
904 buf = malloc(buflen, M_KTRACE, M_WAITOK);
905 strcpy(buf, name);
906 if (seg == UIO_SYSSPACE)
907 bcopy(data, buf + namelen, datalen);
908 else {
909 if (copyin(data, buf + namelen, datalen) != 0) {
910 free(buf, M_KTRACE);
911 return;
912 }
913 }
914 if ((req = ktr_getrequest(KTR_STRUCT_ARRAY)) == NULL) {
915 free(buf, M_KTRACE);
916 return;
917 }
918 ksa = &req->ktr_data.ktr_struct_array;
919 ksa->struct_size = struct_size;
920 req->ktr_buffer = buf;
921 req->ktr_header.ktr_len = buflen;
922 ktr_submitrequest(curthread, req);
923 }
924
925 void
ktrcapfail(enum ktr_cap_violation type,const void * data)926 ktrcapfail(enum ktr_cap_violation type, const void *data)
927 {
928 struct thread *td = curthread;
929 struct ktr_request *req;
930 struct ktr_cap_fail *kcf;
931 union ktr_cap_data *kcd;
932
933 if (__predict_false(td->td_pflags & TDP_INKTRACE))
934 return;
935 if (type != CAPFAIL_SYSCALL &&
936 (td->td_sa.callp->sy_flags & SYF_CAPENABLED) == 0)
937 return;
938
939 req = ktr_getrequest(KTR_CAPFAIL);
940 if (req == NULL)
941 return;
942 kcf = &req->ktr_data.ktr_cap_fail;
943 kcf->cap_type = type;
944 kcf->cap_code = td->td_sa.code;
945 kcf->cap_svflags = td->td_proc->p_sysent->sv_flags;
946 if (data != NULL) {
947 kcd = &kcf->cap_data;
948 switch (type) {
949 case CAPFAIL_NOTCAPABLE:
950 case CAPFAIL_INCREASE:
951 kcd->cap_needed = *(const cap_rights_t *)data;
952 kcd->cap_held = *((const cap_rights_t *)data + 1);
953 break;
954 case CAPFAIL_SYSCALL:
955 case CAPFAIL_SIGNAL:
956 case CAPFAIL_PROTO:
957 kcd->cap_int = *(const int *)data;
958 break;
959 case CAPFAIL_SOCKADDR:
960 kcd->cap_sockaddr = *(const struct sockaddr *)data;
961 break;
962 case CAPFAIL_NAMEI:
963 strlcpy(kcd->cap_path, data, MAXPATHLEN);
964 break;
965 case CAPFAIL_CPUSET:
966 default:
967 break;
968 }
969 }
970 ktr_enqueuerequest(td, req);
971 ktrace_exit(td);
972 }
973
974 void
ktrfault(vm_offset_t vaddr,int type)975 ktrfault(vm_offset_t vaddr, int type)
976 {
977 struct thread *td = curthread;
978 struct ktr_request *req;
979 struct ktr_fault *kf;
980
981 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
982 return;
983
984 req = ktr_getrequest(KTR_FAULT);
985 if (req == NULL)
986 return;
987 kf = &req->ktr_data.ktr_fault;
988 kf->vaddr = vaddr;
989 kf->type = type;
990 ktr_enqueuerequest(td, req);
991 ktrace_exit(td);
992 }
993
994 void
ktrfaultend(int result)995 ktrfaultend(int result)
996 {
997 struct thread *td = curthread;
998 struct ktr_request *req;
999 struct ktr_faultend *kf;
1000
1001 if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
1002 return;
1003
1004 req = ktr_getrequest(KTR_FAULTEND);
1005 if (req == NULL)
1006 return;
1007 kf = &req->ktr_data.ktr_faultend;
1008 kf->result = result;
1009 ktr_enqueuerequest(td, req);
1010 ktrace_exit(td);
1011 }
1012 #endif /* KTRACE */
1013
1014 /* Interface and common routines */
1015
1016 #ifndef _SYS_SYSPROTO_H_
1017 struct ktrace_args {
1018 char *fname;
1019 int ops;
1020 int facs;
1021 int pid;
1022 };
1023 #endif
1024 /* ARGSUSED */
1025 int
sys_ktrace(struct thread * td,struct ktrace_args * uap)1026 sys_ktrace(struct thread *td, struct ktrace_args *uap)
1027 {
1028 #ifdef KTRACE
1029 struct vnode *vp = NULL;
1030 struct proc *p;
1031 struct pgrp *pg;
1032 int facs = uap->facs & ~KTRFAC_ROOT;
1033 int ops = KTROP(uap->ops);
1034 int descend = uap->ops & KTRFLAG_DESCEND;
1035 int ret = 0;
1036 int flags, error = 0;
1037 struct nameidata nd;
1038 struct ktr_io_params *kiop, *old_kiop;
1039
1040 /*
1041 * Need something to (un)trace.
1042 */
1043 if (ops != KTROP_CLEARFILE && facs == 0)
1044 return (EINVAL);
1045
1046 kiop = NULL;
1047 if (ops != KTROP_CLEAR) {
1048 /*
1049 * an operation which requires a file argument.
1050 */
1051 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname);
1052 flags = FREAD | FWRITE | O_NOFOLLOW;
1053 error = vn_open(&nd, &flags, 0, NULL);
1054 if (error)
1055 return (error);
1056 NDFREE_PNBUF(&nd);
1057 vp = nd.ni_vp;
1058 VOP_UNLOCK(vp);
1059 if (vp->v_type != VREG) {
1060 (void)vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
1061 return (EACCES);
1062 }
1063 kiop = ktr_io_params_alloc(td, vp);
1064 }
1065
1066 /*
1067 * Clear all uses of the tracefile.
1068 */
1069 ktrace_enter(td);
1070 if (ops == KTROP_CLEARFILE) {
1071 restart:
1072 sx_slock(&allproc_lock);
1073 FOREACH_PROC_IN_SYSTEM(p) {
1074 old_kiop = NULL;
1075 PROC_LOCK(p);
1076 if (p->p_ktrioparms != NULL &&
1077 p->p_ktrioparms->vp == vp) {
1078 if (ktrcanset(td, p)) {
1079 mtx_lock(&ktrace_mtx);
1080 old_kiop = ktr_freeproc(p);
1081 mtx_unlock(&ktrace_mtx);
1082 } else
1083 error = EPERM;
1084 }
1085 PROC_UNLOCK(p);
1086 if (old_kiop != NULL) {
1087 sx_sunlock(&allproc_lock);
1088 ktr_io_params_free(old_kiop);
1089 goto restart;
1090 }
1091 }
1092 sx_sunlock(&allproc_lock);
1093 goto done;
1094 }
1095 /*
1096 * do it
1097 */
1098 sx_slock(&proctree_lock);
1099 if (uap->pid < 0) {
1100 /*
1101 * by process group
1102 */
1103 pg = pgfind(-uap->pid);
1104 if (pg == NULL) {
1105 sx_sunlock(&proctree_lock);
1106 error = ESRCH;
1107 goto done;
1108 }
1109
1110 /*
1111 * ktrops() may call vrele(). Lock pg_members
1112 * by the proctree_lock rather than pg_mtx.
1113 */
1114 PGRP_UNLOCK(pg);
1115 if (LIST_EMPTY(&pg->pg_members)) {
1116 sx_sunlock(&proctree_lock);
1117 error = ESRCH;
1118 goto done;
1119 }
1120 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1121 PROC_LOCK(p);
1122 if (descend)
1123 ret |= ktrsetchildren(td, p, ops, facs, kiop);
1124 else
1125 ret |= ktrops(td, p, ops, facs, kiop);
1126 }
1127 } else {
1128 /*
1129 * by pid
1130 */
1131 p = pfind(uap->pid);
1132 if (p == NULL) {
1133 error = ESRCH;
1134 sx_sunlock(&proctree_lock);
1135 goto done;
1136 }
1137 if (descend)
1138 ret |= ktrsetchildren(td, p, ops, facs, kiop);
1139 else
1140 ret |= ktrops(td, p, ops, facs, kiop);
1141 }
1142 sx_sunlock(&proctree_lock);
1143 if (!ret)
1144 error = EPERM;
1145 done:
1146 if (kiop != NULL) {
1147 mtx_lock(&ktrace_mtx);
1148 kiop = ktr_io_params_rele(kiop);
1149 mtx_unlock(&ktrace_mtx);
1150 ktr_io_params_free(kiop);
1151 }
1152 ktrace_exit(td);
1153 return (error);
1154 #else /* !KTRACE */
1155 return (ENOSYS);
1156 #endif /* KTRACE */
1157 }
1158
1159 /* ARGSUSED */
1160 int
sys_utrace(struct thread * td,struct utrace_args * uap)1161 sys_utrace(struct thread *td, struct utrace_args *uap)
1162 {
1163
1164 #ifdef KTRACE
1165 struct ktr_request *req;
1166 void *cp;
1167 int error;
1168
1169 if (!KTRPOINT(td, KTR_USER))
1170 return (0);
1171 if (uap->len > KTR_USER_MAXLEN)
1172 return (EINVAL);
1173 cp = malloc(uap->len, M_KTRACE, M_WAITOK);
1174 error = copyin(uap->addr, cp, uap->len);
1175 if (error) {
1176 free(cp, M_KTRACE);
1177 return (error);
1178 }
1179 req = ktr_getrequest(KTR_USER);
1180 if (req == NULL) {
1181 free(cp, M_KTRACE);
1182 return (ENOMEM);
1183 }
1184 req->ktr_buffer = cp;
1185 req->ktr_header.ktr_len = uap->len;
1186 ktr_submitrequest(td, req);
1187 return (0);
1188 #else /* !KTRACE */
1189 return (ENOSYS);
1190 #endif /* KTRACE */
1191 }
1192
1193 #ifdef KTRACE
1194 static int
ktrops(struct thread * td,struct proc * p,int ops,int facs,struct ktr_io_params * new_kiop)1195 ktrops(struct thread *td, struct proc *p, int ops, int facs,
1196 struct ktr_io_params *new_kiop)
1197 {
1198 struct ktr_io_params *old_kiop;
1199
1200 PROC_LOCK_ASSERT(p, MA_OWNED);
1201 if (!ktrcanset(td, p)) {
1202 PROC_UNLOCK(p);
1203 return (0);
1204 }
1205 if ((ops == KTROP_SET && p->p_state == PRS_NEW) ||
1206 p_cansee(td, p) != 0) {
1207 /*
1208 * Disallow setting trace points if the process is being born.
1209 * This avoids races with trace point inheritance in
1210 * ktrprocfork().
1211 */
1212 PROC_UNLOCK(p);
1213 return (0);
1214 }
1215 if ((p->p_flag & P_WEXIT) != 0) {
1216 /*
1217 * There's nothing to do if the process is exiting, but avoid
1218 * signaling an error.
1219 */
1220 PROC_UNLOCK(p);
1221 return (1);
1222 }
1223 old_kiop = NULL;
1224 mtx_lock(&ktrace_mtx);
1225 if (ops == KTROP_SET) {
1226 if (p->p_ktrioparms != NULL &&
1227 p->p_ktrioparms->vp != new_kiop->vp) {
1228 /* if trace file already in use, relinquish below */
1229 old_kiop = ktr_io_params_rele(p->p_ktrioparms);
1230 p->p_ktrioparms = NULL;
1231 }
1232 if (p->p_ktrioparms == NULL) {
1233 p->p_ktrioparms = new_kiop;
1234 ktr_io_params_ref(new_kiop);
1235 }
1236 p->p_traceflag |= facs;
1237 if (priv_check(td, PRIV_KTRACE) == 0)
1238 p->p_traceflag |= KTRFAC_ROOT;
1239 } else {
1240 /* KTROP_CLEAR */
1241 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0)
1242 /* no more tracing */
1243 old_kiop = ktr_freeproc(p);
1244 }
1245 mtx_unlock(&ktrace_mtx);
1246 if ((p->p_traceflag & KTRFAC_MASK) != 0)
1247 ktrprocctor_entered(td, p);
1248 PROC_UNLOCK(p);
1249 ktr_io_params_free(old_kiop);
1250
1251 return (1);
1252 }
1253
1254 static int
ktrsetchildren(struct thread * td,struct proc * top,int ops,int facs,struct ktr_io_params * new_kiop)1255 ktrsetchildren(struct thread *td, struct proc *top, int ops, int facs,
1256 struct ktr_io_params *new_kiop)
1257 {
1258 struct proc *p;
1259 int ret = 0;
1260
1261 p = top;
1262 PROC_LOCK_ASSERT(p, MA_OWNED);
1263 sx_assert(&proctree_lock, SX_LOCKED);
1264 for (;;) {
1265 ret |= ktrops(td, p, ops, facs, new_kiop);
1266 /*
1267 * If this process has children, descend to them next,
1268 * otherwise do any siblings, and if done with this level,
1269 * follow back up the tree (but not past top).
1270 */
1271 if (!LIST_EMPTY(&p->p_children))
1272 p = LIST_FIRST(&p->p_children);
1273 else for (;;) {
1274 if (p == top)
1275 return (ret);
1276 if (LIST_NEXT(p, p_sibling)) {
1277 p = LIST_NEXT(p, p_sibling);
1278 break;
1279 }
1280 p = p->p_pptr;
1281 }
1282 PROC_LOCK(p);
1283 }
1284 /*NOTREACHED*/
1285 }
1286
1287 static void
ktr_writerequest(struct thread * td,struct ktr_request * req)1288 ktr_writerequest(struct thread *td, struct ktr_request *req)
1289 {
1290 struct ktr_io_params *kiop, *kiop1;
1291 struct ktr_header *kth;
1292 struct vnode *vp;
1293 struct proc *p;
1294 struct ucred *cred;
1295 struct uio auio;
1296 struct iovec aiov[3];
1297 struct mount *mp;
1298 off_t lim;
1299 int datalen, buflen;
1300 int error;
1301
1302 p = td->td_proc;
1303
1304 /*
1305 * We reference the kiop for use in I/O in case ktrace is
1306 * disabled on the process as we write out the request.
1307 */
1308 mtx_lock(&ktrace_mtx);
1309 kiop = p->p_ktrioparms;
1310
1311 /*
1312 * If kiop is NULL, it has been cleared out from under this
1313 * request, so just drop it.
1314 */
1315 if (kiop == NULL) {
1316 mtx_unlock(&ktrace_mtx);
1317 return;
1318 }
1319
1320 ktr_io_params_ref(kiop);
1321 vp = kiop->vp;
1322 cred = kiop->cr;
1323 lim = kiop->lim;
1324
1325 KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL"));
1326 mtx_unlock(&ktrace_mtx);
1327
1328 kth = &req->ktr_header;
1329 KASSERT(((u_short)kth->ktr_type & ~KTR_TYPE) < nitems(data_lengths),
1330 ("data_lengths array overflow"));
1331 datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_TYPE];
1332 buflen = kth->ktr_len;
1333 auio.uio_iov = &aiov[0];
1334 auio.uio_offset = 0;
1335 auio.uio_segflg = UIO_SYSSPACE;
1336 auio.uio_rw = UIO_WRITE;
1337 aiov[0].iov_base = (caddr_t)kth;
1338 aiov[0].iov_len = sizeof(struct ktr_header);
1339 auio.uio_resid = sizeof(struct ktr_header);
1340 auio.uio_iovcnt = 1;
1341 auio.uio_td = td;
1342 if (datalen != 0) {
1343 aiov[1].iov_base = (caddr_t)&req->ktr_data;
1344 aiov[1].iov_len = datalen;
1345 auio.uio_resid += datalen;
1346 auio.uio_iovcnt++;
1347 kth->ktr_len += datalen;
1348 }
1349 if (buflen != 0) {
1350 KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write"));
1351 aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer;
1352 aiov[auio.uio_iovcnt].iov_len = buflen;
1353 auio.uio_resid += buflen;
1354 auio.uio_iovcnt++;
1355 }
1356
1357 vn_start_write(vp, &mp, V_WAIT);
1358 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1359 td->td_ktr_io_lim = lim;
1360 #ifdef MAC
1361 error = mac_vnode_check_write(cred, NOCRED, vp);
1362 if (error == 0)
1363 #endif
1364 error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred);
1365 VOP_UNLOCK(vp);
1366 vn_finished_write(mp);
1367 if (error == 0) {
1368 mtx_lock(&ktrace_mtx);
1369 kiop = ktr_io_params_rele(kiop);
1370 mtx_unlock(&ktrace_mtx);
1371 ktr_io_params_free(kiop);
1372 return;
1373 }
1374
1375 /*
1376 * If error encountered, give up tracing on this vnode on this
1377 * process. Other processes might still be suitable for
1378 * writes to this vnode.
1379 */
1380 log(LOG_NOTICE,
1381 "ktrace write failed, errno %d, tracing stopped for pid %d\n",
1382 error, p->p_pid);
1383
1384 kiop1 = NULL;
1385 PROC_LOCK(p);
1386 mtx_lock(&ktrace_mtx);
1387 if (p->p_ktrioparms != NULL && p->p_ktrioparms->vp == vp)
1388 kiop1 = ktr_freeproc(p);
1389 kiop = ktr_io_params_rele(kiop);
1390 mtx_unlock(&ktrace_mtx);
1391 PROC_UNLOCK(p);
1392 ktr_io_params_free(kiop1);
1393 ktr_io_params_free(kiop);
1394 }
1395
1396 /*
1397 * Return true if caller has permission to set the ktracing state
1398 * of target. Essentially, the target can't possess any
1399 * more permissions than the caller. KTRFAC_ROOT signifies that
1400 * root previously set the tracing status on the target process, and
1401 * so, only root may further change it.
1402 */
1403 static int
ktrcanset(struct thread * td,struct proc * targetp)1404 ktrcanset(struct thread *td, struct proc *targetp)
1405 {
1406
1407 PROC_LOCK_ASSERT(targetp, MA_OWNED);
1408 if (targetp->p_traceflag & KTRFAC_ROOT &&
1409 priv_check(td, PRIV_KTRACE))
1410 return (0);
1411
1412 if (p_candebug(td, targetp) != 0)
1413 return (0);
1414
1415 return (1);
1416 }
1417
1418 #endif /* KTRACE */
1419