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