xref: /freebsd/sys/kern/kern_ktrace.c (revision 65a4daeaf3247b7a2c16cf59cdea2ce05987a7cc)
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 	[KTR_ARGS] = 0,
128 	[KTR_ENVS] = 0,
129 };
130 
131 static STAILQ_HEAD(, ktr_request) ktr_free;
132 
133 static SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
134     "KTRACE options");
135 
136 static u_int ktr_requestpool = KTRACE_REQUEST_POOL;
137 TUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool);
138 
139 u_int ktr_geniosize = PAGE_SIZE;
140 SYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RWTUN, &ktr_geniosize,
141     0, "Maximum size of genio event payload");
142 
143 /*
144  * Allow to not to send signal to traced process, in which context the
145  * ktr record is written.  The limit is applied from the process that
146  * set up ktrace, so killing the traced process is not completely fair.
147  */
148 int ktr_filesize_limit_signal = 0;
149 SYSCTL_INT(_kern_ktrace, OID_AUTO, filesize_limit_signal, CTLFLAG_RWTUN,
150     &ktr_filesize_limit_signal, 0,
151     "Send SIGXFSZ to the traced process when the log size limit is exceeded");
152 
153 static int print_message = 1;
154 static struct mtx ktrace_mtx;
155 static struct sx ktrace_sx;
156 
157 struct ktr_io_params {
158 	struct vnode	*vp;
159 	struct ucred	*cr;
160 	off_t		lim;
161 	u_int		refs;
162 };
163 
164 static void ktrace_init(void *dummy);
165 static int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS);
166 static u_int ktrace_resize_pool(u_int oldsize, u_int newsize);
167 static struct ktr_request *ktr_getrequest_entered(struct thread *td, int type);
168 static struct ktr_request *ktr_getrequest(int type);
169 static void ktr_submitrequest(struct thread *td, struct ktr_request *req);
170 static struct ktr_io_params *ktr_freeproc(struct proc *p);
171 static void ktr_freerequest(struct ktr_request *req);
172 static void ktr_freerequest_locked(struct ktr_request *req);
173 static void ktr_writerequest(struct thread *td, struct ktr_request *req);
174 static int ktrcanset(struct thread *,struct proc *);
175 static int ktrsetchildren(struct thread *, struct proc *, int, int,
176     struct ktr_io_params *);
177 static int ktrops(struct thread *, struct proc *, int, int,
178     struct ktr_io_params *);
179 static void ktrprocctor_entered(struct thread *, struct proc *);
180 
181 /*
182  * ktrace itself generates events, such as context switches, which we do not
183  * wish to trace.  Maintain a flag, TDP_INKTRACE, on each thread to determine
184  * whether or not it is in a region where tracing of events should be
185  * suppressed.
186  */
187 static void
ktrace_enter(struct thread * td)188 ktrace_enter(struct thread *td)
189 {
190 
191 	KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set"));
192 	td->td_pflags |= TDP_INKTRACE;
193 }
194 
195 static void
ktrace_exit(struct thread * td)196 ktrace_exit(struct thread *td)
197 {
198 
199 	KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set"));
200 	td->td_pflags &= ~TDP_INKTRACE;
201 }
202 
203 static void
ktrace_assert(struct thread * td)204 ktrace_assert(struct thread *td)
205 {
206 
207 	KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set"));
208 }
209 
210 static void
ast_ktrace(struct thread * td,int tda __unused)211 ast_ktrace(struct thread *td, int tda __unused)
212 {
213 	KTRUSERRET(td);
214 }
215 
216 static void
ktrace_init(void * dummy)217 ktrace_init(void *dummy)
218 {
219 	struct ktr_request *req;
220 	int i;
221 
222 	mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET);
223 	sx_init(&ktrace_sx, "ktrace_sx");
224 	STAILQ_INIT(&ktr_free);
225 	for (i = 0; i < ktr_requestpool; i++) {
226 		req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK |
227 		    M_ZERO);
228 		STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
229 	}
230 	ast_register(TDA_KTRACE, ASTR_ASTF_REQUIRED, 0, ast_ktrace);
231 }
232 SYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL);
233 
234 static int
sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)235 sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS)
236 {
237 	struct thread *td;
238 	u_int newsize, oldsize, wantsize;
239 	int error;
240 
241 	/* Handle easy read-only case first to avoid warnings from GCC. */
242 	if (!req->newptr) {
243 		oldsize = ktr_requestpool;
244 		return (SYSCTL_OUT(req, &oldsize, sizeof(u_int)));
245 	}
246 
247 	error = SYSCTL_IN(req, &wantsize, sizeof(u_int));
248 	if (error)
249 		return (error);
250 	td = curthread;
251 	ktrace_enter(td);
252 	oldsize = ktr_requestpool;
253 	newsize = ktrace_resize_pool(oldsize, wantsize);
254 	ktrace_exit(td);
255 	error = SYSCTL_OUT(req, &oldsize, sizeof(u_int));
256 	if (error)
257 		return (error);
258 	if (wantsize > oldsize && newsize < wantsize)
259 		return (ENOSPC);
260 	return (0);
261 }
262 SYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool,
263     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &ktr_requestpool, 0,
264     sysctl_kern_ktrace_request_pool, "IU",
265     "Pool buffer size for ktrace(1)");
266 
267 static u_int
ktrace_resize_pool(u_int oldsize,u_int newsize)268 ktrace_resize_pool(u_int oldsize, u_int newsize)
269 {
270 	STAILQ_HEAD(, ktr_request) ktr_new;
271 	struct ktr_request *req;
272 	int bound;
273 
274 	print_message = 1;
275 	bound = newsize - oldsize;
276 	if (bound == 0)
277 		return (ktr_requestpool);
278 	if (bound < 0) {
279 		mtx_lock(&ktrace_mtx);
280 		/* Shrink pool down to newsize if possible. */
281 		while (bound++ < 0) {
282 			req = STAILQ_FIRST(&ktr_free);
283 			if (req == NULL)
284 				break;
285 			STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
286 			ktr_requestpool--;
287 			free(req, M_KTRACE);
288 		}
289 	} else {
290 		/* Grow pool up to newsize. */
291 		STAILQ_INIT(&ktr_new);
292 		while (bound-- > 0) {
293 			req = malloc(sizeof(struct ktr_request), M_KTRACE,
294 			    M_WAITOK | M_ZERO);
295 			STAILQ_INSERT_HEAD(&ktr_new, req, ktr_list);
296 		}
297 		mtx_lock(&ktrace_mtx);
298 		STAILQ_CONCAT(&ktr_free, &ktr_new);
299 		ktr_requestpool += (newsize - oldsize);
300 	}
301 	mtx_unlock(&ktrace_mtx);
302 	return (ktr_requestpool);
303 }
304 
305 /* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */
306 CTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) ==
307     (sizeof((struct thread *)NULL)->td_name));
308 
309 static struct ktr_request *
ktr_getrequest_entered(struct thread * td,int type)310 ktr_getrequest_entered(struct thread *td, int type)
311 {
312 	struct ktr_request *req;
313 	struct proc *p = td->td_proc;
314 	int pm;
315 
316 	mtx_lock(&ktrace_mtx);
317 	if (!KTRCHECK(td, type)) {
318 		mtx_unlock(&ktrace_mtx);
319 		return (NULL);
320 	}
321 	req = STAILQ_FIRST(&ktr_free);
322 	if (req != NULL) {
323 		STAILQ_REMOVE_HEAD(&ktr_free, ktr_list);
324 		req->ktr_header.ktr_type = type;
325 		if (p->p_traceflag & KTRFAC_DROP) {
326 			req->ktr_header.ktr_type |= KTR_DROP;
327 			p->p_traceflag &= ~KTRFAC_DROP;
328 		}
329 		mtx_unlock(&ktrace_mtx);
330 		nanotime(&req->ktr_header.ktr_time);
331 		req->ktr_header.ktr_type |= KTR_VERSIONED;
332 		req->ktr_header.ktr_pid = p->p_pid;
333 		req->ktr_header.ktr_tid = td->td_tid;
334 		req->ktr_header.ktr_cpu = PCPU_GET(cpuid);
335 		req->ktr_header.ktr_version = KTR_VERSION1;
336 		bcopy(td->td_name, req->ktr_header.ktr_comm,
337 		    sizeof(req->ktr_header.ktr_comm));
338 		req->ktr_buffer = NULL;
339 		req->ktr_header.ktr_len = 0;
340 	} else {
341 		p->p_traceflag |= KTRFAC_DROP;
342 		pm = print_message;
343 		print_message = 0;
344 		mtx_unlock(&ktrace_mtx);
345 		if (pm)
346 			printf("Out of ktrace request objects.\n");
347 	}
348 	return (req);
349 }
350 
351 static struct ktr_request *
ktr_getrequest(int type)352 ktr_getrequest(int type)
353 {
354 	struct thread *td = curthread;
355 	struct ktr_request *req;
356 
357 	ktrace_enter(td);
358 	req = ktr_getrequest_entered(td, type);
359 	if (req == NULL)
360 		ktrace_exit(td);
361 
362 	return (req);
363 }
364 
365 /*
366  * Some trace generation environments don't permit direct access to VFS,
367  * such as during a context switch where sleeping is not allowed.  Under these
368  * circumstances, queue a request to the thread to be written asynchronously
369  * later.
370  */
371 static void
ktr_enqueuerequest(struct thread * td,struct ktr_request * req)372 ktr_enqueuerequest(struct thread *td, struct ktr_request *req)
373 {
374 
375 	mtx_lock(&ktrace_mtx);
376 	STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list);
377 	mtx_unlock(&ktrace_mtx);
378 	ast_sched(td, TDA_KTRACE);
379 }
380 
381 /*
382  * Drain any pending ktrace records from the per-thread queue to disk.  This
383  * is used both internally before committing other records, and also on
384  * system call return.  We drain all the ones we can find at the time when
385  * drain is requested, but don't keep draining after that as those events
386  * may be approximately "after" the current event.
387  */
388 static void
ktr_drain(struct thread * td)389 ktr_drain(struct thread *td)
390 {
391 	struct ktr_request *queued_req;
392 	STAILQ_HEAD(, ktr_request) local_queue;
393 
394 	ktrace_assert(td);
395 	sx_assert(&ktrace_sx, SX_XLOCKED);
396 
397 	STAILQ_INIT(&local_queue);
398 
399 	if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) {
400 		mtx_lock(&ktrace_mtx);
401 		STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr);
402 		mtx_unlock(&ktrace_mtx);
403 
404 		while ((queued_req = STAILQ_FIRST(&local_queue))) {
405 			STAILQ_REMOVE_HEAD(&local_queue, ktr_list);
406 			ktr_writerequest(td, queued_req);
407 			ktr_freerequest(queued_req);
408 		}
409 	}
410 }
411 
412 /*
413  * Submit a trace record for immediate commit to disk -- to be used only
414  * where entering VFS is OK.  First drain any pending records that may have
415  * been cached in the thread.
416  */
417 static void
ktr_submitrequest(struct thread * td,struct ktr_request * req)418 ktr_submitrequest(struct thread *td, struct ktr_request *req)
419 {
420 
421 	ktrace_assert(td);
422 
423 	sx_xlock(&ktrace_sx);
424 	ktr_drain(td);
425 	ktr_writerequest(td, req);
426 	ktr_freerequest(req);
427 	sx_xunlock(&ktrace_sx);
428 	ktrace_exit(td);
429 }
430 
431 static void
ktr_freerequest(struct ktr_request * req)432 ktr_freerequest(struct ktr_request *req)
433 {
434 
435 	mtx_lock(&ktrace_mtx);
436 	ktr_freerequest_locked(req);
437 	mtx_unlock(&ktrace_mtx);
438 }
439 
440 static void
ktr_freerequest_locked(struct ktr_request * req)441 ktr_freerequest_locked(struct ktr_request *req)
442 {
443 
444 	mtx_assert(&ktrace_mtx, MA_OWNED);
445 	if (req->ktr_buffer != NULL)
446 		free(req->ktr_buffer, M_KTRACE);
447 	STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list);
448 }
449 
450 static void
ktr_io_params_ref(struct ktr_io_params * kiop)451 ktr_io_params_ref(struct ktr_io_params *kiop)
452 {
453 	mtx_assert(&ktrace_mtx, MA_OWNED);
454 	kiop->refs++;
455 }
456 
457 static struct ktr_io_params *
ktr_io_params_rele(struct ktr_io_params * kiop)458 ktr_io_params_rele(struct ktr_io_params *kiop)
459 {
460 	mtx_assert(&ktrace_mtx, MA_OWNED);
461 	if (kiop == NULL)
462 		return (NULL);
463 	KASSERT(kiop->refs > 0, ("kiop ref == 0 %p", kiop));
464 	return (--(kiop->refs) == 0 ? kiop : NULL);
465 }
466 
467 void
ktr_io_params_free(struct ktr_io_params * kiop)468 ktr_io_params_free(struct ktr_io_params *kiop)
469 {
470 	if (kiop == NULL)
471 		return;
472 
473 	MPASS(kiop->refs == 0);
474 	vn_close(kiop->vp, FWRITE, kiop->cr, curthread);
475 	crfree(kiop->cr);
476 	free(kiop, M_KTRACE);
477 }
478 
479 static struct ktr_io_params *
ktr_io_params_alloc(struct thread * td,struct vnode * vp)480 ktr_io_params_alloc(struct thread *td, struct vnode *vp)
481 {
482 	struct ktr_io_params *res;
483 
484 	res = malloc(sizeof(struct ktr_io_params), M_KTRACE, M_WAITOK);
485 	res->vp = vp;
486 	res->cr = crhold(td->td_ucred);
487 	res->lim = lim_cur(td, RLIMIT_FSIZE);
488 	res->refs = 1;
489 	return (res);
490 }
491 
492 /*
493  * Disable tracing for a process and release all associated resources.
494  * The caller is responsible for releasing a reference on the returned
495  * vnode and credentials.
496  */
497 static struct ktr_io_params *
ktr_freeproc(struct proc * p)498 ktr_freeproc(struct proc *p)
499 {
500 	struct ktr_io_params *kiop;
501 	struct ktr_request *req;
502 
503 	PROC_LOCK_ASSERT(p, MA_OWNED);
504 	mtx_assert(&ktrace_mtx, MA_OWNED);
505 	kiop = ktr_io_params_rele(p->p_ktrioparms);
506 	p->p_ktrioparms = NULL;
507 	p->p_traceflag = 0;
508 	while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) {
509 		STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list);
510 		ktr_freerequest_locked(req);
511 	}
512 	return (kiop);
513 }
514 
515 struct vnode *
ktr_get_tracevp(struct proc * p,bool ref)516 ktr_get_tracevp(struct proc *p, bool ref)
517 {
518 	struct vnode *vp;
519 
520 	PROC_LOCK_ASSERT(p, MA_OWNED);
521 
522 	if (p->p_ktrioparms != NULL) {
523 		vp = p->p_ktrioparms->vp;
524 		if (ref)
525 			vrefact(vp);
526 	} else {
527 		vp = NULL;
528 	}
529 	return (vp);
530 }
531 
532 void
ktrsyscall(int code,int narg,syscallarg_t args[])533 ktrsyscall(int code, int narg, syscallarg_t args[])
534 {
535 	struct ktr_request *req;
536 	struct ktr_syscall *ktp;
537 	size_t buflen;
538 	char *buf = NULL;
539 
540 	if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
541 		return;
542 
543 	buflen = sizeof(register_t) * narg;
544 	if (buflen > 0) {
545 		buf = malloc(buflen, M_KTRACE, M_WAITOK);
546 		bcopy(args, buf, buflen);
547 	}
548 	req = ktr_getrequest(KTR_SYSCALL);
549 	if (req == NULL) {
550 		if (buf != NULL)
551 			free(buf, M_KTRACE);
552 		return;
553 	}
554 	ktp = &req->ktr_data.ktr_syscall;
555 	ktp->ktr_code = code;
556 	ktp->ktr_narg = narg;
557 	if (buflen > 0) {
558 		req->ktr_header.ktr_len = buflen;
559 		req->ktr_buffer = buf;
560 	}
561 	ktr_submitrequest(curthread, req);
562 }
563 
564 void
ktrdata(int type,const void * data,size_t len)565 ktrdata(int type, const void *data, size_t len)
566 {
567         struct ktr_request *req;
568         void *buf;
569 
570         if ((req = ktr_getrequest(type)) == NULL)
571                 return;
572         buf = malloc(len, M_KTRACE, M_WAITOK);
573         bcopy(data, buf, len);
574         req->ktr_header.ktr_len = len;
575         req->ktr_buffer = buf;
576         ktr_submitrequest(curthread, req);
577 }
578 
579 void
ktrsysret(int code,int error,register_t retval)580 ktrsysret(int code, int error, register_t retval)
581 {
582 	struct ktr_request *req;
583 	struct ktr_sysret *ktp;
584 
585 	if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
586 		return;
587 
588 	req = ktr_getrequest(KTR_SYSRET);
589 	if (req == NULL)
590 		return;
591 	ktp = &req->ktr_data.ktr_sysret;
592 	ktp->ktr_code = code;
593 	ktp->ktr_error = error;
594 	ktp->ktr_retval = ((error == 0) ? retval: 0);		/* what about val2 ? */
595 	ktr_submitrequest(curthread, req);
596 }
597 
598 /*
599  * When a setuid process execs, disable tracing.
600  *
601  * XXX: We toss any pending asynchronous records.
602  */
603 struct ktr_io_params *
ktrprocexec(struct proc * p)604 ktrprocexec(struct proc *p)
605 {
606 	struct ktr_io_params *kiop;
607 
608 	PROC_LOCK_ASSERT(p, MA_OWNED);
609 
610 	kiop = p->p_ktrioparms;
611 	if (kiop == NULL || priv_check_cred(kiop->cr, PRIV_DEBUG_DIFFCRED) == 0)
612 		return (NULL);
613 
614 	mtx_lock(&ktrace_mtx);
615 	kiop = ktr_freeproc(p);
616 	mtx_unlock(&ktrace_mtx);
617 	return (kiop);
618 }
619 
620 /*
621  * When a process exits, drain per-process asynchronous trace records
622  * and disable tracing.
623  */
624 void
ktrprocexit(struct thread * td)625 ktrprocexit(struct thread *td)
626 {
627 	struct ktr_request *req;
628 	struct proc *p;
629 	struct ktr_io_params *kiop;
630 
631 	p = td->td_proc;
632 	if (p->p_traceflag == 0)
633 		return;
634 
635 	ktrace_enter(td);
636 	req = ktr_getrequest_entered(td, KTR_PROCDTOR);
637 	if (req != NULL)
638 		ktr_enqueuerequest(td, req);
639 	sx_xlock(&ktrace_sx);
640 	ktr_drain(td);
641 	sx_xunlock(&ktrace_sx);
642 	PROC_LOCK(p);
643 	mtx_lock(&ktrace_mtx);
644 	kiop = ktr_freeproc(p);
645 	mtx_unlock(&ktrace_mtx);
646 	PROC_UNLOCK(p);
647 	ktr_io_params_free(kiop);
648 	ktrace_exit(td);
649 }
650 
651 static void
ktrprocctor_entered(struct thread * td,struct proc * p)652 ktrprocctor_entered(struct thread *td, struct proc *p)
653 {
654 	struct ktr_proc_ctor *ktp;
655 	struct ktr_request *req;
656 	struct thread *td2;
657 
658 	ktrace_assert(td);
659 	td2 = FIRST_THREAD_IN_PROC(p);
660 	req = ktr_getrequest_entered(td2, KTR_PROCCTOR);
661 	if (req == NULL)
662 		return;
663 	ktp = &req->ktr_data.ktr_proc_ctor;
664 	ktp->sv_flags = p->p_sysent->sv_flags;
665 	ktr_enqueuerequest(td2, req);
666 }
667 
668 void
ktrprocctor(struct proc * p)669 ktrprocctor(struct proc *p)
670 {
671 	struct thread *td = curthread;
672 
673 	if ((p->p_traceflag & KTRFAC_MASK) == 0)
674 		return;
675 
676 	ktrace_enter(td);
677 	ktrprocctor_entered(td, p);
678 	ktrace_exit(td);
679 }
680 
681 /*
682  * When a process forks, enable tracing in the new process if needed.
683  */
684 void
ktrprocfork(struct proc * p1,struct proc * p2)685 ktrprocfork(struct proc *p1, struct proc *p2)
686 {
687 
688 	MPASS(p2->p_ktrioparms == NULL);
689 	MPASS(p2->p_traceflag == 0);
690 
691 	if (p1->p_traceflag == 0)
692 		return;
693 
694 	PROC_LOCK(p1);
695 	mtx_lock(&ktrace_mtx);
696 	if (p1->p_traceflag & KTRFAC_INHERIT) {
697 		p2->p_traceflag = p1->p_traceflag;
698 		if ((p2->p_ktrioparms = p1->p_ktrioparms) != NULL)
699 			p1->p_ktrioparms->refs++;
700 	}
701 	mtx_unlock(&ktrace_mtx);
702 	PROC_UNLOCK(p1);
703 
704 	ktrprocctor(p2);
705 }
706 
707 /*
708  * When a thread returns, drain any asynchronous records generated by the
709  * system call.
710  */
711 void
ktruserret(struct thread * td)712 ktruserret(struct thread *td)
713 {
714 
715 	ktrace_enter(td);
716 	sx_xlock(&ktrace_sx);
717 	ktr_drain(td);
718 	sx_xunlock(&ktrace_sx);
719 	ktrace_exit(td);
720 }
721 
722 void
ktrnamei(const char * path)723 ktrnamei(const char *path)
724 {
725 	struct ktr_request *req;
726 	int namelen;
727 	char *buf = NULL;
728 
729 	namelen = strlen(path);
730 	if (namelen > 0) {
731 		buf = malloc(namelen, M_KTRACE, M_WAITOK);
732 		bcopy(path, buf, namelen);
733 	}
734 	req = ktr_getrequest(KTR_NAMEI);
735 	if (req == NULL) {
736 		if (buf != NULL)
737 			free(buf, M_KTRACE);
738 		return;
739 	}
740 	if (namelen > 0) {
741 		req->ktr_header.ktr_len = namelen;
742 		req->ktr_buffer = buf;
743 	}
744 	ktr_submitrequest(curthread, req);
745 }
746 
747 void
ktrsysctl(int * name,u_int namelen)748 ktrsysctl(int *name, u_int namelen)
749 {
750 	struct ktr_request *req;
751 	u_int mib[CTL_MAXNAME + 2];
752 	char *mibname;
753 	size_t mibnamelen;
754 	int error;
755 
756 	/* Lookup name of mib. */
757 	KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long"));
758 	mib[0] = 0;
759 	mib[1] = 1;
760 	bcopy(name, mib + 2, namelen * sizeof(*name));
761 	mibnamelen = 128;
762 	mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK);
763 	error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen,
764 	    NULL, 0, &mibnamelen, 0);
765 	if (error) {
766 		free(mibname, M_KTRACE);
767 		return;
768 	}
769 	req = ktr_getrequest(KTR_SYSCTL);
770 	if (req == NULL) {
771 		free(mibname, M_KTRACE);
772 		return;
773 	}
774 	req->ktr_header.ktr_len = mibnamelen;
775 	req->ktr_buffer = mibname;
776 	ktr_submitrequest(curthread, req);
777 }
778 
779 void
ktrgenio(int fd,enum uio_rw rw,struct uio * uio,int error)780 ktrgenio(int fd, enum uio_rw rw, struct uio *uio, int error)
781 {
782 	struct ktr_request *req;
783 	struct ktr_genio *ktg;
784 	int datalen;
785 	char *buf;
786 
787 	if (error != 0 && (rw == UIO_READ || error == EFAULT)) {
788 		freeuio(uio);
789 		return;
790 	}
791 	uio->uio_offset = 0;
792 	uio->uio_rw = UIO_WRITE;
793 	datalen = MIN(uio->uio_resid, ktr_geniosize);
794 	buf = malloc(datalen, M_KTRACE, M_WAITOK);
795 	error = uiomove(buf, datalen, uio);
796 	freeuio(uio);
797 	if (error) {
798 		free(buf, M_KTRACE);
799 		return;
800 	}
801 	req = ktr_getrequest(KTR_GENIO);
802 	if (req == NULL) {
803 		free(buf, M_KTRACE);
804 		return;
805 	}
806 	ktg = &req->ktr_data.ktr_genio;
807 	ktg->ktr_fd = fd;
808 	ktg->ktr_rw = rw;
809 	req->ktr_header.ktr_len = datalen;
810 	req->ktr_buffer = buf;
811 	ktr_submitrequest(curthread, req);
812 }
813 
814 void
ktrpsig(int sig,sig_t action,sigset_t * mask,int code)815 ktrpsig(int sig, sig_t action, sigset_t *mask, int code)
816 {
817 	struct thread *td = curthread;
818 	struct ktr_request *req;
819 	struct ktr_psig	*kp;
820 
821 	req = ktr_getrequest(KTR_PSIG);
822 	if (req == NULL)
823 		return;
824 	kp = &req->ktr_data.ktr_psig;
825 	kp->signo = (char)sig;
826 	kp->action = action;
827 	kp->mask = *mask;
828 	kp->code = code;
829 	ktr_enqueuerequest(td, req);
830 	ktrace_exit(td);
831 }
832 
833 void
ktrcsw(int out,int user,const char * wmesg)834 ktrcsw(int out, int user, const char *wmesg)
835 {
836 	struct thread *td = curthread;
837 	struct ktr_request *req;
838 	struct ktr_csw *kc;
839 
840 	if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
841 		return;
842 
843 	req = ktr_getrequest(KTR_CSW);
844 	if (req == NULL)
845 		return;
846 	kc = &req->ktr_data.ktr_csw;
847 	kc->out = out;
848 	kc->user = user;
849 	if (wmesg != NULL)
850 		strlcpy(kc->wmesg, wmesg, sizeof(kc->wmesg));
851 	else
852 		bzero(kc->wmesg, sizeof(kc->wmesg));
853 	ktr_enqueuerequest(td, req);
854 	ktrace_exit(td);
855 }
856 
857 void
ktrstruct(const char * name,const void * data,size_t datalen)858 ktrstruct(const char *name, const void *data, size_t datalen)
859 {
860 	struct ktr_request *req;
861 	char *buf;
862 	size_t buflen, namelen;
863 
864 	if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
865 		return;
866 
867 	if (data == NULL)
868 		datalen = 0;
869 	namelen = strlen(name) + 1;
870 	buflen = namelen + datalen;
871 	buf = malloc(buflen, M_KTRACE, M_WAITOK);
872 	strcpy(buf, name);
873 	bcopy(data, buf + namelen, datalen);
874 	if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) {
875 		free(buf, M_KTRACE);
876 		return;
877 	}
878 	req->ktr_buffer = buf;
879 	req->ktr_header.ktr_len = buflen;
880 	ktr_submitrequest(curthread, req);
881 }
882 
883 void
ktrstruct_error(const char * name,const void * data,size_t datalen,int error)884 ktrstruct_error(const char *name, const void *data, size_t datalen, int error)
885 {
886 
887 	if (error == 0)
888 		ktrstruct(name, data, datalen);
889 }
890 
891 void
ktrstructarray(const char * name,enum uio_seg seg,const void * data,int num_items,size_t struct_size)892 ktrstructarray(const char *name, enum uio_seg seg, const void *data,
893     int num_items, size_t struct_size)
894 {
895 	struct ktr_request *req;
896 	struct ktr_struct_array *ksa;
897 	char *buf;
898 	size_t buflen, datalen, namelen;
899 	int max_items;
900 
901 	if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
902 		return;
903 	if (num_items < 0)
904 		return;
905 
906 	/* Trim array length to genio size. */
907 	max_items = ktr_geniosize / struct_size;
908 	if (num_items > max_items) {
909 		if (max_items == 0)
910 			num_items = 1;
911 		else
912 			num_items = max_items;
913 	}
914 	datalen = num_items * struct_size;
915 
916 	if (data == NULL)
917 		datalen = 0;
918 
919 	namelen = strlen(name) + 1;
920 	buflen = namelen + datalen;
921 	buf = malloc(buflen, M_KTRACE, M_WAITOK);
922 	strcpy(buf, name);
923 	if (seg == UIO_SYSSPACE)
924 		bcopy(data, buf + namelen, datalen);
925 	else {
926 		if (copyin(data, buf + namelen, datalen) != 0) {
927 			free(buf, M_KTRACE);
928 			return;
929 		}
930 	}
931 	if ((req = ktr_getrequest(KTR_STRUCT_ARRAY)) == NULL) {
932 		free(buf, M_KTRACE);
933 		return;
934 	}
935 	ksa = &req->ktr_data.ktr_struct_array;
936 	ksa->struct_size = struct_size;
937 	req->ktr_buffer = buf;
938 	req->ktr_header.ktr_len = buflen;
939 	ktr_submitrequest(curthread, req);
940 }
941 
942 void
ktrcapfail(enum ktr_cap_violation type,const void * data)943 ktrcapfail(enum ktr_cap_violation type, const void *data)
944 {
945 	struct thread *td = curthread;
946 	struct ktr_request *req;
947 	struct ktr_cap_fail *kcf;
948 	union ktr_cap_data *kcd;
949 
950 	if (__predict_false(td->td_pflags & TDP_INKTRACE))
951 		return;
952 	if (type != CAPFAIL_SYSCALL &&
953 	    (td->td_sa.callp->sy_flags & SYF_CAPENABLED) == 0)
954 		return;
955 
956 	req = ktr_getrequest(KTR_CAPFAIL);
957 	if (req == NULL)
958 		return;
959 	kcf = &req->ktr_data.ktr_cap_fail;
960 	kcf->cap_type = type;
961 	kcf->cap_code = td->td_sa.code;
962 	kcf->cap_svflags = td->td_proc->p_sysent->sv_flags;
963 	if (data != NULL) {
964 		kcd = &kcf->cap_data;
965 		switch (type) {
966 		case CAPFAIL_NOTCAPABLE:
967 		case CAPFAIL_INCREASE:
968 			kcd->cap_needed = *(const cap_rights_t *)data;
969 			kcd->cap_held = *((const cap_rights_t *)data + 1);
970 			break;
971 		case CAPFAIL_SYSCALL:
972 		case CAPFAIL_SIGNAL:
973 		case CAPFAIL_PROTO:
974 			kcd->cap_int = *(const int *)data;
975 			break;
976 		case CAPFAIL_SOCKADDR:
977 			kcd->cap_sockaddr = *(const struct sockaddr *)data;
978 			break;
979 		case CAPFAIL_NAMEI:
980 			strlcpy(kcd->cap_path, data, MAXPATHLEN);
981 			break;
982 		case CAPFAIL_CPUSET:
983 		default:
984 			break;
985 		}
986 	}
987 	ktr_enqueuerequest(td, req);
988 	ktrace_exit(td);
989 }
990 
991 void
ktrfault(vm_offset_t vaddr,int type)992 ktrfault(vm_offset_t vaddr, int type)
993 {
994 	struct thread *td = curthread;
995 	struct ktr_request *req;
996 	struct ktr_fault *kf;
997 
998 	if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
999 		return;
1000 
1001 	req = ktr_getrequest(KTR_FAULT);
1002 	if (req == NULL)
1003 		return;
1004 	kf = &req->ktr_data.ktr_fault;
1005 	kf->vaddr = vaddr;
1006 	kf->type = type;
1007 	ktr_enqueuerequest(td, req);
1008 	ktrace_exit(td);
1009 }
1010 
1011 void
ktrfaultend(int result)1012 ktrfaultend(int result)
1013 {
1014 	struct thread *td = curthread;
1015 	struct ktr_request *req;
1016 	struct ktr_faultend *kf;
1017 
1018 	if (__predict_false(curthread->td_pflags & TDP_INKTRACE))
1019 		return;
1020 
1021 	req = ktr_getrequest(KTR_FAULTEND);
1022 	if (req == NULL)
1023 		return;
1024 	kf = &req->ktr_data.ktr_faultend;
1025 	kf->result = result;
1026 	ktr_enqueuerequest(td, req);
1027 	ktrace_exit(td);
1028 }
1029 #endif /* KTRACE */
1030 
1031 /* Interface and common routines */
1032 
1033 #ifndef _SYS_SYSPROTO_H_
1034 struct ktrace_args {
1035 	char	*fname;
1036 	int	ops;
1037 	int	facs;
1038 	int	pid;
1039 };
1040 #endif
1041 /* ARGSUSED */
1042 int
sys_ktrace(struct thread * td,struct ktrace_args * uap)1043 sys_ktrace(struct thread *td, struct ktrace_args *uap)
1044 {
1045 #ifdef KTRACE
1046 	struct vnode *vp = NULL;
1047 	struct proc *p;
1048 	struct pgrp *pg;
1049 	int facs = uap->facs & ~KTRFAC_ROOT;
1050 	int ops = KTROP(uap->ops);
1051 	int descend = uap->ops & KTRFLAG_DESCEND;
1052 	int ret = 0;
1053 	int flags, error = 0;
1054 	struct nameidata nd;
1055 	struct ktr_io_params *kiop, *old_kiop;
1056 
1057 	/*
1058 	 * Need something to (un)trace.
1059 	 */
1060 	if (ops != KTROP_CLEARFILE && facs == 0)
1061 		return (EINVAL);
1062 
1063 	kiop = NULL;
1064 	if (ops != KTROP_CLEAR) {
1065 		/*
1066 		 * an operation which requires a file argument.
1067 		 */
1068 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname);
1069 		flags = FREAD | FWRITE | O_NOFOLLOW;
1070 		error = vn_open(&nd, &flags, 0, NULL);
1071 		if (error)
1072 			return (error);
1073 		NDFREE_PNBUF(&nd);
1074 		vp = nd.ni_vp;
1075 		VOP_UNLOCK(vp);
1076 		if (vp->v_type != VREG) {
1077 			(void)vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
1078 			return (EACCES);
1079 		}
1080 		kiop = ktr_io_params_alloc(td, vp);
1081 	}
1082 
1083 	/*
1084 	 * Clear all uses of the tracefile.
1085 	 */
1086 	ktrace_enter(td);
1087 	if (ops == KTROP_CLEARFILE) {
1088 restart:
1089 		sx_slock(&allproc_lock);
1090 		FOREACH_PROC_IN_SYSTEM(p) {
1091 			old_kiop = NULL;
1092 			PROC_LOCK(p);
1093 			if (p->p_ktrioparms != NULL &&
1094 			    p->p_ktrioparms->vp == vp) {
1095 				if (ktrcanset(td, p)) {
1096 					mtx_lock(&ktrace_mtx);
1097 					old_kiop = ktr_freeproc(p);
1098 					mtx_unlock(&ktrace_mtx);
1099 				} else
1100 					error = EPERM;
1101 			}
1102 			PROC_UNLOCK(p);
1103 			if (old_kiop != NULL) {
1104 				sx_sunlock(&allproc_lock);
1105 				ktr_io_params_free(old_kiop);
1106 				goto restart;
1107 			}
1108 		}
1109 		sx_sunlock(&allproc_lock);
1110 		goto done;
1111 	}
1112 	/*
1113 	 * do it
1114 	 */
1115 	sx_slock(&proctree_lock);
1116 	if (uap->pid < 0) {
1117 		/*
1118 		 * by process group
1119 		 */
1120 		pg = pgfind(-uap->pid);
1121 		if (pg == NULL) {
1122 			sx_sunlock(&proctree_lock);
1123 			error = ESRCH;
1124 			goto done;
1125 		}
1126 
1127 		/*
1128 		 * ktrops() may call vrele(). Lock pg_members
1129 		 * by the proctree_lock rather than pg_mtx.
1130 		 */
1131 		PGRP_UNLOCK(pg);
1132 		if (LIST_EMPTY(&pg->pg_members)) {
1133 			sx_sunlock(&proctree_lock);
1134 			error = ESRCH;
1135 			goto done;
1136 		}
1137 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1138 			PROC_LOCK(p);
1139 			if (descend)
1140 				ret |= ktrsetchildren(td, p, ops, facs, kiop);
1141 			else
1142 				ret |= ktrops(td, p, ops, facs, kiop);
1143 		}
1144 	} else {
1145 		/*
1146 		 * by pid
1147 		 */
1148 		p = pfind(uap->pid);
1149 		if (p == NULL) {
1150 			error = ESRCH;
1151 			sx_sunlock(&proctree_lock);
1152 			goto done;
1153 		}
1154 		if (descend)
1155 			ret |= ktrsetchildren(td, p, ops, facs, kiop);
1156 		else
1157 			ret |= ktrops(td, p, ops, facs, kiop);
1158 	}
1159 	sx_sunlock(&proctree_lock);
1160 	if (!ret)
1161 		error = EPERM;
1162 done:
1163 	if (kiop != NULL) {
1164 		mtx_lock(&ktrace_mtx);
1165 		kiop = ktr_io_params_rele(kiop);
1166 		mtx_unlock(&ktrace_mtx);
1167 		ktr_io_params_free(kiop);
1168 	}
1169 	ktrace_exit(td);
1170 	return (error);
1171 #else /* !KTRACE */
1172 	return (ENOSYS);
1173 #endif /* KTRACE */
1174 }
1175 
1176 /* ARGSUSED */
1177 int
sys_utrace(struct thread * td,struct utrace_args * uap)1178 sys_utrace(struct thread *td, struct utrace_args *uap)
1179 {
1180 
1181 #ifdef KTRACE
1182 	struct ktr_request *req;
1183 	void *cp;
1184 	int error;
1185 
1186 	if (!KTRPOINT(td, KTR_USER))
1187 		return (0);
1188 	if (uap->len > KTR_USER_MAXLEN)
1189 		return (EINVAL);
1190 	cp = malloc(uap->len, M_KTRACE, M_WAITOK);
1191 	error = copyin(uap->addr, cp, uap->len);
1192 	if (error) {
1193 		free(cp, M_KTRACE);
1194 		return (error);
1195 	}
1196 	req = ktr_getrequest(KTR_USER);
1197 	if (req == NULL) {
1198 		free(cp, M_KTRACE);
1199 		return (ENOMEM);
1200 	}
1201 	req->ktr_buffer = cp;
1202 	req->ktr_header.ktr_len = uap->len;
1203 	ktr_submitrequest(td, req);
1204 	return (0);
1205 #else /* !KTRACE */
1206 	return (ENOSYS);
1207 #endif /* KTRACE */
1208 }
1209 
1210 #ifdef KTRACE
1211 static int
ktrops(struct thread * td,struct proc * p,int ops,int facs,struct ktr_io_params * new_kiop)1212 ktrops(struct thread *td, struct proc *p, int ops, int facs,
1213     struct ktr_io_params *new_kiop)
1214 {
1215 	struct ktr_io_params *old_kiop;
1216 
1217 	PROC_LOCK_ASSERT(p, MA_OWNED);
1218 	if (!ktrcanset(td, p)) {
1219 		PROC_UNLOCK(p);
1220 		return (0);
1221 	}
1222 	if ((ops == KTROP_SET && p->p_state == PRS_NEW) ||
1223 	    p_cansee(td, p) != 0) {
1224 		/*
1225 		 * Disallow setting trace points if the process is being born.
1226 		 * This avoids races with trace point inheritance in
1227 		 * ktrprocfork().
1228 		 */
1229 		PROC_UNLOCK(p);
1230 		return (0);
1231 	}
1232 	if ((p->p_flag & P_WEXIT) != 0) {
1233 		/*
1234 		 * There's nothing to do if the process is exiting, but avoid
1235 		 * signaling an error.
1236 		 */
1237 		PROC_UNLOCK(p);
1238 		return (1);
1239 	}
1240 	old_kiop = NULL;
1241 	mtx_lock(&ktrace_mtx);
1242 	if (ops == KTROP_SET) {
1243 		if (p->p_ktrioparms != NULL &&
1244 		    p->p_ktrioparms->vp != new_kiop->vp) {
1245 			/* if trace file already in use, relinquish below */
1246 			old_kiop = ktr_io_params_rele(p->p_ktrioparms);
1247 			p->p_ktrioparms = NULL;
1248 		}
1249 		if (p->p_ktrioparms == NULL) {
1250 			p->p_ktrioparms = new_kiop;
1251 			ktr_io_params_ref(new_kiop);
1252 		}
1253 		p->p_traceflag |= facs;
1254 		if (priv_check(td, PRIV_KTRACE) == 0)
1255 			p->p_traceflag |= KTRFAC_ROOT;
1256 	} else {
1257 		/* KTROP_CLEAR */
1258 		if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0)
1259 			/* no more tracing */
1260 			old_kiop = ktr_freeproc(p);
1261 	}
1262 	mtx_unlock(&ktrace_mtx);
1263 	if ((p->p_traceflag & KTRFAC_MASK) != 0)
1264 		ktrprocctor_entered(td, p);
1265 	PROC_UNLOCK(p);
1266 	ktr_io_params_free(old_kiop);
1267 
1268 	return (1);
1269 }
1270 
1271 static int
ktrsetchildren(struct thread * td,struct proc * top,int ops,int facs,struct ktr_io_params * new_kiop)1272 ktrsetchildren(struct thread *td, struct proc *top, int ops, int facs,
1273     struct ktr_io_params *new_kiop)
1274 {
1275 	struct proc *p;
1276 	int ret = 0;
1277 
1278 	p = top;
1279 	PROC_LOCK_ASSERT(p, MA_OWNED);
1280 	sx_assert(&proctree_lock, SX_LOCKED);
1281 	for (;;) {
1282 		ret |= ktrops(td, p, ops, facs, new_kiop);
1283 		/*
1284 		 * If this process has children, descend to them next,
1285 		 * otherwise do any siblings, and if done with this level,
1286 		 * follow back up the tree (but not past top).
1287 		 */
1288 		if (!LIST_EMPTY(&p->p_children))
1289 			p = LIST_FIRST(&p->p_children);
1290 		else for (;;) {
1291 			if (p == top)
1292 				return (ret);
1293 			if (LIST_NEXT(p, p_sibling)) {
1294 				p = LIST_NEXT(p, p_sibling);
1295 				break;
1296 			}
1297 			p = p->p_pptr;
1298 		}
1299 		PROC_LOCK(p);
1300 	}
1301 	/*NOTREACHED*/
1302 }
1303 
1304 static void
ktr_writerequest(struct thread * td,struct ktr_request * req)1305 ktr_writerequest(struct thread *td, struct ktr_request *req)
1306 {
1307 	struct ktr_io_params *kiop, *kiop1;
1308 	struct ktr_header *kth;
1309 	struct vnode *vp;
1310 	struct proc *p;
1311 	struct ucred *cred;
1312 	struct uio auio;
1313 	struct iovec aiov[3];
1314 	struct mount *mp;
1315 	off_t lim;
1316 	int datalen, buflen;
1317 	int error;
1318 
1319 	p = td->td_proc;
1320 
1321 	/*
1322 	 * We reference the kiop for use in I/O in case ktrace is
1323 	 * disabled on the process as we write out the request.
1324 	 */
1325 	mtx_lock(&ktrace_mtx);
1326 	kiop = p->p_ktrioparms;
1327 
1328 	/*
1329 	 * If kiop is NULL, it has been cleared out from under this
1330 	 * request, so just drop it.
1331 	 */
1332 	if (kiop == NULL) {
1333 		mtx_unlock(&ktrace_mtx);
1334 		return;
1335 	}
1336 
1337 	ktr_io_params_ref(kiop);
1338 	vp = kiop->vp;
1339 	cred = kiop->cr;
1340 	lim = kiop->lim;
1341 
1342 	KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL"));
1343 	mtx_unlock(&ktrace_mtx);
1344 
1345 	kth = &req->ktr_header;
1346 	KASSERT(((u_short)kth->ktr_type & ~KTR_TYPE) < nitems(data_lengths),
1347 	    ("data_lengths array overflow"));
1348 	datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_TYPE];
1349 	buflen = kth->ktr_len;
1350 	auio.uio_iov = &aiov[0];
1351 	auio.uio_offset = 0;
1352 	auio.uio_segflg = UIO_SYSSPACE;
1353 	auio.uio_rw = UIO_WRITE;
1354 	aiov[0].iov_base = (caddr_t)kth;
1355 	aiov[0].iov_len = sizeof(struct ktr_header);
1356 	auio.uio_resid = sizeof(struct ktr_header);
1357 	auio.uio_iovcnt = 1;
1358 	auio.uio_td = td;
1359 	if (datalen != 0) {
1360 		aiov[1].iov_base = (caddr_t)&req->ktr_data;
1361 		aiov[1].iov_len = datalen;
1362 		auio.uio_resid += datalen;
1363 		auio.uio_iovcnt++;
1364 		kth->ktr_len += datalen;
1365 	}
1366 	if (buflen != 0) {
1367 		KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write"));
1368 		aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer;
1369 		aiov[auio.uio_iovcnt].iov_len = buflen;
1370 		auio.uio_resid += buflen;
1371 		auio.uio_iovcnt++;
1372 	}
1373 
1374 	vn_start_write(vp, &mp, V_WAIT);
1375 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1376 	td->td_ktr_io_lim = lim;
1377 #ifdef MAC
1378 	error = mac_vnode_check_write(cred, NOCRED, vp);
1379 	if (error == 0)
1380 #endif
1381 		error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred);
1382 	VOP_UNLOCK(vp);
1383 	vn_finished_write(mp);
1384 	if (error == 0) {
1385 		mtx_lock(&ktrace_mtx);
1386 		kiop = ktr_io_params_rele(kiop);
1387 		mtx_unlock(&ktrace_mtx);
1388 		ktr_io_params_free(kiop);
1389 		return;
1390 	}
1391 
1392 	/*
1393 	 * If error encountered, give up tracing on this vnode on this
1394 	 * process.  Other processes might still be suitable for
1395 	 * writes to this vnode.
1396 	 */
1397 	log(LOG_NOTICE,
1398 	    "ktrace write failed, errno %d, tracing stopped for pid %d\n",
1399 	    error, p->p_pid);
1400 
1401 	kiop1 = NULL;
1402 	PROC_LOCK(p);
1403 	mtx_lock(&ktrace_mtx);
1404 	if (p->p_ktrioparms != NULL && p->p_ktrioparms->vp == vp)
1405 		kiop1 = ktr_freeproc(p);
1406 	kiop = ktr_io_params_rele(kiop);
1407 	mtx_unlock(&ktrace_mtx);
1408 	PROC_UNLOCK(p);
1409 	ktr_io_params_free(kiop1);
1410 	ktr_io_params_free(kiop);
1411 }
1412 
1413 /*
1414  * Return true if caller has permission to set the ktracing state
1415  * of target.  Essentially, the target can't possess any
1416  * more permissions than the caller.  KTRFAC_ROOT signifies that
1417  * root previously set the tracing status on the target process, and
1418  * so, only root may further change it.
1419  */
1420 static int
ktrcanset(struct thread * td,struct proc * targetp)1421 ktrcanset(struct thread *td, struct proc *targetp)
1422 {
1423 
1424 	PROC_LOCK_ASSERT(targetp, MA_OWNED);
1425 	if (targetp->p_traceflag & KTRFAC_ROOT &&
1426 	    priv_check(td, PRIV_KTRACE))
1427 		return (0);
1428 
1429 	if (p_candebug(td, targetp) != 0)
1430 		return (0);
1431 
1432 	return (1);
1433 }
1434 
1435 #endif /* KTRACE */
1436