xref: /netbsd-src/sys/kern/kern_ktrace.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: kern_ktrace.c,v 1.86 2004/01/16 05:03:02 mrg Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)kern_ktrace.c	8.5 (Berkeley) 5/14/95
32  */
33 
34 #include <sys/cdefs.h>
35 __KERNEL_RCSID(0, "$NetBSD: kern_ktrace.c,v 1.86 2004/01/16 05:03:02 mrg Exp $");
36 
37 #include "opt_ktrace.h"
38 #include "opt_compat_mach.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/proc.h>
43 #include <sys/file.h>
44 #include <sys/namei.h>
45 #include <sys/vnode.h>
46 #include <sys/ktrace.h>
47 #include <sys/malloc.h>
48 #include <sys/syslog.h>
49 #include <sys/filedesc.h>
50 #include <sys/ioctl.h>
51 
52 #include <sys/mount.h>
53 #include <sys/sa.h>
54 #include <sys/syscallargs.h>
55 
56 #ifdef KTRACE
57 
58 int	ktrace_common(struct proc *, int, int, int, struct file *);
59 int	ktrops(struct proc *, struct proc *, int, int, struct file *);
60 int	ktrsetchildren(struct proc *, struct proc *, int, int,
61 	    struct file *);
62 int	ktrcanset(struct proc *, struct proc *);
63 int	ktrsamefile(struct file *, struct file *);
64 
65 /*
66  * "deep" compare of two files for the purposes of clearing a trace.
67  * Returns true if they're the same open file, or if they point at the
68  * same underlying vnode/socket.
69  */
70 
71 int
72 ktrsamefile(f1, f2)
73 	struct file *f1;
74 	struct file *f2;
75 {
76 	return ((f1 == f2) ||
77 	    ((f1 != NULL) && (f2 != NULL) &&
78 		(f1->f_type == f2->f_type) &&
79 		(f1->f_data == f2->f_data)));
80 }
81 
82 void
83 ktrderef(p)
84 	struct proc *p;
85 {
86 	struct file *fp = p->p_tracep;
87 	p->p_traceflag = 0;
88 	if (fp == NULL)
89 		return;
90 	p->p_tracep = NULL;
91 
92 	simple_lock(&fp->f_slock);
93 	FILE_USE(fp);
94 
95 	/*
96 	 * ktrace file descriptor can't be watched (are not visible to
97 	 * userspace), so no kqueue stuff here
98 	 */
99 	closef(fp, NULL);
100 }
101 
102 void
103 ktradref(p)
104 	struct proc *p;
105 {
106 	struct file *fp = p->p_tracep;
107 
108 	fp->f_count++;
109 }
110 
111 void
112 ktrinitheader(kth, p, type)
113 	struct ktr_header *kth;
114 	struct proc *p;
115 	int type;
116 {
117 
118 	memset(kth, 0, sizeof(*kth));
119 	kth->ktr_type = type;
120 	microtime(&kth->ktr_time);
121 	kth->ktr_pid = p->p_pid;
122 	memcpy(kth->ktr_comm, p->p_comm, MAXCOMLEN);
123 }
124 
125 void
126 ktrsyscall(p, code, realcode, callp, args)
127 	struct proc *p;
128 	register_t code;
129 	register_t realcode;
130 	const struct sysent *callp;
131 	register_t args[];
132 {
133 	struct ktr_header kth;
134 	struct ktr_syscall *ktp;
135 	register_t *argp;
136 	int argsize;
137 	size_t len;
138 	u_int i;
139 
140 	if (callp == NULL)
141 		callp = p->p_emul->e_sysent;
142 
143 	argsize = callp[code].sy_argsize;
144 #ifdef _LP64
145 	if (p->p_flag & P_32)
146 		argsize = argsize << 1;
147 #endif
148 	len = sizeof(struct ktr_syscall) + argsize;
149 
150 	p->p_traceflag |= KTRFAC_ACTIVE;
151 	ktrinitheader(&kth, p, KTR_SYSCALL);
152 	ktp = malloc(len, M_TEMP, M_WAITOK);
153 	ktp->ktr_code = realcode;
154 	ktp->ktr_argsize = argsize;
155 	argp = (register_t *)((char *)ktp + sizeof(struct ktr_syscall));
156 	for (i = 0; i < (argsize / sizeof(*argp)); i++)
157 		*argp++ = args[i];
158 	kth.ktr_buf = (caddr_t)ktp;
159 	kth.ktr_len = len;
160 	(void) ktrwrite(p, &kth);
161 	free(ktp, M_TEMP);
162 	p->p_traceflag &= ~KTRFAC_ACTIVE;
163 }
164 
165 void
166 ktrsysret(p, code, error, retval)
167 	struct proc *p;
168 	register_t code;
169 	int error;
170 	register_t *retval;
171 {
172 	struct ktr_header kth;
173 	struct ktr_sysret ktp;
174 
175 	p->p_traceflag |= KTRFAC_ACTIVE;
176 	ktrinitheader(&kth, p, KTR_SYSRET);
177 	ktp.ktr_code = code;
178 	ktp.ktr_eosys = 0;			/* XXX unused */
179 	ktp.ktr_error = error;
180 	ktp.ktr_retval = retval ? retval[0] : 0;
181 	ktp.ktr_retval_1 = retval ? retval[1] : 0;
182 
183 	kth.ktr_buf = (caddr_t)&ktp;
184 	kth.ktr_len = sizeof(struct ktr_sysret);
185 
186 	(void) ktrwrite(p, &kth);
187 	p->p_traceflag &= ~KTRFAC_ACTIVE;
188 }
189 
190 void
191 ktrnamei(p, path)
192 	struct proc *p;
193 	char *path;
194 {
195 	struct ktr_header kth;
196 
197 	p->p_traceflag |= KTRFAC_ACTIVE;
198 	ktrinitheader(&kth, p, KTR_NAMEI);
199 	kth.ktr_len = strlen(path);
200 	kth.ktr_buf = path;
201 
202 	(void) ktrwrite(p, &kth);
203 	p->p_traceflag &= ~KTRFAC_ACTIVE;
204 }
205 
206 void
207 ktremul(p)
208 	struct proc *p;
209 {
210 	struct ktr_header kth;
211 	const char *emul = p->p_emul->e_name;
212 
213 	p->p_traceflag |= KTRFAC_ACTIVE;
214 	ktrinitheader(&kth, p, KTR_EMUL);
215 	kth.ktr_len = strlen(emul);
216 	kth.ktr_buf = (caddr_t)emul;
217 
218 	(void) ktrwrite(p, &kth);
219 	p->p_traceflag &= ~KTRFAC_ACTIVE;
220 }
221 
222 void
223 ktrkmem(struct proc *p, int ktr, const void *buf, size_t len)
224 {
225 	struct ktr_header kth;
226 
227 	p->p_traceflag |= KTRFAC_ACTIVE;
228 	ktrinitheader(&kth, p, ktr);
229 	kth.ktr_len = len;
230 	kth.ktr_buf = buf;
231 
232 	(void)ktrwrite(p, &kth);
233 	p->p_traceflag &= ~KTRFAC_ACTIVE;
234 }
235 
236 void
237 ktrgenio(p, fd, rw, iov, len, error)
238 	struct proc *p;
239 	int fd;
240 	enum uio_rw rw;
241 	struct iovec *iov;
242 	int len;
243 	int error;
244 {
245 	struct ktr_header kth;
246 	struct ktr_genio *ktp;
247 	caddr_t cp;
248 	int resid = len, cnt;
249 	int buflen;
250 
251 	if (error)
252 		return;
253 
254 	p->p_traceflag |= KTRFAC_ACTIVE;
255 
256 	buflen = min(PAGE_SIZE, len + sizeof(struct ktr_genio));
257 
258 	ktrinitheader(&kth, p, KTR_GENIO);
259 	ktp = malloc(buflen, M_TEMP, M_WAITOK);
260 	ktp->ktr_fd = fd;
261 	ktp->ktr_rw = rw;
262 
263 	kth.ktr_buf = (caddr_t)ktp;
264 
265 	cp = (caddr_t)((char *)ktp + sizeof(struct ktr_genio));
266 	buflen -= sizeof(struct ktr_genio);
267 
268 	while (resid > 0) {
269 #if 0 /* XXX NJWLWP */
270 		KDASSERT(p->p_cpu != NULL);
271 		KDASSERT(p->p_cpu == curcpu());
272 #endif
273 		/* XXX NJWLWP */
274 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
275 			preempt(1);
276 
277 		cnt = min(iov->iov_len, buflen);
278 		if (cnt > resid)
279 			cnt = resid;
280 		if (copyin(iov->iov_base, cp, cnt))
281 			break;
282 
283 		kth.ktr_len = cnt + sizeof(struct ktr_genio);
284 
285 		if (__predict_false(ktrwrite(p, &kth) != 0))
286 			break;
287 
288 		iov->iov_base = (caddr_t)iov->iov_base + cnt;
289 		iov->iov_len -= cnt;
290 
291 		if (iov->iov_len == 0)
292 			iov++;
293 
294 		resid -= cnt;
295 	}
296 
297 	free(ktp, M_TEMP);
298 	p->p_traceflag &= ~KTRFAC_ACTIVE;
299 }
300 
301 void
302 ktrpsig(p, sig, action, mask, ksi)
303 	struct proc *p;
304 	int sig;
305 	sig_t action;
306 	const sigset_t *mask;
307 	const ksiginfo_t *ksi;
308 {
309 	struct ktr_header kth;
310 	struct {
311 		struct ktr_psig	kp;
312 		siginfo_t	si;
313 	} kbuf;
314 
315 	p->p_traceflag |= KTRFAC_ACTIVE;
316 	ktrinitheader(&kth, p, KTR_PSIG);
317 	kbuf.kp.signo = (char)sig;
318 	kbuf.kp.action = action;
319 	kbuf.kp.mask = *mask;
320 	kth.ktr_buf = (caddr_t)&kbuf;
321 	if (ksi) {
322 		kbuf.kp.code = KSI_TRAPCODE(ksi);
323 		(void)memset(&kbuf.si, 0, sizeof(kbuf.si));
324 		kbuf.si._info = ksi->ksi_info;
325 		kth.ktr_len = sizeof(kbuf);
326 	} else {
327 		kbuf.kp.code = 0;
328 		kth.ktr_len = sizeof(struct ktr_psig);
329 	}
330 	(void) ktrwrite(p, &kth);
331 	p->p_traceflag &= ~KTRFAC_ACTIVE;
332 }
333 
334 void
335 ktrcsw(p, out, user)
336 	struct proc *p;
337 	int out;
338 	int user;
339 {
340 	struct ktr_header kth;
341 	struct ktr_csw kc;
342 
343 	p->p_traceflag |= KTRFAC_ACTIVE;
344 	ktrinitheader(&kth, p, KTR_CSW);
345 	kc.out = out;
346 	kc.user = user;
347 	kth.ktr_buf = (caddr_t)&kc;
348 	kth.ktr_len = sizeof(struct ktr_csw);
349 
350 	(void) ktrwrite(p, &kth);
351 	p->p_traceflag &= ~KTRFAC_ACTIVE;
352 }
353 
354 void
355 ktruser(p, id, addr, len, ustr)
356 	struct proc *p;
357 	const char *id;
358 	void *addr;
359 	size_t len;
360 	int ustr;
361 {
362 	struct ktr_header kth;
363 	struct ktr_user *ktp;
364 	caddr_t user_dta;
365 
366 	p->p_traceflag |= KTRFAC_ACTIVE;
367 	ktrinitheader(&kth, p, KTR_USER);
368 	ktp = malloc(sizeof(struct ktr_user) + len, M_TEMP, M_WAITOK);
369 	if (ustr) {
370 		if (copyinstr(id, ktp->ktr_id, KTR_USER_MAXIDLEN, NULL) != 0)
371 			ktp->ktr_id[0] = '\0';
372 	} else
373 		strncpy(ktp->ktr_id, id, KTR_USER_MAXIDLEN);
374 	ktp->ktr_id[KTR_USER_MAXIDLEN-1] = '\0';
375 
376 	user_dta = (caddr_t) ((char *)ktp + sizeof(struct ktr_user));
377 	if (copyin(addr, (void *) user_dta, len) != 0)
378 		len = 0;
379 
380 	kth.ktr_buf = (void *)ktp;
381 	kth.ktr_len = sizeof(struct ktr_user) + len;
382 	(void) ktrwrite(p, &kth);
383 
384 	free(ktp, M_TEMP);
385 	p->p_traceflag &= ~KTRFAC_ACTIVE;
386 
387 }
388 
389 void
390 ktrmmsg(p, msgh, size)
391 	struct proc *p;
392 	const void *msgh;
393 	size_t size;
394 {
395 	struct ktr_header kth;
396 	struct ktr_mmsg	*kp;
397 
398 	p->p_traceflag |= KTRFAC_ACTIVE;
399 	ktrinitheader(&kth, p, KTR_MMSG);
400 
401 	kp = (struct ktr_mmsg *)msgh;
402 	kth.ktr_buf = (caddr_t)kp;
403 	kth.ktr_len = size;
404 	(void) ktrwrite(p, &kth);
405 	p->p_traceflag &= ~KTRFAC_ACTIVE;
406 }
407 
408 void
409 ktrmool(p, kaddr, size, uaddr)
410 	struct proc *p;
411 	const void *kaddr;
412 	size_t size;
413 	const void *uaddr;
414 {
415 	struct ktr_header kth;
416 	struct ktr_mool *kp;
417 	struct ktr_mool *buf;
418 
419 	p->p_traceflag |= KTRFAC_ACTIVE;
420 	ktrinitheader(&kth, p, KTR_MOOL);
421 
422 	kp = malloc(size + sizeof(*kp), M_TEMP, M_WAITOK);
423 	kp->uaddr = uaddr;
424 	kp->size = size;
425 	buf = kp + 1; /* Skip uaddr and size */
426 	memcpy(buf, kaddr, size);
427 
428 	kth.ktr_buf = (caddr_t)kp;
429 	kth.ktr_len = size + sizeof(*kp);
430 	(void) ktrwrite(p, &kth);
431 	free(kp, M_TEMP);
432 
433 	p->p_traceflag &= ~KTRFAC_ACTIVE;
434 }
435 
436 
437 /* Interface and common routines */
438 
439 int
440 ktrace_common(curp, ops, facs, pid, fp)
441 	struct proc *curp;
442 	int ops;
443 	int facs;
444 	int pid;
445 	struct file *fp;
446 {
447 	int ret = 0;
448 	int error = 0;
449 	int one = 1;
450 	int descend;
451 	struct proc *p;
452 	struct pgrp *pg;
453 
454 	curp->p_traceflag |= KTRFAC_ACTIVE;
455 	descend = ops & KTRFLAG_DESCEND;
456 	facs = facs & ~((unsigned) KTRFAC_ROOT);
457 
458 	/*
459 	 * Clear all uses of the tracefile
460 	 */
461 	if (KTROP(ops) == KTROP_CLEARFILE) {
462 		proclist_lock_read();
463 		LIST_FOREACH(p, &allproc, p_list) {
464 			if (ktrsamefile(p->p_tracep, fp)) {
465 				if (ktrcanset(curp, p))
466 					ktrderef(p);
467 				else
468 					error = EPERM;
469 			}
470 		}
471 		proclist_unlock_read();
472 		goto done;
473 	}
474 
475 	/*
476 	 * Mark fp non-blocking, to avoid problems from possible deadlocks.
477 	 */
478 
479 	if (fp != NULL) {
480 		fp->f_flag |= FNONBLOCK;
481 		(*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&one, curp);
482 	}
483 
484 	/*
485 	 * need something to (un)trace (XXX - why is this here?)
486 	 */
487 	if (!facs) {
488 		error = EINVAL;
489 		goto done;
490 	}
491 	/*
492 	 * do it
493 	 */
494 	if (pid < 0) {
495 		/*
496 		 * by process group
497 		 */
498 		pg = pg_find(-pid, PFIND_UNLOCK_FAIL);
499 		if (pg == NULL) {
500 			error = ESRCH;
501 			goto done;
502 		}
503 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
504 			if (descend)
505 				ret |= ktrsetchildren(curp, p, ops, facs, fp);
506 			else
507 				ret |= ktrops(curp, p, ops, facs, fp);
508 		}
509 
510 	} else {
511 		/*
512 		 * by pid
513 		 */
514 		p = p_find(pid, PFIND_UNLOCK_FAIL);
515 		if (p == NULL) {
516 			error = ESRCH;
517 			goto done;
518 		}
519 		if (descend)
520 			ret |= ktrsetchildren(curp, p, ops, facs, fp);
521 		else
522 			ret |= ktrops(curp, p, ops, facs, fp);
523 	}
524 	proclist_unlock_read();	/* taken by p{g}_find */
525 	if (!ret)
526 		error = EPERM;
527 done:
528 	curp->p_traceflag &= ~KTRFAC_ACTIVE;
529 	return (error);
530 }
531 
532 /*
533  * ktrace system call
534  */
535 /* ARGSUSED */
536 int
537 sys_fktrace(l, v, retval)
538 	struct lwp *l;
539 	void *v;
540 	register_t *retval;
541 {
542 	struct sys_fktrace_args /* {
543 		syscallarg(int) fd;
544 		syscallarg(int) ops;
545 		syscallarg(int) facs;
546 		syscallarg(int) pid;
547 	} */ *uap = v;
548 	struct proc *curp = l->l_proc;
549 	struct file *fp = NULL;
550 	struct filedesc *fdp = curp->p_fd;
551 	int error;
552 
553 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL)
554 		return (EBADF);
555 
556 	FILE_USE(fp);
557 
558 	if ((fp->f_flag & FWRITE) == 0)
559 		error = EBADF;
560 	else
561 		error = ktrace_common(curp, SCARG(uap, ops),
562 		    SCARG(uap, facs), SCARG(uap, pid), fp);
563 
564 	FILE_UNUSE(fp, curp);
565 
566 	return error;
567 }
568 
569 /*
570  * ktrace system call
571  */
572 /* ARGSUSED */
573 int
574 sys_ktrace(l, v, retval)
575 	struct lwp *l;
576 	void *v;
577 	register_t *retval;
578 {
579 	struct sys_ktrace_args /* {
580 		syscallarg(const char *) fname;
581 		syscallarg(int) ops;
582 		syscallarg(int) facs;
583 		syscallarg(int) pid;
584 	} */ *uap = v;
585 	struct proc *curp = l->l_proc;
586 	struct vnode *vp = NULL;
587 	struct file *fp = NULL;
588 	int fd;
589 	int ops = SCARG(uap, ops);
590 	int error = 0;
591 	struct nameidata nd;
592 
593 	ops = KTROP(ops) | (ops & KTRFLAG_DESCEND);
594 
595 	curp->p_traceflag |= KTRFAC_ACTIVE;
596 	if ((ops & KTROP_CLEAR) == 0) {
597 		/*
598 		 * an operation which requires a file argument.
599 		 */
600 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, SCARG(uap, fname),
601 		    curp);
602 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
603 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
604 			return (error);
605 		}
606 		vp = nd.ni_vp;
607 		VOP_UNLOCK(vp, 0);
608 		if (vp->v_type != VREG) {
609 			(void) vn_close(vp, FREAD|FWRITE, curp->p_ucred, curp);
610 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
611 			return (EACCES);
612 		}
613 		/*
614 		 * XXX This uses up a file descriptor slot in the
615 		 * tracing process for the duration of this syscall.
616 		 * This is not expected to be a problem.  If
617 		 * falloc(NULL, ...) DTRT we could skip that part, but
618 		 * that would require changing its interface to allow
619 		 * the caller to pass in a ucred..
620 		 *
621 		 * This will FILE_USE the fp it returns, if any.
622 		 * Keep it in use until we return.
623 		 */
624 		if ((error = falloc(curp, &fp, &fd)) != 0)
625 			goto done;
626 
627 		fp->f_flag = FWRITE|FAPPEND;
628 		fp->f_type = DTYPE_VNODE;
629 		fp->f_ops = &vnops;
630 		fp->f_data = (caddr_t)vp;
631 		FILE_SET_MATURE(fp);
632 		vp = NULL;
633 	}
634 	error = ktrace_common(curp, SCARG(uap, ops), SCARG(uap, facs),
635 	    SCARG(uap, pid), fp);
636 done:
637 	if (vp != NULL)
638 		(void) vn_close(vp, FWRITE, curp->p_ucred, curp);
639 	if (fp != NULL) {
640 		FILE_UNUSE(fp, curp);	/* release file */
641 		fdrelease(curp, fd); 	/* release fd table slot */
642 	}
643 	return (error);
644 }
645 
646 int
647 ktrops(curp, p, ops, facs, fp)
648 	struct proc *curp;
649 	struct proc *p;
650 	int ops;
651 	int facs;
652 	struct file *fp;
653 {
654 
655 	if (!ktrcanset(curp, p))
656 		return (0);
657 	if (KTROP(ops) == KTROP_SET) {
658 		if (p->p_tracep != fp) {
659 			/*
660 			 * if trace file already in use, relinquish
661 			 */
662 			ktrderef(p);
663 			p->p_tracep = fp;
664 			ktradref(p);
665 		}
666 		p->p_traceflag |= facs;
667 		if (curp->p_ucred->cr_uid == 0)
668 			p->p_traceflag |= KTRFAC_ROOT;
669 	} else {
670 		/* KTROP_CLEAR */
671 		if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) {
672 			/* no more tracing */
673 			ktrderef(p);
674 		}
675 	}
676 
677 	/*
678 	 * Emit an emulation record, every time there is a ktrace
679 	 * change/attach request.
680 	 */
681 	if (KTRPOINT(p, KTR_EMUL))
682 		p->p_traceflag |= KTRFAC_TRC_EMUL;
683 #ifdef __HAVE_SYSCALL_INTERN
684 	(*p->p_emul->e_syscall_intern)(p);
685 #endif
686 
687 	return (1);
688 }
689 
690 int
691 ktrsetchildren(curp, top, ops, facs, fp)
692 	struct proc *curp;
693 	struct proc *top;
694 	int ops;
695 	int facs;
696 	struct file *fp;
697 {
698 	struct proc *p;
699 	int ret = 0;
700 
701 	p = top;
702 	for (;;) {
703 		ret |= ktrops(curp, p, ops, facs, fp);
704 		/*
705 		 * If this process has children, descend to them next,
706 		 * otherwise do any siblings, and if done with this level,
707 		 * follow back up the tree (but not past top).
708 		 */
709 		if (LIST_FIRST(&p->p_children) != NULL) {
710 			p = LIST_FIRST(&p->p_children);
711 			continue;
712 		}
713 		for (;;) {
714 			if (p == top)
715 				return (ret);
716 			if (LIST_NEXT(p, p_sibling) != NULL) {
717 				p = LIST_NEXT(p, p_sibling);
718 				break;
719 			}
720 			p = p->p_pptr;
721 		}
722 	}
723 	/*NOTREACHED*/
724 }
725 
726 int
727 ktrwrite(p, kth)
728 	struct proc *p;
729 	struct ktr_header *kth;
730 {
731 	struct uio auio;
732 	struct iovec aiov[2];
733 	int error, tries;
734 	struct file *fp = p->p_tracep;
735 
736 	if (fp == NULL)
737 		return 0;
738 
739 	if (p->p_traceflag & KTRFAC_TRC_EMUL) {
740 		/* Add emulation trace before first entry for this process */
741 		p->p_traceflag &= ~KTRFAC_TRC_EMUL;
742 		ktremul(p);
743 	}
744 
745 	auio.uio_iov = &aiov[0];
746 	auio.uio_offset = 0;
747 	auio.uio_segflg = UIO_SYSSPACE;
748 	auio.uio_rw = UIO_WRITE;
749 	aiov[0].iov_base = (caddr_t)kth;
750 	aiov[0].iov_len = sizeof(struct ktr_header);
751 	auio.uio_resid = sizeof(struct ktr_header);
752 	auio.uio_iovcnt = 1;
753 	auio.uio_procp = (struct proc *)0;
754 	if (kth->ktr_len > 0) {
755 		auio.uio_iovcnt++;
756 		aiov[1].iov_base = (void *)kth->ktr_buf;
757 		aiov[1].iov_len = kth->ktr_len;
758 		auio.uio_resid += kth->ktr_len;
759 	}
760 
761 	simple_lock(&fp->f_slock);
762 	FILE_USE(fp);
763 
764 	tries = 0;
765 	do {
766 		error = (*fp->f_ops->fo_write)(fp, &fp->f_offset, &auio,
767 		    fp->f_cred, FOF_UPDATE_OFFSET);
768 		tries++;
769 		if (error == EWOULDBLOCK)
770 		  	preempt(1);
771 	} while ((error == EWOULDBLOCK) && (tries < 3));
772 	FILE_UNUSE(fp, NULL);
773 
774 	if (__predict_true(error == 0))
775 		return (0);
776 	/*
777 	 * If error encountered, give up tracing on this vnode.  Don't report
778 	 * EPIPE as this can easily happen with fktrace()/ktruss.
779 	 */
780 	if (error != EPIPE)
781 		log(LOG_NOTICE,
782 		    "ktrace write failed, errno %d, tracing stopped\n",
783 		    error);
784 	proclist_lock_read();
785 	LIST_FOREACH(p, &allproc, p_list) {
786 		if (ktrsamefile(p->p_tracep, fp))
787 			ktrderef(p);
788 	}
789 	proclist_unlock_read();
790 
791 	return (error);
792 }
793 
794 /*
795  * Return true if caller has permission to set the ktracing state
796  * of target.  Essentially, the target can't possess any
797  * more permissions than the caller.  KTRFAC_ROOT signifies that
798  * root previously set the tracing status on the target process, and
799  * so, only root may further change it.
800  *
801  * TODO: check groups.  use caller effective gid.
802  */
803 int
804 ktrcanset(callp, targetp)
805 	struct proc *callp;
806 	struct proc *targetp;
807 {
808 	struct pcred *caller = callp->p_cred;
809 	struct pcred *target = targetp->p_cred;
810 
811 	if ((caller->pc_ucred->cr_uid == target->p_ruid &&
812 	     target->p_ruid == target->p_svuid &&
813 	     caller->p_rgid == target->p_rgid &&	/* XXX */
814 	     target->p_rgid == target->p_svgid &&
815 	     (targetp->p_traceflag & KTRFAC_ROOT) == 0 &&
816 	     (targetp->p_flag & P_SUGID) == 0) ||
817 	     caller->pc_ucred->cr_uid == 0)
818 		return (1);
819 
820 	return (0);
821 }
822 #endif /* KTRACE */
823 
824 /*
825  * Put user defined entry to ktrace records.
826  */
827 int
828 sys_utrace(l, v, retval)
829 	struct lwp *l;
830 	void *v;
831 	register_t *retval;
832 {
833 #ifdef KTRACE
834 	struct sys_utrace_args /* {
835 		syscallarg(const char *) label;
836 		syscallarg(void *) addr;
837 		syscallarg(size_t) len;
838 	} */ *uap = v;
839 	struct proc *p = l->l_proc;
840 	if (!KTRPOINT(p, KTR_USER))
841 		return (0);
842 
843 	if (SCARG(uap, len) > KTR_USER_MAXLEN)
844 		return (EINVAL);
845 
846 	ktruser(p, SCARG(uap, label), SCARG(uap, addr), SCARG(uap, len), 1);
847 
848 	return (0);
849 #else /* !KTRACE */
850 	return ENOSYS;
851 #endif /* KTRACE */
852 }
853