xref: /openbsd-src/sys/kern/kern_ktrace.c (revision db3296cf5c1dd9058ceecc3a29fe4aaa0bd26000)
1 /*	$OpenBSD: kern_ktrace.c,v 1.30 2003/06/02 23:28:05 millert Exp $	*/
2 /*	$NetBSD: kern_ktrace.c,v 1.23 1996/02/09 18:59:36 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)kern_ktrace.c	8.2 (Berkeley) 9/23/93
33  */
34 
35 #ifdef KTRACE
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/proc.h>
40 #include <sys/file.h>
41 #include <sys/namei.h>
42 #include <sys/vnode.h>
43 #include <sys/ktrace.h>
44 #include <sys/malloc.h>
45 #include <sys/syslog.h>
46 
47 #include <sys/mount.h>
48 #include <sys/syscallargs.h>
49 
50 #include <uvm/uvm_extern.h>
51 
52 void ktrinitheader(struct ktr_header *, struct proc *, int);
53 int ktrops(struct proc *, struct proc *, int, int, struct vnode *);
54 int ktrsetchildren(struct proc *, struct proc *, int, int,
55 			struct vnode *);
56 int ktrwrite(struct proc *, struct ktr_header *);
57 int ktrcanset(struct proc *, struct proc *);
58 
59 /*
60  * Change the trace vnode in a correct way (to avoid races).
61  */
62 void
63 ktrsettracevnode(p, newvp)
64 	struct proc *p;
65 	struct vnode *newvp;
66 {
67 	struct vnode *vp;
68 
69 	if (p->p_tracep == newvp)	/* avoid work */
70 		return;
71 
72 	if (newvp != NULL)
73 		VREF(newvp);
74 
75 	vp = p->p_tracep;
76 	p->p_tracep = newvp;
77 
78 	if (vp != NULL)
79 		vrele(vp);
80 }
81 
82 void
83 ktrinitheader(kth, p, type)
84 	struct ktr_header *kth;
85 	struct proc *p;
86 	int type;
87 {
88 	bzero(kth, sizeof (struct ktr_header));
89 	kth->ktr_type = type;
90 	microtime(&kth->ktr_time);
91 	kth->ktr_pid = p->p_pid;
92 	bcopy(p->p_comm, kth->ktr_comm, MAXCOMLEN);
93 }
94 
95 void
96 ktrsyscall(p, code, argsize, args)
97 	struct proc *p;
98 	register_t code;
99 	size_t argsize;
100 	register_t args[];
101 {
102 	struct	ktr_header kth;
103 	struct	ktr_syscall *ktp;
104 	unsigned int len = sizeof(struct ktr_syscall) + argsize;
105 	register_t *argp;
106 	int i;
107 
108 	p->p_traceflag |= KTRFAC_ACTIVE;
109 	ktrinitheader(&kth, p, KTR_SYSCALL);
110 	ktp = malloc(len, M_TEMP, M_WAITOK);
111 	ktp->ktr_code = code;
112 	ktp->ktr_argsize = argsize;
113 	argp = (register_t *)((char *)ktp + sizeof(struct ktr_syscall));
114 	for (i = 0; i < (argsize / sizeof *argp); i++)
115 		*argp++ = args[i];
116 	kth.ktr_buf = (caddr_t)ktp;
117 	kth.ktr_len = len;
118 	ktrwrite(p, &kth);
119 	free(ktp, M_TEMP);
120 	p->p_traceflag &= ~KTRFAC_ACTIVE;
121 }
122 
123 void
124 ktrsysret(p, code, error, retval)
125 	struct proc *p;
126 	register_t code;
127 	int error;
128 	register_t retval;
129 {
130 	struct ktr_header kth;
131 	struct ktr_sysret ktp;
132 
133 	p->p_traceflag |= KTRFAC_ACTIVE;
134 	ktrinitheader(&kth, p, KTR_SYSRET);
135 	ktp.ktr_code = code;
136 	ktp.ktr_error = error;
137 	ktp.ktr_retval = retval;		/* what about val2 ? */
138 
139 	kth.ktr_buf = (caddr_t)&ktp;
140 	kth.ktr_len = sizeof(struct ktr_sysret);
141 
142 	ktrwrite(p, &kth);
143 	p->p_traceflag &= ~KTRFAC_ACTIVE;
144 }
145 
146 void
147 ktrnamei(p, path)
148 	struct proc *p;
149 	char *path;
150 {
151 	struct ktr_header kth;
152 
153 	p->p_traceflag |= KTRFAC_ACTIVE;
154 	ktrinitheader(&kth, p, KTR_NAMEI);
155 	kth.ktr_len = strlen(path);
156 	kth.ktr_buf = path;
157 
158 	ktrwrite(p, &kth);
159 	p->p_traceflag &= ~KTRFAC_ACTIVE;
160 }
161 
162 void
163 ktremul(p, emul)
164 	struct proc *p;
165 	char *emul;
166 {
167 	struct ktr_header kth;
168 
169 	p->p_traceflag |= KTRFAC_ACTIVE;
170 	ktrinitheader(&kth, p, KTR_EMUL);
171 	kth.ktr_len = strlen(emul);
172 	kth.ktr_buf = emul;
173 
174 	ktrwrite(p, &kth);
175 	p->p_traceflag &= ~KTRFAC_ACTIVE;
176 }
177 
178 void
179 ktrgenio(p, fd, rw, iov, len, error)
180 	struct proc *p;
181 	int fd;
182 	enum uio_rw rw;
183 	struct iovec *iov;
184 	int len, error;
185 {
186 	struct ktr_header kth;
187 	struct ktr_genio *ktp;
188 	caddr_t cp;
189 	int resid = len, count;
190 	int buflen;
191 
192 	if (error)
193 		return;
194 
195 	p->p_traceflag |= KTRFAC_ACTIVE;
196 
197 	buflen = min(PAGE_SIZE, len + sizeof(struct ktr_genio));
198 
199 	ktrinitheader(&kth, p, KTR_GENIO);
200 	ktp = malloc(buflen, M_TEMP, M_WAITOK);
201 	ktp->ktr_fd = fd;
202 	ktp->ktr_rw = rw;
203 
204 	kth.ktr_buf = (caddr_t)ktp;
205 
206 	cp = (caddr_t)((char *)ktp + sizeof (struct ktr_genio));
207 	buflen -= sizeof(struct ktr_genio);
208 
209 	while (resid > 0) {
210 		/*
211 		 * Don't allow this process to hog the cpu when doing
212 		 * huge I/O.
213 		 */
214 		if (p->p_schedflags & PSCHED_SHOULDYIELD)
215 			preempt(NULL);
216 
217 		count = min(iov->iov_len, buflen);
218 		if (count > resid)
219 			count = resid;
220 		if (copyin(iov->iov_base, cp, count))
221 			break;
222 
223 		kth.ktr_len = count + sizeof(struct ktr_genio);
224 
225 		if (ktrwrite(p, &kth) != 0)
226 			break;
227 
228 		iov->iov_len -= count;
229 		iov->iov_base = (caddr_t)iov->iov_base + count;
230 
231 		if (iov->iov_len == 0)
232 			iov++;
233 
234 		resid -= count;
235 	}
236 
237 	free(ktp, M_TEMP);
238 	p->p_traceflag &= ~KTRFAC_ACTIVE;
239 
240 }
241 
242 void
243 ktrpsig(p, sig, action, mask, code, si)
244 	struct proc *p;
245 	int sig;
246 	sig_t action;
247 	int mask, code;
248 	siginfo_t *si;
249 {
250 	struct ktr_header kth;
251 	struct ktr_psig kp;
252 
253 	p->p_traceflag |= KTRFAC_ACTIVE;
254 	ktrinitheader(&kth, p, KTR_PSIG);
255 	kp.signo = (char)sig;
256 	kp.action = action;
257 	kp.mask = mask;
258 	kp.code = code;
259 	kp.si = *si;
260 	kth.ktr_buf = (caddr_t)&kp;
261 	kth.ktr_len = sizeof (struct ktr_psig);
262 
263 	ktrwrite(p, &kth);
264 	p->p_traceflag &= ~KTRFAC_ACTIVE;
265 }
266 
267 void
268 ktrcsw(p, out, user)
269 	struct proc *p;
270 	int out, user;
271 {
272 	struct ktr_header kth;
273 	struct	ktr_csw kc;
274 
275 	p->p_traceflag |= KTRFAC_ACTIVE;
276 	ktrinitheader(&kth, p, KTR_CSW);
277 	kc.out = out;
278 	kc.user = user;
279 	kth.ktr_buf = (caddr_t)&kc;
280 	kth.ktr_len = sizeof (struct ktr_csw);
281 
282 	ktrwrite(p, &kth);
283 	p->p_traceflag &= ~KTRFAC_ACTIVE;
284 }
285 
286 /* Interface and common routines */
287 
288 /*
289  * ktrace system call
290  */
291 /* ARGSUSED */
292 int
293 sys_ktrace(curp, v, retval)
294 	struct proc *curp;
295 	void *v;
296 	register_t *retval;
297 {
298 	struct sys_ktrace_args /* {
299 		syscallarg(char *) fname;
300 		syscallarg(int) ops;
301 		syscallarg(int) facs;
302 		syscallarg(pid_t) pid;
303 	} */ *uap = v;
304 	struct vnode *vp = NULL;
305 	struct proc *p = NULL;
306 	struct pgrp *pg;
307 	int facs = SCARG(uap, facs) & ~((unsigned) KTRFAC_ROOT);
308 	int ops = KTROP(SCARG(uap, ops));
309 	int descend = SCARG(uap, ops) & KTRFLAG_DESCEND;
310 	int ret = 0;
311 	int error = 0;
312 	struct nameidata nd;
313 
314 	curp->p_traceflag |= KTRFAC_ACTIVE;
315 	if (ops != KTROP_CLEAR) {
316 		/*
317 		 * an operation which requires a file argument.
318 		 */
319 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, SCARG(uap, fname),
320 		    curp);
321 		if ((error = vn_open(&nd, FREAD|FWRITE|O_NOFOLLOW, 0)) != 0) {
322 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
323 			return (error);
324 		}
325 		vp = nd.ni_vp;
326 
327 		VOP_UNLOCK(vp, 0, curp);
328 		if (vp->v_type != VREG) {
329 			(void) vn_close(vp, FREAD|FWRITE, curp->p_ucred, curp);
330 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
331 			return (EACCES);
332 		}
333 	}
334 	/*
335 	 * Clear all uses of the tracefile
336 	 */
337 	if (ops == KTROP_CLEARFILE) {
338 		for (p = LIST_FIRST(&allproc); p; p = LIST_NEXT(p, p_list)) {
339 			if (p->p_tracep == vp) {
340 				if (ktrcanset(curp, p)) {
341 					p->p_traceflag = 0;
342 					ktrsettracevnode(p, NULL);
343 				} else
344 					error = EPERM;
345 			}
346 		}
347 		goto done;
348 	}
349 	/*
350 	 * need something to (un)trace (XXX - why is this here?)
351 	 */
352 	if (!facs) {
353 		error = EINVAL;
354 		goto done;
355 	}
356 	/*
357 	 * do it
358 	 */
359 	if (SCARG(uap, pid) < 0) {
360 		/*
361 		 * by process group
362 		 */
363 		pg = pgfind(-SCARG(uap, pid));
364 		if (pg == NULL) {
365 			error = ESRCH;
366 			goto done;
367 		}
368 		for (p = pg->pg_members.lh_first; p != 0; p = p->p_pglist.le_next)
369 			if (descend)
370 				ret |= ktrsetchildren(curp, p, ops, facs, vp);
371 			else
372 				ret |= ktrops(curp, p, ops, facs, vp);
373 
374 	} else {
375 		/*
376 		 * by pid
377 		 */
378 		p = pfind(SCARG(uap, pid));
379 		if (p == NULL) {
380 			error = ESRCH;
381 			goto done;
382 		}
383 		if (descend)
384 			ret |= ktrsetchildren(curp, p, ops, facs, vp);
385 		else
386 			ret |= ktrops(curp, p, ops, facs, vp);
387 	}
388 	if (!ret)
389 		error = EPERM;
390 done:
391 	if (vp != NULL)
392 		(void) vn_close(vp, FWRITE, curp->p_ucred, curp);
393 	curp->p_traceflag &= ~KTRFAC_ACTIVE;
394 	return (error);
395 }
396 
397 int
398 ktrops(curp, p, ops, facs, vp)
399 	struct proc *p, *curp;
400 	int ops, facs;
401 	struct vnode *vp;
402 {
403 
404 	if (!ktrcanset(curp, p))
405 		return (0);
406 	if (ops == KTROP_SET) {
407 		ktrsettracevnode(p, vp);
408 		p->p_traceflag |= facs;
409 		if (curp->p_ucred->cr_uid == 0)
410 			p->p_traceflag |= KTRFAC_ROOT;
411 	} else {
412 		/* KTROP_CLEAR */
413 		if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) {
414 			/* no more tracing */
415 			p->p_traceflag = 0;
416 			ktrsettracevnode(p, NULL);
417 		}
418 	}
419 
420 	/*
421 	 * Emit an emulation record, every time there is a ktrace
422 	 * change/attach request.
423 	 */
424 	if (KTRPOINT(p, KTR_EMUL))
425 		ktremul(p, p->p_emul->e_name);
426 
427 	return (1);
428 }
429 
430 int
431 ktrsetchildren(curp, top, ops, facs, vp)
432 	struct proc *curp, *top;
433 	int ops, facs;
434 	struct vnode *vp;
435 {
436 	struct proc *p;
437 	int ret = 0;
438 
439 	p = top;
440 	for (;;) {
441 		ret |= ktrops(curp, p, ops, facs, vp);
442 		/*
443 		 * If this process has children, descend to them next,
444 		 * otherwise do any siblings, and if done with this level,
445 		 * follow back up the tree (but not past top).
446 		 */
447 		if (p->p_children.lh_first)
448 			p = p->p_children.lh_first;
449 		else for (;;) {
450 			if (p == top)
451 				return (ret);
452 			if (p->p_sibling.le_next) {
453 				p = p->p_sibling.le_next;
454 				break;
455 			}
456 			p = p->p_pptr;
457 		}
458 	}
459 	/*NOTREACHED*/
460 }
461 
462 int
463 ktrwrite(p, kth)
464 	struct proc *p;
465 	struct ktr_header *kth;
466 {
467 	struct uio auio;
468 	struct iovec aiov[2];
469 	int error;
470 	struct vnode *vp = p->p_tracep;
471 
472 	if (vp == NULL)
473 		return 0;
474 	auio.uio_iov = &aiov[0];
475 	auio.uio_offset = 0;
476 	auio.uio_segflg = UIO_SYSSPACE;
477 	auio.uio_rw = UIO_WRITE;
478 	aiov[0].iov_base = (caddr_t)kth;
479 	aiov[0].iov_len = sizeof(struct ktr_header);
480 	auio.uio_resid = sizeof(struct ktr_header);
481 	auio.uio_iovcnt = 1;
482 	auio.uio_procp = (struct proc *)0;
483 	if (kth->ktr_len > 0) {
484 		auio.uio_iovcnt++;
485 		aiov[1].iov_base = kth->ktr_buf;
486 		aiov[1].iov_len = kth->ktr_len;
487 		auio.uio_resid += kth->ktr_len;
488 	}
489 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
490 	error = VOP_WRITE(vp, &auio, IO_UNIT|IO_APPEND, p->p_ucred);
491 	VOP_UNLOCK(vp, 0, p);
492 	if (!error)
493 		return 0;
494 	/*
495 	 * If error encountered, give up tracing on this vnode.
496 	 */
497 	log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n",
498 	    error);
499 	for (p = LIST_FIRST(&allproc); p != NULL; p = LIST_NEXT(p, p_list)) {
500 		if (p->p_tracep == vp) {
501 			p->p_traceflag = 0;
502 			ktrsettracevnode(p, NULL);
503 		}
504 	}
505 
506 	return error;
507 }
508 
509 /*
510  * Return true if caller has permission to set the ktracing state
511  * of target.  Essentially, the target can't possess any
512  * more permissions than the caller.  KTRFAC_ROOT signifies that
513  * root previously set the tracing status on the target process, and
514  * so, only root may further change it.
515  *
516  * TODO: check groups.  use caller effective gid.
517  */
518 int
519 ktrcanset(callp, targetp)
520 	struct proc *callp, *targetp;
521 {
522 	struct pcred *caller = callp->p_cred;
523 	struct pcred *target = targetp->p_cred;
524 
525 	if ((caller->pc_ucred->cr_uid == target->p_ruid &&
526 	    target->p_ruid == target->p_svuid &&
527 	    caller->p_rgid == target->p_rgid &&	/* XXX */
528 	    target->p_rgid == target->p_svgid &&
529 	    (targetp->p_traceflag & KTRFAC_ROOT) == 0 &&
530 	    !ISSET(targetp->p_flag, P_SUGID)) ||
531 	    caller->pc_ucred->cr_uid == 0)
532 		return (1);
533 
534 	return (0);
535 }
536 
537 #endif
538