Home | History | Annotate | Line # | Download | only in kern
kern_ktrace.c revision 1.87 
      1 /*	$NetBSD: kern_ktrace.c,v 1.87 2004/02/25 21:34:18 enami 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 1.87 2004/02/25 21:34:18 enami 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 void	ktrinitheader(struct ktr_header *, struct proc *, int);
     59 int	ktrwrite(struct proc *, struct ktr_header *);
     60 int	ktrace_common(struct proc *, int, int, int, struct file *);
     61 int	ktrops(struct proc *, struct proc *, int, int, struct file *);
     62 int	ktrsetchildren(struct proc *, struct proc *, int, int,
     63 	    struct file *);
     64 int	ktrcanset(struct proc *, struct proc *);
     65 int	ktrsamefile(struct file *, struct file *);
     66 
     67 /*
     68  * "deep" compare of two files for the purposes of clearing a trace.
     69  * Returns true if they're the same open file, or if they point at the
     70  * same underlying vnode/socket.
     71  */
     72 
     73 int
     74 ktrsamefile(f1, f2)
     75 	struct file *f1;
     76 	struct file *f2;
     77 {
     78 	return ((f1 == f2) ||
     79 	    ((f1 != NULL) && (f2 != NULL) &&
     80 		(f1->f_type == f2->f_type) &&
     81 		(f1->f_data == f2->f_data)));
     82 }
     83 
     84 void
     85 ktrderef(p)
     86 	struct proc *p;
     87 {
     88 	struct file *fp = p->p_tracep;
     89 	p->p_traceflag = 0;
     90 	if (fp == NULL)
     91 		return;
     92 	p->p_tracep = NULL;
     93 
     94 	simple_lock(&fp->f_slock);
     95 	FILE_USE(fp);
     96 
     97 	/*
     98 	 * ktrace file descriptor can't be watched (are not visible to
     99 	 * userspace), so no kqueue stuff here
    100 	 */
    101 	closef(fp, NULL);
    102 }
    103 
    104 void
    105 ktradref(p)
    106 	struct proc *p;
    107 {
    108 	struct file *fp = p->p_tracep;
    109 
    110 	fp->f_count++;
    111 }
    112 
    113 void
    114 ktrinitheader(kth, p, type)
    115 	struct ktr_header *kth;
    116 	struct proc *p;
    117 	int type;
    118 {
    119 
    120 	memset(kth, 0, sizeof(*kth));
    121 	kth->ktr_type = type;
    122 	microtime(&kth->ktr_time);
    123 	kth->ktr_pid = p->p_pid;
    124 	memcpy(kth->ktr_comm, p->p_comm, MAXCOMLEN);
    125 }
    126 
    127 void
    128 ktrsyscall(p, code, realcode, callp, args)
    129 	struct proc *p;
    130 	register_t code;
    131 	register_t realcode;
    132 	const struct sysent *callp;
    133 	register_t args[];
    134 {
    135 	struct ktr_header kth;
    136 	struct ktr_syscall *ktp;
    137 	register_t *argp;
    138 	int argsize;
    139 	size_t len;
    140 	u_int i;
    141 
    142 	if (callp == NULL)
    143 		callp = p->p_emul->e_sysent;
    144 
    145 	argsize = callp[code].sy_argsize;
    146 #ifdef _LP64
    147 	if (p->p_flag & P_32)
    148 		argsize = argsize << 1;
    149 #endif
    150 	len = sizeof(struct ktr_syscall) + argsize;
    151 
    152 	p->p_traceflag |= KTRFAC_ACTIVE;
    153 	ktrinitheader(&kth, p, KTR_SYSCALL);
    154 	ktp = malloc(len, M_TEMP, M_WAITOK);
    155 	ktp->ktr_code = realcode;
    156 	ktp->ktr_argsize = argsize;
    157 	argp = (register_t *)((char *)ktp + sizeof(struct ktr_syscall));
    158 	for (i = 0; i < (argsize / sizeof(*argp)); i++)
    159 		*argp++ = args[i];
    160 	kth.ktr_buf = (caddr_t)ktp;
    161 	kth.ktr_len = len;
    162 	(void) ktrwrite(p, &kth);
    163 	free(ktp, M_TEMP);
    164 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    165 }
    166 
    167 void
    168 ktrsysret(p, code, error, retval)
    169 	struct proc *p;
    170 	register_t code;
    171 	int error;
    172 	register_t *retval;
    173 {
    174 	struct ktr_header kth;
    175 	struct ktr_sysret ktp;
    176 
    177 	p->p_traceflag |= KTRFAC_ACTIVE;
    178 	ktrinitheader(&kth, p, KTR_SYSRET);
    179 	ktp.ktr_code = code;
    180 	ktp.ktr_eosys = 0;			/* XXX unused */
    181 	ktp.ktr_error = error;
    182 	ktp.ktr_retval = retval ? retval[0] : 0;
    183 	ktp.ktr_retval_1 = retval ? retval[1] : 0;
    184 
    185 	kth.ktr_buf = (caddr_t)&ktp;
    186 	kth.ktr_len = sizeof(struct ktr_sysret);
    187 
    188 	(void) ktrwrite(p, &kth);
    189 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    190 }
    191 
    192 void
    193 ktrnamei(p, path)
    194 	struct proc *p;
    195 	char *path;
    196 {
    197 	struct ktr_header kth;
    198 
    199 	p->p_traceflag |= KTRFAC_ACTIVE;
    200 	ktrinitheader(&kth, p, KTR_NAMEI);
    201 	kth.ktr_len = strlen(path);
    202 	kth.ktr_buf = path;
    203 
    204 	(void) ktrwrite(p, &kth);
    205 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    206 }
    207 
    208 void
    209 ktremul(p)
    210 	struct proc *p;
    211 {
    212 	struct ktr_header kth;
    213 	const char *emul = p->p_emul->e_name;
    214 
    215 	p->p_traceflag |= KTRFAC_ACTIVE;
    216 	ktrinitheader(&kth, p, KTR_EMUL);
    217 	kth.ktr_len = strlen(emul);
    218 	kth.ktr_buf = (caddr_t)emul;
    219 
    220 	(void) ktrwrite(p, &kth);
    221 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    222 }
    223 
    224 void
    225 ktrkmem(struct proc *p, int ktr, const void *buf, size_t len)
    226 {
    227 	struct ktr_header kth;
    228 
    229 	p->p_traceflag |= KTRFAC_ACTIVE;
    230 	ktrinitheader(&kth, p, ktr);
    231 	kth.ktr_len = len;
    232 	kth.ktr_buf = buf;
    233 
    234 	(void)ktrwrite(p, &kth);
    235 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    236 }
    237 
    238 void
    239 ktrgenio(p, fd, rw, iov, len, error)
    240 	struct proc *p;
    241 	int fd;
    242 	enum uio_rw rw;
    243 	struct iovec *iov;
    244 	int len;
    245 	int error;
    246 {
    247 	struct ktr_header kth;
    248 	struct ktr_genio *ktp;
    249 	caddr_t cp;
    250 	int resid = len, cnt;
    251 	int buflen;
    252 
    253 	if (error)
    254 		return;
    255 
    256 	p->p_traceflag |= KTRFAC_ACTIVE;
    257 
    258 	buflen = min(PAGE_SIZE, len + sizeof(struct ktr_genio));
    259 
    260 	ktrinitheader(&kth, p, KTR_GENIO);
    261 	ktp = malloc(buflen, M_TEMP, M_WAITOK);
    262 	ktp->ktr_fd = fd;
    263 	ktp->ktr_rw = rw;
    264 
    265 	kth.ktr_buf = (caddr_t)ktp;
    266 
    267 	cp = (caddr_t)((char *)ktp + sizeof(struct ktr_genio));
    268 	buflen -= sizeof(struct ktr_genio);
    269 
    270 	while (resid > 0) {
    271 #if 0 /* XXX NJWLWP */
    272 		KDASSERT(p->p_cpu != NULL);
    273 		KDASSERT(p->p_cpu == curcpu());
    274 #endif
    275 		/* XXX NJWLWP */
    276 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
    277 			preempt(1);
    278 
    279 		cnt = min(iov->iov_len, buflen);
    280 		if (cnt > resid)
    281 			cnt = resid;
    282 		if (copyin(iov->iov_base, cp, cnt))
    283 			break;
    284 
    285 		kth.ktr_len = cnt + sizeof(struct ktr_genio);
    286 
    287 		if (__predict_false(ktrwrite(p, &kth) != 0))
    288 			break;
    289 
    290 		iov->iov_base = (caddr_t)iov->iov_base + cnt;
    291 		iov->iov_len -= cnt;
    292 
    293 		if (iov->iov_len == 0)
    294 			iov++;
    295 
    296 		resid -= cnt;
    297 	}
    298 
    299 	free(ktp, M_TEMP);
    300 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    301 }
    302 
    303 void
    304 ktrpsig(p, sig, action, mask, ksi)
    305 	struct proc *p;
    306 	int sig;
    307 	sig_t action;
    308 	const sigset_t *mask;
    309 	const ksiginfo_t *ksi;
    310 {
    311 	struct ktr_header kth;
    312 	struct {
    313 		struct ktr_psig	kp;
    314 		siginfo_t	si;
    315 	} kbuf;
    316 
    317 	p->p_traceflag |= KTRFAC_ACTIVE;
    318 	ktrinitheader(&kth, p, KTR_PSIG);
    319 	kbuf.kp.signo = (char)sig;
    320 	kbuf.kp.action = action;
    321 	kbuf.kp.mask = *mask;
    322 	kth.ktr_buf = (caddr_t)&kbuf;
    323 	if (ksi) {
    324 		kbuf.kp.code = KSI_TRAPCODE(ksi);
    325 		(void)memset(&kbuf.si, 0, sizeof(kbuf.si));
    326 		kbuf.si._info = ksi->ksi_info;
    327 		kth.ktr_len = sizeof(kbuf);
    328 	} else {
    329 		kbuf.kp.code = 0;
    330 		kth.ktr_len = sizeof(struct ktr_psig);
    331 	}
    332 	(void) ktrwrite(p, &kth);
    333 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    334 }
    335 
    336 void
    337 ktrcsw(p, out, user)
    338 	struct proc *p;
    339 	int out;
    340 	int user;
    341 {
    342 	struct ktr_header kth;
    343 	struct ktr_csw kc;
    344 
    345 	p->p_traceflag |= KTRFAC_ACTIVE;
    346 	ktrinitheader(&kth, p, KTR_CSW);
    347 	kc.out = out;
    348 	kc.user = user;
    349 	kth.ktr_buf = (caddr_t)&kc;
    350 	kth.ktr_len = sizeof(struct ktr_csw);
    351 
    352 	(void) ktrwrite(p, &kth);
    353 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    354 }
    355 
    356 void
    357 ktruser(p, id, addr, len, ustr)
    358 	struct proc *p;
    359 	const char *id;
    360 	void *addr;
    361 	size_t len;
    362 	int ustr;
    363 {
    364 	struct ktr_header kth;
    365 	struct ktr_user *ktp;
    366 	caddr_t user_dta;
    367 
    368 	p->p_traceflag |= KTRFAC_ACTIVE;
    369 	ktrinitheader(&kth, p, KTR_USER);
    370 	ktp = malloc(sizeof(struct ktr_user) + len, M_TEMP, M_WAITOK);
    371 	if (ustr) {
    372 		if (copyinstr(id, ktp->ktr_id, KTR_USER_MAXIDLEN, NULL) != 0)
    373 			ktp->ktr_id[0] = '\0';
    374 	} else
    375 		strncpy(ktp->ktr_id, id, KTR_USER_MAXIDLEN);
    376 	ktp->ktr_id[KTR_USER_MAXIDLEN-1] = '\0';
    377 
    378 	user_dta = (caddr_t) ((char *)ktp + sizeof(struct ktr_user));
    379 	if (copyin(addr, (void *) user_dta, len) != 0)
    380 		len = 0;
    381 
    382 	kth.ktr_buf = (void *)ktp;
    383 	kth.ktr_len = sizeof(struct ktr_user) + len;
    384 	(void) ktrwrite(p, &kth);
    385 
    386 	free(ktp, M_TEMP);
    387 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    388 
    389 }
    390 
    391 void
    392 ktrmmsg(p, msgh, size)
    393 	struct proc *p;
    394 	const void *msgh;
    395 	size_t size;
    396 {
    397 	struct ktr_header kth;
    398 	struct ktr_mmsg	*kp;
    399 
    400 	p->p_traceflag |= KTRFAC_ACTIVE;
    401 	ktrinitheader(&kth, p, KTR_MMSG);
    402 
    403 	kp = (struct ktr_mmsg *)msgh;
    404 	kth.ktr_buf = (caddr_t)kp;
    405 	kth.ktr_len = size;
    406 	(void) ktrwrite(p, &kth);
    407 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    408 }
    409 
    410 void
    411 ktrmool(p, kaddr, size, uaddr)
    412 	struct proc *p;
    413 	const void *kaddr;
    414 	size_t size;
    415 	const void *uaddr;
    416 {
    417 	struct ktr_header kth;
    418 	struct ktr_mool *kp;
    419 	struct ktr_mool *buf;
    420 
    421 	p->p_traceflag |= KTRFAC_ACTIVE;
    422 	ktrinitheader(&kth, p, KTR_MOOL);
    423 
    424 	kp = malloc(size + sizeof(*kp), M_TEMP, M_WAITOK);
    425 	kp->uaddr = uaddr;
    426 	kp->size = size;
    427 	buf = kp + 1; /* Skip uaddr and size */
    428 	memcpy(buf, kaddr, size);
    429 
    430 	kth.ktr_buf = (caddr_t)kp;
    431 	kth.ktr_len = size + sizeof(*kp);
    432 	(void) ktrwrite(p, &kth);
    433 	free(kp, M_TEMP);
    434 
    435 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    436 }
    437 
    438 
    439 /* Interface and common routines */
    440 
    441 int
    442 ktrace_common(curp, ops, facs, pid, fp)
    443 	struct proc *curp;
    444 	int ops;
    445 	int facs;
    446 	int pid;
    447 	struct file *fp;
    448 {
    449 	int ret = 0;
    450 	int error = 0;
    451 	int one = 1;
    452 	int descend;
    453 	struct proc *p;
    454 	struct pgrp *pg;
    455 
    456 	curp->p_traceflag |= KTRFAC_ACTIVE;
    457 	descend = ops & KTRFLAG_DESCEND;
    458 	facs = facs & ~((unsigned) KTRFAC_ROOT);
    459 
    460 	/*
    461 	 * Clear all uses of the tracefile
    462 	 */
    463 	if (KTROP(ops) == KTROP_CLEARFILE) {
    464 		proclist_lock_read();
    465 		LIST_FOREACH(p, &allproc, p_list) {
    466 			if (ktrsamefile(p->p_tracep, fp)) {
    467 				if (ktrcanset(curp, p))
    468 					ktrderef(p);
    469 				else
    470 					error = EPERM;
    471 			}
    472 		}
    473 		proclist_unlock_read();
    474 		goto done;
    475 	}
    476 
    477 	/*
    478 	 * Mark fp non-blocking, to avoid problems from possible deadlocks.
    479 	 */
    480 
    481 	if (fp != NULL) {
    482 		fp->f_flag |= FNONBLOCK;
    483 		(*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&one, curp);
    484 	}
    485 
    486 	/*
    487 	 * need something to (un)trace (XXX - why is this here?)
    488 	 */
    489 	if (!facs) {
    490 		error = EINVAL;
    491 		goto done;
    492 	}
    493 	/*
    494 	 * do it
    495 	 */
    496 	if (pid < 0) {
    497 		/*
    498 		 * by process group
    499 		 */
    500 		pg = pg_find(-pid, PFIND_UNLOCK_FAIL);
    501 		if (pg == NULL) {
    502 			error = ESRCH;
    503 			goto done;
    504 		}
    505 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
    506 			if (descend)
    507 				ret |= ktrsetchildren(curp, p, ops, facs, fp);
    508 			else
    509 				ret |= ktrops(curp, p, ops, facs, fp);
    510 		}
    511 
    512 	} else {
    513 		/*
    514 		 * by pid
    515 		 */
    516 		p = p_find(pid, PFIND_UNLOCK_FAIL);
    517 		if (p == NULL) {
    518 			error = ESRCH;
    519 			goto done;
    520 		}
    521 		if (descend)
    522 			ret |= ktrsetchildren(curp, p, ops, facs, fp);
    523 		else
    524 			ret |= ktrops(curp, p, ops, facs, fp);
    525 	}
    526 	proclist_unlock_read();	/* taken by p{g}_find */
    527 	if (!ret)
    528 		error = EPERM;
    529 done:
    530 	curp->p_traceflag &= ~KTRFAC_ACTIVE;
    531 	return (error);
    532 }
    533 
    534 /*
    535  * ktrace system call
    536  */
    537 /* ARGSUSED */
    538 int
    539 sys_fktrace(l, v, retval)
    540 	struct lwp *l;
    541 	void *v;
    542 	register_t *retval;
    543 {
    544 	struct sys_fktrace_args /* {
    545 		syscallarg(int) fd;
    546 		syscallarg(int) ops;
    547 		syscallarg(int) facs;
    548 		syscallarg(int) pid;
    549 	} */ *uap = v;
    550 	struct proc *curp = l->l_proc;
    551 	struct file *fp = NULL;
    552 	struct filedesc *fdp = curp->p_fd;
    553 	int error;
    554 
    555 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL)
    556 		return (EBADF);
    557 
    558 	FILE_USE(fp);
    559 
    560 	if ((fp->f_flag & FWRITE) == 0)
    561 		error = EBADF;
    562 	else
    563 		error = ktrace_common(curp, SCARG(uap, ops),
    564 		    SCARG(uap, facs), SCARG(uap, pid), fp);
    565 
    566 	FILE_UNUSE(fp, curp);
    567 
    568 	return error;
    569 }
    570 
    571 /*
    572  * ktrace system call
    573  */
    574 /* ARGSUSED */
    575 int
    576 sys_ktrace(l, v, retval)
    577 	struct lwp *l;
    578 	void *v;
    579 	register_t *retval;
    580 {
    581 	struct sys_ktrace_args /* {
    582 		syscallarg(const char *) fname;
    583 		syscallarg(int) ops;
    584 		syscallarg(int) facs;
    585 		syscallarg(int) pid;
    586 	} */ *uap = v;
    587 	struct proc *curp = l->l_proc;
    588 	struct vnode *vp = NULL;
    589 	struct file *fp = NULL;
    590 	int fd;
    591 	int ops = SCARG(uap, ops);
    592 	int error = 0;
    593 	struct nameidata nd;
    594 
    595 	ops = KTROP(ops) | (ops & KTRFLAG_DESCEND);
    596 
    597 	curp->p_traceflag |= KTRFAC_ACTIVE;
    598 	if ((ops & KTROP_CLEAR) == 0) {
    599 		/*
    600 		 * an operation which requires a file argument.
    601 		 */
    602 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, SCARG(uap, fname),
    603 		    curp);
    604 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
    605 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
    606 			return (error);
    607 		}
    608 		vp = nd.ni_vp;
    609 		VOP_UNLOCK(vp, 0);
    610 		if (vp->v_type != VREG) {
    611 			(void) vn_close(vp, FREAD|FWRITE, curp->p_ucred, curp);
    612 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
    613 			return (EACCES);
    614 		}
    615 		/*
    616 		 * XXX This uses up a file descriptor slot in the
    617 		 * tracing process for the duration of this syscall.
    618 		 * This is not expected to be a problem.  If
    619 		 * falloc(NULL, ...) DTRT we could skip that part, but
    620 		 * that would require changing its interface to allow
    621 		 * the caller to pass in a ucred..
    622 		 *
    623 		 * This will FILE_USE the fp it returns, if any.
    624 		 * Keep it in use until we return.
    625 		 */
    626 		if ((error = falloc(curp, &fp, &fd)) != 0)
    627 			goto done;
    628 
    629 		fp->f_flag = FWRITE|FAPPEND;
    630 		fp->f_type = DTYPE_VNODE;
    631 		fp->f_ops = &vnops;
    632 		fp->f_data = (caddr_t)vp;
    633 		FILE_SET_MATURE(fp);
    634 		vp = NULL;
    635 	}
    636 	error = ktrace_common(curp, SCARG(uap, ops), SCARG(uap, facs),
    637 	    SCARG(uap, pid), fp);
    638 done:
    639 	if (vp != NULL)
    640 		(void) vn_close(vp, FWRITE, curp->p_ucred, curp);
    641 	if (fp != NULL) {
    642 		FILE_UNUSE(fp, curp);	/* release file */
    643 		fdrelease(curp, fd); 	/* release fd table slot */
    644 	}
    645 	return (error);
    646 }
    647 
    648 int
    649 ktrops(curp, p, ops, facs, fp)
    650 	struct proc *curp;
    651 	struct proc *p;
    652 	int ops;
    653 	int facs;
    654 	struct file *fp;
    655 {
    656 
    657 	if (!ktrcanset(curp, p))
    658 		return (0);
    659 	if (KTROP(ops) == KTROP_SET) {
    660 		if (p->p_tracep != fp) {
    661 			/*
    662 			 * if trace file already in use, relinquish
    663 			 */
    664 			ktrderef(p);
    665 			p->p_tracep = fp;
    666 			ktradref(p);
    667 		}
    668 		p->p_traceflag |= facs;
    669 		if (curp->p_ucred->cr_uid == 0)
    670 			p->p_traceflag |= KTRFAC_ROOT;
    671 	} else {
    672 		/* KTROP_CLEAR */
    673 		if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) {
    674 			/* no more tracing */
    675 			ktrderef(p);
    676 		}
    677 	}
    678 
    679 	/*
    680 	 * Emit an emulation record, every time there is a ktrace
    681 	 * change/attach request.
    682 	 */
    683 	if (KTRPOINT(p, KTR_EMUL))
    684 		p->p_traceflag |= KTRFAC_TRC_EMUL;
    685 #ifdef __HAVE_SYSCALL_INTERN
    686 	(*p->p_emul->e_syscall_intern)(p);
    687 #endif
    688 
    689 	return (1);
    690 }
    691 
    692 int
    693 ktrsetchildren(curp, top, ops, facs, fp)
    694 	struct proc *curp;
    695 	struct proc *top;
    696 	int ops;
    697 	int facs;
    698 	struct file *fp;
    699 {
    700 	struct proc *p;
    701 	int ret = 0;
    702 
    703 	p = top;
    704 	for (;;) {
    705 		ret |= ktrops(curp, p, ops, facs, fp);
    706 		/*
    707 		 * If this process has children, descend to them next,
    708 		 * otherwise do any siblings, and if done with this level,
    709 		 * follow back up the tree (but not past top).
    710 		 */
    711 		if (LIST_FIRST(&p->p_children) != NULL) {
    712 			p = LIST_FIRST(&p->p_children);
    713 			continue;
    714 		}
    715 		for (;;) {
    716 			if (p == top)
    717 				return (ret);
    718 			if (LIST_NEXT(p, p_sibling) != NULL) {
    719 				p = LIST_NEXT(p, p_sibling);
    720 				break;
    721 			}
    722 			p = p->p_pptr;
    723 		}
    724 	}
    725 	/*NOTREACHED*/
    726 }
    727 
    728 int
    729 ktrwrite(p, kth)
    730 	struct proc *p;
    731 	struct ktr_header *kth;
    732 {
    733 	struct uio auio;
    734 	struct iovec aiov[2];
    735 	int error, tries;
    736 	struct file *fp = p->p_tracep;
    737 
    738 	if (fp == NULL)
    739 		return 0;
    740 
    741 	if (p->p_traceflag & KTRFAC_TRC_EMUL) {
    742 		/* Add emulation trace before first entry for this process */
    743 		p->p_traceflag &= ~KTRFAC_TRC_EMUL;
    744 		ktremul(p);
    745 	}
    746 
    747 	auio.uio_iov = &aiov[0];
    748 	auio.uio_offset = 0;
    749 	auio.uio_segflg = UIO_SYSSPACE;
    750 	auio.uio_rw = UIO_WRITE;
    751 	aiov[0].iov_base = (caddr_t)kth;
    752 	aiov[0].iov_len = sizeof(struct ktr_header);
    753 	auio.uio_resid = sizeof(struct ktr_header);
    754 	auio.uio_iovcnt = 1;
    755 	auio.uio_procp = (struct proc *)0;
    756 	if (kth->ktr_len > 0) {
    757 		auio.uio_iovcnt++;
    758 		aiov[1].iov_base = (void *)kth->ktr_buf;
    759 		aiov[1].iov_len = kth->ktr_len;
    760 		auio.uio_resid += kth->ktr_len;
    761 	}
    762 
    763 	simple_lock(&fp->f_slock);
    764 	FILE_USE(fp);
    765 
    766 	tries = 0;
    767 	do {
    768 		error = (*fp->f_ops->fo_write)(fp, &fp->f_offset, &auio,
    769 		    fp->f_cred, FOF_UPDATE_OFFSET);
    770 		tries++;
    771 		if (error == EWOULDBLOCK)
    772 		  	preempt(1);
    773 	} while ((error == EWOULDBLOCK) && (tries < 3));
    774 	FILE_UNUSE(fp, NULL);
    775 
    776 	if (__predict_true(error == 0))
    777 		return (0);
    778 	/*
    779 	 * If error encountered, give up tracing on this vnode.  Don't report
    780 	 * EPIPE as this can easily happen with fktrace()/ktruss.
    781 	 */
    782 	if (error != EPIPE)
    783 		log(LOG_NOTICE,
    784 		    "ktrace write failed, errno %d, tracing stopped\n",
    785 		    error);
    786 	proclist_lock_read();
    787 	LIST_FOREACH(p, &allproc, p_list) {
    788 		if (ktrsamefile(p->p_tracep, fp))
    789 			ktrderef(p);
    790 	}
    791 	proclist_unlock_read();
    792 
    793 	return (error);
    794 }
    795 
    796 /*
    797  * Return true if caller has permission to set the ktracing state
    798  * of target.  Essentially, the target can't possess any
    799  * more permissions than the caller.  KTRFAC_ROOT signifies that
    800  * root previously set the tracing status on the target process, and
    801  * so, only root may further change it.
    802  *
    803  * TODO: check groups.  use caller effective gid.
    804  */
    805 int
    806 ktrcanset(callp, targetp)
    807 	struct proc *callp;
    808 	struct proc *targetp;
    809 {
    810 	struct pcred *caller = callp->p_cred;
    811 	struct pcred *target = targetp->p_cred;
    812 
    813 	if ((caller->pc_ucred->cr_uid == target->p_ruid &&
    814 	     target->p_ruid == target->p_svuid &&
    815 	     caller->p_rgid == target->p_rgid &&	/* XXX */
    816 	     target->p_rgid == target->p_svgid &&
    817 	     (targetp->p_traceflag & KTRFAC_ROOT) == 0 &&
    818 	     (targetp->p_flag & P_SUGID) == 0) ||
    819 	     caller->pc_ucred->cr_uid == 0)
    820 		return (1);
    821 
    822 	return (0);
    823 }
    824 #endif /* KTRACE */
    825 
    826 /*
    827  * Put user defined entry to ktrace records.
    828  */
    829 int
    830 sys_utrace(l, v, retval)
    831 	struct lwp *l;
    832 	void *v;
    833 	register_t *retval;
    834 {
    835 #ifdef KTRACE
    836 	struct sys_utrace_args /* {
    837 		syscallarg(const char *) label;
    838 		syscallarg(void *) addr;
    839 		syscallarg(size_t) len;
    840 	} */ *uap = v;
    841 	struct proc *p = l->l_proc;
    842 	if (!KTRPOINT(p, KTR_USER))
    843 		return (0);
    844 
    845 	if (SCARG(uap, len) > KTR_USER_MAXLEN)
    846 		return (EINVAL);
    847 
    848 	ktruser(p, SCARG(uap, label), SCARG(uap, addr), SCARG(uap, len), 1);
    849 
    850 	return (0);
    851 #else /* !KTRACE */
    852 	return ENOSYS;
    853 #endif /* KTRACE */
    854 }
    855