Home | History | Annotate | Line # | Download | only in kern
kern_ktrace.c revision 1.88.2.1
      1 /*	$NetBSD: kern_ktrace.c,v 1.88.2.1 2004/06/24 14:04:46 he 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 1.88.2.1 2004/06/24 14:04:46 he 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(struct file *f1, struct file *f2)
     75 {
     76 
     77 	return ((f1 == f2) ||
     78 	    ((f1 != NULL) && (f2 != NULL) &&
     79 		(f1->f_type == f2->f_type) &&
     80 		(f1->f_data == f2->f_data)));
     81 }
     82 
     83 void
     84 ktrderef(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(struct proc *p)
    104 {
    105 	struct file *fp = p->p_tracep;
    106 
    107 	fp->f_count++;
    108 }
    109 
    110 void
    111 ktrinitheader(struct ktr_header *kth, struct proc *p, int type)
    112 {
    113 
    114 	(void)memset(kth, 0, sizeof(*kth));
    115 	kth->ktr_type = type;
    116 	microtime(&kth->ktr_time);
    117 	kth->ktr_pid = p->p_pid;
    118 	memcpy(kth->ktr_comm, p->p_comm, MAXCOMLEN);
    119 }
    120 
    121 int
    122 ktrsyscall(struct proc *p, register_t code, register_t realcode,
    123     const struct sysent *callp, register_t args[])
    124 {
    125 	struct ktr_header kth;
    126 	struct ktr_syscall *ktp;
    127 	register_t *argp;
    128 	int argsize, error;
    129 	size_t len;
    130 	u_int i;
    131 
    132 	if (callp == NULL)
    133 		callp = p->p_emul->e_sysent;
    134 
    135 	argsize = callp[code].sy_argsize;
    136 #ifdef _LP64
    137 	if (p->p_flag & P_32)
    138 		argsize = argsize << 1;
    139 #endif
    140 	len = sizeof(struct ktr_syscall) + argsize;
    141 
    142 	p->p_traceflag |= KTRFAC_ACTIVE;
    143 	ktrinitheader(&kth, p, KTR_SYSCALL);
    144 	ktp = malloc(len, M_TEMP, M_WAITOK);
    145 	ktp->ktr_code = realcode;
    146 	ktp->ktr_argsize = argsize;
    147 	argp = (register_t *)((char *)ktp + sizeof(struct ktr_syscall));
    148 	for (i = 0; i < (argsize / sizeof(*argp)); i++)
    149 		*argp++ = args[i];
    150 	kth.ktr_buf = (caddr_t)ktp;
    151 	kth.ktr_len = len;
    152 	error = ktrwrite(p, &kth);
    153 	free(ktp, M_TEMP);
    154 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    155 	return error;
    156 }
    157 
    158 int
    159 ktrsysret(struct proc *p, register_t code, int error, register_t *retval)
    160 {
    161 	struct ktr_header kth;
    162 	struct ktr_sysret ktp;
    163 
    164 	p->p_traceflag |= KTRFAC_ACTIVE;
    165 	ktrinitheader(&kth, p, KTR_SYSRET);
    166 	ktp.ktr_code = code;
    167 	ktp.ktr_eosys = 0;			/* XXX unused */
    168 	ktp.ktr_error = error;
    169 	ktp.ktr_retval = retval ? retval[0] : 0;
    170 	ktp.ktr_retval_1 = retval ? retval[1] : 0;
    171 
    172 	kth.ktr_buf = (caddr_t)&ktp;
    173 	kth.ktr_len = sizeof(struct ktr_sysret);
    174 
    175 	error = ktrwrite(p, &kth);
    176 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    177 	return error;
    178 }
    179 
    180 int
    181 ktrnamei(struct proc *p, char *path)
    182 {
    183 	struct ktr_header kth;
    184 	int error;
    185 
    186 	p->p_traceflag |= KTRFAC_ACTIVE;
    187 	ktrinitheader(&kth, p, KTR_NAMEI);
    188 	kth.ktr_len = strlen(path);
    189 	kth.ktr_buf = path;
    190 
    191 	error = ktrwrite(p, &kth);
    192 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    193 	return error;
    194 }
    195 
    196 int
    197 ktremul(struct proc *p)
    198 {
    199 	struct ktr_header kth;
    200 	const char *emul = p->p_emul->e_name;
    201 	int error;
    202 
    203 	p->p_traceflag |= KTRFAC_ACTIVE;
    204 	ktrinitheader(&kth, p, KTR_EMUL);
    205 	kth.ktr_len = strlen(emul);
    206 	kth.ktr_buf = (caddr_t)emul;
    207 
    208 	error = ktrwrite(p, &kth);
    209 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    210 	return error;
    211 }
    212 
    213 int
    214 ktrkmem(struct proc *p, int ktr, const void *buf, size_t len)
    215 {
    216 	struct ktr_header kth;
    217 	int error;
    218 
    219 	p->p_traceflag |= KTRFAC_ACTIVE;
    220 	ktrinitheader(&kth, p, ktr);
    221 	kth.ktr_len = len;
    222 	kth.ktr_buf = buf;
    223 
    224 	error = ktrwrite(p, &kth);
    225 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    226 	return error;
    227 }
    228 
    229 int
    230 ktrgenio(struct proc *p, int fd, enum uio_rw rw, struct iovec *iov,
    231     int len, int error)
    232 {
    233 	struct ktr_header kth;
    234 	struct ktr_genio *ktp;
    235 	caddr_t cp;
    236 	int resid = len, cnt;
    237 	int buflen;
    238 
    239 	if (error)
    240 		return error;
    241 
    242 	p->p_traceflag |= KTRFAC_ACTIVE;
    243 
    244 	buflen = min(PAGE_SIZE, len + sizeof(struct ktr_genio));
    245 
    246 	ktrinitheader(&kth, p, KTR_GENIO);
    247 	ktp = malloc(buflen, M_TEMP, M_WAITOK);
    248 	ktp->ktr_fd = fd;
    249 	ktp->ktr_rw = rw;
    250 
    251 	kth.ktr_buf = (caddr_t)ktp;
    252 
    253 	cp = (caddr_t)((char *)ktp + sizeof(struct ktr_genio));
    254 	buflen -= sizeof(struct ktr_genio);
    255 
    256 	while (resid > 0) {
    257 #if 0 /* XXX NJWLWP */
    258 		KDASSERT(p->p_cpu != NULL);
    259 		KDASSERT(p->p_cpu == curcpu());
    260 #endif
    261 		/* XXX NJWLWP */
    262 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
    263 			preempt(1);
    264 
    265 		cnt = min(iov->iov_len, buflen);
    266 		if (cnt > resid)
    267 			cnt = resid;
    268 		if ((error = copyin(iov->iov_base, cp, cnt)) != 0)
    269 			break;
    270 
    271 		kth.ktr_len = cnt + sizeof(struct ktr_genio);
    272 
    273 		error = ktrwrite(p, &kth);
    274 		if (__predict_false(error != 0))
    275 			break;
    276 
    277 		iov->iov_base = (caddr_t)iov->iov_base + cnt;
    278 		iov->iov_len -= cnt;
    279 
    280 		if (iov->iov_len == 0)
    281 			iov++;
    282 
    283 		resid -= cnt;
    284 	}
    285 
    286 	free(ktp, M_TEMP);
    287 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    288 	return error;
    289 }
    290 
    291 int
    292 ktrpsig(struct proc *p, int sig, sig_t action, const sigset_t *mask,
    293     const ksiginfo_t *ksi)
    294 {
    295 	int error;
    296 
    297 	struct ktr_header kth;
    298 	struct {
    299 		struct ktr_psig	kp;
    300 		siginfo_t	si;
    301 	} kbuf;
    302 
    303 	p->p_traceflag |= KTRFAC_ACTIVE;
    304 	ktrinitheader(&kth, p, KTR_PSIG);
    305 	kbuf.kp.signo = (char)sig;
    306 	kbuf.kp.action = action;
    307 	kbuf.kp.mask = *mask;
    308 	kth.ktr_buf = (caddr_t)&kbuf;
    309 	if (ksi) {
    310 		kbuf.kp.code = KSI_TRAPCODE(ksi);
    311 		(void)memset(&kbuf.si, 0, sizeof(kbuf.si));
    312 		kbuf.si._info = ksi->ksi_info;
    313 		kth.ktr_len = sizeof(kbuf);
    314 	} else {
    315 		kbuf.kp.code = 0;
    316 		kth.ktr_len = sizeof(struct ktr_psig);
    317 	}
    318 	error = ktrwrite(p, &kth);
    319 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    320 	return error;
    321 }
    322 
    323 int
    324 ktrcsw(struct proc *p, int out, int user)
    325 {
    326 	struct ktr_header kth;
    327 	struct ktr_csw kc;
    328 	int error;
    329 
    330 	p->p_traceflag |= KTRFAC_ACTIVE;
    331 	ktrinitheader(&kth, p, KTR_CSW);
    332 	kc.out = out;
    333 	kc.user = user;
    334 	kth.ktr_buf = (caddr_t)&kc;
    335 	kth.ktr_len = sizeof(struct ktr_csw);
    336 
    337 	error = ktrwrite(p, &kth);
    338 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    339 	return error;
    340 }
    341 
    342 int
    343 ktruser(struct proc *p, const char *id, void *addr, size_t len, int ustr)
    344 {
    345 	struct ktr_header kth;
    346 	struct ktr_user *ktp;
    347 	caddr_t user_dta;
    348 	int error;
    349 
    350 	p->p_traceflag |= KTRFAC_ACTIVE;
    351 	ktrinitheader(&kth, p, KTR_USER);
    352 	ktp = malloc(sizeof(struct ktr_user) + len, M_TEMP, M_WAITOK);
    353 	if (ustr) {
    354 		if (copyinstr(id, ktp->ktr_id, KTR_USER_MAXIDLEN, NULL) != 0)
    355 			ktp->ktr_id[0] = '\0';
    356 	} else
    357 		strncpy(ktp->ktr_id, id, KTR_USER_MAXIDLEN);
    358 	ktp->ktr_id[KTR_USER_MAXIDLEN-1] = '\0';
    359 
    360 	user_dta = (caddr_t) ((char *)ktp + sizeof(struct ktr_user));
    361 	if (copyin(addr, (void *) user_dta, len) != 0)
    362 		len = 0;
    363 
    364 	kth.ktr_buf = (void *)ktp;
    365 	kth.ktr_len = sizeof(struct ktr_user) + len;
    366 	error = ktrwrite(p, &kth);
    367 
    368 	free(ktp, M_TEMP);
    369 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    370 	return error;
    371 
    372 }
    373 
    374 int
    375 ktrmmsg(struct proc *p, const void *msgh, size_t size)
    376 {
    377 	struct ktr_header kth;
    378 	struct ktr_mmsg	*kp;
    379 	int error;
    380 
    381 	p->p_traceflag |= KTRFAC_ACTIVE;
    382 	ktrinitheader(&kth, p, KTR_MMSG);
    383 
    384 	kp = (struct ktr_mmsg *)msgh;
    385 	kth.ktr_buf = (caddr_t)kp;
    386 	kth.ktr_len = size;
    387 	error = ktrwrite(p, &kth);
    388 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    389 	return error;
    390 }
    391 
    392 int
    393 ktrmool(struct proc *p, const void *kaddr, size_t size, const void *uaddr)
    394 {
    395 	struct ktr_header kth;
    396 	struct ktr_mool *kp;
    397 	struct ktr_mool *buf;
    398 	int error;
    399 
    400 	p->p_traceflag |= KTRFAC_ACTIVE;
    401 	ktrinitheader(&kth, p, KTR_MOOL);
    402 
    403 	kp = malloc(size + sizeof(*kp), M_TEMP, M_WAITOK);
    404 	kp->uaddr = uaddr;
    405 	kp->size = size;
    406 	buf = kp + 1; /* Skip uaddr and size */
    407 	(void)memcpy(buf, kaddr, size);
    408 
    409 	kth.ktr_buf = (caddr_t)kp;
    410 	kth.ktr_len = size + sizeof(*kp);
    411 	error = ktrwrite(p, &kth);
    412 	free(kp, M_TEMP);
    413 
    414 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    415 	return error;
    416 }
    417 
    418 
    419 /* Interface and common routines */
    420 
    421 int
    422 ktrace_common(struct proc *curp, int ops, int facs, int pid, struct file *fp)
    423 {
    424 	int ret = 0;
    425 	int error = 0;
    426 	int one = 1;
    427 	int descend;
    428 	struct proc *p;
    429 	struct pgrp *pg;
    430 
    431 	curp->p_traceflag |= KTRFAC_ACTIVE;
    432 	descend = ops & KTRFLAG_DESCEND;
    433 	facs = facs & ~((unsigned) KTRFAC_ROOT);
    434 
    435 	/*
    436 	 * Clear all uses of the tracefile
    437 	 */
    438 	if (KTROP(ops) == KTROP_CLEARFILE) {
    439 		proclist_lock_read();
    440 		LIST_FOREACH(p, &allproc, p_list) {
    441 			if (ktrsamefile(p->p_tracep, fp)) {
    442 				if (ktrcanset(curp, p))
    443 					ktrderef(p);
    444 				else
    445 					error = EPERM;
    446 			}
    447 		}
    448 		proclist_unlock_read();
    449 		goto done;
    450 	}
    451 
    452 	/*
    453 	 * Mark fp non-blocking, to avoid problems from possible deadlocks.
    454 	 */
    455 
    456 	if (fp != NULL) {
    457 		fp->f_flag |= FNONBLOCK;
    458 		(*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&one, curp);
    459 	}
    460 
    461 	/*
    462 	 * need something to (un)trace (XXX - why is this here?)
    463 	 */
    464 	if (!facs) {
    465 		error = EINVAL;
    466 		goto done;
    467 	}
    468 	/*
    469 	 * do it
    470 	 */
    471 	if (pid < 0) {
    472 		/*
    473 		 * by process group
    474 		 */
    475 		pg = pg_find(-pid, PFIND_UNLOCK_FAIL);
    476 		if (pg == NULL) {
    477 			error = ESRCH;
    478 			goto done;
    479 		}
    480 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
    481 			if (descend)
    482 				ret |= ktrsetchildren(curp, p, ops, facs, fp);
    483 			else
    484 				ret |= ktrops(curp, p, ops, facs, fp);
    485 		}
    486 
    487 	} else {
    488 		/*
    489 		 * by pid
    490 		 */
    491 		p = p_find(pid, PFIND_UNLOCK_FAIL);
    492 		if (p == NULL) {
    493 			error = ESRCH;
    494 			goto done;
    495 		}
    496 		if (descend)
    497 			ret |= ktrsetchildren(curp, p, ops, facs, fp);
    498 		else
    499 			ret |= ktrops(curp, p, ops, facs, fp);
    500 	}
    501 	proclist_unlock_read();	/* taken by p{g}_find */
    502 	if (!ret)
    503 		error = EPERM;
    504 done:
    505 	curp->p_traceflag &= ~KTRFAC_ACTIVE;
    506 	return (error);
    507 }
    508 
    509 /*
    510  * ktrace system call
    511  */
    512 /* ARGSUSED */
    513 int
    514 sys_fktrace(struct lwp *l, void *v, register_t *retval)
    515 {
    516 	struct sys_fktrace_args /* {
    517 		syscallarg(int) fd;
    518 		syscallarg(int) ops;
    519 		syscallarg(int) facs;
    520 		syscallarg(int) pid;
    521 	} */ *uap = v;
    522 	struct proc *curp = l->l_proc;
    523 	struct file *fp = NULL;
    524 	struct filedesc *fdp = curp->p_fd;
    525 	int error;
    526 
    527 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL)
    528 		return (EBADF);
    529 
    530 	FILE_USE(fp);
    531 
    532 	if ((fp->f_flag & FWRITE) == 0)
    533 		error = EBADF;
    534 	else
    535 		error = ktrace_common(curp, SCARG(uap, ops),
    536 		    SCARG(uap, facs), SCARG(uap, pid), fp);
    537 
    538 	FILE_UNUSE(fp, curp);
    539 
    540 	return error;
    541 }
    542 
    543 /*
    544  * ktrace system call
    545  */
    546 /* ARGSUSED */
    547 int
    548 sys_ktrace(struct lwp *l, void *v, register_t *retval)
    549 {
    550 	struct sys_ktrace_args /* {
    551 		syscallarg(const char *) fname;
    552 		syscallarg(int) ops;
    553 		syscallarg(int) facs;
    554 		syscallarg(int) pid;
    555 	} */ *uap = v;
    556 	struct proc *curp = l->l_proc;
    557 	struct vnode *vp = NULL;
    558 	struct file *fp = NULL;
    559 	int fd;
    560 	int ops = SCARG(uap, ops);
    561 	int error = 0;
    562 	struct nameidata nd;
    563 
    564 	ops = KTROP(ops) | (ops & KTRFLAG_DESCEND);
    565 
    566 	curp->p_traceflag |= KTRFAC_ACTIVE;
    567 	if ((ops & KTROP_CLEAR) == 0) {
    568 		/*
    569 		 * an operation which requires a file argument.
    570 		 */
    571 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, SCARG(uap, fname),
    572 		    curp);
    573 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
    574 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
    575 			return (error);
    576 		}
    577 		vp = nd.ni_vp;
    578 		VOP_UNLOCK(vp, 0);
    579 		if (vp->v_type != VREG) {
    580 			(void) vn_close(vp, FREAD|FWRITE, curp->p_ucred, curp);
    581 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
    582 			return (EACCES);
    583 		}
    584 		/*
    585 		 * XXX This uses up a file descriptor slot in the
    586 		 * tracing process for the duration of this syscall.
    587 		 * This is not expected to be a problem.  If
    588 		 * falloc(NULL, ...) DTRT we could skip that part, but
    589 		 * that would require changing its interface to allow
    590 		 * the caller to pass in a ucred..
    591 		 *
    592 		 * This will FILE_USE the fp it returns, if any.
    593 		 * Keep it in use until we return.
    594 		 */
    595 		if ((error = falloc(curp, &fp, &fd)) != 0)
    596 			goto done;
    597 
    598 		fp->f_flag = FWRITE|FAPPEND;
    599 		fp->f_type = DTYPE_VNODE;
    600 		fp->f_ops = &vnops;
    601 		fp->f_data = (caddr_t)vp;
    602 		FILE_SET_MATURE(fp);
    603 		vp = NULL;
    604 	}
    605 	error = ktrace_common(curp, SCARG(uap, ops), SCARG(uap, facs),
    606 	    SCARG(uap, pid), fp);
    607 done:
    608 	if (vp != NULL)
    609 		(void) vn_close(vp, FWRITE, curp->p_ucred, curp);
    610 	if (fp != NULL) {
    611 		FILE_UNUSE(fp, curp);	/* release file */
    612 		fdrelease(curp, fd); 	/* release fd table slot */
    613 	}
    614 	return (error);
    615 }
    616 
    617 int
    618 ktrops(struct proc *curp, struct proc *p, int ops, int facs,
    619     struct file *fp)
    620 {
    621 
    622 	if (!ktrcanset(curp, p))
    623 		return (0);
    624 	if (KTROP(ops) == KTROP_SET) {
    625 		if (p->p_tracep != fp) {
    626 			/*
    627 			 * if trace file already in use, relinquish
    628 			 */
    629 			ktrderef(p);
    630 			p->p_tracep = fp;
    631 			ktradref(p);
    632 		}
    633 		p->p_traceflag |= facs;
    634 		if (curp->p_ucred->cr_uid == 0)
    635 			p->p_traceflag |= KTRFAC_ROOT;
    636 	} else {
    637 		/* KTROP_CLEAR */
    638 		if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) {
    639 			/* no more tracing */
    640 			ktrderef(p);
    641 		}
    642 	}
    643 
    644 	/*
    645 	 * Emit an emulation record, every time there is a ktrace
    646 	 * change/attach request.
    647 	 */
    648 	if (KTRPOINT(p, KTR_EMUL))
    649 		p->p_traceflag |= KTRFAC_TRC_EMUL;
    650 #ifdef __HAVE_SYSCALL_INTERN
    651 	(*p->p_emul->e_syscall_intern)(p);
    652 #endif
    653 
    654 	return (1);
    655 }
    656 
    657 int
    658 ktrsetchildren(struct proc *curp, struct proc *top, int ops, int facs,
    659     struct file *fp)
    660 {
    661 	struct proc *p;
    662 	int ret = 0;
    663 
    664 	p = top;
    665 	for (;;) {
    666 		ret |= ktrops(curp, p, ops, facs, fp);
    667 		/*
    668 		 * If this process has children, descend to them next,
    669 		 * otherwise do any siblings, and if done with this level,
    670 		 * follow back up the tree (but not past top).
    671 		 */
    672 		if (LIST_FIRST(&p->p_children) != NULL) {
    673 			p = LIST_FIRST(&p->p_children);
    674 			continue;
    675 		}
    676 		for (;;) {
    677 			if (p == top)
    678 				return (ret);
    679 			if (LIST_NEXT(p, p_sibling) != NULL) {
    680 				p = LIST_NEXT(p, p_sibling);
    681 				break;
    682 			}
    683 			p = p->p_pptr;
    684 		}
    685 	}
    686 	/*NOTREACHED*/
    687 }
    688 
    689 int
    690 ktrwrite(struct proc *p, struct ktr_header *kth)
    691 {
    692 	struct uio auio;
    693 	struct iovec aiov[2];
    694 	int error, tries;
    695 	struct file *fp = p->p_tracep;
    696 
    697 	if (fp == NULL)
    698 		return 0;
    699 
    700 	if (p->p_traceflag & KTRFAC_TRC_EMUL) {
    701 		/* Add emulation trace before first entry for this process */
    702 		p->p_traceflag &= ~KTRFAC_TRC_EMUL;
    703 		if ((error = ktremul(p)) != 0)
    704 			return error;
    705 	}
    706 
    707 	auio.uio_iov = &aiov[0];
    708 	auio.uio_offset = 0;
    709 	auio.uio_segflg = UIO_SYSSPACE;
    710 	auio.uio_rw = UIO_WRITE;
    711 	aiov[0].iov_base = (caddr_t)kth;
    712 	aiov[0].iov_len = sizeof(struct ktr_header);
    713 	auio.uio_resid = sizeof(struct ktr_header);
    714 	auio.uio_iovcnt = 1;
    715 	auio.uio_procp = (struct proc *)0;
    716 	if (kth->ktr_len > 0) {
    717 		auio.uio_iovcnt++;
    718 		aiov[1].iov_base = (void *)kth->ktr_buf;
    719 		aiov[1].iov_len = kth->ktr_len;
    720 		auio.uio_resid += kth->ktr_len;
    721 	}
    722 
    723 	simple_lock(&fp->f_slock);
    724 	FILE_USE(fp);
    725 
    726 	tries = 0;
    727 	do {
    728 		error = (*fp->f_ops->fo_write)(fp, &fp->f_offset, &auio,
    729 		    fp->f_cred, FOF_UPDATE_OFFSET);
    730 		tries++;
    731 		if (error == EWOULDBLOCK)
    732 			preempt(1);
    733 	} while ((error == EWOULDBLOCK) && (tries < 3));
    734 	FILE_UNUSE(fp, NULL);
    735 
    736 	if (__predict_true(error == 0))
    737 		return (0);
    738 	/*
    739 	 * If error encountered, give up tracing on this vnode.  Don't report
    740 	 * EPIPE as this can easily happen with fktrace()/ktruss.
    741 	 */
    742 	if (error != EPIPE)
    743 		log(LOG_NOTICE,
    744 		    "ktrace write failed, errno %d, tracing stopped\n",
    745 		    error);
    746 	proclist_lock_read();
    747 	LIST_FOREACH(p, &allproc, p_list) {
    748 		if (ktrsamefile(p->p_tracep, fp))
    749 			ktrderef(p);
    750 	}
    751 	proclist_unlock_read();
    752 
    753 	return (error);
    754 }
    755 
    756 /*
    757  * Return true if caller has permission to set the ktracing state
    758  * of target.  Essentially, the target can't possess any
    759  * more permissions than the caller.  KTRFAC_ROOT signifies that
    760  * root previously set the tracing status on the target process, and
    761  * so, only root may further change it.
    762  *
    763  * TODO: check groups.  use caller effective gid.
    764  */
    765 int
    766 ktrcanset(struct proc *callp, struct proc *targetp)
    767 {
    768 	struct pcred *caller = callp->p_cred;
    769 	struct pcred *target = targetp->p_cred;
    770 
    771 	if ((caller->pc_ucred->cr_uid == target->p_ruid &&
    772 	    target->p_ruid == target->p_svuid &&
    773 	    caller->p_rgid == target->p_rgid &&	/* XXX */
    774 	    target->p_rgid == target->p_svgid &&
    775 	    (targetp->p_traceflag & KTRFAC_ROOT) == 0 &&
    776 	    (targetp->p_flag & P_SUGID) == 0) ||
    777 	    caller->pc_ucred->cr_uid == 0)
    778 		return (1);
    779 
    780 	return (0);
    781 }
    782 #endif /* KTRACE */
    783 
    784 /*
    785  * Put user defined entry to ktrace records.
    786  */
    787 int
    788 sys_utrace(struct lwp *l, void *v, register_t *retval)
    789 {
    790 #ifdef KTRACE
    791 	struct sys_utrace_args /* {
    792 		syscallarg(const char *) label;
    793 		syscallarg(void *) addr;
    794 		syscallarg(size_t) len;
    795 	} */ *uap = v;
    796 	struct proc *p = l->l_proc;
    797 
    798 	if (!KTRPOINT(p, KTR_USER))
    799 		return (0);
    800 
    801 	if (SCARG(uap, len) > KTR_USER_MAXLEN)
    802 		return (EINVAL);
    803 
    804 	ktruser(p, SCARG(uap, label), SCARG(uap, addr), SCARG(uap, len), 1);
    805 
    806 	return (0);
    807 #else /* !KTRACE */
    808 	return ENOSYS;
    809 #endif /* KTRACE */
    810 }
    811