Home | History | Annotate | Line # | Download | only in kern
kern_ktrace.c revision 1.53.2.9
      1 /*	$NetBSD: kern_ktrace.c,v 1.53.2.9 2002/12/11 06:43:04 thorpej 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. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by the University of
     18  *	California, Berkeley and its contributors.
     19  * 4. Neither the name of the University nor the names of its contributors
     20  *    may be used to endorse or promote products derived from this software
     21  *    without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33  * SUCH DAMAGE.
     34  *
     35  *	@(#)kern_ktrace.c	8.5 (Berkeley) 5/14/95
     36  */
     37 
     38 #include <sys/cdefs.h>
     39 __KERNEL_RCSID(0, "$NetBSD: kern_ktrace.c,v 1.53.2.9 2002/12/11 06:43:04 thorpej Exp $");
     40 
     41 #include "opt_ktrace.h"
     42 #include "opt_compat_mach.h"
     43 
     44 #include <sys/param.h>
     45 #include <sys/systm.h>
     46 #include <sys/proc.h>
     47 #include <sys/file.h>
     48 #include <sys/namei.h>
     49 #include <sys/vnode.h>
     50 #include <sys/ktrace.h>
     51 #include <sys/malloc.h>
     52 #include <sys/syslog.h>
     53 #include <sys/filedesc.h>
     54 #include <sys/ioctl.h>
     55 
     56 #include <sys/mount.h>
     57 #include <sys/sa.h>
     58 #include <sys/syscallargs.h>
     59 
     60 #ifdef COMPAT_MACH
     61 #include <compat/mach/mach_types.h>
     62 #include <compat/mach/mach_message.h>
     63 #endif
     64 
     65 #ifdef KTRACE
     66 
     67 int	ktrace_common(struct proc *, int, int, int, struct file *);
     68 void	ktrinitheader(struct ktr_header *, struct proc *, int);
     69 int	ktrops(struct proc *, struct proc *, int, int, struct file *);
     70 int	ktrsetchildren(struct proc *, struct proc *, int, int,
     71 	    struct file *);
     72 int	ktrwrite(struct proc *, struct ktr_header *);
     73 int	ktrcanset(struct proc *, struct proc *);
     74 int	ktrsamefile(struct file *, struct file *);
     75 
     76 /*
     77  * "deep" compare of two files for the purposes of clearing a trace.
     78  * Returns true if they're the same open file, or if they point at the
     79  * same underlying vnode/socket.
     80  */
     81 
     82 int
     83 ktrsamefile(f1, f2)
     84 	struct file *f1;
     85 	struct file *f2;
     86 {
     87 	return ((f1 == f2) ||
     88 	    ((f1 != NULL) && (f2 != NULL) &&
     89 		(f1->f_type == f2->f_type) &&
     90 		(f1->f_data == f2->f_data)));
     91 }
     92 
     93 void
     94 ktrderef(p)
     95 	struct proc *p;
     96 {
     97 	struct file *fp = p->p_tracep;
     98 	p->p_traceflag = 0;
     99 	if (fp == NULL)
    100 		return;
    101 	FILE_USE(fp);
    102 
    103 	/*
    104 	 * ktrace file descriptor can't be watched (are not visible to
    105 	 * userspace), so no kqueue stuff here
    106 	 */
    107 	closef(fp, NULL);
    108 
    109 	p->p_tracep = NULL;
    110 }
    111 
    112 void
    113 ktradref(p)
    114 	struct proc *p;
    115 {
    116 	struct file *fp = p->p_tracep;
    117 
    118 	fp->f_count++;
    119 }
    120 
    121 void
    122 ktrinitheader(kth, p, type)
    123 	struct ktr_header *kth;
    124 	struct proc *p;
    125 	int type;
    126 {
    127 
    128 	memset(kth, 0, sizeof(*kth));
    129 	kth->ktr_type = type;
    130 	microtime(&kth->ktr_time);
    131 	kth->ktr_pid = p->p_pid;
    132 	memcpy(kth->ktr_comm, p->p_comm, MAXCOMLEN);
    133 }
    134 
    135 void
    136 ktrsyscall(p, code, realcode, args)
    137 	struct proc *p;
    138 	register_t code;
    139 	register_t realcode;
    140 	register_t args[];
    141 {
    142 	struct ktr_header kth;
    143 	struct ktr_syscall *ktp;
    144 	register_t *argp;
    145 	int argsize;
    146 	size_t len;
    147 	u_int i;
    148 
    149 	argsize = p->p_emul->e_sysent[code].sy_narg * sizeof (register_t);
    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;		/* what about val2 ? */
    183 
    184 	kth.ktr_buf = (caddr_t)&ktp;
    185 	kth.ktr_len = sizeof(struct ktr_sysret);
    186 
    187 	(void) ktrwrite(p, &kth);
    188 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    189 }
    190 
    191 void
    192 ktrnamei(p, path)
    193 	struct proc *p;
    194 	char *path;
    195 {
    196 	struct ktr_header kth;
    197 
    198 	p->p_traceflag |= KTRFAC_ACTIVE;
    199 	ktrinitheader(&kth, p, KTR_NAMEI);
    200 	kth.ktr_len = strlen(path);
    201 	kth.ktr_buf = path;
    202 
    203 	(void) ktrwrite(p, &kth);
    204 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    205 }
    206 
    207 void
    208 ktremul(p)
    209 	struct proc *p;
    210 {
    211 	struct ktr_header kth;
    212 	const char *emul = p->p_emul->e_name;
    213 
    214 	p->p_traceflag |= KTRFAC_ACTIVE;
    215 	ktrinitheader(&kth, p, KTR_EMUL);
    216 	kth.ktr_len = strlen(emul);
    217 	kth.ktr_buf = (caddr_t)emul;
    218 
    219 	(void) ktrwrite(p, &kth);
    220 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    221 }
    222 
    223 void
    224 ktrgenio(p, fd, rw, iov, len, error)
    225 	struct proc *p;
    226 	int fd;
    227 	enum uio_rw rw;
    228 	struct iovec *iov;
    229 	int len;
    230 	int error;
    231 {
    232 	struct ktr_header kth;
    233 	struct ktr_genio *ktp;
    234 	caddr_t cp;
    235 	int resid = len, cnt;
    236 	int buflen;
    237 
    238 	if (error)
    239 		return;
    240 
    241 	p->p_traceflag |= KTRFAC_ACTIVE;
    242 
    243 	buflen = min(PAGE_SIZE, len + sizeof(struct ktr_genio));
    244 
    245 	ktrinitheader(&kth, p, KTR_GENIO);
    246 	ktp = malloc(buflen, M_TEMP, M_WAITOK);
    247 	ktp->ktr_fd = fd;
    248 	ktp->ktr_rw = rw;
    249 
    250 	kth.ktr_buf = (caddr_t)ktp;
    251 
    252 	cp = (caddr_t)((char *)ktp + sizeof(struct ktr_genio));
    253 	buflen -= sizeof(struct ktr_genio);
    254 
    255 	while (resid > 0) {
    256 #if 0 /* XXX NJWLWP */
    257 		KDASSERT(p->p_cpu != NULL);
    258 		KDASSERT(p->p_cpu == curcpu());
    259 #endif
    260 		/* XXX NJWLWP */
    261 		if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
    262 			preempt(NULL);
    263 
    264 		cnt = min(iov->iov_len, buflen);
    265 		if (cnt > resid)
    266 			cnt = resid;
    267 		if (copyin(iov->iov_base, cp, cnt))
    268 			break;
    269 
    270 		kth.ktr_len = cnt + sizeof(struct ktr_genio);
    271 
    272 		if (__predict_false(ktrwrite(p, &kth) != 0))
    273 			break;
    274 
    275 		iov->iov_base = (caddr_t)iov->iov_base + cnt;
    276 		iov->iov_len -= cnt;
    277 
    278 		if (iov->iov_len == 0)
    279 			iov++;
    280 
    281 		resid -= cnt;
    282 	}
    283 
    284 	free(ktp, M_TEMP);
    285 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    286 }
    287 
    288 void
    289 ktrpsig(p, sig, action, mask, code)
    290 	struct proc *p;
    291 	int sig;
    292 	sig_t action;
    293 	sigset_t *mask;
    294 	int code;
    295 {
    296 	struct ktr_header kth;
    297 	struct ktr_psig	kp;
    298 
    299 	p->p_traceflag |= KTRFAC_ACTIVE;
    300 	ktrinitheader(&kth, p, KTR_PSIG);
    301 	kp.signo = (char)sig;
    302 	kp.action = action;
    303 	kp.mask = *mask;
    304 	kp.code = code;
    305 	kth.ktr_buf = (caddr_t)&kp;
    306 	kth.ktr_len = sizeof(struct ktr_psig);
    307 
    308 	(void) ktrwrite(p, &kth);
    309 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    310 }
    311 
    312 void
    313 ktrcsw(p, out, user)
    314 	struct proc *p;
    315 	int out;
    316 	int user;
    317 {
    318 	struct ktr_header kth;
    319 	struct ktr_csw kc;
    320 
    321 	p->p_traceflag |= KTRFAC_ACTIVE;
    322 	ktrinitheader(&kth, p, KTR_CSW);
    323 	kc.out = out;
    324 	kc.user = user;
    325 	kth.ktr_buf = (caddr_t)&kc;
    326 	kth.ktr_len = sizeof(struct ktr_csw);
    327 
    328 	(void) ktrwrite(p, &kth);
    329 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    330 }
    331 
    332 void
    333 ktruser(p, id, addr, len, ustr)
    334 	struct proc *p;
    335 	const char *id;
    336 	void *addr;
    337 	size_t len;
    338 	int ustr;
    339 {
    340 	struct ktr_header kth;
    341 	struct ktr_user *ktp;
    342 	caddr_t user_dta;
    343 
    344 	p->p_traceflag |= KTRFAC_ACTIVE;
    345 	ktrinitheader(&kth, p, KTR_USER);
    346 	ktp = malloc(sizeof(struct ktr_user) + len, M_TEMP, M_WAITOK);
    347 	if (ustr) {
    348 		if (copyinstr(id, ktp->ktr_id, KTR_USER_MAXIDLEN, NULL) != 0)
    349 			ktp->ktr_id[0] = '\0';
    350 	} else
    351 		strncpy(ktp->ktr_id, id, KTR_USER_MAXIDLEN);
    352 	ktp->ktr_id[KTR_USER_MAXIDLEN-1] = '\0';
    353 
    354 	user_dta = (caddr_t) ((char *)ktp + sizeof(struct ktr_user));
    355 	if (copyin(addr, (void *) user_dta, len) != 0)
    356 		len = 0;
    357 
    358 	kth.ktr_buf = (void *)ktp;
    359 	kth.ktr_len = sizeof(struct ktr_user) + len;
    360 	(void) ktrwrite(p, &kth);
    361 
    362 	free(ktp, M_TEMP);
    363 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    364 
    365 }
    366 
    367 #ifdef COMPAT_MACH
    368 void
    369 ktrmmsg(p, msgh, size)
    370 	struct proc *p;
    371 	const char *msgh;
    372 	size_t size;
    373 {
    374 	struct ktr_header kth;
    375 	struct ktr_mmsg	*kp;
    376 	int error;
    377 
    378 	p->p_traceflag |= KTRFAC_ACTIVE;
    379 	ktrinitheader(&kth, p, KTR_MMSG);
    380 
    381 	kp = (struct ktr_mmsg *)malloc(size, M_TEMP, M_WAITOK);
    382 	if ((error = copyin(msgh, kp, size)) != 0)
    383 		size = 0; /* Still log a message, but empty */
    384 
    385 	kth.ktr_buf = (caddr_t)kp;
    386 	kth.ktr_len = size;
    387 	(void) ktrwrite(p, &kth);
    388 	free(kp, M_TEMP);
    389 	p->p_traceflag &= ~KTRFAC_ACTIVE;
    390 }
    391 #endif /* COMPAT_MACH */
    392 
    393 /* Interface and common routines */
    394 
    395 int
    396 ktrace_common(curp, ops, facs, pid, fp)
    397 	struct proc *curp;
    398 	int ops;
    399 	int facs;
    400 	int pid;
    401 	struct file *fp;
    402 {
    403 	int ret = 0;
    404 	int error = 0;
    405 	int one = 1;
    406 	int descend;
    407 	struct proc *p;
    408 	struct pgrp *pg;
    409 
    410 	curp->p_traceflag |= KTRFAC_ACTIVE;
    411 	descend = ops & KTRFLAG_DESCEND;
    412 	facs = facs & ~((unsigned) KTRFAC_ROOT);
    413 
    414 	/*
    415 	 * Clear all uses of the tracefile
    416 	 */
    417 	if (KTROP(ops) == KTROP_CLEARFILE) {
    418 		proclist_lock_read();
    419 		for (p = LIST_FIRST(&allproc); p != NULL;
    420 		     p = LIST_NEXT(p, p_list)) {
    421 			if (ktrsamefile(p->p_tracep, fp)) {
    422 				if (ktrcanset(curp, p))
    423 					ktrderef(p);
    424 				else
    425 					error = EPERM;
    426 			}
    427 		}
    428 		proclist_unlock_read();
    429 		goto done;
    430 	}
    431 
    432 	/*
    433 	 * Mark fp non-blocking, to avoid problems from possible deadlocks.
    434 	 */
    435 
    436 	if (fp != NULL) {
    437 		fp->f_flag |= FNONBLOCK;
    438 		(*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&one, curp);
    439 	}
    440 
    441 	/*
    442 	 * need something to (un)trace (XXX - why is this here?)
    443 	 */
    444 	if (!facs) {
    445 		error = EINVAL;
    446 		goto done;
    447 	}
    448 	/*
    449 	 * do it
    450 	 */
    451 	if (pid < 0) {
    452 		/*
    453 		 * by process group
    454 		 */
    455 		pg = pgfind(-pid);
    456 		if (pg == NULL) {
    457 			error = ESRCH;
    458 			goto done;
    459 		}
    460 		for (p = LIST_FIRST(&pg->pg_members); p != NULL;
    461 		     p = LIST_NEXT(p, p_pglist)) {
    462 			if (descend)
    463 				ret |= ktrsetchildren(curp, p, ops, facs, fp);
    464 			else
    465 				ret |= ktrops(curp, p, ops, facs, fp);
    466 		}
    467 
    468 	} else {
    469 		/*
    470 		 * by pid
    471 		 */
    472 		p = pfind(pid);
    473 		if (p == NULL) {
    474 			error = ESRCH;
    475 			goto done;
    476 		}
    477 		if (descend)
    478 			ret |= ktrsetchildren(curp, p, ops, facs, fp);
    479 		else
    480 			ret |= ktrops(curp, p, ops, facs, fp);
    481 	}
    482 	if (!ret)
    483 		error = EPERM;
    484 done:
    485 	curp->p_traceflag &= ~KTRFAC_ACTIVE;
    486 	return (error);
    487 }
    488 
    489 /*
    490  * ktrace system call
    491  */
    492 /* ARGSUSED */
    493 int
    494 sys_fktrace(l, v, retval)
    495 	struct lwp *l;
    496 	void *v;
    497 	register_t *retval;
    498 {
    499 	struct sys_fktrace_args /* {
    500 		syscallarg(int) fd;
    501 		syscallarg(int) ops;
    502 		syscallarg(int) facs;
    503 		syscallarg(int) pid;
    504 	} */ *uap = v;
    505 	struct proc *curp = l->l_proc;
    506 	struct file *fp = NULL;
    507 	struct filedesc *fdp = curp->p_fd;
    508 
    509 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL)
    510 		return (EBADF);
    511 
    512 	if ((fp->f_flag & FWRITE) == 0)
    513 		return (EBADF);
    514 
    515 	return ktrace_common(curp, SCARG(uap, ops),
    516 	    SCARG(uap, facs), SCARG(uap, pid), fp);
    517 }
    518 
    519 /*
    520  * ktrace system call
    521  */
    522 /* ARGSUSED */
    523 int
    524 sys_ktrace(l, v, retval)
    525 	struct lwp *l;
    526 	void *v;
    527 	register_t *retval;
    528 {
    529 	struct sys_ktrace_args /* {
    530 		syscallarg(const char *) fname;
    531 		syscallarg(int) ops;
    532 		syscallarg(int) facs;
    533 		syscallarg(int) pid;
    534 	} */ *uap = v;
    535 	struct proc *curp = l->l_proc;
    536 	struct vnode *vp = NULL;
    537 	struct file *fp = NULL;
    538 	int fd;
    539 	int ops = SCARG(uap, ops);
    540 	int error = 0;
    541 	struct nameidata nd;
    542 
    543 	ops = KTROP(ops) | (ops & KTRFLAG_DESCEND);
    544 
    545 	curp->p_traceflag |= KTRFAC_ACTIVE;
    546 	if (ops != KTROP_CLEAR) {
    547 		/*
    548 		 * an operation which requires a file argument.
    549 		 */
    550 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, SCARG(uap, fname),
    551 		    curp);
    552 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
    553 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
    554 			return (error);
    555 		}
    556 		vp = nd.ni_vp;
    557 		VOP_UNLOCK(vp, 0);
    558 		if (vp->v_type != VREG) {
    559 			(void) vn_close(vp, FREAD|FWRITE, curp->p_ucred, curp);
    560 			curp->p_traceflag &= ~KTRFAC_ACTIVE;
    561 			return (EACCES);
    562 		}
    563 		/*
    564 		 * XXX This uses up a file descriptor slot in the
    565 		 * tracing process for the duration of this syscall.
    566 		 * This is not expected to be a problem.  If
    567 		 * falloc(NULL, ...) DTRT we could skip that part, but
    568 		 * that would require changing its interface to allow
    569 		 * the caller to pass in a ucred..
    570 		 *
    571 		 * This will FILE_USE the fp it returns, if any.
    572 		 * Keep it in use until we return.
    573 		 */
    574 		if ((error = falloc(curp, &fp, &fd)) != 0)
    575 			goto done;
    576 
    577 		fp->f_flag = FWRITE|FAPPEND;
    578 		fp->f_type = DTYPE_VNODE;
    579 		fp->f_ops = &vnops;
    580 		fp->f_data = (caddr_t)vp;
    581 		FILE_SET_MATURE(fp);
    582 		vp = NULL;
    583 	}
    584 	error = ktrace_common(curp, SCARG(uap, ops), SCARG(uap, facs),
    585 	    SCARG(uap, pid), fp);
    586 done:
    587 	if (vp != NULL)
    588 		(void) vn_close(vp, FWRITE, curp->p_ucred, curp);
    589 	if (fp != NULL) {
    590 		FILE_UNUSE(fp, curp);	/* release file */
    591 		fdrelease(curp, fd); 	/* release fd table slot */
    592 	}
    593 	return (error);
    594 }
    595 
    596 int
    597 ktrops(curp, p, ops, facs, fp)
    598 	struct proc *curp;
    599 	struct proc *p;
    600 	int ops;
    601 	int facs;
    602 	struct file *fp;
    603 {
    604 
    605 	if (!ktrcanset(curp, p))
    606 		return (0);
    607 	if (KTROP(ops) == KTROP_SET) {
    608 		if (p->p_tracep != fp) {
    609 			/*
    610 			 * if trace file already in use, relinquish
    611 			 */
    612 			ktrderef(p);
    613 			p->p_tracep = fp;
    614 			ktradref(p);
    615 		}
    616 		p->p_traceflag |= facs;
    617 		if (curp->p_ucred->cr_uid == 0)
    618 			p->p_traceflag |= KTRFAC_ROOT;
    619 	} else {
    620 		/* KTROP_CLEAR */
    621 		if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) {
    622 			/* no more tracing */
    623 			ktrderef(p);
    624 		}
    625 	}
    626 
    627 	/*
    628 	 * Emit an emulation record, every time there is a ktrace
    629 	 * change/attach request.
    630 	 */
    631 	if (KTRPOINT(p, KTR_EMUL))
    632 		ktremul(p);
    633 #ifdef __HAVE_SYSCALL_INTERN
    634 	(*p->p_emul->e_syscall_intern)(p);
    635 #endif
    636 
    637 	return (1);
    638 }
    639 
    640 int
    641 ktrsetchildren(curp, top, ops, facs, fp)
    642 	struct proc *curp;
    643 	struct proc *top;
    644 	int ops;
    645 	int facs;
    646 	struct file *fp;
    647 {
    648 	struct proc *p;
    649 	int ret = 0;
    650 
    651 	p = top;
    652 	for (;;) {
    653 		ret |= ktrops(curp, p, ops, facs, fp);
    654 		/*
    655 		 * If this process has children, descend to them next,
    656 		 * otherwise do any siblings, and if done with this level,
    657 		 * follow back up the tree (but not past top).
    658 		 */
    659 		if (LIST_FIRST(&p->p_children) != NULL)
    660 			p = LIST_FIRST(&p->p_children);
    661 		else for (;;) {
    662 			if (p == top)
    663 				return (ret);
    664 			if (LIST_NEXT(p, p_sibling) != NULL) {
    665 				p = LIST_NEXT(p, p_sibling);
    666 				break;
    667 			}
    668 			p = p->p_pptr;
    669 		}
    670 	}
    671 	/*NOTREACHED*/
    672 }
    673 
    674 int
    675 ktrwrite(p, kth)
    676 	struct proc *p;
    677 	struct ktr_header *kth;
    678 {
    679 	struct uio auio;
    680 	struct iovec aiov[2];
    681 	int error, tries;
    682 	struct file *fp = p->p_tracep;
    683 
    684 	if (fp == NULL)
    685 		return 0;
    686 
    687 	auio.uio_iov = &aiov[0];
    688 	auio.uio_offset = 0;
    689 	auio.uio_segflg = UIO_SYSSPACE;
    690 	auio.uio_rw = UIO_WRITE;
    691 	aiov[0].iov_base = (caddr_t)kth;
    692 	aiov[0].iov_len = sizeof(struct ktr_header);
    693 	auio.uio_resid = sizeof(struct ktr_header);
    694 	auio.uio_iovcnt = 1;
    695 	auio.uio_procp = (struct proc *)0;
    696 	if (kth->ktr_len > 0) {
    697 		auio.uio_iovcnt++;
    698 		aiov[1].iov_base = kth->ktr_buf;
    699 		aiov[1].iov_len = kth->ktr_len;
    700 		auio.uio_resid += kth->ktr_len;
    701 	}
    702 
    703 	FILE_USE(fp);
    704 
    705 	tries = 0;
    706 	do {
    707 		error = (*fp->f_ops->fo_write)(fp, &fp->f_offset, &auio,
    708 		    fp->f_cred, FOF_UPDATE_OFFSET);
    709 		tries++;
    710 		if (error == EWOULDBLOCK)
    711 		  	yield();
    712 	} while ((error == EWOULDBLOCK) && (tries < 3));
    713 	FILE_UNUSE(fp, NULL);
    714 
    715 	if (__predict_true(error == 0))
    716 		return (0);
    717 	/*
    718 	 * If error encountered, give up tracing on this vnode.  Don't report
    719 	 * EPIPE as this can easily happen with fktrace()/ktruss.
    720 	 */
    721 	if (error != EPIPE)
    722 		log(LOG_NOTICE,
    723 		    "ktrace write failed, errno %d, tracing stopped\n",
    724 		    error);
    725 	proclist_lock_read();
    726 	for (p = LIST_FIRST(&allproc); p != NULL; p = LIST_NEXT(p, p_list)) {
    727 		if (ktrsamefile(p->p_tracep, fp))
    728 			ktrderef(p);
    729 	}
    730 	proclist_unlock_read();
    731 
    732 	return (error);
    733 }
    734 
    735 /*
    736  * Return true if caller has permission to set the ktracing state
    737  * of target.  Essentially, the target can't possess any
    738  * more permissions than the caller.  KTRFAC_ROOT signifies that
    739  * root previously set the tracing status on the target process, and
    740  * so, only root may further change it.
    741  *
    742  * TODO: check groups.  use caller effective gid.
    743  */
    744 int
    745 ktrcanset(callp, targetp)
    746 	struct proc *callp;
    747 	struct proc *targetp;
    748 {
    749 	struct pcred *caller = callp->p_cred;
    750 	struct pcred *target = targetp->p_cred;
    751 
    752 	if ((caller->pc_ucred->cr_uid == target->p_ruid &&
    753 	     target->p_ruid == target->p_svuid &&
    754 	     caller->p_rgid == target->p_rgid &&	/* XXX */
    755 	     target->p_rgid == target->p_svgid &&
    756 	     (targetp->p_traceflag & KTRFAC_ROOT) == 0 &&
    757 	     (targetp->p_flag & P_SUGID) == 0) ||
    758 	     caller->pc_ucred->cr_uid == 0)
    759 		return (1);
    760 
    761 	return (0);
    762 }
    763 #endif /* KTRACE */
    764 
    765 /*
    766  * Put user defined entry to ktrace records.
    767  */
    768 int
    769 sys_utrace(l, v, retval)
    770 	struct lwp *l;
    771 	void *v;
    772 	register_t *retval;
    773 {
    774 #ifdef KTRACE
    775 	struct sys_utrace_args /* {
    776 		syscallarg(const char *) label;
    777 		syscallarg(void *) addr;
    778 		syscallarg(size_t) len;
    779 	} */ *uap = v;
    780 	struct proc *p = l->l_proc;
    781 	if (!KTRPOINT(p, KTR_USER))
    782 		return (0);
    783 
    784 	if (SCARG(uap, len) > KTR_USER_MAXLEN)
    785 		return (EINVAL);
    786 
    787 	ktruser(p, SCARG(uap, label), SCARG(uap, addr), SCARG(uap, len), 1);
    788 
    789 	return (0);
    790 #else /* !KTRACE */
    791 	return ENOSYS;
    792 #endif /* KTRACE */
    793 }
    794