Home | History | Annotate | Line # | Download | only in kern
vfs_lookup.c revision 1.65.2.1
      1 /*	$NetBSD: vfs_lookup.c,v 1.65.2.1 2005/12/31 11:14:01 yamt Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1989, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  * (c) UNIX System Laboratories, Inc.
      7  * All or some portions of this file are derived from material licensed
      8  * to the University of California by American Telephone and Telegraph
      9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     10  * the permission of UNIX System Laboratories, Inc.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. Neither the name of the University nor the names of its contributors
     21  *    may be used to endorse or promote products derived from this software
     22  *    without specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  * SUCH DAMAGE.
     35  *
     36  *	@(#)vfs_lookup.c	8.10 (Berkeley) 5/27/95
     37  */
     38 
     39 #include <sys/cdefs.h>
     40 __KERNEL_RCSID(0, "$NetBSD: vfs_lookup.c,v 1.65.2.1 2005/12/31 11:14:01 yamt Exp $");
     41 
     42 #include "opt_ktrace.h"
     43 #include "opt_systrace.h"
     44 
     45 #include <sys/param.h>
     46 #include <sys/systm.h>
     47 #include <sys/kernel.h>
     48 #include <sys/syslimits.h>
     49 #include <sys/time.h>
     50 #include <sys/namei.h>
     51 #include <sys/vnode.h>
     52 #include <sys/mount.h>
     53 #include <sys/errno.h>
     54 #include <sys/filedesc.h>
     55 #include <sys/hash.h>
     56 #include <sys/malloc.h>
     57 #include <sys/proc.h>
     58 #include <sys/syslog.h>
     59 
     60 #ifdef KTRACE
     61 #include <sys/ktrace.h>
     62 #endif
     63 #ifdef SYSTRACE
     64 #include <sys/systrace.h>
     65 #endif
     66 
     67 struct pool pnbuf_pool;		/* pathname buffer pool */
     68 struct pool_cache pnbuf_cache;	/* pathname buffer cache */
     69 
     70 /*
     71  * Substitute replacement text for 'magic' strings in symlinks.
     72  * Returns 0 if successful, and returns non-zero if an error
     73  * occurs.  (Currently, the only possible error is running out
     74  * of temporary pathname space.)
     75  *
     76  * Looks for "@<string>" and "@<string>/", where <string> is a
     77  * recognized 'magic' string.  Replaces the "@<string>" with the
     78  * appropriate replacement text.  (Note that in some cases the
     79  * replacement text may have zero length.)
     80  *
     81  * This would have been table driven, but the variance in
     82  * replacement strings (and replacement string lengths) made
     83  * that impractical.
     84  */
     85 #define	VNL(x)							\
     86 	(sizeof(x) - 1)
     87 
     88 #define	VO	'{'
     89 #define	VC	'}'
     90 
     91 #define	MATCH(str)						\
     92 	((termchar == '/' && i + VNL(str) == *len) ||		\
     93 	 (i + VNL(str) < *len &&				\
     94 	  cp[i + VNL(str)] == termchar)) &&			\
     95 	!strncmp((str), &cp[i], VNL(str))
     96 
     97 #define	SUBSTITUTE(m, s, sl)					\
     98 	if ((newlen + (sl)) > MAXPATHLEN)			\
     99 		return (1);					\
    100 	i += VNL(m);						\
    101 	if (termchar != '/')					\
    102 		i++;						\
    103 	memcpy(&tmp[newlen], (s), (sl));			\
    104 	newlen += (sl);						\
    105 	change = 1;						\
    106 	termchar = '/';
    107 
    108 static int
    109 symlink_magic(struct proc *p, char *cp, int *len)
    110 {
    111 	char tmp[MAXPATHLEN];
    112 	int change, i, newlen;
    113 	int termchar = '/';
    114 
    115 	for (change = i = newlen = 0; i < *len; ) {
    116 		if (cp[i] != '@') {
    117 			tmp[newlen++] = cp[i++];
    118 			continue;
    119 		}
    120 
    121 		i++;
    122 
    123 		/* Check for @{var} syntax. */
    124 		if (cp[i] == VO) {
    125 			termchar = VC;
    126 			i++;
    127 		}
    128 
    129 		/*
    130 		 * The following checks should be ordered according
    131 		 * to frequency of use.
    132 		 */
    133 		if (MATCH("machine_arch")) {
    134 			SUBSTITUTE("machine_arch", MACHINE_ARCH,
    135 			    sizeof(MACHINE_ARCH) - 1);
    136 		} else if (MATCH("machine")) {
    137 			SUBSTITUTE("machine", MACHINE,
    138 			    sizeof(MACHINE) - 1);
    139 		} else if (MATCH("hostname")) {
    140 			SUBSTITUTE("hostname", hostname,
    141 			    hostnamelen);
    142 		} else if (MATCH("osrelease")) {
    143 			SUBSTITUTE("osrelease", osrelease,
    144 			    strlen(osrelease));
    145 		} else if (MATCH("emul")) {
    146 			SUBSTITUTE("emul", p->p_emul->e_name,
    147 			    strlen(p->p_emul->e_name));
    148 		} else if (MATCH("kernel_ident")) {
    149 			SUBSTITUTE("kernel_ident", kernel_ident,
    150 			    strlen(kernel_ident));
    151 		} else if (MATCH("domainname")) {
    152 			SUBSTITUTE("domainname", domainname,
    153 			    domainnamelen);
    154 		} else if (MATCH("ostype")) {
    155 			SUBSTITUTE("ostype", ostype,
    156 			    strlen(ostype));
    157 		} else {
    158 			tmp[newlen++] = '@';
    159 			if (termchar == VC)
    160 				tmp[newlen++] = VO;
    161 		}
    162 	}
    163 
    164 	if (! change)
    165 		return (0);
    166 
    167 	memcpy(cp, tmp, newlen);
    168 	*len = newlen;
    169 
    170 	return (0);
    171 }
    172 
    173 #undef VNL
    174 #undef VO
    175 #undef VC
    176 #undef MATCH
    177 #undef SUBSTITUTE
    178 
    179 /*
    180  * Convert a pathname into a pointer to a locked inode.
    181  *
    182  * The FOLLOW flag is set when symbolic links are to be followed
    183  * when they occur at the end of the name translation process.
    184  * Symbolic links are always followed for all other pathname
    185  * components other than the last.
    186  *
    187  * The segflg defines whether the name is to be copied from user
    188  * space or kernel space.
    189  *
    190  * Overall outline of namei:
    191  *
    192  *	copy in name
    193  *	get starting directory
    194  *	while (!done && !error) {
    195  *		call lookup to search path.
    196  *		if symbolic link, massage name in buffer and continue
    197  *	}
    198  */
    199 int
    200 namei(struct nameidata *ndp)
    201 {
    202 	struct cwdinfo *cwdi;		/* pointer to cwd state */
    203 	char *cp;			/* pointer into pathname argument */
    204 	struct vnode *dp;		/* the directory we are searching */
    205 	struct iovec aiov;		/* uio for reading symbolic links */
    206 	struct uio auio;
    207 	int error, linklen;
    208 	struct componentname *cnp = &ndp->ni_cnd;
    209 
    210 #ifdef DIAGNOSTIC
    211 	if (!cnp->cn_cred || !cnp->cn_lwp)
    212 		panic("namei: bad cred/proc");
    213 	if (cnp->cn_nameiop & (~OPMASK))
    214 		panic("namei: nameiop contaminated with flags");
    215 	if (cnp->cn_flags & OPMASK)
    216 		panic("namei: flags contaminated with nameiops");
    217 #endif
    218 	cwdi = cnp->cn_lwp->l_proc->p_cwdi;
    219 
    220 	/*
    221 	 * Get a buffer for the name to be translated, and copy the
    222 	 * name into the buffer.
    223 	 */
    224 	if ((cnp->cn_flags & HASBUF) == 0)
    225 		cnp->cn_pnbuf = PNBUF_GET();
    226 	if (ndp->ni_segflg == UIO_SYSSPACE)
    227 		error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
    228 			    MAXPATHLEN, &ndp->ni_pathlen);
    229 	else
    230 		error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
    231 			    MAXPATHLEN, &ndp->ni_pathlen);
    232 
    233 	/*
    234 	 * POSIX.1 requirement: "" is not a valid file name.
    235 	 */
    236 	if (!error && ndp->ni_pathlen == 1)
    237 		error = ENOENT;
    238 
    239 	if (error) {
    240 		PNBUF_PUT(cnp->cn_pnbuf);
    241 		ndp->ni_vp = NULL;
    242 		return (error);
    243 	}
    244 	ndp->ni_loopcnt = 0;
    245 
    246 #ifdef KTRACE
    247 	if (KTRPOINT(cnp->cn_lwp->l_proc, KTR_NAMEI))
    248 		ktrnamei(cnp->cn_lwp, cnp->cn_pnbuf);
    249 #endif
    250 #ifdef SYSTRACE
    251 	if (ISSET(cnp->cn_lwp->l_proc->p_flag, P_SYSTRACE))
    252 		systrace_namei(ndp);
    253 #endif
    254 
    255 	/*
    256 	 * Get starting point for the translation.
    257 	 */
    258 	if ((ndp->ni_rootdir = cwdi->cwdi_rdir) == NULL)
    259 		ndp->ni_rootdir = rootvnode;
    260 	/*
    261 	 * Check if starting from root directory or current directory.
    262 	 */
    263 	if (cnp->cn_pnbuf[0] == '/') {
    264 		dp = ndp->ni_rootdir;
    265 		VREF(dp);
    266 	} else {
    267 		dp = cwdi->cwdi_cdir;
    268 		VREF(dp);
    269 	}
    270 	for (;;) {
    271 		if (!dp->v_mount)
    272 		{
    273 			/* Give up if the directory is no longer mounted */
    274 			PNBUF_PUT(cnp->cn_pnbuf);
    275 			return (ENOENT);
    276 		}
    277 		cnp->cn_nameptr = cnp->cn_pnbuf;
    278 		ndp->ni_startdir = dp;
    279 		if ((error = lookup(ndp)) != 0) {
    280 			PNBUF_PUT(cnp->cn_pnbuf);
    281 			return (error);
    282 		}
    283 		/*
    284 		 * Check for symbolic link
    285 		 */
    286 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
    287 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
    288 				PNBUF_PUT(cnp->cn_pnbuf);
    289 			else
    290 				cnp->cn_flags |= HASBUF;
    291 			return (0);
    292 		}
    293 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
    294 			VOP_UNLOCK(ndp->ni_dvp, 0);
    295 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
    296 			error = ELOOP;
    297 			break;
    298 		}
    299 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
    300 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
    301 			    cnp->cn_lwp);
    302 			if (error != 0)
    303 				break;
    304 		}
    305 		if (ndp->ni_pathlen > 1)
    306 			cp = PNBUF_GET();
    307 		else
    308 			cp = cnp->cn_pnbuf;
    309 		aiov.iov_base = cp;
    310 		aiov.iov_len = MAXPATHLEN;
    311 		auio.uio_iov = &aiov;
    312 		auio.uio_iovcnt = 1;
    313 		auio.uio_offset = 0;
    314 		auio.uio_rw = UIO_READ;
    315 		auio.uio_resid = MAXPATHLEN;
    316 		UIO_SETUP_SYSSPACE(&auio);
    317 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
    318 		if (error) {
    319 		badlink:
    320 			if (ndp->ni_pathlen > 1)
    321 				PNBUF_PUT(cp);
    322 			break;
    323 		}
    324 		linklen = MAXPATHLEN - auio.uio_resid;
    325 		if (linklen == 0) {
    326 			error = ENOENT;
    327 			goto badlink;
    328 		}
    329 		/*
    330 		 * Do symlink substitution, if appropriate, and
    331 		 * check length for potential overflow.
    332 		 */
    333 		if (((ndp->ni_vp->v_mount->mnt_flag & MNT_MAGICLINKS) &&
    334 		     symlink_magic(cnp->cn_lwp->l_proc, cp, &linklen)) ||
    335 		    (linklen + ndp->ni_pathlen >= MAXPATHLEN)) {
    336 			error = ENAMETOOLONG;
    337 			goto badlink;
    338 		}
    339 		if (ndp->ni_pathlen > 1) {
    340 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
    341 			PNBUF_PUT(cnp->cn_pnbuf);
    342 			cnp->cn_pnbuf = cp;
    343 		} else
    344 			cnp->cn_pnbuf[linklen] = '\0';
    345 		ndp->ni_pathlen += linklen;
    346 		vput(ndp->ni_vp);
    347 		dp = ndp->ni_dvp;
    348 		/*
    349 		 * Check if root directory should replace current directory.
    350 		 */
    351 		if (cnp->cn_pnbuf[0] == '/') {
    352 			vrele(dp);
    353 			dp = ndp->ni_rootdir;
    354 			VREF(dp);
    355 		}
    356 	}
    357 	PNBUF_PUT(cnp->cn_pnbuf);
    358 	vrele(ndp->ni_dvp);
    359 	vput(ndp->ni_vp);
    360 	ndp->ni_vp = NULL;
    361 	return (error);
    362 }
    363 
    364 /*
    365  * Determine the namei hash (for cn_hash) for name.
    366  * If *ep != NULL, hash from name to ep-1.
    367  * If *ep == NULL, hash from name until the first NUL or '/', and
    368  * return the location of this termination character in *ep.
    369  *
    370  * This function returns an equivalent hash to the MI hash32_strn().
    371  * The latter isn't used because in the *ep == NULL case, determining
    372  * the length of the string to the first NUL or `/' and then calling
    373  * hash32_strn() involves unnecessary double-handling of the data.
    374  */
    375 uint32_t
    376 namei_hash(const char *name, const char **ep)
    377 {
    378 	uint32_t	hash;
    379 
    380 	hash = HASH32_STR_INIT;
    381 	if (*ep != NULL) {
    382 		for (; name < *ep; name++)
    383 			hash = hash * 33 + *(const uint8_t *)name;
    384 	} else {
    385 		for (; *name != '\0' && *name != '/'; name++)
    386 			hash = hash * 33 + *(const uint8_t *)name;
    387 		*ep = name;
    388 	}
    389 	return (hash + (hash >> 5));
    390 }
    391 
    392 /*
    393  * Search a pathname.
    394  * This is a very central and rather complicated routine.
    395  *
    396  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
    397  * The starting directory is taken from ni_startdir. The pathname is
    398  * descended until done, or a symbolic link is encountered. The variable
    399  * ni_more is clear if the path is completed; it is set to one if a
    400  * symbolic link needing interpretation is encountered.
    401  *
    402  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
    403  * whether the name is to be looked up, created, renamed, or deleted.
    404  * When CREATE, RENAME, or DELETE is specified, information usable in
    405  * creating, renaming, or deleting a directory entry may be calculated.
    406  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
    407  * locked. If flag has WANTPARENT or'ed into it, the parent directory is
    408  * returned unlocked. Otherwise the parent directory is not returned. If
    409  * the target of the pathname exists and LOCKLEAF is or'ed into the flag
    410  * the target is returned locked, otherwise it is returned unlocked.
    411  * When creating or renaming and LOCKPARENT is specified, the target may not
    412  * be ".".  When deleting and LOCKPARENT is specified, the target may be ".".
    413  *
    414  * Overall outline of lookup:
    415  *
    416  * dirloop:
    417  *	identify next component of name at ndp->ni_ptr
    418  *	handle degenerate case where name is null string
    419  *	if .. and crossing mount points and on mounted filesys, find parent
    420  *	call VOP_LOOKUP routine for next component name
    421  *	    directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
    422  *	    component vnode returned in ni_vp (if it exists), locked.
    423  *	if result vnode is mounted on and crossing mount points,
    424  *	    find mounted on vnode
    425  *	if more components of name, do next level at dirloop
    426  *	return the answer in ni_vp, locked if LOCKLEAF set
    427  *	    if LOCKPARENT set, return locked parent in ni_dvp
    428  *	    if WANTPARENT set, return unlocked parent in ni_dvp
    429  */
    430 int
    431 lookup(struct nameidata *ndp)
    432 {
    433 	const char *cp;			/* pointer into pathname argument */
    434 	struct vnode *dp = 0;		/* the directory we are searching */
    435 	struct vnode *tdp;		/* saved dp */
    436 	struct mount *mp;		/* mount table entry */
    437 	int docache;			/* == 0 do not cache last component */
    438 	int wantparent;			/* 1 => wantparent or lockparent flag */
    439 	int rdonly;			/* lookup read-only flag bit */
    440 	int error = 0;
    441 	int slashes;
    442 	int dpunlocked = 0;		/* dp has already been unlocked */
    443 	struct componentname *cnp = &ndp->ni_cnd;
    444 	struct lwp *l = cnp->cn_lwp;
    445 
    446 	/*
    447 	 * Setup: break out flag bits into variables.
    448 	 */
    449 	wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
    450 	docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
    451 	if (cnp->cn_nameiop == DELETE ||
    452 	    (wantparent && cnp->cn_nameiop != CREATE))
    453 		docache = 0;
    454 	rdonly = cnp->cn_flags & RDONLY;
    455 	ndp->ni_dvp = NULL;
    456 	cnp->cn_flags &= ~ISSYMLINK;
    457 	dp = ndp->ni_startdir;
    458 	ndp->ni_startdir = NULLVP;
    459 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    460 
    461 	/*
    462 	 * If we have a leading string of slashes, remove them, and just make
    463 	 * sure the current node is a directory.
    464 	 */
    465 	cp = cnp->cn_nameptr;
    466 	if (*cp == '/') {
    467 		do {
    468 			cp++;
    469 		} while (*cp == '/');
    470 		ndp->ni_pathlen -= cp - cnp->cn_nameptr;
    471 		cnp->cn_nameptr = cp;
    472 
    473 		if (dp->v_type != VDIR) {
    474 			error = ENOTDIR;
    475 			goto bad;
    476 		}
    477 
    478 		/*
    479 		 * If we've exhausted the path name, then just return the
    480 		 * current node.  If the caller requested the parent node (i.e.
    481 		 * it's a CREATE, DELETE, or RENAME), and we don't have one
    482 		 * (because this is the root directory), then we must fail.
    483 		 */
    484 		if (cnp->cn_nameptr[0] == '\0') {
    485 			if (ndp->ni_dvp == NULL && wantparent) {
    486 				switch (cnp->cn_nameiop) {
    487 				case CREATE:
    488 					error = EEXIST;
    489 					break;
    490 				case DELETE:
    491 				case RENAME:
    492 					error = EBUSY;
    493 					break;
    494 				default:
    495 					KASSERT(0);
    496 				}
    497 				goto bad;
    498 			}
    499 			ndp->ni_vp = dp;
    500 			cnp->cn_flags |= ISLASTCN;
    501 			goto terminal;
    502 		}
    503 	}
    504 
    505 dirloop:
    506 	/*
    507 	 * Search a new directory.
    508 	 *
    509 	 * The cn_hash value is for use by vfs_cache.
    510 	 * The last component of the filename is left accessible via
    511 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    512 	 * the name set the SAVENAME flag. When done, they assume
    513 	 * responsibility for freeing the pathname buffer.
    514 	 */
    515 	cnp->cn_consume = 0;
    516 	cp = NULL;
    517 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
    518 	cnp->cn_namelen = cp - cnp->cn_nameptr;
    519 	if (cnp->cn_namelen > NAME_MAX) {
    520 		error = ENAMETOOLONG;
    521 		goto bad;
    522 	}
    523 #ifdef NAMEI_DIAGNOSTIC
    524 	{ char c = *cp;
    525 	*(char *)cp = '\0';
    526 	printf("{%s}: ", cnp->cn_nameptr);
    527 	*(char *)cp = c; }
    528 #endif /* NAMEI_DIAGNOSTIC */
    529 	ndp->ni_pathlen -= cnp->cn_namelen;
    530 	ndp->ni_next = cp;
    531 	/*
    532 	 * If this component is followed by a slash, then move the pointer to
    533 	 * the next component forward, and remember that this component must be
    534 	 * a directory.
    535 	 */
    536 	if (*cp == '/') {
    537 		do {
    538 			cp++;
    539 		} while (*cp == '/');
    540 		slashes = cp - ndp->ni_next;
    541 		ndp->ni_pathlen -= slashes;
    542 		ndp->ni_next = cp;
    543 		cnp->cn_flags |= REQUIREDIR;
    544 	} else {
    545 		slashes = 0;
    546 		cnp->cn_flags &= ~REQUIREDIR;
    547 	}
    548 	/*
    549 	 * We do special processing on the last component, whether or not it's
    550 	 * a directory.  Cache all intervening lookups, but not the final one.
    551 	 */
    552 	if (*cp == '\0') {
    553 		if (docache)
    554 			cnp->cn_flags |= MAKEENTRY;
    555 		else
    556 			cnp->cn_flags &= ~MAKEENTRY;
    557 		cnp->cn_flags |= ISLASTCN;
    558 	} else {
    559 		cnp->cn_flags |= MAKEENTRY;
    560 		cnp->cn_flags &= ~ISLASTCN;
    561 	}
    562 	if (cnp->cn_namelen == 2 &&
    563 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
    564 		cnp->cn_flags |= ISDOTDOT;
    565 	else
    566 		cnp->cn_flags &= ~ISDOTDOT;
    567 
    568 	/*
    569 	 * Handle "..": two special cases.
    570 	 * 1. If at root directory (e.g. after chroot)
    571 	 *    or at absolute root directory
    572 	 *    then ignore it so can't get out.
    573 	 * 1a. If we have somehow gotten out of a jail, warn
    574 	 *    and also ignore it so we can't get farther out.
    575 	 * 2. If this vnode is the root of a mounted
    576 	 *    filesystem, then replace it with the
    577 	 *    vnode which was mounted on so we take the
    578 	 *    .. in the other file system.
    579 	 */
    580 	if (cnp->cn_flags & ISDOTDOT) {
    581 		struct proc *p = l->l_proc;
    582 
    583 		for (;;) {
    584 			if (dp == ndp->ni_rootdir || dp == rootvnode) {
    585 				ndp->ni_dvp = dp;
    586 				ndp->ni_vp = dp;
    587 				VREF(dp);
    588 				goto nextname;
    589 			}
    590 			if (ndp->ni_rootdir != rootvnode) {
    591 				int retval;
    592 				VOP_UNLOCK(dp, 0);
    593 				retval = vn_isunder(dp, ndp->ni_rootdir, l);
    594 				vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    595 				if (!retval) {
    596 				    /* Oops! We got out of jail! */
    597 				    log(LOG_WARNING,
    598 					"chrooted pid %d uid %d (%s) "
    599 					"detected outside of its chroot\n",
    600 					p->p_pid, p->p_ucred->cr_uid,
    601 					p->p_comm);
    602 				    /* Put us at the jail root. */
    603 				    vput(dp);
    604 				    dp = ndp->ni_rootdir;
    605 				    ndp->ni_dvp = dp;
    606 				    ndp->ni_vp = dp;
    607 				    VREF(dp);
    608 				    VREF(dp);
    609 				    vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    610 				    goto nextname;
    611 				}
    612 			}
    613 			if ((dp->v_flag & VROOT) == 0 ||
    614 			    (cnp->cn_flags & NOCROSSMOUNT))
    615 				break;
    616 			tdp = dp;
    617 			dp = dp->v_mount->mnt_vnodecovered;
    618 			vput(tdp);
    619 			VREF(dp);
    620 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    621 		}
    622 	}
    623 
    624 	/*
    625 	 * We now have a segment name to search for, and a directory to search.
    626 	 */
    627 unionlookup:
    628 	ndp->ni_dvp = dp;
    629 	ndp->ni_vp = NULL;
    630 	cnp->cn_flags &= ~PDIRUNLOCK;
    631 	if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
    632 #ifdef DIAGNOSTIC
    633 		if (ndp->ni_vp != NULL)
    634 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    635 #endif /* DIAGNOSTIC */
    636 #ifdef NAMEI_DIAGNOSTIC
    637 		printf("not found\n");
    638 #endif /* NAMEI_DIAGNOSTIC */
    639 		if ((error == ENOENT) &&
    640 		    (dp->v_flag & VROOT) &&
    641 		    (dp->v_mount->mnt_flag & MNT_UNION)) {
    642 			tdp = dp;
    643 			dp = dp->v_mount->mnt_vnodecovered;
    644 			if (cnp->cn_flags & PDIRUNLOCK)
    645 				vrele(tdp);
    646 			else
    647 				vput(tdp);
    648 			VREF(dp);
    649 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    650 			goto unionlookup;
    651 		}
    652 
    653 		if (cnp->cn_flags & PDIRUNLOCK)
    654 			dpunlocked = 1;
    655 
    656 		if (error != EJUSTRETURN)
    657 			goto bad;
    658 		/*
    659 		 * If this was not the last component, or there were trailing
    660 		 * slashes, and we are not going to create a directory,
    661 		 * then the name must exist.
    662 		 */
    663 		if ((cnp->cn_flags & (REQUIREDIR | CREATEDIR)) == REQUIREDIR) {
    664 			error = ENOENT;
    665 			goto bad;
    666 		}
    667 		/*
    668 		 * If creating and at end of pathname, then can consider
    669 		 * allowing file to be created.
    670 		 */
    671 		if (rdonly) {
    672 			error = EROFS;
    673 			goto bad;
    674 		}
    675 		/*
    676 		 * We return with ni_vp NULL to indicate that the entry
    677 		 * doesn't currently exist, leaving a pointer to the
    678 		 * (possibly locked) directory inode in ndp->ni_dvp.
    679 		 */
    680 		if (cnp->cn_flags & SAVESTART) {
    681 			ndp->ni_startdir = ndp->ni_dvp;
    682 			VREF(ndp->ni_startdir);
    683 		}
    684 		return (0);
    685 	}
    686 #ifdef NAMEI_DIAGNOSTIC
    687 	printf("found\n");
    688 #endif /* NAMEI_DIAGNOSTIC */
    689 
    690 	/*
    691 	 * Take into account any additional components consumed by the
    692 	 * underlying filesystem.  This will include any trailing slashes after
    693 	 * the last component consumed.
    694 	 */
    695 	if (cnp->cn_consume > 0) {
    696 		ndp->ni_pathlen -= cnp->cn_consume - slashes;
    697 		ndp->ni_next += cnp->cn_consume - slashes;
    698 		cnp->cn_consume = 0;
    699 		if (ndp->ni_next[0] == '\0')
    700 			cnp->cn_flags |= ISLASTCN;
    701 	}
    702 
    703 	dp = ndp->ni_vp;
    704 	/*
    705 	 * Check to see if the vnode has been mounted on;
    706 	 * if so find the root of the mounted file system.
    707 	 */
    708 	while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
    709 	       (cnp->cn_flags & NOCROSSMOUNT) == 0) {
    710 		if (vfs_busy(mp, 0, 0))
    711 			continue;
    712 		VOP_UNLOCK(dp, 0);
    713 		error = VFS_ROOT(mp, &tdp);
    714 		vfs_unbusy(mp);
    715 		if (error) {
    716 			dpunlocked = 1;
    717 			goto bad2;
    718 		}
    719 		vrele(dp);
    720 		ndp->ni_vp = dp = tdp;
    721 	}
    722 
    723 	/*
    724 	 * Check for symbolic link.  Back up over any slashes that we skipped,
    725 	 * as we will need them again.
    726 	 */
    727 	if ((dp->v_type == VLNK) && (cnp->cn_flags & (FOLLOW|REQUIREDIR))) {
    728 		ndp->ni_pathlen += slashes;
    729 		ndp->ni_next -= slashes;
    730 		cnp->cn_flags |= ISSYMLINK;
    731 		return (0);
    732 	}
    733 
    734 	/*
    735 	 * Check for directory, if the component was followed by a series of
    736 	 * slashes.
    737 	 */
    738 	if ((dp->v_type != VDIR) && (cnp->cn_flags & REQUIREDIR)) {
    739 		error = ENOTDIR;
    740 		goto bad2;
    741 	}
    742 
    743 nextname:
    744 	/*
    745 	 * Not a symbolic link.  If this was not the last component, then
    746 	 * continue at the next component, else return.
    747 	 */
    748 	if (!(cnp->cn_flags & ISLASTCN)) {
    749 		cnp->cn_nameptr = ndp->ni_next;
    750 		vrele(ndp->ni_dvp);
    751 		goto dirloop;
    752 	}
    753 
    754 terminal:
    755 	/*
    756 	 * Disallow directory write attempts on read-only file systems.
    757 	 */
    758 	if (rdonly &&
    759 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    760 		/*
    761 		 * Disallow directory write attempts on read-only
    762 		 * file systems.
    763 		 */
    764 		error = EROFS;
    765 		goto bad2;
    766 	}
    767 	if (ndp->ni_dvp != NULL) {
    768 		if (cnp->cn_flags & SAVESTART) {
    769 			ndp->ni_startdir = ndp->ni_dvp;
    770 			VREF(ndp->ni_startdir);
    771 		}
    772 		if (!wantparent)
    773 			vrele(ndp->ni_dvp);
    774 	}
    775 	if ((cnp->cn_flags & LOCKLEAF) == 0)
    776 		VOP_UNLOCK(dp, 0);
    777 	return (0);
    778 
    779 bad2:
    780 	if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
    781 			((cnp->cn_flags & PDIRUNLOCK) == 0))
    782 		VOP_UNLOCK(ndp->ni_dvp, 0);
    783 	vrele(ndp->ni_dvp);
    784 bad:
    785 	if (dpunlocked)
    786 		vrele(dp);
    787 	else
    788 		vput(dp);
    789 	ndp->ni_vp = NULL;
    790 	return (error);
    791 }
    792 
    793 /*
    794  * Reacquire a path name component.
    795  */
    796 int
    797 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
    798 {
    799 	struct vnode *dp = 0;		/* the directory we are searching */
    800 	int wantparent;			/* 1 => wantparent or lockparent flag */
    801 	int rdonly;			/* lookup read-only flag bit */
    802 	int error = 0;
    803 #ifdef DEBUG
    804 	u_long newhash;			/* DEBUG: check name hash */
    805 	const char *cp;			/* DEBUG: check name ptr/len */
    806 #endif /* DEBUG */
    807 
    808 	/*
    809 	 * Setup: break out flag bits into variables.
    810 	 */
    811 	wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
    812 	rdonly = cnp->cn_flags & RDONLY;
    813 	cnp->cn_flags &= ~ISSYMLINK;
    814 	dp = dvp;
    815 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    816 
    817 /* dirloop: */
    818 	/*
    819 	 * Search a new directory.
    820 	 *
    821 	 * The cn_hash value is for use by vfs_cache.
    822 	 * The last component of the filename is left accessible via
    823 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    824 	 * the name set the SAVENAME flag. When done, they assume
    825 	 * responsibility for freeing the pathname buffer.
    826 	 */
    827 #ifdef DEBUG
    828 	cp = NULL;
    829 	newhash = namei_hash(cnp->cn_nameptr, &cp);
    830 	if (newhash != cnp->cn_hash)
    831 		panic("relookup: bad hash");
    832 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
    833 		panic("relookup: bad len");
    834 	while (*cp == '/')
    835 		cp++;
    836 	if (*cp != 0)
    837 		panic("relookup: not last component");
    838 #endif /* DEBUG */
    839 #ifdef NAMEI_DIAGNOSTIC
    840 	printf("{%s}: ", cnp->cn_nameptr);
    841 #endif /* NAMEI_DIAGNOSTIC */
    842 
    843 	/*
    844 	 * Check for degenerate name (e.g. / or "")
    845 	 * which is a way of talking about a directory,
    846 	 * e.g. like "/." or ".".
    847 	 */
    848 	if (cnp->cn_nameptr[0] == '\0')
    849 		panic("relookup: null name");
    850 
    851 	if (cnp->cn_flags & ISDOTDOT)
    852 		panic("relookup: lookup on dot-dot");
    853 
    854 	/*
    855 	 * We now have a segment name to search for, and a directory to search.
    856 	 */
    857 	if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
    858 #ifdef DIAGNOSTIC
    859 		if (*vpp != NULL)
    860 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    861 #endif
    862 		if (error != EJUSTRETURN)
    863 			goto bad;
    864 		/*
    865 		 * If creating and at end of pathname, then can consider
    866 		 * allowing file to be created.
    867 		 */
    868 		if (rdonly) {
    869 			error = EROFS;
    870 			goto bad;
    871 		}
    872 		/* ASSERT(dvp == ndp->ni_startdir) */
    873 		if (cnp->cn_flags & SAVESTART)
    874 			VREF(dvp);
    875 		/*
    876 		 * We return with ni_vp NULL to indicate that the entry
    877 		 * doesn't currently exist, leaving a pointer to the
    878 		 * (possibly locked) directory inode in ndp->ni_dvp.
    879 		 */
    880 		return (0);
    881 	}
    882 	dp = *vpp;
    883 
    884 #ifdef DIAGNOSTIC
    885 	/*
    886 	 * Check for symbolic link
    887 	 */
    888 	if (dp->v_type == VLNK && (cnp->cn_flags & FOLLOW))
    889 		panic("relookup: symlink found");
    890 #endif
    891 
    892 	/*
    893 	 * Check for read-only file systems.
    894 	 */
    895 	if (rdonly &&
    896 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    897 		error = EROFS;
    898 		goto bad2;
    899 	}
    900 	/* ASSERT(dvp == ndp->ni_startdir) */
    901 	if (cnp->cn_flags & SAVESTART)
    902 		VREF(dvp);
    903 	if (!wantparent)
    904 		vrele(dvp);
    905 	if ((cnp->cn_flags & LOCKLEAF) == 0)
    906 		VOP_UNLOCK(dp, 0);
    907 	return (0);
    908 
    909 bad2:
    910 	if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
    911 		VOP_UNLOCK(dvp, 0);
    912 	vrele(dvp);
    913 bad:
    914 	vput(dp);
    915 	*vpp = NULL;
    916 	return (error);
    917 }
    918