Home | History | Annotate | Line # | Download | only in kern
vfs_lookup.c revision 1.65.4.1
      1 /*	$NetBSD: vfs_lookup.c,v 1.65.4.1 2006/09/09 02:57:17 rpaulo 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 1.65.4.1 2006/09/09 02:57:17 rpaulo Exp $");
     41 
     42 #include "opt_ktrace.h"
     43 #include "opt_systrace.h"
     44 #include "opt_magiclinks.h"
     45 
     46 #include <sys/param.h>
     47 #include <sys/systm.h>
     48 #include <sys/kernel.h>
     49 #include <sys/syslimits.h>
     50 #include <sys/time.h>
     51 #include <sys/namei.h>
     52 #include <sys/vnode.h>
     53 #include <sys/mount.h>
     54 #include <sys/errno.h>
     55 #include <sys/filedesc.h>
     56 #include <sys/hash.h>
     57 #include <sys/malloc.h>
     58 #include <sys/proc.h>
     59 #include <sys/syslog.h>
     60 #include <sys/kauth.h>
     61 
     62 #ifdef KTRACE
     63 #include <sys/ktrace.h>
     64 #endif
     65 #ifdef SYSTRACE
     66 #include <sys/systrace.h>
     67 #endif
     68 
     69 #ifndef MAGICLINKS
     70 #define MAGICLINKS 0
     71 #endif
     72 
     73 int vfs_magiclinks = MAGICLINKS;
     74 
     75 struct pool pnbuf_pool;		/* pathname buffer pool */
     76 struct pool_cache pnbuf_cache;	/* pathname buffer cache */
     77 
     78 /*
     79  * Substitute replacement text for 'magic' strings in symlinks.
     80  * Returns 0 if successful, and returns non-zero if an error
     81  * occurs.  (Currently, the only possible error is running out
     82  * of temporary pathname space.)
     83  *
     84  * Looks for "@<string>" and "@<string>/", where <string> is a
     85  * recognized 'magic' string.  Replaces the "@<string>" with the
     86  * appropriate replacement text.  (Note that in some cases the
     87  * replacement text may have zero length.)
     88  *
     89  * This would have been table driven, but the variance in
     90  * replacement strings (and replacement string lengths) made
     91  * that impractical.
     92  */
     93 #define	VNL(x)							\
     94 	(sizeof(x) - 1)
     95 
     96 #define	VO	'{'
     97 #define	VC	'}'
     98 
     99 #define	MATCH(str)						\
    100 	((termchar == '/' && i + VNL(str) == *len) ||		\
    101 	 (i + VNL(str) < *len &&				\
    102 	  cp[i + VNL(str)] == termchar)) &&			\
    103 	!strncmp((str), &cp[i], VNL(str))
    104 
    105 #define	SUBSTITUTE(m, s, sl)					\
    106 	if ((newlen + (sl)) > MAXPATHLEN)			\
    107 		return (1);					\
    108 	i += VNL(m);						\
    109 	if (termchar != '/')					\
    110 		i++;						\
    111 	memcpy(&tmp[newlen], (s), (sl));			\
    112 	newlen += (sl);						\
    113 	change = 1;						\
    114 	termchar = '/';
    115 
    116 static int
    117 symlink_magic(struct proc *p, char *cp, int *len)
    118 {
    119 	char *tmp;
    120 	int change, i, newlen;
    121 	int termchar = '/';
    122 
    123 	tmp = PNBUF_GET();
    124 	for (change = i = newlen = 0; i < *len; ) {
    125 		if (cp[i] != '@') {
    126 			tmp[newlen++] = cp[i++];
    127 			continue;
    128 		}
    129 
    130 		i++;
    131 
    132 		/* Check for @{var} syntax. */
    133 		if (cp[i] == VO) {
    134 			termchar = VC;
    135 			i++;
    136 		}
    137 
    138 		/*
    139 		 * The following checks should be ordered according
    140 		 * to frequency of use.
    141 		 */
    142 		if (MATCH("machine_arch")) {
    143 			SUBSTITUTE("machine_arch", MACHINE_ARCH,
    144 			    sizeof(MACHINE_ARCH) - 1);
    145 		} else if (MATCH("machine")) {
    146 			SUBSTITUTE("machine", MACHINE,
    147 			    sizeof(MACHINE) - 1);
    148 		} else if (MATCH("hostname")) {
    149 			SUBSTITUTE("hostname", hostname,
    150 			    hostnamelen);
    151 		} else if (MATCH("osrelease")) {
    152 			SUBSTITUTE("osrelease", osrelease,
    153 			    strlen(osrelease));
    154 		} else if (MATCH("emul")) {
    155 			SUBSTITUTE("emul", p->p_emul->e_name,
    156 			    strlen(p->p_emul->e_name));
    157 		} else if (MATCH("kernel_ident")) {
    158 			SUBSTITUTE("kernel_ident", kernel_ident,
    159 			    strlen(kernel_ident));
    160 		} else if (MATCH("domainname")) {
    161 			SUBSTITUTE("domainname", domainname,
    162 			    domainnamelen);
    163 		} else if (MATCH("ostype")) {
    164 			SUBSTITUTE("ostype", ostype,
    165 			    strlen(ostype));
    166 		} else {
    167 			tmp[newlen++] = '@';
    168 			if (termchar == VC)
    169 				tmp[newlen++] = VO;
    170 		}
    171 	}
    172 
    173 	if (change) {
    174 		memcpy(cp, tmp, newlen);
    175 		*len = newlen;
    176 	}
    177 	PNBUF_PUT(tmp);
    178 
    179 	return (0);
    180 }
    181 
    182 #undef VNL
    183 #undef VO
    184 #undef VC
    185 #undef MATCH
    186 #undef SUBSTITUTE
    187 
    188 /*
    189  * Convert a pathname into a pointer to a locked vnode.
    190  *
    191  * The FOLLOW flag is set when symbolic links are to be followed
    192  * when they occur at the end of the name translation process.
    193  * Symbolic links are always followed for all other pathname
    194  * components other than the last.
    195  *
    196  * The segflg defines whether the name is to be copied from user
    197  * space or kernel space.
    198  *
    199  * Overall outline of namei:
    200  *
    201  *	copy in name
    202  *	get starting directory
    203  *	while (!done && !error) {
    204  *		call lookup to search path.
    205  *		if symbolic link, massage name in buffer and continue
    206  *	}
    207  */
    208 int
    209 namei(struct nameidata *ndp)
    210 {
    211 	struct cwdinfo *cwdi;		/* pointer to cwd state */
    212 	char *cp;			/* pointer into pathname argument */
    213 	struct vnode *dp;		/* the directory we are searching */
    214 	struct iovec aiov;		/* uio for reading symbolic links */
    215 	struct uio auio;
    216 	int error, linklen;
    217 	struct componentname *cnp = &ndp->ni_cnd;
    218 
    219 #ifdef DIAGNOSTIC
    220 	if (!cnp->cn_cred || !cnp->cn_lwp)
    221 		panic("namei: bad cred/proc");
    222 	if (cnp->cn_nameiop & (~OPMASK))
    223 		panic("namei: nameiop contaminated with flags");
    224 	if (cnp->cn_flags & OPMASK)
    225 		panic("namei: flags contaminated with nameiops");
    226 #endif
    227 	cwdi = cnp->cn_lwp->l_proc->p_cwdi;
    228 
    229 	/*
    230 	 * Get a buffer for the name to be translated, and copy the
    231 	 * name into the buffer.
    232 	 */
    233 	if ((cnp->cn_flags & HASBUF) == 0)
    234 		cnp->cn_pnbuf = PNBUF_GET();
    235 	if (ndp->ni_segflg == UIO_SYSSPACE)
    236 		error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
    237 			    MAXPATHLEN, &ndp->ni_pathlen);
    238 	else
    239 		error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
    240 			    MAXPATHLEN, &ndp->ni_pathlen);
    241 
    242 	/*
    243 	 * POSIX.1 requirement: "" is not a valid file name.
    244 	 */
    245 	if (!error && ndp->ni_pathlen == 1)
    246 		error = ENOENT;
    247 
    248 	if (error) {
    249 		PNBUF_PUT(cnp->cn_pnbuf);
    250 		ndp->ni_vp = NULL;
    251 		return (error);
    252 	}
    253 	ndp->ni_loopcnt = 0;
    254 
    255 #ifdef KTRACE
    256 	if (KTRPOINT(cnp->cn_lwp->l_proc, KTR_NAMEI))
    257 		ktrnamei(cnp->cn_lwp, cnp->cn_pnbuf);
    258 #endif
    259 #ifdef SYSTRACE
    260 	if (ISSET(cnp->cn_lwp->l_proc->p_flag, P_SYSTRACE))
    261 		systrace_namei(ndp);
    262 #endif
    263 
    264 	/*
    265 	 * Get starting point for the translation.
    266 	 */
    267 	if ((ndp->ni_rootdir = cwdi->cwdi_rdir) == NULL)
    268 		ndp->ni_rootdir = rootvnode;
    269 	/*
    270 	 * Check if starting from root directory or current directory.
    271 	 */
    272 	if (cnp->cn_pnbuf[0] == '/') {
    273 		dp = ndp->ni_rootdir;
    274 		VREF(dp);
    275 	} else {
    276 		dp = cwdi->cwdi_cdir;
    277 		VREF(dp);
    278 	}
    279 	for (;;) {
    280 		if (!dp->v_mount)
    281 		{
    282 			/* Give up if the directory is no longer mounted */
    283 			PNBUF_PUT(cnp->cn_pnbuf);
    284 			return (ENOENT);
    285 		}
    286 		cnp->cn_nameptr = cnp->cn_pnbuf;
    287 		ndp->ni_startdir = dp;
    288 		if ((error = lookup(ndp)) != 0) {
    289 			PNBUF_PUT(cnp->cn_pnbuf);
    290 			return (error);
    291 		}
    292 		/*
    293 		 * Check for symbolic link
    294 		 */
    295 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
    296 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
    297 				PNBUF_PUT(cnp->cn_pnbuf);
    298 			else
    299 				cnp->cn_flags |= HASBUF;
    300 			return (0);
    301 		}
    302 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
    303 			VOP_UNLOCK(ndp->ni_dvp, 0);
    304 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
    305 			error = ELOOP;
    306 			break;
    307 		}
    308 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
    309 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
    310 			    cnp->cn_lwp);
    311 			if (error != 0)
    312 				break;
    313 		}
    314 		if (ndp->ni_pathlen > 1)
    315 			cp = PNBUF_GET();
    316 		else
    317 			cp = cnp->cn_pnbuf;
    318 		aiov.iov_base = cp;
    319 		aiov.iov_len = MAXPATHLEN;
    320 		auio.uio_iov = &aiov;
    321 		auio.uio_iovcnt = 1;
    322 		auio.uio_offset = 0;
    323 		auio.uio_rw = UIO_READ;
    324 		auio.uio_resid = MAXPATHLEN;
    325 		UIO_SETUP_SYSSPACE(&auio);
    326 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
    327 		if (error) {
    328 		badlink:
    329 			if (ndp->ni_pathlen > 1)
    330 				PNBUF_PUT(cp);
    331 			break;
    332 		}
    333 		linklen = MAXPATHLEN - auio.uio_resid;
    334 		if (linklen == 0) {
    335 			error = ENOENT;
    336 			goto badlink;
    337 		}
    338 		/*
    339 		 * Do symlink substitution, if appropriate, and
    340 		 * check length for potential overflow.
    341 		 */
    342 		if ((vfs_magiclinks &&
    343 		     symlink_magic(cnp->cn_lwp->l_proc, cp, &linklen)) ||
    344 		    (linklen + ndp->ni_pathlen >= MAXPATHLEN)) {
    345 			error = ENAMETOOLONG;
    346 			goto badlink;
    347 		}
    348 		if (ndp->ni_pathlen > 1) {
    349 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
    350 			PNBUF_PUT(cnp->cn_pnbuf);
    351 			cnp->cn_pnbuf = cp;
    352 		} else
    353 			cnp->cn_pnbuf[linklen] = '\0';
    354 		ndp->ni_pathlen += linklen;
    355 		vput(ndp->ni_vp);
    356 		dp = ndp->ni_dvp;
    357 		/*
    358 		 * Check if root directory should replace current directory.
    359 		 */
    360 		if (cnp->cn_pnbuf[0] == '/') {
    361 			vrele(dp);
    362 			dp = ndp->ni_rootdir;
    363 			VREF(dp);
    364 		}
    365 	}
    366 	PNBUF_PUT(cnp->cn_pnbuf);
    367 	vrele(ndp->ni_dvp);
    368 	vput(ndp->ni_vp);
    369 	ndp->ni_vp = NULL;
    370 	return (error);
    371 }
    372 
    373 /*
    374  * Determine the namei hash (for cn_hash) for name.
    375  * If *ep != NULL, hash from name to ep-1.
    376  * If *ep == NULL, hash from name until the first NUL or '/', and
    377  * return the location of this termination character in *ep.
    378  *
    379  * This function returns an equivalent hash to the MI hash32_strn().
    380  * The latter isn't used because in the *ep == NULL case, determining
    381  * the length of the string to the first NUL or `/' and then calling
    382  * hash32_strn() involves unnecessary double-handling of the data.
    383  */
    384 uint32_t
    385 namei_hash(const char *name, const char **ep)
    386 {
    387 	uint32_t	hash;
    388 
    389 	hash = HASH32_STR_INIT;
    390 	if (*ep != NULL) {
    391 		for (; name < *ep; name++)
    392 			hash = hash * 33 + *(const uint8_t *)name;
    393 	} else {
    394 		for (; *name != '\0' && *name != '/'; name++)
    395 			hash = hash * 33 + *(const uint8_t *)name;
    396 		*ep = name;
    397 	}
    398 	return (hash + (hash >> 5));
    399 }
    400 
    401 /*
    402  * Search a pathname.
    403  * This is a very central and rather complicated routine.
    404  *
    405  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
    406  * The starting directory is taken from ni_startdir. The pathname is
    407  * descended until done, or a symbolic link is encountered. The variable
    408  * ni_more is clear if the path is completed; it is set to one if a
    409  * symbolic link needing interpretation is encountered.
    410  *
    411  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
    412  * whether the name is to be looked up, created, renamed, or deleted.
    413  * When CREATE, RENAME, or DELETE is specified, information usable in
    414  * creating, renaming, or deleting a directory entry may be calculated.
    415  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
    416  * locked. If flag has WANTPARENT or'ed into it, the parent directory is
    417  * returned unlocked. Otherwise the parent directory is not returned. If
    418  * the target of the pathname exists and LOCKLEAF is or'ed into the flag
    419  * the target is returned locked, otherwise it is returned unlocked.
    420  * When creating or renaming and LOCKPARENT is specified, the target may not
    421  * be ".".  When deleting and LOCKPARENT is specified, the target may be ".".
    422  *
    423  * Overall outline of lookup:
    424  *
    425  * dirloop:
    426  *	identify next component of name at ndp->ni_ptr
    427  *	handle degenerate case where name is null string
    428  *	if .. and crossing mount points and on mounted filesys, find parent
    429  *	call VOP_LOOKUP routine for next component name
    430  *	    directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
    431  *	    component vnode returned in ni_vp (if it exists), locked.
    432  *	if result vnode is mounted on and crossing mount points,
    433  *	    find mounted on vnode
    434  *	if more components of name, do next level at dirloop
    435  *	return the answer in ni_vp, locked if LOCKLEAF set
    436  *	    if LOCKPARENT set, return locked parent in ni_dvp
    437  *	    if WANTPARENT set, return unlocked parent in ni_dvp
    438  */
    439 int
    440 lookup(struct nameidata *ndp)
    441 {
    442 	const char *cp;			/* pointer into pathname argument */
    443 	struct vnode *dp = 0;		/* the directory we are searching */
    444 	struct vnode *tdp;		/* saved dp */
    445 	struct mount *mp;		/* mount table entry */
    446 	int docache;			/* == 0 do not cache last component */
    447 	int wantparent;			/* 1 => wantparent or lockparent flag */
    448 	int rdonly;			/* lookup read-only flag bit */
    449 	int error = 0;
    450 	int slashes;
    451 	int dpunlocked = 0;		/* dp has already been unlocked */
    452 	struct componentname *cnp = &ndp->ni_cnd;
    453 	struct lwp *l = cnp->cn_lwp;
    454 
    455 	/*
    456 	 * Setup: break out flag bits into variables.
    457 	 */
    458 	wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
    459 	docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
    460 	if (cnp->cn_nameiop == DELETE ||
    461 	    (wantparent && cnp->cn_nameiop != CREATE))
    462 		docache = 0;
    463 	rdonly = cnp->cn_flags & RDONLY;
    464 	ndp->ni_dvp = NULL;
    465 	cnp->cn_flags &= ~ISSYMLINK;
    466 	dp = ndp->ni_startdir;
    467 	ndp->ni_startdir = NULLVP;
    468 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    469 
    470 	/*
    471 	 * If we have a leading string of slashes, remove them, and just make
    472 	 * sure the current node is a directory.
    473 	 */
    474 	cp = cnp->cn_nameptr;
    475 	if (*cp == '/') {
    476 		do {
    477 			cp++;
    478 		} while (*cp == '/');
    479 		ndp->ni_pathlen -= cp - cnp->cn_nameptr;
    480 		cnp->cn_nameptr = cp;
    481 
    482 		if (dp->v_type != VDIR) {
    483 			error = ENOTDIR;
    484 			goto bad;
    485 		}
    486 
    487 		/*
    488 		 * If we've exhausted the path name, then just return the
    489 		 * current node.  If the caller requested the parent node (i.e.
    490 		 * it's a CREATE, DELETE, or RENAME), and we don't have one
    491 		 * (because this is the root directory), then we must fail.
    492 		 */
    493 		if (cnp->cn_nameptr[0] == '\0') {
    494 			if (ndp->ni_dvp == NULL && wantparent) {
    495 				switch (cnp->cn_nameiop) {
    496 				case CREATE:
    497 					error = EEXIST;
    498 					break;
    499 				case DELETE:
    500 				case RENAME:
    501 					error = EBUSY;
    502 					break;
    503 				default:
    504 					KASSERT(0);
    505 				}
    506 				goto bad;
    507 			}
    508 			ndp->ni_vp = dp;
    509 			cnp->cn_flags |= ISLASTCN;
    510 			goto terminal;
    511 		}
    512 	}
    513 
    514 dirloop:
    515 	/*
    516 	 * Search a new directory.
    517 	 *
    518 	 * The cn_hash value is for use by vfs_cache.
    519 	 * The last component of the filename is left accessible via
    520 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    521 	 * the name set the SAVENAME flag. When done, they assume
    522 	 * responsibility for freeing the pathname buffer.
    523 	 */
    524 	cnp->cn_consume = 0;
    525 	cp = NULL;
    526 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
    527 	cnp->cn_namelen = cp - cnp->cn_nameptr;
    528 	if (cnp->cn_namelen > NAME_MAX) {
    529 		error = ENAMETOOLONG;
    530 		goto bad;
    531 	}
    532 #ifdef NAMEI_DIAGNOSTIC
    533 	{ char c = *cp;
    534 	*(char *)cp = '\0';
    535 	printf("{%s}: ", cnp->cn_nameptr);
    536 	*(char *)cp = c; }
    537 #endif /* NAMEI_DIAGNOSTIC */
    538 	ndp->ni_pathlen -= cnp->cn_namelen;
    539 	ndp->ni_next = cp;
    540 	/*
    541 	 * If this component is followed by a slash, then move the pointer to
    542 	 * the next component forward, and remember that this component must be
    543 	 * a directory.
    544 	 */
    545 	if (*cp == '/') {
    546 		do {
    547 			cp++;
    548 		} while (*cp == '/');
    549 		slashes = cp - ndp->ni_next;
    550 		ndp->ni_pathlen -= slashes;
    551 		ndp->ni_next = cp;
    552 		cnp->cn_flags |= REQUIREDIR;
    553 	} else {
    554 		slashes = 0;
    555 		cnp->cn_flags &= ~REQUIREDIR;
    556 	}
    557 	/*
    558 	 * We do special processing on the last component, whether or not it's
    559 	 * a directory.  Cache all intervening lookups, but not the final one.
    560 	 */
    561 	if (*cp == '\0') {
    562 		if (docache)
    563 			cnp->cn_flags |= MAKEENTRY;
    564 		else
    565 			cnp->cn_flags &= ~MAKEENTRY;
    566 		cnp->cn_flags |= ISLASTCN;
    567 	} else {
    568 		cnp->cn_flags |= MAKEENTRY;
    569 		cnp->cn_flags &= ~ISLASTCN;
    570 	}
    571 	if (cnp->cn_namelen == 2 &&
    572 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
    573 		cnp->cn_flags |= ISDOTDOT;
    574 	else
    575 		cnp->cn_flags &= ~ISDOTDOT;
    576 
    577 	/*
    578 	 * Handle "..": two special cases.
    579 	 * 1. If at root directory (e.g. after chroot)
    580 	 *    or at absolute root directory
    581 	 *    then ignore it so can't get out.
    582 	 * 1a. If we have somehow gotten out of a jail, warn
    583 	 *    and also ignore it so we can't get farther out.
    584 	 * 2. If this vnode is the root of a mounted
    585 	 *    filesystem, then replace it with the
    586 	 *    vnode which was mounted on so we take the
    587 	 *    .. in the other file system.
    588 	 */
    589 	if (cnp->cn_flags & ISDOTDOT) {
    590 		struct proc *p = l->l_proc;
    591 
    592 		for (;;) {
    593 			if (dp == ndp->ni_rootdir || dp == rootvnode) {
    594 				ndp->ni_dvp = dp;
    595 				ndp->ni_vp = dp;
    596 				VREF(dp);
    597 				goto nextname;
    598 			}
    599 			if (ndp->ni_rootdir != rootvnode) {
    600 				int retval;
    601 				VOP_UNLOCK(dp, 0);
    602 				retval = vn_isunder(dp, ndp->ni_rootdir, l);
    603 				vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    604 				if (!retval) {
    605 				    /* Oops! We got out of jail! */
    606 				    log(LOG_WARNING,
    607 					"chrooted pid %d uid %d (%s) "
    608 					"detected outside of its chroot\n",
    609 					p->p_pid, kauth_cred_geteuid(l->l_cred),
    610 					p->p_comm);
    611 				    /* Put us at the jail root. */
    612 				    vput(dp);
    613 				    dp = ndp->ni_rootdir;
    614 				    ndp->ni_dvp = dp;
    615 				    ndp->ni_vp = dp;
    616 				    VREF(dp);
    617 				    VREF(dp);
    618 				    vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    619 				    goto nextname;
    620 				}
    621 			}
    622 			if ((dp->v_flag & VROOT) == 0 ||
    623 			    (cnp->cn_flags & NOCROSSMOUNT))
    624 				break;
    625 			tdp = dp;
    626 			dp = dp->v_mount->mnt_vnodecovered;
    627 			vput(tdp);
    628 			VREF(dp);
    629 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    630 		}
    631 	}
    632 
    633 	/*
    634 	 * We now have a segment name to search for, and a directory to search.
    635 	 */
    636 unionlookup:
    637 	ndp->ni_dvp = dp;
    638 	ndp->ni_vp = NULL;
    639 	cnp->cn_flags &= ~PDIRUNLOCK;
    640 	if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
    641 #ifdef DIAGNOSTIC
    642 		if (ndp->ni_vp != NULL)
    643 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    644 #endif /* DIAGNOSTIC */
    645 #ifdef NAMEI_DIAGNOSTIC
    646 		printf("not found\n");
    647 #endif /* NAMEI_DIAGNOSTIC */
    648 		if ((error == ENOENT) &&
    649 		    (dp->v_flag & VROOT) &&
    650 		    (dp->v_mount->mnt_flag & MNT_UNION)) {
    651 			tdp = dp;
    652 			dp = dp->v_mount->mnt_vnodecovered;
    653 			if (cnp->cn_flags & PDIRUNLOCK)
    654 				vrele(tdp);
    655 			else
    656 				vput(tdp);
    657 			VREF(dp);
    658 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    659 			goto unionlookup;
    660 		}
    661 
    662 		if (cnp->cn_flags & PDIRUNLOCK)
    663 			dpunlocked = 1;
    664 
    665 		if (error != EJUSTRETURN)
    666 			goto bad;
    667 		/*
    668 		 * If this was not the last component, or there were trailing
    669 		 * slashes, and we are not going to create a directory,
    670 		 * then the name must exist.
    671 		 */
    672 		if ((cnp->cn_flags & (REQUIREDIR | CREATEDIR)) == REQUIREDIR) {
    673 			error = ENOENT;
    674 			goto bad;
    675 		}
    676 		/*
    677 		 * If creating and at end of pathname, then can consider
    678 		 * allowing file to be created.
    679 		 */
    680 		if (rdonly) {
    681 			error = EROFS;
    682 			goto bad;
    683 		}
    684 		/*
    685 		 * We return with ni_vp NULL to indicate that the entry
    686 		 * doesn't currently exist, leaving a pointer to the
    687 		 * (possibly locked) directory vnode in ndp->ni_dvp.
    688 		 */
    689 		if (cnp->cn_flags & SAVESTART) {
    690 			ndp->ni_startdir = ndp->ni_dvp;
    691 			VREF(ndp->ni_startdir);
    692 		}
    693 		return (0);
    694 	}
    695 #ifdef NAMEI_DIAGNOSTIC
    696 	printf("found\n");
    697 #endif /* NAMEI_DIAGNOSTIC */
    698 
    699 	/*
    700 	 * Take into account any additional components consumed by the
    701 	 * underlying filesystem.  This will include any trailing slashes after
    702 	 * the last component consumed.
    703 	 */
    704 	if (cnp->cn_consume > 0) {
    705 		ndp->ni_pathlen -= cnp->cn_consume - slashes;
    706 		ndp->ni_next += cnp->cn_consume - slashes;
    707 		cnp->cn_consume = 0;
    708 		if (ndp->ni_next[0] == '\0')
    709 			cnp->cn_flags |= ISLASTCN;
    710 	}
    711 
    712 	dp = ndp->ni_vp;
    713 	/*
    714 	 * Check to see if the vnode has been mounted on;
    715 	 * if so find the root of the mounted file system.
    716 	 */
    717 	while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
    718 	       (cnp->cn_flags & NOCROSSMOUNT) == 0) {
    719 		if (vfs_busy(mp, 0, 0))
    720 			continue;
    721 		VOP_UNLOCK(dp, 0);
    722 		error = VFS_ROOT(mp, &tdp);
    723 		vfs_unbusy(mp);
    724 		if (error) {
    725 			dpunlocked = 1;
    726 			goto bad2;
    727 		}
    728 		vrele(dp);
    729 		ndp->ni_vp = dp = tdp;
    730 	}
    731 
    732 	/*
    733 	 * Check for symbolic link.  Back up over any slashes that we skipped,
    734 	 * as we will need them again.
    735 	 */
    736 	if ((dp->v_type == VLNK) && (cnp->cn_flags & (FOLLOW|REQUIREDIR))) {
    737 		ndp->ni_pathlen += slashes;
    738 		ndp->ni_next -= slashes;
    739 		cnp->cn_flags |= ISSYMLINK;
    740 		return (0);
    741 	}
    742 
    743 	/*
    744 	 * Check for directory, if the component was followed by a series of
    745 	 * slashes.
    746 	 */
    747 	if ((dp->v_type != VDIR) && (cnp->cn_flags & REQUIREDIR)) {
    748 		error = ENOTDIR;
    749 		goto bad2;
    750 	}
    751 
    752 nextname:
    753 	/*
    754 	 * Not a symbolic link.  If this was not the last component, then
    755 	 * continue at the next component, else return.
    756 	 */
    757 	if (!(cnp->cn_flags & ISLASTCN)) {
    758 		cnp->cn_nameptr = ndp->ni_next;
    759 		vrele(ndp->ni_dvp);
    760 		goto dirloop;
    761 	}
    762 
    763 terminal:
    764 	/*
    765 	 * Disallow directory write attempts on read-only file systems.
    766 	 */
    767 	if (rdonly &&
    768 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    769 		/*
    770 		 * Disallow directory write attempts on read-only
    771 		 * file systems.
    772 		 */
    773 		error = EROFS;
    774 		goto bad2;
    775 	}
    776 	if (ndp->ni_dvp != NULL) {
    777 		if (cnp->cn_flags & SAVESTART) {
    778 			ndp->ni_startdir = ndp->ni_dvp;
    779 			VREF(ndp->ni_startdir);
    780 		}
    781 		if (!wantparent)
    782 			vrele(ndp->ni_dvp);
    783 	}
    784 	if ((cnp->cn_flags & LOCKLEAF) == 0)
    785 		VOP_UNLOCK(dp, 0);
    786 	return (0);
    787 
    788 bad2:
    789 	if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
    790 			((cnp->cn_flags & PDIRUNLOCK) == 0))
    791 		VOP_UNLOCK(ndp->ni_dvp, 0);
    792 	vrele(ndp->ni_dvp);
    793 bad:
    794 	if (dpunlocked)
    795 		vrele(dp);
    796 	else
    797 		vput(dp);
    798 	ndp->ni_vp = NULL;
    799 	return (error);
    800 }
    801 
    802 /*
    803  * Reacquire a path name component.
    804  */
    805 int
    806 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
    807 {
    808 	struct vnode *dp = 0;		/* the directory we are searching */
    809 	int wantparent;			/* 1 => wantparent or lockparent flag */
    810 	int rdonly;			/* lookup read-only flag bit */
    811 	int error = 0;
    812 #ifdef DEBUG
    813 	u_long newhash;			/* DEBUG: check name hash */
    814 	const char *cp;			/* DEBUG: check name ptr/len */
    815 #endif /* DEBUG */
    816 
    817 	/*
    818 	 * Setup: break out flag bits into variables.
    819 	 */
    820 	wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
    821 	rdonly = cnp->cn_flags & RDONLY;
    822 	cnp->cn_flags &= ~ISSYMLINK;
    823 	dp = dvp;
    824 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    825 
    826 /* dirloop: */
    827 	/*
    828 	 * Search a new directory.
    829 	 *
    830 	 * The cn_hash value is for use by vfs_cache.
    831 	 * The last component of the filename is left accessible via
    832 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    833 	 * the name set the SAVENAME flag. When done, they assume
    834 	 * responsibility for freeing the pathname buffer.
    835 	 */
    836 #ifdef DEBUG
    837 	cp = NULL;
    838 	newhash = namei_hash(cnp->cn_nameptr, &cp);
    839 	if (newhash != cnp->cn_hash)
    840 		panic("relookup: bad hash");
    841 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
    842 		panic("relookup: bad len");
    843 	while (*cp == '/')
    844 		cp++;
    845 	if (*cp != 0)
    846 		panic("relookup: not last component");
    847 #endif /* DEBUG */
    848 #ifdef NAMEI_DIAGNOSTIC
    849 	printf("{%s}: ", cnp->cn_nameptr);
    850 #endif /* NAMEI_DIAGNOSTIC */
    851 
    852 	/*
    853 	 * Check for degenerate name (e.g. / or "")
    854 	 * which is a way of talking about a directory,
    855 	 * e.g. like "/." or ".".
    856 	 */
    857 	if (cnp->cn_nameptr[0] == '\0')
    858 		panic("relookup: null name");
    859 
    860 	if (cnp->cn_flags & ISDOTDOT)
    861 		panic("relookup: lookup on dot-dot");
    862 
    863 	/*
    864 	 * We now have a segment name to search for, and a directory to search.
    865 	 */
    866 	if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
    867 #ifdef DIAGNOSTIC
    868 		if (*vpp != NULL)
    869 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    870 #endif
    871 		if (error != EJUSTRETURN)
    872 			goto bad;
    873 		/*
    874 		 * If creating and at end of pathname, then can consider
    875 		 * allowing file to be created.
    876 		 */
    877 		if (rdonly) {
    878 			error = EROFS;
    879 			goto bad;
    880 		}
    881 		/* ASSERT(dvp == ndp->ni_startdir) */
    882 		if (cnp->cn_flags & SAVESTART)
    883 			VREF(dvp);
    884 		/*
    885 		 * We return with ni_vp NULL to indicate that the entry
    886 		 * doesn't currently exist, leaving a pointer to the
    887 		 * (possibly locked) directory vnode in ndp->ni_dvp.
    888 		 */
    889 		return (0);
    890 	}
    891 	dp = *vpp;
    892 
    893 #ifdef DIAGNOSTIC
    894 	/*
    895 	 * Check for symbolic link
    896 	 */
    897 	if (dp->v_type == VLNK && (cnp->cn_flags & FOLLOW))
    898 		panic("relookup: symlink found");
    899 #endif
    900 
    901 	/*
    902 	 * Check for read-only file systems.
    903 	 */
    904 	if (rdonly &&
    905 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    906 		error = EROFS;
    907 		goto bad2;
    908 	}
    909 	/* ASSERT(dvp == ndp->ni_startdir) */
    910 	if (cnp->cn_flags & SAVESTART)
    911 		VREF(dvp);
    912 	if (!wantparent)
    913 		vrele(dvp);
    914 	if ((cnp->cn_flags & LOCKLEAF) == 0)
    915 		VOP_UNLOCK(dp, 0);
    916 	return (0);
    917 
    918 bad2:
    919 	if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
    920 		VOP_UNLOCK(dvp, 0);
    921 	vrele(dvp);
    922 bad:
    923 	vput(dp);
    924 	*vpp = NULL;
    925 	return (error);
    926 }
    927