Home | History | Annotate | Line # | Download | only in kern
vfs_lookup.c revision 1.72
      1 /*	$NetBSD: vfs_lookup.c,v 1.72 2006/11/04 10:14:00 elad 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 1.72 2006/11/04 10:14:00 elad 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 if (MATCH("uid")) {
    167 			char uidtmp[11]; /* XXX elad */
    168 
    169 			(void)snprintf(uidtmp, sizeof(uidtmp), "%u",
    170 			    kauth_cred_geteuid(kauth_cred_get()));
    171 			SUBSTITUTE("uid", uidtmp, strlen(uidtmp));
    172 		} else {
    173 			tmp[newlen++] = '@';
    174 			if (termchar == VC)
    175 				tmp[newlen++] = VO;
    176 		}
    177 	}
    178 
    179 	if (change) {
    180 		memcpy(cp, tmp, newlen);
    181 		*len = newlen;
    182 	}
    183 	PNBUF_PUT(tmp);
    184 
    185 	return (0);
    186 }
    187 
    188 #undef VNL
    189 #undef VO
    190 #undef VC
    191 #undef MATCH
    192 #undef SUBSTITUTE
    193 
    194 /*
    195  * Convert a pathname into a pointer to a locked vnode.
    196  *
    197  * The FOLLOW flag is set when symbolic links are to be followed
    198  * when they occur at the end of the name translation process.
    199  * Symbolic links are always followed for all other pathname
    200  * components other than the last.
    201  *
    202  * The segflg defines whether the name is to be copied from user
    203  * space or kernel space.
    204  *
    205  * Overall outline of namei:
    206  *
    207  *	copy in name
    208  *	get starting directory
    209  *	while (!done && !error) {
    210  *		call lookup to search path.
    211  *		if symbolic link, massage name in buffer and continue
    212  *	}
    213  */
    214 int
    215 namei(struct nameidata *ndp)
    216 {
    217 	struct cwdinfo *cwdi;		/* pointer to cwd state */
    218 	char *cp;			/* pointer into pathname argument */
    219 	struct vnode *dp;		/* the directory we are searching */
    220 	struct iovec aiov;		/* uio for reading symbolic links */
    221 	struct uio auio;
    222 	int error, linklen;
    223 	struct componentname *cnp = &ndp->ni_cnd;
    224 
    225 #ifdef DIAGNOSTIC
    226 	if (!cnp->cn_cred || !cnp->cn_lwp)
    227 		panic("namei: bad cred/proc");
    228 	if (cnp->cn_nameiop & (~OPMASK))
    229 		panic("namei: nameiop contaminated with flags");
    230 	if (cnp->cn_flags & OPMASK)
    231 		panic("namei: flags contaminated with nameiops");
    232 #endif
    233 	cwdi = cnp->cn_lwp->l_proc->p_cwdi;
    234 
    235 	/*
    236 	 * Get a buffer for the name to be translated, and copy the
    237 	 * name into the buffer.
    238 	 */
    239 	if ((cnp->cn_flags & HASBUF) == 0)
    240 		cnp->cn_pnbuf = PNBUF_GET();
    241 	if (ndp->ni_segflg == UIO_SYSSPACE)
    242 		error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
    243 			    MAXPATHLEN, &ndp->ni_pathlen);
    244 	else
    245 		error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
    246 			    MAXPATHLEN, &ndp->ni_pathlen);
    247 
    248 	/*
    249 	 * POSIX.1 requirement: "" is not a valid file name.
    250 	 */
    251 	if (!error && ndp->ni_pathlen == 1)
    252 		error = ENOENT;
    253 
    254 	if (error) {
    255 		PNBUF_PUT(cnp->cn_pnbuf);
    256 		ndp->ni_vp = NULL;
    257 		return (error);
    258 	}
    259 	ndp->ni_loopcnt = 0;
    260 
    261 #ifdef KTRACE
    262 	if (KTRPOINT(cnp->cn_lwp->l_proc, KTR_NAMEI))
    263 		ktrnamei(cnp->cn_lwp, cnp->cn_pnbuf);
    264 #endif
    265 #ifdef SYSTRACE
    266 	if (ISSET(cnp->cn_lwp->l_proc->p_flag, P_SYSTRACE))
    267 		systrace_namei(ndp);
    268 #endif
    269 
    270 	/*
    271 	 * Get starting point for the translation.
    272 	 */
    273 	if ((ndp->ni_rootdir = cwdi->cwdi_rdir) == NULL)
    274 		ndp->ni_rootdir = rootvnode;
    275 	/*
    276 	 * Check if starting from root directory or current directory.
    277 	 */
    278 	if (cnp->cn_pnbuf[0] == '/') {
    279 		dp = ndp->ni_rootdir;
    280 		VREF(dp);
    281 	} else {
    282 		dp = cwdi->cwdi_cdir;
    283 		VREF(dp);
    284 	}
    285 	for (;;) {
    286 		if (!dp->v_mount)
    287 		{
    288 			/* Give up if the directory is no longer mounted */
    289 			PNBUF_PUT(cnp->cn_pnbuf);
    290 			return (ENOENT);
    291 		}
    292 		cnp->cn_nameptr = cnp->cn_pnbuf;
    293 		ndp->ni_startdir = dp;
    294 		if ((error = lookup(ndp)) != 0) {
    295 			PNBUF_PUT(cnp->cn_pnbuf);
    296 			return (error);
    297 		}
    298 		/*
    299 		 * Check for symbolic link
    300 		 */
    301 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
    302 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
    303 				PNBUF_PUT(cnp->cn_pnbuf);
    304 			else
    305 				cnp->cn_flags |= HASBUF;
    306 			return (0);
    307 		}
    308 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
    309 			VOP_UNLOCK(ndp->ni_dvp, 0);
    310 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
    311 			error = ELOOP;
    312 			break;
    313 		}
    314 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
    315 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
    316 			    cnp->cn_lwp);
    317 			if (error != 0)
    318 				break;
    319 		}
    320 		if (ndp->ni_pathlen > 1)
    321 			cp = PNBUF_GET();
    322 		else
    323 			cp = cnp->cn_pnbuf;
    324 		aiov.iov_base = cp;
    325 		aiov.iov_len = MAXPATHLEN;
    326 		auio.uio_iov = &aiov;
    327 		auio.uio_iovcnt = 1;
    328 		auio.uio_offset = 0;
    329 		auio.uio_rw = UIO_READ;
    330 		auio.uio_resid = MAXPATHLEN;
    331 		UIO_SETUP_SYSSPACE(&auio);
    332 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
    333 		if (error) {
    334 		badlink:
    335 			if (ndp->ni_pathlen > 1)
    336 				PNBUF_PUT(cp);
    337 			break;
    338 		}
    339 		linklen = MAXPATHLEN - auio.uio_resid;
    340 		if (linklen == 0) {
    341 			error = ENOENT;
    342 			goto badlink;
    343 		}
    344 		/*
    345 		 * Do symlink substitution, if appropriate, and
    346 		 * check length for potential overflow.
    347 		 */
    348 		if ((vfs_magiclinks &&
    349 		     symlink_magic(cnp->cn_lwp->l_proc, cp, &linklen)) ||
    350 		    (linklen + ndp->ni_pathlen >= MAXPATHLEN)) {
    351 			error = ENAMETOOLONG;
    352 			goto badlink;
    353 		}
    354 		if (ndp->ni_pathlen > 1) {
    355 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
    356 			PNBUF_PUT(cnp->cn_pnbuf);
    357 			cnp->cn_pnbuf = cp;
    358 		} else
    359 			cnp->cn_pnbuf[linklen] = '\0';
    360 		ndp->ni_pathlen += linklen;
    361 		vput(ndp->ni_vp);
    362 		dp = ndp->ni_dvp;
    363 		/*
    364 		 * Check if root directory should replace current directory.
    365 		 */
    366 		if (cnp->cn_pnbuf[0] == '/') {
    367 			vrele(dp);
    368 			dp = ndp->ni_rootdir;
    369 			VREF(dp);
    370 		}
    371 	}
    372 	PNBUF_PUT(cnp->cn_pnbuf);
    373 	vrele(ndp->ni_dvp);
    374 	vput(ndp->ni_vp);
    375 	ndp->ni_vp = NULL;
    376 	return (error);
    377 }
    378 
    379 /*
    380  * Determine the namei hash (for cn_hash) for name.
    381  * If *ep != NULL, hash from name to ep-1.
    382  * If *ep == NULL, hash from name until the first NUL or '/', and
    383  * return the location of this termination character in *ep.
    384  *
    385  * This function returns an equivalent hash to the MI hash32_strn().
    386  * The latter isn't used because in the *ep == NULL case, determining
    387  * the length of the string to the first NUL or `/' and then calling
    388  * hash32_strn() involves unnecessary double-handling of the data.
    389  */
    390 uint32_t
    391 namei_hash(const char *name, const char **ep)
    392 {
    393 	uint32_t	hash;
    394 
    395 	hash = HASH32_STR_INIT;
    396 	if (*ep != NULL) {
    397 		for (; name < *ep; name++)
    398 			hash = hash * 33 + *(const uint8_t *)name;
    399 	} else {
    400 		for (; *name != '\0' && *name != '/'; name++)
    401 			hash = hash * 33 + *(const uint8_t *)name;
    402 		*ep = name;
    403 	}
    404 	return (hash + (hash >> 5));
    405 }
    406 
    407 /*
    408  * Search a pathname.
    409  * This is a very central and rather complicated routine.
    410  *
    411  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
    412  * The starting directory is taken from ni_startdir. The pathname is
    413  * descended until done, or a symbolic link is encountered. The variable
    414  * ni_more is clear if the path is completed; it is set to one if a
    415  * symbolic link needing interpretation is encountered.
    416  *
    417  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
    418  * whether the name is to be looked up, created, renamed, or deleted.
    419  * When CREATE, RENAME, or DELETE is specified, information usable in
    420  * creating, renaming, or deleting a directory entry may be calculated.
    421  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
    422  * locked. If flag has WANTPARENT or'ed into it, the parent directory is
    423  * returned unlocked. Otherwise the parent directory is not returned. If
    424  * the target of the pathname exists and LOCKLEAF is or'ed into the flag
    425  * the target is returned locked, otherwise it is returned unlocked.
    426  * When creating or renaming and LOCKPARENT is specified, the target may not
    427  * be ".".  When deleting and LOCKPARENT is specified, the target may be ".".
    428  *
    429  * Overall outline of lookup:
    430  *
    431  * dirloop:
    432  *	identify next component of name at ndp->ni_ptr
    433  *	handle degenerate case where name is null string
    434  *	if .. and crossing mount points and on mounted filesys, find parent
    435  *	call VOP_LOOKUP routine for next component name
    436  *	    directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
    437  *	    component vnode returned in ni_vp (if it exists), locked.
    438  *	if result vnode is mounted on and crossing mount points,
    439  *	    find mounted on vnode
    440  *	if more components of name, do next level at dirloop
    441  *	return the answer in ni_vp, locked if LOCKLEAF set
    442  *	    if LOCKPARENT set, return locked parent in ni_dvp
    443  *	    if WANTPARENT set, return unlocked parent in ni_dvp
    444  */
    445 int
    446 lookup(struct nameidata *ndp)
    447 {
    448 	const char *cp;			/* pointer into pathname argument */
    449 	struct vnode *dp = 0;		/* the directory we are searching */
    450 	struct vnode *tdp;		/* saved dp */
    451 	struct mount *mp;		/* mount table entry */
    452 	int docache;			/* == 0 do not cache last component */
    453 	int wantparent;			/* 1 => wantparent or lockparent flag */
    454 	int rdonly;			/* lookup read-only flag bit */
    455 	int error = 0;
    456 	int slashes;
    457 	int dpunlocked = 0;		/* dp has already been unlocked */
    458 	struct componentname *cnp = &ndp->ni_cnd;
    459 	struct lwp *l = cnp->cn_lwp;
    460 
    461 	/*
    462 	 * Setup: break out flag bits into variables.
    463 	 */
    464 	wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
    465 	docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
    466 	if (cnp->cn_nameiop == DELETE ||
    467 	    (wantparent && cnp->cn_nameiop != CREATE))
    468 		docache = 0;
    469 	rdonly = cnp->cn_flags & RDONLY;
    470 	ndp->ni_dvp = NULL;
    471 	cnp->cn_flags &= ~ISSYMLINK;
    472 	dp = ndp->ni_startdir;
    473 	ndp->ni_startdir = NULLVP;
    474 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    475 
    476 	/*
    477 	 * If we have a leading string of slashes, remove them, and just make
    478 	 * sure the current node is a directory.
    479 	 */
    480 	cp = cnp->cn_nameptr;
    481 	if (*cp == '/') {
    482 		do {
    483 			cp++;
    484 		} while (*cp == '/');
    485 		ndp->ni_pathlen -= cp - cnp->cn_nameptr;
    486 		cnp->cn_nameptr = cp;
    487 
    488 		if (dp->v_type != VDIR) {
    489 			error = ENOTDIR;
    490 			goto bad;
    491 		}
    492 
    493 		/*
    494 		 * If we've exhausted the path name, then just return the
    495 		 * current node.  If the caller requested the parent node (i.e.
    496 		 * it's a CREATE, DELETE, or RENAME), and we don't have one
    497 		 * (because this is the root directory), then we must fail.
    498 		 */
    499 		if (cnp->cn_nameptr[0] == '\0') {
    500 			if (ndp->ni_dvp == NULL && wantparent) {
    501 				switch (cnp->cn_nameiop) {
    502 				case CREATE:
    503 					error = EEXIST;
    504 					break;
    505 				case DELETE:
    506 				case RENAME:
    507 					error = EBUSY;
    508 					break;
    509 				default:
    510 					KASSERT(0);
    511 				}
    512 				goto bad;
    513 			}
    514 			ndp->ni_vp = dp;
    515 			cnp->cn_flags |= ISLASTCN;
    516 			goto terminal;
    517 		}
    518 	}
    519 
    520 dirloop:
    521 	/*
    522 	 * Search a new directory.
    523 	 *
    524 	 * The cn_hash value is for use by vfs_cache.
    525 	 * The last component of the filename is left accessible via
    526 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    527 	 * the name set the SAVENAME flag. When done, they assume
    528 	 * responsibility for freeing the pathname buffer.
    529 	 */
    530 	cnp->cn_consume = 0;
    531 	cp = NULL;
    532 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
    533 	cnp->cn_namelen = cp - cnp->cn_nameptr;
    534 	if (cnp->cn_namelen > NAME_MAX) {
    535 		error = ENAMETOOLONG;
    536 		goto bad;
    537 	}
    538 #ifdef NAMEI_DIAGNOSTIC
    539 	{ char c = *cp;
    540 	*(char *)cp = '\0';
    541 	printf("{%s}: ", cnp->cn_nameptr);
    542 	*(char *)cp = c; }
    543 #endif /* NAMEI_DIAGNOSTIC */
    544 	ndp->ni_pathlen -= cnp->cn_namelen;
    545 	ndp->ni_next = cp;
    546 	/*
    547 	 * If this component is followed by a slash, then move the pointer to
    548 	 * the next component forward, and remember that this component must be
    549 	 * a directory.
    550 	 */
    551 	if (*cp == '/') {
    552 		do {
    553 			cp++;
    554 		} while (*cp == '/');
    555 		slashes = cp - ndp->ni_next;
    556 		ndp->ni_pathlen -= slashes;
    557 		ndp->ni_next = cp;
    558 		cnp->cn_flags |= REQUIREDIR;
    559 	} else {
    560 		slashes = 0;
    561 		cnp->cn_flags &= ~REQUIREDIR;
    562 	}
    563 	/*
    564 	 * We do special processing on the last component, whether or not it's
    565 	 * a directory.  Cache all intervening lookups, but not the final one.
    566 	 */
    567 	if (*cp == '\0') {
    568 		if (docache)
    569 			cnp->cn_flags |= MAKEENTRY;
    570 		else
    571 			cnp->cn_flags &= ~MAKEENTRY;
    572 		cnp->cn_flags |= ISLASTCN;
    573 	} else {
    574 		cnp->cn_flags |= MAKEENTRY;
    575 		cnp->cn_flags &= ~ISLASTCN;
    576 	}
    577 	if (cnp->cn_namelen == 2 &&
    578 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
    579 		cnp->cn_flags |= ISDOTDOT;
    580 	else
    581 		cnp->cn_flags &= ~ISDOTDOT;
    582 
    583 	/*
    584 	 * Handle "..": two special cases.
    585 	 * 1. If at root directory (e.g. after chroot)
    586 	 *    or at absolute root directory
    587 	 *    then ignore it so can't get out.
    588 	 * 1a. If we have somehow gotten out of a jail, warn
    589 	 *    and also ignore it so we can't get farther out.
    590 	 * 2. If this vnode is the root of a mounted
    591 	 *    filesystem, then replace it with the
    592 	 *    vnode which was mounted on so we take the
    593 	 *    .. in the other file system.
    594 	 */
    595 	if (cnp->cn_flags & ISDOTDOT) {
    596 		struct proc *p = l->l_proc;
    597 
    598 		for (;;) {
    599 			if (dp == ndp->ni_rootdir || dp == rootvnode) {
    600 				ndp->ni_dvp = dp;
    601 				ndp->ni_vp = dp;
    602 				VREF(dp);
    603 				goto nextname;
    604 			}
    605 			if (ndp->ni_rootdir != rootvnode) {
    606 				int retval;
    607 				VOP_UNLOCK(dp, 0);
    608 				retval = vn_isunder(dp, ndp->ni_rootdir, l);
    609 				vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    610 				if (!retval) {
    611 				    /* Oops! We got out of jail! */
    612 				    log(LOG_WARNING,
    613 					"chrooted pid %d uid %d (%s) "
    614 					"detected outside of its chroot\n",
    615 					p->p_pid, kauth_cred_geteuid(l->l_cred),
    616 					p->p_comm);
    617 				    /* Put us at the jail root. */
    618 				    vput(dp);
    619 				    dp = ndp->ni_rootdir;
    620 				    ndp->ni_dvp = dp;
    621 				    ndp->ni_vp = dp;
    622 				    VREF(dp);
    623 				    VREF(dp);
    624 				    vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    625 				    goto nextname;
    626 				}
    627 			}
    628 			if ((dp->v_flag & VROOT) == 0 ||
    629 			    (cnp->cn_flags & NOCROSSMOUNT))
    630 				break;
    631 			tdp = dp;
    632 			dp = dp->v_mount->mnt_vnodecovered;
    633 			vput(tdp);
    634 			VREF(dp);
    635 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    636 		}
    637 	}
    638 
    639 	/*
    640 	 * We now have a segment name to search for, and a directory to search.
    641 	 */
    642 unionlookup:
    643 	ndp->ni_dvp = dp;
    644 	ndp->ni_vp = NULL;
    645 	cnp->cn_flags &= ~PDIRUNLOCK;
    646 	if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
    647 #ifdef DIAGNOSTIC
    648 		if (ndp->ni_vp != NULL)
    649 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    650 #endif /* DIAGNOSTIC */
    651 #ifdef NAMEI_DIAGNOSTIC
    652 		printf("not found\n");
    653 #endif /* NAMEI_DIAGNOSTIC */
    654 		if ((error == ENOENT) &&
    655 		    (dp->v_flag & VROOT) &&
    656 		    (dp->v_mount->mnt_flag & MNT_UNION)) {
    657 			tdp = dp;
    658 			dp = dp->v_mount->mnt_vnodecovered;
    659 			if (cnp->cn_flags & PDIRUNLOCK)
    660 				vrele(tdp);
    661 			else
    662 				vput(tdp);
    663 			VREF(dp);
    664 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    665 			goto unionlookup;
    666 		}
    667 
    668 		if (cnp->cn_flags & PDIRUNLOCK)
    669 			dpunlocked = 1;
    670 
    671 		if (error != EJUSTRETURN)
    672 			goto bad;
    673 		/*
    674 		 * If this was not the last component, or there were trailing
    675 		 * slashes, and we are not going to create a directory,
    676 		 * then the name must exist.
    677 		 */
    678 		if ((cnp->cn_flags & (REQUIREDIR | CREATEDIR)) == REQUIREDIR) {
    679 			error = ENOENT;
    680 			goto bad;
    681 		}
    682 		/*
    683 		 * If creating and at end of pathname, then can consider
    684 		 * allowing file to be created.
    685 		 */
    686 		if (rdonly) {
    687 			error = EROFS;
    688 			goto bad;
    689 		}
    690 		/*
    691 		 * We return with ni_vp NULL to indicate that the entry
    692 		 * doesn't currently exist, leaving a pointer to the
    693 		 * (possibly locked) directory vnode in ndp->ni_dvp.
    694 		 */
    695 		if (cnp->cn_flags & SAVESTART) {
    696 			ndp->ni_startdir = ndp->ni_dvp;
    697 			VREF(ndp->ni_startdir);
    698 		}
    699 		return (0);
    700 	}
    701 #ifdef NAMEI_DIAGNOSTIC
    702 	printf("found\n");
    703 #endif /* NAMEI_DIAGNOSTIC */
    704 
    705 	/*
    706 	 * Take into account any additional components consumed by the
    707 	 * underlying filesystem.  This will include any trailing slashes after
    708 	 * the last component consumed.
    709 	 */
    710 	if (cnp->cn_consume > 0) {
    711 		ndp->ni_pathlen -= cnp->cn_consume - slashes;
    712 		ndp->ni_next += cnp->cn_consume - slashes;
    713 		cnp->cn_consume = 0;
    714 		if (ndp->ni_next[0] == '\0')
    715 			cnp->cn_flags |= ISLASTCN;
    716 	}
    717 
    718 	dp = ndp->ni_vp;
    719 	/*
    720 	 * Check to see if the vnode has been mounted on;
    721 	 * if so find the root of the mounted file system.
    722 	 */
    723 	while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
    724 	       (cnp->cn_flags & NOCROSSMOUNT) == 0) {
    725 		if (vfs_busy(mp, 0, 0))
    726 			continue;
    727 		VOP_UNLOCK(dp, 0);
    728 		error = VFS_ROOT(mp, &tdp);
    729 		vfs_unbusy(mp);
    730 		if (error) {
    731 			dpunlocked = 1;
    732 			goto bad2;
    733 		}
    734 		vrele(dp);
    735 		ndp->ni_vp = dp = tdp;
    736 	}
    737 
    738 	/*
    739 	 * Check for symbolic link.  Back up over any slashes that we skipped,
    740 	 * as we will need them again.
    741 	 */
    742 	if ((dp->v_type == VLNK) && (cnp->cn_flags & (FOLLOW|REQUIREDIR))) {
    743 		ndp->ni_pathlen += slashes;
    744 		ndp->ni_next -= slashes;
    745 		cnp->cn_flags |= ISSYMLINK;
    746 		return (0);
    747 	}
    748 
    749 	/*
    750 	 * Check for directory, if the component was followed by a series of
    751 	 * slashes.
    752 	 */
    753 	if ((dp->v_type != VDIR) && (cnp->cn_flags & REQUIREDIR)) {
    754 		error = ENOTDIR;
    755 		goto bad2;
    756 	}
    757 
    758 nextname:
    759 	/*
    760 	 * Not a symbolic link.  If this was not the last component, then
    761 	 * continue at the next component, else return.
    762 	 */
    763 	if (!(cnp->cn_flags & ISLASTCN)) {
    764 		cnp->cn_nameptr = ndp->ni_next;
    765 		vrele(ndp->ni_dvp);
    766 		goto dirloop;
    767 	}
    768 
    769 terminal:
    770 	/*
    771 	 * Disallow directory write attempts on read-only file systems.
    772 	 */
    773 	if (rdonly &&
    774 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    775 		/*
    776 		 * Disallow directory write attempts on read-only
    777 		 * file systems.
    778 		 */
    779 		error = EROFS;
    780 		goto bad2;
    781 	}
    782 	if (ndp->ni_dvp != NULL) {
    783 		if (cnp->cn_flags & SAVESTART) {
    784 			ndp->ni_startdir = ndp->ni_dvp;
    785 			VREF(ndp->ni_startdir);
    786 		}
    787 		if (!wantparent)
    788 			vrele(ndp->ni_dvp);
    789 	}
    790 	if ((cnp->cn_flags & LOCKLEAF) == 0)
    791 		VOP_UNLOCK(dp, 0);
    792 	return (0);
    793 
    794 bad2:
    795 	if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
    796 			((cnp->cn_flags & PDIRUNLOCK) == 0))
    797 		VOP_UNLOCK(ndp->ni_dvp, 0);
    798 	vrele(ndp->ni_dvp);
    799 bad:
    800 	if (dpunlocked)
    801 		vrele(dp);
    802 	else
    803 		vput(dp);
    804 	ndp->ni_vp = NULL;
    805 	return (error);
    806 }
    807 
    808 /*
    809  * Reacquire a path name component.
    810  */
    811 int
    812 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
    813 {
    814 	struct vnode *dp = 0;		/* the directory we are searching */
    815 	int wantparent;			/* 1 => wantparent or lockparent flag */
    816 	int rdonly;			/* lookup read-only flag bit */
    817 	int error = 0;
    818 #ifdef DEBUG
    819 	u_long newhash;			/* DEBUG: check name hash */
    820 	const char *cp;			/* DEBUG: check name ptr/len */
    821 #endif /* DEBUG */
    822 
    823 	/*
    824 	 * Setup: break out flag bits into variables.
    825 	 */
    826 	wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
    827 	rdonly = cnp->cn_flags & RDONLY;
    828 	cnp->cn_flags &= ~ISSYMLINK;
    829 	dp = dvp;
    830 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    831 
    832 /* dirloop: */
    833 	/*
    834 	 * Search a new directory.
    835 	 *
    836 	 * The cn_hash value is for use by vfs_cache.
    837 	 * The last component of the filename is left accessible via
    838 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    839 	 * the name set the SAVENAME flag. When done, they assume
    840 	 * responsibility for freeing the pathname buffer.
    841 	 */
    842 #ifdef DEBUG
    843 	cp = NULL;
    844 	newhash = namei_hash(cnp->cn_nameptr, &cp);
    845 	if (newhash != cnp->cn_hash)
    846 		panic("relookup: bad hash");
    847 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
    848 		panic("relookup: bad len");
    849 	while (*cp == '/')
    850 		cp++;
    851 	if (*cp != 0)
    852 		panic("relookup: not last component");
    853 #endif /* DEBUG */
    854 #ifdef NAMEI_DIAGNOSTIC
    855 	printf("{%s}: ", cnp->cn_nameptr);
    856 #endif /* NAMEI_DIAGNOSTIC */
    857 
    858 	/*
    859 	 * Check for degenerate name (e.g. / or "")
    860 	 * which is a way of talking about a directory,
    861 	 * e.g. like "/." or ".".
    862 	 */
    863 	if (cnp->cn_nameptr[0] == '\0')
    864 		panic("relookup: null name");
    865 
    866 	if (cnp->cn_flags & ISDOTDOT)
    867 		panic("relookup: lookup on dot-dot");
    868 
    869 	/*
    870 	 * We now have a segment name to search for, and a directory to search.
    871 	 */
    872 	if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
    873 #ifdef DIAGNOSTIC
    874 		if (*vpp != NULL)
    875 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    876 #endif
    877 		if (error != EJUSTRETURN)
    878 			goto bad;
    879 		/*
    880 		 * If creating and at end of pathname, then can consider
    881 		 * allowing file to be created.
    882 		 */
    883 		if (rdonly) {
    884 			error = EROFS;
    885 			goto bad;
    886 		}
    887 		/* ASSERT(dvp == ndp->ni_startdir) */
    888 		if (cnp->cn_flags & SAVESTART)
    889 			VREF(dvp);
    890 		/*
    891 		 * We return with ni_vp NULL to indicate that the entry
    892 		 * doesn't currently exist, leaving a pointer to the
    893 		 * (possibly locked) directory vnode in ndp->ni_dvp.
    894 		 */
    895 		return (0);
    896 	}
    897 	dp = *vpp;
    898 
    899 #ifdef DIAGNOSTIC
    900 	/*
    901 	 * Check for symbolic link
    902 	 */
    903 	if (dp->v_type == VLNK && (cnp->cn_flags & FOLLOW))
    904 		panic("relookup: symlink found");
    905 #endif
    906 
    907 	/*
    908 	 * Check for read-only file systems.
    909 	 */
    910 	if (rdonly &&
    911 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    912 		error = EROFS;
    913 		goto bad2;
    914 	}
    915 	/* ASSERT(dvp == ndp->ni_startdir) */
    916 	if (cnp->cn_flags & SAVESTART)
    917 		VREF(dvp);
    918 	if (!wantparent)
    919 		vrele(dvp);
    920 	if ((cnp->cn_flags & LOCKLEAF) == 0)
    921 		VOP_UNLOCK(dp, 0);
    922 	return (0);
    923 
    924 bad2:
    925 	if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
    926 		VOP_UNLOCK(dvp, 0);
    927 	vrele(dvp);
    928 bad:
    929 	vput(dp);
    930 	*vpp = NULL;
    931 	return (error);
    932 }
    933