Home | History | Annotate | Line # | Download | only in kern
vfs_lookup.c revision 1.62.2.3
      1 /*	$NetBSD: vfs_lookup.c,v 1.62.2.3 2007/02/26 09:11:22 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 1.62.2.3 2007/02/26 09:11:22 yamt 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 struct pathname_internal {
     74 	char *pathbuf;
     75 	bool needfree;
     76 };
     77 
     78 int vfs_magiclinks = MAGICLINKS;
     79 
     80 struct pool pnbuf_pool;		/* pathname buffer pool */
     81 struct pool_cache pnbuf_cache;	/* pathname buffer cache */
     82 
     83 /*
     84  * Substitute replacement text for 'magic' strings in symlinks.
     85  * Returns 0 if successful, and returns non-zero if an error
     86  * occurs.  (Currently, the only possible error is running out
     87  * of temporary pathname space.)
     88  *
     89  * Looks for "@<string>" and "@<string>/", where <string> is a
     90  * recognized 'magic' string.  Replaces the "@<string>" with the
     91  * appropriate replacement text.  (Note that in some cases the
     92  * replacement text may have zero length.)
     93  *
     94  * This would have been table driven, but the variance in
     95  * replacement strings (and replacement string lengths) made
     96  * that impractical.
     97  */
     98 #define	VNL(x)							\
     99 	(sizeof(x) - 1)
    100 
    101 #define	VO	'{'
    102 #define	VC	'}'
    103 
    104 #define	MATCH(str)						\
    105 	((termchar == '/' && i + VNL(str) == *len) ||		\
    106 	 (i + VNL(str) < *len &&				\
    107 	  cp[i + VNL(str)] == termchar)) &&			\
    108 	!strncmp((str), &cp[i], VNL(str))
    109 
    110 #define	SUBSTITUTE(m, s, sl)					\
    111 	if ((newlen + (sl)) > MAXPATHLEN)			\
    112 		return (1);					\
    113 	i += VNL(m);						\
    114 	if (termchar != '/')					\
    115 		i++;						\
    116 	memcpy(&tmp[newlen], (s), (sl));			\
    117 	newlen += (sl);						\
    118 	change = 1;						\
    119 	termchar = '/';
    120 
    121 static int
    122 symlink_magic(struct proc *p, char *cp, int *len)
    123 {
    124 	char *tmp;
    125 	int change, i, newlen;
    126 	int termchar = '/';
    127 
    128 	tmp = PNBUF_GET();
    129 	for (change = i = newlen = 0; i < *len; ) {
    130 		if (cp[i] != '@') {
    131 			tmp[newlen++] = cp[i++];
    132 			continue;
    133 		}
    134 
    135 		i++;
    136 
    137 		/* Check for @{var} syntax. */
    138 		if (cp[i] == VO) {
    139 			termchar = VC;
    140 			i++;
    141 		}
    142 
    143 		/*
    144 		 * The following checks should be ordered according
    145 		 * to frequency of use.
    146 		 */
    147 		if (MATCH("machine_arch")) {
    148 			SUBSTITUTE("machine_arch", MACHINE_ARCH,
    149 			    sizeof(MACHINE_ARCH) - 1);
    150 		} else if (MATCH("machine")) {
    151 			SUBSTITUTE("machine", MACHINE,
    152 			    sizeof(MACHINE) - 1);
    153 		} else if (MATCH("hostname")) {
    154 			SUBSTITUTE("hostname", hostname,
    155 			    hostnamelen);
    156 		} else if (MATCH("osrelease")) {
    157 			SUBSTITUTE("osrelease", osrelease,
    158 			    strlen(osrelease));
    159 		} else if (MATCH("emul")) {
    160 			SUBSTITUTE("emul", p->p_emul->e_name,
    161 			    strlen(p->p_emul->e_name));
    162 		} else if (MATCH("kernel_ident")) {
    163 			SUBSTITUTE("kernel_ident", kernel_ident,
    164 			    strlen(kernel_ident));
    165 		} else if (MATCH("domainname")) {
    166 			SUBSTITUTE("domainname", domainname,
    167 			    domainnamelen);
    168 		} else if (MATCH("ostype")) {
    169 			SUBSTITUTE("ostype", ostype,
    170 			    strlen(ostype));
    171 		} else if (MATCH("uid")) {
    172 			char uidtmp[11]; /* XXX elad */
    173 
    174 			(void)snprintf(uidtmp, sizeof(uidtmp), "%u",
    175 			    kauth_cred_geteuid(kauth_cred_get()));
    176 			SUBSTITUTE("uid", uidtmp, strlen(uidtmp));
    177 		} else {
    178 			tmp[newlen++] = '@';
    179 			if (termchar == VC)
    180 				tmp[newlen++] = VO;
    181 		}
    182 	}
    183 
    184 	if (change) {
    185 		memcpy(cp, tmp, newlen);
    186 		*len = newlen;
    187 	}
    188 	PNBUF_PUT(tmp);
    189 
    190 	return (0);
    191 }
    192 
    193 #undef VNL
    194 #undef VO
    195 #undef VC
    196 #undef MATCH
    197 #undef SUBSTITUTE
    198 
    199 int
    200 pathname_get(const char *dirp, enum uio_seg segflg, pathname_t *path)
    201 {
    202 	int error;
    203 
    204 	if (dirp == NULL)
    205 		return (EFAULT);
    206 
    207 	*path = malloc(sizeof(struct pathname_internal), M_TEMP,
    208 	    M_ZERO|M_WAITOK);
    209 
    210 	if (segflg == UIO_USERSPACE) {
    211 		(*path)->pathbuf = PNBUF_GET();
    212 		error = copyinstr(dirp, (*path)->pathbuf, MAXPATHLEN,
    213 		    NULL);
    214 		if (error) {
    215 			PNBUF_PUT((*path)->pathbuf);
    216 			free(*path, M_TEMP);
    217 			*path = NULL;
    218 			return (error);
    219 		}
    220 		(*path)->needfree = true;
    221 	} else {
    222 		(*path)->pathbuf = __UNCONST(dirp);
    223 		(*path)->needfree = false;
    224 	}
    225 
    226 	return (0);
    227 }
    228 
    229 const char *
    230 pathname_path(pathname_t path)
    231 {
    232 	KASSERT(path != NULL);
    233 	return (path->pathbuf);
    234 }
    235 
    236 void
    237 pathname_put(pathname_t path)
    238 {
    239 	if (path != NULL) {
    240 		if (path->pathbuf != NULL && path->needfree)
    241 			PNBUF_PUT(path->pathbuf);
    242 		free(path, M_TEMP);
    243 	}
    244 }
    245 
    246 /*
    247  * Convert a pathname into a pointer to a locked vnode.
    248  *
    249  * The FOLLOW flag is set when symbolic links are to be followed
    250  * when they occur at the end of the name translation process.
    251  * Symbolic links are always followed for all other pathname
    252  * components other than the last.
    253  *
    254  * The segflg defines whether the name is to be copied from user
    255  * space or kernel space.
    256  *
    257  * Overall outline of namei:
    258  *
    259  *	copy in name
    260  *	get starting directory
    261  *	while (!done && !error) {
    262  *		call lookup to search path.
    263  *		if symbolic link, massage name in buffer and continue
    264  *	}
    265  */
    266 int
    267 namei(struct nameidata *ndp)
    268 {
    269 	struct cwdinfo *cwdi;		/* pointer to cwd state */
    270 	char *cp;			/* pointer into pathname argument */
    271 	struct vnode *dp;		/* the directory we are searching */
    272 	struct iovec aiov;		/* uio for reading symbolic links */
    273 	struct uio auio;
    274 	int error, linklen;
    275 	struct componentname *cnp = &ndp->ni_cnd;
    276 
    277 #ifdef DIAGNOSTIC
    278 	if (!cnp->cn_cred || !cnp->cn_lwp)
    279 		panic("namei: bad cred/proc");
    280 	if (cnp->cn_nameiop & (~OPMASK))
    281 		panic("namei: nameiop contaminated with flags");
    282 	if (cnp->cn_flags & OPMASK)
    283 		panic("namei: flags contaminated with nameiops");
    284 #endif
    285 	cwdi = cnp->cn_lwp->l_proc->p_cwdi;
    286 
    287 	/*
    288 	 * Get a buffer for the name to be translated, and copy the
    289 	 * name into the buffer.
    290 	 */
    291 	if ((cnp->cn_flags & HASBUF) == 0)
    292 		cnp->cn_pnbuf = PNBUF_GET();
    293 	if (ndp->ni_segflg == UIO_SYSSPACE)
    294 		error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
    295 			    MAXPATHLEN, &ndp->ni_pathlen);
    296 	else
    297 		error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
    298 			    MAXPATHLEN, &ndp->ni_pathlen);
    299 
    300 	/*
    301 	 * POSIX.1 requirement: "" is not a valid file name.
    302 	 */
    303 	if (!error && ndp->ni_pathlen == 1)
    304 		error = ENOENT;
    305 
    306 	if (error) {
    307 		PNBUF_PUT(cnp->cn_pnbuf);
    308 		ndp->ni_vp = NULL;
    309 		return (error);
    310 	}
    311 	ndp->ni_loopcnt = 0;
    312 
    313 #ifdef KTRACE
    314 	if (KTRPOINT(cnp->cn_lwp->l_proc, KTR_NAMEI))
    315 		ktrnamei(cnp->cn_lwp, cnp->cn_pnbuf);
    316 #endif
    317 #ifdef SYSTRACE
    318 	if (ISSET(cnp->cn_lwp->l_proc->p_flag, PK_SYSTRACE))
    319 		systrace_namei(ndp);
    320 #endif
    321 
    322 	/*
    323 	 * Get starting point for the translation.
    324 	 */
    325 	if ((ndp->ni_rootdir = cwdi->cwdi_rdir) == NULL)
    326 		ndp->ni_rootdir = rootvnode;
    327 	/*
    328 	 * Check if starting from root directory or current directory.
    329 	 */
    330 	if (cnp->cn_pnbuf[0] == '/') {
    331 		dp = ndp->ni_rootdir;
    332 		VREF(dp);
    333 	} else {
    334 		dp = cwdi->cwdi_cdir;
    335 		VREF(dp);
    336 	}
    337 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    338 	for (;;) {
    339 		if (!dp->v_mount) {
    340 			/* Give up if the directory is no longer mounted */
    341 			vput(dp);
    342 			PNBUF_PUT(cnp->cn_pnbuf);
    343 			return (ENOENT);
    344 		}
    345 		cnp->cn_nameptr = cnp->cn_pnbuf;
    346 		ndp->ni_startdir = dp;
    347 		error = lookup(ndp);
    348 		if (error != 0) {
    349 			if (ndp->ni_dvp) {
    350 				vput(ndp->ni_dvp);
    351 			}
    352 			PNBUF_PUT(cnp->cn_pnbuf);
    353 			return (error);
    354 		}
    355 
    356 		/*
    357 		 * Check for symbolic link
    358 		 */
    359 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
    360 			if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) {
    361 				if (ndp->ni_dvp == ndp->ni_vp) {
    362 					vrele(ndp->ni_dvp);
    363 				} else {
    364 					vput(ndp->ni_dvp);
    365 				}
    366 			}
    367 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
    368 				PNBUF_PUT(cnp->cn_pnbuf);
    369 			else
    370 				cnp->cn_flags |= HASBUF;
    371 			return (0);
    372 		}
    373 
    374 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
    375 			error = ELOOP;
    376 			break;
    377 		}
    378 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
    379 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
    380 			    cnp->cn_lwp);
    381 			if (error != 0)
    382 				break;
    383 		}
    384 		if (ndp->ni_pathlen > 1)
    385 			cp = PNBUF_GET();
    386 		else
    387 			cp = cnp->cn_pnbuf;
    388 		aiov.iov_base = cp;
    389 		aiov.iov_len = MAXPATHLEN;
    390 		auio.uio_iov = &aiov;
    391 		auio.uio_iovcnt = 1;
    392 		auio.uio_offset = 0;
    393 		auio.uio_rw = UIO_READ;
    394 		auio.uio_resid = MAXPATHLEN;
    395 		UIO_SETUP_SYSSPACE(&auio);
    396 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
    397 		if (error) {
    398 badlink:
    399 			if (ndp->ni_pathlen > 1)
    400 				PNBUF_PUT(cp);
    401 			break;
    402 		}
    403 		linklen = MAXPATHLEN - auio.uio_resid;
    404 		if (linklen == 0) {
    405 			error = ENOENT;
    406 			goto badlink;
    407 		}
    408 
    409 		/*
    410 		 * Do symlink substitution, if appropriate, and
    411 		 * check length for potential overflow.
    412 		 */
    413 		if ((vfs_magiclinks &&
    414 		     symlink_magic(cnp->cn_lwp->l_proc, cp, &linklen)) ||
    415 		    (linklen + ndp->ni_pathlen >= MAXPATHLEN)) {
    416 			error = ENAMETOOLONG;
    417 			goto badlink;
    418 		}
    419 		if (ndp->ni_pathlen > 1) {
    420 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
    421 			PNBUF_PUT(cnp->cn_pnbuf);
    422 			cnp->cn_pnbuf = cp;
    423 		} else
    424 			cnp->cn_pnbuf[linklen] = '\0';
    425 		ndp->ni_pathlen += linklen;
    426 		vput(ndp->ni_vp);
    427 		dp = ndp->ni_dvp;
    428 
    429 		/*
    430 		 * Check if root directory should replace current directory.
    431 		 */
    432 		if (cnp->cn_pnbuf[0] == '/') {
    433 			vput(dp);
    434 			dp = ndp->ni_rootdir;
    435 			VREF(dp);
    436 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    437 		}
    438 	}
    439 	KASSERT(ndp->ni_dvp != ndp->ni_vp);
    440 	vput(ndp->ni_dvp);
    441 	vput(ndp->ni_vp);
    442 	ndp->ni_vp = NULL;
    443 	PNBUF_PUT(cnp->cn_pnbuf);
    444 	return (error);
    445 }
    446 
    447 /*
    448  * Determine the namei hash (for cn_hash) for name.
    449  * If *ep != NULL, hash from name to ep-1.
    450  * If *ep == NULL, hash from name until the first NUL or '/', and
    451  * return the location of this termination character in *ep.
    452  *
    453  * This function returns an equivalent hash to the MI hash32_strn().
    454  * The latter isn't used because in the *ep == NULL case, determining
    455  * the length of the string to the first NUL or `/' and then calling
    456  * hash32_strn() involves unnecessary double-handling of the data.
    457  */
    458 uint32_t
    459 namei_hash(const char *name, const char **ep)
    460 {
    461 	uint32_t	hash;
    462 
    463 	hash = HASH32_STR_INIT;
    464 	if (*ep != NULL) {
    465 		for (; name < *ep; name++)
    466 			hash = hash * 33 + *(const uint8_t *)name;
    467 	} else {
    468 		for (; *name != '\0' && *name != '/'; name++)
    469 			hash = hash * 33 + *(const uint8_t *)name;
    470 		*ep = name;
    471 	}
    472 	return (hash + (hash >> 5));
    473 }
    474 
    475 /*
    476  * Search a pathname.
    477  * This is a very central and rather complicated routine.
    478  *
    479  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
    480  * The starting directory is taken from ni_startdir. The pathname is
    481  * descended until done, or a symbolic link is encountered. The variable
    482  * ni_more is clear if the path is completed; it is set to one if a
    483  * symbolic link needing interpretation is encountered.
    484  *
    485  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
    486  * whether the name is to be looked up, created, renamed, or deleted.
    487  * When CREATE, RENAME, or DELETE is specified, information usable in
    488  * creating, renaming, or deleting a directory entry may be calculated.
    489  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
    490  * locked.  Otherwise the parent directory is not returned. If the target
    491  * of the pathname exists and LOCKLEAF is or'ed into the flag the target
    492  * is returned locked, otherwise it is returned unlocked.  When creating
    493  * or renaming and LOCKPARENT is specified, the target may not be ".".
    494  * When deleting and LOCKPARENT is specified, the target may be ".".
    495  *
    496  * Overall outline of lookup:
    497  *
    498  * dirloop:
    499  *	identify next component of name at ndp->ni_ptr
    500  *	handle degenerate case where name is null string
    501  *	if .. and crossing mount points and on mounted filesys, find parent
    502  *	call VOP_LOOKUP routine for next component name
    503  *	    directory vnode returned in ni_dvp, locked.
    504  *	    component vnode returned in ni_vp (if it exists), locked.
    505  *	if result vnode is mounted on and crossing mount points,
    506  *	    find mounted on vnode
    507  *	if more components of name, do next level at dirloop
    508  *	return the answer in ni_vp, locked if LOCKLEAF set
    509  *	    if LOCKPARENT set, return locked parent in ni_dvp
    510  */
    511 int
    512 lookup(struct nameidata *ndp)
    513 {
    514 	const char *cp;			/* pointer into pathname argument */
    515 	struct vnode *dp = 0;		/* the directory we are searching */
    516 	struct vnode *tdp;		/* saved dp */
    517 	struct mount *mp;		/* mount table entry */
    518 	int docache;			/* == 0 do not cache last component */
    519 	int rdonly;			/* lookup read-only flag bit */
    520 	int error = 0;
    521 	int slashes;
    522 	struct componentname *cnp = &ndp->ni_cnd;
    523 	struct lwp *l = cnp->cn_lwp;
    524 
    525 	/*
    526 	 * Setup: break out flag bits into variables.
    527 	 */
    528 	docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
    529 	if (cnp->cn_nameiop == DELETE)
    530 		docache = 0;
    531 	rdonly = cnp->cn_flags & RDONLY;
    532 	ndp->ni_dvp = NULL;
    533 	cnp->cn_flags &= ~ISSYMLINK;
    534 	dp = ndp->ni_startdir;
    535 	ndp->ni_startdir = NULLVP;
    536 
    537 	/*
    538 	 * If we have a leading string of slashes, remove them, and just make
    539 	 * sure the current node is a directory.
    540 	 */
    541 	cp = cnp->cn_nameptr;
    542 	if (*cp == '/') {
    543 		do {
    544 			cp++;
    545 		} while (*cp == '/');
    546 		ndp->ni_pathlen -= cp - cnp->cn_nameptr;
    547 		cnp->cn_nameptr = cp;
    548 
    549 		if (dp->v_type != VDIR) {
    550 			error = ENOTDIR;
    551 			vput(dp);
    552 			goto bad;
    553 		}
    554 
    555 		/*
    556 		 * If we've exhausted the path name, then just return the
    557 		 * current node.  If the caller requested the parent node (i.e.
    558 		 * it's a CREATE, DELETE, or RENAME), and we don't have one
    559 		 * (because this is the root directory), then we must fail.
    560 		 */
    561 		if (cnp->cn_nameptr[0] == '\0') {
    562 			if (ndp->ni_dvp == NULL && cnp->cn_nameiop != LOOKUP) {
    563 				switch (cnp->cn_nameiop) {
    564 				case CREATE:
    565 					error = EEXIST;
    566 					break;
    567 				case DELETE:
    568 				case RENAME:
    569 					error = EBUSY;
    570 					break;
    571 				default:
    572 					KASSERT(0);
    573 				}
    574 				vput(dp);
    575 				goto bad;
    576 			}
    577 			ndp->ni_vp = dp;
    578 			cnp->cn_flags |= ISLASTCN;
    579 			goto terminal;
    580 		}
    581 	}
    582 
    583 dirloop:
    584 	/*
    585 	 * Search a new directory.
    586 	 *
    587 	 * The cn_hash value is for use by vfs_cache.
    588 	 * The last component of the filename is left accessible via
    589 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    590 	 * the name set the SAVENAME flag. When done, they assume
    591 	 * responsibility for freeing the pathname buffer.
    592 	 *
    593 	 * At this point, our only vnode state is that "dp" is held and locked.
    594 	 */
    595 	cnp->cn_consume = 0;
    596 	cp = NULL;
    597 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
    598 	cnp->cn_namelen = cp - cnp->cn_nameptr;
    599 	if (cnp->cn_namelen > NAME_MAX) {
    600 		vput(dp);
    601 		error = ENAMETOOLONG;
    602 		ndp->ni_dvp = NULL;
    603 		goto bad;
    604 	}
    605 #ifdef NAMEI_DIAGNOSTIC
    606 	{ char c = *cp;
    607 	*(char *)cp = '\0';
    608 	printf("{%s}: ", cnp->cn_nameptr);
    609 	*(char *)cp = c; }
    610 #endif /* NAMEI_DIAGNOSTIC */
    611 	ndp->ni_pathlen -= cnp->cn_namelen;
    612 	ndp->ni_next = cp;
    613 	/*
    614 	 * If this component is followed by a slash, then move the pointer to
    615 	 * the next component forward, and remember that this component must be
    616 	 * a directory.
    617 	 */
    618 	if (*cp == '/') {
    619 		do {
    620 			cp++;
    621 		} while (*cp == '/');
    622 		slashes = cp - ndp->ni_next;
    623 		ndp->ni_pathlen -= slashes;
    624 		ndp->ni_next = cp;
    625 		cnp->cn_flags |= REQUIREDIR;
    626 	} else {
    627 		slashes = 0;
    628 		cnp->cn_flags &= ~REQUIREDIR;
    629 	}
    630 	/*
    631 	 * We do special processing on the last component, whether or not it's
    632 	 * a directory.  Cache all intervening lookups, but not the final one.
    633 	 */
    634 	if (*cp == '\0') {
    635 		if (docache)
    636 			cnp->cn_flags |= MAKEENTRY;
    637 		else
    638 			cnp->cn_flags &= ~MAKEENTRY;
    639 		cnp->cn_flags |= ISLASTCN;
    640 	} else {
    641 		cnp->cn_flags |= MAKEENTRY;
    642 		cnp->cn_flags &= ~ISLASTCN;
    643 	}
    644 	if (cnp->cn_namelen == 2 &&
    645 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
    646 		cnp->cn_flags |= ISDOTDOT;
    647 	else
    648 		cnp->cn_flags &= ~ISDOTDOT;
    649 
    650 	/*
    651 	 * Handle "..": two special cases.
    652 	 * 1. If at root directory (e.g. after chroot)
    653 	 *    or at absolute root directory
    654 	 *    then ignore it so can't get out.
    655 	 * 1a. If we have somehow gotten out of a jail, warn
    656 	 *    and also ignore it so we can't get farther out.
    657 	 * 2. If this vnode is the root of a mounted
    658 	 *    filesystem, then replace it with the
    659 	 *    vnode which was mounted on so we take the
    660 	 *    .. in the other file system.
    661 	 */
    662 	if (cnp->cn_flags & ISDOTDOT) {
    663 		struct proc *p = l->l_proc;
    664 
    665 		for (;;) {
    666 			if (dp == ndp->ni_rootdir || dp == rootvnode) {
    667 				ndp->ni_dvp = dp;
    668 				ndp->ni_vp = dp;
    669 				VREF(dp);
    670 				goto nextname;
    671 			}
    672 			if (ndp->ni_rootdir != rootvnode) {
    673 				int retval;
    674 
    675 				VOP_UNLOCK(dp, 0);
    676 				retval = vn_isunder(dp, ndp->ni_rootdir, l);
    677 				vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    678 				if (!retval) {
    679 				    /* Oops! We got out of jail! */
    680 				    log(LOG_WARNING,
    681 					"chrooted pid %d uid %d (%s) "
    682 					"detected outside of its chroot\n",
    683 					p->p_pid, kauth_cred_geteuid(l->l_cred),
    684 					p->p_comm);
    685 				    /* Put us at the jail root. */
    686 				    vput(dp);
    687 				    dp = ndp->ni_rootdir;
    688 				    ndp->ni_dvp = dp;
    689 				    ndp->ni_vp = dp;
    690 				    VREF(dp);
    691 				    VREF(dp);
    692 				    vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    693 				    goto nextname;
    694 				}
    695 			}
    696 			if ((dp->v_flag & VROOT) == 0 ||
    697 			    (cnp->cn_flags & NOCROSSMOUNT))
    698 				break;
    699 			tdp = dp;
    700 			dp = dp->v_mount->mnt_vnodecovered;
    701 			vput(tdp);
    702 			VREF(dp);
    703 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    704 		}
    705 	}
    706 
    707 	/*
    708 	 * We now have a segment name to search for, and a directory to search.
    709 	 * Again, our only vnode state is that "dp" is held and locked.
    710 	 */
    711 unionlookup:
    712 	ndp->ni_dvp = dp;
    713 	ndp->ni_vp = NULL;
    714 	error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
    715 	if (error != 0) {
    716 #ifdef DIAGNOSTIC
    717 		if (ndp->ni_vp != NULL)
    718 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    719 #endif /* DIAGNOSTIC */
    720 #ifdef NAMEI_DIAGNOSTIC
    721 		printf("not found\n");
    722 #endif /* NAMEI_DIAGNOSTIC */
    723 		if ((error == ENOENT) &&
    724 		    (dp->v_flag & VROOT) &&
    725 		    (dp->v_mount->mnt_flag & MNT_UNION)) {
    726 			tdp = dp;
    727 			dp = dp->v_mount->mnt_vnodecovered;
    728 			vput(tdp);
    729 			VREF(dp);
    730 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    731 			goto unionlookup;
    732 		}
    733 
    734 		if (error != EJUSTRETURN)
    735 			goto bad;
    736 
    737 		/*
    738 		 * If this was not the last component, or there were trailing
    739 		 * slashes, and we are not going to create a directory,
    740 		 * then the name must exist.
    741 		 */
    742 		if ((cnp->cn_flags & (REQUIREDIR | CREATEDIR)) == REQUIREDIR) {
    743 			error = ENOENT;
    744 			goto bad;
    745 		}
    746 
    747 		/*
    748 		 * If creating and at end of pathname, then can consider
    749 		 * allowing file to be created.
    750 		 */
    751 		if (rdonly) {
    752 			error = EROFS;
    753 			goto bad;
    754 		}
    755 
    756 		/*
    757 		 * We return with ni_vp NULL to indicate that the entry
    758 		 * doesn't currently exist, leaving a pointer to the
    759 		 * (possibly locked) directory vnode in ndp->ni_dvp.
    760 		 */
    761 		if (cnp->cn_flags & SAVESTART) {
    762 			ndp->ni_startdir = ndp->ni_dvp;
    763 			VREF(ndp->ni_startdir);
    764 		}
    765 		return (0);
    766 	}
    767 #ifdef NAMEI_DIAGNOSTIC
    768 	printf("found\n");
    769 #endif /* NAMEI_DIAGNOSTIC */
    770 
    771 	/*
    772 	 * Take into account any additional components consumed by the
    773 	 * underlying filesystem.  This will include any trailing slashes after
    774 	 * the last component consumed.
    775 	 */
    776 	if (cnp->cn_consume > 0) {
    777 		ndp->ni_pathlen -= cnp->cn_consume - slashes;
    778 		ndp->ni_next += cnp->cn_consume - slashes;
    779 		cnp->cn_consume = 0;
    780 		if (ndp->ni_next[0] == '\0')
    781 			cnp->cn_flags |= ISLASTCN;
    782 	}
    783 
    784 	dp = ndp->ni_vp;
    785 
    786 	/*
    787 	 * "dp" and "ndp->ni_dvp" are both locked and held,
    788 	 * and may be the same vnode.
    789 	 */
    790 
    791 	/*
    792 	 * Check to see if the vnode has been mounted on;
    793 	 * if so find the root of the mounted file system.
    794 	 */
    795 	while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
    796 	       (cnp->cn_flags & NOCROSSMOUNT) == 0) {
    797 		if (vfs_busy(mp, 0, 0))
    798 			continue;
    799 
    800 		KASSERT(ndp->ni_dvp != dp);
    801 		VOP_UNLOCK(ndp->ni_dvp, 0);
    802 		vput(dp);
    803 		error = VFS_ROOT(mp, &tdp);
    804 		vfs_unbusy(mp);
    805 		if (error) {
    806 			vn_lock(ndp->ni_dvp, LK_EXCLUSIVE | LK_RETRY);
    807 			goto bad;
    808 		}
    809 		VOP_UNLOCK(tdp, 0);
    810 		ndp->ni_vp = dp = tdp;
    811 		vn_lock(ndp->ni_dvp, LK_EXCLUSIVE | LK_RETRY);
    812 		vn_lock(ndp->ni_vp, LK_EXCLUSIVE | LK_RETRY);
    813 	}
    814 
    815 	/*
    816 	 * Check for symbolic link.  Back up over any slashes that we skipped,
    817 	 * as we will need them again.
    818 	 */
    819 	if ((dp->v_type == VLNK) && (cnp->cn_flags & (FOLLOW|REQUIREDIR))) {
    820 		ndp->ni_pathlen += slashes;
    821 		ndp->ni_next -= slashes;
    822 		cnp->cn_flags |= ISSYMLINK;
    823 		return (0);
    824 	}
    825 
    826 	/*
    827 	 * Check for directory, if the component was followed by a series of
    828 	 * slashes.
    829 	 */
    830 	if ((dp->v_type != VDIR) && (cnp->cn_flags & REQUIREDIR)) {
    831 		error = ENOTDIR;
    832 		KASSERT(dp != ndp->ni_dvp);
    833 		vput(dp);
    834 		goto bad;
    835 	}
    836 
    837 nextname:
    838 
    839 	/*
    840 	 * Not a symbolic link.  If this was not the last component, then
    841 	 * continue at the next component, else return.
    842 	 */
    843 	if (!(cnp->cn_flags & ISLASTCN)) {
    844 		cnp->cn_nameptr = ndp->ni_next;
    845 		if (ndp->ni_dvp == dp) {
    846 			vrele(ndp->ni_dvp);
    847 		} else {
    848 			vput(ndp->ni_dvp);
    849 		}
    850 		goto dirloop;
    851 	}
    852 
    853 terminal:
    854 
    855 	/*
    856 	 * Disallow directory write attempts on read-only file systems.
    857 	 */
    858 	if (rdonly &&
    859 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    860 
    861 		/*
    862 		 * Disallow directory write attempts on read-only
    863 		 * file systems.
    864 		 */
    865 		error = EROFS;
    866 		if (dp != ndp->ni_dvp) {
    867 			vput(dp);
    868 		}
    869 		goto bad;
    870 	}
    871 	if (ndp->ni_dvp != NULL) {
    872 		if (cnp->cn_flags & SAVESTART) {
    873 			ndp->ni_startdir = ndp->ni_dvp;
    874 			VREF(ndp->ni_startdir);
    875 		}
    876 	}
    877 	if ((cnp->cn_flags & LOCKLEAF) == 0) {
    878 		VOP_UNLOCK(dp, 0);
    879 	}
    880 	return (0);
    881 
    882 bad:
    883 	ndp->ni_vp = NULL;
    884 	return (error);
    885 }
    886 
    887 /*
    888  * Reacquire a path name component.
    889  * dvp is locked on entry and exit.
    890  * *vpp is locked on exit unless it's NULL.
    891  */
    892 int
    893 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
    894 {
    895 	int rdonly;			/* lookup read-only flag bit */
    896 	int error = 0;
    897 #ifdef DEBUG
    898 	uint32_t newhash;		/* DEBUG: check name hash */
    899 	const char *cp;			/* DEBUG: check name ptr/len */
    900 #endif /* DEBUG */
    901 
    902 	/*
    903 	 * Setup: break out flag bits into variables.
    904 	 */
    905 	rdonly = cnp->cn_flags & RDONLY;
    906 	cnp->cn_flags &= ~ISSYMLINK;
    907 
    908 	/*
    909 	 * Search a new directory.
    910 	 *
    911 	 * The cn_hash value is for use by vfs_cache.
    912 	 * The last component of the filename is left accessible via
    913 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    914 	 * the name set the SAVENAME flag. When done, they assume
    915 	 * responsibility for freeing the pathname buffer.
    916 	 */
    917 #ifdef DEBUG
    918 	cp = NULL;
    919 	newhash = namei_hash(cnp->cn_nameptr, &cp);
    920 	if ((uint32_t)newhash != (uint32_t)cnp->cn_hash)
    921 		panic("relookup: bad hash");
    922 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
    923 		panic("relookup: bad len");
    924 	while (*cp == '/')
    925 		cp++;
    926 	if (*cp != 0)
    927 		panic("relookup: not last component");
    928 #endif /* DEBUG */
    929 
    930 	/*
    931 	 * Check for degenerate name (e.g. / or "")
    932 	 * which is a way of talking about a directory,
    933 	 * e.g. like "/." or ".".
    934 	 */
    935 	if (cnp->cn_nameptr[0] == '\0')
    936 		panic("relookup: null name");
    937 
    938 	if (cnp->cn_flags & ISDOTDOT)
    939 		panic("relookup: lookup on dot-dot");
    940 
    941 	/*
    942 	 * We now have a segment name to search for, and a directory to search.
    943 	 */
    944 	if ((error = VOP_LOOKUP(dvp, vpp, cnp)) != 0) {
    945 #ifdef DIAGNOSTIC
    946 		if (*vpp != NULL)
    947 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    948 #endif
    949 		if (error != EJUSTRETURN)
    950 			goto bad;
    951 	}
    952 
    953 #ifdef DIAGNOSTIC
    954 	/*
    955 	 * Check for symbolic link
    956 	 */
    957 	if (*vpp && (*vpp)->v_type == VLNK && (cnp->cn_flags & FOLLOW))
    958 		panic("relookup: symlink found");
    959 #endif
    960 
    961 	/*
    962 	 * Check for read-only file systems.
    963 	 */
    964 	if (rdonly && cnp->cn_nameiop != LOOKUP) {
    965 		error = EROFS;
    966 		if (*vpp) {
    967 			vput(*vpp);
    968 		}
    969 		goto bad;
    970 	}
    971 	if (cnp->cn_flags & SAVESTART)
    972 		VREF(dvp);
    973 	return (0);
    974 
    975 bad:
    976 	*vpp = NULL;
    977 	return (error);
    978 }
    979