Home | History | Annotate | Line # | Download | only in kern
vfs_lookup.c revision 1.99
      1 /*	$NetBSD: vfs_lookup.c,v 1.99 2007/11/07 00:23:25 ad 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 1.99 2007/11/07 00:23:25 ad Exp $");
     41 
     42 #include "opt_systrace.h"
     43 #include "opt_magiclinks.h"
     44 
     45 #include <sys/param.h>
     46 #include <sys/systm.h>
     47 #include <sys/kernel.h>
     48 #include <sys/syslimits.h>
     49 #include <sys/time.h>
     50 #include <sys/namei.h>
     51 #include <sys/vnode.h>
     52 #include <sys/mount.h>
     53 #include <sys/errno.h>
     54 #include <sys/filedesc.h>
     55 #include <sys/hash.h>
     56 #include <sys/malloc.h>
     57 #include <sys/proc.h>
     58 #include <sys/syslog.h>
     59 #include <sys/kauth.h>
     60 #include <sys/ktrace.h>
     61 
     62 #ifdef SYSTRACE
     63 #include <sys/systrace.h>
     64 #endif
     65 
     66 #ifndef MAGICLINKS
     67 #define MAGICLINKS 0
     68 #endif
     69 
     70 struct pathname_internal {
     71 	char *pathbuf;
     72 	bool needfree;
     73 };
     74 
     75 int vfs_magiclinks = MAGICLINKS;
     76 
     77 pool_cache_t pnbuf_cache;	/* pathname buffer cache */
     78 
     79 /*
     80  * Substitute replacement text for 'magic' strings in symlinks.
     81  * Returns 0 if successful, and returns non-zero if an error
     82  * occurs.  (Currently, the only possible error is running out
     83  * of temporary pathname space.)
     84  *
     85  * Looks for "@<string>" and "@<string>/", where <string> is a
     86  * recognized 'magic' string.  Replaces the "@<string>" with the
     87  * appropriate replacement text.  (Note that in some cases the
     88  * replacement text may have zero length.)
     89  *
     90  * This would have been table driven, but the variance in
     91  * replacement strings (and replacement string lengths) made
     92  * that impractical.
     93  */
     94 #define	VNL(x)							\
     95 	(sizeof(x) - 1)
     96 
     97 #define	VO	'{'
     98 #define	VC	'}'
     99 
    100 #define	MATCH(str)						\
    101 	((termchar == '/' && i + VNL(str) == *len) ||		\
    102 	 (i + VNL(str) < *len &&				\
    103 	  cp[i + VNL(str)] == termchar)) &&			\
    104 	!strncmp((str), &cp[i], VNL(str))
    105 
    106 #define	SUBSTITUTE(m, s, sl)					\
    107 	if ((newlen + (sl)) > MAXPATHLEN)			\
    108 		return (1);					\
    109 	i += VNL(m);						\
    110 	if (termchar != '/')					\
    111 		i++;						\
    112 	memcpy(&tmp[newlen], (s), (sl));			\
    113 	newlen += (sl);						\
    114 	change = 1;						\
    115 	termchar = '/';
    116 
    117 static int
    118 symlink_magic(struct proc *p, char *cp, int *len)
    119 {
    120 	char *tmp;
    121 	int change, i, newlen;
    122 	int termchar = '/';
    123 
    124 	tmp = PNBUF_GET();
    125 	for (change = i = newlen = 0; i < *len; ) {
    126 		if (cp[i] != '@') {
    127 			tmp[newlen++] = cp[i++];
    128 			continue;
    129 		}
    130 
    131 		i++;
    132 
    133 		/* Check for @{var} syntax. */
    134 		if (cp[i] == VO) {
    135 			termchar = VC;
    136 			i++;
    137 		}
    138 
    139 		/*
    140 		 * The following checks should be ordered according
    141 		 * to frequency of use.
    142 		 */
    143 		if (MATCH("machine_arch")) {
    144 			SUBSTITUTE("machine_arch", MACHINE_ARCH,
    145 			    sizeof(MACHINE_ARCH) - 1);
    146 		} else if (MATCH("machine")) {
    147 			SUBSTITUTE("machine", MACHINE,
    148 			    sizeof(MACHINE) - 1);
    149 		} else if (MATCH("hostname")) {
    150 			SUBSTITUTE("hostname", hostname,
    151 			    hostnamelen);
    152 		} else if (MATCH("osrelease")) {
    153 			SUBSTITUTE("osrelease", osrelease,
    154 			    strlen(osrelease));
    155 		} else if (MATCH("emul")) {
    156 			SUBSTITUTE("emul", p->p_emul->e_name,
    157 			    strlen(p->p_emul->e_name));
    158 		} else if (MATCH("kernel_ident")) {
    159 			SUBSTITUTE("kernel_ident", kernel_ident,
    160 			    strlen(kernel_ident));
    161 		} else if (MATCH("domainname")) {
    162 			SUBSTITUTE("domainname", domainname,
    163 			    domainnamelen);
    164 		} else if (MATCH("ostype")) {
    165 			SUBSTITUTE("ostype", ostype,
    166 			    strlen(ostype));
    167 		} else if (MATCH("uid")) {
    168 			char uidtmp[11]; /* XXX elad */
    169 
    170 			(void)snprintf(uidtmp, sizeof(uidtmp), "%u",
    171 			    kauth_cred_geteuid(kauth_cred_get()));
    172 			SUBSTITUTE("uid", uidtmp, strlen(uidtmp));
    173 		} else {
    174 			tmp[newlen++] = '@';
    175 			if (termchar == VC)
    176 				tmp[newlen++] = VO;
    177 		}
    178 	}
    179 
    180 	if (change) {
    181 		memcpy(cp, tmp, newlen);
    182 		*len = newlen;
    183 	}
    184 	PNBUF_PUT(tmp);
    185 
    186 	return (0);
    187 }
    188 
    189 #undef VNL
    190 #undef VO
    191 #undef VC
    192 #undef MATCH
    193 #undef SUBSTITUTE
    194 
    195 /*
    196  * Convert a pathname into a pointer to a locked vnode.
    197  *
    198  * The FOLLOW flag is set when symbolic links are to be followed
    199  * when they occur at the end of the name translation process.
    200  * Symbolic links are always followed for all other pathname
    201  * components other than the last.
    202  *
    203  * The segflg defines whether the name is to be copied from user
    204  * space or kernel space.
    205  *
    206  * Overall outline of namei:
    207  *
    208  *	copy in name
    209  *	get starting directory
    210  *	while (!done && !error) {
    211  *		call lookup to search path.
    212  *		if symbolic link, massage name in buffer and continue
    213  *	}
    214  */
    215 int
    216 namei(struct nameidata *ndp)
    217 {
    218 	struct cwdinfo *cwdi;		/* pointer to cwd state */
    219 	char *cp;			/* pointer into pathname argument */
    220 	struct vnode *dp;		/* the directory we are searching */
    221 	struct iovec aiov;		/* uio for reading symbolic links */
    222 	struct uio auio;
    223 	int error, linklen;
    224 	struct componentname *cnp = &ndp->ni_cnd;
    225 
    226 #ifdef DIAGNOSTIC
    227 	if (!cnp->cn_cred || !cnp->cn_lwp)
    228 		panic("namei: bad cred/proc");
    229 	if (cnp->cn_nameiop & (~OPMASK))
    230 		panic("namei: nameiop contaminated with flags");
    231 	if (cnp->cn_flags & OPMASK)
    232 		panic("namei: flags contaminated with nameiops");
    233 #endif
    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     emul_retry:
    242 	if (ndp->ni_segflg == UIO_SYSSPACE)
    243 		error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
    244 			    MAXPATHLEN, &ndp->ni_pathlen);
    245 	else
    246 		error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
    247 			    MAXPATHLEN, &ndp->ni_pathlen);
    248 
    249 	/*
    250 	 * POSIX.1 requirement: "" is not a valid file name.
    251 	 */
    252 	if (!error && ndp->ni_pathlen == 1)
    253 		error = ENOENT;
    254 
    255 	if (error) {
    256 		PNBUF_PUT(cnp->cn_pnbuf);
    257 		ndp->ni_vp = NULL;
    258 		return (error);
    259 	}
    260 	ndp->ni_loopcnt = 0;
    261 
    262 	/*
    263 	 * Get root directory for the translation.
    264 	 */
    265 	cwdi = cnp->cn_lwp->l_proc->p_cwdi;
    266 	rw_enter(&cwdi->cwdi_lock, RW_READER);
    267 	dp = cwdi->cwdi_rdir;
    268 	if (dp == NULL)
    269 		dp = rootvnode;
    270 	ndp->ni_rootdir = dp;
    271 
    272 	/*
    273 	 * Check if starting from root directory or current directory.
    274 	 */
    275 	if (cnp->cn_pnbuf[0] == '/') {
    276 		if (cnp->cn_flags & TRYEMULROOT) {
    277 			if (cnp->cn_flags & EMULROOTSET) {
    278 				/* Called from (eg) emul_find_interp() */
    279 				dp = ndp->ni_erootdir;
    280 			} else {
    281 				if (cwdi->cwdi_edir == NULL
    282 				    || (cnp->cn_pnbuf[1] == '.'
    283 					   && cnp->cn_pnbuf[2] == '.'
    284 					   && cnp->cn_pnbuf[3] == '/')) {
    285 					ndp->ni_erootdir = NULL;
    286 				} else {
    287 					dp = cwdi->cwdi_edir;
    288 					ndp->ni_erootdir = dp;
    289 				}
    290 			}
    291 		} else
    292 			ndp->ni_erootdir = NULL;
    293 	} else {
    294 		dp = cwdi->cwdi_cdir;
    295 		ndp->ni_erootdir = NULL;
    296 	}
    297 	VREF(dp);
    298 	rw_exit(&cwdi->cwdi_lock);
    299 
    300 	if (ktrpoint(KTR_NAMEI)) {
    301 		if (ndp->ni_erootdir != NULL) {
    302 			/*
    303 			 * To make any sense, the trace entry need to have the
    304 			 * text of the emulation path prepended.
    305 			 * Usually we can get this from the current process,
    306 			 * but when called from emul_find_interp() it is only
    307 			 * in the exec_package - so we get it passed in ni_next
    308 			 * (this is a hack).
    309 			 */
    310 			const char *emul_path;
    311 			if (cnp->cn_flags & EMULROOTSET)
    312 				emul_path = ndp->ni_next;
    313 			else
    314 				emul_path = cnp->cn_lwp->l_proc->p_emul->e_path;
    315 			ktrnamei2(emul_path, strlen(emul_path),
    316 			    cnp->cn_pnbuf, ndp->ni_pathlen);
    317 		} else
    318 			ktrnamei(cnp->cn_pnbuf, ndp->ni_pathlen);
    319 	}
    320 
    321 #ifdef SYSTRACE
    322 	if (ISSET(cnp->cn_lwp->l_proc->p_flag, PK_SYSTRACE))
    323 		systrace_namei(ndp);
    324 #endif
    325 
    326 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    327 	/* Loop through symbolic links */
    328 	for (;;) {
    329 		if (!dp->v_mount) {
    330 			/* Give up if the directory is no longer mounted */
    331 			vput(dp);
    332 			PNBUF_PUT(cnp->cn_pnbuf);
    333 			return (ENOENT);
    334 		}
    335 		cnp->cn_nameptr = cnp->cn_pnbuf;
    336 		ndp->ni_startdir = dp;
    337 		error = lookup(ndp);
    338 		if (error != 0) {
    339 			if (ndp->ni_dvp) {
    340 				vput(ndp->ni_dvp);
    341 			}
    342 			if (ndp->ni_erootdir != NULL) {
    343 				/* Retry the whole thing from the normal root */
    344 				cnp->cn_flags &= ~TRYEMULROOT;
    345 				goto emul_retry;
    346 			}
    347 			PNBUF_PUT(cnp->cn_pnbuf);
    348 			return (error);
    349 		}
    350 
    351 		/*
    352 		 * Check for symbolic link
    353 		 */
    354 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
    355 			if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) {
    356 				if (ndp->ni_dvp == ndp->ni_vp) {
    357 					vrele(ndp->ni_dvp);
    358 				} else {
    359 					vput(ndp->ni_dvp);
    360 				}
    361 			}
    362 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
    363 				PNBUF_PUT(cnp->cn_pnbuf);
    364 			else
    365 				cnp->cn_flags |= HASBUF;
    366 			return (0);
    367 		}
    368 
    369 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
    370 			error = ELOOP;
    371 			break;
    372 		}
    373 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
    374 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
    375 			    cnp->cn_lwp);
    376 			if (error != 0)
    377 				break;
    378 		}
    379 		if (ndp->ni_pathlen > 1)
    380 			cp = PNBUF_GET();
    381 		else
    382 			cp = cnp->cn_pnbuf;
    383 		aiov.iov_base = cp;
    384 		aiov.iov_len = MAXPATHLEN;
    385 		auio.uio_iov = &aiov;
    386 		auio.uio_iovcnt = 1;
    387 		auio.uio_offset = 0;
    388 		auio.uio_rw = UIO_READ;
    389 		auio.uio_resid = MAXPATHLEN;
    390 		UIO_SETUP_SYSSPACE(&auio);
    391 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
    392 		if (error) {
    393 badlink:
    394 			if (ndp->ni_pathlen > 1)
    395 				PNBUF_PUT(cp);
    396 			break;
    397 		}
    398 		linklen = MAXPATHLEN - auio.uio_resid;
    399 		if (linklen == 0) {
    400 			error = ENOENT;
    401 			goto badlink;
    402 		}
    403 
    404 		/*
    405 		 * Do symlink substitution, if appropriate, and
    406 		 * check length for potential overflow.
    407 		 */
    408 		if ((vfs_magiclinks &&
    409 		     symlink_magic(cnp->cn_lwp->l_proc, cp, &linklen)) ||
    410 		    (linklen + ndp->ni_pathlen >= MAXPATHLEN)) {
    411 			error = ENAMETOOLONG;
    412 			goto badlink;
    413 		}
    414 		if (ndp->ni_pathlen > 1) {
    415 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
    416 			PNBUF_PUT(cnp->cn_pnbuf);
    417 			cnp->cn_pnbuf = cp;
    418 		} else
    419 			cnp->cn_pnbuf[linklen] = '\0';
    420 		ndp->ni_pathlen += linklen;
    421 		vput(ndp->ni_vp);
    422 		dp = ndp->ni_dvp;
    423 
    424 		/*
    425 		 * Check if root directory should replace current directory.
    426 		 */
    427 		if (cnp->cn_pnbuf[0] == '/') {
    428 			vput(dp);
    429 			/* Keep absolute symbolic links inside emulation root */
    430 			dp = ndp->ni_erootdir;
    431 			if (dp == NULL || (cnp->cn_pnbuf[1] == '.'
    432 			    && cnp->cn_pnbuf[2] == '.'
    433 			    && cnp->cn_pnbuf[3] == '/')) {
    434 				ndp->ni_erootdir = NULL;
    435 				dp = ndp->ni_rootdir;
    436 			}
    437 			VREF(dp);
    438 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    439 		}
    440 	}
    441 	/* Failed to process a symbolic link */
    442 	KASSERT(ndp->ni_dvp != ndp->ni_vp);
    443 	vput(ndp->ni_dvp);
    444 	vput(ndp->ni_vp);
    445 	ndp->ni_vp = NULL;
    446 	PNBUF_PUT(cnp->cn_pnbuf);
    447 	return (error);
    448 }
    449 
    450 /*
    451  * Determine the namei hash (for cn_hash) for name.
    452  * If *ep != NULL, hash from name to ep-1.
    453  * If *ep == NULL, hash from name until the first NUL or '/', and
    454  * return the location of this termination character in *ep.
    455  *
    456  * This function returns an equivalent hash to the MI hash32_strn().
    457  * The latter isn't used because in the *ep == NULL case, determining
    458  * the length of the string to the first NUL or `/' and then calling
    459  * hash32_strn() involves unnecessary double-handling of the data.
    460  */
    461 uint32_t
    462 namei_hash(const char *name, const char **ep)
    463 {
    464 	uint32_t	hash;
    465 
    466 	hash = HASH32_STR_INIT;
    467 	if (*ep != NULL) {
    468 		for (; name < *ep; name++)
    469 			hash = hash * 33 + *(const uint8_t *)name;
    470 	} else {
    471 		for (; *name != '\0' && *name != '/'; name++)
    472 			hash = hash * 33 + *(const uint8_t *)name;
    473 		*ep = name;
    474 	}
    475 	return (hash + (hash >> 5));
    476 }
    477 
    478 /*
    479  * Search a pathname.
    480  * This is a very central and rather complicated routine.
    481  *
    482  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
    483  * The starting directory is taken from ni_startdir. The pathname is
    484  * descended until done, or a symbolic link is encountered. The variable
    485  * ni_more is clear if the path is completed; it is set to one if a
    486  * symbolic link needing interpretation is encountered.
    487  *
    488  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
    489  * whether the name is to be looked up, created, renamed, or deleted.
    490  * When CREATE, RENAME, or DELETE is specified, information usable in
    491  * creating, renaming, or deleting a directory entry may be calculated.
    492  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
    493  * locked.  Otherwise the parent directory is not returned. If the target
    494  * of the pathname exists and LOCKLEAF is or'ed into the flag the target
    495  * is returned locked, otherwise it is returned unlocked.  When creating
    496  * or renaming and LOCKPARENT is specified, the target may not be ".".
    497  * When deleting and LOCKPARENT is specified, the target may be ".".
    498  *
    499  * Overall outline of lookup:
    500  *
    501  * dirloop:
    502  *	identify next component of name at ndp->ni_ptr
    503  *	handle degenerate case where name is null string
    504  *	if .. and crossing mount points and on mounted filesys, find parent
    505  *	call VOP_LOOKUP routine for next component name
    506  *	    directory vnode returned in ni_dvp, locked.
    507  *	    component vnode returned in ni_vp (if it exists), locked.
    508  *	if result vnode is mounted on and crossing mount points,
    509  *	    find mounted on vnode
    510  *	if more components of name, do next level at dirloop
    511  *	return the answer in ni_vp, locked if LOCKLEAF set
    512  *	    if LOCKPARENT set, return locked parent in ni_dvp
    513  */
    514 int
    515 lookup(struct nameidata *ndp)
    516 {
    517 	const char *cp;			/* pointer into pathname argument */
    518 	struct vnode *dp = 0;		/* the directory we are searching */
    519 	struct vnode *tdp;		/* saved dp */
    520 	struct mount *mp;		/* mount table entry */
    521 	int docache;			/* == 0 do not cache last component */
    522 	int rdonly;			/* lookup read-only flag bit */
    523 	int error = 0;
    524 	int slashes;
    525 	struct componentname *cnp = &ndp->ni_cnd;
    526 	struct lwp *l = cnp->cn_lwp;
    527 
    528 	/*
    529 	 * Setup: break out flag bits into variables.
    530 	 */
    531 	docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
    532 	if (cnp->cn_nameiop == DELETE)
    533 		docache = 0;
    534 	rdonly = cnp->cn_flags & RDONLY;
    535 	ndp->ni_dvp = NULL;
    536 	cnp->cn_flags &= ~ISSYMLINK;
    537 	dp = ndp->ni_startdir;
    538 	ndp->ni_startdir = NULLVP;
    539 
    540 	/*
    541 	 * If we have a leading string of slashes, remove them, and just make
    542 	 * sure the current node is a directory.
    543 	 */
    544 	cp = cnp->cn_nameptr;
    545 	if (*cp == '/') {
    546 		do {
    547 			cp++;
    548 		} while (*cp == '/');
    549 		ndp->ni_pathlen -= cp - cnp->cn_nameptr;
    550 		cnp->cn_nameptr = cp;
    551 
    552 		if (dp->v_type != VDIR) {
    553 			error = ENOTDIR;
    554 			vput(dp);
    555 			goto bad;
    556 		}
    557 
    558 		/*
    559 		 * If we've exhausted the path name, then just return the
    560 		 * current node.
    561 		 */
    562 		if (cnp->cn_nameptr[0] == '\0') {
    563 			ndp->ni_vp = dp;
    564 			cnp->cn_flags |= ISLASTCN;
    565 			goto terminal;
    566 		}
    567 	}
    568 
    569 dirloop:
    570 	/*
    571 	 * Search a new directory.
    572 	 *
    573 	 * The cn_hash value is for use by vfs_cache.
    574 	 * The last component of the filename is left accessible via
    575 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    576 	 * the name set the SAVENAME flag. When done, they assume
    577 	 * responsibility for freeing the pathname buffer.
    578 	 *
    579 	 * At this point, our only vnode state is that "dp" is held and locked.
    580 	 */
    581 	cnp->cn_consume = 0;
    582 	cp = NULL;
    583 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
    584 	cnp->cn_namelen = cp - cnp->cn_nameptr;
    585 	if (cnp->cn_namelen > NAME_MAX) {
    586 		vput(dp);
    587 		error = ENAMETOOLONG;
    588 		ndp->ni_dvp = NULL;
    589 		goto bad;
    590 	}
    591 #ifdef NAMEI_DIAGNOSTIC
    592 	{ char c = *cp;
    593 	*(char *)cp = '\0';
    594 	printf("{%s}: ", cnp->cn_nameptr);
    595 	*(char *)cp = c; }
    596 #endif /* NAMEI_DIAGNOSTIC */
    597 	ndp->ni_pathlen -= cnp->cn_namelen;
    598 	ndp->ni_next = cp;
    599 	/*
    600 	 * If this component is followed by a slash, then move the pointer to
    601 	 * the next component forward, and remember that this component must be
    602 	 * a directory.
    603 	 */
    604 	if (*cp == '/') {
    605 		do {
    606 			cp++;
    607 		} while (*cp == '/');
    608 		slashes = cp - ndp->ni_next;
    609 		ndp->ni_pathlen -= slashes;
    610 		ndp->ni_next = cp;
    611 		cnp->cn_flags |= REQUIREDIR;
    612 	} else {
    613 		slashes = 0;
    614 		cnp->cn_flags &= ~REQUIREDIR;
    615 	}
    616 	/*
    617 	 * We do special processing on the last component, whether or not it's
    618 	 * a directory.  Cache all intervening lookups, but not the final one.
    619 	 */
    620 	if (*cp == '\0') {
    621 		if (docache)
    622 			cnp->cn_flags |= MAKEENTRY;
    623 		else
    624 			cnp->cn_flags &= ~MAKEENTRY;
    625 		cnp->cn_flags |= ISLASTCN;
    626 	} else {
    627 		cnp->cn_flags |= MAKEENTRY;
    628 		cnp->cn_flags &= ~ISLASTCN;
    629 	}
    630 	if (cnp->cn_namelen == 2 &&
    631 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
    632 		cnp->cn_flags |= ISDOTDOT;
    633 	else
    634 		cnp->cn_flags &= ~ISDOTDOT;
    635 
    636 	/*
    637 	 * Handle "..": two special cases.
    638 	 * 1. If at root directory (e.g. after chroot)
    639 	 *    or at absolute root directory
    640 	 *    then ignore it so can't get out.
    641 	 * 1a. If at the root of the emulation filesystem go to the real
    642 	 *    root. So "/../<path>" is always absolute.
    643 	 * 1b. If we have somehow gotten out of a jail, warn
    644 	 *    and also ignore it so we can't get farther out.
    645 	 * 2. If this vnode is the root of a mounted
    646 	 *    filesystem, then replace it with the
    647 	 *    vnode which was mounted on so we take the
    648 	 *    .. in the other file system.
    649 	 */
    650 	if (cnp->cn_flags & ISDOTDOT) {
    651 		struct proc *p = l->l_proc;
    652 
    653 		for (;;) {
    654 			if (dp == ndp->ni_rootdir || dp == rootvnode) {
    655 				ndp->ni_dvp = dp;
    656 				ndp->ni_vp = dp;
    657 				VREF(dp);
    658 				goto nextname;
    659 			}
    660 			if (ndp->ni_rootdir != rootvnode) {
    661 				int retval;
    662 
    663 				VOP_UNLOCK(dp, 0);
    664 				retval = vn_isunder(dp, ndp->ni_rootdir, l);
    665 				vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    666 				if (!retval) {
    667 				    /* Oops! We got out of jail! */
    668 				    log(LOG_WARNING,
    669 					"chrooted pid %d uid %d (%s) "
    670 					"detected outside of its chroot\n",
    671 					p->p_pid, kauth_cred_geteuid(l->l_cred),
    672 					p->p_comm);
    673 				    /* Put us at the jail root. */
    674 				    vput(dp);
    675 				    dp = ndp->ni_rootdir;
    676 				    ndp->ni_dvp = dp;
    677 				    ndp->ni_vp = dp;
    678 				    VREF(dp);
    679 				    VREF(dp);
    680 				    vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    681 				    goto nextname;
    682 				}
    683 			}
    684 			if ((dp->v_vflag & VV_ROOT) == 0 ||
    685 			    (cnp->cn_flags & NOCROSSMOUNT))
    686 				break;
    687 			tdp = dp;
    688 			dp = dp->v_mount->mnt_vnodecovered;
    689 			vput(tdp);
    690 			VREF(dp);
    691 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    692 		}
    693 	}
    694 
    695 	/*
    696 	 * We now have a segment name to search for, and a directory to search.
    697 	 * Again, our only vnode state is that "dp" is held and locked.
    698 	 */
    699 unionlookup:
    700 	ndp->ni_dvp = dp;
    701 	ndp->ni_vp = NULL;
    702 	error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
    703 	if (error != 0) {
    704 #ifdef DIAGNOSTIC
    705 		if (ndp->ni_vp != NULL)
    706 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    707 #endif /* DIAGNOSTIC */
    708 #ifdef NAMEI_DIAGNOSTIC
    709 		printf("not found\n");
    710 #endif /* NAMEI_DIAGNOSTIC */
    711 		if ((error == ENOENT) &&
    712 		    (dp->v_vflag & VV_ROOT) &&
    713 		    (dp->v_mount->mnt_flag & MNT_UNION)) {
    714 			tdp = dp;
    715 			dp = dp->v_mount->mnt_vnodecovered;
    716 			vput(tdp);
    717 			VREF(dp);
    718 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    719 			goto unionlookup;
    720 		}
    721 
    722 		if (error != EJUSTRETURN)
    723 			goto bad;
    724 
    725 		/*
    726 		 * If this was not the last component, or there were trailing
    727 		 * slashes, and we are not going to create a directory,
    728 		 * then the name must exist.
    729 		 */
    730 		if ((cnp->cn_flags & (REQUIREDIR | CREATEDIR)) == REQUIREDIR) {
    731 			error = ENOENT;
    732 			goto bad;
    733 		}
    734 
    735 		/*
    736 		 * If creating and at end of pathname, then can consider
    737 		 * allowing file to be created.
    738 		 */
    739 		if (rdonly) {
    740 			error = EROFS;
    741 			goto bad;
    742 		}
    743 
    744 		/*
    745 		 * We return with ni_vp NULL to indicate that the entry
    746 		 * doesn't currently exist, leaving a pointer to the
    747 		 * (possibly locked) directory vnode in ndp->ni_dvp.
    748 		 */
    749 		if (cnp->cn_flags & SAVESTART) {
    750 			ndp->ni_startdir = ndp->ni_dvp;
    751 			VREF(ndp->ni_startdir);
    752 		}
    753 		return (0);
    754 	}
    755 #ifdef NAMEI_DIAGNOSTIC
    756 	printf("found\n");
    757 #endif /* NAMEI_DIAGNOSTIC */
    758 
    759 	/*
    760 	 * Take into account any additional components consumed by the
    761 	 * underlying filesystem.  This will include any trailing slashes after
    762 	 * the last component consumed.
    763 	 */
    764 	if (cnp->cn_consume > 0) {
    765 		ndp->ni_pathlen -= cnp->cn_consume - slashes;
    766 		ndp->ni_next += cnp->cn_consume - slashes;
    767 		cnp->cn_consume = 0;
    768 		if (ndp->ni_next[0] == '\0')
    769 			cnp->cn_flags |= ISLASTCN;
    770 	}
    771 
    772 	dp = ndp->ni_vp;
    773 
    774 	/*
    775 	 * "dp" and "ndp->ni_dvp" are both locked and held,
    776 	 * and may be the same vnode.
    777 	 */
    778 
    779 	/*
    780 	 * Check to see if the vnode has been mounted on;
    781 	 * if so find the root of the mounted file system.
    782 	 */
    783 	while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
    784 	       (cnp->cn_flags & NOCROSSMOUNT) == 0) {
    785 		if (vfs_busy(mp, 0, 0))
    786 			continue;
    787 
    788 		KASSERT(ndp->ni_dvp != dp);
    789 		VOP_UNLOCK(ndp->ni_dvp, 0);
    790 		vput(dp);
    791 		error = VFS_ROOT(mp, &tdp);
    792 		vfs_unbusy(mp);
    793 		if (error) {
    794 			vn_lock(ndp->ni_dvp, LK_EXCLUSIVE | LK_RETRY);
    795 			goto bad;
    796 		}
    797 		VOP_UNLOCK(tdp, 0);
    798 		ndp->ni_vp = dp = tdp;
    799 		vn_lock(ndp->ni_dvp, LK_EXCLUSIVE | LK_RETRY);
    800 		vn_lock(ndp->ni_vp, LK_EXCLUSIVE | LK_RETRY);
    801 	}
    802 
    803 	/*
    804 	 * Check for symbolic link.  Back up over any slashes that we skipped,
    805 	 * as we will need them again.
    806 	 */
    807 	if ((dp->v_type == VLNK) && (cnp->cn_flags & (FOLLOW|REQUIREDIR))) {
    808 		ndp->ni_pathlen += slashes;
    809 		ndp->ni_next -= slashes;
    810 		cnp->cn_flags |= ISSYMLINK;
    811 		return (0);
    812 	}
    813 
    814 	/*
    815 	 * Check for directory, if the component was followed by a series of
    816 	 * slashes.
    817 	 */
    818 	if ((dp->v_type != VDIR) && (cnp->cn_flags & REQUIREDIR)) {
    819 		error = ENOTDIR;
    820 		KASSERT(dp != ndp->ni_dvp);
    821 		vput(dp);
    822 		goto bad;
    823 	}
    824 
    825 nextname:
    826 
    827 	/*
    828 	 * Not a symbolic link.  If this was not the last component, then
    829 	 * continue at the next component, else return.
    830 	 */
    831 	if (!(cnp->cn_flags & ISLASTCN)) {
    832 		cnp->cn_nameptr = ndp->ni_next;
    833 		if (ndp->ni_dvp == dp) {
    834 			vrele(ndp->ni_dvp);
    835 		} else {
    836 			vput(ndp->ni_dvp);
    837 		}
    838 		goto dirloop;
    839 	}
    840 
    841 terminal:
    842 	if (dp == ndp->ni_erootdir) {
    843 		/*
    844 		 * We are about to return the emulation root.
    845 		 * This isn't a good idea because code might repeatedly
    846 		 * lookup ".." until the file matches that returned
    847 		 * for "/" and loop forever.
    848 		 * So convert it to the real root.
    849 		 */
    850 		if (ndp->ni_dvp == dp)
    851 			vrele(dp);
    852 		else
    853 			if (ndp->ni_dvp != NULL)
    854 				vput(ndp->ni_dvp);
    855 		ndp->ni_dvp = NULL;
    856 		vput(dp);
    857 		dp = ndp->ni_rootdir;
    858 		VREF(dp);
    859 		vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
    860 		ndp->ni_vp = dp;
    861 	}
    862 
    863 	/*
    864 	 * If the caller requested the parent node (i.e.
    865 	 * it's a CREATE, DELETE, or RENAME), and we don't have one
    866 	 * (because this is the root directory), then we must fail.
    867 	 */
    868 	if (ndp->ni_dvp == NULL && cnp->cn_nameiop != LOOKUP) {
    869 		switch (cnp->cn_nameiop) {
    870 		case CREATE:
    871 			error = EEXIST;
    872 			break;
    873 		case DELETE:
    874 		case RENAME:
    875 			error = EBUSY;
    876 			break;
    877 		default:
    878 			KASSERT(0);
    879 		}
    880 		vput(dp);
    881 		goto bad;
    882 	}
    883 
    884 	/*
    885 	 * Disallow directory write attempts on read-only lookups.
    886 	 * Prefers EEXIST over EROFS for the CREATE case.
    887 	 */
    888 	if (rdonly &&
    889 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
    890 		error = EROFS;
    891 		if (dp != ndp->ni_dvp) {
    892 			vput(dp);
    893 		}
    894 		goto bad;
    895 	}
    896 	if (ndp->ni_dvp != NULL) {
    897 		if (cnp->cn_flags & SAVESTART) {
    898 			ndp->ni_startdir = ndp->ni_dvp;
    899 			VREF(ndp->ni_startdir);
    900 		}
    901 	}
    902 	if ((cnp->cn_flags & LOCKLEAF) == 0) {
    903 		VOP_UNLOCK(dp, 0);
    904 	}
    905 	return (0);
    906 
    907 bad:
    908 	ndp->ni_vp = NULL;
    909 	return (error);
    910 }
    911 
    912 /*
    913  * Reacquire a path name component.
    914  * dvp is locked on entry and exit.
    915  * *vpp is locked on exit unless it's NULL.
    916  */
    917 int
    918 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
    919 {
    920 	int rdonly;			/* lookup read-only flag bit */
    921 	int error = 0;
    922 #ifdef DEBUG
    923 	uint32_t newhash;		/* DEBUG: check name hash */
    924 	const char *cp;			/* DEBUG: check name ptr/len */
    925 #endif /* DEBUG */
    926 
    927 	/*
    928 	 * Setup: break out flag bits into variables.
    929 	 */
    930 	rdonly = cnp->cn_flags & RDONLY;
    931 	cnp->cn_flags &= ~ISSYMLINK;
    932 
    933 	/*
    934 	 * Search a new directory.
    935 	 *
    936 	 * The cn_hash value is for use by vfs_cache.
    937 	 * The last component of the filename is left accessible via
    938 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    939 	 * the name set the SAVENAME flag. When done, they assume
    940 	 * responsibility for freeing the pathname buffer.
    941 	 */
    942 #ifdef DEBUG
    943 	cp = NULL;
    944 	newhash = namei_hash(cnp->cn_nameptr, &cp);
    945 	if ((uint32_t)newhash != (uint32_t)cnp->cn_hash)
    946 		panic("relookup: bad hash");
    947 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
    948 		panic("relookup: bad len");
    949 	while (*cp == '/')
    950 		cp++;
    951 	if (*cp != 0)
    952 		panic("relookup: not last component");
    953 #endif /* DEBUG */
    954 
    955 	/*
    956 	 * Check for degenerate name (e.g. / or "")
    957 	 * which is a way of talking about a directory,
    958 	 * e.g. like "/." or ".".
    959 	 */
    960 	if (cnp->cn_nameptr[0] == '\0')
    961 		panic("relookup: null name");
    962 
    963 	if (cnp->cn_flags & ISDOTDOT)
    964 		panic("relookup: lookup on dot-dot");
    965 
    966 	/*
    967 	 * We now have a segment name to search for, and a directory to search.
    968 	 */
    969 	if ((error = VOP_LOOKUP(dvp, vpp, cnp)) != 0) {
    970 #ifdef DIAGNOSTIC
    971 		if (*vpp != NULL)
    972 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    973 #endif
    974 		if (error != EJUSTRETURN)
    975 			goto bad;
    976 	}
    977 
    978 #ifdef DIAGNOSTIC
    979 	/*
    980 	 * Check for symbolic link
    981 	 */
    982 	if (*vpp && (*vpp)->v_type == VLNK && (cnp->cn_flags & FOLLOW))
    983 		panic("relookup: symlink found");
    984 #endif
    985 
    986 	/*
    987 	 * Check for read-only lookups.
    988 	 */
    989 	if (rdonly && cnp->cn_nameiop != LOOKUP) {
    990 		error = EROFS;
    991 		if (*vpp) {
    992 			vput(*vpp);
    993 		}
    994 		goto bad;
    995 	}
    996 	if (cnp->cn_flags & SAVESTART)
    997 		VREF(dvp);
    998 	return (0);
    999 
   1000 bad:
   1001 	*vpp = NULL;
   1002 	return (error);
   1003 }
   1004