Home | History | Annotate | Line # | Download | only in kern
vfs_lookup.c revision 1.131
      1 /*	$NetBSD: vfs_lookup.c,v 1.131 2011/01/04 07:43:42 dholland 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 1.131 2011/01/04 07:43:42 dholland Exp $");
     41 
     42 #include "opt_magiclinks.h"
     43 
     44 #include <sys/param.h>
     45 #include <sys/systm.h>
     46 #include <sys/kernel.h>
     47 #include <sys/syslimits.h>
     48 #include <sys/time.h>
     49 #include <sys/namei.h>
     50 #include <sys/vnode.h>
     51 #include <sys/mount.h>
     52 #include <sys/errno.h>
     53 #include <sys/filedesc.h>
     54 #include <sys/hash.h>
     55 #include <sys/proc.h>
     56 #include <sys/syslog.h>
     57 #include <sys/kauth.h>
     58 #include <sys/ktrace.h>
     59 
     60 #ifndef MAGICLINKS
     61 #define MAGICLINKS 0
     62 #endif
     63 
     64 int vfs_magiclinks = MAGICLINKS;
     65 
     66 pool_cache_t pnbuf_cache;	/* pathname buffer cache */
     67 
     68 /*
     69  * Substitute replacement text for 'magic' strings in symlinks.
     70  * Returns 0 if successful, and returns non-zero if an error
     71  * occurs.  (Currently, the only possible error is running out
     72  * of temporary pathname space.)
     73  *
     74  * Looks for "@<string>" and "@<string>/", where <string> is a
     75  * recognized 'magic' string.  Replaces the "@<string>" with the
     76  * appropriate replacement text.  (Note that in some cases the
     77  * replacement text may have zero length.)
     78  *
     79  * This would have been table driven, but the variance in
     80  * replacement strings (and replacement string lengths) made
     81  * that impractical.
     82  */
     83 #define	VNL(x)							\
     84 	(sizeof(x) - 1)
     85 
     86 #define	VO	'{'
     87 #define	VC	'}'
     88 
     89 #define	MATCH(str)						\
     90 	((termchar == '/' && i + VNL(str) == *len) ||		\
     91 	 (i + VNL(str) < *len &&				\
     92 	  cp[i + VNL(str)] == termchar)) &&			\
     93 	!strncmp((str), &cp[i], VNL(str))
     94 
     95 #define	SUBSTITUTE(m, s, sl)					\
     96 	if ((newlen + (sl)) >= MAXPATHLEN)			\
     97 		return 1;					\
     98 	i += VNL(m);						\
     99 	if (termchar != '/')					\
    100 		i++;						\
    101 	(void)memcpy(&tmp[newlen], (s), (sl));			\
    102 	newlen += (sl);						\
    103 	change = 1;						\
    104 	termchar = '/';
    105 
    106 static int
    107 symlink_magic(struct proc *p, char *cp, size_t *len)
    108 {
    109 	char *tmp;
    110 	size_t change, i, newlen, slen;
    111 	char termchar = '/';
    112 	char idtmp[11]; /* enough for 32 bit *unsigned* integer */
    113 
    114 
    115 	tmp = PNBUF_GET();
    116 	for (change = i = newlen = 0; i < *len; ) {
    117 		if (cp[i] != '@') {
    118 			tmp[newlen++] = cp[i++];
    119 			continue;
    120 		}
    121 
    122 		i++;
    123 
    124 		/* Check for @{var} syntax. */
    125 		if (cp[i] == VO) {
    126 			termchar = VC;
    127 			i++;
    128 		}
    129 
    130 		/*
    131 		 * The following checks should be ordered according
    132 		 * to frequency of use.
    133 		 */
    134 		if (MATCH("machine_arch")) {
    135 			slen = VNL(MACHINE_ARCH);
    136 			SUBSTITUTE("machine_arch", MACHINE_ARCH, slen);
    137 		} else if (MATCH("machine")) {
    138 			slen = VNL(MACHINE);
    139 			SUBSTITUTE("machine", MACHINE, slen);
    140 		} else if (MATCH("hostname")) {
    141 			SUBSTITUTE("hostname", hostname, hostnamelen);
    142 		} else if (MATCH("osrelease")) {
    143 			slen = strlen(osrelease);
    144 			SUBSTITUTE("osrelease", osrelease, slen);
    145 		} else if (MATCH("emul")) {
    146 			slen = strlen(p->p_emul->e_name);
    147 			SUBSTITUTE("emul", p->p_emul->e_name, slen);
    148 		} else if (MATCH("kernel_ident")) {
    149 			slen = strlen(kernel_ident);
    150 			SUBSTITUTE("kernel_ident", kernel_ident, slen);
    151 		} else if (MATCH("domainname")) {
    152 			SUBSTITUTE("domainname", domainname, domainnamelen);
    153 		} else if (MATCH("ostype")) {
    154 			slen = strlen(ostype);
    155 			SUBSTITUTE("ostype", ostype, slen);
    156 		} else if (MATCH("uid")) {
    157 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
    158 			    kauth_cred_geteuid(kauth_cred_get()));
    159 			SUBSTITUTE("uid", idtmp, slen);
    160 		} else if (MATCH("ruid")) {
    161 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
    162 			    kauth_cred_getuid(kauth_cred_get()));
    163 			SUBSTITUTE("ruid", idtmp, slen);
    164 		} else if (MATCH("gid")) {
    165 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
    166 			    kauth_cred_getegid(kauth_cred_get()));
    167 			SUBSTITUTE("gid", idtmp, slen);
    168 		} else if (MATCH("rgid")) {
    169 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
    170 			    kauth_cred_getgid(kauth_cred_get()));
    171 			SUBSTITUTE("rgid", idtmp, slen);
    172 		} else {
    173 			tmp[newlen++] = '@';
    174 			if (termchar == VC)
    175 				tmp[newlen++] = VO;
    176 		}
    177 	}
    178 
    179 	if (change) {
    180 		(void)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 
    196 /*
    197  * Determine the namei hash (for cn_hash) for name.
    198  * If *ep != NULL, hash from name to ep-1.
    199  * If *ep == NULL, hash from name until the first NUL or '/', and
    200  * return the location of this termination character in *ep.
    201  *
    202  * This function returns an equivalent hash to the MI hash32_strn().
    203  * The latter isn't used because in the *ep == NULL case, determining
    204  * the length of the string to the first NUL or `/' and then calling
    205  * hash32_strn() involves unnecessary double-handling of the data.
    206  */
    207 uint32_t
    208 namei_hash(const char *name, const char **ep)
    209 {
    210 	uint32_t	hash;
    211 
    212 	hash = HASH32_STR_INIT;
    213 	if (*ep != NULL) {
    214 		for (; name < *ep; name++)
    215 			hash = hash * 33 + *(const uint8_t *)name;
    216 	} else {
    217 		for (; *name != '\0' && *name != '/'; name++)
    218 			hash = hash * 33 + *(const uint8_t *)name;
    219 		*ep = name;
    220 	}
    221 	return (hash + (hash >> 5));
    222 }
    223 
    224 ////////////////////////////////////////////////////////////
    225 
    226 /*
    227  * Sealed abstraction for pathnames.
    228  *
    229  * System-call-layer level code that is going to call namei should
    230  * first create a pathbuf and adjust all the bells and whistles on it
    231  * as needed by context
    232  */
    233 
    234 struct pathbuf {
    235 	char *pb_path;
    236 	char *pb_pathcopy;
    237 	unsigned pb_pathcopyuses;
    238 };
    239 
    240 static struct pathbuf *
    241 pathbuf_create_raw(void)
    242 {
    243 	struct pathbuf *pb;
    244 
    245 	pb = kmem_alloc(sizeof(*pb), KM_SLEEP);
    246 	if (pb == NULL) {
    247 		return NULL;
    248 	}
    249 	pb->pb_path = PNBUF_GET();
    250 	if (pb->pb_path == NULL) {
    251 		kmem_free(pb, sizeof(*pb));
    252 		return NULL;
    253 	}
    254 	pb->pb_pathcopy = NULL;
    255 	pb->pb_pathcopyuses = 0;
    256 	return pb;
    257 }
    258 
    259 void
    260 pathbuf_destroy(struct pathbuf *pb)
    261 {
    262 	KASSERT(pb->pb_pathcopyuses == 0);
    263 	KASSERT(pb->pb_pathcopy == NULL);
    264 	PNBUF_PUT(pb->pb_path);
    265 	kmem_free(pb, sizeof(*pb));
    266 }
    267 
    268 struct pathbuf *
    269 pathbuf_assimilate(char *pnbuf)
    270 {
    271 	struct pathbuf *pb;
    272 
    273 	pb = kmem_alloc(sizeof(*pb), KM_SLEEP);
    274 	if (pb == NULL) {
    275 		return NULL;
    276 	}
    277 	pb->pb_path = pnbuf;
    278 	pb->pb_pathcopy = NULL;
    279 	pb->pb_pathcopyuses = 0;
    280 	return pb;
    281 }
    282 
    283 struct pathbuf *
    284 pathbuf_create(const char *path)
    285 {
    286 	struct pathbuf *pb;
    287 	int error;
    288 
    289 	pb = pathbuf_create_raw();
    290 	if (pb == NULL) {
    291 		return NULL;
    292 	}
    293 	error = copystr(path, pb->pb_path, PATH_MAX, NULL);
    294 	if (error != 0) {
    295 		KASSERT(!"kernel path too long in pathbuf_create");
    296 		/* make sure it's null-terminated, just in case */
    297 		pb->pb_path[PATH_MAX-1] = '\0';
    298 	}
    299 	return pb;
    300 }
    301 
    302 int
    303 pathbuf_copyin(const char *userpath, struct pathbuf **ret)
    304 {
    305 	struct pathbuf *pb;
    306 	int error;
    307 
    308 	pb = pathbuf_create_raw();
    309 	if (pb == NULL) {
    310 		return ENOMEM;
    311 	}
    312 	error = copyinstr(userpath, pb->pb_path, PATH_MAX, NULL);
    313 	if (error) {
    314 		pathbuf_destroy(pb);
    315 		return error;
    316 	}
    317 	*ret = pb;
    318 	return 0;
    319 }
    320 
    321 /*
    322  * XXX should not exist
    323  */
    324 int
    325 pathbuf_maybe_copyin(const char *path, enum uio_seg seg, struct pathbuf **ret)
    326 {
    327 	if (seg == UIO_USERSPACE) {
    328 		return pathbuf_copyin(path, ret);
    329 	} else {
    330 		*ret = pathbuf_create(path);
    331 		if (*ret == NULL) {
    332 			return ENOMEM;
    333 		}
    334 		return 0;
    335 	}
    336 }
    337 
    338 /*
    339  * Get a copy of the path buffer as it currently exists. If this is
    340  * called after namei starts the results may be arbitrary.
    341  */
    342 void
    343 pathbuf_copystring(const struct pathbuf *pb, char *buf, size_t maxlen)
    344 {
    345 	strlcpy(buf, pb->pb_path, maxlen);
    346 }
    347 
    348 /*
    349  * These two functions allow access to a saved copy of the original
    350  * path string. The first copy should be gotten before namei is
    351  * called. Each copy that is gotten should be put back.
    352  */
    353 
    354 const char *
    355 pathbuf_stringcopy_get(struct pathbuf *pb)
    356 {
    357 	if (pb->pb_pathcopyuses == 0) {
    358 		pb->pb_pathcopy = PNBUF_GET();
    359 		strcpy(pb->pb_pathcopy, pb->pb_path);
    360 	}
    361 	pb->pb_pathcopyuses++;
    362 	return pb->pb_pathcopy;
    363 }
    364 
    365 void
    366 pathbuf_stringcopy_put(struct pathbuf *pb, const char *str)
    367 {
    368 	KASSERT(str == pb->pb_pathcopy);
    369 	KASSERT(pb->pb_pathcopyuses > 0);
    370 	pb->pb_pathcopyuses--;
    371 	if (pb->pb_pathcopyuses == 0) {
    372 		PNBUF_PUT(pb->pb_pathcopy);
    373 		pb->pb_pathcopy = NULL;
    374 	}
    375 }
    376 
    377 
    378 ////////////////////////////////////////////////////////////
    379 
    380 /*
    381  * Convert a pathname into a pointer to a locked vnode.
    382  *
    383  * The FOLLOW flag is set when symbolic links are to be followed
    384  * when they occur at the end of the name translation process.
    385  * Symbolic links are always followed for all other pathname
    386  * components other than the last.
    387  *
    388  * The segflg defines whether the name is to be copied from user
    389  * space or kernel space.
    390  *
    391  * Overall outline of namei:
    392  *
    393  *	copy in name
    394  *	get starting directory
    395  *	while (!done && !error) {
    396  *		call lookup to search path.
    397  *		if symbolic link, massage name in buffer and continue
    398  *	}
    399  */
    400 
    401 /*
    402  * Internal state for a namei operation.
    403  */
    404 struct namei_state {
    405 	struct nameidata *ndp;
    406 	struct componentname *cnp;
    407 
    408 	/* used by the pieces of namei */
    409 	struct vnode *namei_startdir; /* The directory namei() starts from. */
    410 
    411 	/* used by the pieces of lookup */
    412 	int lookup_alldone;
    413 
    414 	int docache;			/* == 0 do not cache last component */
    415 	int rdonly;			/* lookup read-only flag bit */
    416 	struct vnode *dp;		/* the directory we are searching */
    417 	int slashes;
    418 };
    419 
    420 /* XXX reorder things to make this decl unnecessary */
    421 static int do_lookup(struct namei_state *state, struct vnode *startdir);
    422 
    423 
    424 /*
    425  * Initialize the namei working state.
    426  */
    427 static void
    428 namei_init(struct namei_state *state, struct nameidata *ndp)
    429 {
    430 	state->ndp = ndp;
    431 	state->cnp = &ndp->ni_cnd;
    432 	KASSERT((state->cnp->cn_flags & INRELOOKUP) == 0);
    433 
    434 	state->namei_startdir = NULL;
    435 
    436 	state->lookup_alldone = 0;
    437 
    438 	state->docache = 0;
    439 	state->rdonly = 0;
    440 	state->dp = NULL;
    441 	state->slashes = 0;
    442 }
    443 
    444 /*
    445  * Clean up the working namei state, leaving things ready for return
    446  * from namei.
    447  */
    448 static void
    449 namei_cleanup(struct namei_state *state)
    450 {
    451 	KASSERT(state->cnp == &state->ndp->ni_cnd);
    452 
    453 	//KASSERT(state->namei_startdir == NULL); 	// not yet
    454 
    455 	/* nothing for now */
    456 	(void)state;
    457 }
    458 
    459 //////////////////////////////
    460 
    461 /*
    462  * Start up namei. Early portion.
    463  *
    464  * This is divided from namei_start2 by the emul_retry: point.
    465  */
    466 static void
    467 namei_start1(struct namei_state *state)
    468 {
    469 
    470 #ifdef DIAGNOSTIC
    471 	if (!state->cnp->cn_cred)
    472 		panic("namei: bad cred/proc");
    473 	if (state->cnp->cn_nameiop & (~OPMASK))
    474 		panic("namei: nameiop contaminated with flags");
    475 	if (state->cnp->cn_flags & OPMASK)
    476 		panic("namei: flags contaminated with nameiops");
    477 #endif
    478 
    479 	/*
    480 	 * The buffer for name translation shall be the one inside the
    481 	 * pathbuf.
    482 	 */
    483 	state->ndp->ni_pnbuf = state->ndp->ni_pathbuf->pb_path;
    484 }
    485 
    486 /*
    487  * Start up namei. Copy the path, find the root dir and cwd, establish
    488  * the starting directory for lookup, and lock it.
    489  */
    490 static int
    491 namei_start2(struct namei_state *state)
    492 {
    493 	struct nameidata *ndp = state->ndp;
    494 	struct componentname *cnp = state->cnp;
    495 
    496 	struct cwdinfo *cwdi;		/* pointer to cwd state */
    497 	struct lwp *self = curlwp;	/* thread doing namei() */
    498 
    499 	/* length includes null terminator (was originally from copyinstr) */
    500 	ndp->ni_pathlen = strlen(ndp->ni_pnbuf) + 1;
    501 
    502 	/*
    503 	 * POSIX.1 requirement: "" is not a valid file name.
    504 	 */
    505 	if (ndp->ni_pathlen == 1) {
    506 		ndp->ni_vp = NULL;
    507 		return ENOENT;
    508 	}
    509 
    510 	ndp->ni_loopcnt = 0;
    511 
    512 	/*
    513 	 * Get root directory for the translation.
    514 	 */
    515 	cwdi = self->l_proc->p_cwdi;
    516 	rw_enter(&cwdi->cwdi_lock, RW_READER);
    517 	state->namei_startdir = cwdi->cwdi_rdir;
    518 	if (state->namei_startdir == NULL)
    519 		state->namei_startdir = rootvnode;
    520 	ndp->ni_rootdir = state->namei_startdir;
    521 
    522 	/*
    523 	 * Check if starting from root directory or current directory.
    524 	 */
    525 	if (ndp->ni_pnbuf[0] == '/') {
    526 		if (cnp->cn_flags & TRYEMULROOT) {
    527 			if (cnp->cn_flags & EMULROOTSET) {
    528 				/* Called from (eg) emul_find_interp() */
    529 				state->namei_startdir = ndp->ni_erootdir;
    530 			} else {
    531 				if (cwdi->cwdi_edir == NULL
    532 				    || (ndp->ni_pnbuf[1] == '.'
    533 					   && ndp->ni_pnbuf[2] == '.'
    534 					   && ndp->ni_pnbuf[3] == '/')) {
    535 					ndp->ni_erootdir = NULL;
    536 				} else {
    537 					state->namei_startdir = cwdi->cwdi_edir;
    538 					ndp->ni_erootdir = state->namei_startdir;
    539 				}
    540 			}
    541 		} else {
    542 			ndp->ni_erootdir = NULL;
    543 			if (cnp->cn_flags & NOCHROOT)
    544 				state->namei_startdir = ndp->ni_rootdir = rootvnode;
    545 		}
    546 	} else {
    547 		state->namei_startdir = cwdi->cwdi_cdir;
    548 		ndp->ni_erootdir = NULL;
    549 	}
    550 	vref(state->namei_startdir);
    551 	rw_exit(&cwdi->cwdi_lock);
    552 
    553 	/*
    554 	 * Ktrace it.
    555 	 */
    556 	if (ktrpoint(KTR_NAMEI)) {
    557 		if (ndp->ni_erootdir != NULL) {
    558 			/*
    559 			 * To make any sense, the trace entry need to have the
    560 			 * text of the emulation path prepended.
    561 			 * Usually we can get this from the current process,
    562 			 * but when called from emul_find_interp() it is only
    563 			 * in the exec_package - so we get it passed in ni_next
    564 			 * (this is a hack).
    565 			 */
    566 			const char *emul_path;
    567 			if (cnp->cn_flags & EMULROOTSET)
    568 				emul_path = ndp->ni_next;
    569 			else
    570 				emul_path = self->l_proc->p_emul->e_path;
    571 			ktrnamei2(emul_path, strlen(emul_path),
    572 			    ndp->ni_pnbuf, ndp->ni_pathlen);
    573 		} else
    574 			ktrnamei(ndp->ni_pnbuf, ndp->ni_pathlen);
    575 	}
    576 
    577 	vn_lock(state->namei_startdir, LK_EXCLUSIVE | LK_RETRY);
    578 
    579 	return 0;
    580 }
    581 
    582 /*
    583  * Undo namei_start: unlock and release the current lookup directory,
    584  * and discard the path buffer.
    585  */
    586 static void
    587 namei_end(struct namei_state *state)
    588 {
    589 	vput(state->namei_startdir);
    590 }
    591 
    592 /*
    593  * Check for being at a symlink.
    594  */
    595 static inline int
    596 namei_atsymlink(struct namei_state *state)
    597 {
    598 	return (state->cnp->cn_flags & ISSYMLINK) != 0;
    599 }
    600 
    601 /*
    602  * Follow a symlink.
    603  */
    604 static inline int
    605 namei_follow(struct namei_state *state)
    606 {
    607 	struct nameidata *ndp = state->ndp;
    608 	struct componentname *cnp = state->cnp;
    609 
    610 	struct lwp *self = curlwp;	/* thread doing namei() */
    611 	struct iovec aiov;		/* uio for reading symbolic links */
    612 	struct uio auio;
    613 	char *cp;			/* pointer into pathname argument */
    614 	size_t linklen;
    615 	int error;
    616 
    617 	if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
    618 		return ELOOP;
    619 	}
    620 	if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
    621 		error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred);
    622 		if (error != 0)
    623 			return error;
    624 	}
    625 
    626 	/* FUTURE: fix this to not use a second buffer */
    627 	cp = PNBUF_GET();
    628 	aiov.iov_base = cp;
    629 	aiov.iov_len = MAXPATHLEN;
    630 	auio.uio_iov = &aiov;
    631 	auio.uio_iovcnt = 1;
    632 	auio.uio_offset = 0;
    633 	auio.uio_rw = UIO_READ;
    634 	auio.uio_resid = MAXPATHLEN;
    635 	UIO_SETUP_SYSSPACE(&auio);
    636 	error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
    637 	if (error) {
    638 		PNBUF_PUT(cp);
    639 		return error;
    640 	}
    641 	linklen = MAXPATHLEN - auio.uio_resid;
    642 	if (linklen == 0) {
    643 		PNBUF_PUT(cp);
    644 		return ENOENT;
    645 	}
    646 
    647 	/*
    648 	 * Do symlink substitution, if appropriate, and
    649 	 * check length for potential overflow.
    650 	 */
    651 	if ((vfs_magiclinks &&
    652 	     symlink_magic(self->l_proc, cp, &linklen)) ||
    653 	    (linklen + ndp->ni_pathlen >= MAXPATHLEN)) {
    654 		PNBUF_PUT(cp);
    655 		return ENAMETOOLONG;
    656 	}
    657 	if (ndp->ni_pathlen > 1) {
    658 		/* includes a null-terminator */
    659 		memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
    660 	} else {
    661 		cp[linklen] = '\0';
    662 	}
    663 	ndp->ni_pathlen += linklen;
    664 	memcpy(ndp->ni_pnbuf, cp, ndp->ni_pathlen);
    665 	PNBUF_PUT(cp);
    666 	vput(ndp->ni_vp);
    667 	state->namei_startdir = ndp->ni_dvp;
    668 
    669 	/*
    670 	 * Check if root directory should replace current directory.
    671 	 */
    672 	if (ndp->ni_pnbuf[0] == '/') {
    673 		vput(state->namei_startdir);
    674 		/* Keep absolute symbolic links inside emulation root */
    675 		state->namei_startdir = ndp->ni_erootdir;
    676 		if (state->namei_startdir == NULL ||
    677 		    (ndp->ni_pnbuf[1] == '.'
    678 		     && ndp->ni_pnbuf[2] == '.'
    679 		     && ndp->ni_pnbuf[3] == '/')) {
    680 			ndp->ni_erootdir = NULL;
    681 			state->namei_startdir = ndp->ni_rootdir;
    682 		}
    683 		vref(state->namei_startdir);
    684 		vn_lock(state->namei_startdir, LK_EXCLUSIVE | LK_RETRY);
    685 	}
    686 
    687 	return 0;
    688 }
    689 
    690 //////////////////////////////
    691 
    692 /*
    693  * Search a pathname.
    694  * This is a very central and rather complicated routine.
    695  *
    696  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
    697  * The starting directory is passed in. The pathname is descended
    698  * until done, or a symbolic link is encountered. The variable ni_more
    699  * is clear if the path is completed; it is set to one if a symbolic
    700  * link needing interpretation is encountered.
    701  *
    702  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
    703  * whether the name is to be looked up, created, renamed, or deleted.
    704  * When CREATE, RENAME, or DELETE is specified, information usable in
    705  * creating, renaming, or deleting a directory entry may be calculated.
    706  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
    707  * locked.  Otherwise the parent directory is not returned. If the target
    708  * of the pathname exists and LOCKLEAF is or'ed into the flag the target
    709  * is returned locked, otherwise it is returned unlocked.  When creating
    710  * or renaming and LOCKPARENT is specified, the target may not be ".".
    711  * When deleting and LOCKPARENT is specified, the target may be ".".
    712  *
    713  * Overall outline of lookup:
    714  *
    715  * dirloop:
    716  *	identify next component of name at ndp->ni_ptr
    717  *	handle degenerate case where name is null string
    718  *	if .. and crossing mount points and on mounted filesys, find parent
    719  *	call VOP_LOOKUP routine for next component name
    720  *	    directory vnode returned in ni_dvp, locked.
    721  *	    component vnode returned in ni_vp (if it exists), locked.
    722  *	if result vnode is mounted on and crossing mount points,
    723  *	    find mounted on vnode
    724  *	if more components of name, do next level at dirloop
    725  *	return the answer in ni_vp, locked if LOCKLEAF set
    726  *	    if LOCKPARENT set, return locked parent in ni_dvp
    727  */
    728 
    729 /*
    730  * Begin lookup().
    731  */
    732 static int
    733 lookup_start(struct namei_state *state, struct vnode *startdir)
    734 {
    735 	const char *cp;			/* pointer into pathname argument */
    736 
    737 	struct componentname *cnp = state->cnp;
    738 	struct nameidata *ndp = state->ndp;
    739 
    740 	KASSERT(cnp == &ndp->ni_cnd);
    741 
    742 	state->lookup_alldone = 0;
    743 	state->dp = NULL;
    744 
    745 	/*
    746 	 * Setup: break out flag bits into variables.
    747 	 */
    748 	state->docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
    749 	if (cnp->cn_nameiop == DELETE)
    750 		state->docache = 0;
    751 	state->rdonly = cnp->cn_flags & RDONLY;
    752 	ndp->ni_dvp = NULL;
    753 	cnp->cn_flags &= ~ISSYMLINK;
    754 	state->dp = startdir;
    755 
    756 	/*
    757 	 * If we have a leading string of slashes, remove them, and just make
    758 	 * sure the current node is a directory.
    759 	 */
    760 	cp = cnp->cn_nameptr;
    761 	if (*cp == '/') {
    762 		do {
    763 			cp++;
    764 		} while (*cp == '/');
    765 		ndp->ni_pathlen -= cp - cnp->cn_nameptr;
    766 		cnp->cn_nameptr = cp;
    767 
    768 		if (state->dp->v_type != VDIR) {
    769 			vput(state->dp);
    770 			return ENOTDIR;
    771 		}
    772 
    773 		/*
    774 		 * If we've exhausted the path name, then just return the
    775 		 * current node.
    776 		 */
    777 		if (cnp->cn_nameptr[0] == '\0') {
    778 			ndp->ni_vp = state->dp;
    779 			cnp->cn_flags |= ISLASTCN;
    780 
    781 			/* bleh */
    782 			state->lookup_alldone = 1;
    783 			return 0;
    784 		}
    785 	}
    786 
    787 	return 0;
    788 }
    789 
    790 static int
    791 lookup_parsepath(struct namei_state *state)
    792 {
    793 	const char *cp;			/* pointer into pathname argument */
    794 
    795 	struct componentname *cnp = state->cnp;
    796 	struct nameidata *ndp = state->ndp;
    797 
    798 	KASSERT(cnp == &ndp->ni_cnd);
    799 
    800 	/*
    801 	 * Search a new directory.
    802 	 *
    803 	 * The cn_hash value is for use by vfs_cache.
    804 	 * The last component of the filename is left accessible via
    805 	 * cnp->cn_nameptr for callers that need the name. Callers needing
    806 	 * the name set the SAVENAME flag. When done, they assume
    807 	 * responsibility for freeing the pathname buffer.
    808 	 *
    809 	 * At this point, our only vnode state is that "dp" is held and locked.
    810 	 */
    811 	cnp->cn_consume = 0;
    812 	cp = NULL;
    813 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
    814 	cnp->cn_namelen = cp - cnp->cn_nameptr;
    815 	if (cnp->cn_namelen > NAME_MAX) {
    816 		vput(state->dp);
    817 		ndp->ni_dvp = NULL;
    818 		return ENAMETOOLONG;
    819 	}
    820 #ifdef NAMEI_DIAGNOSTIC
    821 	{ char c = *cp;
    822 	*(char *)cp = '\0';
    823 	printf("{%s}: ", cnp->cn_nameptr);
    824 	*(char *)cp = c; }
    825 #endif /* NAMEI_DIAGNOSTIC */
    826 	ndp->ni_pathlen -= cnp->cn_namelen;
    827 	ndp->ni_next = cp;
    828 	/*
    829 	 * If this component is followed by a slash, then move the pointer to
    830 	 * the next component forward, and remember that this component must be
    831 	 * a directory.
    832 	 */
    833 	if (*cp == '/') {
    834 		do {
    835 			cp++;
    836 		} while (*cp == '/');
    837 		state->slashes = cp - ndp->ni_next;
    838 		ndp->ni_pathlen -= state->slashes;
    839 		ndp->ni_next = cp;
    840 		cnp->cn_flags |= REQUIREDIR;
    841 	} else {
    842 		state->slashes = 0;
    843 		cnp->cn_flags &= ~REQUIREDIR;
    844 	}
    845 	/*
    846 	 * We do special processing on the last component, whether or not it's
    847 	 * a directory.  Cache all intervening lookups, but not the final one.
    848 	 */
    849 	if (*cp == '\0') {
    850 		if (state->docache)
    851 			cnp->cn_flags |= MAKEENTRY;
    852 		else
    853 			cnp->cn_flags &= ~MAKEENTRY;
    854 		cnp->cn_flags |= ISLASTCN;
    855 	} else {
    856 		cnp->cn_flags |= MAKEENTRY;
    857 		cnp->cn_flags &= ~ISLASTCN;
    858 	}
    859 	if (cnp->cn_namelen == 2 &&
    860 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
    861 		cnp->cn_flags |= ISDOTDOT;
    862 	else
    863 		cnp->cn_flags &= ~ISDOTDOT;
    864 
    865 	return 0;
    866 }
    867 
    868 static int
    869 lookup_once(struct namei_state *state)
    870 {
    871 	struct vnode *tdp;		/* saved dp */
    872 	struct mount *mp;		/* mount table entry */
    873 	struct lwp *l = curlwp;
    874 	int error;
    875 
    876 	struct componentname *cnp = state->cnp;
    877 	struct nameidata *ndp = state->ndp;
    878 
    879 	KASSERT(cnp == &ndp->ni_cnd);
    880 
    881 	/*
    882 	 * Handle "..": two special cases.
    883 	 * 1. If at root directory (e.g. after chroot)
    884 	 *    or at absolute root directory
    885 	 *    then ignore it so can't get out.
    886 	 * 1a. If at the root of the emulation filesystem go to the real
    887 	 *    root. So "/../<path>" is always absolute.
    888 	 * 1b. If we have somehow gotten out of a jail, warn
    889 	 *    and also ignore it so we can't get farther out.
    890 	 * 2. If this vnode is the root of a mounted
    891 	 *    filesystem, then replace it with the
    892 	 *    vnode which was mounted on so we take the
    893 	 *    .. in the other file system.
    894 	 */
    895 	if (cnp->cn_flags & ISDOTDOT) {
    896 		struct proc *p = l->l_proc;
    897 
    898 		for (;;) {
    899 			if (state->dp == ndp->ni_rootdir || state->dp == rootvnode) {
    900 				ndp->ni_dvp = state->dp;
    901 				ndp->ni_vp = state->dp;
    902 				vref(state->dp);
    903 				return 0;
    904 			}
    905 			if (ndp->ni_rootdir != rootvnode) {
    906 				int retval;
    907 
    908 				VOP_UNLOCK(state->dp);
    909 				retval = vn_isunder(state->dp, ndp->ni_rootdir, l);
    910 				vn_lock(state->dp, LK_EXCLUSIVE | LK_RETRY);
    911 				if (!retval) {
    912 				    /* Oops! We got out of jail! */
    913 				    log(LOG_WARNING,
    914 					"chrooted pid %d uid %d (%s) "
    915 					"detected outside of its chroot\n",
    916 					p->p_pid, kauth_cred_geteuid(l->l_cred),
    917 					p->p_comm);
    918 				    /* Put us at the jail root. */
    919 				    vput(state->dp);
    920 				    state->dp = ndp->ni_rootdir;
    921 				    ndp->ni_dvp = state->dp;
    922 				    ndp->ni_vp = state->dp;
    923 				    vref(state->dp);
    924 				    vref(state->dp);
    925 				    vn_lock(state->dp, LK_EXCLUSIVE | LK_RETRY);
    926 				    return 0;
    927 				}
    928 			}
    929 			if ((state->dp->v_vflag & VV_ROOT) == 0 ||
    930 			    (cnp->cn_flags & NOCROSSMOUNT))
    931 				break;
    932 			tdp = state->dp;
    933 			state->dp = state->dp->v_mount->mnt_vnodecovered;
    934 			vput(tdp);
    935 			vref(state->dp);
    936 			vn_lock(state->dp, LK_EXCLUSIVE | LK_RETRY);
    937 		}
    938 	}
    939 
    940 	/*
    941 	 * We now have a segment name to search for, and a directory to search.
    942 	 * Again, our only vnode state is that "dp" is held and locked.
    943 	 */
    944 unionlookup:
    945 	ndp->ni_dvp = state->dp;
    946 	ndp->ni_vp = NULL;
    947 	error = VOP_LOOKUP(state->dp, &ndp->ni_vp, cnp);
    948 	if (error != 0) {
    949 #ifdef DIAGNOSTIC
    950 		if (ndp->ni_vp != NULL)
    951 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
    952 #endif /* DIAGNOSTIC */
    953 #ifdef NAMEI_DIAGNOSTIC
    954 		printf("not found\n");
    955 #endif /* NAMEI_DIAGNOSTIC */
    956 		if ((error == ENOENT) &&
    957 		    (state->dp->v_vflag & VV_ROOT) &&
    958 		    (state->dp->v_mount->mnt_flag & MNT_UNION)) {
    959 			tdp = state->dp;
    960 			state->dp = state->dp->v_mount->mnt_vnodecovered;
    961 			vput(tdp);
    962 			vref(state->dp);
    963 			vn_lock(state->dp, LK_EXCLUSIVE | LK_RETRY);
    964 			goto unionlookup;
    965 		}
    966 
    967 		if (error != EJUSTRETURN)
    968 			return error;
    969 
    970 		/*
    971 		 * If this was not the last component, or there were trailing
    972 		 * slashes, and we are not going to create a directory,
    973 		 * then the name must exist.
    974 		 */
    975 		if ((cnp->cn_flags & (REQUIREDIR | CREATEDIR)) == REQUIREDIR) {
    976 			return ENOENT;
    977 		}
    978 
    979 		/*
    980 		 * If creating and at end of pathname, then can consider
    981 		 * allowing file to be created.
    982 		 */
    983 		if (state->rdonly) {
    984 			return EROFS;
    985 		}
    986 
    987 		/*
    988 		 * We return with ni_vp NULL to indicate that the entry
    989 		 * doesn't currently exist, leaving a pointer to the
    990 		 * (possibly locked) directory vnode in ndp->ni_dvp.
    991 		 */
    992 		state->lookup_alldone = 1;
    993 		return (0);
    994 	}
    995 #ifdef NAMEI_DIAGNOSTIC
    996 	printf("found\n");
    997 #endif /* NAMEI_DIAGNOSTIC */
    998 
    999 	/*
   1000 	 * Take into account any additional components consumed by the
   1001 	 * underlying filesystem.  This will include any trailing slashes after
   1002 	 * the last component consumed.
   1003 	 */
   1004 	if (cnp->cn_consume > 0) {
   1005 		ndp->ni_pathlen -= cnp->cn_consume - state->slashes;
   1006 		ndp->ni_next += cnp->cn_consume - state->slashes;
   1007 		cnp->cn_consume = 0;
   1008 		if (ndp->ni_next[0] == '\0')
   1009 			cnp->cn_flags |= ISLASTCN;
   1010 	}
   1011 
   1012 	state->dp = ndp->ni_vp;
   1013 
   1014 	/*
   1015 	 * "state->dp" and "ndp->ni_dvp" are both locked and held,
   1016 	 * and may be the same vnode.
   1017 	 */
   1018 
   1019 	/*
   1020 	 * Check to see if the vnode has been mounted on;
   1021 	 * if so find the root of the mounted file system.
   1022 	 */
   1023 	while (state->dp->v_type == VDIR && (mp = state->dp->v_mountedhere) &&
   1024 	       (cnp->cn_flags & NOCROSSMOUNT) == 0) {
   1025 		error = vfs_busy(mp, NULL);
   1026 		if (error != 0) {
   1027 			vput(state->dp);
   1028 			return error;
   1029 		}
   1030 		KASSERT(ndp->ni_dvp != state->dp);
   1031 		VOP_UNLOCK(ndp->ni_dvp);
   1032 		vput(state->dp);
   1033 		error = VFS_ROOT(mp, &tdp);
   1034 		vfs_unbusy(mp, false, NULL);
   1035 		if (error) {
   1036 			vn_lock(ndp->ni_dvp, LK_EXCLUSIVE | LK_RETRY);
   1037 			return error;
   1038 		}
   1039 		VOP_UNLOCK(tdp);
   1040 		ndp->ni_vp = state->dp = tdp;
   1041 		vn_lock(ndp->ni_dvp, LK_EXCLUSIVE | LK_RETRY);
   1042 		vn_lock(ndp->ni_vp, LK_EXCLUSIVE | LK_RETRY);
   1043 	}
   1044 
   1045 	return 0;
   1046 }
   1047 
   1048 static int
   1049 do_lookup(struct namei_state *state, struct vnode *startdir)
   1050 {
   1051 	int error = 0;
   1052 
   1053 	struct componentname *cnp = state->cnp;
   1054 	struct nameidata *ndp = state->ndp;
   1055 
   1056 	KASSERT(cnp == &ndp->ni_cnd);
   1057 
   1058 	error = lookup_start(state, startdir);
   1059 	if (error) {
   1060 		goto bad;
   1061 	}
   1062 	// XXX: this case should not be necessary given proper handling
   1063 	// of slashes elsewhere.
   1064 	if (state->lookup_alldone) {
   1065 		goto terminal;
   1066 	}
   1067 
   1068 dirloop:
   1069 	error = lookup_parsepath(state);
   1070 	if (error) {
   1071 		goto bad;
   1072 	}
   1073 
   1074 	error = lookup_once(state);
   1075 	if (error) {
   1076 		goto bad;
   1077 	}
   1078 	// XXX ought to be able to avoid this case too
   1079 	if (state->lookup_alldone) {
   1080 		/* this should NOT be "goto terminal;" */
   1081 		return 0;
   1082 	}
   1083 
   1084 	/*
   1085 	 * Check for symbolic link.  Back up over any slashes that we skipped,
   1086 	 * as we will need them again.
   1087 	 */
   1088 	if ((state->dp->v_type == VLNK) && (cnp->cn_flags & (FOLLOW|REQUIREDIR))) {
   1089 		ndp->ni_pathlen += state->slashes;
   1090 		ndp->ni_next -= state->slashes;
   1091 		cnp->cn_flags |= ISSYMLINK;
   1092 		return (0);
   1093 	}
   1094 
   1095 	/*
   1096 	 * Check for directory, if the component was followed by a series of
   1097 	 * slashes.
   1098 	 */
   1099 	if ((state->dp->v_type != VDIR) && (cnp->cn_flags & REQUIREDIR)) {
   1100 		error = ENOTDIR;
   1101 		KASSERT(state->dp != ndp->ni_dvp);
   1102 		vput(state->dp);
   1103 		goto bad;
   1104 	}
   1105 
   1106 	/*
   1107 	 * Not a symbolic link.  If this was not the last component, then
   1108 	 * continue at the next component, else return.
   1109 	 */
   1110 	if (!(cnp->cn_flags & ISLASTCN)) {
   1111 		cnp->cn_nameptr = ndp->ni_next;
   1112 		if (ndp->ni_dvp == state->dp) {
   1113 			vrele(ndp->ni_dvp);
   1114 		} else {
   1115 			vput(ndp->ni_dvp);
   1116 		}
   1117 		goto dirloop;
   1118 	}
   1119 
   1120 terminal:
   1121 	if (state->dp == ndp->ni_erootdir) {
   1122 		/*
   1123 		 * We are about to return the emulation root.
   1124 		 * This isn't a good idea because code might repeatedly
   1125 		 * lookup ".." until the file matches that returned
   1126 		 * for "/" and loop forever.
   1127 		 * So convert it to the real root.
   1128 		 */
   1129 		if (ndp->ni_dvp == state->dp)
   1130 			vrele(state->dp);
   1131 		else
   1132 			if (ndp->ni_dvp != NULL)
   1133 				vput(ndp->ni_dvp);
   1134 		ndp->ni_dvp = NULL;
   1135 		vput(state->dp);
   1136 		state->dp = ndp->ni_rootdir;
   1137 		vref(state->dp);
   1138 		vn_lock(state->dp, LK_EXCLUSIVE | LK_RETRY);
   1139 		ndp->ni_vp = state->dp;
   1140 	}
   1141 
   1142 	/*
   1143 	 * If the caller requested the parent node (i.e.
   1144 	 * it's a CREATE, DELETE, or RENAME), and we don't have one
   1145 	 * (because this is the root directory), then we must fail.
   1146 	 */
   1147 	if (ndp->ni_dvp == NULL && cnp->cn_nameiop != LOOKUP) {
   1148 		switch (cnp->cn_nameiop) {
   1149 		case CREATE:
   1150 			error = EEXIST;
   1151 			break;
   1152 		case DELETE:
   1153 		case RENAME:
   1154 			error = EBUSY;
   1155 			break;
   1156 		default:
   1157 			KASSERT(0);
   1158 		}
   1159 		vput(state->dp);
   1160 		goto bad;
   1161 	}
   1162 
   1163 	/*
   1164 	 * Disallow directory write attempts on read-only lookups.
   1165 	 * Prefers EEXIST over EROFS for the CREATE case.
   1166 	 */
   1167 	if (state->rdonly &&
   1168 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
   1169 		error = EROFS;
   1170 		if (state->dp != ndp->ni_dvp) {
   1171 			vput(state->dp);
   1172 		}
   1173 		goto bad;
   1174 	}
   1175 	if ((cnp->cn_flags & LOCKLEAF) == 0) {
   1176 		VOP_UNLOCK(state->dp);
   1177 	}
   1178 	return (0);
   1179 
   1180 bad:
   1181 	ndp->ni_vp = NULL;
   1182 	return (error);
   1183 }
   1184 
   1185 //////////////////////////////
   1186 
   1187 static int
   1188 do_namei(struct namei_state *state)
   1189 {
   1190 	int error;
   1191 
   1192 	struct nameidata *ndp = state->ndp;
   1193 	struct componentname *cnp = state->cnp;
   1194 	const char *savepath = NULL;
   1195 
   1196 	KASSERT(cnp == &ndp->ni_cnd);
   1197 
   1198 	namei_start1(state);
   1199 
   1200 	if (cnp->cn_flags & TRYEMULROOT) {
   1201 		savepath = pathbuf_stringcopy_get(ndp->ni_pathbuf);
   1202 	}
   1203 
   1204     emul_retry:
   1205 
   1206 	if (savepath != NULL) {
   1207 		/* kinda gross */
   1208 		strcpy(ndp->ni_pathbuf->pb_path, savepath);
   1209 		pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
   1210 		savepath = NULL;
   1211 	}
   1212 
   1213 	error = namei_start2(state);
   1214 	if (error) {
   1215 		if (savepath != NULL) {
   1216 			pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
   1217 		}
   1218 		return error;
   1219 	}
   1220 
   1221 	/* Loop through symbolic links */
   1222 	for (;;) {
   1223 		if (state->namei_startdir->v_mount == NULL) {
   1224 			/* Give up if the directory is no longer mounted */
   1225 			if (savepath != NULL) {
   1226 				pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
   1227 			}
   1228 			namei_end(state);
   1229 			return (ENOENT);
   1230 		}
   1231 		cnp->cn_nameptr = ndp->ni_pnbuf;
   1232 		error = do_lookup(state, state->namei_startdir);
   1233 		if (error != 0) {
   1234 			/* XXX this should use namei_end() */
   1235 			if (ndp->ni_dvp) {
   1236 				vput(ndp->ni_dvp);
   1237 			}
   1238 			if (ndp->ni_erootdir != NULL) {
   1239 				/* Retry the whole thing from the normal root */
   1240 				cnp->cn_flags &= ~TRYEMULROOT;
   1241 				goto emul_retry;
   1242 			}
   1243 			KASSERT(savepath == NULL);
   1244 			return (error);
   1245 		}
   1246 
   1247 		/*
   1248 		 * Check for symbolic link
   1249 		 */
   1250 		if (namei_atsymlink(state)) {
   1251 			error = namei_follow(state);
   1252 			if (error) {
   1253 				KASSERT(ndp->ni_dvp != ndp->ni_vp);
   1254 				vput(ndp->ni_dvp);
   1255 				vput(ndp->ni_vp);
   1256 				ndp->ni_vp = NULL;
   1257 				if (savepath != NULL) {
   1258 					pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
   1259 				}
   1260 				return error;
   1261 			}
   1262 		}
   1263 		else {
   1264 			break;
   1265 		}
   1266 	}
   1267 
   1268 	/*
   1269 	 * Done
   1270 	 */
   1271 
   1272 	if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) {
   1273 		if (ndp->ni_dvp == ndp->ni_vp) {
   1274 			vrele(ndp->ni_dvp);
   1275 		} else {
   1276 			vput(ndp->ni_dvp);
   1277 		}
   1278 	}
   1279 
   1280 	if (savepath != NULL) {
   1281 		pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
   1282 	}
   1283 
   1284 	return 0;
   1285 }
   1286 
   1287 int
   1288 namei(struct nameidata *ndp)
   1289 {
   1290 	struct namei_state state;
   1291 	int error;
   1292 
   1293 	namei_init(&state, ndp);
   1294 	error = do_namei(&state);
   1295 	namei_cleanup(&state);
   1296 
   1297 	return error;
   1298 }
   1299 
   1300 ////////////////////////////////////////////////////////////
   1301 
   1302 /*
   1303  * Externally visible interfaces used by nfsd (bletch, yuk, XXX)
   1304  *
   1305  * The "index" version differs from the "main" version in that it's
   1306  * called from a different place in a different context. For now I
   1307  * want to be able to shuffle code in from one call site without
   1308  * affecting the other.
   1309  */
   1310 
   1311 int
   1312 lookup_for_nfsd(struct nameidata *ndp, struct vnode *dp, int neverfollow)
   1313 {
   1314 	struct namei_state state;
   1315 	int error;
   1316 
   1317 	struct iovec aiov;
   1318 	struct uio auio;
   1319 	int linklen;
   1320 	char *cp;
   1321 
   1322 	ndp->ni_pnbuf = ndp->ni_pathbuf->pb_path;
   1323 	namei_init(&state, ndp);
   1324 
   1325 	/*
   1326 	 * BEGIN wodge of code from nfsd
   1327 	 */
   1328 
   1329 	vref(dp);
   1330 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
   1331 
   1332     for (;;) {
   1333 
   1334 	state.cnp->cn_nameptr = state.ndp->ni_pnbuf;
   1335 
   1336 	/*
   1337 	 * END wodge of code from nfsd
   1338 	 */
   1339 
   1340 	error = do_lookup(&state, dp);
   1341 	if (error) {
   1342 		/* BEGIN from nfsd */
   1343 		if (ndp->ni_dvp) {
   1344 			vput(ndp->ni_dvp);
   1345 		}
   1346 		/* END from nfsd */
   1347 		namei_cleanup(&state);
   1348 		return error;
   1349 	}
   1350 
   1351 	/*
   1352 	 * BEGIN wodge of code from nfsd
   1353 	 */
   1354 
   1355 	/*
   1356 	 * Check for encountering a symbolic link
   1357 	 */
   1358 	if ((state.cnp->cn_flags & ISSYMLINK) == 0) {
   1359 		if ((state.cnp->cn_flags & LOCKPARENT) == 0 && state.ndp->ni_dvp) {
   1360 			if (state.ndp->ni_dvp == state.ndp->ni_vp) {
   1361 				vrele(state.ndp->ni_dvp);
   1362 			} else {
   1363 				vput(state.ndp->ni_dvp);
   1364 			}
   1365 		}
   1366 		return (0);
   1367 	} else {
   1368 		if (neverfollow) {
   1369 			error = EINVAL;
   1370 			goto out;
   1371 		}
   1372 		if (state.ndp->ni_loopcnt++ >= MAXSYMLINKS) {
   1373 			error = ELOOP;
   1374 			goto out;
   1375 		}
   1376 		if (state.ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
   1377 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, state.cnp->cn_cred);
   1378 			if (error != 0)
   1379 				goto out;
   1380 		}
   1381 		cp = PNBUF_GET();
   1382 		aiov.iov_base = cp;
   1383 		aiov.iov_len = MAXPATHLEN;
   1384 		auio.uio_iov = &aiov;
   1385 		auio.uio_iovcnt = 1;
   1386 		auio.uio_offset = 0;
   1387 		auio.uio_rw = UIO_READ;
   1388 		auio.uio_resid = MAXPATHLEN;
   1389 		UIO_SETUP_SYSSPACE(&auio);
   1390 		error = VOP_READLINK(ndp->ni_vp, &auio, state.cnp->cn_cred);
   1391 		if (error) {
   1392 			PNBUF_PUT(cp);
   1393 			goto out;
   1394 		}
   1395 		linklen = MAXPATHLEN - auio.uio_resid;
   1396 		if (linklen == 0) {
   1397 			PNBUF_PUT(cp);
   1398 			error = ENOENT;
   1399 			goto out;
   1400 		}
   1401 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
   1402 			PNBUF_PUT(cp);
   1403 			error = ENAMETOOLONG;
   1404 			goto out;
   1405 		}
   1406 		if (ndp->ni_pathlen > 1) {
   1407 			/* includes a null-terminator */
   1408 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
   1409 		} else {
   1410 			cp[linklen] = '\0';
   1411 		}
   1412 		state.ndp->ni_pathlen += linklen;
   1413 		memcpy(state.ndp->ni_pnbuf, cp, state.ndp->ni_pathlen);
   1414 		PNBUF_PUT(cp);
   1415 		vput(state.ndp->ni_vp);
   1416 		dp = state.ndp->ni_dvp;
   1417 
   1418 		/*
   1419 		 * Check if root directory should replace current directory.
   1420 		 */
   1421 		if (state.ndp->ni_pnbuf[0] == '/') {
   1422 			vput(dp);
   1423 			dp = ndp->ni_rootdir;
   1424 			vref(dp);
   1425 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
   1426 		}
   1427 	}
   1428 
   1429     }
   1430  out:
   1431 	vput(state.ndp->ni_vp);
   1432 	vput(state.ndp->ni_dvp);
   1433 	state.ndp->ni_vp = NULL;
   1434 
   1435 	/*
   1436 	 * END wodge of code from nfsd
   1437 	 */
   1438 	namei_cleanup(&state);
   1439 
   1440 	return error;
   1441 }
   1442 
   1443 int
   1444 lookup_for_nfsd_index(struct nameidata *ndp, struct vnode *startdir)
   1445 {
   1446 	struct namei_state state;
   1447 	int error;
   1448 
   1449 	vref(startdir);
   1450 
   1451 	ndp->ni_pnbuf = ndp->ni_pathbuf->pb_path;
   1452 	ndp->ni_cnd.cn_nameptr = ndp->ni_pnbuf;
   1453 
   1454 	namei_init(&state, ndp);
   1455 	error = do_lookup(&state, startdir);
   1456 	namei_cleanup(&state);
   1457 
   1458 	return error;
   1459 }
   1460 
   1461 ////////////////////////////////////////////////////////////
   1462 
   1463 /*
   1464  * Reacquire a path name component.
   1465  * dvp is locked on entry and exit.
   1466  * *vpp is locked on exit unless it's NULL.
   1467  */
   1468 int
   1469 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, int dummy)
   1470 {
   1471 	int rdonly;			/* lookup read-only flag bit */
   1472 	int error = 0;
   1473 #ifdef DEBUG
   1474 	uint32_t newhash;		/* DEBUG: check name hash */
   1475 	const char *cp;			/* DEBUG: check name ptr/len */
   1476 #endif /* DEBUG */
   1477 
   1478 	(void)dummy;
   1479 
   1480 	/*
   1481 	 * Setup: break out flag bits into variables.
   1482 	 */
   1483 	rdonly = cnp->cn_flags & RDONLY;
   1484 	cnp->cn_flags &= ~ISSYMLINK;
   1485 
   1486 	/*
   1487 	 * Search a new directory.
   1488 	 *
   1489 	 * The cn_hash value is for use by vfs_cache.
   1490 	 * The last component of the filename is left accessible via
   1491 	 * cnp->cn_nameptr for callers that need the name. Callers needing
   1492 	 * the name set the SAVENAME flag. When done, they assume
   1493 	 * responsibility for freeing the pathname buffer.
   1494 	 */
   1495 #ifdef DEBUG
   1496 	cp = NULL;
   1497 	newhash = namei_hash(cnp->cn_nameptr, &cp);
   1498 	if ((uint32_t)newhash != (uint32_t)cnp->cn_hash)
   1499 		panic("relookup: bad hash");
   1500 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
   1501 		panic("relookup: bad len");
   1502 	while (*cp == '/')
   1503 		cp++;
   1504 	if (*cp != 0)
   1505 		panic("relookup: not last component");
   1506 #endif /* DEBUG */
   1507 
   1508 	/*
   1509 	 * Check for degenerate name (e.g. / or "")
   1510 	 * which is a way of talking about a directory,
   1511 	 * e.g. like "/." or ".".
   1512 	 */
   1513 	if (cnp->cn_nameptr[0] == '\0')
   1514 		panic("relookup: null name");
   1515 
   1516 	if (cnp->cn_flags & ISDOTDOT)
   1517 		panic("relookup: lookup on dot-dot");
   1518 
   1519 	/*
   1520 	 * We now have a segment name to search for, and a directory to search.
   1521 	 */
   1522 	cnp->cn_flags |= INRELOOKUP;
   1523 	error = VOP_LOOKUP(dvp, vpp, cnp);
   1524 	cnp->cn_flags &= ~INRELOOKUP;
   1525 	if ((error) != 0) {
   1526 #ifdef DIAGNOSTIC
   1527 		if (*vpp != NULL)
   1528 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
   1529 #endif
   1530 		if (error != EJUSTRETURN)
   1531 			goto bad;
   1532 	}
   1533 
   1534 #ifdef DIAGNOSTIC
   1535 	/*
   1536 	 * Check for symbolic link
   1537 	 */
   1538 	if (*vpp && (*vpp)->v_type == VLNK && (cnp->cn_flags & FOLLOW))
   1539 		panic("relookup: symlink found");
   1540 #endif
   1541 
   1542 	/*
   1543 	 * Check for read-only lookups.
   1544 	 */
   1545 	if (rdonly && cnp->cn_nameiop != LOOKUP) {
   1546 		error = EROFS;
   1547 		if (*vpp) {
   1548 			vput(*vpp);
   1549 		}
   1550 		goto bad;
   1551 	}
   1552 	return (0);
   1553 
   1554 bad:
   1555 	*vpp = NULL;
   1556 	return (error);
   1557 }
   1558 
   1559 /*
   1560  * namei_simple - simple forms of namei.
   1561  *
   1562  * These are wrappers to allow the simple case callers of namei to be
   1563  * left alone while everything else changes under them.
   1564  */
   1565 
   1566 /* Flags */
   1567 struct namei_simple_flags_type {
   1568 	int dummy;
   1569 };
   1570 static const struct namei_simple_flags_type ns_nn, ns_nt, ns_fn, ns_ft;
   1571 const namei_simple_flags_t NSM_NOFOLLOW_NOEMULROOT = &ns_nn;
   1572 const namei_simple_flags_t NSM_NOFOLLOW_TRYEMULROOT = &ns_nt;
   1573 const namei_simple_flags_t NSM_FOLLOW_NOEMULROOT = &ns_fn;
   1574 const namei_simple_flags_t NSM_FOLLOW_TRYEMULROOT = &ns_ft;
   1575 
   1576 static
   1577 int
   1578 namei_simple_convert_flags(namei_simple_flags_t sflags)
   1579 {
   1580 	if (sflags == NSM_NOFOLLOW_NOEMULROOT)
   1581 		return NOFOLLOW | 0;
   1582 	if (sflags == NSM_NOFOLLOW_TRYEMULROOT)
   1583 		return NOFOLLOW | TRYEMULROOT;
   1584 	if (sflags == NSM_FOLLOW_NOEMULROOT)
   1585 		return FOLLOW | 0;
   1586 	if (sflags == NSM_FOLLOW_TRYEMULROOT)
   1587 		return FOLLOW | TRYEMULROOT;
   1588 	panic("namei_simple_convert_flags: bogus sflags\n");
   1589 	return 0;
   1590 }
   1591 
   1592 int
   1593 namei_simple_kernel(const char *path, namei_simple_flags_t sflags,
   1594 			struct vnode **vp_ret)
   1595 {
   1596 	struct nameidata nd;
   1597 	struct pathbuf *pb;
   1598 	int err;
   1599 
   1600 	pb = pathbuf_create(path);
   1601 	if (pb == NULL) {
   1602 		return ENOMEM;
   1603 	}
   1604 
   1605 	NDINIT(&nd,
   1606 		LOOKUP,
   1607 		namei_simple_convert_flags(sflags),
   1608 		pb);
   1609 	err = namei(&nd);
   1610 	if (err != 0) {
   1611 		pathbuf_destroy(pb);
   1612 		return err;
   1613 	}
   1614 	*vp_ret = nd.ni_vp;
   1615 	pathbuf_destroy(pb);
   1616 	return 0;
   1617 }
   1618 
   1619 int
   1620 namei_simple_user(const char *path, namei_simple_flags_t sflags,
   1621 			struct vnode **vp_ret)
   1622 {
   1623 	struct pathbuf *pb;
   1624 	struct nameidata nd;
   1625 	int err;
   1626 
   1627 	err = pathbuf_copyin(path, &pb);
   1628 	if (err) {
   1629 		return err;
   1630 	}
   1631 
   1632 	NDINIT(&nd,
   1633 		LOOKUP,
   1634 		namei_simple_convert_flags(sflags),
   1635 		pb);
   1636 	err = namei(&nd);
   1637 	if (err != 0) {
   1638 		pathbuf_destroy(pb);
   1639 		return err;
   1640 	}
   1641 	*vp_ret = nd.ni_vp;
   1642 	pathbuf_destroy(pb);
   1643 	return 0;
   1644 }
   1645