xref: /netbsd-src/sys/kern/vfs_lookup.c (revision b757af438b42b93f8c6571f026d8b8ef3eaf5fc9)
1 /*	$NetBSD: vfs_lookup.c,v 1.192 2011/09/27 02:10:55 christos 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 1.192 2011/09/27 02:10:55 christos 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 #include <sys/dirent.h>
60 
61 #ifndef MAGICLINKS
62 #define MAGICLINKS 0
63 #endif
64 
65 int vfs_magiclinks = MAGICLINKS;
66 
67 __CTASSERT(MAXNAMLEN == NAME_MAX);
68 
69 /*
70  * Substitute replacement text for 'magic' strings in symlinks.
71  * Returns 0 if successful, and returns non-zero if an error
72  * occurs.  (Currently, the only possible error is running out
73  * of temporary pathname space.)
74  *
75  * Looks for "@<string>" and "@<string>/", where <string> is a
76  * recognized 'magic' string.  Replaces the "@<string>" with the
77  * appropriate replacement text.  (Note that in some cases the
78  * replacement text may have zero length.)
79  *
80  * This would have been table driven, but the variance in
81  * replacement strings (and replacement string lengths) made
82  * that impractical.
83  */
84 #define	VNL(x)							\
85 	(sizeof(x) - 1)
86 
87 #define	VO	'{'
88 #define	VC	'}'
89 
90 #define	MATCH(str)						\
91 	((termchar == '/' && i + VNL(str) == *len) ||		\
92 	 (i + VNL(str) < *len &&				\
93 	  cp[i + VNL(str)] == termchar)) &&			\
94 	!strncmp((str), &cp[i], VNL(str))
95 
96 #define	SUBSTITUTE(m, s, sl)					\
97 	if ((newlen + (sl)) >= MAXPATHLEN)			\
98 		return 1;					\
99 	i += VNL(m);						\
100 	if (termchar != '/')					\
101 		i++;						\
102 	(void)memcpy(&tmp[newlen], (s), (sl));			\
103 	newlen += (sl);						\
104 	change = 1;						\
105 	termchar = '/';
106 
107 static int
108 symlink_magic(struct proc *p, char *cp, size_t *len)
109 {
110 	char *tmp;
111 	size_t change, i, newlen, slen;
112 	char termchar = '/';
113 	char idtmp[11]; /* enough for 32 bit *unsigned* integer */
114 
115 
116 	tmp = PNBUF_GET();
117 	for (change = i = newlen = 0; i < *len; ) {
118 		if (cp[i] != '@') {
119 			tmp[newlen++] = cp[i++];
120 			continue;
121 		}
122 
123 		i++;
124 
125 		/* Check for @{var} syntax. */
126 		if (cp[i] == VO) {
127 			termchar = VC;
128 			i++;
129 		}
130 
131 		/*
132 		 * The following checks should be ordered according
133 		 * to frequency of use.
134 		 */
135 		if (MATCH("machine_arch")) {
136 			slen = VNL(MACHINE_ARCH);
137 			SUBSTITUTE("machine_arch", MACHINE_ARCH, slen);
138 		} else if (MATCH("machine")) {
139 			slen = VNL(MACHINE);
140 			SUBSTITUTE("machine", MACHINE, slen);
141 		} else if (MATCH("hostname")) {
142 			SUBSTITUTE("hostname", hostname, hostnamelen);
143 		} else if (MATCH("osrelease")) {
144 			slen = strlen(osrelease);
145 			SUBSTITUTE("osrelease", osrelease, slen);
146 		} else if (MATCH("emul")) {
147 			slen = strlen(p->p_emul->e_name);
148 			SUBSTITUTE("emul", p->p_emul->e_name, slen);
149 		} else if (MATCH("kernel_ident")) {
150 			slen = strlen(kernel_ident);
151 			SUBSTITUTE("kernel_ident", kernel_ident, slen);
152 		} else if (MATCH("domainname")) {
153 			SUBSTITUTE("domainname", domainname, domainnamelen);
154 		} else if (MATCH("ostype")) {
155 			slen = strlen(ostype);
156 			SUBSTITUTE("ostype", ostype, slen);
157 		} else if (MATCH("uid")) {
158 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
159 			    kauth_cred_geteuid(kauth_cred_get()));
160 			SUBSTITUTE("uid", idtmp, slen);
161 		} else if (MATCH("ruid")) {
162 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
163 			    kauth_cred_getuid(kauth_cred_get()));
164 			SUBSTITUTE("ruid", idtmp, slen);
165 		} else if (MATCH("gid")) {
166 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
167 			    kauth_cred_getegid(kauth_cred_get()));
168 			SUBSTITUTE("gid", idtmp, slen);
169 		} else if (MATCH("rgid")) {
170 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
171 			    kauth_cred_getgid(kauth_cred_get()));
172 			SUBSTITUTE("rgid", idtmp, slen);
173 		} else {
174 			tmp[newlen++] = '@';
175 			if (termchar == VC)
176 				tmp[newlen++] = VO;
177 		}
178 	}
179 
180 	if (change) {
181 		(void)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 
197 /*
198  * Determine the namei hash (for cn_hash) for name.
199  * If *ep != NULL, hash from name to ep-1.
200  * If *ep == NULL, hash from name until the first NUL or '/', and
201  * return the location of this termination character in *ep.
202  *
203  * This function returns an equivalent hash to the MI hash32_strn().
204  * The latter isn't used because in the *ep == NULL case, determining
205  * the length of the string to the first NUL or `/' and then calling
206  * hash32_strn() involves unnecessary double-handling of the data.
207  */
208 uint32_t
209 namei_hash(const char *name, const char **ep)
210 {
211 	uint32_t	hash;
212 
213 	hash = HASH32_STR_INIT;
214 	if (*ep != NULL) {
215 		for (; name < *ep; name++)
216 			hash = hash * 33 + *(const uint8_t *)name;
217 	} else {
218 		for (; *name != '\0' && *name != '/'; name++)
219 			hash = hash * 33 + *(const uint8_t *)name;
220 		*ep = name;
221 	}
222 	return (hash + (hash >> 5));
223 }
224 
225 ////////////////////////////////////////////////////////////
226 
227 /*
228  * Sealed abstraction for pathnames.
229  *
230  * System-call-layer level code that is going to call namei should
231  * first create a pathbuf and adjust all the bells and whistles on it
232  * as needed by context.
233  */
234 
235 struct pathbuf {
236 	char *pb_path;
237 	char *pb_pathcopy;
238 	unsigned pb_pathcopyuses;
239 };
240 
241 static struct pathbuf *
242 pathbuf_create_raw(void)
243 {
244 	struct pathbuf *pb;
245 
246 	pb = kmem_alloc(sizeof(*pb), KM_SLEEP);
247 	if (pb == NULL) {
248 		return NULL;
249 	}
250 	pb->pb_path = PNBUF_GET();
251 	if (pb->pb_path == NULL) {
252 		kmem_free(pb, sizeof(*pb));
253 		return NULL;
254 	}
255 	pb->pb_pathcopy = NULL;
256 	pb->pb_pathcopyuses = 0;
257 	return pb;
258 }
259 
260 void
261 pathbuf_destroy(struct pathbuf *pb)
262 {
263 	KASSERT(pb->pb_pathcopyuses == 0);
264 	KASSERT(pb->pb_pathcopy == NULL);
265 	PNBUF_PUT(pb->pb_path);
266 	kmem_free(pb, sizeof(*pb));
267 }
268 
269 struct pathbuf *
270 pathbuf_assimilate(char *pnbuf)
271 {
272 	struct pathbuf *pb;
273 
274 	pb = kmem_alloc(sizeof(*pb), KM_SLEEP);
275 	if (pb == NULL) {
276 		return NULL;
277 	}
278 	pb->pb_path = pnbuf;
279 	pb->pb_pathcopy = NULL;
280 	pb->pb_pathcopyuses = 0;
281 	return pb;
282 }
283 
284 struct pathbuf *
285 pathbuf_create(const char *path)
286 {
287 	struct pathbuf *pb;
288 	int error;
289 
290 	pb = pathbuf_create_raw();
291 	if (pb == NULL) {
292 		return NULL;
293 	}
294 	error = copystr(path, pb->pb_path, PATH_MAX, NULL);
295 	if (error != 0) {
296 		KASSERT(!"kernel path too long in pathbuf_create");
297 		/* make sure it's null-terminated, just in case */
298 		pb->pb_path[PATH_MAX-1] = '\0';
299 	}
300 	return pb;
301 }
302 
303 int
304 pathbuf_copyin(const char *userpath, struct pathbuf **ret)
305 {
306 	struct pathbuf *pb;
307 	int error;
308 
309 	pb = pathbuf_create_raw();
310 	if (pb == NULL) {
311 		return ENOMEM;
312 	}
313 	error = copyinstr(userpath, pb->pb_path, PATH_MAX, NULL);
314 	if (error) {
315 		pathbuf_destroy(pb);
316 		return error;
317 	}
318 	*ret = pb;
319 	return 0;
320 }
321 
322 /*
323  * XXX should not exist:
324  *   1. whether a pointer is kernel or user should be statically checkable.
325  *   2. copyin should be handled by the upper part of the syscall layer,
326  *      not in here.
327  */
328 int
329 pathbuf_maybe_copyin(const char *path, enum uio_seg seg, struct pathbuf **ret)
330 {
331 	if (seg == UIO_USERSPACE) {
332 		return pathbuf_copyin(path, ret);
333 	} else {
334 		*ret = pathbuf_create(path);
335 		if (*ret == NULL) {
336 			return ENOMEM;
337 		}
338 		return 0;
339 	}
340 }
341 
342 /*
343  * Get a copy of the path buffer as it currently exists. If this is
344  * called after namei starts the results may be arbitrary.
345  */
346 void
347 pathbuf_copystring(const struct pathbuf *pb, char *buf, size_t maxlen)
348 {
349 	strlcpy(buf, pb->pb_path, maxlen);
350 }
351 
352 /*
353  * These two functions allow access to a saved copy of the original
354  * path string. The first copy should be gotten before namei is
355  * called. Each copy that is gotten should be put back.
356  */
357 
358 const char *
359 pathbuf_stringcopy_get(struct pathbuf *pb)
360 {
361 	if (pb->pb_pathcopyuses == 0) {
362 		pb->pb_pathcopy = PNBUF_GET();
363 		strcpy(pb->pb_pathcopy, pb->pb_path);
364 	}
365 	pb->pb_pathcopyuses++;
366 	return pb->pb_pathcopy;
367 }
368 
369 void
370 pathbuf_stringcopy_put(struct pathbuf *pb, const char *str)
371 {
372 	KASSERT(str == pb->pb_pathcopy);
373 	KASSERT(pb->pb_pathcopyuses > 0);
374 	pb->pb_pathcopyuses--;
375 	if (pb->pb_pathcopyuses == 0) {
376 		PNBUF_PUT(pb->pb_pathcopy);
377 		pb->pb_pathcopy = NULL;
378 	}
379 }
380 
381 
382 ////////////////////////////////////////////////////////////
383 
384 /*
385  * namei: convert a pathname into a pointer to a (maybe-locked) vnode,
386  * and maybe also its parent directory vnode, and assorted other guff.
387  * See namei(9) for the interface documentation.
388  *
389  *
390  * The FOLLOW flag is set when symbolic links are to be followed
391  * when they occur at the end of the name translation process.
392  * Symbolic links are always followed for all other pathname
393  * components other than the last.
394  *
395  * The segflg defines whether the name is to be copied from user
396  * space or kernel space.
397  *
398  * Overall outline of namei:
399  *
400  *	copy in name
401  *	get starting directory
402  *	while (!done && !error) {
403  *		call lookup to search path.
404  *		if symbolic link, massage name in buffer and continue
405  *	}
406  */
407 
408 /*
409  * Search a pathname.
410  * This is a very central and rather complicated routine.
411  *
412  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
413  * The starting directory is passed in. The pathname is descended
414  * until done, or a symbolic link is encountered. The variable ni_more
415  * is clear if the path is completed; it is set to one if a symbolic
416  * link needing interpretation is encountered.
417  *
418  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
419  * whether the name is to be looked up, created, renamed, or deleted.
420  * When CREATE, RENAME, or DELETE is specified, information usable in
421  * creating, renaming, or deleting a directory entry may be calculated.
422  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
423  * locked.  Otherwise the parent directory is not returned. If the target
424  * of the pathname exists and LOCKLEAF is or'ed into the flag the target
425  * is returned locked, otherwise it is returned unlocked.  When creating
426  * or renaming and LOCKPARENT is specified, the target may not be ".".
427  * When deleting and LOCKPARENT is specified, the target may be ".".
428  *
429  * Overall outline of lookup:
430  *
431  * dirloop:
432  *	identify next component of name at ndp->ni_ptr
433  *	handle degenerate case where name is null string
434  *	if .. and crossing mount points and on mounted filesys, find parent
435  *	call VOP_LOOKUP routine for next component name
436  *	    directory vnode returned in ni_dvp, locked.
437  *	    component vnode returned in ni_vp (if it exists), locked.
438  *	if result vnode is mounted on and crossing mount points,
439  *	    find mounted on vnode
440  *	if more components of name, do next level at dirloop
441  *	return the answer in ni_vp, locked if LOCKLEAF set
442  *	    if LOCKPARENT set, return locked parent in ni_dvp
443  */
444 
445 
446 /*
447  * Internal state for a namei operation.
448  *
449  * cnp is always equal to &ndp->ni_cnp.
450  */
451 struct namei_state {
452 	struct nameidata *ndp;
453 	struct componentname *cnp;
454 
455 	int docache;			/* == 0 do not cache last component */
456 	int rdonly;			/* lookup read-only flag bit */
457 	int slashes;
458 
459 	unsigned attempt_retry:1;	/* true if error allows emul retry */
460 };
461 
462 
463 /*
464  * Initialize the namei working state.
465  */
466 static void
467 namei_init(struct namei_state *state, struct nameidata *ndp)
468 {
469 	state->ndp = ndp;
470 	state->cnp = &ndp->ni_cnd;
471 	KASSERT((state->cnp->cn_flags & INRELOOKUP) == 0);
472 
473 	state->docache = 0;
474 	state->rdonly = 0;
475 	state->slashes = 0;
476 
477 #ifdef DIAGNOSTIC
478 	if (!state->cnp->cn_cred)
479 		panic("namei: bad cred/proc");
480 	if (state->cnp->cn_nameiop & (~OPMASK))
481 		panic("namei: nameiop contaminated with flags");
482 	if (state->cnp->cn_flags & OPMASK)
483 		panic("namei: flags contaminated with nameiops");
484 #endif
485 
486 	/*
487 	 * The buffer for name translation shall be the one inside the
488 	 * pathbuf.
489 	 */
490 	state->ndp->ni_pnbuf = state->ndp->ni_pathbuf->pb_path;
491 }
492 
493 /*
494  * Clean up the working namei state, leaving things ready for return
495  * from namei.
496  */
497 static void
498 namei_cleanup(struct namei_state *state)
499 {
500 	KASSERT(state->cnp == &state->ndp->ni_cnd);
501 
502 	/* nothing for now */
503 	(void)state;
504 }
505 
506 //////////////////////////////
507 
508 /*
509  * Get the directory context.
510  * Initializes the rootdir and erootdir state and returns a reference
511  * to the starting dir.
512  */
513 static struct vnode *
514 namei_getstartdir(struct namei_state *state)
515 {
516 	struct nameidata *ndp = state->ndp;
517 	struct componentname *cnp = state->cnp;
518 	struct cwdinfo *cwdi;		/* pointer to cwd state */
519 	struct lwp *self = curlwp;	/* thread doing namei() */
520 	struct vnode *rootdir, *erootdir, *curdir, *startdir;
521 
522 	cwdi = self->l_proc->p_cwdi;
523 	rw_enter(&cwdi->cwdi_lock, RW_READER);
524 
525 	/* root dir */
526 	if (cwdi->cwdi_rdir == NULL || (cnp->cn_flags & NOCHROOT)) {
527 		rootdir = rootvnode;
528 	} else {
529 		rootdir = cwdi->cwdi_rdir;
530 	}
531 
532 	/* emulation root dir, if any */
533 	if ((cnp->cn_flags & TRYEMULROOT) == 0) {
534 		/* if we don't want it, don't fetch it */
535 		erootdir = NULL;
536 	} else if (cnp->cn_flags & EMULROOTSET) {
537 		/* explicitly set emulroot; "/../" doesn't override this */
538 		erootdir = ndp->ni_erootdir;
539 	} else if (!strncmp(ndp->ni_pnbuf, "/../", 4)) {
540 		/* explicit reference to real rootdir */
541 		erootdir = NULL;
542 	} else {
543 		/* may be null */
544 		erootdir = cwdi->cwdi_edir;
545 	}
546 
547 	/* current dir */
548 	curdir = cwdi->cwdi_cdir;
549 
550 	if (ndp->ni_pnbuf[0] != '/') {
551 		startdir = curdir;
552 		erootdir = NULL;
553 	} else if (cnp->cn_flags & TRYEMULROOT && erootdir != NULL) {
554 		startdir = erootdir;
555 	} else {
556 		startdir = rootdir;
557 		erootdir = NULL;
558 	}
559 
560 	state->ndp->ni_rootdir = rootdir;
561 	state->ndp->ni_erootdir = erootdir;
562 
563 	/*
564 	 * Get a reference to the start dir so we can safely unlock cwdi.
565 	 *
566 	 * XXX: should we hold references to rootdir and erootdir while
567 	 * we're running? What happens if a multithreaded process chroots
568 	 * during namei?
569 	 */
570 	vref(startdir);
571 
572 	rw_exit(&cwdi->cwdi_lock);
573 	return startdir;
574 }
575 
576 /*
577  * Get the directory context for the nfsd case, in parallel to
578  * getstartdir. Initializes the rootdir and erootdir state and
579  * returns a reference to the passed-in starting dir.
580  */
581 static struct vnode *
582 namei_getstartdir_for_nfsd(struct namei_state *state, struct vnode *startdir)
583 {
584 	/* always use the real root, and never set an emulation root */
585 	state->ndp->ni_rootdir = rootvnode;
586 	state->ndp->ni_erootdir = NULL;
587 
588 	vref(startdir);
589 	return startdir;
590 }
591 
592 
593 /*
594  * Ktrace the namei operation.
595  */
596 static void
597 namei_ktrace(struct namei_state *state)
598 {
599 	struct nameidata *ndp = state->ndp;
600 	struct componentname *cnp = state->cnp;
601 	struct lwp *self = curlwp;	/* thread doing namei() */
602 	const char *emul_path;
603 
604 	if (ktrpoint(KTR_NAMEI)) {
605 		if (ndp->ni_erootdir != NULL) {
606 			/*
607 			 * To make any sense, the trace entry need to have the
608 			 * text of the emulation path prepended.
609 			 * Usually we can get this from the current process,
610 			 * but when called from emul_find_interp() it is only
611 			 * in the exec_package - so we get it passed in ni_next
612 			 * (this is a hack).
613 			 */
614 			if (cnp->cn_flags & EMULROOTSET)
615 				emul_path = ndp->ni_next;
616 			else
617 				emul_path = self->l_proc->p_emul->e_path;
618 			ktrnamei2(emul_path, strlen(emul_path),
619 			    ndp->ni_pnbuf, ndp->ni_pathlen);
620 		} else
621 			ktrnamei(ndp->ni_pnbuf, ndp->ni_pathlen);
622 	}
623 }
624 
625 /*
626  * Start up namei. Find the root dir and cwd, establish the starting
627  * directory for lookup, and lock it. Also calls ktrace when
628  * appropriate.
629  */
630 static int
631 namei_start(struct namei_state *state, struct vnode *forcecwd,
632 	    struct vnode **startdir_ret)
633 {
634 	struct nameidata *ndp = state->ndp;
635 	struct vnode *startdir;
636 
637 	/* length includes null terminator (was originally from copyinstr) */
638 	ndp->ni_pathlen = strlen(ndp->ni_pnbuf) + 1;
639 
640 	/*
641 	 * POSIX.1 requirement: "" is not a valid file name.
642 	 */
643 	if (ndp->ni_pathlen == 1) {
644 		return ENOENT;
645 	}
646 
647 	ndp->ni_loopcnt = 0;
648 
649 	/* Get starting directory, set up root, and ktrace. */
650 	if (forcecwd != NULL) {
651 		startdir = namei_getstartdir_for_nfsd(state, forcecwd);
652 		/* no ktrace */
653 	} else {
654 		startdir = namei_getstartdir(state);
655 		namei_ktrace(state);
656 	}
657 
658 	vn_lock(startdir, LK_EXCLUSIVE | LK_RETRY);
659 
660 	*startdir_ret = startdir;
661 	return 0;
662 }
663 
664 /*
665  * Check for being at a symlink that we're going to follow.
666  */
667 static inline int
668 namei_atsymlink(struct namei_state *state, struct vnode *foundobj)
669 {
670 	return (foundobj->v_type == VLNK) &&
671 		(state->cnp->cn_flags & (FOLLOW|REQUIREDIR));
672 }
673 
674 /*
675  * Follow a symlink.
676  *
677  * Updates searchdir. inhibitmagic causes magic symlinks to not be
678  * interpreted; this is used by nfsd.
679  *
680  * Unlocks foundobj on success (ugh)
681  */
682 static inline int
683 namei_follow(struct namei_state *state, int inhibitmagic,
684 	     struct vnode *searchdir, struct vnode *foundobj,
685 	     struct vnode **newsearchdir_ret)
686 {
687 	struct nameidata *ndp = state->ndp;
688 	struct componentname *cnp = state->cnp;
689 
690 	struct lwp *self = curlwp;	/* thread doing namei() */
691 	struct iovec aiov;		/* uio for reading symbolic links */
692 	struct uio auio;
693 	char *cp;			/* pointer into pathname argument */
694 	size_t linklen;
695 	int error;
696 
697 	KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
698 	KASSERT(VOP_ISLOCKED(foundobj) == LK_EXCLUSIVE);
699 	if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
700 		return ELOOP;
701 	}
702 	if (foundobj->v_mount->mnt_flag & MNT_SYMPERM) {
703 		error = VOP_ACCESS(foundobj, VEXEC, cnp->cn_cred);
704 		if (error != 0)
705 			return error;
706 	}
707 
708 	/* FUTURE: fix this to not use a second buffer */
709 	cp = PNBUF_GET();
710 	aiov.iov_base = cp;
711 	aiov.iov_len = MAXPATHLEN;
712 	auio.uio_iov = &aiov;
713 	auio.uio_iovcnt = 1;
714 	auio.uio_offset = 0;
715 	auio.uio_rw = UIO_READ;
716 	auio.uio_resid = MAXPATHLEN;
717 	UIO_SETUP_SYSSPACE(&auio);
718 	error = VOP_READLINK(foundobj, &auio, cnp->cn_cred);
719 	if (error) {
720 		PNBUF_PUT(cp);
721 		return error;
722 	}
723 	linklen = MAXPATHLEN - auio.uio_resid;
724 	if (linklen == 0) {
725 		PNBUF_PUT(cp);
726 		return ENOENT;
727 	}
728 
729 	/*
730 	 * Do symlink substitution, if appropriate, and
731 	 * check length for potential overflow.
732 	 *
733 	 * Inhibit symlink substitution for nfsd.
734 	 * XXX: This is how it was before; is that a bug or a feature?
735 	 */
736 	if ((!inhibitmagic && vfs_magiclinks &&
737 	     symlink_magic(self->l_proc, cp, &linklen)) ||
738 	    (linklen + ndp->ni_pathlen >= MAXPATHLEN)) {
739 		PNBUF_PUT(cp);
740 		return ENAMETOOLONG;
741 	}
742 	if (ndp->ni_pathlen > 1) {
743 		/* includes a null-terminator */
744 		memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
745 	} else {
746 		cp[linklen] = '\0';
747 	}
748 	ndp->ni_pathlen += linklen;
749 	memcpy(ndp->ni_pnbuf, cp, ndp->ni_pathlen);
750 	PNBUF_PUT(cp);
751 
752 	/* we're now starting from the beginning of the buffer again */
753 	cnp->cn_nameptr = ndp->ni_pnbuf;
754 
755 	/* must unlock this before relocking searchdir */
756 	VOP_UNLOCK(foundobj);
757 
758 	/*
759 	 * Check if root directory should replace current directory.
760 	 */
761 	if (ndp->ni_pnbuf[0] == '/') {
762 		vput(searchdir);
763 		/* Keep absolute symbolic links inside emulation root */
764 		searchdir = ndp->ni_erootdir;
765 		if (searchdir == NULL ||
766 		    (ndp->ni_pnbuf[1] == '.'
767 		     && ndp->ni_pnbuf[2] == '.'
768 		     && ndp->ni_pnbuf[3] == '/')) {
769 			ndp->ni_erootdir = NULL;
770 			searchdir = ndp->ni_rootdir;
771 		}
772 		vref(searchdir);
773 		vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
774 		while (cnp->cn_nameptr[0] == '/') {
775 			cnp->cn_nameptr++;
776 			ndp->ni_pathlen--;
777 		}
778 	}
779 
780 	*newsearchdir_ret = searchdir;
781 	KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
782 	return 0;
783 }
784 
785 //////////////////////////////
786 
787 /*
788  * Inspect the leading path component and update the state accordingly.
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 the search dir
810 	 * is held and locked.
811 	 */
812 	cnp->cn_consume = 0;
813 	cp = NULL;
814 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
815 	cnp->cn_namelen = cp - cnp->cn_nameptr;
816 	if (cnp->cn_namelen > KERNEL_NAME_MAX) {
817 		return ENAMETOOLONG;
818 	}
819 #ifdef NAMEI_DIAGNOSTIC
820 	{ char c = *cp;
821 	*(char *)cp = '\0';
822 	printf("{%s}: ", cnp->cn_nameptr);
823 	*(char *)cp = c; }
824 #endif /* NAMEI_DIAGNOSTIC */
825 	ndp->ni_pathlen -= cnp->cn_namelen;
826 	ndp->ni_next = cp;
827 	/*
828 	 * If this component is followed by a slash, then move the pointer to
829 	 * the next component forward, and remember that this component must be
830 	 * a directory.
831 	 */
832 	if (*cp == '/') {
833 		do {
834 			cp++;
835 		} while (*cp == '/');
836 		state->slashes = cp - ndp->ni_next;
837 		ndp->ni_pathlen -= state->slashes;
838 		ndp->ni_next = cp;
839 		cnp->cn_flags |= REQUIREDIR;
840 	} else {
841 		state->slashes = 0;
842 		cnp->cn_flags &= ~REQUIREDIR;
843 	}
844 	/*
845 	 * We do special processing on the last component, whether or not it's
846 	 * a directory.  Cache all intervening lookups, but not the final one.
847 	 */
848 	if (*cp == '\0') {
849 		if (state->docache)
850 			cnp->cn_flags |= MAKEENTRY;
851 		else
852 			cnp->cn_flags &= ~MAKEENTRY;
853 		cnp->cn_flags |= ISLASTCN;
854 	} else {
855 		cnp->cn_flags |= MAKEENTRY;
856 		cnp->cn_flags &= ~ISLASTCN;
857 	}
858 	if (cnp->cn_namelen == 2 &&
859 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
860 		cnp->cn_flags |= ISDOTDOT;
861 	else
862 		cnp->cn_flags &= ~ISDOTDOT;
863 
864 	return 0;
865 }
866 
867 /*
868  * Call VOP_LOOKUP for a single lookup; return a new search directory
869  * (used when crossing mountpoints up or searching union mounts down) and
870  * the found object, which for create operations may be NULL on success.
871  */
872 static int
873 lookup_once(struct namei_state *state,
874 	    struct vnode *searchdir,
875 	    struct vnode **newsearchdir_ret,
876 	    struct vnode **foundobj_ret)
877 {
878 	struct vnode *tmpvn;		/* scratch vnode */
879 	struct vnode *foundobj;		/* result */
880 	struct mount *mp;		/* mount table entry */
881 	struct lwp *l = curlwp;
882 	int error;
883 
884 	struct componentname *cnp = state->cnp;
885 	struct nameidata *ndp = state->ndp;
886 
887 	KASSERT(cnp == &ndp->ni_cnd);
888 	KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
889 	*newsearchdir_ret = searchdir;
890 
891 	/*
892 	 * Handle "..": two special cases.
893 	 * 1. If at root directory (e.g. after chroot)
894 	 *    or at absolute root directory
895 	 *    then ignore it so can't get out.
896 	 * 1a. If at the root of the emulation filesystem go to the real
897 	 *    root. So "/../<path>" is always absolute.
898 	 * 1b. If we have somehow gotten out of a jail, warn
899 	 *    and also ignore it so we can't get farther out.
900 	 * 2. If this vnode is the root of a mounted
901 	 *    filesystem, then replace it with the
902 	 *    vnode which was mounted on so we take the
903 	 *    .. in the other file system.
904 	 */
905 	if (cnp->cn_flags & ISDOTDOT) {
906 		struct proc *p = l->l_proc;
907 
908 		for (;;) {
909 			if (searchdir == ndp->ni_rootdir ||
910 			    searchdir == rootvnode) {
911 				foundobj = searchdir;
912 				vref(foundobj);
913 				*foundobj_ret = foundobj;
914 				error = 0;
915 				goto done;
916 			}
917 			if (ndp->ni_rootdir != rootvnode) {
918 				int retval;
919 
920 				VOP_UNLOCK(searchdir);
921 				retval = vn_isunder(searchdir, ndp->ni_rootdir, l);
922 				vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
923 				if (!retval) {
924 				    /* Oops! We got out of jail! */
925 				    log(LOG_WARNING,
926 					"chrooted pid %d uid %d (%s) "
927 					"detected outside of its chroot\n",
928 					p->p_pid, kauth_cred_geteuid(l->l_cred),
929 					p->p_comm);
930 				    /* Put us at the jail root. */
931 				    vput(searchdir);
932 				    searchdir = NULL;
933 				    foundobj = ndp->ni_rootdir;
934 				    vref(foundobj);
935 				    vref(foundobj);
936 				    vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
937 				    *newsearchdir_ret = foundobj;
938 				    *foundobj_ret = foundobj;
939 				    error = 0;
940 				    goto done;
941 				}
942 			}
943 			if ((searchdir->v_vflag & VV_ROOT) == 0 ||
944 			    (cnp->cn_flags & NOCROSSMOUNT))
945 				break;
946 			tmpvn = searchdir;
947 			searchdir = searchdir->v_mount->mnt_vnodecovered;
948 			vref(searchdir);
949 			vput(tmpvn);
950 			vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
951 			*newsearchdir_ret = searchdir;
952 		}
953 	}
954 
955 	/*
956 	 * We now have a segment name to search for, and a directory to search.
957 	 * Our vnode state here is that "searchdir" is held and locked.
958 	 */
959 unionlookup:
960 	foundobj = NULL;
961 	error = VOP_LOOKUP(searchdir, &foundobj, cnp);
962 
963 	if (error != 0) {
964 #ifdef DIAGNOSTIC
965 		if (foundobj != NULL)
966 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
967 #endif /* DIAGNOSTIC */
968 #ifdef NAMEI_DIAGNOSTIC
969 		printf("not found\n");
970 #endif /* NAMEI_DIAGNOSTIC */
971 		if ((error == ENOENT) &&
972 		    (searchdir->v_vflag & VV_ROOT) &&
973 		    (searchdir->v_mount->mnt_flag & MNT_UNION)) {
974 			tmpvn = searchdir;
975 			searchdir = searchdir->v_mount->mnt_vnodecovered;
976 			vref(searchdir);
977 			vput(tmpvn);
978 			vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
979 			*newsearchdir_ret = searchdir;
980 			goto unionlookup;
981 		}
982 
983 		if (error != EJUSTRETURN)
984 			goto done;
985 
986 		/*
987 		 * If this was not the last component, or there were trailing
988 		 * slashes, and we are not going to create a directory,
989 		 * then the name must exist.
990 		 */
991 		if ((cnp->cn_flags & (REQUIREDIR | CREATEDIR)) == REQUIREDIR) {
992 			error = ENOENT;
993 			goto done;
994 		}
995 
996 		/*
997 		 * If creating and at end of pathname, then can consider
998 		 * allowing file to be created.
999 		 */
1000 		if (state->rdonly) {
1001 			error = EROFS;
1002 			goto done;
1003 		}
1004 
1005 		/*
1006 		 * We return success and a NULL foundobj to indicate
1007 		 * that the entry doesn't currently exist, leaving a
1008 		 * pointer to the (normally, locked) directory vnode
1009 		 * as searchdir.
1010 		 */
1011 		*foundobj_ret = NULL;
1012 		error = 0;
1013 		goto done;
1014 	}
1015 #ifdef NAMEI_DIAGNOSTIC
1016 	printf("found\n");
1017 #endif /* NAMEI_DIAGNOSTIC */
1018 
1019 	/*
1020 	 * Take into account any additional components consumed by the
1021 	 * underlying filesystem.  This will include any trailing slashes after
1022 	 * the last component consumed.
1023 	 */
1024 	if (cnp->cn_consume > 0) {
1025 		ndp->ni_pathlen -= cnp->cn_consume - state->slashes;
1026 		ndp->ni_next += cnp->cn_consume - state->slashes;
1027 		cnp->cn_consume = 0;
1028 		if (ndp->ni_next[0] == '\0')
1029 			cnp->cn_flags |= ISLASTCN;
1030 	}
1031 
1032 	/*
1033 	 * "foundobj" and "searchdir" are both locked and held,
1034 	 * and may be the same vnode.
1035 	 */
1036 
1037 	/*
1038 	 * Check to see if the vnode has been mounted on;
1039 	 * if so find the root of the mounted file system.
1040 	 */
1041 	while (foundobj->v_type == VDIR &&
1042 	       (mp = foundobj->v_mountedhere) != NULL &&
1043 	       (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1044 		error = vfs_busy(mp, NULL);
1045 		if (error != 0) {
1046 			if (searchdir != foundobj) {
1047 				vput(foundobj);
1048 			} else {
1049 				vrele(foundobj);
1050 			}
1051 			goto done;
1052 		}
1053 		if (searchdir != foundobj) {
1054 			VOP_UNLOCK(searchdir);
1055 		}
1056 		vput(foundobj);
1057 		error = VFS_ROOT(mp, &foundobj);
1058 		vfs_unbusy(mp, false, NULL);
1059 		if (error) {
1060 			vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
1061 			goto done;
1062 		}
1063 		/*
1064 		 * avoid locking vnodes from two filesystems because it's
1065 		 * prune to deadlock.  eg. when using puffs.
1066 		 * also, it isn't a good idea to propagate slowness of a
1067 		 * filesystem up to the root directory.
1068 		 * for now, only handle the common case.  (ie. foundobj is VDIR)
1069 		 */
1070 		if (foundobj->v_type == VDIR) {
1071 			vrele(searchdir);
1072 			*newsearchdir_ret = searchdir = foundobj;
1073 			vref(searchdir);
1074 		} else {
1075 			VOP_UNLOCK(foundobj);
1076 			vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
1077 			vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
1078 		}
1079 	}
1080 
1081 	*foundobj_ret = foundobj;
1082 	error = 0;
1083 done:
1084 	KASSERT(VOP_ISLOCKED(*newsearchdir_ret) == LK_EXCLUSIVE);
1085 	/*
1086 	 * *foundobj_ret is valid only if error == 0.
1087 	 */
1088 	KASSERT(error != 0 || *foundobj_ret == NULL ||
1089 	    VOP_ISLOCKED(*foundobj_ret) == LK_EXCLUSIVE);
1090 	return error;
1091 }
1092 
1093 //////////////////////////////
1094 
1095 /*
1096  * Do a complete path search from a single root directory.
1097  * (This is called up to twice if TRYEMULROOT is in effect.)
1098  */
1099 static int
1100 namei_oneroot(struct namei_state *state, struct vnode *forcecwd,
1101 	 int neverfollow, int inhibitmagic)
1102 {
1103 	struct nameidata *ndp = state->ndp;
1104 	struct componentname *cnp = state->cnp;
1105 	struct vnode *searchdir, *foundobj;
1106 	int error;
1107 
1108 	error = namei_start(state, forcecwd, &searchdir);
1109 	if (error) {
1110 		ndp->ni_dvp = NULL;
1111 		ndp->ni_vp = NULL;
1112 		return error;
1113 	}
1114 	KASSERT(searchdir->v_type == VDIR);
1115 
1116 	/*
1117 	 * Setup: break out flag bits into variables.
1118 	 */
1119 	state->docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
1120 	if (cnp->cn_nameiop == DELETE)
1121 		state->docache = 0;
1122 	state->rdonly = cnp->cn_flags & RDONLY;
1123 
1124 	/*
1125 	 * Keep going until we run out of path components.
1126 	 */
1127 	cnp->cn_nameptr = ndp->ni_pnbuf;
1128 
1129 	/* drop leading slashes (already used them to choose startdir) */
1130 	while (cnp->cn_nameptr[0] == '/') {
1131 		cnp->cn_nameptr++;
1132 		ndp->ni_pathlen--;
1133 	}
1134 	/* was it just "/"? */
1135 	if (cnp->cn_nameptr[0] == '\0') {
1136 		foundobj = searchdir;
1137 		searchdir = NULL;
1138 		cnp->cn_flags |= ISLASTCN;
1139 
1140 		/* bleh */
1141 		goto skiploop;
1142 	}
1143 
1144 	for (;;) {
1145 
1146 		/*
1147 		 * If the directory we're on is unmounted, bail out.
1148 		 * XXX: should this also check if it's unlinked?
1149 		 * XXX: yes it should... but how?
1150 		 */
1151 		if (searchdir->v_mount == NULL) {
1152 			vput(searchdir);
1153 			ndp->ni_dvp = NULL;
1154 			ndp->ni_vp = NULL;
1155 			return (ENOENT);
1156 		}
1157 
1158 		/*
1159 		 * Look up the next path component.
1160 		 * (currently, this may consume more than one)
1161 		 */
1162 
1163 		/* There should be no slashes here. */
1164 		KASSERT(cnp->cn_nameptr[0] != '/');
1165 
1166 		/* and we shouldn't have looped around if we were done */
1167 		KASSERT(cnp->cn_nameptr[0] != '\0');
1168 
1169 		error = lookup_parsepath(state);
1170 		if (error) {
1171 			vput(searchdir);
1172 			ndp->ni_dvp = NULL;
1173 			ndp->ni_vp = NULL;
1174 			state->attempt_retry = 1;
1175 			return (error);
1176 		}
1177 
1178 		error = lookup_once(state, searchdir, &searchdir, &foundobj);
1179 		if (error) {
1180 			vput(searchdir);
1181 			ndp->ni_dvp = NULL;
1182 			ndp->ni_vp = NULL;
1183 			/*
1184 			 * Note that if we're doing TRYEMULROOT we can
1185 			 * retry with the normal root. Where this is
1186 			 * currently set matches previous practice,
1187 			 * but the previous practice didn't make much
1188 			 * sense and somebody should sit down and
1189 			 * figure out which cases should cause retry
1190 			 * and which shouldn't. XXX.
1191 			 */
1192 			state->attempt_retry = 1;
1193 			return (error);
1194 		}
1195 
1196 		if (foundobj == NULL) {
1197 			/*
1198 			 * Success with no object returned means we're
1199 			 * creating something and it isn't already
1200 			 * there. Break out of the main loop now so
1201 			 * the code below doesn't have to test for
1202 			 * foundobj == NULL.
1203 			 */
1204 			break;
1205 		}
1206 
1207 		/*
1208 		 * Check for symbolic link. If we've reached one,
1209 		 * follow it, unless we aren't supposed to. Back up
1210 		 * over any slashes that we skipped, as we will need
1211 		 * them again.
1212 		 */
1213 		if (namei_atsymlink(state, foundobj)) {
1214 			ndp->ni_pathlen += state->slashes;
1215 			ndp->ni_next -= state->slashes;
1216 			if (neverfollow) {
1217 				error = EINVAL;
1218 			} else {
1219 				/*
1220 				 * dholland 20110410: if we're at a
1221 				 * union mount it might make sense to
1222 				 * use the top of the union stack here
1223 				 * rather than the layer we found the
1224 				 * symlink in. (FUTURE)
1225 				 */
1226 				error = namei_follow(state, inhibitmagic,
1227 						     searchdir, foundobj,
1228 						     &searchdir);
1229 			}
1230 			if (error) {
1231 				KASSERT(searchdir != foundobj);
1232 				vput(searchdir);
1233 				vput(foundobj);
1234 				ndp->ni_dvp = NULL;
1235 				ndp->ni_vp = NULL;
1236 				return error;
1237 			}
1238 			/* namei_follow unlocks it (ugh) so rele, not put */
1239 			vrele(foundobj);
1240 			foundobj = NULL;
1241 
1242 			/*
1243 			 * If we followed a symlink to `/' and there
1244 			 * are no more components after the symlink,
1245 			 * we're done with the loop and what we found
1246 			 * is the searchdir.
1247 			 */
1248 			if (cnp->cn_nameptr[0] == '\0') {
1249 				foundobj = searchdir;
1250 				searchdir = NULL;
1251 				cnp->cn_flags |= ISLASTCN;
1252 				break;
1253 			}
1254 
1255 			continue;
1256 		}
1257 
1258 		/*
1259 		 * Not a symbolic link.
1260 		 *
1261 		 * Check for directory, if the component was
1262 		 * followed by a series of slashes.
1263 		 */
1264 		if ((foundobj->v_type != VDIR) &&
1265 		    (cnp->cn_flags & REQUIREDIR)) {
1266 			if (searchdir == foundobj) {
1267 				vrele(searchdir);
1268 			} else {
1269 				vput(searchdir);
1270 			}
1271 			vput(foundobj);
1272 			ndp->ni_dvp = NULL;
1273 			ndp->ni_vp = NULL;
1274 			state->attempt_retry = 1;
1275 			return ENOTDIR;
1276 		}
1277 
1278 		/*
1279 		 * Stop if we've reached the last component.
1280 		 */
1281 		if (cnp->cn_flags & ISLASTCN) {
1282 			break;
1283 		}
1284 
1285 		/*
1286 		 * Continue with the next component.
1287 		 */
1288 		cnp->cn_nameptr = ndp->ni_next;
1289 		if (searchdir == foundobj) {
1290 			vrele(searchdir);
1291 		} else {
1292 			vput(searchdir);
1293 		}
1294 		searchdir = foundobj;
1295 		foundobj = NULL;
1296 	}
1297 
1298  skiploop:
1299 
1300 	if (foundobj != NULL) {
1301 		if (foundobj == ndp->ni_erootdir) {
1302 			/*
1303 			 * We are about to return the emulation root.
1304 			 * This isn't a good idea because code might
1305 			 * repeatedly lookup ".." until the file
1306 			 * matches that returned for "/" and loop
1307 			 * forever.  So convert it to the real root.
1308 			 */
1309 			if (searchdir != NULL) {
1310 				if (searchdir == foundobj)
1311 					vrele(searchdir);
1312 				else
1313 					vput(searchdir);
1314 				searchdir = NULL;
1315 			}
1316 			vput(foundobj);
1317 			foundobj = ndp->ni_rootdir;
1318 			vref(foundobj);
1319 			vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
1320 		}
1321 
1322 		/*
1323 		 * If the caller requested the parent node (i.e. it's
1324 		 * a CREATE, DELETE, or RENAME), and we don't have one
1325 		 * (because this is the root directory, or we crossed
1326 		 * a mount point), then we must fail.
1327 		 */
1328 		if (cnp->cn_nameiop != LOOKUP &&
1329 		    (searchdir == NULL ||
1330 		     searchdir->v_mount != foundobj->v_mount)) {
1331 			if (searchdir) {
1332 				vput(searchdir);
1333 			}
1334 			vput(foundobj);
1335 			foundobj = NULL;
1336 			ndp->ni_dvp = NULL;
1337 			ndp->ni_vp = NULL;
1338 			state->attempt_retry = 1;
1339 
1340 			switch (cnp->cn_nameiop) {
1341 			    case CREATE:
1342 				return EEXIST;
1343 			    case DELETE:
1344 			    case RENAME:
1345 				return EBUSY;
1346 			    default:
1347 				break;
1348 			}
1349 			panic("Invalid nameiop\n");
1350 		}
1351 
1352 		/*
1353 		 * Disallow directory write attempts on read-only lookups.
1354 		 * Prefers EEXIST over EROFS for the CREATE case.
1355 		 */
1356 		if (state->rdonly &&
1357 		    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1358 			if (searchdir) {
1359 				if (foundobj != searchdir) {
1360 					vput(searchdir);
1361 				} else {
1362 					vrele(searchdir);
1363 				}
1364 				searchdir = NULL;
1365 			}
1366 			vput(foundobj);
1367 			foundobj = NULL;
1368 			ndp->ni_dvp = NULL;
1369 			ndp->ni_vp = NULL;
1370 			state->attempt_retry = 1;
1371 			return EROFS;
1372 		}
1373 		if ((cnp->cn_flags & LOCKLEAF) == 0) {
1374 			/*
1375 			 * Note: if LOCKPARENT but not LOCKLEAF is
1376 			 * set, and searchdir == foundobj, this code
1377 			 * necessarily unlocks the parent as well as
1378 			 * the leaf. That is, just because you specify
1379 			 * LOCKPARENT doesn't mean you necessarily get
1380 			 * a locked parent vnode. The code in
1381 			 * vfs_syscalls.c, and possibly elsewhere,
1382 			 * that uses this combination "knows" this, so
1383 			 * it can't be safely changed. Feh. XXX
1384 			 */
1385 			VOP_UNLOCK(foundobj);
1386 		}
1387 	}
1388 
1389 	/*
1390 	 * Done.
1391 	 */
1392 
1393 	/*
1394 	 * If LOCKPARENT is not set, the parent directory isn't returned.
1395 	 */
1396 	if ((cnp->cn_flags & LOCKPARENT) == 0 && searchdir != NULL) {
1397 		if (searchdir == foundobj) {
1398 			vrele(searchdir);
1399 		} else {
1400 			vput(searchdir);
1401 		}
1402 		searchdir = NULL;
1403 	}
1404 
1405 	ndp->ni_dvp = searchdir;
1406 	ndp->ni_vp = foundobj;
1407 	return 0;
1408 }
1409 
1410 /*
1411  * Do namei; wrapper layer that handles TRYEMULROOT.
1412  */
1413 static int
1414 namei_tryemulroot(struct namei_state *state, struct vnode *forcecwd,
1415 	 int neverfollow, int inhibitmagic)
1416 {
1417 	int error;
1418 
1419 	struct nameidata *ndp = state->ndp;
1420 	struct componentname *cnp = state->cnp;
1421 	const char *savepath = NULL;
1422 
1423 	KASSERT(cnp == &ndp->ni_cnd);
1424 
1425 	if (cnp->cn_flags & TRYEMULROOT) {
1426 		savepath = pathbuf_stringcopy_get(ndp->ni_pathbuf);
1427 	}
1428 
1429     emul_retry:
1430 	state->attempt_retry = 0;
1431 
1432 	error = namei_oneroot(state, forcecwd, neverfollow, inhibitmagic);
1433 	if (error) {
1434 		/*
1435 		 * Once namei has started up, the existence of ni_erootdir
1436 		 * tells us whether we're working from an emulation root.
1437 		 * The TRYEMULROOT flag isn't necessarily authoritative.
1438 		 */
1439 		if (ndp->ni_erootdir != NULL && state->attempt_retry) {
1440 			/* Retry the whole thing using the normal root */
1441 			cnp->cn_flags &= ~TRYEMULROOT;
1442 			state->attempt_retry = 0;
1443 
1444 			/* kinda gross */
1445 			strcpy(ndp->ni_pathbuf->pb_path, savepath);
1446 			pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
1447 			savepath = NULL;
1448 
1449 			goto emul_retry;
1450 		}
1451 	}
1452 	if (savepath != NULL) {
1453 		pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
1454 	}
1455 	return error;
1456 }
1457 
1458 /*
1459  * External interface.
1460  */
1461 int
1462 namei(struct nameidata *ndp)
1463 {
1464 	struct namei_state state;
1465 	int error;
1466 
1467 	namei_init(&state, ndp);
1468 	error = namei_tryemulroot(&state, NULL,
1469 				  0/*!neverfollow*/, 0/*!inhibitmagic*/);
1470 	namei_cleanup(&state);
1471 
1472 	if (error) {
1473 		/* make sure no stray refs leak out */
1474 		KASSERT(ndp->ni_dvp == NULL);
1475 		KASSERT(ndp->ni_vp == NULL);
1476 	}
1477 
1478 	return error;
1479 }
1480 
1481 ////////////////////////////////////////////////////////////
1482 
1483 /*
1484  * External interface used by nfsd. This is basically different from
1485  * namei only in that it has the ability to pass in the "current
1486  * directory", and uses an extra flag "neverfollow" for which there's
1487  * no physical flag defined in namei.h. (There used to be a cut&paste
1488  * copy of about half of namei in nfsd to allow these minor
1489  * adjustments to exist.)
1490  *
1491  * XXX: the namei interface should be adjusted so nfsd can just use
1492  * ordinary namei().
1493  */
1494 int
1495 lookup_for_nfsd(struct nameidata *ndp, struct vnode *forcecwd, int neverfollow)
1496 {
1497 	struct namei_state state;
1498 	int error;
1499 
1500 	namei_init(&state, ndp);
1501 	error = namei_tryemulroot(&state, forcecwd,
1502 				  neverfollow, 1/*inhibitmagic*/);
1503 	namei_cleanup(&state);
1504 
1505 	if (error) {
1506 		/* make sure no stray refs leak out */
1507 		KASSERT(ndp->ni_dvp == NULL);
1508 		KASSERT(ndp->ni_vp == NULL);
1509 	}
1510 
1511 	return error;
1512 }
1513 
1514 /*
1515  * A second external interface used by nfsd. This turns out to be a
1516  * single lookup used by the WebNFS code (ha!) to get "index.html" or
1517  * equivalent when asked for a directory. It should eventually evolve
1518  * into some kind of namei_once() call; for the time being it's kind
1519  * of a mess. XXX.
1520  *
1521  * dholland 20110109: I don't think it works, and I don't think it
1522  * worked before I started hacking and slashing either, and I doubt
1523  * anyone will ever notice.
1524  */
1525 
1526 /*
1527  * Internals. This calls lookup_once() after setting up the assorted
1528  * pieces of state the way they ought to be.
1529  */
1530 static int
1531 do_lookup_for_nfsd_index(struct namei_state *state, struct vnode *startdir)
1532 {
1533 	int error = 0;
1534 
1535 	struct componentname *cnp = state->cnp;
1536 	struct nameidata *ndp = state->ndp;
1537 	struct vnode *foundobj;
1538 	const char *cp;			/* pointer into pathname argument */
1539 
1540 	KASSERT(cnp == &ndp->ni_cnd);
1541 
1542 	cnp->cn_nameptr = ndp->ni_pnbuf;
1543 	state->docache = 1;
1544 	state->rdonly = cnp->cn_flags & RDONLY;
1545 	ndp->ni_dvp = NULL;
1546 
1547 	cnp->cn_consume = 0;
1548 	cp = NULL;
1549 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
1550 	cnp->cn_namelen = cp - cnp->cn_nameptr;
1551 	KASSERT(cnp->cn_namelen <= KERNEL_NAME_MAX);
1552 	ndp->ni_pathlen -= cnp->cn_namelen;
1553 	ndp->ni_next = cp;
1554 	state->slashes = 0;
1555 	cnp->cn_flags &= ~REQUIREDIR;
1556 	cnp->cn_flags |= MAKEENTRY|ISLASTCN;
1557 
1558 	if (cnp->cn_namelen == 2 &&
1559 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
1560 		cnp->cn_flags |= ISDOTDOT;
1561 	else
1562 		cnp->cn_flags &= ~ISDOTDOT;
1563 
1564 	/*
1565 	 * Because lookup_once can change the startdir, we need our
1566 	 * own reference to it to avoid consuming the caller's.
1567 	 */
1568 	vref(startdir);
1569 	vn_lock(startdir, LK_EXCLUSIVE | LK_RETRY);
1570 	error = lookup_once(state, startdir, &startdir, &foundobj);
1571 	if (error == 0 && startdir == foundobj) {
1572 		vrele(startdir);
1573 	} else {
1574 		vput(startdir);
1575 	}
1576 	if (error) {
1577 		goto bad;
1578 	}
1579 	ndp->ni_vp = foundobj;
1580 
1581 	if (foundobj == NULL) {
1582 		return 0;
1583 	}
1584 
1585 	KASSERT((cnp->cn_flags & LOCKPARENT) == 0);
1586 	if ((cnp->cn_flags & LOCKLEAF) == 0) {
1587 		VOP_UNLOCK(foundobj);
1588 	}
1589 	return (0);
1590 
1591 bad:
1592 	ndp->ni_vp = NULL;
1593 	return (error);
1594 }
1595 
1596 /*
1597  * External interface. The partitioning between this function and the
1598  * above isn't very clear - the above function exists mostly so code
1599  * that uses "state->" can be shuffled around without having to change
1600  * it to "state.".
1601  */
1602 int
1603 lookup_for_nfsd_index(struct nameidata *ndp, struct vnode *startdir)
1604 {
1605 	struct namei_state state;
1606 	int error;
1607 
1608 	/*
1609 	 * Note: the name sent in here (is not|should not be) allowed
1610 	 * to contain a slash.
1611 	 */
1612 	if (strlen(ndp->ni_pathbuf->pb_path) > KERNEL_NAME_MAX) {
1613 		return ENAMETOOLONG;
1614 	}
1615 	if (strchr(ndp->ni_pathbuf->pb_path, '/')) {
1616 		return EINVAL;
1617 	}
1618 
1619 	ndp->ni_pathlen = strlen(ndp->ni_pathbuf->pb_path) + 1;
1620 	ndp->ni_pnbuf = NULL;
1621 	ndp->ni_cnd.cn_nameptr = NULL;
1622 
1623 	namei_init(&state, ndp);
1624 	error = do_lookup_for_nfsd_index(&state, startdir);
1625 	namei_cleanup(&state);
1626 
1627 	return error;
1628 }
1629 
1630 ////////////////////////////////////////////////////////////
1631 
1632 /*
1633  * Reacquire a path name component.
1634  * dvp is locked on entry and exit.
1635  * *vpp is locked on exit unless it's NULL.
1636  */
1637 int
1638 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, int dummy)
1639 {
1640 	int rdonly;			/* lookup read-only flag bit */
1641 	int error = 0;
1642 #ifdef DEBUG
1643 	uint32_t newhash;		/* DEBUG: check name hash */
1644 	const char *cp;			/* DEBUG: check name ptr/len */
1645 #endif /* DEBUG */
1646 
1647 	(void)dummy;
1648 
1649 	/*
1650 	 * Setup: break out flag bits into variables.
1651 	 */
1652 	rdonly = cnp->cn_flags & RDONLY;
1653 
1654 	/*
1655 	 * Search a new directory.
1656 	 *
1657 	 * The cn_hash value is for use by vfs_cache.
1658 	 * The last component of the filename is left accessible via
1659 	 * cnp->cn_nameptr for callers that need the name. Callers needing
1660 	 * the name set the SAVENAME flag. When done, they assume
1661 	 * responsibility for freeing the pathname buffer.
1662 	 */
1663 #ifdef DEBUG
1664 	cp = NULL;
1665 	newhash = namei_hash(cnp->cn_nameptr, &cp);
1666 	if ((uint32_t)newhash != (uint32_t)cnp->cn_hash)
1667 		panic("relookup: bad hash");
1668 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
1669 		panic("relookup: bad len");
1670 	while (*cp == '/')
1671 		cp++;
1672 	if (*cp != 0)
1673 		panic("relookup: not last component");
1674 #endif /* DEBUG */
1675 
1676 	/*
1677 	 * Check for degenerate name (e.g. / or "")
1678 	 * which is a way of talking about a directory,
1679 	 * e.g. like "/." or ".".
1680 	 */
1681 	if (cnp->cn_nameptr[0] == '\0')
1682 		panic("relookup: null name");
1683 
1684 	if (cnp->cn_flags & ISDOTDOT)
1685 		panic("relookup: lookup on dot-dot");
1686 
1687 	/*
1688 	 * We now have a segment name to search for, and a directory to search.
1689 	 */
1690 	cnp->cn_flags |= INRELOOKUP;
1691 	error = VOP_LOOKUP(dvp, vpp, cnp);
1692 	cnp->cn_flags &= ~INRELOOKUP;
1693 	if ((error) != 0) {
1694 #ifdef DIAGNOSTIC
1695 		if (*vpp != NULL)
1696 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
1697 #endif
1698 		if (error != EJUSTRETURN)
1699 			goto bad;
1700 	}
1701 
1702 #ifdef DIAGNOSTIC
1703 	/*
1704 	 * Check for symbolic link
1705 	 */
1706 	if (*vpp && (*vpp)->v_type == VLNK && (cnp->cn_flags & FOLLOW))
1707 		panic("relookup: symlink found");
1708 #endif
1709 
1710 	/*
1711 	 * Check for read-only lookups.
1712 	 */
1713 	if (rdonly && cnp->cn_nameiop != LOOKUP) {
1714 		error = EROFS;
1715 		if (*vpp) {
1716 			vput(*vpp);
1717 		}
1718 		goto bad;
1719 	}
1720 	return (0);
1721 
1722 bad:
1723 	*vpp = NULL;
1724 	return (error);
1725 }
1726 
1727 /*
1728  * namei_simple - simple forms of namei.
1729  *
1730  * These are wrappers to allow the simple case callers of namei to be
1731  * left alone while everything else changes under them.
1732  */
1733 
1734 /* Flags */
1735 struct namei_simple_flags_type {
1736 	int dummy;
1737 };
1738 static const struct namei_simple_flags_type ns_nn, ns_nt, ns_fn, ns_ft;
1739 const namei_simple_flags_t NSM_NOFOLLOW_NOEMULROOT = &ns_nn;
1740 const namei_simple_flags_t NSM_NOFOLLOW_TRYEMULROOT = &ns_nt;
1741 const namei_simple_flags_t NSM_FOLLOW_NOEMULROOT = &ns_fn;
1742 const namei_simple_flags_t NSM_FOLLOW_TRYEMULROOT = &ns_ft;
1743 
1744 static
1745 int
1746 namei_simple_convert_flags(namei_simple_flags_t sflags)
1747 {
1748 	if (sflags == NSM_NOFOLLOW_NOEMULROOT)
1749 		return NOFOLLOW | 0;
1750 	if (sflags == NSM_NOFOLLOW_TRYEMULROOT)
1751 		return NOFOLLOW | TRYEMULROOT;
1752 	if (sflags == NSM_FOLLOW_NOEMULROOT)
1753 		return FOLLOW | 0;
1754 	if (sflags == NSM_FOLLOW_TRYEMULROOT)
1755 		return FOLLOW | TRYEMULROOT;
1756 	panic("namei_simple_convert_flags: bogus sflags\n");
1757 	return 0;
1758 }
1759 
1760 int
1761 namei_simple_kernel(const char *path, namei_simple_flags_t sflags,
1762 			struct vnode **vp_ret)
1763 {
1764 	struct nameidata nd;
1765 	struct pathbuf *pb;
1766 	int err;
1767 
1768 	pb = pathbuf_create(path);
1769 	if (pb == NULL) {
1770 		return ENOMEM;
1771 	}
1772 
1773 	NDINIT(&nd,
1774 		LOOKUP,
1775 		namei_simple_convert_flags(sflags),
1776 		pb);
1777 	err = namei(&nd);
1778 	if (err != 0) {
1779 		pathbuf_destroy(pb);
1780 		return err;
1781 	}
1782 	*vp_ret = nd.ni_vp;
1783 	pathbuf_destroy(pb);
1784 	return 0;
1785 }
1786 
1787 int
1788 namei_simple_user(const char *path, namei_simple_flags_t sflags,
1789 			struct vnode **vp_ret)
1790 {
1791 	struct pathbuf *pb;
1792 	struct nameidata nd;
1793 	int err;
1794 
1795 	err = pathbuf_copyin(path, &pb);
1796 	if (err) {
1797 		return err;
1798 	}
1799 
1800 	NDINIT(&nd,
1801 		LOOKUP,
1802 		namei_simple_convert_flags(sflags),
1803 		pb);
1804 	err = namei(&nd);
1805 	if (err != 0) {
1806 		pathbuf_destroy(pb);
1807 		return err;
1808 	}
1809 	*vp_ret = nd.ni_vp;
1810 	pathbuf_destroy(pb);
1811 	return 0;
1812 }
1813