xref: /netbsd-src/sys/kern/vfs_lookup.c (revision f32144615845cc98528c78d4f24930c3b8fe6a8f)
1 /*	$NetBSD: vfs_lookup.c,v 1.175 2011/04/14 15:29:25 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vfs_lookup.c	8.10 (Berkeley) 5/27/95
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: vfs_lookup.c,v 1.175 2011/04/14 15:29:25 yamt 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 /*
67  * Substitute replacement text for 'magic' strings in symlinks.
68  * Returns 0 if successful, and returns non-zero if an error
69  * occurs.  (Currently, the only possible error is running out
70  * of temporary pathname space.)
71  *
72  * Looks for "@<string>" and "@<string>/", where <string> is a
73  * recognized 'magic' string.  Replaces the "@<string>" with the
74  * appropriate replacement text.  (Note that in some cases the
75  * replacement text may have zero length.)
76  *
77  * This would have been table driven, but the variance in
78  * replacement strings (and replacement string lengths) made
79  * that impractical.
80  */
81 #define	VNL(x)							\
82 	(sizeof(x) - 1)
83 
84 #define	VO	'{'
85 #define	VC	'}'
86 
87 #define	MATCH(str)						\
88 	((termchar == '/' && i + VNL(str) == *len) ||		\
89 	 (i + VNL(str) < *len &&				\
90 	  cp[i + VNL(str)] == termchar)) &&			\
91 	!strncmp((str), &cp[i], VNL(str))
92 
93 #define	SUBSTITUTE(m, s, sl)					\
94 	if ((newlen + (sl)) >= MAXPATHLEN)			\
95 		return 1;					\
96 	i += VNL(m);						\
97 	if (termchar != '/')					\
98 		i++;						\
99 	(void)memcpy(&tmp[newlen], (s), (sl));			\
100 	newlen += (sl);						\
101 	change = 1;						\
102 	termchar = '/';
103 
104 static int
105 symlink_magic(struct proc *p, char *cp, size_t *len)
106 {
107 	char *tmp;
108 	size_t change, i, newlen, slen;
109 	char termchar = '/';
110 	char idtmp[11]; /* enough for 32 bit *unsigned* integer */
111 
112 
113 	tmp = PNBUF_GET();
114 	for (change = i = newlen = 0; i < *len; ) {
115 		if (cp[i] != '@') {
116 			tmp[newlen++] = cp[i++];
117 			continue;
118 		}
119 
120 		i++;
121 
122 		/* Check for @{var} syntax. */
123 		if (cp[i] == VO) {
124 			termchar = VC;
125 			i++;
126 		}
127 
128 		/*
129 		 * The following checks should be ordered according
130 		 * to frequency of use.
131 		 */
132 		if (MATCH("machine_arch")) {
133 			slen = VNL(MACHINE_ARCH);
134 			SUBSTITUTE("machine_arch", MACHINE_ARCH, slen);
135 		} else if (MATCH("machine")) {
136 			slen = VNL(MACHINE);
137 			SUBSTITUTE("machine", MACHINE, slen);
138 		} else if (MATCH("hostname")) {
139 			SUBSTITUTE("hostname", hostname, hostnamelen);
140 		} else if (MATCH("osrelease")) {
141 			slen = strlen(osrelease);
142 			SUBSTITUTE("osrelease", osrelease, slen);
143 		} else if (MATCH("emul")) {
144 			slen = strlen(p->p_emul->e_name);
145 			SUBSTITUTE("emul", p->p_emul->e_name, slen);
146 		} else if (MATCH("kernel_ident")) {
147 			slen = strlen(kernel_ident);
148 			SUBSTITUTE("kernel_ident", kernel_ident, slen);
149 		} else if (MATCH("domainname")) {
150 			SUBSTITUTE("domainname", domainname, domainnamelen);
151 		} else if (MATCH("ostype")) {
152 			slen = strlen(ostype);
153 			SUBSTITUTE("ostype", ostype, slen);
154 		} else if (MATCH("uid")) {
155 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
156 			    kauth_cred_geteuid(kauth_cred_get()));
157 			SUBSTITUTE("uid", idtmp, slen);
158 		} else if (MATCH("ruid")) {
159 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
160 			    kauth_cred_getuid(kauth_cred_get()));
161 			SUBSTITUTE("ruid", idtmp, slen);
162 		} else if (MATCH("gid")) {
163 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
164 			    kauth_cred_getegid(kauth_cred_get()));
165 			SUBSTITUTE("gid", idtmp, slen);
166 		} else if (MATCH("rgid")) {
167 			slen = snprintf(idtmp, sizeof(idtmp), "%u",
168 			    kauth_cred_getgid(kauth_cred_get()));
169 			SUBSTITUTE("rgid", idtmp, slen);
170 		} else {
171 			tmp[newlen++] = '@';
172 			if (termchar == VC)
173 				tmp[newlen++] = VO;
174 		}
175 	}
176 
177 	if (change) {
178 		(void)memcpy(cp, tmp, newlen);
179 		*len = newlen;
180 	}
181 	PNBUF_PUT(tmp);
182 
183 	return 0;
184 }
185 
186 #undef VNL
187 #undef VO
188 #undef VC
189 #undef MATCH
190 #undef SUBSTITUTE
191 
192 ////////////////////////////////////////////////////////////
193 
194 /*
195  * Determine the namei hash (for cn_hash) for name.
196  * If *ep != NULL, hash from name to ep-1.
197  * If *ep == NULL, hash from name until the first NUL or '/', and
198  * return the location of this termination character in *ep.
199  *
200  * This function returns an equivalent hash to the MI hash32_strn().
201  * The latter isn't used because in the *ep == NULL case, determining
202  * the length of the string to the first NUL or `/' and then calling
203  * hash32_strn() involves unnecessary double-handling of the data.
204  */
205 uint32_t
206 namei_hash(const char *name, const char **ep)
207 {
208 	uint32_t	hash;
209 
210 	hash = HASH32_STR_INIT;
211 	if (*ep != NULL) {
212 		for (; name < *ep; name++)
213 			hash = hash * 33 + *(const uint8_t *)name;
214 	} else {
215 		for (; *name != '\0' && *name != '/'; name++)
216 			hash = hash * 33 + *(const uint8_t *)name;
217 		*ep = name;
218 	}
219 	return (hash + (hash >> 5));
220 }
221 
222 ////////////////////////////////////////////////////////////
223 
224 /*
225  * Sealed abstraction for pathnames.
226  *
227  * System-call-layer level code that is going to call namei should
228  * first create a pathbuf and adjust all the bells and whistles on it
229  * as needed by context
230  */
231 
232 struct pathbuf {
233 	char *pb_path;
234 	char *pb_pathcopy;
235 	unsigned pb_pathcopyuses;
236 };
237 
238 static struct pathbuf *
239 pathbuf_create_raw(void)
240 {
241 	struct pathbuf *pb;
242 
243 	pb = kmem_alloc(sizeof(*pb), KM_SLEEP);
244 	if (pb == NULL) {
245 		return NULL;
246 	}
247 	pb->pb_path = PNBUF_GET();
248 	if (pb->pb_path == NULL) {
249 		kmem_free(pb, sizeof(*pb));
250 		return NULL;
251 	}
252 	pb->pb_pathcopy = NULL;
253 	pb->pb_pathcopyuses = 0;
254 	return pb;
255 }
256 
257 void
258 pathbuf_destroy(struct pathbuf *pb)
259 {
260 	KASSERT(pb->pb_pathcopyuses == 0);
261 	KASSERT(pb->pb_pathcopy == NULL);
262 	PNBUF_PUT(pb->pb_path);
263 	kmem_free(pb, sizeof(*pb));
264 }
265 
266 struct pathbuf *
267 pathbuf_assimilate(char *pnbuf)
268 {
269 	struct pathbuf *pb;
270 
271 	pb = kmem_alloc(sizeof(*pb), KM_SLEEP);
272 	if (pb == NULL) {
273 		return NULL;
274 	}
275 	pb->pb_path = pnbuf;
276 	pb->pb_pathcopy = NULL;
277 	pb->pb_pathcopyuses = 0;
278 	return pb;
279 }
280 
281 struct pathbuf *
282 pathbuf_create(const char *path)
283 {
284 	struct pathbuf *pb;
285 	int error;
286 
287 	pb = pathbuf_create_raw();
288 	if (pb == NULL) {
289 		return NULL;
290 	}
291 	error = copystr(path, pb->pb_path, PATH_MAX, NULL);
292 	if (error != 0) {
293 		KASSERT(!"kernel path too long in pathbuf_create");
294 		/* make sure it's null-terminated, just in case */
295 		pb->pb_path[PATH_MAX-1] = '\0';
296 	}
297 	return pb;
298 }
299 
300 int
301 pathbuf_copyin(const char *userpath, struct pathbuf **ret)
302 {
303 	struct pathbuf *pb;
304 	int error;
305 
306 	pb = pathbuf_create_raw();
307 	if (pb == NULL) {
308 		return ENOMEM;
309 	}
310 	error = copyinstr(userpath, pb->pb_path, PATH_MAX, NULL);
311 	if (error) {
312 		pathbuf_destroy(pb);
313 		return error;
314 	}
315 	*ret = pb;
316 	return 0;
317 }
318 
319 /*
320  * XXX should not exist:
321  *   1. whether a pointer is kernel or user should be statically checkable
322  *   2. copyin should be handled by the upper part of the syscall layer,
323  *      not in here.
324  */
325 int
326 pathbuf_maybe_copyin(const char *path, enum uio_seg seg, struct pathbuf **ret)
327 {
328 	if (seg == UIO_USERSPACE) {
329 		return pathbuf_copyin(path, ret);
330 	} else {
331 		*ret = pathbuf_create(path);
332 		if (*ret == NULL) {
333 			return ENOMEM;
334 		}
335 		return 0;
336 	}
337 }
338 
339 /*
340  * Get a copy of the path buffer as it currently exists. If this is
341  * called after namei starts the results may be arbitrary.
342  */
343 void
344 pathbuf_copystring(const struct pathbuf *pb, char *buf, size_t maxlen)
345 {
346 	strlcpy(buf, pb->pb_path, maxlen);
347 }
348 
349 /*
350  * These two functions allow access to a saved copy of the original
351  * path string. The first copy should be gotten before namei is
352  * called. Each copy that is gotten should be put back.
353  */
354 
355 const char *
356 pathbuf_stringcopy_get(struct pathbuf *pb)
357 {
358 	if (pb->pb_pathcopyuses == 0) {
359 		pb->pb_pathcopy = PNBUF_GET();
360 		strcpy(pb->pb_pathcopy, pb->pb_path);
361 	}
362 	pb->pb_pathcopyuses++;
363 	return pb->pb_pathcopy;
364 }
365 
366 void
367 pathbuf_stringcopy_put(struct pathbuf *pb, const char *str)
368 {
369 	KASSERT(str == pb->pb_pathcopy);
370 	KASSERT(pb->pb_pathcopyuses > 0);
371 	pb->pb_pathcopyuses--;
372 	if (pb->pb_pathcopyuses == 0) {
373 		PNBUF_PUT(pb->pb_pathcopy);
374 		pb->pb_pathcopy = NULL;
375 	}
376 }
377 
378 
379 ////////////////////////////////////////////////////////////
380 
381 /*
382  * namei: convert a pathname into a pointer to a (maybe-locked) vnode,
383  * and maybe also its parent directory vnode, and assorted other guff.
384  * See namei(9) for the interface documentation.
385  *
386  *
387  * The FOLLOW flag is set when symbolic links are to be followed
388  * when they occur at the end of the name translation process.
389  * Symbolic links are always followed for all other pathname
390  * components other than the last.
391  *
392  * The segflg defines whether the name is to be copied from user
393  * space or kernel space.
394  *
395  * Overall outline of namei:
396  *
397  *	copy in name
398  *	get starting directory
399  *	while (!done && !error) {
400  *		call lookup to search path.
401  *		if symbolic link, massage name in buffer and continue
402  *	}
403  */
404 
405 /*
406  * Search a pathname.
407  * This is a very central and rather complicated routine.
408  *
409  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
410  * The starting directory is passed in. The pathname is descended
411  * until done, or a symbolic link is encountered. The variable ni_more
412  * is clear if the path is completed; it is set to one if a symbolic
413  * link needing interpretation is encountered.
414  *
415  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
416  * whether the name is to be looked up, created, renamed, or deleted.
417  * When CREATE, RENAME, or DELETE is specified, information usable in
418  * creating, renaming, or deleting a directory entry may be calculated.
419  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
420  * locked.  Otherwise the parent directory is not returned. If the target
421  * of the pathname exists and LOCKLEAF is or'ed into the flag the target
422  * is returned locked, otherwise it is returned unlocked.  When creating
423  * or renaming and LOCKPARENT is specified, the target may not be ".".
424  * When deleting and LOCKPARENT is specified, the target may be ".".
425  *
426  * Overall outline of lookup:
427  *
428  * dirloop:
429  *	identify next component of name at ndp->ni_ptr
430  *	handle degenerate case where name is null string
431  *	if .. and crossing mount points and on mounted filesys, find parent
432  *	call VOP_LOOKUP routine for next component name
433  *	    directory vnode returned in ni_dvp, locked.
434  *	    component vnode returned in ni_vp (if it exists), locked.
435  *	if result vnode is mounted on and crossing mount points,
436  *	    find mounted on vnode
437  *	if more components of name, do next level at dirloop
438  *	return the answer in ni_vp, locked if LOCKLEAF set
439  *	    if LOCKPARENT set, return locked parent in ni_dvp
440  */
441 
442 
443 /*
444  * Internal state for a namei operation.
445  *
446  * cnp is always equal to &ndp->ni_cnp.
447  */
448 struct namei_state {
449 	struct nameidata *ndp;
450 	struct componentname *cnp;
451 
452 	int docache;			/* == 0 do not cache last component */
453 	int rdonly;			/* lookup read-only flag bit */
454 	int slashes;
455 
456 	unsigned attempt_retry:1;	/* true if error allows emul retry */
457 };
458 
459 
460 /*
461  * Initialize the namei working state.
462  */
463 static void
464 namei_init(struct namei_state *state, struct nameidata *ndp)
465 {
466 	state->ndp = ndp;
467 	state->cnp = &ndp->ni_cnd;
468 	KASSERT((state->cnp->cn_flags & INRELOOKUP) == 0);
469 
470 	state->docache = 0;
471 	state->rdonly = 0;
472 	state->slashes = 0;
473 
474 #ifdef DIAGNOSTIC
475 	if (!state->cnp->cn_cred)
476 		panic("namei: bad cred/proc");
477 	if (state->cnp->cn_nameiop & (~OPMASK))
478 		panic("namei: nameiop contaminated with flags");
479 	if (state->cnp->cn_flags & OPMASK)
480 		panic("namei: flags contaminated with nameiops");
481 #endif
482 
483 	/*
484 	 * The buffer for name translation shall be the one inside the
485 	 * pathbuf.
486 	 */
487 	state->ndp->ni_pnbuf = state->ndp->ni_pathbuf->pb_path;
488 }
489 
490 /*
491  * Clean up the working namei state, leaving things ready for return
492  * from namei.
493  */
494 static void
495 namei_cleanup(struct namei_state *state)
496 {
497 	KASSERT(state->cnp == &state->ndp->ni_cnd);
498 
499 	/* nothing for now */
500 	(void)state;
501 }
502 
503 //////////////////////////////
504 
505 /*
506  * Get the directory context.
507  * Initializes the rootdir and erootdir state and returns a reference
508  * to the starting dir.
509  */
510 static struct vnode *
511 namei_getstartdir(struct namei_state *state)
512 {
513 	struct nameidata *ndp = state->ndp;
514 	struct componentname *cnp = state->cnp;
515 	struct cwdinfo *cwdi;		/* pointer to cwd state */
516 	struct lwp *self = curlwp;	/* thread doing namei() */
517 	struct vnode *rootdir, *erootdir, *curdir, *startdir;
518 
519 	cwdi = self->l_proc->p_cwdi;
520 	rw_enter(&cwdi->cwdi_lock, RW_READER);
521 
522 	/* root dir */
523 	if (cwdi->cwdi_rdir == NULL || (cnp->cn_flags & NOCHROOT)) {
524 		rootdir = rootvnode;
525 	} else {
526 		rootdir = cwdi->cwdi_rdir;
527 	}
528 
529 	/* emulation root dir, if any */
530 	if ((cnp->cn_flags & TRYEMULROOT) == 0) {
531 		/* if we don't want it, don't fetch it */
532 		erootdir = NULL;
533 	} else if (cnp->cn_flags & EMULROOTSET) {
534 		/* explicitly set emulroot; "/../" doesn't override this */
535 		erootdir = ndp->ni_erootdir;
536 	} else if (!strncmp(ndp->ni_pnbuf, "/../", 4)) {
537 		/* explicit reference to real rootdir */
538 		erootdir = NULL;
539 	} else {
540 		/* may be null */
541 		erootdir = cwdi->cwdi_edir;
542 	}
543 
544 	/* current dir */
545 	curdir = cwdi->cwdi_cdir;
546 
547 	if (ndp->ni_pnbuf[0] != '/') {
548 		startdir = curdir;
549 		erootdir = NULL;
550 	} else if (cnp->cn_flags & TRYEMULROOT && erootdir != NULL) {
551 		startdir = erootdir;
552 	} else {
553 		startdir = rootdir;
554 		erootdir = NULL;
555 	}
556 
557 	state->ndp->ni_rootdir = rootdir;
558 	state->ndp->ni_erootdir = erootdir;
559 
560 	/*
561 	 * Get a reference to the start dir so we can safely unlock cwdi.
562 	 *
563 	 * XXX: should we hold references to rootdir and erootdir while
564 	 * we're running? What happens if a multithreaded process chroots
565 	 * during namei?
566 	 */
567 	vref(startdir);
568 
569 	rw_exit(&cwdi->cwdi_lock);
570 	return startdir;
571 }
572 
573 /*
574  * Get the directory context for the nfsd case, in parallel to
575  * getstartdir. Initializes the rootdir and erootdir state and
576  * returns a reference to the passed-in starting dir.
577  */
578 static struct vnode *
579 namei_getstartdir_for_nfsd(struct namei_state *state, struct vnode *startdir)
580 {
581 	/* always use the real root, and never set an emulation root */
582 	state->ndp->ni_rootdir = rootvnode;
583 	state->ndp->ni_erootdir = NULL;
584 
585 	vref(startdir);
586 	return startdir;
587 }
588 
589 
590 /*
591  * Ktrace the namei operation.
592  */
593 static void
594 namei_ktrace(struct namei_state *state)
595 {
596 	struct nameidata *ndp = state->ndp;
597 	struct componentname *cnp = state->cnp;
598 	struct lwp *self = curlwp;	/* thread doing namei() */
599 	const char *emul_path;
600 
601 	if (ktrpoint(KTR_NAMEI)) {
602 		if (ndp->ni_erootdir != NULL) {
603 			/*
604 			 * To make any sense, the trace entry need to have the
605 			 * text of the emulation path prepended.
606 			 * Usually we can get this from the current process,
607 			 * but when called from emul_find_interp() it is only
608 			 * in the exec_package - so we get it passed in ni_next
609 			 * (this is a hack).
610 			 */
611 			if (cnp->cn_flags & EMULROOTSET)
612 				emul_path = ndp->ni_next;
613 			else
614 				emul_path = self->l_proc->p_emul->e_path;
615 			ktrnamei2(emul_path, strlen(emul_path),
616 			    ndp->ni_pnbuf, ndp->ni_pathlen);
617 		} else
618 			ktrnamei(ndp->ni_pnbuf, ndp->ni_pathlen);
619 	}
620 }
621 
622 /*
623  * Start up namei. Find the root dir and cwd, establish the starting
624  * directory for lookup, and lock it. Also calls ktrace when
625  * appropriate.
626  */
627 static int
628 namei_start(struct namei_state *state, struct vnode *forcecwd,
629 	    struct vnode **startdir_ret)
630 {
631 	struct nameidata *ndp = state->ndp;
632 	struct vnode *startdir;
633 
634 	/* length includes null terminator (was originally from copyinstr) */
635 	ndp->ni_pathlen = strlen(ndp->ni_pnbuf) + 1;
636 
637 	/*
638 	 * POSIX.1 requirement: "" is not a valid file name.
639 	 */
640 	if (ndp->ni_pathlen == 1) {
641 		return ENOENT;
642 	}
643 
644 	ndp->ni_loopcnt = 0;
645 
646 	/* Get starting directory, set up root, and ktrace. */
647 	if (forcecwd != NULL) {
648 		startdir = namei_getstartdir_for_nfsd(state, forcecwd);
649 		/* no ktrace */
650 	} else {
651 		startdir = namei_getstartdir(state);
652 		namei_ktrace(state);
653 	}
654 
655 	vn_lock(startdir, LK_EXCLUSIVE | LK_RETRY);
656 
657 	*startdir_ret = startdir;
658 	return 0;
659 }
660 
661 /*
662  * Check for being at a symlink that we're going to follow.
663  */
664 static inline int
665 namei_atsymlink(struct namei_state *state, struct vnode *foundobj)
666 {
667 	return (foundobj->v_type == VLNK) &&
668 		(state->cnp->cn_flags & (FOLLOW|REQUIREDIR));
669 }
670 
671 /*
672  * Follow a symlink.
673  *
674  * Updates searchdir. inhibitmagic causes magic symlinks to not be
675  * interpreted; this is used by nfsd.
676  *
677  * Unlocks foundobj on success (ugh)
678  */
679 static inline int
680 namei_follow(struct namei_state *state, int inhibitmagic,
681 	     struct vnode *searchdir, struct vnode *foundobj,
682 	     struct vnode **newsearchdir_ret)
683 {
684 	struct nameidata *ndp = state->ndp;
685 	struct componentname *cnp = state->cnp;
686 
687 	struct lwp *self = curlwp;	/* thread doing namei() */
688 	struct iovec aiov;		/* uio for reading symbolic links */
689 	struct uio auio;
690 	char *cp;			/* pointer into pathname argument */
691 	size_t linklen;
692 	int error;
693 
694 	KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
695 	KASSERT(VOP_ISLOCKED(foundobj) == LK_EXCLUSIVE);
696 	if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
697 		return ELOOP;
698 	}
699 	if (foundobj->v_mount->mnt_flag & MNT_SYMPERM) {
700 		error = VOP_ACCESS(foundobj, VEXEC, cnp->cn_cred);
701 		if (error != 0)
702 			return error;
703 	}
704 
705 	/* FUTURE: fix this to not use a second buffer */
706 	cp = PNBUF_GET();
707 	aiov.iov_base = cp;
708 	aiov.iov_len = MAXPATHLEN;
709 	auio.uio_iov = &aiov;
710 	auio.uio_iovcnt = 1;
711 	auio.uio_offset = 0;
712 	auio.uio_rw = UIO_READ;
713 	auio.uio_resid = MAXPATHLEN;
714 	UIO_SETUP_SYSSPACE(&auio);
715 	error = VOP_READLINK(foundobj, &auio, cnp->cn_cred);
716 	if (error) {
717 		PNBUF_PUT(cp);
718 		return error;
719 	}
720 	linklen = MAXPATHLEN - auio.uio_resid;
721 	if (linklen == 0) {
722 		PNBUF_PUT(cp);
723 		return ENOENT;
724 	}
725 
726 	/*
727 	 * Do symlink substitution, if appropriate, and
728 	 * check length for potential overflow.
729 	 *
730 	 * Inhibit symlink substitution for nfsd.
731 	 * XXX: This is how it was before; is that a bug or a feature?
732 	 */
733 	if ((!inhibitmagic && vfs_magiclinks &&
734 	     symlink_magic(self->l_proc, cp, &linklen)) ||
735 	    (linklen + ndp->ni_pathlen >= MAXPATHLEN)) {
736 		PNBUF_PUT(cp);
737 		return ENAMETOOLONG;
738 	}
739 	if (ndp->ni_pathlen > 1) {
740 		/* includes a null-terminator */
741 		memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
742 	} else {
743 		cp[linklen] = '\0';
744 	}
745 	ndp->ni_pathlen += linklen;
746 	memcpy(ndp->ni_pnbuf, cp, ndp->ni_pathlen);
747 	PNBUF_PUT(cp);
748 
749 	/* we're now starting from the beginning of the buffer again */
750 	cnp->cn_nameptr = ndp->ni_pnbuf;
751 
752 	/* must unlock this before relocking searchdir */
753 	VOP_UNLOCK(foundobj);
754 
755 	/*
756 	 * Check if root directory should replace current directory.
757 	 */
758 	if (ndp->ni_pnbuf[0] == '/') {
759 		vput(searchdir);
760 		/* Keep absolute symbolic links inside emulation root */
761 		searchdir = ndp->ni_erootdir;
762 		if (searchdir == NULL ||
763 		    (ndp->ni_pnbuf[1] == '.'
764 		     && ndp->ni_pnbuf[2] == '.'
765 		     && ndp->ni_pnbuf[3] == '/')) {
766 			ndp->ni_erootdir = NULL;
767 			searchdir = ndp->ni_rootdir;
768 		}
769 		vref(searchdir);
770 		vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
771 	}
772 
773 	*newsearchdir_ret = searchdir;
774 	KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
775 	return 0;
776 }
777 
778 //////////////////////////////
779 
780 /*
781  * Inspect the leading path component and update the state accordingly.
782  */
783 static int
784 lookup_parsepath(struct namei_state *state)
785 {
786 	const char *cp;			/* pointer into pathname argument */
787 
788 	struct componentname *cnp = state->cnp;
789 	struct nameidata *ndp = state->ndp;
790 
791 	KASSERT(cnp == &ndp->ni_cnd);
792 
793 	/*
794 	 * Search a new directory.
795 	 *
796 	 * The cn_hash value is for use by vfs_cache.
797 	 * The last component of the filename is left accessible via
798 	 * cnp->cn_nameptr for callers that need the name. Callers needing
799 	 * the name set the SAVENAME flag. When done, they assume
800 	 * responsibility for freeing the pathname buffer.
801 	 *
802 	 * At this point, our only vnode state is that the search dir
803 	 * is held and locked.
804 	 */
805 	cnp->cn_consume = 0;
806 	cp = NULL;
807 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
808 	cnp->cn_namelen = cp - cnp->cn_nameptr;
809 	if (cnp->cn_namelen > NAME_MAX) {
810 		return ENAMETOOLONG;
811 	}
812 #ifdef NAMEI_DIAGNOSTIC
813 	{ char c = *cp;
814 	*(char *)cp = '\0';
815 	printf("{%s}: ", cnp->cn_nameptr);
816 	*(char *)cp = c; }
817 #endif /* NAMEI_DIAGNOSTIC */
818 	ndp->ni_pathlen -= cnp->cn_namelen;
819 	ndp->ni_next = cp;
820 	/*
821 	 * If this component is followed by a slash, then move the pointer to
822 	 * the next component forward, and remember that this component must be
823 	 * a directory.
824 	 */
825 	if (*cp == '/') {
826 		do {
827 			cp++;
828 		} while (*cp == '/');
829 		state->slashes = cp - ndp->ni_next;
830 		ndp->ni_pathlen -= state->slashes;
831 		ndp->ni_next = cp;
832 		cnp->cn_flags |= REQUIREDIR;
833 	} else {
834 		state->slashes = 0;
835 		cnp->cn_flags &= ~REQUIREDIR;
836 	}
837 	/*
838 	 * We do special processing on the last component, whether or not it's
839 	 * a directory.  Cache all intervening lookups, but not the final one.
840 	 */
841 	if (*cp == '\0') {
842 		if (state->docache)
843 			cnp->cn_flags |= MAKEENTRY;
844 		else
845 			cnp->cn_flags &= ~MAKEENTRY;
846 		cnp->cn_flags |= ISLASTCN;
847 	} else {
848 		cnp->cn_flags |= MAKEENTRY;
849 		cnp->cn_flags &= ~ISLASTCN;
850 	}
851 	if (cnp->cn_namelen == 2 &&
852 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
853 		cnp->cn_flags |= ISDOTDOT;
854 	else
855 		cnp->cn_flags &= ~ISDOTDOT;
856 
857 	return 0;
858 }
859 
860 /*
861  * Call VOP_LOOKUP for a single lookup; return a new search directory
862  * (used when crossing mountpoints up or searching union mounts down) and
863  * the found object, which for create operations may be NULL on success.
864  */
865 static int
866 lookup_once(struct namei_state *state,
867 	    struct vnode *searchdir,
868 	    struct vnode **newsearchdir_ret,
869 	    struct vnode **foundobj_ret)
870 {
871 	struct vnode *tmpvn;		/* scratch vnode */
872 	struct vnode *foundobj;		/* result */
873 	struct mount *mp;		/* mount table entry */
874 	struct lwp *l = curlwp;
875 	int error;
876 
877 	struct componentname *cnp = state->cnp;
878 	struct nameidata *ndp = state->ndp;
879 
880 	KASSERT(cnp == &ndp->ni_cnd);
881 	KASSERT(VOP_ISLOCKED(searchdir) == LK_EXCLUSIVE);
882 	*newsearchdir_ret = searchdir;
883 
884 	/*
885 	 * Handle "..": two special cases.
886 	 * 1. If at root directory (e.g. after chroot)
887 	 *    or at absolute root directory
888 	 *    then ignore it so can't get out.
889 	 * 1a. If at the root of the emulation filesystem go to the real
890 	 *    root. So "/../<path>" is always absolute.
891 	 * 1b. If we have somehow gotten out of a jail, warn
892 	 *    and also ignore it so we can't get farther out.
893 	 * 2. If this vnode is the root of a mounted
894 	 *    filesystem, then replace it with the
895 	 *    vnode which was mounted on so we take the
896 	 *    .. in the other file system.
897 	 */
898 	if (cnp->cn_flags & ISDOTDOT) {
899 		struct proc *p = l->l_proc;
900 
901 		for (;;) {
902 			if (searchdir == ndp->ni_rootdir ||
903 			    searchdir == rootvnode) {
904 				foundobj = searchdir;
905 				vref(foundobj);
906 				*foundobj_ret = foundobj;
907 				error = 0;
908 				goto done;
909 			}
910 			if (ndp->ni_rootdir != rootvnode) {
911 				int retval;
912 
913 				VOP_UNLOCK(searchdir);
914 				retval = vn_isunder(searchdir, ndp->ni_rootdir, l);
915 				vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
916 				if (!retval) {
917 				    /* Oops! We got out of jail! */
918 				    log(LOG_WARNING,
919 					"chrooted pid %d uid %d (%s) "
920 					"detected outside of its chroot\n",
921 					p->p_pid, kauth_cred_geteuid(l->l_cred),
922 					p->p_comm);
923 				    /* Put us at the jail root. */
924 				    vput(searchdir);
925 				    searchdir = NULL;
926 				    foundobj = ndp->ni_rootdir;
927 				    vref(foundobj);
928 				    vref(foundobj);
929 				    vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
930 				    *newsearchdir_ret = foundobj;
931 				    *foundobj_ret = foundobj;
932 				    error = 0;
933 				    goto done;
934 				}
935 			}
936 			if ((searchdir->v_vflag & VV_ROOT) == 0 ||
937 			    (cnp->cn_flags & NOCROSSMOUNT))
938 				break;
939 			tmpvn = searchdir;
940 			searchdir = searchdir->v_mount->mnt_vnodecovered;
941 			vref(searchdir);
942 			vput(tmpvn);
943 			vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
944 			*newsearchdir_ret = searchdir;
945 		}
946 	}
947 
948 	/*
949 	 * We now have a segment name to search for, and a directory to search.
950 	 * Our vnode state here is that "searchdir" is held and locked.
951 	 */
952 unionlookup:
953 	foundobj = NULL;
954 	error = VOP_LOOKUP(searchdir, &foundobj, cnp);
955 
956 	if (error != 0) {
957 #ifdef DIAGNOSTIC
958 		if (foundobj != NULL)
959 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
960 #endif /* DIAGNOSTIC */
961 #ifdef NAMEI_DIAGNOSTIC
962 		printf("not found\n");
963 #endif /* NAMEI_DIAGNOSTIC */
964 		if ((error == ENOENT) &&
965 		    (searchdir->v_vflag & VV_ROOT) &&
966 		    (searchdir->v_mount->mnt_flag & MNT_UNION)) {
967 			tmpvn = searchdir;
968 			searchdir = searchdir->v_mount->mnt_vnodecovered;
969 			vref(searchdir);
970 			vput(tmpvn);
971 			vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
972 			*newsearchdir_ret = searchdir;
973 			goto unionlookup;
974 		}
975 
976 		if (error != EJUSTRETURN)
977 			goto done;
978 
979 		/*
980 		 * If this was not the last component, or there were trailing
981 		 * slashes, and we are not going to create a directory,
982 		 * then the name must exist.
983 		 */
984 		if ((cnp->cn_flags & (REQUIREDIR | CREATEDIR)) == REQUIREDIR) {
985 			error = ENOENT;
986 			goto done;
987 		}
988 
989 		/*
990 		 * If creating and at end of pathname, then can consider
991 		 * allowing file to be created.
992 		 */
993 		if (state->rdonly) {
994 			error = EROFS;
995 			goto done;
996 		}
997 
998 		/*
999 		 * We return success and a NULL foundobj to indicate
1000 		 * that the entry doesn't currently exist, leaving a
1001 		 * pointer to the (normally, locked) directory vnode
1002 		 * as searchdir.
1003 		 */
1004 		*foundobj_ret = NULL;
1005 		error = 0;
1006 		goto done;
1007 	}
1008 #ifdef NAMEI_DIAGNOSTIC
1009 	printf("found\n");
1010 #endif /* NAMEI_DIAGNOSTIC */
1011 
1012 	/*
1013 	 * Take into account any additional components consumed by the
1014 	 * underlying filesystem.  This will include any trailing slashes after
1015 	 * the last component consumed.
1016 	 */
1017 	if (cnp->cn_consume > 0) {
1018 		ndp->ni_pathlen -= cnp->cn_consume - state->slashes;
1019 		ndp->ni_next += cnp->cn_consume - state->slashes;
1020 		cnp->cn_consume = 0;
1021 		if (ndp->ni_next[0] == '\0')
1022 			cnp->cn_flags |= ISLASTCN;
1023 	}
1024 
1025 	/*
1026 	 * "foundobj" and "searchdir" are both locked and held,
1027 	 * and may be the same vnode.
1028 	 */
1029 
1030 	/*
1031 	 * Check to see if the vnode has been mounted on;
1032 	 * if so find the root of the mounted file system.
1033 	 */
1034 	while (foundobj->v_type == VDIR &&
1035 	       (mp = foundobj->v_mountedhere) != NULL &&
1036 	       (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1037 		error = vfs_busy(mp, NULL);
1038 		if (error != 0) {
1039 			vput(foundobj);
1040 			goto done;
1041 		}
1042 		KASSERT(searchdir != foundobj);
1043 		VOP_UNLOCK(searchdir);
1044 		vput(foundobj);
1045 		error = VFS_ROOT(mp, &foundobj);
1046 		vfs_unbusy(mp, false, NULL);
1047 		if (error) {
1048 			vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
1049 			goto done;
1050 		}
1051 		VOP_UNLOCK(foundobj);
1052 		vn_lock(searchdir, LK_EXCLUSIVE | LK_RETRY);
1053 		vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
1054 	}
1055 
1056 	*foundobj_ret = foundobj;
1057 	error = 0;
1058 done:
1059 	KASSERT(VOP_ISLOCKED(*newsearchdir_ret) == LK_EXCLUSIVE);
1060 	/*
1061 	 * *foundobj_ret is valid only if error == 0.
1062 	 */
1063 	KASSERT(error != 0 || *foundobj_ret == NULL ||
1064 	    VOP_ISLOCKED(*foundobj_ret) == LK_EXCLUSIVE);
1065 	return error;
1066 }
1067 
1068 //////////////////////////////
1069 
1070 /*
1071  * Do a complete path search from a single root directory.
1072  * (This is called up to twice if TRYEMULROOT is in effect.)
1073  */
1074 static int
1075 namei_oneroot(struct namei_state *state, struct vnode *forcecwd,
1076 	 int neverfollow, int inhibitmagic)
1077 {
1078 	struct nameidata *ndp = state->ndp;
1079 	struct componentname *cnp = state->cnp;
1080 	struct vnode *searchdir, *foundobj;
1081 	const char *cp;
1082 	int error;
1083 
1084 	error = namei_start(state, forcecwd, &searchdir);
1085 	if (error) {
1086 		ndp->ni_dvp = NULL;
1087 		ndp->ni_vp = NULL;
1088 		return error;
1089 	}
1090 
1091 	/*
1092 	 * Setup: break out flag bits into variables.
1093 	 */
1094 	state->docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
1095 	if (cnp->cn_nameiop == DELETE)
1096 		state->docache = 0;
1097 	state->rdonly = cnp->cn_flags & RDONLY;
1098 
1099 	/*
1100 	 * Keep going until we run out of path components.
1101 	 */
1102 	cnp->cn_nameptr = ndp->ni_pnbuf;
1103 	for (;;) {
1104 
1105 		/*
1106 		 * If the directory we're on is unmounted, bail out.
1107 		 * XXX: should this also check if it's unlinked?
1108 		 */
1109 		if (searchdir->v_mount == NULL) {
1110 			vput(searchdir);
1111 			ndp->ni_dvp = NULL;
1112 			ndp->ni_vp = NULL;
1113 			return (ENOENT);
1114 		}
1115 
1116 		/*
1117 		 * Look up the next path component.
1118 		 * (currently, this may consume more than one)
1119 		 */
1120 
1121 		cnp->cn_flags &= ~ISSYMLINK;
1122 
1123     dirloop:
1124 		/*
1125 		 * If we have a leading string of slashes, remove
1126 		 * them, and just make sure the current node is a
1127 		 * directory.
1128 		 */
1129 		cp = cnp->cn_nameptr;
1130 		if (*cp == '/') {
1131 			do {
1132 				cp++;
1133 			} while (*cp == '/');
1134 			ndp->ni_pathlen -= cp - cnp->cn_nameptr;
1135 			cnp->cn_nameptr = cp;
1136 
1137 			if (searchdir->v_type != VDIR) {
1138 				vput(searchdir);
1139 				ndp->ni_dvp = NULL;
1140 				ndp->ni_vp = NULL;
1141 				state->attempt_retry = 1;
1142 				return ENOTDIR;
1143 			}
1144 		}
1145 
1146 		/*
1147 		 * If we've exhausted the path name, then just return the
1148 		 * current node.
1149 		 */
1150 		if (cnp->cn_nameptr[0] == '\0') {
1151 			vref(searchdir);
1152 			foundobj = searchdir;
1153 			cnp->cn_flags |= ISLASTCN;
1154 
1155 			/* bleh */
1156 			goto terminal;
1157 		}
1158 
1159 		error = lookup_parsepath(state);
1160 		if (error) {
1161 			vput(searchdir);
1162 			ndp->ni_dvp = NULL;
1163 			ndp->ni_vp = NULL;
1164 			state->attempt_retry = 1;
1165 			return (error);
1166 		}
1167 
1168 		error = lookup_once(state, searchdir, &searchdir, &foundobj);
1169 		if (error) {
1170 			vput(searchdir);
1171 			ndp->ni_dvp = NULL;
1172 			ndp->ni_vp = NULL;
1173 			/*
1174 			 * Note that if we're doing TRYEMULROOT we can
1175 			 * retry with the normal root. Where this is
1176 			 * currently set matches previous practice,
1177 			 * but the previous practice didn't make much
1178 			 * sense and somebody should sit down and
1179 			 * figure out which cases should cause retry
1180 			 * and which shouldn't. XXX.
1181 			 */
1182 			state->attempt_retry = 1;
1183 			return (error);
1184 		}
1185 
1186 		if (foundobj == NULL) {
1187 			/*
1188 			 * Success with no object returned means we're
1189 			 * creating something and it isn't already
1190 			 * there. Break out of the main loop now so
1191 			 * the code below doesn't have to test for
1192 			 * foundobj == NULL.
1193 			 */
1194 			break;
1195 		}
1196 
1197 		/*
1198 		 * Check for symbolic link. If we've reached one,
1199 		 * follow it, unless we aren't supposed to. Back up
1200 		 * over any slashes that we skipped, as we will need
1201 		 * them again.
1202 		 */
1203 		if (namei_atsymlink(state, foundobj)) {
1204 			ndp->ni_pathlen += state->slashes;
1205 			ndp->ni_next -= state->slashes;
1206 			cnp->cn_flags |= ISSYMLINK;
1207 			if (neverfollow) {
1208 				error = EINVAL;
1209 			} else {
1210 				/*
1211 				 * dholland 20110410: if we're at a
1212 				 * union mount it might make sense to
1213 				 * use the top of the union stack here
1214 				 * rather than the layer we found the
1215 				 * symlink in. (FUTURE)
1216 				 */
1217 				error = namei_follow(state, inhibitmagic,
1218 						     searchdir, foundobj,
1219 						     &searchdir);
1220 			}
1221 			if (error) {
1222 				KASSERT(searchdir != foundobj);
1223 				vput(searchdir);
1224 				vput(foundobj);
1225 				ndp->ni_dvp = NULL;
1226 				ndp->ni_vp = NULL;
1227 				return error;
1228 			}
1229 			/* namei_follow unlocks it (ugh) so rele, not put */
1230 			vrele(foundobj);
1231 			foundobj = NULL;
1232 			continue;
1233 		}
1234 
1235 		/*
1236 		 * Check for directory, if the component was
1237 		 * followed by a series of slashes.
1238 		 */
1239 		if ((foundobj->v_type != VDIR) && (cnp->cn_flags & REQUIREDIR)) {
1240 			KASSERT(foundobj != searchdir);
1241 			if (searchdir) {
1242 				vput(searchdir);
1243 			}
1244 			vput(foundobj);
1245 			ndp->ni_dvp = NULL;
1246 			ndp->ni_vp = NULL;
1247 			state->attempt_retry = 1;
1248 			return ENOTDIR;
1249 		}
1250 
1251 		/*
1252 		 * Not a symbolic link.  If this was not the
1253 		 * last component, then continue at the next
1254 		 * component, else return.
1255 		 */
1256 		if (!(cnp->cn_flags & ISLASTCN)) {
1257 			cnp->cn_nameptr = ndp->ni_next;
1258 			if (searchdir == foundobj) {
1259 				vrele(searchdir);
1260 			} else {
1261 				vput(searchdir);
1262 			}
1263 			searchdir = foundobj;
1264 			foundobj = NULL;
1265 			goto dirloop;
1266 		}
1267 
1268     terminal:
1269 		if (foundobj == ndp->ni_erootdir) {
1270 			/*
1271 			 * We are about to return the emulation root.
1272 			 * This isn't a good idea because code might
1273 			 * repeatedly lookup ".." until the file
1274 			 * matches that returned for "/" and loop
1275 			 * forever.  So convert it to the real root.
1276 			 */
1277 			if (searchdir != NULL) {
1278 				if (searchdir == foundobj)
1279 					vrele(searchdir);
1280 				else
1281 					vput(searchdir);
1282 				searchdir = NULL;
1283 			}
1284 			vput(foundobj);
1285 			foundobj = ndp->ni_rootdir;
1286 			vref(foundobj);
1287 			vn_lock(foundobj, LK_EXCLUSIVE | LK_RETRY);
1288 		}
1289 
1290 		/*
1291 		 * If the caller requested the parent node (i.e. it's
1292 		 * a CREATE, DELETE, or RENAME), and we don't have one
1293 		 * (because this is the root directory, or we crossed
1294 		 * a mount point), then we must fail.
1295 		 */
1296 		if (cnp->cn_nameiop != LOOKUP &&
1297 		    (searchdir == NULL ||
1298 		     searchdir->v_mount != foundobj->v_mount)) {
1299 			if (searchdir) {
1300 				vput(searchdir);
1301 			}
1302 			vput(foundobj);
1303 			foundobj = NULL;
1304 			ndp->ni_dvp = NULL;
1305 			ndp->ni_vp = NULL;
1306 			state->attempt_retry = 1;
1307 
1308 			switch (cnp->cn_nameiop) {
1309 			    case CREATE:
1310 				return EEXIST;
1311 			    case DELETE:
1312 			    case RENAME:
1313 				return EBUSY;
1314 			    default:
1315 				break;
1316 			}
1317 			panic("Invalid nameiop\n");
1318 		}
1319 
1320 		/*
1321 		 * Disallow directory write attempts on read-only lookups.
1322 		 * Prefers EEXIST over EROFS for the CREATE case.
1323 		 */
1324 		if (state->rdonly &&
1325 		    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1326 			if (searchdir) {
1327 				if (foundobj != searchdir) {
1328 					vput(searchdir);
1329 				} else {
1330 					vrele(searchdir);
1331 				}
1332 				searchdir = NULL;
1333 			}
1334 			vput(foundobj);
1335 			foundobj = NULL;
1336 			ndp->ni_dvp = NULL;
1337 			ndp->ni_vp = NULL;
1338 			state->attempt_retry = 1;
1339 			return EROFS;
1340 		}
1341 		if ((cnp->cn_flags & LOCKLEAF) == 0) {
1342 			/*
1343 			 * Note: if LOCKPARENT but not LOCKLEAF is
1344 			 * set, and searchdir == foundobj, this code
1345 			 * necessarily unlocks the parent as well as
1346 			 * the leaf. That is, just because you specify
1347 			 * LOCKPARENT doesn't mean you necessarily get
1348 			 * a locked parent vnode. The code in
1349 			 * vfs_syscalls.c, and possibly elsewhere,
1350 			 * that uses this combination "knows" this, so
1351 			 * it can't be safely changed. Feh. XXX
1352 			 */
1353 			VOP_UNLOCK(foundobj);
1354 		}
1355 
1356 		break;
1357 	}
1358 
1359 	/*
1360 	 * Done.
1361 	 */
1362 
1363 	/*
1364 	 * If LOCKPARENT is not set, the parent directory isn't returned.
1365 	 */
1366 	if ((cnp->cn_flags & LOCKPARENT) == 0 && searchdir != NULL) {
1367 		if (searchdir == foundobj) {
1368 			vrele(searchdir);
1369 		} else {
1370 			vput(searchdir);
1371 		}
1372 		searchdir = NULL;
1373 	}
1374 
1375 	ndp->ni_dvp = searchdir;
1376 	ndp->ni_vp = foundobj;
1377 	return 0;
1378 }
1379 
1380 /*
1381  * Do namei; wrapper layer that handles TRYEMULROOT.
1382  */
1383 static int
1384 namei_tryemulroot(struct namei_state *state, struct vnode *forcecwd,
1385 	 int neverfollow, int inhibitmagic)
1386 {
1387 	int error;
1388 
1389 	struct nameidata *ndp = state->ndp;
1390 	struct componentname *cnp = state->cnp;
1391 	const char *savepath = NULL;
1392 
1393 	KASSERT(cnp == &ndp->ni_cnd);
1394 
1395 	if (cnp->cn_flags & TRYEMULROOT) {
1396 		savepath = pathbuf_stringcopy_get(ndp->ni_pathbuf);
1397 	}
1398 
1399     emul_retry:
1400 	state->attempt_retry = 0;
1401 
1402 	error = namei_oneroot(state, forcecwd, neverfollow, inhibitmagic);
1403 	if (error) {
1404 		/*
1405 		 * Once namei has started up, the existence of ni_erootdir
1406 		 * tells us whether we're working from an emulation root.
1407 		 * The TRYEMULROOT flag isn't necessarily authoritative.
1408 		 */
1409 		if (ndp->ni_erootdir != NULL && state->attempt_retry) {
1410 			/* Retry the whole thing using the normal root */
1411 			cnp->cn_flags &= ~TRYEMULROOT;
1412 			state->attempt_retry = 0;
1413 
1414 			/* kinda gross */
1415 			strcpy(ndp->ni_pathbuf->pb_path, savepath);
1416 			pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
1417 			savepath = NULL;
1418 
1419 			goto emul_retry;
1420 		}
1421 	}
1422 	if (savepath != NULL) {
1423 		pathbuf_stringcopy_put(ndp->ni_pathbuf, savepath);
1424 	}
1425 	return error;
1426 }
1427 
1428 /*
1429  * External interface.
1430  */
1431 int
1432 namei(struct nameidata *ndp)
1433 {
1434 	struct namei_state state;
1435 	int error;
1436 
1437 	namei_init(&state, ndp);
1438 	error = namei_tryemulroot(&state, NULL,
1439 				  0/*!neverfollow*/, 0/*!inhibitmagic*/);
1440 	namei_cleanup(&state);
1441 
1442 	if (error) {
1443 		/* make sure no stray refs leak out */
1444 		KASSERT(ndp->ni_dvp == NULL);
1445 		KASSERT(ndp->ni_vp == NULL);
1446 	}
1447 
1448 	return error;
1449 }
1450 
1451 ////////////////////////////////////////////////////////////
1452 
1453 /*
1454  * External interface used by nfsd. This is basically different from
1455  * namei only in that it has the ability to pass in the "current
1456  * directory", and uses an extra flag "neverfollow" for which there's
1457  * no physical flag defined in namei.h. (There used to be a cut&paste
1458  * copy of about half of namei in nfsd to allow these minor
1459  * adjustments to exist.)
1460  *
1461  * XXX: the namei interface should be adjusted so nfsd can just use
1462  * ordinary namei().
1463  */
1464 int
1465 lookup_for_nfsd(struct nameidata *ndp, struct vnode *forcecwd, int neverfollow)
1466 {
1467 	struct namei_state state;
1468 	int error;
1469 
1470 	namei_init(&state, ndp);
1471 	error = namei_tryemulroot(&state, forcecwd,
1472 				  neverfollow, 1/*inhibitmagic*/);
1473 	namei_cleanup(&state);
1474 
1475 	if (error) {
1476 		/* make sure no stray refs leak out */
1477 		KASSERT(ndp->ni_dvp == NULL);
1478 		KASSERT(ndp->ni_vp == NULL);
1479 	}
1480 
1481 	return error;
1482 }
1483 
1484 /*
1485  * A second external interface used by nfsd. This turns out to be a
1486  * single lookup used by the WebNFS code (ha!) to get "index.html" or
1487  * equivalent when asked for a directory. It should eventually evolve
1488  * into some kind of namei_once() call; for the time being it's kind
1489  * of a mess. XXX.
1490  *
1491  * dholland 20110109: I don't think it works, and I don't think it
1492  * worked before I started hacking and slashing either, and I doubt
1493  * anyone will ever notice.
1494  */
1495 
1496 /*
1497  * Internals. This calls lookup_once() after setting up the assorted
1498  * pieces of state the way they ought to be.
1499  */
1500 static int
1501 do_lookup_for_nfsd_index(struct namei_state *state, struct vnode *startdir)
1502 {
1503 	int error = 0;
1504 
1505 	struct componentname *cnp = state->cnp;
1506 	struct nameidata *ndp = state->ndp;
1507 	struct vnode *foundobj;
1508 	const char *cp;			/* pointer into pathname argument */
1509 
1510 	KASSERT(cnp == &ndp->ni_cnd);
1511 
1512 	cnp->cn_nameptr = ndp->ni_pnbuf;
1513 	state->docache = 1;
1514 	state->rdonly = cnp->cn_flags & RDONLY;
1515 	ndp->ni_dvp = NULL;
1516 	cnp->cn_flags &= ~ISSYMLINK;
1517 
1518 	cnp->cn_consume = 0;
1519 	cp = NULL;
1520 	cnp->cn_hash = namei_hash(cnp->cn_nameptr, &cp);
1521 	cnp->cn_namelen = cp - cnp->cn_nameptr;
1522 	KASSERT(cnp->cn_namelen <= NAME_MAX);
1523 	ndp->ni_pathlen -= cnp->cn_namelen;
1524 	ndp->ni_next = cp;
1525 	state->slashes = 0;
1526 	cnp->cn_flags &= ~REQUIREDIR;
1527 	cnp->cn_flags |= MAKEENTRY|ISLASTCN;
1528 
1529 	if (cnp->cn_namelen == 2 &&
1530 	    cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
1531 		cnp->cn_flags |= ISDOTDOT;
1532 	else
1533 		cnp->cn_flags &= ~ISDOTDOT;
1534 
1535 	/*
1536 	 * Because lookup_once can change the startdir, we need our
1537 	 * own reference to it to avoid consuming the caller's.
1538 	 */
1539 	vref(startdir);
1540 	vn_lock(startdir, LK_EXCLUSIVE | LK_RETRY);
1541 	error = lookup_once(state, startdir, &startdir, &foundobj);
1542 	vput(startdir);
1543 	if (error) {
1544 		goto bad;
1545 	}
1546 	ndp->ni_vp = foundobj;
1547 
1548 	if (foundobj == NULL) {
1549 		return 0;
1550 	}
1551 
1552 	KASSERT((cnp->cn_flags & LOCKPARENT) == 0);
1553 	if ((cnp->cn_flags & LOCKLEAF) == 0) {
1554 		VOP_UNLOCK(foundobj);
1555 	}
1556 	return (0);
1557 
1558 bad:
1559 	ndp->ni_vp = NULL;
1560 	return (error);
1561 }
1562 
1563 /*
1564  * External interface. The partitioning between this function and the
1565  * above isn't very clear - the above function exists mostly so code
1566  * that uses "state->" can be shuffled around without having to change
1567  * it to "state.".
1568  */
1569 int
1570 lookup_for_nfsd_index(struct nameidata *ndp, struct vnode *startdir)
1571 {
1572 	struct namei_state state;
1573 	int error;
1574 
1575 	/*
1576 	 * Note: the name sent in here (is not|should not be) allowed
1577 	 * to contain a slash.
1578 	 */
1579 	if (strlen(ndp->ni_pathbuf->pb_path) > NAME_MAX) {
1580 		return ENAMETOOLONG;
1581 	}
1582 	if (strchr(ndp->ni_pathbuf->pb_path, '/')) {
1583 		return EINVAL;
1584 	}
1585 
1586 	ndp->ni_pathlen = strlen(ndp->ni_pathbuf->pb_path) + 1;
1587 	ndp->ni_pnbuf = NULL;
1588 	ndp->ni_cnd.cn_nameptr = NULL;
1589 
1590 	namei_init(&state, ndp);
1591 	error = do_lookup_for_nfsd_index(&state, startdir);
1592 	namei_cleanup(&state);
1593 
1594 	return error;
1595 }
1596 
1597 ////////////////////////////////////////////////////////////
1598 
1599 /*
1600  * Reacquire a path name component.
1601  * dvp is locked on entry and exit.
1602  * *vpp is locked on exit unless it's NULL.
1603  */
1604 int
1605 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, int dummy)
1606 {
1607 	int rdonly;			/* lookup read-only flag bit */
1608 	int error = 0;
1609 #ifdef DEBUG
1610 	uint32_t newhash;		/* DEBUG: check name hash */
1611 	const char *cp;			/* DEBUG: check name ptr/len */
1612 #endif /* DEBUG */
1613 
1614 	(void)dummy;
1615 
1616 	/*
1617 	 * Setup: break out flag bits into variables.
1618 	 */
1619 	rdonly = cnp->cn_flags & RDONLY;
1620 	cnp->cn_flags &= ~ISSYMLINK;
1621 
1622 	/*
1623 	 * Search a new directory.
1624 	 *
1625 	 * The cn_hash value is for use by vfs_cache.
1626 	 * The last component of the filename is left accessible via
1627 	 * cnp->cn_nameptr for callers that need the name. Callers needing
1628 	 * the name set the SAVENAME flag. When done, they assume
1629 	 * responsibility for freeing the pathname buffer.
1630 	 */
1631 #ifdef DEBUG
1632 	cp = NULL;
1633 	newhash = namei_hash(cnp->cn_nameptr, &cp);
1634 	if ((uint32_t)newhash != (uint32_t)cnp->cn_hash)
1635 		panic("relookup: bad hash");
1636 	if (cnp->cn_namelen != cp - cnp->cn_nameptr)
1637 		panic("relookup: bad len");
1638 	while (*cp == '/')
1639 		cp++;
1640 	if (*cp != 0)
1641 		panic("relookup: not last component");
1642 #endif /* DEBUG */
1643 
1644 	/*
1645 	 * Check for degenerate name (e.g. / or "")
1646 	 * which is a way of talking about a directory,
1647 	 * e.g. like "/." or ".".
1648 	 */
1649 	if (cnp->cn_nameptr[0] == '\0')
1650 		panic("relookup: null name");
1651 
1652 	if (cnp->cn_flags & ISDOTDOT)
1653 		panic("relookup: lookup on dot-dot");
1654 
1655 	/*
1656 	 * We now have a segment name to search for, and a directory to search.
1657 	 */
1658 	cnp->cn_flags |= INRELOOKUP;
1659 	error = VOP_LOOKUP(dvp, vpp, cnp);
1660 	cnp->cn_flags &= ~INRELOOKUP;
1661 	if ((error) != 0) {
1662 #ifdef DIAGNOSTIC
1663 		if (*vpp != NULL)
1664 			panic("leaf `%s' should be empty", cnp->cn_nameptr);
1665 #endif
1666 		if (error != EJUSTRETURN)
1667 			goto bad;
1668 	}
1669 
1670 #ifdef DIAGNOSTIC
1671 	/*
1672 	 * Check for symbolic link
1673 	 */
1674 	if (*vpp && (*vpp)->v_type == VLNK && (cnp->cn_flags & FOLLOW))
1675 		panic("relookup: symlink found");
1676 #endif
1677 
1678 	/*
1679 	 * Check for read-only lookups.
1680 	 */
1681 	if (rdonly && cnp->cn_nameiop != LOOKUP) {
1682 		error = EROFS;
1683 		if (*vpp) {
1684 			vput(*vpp);
1685 		}
1686 		goto bad;
1687 	}
1688 	return (0);
1689 
1690 bad:
1691 	*vpp = NULL;
1692 	return (error);
1693 }
1694 
1695 /*
1696  * namei_simple - simple forms of namei.
1697  *
1698  * These are wrappers to allow the simple case callers of namei to be
1699  * left alone while everything else changes under them.
1700  */
1701 
1702 /* Flags */
1703 struct namei_simple_flags_type {
1704 	int dummy;
1705 };
1706 static const struct namei_simple_flags_type ns_nn, ns_nt, ns_fn, ns_ft;
1707 const namei_simple_flags_t NSM_NOFOLLOW_NOEMULROOT = &ns_nn;
1708 const namei_simple_flags_t NSM_NOFOLLOW_TRYEMULROOT = &ns_nt;
1709 const namei_simple_flags_t NSM_FOLLOW_NOEMULROOT = &ns_fn;
1710 const namei_simple_flags_t NSM_FOLLOW_TRYEMULROOT = &ns_ft;
1711 
1712 static
1713 int
1714 namei_simple_convert_flags(namei_simple_flags_t sflags)
1715 {
1716 	if (sflags == NSM_NOFOLLOW_NOEMULROOT)
1717 		return NOFOLLOW | 0;
1718 	if (sflags == NSM_NOFOLLOW_TRYEMULROOT)
1719 		return NOFOLLOW | TRYEMULROOT;
1720 	if (sflags == NSM_FOLLOW_NOEMULROOT)
1721 		return FOLLOW | 0;
1722 	if (sflags == NSM_FOLLOW_TRYEMULROOT)
1723 		return FOLLOW | TRYEMULROOT;
1724 	panic("namei_simple_convert_flags: bogus sflags\n");
1725 	return 0;
1726 }
1727 
1728 int
1729 namei_simple_kernel(const char *path, namei_simple_flags_t sflags,
1730 			struct vnode **vp_ret)
1731 {
1732 	struct nameidata nd;
1733 	struct pathbuf *pb;
1734 	int err;
1735 
1736 	pb = pathbuf_create(path);
1737 	if (pb == NULL) {
1738 		return ENOMEM;
1739 	}
1740 
1741 	NDINIT(&nd,
1742 		LOOKUP,
1743 		namei_simple_convert_flags(sflags),
1744 		pb);
1745 	err = namei(&nd);
1746 	if (err != 0) {
1747 		pathbuf_destroy(pb);
1748 		return err;
1749 	}
1750 	*vp_ret = nd.ni_vp;
1751 	pathbuf_destroy(pb);
1752 	return 0;
1753 }
1754 
1755 int
1756 namei_simple_user(const char *path, namei_simple_flags_t sflags,
1757 			struct vnode **vp_ret)
1758 {
1759 	struct pathbuf *pb;
1760 	struct nameidata nd;
1761 	int err;
1762 
1763 	err = pathbuf_copyin(path, &pb);
1764 	if (err) {
1765 		return err;
1766 	}
1767 
1768 	NDINIT(&nd,
1769 		LOOKUP,
1770 		namei_simple_convert_flags(sflags),
1771 		pb);
1772 	err = namei(&nd);
1773 	if (err != 0) {
1774 		pathbuf_destroy(pb);
1775 		return err;
1776 	}
1777 	*vp_ret = nd.ni_vp;
1778 	pathbuf_destroy(pb);
1779 	return 0;
1780 }
1781