xref: /freebsd/sys/kern/vfs_cache.c (revision 48c5129f93c5eb5419c87b08e4677d51513f1dc0)
1 /*-
2  * Copyright (c) 1989, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Poul-Henning Kamp of the FreeBSD Project.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)vfs_cache.c	8.5 (Berkeley) 3/22/95
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include "opt_kdtrace.h"
39 #include "opt_ktrace.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/filedesc.h>
44 #include <sys/fnv_hash.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/fcntl.h>
49 #include <sys/mount.h>
50 #include <sys/namei.h>
51 #include <sys/proc.h>
52 #include <sys/rwlock.h>
53 #include <sys/sdt.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysproto.h>
57 #include <sys/vnode.h>
58 #ifdef KTRACE
59 #include <sys/ktrace.h>
60 #endif
61 
62 #include <vm/uma.h>
63 
64 SDT_PROVIDER_DECLARE(vfs);
65 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, done, "struct vnode *", "char *",
66     "struct vnode *");
67 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, done, "struct vnode *",
68     "char *");
69 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, entry, "struct vnode *");
70 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, hit, "struct vnode *",
71     "struct char *", "struct vnode *");
72 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, miss, "struct vnode *");
73 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, return, "int",
74     "struct vnode *", "struct char *");
75 SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, hit, "struct vnode *", "char *",
76     "struct vnode *");
77 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit_negative, hit-negative,
78     "struct vnode *", "char *");
79 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, miss, "struct vnode *",
80     "char *");
81 SDT_PROBE_DEFINE1(vfs, namecache, purge, done, done, "struct vnode *");
82 SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, done, "struct vnode *");
83 SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, done, "struct mount *");
84 SDT_PROBE_DEFINE3(vfs, namecache, zap, done, done, "struct vnode *", "char *",
85     "struct vnode *");
86 SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, done, "struct vnode *",
87     "char *");
88 
89 /*
90  * This structure describes the elements in the cache of recent
91  * names looked up by namei.
92  */
93 
94 struct	namecache {
95 	LIST_ENTRY(namecache) nc_hash;	/* hash chain */
96 	LIST_ENTRY(namecache) nc_src;	/* source vnode list */
97 	TAILQ_ENTRY(namecache) nc_dst;	/* destination vnode list */
98 	struct	vnode *nc_dvp;		/* vnode of parent of name */
99 	struct	vnode *nc_vp;		/* vnode the name refers to */
100 	u_char	nc_flag;		/* flag bits */
101 	u_char	nc_nlen;		/* length of name */
102 	char	nc_name[0];		/* segment name + nul */
103 };
104 
105 /*
106  * struct namecache_ts repeats struct namecache layout up to the
107  * nc_nlen member.
108  * struct namecache_ts is used in place of struct namecache when time(s) need
109  * to be stored.  The nc_dotdottime field is used when a cache entry is mapping
110  * both a non-dotdot directory name plus dotdot for the directory's
111  * parent.
112  */
113 struct	namecache_ts {
114 	LIST_ENTRY(namecache) nc_hash;	/* hash chain */
115 	LIST_ENTRY(namecache) nc_src;	/* source vnode list */
116 	TAILQ_ENTRY(namecache) nc_dst;	/* destination vnode list */
117 	struct	vnode *nc_dvp;		/* vnode of parent of name */
118 	struct	vnode *nc_vp;		/* vnode the name refers to */
119 	u_char	nc_flag;		/* flag bits */
120 	u_char	nc_nlen;		/* length of name */
121 	struct	timespec nc_time;	/* timespec provided by fs */
122 	struct	timespec nc_dotdottime;	/* dotdot timespec provided by fs */
123 	int	nc_ticks;		/* ticks value when entry was added */
124 	char	nc_name[0];		/* segment name + nul */
125 };
126 
127 /*
128  * Flags in namecache.nc_flag
129  */
130 #define NCF_WHITE	0x01
131 #define NCF_ISDOTDOT	0x02
132 #define	NCF_TS		0x04
133 #define	NCF_DTS		0x08
134 
135 /*
136  * Name caching works as follows:
137  *
138  * Names found by directory scans are retained in a cache
139  * for future reference.  It is managed LRU, so frequently
140  * used names will hang around.  Cache is indexed by hash value
141  * obtained from (vp, name) where vp refers to the directory
142  * containing name.
143  *
144  * If it is a "negative" entry, (i.e. for a name that is known NOT to
145  * exist) the vnode pointer will be NULL.
146  *
147  * Upon reaching the last segment of a path, if the reference
148  * is for DELETE, or NOCACHE is set (rewrite), and the
149  * name is located in the cache, it will be dropped.
150  */
151 
152 /*
153  * Structures associated with name cacheing.
154  */
155 #define NCHHASH(hash) \
156 	(&nchashtbl[(hash) & nchash])
157 static LIST_HEAD(nchashhead, namecache) *nchashtbl;	/* Hash Table */
158 static TAILQ_HEAD(, namecache) ncneg;	/* Hash Table */
159 static u_long	nchash;			/* size of hash table */
160 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0,
161     "Size of namecache hash table");
162 static u_long	ncnegfactor = 16;	/* ratio of negative entries */
163 SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0,
164     "Ratio of negative namecache entries");
165 static u_long	numneg;			/* number of negative entries allocated */
166 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0,
167     "Number of negative entries in namecache");
168 static u_long	numcache;		/* number of cache entries allocated */
169 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0,
170     "Number of namecache entries");
171 static u_long	numcachehv;		/* number of cache entries with vnodes held */
172 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0,
173     "Number of namecache entries with vnodes held");
174 static u_int	ncsizefactor = 2;
175 SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0,
176     "Size factor for namecache");
177 
178 struct nchstats	nchstats;		/* cache effectiveness statistics */
179 
180 static struct rwlock cache_lock;
181 RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
182 
183 #define	CACHE_UPGRADE_LOCK()	rw_try_upgrade(&cache_lock)
184 #define	CACHE_RLOCK()		rw_rlock(&cache_lock)
185 #define	CACHE_RUNLOCK()		rw_runlock(&cache_lock)
186 #define	CACHE_WLOCK()		rw_wlock(&cache_lock)
187 #define	CACHE_WUNLOCK()		rw_wunlock(&cache_lock)
188 
189 /*
190  * UMA zones for the VFS cache.
191  *
192  * The small cache is used for entries with short names, which are the
193  * most common.  The large cache is used for entries which are too big to
194  * fit in the small cache.
195  */
196 static uma_zone_t cache_zone_small;
197 static uma_zone_t cache_zone_small_ts;
198 static uma_zone_t cache_zone_large;
199 static uma_zone_t cache_zone_large_ts;
200 
201 #define	CACHE_PATH_CUTOFF	35
202 
203 static struct namecache *
204 cache_alloc(int len, int ts)
205 {
206 
207 	if (len > CACHE_PATH_CUTOFF) {
208 		if (ts)
209 			return (uma_zalloc(cache_zone_large_ts, M_WAITOK));
210 		else
211 			return (uma_zalloc(cache_zone_large, M_WAITOK));
212 	}
213 	if (ts)
214 		return (uma_zalloc(cache_zone_small_ts, M_WAITOK));
215 	else
216 		return (uma_zalloc(cache_zone_small, M_WAITOK));
217 }
218 
219 static void
220 cache_free(struct namecache *ncp)
221 {
222 	int ts;
223 
224 	if (ncp == NULL)
225 		return;
226 	ts = ncp->nc_flag & NCF_TS;
227 	if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) {
228 		if (ts)
229 			uma_zfree(cache_zone_small_ts, ncp);
230 		else
231 			uma_zfree(cache_zone_small, ncp);
232 	} else if (ts)
233 		uma_zfree(cache_zone_large_ts, ncp);
234 	else
235 		uma_zfree(cache_zone_large, ncp);
236 }
237 
238 static char *
239 nc_get_name(struct namecache *ncp)
240 {
241 	struct namecache_ts *ncp_ts;
242 
243 	if ((ncp->nc_flag & NCF_TS) == 0)
244 		return (ncp->nc_name);
245 	ncp_ts = (struct namecache_ts *)ncp;
246 	return (ncp_ts->nc_name);
247 }
248 
249 static void
250 cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp)
251 {
252 
253 	KASSERT((ncp->nc_flag & NCF_TS) != 0 ||
254 	    (tsp == NULL && ticksp == NULL),
255 	    ("No NCF_TS"));
256 
257 	if (tsp != NULL)
258 		*tsp = ((struct namecache_ts *)ncp)->nc_time;
259 	if (ticksp != NULL)
260 		*ticksp = ((struct namecache_ts *)ncp)->nc_ticks;
261 }
262 
263 static int	doingcache = 1;		/* 1 => enable the cache */
264 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
265     "VFS namecache enabled");
266 
267 /* Export size information to userland */
268 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0,
269     sizeof(struct namecache), "sizeof(struct namecache)");
270 
271 /*
272  * The new name cache statistics
273  */
274 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0,
275     "Name cache statistics");
276 #define STATNODE(mode, name, var, descr) \
277 	SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, descr);
278 STATNODE(CTLFLAG_RD, numneg, &numneg, "Number of negative cache entries");
279 STATNODE(CTLFLAG_RD, numcache, &numcache, "Number of cache entries");
280 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls,
281     "Number of cache lookups");
282 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits,
283     "Number of '.' hits");
284 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits,
285     "Number of '..' hits");
286 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks,
287     "Number of checks in lookup");
288 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss,
289     "Number of cache misses");
290 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap,
291     "Number of cache misses we do not want to cache");
292 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps,
293     "Number of cache hits (positive) we do not want to cache");
294 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits,
295     "Number of cache hits (positive)");
296 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps,
297     "Number of cache hits (negative) we do not want to cache");
298 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits,
299     "Number of cache hits (negative)");
300 static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades,
301     "Number of updates of the cache after lookup (write lock + retry)");
302 
303 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
304     &nchstats, sizeof(nchstats), "LU",
305     "VFS cache effectiveness statistics");
306 
307 
308 
309 static void cache_zap(struct namecache *ncp);
310 static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
311     u_int *buflen);
312 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
313     char *buf, char **retbuf, u_int buflen);
314 
315 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
316 
317 #ifdef DIAGNOSTIC
318 /*
319  * Grab an atomic snapshot of the name cache hash chain lengths
320  */
321 static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
322     "hash table stats");
323 
324 static int
325 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
326 {
327 	int error;
328 	struct nchashhead *ncpp;
329 	struct namecache *ncp;
330 	int n_nchash;
331 	int count;
332 
333 	n_nchash = nchash + 1;	/* nchash is max index, not count */
334 	if (!req->oldptr)
335 		return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
336 
337 	/* Scan hash tables for applicable entries */
338 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
339 		CACHE_RLOCK();
340 		count = 0;
341 		LIST_FOREACH(ncp, ncpp, nc_hash) {
342 			count++;
343 		}
344 		CACHE_RUNLOCK();
345 		error = SYSCTL_OUT(req, &count, sizeof(count));
346 		if (error)
347 			return (error);
348 	}
349 	return (0);
350 }
351 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
352     CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
353     "nchash chain lengths");
354 
355 static int
356 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
357 {
358 	int error;
359 	struct nchashhead *ncpp;
360 	struct namecache *ncp;
361 	int n_nchash;
362 	int count, maxlength, used, pct;
363 
364 	if (!req->oldptr)
365 		return SYSCTL_OUT(req, 0, 4 * sizeof(int));
366 
367 	n_nchash = nchash + 1;	/* nchash is max index, not count */
368 	used = 0;
369 	maxlength = 0;
370 
371 	/* Scan hash tables for applicable entries */
372 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
373 		count = 0;
374 		CACHE_RLOCK();
375 		LIST_FOREACH(ncp, ncpp, nc_hash) {
376 			count++;
377 		}
378 		CACHE_RUNLOCK();
379 		if (count)
380 			used++;
381 		if (maxlength < count)
382 			maxlength = count;
383 	}
384 	n_nchash = nchash + 1;
385 	pct = (used * 100) / (n_nchash / 100);
386 	error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
387 	if (error)
388 		return (error);
389 	error = SYSCTL_OUT(req, &used, sizeof(used));
390 	if (error)
391 		return (error);
392 	error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
393 	if (error)
394 		return (error);
395 	error = SYSCTL_OUT(req, &pct, sizeof(pct));
396 	if (error)
397 		return (error);
398 	return (0);
399 }
400 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
401     CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
402     "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)");
403 #endif
404 
405 /*
406  * cache_zap():
407  *
408  *   Removes a namecache entry from cache, whether it contains an actual
409  *   pointer to a vnode or if it is just a negative cache entry.
410  */
411 static void
412 cache_zap(ncp)
413 	struct namecache *ncp;
414 {
415 	struct vnode *vp;
416 
417 	rw_assert(&cache_lock, RA_WLOCKED);
418 	CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
419 #ifdef KDTRACE_HOOKS
420 	if (ncp->nc_vp != NULL) {
421 		SDT_PROBE(vfs, namecache, zap, done, ncp->nc_dvp,
422 		    nc_get_name(ncp), ncp->nc_vp, 0, 0);
423 	} else {
424 		SDT_PROBE(vfs, namecache, zap_negative, done, ncp->nc_dvp,
425 		    nc_get_name(ncp), 0, 0, 0);
426 	}
427 #endif
428 	vp = NULL;
429 	LIST_REMOVE(ncp, nc_hash);
430 	if (ncp->nc_flag & NCF_ISDOTDOT) {
431 		if (ncp == ncp->nc_dvp->v_cache_dd)
432 			ncp->nc_dvp->v_cache_dd = NULL;
433 	} else {
434 		LIST_REMOVE(ncp, nc_src);
435 		if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
436 			vp = ncp->nc_dvp;
437 			numcachehv--;
438 		}
439 	}
440 	if (ncp->nc_vp) {
441 		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
442 		if (ncp == ncp->nc_vp->v_cache_dd)
443 			ncp->nc_vp->v_cache_dd = NULL;
444 	} else {
445 		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
446 		numneg--;
447 	}
448 	numcache--;
449 	cache_free(ncp);
450 	if (vp)
451 		vdrop(vp);
452 }
453 
454 /*
455  * Lookup an entry in the cache
456  *
457  * Lookup is called with dvp pointing to the directory to search,
458  * cnp pointing to the name of the entry being sought. If the lookup
459  * succeeds, the vnode is returned in *vpp, and a status of -1 is
460  * returned. If the lookup determines that the name does not exist
461  * (negative cacheing), a status of ENOENT is returned. If the lookup
462  * fails, a status of zero is returned.  If the directory vnode is
463  * recycled out from under us due to a forced unmount, a status of
464  * ENOENT is returned.
465  *
466  * vpp is locked and ref'd on return.  If we're looking up DOTDOT, dvp is
467  * unlocked.  If we're looking up . an extra ref is taken, but the lock is
468  * not recursively acquired.
469  */
470 
471 int
472 cache_lookup(dvp, vpp, cnp, tsp, ticksp)
473 	struct vnode *dvp;
474 	struct vnode **vpp;
475 	struct componentname *cnp;
476 	struct timespec *tsp;
477 	int *ticksp;
478 {
479 	struct namecache *ncp;
480 	uint32_t hash;
481 	int error, ltype, wlocked;
482 
483 	if (!doingcache) {
484 		cnp->cn_flags &= ~MAKEENTRY;
485 		return (0);
486 	}
487 retry:
488 	CACHE_RLOCK();
489 	wlocked = 0;
490 	numcalls++;
491 	error = 0;
492 
493 retry_wlocked:
494 	if (cnp->cn_nameptr[0] == '.') {
495 		if (cnp->cn_namelen == 1) {
496 			*vpp = dvp;
497 			CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
498 			    dvp, cnp->cn_nameptr);
499 			dothits++;
500 			SDT_PROBE(vfs, namecache, lookup, hit, dvp, ".",
501 			    *vpp, 0, 0);
502 			if (tsp != NULL)
503 				timespecclear(tsp);
504 			if (ticksp != NULL)
505 				*ticksp = ticks;
506 			goto success;
507 		}
508 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
509 			dotdothits++;
510 			if (dvp->v_cache_dd == NULL) {
511 				SDT_PROBE(vfs, namecache, lookup, miss, dvp,
512 				    "..", NULL, 0, 0);
513 				goto unlock;
514 			}
515 			if ((cnp->cn_flags & MAKEENTRY) == 0) {
516 				if (!wlocked && !CACHE_UPGRADE_LOCK())
517 					goto wlock;
518 				if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
519 					cache_zap(dvp->v_cache_dd);
520 				dvp->v_cache_dd = NULL;
521 				CACHE_WUNLOCK();
522 				return (0);
523 			}
524 			ncp = dvp->v_cache_dd;
525 			if (ncp->nc_flag & NCF_ISDOTDOT)
526 				*vpp = ncp->nc_vp;
527 			else
528 				*vpp = ncp->nc_dvp;
529 			/* Return failure if negative entry was found. */
530 			if (*vpp == NULL)
531 				goto negative_success;
532 			CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
533 			    dvp, cnp->cn_nameptr, *vpp);
534 			SDT_PROBE(vfs, namecache, lookup, hit, dvp, "..",
535 			    *vpp, 0, 0);
536 			cache_out_ts(ncp, tsp, ticksp);
537 			if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
538 			    NCF_DTS && tsp != NULL)
539 				*tsp = ((struct namecache_ts *)ncp)->
540 				    nc_dotdottime;
541 			goto success;
542 		}
543 	}
544 
545 	hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
546 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
547 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
548 		numchecks++;
549 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
550 		    !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen))
551 			break;
552 	}
553 
554 	/* We failed to find an entry */
555 	if (ncp == NULL) {
556 		SDT_PROBE(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
557 		    NULL, 0, 0);
558 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
559 			nummisszap++;
560 		} else {
561 			nummiss++;
562 		}
563 		nchstats.ncs_miss++;
564 		goto unlock;
565 	}
566 
567 	/* We don't want to have an entry, so dump it */
568 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
569 		numposzaps++;
570 		nchstats.ncs_badhits++;
571 		if (!wlocked && !CACHE_UPGRADE_LOCK())
572 			goto wlock;
573 		cache_zap(ncp);
574 		CACHE_WUNLOCK();
575 		return (0);
576 	}
577 
578 	/* We found a "positive" match, return the vnode */
579 	if (ncp->nc_vp) {
580 		numposhits++;
581 		nchstats.ncs_goodhits++;
582 		*vpp = ncp->nc_vp;
583 		CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
584 		    dvp, cnp->cn_nameptr, *vpp, ncp);
585 		SDT_PROBE(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp),
586 		    *vpp, 0, 0);
587 		cache_out_ts(ncp, tsp, ticksp);
588 		goto success;
589 	}
590 
591 negative_success:
592 	/* We found a negative match, and want to create it, so purge */
593 	if (cnp->cn_nameiop == CREATE) {
594 		numnegzaps++;
595 		nchstats.ncs_badhits++;
596 		if (!wlocked && !CACHE_UPGRADE_LOCK())
597 			goto wlock;
598 		cache_zap(ncp);
599 		CACHE_WUNLOCK();
600 		return (0);
601 	}
602 
603 	if (!wlocked && !CACHE_UPGRADE_LOCK())
604 		goto wlock;
605 	numneghits++;
606 	/*
607 	 * We found a "negative" match, so we shift it to the end of
608 	 * the "negative" cache entries queue to satisfy LRU.  Also,
609 	 * check to see if the entry is a whiteout; indicate this to
610 	 * the componentname, if so.
611 	 */
612 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
613 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
614 	nchstats.ncs_neghits++;
615 	if (ncp->nc_flag & NCF_WHITE)
616 		cnp->cn_flags |= ISWHITEOUT;
617 	SDT_PROBE(vfs, namecache, lookup, hit_negative, dvp, nc_get_name(ncp),
618 	    0, 0, 0);
619 	cache_out_ts(ncp, tsp, ticksp);
620 	CACHE_WUNLOCK();
621 	return (ENOENT);
622 
623 wlock:
624 	/*
625 	 * We need to update the cache after our lookup, so upgrade to
626 	 * a write lock and retry the operation.
627 	 */
628 	CACHE_RUNLOCK();
629 	CACHE_WLOCK();
630 	numupgrades++;
631 	wlocked = 1;
632 	goto retry_wlocked;
633 
634 success:
635 	/*
636 	 * On success we return a locked and ref'd vnode as per the lookup
637 	 * protocol.
638 	 */
639 	if (dvp == *vpp) {   /* lookup on "." */
640 		VREF(*vpp);
641 		if (wlocked)
642 			CACHE_WUNLOCK();
643 		else
644 			CACHE_RUNLOCK();
645 		/*
646 		 * When we lookup "." we still can be asked to lock it
647 		 * differently...
648 		 */
649 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
650 		if (ltype != VOP_ISLOCKED(*vpp)) {
651 			if (ltype == LK_EXCLUSIVE) {
652 				vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
653 				if ((*vpp)->v_iflag & VI_DOOMED) {
654 					/* forced unmount */
655 					vrele(*vpp);
656 					*vpp = NULL;
657 					return (ENOENT);
658 				}
659 			} else
660 				vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
661 		}
662 		return (-1);
663 	}
664 	ltype = 0;	/* silence gcc warning */
665 	if (cnp->cn_flags & ISDOTDOT) {
666 		ltype = VOP_ISLOCKED(dvp);
667 		VOP_UNLOCK(dvp, 0);
668 	}
669 	VI_LOCK(*vpp);
670 	if (wlocked)
671 		CACHE_WUNLOCK();
672 	else
673 		CACHE_RUNLOCK();
674 	error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
675 	if (cnp->cn_flags & ISDOTDOT) {
676 		vn_lock(dvp, ltype | LK_RETRY);
677 		if (dvp->v_iflag & VI_DOOMED) {
678 			if (error == 0)
679 				vput(*vpp);
680 			*vpp = NULL;
681 			return (ENOENT);
682 		}
683 	}
684 	if (error) {
685 		*vpp = NULL;
686 		goto retry;
687 	}
688 	if ((cnp->cn_flags & ISLASTCN) &&
689 	    (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
690 		ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
691 	}
692 	return (-1);
693 
694 unlock:
695 	if (wlocked)
696 		CACHE_WUNLOCK();
697 	else
698 		CACHE_RUNLOCK();
699 	return (0);
700 }
701 
702 /*
703  * Add an entry to the cache.
704  */
705 void
706 cache_enter_time(dvp, vp, cnp, tsp, dtsp)
707 	struct vnode *dvp;
708 	struct vnode *vp;
709 	struct componentname *cnp;
710 	struct timespec *tsp;
711 	struct timespec *dtsp;
712 {
713 	struct namecache *ncp, *n2;
714 	struct namecache_ts *n3;
715 	struct nchashhead *ncpp;
716 	uint32_t hash;
717 	int flag;
718 	int hold;
719 	int zap;
720 	int len;
721 
722 	CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
723 	VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
724 	    ("cache_enter: Adding a doomed vnode"));
725 	VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
726 	    ("cache_enter: Doomed vnode used as src"));
727 
728 	if (!doingcache)
729 		return;
730 
731 	/*
732 	 * Avoid blowout in namecache entries.
733 	 */
734 	if (numcache >= desiredvnodes * ncsizefactor)
735 		return;
736 
737 	flag = 0;
738 	if (cnp->cn_nameptr[0] == '.') {
739 		if (cnp->cn_namelen == 1)
740 			return;
741 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
742 			CACHE_WLOCK();
743 			/*
744 			 * If dotdot entry already exists, just retarget it
745 			 * to new parent vnode, otherwise continue with new
746 			 * namecache entry allocation.
747 			 */
748 			if ((ncp = dvp->v_cache_dd) != NULL &&
749 			    ncp->nc_flag & NCF_ISDOTDOT) {
750 				KASSERT(ncp->nc_dvp == dvp,
751 				    ("wrong isdotdot parent"));
752 				if (ncp->nc_vp != NULL)
753 					TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
754 					    ncp, nc_dst);
755 				else
756 					TAILQ_REMOVE(&ncneg, ncp, nc_dst);
757 				if (vp != NULL)
758 					TAILQ_INSERT_HEAD(&vp->v_cache_dst,
759 					    ncp, nc_dst);
760 				else
761 					TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
762 				ncp->nc_vp = vp;
763 				CACHE_WUNLOCK();
764 				return;
765 			}
766 			dvp->v_cache_dd = NULL;
767 			SDT_PROBE(vfs, namecache, enter, done, dvp, "..", vp,
768 			    0, 0);
769 			CACHE_WUNLOCK();
770 			flag = NCF_ISDOTDOT;
771 		}
772 	}
773 
774 	hold = 0;
775 	zap = 0;
776 
777 	/*
778 	 * Calculate the hash key and setup as much of the new
779 	 * namecache entry as possible before acquiring the lock.
780 	 */
781 	ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
782 	ncp->nc_vp = vp;
783 	ncp->nc_dvp = dvp;
784 	ncp->nc_flag = flag;
785 	if (tsp != NULL) {
786 		n3 = (struct namecache_ts *)ncp;
787 		n3->nc_time = *tsp;
788 		n3->nc_ticks = ticks;
789 		n3->nc_flag |= NCF_TS;
790 		if (dtsp != NULL) {
791 			n3->nc_dotdottime = *dtsp;
792 			n3->nc_flag |= NCF_DTS;
793 		}
794 	}
795 	len = ncp->nc_nlen = cnp->cn_namelen;
796 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
797 	strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1);
798 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
799 	CACHE_WLOCK();
800 
801 	/*
802 	 * See if this vnode or negative entry is already in the cache
803 	 * with this name.  This can happen with concurrent lookups of
804 	 * the same path name.
805 	 */
806 	ncpp = NCHHASH(hash);
807 	LIST_FOREACH(n2, ncpp, nc_hash) {
808 		if (n2->nc_dvp == dvp &&
809 		    n2->nc_nlen == cnp->cn_namelen &&
810 		    !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) {
811 			if (tsp != NULL) {
812 				KASSERT((n2->nc_flag & NCF_TS) != 0,
813 				    ("no NCF_TS"));
814 				n3 = (struct namecache_ts *)n2;
815 				n3->nc_time =
816 				    ((struct namecache_ts *)ncp)->nc_time;
817 				n3->nc_ticks =
818 				    ((struct namecache_ts *)ncp)->nc_ticks;
819 				if (dtsp != NULL) {
820 					n3->nc_dotdottime =
821 					    ((struct namecache_ts *)ncp)->
822 					    nc_dotdottime;
823 					n3->nc_flag |= NCF_DTS;
824 				}
825 			}
826 			CACHE_WUNLOCK();
827 			cache_free(ncp);
828 			return;
829 		}
830 	}
831 
832 	if (flag == NCF_ISDOTDOT) {
833 		/*
834 		 * See if we are trying to add .. entry, but some other lookup
835 		 * has populated v_cache_dd pointer already.
836 		 */
837 		if (dvp->v_cache_dd != NULL) {
838 		    CACHE_WUNLOCK();
839 		    cache_free(ncp);
840 		    return;
841 		}
842 		KASSERT(vp == NULL || vp->v_type == VDIR,
843 		    ("wrong vnode type %p", vp));
844 		dvp->v_cache_dd = ncp;
845 	}
846 
847 	numcache++;
848 	if (!vp) {
849 		numneg++;
850 		if (cnp->cn_flags & ISWHITEOUT)
851 			ncp->nc_flag |= NCF_WHITE;
852 	} else if (vp->v_type == VDIR) {
853 		if (flag != NCF_ISDOTDOT) {
854 			/*
855 			 * For this case, the cache entry maps both the
856 			 * directory name in it and the name ".." for the
857 			 * directory's parent.
858 			 */
859 			if ((n2 = vp->v_cache_dd) != NULL &&
860 			    (n2->nc_flag & NCF_ISDOTDOT) != 0)
861 				cache_zap(n2);
862 			vp->v_cache_dd = ncp;
863 		}
864 	} else {
865 		vp->v_cache_dd = NULL;
866 	}
867 
868 	/*
869 	 * Insert the new namecache entry into the appropriate chain
870 	 * within the cache entries table.
871 	 */
872 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
873 	if (flag != NCF_ISDOTDOT) {
874 		if (LIST_EMPTY(&dvp->v_cache_src)) {
875 			hold = 1;
876 			numcachehv++;
877 		}
878 		LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
879 	}
880 
881 	/*
882 	 * If the entry is "negative", we place it into the
883 	 * "negative" cache queue, otherwise, we place it into the
884 	 * destination vnode's cache entries queue.
885 	 */
886 	if (vp) {
887 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
888 		SDT_PROBE(vfs, namecache, enter, done, dvp, nc_get_name(ncp),
889 		    vp, 0, 0);
890 	} else {
891 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
892 		SDT_PROBE(vfs, namecache, enter_negative, done, dvp,
893 		    nc_get_name(ncp), 0, 0, 0);
894 	}
895 	if (numneg * ncnegfactor > numcache) {
896 		ncp = TAILQ_FIRST(&ncneg);
897 		zap = 1;
898 	}
899 	if (hold)
900 		vhold(dvp);
901 	if (zap)
902 		cache_zap(ncp);
903 	CACHE_WUNLOCK();
904 }
905 
906 /*
907  * Name cache initialization, from vfs_init() when we are booting
908  */
909 static void
910 nchinit(void *dummy __unused)
911 {
912 
913 	TAILQ_INIT(&ncneg);
914 
915 	cache_zone_small = uma_zcreate("S VFS Cache",
916 	    sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
917 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
918 	cache_zone_small_ts = uma_zcreate("STS VFS Cache",
919 	    sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
920 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
921 	cache_zone_large = uma_zcreate("L VFS Cache",
922 	    sizeof(struct namecache) + NAME_MAX + 1,
923 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
924 	cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
925 	    sizeof(struct namecache_ts) + NAME_MAX + 1,
926 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
927 
928 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
929 }
930 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
931 
932 
933 /*
934  * Invalidate all entries to a particular vnode.
935  */
936 void
937 cache_purge(vp)
938 	struct vnode *vp;
939 {
940 
941 	CTR1(KTR_VFS, "cache_purge(%p)", vp);
942 	SDT_PROBE(vfs, namecache, purge, done, vp, 0, 0, 0, 0);
943 	CACHE_WLOCK();
944 	while (!LIST_EMPTY(&vp->v_cache_src))
945 		cache_zap(LIST_FIRST(&vp->v_cache_src));
946 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
947 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
948 	if (vp->v_cache_dd != NULL) {
949 		KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
950 		   ("lost dotdot link"));
951 		cache_zap(vp->v_cache_dd);
952 	}
953 	KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
954 	CACHE_WUNLOCK();
955 }
956 
957 /*
958  * Invalidate all negative entries for a particular directory vnode.
959  */
960 void
961 cache_purge_negative(vp)
962 	struct vnode *vp;
963 {
964 	struct namecache *cp, *ncp;
965 
966 	CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
967 	SDT_PROBE(vfs, namecache, purge_negative, done, vp, 0, 0, 0, 0);
968 	CACHE_WLOCK();
969 	LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
970 		if (cp->nc_vp == NULL)
971 			cache_zap(cp);
972 	}
973 	CACHE_WUNLOCK();
974 }
975 
976 /*
977  * Flush all entries referencing a particular filesystem.
978  */
979 void
980 cache_purgevfs(mp)
981 	struct mount *mp;
982 {
983 	struct nchashhead *ncpp;
984 	struct namecache *ncp, *nnp;
985 
986 	/* Scan hash tables for applicable entries */
987 	SDT_PROBE(vfs, namecache, purgevfs, done, mp, 0, 0, 0, 0);
988 	CACHE_WLOCK();
989 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
990 		LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
991 			if (ncp->nc_dvp->v_mount == mp)
992 				cache_zap(ncp);
993 		}
994 	}
995 	CACHE_WUNLOCK();
996 }
997 
998 /*
999  * Perform canonical checks and cache lookup and pass on to filesystem
1000  * through the vop_cachedlookup only if needed.
1001  */
1002 
1003 int
1004 vfs_cache_lookup(ap)
1005 	struct vop_lookup_args /* {
1006 		struct vnode *a_dvp;
1007 		struct vnode **a_vpp;
1008 		struct componentname *a_cnp;
1009 	} */ *ap;
1010 {
1011 	struct vnode *dvp;
1012 	int error;
1013 	struct vnode **vpp = ap->a_vpp;
1014 	struct componentname *cnp = ap->a_cnp;
1015 	struct ucred *cred = cnp->cn_cred;
1016 	int flags = cnp->cn_flags;
1017 	struct thread *td = cnp->cn_thread;
1018 
1019 	*vpp = NULL;
1020 	dvp = ap->a_dvp;
1021 
1022 	if (dvp->v_type != VDIR)
1023 		return (ENOTDIR);
1024 
1025 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1026 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1027 		return (EROFS);
1028 
1029 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
1030 	if (error)
1031 		return (error);
1032 
1033 	error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
1034 	if (error == 0)
1035 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
1036 	if (error == -1)
1037 		return (0);
1038 	return (error);
1039 }
1040 
1041 
1042 #ifndef _SYS_SYSPROTO_H_
1043 struct  __getcwd_args {
1044 	u_char	*buf;
1045 	u_int	buflen;
1046 };
1047 #endif
1048 
1049 /*
1050  * XXX All of these sysctls would probably be more productive dead.
1051  */
1052 static int disablecwd;
1053 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
1054    "Disable the getcwd syscall");
1055 
1056 /* Implementation of the getcwd syscall. */
1057 int
1058 sys___getcwd(td, uap)
1059 	struct thread *td;
1060 	struct __getcwd_args *uap;
1061 {
1062 
1063 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
1064 }
1065 
1066 int
1067 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
1068 {
1069 	char *bp, *tmpbuf;
1070 	struct filedesc *fdp;
1071 	struct vnode *cdir, *rdir;
1072 	int error;
1073 
1074 	if (disablecwd)
1075 		return (ENODEV);
1076 	if (buflen < 2)
1077 		return (EINVAL);
1078 	if (buflen > MAXPATHLEN)
1079 		buflen = MAXPATHLEN;
1080 
1081 	tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
1082 	fdp = td->td_proc->p_fd;
1083 	FILEDESC_SLOCK(fdp);
1084 	cdir = fdp->fd_cdir;
1085 	VREF(cdir);
1086 	rdir = fdp->fd_rdir;
1087 	VREF(rdir);
1088 	FILEDESC_SUNLOCK(fdp);
1089 	error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
1090 	vrele(rdir);
1091 	vrele(cdir);
1092 
1093 	if (!error) {
1094 		if (bufseg == UIO_SYSSPACE)
1095 			bcopy(bp, buf, strlen(bp) + 1);
1096 		else
1097 			error = copyout(bp, buf, strlen(bp) + 1);
1098 #ifdef KTRACE
1099 	if (KTRPOINT(curthread, KTR_NAMEI))
1100 		ktrnamei(bp);
1101 #endif
1102 	}
1103 	free(tmpbuf, M_TEMP);
1104 	return (error);
1105 }
1106 
1107 /*
1108  * Thus begins the fullpath magic.
1109  */
1110 
1111 #undef STATNODE
1112 #define STATNODE(name, descr)						\
1113 	static u_int name;						\
1114 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr)
1115 
1116 static int disablefullpath;
1117 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
1118     "Disable the vn_fullpath function");
1119 
1120 /* These count for kern___getcwd(), too. */
1121 STATNODE(numfullpathcalls, "Number of fullpath search calls");
1122 STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
1123 STATNODE(numfullpathfail2,
1124     "Number of fullpath search errors (VOP_VPTOCNP failures)");
1125 STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
1126 STATNODE(numfullpathfound, "Number of successful fullpath calls");
1127 
1128 /*
1129  * Retrieve the full filesystem path that correspond to a vnode from the name
1130  * cache (if available)
1131  */
1132 int
1133 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
1134 {
1135 	char *buf;
1136 	struct filedesc *fdp;
1137 	struct vnode *rdir;
1138 	int error;
1139 
1140 	if (disablefullpath)
1141 		return (ENODEV);
1142 	if (vn == NULL)
1143 		return (EINVAL);
1144 
1145 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1146 	fdp = td->td_proc->p_fd;
1147 	FILEDESC_SLOCK(fdp);
1148 	rdir = fdp->fd_rdir;
1149 	VREF(rdir);
1150 	FILEDESC_SUNLOCK(fdp);
1151 	error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1152 	vrele(rdir);
1153 
1154 	if (!error)
1155 		*freebuf = buf;
1156 	else
1157 		free(buf, M_TEMP);
1158 	return (error);
1159 }
1160 
1161 /*
1162  * This function is similar to vn_fullpath, but it attempts to lookup the
1163  * pathname relative to the global root mount point.  This is required for the
1164  * auditing sub-system, as audited pathnames must be absolute, relative to the
1165  * global root mount point.
1166  */
1167 int
1168 vn_fullpath_global(struct thread *td, struct vnode *vn,
1169     char **retbuf, char **freebuf)
1170 {
1171 	char *buf;
1172 	int error;
1173 
1174 	if (disablefullpath)
1175 		return (ENODEV);
1176 	if (vn == NULL)
1177 		return (EINVAL);
1178 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1179 	error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1180 	if (!error)
1181 		*freebuf = buf;
1182 	else
1183 		free(buf, M_TEMP);
1184 	return (error);
1185 }
1186 
1187 int
1188 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1189 {
1190 	int error;
1191 
1192 	CACHE_RLOCK();
1193 	error = vn_vptocnp_locked(vp, cred, buf, buflen);
1194 	if (error == 0)
1195 		CACHE_RUNLOCK();
1196 	return (error);
1197 }
1198 
1199 static int
1200 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1201     u_int *buflen)
1202 {
1203 	struct vnode *dvp;
1204 	struct namecache *ncp;
1205 	int error;
1206 
1207 	TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1208 		if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1209 			break;
1210 	}
1211 	if (ncp != NULL) {
1212 		if (*buflen < ncp->nc_nlen) {
1213 			CACHE_RUNLOCK();
1214 			vrele(*vp);
1215 			numfullpathfail4++;
1216 			error = ENOMEM;
1217 			SDT_PROBE(vfs, namecache, fullpath, return, error,
1218 			    vp, NULL, 0, 0);
1219 			return (error);
1220 		}
1221 		*buflen -= ncp->nc_nlen;
1222 		memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen);
1223 		SDT_PROBE(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1224 		    nc_get_name(ncp), vp, 0, 0);
1225 		dvp = *vp;
1226 		*vp = ncp->nc_dvp;
1227 		vref(*vp);
1228 		CACHE_RUNLOCK();
1229 		vrele(dvp);
1230 		CACHE_RLOCK();
1231 		return (0);
1232 	}
1233 	SDT_PROBE(vfs, namecache, fullpath, miss, vp, 0, 0, 0, 0);
1234 
1235 	CACHE_RUNLOCK();
1236 	vn_lock(*vp, LK_SHARED | LK_RETRY);
1237 	error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1238 	vput(*vp);
1239 	if (error) {
1240 		numfullpathfail2++;
1241 		SDT_PROBE(vfs, namecache, fullpath, return,  error, vp,
1242 		    NULL, 0, 0);
1243 		return (error);
1244 	}
1245 
1246 	*vp = dvp;
1247 	CACHE_RLOCK();
1248 	if (dvp->v_iflag & VI_DOOMED) {
1249 		/* forced unmount */
1250 		CACHE_RUNLOCK();
1251 		vrele(dvp);
1252 		error = ENOENT;
1253 		SDT_PROBE(vfs, namecache, fullpath, return, error, vp,
1254 		    NULL, 0, 0);
1255 		return (error);
1256 	}
1257 	/*
1258 	 * *vp has its use count incremented still.
1259 	 */
1260 
1261 	return (0);
1262 }
1263 
1264 /*
1265  * The magic behind kern___getcwd() and vn_fullpath().
1266  */
1267 static int
1268 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1269     char *buf, char **retbuf, u_int buflen)
1270 {
1271 	int error, slash_prefixed;
1272 #ifdef KDTRACE_HOOKS
1273 	struct vnode *startvp = vp;
1274 #endif
1275 	struct vnode *vp1;
1276 
1277 	buflen--;
1278 	buf[buflen] = '\0';
1279 	error = 0;
1280 	slash_prefixed = 0;
1281 
1282 	SDT_PROBE(vfs, namecache, fullpath, entry, vp, 0, 0, 0, 0);
1283 	numfullpathcalls++;
1284 	vref(vp);
1285 	CACHE_RLOCK();
1286 	if (vp->v_type != VDIR) {
1287 		error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1288 		if (error)
1289 			return (error);
1290 		if (buflen == 0) {
1291 			CACHE_RUNLOCK();
1292 			vrele(vp);
1293 			return (ENOMEM);
1294 		}
1295 		buf[--buflen] = '/';
1296 		slash_prefixed = 1;
1297 	}
1298 	while (vp != rdir && vp != rootvnode) {
1299 		if (vp->v_vflag & VV_ROOT) {
1300 			if (vp->v_iflag & VI_DOOMED) {	/* forced unmount */
1301 				CACHE_RUNLOCK();
1302 				vrele(vp);
1303 				error = ENOENT;
1304 				SDT_PROBE(vfs, namecache, fullpath, return,
1305 				    error, vp, NULL, 0, 0);
1306 				break;
1307 			}
1308 			vp1 = vp->v_mount->mnt_vnodecovered;
1309 			vref(vp1);
1310 			CACHE_RUNLOCK();
1311 			vrele(vp);
1312 			vp = vp1;
1313 			CACHE_RLOCK();
1314 			continue;
1315 		}
1316 		if (vp->v_type != VDIR) {
1317 			CACHE_RUNLOCK();
1318 			vrele(vp);
1319 			numfullpathfail1++;
1320 			error = ENOTDIR;
1321 			SDT_PROBE(vfs, namecache, fullpath, return,
1322 			    error, vp, NULL, 0, 0);
1323 			break;
1324 		}
1325 		error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1326 		if (error)
1327 			break;
1328 		if (buflen == 0) {
1329 			CACHE_RUNLOCK();
1330 			vrele(vp);
1331 			error = ENOMEM;
1332 			SDT_PROBE(vfs, namecache, fullpath, return, error,
1333 			    startvp, NULL, 0, 0);
1334 			break;
1335 		}
1336 		buf[--buflen] = '/';
1337 		slash_prefixed = 1;
1338 	}
1339 	if (error)
1340 		return (error);
1341 	if (!slash_prefixed) {
1342 		if (buflen == 0) {
1343 			CACHE_RUNLOCK();
1344 			vrele(vp);
1345 			numfullpathfail4++;
1346 			SDT_PROBE(vfs, namecache, fullpath, return, ENOMEM,
1347 			    startvp, NULL, 0, 0);
1348 			return (ENOMEM);
1349 		}
1350 		buf[--buflen] = '/';
1351 	}
1352 	numfullpathfound++;
1353 	CACHE_RUNLOCK();
1354 	vrele(vp);
1355 
1356 	SDT_PROBE(vfs, namecache, fullpath, return, 0, startvp, buf + buflen,
1357 	    0, 0);
1358 	*retbuf = buf + buflen;
1359 	return (0);
1360 }
1361 
1362 int
1363 vn_commname(struct vnode *vp, char *buf, u_int buflen)
1364 {
1365 	struct namecache *ncp;
1366 	int l;
1367 
1368 	CACHE_RLOCK();
1369 	TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1370 		if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1371 			break;
1372 	if (ncp == NULL) {
1373 		CACHE_RUNLOCK();
1374 		return (ENOENT);
1375 	}
1376 	l = min(ncp->nc_nlen, buflen - 1);
1377 	memcpy(buf, nc_get_name(ncp), l);
1378 	CACHE_RUNLOCK();
1379 	buf[l] = '\0';
1380 	return (0);
1381 }
1382 
1383 /* ABI compat shims for old kernel modules. */
1384 #undef cache_enter
1385 
1386 void	cache_enter(struct vnode *dvp, struct vnode *vp,
1387 	    struct componentname *cnp);
1388 
1389 void
1390 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
1391 {
1392 
1393 	cache_enter_time(dvp, vp, cnp, NULL, NULL);
1394 }
1395 
1396 /*
1397  * This function updates path string to vnode's full global path
1398  * and checks the size of the new path string against the pathlen argument.
1399  *
1400  * Requires a locked, referenced vnode and GIANT lock held.
1401  * Vnode is re-locked on success or ENODEV, otherwise unlocked.
1402  *
1403  * If sysctl debug.disablefullpath is set, ENODEV is returned,
1404  * vnode is left locked and path remain untouched.
1405  *
1406  * If vp is a directory, the call to vn_fullpath_global() always succeeds
1407  * because it falls back to the ".." lookup if the namecache lookup fails.
1408  */
1409 int
1410 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
1411     u_int pathlen)
1412 {
1413 	struct nameidata nd;
1414 	struct vnode *vp1;
1415 	char *rpath, *fbuf;
1416 	int error;
1417 
1418 	ASSERT_VOP_ELOCKED(vp, __func__);
1419 
1420 	/* Return ENODEV if sysctl debug.disablefullpath==1 */
1421 	if (disablefullpath)
1422 		return (ENODEV);
1423 
1424 	/* Construct global filesystem path from vp. */
1425 	VOP_UNLOCK(vp, 0);
1426 	error = vn_fullpath_global(td, vp, &rpath, &fbuf);
1427 
1428 	if (error != 0) {
1429 		vrele(vp);
1430 		return (error);
1431 	}
1432 
1433 	if (strlen(rpath) >= pathlen) {
1434 		vrele(vp);
1435 		error = ENAMETOOLONG;
1436 		goto out;
1437 	}
1438 
1439 	/*
1440 	 * Re-lookup the vnode by path to detect a possible rename.
1441 	 * As a side effect, the vnode is relocked.
1442 	 * If vnode was renamed, return ENOENT.
1443 	 */
1444 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1445 	    UIO_SYSSPACE, path, td);
1446 	error = namei(&nd);
1447 	if (error != 0) {
1448 		vrele(vp);
1449 		goto out;
1450 	}
1451 	NDFREE(&nd, NDF_ONLY_PNBUF);
1452 	vp1 = nd.ni_vp;
1453 	vrele(vp);
1454 	if (vp1 == vp)
1455 		strcpy(path, rpath);
1456 	else {
1457 		vput(vp1);
1458 		error = ENOENT;
1459 	}
1460 
1461 out:
1462 	free(fbuf, M_TEMP);
1463 	return (error);
1464 }
1465