xref: /freebsd/sys/kern/vfs_cache.c (revision c9dbb1cc52b063bbd9ab078a7afc89a8696da659)
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_ktrace.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/filedesc.h>
43 #include <sys/fnv_hash.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/fcntl.h>
48 #include <sys/mount.h>
49 #include <sys/namei.h>
50 #include <sys/proc.h>
51 #include <sys/rwlock.h>
52 #include <sys/sdt.h>
53 #include <sys/syscallsubr.h>
54 #include <sys/sysctl.h>
55 #include <sys/sysproto.h>
56 #include <sys/vnode.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60 
61 #include <vm/uma.h>
62 
63 SDT_PROVIDER_DECLARE(vfs);
64 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
65     "struct vnode *");
66 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
67     "char *");
68 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *");
69 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *",
70     "char *", "struct vnode *");
71 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *");
72 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int",
73     "struct vnode *", "char *");
74 SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *",
75     "struct vnode *");
76 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative,
77     "struct vnode *", "char *");
78 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
79     "char *");
80 SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *");
81 SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *");
82 SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *");
83 SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *",
84     "struct vnode *");
85 SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *",
86     "char *");
87 
88 /*
89  * This structure describes the elements in the cache of recent
90  * names looked up by namei.
91  */
92 
93 struct	namecache {
94 	LIST_ENTRY(namecache) nc_hash;	/* hash chain */
95 	LIST_ENTRY(namecache) nc_src;	/* source vnode list */
96 	TAILQ_ENTRY(namecache) nc_dst;	/* destination vnode list */
97 	struct	vnode *nc_dvp;		/* vnode of parent of name */
98 	struct	vnode *nc_vp;		/* vnode the name refers to */
99 	u_char	nc_flag;		/* flag bits */
100 	u_char	nc_nlen;		/* length of name */
101 	char	nc_name[0];		/* segment name + nul */
102 };
103 
104 /*
105  * struct namecache_ts repeats struct namecache layout up to the
106  * nc_nlen member.
107  * struct namecache_ts is used in place of struct namecache when time(s) need
108  * to be stored.  The nc_dotdottime field is used when a cache entry is mapping
109  * both a non-dotdot directory name plus dotdot for the directory's
110  * parent.
111  */
112 struct	namecache_ts {
113 	LIST_ENTRY(namecache) nc_hash;	/* hash chain */
114 	LIST_ENTRY(namecache) nc_src;	/* source vnode list */
115 	TAILQ_ENTRY(namecache) nc_dst;	/* destination vnode list */
116 	struct	vnode *nc_dvp;		/* vnode of parent of name */
117 	struct	vnode *nc_vp;		/* vnode the name refers to */
118 	u_char	nc_flag;		/* flag bits */
119 	u_char	nc_nlen;		/* length of name */
120 	struct	timespec nc_time;	/* timespec provided by fs */
121 	struct	timespec nc_dotdottime;	/* dotdot timespec provided by fs */
122 	int	nc_ticks;		/* ticks value when entry was added */
123 	char	nc_name[0];		/* segment name + nul */
124 };
125 
126 /*
127  * Flags in namecache.nc_flag
128  */
129 #define NCF_WHITE	0x01
130 #define NCF_ISDOTDOT	0x02
131 #define	NCF_TS		0x04
132 #define	NCF_DTS		0x08
133 
134 /*
135  * Name caching works as follows:
136  *
137  * Names found by directory scans are retained in a cache
138  * for future reference.  It is managed LRU, so frequently
139  * used names will hang around.  Cache is indexed by hash value
140  * obtained from (vp, name) where vp refers to the directory
141  * containing name.
142  *
143  * If it is a "negative" entry, (i.e. for a name that is known NOT to
144  * exist) the vnode pointer will be NULL.
145  *
146  * Upon reaching the last segment of a path, if the reference
147  * is for DELETE, or NOCACHE is set (rewrite), and the
148  * name is located in the cache, it will be dropped.
149  */
150 
151 /*
152  * Structures associated with name cacheing.
153  */
154 #define NCHHASH(hash) \
155 	(&nchashtbl[(hash) & nchash])
156 static LIST_HEAD(nchashhead, namecache) *nchashtbl;	/* Hash Table */
157 static TAILQ_HEAD(, namecache) ncneg;	/* Hash Table */
158 static u_long	nchash;			/* size of hash table */
159 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0,
160     "Size of namecache hash table");
161 static u_long	ncnegfactor = 16;	/* ratio of negative entries */
162 SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0,
163     "Ratio of negative namecache entries");
164 static u_long	numneg;			/* number of negative entries allocated */
165 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0,
166     "Number of negative entries in namecache");
167 static u_long	numcache;		/* number of cache entries allocated */
168 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0,
169     "Number of namecache entries");
170 static u_long	numcachehv;		/* number of cache entries with vnodes held */
171 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0,
172     "Number of namecache entries with vnodes held");
173 static u_int	ncsizefactor = 2;
174 SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0,
175     "Size factor for namecache");
176 
177 struct nchstats	nchstats;		/* cache effectiveness statistics */
178 
179 static struct rwlock cache_lock;
180 RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
181 
182 #define	CACHE_UPGRADE_LOCK()	rw_try_upgrade(&cache_lock)
183 #define	CACHE_RLOCK()		rw_rlock(&cache_lock)
184 #define	CACHE_RUNLOCK()		rw_runlock(&cache_lock)
185 #define	CACHE_WLOCK()		rw_wlock(&cache_lock)
186 #define	CACHE_WUNLOCK()		rw_wunlock(&cache_lock)
187 
188 /*
189  * UMA zones for the VFS cache.
190  *
191  * The small cache is used for entries with short names, which are the
192  * most common.  The large cache is used for entries which are too big to
193  * fit in the small cache.
194  */
195 static uma_zone_t cache_zone_small;
196 static uma_zone_t cache_zone_small_ts;
197 static uma_zone_t cache_zone_large;
198 static uma_zone_t cache_zone_large_ts;
199 
200 #define	CACHE_PATH_CUTOFF	35
201 
202 static struct namecache *
203 cache_alloc(int len, int ts)
204 {
205 
206 	if (len > CACHE_PATH_CUTOFF) {
207 		if (ts)
208 			return (uma_zalloc(cache_zone_large_ts, M_WAITOK));
209 		else
210 			return (uma_zalloc(cache_zone_large, M_WAITOK));
211 	}
212 	if (ts)
213 		return (uma_zalloc(cache_zone_small_ts, M_WAITOK));
214 	else
215 		return (uma_zalloc(cache_zone_small, M_WAITOK));
216 }
217 
218 static void
219 cache_free(struct namecache *ncp)
220 {
221 	int ts;
222 
223 	if (ncp == NULL)
224 		return;
225 	ts = ncp->nc_flag & NCF_TS;
226 	if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) {
227 		if (ts)
228 			uma_zfree(cache_zone_small_ts, ncp);
229 		else
230 			uma_zfree(cache_zone_small, ncp);
231 	} else if (ts)
232 		uma_zfree(cache_zone_large_ts, ncp);
233 	else
234 		uma_zfree(cache_zone_large, ncp);
235 }
236 
237 static char *
238 nc_get_name(struct namecache *ncp)
239 {
240 	struct namecache_ts *ncp_ts;
241 
242 	if ((ncp->nc_flag & NCF_TS) == 0)
243 		return (ncp->nc_name);
244 	ncp_ts = (struct namecache_ts *)ncp;
245 	return (ncp_ts->nc_name);
246 }
247 
248 static void
249 cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp)
250 {
251 
252 	KASSERT((ncp->nc_flag & NCF_TS) != 0 ||
253 	    (tsp == NULL && ticksp == NULL),
254 	    ("No NCF_TS"));
255 
256 	if (tsp != NULL)
257 		*tsp = ((struct namecache_ts *)ncp)->nc_time;
258 	if (ticksp != NULL)
259 		*ticksp = ((struct namecache_ts *)ncp)->nc_ticks;
260 }
261 
262 static int	doingcache = 1;		/* 1 => enable the cache */
263 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
264     "VFS namecache enabled");
265 
266 /* Export size information to userland */
267 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR,
268     sizeof(struct namecache), "sizeof(struct namecache)");
269 
270 /*
271  * The new name cache statistics
272  */
273 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0,
274     "Name cache statistics");
275 #define STATNODE(mode, name, var, descr) \
276 	SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, descr);
277 STATNODE(CTLFLAG_RD, numneg, &numneg, "Number of negative cache entries");
278 STATNODE(CTLFLAG_RD, numcache, &numcache, "Number of cache entries");
279 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls,
280     "Number of cache lookups");
281 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits,
282     "Number of '.' hits");
283 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits,
284     "Number of '..' hits");
285 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks,
286     "Number of checks in lookup");
287 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss,
288     "Number of cache misses");
289 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap,
290     "Number of cache misses we do not want to cache");
291 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps,
292     "Number of cache hits (positive) we do not want to cache");
293 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits,
294     "Number of cache hits (positive)");
295 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps,
296     "Number of cache hits (negative) we do not want to cache");
297 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits,
298     "Number of cache hits (negative)");
299 static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades,
300     "Number of updates of the cache after lookup (write lock + retry)");
301 
302 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
303     &nchstats, sizeof(nchstats), "LU",
304     "VFS cache effectiveness statistics");
305 
306 
307 
308 static void cache_zap(struct namecache *ncp);
309 static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
310     u_int *buflen);
311 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
312     char *buf, char **retbuf, u_int buflen);
313 
314 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
315 
316 #ifdef DIAGNOSTIC
317 /*
318  * Grab an atomic snapshot of the name cache hash chain lengths
319  */
320 static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
321     "hash table stats");
322 
323 static int
324 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
325 {
326 	struct nchashhead *ncpp;
327 	struct namecache *ncp;
328 	int i, error, n_nchash, *cntbuf;
329 
330 retry:
331 	n_nchash = nchash + 1;	/* nchash is max index, not count */
332 	if (req->oldptr == NULL)
333 		return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
334 	cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK);
335 	CACHE_RLOCK();
336 	if (n_nchash != nchash + 1) {
337 		CACHE_RUNLOCK();
338 		free(cntbuf, M_TEMP);
339 		goto retry;
340 	}
341 	/* Scan hash tables counting entries */
342 	for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++)
343 		LIST_FOREACH(ncp, ncpp, nc_hash)
344 			cntbuf[i]++;
345 	CACHE_RUNLOCK();
346 	for (error = 0, i = 0; i < n_nchash; i++)
347 		if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0)
348 			break;
349 	free(cntbuf, M_TEMP);
350 	return (error);
351 }
352 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
353     CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
354     "nchash chain lengths");
355 
356 static int
357 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
358 {
359 	int error;
360 	struct nchashhead *ncpp;
361 	struct namecache *ncp;
362 	int n_nchash;
363 	int count, maxlength, used, pct;
364 
365 	if (!req->oldptr)
366 		return SYSCTL_OUT(req, 0, 4 * sizeof(int));
367 
368 	CACHE_RLOCK();
369 	n_nchash = nchash + 1;	/* nchash is max index, not count */
370 	used = 0;
371 	maxlength = 0;
372 
373 	/* Scan hash tables for applicable entries */
374 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
375 		count = 0;
376 		LIST_FOREACH(ncp, ncpp, nc_hash) {
377 			count++;
378 		}
379 		if (count)
380 			used++;
381 		if (maxlength < count)
382 			maxlength = count;
383 	}
384 	n_nchash = nchash + 1;
385 	CACHE_RUNLOCK();
386 	pct = (used * 100) / (n_nchash / 100);
387 	error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
388 	if (error)
389 		return (error);
390 	error = SYSCTL_OUT(req, &used, sizeof(used));
391 	if (error)
392 		return (error);
393 	error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
394 	if (error)
395 		return (error);
396 	error = SYSCTL_OUT(req, &pct, sizeof(pct));
397 	if (error)
398 		return (error);
399 	return (0);
400 }
401 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
402     CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
403     "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)");
404 #endif
405 
406 /*
407  * cache_zap():
408  *
409  *   Removes a namecache entry from cache, whether it contains an actual
410  *   pointer to a vnode or if it is just a negative cache entry.
411  */
412 static void
413 cache_zap(ncp)
414 	struct namecache *ncp;
415 {
416 	struct vnode *vp;
417 
418 	rw_assert(&cache_lock, RA_WLOCKED);
419 	CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
420 #ifdef KDTRACE_HOOKS
421 	if (ncp->nc_vp != NULL) {
422 		SDT_PROBE(vfs, namecache, zap, done, ncp->nc_dvp,
423 		    nc_get_name(ncp), ncp->nc_vp, 0, 0);
424 	} else {
425 		SDT_PROBE(vfs, namecache, zap_negative, done, ncp->nc_dvp,
426 		    nc_get_name(ncp), 0, 0, 0);
427 	}
428 #endif
429 	vp = NULL;
430 	LIST_REMOVE(ncp, nc_hash);
431 	if (ncp->nc_flag & NCF_ISDOTDOT) {
432 		if (ncp == ncp->nc_dvp->v_cache_dd)
433 			ncp->nc_dvp->v_cache_dd = NULL;
434 	} else {
435 		LIST_REMOVE(ncp, nc_src);
436 		if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
437 			vp = ncp->nc_dvp;
438 			numcachehv--;
439 		}
440 	}
441 	if (ncp->nc_vp) {
442 		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
443 		if (ncp == ncp->nc_vp->v_cache_dd)
444 			ncp->nc_vp->v_cache_dd = NULL;
445 	} else {
446 		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
447 		numneg--;
448 	}
449 	numcache--;
450 	cache_free(ncp);
451 	if (vp)
452 		vdrop(vp);
453 }
454 
455 /*
456  * Lookup an entry in the cache
457  *
458  * Lookup is called with dvp pointing to the directory to search,
459  * cnp pointing to the name of the entry being sought. If the lookup
460  * succeeds, the vnode is returned in *vpp, and a status of -1 is
461  * returned. If the lookup determines that the name does not exist
462  * (negative cacheing), a status of ENOENT is returned. If the lookup
463  * fails, a status of zero is returned.  If the directory vnode is
464  * recycled out from under us due to a forced unmount, a status of
465  * ENOENT is returned.
466  *
467  * vpp is locked and ref'd on return.  If we're looking up DOTDOT, dvp is
468  * unlocked.  If we're looking up . an extra ref is taken, but the lock is
469  * not recursively acquired.
470  */
471 
472 int
473 cache_lookup(dvp, vpp, cnp, tsp, ticksp)
474 	struct vnode *dvp;
475 	struct vnode **vpp;
476 	struct componentname *cnp;
477 	struct timespec *tsp;
478 	int *ticksp;
479 {
480 	struct namecache *ncp;
481 	uint32_t hash;
482 	int error, ltype, wlocked;
483 
484 	if (!doingcache) {
485 		cnp->cn_flags &= ~MAKEENTRY;
486 		return (0);
487 	}
488 retry:
489 	CACHE_RLOCK();
490 	wlocked = 0;
491 	numcalls++;
492 	error = 0;
493 
494 retry_wlocked:
495 	if (cnp->cn_nameptr[0] == '.') {
496 		if (cnp->cn_namelen == 1) {
497 			*vpp = dvp;
498 			CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
499 			    dvp, cnp->cn_nameptr);
500 			dothits++;
501 			SDT_PROBE(vfs, namecache, lookup, hit, dvp, ".",
502 			    *vpp, 0, 0);
503 			if (tsp != NULL)
504 				timespecclear(tsp);
505 			if (ticksp != NULL)
506 				*ticksp = ticks;
507 			goto success;
508 		}
509 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
510 			dotdothits++;
511 			if (dvp->v_cache_dd == NULL) {
512 				SDT_PROBE(vfs, namecache, lookup, miss, dvp,
513 				    "..", NULL, 0, 0);
514 				goto unlock;
515 			}
516 			if ((cnp->cn_flags & MAKEENTRY) == 0) {
517 				if (!wlocked && !CACHE_UPGRADE_LOCK())
518 					goto wlock;
519 				if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
520 					cache_zap(dvp->v_cache_dd);
521 				dvp->v_cache_dd = NULL;
522 				CACHE_WUNLOCK();
523 				return (0);
524 			}
525 			ncp = dvp->v_cache_dd;
526 			if (ncp->nc_flag & NCF_ISDOTDOT)
527 				*vpp = ncp->nc_vp;
528 			else
529 				*vpp = ncp->nc_dvp;
530 			/* Return failure if negative entry was found. */
531 			if (*vpp == NULL)
532 				goto negative_success;
533 			CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
534 			    dvp, cnp->cn_nameptr, *vpp);
535 			SDT_PROBE(vfs, namecache, lookup, hit, dvp, "..",
536 			    *vpp, 0, 0);
537 			cache_out_ts(ncp, tsp, ticksp);
538 			if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
539 			    NCF_DTS && tsp != NULL)
540 				*tsp = ((struct namecache_ts *)ncp)->
541 				    nc_dotdottime;
542 			goto success;
543 		}
544 	}
545 
546 	hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
547 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
548 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
549 		numchecks++;
550 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
551 		    !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen))
552 			break;
553 	}
554 
555 	/* We failed to find an entry */
556 	if (ncp == NULL) {
557 		SDT_PROBE(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
558 		    NULL, 0, 0);
559 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
560 			nummisszap++;
561 		} else {
562 			nummiss++;
563 		}
564 		nchstats.ncs_miss++;
565 		goto unlock;
566 	}
567 
568 	/* We don't want to have an entry, so dump it */
569 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
570 		numposzaps++;
571 		nchstats.ncs_badhits++;
572 		if (!wlocked && !CACHE_UPGRADE_LOCK())
573 			goto wlock;
574 		cache_zap(ncp);
575 		CACHE_WUNLOCK();
576 		return (0);
577 	}
578 
579 	/* We found a "positive" match, return the vnode */
580 	if (ncp->nc_vp) {
581 		numposhits++;
582 		nchstats.ncs_goodhits++;
583 		*vpp = ncp->nc_vp;
584 		CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
585 		    dvp, cnp->cn_nameptr, *vpp, ncp);
586 		SDT_PROBE(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp),
587 		    *vpp, 0, 0);
588 		cache_out_ts(ncp, tsp, ticksp);
589 		goto success;
590 	}
591 
592 negative_success:
593 	/* We found a negative match, and want to create it, so purge */
594 	if (cnp->cn_nameiop == CREATE) {
595 		numnegzaps++;
596 		nchstats.ncs_badhits++;
597 		if (!wlocked && !CACHE_UPGRADE_LOCK())
598 			goto wlock;
599 		cache_zap(ncp);
600 		CACHE_WUNLOCK();
601 		return (0);
602 	}
603 
604 	if (!wlocked && !CACHE_UPGRADE_LOCK())
605 		goto wlock;
606 	numneghits++;
607 	/*
608 	 * We found a "negative" match, so we shift it to the end of
609 	 * the "negative" cache entries queue to satisfy LRU.  Also,
610 	 * check to see if the entry is a whiteout; indicate this to
611 	 * the componentname, if so.
612 	 */
613 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
614 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
615 	nchstats.ncs_neghits++;
616 	if (ncp->nc_flag & NCF_WHITE)
617 		cnp->cn_flags |= ISWHITEOUT;
618 	SDT_PROBE(vfs, namecache, lookup, hit__negative, dvp, nc_get_name(ncp),
619 	    0, 0, 0);
620 	cache_out_ts(ncp, tsp, ticksp);
621 	CACHE_WUNLOCK();
622 	return (ENOENT);
623 
624 wlock:
625 	/*
626 	 * We need to update the cache after our lookup, so upgrade to
627 	 * a write lock and retry the operation.
628 	 */
629 	CACHE_RUNLOCK();
630 	CACHE_WLOCK();
631 	numupgrades++;
632 	wlocked = 1;
633 	goto retry_wlocked;
634 
635 success:
636 	/*
637 	 * On success we return a locked and ref'd vnode as per the lookup
638 	 * protocol.
639 	 */
640 	if (dvp == *vpp) {   /* lookup on "." */
641 		VREF(*vpp);
642 		if (wlocked)
643 			CACHE_WUNLOCK();
644 		else
645 			CACHE_RUNLOCK();
646 		/*
647 		 * When we lookup "." we still can be asked to lock it
648 		 * differently...
649 		 */
650 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
651 		if (ltype != VOP_ISLOCKED(*vpp)) {
652 			if (ltype == LK_EXCLUSIVE) {
653 				vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
654 				if ((*vpp)->v_iflag & VI_DOOMED) {
655 					/* forced unmount */
656 					vrele(*vpp);
657 					*vpp = NULL;
658 					return (ENOENT);
659 				}
660 			} else
661 				vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
662 		}
663 		return (-1);
664 	}
665 	ltype = 0;	/* silence gcc warning */
666 	if (cnp->cn_flags & ISDOTDOT) {
667 		ltype = VOP_ISLOCKED(dvp);
668 		VOP_UNLOCK(dvp, 0);
669 	}
670 	vhold(*vpp);
671 	if (wlocked)
672 		CACHE_WUNLOCK();
673 	else
674 		CACHE_RUNLOCK();
675 	error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread);
676 	if (cnp->cn_flags & ISDOTDOT) {
677 		vn_lock(dvp, ltype | LK_RETRY);
678 		if (dvp->v_iflag & VI_DOOMED) {
679 			if (error == 0)
680 				vput(*vpp);
681 			*vpp = NULL;
682 			return (ENOENT);
683 		}
684 	}
685 	if (error) {
686 		*vpp = NULL;
687 		goto retry;
688 	}
689 	if ((cnp->cn_flags & ISLASTCN) &&
690 	    (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
691 		ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
692 	}
693 	return (-1);
694 
695 unlock:
696 	if (wlocked)
697 		CACHE_WUNLOCK();
698 	else
699 		CACHE_RUNLOCK();
700 	return (0);
701 }
702 
703 /*
704  * Add an entry to the cache.
705  */
706 void
707 cache_enter_time(dvp, vp, cnp, tsp, dtsp)
708 	struct vnode *dvp;
709 	struct vnode *vp;
710 	struct componentname *cnp;
711 	struct timespec *tsp;
712 	struct timespec *dtsp;
713 {
714 	struct namecache *ncp, *n2;
715 	struct namecache_ts *n3;
716 	struct nchashhead *ncpp;
717 	uint32_t hash;
718 	int flag;
719 	int hold;
720 	int zap;
721 	int len;
722 
723 	CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
724 	VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
725 	    ("cache_enter: Adding a doomed vnode"));
726 	VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
727 	    ("cache_enter: Doomed vnode used as src"));
728 
729 	if (!doingcache)
730 		return;
731 
732 	/*
733 	 * Avoid blowout in namecache entries.
734 	 */
735 	if (numcache >= desiredvnodes * ncsizefactor)
736 		return;
737 
738 	flag = 0;
739 	if (cnp->cn_nameptr[0] == '.') {
740 		if (cnp->cn_namelen == 1)
741 			return;
742 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
743 			CACHE_WLOCK();
744 			/*
745 			 * If dotdot entry already exists, just retarget it
746 			 * to new parent vnode, otherwise continue with new
747 			 * namecache entry allocation.
748 			 */
749 			if ((ncp = dvp->v_cache_dd) != NULL &&
750 			    ncp->nc_flag & NCF_ISDOTDOT) {
751 				KASSERT(ncp->nc_dvp == dvp,
752 				    ("wrong isdotdot parent"));
753 				if (ncp->nc_vp != NULL) {
754 					TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
755 					    ncp, nc_dst);
756 				} else {
757 					TAILQ_REMOVE(&ncneg, ncp, nc_dst);
758 					numneg--;
759 				}
760 				if (vp != NULL) {
761 					TAILQ_INSERT_HEAD(&vp->v_cache_dst,
762 					    ncp, nc_dst);
763 				} else {
764 					TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
765 					numneg++;
766 				}
767 				ncp->nc_vp = vp;
768 				CACHE_WUNLOCK();
769 				return;
770 			}
771 			dvp->v_cache_dd = NULL;
772 			SDT_PROBE(vfs, namecache, enter, done, dvp, "..", vp,
773 			    0, 0);
774 			CACHE_WUNLOCK();
775 			flag = NCF_ISDOTDOT;
776 		}
777 	}
778 
779 	hold = 0;
780 	zap = 0;
781 
782 	/*
783 	 * Calculate the hash key and setup as much of the new
784 	 * namecache entry as possible before acquiring the lock.
785 	 */
786 	ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
787 	ncp->nc_vp = vp;
788 	ncp->nc_dvp = dvp;
789 	ncp->nc_flag = flag;
790 	if (tsp != NULL) {
791 		n3 = (struct namecache_ts *)ncp;
792 		n3->nc_time = *tsp;
793 		n3->nc_ticks = ticks;
794 		n3->nc_flag |= NCF_TS;
795 		if (dtsp != NULL) {
796 			n3->nc_dotdottime = *dtsp;
797 			n3->nc_flag |= NCF_DTS;
798 		}
799 	}
800 	len = ncp->nc_nlen = cnp->cn_namelen;
801 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
802 	strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1);
803 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
804 	CACHE_WLOCK();
805 
806 	/*
807 	 * See if this vnode or negative entry is already in the cache
808 	 * with this name.  This can happen with concurrent lookups of
809 	 * the same path name.
810 	 */
811 	ncpp = NCHHASH(hash);
812 	LIST_FOREACH(n2, ncpp, nc_hash) {
813 		if (n2->nc_dvp == dvp &&
814 		    n2->nc_nlen == cnp->cn_namelen &&
815 		    !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) {
816 			if (tsp != NULL) {
817 				KASSERT((n2->nc_flag & NCF_TS) != 0,
818 				    ("no NCF_TS"));
819 				n3 = (struct namecache_ts *)n2;
820 				n3->nc_time =
821 				    ((struct namecache_ts *)ncp)->nc_time;
822 				n3->nc_ticks =
823 				    ((struct namecache_ts *)ncp)->nc_ticks;
824 				if (dtsp != NULL) {
825 					n3->nc_dotdottime =
826 					    ((struct namecache_ts *)ncp)->
827 					    nc_dotdottime;
828 					n3->nc_flag |= NCF_DTS;
829 				}
830 			}
831 			CACHE_WUNLOCK();
832 			cache_free(ncp);
833 			return;
834 		}
835 	}
836 
837 	if (flag == NCF_ISDOTDOT) {
838 		/*
839 		 * See if we are trying to add .. entry, but some other lookup
840 		 * has populated v_cache_dd pointer already.
841 		 */
842 		if (dvp->v_cache_dd != NULL) {
843 		    CACHE_WUNLOCK();
844 		    cache_free(ncp);
845 		    return;
846 		}
847 		KASSERT(vp == NULL || vp->v_type == VDIR,
848 		    ("wrong vnode type %p", vp));
849 		dvp->v_cache_dd = ncp;
850 	}
851 
852 	numcache++;
853 	if (!vp) {
854 		numneg++;
855 		if (cnp->cn_flags & ISWHITEOUT)
856 			ncp->nc_flag |= NCF_WHITE;
857 	} else if (vp->v_type == VDIR) {
858 		if (flag != NCF_ISDOTDOT) {
859 			/*
860 			 * For this case, the cache entry maps both the
861 			 * directory name in it and the name ".." for the
862 			 * directory's parent.
863 			 */
864 			if ((n2 = vp->v_cache_dd) != NULL &&
865 			    (n2->nc_flag & NCF_ISDOTDOT) != 0)
866 				cache_zap(n2);
867 			vp->v_cache_dd = ncp;
868 		}
869 	} else {
870 		vp->v_cache_dd = NULL;
871 	}
872 
873 	/*
874 	 * Insert the new namecache entry into the appropriate chain
875 	 * within the cache entries table.
876 	 */
877 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
878 	if (flag != NCF_ISDOTDOT) {
879 		if (LIST_EMPTY(&dvp->v_cache_src)) {
880 			hold = 1;
881 			numcachehv++;
882 		}
883 		LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
884 	}
885 
886 	/*
887 	 * If the entry is "negative", we place it into the
888 	 * "negative" cache queue, otherwise, we place it into the
889 	 * destination vnode's cache entries queue.
890 	 */
891 	if (vp) {
892 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
893 		SDT_PROBE(vfs, namecache, enter, done, dvp, nc_get_name(ncp),
894 		    vp, 0, 0);
895 	} else {
896 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
897 		SDT_PROBE(vfs, namecache, enter_negative, done, dvp,
898 		    nc_get_name(ncp), 0, 0, 0);
899 	}
900 	if (numneg * ncnegfactor > numcache) {
901 		ncp = TAILQ_FIRST(&ncneg);
902 		KASSERT(ncp->nc_vp == NULL, ("ncp %p vp %p on ncneg",
903 		    ncp, ncp->nc_vp));
904 		zap = 1;
905 	}
906 	if (hold)
907 		vhold(dvp);
908 	if (zap)
909 		cache_zap(ncp);
910 	CACHE_WUNLOCK();
911 }
912 
913 /*
914  * Name cache initialization, from vfs_init() when we are booting
915  */
916 static void
917 nchinit(void *dummy __unused)
918 {
919 
920 	TAILQ_INIT(&ncneg);
921 
922 	cache_zone_small = uma_zcreate("S VFS Cache",
923 	    sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
924 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
925 	cache_zone_small_ts = uma_zcreate("STS VFS Cache",
926 	    sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
927 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
928 	cache_zone_large = uma_zcreate("L VFS Cache",
929 	    sizeof(struct namecache) + NAME_MAX + 1,
930 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
931 	cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
932 	    sizeof(struct namecache_ts) + NAME_MAX + 1,
933 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
934 
935 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
936 }
937 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
938 
939 void
940 cache_changesize(int newmaxvnodes)
941 {
942 	struct nchashhead *new_nchashtbl, *old_nchashtbl;
943 	u_long new_nchash, old_nchash;
944 	struct namecache *ncp;
945 	uint32_t hash;
946 	int i;
947 
948 	new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash);
949 	/* If same hash table size, nothing to do */
950 	if (nchash == new_nchash) {
951 		free(new_nchashtbl, M_VFSCACHE);
952 		return;
953 	}
954 	/*
955 	 * Move everything from the old hash table to the new table.
956 	 * None of the namecache entries in the table can be removed
957 	 * because to do so, they have to be removed from the hash table.
958 	 */
959 	CACHE_WLOCK();
960 	old_nchashtbl = nchashtbl;
961 	old_nchash = nchash;
962 	nchashtbl = new_nchashtbl;
963 	nchash = new_nchash;
964 	for (i = 0; i <= old_nchash; i++) {
965 		while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) {
966 			hash = fnv_32_buf(nc_get_name(ncp), ncp->nc_nlen,
967 			    FNV1_32_INIT);
968 			hash = fnv_32_buf(&ncp->nc_dvp, sizeof(ncp->nc_dvp),
969 			    hash);
970 			LIST_REMOVE(ncp, nc_hash);
971 			LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash);
972 		}
973 	}
974 	CACHE_WUNLOCK();
975 	free(old_nchashtbl, M_VFSCACHE);
976 }
977 
978 /*
979  * Invalidate all entries to a particular vnode.
980  */
981 void
982 cache_purge(vp)
983 	struct vnode *vp;
984 {
985 
986 	CTR1(KTR_VFS, "cache_purge(%p)", vp);
987 	SDT_PROBE(vfs, namecache, purge, done, vp, 0, 0, 0, 0);
988 	CACHE_WLOCK();
989 	while (!LIST_EMPTY(&vp->v_cache_src))
990 		cache_zap(LIST_FIRST(&vp->v_cache_src));
991 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
992 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
993 	if (vp->v_cache_dd != NULL) {
994 		KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
995 		   ("lost dotdot link"));
996 		cache_zap(vp->v_cache_dd);
997 	}
998 	KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
999 	CACHE_WUNLOCK();
1000 }
1001 
1002 /*
1003  * Invalidate all negative entries for a particular directory vnode.
1004  */
1005 void
1006 cache_purge_negative(vp)
1007 	struct vnode *vp;
1008 {
1009 	struct namecache *cp, *ncp;
1010 
1011 	CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
1012 	SDT_PROBE(vfs, namecache, purge_negative, done, vp, 0, 0, 0, 0);
1013 	CACHE_WLOCK();
1014 	LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
1015 		if (cp->nc_vp == NULL)
1016 			cache_zap(cp);
1017 	}
1018 	CACHE_WUNLOCK();
1019 }
1020 
1021 /*
1022  * Flush all entries referencing a particular filesystem.
1023  */
1024 void
1025 cache_purgevfs(mp)
1026 	struct mount *mp;
1027 {
1028 	struct nchashhead *ncpp;
1029 	struct namecache *ncp, *nnp;
1030 
1031 	/* Scan hash tables for applicable entries */
1032 	SDT_PROBE(vfs, namecache, purgevfs, done, mp, 0, 0, 0, 0);
1033 	CACHE_WLOCK();
1034 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
1035 		LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
1036 			if (ncp->nc_dvp->v_mount == mp)
1037 				cache_zap(ncp);
1038 		}
1039 	}
1040 	CACHE_WUNLOCK();
1041 }
1042 
1043 /*
1044  * Perform canonical checks and cache lookup and pass on to filesystem
1045  * through the vop_cachedlookup only if needed.
1046  */
1047 
1048 int
1049 vfs_cache_lookup(ap)
1050 	struct vop_lookup_args /* {
1051 		struct vnode *a_dvp;
1052 		struct vnode **a_vpp;
1053 		struct componentname *a_cnp;
1054 	} */ *ap;
1055 {
1056 	struct vnode *dvp;
1057 	int error;
1058 	struct vnode **vpp = ap->a_vpp;
1059 	struct componentname *cnp = ap->a_cnp;
1060 	struct ucred *cred = cnp->cn_cred;
1061 	int flags = cnp->cn_flags;
1062 	struct thread *td = cnp->cn_thread;
1063 
1064 	*vpp = NULL;
1065 	dvp = ap->a_dvp;
1066 
1067 	if (dvp->v_type != VDIR)
1068 		return (ENOTDIR);
1069 
1070 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1071 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1072 		return (EROFS);
1073 
1074 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
1075 	if (error)
1076 		return (error);
1077 
1078 	error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
1079 	if (error == 0)
1080 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
1081 	if (error == -1)
1082 		return (0);
1083 	return (error);
1084 }
1085 
1086 /*
1087  * XXX All of these sysctls would probably be more productive dead.
1088  */
1089 static int disablecwd;
1090 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
1091    "Disable the getcwd syscall");
1092 
1093 /* Implementation of the getcwd syscall. */
1094 int
1095 sys___getcwd(td, uap)
1096 	struct thread *td;
1097 	struct __getcwd_args *uap;
1098 {
1099 
1100 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen,
1101 	    MAXPATHLEN));
1102 }
1103 
1104 int
1105 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen,
1106     u_int path_max)
1107 {
1108 	char *bp, *tmpbuf;
1109 	struct filedesc *fdp;
1110 	struct vnode *cdir, *rdir;
1111 	int error;
1112 
1113 	if (disablecwd)
1114 		return (ENODEV);
1115 	if (buflen < 2)
1116 		return (EINVAL);
1117 	if (buflen > path_max)
1118 		buflen = path_max;
1119 
1120 	tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
1121 	fdp = td->td_proc->p_fd;
1122 	FILEDESC_SLOCK(fdp);
1123 	cdir = fdp->fd_cdir;
1124 	VREF(cdir);
1125 	rdir = fdp->fd_rdir;
1126 	VREF(rdir);
1127 	FILEDESC_SUNLOCK(fdp);
1128 	error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
1129 	vrele(rdir);
1130 	vrele(cdir);
1131 
1132 	if (!error) {
1133 		if (bufseg == UIO_SYSSPACE)
1134 			bcopy(bp, buf, strlen(bp) + 1);
1135 		else
1136 			error = copyout(bp, buf, strlen(bp) + 1);
1137 #ifdef KTRACE
1138 	if (KTRPOINT(curthread, KTR_NAMEI))
1139 		ktrnamei(bp);
1140 #endif
1141 	}
1142 	free(tmpbuf, M_TEMP);
1143 	return (error);
1144 }
1145 
1146 /*
1147  * Thus begins the fullpath magic.
1148  */
1149 
1150 #undef STATNODE
1151 #define STATNODE(name, descr)						\
1152 	static u_int name;						\
1153 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr)
1154 
1155 static int disablefullpath;
1156 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
1157     "Disable the vn_fullpath function");
1158 
1159 /* These count for kern___getcwd(), too. */
1160 STATNODE(numfullpathcalls, "Number of fullpath search calls");
1161 STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
1162 STATNODE(numfullpathfail2,
1163     "Number of fullpath search errors (VOP_VPTOCNP failures)");
1164 STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
1165 STATNODE(numfullpathfound, "Number of successful fullpath calls");
1166 
1167 /*
1168  * Retrieve the full filesystem path that correspond to a vnode from the name
1169  * cache (if available)
1170  */
1171 int
1172 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
1173 {
1174 	char *buf;
1175 	struct filedesc *fdp;
1176 	struct vnode *rdir;
1177 	int error;
1178 
1179 	if (disablefullpath)
1180 		return (ENODEV);
1181 	if (vn == NULL)
1182 		return (EINVAL);
1183 
1184 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1185 	fdp = td->td_proc->p_fd;
1186 	FILEDESC_SLOCK(fdp);
1187 	rdir = fdp->fd_rdir;
1188 	VREF(rdir);
1189 	FILEDESC_SUNLOCK(fdp);
1190 	error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1191 	vrele(rdir);
1192 
1193 	if (!error)
1194 		*freebuf = buf;
1195 	else
1196 		free(buf, M_TEMP);
1197 	return (error);
1198 }
1199 
1200 /*
1201  * This function is similar to vn_fullpath, but it attempts to lookup the
1202  * pathname relative to the global root mount point.  This is required for the
1203  * auditing sub-system, as audited pathnames must be absolute, relative to the
1204  * global root mount point.
1205  */
1206 int
1207 vn_fullpath_global(struct thread *td, struct vnode *vn,
1208     char **retbuf, char **freebuf)
1209 {
1210 	char *buf;
1211 	int error;
1212 
1213 	if (disablefullpath)
1214 		return (ENODEV);
1215 	if (vn == NULL)
1216 		return (EINVAL);
1217 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1218 	error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1219 	if (!error)
1220 		*freebuf = buf;
1221 	else
1222 		free(buf, M_TEMP);
1223 	return (error);
1224 }
1225 
1226 int
1227 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1228 {
1229 	int error;
1230 
1231 	CACHE_RLOCK();
1232 	error = vn_vptocnp_locked(vp, cred, buf, buflen);
1233 	if (error == 0)
1234 		CACHE_RUNLOCK();
1235 	return (error);
1236 }
1237 
1238 static int
1239 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1240     u_int *buflen)
1241 {
1242 	struct vnode *dvp;
1243 	struct namecache *ncp;
1244 	int error;
1245 
1246 	TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1247 		if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1248 			break;
1249 	}
1250 	if (ncp != NULL) {
1251 		if (*buflen < ncp->nc_nlen) {
1252 			CACHE_RUNLOCK();
1253 			vrele(*vp);
1254 			numfullpathfail4++;
1255 			error = ENOMEM;
1256 			SDT_PROBE(vfs, namecache, fullpath, return, error,
1257 			    vp, NULL, 0, 0);
1258 			return (error);
1259 		}
1260 		*buflen -= ncp->nc_nlen;
1261 		memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen);
1262 		SDT_PROBE(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1263 		    nc_get_name(ncp), vp, 0, 0);
1264 		dvp = *vp;
1265 		*vp = ncp->nc_dvp;
1266 		vref(*vp);
1267 		CACHE_RUNLOCK();
1268 		vrele(dvp);
1269 		CACHE_RLOCK();
1270 		return (0);
1271 	}
1272 	SDT_PROBE(vfs, namecache, fullpath, miss, vp, 0, 0, 0, 0);
1273 
1274 	CACHE_RUNLOCK();
1275 	vn_lock(*vp, LK_SHARED | LK_RETRY);
1276 	error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1277 	vput(*vp);
1278 	if (error) {
1279 		numfullpathfail2++;
1280 		SDT_PROBE(vfs, namecache, fullpath, return,  error, vp,
1281 		    NULL, 0, 0);
1282 		return (error);
1283 	}
1284 
1285 	*vp = dvp;
1286 	CACHE_RLOCK();
1287 	if (dvp->v_iflag & VI_DOOMED) {
1288 		/* forced unmount */
1289 		CACHE_RUNLOCK();
1290 		vrele(dvp);
1291 		error = ENOENT;
1292 		SDT_PROBE(vfs, namecache, fullpath, return, error, vp,
1293 		    NULL, 0, 0);
1294 		return (error);
1295 	}
1296 	/*
1297 	 * *vp has its use count incremented still.
1298 	 */
1299 
1300 	return (0);
1301 }
1302 
1303 /*
1304  * The magic behind kern___getcwd() and vn_fullpath().
1305  */
1306 static int
1307 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1308     char *buf, char **retbuf, u_int buflen)
1309 {
1310 	int error, slash_prefixed;
1311 #ifdef KDTRACE_HOOKS
1312 	struct vnode *startvp = vp;
1313 #endif
1314 	struct vnode *vp1;
1315 
1316 	buflen--;
1317 	buf[buflen] = '\0';
1318 	error = 0;
1319 	slash_prefixed = 0;
1320 
1321 	SDT_PROBE(vfs, namecache, fullpath, entry, vp, 0, 0, 0, 0);
1322 	numfullpathcalls++;
1323 	vref(vp);
1324 	CACHE_RLOCK();
1325 	if (vp->v_type != VDIR) {
1326 		error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1327 		if (error)
1328 			return (error);
1329 		if (buflen == 0) {
1330 			CACHE_RUNLOCK();
1331 			vrele(vp);
1332 			return (ENOMEM);
1333 		}
1334 		buf[--buflen] = '/';
1335 		slash_prefixed = 1;
1336 	}
1337 	while (vp != rdir && vp != rootvnode) {
1338 		if (vp->v_vflag & VV_ROOT) {
1339 			if (vp->v_iflag & VI_DOOMED) {	/* forced unmount */
1340 				CACHE_RUNLOCK();
1341 				vrele(vp);
1342 				error = ENOENT;
1343 				SDT_PROBE(vfs, namecache, fullpath, return,
1344 				    error, vp, NULL, 0, 0);
1345 				break;
1346 			}
1347 			vp1 = vp->v_mount->mnt_vnodecovered;
1348 			vref(vp1);
1349 			CACHE_RUNLOCK();
1350 			vrele(vp);
1351 			vp = vp1;
1352 			CACHE_RLOCK();
1353 			continue;
1354 		}
1355 		if (vp->v_type != VDIR) {
1356 			CACHE_RUNLOCK();
1357 			vrele(vp);
1358 			numfullpathfail1++;
1359 			error = ENOTDIR;
1360 			SDT_PROBE(vfs, namecache, fullpath, return,
1361 			    error, vp, NULL, 0, 0);
1362 			break;
1363 		}
1364 		error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1365 		if (error)
1366 			break;
1367 		if (buflen == 0) {
1368 			CACHE_RUNLOCK();
1369 			vrele(vp);
1370 			error = ENOMEM;
1371 			SDT_PROBE(vfs, namecache, fullpath, return, error,
1372 			    startvp, NULL, 0, 0);
1373 			break;
1374 		}
1375 		buf[--buflen] = '/';
1376 		slash_prefixed = 1;
1377 	}
1378 	if (error)
1379 		return (error);
1380 	if (!slash_prefixed) {
1381 		if (buflen == 0) {
1382 			CACHE_RUNLOCK();
1383 			vrele(vp);
1384 			numfullpathfail4++;
1385 			SDT_PROBE(vfs, namecache, fullpath, return, ENOMEM,
1386 			    startvp, NULL, 0, 0);
1387 			return (ENOMEM);
1388 		}
1389 		buf[--buflen] = '/';
1390 	}
1391 	numfullpathfound++;
1392 	CACHE_RUNLOCK();
1393 	vrele(vp);
1394 
1395 	SDT_PROBE(vfs, namecache, fullpath, return, 0, startvp, buf + buflen,
1396 	    0, 0);
1397 	*retbuf = buf + buflen;
1398 	return (0);
1399 }
1400 
1401 struct vnode *
1402 vn_dir_dd_ino(struct vnode *vp)
1403 {
1404 	struct namecache *ncp;
1405 	struct vnode *ddvp;
1406 
1407 	ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
1408 	CACHE_RLOCK();
1409 	TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
1410 		if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
1411 			continue;
1412 		ddvp = ncp->nc_dvp;
1413 		vhold(ddvp);
1414 		CACHE_RUNLOCK();
1415 		if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread))
1416 			return (NULL);
1417 		return (ddvp);
1418 	}
1419 	CACHE_RUNLOCK();
1420 	return (NULL);
1421 }
1422 
1423 int
1424 vn_commname(struct vnode *vp, char *buf, u_int buflen)
1425 {
1426 	struct namecache *ncp;
1427 	int l;
1428 
1429 	CACHE_RLOCK();
1430 	TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1431 		if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1432 			break;
1433 	if (ncp == NULL) {
1434 		CACHE_RUNLOCK();
1435 		return (ENOENT);
1436 	}
1437 	l = min(ncp->nc_nlen, buflen - 1);
1438 	memcpy(buf, nc_get_name(ncp), l);
1439 	CACHE_RUNLOCK();
1440 	buf[l] = '\0';
1441 	return (0);
1442 }
1443 
1444 /* ABI compat shims for old kernel modules. */
1445 #undef cache_enter
1446 
1447 void	cache_enter(struct vnode *dvp, struct vnode *vp,
1448 	    struct componentname *cnp);
1449 
1450 void
1451 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
1452 {
1453 
1454 	cache_enter_time(dvp, vp, cnp, NULL, NULL);
1455 }
1456 
1457 /*
1458  * This function updates path string to vnode's full global path
1459  * and checks the size of the new path string against the pathlen argument.
1460  *
1461  * Requires a locked, referenced vnode.
1462  * Vnode is re-locked on success or ENODEV, otherwise unlocked.
1463  *
1464  * If sysctl debug.disablefullpath is set, ENODEV is returned,
1465  * vnode is left locked and path remain untouched.
1466  *
1467  * If vp is a directory, the call to vn_fullpath_global() always succeeds
1468  * because it falls back to the ".." lookup if the namecache lookup fails.
1469  */
1470 int
1471 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
1472     u_int pathlen)
1473 {
1474 	struct nameidata nd;
1475 	struct vnode *vp1;
1476 	char *rpath, *fbuf;
1477 	int error;
1478 
1479 	ASSERT_VOP_ELOCKED(vp, __func__);
1480 
1481 	/* Return ENODEV if sysctl debug.disablefullpath==1 */
1482 	if (disablefullpath)
1483 		return (ENODEV);
1484 
1485 	/* Construct global filesystem path from vp. */
1486 	VOP_UNLOCK(vp, 0);
1487 	error = vn_fullpath_global(td, vp, &rpath, &fbuf);
1488 
1489 	if (error != 0) {
1490 		vrele(vp);
1491 		return (error);
1492 	}
1493 
1494 	if (strlen(rpath) >= pathlen) {
1495 		vrele(vp);
1496 		error = ENAMETOOLONG;
1497 		goto out;
1498 	}
1499 
1500 	/*
1501 	 * Re-lookup the vnode by path to detect a possible rename.
1502 	 * As a side effect, the vnode is relocked.
1503 	 * If vnode was renamed, return ENOENT.
1504 	 */
1505 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1506 	    UIO_SYSSPACE, path, td);
1507 	error = namei(&nd);
1508 	if (error != 0) {
1509 		vrele(vp);
1510 		goto out;
1511 	}
1512 	NDFREE(&nd, NDF_ONLY_PNBUF);
1513 	vp1 = nd.ni_vp;
1514 	vrele(vp);
1515 	if (vp1 == vp)
1516 		strcpy(path, rpath);
1517 	else {
1518 		vput(vp1);
1519 		error = ENOENT;
1520 	}
1521 
1522 out:
1523 	free(fbuf, M_TEMP);
1524 	return (error);
1525 }
1526