2 * Copyright (c) 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Poul-Henning Kamp of the FreeBSD Project.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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 * 3. 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.
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
32 * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
38 #include "opt_ktrace.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/counter.h>
43 #include <sys/filedesc.h>
44 #include <sys/fnv_hash.h>
45 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/fcntl.h>
49 #include <sys/mount.h>
50 #include <sys/namei.h>
52 #include <sys/rwlock.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysproto.h>
58 #include <sys/vnode.h>
60 #include <sys/ktrace.h>
65 SDT_PROVIDER_DECLARE(vfs);
66 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
68 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
70 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *");
71 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *",
72 "char *", "struct vnode *");
73 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *");
74 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int",
75 "struct vnode *", "char *");
76 SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *",
78 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative,
79 "struct vnode *", "char *");
80 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
82 SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *");
83 SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *");
84 SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *");
85 SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *",
87 SDT_PROBE_DEFINE3(vfs, namecache, zap_negative, done, "struct vnode *",
89 SDT_PROBE_DEFINE3(vfs, namecache, shrink_negative, done, "struct vnode *",
93 * This structure describes the elements in the cache of recent
94 * names looked up by namei.
98 LIST_ENTRY(namecache) nc_hash; /* hash chain */
99 LIST_ENTRY(namecache) nc_src; /* source vnode list */
100 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
101 struct vnode *nc_dvp; /* vnode of parent of name */
103 struct vnode *nu_vp; /* vnode the name refers to */
104 u_int nu_neghits; /* negative entry hits */
106 u_char nc_flag; /* flag bits */
107 u_char nc_nlen; /* length of name */
108 char nc_name[0]; /* segment name + nul */
112 * struct namecache_ts repeats struct namecache layout up to the
114 * struct namecache_ts is used in place of struct namecache when time(s) need
115 * to be stored. The nc_dotdottime field is used when a cache entry is mapping
116 * both a non-dotdot directory name plus dotdot for the directory's
119 struct namecache_ts {
120 LIST_ENTRY(namecache) nc_hash; /* hash chain */
121 LIST_ENTRY(namecache) nc_src; /* source vnode list */
122 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
123 struct vnode *nc_dvp; /* vnode of parent of name */
125 struct vnode *nu_vp; /* vnode the name refers to */
126 u_int nu_neghits; /* negative entry hits */
128 u_char nc_flag; /* flag bits */
129 u_char nc_nlen; /* length of name */
130 struct timespec nc_time; /* timespec provided by fs */
131 struct timespec nc_dotdottime; /* dotdot timespec provided by fs */
132 int nc_ticks; /* ticks value when entry was added */
133 char nc_name[0]; /* segment name + nul */
136 #define nc_vp n_un.nu_vp
137 #define nc_neghits n_un.nu_neghits
140 * Flags in namecache.nc_flag
142 #define NCF_WHITE 0x01
143 #define NCF_ISDOTDOT 0x02
146 #define NCF_DVDROP 0x10
147 #define NCF_NEGATIVE 0x20
148 #define NCF_HOTNEGATIVE 0x40
151 * Name caching works as follows:
153 * Names found by directory scans are retained in a cache
154 * for future reference. It is managed LRU, so frequently
155 * used names will hang around. Cache is indexed by hash value
156 * obtained from (vp, name) where vp refers to the directory
159 * If it is a "negative" entry, (i.e. for a name that is known NOT to
160 * exist) the vnode pointer will be NULL.
162 * Upon reaching the last segment of a path, if the reference
163 * is for DELETE, or NOCACHE is set (rewrite), and the
164 * name is located in the cache, it will be dropped.
166 * These locks are used (in the order in which they can be taken):
168 * vnodelock mtx vnode lists and v_cache_dd field protection
169 * bucketlock rwlock for access to given set of hash buckets
170 * neglist mtx negative entry LRU management
172 * Additionally, ncneg_shrink_lock mtx is used to have at most one thread
173 * shrinking the LRU list.
175 * It is legal to take multiple vnodelock and bucketlock locks. The locking
176 * order is lower address first. Both are recursive.
178 * "." lookups are lockless.
180 * ".." and vnode -> name lookups require vnodelock.
182 * name -> vnode lookup requires the relevant bucketlock to be held for reading.
184 * Insertions and removals of entries require involved vnodes and bucketlocks
185 * to be write-locked to prevent other threads from seeing the entry.
187 * Some lookups result in removal of the found entry (e.g. getting rid of a
188 * negative entry with the intent to create a positive one), which poses a
189 * problem when multiple threads reach the state. Similarly, two different
190 * threads can purge two different vnodes and try to remove the same name.
192 * If the already held vnode lock is lower than the second required lock, we
193 * can just take the other lock. However, in the opposite case, this could
194 * deadlock. As such, this is resolved by trylocking and if that fails unlocking
195 * the first node, locking everything in order and revalidating the state.
199 * Structures associated with name caching.
201 #define NCHHASH(hash) \
202 (&nchashtbl[(hash) & nchash])
203 static __read_mostly LIST_HEAD(nchashhead, namecache) *nchashtbl;/* Hash Table */
204 static u_long __read_mostly nchash; /* size of hash table */
205 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0,
206 "Size of namecache hash table");
207 static u_long __read_mostly ncnegfactor = 16; /* ratio of negative entries */
208 SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0,
209 "Ratio of negative namecache entries");
210 static u_long __exclusive_cache_line numneg; /* number of negative entries allocated */
211 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0,
212 "Number of negative entries in namecache");
213 static u_long __exclusive_cache_line numcache;/* number of cache entries allocated */
214 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0,
215 "Number of namecache entries");
216 static u_long __exclusive_cache_line numcachehv;/* number of cache entries with vnodes held */
217 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0,
218 "Number of namecache entries with vnodes held");
219 u_int __read_mostly ncsizefactor = 2;
220 SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0,
221 "Size factor for namecache");
222 static u_int __read_mostly ncpurgeminvnodes;
223 SYSCTL_UINT(_vfs, OID_AUTO, ncpurgeminvnodes, CTLFLAG_RW, &ncpurgeminvnodes, 0,
224 "Number of vnodes below which purgevfs ignores the request");
225 static u_int __read_mostly ncneghitsrequeue = 8;
226 SYSCTL_UINT(_vfs, OID_AUTO, ncneghitsrequeue, CTLFLAG_RW, &ncneghitsrequeue, 0,
227 "Number of hits to requeue a negative entry in the LRU list");
229 struct nchstats nchstats; /* cache effectiveness statistics */
231 static struct mtx ncneg_shrink_lock;
232 static int shrink_list_turn;
236 TAILQ_HEAD(, namecache) nl_list;
237 } __aligned(CACHE_LINE_SIZE);
239 static struct neglist __read_mostly *neglists;
240 static struct neglist ncneg_hot;
242 #define numneglists (ncneghash + 1)
243 static u_int __read_mostly ncneghash;
244 static inline struct neglist *
245 NCP2NEGLIST(struct namecache *ncp)
248 return (&neglists[(((uintptr_t)(ncp) >> 8) & ncneghash)]);
251 #define numbucketlocks (ncbuckethash + 1)
252 static u_int __read_mostly ncbuckethash;
253 static struct rwlock_padalign __read_mostly *bucketlocks;
254 #define HASH2BUCKETLOCK(hash) \
255 ((struct rwlock *)(&bucketlocks[((hash) & ncbuckethash)]))
257 #define numvnodelocks (ncvnodehash + 1)
258 static u_int __read_mostly ncvnodehash;
259 static struct mtx __read_mostly *vnodelocks;
260 static inline struct mtx *
261 VP2VNODELOCK(struct vnode *vp)
264 return (&vnodelocks[(((uintptr_t)(vp) >> 8) & ncvnodehash)]);
268 * UMA zones for the VFS cache.
270 * The small cache is used for entries with short names, which are the
271 * most common. The large cache is used for entries which are too big to
272 * fit in the small cache.
274 static uma_zone_t __read_mostly cache_zone_small;
275 static uma_zone_t __read_mostly cache_zone_small_ts;
276 static uma_zone_t __read_mostly cache_zone_large;
277 static uma_zone_t __read_mostly cache_zone_large_ts;
279 #define CACHE_PATH_CUTOFF 35
281 static struct namecache *
282 cache_alloc(int len, int ts)
285 if (len > CACHE_PATH_CUTOFF) {
287 return (uma_zalloc(cache_zone_large_ts, M_WAITOK));
289 return (uma_zalloc(cache_zone_large, M_WAITOK));
292 return (uma_zalloc(cache_zone_small_ts, M_WAITOK));
294 return (uma_zalloc(cache_zone_small, M_WAITOK));
298 cache_free(struct namecache *ncp)
304 ts = ncp->nc_flag & NCF_TS;
305 if ((ncp->nc_flag & NCF_DVDROP) != 0)
307 if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) {
309 uma_zfree(cache_zone_small_ts, ncp);
311 uma_zfree(cache_zone_small, ncp);
313 uma_zfree(cache_zone_large_ts, ncp);
315 uma_zfree(cache_zone_large, ncp);
319 nc_get_name(struct namecache *ncp)
321 struct namecache_ts *ncp_ts;
323 if ((ncp->nc_flag & NCF_TS) == 0)
324 return (ncp->nc_name);
325 ncp_ts = (struct namecache_ts *)ncp;
326 return (ncp_ts->nc_name);
330 cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp)
333 KASSERT((ncp->nc_flag & NCF_TS) != 0 ||
334 (tsp == NULL && ticksp == NULL),
338 *tsp = ((struct namecache_ts *)ncp)->nc_time;
340 *ticksp = ((struct namecache_ts *)ncp)->nc_ticks;
343 static int __read_mostly doingcache = 1; /* 1 => enable the cache */
344 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
345 "VFS namecache enabled");
347 /* Export size information to userland */
348 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR,
349 sizeof(struct namecache), "sizeof(struct namecache)");
352 * The new name cache statistics
354 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0,
355 "Name cache statistics");
356 #define STATNODE_ULONG(name, descr) \
357 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr);
358 #define STATNODE_COUNTER(name, descr) \
359 static counter_u64_t __read_mostly name; \
360 SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, descr);
361 STATNODE_ULONG(numneg, "Number of negative cache entries");
362 STATNODE_ULONG(numcache, "Number of cache entries");
363 STATNODE_COUNTER(numcalls, "Number of cache lookups");
364 STATNODE_COUNTER(dothits, "Number of '.' hits");
365 STATNODE_COUNTER(dotdothits, "Number of '..' hits");
366 STATNODE_COUNTER(numchecks, "Number of checks in lookup");
367 STATNODE_COUNTER(nummiss, "Number of cache misses");
368 STATNODE_COUNTER(nummisszap, "Number of cache misses we do not want to cache");
369 STATNODE_COUNTER(numposzaps,
370 "Number of cache hits (positive) we do not want to cache");
371 STATNODE_COUNTER(numposhits, "Number of cache hits (positive)");
372 STATNODE_COUNTER(numnegzaps,
373 "Number of cache hits (negative) we do not want to cache");
374 STATNODE_COUNTER(numneghits, "Number of cache hits (negative)");
375 /* These count for kern___getcwd(), too. */
376 STATNODE_COUNTER(numfullpathcalls, "Number of fullpath search calls");
377 STATNODE_COUNTER(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
378 STATNODE_COUNTER(numfullpathfail2,
379 "Number of fullpath search errors (VOP_VPTOCNP failures)");
380 STATNODE_COUNTER(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
381 STATNODE_COUNTER(numfullpathfound, "Number of successful fullpath calls");
382 static long zap_and_exit_bucket_fail; STATNODE_ULONG(zap_and_exit_bucket_fail,
383 "Number of times zap_and_exit failed to lock");
384 static long cache_lock_vnodes_cel_3_failures;
385 STATNODE_ULONG(cache_lock_vnodes_cel_3_failures,
386 "Number of times 3-way vnode locking failed");
388 static void cache_zap_locked(struct namecache *ncp, bool neg_locked);
389 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
390 char *buf, char **retbuf, u_int buflen);
392 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
394 static int cache_yield;
395 SYSCTL_INT(_vfs_cache, OID_AUTO, yield, CTLFLAG_RD, &cache_yield, 0,
396 "Number of times cache called yield");
399 cache_maybe_yield(void)
402 if (should_yield()) {
404 kern_yield(PRI_USER);
409 cache_assert_vlp_locked(struct mtx *vlp)
413 mtx_assert(vlp, MA_OWNED);
417 cache_assert_vnode_locked(struct vnode *vp)
421 vlp = VP2VNODELOCK(vp);
422 cache_assert_vlp_locked(vlp);
426 cache_get_hash(char *name, u_char len, struct vnode *dvp)
430 hash = fnv_32_buf(name, len, FNV1_32_INIT);
431 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
435 static inline struct rwlock *
436 NCP2BUCKETLOCK(struct namecache *ncp)
440 hash = cache_get_hash(nc_get_name(ncp), ncp->nc_nlen, ncp->nc_dvp);
441 return (HASH2BUCKETLOCK(hash));
446 cache_assert_bucket_locked(struct namecache *ncp, int mode)
450 blp = NCP2BUCKETLOCK(ncp);
451 rw_assert(blp, mode);
454 #define cache_assert_bucket_locked(x, y) do { } while (0)
457 #define cache_sort(x, y) _cache_sort((void **)(x), (void **)(y))
459 _cache_sort(void **p1, void **p2)
471 cache_lock_all_buckets(void)
475 for (i = 0; i < numbucketlocks; i++)
476 rw_wlock(&bucketlocks[i]);
480 cache_unlock_all_buckets(void)
484 for (i = 0; i < numbucketlocks; i++)
485 rw_wunlock(&bucketlocks[i]);
489 cache_lock_all_vnodes(void)
493 for (i = 0; i < numvnodelocks; i++)
494 mtx_lock(&vnodelocks[i]);
498 cache_unlock_all_vnodes(void)
502 for (i = 0; i < numvnodelocks; i++)
503 mtx_unlock(&vnodelocks[i]);
507 cache_trylock_vnodes(struct mtx *vlp1, struct mtx *vlp2)
510 cache_sort(&vlp1, &vlp2);
514 if (!mtx_trylock(vlp1))
517 if (!mtx_trylock(vlp2)) {
527 cache_unlock_vnodes(struct mtx *vlp1, struct mtx *vlp2)
530 MPASS(vlp1 != NULL || vlp2 != NULL);
539 sysctl_nchstats(SYSCTL_HANDLER_ARGS)
541 struct nchstats snap;
543 if (req->oldptr == NULL)
544 return (SYSCTL_OUT(req, 0, sizeof(snap)));
547 snap.ncs_goodhits = counter_u64_fetch(numposhits);
548 snap.ncs_neghits = counter_u64_fetch(numneghits);
549 snap.ncs_badhits = counter_u64_fetch(numposzaps) +
550 counter_u64_fetch(numnegzaps);
551 snap.ncs_miss = counter_u64_fetch(nummisszap) +
552 counter_u64_fetch(nummiss);
554 return (SYSCTL_OUT(req, &snap, sizeof(snap)));
556 SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD |
557 CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU",
558 "VFS cache effectiveness statistics");
562 * Grab an atomic snapshot of the name cache hash chain lengths
564 static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
568 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
570 struct nchashhead *ncpp;
571 struct namecache *ncp;
572 int i, error, n_nchash, *cntbuf;
575 n_nchash = nchash + 1; /* nchash is max index, not count */
576 if (req->oldptr == NULL)
577 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
578 cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK);
579 cache_lock_all_buckets();
580 if (n_nchash != nchash + 1) {
581 cache_unlock_all_buckets();
582 free(cntbuf, M_TEMP);
585 /* Scan hash tables counting entries */
586 for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++)
587 LIST_FOREACH(ncp, ncpp, nc_hash)
589 cache_unlock_all_buckets();
590 for (error = 0, i = 0; i < n_nchash; i++)
591 if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0)
593 free(cntbuf, M_TEMP);
596 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
597 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
598 "nchash chain lengths");
601 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
604 struct nchashhead *ncpp;
605 struct namecache *ncp;
607 int count, maxlength, used, pct;
610 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
612 cache_lock_all_buckets();
613 n_nchash = nchash + 1; /* nchash is max index, not count */
617 /* Scan hash tables for applicable entries */
618 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
620 LIST_FOREACH(ncp, ncpp, nc_hash) {
625 if (maxlength < count)
628 n_nchash = nchash + 1;
629 cache_unlock_all_buckets();
630 pct = (used * 100) / (n_nchash / 100);
631 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
634 error = SYSCTL_OUT(req, &used, sizeof(used));
637 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
640 error = SYSCTL_OUT(req, &pct, sizeof(pct));
645 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
646 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
647 "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)");
651 * Negative entries management
653 * A variation of LRU scheme is used. New entries are hashed into one of
654 * numneglists cold lists. Entries get promoted to the hot list on first hit.
655 * Partial LRU for the hot list is maintained by requeueing them every
656 * ncneghitsrequeue hits.
658 * The shrinker will demote hot list head and evict from the cold list in a
659 * round-robin manner.
662 cache_negative_hit(struct namecache *ncp)
664 struct neglist *neglist;
667 MPASS(ncp->nc_flag & NCF_NEGATIVE);
668 hits = atomic_fetchadd_int(&ncp->nc_neghits, 1);
669 if (ncp->nc_flag & NCF_HOTNEGATIVE) {
670 if ((hits % ncneghitsrequeue) != 0)
672 mtx_lock(&ncneg_hot.nl_lock);
673 if (ncp->nc_flag & NCF_HOTNEGATIVE) {
674 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst);
675 TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst);
676 mtx_unlock(&ncneg_hot.nl_lock);
680 * The shrinker cleared the flag and removed the entry from
681 * the hot list. Put it back.
684 mtx_lock(&ncneg_hot.nl_lock);
686 neglist = NCP2NEGLIST(ncp);
687 mtx_lock(&neglist->nl_lock);
688 if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) {
689 TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst);
690 TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst);
691 ncp->nc_flag |= NCF_HOTNEGATIVE;
693 mtx_unlock(&neglist->nl_lock);
694 mtx_unlock(&ncneg_hot.nl_lock);
698 cache_negative_insert(struct namecache *ncp, bool neg_locked)
700 struct neglist *neglist;
702 MPASS(ncp->nc_flag & NCF_NEGATIVE);
703 cache_assert_bucket_locked(ncp, RA_WLOCKED);
704 neglist = NCP2NEGLIST(ncp);
706 mtx_lock(&neglist->nl_lock);
708 mtx_assert(&neglist->nl_lock, MA_OWNED);
710 TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst);
712 mtx_unlock(&neglist->nl_lock);
713 atomic_add_rel_long(&numneg, 1);
717 cache_negative_remove(struct namecache *ncp, bool neg_locked)
719 struct neglist *neglist;
720 bool hot_locked = false;
721 bool list_locked = false;
723 MPASS(ncp->nc_flag & NCF_NEGATIVE);
724 cache_assert_bucket_locked(ncp, RA_WLOCKED);
725 neglist = NCP2NEGLIST(ncp);
727 if (ncp->nc_flag & NCF_HOTNEGATIVE) {
729 mtx_lock(&ncneg_hot.nl_lock);
730 if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) {
732 mtx_lock(&neglist->nl_lock);
736 mtx_lock(&neglist->nl_lock);
739 mtx_assert(&neglist->nl_lock, MA_OWNED);
740 mtx_assert(&ncneg_hot.nl_lock, MA_OWNED);
742 if (ncp->nc_flag & NCF_HOTNEGATIVE) {
743 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst);
745 TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst);
748 mtx_unlock(&neglist->nl_lock);
750 mtx_unlock(&ncneg_hot.nl_lock);
751 atomic_subtract_rel_long(&numneg, 1);
755 cache_negative_shrink_select(int start, struct namecache **ncpp,
756 struct neglist **neglistpp)
758 struct neglist *neglist;
759 struct namecache *ncp;
764 for (i = start; i < numneglists; i++) {
765 neglist = &neglists[i];
766 if (TAILQ_FIRST(&neglist->nl_list) == NULL)
768 mtx_lock(&neglist->nl_lock);
769 ncp = TAILQ_FIRST(&neglist->nl_list);
772 mtx_unlock(&neglist->nl_lock);
775 *neglistpp = neglist;
780 cache_negative_zap_one(void)
782 struct namecache *ncp, *ncp2;
783 struct neglist *neglist;
787 if (!mtx_trylock(&ncneg_shrink_lock))
790 mtx_lock(&ncneg_hot.nl_lock);
791 ncp = TAILQ_FIRST(&ncneg_hot.nl_list);
793 neglist = NCP2NEGLIST(ncp);
794 mtx_lock(&neglist->nl_lock);
795 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst);
796 TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst);
797 ncp->nc_flag &= ~NCF_HOTNEGATIVE;
798 mtx_unlock(&neglist->nl_lock);
801 cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist);
803 if (shrink_list_turn == numneglists)
804 shrink_list_turn = 0;
805 if (ncp == NULL && shrink_list_turn == 0)
806 cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist);
808 mtx_unlock(&ncneg_hot.nl_lock);
812 MPASS(ncp->nc_flag & NCF_NEGATIVE);
813 dvlp = VP2VNODELOCK(ncp->nc_dvp);
814 blp = NCP2BUCKETLOCK(ncp);
815 mtx_unlock(&neglist->nl_lock);
816 mtx_unlock(&ncneg_hot.nl_lock);
819 mtx_lock(&ncneg_hot.nl_lock);
820 mtx_lock(&neglist->nl_lock);
821 ncp2 = TAILQ_FIRST(&neglist->nl_list);
822 if (ncp != ncp2 || dvlp != VP2VNODELOCK(ncp2->nc_dvp) ||
823 blp != NCP2BUCKETLOCK(ncp2) || !(ncp2->nc_flag & NCF_NEGATIVE)) {
827 SDT_PROBE3(vfs, namecache, shrink_negative, done, ncp->nc_dvp,
828 nc_get_name(ncp), ncp->nc_neghits);
830 cache_zap_locked(ncp, true);
832 mtx_unlock(&neglist->nl_lock);
833 mtx_unlock(&ncneg_hot.nl_lock);
837 mtx_unlock(&ncneg_shrink_lock);
842 * cache_zap_locked():
844 * Removes a namecache entry from cache, whether it contains an actual
845 * pointer to a vnode or if it is just a negative cache entry.
848 cache_zap_locked(struct namecache *ncp, bool neg_locked)
851 if (!(ncp->nc_flag & NCF_NEGATIVE))
852 cache_assert_vnode_locked(ncp->nc_vp);
853 cache_assert_vnode_locked(ncp->nc_dvp);
854 cache_assert_bucket_locked(ncp, RA_WLOCKED);
856 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp,
857 (ncp->nc_flag & NCF_NEGATIVE) ? NULL : ncp->nc_vp);
858 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
859 SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp,
860 nc_get_name(ncp), ncp->nc_vp);
862 SDT_PROBE3(vfs, namecache, zap_negative, done, ncp->nc_dvp,
863 nc_get_name(ncp), ncp->nc_neghits);
865 LIST_REMOVE(ncp, nc_hash);
866 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
867 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
868 if (ncp == ncp->nc_vp->v_cache_dd)
869 ncp->nc_vp->v_cache_dd = NULL;
871 cache_negative_remove(ncp, neg_locked);
873 if (ncp->nc_flag & NCF_ISDOTDOT) {
874 if (ncp == ncp->nc_dvp->v_cache_dd)
875 ncp->nc_dvp->v_cache_dd = NULL;
877 LIST_REMOVE(ncp, nc_src);
878 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
879 ncp->nc_flag |= NCF_DVDROP;
880 atomic_subtract_rel_long(&numcachehv, 1);
883 atomic_subtract_rel_long(&numcache, 1);
887 cache_zap_negative_locked_vnode_kl(struct namecache *ncp, struct vnode *vp)
891 MPASS(ncp->nc_dvp == vp);
892 MPASS(ncp->nc_flag & NCF_NEGATIVE);
893 cache_assert_vnode_locked(vp);
895 blp = NCP2BUCKETLOCK(ncp);
897 cache_zap_locked(ncp, false);
902 cache_zap_locked_vnode_kl2(struct namecache *ncp, struct vnode *vp,
905 struct mtx *pvlp, *vlp1, *vlp2, *to_unlock;
908 MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp);
909 cache_assert_vnode_locked(vp);
911 if (ncp->nc_flag & NCF_NEGATIVE) {
916 cache_zap_negative_locked_vnode_kl(ncp, vp);
920 pvlp = VP2VNODELOCK(vp);
921 blp = NCP2BUCKETLOCK(ncp);
922 vlp1 = VP2VNODELOCK(ncp->nc_dvp);
923 vlp2 = VP2VNODELOCK(ncp->nc_vp);
925 if (*vlpp == vlp1 || *vlpp == vlp2) {
933 cache_sort(&vlp1, &vlp2);
938 if (!mtx_trylock(vlp1))
944 cache_zap_locked(ncp, false);
946 if (to_unlock != NULL)
947 mtx_unlock(to_unlock);
954 MPASS(*vlpp == NULL);
960 cache_zap_locked_vnode(struct namecache *ncp, struct vnode *vp)
962 struct mtx *pvlp, *vlp1, *vlp2, *to_unlock;
966 MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp);
967 cache_assert_vnode_locked(vp);
969 pvlp = VP2VNODELOCK(vp);
970 if (ncp->nc_flag & NCF_NEGATIVE) {
971 cache_zap_negative_locked_vnode_kl(ncp, vp);
975 blp = NCP2BUCKETLOCK(ncp);
976 vlp1 = VP2VNODELOCK(ncp->nc_dvp);
977 vlp2 = VP2VNODELOCK(ncp->nc_vp);
978 cache_sort(&vlp1, &vlp2);
983 if (!mtx_trylock(vlp1)) {
990 cache_zap_locked(ncp, false);
992 mtx_unlock(to_unlock);
999 cache_zap_rlocked_bucket(struct namecache *ncp, struct rwlock *blp)
1001 struct mtx *dvlp, *vlp;
1003 cache_assert_bucket_locked(ncp, RA_RLOCKED);
1005 dvlp = VP2VNODELOCK(ncp->nc_dvp);
1007 if (!(ncp->nc_flag & NCF_NEGATIVE))
1008 vlp = VP2VNODELOCK(ncp->nc_vp);
1009 if (cache_trylock_vnodes(dvlp, vlp) == 0) {
1012 cache_zap_locked(ncp, false);
1014 cache_unlock_vnodes(dvlp, vlp);
1023 cache_zap_wlocked_bucket_kl(struct namecache *ncp, struct rwlock *blp,
1024 struct mtx **vlpp1, struct mtx **vlpp2)
1026 struct mtx *dvlp, *vlp;
1028 cache_assert_bucket_locked(ncp, RA_WLOCKED);
1030 dvlp = VP2VNODELOCK(ncp->nc_dvp);
1032 if (!(ncp->nc_flag & NCF_NEGATIVE))
1033 vlp = VP2VNODELOCK(ncp->nc_vp);
1034 cache_sort(&dvlp, &vlp);
1036 if (*vlpp1 == dvlp && *vlpp2 == vlp) {
1037 cache_zap_locked(ncp, false);
1038 cache_unlock_vnodes(dvlp, vlp);
1051 if (cache_trylock_vnodes(dvlp, vlp) == 0) {
1052 cache_zap_locked(ncp, false);
1053 cache_unlock_vnodes(dvlp, vlp);
1068 cache_lookup_unlock(struct rwlock *blp, struct mtx *vlp)
1073 mtx_assert(vlp, MA_NOTOWNED);
1080 * Lookup an entry in the cache
1082 * Lookup is called with dvp pointing to the directory to search,
1083 * cnp pointing to the name of the entry being sought. If the lookup
1084 * succeeds, the vnode is returned in *vpp, and a status of -1 is
1085 * returned. If the lookup determines that the name does not exist
1086 * (negative caching), a status of ENOENT is returned. If the lookup
1087 * fails, a status of zero is returned. If the directory vnode is
1088 * recycled out from under us due to a forced unmount, a status of
1089 * ENOENT is returned.
1091 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
1092 * unlocked. If we're looking up . an extra ref is taken, but the lock is
1093 * not recursively acquired.
1097 cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1098 struct timespec *tsp, int *ticksp)
1100 struct namecache *ncp;
1102 struct mtx *dvlp, *dvlp2;
1106 if (__predict_false(!doingcache)) {
1107 cnp->cn_flags &= ~MAKEENTRY;
1112 dvlp = VP2VNODELOCK(dvp);
1114 counter_u64_add(numcalls, 1);
1116 if (cnp->cn_nameptr[0] == '.') {
1117 if (cnp->cn_namelen == 1) {
1119 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
1120 dvp, cnp->cn_nameptr);
1121 counter_u64_add(dothits, 1);
1122 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp);
1129 * When we lookup "." we still can be asked to lock it
1132 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
1133 if (ltype != VOP_ISLOCKED(*vpp)) {
1134 if (ltype == LK_EXCLUSIVE) {
1135 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
1136 if ((*vpp)->v_iflag & VI_DOOMED) {
1137 /* forced unmount */
1143 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
1147 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
1148 counter_u64_add(dotdothits, 1);
1152 ncp = dvp->v_cache_dd;
1154 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
1159 if ((cnp->cn_flags & MAKEENTRY) == 0) {
1160 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) {
1161 if (ncp->nc_dvp != dvp)
1162 panic("dvp %p v_cache_dd %p\n", dvp, ncp);
1163 if (!cache_zap_locked_vnode_kl2(ncp,
1166 MPASS(dvp->v_cache_dd == NULL);
1172 dvp->v_cache_dd = NULL;
1179 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) {
1180 if (ncp->nc_flag & NCF_NEGATIVE)
1186 /* Return failure if negative entry was found. */
1188 goto negative_success;
1189 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
1190 dvp, cnp->cn_nameptr, *vpp);
1191 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..",
1193 cache_out_ts(ncp, tsp, ticksp);
1194 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
1195 NCF_DTS && tsp != NULL)
1196 *tsp = ((struct namecache_ts *)ncp)->
1202 hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp);
1203 blp = HASH2BUCKETLOCK(hash);
1206 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
1207 counter_u64_add(numchecks, 1);
1208 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
1209 !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen))
1213 /* We failed to find an entry */
1215 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
1217 if ((cnp->cn_flags & MAKEENTRY) == 0) {
1218 counter_u64_add(nummisszap, 1);
1220 counter_u64_add(nummiss, 1);
1225 /* We don't want to have an entry, so dump it */
1226 if ((cnp->cn_flags & MAKEENTRY) == 0) {
1227 counter_u64_add(numposzaps, 1);
1231 /* We found a "positive" match, return the vnode */
1232 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
1233 counter_u64_add(numposhits, 1);
1235 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
1236 dvp, cnp->cn_nameptr, *vpp, ncp);
1237 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp),
1239 cache_out_ts(ncp, tsp, ticksp);
1244 /* We found a negative match, and want to create it, so purge */
1245 if (cnp->cn_nameiop == CREATE) {
1246 counter_u64_add(numnegzaps, 1);
1250 counter_u64_add(numneghits, 1);
1251 cache_negative_hit(ncp);
1252 if (ncp->nc_flag & NCF_WHITE)
1253 cnp->cn_flags |= ISWHITEOUT;
1254 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp,
1256 cache_out_ts(ncp, tsp, ticksp);
1257 cache_lookup_unlock(blp, dvlp);
1262 * On success we return a locked and ref'd vnode as per the lookup
1266 ltype = 0; /* silence gcc warning */
1267 if (cnp->cn_flags & ISDOTDOT) {
1268 ltype = VOP_ISLOCKED(dvp);
1272 cache_lookup_unlock(blp, dvlp);
1273 error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread);
1274 if (cnp->cn_flags & ISDOTDOT) {
1275 vn_lock(dvp, ltype | LK_RETRY);
1276 if (dvp->v_iflag & VI_DOOMED) {
1287 if ((cnp->cn_flags & ISLASTCN) &&
1288 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
1289 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
1294 cache_lookup_unlock(blp, dvlp);
1299 error = cache_zap_rlocked_bucket(ncp, blp);
1301 error = cache_zap_locked_vnode(ncp, dvp);
1303 zap_and_exit_bucket_fail++;
1304 cache_maybe_yield();
1311 struct celockstate {
1313 struct rwlock *blp[2];
1315 CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3));
1316 CTASSERT((nitems(((struct celockstate *)0)->blp) == 2));
1319 cache_celockstate_init(struct celockstate *cel)
1322 bzero(cel, sizeof(*cel));
1326 cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp,
1329 struct mtx *vlp1, *vlp2;
1331 MPASS(cel->vlp[0] == NULL);
1332 MPASS(cel->vlp[1] == NULL);
1333 MPASS(cel->vlp[2] == NULL);
1335 MPASS(vp != NULL || dvp != NULL);
1337 vlp1 = VP2VNODELOCK(vp);
1338 vlp2 = VP2VNODELOCK(dvp);
1339 cache_sort(&vlp1, &vlp2);
1350 cache_unlock_vnodes_cel(struct celockstate *cel)
1353 MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL);
1355 if (cel->vlp[0] != NULL)
1356 mtx_unlock(cel->vlp[0]);
1357 if (cel->vlp[1] != NULL)
1358 mtx_unlock(cel->vlp[1]);
1359 if (cel->vlp[2] != NULL)
1360 mtx_unlock(cel->vlp[2]);
1364 cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp)
1369 cache_assert_vlp_locked(cel->vlp[0]);
1370 cache_assert_vlp_locked(cel->vlp[1]);
1371 MPASS(cel->vlp[2] == NULL);
1374 vlp = VP2VNODELOCK(vp);
1377 if (vlp >= cel->vlp[1]) {
1380 if (mtx_trylock(vlp))
1382 cache_lock_vnodes_cel_3_failures++;
1383 cache_unlock_vnodes_cel(cel);
1384 if (vlp < cel->vlp[0]) {
1386 mtx_lock(cel->vlp[0]);
1387 mtx_lock(cel->vlp[1]);
1389 if (cel->vlp[0] != NULL)
1390 mtx_lock(cel->vlp[0]);
1392 mtx_lock(cel->vlp[1]);
1402 cache_lock_buckets_cel(struct celockstate *cel, struct rwlock *blp1,
1403 struct rwlock *blp2)
1406 MPASS(cel->blp[0] == NULL);
1407 MPASS(cel->blp[1] == NULL);
1409 cache_sort(&blp1, &blp2);
1420 cache_unlock_buckets_cel(struct celockstate *cel)
1423 if (cel->blp[0] != NULL)
1424 rw_wunlock(cel->blp[0]);
1425 rw_wunlock(cel->blp[1]);
1429 * Lock part of the cache affected by the insertion.
1431 * This means vnodelocks for dvp, vp and the relevant bucketlock.
1432 * However, insertion can result in removal of an old entry. In this
1433 * case we have an additional vnode and bucketlock pair to lock. If the
1434 * entry is negative, ncelock is locked instead of the vnode.
1436 * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while
1437 * preserving the locking order (smaller address first).
1440 cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp,
1443 struct namecache *ncp;
1444 struct rwlock *blps[2];
1446 blps[0] = HASH2BUCKETLOCK(hash);
1449 cache_lock_vnodes_cel(cel, dvp, vp);
1450 if (vp == NULL || vp->v_type != VDIR)
1452 ncp = vp->v_cache_dd;
1455 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1457 MPASS(ncp->nc_dvp == vp);
1458 blps[1] = NCP2BUCKETLOCK(ncp);
1459 if (ncp->nc_flag & NCF_NEGATIVE)
1461 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp))
1464 * All vnodes got re-locked. Re-validate the state and if
1465 * nothing changed we are done. Otherwise restart.
1467 if (ncp == vp->v_cache_dd &&
1468 (ncp->nc_flag & NCF_ISDOTDOT) != 0 &&
1469 blps[1] == NCP2BUCKETLOCK(ncp) &&
1470 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2])
1472 cache_unlock_vnodes_cel(cel);
1477 cache_lock_buckets_cel(cel, blps[0], blps[1]);
1481 cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp,
1484 struct namecache *ncp;
1485 struct rwlock *blps[2];
1487 blps[0] = HASH2BUCKETLOCK(hash);
1490 cache_lock_vnodes_cel(cel, dvp, vp);
1491 ncp = dvp->v_cache_dd;
1494 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1496 MPASS(ncp->nc_dvp == dvp);
1497 blps[1] = NCP2BUCKETLOCK(ncp);
1498 if (ncp->nc_flag & NCF_NEGATIVE)
1500 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp))
1502 if (ncp == dvp->v_cache_dd &&
1503 (ncp->nc_flag & NCF_ISDOTDOT) != 0 &&
1504 blps[1] == NCP2BUCKETLOCK(ncp) &&
1505 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2])
1507 cache_unlock_vnodes_cel(cel);
1512 cache_lock_buckets_cel(cel, blps[0], blps[1]);
1516 cache_enter_unlock(struct celockstate *cel)
1519 cache_unlock_buckets_cel(cel);
1520 cache_unlock_vnodes_cel(cel);
1524 * Add an entry to the cache.
1527 cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp,
1528 struct timespec *tsp, struct timespec *dtsp)
1530 struct celockstate cel;
1531 struct namecache *ncp, *n2, *ndd;
1532 struct namecache_ts *n3;
1533 struct nchashhead *ncpp;
1534 struct neglist *neglist;
1540 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
1541 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
1542 ("cache_enter: Adding a doomed vnode"));
1543 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
1544 ("cache_enter: Doomed vnode used as src"));
1546 if (__predict_false(!doingcache))
1550 * Avoid blowout in namecache entries.
1552 if (__predict_false(numcache >= desiredvnodes * ncsizefactor))
1555 cache_celockstate_init(&cel);
1558 if (cnp->cn_nameptr[0] == '.') {
1559 if (cnp->cn_namelen == 1)
1561 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
1562 len = cnp->cn_namelen;
1563 hash = cache_get_hash(cnp->cn_nameptr, len, dvp);
1564 cache_enter_lock_dd(&cel, dvp, vp, hash);
1566 * If dotdot entry already exists, just retarget it
1567 * to new parent vnode, otherwise continue with new
1568 * namecache entry allocation.
1570 if ((ncp = dvp->v_cache_dd) != NULL &&
1571 ncp->nc_flag & NCF_ISDOTDOT) {
1572 KASSERT(ncp->nc_dvp == dvp,
1573 ("wrong isdotdot parent"));
1575 if (ncp->nc_flag & NCF_NEGATIVE || vp == NULL) {
1576 neglist = NCP2NEGLIST(ncp);
1577 mtx_lock(&ncneg_hot.nl_lock);
1578 mtx_lock(&neglist->nl_lock);
1581 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
1582 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
1585 cache_negative_remove(ncp, true);
1588 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
1590 ncp->nc_flag &= ~(NCF_NEGATIVE|NCF_HOTNEGATIVE);
1592 ncp->nc_flag &= ~(NCF_HOTNEGATIVE);
1593 ncp->nc_flag |= NCF_NEGATIVE;
1594 cache_negative_insert(ncp, true);
1597 mtx_unlock(&neglist->nl_lock);
1598 mtx_unlock(&ncneg_hot.nl_lock);
1601 cache_enter_unlock(&cel);
1604 dvp->v_cache_dd = NULL;
1605 cache_enter_unlock(&cel);
1606 cache_celockstate_init(&cel);
1607 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp);
1608 flag = NCF_ISDOTDOT;
1613 * Calculate the hash key and setup as much of the new
1614 * namecache entry as possible before acquiring the lock.
1616 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
1617 ncp->nc_flag = flag;
1620 ncp->nc_flag |= NCF_NEGATIVE;
1623 n3 = (struct namecache_ts *)ncp;
1625 n3->nc_ticks = ticks;
1626 n3->nc_flag |= NCF_TS;
1628 n3->nc_dotdottime = *dtsp;
1629 n3->nc_flag |= NCF_DTS;
1632 len = ncp->nc_nlen = cnp->cn_namelen;
1633 hash = cache_get_hash(cnp->cn_nameptr, len, dvp);
1634 strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1);
1635 cache_enter_lock(&cel, dvp, vp, hash);
1638 * See if this vnode or negative entry is already in the cache
1639 * with this name. This can happen with concurrent lookups of
1640 * the same path name.
1642 ncpp = NCHHASH(hash);
1643 LIST_FOREACH(n2, ncpp, nc_hash) {
1644 if (n2->nc_dvp == dvp &&
1645 n2->nc_nlen == cnp->cn_namelen &&
1646 !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) {
1648 KASSERT((n2->nc_flag & NCF_TS) != 0,
1650 n3 = (struct namecache_ts *)n2;
1652 ((struct namecache_ts *)ncp)->nc_time;
1654 ((struct namecache_ts *)ncp)->nc_ticks;
1657 ((struct namecache_ts *)ncp)->
1659 if (ncp->nc_flag & NCF_NEGATIVE)
1660 mtx_lock(&ncneg_hot.nl_lock);
1661 n3->nc_flag |= NCF_DTS;
1662 if (ncp->nc_flag & NCF_NEGATIVE)
1663 mtx_unlock(&ncneg_hot.nl_lock);
1666 goto out_unlock_free;
1670 if (flag == NCF_ISDOTDOT) {
1672 * See if we are trying to add .. entry, but some other lookup
1673 * has populated v_cache_dd pointer already.
1675 if (dvp->v_cache_dd != NULL)
1676 goto out_unlock_free;
1677 KASSERT(vp == NULL || vp->v_type == VDIR,
1678 ("wrong vnode type %p", vp));
1679 dvp->v_cache_dd = ncp;
1682 atomic_add_rel_long(&numcache, 1);
1684 if (vp->v_type == VDIR) {
1685 if (flag != NCF_ISDOTDOT) {
1687 * For this case, the cache entry maps both the
1688 * directory name in it and the name ".." for the
1689 * directory's parent.
1691 if ((ndd = vp->v_cache_dd) != NULL) {
1692 if ((ndd->nc_flag & NCF_ISDOTDOT) != 0)
1693 cache_zap_locked(ndd, false);
1697 vp->v_cache_dd = ncp;
1700 vp->v_cache_dd = NULL;
1704 if (flag != NCF_ISDOTDOT) {
1705 if (LIST_EMPTY(&dvp->v_cache_src)) {
1707 atomic_add_rel_long(&numcachehv, 1);
1709 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
1713 * Insert the new namecache entry into the appropriate chain
1714 * within the cache entries table.
1716 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
1719 * If the entry is "negative", we place it into the
1720 * "negative" cache queue, otherwise, we place it into the
1721 * destination vnode's cache entries queue.
1724 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
1725 SDT_PROBE3(vfs, namecache, enter, done, dvp, nc_get_name(ncp),
1728 if (cnp->cn_flags & ISWHITEOUT)
1729 ncp->nc_flag |= NCF_WHITE;
1730 cache_negative_insert(ncp, false);
1731 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp,
1734 cache_enter_unlock(&cel);
1735 if (numneg * ncnegfactor > numcache)
1736 cache_negative_zap_one();
1740 cache_enter_unlock(&cel);
1746 cache_roundup_2(u_int val)
1750 for (res = 1; res <= val; res <<= 1)
1757 * Name cache initialization, from vfs_init() when we are booting
1760 nchinit(void *dummy __unused)
1764 cache_zone_small = uma_zcreate("S VFS Cache",
1765 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
1766 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
1767 cache_zone_small_ts = uma_zcreate("STS VFS Cache",
1768 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
1769 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
1770 cache_zone_large = uma_zcreate("L VFS Cache",
1771 sizeof(struct namecache) + NAME_MAX + 1,
1772 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
1773 cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
1774 sizeof(struct namecache_ts) + NAME_MAX + 1,
1775 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
1777 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
1778 ncbuckethash = cache_roundup_2(mp_ncpus * 64) - 1;
1779 if (ncbuckethash > nchash)
1780 ncbuckethash = nchash;
1781 bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE,
1783 for (i = 0; i < numbucketlocks; i++)
1784 rw_init_flags(&bucketlocks[i], "ncbuc", RW_DUPOK | RW_RECURSE);
1785 ncvnodehash = cache_roundup_2(mp_ncpus * 64) - 1;
1786 vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE,
1788 for (i = 0; i < numvnodelocks; i++)
1789 mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE);
1790 ncpurgeminvnodes = numbucketlocks;
1793 neglists = malloc(sizeof(*neglists) * numneglists, M_VFSCACHE,
1795 for (i = 0; i < numneglists; i++) {
1796 mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF);
1797 TAILQ_INIT(&neglists[i].nl_list);
1799 mtx_init(&ncneg_hot.nl_lock, "ncneglh", NULL, MTX_DEF);
1800 TAILQ_INIT(&ncneg_hot.nl_list);
1802 mtx_init(&ncneg_shrink_lock, "ncnegs", NULL, MTX_DEF);
1804 numcalls = counter_u64_alloc(M_WAITOK);
1805 dothits = counter_u64_alloc(M_WAITOK);
1806 dotdothits = counter_u64_alloc(M_WAITOK);
1807 numchecks = counter_u64_alloc(M_WAITOK);
1808 nummiss = counter_u64_alloc(M_WAITOK);
1809 nummisszap = counter_u64_alloc(M_WAITOK);
1810 numposzaps = counter_u64_alloc(M_WAITOK);
1811 numposhits = counter_u64_alloc(M_WAITOK);
1812 numnegzaps = counter_u64_alloc(M_WAITOK);
1813 numneghits = counter_u64_alloc(M_WAITOK);
1814 numfullpathcalls = counter_u64_alloc(M_WAITOK);
1815 numfullpathfail1 = counter_u64_alloc(M_WAITOK);
1816 numfullpathfail2 = counter_u64_alloc(M_WAITOK);
1817 numfullpathfail4 = counter_u64_alloc(M_WAITOK);
1818 numfullpathfound = counter_u64_alloc(M_WAITOK);
1820 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
1823 cache_changesize(int newmaxvnodes)
1825 struct nchashhead *new_nchashtbl, *old_nchashtbl;
1826 u_long new_nchash, old_nchash;
1827 struct namecache *ncp;
1831 newmaxvnodes = cache_roundup_2(newmaxvnodes * 2);
1832 if (newmaxvnodes < numbucketlocks)
1833 newmaxvnodes = numbucketlocks;
1835 new_nchashtbl = hashinit(newmaxvnodes, M_VFSCACHE, &new_nchash);
1836 /* If same hash table size, nothing to do */
1837 if (nchash == new_nchash) {
1838 free(new_nchashtbl, M_VFSCACHE);
1842 * Move everything from the old hash table to the new table.
1843 * None of the namecache entries in the table can be removed
1844 * because to do so, they have to be removed from the hash table.
1846 cache_lock_all_vnodes();
1847 cache_lock_all_buckets();
1848 old_nchashtbl = nchashtbl;
1849 old_nchash = nchash;
1850 nchashtbl = new_nchashtbl;
1851 nchash = new_nchash;
1852 for (i = 0; i <= old_nchash; i++) {
1853 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) {
1854 hash = cache_get_hash(nc_get_name(ncp), ncp->nc_nlen,
1856 LIST_REMOVE(ncp, nc_hash);
1857 LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash);
1860 cache_unlock_all_buckets();
1861 cache_unlock_all_vnodes();
1862 free(old_nchashtbl, M_VFSCACHE);
1866 * Invalidate all entries to a particular vnode.
1869 cache_purge(struct vnode *vp)
1871 TAILQ_HEAD(, namecache) ncps;
1872 struct namecache *ncp, *nnp;
1873 struct mtx *vlp, *vlp2;
1875 CTR1(KTR_VFS, "cache_purge(%p)", vp);
1876 SDT_PROBE1(vfs, namecache, purge, done, vp);
1877 if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) &&
1878 vp->v_cache_dd == NULL)
1881 vlp = VP2VNODELOCK(vp);
1885 while (!LIST_EMPTY(&vp->v_cache_src)) {
1886 ncp = LIST_FIRST(&vp->v_cache_src);
1887 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1889 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1891 while (!TAILQ_EMPTY(&vp->v_cache_dst)) {
1892 ncp = TAILQ_FIRST(&vp->v_cache_dst);
1893 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1895 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1897 ncp = vp->v_cache_dd;
1899 KASSERT(ncp->nc_flag & NCF_ISDOTDOT,
1900 ("lost dotdot link"));
1901 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1903 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1905 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
1909 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
1915 * Invalidate all negative entries for a particular directory vnode.
1918 cache_purge_negative(struct vnode *vp)
1920 TAILQ_HEAD(, namecache) ncps;
1921 struct namecache *ncp, *nnp;
1924 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
1925 SDT_PROBE1(vfs, namecache, purge_negative, done, vp);
1927 vlp = VP2VNODELOCK(vp);
1929 LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) {
1930 if (!(ncp->nc_flag & NCF_NEGATIVE))
1932 cache_zap_negative_locked_vnode_kl(ncp, vp);
1933 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1936 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
1942 * Flush all entries referencing a particular filesystem.
1945 cache_purgevfs(struct mount *mp, bool force)
1947 TAILQ_HEAD(, namecache) ncps;
1948 struct mtx *vlp1, *vlp2;
1950 struct nchashhead *bucket;
1951 struct namecache *ncp, *nnp;
1952 u_long i, j, n_nchash;
1955 /* Scan hash tables for applicable entries */
1956 SDT_PROBE1(vfs, namecache, purgevfs, done, mp);
1957 if (!force && mp->mnt_nvnodelistsize <= ncpurgeminvnodes)
1960 n_nchash = nchash + 1;
1962 for (i = 0; i < numbucketlocks; i++) {
1963 blp = (struct rwlock *)&bucketlocks[i];
1965 for (j = i; j < n_nchash; j += numbucketlocks) {
1967 bucket = &nchashtbl[j];
1968 LIST_FOREACH_SAFE(ncp, bucket, nc_hash, nnp) {
1969 cache_assert_bucket_locked(ncp, RA_WLOCKED);
1970 if (ncp->nc_dvp->v_mount != mp)
1972 error = cache_zap_wlocked_bucket_kl(ncp, blp,
1976 TAILQ_INSERT_HEAD(&ncps, ncp, nc_dst);
1980 if (vlp1 == NULL && vlp2 == NULL)
1981 cache_maybe_yield();
1988 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
1994 * Perform canonical checks and cache lookup and pass on to filesystem
1995 * through the vop_cachedlookup only if needed.
1999 vfs_cache_lookup(struct vop_lookup_args *ap)
2003 struct vnode **vpp = ap->a_vpp;
2004 struct componentname *cnp = ap->a_cnp;
2005 struct ucred *cred = cnp->cn_cred;
2006 int flags = cnp->cn_flags;
2007 struct thread *td = cnp->cn_thread;
2012 if (dvp->v_type != VDIR)
2015 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
2016 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
2019 error = VOP_ACCESS(dvp, VEXEC, cred, td);
2023 error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
2025 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
2032 * XXX All of these sysctls would probably be more productive dead.
2034 static int __read_mostly disablecwd;
2035 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
2036 "Disable the getcwd syscall");
2038 /* Implementation of the getcwd syscall. */
2040 sys___getcwd(struct thread *td, struct __getcwd_args *uap)
2043 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen,
2048 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen,
2052 struct filedesc *fdp;
2053 struct vnode *cdir, *rdir;
2056 if (__predict_false(disablecwd))
2058 if (__predict_false(buflen < 2))
2060 if (buflen > path_max)
2063 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
2064 fdp = td->td_proc->p_fd;
2065 FILEDESC_SLOCK(fdp);
2066 cdir = fdp->fd_cdir;
2068 rdir = fdp->fd_rdir;
2070 FILEDESC_SUNLOCK(fdp);
2071 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
2076 if (bufseg == UIO_SYSSPACE)
2077 bcopy(bp, buf, strlen(bp) + 1);
2079 error = copyout(bp, buf, strlen(bp) + 1);
2081 if (KTRPOINT(curthread, KTR_NAMEI))
2085 free(tmpbuf, M_TEMP);
2090 * Thus begins the fullpath magic.
2093 static int __read_mostly disablefullpath;
2094 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
2095 "Disable the vn_fullpath function");
2098 * Retrieve the full filesystem path that correspond to a vnode from the name
2099 * cache (if available)
2102 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
2105 struct filedesc *fdp;
2109 if (__predict_false(disablefullpath))
2111 if (__predict_false(vn == NULL))
2114 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
2115 fdp = td->td_proc->p_fd;
2116 FILEDESC_SLOCK(fdp);
2117 rdir = fdp->fd_rdir;
2119 FILEDESC_SUNLOCK(fdp);
2120 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
2131 * This function is similar to vn_fullpath, but it attempts to lookup the
2132 * pathname relative to the global root mount point. This is required for the
2133 * auditing sub-system, as audited pathnames must be absolute, relative to the
2134 * global root mount point.
2137 vn_fullpath_global(struct thread *td, struct vnode *vn,
2138 char **retbuf, char **freebuf)
2143 if (__predict_false(disablefullpath))
2145 if (__predict_false(vn == NULL))
2147 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
2148 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
2157 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
2160 struct namecache *ncp;
2164 vlp = VP2VNODELOCK(*vp);
2166 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
2167 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
2171 if (*buflen < ncp->nc_nlen) {
2174 counter_u64_add(numfullpathfail4, 1);
2176 SDT_PROBE3(vfs, namecache, fullpath, return, error,
2180 *buflen -= ncp->nc_nlen;
2181 memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen);
2182 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp,
2183 nc_get_name(ncp), vp);
2191 SDT_PROBE1(vfs, namecache, fullpath, miss, vp);
2194 vn_lock(*vp, LK_SHARED | LK_RETRY);
2195 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
2198 counter_u64_add(numfullpathfail2, 1);
2199 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
2204 if (dvp->v_iflag & VI_DOOMED) {
2205 /* forced unmount */
2208 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
2212 * *vp has its use count incremented still.
2219 * The magic behind kern___getcwd() and vn_fullpath().
2222 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
2223 char *buf, char **retbuf, u_int buflen)
2225 int error, slash_prefixed;
2226 #ifdef KDTRACE_HOOKS
2227 struct vnode *startvp = vp;
2236 SDT_PROBE1(vfs, namecache, fullpath, entry, vp);
2237 counter_u64_add(numfullpathcalls, 1);
2239 if (vp->v_type != VDIR) {
2240 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen);
2247 buf[--buflen] = '/';
2250 while (vp != rdir && vp != rootvnode) {
2252 * The vp vnode must be already fully constructed,
2253 * since it is either found in namecache or obtained
2254 * from VOP_VPTOCNP(). We may test for VV_ROOT safely
2255 * without obtaining the vnode lock.
2257 if ((vp->v_vflag & VV_ROOT) != 0) {
2258 vn_lock(vp, LK_RETRY | LK_SHARED);
2261 * With the vnode locked, check for races with
2262 * unmount, forced or not. Note that we
2263 * already verified that vp is not equal to
2264 * the root vnode, which means that
2265 * mnt_vnodecovered can be NULL only for the
2268 if ((vp->v_iflag & VI_DOOMED) != 0 ||
2269 (vp1 = vp->v_mount->mnt_vnodecovered) == NULL ||
2270 vp1->v_mountedhere != vp->v_mount) {
2273 SDT_PROBE3(vfs, namecache, fullpath, return,
2283 if (vp->v_type != VDIR) {
2285 counter_u64_add(numfullpathfail1, 1);
2287 SDT_PROBE3(vfs, namecache, fullpath, return,
2291 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen);
2297 SDT_PROBE3(vfs, namecache, fullpath, return, error,
2301 buf[--buflen] = '/';
2306 if (!slash_prefixed) {
2309 counter_u64_add(numfullpathfail4, 1);
2310 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM,
2314 buf[--buflen] = '/';
2316 counter_u64_add(numfullpathfound, 1);
2319 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen);
2320 *retbuf = buf + buflen;
2325 vn_dir_dd_ino(struct vnode *vp)
2327 struct namecache *ncp;
2331 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
2332 vlp = VP2VNODELOCK(vp);
2334 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
2335 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
2340 if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread))
2349 vn_commname(struct vnode *vp, char *buf, u_int buflen)
2351 struct namecache *ncp;
2355 vlp = VP2VNODELOCK(vp);
2357 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
2358 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
2364 l = min(ncp->nc_nlen, buflen - 1);
2365 memcpy(buf, nc_get_name(ncp), l);
2371 /* ABI compat shims for old kernel modules. */
2374 void cache_enter(struct vnode *dvp, struct vnode *vp,
2375 struct componentname *cnp);
2378 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2381 cache_enter_time(dvp, vp, cnp, NULL, NULL);
2385 * This function updates path string to vnode's full global path
2386 * and checks the size of the new path string against the pathlen argument.
2388 * Requires a locked, referenced vnode.
2389 * Vnode is re-locked on success or ENODEV, otherwise unlocked.
2391 * If sysctl debug.disablefullpath is set, ENODEV is returned,
2392 * vnode is left locked and path remain untouched.
2394 * If vp is a directory, the call to vn_fullpath_global() always succeeds
2395 * because it falls back to the ".." lookup if the namecache lookup fails.
2398 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
2401 struct nameidata nd;
2406 ASSERT_VOP_ELOCKED(vp, __func__);
2408 /* Return ENODEV if sysctl debug.disablefullpath==1 */
2409 if (__predict_false(disablefullpath))
2412 /* Construct global filesystem path from vp. */
2414 error = vn_fullpath_global(td, vp, &rpath, &fbuf);
2421 if (strlen(rpath) >= pathlen) {
2423 error = ENAMETOOLONG;
2428 * Re-lookup the vnode by path to detect a possible rename.
2429 * As a side effect, the vnode is relocked.
2430 * If vnode was renamed, return ENOENT.
2432 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
2433 UIO_SYSSPACE, path, td);
2439 NDFREE(&nd, NDF_ONLY_PNBUF);
2443 strcpy(path, rpath);