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 * 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.
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/filedesc.h>
43 #include <sys/fnv_hash.h>
44 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/fcntl.h>
48 #include <sys/mount.h>
49 #include <sys/namei.h>
51 #include <sys/rwlock.h>
53 #include <sys/syscallsubr.h>
54 #include <sys/sysctl.h>
55 #include <sys/sysproto.h>
56 #include <sys/vnode.h>
58 #include <sys/ktrace.h>
63 SDT_PROVIDER_DECLARE(vfs);
64 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
66 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
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 *",
76 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative,
77 "struct vnode *", "char *");
78 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
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 *",
85 SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *",
89 * This structure describes the elements in the cache of recent
90 * names looked up by namei.
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 */
105 * struct namecache_ts repeats struct namecache layout up to the
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
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 */
127 * Flags in namecache.nc_flag
129 #define NCF_WHITE 0x01
130 #define NCF_ISDOTDOT 0x02
135 * Name caching works as follows:
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
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.
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.
152 * Structures associated with name cacheing.
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");
177 struct nchstats nchstats; /* cache effectiveness statistics */
179 static struct rwlock cache_lock;
180 RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
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)
189 * UMA zones for the VFS cache.
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.
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;
200 #define CACHE_PATH_CUTOFF 35
202 static struct namecache *
203 cache_alloc(int len, int ts)
206 if (len > CACHE_PATH_CUTOFF) {
208 return (uma_zalloc(cache_zone_large_ts, M_WAITOK));
210 return (uma_zalloc(cache_zone_large, M_WAITOK));
213 return (uma_zalloc(cache_zone_small_ts, M_WAITOK));
215 return (uma_zalloc(cache_zone_small, M_WAITOK));
219 cache_free(struct namecache *ncp)
225 ts = ncp->nc_flag & NCF_TS;
226 if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) {
228 uma_zfree(cache_zone_small_ts, ncp);
230 uma_zfree(cache_zone_small, ncp);
232 uma_zfree(cache_zone_large_ts, ncp);
234 uma_zfree(cache_zone_large, ncp);
238 nc_get_name(struct namecache *ncp)
240 struct namecache_ts *ncp_ts;
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);
249 cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp)
252 KASSERT((ncp->nc_flag & NCF_TS) != 0 ||
253 (tsp == NULL && ticksp == NULL),
257 *tsp = ((struct namecache_ts *)ncp)->nc_time;
259 *ticksp = ((struct namecache_ts *)ncp)->nc_ticks;
262 static int doingcache = 1; /* 1 => enable the cache */
263 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
264 "VFS namecache enabled");
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)");
271 * The new name cache statistics
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)");
302 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
303 &nchstats, sizeof(nchstats), "LU",
304 "VFS cache effectiveness statistics");
308 static void cache_zap(struct namecache *ncp);
309 static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
311 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
312 char *buf, char **retbuf, u_int buflen);
314 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
318 * Grab an atomic snapshot of the name cache hash chain lengths
320 static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
324 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
326 struct nchashhead *ncpp;
327 struct namecache *ncp;
328 int i, error, n_nchash, *cntbuf;
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);
336 if (n_nchash != nchash + 1) {
338 free(cntbuf, M_TEMP);
341 /* Scan hash tables counting entries */
342 for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++)
343 LIST_FOREACH(ncp, ncpp, nc_hash)
346 for (error = 0, i = 0; i < n_nchash; i++)
347 if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0)
349 free(cntbuf, M_TEMP);
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");
357 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
360 struct nchashhead *ncpp;
361 struct namecache *ncp;
363 int count, maxlength, used, pct;
366 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
369 n_nchash = nchash + 1; /* nchash is max index, not count */
373 /* Scan hash tables for applicable entries */
374 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
376 LIST_FOREACH(ncp, ncpp, nc_hash) {
381 if (maxlength < count)
384 n_nchash = nchash + 1;
386 pct = (used * 100) / (n_nchash / 100);
387 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
390 error = SYSCTL_OUT(req, &used, sizeof(used));
393 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
396 error = SYSCTL_OUT(req, &pct, sizeof(pct));
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)");
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.
414 struct namecache *ncp;
418 rw_assert(&cache_lock, RA_WLOCKED);
419 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
420 if (ncp->nc_vp != NULL) {
421 SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp,
422 nc_get_name(ncp), ncp->nc_vp);
424 SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp,
428 LIST_REMOVE(ncp, nc_hash);
429 if (ncp->nc_flag & NCF_ISDOTDOT) {
430 if (ncp == ncp->nc_dvp->v_cache_dd)
431 ncp->nc_dvp->v_cache_dd = NULL;
433 LIST_REMOVE(ncp, nc_src);
434 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
440 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
441 if (ncp == ncp->nc_vp->v_cache_dd)
442 ncp->nc_vp->v_cache_dd = NULL;
444 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
454 * Lookup an entry in the cache
456 * Lookup is called with dvp pointing to the directory to search,
457 * cnp pointing to the name of the entry being sought. If the lookup
458 * succeeds, the vnode is returned in *vpp, and a status of -1 is
459 * returned. If the lookup determines that the name does not exist
460 * (negative cacheing), a status of ENOENT is returned. If the lookup
461 * fails, a status of zero is returned. If the directory vnode is
462 * recycled out from under us due to a forced unmount, a status of
463 * ENOENT is returned.
465 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
466 * unlocked. If we're looking up . an extra ref is taken, but the lock is
467 * not recursively acquired.
471 cache_lookup(dvp, vpp, cnp, tsp, ticksp)
474 struct componentname *cnp;
475 struct timespec *tsp;
478 struct namecache *ncp;
480 int error, ltype, wlocked;
483 cnp->cn_flags &= ~MAKEENTRY;
493 if (cnp->cn_nameptr[0] == '.') {
494 if (cnp->cn_namelen == 1) {
496 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
497 dvp, cnp->cn_nameptr);
499 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp);
506 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
508 if (dvp->v_cache_dd == NULL) {
509 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
513 if ((cnp->cn_flags & MAKEENTRY) == 0) {
514 if (!wlocked && !CACHE_UPGRADE_LOCK())
516 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
517 cache_zap(dvp->v_cache_dd);
518 dvp->v_cache_dd = NULL;
522 ncp = dvp->v_cache_dd;
523 if (ncp->nc_flag & NCF_ISDOTDOT)
527 /* Return failure if negative entry was found. */
529 goto negative_success;
530 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
531 dvp, cnp->cn_nameptr, *vpp);
532 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..",
534 cache_out_ts(ncp, tsp, ticksp);
535 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
536 NCF_DTS && tsp != NULL)
537 *tsp = ((struct namecache_ts *)ncp)->
543 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
544 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
545 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
547 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
548 !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen))
552 /* We failed to find an entry */
554 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
556 if ((cnp->cn_flags & MAKEENTRY) == 0) {
565 /* We don't want to have an entry, so dump it */
566 if ((cnp->cn_flags & MAKEENTRY) == 0) {
568 nchstats.ncs_badhits++;
569 if (!wlocked && !CACHE_UPGRADE_LOCK())
576 /* We found a "positive" match, return the vnode */
579 nchstats.ncs_goodhits++;
581 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
582 dvp, cnp->cn_nameptr, *vpp, ncp);
583 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp),
585 cache_out_ts(ncp, tsp, ticksp);
590 /* We found a negative match, and want to create it, so purge */
591 if (cnp->cn_nameiop == CREATE) {
593 nchstats.ncs_badhits++;
594 if (!wlocked && !CACHE_UPGRADE_LOCK())
601 if (!wlocked && !CACHE_UPGRADE_LOCK())
605 * We found a "negative" match, so we shift it to the end of
606 * the "negative" cache entries queue to satisfy LRU. Also,
607 * check to see if the entry is a whiteout; indicate this to
608 * the componentname, if so.
610 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
611 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
612 nchstats.ncs_neghits++;
613 if (ncp->nc_flag & NCF_WHITE)
614 cnp->cn_flags |= ISWHITEOUT;
615 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp,
617 cache_out_ts(ncp, tsp, ticksp);
623 * We need to update the cache after our lookup, so upgrade to
624 * a write lock and retry the operation.
634 * On success we return a locked and ref'd vnode as per the lookup
637 if (dvp == *vpp) { /* lookup on "." */
644 * When we lookup "." we still can be asked to lock it
647 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
648 if (ltype != VOP_ISLOCKED(*vpp)) {
649 if (ltype == LK_EXCLUSIVE) {
650 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
651 if ((*vpp)->v_iflag & VI_DOOMED) {
658 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
662 ltype = 0; /* silence gcc warning */
663 if (cnp->cn_flags & ISDOTDOT) {
664 ltype = VOP_ISLOCKED(dvp);
672 error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread);
673 if (cnp->cn_flags & ISDOTDOT) {
674 vn_lock(dvp, ltype | LK_RETRY);
675 if (dvp->v_iflag & VI_DOOMED) {
686 if ((cnp->cn_flags & ISLASTCN) &&
687 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
688 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
701 * Add an entry to the cache.
704 cache_enter_time(dvp, vp, cnp, tsp, dtsp)
707 struct componentname *cnp;
708 struct timespec *tsp;
709 struct timespec *dtsp;
711 struct namecache *ncp, *n2;
712 struct namecache_ts *n3;
713 struct nchashhead *ncpp;
720 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
721 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
722 ("cache_enter: Adding a doomed vnode"));
723 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
724 ("cache_enter: Doomed vnode used as src"));
730 * Avoid blowout in namecache entries.
732 if (numcache >= desiredvnodes * ncsizefactor)
736 if (cnp->cn_nameptr[0] == '.') {
737 if (cnp->cn_namelen == 1)
739 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
742 * If dotdot entry already exists, just retarget it
743 * to new parent vnode, otherwise continue with new
744 * namecache entry allocation.
746 if ((ncp = dvp->v_cache_dd) != NULL &&
747 ncp->nc_flag & NCF_ISDOTDOT) {
748 KASSERT(ncp->nc_dvp == dvp,
749 ("wrong isdotdot parent"));
750 if (ncp->nc_vp != NULL) {
751 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
754 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
758 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
761 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
768 dvp->v_cache_dd = NULL;
769 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp);
779 * Calculate the hash key and setup as much of the new
780 * namecache entry as possible before acquiring the lock.
782 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
787 n3 = (struct namecache_ts *)ncp;
789 n3->nc_ticks = ticks;
790 n3->nc_flag |= NCF_TS;
792 n3->nc_dotdottime = *dtsp;
793 n3->nc_flag |= NCF_DTS;
796 len = ncp->nc_nlen = cnp->cn_namelen;
797 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
798 strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1);
799 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
803 * See if this vnode or negative entry is already in the cache
804 * with this name. This can happen with concurrent lookups of
805 * the same path name.
807 ncpp = NCHHASH(hash);
808 LIST_FOREACH(n2, ncpp, nc_hash) {
809 if (n2->nc_dvp == dvp &&
810 n2->nc_nlen == cnp->cn_namelen &&
811 !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) {
813 KASSERT((n2->nc_flag & NCF_TS) != 0,
815 n3 = (struct namecache_ts *)n2;
817 ((struct namecache_ts *)ncp)->nc_time;
819 ((struct namecache_ts *)ncp)->nc_ticks;
822 ((struct namecache_ts *)ncp)->
824 n3->nc_flag |= NCF_DTS;
833 if (flag == NCF_ISDOTDOT) {
835 * See if we are trying to add .. entry, but some other lookup
836 * has populated v_cache_dd pointer already.
838 if (dvp->v_cache_dd != NULL) {
843 KASSERT(vp == NULL || vp->v_type == VDIR,
844 ("wrong vnode type %p", vp));
845 dvp->v_cache_dd = ncp;
851 if (cnp->cn_flags & ISWHITEOUT)
852 ncp->nc_flag |= NCF_WHITE;
853 } else if (vp->v_type == VDIR) {
854 if (flag != NCF_ISDOTDOT) {
856 * For this case, the cache entry maps both the
857 * directory name in it and the name ".." for the
858 * directory's parent.
860 if ((n2 = vp->v_cache_dd) != NULL &&
861 (n2->nc_flag & NCF_ISDOTDOT) != 0)
863 vp->v_cache_dd = ncp;
866 vp->v_cache_dd = NULL;
870 * Insert the new namecache entry into the appropriate chain
871 * within the cache entries table.
873 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
874 if (flag != NCF_ISDOTDOT) {
875 if (LIST_EMPTY(&dvp->v_cache_src)) {
879 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
883 * If the entry is "negative", we place it into the
884 * "negative" cache queue, otherwise, we place it into the
885 * destination vnode's cache entries queue.
888 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
889 SDT_PROBE3(vfs, namecache, enter, done, dvp, nc_get_name(ncp),
892 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
893 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp,
896 if (numneg * ncnegfactor > numcache) {
897 ncp = TAILQ_FIRST(&ncneg);
898 KASSERT(ncp->nc_vp == NULL, ("ncp %p vp %p on ncneg",
910 * Name cache initialization, from vfs_init() when we are booting
913 nchinit(void *dummy __unused)
918 cache_zone_small = uma_zcreate("S VFS Cache",
919 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
920 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
921 cache_zone_small_ts = uma_zcreate("STS VFS Cache",
922 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
923 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
924 cache_zone_large = uma_zcreate("L VFS Cache",
925 sizeof(struct namecache) + NAME_MAX + 1,
926 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
927 cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
928 sizeof(struct namecache_ts) + NAME_MAX + 1,
929 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
931 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
933 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
936 cache_changesize(int newmaxvnodes)
938 struct nchashhead *new_nchashtbl, *old_nchashtbl;
939 u_long new_nchash, old_nchash;
940 struct namecache *ncp;
944 new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash);
945 /* If same hash table size, nothing to do */
946 if (nchash == new_nchash) {
947 free(new_nchashtbl, M_VFSCACHE);
951 * Move everything from the old hash table to the new table.
952 * None of the namecache entries in the table can be removed
953 * because to do so, they have to be removed from the hash table.
956 old_nchashtbl = nchashtbl;
958 nchashtbl = new_nchashtbl;
960 for (i = 0; i <= old_nchash; i++) {
961 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) {
962 hash = fnv_32_buf(nc_get_name(ncp), ncp->nc_nlen,
964 hash = fnv_32_buf(&ncp->nc_dvp, sizeof(ncp->nc_dvp),
966 LIST_REMOVE(ncp, nc_hash);
967 LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash);
971 free(old_nchashtbl, M_VFSCACHE);
975 * Invalidate all entries to a particular vnode.
982 CTR1(KTR_VFS, "cache_purge(%p)", vp);
983 SDT_PROBE1(vfs, namecache, purge, done, vp);
985 while (!LIST_EMPTY(&vp->v_cache_src))
986 cache_zap(LIST_FIRST(&vp->v_cache_src));
987 while (!TAILQ_EMPTY(&vp->v_cache_dst))
988 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
989 if (vp->v_cache_dd != NULL) {
990 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
991 ("lost dotdot link"));
992 cache_zap(vp->v_cache_dd);
994 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
999 * Invalidate all negative entries for a particular directory vnode.
1002 cache_purge_negative(vp)
1005 struct namecache *cp, *ncp;
1007 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
1008 SDT_PROBE1(vfs, namecache, purge_negative, done, vp);
1010 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
1011 if (cp->nc_vp == NULL)
1018 * Flush all entries referencing a particular filesystem.
1024 struct nchashhead *ncpp;
1025 struct namecache *ncp, *nnp;
1027 /* Scan hash tables for applicable entries */
1028 SDT_PROBE1(vfs, namecache, purgevfs, done, mp);
1030 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
1031 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
1032 if (ncp->nc_dvp->v_mount == mp)
1040 * Perform canonical checks and cache lookup and pass on to filesystem
1041 * through the vop_cachedlookup only if needed.
1045 vfs_cache_lookup(ap)
1046 struct vop_lookup_args /* {
1047 struct vnode *a_dvp;
1048 struct vnode **a_vpp;
1049 struct componentname *a_cnp;
1054 struct vnode **vpp = ap->a_vpp;
1055 struct componentname *cnp = ap->a_cnp;
1056 struct ucred *cred = cnp->cn_cred;
1057 int flags = cnp->cn_flags;
1058 struct thread *td = cnp->cn_thread;
1063 if (dvp->v_type != VDIR)
1066 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1067 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1070 error = VOP_ACCESS(dvp, VEXEC, cred, td);
1074 error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
1076 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
1083 * XXX All of these sysctls would probably be more productive dead.
1085 static int disablecwd;
1086 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
1087 "Disable the getcwd syscall");
1089 /* Implementation of the getcwd syscall. */
1091 sys___getcwd(td, uap)
1093 struct __getcwd_args *uap;
1096 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen,
1101 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen,
1105 struct filedesc *fdp;
1106 struct vnode *cdir, *rdir;
1113 if (buflen > path_max)
1116 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
1117 fdp = td->td_proc->p_fd;
1118 FILEDESC_SLOCK(fdp);
1119 cdir = fdp->fd_cdir;
1121 rdir = fdp->fd_rdir;
1123 FILEDESC_SUNLOCK(fdp);
1124 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
1129 if (bufseg == UIO_SYSSPACE)
1130 bcopy(bp, buf, strlen(bp) + 1);
1132 error = copyout(bp, buf, strlen(bp) + 1);
1134 if (KTRPOINT(curthread, KTR_NAMEI))
1138 free(tmpbuf, M_TEMP);
1143 * Thus begins the fullpath magic.
1147 #define STATNODE(name, descr) \
1148 static u_int name; \
1149 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr)
1151 static int disablefullpath;
1152 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
1153 "Disable the vn_fullpath function");
1155 /* These count for kern___getcwd(), too. */
1156 STATNODE(numfullpathcalls, "Number of fullpath search calls");
1157 STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
1158 STATNODE(numfullpathfail2,
1159 "Number of fullpath search errors (VOP_VPTOCNP failures)");
1160 STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
1161 STATNODE(numfullpathfound, "Number of successful fullpath calls");
1164 * Retrieve the full filesystem path that correspond to a vnode from the name
1165 * cache (if available)
1168 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
1171 struct filedesc *fdp;
1175 if (disablefullpath)
1180 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1181 fdp = td->td_proc->p_fd;
1182 FILEDESC_SLOCK(fdp);
1183 rdir = fdp->fd_rdir;
1185 FILEDESC_SUNLOCK(fdp);
1186 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1197 * This function is similar to vn_fullpath, but it attempts to lookup the
1198 * pathname relative to the global root mount point. This is required for the
1199 * auditing sub-system, as audited pathnames must be absolute, relative to the
1200 * global root mount point.
1203 vn_fullpath_global(struct thread *td, struct vnode *vn,
1204 char **retbuf, char **freebuf)
1209 if (disablefullpath)
1213 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1214 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1223 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1228 error = vn_vptocnp_locked(vp, cred, buf, buflen);
1235 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1239 struct namecache *ncp;
1242 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1243 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1247 if (*buflen < ncp->nc_nlen) {
1252 SDT_PROBE3(vfs, namecache, fullpath, return, error,
1256 *buflen -= ncp->nc_nlen;
1257 memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen);
1258 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1259 nc_get_name(ncp), vp);
1268 SDT_PROBE1(vfs, namecache, fullpath, miss, vp);
1271 vn_lock(*vp, LK_SHARED | LK_RETRY);
1272 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1276 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
1282 if (dvp->v_iflag & VI_DOOMED) {
1283 /* forced unmount */
1287 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
1291 * *vp has its use count incremented still.
1298 * The magic behind kern___getcwd() and vn_fullpath().
1301 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1302 char *buf, char **retbuf, u_int buflen)
1304 int error, slash_prefixed;
1305 #ifdef KDTRACE_HOOKS
1306 struct vnode *startvp = vp;
1315 SDT_PROBE1(vfs, namecache, fullpath, entry, vp);
1319 if (vp->v_type != VDIR) {
1320 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1328 buf[--buflen] = '/';
1331 while (vp != rdir && vp != rootvnode) {
1332 if (vp->v_vflag & VV_ROOT) {
1333 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */
1337 SDT_PROBE3(vfs, namecache, fullpath, return,
1341 vp1 = vp->v_mount->mnt_vnodecovered;
1349 if (vp->v_type != VDIR) {
1354 SDT_PROBE3(vfs, namecache, fullpath, return,
1358 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1365 SDT_PROBE3(vfs, namecache, fullpath, return, error,
1369 buf[--buflen] = '/';
1374 if (!slash_prefixed) {
1379 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM,
1383 buf[--buflen] = '/';
1389 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen);
1390 *retbuf = buf + buflen;
1395 vn_dir_dd_ino(struct vnode *vp)
1397 struct namecache *ncp;
1400 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
1402 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
1403 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
1408 if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread))
1417 vn_commname(struct vnode *vp, char *buf, u_int buflen)
1419 struct namecache *ncp;
1423 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1424 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1430 l = min(ncp->nc_nlen, buflen - 1);
1431 memcpy(buf, nc_get_name(ncp), l);
1437 /* ABI compat shims for old kernel modules. */
1440 void cache_enter(struct vnode *dvp, struct vnode *vp,
1441 struct componentname *cnp);
1444 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
1447 cache_enter_time(dvp, vp, cnp, NULL, NULL);
1451 * This function updates path string to vnode's full global path
1452 * and checks the size of the new path string against the pathlen argument.
1454 * Requires a locked, referenced vnode.
1455 * Vnode is re-locked on success or ENODEV, otherwise unlocked.
1457 * If sysctl debug.disablefullpath is set, ENODEV is returned,
1458 * vnode is left locked and path remain untouched.
1460 * If vp is a directory, the call to vn_fullpath_global() always succeeds
1461 * because it falls back to the ".." lookup if the namecache lookup fails.
1464 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
1467 struct nameidata nd;
1472 ASSERT_VOP_ELOCKED(vp, __func__);
1474 /* Return ENODEV if sysctl debug.disablefullpath==1 */
1475 if (disablefullpath)
1478 /* Construct global filesystem path from vp. */
1480 error = vn_fullpath_global(td, vp, &rpath, &fbuf);
1487 if (strlen(rpath) >= pathlen) {
1489 error = ENAMETOOLONG;
1494 * Re-lookup the vnode by path to detect a possible rename.
1495 * As a side effect, the vnode is relocked.
1496 * If vnode was renamed, return ENOENT.
1498 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1499 UIO_SYSSPACE, path, td);
1505 NDFREE(&nd, NDF_ONLY_PNBUF);
1509 strcpy(path, rpath);