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 <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
42 #include <sys/mutex.h>
43 #include <sys/sysctl.h>
44 #include <sys/mount.h>
45 #include <sys/vnode.h>
46 #include <sys/namei.h>
47 #include <sys/malloc.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysproto.h>
51 #include <sys/filedesc.h>
52 #include <sys/fnv_hash.h>
57 * This structure describes the elements in the cache of recent
58 * names looked up by namei.
62 LIST_ENTRY(namecache) nc_hash; /* hash chain */
63 LIST_ENTRY(namecache) nc_src; /* source vnode list */
64 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
65 struct vnode *nc_dvp; /* vnode of parent of name */
66 struct vnode *nc_vp; /* vnode the name refers to */
67 u_char nc_flag; /* flag bits */
68 u_char nc_nlen; /* length of name */
69 char nc_name[0]; /* segment name */
73 * Name caching works as follows:
75 * Names found by directory scans are retained in a cache
76 * for future reference. It is managed LRU, so frequently
77 * used names will hang around. Cache is indexed by hash value
78 * obtained from (vp, name) where vp refers to the directory
81 * If it is a "negative" entry, (i.e. for a name that is known NOT to
82 * exist) the vnode pointer will be NULL.
84 * Upon reaching the last segment of a path, if the reference
85 * is for DELETE, or NOCACHE is set (rewrite), and the
86 * name is located in the cache, it will be dropped.
90 * Structures associated with name cacheing.
92 #define NCHHASH(hash) \
93 (&nchashtbl[(hash) & nchash])
94 static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
95 static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
96 static u_long nchash; /* size of hash table */
97 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
98 static u_long ncnegfactor = 16; /* ratio of negative entries */
99 SYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
100 static u_long numneg; /* number of cache entries allocated */
101 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
102 static u_long numcache; /* number of cache entries allocated */
103 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
104 static u_long numcachehv; /* number of cache entries with vnodes held */
105 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, "");
107 static u_long numcachepl; /* number of cache purge for leaf entries */
108 SYSCTL_ULONG(_debug, OID_AUTO, numcachepl, CTLFLAG_RD, &numcachepl, 0, "");
110 struct nchstats nchstats; /* cache effectiveness statistics */
112 static struct mtx cache_lock;
113 MTX_SYSINIT(vfscache, &cache_lock, "Name Cache", MTX_DEF);
115 #define CACHE_LOCK() mtx_lock(&cache_lock)
116 #define CACHE_UNLOCK() mtx_unlock(&cache_lock)
119 * UMA zones for the VFS cache.
121 * The small cache is used for entries with short names, which are the
122 * most common. The large cache is used for entries which are too big to
123 * fit in the small cache.
125 static uma_zone_t cache_zone_small;
126 static uma_zone_t cache_zone_large;
128 #define CACHE_PATH_CUTOFF 32
129 #define CACHE_ZONE_SMALL (sizeof(struct namecache) + CACHE_PATH_CUTOFF)
130 #define CACHE_ZONE_LARGE (sizeof(struct namecache) + NAME_MAX)
132 #define cache_alloc(len) uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \
133 cache_zone_small : cache_zone_large, M_WAITOK)
134 #define cache_free(ncp) do { \
136 uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \
137 cache_zone_small : cache_zone_large, (ncp)); \
140 static int doingcache = 1; /* 1 => enable the cache */
141 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
143 /* Export size information to userland */
144 SYSCTL_INT(_debug, OID_AUTO, vnsize, CTLFLAG_RD, 0, sizeof(struct vnode), "");
145 SYSCTL_INT(_debug, OID_AUTO, ncsize, CTLFLAG_RD, 0, sizeof(struct namecache), "");
148 * The new name cache statistics
150 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
151 #define STATNODE(mode, name, var) \
152 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
153 STATNODE(CTLFLAG_RD, numneg, &numneg);
154 STATNODE(CTLFLAG_RD, numcache, &numcache);
155 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
156 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
157 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
158 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
159 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
160 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
161 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
162 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
163 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
164 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
166 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD, &nchstats,
167 sizeof(nchstats), "LU", "VFS cache effectiveness statistics");
171 static void cache_zap(struct namecache *ncp);
172 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
173 char *buf, char **retbuf, u_int buflen);
175 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
178 * Flags in namecache.nc_flag
183 * Grab an atomic snapshot of the name cache hash chain lengths
185 SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
188 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
191 struct nchashhead *ncpp;
192 struct namecache *ncp;
196 n_nchash = nchash + 1; /* nchash is max index, not count */
198 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
200 /* Scan hash tables for applicable entries */
201 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
203 LIST_FOREACH(ncp, ncpp, nc_hash) {
206 error = SYSCTL_OUT(req, &count, sizeof(count));
212 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD,
213 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths");
216 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
219 struct nchashhead *ncpp;
220 struct namecache *ncp;
222 int count, maxlength, used, pct;
225 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
227 n_nchash = nchash + 1; /* nchash is max index, not count */
231 /* Scan hash tables for applicable entries */
232 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
234 LIST_FOREACH(ncp, ncpp, nc_hash) {
239 if (maxlength < count)
242 n_nchash = nchash + 1;
243 pct = (used * 100 * 100) / n_nchash;
244 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
247 error = SYSCTL_OUT(req, &used, sizeof(used));
250 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
253 error = SYSCTL_OUT(req, &pct, sizeof(pct));
258 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD,
259 0, 0, sysctl_debug_hashstat_nchash, "I", "nchash chain lengths");
264 * Removes a namecache entry from cache, whether it contains an actual
265 * pointer to a vnode or if it is just a negative cache entry.
269 struct namecache *ncp;
273 mtx_assert(&cache_lock, MA_OWNED);
275 LIST_REMOVE(ncp, nc_hash);
276 LIST_REMOVE(ncp, nc_src);
277 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
282 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
284 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
296 * Test whether this (directory) vnode's namei cache entry contains
297 * subdirectories or not. Used to determine whether the directory is
298 * a leaf in the namei cache or not. Note: the directory may still
299 * contain files in the namei cache.
301 * Returns 0 if the directory is a leaf, -1 if it isn't.
304 cache_leaf_test(struct vnode *vp)
306 struct namecache *ncpc;
311 for (ncpc = LIST_FIRST(&vp->v_cache_src);
313 ncpc = LIST_NEXT(ncpc, nc_src)
315 if (ncpc->nc_vp != NULL && ncpc->nc_vp->v_type == VDIR) {
325 * Lookup an entry in the cache
327 * Lookup is called with dvp pointing to the directory to search,
328 * cnp pointing to the name of the entry being sought. If the lookup
329 * succeeds, the vnode is returned in *vpp, and a status of -1 is
330 * returned. If the lookup determines that the name does not exist
331 * (negative cacheing), a status of ENOENT is returned. If the lookup
332 * fails, a status of zero is returned.
334 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
335 * unlocked. If we're looking up . an extra ref is taken, but the lock is
336 * not recursively acquired.
340 cache_lookup(dvp, vpp, cnp)
343 struct componentname *cnp;
345 struct namecache *ncp;
350 cnp->cn_flags &= ~MAKEENTRY;
357 if (cnp->cn_nameptr[0] == '.') {
358 if (cnp->cn_namelen == 1) {
363 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
365 if (dvp->v_dd == NULL ||
366 (cnp->cn_flags & MAKEENTRY) == 0) {
375 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
376 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
377 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
379 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
380 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
384 /* We failed to find an entry */
386 if ((cnp->cn_flags & MAKEENTRY) == 0) {
396 /* We don't want to have an entry, so dump it */
397 if ((cnp->cn_flags & MAKEENTRY) == 0) {
399 nchstats.ncs_badhits++;
405 /* We found a "positive" match, return the vnode */
408 nchstats.ncs_goodhits++;
413 /* We found a negative match, and want to create it, so purge */
414 if (cnp->cn_nameiop == CREATE) {
416 nchstats.ncs_badhits++;
424 * We found a "negative" match, so we shift it to the end of
425 * the "negative" cache entries queue to satisfy LRU. Also,
426 * check to see if the entry is a whiteout; indicate this to
427 * the componentname, if so.
429 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
430 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
431 nchstats.ncs_neghits++;
432 if (ncp->nc_flag & NCF_WHITE)
433 cnp->cn_flags |= ISWHITEOUT;
439 * On success we return a locked and ref'd vnode as per the lookup
442 if (dvp == *vpp) { /* lookup on "." */
447 if (cnp->cn_flags & ISDOTDOT)
448 VOP_UNLOCK(dvp, 0, cnp->cn_thread);
451 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
452 if (cnp->cn_flags & ISDOTDOT)
453 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, cnp->cn_thread);
462 * Add an entry to the cache.
465 cache_enter(dvp, vp, cnp)
468 struct componentname *cnp;
470 struct namecache *ncp;
471 struct nchashhead *ncpp;
480 if (cnp->cn_nameptr[0] == '.') {
481 if (cnp->cn_namelen == 1) {
484 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
492 ncp = cache_alloc(cnp->cn_namelen);
497 ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
498 } else if (vp->v_type == VDIR) {
505 * Set the rest of the namecache entry elements, calculate it's
506 * hash key and insert it into the appropriate chain within
507 * the cache entries table.
511 len = ncp->nc_nlen = cnp->cn_namelen;
512 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
513 bcopy(cnp->cn_nameptr, ncp->nc_name, len);
514 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
515 ncpp = NCHHASH(hash);
516 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
517 if (LIST_EMPTY(&dvp->v_cache_src)) {
521 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
523 * If the entry is "negative", we place it into the
524 * "negative" cache queue, otherwise, we place it into the
525 * destination vnode's cache entries queue.
528 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
530 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
532 if (numneg * ncnegfactor > numcache) {
533 ncp = TAILQ_FIRST(&ncneg);
544 * Name cache initialization, from vfs_init() when we are booting
547 nchinit(void *dummy __unused)
552 cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
553 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
554 cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
555 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
557 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
559 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL)
563 * Invalidate all entries to a particular vnode.
569 struct namecache *ncp;
572 while (!LIST_EMPTY(&vp->v_cache_src)) {
575 ncp = LIST_FIRST(&vp->v_cache_src);
577 * We must reset v_dd of any children so they don't
578 * continue to point to us.
580 if ((cvp = ncp->nc_vp) && cvp->v_dd == vp)
584 while (!TAILQ_EMPTY(&vp->v_cache_dst))
585 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
591 * Flush all entries referencing a particular filesystem.
593 * Since we need to check it anyway, we will flush all the invalid
594 * entries at the same time.
600 struct nchashhead *ncpp;
601 struct namecache *ncp, *nnp;
602 struct nchashhead mplist;
607 /* Scan hash tables for applicable entries */
609 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
610 for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
611 nnp = LIST_NEXT(ncp, nc_hash);
612 if (ncp->nc_dvp->v_mount == mp) {
613 LIST_REMOVE(ncp, nc_hash);
614 LIST_INSERT_HEAD(&mplist, ncp, nc_hash);
618 while (!LIST_EMPTY(&mplist))
619 cache_zap(LIST_FIRST(&mplist));
624 * Perform canonical checks and cache lookup and pass on to filesystem
625 * through the vop_cachedlookup only if needed.
630 struct vop_lookup_args /* {
632 struct vnode **a_vpp;
633 struct componentname *a_cnp;
638 struct vnode **vpp = ap->a_vpp;
639 struct componentname *cnp = ap->a_cnp;
640 struct ucred *cred = cnp->cn_cred;
641 int flags = cnp->cn_flags;
642 struct thread *td = cnp->cn_thread;
647 if (dvp->v_type != VDIR)
650 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
651 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
654 error = VOP_ACCESS(dvp, VEXEC, cred, td);
658 error = cache_lookup(dvp, vpp, cnp);
660 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
667 #ifndef _SYS_SYSPROTO_H_
668 struct __getcwd_args {
675 * XXX All of these sysctls would probably be more productive dead.
677 static int disablecwd;
678 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
679 "Disable the getcwd syscall");
681 /* Implementation of the getcwd syscall */
685 struct __getcwd_args *uap;
688 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
692 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
695 struct filedesc *fdp;
702 if (buflen > MAXPATHLEN)
705 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
706 fdp = td->td_proc->p_fd;
709 error = vn_fullpath1(td, fdp->fd_cdir, fdp->fd_rdir, tmpbuf,
711 FILEDESC_UNLOCK(fdp);
715 if (bufseg == UIO_SYSSPACE)
716 bcopy(bp, buf, strlen(bp) + 1);
718 error = copyout(bp, buf, strlen(bp) + 1);
720 free(tmpbuf, M_TEMP);
725 * Thus begins the fullpath magic.
729 #define STATNODE(name) \
731 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
733 static int disablefullpath;
734 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
735 "Disable the vn_fullpath function");
737 /* These count for kern___getcwd(), too. */
738 STATNODE(numfullpathcalls);
739 STATNODE(numfullpathfail1);
740 STATNODE(numfullpathfail2);
741 STATNODE(numfullpathfail4);
742 STATNODE(numfullpathfound);
745 * Retrieve the full filesystem path that correspond to a vnode from the name
746 * cache (if available)
749 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
752 struct filedesc *fdp;
760 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
761 fdp = td->td_proc->p_fd;
764 error = vn_fullpath1(td, vn, fdp->fd_rdir, buf, retbuf, MAXPATHLEN);
765 FILEDESC_UNLOCK(fdp);
776 * The magic behind kern___getcwd() and vn_fullpath().
779 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
780 char *buf, char **retbuf, u_int buflen)
783 int error, i, slash_prefixed;
784 struct namecache *ncp;
786 mtx_assert(&Giant, MA_OWNED);
788 bp = buf + buflen - 1;
795 if (vp->v_type != VDIR) {
796 ncp = TAILQ_FIRST(&vp->v_cache_dst);
802 for (i = ncp->nc_nlen - 1; i >= 0 && bp > buf; i--)
803 *--bp = ncp->nc_name[i];
813 while (vp != rdir && vp != rootvnode) {
814 if (vp->v_vflag & VV_ROOT) {
815 if (vp->v_mount == NULL) { /* forced unmount */
819 vp = vp->v_mount->mnt_vnodecovered;
822 if (vp->v_dd == NULL) {
827 ncp = TAILQ_FIRST(&vp->v_cache_dst);
833 MPASS(ncp->nc_dvp == vp->v_dd);
834 for (i = ncp->nc_nlen - 1; i >= 0 && bp != buf; i--)
835 *--bp = ncp->nc_name[i];
849 if (!slash_prefixed) {