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. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.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/sysproto.h>
50 #include <sys/filedesc.h>
53 * Name caching works as follows:
55 * Names found by directory scans are retained in a cache
56 * for future reference. It is managed LRU, so frequently
57 * used names will hang around. Cache is indexed by hash value
58 * obtained from (vp, name) where vp refers to the directory
61 * If it is a "negative" entry, (i.e. for a name that is known NOT to
62 * exist) the vnode pointer will be NULL.
64 * Upon reaching the last segment of a path, if the reference
65 * is for DELETE, or NOCACHE is set (rewrite), and the
66 * name is located in the cache, it will be dropped.
70 * Structures associated with name cacheing.
72 #define NCHHASH(dvp, hash) \
73 (&nchashtbl[((dvp)->v_id + (hash)) & nchash])
74 static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
75 static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
76 static u_long nchash; /* size of hash table */
77 SYSCTL_INT(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
78 static u_long ncnegfactor = 16; /* ratio of negative entries */
79 SYSCTL_INT(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
80 static u_long numneg; /* number of cache entries allocated */
81 SYSCTL_INT(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
82 static u_long numcache; /* number of cache entries allocated */
83 SYSCTL_INT(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
84 struct nchstats nchstats; /* cache effectiveness statistics */
86 static int doingcache = 1; /* 1 => enable the cache */
87 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
88 SYSCTL_INT(_debug, OID_AUTO, vnsize, CTLFLAG_RD, 0, sizeof(struct vnode), "");
89 SYSCTL_INT(_debug, OID_AUTO, ncsize, CTLFLAG_RD, 0, sizeof(struct namecache), "");
92 * The new name cache statistics
94 SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
95 #define STATNODE(mode, name, var) \
96 SYSCTL_INT(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
97 STATNODE(CTLFLAG_RD, numneg, &numneg);
98 STATNODE(CTLFLAG_RD, numcache, &numcache);
99 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
100 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
101 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
102 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
103 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
104 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
105 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
106 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
107 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
108 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
111 static void cache_zap __P((struct namecache *ncp));
113 MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
116 * Flags in namecache.nc_flag
120 * Delete an entry from its hash list and move it to the front
121 * of the LRU list for immediate reuse.
125 struct namecache *ncp;
127 LIST_REMOVE(ncp, nc_hash);
128 LIST_REMOVE(ncp, nc_src);
129 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src))
132 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
134 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
138 free(ncp, M_VFSCACHE);
142 * Lookup an entry in the cache
144 * We don't do this if the segment name is long, simply so the cache
145 * can avoid holding long names (which would either waste space, or
146 * add greatly to the complexity).
148 * Lookup is called with dvp pointing to the directory to search,
149 * cnp pointing to the name of the entry being sought. If the lookup
150 * succeeds, the vnode is returned in *vpp, and a status of -1 is
151 * returned. If the lookup determines that the name does not exist
152 * (negative cacheing), a status of ENOENT is returned. If the lookup
153 * fails, a status of zero is returned.
157 cache_lookup(dvp, vpp, cnp)
160 struct componentname *cnp;
162 struct namecache *ncp;
168 cnp->cn_flags &= ~MAKEENTRY;
174 if (cnp->cn_nameptr[0] == '.') {
175 if (cnp->cn_namelen == 1) {
180 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
182 if (dvp->v_dd->v_id != dvp->v_ddid ||
183 (cnp->cn_flags & MAKEENTRY) == 0) {
193 len = cnp->cn_namelen;
194 for (cp = cnp->cn_nameptr; len; len--, cp++)
196 LIST_FOREACH(ncp, (NCHHASH(dvp, hash)), nc_hash) {
198 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
199 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
203 /* We failed to find an entry */
205 if ((cnp->cn_flags & MAKEENTRY) == 0) {
214 /* We don't want to have an entry, so dump it */
215 if ((cnp->cn_flags & MAKEENTRY) == 0) {
217 nchstats.ncs_badhits++;
222 /* We found a "positive" match, return the vnode */
225 nchstats.ncs_goodhits++;
230 /* We found a negative match, and want to create it, so purge */
231 if (cnp->cn_nameiop == CREATE) {
233 nchstats.ncs_badhits++;
240 * We found a "negative" match, ENOENT notifies client of this match.
241 * The nc_vpid field records whether this is a whiteout.
243 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
244 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
245 nchstats.ncs_neghits++;
246 if (ncp->nc_flag & NCF_WHITE)
247 cnp->cn_flags |= ISWHITEOUT;
252 * Add an entry to the cache.
255 cache_enter(dvp, vp, cnp)
258 struct componentname *cnp;
260 struct namecache *ncp;
261 struct nchashhead *ncpp;
269 if (cnp->cn_nameptr[0] == '.') {
270 if (cnp->cn_namelen == 1) {
273 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
276 dvp->v_ddid = vp->v_id;
285 ncp = (struct namecache *)
286 malloc(sizeof *ncp + cnp->cn_namelen, M_VFSCACHE, M_WAITOK);
287 bzero((char *)ncp, sizeof *ncp);
291 ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
292 } else if (vp->v_type == VDIR) {
294 vp->v_ddid = dvp->v_id;
298 * Fill in cache info, if vp is NULL this is a "negative" cache entry.
299 * For negative entries, we have to record whether it is a whiteout.
300 * the whiteout flag is stored in the nc_vpid field which is
305 len = ncp->nc_nlen = cnp->cn_namelen;
308 for (cp = cnp->cn_nameptr; len; len--, cp++, dp++)
310 ncpp = NCHHASH(dvp, hash);
311 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
312 if (LIST_EMPTY(&dvp->v_cache_src))
314 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
316 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
318 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
320 if (numneg * ncnegfactor > numcache) {
321 ncp = TAILQ_FIRST(&ncneg);
327 * Name cache initialization, from vfs_init() when we are booting
334 nchashtbl = hashinit(desiredvnodes*2, M_VFSCACHE, &nchash);
338 * Invalidate all entries to a particular vnode.
340 * Remove all entries in the namecache relating to this vnode and
341 * change the v_id. We take the v_id from a global counter, since
342 * it becomes a handy sequence number in crash-dumps that way.
343 * No valid vnode will ever have (v_id == 0).
345 * XXX: Only time and the size of v_id prevents this from failing:
346 * XXX: In theory we should hunt down all (struct vnode*, v_id)
347 * XXX: soft references and nuke them, at least on the global
348 * XXX: v_id wraparound. The period of resistance can be extended
349 * XXX: by incrementing each vnodes v_id individually instead of
350 * XXX: using the global v_id.
357 static u_long nextid;
359 while (!LIST_EMPTY(&vp->v_cache_src))
360 cache_zap(LIST_FIRST(&vp->v_cache_src));
361 while (!TAILQ_EMPTY(&vp->v_cache_dst))
362 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
366 while (nextid == vp->v_id || !nextid);
373 * Flush all entries referencing a particular filesystem.
375 * Since we need to check it anyway, we will flush all the invalid
376 * entries at the same time.
382 struct nchashhead *ncpp;
383 struct namecache *ncp, *nnp;
385 /* Scan hash tables for applicable entries */
386 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
387 for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
388 nnp = LIST_NEXT(ncp, nc_hash);
389 if (ncp->nc_dvp->v_mount == mp) {
397 * Perform canonical checks and cache lookup and pass on to filesystem
398 * through the vop_cachedlookup only if needed.
403 struct vop_lookup_args /* {
405 struct vnode **a_vpp;
406 struct componentname *a_cnp;
413 struct vnode **vpp = ap->a_vpp;
414 struct componentname *cnp = ap->a_cnp;
415 struct ucred *cred = cnp->cn_cred;
416 int flags = cnp->cn_flags;
417 struct proc *p = cnp->cn_proc;
418 u_long vpid; /* capability number of vnode */
422 lockparent = flags & LOCKPARENT;
424 if (vdp->v_type != VDIR)
427 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
428 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
431 error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc);
436 error = cache_lookup(vdp, vpp, cnp);
439 return (VOP_CACHEDLOOKUP(ap->a_dvp, ap->a_vpp, ap->a_cnp));
447 if (pdp == vdp) { /* lookup on "." */
450 } else if (flags & ISDOTDOT) {
451 VOP_UNLOCK(pdp, 0, p);
452 error = vget(vdp, LK_EXCLUSIVE, p);
453 if (!error && lockparent && (flags & ISLASTCN))
454 error = vn_lock(pdp, LK_EXCLUSIVE, p);
456 error = vget(vdp, LK_EXCLUSIVE, p);
457 if (!lockparent || error || !(flags & ISLASTCN))
458 VOP_UNLOCK(pdp, 0, p);
461 * Check that the capability number did not change
462 * while we were waiting for the lock.
465 if (vpid == vdp->v_id)
468 if (lockparent && pdp != vdp && (flags & ISLASTCN))
469 VOP_UNLOCK(pdp, 0, p);
471 error = vn_lock(pdp, LK_EXCLUSIVE, p);
474 return (VOP_CACHEDLOOKUP(ap->a_dvp, ap->a_vpp, ap->a_cnp));
478 #ifndef _SYS_SYSPROTO_H_
479 struct __getcwd_args {
485 #define STATNODE(mode, name, var) \
486 SYSCTL_INT(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
488 static int disablecwd;
489 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, "");
491 static u_long numcwdcalls; STATNODE(CTLFLAG_RD, numcwdcalls, &numcwdcalls);
492 static u_long numcwdfail1; STATNODE(CTLFLAG_RD, numcwdfail1, &numcwdfail1);
493 static u_long numcwdfail2; STATNODE(CTLFLAG_RD, numcwdfail2, &numcwdfail2);
494 static u_long numcwdfail3; STATNODE(CTLFLAG_RD, numcwdfail3, &numcwdfail3);
495 static u_long numcwdfail4; STATNODE(CTLFLAG_RD, numcwdfail4, &numcwdfail4);
496 static u_long numcwdfound; STATNODE(CTLFLAG_RD, numcwdfound, &numcwdfound);
500 struct __getcwd_args *uap;
503 int error, i, slash_prefixed;
504 struct filedesc *fdp;
505 struct namecache *ncp;
513 if (uap->buflen > MAXPATHLEN)
514 uap->buflen = MAXPATHLEN;
515 buf = bp = malloc(uap->buflen, M_TEMP, M_WAITOK);
516 bp += uap->buflen - 1;
520 for (vp = fdp->fd_cdir; vp != fdp->fd_rdir && vp != rootvnode;) {
521 if (vp->v_flag & VROOT) {
522 if (vp->v_mount == NULL) /* forced unmount */
524 vp = vp->v_mount->mnt_vnodecovered;
527 if (vp->v_dd->v_id != vp->v_ddid) {
532 ncp = TAILQ_FIRST(&vp->v_cache_dst);
538 if (ncp->nc_dvp != vp->v_dd) {
543 for (i = ncp->nc_nlen - 1; i >= 0; i--) {
549 *--bp = ncp->nc_name[i];
560 if (!slash_prefixed) {
569 error = copyout(bp, uap->buf, strlen(bp) + 1);