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 * This structure describes the elements in the cache of recent
54 * names looked up by namei.
58 LIST_ENTRY(namecache) nc_hash; /* hash chain */
59 LIST_ENTRY(namecache) nc_src; /* source vnode list */
60 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
61 struct vnode *nc_dvp; /* vnode of parent of name */
62 struct vnode *nc_vp; /* vnode the name refers to */
63 u_char nc_flag; /* flag bits */
64 u_char nc_nlen; /* length of name */
65 char nc_name[0]; /* segment name */
69 * Name caching works as follows:
71 * Names found by directory scans are retained in a cache
72 * for future reference. It is managed LRU, so frequently
73 * used names will hang around. Cache is indexed by hash value
74 * obtained from (vp, name) where vp refers to the directory
77 * If it is a "negative" entry, (i.e. for a name that is known NOT to
78 * exist) the vnode pointer will be NULL.
80 * Upon reaching the last segment of a path, if the reference
81 * is for DELETE, or NOCACHE is set (rewrite), and the
82 * name is located in the cache, it will be dropped.
86 * Structures associated with name cacheing.
88 #define NCHHASH(dvp, hash) \
89 (&nchashtbl[((dvp)->v_id + (hash)) & nchash])
90 static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
91 static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
92 static u_long nchash; /* size of hash table */
93 SYSCTL_INT(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
94 static u_long ncnegfactor = 16; /* ratio of negative entries */
95 SYSCTL_INT(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
96 static u_long numneg; /* number of cache entries allocated */
97 SYSCTL_INT(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
98 static u_long numcache; /* number of cache entries allocated */
99 SYSCTL_INT(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
100 struct nchstats nchstats; /* cache effectiveness statistics */
102 static int doingcache = 1; /* 1 => enable the cache */
103 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
104 SYSCTL_INT(_debug, OID_AUTO, vnsize, CTLFLAG_RD, 0, sizeof(struct vnode), "");
105 SYSCTL_INT(_debug, OID_AUTO, ncsize, CTLFLAG_RD, 0, sizeof(struct namecache), "");
108 * The new name cache statistics
110 SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
111 #define STATNODE(mode, name, var) \
112 SYSCTL_INT(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
113 STATNODE(CTLFLAG_RD, numneg, &numneg);
114 STATNODE(CTLFLAG_RD, numcache, &numcache);
115 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
116 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
117 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
118 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
119 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
120 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
121 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
122 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
123 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
124 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
127 static void cache_zap __P((struct namecache *ncp));
129 MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
132 * Flags in namecache.nc_flag
136 * Delete an entry from its hash list and move it to the front
137 * of the LRU list for immediate reuse.
141 struct namecache *ncp;
143 LIST_REMOVE(ncp, nc_hash);
144 LIST_REMOVE(ncp, nc_src);
145 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src))
148 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
150 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
154 free(ncp, M_VFSCACHE);
158 * Lookup an entry in the cache
160 * We don't do this if the segment name is long, simply so the cache
161 * can avoid holding long names (which would either waste space, or
162 * add greatly to the complexity).
164 * Lookup is called with dvp pointing to the directory to search,
165 * cnp pointing to the name of the entry being sought. If the lookup
166 * succeeds, the vnode is returned in *vpp, and a status of -1 is
167 * returned. If the lookup determines that the name does not exist
168 * (negative cacheing), a status of ENOENT is returned. If the lookup
169 * fails, a status of zero is returned.
173 cache_lookup(dvp, vpp, cnp)
176 struct componentname *cnp;
178 struct namecache *ncp;
184 cnp->cn_flags &= ~MAKEENTRY;
190 if (cnp->cn_nameptr[0] == '.') {
191 if (cnp->cn_namelen == 1) {
196 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
198 if (dvp->v_dd->v_id != dvp->v_ddid ||
199 (cnp->cn_flags & MAKEENTRY) == 0) {
209 len = cnp->cn_namelen;
210 for (cp = cnp->cn_nameptr; len; len--, cp++)
212 LIST_FOREACH(ncp, (NCHHASH(dvp, hash)), nc_hash) {
214 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
215 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
219 /* We failed to find an entry */
221 if ((cnp->cn_flags & MAKEENTRY) == 0) {
230 /* We don't want to have an entry, so dump it */
231 if ((cnp->cn_flags & MAKEENTRY) == 0) {
233 nchstats.ncs_badhits++;
238 /* We found a "positive" match, return the vnode */
241 nchstats.ncs_goodhits++;
246 /* We found a negative match, and want to create it, so purge */
247 if (cnp->cn_nameiop == CREATE) {
249 nchstats.ncs_badhits++;
256 * We found a "negative" match, ENOENT notifies client of this match.
257 * The nc_vpid field records whether this is a whiteout.
259 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
260 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
261 nchstats.ncs_neghits++;
262 if (ncp->nc_flag & NCF_WHITE)
263 cnp->cn_flags |= ISWHITEOUT;
268 * Add an entry to the cache.
271 cache_enter(dvp, vp, cnp)
274 struct componentname *cnp;
276 struct namecache *ncp;
277 struct nchashhead *ncpp;
285 if (cnp->cn_nameptr[0] == '.') {
286 if (cnp->cn_namelen == 1) {
289 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
292 dvp->v_ddid = vp->v_id;
301 ncp = (struct namecache *)
302 malloc(sizeof *ncp + cnp->cn_namelen, M_VFSCACHE, M_WAITOK);
303 bzero((char *)ncp, sizeof *ncp);
307 ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
308 } else if (vp->v_type == VDIR) {
310 vp->v_ddid = dvp->v_id;
314 * Fill in cache info, if vp is NULL this is a "negative" cache entry.
315 * For negative entries, we have to record whether it is a whiteout.
316 * the whiteout flag is stored in the nc_vpid field which is
321 len = ncp->nc_nlen = cnp->cn_namelen;
324 for (cp = cnp->cn_nameptr; len; len--, cp++, dp++)
326 ncpp = NCHHASH(dvp, hash);
327 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
328 if (LIST_EMPTY(&dvp->v_cache_src))
330 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
332 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
334 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
336 if (numneg * ncnegfactor > numcache) {
337 ncp = TAILQ_FIRST(&ncneg);
343 * Name cache initialization, from vfs_init() when we are booting
350 nchashtbl = hashinit(desiredvnodes*2, M_VFSCACHE, &nchash);
354 * Invalidate all entries to a particular vnode.
356 * Remove all entries in the namecache relating to this vnode and
357 * change the v_id. We take the v_id from a global counter, since
358 * it becomes a handy sequence number in crash-dumps that way.
359 * No valid vnode will ever have (v_id == 0).
361 * XXX: Only time and the size of v_id prevents this from failing:
362 * XXX: In theory we should hunt down all (struct vnode*, v_id)
363 * XXX: soft references and nuke them, at least on the global
364 * XXX: v_id wraparound. The period of resistance can be extended
365 * XXX: by incrementing each vnodes v_id individually instead of
366 * XXX: using the global v_id.
373 static u_long nextid;
375 while (!LIST_EMPTY(&vp->v_cache_src))
376 cache_zap(LIST_FIRST(&vp->v_cache_src));
377 while (!TAILQ_EMPTY(&vp->v_cache_dst))
378 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
382 while (nextid == vp->v_id || !nextid);
389 * Flush all entries referencing a particular filesystem.
391 * Since we need to check it anyway, we will flush all the invalid
392 * entries at the same time.
398 struct nchashhead *ncpp;
399 struct namecache *ncp, *nnp;
401 /* Scan hash tables for applicable entries */
402 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
403 for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
404 nnp = LIST_NEXT(ncp, nc_hash);
405 if (ncp->nc_dvp->v_mount == mp) {
413 * Perform canonical checks and cache lookup and pass on to filesystem
414 * through the vop_cachedlookup only if needed.
419 struct vop_lookup_args /* {
421 struct vnode **a_vpp;
422 struct componentname *a_cnp;
429 struct vnode **vpp = ap->a_vpp;
430 struct componentname *cnp = ap->a_cnp;
431 struct ucred *cred = cnp->cn_cred;
432 int flags = cnp->cn_flags;
433 struct proc *p = cnp->cn_proc;
434 u_long vpid; /* capability number of vnode */
438 lockparent = flags & LOCKPARENT;
440 if (vdp->v_type != VDIR)
443 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
444 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
447 error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc);
452 error = cache_lookup(vdp, vpp, cnp);
455 return (VOP_CACHEDLOOKUP(ap->a_dvp, ap->a_vpp, ap->a_cnp));
463 if (pdp == vdp) { /* lookup on "." */
466 } else if (flags & ISDOTDOT) {
467 VOP_UNLOCK(pdp, 0, p);
468 error = vget(vdp, LK_EXCLUSIVE, p);
469 if (!error && lockparent && (flags & ISLASTCN))
470 error = vn_lock(pdp, LK_EXCLUSIVE, p);
472 error = vget(vdp, LK_EXCLUSIVE, p);
473 if (!lockparent || error || !(flags & ISLASTCN))
474 VOP_UNLOCK(pdp, 0, p);
477 * Check that the capability number did not change
478 * while we were waiting for the lock.
481 if (vpid == vdp->v_id)
484 if (lockparent && pdp != vdp && (flags & ISLASTCN))
485 VOP_UNLOCK(pdp, 0, p);
487 error = vn_lock(pdp, LK_EXCLUSIVE, p);
490 return (VOP_CACHEDLOOKUP(ap->a_dvp, ap->a_vpp, ap->a_cnp));
494 #ifndef _SYS_SYSPROTO_H_
495 struct __getcwd_args {
501 #define STATNODE(mode, name, var) \
502 SYSCTL_INT(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
504 static int disablecwd;
505 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, "");
507 static u_long numcwdcalls; STATNODE(CTLFLAG_RD, numcwdcalls, &numcwdcalls);
508 static u_long numcwdfail1; STATNODE(CTLFLAG_RD, numcwdfail1, &numcwdfail1);
509 static u_long numcwdfail2; STATNODE(CTLFLAG_RD, numcwdfail2, &numcwdfail2);
510 static u_long numcwdfail3; STATNODE(CTLFLAG_RD, numcwdfail3, &numcwdfail3);
511 static u_long numcwdfail4; STATNODE(CTLFLAG_RD, numcwdfail4, &numcwdfail4);
512 static u_long numcwdfound; STATNODE(CTLFLAG_RD, numcwdfound, &numcwdfound);
516 struct __getcwd_args *uap;
519 int error, i, slash_prefixed;
520 struct filedesc *fdp;
521 struct namecache *ncp;
529 if (uap->buflen > MAXPATHLEN)
530 uap->buflen = MAXPATHLEN;
531 buf = bp = malloc(uap->buflen, M_TEMP, M_WAITOK);
532 bp += uap->buflen - 1;
536 for (vp = fdp->fd_cdir; vp != fdp->fd_rdir && vp != rootvnode;) {
537 if (vp->v_flag & VROOT) {
538 if (vp->v_mount == NULL) /* forced unmount */
540 vp = vp->v_mount->mnt_vnodecovered;
543 if (vp->v_dd->v_id != vp->v_ddid) {
548 ncp = TAILQ_FIRST(&vp->v_cache_dst);
554 if (ncp->nc_dvp != vp->v_dd) {
559 for (i = ncp->nc_nlen - 1; i >= 0; i--) {
565 *--bp = ncp->nc_name[i];
576 if (!slash_prefixed) {
585 error = copyout(bp, uap->buf, strlen(bp) + 1);