FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_cache.c
1 /* $OpenBSD: vfs_cache.c,v 1.58 2022/08/14 01:58:28 jsg Exp $ */
2 /* $NetBSD: vfs_cache.c,v 1.13 1996/02/04 02:18:09 christos Exp $ */
3
4 /*
5 * Copyright (c) 1989, 1993
6 * The Regents of the University of California. All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
19 *
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
30 * SUCH DAMAGE.
31 *
32 * @(#)vfs_cache.c 8.3 (Berkeley) 8/22/94
33 */
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/vnode.h>
38 #include <sys/lock.h>
39 #include <sys/namei.h>
40 #include <sys/errno.h>
41 #include <sys/pool.h>
42
43 /*
44 * TODO: namecache access should really be locked.
45 */
46
47 /*
48 * For simplicity (and economy of storage), names longer than
49 * a maximum length of NAMECACHE_MAXLEN are not cached; they occur
50 * infrequently in any case, and are almost never of interest.
51 *
52 * Upon reaching the last segment of a path, if the reference
53 * is for DELETE, or NOCACHE is set (rewrite), and the
54 * name is located in the cache, it will be dropped.
55 */
56
57 /*
58 * Structures associated with name caching.
59 */
60 long numcache; /* total number of cache entries allocated */
61 long numneg; /* number of negative cache entries */
62
63 TAILQ_HEAD(, namecache) nclruhead; /* Regular Entry LRU chain */
64 TAILQ_HEAD(, namecache) nclruneghead; /* Negative Entry LRU chain */
65 struct nchstats nchstats; /* cache effectiveness statistics */
66
67 int doingcache = 1; /* 1 => enable the cache */
68
69 struct pool nch_pool;
70
71 void cache_zap(struct namecache *);
72 u_long nextvnodeid;
73
74 static inline int
75 namecache_compare(const struct namecache *n1, const struct namecache *n2)
76 {
77 if (n1->nc_nlen == n2->nc_nlen)
78 return (memcmp(n1->nc_name, n2->nc_name, n1->nc_nlen));
79 else
80 return (n1->nc_nlen - n2->nc_nlen);
81 }
82
83 RBT_PROTOTYPE(namecache_rb_cache, namecache, n_rbcache, namecache_compare);
84 RBT_GENERATE(namecache_rb_cache, namecache, n_rbcache, namecache_compare);
85
86 void
87 cache_tree_init(struct namecache_rb_cache *tree)
88 {
89 RBT_INIT(namecache_rb_cache, tree);
90 }
91
92 /*
93 * blow away a namecache entry
94 */
95 void
96 cache_zap(struct namecache *ncp)
97 {
98 struct vnode *dvp = NULL;
99
100 if (ncp->nc_vp != NULL) {
101 TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
102 numcache--;
103 } else {
104 TAILQ_REMOVE(&nclruneghead, ncp, nc_neg);
105 numneg--;
106 }
107 if (ncp->nc_dvp) {
108 RBT_REMOVE(namecache_rb_cache, &ncp->nc_dvp->v_nc_tree, ncp);
109 if (RBT_EMPTY(namecache_rb_cache, &ncp->nc_dvp->v_nc_tree))
110 dvp = ncp->nc_dvp;
111 }
112 if (ncp->nc_vp && (ncp->nc_vpid == ncp->nc_vp->v_id)) {
113 if (ncp->nc_vp != ncp->nc_dvp &&
114 ncp->nc_vp->v_type == VDIR &&
115 (ncp->nc_nlen > 2 ||
116 (ncp->nc_nlen > 1 &&
117 ncp->nc_name[1] != '.') ||
118 (ncp->nc_nlen > 0 &&
119 ncp->nc_name[0] != '.'))) {
120 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_me);
121 }
122 }
123 pool_put(&nch_pool, ncp);
124 if (dvp)
125 vdrop(dvp);
126 }
127
128 /*
129 * Look for a name in the cache.
130 * dvp points to the directory to search. The componentname cnp holds
131 * the information on the entry being sought, such as its length
132 * and its name. If the lookup succeeds, vpp is set to point to the vnode
133 * and an error of 0 is returned. If the lookup determines the name does
134 * not exist (negative caching) an error of ENOENT is returned. If the
135 * lookup fails, an error of -1 is returned.
136 */
137 int
138 cache_lookup(struct vnode *dvp, struct vnode **vpp,
139 struct componentname *cnp)
140 {
141 struct namecache *ncp;
142 struct namecache n;
143 struct vnode *vp;
144 u_long vpid;
145 int error;
146
147 *vpp = NULL;
148
149 if (!doingcache) {
150 cnp->cn_flags &= ~MAKEENTRY;
151 return (-1);
152 }
153 if (cnp->cn_namelen > NAMECACHE_MAXLEN) {
154 nchstats.ncs_long++;
155 cnp->cn_flags &= ~MAKEENTRY;
156 return (-1);
157 }
158
159 /* lookup in directory vnode's redblack tree */
160 n.nc_nlen = cnp->cn_namelen;
161 memcpy(n.nc_name, cnp->cn_nameptr, n.nc_nlen);
162 ncp = RBT_FIND(namecache_rb_cache, &dvp->v_nc_tree, &n);
163
164 if (ncp == NULL) {
165 nchstats.ncs_miss++;
166 return (-1);
167 }
168 if ((cnp->cn_flags & MAKEENTRY) == 0) {
169 nchstats.ncs_badhits++;
170 goto remove;
171 } else if (ncp->nc_vp == NULL) {
172 if (cnp->cn_nameiop != CREATE ||
173 (cnp->cn_flags & ISLASTCN) == 0) {
174 nchstats.ncs_neghits++;
175 /*
176 * Move this slot to end of the negative LRU chain,
177 */
178 if (TAILQ_NEXT(ncp, nc_neg) != NULL) {
179 TAILQ_REMOVE(&nclruneghead, ncp, nc_neg);
180 TAILQ_INSERT_TAIL(&nclruneghead, ncp,
181 nc_neg);
182 }
183 return (ENOENT);
184 } else {
185 nchstats.ncs_badhits++;
186 goto remove;
187 }
188 } else if (ncp->nc_vpid != ncp->nc_vp->v_id) {
189 nchstats.ncs_falsehits++;
190 goto remove;
191 }
192
193 /*
194 * Move this slot to end of the regular LRU chain.
195 */
196 if (TAILQ_NEXT(ncp, nc_lru) != NULL) {
197 TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
198 TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
199 }
200
201 vp = ncp->nc_vp;
202 vpid = vp->v_id;
203 if (vp == dvp) { /* lookup on "." */
204 vref(dvp);
205 error = 0;
206 } else if (cnp->cn_flags & ISDOTDOT) {
207 VOP_UNLOCK(dvp);
208 cnp->cn_flags |= PDIRUNLOCK;
209 error = vget(vp, LK_EXCLUSIVE);
210 /*
211 * If the above vget() succeeded and both LOCKPARENT and
212 * ISLASTCN is set, lock the directory vnode as well.
213 */
214 if (!error && (~cnp->cn_flags & (LOCKPARENT|ISLASTCN)) == 0) {
215 if ((error = vn_lock(dvp, LK_EXCLUSIVE)) != 0) {
216 vput(vp);
217 return (error);
218 }
219 cnp->cn_flags &= ~PDIRUNLOCK;
220 }
221 } else {
222 error = vget(vp, LK_EXCLUSIVE);
223 /*
224 * If the above vget() failed or either of LOCKPARENT or
225 * ISLASTCN is set, unlock the directory vnode.
226 */
227 if (error || (~cnp->cn_flags & (LOCKPARENT|ISLASTCN)) != 0) {
228 VOP_UNLOCK(dvp);
229 cnp->cn_flags |= PDIRUNLOCK;
230 }
231 }
232
233 /*
234 * Check that the lock succeeded, and that the capability number did
235 * not change while we were waiting for the lock.
236 */
237 if (error || vpid != vp->v_id) {
238 if (!error) {
239 vput(vp);
240 nchstats.ncs_falsehits++;
241 } else
242 nchstats.ncs_badhits++;
243 /*
244 * The parent needs to be locked when we return to VOP_LOOKUP().
245 * The `.' case here should be extremely rare (if it can happen
246 * at all), so we don't bother optimizing out the unlock/relock.
247 */
248 if (vp == dvp || error ||
249 (~cnp->cn_flags & (LOCKPARENT|ISLASTCN)) != 0) {
250 if ((error = vn_lock(dvp, LK_EXCLUSIVE)) != 0)
251 return (error);
252 cnp->cn_flags &= ~PDIRUNLOCK;
253 }
254 return (-1);
255 }
256
257 nchstats.ncs_goodhits++;
258 *vpp = vp;
259 return (0);
260
261 remove:
262 /*
263 * Last component and we are renaming or deleting,
264 * the cache entry is invalid, or otherwise don't
265 * want cache entry to exist.
266 */
267 cache_zap(ncp);
268 return (-1);
269 }
270
271 /*
272 * Scan cache looking for name of directory entry pointing at vp.
273 *
274 * Fill in dvpp.
275 *
276 * If bufp is non-NULL, also place the name in the buffer which starts
277 * at bufp, immediately before *bpp, and move bpp backwards to point
278 * at the start of it. (Yes, this is a little baroque, but it's done
279 * this way to cater to the whims of getcwd).
280 *
281 * Returns 0 on success, -1 on cache miss, positive errno on failure.
282 *
283 * TODO: should we return *dvpp locked?
284 */
285
286 int
287 cache_revlookup(struct vnode *vp, struct vnode **dvpp, char **bpp, char *bufp)
288 {
289 struct namecache *ncp;
290 struct vnode *dvp = NULL;
291 char *bp;
292
293 if (!doingcache)
294 goto out;
295 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_me) {
296 dvp = ncp->nc_dvp;
297 if (dvp && dvp != vp && ncp->nc_dvpid == dvp->v_id)
298 goto found;
299 }
300 goto miss;
301 found:
302 #ifdef DIAGNOSTIC
303 if (ncp->nc_nlen == 1 &&
304 ncp->nc_name[0] == '.')
305 panic("cache_revlookup: found entry for .");
306 if (ncp->nc_nlen == 2 &&
307 ncp->nc_name[0] == '.' &&
308 ncp->nc_name[1] == '.')
309 panic("cache_revlookup: found entry for ..");
310 #endif
311 nchstats.ncs_revhits++;
312
313 if (bufp != NULL) {
314 bp = *bpp;
315 bp -= ncp->nc_nlen;
316 if (bp <= bufp) {
317 *dvpp = NULL;
318 return (ERANGE);
319 }
320 memcpy(bp, ncp->nc_name, ncp->nc_nlen);
321 *bpp = bp;
322 }
323
324 *dvpp = dvp;
325
326 /*
327 * XXX: Should we vget() here to have more
328 * consistent semantics with cache_lookup()?
329 */
330 return (0);
331
332 miss:
333 nchstats.ncs_revmiss++;
334 out:
335 *dvpp = NULL;
336 return (-1);
337 }
338
339 /*
340 * Add an entry to the cache
341 */
342 void
343 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
344 {
345 struct namecache *ncp, *lncp;
346
347 if (!doingcache || cnp->cn_namelen > NAMECACHE_MAXLEN)
348 return;
349
350 /*
351 * allocate, or recycle (free and allocate) an ncp.
352 */
353 if (numcache >= initialvnodes) {
354 if ((ncp = TAILQ_FIRST(&nclruhead)) != NULL)
355 cache_zap(ncp);
356 else if ((ncp = TAILQ_FIRST(&nclruneghead)) != NULL)
357 cache_zap(ncp);
358 else
359 panic("wtf? leak?");
360 }
361 ncp = pool_get(&nch_pool, PR_WAITOK|PR_ZERO);
362
363 /* grab the vnode we just found */
364 ncp->nc_vp = vp;
365 if (vp)
366 ncp->nc_vpid = vp->v_id;
367
368 /* fill in cache info */
369 ncp->nc_dvp = dvp;
370 ncp->nc_dvpid = dvp->v_id;
371 ncp->nc_nlen = cnp->cn_namelen;
372 memcpy(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen);
373 if (RBT_EMPTY(namecache_rb_cache, &dvp->v_nc_tree)) {
374 vhold(dvp);
375 }
376 if ((lncp = RBT_INSERT(namecache_rb_cache, &dvp->v_nc_tree, ncp))
377 != NULL) {
378 /* someone has raced us and added a different entry
379 * for the same vnode (different ncp) - we don't need
380 * this entry, so free it and we are done.
381 */
382 pool_put(&nch_pool, ncp);
383 /* we know now dvp->v_nc_tree is not empty, no need
384 * to vdrop here
385 */
386 goto done;
387 }
388 if (vp) {
389 TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
390 numcache++;
391 /* don't put . or .. in the reverse map */
392 if (vp != dvp && vp->v_type == VDIR &&
393 (ncp->nc_nlen > 2 ||
394 (ncp->nc_nlen > 1 &&
395 ncp->nc_name[1] != '.') ||
396 (ncp->nc_nlen > 0 &&
397 ncp->nc_name[0] != '.')))
398 TAILQ_INSERT_TAIL(&vp->v_cache_dst, ncp,
399 nc_me);
400 } else {
401 TAILQ_INSERT_TAIL(&nclruneghead, ncp, nc_neg);
402 numneg++;
403 }
404 if (numneg > initialvnodes) {
405 if ((ncp = TAILQ_FIRST(&nclruneghead))
406 != NULL)
407 cache_zap(ncp);
408 }
409 done:
410 return;
411 }
412
413
414 /*
415 * Name cache initialization, from vfs_init() when we are booting
416 */
417 void
418 nchinit(void)
419 {
420 TAILQ_INIT(&nclruhead);
421 TAILQ_INIT(&nclruneghead);
422 pool_init(&nch_pool, sizeof(struct namecache), 0, IPL_NONE, PR_WAITOK,
423 "nchpl", NULL);
424 }
425
426 /*
427 * Cache flush, a particular vnode; called when a vnode is renamed to
428 * hide entries that would now be invalid
429 */
430 void
431 cache_purge(struct vnode *vp)
432 {
433 struct namecache *ncp;
434
435 /* We should never have destinations cached for a non-VDIR vnode. */
436 KASSERT(vp->v_type == VDIR || TAILQ_EMPTY(&vp->v_cache_dst));
437
438 while ((ncp = TAILQ_FIRST(&vp->v_cache_dst)))
439 cache_zap(ncp);
440 while ((ncp = RBT_ROOT(namecache_rb_cache, &vp->v_nc_tree)))
441 cache_zap(ncp);
442
443 /* XXX this blows goats */
444 vp->v_id = ++nextvnodeid;
445 if (vp->v_id == 0)
446 vp->v_id = ++nextvnodeid;
447 }
448
449 /*
450 * Cache flush, a whole filesystem; called when filesys is umounted to
451 * remove entries that would now be invalid
452 */
453 void
454 cache_purgevfs(struct mount *mp)
455 {
456 struct namecache *ncp, *nxtcp;
457
458 /* whack the regular entries */
459 TAILQ_FOREACH_SAFE(ncp, &nclruhead, nc_lru, nxtcp) {
460 if (ncp->nc_dvp == NULL || ncp->nc_dvp->v_mount != mp)
461 continue;
462 /* free the resources we had */
463 cache_zap(ncp);
464 }
465 /* whack the negative entries */
466 TAILQ_FOREACH_SAFE(ncp, &nclruneghead, nc_neg, nxtcp) {
467 if (ncp->nc_dvp == NULL || ncp->nc_dvp->v_mount != mp)
468 continue;
469 /* free the resources we had */
470 cache_zap(ncp);
471 }
472 }
Cache object: 1286ef1464c1d0dc0d2292c5b35e5dfa
|