FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_cache.c
1 /*-
2 * Copyright (c) 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Poul-Henning Kamp of the FreeBSD Project.
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 * 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.
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 8.5 (Berkeley) 3/22/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/8.0/sys/kern/vfs_cache.c 196207 2009-08-14 11:06:58Z kib $");
37
38 #include "opt_kdtrace.h"
39 #include "opt_ktrace.h"
40
41 #include <sys/param.h>
42 #include <sys/filedesc.h>
43 #include <sys/fnv_hash.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/namei.h>
49 #include <sys/proc.h>
50 #include <sys/rwlock.h>
51 #include <sys/sdt.h>
52 #include <sys/syscallsubr.h>
53 #include <sys/sysctl.h>
54 #include <sys/sysproto.h>
55 #include <sys/systm.h>
56 #include <sys/vnode.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60
61 #include <vm/uma.h>
62
63 SDT_PROVIDER_DECLARE(vfs);
64 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
65 "struct vnode *");
66 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
67 "char *");
68 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *");
69 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *",
70 "struct char *", "struct vnode *");
71 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *");
72 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", "struct vnode *",
73 "struct char *");
74 SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *",
75 "struct vnode *");
76 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit_negative, "struct vnode *",
77 "char *");
78 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
79 "char *");
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 *",
84 "struct vnode *");
85 SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *",
86 "char *");
87
88 /*
89 * This structure describes the elements in the cache of recent
90 * names looked up by namei.
91 */
92
93 struct namecache {
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 */
102 };
103
104 /*
105 * Name caching works as follows:
106 *
107 * Names found by directory scans are retained in a cache
108 * for future reference. It is managed LRU, so frequently
109 * used names will hang around. Cache is indexed by hash value
110 * obtained from (vp, name) where vp refers to the directory
111 * containing name.
112 *
113 * If it is a "negative" entry, (i.e. for a name that is known NOT to
114 * exist) the vnode pointer will be NULL.
115 *
116 * Upon reaching the last segment of a path, if the reference
117 * is for DELETE, or NOCACHE is set (rewrite), and the
118 * name is located in the cache, it will be dropped.
119 */
120
121 /*
122 * Structures associated with name cacheing.
123 */
124 #define NCHHASH(hash) \
125 (&nchashtbl[(hash) & nchash])
126 static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
127 static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
128 static u_long nchash; /* size of hash table */
129 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
130 static u_long ncnegfactor = 16; /* ratio of negative entries */
131 SYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
132 static u_long numneg; /* number of cache entries allocated */
133 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
134 static u_long numcache; /* number of cache entries allocated */
135 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
136 static u_long numcachehv; /* number of cache entries with vnodes held */
137 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, "");
138 #if 0
139 static u_long numcachepl; /* number of cache purge for leaf entries */
140 SYSCTL_ULONG(_debug, OID_AUTO, numcachepl, CTLFLAG_RD, &numcachepl, 0, "");
141 #endif
142 struct nchstats nchstats; /* cache effectiveness statistics */
143
144 static struct rwlock cache_lock;
145 RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
146
147 #define CACHE_UPGRADE_LOCK() rw_try_upgrade(&cache_lock)
148 #define CACHE_RLOCK() rw_rlock(&cache_lock)
149 #define CACHE_RUNLOCK() rw_runlock(&cache_lock)
150 #define CACHE_WLOCK() rw_wlock(&cache_lock)
151 #define CACHE_WUNLOCK() rw_wunlock(&cache_lock)
152
153 /*
154 * UMA zones for the VFS cache.
155 *
156 * The small cache is used for entries with short names, which are the
157 * most common. The large cache is used for entries which are too big to
158 * fit in the small cache.
159 */
160 static uma_zone_t cache_zone_small;
161 static uma_zone_t cache_zone_large;
162
163 #define CACHE_PATH_CUTOFF 35
164 #define CACHE_ZONE_SMALL (sizeof(struct namecache) + CACHE_PATH_CUTOFF \
165 + 1)
166 #define CACHE_ZONE_LARGE (sizeof(struct namecache) + NAME_MAX + 1)
167
168 #define cache_alloc(len) uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \
169 cache_zone_small : cache_zone_large, M_WAITOK)
170 #define cache_free(ncp) do { \
171 if (ncp != NULL) \
172 uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \
173 cache_zone_small : cache_zone_large, (ncp)); \
174 } while (0)
175
176 static int doingcache = 1; /* 1 => enable the cache */
177 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
178
179 /* Export size information to userland */
180 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0,
181 sizeof(struct namecache), "");
182
183 /*
184 * The new name cache statistics
185 */
186 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
187 #define STATNODE(mode, name, var) \
188 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
189 STATNODE(CTLFLAG_RD, numneg, &numneg);
190 STATNODE(CTLFLAG_RD, numcache, &numcache);
191 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
192 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
193 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
194 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
195 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
196 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
197 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
198 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
199 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
200 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
201 static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades);
202
203 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
204 &nchstats, sizeof(nchstats), "LU", "VFS cache effectiveness statistics");
205
206
207
208 static void cache_zap(struct namecache *ncp);
209 static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
210 u_int *buflen);
211 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
212 char *buf, char **retbuf, u_int buflen);
213
214 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
215
216 /*
217 * Flags in namecache.nc_flag
218 */
219 #define NCF_WHITE 0x01
220 #define NCF_ISDOTDOT 0x02
221
222 #ifdef DIAGNOSTIC
223 /*
224 * Grab an atomic snapshot of the name cache hash chain lengths
225 */
226 SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
227
228 static int
229 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
230 {
231 int error;
232 struct nchashhead *ncpp;
233 struct namecache *ncp;
234 int n_nchash;
235 int count;
236
237 n_nchash = nchash + 1; /* nchash is max index, not count */
238 if (!req->oldptr)
239 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
240
241 /* Scan hash tables for applicable entries */
242 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
243 CACHE_RLOCK();
244 count = 0;
245 LIST_FOREACH(ncp, ncpp, nc_hash) {
246 count++;
247 }
248 CACHE_RUNLOCK();
249 error = SYSCTL_OUT(req, &count, sizeof(count));
250 if (error)
251 return (error);
252 }
253 return (0);
254 }
255 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
256 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
257 "nchash chain lengths");
258
259 static int
260 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
261 {
262 int error;
263 struct nchashhead *ncpp;
264 struct namecache *ncp;
265 int n_nchash;
266 int count, maxlength, used, pct;
267
268 if (!req->oldptr)
269 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
270
271 n_nchash = nchash + 1; /* nchash is max index, not count */
272 used = 0;
273 maxlength = 0;
274
275 /* Scan hash tables for applicable entries */
276 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
277 count = 0;
278 CACHE_RLOCK();
279 LIST_FOREACH(ncp, ncpp, nc_hash) {
280 count++;
281 }
282 CACHE_RUNLOCK();
283 if (count)
284 used++;
285 if (maxlength < count)
286 maxlength = count;
287 }
288 n_nchash = nchash + 1;
289 pct = (used * 100 * 100) / n_nchash;
290 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
291 if (error)
292 return (error);
293 error = SYSCTL_OUT(req, &used, sizeof(used));
294 if (error)
295 return (error);
296 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
297 if (error)
298 return (error);
299 error = SYSCTL_OUT(req, &pct, sizeof(pct));
300 if (error)
301 return (error);
302 return (0);
303 }
304 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
305 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
306 "nchash chain lengths");
307 #endif
308
309 /*
310 * cache_zap():
311 *
312 * Removes a namecache entry from cache, whether it contains an actual
313 * pointer to a vnode or if it is just a negative cache entry.
314 */
315 static void
316 cache_zap(ncp)
317 struct namecache *ncp;
318 {
319 struct vnode *vp;
320
321 rw_assert(&cache_lock, RA_WLOCKED);
322 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
323 #ifdef KDTRACE_HOOKS
324 if (ncp->nc_vp != NULL) {
325 SDT_PROBE(vfs, namecache, zap, done, ncp->nc_dvp,
326 ncp->nc_name, ncp->nc_vp, 0, 0);
327 } else {
328 SDT_PROBE(vfs, namecache, zap_negative, done, ncp->nc_dvp,
329 ncp->nc_name, 0, 0, 0);
330 }
331 #endif
332 vp = NULL;
333 LIST_REMOVE(ncp, nc_hash);
334 if (ncp->nc_flag & NCF_ISDOTDOT) {
335 if (ncp == ncp->nc_dvp->v_cache_dd)
336 ncp->nc_dvp->v_cache_dd = NULL;
337 } else {
338 LIST_REMOVE(ncp, nc_src);
339 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
340 vp = ncp->nc_dvp;
341 numcachehv--;
342 }
343 }
344 if (ncp->nc_vp) {
345 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
346 if (ncp == ncp->nc_vp->v_cache_dd)
347 ncp->nc_vp->v_cache_dd = NULL;
348 } else {
349 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
350 numneg--;
351 }
352 numcache--;
353 cache_free(ncp);
354 if (vp)
355 vdrop(vp);
356 }
357
358 /*
359 * Lookup an entry in the cache
360 *
361 * Lookup is called with dvp pointing to the directory to search,
362 * cnp pointing to the name of the entry being sought. If the lookup
363 * succeeds, the vnode is returned in *vpp, and a status of -1 is
364 * returned. If the lookup determines that the name does not exist
365 * (negative cacheing), a status of ENOENT is returned. If the lookup
366 * fails, a status of zero is returned. If the directory vnode is
367 * recycled out from under us due to a forced unmount, a status of
368 * ENOENT is returned.
369 *
370 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
371 * unlocked. If we're looking up . an extra ref is taken, but the lock is
372 * not recursively acquired.
373 */
374
375 int
376 cache_lookup(dvp, vpp, cnp)
377 struct vnode *dvp;
378 struct vnode **vpp;
379 struct componentname *cnp;
380 {
381 struct namecache *ncp;
382 u_int32_t hash;
383 int error, ltype, wlocked;
384
385 if (!doingcache) {
386 cnp->cn_flags &= ~MAKEENTRY;
387 return (0);
388 }
389 retry:
390 CACHE_RLOCK();
391 wlocked = 0;
392 numcalls++;
393 error = 0;
394
395 retry_wlocked:
396 if (cnp->cn_nameptr[0] == '.') {
397 if (cnp->cn_namelen == 1) {
398 *vpp = dvp;
399 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
400 dvp, cnp->cn_nameptr);
401 dothits++;
402 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ".",
403 *vpp, 0, 0);
404 goto success;
405 }
406 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
407 dotdothits++;
408 if (dvp->v_cache_dd == NULL) {
409 SDT_PROBE(vfs, namecache, lookup, miss, dvp,
410 "..", NULL, 0, 0);
411 goto unlock;
412 }
413 if ((cnp->cn_flags & MAKEENTRY) == 0) {
414 if (!wlocked && !CACHE_UPGRADE_LOCK())
415 goto wlock;
416 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
417 cache_zap(dvp->v_cache_dd);
418 dvp->v_cache_dd = NULL;
419 CACHE_WUNLOCK();
420 return (0);
421 }
422 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
423 *vpp = dvp->v_cache_dd->nc_vp;
424 else
425 *vpp = dvp->v_cache_dd->nc_dvp;
426 /* Return failure if negative entry was found. */
427 if (*vpp == NULL) {
428 ncp = dvp->v_cache_dd;
429 goto negative_success;
430 }
431 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
432 dvp, cnp->cn_nameptr, *vpp);
433 SDT_PROBE(vfs, namecache, lookup, hit, dvp, "..",
434 *vpp, 0, 0);
435 goto success;
436 }
437 }
438
439 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
440 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
441 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
442 numchecks++;
443 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
444 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
445 break;
446 }
447
448 /* We failed to find an entry */
449 if (ncp == NULL) {
450 SDT_PROBE(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
451 NULL, 0, 0);
452 if ((cnp->cn_flags & MAKEENTRY) == 0) {
453 nummisszap++;
454 } else {
455 nummiss++;
456 }
457 nchstats.ncs_miss++;
458 goto unlock;
459 }
460
461 /* We don't want to have an entry, so dump it */
462 if ((cnp->cn_flags & MAKEENTRY) == 0) {
463 numposzaps++;
464 nchstats.ncs_badhits++;
465 if (!wlocked && !CACHE_UPGRADE_LOCK())
466 goto wlock;
467 cache_zap(ncp);
468 CACHE_WUNLOCK();
469 return (0);
470 }
471
472 /* We found a "positive" match, return the vnode */
473 if (ncp->nc_vp) {
474 numposhits++;
475 nchstats.ncs_goodhits++;
476 *vpp = ncp->nc_vp;
477 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
478 dvp, cnp->cn_nameptr, *vpp, ncp);
479 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ncp->nc_name,
480 *vpp, 0, 0);
481 goto success;
482 }
483
484 negative_success:
485 /* We found a negative match, and want to create it, so purge */
486 if (cnp->cn_nameiop == CREATE) {
487 numnegzaps++;
488 nchstats.ncs_badhits++;
489 if (!wlocked && !CACHE_UPGRADE_LOCK())
490 goto wlock;
491 cache_zap(ncp);
492 CACHE_WUNLOCK();
493 return (0);
494 }
495
496 if (!wlocked && !CACHE_UPGRADE_LOCK())
497 goto wlock;
498 numneghits++;
499 /*
500 * We found a "negative" match, so we shift it to the end of
501 * the "negative" cache entries queue to satisfy LRU. Also,
502 * check to see if the entry is a whiteout; indicate this to
503 * the componentname, if so.
504 */
505 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
506 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
507 nchstats.ncs_neghits++;
508 if (ncp->nc_flag & NCF_WHITE)
509 cnp->cn_flags |= ISWHITEOUT;
510 SDT_PROBE(vfs, namecache, lookup, hit_negative, dvp, ncp->nc_name,
511 0, 0, 0);
512 CACHE_WUNLOCK();
513 return (ENOENT);
514
515 wlock:
516 /*
517 * We need to update the cache after our lookup, so upgrade to
518 * a write lock and retry the operation.
519 */
520 CACHE_RUNLOCK();
521 CACHE_WLOCK();
522 numupgrades++;
523 wlocked = 1;
524 goto retry_wlocked;
525
526 success:
527 /*
528 * On success we return a locked and ref'd vnode as per the lookup
529 * protocol.
530 */
531 if (dvp == *vpp) { /* lookup on "." */
532 VREF(*vpp);
533 if (wlocked)
534 CACHE_WUNLOCK();
535 else
536 CACHE_RUNLOCK();
537 /*
538 * When we lookup "." we still can be asked to lock it
539 * differently...
540 */
541 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
542 if (ltype != VOP_ISLOCKED(*vpp)) {
543 if (ltype == LK_EXCLUSIVE) {
544 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
545 if ((*vpp)->v_iflag & VI_DOOMED) {
546 /* forced unmount */
547 vrele(*vpp);
548 *vpp = NULL;
549 return (ENOENT);
550 }
551 } else
552 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
553 }
554 return (-1);
555 }
556 ltype = 0; /* silence gcc warning */
557 if (cnp->cn_flags & ISDOTDOT) {
558 ltype = VOP_ISLOCKED(dvp);
559 VOP_UNLOCK(dvp, 0);
560 }
561 VI_LOCK(*vpp);
562 if (wlocked)
563 CACHE_WUNLOCK();
564 else
565 CACHE_RUNLOCK();
566 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
567 if (cnp->cn_flags & ISDOTDOT) {
568 vn_lock(dvp, ltype | LK_RETRY);
569 if (dvp->v_iflag & VI_DOOMED) {
570 if (error == 0)
571 vput(*vpp);
572 *vpp = NULL;
573 return (ENOENT);
574 }
575 }
576 if (error) {
577 *vpp = NULL;
578 goto retry;
579 }
580 if ((cnp->cn_flags & ISLASTCN) &&
581 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
582 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
583 }
584 return (-1);
585
586 unlock:
587 if (wlocked)
588 CACHE_WUNLOCK();
589 else
590 CACHE_RUNLOCK();
591 return (0);
592 }
593
594 /*
595 * Add an entry to the cache.
596 */
597 void
598 cache_enter(dvp, vp, cnp)
599 struct vnode *dvp;
600 struct vnode *vp;
601 struct componentname *cnp;
602 {
603 struct namecache *ncp, *n2;
604 struct nchashhead *ncpp;
605 u_int32_t hash;
606 int flag;
607 int hold;
608 int zap;
609 int len;
610
611 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
612 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
613 ("cahe_enter: Adding a doomed vnode"));
614
615 if (!doingcache)
616 return;
617
618 /*
619 * Avoid blowout in namecache entries.
620 */
621 if (numcache >= desiredvnodes * 2)
622 return;
623
624 flag = 0;
625 if (cnp->cn_nameptr[0] == '.') {
626 if (cnp->cn_namelen == 1)
627 return;
628 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
629 CACHE_WLOCK();
630 /*
631 * If dotdot entry already exists, just retarget it
632 * to new parent vnode, otherwise continue with new
633 * namecache entry allocation.
634 */
635 if ((ncp = dvp->v_cache_dd) != NULL &&
636 ncp->nc_flag & NCF_ISDOTDOT) {
637 KASSERT(ncp->nc_dvp == dvp,
638 ("wrong isdotdot parent"));
639 if (ncp->nc_vp != NULL)
640 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
641 ncp, nc_dst);
642 else
643 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
644 if (vp != NULL)
645 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
646 ncp, nc_dst);
647 else
648 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
649 ncp->nc_vp = vp;
650 CACHE_WUNLOCK();
651 return;
652 }
653 dvp->v_cache_dd = NULL;
654 SDT_PROBE(vfs, namecache, enter, done, dvp, "..", vp,
655 0, 0);
656 CACHE_WUNLOCK();
657 flag = NCF_ISDOTDOT;
658 }
659 }
660
661 hold = 0;
662 zap = 0;
663
664 /*
665 * Calculate the hash key and setup as much of the new
666 * namecache entry as possible before acquiring the lock.
667 */
668 ncp = cache_alloc(cnp->cn_namelen);
669 ncp->nc_vp = vp;
670 ncp->nc_dvp = dvp;
671 ncp->nc_flag = flag;
672 len = ncp->nc_nlen = cnp->cn_namelen;
673 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
674 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1);
675 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
676 CACHE_WLOCK();
677
678 /*
679 * See if this vnode or negative entry is already in the cache
680 * with this name. This can happen with concurrent lookups of
681 * the same path name.
682 */
683 ncpp = NCHHASH(hash);
684 LIST_FOREACH(n2, ncpp, nc_hash) {
685 if (n2->nc_dvp == dvp &&
686 n2->nc_nlen == cnp->cn_namelen &&
687 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
688 CACHE_WUNLOCK();
689 cache_free(ncp);
690 return;
691 }
692 }
693
694 if (flag == NCF_ISDOTDOT) {
695 /*
696 * See if we are trying to add .. entry, but some other lookup
697 * has populated v_cache_dd pointer already.
698 */
699 if (dvp->v_cache_dd != NULL) {
700 CACHE_WUNLOCK();
701 cache_free(ncp);
702 return;
703 }
704 KASSERT(vp == NULL || vp->v_type == VDIR,
705 ("wrong vnode type %p", vp));
706 dvp->v_cache_dd = ncp;
707 }
708
709 numcache++;
710 if (!vp) {
711 numneg++;
712 if (cnp->cn_flags & ISWHITEOUT)
713 ncp->nc_flag |= NCF_WHITE;
714 } else if (vp->v_type == VDIR) {
715 if (flag != NCF_ISDOTDOT) {
716 if ((n2 = vp->v_cache_dd) != NULL &&
717 (n2->nc_flag & NCF_ISDOTDOT) != 0)
718 cache_zap(n2);
719 vp->v_cache_dd = ncp;
720 }
721 } else {
722 vp->v_cache_dd = NULL;
723 }
724
725 /*
726 * Insert the new namecache entry into the appropriate chain
727 * within the cache entries table.
728 */
729 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
730 if (flag != NCF_ISDOTDOT) {
731 if (LIST_EMPTY(&dvp->v_cache_src)) {
732 hold = 1;
733 numcachehv++;
734 }
735 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
736 }
737
738 /*
739 * If the entry is "negative", we place it into the
740 * "negative" cache queue, otherwise, we place it into the
741 * destination vnode's cache entries queue.
742 */
743 if (vp) {
744 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
745 SDT_PROBE(vfs, namecache, enter, done, dvp, ncp->nc_name, vp,
746 0, 0);
747 } else {
748 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
749 SDT_PROBE(vfs, namecache, enter_negative, done, dvp,
750 ncp->nc_name, 0, 0, 0);
751 }
752 if (numneg * ncnegfactor > numcache) {
753 ncp = TAILQ_FIRST(&ncneg);
754 zap = 1;
755 }
756 if (hold)
757 vhold(dvp);
758 if (zap)
759 cache_zap(ncp);
760 CACHE_WUNLOCK();
761 }
762
763 /*
764 * Name cache initialization, from vfs_init() when we are booting
765 */
766 static void
767 nchinit(void *dummy __unused)
768 {
769
770 TAILQ_INIT(&ncneg);
771
772 cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
773 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
774 cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
775 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
776
777 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
778 }
779 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
780
781
782 /*
783 * Invalidate all entries to a particular vnode.
784 */
785 void
786 cache_purge(vp)
787 struct vnode *vp;
788 {
789
790 CTR1(KTR_VFS, "cache_purge(%p)", vp);
791 SDT_PROBE(vfs, namecache, purge, done, vp, 0, 0, 0, 0);
792 CACHE_WLOCK();
793 while (!LIST_EMPTY(&vp->v_cache_src))
794 cache_zap(LIST_FIRST(&vp->v_cache_src));
795 while (!TAILQ_EMPTY(&vp->v_cache_dst))
796 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
797 if (vp->v_cache_dd != NULL) {
798 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
799 ("lost dotdot link"));
800 cache_zap(vp->v_cache_dd);
801 }
802 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
803 CACHE_WUNLOCK();
804 }
805
806 /*
807 * Invalidate all negative entries for a particular directory vnode.
808 */
809 void
810 cache_purge_negative(vp)
811 struct vnode *vp;
812 {
813 struct namecache *cp, *ncp;
814
815 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
816 SDT_PROBE(vfs, namecache, purge_negative, done, vp, 0, 0, 0, 0);
817 CACHE_WLOCK();
818 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
819 if (cp->nc_vp == NULL)
820 cache_zap(cp);
821 }
822 CACHE_WUNLOCK();
823 }
824
825 /*
826 * Flush all entries referencing a particular filesystem.
827 */
828 void
829 cache_purgevfs(mp)
830 struct mount *mp;
831 {
832 struct nchashhead *ncpp;
833 struct namecache *ncp, *nnp;
834
835 /* Scan hash tables for applicable entries */
836 SDT_PROBE(vfs, namecache, purgevfs, done, mp, 0, 0, 0, 0);
837 CACHE_WLOCK();
838 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
839 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
840 if (ncp->nc_dvp->v_mount == mp)
841 cache_zap(ncp);
842 }
843 }
844 CACHE_WUNLOCK();
845 }
846
847 /*
848 * Perform canonical checks and cache lookup and pass on to filesystem
849 * through the vop_cachedlookup only if needed.
850 */
851
852 int
853 vfs_cache_lookup(ap)
854 struct vop_lookup_args /* {
855 struct vnode *a_dvp;
856 struct vnode **a_vpp;
857 struct componentname *a_cnp;
858 } */ *ap;
859 {
860 struct vnode *dvp;
861 int error;
862 struct vnode **vpp = ap->a_vpp;
863 struct componentname *cnp = ap->a_cnp;
864 struct ucred *cred = cnp->cn_cred;
865 int flags = cnp->cn_flags;
866 struct thread *td = cnp->cn_thread;
867
868 *vpp = NULL;
869 dvp = ap->a_dvp;
870
871 if (dvp->v_type != VDIR)
872 return (ENOTDIR);
873
874 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
875 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
876 return (EROFS);
877
878 error = VOP_ACCESS(dvp, VEXEC, cred, td);
879 if (error)
880 return (error);
881
882 error = cache_lookup(dvp, vpp, cnp);
883 if (error == 0)
884 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
885 if (error == -1)
886 return (0);
887 return (error);
888 }
889
890
891 #ifndef _SYS_SYSPROTO_H_
892 struct __getcwd_args {
893 u_char *buf;
894 u_int buflen;
895 };
896 #endif
897
898 /*
899 * XXX All of these sysctls would probably be more productive dead.
900 */
901 static int disablecwd;
902 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
903 "Disable the getcwd syscall");
904
905 /* Implementation of the getcwd syscall. */
906 int
907 __getcwd(td, uap)
908 struct thread *td;
909 struct __getcwd_args *uap;
910 {
911
912 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
913 }
914
915 int
916 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
917 {
918 char *bp, *tmpbuf;
919 struct filedesc *fdp;
920 struct vnode *cdir, *rdir;
921 int error, vfslocked;
922
923 if (disablecwd)
924 return (ENODEV);
925 if (buflen < 2)
926 return (EINVAL);
927 if (buflen > MAXPATHLEN)
928 buflen = MAXPATHLEN;
929
930 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
931 fdp = td->td_proc->p_fd;
932 FILEDESC_SLOCK(fdp);
933 cdir = fdp->fd_cdir;
934 VREF(cdir);
935 rdir = fdp->fd_rdir;
936 VREF(rdir);
937 FILEDESC_SUNLOCK(fdp);
938 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
939 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
940 vrele(rdir);
941 VFS_UNLOCK_GIANT(vfslocked);
942 vfslocked = VFS_LOCK_GIANT(cdir->v_mount);
943 vrele(cdir);
944 VFS_UNLOCK_GIANT(vfslocked);
945
946 if (!error) {
947 if (bufseg == UIO_SYSSPACE)
948 bcopy(bp, buf, strlen(bp) + 1);
949 else
950 error = copyout(bp, buf, strlen(bp) + 1);
951 #ifdef KTRACE
952 if (KTRPOINT(curthread, KTR_NAMEI))
953 ktrnamei(bp);
954 #endif
955 }
956 free(tmpbuf, M_TEMP);
957 return (error);
958 }
959
960 /*
961 * Thus begins the fullpath magic.
962 */
963
964 #undef STATNODE
965 #define STATNODE(name) \
966 static u_int name; \
967 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
968
969 static int disablefullpath;
970 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
971 "Disable the vn_fullpath function");
972
973 /* These count for kern___getcwd(), too. */
974 STATNODE(numfullpathcalls);
975 STATNODE(numfullpathfail1);
976 STATNODE(numfullpathfail2);
977 STATNODE(numfullpathfail4);
978 STATNODE(numfullpathfound);
979
980 /*
981 * Retrieve the full filesystem path that correspond to a vnode from the name
982 * cache (if available)
983 */
984 int
985 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
986 {
987 char *buf;
988 struct filedesc *fdp;
989 struct vnode *rdir;
990 int error, vfslocked;
991
992 if (disablefullpath)
993 return (ENODEV);
994 if (vn == NULL)
995 return (EINVAL);
996
997 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
998 fdp = td->td_proc->p_fd;
999 FILEDESC_SLOCK(fdp);
1000 rdir = fdp->fd_rdir;
1001 VREF(rdir);
1002 FILEDESC_SUNLOCK(fdp);
1003 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1004 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
1005 vrele(rdir);
1006 VFS_UNLOCK_GIANT(vfslocked);
1007
1008 if (!error)
1009 *freebuf = buf;
1010 else
1011 free(buf, M_TEMP);
1012 return (error);
1013 }
1014
1015 /*
1016 * This function is similar to vn_fullpath, but it attempts to lookup the
1017 * pathname relative to the global root mount point. This is required for the
1018 * auditing sub-system, as audited pathnames must be absolute, relative to the
1019 * global root mount point.
1020 */
1021 int
1022 vn_fullpath_global(struct thread *td, struct vnode *vn,
1023 char **retbuf, char **freebuf)
1024 {
1025 char *buf;
1026 int error;
1027
1028 if (disablefullpath)
1029 return (ENODEV);
1030 if (vn == NULL)
1031 return (EINVAL);
1032 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1033 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1034 if (!error)
1035 *freebuf = buf;
1036 else
1037 free(buf, M_TEMP);
1038 return (error);
1039 }
1040
1041 int
1042 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1043 {
1044 int error;
1045
1046 CACHE_RLOCK();
1047 error = vn_vptocnp_locked(vp, cred, buf, buflen);
1048 if (error == 0) {
1049 /*
1050 * vn_vptocnp_locked() dropped hold acquired by
1051 * VOP_VPTOCNP immediately after locking the
1052 * cache. Since we are going to drop the cache rlock,
1053 * re-hold the result.
1054 */
1055 vhold(*vp);
1056 CACHE_RUNLOCK();
1057 }
1058 return (error);
1059 }
1060
1061 static int
1062 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1063 u_int *buflen)
1064 {
1065 struct vnode *dvp;
1066 struct namecache *ncp;
1067 int error, vfslocked;
1068
1069 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1070 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1071 break;
1072 }
1073 if (ncp != NULL) {
1074 if (*buflen < ncp->nc_nlen) {
1075 CACHE_RUNLOCK();
1076 numfullpathfail4++;
1077 error = ENOMEM;
1078 SDT_PROBE(vfs, namecache, fullpath, return, error,
1079 vp, NULL, 0, 0);
1080 return (error);
1081 }
1082 *buflen -= ncp->nc_nlen;
1083 memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen);
1084 SDT_PROBE(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1085 ncp->nc_name, vp, 0, 0);
1086 *vp = ncp->nc_dvp;
1087 return (0);
1088 }
1089 SDT_PROBE(vfs, namecache, fullpath, miss, vp, 0, 0, 0, 0);
1090
1091 vhold(*vp);
1092 CACHE_RUNLOCK();
1093 vfslocked = VFS_LOCK_GIANT((*vp)->v_mount);
1094 vn_lock(*vp, LK_SHARED | LK_RETRY);
1095 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1096 VOP_UNLOCK(*vp, 0);
1097 vdrop(*vp);
1098 VFS_UNLOCK_GIANT(vfslocked);
1099 if (error) {
1100 numfullpathfail2++;
1101 SDT_PROBE(vfs, namecache, fullpath, return, error, vp,
1102 NULL, 0, 0);
1103 return (error);
1104 }
1105
1106 *vp = dvp;
1107 CACHE_RLOCK();
1108 if ((*vp)->v_iflag & VI_DOOMED) {
1109 /* forced unmount */
1110 CACHE_RUNLOCK();
1111 vdrop(*vp);
1112 error = ENOENT;
1113 SDT_PROBE(vfs, namecache, fullpath, return, error, vp,
1114 NULL, 0, 0);
1115 return (error);
1116 }
1117 vdrop(*vp);
1118
1119 return (0);
1120 }
1121
1122 /*
1123 * The magic behind kern___getcwd() and vn_fullpath().
1124 */
1125 static int
1126 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1127 char *buf, char **retbuf, u_int buflen)
1128 {
1129 int error, slash_prefixed;
1130 #ifdef KDTRACE_HOOKS
1131 struct vnode *startvp = vp;
1132 #endif
1133
1134 buflen--;
1135 buf[buflen] = '\0';
1136 error = 0;
1137 slash_prefixed = 0;
1138
1139 SDT_PROBE(vfs, namecache, fullpath, entry, vp, 0, 0, 0, 0);
1140 numfullpathcalls++;
1141 CACHE_RLOCK();
1142 if (vp->v_type != VDIR) {
1143 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1144 if (error)
1145 return (error);
1146 if (buflen == 0) {
1147 CACHE_RUNLOCK();
1148 return (ENOMEM);
1149 }
1150 buf[--buflen] = '/';
1151 slash_prefixed = 1;
1152 }
1153 while (vp != rdir && vp != rootvnode) {
1154 if (vp->v_vflag & VV_ROOT) {
1155 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */
1156 CACHE_RUNLOCK();
1157 error = ENOENT;
1158 SDT_PROBE(vfs, namecache, fullpath, return,
1159 error, vp, NULL, 0, 0);
1160 break;
1161 }
1162 vp = vp->v_mount->mnt_vnodecovered;
1163 continue;
1164 }
1165 if (vp->v_type != VDIR) {
1166 CACHE_RUNLOCK();
1167 numfullpathfail1++;
1168 error = ENOTDIR;
1169 SDT_PROBE(vfs, namecache, fullpath, return,
1170 error, vp, NULL, 0, 0);
1171 break;
1172 }
1173 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1174 if (error)
1175 break;
1176 if (buflen == 0) {
1177 CACHE_RUNLOCK();
1178 error = ENOMEM;
1179 SDT_PROBE(vfs, namecache, fullpath, return, error,
1180 startvp, NULL, 0, 0);
1181 break;
1182 }
1183 buf[--buflen] = '/';
1184 slash_prefixed = 1;
1185 }
1186 if (error)
1187 return (error);
1188 if (!slash_prefixed) {
1189 if (buflen == 0) {
1190 CACHE_RUNLOCK();
1191 numfullpathfail4++;
1192 SDT_PROBE(vfs, namecache, fullpath, return, ENOMEM,
1193 startvp, NULL, 0, 0);
1194 return (ENOMEM);
1195 }
1196 buf[--buflen] = '/';
1197 }
1198 numfullpathfound++;
1199 CACHE_RUNLOCK();
1200
1201 SDT_PROBE(vfs, namecache, fullpath, return, 0, startvp, buf + buflen,
1202 0, 0);
1203 *retbuf = buf + buflen;
1204 return (0);
1205 }
1206
1207 int
1208 vn_commname(struct vnode *vp, char *buf, u_int buflen)
1209 {
1210 struct namecache *ncp;
1211 int l;
1212
1213 CACHE_RLOCK();
1214 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1215 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1216 break;
1217 if (ncp == NULL) {
1218 CACHE_RUNLOCK();
1219 return (ENOENT);
1220 }
1221 l = min(ncp->nc_nlen, buflen - 1);
1222 memcpy(buf, ncp->nc_name, l);
1223 CACHE_RUNLOCK();
1224 buf[l] = '\0';
1225 return (0);
1226 }
Cache object: 0eeeec88a0016427e8a82a407ad2804e
|