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 * 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 8.5 (Berkeley) 3/22/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/11.2/sys/kern/vfs_cache.c 331722 2018-03-29 02:50:57Z eadler $");
37
38 #include "opt_ktrace.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/counter.h>
43 #include <sys/filedesc.h>
44 #include <sys/fnv_hash.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/fcntl.h>
49 #include <sys/mount.h>
50 #include <sys/namei.h>
51 #include <sys/proc.h>
52 #include <sys/rwlock.h>
53 #include <sys/sdt.h>
54 #include <sys/smp.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysproto.h>
58 #include <sys/vnode.h>
59 #ifdef KTRACE
60 #include <sys/ktrace.h>
61 #endif
62
63 #include <vm/uma.h>
64
65 SDT_PROVIDER_DECLARE(vfs);
66 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
67 "struct vnode *");
68 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
69 "char *");
70 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *");
71 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *",
72 "char *", "struct vnode *");
73 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *");
74 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int",
75 "struct vnode *", "char *");
76 SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *",
77 "struct vnode *");
78 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative,
79 "struct vnode *", "char *");
80 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
81 "char *");
82 SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *");
83 SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *");
84 SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *");
85 SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *",
86 "struct vnode *");
87 SDT_PROBE_DEFINE3(vfs, namecache, zap_negative, done, "struct vnode *",
88 "char *", "int");
89 SDT_PROBE_DEFINE3(vfs, namecache, shrink_negative, done, "struct vnode *",
90 "char *", "int");
91
92 /*
93 * This structure describes the elements in the cache of recent
94 * names looked up by namei.
95 */
96
97 struct namecache {
98 LIST_ENTRY(namecache) nc_hash; /* hash chain */
99 LIST_ENTRY(namecache) nc_src; /* source vnode list */
100 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
101 struct vnode *nc_dvp; /* vnode of parent of name */
102 union {
103 struct vnode *nu_vp; /* vnode the name refers to */
104 u_int nu_neghits; /* negative entry hits */
105 } n_un;
106 u_char nc_flag; /* flag bits */
107 u_char nc_nlen; /* length of name */
108 char nc_name[0]; /* segment name + nul */
109 };
110
111 /*
112 * struct namecache_ts repeats struct namecache layout up to the
113 * nc_nlen member.
114 * struct namecache_ts is used in place of struct namecache when time(s) need
115 * to be stored. The nc_dotdottime field is used when a cache entry is mapping
116 * both a non-dotdot directory name plus dotdot for the directory's
117 * parent.
118 */
119 struct namecache_ts {
120 struct timespec nc_time; /* timespec provided by fs */
121 struct timespec nc_dotdottime; /* dotdot timespec provided by fs */
122 int nc_ticks; /* ticks value when entry was added */
123 struct namecache nc_nc;
124 };
125
126 #define nc_vp n_un.nu_vp
127 #define nc_neghits n_un.nu_neghits
128
129 /*
130 * Flags in namecache.nc_flag
131 */
132 #define NCF_WHITE 0x01
133 #define NCF_ISDOTDOT 0x02
134 #define NCF_TS 0x04
135 #define NCF_DTS 0x08
136 #define NCF_DVDROP 0x10
137 #define NCF_NEGATIVE 0x20
138 #define NCF_HOTNEGATIVE 0x40
139
140 /*
141 * Name caching works as follows:
142 *
143 * Names found by directory scans are retained in a cache
144 * for future reference. It is managed LRU, so frequently
145 * used names will hang around. Cache is indexed by hash value
146 * obtained from (vp, name) where vp refers to the directory
147 * containing name.
148 *
149 * If it is a "negative" entry, (i.e. for a name that is known NOT to
150 * exist) the vnode pointer will be NULL.
151 *
152 * Upon reaching the last segment of a path, if the reference
153 * is for DELETE, or NOCACHE is set (rewrite), and the
154 * name is located in the cache, it will be dropped.
155 *
156 * These locks are used (in the order in which they can be taken):
157 * NAME TYPE ROLE
158 * vnodelock mtx vnode lists and v_cache_dd field protection
159 * bucketlock rwlock for access to given set of hash buckets
160 * neglist mtx negative entry LRU management
161 *
162 * Additionally, ncneg_shrink_lock mtx is used to have at most one thread
163 * shrinking the LRU list.
164 *
165 * It is legal to take multiple vnodelock and bucketlock locks. The locking
166 * order is lower address first. Both are recursive.
167 *
168 * "." lookups are lockless.
169 *
170 * ".." and vnode -> name lookups require vnodelock.
171 *
172 * name -> vnode lookup requires the relevant bucketlock to be held for reading.
173 *
174 * Insertions and removals of entries require involved vnodes and bucketlocks
175 * to be write-locked to prevent other threads from seeing the entry.
176 *
177 * Some lookups result in removal of the found entry (e.g. getting rid of a
178 * negative entry with the intent to create a positive one), which poses a
179 * problem when multiple threads reach the state. Similarly, two different
180 * threads can purge two different vnodes and try to remove the same name.
181 *
182 * If the already held vnode lock is lower than the second required lock, we
183 * can just take the other lock. However, in the opposite case, this could
184 * deadlock. As such, this is resolved by trylocking and if that fails unlocking
185 * the first node, locking everything in order and revalidating the state.
186 */
187
188 /*
189 * Structures associated with name caching.
190 */
191 #define NCHHASH(hash) \
192 (&nchashtbl[(hash) & nchash])
193 static __read_mostly LIST_HEAD(nchashhead, namecache) *nchashtbl;/* Hash Table */
194 static u_long __read_mostly nchash; /* size of hash table */
195 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0,
196 "Size of namecache hash table");
197 static u_long __read_mostly ncnegfactor = 12; /* ratio of negative entries */
198 SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0,
199 "Ratio of negative namecache entries");
200 static u_long __exclusive_cache_line numneg; /* number of negative entries allocated */
201 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0,
202 "Number of negative entries in namecache");
203 static u_long __exclusive_cache_line numcache;/* number of cache entries allocated */
204 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0,
205 "Number of namecache entries");
206 static u_long __exclusive_cache_line numcachehv;/* number of cache entries with vnodes held */
207 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0,
208 "Number of namecache entries with vnodes held");
209 u_int __read_mostly ncsizefactor = 2;
210 SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0,
211 "Size factor for namecache");
212 static u_int __read_mostly ncpurgeminvnodes;
213 SYSCTL_UINT(_vfs, OID_AUTO, ncpurgeminvnodes, CTLFLAG_RW, &ncpurgeminvnodes, 0,
214 "Number of vnodes below which purgevfs ignores the request");
215 static u_int __read_mostly ncneghitsrequeue = 8;
216 SYSCTL_UINT(_vfs, OID_AUTO, ncneghitsrequeue, CTLFLAG_RW, &ncneghitsrequeue, 0,
217 "Number of hits to requeue a negative entry in the LRU list");
218
219 struct nchstats nchstats; /* cache effectiveness statistics */
220
221 static struct mtx ncneg_shrink_lock;
222 static int shrink_list_turn;
223
224 struct neglist {
225 struct mtx nl_lock;
226 TAILQ_HEAD(, namecache) nl_list;
227 } __aligned(CACHE_LINE_SIZE);
228
229 static struct neglist __read_mostly *neglists;
230 static struct neglist ncneg_hot;
231
232 #define numneglists (ncneghash + 1)
233 static u_int __read_mostly ncneghash;
234 static inline struct neglist *
235 NCP2NEGLIST(struct namecache *ncp)
236 {
237
238 return (&neglists[(((uintptr_t)(ncp) >> 8) & ncneghash)]);
239 }
240
241 #define numbucketlocks (ncbuckethash + 1)
242 static u_int __read_mostly ncbuckethash;
243 static struct rwlock_padalign __read_mostly *bucketlocks;
244 #define HASH2BUCKETLOCK(hash) \
245 ((struct rwlock *)(&bucketlocks[((hash) & ncbuckethash)]))
246
247 #define numvnodelocks (ncvnodehash + 1)
248 static u_int __read_mostly ncvnodehash;
249 static struct mtx __read_mostly *vnodelocks;
250 static inline struct mtx *
251 VP2VNODELOCK(struct vnode *vp)
252 {
253
254 return (&vnodelocks[(((uintptr_t)(vp) >> 8) & ncvnodehash)]);
255 }
256
257 /*
258 * UMA zones for the VFS cache.
259 *
260 * The small cache is used for entries with short names, which are the
261 * most common. The large cache is used for entries which are too big to
262 * fit in the small cache.
263 */
264 static uma_zone_t __read_mostly cache_zone_small;
265 static uma_zone_t __read_mostly cache_zone_small_ts;
266 static uma_zone_t __read_mostly cache_zone_large;
267 static uma_zone_t __read_mostly cache_zone_large_ts;
268
269 #define CACHE_PATH_CUTOFF 35
270
271 static struct namecache *
272 cache_alloc(int len, int ts)
273 {
274 struct namecache_ts *ncp_ts;
275 struct namecache *ncp;
276
277 if (__predict_false(ts)) {
278 if (len <= CACHE_PATH_CUTOFF)
279 ncp_ts = uma_zalloc(cache_zone_small_ts, M_WAITOK);
280 else
281 ncp_ts = uma_zalloc(cache_zone_large_ts, M_WAITOK);
282 ncp = &ncp_ts->nc_nc;
283 } else {
284 if (len <= CACHE_PATH_CUTOFF)
285 ncp = uma_zalloc(cache_zone_small, M_WAITOK);
286 else
287 ncp = uma_zalloc(cache_zone_large, M_WAITOK);
288 }
289 return (ncp);
290 }
291
292 static void
293 cache_free(struct namecache *ncp)
294 {
295 struct namecache_ts *ncp_ts;
296
297 if (ncp == NULL)
298 return;
299 if ((ncp->nc_flag & NCF_DVDROP) != 0)
300 vdrop(ncp->nc_dvp);
301 if (__predict_false(ncp->nc_flag & NCF_TS)) {
302 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc);
303 if (ncp->nc_nlen <= CACHE_PATH_CUTOFF)
304 uma_zfree(cache_zone_small_ts, ncp_ts);
305 else
306 uma_zfree(cache_zone_large_ts, ncp_ts);
307 } else {
308 if (ncp->nc_nlen <= CACHE_PATH_CUTOFF)
309 uma_zfree(cache_zone_small, ncp);
310 else
311 uma_zfree(cache_zone_large, ncp);
312 }
313 }
314
315 static void
316 cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp)
317 {
318 struct namecache_ts *ncp_ts;
319
320 KASSERT((ncp->nc_flag & NCF_TS) != 0 ||
321 (tsp == NULL && ticksp == NULL),
322 ("No NCF_TS"));
323
324 if (tsp == NULL && ticksp == NULL)
325 return;
326
327 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc);
328 if (tsp != NULL)
329 *tsp = ncp_ts->nc_time;
330 if (ticksp != NULL)
331 *ticksp = ncp_ts->nc_ticks;
332 }
333
334 static int __read_mostly doingcache = 1; /* 1 => enable the cache */
335 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
336 "VFS namecache enabled");
337
338 /* Export size information to userland */
339 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR,
340 sizeof(struct namecache), "sizeof(struct namecache)");
341
342 /*
343 * The new name cache statistics
344 */
345 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0,
346 "Name cache statistics");
347 #define STATNODE_ULONG(name, descr) \
348 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr);
349 #define STATNODE_COUNTER(name, descr) \
350 static counter_u64_t __read_mostly name; \
351 SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, descr);
352 STATNODE_ULONG(numneg, "Number of negative cache entries");
353 STATNODE_ULONG(numcache, "Number of cache entries");
354 STATNODE_COUNTER(numcalls, "Number of cache lookups");
355 STATNODE_COUNTER(dothits, "Number of '.' hits");
356 STATNODE_COUNTER(dotdothits, "Number of '..' hits");
357 STATNODE_COUNTER(numchecks, "Number of checks in lookup");
358 STATNODE_COUNTER(nummiss, "Number of cache misses");
359 STATNODE_COUNTER(nummisszap, "Number of cache misses we do not want to cache");
360 STATNODE_COUNTER(numposzaps,
361 "Number of cache hits (positive) we do not want to cache");
362 STATNODE_COUNTER(numposhits, "Number of cache hits (positive)");
363 STATNODE_COUNTER(numnegzaps,
364 "Number of cache hits (negative) we do not want to cache");
365 STATNODE_COUNTER(numneghits, "Number of cache hits (negative)");
366 /* These count for kern___getcwd(), too. */
367 STATNODE_COUNTER(numfullpathcalls, "Number of fullpath search calls");
368 STATNODE_COUNTER(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
369 STATNODE_COUNTER(numfullpathfail2,
370 "Number of fullpath search errors (VOP_VPTOCNP failures)");
371 STATNODE_COUNTER(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
372 STATNODE_COUNTER(numfullpathfound, "Number of successful fullpath calls");
373 static long zap_and_exit_bucket_fail; STATNODE_ULONG(zap_and_exit_bucket_fail,
374 "Number of times zap_and_exit failed to lock");
375 static long cache_lock_vnodes_cel_3_failures;
376 STATNODE_ULONG(cache_lock_vnodes_cel_3_failures,
377 "Number of times 3-way vnode locking failed");
378
379 static void cache_zap_locked(struct namecache *ncp, bool neg_locked);
380 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
381 char *buf, char **retbuf, u_int buflen);
382
383 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
384
385 static int cache_yield;
386 SYSCTL_INT(_vfs_cache, OID_AUTO, yield, CTLFLAG_RD, &cache_yield, 0,
387 "Number of times cache called yield");
388
389 static void
390 cache_maybe_yield(void)
391 {
392
393 if (should_yield()) {
394 cache_yield++;
395 kern_yield(PRI_USER);
396 }
397 }
398
399 static inline void
400 cache_assert_vlp_locked(struct mtx *vlp)
401 {
402
403 if (vlp != NULL)
404 mtx_assert(vlp, MA_OWNED);
405 }
406
407 static inline void
408 cache_assert_vnode_locked(struct vnode *vp)
409 {
410 struct mtx *vlp;
411
412 vlp = VP2VNODELOCK(vp);
413 cache_assert_vlp_locked(vlp);
414 }
415
416 static uint32_t
417 cache_get_hash(char *name, u_char len, struct vnode *dvp)
418 {
419 uint32_t hash;
420
421 hash = fnv_32_buf(name, len, FNV1_32_INIT);
422 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
423 return (hash);
424 }
425
426 static inline struct rwlock *
427 NCP2BUCKETLOCK(struct namecache *ncp)
428 {
429 uint32_t hash;
430
431 hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp);
432 return (HASH2BUCKETLOCK(hash));
433 }
434
435 #ifdef INVARIANTS
436 static void
437 cache_assert_bucket_locked(struct namecache *ncp, int mode)
438 {
439 struct rwlock *blp;
440
441 blp = NCP2BUCKETLOCK(ncp);
442 rw_assert(blp, mode);
443 }
444 #else
445 #define cache_assert_bucket_locked(x, y) do { } while (0)
446 #endif
447
448 #define cache_sort(x, y) _cache_sort((void **)(x), (void **)(y))
449 static void
450 _cache_sort(void **p1, void **p2)
451 {
452 void *tmp;
453
454 if (*p1 > *p2) {
455 tmp = *p2;
456 *p2 = *p1;
457 *p1 = tmp;
458 }
459 }
460
461 static void
462 cache_lock_all_buckets(void)
463 {
464 u_int i;
465
466 for (i = 0; i < numbucketlocks; i++)
467 rw_wlock(&bucketlocks[i]);
468 }
469
470 static void
471 cache_unlock_all_buckets(void)
472 {
473 u_int i;
474
475 for (i = 0; i < numbucketlocks; i++)
476 rw_wunlock(&bucketlocks[i]);
477 }
478
479 static void
480 cache_lock_all_vnodes(void)
481 {
482 u_int i;
483
484 for (i = 0; i < numvnodelocks; i++)
485 mtx_lock(&vnodelocks[i]);
486 }
487
488 static void
489 cache_unlock_all_vnodes(void)
490 {
491 u_int i;
492
493 for (i = 0; i < numvnodelocks; i++)
494 mtx_unlock(&vnodelocks[i]);
495 }
496
497 static int
498 cache_trylock_vnodes(struct mtx *vlp1, struct mtx *vlp2)
499 {
500
501 cache_sort(&vlp1, &vlp2);
502 MPASS(vlp2 != NULL);
503
504 if (vlp1 != NULL) {
505 if (!mtx_trylock(vlp1))
506 return (EAGAIN);
507 }
508 if (!mtx_trylock(vlp2)) {
509 if (vlp1 != NULL)
510 mtx_unlock(vlp1);
511 return (EAGAIN);
512 }
513
514 return (0);
515 }
516
517 static void
518 cache_unlock_vnodes(struct mtx *vlp1, struct mtx *vlp2)
519 {
520
521 MPASS(vlp1 != NULL || vlp2 != NULL);
522
523 if (vlp1 != NULL)
524 mtx_unlock(vlp1);
525 if (vlp2 != NULL)
526 mtx_unlock(vlp2);
527 }
528
529 static int
530 sysctl_nchstats(SYSCTL_HANDLER_ARGS)
531 {
532 struct nchstats snap;
533
534 if (req->oldptr == NULL)
535 return (SYSCTL_OUT(req, 0, sizeof(snap)));
536
537 snap = nchstats;
538 snap.ncs_goodhits = counter_u64_fetch(numposhits);
539 snap.ncs_neghits = counter_u64_fetch(numneghits);
540 snap.ncs_badhits = counter_u64_fetch(numposzaps) +
541 counter_u64_fetch(numnegzaps);
542 snap.ncs_miss = counter_u64_fetch(nummisszap) +
543 counter_u64_fetch(nummiss);
544
545 return (SYSCTL_OUT(req, &snap, sizeof(snap)));
546 }
547 SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD |
548 CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU",
549 "VFS cache effectiveness statistics");
550
551 #ifdef DIAGNOSTIC
552 /*
553 * Grab an atomic snapshot of the name cache hash chain lengths
554 */
555 static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
556 "hash table stats");
557
558 static int
559 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
560 {
561 struct nchashhead *ncpp;
562 struct namecache *ncp;
563 int i, error, n_nchash, *cntbuf;
564
565 retry:
566 n_nchash = nchash + 1; /* nchash is max index, not count */
567 if (req->oldptr == NULL)
568 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
569 cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK);
570 cache_lock_all_buckets();
571 if (n_nchash != nchash + 1) {
572 cache_unlock_all_buckets();
573 free(cntbuf, M_TEMP);
574 goto retry;
575 }
576 /* Scan hash tables counting entries */
577 for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++)
578 LIST_FOREACH(ncp, ncpp, nc_hash)
579 cntbuf[i]++;
580 cache_unlock_all_buckets();
581 for (error = 0, i = 0; i < n_nchash; i++)
582 if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0)
583 break;
584 free(cntbuf, M_TEMP);
585 return (error);
586 }
587 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
588 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
589 "nchash chain lengths");
590
591 static int
592 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
593 {
594 int error;
595 struct nchashhead *ncpp;
596 struct namecache *ncp;
597 int n_nchash;
598 int count, maxlength, used, pct;
599
600 if (!req->oldptr)
601 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
602
603 cache_lock_all_buckets();
604 n_nchash = nchash + 1; /* nchash is max index, not count */
605 used = 0;
606 maxlength = 0;
607
608 /* Scan hash tables for applicable entries */
609 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
610 count = 0;
611 LIST_FOREACH(ncp, ncpp, nc_hash) {
612 count++;
613 }
614 if (count)
615 used++;
616 if (maxlength < count)
617 maxlength = count;
618 }
619 n_nchash = nchash + 1;
620 cache_unlock_all_buckets();
621 pct = (used * 100) / (n_nchash / 100);
622 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
623 if (error)
624 return (error);
625 error = SYSCTL_OUT(req, &used, sizeof(used));
626 if (error)
627 return (error);
628 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
629 if (error)
630 return (error);
631 error = SYSCTL_OUT(req, &pct, sizeof(pct));
632 if (error)
633 return (error);
634 return (0);
635 }
636 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
637 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
638 "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)");
639 #endif
640
641 /*
642 * Negative entries management
643 *
644 * A variation of LRU scheme is used. New entries are hashed into one of
645 * numneglists cold lists. Entries get promoted to the hot list on first hit.
646 * Partial LRU for the hot list is maintained by requeueing them every
647 * ncneghitsrequeue hits.
648 *
649 * The shrinker will demote hot list head and evict from the cold list in a
650 * round-robin manner.
651 */
652 static void
653 cache_negative_hit(struct namecache *ncp)
654 {
655 struct neglist *neglist;
656 u_int hits;
657
658 MPASS(ncp->nc_flag & NCF_NEGATIVE);
659 hits = atomic_fetchadd_int(&ncp->nc_neghits, 1);
660 if (ncp->nc_flag & NCF_HOTNEGATIVE) {
661 if ((hits % ncneghitsrequeue) != 0)
662 return;
663 mtx_lock(&ncneg_hot.nl_lock);
664 if (ncp->nc_flag & NCF_HOTNEGATIVE) {
665 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst);
666 TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst);
667 mtx_unlock(&ncneg_hot.nl_lock);
668 return;
669 }
670 /*
671 * The shrinker cleared the flag and removed the entry from
672 * the hot list. Put it back.
673 */
674 } else {
675 mtx_lock(&ncneg_hot.nl_lock);
676 }
677 neglist = NCP2NEGLIST(ncp);
678 mtx_lock(&neglist->nl_lock);
679 if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) {
680 TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst);
681 TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst);
682 ncp->nc_flag |= NCF_HOTNEGATIVE;
683 }
684 mtx_unlock(&neglist->nl_lock);
685 mtx_unlock(&ncneg_hot.nl_lock);
686 }
687
688 static void
689 cache_negative_insert(struct namecache *ncp, bool neg_locked)
690 {
691 struct neglist *neglist;
692
693 MPASS(ncp->nc_flag & NCF_NEGATIVE);
694 cache_assert_bucket_locked(ncp, RA_WLOCKED);
695 neglist = NCP2NEGLIST(ncp);
696 if (!neg_locked) {
697 mtx_lock(&neglist->nl_lock);
698 } else {
699 mtx_assert(&neglist->nl_lock, MA_OWNED);
700 }
701 TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst);
702 if (!neg_locked)
703 mtx_unlock(&neglist->nl_lock);
704 atomic_add_rel_long(&numneg, 1);
705 }
706
707 static void
708 cache_negative_remove(struct namecache *ncp, bool neg_locked)
709 {
710 struct neglist *neglist;
711 bool hot_locked = false;
712 bool list_locked = false;
713
714 MPASS(ncp->nc_flag & NCF_NEGATIVE);
715 cache_assert_bucket_locked(ncp, RA_WLOCKED);
716 neglist = NCP2NEGLIST(ncp);
717 if (!neg_locked) {
718 if (ncp->nc_flag & NCF_HOTNEGATIVE) {
719 hot_locked = true;
720 mtx_lock(&ncneg_hot.nl_lock);
721 if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) {
722 list_locked = true;
723 mtx_lock(&neglist->nl_lock);
724 }
725 } else {
726 list_locked = true;
727 mtx_lock(&neglist->nl_lock);
728 }
729 }
730 if (ncp->nc_flag & NCF_HOTNEGATIVE) {
731 mtx_assert(&ncneg_hot.nl_lock, MA_OWNED);
732 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst);
733 } else {
734 mtx_assert(&neglist->nl_lock, MA_OWNED);
735 TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst);
736 }
737 if (list_locked)
738 mtx_unlock(&neglist->nl_lock);
739 if (hot_locked)
740 mtx_unlock(&ncneg_hot.nl_lock);
741 atomic_subtract_rel_long(&numneg, 1);
742 }
743
744 static void
745 cache_negative_shrink_select(int start, struct namecache **ncpp,
746 struct neglist **neglistpp)
747 {
748 struct neglist *neglist;
749 struct namecache *ncp;
750 int i;
751
752 *ncpp = ncp = NULL;
753
754 for (i = start; i < numneglists; i++) {
755 neglist = &neglists[i];
756 if (TAILQ_FIRST(&neglist->nl_list) == NULL)
757 continue;
758 mtx_lock(&neglist->nl_lock);
759 ncp = TAILQ_FIRST(&neglist->nl_list);
760 if (ncp != NULL)
761 break;
762 mtx_unlock(&neglist->nl_lock);
763 }
764
765 *neglistpp = neglist;
766 *ncpp = ncp;
767 }
768
769 static void
770 cache_negative_zap_one(void)
771 {
772 struct namecache *ncp, *ncp2;
773 struct neglist *neglist;
774 struct mtx *dvlp;
775 struct rwlock *blp;
776
777 if (!mtx_trylock(&ncneg_shrink_lock))
778 return;
779
780 mtx_lock(&ncneg_hot.nl_lock);
781 ncp = TAILQ_FIRST(&ncneg_hot.nl_list);
782 if (ncp != NULL) {
783 neglist = NCP2NEGLIST(ncp);
784 mtx_lock(&neglist->nl_lock);
785 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst);
786 TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst);
787 ncp->nc_flag &= ~NCF_HOTNEGATIVE;
788 mtx_unlock(&neglist->nl_lock);
789 }
790
791 cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist);
792 shrink_list_turn++;
793 if (shrink_list_turn == numneglists)
794 shrink_list_turn = 0;
795 if (ncp == NULL && shrink_list_turn == 0)
796 cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist);
797 if (ncp == NULL) {
798 mtx_unlock(&ncneg_hot.nl_lock);
799 goto out;
800 }
801
802 MPASS(ncp->nc_flag & NCF_NEGATIVE);
803 dvlp = VP2VNODELOCK(ncp->nc_dvp);
804 blp = NCP2BUCKETLOCK(ncp);
805 mtx_unlock(&neglist->nl_lock);
806 mtx_unlock(&ncneg_hot.nl_lock);
807 mtx_lock(dvlp);
808 rw_wlock(blp);
809 mtx_lock(&neglist->nl_lock);
810 ncp2 = TAILQ_FIRST(&neglist->nl_list);
811 if (ncp != ncp2 || dvlp != VP2VNODELOCK(ncp2->nc_dvp) ||
812 blp != NCP2BUCKETLOCK(ncp2) || !(ncp2->nc_flag & NCF_NEGATIVE)) {
813 ncp = NULL;
814 goto out_unlock_all;
815 }
816 SDT_PROBE3(vfs, namecache, shrink_negative, done, ncp->nc_dvp,
817 ncp->nc_name, ncp->nc_neghits);
818
819 cache_zap_locked(ncp, true);
820 out_unlock_all:
821 mtx_unlock(&neglist->nl_lock);
822 rw_wunlock(blp);
823 mtx_unlock(dvlp);
824 out:
825 mtx_unlock(&ncneg_shrink_lock);
826 cache_free(ncp);
827 }
828
829 /*
830 * cache_zap_locked():
831 *
832 * Removes a namecache entry from cache, whether it contains an actual
833 * pointer to a vnode or if it is just a negative cache entry.
834 */
835 static void
836 cache_zap_locked(struct namecache *ncp, bool neg_locked)
837 {
838
839 if (!(ncp->nc_flag & NCF_NEGATIVE))
840 cache_assert_vnode_locked(ncp->nc_vp);
841 cache_assert_vnode_locked(ncp->nc_dvp);
842 cache_assert_bucket_locked(ncp, RA_WLOCKED);
843
844 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp,
845 (ncp->nc_flag & NCF_NEGATIVE) ? NULL : ncp->nc_vp);
846 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
847 SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp,
848 ncp->nc_name, ncp->nc_vp);
849 } else {
850 SDT_PROBE3(vfs, namecache, zap_negative, done, ncp->nc_dvp,
851 ncp->nc_name, ncp->nc_neghits);
852 }
853 LIST_REMOVE(ncp, nc_hash);
854 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
855 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
856 if (ncp == ncp->nc_vp->v_cache_dd)
857 ncp->nc_vp->v_cache_dd = NULL;
858 } else {
859 cache_negative_remove(ncp, neg_locked);
860 }
861 if (ncp->nc_flag & NCF_ISDOTDOT) {
862 if (ncp == ncp->nc_dvp->v_cache_dd)
863 ncp->nc_dvp->v_cache_dd = NULL;
864 } else {
865 LIST_REMOVE(ncp, nc_src);
866 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
867 ncp->nc_flag |= NCF_DVDROP;
868 atomic_subtract_rel_long(&numcachehv, 1);
869 }
870 }
871 atomic_subtract_rel_long(&numcache, 1);
872 }
873
874 static void
875 cache_zap_negative_locked_vnode_kl(struct namecache *ncp, struct vnode *vp)
876 {
877 struct rwlock *blp;
878
879 MPASS(ncp->nc_dvp == vp);
880 MPASS(ncp->nc_flag & NCF_NEGATIVE);
881 cache_assert_vnode_locked(vp);
882
883 blp = NCP2BUCKETLOCK(ncp);
884 rw_wlock(blp);
885 cache_zap_locked(ncp, false);
886 rw_wunlock(blp);
887 }
888
889 static bool
890 cache_zap_locked_vnode_kl2(struct namecache *ncp, struct vnode *vp,
891 struct mtx **vlpp)
892 {
893 struct mtx *pvlp, *vlp1, *vlp2, *to_unlock;
894 struct rwlock *blp;
895
896 MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp);
897 cache_assert_vnode_locked(vp);
898
899 if (ncp->nc_flag & NCF_NEGATIVE) {
900 if (*vlpp != NULL) {
901 mtx_unlock(*vlpp);
902 *vlpp = NULL;
903 }
904 cache_zap_negative_locked_vnode_kl(ncp, vp);
905 return (true);
906 }
907
908 pvlp = VP2VNODELOCK(vp);
909 blp = NCP2BUCKETLOCK(ncp);
910 vlp1 = VP2VNODELOCK(ncp->nc_dvp);
911 vlp2 = VP2VNODELOCK(ncp->nc_vp);
912
913 if (*vlpp == vlp1 || *vlpp == vlp2) {
914 to_unlock = *vlpp;
915 *vlpp = NULL;
916 } else {
917 if (*vlpp != NULL) {
918 mtx_unlock(*vlpp);
919 *vlpp = NULL;
920 }
921 cache_sort(&vlp1, &vlp2);
922 if (vlp1 == pvlp) {
923 mtx_lock(vlp2);
924 to_unlock = vlp2;
925 } else {
926 if (!mtx_trylock(vlp1))
927 goto out_relock;
928 to_unlock = vlp1;
929 }
930 }
931 rw_wlock(blp);
932 cache_zap_locked(ncp, false);
933 rw_wunlock(blp);
934 if (to_unlock != NULL)
935 mtx_unlock(to_unlock);
936 return (true);
937
938 out_relock:
939 mtx_unlock(vlp2);
940 mtx_lock(vlp1);
941 mtx_lock(vlp2);
942 MPASS(*vlpp == NULL);
943 *vlpp = vlp1;
944 return (false);
945 }
946
947 static int
948 cache_zap_locked_vnode(struct namecache *ncp, struct vnode *vp)
949 {
950 struct mtx *pvlp, *vlp1, *vlp2, *to_unlock;
951 struct rwlock *blp;
952 int error = 0;
953
954 MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp);
955 cache_assert_vnode_locked(vp);
956
957 pvlp = VP2VNODELOCK(vp);
958 if (ncp->nc_flag & NCF_NEGATIVE) {
959 cache_zap_negative_locked_vnode_kl(ncp, vp);
960 goto out;
961 }
962
963 blp = NCP2BUCKETLOCK(ncp);
964 vlp1 = VP2VNODELOCK(ncp->nc_dvp);
965 vlp2 = VP2VNODELOCK(ncp->nc_vp);
966 cache_sort(&vlp1, &vlp2);
967 if (vlp1 == pvlp) {
968 mtx_lock(vlp2);
969 to_unlock = vlp2;
970 } else {
971 if (!mtx_trylock(vlp1)) {
972 error = EAGAIN;
973 goto out;
974 }
975 to_unlock = vlp1;
976 }
977 rw_wlock(blp);
978 cache_zap_locked(ncp, false);
979 rw_wunlock(blp);
980 mtx_unlock(to_unlock);
981 out:
982 mtx_unlock(pvlp);
983 return (error);
984 }
985
986 static int
987 cache_zap_wlocked_bucket(struct namecache *ncp, struct rwlock *blp)
988 {
989 struct mtx *dvlp, *vlp;
990
991 cache_assert_bucket_locked(ncp, RA_WLOCKED);
992
993 dvlp = VP2VNODELOCK(ncp->nc_dvp);
994 vlp = NULL;
995 if (!(ncp->nc_flag & NCF_NEGATIVE))
996 vlp = VP2VNODELOCK(ncp->nc_vp);
997 if (cache_trylock_vnodes(dvlp, vlp) == 0) {
998 cache_zap_locked(ncp, false);
999 rw_wunlock(blp);
1000 cache_unlock_vnodes(dvlp, vlp);
1001 return (0);
1002 }
1003
1004 rw_wunlock(blp);
1005 return (EAGAIN);
1006 }
1007
1008 static int
1009 cache_zap_rlocked_bucket(struct namecache *ncp, struct rwlock *blp)
1010 {
1011 struct mtx *dvlp, *vlp;
1012
1013 cache_assert_bucket_locked(ncp, RA_RLOCKED);
1014
1015 dvlp = VP2VNODELOCK(ncp->nc_dvp);
1016 vlp = NULL;
1017 if (!(ncp->nc_flag & NCF_NEGATIVE))
1018 vlp = VP2VNODELOCK(ncp->nc_vp);
1019 if (cache_trylock_vnodes(dvlp, vlp) == 0) {
1020 rw_runlock(blp);
1021 rw_wlock(blp);
1022 cache_zap_locked(ncp, false);
1023 rw_wunlock(blp);
1024 cache_unlock_vnodes(dvlp, vlp);
1025 return (0);
1026 }
1027
1028 rw_runlock(blp);
1029 return (EAGAIN);
1030 }
1031
1032 static int
1033 cache_zap_wlocked_bucket_kl(struct namecache *ncp, struct rwlock *blp,
1034 struct mtx **vlpp1, struct mtx **vlpp2)
1035 {
1036 struct mtx *dvlp, *vlp;
1037
1038 cache_assert_bucket_locked(ncp, RA_WLOCKED);
1039
1040 dvlp = VP2VNODELOCK(ncp->nc_dvp);
1041 vlp = NULL;
1042 if (!(ncp->nc_flag & NCF_NEGATIVE))
1043 vlp = VP2VNODELOCK(ncp->nc_vp);
1044 cache_sort(&dvlp, &vlp);
1045
1046 if (*vlpp1 == dvlp && *vlpp2 == vlp) {
1047 cache_zap_locked(ncp, false);
1048 cache_unlock_vnodes(dvlp, vlp);
1049 *vlpp1 = NULL;
1050 *vlpp2 = NULL;
1051 return (0);
1052 }
1053
1054 if (*vlpp1 != NULL)
1055 mtx_unlock(*vlpp1);
1056 if (*vlpp2 != NULL)
1057 mtx_unlock(*vlpp2);
1058 *vlpp1 = NULL;
1059 *vlpp2 = NULL;
1060
1061 if (cache_trylock_vnodes(dvlp, vlp) == 0) {
1062 cache_zap_locked(ncp, false);
1063 cache_unlock_vnodes(dvlp, vlp);
1064 return (0);
1065 }
1066
1067 rw_wunlock(blp);
1068 *vlpp1 = dvlp;
1069 *vlpp2 = vlp;
1070 if (*vlpp1 != NULL)
1071 mtx_lock(*vlpp1);
1072 mtx_lock(*vlpp2);
1073 rw_wlock(blp);
1074 return (EAGAIN);
1075 }
1076
1077 static void
1078 cache_lookup_unlock(struct rwlock *blp, struct mtx *vlp)
1079 {
1080
1081 if (blp != NULL) {
1082 rw_runlock(blp);
1083 } else {
1084 mtx_unlock(vlp);
1085 }
1086 }
1087
1088 static int __noinline
1089 cache_lookup_dot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1090 struct timespec *tsp, int *ticksp)
1091 {
1092 int ltype;
1093
1094 *vpp = dvp;
1095 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
1096 dvp, cnp->cn_nameptr);
1097 counter_u64_add(dothits, 1);
1098 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp);
1099 if (tsp != NULL)
1100 timespecclear(tsp);
1101 if (ticksp != NULL)
1102 *ticksp = ticks;
1103 vrefact(*vpp);
1104 /*
1105 * When we lookup "." we still can be asked to lock it
1106 * differently...
1107 */
1108 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
1109 if (ltype != VOP_ISLOCKED(*vpp)) {
1110 if (ltype == LK_EXCLUSIVE) {
1111 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
1112 if ((*vpp)->v_iflag & VI_DOOMED) {
1113 /* forced unmount */
1114 vrele(*vpp);
1115 *vpp = NULL;
1116 return (ENOENT);
1117 }
1118 } else
1119 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
1120 }
1121 return (-1);
1122 }
1123
1124 /*
1125 * Lookup an entry in the cache
1126 *
1127 * Lookup is called with dvp pointing to the directory to search,
1128 * cnp pointing to the name of the entry being sought. If the lookup
1129 * succeeds, the vnode is returned in *vpp, and a status of -1 is
1130 * returned. If the lookup determines that the name does not exist
1131 * (negative caching), a status of ENOENT is returned. If the lookup
1132 * fails, a status of zero is returned. If the directory vnode is
1133 * recycled out from under us due to a forced unmount, a status of
1134 * ENOENT is returned.
1135 *
1136 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
1137 * unlocked. If we're looking up . an extra ref is taken, but the lock is
1138 * not recursively acquired.
1139 */
1140
1141 static __noinline int
1142 cache_lookup_nomakeentry(struct vnode *dvp, struct vnode **vpp,
1143 struct componentname *cnp, struct timespec *tsp, int *ticksp)
1144 {
1145 struct namecache *ncp;
1146 struct rwlock *blp;
1147 struct mtx *dvlp, *dvlp2;
1148 uint32_t hash;
1149 int error;
1150
1151 if (cnp->cn_namelen == 2 &&
1152 cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') {
1153 counter_u64_add(dotdothits, 1);
1154 dvlp = VP2VNODELOCK(dvp);
1155 dvlp2 = NULL;
1156 mtx_lock(dvlp);
1157 retry_dotdot:
1158 ncp = dvp->v_cache_dd;
1159 if (ncp == NULL) {
1160 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
1161 "..", NULL);
1162 mtx_unlock(dvlp);
1163 if (dvlp2 != NULL)
1164 mtx_unlock(dvlp2);
1165 return (0);
1166 }
1167 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) {
1168 if (ncp->nc_dvp != dvp)
1169 panic("dvp %p v_cache_dd %p\n", dvp, ncp);
1170 if (!cache_zap_locked_vnode_kl2(ncp,
1171 dvp, &dvlp2))
1172 goto retry_dotdot;
1173 MPASS(dvp->v_cache_dd == NULL);
1174 mtx_unlock(dvlp);
1175 if (dvlp2 != NULL)
1176 mtx_unlock(dvlp2);
1177 cache_free(ncp);
1178 } else {
1179 dvp->v_cache_dd = NULL;
1180 mtx_unlock(dvlp);
1181 if (dvlp2 != NULL)
1182 mtx_unlock(dvlp2);
1183 }
1184 return (0);
1185 }
1186
1187 hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp);
1188 blp = HASH2BUCKETLOCK(hash);
1189 retry:
1190 if (LIST_EMPTY(NCHHASH(hash)))
1191 goto out_no_entry;
1192
1193 rw_wlock(blp);
1194
1195 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
1196 counter_u64_add(numchecks, 1);
1197 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
1198 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
1199 break;
1200 }
1201
1202 /* We failed to find an entry */
1203 if (ncp == NULL) {
1204 rw_wunlock(blp);
1205 goto out_no_entry;
1206 }
1207
1208 counter_u64_add(numposzaps, 1);
1209
1210 error = cache_zap_wlocked_bucket(ncp, blp);
1211 if (error != 0) {
1212 zap_and_exit_bucket_fail++;
1213 cache_maybe_yield();
1214 goto retry;
1215 }
1216 cache_free(ncp);
1217 return (0);
1218 out_no_entry:
1219 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL);
1220 counter_u64_add(nummisszap, 1);
1221 return (0);
1222 }
1223
1224 int
1225 cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1226 struct timespec *tsp, int *ticksp)
1227 {
1228 struct namecache_ts *ncp_ts;
1229 struct namecache *ncp;
1230 struct rwlock *blp;
1231 struct mtx *dvlp, *dvlp2;
1232 uint32_t hash;
1233 int error, ltype;
1234
1235 if (__predict_false(!doingcache)) {
1236 cnp->cn_flags &= ~MAKEENTRY;
1237 return (0);
1238 }
1239
1240 counter_u64_add(numcalls, 1);
1241
1242 if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.'))
1243 return (cache_lookup_dot(dvp, vpp, cnp, tsp, ticksp));
1244
1245 if ((cnp->cn_flags & MAKEENTRY) == 0)
1246 return (cache_lookup_nomakeentry(dvp, vpp, cnp, tsp, ticksp));
1247
1248 retry:
1249 blp = NULL;
1250 error = 0;
1251 if (cnp->cn_namelen == 2 &&
1252 cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') {
1253 counter_u64_add(dotdothits, 1);
1254 dvlp = VP2VNODELOCK(dvp);
1255 dvlp2 = NULL;
1256 mtx_lock(dvlp);
1257 ncp = dvp->v_cache_dd;
1258 if (ncp == NULL) {
1259 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
1260 "..", NULL);
1261 mtx_unlock(dvlp);
1262 return (0);
1263 }
1264 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) {
1265 if (ncp->nc_flag & NCF_NEGATIVE)
1266 *vpp = NULL;
1267 else
1268 *vpp = ncp->nc_vp;
1269 } else
1270 *vpp = ncp->nc_dvp;
1271 /* Return failure if negative entry was found. */
1272 if (*vpp == NULL)
1273 goto negative_success;
1274 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
1275 dvp, cnp->cn_nameptr, *vpp);
1276 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..",
1277 *vpp);
1278 cache_out_ts(ncp, tsp, ticksp);
1279 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
1280 NCF_DTS && tsp != NULL) {
1281 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc);
1282 *tsp = ncp_ts->nc_dotdottime;
1283 }
1284 goto success;
1285 }
1286
1287 hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp);
1288 blp = HASH2BUCKETLOCK(hash);
1289 rw_rlock(blp);
1290
1291 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
1292 counter_u64_add(numchecks, 1);
1293 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
1294 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
1295 break;
1296 }
1297
1298 /* We failed to find an entry */
1299 if (ncp == NULL) {
1300 rw_runlock(blp);
1301 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
1302 NULL);
1303 counter_u64_add(nummiss, 1);
1304 return (0);
1305 }
1306
1307 /* We found a "positive" match, return the vnode */
1308 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
1309 counter_u64_add(numposhits, 1);
1310 *vpp = ncp->nc_vp;
1311 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
1312 dvp, cnp->cn_nameptr, *vpp, ncp);
1313 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name,
1314 *vpp);
1315 cache_out_ts(ncp, tsp, ticksp);
1316 goto success;
1317 }
1318
1319 negative_success:
1320 /* We found a negative match, and want to create it, so purge */
1321 if (cnp->cn_nameiop == CREATE) {
1322 counter_u64_add(numnegzaps, 1);
1323 goto zap_and_exit;
1324 }
1325
1326 counter_u64_add(numneghits, 1);
1327 cache_negative_hit(ncp);
1328 if (ncp->nc_flag & NCF_WHITE)
1329 cnp->cn_flags |= ISWHITEOUT;
1330 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp,
1331 ncp->nc_name);
1332 cache_out_ts(ncp, tsp, ticksp);
1333 cache_lookup_unlock(blp, dvlp);
1334 return (ENOENT);
1335
1336 success:
1337 /*
1338 * On success we return a locked and ref'd vnode as per the lookup
1339 * protocol.
1340 */
1341 MPASS(dvp != *vpp);
1342 ltype = 0; /* silence gcc warning */
1343 if (cnp->cn_flags & ISDOTDOT) {
1344 ltype = VOP_ISLOCKED(dvp);
1345 VOP_UNLOCK(dvp, 0);
1346 }
1347 vhold(*vpp);
1348 cache_lookup_unlock(blp, dvlp);
1349 error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread);
1350 if (cnp->cn_flags & ISDOTDOT) {
1351 vn_lock(dvp, ltype | LK_RETRY);
1352 if (dvp->v_iflag & VI_DOOMED) {
1353 if (error == 0)
1354 vput(*vpp);
1355 *vpp = NULL;
1356 return (ENOENT);
1357 }
1358 }
1359 if (error) {
1360 *vpp = NULL;
1361 goto retry;
1362 }
1363 if ((cnp->cn_flags & ISLASTCN) &&
1364 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
1365 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
1366 }
1367 return (-1);
1368
1369 zap_and_exit:
1370 if (blp != NULL)
1371 error = cache_zap_rlocked_bucket(ncp, blp);
1372 else
1373 error = cache_zap_locked_vnode(ncp, dvp);
1374 if (error != 0) {
1375 zap_and_exit_bucket_fail++;
1376 cache_maybe_yield();
1377 goto retry;
1378 }
1379 cache_free(ncp);
1380 return (0);
1381 }
1382
1383 struct celockstate {
1384 struct mtx *vlp[3];
1385 struct rwlock *blp[2];
1386 };
1387 CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3));
1388 CTASSERT((nitems(((struct celockstate *)0)->blp) == 2));
1389
1390 static inline void
1391 cache_celockstate_init(struct celockstate *cel)
1392 {
1393
1394 bzero(cel, sizeof(*cel));
1395 }
1396
1397 static void
1398 cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp,
1399 struct vnode *dvp)
1400 {
1401 struct mtx *vlp1, *vlp2;
1402
1403 MPASS(cel->vlp[0] == NULL);
1404 MPASS(cel->vlp[1] == NULL);
1405 MPASS(cel->vlp[2] == NULL);
1406
1407 MPASS(vp != NULL || dvp != NULL);
1408
1409 vlp1 = VP2VNODELOCK(vp);
1410 vlp2 = VP2VNODELOCK(dvp);
1411 cache_sort(&vlp1, &vlp2);
1412
1413 if (vlp1 != NULL) {
1414 mtx_lock(vlp1);
1415 cel->vlp[0] = vlp1;
1416 }
1417 mtx_lock(vlp2);
1418 cel->vlp[1] = vlp2;
1419 }
1420
1421 static void
1422 cache_unlock_vnodes_cel(struct celockstate *cel)
1423 {
1424
1425 MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL);
1426
1427 if (cel->vlp[0] != NULL)
1428 mtx_unlock(cel->vlp[0]);
1429 if (cel->vlp[1] != NULL)
1430 mtx_unlock(cel->vlp[1]);
1431 if (cel->vlp[2] != NULL)
1432 mtx_unlock(cel->vlp[2]);
1433 }
1434
1435 static bool
1436 cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp)
1437 {
1438 struct mtx *vlp;
1439 bool ret;
1440
1441 cache_assert_vlp_locked(cel->vlp[0]);
1442 cache_assert_vlp_locked(cel->vlp[1]);
1443 MPASS(cel->vlp[2] == NULL);
1444
1445 MPASS(vp != NULL);
1446 vlp = VP2VNODELOCK(vp);
1447
1448 ret = true;
1449 if (vlp >= cel->vlp[1]) {
1450 mtx_lock(vlp);
1451 } else {
1452 if (mtx_trylock(vlp))
1453 goto out;
1454 cache_lock_vnodes_cel_3_failures++;
1455 cache_unlock_vnodes_cel(cel);
1456 if (vlp < cel->vlp[0]) {
1457 mtx_lock(vlp);
1458 mtx_lock(cel->vlp[0]);
1459 mtx_lock(cel->vlp[1]);
1460 } else {
1461 if (cel->vlp[0] != NULL)
1462 mtx_lock(cel->vlp[0]);
1463 mtx_lock(vlp);
1464 mtx_lock(cel->vlp[1]);
1465 }
1466 ret = false;
1467 }
1468 out:
1469 cel->vlp[2] = vlp;
1470 return (ret);
1471 }
1472
1473 static void
1474 cache_lock_buckets_cel(struct celockstate *cel, struct rwlock *blp1,
1475 struct rwlock *blp2)
1476 {
1477
1478 MPASS(cel->blp[0] == NULL);
1479 MPASS(cel->blp[1] == NULL);
1480
1481 cache_sort(&blp1, &blp2);
1482
1483 if (blp1 != NULL) {
1484 rw_wlock(blp1);
1485 cel->blp[0] = blp1;
1486 }
1487 rw_wlock(blp2);
1488 cel->blp[1] = blp2;
1489 }
1490
1491 static void
1492 cache_unlock_buckets_cel(struct celockstate *cel)
1493 {
1494
1495 if (cel->blp[0] != NULL)
1496 rw_wunlock(cel->blp[0]);
1497 rw_wunlock(cel->blp[1]);
1498 }
1499
1500 /*
1501 * Lock part of the cache affected by the insertion.
1502 *
1503 * This means vnodelocks for dvp, vp and the relevant bucketlock.
1504 * However, insertion can result in removal of an old entry. In this
1505 * case we have an additional vnode and bucketlock pair to lock. If the
1506 * entry is negative, ncelock is locked instead of the vnode.
1507 *
1508 * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while
1509 * preserving the locking order (smaller address first).
1510 */
1511 static void
1512 cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp,
1513 uint32_t hash)
1514 {
1515 struct namecache *ncp;
1516 struct rwlock *blps[2];
1517
1518 blps[0] = HASH2BUCKETLOCK(hash);
1519 for (;;) {
1520 blps[1] = NULL;
1521 cache_lock_vnodes_cel(cel, dvp, vp);
1522 if (vp == NULL || vp->v_type != VDIR)
1523 break;
1524 ncp = vp->v_cache_dd;
1525 if (ncp == NULL)
1526 break;
1527 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1528 break;
1529 MPASS(ncp->nc_dvp == vp);
1530 blps[1] = NCP2BUCKETLOCK(ncp);
1531 if (ncp->nc_flag & NCF_NEGATIVE)
1532 break;
1533 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp))
1534 break;
1535 /*
1536 * All vnodes got re-locked. Re-validate the state and if
1537 * nothing changed we are done. Otherwise restart.
1538 */
1539 if (ncp == vp->v_cache_dd &&
1540 (ncp->nc_flag & NCF_ISDOTDOT) != 0 &&
1541 blps[1] == NCP2BUCKETLOCK(ncp) &&
1542 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2])
1543 break;
1544 cache_unlock_vnodes_cel(cel);
1545 cel->vlp[0] = NULL;
1546 cel->vlp[1] = NULL;
1547 cel->vlp[2] = NULL;
1548 }
1549 cache_lock_buckets_cel(cel, blps[0], blps[1]);
1550 }
1551
1552 static void
1553 cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp,
1554 uint32_t hash)
1555 {
1556 struct namecache *ncp;
1557 struct rwlock *blps[2];
1558
1559 blps[0] = HASH2BUCKETLOCK(hash);
1560 for (;;) {
1561 blps[1] = NULL;
1562 cache_lock_vnodes_cel(cel, dvp, vp);
1563 ncp = dvp->v_cache_dd;
1564 if (ncp == NULL)
1565 break;
1566 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1567 break;
1568 MPASS(ncp->nc_dvp == dvp);
1569 blps[1] = NCP2BUCKETLOCK(ncp);
1570 if (ncp->nc_flag & NCF_NEGATIVE)
1571 break;
1572 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp))
1573 break;
1574 if (ncp == dvp->v_cache_dd &&
1575 (ncp->nc_flag & NCF_ISDOTDOT) != 0 &&
1576 blps[1] == NCP2BUCKETLOCK(ncp) &&
1577 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2])
1578 break;
1579 cache_unlock_vnodes_cel(cel);
1580 cel->vlp[0] = NULL;
1581 cel->vlp[1] = NULL;
1582 cel->vlp[2] = NULL;
1583 }
1584 cache_lock_buckets_cel(cel, blps[0], blps[1]);
1585 }
1586
1587 static void
1588 cache_enter_unlock(struct celockstate *cel)
1589 {
1590
1591 cache_unlock_buckets_cel(cel);
1592 cache_unlock_vnodes_cel(cel);
1593 }
1594
1595 /*
1596 * Add an entry to the cache.
1597 */
1598 void
1599 cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp,
1600 struct timespec *tsp, struct timespec *dtsp)
1601 {
1602 struct celockstate cel;
1603 struct namecache *ncp, *n2, *ndd;
1604 struct namecache_ts *ncp_ts, *n2_ts;
1605 struct nchashhead *ncpp;
1606 struct neglist *neglist;
1607 uint32_t hash;
1608 int flag;
1609 int len;
1610 bool neg_locked;
1611 int lnumcache;
1612
1613 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
1614 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
1615 ("cache_enter: Adding a doomed vnode"));
1616 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
1617 ("cache_enter: Doomed vnode used as src"));
1618
1619 if (__predict_false(!doingcache))
1620 return;
1621
1622 /*
1623 * Avoid blowout in namecache entries.
1624 */
1625 if (__predict_false(numcache >= desiredvnodes * ncsizefactor))
1626 return;
1627
1628 cache_celockstate_init(&cel);
1629 ndd = NULL;
1630 flag = 0;
1631 if (cnp->cn_nameptr[0] == '.') {
1632 if (cnp->cn_namelen == 1)
1633 return;
1634 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
1635 len = cnp->cn_namelen;
1636 hash = cache_get_hash(cnp->cn_nameptr, len, dvp);
1637 cache_enter_lock_dd(&cel, dvp, vp, hash);
1638 /*
1639 * If dotdot entry already exists, just retarget it
1640 * to new parent vnode, otherwise continue with new
1641 * namecache entry allocation.
1642 */
1643 if ((ncp = dvp->v_cache_dd) != NULL &&
1644 ncp->nc_flag & NCF_ISDOTDOT) {
1645 KASSERT(ncp->nc_dvp == dvp,
1646 ("wrong isdotdot parent"));
1647 neg_locked = false;
1648 if (ncp->nc_flag & NCF_NEGATIVE || vp == NULL) {
1649 neglist = NCP2NEGLIST(ncp);
1650 mtx_lock(&ncneg_hot.nl_lock);
1651 mtx_lock(&neglist->nl_lock);
1652 neg_locked = true;
1653 }
1654 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
1655 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
1656 ncp, nc_dst);
1657 } else {
1658 cache_negative_remove(ncp, true);
1659 }
1660 if (vp != NULL) {
1661 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
1662 ncp, nc_dst);
1663 ncp->nc_flag &= ~(NCF_NEGATIVE|NCF_HOTNEGATIVE);
1664 } else {
1665 ncp->nc_flag &= ~(NCF_HOTNEGATIVE);
1666 ncp->nc_flag |= NCF_NEGATIVE;
1667 cache_negative_insert(ncp, true);
1668 }
1669 if (neg_locked) {
1670 mtx_unlock(&neglist->nl_lock);
1671 mtx_unlock(&ncneg_hot.nl_lock);
1672 }
1673 ncp->nc_vp = vp;
1674 cache_enter_unlock(&cel);
1675 return;
1676 }
1677 dvp->v_cache_dd = NULL;
1678 cache_enter_unlock(&cel);
1679 cache_celockstate_init(&cel);
1680 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp);
1681 flag = NCF_ISDOTDOT;
1682 }
1683 }
1684
1685 /*
1686 * Calculate the hash key and setup as much of the new
1687 * namecache entry as possible before acquiring the lock.
1688 */
1689 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
1690 ncp->nc_flag = flag;
1691 ncp->nc_vp = vp;
1692 if (vp == NULL)
1693 ncp->nc_flag |= NCF_NEGATIVE;
1694 ncp->nc_dvp = dvp;
1695 if (tsp != NULL) {
1696 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc);
1697 ncp_ts->nc_time = *tsp;
1698 ncp_ts->nc_ticks = ticks;
1699 ncp_ts->nc_nc.nc_flag |= NCF_TS;
1700 if (dtsp != NULL) {
1701 ncp_ts->nc_dotdottime = *dtsp;
1702 ncp_ts->nc_nc.nc_flag |= NCF_DTS;
1703 }
1704 }
1705 len = ncp->nc_nlen = cnp->cn_namelen;
1706 hash = cache_get_hash(cnp->cn_nameptr, len, dvp);
1707 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1);
1708 cache_enter_lock(&cel, dvp, vp, hash);
1709
1710 /*
1711 * See if this vnode or negative entry is already in the cache
1712 * with this name. This can happen with concurrent lookups of
1713 * the same path name.
1714 */
1715 ncpp = NCHHASH(hash);
1716 LIST_FOREACH(n2, ncpp, nc_hash) {
1717 if (n2->nc_dvp == dvp &&
1718 n2->nc_nlen == cnp->cn_namelen &&
1719 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
1720 if (tsp != NULL) {
1721 KASSERT((n2->nc_flag & NCF_TS) != 0,
1722 ("no NCF_TS"));
1723 n2_ts = __containerof(n2, struct namecache_ts, nc_nc);
1724 n2_ts->nc_time = ncp_ts->nc_time;
1725 n2_ts->nc_ticks = ncp_ts->nc_ticks;
1726 if (dtsp != NULL) {
1727 n2_ts->nc_dotdottime = ncp_ts->nc_dotdottime;
1728 if (ncp->nc_flag & NCF_NEGATIVE)
1729 mtx_lock(&ncneg_hot.nl_lock);
1730 n2_ts->nc_nc.nc_flag |= NCF_DTS;
1731 if (ncp->nc_flag & NCF_NEGATIVE)
1732 mtx_unlock(&ncneg_hot.nl_lock);
1733 }
1734 }
1735 goto out_unlock_free;
1736 }
1737 }
1738
1739 if (flag == NCF_ISDOTDOT) {
1740 /*
1741 * See if we are trying to add .. entry, but some other lookup
1742 * has populated v_cache_dd pointer already.
1743 */
1744 if (dvp->v_cache_dd != NULL)
1745 goto out_unlock_free;
1746 KASSERT(vp == NULL || vp->v_type == VDIR,
1747 ("wrong vnode type %p", vp));
1748 dvp->v_cache_dd = ncp;
1749 }
1750
1751 if (vp != NULL) {
1752 if (vp->v_type == VDIR) {
1753 if (flag != NCF_ISDOTDOT) {
1754 /*
1755 * For this case, the cache entry maps both the
1756 * directory name in it and the name ".." for the
1757 * directory's parent.
1758 */
1759 if ((ndd = vp->v_cache_dd) != NULL) {
1760 if ((ndd->nc_flag & NCF_ISDOTDOT) != 0)
1761 cache_zap_locked(ndd, false);
1762 else
1763 ndd = NULL;
1764 }
1765 vp->v_cache_dd = ncp;
1766 }
1767 } else {
1768 vp->v_cache_dd = NULL;
1769 }
1770 }
1771
1772 if (flag != NCF_ISDOTDOT) {
1773 if (LIST_EMPTY(&dvp->v_cache_src)) {
1774 vhold(dvp);
1775 atomic_add_rel_long(&numcachehv, 1);
1776 }
1777 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
1778 }
1779
1780 /*
1781 * Insert the new namecache entry into the appropriate chain
1782 * within the cache entries table.
1783 */
1784 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
1785
1786 /*
1787 * If the entry is "negative", we place it into the
1788 * "negative" cache queue, otherwise, we place it into the
1789 * destination vnode's cache entries queue.
1790 */
1791 if (vp != NULL) {
1792 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
1793 SDT_PROBE3(vfs, namecache, enter, done, dvp, ncp->nc_name,
1794 vp);
1795 } else {
1796 if (cnp->cn_flags & ISWHITEOUT)
1797 ncp->nc_flag |= NCF_WHITE;
1798 cache_negative_insert(ncp, false);
1799 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp,
1800 ncp->nc_name);
1801 }
1802 cache_enter_unlock(&cel);
1803 lnumcache = atomic_fetchadd_long(&numcache, 1) + 1;
1804 if (numneg * ncnegfactor > lnumcache)
1805 cache_negative_zap_one();
1806 cache_free(ndd);
1807 return;
1808 out_unlock_free:
1809 cache_enter_unlock(&cel);
1810 cache_free(ncp);
1811 return;
1812 }
1813
1814 static u_int
1815 cache_roundup_2(u_int val)
1816 {
1817 u_int res;
1818
1819 for (res = 1; res <= val; res <<= 1)
1820 continue;
1821
1822 return (res);
1823 }
1824
1825 /*
1826 * Name cache initialization, from vfs_init() when we are booting
1827 */
1828 static void
1829 nchinit(void *dummy __unused)
1830 {
1831 u_int i;
1832
1833 cache_zone_small = uma_zcreate("S VFS Cache",
1834 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
1835 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache),
1836 UMA_ZONE_ZINIT);
1837 cache_zone_small_ts = uma_zcreate("STS VFS Cache",
1838 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
1839 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts),
1840 UMA_ZONE_ZINIT);
1841 cache_zone_large = uma_zcreate("L VFS Cache",
1842 sizeof(struct namecache) + NAME_MAX + 1,
1843 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache),
1844 UMA_ZONE_ZINIT);
1845 cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
1846 sizeof(struct namecache_ts) + NAME_MAX + 1,
1847 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts),
1848 UMA_ZONE_ZINIT);
1849
1850 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
1851 ncbuckethash = cache_roundup_2(mp_ncpus * 64) - 1;
1852 if (ncbuckethash > nchash)
1853 ncbuckethash = nchash;
1854 bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE,
1855 M_WAITOK | M_ZERO);
1856 for (i = 0; i < numbucketlocks; i++)
1857 rw_init_flags(&bucketlocks[i], "ncbuc", RW_DUPOK | RW_RECURSE);
1858 ncvnodehash = cache_roundup_2(mp_ncpus * 64) - 1;
1859 vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE,
1860 M_WAITOK | M_ZERO);
1861 for (i = 0; i < numvnodelocks; i++)
1862 mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE);
1863 ncpurgeminvnodes = numbucketlocks;
1864
1865 ncneghash = 3;
1866 neglists = malloc(sizeof(*neglists) * numneglists, M_VFSCACHE,
1867 M_WAITOK | M_ZERO);
1868 for (i = 0; i < numneglists; i++) {
1869 mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF);
1870 TAILQ_INIT(&neglists[i].nl_list);
1871 }
1872 mtx_init(&ncneg_hot.nl_lock, "ncneglh", NULL, MTX_DEF);
1873 TAILQ_INIT(&ncneg_hot.nl_list);
1874
1875 mtx_init(&ncneg_shrink_lock, "ncnegs", NULL, MTX_DEF);
1876
1877 numcalls = counter_u64_alloc(M_WAITOK);
1878 dothits = counter_u64_alloc(M_WAITOK);
1879 dotdothits = counter_u64_alloc(M_WAITOK);
1880 numchecks = counter_u64_alloc(M_WAITOK);
1881 nummiss = counter_u64_alloc(M_WAITOK);
1882 nummisszap = counter_u64_alloc(M_WAITOK);
1883 numposzaps = counter_u64_alloc(M_WAITOK);
1884 numposhits = counter_u64_alloc(M_WAITOK);
1885 numnegzaps = counter_u64_alloc(M_WAITOK);
1886 numneghits = counter_u64_alloc(M_WAITOK);
1887 numfullpathcalls = counter_u64_alloc(M_WAITOK);
1888 numfullpathfail1 = counter_u64_alloc(M_WAITOK);
1889 numfullpathfail2 = counter_u64_alloc(M_WAITOK);
1890 numfullpathfail4 = counter_u64_alloc(M_WAITOK);
1891 numfullpathfound = counter_u64_alloc(M_WAITOK);
1892 }
1893 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
1894
1895 void
1896 cache_changesize(int newmaxvnodes)
1897 {
1898 struct nchashhead *new_nchashtbl, *old_nchashtbl;
1899 u_long new_nchash, old_nchash;
1900 struct namecache *ncp;
1901 uint32_t hash;
1902 int i;
1903
1904 newmaxvnodes = cache_roundup_2(newmaxvnodes * 2);
1905 if (newmaxvnodes < numbucketlocks)
1906 newmaxvnodes = numbucketlocks;
1907
1908 new_nchashtbl = hashinit(newmaxvnodes, M_VFSCACHE, &new_nchash);
1909 /* If same hash table size, nothing to do */
1910 if (nchash == new_nchash) {
1911 free(new_nchashtbl, M_VFSCACHE);
1912 return;
1913 }
1914 /*
1915 * Move everything from the old hash table to the new table.
1916 * None of the namecache entries in the table can be removed
1917 * because to do so, they have to be removed from the hash table.
1918 */
1919 cache_lock_all_vnodes();
1920 cache_lock_all_buckets();
1921 old_nchashtbl = nchashtbl;
1922 old_nchash = nchash;
1923 nchashtbl = new_nchashtbl;
1924 nchash = new_nchash;
1925 for (i = 0; i <= old_nchash; i++) {
1926 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) {
1927 hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen,
1928 ncp->nc_dvp);
1929 LIST_REMOVE(ncp, nc_hash);
1930 LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash);
1931 }
1932 }
1933 cache_unlock_all_buckets();
1934 cache_unlock_all_vnodes();
1935 free(old_nchashtbl, M_VFSCACHE);
1936 }
1937
1938 /*
1939 * Invalidate all entries to a particular vnode.
1940 */
1941 void
1942 cache_purge(struct vnode *vp)
1943 {
1944 TAILQ_HEAD(, namecache) ncps;
1945 struct namecache *ncp, *nnp;
1946 struct mtx *vlp, *vlp2;
1947
1948 CTR1(KTR_VFS, "cache_purge(%p)", vp);
1949 SDT_PROBE1(vfs, namecache, purge, done, vp);
1950 if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) &&
1951 vp->v_cache_dd == NULL)
1952 return;
1953 TAILQ_INIT(&ncps);
1954 vlp = VP2VNODELOCK(vp);
1955 vlp2 = NULL;
1956 mtx_lock(vlp);
1957 retry:
1958 while (!LIST_EMPTY(&vp->v_cache_src)) {
1959 ncp = LIST_FIRST(&vp->v_cache_src);
1960 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1961 goto retry;
1962 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1963 }
1964 while (!TAILQ_EMPTY(&vp->v_cache_dst)) {
1965 ncp = TAILQ_FIRST(&vp->v_cache_dst);
1966 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1967 goto retry;
1968 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1969 }
1970 ncp = vp->v_cache_dd;
1971 if (ncp != NULL) {
1972 KASSERT(ncp->nc_flag & NCF_ISDOTDOT,
1973 ("lost dotdot link"));
1974 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1975 goto retry;
1976 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1977 }
1978 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
1979 mtx_unlock(vlp);
1980 if (vlp2 != NULL)
1981 mtx_unlock(vlp2);
1982 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
1983 cache_free(ncp);
1984 }
1985 }
1986
1987 /*
1988 * Invalidate all negative entries for a particular directory vnode.
1989 */
1990 void
1991 cache_purge_negative(struct vnode *vp)
1992 {
1993 TAILQ_HEAD(, namecache) ncps;
1994 struct namecache *ncp, *nnp;
1995 struct mtx *vlp;
1996
1997 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
1998 SDT_PROBE1(vfs, namecache, purge_negative, done, vp);
1999 if (LIST_EMPTY(&vp->v_cache_src))
2000 return;
2001 TAILQ_INIT(&ncps);
2002 vlp = VP2VNODELOCK(vp);
2003 mtx_lock(vlp);
2004 LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) {
2005 if (!(ncp->nc_flag & NCF_NEGATIVE))
2006 continue;
2007 cache_zap_negative_locked_vnode_kl(ncp, vp);
2008 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
2009 }
2010 mtx_unlock(vlp);
2011 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
2012 cache_free(ncp);
2013 }
2014 }
2015
2016 /*
2017 * Flush all entries referencing a particular filesystem.
2018 */
2019 void
2020 cache_purgevfs(struct mount *mp, bool force)
2021 {
2022 TAILQ_HEAD(, namecache) ncps;
2023 struct mtx *vlp1, *vlp2;
2024 struct rwlock *blp;
2025 struct nchashhead *bucket;
2026 struct namecache *ncp, *nnp;
2027 u_long i, j, n_nchash;
2028 int error;
2029
2030 /* Scan hash tables for applicable entries */
2031 SDT_PROBE1(vfs, namecache, purgevfs, done, mp);
2032 if (!force && mp->mnt_nvnodelistsize <= ncpurgeminvnodes)
2033 return;
2034 TAILQ_INIT(&ncps);
2035 n_nchash = nchash + 1;
2036 vlp1 = vlp2 = NULL;
2037 for (i = 0; i < numbucketlocks; i++) {
2038 blp = (struct rwlock *)&bucketlocks[i];
2039 rw_wlock(blp);
2040 for (j = i; j < n_nchash; j += numbucketlocks) {
2041 retry:
2042 bucket = &nchashtbl[j];
2043 LIST_FOREACH_SAFE(ncp, bucket, nc_hash, nnp) {
2044 cache_assert_bucket_locked(ncp, RA_WLOCKED);
2045 if (ncp->nc_dvp->v_mount != mp)
2046 continue;
2047 error = cache_zap_wlocked_bucket_kl(ncp, blp,
2048 &vlp1, &vlp2);
2049 if (error != 0)
2050 goto retry;
2051 TAILQ_INSERT_HEAD(&ncps, ncp, nc_dst);
2052 }
2053 }
2054 rw_wunlock(blp);
2055 if (vlp1 == NULL && vlp2 == NULL)
2056 cache_maybe_yield();
2057 }
2058 if (vlp1 != NULL)
2059 mtx_unlock(vlp1);
2060 if (vlp2 != NULL)
2061 mtx_unlock(vlp2);
2062
2063 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
2064 cache_free(ncp);
2065 }
2066 }
2067
2068 /*
2069 * Perform canonical checks and cache lookup and pass on to filesystem
2070 * through the vop_cachedlookup only if needed.
2071 */
2072
2073 int
2074 vfs_cache_lookup(struct vop_lookup_args *ap)
2075 {
2076 struct vnode *dvp;
2077 int error;
2078 struct vnode **vpp = ap->a_vpp;
2079 struct componentname *cnp = ap->a_cnp;
2080 struct ucred *cred = cnp->cn_cred;
2081 int flags = cnp->cn_flags;
2082 struct thread *td = cnp->cn_thread;
2083
2084 *vpp = NULL;
2085 dvp = ap->a_dvp;
2086
2087 if (dvp->v_type != VDIR)
2088 return (ENOTDIR);
2089
2090 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
2091 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
2092 return (EROFS);
2093
2094 error = VOP_ACCESS(dvp, VEXEC, cred, td);
2095 if (error)
2096 return (error);
2097
2098 error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
2099 if (error == 0)
2100 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
2101 if (error == -1)
2102 return (0);
2103 return (error);
2104 }
2105
2106 /*
2107 * XXX All of these sysctls would probably be more productive dead.
2108 */
2109 static int __read_mostly disablecwd;
2110 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
2111 "Disable the getcwd syscall");
2112
2113 /* Implementation of the getcwd syscall. */
2114 int
2115 sys___getcwd(struct thread *td, struct __getcwd_args *uap)
2116 {
2117
2118 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen,
2119 MAXPATHLEN));
2120 }
2121
2122 int
2123 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen,
2124 u_int path_max)
2125 {
2126 char *bp, *tmpbuf;
2127 struct filedesc *fdp;
2128 struct vnode *cdir, *rdir;
2129 int error;
2130
2131 if (__predict_false(disablecwd))
2132 return (ENODEV);
2133 if (__predict_false(buflen < 2))
2134 return (EINVAL);
2135 if (buflen > path_max)
2136 buflen = path_max;
2137
2138 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
2139 fdp = td->td_proc->p_fd;
2140 FILEDESC_SLOCK(fdp);
2141 cdir = fdp->fd_cdir;
2142 vrefact(cdir);
2143 rdir = fdp->fd_rdir;
2144 vrefact(rdir);
2145 FILEDESC_SUNLOCK(fdp);
2146 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
2147 vrele(rdir);
2148 vrele(cdir);
2149
2150 if (!error) {
2151 if (bufseg == UIO_SYSSPACE)
2152 bcopy(bp, buf, strlen(bp) + 1);
2153 else
2154 error = copyout(bp, buf, strlen(bp) + 1);
2155 #ifdef KTRACE
2156 if (KTRPOINT(curthread, KTR_NAMEI))
2157 ktrnamei(bp);
2158 #endif
2159 }
2160 free(tmpbuf, M_TEMP);
2161 return (error);
2162 }
2163
2164 /*
2165 * Thus begins the fullpath magic.
2166 */
2167
2168 static int __read_mostly disablefullpath;
2169 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
2170 "Disable the vn_fullpath function");
2171
2172 /*
2173 * Retrieve the full filesystem path that correspond to a vnode from the name
2174 * cache (if available)
2175 */
2176 int
2177 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
2178 {
2179 char *buf;
2180 struct filedesc *fdp;
2181 struct vnode *rdir;
2182 int error;
2183
2184 if (__predict_false(disablefullpath))
2185 return (ENODEV);
2186 if (__predict_false(vn == NULL))
2187 return (EINVAL);
2188
2189 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
2190 fdp = td->td_proc->p_fd;
2191 FILEDESC_SLOCK(fdp);
2192 rdir = fdp->fd_rdir;
2193 vrefact(rdir);
2194 FILEDESC_SUNLOCK(fdp);
2195 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
2196 vrele(rdir);
2197
2198 if (!error)
2199 *freebuf = buf;
2200 else
2201 free(buf, M_TEMP);
2202 return (error);
2203 }
2204
2205 /*
2206 * This function is similar to vn_fullpath, but it attempts to lookup the
2207 * pathname relative to the global root mount point. This is required for the
2208 * auditing sub-system, as audited pathnames must be absolute, relative to the
2209 * global root mount point.
2210 */
2211 int
2212 vn_fullpath_global(struct thread *td, struct vnode *vn,
2213 char **retbuf, char **freebuf)
2214 {
2215 char *buf;
2216 int error;
2217
2218 if (__predict_false(disablefullpath))
2219 return (ENODEV);
2220 if (__predict_false(vn == NULL))
2221 return (EINVAL);
2222 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
2223 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
2224 if (!error)
2225 *freebuf = buf;
2226 else
2227 free(buf, M_TEMP);
2228 return (error);
2229 }
2230
2231 int
2232 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
2233 {
2234 struct vnode *dvp;
2235 struct namecache *ncp;
2236 struct mtx *vlp;
2237 int error;
2238
2239 vlp = VP2VNODELOCK(*vp);
2240 mtx_lock(vlp);
2241 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
2242 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
2243 break;
2244 }
2245 if (ncp != NULL) {
2246 if (*buflen < ncp->nc_nlen) {
2247 mtx_unlock(vlp);
2248 vrele(*vp);
2249 counter_u64_add(numfullpathfail4, 1);
2250 error = ENOMEM;
2251 SDT_PROBE3(vfs, namecache, fullpath, return, error,
2252 vp, NULL);
2253 return (error);
2254 }
2255 *buflen -= ncp->nc_nlen;
2256 memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen);
2257 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp,
2258 ncp->nc_name, vp);
2259 dvp = *vp;
2260 *vp = ncp->nc_dvp;
2261 vref(*vp);
2262 mtx_unlock(vlp);
2263 vrele(dvp);
2264 return (0);
2265 }
2266 SDT_PROBE1(vfs, namecache, fullpath, miss, vp);
2267
2268 mtx_unlock(vlp);
2269 vn_lock(*vp, LK_SHARED | LK_RETRY);
2270 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
2271 vput(*vp);
2272 if (error) {
2273 counter_u64_add(numfullpathfail2, 1);
2274 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
2275 return (error);
2276 }
2277
2278 *vp = dvp;
2279 if (dvp->v_iflag & VI_DOOMED) {
2280 /* forced unmount */
2281 vrele(dvp);
2282 error = ENOENT;
2283 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
2284 return (error);
2285 }
2286 /*
2287 * *vp has its use count incremented still.
2288 */
2289
2290 return (0);
2291 }
2292
2293 /*
2294 * The magic behind kern___getcwd() and vn_fullpath().
2295 */
2296 static int
2297 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
2298 char *buf, char **retbuf, u_int buflen)
2299 {
2300 int error, slash_prefixed;
2301 #ifdef KDTRACE_HOOKS
2302 struct vnode *startvp = vp;
2303 #endif
2304 struct vnode *vp1;
2305
2306 buflen--;
2307 buf[buflen] = '\0';
2308 error = 0;
2309 slash_prefixed = 0;
2310
2311 SDT_PROBE1(vfs, namecache, fullpath, entry, vp);
2312 counter_u64_add(numfullpathcalls, 1);
2313 vref(vp);
2314 if (vp->v_type != VDIR) {
2315 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen);
2316 if (error)
2317 return (error);
2318 if (buflen == 0) {
2319 vrele(vp);
2320 return (ENOMEM);
2321 }
2322 buf[--buflen] = '/';
2323 slash_prefixed = 1;
2324 }
2325 while (vp != rdir && vp != rootvnode) {
2326 /*
2327 * The vp vnode must be already fully constructed,
2328 * since it is either found in namecache or obtained
2329 * from VOP_VPTOCNP(). We may test for VV_ROOT safely
2330 * without obtaining the vnode lock.
2331 */
2332 if ((vp->v_vflag & VV_ROOT) != 0) {
2333 vn_lock(vp, LK_RETRY | LK_SHARED);
2334
2335 /*
2336 * With the vnode locked, check for races with
2337 * unmount, forced or not. Note that we
2338 * already verified that vp is not equal to
2339 * the root vnode, which means that
2340 * mnt_vnodecovered can be NULL only for the
2341 * case of unmount.
2342 */
2343 if ((vp->v_iflag & VI_DOOMED) != 0 ||
2344 (vp1 = vp->v_mount->mnt_vnodecovered) == NULL ||
2345 vp1->v_mountedhere != vp->v_mount) {
2346 vput(vp);
2347 error = ENOENT;
2348 SDT_PROBE3(vfs, namecache, fullpath, return,
2349 error, vp, NULL);
2350 break;
2351 }
2352
2353 vref(vp1);
2354 vput(vp);
2355 vp = vp1;
2356 continue;
2357 }
2358 if (vp->v_type != VDIR) {
2359 vrele(vp);
2360 counter_u64_add(numfullpathfail1, 1);
2361 error = ENOTDIR;
2362 SDT_PROBE3(vfs, namecache, fullpath, return,
2363 error, vp, NULL);
2364 break;
2365 }
2366 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen);
2367 if (error)
2368 break;
2369 if (buflen == 0) {
2370 vrele(vp);
2371 error = ENOMEM;
2372 SDT_PROBE3(vfs, namecache, fullpath, return, error,
2373 startvp, NULL);
2374 break;
2375 }
2376 buf[--buflen] = '/';
2377 slash_prefixed = 1;
2378 }
2379 if (error)
2380 return (error);
2381 if (!slash_prefixed) {
2382 if (buflen == 0) {
2383 vrele(vp);
2384 counter_u64_add(numfullpathfail4, 1);
2385 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM,
2386 startvp, NULL);
2387 return (ENOMEM);
2388 }
2389 buf[--buflen] = '/';
2390 }
2391 counter_u64_add(numfullpathfound, 1);
2392 vrele(vp);
2393
2394 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen);
2395 *retbuf = buf + buflen;
2396 return (0);
2397 }
2398
2399 struct vnode *
2400 vn_dir_dd_ino(struct vnode *vp)
2401 {
2402 struct namecache *ncp;
2403 struct vnode *ddvp;
2404 struct mtx *vlp;
2405
2406 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
2407 vlp = VP2VNODELOCK(vp);
2408 mtx_lock(vlp);
2409 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
2410 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
2411 continue;
2412 ddvp = ncp->nc_dvp;
2413 vhold(ddvp);
2414 mtx_unlock(vlp);
2415 if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread))
2416 return (NULL);
2417 return (ddvp);
2418 }
2419 mtx_unlock(vlp);
2420 return (NULL);
2421 }
2422
2423 int
2424 vn_commname(struct vnode *vp, char *buf, u_int buflen)
2425 {
2426 struct namecache *ncp;
2427 struct mtx *vlp;
2428 int l;
2429
2430 vlp = VP2VNODELOCK(vp);
2431 mtx_lock(vlp);
2432 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
2433 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
2434 break;
2435 if (ncp == NULL) {
2436 mtx_unlock(vlp);
2437 return (ENOENT);
2438 }
2439 l = min(ncp->nc_nlen, buflen - 1);
2440 memcpy(buf, ncp->nc_name, l);
2441 mtx_unlock(vlp);
2442 buf[l] = '\0';
2443 return (0);
2444 }
2445
2446 /* ABI compat shims for old kernel modules. */
2447 #undef cache_enter
2448
2449 void cache_enter(struct vnode *dvp, struct vnode *vp,
2450 struct componentname *cnp);
2451
2452 void
2453 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2454 {
2455
2456 cache_enter_time(dvp, vp, cnp, NULL, NULL);
2457 }
2458
2459 /*
2460 * This function updates path string to vnode's full global path
2461 * and checks the size of the new path string against the pathlen argument.
2462 *
2463 * Requires a locked, referenced vnode.
2464 * Vnode is re-locked on success or ENODEV, otherwise unlocked.
2465 *
2466 * If sysctl debug.disablefullpath is set, ENODEV is returned,
2467 * vnode is left locked and path remain untouched.
2468 *
2469 * If vp is a directory, the call to vn_fullpath_global() always succeeds
2470 * because it falls back to the ".." lookup if the namecache lookup fails.
2471 */
2472 int
2473 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
2474 u_int pathlen)
2475 {
2476 struct nameidata nd;
2477 struct vnode *vp1;
2478 char *rpath, *fbuf;
2479 int error;
2480
2481 ASSERT_VOP_ELOCKED(vp, __func__);
2482
2483 /* Return ENODEV if sysctl debug.disablefullpath==1 */
2484 if (__predict_false(disablefullpath))
2485 return (ENODEV);
2486
2487 /* Construct global filesystem path from vp. */
2488 VOP_UNLOCK(vp, 0);
2489 error = vn_fullpath_global(td, vp, &rpath, &fbuf);
2490
2491 if (error != 0) {
2492 vrele(vp);
2493 return (error);
2494 }
2495
2496 if (strlen(rpath) >= pathlen) {
2497 vrele(vp);
2498 error = ENAMETOOLONG;
2499 goto out;
2500 }
2501
2502 /*
2503 * Re-lookup the vnode by path to detect a possible rename.
2504 * As a side effect, the vnode is relocked.
2505 * If vnode was renamed, return ENOENT.
2506 */
2507 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
2508 UIO_SYSSPACE, path, td);
2509 error = namei(&nd);
2510 if (error != 0) {
2511 vrele(vp);
2512 goto out;
2513 }
2514 NDFREE(&nd, NDF_ONLY_PNBUF);
2515 vp1 = nd.ni_vp;
2516 vrele(vp);
2517 if (vp1 == vp)
2518 strcpy(path, rpath);
2519 else {
2520 vput(vp1);
2521 error = ENOENT;
2522 }
2523
2524 out:
2525 free(fbuf, M_TEMP);
2526 return (error);
2527 }
Cache object: 1a6683e721f4ce952d7a400e53f30072
|