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("$FreeBSD$");
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 (dvp, name) where dvp 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 *
NCP2NEGLIST(struct namecache * ncp)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 *
VP2VNODELOCK(struct vnode * vp)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 *
cache_alloc(int len,int ts)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
cache_free(struct namecache * ncp)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
cache_out_ts(struct namecache * ncp,struct timespec * tsp,int * ticksp)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
cache_maybe_yield(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
cache_assert_vlp_locked(struct mtx * vlp)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
cache_assert_vnode_locked(struct vnode * vp)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
cache_get_hash(char * name,u_char len,struct vnode * dvp)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 *
NCP2BUCKETLOCK(struct namecache * ncp)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
cache_assert_bucket_locked(struct namecache * ncp,int mode)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
_cache_sort(void ** p1,void ** p2)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
cache_lock_all_buckets(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
cache_unlock_all_buckets(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
cache_lock_all_vnodes(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
cache_unlock_all_vnodes(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
cache_trylock_vnodes(struct mtx * vlp1,struct mtx * vlp2)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
cache_unlock_vnodes(struct mtx * vlp1,struct mtx * vlp2)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
sysctl_nchstats(SYSCTL_HANDLER_ARGS)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
sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)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
sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)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
cache_negative_hit(struct namecache * ncp)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
cache_negative_insert(struct namecache * ncp,bool neg_locked)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
cache_negative_remove(struct namecache * ncp,bool neg_locked)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
cache_negative_shrink_select(int start,struct namecache ** ncpp,struct neglist ** neglistpp)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
cache_negative_zap_one(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
cache_zap_locked(struct namecache * ncp,bool neg_locked)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
cache_zap_negative_locked_vnode_kl(struct namecache * ncp,struct vnode * vp)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
cache_zap_locked_vnode_kl2(struct namecache * ncp,struct vnode * vp,struct mtx ** vlpp)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
cache_zap_locked_vnode(struct namecache * ncp,struct vnode * vp)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
cache_zap_wlocked_bucket(struct namecache * ncp,struct rwlock * blp)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
cache_zap_rlocked_bucket(struct namecache * ncp,struct rwlock * blp)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
cache_zap_wlocked_bucket_kl(struct namecache * ncp,struct rwlock * blp,struct mtx ** vlpp1,struct mtx ** vlpp2)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
cache_lookup_unlock(struct rwlock * blp,struct mtx * vlp)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
cache_lookup_dot(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,struct timespec * tsp,int * ticksp)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 static __noinline int
cache_lookup_nomakeentry(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,struct timespec * tsp,int * ticksp)1125 cache_lookup_nomakeentry(struct vnode *dvp, struct vnode **vpp,
1126 struct componentname *cnp, struct timespec *tsp, int *ticksp)
1127 {
1128 struct namecache *ncp;
1129 struct rwlock *blp;
1130 struct mtx *dvlp, *dvlp2;
1131 uint32_t hash;
1132 int error;
1133
1134 if (cnp->cn_namelen == 2 &&
1135 cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') {
1136 counter_u64_add(dotdothits, 1);
1137 dvlp = VP2VNODELOCK(dvp);
1138 dvlp2 = NULL;
1139 mtx_lock(dvlp);
1140 retry_dotdot:
1141 ncp = dvp->v_cache_dd;
1142 if (ncp == NULL) {
1143 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
1144 "..", NULL);
1145 mtx_unlock(dvlp);
1146 if (dvlp2 != NULL)
1147 mtx_unlock(dvlp2);
1148 return (0);
1149 }
1150 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) {
1151 if (ncp->nc_dvp != dvp)
1152 panic("dvp %p v_cache_dd %p\n", dvp, ncp);
1153 if (!cache_zap_locked_vnode_kl2(ncp,
1154 dvp, &dvlp2))
1155 goto retry_dotdot;
1156 MPASS(dvp->v_cache_dd == NULL);
1157 mtx_unlock(dvlp);
1158 if (dvlp2 != NULL)
1159 mtx_unlock(dvlp2);
1160 cache_free(ncp);
1161 } else {
1162 dvp->v_cache_dd = NULL;
1163 mtx_unlock(dvlp);
1164 if (dvlp2 != NULL)
1165 mtx_unlock(dvlp2);
1166 }
1167 return (0);
1168 }
1169
1170 hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp);
1171 blp = HASH2BUCKETLOCK(hash);
1172 retry:
1173 if (LIST_EMPTY(NCHHASH(hash)))
1174 goto out_no_entry;
1175
1176 rw_wlock(blp);
1177
1178 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
1179 counter_u64_add(numchecks, 1);
1180 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
1181 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
1182 break;
1183 }
1184
1185 /* We failed to find an entry */
1186 if (ncp == NULL) {
1187 rw_wunlock(blp);
1188 goto out_no_entry;
1189 }
1190
1191 counter_u64_add(numposzaps, 1);
1192
1193 error = cache_zap_wlocked_bucket(ncp, blp);
1194 if (error != 0) {
1195 zap_and_exit_bucket_fail++;
1196 cache_maybe_yield();
1197 goto retry;
1198 }
1199 cache_free(ncp);
1200 return (0);
1201 out_no_entry:
1202 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL);
1203 counter_u64_add(nummisszap, 1);
1204 return (0);
1205 }
1206
1207 /**
1208 * Lookup a name in the name cache
1209 *
1210 * # Arguments
1211 *
1212 * - dvp: Parent directory in which to search.
1213 * - vpp: Return argument. Will contain desired vnode on cache hit.
1214 * - cnp: Parameters of the name search. The most interesting bits of
1215 * the cn_flags field have the following meanings:
1216 * - MAKEENTRY: If clear, free an entry from the cache rather than look
1217 * it up.
1218 * - ISDOTDOT: Must be set if and only if cn_nameptr == ".."
1219 * - tsp: Return storage for cache timestamp. On a successful (positive
1220 * or negative) lookup, tsp will be filled with any timespec that
1221 * was stored when this cache entry was created. However, it will
1222 * be clear for "." entries.
1223 * - ticks: Return storage for alternate cache timestamp. On a successful
1224 * (positive or negative) lookup, it will contain the ticks value
1225 * that was current when the cache entry was created, unless cnp
1226 * was ".".
1227 *
1228 * # Returns
1229 *
1230 * - -1: A positive cache hit. vpp will contain the desired vnode.
1231 * - ENOENT: A negative cache hit, or dvp was recycled out from under us due
1232 * to a forced unmount. vpp will not be modified. If the entry
1233 * is a whiteout, then the ISWHITEOUT flag will be set in
1234 * cnp->cn_flags.
1235 * - 0: A cache miss. vpp will not be modified.
1236 *
1237 * # Locking
1238 *
1239 * On a cache hit, vpp will be returned locked and ref'd. If we're looking up
1240 * .., dvp is unlocked. If we're looking up . an extra ref is taken, but the
1241 * lock is not recursively acquired.
1242 */
1243 int
cache_lookup(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,struct timespec * tsp,int * ticksp)1244 cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1245 struct timespec *tsp, int *ticksp)
1246 {
1247 struct namecache_ts *ncp_ts;
1248 struct namecache *ncp;
1249 struct rwlock *blp;
1250 struct mtx *dvlp, *dvlp2;
1251 uint32_t hash;
1252 int error, ltype;
1253
1254 if (__predict_false(!doingcache)) {
1255 cnp->cn_flags &= ~MAKEENTRY;
1256 return (0);
1257 }
1258
1259 counter_u64_add(numcalls, 1);
1260
1261 if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.'))
1262 return (cache_lookup_dot(dvp, vpp, cnp, tsp, ticksp));
1263
1264 if ((cnp->cn_flags & MAKEENTRY) == 0)
1265 return (cache_lookup_nomakeentry(dvp, vpp, cnp, tsp, ticksp));
1266
1267 retry:
1268 blp = NULL;
1269 error = 0;
1270 if (cnp->cn_namelen == 2 &&
1271 cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') {
1272 counter_u64_add(dotdothits, 1);
1273 dvlp = VP2VNODELOCK(dvp);
1274 dvlp2 = NULL;
1275 mtx_lock(dvlp);
1276 ncp = dvp->v_cache_dd;
1277 if (ncp == NULL) {
1278 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
1279 "..", NULL);
1280 mtx_unlock(dvlp);
1281 return (0);
1282 }
1283 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) {
1284 if (ncp->nc_flag & NCF_NEGATIVE)
1285 *vpp = NULL;
1286 else
1287 *vpp = ncp->nc_vp;
1288 } else
1289 *vpp = ncp->nc_dvp;
1290 /* Return failure if negative entry was found. */
1291 if (*vpp == NULL)
1292 goto negative_success;
1293 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
1294 dvp, cnp->cn_nameptr, *vpp);
1295 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..",
1296 *vpp);
1297 cache_out_ts(ncp, tsp, ticksp);
1298 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
1299 NCF_DTS && tsp != NULL) {
1300 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc);
1301 *tsp = ncp_ts->nc_dotdottime;
1302 }
1303 goto success;
1304 }
1305
1306 hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp);
1307 blp = HASH2BUCKETLOCK(hash);
1308 rw_rlock(blp);
1309
1310 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
1311 counter_u64_add(numchecks, 1);
1312 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
1313 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
1314 break;
1315 }
1316
1317 /* We failed to find an entry */
1318 if (ncp == NULL) {
1319 rw_runlock(blp);
1320 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
1321 NULL);
1322 counter_u64_add(nummiss, 1);
1323 return (0);
1324 }
1325
1326 /* We found a "positive" match, return the vnode */
1327 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
1328 counter_u64_add(numposhits, 1);
1329 *vpp = ncp->nc_vp;
1330 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
1331 dvp, cnp->cn_nameptr, *vpp, ncp);
1332 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name,
1333 *vpp);
1334 cache_out_ts(ncp, tsp, ticksp);
1335 goto success;
1336 }
1337
1338 negative_success:
1339 /* We found a negative match, and want to create it, so purge */
1340 if (cnp->cn_nameiop == CREATE) {
1341 counter_u64_add(numnegzaps, 1);
1342 goto zap_and_exit;
1343 }
1344
1345 counter_u64_add(numneghits, 1);
1346 cache_negative_hit(ncp);
1347 if (ncp->nc_flag & NCF_WHITE)
1348 cnp->cn_flags |= ISWHITEOUT;
1349 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp,
1350 ncp->nc_name);
1351 cache_out_ts(ncp, tsp, ticksp);
1352 cache_lookup_unlock(blp, dvlp);
1353 return (ENOENT);
1354
1355 success:
1356 /*
1357 * On success we return a locked and ref'd vnode as per the lookup
1358 * protocol.
1359 */
1360 MPASS(dvp != *vpp);
1361 ltype = 0; /* silence gcc warning */
1362 if (cnp->cn_flags & ISDOTDOT) {
1363 ltype = VOP_ISLOCKED(dvp);
1364 VOP_UNLOCK(dvp, 0);
1365 }
1366 vhold(*vpp);
1367 cache_lookup_unlock(blp, dvlp);
1368 error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread);
1369 if (cnp->cn_flags & ISDOTDOT) {
1370 vn_lock(dvp, ltype | LK_RETRY);
1371 if (dvp->v_iflag & VI_DOOMED) {
1372 if (error == 0)
1373 vput(*vpp);
1374 *vpp = NULL;
1375 return (ENOENT);
1376 }
1377 }
1378 if (error) {
1379 *vpp = NULL;
1380 goto retry;
1381 }
1382 if ((cnp->cn_flags & ISLASTCN) &&
1383 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
1384 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
1385 }
1386 return (-1);
1387
1388 zap_and_exit:
1389 if (blp != NULL)
1390 error = cache_zap_rlocked_bucket(ncp, blp);
1391 else
1392 error = cache_zap_locked_vnode(ncp, dvp);
1393 if (error != 0) {
1394 zap_and_exit_bucket_fail++;
1395 cache_maybe_yield();
1396 goto retry;
1397 }
1398 cache_free(ncp);
1399 return (0);
1400 }
1401
1402 struct celockstate {
1403 struct mtx *vlp[3];
1404 struct rwlock *blp[2];
1405 };
1406 CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3));
1407 CTASSERT((nitems(((struct celockstate *)0)->blp) == 2));
1408
1409 static inline void
cache_celockstate_init(struct celockstate * cel)1410 cache_celockstate_init(struct celockstate *cel)
1411 {
1412
1413 bzero(cel, sizeof(*cel));
1414 }
1415
1416 static void
cache_lock_vnodes_cel(struct celockstate * cel,struct vnode * vp,struct vnode * dvp)1417 cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp,
1418 struct vnode *dvp)
1419 {
1420 struct mtx *vlp1, *vlp2;
1421
1422 MPASS(cel->vlp[0] == NULL);
1423 MPASS(cel->vlp[1] == NULL);
1424 MPASS(cel->vlp[2] == NULL);
1425
1426 MPASS(vp != NULL || dvp != NULL);
1427
1428 vlp1 = VP2VNODELOCK(vp);
1429 vlp2 = VP2VNODELOCK(dvp);
1430 cache_sort(&vlp1, &vlp2);
1431
1432 if (vlp1 != NULL) {
1433 mtx_lock(vlp1);
1434 cel->vlp[0] = vlp1;
1435 }
1436 mtx_lock(vlp2);
1437 cel->vlp[1] = vlp2;
1438 }
1439
1440 static void
cache_unlock_vnodes_cel(struct celockstate * cel)1441 cache_unlock_vnodes_cel(struct celockstate *cel)
1442 {
1443
1444 MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL);
1445
1446 if (cel->vlp[0] != NULL)
1447 mtx_unlock(cel->vlp[0]);
1448 if (cel->vlp[1] != NULL)
1449 mtx_unlock(cel->vlp[1]);
1450 if (cel->vlp[2] != NULL)
1451 mtx_unlock(cel->vlp[2]);
1452 }
1453
1454 static bool
cache_lock_vnodes_cel_3(struct celockstate * cel,struct vnode * vp)1455 cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp)
1456 {
1457 struct mtx *vlp;
1458 bool ret;
1459
1460 cache_assert_vlp_locked(cel->vlp[0]);
1461 cache_assert_vlp_locked(cel->vlp[1]);
1462 MPASS(cel->vlp[2] == NULL);
1463
1464 MPASS(vp != NULL);
1465 vlp = VP2VNODELOCK(vp);
1466
1467 ret = true;
1468 if (vlp >= cel->vlp[1]) {
1469 mtx_lock(vlp);
1470 } else {
1471 if (mtx_trylock(vlp))
1472 goto out;
1473 cache_lock_vnodes_cel_3_failures++;
1474 cache_unlock_vnodes_cel(cel);
1475 if (vlp < cel->vlp[0]) {
1476 mtx_lock(vlp);
1477 mtx_lock(cel->vlp[0]);
1478 mtx_lock(cel->vlp[1]);
1479 } else {
1480 if (cel->vlp[0] != NULL)
1481 mtx_lock(cel->vlp[0]);
1482 mtx_lock(vlp);
1483 mtx_lock(cel->vlp[1]);
1484 }
1485 ret = false;
1486 }
1487 out:
1488 cel->vlp[2] = vlp;
1489 return (ret);
1490 }
1491
1492 static void
cache_lock_buckets_cel(struct celockstate * cel,struct rwlock * blp1,struct rwlock * blp2)1493 cache_lock_buckets_cel(struct celockstate *cel, struct rwlock *blp1,
1494 struct rwlock *blp2)
1495 {
1496
1497 MPASS(cel->blp[0] == NULL);
1498 MPASS(cel->blp[1] == NULL);
1499
1500 cache_sort(&blp1, &blp2);
1501
1502 if (blp1 != NULL) {
1503 rw_wlock(blp1);
1504 cel->blp[0] = blp1;
1505 }
1506 rw_wlock(blp2);
1507 cel->blp[1] = blp2;
1508 }
1509
1510 static void
cache_unlock_buckets_cel(struct celockstate * cel)1511 cache_unlock_buckets_cel(struct celockstate *cel)
1512 {
1513
1514 if (cel->blp[0] != NULL)
1515 rw_wunlock(cel->blp[0]);
1516 rw_wunlock(cel->blp[1]);
1517 }
1518
1519 /*
1520 * Lock part of the cache affected by the insertion.
1521 *
1522 * This means vnodelocks for dvp, vp and the relevant bucketlock.
1523 * However, insertion can result in removal of an old entry. In this
1524 * case we have an additional vnode and bucketlock pair to lock. If the
1525 * entry is negative, ncelock is locked instead of the vnode.
1526 *
1527 * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while
1528 * preserving the locking order (smaller address first).
1529 */
1530 static void
cache_enter_lock(struct celockstate * cel,struct vnode * dvp,struct vnode * vp,uint32_t hash)1531 cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp,
1532 uint32_t hash)
1533 {
1534 struct namecache *ncp;
1535 struct rwlock *blps[2];
1536
1537 blps[0] = HASH2BUCKETLOCK(hash);
1538 for (;;) {
1539 blps[1] = NULL;
1540 cache_lock_vnodes_cel(cel, dvp, vp);
1541 if (vp == NULL || vp->v_type != VDIR)
1542 break;
1543 ncp = vp->v_cache_dd;
1544 if (ncp == NULL)
1545 break;
1546 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1547 break;
1548 MPASS(ncp->nc_dvp == vp);
1549 blps[1] = NCP2BUCKETLOCK(ncp);
1550 if (ncp->nc_flag & NCF_NEGATIVE)
1551 break;
1552 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp))
1553 break;
1554 /*
1555 * All vnodes got re-locked. Re-validate the state and if
1556 * nothing changed we are done. Otherwise restart.
1557 */
1558 if (ncp == vp->v_cache_dd &&
1559 (ncp->nc_flag & NCF_ISDOTDOT) != 0 &&
1560 blps[1] == NCP2BUCKETLOCK(ncp) &&
1561 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2])
1562 break;
1563 cache_unlock_vnodes_cel(cel);
1564 cel->vlp[0] = NULL;
1565 cel->vlp[1] = NULL;
1566 cel->vlp[2] = NULL;
1567 }
1568 cache_lock_buckets_cel(cel, blps[0], blps[1]);
1569 }
1570
1571 static void
cache_enter_lock_dd(struct celockstate * cel,struct vnode * dvp,struct vnode * vp,uint32_t hash)1572 cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp,
1573 uint32_t hash)
1574 {
1575 struct namecache *ncp;
1576 struct rwlock *blps[2];
1577
1578 blps[0] = HASH2BUCKETLOCK(hash);
1579 for (;;) {
1580 blps[1] = NULL;
1581 cache_lock_vnodes_cel(cel, dvp, vp);
1582 ncp = dvp->v_cache_dd;
1583 if (ncp == NULL)
1584 break;
1585 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1586 break;
1587 MPASS(ncp->nc_dvp == dvp);
1588 blps[1] = NCP2BUCKETLOCK(ncp);
1589 if (ncp->nc_flag & NCF_NEGATIVE)
1590 break;
1591 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp))
1592 break;
1593 if (ncp == dvp->v_cache_dd &&
1594 (ncp->nc_flag & NCF_ISDOTDOT) != 0 &&
1595 blps[1] == NCP2BUCKETLOCK(ncp) &&
1596 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2])
1597 break;
1598 cache_unlock_vnodes_cel(cel);
1599 cel->vlp[0] = NULL;
1600 cel->vlp[1] = NULL;
1601 cel->vlp[2] = NULL;
1602 }
1603 cache_lock_buckets_cel(cel, blps[0], blps[1]);
1604 }
1605
1606 static void
cache_enter_unlock(struct celockstate * cel)1607 cache_enter_unlock(struct celockstate *cel)
1608 {
1609
1610 cache_unlock_buckets_cel(cel);
1611 cache_unlock_vnodes_cel(cel);
1612 }
1613
1614 /*
1615 * Add an entry to the cache.
1616 */
1617 void
cache_enter_time(struct vnode * dvp,struct vnode * vp,struct componentname * cnp,struct timespec * tsp,struct timespec * dtsp)1618 cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp,
1619 struct timespec *tsp, struct timespec *dtsp)
1620 {
1621 struct celockstate cel;
1622 struct namecache *ncp, *n2, *ndd;
1623 struct namecache_ts *ncp_ts, *n2_ts;
1624 struct nchashhead *ncpp;
1625 struct neglist *neglist;
1626 uint32_t hash;
1627 int flag;
1628 int len;
1629 bool neg_locked;
1630 int lnumcache;
1631
1632 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
1633 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
1634 ("cache_enter: Adding a doomed vnode"));
1635 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
1636 ("cache_enter: Doomed vnode used as src"));
1637
1638 if (__predict_false(!doingcache))
1639 return;
1640
1641 /*
1642 * Avoid blowout in namecache entries.
1643 */
1644 if (__predict_false(numcache >= desiredvnodes * ncsizefactor))
1645 return;
1646
1647 cache_celockstate_init(&cel);
1648 ndd = NULL;
1649 flag = 0;
1650 if (cnp->cn_nameptr[0] == '.') {
1651 if (cnp->cn_namelen == 1)
1652 return;
1653 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
1654 len = cnp->cn_namelen;
1655 hash = cache_get_hash(cnp->cn_nameptr, len, dvp);
1656 cache_enter_lock_dd(&cel, dvp, vp, hash);
1657 /*
1658 * If dotdot entry already exists, just retarget it
1659 * to new parent vnode, otherwise continue with new
1660 * namecache entry allocation.
1661 */
1662 if ((ncp = dvp->v_cache_dd) != NULL &&
1663 ncp->nc_flag & NCF_ISDOTDOT) {
1664 KASSERT(ncp->nc_dvp == dvp,
1665 ("wrong isdotdot parent"));
1666 neg_locked = false;
1667 if (ncp->nc_flag & NCF_NEGATIVE || vp == NULL) {
1668 neglist = NCP2NEGLIST(ncp);
1669 mtx_lock(&ncneg_hot.nl_lock);
1670 mtx_lock(&neglist->nl_lock);
1671 neg_locked = true;
1672 }
1673 if (!(ncp->nc_flag & NCF_NEGATIVE)) {
1674 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
1675 ncp, nc_dst);
1676 } else {
1677 cache_negative_remove(ncp, true);
1678 }
1679 if (vp != NULL) {
1680 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
1681 ncp, nc_dst);
1682 ncp->nc_flag &= ~(NCF_NEGATIVE|NCF_HOTNEGATIVE);
1683 } else {
1684 ncp->nc_flag &= ~(NCF_HOTNEGATIVE);
1685 ncp->nc_flag |= NCF_NEGATIVE;
1686 cache_negative_insert(ncp, true);
1687 }
1688 if (neg_locked) {
1689 mtx_unlock(&neglist->nl_lock);
1690 mtx_unlock(&ncneg_hot.nl_lock);
1691 }
1692 ncp->nc_vp = vp;
1693 cache_enter_unlock(&cel);
1694 return;
1695 }
1696 dvp->v_cache_dd = NULL;
1697 cache_enter_unlock(&cel);
1698 cache_celockstate_init(&cel);
1699 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp);
1700 flag = NCF_ISDOTDOT;
1701 }
1702 }
1703
1704 /*
1705 * Calculate the hash key and setup as much of the new
1706 * namecache entry as possible before acquiring the lock.
1707 */
1708 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
1709 ncp->nc_flag = flag;
1710 ncp->nc_vp = vp;
1711 if (vp == NULL)
1712 ncp->nc_flag |= NCF_NEGATIVE;
1713 ncp->nc_dvp = dvp;
1714 if (tsp != NULL) {
1715 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc);
1716 ncp_ts->nc_time = *tsp;
1717 ncp_ts->nc_ticks = ticks;
1718 ncp_ts->nc_nc.nc_flag |= NCF_TS;
1719 if (dtsp != NULL) {
1720 ncp_ts->nc_dotdottime = *dtsp;
1721 ncp_ts->nc_nc.nc_flag |= NCF_DTS;
1722 }
1723 }
1724 len = ncp->nc_nlen = cnp->cn_namelen;
1725 hash = cache_get_hash(cnp->cn_nameptr, len, dvp);
1726 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1);
1727 cache_enter_lock(&cel, dvp, vp, hash);
1728
1729 /*
1730 * See if this vnode or negative entry is already in the cache
1731 * with this name. This can happen with concurrent lookups of
1732 * the same path name.
1733 */
1734 ncpp = NCHHASH(hash);
1735 LIST_FOREACH(n2, ncpp, nc_hash) {
1736 if (n2->nc_dvp == dvp &&
1737 n2->nc_nlen == cnp->cn_namelen &&
1738 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
1739 if (tsp != NULL) {
1740 KASSERT((n2->nc_flag & NCF_TS) != 0,
1741 ("no NCF_TS"));
1742 n2_ts = __containerof(n2, struct namecache_ts, nc_nc);
1743 n2_ts->nc_time = ncp_ts->nc_time;
1744 n2_ts->nc_ticks = ncp_ts->nc_ticks;
1745 if (dtsp != NULL) {
1746 n2_ts->nc_dotdottime = ncp_ts->nc_dotdottime;
1747 if (ncp->nc_flag & NCF_NEGATIVE)
1748 mtx_lock(&ncneg_hot.nl_lock);
1749 n2_ts->nc_nc.nc_flag |= NCF_DTS;
1750 if (ncp->nc_flag & NCF_NEGATIVE)
1751 mtx_unlock(&ncneg_hot.nl_lock);
1752 }
1753 }
1754 goto out_unlock_free;
1755 }
1756 }
1757
1758 if (flag == NCF_ISDOTDOT) {
1759 /*
1760 * See if we are trying to add .. entry, but some other lookup
1761 * has populated v_cache_dd pointer already.
1762 */
1763 if (dvp->v_cache_dd != NULL)
1764 goto out_unlock_free;
1765 KASSERT(vp == NULL || vp->v_type == VDIR,
1766 ("wrong vnode type %p", vp));
1767 dvp->v_cache_dd = ncp;
1768 }
1769
1770 if (vp != NULL) {
1771 if (vp->v_type == VDIR) {
1772 if (flag != NCF_ISDOTDOT) {
1773 /*
1774 * For this case, the cache entry maps both the
1775 * directory name in it and the name ".." for the
1776 * directory's parent.
1777 */
1778 if ((ndd = vp->v_cache_dd) != NULL) {
1779 if ((ndd->nc_flag & NCF_ISDOTDOT) != 0)
1780 cache_zap_locked(ndd, false);
1781 else
1782 ndd = NULL;
1783 }
1784 vp->v_cache_dd = ncp;
1785 }
1786 } else {
1787 vp->v_cache_dd = NULL;
1788 }
1789 }
1790
1791 if (flag != NCF_ISDOTDOT) {
1792 if (LIST_EMPTY(&dvp->v_cache_src)) {
1793 vhold(dvp);
1794 atomic_add_rel_long(&numcachehv, 1);
1795 }
1796 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
1797 }
1798
1799 /*
1800 * Insert the new namecache entry into the appropriate chain
1801 * within the cache entries table.
1802 */
1803 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
1804
1805 /*
1806 * If the entry is "negative", we place it into the
1807 * "negative" cache queue, otherwise, we place it into the
1808 * destination vnode's cache entries queue.
1809 */
1810 if (vp != NULL) {
1811 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
1812 SDT_PROBE3(vfs, namecache, enter, done, dvp, ncp->nc_name,
1813 vp);
1814 } else {
1815 if (cnp->cn_flags & ISWHITEOUT)
1816 ncp->nc_flag |= NCF_WHITE;
1817 cache_negative_insert(ncp, false);
1818 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp,
1819 ncp->nc_name);
1820 }
1821 cache_enter_unlock(&cel);
1822 lnumcache = atomic_fetchadd_long(&numcache, 1) + 1;
1823 if (numneg * ncnegfactor > lnumcache)
1824 cache_negative_zap_one();
1825 cache_free(ndd);
1826 return;
1827 out_unlock_free:
1828 cache_enter_unlock(&cel);
1829 cache_free(ncp);
1830 return;
1831 }
1832
1833 static u_int
cache_roundup_2(u_int val)1834 cache_roundup_2(u_int val)
1835 {
1836 u_int res;
1837
1838 for (res = 1; res <= val; res <<= 1)
1839 continue;
1840
1841 return (res);
1842 }
1843
1844 /*
1845 * Name cache initialization, from vfs_init() when we are booting
1846 */
1847 static void
nchinit(void * dummy __unused)1848 nchinit(void *dummy __unused)
1849 {
1850 u_int i;
1851
1852 cache_zone_small = uma_zcreate("S VFS Cache",
1853 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
1854 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache),
1855 UMA_ZONE_ZINIT);
1856 cache_zone_small_ts = uma_zcreate("STS VFS Cache",
1857 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
1858 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts),
1859 UMA_ZONE_ZINIT);
1860 cache_zone_large = uma_zcreate("L VFS Cache",
1861 sizeof(struct namecache) + NAME_MAX + 1,
1862 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache),
1863 UMA_ZONE_ZINIT);
1864 cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
1865 sizeof(struct namecache_ts) + NAME_MAX + 1,
1866 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts),
1867 UMA_ZONE_ZINIT);
1868
1869 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
1870 ncbuckethash = cache_roundup_2(mp_ncpus * 64) - 1;
1871 if (ncbuckethash > nchash)
1872 ncbuckethash = nchash;
1873 bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE,
1874 M_WAITOK | M_ZERO);
1875 for (i = 0; i < numbucketlocks; i++)
1876 rw_init_flags(&bucketlocks[i], "ncbuc", RW_DUPOK | RW_RECURSE);
1877 ncvnodehash = cache_roundup_2(mp_ncpus * 64) - 1;
1878 vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE,
1879 M_WAITOK | M_ZERO);
1880 for (i = 0; i < numvnodelocks; i++)
1881 mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE);
1882 ncpurgeminvnodes = numbucketlocks;
1883
1884 ncneghash = 3;
1885 neglists = malloc(sizeof(*neglists) * numneglists, M_VFSCACHE,
1886 M_WAITOK | M_ZERO);
1887 for (i = 0; i < numneglists; i++) {
1888 mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF);
1889 TAILQ_INIT(&neglists[i].nl_list);
1890 }
1891 mtx_init(&ncneg_hot.nl_lock, "ncneglh", NULL, MTX_DEF);
1892 TAILQ_INIT(&ncneg_hot.nl_list);
1893
1894 mtx_init(&ncneg_shrink_lock, "ncnegs", NULL, MTX_DEF);
1895
1896 numcalls = counter_u64_alloc(M_WAITOK);
1897 dothits = counter_u64_alloc(M_WAITOK);
1898 dotdothits = counter_u64_alloc(M_WAITOK);
1899 numchecks = counter_u64_alloc(M_WAITOK);
1900 nummiss = counter_u64_alloc(M_WAITOK);
1901 nummisszap = counter_u64_alloc(M_WAITOK);
1902 numposzaps = counter_u64_alloc(M_WAITOK);
1903 numposhits = counter_u64_alloc(M_WAITOK);
1904 numnegzaps = counter_u64_alloc(M_WAITOK);
1905 numneghits = counter_u64_alloc(M_WAITOK);
1906 numfullpathcalls = counter_u64_alloc(M_WAITOK);
1907 numfullpathfail1 = counter_u64_alloc(M_WAITOK);
1908 numfullpathfail2 = counter_u64_alloc(M_WAITOK);
1909 numfullpathfail4 = counter_u64_alloc(M_WAITOK);
1910 numfullpathfound = counter_u64_alloc(M_WAITOK);
1911 }
1912 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
1913
1914 void
cache_changesize(int newmaxvnodes)1915 cache_changesize(int newmaxvnodes)
1916 {
1917 struct nchashhead *new_nchashtbl, *old_nchashtbl;
1918 u_long new_nchash, old_nchash;
1919 struct namecache *ncp;
1920 uint32_t hash;
1921 int i;
1922
1923 newmaxvnodes = cache_roundup_2(newmaxvnodes * 2);
1924 if (newmaxvnodes < numbucketlocks)
1925 newmaxvnodes = numbucketlocks;
1926
1927 new_nchashtbl = hashinit(newmaxvnodes, M_VFSCACHE, &new_nchash);
1928 /* If same hash table size, nothing to do */
1929 if (nchash == new_nchash) {
1930 free(new_nchashtbl, M_VFSCACHE);
1931 return;
1932 }
1933 /*
1934 * Move everything from the old hash table to the new table.
1935 * None of the namecache entries in the table can be removed
1936 * because to do so, they have to be removed from the hash table.
1937 */
1938 cache_lock_all_vnodes();
1939 cache_lock_all_buckets();
1940 old_nchashtbl = nchashtbl;
1941 old_nchash = nchash;
1942 nchashtbl = new_nchashtbl;
1943 nchash = new_nchash;
1944 for (i = 0; i <= old_nchash; i++) {
1945 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) {
1946 hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen,
1947 ncp->nc_dvp);
1948 LIST_REMOVE(ncp, nc_hash);
1949 LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash);
1950 }
1951 }
1952 cache_unlock_all_buckets();
1953 cache_unlock_all_vnodes();
1954 free(old_nchashtbl, M_VFSCACHE);
1955 }
1956
1957 /*
1958 * Invalidate all entries to a particular vnode.
1959 */
1960 void
cache_purge(struct vnode * vp)1961 cache_purge(struct vnode *vp)
1962 {
1963 TAILQ_HEAD(, namecache) ncps;
1964 struct namecache *ncp, *nnp;
1965 struct mtx *vlp, *vlp2;
1966
1967 CTR1(KTR_VFS, "cache_purge(%p)", vp);
1968 SDT_PROBE1(vfs, namecache, purge, done, vp);
1969 if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) &&
1970 vp->v_cache_dd == NULL)
1971 return;
1972 TAILQ_INIT(&ncps);
1973 vlp = VP2VNODELOCK(vp);
1974 vlp2 = NULL;
1975 mtx_lock(vlp);
1976 retry:
1977 while (!LIST_EMPTY(&vp->v_cache_src)) {
1978 ncp = LIST_FIRST(&vp->v_cache_src);
1979 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1980 goto retry;
1981 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1982 }
1983 while (!TAILQ_EMPTY(&vp->v_cache_dst)) {
1984 ncp = TAILQ_FIRST(&vp->v_cache_dst);
1985 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1986 goto retry;
1987 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1988 }
1989 ncp = vp->v_cache_dd;
1990 if (ncp != NULL) {
1991 KASSERT(ncp->nc_flag & NCF_ISDOTDOT,
1992 ("lost dotdot link"));
1993 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2))
1994 goto retry;
1995 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
1996 }
1997 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
1998 mtx_unlock(vlp);
1999 if (vlp2 != NULL)
2000 mtx_unlock(vlp2);
2001 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
2002 cache_free(ncp);
2003 }
2004 }
2005
2006 /*
2007 * Invalidate all negative entries for a particular directory vnode.
2008 */
2009 void
cache_purge_negative(struct vnode * vp)2010 cache_purge_negative(struct vnode *vp)
2011 {
2012 TAILQ_HEAD(, namecache) ncps;
2013 struct namecache *ncp, *nnp;
2014 struct mtx *vlp;
2015
2016 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
2017 SDT_PROBE1(vfs, namecache, purge_negative, done, vp);
2018 if (LIST_EMPTY(&vp->v_cache_src))
2019 return;
2020 TAILQ_INIT(&ncps);
2021 vlp = VP2VNODELOCK(vp);
2022 mtx_lock(vlp);
2023 LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) {
2024 if (!(ncp->nc_flag & NCF_NEGATIVE))
2025 continue;
2026 cache_zap_negative_locked_vnode_kl(ncp, vp);
2027 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst);
2028 }
2029 mtx_unlock(vlp);
2030 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
2031 cache_free(ncp);
2032 }
2033 }
2034
2035 /*
2036 * Flush all entries referencing a particular filesystem.
2037 */
2038 void
cache_purgevfs(struct mount * mp,bool force)2039 cache_purgevfs(struct mount *mp, bool force)
2040 {
2041 TAILQ_HEAD(, namecache) ncps;
2042 struct mtx *vlp1, *vlp2;
2043 struct rwlock *blp;
2044 struct nchashhead *bucket;
2045 struct namecache *ncp, *nnp;
2046 u_long i, j, n_nchash;
2047 int error;
2048
2049 /* Scan hash tables for applicable entries */
2050 SDT_PROBE1(vfs, namecache, purgevfs, done, mp);
2051 if (!force && mp->mnt_nvnodelistsize <= ncpurgeminvnodes)
2052 return;
2053 TAILQ_INIT(&ncps);
2054 n_nchash = nchash + 1;
2055 vlp1 = vlp2 = NULL;
2056 for (i = 0; i < numbucketlocks; i++) {
2057 blp = (struct rwlock *)&bucketlocks[i];
2058 rw_wlock(blp);
2059 for (j = i; j < n_nchash; j += numbucketlocks) {
2060 retry:
2061 bucket = &nchashtbl[j];
2062 LIST_FOREACH_SAFE(ncp, bucket, nc_hash, nnp) {
2063 cache_assert_bucket_locked(ncp, RA_WLOCKED);
2064 if (ncp->nc_dvp->v_mount != mp)
2065 continue;
2066 error = cache_zap_wlocked_bucket_kl(ncp, blp,
2067 &vlp1, &vlp2);
2068 if (error != 0)
2069 goto retry;
2070 TAILQ_INSERT_HEAD(&ncps, ncp, nc_dst);
2071 }
2072 }
2073 rw_wunlock(blp);
2074 if (vlp1 == NULL && vlp2 == NULL)
2075 cache_maybe_yield();
2076 }
2077 if (vlp1 != NULL)
2078 mtx_unlock(vlp1);
2079 if (vlp2 != NULL)
2080 mtx_unlock(vlp2);
2081
2082 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) {
2083 cache_free(ncp);
2084 }
2085 }
2086
2087 /*
2088 * Perform canonical checks and cache lookup and pass on to filesystem
2089 * through the vop_cachedlookup only if needed.
2090 */
2091
2092 int
vfs_cache_lookup(struct vop_lookup_args * ap)2093 vfs_cache_lookup(struct vop_lookup_args *ap)
2094 {
2095 struct vnode *dvp;
2096 int error;
2097 struct vnode **vpp = ap->a_vpp;
2098 struct componentname *cnp = ap->a_cnp;
2099 int flags = cnp->cn_flags;
2100
2101 *vpp = NULL;
2102 dvp = ap->a_dvp;
2103
2104 if (dvp->v_type != VDIR)
2105 return (ENOTDIR);
2106
2107 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
2108 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
2109 return (EROFS);
2110
2111 error = vn_dir_check_exec(dvp, cnp);
2112 if (error != 0)
2113 return (error);
2114
2115 error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
2116 if (error == 0)
2117 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
2118 if (error == -1)
2119 return (0);
2120 return (error);
2121 }
2122
2123 /*
2124 * XXX All of these sysctls would probably be more productive dead.
2125 */
2126 static int __read_mostly disablecwd;
2127 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
2128 "Disable the getcwd syscall");
2129
2130 /* Implementation of the getcwd syscall. */
2131 int
sys___getcwd(struct thread * td,struct __getcwd_args * uap)2132 sys___getcwd(struct thread *td, struct __getcwd_args *uap)
2133 {
2134
2135 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen,
2136 MAXPATHLEN));
2137 }
2138
2139 int
kern___getcwd(struct thread * td,char * buf,enum uio_seg bufseg,u_int buflen,u_int path_max)2140 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen,
2141 u_int path_max)
2142 {
2143 char *bp, *tmpbuf;
2144 struct filedesc *fdp;
2145 struct vnode *cdir, *rdir;
2146 int error;
2147
2148 if (__predict_false(disablecwd))
2149 return (ENODEV);
2150 if (__predict_false(buflen < 2))
2151 return (EINVAL);
2152 if (buflen > path_max)
2153 buflen = path_max;
2154
2155 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
2156 fdp = td->td_proc->p_fd;
2157 FILEDESC_SLOCK(fdp);
2158 cdir = fdp->fd_cdir;
2159 vrefact(cdir);
2160 rdir = fdp->fd_rdir;
2161 vrefact(rdir);
2162 FILEDESC_SUNLOCK(fdp);
2163 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
2164 vrele(rdir);
2165 vrele(cdir);
2166
2167 if (!error) {
2168 if (bufseg == UIO_SYSSPACE)
2169 bcopy(bp, buf, strlen(bp) + 1);
2170 else
2171 error = copyout(bp, buf, strlen(bp) + 1);
2172 #ifdef KTRACE
2173 if (KTRPOINT(curthread, KTR_NAMEI))
2174 ktrnamei(bp);
2175 #endif
2176 }
2177 free(tmpbuf, M_TEMP);
2178 return (error);
2179 }
2180
2181 /*
2182 * Thus begins the fullpath magic.
2183 */
2184
2185 static int __read_mostly disablefullpath;
2186 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
2187 "Disable the vn_fullpath function");
2188
2189 /*
2190 * Retrieve the full filesystem path that correspond to a vnode from the name
2191 * cache (if available)
2192 */
2193 int
vn_fullpath(struct thread * td,struct vnode * vn,char ** retbuf,char ** freebuf)2194 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
2195 {
2196 char *buf;
2197 struct filedesc *fdp;
2198 struct vnode *rdir;
2199 int error;
2200
2201 if (__predict_false(disablefullpath))
2202 return (ENODEV);
2203 if (__predict_false(vn == NULL))
2204 return (EINVAL);
2205
2206 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
2207 fdp = td->td_proc->p_fd;
2208 FILEDESC_SLOCK(fdp);
2209 rdir = fdp->fd_rdir;
2210 vrefact(rdir);
2211 FILEDESC_SUNLOCK(fdp);
2212 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
2213 vrele(rdir);
2214
2215 if (!error)
2216 *freebuf = buf;
2217 else
2218 free(buf, M_TEMP);
2219 return (error);
2220 }
2221
2222 /*
2223 * This function is similar to vn_fullpath, but it attempts to lookup the
2224 * pathname relative to the global root mount point. This is required for the
2225 * auditing sub-system, as audited pathnames must be absolute, relative to the
2226 * global root mount point.
2227 */
2228 int
vn_fullpath_global(struct thread * td,struct vnode * vn,char ** retbuf,char ** freebuf)2229 vn_fullpath_global(struct thread *td, struct vnode *vn,
2230 char **retbuf, char **freebuf)
2231 {
2232 char *buf;
2233 int error;
2234
2235 if (__predict_false(disablefullpath))
2236 return (ENODEV);
2237 if (__predict_false(vn == NULL))
2238 return (EINVAL);
2239 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
2240 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
2241 if (!error)
2242 *freebuf = buf;
2243 else
2244 free(buf, M_TEMP);
2245 return (error);
2246 }
2247
2248 int
vn_vptocnp(struct vnode ** vp,struct ucred * cred,char * buf,u_int * buflen)2249 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
2250 {
2251 struct vnode *dvp;
2252 struct namecache *ncp;
2253 struct mtx *vlp;
2254 int error;
2255
2256 vlp = VP2VNODELOCK(*vp);
2257 mtx_lock(vlp);
2258 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
2259 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
2260 break;
2261 }
2262 if (ncp != NULL) {
2263 if (*buflen < ncp->nc_nlen) {
2264 mtx_unlock(vlp);
2265 vrele(*vp);
2266 counter_u64_add(numfullpathfail4, 1);
2267 error = ENOMEM;
2268 SDT_PROBE3(vfs, namecache, fullpath, return, error,
2269 vp, NULL);
2270 return (error);
2271 }
2272 *buflen -= ncp->nc_nlen;
2273 memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen);
2274 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp,
2275 ncp->nc_name, vp);
2276 dvp = *vp;
2277 *vp = ncp->nc_dvp;
2278 vref(*vp);
2279 mtx_unlock(vlp);
2280 vrele(dvp);
2281 return (0);
2282 }
2283 SDT_PROBE1(vfs, namecache, fullpath, miss, vp);
2284
2285 mtx_unlock(vlp);
2286 vn_lock(*vp, LK_SHARED | LK_RETRY);
2287 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
2288 vput(*vp);
2289 if (error) {
2290 counter_u64_add(numfullpathfail2, 1);
2291 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
2292 return (error);
2293 }
2294
2295 *vp = dvp;
2296 if (dvp->v_iflag & VI_DOOMED) {
2297 /* forced unmount */
2298 vrele(dvp);
2299 error = ENOENT;
2300 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
2301 return (error);
2302 }
2303 /*
2304 * *vp has its use count incremented still.
2305 */
2306
2307 return (0);
2308 }
2309
2310 /*
2311 * The magic behind kern___getcwd() and vn_fullpath().
2312 */
2313 static int
vn_fullpath1(struct thread * td,struct vnode * vp,struct vnode * rdir,char * buf,char ** retbuf,u_int buflen)2314 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
2315 char *buf, char **retbuf, u_int buflen)
2316 {
2317 int error, slash_prefixed;
2318 #ifdef KDTRACE_HOOKS
2319 struct vnode *startvp = vp;
2320 #endif
2321 struct vnode *vp1;
2322
2323 buflen--;
2324 buf[buflen] = '\0';
2325 error = 0;
2326 slash_prefixed = 0;
2327
2328 SDT_PROBE1(vfs, namecache, fullpath, entry, vp);
2329 counter_u64_add(numfullpathcalls, 1);
2330 vref(vp);
2331 if (vp->v_type != VDIR) {
2332 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen);
2333 if (error)
2334 return (error);
2335 if (buflen == 0) {
2336 vrele(vp);
2337 return (ENOMEM);
2338 }
2339 buf[--buflen] = '/';
2340 slash_prefixed = 1;
2341 }
2342 while (vp != rdir && vp != rootvnode) {
2343 /*
2344 * The vp vnode must be already fully constructed,
2345 * since it is either found in namecache or obtained
2346 * from VOP_VPTOCNP(). We may test for VV_ROOT safely
2347 * without obtaining the vnode lock.
2348 */
2349 if ((vp->v_vflag & VV_ROOT) != 0) {
2350 vn_lock(vp, LK_RETRY | LK_SHARED);
2351
2352 /*
2353 * With the vnode locked, check for races with
2354 * unmount, forced or not. Note that we
2355 * already verified that vp is not equal to
2356 * the root vnode, which means that
2357 * mnt_vnodecovered can be NULL only for the
2358 * case of unmount.
2359 */
2360 if ((vp->v_iflag & VI_DOOMED) != 0 ||
2361 (vp1 = vp->v_mount->mnt_vnodecovered) == NULL ||
2362 vp1->v_mountedhere != vp->v_mount) {
2363 vput(vp);
2364 error = ENOENT;
2365 SDT_PROBE3(vfs, namecache, fullpath, return,
2366 error, vp, NULL);
2367 break;
2368 }
2369
2370 vref(vp1);
2371 vput(vp);
2372 vp = vp1;
2373 continue;
2374 }
2375 if (vp->v_type != VDIR) {
2376 vrele(vp);
2377 counter_u64_add(numfullpathfail1, 1);
2378 error = ENOTDIR;
2379 SDT_PROBE3(vfs, namecache, fullpath, return,
2380 error, vp, NULL);
2381 break;
2382 }
2383 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen);
2384 if (error)
2385 break;
2386 if (buflen == 0) {
2387 vrele(vp);
2388 error = ENOMEM;
2389 SDT_PROBE3(vfs, namecache, fullpath, return, error,
2390 startvp, NULL);
2391 break;
2392 }
2393 buf[--buflen] = '/';
2394 slash_prefixed = 1;
2395 }
2396 if (error)
2397 return (error);
2398 if (!slash_prefixed) {
2399 if (buflen == 0) {
2400 vrele(vp);
2401 counter_u64_add(numfullpathfail4, 1);
2402 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM,
2403 startvp, NULL);
2404 return (ENOMEM);
2405 }
2406 buf[--buflen] = '/';
2407 }
2408 counter_u64_add(numfullpathfound, 1);
2409 vrele(vp);
2410
2411 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen);
2412 *retbuf = buf + buflen;
2413 return (0);
2414 }
2415
2416 struct vnode *
vn_dir_dd_ino(struct vnode * vp)2417 vn_dir_dd_ino(struct vnode *vp)
2418 {
2419 struct namecache *ncp;
2420 struct vnode *ddvp;
2421 struct mtx *vlp;
2422
2423 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
2424 vlp = VP2VNODELOCK(vp);
2425 mtx_lock(vlp);
2426 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
2427 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
2428 continue;
2429 ddvp = ncp->nc_dvp;
2430 vhold(ddvp);
2431 mtx_unlock(vlp);
2432 if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread))
2433 return (NULL);
2434 return (ddvp);
2435 }
2436 mtx_unlock(vlp);
2437 return (NULL);
2438 }
2439
2440 int
vn_commname(struct vnode * vp,char * buf,u_int buflen)2441 vn_commname(struct vnode *vp, char *buf, u_int buflen)
2442 {
2443 struct namecache *ncp;
2444 struct mtx *vlp;
2445 int l;
2446
2447 vlp = VP2VNODELOCK(vp);
2448 mtx_lock(vlp);
2449 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
2450 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
2451 break;
2452 if (ncp == NULL) {
2453 mtx_unlock(vlp);
2454 return (ENOENT);
2455 }
2456 l = min(ncp->nc_nlen, buflen - 1);
2457 memcpy(buf, ncp->nc_name, l);
2458 mtx_unlock(vlp);
2459 buf[l] = '\0';
2460 return (0);
2461 }
2462
2463 /* ABI compat shims for old kernel modules. */
2464 #undef cache_enter
2465
2466 void cache_enter(struct vnode *dvp, struct vnode *vp,
2467 struct componentname *cnp);
2468
2469 void
cache_enter(struct vnode * dvp,struct vnode * vp,struct componentname * cnp)2470 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2471 {
2472
2473 cache_enter_time(dvp, vp, cnp, NULL, NULL);
2474 }
2475
2476 /*
2477 * This function updates path string to vnode's full global path
2478 * and checks the size of the new path string against the pathlen argument.
2479 *
2480 * Requires a locked, referenced vnode.
2481 * Vnode is re-locked on success or ENODEV, otherwise unlocked.
2482 *
2483 * If sysctl debug.disablefullpath is set, ENODEV is returned,
2484 * vnode is left locked and path remain untouched.
2485 *
2486 * If vp is a directory, the call to vn_fullpath_global() always succeeds
2487 * because it falls back to the ".." lookup if the namecache lookup fails.
2488 */
2489 int
vn_path_to_global_path(struct thread * td,struct vnode * vp,char * path,u_int pathlen)2490 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
2491 u_int pathlen)
2492 {
2493 struct nameidata nd;
2494 struct vnode *vp1;
2495 char *rpath, *fbuf;
2496 int error;
2497
2498 ASSERT_VOP_ELOCKED(vp, __func__);
2499
2500 /* Return ENODEV if sysctl debug.disablefullpath==1 */
2501 if (__predict_false(disablefullpath))
2502 return (ENODEV);
2503
2504 /* Construct global filesystem path from vp. */
2505 VOP_UNLOCK(vp, 0);
2506 error = vn_fullpath_global(td, vp, &rpath, &fbuf);
2507
2508 if (error != 0) {
2509 vrele(vp);
2510 return (error);
2511 }
2512
2513 if (strlen(rpath) >= pathlen) {
2514 vrele(vp);
2515 error = ENAMETOOLONG;
2516 goto out;
2517 }
2518
2519 /*
2520 * Re-lookup the vnode by path to detect a possible rename.
2521 * As a side effect, the vnode is relocked.
2522 * If vnode was renamed, return ENOENT.
2523 */
2524 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
2525 UIO_SYSSPACE, path, td);
2526 error = namei(&nd);
2527 if (error != 0) {
2528 vrele(vp);
2529 goto out;
2530 }
2531 NDFREE(&nd, NDF_ONLY_PNBUF);
2532 vp1 = nd.ni_vp;
2533 vrele(vp);
2534 if (vp1 == vp)
2535 strcpy(path, rpath);
2536 else {
2537 vput(vp1);
2538 error = ENOENT;
2539 }
2540
2541 out:
2542 free(fbuf, M_TEMP);
2543 return (error);
2544 }
2545