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 (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 = 16; /* 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_rlocked_bucket(struct namecache *ncp, struct rwlock *blp) 988 { 989 struct mtx *dvlp, *vlp; 990 991 cache_assert_bucket_locked(ncp, RA_RLOCKED); 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 rw_runlock(blp); 999 rw_wlock(blp); 1000 cache_zap_locked(ncp, false); 1001 rw_wunlock(blp); 1002 cache_unlock_vnodes(dvlp, vlp); 1003 return (0); 1004 } 1005 1006 rw_runlock(blp); 1007 return (EAGAIN); 1008 } 1009 1010 static int 1011 cache_zap_wlocked_bucket_kl(struct namecache *ncp, struct rwlock *blp, 1012 struct mtx **vlpp1, struct mtx **vlpp2) 1013 { 1014 struct mtx *dvlp, *vlp; 1015 1016 cache_assert_bucket_locked(ncp, RA_WLOCKED); 1017 1018 dvlp = VP2VNODELOCK(ncp->nc_dvp); 1019 vlp = NULL; 1020 if (!(ncp->nc_flag & NCF_NEGATIVE)) 1021 vlp = VP2VNODELOCK(ncp->nc_vp); 1022 cache_sort(&dvlp, &vlp); 1023 1024 if (*vlpp1 == dvlp && *vlpp2 == vlp) { 1025 cache_zap_locked(ncp, false); 1026 cache_unlock_vnodes(dvlp, vlp); 1027 *vlpp1 = NULL; 1028 *vlpp2 = NULL; 1029 return (0); 1030 } 1031 1032 if (*vlpp1 != NULL) 1033 mtx_unlock(*vlpp1); 1034 if (*vlpp2 != NULL) 1035 mtx_unlock(*vlpp2); 1036 *vlpp1 = NULL; 1037 *vlpp2 = NULL; 1038 1039 if (cache_trylock_vnodes(dvlp, vlp) == 0) { 1040 cache_zap_locked(ncp, false); 1041 cache_unlock_vnodes(dvlp, vlp); 1042 return (0); 1043 } 1044 1045 rw_wunlock(blp); 1046 *vlpp1 = dvlp; 1047 *vlpp2 = vlp; 1048 if (*vlpp1 != NULL) 1049 mtx_lock(*vlpp1); 1050 mtx_lock(*vlpp2); 1051 rw_wlock(blp); 1052 return (EAGAIN); 1053 } 1054 1055 static void 1056 cache_lookup_unlock(struct rwlock *blp, struct mtx *vlp) 1057 { 1058 1059 if (blp != NULL) { 1060 rw_runlock(blp); 1061 mtx_assert(vlp, MA_NOTOWNED); 1062 } else { 1063 mtx_unlock(vlp); 1064 } 1065 } 1066 1067 static int __noinline 1068 cache_lookup_dot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1069 struct timespec *tsp, int *ticksp) 1070 { 1071 int ltype; 1072 1073 *vpp = dvp; 1074 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .", 1075 dvp, cnp->cn_nameptr); 1076 counter_u64_add(dothits, 1); 1077 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp); 1078 if (tsp != NULL) 1079 timespecclear(tsp); 1080 if (ticksp != NULL) 1081 *ticksp = ticks; 1082 vrefact(*vpp); 1083 /* 1084 * When we lookup "." we still can be asked to lock it 1085 * differently... 1086 */ 1087 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 1088 if (ltype != VOP_ISLOCKED(*vpp)) { 1089 if (ltype == LK_EXCLUSIVE) { 1090 vn_lock(*vpp, LK_UPGRADE | LK_RETRY); 1091 if ((*vpp)->v_iflag & VI_DOOMED) { 1092 /* forced unmount */ 1093 vrele(*vpp); 1094 *vpp = NULL; 1095 return (ENOENT); 1096 } 1097 } else 1098 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); 1099 } 1100 return (-1); 1101 } 1102 1103 /* 1104 * Lookup an entry in the cache 1105 * 1106 * Lookup is called with dvp pointing to the directory to search, 1107 * cnp pointing to the name of the entry being sought. If the lookup 1108 * succeeds, the vnode is returned in *vpp, and a status of -1 is 1109 * returned. If the lookup determines that the name does not exist 1110 * (negative caching), a status of ENOENT is returned. If the lookup 1111 * fails, a status of zero is returned. If the directory vnode is 1112 * recycled out from under us due to a forced unmount, a status of 1113 * ENOENT is returned. 1114 * 1115 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is 1116 * unlocked. If we're looking up . an extra ref is taken, but the lock is 1117 * not recursively acquired. 1118 */ 1119 1120 int 1121 cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1122 struct timespec *tsp, int *ticksp) 1123 { 1124 struct namecache_ts *ncp_ts; 1125 struct namecache *ncp; 1126 struct rwlock *blp; 1127 struct mtx *dvlp, *dvlp2; 1128 uint32_t hash; 1129 int error, ltype; 1130 1131 if (__predict_false(!doingcache)) { 1132 cnp->cn_flags &= ~MAKEENTRY; 1133 return (0); 1134 } 1135 retry: 1136 blp = NULL; 1137 dvlp = VP2VNODELOCK(dvp); 1138 error = 0; 1139 counter_u64_add(numcalls, 1); 1140 1141 if (cnp->cn_nameptr[0] == '.') { 1142 if (cnp->cn_namelen == 1) 1143 return (cache_lookup_dot(dvp, vpp, cnp, tsp, ticksp)); 1144 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 1145 counter_u64_add(dotdothits, 1); 1146 dvlp2 = NULL; 1147 mtx_lock(dvlp); 1148 retry_dotdot: 1149 ncp = dvp->v_cache_dd; 1150 if (ncp == NULL) { 1151 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, 1152 "..", NULL); 1153 mtx_unlock(dvlp); 1154 return (0); 1155 } 1156 if ((cnp->cn_flags & MAKEENTRY) == 0) { 1157 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { 1158 if (ncp->nc_dvp != dvp) 1159 panic("dvp %p v_cache_dd %p\n", dvp, ncp); 1160 if (!cache_zap_locked_vnode_kl2(ncp, 1161 dvp, &dvlp2)) 1162 goto retry_dotdot; 1163 MPASS(dvp->v_cache_dd == NULL); 1164 mtx_unlock(dvlp); 1165 if (dvlp2 != NULL) 1166 mtx_unlock(dvlp2); 1167 cache_free(ncp); 1168 } else { 1169 dvp->v_cache_dd = NULL; 1170 mtx_unlock(dvlp); 1171 if (dvlp2 != NULL) 1172 mtx_unlock(dvlp2); 1173 } 1174 return (0); 1175 } 1176 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { 1177 if (ncp->nc_flag & NCF_NEGATIVE) 1178 *vpp = NULL; 1179 else 1180 *vpp = ncp->nc_vp; 1181 } else 1182 *vpp = ncp->nc_dvp; 1183 /* Return failure if negative entry was found. */ 1184 if (*vpp == NULL) 1185 goto negative_success; 1186 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..", 1187 dvp, cnp->cn_nameptr, *vpp); 1188 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..", 1189 *vpp); 1190 cache_out_ts(ncp, tsp, ticksp); 1191 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == 1192 NCF_DTS && tsp != NULL) { 1193 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); 1194 *tsp = ncp_ts->nc_dotdottime; 1195 } 1196 goto success; 1197 } 1198 } 1199 1200 hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); 1201 blp = HASH2BUCKETLOCK(hash); 1202 rw_rlock(blp); 1203 1204 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { 1205 counter_u64_add(numchecks, 1); 1206 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && 1207 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) 1208 break; 1209 } 1210 1211 /* We failed to find an entry */ 1212 if (ncp == NULL) { 1213 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, 1214 NULL); 1215 if ((cnp->cn_flags & MAKEENTRY) == 0) { 1216 counter_u64_add(nummisszap, 1); 1217 } else { 1218 counter_u64_add(nummiss, 1); 1219 } 1220 cache_lookup_unlock(blp, dvlp); 1221 return (0); 1222 } 1223 1224 /* We don't want to have an entry, so dump it */ 1225 if ((cnp->cn_flags & MAKEENTRY) == 0) { 1226 counter_u64_add(numposzaps, 1); 1227 goto zap_and_exit; 1228 } 1229 1230 /* We found a "positive" match, return the vnode */ 1231 if (!(ncp->nc_flag & NCF_NEGATIVE)) { 1232 counter_u64_add(numposhits, 1); 1233 *vpp = ncp->nc_vp; 1234 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p", 1235 dvp, cnp->cn_nameptr, *vpp, ncp); 1236 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, 1237 *vpp); 1238 cache_out_ts(ncp, tsp, ticksp); 1239 goto success; 1240 } 1241 1242 negative_success: 1243 /* We found a negative match, and want to create it, so purge */ 1244 if (cnp->cn_nameiop == CREATE) { 1245 counter_u64_add(numnegzaps, 1); 1246 goto zap_and_exit; 1247 } 1248 1249 counter_u64_add(numneghits, 1); 1250 cache_negative_hit(ncp); 1251 if (ncp->nc_flag & NCF_WHITE) 1252 cnp->cn_flags |= ISWHITEOUT; 1253 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, 1254 ncp->nc_name); 1255 cache_out_ts(ncp, tsp, ticksp); 1256 cache_lookup_unlock(blp, dvlp); 1257 return (ENOENT); 1258 1259 success: 1260 /* 1261 * On success we return a locked and ref'd vnode as per the lookup 1262 * protocol. 1263 */ 1264 MPASS(dvp != *vpp); 1265 ltype = 0; /* silence gcc warning */ 1266 if (cnp->cn_flags & ISDOTDOT) { 1267 ltype = VOP_ISLOCKED(dvp); 1268 VOP_UNLOCK(dvp, 0); 1269 } 1270 vhold(*vpp); 1271 cache_lookup_unlock(blp, dvlp); 1272 error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread); 1273 if (cnp->cn_flags & ISDOTDOT) { 1274 vn_lock(dvp, ltype | LK_RETRY); 1275 if (dvp->v_iflag & VI_DOOMED) { 1276 if (error == 0) 1277 vput(*vpp); 1278 *vpp = NULL; 1279 return (ENOENT); 1280 } 1281 } 1282 if (error) { 1283 *vpp = NULL; 1284 goto retry; 1285 } 1286 if ((cnp->cn_flags & ISLASTCN) && 1287 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) { 1288 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup"); 1289 } 1290 return (-1); 1291 1292 zap_and_exit: 1293 if (blp != NULL) 1294 error = cache_zap_rlocked_bucket(ncp, blp); 1295 else 1296 error = cache_zap_locked_vnode(ncp, dvp); 1297 if (error != 0) { 1298 zap_and_exit_bucket_fail++; 1299 cache_maybe_yield(); 1300 goto retry; 1301 } 1302 cache_free(ncp); 1303 return (0); 1304 } 1305 1306 struct celockstate { 1307 struct mtx *vlp[3]; 1308 struct rwlock *blp[2]; 1309 }; 1310 CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3)); 1311 CTASSERT((nitems(((struct celockstate *)0)->blp) == 2)); 1312 1313 static inline void 1314 cache_celockstate_init(struct celockstate *cel) 1315 { 1316 1317 bzero(cel, sizeof(*cel)); 1318 } 1319 1320 static void 1321 cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp, 1322 struct vnode *dvp) 1323 { 1324 struct mtx *vlp1, *vlp2; 1325 1326 MPASS(cel->vlp[0] == NULL); 1327 MPASS(cel->vlp[1] == NULL); 1328 MPASS(cel->vlp[2] == NULL); 1329 1330 MPASS(vp != NULL || dvp != NULL); 1331 1332 vlp1 = VP2VNODELOCK(vp); 1333 vlp2 = VP2VNODELOCK(dvp); 1334 cache_sort(&vlp1, &vlp2); 1335 1336 if (vlp1 != NULL) { 1337 mtx_lock(vlp1); 1338 cel->vlp[0] = vlp1; 1339 } 1340 mtx_lock(vlp2); 1341 cel->vlp[1] = vlp2; 1342 } 1343 1344 static void 1345 cache_unlock_vnodes_cel(struct celockstate *cel) 1346 { 1347 1348 MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL); 1349 1350 if (cel->vlp[0] != NULL) 1351 mtx_unlock(cel->vlp[0]); 1352 if (cel->vlp[1] != NULL) 1353 mtx_unlock(cel->vlp[1]); 1354 if (cel->vlp[2] != NULL) 1355 mtx_unlock(cel->vlp[2]); 1356 } 1357 1358 static bool 1359 cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp) 1360 { 1361 struct mtx *vlp; 1362 bool ret; 1363 1364 cache_assert_vlp_locked(cel->vlp[0]); 1365 cache_assert_vlp_locked(cel->vlp[1]); 1366 MPASS(cel->vlp[2] == NULL); 1367 1368 MPASS(vp != NULL); 1369 vlp = VP2VNODELOCK(vp); 1370 1371 ret = true; 1372 if (vlp >= cel->vlp[1]) { 1373 mtx_lock(vlp); 1374 } else { 1375 if (mtx_trylock(vlp)) 1376 goto out; 1377 cache_lock_vnodes_cel_3_failures++; 1378 cache_unlock_vnodes_cel(cel); 1379 if (vlp < cel->vlp[0]) { 1380 mtx_lock(vlp); 1381 mtx_lock(cel->vlp[0]); 1382 mtx_lock(cel->vlp[1]); 1383 } else { 1384 if (cel->vlp[0] != NULL) 1385 mtx_lock(cel->vlp[0]); 1386 mtx_lock(vlp); 1387 mtx_lock(cel->vlp[1]); 1388 } 1389 ret = false; 1390 } 1391 out: 1392 cel->vlp[2] = vlp; 1393 return (ret); 1394 } 1395 1396 static void 1397 cache_lock_buckets_cel(struct celockstate *cel, struct rwlock *blp1, 1398 struct rwlock *blp2) 1399 { 1400 1401 MPASS(cel->blp[0] == NULL); 1402 MPASS(cel->blp[1] == NULL); 1403 1404 cache_sort(&blp1, &blp2); 1405 1406 if (blp1 != NULL) { 1407 rw_wlock(blp1); 1408 cel->blp[0] = blp1; 1409 } 1410 rw_wlock(blp2); 1411 cel->blp[1] = blp2; 1412 } 1413 1414 static void 1415 cache_unlock_buckets_cel(struct celockstate *cel) 1416 { 1417 1418 if (cel->blp[0] != NULL) 1419 rw_wunlock(cel->blp[0]); 1420 rw_wunlock(cel->blp[1]); 1421 } 1422 1423 /* 1424 * Lock part of the cache affected by the insertion. 1425 * 1426 * This means vnodelocks for dvp, vp and the relevant bucketlock. 1427 * However, insertion can result in removal of an old entry. In this 1428 * case we have an additional vnode and bucketlock pair to lock. If the 1429 * entry is negative, ncelock is locked instead of the vnode. 1430 * 1431 * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while 1432 * preserving the locking order (smaller address first). 1433 */ 1434 static void 1435 cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, 1436 uint32_t hash) 1437 { 1438 struct namecache *ncp; 1439 struct rwlock *blps[2]; 1440 1441 blps[0] = HASH2BUCKETLOCK(hash); 1442 for (;;) { 1443 blps[1] = NULL; 1444 cache_lock_vnodes_cel(cel, dvp, vp); 1445 if (vp == NULL || vp->v_type != VDIR) 1446 break; 1447 ncp = vp->v_cache_dd; 1448 if (ncp == NULL) 1449 break; 1450 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 1451 break; 1452 MPASS(ncp->nc_dvp == vp); 1453 blps[1] = NCP2BUCKETLOCK(ncp); 1454 if (ncp->nc_flag & NCF_NEGATIVE) 1455 break; 1456 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) 1457 break; 1458 /* 1459 * All vnodes got re-locked. Re-validate the state and if 1460 * nothing changed we are done. Otherwise restart. 1461 */ 1462 if (ncp == vp->v_cache_dd && 1463 (ncp->nc_flag & NCF_ISDOTDOT) != 0 && 1464 blps[1] == NCP2BUCKETLOCK(ncp) && 1465 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) 1466 break; 1467 cache_unlock_vnodes_cel(cel); 1468 cel->vlp[0] = NULL; 1469 cel->vlp[1] = NULL; 1470 cel->vlp[2] = NULL; 1471 } 1472 cache_lock_buckets_cel(cel, blps[0], blps[1]); 1473 } 1474 1475 static void 1476 cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, 1477 uint32_t hash) 1478 { 1479 struct namecache *ncp; 1480 struct rwlock *blps[2]; 1481 1482 blps[0] = HASH2BUCKETLOCK(hash); 1483 for (;;) { 1484 blps[1] = NULL; 1485 cache_lock_vnodes_cel(cel, dvp, vp); 1486 ncp = dvp->v_cache_dd; 1487 if (ncp == NULL) 1488 break; 1489 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 1490 break; 1491 MPASS(ncp->nc_dvp == dvp); 1492 blps[1] = NCP2BUCKETLOCK(ncp); 1493 if (ncp->nc_flag & NCF_NEGATIVE) 1494 break; 1495 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) 1496 break; 1497 if (ncp == dvp->v_cache_dd && 1498 (ncp->nc_flag & NCF_ISDOTDOT) != 0 && 1499 blps[1] == NCP2BUCKETLOCK(ncp) && 1500 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) 1501 break; 1502 cache_unlock_vnodes_cel(cel); 1503 cel->vlp[0] = NULL; 1504 cel->vlp[1] = NULL; 1505 cel->vlp[2] = NULL; 1506 } 1507 cache_lock_buckets_cel(cel, blps[0], blps[1]); 1508 } 1509 1510 static void 1511 cache_enter_unlock(struct celockstate *cel) 1512 { 1513 1514 cache_unlock_buckets_cel(cel); 1515 cache_unlock_vnodes_cel(cel); 1516 } 1517 1518 /* 1519 * Add an entry to the cache. 1520 */ 1521 void 1522 cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, 1523 struct timespec *tsp, struct timespec *dtsp) 1524 { 1525 struct celockstate cel; 1526 struct namecache *ncp, *n2, *ndd; 1527 struct namecache_ts *ncp_ts, *n2_ts; 1528 struct nchashhead *ncpp; 1529 struct neglist *neglist; 1530 uint32_t hash; 1531 int flag; 1532 int len; 1533 bool neg_locked; 1534 1535 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr); 1536 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp, 1537 ("cache_enter: Adding a doomed vnode")); 1538 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp, 1539 ("cache_enter: Doomed vnode used as src")); 1540 1541 if (__predict_false(!doingcache)) 1542 return; 1543 1544 /* 1545 * Avoid blowout in namecache entries. 1546 */ 1547 if (__predict_false(numcache >= desiredvnodes * ncsizefactor)) 1548 return; 1549 1550 cache_celockstate_init(&cel); 1551 ndd = NULL; 1552 flag = 0; 1553 if (cnp->cn_nameptr[0] == '.') { 1554 if (cnp->cn_namelen == 1) 1555 return; 1556 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 1557 len = cnp->cn_namelen; 1558 hash = cache_get_hash(cnp->cn_nameptr, len, dvp); 1559 cache_enter_lock_dd(&cel, dvp, vp, hash); 1560 /* 1561 * If dotdot entry already exists, just retarget it 1562 * to new parent vnode, otherwise continue with new 1563 * namecache entry allocation. 1564 */ 1565 if ((ncp = dvp->v_cache_dd) != NULL && 1566 ncp->nc_flag & NCF_ISDOTDOT) { 1567 KASSERT(ncp->nc_dvp == dvp, 1568 ("wrong isdotdot parent")); 1569 neg_locked = false; 1570 if (ncp->nc_flag & NCF_NEGATIVE || vp == NULL) { 1571 neglist = NCP2NEGLIST(ncp); 1572 mtx_lock(&ncneg_hot.nl_lock); 1573 mtx_lock(&neglist->nl_lock); 1574 neg_locked = true; 1575 } 1576 if (!(ncp->nc_flag & NCF_NEGATIVE)) { 1577 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, 1578 ncp, nc_dst); 1579 } else { 1580 cache_negative_remove(ncp, true); 1581 } 1582 if (vp != NULL) { 1583 TAILQ_INSERT_HEAD(&vp->v_cache_dst, 1584 ncp, nc_dst); 1585 ncp->nc_flag &= ~(NCF_NEGATIVE|NCF_HOTNEGATIVE); 1586 } else { 1587 ncp->nc_flag &= ~(NCF_HOTNEGATIVE); 1588 ncp->nc_flag |= NCF_NEGATIVE; 1589 cache_negative_insert(ncp, true); 1590 } 1591 if (neg_locked) { 1592 mtx_unlock(&neglist->nl_lock); 1593 mtx_unlock(&ncneg_hot.nl_lock); 1594 } 1595 ncp->nc_vp = vp; 1596 cache_enter_unlock(&cel); 1597 return; 1598 } 1599 dvp->v_cache_dd = NULL; 1600 cache_enter_unlock(&cel); 1601 cache_celockstate_init(&cel); 1602 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp); 1603 flag = NCF_ISDOTDOT; 1604 } 1605 } 1606 1607 /* 1608 * Calculate the hash key and setup as much of the new 1609 * namecache entry as possible before acquiring the lock. 1610 */ 1611 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); 1612 ncp->nc_flag = flag; 1613 ncp->nc_vp = vp; 1614 if (vp == NULL) 1615 ncp->nc_flag |= NCF_NEGATIVE; 1616 ncp->nc_dvp = dvp; 1617 if (tsp != NULL) { 1618 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); 1619 ncp_ts->nc_time = *tsp; 1620 ncp_ts->nc_ticks = ticks; 1621 ncp_ts->nc_nc.nc_flag |= NCF_TS; 1622 if (dtsp != NULL) { 1623 ncp_ts->nc_dotdottime = *dtsp; 1624 ncp_ts->nc_nc.nc_flag |= NCF_DTS; 1625 } 1626 } 1627 len = ncp->nc_nlen = cnp->cn_namelen; 1628 hash = cache_get_hash(cnp->cn_nameptr, len, dvp); 1629 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1); 1630 cache_enter_lock(&cel, dvp, vp, hash); 1631 1632 /* 1633 * See if this vnode or negative entry is already in the cache 1634 * with this name. This can happen with concurrent lookups of 1635 * the same path name. 1636 */ 1637 ncpp = NCHHASH(hash); 1638 LIST_FOREACH(n2, ncpp, nc_hash) { 1639 if (n2->nc_dvp == dvp && 1640 n2->nc_nlen == cnp->cn_namelen && 1641 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) { 1642 if (tsp != NULL) { 1643 KASSERT((n2->nc_flag & NCF_TS) != 0, 1644 ("no NCF_TS")); 1645 n2_ts = __containerof(n2, struct namecache_ts, nc_nc); 1646 n2_ts->nc_time = ncp_ts->nc_time; 1647 n2_ts->nc_ticks = ncp_ts->nc_ticks; 1648 if (dtsp != NULL) { 1649 n2_ts->nc_dotdottime = ncp_ts->nc_dotdottime; 1650 if (ncp->nc_flag & NCF_NEGATIVE) 1651 mtx_lock(&ncneg_hot.nl_lock); 1652 n2_ts->nc_nc.nc_flag |= NCF_DTS; 1653 if (ncp->nc_flag & NCF_NEGATIVE) 1654 mtx_unlock(&ncneg_hot.nl_lock); 1655 } 1656 } 1657 goto out_unlock_free; 1658 } 1659 } 1660 1661 if (flag == NCF_ISDOTDOT) { 1662 /* 1663 * See if we are trying to add .. entry, but some other lookup 1664 * has populated v_cache_dd pointer already. 1665 */ 1666 if (dvp->v_cache_dd != NULL) 1667 goto out_unlock_free; 1668 KASSERT(vp == NULL || vp->v_type == VDIR, 1669 ("wrong vnode type %p", vp)); 1670 dvp->v_cache_dd = ncp; 1671 } 1672 1673 atomic_add_rel_long(&numcache, 1); 1674 if (vp != NULL) { 1675 if (vp->v_type == VDIR) { 1676 if (flag != NCF_ISDOTDOT) { 1677 /* 1678 * For this case, the cache entry maps both the 1679 * directory name in it and the name ".." for the 1680 * directory's parent. 1681 */ 1682 if ((ndd = vp->v_cache_dd) != NULL) { 1683 if ((ndd->nc_flag & NCF_ISDOTDOT) != 0) 1684 cache_zap_locked(ndd, false); 1685 else 1686 ndd = NULL; 1687 } 1688 vp->v_cache_dd = ncp; 1689 } 1690 } else { 1691 vp->v_cache_dd = NULL; 1692 } 1693 } 1694 1695 if (flag != NCF_ISDOTDOT) { 1696 if (LIST_EMPTY(&dvp->v_cache_src)) { 1697 vhold(dvp); 1698 atomic_add_rel_long(&numcachehv, 1); 1699 } 1700 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); 1701 } 1702 1703 /* 1704 * Insert the new namecache entry into the appropriate chain 1705 * within the cache entries table. 1706 */ 1707 LIST_INSERT_HEAD(ncpp, ncp, nc_hash); 1708 1709 /* 1710 * If the entry is "negative", we place it into the 1711 * "negative" cache queue, otherwise, we place it into the 1712 * destination vnode's cache entries queue. 1713 */ 1714 if (vp != NULL) { 1715 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); 1716 SDT_PROBE3(vfs, namecache, enter, done, dvp, ncp->nc_name, 1717 vp); 1718 } else { 1719 if (cnp->cn_flags & ISWHITEOUT) 1720 ncp->nc_flag |= NCF_WHITE; 1721 cache_negative_insert(ncp, false); 1722 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp, 1723 ncp->nc_name); 1724 } 1725 cache_enter_unlock(&cel); 1726 if (numneg * ncnegfactor > numcache) 1727 cache_negative_zap_one(); 1728 cache_free(ndd); 1729 return; 1730 out_unlock_free: 1731 cache_enter_unlock(&cel); 1732 cache_free(ncp); 1733 return; 1734 } 1735 1736 static u_int 1737 cache_roundup_2(u_int val) 1738 { 1739 u_int res; 1740 1741 for (res = 1; res <= val; res <<= 1) 1742 continue; 1743 1744 return (res); 1745 } 1746 1747 /* 1748 * Name cache initialization, from vfs_init() when we are booting 1749 */ 1750 static void 1751 nchinit(void *dummy __unused) 1752 { 1753 u_int i; 1754 1755 cache_zone_small = uma_zcreate("S VFS Cache", 1756 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1, 1757 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache), 1758 UMA_ZONE_ZINIT); 1759 cache_zone_small_ts = uma_zcreate("STS VFS Cache", 1760 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1, 1761 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts), 1762 UMA_ZONE_ZINIT); 1763 cache_zone_large = uma_zcreate("L VFS Cache", 1764 sizeof(struct namecache) + NAME_MAX + 1, 1765 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache), 1766 UMA_ZONE_ZINIT); 1767 cache_zone_large_ts = uma_zcreate("LTS VFS Cache", 1768 sizeof(struct namecache_ts) + NAME_MAX + 1, 1769 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts), 1770 UMA_ZONE_ZINIT); 1771 1772 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash); 1773 ncbuckethash = cache_roundup_2(mp_ncpus * 64) - 1; 1774 if (ncbuckethash > nchash) 1775 ncbuckethash = nchash; 1776 bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE, 1777 M_WAITOK | M_ZERO); 1778 for (i = 0; i < numbucketlocks; i++) 1779 rw_init_flags(&bucketlocks[i], "ncbuc", RW_DUPOK | RW_RECURSE); 1780 ncvnodehash = cache_roundup_2(mp_ncpus * 64) - 1; 1781 vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE, 1782 M_WAITOK | M_ZERO); 1783 for (i = 0; i < numvnodelocks; i++) 1784 mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE); 1785 ncpurgeminvnodes = numbucketlocks; 1786 1787 ncneghash = 3; 1788 neglists = malloc(sizeof(*neglists) * numneglists, M_VFSCACHE, 1789 M_WAITOK | M_ZERO); 1790 for (i = 0; i < numneglists; i++) { 1791 mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF); 1792 TAILQ_INIT(&neglists[i].nl_list); 1793 } 1794 mtx_init(&ncneg_hot.nl_lock, "ncneglh", NULL, MTX_DEF); 1795 TAILQ_INIT(&ncneg_hot.nl_list); 1796 1797 mtx_init(&ncneg_shrink_lock, "ncnegs", NULL, MTX_DEF); 1798 1799 numcalls = counter_u64_alloc(M_WAITOK); 1800 dothits = counter_u64_alloc(M_WAITOK); 1801 dotdothits = counter_u64_alloc(M_WAITOK); 1802 numchecks = counter_u64_alloc(M_WAITOK); 1803 nummiss = counter_u64_alloc(M_WAITOK); 1804 nummisszap = counter_u64_alloc(M_WAITOK); 1805 numposzaps = counter_u64_alloc(M_WAITOK); 1806 numposhits = counter_u64_alloc(M_WAITOK); 1807 numnegzaps = counter_u64_alloc(M_WAITOK); 1808 numneghits = counter_u64_alloc(M_WAITOK); 1809 numfullpathcalls = counter_u64_alloc(M_WAITOK); 1810 numfullpathfail1 = counter_u64_alloc(M_WAITOK); 1811 numfullpathfail2 = counter_u64_alloc(M_WAITOK); 1812 numfullpathfail4 = counter_u64_alloc(M_WAITOK); 1813 numfullpathfound = counter_u64_alloc(M_WAITOK); 1814 } 1815 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); 1816 1817 void 1818 cache_changesize(int newmaxvnodes) 1819 { 1820 struct nchashhead *new_nchashtbl, *old_nchashtbl; 1821 u_long new_nchash, old_nchash; 1822 struct namecache *ncp; 1823 uint32_t hash; 1824 int i; 1825 1826 newmaxvnodes = cache_roundup_2(newmaxvnodes * 2); 1827 if (newmaxvnodes < numbucketlocks) 1828 newmaxvnodes = numbucketlocks; 1829 1830 new_nchashtbl = hashinit(newmaxvnodes, M_VFSCACHE, &new_nchash); 1831 /* If same hash table size, nothing to do */ 1832 if (nchash == new_nchash) { 1833 free(new_nchashtbl, M_VFSCACHE); 1834 return; 1835 } 1836 /* 1837 * Move everything from the old hash table to the new table. 1838 * None of the namecache entries in the table can be removed 1839 * because to do so, they have to be removed from the hash table. 1840 */ 1841 cache_lock_all_vnodes(); 1842 cache_lock_all_buckets(); 1843 old_nchashtbl = nchashtbl; 1844 old_nchash = nchash; 1845 nchashtbl = new_nchashtbl; 1846 nchash = new_nchash; 1847 for (i = 0; i <= old_nchash; i++) { 1848 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) { 1849 hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, 1850 ncp->nc_dvp); 1851 LIST_REMOVE(ncp, nc_hash); 1852 LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash); 1853 } 1854 } 1855 cache_unlock_all_buckets(); 1856 cache_unlock_all_vnodes(); 1857 free(old_nchashtbl, M_VFSCACHE); 1858 } 1859 1860 /* 1861 * Invalidate all entries to a particular vnode. 1862 */ 1863 void 1864 cache_purge(struct vnode *vp) 1865 { 1866 TAILQ_HEAD(, namecache) ncps; 1867 struct namecache *ncp, *nnp; 1868 struct mtx *vlp, *vlp2; 1869 1870 CTR1(KTR_VFS, "cache_purge(%p)", vp); 1871 SDT_PROBE1(vfs, namecache, purge, done, vp); 1872 if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) && 1873 vp->v_cache_dd == NULL) 1874 return; 1875 TAILQ_INIT(&ncps); 1876 vlp = VP2VNODELOCK(vp); 1877 vlp2 = NULL; 1878 mtx_lock(vlp); 1879 retry: 1880 while (!LIST_EMPTY(&vp->v_cache_src)) { 1881 ncp = LIST_FIRST(&vp->v_cache_src); 1882 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) 1883 goto retry; 1884 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); 1885 } 1886 while (!TAILQ_EMPTY(&vp->v_cache_dst)) { 1887 ncp = TAILQ_FIRST(&vp->v_cache_dst); 1888 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) 1889 goto retry; 1890 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); 1891 } 1892 ncp = vp->v_cache_dd; 1893 if (ncp != NULL) { 1894 KASSERT(ncp->nc_flag & NCF_ISDOTDOT, 1895 ("lost dotdot link")); 1896 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) 1897 goto retry; 1898 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); 1899 } 1900 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); 1901 mtx_unlock(vlp); 1902 if (vlp2 != NULL) 1903 mtx_unlock(vlp2); 1904 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { 1905 cache_free(ncp); 1906 } 1907 } 1908 1909 /* 1910 * Invalidate all negative entries for a particular directory vnode. 1911 */ 1912 void 1913 cache_purge_negative(struct vnode *vp) 1914 { 1915 TAILQ_HEAD(, namecache) ncps; 1916 struct namecache *ncp, *nnp; 1917 struct mtx *vlp; 1918 1919 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp); 1920 SDT_PROBE1(vfs, namecache, purge_negative, done, vp); 1921 TAILQ_INIT(&ncps); 1922 vlp = VP2VNODELOCK(vp); 1923 mtx_lock(vlp); 1924 LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) { 1925 if (!(ncp->nc_flag & NCF_NEGATIVE)) 1926 continue; 1927 cache_zap_negative_locked_vnode_kl(ncp, vp); 1928 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); 1929 } 1930 mtx_unlock(vlp); 1931 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { 1932 cache_free(ncp); 1933 } 1934 } 1935 1936 /* 1937 * Flush all entries referencing a particular filesystem. 1938 */ 1939 void 1940 cache_purgevfs(struct mount *mp, bool force) 1941 { 1942 TAILQ_HEAD(, namecache) ncps; 1943 struct mtx *vlp1, *vlp2; 1944 struct rwlock *blp; 1945 struct nchashhead *bucket; 1946 struct namecache *ncp, *nnp; 1947 u_long i, j, n_nchash; 1948 int error; 1949 1950 /* Scan hash tables for applicable entries */ 1951 SDT_PROBE1(vfs, namecache, purgevfs, done, mp); 1952 if (!force && mp->mnt_nvnodelistsize <= ncpurgeminvnodes) 1953 return; 1954 TAILQ_INIT(&ncps); 1955 n_nchash = nchash + 1; 1956 vlp1 = vlp2 = NULL; 1957 for (i = 0; i < numbucketlocks; i++) { 1958 blp = (struct rwlock *)&bucketlocks[i]; 1959 rw_wlock(blp); 1960 for (j = i; j < n_nchash; j += numbucketlocks) { 1961 retry: 1962 bucket = &nchashtbl[j]; 1963 LIST_FOREACH_SAFE(ncp, bucket, nc_hash, nnp) { 1964 cache_assert_bucket_locked(ncp, RA_WLOCKED); 1965 if (ncp->nc_dvp->v_mount != mp) 1966 continue; 1967 error = cache_zap_wlocked_bucket_kl(ncp, blp, 1968 &vlp1, &vlp2); 1969 if (error != 0) 1970 goto retry; 1971 TAILQ_INSERT_HEAD(&ncps, ncp, nc_dst); 1972 } 1973 } 1974 rw_wunlock(blp); 1975 if (vlp1 == NULL && vlp2 == NULL) 1976 cache_maybe_yield(); 1977 } 1978 if (vlp1 != NULL) 1979 mtx_unlock(vlp1); 1980 if (vlp2 != NULL) 1981 mtx_unlock(vlp2); 1982 1983 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { 1984 cache_free(ncp); 1985 } 1986 } 1987 1988 /* 1989 * Perform canonical checks and cache lookup and pass on to filesystem 1990 * through the vop_cachedlookup only if needed. 1991 */ 1992 1993 int 1994 vfs_cache_lookup(struct vop_lookup_args *ap) 1995 { 1996 struct vnode *dvp; 1997 int error; 1998 struct vnode **vpp = ap->a_vpp; 1999 struct componentname *cnp = ap->a_cnp; 2000 struct ucred *cred = cnp->cn_cred; 2001 int flags = cnp->cn_flags; 2002 struct thread *td = cnp->cn_thread; 2003 2004 *vpp = NULL; 2005 dvp = ap->a_dvp; 2006 2007 if (dvp->v_type != VDIR) 2008 return (ENOTDIR); 2009 2010 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 2011 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 2012 return (EROFS); 2013 2014 error = VOP_ACCESS(dvp, VEXEC, cred, td); 2015 if (error) 2016 return (error); 2017 2018 error = cache_lookup(dvp, vpp, cnp, NULL, NULL); 2019 if (error == 0) 2020 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); 2021 if (error == -1) 2022 return (0); 2023 return (error); 2024 } 2025 2026 /* 2027 * XXX All of these sysctls would probably be more productive dead. 2028 */ 2029 static int __read_mostly disablecwd; 2030 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, 2031 "Disable the getcwd syscall"); 2032 2033 /* Implementation of the getcwd syscall. */ 2034 int 2035 sys___getcwd(struct thread *td, struct __getcwd_args *uap) 2036 { 2037 2038 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen, 2039 MAXPATHLEN)); 2040 } 2041 2042 int 2043 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, size_t buflen, 2044 size_t path_max) 2045 { 2046 char *bp, *tmpbuf; 2047 struct filedesc *fdp; 2048 struct vnode *cdir, *rdir; 2049 int error; 2050 2051 if (__predict_false(disablecwd)) 2052 return (ENODEV); 2053 if (__predict_false(buflen < 2)) 2054 return (EINVAL); 2055 if (buflen > path_max) 2056 buflen = path_max; 2057 2058 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK); 2059 fdp = td->td_proc->p_fd; 2060 FILEDESC_SLOCK(fdp); 2061 cdir = fdp->fd_cdir; 2062 vrefact(cdir); 2063 rdir = fdp->fd_rdir; 2064 vrefact(rdir); 2065 FILEDESC_SUNLOCK(fdp); 2066 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen); 2067 vrele(rdir); 2068 vrele(cdir); 2069 2070 if (!error) { 2071 if (bufseg == UIO_SYSSPACE) 2072 bcopy(bp, buf, strlen(bp) + 1); 2073 else 2074 error = copyout(bp, buf, strlen(bp) + 1); 2075 #ifdef KTRACE 2076 if (KTRPOINT(curthread, KTR_NAMEI)) 2077 ktrnamei(bp); 2078 #endif 2079 } 2080 free(tmpbuf, M_TEMP); 2081 return (error); 2082 } 2083 2084 /* 2085 * Thus begins the fullpath magic. 2086 */ 2087 2088 static int __read_mostly disablefullpath; 2089 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0, 2090 "Disable the vn_fullpath function"); 2091 2092 /* 2093 * Retrieve the full filesystem path that correspond to a vnode from the name 2094 * cache (if available) 2095 */ 2096 int 2097 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) 2098 { 2099 char *buf; 2100 struct filedesc *fdp; 2101 struct vnode *rdir; 2102 int error; 2103 2104 if (__predict_false(disablefullpath)) 2105 return (ENODEV); 2106 if (__predict_false(vn == NULL)) 2107 return (EINVAL); 2108 2109 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 2110 fdp = td->td_proc->p_fd; 2111 FILEDESC_SLOCK(fdp); 2112 rdir = fdp->fd_rdir; 2113 vrefact(rdir); 2114 FILEDESC_SUNLOCK(fdp); 2115 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN); 2116 vrele(rdir); 2117 2118 if (!error) 2119 *freebuf = buf; 2120 else 2121 free(buf, M_TEMP); 2122 return (error); 2123 } 2124 2125 /* 2126 * This function is similar to vn_fullpath, but it attempts to lookup the 2127 * pathname relative to the global root mount point. This is required for the 2128 * auditing sub-system, as audited pathnames must be absolute, relative to the 2129 * global root mount point. 2130 */ 2131 int 2132 vn_fullpath_global(struct thread *td, struct vnode *vn, 2133 char **retbuf, char **freebuf) 2134 { 2135 char *buf; 2136 int error; 2137 2138 if (__predict_false(disablefullpath)) 2139 return (ENODEV); 2140 if (__predict_false(vn == NULL)) 2141 return (EINVAL); 2142 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 2143 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN); 2144 if (!error) 2145 *freebuf = buf; 2146 else 2147 free(buf, M_TEMP); 2148 return (error); 2149 } 2150 2151 int 2152 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen) 2153 { 2154 struct vnode *dvp; 2155 struct namecache *ncp; 2156 struct mtx *vlp; 2157 int error; 2158 2159 vlp = VP2VNODELOCK(*vp); 2160 mtx_lock(vlp); 2161 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) { 2162 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 2163 break; 2164 } 2165 if (ncp != NULL) { 2166 if (*buflen < ncp->nc_nlen) { 2167 mtx_unlock(vlp); 2168 vrele(*vp); 2169 counter_u64_add(numfullpathfail4, 1); 2170 error = ENOMEM; 2171 SDT_PROBE3(vfs, namecache, fullpath, return, error, 2172 vp, NULL); 2173 return (error); 2174 } 2175 *buflen -= ncp->nc_nlen; 2176 memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen); 2177 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp, 2178 ncp->nc_name, vp); 2179 dvp = *vp; 2180 *vp = ncp->nc_dvp; 2181 vref(*vp); 2182 mtx_unlock(vlp); 2183 vrele(dvp); 2184 return (0); 2185 } 2186 SDT_PROBE1(vfs, namecache, fullpath, miss, vp); 2187 2188 mtx_unlock(vlp); 2189 vn_lock(*vp, LK_SHARED | LK_RETRY); 2190 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen); 2191 vput(*vp); 2192 if (error) { 2193 counter_u64_add(numfullpathfail2, 1); 2194 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); 2195 return (error); 2196 } 2197 2198 *vp = dvp; 2199 if (dvp->v_iflag & VI_DOOMED) { 2200 /* forced unmount */ 2201 vrele(dvp); 2202 error = ENOENT; 2203 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); 2204 return (error); 2205 } 2206 /* 2207 * *vp has its use count incremented still. 2208 */ 2209 2210 return (0); 2211 } 2212 2213 /* 2214 * The magic behind kern___getcwd() and vn_fullpath(). 2215 */ 2216 static int 2217 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 2218 char *buf, char **retbuf, u_int buflen) 2219 { 2220 int error, slash_prefixed; 2221 #ifdef KDTRACE_HOOKS 2222 struct vnode *startvp = vp; 2223 #endif 2224 struct vnode *vp1; 2225 2226 buflen--; 2227 buf[buflen] = '\0'; 2228 error = 0; 2229 slash_prefixed = 0; 2230 2231 SDT_PROBE1(vfs, namecache, fullpath, entry, vp); 2232 counter_u64_add(numfullpathcalls, 1); 2233 vref(vp); 2234 if (vp->v_type != VDIR) { 2235 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen); 2236 if (error) 2237 return (error); 2238 if (buflen == 0) { 2239 vrele(vp); 2240 return (ENOMEM); 2241 } 2242 buf[--buflen] = '/'; 2243 slash_prefixed = 1; 2244 } 2245 while (vp != rdir && vp != rootvnode) { 2246 /* 2247 * The vp vnode must be already fully constructed, 2248 * since it is either found in namecache or obtained 2249 * from VOP_VPTOCNP(). We may test for VV_ROOT safely 2250 * without obtaining the vnode lock. 2251 */ 2252 if ((vp->v_vflag & VV_ROOT) != 0) { 2253 vn_lock(vp, LK_RETRY | LK_SHARED); 2254 2255 /* 2256 * With the vnode locked, check for races with 2257 * unmount, forced or not. Note that we 2258 * already verified that vp is not equal to 2259 * the root vnode, which means that 2260 * mnt_vnodecovered can be NULL only for the 2261 * case of unmount. 2262 */ 2263 if ((vp->v_iflag & VI_DOOMED) != 0 || 2264 (vp1 = vp->v_mount->mnt_vnodecovered) == NULL || 2265 vp1->v_mountedhere != vp->v_mount) { 2266 vput(vp); 2267 error = ENOENT; 2268 SDT_PROBE3(vfs, namecache, fullpath, return, 2269 error, vp, NULL); 2270 break; 2271 } 2272 2273 vref(vp1); 2274 vput(vp); 2275 vp = vp1; 2276 continue; 2277 } 2278 if (vp->v_type != VDIR) { 2279 vrele(vp); 2280 counter_u64_add(numfullpathfail1, 1); 2281 error = ENOTDIR; 2282 SDT_PROBE3(vfs, namecache, fullpath, return, 2283 error, vp, NULL); 2284 break; 2285 } 2286 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen); 2287 if (error) 2288 break; 2289 if (buflen == 0) { 2290 vrele(vp); 2291 error = ENOMEM; 2292 SDT_PROBE3(vfs, namecache, fullpath, return, error, 2293 startvp, NULL); 2294 break; 2295 } 2296 buf[--buflen] = '/'; 2297 slash_prefixed = 1; 2298 } 2299 if (error) 2300 return (error); 2301 if (!slash_prefixed) { 2302 if (buflen == 0) { 2303 vrele(vp); 2304 counter_u64_add(numfullpathfail4, 1); 2305 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM, 2306 startvp, NULL); 2307 return (ENOMEM); 2308 } 2309 buf[--buflen] = '/'; 2310 } 2311 counter_u64_add(numfullpathfound, 1); 2312 vrele(vp); 2313 2314 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen); 2315 *retbuf = buf + buflen; 2316 return (0); 2317 } 2318 2319 struct vnode * 2320 vn_dir_dd_ino(struct vnode *vp) 2321 { 2322 struct namecache *ncp; 2323 struct vnode *ddvp; 2324 struct mtx *vlp; 2325 2326 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino"); 2327 vlp = VP2VNODELOCK(vp); 2328 mtx_lock(vlp); 2329 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) { 2330 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) 2331 continue; 2332 ddvp = ncp->nc_dvp; 2333 vhold(ddvp); 2334 mtx_unlock(vlp); 2335 if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread)) 2336 return (NULL); 2337 return (ddvp); 2338 } 2339 mtx_unlock(vlp); 2340 return (NULL); 2341 } 2342 2343 int 2344 vn_commname(struct vnode *vp, char *buf, u_int buflen) 2345 { 2346 struct namecache *ncp; 2347 struct mtx *vlp; 2348 int l; 2349 2350 vlp = VP2VNODELOCK(vp); 2351 mtx_lock(vlp); 2352 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) 2353 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 2354 break; 2355 if (ncp == NULL) { 2356 mtx_unlock(vlp); 2357 return (ENOENT); 2358 } 2359 l = min(ncp->nc_nlen, buflen - 1); 2360 memcpy(buf, ncp->nc_name, l); 2361 mtx_unlock(vlp); 2362 buf[l] = '\0'; 2363 return (0); 2364 } 2365 2366 /* ABI compat shims for old kernel modules. */ 2367 #undef cache_enter 2368 2369 void cache_enter(struct vnode *dvp, struct vnode *vp, 2370 struct componentname *cnp); 2371 2372 void 2373 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2374 { 2375 2376 cache_enter_time(dvp, vp, cnp, NULL, NULL); 2377 } 2378 2379 /* 2380 * This function updates path string to vnode's full global path 2381 * and checks the size of the new path string against the pathlen argument. 2382 * 2383 * Requires a locked, referenced vnode. 2384 * Vnode is re-locked on success or ENODEV, otherwise unlocked. 2385 * 2386 * If sysctl debug.disablefullpath is set, ENODEV is returned, 2387 * vnode is left locked and path remain untouched. 2388 * 2389 * If vp is a directory, the call to vn_fullpath_global() always succeeds 2390 * because it falls back to the ".." lookup if the namecache lookup fails. 2391 */ 2392 int 2393 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, 2394 u_int pathlen) 2395 { 2396 struct nameidata nd; 2397 struct vnode *vp1; 2398 char *rpath, *fbuf; 2399 int error; 2400 2401 ASSERT_VOP_ELOCKED(vp, __func__); 2402 2403 /* Return ENODEV if sysctl debug.disablefullpath==1 */ 2404 if (__predict_false(disablefullpath)) 2405 return (ENODEV); 2406 2407 /* Construct global filesystem path from vp. */ 2408 VOP_UNLOCK(vp, 0); 2409 error = vn_fullpath_global(td, vp, &rpath, &fbuf); 2410 2411 if (error != 0) { 2412 vrele(vp); 2413 return (error); 2414 } 2415 2416 if (strlen(rpath) >= pathlen) { 2417 vrele(vp); 2418 error = ENAMETOOLONG; 2419 goto out; 2420 } 2421 2422 /* 2423 * Re-lookup the vnode by path to detect a possible rename. 2424 * As a side effect, the vnode is relocked. 2425 * If vnode was renamed, return ENOENT. 2426 */ 2427 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 2428 UIO_SYSSPACE, path, td); 2429 error = namei(&nd); 2430 if (error != 0) { 2431 vrele(vp); 2432 goto out; 2433 } 2434 NDFREE(&nd, NDF_ONLY_PNBUF); 2435 vp1 = nd.ni_vp; 2436 vrele(vp); 2437 if (vp1 == vp) 2438 strcpy(path, rpath); 2439 else { 2440 vput(vp1); 2441 error = ENOENT; 2442 } 2443 2444 out: 2445 free(fbuf, M_TEMP); 2446 return (error); 2447 } 2448