vfs_cache.c revision 330897
1/*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Poul-Henning Kamp of the FreeBSD Project. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: stable/11/sys/kern/vfs_cache.c 330897 2018-03-14 03:19:51Z eadler $"); 39 40#include "opt_ktrace.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/counter.h> 45#include <sys/filedesc.h> 46#include <sys/fnv_hash.h> 47#include <sys/kernel.h> 48#include <sys/lock.h> 49#include <sys/malloc.h> 50#include <sys/fcntl.h> 51#include <sys/mount.h> 52#include <sys/namei.h> 53#include <sys/proc.h> 54#include <sys/rwlock.h> 55#include <sys/sdt.h> 56#include <sys/smp.h> 57#include <sys/syscallsubr.h> 58#include <sys/sysctl.h> 59#include <sys/sysproto.h> 60#include <sys/vnode.h> 61#ifdef KTRACE 62#include <sys/ktrace.h> 63#endif 64 65#include <vm/uma.h> 66 67SDT_PROVIDER_DECLARE(vfs); 68SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *", 69 "struct vnode *"); 70SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *", 71 "char *"); 72SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *"); 73SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *", 74 "char *", "struct vnode *"); 75SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *"); 76SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", 77 "struct vnode *", "char *"); 78SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *", 79 "struct vnode *"); 80SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative, 81 "struct vnode *", "char *"); 82SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *", 83 "char *"); 84SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *"); 85SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *"); 86SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *"); 87SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *", 88 "struct vnode *"); 89SDT_PROBE_DEFINE3(vfs, namecache, zap_negative, done, "struct vnode *", 90 "char *", "int"); 91SDT_PROBE_DEFINE3(vfs, namecache, shrink_negative, done, "struct vnode *", 92 "char *", "int"); 93 94/* 95 * This structure describes the elements in the cache of recent 96 * names looked up by namei. 97 */ 98 99struct namecache { 100 LIST_ENTRY(namecache) nc_hash; /* hash chain */ 101 LIST_ENTRY(namecache) nc_src; /* source vnode list */ 102 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ 103 struct vnode *nc_dvp; /* vnode of parent of name */ 104 union { 105 struct vnode *nu_vp; /* vnode the name refers to */ 106 u_int nu_neghits; /* negative entry hits */ 107 } n_un; 108 u_char nc_flag; /* flag bits */ 109 u_char nc_nlen; /* length of name */ 110 char nc_name[0]; /* segment name + nul */ 111}; 112 113/* 114 * struct namecache_ts repeats struct namecache layout up to the 115 * nc_nlen member. 116 * struct namecache_ts is used in place of struct namecache when time(s) need 117 * to be stored. The nc_dotdottime field is used when a cache entry is mapping 118 * both a non-dotdot directory name plus dotdot for the directory's 119 * parent. 120 */ 121struct namecache_ts { 122 struct timespec nc_time; /* timespec provided by fs */ 123 struct timespec nc_dotdottime; /* dotdot timespec provided by fs */ 124 int nc_ticks; /* ticks value when entry was added */ 125 struct namecache nc_nc; 126}; 127 128#define nc_vp n_un.nu_vp 129#define nc_neghits n_un.nu_neghits 130 131/* 132 * Flags in namecache.nc_flag 133 */ 134#define NCF_WHITE 0x01 135#define NCF_ISDOTDOT 0x02 136#define NCF_TS 0x04 137#define NCF_DTS 0x08 138#define NCF_DVDROP 0x10 139#define NCF_NEGATIVE 0x20 140#define NCF_HOTNEGATIVE 0x40 141 142/* 143 * Name caching works as follows: 144 * 145 * Names found by directory scans are retained in a cache 146 * for future reference. It is managed LRU, so frequently 147 * used names will hang around. Cache is indexed by hash value 148 * obtained from (vp, name) where vp refers to the directory 149 * containing name. 150 * 151 * If it is a "negative" entry, (i.e. for a name that is known NOT to 152 * exist) the vnode pointer will be NULL. 153 * 154 * Upon reaching the last segment of a path, if the reference 155 * is for DELETE, or NOCACHE is set (rewrite), and the 156 * name is located in the cache, it will be dropped. 157 * 158 * These locks are used (in the order in which they can be taken): 159 * NAME TYPE ROLE 160 * vnodelock mtx vnode lists and v_cache_dd field protection 161 * bucketlock rwlock for access to given set of hash buckets 162 * neglist mtx negative entry LRU management 163 * 164 * Additionally, ncneg_shrink_lock mtx is used to have at most one thread 165 * shrinking the LRU list. 166 * 167 * It is legal to take multiple vnodelock and bucketlock locks. The locking 168 * order is lower address first. Both are recursive. 169 * 170 * "." lookups are lockless. 171 * 172 * ".." and vnode -> name lookups require vnodelock. 173 * 174 * name -> vnode lookup requires the relevant bucketlock to be held for reading. 175 * 176 * Insertions and removals of entries require involved vnodes and bucketlocks 177 * to be write-locked to prevent other threads from seeing the entry. 178 * 179 * Some lookups result in removal of the found entry (e.g. getting rid of a 180 * negative entry with the intent to create a positive one), which poses a 181 * problem when multiple threads reach the state. Similarly, two different 182 * threads can purge two different vnodes and try to remove the same name. 183 * 184 * If the already held vnode lock is lower than the second required lock, we 185 * can just take the other lock. However, in the opposite case, this could 186 * deadlock. As such, this is resolved by trylocking and if that fails unlocking 187 * the first node, locking everything in order and revalidating the state. 188 */ 189 190/* 191 * Structures associated with name caching. 192 */ 193#define NCHHASH(hash) \ 194 (&nchashtbl[(hash) & nchash]) 195static __read_mostly LIST_HEAD(nchashhead, namecache) *nchashtbl;/* Hash Table */ 196static u_long __read_mostly nchash; /* size of hash table */ 197SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, 198 "Size of namecache hash table"); 199static u_long __read_mostly ncnegfactor = 12; /* ratio of negative entries */ 200SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, 201 "Ratio of negative namecache entries"); 202static u_long __exclusive_cache_line numneg; /* number of negative entries allocated */ 203SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, 204 "Number of negative entries in namecache"); 205static u_long __exclusive_cache_line numcache;/* number of cache entries allocated */ 206SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, 207 "Number of namecache entries"); 208static u_long __exclusive_cache_line numcachehv;/* number of cache entries with vnodes held */ 209SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, 210 "Number of namecache entries with vnodes held"); 211u_int __read_mostly ncsizefactor = 2; 212SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0, 213 "Size factor for namecache"); 214static u_int __read_mostly ncpurgeminvnodes; 215SYSCTL_UINT(_vfs, OID_AUTO, ncpurgeminvnodes, CTLFLAG_RW, &ncpurgeminvnodes, 0, 216 "Number of vnodes below which purgevfs ignores the request"); 217static u_int __read_mostly ncneghitsrequeue = 8; 218SYSCTL_UINT(_vfs, OID_AUTO, ncneghitsrequeue, CTLFLAG_RW, &ncneghitsrequeue, 0, 219 "Number of hits to requeue a negative entry in the LRU list"); 220 221struct nchstats nchstats; /* cache effectiveness statistics */ 222 223static struct mtx ncneg_shrink_lock; 224static int shrink_list_turn; 225 226struct neglist { 227 struct mtx nl_lock; 228 TAILQ_HEAD(, namecache) nl_list; 229} __aligned(CACHE_LINE_SIZE); 230 231static struct neglist __read_mostly *neglists; 232static struct neglist ncneg_hot; 233 234#define numneglists (ncneghash + 1) 235static u_int __read_mostly ncneghash; 236static inline struct neglist * 237NCP2NEGLIST(struct namecache *ncp) 238{ 239 240 return (&neglists[(((uintptr_t)(ncp) >> 8) & ncneghash)]); 241} 242 243#define numbucketlocks (ncbuckethash + 1) 244static u_int __read_mostly ncbuckethash; 245static struct rwlock_padalign __read_mostly *bucketlocks; 246#define HASH2BUCKETLOCK(hash) \ 247 ((struct rwlock *)(&bucketlocks[((hash) & ncbuckethash)])) 248 249#define numvnodelocks (ncvnodehash + 1) 250static u_int __read_mostly ncvnodehash; 251static struct mtx __read_mostly *vnodelocks; 252static inline struct mtx * 253VP2VNODELOCK(struct vnode *vp) 254{ 255 256 return (&vnodelocks[(((uintptr_t)(vp) >> 8) & ncvnodehash)]); 257} 258 259/* 260 * UMA zones for the VFS cache. 261 * 262 * The small cache is used for entries with short names, which are the 263 * most common. The large cache is used for entries which are too big to 264 * fit in the small cache. 265 */ 266static uma_zone_t __read_mostly cache_zone_small; 267static uma_zone_t __read_mostly cache_zone_small_ts; 268static uma_zone_t __read_mostly cache_zone_large; 269static uma_zone_t __read_mostly cache_zone_large_ts; 270 271#define CACHE_PATH_CUTOFF 35 272 273static struct namecache * 274cache_alloc(int len, int ts) 275{ 276 struct namecache_ts *ncp_ts; 277 struct namecache *ncp; 278 279 if (__predict_false(ts)) { 280 if (len <= CACHE_PATH_CUTOFF) 281 ncp_ts = uma_zalloc(cache_zone_small_ts, M_WAITOK); 282 else 283 ncp_ts = uma_zalloc(cache_zone_large_ts, M_WAITOK); 284 ncp = &ncp_ts->nc_nc; 285 } else { 286 if (len <= CACHE_PATH_CUTOFF) 287 ncp = uma_zalloc(cache_zone_small, M_WAITOK); 288 else 289 ncp = uma_zalloc(cache_zone_large, M_WAITOK); 290 } 291 return (ncp); 292} 293 294static void 295cache_free(struct namecache *ncp) 296{ 297 struct namecache_ts *ncp_ts; 298 299 if (ncp == NULL) 300 return; 301 if ((ncp->nc_flag & NCF_DVDROP) != 0) 302 vdrop(ncp->nc_dvp); 303 if (__predict_false(ncp->nc_flag & NCF_TS)) { 304 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); 305 if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) 306 uma_zfree(cache_zone_small_ts, ncp_ts); 307 else 308 uma_zfree(cache_zone_large_ts, ncp_ts); 309 } else { 310 if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) 311 uma_zfree(cache_zone_small, ncp); 312 else 313 uma_zfree(cache_zone_large, ncp); 314 } 315} 316 317static void 318cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp) 319{ 320 struct namecache_ts *ncp_ts; 321 322 KASSERT((ncp->nc_flag & NCF_TS) != 0 || 323 (tsp == NULL && ticksp == NULL), 324 ("No NCF_TS")); 325 326 if (tsp == NULL && ticksp == NULL) 327 return; 328 329 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); 330 if (tsp != NULL) 331 *tsp = ncp_ts->nc_time; 332 if (ticksp != NULL) 333 *ticksp = ncp_ts->nc_ticks; 334} 335 336static int __read_mostly doingcache = 1; /* 1 => enable the cache */ 337SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, 338 "VFS namecache enabled"); 339 340/* Export size information to userland */ 341SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 342 sizeof(struct namecache), "sizeof(struct namecache)"); 343 344/* 345 * The new name cache statistics 346 */ 347static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, 348 "Name cache statistics"); 349#define STATNODE_ULONG(name, descr) \ 350 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr); 351#define STATNODE_COUNTER(name, descr) \ 352 static counter_u64_t __read_mostly name; \ 353 SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, descr); 354STATNODE_ULONG(numneg, "Number of negative cache entries"); 355STATNODE_ULONG(numcache, "Number of cache entries"); 356STATNODE_COUNTER(numcalls, "Number of cache lookups"); 357STATNODE_COUNTER(dothits, "Number of '.' hits"); 358STATNODE_COUNTER(dotdothits, "Number of '..' hits"); 359STATNODE_COUNTER(numchecks, "Number of checks in lookup"); 360STATNODE_COUNTER(nummiss, "Number of cache misses"); 361STATNODE_COUNTER(nummisszap, "Number of cache misses we do not want to cache"); 362STATNODE_COUNTER(numposzaps, 363 "Number of cache hits (positive) we do not want to cache"); 364STATNODE_COUNTER(numposhits, "Number of cache hits (positive)"); 365STATNODE_COUNTER(numnegzaps, 366 "Number of cache hits (negative) we do not want to cache"); 367STATNODE_COUNTER(numneghits, "Number of cache hits (negative)"); 368/* These count for kern___getcwd(), too. */ 369STATNODE_COUNTER(numfullpathcalls, "Number of fullpath search calls"); 370STATNODE_COUNTER(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)"); 371STATNODE_COUNTER(numfullpathfail2, 372 "Number of fullpath search errors (VOP_VPTOCNP failures)"); 373STATNODE_COUNTER(numfullpathfail4, "Number of fullpath search errors (ENOMEM)"); 374STATNODE_COUNTER(numfullpathfound, "Number of successful fullpath calls"); 375static long zap_and_exit_bucket_fail; STATNODE_ULONG(zap_and_exit_bucket_fail, 376 "Number of times zap_and_exit failed to lock"); 377static long cache_lock_vnodes_cel_3_failures; 378STATNODE_ULONG(cache_lock_vnodes_cel_3_failures, 379 "Number of times 3-way vnode locking failed"); 380 381static void cache_zap_locked(struct namecache *ncp, bool neg_locked); 382static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 383 char *buf, char **retbuf, u_int buflen); 384 385static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); 386 387static int cache_yield; 388SYSCTL_INT(_vfs_cache, OID_AUTO, yield, CTLFLAG_RD, &cache_yield, 0, 389 "Number of times cache called yield"); 390 391static void 392cache_maybe_yield(void) 393{ 394 395 if (should_yield()) { 396 cache_yield++; 397 kern_yield(PRI_USER); 398 } 399} 400 401static inline void 402cache_assert_vlp_locked(struct mtx *vlp) 403{ 404 405 if (vlp != NULL) 406 mtx_assert(vlp, MA_OWNED); 407} 408 409static inline void 410cache_assert_vnode_locked(struct vnode *vp) 411{ 412 struct mtx *vlp; 413 414 vlp = VP2VNODELOCK(vp); 415 cache_assert_vlp_locked(vlp); 416} 417 418static uint32_t 419cache_get_hash(char *name, u_char len, struct vnode *dvp) 420{ 421 uint32_t hash; 422 423 hash = fnv_32_buf(name, len, FNV1_32_INIT); 424 hash = fnv_32_buf(&dvp, sizeof(dvp), hash); 425 return (hash); 426} 427 428static inline struct rwlock * 429NCP2BUCKETLOCK(struct namecache *ncp) 430{ 431 uint32_t hash; 432 433 hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); 434 return (HASH2BUCKETLOCK(hash)); 435} 436 437#ifdef INVARIANTS 438static void 439cache_assert_bucket_locked(struct namecache *ncp, int mode) 440{ 441 struct rwlock *blp; 442 443 blp = NCP2BUCKETLOCK(ncp); 444 rw_assert(blp, mode); 445} 446#else 447#define cache_assert_bucket_locked(x, y) do { } while (0) 448#endif 449 450#define cache_sort(x, y) _cache_sort((void **)(x), (void **)(y)) 451static void 452_cache_sort(void **p1, void **p2) 453{ 454 void *tmp; 455 456 if (*p1 > *p2) { 457 tmp = *p2; 458 *p2 = *p1; 459 *p1 = tmp; 460 } 461} 462 463static void 464cache_lock_all_buckets(void) 465{ 466 u_int i; 467 468 for (i = 0; i < numbucketlocks; i++) 469 rw_wlock(&bucketlocks[i]); 470} 471 472static void 473cache_unlock_all_buckets(void) 474{ 475 u_int i; 476 477 for (i = 0; i < numbucketlocks; i++) 478 rw_wunlock(&bucketlocks[i]); 479} 480 481static void 482cache_lock_all_vnodes(void) 483{ 484 u_int i; 485 486 for (i = 0; i < numvnodelocks; i++) 487 mtx_lock(&vnodelocks[i]); 488} 489 490static void 491cache_unlock_all_vnodes(void) 492{ 493 u_int i; 494 495 for (i = 0; i < numvnodelocks; i++) 496 mtx_unlock(&vnodelocks[i]); 497} 498 499static int 500cache_trylock_vnodes(struct mtx *vlp1, struct mtx *vlp2) 501{ 502 503 cache_sort(&vlp1, &vlp2); 504 MPASS(vlp2 != NULL); 505 506 if (vlp1 != NULL) { 507 if (!mtx_trylock(vlp1)) 508 return (EAGAIN); 509 } 510 if (!mtx_trylock(vlp2)) { 511 if (vlp1 != NULL) 512 mtx_unlock(vlp1); 513 return (EAGAIN); 514 } 515 516 return (0); 517} 518 519static void 520cache_unlock_vnodes(struct mtx *vlp1, struct mtx *vlp2) 521{ 522 523 MPASS(vlp1 != NULL || vlp2 != NULL); 524 525 if (vlp1 != NULL) 526 mtx_unlock(vlp1); 527 if (vlp2 != NULL) 528 mtx_unlock(vlp2); 529} 530 531static int 532sysctl_nchstats(SYSCTL_HANDLER_ARGS) 533{ 534 struct nchstats snap; 535 536 if (req->oldptr == NULL) 537 return (SYSCTL_OUT(req, 0, sizeof(snap))); 538 539 snap = nchstats; 540 snap.ncs_goodhits = counter_u64_fetch(numposhits); 541 snap.ncs_neghits = counter_u64_fetch(numneghits); 542 snap.ncs_badhits = counter_u64_fetch(numposzaps) + 543 counter_u64_fetch(numnegzaps); 544 snap.ncs_miss = counter_u64_fetch(nummisszap) + 545 counter_u64_fetch(nummiss); 546 547 return (SYSCTL_OUT(req, &snap, sizeof(snap))); 548} 549SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD | 550 CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU", 551 "VFS cache effectiveness statistics"); 552 553#ifdef DIAGNOSTIC 554/* 555 * Grab an atomic snapshot of the name cache hash chain lengths 556 */ 557static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, 558 "hash table stats"); 559 560static int 561sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) 562{ 563 struct nchashhead *ncpp; 564 struct namecache *ncp; 565 int i, error, n_nchash, *cntbuf; 566 567retry: 568 n_nchash = nchash + 1; /* nchash is max index, not count */ 569 if (req->oldptr == NULL) 570 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); 571 cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK); 572 cache_lock_all_buckets(); 573 if (n_nchash != nchash + 1) { 574 cache_unlock_all_buckets(); 575 free(cntbuf, M_TEMP); 576 goto retry; 577 } 578 /* Scan hash tables counting entries */ 579 for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++) 580 LIST_FOREACH(ncp, ncpp, nc_hash) 581 cntbuf[i]++; 582 cache_unlock_all_buckets(); 583 for (error = 0, i = 0; i < n_nchash; i++) 584 if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0) 585 break; 586 free(cntbuf, M_TEMP); 587 return (error); 588} 589SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD| 590 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", 591 "nchash chain lengths"); 592 593static int 594sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) 595{ 596 int error; 597 struct nchashhead *ncpp; 598 struct namecache *ncp; 599 int n_nchash; 600 int count, maxlength, used, pct; 601 602 if (!req->oldptr) 603 return SYSCTL_OUT(req, 0, 4 * sizeof(int)); 604 605 cache_lock_all_buckets(); 606 n_nchash = nchash + 1; /* nchash is max index, not count */ 607 used = 0; 608 maxlength = 0; 609 610 /* Scan hash tables for applicable entries */ 611 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { 612 count = 0; 613 LIST_FOREACH(ncp, ncpp, nc_hash) { 614 count++; 615 } 616 if (count) 617 used++; 618 if (maxlength < count) 619 maxlength = count; 620 } 621 n_nchash = nchash + 1; 622 cache_unlock_all_buckets(); 623 pct = (used * 100) / (n_nchash / 100); 624 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); 625 if (error) 626 return (error); 627 error = SYSCTL_OUT(req, &used, sizeof(used)); 628 if (error) 629 return (error); 630 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); 631 if (error) 632 return (error); 633 error = SYSCTL_OUT(req, &pct, sizeof(pct)); 634 if (error) 635 return (error); 636 return (0); 637} 638SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD| 639 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I", 640 "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)"); 641#endif 642 643/* 644 * Negative entries management 645 * 646 * A variation of LRU scheme is used. New entries are hashed into one of 647 * numneglists cold lists. Entries get promoted to the hot list on first hit. 648 * Partial LRU for the hot list is maintained by requeueing them every 649 * ncneghitsrequeue hits. 650 * 651 * The shrinker will demote hot list head and evict from the cold list in a 652 * round-robin manner. 653 */ 654static void 655cache_negative_hit(struct namecache *ncp) 656{ 657 struct neglist *neglist; 658 u_int hits; 659 660 MPASS(ncp->nc_flag & NCF_NEGATIVE); 661 hits = atomic_fetchadd_int(&ncp->nc_neghits, 1); 662 if (ncp->nc_flag & NCF_HOTNEGATIVE) { 663 if ((hits % ncneghitsrequeue) != 0) 664 return; 665 mtx_lock(&ncneg_hot.nl_lock); 666 if (ncp->nc_flag & NCF_HOTNEGATIVE) { 667 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); 668 TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst); 669 mtx_unlock(&ncneg_hot.nl_lock); 670 return; 671 } 672 /* 673 * The shrinker cleared the flag and removed the entry from 674 * the hot list. Put it back. 675 */ 676 } else { 677 mtx_lock(&ncneg_hot.nl_lock); 678 } 679 neglist = NCP2NEGLIST(ncp); 680 mtx_lock(&neglist->nl_lock); 681 if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) { 682 TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); 683 TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst); 684 ncp->nc_flag |= NCF_HOTNEGATIVE; 685 } 686 mtx_unlock(&neglist->nl_lock); 687 mtx_unlock(&ncneg_hot.nl_lock); 688} 689 690static void 691cache_negative_insert(struct namecache *ncp, bool neg_locked) 692{ 693 struct neglist *neglist; 694 695 MPASS(ncp->nc_flag & NCF_NEGATIVE); 696 cache_assert_bucket_locked(ncp, RA_WLOCKED); 697 neglist = NCP2NEGLIST(ncp); 698 if (!neg_locked) { 699 mtx_lock(&neglist->nl_lock); 700 } else { 701 mtx_assert(&neglist->nl_lock, MA_OWNED); 702 } 703 TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst); 704 if (!neg_locked) 705 mtx_unlock(&neglist->nl_lock); 706 atomic_add_rel_long(&numneg, 1); 707} 708 709static void 710cache_negative_remove(struct namecache *ncp, bool neg_locked) 711{ 712 struct neglist *neglist; 713 bool hot_locked = false; 714 bool list_locked = false; 715 716 MPASS(ncp->nc_flag & NCF_NEGATIVE); 717 cache_assert_bucket_locked(ncp, RA_WLOCKED); 718 neglist = NCP2NEGLIST(ncp); 719 if (!neg_locked) { 720 if (ncp->nc_flag & NCF_HOTNEGATIVE) { 721 hot_locked = true; 722 mtx_lock(&ncneg_hot.nl_lock); 723 if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) { 724 list_locked = true; 725 mtx_lock(&neglist->nl_lock); 726 } 727 } else { 728 list_locked = true; 729 mtx_lock(&neglist->nl_lock); 730 } 731 } 732 if (ncp->nc_flag & NCF_HOTNEGATIVE) { 733 mtx_assert(&ncneg_hot.nl_lock, MA_OWNED); 734 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); 735 } else { 736 mtx_assert(&neglist->nl_lock, MA_OWNED); 737 TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); 738 } 739 if (list_locked) 740 mtx_unlock(&neglist->nl_lock); 741 if (hot_locked) 742 mtx_unlock(&ncneg_hot.nl_lock); 743 atomic_subtract_rel_long(&numneg, 1); 744} 745 746static void 747cache_negative_shrink_select(int start, struct namecache **ncpp, 748 struct neglist **neglistpp) 749{ 750 struct neglist *neglist; 751 struct namecache *ncp; 752 int i; 753 754 *ncpp = ncp = NULL; 755 756 for (i = start; i < numneglists; i++) { 757 neglist = &neglists[i]; 758 if (TAILQ_FIRST(&neglist->nl_list) == NULL) 759 continue; 760 mtx_lock(&neglist->nl_lock); 761 ncp = TAILQ_FIRST(&neglist->nl_list); 762 if (ncp != NULL) 763 break; 764 mtx_unlock(&neglist->nl_lock); 765 } 766 767 *neglistpp = neglist; 768 *ncpp = ncp; 769} 770 771static void 772cache_negative_zap_one(void) 773{ 774 struct namecache *ncp, *ncp2; 775 struct neglist *neglist; 776 struct mtx *dvlp; 777 struct rwlock *blp; 778 779 if (!mtx_trylock(&ncneg_shrink_lock)) 780 return; 781 782 mtx_lock(&ncneg_hot.nl_lock); 783 ncp = TAILQ_FIRST(&ncneg_hot.nl_list); 784 if (ncp != NULL) { 785 neglist = NCP2NEGLIST(ncp); 786 mtx_lock(&neglist->nl_lock); 787 TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); 788 TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst); 789 ncp->nc_flag &= ~NCF_HOTNEGATIVE; 790 mtx_unlock(&neglist->nl_lock); 791 } 792 793 cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist); 794 shrink_list_turn++; 795 if (shrink_list_turn == numneglists) 796 shrink_list_turn = 0; 797 if (ncp == NULL && shrink_list_turn == 0) 798 cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist); 799 if (ncp == NULL) { 800 mtx_unlock(&ncneg_hot.nl_lock); 801 goto out; 802 } 803 804 MPASS(ncp->nc_flag & NCF_NEGATIVE); 805 dvlp = VP2VNODELOCK(ncp->nc_dvp); 806 blp = NCP2BUCKETLOCK(ncp); 807 mtx_unlock(&neglist->nl_lock); 808 mtx_unlock(&ncneg_hot.nl_lock); 809 mtx_lock(dvlp); 810 rw_wlock(blp); 811 mtx_lock(&neglist->nl_lock); 812 ncp2 = TAILQ_FIRST(&neglist->nl_list); 813 if (ncp != ncp2 || dvlp != VP2VNODELOCK(ncp2->nc_dvp) || 814 blp != NCP2BUCKETLOCK(ncp2) || !(ncp2->nc_flag & NCF_NEGATIVE)) { 815 ncp = NULL; 816 goto out_unlock_all; 817 } 818 SDT_PROBE3(vfs, namecache, shrink_negative, done, ncp->nc_dvp, 819 ncp->nc_name, ncp->nc_neghits); 820 821 cache_zap_locked(ncp, true); 822out_unlock_all: 823 mtx_unlock(&neglist->nl_lock); 824 rw_wunlock(blp); 825 mtx_unlock(dvlp); 826out: 827 mtx_unlock(&ncneg_shrink_lock); 828 cache_free(ncp); 829} 830 831/* 832 * cache_zap_locked(): 833 * 834 * Removes a namecache entry from cache, whether it contains an actual 835 * pointer to a vnode or if it is just a negative cache entry. 836 */ 837static void 838cache_zap_locked(struct namecache *ncp, bool neg_locked) 839{ 840 841 if (!(ncp->nc_flag & NCF_NEGATIVE)) 842 cache_assert_vnode_locked(ncp->nc_vp); 843 cache_assert_vnode_locked(ncp->nc_dvp); 844 cache_assert_bucket_locked(ncp, RA_WLOCKED); 845 846 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, 847 (ncp->nc_flag & NCF_NEGATIVE) ? NULL : ncp->nc_vp); 848 if (!(ncp->nc_flag & NCF_NEGATIVE)) { 849 SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp, 850 ncp->nc_name, ncp->nc_vp); 851 } else { 852 SDT_PROBE3(vfs, namecache, zap_negative, done, ncp->nc_dvp, 853 ncp->nc_name, ncp->nc_neghits); 854 } 855 LIST_REMOVE(ncp, nc_hash); 856 if (!(ncp->nc_flag & NCF_NEGATIVE)) { 857 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); 858 if (ncp == ncp->nc_vp->v_cache_dd) 859 ncp->nc_vp->v_cache_dd = NULL; 860 } else { 861 cache_negative_remove(ncp, neg_locked); 862 } 863 if (ncp->nc_flag & NCF_ISDOTDOT) { 864 if (ncp == ncp->nc_dvp->v_cache_dd) 865 ncp->nc_dvp->v_cache_dd = NULL; 866 } else { 867 LIST_REMOVE(ncp, nc_src); 868 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) { 869 ncp->nc_flag |= NCF_DVDROP; 870 atomic_subtract_rel_long(&numcachehv, 1); 871 } 872 } 873 atomic_subtract_rel_long(&numcache, 1); 874} 875 876static void 877cache_zap_negative_locked_vnode_kl(struct namecache *ncp, struct vnode *vp) 878{ 879 struct rwlock *blp; 880 881 MPASS(ncp->nc_dvp == vp); 882 MPASS(ncp->nc_flag & NCF_NEGATIVE); 883 cache_assert_vnode_locked(vp); 884 885 blp = NCP2BUCKETLOCK(ncp); 886 rw_wlock(blp); 887 cache_zap_locked(ncp, false); 888 rw_wunlock(blp); 889} 890 891static bool 892cache_zap_locked_vnode_kl2(struct namecache *ncp, struct vnode *vp, 893 struct mtx **vlpp) 894{ 895 struct mtx *pvlp, *vlp1, *vlp2, *to_unlock; 896 struct rwlock *blp; 897 898 MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp); 899 cache_assert_vnode_locked(vp); 900 901 if (ncp->nc_flag & NCF_NEGATIVE) { 902 if (*vlpp != NULL) { 903 mtx_unlock(*vlpp); 904 *vlpp = NULL; 905 } 906 cache_zap_negative_locked_vnode_kl(ncp, vp); 907 return (true); 908 } 909 910 pvlp = VP2VNODELOCK(vp); 911 blp = NCP2BUCKETLOCK(ncp); 912 vlp1 = VP2VNODELOCK(ncp->nc_dvp); 913 vlp2 = VP2VNODELOCK(ncp->nc_vp); 914 915 if (*vlpp == vlp1 || *vlpp == vlp2) { 916 to_unlock = *vlpp; 917 *vlpp = NULL; 918 } else { 919 if (*vlpp != NULL) { 920 mtx_unlock(*vlpp); 921 *vlpp = NULL; 922 } 923 cache_sort(&vlp1, &vlp2); 924 if (vlp1 == pvlp) { 925 mtx_lock(vlp2); 926 to_unlock = vlp2; 927 } else { 928 if (!mtx_trylock(vlp1)) 929 goto out_relock; 930 to_unlock = vlp1; 931 } 932 } 933 rw_wlock(blp); 934 cache_zap_locked(ncp, false); 935 rw_wunlock(blp); 936 if (to_unlock != NULL) 937 mtx_unlock(to_unlock); 938 return (true); 939 940out_relock: 941 mtx_unlock(vlp2); 942 mtx_lock(vlp1); 943 mtx_lock(vlp2); 944 MPASS(*vlpp == NULL); 945 *vlpp = vlp1; 946 return (false); 947} 948 949static int 950cache_zap_locked_vnode(struct namecache *ncp, struct vnode *vp) 951{ 952 struct mtx *pvlp, *vlp1, *vlp2, *to_unlock; 953 struct rwlock *blp; 954 int error = 0; 955 956 MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp); 957 cache_assert_vnode_locked(vp); 958 959 pvlp = VP2VNODELOCK(vp); 960 if (ncp->nc_flag & NCF_NEGATIVE) { 961 cache_zap_negative_locked_vnode_kl(ncp, vp); 962 goto out; 963 } 964 965 blp = NCP2BUCKETLOCK(ncp); 966 vlp1 = VP2VNODELOCK(ncp->nc_dvp); 967 vlp2 = VP2VNODELOCK(ncp->nc_vp); 968 cache_sort(&vlp1, &vlp2); 969 if (vlp1 == pvlp) { 970 mtx_lock(vlp2); 971 to_unlock = vlp2; 972 } else { 973 if (!mtx_trylock(vlp1)) { 974 error = EAGAIN; 975 goto out; 976 } 977 to_unlock = vlp1; 978 } 979 rw_wlock(blp); 980 cache_zap_locked(ncp, false); 981 rw_wunlock(blp); 982 mtx_unlock(to_unlock); 983out: 984 mtx_unlock(pvlp); 985 return (error); 986} 987 988static int 989cache_zap_wlocked_bucket(struct namecache *ncp, struct rwlock *blp) 990{ 991 struct mtx *dvlp, *vlp; 992 993 cache_assert_bucket_locked(ncp, RA_WLOCKED); 994 995 dvlp = VP2VNODELOCK(ncp->nc_dvp); 996 vlp = NULL; 997 if (!(ncp->nc_flag & NCF_NEGATIVE)) 998 vlp = VP2VNODELOCK(ncp->nc_vp); 999 if (cache_trylock_vnodes(dvlp, vlp) == 0) { 1000 cache_zap_locked(ncp, false); 1001 rw_wunlock(blp); 1002 cache_unlock_vnodes(dvlp, vlp); 1003 return (0); 1004 } 1005 1006 rw_wunlock(blp); 1007 return (EAGAIN); 1008} 1009 1010static int 1011cache_zap_rlocked_bucket(struct namecache *ncp, struct rwlock *blp) 1012{ 1013 struct mtx *dvlp, *vlp; 1014 1015 cache_assert_bucket_locked(ncp, RA_RLOCKED); 1016 1017 dvlp = VP2VNODELOCK(ncp->nc_dvp); 1018 vlp = NULL; 1019 if (!(ncp->nc_flag & NCF_NEGATIVE)) 1020 vlp = VP2VNODELOCK(ncp->nc_vp); 1021 if (cache_trylock_vnodes(dvlp, vlp) == 0) { 1022 rw_runlock(blp); 1023 rw_wlock(blp); 1024 cache_zap_locked(ncp, false); 1025 rw_wunlock(blp); 1026 cache_unlock_vnodes(dvlp, vlp); 1027 return (0); 1028 } 1029 1030 rw_runlock(blp); 1031 return (EAGAIN); 1032} 1033 1034static int 1035cache_zap_wlocked_bucket_kl(struct namecache *ncp, struct rwlock *blp, 1036 struct mtx **vlpp1, struct mtx **vlpp2) 1037{ 1038 struct mtx *dvlp, *vlp; 1039 1040 cache_assert_bucket_locked(ncp, RA_WLOCKED); 1041 1042 dvlp = VP2VNODELOCK(ncp->nc_dvp); 1043 vlp = NULL; 1044 if (!(ncp->nc_flag & NCF_NEGATIVE)) 1045 vlp = VP2VNODELOCK(ncp->nc_vp); 1046 cache_sort(&dvlp, &vlp); 1047 1048 if (*vlpp1 == dvlp && *vlpp2 == vlp) { 1049 cache_zap_locked(ncp, false); 1050 cache_unlock_vnodes(dvlp, vlp); 1051 *vlpp1 = NULL; 1052 *vlpp2 = NULL; 1053 return (0); 1054 } 1055 1056 if (*vlpp1 != NULL) 1057 mtx_unlock(*vlpp1); 1058 if (*vlpp2 != NULL) 1059 mtx_unlock(*vlpp2); 1060 *vlpp1 = NULL; 1061 *vlpp2 = NULL; 1062 1063 if (cache_trylock_vnodes(dvlp, vlp) == 0) { 1064 cache_zap_locked(ncp, false); 1065 cache_unlock_vnodes(dvlp, vlp); 1066 return (0); 1067 } 1068 1069 rw_wunlock(blp); 1070 *vlpp1 = dvlp; 1071 *vlpp2 = vlp; 1072 if (*vlpp1 != NULL) 1073 mtx_lock(*vlpp1); 1074 mtx_lock(*vlpp2); 1075 rw_wlock(blp); 1076 return (EAGAIN); 1077} 1078 1079static void 1080cache_lookup_unlock(struct rwlock *blp, struct mtx *vlp) 1081{ 1082 1083 if (blp != NULL) { 1084 rw_runlock(blp); 1085 } else { 1086 mtx_unlock(vlp); 1087 } 1088} 1089 1090static int __noinline 1091cache_lookup_dot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1092 struct timespec *tsp, int *ticksp) 1093{ 1094 int ltype; 1095 1096 *vpp = dvp; 1097 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .", 1098 dvp, cnp->cn_nameptr); 1099 counter_u64_add(dothits, 1); 1100 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp); 1101 if (tsp != NULL) 1102 timespecclear(tsp); 1103 if (ticksp != NULL) 1104 *ticksp = ticks; 1105 vrefact(*vpp); 1106 /* 1107 * When we lookup "." we still can be asked to lock it 1108 * differently... 1109 */ 1110 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 1111 if (ltype != VOP_ISLOCKED(*vpp)) { 1112 if (ltype == LK_EXCLUSIVE) { 1113 vn_lock(*vpp, LK_UPGRADE | LK_RETRY); 1114 if ((*vpp)->v_iflag & VI_DOOMED) { 1115 /* forced unmount */ 1116 vrele(*vpp); 1117 *vpp = NULL; 1118 return (ENOENT); 1119 } 1120 } else 1121 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); 1122 } 1123 return (-1); 1124} 1125 1126/* 1127 * Lookup an entry in the cache 1128 * 1129 * Lookup is called with dvp pointing to the directory to search, 1130 * cnp pointing to the name of the entry being sought. If the lookup 1131 * succeeds, the vnode is returned in *vpp, and a status of -1 is 1132 * returned. If the lookup determines that the name does not exist 1133 * (negative caching), a status of ENOENT is returned. If the lookup 1134 * fails, a status of zero is returned. If the directory vnode is 1135 * recycled out from under us due to a forced unmount, a status of 1136 * ENOENT is returned. 1137 * 1138 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is 1139 * unlocked. If we're looking up . an extra ref is taken, but the lock is 1140 * not recursively acquired. 1141 */ 1142 1143static __noinline int 1144cache_lookup_nomakeentry(struct vnode *dvp, struct vnode **vpp, 1145 struct componentname *cnp, struct timespec *tsp, int *ticksp) 1146{ 1147 struct namecache *ncp; 1148 struct rwlock *blp; 1149 struct mtx *dvlp, *dvlp2; 1150 uint32_t hash; 1151 int error; 1152 1153 if (cnp->cn_namelen == 2 && 1154 cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') { 1155 counter_u64_add(dotdothits, 1); 1156 dvlp = VP2VNODELOCK(dvp); 1157 dvlp2 = NULL; 1158 mtx_lock(dvlp); 1159retry_dotdot: 1160 ncp = dvp->v_cache_dd; 1161 if (ncp == NULL) { 1162 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, 1163 "..", NULL); 1164 mtx_unlock(dvlp); 1165 if (dvlp2 != NULL) 1166 mtx_unlock(dvlp2); 1167 return (0); 1168 } 1169 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { 1170 if (ncp->nc_dvp != dvp) 1171 panic("dvp %p v_cache_dd %p\n", dvp, ncp); 1172 if (!cache_zap_locked_vnode_kl2(ncp, 1173 dvp, &dvlp2)) 1174 goto retry_dotdot; 1175 MPASS(dvp->v_cache_dd == NULL); 1176 mtx_unlock(dvlp); 1177 if (dvlp2 != NULL) 1178 mtx_unlock(dvlp2); 1179 cache_free(ncp); 1180 } else { 1181 dvp->v_cache_dd = NULL; 1182 mtx_unlock(dvlp); 1183 if (dvlp2 != NULL) 1184 mtx_unlock(dvlp2); 1185 } 1186 return (0); 1187 } 1188 1189 hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); 1190 blp = HASH2BUCKETLOCK(hash); 1191retry: 1192 if (LIST_EMPTY(NCHHASH(hash))) 1193 goto out_no_entry; 1194 1195 rw_wlock(blp); 1196 1197 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { 1198 counter_u64_add(numchecks, 1); 1199 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && 1200 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) 1201 break; 1202 } 1203 1204 /* We failed to find an entry */ 1205 if (ncp == NULL) { 1206 rw_wunlock(blp); 1207 goto out_no_entry; 1208 } 1209 1210 counter_u64_add(numposzaps, 1); 1211 1212 error = cache_zap_wlocked_bucket(ncp, blp); 1213 if (error != 0) { 1214 zap_and_exit_bucket_fail++; 1215 cache_maybe_yield(); 1216 goto retry; 1217 } 1218 cache_free(ncp); 1219 return (0); 1220out_no_entry: 1221 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL); 1222 counter_u64_add(nummisszap, 1); 1223 return (0); 1224} 1225 1226int 1227cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, 1228 struct timespec *tsp, int *ticksp) 1229{ 1230 struct namecache_ts *ncp_ts; 1231 struct namecache *ncp; 1232 struct rwlock *blp; 1233 struct mtx *dvlp, *dvlp2; 1234 uint32_t hash; 1235 int error, ltype; 1236 1237 if (__predict_false(!doingcache)) { 1238 cnp->cn_flags &= ~MAKEENTRY; 1239 return (0); 1240 } 1241 1242 counter_u64_add(numcalls, 1); 1243 1244 if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.')) 1245 return (cache_lookup_dot(dvp, vpp, cnp, tsp, ticksp)); 1246 1247 if ((cnp->cn_flags & MAKEENTRY) == 0) 1248 return (cache_lookup_nomakeentry(dvp, vpp, cnp, tsp, ticksp)); 1249 1250retry: 1251 blp = NULL; 1252 error = 0; 1253 if (cnp->cn_namelen == 2 && 1254 cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') { 1255 counter_u64_add(dotdothits, 1); 1256 dvlp = VP2VNODELOCK(dvp); 1257 dvlp2 = NULL; 1258 mtx_lock(dvlp); 1259 ncp = dvp->v_cache_dd; 1260 if (ncp == NULL) { 1261 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, 1262 "..", NULL); 1263 mtx_unlock(dvlp); 1264 return (0); 1265 } 1266 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { 1267 if (ncp->nc_flag & NCF_NEGATIVE) 1268 *vpp = NULL; 1269 else 1270 *vpp = ncp->nc_vp; 1271 } else 1272 *vpp = ncp->nc_dvp; 1273 /* Return failure if negative entry was found. */ 1274 if (*vpp == NULL) 1275 goto negative_success; 1276 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..", 1277 dvp, cnp->cn_nameptr, *vpp); 1278 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..", 1279 *vpp); 1280 cache_out_ts(ncp, tsp, ticksp); 1281 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == 1282 NCF_DTS && tsp != NULL) { 1283 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); 1284 *tsp = ncp_ts->nc_dotdottime; 1285 } 1286 goto success; 1287 } 1288 1289 hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); 1290 blp = HASH2BUCKETLOCK(hash); 1291 rw_rlock(blp); 1292 1293 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { 1294 counter_u64_add(numchecks, 1); 1295 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && 1296 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) 1297 break; 1298 } 1299 1300 /* We failed to find an entry */ 1301 if (ncp == NULL) { 1302 rw_runlock(blp); 1303 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, 1304 NULL); 1305 counter_u64_add(nummiss, 1); 1306 return (0); 1307 } 1308 1309 /* We found a "positive" match, return the vnode */ 1310 if (!(ncp->nc_flag & NCF_NEGATIVE)) { 1311 counter_u64_add(numposhits, 1); 1312 *vpp = ncp->nc_vp; 1313 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p", 1314 dvp, cnp->cn_nameptr, *vpp, ncp); 1315 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, 1316 *vpp); 1317 cache_out_ts(ncp, tsp, ticksp); 1318 goto success; 1319 } 1320 1321negative_success: 1322 /* We found a negative match, and want to create it, so purge */ 1323 if (cnp->cn_nameiop == CREATE) { 1324 counter_u64_add(numnegzaps, 1); 1325 goto zap_and_exit; 1326 } 1327 1328 counter_u64_add(numneghits, 1); 1329 cache_negative_hit(ncp); 1330 if (ncp->nc_flag & NCF_WHITE) 1331 cnp->cn_flags |= ISWHITEOUT; 1332 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, 1333 ncp->nc_name); 1334 cache_out_ts(ncp, tsp, ticksp); 1335 cache_lookup_unlock(blp, dvlp); 1336 return (ENOENT); 1337 1338success: 1339 /* 1340 * On success we return a locked and ref'd vnode as per the lookup 1341 * protocol. 1342 */ 1343 MPASS(dvp != *vpp); 1344 ltype = 0; /* silence gcc warning */ 1345 if (cnp->cn_flags & ISDOTDOT) { 1346 ltype = VOP_ISLOCKED(dvp); 1347 VOP_UNLOCK(dvp, 0); 1348 } 1349 vhold(*vpp); 1350 cache_lookup_unlock(blp, dvlp); 1351 error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread); 1352 if (cnp->cn_flags & ISDOTDOT) { 1353 vn_lock(dvp, ltype | LK_RETRY); 1354 if (dvp->v_iflag & VI_DOOMED) { 1355 if (error == 0) 1356 vput(*vpp); 1357 *vpp = NULL; 1358 return (ENOENT); 1359 } 1360 } 1361 if (error) { 1362 *vpp = NULL; 1363 goto retry; 1364 } 1365 if ((cnp->cn_flags & ISLASTCN) && 1366 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) { 1367 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup"); 1368 } 1369 return (-1); 1370 1371zap_and_exit: 1372 if (blp != NULL) 1373 error = cache_zap_rlocked_bucket(ncp, blp); 1374 else 1375 error = cache_zap_locked_vnode(ncp, dvp); 1376 if (error != 0) { 1377 zap_and_exit_bucket_fail++; 1378 cache_maybe_yield(); 1379 goto retry; 1380 } 1381 cache_free(ncp); 1382 return (0); 1383} 1384 1385struct celockstate { 1386 struct mtx *vlp[3]; 1387 struct rwlock *blp[2]; 1388}; 1389CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3)); 1390CTASSERT((nitems(((struct celockstate *)0)->blp) == 2)); 1391 1392static inline void 1393cache_celockstate_init(struct celockstate *cel) 1394{ 1395 1396 bzero(cel, sizeof(*cel)); 1397} 1398 1399static void 1400cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp, 1401 struct vnode *dvp) 1402{ 1403 struct mtx *vlp1, *vlp2; 1404 1405 MPASS(cel->vlp[0] == NULL); 1406 MPASS(cel->vlp[1] == NULL); 1407 MPASS(cel->vlp[2] == NULL); 1408 1409 MPASS(vp != NULL || dvp != NULL); 1410 1411 vlp1 = VP2VNODELOCK(vp); 1412 vlp2 = VP2VNODELOCK(dvp); 1413 cache_sort(&vlp1, &vlp2); 1414 1415 if (vlp1 != NULL) { 1416 mtx_lock(vlp1); 1417 cel->vlp[0] = vlp1; 1418 } 1419 mtx_lock(vlp2); 1420 cel->vlp[1] = vlp2; 1421} 1422 1423static void 1424cache_unlock_vnodes_cel(struct celockstate *cel) 1425{ 1426 1427 MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL); 1428 1429 if (cel->vlp[0] != NULL) 1430 mtx_unlock(cel->vlp[0]); 1431 if (cel->vlp[1] != NULL) 1432 mtx_unlock(cel->vlp[1]); 1433 if (cel->vlp[2] != NULL) 1434 mtx_unlock(cel->vlp[2]); 1435} 1436 1437static bool 1438cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp) 1439{ 1440 struct mtx *vlp; 1441 bool ret; 1442 1443 cache_assert_vlp_locked(cel->vlp[0]); 1444 cache_assert_vlp_locked(cel->vlp[1]); 1445 MPASS(cel->vlp[2] == NULL); 1446 1447 MPASS(vp != NULL); 1448 vlp = VP2VNODELOCK(vp); 1449 1450 ret = true; 1451 if (vlp >= cel->vlp[1]) { 1452 mtx_lock(vlp); 1453 } else { 1454 if (mtx_trylock(vlp)) 1455 goto out; 1456 cache_lock_vnodes_cel_3_failures++; 1457 cache_unlock_vnodes_cel(cel); 1458 if (vlp < cel->vlp[0]) { 1459 mtx_lock(vlp); 1460 mtx_lock(cel->vlp[0]); 1461 mtx_lock(cel->vlp[1]); 1462 } else { 1463 if (cel->vlp[0] != NULL) 1464 mtx_lock(cel->vlp[0]); 1465 mtx_lock(vlp); 1466 mtx_lock(cel->vlp[1]); 1467 } 1468 ret = false; 1469 } 1470out: 1471 cel->vlp[2] = vlp; 1472 return (ret); 1473} 1474 1475static void 1476cache_lock_buckets_cel(struct celockstate *cel, struct rwlock *blp1, 1477 struct rwlock *blp2) 1478{ 1479 1480 MPASS(cel->blp[0] == NULL); 1481 MPASS(cel->blp[1] == NULL); 1482 1483 cache_sort(&blp1, &blp2); 1484 1485 if (blp1 != NULL) { 1486 rw_wlock(blp1); 1487 cel->blp[0] = blp1; 1488 } 1489 rw_wlock(blp2); 1490 cel->blp[1] = blp2; 1491} 1492 1493static void 1494cache_unlock_buckets_cel(struct celockstate *cel) 1495{ 1496 1497 if (cel->blp[0] != NULL) 1498 rw_wunlock(cel->blp[0]); 1499 rw_wunlock(cel->blp[1]); 1500} 1501 1502/* 1503 * Lock part of the cache affected by the insertion. 1504 * 1505 * This means vnodelocks for dvp, vp and the relevant bucketlock. 1506 * However, insertion can result in removal of an old entry. In this 1507 * case we have an additional vnode and bucketlock pair to lock. If the 1508 * entry is negative, ncelock is locked instead of the vnode. 1509 * 1510 * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while 1511 * preserving the locking order (smaller address first). 1512 */ 1513static void 1514cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, 1515 uint32_t hash) 1516{ 1517 struct namecache *ncp; 1518 struct rwlock *blps[2]; 1519 1520 blps[0] = HASH2BUCKETLOCK(hash); 1521 for (;;) { 1522 blps[1] = NULL; 1523 cache_lock_vnodes_cel(cel, dvp, vp); 1524 if (vp == NULL || vp->v_type != VDIR) 1525 break; 1526 ncp = vp->v_cache_dd; 1527 if (ncp == NULL) 1528 break; 1529 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 1530 break; 1531 MPASS(ncp->nc_dvp == vp); 1532 blps[1] = NCP2BUCKETLOCK(ncp); 1533 if (ncp->nc_flag & NCF_NEGATIVE) 1534 break; 1535 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) 1536 break; 1537 /* 1538 * All vnodes got re-locked. Re-validate the state and if 1539 * nothing changed we are done. Otherwise restart. 1540 */ 1541 if (ncp == vp->v_cache_dd && 1542 (ncp->nc_flag & NCF_ISDOTDOT) != 0 && 1543 blps[1] == NCP2BUCKETLOCK(ncp) && 1544 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) 1545 break; 1546 cache_unlock_vnodes_cel(cel); 1547 cel->vlp[0] = NULL; 1548 cel->vlp[1] = NULL; 1549 cel->vlp[2] = NULL; 1550 } 1551 cache_lock_buckets_cel(cel, blps[0], blps[1]); 1552} 1553 1554static void 1555cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, 1556 uint32_t hash) 1557{ 1558 struct namecache *ncp; 1559 struct rwlock *blps[2]; 1560 1561 blps[0] = HASH2BUCKETLOCK(hash); 1562 for (;;) { 1563 blps[1] = NULL; 1564 cache_lock_vnodes_cel(cel, dvp, vp); 1565 ncp = dvp->v_cache_dd; 1566 if (ncp == NULL) 1567 break; 1568 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 1569 break; 1570 MPASS(ncp->nc_dvp == dvp); 1571 blps[1] = NCP2BUCKETLOCK(ncp); 1572 if (ncp->nc_flag & NCF_NEGATIVE) 1573 break; 1574 if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) 1575 break; 1576 if (ncp == dvp->v_cache_dd && 1577 (ncp->nc_flag & NCF_ISDOTDOT) != 0 && 1578 blps[1] == NCP2BUCKETLOCK(ncp) && 1579 VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) 1580 break; 1581 cache_unlock_vnodes_cel(cel); 1582 cel->vlp[0] = NULL; 1583 cel->vlp[1] = NULL; 1584 cel->vlp[2] = NULL; 1585 } 1586 cache_lock_buckets_cel(cel, blps[0], blps[1]); 1587} 1588 1589static void 1590cache_enter_unlock(struct celockstate *cel) 1591{ 1592 1593 cache_unlock_buckets_cel(cel); 1594 cache_unlock_vnodes_cel(cel); 1595} 1596 1597/* 1598 * Add an entry to the cache. 1599 */ 1600void 1601cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, 1602 struct timespec *tsp, struct timespec *dtsp) 1603{ 1604 struct celockstate cel; 1605 struct namecache *ncp, *n2, *ndd; 1606 struct namecache_ts *ncp_ts, *n2_ts; 1607 struct nchashhead *ncpp; 1608 struct neglist *neglist; 1609 uint32_t hash; 1610 int flag; 1611 int len; 1612 bool neg_locked; 1613 int lnumcache; 1614 1615 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr); 1616 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp, 1617 ("cache_enter: Adding a doomed vnode")); 1618 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp, 1619 ("cache_enter: Doomed vnode used as src")); 1620 1621 if (__predict_false(!doingcache)) 1622 return; 1623 1624 /* 1625 * Avoid blowout in namecache entries. 1626 */ 1627 if (__predict_false(numcache >= desiredvnodes * ncsizefactor)) 1628 return; 1629 1630 cache_celockstate_init(&cel); 1631 ndd = NULL; 1632 flag = 0; 1633 if (cnp->cn_nameptr[0] == '.') { 1634 if (cnp->cn_namelen == 1) 1635 return; 1636 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 1637 len = cnp->cn_namelen; 1638 hash = cache_get_hash(cnp->cn_nameptr, len, dvp); 1639 cache_enter_lock_dd(&cel, dvp, vp, hash); 1640 /* 1641 * If dotdot entry already exists, just retarget it 1642 * to new parent vnode, otherwise continue with new 1643 * namecache entry allocation. 1644 */ 1645 if ((ncp = dvp->v_cache_dd) != NULL && 1646 ncp->nc_flag & NCF_ISDOTDOT) { 1647 KASSERT(ncp->nc_dvp == dvp, 1648 ("wrong isdotdot parent")); 1649 neg_locked = false; 1650 if (ncp->nc_flag & NCF_NEGATIVE || vp == NULL) { 1651 neglist = NCP2NEGLIST(ncp); 1652 mtx_lock(&ncneg_hot.nl_lock); 1653 mtx_lock(&neglist->nl_lock); 1654 neg_locked = true; 1655 } 1656 if (!(ncp->nc_flag & NCF_NEGATIVE)) { 1657 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, 1658 ncp, nc_dst); 1659 } else { 1660 cache_negative_remove(ncp, true); 1661 } 1662 if (vp != NULL) { 1663 TAILQ_INSERT_HEAD(&vp->v_cache_dst, 1664 ncp, nc_dst); 1665 ncp->nc_flag &= ~(NCF_NEGATIVE|NCF_HOTNEGATIVE); 1666 } else { 1667 ncp->nc_flag &= ~(NCF_HOTNEGATIVE); 1668 ncp->nc_flag |= NCF_NEGATIVE; 1669 cache_negative_insert(ncp, true); 1670 } 1671 if (neg_locked) { 1672 mtx_unlock(&neglist->nl_lock); 1673 mtx_unlock(&ncneg_hot.nl_lock); 1674 } 1675 ncp->nc_vp = vp; 1676 cache_enter_unlock(&cel); 1677 return; 1678 } 1679 dvp->v_cache_dd = NULL; 1680 cache_enter_unlock(&cel); 1681 cache_celockstate_init(&cel); 1682 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp); 1683 flag = NCF_ISDOTDOT; 1684 } 1685 } 1686 1687 /* 1688 * Calculate the hash key and setup as much of the new 1689 * namecache entry as possible before acquiring the lock. 1690 */ 1691 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); 1692 ncp->nc_flag = flag; 1693 ncp->nc_vp = vp; 1694 if (vp == NULL) 1695 ncp->nc_flag |= NCF_NEGATIVE; 1696 ncp->nc_dvp = dvp; 1697 if (tsp != NULL) { 1698 ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); 1699 ncp_ts->nc_time = *tsp; 1700 ncp_ts->nc_ticks = ticks; 1701 ncp_ts->nc_nc.nc_flag |= NCF_TS; 1702 if (dtsp != NULL) { 1703 ncp_ts->nc_dotdottime = *dtsp; 1704 ncp_ts->nc_nc.nc_flag |= NCF_DTS; 1705 } 1706 } 1707 len = ncp->nc_nlen = cnp->cn_namelen; 1708 hash = cache_get_hash(cnp->cn_nameptr, len, dvp); 1709 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1); 1710 cache_enter_lock(&cel, dvp, vp, hash); 1711 1712 /* 1713 * See if this vnode or negative entry is already in the cache 1714 * with this name. This can happen with concurrent lookups of 1715 * the same path name. 1716 */ 1717 ncpp = NCHHASH(hash); 1718 LIST_FOREACH(n2, ncpp, nc_hash) { 1719 if (n2->nc_dvp == dvp && 1720 n2->nc_nlen == cnp->cn_namelen && 1721 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) { 1722 if (tsp != NULL) { 1723 KASSERT((n2->nc_flag & NCF_TS) != 0, 1724 ("no NCF_TS")); 1725 n2_ts = __containerof(n2, struct namecache_ts, nc_nc); 1726 n2_ts->nc_time = ncp_ts->nc_time; 1727 n2_ts->nc_ticks = ncp_ts->nc_ticks; 1728 if (dtsp != NULL) { 1729 n2_ts->nc_dotdottime = ncp_ts->nc_dotdottime; 1730 if (ncp->nc_flag & NCF_NEGATIVE) 1731 mtx_lock(&ncneg_hot.nl_lock); 1732 n2_ts->nc_nc.nc_flag |= NCF_DTS; 1733 if (ncp->nc_flag & NCF_NEGATIVE) 1734 mtx_unlock(&ncneg_hot.nl_lock); 1735 } 1736 } 1737 goto out_unlock_free; 1738 } 1739 } 1740 1741 if (flag == NCF_ISDOTDOT) { 1742 /* 1743 * See if we are trying to add .. entry, but some other lookup 1744 * has populated v_cache_dd pointer already. 1745 */ 1746 if (dvp->v_cache_dd != NULL) 1747 goto out_unlock_free; 1748 KASSERT(vp == NULL || vp->v_type == VDIR, 1749 ("wrong vnode type %p", vp)); 1750 dvp->v_cache_dd = ncp; 1751 } 1752 1753 if (vp != NULL) { 1754 if (vp->v_type == VDIR) { 1755 if (flag != NCF_ISDOTDOT) { 1756 /* 1757 * For this case, the cache entry maps both the 1758 * directory name in it and the name ".." for the 1759 * directory's parent. 1760 */ 1761 if ((ndd = vp->v_cache_dd) != NULL) { 1762 if ((ndd->nc_flag & NCF_ISDOTDOT) != 0) 1763 cache_zap_locked(ndd, false); 1764 else 1765 ndd = NULL; 1766 } 1767 vp->v_cache_dd = ncp; 1768 } 1769 } else { 1770 vp->v_cache_dd = NULL; 1771 } 1772 } 1773 1774 if (flag != NCF_ISDOTDOT) { 1775 if (LIST_EMPTY(&dvp->v_cache_src)) { 1776 vhold(dvp); 1777 atomic_add_rel_long(&numcachehv, 1); 1778 } 1779 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); 1780 } 1781 1782 /* 1783 * Insert the new namecache entry into the appropriate chain 1784 * within the cache entries table. 1785 */ 1786 LIST_INSERT_HEAD(ncpp, ncp, nc_hash); 1787 1788 /* 1789 * If the entry is "negative", we place it into the 1790 * "negative" cache queue, otherwise, we place it into the 1791 * destination vnode's cache entries queue. 1792 */ 1793 if (vp != NULL) { 1794 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); 1795 SDT_PROBE3(vfs, namecache, enter, done, dvp, ncp->nc_name, 1796 vp); 1797 } else { 1798 if (cnp->cn_flags & ISWHITEOUT) 1799 ncp->nc_flag |= NCF_WHITE; 1800 cache_negative_insert(ncp, false); 1801 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp, 1802 ncp->nc_name); 1803 } 1804 cache_enter_unlock(&cel); 1805 lnumcache = atomic_fetchadd_long(&numcache, 1) + 1; 1806 if (numneg * ncnegfactor > lnumcache) 1807 cache_negative_zap_one(); 1808 cache_free(ndd); 1809 return; 1810out_unlock_free: 1811 cache_enter_unlock(&cel); 1812 cache_free(ncp); 1813 return; 1814} 1815 1816static u_int 1817cache_roundup_2(u_int val) 1818{ 1819 u_int res; 1820 1821 for (res = 1; res <= val; res <<= 1) 1822 continue; 1823 1824 return (res); 1825} 1826 1827/* 1828 * Name cache initialization, from vfs_init() when we are booting 1829 */ 1830static void 1831nchinit(void *dummy __unused) 1832{ 1833 u_int i; 1834 1835 cache_zone_small = uma_zcreate("S VFS Cache", 1836 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1, 1837 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache), 1838 UMA_ZONE_ZINIT); 1839 cache_zone_small_ts = uma_zcreate("STS VFS Cache", 1840 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1, 1841 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts), 1842 UMA_ZONE_ZINIT); 1843 cache_zone_large = uma_zcreate("L VFS Cache", 1844 sizeof(struct namecache) + NAME_MAX + 1, 1845 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache), 1846 UMA_ZONE_ZINIT); 1847 cache_zone_large_ts = uma_zcreate("LTS VFS Cache", 1848 sizeof(struct namecache_ts) + NAME_MAX + 1, 1849 NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts), 1850 UMA_ZONE_ZINIT); 1851 1852 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash); 1853 ncbuckethash = cache_roundup_2(mp_ncpus * 64) - 1; 1854 if (ncbuckethash > nchash) 1855 ncbuckethash = nchash; 1856 bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE, 1857 M_WAITOK | M_ZERO); 1858 for (i = 0; i < numbucketlocks; i++) 1859 rw_init_flags(&bucketlocks[i], "ncbuc", RW_DUPOK | RW_RECURSE); 1860 ncvnodehash = cache_roundup_2(mp_ncpus * 64) - 1; 1861 vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE, 1862 M_WAITOK | M_ZERO); 1863 for (i = 0; i < numvnodelocks; i++) 1864 mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE); 1865 ncpurgeminvnodes = numbucketlocks; 1866 1867 ncneghash = 3; 1868 neglists = malloc(sizeof(*neglists) * numneglists, M_VFSCACHE, 1869 M_WAITOK | M_ZERO); 1870 for (i = 0; i < numneglists; i++) { 1871 mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF); 1872 TAILQ_INIT(&neglists[i].nl_list); 1873 } 1874 mtx_init(&ncneg_hot.nl_lock, "ncneglh", NULL, MTX_DEF); 1875 TAILQ_INIT(&ncneg_hot.nl_list); 1876 1877 mtx_init(&ncneg_shrink_lock, "ncnegs", NULL, MTX_DEF); 1878 1879 numcalls = counter_u64_alloc(M_WAITOK); 1880 dothits = counter_u64_alloc(M_WAITOK); 1881 dotdothits = counter_u64_alloc(M_WAITOK); 1882 numchecks = counter_u64_alloc(M_WAITOK); 1883 nummiss = counter_u64_alloc(M_WAITOK); 1884 nummisszap = counter_u64_alloc(M_WAITOK); 1885 numposzaps = counter_u64_alloc(M_WAITOK); 1886 numposhits = counter_u64_alloc(M_WAITOK); 1887 numnegzaps = counter_u64_alloc(M_WAITOK); 1888 numneghits = counter_u64_alloc(M_WAITOK); 1889 numfullpathcalls = counter_u64_alloc(M_WAITOK); 1890 numfullpathfail1 = counter_u64_alloc(M_WAITOK); 1891 numfullpathfail2 = counter_u64_alloc(M_WAITOK); 1892 numfullpathfail4 = counter_u64_alloc(M_WAITOK); 1893 numfullpathfound = counter_u64_alloc(M_WAITOK); 1894} 1895SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); 1896 1897void 1898cache_changesize(int newmaxvnodes) 1899{ 1900 struct nchashhead *new_nchashtbl, *old_nchashtbl; 1901 u_long new_nchash, old_nchash; 1902 struct namecache *ncp; 1903 uint32_t hash; 1904 int i; 1905 1906 newmaxvnodes = cache_roundup_2(newmaxvnodes * 2); 1907 if (newmaxvnodes < numbucketlocks) 1908 newmaxvnodes = numbucketlocks; 1909 1910 new_nchashtbl = hashinit(newmaxvnodes, M_VFSCACHE, &new_nchash); 1911 /* If same hash table size, nothing to do */ 1912 if (nchash == new_nchash) { 1913 free(new_nchashtbl, M_VFSCACHE); 1914 return; 1915 } 1916 /* 1917 * Move everything from the old hash table to the new table. 1918 * None of the namecache entries in the table can be removed 1919 * because to do so, they have to be removed from the hash table. 1920 */ 1921 cache_lock_all_vnodes(); 1922 cache_lock_all_buckets(); 1923 old_nchashtbl = nchashtbl; 1924 old_nchash = nchash; 1925 nchashtbl = new_nchashtbl; 1926 nchash = new_nchash; 1927 for (i = 0; i <= old_nchash; i++) { 1928 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) { 1929 hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, 1930 ncp->nc_dvp); 1931 LIST_REMOVE(ncp, nc_hash); 1932 LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash); 1933 } 1934 } 1935 cache_unlock_all_buckets(); 1936 cache_unlock_all_vnodes(); 1937 free(old_nchashtbl, M_VFSCACHE); 1938} 1939 1940/* 1941 * Invalidate all entries to a particular vnode. 1942 */ 1943void 1944cache_purge(struct vnode *vp) 1945{ 1946 TAILQ_HEAD(, namecache) ncps; 1947 struct namecache *ncp, *nnp; 1948 struct mtx *vlp, *vlp2; 1949 1950 CTR1(KTR_VFS, "cache_purge(%p)", vp); 1951 SDT_PROBE1(vfs, namecache, purge, done, vp); 1952 if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) && 1953 vp->v_cache_dd == NULL) 1954 return; 1955 TAILQ_INIT(&ncps); 1956 vlp = VP2VNODELOCK(vp); 1957 vlp2 = NULL; 1958 mtx_lock(vlp); 1959retry: 1960 while (!LIST_EMPTY(&vp->v_cache_src)) { 1961 ncp = LIST_FIRST(&vp->v_cache_src); 1962 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) 1963 goto retry; 1964 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); 1965 } 1966 while (!TAILQ_EMPTY(&vp->v_cache_dst)) { 1967 ncp = TAILQ_FIRST(&vp->v_cache_dst); 1968 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) 1969 goto retry; 1970 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); 1971 } 1972 ncp = vp->v_cache_dd; 1973 if (ncp != NULL) { 1974 KASSERT(ncp->nc_flag & NCF_ISDOTDOT, 1975 ("lost dotdot link")); 1976 if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) 1977 goto retry; 1978 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); 1979 } 1980 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); 1981 mtx_unlock(vlp); 1982 if (vlp2 != NULL) 1983 mtx_unlock(vlp2); 1984 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { 1985 cache_free(ncp); 1986 } 1987} 1988 1989/* 1990 * Invalidate all negative entries for a particular directory vnode. 1991 */ 1992void 1993cache_purge_negative(struct vnode *vp) 1994{ 1995 TAILQ_HEAD(, namecache) ncps; 1996 struct namecache *ncp, *nnp; 1997 struct mtx *vlp; 1998 1999 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp); 2000 SDT_PROBE1(vfs, namecache, purge_negative, done, vp); 2001 if (LIST_EMPTY(&vp->v_cache_src)) 2002 return; 2003 TAILQ_INIT(&ncps); 2004 vlp = VP2VNODELOCK(vp); 2005 mtx_lock(vlp); 2006 LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) { 2007 if (!(ncp->nc_flag & NCF_NEGATIVE)) 2008 continue; 2009 cache_zap_negative_locked_vnode_kl(ncp, vp); 2010 TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); 2011 } 2012 mtx_unlock(vlp); 2013 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { 2014 cache_free(ncp); 2015 } 2016} 2017 2018/* 2019 * Flush all entries referencing a particular filesystem. 2020 */ 2021void 2022cache_purgevfs(struct mount *mp, bool force) 2023{ 2024 TAILQ_HEAD(, namecache) ncps; 2025 struct mtx *vlp1, *vlp2; 2026 struct rwlock *blp; 2027 struct nchashhead *bucket; 2028 struct namecache *ncp, *nnp; 2029 u_long i, j, n_nchash; 2030 int error; 2031 2032 /* Scan hash tables for applicable entries */ 2033 SDT_PROBE1(vfs, namecache, purgevfs, done, mp); 2034 if (!force && mp->mnt_nvnodelistsize <= ncpurgeminvnodes) 2035 return; 2036 TAILQ_INIT(&ncps); 2037 n_nchash = nchash + 1; 2038 vlp1 = vlp2 = NULL; 2039 for (i = 0; i < numbucketlocks; i++) { 2040 blp = (struct rwlock *)&bucketlocks[i]; 2041 rw_wlock(blp); 2042 for (j = i; j < n_nchash; j += numbucketlocks) { 2043retry: 2044 bucket = &nchashtbl[j]; 2045 LIST_FOREACH_SAFE(ncp, bucket, nc_hash, nnp) { 2046 cache_assert_bucket_locked(ncp, RA_WLOCKED); 2047 if (ncp->nc_dvp->v_mount != mp) 2048 continue; 2049 error = cache_zap_wlocked_bucket_kl(ncp, blp, 2050 &vlp1, &vlp2); 2051 if (error != 0) 2052 goto retry; 2053 TAILQ_INSERT_HEAD(&ncps, ncp, nc_dst); 2054 } 2055 } 2056 rw_wunlock(blp); 2057 if (vlp1 == NULL && vlp2 == NULL) 2058 cache_maybe_yield(); 2059 } 2060 if (vlp1 != NULL) 2061 mtx_unlock(vlp1); 2062 if (vlp2 != NULL) 2063 mtx_unlock(vlp2); 2064 2065 TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { 2066 cache_free(ncp); 2067 } 2068} 2069 2070/* 2071 * Perform canonical checks and cache lookup and pass on to filesystem 2072 * through the vop_cachedlookup only if needed. 2073 */ 2074 2075int 2076vfs_cache_lookup(struct vop_lookup_args *ap) 2077{ 2078 struct vnode *dvp; 2079 int error; 2080 struct vnode **vpp = ap->a_vpp; 2081 struct componentname *cnp = ap->a_cnp; 2082 struct ucred *cred = cnp->cn_cred; 2083 int flags = cnp->cn_flags; 2084 struct thread *td = cnp->cn_thread; 2085 2086 *vpp = NULL; 2087 dvp = ap->a_dvp; 2088 2089 if (dvp->v_type != VDIR) 2090 return (ENOTDIR); 2091 2092 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 2093 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 2094 return (EROFS); 2095 2096 error = VOP_ACCESS(dvp, VEXEC, cred, td); 2097 if (error) 2098 return (error); 2099 2100 error = cache_lookup(dvp, vpp, cnp, NULL, NULL); 2101 if (error == 0) 2102 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); 2103 if (error == -1) 2104 return (0); 2105 return (error); 2106} 2107 2108/* 2109 * XXX All of these sysctls would probably be more productive dead. 2110 */ 2111static int __read_mostly disablecwd; 2112SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, 2113 "Disable the getcwd syscall"); 2114 2115/* Implementation of the getcwd syscall. */ 2116int 2117sys___getcwd(struct thread *td, struct __getcwd_args *uap) 2118{ 2119 2120 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen, 2121 MAXPATHLEN)); 2122} 2123 2124int 2125kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen, 2126 u_int path_max) 2127{ 2128 char *bp, *tmpbuf; 2129 struct filedesc *fdp; 2130 struct vnode *cdir, *rdir; 2131 int error; 2132 2133 if (__predict_false(disablecwd)) 2134 return (ENODEV); 2135 if (__predict_false(buflen < 2)) 2136 return (EINVAL); 2137 if (buflen > path_max) 2138 buflen = path_max; 2139 2140 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK); 2141 fdp = td->td_proc->p_fd; 2142 FILEDESC_SLOCK(fdp); 2143 cdir = fdp->fd_cdir; 2144 vrefact(cdir); 2145 rdir = fdp->fd_rdir; 2146 vrefact(rdir); 2147 FILEDESC_SUNLOCK(fdp); 2148 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen); 2149 vrele(rdir); 2150 vrele(cdir); 2151 2152 if (!error) { 2153 if (bufseg == UIO_SYSSPACE) 2154 bcopy(bp, buf, strlen(bp) + 1); 2155 else 2156 error = copyout(bp, buf, strlen(bp) + 1); 2157#ifdef KTRACE 2158 if (KTRPOINT(curthread, KTR_NAMEI)) 2159 ktrnamei(bp); 2160#endif 2161 } 2162 free(tmpbuf, M_TEMP); 2163 return (error); 2164} 2165 2166/* 2167 * Thus begins the fullpath magic. 2168 */ 2169 2170static int __read_mostly disablefullpath; 2171SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0, 2172 "Disable the vn_fullpath function"); 2173 2174/* 2175 * Retrieve the full filesystem path that correspond to a vnode from the name 2176 * cache (if available) 2177 */ 2178int 2179vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) 2180{ 2181 char *buf; 2182 struct filedesc *fdp; 2183 struct vnode *rdir; 2184 int error; 2185 2186 if (__predict_false(disablefullpath)) 2187 return (ENODEV); 2188 if (__predict_false(vn == NULL)) 2189 return (EINVAL); 2190 2191 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 2192 fdp = td->td_proc->p_fd; 2193 FILEDESC_SLOCK(fdp); 2194 rdir = fdp->fd_rdir; 2195 vrefact(rdir); 2196 FILEDESC_SUNLOCK(fdp); 2197 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN); 2198 vrele(rdir); 2199 2200 if (!error) 2201 *freebuf = buf; 2202 else 2203 free(buf, M_TEMP); 2204 return (error); 2205} 2206 2207/* 2208 * This function is similar to vn_fullpath, but it attempts to lookup the 2209 * pathname relative to the global root mount point. This is required for the 2210 * auditing sub-system, as audited pathnames must be absolute, relative to the 2211 * global root mount point. 2212 */ 2213int 2214vn_fullpath_global(struct thread *td, struct vnode *vn, 2215 char **retbuf, char **freebuf) 2216{ 2217 char *buf; 2218 int error; 2219 2220 if (__predict_false(disablefullpath)) 2221 return (ENODEV); 2222 if (__predict_false(vn == NULL)) 2223 return (EINVAL); 2224 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 2225 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN); 2226 if (!error) 2227 *freebuf = buf; 2228 else 2229 free(buf, M_TEMP); 2230 return (error); 2231} 2232 2233int 2234vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen) 2235{ 2236 struct vnode *dvp; 2237 struct namecache *ncp; 2238 struct mtx *vlp; 2239 int error; 2240 2241 vlp = VP2VNODELOCK(*vp); 2242 mtx_lock(vlp); 2243 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) { 2244 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 2245 break; 2246 } 2247 if (ncp != NULL) { 2248 if (*buflen < ncp->nc_nlen) { 2249 mtx_unlock(vlp); 2250 vrele(*vp); 2251 counter_u64_add(numfullpathfail4, 1); 2252 error = ENOMEM; 2253 SDT_PROBE3(vfs, namecache, fullpath, return, error, 2254 vp, NULL); 2255 return (error); 2256 } 2257 *buflen -= ncp->nc_nlen; 2258 memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen); 2259 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp, 2260 ncp->nc_name, vp); 2261 dvp = *vp; 2262 *vp = ncp->nc_dvp; 2263 vref(*vp); 2264 mtx_unlock(vlp); 2265 vrele(dvp); 2266 return (0); 2267 } 2268 SDT_PROBE1(vfs, namecache, fullpath, miss, vp); 2269 2270 mtx_unlock(vlp); 2271 vn_lock(*vp, LK_SHARED | LK_RETRY); 2272 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen); 2273 vput(*vp); 2274 if (error) { 2275 counter_u64_add(numfullpathfail2, 1); 2276 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); 2277 return (error); 2278 } 2279 2280 *vp = dvp; 2281 if (dvp->v_iflag & VI_DOOMED) { 2282 /* forced unmount */ 2283 vrele(dvp); 2284 error = ENOENT; 2285 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); 2286 return (error); 2287 } 2288 /* 2289 * *vp has its use count incremented still. 2290 */ 2291 2292 return (0); 2293} 2294 2295/* 2296 * The magic behind kern___getcwd() and vn_fullpath(). 2297 */ 2298static int 2299vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 2300 char *buf, char **retbuf, u_int buflen) 2301{ 2302 int error, slash_prefixed; 2303#ifdef KDTRACE_HOOKS 2304 struct vnode *startvp = vp; 2305#endif 2306 struct vnode *vp1; 2307 2308 buflen--; 2309 buf[buflen] = '\0'; 2310 error = 0; 2311 slash_prefixed = 0; 2312 2313 SDT_PROBE1(vfs, namecache, fullpath, entry, vp); 2314 counter_u64_add(numfullpathcalls, 1); 2315 vref(vp); 2316 if (vp->v_type != VDIR) { 2317 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen); 2318 if (error) 2319 return (error); 2320 if (buflen == 0) { 2321 vrele(vp); 2322 return (ENOMEM); 2323 } 2324 buf[--buflen] = '/'; 2325 slash_prefixed = 1; 2326 } 2327 while (vp != rdir && vp != rootvnode) { 2328 /* 2329 * The vp vnode must be already fully constructed, 2330 * since it is either found in namecache or obtained 2331 * from VOP_VPTOCNP(). We may test for VV_ROOT safely 2332 * without obtaining the vnode lock. 2333 */ 2334 if ((vp->v_vflag & VV_ROOT) != 0) { 2335 vn_lock(vp, LK_RETRY | LK_SHARED); 2336 2337 /* 2338 * With the vnode locked, check for races with 2339 * unmount, forced or not. Note that we 2340 * already verified that vp is not equal to 2341 * the root vnode, which means that 2342 * mnt_vnodecovered can be NULL only for the 2343 * case of unmount. 2344 */ 2345 if ((vp->v_iflag & VI_DOOMED) != 0 || 2346 (vp1 = vp->v_mount->mnt_vnodecovered) == NULL || 2347 vp1->v_mountedhere != vp->v_mount) { 2348 vput(vp); 2349 error = ENOENT; 2350 SDT_PROBE3(vfs, namecache, fullpath, return, 2351 error, vp, NULL); 2352 break; 2353 } 2354 2355 vref(vp1); 2356 vput(vp); 2357 vp = vp1; 2358 continue; 2359 } 2360 if (vp->v_type != VDIR) { 2361 vrele(vp); 2362 counter_u64_add(numfullpathfail1, 1); 2363 error = ENOTDIR; 2364 SDT_PROBE3(vfs, namecache, fullpath, return, 2365 error, vp, NULL); 2366 break; 2367 } 2368 error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen); 2369 if (error) 2370 break; 2371 if (buflen == 0) { 2372 vrele(vp); 2373 error = ENOMEM; 2374 SDT_PROBE3(vfs, namecache, fullpath, return, error, 2375 startvp, NULL); 2376 break; 2377 } 2378 buf[--buflen] = '/'; 2379 slash_prefixed = 1; 2380 } 2381 if (error) 2382 return (error); 2383 if (!slash_prefixed) { 2384 if (buflen == 0) { 2385 vrele(vp); 2386 counter_u64_add(numfullpathfail4, 1); 2387 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM, 2388 startvp, NULL); 2389 return (ENOMEM); 2390 } 2391 buf[--buflen] = '/'; 2392 } 2393 counter_u64_add(numfullpathfound, 1); 2394 vrele(vp); 2395 2396 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen); 2397 *retbuf = buf + buflen; 2398 return (0); 2399} 2400 2401struct vnode * 2402vn_dir_dd_ino(struct vnode *vp) 2403{ 2404 struct namecache *ncp; 2405 struct vnode *ddvp; 2406 struct mtx *vlp; 2407 2408 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino"); 2409 vlp = VP2VNODELOCK(vp); 2410 mtx_lock(vlp); 2411 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) { 2412 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) 2413 continue; 2414 ddvp = ncp->nc_dvp; 2415 vhold(ddvp); 2416 mtx_unlock(vlp); 2417 if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread)) 2418 return (NULL); 2419 return (ddvp); 2420 } 2421 mtx_unlock(vlp); 2422 return (NULL); 2423} 2424 2425int 2426vn_commname(struct vnode *vp, char *buf, u_int buflen) 2427{ 2428 struct namecache *ncp; 2429 struct mtx *vlp; 2430 int l; 2431 2432 vlp = VP2VNODELOCK(vp); 2433 mtx_lock(vlp); 2434 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) 2435 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 2436 break; 2437 if (ncp == NULL) { 2438 mtx_unlock(vlp); 2439 return (ENOENT); 2440 } 2441 l = min(ncp->nc_nlen, buflen - 1); 2442 memcpy(buf, ncp->nc_name, l); 2443 mtx_unlock(vlp); 2444 buf[l] = '\0'; 2445 return (0); 2446} 2447 2448/* ABI compat shims for old kernel modules. */ 2449#undef cache_enter 2450 2451void cache_enter(struct vnode *dvp, struct vnode *vp, 2452 struct componentname *cnp); 2453 2454void 2455cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 2456{ 2457 2458 cache_enter_time(dvp, vp, cnp, NULL, NULL); 2459} 2460 2461/* 2462 * This function updates path string to vnode's full global path 2463 * and checks the size of the new path string against the pathlen argument. 2464 * 2465 * Requires a locked, referenced vnode. 2466 * Vnode is re-locked on success or ENODEV, otherwise unlocked. 2467 * 2468 * If sysctl debug.disablefullpath is set, ENODEV is returned, 2469 * vnode is left locked and path remain untouched. 2470 * 2471 * If vp is a directory, the call to vn_fullpath_global() always succeeds 2472 * because it falls back to the ".." lookup if the namecache lookup fails. 2473 */ 2474int 2475vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, 2476 u_int pathlen) 2477{ 2478 struct nameidata nd; 2479 struct vnode *vp1; 2480 char *rpath, *fbuf; 2481 int error; 2482 2483 ASSERT_VOP_ELOCKED(vp, __func__); 2484 2485 /* Return ENODEV if sysctl debug.disablefullpath==1 */ 2486 if (__predict_false(disablefullpath)) 2487 return (ENODEV); 2488 2489 /* Construct global filesystem path from vp. */ 2490 VOP_UNLOCK(vp, 0); 2491 error = vn_fullpath_global(td, vp, &rpath, &fbuf); 2492 2493 if (error != 0) { 2494 vrele(vp); 2495 return (error); 2496 } 2497 2498 if (strlen(rpath) >= pathlen) { 2499 vrele(vp); 2500 error = ENAMETOOLONG; 2501 goto out; 2502 } 2503 2504 /* 2505 * Re-lookup the vnode by path to detect a possible rename. 2506 * As a side effect, the vnode is relocked. 2507 * If vnode was renamed, return ENOENT. 2508 */ 2509 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 2510 UIO_SYSSPACE, path, td); 2511 error = namei(&nd); 2512 if (error != 0) { 2513 vrele(vp); 2514 goto out; 2515 } 2516 NDFREE(&nd, NDF_ONLY_PNBUF); 2517 vp1 = nd.ni_vp; 2518 vrele(vp); 2519 if (vp1 == vp) 2520 strcpy(path, rpath); 2521 else { 2522 vput(vp1); 2523 error = ENOENT; 2524 } 2525 2526out: 2527 free(fbuf, M_TEMP); 2528 return (error); 2529} 2530