Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_cache.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * Copyright (c) 1989, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Poul-Henning Kamp of the FreeBSD Project. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 38 #include "opt_kdtrace.h" 39 #include "opt_ktrace.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.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/mount.h> 49 #include <sys/namei.h> 50 #include <sys/proc.h> 51 #include <sys/rwlock.h> 52 #include <sys/sdt.h> 53 #include <sys/syscallsubr.h> 54 #include <sys/sysctl.h> 55 #include <sys/sysproto.h> 56 #include <sys/vnode.h> 57 #ifdef KTRACE 58 #include <sys/ktrace.h> 59 #endif 60 61 #include <vm/uma.h> 62 63 SDT_PROVIDER_DECLARE(vfs); 64 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, done, "struct vnode *", "char *", 65 "struct vnode *"); 66 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, done, "struct vnode *", 67 "char *"); 68 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, entry, "struct vnode *"); 69 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, hit, "struct vnode *", 70 "char *", "struct vnode *"); 71 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, miss, "struct vnode *"); 72 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, return, "int", 73 "struct vnode *", "char *"); 74 SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, hit, "struct vnode *", "char *", 75 "struct vnode *"); 76 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit_negative, hit-negative, 77 "struct vnode *", "char *"); 78 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, miss, "struct vnode *", 79 "char *"); 80 SDT_PROBE_DEFINE1(vfs, namecache, purge, done, done, "struct vnode *"); 81 SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, done, "struct vnode *"); 82 SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, done, "struct mount *"); 83 SDT_PROBE_DEFINE3(vfs, namecache, zap, done, done, "struct vnode *", "char *", 84 "struct vnode *"); 85 SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, done, "struct vnode *", 86 "char *"); 87 88 /* 89 * This structure describes the elements in the cache of recent 90 * names looked up by namei. 91 */ 92 93 struct namecache { 94 LIST_ENTRY(namecache) nc_hash; /* hash chain */ 95 LIST_ENTRY(namecache) nc_src; /* source vnode list */ 96 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ 97 struct vnode *nc_dvp; /* vnode of parent of name */ 98 struct vnode *nc_vp; /* vnode the name refers to */ 99 u_char nc_flag; /* flag bits */ 100 u_char nc_nlen; /* length of name */ 101 char nc_name[0]; /* segment name + nul */ 102 }; 103 104 /* 105 * struct namecache_ts repeats struct namecache layout up to the 106 * nc_nlen member. 107 * struct namecache_ts is used in place of struct namecache when time(s) need 108 * to be stored. The nc_dotdottime field is used when a cache entry is mapping 109 * both a non-dotdot directory name plus dotdot for the directory's 110 * parent. 111 */ 112 struct namecache_ts { 113 LIST_ENTRY(namecache) nc_hash; /* hash chain */ 114 LIST_ENTRY(namecache) nc_src; /* source vnode list */ 115 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ 116 struct vnode *nc_dvp; /* vnode of parent of name */ 117 struct vnode *nc_vp; /* vnode the name refers to */ 118 u_char nc_flag; /* flag bits */ 119 u_char nc_nlen; /* length of name */ 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 char nc_name[0]; /* segment name + nul */ 124 }; 125 126 /* 127 * Flags in namecache.nc_flag 128 */ 129 #define NCF_WHITE 0x01 130 #define NCF_ISDOTDOT 0x02 131 #define NCF_TS 0x04 132 #define NCF_DTS 0x08 133 134 /* 135 * Name caching works as follows: 136 * 137 * Names found by directory scans are retained in a cache 138 * for future reference. It is managed LRU, so frequently 139 * used names will hang around. Cache is indexed by hash value 140 * obtained from (vp, name) where vp refers to the directory 141 * containing name. 142 * 143 * If it is a "negative" entry, (i.e. for a name that is known NOT to 144 * exist) the vnode pointer will be NULL. 145 * 146 * Upon reaching the last segment of a path, if the reference 147 * is for DELETE, or NOCACHE is set (rewrite), and the 148 * name is located in the cache, it will be dropped. 149 */ 150 151 /* 152 * Structures associated with name cacheing. 153 */ 154 #define NCHHASH(hash) \ 155 (&nchashtbl[(hash) & nchash]) 156 static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */ 157 static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */ 158 static u_long nchash; /* size of hash table */ 159 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, 160 "Size of namecache hash table"); 161 static u_long ncnegfactor = 16; /* ratio of negative entries */ 162 /* _debug sysctl left for backward compatibility */ 163 SYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, ""); 164 SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, 165 "Ratio of negative namecache entries"); 166 static u_long numneg; /* number of negative entries allocated */ 167 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, 168 "Number of negative entries in namecache"); 169 static u_long numcache; /* number of cache entries allocated */ 170 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, 171 "Number of namecache entries"); 172 static u_long numcachehv; /* number of cache entries with vnodes held */ 173 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, 174 "Number of namecache entries with vnodes held"); 175 static u_int ncsizefactor = 2; 176 SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0, 177 "Size factor for namecache"); 178 179 struct nchstats nchstats; /* cache effectiveness statistics */ 180 181 static struct rwlock cache_lock; 182 RW_SYSINIT(vfscache, &cache_lock, "Name Cache"); 183 184 #define CACHE_UPGRADE_LOCK() rw_try_upgrade(&cache_lock) 185 #define CACHE_RLOCK() rw_rlock(&cache_lock) 186 #define CACHE_RUNLOCK() rw_runlock(&cache_lock) 187 #define CACHE_WLOCK() rw_wlock(&cache_lock) 188 #define CACHE_WUNLOCK() rw_wunlock(&cache_lock) 189 190 /* 191 * UMA zones for the VFS cache. 192 * 193 * The small cache is used for entries with short names, which are the 194 * most common. The large cache is used for entries which are too big to 195 * fit in the small cache. 196 */ 197 static uma_zone_t cache_zone_small; 198 static uma_zone_t cache_zone_small_ts; 199 static uma_zone_t cache_zone_large; 200 static uma_zone_t cache_zone_large_ts; 201 202 #define CACHE_PATH_CUTOFF 35 203 204 static struct namecache * 205 cache_alloc(int len, int ts) 206 { 207 208 if (len > CACHE_PATH_CUTOFF) { 209 if (ts) 210 return (uma_zalloc(cache_zone_large_ts, M_WAITOK)); 211 else 212 return (uma_zalloc(cache_zone_large, M_WAITOK)); 213 } 214 if (ts) 215 return (uma_zalloc(cache_zone_small_ts, M_WAITOK)); 216 else 217 return (uma_zalloc(cache_zone_small, M_WAITOK)); 218 } 219 220 static void 221 cache_free(struct namecache *ncp) 222 { 223 int ts; 224 225 if (ncp == NULL) 226 return; 227 ts = ncp->nc_flag & NCF_TS; 228 if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) { 229 if (ts) 230 uma_zfree(cache_zone_small_ts, ncp); 231 else 232 uma_zfree(cache_zone_small, ncp); 233 } else if (ts) 234 uma_zfree(cache_zone_large_ts, ncp); 235 else 236 uma_zfree(cache_zone_large, ncp); 237 } 238 239 static char * 240 nc_get_name(struct namecache *ncp) 241 { 242 struct namecache_ts *ncp_ts; 243 244 if ((ncp->nc_flag & NCF_TS) == 0) 245 return (ncp->nc_name); 246 ncp_ts = (struct namecache_ts *)ncp; 247 return (ncp_ts->nc_name); 248 } 249 250 static void 251 cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp) 252 { 253 254 KASSERT((ncp->nc_flag & NCF_TS) != 0 || 255 (tsp == NULL && ticksp == NULL), 256 ("No NCF_TS")); 257 258 if (tsp != NULL) 259 *tsp = ((struct namecache_ts *)ncp)->nc_time; 260 if (ticksp != NULL) 261 *ticksp = ((struct namecache_ts *)ncp)->nc_ticks; 262 } 263 264 static int doingcache = 1; /* 1 => enable the cache */ 265 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, 266 "VFS namecache enabled"); 267 268 /* Export size information to userland */ 269 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0, 270 sizeof(struct namecache), "sizeof(struct namecache)"); 271 272 /* 273 * The new name cache statistics 274 */ 275 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, 276 "Name cache statistics"); 277 #define STATNODE(mode, name, var, descr) \ 278 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, descr); 279 STATNODE(CTLFLAG_RD, numneg, &numneg, "Number of negative cache entries"); 280 STATNODE(CTLFLAG_RD, numcache, &numcache, "Number of cache entries"); 281 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls, 282 "Number of cache lookups"); 283 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits, 284 "Number of '.' hits"); 285 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits, 286 "Number of '..' hits"); 287 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks, 288 "Number of checks in lookup"); 289 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss, 290 "Number of cache misses"); 291 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap, 292 "Number of cache misses we do not want to cache"); 293 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps, 294 "Number of cache hits (positive) we do not want to cache"); 295 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits, 296 "Number of cache hits (positive)"); 297 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps, 298 "Number of cache hits (negative) we do not want to cache"); 299 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits, 300 "Number of cache hits (negative)"); 301 static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades, 302 "Number of updates of the cache after lookup (write lock + retry)"); 303 304 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE, 305 &nchstats, sizeof(nchstats), "LU", 306 "VFS cache effectiveness statistics"); 307 308 309 310 static void cache_zap(struct namecache *ncp); 311 static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf, 312 u_int *buflen); 313 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 314 char *buf, char **retbuf, u_int buflen); 315 316 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); 317 318 #ifdef DIAGNOSTIC 319 /* 320 * Grab an atomic snapshot of the name cache hash chain lengths 321 */ 322 SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats"); 323 324 static int 325 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) 326 { 327 int error; 328 struct nchashhead *ncpp; 329 struct namecache *ncp; 330 int n_nchash; 331 int count; 332 333 n_nchash = nchash + 1; /* nchash is max index, not count */ 334 if (!req->oldptr) 335 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); 336 337 /* Scan hash tables for applicable entries */ 338 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { 339 CACHE_RLOCK(); 340 count = 0; 341 LIST_FOREACH(ncp, ncpp, nc_hash) { 342 count++; 343 } 344 CACHE_RUNLOCK(); 345 error = SYSCTL_OUT(req, &count, sizeof(count)); 346 if (error) 347 return (error); 348 } 349 return (0); 350 } 351 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD| 352 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", 353 "nchash chain lengths"); 354 355 static int 356 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) 357 { 358 int error; 359 struct nchashhead *ncpp; 360 struct namecache *ncp; 361 int n_nchash; 362 int count, maxlength, used, pct; 363 364 if (!req->oldptr) 365 return SYSCTL_OUT(req, 0, 4 * sizeof(int)); 366 367 n_nchash = nchash + 1; /* nchash is max index, not count */ 368 used = 0; 369 maxlength = 0; 370 371 /* Scan hash tables for applicable entries */ 372 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { 373 count = 0; 374 CACHE_RLOCK(); 375 LIST_FOREACH(ncp, ncpp, nc_hash) { 376 count++; 377 } 378 CACHE_RUNLOCK(); 379 if (count) 380 used++; 381 if (maxlength < count) 382 maxlength = count; 383 } 384 n_nchash = nchash + 1; 385 pct = (used * 100 * 100) / n_nchash; 386 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); 387 if (error) 388 return (error); 389 error = SYSCTL_OUT(req, &used, sizeof(used)); 390 if (error) 391 return (error); 392 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); 393 if (error) 394 return (error); 395 error = SYSCTL_OUT(req, &pct, sizeof(pct)); 396 if (error) 397 return (error); 398 return (0); 399 } 400 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD| 401 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I", 402 "nchash chain lengths"); 403 #endif 404 405 /* 406 * cache_zap(): 407 * 408 * Removes a namecache entry from cache, whether it contains an actual 409 * pointer to a vnode or if it is just a negative cache entry. 410 */ 411 static void 412 cache_zap(ncp) 413 struct namecache *ncp; 414 { 415 struct vnode *vp; 416 417 rw_assert(&cache_lock, RA_WLOCKED); 418 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp); 419 #ifdef KDTRACE_HOOKS 420 if (ncp->nc_vp != NULL) { 421 SDT_PROBE(vfs, namecache, zap, done, ncp->nc_dvp, 422 nc_get_name(ncp), ncp->nc_vp, 0, 0); 423 } else { 424 SDT_PROBE(vfs, namecache, zap_negative, done, ncp->nc_dvp, 425 nc_get_name(ncp), 0, 0, 0); 426 } 427 #endif 428 vp = NULL; 429 LIST_REMOVE(ncp, nc_hash); 430 if (ncp->nc_flag & NCF_ISDOTDOT) { 431 if (ncp == ncp->nc_dvp->v_cache_dd) 432 ncp->nc_dvp->v_cache_dd = NULL; 433 } else { 434 LIST_REMOVE(ncp, nc_src); 435 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) { 436 vp = ncp->nc_dvp; 437 numcachehv--; 438 } 439 } 440 if (ncp->nc_vp) { 441 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); 442 if (ncp == ncp->nc_vp->v_cache_dd) 443 ncp->nc_vp->v_cache_dd = NULL; 444 } else { 445 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 446 numneg--; 447 } 448 numcache--; 449 cache_free(ncp); 450 if (vp) 451 vdrop(vp); 452 } 453 454 /* 455 * Lookup an entry in the cache 456 * 457 * Lookup is called with dvp pointing to the directory to search, 458 * cnp pointing to the name of the entry being sought. If the lookup 459 * succeeds, the vnode is returned in *vpp, and a status of -1 is 460 * returned. If the lookup determines that the name does not exist 461 * (negative cacheing), a status of ENOENT is returned. If the lookup 462 * fails, a status of zero is returned. If the directory vnode is 463 * recycled out from under us due to a forced unmount, a status of 464 * ENOENT is returned. 465 * 466 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is 467 * unlocked. If we're looking up . an extra ref is taken, but the lock is 468 * not recursively acquired. 469 */ 470 471 int 472 cache_lookup_times(dvp, vpp, cnp, tsp, ticksp) 473 struct vnode *dvp; 474 struct vnode **vpp; 475 struct componentname *cnp; 476 struct timespec *tsp; 477 int *ticksp; 478 { 479 struct namecache *ncp; 480 u_int32_t hash; 481 int error, ltype, wlocked; 482 483 if (!doingcache) { 484 cnp->cn_flags &= ~MAKEENTRY; 485 return (0); 486 } 487 retry: 488 CACHE_RLOCK(); 489 wlocked = 0; 490 numcalls++; 491 error = 0; 492 493 retry_wlocked: 494 if (cnp->cn_nameptr[0] == '.') { 495 if (cnp->cn_namelen == 1) { 496 *vpp = dvp; 497 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .", 498 dvp, cnp->cn_nameptr); 499 dothits++; 500 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ".", 501 *vpp, 0, 0); 502 if (tsp != NULL) 503 timespecclear(tsp); 504 if (ticksp != NULL) 505 *ticksp = ticks; 506 goto success; 507 } 508 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 509 dotdothits++; 510 if (dvp->v_cache_dd == NULL) { 511 SDT_PROBE(vfs, namecache, lookup, miss, dvp, 512 "..", NULL, 0, 0); 513 goto unlock; 514 } 515 if ((cnp->cn_flags & MAKEENTRY) == 0) { 516 if (!wlocked && !CACHE_UPGRADE_LOCK()) 517 goto wlock; 518 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT) 519 cache_zap(dvp->v_cache_dd); 520 dvp->v_cache_dd = NULL; 521 CACHE_WUNLOCK(); 522 return (0); 523 } 524 ncp = dvp->v_cache_dd; 525 if (ncp->nc_flag & NCF_ISDOTDOT) 526 *vpp = ncp->nc_vp; 527 else 528 *vpp = ncp->nc_dvp; 529 /* Return failure if negative entry was found. */ 530 if (*vpp == NULL) 531 goto negative_success; 532 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..", 533 dvp, cnp->cn_nameptr, *vpp); 534 SDT_PROBE(vfs, namecache, lookup, hit, dvp, "..", 535 *vpp, 0, 0); 536 cache_out_ts(ncp, tsp, ticksp); 537 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == 538 NCF_DTS && tsp != NULL) 539 *tsp = ((struct namecache_ts *)ncp)-> 540 nc_dotdottime; 541 goto success; 542 } 543 } 544 545 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT); 546 hash = fnv_32_buf(&dvp, sizeof(dvp), hash); 547 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { 548 numchecks++; 549 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && 550 !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen)) 551 break; 552 } 553 554 /* We failed to find an entry */ 555 if (ncp == NULL) { 556 SDT_PROBE(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, 557 NULL, 0, 0); 558 if ((cnp->cn_flags & MAKEENTRY) == 0) { 559 nummisszap++; 560 } else { 561 nummiss++; 562 } 563 nchstats.ncs_miss++; 564 goto unlock; 565 } 566 567 /* We don't want to have an entry, so dump it */ 568 if ((cnp->cn_flags & MAKEENTRY) == 0) { 569 numposzaps++; 570 nchstats.ncs_badhits++; 571 if (!wlocked && !CACHE_UPGRADE_LOCK()) 572 goto wlock; 573 cache_zap(ncp); 574 CACHE_WUNLOCK(); 575 return (0); 576 } 577 578 /* We found a "positive" match, return the vnode */ 579 if (ncp->nc_vp) { 580 numposhits++; 581 nchstats.ncs_goodhits++; 582 *vpp = ncp->nc_vp; 583 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p", 584 dvp, cnp->cn_nameptr, *vpp, ncp); 585 SDT_PROBE(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp), 586 *vpp, 0, 0); 587 cache_out_ts(ncp, tsp, ticksp); 588 goto success; 589 } 590 591 negative_success: 592 /* We found a negative match, and want to create it, so purge */ 593 if (cnp->cn_nameiop == CREATE) { 594 numnegzaps++; 595 nchstats.ncs_badhits++; 596 if (!wlocked && !CACHE_UPGRADE_LOCK()) 597 goto wlock; 598 cache_zap(ncp); 599 CACHE_WUNLOCK(); 600 return (0); 601 } 602 603 if (!wlocked && !CACHE_UPGRADE_LOCK()) 604 goto wlock; 605 numneghits++; 606 /* 607 * We found a "negative" match, so we shift it to the end of 608 * the "negative" cache entries queue to satisfy LRU. Also, 609 * check to see if the entry is a whiteout; indicate this to 610 * the componentname, if so. 611 */ 612 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 613 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 614 nchstats.ncs_neghits++; 615 if (ncp->nc_flag & NCF_WHITE) 616 cnp->cn_flags |= ISWHITEOUT; 617 SDT_PROBE(vfs, namecache, lookup, hit_negative, dvp, nc_get_name(ncp), 618 0, 0, 0); 619 cache_out_ts(ncp, tsp, ticksp); 620 CACHE_WUNLOCK(); 621 return (ENOENT); 622 623 wlock: 624 /* 625 * We need to update the cache after our lookup, so upgrade to 626 * a write lock and retry the operation. 627 */ 628 CACHE_RUNLOCK(); 629 CACHE_WLOCK(); 630 numupgrades++; 631 wlocked = 1; 632 goto retry_wlocked; 633 634 success: 635 /* 636 * On success we return a locked and ref'd vnode as per the lookup 637 * protocol. 638 */ 639 if (dvp == *vpp) { /* lookup on "." */ 640 VREF(*vpp); 641 if (wlocked) 642 CACHE_WUNLOCK(); 643 else 644 CACHE_RUNLOCK(); 645 /* 646 * When we lookup "." we still can be asked to lock it 647 * differently... 648 */ 649 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 650 if (ltype != VOP_ISLOCKED(*vpp)) { 651 if (ltype == LK_EXCLUSIVE) { 652 vn_lock(*vpp, LK_UPGRADE | LK_RETRY); 653 if ((*vpp)->v_iflag & VI_DOOMED) { 654 /* forced unmount */ 655 vrele(*vpp); 656 *vpp = NULL; 657 return (ENOENT); 658 } 659 } else 660 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); 661 } 662 return (-1); 663 } 664 ltype = 0; /* silence gcc warning */ 665 if (cnp->cn_flags & ISDOTDOT) { 666 ltype = VOP_ISLOCKED(dvp); 667 VOP_UNLOCK(dvp, 0); 668 } 669 VI_LOCK(*vpp); 670 if (wlocked) 671 CACHE_WUNLOCK(); 672 else 673 CACHE_RUNLOCK(); 674 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread); 675 if (cnp->cn_flags & ISDOTDOT) { 676 vn_lock(dvp, ltype | LK_RETRY); 677 if (dvp->v_iflag & VI_DOOMED) { 678 if (error == 0) 679 vput(*vpp); 680 *vpp = NULL; 681 return (ENOENT); 682 } 683 } 684 if (error) { 685 *vpp = NULL; 686 goto retry; 687 } 688 if ((cnp->cn_flags & ISLASTCN) && 689 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) { 690 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup"); 691 } 692 return (-1); 693 694 unlock: 695 if (wlocked) 696 CACHE_WUNLOCK(); 697 else 698 CACHE_RUNLOCK(); 699 return (0); 700 } 701 702 /* 703 * Add an entry to the cache. 704 */ 705 void 706 cache_enter_time(dvp, vp, cnp, tsp, dtsp) 707 struct vnode *dvp; 708 struct vnode *vp; 709 struct componentname *cnp; 710 struct timespec *tsp; 711 struct timespec *dtsp; 712 { 713 struct namecache *ncp, *n2; 714 struct namecache_ts *n3; 715 struct nchashhead *ncpp; 716 u_int32_t hash; 717 int flag; 718 int hold; 719 int zap; 720 int len; 721 722 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr); 723 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp, 724 ("cache_enter: Adding a doomed vnode")); 725 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp, 726 ("cache_enter: Doomed vnode used as src")); 727 728 if (!doingcache) 729 return; 730 731 /* 732 * Avoid blowout in namecache entries. 733 */ 734 if (numcache >= desiredvnodes * ncsizefactor) 735 return; 736 737 flag = 0; 738 if (cnp->cn_nameptr[0] == '.') { 739 if (cnp->cn_namelen == 1) 740 return; 741 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 742 CACHE_WLOCK(); 743 /* 744 * If dotdot entry already exists, just retarget it 745 * to new parent vnode, otherwise continue with new 746 * namecache entry allocation. 747 */ 748 if ((ncp = dvp->v_cache_dd) != NULL && 749 ncp->nc_flag & NCF_ISDOTDOT) { 750 KASSERT(ncp->nc_dvp == dvp, 751 ("wrong isdotdot parent")); 752 if (ncp->nc_vp != NULL) 753 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, 754 ncp, nc_dst); 755 else 756 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 757 if (vp != NULL) 758 TAILQ_INSERT_HEAD(&vp->v_cache_dst, 759 ncp, nc_dst); 760 else 761 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 762 ncp->nc_vp = vp; 763 CACHE_WUNLOCK(); 764 return; 765 } 766 dvp->v_cache_dd = NULL; 767 SDT_PROBE(vfs, namecache, enter, done, dvp, "..", vp, 768 0, 0); 769 CACHE_WUNLOCK(); 770 flag = NCF_ISDOTDOT; 771 } 772 } 773 774 hold = 0; 775 zap = 0; 776 777 /* 778 * Calculate the hash key and setup as much of the new 779 * namecache entry as possible before acquiring the lock. 780 */ 781 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); 782 ncp->nc_vp = vp; 783 ncp->nc_dvp = dvp; 784 ncp->nc_flag = flag; 785 if (tsp != NULL) { 786 n3 = (struct namecache_ts *)ncp; 787 n3->nc_time = *tsp; 788 n3->nc_ticks = ticks; 789 n3->nc_flag |= NCF_TS; 790 if (dtsp != NULL) { 791 n3->nc_dotdottime = *dtsp; 792 n3->nc_flag |= NCF_DTS; 793 } 794 } 795 len = ncp->nc_nlen = cnp->cn_namelen; 796 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT); 797 strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1); 798 hash = fnv_32_buf(&dvp, sizeof(dvp), hash); 799 CACHE_WLOCK(); 800 801 /* 802 * See if this vnode or negative entry is already in the cache 803 * with this name. This can happen with concurrent lookups of 804 * the same path name. 805 */ 806 ncpp = NCHHASH(hash); 807 LIST_FOREACH(n2, ncpp, nc_hash) { 808 if (n2->nc_dvp == dvp && 809 n2->nc_nlen == cnp->cn_namelen && 810 !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) { 811 if (tsp != NULL) { 812 KASSERT((n2->nc_flag & NCF_TS) != 0, 813 ("no NCF_TS")); 814 n3 = (struct namecache_ts *)n2; 815 n3->nc_time = 816 ((struct namecache_ts *)ncp)->nc_time; 817 n3->nc_ticks = 818 ((struct namecache_ts *)ncp)->nc_ticks; 819 if (dtsp != NULL) { 820 n3->nc_dotdottime = 821 ((struct namecache_ts *)ncp)-> 822 nc_dotdottime; 823 n3->nc_flag |= NCF_DTS; 824 } 825 } 826 CACHE_WUNLOCK(); 827 cache_free(ncp); 828 return; 829 } 830 } 831 832 if (flag == NCF_ISDOTDOT) { 833 /* 834 * See if we are trying to add .. entry, but some other lookup 835 * has populated v_cache_dd pointer already. 836 */ 837 if (dvp->v_cache_dd != NULL) { 838 CACHE_WUNLOCK(); 839 cache_free(ncp); 840 return; 841 } 842 KASSERT(vp == NULL || vp->v_type == VDIR, 843 ("wrong vnode type %p", vp)); 844 dvp->v_cache_dd = ncp; 845 } 846 847 numcache++; 848 if (!vp) { 849 numneg++; 850 if (cnp->cn_flags & ISWHITEOUT) 851 ncp->nc_flag |= NCF_WHITE; 852 } else if (vp->v_type == VDIR) { 853 if (flag != NCF_ISDOTDOT) { 854 /* 855 * For this case, the cache entry maps both the 856 * directory name in it and the name ".." for the 857 * directory's parent. 858 */ 859 if ((n2 = vp->v_cache_dd) != NULL && 860 (n2->nc_flag & NCF_ISDOTDOT) != 0) 861 cache_zap(n2); 862 vp->v_cache_dd = ncp; 863 } 864 } else { 865 vp->v_cache_dd = NULL; 866 } 867 868 /* 869 * Insert the new namecache entry into the appropriate chain 870 * within the cache entries table. 871 */ 872 LIST_INSERT_HEAD(ncpp, ncp, nc_hash); 873 if (flag != NCF_ISDOTDOT) { 874 if (LIST_EMPTY(&dvp->v_cache_src)) { 875 hold = 1; 876 numcachehv++; 877 } 878 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); 879 } 880 881 /* 882 * If the entry is "negative", we place it into the 883 * "negative" cache queue, otherwise, we place it into the 884 * destination vnode's cache entries queue. 885 */ 886 if (vp) { 887 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); 888 SDT_PROBE(vfs, namecache, enter, done, dvp, nc_get_name(ncp), 889 vp, 0, 0); 890 } else { 891 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 892 SDT_PROBE(vfs, namecache, enter_negative, done, dvp, 893 nc_get_name(ncp), 0, 0, 0); 894 } 895 if (numneg * ncnegfactor > numcache) { 896 ncp = TAILQ_FIRST(&ncneg); 897 zap = 1; 898 } 899 if (hold) 900 vhold(dvp); 901 if (zap) 902 cache_zap(ncp); 903 CACHE_WUNLOCK(); 904 } 905 906 /* 907 * Name cache initialization, from vfs_init() when we are booting 908 */ 909 static void 910 nchinit(void *dummy __unused) 911 { 912 913 TAILQ_INIT(&ncneg); 914 915 cache_zone_small = uma_zcreate("S VFS Cache", 916 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1, 917 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 918 cache_zone_small_ts = uma_zcreate("STS VFS Cache", 919 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1, 920 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 921 cache_zone_large = uma_zcreate("L VFS Cache", 922 sizeof(struct namecache) + NAME_MAX + 1, 923 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 924 cache_zone_large_ts = uma_zcreate("LTS VFS Cache", 925 sizeof(struct namecache_ts) + NAME_MAX + 1, 926 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 927 928 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash); 929 } 930 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); 931 932 933 /* 934 * Invalidate all entries to a particular vnode. 935 */ 936 void 937 cache_purge(vp) 938 struct vnode *vp; 939 { 940 941 CTR1(KTR_VFS, "cache_purge(%p)", vp); 942 SDT_PROBE(vfs, namecache, purge, done, vp, 0, 0, 0, 0); 943 CACHE_WLOCK(); 944 while (!LIST_EMPTY(&vp->v_cache_src)) 945 cache_zap(LIST_FIRST(&vp->v_cache_src)); 946 while (!TAILQ_EMPTY(&vp->v_cache_dst)) 947 cache_zap(TAILQ_FIRST(&vp->v_cache_dst)); 948 if (vp->v_cache_dd != NULL) { 949 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT, 950 ("lost dotdot link")); 951 cache_zap(vp->v_cache_dd); 952 } 953 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); 954 CACHE_WUNLOCK(); 955 } 956 957 /* 958 * Invalidate all negative entries for a particular directory vnode. 959 */ 960 void 961 cache_purge_negative(vp) 962 struct vnode *vp; 963 { 964 struct namecache *cp, *ncp; 965 966 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp); 967 SDT_PROBE(vfs, namecache, purge_negative, done, vp, 0, 0, 0, 0); 968 CACHE_WLOCK(); 969 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) { 970 if (cp->nc_vp == NULL) 971 cache_zap(cp); 972 } 973 CACHE_WUNLOCK(); 974 } 975 976 /* 977 * Flush all entries referencing a particular filesystem. 978 */ 979 void 980 cache_purgevfs(mp) 981 struct mount *mp; 982 { 983 struct nchashhead *ncpp; 984 struct namecache *ncp, *nnp; 985 986 /* Scan hash tables for applicable entries */ 987 SDT_PROBE(vfs, namecache, purgevfs, done, mp, 0, 0, 0, 0); 988 CACHE_WLOCK(); 989 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) { 990 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) { 991 if (ncp->nc_dvp->v_mount == mp) 992 cache_zap(ncp); 993 } 994 } 995 CACHE_WUNLOCK(); 996 } 997 998 /* 999 * Perform canonical checks and cache lookup and pass on to filesystem 1000 * through the vop_cachedlookup only if needed. 1001 */ 1002 1003 int 1004 vfs_cache_lookup(ap) 1005 struct vop_lookup_args /* { 1006 struct vnode *a_dvp; 1007 struct vnode **a_vpp; 1008 struct componentname *a_cnp; 1009 } */ *ap; 1010 { 1011 struct vnode *dvp; 1012 int error; 1013 struct vnode **vpp = ap->a_vpp; 1014 struct componentname *cnp = ap->a_cnp; 1015 struct ucred *cred = cnp->cn_cred; 1016 int flags = cnp->cn_flags; 1017 struct thread *td = cnp->cn_thread; 1018 1019 *vpp = NULL; 1020 dvp = ap->a_dvp; 1021 1022 if (dvp->v_type != VDIR) 1023 return (ENOTDIR); 1024 1025 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 1026 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1027 return (EROFS); 1028 1029 error = VOP_ACCESS(dvp, VEXEC, cred, td); 1030 if (error) 1031 return (error); 1032 1033 error = cache_lookup(dvp, vpp, cnp); 1034 if (error == 0) 1035 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); 1036 if (error == -1) 1037 return (0); 1038 return (error); 1039 } 1040 1041 1042 #ifndef _SYS_SYSPROTO_H_ 1043 struct __getcwd_args { 1044 u_char *buf; 1045 u_int buflen; 1046 }; 1047 #endif 1048 1049 /* 1050 * XXX All of these sysctls would probably be more productive dead. 1051 */ 1052 static int disablecwd; 1053 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, 1054 "Disable the getcwd syscall"); 1055 1056 /* Implementation of the getcwd syscall. */ 1057 int 1058 __getcwd(td, uap) 1059 struct thread *td; 1060 struct __getcwd_args *uap; 1061 { 1062 1063 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen)); 1064 } 1065 1066 int 1067 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen) 1068 { 1069 char *bp, *tmpbuf; 1070 struct filedesc *fdp; 1071 struct vnode *cdir, *rdir; 1072 int error, vfslocked; 1073 1074 if (disablecwd) 1075 return (ENODEV); 1076 if (buflen < 2) 1077 return (EINVAL); 1078 if (buflen > MAXPATHLEN) 1079 buflen = MAXPATHLEN; 1080 1081 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK); 1082 fdp = td->td_proc->p_fd; 1083 FILEDESC_SLOCK(fdp); 1084 cdir = fdp->fd_cdir; 1085 VREF(cdir); 1086 rdir = fdp->fd_rdir; 1087 VREF(rdir); 1088 FILEDESC_SUNLOCK(fdp); 1089 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen); 1090 vfslocked = VFS_LOCK_GIANT(rdir->v_mount); 1091 vrele(rdir); 1092 VFS_UNLOCK_GIANT(vfslocked); 1093 vfslocked = VFS_LOCK_GIANT(cdir->v_mount); 1094 vrele(cdir); 1095 VFS_UNLOCK_GIANT(vfslocked); 1096 1097 if (!error) { 1098 if (bufseg == UIO_SYSSPACE) 1099 bcopy(bp, buf, strlen(bp) + 1); 1100 else 1101 error = copyout(bp, buf, strlen(bp) + 1); 1102 #ifdef KTRACE 1103 if (KTRPOINT(curthread, KTR_NAMEI)) 1104 ktrnamei(bp); 1105 #endif 1106 } 1107 free(tmpbuf, M_TEMP); 1108 return (error); 1109 } 1110 1111 /* 1112 * Thus begins the fullpath magic. 1113 */ 1114 1115 #undef STATNODE 1116 #define STATNODE(name, descr) \ 1117 static u_int name; \ 1118 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr) 1119 1120 static int disablefullpath; 1121 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0, 1122 "Disable the vn_fullpath function"); 1123 1124 /* These count for kern___getcwd(), too. */ 1125 STATNODE(numfullpathcalls, "Number of fullpath search calls"); 1126 STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)"); 1127 STATNODE(numfullpathfail2, 1128 "Number of fullpath search errors (VOP_VPTOCNP failures)"); 1129 STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)"); 1130 STATNODE(numfullpathfound, "Number of successful fullpath calls"); 1131 1132 /* 1133 * Retrieve the full filesystem path that correspond to a vnode from the name 1134 * cache (if available) 1135 */ 1136 int 1137 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) 1138 { 1139 char *buf; 1140 struct filedesc *fdp; 1141 struct vnode *rdir; 1142 int error, vfslocked; 1143 1144 if (disablefullpath) 1145 return (ENODEV); 1146 if (vn == NULL) 1147 return (EINVAL); 1148 1149 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1150 fdp = td->td_proc->p_fd; 1151 FILEDESC_SLOCK(fdp); 1152 rdir = fdp->fd_rdir; 1153 VREF(rdir); 1154 FILEDESC_SUNLOCK(fdp); 1155 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN); 1156 vfslocked = VFS_LOCK_GIANT(rdir->v_mount); 1157 vrele(rdir); 1158 VFS_UNLOCK_GIANT(vfslocked); 1159 1160 if (!error) 1161 *freebuf = buf; 1162 else 1163 free(buf, M_TEMP); 1164 return (error); 1165 } 1166 1167 /* 1168 * This function is similar to vn_fullpath, but it attempts to lookup the 1169 * pathname relative to the global root mount point. This is required for the 1170 * auditing sub-system, as audited pathnames must be absolute, relative to the 1171 * global root mount point. 1172 */ 1173 int 1174 vn_fullpath_global(struct thread *td, struct vnode *vn, 1175 char **retbuf, char **freebuf) 1176 { 1177 char *buf; 1178 int error; 1179 1180 if (disablefullpath) 1181 return (ENODEV); 1182 if (vn == NULL) 1183 return (EINVAL); 1184 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1185 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN); 1186 if (!error) 1187 *freebuf = buf; 1188 else 1189 free(buf, M_TEMP); 1190 return (error); 1191 } 1192 1193 int 1194 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen) 1195 { 1196 int error; 1197 1198 CACHE_RLOCK(); 1199 error = vn_vptocnp_locked(vp, cred, buf, buflen); 1200 if (error == 0) { 1201 /* 1202 * vn_vptocnp_locked() dropped hold acquired by 1203 * VOP_VPTOCNP immediately after locking the 1204 * cache. Since we are going to drop the cache rlock, 1205 * re-hold the result. 1206 */ 1207 vhold(*vp); 1208 CACHE_RUNLOCK(); 1209 } 1210 return (error); 1211 } 1212 1213 static int 1214 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf, 1215 u_int *buflen) 1216 { 1217 struct vnode *dvp; 1218 struct namecache *ncp; 1219 int error, vfslocked; 1220 1221 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) { 1222 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 1223 break; 1224 } 1225 if (ncp != NULL) { 1226 if (*buflen < ncp->nc_nlen) { 1227 CACHE_RUNLOCK(); 1228 numfullpathfail4++; 1229 error = ENOMEM; 1230 SDT_PROBE(vfs, namecache, fullpath, return, error, 1231 vp, NULL, 0, 0); 1232 return (error); 1233 } 1234 *buflen -= ncp->nc_nlen; 1235 memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen); 1236 SDT_PROBE(vfs, namecache, fullpath, hit, ncp->nc_dvp, 1237 nc_get_name(ncp), vp, 0, 0); 1238 *vp = ncp->nc_dvp; 1239 return (0); 1240 } 1241 SDT_PROBE(vfs, namecache, fullpath, miss, vp, 0, 0, 0, 0); 1242 1243 vhold(*vp); 1244 CACHE_RUNLOCK(); 1245 vfslocked = VFS_LOCK_GIANT((*vp)->v_mount); 1246 vn_lock(*vp, LK_SHARED | LK_RETRY); 1247 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen); 1248 VOP_UNLOCK(*vp, 0); 1249 vdrop(*vp); 1250 VFS_UNLOCK_GIANT(vfslocked); 1251 if (error) { 1252 numfullpathfail2++; 1253 SDT_PROBE(vfs, namecache, fullpath, return, error, vp, 1254 NULL, 0, 0); 1255 return (error); 1256 } 1257 1258 *vp = dvp; 1259 CACHE_RLOCK(); 1260 if ((*vp)->v_iflag & VI_DOOMED) { 1261 /* forced unmount */ 1262 CACHE_RUNLOCK(); 1263 vdrop(*vp); 1264 error = ENOENT; 1265 SDT_PROBE(vfs, namecache, fullpath, return, error, vp, 1266 NULL, 0, 0); 1267 return (error); 1268 } 1269 vdrop(*vp); 1270 1271 return (0); 1272 } 1273 1274 /* 1275 * The magic behind kern___getcwd() and vn_fullpath(). 1276 */ 1277 static int 1278 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 1279 char *buf, char **retbuf, u_int buflen) 1280 { 1281 int error, slash_prefixed; 1282 #ifdef KDTRACE_HOOKS 1283 struct vnode *startvp = vp; 1284 #endif 1285 1286 buflen--; 1287 buf[buflen] = '\0'; 1288 error = 0; 1289 slash_prefixed = 0; 1290 1291 SDT_PROBE(vfs, namecache, fullpath, entry, vp, 0, 0, 0, 0); 1292 numfullpathcalls++; 1293 CACHE_RLOCK(); 1294 if (vp->v_type != VDIR) { 1295 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen); 1296 if (error) 1297 return (error); 1298 if (buflen == 0) { 1299 CACHE_RUNLOCK(); 1300 return (ENOMEM); 1301 } 1302 buf[--buflen] = '/'; 1303 slash_prefixed = 1; 1304 } 1305 while (vp != rdir && vp != rootvnode) { 1306 if (vp->v_vflag & VV_ROOT) { 1307 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */ 1308 CACHE_RUNLOCK(); 1309 error = ENOENT; 1310 SDT_PROBE(vfs, namecache, fullpath, return, 1311 error, vp, NULL, 0, 0); 1312 break; 1313 } 1314 vp = vp->v_mount->mnt_vnodecovered; 1315 continue; 1316 } 1317 if (vp->v_type != VDIR) { 1318 CACHE_RUNLOCK(); 1319 numfullpathfail1++; 1320 error = ENOTDIR; 1321 SDT_PROBE(vfs, namecache, fullpath, return, 1322 error, vp, NULL, 0, 0); 1323 break; 1324 } 1325 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen); 1326 if (error) 1327 break; 1328 if (buflen == 0) { 1329 CACHE_RUNLOCK(); 1330 error = ENOMEM; 1331 SDT_PROBE(vfs, namecache, fullpath, return, error, 1332 startvp, NULL, 0, 0); 1333 break; 1334 } 1335 buf[--buflen] = '/'; 1336 slash_prefixed = 1; 1337 } 1338 if (error) 1339 return (error); 1340 if (!slash_prefixed) { 1341 if (buflen == 0) { 1342 CACHE_RUNLOCK(); 1343 numfullpathfail4++; 1344 SDT_PROBE(vfs, namecache, fullpath, return, ENOMEM, 1345 startvp, NULL, 0, 0); 1346 return (ENOMEM); 1347 } 1348 buf[--buflen] = '/'; 1349 } 1350 numfullpathfound++; 1351 CACHE_RUNLOCK(); 1352 1353 SDT_PROBE(vfs, namecache, fullpath, return, 0, startvp, buf + buflen, 1354 0, 0); 1355 *retbuf = buf + buflen; 1356 return (0); 1357 } 1358 1359 int 1360 vn_commname(struct vnode *vp, char *buf, u_int buflen) 1361 { 1362 struct namecache *ncp; 1363 int l; 1364 1365 CACHE_RLOCK(); 1366 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) 1367 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 1368 break; 1369 if (ncp == NULL) { 1370 CACHE_RUNLOCK(); 1371 return (ENOENT); 1372 } 1373 l = min(ncp->nc_nlen, buflen - 1); 1374 memcpy(buf, nc_get_name(ncp), l); 1375 CACHE_RUNLOCK(); 1376 buf[l] = '\0'; 1377 return (0); 1378 } 1379 1380 /* ABI compat shims for old kernel modules. */ 1381 #undef cache_enter 1382 #undef cache_lookup 1383 1384 void cache_enter(struct vnode *dvp, struct vnode *vp, 1385 struct componentname *cnp); 1386 int cache_lookup(struct vnode *dvp, struct vnode **vpp, 1387 struct componentname *cnp); 1388 1389 void 1390 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 1391 { 1392 1393 cache_enter_time(dvp, vp, cnp, NULL, NULL); 1394 } 1395 1396 int 1397 cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) 1398 { 1399 1400 return (cache_lookup_times(dvp, vpp, cnp, NULL, NULL)); 1401 }
Cache object: fd14c559d4daa818cbf70ed5e1b90c8b
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]