vfs_cache.c revision 302237
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: stable/10/sys/kern/vfs_cache.c 302237 2016-06-27 22:10:07Z bdrewery $"); 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/fcntl.h> 49#include <sys/mount.h> 50#include <sys/namei.h> 51#include <sys/proc.h> 52#include <sys/rwlock.h> 53#include <sys/sdt.h> 54#include <sys/syscallsubr.h> 55#include <sys/sysctl.h> 56#include <sys/sysproto.h> 57#include <sys/vnode.h> 58#ifdef KTRACE 59#include <sys/ktrace.h> 60#endif 61 62#include <vm/uma.h> 63 64SDT_PROVIDER_DECLARE(vfs); 65SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *", 66 "struct vnode *"); 67SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *", 68 "char *"); 69SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *"); 70SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *", 71 "char *", "struct vnode *"); 72SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *"); 73SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", 74 "struct vnode *", "char *"); 75SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *", 76 "struct vnode *"); 77SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative, 78 "struct vnode *", "char *"); 79SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *", 80 "char *"); 81SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *"); 82SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *"); 83SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *"); 84SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *", 85 "struct vnode *"); 86SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *", 87 "char *"); 88 89/* 90 * This structure describes the elements in the cache of recent 91 * names looked up by namei. 92 */ 93 94struct namecache { 95 LIST_ENTRY(namecache) nc_hash; /* hash chain */ 96 LIST_ENTRY(namecache) nc_src; /* source vnode list */ 97 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ 98 struct vnode *nc_dvp; /* vnode of parent of name */ 99 struct vnode *nc_vp; /* vnode the name refers to */ 100 u_char nc_flag; /* flag bits */ 101 u_char nc_nlen; /* length of name */ 102 char nc_name[0]; /* segment name + nul */ 103}; 104 105/* 106 * struct namecache_ts repeats struct namecache layout up to the 107 * nc_nlen member. 108 * struct namecache_ts is used in place of struct namecache when time(s) need 109 * to be stored. The nc_dotdottime field is used when a cache entry is mapping 110 * both a non-dotdot directory name plus dotdot for the directory's 111 * parent. 112 */ 113struct namecache_ts { 114 LIST_ENTRY(namecache) nc_hash; /* hash chain */ 115 LIST_ENTRY(namecache) nc_src; /* source vnode list */ 116 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ 117 struct vnode *nc_dvp; /* vnode of parent of name */ 118 struct vnode *nc_vp; /* vnode the name refers to */ 119 u_char nc_flag; /* flag bits */ 120 u_char nc_nlen; /* length of name */ 121 struct timespec nc_time; /* timespec provided by fs */ 122 struct timespec nc_dotdottime; /* dotdot timespec provided by fs */ 123 int nc_ticks; /* ticks value when entry was added */ 124 char nc_name[0]; /* segment name + nul */ 125}; 126 127/* 128 * Flags in namecache.nc_flag 129 */ 130#define NCF_WHITE 0x01 131#define NCF_ISDOTDOT 0x02 132#define NCF_TS 0x04 133#define NCF_DTS 0x08 134 135/* 136 * Name caching works as follows: 137 * 138 * Names found by directory scans are retained in a cache 139 * for future reference. It is managed LRU, so frequently 140 * used names will hang around. Cache is indexed by hash value 141 * obtained from (vp, name) where vp refers to the directory 142 * containing name. 143 * 144 * If it is a "negative" entry, (i.e. for a name that is known NOT to 145 * exist) the vnode pointer will be NULL. 146 * 147 * Upon reaching the last segment of a path, if the reference 148 * is for DELETE, or NOCACHE is set (rewrite), and the 149 * name is located in the cache, it will be dropped. 150 */ 151 152/* 153 * Structures associated with name caching. 154 */ 155#define NCHHASH(hash) \ 156 (&nchashtbl[(hash) & nchash]) 157static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */ 158static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */ 159static u_long nchash; /* size of hash table */ 160SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, 161 "Size of namecache hash table"); 162static u_long ncnegfactor = 16; /* ratio of negative entries */ 163SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, 164 "Ratio of negative namecache entries"); 165static u_long numneg; /* number of negative entries allocated */ 166SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, 167 "Number of negative entries in namecache"); 168static u_long numcache; /* number of cache entries allocated */ 169SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, 170 "Number of namecache entries"); 171static u_long numcachehv; /* number of cache entries with vnodes held */ 172SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, 173 "Number of namecache entries with vnodes held"); 174static u_int ncsizefactor = 2; 175SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0, 176 "Size factor for namecache"); 177 178struct nchstats nchstats; /* cache effectiveness statistics */ 179 180static struct rwlock cache_lock; 181RW_SYSINIT(vfscache, &cache_lock, "Name Cache"); 182 183#define CACHE_UPGRADE_LOCK() rw_try_upgrade(&cache_lock) 184#define CACHE_RLOCK() rw_rlock(&cache_lock) 185#define CACHE_RUNLOCK() rw_runlock(&cache_lock) 186#define CACHE_WLOCK() rw_wlock(&cache_lock) 187#define CACHE_WUNLOCK() rw_wunlock(&cache_lock) 188 189/* 190 * UMA zones for the VFS cache. 191 * 192 * The small cache is used for entries with short names, which are the 193 * most common. The large cache is used for entries which are too big to 194 * fit in the small cache. 195 */ 196static uma_zone_t cache_zone_small; 197static uma_zone_t cache_zone_small_ts; 198static uma_zone_t cache_zone_large; 199static uma_zone_t cache_zone_large_ts; 200 201#define CACHE_PATH_CUTOFF 35 202 203static struct namecache * 204cache_alloc(int len, int ts) 205{ 206 207 if (len > CACHE_PATH_CUTOFF) { 208 if (ts) 209 return (uma_zalloc(cache_zone_large_ts, M_WAITOK)); 210 else 211 return (uma_zalloc(cache_zone_large, M_WAITOK)); 212 } 213 if (ts) 214 return (uma_zalloc(cache_zone_small_ts, M_WAITOK)); 215 else 216 return (uma_zalloc(cache_zone_small, M_WAITOK)); 217} 218 219static void 220cache_free(struct namecache *ncp) 221{ 222 int ts; 223 224 if (ncp == NULL) 225 return; 226 ts = ncp->nc_flag & NCF_TS; 227 if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) { 228 if (ts) 229 uma_zfree(cache_zone_small_ts, ncp); 230 else 231 uma_zfree(cache_zone_small, ncp); 232 } else if (ts) 233 uma_zfree(cache_zone_large_ts, ncp); 234 else 235 uma_zfree(cache_zone_large, ncp); 236} 237 238static char * 239nc_get_name(struct namecache *ncp) 240{ 241 struct namecache_ts *ncp_ts; 242 243 if ((ncp->nc_flag & NCF_TS) == 0) 244 return (ncp->nc_name); 245 ncp_ts = (struct namecache_ts *)ncp; 246 return (ncp_ts->nc_name); 247} 248 249static void 250cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp) 251{ 252 253 KASSERT((ncp->nc_flag & NCF_TS) != 0 || 254 (tsp == NULL && ticksp == NULL), 255 ("No NCF_TS")); 256 257 if (tsp != NULL) 258 *tsp = ((struct namecache_ts *)ncp)->nc_time; 259 if (ticksp != NULL) 260 *ticksp = ((struct namecache_ts *)ncp)->nc_ticks; 261} 262 263static int doingcache = 1; /* 1 => enable the cache */ 264SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, 265 "VFS namecache enabled"); 266 267/* Export size information to userland */ 268SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, 269 sizeof(struct namecache), "sizeof(struct namecache)"); 270 271/* 272 * The new name cache statistics 273 */ 274static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, 275 "Name cache statistics"); 276#define STATNODE(mode, name, var, descr) \ 277 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, descr); 278STATNODE(CTLFLAG_RD, numneg, &numneg, "Number of negative cache entries"); 279STATNODE(CTLFLAG_RD, numcache, &numcache, "Number of cache entries"); 280static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls, 281 "Number of cache lookups"); 282static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits, 283 "Number of '.' hits"); 284static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits, 285 "Number of '..' hits"); 286static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks, 287 "Number of checks in lookup"); 288static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss, 289 "Number of cache misses"); 290static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap, 291 "Number of cache misses we do not want to cache"); 292static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps, 293 "Number of cache hits (positive) we do not want to cache"); 294static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits, 295 "Number of cache hits (positive)"); 296static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps, 297 "Number of cache hits (negative) we do not want to cache"); 298static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits, 299 "Number of cache hits (negative)"); 300static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades, 301 "Number of updates of the cache after lookup (write lock + retry)"); 302 303SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE, 304 &nchstats, sizeof(nchstats), "LU", 305 "VFS cache effectiveness statistics"); 306 307 308 309static void cache_zap(struct namecache *ncp); 310static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf, 311 u_int *buflen); 312static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 313 char *buf, char **retbuf, u_int buflen); 314 315static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); 316 317#ifdef DIAGNOSTIC 318/* 319 * Grab an atomic snapshot of the name cache hash chain lengths 320 */ 321static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, 322 "hash table stats"); 323 324static int 325sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) 326{ 327 struct nchashhead *ncpp; 328 struct namecache *ncp; 329 int i, error, n_nchash, *cntbuf; 330 331retry: 332 n_nchash = nchash + 1; /* nchash is max index, not count */ 333 if (!req->oldptr) 334 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); 335 cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK); 336 CACHE_RLOCK(); 337 if (n_nchash != nchash + 1) { 338 CACHE_RUNLOCK(); 339 free(cntbuf, M_TEMP); 340 goto retry; 341 } 342 /* Scan hash tables counting entries */ 343 for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++) 344 LIST_FOREACH(ncp, ncpp, nc_hash) 345 cntbuf[i]++; 346 CACHE_RUNLOCK(); 347 for (error = 0, i = 0; i < n_nchash; i++) 348 if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0) 349 break; 350 free(cntbuf, M_TEMP); 351 return (error); 352} 353SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD| 354 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", 355 "nchash chain lengths"); 356 357static int 358sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) 359{ 360 int error; 361 struct nchashhead *ncpp; 362 struct namecache *ncp; 363 int n_nchash; 364 int count, maxlength, used, pct; 365 366 if (!req->oldptr) 367 return SYSCTL_OUT(req, 0, 4 * sizeof(int)); 368 369 n_nchash = nchash + 1; /* nchash is max index, not count */ 370 used = 0; 371 maxlength = 0; 372 373 /* Scan hash tables for applicable entries */ 374 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { 375 count = 0; 376 CACHE_RLOCK(); 377 LIST_FOREACH(ncp, ncpp, nc_hash) { 378 count++; 379 } 380 CACHE_RUNLOCK(); 381 if (count) 382 used++; 383 if (maxlength < count) 384 maxlength = count; 385 } 386 n_nchash = nchash + 1; 387 pct = (used * 100) / (n_nchash / 100); 388 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); 389 if (error) 390 return (error); 391 error = SYSCTL_OUT(req, &used, sizeof(used)); 392 if (error) 393 return (error); 394 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); 395 if (error) 396 return (error); 397 error = SYSCTL_OUT(req, &pct, sizeof(pct)); 398 if (error) 399 return (error); 400 return (0); 401} 402SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD| 403 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I", 404 "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)"); 405#endif 406 407/* 408 * cache_zap(): 409 * 410 * Removes a namecache entry from cache, whether it contains an actual 411 * pointer to a vnode or if it is just a negative cache entry. 412 */ 413static void 414cache_zap(ncp) 415 struct namecache *ncp; 416{ 417 struct vnode *vp; 418 419 rw_assert(&cache_lock, RA_WLOCKED); 420 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp); 421 if (ncp->nc_vp != NULL) { 422 SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp, 423 nc_get_name(ncp), ncp->nc_vp); 424 } else { 425 SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp, 426 nc_get_name(ncp)); 427 } 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 caching), 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 471int 472cache_lookup(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 uint32_t hash; 481 int error, ltype, wlocked; 482 483 if (!doingcache) { 484 cnp->cn_flags &= ~MAKEENTRY; 485 return (0); 486 } 487retry: 488 CACHE_RLOCK(); 489 wlocked = 0; 490 numcalls++; 491 error = 0; 492 493retry_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_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp); 501 if (tsp != NULL) 502 timespecclear(tsp); 503 if (ticksp != NULL) 504 *ticksp = ticks; 505 goto success; 506 } 507 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 508 dotdothits++; 509 if (dvp->v_cache_dd == NULL) { 510 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, 511 "..", NULL); 512 goto unlock; 513 } 514 if ((cnp->cn_flags & MAKEENTRY) == 0) { 515 if (!wlocked && !CACHE_UPGRADE_LOCK()) 516 goto wlock; 517 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT) 518 cache_zap(dvp->v_cache_dd); 519 dvp->v_cache_dd = NULL; 520 CACHE_WUNLOCK(); 521 return (0); 522 } 523 ncp = dvp->v_cache_dd; 524 if (ncp->nc_flag & NCF_ISDOTDOT) 525 *vpp = ncp->nc_vp; 526 else 527 *vpp = ncp->nc_dvp; 528 /* Return failure if negative entry was found. */ 529 if (*vpp == NULL) 530 goto negative_success; 531 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..", 532 dvp, cnp->cn_nameptr, *vpp); 533 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..", 534 *vpp); 535 cache_out_ts(ncp, tsp, ticksp); 536 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == 537 NCF_DTS && tsp != NULL) 538 *tsp = ((struct namecache_ts *)ncp)-> 539 nc_dotdottime; 540 goto success; 541 } 542 } 543 544 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT); 545 hash = fnv_32_buf(&dvp, sizeof(dvp), hash); 546 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { 547 numchecks++; 548 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && 549 !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen)) 550 break; 551 } 552 553 /* We failed to find an entry */ 554 if (ncp == NULL) { 555 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, 556 NULL); 557 if ((cnp->cn_flags & MAKEENTRY) == 0) { 558 nummisszap++; 559 } else { 560 nummiss++; 561 } 562 nchstats.ncs_miss++; 563 goto unlock; 564 } 565 566 /* We don't want to have an entry, so dump it */ 567 if ((cnp->cn_flags & MAKEENTRY) == 0) { 568 numposzaps++; 569 nchstats.ncs_badhits++; 570 if (!wlocked && !CACHE_UPGRADE_LOCK()) 571 goto wlock; 572 cache_zap(ncp); 573 CACHE_WUNLOCK(); 574 return (0); 575 } 576 577 /* We found a "positive" match, return the vnode */ 578 if (ncp->nc_vp) { 579 numposhits++; 580 nchstats.ncs_goodhits++; 581 *vpp = ncp->nc_vp; 582 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p", 583 dvp, cnp->cn_nameptr, *vpp, ncp); 584 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp), 585 *vpp); 586 cache_out_ts(ncp, tsp, ticksp); 587 goto success; 588 } 589 590negative_success: 591 /* We found a negative match, and want to create it, so purge */ 592 if (cnp->cn_nameiop == CREATE) { 593 numnegzaps++; 594 nchstats.ncs_badhits++; 595 if (!wlocked && !CACHE_UPGRADE_LOCK()) 596 goto wlock; 597 cache_zap(ncp); 598 CACHE_WUNLOCK(); 599 return (0); 600 } 601 602 if (!wlocked && !CACHE_UPGRADE_LOCK()) 603 goto wlock; 604 numneghits++; 605 /* 606 * We found a "negative" match, so we shift it to the end of 607 * the "negative" cache entries queue to satisfy LRU. Also, 608 * check to see if the entry is a whiteout; indicate this to 609 * the componentname, if so. 610 */ 611 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 612 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 613 nchstats.ncs_neghits++; 614 if (ncp->nc_flag & NCF_WHITE) 615 cnp->cn_flags |= ISWHITEOUT; 616 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, 617 nc_get_name(ncp)); 618 cache_out_ts(ncp, tsp, ticksp); 619 CACHE_WUNLOCK(); 620 return (ENOENT); 621 622wlock: 623 /* 624 * We need to update the cache after our lookup, so upgrade to 625 * a write lock and retry the operation. 626 */ 627 CACHE_RUNLOCK(); 628 CACHE_WLOCK(); 629 numupgrades++; 630 wlocked = 1; 631 goto retry_wlocked; 632 633success: 634 /* 635 * On success we return a locked and ref'd vnode as per the lookup 636 * protocol. 637 */ 638 if (dvp == *vpp) { /* lookup on "." */ 639 VREF(*vpp); 640 if (wlocked) 641 CACHE_WUNLOCK(); 642 else 643 CACHE_RUNLOCK(); 644 /* 645 * When we lookup "." we still can be asked to lock it 646 * differently... 647 */ 648 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 649 if (ltype != VOP_ISLOCKED(*vpp)) { 650 if (ltype == LK_EXCLUSIVE) { 651 vn_lock(*vpp, LK_UPGRADE | LK_RETRY); 652 if ((*vpp)->v_iflag & VI_DOOMED) { 653 /* forced unmount */ 654 vrele(*vpp); 655 *vpp = NULL; 656 return (ENOENT); 657 } 658 } else 659 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); 660 } 661 return (-1); 662 } 663 ltype = 0; /* silence gcc warning */ 664 if (cnp->cn_flags & ISDOTDOT) { 665 ltype = VOP_ISLOCKED(dvp); 666 VOP_UNLOCK(dvp, 0); 667 } 668 VI_LOCK(*vpp); 669 if (wlocked) 670 CACHE_WUNLOCK(); 671 else 672 CACHE_RUNLOCK(); 673 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread); 674 if (cnp->cn_flags & ISDOTDOT) { 675 vn_lock(dvp, ltype | LK_RETRY); 676 if (dvp->v_iflag & VI_DOOMED) { 677 if (error == 0) 678 vput(*vpp); 679 *vpp = NULL; 680 return (ENOENT); 681 } 682 } 683 if (error) { 684 *vpp = NULL; 685 goto retry; 686 } 687 if ((cnp->cn_flags & ISLASTCN) && 688 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) { 689 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup"); 690 } 691 return (-1); 692 693unlock: 694 if (wlocked) 695 CACHE_WUNLOCK(); 696 else 697 CACHE_RUNLOCK(); 698 return (0); 699} 700 701/* 702 * Add an entry to the cache. 703 */ 704void 705cache_enter_time(dvp, vp, cnp, tsp, dtsp) 706 struct vnode *dvp; 707 struct vnode *vp; 708 struct componentname *cnp; 709 struct timespec *tsp; 710 struct timespec *dtsp; 711{ 712 struct namecache *ncp, *n2; 713 struct namecache_ts *n3; 714 struct nchashhead *ncpp; 715 uint32_t hash; 716 int flag; 717 int hold; 718 int zap; 719 int len; 720 721 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr); 722 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp, 723 ("cache_enter: Adding a doomed vnode")); 724 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp, 725 ("cache_enter: Doomed vnode used as src")); 726 727 if (!doingcache) 728 return; 729 730 /* 731 * Avoid blowout in namecache entries. 732 */ 733 if (numcache >= desiredvnodes * ncsizefactor) 734 return; 735 736 flag = 0; 737 if (cnp->cn_nameptr[0] == '.') { 738 if (cnp->cn_namelen == 1) 739 return; 740 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 741 CACHE_WLOCK(); 742 /* 743 * If dotdot entry already exists, just retarget it 744 * to new parent vnode, otherwise continue with new 745 * namecache entry allocation. 746 */ 747 if ((ncp = dvp->v_cache_dd) != NULL && 748 ncp->nc_flag & NCF_ISDOTDOT) { 749 KASSERT(ncp->nc_dvp == dvp, 750 ("wrong isdotdot parent")); 751 if (ncp->nc_vp != NULL) { 752 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, 753 ncp, nc_dst); 754 } else { 755 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 756 numneg--; 757 } 758 if (vp != NULL) { 759 TAILQ_INSERT_HEAD(&vp->v_cache_dst, 760 ncp, nc_dst); 761 } else { 762 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 763 numneg++; 764 } 765 ncp->nc_vp = vp; 766 CACHE_WUNLOCK(); 767 return; 768 } 769 dvp->v_cache_dd = NULL; 770 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp); 771 CACHE_WUNLOCK(); 772 flag = NCF_ISDOTDOT; 773 } 774 } 775 776 hold = 0; 777 zap = 0; 778 779 /* 780 * Calculate the hash key and setup as much of the new 781 * namecache entry as possible before acquiring the lock. 782 */ 783 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); 784 ncp->nc_vp = vp; 785 ncp->nc_dvp = dvp; 786 ncp->nc_flag = flag; 787 if (tsp != NULL) { 788 n3 = (struct namecache_ts *)ncp; 789 n3->nc_time = *tsp; 790 n3->nc_ticks = ticks; 791 n3->nc_flag |= NCF_TS; 792 if (dtsp != NULL) { 793 n3->nc_dotdottime = *dtsp; 794 n3->nc_flag |= NCF_DTS; 795 } 796 } 797 len = ncp->nc_nlen = cnp->cn_namelen; 798 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT); 799 strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1); 800 hash = fnv_32_buf(&dvp, sizeof(dvp), hash); 801 CACHE_WLOCK(); 802 803 /* 804 * See if this vnode or negative entry is already in the cache 805 * with this name. This can happen with concurrent lookups of 806 * the same path name. 807 */ 808 ncpp = NCHHASH(hash); 809 LIST_FOREACH(n2, ncpp, nc_hash) { 810 if (n2->nc_dvp == dvp && 811 n2->nc_nlen == cnp->cn_namelen && 812 !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) { 813 if (tsp != NULL) { 814 KASSERT((n2->nc_flag & NCF_TS) != 0, 815 ("no NCF_TS")); 816 n3 = (struct namecache_ts *)n2; 817 n3->nc_time = 818 ((struct namecache_ts *)ncp)->nc_time; 819 n3->nc_ticks = 820 ((struct namecache_ts *)ncp)->nc_ticks; 821 if (dtsp != NULL) { 822 n3->nc_dotdottime = 823 ((struct namecache_ts *)ncp)-> 824 nc_dotdottime; 825 n3->nc_flag |= NCF_DTS; 826 } 827 } 828 CACHE_WUNLOCK(); 829 cache_free(ncp); 830 return; 831 } 832 } 833 834 if (flag == NCF_ISDOTDOT) { 835 /* 836 * See if we are trying to add .. entry, but some other lookup 837 * has populated v_cache_dd pointer already. 838 */ 839 if (dvp->v_cache_dd != NULL) { 840 CACHE_WUNLOCK(); 841 cache_free(ncp); 842 return; 843 } 844 KASSERT(vp == NULL || vp->v_type == VDIR, 845 ("wrong vnode type %p", vp)); 846 dvp->v_cache_dd = ncp; 847 } 848 849 numcache++; 850 if (!vp) { 851 numneg++; 852 if (cnp->cn_flags & ISWHITEOUT) 853 ncp->nc_flag |= NCF_WHITE; 854 } else if (vp->v_type == VDIR) { 855 if (flag != NCF_ISDOTDOT) { 856 /* 857 * For this case, the cache entry maps both the 858 * directory name in it and the name ".." for the 859 * directory's parent. 860 */ 861 if ((n2 = vp->v_cache_dd) != NULL && 862 (n2->nc_flag & NCF_ISDOTDOT) != 0) 863 cache_zap(n2); 864 vp->v_cache_dd = ncp; 865 } 866 } else { 867 vp->v_cache_dd = NULL; 868 } 869 870 /* 871 * Insert the new namecache entry into the appropriate chain 872 * within the cache entries table. 873 */ 874 LIST_INSERT_HEAD(ncpp, ncp, nc_hash); 875 if (flag != NCF_ISDOTDOT) { 876 if (LIST_EMPTY(&dvp->v_cache_src)) { 877 hold = 1; 878 numcachehv++; 879 } 880 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); 881 } 882 883 /* 884 * If the entry is "negative", we place it into the 885 * "negative" cache queue, otherwise, we place it into the 886 * destination vnode's cache entries queue. 887 */ 888 if (vp) { 889 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); 890 SDT_PROBE3(vfs, namecache, enter, done, dvp, nc_get_name(ncp), 891 vp); 892 } else { 893 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 894 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp, 895 nc_get_name(ncp)); 896 } 897 if (numneg * ncnegfactor > numcache) { 898 ncp = TAILQ_FIRST(&ncneg); 899 KASSERT(ncp->nc_vp == NULL, ("ncp %p vp %p on ncneg", 900 ncp, ncp->nc_vp)); 901 zap = 1; 902 } 903 if (hold) 904 vhold(dvp); 905 if (zap) 906 cache_zap(ncp); 907 CACHE_WUNLOCK(); 908} 909 910/* 911 * Name cache initialization, from vfs_init() when we are booting 912 */ 913static void 914nchinit(void *dummy __unused) 915{ 916 917 TAILQ_INIT(&ncneg); 918 919 cache_zone_small = uma_zcreate("S VFS Cache", 920 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1, 921 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 922 cache_zone_small_ts = uma_zcreate("STS VFS Cache", 923 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1, 924 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 925 cache_zone_large = uma_zcreate("L VFS Cache", 926 sizeof(struct namecache) + NAME_MAX + 1, 927 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 928 cache_zone_large_ts = uma_zcreate("LTS VFS Cache", 929 sizeof(struct namecache_ts) + NAME_MAX + 1, 930 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 931 932 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash); 933} 934SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); 935 936void 937cache_changesize(int newmaxvnodes) 938{ 939 struct nchashhead *new_nchashtbl, *old_nchashtbl; 940 u_long new_nchash, old_nchash; 941 struct namecache *ncp; 942 uint32_t hash; 943 int i; 944 945 new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash); 946 /* If same hash table size, nothing to do */ 947 if (nchash == new_nchash) { 948 free(new_nchashtbl, M_VFSCACHE); 949 return; 950 } 951 /* 952 * Move everything from the old hash table to the new table. 953 * None of the namecache entries in the table can be removed 954 * because to do so, they have to be removed from the hash table. 955 */ 956 CACHE_WLOCK(); 957 old_nchashtbl = nchashtbl; 958 old_nchash = nchash; 959 nchashtbl = new_nchashtbl; 960 nchash = new_nchash; 961 for (i = 0; i <= old_nchash; i++) { 962 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) { 963 hash = fnv_32_buf(nc_get_name(ncp), ncp->nc_nlen, 964 FNV1_32_INIT); 965 hash = fnv_32_buf(&ncp->nc_dvp, sizeof(ncp->nc_dvp), 966 hash); 967 LIST_REMOVE(ncp, nc_hash); 968 LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash); 969 } 970 } 971 CACHE_WUNLOCK(); 972 free(old_nchashtbl, M_VFSCACHE); 973} 974 975/* 976 * Invalidate all entries to a particular vnode. 977 */ 978void 979cache_purge(vp) 980 struct vnode *vp; 981{ 982 983 CTR1(KTR_VFS, "cache_purge(%p)", vp); 984 SDT_PROBE1(vfs, namecache, purge, done, vp); 985 CACHE_WLOCK(); 986 while (!LIST_EMPTY(&vp->v_cache_src)) 987 cache_zap(LIST_FIRST(&vp->v_cache_src)); 988 while (!TAILQ_EMPTY(&vp->v_cache_dst)) 989 cache_zap(TAILQ_FIRST(&vp->v_cache_dst)); 990 if (vp->v_cache_dd != NULL) { 991 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT, 992 ("lost dotdot link")); 993 cache_zap(vp->v_cache_dd); 994 } 995 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); 996 CACHE_WUNLOCK(); 997} 998 999/* 1000 * Invalidate all negative entries for a particular directory vnode. 1001 */ 1002void 1003cache_purge_negative(vp) 1004 struct vnode *vp; 1005{ 1006 struct namecache *cp, *ncp; 1007 1008 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp); 1009 SDT_PROBE1(vfs, namecache, purge_negative, done, vp); 1010 CACHE_WLOCK(); 1011 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) { 1012 if (cp->nc_vp == NULL) 1013 cache_zap(cp); 1014 } 1015 CACHE_WUNLOCK(); 1016} 1017 1018/* 1019 * Flush all entries referencing a particular filesystem. 1020 */ 1021void 1022cache_purgevfs(mp) 1023 struct mount *mp; 1024{ 1025 struct nchashhead *ncpp; 1026 struct namecache *ncp, *nnp; 1027 1028 /* Scan hash tables for applicable entries */ 1029 SDT_PROBE1(vfs, namecache, purgevfs, done, mp); 1030 CACHE_WLOCK(); 1031 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) { 1032 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) { 1033 if (ncp->nc_dvp->v_mount == mp) 1034 cache_zap(ncp); 1035 } 1036 } 1037 CACHE_WUNLOCK(); 1038} 1039 1040/* 1041 * Perform canonical checks and cache lookup and pass on to filesystem 1042 * through the vop_cachedlookup only if needed. 1043 */ 1044 1045int 1046vfs_cache_lookup(ap) 1047 struct vop_lookup_args /* { 1048 struct vnode *a_dvp; 1049 struct vnode **a_vpp; 1050 struct componentname *a_cnp; 1051 } */ *ap; 1052{ 1053 struct vnode *dvp; 1054 int error; 1055 struct vnode **vpp = ap->a_vpp; 1056 struct componentname *cnp = ap->a_cnp; 1057 struct ucred *cred = cnp->cn_cred; 1058 int flags = cnp->cn_flags; 1059 struct thread *td = cnp->cn_thread; 1060 1061 *vpp = NULL; 1062 dvp = ap->a_dvp; 1063 1064 if (dvp->v_type != VDIR) 1065 return (ENOTDIR); 1066 1067 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 1068 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 1069 return (EROFS); 1070 1071 error = VOP_ACCESS(dvp, VEXEC, cred, td); 1072 if (error) 1073 return (error); 1074 1075 error = cache_lookup(dvp, vpp, cnp, NULL, NULL); 1076 if (error == 0) 1077 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); 1078 if (error == -1) 1079 return (0); 1080 return (error); 1081} 1082 1083/* 1084 * XXX All of these sysctls would probably be more productive dead. 1085 */ 1086static int disablecwd; 1087SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, 1088 "Disable the getcwd syscall"); 1089 1090/* Implementation of the getcwd syscall. */ 1091int 1092sys___getcwd(td, uap) 1093 struct thread *td; 1094 struct __getcwd_args *uap; 1095{ 1096 1097 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen)); 1098} 1099 1100int 1101kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen) 1102{ 1103 char *bp, *tmpbuf; 1104 struct filedesc *fdp; 1105 struct vnode *cdir, *rdir; 1106 int error; 1107 1108 if (disablecwd) 1109 return (ENODEV); 1110 if (buflen < 2) 1111 return (EINVAL); 1112 if (buflen > MAXPATHLEN) 1113 buflen = MAXPATHLEN; 1114 1115 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK); 1116 fdp = td->td_proc->p_fd; 1117 FILEDESC_SLOCK(fdp); 1118 cdir = fdp->fd_cdir; 1119 VREF(cdir); 1120 rdir = fdp->fd_rdir; 1121 VREF(rdir); 1122 FILEDESC_SUNLOCK(fdp); 1123 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen); 1124 vrele(rdir); 1125 vrele(cdir); 1126 1127 if (!error) { 1128 if (bufseg == UIO_SYSSPACE) 1129 bcopy(bp, buf, strlen(bp) + 1); 1130 else 1131 error = copyout(bp, buf, strlen(bp) + 1); 1132#ifdef KTRACE 1133 if (KTRPOINT(curthread, KTR_NAMEI)) 1134 ktrnamei(bp); 1135#endif 1136 } 1137 free(tmpbuf, M_TEMP); 1138 return (error); 1139} 1140 1141/* 1142 * Thus begins the fullpath magic. 1143 */ 1144 1145#undef STATNODE 1146#define STATNODE(name, descr) \ 1147 static u_int name; \ 1148 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr) 1149 1150static int disablefullpath; 1151SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0, 1152 "Disable the vn_fullpath function"); 1153 1154/* These count for kern___getcwd(), too. */ 1155STATNODE(numfullpathcalls, "Number of fullpath search calls"); 1156STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)"); 1157STATNODE(numfullpathfail2, 1158 "Number of fullpath search errors (VOP_VPTOCNP failures)"); 1159STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)"); 1160STATNODE(numfullpathfound, "Number of successful fullpath calls"); 1161 1162/* 1163 * Retrieve the full filesystem path that correspond to a vnode from the name 1164 * cache (if available) 1165 */ 1166int 1167vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) 1168{ 1169 char *buf; 1170 struct filedesc *fdp; 1171 struct vnode *rdir; 1172 int error; 1173 1174 if (disablefullpath) 1175 return (ENODEV); 1176 if (vn == NULL) 1177 return (EINVAL); 1178 1179 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1180 fdp = td->td_proc->p_fd; 1181 FILEDESC_SLOCK(fdp); 1182 rdir = fdp->fd_rdir; 1183 VREF(rdir); 1184 FILEDESC_SUNLOCK(fdp); 1185 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN); 1186 vrele(rdir); 1187 1188 if (!error) 1189 *freebuf = buf; 1190 else 1191 free(buf, M_TEMP); 1192 return (error); 1193} 1194 1195/* 1196 * This function is similar to vn_fullpath, but it attempts to lookup the 1197 * pathname relative to the global root mount point. This is required for the 1198 * auditing sub-system, as audited pathnames must be absolute, relative to the 1199 * global root mount point. 1200 */ 1201int 1202vn_fullpath_global(struct thread *td, struct vnode *vn, 1203 char **retbuf, char **freebuf) 1204{ 1205 char *buf; 1206 int error; 1207 1208 if (disablefullpath) 1209 return (ENODEV); 1210 if (vn == NULL) 1211 return (EINVAL); 1212 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1213 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN); 1214 if (!error) 1215 *freebuf = buf; 1216 else 1217 free(buf, M_TEMP); 1218 return (error); 1219} 1220 1221int 1222vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen) 1223{ 1224 int error; 1225 1226 CACHE_RLOCK(); 1227 error = vn_vptocnp_locked(vp, cred, buf, buflen); 1228 if (error == 0) 1229 CACHE_RUNLOCK(); 1230 return (error); 1231} 1232 1233static int 1234vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf, 1235 u_int *buflen) 1236{ 1237 struct vnode *dvp; 1238 struct namecache *ncp; 1239 int error; 1240 1241 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) { 1242 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 1243 break; 1244 } 1245 if (ncp != NULL) { 1246 if (*buflen < ncp->nc_nlen) { 1247 CACHE_RUNLOCK(); 1248 vrele(*vp); 1249 numfullpathfail4++; 1250 error = ENOMEM; 1251 SDT_PROBE3(vfs, namecache, fullpath, return, error, 1252 vp, NULL); 1253 return (error); 1254 } 1255 *buflen -= ncp->nc_nlen; 1256 memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen); 1257 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp, 1258 nc_get_name(ncp), vp); 1259 dvp = *vp; 1260 *vp = ncp->nc_dvp; 1261 vref(*vp); 1262 CACHE_RUNLOCK(); 1263 vrele(dvp); 1264 CACHE_RLOCK(); 1265 return (0); 1266 } 1267 SDT_PROBE1(vfs, namecache, fullpath, miss, vp); 1268 1269 CACHE_RUNLOCK(); 1270 vn_lock(*vp, LK_SHARED | LK_RETRY); 1271 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen); 1272 vput(*vp); 1273 if (error) { 1274 numfullpathfail2++; 1275 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); 1276 return (error); 1277 } 1278 1279 *vp = dvp; 1280 CACHE_RLOCK(); 1281 if (dvp->v_iflag & VI_DOOMED) { 1282 /* forced unmount */ 1283 CACHE_RUNLOCK(); 1284 vrele(dvp); 1285 error = ENOENT; 1286 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); 1287 return (error); 1288 } 1289 /* 1290 * *vp has its use count incremented still. 1291 */ 1292 1293 return (0); 1294} 1295 1296/* 1297 * The magic behind kern___getcwd() and vn_fullpath(). 1298 */ 1299static int 1300vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 1301 char *buf, char **retbuf, u_int buflen) 1302{ 1303 int error, slash_prefixed; 1304#ifdef KDTRACE_HOOKS 1305 struct vnode *startvp = vp; 1306#endif 1307 struct vnode *vp1; 1308 1309 buflen--; 1310 buf[buflen] = '\0'; 1311 error = 0; 1312 slash_prefixed = 0; 1313 1314 SDT_PROBE1(vfs, namecache, fullpath, entry, vp); 1315 numfullpathcalls++; 1316 vref(vp); 1317 CACHE_RLOCK(); 1318 if (vp->v_type != VDIR) { 1319 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen); 1320 if (error) 1321 return (error); 1322 if (buflen == 0) { 1323 CACHE_RUNLOCK(); 1324 vrele(vp); 1325 return (ENOMEM); 1326 } 1327 buf[--buflen] = '/'; 1328 slash_prefixed = 1; 1329 } 1330 while (vp != rdir && vp != rootvnode) { 1331 if (vp->v_vflag & VV_ROOT) { 1332 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */ 1333 CACHE_RUNLOCK(); 1334 vrele(vp); 1335 error = ENOENT; 1336 SDT_PROBE3(vfs, namecache, fullpath, return, 1337 error, vp, NULL); 1338 break; 1339 } 1340 vp1 = vp->v_mount->mnt_vnodecovered; 1341 vref(vp1); 1342 CACHE_RUNLOCK(); 1343 vrele(vp); 1344 vp = vp1; 1345 CACHE_RLOCK(); 1346 continue; 1347 } 1348 if (vp->v_type != VDIR) { 1349 CACHE_RUNLOCK(); 1350 vrele(vp); 1351 numfullpathfail1++; 1352 error = ENOTDIR; 1353 SDT_PROBE3(vfs, namecache, fullpath, return, 1354 error, vp, NULL); 1355 break; 1356 } 1357 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen); 1358 if (error) 1359 break; 1360 if (buflen == 0) { 1361 CACHE_RUNLOCK(); 1362 vrele(vp); 1363 error = ENOMEM; 1364 SDT_PROBE3(vfs, namecache, fullpath, return, error, 1365 startvp, NULL); 1366 break; 1367 } 1368 buf[--buflen] = '/'; 1369 slash_prefixed = 1; 1370 } 1371 if (error) 1372 return (error); 1373 if (!slash_prefixed) { 1374 if (buflen == 0) { 1375 CACHE_RUNLOCK(); 1376 vrele(vp); 1377 numfullpathfail4++; 1378 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM, 1379 startvp, NULL); 1380 return (ENOMEM); 1381 } 1382 buf[--buflen] = '/'; 1383 } 1384 numfullpathfound++; 1385 CACHE_RUNLOCK(); 1386 vrele(vp); 1387 1388 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen); 1389 *retbuf = buf + buflen; 1390 return (0); 1391} 1392 1393struct vnode * 1394vn_dir_dd_ino(struct vnode *vp) 1395{ 1396 struct namecache *ncp; 1397 struct vnode *ddvp; 1398 1399 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino"); 1400 CACHE_RLOCK(); 1401 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) { 1402 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) 1403 continue; 1404 ddvp = ncp->nc_dvp; 1405 VI_LOCK(ddvp); 1406 CACHE_RUNLOCK(); 1407 if (vget(ddvp, LK_INTERLOCK | LK_SHARED | LK_NOWAIT, curthread)) 1408 return (NULL); 1409 return (ddvp); 1410 } 1411 CACHE_RUNLOCK(); 1412 return (NULL); 1413} 1414 1415int 1416vn_commname(struct vnode *vp, char *buf, u_int buflen) 1417{ 1418 struct namecache *ncp; 1419 int l; 1420 1421 CACHE_RLOCK(); 1422 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) 1423 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) 1424 break; 1425 if (ncp == NULL) { 1426 CACHE_RUNLOCK(); 1427 return (ENOENT); 1428 } 1429 l = min(ncp->nc_nlen, buflen - 1); 1430 memcpy(buf, nc_get_name(ncp), l); 1431 CACHE_RUNLOCK(); 1432 buf[l] = '\0'; 1433 return (0); 1434} 1435 1436/* ABI compat shims for old kernel modules. */ 1437#undef cache_enter 1438 1439void cache_enter(struct vnode *dvp, struct vnode *vp, 1440 struct componentname *cnp); 1441 1442void 1443cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) 1444{ 1445 1446 cache_enter_time(dvp, vp, cnp, NULL, NULL); 1447} 1448 1449/* 1450 * This function updates path string to vnode's full global path 1451 * and checks the size of the new path string against the pathlen argument. 1452 * 1453 * Requires a locked, referenced vnode and GIANT lock held. 1454 * Vnode is re-locked on success or ENODEV, otherwise unlocked. 1455 * 1456 * If sysctl debug.disablefullpath is set, ENODEV is returned, 1457 * vnode is left locked and path remain untouched. 1458 * 1459 * If vp is a directory, the call to vn_fullpath_global() always succeeds 1460 * because it falls back to the ".." lookup if the namecache lookup fails. 1461 */ 1462int 1463vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, 1464 u_int pathlen) 1465{ 1466 struct nameidata nd; 1467 struct vnode *vp1; 1468 char *rpath, *fbuf; 1469 int error; 1470 1471 ASSERT_VOP_ELOCKED(vp, __func__); 1472 1473 /* Return ENODEV if sysctl debug.disablefullpath==1 */ 1474 if (disablefullpath) 1475 return (ENODEV); 1476 1477 /* Construct global filesystem path from vp. */ 1478 VOP_UNLOCK(vp, 0); 1479 error = vn_fullpath_global(td, vp, &rpath, &fbuf); 1480 1481 if (error != 0) { 1482 vrele(vp); 1483 return (error); 1484 } 1485 1486 if (strlen(rpath) >= pathlen) { 1487 vrele(vp); 1488 error = ENAMETOOLONG; 1489 goto out; 1490 } 1491 1492 /* 1493 * Re-lookup the vnode by path to detect a possible rename. 1494 * As a side effect, the vnode is relocked. 1495 * If vnode was renamed, return ENOENT. 1496 */ 1497 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1498 UIO_SYSSPACE, path, td); 1499 error = namei(&nd); 1500 if (error != 0) { 1501 vrele(vp); 1502 goto out; 1503 } 1504 NDFREE(&nd, NDF_ONLY_PNBUF); 1505 vp1 = nd.ni_vp; 1506 vrele(vp); 1507 if (vp1 == vp) 1508 strcpy(path, rpath); 1509 else { 1510 vput(vp1); 1511 error = ENOENT; 1512 } 1513 1514out: 1515 free(fbuf, M_TEMP); 1516 return (error); 1517} 1518