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 <sys/param.h> 39 #include <sys/filedesc.h> 40 #include <sys/fnv_hash.h> 41 #include <sys/kernel.h> 42 #include <sys/lock.h> 43 #include <sys/malloc.h> 44 #include <sys/mount.h> 45 #include <sys/mutex.h> 46 #include <sys/namei.h> 47 #include <sys/proc.h> 48 #include <sys/syscallsubr.h> 49 #include <sys/sysctl.h> 50 #include <sys/sysproto.h> 51 #include <sys/systm.h> 52 #include <sys/vnode.h> 53 54 #include <vm/uma.h> 55 56 /* 57 * This structure describes the elements in the cache of recent 58 * names looked up by namei. 59 */ 60 61 struct namecache { 62 LIST_ENTRY(namecache) nc_hash; /* hash chain */ 63 LIST_ENTRY(namecache) nc_src; /* source vnode list */ 64 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ 65 struct vnode *nc_dvp; /* vnode of parent of name */ 66 struct vnode *nc_vp; /* vnode the name refers to */ 67 u_char nc_flag; /* flag bits */ 68 u_char nc_nlen; /* length of name */ 69 char nc_name[0]; /* segment name */ 70 }; 71 72 /* 73 * Name caching works as follows: 74 * 75 * Names found by directory scans are retained in a cache 76 * for future reference. It is managed LRU, so frequently 77 * used names will hang around. Cache is indexed by hash value 78 * obtained from (vp, name) where vp refers to the directory 79 * containing name. 80 * 81 * If it is a "negative" entry, (i.e. for a name that is known NOT to 82 * exist) the vnode pointer will be NULL. 83 * 84 * Upon reaching the last segment of a path, if the reference 85 * is for DELETE, or NOCACHE is set (rewrite), and the 86 * name is located in the cache, it will be dropped. 87 */ 88 89 /* 90 * Structures associated with name cacheing. 91 */ 92 #define NCHHASH(hash) \ 93 (&nchashtbl[(hash) & nchash]) 94 static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */ 95 static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */ 96 static u_long nchash; /* size of hash table */ 97 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, ""); 98 static u_long ncnegfactor = 16; /* ratio of negative entries */ 99 SYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, ""); 100 static u_long numneg; /* number of cache entries allocated */ 101 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, ""); 102 static u_long numcache; /* number of cache entries allocated */ 103 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, ""); 104 static u_long numcachehv; /* number of cache entries with vnodes held */ 105 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, ""); 106 #if 0 107 static u_long numcachepl; /* number of cache purge for leaf entries */ 108 SYSCTL_ULONG(_debug, OID_AUTO, numcachepl, CTLFLAG_RD, &numcachepl, 0, ""); 109 #endif 110 struct nchstats nchstats; /* cache effectiveness statistics */ 111 112 static struct mtx cache_lock; 113 MTX_SYSINIT(vfscache, &cache_lock, "Name Cache", MTX_DEF); 114 115 #define CACHE_LOCK() mtx_lock(&cache_lock) 116 #define CACHE_UNLOCK() mtx_unlock(&cache_lock) 117 118 /* 119 * UMA zones for the VFS cache. 120 * 121 * The small cache is used for entries with short names, which are the 122 * most common. The large cache is used for entries which are too big to 123 * fit in the small cache. 124 */ 125 static uma_zone_t cache_zone_small; 126 static uma_zone_t cache_zone_large; 127 128 #define CACHE_PATH_CUTOFF 32 129 #define CACHE_ZONE_SMALL (sizeof(struct namecache) + CACHE_PATH_CUTOFF) 130 #define CACHE_ZONE_LARGE (sizeof(struct namecache) + NAME_MAX) 131 132 #define cache_alloc(len) uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \ 133 cache_zone_small : cache_zone_large, M_WAITOK) 134 #define cache_free(ncp) do { \ 135 if (ncp != NULL) \ 136 uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \ 137 cache_zone_small : cache_zone_large, (ncp)); \ 138 } while (0) 139 140 static int doingcache = 1; /* 1 => enable the cache */ 141 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, ""); 142 143 /* Export size information to userland */ 144 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0, 145 sizeof(struct namecache), ""); 146 147 /* 148 * The new name cache statistics 149 */ 150 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics"); 151 #define STATNODE(mode, name, var) \ 152 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, ""); 153 STATNODE(CTLFLAG_RD, numneg, &numneg); 154 STATNODE(CTLFLAG_RD, numcache, &numcache); 155 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls); 156 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits); 157 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits); 158 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks); 159 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss); 160 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap); 161 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps); 162 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits); 163 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps); 164 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits); 165 166 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD, &nchstats, 167 sizeof(nchstats), "LU", "VFS cache effectiveness statistics"); 168 169 170 171 static void cache_zap(struct namecache *ncp); 172 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 173 char *buf, char **retbuf, u_int buflen); 174 175 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); 176 177 /* 178 * Flags in namecache.nc_flag 179 */ 180 #define NCF_WHITE 1 181 182 /* 183 * Grab an atomic snapshot of the name cache hash chain lengths 184 */ 185 SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats"); 186 187 static int 188 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) 189 { 190 int error; 191 struct nchashhead *ncpp; 192 struct namecache *ncp; 193 int n_nchash; 194 int count; 195 196 n_nchash = nchash + 1; /* nchash is max index, not count */ 197 if (!req->oldptr) 198 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); 199 200 /* Scan hash tables for applicable entries */ 201 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { 202 CACHE_LOCK(); 203 count = 0; 204 LIST_FOREACH(ncp, ncpp, nc_hash) { 205 count++; 206 } 207 CACHE_UNLOCK(); 208 error = SYSCTL_OUT(req, &count, sizeof(count)); 209 if (error) 210 return (error); 211 } 212 return (0); 213 } 214 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD, 215 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths"); 216 217 static int 218 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) 219 { 220 int error; 221 struct nchashhead *ncpp; 222 struct namecache *ncp; 223 int n_nchash; 224 int count, maxlength, used, pct; 225 226 if (!req->oldptr) 227 return SYSCTL_OUT(req, 0, 4 * sizeof(int)); 228 229 n_nchash = nchash + 1; /* nchash is max index, not count */ 230 used = 0; 231 maxlength = 0; 232 233 /* Scan hash tables for applicable entries */ 234 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { 235 count = 0; 236 CACHE_LOCK(); 237 LIST_FOREACH(ncp, ncpp, nc_hash) { 238 count++; 239 } 240 CACHE_UNLOCK(); 241 if (count) 242 used++; 243 if (maxlength < count) 244 maxlength = count; 245 } 246 n_nchash = nchash + 1; 247 pct = (used * 100 * 100) / n_nchash; 248 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); 249 if (error) 250 return (error); 251 error = SYSCTL_OUT(req, &used, sizeof(used)); 252 if (error) 253 return (error); 254 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); 255 if (error) 256 return (error); 257 error = SYSCTL_OUT(req, &pct, sizeof(pct)); 258 if (error) 259 return (error); 260 return (0); 261 } 262 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD, 263 0, 0, sysctl_debug_hashstat_nchash, "I", "nchash chain lengths"); 264 265 /* 266 * cache_zap(): 267 * 268 * Removes a namecache entry from cache, whether it contains an actual 269 * pointer to a vnode or if it is just a negative cache entry. 270 */ 271 static void 272 cache_zap(ncp) 273 struct namecache *ncp; 274 { 275 struct vnode *vp; 276 277 mtx_assert(&cache_lock, MA_OWNED); 278 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp); 279 vp = NULL; 280 LIST_REMOVE(ncp, nc_hash); 281 LIST_REMOVE(ncp, nc_src); 282 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) { 283 vp = ncp->nc_dvp; 284 numcachehv--; 285 } 286 if (ncp->nc_vp) { 287 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); 288 ncp->nc_vp->v_dd = NULL; 289 } else { 290 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 291 numneg--; 292 } 293 numcache--; 294 cache_free(ncp); 295 if (vp) 296 vdrop(vp); 297 } 298 299 /* 300 * Lookup an entry in the cache 301 * 302 * Lookup is called with dvp pointing to the directory to search, 303 * cnp pointing to the name of the entry being sought. If the lookup 304 * succeeds, the vnode is returned in *vpp, and a status of -1 is 305 * returned. If the lookup determines that the name does not exist 306 * (negative cacheing), a status of ENOENT is returned. If the lookup 307 * fails, a status of zero is returned. If the directory vnode is 308 * recycled out from under us due to a forced unmount, a status of 309 * EBADF is returned. 310 * 311 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is 312 * unlocked. If we're looking up . an extra ref is taken, but the lock is 313 * not recursively acquired. 314 */ 315 316 int 317 cache_lookup(dvp, vpp, cnp) 318 struct vnode *dvp; 319 struct vnode **vpp; 320 struct componentname *cnp; 321 { 322 struct namecache *ncp; 323 u_int32_t hash; 324 int error, ltype; 325 326 if (!doingcache) { 327 cnp->cn_flags &= ~MAKEENTRY; 328 return (0); 329 } 330 retry: 331 CACHE_LOCK(); 332 numcalls++; 333 334 if (cnp->cn_nameptr[0] == '.') { 335 if (cnp->cn_namelen == 1) { 336 *vpp = dvp; 337 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .", 338 dvp, cnp->cn_nameptr); 339 dothits++; 340 goto success; 341 } 342 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 343 dotdothits++; 344 if (dvp->v_dd == NULL || 345 (cnp->cn_flags & MAKEENTRY) == 0) { 346 CACHE_UNLOCK(); 347 return (0); 348 } 349 *vpp = dvp->v_dd; 350 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..", 351 dvp, cnp->cn_nameptr, *vpp); 352 goto success; 353 } 354 } 355 356 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT); 357 hash = fnv_32_buf(&dvp, sizeof(dvp), hash); 358 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { 359 numchecks++; 360 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && 361 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) 362 break; 363 } 364 365 /* We failed to find an entry */ 366 if (ncp == 0) { 367 if ((cnp->cn_flags & MAKEENTRY) == 0) { 368 nummisszap++; 369 } else { 370 nummiss++; 371 } 372 nchstats.ncs_miss++; 373 CACHE_UNLOCK(); 374 return (0); 375 } 376 377 /* We don't want to have an entry, so dump it */ 378 if ((cnp->cn_flags & MAKEENTRY) == 0) { 379 numposzaps++; 380 nchstats.ncs_badhits++; 381 cache_zap(ncp); 382 CACHE_UNLOCK(); 383 return (0); 384 } 385 386 /* We found a "positive" match, return the vnode */ 387 if (ncp->nc_vp) { 388 numposhits++; 389 nchstats.ncs_goodhits++; 390 *vpp = ncp->nc_vp; 391 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p", 392 dvp, cnp->cn_nameptr, *vpp, ncp); 393 goto success; 394 } 395 396 /* We found a negative match, and want to create it, so purge */ 397 if (cnp->cn_nameiop == CREATE) { 398 numnegzaps++; 399 nchstats.ncs_badhits++; 400 cache_zap(ncp); 401 CACHE_UNLOCK(); 402 return (0); 403 } 404 405 numneghits++; 406 /* 407 * We found a "negative" match, so we shift it to the end of 408 * the "negative" cache entries queue to satisfy LRU. Also, 409 * check to see if the entry is a whiteout; indicate this to 410 * the componentname, if so. 411 */ 412 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 413 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 414 nchstats.ncs_neghits++; 415 if (ncp->nc_flag & NCF_WHITE) 416 cnp->cn_flags |= ISWHITEOUT; 417 CACHE_UNLOCK(); 418 return (ENOENT); 419 420 success: 421 /* 422 * On success we return a locked and ref'd vnode as per the lookup 423 * protocol. 424 */ 425 if (dvp == *vpp) { /* lookup on "." */ 426 VREF(*vpp); 427 CACHE_UNLOCK(); 428 /* 429 * When we lookup "." we still can be asked to lock it 430 * differently... 431 */ 432 ltype = cnp->cn_lkflags & LK_TYPE_MASK; 433 if (ltype != VOP_ISLOCKED(*vpp)) { 434 if (ltype == LK_EXCLUSIVE) { 435 vn_lock(*vpp, LK_UPGRADE | LK_RETRY); 436 if ((*vpp)->v_iflag & VI_DOOMED) { 437 /* forced unmount */ 438 vrele(*vpp); 439 *vpp = NULL; 440 return (EBADF); 441 } 442 } else 443 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); 444 } 445 return (-1); 446 } 447 ltype = 0; /* silence gcc warning */ 448 if (cnp->cn_flags & ISDOTDOT) { 449 ltype = VOP_ISLOCKED(dvp); 450 VOP_UNLOCK(dvp, 0); 451 } 452 VI_LOCK(*vpp); 453 CACHE_UNLOCK(); 454 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread); 455 if (cnp->cn_flags & ISDOTDOT) 456 vn_lock(dvp, ltype | LK_RETRY); 457 if (error) { 458 *vpp = NULL; 459 goto retry; 460 } 461 if ((cnp->cn_flags & ISLASTCN) && 462 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) { 463 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup"); 464 } 465 return (-1); 466 } 467 468 /* 469 * Add an entry to the cache. 470 */ 471 void 472 cache_enter(dvp, vp, cnp) 473 struct vnode *dvp; 474 struct vnode *vp; 475 struct componentname *cnp; 476 { 477 struct namecache *ncp; 478 struct nchashhead *ncpp; 479 u_int32_t hash; 480 int hold; 481 int zap; 482 int len; 483 484 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr); 485 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp, 486 ("cahe_enter: Adding a doomed vnode")); 487 488 if (!doingcache) 489 return; 490 491 if (cnp->cn_nameptr[0] == '.') { 492 if (cnp->cn_namelen == 1) { 493 return; 494 } 495 /* 496 * For dotdot lookups only cache the v_dd pointer if the 497 * directory has a link back to its parent via v_cache_dst. 498 * Without this an unlinked directory would keep a soft 499 * reference to its parent which could not be NULLd at 500 * cache_purge() time. 501 */ 502 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 503 CACHE_LOCK(); 504 if (!TAILQ_EMPTY(&dvp->v_cache_dst)) 505 dvp->v_dd = vp; 506 CACHE_UNLOCK(); 507 return; 508 } 509 } 510 511 hold = 0; 512 zap = 0; 513 514 /* 515 * Calculate the hash key and setup as much of the new 516 * namecache entry as possible before acquiring the lock. 517 */ 518 ncp = cache_alloc(cnp->cn_namelen); 519 ncp->nc_vp = vp; 520 ncp->nc_dvp = dvp; 521 len = ncp->nc_nlen = cnp->cn_namelen; 522 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT); 523 bcopy(cnp->cn_nameptr, ncp->nc_name, len); 524 hash = fnv_32_buf(&dvp, sizeof(dvp), hash); 525 CACHE_LOCK(); 526 527 /* 528 * See if this vnode is already in the cache with this name. 529 * This can happen with concurrent lookups of the same path 530 * name. 531 */ 532 if (vp) { 533 struct namecache *n2; 534 535 TAILQ_FOREACH(n2, &vp->v_cache_dst, nc_dst) { 536 if (n2->nc_dvp == dvp && 537 n2->nc_nlen == cnp->cn_namelen && 538 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) { 539 CACHE_UNLOCK(); 540 cache_free(ncp); 541 return; 542 } 543 } 544 } 545 546 numcache++; 547 if (!vp) { 548 numneg++; 549 ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0; 550 } else if (vp->v_type == VDIR) { 551 vp->v_dd = dvp; 552 } else { 553 vp->v_dd = NULL; 554 } 555 556 /* 557 * Insert the new namecache entry into the appropriate chain 558 * within the cache entries table. 559 */ 560 ncpp = NCHHASH(hash); 561 LIST_INSERT_HEAD(ncpp, ncp, nc_hash); 562 if (LIST_EMPTY(&dvp->v_cache_src)) { 563 hold = 1; 564 numcachehv++; 565 } 566 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); 567 /* 568 * If the entry is "negative", we place it into the 569 * "negative" cache queue, otherwise, we place it into the 570 * destination vnode's cache entries queue. 571 */ 572 if (vp) { 573 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); 574 } else { 575 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 576 } 577 if (numneg * ncnegfactor > numcache) { 578 ncp = TAILQ_FIRST(&ncneg); 579 zap = 1; 580 } 581 if (hold) 582 vhold(dvp); 583 if (zap) 584 cache_zap(ncp); 585 CACHE_UNLOCK(); 586 } 587 588 /* 589 * Name cache initialization, from vfs_init() when we are booting 590 */ 591 static void 592 nchinit(void *dummy __unused) 593 { 594 595 TAILQ_INIT(&ncneg); 596 597 cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL, 598 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 599 cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL, 600 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT); 601 602 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash); 603 } 604 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); 605 606 607 /* 608 * Invalidate all entries to a particular vnode. 609 */ 610 void 611 cache_purge(vp) 612 struct vnode *vp; 613 { 614 615 CTR1(KTR_VFS, "cache_purge(%p)", vp); 616 CACHE_LOCK(); 617 while (!LIST_EMPTY(&vp->v_cache_src)) 618 cache_zap(LIST_FIRST(&vp->v_cache_src)); 619 while (!TAILQ_EMPTY(&vp->v_cache_dst)) 620 cache_zap(TAILQ_FIRST(&vp->v_cache_dst)); 621 vp->v_dd = NULL; 622 CACHE_UNLOCK(); 623 } 624 625 /* 626 * Flush all entries referencing a particular filesystem. 627 */ 628 void 629 cache_purgevfs(mp) 630 struct mount *mp; 631 { 632 struct nchashhead *ncpp; 633 struct namecache *ncp, *nnp; 634 635 /* Scan hash tables for applicable entries */ 636 CACHE_LOCK(); 637 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) { 638 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) { 639 if (ncp->nc_dvp->v_mount == mp) 640 cache_zap(ncp); 641 } 642 } 643 CACHE_UNLOCK(); 644 } 645 646 /* 647 * Perform canonical checks and cache lookup and pass on to filesystem 648 * through the vop_cachedlookup only if needed. 649 */ 650 651 int 652 vfs_cache_lookup(ap) 653 struct vop_lookup_args /* { 654 struct vnode *a_dvp; 655 struct vnode **a_vpp; 656 struct componentname *a_cnp; 657 } */ *ap; 658 { 659 struct vnode *dvp; 660 int error; 661 struct vnode **vpp = ap->a_vpp; 662 struct componentname *cnp = ap->a_cnp; 663 struct ucred *cred = cnp->cn_cred; 664 int flags = cnp->cn_flags; 665 struct thread *td = cnp->cn_thread; 666 667 *vpp = NULL; 668 dvp = ap->a_dvp; 669 670 if (dvp->v_type != VDIR) 671 return (ENOTDIR); 672 673 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 674 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 675 return (EROFS); 676 677 error = VOP_ACCESS(dvp, VEXEC, cred, td); 678 if (error) 679 return (error); 680 681 error = cache_lookup(dvp, vpp, cnp); 682 if (error == 0) 683 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); 684 if (error == -1) 685 return (0); 686 return (error); 687 } 688 689 690 #ifndef _SYS_SYSPROTO_H_ 691 struct __getcwd_args { 692 u_char *buf; 693 u_int buflen; 694 }; 695 #endif 696 697 /* 698 * XXX All of these sysctls would probably be more productive dead. 699 */ 700 static int disablecwd; 701 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, 702 "Disable the getcwd syscall"); 703 704 /* Implementation of the getcwd syscall. */ 705 int 706 __getcwd(td, uap) 707 struct thread *td; 708 struct __getcwd_args *uap; 709 { 710 711 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen)); 712 } 713 714 int 715 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen) 716 { 717 char *bp, *tmpbuf; 718 struct filedesc *fdp; 719 struct vnode *cdir, *rdir; 720 int error, vfslocked; 721 722 if (disablecwd) 723 return (ENODEV); 724 if (buflen < 2) 725 return (EINVAL); 726 if (buflen > MAXPATHLEN) 727 buflen = MAXPATHLEN; 728 729 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK); 730 fdp = td->td_proc->p_fd; 731 FILEDESC_SLOCK(fdp); 732 cdir = fdp->fd_cdir; 733 VREF(cdir); 734 rdir = fdp->fd_rdir; 735 VREF(rdir); 736 FILEDESC_SUNLOCK(fdp); 737 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen); 738 vfslocked = VFS_LOCK_GIANT(rdir->v_mount); 739 vrele(rdir); 740 VFS_UNLOCK_GIANT(vfslocked); 741 vfslocked = VFS_LOCK_GIANT(cdir->v_mount); 742 vrele(cdir); 743 VFS_UNLOCK_GIANT(vfslocked); 744 745 if (!error) { 746 if (bufseg == UIO_SYSSPACE) 747 bcopy(bp, buf, strlen(bp) + 1); 748 else 749 error = copyout(bp, buf, strlen(bp) + 1); 750 } 751 free(tmpbuf, M_TEMP); 752 return (error); 753 } 754 755 /* 756 * Thus begins the fullpath magic. 757 */ 758 759 #undef STATNODE 760 #define STATNODE(name) \ 761 static u_int name; \ 762 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "") 763 764 static int disablefullpath; 765 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0, 766 "Disable the vn_fullpath function"); 767 768 /* These count for kern___getcwd(), too. */ 769 STATNODE(numfullpathcalls); 770 STATNODE(numfullpathfail1); 771 STATNODE(numfullpathfail2); 772 STATNODE(numfullpathfail4); 773 STATNODE(numfullpathfound); 774 775 /* 776 * Retrieve the full filesystem path that correspond to a vnode from the name 777 * cache (if available) 778 */ 779 int 780 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) 781 { 782 char *buf; 783 struct filedesc *fdp; 784 struct vnode *rdir; 785 int error, vfslocked; 786 787 if (disablefullpath) 788 return (ENODEV); 789 if (vn == NULL) 790 return (EINVAL); 791 792 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 793 fdp = td->td_proc->p_fd; 794 FILEDESC_SLOCK(fdp); 795 rdir = fdp->fd_rdir; 796 VREF(rdir); 797 FILEDESC_SUNLOCK(fdp); 798 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN); 799 vfslocked = VFS_LOCK_GIANT(rdir->v_mount); 800 vrele(rdir); 801 VFS_UNLOCK_GIANT(vfslocked); 802 803 if (!error) 804 *freebuf = buf; 805 else 806 free(buf, M_TEMP); 807 return (error); 808 } 809 810 /* 811 * This function is similar to vn_fullpath, but it attempts to lookup the 812 * pathname relative to the global root mount point. This is required for the 813 * auditing sub-system, as audited pathnames must be absolute, relative to the 814 * global root mount point. 815 */ 816 int 817 vn_fullpath_global(struct thread *td, struct vnode *vn, 818 char **retbuf, char **freebuf) 819 { 820 char *buf; 821 int error; 822 823 if (disablefullpath) 824 return (ENODEV); 825 if (vn == NULL) 826 return (EINVAL); 827 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 828 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN); 829 if (!error) 830 *freebuf = buf; 831 else 832 free(buf, M_TEMP); 833 return (error); 834 } 835 836 /* 837 * The magic behind kern___getcwd() and vn_fullpath(). 838 */ 839 static int 840 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, 841 char *buf, char **retbuf, u_int buflen) 842 { 843 char *bp; 844 int error, i, slash_prefixed; 845 struct namecache *ncp; 846 847 bp = buf + buflen - 1; 848 *bp = '\0'; 849 error = 0; 850 slash_prefixed = 0; 851 852 CACHE_LOCK(); 853 numfullpathcalls++; 854 if (vp->v_type != VDIR) { 855 ncp = TAILQ_FIRST(&vp->v_cache_dst); 856 if (!ncp) { 857 numfullpathfail2++; 858 CACHE_UNLOCK(); 859 return (ENOENT); 860 } 861 for (i = ncp->nc_nlen - 1; i >= 0 && bp > buf; i--) 862 *--bp = ncp->nc_name[i]; 863 if (bp == buf) { 864 numfullpathfail4++; 865 CACHE_UNLOCK(); 866 return (ENOMEM); 867 } 868 *--bp = '/'; 869 slash_prefixed = 1; 870 vp = ncp->nc_dvp; 871 } 872 while (vp != rdir && vp != rootvnode) { 873 if (vp->v_vflag & VV_ROOT) { 874 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */ 875 error = EBADF; 876 break; 877 } 878 vp = vp->v_mount->mnt_vnodecovered; 879 continue; 880 } 881 if (vp->v_dd == NULL) { 882 numfullpathfail1++; 883 error = ENOTDIR; 884 break; 885 } 886 ncp = TAILQ_FIRST(&vp->v_cache_dst); 887 if (!ncp) { 888 numfullpathfail2++; 889 error = ENOENT; 890 break; 891 } 892 MPASS(ncp->nc_dvp == vp->v_dd); 893 for (i = ncp->nc_nlen - 1; i >= 0 && bp != buf; i--) 894 *--bp = ncp->nc_name[i]; 895 if (bp == buf) { 896 numfullpathfail4++; 897 error = ENOMEM; 898 break; 899 } 900 *--bp = '/'; 901 slash_prefixed = 1; 902 vp = ncp->nc_dvp; 903 } 904 if (error) { 905 CACHE_UNLOCK(); 906 return (error); 907 } 908 if (!slash_prefixed) { 909 if (bp == buf) { 910 numfullpathfail4++; 911 CACHE_UNLOCK(); 912 return (ENOMEM); 913 } else { 914 *--bp = '/'; 915 } 916 } 917 numfullpathfound++; 918 CACHE_UNLOCK(); 919 920 *retbuf = bp; 921 return (0); 922 } 923 924 int 925 vn_commname(struct vnode *vp, char *buf, u_int buflen) 926 { 927 struct namecache *ncp; 928 int l; 929 930 CACHE_LOCK(); 931 ncp = TAILQ_FIRST(&vp->v_cache_dst); 932 if (!ncp) { 933 CACHE_UNLOCK(); 934 return (ENOENT); 935 } 936 l = min(ncp->nc_nlen, buflen - 1); 937 memcpy(buf, ncp->nc_name, l); 938 CACHE_UNLOCK(); 939 buf[l] = '\0'; 940 return (0); 941 } 942