1 /* 2 * Copyright (c) 1989, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Poul-Henning Kamp of the FreeBSD Project. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95 37 * $Id: vfs_cache.c,v 1.32 1997/09/24 07:46:52 phk Exp $ 38 */ 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/sysctl.h> 44 #include <sys/mount.h> 45 #include <sys/vnode.h> 46 #include <sys/namei.h> 47 #include <sys/malloc.h> 48 49 50 /* 51 * Name caching works as follows: 52 * 53 * Names found by directory scans are retained in a cache 54 * for future reference. It is managed LRU, so frequently 55 * used names will hang around. Cache is indexed by hash value 56 * obtained from (vp, name) where vp refers to the directory 57 * containing name. 58 * 59 * If it is a "negative" entry, (i.e. for a name that is known NOT to 60 * exist) the vnode pointer will be NULL. 61 * 62 * Upon reaching the last segment of a path, if the reference 63 * is for DELETE, or NOCACHE is set (rewrite), and the 64 * name is located in the cache, it will be dropped. 65 */ 66 67 /* 68 * Structures associated with name cacheing. 69 */ 70 #define NCHHASH(dvp, cnp) \ 71 (&nchashtbl[((dvp)->v_id + (cnp)->cn_hash) & nchash]) 72 static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */ 73 static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */ 74 static u_long nchash; /* size of hash table */ 75 SYSCTL_INT(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, ""); 76 static u_long ncnegfactor = 16; /* ratio of negative entries */ 77 SYSCTL_INT(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, ""); 78 static u_long numneg; /* number of cache entries allocated */ 79 SYSCTL_INT(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, ""); 80 static u_long numcache; /* number of cache entries allocated */ 81 SYSCTL_INT(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, ""); 82 struct nchstats nchstats; /* cache effectiveness statistics */ 83 84 static int doingcache = 1; /* 1 => enable the cache */ 85 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, ""); 86 SYSCTL_INT(_debug, OID_AUTO, vnsize, CTLFLAG_RD, 0, sizeof(struct vnode), ""); 87 SYSCTL_INT(_debug, OID_AUTO, ncsize, CTLFLAG_RD, 0, sizeof(struct namecache), ""); 88 89 /* 90 * The new name cache statistics 91 */ 92 SYSCTL_NODE(_vfs, CTL_VFS, cache, CTLFLAG_RW, 0, 93 "Name cache statistics"); 94 #define STATNODE(mode, name, var) \ 95 SYSCTL_INT(_vfs_cache, OID_AUTO, name, mode, var, 0, ""); 96 STATNODE(CTLFLAG_RD, numneg, &numneg); 97 STATNODE(CTLFLAG_RD, numcache, &numcache); 98 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls); 99 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits); 100 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits); 101 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks); 102 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss); 103 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap); 104 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps); 105 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits); 106 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps); 107 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits); 108 109 110 static void cache_zap __P((struct namecache *ncp)); 111 112 /* 113 * Flags in namecache.nc_flag 114 */ 115 #define NCF_WHITE 1 116 /* 117 * Delete an entry from its hash list and move it to the front 118 * of the LRU list for immediate reuse. 119 */ 120 static void 121 cache_zap(ncp) 122 struct namecache *ncp; 123 { 124 LIST_REMOVE(ncp, nc_hash); 125 LIST_REMOVE(ncp, nc_src); 126 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) 127 vdrop(ncp->nc_dvp); 128 if (ncp->nc_vp) { 129 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); 130 } else { 131 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 132 numneg--; 133 } 134 numcache--; 135 free(ncp, M_CACHE); 136 } 137 138 /* 139 * Lookup an entry in the cache 140 * 141 * We don't do this if the segment name is long, simply so the cache 142 * can avoid holding long names (which would either waste space, or 143 * add greatly to the complexity). 144 * 145 * Lookup is called with dvp pointing to the directory to search, 146 * cnp pointing to the name of the entry being sought. If the lookup 147 * succeeds, the vnode is returned in *vpp, and a status of -1 is 148 * returned. If the lookup determines that the name does not exist 149 * (negative cacheing), a status of ENOENT is returned. If the lookup 150 * fails, a status of zero is returned. 151 */ 152 153 int 154 cache_lookup(dvp, vpp, cnp) 155 struct vnode *dvp; 156 struct vnode **vpp; 157 struct componentname *cnp; 158 { 159 register struct namecache *ncp, *nnp; 160 register struct nchashhead *ncpp; 161 162 if (!doingcache) { 163 cnp->cn_flags &= ~MAKEENTRY; 164 return (0); 165 } 166 167 numcalls++; 168 169 if (cnp->cn_nameptr[0] == '.') { 170 if (cnp->cn_namelen == 1) { 171 *vpp = dvp; 172 dothits++; 173 return (-1); 174 } 175 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 176 dotdothits++; 177 if (dvp->v_dd->v_id != dvp->v_ddid || 178 (cnp->cn_flags & MAKEENTRY) == 0) { 179 dvp->v_ddid = 0; 180 return (0); 181 } 182 *vpp = dvp->v_dd; 183 return (-1); 184 } 185 } 186 187 LIST_FOREACH(ncp, (NCHHASH(dvp, cnp)), nc_hash) { 188 numchecks++; 189 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && 190 !bcmp(ncp->nc_name, cnp->cn_nameptr, (u_int)ncp->nc_nlen)) 191 break; 192 } 193 194 /* We failed to find an entry */ 195 if (ncp == 0) { 196 if ((cnp->cn_flags & MAKEENTRY) == 0) { 197 nummisszap++; 198 } else { 199 nummiss++; 200 } 201 nchstats.ncs_miss++; 202 return (0); 203 } 204 205 /* We don't want to have an entry, so dump it */ 206 if ((cnp->cn_flags & MAKEENTRY) == 0) { 207 numposzaps++; 208 nchstats.ncs_badhits++; 209 cache_zap(ncp); 210 return (0); 211 } 212 213 /* We found a "positive" match, return the vnode */ 214 if (ncp->nc_vp) { 215 numposhits++; 216 nchstats.ncs_goodhits++; 217 *vpp = ncp->nc_vp; 218 return (-1); 219 } 220 221 /* We found a negative match, and want to create it, so purge */ 222 if (cnp->cn_nameiop == CREATE) { 223 numnegzaps++; 224 nchstats.ncs_badhits++; 225 cache_zap(ncp); 226 return (0); 227 } 228 229 numneghits++; 230 /* 231 * We found a "negative" match, ENOENT notifies client of this match. 232 * The nc_vpid field records whether this is a whiteout. 233 */ 234 TAILQ_REMOVE(&ncneg, ncp, nc_dst); 235 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 236 nchstats.ncs_neghits++; 237 if (ncp->nc_flag & NCF_WHITE) 238 cnp->cn_flags |= ISWHITEOUT; 239 return (ENOENT); 240 } 241 242 /* 243 * Add an entry to the cache. 244 */ 245 void 246 cache_enter(dvp, vp, cnp) 247 struct vnode *dvp; 248 struct vnode *vp; 249 struct componentname *cnp; 250 { 251 register struct namecache *ncp; 252 register struct nchashhead *ncpp; 253 254 if (!doingcache) 255 return; 256 257 if (cnp->cn_nameptr[0] == '.') { 258 if (cnp->cn_namelen == 1) { 259 return; 260 } 261 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { 262 if (vp) { 263 dvp->v_dd = vp; 264 dvp->v_ddid = vp->v_id; 265 } else { 266 dvp->v_dd = dvp; 267 dvp->v_ddid = 0; 268 } 269 return; 270 } 271 } 272 273 ncp = (struct namecache *) 274 malloc(sizeof *ncp + cnp->cn_namelen, M_CACHE, M_WAITOK); 275 bzero((char *)ncp, sizeof *ncp); 276 numcache++; 277 if (!vp) { 278 numneg++; 279 ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0; 280 } else if (vp->v_type == VDIR) { 281 vp->v_dd = dvp; 282 vp->v_ddid = dvp->v_id; 283 } 284 285 /* 286 * Fill in cache info, if vp is NULL this is a "negative" cache entry. 287 * For negative entries, we have to record whether it is a whiteout. 288 * the whiteout flag is stored in the nc_vpid field which is 289 * otherwise unused. 290 */ 291 ncp->nc_vp = vp; 292 ncp->nc_dvp = dvp; 293 ncp->nc_nlen = cnp->cn_namelen; 294 bcopy(cnp->cn_nameptr, ncp->nc_name, (unsigned)ncp->nc_nlen); 295 ncpp = NCHHASH(dvp, cnp); 296 LIST_INSERT_HEAD(ncpp, ncp, nc_hash); 297 if (LIST_EMPTY(&dvp->v_cache_src)) 298 vhold(dvp); 299 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); 300 if (vp) { 301 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); 302 } else { 303 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst); 304 } 305 if (numneg*ncnegfactor > numcache) { 306 ncp = TAILQ_FIRST(&ncneg); 307 cache_zap(ncp); 308 } 309 } 310 311 /* 312 * Name cache initialization, from vfs_init() when we are booting 313 */ 314 void 315 nchinit() 316 { 317 318 TAILQ_INIT(&ncneg); 319 nchashtbl = hashinit(desiredvnodes*2, M_CACHE, &nchash); 320 } 321 322 /* 323 * Invalidate all entries to particular vnode. 324 * 325 * We actually just increment the v_id, that will do it. The stale entries 326 * will be purged by lookup as they get found. If the v_id wraps around, we 327 * need to ditch the entire cache, to avoid confusion. No valid vnode will 328 * ever have (v_id == 0). 329 */ 330 void 331 cache_purge(vp) 332 struct vnode *vp; 333 { 334 struct namecache *ncp; 335 struct nchashhead *ncpp; 336 static u_long nextid; 337 338 while (!LIST_EMPTY(&vp->v_cache_src)) 339 cache_zap(LIST_FIRST(&vp->v_cache_src)); 340 while (!TAILQ_EMPTY(&vp->v_cache_dst)) 341 cache_zap(TAILQ_FIRST(&vp->v_cache_dst)); 342 343 nextid++; 344 while (nextid == vp->v_id || !nextid) 345 continue; 346 vp->v_id = nextid; 347 vp->v_dd = vp; 348 vp->v_ddid = 0; 349 } 350 351 /* 352 * Flush all entries referencing a particular filesystem. 353 * 354 * Since we need to check it anyway, we will flush all the invalid 355 * entries at the same time. 356 */ 357 void 358 cache_purgevfs(mp) 359 struct mount *mp; 360 { 361 struct nchashhead *ncpp; 362 struct namecache *ncp, *nnp; 363 364 /* Scan hash tables for applicable entries */ 365 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) { 366 for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) { 367 nnp = LIST_NEXT(ncp, nc_hash); 368 if (ncp->nc_dvp->v_mount == mp) { 369 cache_zap(ncp); 370 } 371 } 372 } 373 } 374 375 /* 376 * Perform canonical checks and cache lookup and pass on to filesystem 377 * through the vop_cachedlookup only if needed. 378 */ 379 380 int 381 vfs_cache_lookup(ap) 382 struct vop_lookup_args /* { 383 struct vnode *a_dvp; 384 struct vnode **a_vpp; 385 struct componentname *a_cnp; 386 } */ *ap; 387 { 388 struct vnode *vdp; 389 struct vnode *pdp; 390 int lockparent; 391 int error; 392 struct vnode **vpp = ap->a_vpp; 393 struct componentname *cnp = ap->a_cnp; 394 struct ucred *cred = cnp->cn_cred; 395 int flags = cnp->cn_flags; 396 struct proc *p = cnp->cn_proc; 397 u_long vpid; /* capability number of vnode */ 398 399 *vpp = NULL; 400 vdp = ap->a_dvp; 401 lockparent = flags & LOCKPARENT; 402 403 if (vdp->v_type != VDIR) 404 return (ENOTDIR); 405 406 if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) && 407 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 408 return (EROFS); 409 410 error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc); 411 412 if (error) 413 return (error); 414 415 error = cache_lookup(vdp, vpp, cnp); 416 417 if (!error) 418 return (VCALL(vdp, VOFFSET(vop_cachedlookup), 419 (struct vop_cachedlookup_args *)ap)); 420 421 if (error == ENOENT) 422 return (error); 423 424 pdp = vdp; 425 vdp = *vpp; 426 vpid = vdp->v_id; 427 if (pdp == vdp) { /* lookup on "." */ 428 VREF(vdp); 429 error = 0; 430 } else if (flags & ISDOTDOT) { 431 VOP_UNLOCK(pdp, 0, p); 432 error = vget(vdp, LK_EXCLUSIVE, p); 433 if (!error && lockparent && (flags & ISLASTCN)) 434 error = vn_lock(pdp, LK_EXCLUSIVE, p); 435 } else { 436 error = vget(vdp, LK_EXCLUSIVE, p); 437 if (!lockparent || error || !(flags & ISLASTCN)) 438 VOP_UNLOCK(pdp, 0, p); 439 } 440 /* 441 * Check that the capability number did not change 442 * while we were waiting for the lock. 443 */ 444 if (!error) { 445 if (vpid == vdp->v_id) 446 return (0); 447 vput(vdp); 448 if (lockparent && pdp != vdp && (flags & ISLASTCN)) 449 VOP_UNLOCK(pdp, 0, p); 450 } 451 error = vn_lock(pdp, LK_EXCLUSIVE, p); 452 if (error) 453 return (error); 454 return (VCALL(vdp, VOFFSET(vop_cachedlookup), 455 (struct vop_cachedlookup_args *)ap)); 456 } 457