Home | History | Annotate | Line # | Download | only in kern
vfs_cache.c revision 1.28
      1 /*	$NetBSD: vfs_cache.c,v 1.28 2000/11/24 07:25:51 chs Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1989, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by the University of
     18  *	California, Berkeley and its contributors.
     19  * 4. Neither the name of the University nor the names of its contributors
     20  *    may be used to endorse or promote products derived from this software
     21  *    without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33  * SUCH DAMAGE.
     34  *
     35  *	@(#)vfs_cache.c	8.3 (Berkeley) 8/22/94
     36  */
     37 
     38 #include "opt_ddb.h"
     39 
     40 #include <sys/param.h>
     41 #include <sys/systm.h>
     42 #include <sys/time.h>
     43 #include <sys/mount.h>
     44 #include <sys/vnode.h>
     45 #include <sys/namei.h>
     46 #include <sys/errno.h>
     47 #include <sys/malloc.h>
     48 #include <sys/pool.h>
     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 (dvp, name) where dvp refers to the directory
     57  * containing name.
     58  *
     59  * For simplicity (and economy of storage), names longer than
     60  * a maximum length of NCHNAMLEN are not cached; they occur
     61  * infrequently in any case, and are almost never of interest.
     62  *
     63  * Upon reaching the last segment of a path, if the reference
     64  * is for DELETE, or NOCACHE is set (rewrite), and the
     65  * name is located in the cache, it will be dropped.
     66  * The entry is dropped also when it was not possible to lock
     67  * the cached vnode, either because vget() failed or the generation
     68  * number has changed while waiting for the lock.
     69  */
     70 
     71 /*
     72  * Structures associated with name cacheing.
     73  */
     74 LIST_HEAD(nchashhead, namecache) *nchashtbl;
     75 u_long	nchash;				/* size of hash table - 1 */
     76 long	numcache;			/* number of cache entries allocated */
     77 
     78 LIST_HEAD(ncvhashhead, namecache) *ncvhashtbl;
     79 u_long	ncvhash;			/* size of hash table - 1 */
     80 
     81 TAILQ_HEAD(, namecache) nclruhead;		/* LRU chain */
     82 struct	nchstats nchstats;		/* cache effectiveness statistics */
     83 
     84 struct pool namecache_pool;
     85 
     86 int doingcache = 1;			/* 1 => enable the cache */
     87 
     88 /*
     89  * Look for a the name in the cache. We don't do this
     90  * if the segment name is long, simply so the cache can avoid
     91  * holding long names (which would either waste space, or
     92  * add greatly to the complexity).
     93  *
     94  * Lookup is called with ni_dvp pointing to the directory to search,
     95  * ni_ptr pointing to the name of the entry being sought, ni_namelen
     96  * tells the length of the name, and ni_hash contains a hash of
     97  * the name. If the lookup succeeds, the vnode is locked, stored in ni_vp
     98  * and a status of zero is returned. If the locking fails for whatever
     99  * reason, the vnode is unlocked and the error is returned to caller.
    100  * If the lookup determines that the name does not exist (negative cacheing),
    101  * a status of ENOENT is returned. If the lookup fails, a status of -1
    102  * is returned.
    103  */
    104 int
    105 cache_lookup(dvp, vpp, cnp)
    106 	struct vnode *dvp;
    107 	struct vnode **vpp;
    108 	struct componentname *cnp;
    109 {
    110 	struct namecache *ncp;
    111 	struct nchashhead *ncpp;
    112 	struct vnode *vp;
    113 	int vpid, error;
    114 
    115 	if (!doingcache) {
    116 		cnp->cn_flags &= ~MAKEENTRY;
    117 		*vpp = 0;
    118 		return (-1);
    119 	}
    120 	if (cnp->cn_namelen > NCHNAMLEN) {
    121 		nchstats.ncs_long++;
    122 		cnp->cn_flags &= ~MAKEENTRY;
    123 		*vpp = 0;
    124 		return (-1);
    125 	}
    126 	ncpp = &nchashtbl[(cnp->cn_hash ^ dvp->v_id) & nchash];
    127 	LIST_FOREACH(ncp, ncpp, nc_hash) {
    128 		if (ncp->nc_dvp == dvp &&
    129 		    ncp->nc_dvpid == dvp->v_id &&
    130 		    ncp->nc_nlen == cnp->cn_namelen &&
    131 		    !memcmp(ncp->nc_name, cnp->cn_nameptr, (u_int)ncp->nc_nlen))
    132 			break;
    133 	}
    134 	if (ncp == 0) {
    135 		nchstats.ncs_miss++;
    136 		*vpp = 0;
    137 		return (-1);
    138 	}
    139 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
    140 		nchstats.ncs_badhits++;
    141 		goto remove;
    142 	} else if (ncp->nc_vp == NULL) {
    143 		/*
    144 		 * Restore the ISWHITEOUT flag saved earlier.
    145 		 */
    146 		cnp->cn_flags |= ncp->nc_vpid;
    147 		if (cnp->cn_nameiop != CREATE ||
    148 		    (cnp->cn_flags & ISLASTCN) == 0) {
    149 			nchstats.ncs_neghits++;
    150 			/*
    151 			 * Move this slot to end of LRU chain,
    152 			 * if not already there.
    153 			 */
    154 			if (ncp->nc_lru.tqe_next != 0) {
    155 				TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
    156 				TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
    157 			}
    158 			return (ENOENT);
    159 		} else {
    160 			nchstats.ncs_badhits++;
    161 			goto remove;
    162 		}
    163 	} else if (ncp->nc_vpid != ncp->nc_vp->v_id) {
    164 		nchstats.ncs_falsehits++;
    165 		goto remove;
    166 	}
    167 
    168 	vp = ncp->nc_vp;
    169 	vpid = vp->v_id;
    170 	if (vp == dvp) {	/* lookup on "." */
    171 		VREF(dvp);
    172 		error = 0;
    173 	} else if (cnp->cn_flags & ISDOTDOT) {
    174 		VOP_UNLOCK(dvp, 0);
    175 		cnp->cn_flags |= PDIRUNLOCK;
    176 		error = vget(vp, LK_EXCLUSIVE);
    177 		/* if the above vget() succeeded and both LOCKPARENT and
    178 		 * ISLASTCN is set, lock the directory vnode as well */
    179 		if (!error && (~cnp->cn_flags & (LOCKPARENT|ISLASTCN)) == 0) {
    180 			if ((error = vn_lock(dvp, LK_EXCLUSIVE)) != 0) {
    181 				vput(vp);
    182 				return (error);
    183 			}
    184 			cnp->cn_flags &= ~PDIRUNLOCK;
    185 		}
    186 	} else {
    187 		error = vget(vp, LK_EXCLUSIVE);
    188 		/* if the above vget() failed or either of LOCKPARENT or
    189 		 * ISLASTCN is set, unlock the directory vnode */
    190 		if (error || (~cnp->cn_flags & (LOCKPARENT|ISLASTCN)) != 0) {
    191 			VOP_UNLOCK(dvp, 0);
    192 			cnp->cn_flags |= PDIRUNLOCK;
    193 		}
    194 	}
    195 
    196 	/*
    197 	 * Check that the lock succeeded, and that the capability number did
    198 	 * not change while we were waiting for the lock.
    199 	 */
    200 	if (error || vpid != vp->v_id) {
    201 		if (!error) {
    202 			vput(vp);
    203 			nchstats.ncs_falsehits++;
    204 		} else
    205 			nchstats.ncs_badhits++;
    206 		/*
    207 		 * The parent needs to be locked when we return to VOP_LOOKUP().
    208 		 * The `.' case here should be extremely rare (if it can happen
    209 		 * at all), so we don't bother optimizing out the unlock/relock.
    210 		 */
    211 		if (vp == dvp ||
    212 		    error || (~cnp->cn_flags & (LOCKPARENT|ISLASTCN)) != 0) {
    213 			if ((error = vn_lock(dvp, LK_EXCLUSIVE)) != 0)
    214 				return (error);
    215 			cnp->cn_flags &= ~PDIRUNLOCK;
    216 		}
    217 		goto remove;
    218 	}
    219 
    220 	nchstats.ncs_goodhits++;
    221 	/*
    222 	 * move this slot to end of LRU chain, if not already there
    223 	 */
    224 	if (ncp->nc_lru.tqe_next != 0) {
    225 		TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
    226 		TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
    227 	}
    228 	*vpp = vp;
    229 	return (0);
    230 
    231 remove:
    232 	/*
    233 	 * Last component and we are renaming or deleting,
    234 	 * the cache entry is invalid, or otherwise don't
    235 	 * want cache entry to exist.
    236 	 */
    237 	TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
    238 	LIST_REMOVE(ncp, nc_hash);
    239 	ncp->nc_hash.le_prev = 0;
    240 	if (ncp->nc_vhash.le_prev != NULL) {
    241 		LIST_REMOVE(ncp, nc_vhash);
    242 		ncp->nc_vhash.le_prev = 0;
    243 	}
    244 	TAILQ_INSERT_HEAD(&nclruhead, ncp, nc_lru);
    245 	*vpp = 0;
    246 	return (-1);
    247 }
    248 
    249 /*
    250  * Scan cache looking for name of directory entry pointing at vp.
    251  *
    252  * Fill in dvpp.
    253  *
    254  * If bufp is non-NULL, also place the name in the buffer which starts
    255  * at bufp, immediately before *bpp, and move bpp backwards to point
    256  * at the start of it.  (Yes, this is a little baroque, but it's done
    257  * this way to cater to the whims of getcwd).
    258  *
    259  * Returns 0 on success, -1 on cache miss, positive errno on failure.
    260  */
    261 int
    262 cache_revlookup (vp, dvpp, bpp, bufp)
    263 	struct vnode *vp, **dvpp;
    264 	char **bpp;
    265 	char *bufp;
    266 {
    267 	struct namecache *ncp;
    268 	struct vnode *dvp;
    269 	struct ncvhashhead *nvcpp;
    270 
    271 	if (!doingcache)
    272 		goto out;
    273 
    274 	nvcpp = &ncvhashtbl[(vp->v_id & ncvhash)];
    275 
    276 	LIST_FOREACH(ncp, nvcpp, nc_vhash) {
    277 		if ((ncp->nc_vp == vp) &&
    278 		    (ncp->nc_vpid == vp->v_id) &&
    279 		    ((dvp = ncp->nc_dvp) != 0) &&
    280 		    (dvp != vp) && 		/* avoid pesky . entries.. */
    281 		    (dvp->v_id == ncp->nc_dvpid))
    282 		{
    283 			char *bp;
    284 
    285 #ifdef DIAGNOSTIC
    286 			if ((ncp->nc_nlen == 1) &&
    287 			    (ncp->nc_name[0] == '.'))
    288 				panic("cache_revlookup: found entry for .");
    289 
    290 			if ((ncp->nc_nlen == 2) &&
    291 			    (ncp->nc_name[0] == '.') &&
    292 			    (ncp->nc_name[1] == '.'))
    293 				panic("cache_revlookup: found entry for ..");
    294 #endif
    295 			nchstats.ncs_revhits++;
    296 
    297 			if (bufp) {
    298 				bp = *bpp;
    299 				bp -= ncp->nc_nlen;
    300 				if (bp <= bufp) {
    301 					*dvpp = 0;
    302 					return ERANGE;
    303 				}
    304 				memcpy(bp, ncp->nc_name, ncp->nc_nlen);
    305 				*bpp = bp;
    306 			}
    307 
    308 			/* XXX MP: how do we know dvp won't evaporate? */
    309 			*dvpp = dvp;
    310 			return 0;
    311 		}
    312 	}
    313 	nchstats.ncs_revmiss++;
    314  out:
    315 	*dvpp = 0;
    316 	return -1;
    317 }
    318 
    319 /*
    320  * Add an entry to the cache
    321  */
    322 void
    323 cache_enter(dvp, vp, cnp)
    324 	struct vnode *dvp;
    325 	struct vnode *vp;
    326 	struct componentname *cnp;
    327 {
    328 	struct namecache *ncp;
    329 	struct nchashhead *ncpp;
    330 	struct ncvhashhead *nvcpp;
    331 
    332 #ifdef DIAGNOSTIC
    333 	if (cnp->cn_namelen > NCHNAMLEN)
    334 		panic("cache_enter: name too long");
    335 #endif
    336 	if (!doingcache)
    337 		return;
    338 	/*
    339 	 * Free the cache slot at head of lru chain.
    340 	 */
    341 	if (numcache < numvnodes) {
    342 		ncp = pool_get(&namecache_pool, PR_WAITOK);
    343 		memset(ncp, 0, sizeof(*ncp));
    344 		numcache++;
    345 	} else if ((ncp = TAILQ_FIRST(&nclruhead)) != NULL) {
    346 		TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
    347 		if (ncp->nc_hash.le_prev != 0) {
    348 			LIST_REMOVE(ncp, nc_hash);
    349 			ncp->nc_hash.le_prev = 0;
    350 		}
    351 		if (ncp->nc_vhash.le_prev != 0) {
    352 			LIST_REMOVE(ncp, nc_vhash);
    353 			ncp->nc_vhash.le_prev = 0;
    354 		}
    355 	} else
    356 		return;
    357 	/* grab the vnode we just found */
    358 	ncp->nc_vp = vp;
    359 	if (vp)
    360 		ncp->nc_vpid = vp->v_id;
    361 	else {
    362 		/*
    363 		 * For negative hits, save the ISWHITEOUT flag so we can
    364 		 * restore it later when the cache entry is used again.
    365 		 */
    366 		ncp->nc_vpid = cnp->cn_flags & ISWHITEOUT;
    367 	}
    368 	/* fill in cache info */
    369 	ncp->nc_dvp = dvp;
    370 	ncp->nc_dvpid = dvp->v_id;
    371 	ncp->nc_nlen = cnp->cn_namelen;
    372 	memcpy(ncp->nc_name, cnp->cn_nameptr, (unsigned)ncp->nc_nlen);
    373 	TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
    374 	ncpp = &nchashtbl[(cnp->cn_hash ^ dvp->v_id) & nchash];
    375 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
    376 
    377 	ncp->nc_vhash.le_prev = 0;
    378 	ncp->nc_vhash.le_next = 0;
    379 
    380 	/*
    381 	 * Create reverse-cache entries (used in getcwd) for directories.
    382 	 */
    383 	if (vp &&
    384 	    (vp != dvp) &&
    385 	    (vp->v_type == VDIR) &&
    386 	    ((ncp->nc_nlen > 2) ||
    387 	     ((ncp->nc_nlen == 2) && (ncp->nc_name[0] != '.') && (ncp->nc_name[1] != '.')) ||
    388 	     ((ncp->nc_nlen == 1) && (ncp->nc_name[0] != '.'))))
    389 	{
    390 		nvcpp = &ncvhashtbl[(vp->v_id & ncvhash)];
    391 		LIST_INSERT_HEAD(nvcpp, ncp, nc_vhash);
    392 	}
    393 
    394 }
    395 
    396 /*
    397  * Name cache initialization, from vfs_init() when we are booting
    398  */
    399 void
    400 nchinit()
    401 {
    402 
    403 	TAILQ_INIT(&nclruhead);
    404 	nchashtbl =
    405 	    hashinit(desiredvnodes, HASH_LIST, M_CACHE, M_WAITOK, &nchash);
    406 	ncvhashtbl =
    407 	    hashinit(desiredvnodes/8, HASH_LIST, M_CACHE, M_WAITOK, &ncvhash);
    408 	pool_init(&namecache_pool, sizeof(struct namecache), 0, 0, 0,
    409 	    "ncachepl", 0, pool_page_alloc_nointr, pool_page_free_nointr,
    410 	    M_CACHE);
    411 }
    412 
    413 /*
    414  * Cache flush, a particular vnode; called when a vnode is renamed to
    415  * hide entries that would now be invalid
    416  */
    417 void
    418 cache_purge(vp)
    419 	struct vnode *vp;
    420 {
    421 	struct namecache *ncp;
    422 	struct nchashhead *ncpp;
    423 	static u_long nextvnodeid;
    424 
    425 	vp->v_id = ++nextvnodeid;
    426 	if (nextvnodeid != 0)
    427 		return;
    428 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
    429 		LIST_FOREACH(ncp, ncpp, nc_hash) {
    430 			ncp->nc_vpid = 0;
    431 			ncp->nc_dvpid = 0;
    432 		}
    433 	}
    434 	vp->v_id = ++nextvnodeid;
    435 }
    436 
    437 /*
    438  * Cache flush, a whole filesystem; called when filesys is umounted to
    439  * remove entries that would now be invalid.
    440  */
    441 void
    442 cache_purgevfs(mp)
    443 	struct mount *mp;
    444 {
    445 	struct namecache *ncp, *nxtcp;
    446 
    447 	for (ncp = TAILQ_FIRST(&nclruhead); ncp != NULL; ncp = nxtcp) {
    448 		nxtcp = TAILQ_NEXT(ncp, nc_lru);
    449 		if (ncp->nc_dvp == NULL || ncp->nc_dvp->v_mount != mp) {
    450 			continue;
    451 		}
    452 		/* free the resources we had */
    453 		ncp->nc_vp = NULL;
    454 		ncp->nc_dvp = NULL;
    455 		TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
    456 		if (ncp->nc_hash.le_prev != 0) {
    457 			LIST_REMOVE(ncp, nc_hash);
    458 			ncp->nc_hash.le_prev = 0;
    459 		}
    460 		if (ncp->nc_vhash.le_prev != 0) {
    461 			LIST_REMOVE(ncp, nc_vhash);
    462 			ncp->nc_vhash.le_prev = 0;
    463 		}
    464 		TAILQ_INSERT_HEAD(&nclruhead, ncp, nc_lru);
    465 	}
    466 }
    467 
    468 #ifdef DDB
    469 void
    470 namecache_print(struct vnode *vp, void (*pr)(const char *, ...))
    471 {
    472 	struct vnode *dvp = NULL;
    473 	struct namecache *ncp;
    474 
    475 	TAILQ_FOREACH(ncp, &nclruhead, nc_lru) {
    476 		if (ncp->nc_vp == vp && ncp->nc_vpid == vp->v_id) {
    477 			(*pr)("name %.*s\n", ncp->nc_nlen, ncp->nc_name);
    478 			dvp = ncp->nc_dvp;
    479 		}
    480 	}
    481 	if (dvp == NULL) {
    482 		(*pr)("name not found\n");
    483 		return;
    484 	}
    485 	vp = dvp;
    486 	TAILQ_FOREACH(ncp, &nclruhead, nc_lru) {
    487 		if (ncp->nc_vp == vp && ncp->nc_vpid == vp->v_id) {
    488 			(*pr)("parent %.*s\n", ncp->nc_nlen, ncp->nc_name);
    489 		}
    490 	}
    491 }
    492 #endif
    493