xref: /freebsd/sys/kern/vfs_cache.c (revision 33b77e2decd50e53798014b70bf7ca3bdc4c0c7e)
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 1.36 1997/11/07 08:53:05 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;
160 
161 	if (!doingcache) {
162 		cnp->cn_flags &= ~MAKEENTRY;
163 		return (0);
164 	}
165 
166 	numcalls++;
167 
168 	if (cnp->cn_nameptr[0] == '.') {
169 		if (cnp->cn_namelen == 1) {
170 			*vpp = dvp;
171 			dothits++;
172 			return (-1);
173 		}
174 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
175 			dotdothits++;
176 			if (dvp->v_dd->v_id != dvp->v_ddid ||
177 			    (cnp->cn_flags & MAKEENTRY) == 0) {
178 				dvp->v_ddid = 0;
179 				return (0);
180 			}
181 			*vpp = dvp->v_dd;
182 			return (-1);
183 		}
184 	}
185 
186 	LIST_FOREACH(ncp, (NCHHASH(dvp, cnp)), nc_hash) {
187 		numchecks++;
188 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
189 		    !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
190 			break;
191 	}
192 
193 	/* We failed to find an entry */
194 	if (ncp == 0) {
195 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
196 			nummisszap++;
197 		} else {
198 			nummiss++;
199 		}
200 		nchstats.ncs_miss++;
201 		return (0);
202 	}
203 
204 	/* We don't want to have an entry, so dump it */
205 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
206 		numposzaps++;
207 		nchstats.ncs_badhits++;
208 		cache_zap(ncp);
209 		return (0);
210 	}
211 
212 	/* We found a "positive" match, return the vnode */
213         if (ncp->nc_vp) {
214 		numposhits++;
215 		nchstats.ncs_goodhits++;
216 		*vpp = ncp->nc_vp;
217 		return (-1);
218 	}
219 
220 	/* We found a negative match, and want to create it, so purge */
221 	if (cnp->cn_nameiop == CREATE) {
222 		numnegzaps++;
223 		nchstats.ncs_badhits++;
224 		cache_zap(ncp);
225 		return (0);
226 	}
227 
228 	numneghits++;
229 	/*
230 	 * We found a "negative" match, ENOENT notifies client of this match.
231 	 * The nc_vpid field records whether this is a whiteout.
232 	 */
233 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
234 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
235 	nchstats.ncs_neghits++;
236 	if (ncp->nc_flag & NCF_WHITE)
237 		cnp->cn_flags |= ISWHITEOUT;
238 	return (ENOENT);
239 }
240 
241 /*
242  * Add an entry to the cache.
243  */
244 void
245 cache_enter(dvp, vp, cnp)
246 	struct vnode *dvp;
247 	struct vnode *vp;
248 	struct componentname *cnp;
249 {
250 	register struct namecache *ncp;
251 	register struct nchashhead *ncpp;
252 
253 	if (!doingcache)
254 		return;
255 
256 	if (cnp->cn_nameptr[0] == '.') {
257 		if (cnp->cn_namelen == 1) {
258 			return;
259 		}
260 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
261 			if (vp) {
262 				dvp->v_dd = vp;
263 				dvp->v_ddid = vp->v_id;
264 			} else {
265 				dvp->v_dd = dvp;
266 				dvp->v_ddid = 0;
267 			}
268 			return;
269 		}
270 	}
271 
272 	ncp = (struct namecache *)
273 		malloc(sizeof *ncp + cnp->cn_namelen, M_CACHE, M_WAITOK);
274 	bzero((char *)ncp, sizeof *ncp);
275 	numcache++;
276 	if (!vp) {
277 		numneg++;
278 		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
279 	} else if (vp->v_type == VDIR) {
280 		vp->v_dd = dvp;
281 		vp->v_ddid = dvp->v_id;
282 	}
283 
284 	/*
285 	 * Fill in cache info, if vp is NULL this is a "negative" cache entry.
286 	 * For negative entries, we have to record whether it is a whiteout.
287 	 * the whiteout flag is stored in the nc_vpid field which is
288 	 * otherwise unused.
289 	 */
290 	ncp->nc_vp = vp;
291 	ncp->nc_dvp = dvp;
292 	ncp->nc_nlen = cnp->cn_namelen;
293 	bcopy(cnp->cn_nameptr, ncp->nc_name, ncp->nc_nlen);
294 	ncpp = NCHHASH(dvp, cnp);
295 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
296 	if (LIST_EMPTY(&dvp->v_cache_src))
297 		vhold(dvp);
298 	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
299 	if (vp) {
300 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
301 	} else {
302 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
303 	}
304 	if (numneg*ncnegfactor > numcache) {
305 		ncp = TAILQ_FIRST(&ncneg);
306 		cache_zap(ncp);
307 	}
308 }
309 
310 /*
311  * Name cache initialization, from vfs_init() when we are booting
312  */
313 void
314 nchinit()
315 {
316 
317 	TAILQ_INIT(&ncneg);
318 	nchashtbl = hashinit(desiredvnodes*2, M_CACHE, &nchash);
319 }
320 
321 /*
322  * Invalidate all entries to particular vnode.
323  *
324  * We actually just increment the v_id, that will do it. The stale entries
325  * will be purged by lookup as they get found. If the v_id wraps around, we
326  * need to ditch the entire cache, to avoid confusion. No valid vnode will
327  * ever have (v_id == 0).
328  */
329 void
330 cache_purge(vp)
331 	struct vnode *vp;
332 {
333 	static u_long nextid;
334 
335 	while (!LIST_EMPTY(&vp->v_cache_src))
336 		cache_zap(LIST_FIRST(&vp->v_cache_src));
337 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
338 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
339 
340 	nextid++;
341 	while (nextid == vp->v_id || !nextid)
342 		continue;
343 	vp->v_id = nextid;
344 	vp->v_dd = vp;
345 	vp->v_ddid = 0;
346 }
347 
348 /*
349  * Flush all entries referencing a particular filesystem.
350  *
351  * Since we need to check it anyway, we will flush all the invalid
352  * entries at the same time.
353  */
354 void
355 cache_purgevfs(mp)
356 	struct mount *mp;
357 {
358 	struct nchashhead *ncpp;
359 	struct namecache *ncp, *nnp;
360 
361 	/* Scan hash tables for applicable entries */
362 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
363 		for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
364 			nnp = LIST_NEXT(ncp, nc_hash);
365 			if (ncp->nc_dvp->v_mount == mp) {
366 				cache_zap(ncp);
367 			}
368 		}
369 	}
370 }
371 
372 /*
373  * Perform canonical checks and cache lookup and pass on to filesystem
374  * through the vop_cachedlookup only if needed.
375  */
376 
377 int
378 vfs_cache_lookup(ap)
379 	struct vop_lookup_args /* {
380 		struct vnode *a_dvp;
381 		struct vnode **a_vpp;
382 		struct componentname *a_cnp;
383 	} */ *ap;
384 {
385 	struct vnode *vdp;
386 	struct vnode *pdp;
387 	int lockparent;
388 	int error;
389 	struct vnode **vpp = ap->a_vpp;
390 	struct componentname *cnp = ap->a_cnp;
391 	struct ucred *cred = cnp->cn_cred;
392 	int flags = cnp->cn_flags;
393 	struct proc *p = cnp->cn_proc;
394 	u_long vpid;	/* capability number of vnode */
395 
396 	*vpp = NULL;
397 	vdp = ap->a_dvp;
398 	lockparent = flags & LOCKPARENT;
399 
400 	if (vdp->v_type != VDIR)
401                 return (ENOTDIR);
402 
403 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
404 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
405 		return (EROFS);
406 
407 	error = VOP_ACCESS(vdp, VEXEC, cred, cnp->cn_proc);
408 
409 	if (error)
410 		return (error);
411 
412 	error = cache_lookup(vdp, vpp, cnp);
413 
414 	if (!error)
415 		return (VOP_CACHEDLOOKUP(ap->a_dvp, ap->a_vpp, ap->a_cnp));
416 
417 	if (error == ENOENT)
418 		return (error);
419 
420 	pdp = vdp;
421 	vdp = *vpp;
422 	vpid = vdp->v_id;
423 	if (pdp == vdp) {   /* lookup on "." */
424 		VREF(vdp);
425 		error = 0;
426 	} else if (flags & ISDOTDOT) {
427 		VOP_UNLOCK(pdp, 0, p);
428 		error = vget(vdp, LK_EXCLUSIVE, p);
429 		if (!error && lockparent && (flags & ISLASTCN))
430 			error = vn_lock(pdp, LK_EXCLUSIVE, p);
431 	} else {
432 		error = vget(vdp, LK_EXCLUSIVE, p);
433 		if (!lockparent || error || !(flags & ISLASTCN))
434 			VOP_UNLOCK(pdp, 0, p);
435 	}
436 	/*
437 	 * Check that the capability number did not change
438 	 * while we were waiting for the lock.
439 	 */
440 	if (!error) {
441 		if (vpid == vdp->v_id)
442 			return (0);
443 		vput(vdp);
444 		if (lockparent && pdp != vdp && (flags & ISLASTCN))
445 			VOP_UNLOCK(pdp, 0, p);
446 	}
447 	error = vn_lock(pdp, LK_EXCLUSIVE, p);
448 	if (error)
449 		return (error);
450 	return (VOP_CACHEDLOOKUP(ap->a_dvp, ap->a_vpp, ap->a_cnp));
451 }
452