xref: /freebsd/sys/kern/vfs_cache.c (revision 52ec752989b2e6d4e9a59a8ff25d8ff596d85e62)
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  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/mutex.h>
47 #include <sys/sysctl.h>
48 #include <sys/mount.h>
49 #include <sys/vnode.h>
50 #include <sys/namei.h>
51 #include <sys/malloc.h>
52 #include <sys/syscallsubr.h>
53 #include <sys/sysproto.h>
54 #include <sys/proc.h>
55 #include <sys/filedesc.h>
56 #include <sys/fnv_hash.h>
57 
58 #include <vm/uma.h>
59 
60 /*
61  * This structure describes the elements in the cache of recent
62  * names looked up by namei.
63  */
64 
65 struct	namecache {
66 	LIST_ENTRY(namecache) nc_hash;	/* hash chain */
67 	LIST_ENTRY(namecache) nc_src;	/* source vnode list */
68 	TAILQ_ENTRY(namecache) nc_dst;	/* destination vnode list */
69 	struct	vnode *nc_dvp;		/* vnode of parent of name */
70 	struct	vnode *nc_vp;		/* vnode the name refers to */
71 	u_char	nc_flag;		/* flag bits */
72 	u_char	nc_nlen;		/* length of name */
73 	char	nc_name[0];		/* segment name */
74 };
75 
76 /*
77  * Name caching works as follows:
78  *
79  * Names found by directory scans are retained in a cache
80  * for future reference.  It is managed LRU, so frequently
81  * used names will hang around.  Cache is indexed by hash value
82  * obtained from (vp, name) where vp refers to the directory
83  * containing name.
84  *
85  * If it is a "negative" entry, (i.e. for a name that is known NOT to
86  * exist) the vnode pointer will be NULL.
87  *
88  * Upon reaching the last segment of a path, if the reference
89  * is for DELETE, or NOCACHE is set (rewrite), and the
90  * name is located in the cache, it will be dropped.
91  */
92 
93 /*
94  * Structures associated with name cacheing.
95  */
96 #define NCHHASH(hash) \
97 	(&nchashtbl[(hash) & nchash])
98 static LIST_HEAD(nchashhead, namecache) *nchashtbl;	/* Hash Table */
99 static TAILQ_HEAD(, namecache) ncneg;	/* Hash Table */
100 static u_long	nchash;			/* size of hash table */
101 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
102 static u_long	ncnegfactor = 16;	/* ratio of negative entries */
103 SYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
104 static u_long	numneg;			/* number of cache entries allocated */
105 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
106 static u_long	numcache;		/* number of cache entries allocated */
107 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
108 static u_long	numcachehv;		/* number of cache entries with vnodes held */
109 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, "");
110 #if 0
111 static u_long	numcachepl;		/* number of cache purge for leaf entries */
112 SYSCTL_ULONG(_debug, OID_AUTO, numcachepl, CTLFLAG_RD, &numcachepl, 0, "");
113 #endif
114 struct	nchstats nchstats;		/* cache effectiveness statistics */
115 
116 struct mtx cache_lock;
117 MTX_SYSINIT(vfscache, &cache_lock, "Name Cache", MTX_DEF);
118 
119 #define	CACHE_LOCK()	mtx_lock(&cache_lock)
120 #define	CACHE_UNLOCK()	mtx_unlock(&cache_lock)
121 
122 /*
123  * UMA zones for the VFS cache.
124  *
125  * The small cache is used for entries with short names, which are the
126  * most common.  The large cache is used for entries which are too big to
127  * fit in the small cache.
128  */
129 static uma_zone_t cache_zone_small;
130 static uma_zone_t cache_zone_large;
131 
132 #define	CACHE_PATH_CUTOFF	32
133 #define	CACHE_ZONE_SMALL	(sizeof(struct namecache) + CACHE_PATH_CUTOFF)
134 #define	CACHE_ZONE_LARGE	(sizeof(struct namecache) + NAME_MAX)
135 
136 #define cache_alloc(len)	uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \
137 	cache_zone_small : cache_zone_large, M_WAITOK)
138 #define cache_free(ncp)		do { \
139 	if (ncp != NULL) \
140 		uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \
141 		    cache_zone_small : cache_zone_large, (ncp)); \
142 } while (0)
143 
144 static int	doingcache = 1;		/* 1 => enable the cache */
145 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
146 
147 /* Export size information to userland */
148 SYSCTL_INT(_debug, OID_AUTO, vnsize, CTLFLAG_RD, 0, sizeof(struct vnode), "");
149 SYSCTL_INT(_debug, OID_AUTO, ncsize, CTLFLAG_RD, 0, sizeof(struct namecache), "");
150 
151 /*
152  * The new name cache statistics
153  */
154 SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
155 #define STATNODE(mode, name, var) \
156 	SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
157 STATNODE(CTLFLAG_RD, numneg, &numneg);
158 STATNODE(CTLFLAG_RD, numcache, &numcache);
159 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
160 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
161 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
162 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
163 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
164 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
165 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
166 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
167 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
168 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
169 
170 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD, &nchstats,
171 	sizeof(nchstats), "LU", "VFS cache effectiveness statistics");
172 
173 
174 
175 static void cache_zap(struct namecache *ncp, int locked);
176 
177 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
178 
179 /*
180  * Flags in namecache.nc_flag
181  */
182 #define NCF_WHITE	1
183 
184 /*
185  * Grab an atomic snapshot of the name cache hash chain lengths
186  */
187 SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
188 
189 static int
190 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
191 {
192 	int error;
193 	struct nchashhead *ncpp;
194 	struct namecache *ncp;
195 	int n_nchash;
196 	int count;
197 
198 	n_nchash = nchash + 1;	/* nchash is max index, not count */
199 	if (!req->oldptr)
200 		return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
201 
202 	/* Scan hash tables for applicable entries */
203 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
204 		count = 0;
205 		LIST_FOREACH(ncp, ncpp, nc_hash) {
206 			count++;
207 		}
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 		LIST_FOREACH(ncp, ncpp, nc_hash) {
237 			count++;
238 		}
239 		if (count)
240 			used++;
241 		if (maxlength < count)
242 			maxlength = count;
243 	}
244 	n_nchash = nchash + 1;
245 	pct = (used * 100 * 100) / n_nchash;
246 	error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
247 	if (error)
248 		return (error);
249 	error = SYSCTL_OUT(req, &used, sizeof(used));
250 	if (error)
251 		return (error);
252 	error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
253 	if (error)
254 		return (error);
255 	error = SYSCTL_OUT(req, &pct, sizeof(pct));
256 	if (error)
257 		return (error);
258 	return (0);
259 }
260 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD,
261 	0, 0, sysctl_debug_hashstat_nchash, "I", "nchash chain lengths");
262 
263 /*
264  * cache_zap():
265  *
266  *   Removes a namecache entry from cache, whether it contains an actual
267  *   pointer to a vnode or if it is just a negative cache entry.
268  */
269 static void
270 cache_zap(ncp, locked)
271 	struct namecache *ncp;
272 	int locked;
273 {
274 	struct vnode *vp;
275 
276 	vp = NULL;
277 	if (!locked)
278 		CACHE_LOCK();
279 	LIST_REMOVE(ncp, nc_hash);
280 	LIST_REMOVE(ncp, nc_src);
281 	if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
282 		vp = ncp->nc_dvp;
283 		numcachehv--;
284 	}
285 	if (ncp->nc_vp) {
286 		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
287 	} else {
288 		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
289 		numneg--;
290 	}
291 	numcache--;
292 	CACHE_UNLOCK();
293 	cache_free(ncp);
294 	if (vp)
295 		vdrop(vp);
296 	if (locked)
297 		CACHE_LOCK();
298 }
299 
300 /*
301  * cache_leaf_test()
302  *
303  *      Test whether this (directory) vnode's namei cache entry contains
304  *      subdirectories or not.  Used to determine whether the directory is
305  *      a leaf in the namei cache or not.  Note: the directory may still
306  *      contain files in the namei cache.
307  *
308  *      Returns 0 if the directory is a leaf, -1 if it isn't.
309  */
310 int
311 cache_leaf_test(struct vnode *vp)
312 {
313 	struct namecache *ncpc;
314 	int leaf;
315 
316 	leaf = 0;
317 	CACHE_LOCK();
318 	for (ncpc = LIST_FIRST(&vp->v_cache_src);
319 	     ncpc != NULL;
320 	     ncpc = LIST_NEXT(ncpc, nc_src)
321 	 ) {
322 		if (ncpc->nc_vp != NULL && ncpc->nc_vp->v_type == VDIR) {
323 			leaf = -1;
324 			break;
325 		}
326 	}
327 	CACHE_UNLOCK();
328 	return (leaf);
329 }
330 
331 /*
332  * Lookup an entry in the cache
333  *
334  * Lookup is called with dvp pointing to the directory to search,
335  * cnp pointing to the name of the entry being sought. If the lookup
336  * succeeds, the vnode is returned in *vpp, and a status of -1 is
337  * returned. If the lookup determines that the name does not exist
338  * (negative cacheing), a status of ENOENT is returned. If the lookup
339  * fails, a status of zero is returned.
340  */
341 
342 int
343 cache_lookup(dvp, vpp, cnp)
344 	struct vnode *dvp;
345 	struct vnode **vpp;
346 	struct componentname *cnp;
347 {
348 	struct namecache *ncp;
349 	u_int32_t hash;
350 
351 	if (!doingcache) {
352 		cnp->cn_flags &= ~MAKEENTRY;
353 		return (0);
354 	}
355 
356 	CACHE_LOCK();
357 	numcalls++;
358 
359 	if (cnp->cn_nameptr[0] == '.') {
360 		if (cnp->cn_namelen == 1) {
361 			*vpp = dvp;
362 			dothits++;
363 			CACHE_UNLOCK();
364 			return (-1);
365 		}
366 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
367 			dotdothits++;
368 			if (dvp->v_dd->v_id != dvp->v_ddid ||
369 			    (cnp->cn_flags & MAKEENTRY) == 0) {
370 				dvp->v_ddid = 0;
371 				CACHE_UNLOCK();
372 				return (0);
373 			}
374 			*vpp = dvp->v_dd;
375 			CACHE_UNLOCK();
376 			return (-1);
377 		}
378 	}
379 
380 	hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
381 	hash = fnv_32_buf(&dvp->v_id, sizeof(dvp->v_id), hash);
382 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
383 		numchecks++;
384 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
385 		    !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
386 			break;
387 	}
388 
389 	/* We failed to find an entry */
390 	if (ncp == 0) {
391 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
392 			nummisszap++;
393 		} else {
394 			nummiss++;
395 		}
396 		nchstats.ncs_miss++;
397 		CACHE_UNLOCK();
398 		return (0);
399 	}
400 
401 	/* We don't want to have an entry, so dump it */
402 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
403 		numposzaps++;
404 		nchstats.ncs_badhits++;
405 		CACHE_UNLOCK();
406 		cache_zap(ncp, 0);
407 		return (0);
408 	}
409 
410 	/* We found a "positive" match, return the vnode */
411 	if (ncp->nc_vp) {
412 		numposhits++;
413 		nchstats.ncs_goodhits++;
414 		*vpp = ncp->nc_vp;
415 		CACHE_UNLOCK();
416 		return (-1);
417 	}
418 
419 	/* We found a negative match, and want to create it, so purge */
420 	if (cnp->cn_nameiop == CREATE) {
421 		numnegzaps++;
422 		nchstats.ncs_badhits++;
423 		CACHE_UNLOCK();
424 		cache_zap(ncp, 0);
425 		return (0);
426 	}
427 
428 	numneghits++;
429 	/*
430 	 * We found a "negative" match, so we shift it to the end of
431 	 * the "negative" cache entries queue to satisfy LRU.  Also,
432 	 * check to see if the entry is a whiteout; indicate this to
433 	 * the componentname, if so.
434 	 */
435 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
436 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
437 	nchstats.ncs_neghits++;
438 	if (ncp->nc_flag & NCF_WHITE)
439 		cnp->cn_flags |= ISWHITEOUT;
440 	CACHE_UNLOCK();
441 	return (ENOENT);
442 }
443 
444 /*
445  * Add an entry to the cache.
446  */
447 void
448 cache_enter(dvp, vp, cnp)
449 	struct vnode *dvp;
450 	struct vnode *vp;
451 	struct componentname *cnp;
452 {
453 	struct namecache *ncp;
454 	struct nchashhead *ncpp;
455 	u_int32_t hash;
456 	int hold;
457 	int zap;
458 	int len;
459 
460 	if (!doingcache)
461 		return;
462 
463 	if (cnp->cn_nameptr[0] == '.') {
464 		if (cnp->cn_namelen == 1) {
465 			return;
466 		}
467 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
468 			if (vp) {
469 				dvp->v_dd = vp;
470 				dvp->v_ddid = vp->v_id;
471 			} else {
472 				dvp->v_dd = dvp;
473 				dvp->v_ddid = 0;
474 			}
475 			return;
476 		}
477 	}
478 
479 	hold = 0;
480 	zap = 0;
481 	ncp = cache_alloc(cnp->cn_namelen);
482 	CACHE_LOCK();
483 	numcache++;
484 	if (!vp) {
485 		numneg++;
486 		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
487 	} else if (vp->v_type == VDIR) {
488 		vp->v_dd = dvp;
489 		vp->v_ddid = dvp->v_id;
490 	}
491 
492 	/*
493 	 * Set the rest of the namecache entry elements, calculate it's
494 	 * hash key and insert it into the appropriate chain within
495 	 * the cache entries table.
496 	 */
497 	ncp->nc_vp = vp;
498 	ncp->nc_dvp = dvp;
499 	len = ncp->nc_nlen = cnp->cn_namelen;
500 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
501 	bcopy(cnp->cn_nameptr, ncp->nc_name, len);
502 	hash = fnv_32_buf(&dvp->v_id, sizeof(dvp->v_id), hash);
503 	ncpp = NCHHASH(hash);
504 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
505 	if (LIST_EMPTY(&dvp->v_cache_src)) {
506 		hold = 1;
507 		numcachehv++;
508 	}
509 	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
510 	/*
511 	 * If the entry is "negative", we place it into the
512 	 * "negative" cache queue, otherwise, we place it into the
513 	 * destination vnode's cache entries queue.
514 	 */
515 	if (vp) {
516 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
517 	} else {
518 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
519 	}
520 	if (numneg * ncnegfactor > numcache) {
521 		ncp = TAILQ_FIRST(&ncneg);
522 		zap = 1;
523 	}
524 	CACHE_UNLOCK();
525 	if (hold)
526 		vhold(dvp);
527 	if (zap)
528 		cache_zap(ncp, 0);
529 }
530 
531 /*
532  * Name cache initialization, from vfs_init() when we are booting
533  */
534 static void
535 nchinit(void *dummy __unused)
536 {
537 
538 	TAILQ_INIT(&ncneg);
539 
540 	cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
541 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
542 	cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
543 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
544 
545 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
546 }
547 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL)
548 
549 
550 /*
551  * Invalidate all entries to a particular vnode.
552  *
553  * Remove all entries in the namecache relating to this vnode and
554  * change the v_id.  We take the v_id from a global counter, since
555  * it becomes a handy sequence number in crash-dumps that way.
556  * No valid vnode will ever have (v_id == 0).
557  *
558  * XXX: Only time and the size of v_id prevents this from failing:
559  * XXX: In theory we should hunt down all (struct vnode*, v_id)
560  * XXX: soft references and nuke them, at least on the global
561  * XXX: v_id wraparound.  The period of resistance can be extended
562  * XXX: by incrementing each vnodes v_id individually instead of
563  * XXX: using the global v_id.
564  */
565 
566 /*
567  * XXX This is sometimes called when a vnode may still be re-used, in which
568  * case v_dd may be invalid.  Need to look this up.
569  */
570 void
571 cache_purge(vp)
572 	struct vnode *vp;
573 {
574 	static u_long nextid;
575 
576 	CACHE_LOCK();
577 	while (!LIST_EMPTY(&vp->v_cache_src))
578 		cache_zap(LIST_FIRST(&vp->v_cache_src), 1);
579 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
580 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst), 1);
581 
582 	do
583 		nextid++;
584 	while (nextid == vp->v_id || !nextid);
585 	vp->v_id = nextid;
586 	vp->v_dd = vp;
587 	vp->v_ddid = 0;
588 	CACHE_UNLOCK();
589 }
590 
591 /*
592  * Flush all entries referencing a particular filesystem.
593  *
594  * Since we need to check it anyway, we will flush all the invalid
595  * entries at the same time.
596  */
597 void
598 cache_purgevfs(mp)
599 	struct mount *mp;
600 {
601 	struct nchashhead *ncpp;
602 	struct namecache *ncp, *nnp;
603 	struct nchashhead mplist;
604 
605 	LIST_INIT(&mplist);
606 	ncp = NULL;
607 
608 	/* Scan hash tables for applicable entries */
609 	CACHE_LOCK();
610 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
611 		for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
612 			nnp = LIST_NEXT(ncp, nc_hash);
613 			if (ncp->nc_dvp->v_mount == mp) {
614 				LIST_REMOVE(ncp, nc_hash);
615 				LIST_INSERT_HEAD(&mplist, ncp, nc_hash);
616 			}
617 		}
618 	}
619 	CACHE_UNLOCK();
620 	while (!LIST_EMPTY(&mplist))
621 		cache_zap(LIST_FIRST(&mplist), 0);
622 }
623 
624 /*
625  * Perform canonical checks and cache lookup and pass on to filesystem
626  * through the vop_cachedlookup only if needed.
627  */
628 
629 int
630 vfs_cache_lookup(ap)
631 	struct vop_lookup_args /* {
632 		struct vnode *a_dvp;
633 		struct vnode **a_vpp;
634 		struct componentname *a_cnp;
635 	} */ *ap;
636 {
637 	struct vnode *dvp, *vp;
638 	int lockparent;
639 	int error;
640 	struct vnode **vpp = ap->a_vpp;
641 	struct componentname *cnp = ap->a_cnp;
642 	struct ucred *cred = cnp->cn_cred;
643 	int flags = cnp->cn_flags;
644 	struct thread *td = cnp->cn_thread;
645 	u_long vpid;	/* capability number of vnode */
646 
647 	*vpp = NULL;
648 	dvp = ap->a_dvp;
649 	lockparent = flags & LOCKPARENT;
650 
651 	if (dvp->v_type != VDIR)
652 		return (ENOTDIR);
653 
654 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
655 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
656 		return (EROFS);
657 
658 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
659 
660 	if (error)
661 		return (error);
662 
663 	error = cache_lookup(dvp, vpp, cnp);
664 
665 #ifdef LOOKUP_SHARED
666 	if (!error) {
667 		/* We do this because the rest of the system now expects to get
668 		 * a shared lock, which is later upgraded if LOCKSHARED is not
669 		 * set.  We have so many cases here because of bugs that yield
670 		 * inconsistant lock states.  This all badly needs to be fixed
671 		 */
672 		error = VOP_CACHEDLOOKUP(dvp, vpp, cnp);
673 		if (!error) {
674 			int flock;
675 
676 			flock = VOP_ISLOCKED(*vpp, td);
677 			if (flock != LK_EXCLUSIVE) {
678 				if (flock == 0) {
679 					if ((flags & ISLASTCN) &&
680 					    (flags & LOCKSHARED))
681 						VOP_LOCK(*vpp, LK_SHARED, td);
682 					else
683 						VOP_LOCK(*vpp, LK_EXCLUSIVE, td);
684 				}
685 			} else if ((flags & ISLASTCN) && (flags & LOCKSHARED))
686 				VOP_LOCK(*vpp, LK_DOWNGRADE, td);
687 		}
688 		return (error);
689 	}
690 #else
691 	if (!error)
692 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
693 #endif
694 
695 	if (error == ENOENT)
696 		return (error);
697 
698 	vp = *vpp;
699 	vpid = vp->v_id;
700 	cnp->cn_flags &= ~PDIRUNLOCK;
701 	if (dvp == vp) {   /* lookup on "." */
702 		VREF(vp);
703 		error = 0;
704 	} else if (flags & ISDOTDOT) {
705 		VOP_UNLOCK(dvp, 0, td);
706 		cnp->cn_flags |= PDIRUNLOCK;
707 #ifdef LOOKUP_SHARED
708 		if ((flags & ISLASTCN) && (flags & LOCKSHARED))
709 			error = vget(vp, LK_SHARED, td);
710 		else
711 			error = vget(vp, LK_EXCLUSIVE, td);
712 #else
713 		error = vget(vp, LK_EXCLUSIVE, td);
714 #endif
715 
716 		if (!error && lockparent && (flags & ISLASTCN)) {
717 			if ((error = vn_lock(dvp, LK_EXCLUSIVE, td)) == 0)
718 				cnp->cn_flags &= ~PDIRUNLOCK;
719 		}
720 	} else {
721 #ifdef LOOKUP_SHARED
722 		if ((flags & ISLASTCN) && (flags & LOCKSHARED))
723 			error = vget(vp, LK_SHARED, td);
724 		else
725 			error = vget(vp, LK_EXCLUSIVE, td);
726 #else
727 		error = vget(vp, LK_EXCLUSIVE, td);
728 #endif
729 		if (!lockparent || error || !(flags & ISLASTCN)) {
730 			VOP_UNLOCK(dvp, 0, td);
731 			cnp->cn_flags |= PDIRUNLOCK;
732 		}
733 	}
734 	/*
735 	 * Check that the capability number did not change
736 	 * while we were waiting for the lock.
737 	 */
738 	if (!error) {
739 		if (vpid == vp->v_id)
740 			return (0);
741 		vput(vp);
742 		if (lockparent && dvp != vp && (flags & ISLASTCN)) {
743 			VOP_UNLOCK(dvp, 0, td);
744 			cnp->cn_flags |= PDIRUNLOCK;
745 		}
746 	}
747 	if (cnp->cn_flags & PDIRUNLOCK) {
748 		error = vn_lock(dvp, LK_EXCLUSIVE, td);
749 		if (error)
750 			return (error);
751 		cnp->cn_flags &= ~PDIRUNLOCK;
752 	}
753 #ifdef LOOKUP_SHARED
754 	error = VOP_CACHEDLOOKUP(dvp, vpp, cnp);
755 
756 	if (!error) {
757 		int flock = 0;
758 
759 		flock = VOP_ISLOCKED(*vpp, td);
760 		if (flock != LK_EXCLUSIVE) {
761 			if (flock == 0) {
762 				if ((flags & ISLASTCN) && (flags & LOCKSHARED))
763 					VOP_LOCK(*vpp, LK_SHARED, td);
764 				else
765 					VOP_LOCK(*vpp, LK_EXCLUSIVE, td);
766 			}
767 		} else if ((flags & ISLASTCN) && (flags & LOCKSHARED))
768 			VOP_LOCK(*vpp, LK_DOWNGRADE, td);
769 	}
770 
771 	return (error);
772 #else
773 	return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
774 #endif
775 }
776 
777 
778 #ifndef _SYS_SYSPROTO_H_
779 struct  __getcwd_args {
780 	u_char	*buf;
781 	u_int	buflen;
782 };
783 #endif
784 
785 /*
786  * XXX All of these sysctls would probably be more productive dead.
787  */
788 static int disablecwd;
789 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
790    "Disable the getcwd syscall");
791 
792 /* Various statistics for the getcwd syscall */
793 static u_long numcwdcalls; STATNODE(CTLFLAG_RD, numcwdcalls, &numcwdcalls);
794 static u_long numcwdfail1; STATNODE(CTLFLAG_RD, numcwdfail1, &numcwdfail1);
795 static u_long numcwdfail2; STATNODE(CTLFLAG_RD, numcwdfail2, &numcwdfail2);
796 static u_long numcwdfail3; STATNODE(CTLFLAG_RD, numcwdfail3, &numcwdfail3);
797 static u_long numcwdfail4; STATNODE(CTLFLAG_RD, numcwdfail4, &numcwdfail4);
798 static u_long numcwdfound; STATNODE(CTLFLAG_RD, numcwdfound, &numcwdfound);
799 
800 /* Implementation of the getcwd syscall */
801 int
802 __getcwd(td, uap)
803 	struct thread *td;
804 	struct __getcwd_args *uap;
805 {
806 
807 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
808 }
809 
810 int
811 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
812 {
813 	char *bp, *tmpbuf;
814 	int error, i, slash_prefixed;
815 	struct filedesc *fdp;
816 	struct namecache *ncp;
817 	struct vnode *vp;
818 
819 	numcwdcalls++;
820 	if (disablecwd)
821 		return (ENODEV);
822 	if (buflen < 2)
823 		return (EINVAL);
824 	if (buflen > MAXPATHLEN)
825 		buflen = MAXPATHLEN;
826 	error = 0;
827 	tmpbuf = bp = malloc(buflen, M_TEMP, M_WAITOK);
828 	bp += buflen - 1;
829 	*bp = '\0';
830 	fdp = td->td_proc->p_fd;
831 	slash_prefixed = 0;
832 	FILEDESC_LOCK(fdp);
833 	for (vp = fdp->fd_cdir; vp != fdp->fd_rdir && vp != rootvnode;) {
834 		if (vp->v_vflag & VV_ROOT) {
835 			if (vp->v_mount == NULL) {	/* forced unmount */
836 				FILEDESC_UNLOCK(fdp);
837 				free(tmpbuf, M_TEMP);
838 				return (EBADF);
839 			}
840 			vp = vp->v_mount->mnt_vnodecovered;
841 			continue;
842 		}
843 		if (vp->v_dd->v_id != vp->v_ddid) {
844 			FILEDESC_UNLOCK(fdp);
845 			numcwdfail1++;
846 			free(tmpbuf, M_TEMP);
847 			return (ENOTDIR);
848 		}
849 		CACHE_LOCK();
850 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
851 		if (!ncp) {
852 			numcwdfail2++;
853 			CACHE_UNLOCK();
854 			FILEDESC_UNLOCK(fdp);
855 			free(tmpbuf, M_TEMP);
856 			return (ENOENT);
857 		}
858 		if (ncp->nc_dvp != vp->v_dd) {
859 			numcwdfail3++;
860 			CACHE_UNLOCK();
861 			FILEDESC_UNLOCK(fdp);
862 			free(tmpbuf, M_TEMP);
863 			return (EBADF);
864 		}
865 		for (i = ncp->nc_nlen - 1; i >= 0; i--) {
866 			if (bp == tmpbuf) {
867 				numcwdfail4++;
868 				CACHE_UNLOCK();
869 				FILEDESC_UNLOCK(fdp);
870 				free(tmpbuf, M_TEMP);
871 				return (ENOMEM);
872 			}
873 			*--bp = ncp->nc_name[i];
874 		}
875 		if (bp == tmpbuf) {
876 			numcwdfail4++;
877 			CACHE_UNLOCK();
878 			FILEDESC_UNLOCK(fdp);
879 			free(tmpbuf, M_TEMP);
880 			return (ENOMEM);
881 		}
882 		*--bp = '/';
883 		slash_prefixed = 1;
884 		vp = vp->v_dd;
885 		CACHE_UNLOCK();
886 	}
887 	FILEDESC_UNLOCK(fdp);
888 	if (!slash_prefixed) {
889 		if (bp == tmpbuf) {
890 			numcwdfail4++;
891 			free(tmpbuf, M_TEMP);
892 			return (ENOMEM);
893 		}
894 		*--bp = '/';
895 	}
896 	numcwdfound++;
897 	if (bufseg == UIO_SYSSPACE)
898 		bcopy(bp, buf, strlen(bp) + 1);
899 	else
900 		error = copyout(bp, buf, strlen(bp) + 1);
901 	free(tmpbuf, M_TEMP);
902 	return (error);
903 }
904 
905 /*
906  * Thus begins the fullpath magic.
907  */
908 
909 #undef STATNODE
910 #define STATNODE(name)							\
911 	static u_int name;						\
912 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
913 
914 static int disablefullpath;
915 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
916 	"Disable the vn_fullpath function");
917 
918 STATNODE(numfullpathcalls);
919 STATNODE(numfullpathfail1);
920 STATNODE(numfullpathfail2);
921 STATNODE(numfullpathfail3);
922 STATNODE(numfullpathfail4);
923 STATNODE(numfullpathfound);
924 
925 /*
926  * Retrieve the full filesystem path that correspond to a vnode from the name
927  * cache (if available)
928  */
929 int
930 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
931 {
932 	char *bp, *buf;
933 	int i, slash_prefixed;
934 	struct filedesc *fdp;
935 	struct namecache *ncp;
936 	struct vnode *vp;
937 
938 	numfullpathcalls++;
939 	if (disablefullpath)
940 		return (ENODEV);
941 	if (vn == NULL)
942 		return (EINVAL);
943 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
944 	bp = buf + MAXPATHLEN - 1;
945 	*bp = '\0';
946 	fdp = td->td_proc->p_fd;
947 	slash_prefixed = 0;
948 	ASSERT_VOP_LOCKED(vn, "vn_fullpath");
949 	FILEDESC_LOCK(fdp);
950 	for (vp = vn; vp != fdp->fd_rdir && vp != rootvnode;) {
951 		if (vp->v_vflag & VV_ROOT) {
952 			if (vp->v_mount == NULL) {	/* forced unmount */
953 				FILEDESC_UNLOCK(fdp);
954 				free(buf, M_TEMP);
955 				return (EBADF);
956 			}
957 			vp = vp->v_mount->mnt_vnodecovered;
958 			continue;
959 		}
960 		if (vp != vn && vp->v_dd->v_id != vp->v_ddid) {
961 			FILEDESC_UNLOCK(fdp);
962 			free(buf, M_TEMP);
963 			numfullpathfail1++;
964 			return (ENOTDIR);
965 		}
966 		CACHE_LOCK();
967 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
968 		if (!ncp) {
969 			numfullpathfail2++;
970 			CACHE_UNLOCK();
971 			FILEDESC_UNLOCK(fdp);
972 			free(buf, M_TEMP);
973 			return (ENOENT);
974 		}
975 		if (vp != vn && ncp->nc_dvp != vp->v_dd) {
976 			numfullpathfail3++;
977 			CACHE_UNLOCK();
978 			FILEDESC_UNLOCK(fdp);
979 			free(buf, M_TEMP);
980 			return (EBADF);
981 		}
982 		for (i = ncp->nc_nlen - 1; i >= 0; i--) {
983 			if (bp == buf) {
984 				numfullpathfail4++;
985 				CACHE_UNLOCK();
986 				FILEDESC_UNLOCK(fdp);
987 				free(buf, M_TEMP);
988 				return (ENOMEM);
989 			}
990 			*--bp = ncp->nc_name[i];
991 		}
992 		if (bp == buf) {
993 			numfullpathfail4++;
994 			CACHE_UNLOCK();
995 			FILEDESC_UNLOCK(fdp);
996 			free(buf, M_TEMP);
997 			return (ENOMEM);
998 		}
999 		*--bp = '/';
1000 		slash_prefixed = 1;
1001 		vp = ncp->nc_dvp;
1002 		CACHE_UNLOCK();
1003 	}
1004 	if (!slash_prefixed) {
1005 		if (bp == buf) {
1006 			numfullpathfail4++;
1007 			FILEDESC_UNLOCK(fdp);
1008 			free(buf, M_TEMP);
1009 			return (ENOMEM);
1010 		}
1011 		*--bp = '/';
1012 	}
1013 	FILEDESC_UNLOCK(fdp);
1014 	numfullpathfound++;
1015 	*retbuf = bp;
1016 	*freebuf = buf;
1017 	return (0);
1018 }
1019