xref: /freebsd/sys/kern/vfs_cache.c (revision f4f6abcb4ea1f4c91714febe157d0eaaf3e236ec)
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/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.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 #include <sys/syscallsubr.h>
49 #include <sys/sysproto.h>
50 #include <sys/proc.h>
51 #include <sys/filedesc.h>
52 #include <sys/fnv_hash.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, OID_AUTO, vnsize, CTLFLAG_RD, 0, sizeof(struct vnode), "");
145 SYSCTL_INT(_debug, OID_AUTO, ncsize, CTLFLAG_RD, 0, 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 		count = 0;
203 		LIST_FOREACH(ncp, ncpp, nc_hash) {
204 			count++;
205 		}
206 		error = SYSCTL_OUT(req, &count, sizeof(count));
207 		if (error)
208 			return (error);
209 	}
210 	return (0);
211 }
212 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD,
213 	0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths");
214 
215 static int
216 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
217 {
218 	int error;
219 	struct nchashhead *ncpp;
220 	struct namecache *ncp;
221 	int n_nchash;
222 	int count, maxlength, used, pct;
223 
224 	if (!req->oldptr)
225 		return SYSCTL_OUT(req, 0, 4 * sizeof(int));
226 
227 	n_nchash = nchash + 1;	/* nchash is max index, not count */
228 	used = 0;
229 	maxlength = 0;
230 
231 	/* Scan hash tables for applicable entries */
232 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
233 		count = 0;
234 		LIST_FOREACH(ncp, ncpp, nc_hash) {
235 			count++;
236 		}
237 		if (count)
238 			used++;
239 		if (maxlength < count)
240 			maxlength = count;
241 	}
242 	n_nchash = nchash + 1;
243 	pct = (used * 100 * 100) / n_nchash;
244 	error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
245 	if (error)
246 		return (error);
247 	error = SYSCTL_OUT(req, &used, sizeof(used));
248 	if (error)
249 		return (error);
250 	error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
251 	if (error)
252 		return (error);
253 	error = SYSCTL_OUT(req, &pct, sizeof(pct));
254 	if (error)
255 		return (error);
256 	return (0);
257 }
258 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD,
259 	0, 0, sysctl_debug_hashstat_nchash, "I", "nchash chain lengths");
260 
261 /*
262  * cache_zap():
263  *
264  *   Removes a namecache entry from cache, whether it contains an actual
265  *   pointer to a vnode or if it is just a negative cache entry.
266  */
267 static void
268 cache_zap(ncp)
269 	struct namecache *ncp;
270 {
271 	struct vnode *vp;
272 
273 	mtx_assert(&cache_lock, MA_OWNED);
274 	vp = NULL;
275 	LIST_REMOVE(ncp, nc_hash);
276 	LIST_REMOVE(ncp, nc_src);
277 	if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
278 		vp = ncp->nc_dvp;
279 		numcachehv--;
280 	}
281 	if (ncp->nc_vp) {
282 		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
283 	} else {
284 		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
285 		numneg--;
286 	}
287 	numcache--;
288 	cache_free(ncp);
289 	if (vp)
290 		vdrop(vp);
291 }
292 
293 /*
294  * cache_leaf_test()
295  *
296  *      Test whether this (directory) vnode's namei cache entry contains
297  *      subdirectories or not.  Used to determine whether the directory is
298  *      a leaf in the namei cache or not.  Note: the directory may still
299  *      contain files in the namei cache.
300  *
301  *      Returns 0 if the directory is a leaf, -1 if it isn't.
302  */
303 int
304 cache_leaf_test(struct vnode *vp)
305 {
306 	struct namecache *ncpc;
307 	int leaf;
308 
309 	leaf = 0;
310 	CACHE_LOCK();
311 	for (ncpc = LIST_FIRST(&vp->v_cache_src);
312 	     ncpc != NULL;
313 	     ncpc = LIST_NEXT(ncpc, nc_src)
314 	 ) {
315 		if (ncpc->nc_vp != NULL && ncpc->nc_vp->v_type == VDIR) {
316 			leaf = -1;
317 			break;
318 		}
319 	}
320 	CACHE_UNLOCK();
321 	return (leaf);
322 }
323 
324 /*
325  * Lookup an entry in the cache
326  *
327  * Lookup is called with dvp pointing to the directory to search,
328  * cnp pointing to the name of the entry being sought. If the lookup
329  * succeeds, the vnode is returned in *vpp, and a status of -1 is
330  * returned. If the lookup determines that the name does not exist
331  * (negative cacheing), a status of ENOENT is returned. If the lookup
332  * fails, a status of zero is returned.
333  *
334  * vpp is locked and ref'd on return.  If we're looking up DOTDOT, dvp is
335  * unlocked.  If we're looking up . an extra ref is taken, but the lock is
336  * not recursively acquired.
337  */
338 
339 int
340 cache_lookup(dvp, vpp, cnp)
341 	struct vnode *dvp;
342 	struct vnode **vpp;
343 	struct componentname *cnp;
344 {
345 	struct namecache *ncp;
346 	u_int32_t hash;
347 
348 	if (!doingcache) {
349 		cnp->cn_flags &= ~MAKEENTRY;
350 		return (0);
351 	}
352 retry:
353 	CACHE_LOCK();
354 	numcalls++;
355 
356 	if (cnp->cn_nameptr[0] == '.') {
357 		if (cnp->cn_namelen == 1) {
358 			*vpp = dvp;
359 			dothits++;
360 			goto success;
361 		}
362 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
363 			dotdothits++;
364 			if (dvp->v_dd == NULL ||
365 			    (cnp->cn_flags & MAKEENTRY) == 0) {
366 				CACHE_UNLOCK();
367 				return (0);
368 			}
369 			*vpp = dvp->v_dd;
370 			goto success;
371 		}
372 	}
373 
374 	hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
375 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
376 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
377 		numchecks++;
378 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
379 		    !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
380 			break;
381 	}
382 
383 	/* We failed to find an entry */
384 	if (ncp == 0) {
385 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
386 			nummisszap++;
387 		} else {
388 			nummiss++;
389 		}
390 		nchstats.ncs_miss++;
391 		CACHE_UNLOCK();
392 		return (0);
393 	}
394 
395 	/* We don't want to have an entry, so dump it */
396 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
397 		numposzaps++;
398 		nchstats.ncs_badhits++;
399 		cache_zap(ncp);
400 		CACHE_UNLOCK();
401 		return (0);
402 	}
403 
404 	/* We found a "positive" match, return the vnode */
405 	if (ncp->nc_vp) {
406 		numposhits++;
407 		nchstats.ncs_goodhits++;
408 		*vpp = ncp->nc_vp;
409 		goto success;
410 	}
411 
412 	/* We found a negative match, and want to create it, so purge */
413 	if (cnp->cn_nameiop == CREATE) {
414 		numnegzaps++;
415 		nchstats.ncs_badhits++;
416 		cache_zap(ncp);
417 		CACHE_UNLOCK();
418 		return (0);
419 	}
420 
421 	numneghits++;
422 	/*
423 	 * We found a "negative" match, so we shift it to the end of
424 	 * the "negative" cache entries queue to satisfy LRU.  Also,
425 	 * check to see if the entry is a whiteout; indicate this to
426 	 * the componentname, if so.
427 	 */
428 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
429 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
430 	nchstats.ncs_neghits++;
431 	if (ncp->nc_flag & NCF_WHITE)
432 		cnp->cn_flags |= ISWHITEOUT;
433 	CACHE_UNLOCK();
434 	return (ENOENT);
435 
436 success:
437 	/*
438 	 * On success we return a locked and ref'd vnode as per the lookup
439 	 * protocol.
440 	 */
441 	if (dvp == *vpp) {   /* lookup on "." */
442 		VREF(*vpp);
443 		CACHE_UNLOCK();
444 		return (-1);
445 	}
446 	if (cnp->cn_flags & ISDOTDOT)
447 		VOP_UNLOCK(dvp, 0, cnp->cn_thread);
448 	VI_LOCK(*vpp);
449 	CACHE_UNLOCK();
450 	if (vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread)) {
451 		if (cnp->cn_flags & ISDOTDOT)
452 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, cnp->cn_thread);
453 		*vpp = NULL;
454 		goto retry;
455 	}
456 	return (-1);
457 }
458 
459 /*
460  * Add an entry to the cache.
461  */
462 void
463 cache_enter(dvp, vp, cnp)
464 	struct vnode *dvp;
465 	struct vnode *vp;
466 	struct componentname *cnp;
467 {
468 	struct namecache *ncp;
469 	struct nchashhead *ncpp;
470 	u_int32_t hash;
471 	int hold;
472 	int zap;
473 	int len;
474 
475 	if (!doingcache)
476 		return;
477 
478 	if (cnp->cn_nameptr[0] == '.') {
479 		if (cnp->cn_namelen == 1) {
480 			return;
481 		}
482 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
483 			dvp->v_dd = vp;
484 			return;
485 		}
486 	}
487 
488 	hold = 0;
489 	zap = 0;
490 	ncp = cache_alloc(cnp->cn_namelen);
491 	CACHE_LOCK();
492 	numcache++;
493 	if (!vp) {
494 		numneg++;
495 		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
496 	} else if (vp->v_type == VDIR) {
497 		vp->v_dd = dvp;
498 	} else {
499 		vp->v_dd = NULL;
500 	}
501 
502 	/*
503 	 * Set the rest of the namecache entry elements, calculate it's
504 	 * hash key and insert it into the appropriate chain within
505 	 * the cache entries table.
506 	 */
507 	ncp->nc_vp = vp;
508 	ncp->nc_dvp = dvp;
509 	len = ncp->nc_nlen = cnp->cn_namelen;
510 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
511 	bcopy(cnp->cn_nameptr, ncp->nc_name, len);
512 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
513 	ncpp = NCHHASH(hash);
514 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
515 	if (LIST_EMPTY(&dvp->v_cache_src)) {
516 		hold = 1;
517 		numcachehv++;
518 	}
519 	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
520 	/*
521 	 * If the entry is "negative", we place it into the
522 	 * "negative" cache queue, otherwise, we place it into the
523 	 * destination vnode's cache entries queue.
524 	 */
525 	if (vp) {
526 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
527 	} else {
528 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
529 	}
530 	if (numneg * ncnegfactor > numcache) {
531 		ncp = TAILQ_FIRST(&ncneg);
532 		zap = 1;
533 	}
534 	if (hold)
535 		vhold(dvp);
536 	if (zap)
537 		cache_zap(ncp);
538 	CACHE_UNLOCK();
539 }
540 
541 /*
542  * Name cache initialization, from vfs_init() when we are booting
543  */
544 static void
545 nchinit(void *dummy __unused)
546 {
547 
548 	TAILQ_INIT(&ncneg);
549 
550 	cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
551 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
552 	cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
553 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
554 
555 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
556 }
557 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL)
558 
559 
560 /*
561  * Invalidate all entries to a particular vnode.
562  */
563 void
564 cache_purge(vp)
565 	struct vnode *vp;
566 {
567 	struct namecache *ncp;
568 
569 	CACHE_LOCK();
570 	while (!LIST_EMPTY(&vp->v_cache_src)) {
571 		struct vnode *cvp;
572 
573 		ncp = LIST_FIRST(&vp->v_cache_src);
574 		/*
575 		 * We must reset v_dd of any children so they don't
576 		 * continue to point to us.
577 		 */
578 		if ((cvp = ncp->nc_vp) && cvp->v_dd == vp)
579 			cvp->v_dd = NULL;
580 		cache_zap(ncp);
581 	}
582 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
583 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
584 	vp->v_dd = NULL;
585 	CACHE_UNLOCK();
586 }
587 
588 /*
589  * Flush all entries referencing a particular filesystem.
590  *
591  * Since we need to check it anyway, we will flush all the invalid
592  * entries at the same time.
593  */
594 void
595 cache_purgevfs(mp)
596 	struct mount *mp;
597 {
598 	struct nchashhead *ncpp;
599 	struct namecache *ncp, *nnp;
600 	struct nchashhead mplist;
601 
602 	LIST_INIT(&mplist);
603 	ncp = NULL;
604 
605 	/* Scan hash tables for applicable entries */
606 	CACHE_LOCK();
607 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
608 		for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
609 			nnp = LIST_NEXT(ncp, nc_hash);
610 			if (ncp->nc_dvp->v_mount == mp) {
611 				LIST_REMOVE(ncp, nc_hash);
612 				LIST_INSERT_HEAD(&mplist, ncp, nc_hash);
613 			}
614 		}
615 	}
616 	while (!LIST_EMPTY(&mplist))
617 		cache_zap(LIST_FIRST(&mplist));
618 	CACHE_UNLOCK();
619 }
620 
621 /*
622  * Perform canonical checks and cache lookup and pass on to filesystem
623  * through the vop_cachedlookup only if needed.
624  */
625 
626 int
627 vfs_cache_lookup(ap)
628 	struct vop_lookup_args /* {
629 		struct vnode *a_dvp;
630 		struct vnode **a_vpp;
631 		struct componentname *a_cnp;
632 	} */ *ap;
633 {
634 	struct vnode *dvp;
635 	int error;
636 	struct vnode **vpp = ap->a_vpp;
637 	struct componentname *cnp = ap->a_cnp;
638 	struct ucred *cred = cnp->cn_cred;
639 	int flags = cnp->cn_flags;
640 	struct thread *td = cnp->cn_thread;
641 
642 	*vpp = NULL;
643 	dvp = ap->a_dvp;
644 
645 	if (dvp->v_type != VDIR)
646 		return (ENOTDIR);
647 
648 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
649 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
650 		return (EROFS);
651 
652 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
653 	if (error)
654 		return (error);
655 
656 	error = cache_lookup(dvp, vpp, cnp);
657 	if (error == 0)
658 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
659 	if (error == ENOENT)
660 		return (error);
661 	return (0);
662 }
663 
664 
665 #ifndef _SYS_SYSPROTO_H_
666 struct  __getcwd_args {
667 	u_char	*buf;
668 	u_int	buflen;
669 };
670 #endif
671 
672 /*
673  * XXX All of these sysctls would probably be more productive dead.
674  */
675 static int disablecwd;
676 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
677    "Disable the getcwd syscall");
678 
679 /* Implementation of the getcwd syscall */
680 int
681 __getcwd(td, uap)
682 	struct thread *td;
683 	struct __getcwd_args *uap;
684 {
685 
686 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
687 }
688 
689 int
690 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
691 {
692 	char *bp, *tmpbuf;
693 	struct filedesc *fdp;
694 	int error;
695 
696 	if (disablecwd)
697 		return (ENODEV);
698 	if (buflen < 2)
699 		return (EINVAL);
700 	if (buflen > MAXPATHLEN)
701 		buflen = MAXPATHLEN;
702 
703 	tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
704 	fdp = td->td_proc->p_fd;
705 	mtx_lock(&Giant);
706 	FILEDESC_LOCK(fdp);
707 	error = vn_fullpath1(td, fdp->fd_cdir, fdp->fd_rdir, tmpbuf,
708 	    &bp, buflen);
709 	FILEDESC_UNLOCK(fdp);
710 	mtx_unlock(&Giant);
711 
712 	if (!error) {
713 		if (bufseg == UIO_SYSSPACE)
714 			bcopy(bp, buf, strlen(bp) + 1);
715 		else
716 			error = copyout(bp, buf, strlen(bp) + 1);
717 	}
718 	free(tmpbuf, M_TEMP);
719 	return (error);
720 }
721 
722 /*
723  * Thus begins the fullpath magic.
724  */
725 
726 #undef STATNODE
727 #define STATNODE(name)							\
728 	static u_int name;						\
729 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
730 
731 static int disablefullpath;
732 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
733 	"Disable the vn_fullpath function");
734 
735 /* These count for kern___getcwd(), too. */
736 STATNODE(numfullpathcalls);
737 STATNODE(numfullpathfail1);
738 STATNODE(numfullpathfail2);
739 STATNODE(numfullpathfail4);
740 STATNODE(numfullpathfound);
741 
742 /*
743  * Retrieve the full filesystem path that correspond to a vnode from the name
744  * cache (if available)
745  */
746 int
747 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
748 {
749 	char *buf;
750 	struct filedesc *fdp;
751 	int error;
752 
753 	if (disablefullpath)
754 		return (ENODEV);
755 	if (vn == NULL)
756 		return (EINVAL);
757 
758 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
759 	fdp = td->td_proc->p_fd;
760 	mtx_lock(&Giant);
761 	FILEDESC_LOCK(fdp);
762 	error = vn_fullpath1(td, vn, fdp->fd_rdir, buf, retbuf, MAXPATHLEN);
763 	FILEDESC_UNLOCK(fdp);
764 	mtx_unlock(&Giant);
765 
766 	if (!error)
767 		*freebuf = buf;
768 	else
769 		free(buf, M_TEMP);
770 	return (error);
771 }
772 
773 /*
774  * The magic behind kern___getcwd() and vn_fullpath().
775  */
776 static int
777 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
778     char *buf, char **retbuf, u_int buflen)
779 {
780 	char *bp;
781 	int error, i, slash_prefixed;
782 	struct namecache *ncp;
783 
784 	mtx_assert(&Giant, MA_OWNED);
785 
786 	bp = buf + buflen - 1;
787 	*bp = '\0';
788 	error = 0;
789 	slash_prefixed = 0;
790 
791 	CACHE_LOCK();
792 	numfullpathcalls++;
793 	if (vp->v_type != VDIR) {
794 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
795 		if (!ncp) {
796 			numfullpathfail2++;
797 			CACHE_UNLOCK();
798 			return (ENOENT);
799 		}
800 		for (i = ncp->nc_nlen - 1; i >= 0 && bp > buf; i--)
801 			*--bp = ncp->nc_name[i];
802 		if (bp == buf) {
803 			numfullpathfail4++;
804 			CACHE_UNLOCK();
805 			return (ENOMEM);
806 		}
807 		*--bp = '/';
808 		slash_prefixed = 1;
809 		vp = ncp->nc_dvp;
810 	}
811 	while (vp != rdir && vp != rootvnode) {
812 		if (vp->v_vflag & VV_ROOT) {
813 			if (vp->v_mount == NULL) {	/* forced unmount */
814 				error = EBADF;
815 				break;
816 			}
817 			vp = vp->v_mount->mnt_vnodecovered;
818 			continue;
819 		}
820 		if (vp->v_dd == NULL) {
821 			numfullpathfail1++;
822 			error = ENOTDIR;
823 			break;
824 		}
825 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
826 		if (!ncp) {
827 			numfullpathfail2++;
828 			error = ENOENT;
829 			break;
830 		}
831 		MPASS(ncp->nc_dvp == vp->v_dd);
832 		for (i = ncp->nc_nlen - 1; i >= 0 && bp != buf; i--)
833 			*--bp = ncp->nc_name[i];
834 		if (bp == buf) {
835 			numfullpathfail4++;
836 			error = ENOMEM;
837 			break;
838 		}
839 		*--bp = '/';
840 		slash_prefixed = 1;
841 		vp = ncp->nc_dvp;
842 	}
843 	if (error) {
844 		CACHE_UNLOCK();
845 		return (error);
846 	}
847 	if (!slash_prefixed) {
848 		if (bp == buf) {
849 			numfullpathfail4++;
850 			CACHE_UNLOCK();
851 			return (ENOMEM);
852 		} else {
853 			*--bp = '/';
854 		}
855 	}
856 	numfullpathfound++;
857 	CACHE_UNLOCK();
858 
859 	*retbuf = bp;
860 	return (0);
861 }
862