xref: /freebsd/sys/kern/vfs_cache.c (revision 10f0bcab61ef441cb5af32fb706688d8cbd55dc0)
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_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0,
145 	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 	CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
275 	vp = NULL;
276 	LIST_REMOVE(ncp, nc_hash);
277 	LIST_REMOVE(ncp, nc_src);
278 	if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
279 		vp = ncp->nc_dvp;
280 		numcachehv--;
281 	}
282 	if (ncp->nc_vp) {
283 		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
284 		ncp->nc_vp->v_dd = NULL;
285 	} else {
286 		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
287 		numneg--;
288 	}
289 	numcache--;
290 	cache_free(ncp);
291 	if (vp)
292 		vdrop(vp);
293 }
294 
295 /*
296  * Lookup an entry in the cache
297  *
298  * Lookup is called with dvp pointing to the directory to search,
299  * cnp pointing to the name of the entry being sought. If the lookup
300  * succeeds, the vnode is returned in *vpp, and a status of -1 is
301  * returned. If the lookup determines that the name does not exist
302  * (negative cacheing), a status of ENOENT is returned. If the lookup
303  * fails, a status of zero is returned.
304  *
305  * vpp is locked and ref'd on return.  If we're looking up DOTDOT, dvp is
306  * unlocked.  If we're looking up . an extra ref is taken, but the lock is
307  * not recursively acquired.
308  */
309 
310 int
311 cache_lookup(dvp, vpp, cnp)
312 	struct vnode *dvp;
313 	struct vnode **vpp;
314 	struct componentname *cnp;
315 {
316 	struct namecache *ncp;
317 	u_int32_t hash;
318 	int error, ltype;
319 
320 	if (!doingcache) {
321 		cnp->cn_flags &= ~MAKEENTRY;
322 		return (0);
323 	}
324 retry:
325 	CACHE_LOCK();
326 	numcalls++;
327 
328 	if (cnp->cn_nameptr[0] == '.') {
329 		if (cnp->cn_namelen == 1) {
330 			*vpp = dvp;
331 			CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
332 			    dvp, cnp->cn_nameptr);
333 			dothits++;
334 			goto success;
335 		}
336 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
337 			dotdothits++;
338 			if (dvp->v_dd == NULL ||
339 			    (cnp->cn_flags & MAKEENTRY) == 0) {
340 				CACHE_UNLOCK();
341 				return (0);
342 			}
343 			*vpp = dvp->v_dd;
344 			CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
345 			    dvp, cnp->cn_nameptr, *vpp);
346 			goto success;
347 		}
348 	}
349 
350 	hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
351 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
352 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
353 		numchecks++;
354 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
355 		    !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
356 			break;
357 	}
358 
359 	/* We failed to find an entry */
360 	if (ncp == 0) {
361 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
362 			nummisszap++;
363 		} else {
364 			nummiss++;
365 		}
366 		nchstats.ncs_miss++;
367 		CACHE_UNLOCK();
368 		return (0);
369 	}
370 
371 	/* We don't want to have an entry, so dump it */
372 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
373 		numposzaps++;
374 		nchstats.ncs_badhits++;
375 		cache_zap(ncp);
376 		CACHE_UNLOCK();
377 		return (0);
378 	}
379 
380 	/* We found a "positive" match, return the vnode */
381 	if (ncp->nc_vp) {
382 		numposhits++;
383 		nchstats.ncs_goodhits++;
384 		*vpp = ncp->nc_vp;
385 		CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
386 		    dvp, cnp->cn_nameptr, *vpp, ncp);
387 		goto success;
388 	}
389 
390 	/* We found a negative match, and want to create it, so purge */
391 	if (cnp->cn_nameiop == CREATE) {
392 		numnegzaps++;
393 		nchstats.ncs_badhits++;
394 		cache_zap(ncp);
395 		CACHE_UNLOCK();
396 		return (0);
397 	}
398 
399 	numneghits++;
400 	/*
401 	 * We found a "negative" match, so we shift it to the end of
402 	 * the "negative" cache entries queue to satisfy LRU.  Also,
403 	 * check to see if the entry is a whiteout; indicate this to
404 	 * the componentname, if so.
405 	 */
406 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
407 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
408 	nchstats.ncs_neghits++;
409 	if (ncp->nc_flag & NCF_WHITE)
410 		cnp->cn_flags |= ISWHITEOUT;
411 	CACHE_UNLOCK();
412 	return (ENOENT);
413 
414 success:
415 	/*
416 	 * On success we return a locked and ref'd vnode as per the lookup
417 	 * protocol.
418 	 */
419 	if (dvp == *vpp) {   /* lookup on "." */
420 		VREF(*vpp);
421 		CACHE_UNLOCK();
422 		/*
423 		 * When we lookup "." we still can be asked to lock it
424 		 * differently...
425 		 */
426 		ltype = cnp->cn_lkflags & (LK_SHARED | LK_EXCLUSIVE);
427 		if (ltype == VOP_ISLOCKED(*vpp))
428 			return (-1);
429 		else if (ltype == LK_EXCLUSIVE)
430 			vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
431 		return (-1);
432 	}
433 	ltype = 0;	/* silence gcc warning */
434 	if (cnp->cn_flags & ISDOTDOT) {
435 		ltype = VOP_ISLOCKED(dvp);
436 		VOP_UNLOCK(dvp, 0);
437 	}
438 	VI_LOCK(*vpp);
439 	CACHE_UNLOCK();
440 	error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
441 	if (cnp->cn_flags & ISDOTDOT)
442 		vn_lock(dvp, ltype | LK_RETRY);
443 	if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_lkflags & LK_EXCLUSIVE))
444 		ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
445 	if (error) {
446 		*vpp = NULL;
447 		goto retry;
448 	}
449 	return (-1);
450 }
451 
452 /*
453  * Add an entry to the cache.
454  */
455 void
456 cache_enter(dvp, vp, cnp)
457 	struct vnode *dvp;
458 	struct vnode *vp;
459 	struct componentname *cnp;
460 {
461 	struct namecache *ncp;
462 	struct nchashhead *ncpp;
463 	u_int32_t hash;
464 	int hold;
465 	int zap;
466 	int len;
467 
468 	CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
469 	VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
470 	    ("cahe_enter: Adding a doomed vnode"));
471 
472 	if (!doingcache)
473 		return;
474 
475 	if (cnp->cn_nameptr[0] == '.') {
476 		if (cnp->cn_namelen == 1) {
477 			return;
478 		}
479 		/*
480 		 * For dotdot lookups only cache the v_dd pointer if the
481 		 * directory has a link back to its parent via v_cache_dst.
482 		 * Without this an unlinked directory would keep a soft
483 		 * reference to its parent which could not be NULLd at
484 		 * cache_purge() time.
485 		 */
486 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
487 			CACHE_LOCK();
488 			if (!TAILQ_EMPTY(&dvp->v_cache_dst))
489 				dvp->v_dd = vp;
490 			CACHE_UNLOCK();
491 			return;
492 		}
493 	}
494 
495 	hold = 0;
496 	zap = 0;
497 	ncp = cache_alloc(cnp->cn_namelen);
498 	CACHE_LOCK();
499 	numcache++;
500 	if (!vp) {
501 		numneg++;
502 		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
503 	} else if (vp->v_type == VDIR) {
504 		vp->v_dd = dvp;
505 	} else {
506 		vp->v_dd = NULL;
507 	}
508 
509 	/*
510 	 * Set the rest of the namecache entry elements, calculate it's
511 	 * hash key and insert it into the appropriate chain within
512 	 * the cache entries table.
513 	 */
514 	ncp->nc_vp = vp;
515 	ncp->nc_dvp = dvp;
516 	len = ncp->nc_nlen = cnp->cn_namelen;
517 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
518 	bcopy(cnp->cn_nameptr, ncp->nc_name, len);
519 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
520 	ncpp = NCHHASH(hash);
521 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
522 	if (LIST_EMPTY(&dvp->v_cache_src)) {
523 		hold = 1;
524 		numcachehv++;
525 	}
526 	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
527 	/*
528 	 * If the entry is "negative", we place it into the
529 	 * "negative" cache queue, otherwise, we place it into the
530 	 * destination vnode's cache entries queue.
531 	 */
532 	if (vp) {
533 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
534 	} else {
535 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
536 	}
537 	if (numneg * ncnegfactor > numcache) {
538 		ncp = TAILQ_FIRST(&ncneg);
539 		zap = 1;
540 	}
541 	if (hold)
542 		vhold(dvp);
543 	if (zap)
544 		cache_zap(ncp);
545 	CACHE_UNLOCK();
546 }
547 
548 /*
549  * Name cache initialization, from vfs_init() when we are booting
550  */
551 static void
552 nchinit(void *dummy __unused)
553 {
554 
555 	TAILQ_INIT(&ncneg);
556 
557 	cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
558 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
559 	cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
560 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
561 
562 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
563 }
564 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
565 
566 
567 /*
568  * Invalidate all entries to a particular vnode.
569  */
570 void
571 cache_purge(vp)
572 	struct vnode *vp;
573 {
574 
575 	CTR1(KTR_VFS, "cache_purge(%p)", vp);
576 	CACHE_LOCK();
577 	while (!LIST_EMPTY(&vp->v_cache_src))
578 		cache_zap(LIST_FIRST(&vp->v_cache_src));
579 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
580 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
581 	vp->v_dd = NULL;
582 	CACHE_UNLOCK();
583 }
584 
585 /*
586  * Flush all entries referencing a particular filesystem.
587  */
588 void
589 cache_purgevfs(mp)
590 	struct mount *mp;
591 {
592 	struct nchashhead *ncpp;
593 	struct namecache *ncp, *nnp;
594 
595 	/* Scan hash tables for applicable entries */
596 	CACHE_LOCK();
597 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
598 		LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
599 			if (ncp->nc_dvp->v_mount == mp)
600 				cache_zap(ncp);
601 		}
602 	}
603 	CACHE_UNLOCK();
604 }
605 
606 /*
607  * Perform canonical checks and cache lookup and pass on to filesystem
608  * through the vop_cachedlookup only if needed.
609  */
610 
611 int
612 vfs_cache_lookup(ap)
613 	struct vop_lookup_args /* {
614 		struct vnode *a_dvp;
615 		struct vnode **a_vpp;
616 		struct componentname *a_cnp;
617 	} */ *ap;
618 {
619 	struct vnode *dvp;
620 	int error;
621 	struct vnode **vpp = ap->a_vpp;
622 	struct componentname *cnp = ap->a_cnp;
623 	struct ucred *cred = cnp->cn_cred;
624 	int flags = cnp->cn_flags;
625 	struct thread *td = cnp->cn_thread;
626 
627 	*vpp = NULL;
628 	dvp = ap->a_dvp;
629 
630 	if (dvp->v_type != VDIR)
631 		return (ENOTDIR);
632 
633 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
634 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
635 		return (EROFS);
636 
637 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
638 	if (error)
639 		return (error);
640 
641 	error = cache_lookup(dvp, vpp, cnp);
642 	if (error == 0)
643 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
644 	if (error == ENOENT)
645 		return (error);
646 	return (0);
647 }
648 
649 
650 #ifndef _SYS_SYSPROTO_H_
651 struct  __getcwd_args {
652 	u_char	*buf;
653 	u_int	buflen;
654 };
655 #endif
656 
657 /*
658  * XXX All of these sysctls would probably be more productive dead.
659  */
660 static int disablecwd;
661 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
662    "Disable the getcwd syscall");
663 
664 /* Implementation of the getcwd syscall. */
665 int
666 __getcwd(td, uap)
667 	struct thread *td;
668 	struct __getcwd_args *uap;
669 {
670 
671 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
672 }
673 
674 int
675 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
676 {
677 	char *bp, *tmpbuf;
678 	struct filedesc *fdp;
679 	int error;
680 
681 	if (disablecwd)
682 		return (ENODEV);
683 	if (buflen < 2)
684 		return (EINVAL);
685 	if (buflen > MAXPATHLEN)
686 		buflen = MAXPATHLEN;
687 
688 	tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
689 	fdp = td->td_proc->p_fd;
690 	FILEDESC_SLOCK(fdp);
691 	error = vn_fullpath1(td, fdp->fd_cdir, fdp->fd_rdir, tmpbuf,
692 	    &bp, buflen);
693 	FILEDESC_SUNLOCK(fdp);
694 
695 	if (!error) {
696 		if (bufseg == UIO_SYSSPACE)
697 			bcopy(bp, buf, strlen(bp) + 1);
698 		else
699 			error = copyout(bp, buf, strlen(bp) + 1);
700 	}
701 	free(tmpbuf, M_TEMP);
702 	return (error);
703 }
704 
705 /*
706  * Thus begins the fullpath magic.
707  */
708 
709 #undef STATNODE
710 #define STATNODE(name)							\
711 	static u_int name;						\
712 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
713 
714 static int disablefullpath;
715 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
716 	"Disable the vn_fullpath function");
717 
718 /* These count for kern___getcwd(), too. */
719 STATNODE(numfullpathcalls);
720 STATNODE(numfullpathfail1);
721 STATNODE(numfullpathfail2);
722 STATNODE(numfullpathfail4);
723 STATNODE(numfullpathfound);
724 
725 /*
726  * Retrieve the full filesystem path that correspond to a vnode from the name
727  * cache (if available)
728  */
729 int
730 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
731 {
732 	char *buf;
733 	struct filedesc *fdp;
734 	int error;
735 
736 	if (disablefullpath)
737 		return (ENODEV);
738 	if (vn == NULL)
739 		return (EINVAL);
740 
741 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
742 	fdp = td->td_proc->p_fd;
743 	FILEDESC_SLOCK(fdp);
744 	error = vn_fullpath1(td, vn, fdp->fd_rdir, buf, retbuf, MAXPATHLEN);
745 	FILEDESC_SUNLOCK(fdp);
746 
747 	if (!error)
748 		*freebuf = buf;
749 	else
750 		free(buf, M_TEMP);
751 	return (error);
752 }
753 
754 /*
755  * The magic behind kern___getcwd() and vn_fullpath().
756  */
757 static int
758 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
759     char *buf, char **retbuf, u_int buflen)
760 {
761 	char *bp;
762 	int error, i, slash_prefixed;
763 	struct namecache *ncp;
764 
765 	bp = buf + buflen - 1;
766 	*bp = '\0';
767 	error = 0;
768 	slash_prefixed = 0;
769 
770 	CACHE_LOCK();
771 	numfullpathcalls++;
772 	if (vp->v_type != VDIR) {
773 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
774 		if (!ncp) {
775 			numfullpathfail2++;
776 			CACHE_UNLOCK();
777 			return (ENOENT);
778 		}
779 		for (i = ncp->nc_nlen - 1; i >= 0 && bp > buf; i--)
780 			*--bp = ncp->nc_name[i];
781 		if (bp == buf) {
782 			numfullpathfail4++;
783 			CACHE_UNLOCK();
784 			return (ENOMEM);
785 		}
786 		*--bp = '/';
787 		slash_prefixed = 1;
788 		vp = ncp->nc_dvp;
789 	}
790 	while (vp != rdir && vp != rootvnode) {
791 		if (vp->v_vflag & VV_ROOT) {
792 			if (vp->v_iflag & VI_DOOMED) {	/* forced unmount */
793 				error = EBADF;
794 				break;
795 			}
796 			vp = vp->v_mount->mnt_vnodecovered;
797 			continue;
798 		}
799 		if (vp->v_dd == NULL) {
800 			numfullpathfail1++;
801 			error = ENOTDIR;
802 			break;
803 		}
804 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
805 		if (!ncp) {
806 			numfullpathfail2++;
807 			error = ENOENT;
808 			break;
809 		}
810 		MPASS(ncp->nc_dvp == vp->v_dd);
811 		for (i = ncp->nc_nlen - 1; i >= 0 && bp != buf; i--)
812 			*--bp = ncp->nc_name[i];
813 		if (bp == buf) {
814 			numfullpathfail4++;
815 			error = ENOMEM;
816 			break;
817 		}
818 		*--bp = '/';
819 		slash_prefixed = 1;
820 		vp = ncp->nc_dvp;
821 	}
822 	if (error) {
823 		CACHE_UNLOCK();
824 		return (error);
825 	}
826 	if (!slash_prefixed) {
827 		if (bp == buf) {
828 			numfullpathfail4++;
829 			CACHE_UNLOCK();
830 			return (ENOMEM);
831 		} else {
832 			*--bp = '/';
833 		}
834 	}
835 	numfullpathfound++;
836 	CACHE_UNLOCK();
837 
838 	*retbuf = bp;
839 	return (0);
840 }
841