xref: /freebsd/sys/kern/vfs_cache.c (revision 282a3889ebf826db9839be296ff1dd903f6d6d6e)
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 		return (-1);
423 	}
424 	ltype = 0;	/* silence gcc warning */
425 	if (cnp->cn_flags & ISDOTDOT) {
426 		ltype = VOP_ISLOCKED(dvp, cnp->cn_thread);
427 		VOP_UNLOCK(dvp, 0, cnp->cn_thread);
428 	}
429 	VI_LOCK(*vpp);
430 	CACHE_UNLOCK();
431 	error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
432 	if (cnp->cn_flags & ISDOTDOT)
433 		vn_lock(dvp, ltype | LK_RETRY, cnp->cn_thread);
434 	if (error) {
435 		*vpp = NULL;
436 		goto retry;
437 	}
438 	return (-1);
439 }
440 
441 /*
442  * Add an entry to the cache.
443  */
444 void
445 cache_enter(dvp, vp, cnp)
446 	struct vnode *dvp;
447 	struct vnode *vp;
448 	struct componentname *cnp;
449 {
450 	struct namecache *ncp;
451 	struct nchashhead *ncpp;
452 	u_int32_t hash;
453 	int hold;
454 	int zap;
455 	int len;
456 
457 	CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
458 	VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
459 	    ("cahe_enter: Adding a doomed vnode"));
460 
461 	if (!doingcache)
462 		return;
463 
464 	if (cnp->cn_nameptr[0] == '.') {
465 		if (cnp->cn_namelen == 1) {
466 			return;
467 		}
468 		/*
469 		 * For dotdot lookups only cache the v_dd pointer if the
470 		 * directory has a link back to its parent via v_cache_dst.
471 		 * Without this an unlinked directory would keep a soft
472 		 * reference to its parent which could not be NULLd at
473 		 * cache_purge() time.
474 		 */
475 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
476 			CACHE_LOCK();
477 			if (!TAILQ_EMPTY(&dvp->v_cache_dst))
478 				dvp->v_dd = vp;
479 			CACHE_UNLOCK();
480 			return;
481 		}
482 	}
483 
484 	hold = 0;
485 	zap = 0;
486 	ncp = cache_alloc(cnp->cn_namelen);
487 	CACHE_LOCK();
488 	numcache++;
489 	if (!vp) {
490 		numneg++;
491 		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
492 	} else if (vp->v_type == VDIR) {
493 		vp->v_dd = dvp;
494 	} else {
495 		vp->v_dd = NULL;
496 	}
497 
498 	/*
499 	 * Set the rest of the namecache entry elements, calculate it's
500 	 * hash key and insert it into the appropriate chain within
501 	 * the cache entries table.
502 	 */
503 	ncp->nc_vp = vp;
504 	ncp->nc_dvp = dvp;
505 	len = ncp->nc_nlen = cnp->cn_namelen;
506 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
507 	bcopy(cnp->cn_nameptr, ncp->nc_name, len);
508 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
509 	ncpp = NCHHASH(hash);
510 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
511 	if (LIST_EMPTY(&dvp->v_cache_src)) {
512 		hold = 1;
513 		numcachehv++;
514 	}
515 	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
516 	/*
517 	 * If the entry is "negative", we place it into the
518 	 * "negative" cache queue, otherwise, we place it into the
519 	 * destination vnode's cache entries queue.
520 	 */
521 	if (vp) {
522 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
523 	} else {
524 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
525 	}
526 	if (numneg * ncnegfactor > numcache) {
527 		ncp = TAILQ_FIRST(&ncneg);
528 		zap = 1;
529 	}
530 	if (hold)
531 		vhold(dvp);
532 	if (zap)
533 		cache_zap(ncp);
534 	CACHE_UNLOCK();
535 }
536 
537 /*
538  * Name cache initialization, from vfs_init() when we are booting
539  */
540 static void
541 nchinit(void *dummy __unused)
542 {
543 
544 	TAILQ_INIT(&ncneg);
545 
546 	cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
547 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
548 	cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
549 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
550 
551 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
552 }
553 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL)
554 
555 
556 /*
557  * Invalidate all entries to a particular vnode.
558  */
559 void
560 cache_purge(vp)
561 	struct vnode *vp;
562 {
563 
564 	CTR1(KTR_VFS, "cache_purge(%p)", vp);
565 	CACHE_LOCK();
566 	while (!LIST_EMPTY(&vp->v_cache_src))
567 		cache_zap(LIST_FIRST(&vp->v_cache_src));
568 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
569 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
570 	vp->v_dd = NULL;
571 	CACHE_UNLOCK();
572 }
573 
574 /*
575  * Flush all entries referencing a particular filesystem.
576  */
577 void
578 cache_purgevfs(mp)
579 	struct mount *mp;
580 {
581 	struct nchashhead *ncpp;
582 	struct namecache *ncp, *nnp;
583 
584 	/* Scan hash tables for applicable entries */
585 	CACHE_LOCK();
586 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
587 		LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
588 			if (ncp->nc_dvp->v_mount == mp)
589 				cache_zap(ncp);
590 		}
591 	}
592 	CACHE_UNLOCK();
593 }
594 
595 /*
596  * Perform canonical checks and cache lookup and pass on to filesystem
597  * through the vop_cachedlookup only if needed.
598  */
599 
600 int
601 vfs_cache_lookup(ap)
602 	struct vop_lookup_args /* {
603 		struct vnode *a_dvp;
604 		struct vnode **a_vpp;
605 		struct componentname *a_cnp;
606 	} */ *ap;
607 {
608 	struct vnode *dvp;
609 	int error;
610 	struct vnode **vpp = ap->a_vpp;
611 	struct componentname *cnp = ap->a_cnp;
612 	struct ucred *cred = cnp->cn_cred;
613 	int flags = cnp->cn_flags;
614 	struct thread *td = cnp->cn_thread;
615 
616 	*vpp = NULL;
617 	dvp = ap->a_dvp;
618 
619 	if (dvp->v_type != VDIR)
620 		return (ENOTDIR);
621 
622 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
623 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
624 		return (EROFS);
625 
626 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
627 	if (error)
628 		return (error);
629 
630 	error = cache_lookup(dvp, vpp, cnp);
631 	if (error == 0)
632 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
633 	if (error == ENOENT)
634 		return (error);
635 	return (0);
636 }
637 
638 
639 #ifndef _SYS_SYSPROTO_H_
640 struct  __getcwd_args {
641 	u_char	*buf;
642 	u_int	buflen;
643 };
644 #endif
645 
646 /*
647  * XXX All of these sysctls would probably be more productive dead.
648  */
649 static int disablecwd;
650 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
651    "Disable the getcwd syscall");
652 
653 /* Implementation of the getcwd syscall. */
654 int
655 __getcwd(td, uap)
656 	struct thread *td;
657 	struct __getcwd_args *uap;
658 {
659 
660 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
661 }
662 
663 int
664 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
665 {
666 	char *bp, *tmpbuf;
667 	struct filedesc *fdp;
668 	int error;
669 
670 	if (disablecwd)
671 		return (ENODEV);
672 	if (buflen < 2)
673 		return (EINVAL);
674 	if (buflen > MAXPATHLEN)
675 		buflen = MAXPATHLEN;
676 
677 	tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
678 	fdp = td->td_proc->p_fd;
679 	mtx_lock(&Giant);
680 	FILEDESC_SLOCK(fdp);
681 	error = vn_fullpath1(td, fdp->fd_cdir, fdp->fd_rdir, tmpbuf,
682 	    &bp, buflen);
683 	FILEDESC_SUNLOCK(fdp);
684 	mtx_unlock(&Giant);
685 
686 	if (!error) {
687 		if (bufseg == UIO_SYSSPACE)
688 			bcopy(bp, buf, strlen(bp) + 1);
689 		else
690 			error = copyout(bp, buf, strlen(bp) + 1);
691 	}
692 	free(tmpbuf, M_TEMP);
693 	return (error);
694 }
695 
696 /*
697  * Thus begins the fullpath magic.
698  */
699 
700 #undef STATNODE
701 #define STATNODE(name)							\
702 	static u_int name;						\
703 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
704 
705 static int disablefullpath;
706 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
707 	"Disable the vn_fullpath function");
708 
709 /* These count for kern___getcwd(), too. */
710 STATNODE(numfullpathcalls);
711 STATNODE(numfullpathfail1);
712 STATNODE(numfullpathfail2);
713 STATNODE(numfullpathfail4);
714 STATNODE(numfullpathfound);
715 
716 /*
717  * Retrieve the full filesystem path that correspond to a vnode from the name
718  * cache (if available)
719  */
720 int
721 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
722 {
723 	char *buf;
724 	struct filedesc *fdp;
725 	int error;
726 
727 	if (disablefullpath)
728 		return (ENODEV);
729 	if (vn == NULL)
730 		return (EINVAL);
731 
732 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
733 	fdp = td->td_proc->p_fd;
734 	FILEDESC_SLOCK(fdp);
735 	error = vn_fullpath1(td, vn, fdp->fd_rdir, buf, retbuf, MAXPATHLEN);
736 	FILEDESC_SUNLOCK(fdp);
737 
738 	if (!error)
739 		*freebuf = buf;
740 	else
741 		free(buf, M_TEMP);
742 	return (error);
743 }
744 
745 /*
746  * The magic behind kern___getcwd() and vn_fullpath().
747  */
748 static int
749 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
750     char *buf, char **retbuf, u_int buflen)
751 {
752 	char *bp;
753 	int error, i, slash_prefixed;
754 	struct namecache *ncp;
755 
756 	bp = buf + buflen - 1;
757 	*bp = '\0';
758 	error = 0;
759 	slash_prefixed = 0;
760 
761 	CACHE_LOCK();
762 	numfullpathcalls++;
763 	if (vp->v_type != VDIR) {
764 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
765 		if (!ncp) {
766 			numfullpathfail2++;
767 			CACHE_UNLOCK();
768 			return (ENOENT);
769 		}
770 		for (i = ncp->nc_nlen - 1; i >= 0 && bp > buf; i--)
771 			*--bp = ncp->nc_name[i];
772 		if (bp == buf) {
773 			numfullpathfail4++;
774 			CACHE_UNLOCK();
775 			return (ENOMEM);
776 		}
777 		*--bp = '/';
778 		slash_prefixed = 1;
779 		vp = ncp->nc_dvp;
780 	}
781 	while (vp != rdir && vp != rootvnode) {
782 		if (vp->v_vflag & VV_ROOT) {
783 			if (vp->v_iflag & VI_DOOMED) {	/* forced unmount */
784 				error = EBADF;
785 				break;
786 			}
787 			vp = vp->v_mount->mnt_vnodecovered;
788 			continue;
789 		}
790 		if (vp->v_dd == NULL) {
791 			numfullpathfail1++;
792 			error = ENOTDIR;
793 			break;
794 		}
795 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
796 		if (!ncp) {
797 			numfullpathfail2++;
798 			error = ENOENT;
799 			break;
800 		}
801 		MPASS(ncp->nc_dvp == vp->v_dd);
802 		for (i = ncp->nc_nlen - 1; i >= 0 && bp != buf; i--)
803 			*--bp = ncp->nc_name[i];
804 		if (bp == buf) {
805 			numfullpathfail4++;
806 			error = ENOMEM;
807 			break;
808 		}
809 		*--bp = '/';
810 		slash_prefixed = 1;
811 		vp = ncp->nc_dvp;
812 	}
813 	if (error) {
814 		CACHE_UNLOCK();
815 		return (error);
816 	}
817 	if (!slash_prefixed) {
818 		if (bp == buf) {
819 			numfullpathfail4++;
820 			CACHE_UNLOCK();
821 			return (ENOMEM);
822 		} else {
823 			*--bp = '/';
824 		}
825 	}
826 	numfullpathfound++;
827 	CACHE_UNLOCK();
828 
829 	*retbuf = bp;
830 	return (0);
831 }
832