xref: /freebsd/sys/kern/vfs_cache.c (revision 54ebdd631db8c0bba2baab0155f603a8b5cf014a)
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/filedesc.h>
40 #include <sys/fnv_hash.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/mutex.h>
46 #include <sys/namei.h>
47 #include <sys/proc.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysctl.h>
50 #include <sys/sysproto.h>
51 #include <sys/systm.h>
52 #include <sys/vnode.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_vptocnp(struct vnode **vp, char **bp, char *buf, u_int *buflen);
173 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
174     char *buf, char **retbuf, u_int buflen);
175 
176 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
177 
178 /*
179  * Flags in namecache.nc_flag
180  */
181 #define NCF_WHITE	1
182 
183 /*
184  * Grab an atomic snapshot of the name cache hash chain lengths
185  */
186 SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
187 
188 static int
189 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
190 {
191 	int error;
192 	struct nchashhead *ncpp;
193 	struct namecache *ncp;
194 	int n_nchash;
195 	int count;
196 
197 	n_nchash = nchash + 1;	/* nchash is max index, not count */
198 	if (!req->oldptr)
199 		return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
200 
201 	/* Scan hash tables for applicable entries */
202 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
203 		CACHE_LOCK();
204 		count = 0;
205 		LIST_FOREACH(ncp, ncpp, nc_hash) {
206 			count++;
207 		}
208 		CACHE_UNLOCK();
209 		error = SYSCTL_OUT(req, &count, sizeof(count));
210 		if (error)
211 			return (error);
212 	}
213 	return (0);
214 }
215 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD,
216 	0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths");
217 
218 static int
219 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
220 {
221 	int error;
222 	struct nchashhead *ncpp;
223 	struct namecache *ncp;
224 	int n_nchash;
225 	int count, maxlength, used, pct;
226 
227 	if (!req->oldptr)
228 		return SYSCTL_OUT(req, 0, 4 * sizeof(int));
229 
230 	n_nchash = nchash + 1;	/* nchash is max index, not count */
231 	used = 0;
232 	maxlength = 0;
233 
234 	/* Scan hash tables for applicable entries */
235 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
236 		count = 0;
237 		CACHE_LOCK();
238 		LIST_FOREACH(ncp, ncpp, nc_hash) {
239 			count++;
240 		}
241 		CACHE_UNLOCK();
242 		if (count)
243 			used++;
244 		if (maxlength < count)
245 			maxlength = count;
246 	}
247 	n_nchash = nchash + 1;
248 	pct = (used * 100 * 100) / n_nchash;
249 	error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
250 	if (error)
251 		return (error);
252 	error = SYSCTL_OUT(req, &used, sizeof(used));
253 	if (error)
254 		return (error);
255 	error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
256 	if (error)
257 		return (error);
258 	error = SYSCTL_OUT(req, &pct, sizeof(pct));
259 	if (error)
260 		return (error);
261 	return (0);
262 }
263 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD,
264 	0, 0, sysctl_debug_hashstat_nchash, "I", "nchash chain lengths");
265 
266 /*
267  * cache_zap():
268  *
269  *   Removes a namecache entry from cache, whether it contains an actual
270  *   pointer to a vnode or if it is just a negative cache entry.
271  */
272 static void
273 cache_zap(ncp)
274 	struct namecache *ncp;
275 {
276 	struct vnode *vp;
277 
278 	mtx_assert(&cache_lock, MA_OWNED);
279 	CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
280 	vp = NULL;
281 	LIST_REMOVE(ncp, nc_hash);
282 	LIST_REMOVE(ncp, nc_src);
283 	if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
284 		vp = ncp->nc_dvp;
285 		numcachehv--;
286 	}
287 	if (ncp->nc_vp) {
288 		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
289 		ncp->nc_vp->v_dd = NULL;
290 	} else {
291 		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
292 		numneg--;
293 	}
294 	numcache--;
295 	cache_free(ncp);
296 	if (vp)
297 		vdrop(vp);
298 }
299 
300 /*
301  * Lookup an entry in the cache
302  *
303  * Lookup is called with dvp pointing to the directory to search,
304  * cnp pointing to the name of the entry being sought. If the lookup
305  * succeeds, the vnode is returned in *vpp, and a status of -1 is
306  * returned. If the lookup determines that the name does not exist
307  * (negative cacheing), a status of ENOENT is returned. If the lookup
308  * fails, a status of zero is returned.  If the directory vnode is
309  * recycled out from under us due to a forced unmount, a status of
310  * EBADF is returned.
311  *
312  * vpp is locked and ref'd on return.  If we're looking up DOTDOT, dvp is
313  * unlocked.  If we're looking up . an extra ref is taken, but the lock is
314  * not recursively acquired.
315  */
316 
317 int
318 cache_lookup(dvp, vpp, cnp)
319 	struct vnode *dvp;
320 	struct vnode **vpp;
321 	struct componentname *cnp;
322 {
323 	struct namecache *ncp;
324 	u_int32_t hash;
325 	int error, ltype;
326 
327 	if (!doingcache) {
328 		cnp->cn_flags &= ~MAKEENTRY;
329 		return (0);
330 	}
331 retry:
332 	CACHE_LOCK();
333 	numcalls++;
334 
335 	if (cnp->cn_nameptr[0] == '.') {
336 		if (cnp->cn_namelen == 1) {
337 			*vpp = dvp;
338 			CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
339 			    dvp, cnp->cn_nameptr);
340 			dothits++;
341 			goto success;
342 		}
343 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
344 			dotdothits++;
345 			if (dvp->v_dd == NULL ||
346 			    (cnp->cn_flags & MAKEENTRY) == 0) {
347 				CACHE_UNLOCK();
348 				return (0);
349 			}
350 			*vpp = dvp->v_dd;
351 			CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
352 			    dvp, cnp->cn_nameptr, *vpp);
353 			goto success;
354 		}
355 	}
356 
357 	hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
358 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
359 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
360 		numchecks++;
361 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
362 		    !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
363 			break;
364 	}
365 
366 	/* We failed to find an entry */
367 	if (ncp == 0) {
368 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
369 			nummisszap++;
370 		} else {
371 			nummiss++;
372 		}
373 		nchstats.ncs_miss++;
374 		CACHE_UNLOCK();
375 		return (0);
376 	}
377 
378 	/* We don't want to have an entry, so dump it */
379 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
380 		numposzaps++;
381 		nchstats.ncs_badhits++;
382 		cache_zap(ncp);
383 		CACHE_UNLOCK();
384 		return (0);
385 	}
386 
387 	/* We found a "positive" match, return the vnode */
388 	if (ncp->nc_vp) {
389 		numposhits++;
390 		nchstats.ncs_goodhits++;
391 		*vpp = ncp->nc_vp;
392 		CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
393 		    dvp, cnp->cn_nameptr, *vpp, ncp);
394 		goto success;
395 	}
396 
397 	/* We found a negative match, and want to create it, so purge */
398 	if (cnp->cn_nameiop == CREATE) {
399 		numnegzaps++;
400 		nchstats.ncs_badhits++;
401 		cache_zap(ncp);
402 		CACHE_UNLOCK();
403 		return (0);
404 	}
405 
406 	numneghits++;
407 	/*
408 	 * We found a "negative" match, so we shift it to the end of
409 	 * the "negative" cache entries queue to satisfy LRU.  Also,
410 	 * check to see if the entry is a whiteout; indicate this to
411 	 * the componentname, if so.
412 	 */
413 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
414 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
415 	nchstats.ncs_neghits++;
416 	if (ncp->nc_flag & NCF_WHITE)
417 		cnp->cn_flags |= ISWHITEOUT;
418 	CACHE_UNLOCK();
419 	return (ENOENT);
420 
421 success:
422 	/*
423 	 * On success we return a locked and ref'd vnode as per the lookup
424 	 * protocol.
425 	 */
426 	if (dvp == *vpp) {   /* lookup on "." */
427 		VREF(*vpp);
428 		CACHE_UNLOCK();
429 		/*
430 		 * When we lookup "." we still can be asked to lock it
431 		 * differently...
432 		 */
433 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
434 		if (ltype != VOP_ISLOCKED(*vpp)) {
435 			if (ltype == LK_EXCLUSIVE) {
436 				vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
437 				if ((*vpp)->v_iflag & VI_DOOMED) {
438 					/* forced unmount */
439 					vrele(*vpp);
440 					*vpp = NULL;
441 					return (EBADF);
442 				}
443 			} else
444 				vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
445 		}
446 		return (-1);
447 	}
448 	ltype = 0;	/* silence gcc warning */
449 	if (cnp->cn_flags & ISDOTDOT) {
450 		ltype = VOP_ISLOCKED(dvp);
451 		VOP_UNLOCK(dvp, 0);
452 	}
453 	VI_LOCK(*vpp);
454 	CACHE_UNLOCK();
455 	error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
456 	if (cnp->cn_flags & ISDOTDOT)
457 		vn_lock(dvp, ltype | LK_RETRY);
458 	if (error) {
459 		*vpp = NULL;
460 		goto retry;
461 	}
462 	if ((cnp->cn_flags & ISLASTCN) &&
463 	    (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
464 		ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
465 	}
466 	return (-1);
467 }
468 
469 /*
470  * Add an entry to the cache.
471  */
472 void
473 cache_enter(dvp, vp, cnp)
474 	struct vnode *dvp;
475 	struct vnode *vp;
476 	struct componentname *cnp;
477 {
478 	struct namecache *ncp, *n2;
479 	struct nchashhead *ncpp;
480 	u_int32_t hash;
481 	int hold;
482 	int zap;
483 	int len;
484 
485 	CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
486 	VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
487 	    ("cahe_enter: Adding a doomed vnode"));
488 
489 	if (!doingcache)
490 		return;
491 
492 	if (cnp->cn_nameptr[0] == '.') {
493 		if (cnp->cn_namelen == 1) {
494 			return;
495 		}
496 		/*
497 		 * For dotdot lookups only cache the v_dd pointer if the
498 		 * directory has a link back to its parent via v_cache_dst.
499 		 * Without this an unlinked directory would keep a soft
500 		 * reference to its parent which could not be NULLd at
501 		 * cache_purge() time.
502 		 */
503 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
504 			CACHE_LOCK();
505 			if (!TAILQ_EMPTY(&dvp->v_cache_dst))
506 				dvp->v_dd = vp;
507 			CACHE_UNLOCK();
508 			return;
509 		}
510 	}
511 
512 	hold = 0;
513 	zap = 0;
514 
515 	/*
516 	 * Calculate the hash key and setup as much of the new
517 	 * namecache entry as possible before acquiring the lock.
518 	 */
519 	ncp = cache_alloc(cnp->cn_namelen);
520 	ncp->nc_vp = vp;
521 	ncp->nc_dvp = dvp;
522 	len = ncp->nc_nlen = cnp->cn_namelen;
523 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
524 	bcopy(cnp->cn_nameptr, ncp->nc_name, len);
525 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
526 	CACHE_LOCK();
527 
528 	/*
529 	 * See if this vnode is already in the cache with this name.
530 	 * This can happen with concurrent lookups of the same path
531 	 * name.
532 	 */
533 	if (vp) {
534 		TAILQ_FOREACH(n2, &vp->v_cache_dst, nc_dst) {
535 			if (n2->nc_dvp == dvp &&
536 			    n2->nc_nlen == cnp->cn_namelen &&
537 			    !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
538 				CACHE_UNLOCK();
539 				cache_free(ncp);
540 				return;
541 			}
542 		}
543 	} else {
544 		TAILQ_FOREACH(n2, &ncneg, nc_dst) {
545 			if (n2->nc_nlen == cnp->cn_namelen &&
546 			    !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
547 				CACHE_UNLOCK();
548 				cache_free(ncp);
549 				return;
550 			}
551 		}
552 	}
553 
554 	numcache++;
555 	if (!vp) {
556 		numneg++;
557 		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
558 	} else if (vp->v_type == VDIR) {
559 		vp->v_dd = dvp;
560 	} else {
561 		vp->v_dd = NULL;
562 	}
563 
564 	/*
565 	 * Insert the new namecache entry into the appropriate chain
566 	 * within the cache entries table.
567 	 */
568 	ncpp = NCHHASH(hash);
569 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
570 	if (LIST_EMPTY(&dvp->v_cache_src)) {
571 		hold = 1;
572 		numcachehv++;
573 	}
574 	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
575 	/*
576 	 * If the entry is "negative", we place it into the
577 	 * "negative" cache queue, otherwise, we place it into the
578 	 * destination vnode's cache entries queue.
579 	 */
580 	if (vp) {
581 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
582 	} else {
583 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
584 	}
585 	if (numneg * ncnegfactor > numcache) {
586 		ncp = TAILQ_FIRST(&ncneg);
587 		zap = 1;
588 	}
589 	if (hold)
590 		vhold(dvp);
591 	if (zap)
592 		cache_zap(ncp);
593 	CACHE_UNLOCK();
594 }
595 
596 /*
597  * Name cache initialization, from vfs_init() when we are booting
598  */
599 static void
600 nchinit(void *dummy __unused)
601 {
602 
603 	TAILQ_INIT(&ncneg);
604 
605 	cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
606 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
607 	cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
608 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
609 
610 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
611 }
612 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
613 
614 
615 /*
616  * Invalidate all entries to a particular vnode.
617  */
618 void
619 cache_purge(vp)
620 	struct vnode *vp;
621 {
622 
623 	CTR1(KTR_VFS, "cache_purge(%p)", vp);
624 	CACHE_LOCK();
625 	while (!LIST_EMPTY(&vp->v_cache_src))
626 		cache_zap(LIST_FIRST(&vp->v_cache_src));
627 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
628 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
629 	vp->v_dd = NULL;
630 	CACHE_UNLOCK();
631 }
632 
633 /*
634  * Flush all entries referencing a particular filesystem.
635  */
636 void
637 cache_purgevfs(mp)
638 	struct mount *mp;
639 {
640 	struct nchashhead *ncpp;
641 	struct namecache *ncp, *nnp;
642 
643 	/* Scan hash tables for applicable entries */
644 	CACHE_LOCK();
645 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
646 		LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
647 			if (ncp->nc_dvp->v_mount == mp)
648 				cache_zap(ncp);
649 		}
650 	}
651 	CACHE_UNLOCK();
652 }
653 
654 /*
655  * Perform canonical checks and cache lookup and pass on to filesystem
656  * through the vop_cachedlookup only if needed.
657  */
658 
659 int
660 vfs_cache_lookup(ap)
661 	struct vop_lookup_args /* {
662 		struct vnode *a_dvp;
663 		struct vnode **a_vpp;
664 		struct componentname *a_cnp;
665 	} */ *ap;
666 {
667 	struct vnode *dvp;
668 	int error;
669 	struct vnode **vpp = ap->a_vpp;
670 	struct componentname *cnp = ap->a_cnp;
671 	struct ucred *cred = cnp->cn_cred;
672 	int flags = cnp->cn_flags;
673 	struct thread *td = cnp->cn_thread;
674 
675 	*vpp = NULL;
676 	dvp = ap->a_dvp;
677 
678 	if (dvp->v_type != VDIR)
679 		return (ENOTDIR);
680 
681 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
682 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
683 		return (EROFS);
684 
685 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
686 	if (error)
687 		return (error);
688 
689 	error = cache_lookup(dvp, vpp, cnp);
690 	if (error == 0)
691 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
692 	if (error == -1)
693 		return (0);
694 	return (error);
695 }
696 
697 
698 #ifndef _SYS_SYSPROTO_H_
699 struct  __getcwd_args {
700 	u_char	*buf;
701 	u_int	buflen;
702 };
703 #endif
704 
705 /*
706  * XXX All of these sysctls would probably be more productive dead.
707  */
708 static int disablecwd;
709 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
710    "Disable the getcwd syscall");
711 
712 /* Implementation of the getcwd syscall. */
713 int
714 __getcwd(td, uap)
715 	struct thread *td;
716 	struct __getcwd_args *uap;
717 {
718 
719 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
720 }
721 
722 int
723 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
724 {
725 	char *bp, *tmpbuf;
726 	struct filedesc *fdp;
727 	struct vnode *cdir, *rdir;
728 	int error, vfslocked;
729 
730 	if (disablecwd)
731 		return (ENODEV);
732 	if (buflen < 2)
733 		return (EINVAL);
734 	if (buflen > MAXPATHLEN)
735 		buflen = MAXPATHLEN;
736 
737 	tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
738 	fdp = td->td_proc->p_fd;
739 	FILEDESC_SLOCK(fdp);
740 	cdir = fdp->fd_cdir;
741 	VREF(cdir);
742 	rdir = fdp->fd_rdir;
743 	VREF(rdir);
744 	FILEDESC_SUNLOCK(fdp);
745 	error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
746 	vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
747 	vrele(rdir);
748 	VFS_UNLOCK_GIANT(vfslocked);
749 	vfslocked = VFS_LOCK_GIANT(cdir->v_mount);
750 	vrele(cdir);
751 	VFS_UNLOCK_GIANT(vfslocked);
752 
753 	if (!error) {
754 		if (bufseg == UIO_SYSSPACE)
755 			bcopy(bp, buf, strlen(bp) + 1);
756 		else
757 			error = copyout(bp, buf, strlen(bp) + 1);
758 	}
759 	free(tmpbuf, M_TEMP);
760 	return (error);
761 }
762 
763 /*
764  * Thus begins the fullpath magic.
765  */
766 
767 #undef STATNODE
768 #define STATNODE(name)							\
769 	static u_int name;						\
770 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
771 
772 static int disablefullpath;
773 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
774 	"Disable the vn_fullpath function");
775 
776 /* These count for kern___getcwd(), too. */
777 STATNODE(numfullpathcalls);
778 STATNODE(numfullpathfail1);
779 STATNODE(numfullpathfail2);
780 STATNODE(numfullpathfail4);
781 STATNODE(numfullpathfound);
782 
783 /*
784  * Retrieve the full filesystem path that correspond to a vnode from the name
785  * cache (if available)
786  */
787 int
788 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
789 {
790 	char *buf;
791 	struct filedesc *fdp;
792 	struct vnode *rdir;
793 	int error, vfslocked;
794 
795 	if (disablefullpath)
796 		return (ENODEV);
797 	if (vn == NULL)
798 		return (EINVAL);
799 
800 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
801 	fdp = td->td_proc->p_fd;
802 	FILEDESC_SLOCK(fdp);
803 	rdir = fdp->fd_rdir;
804 	VREF(rdir);
805 	FILEDESC_SUNLOCK(fdp);
806 	error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
807 	vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
808 	vrele(rdir);
809 	VFS_UNLOCK_GIANT(vfslocked);
810 
811 	if (!error)
812 		*freebuf = buf;
813 	else
814 		free(buf, M_TEMP);
815 	return (error);
816 }
817 
818 /*
819  * This function is similar to vn_fullpath, but it attempts to lookup the
820  * pathname relative to the global root mount point.  This is required for the
821  * auditing sub-system, as audited pathnames must be absolute, relative to the
822  * global root mount point.
823  */
824 int
825 vn_fullpath_global(struct thread *td, struct vnode *vn,
826     char **retbuf, char **freebuf)
827 {
828 	char *buf;
829 	int error;
830 
831 	if (disablefullpath)
832 		return (ENODEV);
833 	if (vn == NULL)
834 		return (EINVAL);
835 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
836 	error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
837 	if (!error)
838 		*freebuf = buf;
839 	else
840 		free(buf, M_TEMP);
841 	return (error);
842 }
843 
844 static int
845 vn_vptocnp(struct vnode **vp, char **bp, char *buf, u_int *buflen)
846 {
847 	struct vnode *dvp;
848 	int error, vfslocked;
849 
850 	vhold(*vp);
851 	CACHE_UNLOCK();
852 	vfslocked = VFS_LOCK_GIANT((*vp)->v_mount);
853 	vn_lock(*vp, LK_SHARED | LK_RETRY);
854 	vdrop(*vp);
855 	error = VOP_VPTOCNP(*vp, &dvp, buf, buflen);
856 	VOP_UNLOCK(*vp, 0);
857 	VFS_UNLOCK_GIANT(vfslocked);
858 	if (error) {
859 		numfullpathfail2++;
860 		return (error);
861 	}
862 	*bp = buf + *buflen;
863 	*vp = dvp;
864 	CACHE_LOCK();
865 	if ((*vp)->v_iflag & VI_DOOMED) {
866 		/* forced unmount */
867 		CACHE_UNLOCK();
868 		vdrop(*vp);
869 		return (ENOENT);
870 	}
871 	vdrop(*vp);
872 
873 	return (0);
874 }
875 
876 /*
877  * The magic behind kern___getcwd() and vn_fullpath().
878  */
879 static int
880 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
881     char *buf, char **retbuf, u_int buflen)
882 {
883 	char *bp;
884 	int error, i, slash_prefixed;
885 	struct namecache *ncp;
886 
887 	buflen--;
888 	bp = buf + buflen;
889 	*bp = '\0';
890 	error = 0;
891 	slash_prefixed = 0;
892 
893 	CACHE_LOCK();
894 	numfullpathcalls++;
895 	if (vp->v_type != VDIR) {
896 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
897 		if (ncp != NULL) {
898 			for (i = ncp->nc_nlen - 1; i >= 0 && bp > buf; i--)
899 				*--bp = ncp->nc_name[i];
900 			if (bp == buf) {
901 				numfullpathfail4++;
902 				CACHE_UNLOCK();
903 				return (ENOMEM);
904 			}
905 			vp = ncp->nc_dvp;
906 		} else {
907 			error = vn_vptocnp(&vp, &bp, buf, &buflen);
908 			if (error) {
909 				return (error);
910 			}
911 		}
912 		*--bp = '/';
913 		buflen--;
914 		if (buflen < 0) {
915 			numfullpathfail4++;
916 			CACHE_UNLOCK();
917 			return (ENOMEM);
918 		}
919 		slash_prefixed = 1;
920 	}
921 	while (vp != rdir && vp != rootvnode) {
922 		if (vp->v_vflag & VV_ROOT) {
923 			if (vp->v_iflag & VI_DOOMED) {	/* forced unmount */
924 				CACHE_UNLOCK();
925 				error = EBADF;
926 				break;
927 			}
928 			vp = vp->v_mount->mnt_vnodecovered;
929 			continue;
930 		}
931 		if (vp->v_type != VDIR) {
932 			numfullpathfail1++;
933 			CACHE_UNLOCK();
934 			error = ENOTDIR;
935 			break;
936 		}
937 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
938 		if (ncp != NULL) {
939 			MPASS(ncp->nc_dvp == vp->v_dd);
940 			buflen -= ncp->nc_nlen - 1;
941 			for (i = ncp->nc_nlen - 1; i >= 0 && bp != buf; i--)
942 				*--bp = ncp->nc_name[i];
943 			if (bp == buf) {
944 				numfullpathfail4++;
945 				CACHE_UNLOCK();
946 				error = ENOMEM;
947 				break;
948 			}
949 			vp = ncp->nc_dvp;
950 		} else {
951 			error = vn_vptocnp(&vp, &bp, buf, &buflen);
952 			if (error) {
953 				break;
954 			}
955 		}
956 		*--bp = '/';
957 		buflen--;
958 		if (buflen < 0) {
959 			numfullpathfail4++;
960 			CACHE_UNLOCK();
961 			error = ENOMEM;
962 			break;
963 		}
964 		slash_prefixed = 1;
965 	}
966 	if (error)
967 		return (error);
968 	if (!slash_prefixed) {
969 		if (bp == buf) {
970 			numfullpathfail4++;
971 			CACHE_UNLOCK();
972 			return (ENOMEM);
973 		} else {
974 			*--bp = '/';
975 		}
976 	}
977 	numfullpathfound++;
978 	CACHE_UNLOCK();
979 
980 	*retbuf = bp;
981 	return (0);
982 }
983 
984 int
985 vn_commname(struct vnode *vp, char *buf, u_int buflen)
986 {
987 	struct namecache *ncp;
988 	int l;
989 
990 	CACHE_LOCK();
991 	ncp = TAILQ_FIRST(&vp->v_cache_dst);
992 	if (!ncp) {
993 		CACHE_UNLOCK();
994 		return (ENOENT);
995 	}
996 	l = min(ncp->nc_nlen, buflen - 1);
997 	memcpy(buf, ncp->nc_name, l);
998 	CACHE_UNLOCK();
999 	buf[l] = '\0';
1000 	return (0);
1001 }
1002