xref: /freebsd/sys/kern/vfs_cache.c (revision 6b3455a7665208c366849f0b2b3bc916fb97516e)
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 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 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, int locked);
172 
173 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
174 
175 /*
176  * Flags in namecache.nc_flag
177  */
178 #define NCF_WHITE	1
179 
180 /*
181  * Grab an atomic snapshot of the name cache hash chain lengths
182  */
183 SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
184 
185 static int
186 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
187 {
188 	int error;
189 	struct nchashhead *ncpp;
190 	struct namecache *ncp;
191 	int n_nchash;
192 	int count;
193 
194 	n_nchash = nchash + 1;	/* nchash is max index, not count */
195 	if (!req->oldptr)
196 		return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
197 
198 	/* Scan hash tables for applicable entries */
199 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
200 		count = 0;
201 		LIST_FOREACH(ncp, ncpp, nc_hash) {
202 			count++;
203 		}
204 		error = SYSCTL_OUT(req, &count, sizeof(count));
205 		if (error)
206 			return (error);
207 	}
208 	return (0);
209 }
210 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD,
211 	0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths");
212 
213 static int
214 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
215 {
216 	int error;
217 	struct nchashhead *ncpp;
218 	struct namecache *ncp;
219 	int n_nchash;
220 	int count, maxlength, used, pct;
221 
222 	if (!req->oldptr)
223 		return SYSCTL_OUT(req, 0, 4 * sizeof(int));
224 
225 	n_nchash = nchash + 1;	/* nchash is max index, not count */
226 	used = 0;
227 	maxlength = 0;
228 
229 	/* Scan hash tables for applicable entries */
230 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
231 		count = 0;
232 		LIST_FOREACH(ncp, ncpp, nc_hash) {
233 			count++;
234 		}
235 		if (count)
236 			used++;
237 		if (maxlength < count)
238 			maxlength = count;
239 	}
240 	n_nchash = nchash + 1;
241 	pct = (used * 100 * 100) / n_nchash;
242 	error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
243 	if (error)
244 		return (error);
245 	error = SYSCTL_OUT(req, &used, sizeof(used));
246 	if (error)
247 		return (error);
248 	error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
249 	if (error)
250 		return (error);
251 	error = SYSCTL_OUT(req, &pct, sizeof(pct));
252 	if (error)
253 		return (error);
254 	return (0);
255 }
256 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD,
257 	0, 0, sysctl_debug_hashstat_nchash, "I", "nchash chain lengths");
258 
259 /*
260  * cache_zap():
261  *
262  *   Removes a namecache entry from cache, whether it contains an actual
263  *   pointer to a vnode or if it is just a negative cache entry.
264  */
265 static void
266 cache_zap(ncp, locked)
267 	struct namecache *ncp;
268 	int locked;
269 {
270 	struct vnode *vp;
271 
272 	vp = NULL;
273 	if (!locked)
274 		CACHE_LOCK();
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_UNLOCK();
289 	cache_free(ncp);
290 	if (vp)
291 		vdrop(vp);
292 	if (locked)
293 		CACHE_LOCK();
294 }
295 
296 /*
297  * cache_leaf_test()
298  *
299  *      Test whether this (directory) vnode's namei cache entry contains
300  *      subdirectories or not.  Used to determine whether the directory is
301  *      a leaf in the namei cache or not.  Note: the directory may still
302  *      contain files in the namei cache.
303  *
304  *      Returns 0 if the directory is a leaf, -1 if it isn't.
305  */
306 int
307 cache_leaf_test(struct vnode *vp)
308 {
309 	struct namecache *ncpc;
310 	int leaf;
311 
312 	leaf = 0;
313 	CACHE_LOCK();
314 	for (ncpc = LIST_FIRST(&vp->v_cache_src);
315 	     ncpc != NULL;
316 	     ncpc = LIST_NEXT(ncpc, nc_src)
317 	 ) {
318 		if (ncpc->nc_vp != NULL && ncpc->nc_vp->v_type == VDIR) {
319 			leaf = -1;
320 			break;
321 		}
322 	}
323 	CACHE_UNLOCK();
324 	return (leaf);
325 }
326 
327 /*
328  * Lookup an entry in the cache
329  *
330  * Lookup is called with dvp pointing to the directory to search,
331  * cnp pointing to the name of the entry being sought. If the lookup
332  * succeeds, the vnode is returned in *vpp, and a status of -1 is
333  * returned. If the lookup determines that the name does not exist
334  * (negative cacheing), a status of ENOENT is returned. If the lookup
335  * fails, a status of zero is returned.
336  */
337 
338 int
339 cache_lookup(dvp, vpp, cnp)
340 	struct vnode *dvp;
341 	struct vnode **vpp;
342 	struct componentname *cnp;
343 {
344 	struct namecache *ncp;
345 	u_int32_t hash;
346 
347 	if (!doingcache) {
348 		cnp->cn_flags &= ~MAKEENTRY;
349 		return (0);
350 	}
351 
352 	CACHE_LOCK();
353 	numcalls++;
354 
355 	if (cnp->cn_nameptr[0] == '.') {
356 		if (cnp->cn_namelen == 1) {
357 			*vpp = dvp;
358 			dothits++;
359 			CACHE_UNLOCK();
360 			return (-1);
361 		}
362 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
363 			dotdothits++;
364 			if (dvp->v_dd->v_id != dvp->v_ddid ||
365 			    (cnp->cn_flags & MAKEENTRY) == 0) {
366 				dvp->v_ddid = 0;
367 				CACHE_UNLOCK();
368 				return (0);
369 			}
370 			*vpp = dvp->v_dd;
371 			CACHE_UNLOCK();
372 			return (-1);
373 		}
374 	}
375 
376 	hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
377 	hash = fnv_32_buf(&dvp->v_id, sizeof(dvp->v_id), hash);
378 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
379 		numchecks++;
380 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
381 		    !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
382 			break;
383 	}
384 
385 	/* We failed to find an entry */
386 	if (ncp == 0) {
387 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
388 			nummisszap++;
389 		} else {
390 			nummiss++;
391 		}
392 		nchstats.ncs_miss++;
393 		CACHE_UNLOCK();
394 		return (0);
395 	}
396 
397 	/* We don't want to have an entry, so dump it */
398 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
399 		numposzaps++;
400 		nchstats.ncs_badhits++;
401 		CACHE_UNLOCK();
402 		cache_zap(ncp, 0);
403 		return (0);
404 	}
405 
406 	/* We found a "positive" match, return the vnode */
407 	if (ncp->nc_vp) {
408 		numposhits++;
409 		nchstats.ncs_goodhits++;
410 		*vpp = ncp->nc_vp;
411 		CACHE_UNLOCK();
412 		return (-1);
413 	}
414 
415 	/* We found a negative match, and want to create it, so purge */
416 	if (cnp->cn_nameiop == CREATE) {
417 		numnegzaps++;
418 		nchstats.ncs_badhits++;
419 		CACHE_UNLOCK();
420 		cache_zap(ncp, 0);
421 		return (0);
422 	}
423 
424 	numneghits++;
425 	/*
426 	 * We found a "negative" match, so we shift it to the end of
427 	 * the "negative" cache entries queue to satisfy LRU.  Also,
428 	 * check to see if the entry is a whiteout; indicate this to
429 	 * the componentname, if so.
430 	 */
431 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
432 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
433 	nchstats.ncs_neghits++;
434 	if (ncp->nc_flag & NCF_WHITE)
435 		cnp->cn_flags |= ISWHITEOUT;
436 	CACHE_UNLOCK();
437 	return (ENOENT);
438 }
439 
440 /*
441  * Add an entry to the cache.
442  */
443 void
444 cache_enter(dvp, vp, cnp)
445 	struct vnode *dvp;
446 	struct vnode *vp;
447 	struct componentname *cnp;
448 {
449 	struct namecache *ncp;
450 	struct nchashhead *ncpp;
451 	u_int32_t hash;
452 	int hold;
453 	int zap;
454 	int len;
455 
456 	if (!doingcache)
457 		return;
458 
459 	if (cnp->cn_nameptr[0] == '.') {
460 		if (cnp->cn_namelen == 1) {
461 			return;
462 		}
463 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
464 			if (vp) {
465 				dvp->v_dd = vp;
466 				dvp->v_ddid = vp->v_id;
467 			} else {
468 				dvp->v_dd = dvp;
469 				dvp->v_ddid = 0;
470 			}
471 			return;
472 		}
473 	}
474 
475 	hold = 0;
476 	zap = 0;
477 	ncp = cache_alloc(cnp->cn_namelen);
478 	CACHE_LOCK();
479 	numcache++;
480 	if (!vp) {
481 		numneg++;
482 		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
483 	} else if (vp->v_type == VDIR) {
484 		vp->v_dd = dvp;
485 		vp->v_ddid = dvp->v_id;
486 	}
487 
488 	/*
489 	 * Set the rest of the namecache entry elements, calculate it's
490 	 * hash key and insert it into the appropriate chain within
491 	 * the cache entries table.
492 	 */
493 	ncp->nc_vp = vp;
494 	ncp->nc_dvp = dvp;
495 	len = ncp->nc_nlen = cnp->cn_namelen;
496 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
497 	bcopy(cnp->cn_nameptr, ncp->nc_name, len);
498 	hash = fnv_32_buf(&dvp->v_id, sizeof(dvp->v_id), hash);
499 	ncpp = NCHHASH(hash);
500 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
501 	if (LIST_EMPTY(&dvp->v_cache_src)) {
502 		hold = 1;
503 		numcachehv++;
504 	}
505 	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
506 	/*
507 	 * If the entry is "negative", we place it into the
508 	 * "negative" cache queue, otherwise, we place it into the
509 	 * destination vnode's cache entries queue.
510 	 */
511 	if (vp) {
512 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
513 	} else {
514 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
515 	}
516 	if (numneg * ncnegfactor > numcache) {
517 		ncp = TAILQ_FIRST(&ncneg);
518 		zap = 1;
519 	}
520 	CACHE_UNLOCK();
521 	if (hold)
522 		vhold(dvp);
523 	if (zap)
524 		cache_zap(ncp, 0);
525 }
526 
527 /*
528  * Name cache initialization, from vfs_init() when we are booting
529  */
530 static void
531 nchinit(void *dummy __unused)
532 {
533 
534 	TAILQ_INIT(&ncneg);
535 
536 	cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
537 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
538 	cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
539 	    NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
540 
541 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
542 }
543 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL)
544 
545 
546 /*
547  * Invalidate all entries to a particular vnode.
548  *
549  * Remove all entries in the namecache relating to this vnode and
550  * change the v_id.  We take the v_id from a global counter, since
551  * it becomes a handy sequence number in crash-dumps that way.
552  * No valid vnode will ever have (v_id == 0).
553  *
554  * XXX: Only time and the size of v_id prevents this from failing:
555  * XXX: In theory we should hunt down all (struct vnode*, v_id)
556  * XXX: soft references and nuke them, at least on the global
557  * XXX: v_id wraparound.  The period of resistance can be extended
558  * XXX: by incrementing each vnodes v_id individually instead of
559  * XXX: using the global v_id.
560  */
561 
562 /*
563  * XXX This is sometimes called when a vnode may still be re-used, in which
564  * case v_dd may be invalid.  Need to look this up.
565  */
566 void
567 cache_purge(vp)
568 	struct vnode *vp;
569 {
570 	static u_long nextid;
571 
572 	CACHE_LOCK();
573 	while (!LIST_EMPTY(&vp->v_cache_src))
574 		cache_zap(LIST_FIRST(&vp->v_cache_src), 1);
575 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
576 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst), 1);
577 
578 	do
579 		nextid++;
580 	while (nextid == vp->v_id || !nextid);
581 	vp->v_id = nextid;
582 	vp->v_dd = vp;
583 	vp->v_ddid = 0;
584 	CACHE_UNLOCK();
585 }
586 
587 /*
588  * Flush all entries referencing a particular filesystem.
589  *
590  * Since we need to check it anyway, we will flush all the invalid
591  * entries at the same time.
592  */
593 void
594 cache_purgevfs(mp)
595 	struct mount *mp;
596 {
597 	struct nchashhead *ncpp;
598 	struct namecache *ncp, *nnp;
599 	struct nchashhead mplist;
600 
601 	LIST_INIT(&mplist);
602 	ncp = NULL;
603 
604 	/* Scan hash tables for applicable entries */
605 	CACHE_LOCK();
606 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
607 		for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
608 			nnp = LIST_NEXT(ncp, nc_hash);
609 			if (ncp->nc_dvp->v_mount == mp) {
610 				LIST_REMOVE(ncp, nc_hash);
611 				LIST_INSERT_HEAD(&mplist, ncp, nc_hash);
612 			}
613 		}
614 	}
615 	CACHE_UNLOCK();
616 	while (!LIST_EMPTY(&mplist))
617 		cache_zap(LIST_FIRST(&mplist), 0);
618 }
619 
620 /*
621  * Perform canonical checks and cache lookup and pass on to filesystem
622  * through the vop_cachedlookup only if needed.
623  */
624 
625 int
626 vfs_cache_lookup(ap)
627 	struct vop_lookup_args /* {
628 		struct vnode *a_dvp;
629 		struct vnode **a_vpp;
630 		struct componentname *a_cnp;
631 	} */ *ap;
632 {
633 	struct vnode *dvp, *vp;
634 	int lockparent;
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 	u_long vpid;	/* capability number of vnode */
642 
643 	*vpp = NULL;
644 	dvp = ap->a_dvp;
645 	lockparent = flags & LOCKPARENT;
646 
647 	if (dvp->v_type != VDIR)
648 		return (ENOTDIR);
649 
650 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
651 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
652 		return (EROFS);
653 
654 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
655 
656 	if (error)
657 		return (error);
658 
659 	error = cache_lookup(dvp, vpp, cnp);
660 
661 #ifdef LOOKUP_SHARED
662 	if (!error) {
663 		/* We do this because the rest of the system now expects to get
664 		 * a shared lock, which is later upgraded if LOCKSHARED is not
665 		 * set.  We have so many cases here because of bugs that yield
666 		 * inconsistant lock states.  This all badly needs to be fixed
667 		 */
668 		error = VOP_CACHEDLOOKUP(dvp, vpp, cnp);
669 		if (!error) {
670 			int flock;
671 
672 			flock = VOP_ISLOCKED(*vpp, td);
673 			if (flock != LK_EXCLUSIVE) {
674 				if (flock == 0) {
675 					if ((flags & ISLASTCN) &&
676 					    (flags & LOCKSHARED))
677 						VOP_LOCK(*vpp, LK_SHARED, td);
678 					else
679 						VOP_LOCK(*vpp, LK_EXCLUSIVE, td);
680 				}
681 			} else if ((flags & ISLASTCN) && (flags & LOCKSHARED))
682 				VOP_LOCK(*vpp, LK_DOWNGRADE, td);
683 		}
684 		return (error);
685 	}
686 #else
687 	if (!error)
688 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
689 #endif
690 
691 	if (error == ENOENT)
692 		return (error);
693 
694 	vp = *vpp;
695 	vpid = vp->v_id;
696 	cnp->cn_flags &= ~PDIRUNLOCK;
697 	if (dvp == vp) {   /* lookup on "." */
698 		VREF(vp);
699 		error = 0;
700 	} else if (flags & ISDOTDOT) {
701 		VOP_UNLOCK(dvp, 0, td);
702 		cnp->cn_flags |= PDIRUNLOCK;
703 #ifdef LOOKUP_SHARED
704 		if ((flags & ISLASTCN) && (flags & LOCKSHARED))
705 			error = vget(vp, LK_SHARED, td);
706 		else
707 			error = vget(vp, LK_EXCLUSIVE, td);
708 #else
709 		error = vget(vp, LK_EXCLUSIVE, td);
710 #endif
711 
712 		if (!error && lockparent && (flags & ISLASTCN)) {
713 			if ((error = vn_lock(dvp, LK_EXCLUSIVE, td)) == 0)
714 				cnp->cn_flags &= ~PDIRUNLOCK;
715 		}
716 	} else {
717 #ifdef LOOKUP_SHARED
718 		if ((flags & ISLASTCN) && (flags & LOCKSHARED))
719 			error = vget(vp, LK_SHARED, td);
720 		else
721 			error = vget(vp, LK_EXCLUSIVE, td);
722 #else
723 		error = vget(vp, LK_EXCLUSIVE, td);
724 #endif
725 		if (!lockparent || error || !(flags & ISLASTCN)) {
726 			VOP_UNLOCK(dvp, 0, td);
727 			cnp->cn_flags |= PDIRUNLOCK;
728 		}
729 	}
730 	/*
731 	 * Check that the capability number did not change
732 	 * while we were waiting for the lock.
733 	 */
734 	if (!error) {
735 		if (vpid == vp->v_id)
736 			return (0);
737 		vput(vp);
738 		if (lockparent && dvp != vp && (flags & ISLASTCN)) {
739 			VOP_UNLOCK(dvp, 0, td);
740 			cnp->cn_flags |= PDIRUNLOCK;
741 		}
742 	}
743 	if (cnp->cn_flags & PDIRUNLOCK) {
744 		error = vn_lock(dvp, LK_EXCLUSIVE, td);
745 		if (error)
746 			return (error);
747 		cnp->cn_flags &= ~PDIRUNLOCK;
748 	}
749 #ifdef LOOKUP_SHARED
750 	error = VOP_CACHEDLOOKUP(dvp, vpp, cnp);
751 
752 	if (!error) {
753 		int flock = 0;
754 
755 		flock = VOP_ISLOCKED(*vpp, td);
756 		if (flock != LK_EXCLUSIVE) {
757 			if (flock == 0) {
758 				if ((flags & ISLASTCN) && (flags & LOCKSHARED))
759 					VOP_LOCK(*vpp, LK_SHARED, td);
760 				else
761 					VOP_LOCK(*vpp, LK_EXCLUSIVE, td);
762 			}
763 		} else if ((flags & ISLASTCN) && (flags & LOCKSHARED))
764 			VOP_LOCK(*vpp, LK_DOWNGRADE, td);
765 	}
766 
767 	return (error);
768 #else
769 	return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
770 #endif
771 }
772 
773 
774 #ifndef _SYS_SYSPROTO_H_
775 struct  __getcwd_args {
776 	u_char	*buf;
777 	u_int	buflen;
778 };
779 #endif
780 
781 /*
782  * XXX All of these sysctls would probably be more productive dead.
783  */
784 static int disablecwd;
785 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
786    "Disable the getcwd syscall");
787 
788 /* Various statistics for the getcwd syscall */
789 static u_long numcwdcalls; STATNODE(CTLFLAG_RD, numcwdcalls, &numcwdcalls);
790 static u_long numcwdfail1; STATNODE(CTLFLAG_RD, numcwdfail1, &numcwdfail1);
791 static u_long numcwdfail2; STATNODE(CTLFLAG_RD, numcwdfail2, &numcwdfail2);
792 static u_long numcwdfail3; STATNODE(CTLFLAG_RD, numcwdfail3, &numcwdfail3);
793 static u_long numcwdfail4; STATNODE(CTLFLAG_RD, numcwdfail4, &numcwdfail4);
794 static u_long numcwdfound; STATNODE(CTLFLAG_RD, numcwdfound, &numcwdfound);
795 
796 /* Implementation of the getcwd syscall */
797 int
798 __getcwd(td, uap)
799 	struct thread *td;
800 	struct __getcwd_args *uap;
801 {
802 
803 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
804 }
805 
806 int
807 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
808 {
809 	char *bp, *tmpbuf;
810 	int error, i, slash_prefixed;
811 	struct filedesc *fdp;
812 	struct namecache *ncp;
813 	struct vnode *vp;
814 
815 	numcwdcalls++;
816 	if (disablecwd)
817 		return (ENODEV);
818 	if (buflen < 2)
819 		return (EINVAL);
820 	if (buflen > MAXPATHLEN)
821 		buflen = MAXPATHLEN;
822 	error = 0;
823 	tmpbuf = bp = malloc(buflen, M_TEMP, M_WAITOK);
824 	bp += buflen - 1;
825 	*bp = '\0';
826 	fdp = td->td_proc->p_fd;
827 	slash_prefixed = 0;
828 	FILEDESC_LOCK(fdp);
829 	for (vp = fdp->fd_cdir; vp != fdp->fd_rdir && vp != rootvnode;) {
830 		if (vp->v_vflag & VV_ROOT) {
831 			if (vp->v_mount == NULL) {	/* forced unmount */
832 				FILEDESC_UNLOCK(fdp);
833 				free(tmpbuf, M_TEMP);
834 				return (EBADF);
835 			}
836 			vp = vp->v_mount->mnt_vnodecovered;
837 			continue;
838 		}
839 		if (vp->v_dd->v_id != vp->v_ddid) {
840 			FILEDESC_UNLOCK(fdp);
841 			numcwdfail1++;
842 			free(tmpbuf, M_TEMP);
843 			return (ENOTDIR);
844 		}
845 		CACHE_LOCK();
846 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
847 		if (!ncp) {
848 			numcwdfail2++;
849 			CACHE_UNLOCK();
850 			FILEDESC_UNLOCK(fdp);
851 			free(tmpbuf, M_TEMP);
852 			return (ENOENT);
853 		}
854 		if (ncp->nc_dvp != vp->v_dd) {
855 			numcwdfail3++;
856 			CACHE_UNLOCK();
857 			FILEDESC_UNLOCK(fdp);
858 			free(tmpbuf, M_TEMP);
859 			return (EBADF);
860 		}
861 		for (i = ncp->nc_nlen - 1; i >= 0; i--) {
862 			if (bp == tmpbuf) {
863 				numcwdfail4++;
864 				CACHE_UNLOCK();
865 				FILEDESC_UNLOCK(fdp);
866 				free(tmpbuf, M_TEMP);
867 				return (ENOMEM);
868 			}
869 			*--bp = ncp->nc_name[i];
870 		}
871 		if (bp == tmpbuf) {
872 			numcwdfail4++;
873 			CACHE_UNLOCK();
874 			FILEDESC_UNLOCK(fdp);
875 			free(tmpbuf, M_TEMP);
876 			return (ENOMEM);
877 		}
878 		*--bp = '/';
879 		slash_prefixed = 1;
880 		vp = vp->v_dd;
881 		CACHE_UNLOCK();
882 	}
883 	FILEDESC_UNLOCK(fdp);
884 	if (!slash_prefixed) {
885 		if (bp == tmpbuf) {
886 			numcwdfail4++;
887 			free(tmpbuf, M_TEMP);
888 			return (ENOMEM);
889 		}
890 		*--bp = '/';
891 	}
892 	numcwdfound++;
893 	if (bufseg == UIO_SYSSPACE)
894 		bcopy(bp, buf, strlen(bp) + 1);
895 	else
896 		error = copyout(bp, buf, strlen(bp) + 1);
897 	free(tmpbuf, M_TEMP);
898 	return (error);
899 }
900 
901 /*
902  * Thus begins the fullpath magic.
903  */
904 
905 #undef STATNODE
906 #define STATNODE(name)							\
907 	static u_int name;						\
908 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
909 
910 static int disablefullpath;
911 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
912 	"Disable the vn_fullpath function");
913 
914 STATNODE(numfullpathcalls);
915 STATNODE(numfullpathfail1);
916 STATNODE(numfullpathfail2);
917 STATNODE(numfullpathfail3);
918 STATNODE(numfullpathfail4);
919 STATNODE(numfullpathfound);
920 
921 /*
922  * Retrieve the full filesystem path that correspond to a vnode from the name
923  * cache (if available)
924  */
925 int
926 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
927 {
928 	char *bp, *buf;
929 	int i, slash_prefixed;
930 	struct filedesc *fdp;
931 	struct namecache *ncp;
932 	struct vnode *vp;
933 
934 	numfullpathcalls++;
935 	if (disablefullpath)
936 		return (ENODEV);
937 	if (vn == NULL)
938 		return (EINVAL);
939 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
940 	bp = buf + MAXPATHLEN - 1;
941 	*bp = '\0';
942 	fdp = td->td_proc->p_fd;
943 	slash_prefixed = 0;
944 	ASSERT_VOP_LOCKED(vn, "vn_fullpath");
945 	FILEDESC_LOCK(fdp);
946 	for (vp = vn; vp != fdp->fd_rdir && vp != rootvnode;) {
947 		if (vp->v_vflag & VV_ROOT) {
948 			if (vp->v_mount == NULL) {	/* forced unmount */
949 				FILEDESC_UNLOCK(fdp);
950 				free(buf, M_TEMP);
951 				return (EBADF);
952 			}
953 			vp = vp->v_mount->mnt_vnodecovered;
954 			continue;
955 		}
956 		if (vp != vn && vp->v_dd->v_id != vp->v_ddid) {
957 			FILEDESC_UNLOCK(fdp);
958 			free(buf, M_TEMP);
959 			numfullpathfail1++;
960 			return (ENOTDIR);
961 		}
962 		CACHE_LOCK();
963 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
964 		if (!ncp) {
965 			numfullpathfail2++;
966 			CACHE_UNLOCK();
967 			FILEDESC_UNLOCK(fdp);
968 			free(buf, M_TEMP);
969 			return (ENOENT);
970 		}
971 		if (vp != vn && ncp->nc_dvp != vp->v_dd) {
972 			numfullpathfail3++;
973 			CACHE_UNLOCK();
974 			FILEDESC_UNLOCK(fdp);
975 			free(buf, M_TEMP);
976 			return (EBADF);
977 		}
978 		for (i = ncp->nc_nlen - 1; i >= 0; i--) {
979 			if (bp == buf) {
980 				numfullpathfail4++;
981 				CACHE_UNLOCK();
982 				FILEDESC_UNLOCK(fdp);
983 				free(buf, M_TEMP);
984 				return (ENOMEM);
985 			}
986 			*--bp = ncp->nc_name[i];
987 		}
988 		if (bp == buf) {
989 			numfullpathfail4++;
990 			CACHE_UNLOCK();
991 			FILEDESC_UNLOCK(fdp);
992 			free(buf, M_TEMP);
993 			return (ENOMEM);
994 		}
995 		*--bp = '/';
996 		slash_prefixed = 1;
997 		vp = ncp->nc_dvp;
998 		CACHE_UNLOCK();
999 	}
1000 	if (!slash_prefixed) {
1001 		if (bp == buf) {
1002 			numfullpathfail4++;
1003 			FILEDESC_UNLOCK(fdp);
1004 			free(buf, M_TEMP);
1005 			return (ENOMEM);
1006 		}
1007 		*--bp = '/';
1008 	}
1009 	FILEDESC_UNLOCK(fdp);
1010 	numfullpathfound++;
1011 	*retbuf = bp;
1012 	*freebuf = buf;
1013 	return (0);
1014 }
1015