xref: /freebsd/sys/kern/vfs_cache.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
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  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vfs_cache.c	8.5 (Berkeley) 3/22/95
37  * $FreeBSD$
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/sysctl.h>
46 #include <sys/mount.h>
47 #include <sys/vnode.h>
48 #include <sys/namei.h>
49 #include <sys/malloc.h>
50 #include <sys/syscallsubr.h>
51 #include <sys/sysproto.h>
52 #include <sys/proc.h>
53 #include <sys/filedesc.h>
54 #include <sys/fnv_hash.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 int	doingcache = 1;		/* 1 => enable the cache */
113 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
114 
115 /* Export size information to userland */
116 SYSCTL_INT(_debug, OID_AUTO, vnsize, CTLFLAG_RD, 0, sizeof(struct vnode), "");
117 SYSCTL_INT(_debug, OID_AUTO, ncsize, CTLFLAG_RD, 0, sizeof(struct namecache), "");
118 
119 /*
120  * The new name cache statistics
121  */
122 SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
123 #define STATNODE(mode, name, var) \
124 	SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
125 STATNODE(CTLFLAG_RD, numneg, &numneg);
126 STATNODE(CTLFLAG_RD, numcache, &numcache);
127 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
128 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
129 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
130 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
131 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
132 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
133 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
134 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
135 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
136 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
137 
138 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD, &nchstats,
139         sizeof(nchstats), "LU", "VFS cache effectiveness statistics");
140 
141 
142 
143 static void cache_zap(struct namecache *ncp);
144 
145 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
146 
147 /*
148  * Flags in namecache.nc_flag
149  */
150 #define NCF_WHITE	1
151 
152 /*
153  * Grab an atomic snapshot of the name cache hash chain lengths
154  */
155 SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
156 
157 static int
158 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
159 {
160 	int error;
161 	struct nchashhead *ncpp;
162 	struct namecache *ncp;
163 	int n_nchash;
164 	int count;
165 
166 	n_nchash = nchash + 1;	/* nchash is max index, not count */
167 	if (!req->oldptr)
168 		return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
169 
170 	/* Scan hash tables for applicable entries */
171 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
172 		count = 0;
173 		LIST_FOREACH(ncp, ncpp, nc_hash) {
174 			count++;
175 		}
176 		error = SYSCTL_OUT(req, &count, sizeof(count));
177 		if (error)
178 			return (error);
179 	}
180 	return (0);
181 }
182 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD,
183 	0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths");
184 
185 static int
186 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
187 {
188 	int error;
189 	struct nchashhead *ncpp;
190 	struct namecache *ncp;
191 	int n_nchash;
192 	int count, maxlength, used, pct;
193 
194 	if (!req->oldptr)
195 		return SYSCTL_OUT(req, 0, 4 * sizeof(int));
196 
197 	n_nchash = nchash + 1;	/* nchash is max index, not count */
198 	used = 0;
199 	maxlength = 0;
200 
201 	/* Scan hash tables for applicable entries */
202 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
203 		count = 0;
204 		LIST_FOREACH(ncp, ncpp, nc_hash) {
205 			count++;
206 		}
207 		if (count)
208 			used++;
209 		if (maxlength < count)
210 			maxlength = count;
211 	}
212 	n_nchash = nchash + 1;
213 	pct = (used * 100 * 100) / n_nchash;
214 	error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
215 	if (error)
216 		return (error);
217 	error = SYSCTL_OUT(req, &used, sizeof(used));
218 	if (error)
219 		return (error);
220 	error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
221 	if (error)
222 		return (error);
223 	error = SYSCTL_OUT(req, &pct, sizeof(pct));
224 	if (error)
225 		return (error);
226 	return (0);
227 }
228 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD,
229 	0, 0, sysctl_debug_hashstat_nchash, "I", "nchash chain lengths");
230 
231 /*
232  * cache_zap():
233  *
234  *   Removes a namecache entry from cache, whether it contains an actual
235  *   pointer to a vnode or if it is just a negative cache entry.
236  */
237 static void
238 cache_zap(ncp)
239 	struct namecache *ncp;
240 {
241 	LIST_REMOVE(ncp, nc_hash);
242 	LIST_REMOVE(ncp, nc_src);
243 	if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
244 		vdrop(ncp->nc_dvp);
245 		numcachehv--;
246 	}
247 	if (ncp->nc_vp) {
248 		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
249 	} else {
250 		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
251 		numneg--;
252 	}
253 	numcache--;
254 	free(ncp, M_VFSCACHE);
255 }
256 
257 /*
258  * cache_leaf_test()
259  *
260  *      Test whether this (directory) vnode's namei cache entry contains
261  *      subdirectories or not.  Used to determine whether the directory is
262  *      a leaf in the namei cache or not.  Note: the directory may still
263  *      contain files in the namei cache.
264  *
265  *      Returns 0 if the directory is a leaf, -1 if it isn't.
266  */
267 int
268 cache_leaf_test(struct vnode *vp)
269 {
270 	struct namecache *ncpc;
271 
272 	for (ncpc = LIST_FIRST(&vp->v_cache_src);
273 	     ncpc != NULL;
274 	     ncpc = LIST_NEXT(ncpc, nc_src)
275 	) {
276 		if (ncpc->nc_vp != NULL && ncpc->nc_vp->v_type == VDIR)
277 			return(-1);
278 	}
279 	return(0);
280 }
281 
282 /*
283  * Lookup an entry in the cache
284  *
285  * Lookup is called with dvp pointing to the directory to search,
286  * cnp pointing to the name of the entry being sought. If the lookup
287  * succeeds, the vnode is returned in *vpp, and a status of -1 is
288  * returned. If the lookup determines that the name does not exist
289  * (negative cacheing), a status of ENOENT is returned. If the lookup
290  * fails, a status of zero is returned.
291  */
292 
293 int
294 cache_lookup(dvp, vpp, cnp)
295 	struct vnode *dvp;
296 	struct vnode **vpp;
297 	struct componentname *cnp;
298 {
299 	struct namecache *ncp;
300 	u_int32_t hash;
301 
302 	if (!doingcache) {
303 		cnp->cn_flags &= ~MAKEENTRY;
304 		return (0);
305 	}
306 
307 	numcalls++;
308 
309 	if (cnp->cn_nameptr[0] == '.') {
310 		if (cnp->cn_namelen == 1) {
311 			*vpp = dvp;
312 			dothits++;
313 			return (-1);
314 		}
315 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
316 			dotdothits++;
317 			if (dvp->v_dd->v_id != dvp->v_ddid ||
318 			    (cnp->cn_flags & MAKEENTRY) == 0) {
319 				dvp->v_ddid = 0;
320 				return (0);
321 			}
322 			*vpp = dvp->v_dd;
323 			return (-1);
324 		}
325 	}
326 
327 	hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
328 	hash = fnv_32_buf(&dvp->v_id, sizeof(dvp->v_id), hash);
329 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
330 		numchecks++;
331 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
332 		    !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
333 			break;
334 	}
335 
336 	/* We failed to find an entry */
337 	if (ncp == 0) {
338 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
339 			nummisszap++;
340 		} else {
341 			nummiss++;
342 		}
343 		nchstats.ncs_miss++;
344 		return (0);
345 	}
346 
347 	/* We don't want to have an entry, so dump it */
348 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
349 		numposzaps++;
350 		nchstats.ncs_badhits++;
351 		cache_zap(ncp);
352 		return (0);
353 	}
354 
355 	/* We found a "positive" match, return the vnode */
356         if (ncp->nc_vp) {
357 		numposhits++;
358 		nchstats.ncs_goodhits++;
359 		*vpp = ncp->nc_vp;
360 		return (-1);
361 	}
362 
363 	/* We found a negative match, and want to create it, so purge */
364 	if (cnp->cn_nameiop == CREATE) {
365 		numnegzaps++;
366 		nchstats.ncs_badhits++;
367 		cache_zap(ncp);
368 		return (0);
369 	}
370 
371 	numneghits++;
372 	/*
373 	 * We found a "negative" match, so we shift it to the end of
374 	 * the "negative" cache entries queue to satisfy LRU.  Also,
375 	 * check to see if the entry is a whiteout; indicate this to
376 	 * the componentname, if so.
377 	 */
378 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
379 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
380 	nchstats.ncs_neghits++;
381 	if (ncp->nc_flag & NCF_WHITE)
382 		cnp->cn_flags |= ISWHITEOUT;
383 	return (ENOENT);
384 }
385 
386 /*
387  * Add an entry to the cache.
388  */
389 void
390 cache_enter(dvp, vp, cnp)
391 	struct vnode *dvp;
392 	struct vnode *vp;
393 	struct componentname *cnp;
394 {
395 	struct namecache *ncp;
396 	struct nchashhead *ncpp;
397 	u_int32_t hash;
398 	int len;
399 
400 	if (!doingcache)
401 		return;
402 
403 	if (cnp->cn_nameptr[0] == '.') {
404 		if (cnp->cn_namelen == 1) {
405 			return;
406 		}
407 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
408 			if (vp) {
409 				dvp->v_dd = vp;
410 				dvp->v_ddid = vp->v_id;
411 			} else {
412 				dvp->v_dd = dvp;
413 				dvp->v_ddid = 0;
414 			}
415 			return;
416 		}
417 	}
418 
419 	ncp = (struct namecache *)
420 		malloc(sizeof *ncp + cnp->cn_namelen, M_VFSCACHE, M_WAITOK);
421 	bzero((char *)ncp, sizeof *ncp);
422 	numcache++;
423 	if (!vp) {
424 		numneg++;
425 		ncp->nc_flag = cnp->cn_flags & ISWHITEOUT ? NCF_WHITE : 0;
426 	} else if (vp->v_type == VDIR) {
427 		vp->v_dd = dvp;
428 		vp->v_ddid = dvp->v_id;
429 	}
430 
431 	/*
432 	 * Set the rest of the namecache entry elements, calculate it's
433 	 * hash key and insert it into the appropriate chain within
434 	 * the cache entries table.
435 	 */
436 	ncp->nc_vp = vp;
437 	ncp->nc_dvp = dvp;
438 	len = ncp->nc_nlen = cnp->cn_namelen;
439 	hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
440 	bcopy(cnp->cn_nameptr, ncp->nc_name, len);
441 	hash = fnv_32_buf(&dvp->v_id, sizeof(dvp->v_id), hash);
442 	ncpp = NCHHASH(hash);
443 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
444 	if (LIST_EMPTY(&dvp->v_cache_src)) {
445 		vhold(dvp);
446 		numcachehv++;
447 	}
448 	LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
449 	/*
450 	 * If the entry is "negative", we place it into the
451 	 * "negative" cache queue, otherwise, we place it into the
452 	 * destination vnode's cache entries queue.
453 	 */
454 	if (vp) {
455 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
456 	} else {
457 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
458 	}
459 	if (numneg * ncnegfactor > numcache) {
460 		ncp = TAILQ_FIRST(&ncneg);
461 		cache_zap(ncp);
462 	}
463 }
464 
465 /*
466  * Name cache initialization, from vfs_init() when we are booting
467  */
468 static void
469 nchinit(void *dummy __unused)
470 {
471 
472 	TAILQ_INIT(&ncneg);
473 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
474 }
475 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL)
476 
477 
478 /*
479  * Invalidate all entries to a particular vnode.
480  *
481  * Remove all entries in the namecache relating to this vnode and
482  * change the v_id.  We take the v_id from a global counter, since
483  * it becomes a handy sequence number in crash-dumps that way.
484  * No valid vnode will ever have (v_id == 0).
485  *
486  * XXX: Only time and the size of v_id prevents this from failing:
487  * XXX: In theory we should hunt down all (struct vnode*, v_id)
488  * XXX: soft references and nuke them, at least on the global
489  * XXX: v_id wraparound.  The period of resistance can be extended
490  * XXX: by incrementing each vnodes v_id individually instead of
491  * XXX: using the global v_id.
492  */
493 
494 void
495 cache_purge(vp)
496 	struct vnode *vp;
497 {
498 	static u_long nextid;
499 
500 	while (!LIST_EMPTY(&vp->v_cache_src))
501 		cache_zap(LIST_FIRST(&vp->v_cache_src));
502 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
503 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
504 
505 	do
506 		nextid++;
507 	while (nextid == vp->v_id || !nextid);
508 	vp->v_id = nextid;
509 	vp->v_dd = vp;
510 	vp->v_ddid = 0;
511 }
512 
513 /*
514  * Flush all entries referencing a particular filesystem.
515  *
516  * Since we need to check it anyway, we will flush all the invalid
517  * entries at the same time.
518  */
519 void
520 cache_purgevfs(mp)
521 	struct mount *mp;
522 {
523 	struct nchashhead *ncpp;
524 	struct namecache *ncp, *nnp;
525 
526 	/* Scan hash tables for applicable entries */
527 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
528 		for (ncp = LIST_FIRST(ncpp); ncp != 0; ncp = nnp) {
529 			nnp = LIST_NEXT(ncp, nc_hash);
530 			if (ncp->nc_dvp->v_mount == mp) {
531 				cache_zap(ncp);
532 			}
533 		}
534 	}
535 }
536 
537 /*
538  * Perform canonical checks and cache lookup and pass on to filesystem
539  * through the vop_cachedlookup only if needed.
540  */
541 
542 int
543 vfs_cache_lookup(ap)
544 	struct vop_lookup_args /* {
545 		struct vnode *a_dvp;
546 		struct vnode **a_vpp;
547 		struct componentname *a_cnp;
548 	} */ *ap;
549 {
550 	struct vnode *dvp, *vp;
551 	int lockparent;
552 	int error;
553 	struct vnode **vpp = ap->a_vpp;
554 	struct componentname *cnp = ap->a_cnp;
555 	struct ucred *cred = cnp->cn_cred;
556 	int flags = cnp->cn_flags;
557 	struct thread *td = cnp->cn_thread;
558 	u_long vpid;	/* capability number of vnode */
559 
560 	*vpp = NULL;
561 	dvp = ap->a_dvp;
562 	lockparent = flags & LOCKPARENT;
563 
564 	if (dvp->v_type != VDIR)
565                 return (ENOTDIR);
566 
567 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
568 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
569 		return (EROFS);
570 
571 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
572 
573 	if (error)
574 		return (error);
575 
576 	error = cache_lookup(dvp, vpp, cnp);
577 
578 #ifdef LOOKUP_SHARED
579 	if (!error) {
580 		/* We do this because the rest of the system now expects to get
581 		 * a shared lock, which is later upgraded if LOCKSHARED is not
582 		 * set.  We have so many cases here because of bugs that yield
583 		 * inconsistant lock states.  This all badly needs to be fixed
584 		 */
585 		error = VOP_CACHEDLOOKUP(dvp, vpp, cnp);
586 		if (!error) {
587 			int flock;
588 
589 			flock = VOP_ISLOCKED(*vpp, td);
590 			if (flock != LK_EXCLUSIVE) {
591 				if (flock == 0) {
592 					if ((flags & ISLASTCN) &&
593 					    (flags & LOCKSHARED))
594 						VOP_LOCK(*vpp, LK_SHARED, td);
595 					else
596 						VOP_LOCK(*vpp, LK_EXCLUSIVE, td);
597 				}
598 			} else if ((flags & ISLASTCN) && (flags & LOCKSHARED))
599 				VOP_LOCK(*vpp, LK_DOWNGRADE, td);
600 		}
601 		return (error);
602 	}
603 #else
604 	if (!error)
605 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
606 #endif
607 
608 	if (error == ENOENT)
609 		return (error);
610 
611 	vp = *vpp;
612 	vpid = vp->v_id;
613 	cnp->cn_flags &= ~PDIRUNLOCK;
614 	if (dvp == vp) {   /* lookup on "." */
615 		VREF(vp);
616 		error = 0;
617 	} else if (flags & ISDOTDOT) {
618 		VOP_UNLOCK(dvp, 0, td);
619 		cnp->cn_flags |= PDIRUNLOCK;
620 #ifdef LOOKUP_SHARED
621 		if ((flags & ISLASTCN) && (flags & LOCKSHARED))
622 			error = vget(vp, LK_SHARED, td);
623 		else
624 			error = vget(vp, LK_EXCLUSIVE, td);
625 #else
626 		error = vget(vp, LK_EXCLUSIVE, td);
627 #endif
628 
629 		if (!error && lockparent && (flags & ISLASTCN)) {
630 			if ((error = vn_lock(dvp, LK_EXCLUSIVE, td)) == 0)
631 				cnp->cn_flags &= ~PDIRUNLOCK;
632 		}
633 	} else {
634 #ifdef LOOKUP_SHARED
635 		if ((flags & ISLASTCN) && (flags & LOCKSHARED))
636 			error = vget(vp, LK_SHARED, td);
637 		else
638 			error = vget(vp, LK_EXCLUSIVE, td);
639 #else
640 		error = vget(vp, LK_EXCLUSIVE, td);
641 #endif
642 		if (!lockparent || error || !(flags & ISLASTCN)) {
643 			VOP_UNLOCK(dvp, 0, td);
644 			cnp->cn_flags |= PDIRUNLOCK;
645 		}
646 	}
647 	/*
648 	 * Check that the capability number did not change
649 	 * while we were waiting for the lock.
650 	 */
651 	if (!error) {
652 		if (vpid == vp->v_id)
653 			return (0);
654 		vput(vp);
655 		if (lockparent && dvp != vp && (flags & ISLASTCN)) {
656 			VOP_UNLOCK(dvp, 0, td);
657 			cnp->cn_flags |= PDIRUNLOCK;
658 		}
659 	}
660 	if (cnp->cn_flags & PDIRUNLOCK) {
661 		error = vn_lock(dvp, LK_EXCLUSIVE, td);
662 		if (error)
663 			return (error);
664 		cnp->cn_flags &= ~PDIRUNLOCK;
665 	}
666 #ifdef LOOKUP_SHARED
667 	error = VOP_CACHEDLOOKUP(dvp, vpp, cnp);
668 
669 	if (!error) {
670 		int flock = 0;
671 
672 		flock = VOP_ISLOCKED(*vpp, td);
673 		if (flock != LK_EXCLUSIVE) {
674 			if (flock == 0) {
675 				if ((flags & ISLASTCN) && (flags & LOCKSHARED))
676 					VOP_LOCK(*vpp, LK_SHARED, td);
677 				else
678 					VOP_LOCK(*vpp, LK_EXCLUSIVE, td);
679 			}
680 		} else if ((flags & ISLASTCN) && (flags & LOCKSHARED))
681 			VOP_LOCK(*vpp, LK_DOWNGRADE, td);
682 	}
683 
684 	return (error);
685 #else
686 	return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
687 #endif
688 }
689 
690 
691 #ifndef _SYS_SYSPROTO_H_
692 struct  __getcwd_args {
693 	u_char	*buf;
694 	u_int	buflen;
695 };
696 #endif
697 
698 /*
699  * XXX All of these sysctls would probably be more productive dead.
700  */
701 static int disablecwd;
702 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
703    "Disable the getcwd syscall");
704 
705 /* Various statistics for the getcwd syscall */
706 static u_long numcwdcalls; STATNODE(CTLFLAG_RD, numcwdcalls, &numcwdcalls);
707 static u_long numcwdfail1; STATNODE(CTLFLAG_RD, numcwdfail1, &numcwdfail1);
708 static u_long numcwdfail2; STATNODE(CTLFLAG_RD, numcwdfail2, &numcwdfail2);
709 static u_long numcwdfail3; STATNODE(CTLFLAG_RD, numcwdfail3, &numcwdfail3);
710 static u_long numcwdfail4; STATNODE(CTLFLAG_RD, numcwdfail4, &numcwdfail4);
711 static u_long numcwdfound; STATNODE(CTLFLAG_RD, numcwdfound, &numcwdfound);
712 
713 /* Implementation of the getcwd syscall */
714 int
715 __getcwd(td, uap)
716 	struct thread *td;
717 	struct __getcwd_args *uap;
718 {
719 
720 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
721 }
722 
723 int
724 kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
725 {
726 	char *bp, *tmpbuf;
727 	int error, i, slash_prefixed;
728 	struct filedesc *fdp;
729 	struct namecache *ncp;
730 	struct vnode *vp;
731 
732 	numcwdcalls++;
733 	if (disablecwd)
734 		return (ENODEV);
735 	if (buflen < 2)
736 		return (EINVAL);
737 	if (buflen > MAXPATHLEN)
738 		buflen = MAXPATHLEN;
739 	error = 0;
740 	tmpbuf = bp = malloc(buflen, M_TEMP, M_WAITOK);
741 	bp += buflen - 1;
742 	*bp = '\0';
743 	fdp = td->td_proc->p_fd;
744 	slash_prefixed = 0;
745 	FILEDESC_LOCK(fdp);
746 	mp_fixme("No vnode locking done!");
747 	for (vp = fdp->fd_cdir; vp != fdp->fd_rdir && vp != rootvnode;) {
748 		if (vp->v_vflag & VV_ROOT) {
749 			if (vp->v_mount == NULL) {	/* forced unmount */
750 				FILEDESC_UNLOCK(fdp);
751 				free(tmpbuf, M_TEMP);
752 				return (EBADF);
753 			}
754 			vp = vp->v_mount->mnt_vnodecovered;
755 			continue;
756 		}
757 		if (vp->v_dd->v_id != vp->v_ddid) {
758 			FILEDESC_UNLOCK(fdp);
759 			numcwdfail1++;
760 			free(tmpbuf, M_TEMP);
761 			return (ENOTDIR);
762 		}
763 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
764 		if (!ncp) {
765 			FILEDESC_UNLOCK(fdp);
766 			numcwdfail2++;
767 			free(tmpbuf, M_TEMP);
768 			return (ENOENT);
769 		}
770 		if (ncp->nc_dvp != vp->v_dd) {
771 			FILEDESC_UNLOCK(fdp);
772 			numcwdfail3++;
773 			free(tmpbuf, M_TEMP);
774 			return (EBADF);
775 		}
776 		for (i = ncp->nc_nlen - 1; i >= 0; i--) {
777 			if (bp == tmpbuf) {
778 				FILEDESC_UNLOCK(fdp);
779 				numcwdfail4++;
780 				free(tmpbuf, M_TEMP);
781 				return (ENOMEM);
782 			}
783 			*--bp = ncp->nc_name[i];
784 		}
785 		if (bp == tmpbuf) {
786 			FILEDESC_UNLOCK(fdp);
787 			numcwdfail4++;
788 			free(tmpbuf, M_TEMP);
789 			return (ENOMEM);
790 		}
791 		*--bp = '/';
792 		slash_prefixed = 1;
793 		vp = vp->v_dd;
794 	}
795 	FILEDESC_UNLOCK(fdp);
796 	if (!slash_prefixed) {
797 		if (bp == tmpbuf) {
798 			numcwdfail4++;
799 			free(tmpbuf, M_TEMP);
800 			return (ENOMEM);
801 		}
802 		*--bp = '/';
803 	}
804 	numcwdfound++;
805 	if (bufseg == UIO_SYSSPACE)
806 		bcopy(bp, buf, strlen(bp) + 1);
807 	else
808 		error = copyout(bp, buf, strlen(bp) + 1);
809 	free(tmpbuf, M_TEMP);
810 	return (error);
811 }
812 
813 /*
814  * Thus begins the fullpath magic.
815  */
816 
817 #undef STATNODE
818 #define STATNODE(name)							\
819 	static u_int name;						\
820 	SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
821 
822 static int disablefullpath;
823 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
824 	"Disable the vn_fullpath function");
825 
826 STATNODE(numfullpathcalls);
827 STATNODE(numfullpathfail1);
828 STATNODE(numfullpathfail2);
829 STATNODE(numfullpathfail3);
830 STATNODE(numfullpathfail4);
831 STATNODE(numfullpathfound);
832 
833 /*
834  * Retrieve the full filesystem path that correspond to a vnode from the name
835  * cache (if available)
836  */
837 int
838 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
839 {
840 	char *bp, *buf;
841 	int i, slash_prefixed;
842 	struct filedesc *fdp;
843 	struct namecache *ncp;
844 	struct vnode *vp;
845 
846 	numfullpathcalls++;
847 	if (disablefullpath)
848 		return (ENODEV);
849 	if (vn == NULL)
850 		return (EINVAL);
851 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
852 	bp = buf + MAXPATHLEN - 1;
853 	*bp = '\0';
854 	fdp = td->td_proc->p_fd;
855 	slash_prefixed = 0;
856 	FILEDESC_LOCK(fdp);
857 	for (vp = vn; vp != fdp->fd_rdir && vp != rootvnode;) {
858 		ASSERT_VOP_LOCKED(vp, "vn_fullpath");
859 		if (vp->v_vflag & VV_ROOT) {
860 			if (vp->v_mount == NULL) {	/* forced unmount */
861 				FILEDESC_UNLOCK(fdp);
862 				free(buf, M_TEMP);
863 				return (EBADF);
864 			}
865 			vp = vp->v_mount->mnt_vnodecovered;
866 			continue;
867 		}
868 		if (vp != vn && vp->v_dd->v_id != vp->v_ddid) {
869 			FILEDESC_UNLOCK(fdp);
870 			numfullpathfail1++;
871 			free(buf, M_TEMP);
872 			return (ENOTDIR);
873 		}
874 		ncp = TAILQ_FIRST(&vp->v_cache_dst);
875 		if (!ncp) {
876 			FILEDESC_UNLOCK(fdp);
877 			numfullpathfail2++;
878 			free(buf, M_TEMP);
879 			return (ENOENT);
880 		}
881 		if (vp != vn && ncp->nc_dvp != vp->v_dd) {
882 			FILEDESC_UNLOCK(fdp);
883 			numfullpathfail3++;
884 			free(buf, M_TEMP);
885 			return (EBADF);
886 		}
887 		for (i = ncp->nc_nlen - 1; i >= 0; i--) {
888 			if (bp == buf) {
889 				FILEDESC_UNLOCK(fdp);
890 				numfullpathfail4++;
891 				free(buf, M_TEMP);
892 				return (ENOMEM);
893 			}
894 			*--bp = ncp->nc_name[i];
895 		}
896 		if (bp == buf) {
897 			FILEDESC_UNLOCK(fdp);
898 			numfullpathfail4++;
899 			free(buf, M_TEMP);
900 			return (ENOMEM);
901 		}
902 		*--bp = '/';
903 		slash_prefixed = 1;
904 		vp = ncp->nc_dvp;
905 	}
906 	if (!slash_prefixed) {
907 		if (bp == buf) {
908 			FILEDESC_UNLOCK(fdp);
909 			numfullpathfail4++;
910 			free(buf, M_TEMP);
911 			return (ENOMEM);
912 		}
913 		*--bp = '/';
914 	}
915 	FILEDESC_UNLOCK(fdp);
916 	numfullpathfound++;
917 	*retbuf = bp;
918 	*freebuf = buf;
919 	return (0);
920 }
921