xref: /freebsd/sys/kern/vfs_cache.c (revision 86c9d9918f1db7cdd968b60f8902466887bcd9e9)
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 "opt_ktrace.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/counter.h>
43 #include <sys/filedesc.h>
44 #include <sys/fnv_hash.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/fcntl.h>
49 #include <sys/mount.h>
50 #include <sys/namei.h>
51 #include <sys/proc.h>
52 #include <sys/rwlock.h>
53 #include <sys/sdt.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysproto.h>
57 #include <sys/vnode.h>
58 #ifdef KTRACE
59 #include <sys/ktrace.h>
60 #endif
61 
62 #include <vm/uma.h>
63 
64 SDT_PROVIDER_DECLARE(vfs);
65 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
66     "struct vnode *");
67 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
68     "char *");
69 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *");
70 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *",
71     "char *", "struct vnode *");
72 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *");
73 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int",
74     "struct vnode *", "char *");
75 SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *",
76     "struct vnode *");
77 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative,
78     "struct vnode *", "char *");
79 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
80     "char *");
81 SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *");
82 SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *");
83 SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *");
84 SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *",
85     "struct vnode *");
86 SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *",
87     "char *");
88 
89 /*
90  * This structure describes the elements in the cache of recent
91  * names looked up by namei.
92  */
93 
94 struct	namecache {
95 	LIST_ENTRY(namecache) nc_hash;	/* hash chain */
96 	LIST_ENTRY(namecache) nc_src;	/* source vnode list */
97 	TAILQ_ENTRY(namecache) nc_dst;	/* destination vnode list */
98 	struct	vnode *nc_dvp;		/* vnode of parent of name */
99 	struct	vnode *nc_vp;		/* vnode the name refers to */
100 	u_char	nc_flag;		/* flag bits */
101 	u_char	nc_nlen;		/* length of name */
102 	char	nc_name[0];		/* segment name + nul */
103 };
104 
105 /*
106  * struct namecache_ts repeats struct namecache layout up to the
107  * nc_nlen member.
108  * struct namecache_ts is used in place of struct namecache when time(s) need
109  * to be stored.  The nc_dotdottime field is used when a cache entry is mapping
110  * both a non-dotdot directory name plus dotdot for the directory's
111  * parent.
112  */
113 struct	namecache_ts {
114 	LIST_ENTRY(namecache) nc_hash;	/* hash chain */
115 	LIST_ENTRY(namecache) nc_src;	/* source vnode list */
116 	TAILQ_ENTRY(namecache) nc_dst;	/* destination vnode list */
117 	struct	vnode *nc_dvp;		/* vnode of parent of name */
118 	struct	vnode *nc_vp;		/* vnode the name refers to */
119 	u_char	nc_flag;		/* flag bits */
120 	u_char	nc_nlen;		/* length of name */
121 	struct	timespec nc_time;	/* timespec provided by fs */
122 	struct	timespec nc_dotdottime;	/* dotdot timespec provided by fs */
123 	int	nc_ticks;		/* ticks value when entry was added */
124 	char	nc_name[0];		/* segment name + nul */
125 };
126 
127 /*
128  * Flags in namecache.nc_flag
129  */
130 #define NCF_WHITE	0x01
131 #define NCF_ISDOTDOT	0x02
132 #define	NCF_TS		0x04
133 #define	NCF_DTS		0x08
134 
135 /*
136  * Name caching works as follows:
137  *
138  * Names found by directory scans are retained in a cache
139  * for future reference.  It is managed LRU, so frequently
140  * used names will hang around.  Cache is indexed by hash value
141  * obtained from (vp, name) where vp refers to the directory
142  * containing name.
143  *
144  * If it is a "negative" entry, (i.e. for a name that is known NOT to
145  * exist) the vnode pointer will be NULL.
146  *
147  * Upon reaching the last segment of a path, if the reference
148  * is for DELETE, or NOCACHE is set (rewrite), and the
149  * name is located in the cache, it will be dropped.
150  */
151 
152 /*
153  * Structures associated with name caching.
154  */
155 #define NCHHASH(hash) \
156 	(&nchashtbl[(hash) & nchash])
157 static LIST_HEAD(nchashhead, namecache) *nchashtbl;	/* Hash Table */
158 static TAILQ_HEAD(, namecache) ncneg;	/* Hash Table */
159 static u_long	nchash;			/* size of hash table */
160 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0,
161     "Size of namecache hash table");
162 static u_long	ncnegfactor = 16;	/* ratio of negative entries */
163 SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0,
164     "Ratio of negative namecache entries");
165 static u_long	numneg;			/* number of negative entries allocated */
166 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0,
167     "Number of negative entries in namecache");
168 static u_long	numcache;		/* number of cache entries allocated */
169 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0,
170     "Number of namecache entries");
171 static u_long	numcachehv;		/* number of cache entries with vnodes held */
172 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0,
173     "Number of namecache entries with vnodes held");
174 u_int	ncsizefactor = 2;
175 SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0,
176     "Size factor for namecache");
177 
178 struct nchstats	nchstats;		/* cache effectiveness statistics */
179 
180 static struct rwlock cache_lock;
181 RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
182 
183 #define	CACHE_UPGRADE_LOCK()	rw_try_upgrade(&cache_lock)
184 #define	CACHE_RLOCK()		rw_rlock(&cache_lock)
185 #define	CACHE_RUNLOCK()		rw_runlock(&cache_lock)
186 #define	CACHE_WLOCK()		rw_wlock(&cache_lock)
187 #define	CACHE_WUNLOCK()		rw_wunlock(&cache_lock)
188 
189 /*
190  * UMA zones for the VFS cache.
191  *
192  * The small cache is used for entries with short names, which are the
193  * most common.  The large cache is used for entries which are too big to
194  * fit in the small cache.
195  */
196 static uma_zone_t cache_zone_small;
197 static uma_zone_t cache_zone_small_ts;
198 static uma_zone_t cache_zone_large;
199 static uma_zone_t cache_zone_large_ts;
200 
201 #define	CACHE_PATH_CUTOFF	35
202 
203 static struct namecache *
204 cache_alloc(int len, int ts)
205 {
206 
207 	if (len > CACHE_PATH_CUTOFF) {
208 		if (ts)
209 			return (uma_zalloc(cache_zone_large_ts, M_WAITOK));
210 		else
211 			return (uma_zalloc(cache_zone_large, M_WAITOK));
212 	}
213 	if (ts)
214 		return (uma_zalloc(cache_zone_small_ts, M_WAITOK));
215 	else
216 		return (uma_zalloc(cache_zone_small, M_WAITOK));
217 }
218 
219 static void
220 cache_free(struct namecache *ncp)
221 {
222 	int ts;
223 
224 	if (ncp == NULL)
225 		return;
226 	ts = ncp->nc_flag & NCF_TS;
227 	if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) {
228 		if (ts)
229 			uma_zfree(cache_zone_small_ts, ncp);
230 		else
231 			uma_zfree(cache_zone_small, ncp);
232 	} else if (ts)
233 		uma_zfree(cache_zone_large_ts, ncp);
234 	else
235 		uma_zfree(cache_zone_large, ncp);
236 }
237 
238 static char *
239 nc_get_name(struct namecache *ncp)
240 {
241 	struct namecache_ts *ncp_ts;
242 
243 	if ((ncp->nc_flag & NCF_TS) == 0)
244 		return (ncp->nc_name);
245 	ncp_ts = (struct namecache_ts *)ncp;
246 	return (ncp_ts->nc_name);
247 }
248 
249 static void
250 cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp)
251 {
252 
253 	KASSERT((ncp->nc_flag & NCF_TS) != 0 ||
254 	    (tsp == NULL && ticksp == NULL),
255 	    ("No NCF_TS"));
256 
257 	if (tsp != NULL)
258 		*tsp = ((struct namecache_ts *)ncp)->nc_time;
259 	if (ticksp != NULL)
260 		*ticksp = ((struct namecache_ts *)ncp)->nc_ticks;
261 }
262 
263 static int	doingcache = 1;		/* 1 => enable the cache */
264 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
265     "VFS namecache enabled");
266 
267 /* Export size information to userland */
268 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR,
269     sizeof(struct namecache), "sizeof(struct namecache)");
270 
271 /*
272  * The new name cache statistics
273  */
274 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0,
275     "Name cache statistics");
276 #define STATNODE_ULONG(name, descr)	\
277 	SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr);
278 #define STATNODE_COUNTER(name, descr)	\
279 	static counter_u64_t name;	\
280 	SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, descr);
281 STATNODE_ULONG(numneg, "Number of negative cache entries");
282 STATNODE_ULONG(numcache, "Number of cache entries");
283 STATNODE_COUNTER(numcalls, "Number of cache lookups");
284 STATNODE_COUNTER(dothits, "Number of '.' hits");
285 STATNODE_COUNTER(dotdothits, "Number of '..' hits");
286 STATNODE_COUNTER(numchecks, "Number of checks in lookup");
287 STATNODE_COUNTER(nummiss, "Number of cache misses");
288 STATNODE_COUNTER(nummisszap, "Number of cache misses we do not want to cache");
289 STATNODE_COUNTER(numposzaps,
290     "Number of cache hits (positive) we do not want to cache");
291 STATNODE_COUNTER(numposhits, "Number of cache hits (positive)");
292 STATNODE_COUNTER(numnegzaps,
293     "Number of cache hits (negative) we do not want to cache");
294 STATNODE_COUNTER(numneghits, "Number of cache hits (negative)");
295 /* These count for kern___getcwd(), too. */
296 STATNODE_COUNTER(numfullpathcalls, "Number of fullpath search calls");
297 STATNODE_COUNTER(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
298 STATNODE_COUNTER(numfullpathfail2,
299     "Number of fullpath search errors (VOP_VPTOCNP failures)");
300 STATNODE_COUNTER(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
301 STATNODE_COUNTER(numfullpathfound, "Number of successful fullpath calls");
302 static long numupgrades; STATNODE_ULONG(numupgrades,
303     "Number of updates of the cache after lookup (write lock + retry)");
304 
305 static void cache_zap(struct namecache *ncp);
306 static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
307     u_int *buflen);
308 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
309     char *buf, char **retbuf, u_int buflen);
310 
311 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
312 
313 static uint32_t
314 cache_get_hash(char *name, u_char len, struct vnode *dvp)
315 {
316 	uint32_t hash;
317 
318 	hash = fnv_32_buf(name, len, FNV1_32_INIT);
319 	hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
320 	return (hash);
321 }
322 
323 static int
324 sysctl_nchstats(SYSCTL_HANDLER_ARGS)
325 {
326 	struct nchstats snap;
327 
328 	if (req->oldptr == NULL)
329 		return (SYSCTL_OUT(req, 0, sizeof(snap)));
330 
331 	snap = nchstats;
332 	snap.ncs_goodhits = counter_u64_fetch(numposhits);
333 	snap.ncs_neghits = counter_u64_fetch(numneghits);
334 	snap.ncs_badhits = counter_u64_fetch(numposzaps) +
335 	    counter_u64_fetch(numnegzaps);
336 	snap.ncs_miss = counter_u64_fetch(nummisszap) +
337 	    counter_u64_fetch(nummiss);
338 
339 	return (SYSCTL_OUT(req, &snap, sizeof(snap)));
340 }
341 SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD |
342     CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU",
343     "VFS cache effectiveness statistics");
344 
345 #ifdef DIAGNOSTIC
346 /*
347  * Grab an atomic snapshot of the name cache hash chain lengths
348  */
349 static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
350     "hash table stats");
351 
352 static int
353 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
354 {
355 	struct nchashhead *ncpp;
356 	struct namecache *ncp;
357 	int i, error, n_nchash, *cntbuf;
358 
359 retry:
360 	n_nchash = nchash + 1;	/* nchash is max index, not count */
361 	if (req->oldptr == NULL)
362 		return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
363 	cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK);
364 	CACHE_RLOCK();
365 	if (n_nchash != nchash + 1) {
366 		CACHE_RUNLOCK();
367 		free(cntbuf, M_TEMP);
368 		goto retry;
369 	}
370 	/* Scan hash tables counting entries */
371 	for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++)
372 		LIST_FOREACH(ncp, ncpp, nc_hash)
373 			cntbuf[i]++;
374 	CACHE_RUNLOCK();
375 	for (error = 0, i = 0; i < n_nchash; i++)
376 		if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0)
377 			break;
378 	free(cntbuf, M_TEMP);
379 	return (error);
380 }
381 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
382     CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
383     "nchash chain lengths");
384 
385 static int
386 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
387 {
388 	int error;
389 	struct nchashhead *ncpp;
390 	struct namecache *ncp;
391 	int n_nchash;
392 	int count, maxlength, used, pct;
393 
394 	if (!req->oldptr)
395 		return SYSCTL_OUT(req, 0, 4 * sizeof(int));
396 
397 	CACHE_RLOCK();
398 	n_nchash = nchash + 1;	/* nchash is max index, not count */
399 	used = 0;
400 	maxlength = 0;
401 
402 	/* Scan hash tables for applicable entries */
403 	for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
404 		count = 0;
405 		LIST_FOREACH(ncp, ncpp, nc_hash) {
406 			count++;
407 		}
408 		if (count)
409 			used++;
410 		if (maxlength < count)
411 			maxlength = count;
412 	}
413 	n_nchash = nchash + 1;
414 	CACHE_RUNLOCK();
415 	pct = (used * 100) / (n_nchash / 100);
416 	error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
417 	if (error)
418 		return (error);
419 	error = SYSCTL_OUT(req, &used, sizeof(used));
420 	if (error)
421 		return (error);
422 	error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
423 	if (error)
424 		return (error);
425 	error = SYSCTL_OUT(req, &pct, sizeof(pct));
426 	if (error)
427 		return (error);
428 	return (0);
429 }
430 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
431     CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
432     "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)");
433 #endif
434 
435 /*
436  * cache_zap():
437  *
438  *   Removes a namecache entry from cache, whether it contains an actual
439  *   pointer to a vnode or if it is just a negative cache entry.
440  */
441 static void
442 cache_zap(struct namecache *ncp)
443 {
444 	struct vnode *vp;
445 
446 	rw_assert(&cache_lock, RA_WLOCKED);
447 	CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
448 	if (ncp->nc_vp != NULL) {
449 		SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp,
450 		    nc_get_name(ncp), ncp->nc_vp);
451 	} else {
452 		SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp,
453 		    nc_get_name(ncp));
454 	}
455 	vp = NULL;
456 	LIST_REMOVE(ncp, nc_hash);
457 	if (ncp->nc_flag & NCF_ISDOTDOT) {
458 		if (ncp == ncp->nc_dvp->v_cache_dd)
459 			ncp->nc_dvp->v_cache_dd = NULL;
460 	} else {
461 		LIST_REMOVE(ncp, nc_src);
462 		if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
463 			vp = ncp->nc_dvp;
464 			numcachehv--;
465 		}
466 	}
467 	if (ncp->nc_vp) {
468 		TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
469 		if (ncp == ncp->nc_vp->v_cache_dd)
470 			ncp->nc_vp->v_cache_dd = NULL;
471 	} else {
472 		TAILQ_REMOVE(&ncneg, ncp, nc_dst);
473 		numneg--;
474 	}
475 	numcache--;
476 	cache_free(ncp);
477 	if (vp != NULL)
478 		vdrop(vp);
479 }
480 
481 /*
482  * Lookup an entry in the cache
483  *
484  * Lookup is called with dvp pointing to the directory to search,
485  * cnp pointing to the name of the entry being sought. If the lookup
486  * succeeds, the vnode is returned in *vpp, and a status of -1 is
487  * returned. If the lookup determines that the name does not exist
488  * (negative caching), a status of ENOENT is returned. If the lookup
489  * fails, a status of zero is returned.  If the directory vnode is
490  * recycled out from under us due to a forced unmount, a status of
491  * ENOENT is returned.
492  *
493  * vpp is locked and ref'd on return.  If we're looking up DOTDOT, dvp is
494  * unlocked.  If we're looking up . an extra ref is taken, but the lock is
495  * not recursively acquired.
496  */
497 
498 int
499 cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
500     struct timespec *tsp, int *ticksp)
501 {
502 	struct namecache *ncp;
503 	uint32_t hash;
504 	int error, ltype, wlocked;
505 
506 	if (!doingcache) {
507 		cnp->cn_flags &= ~MAKEENTRY;
508 		return (0);
509 	}
510 retry:
511 	wlocked = 0;
512 	counter_u64_add(numcalls, 1);
513 	error = 0;
514 
515 retry_wlocked:
516 	if (cnp->cn_nameptr[0] == '.') {
517 		if (cnp->cn_namelen == 1) {
518 			*vpp = dvp;
519 			CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
520 			    dvp, cnp->cn_nameptr);
521 			counter_u64_add(dothits, 1);
522 			SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp);
523 			if (tsp != NULL)
524 				timespecclear(tsp);
525 			if (ticksp != NULL)
526 				*ticksp = ticks;
527 			VREF(*vpp);
528 			/*
529 			 * When we lookup "." we still can be asked to lock it
530 			 * differently...
531 			 */
532 			ltype = cnp->cn_lkflags & LK_TYPE_MASK;
533 			if (ltype != VOP_ISLOCKED(*vpp)) {
534 				if (ltype == LK_EXCLUSIVE) {
535 					vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
536 					if ((*vpp)->v_iflag & VI_DOOMED) {
537 						/* forced unmount */
538 						vrele(*vpp);
539 						*vpp = NULL;
540 						return (ENOENT);
541 					}
542 				} else
543 					vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
544 			}
545 			return (-1);
546 		}
547 		if (!wlocked)
548 			CACHE_RLOCK();
549 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
550 			counter_u64_add(dotdothits, 1);
551 			if (dvp->v_cache_dd == NULL) {
552 				SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
553 				    "..", NULL);
554 				goto unlock;
555 			}
556 			if ((cnp->cn_flags & MAKEENTRY) == 0) {
557 				if (!wlocked && !CACHE_UPGRADE_LOCK())
558 					goto wlock;
559 				if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
560 					cache_zap(dvp->v_cache_dd);
561 				dvp->v_cache_dd = NULL;
562 				CACHE_WUNLOCK();
563 				return (0);
564 			}
565 			ncp = dvp->v_cache_dd;
566 			if (ncp->nc_flag & NCF_ISDOTDOT)
567 				*vpp = ncp->nc_vp;
568 			else
569 				*vpp = ncp->nc_dvp;
570 			/* Return failure if negative entry was found. */
571 			if (*vpp == NULL)
572 				goto negative_success;
573 			CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
574 			    dvp, cnp->cn_nameptr, *vpp);
575 			SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..",
576 			    *vpp);
577 			cache_out_ts(ncp, tsp, ticksp);
578 			if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
579 			    NCF_DTS && tsp != NULL)
580 				*tsp = ((struct namecache_ts *)ncp)->
581 				    nc_dotdottime;
582 			goto success;
583 		}
584 	} else if (!wlocked)
585 		CACHE_RLOCK();
586 
587 	hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp);
588 	LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
589 		counter_u64_add(numchecks, 1);
590 		if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
591 		    !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen))
592 			break;
593 	}
594 
595 	/* We failed to find an entry */
596 	if (ncp == NULL) {
597 		SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
598 		    NULL);
599 		if ((cnp->cn_flags & MAKEENTRY) == 0) {
600 			counter_u64_add(nummisszap, 1);
601 		} else {
602 			counter_u64_add(nummiss, 1);
603 		}
604 		goto unlock;
605 	}
606 
607 	/* We don't want to have an entry, so dump it */
608 	if ((cnp->cn_flags & MAKEENTRY) == 0) {
609 		counter_u64_add(numposzaps, 1);
610 		if (!wlocked && !CACHE_UPGRADE_LOCK())
611 			goto wlock;
612 		cache_zap(ncp);
613 		CACHE_WUNLOCK();
614 		return (0);
615 	}
616 
617 	/* We found a "positive" match, return the vnode */
618 	if (ncp->nc_vp) {
619 		counter_u64_add(numposhits, 1);
620 		*vpp = ncp->nc_vp;
621 		CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
622 		    dvp, cnp->cn_nameptr, *vpp, ncp);
623 		SDT_PROBE3(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp),
624 		    *vpp);
625 		cache_out_ts(ncp, tsp, ticksp);
626 		goto success;
627 	}
628 
629 negative_success:
630 	/* We found a negative match, and want to create it, so purge */
631 	if (cnp->cn_nameiop == CREATE) {
632 		counter_u64_add(numnegzaps, 1);
633 		if (!wlocked && !CACHE_UPGRADE_LOCK())
634 			goto wlock;
635 		cache_zap(ncp);
636 		CACHE_WUNLOCK();
637 		return (0);
638 	}
639 
640 	if (!wlocked && !CACHE_UPGRADE_LOCK())
641 		goto wlock;
642 	counter_u64_add(numneghits, 1);
643 	/*
644 	 * We found a "negative" match, so we shift it to the end of
645 	 * the "negative" cache entries queue to satisfy LRU.  Also,
646 	 * check to see if the entry is a whiteout; indicate this to
647 	 * the componentname, if so.
648 	 */
649 	TAILQ_REMOVE(&ncneg, ncp, nc_dst);
650 	TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
651 	if (ncp->nc_flag & NCF_WHITE)
652 		cnp->cn_flags |= ISWHITEOUT;
653 	SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp,
654 	    nc_get_name(ncp));
655 	cache_out_ts(ncp, tsp, ticksp);
656 	CACHE_WUNLOCK();
657 	return (ENOENT);
658 
659 wlock:
660 	/*
661 	 * We need to update the cache after our lookup, so upgrade to
662 	 * a write lock and retry the operation.
663 	 */
664 	CACHE_RUNLOCK();
665 	CACHE_WLOCK();
666 	numupgrades++;
667 	wlocked = 1;
668 	goto retry_wlocked;
669 
670 success:
671 	/*
672 	 * On success we return a locked and ref'd vnode as per the lookup
673 	 * protocol.
674 	 */
675 	MPASS(dvp != *vpp);
676 	ltype = 0;	/* silence gcc warning */
677 	if (cnp->cn_flags & ISDOTDOT) {
678 		ltype = VOP_ISLOCKED(dvp);
679 		VOP_UNLOCK(dvp, 0);
680 	}
681 	vhold(*vpp);
682 	if (wlocked)
683 		CACHE_WUNLOCK();
684 	else
685 		CACHE_RUNLOCK();
686 	error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread);
687 	if (cnp->cn_flags & ISDOTDOT) {
688 		vn_lock(dvp, ltype | LK_RETRY);
689 		if (dvp->v_iflag & VI_DOOMED) {
690 			if (error == 0)
691 				vput(*vpp);
692 			*vpp = NULL;
693 			return (ENOENT);
694 		}
695 	}
696 	if (error) {
697 		*vpp = NULL;
698 		goto retry;
699 	}
700 	if ((cnp->cn_flags & ISLASTCN) &&
701 	    (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
702 		ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
703 	}
704 	return (-1);
705 
706 unlock:
707 	if (wlocked)
708 		CACHE_WUNLOCK();
709 	else
710 		CACHE_RUNLOCK();
711 	return (0);
712 }
713 
714 /*
715  * Add an entry to the cache.
716  */
717 void
718 cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp,
719     struct timespec *tsp, struct timespec *dtsp)
720 {
721 	struct namecache *ncp, *n2;
722 	struct namecache_ts *n3;
723 	struct nchashhead *ncpp;
724 	uint32_t hash;
725 	int flag;
726 	int len;
727 
728 	CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
729 	VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
730 	    ("cache_enter: Adding a doomed vnode"));
731 	VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
732 	    ("cache_enter: Doomed vnode used as src"));
733 
734 	if (!doingcache)
735 		return;
736 
737 	/*
738 	 * Avoid blowout in namecache entries.
739 	 */
740 	if (numcache >= desiredvnodes * ncsizefactor)
741 		return;
742 
743 	flag = 0;
744 	if (cnp->cn_nameptr[0] == '.') {
745 		if (cnp->cn_namelen == 1)
746 			return;
747 		if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
748 			CACHE_WLOCK();
749 			/*
750 			 * If dotdot entry already exists, just retarget it
751 			 * to new parent vnode, otherwise continue with new
752 			 * namecache entry allocation.
753 			 */
754 			if ((ncp = dvp->v_cache_dd) != NULL &&
755 			    ncp->nc_flag & NCF_ISDOTDOT) {
756 				KASSERT(ncp->nc_dvp == dvp,
757 				    ("wrong isdotdot parent"));
758 				if (ncp->nc_vp != NULL) {
759 					TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
760 					    ncp, nc_dst);
761 				} else {
762 					TAILQ_REMOVE(&ncneg, ncp, nc_dst);
763 					numneg--;
764 				}
765 				if (vp != NULL) {
766 					TAILQ_INSERT_HEAD(&vp->v_cache_dst,
767 					    ncp, nc_dst);
768 				} else {
769 					TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
770 					numneg++;
771 				}
772 				ncp->nc_vp = vp;
773 				CACHE_WUNLOCK();
774 				return;
775 			}
776 			dvp->v_cache_dd = NULL;
777 			SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp);
778 			CACHE_WUNLOCK();
779 			flag = NCF_ISDOTDOT;
780 		}
781 	}
782 
783 	/*
784 	 * Calculate the hash key and setup as much of the new
785 	 * namecache entry as possible before acquiring the lock.
786 	 */
787 	ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
788 	ncp->nc_vp = vp;
789 	ncp->nc_dvp = dvp;
790 	ncp->nc_flag = flag;
791 	if (tsp != NULL) {
792 		n3 = (struct namecache_ts *)ncp;
793 		n3->nc_time = *tsp;
794 		n3->nc_ticks = ticks;
795 		n3->nc_flag |= NCF_TS;
796 		if (dtsp != NULL) {
797 			n3->nc_dotdottime = *dtsp;
798 			n3->nc_flag |= NCF_DTS;
799 		}
800 	}
801 	len = ncp->nc_nlen = cnp->cn_namelen;
802 	hash = cache_get_hash(cnp->cn_nameptr, len, dvp);
803 	strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1);
804 	CACHE_WLOCK();
805 
806 	/*
807 	 * See if this vnode or negative entry is already in the cache
808 	 * with this name.  This can happen with concurrent lookups of
809 	 * the same path name.
810 	 */
811 	ncpp = NCHHASH(hash);
812 	LIST_FOREACH(n2, ncpp, nc_hash) {
813 		if (n2->nc_dvp == dvp &&
814 		    n2->nc_nlen == cnp->cn_namelen &&
815 		    !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) {
816 			if (tsp != NULL) {
817 				KASSERT((n2->nc_flag & NCF_TS) != 0,
818 				    ("no NCF_TS"));
819 				n3 = (struct namecache_ts *)n2;
820 				n3->nc_time =
821 				    ((struct namecache_ts *)ncp)->nc_time;
822 				n3->nc_ticks =
823 				    ((struct namecache_ts *)ncp)->nc_ticks;
824 				if (dtsp != NULL) {
825 					n3->nc_dotdottime =
826 					    ((struct namecache_ts *)ncp)->
827 					    nc_dotdottime;
828 					n3->nc_flag |= NCF_DTS;
829 				}
830 			}
831 			CACHE_WUNLOCK();
832 			cache_free(ncp);
833 			return;
834 		}
835 	}
836 
837 	if (flag == NCF_ISDOTDOT) {
838 		/*
839 		 * See if we are trying to add .. entry, but some other lookup
840 		 * has populated v_cache_dd pointer already.
841 		 */
842 		if (dvp->v_cache_dd != NULL) {
843 			CACHE_WUNLOCK();
844 			cache_free(ncp);
845 			return;
846 		}
847 		KASSERT(vp == NULL || vp->v_type == VDIR,
848 		    ("wrong vnode type %p", vp));
849 		dvp->v_cache_dd = ncp;
850 	}
851 
852 	numcache++;
853 	if (vp != NULL) {
854 		if (vp->v_type == VDIR) {
855 			if (flag != NCF_ISDOTDOT) {
856 				/*
857 				 * For this case, the cache entry maps both the
858 				 * directory name in it and the name ".." for the
859 				 * directory's parent.
860 				 */
861 				if ((n2 = vp->v_cache_dd) != NULL &&
862 				    (n2->nc_flag & NCF_ISDOTDOT) != 0)
863 					cache_zap(n2);
864 				vp->v_cache_dd = ncp;
865 			}
866 		} else {
867 			vp->v_cache_dd = NULL;
868 		}
869 	}
870 
871 	/*
872 	 * Insert the new namecache entry into the appropriate chain
873 	 * within the cache entries table.
874 	 */
875 	LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
876 	if (flag != NCF_ISDOTDOT) {
877 		if (LIST_EMPTY(&dvp->v_cache_src)) {
878 			vhold(dvp);
879 			numcachehv++;
880 		}
881 		LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
882 	}
883 
884 	/*
885 	 * If the entry is "negative", we place it into the
886 	 * "negative" cache queue, otherwise, we place it into the
887 	 * destination vnode's cache entries queue.
888 	 */
889 	if (vp != NULL) {
890 		TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
891 		SDT_PROBE3(vfs, namecache, enter, done, dvp, nc_get_name(ncp),
892 		    vp);
893 	} else {
894 		if (cnp->cn_flags & ISWHITEOUT)
895 			ncp->nc_flag |= NCF_WHITE;
896 		TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
897 		numneg++;
898 		SDT_PROBE2(vfs, namecache, enter_negative, done, dvp,
899 		    nc_get_name(ncp));
900 	}
901 	if (numneg * ncnegfactor > numcache) {
902 		ncp = TAILQ_FIRST(&ncneg);
903 		KASSERT(ncp->nc_vp == NULL, ("ncp %p vp %p on ncneg",
904 		    ncp, ncp->nc_vp));
905 		cache_zap(ncp);
906 	}
907 	CACHE_WUNLOCK();
908 }
909 
910 /*
911  * Name cache initialization, from vfs_init() when we are booting
912  */
913 static void
914 nchinit(void *dummy __unused)
915 {
916 
917 	TAILQ_INIT(&ncneg);
918 
919 	cache_zone_small = uma_zcreate("S VFS Cache",
920 	    sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
921 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
922 	cache_zone_small_ts = uma_zcreate("STS VFS Cache",
923 	    sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
924 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
925 	cache_zone_large = uma_zcreate("L VFS Cache",
926 	    sizeof(struct namecache) + NAME_MAX + 1,
927 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
928 	cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
929 	    sizeof(struct namecache_ts) + NAME_MAX + 1,
930 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
931 
932 	nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
933 
934 	numcalls = counter_u64_alloc(M_WAITOK);
935 	dothits = counter_u64_alloc(M_WAITOK);
936 	dotdothits = counter_u64_alloc(M_WAITOK);
937 	numchecks = counter_u64_alloc(M_WAITOK);
938 	nummiss = counter_u64_alloc(M_WAITOK);
939 	nummisszap = counter_u64_alloc(M_WAITOK);
940 	numposzaps = counter_u64_alloc(M_WAITOK);
941 	numposhits = counter_u64_alloc(M_WAITOK);
942 	numnegzaps = counter_u64_alloc(M_WAITOK);
943 	numneghits = counter_u64_alloc(M_WAITOK);
944 	numfullpathcalls = counter_u64_alloc(M_WAITOK);
945 	numfullpathfail1 = counter_u64_alloc(M_WAITOK);
946 	numfullpathfail2 = counter_u64_alloc(M_WAITOK);
947 	numfullpathfail4 = counter_u64_alloc(M_WAITOK);
948 	numfullpathfound = counter_u64_alloc(M_WAITOK);
949 }
950 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
951 
952 void
953 cache_changesize(int newmaxvnodes)
954 {
955 	struct nchashhead *new_nchashtbl, *old_nchashtbl;
956 	u_long new_nchash, old_nchash;
957 	struct namecache *ncp;
958 	uint32_t hash;
959 	int i;
960 
961 	new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash);
962 	/* If same hash table size, nothing to do */
963 	if (nchash == new_nchash) {
964 		free(new_nchashtbl, M_VFSCACHE);
965 		return;
966 	}
967 	/*
968 	 * Move everything from the old hash table to the new table.
969 	 * None of the namecache entries in the table can be removed
970 	 * because to do so, they have to be removed from the hash table.
971 	 */
972 	CACHE_WLOCK();
973 	old_nchashtbl = nchashtbl;
974 	old_nchash = nchash;
975 	nchashtbl = new_nchashtbl;
976 	nchash = new_nchash;
977 	for (i = 0; i <= old_nchash; i++) {
978 		while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) {
979 			hash = cache_get_hash(nc_get_name(ncp), ncp->nc_nlen,
980 			    ncp->nc_dvp);
981 			LIST_REMOVE(ncp, nc_hash);
982 			LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash);
983 		}
984 	}
985 	CACHE_WUNLOCK();
986 	free(old_nchashtbl, M_VFSCACHE);
987 }
988 
989 /*
990  * Invalidate all entries to a particular vnode.
991  */
992 void
993 cache_purge(struct vnode *vp)
994 {
995 
996 	CTR1(KTR_VFS, "cache_purge(%p)", vp);
997 	SDT_PROBE1(vfs, namecache, purge, done, vp);
998 	CACHE_WLOCK();
999 	while (!LIST_EMPTY(&vp->v_cache_src))
1000 		cache_zap(LIST_FIRST(&vp->v_cache_src));
1001 	while (!TAILQ_EMPTY(&vp->v_cache_dst))
1002 		cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
1003 	if (vp->v_cache_dd != NULL) {
1004 		KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
1005 		   ("lost dotdot link"));
1006 		cache_zap(vp->v_cache_dd);
1007 	}
1008 	KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
1009 	CACHE_WUNLOCK();
1010 }
1011 
1012 /*
1013  * Invalidate all negative entries for a particular directory vnode.
1014  */
1015 void
1016 cache_purge_negative(struct vnode *vp)
1017 {
1018 	struct namecache *cp, *ncp;
1019 
1020 	CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
1021 	SDT_PROBE1(vfs, namecache, purge_negative, done, vp);
1022 	CACHE_WLOCK();
1023 	LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
1024 		if (cp->nc_vp == NULL)
1025 			cache_zap(cp);
1026 	}
1027 	CACHE_WUNLOCK();
1028 }
1029 
1030 /*
1031  * Flush all entries referencing a particular filesystem.
1032  */
1033 void
1034 cache_purgevfs(struct mount *mp)
1035 {
1036 	struct nchashhead *ncpp;
1037 	struct namecache *ncp, *nnp;
1038 
1039 	/* Scan hash tables for applicable entries */
1040 	SDT_PROBE1(vfs, namecache, purgevfs, done, mp);
1041 	CACHE_WLOCK();
1042 	for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
1043 		LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
1044 			if (ncp->nc_dvp->v_mount == mp)
1045 				cache_zap(ncp);
1046 		}
1047 	}
1048 	CACHE_WUNLOCK();
1049 }
1050 
1051 /*
1052  * Perform canonical checks and cache lookup and pass on to filesystem
1053  * through the vop_cachedlookup only if needed.
1054  */
1055 
1056 int
1057 vfs_cache_lookup(struct vop_lookup_args *ap)
1058 {
1059 	struct vnode *dvp;
1060 	int error;
1061 	struct vnode **vpp = ap->a_vpp;
1062 	struct componentname *cnp = ap->a_cnp;
1063 	struct ucred *cred = cnp->cn_cred;
1064 	int flags = cnp->cn_flags;
1065 	struct thread *td = cnp->cn_thread;
1066 
1067 	*vpp = NULL;
1068 	dvp = ap->a_dvp;
1069 
1070 	if (dvp->v_type != VDIR)
1071 		return (ENOTDIR);
1072 
1073 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1074 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1075 		return (EROFS);
1076 
1077 	error = VOP_ACCESS(dvp, VEXEC, cred, td);
1078 	if (error)
1079 		return (error);
1080 
1081 	error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
1082 	if (error == 0)
1083 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
1084 	if (error == -1)
1085 		return (0);
1086 	return (error);
1087 }
1088 
1089 /*
1090  * XXX All of these sysctls would probably be more productive dead.
1091  */
1092 static int disablecwd;
1093 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
1094    "Disable the getcwd syscall");
1095 
1096 /* Implementation of the getcwd syscall. */
1097 int
1098 sys___getcwd(struct thread *td, struct __getcwd_args *uap)
1099 {
1100 
1101 	return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen,
1102 	    MAXPATHLEN));
1103 }
1104 
1105 int
1106 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen,
1107     u_int path_max)
1108 {
1109 	char *bp, *tmpbuf;
1110 	struct filedesc *fdp;
1111 	struct vnode *cdir, *rdir;
1112 	int error;
1113 
1114 	if (disablecwd)
1115 		return (ENODEV);
1116 	if (buflen < 2)
1117 		return (EINVAL);
1118 	if (buflen > path_max)
1119 		buflen = path_max;
1120 
1121 	tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
1122 	fdp = td->td_proc->p_fd;
1123 	FILEDESC_SLOCK(fdp);
1124 	cdir = fdp->fd_cdir;
1125 	VREF(cdir);
1126 	rdir = fdp->fd_rdir;
1127 	VREF(rdir);
1128 	FILEDESC_SUNLOCK(fdp);
1129 	error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
1130 	vrele(rdir);
1131 	vrele(cdir);
1132 
1133 	if (!error) {
1134 		if (bufseg == UIO_SYSSPACE)
1135 			bcopy(bp, buf, strlen(bp) + 1);
1136 		else
1137 			error = copyout(bp, buf, strlen(bp) + 1);
1138 #ifdef KTRACE
1139 	if (KTRPOINT(curthread, KTR_NAMEI))
1140 		ktrnamei(bp);
1141 #endif
1142 	}
1143 	free(tmpbuf, M_TEMP);
1144 	return (error);
1145 }
1146 
1147 /*
1148  * Thus begins the fullpath magic.
1149  */
1150 
1151 static int disablefullpath;
1152 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
1153     "Disable the vn_fullpath function");
1154 
1155 /*
1156  * Retrieve the full filesystem path that correspond to a vnode from the name
1157  * cache (if available)
1158  */
1159 int
1160 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
1161 {
1162 	char *buf;
1163 	struct filedesc *fdp;
1164 	struct vnode *rdir;
1165 	int error;
1166 
1167 	if (disablefullpath)
1168 		return (ENODEV);
1169 	if (vn == NULL)
1170 		return (EINVAL);
1171 
1172 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1173 	fdp = td->td_proc->p_fd;
1174 	FILEDESC_SLOCK(fdp);
1175 	rdir = fdp->fd_rdir;
1176 	VREF(rdir);
1177 	FILEDESC_SUNLOCK(fdp);
1178 	error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1179 	vrele(rdir);
1180 
1181 	if (!error)
1182 		*freebuf = buf;
1183 	else
1184 		free(buf, M_TEMP);
1185 	return (error);
1186 }
1187 
1188 /*
1189  * This function is similar to vn_fullpath, but it attempts to lookup the
1190  * pathname relative to the global root mount point.  This is required for the
1191  * auditing sub-system, as audited pathnames must be absolute, relative to the
1192  * global root mount point.
1193  */
1194 int
1195 vn_fullpath_global(struct thread *td, struct vnode *vn,
1196     char **retbuf, char **freebuf)
1197 {
1198 	char *buf;
1199 	int error;
1200 
1201 	if (disablefullpath)
1202 		return (ENODEV);
1203 	if (vn == NULL)
1204 		return (EINVAL);
1205 	buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1206 	error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1207 	if (!error)
1208 		*freebuf = buf;
1209 	else
1210 		free(buf, M_TEMP);
1211 	return (error);
1212 }
1213 
1214 int
1215 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1216 {
1217 	int error;
1218 
1219 	CACHE_RLOCK();
1220 	error = vn_vptocnp_locked(vp, cred, buf, buflen);
1221 	if (error == 0)
1222 		CACHE_RUNLOCK();
1223 	return (error);
1224 }
1225 
1226 static int
1227 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1228     u_int *buflen)
1229 {
1230 	struct vnode *dvp;
1231 	struct namecache *ncp;
1232 	int error;
1233 
1234 	TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1235 		if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1236 			break;
1237 	}
1238 	if (ncp != NULL) {
1239 		if (*buflen < ncp->nc_nlen) {
1240 			CACHE_RUNLOCK();
1241 			vrele(*vp);
1242 			counter_u64_add(numfullpathfail4, 1);
1243 			error = ENOMEM;
1244 			SDT_PROBE3(vfs, namecache, fullpath, return, error,
1245 			    vp, NULL);
1246 			return (error);
1247 		}
1248 		*buflen -= ncp->nc_nlen;
1249 		memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen);
1250 		SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1251 		    nc_get_name(ncp), vp);
1252 		dvp = *vp;
1253 		*vp = ncp->nc_dvp;
1254 		vref(*vp);
1255 		CACHE_RUNLOCK();
1256 		vrele(dvp);
1257 		CACHE_RLOCK();
1258 		return (0);
1259 	}
1260 	SDT_PROBE1(vfs, namecache, fullpath, miss, vp);
1261 
1262 	CACHE_RUNLOCK();
1263 	vn_lock(*vp, LK_SHARED | LK_RETRY);
1264 	error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1265 	vput(*vp);
1266 	if (error) {
1267 		counter_u64_add(numfullpathfail2, 1);
1268 		SDT_PROBE3(vfs, namecache, fullpath, return,  error, vp, NULL);
1269 		return (error);
1270 	}
1271 
1272 	*vp = dvp;
1273 	CACHE_RLOCK();
1274 	if (dvp->v_iflag & VI_DOOMED) {
1275 		/* forced unmount */
1276 		CACHE_RUNLOCK();
1277 		vrele(dvp);
1278 		error = ENOENT;
1279 		SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
1280 		return (error);
1281 	}
1282 	/*
1283 	 * *vp has its use count incremented still.
1284 	 */
1285 
1286 	return (0);
1287 }
1288 
1289 /*
1290  * The magic behind kern___getcwd() and vn_fullpath().
1291  */
1292 static int
1293 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1294     char *buf, char **retbuf, u_int buflen)
1295 {
1296 	int error, slash_prefixed;
1297 #ifdef KDTRACE_HOOKS
1298 	struct vnode *startvp = vp;
1299 #endif
1300 	struct vnode *vp1;
1301 
1302 	buflen--;
1303 	buf[buflen] = '\0';
1304 	error = 0;
1305 	slash_prefixed = 0;
1306 
1307 	SDT_PROBE1(vfs, namecache, fullpath, entry, vp);
1308 	counter_u64_add(numfullpathcalls, 1);
1309 	vref(vp);
1310 	CACHE_RLOCK();
1311 	if (vp->v_type != VDIR) {
1312 		error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1313 		if (error)
1314 			return (error);
1315 		if (buflen == 0) {
1316 			CACHE_RUNLOCK();
1317 			vrele(vp);
1318 			return (ENOMEM);
1319 		}
1320 		buf[--buflen] = '/';
1321 		slash_prefixed = 1;
1322 	}
1323 	while (vp != rdir && vp != rootvnode) {
1324 		if (vp->v_vflag & VV_ROOT) {
1325 			if (vp->v_iflag & VI_DOOMED) {	/* forced unmount */
1326 				CACHE_RUNLOCK();
1327 				vrele(vp);
1328 				error = ENOENT;
1329 				SDT_PROBE3(vfs, namecache, fullpath, return,
1330 				    error, vp, NULL);
1331 				break;
1332 			}
1333 			vp1 = vp->v_mount->mnt_vnodecovered;
1334 			vref(vp1);
1335 			CACHE_RUNLOCK();
1336 			vrele(vp);
1337 			vp = vp1;
1338 			CACHE_RLOCK();
1339 			continue;
1340 		}
1341 		if (vp->v_type != VDIR) {
1342 			CACHE_RUNLOCK();
1343 			vrele(vp);
1344 			counter_u64_add(numfullpathfail1, 1);
1345 			error = ENOTDIR;
1346 			SDT_PROBE3(vfs, namecache, fullpath, return,
1347 			    error, vp, NULL);
1348 			break;
1349 		}
1350 		error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1351 		if (error)
1352 			break;
1353 		if (buflen == 0) {
1354 			CACHE_RUNLOCK();
1355 			vrele(vp);
1356 			error = ENOMEM;
1357 			SDT_PROBE3(vfs, namecache, fullpath, return, error,
1358 			    startvp, NULL);
1359 			break;
1360 		}
1361 		buf[--buflen] = '/';
1362 		slash_prefixed = 1;
1363 	}
1364 	if (error)
1365 		return (error);
1366 	if (!slash_prefixed) {
1367 		if (buflen == 0) {
1368 			CACHE_RUNLOCK();
1369 			vrele(vp);
1370 			counter_u64_add(numfullpathfail4, 1);
1371 			SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM,
1372 			    startvp, NULL);
1373 			return (ENOMEM);
1374 		}
1375 		buf[--buflen] = '/';
1376 	}
1377 	counter_u64_add(numfullpathfound, 1);
1378 	CACHE_RUNLOCK();
1379 	vrele(vp);
1380 
1381 	SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen);
1382 	*retbuf = buf + buflen;
1383 	return (0);
1384 }
1385 
1386 struct vnode *
1387 vn_dir_dd_ino(struct vnode *vp)
1388 {
1389 	struct namecache *ncp;
1390 	struct vnode *ddvp;
1391 
1392 	ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
1393 	CACHE_RLOCK();
1394 	TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
1395 		if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
1396 			continue;
1397 		ddvp = ncp->nc_dvp;
1398 		vhold(ddvp);
1399 		CACHE_RUNLOCK();
1400 		if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread))
1401 			return (NULL);
1402 		return (ddvp);
1403 	}
1404 	CACHE_RUNLOCK();
1405 	return (NULL);
1406 }
1407 
1408 int
1409 vn_commname(struct vnode *vp, char *buf, u_int buflen)
1410 {
1411 	struct namecache *ncp;
1412 	int l;
1413 
1414 	CACHE_RLOCK();
1415 	TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1416 		if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1417 			break;
1418 	if (ncp == NULL) {
1419 		CACHE_RUNLOCK();
1420 		return (ENOENT);
1421 	}
1422 	l = min(ncp->nc_nlen, buflen - 1);
1423 	memcpy(buf, nc_get_name(ncp), l);
1424 	CACHE_RUNLOCK();
1425 	buf[l] = '\0';
1426 	return (0);
1427 }
1428 
1429 /* ABI compat shims for old kernel modules. */
1430 #undef cache_enter
1431 
1432 void	cache_enter(struct vnode *dvp, struct vnode *vp,
1433 	    struct componentname *cnp);
1434 
1435 void
1436 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
1437 {
1438 
1439 	cache_enter_time(dvp, vp, cnp, NULL, NULL);
1440 }
1441 
1442 /*
1443  * This function updates path string to vnode's full global path
1444  * and checks the size of the new path string against the pathlen argument.
1445  *
1446  * Requires a locked, referenced vnode.
1447  * Vnode is re-locked on success or ENODEV, otherwise unlocked.
1448  *
1449  * If sysctl debug.disablefullpath is set, ENODEV is returned,
1450  * vnode is left locked and path remain untouched.
1451  *
1452  * If vp is a directory, the call to vn_fullpath_global() always succeeds
1453  * because it falls back to the ".." lookup if the namecache lookup fails.
1454  */
1455 int
1456 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
1457     u_int pathlen)
1458 {
1459 	struct nameidata nd;
1460 	struct vnode *vp1;
1461 	char *rpath, *fbuf;
1462 	int error;
1463 
1464 	ASSERT_VOP_ELOCKED(vp, __func__);
1465 
1466 	/* Return ENODEV if sysctl debug.disablefullpath==1 */
1467 	if (disablefullpath)
1468 		return (ENODEV);
1469 
1470 	/* Construct global filesystem path from vp. */
1471 	VOP_UNLOCK(vp, 0);
1472 	error = vn_fullpath_global(td, vp, &rpath, &fbuf);
1473 
1474 	if (error != 0) {
1475 		vrele(vp);
1476 		return (error);
1477 	}
1478 
1479 	if (strlen(rpath) >= pathlen) {
1480 		vrele(vp);
1481 		error = ENAMETOOLONG;
1482 		goto out;
1483 	}
1484 
1485 	/*
1486 	 * Re-lookup the vnode by path to detect a possible rename.
1487 	 * As a side effect, the vnode is relocked.
1488 	 * If vnode was renamed, return ENOENT.
1489 	 */
1490 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1491 	    UIO_SYSSPACE, path, td);
1492 	error = namei(&nd);
1493 	if (error != 0) {
1494 		vrele(vp);
1495 		goto out;
1496 	}
1497 	NDFREE(&nd, NDF_ONLY_PNBUF);
1498 	vp1 = nd.ni_vp;
1499 	vrele(vp);
1500 	if (vp1 == vp)
1501 		strcpy(path, rpath);
1502 	else {
1503 		vput(vp1);
1504 		error = ENOENT;
1505 	}
1506 
1507 out:
1508 	free(fbuf, M_TEMP);
1509 	return (error);
1510 }
1511