FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_cache.c
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_kdtrace.h"
39 #include "opt_ktrace.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.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 cacheing.
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 static 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(mode, name, var, descr) \
277 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, descr);
278 STATNODE(CTLFLAG_RD, numneg, &numneg, "Number of negative cache entries");
279 STATNODE(CTLFLAG_RD, numcache, &numcache, "Number of cache entries");
280 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls,
281 "Number of cache lookups");
282 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits,
283 "Number of '.' hits");
284 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits,
285 "Number of '..' hits");
286 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks,
287 "Number of checks in lookup");
288 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss,
289 "Number of cache misses");
290 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap,
291 "Number of cache misses we do not want to cache");
292 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps,
293 "Number of cache hits (positive) we do not want to cache");
294 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits,
295 "Number of cache hits (positive)");
296 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps,
297 "Number of cache hits (negative) we do not want to cache");
298 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits,
299 "Number of cache hits (negative)");
300 static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades,
301 "Number of updates of the cache after lookup (write lock + retry)");
302
303 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
304 &nchstats, sizeof(nchstats), "LU",
305 "VFS cache effectiveness statistics");
306
307
308
309 static void cache_zap(struct namecache *ncp);
310 static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
311 u_int *buflen);
312 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
313 char *buf, char **retbuf, u_int buflen);
314
315 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
316
317 #ifdef DIAGNOSTIC
318 /*
319 * Grab an atomic snapshot of the name cache hash chain lengths
320 */
321 static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
322 "hash table stats");
323
324 static int
325 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
326 {
327 int error;
328 struct nchashhead *ncpp;
329 struct namecache *ncp;
330 int n_nchash;
331 int count;
332
333 n_nchash = nchash + 1; /* nchash is max index, not count */
334 if (!req->oldptr)
335 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
336
337 /* Scan hash tables for applicable entries */
338 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
339 CACHE_RLOCK();
340 count = 0;
341 LIST_FOREACH(ncp, ncpp, nc_hash) {
342 count++;
343 }
344 CACHE_RUNLOCK();
345 error = SYSCTL_OUT(req, &count, sizeof(count));
346 if (error)
347 return (error);
348 }
349 return (0);
350 }
351 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
352 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
353 "nchash chain lengths");
354
355 static int
356 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
357 {
358 int error;
359 struct nchashhead *ncpp;
360 struct namecache *ncp;
361 int n_nchash;
362 int count, maxlength, used, pct;
363
364 if (!req->oldptr)
365 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
366
367 n_nchash = nchash + 1; /* nchash is max index, not count */
368 used = 0;
369 maxlength = 0;
370
371 /* Scan hash tables for applicable entries */
372 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
373 count = 0;
374 CACHE_RLOCK();
375 LIST_FOREACH(ncp, ncpp, nc_hash) {
376 count++;
377 }
378 CACHE_RUNLOCK();
379 if (count)
380 used++;
381 if (maxlength < count)
382 maxlength = count;
383 }
384 n_nchash = nchash + 1;
385 pct = (used * 100 * 100) / n_nchash;
386 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
387 if (error)
388 return (error);
389 error = SYSCTL_OUT(req, &used, sizeof(used));
390 if (error)
391 return (error);
392 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
393 if (error)
394 return (error);
395 error = SYSCTL_OUT(req, &pct, sizeof(pct));
396 if (error)
397 return (error);
398 return (0);
399 }
400 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
401 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
402 "nchash chain lengths");
403 #endif
404
405 /*
406 * cache_zap():
407 *
408 * Removes a namecache entry from cache, whether it contains an actual
409 * pointer to a vnode or if it is just a negative cache entry.
410 */
411 static void
412 cache_zap(ncp)
413 struct namecache *ncp;
414 {
415 struct vnode *vp;
416
417 rw_assert(&cache_lock, RA_WLOCKED);
418 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
419 #ifdef KDTRACE_HOOKS
420 if (ncp->nc_vp != NULL) {
421 SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp,
422 nc_get_name(ncp), ncp->nc_vp);
423 } else {
424 SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp,
425 nc_get_name(ncp));
426 }
427 #endif
428 vp = NULL;
429 LIST_REMOVE(ncp, nc_hash);
430 if (ncp->nc_flag & NCF_ISDOTDOT) {
431 if (ncp == ncp->nc_dvp->v_cache_dd)
432 ncp->nc_dvp->v_cache_dd = NULL;
433 } else {
434 LIST_REMOVE(ncp, nc_src);
435 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
436 vp = ncp->nc_dvp;
437 numcachehv--;
438 }
439 }
440 if (ncp->nc_vp) {
441 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
442 if (ncp == ncp->nc_vp->v_cache_dd)
443 ncp->nc_vp->v_cache_dd = NULL;
444 } else {
445 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
446 numneg--;
447 }
448 numcache--;
449 cache_free(ncp);
450 if (vp)
451 vdrop(vp);
452 }
453
454 /*
455 * Lookup an entry in the cache
456 *
457 * Lookup is called with dvp pointing to the directory to search,
458 * cnp pointing to the name of the entry being sought. If the lookup
459 * succeeds, the vnode is returned in *vpp, and a status of -1 is
460 * returned. If the lookup determines that the name does not exist
461 * (negative cacheing), a status of ENOENT is returned. If the lookup
462 * fails, a status of zero is returned. If the directory vnode is
463 * recycled out from under us due to a forced unmount, a status of
464 * ENOENT is returned.
465 *
466 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
467 * unlocked. If we're looking up . an extra ref is taken, but the lock is
468 * not recursively acquired.
469 */
470
471 int
472 cache_lookup_times(dvp, vpp, cnp, tsp, ticksp)
473 struct vnode *dvp;
474 struct vnode **vpp;
475 struct componentname *cnp;
476 struct timespec *tsp;
477 int *ticksp;
478 {
479 struct namecache *ncp;
480 uint32_t hash;
481 int error, ltype, wlocked;
482
483 if (!doingcache) {
484 cnp->cn_flags &= ~MAKEENTRY;
485 return (0);
486 }
487 retry:
488 CACHE_RLOCK();
489 wlocked = 0;
490 numcalls++;
491 error = 0;
492
493 retry_wlocked:
494 if (cnp->cn_nameptr[0] == '.') {
495 if (cnp->cn_namelen == 1) {
496 *vpp = dvp;
497 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
498 dvp, cnp->cn_nameptr);
499 dothits++;
500 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp);
501 if (tsp != NULL)
502 timespecclear(tsp);
503 if (ticksp != NULL)
504 *ticksp = ticks;
505 goto success;
506 }
507 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
508 dotdothits++;
509 if (dvp->v_cache_dd == NULL) {
510 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
511 "..", NULL);
512 goto unlock;
513 }
514 if ((cnp->cn_flags & MAKEENTRY) == 0) {
515 if (!wlocked && !CACHE_UPGRADE_LOCK())
516 goto wlock;
517 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
518 cache_zap(dvp->v_cache_dd);
519 dvp->v_cache_dd = NULL;
520 CACHE_WUNLOCK();
521 return (0);
522 }
523 ncp = dvp->v_cache_dd;
524 if (ncp->nc_flag & NCF_ISDOTDOT)
525 *vpp = ncp->nc_vp;
526 else
527 *vpp = ncp->nc_dvp;
528 /* Return failure if negative entry was found. */
529 if (*vpp == NULL)
530 goto negative_success;
531 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
532 dvp, cnp->cn_nameptr, *vpp);
533 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..",
534 *vpp);
535 cache_out_ts(ncp, tsp, ticksp);
536 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
537 NCF_DTS && tsp != NULL)
538 *tsp = ((struct namecache_ts *)ncp)->
539 nc_dotdottime;
540 goto success;
541 }
542 }
543
544 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
545 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
546 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
547 numchecks++;
548 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
549 !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen))
550 break;
551 }
552
553 /* We failed to find an entry */
554 if (ncp == NULL) {
555 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
556 NULL);
557 if ((cnp->cn_flags & MAKEENTRY) == 0) {
558 nummisszap++;
559 } else {
560 nummiss++;
561 }
562 nchstats.ncs_miss++;
563 goto unlock;
564 }
565
566 /* We don't want to have an entry, so dump it */
567 if ((cnp->cn_flags & MAKEENTRY) == 0) {
568 numposzaps++;
569 nchstats.ncs_badhits++;
570 if (!wlocked && !CACHE_UPGRADE_LOCK())
571 goto wlock;
572 cache_zap(ncp);
573 CACHE_WUNLOCK();
574 return (0);
575 }
576
577 /* We found a "positive" match, return the vnode */
578 if (ncp->nc_vp) {
579 numposhits++;
580 nchstats.ncs_goodhits++;
581 *vpp = ncp->nc_vp;
582 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
583 dvp, cnp->cn_nameptr, *vpp, ncp);
584 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp),
585 *vpp);
586 cache_out_ts(ncp, tsp, ticksp);
587 goto success;
588 }
589
590 negative_success:
591 /* We found a negative match, and want to create it, so purge */
592 if (cnp->cn_nameiop == CREATE) {
593 numnegzaps++;
594 nchstats.ncs_badhits++;
595 if (!wlocked && !CACHE_UPGRADE_LOCK())
596 goto wlock;
597 cache_zap(ncp);
598 CACHE_WUNLOCK();
599 return (0);
600 }
601
602 if (!wlocked && !CACHE_UPGRADE_LOCK())
603 goto wlock;
604 numneghits++;
605 /*
606 * We found a "negative" match, so we shift it to the end of
607 * the "negative" cache entries queue to satisfy LRU. Also,
608 * check to see if the entry is a whiteout; indicate this to
609 * the componentname, if so.
610 */
611 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
612 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
613 nchstats.ncs_neghits++;
614 if (ncp->nc_flag & NCF_WHITE)
615 cnp->cn_flags |= ISWHITEOUT;
616 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp,
617 nc_get_name(ncp));
618 cache_out_ts(ncp, tsp, ticksp);
619 CACHE_WUNLOCK();
620 return (ENOENT);
621
622 wlock:
623 /*
624 * We need to update the cache after our lookup, so upgrade to
625 * a write lock and retry the operation.
626 */
627 CACHE_RUNLOCK();
628 CACHE_WLOCK();
629 numupgrades++;
630 wlocked = 1;
631 goto retry_wlocked;
632
633 success:
634 /*
635 * On success we return a locked and ref'd vnode as per the lookup
636 * protocol.
637 */
638 if (dvp == *vpp) { /* lookup on "." */
639 VREF(*vpp);
640 if (wlocked)
641 CACHE_WUNLOCK();
642 else
643 CACHE_RUNLOCK();
644 /*
645 * When we lookup "." we still can be asked to lock it
646 * differently...
647 */
648 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
649 if (ltype != VOP_ISLOCKED(*vpp)) {
650 if (ltype == LK_EXCLUSIVE) {
651 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
652 if ((*vpp)->v_iflag & VI_DOOMED) {
653 /* forced unmount */
654 vrele(*vpp);
655 *vpp = NULL;
656 return (ENOENT);
657 }
658 } else
659 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
660 }
661 return (-1);
662 }
663 ltype = 0; /* silence gcc warning */
664 if (cnp->cn_flags & ISDOTDOT) {
665 ltype = VOP_ISLOCKED(dvp);
666 VOP_UNLOCK(dvp, 0);
667 }
668 VI_LOCK(*vpp);
669 if (wlocked)
670 CACHE_WUNLOCK();
671 else
672 CACHE_RUNLOCK();
673 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
674 if (cnp->cn_flags & ISDOTDOT) {
675 vn_lock(dvp, ltype | LK_RETRY);
676 if (dvp->v_iflag & VI_DOOMED) {
677 if (error == 0)
678 vput(*vpp);
679 *vpp = NULL;
680 return (ENOENT);
681 }
682 }
683 if (error) {
684 *vpp = NULL;
685 goto retry;
686 }
687 if ((cnp->cn_flags & ISLASTCN) &&
688 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
689 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
690 }
691 return (-1);
692
693 unlock:
694 if (wlocked)
695 CACHE_WUNLOCK();
696 else
697 CACHE_RUNLOCK();
698 return (0);
699 }
700
701 /*
702 * Add an entry to the cache.
703 */
704 void
705 cache_enter_time(dvp, vp, cnp, tsp, dtsp)
706 struct vnode *dvp;
707 struct vnode *vp;
708 struct componentname *cnp;
709 struct timespec *tsp;
710 struct timespec *dtsp;
711 {
712 struct namecache *ncp, *n2;
713 struct namecache_ts *n3;
714 struct nchashhead *ncpp;
715 uint32_t hash;
716 int flag;
717 int hold;
718 int zap;
719 int len;
720
721 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
722 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
723 ("cache_enter: Adding a doomed vnode"));
724 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
725 ("cache_enter: Doomed vnode used as src"));
726
727 if (!doingcache)
728 return;
729
730 /*
731 * Avoid blowout in namecache entries.
732 */
733 if (numcache >= desiredvnodes * ncsizefactor)
734 return;
735
736 flag = 0;
737 if (cnp->cn_nameptr[0] == '.') {
738 if (cnp->cn_namelen == 1)
739 return;
740 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
741 CACHE_WLOCK();
742 /*
743 * If dotdot entry already exists, just retarget it
744 * to new parent vnode, otherwise continue with new
745 * namecache entry allocation.
746 */
747 if ((ncp = dvp->v_cache_dd) != NULL &&
748 ncp->nc_flag & NCF_ISDOTDOT) {
749 KASSERT(ncp->nc_dvp == dvp,
750 ("wrong isdotdot parent"));
751 if (ncp->nc_vp != NULL) {
752 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
753 ncp, nc_dst);
754 } else {
755 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
756 numneg--;
757 }
758 if (vp != NULL) {
759 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
760 ncp, nc_dst);
761 } else {
762 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
763 numneg++;
764 }
765 ncp->nc_vp = vp;
766 CACHE_WUNLOCK();
767 return;
768 }
769 dvp->v_cache_dd = NULL;
770 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp);
771 CACHE_WUNLOCK();
772 flag = NCF_ISDOTDOT;
773 }
774 }
775
776 hold = 0;
777 zap = 0;
778
779 /*
780 * Calculate the hash key and setup as much of the new
781 * namecache entry as possible before acquiring the lock.
782 */
783 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
784 ncp->nc_vp = vp;
785 ncp->nc_dvp = dvp;
786 ncp->nc_flag = flag;
787 if (tsp != NULL) {
788 n3 = (struct namecache_ts *)ncp;
789 n3->nc_time = *tsp;
790 n3->nc_ticks = ticks;
791 n3->nc_flag |= NCF_TS;
792 if (dtsp != NULL) {
793 n3->nc_dotdottime = *dtsp;
794 n3->nc_flag |= NCF_DTS;
795 }
796 }
797 len = ncp->nc_nlen = cnp->cn_namelen;
798 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
799 strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1);
800 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
801 CACHE_WLOCK();
802
803 /*
804 * See if this vnode or negative entry is already in the cache
805 * with this name. This can happen with concurrent lookups of
806 * the same path name.
807 */
808 ncpp = NCHHASH(hash);
809 LIST_FOREACH(n2, ncpp, nc_hash) {
810 if (n2->nc_dvp == dvp &&
811 n2->nc_nlen == cnp->cn_namelen &&
812 !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) {
813 if (tsp != NULL) {
814 KASSERT((n2->nc_flag & NCF_TS) != 0,
815 ("no NCF_TS"));
816 n3 = (struct namecache_ts *)n2;
817 n3->nc_time =
818 ((struct namecache_ts *)ncp)->nc_time;
819 n3->nc_ticks =
820 ((struct namecache_ts *)ncp)->nc_ticks;
821 if (dtsp != NULL) {
822 n3->nc_dotdottime =
823 ((struct namecache_ts *)ncp)->
824 nc_dotdottime;
825 n3->nc_flag |= NCF_DTS;
826 }
827 }
828 CACHE_WUNLOCK();
829 cache_free(ncp);
830 return;
831 }
832 }
833
834 if (flag == NCF_ISDOTDOT) {
835 /*
836 * See if we are trying to add .. entry, but some other lookup
837 * has populated v_cache_dd pointer already.
838 */
839 if (dvp->v_cache_dd != NULL) {
840 CACHE_WUNLOCK();
841 cache_free(ncp);
842 return;
843 }
844 KASSERT(vp == NULL || vp->v_type == VDIR,
845 ("wrong vnode type %p", vp));
846 dvp->v_cache_dd = ncp;
847 }
848
849 numcache++;
850 if (!vp) {
851 numneg++;
852 if (cnp->cn_flags & ISWHITEOUT)
853 ncp->nc_flag |= NCF_WHITE;
854 } else 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 * Insert the new namecache entry into the appropriate chain
872 * within the cache entries table.
873 */
874 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
875 if (flag != NCF_ISDOTDOT) {
876 if (LIST_EMPTY(&dvp->v_cache_src)) {
877 hold = 1;
878 numcachehv++;
879 }
880 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
881 }
882
883 /*
884 * If the entry is "negative", we place it into the
885 * "negative" cache queue, otherwise, we place it into the
886 * destination vnode's cache entries queue.
887 */
888 if (vp) {
889 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
890 SDT_PROBE3(vfs, namecache, enter, done, dvp, nc_get_name(ncp),
891 vp);
892 } else {
893 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
894 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp,
895 nc_get_name(ncp));
896 }
897 if (numneg * ncnegfactor > numcache) {
898 ncp = TAILQ_FIRST(&ncneg);
899 KASSERT(ncp->nc_vp == NULL, ("ncp %p vp %p on ncneg",
900 ncp, ncp->nc_vp));
901 zap = 1;
902 }
903 if (hold)
904 vhold(dvp);
905 if (zap)
906 cache_zap(ncp);
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 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
935
936
937 /*
938 * Invalidate all entries to a particular vnode.
939 */
940 void
941 cache_purge(vp)
942 struct vnode *vp;
943 {
944
945 CTR1(KTR_VFS, "cache_purge(%p)", vp);
946 SDT_PROBE1(vfs, namecache, purge, done, vp);
947 CACHE_WLOCK();
948 while (!LIST_EMPTY(&vp->v_cache_src))
949 cache_zap(LIST_FIRST(&vp->v_cache_src));
950 while (!TAILQ_EMPTY(&vp->v_cache_dst))
951 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
952 if (vp->v_cache_dd != NULL) {
953 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
954 ("lost dotdot link"));
955 cache_zap(vp->v_cache_dd);
956 }
957 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
958 CACHE_WUNLOCK();
959 }
960
961 /*
962 * Invalidate all negative entries for a particular directory vnode.
963 */
964 void
965 cache_purge_negative(vp)
966 struct vnode *vp;
967 {
968 struct namecache *cp, *ncp;
969
970 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
971 SDT_PROBE1(vfs, namecache, purge_negative, done, vp);
972 CACHE_WLOCK();
973 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
974 if (cp->nc_vp == NULL)
975 cache_zap(cp);
976 }
977 CACHE_WUNLOCK();
978 }
979
980 /*
981 * Flush all entries referencing a particular filesystem.
982 */
983 void
984 cache_purgevfs(mp)
985 struct mount *mp;
986 {
987 struct nchashhead *ncpp;
988 struct namecache *ncp, *nnp;
989
990 /* Scan hash tables for applicable entries */
991 SDT_PROBE1(vfs, namecache, purgevfs, done, mp);
992 CACHE_WLOCK();
993 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
994 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
995 if (ncp->nc_dvp->v_mount == mp)
996 cache_zap(ncp);
997 }
998 }
999 CACHE_WUNLOCK();
1000 }
1001
1002 /*
1003 * Perform canonical checks and cache lookup and pass on to filesystem
1004 * through the vop_cachedlookup only if needed.
1005 */
1006
1007 int
1008 vfs_cache_lookup(ap)
1009 struct vop_lookup_args /* {
1010 struct vnode *a_dvp;
1011 struct vnode **a_vpp;
1012 struct componentname *a_cnp;
1013 } */ *ap;
1014 {
1015 struct vnode *dvp;
1016 int error;
1017 struct vnode **vpp = ap->a_vpp;
1018 struct componentname *cnp = ap->a_cnp;
1019 struct ucred *cred = cnp->cn_cred;
1020 int flags = cnp->cn_flags;
1021 struct thread *td = cnp->cn_thread;
1022
1023 *vpp = NULL;
1024 dvp = ap->a_dvp;
1025
1026 if (dvp->v_type != VDIR)
1027 return (ENOTDIR);
1028
1029 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1030 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1031 return (EROFS);
1032
1033 error = VOP_ACCESS(dvp, VEXEC, cred, td);
1034 if (error)
1035 return (error);
1036
1037 error = cache_lookup(dvp, vpp, cnp);
1038 if (error == 0)
1039 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
1040 if (error == -1)
1041 return (0);
1042 return (error);
1043 }
1044
1045 /*
1046 * XXX All of these sysctls would probably be more productive dead.
1047 */
1048 static int disablecwd;
1049 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
1050 "Disable the getcwd syscall");
1051
1052 /* Implementation of the getcwd syscall. */
1053 int
1054 sys___getcwd(td, uap)
1055 struct thread *td;
1056 struct __getcwd_args *uap;
1057 {
1058
1059 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
1060 }
1061
1062 int
1063 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen)
1064 {
1065 char *bp, *tmpbuf;
1066 struct filedesc *fdp;
1067 struct vnode *cdir, *rdir;
1068 int error, vfslocked;
1069
1070 if (disablecwd)
1071 return (ENODEV);
1072 if (buflen < 2)
1073 return (EINVAL);
1074 if (buflen > MAXPATHLEN)
1075 buflen = MAXPATHLEN;
1076
1077 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
1078 fdp = td->td_proc->p_fd;
1079 FILEDESC_SLOCK(fdp);
1080 cdir = fdp->fd_cdir;
1081 VREF(cdir);
1082 rdir = fdp->fd_rdir;
1083 VREF(rdir);
1084 FILEDESC_SUNLOCK(fdp);
1085 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
1086 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
1087 vrele(rdir);
1088 VFS_UNLOCK_GIANT(vfslocked);
1089 vfslocked = VFS_LOCK_GIANT(cdir->v_mount);
1090 vrele(cdir);
1091 VFS_UNLOCK_GIANT(vfslocked);
1092
1093 if (!error) {
1094 if (bufseg == UIO_SYSSPACE)
1095 bcopy(bp, buf, strlen(bp) + 1);
1096 else
1097 error = copyout(bp, buf, strlen(bp) + 1);
1098 #ifdef KTRACE
1099 if (KTRPOINT(curthread, KTR_NAMEI))
1100 ktrnamei(bp);
1101 #endif
1102 }
1103 free(tmpbuf, M_TEMP);
1104 return (error);
1105 }
1106
1107 /*
1108 * Thus begins the fullpath magic.
1109 */
1110
1111 #undef STATNODE
1112 #define STATNODE(name, descr) \
1113 static u_int name; \
1114 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr)
1115
1116 static int disablefullpath;
1117 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
1118 "Disable the vn_fullpath function");
1119
1120 /* These count for kern___getcwd(), too. */
1121 STATNODE(numfullpathcalls, "Number of fullpath search calls");
1122 STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
1123 STATNODE(numfullpathfail2,
1124 "Number of fullpath search errors (VOP_VPTOCNP failures)");
1125 STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
1126 STATNODE(numfullpathfound, "Number of successful fullpath calls");
1127
1128 /*
1129 * Retrieve the full filesystem path that correspond to a vnode from the name
1130 * cache (if available)
1131 */
1132 int
1133 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
1134 {
1135 char *buf;
1136 struct filedesc *fdp;
1137 struct vnode *rdir;
1138 int error, vfslocked;
1139
1140 if (disablefullpath)
1141 return (ENODEV);
1142 if (vn == NULL)
1143 return (EINVAL);
1144
1145 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1146 fdp = td->td_proc->p_fd;
1147 FILEDESC_SLOCK(fdp);
1148 rdir = fdp->fd_rdir;
1149 VREF(rdir);
1150 FILEDESC_SUNLOCK(fdp);
1151 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1152 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
1153 vrele(rdir);
1154 VFS_UNLOCK_GIANT(vfslocked);
1155
1156 if (!error)
1157 *freebuf = buf;
1158 else
1159 free(buf, M_TEMP);
1160 return (error);
1161 }
1162
1163 /*
1164 * This function is similar to vn_fullpath, but it attempts to lookup the
1165 * pathname relative to the global root mount point. This is required for the
1166 * auditing sub-system, as audited pathnames must be absolute, relative to the
1167 * global root mount point.
1168 */
1169 int
1170 vn_fullpath_global(struct thread *td, struct vnode *vn,
1171 char **retbuf, char **freebuf)
1172 {
1173 char *buf;
1174 int error;
1175
1176 if (disablefullpath)
1177 return (ENODEV);
1178 if (vn == NULL)
1179 return (EINVAL);
1180 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1181 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1182 if (!error)
1183 *freebuf = buf;
1184 else
1185 free(buf, M_TEMP);
1186 return (error);
1187 }
1188
1189 int
1190 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1191 {
1192 int error;
1193
1194 CACHE_RLOCK();
1195 error = vn_vptocnp_locked(vp, cred, buf, buflen);
1196 if (error == 0)
1197 CACHE_RUNLOCK();
1198 return (error);
1199 }
1200
1201 static int
1202 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1203 u_int *buflen)
1204 {
1205 struct vnode *dvp;
1206 struct namecache *ncp;
1207 int error, vfslocked;
1208
1209 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1210 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1211 break;
1212 }
1213 if (ncp != NULL) {
1214 if (*buflen < ncp->nc_nlen) {
1215 CACHE_RUNLOCK();
1216 vfslocked = VFS_LOCK_GIANT((*vp)->v_mount);
1217 vrele(*vp);
1218 VFS_UNLOCK_GIANT(vfslocked);
1219 numfullpathfail4++;
1220 error = ENOMEM;
1221 SDT_PROBE3(vfs, namecache, fullpath, return, error,
1222 vp, NULL);
1223 return (error);
1224 }
1225 *buflen -= ncp->nc_nlen;
1226 memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen);
1227 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1228 nc_get_name(ncp), vp);
1229 dvp = *vp;
1230 *vp = ncp->nc_dvp;
1231 vref(*vp);
1232 CACHE_RUNLOCK();
1233 vfslocked = VFS_LOCK_GIANT(dvp->v_mount);
1234 vrele(dvp);
1235 VFS_UNLOCK_GIANT(vfslocked);
1236 CACHE_RLOCK();
1237 return (0);
1238 }
1239 SDT_PROBE1(vfs, namecache, fullpath, miss, vp);
1240
1241 CACHE_RUNLOCK();
1242 vfslocked = VFS_LOCK_GIANT((*vp)->v_mount);
1243 vn_lock(*vp, LK_SHARED | LK_RETRY);
1244 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1245 vput(*vp);
1246 VFS_UNLOCK_GIANT(vfslocked);
1247 if (error) {
1248 numfullpathfail2++;
1249 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
1250 return (error);
1251 }
1252
1253 *vp = dvp;
1254 CACHE_RLOCK();
1255 if (dvp->v_iflag & VI_DOOMED) {
1256 /* forced unmount */
1257 CACHE_RUNLOCK();
1258 vfslocked = VFS_LOCK_GIANT(dvp->v_mount);
1259 vrele(dvp);
1260 VFS_UNLOCK_GIANT(vfslocked);
1261 error = ENOENT;
1262 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
1263 return (error);
1264 }
1265 /*
1266 * *vp has its use count incremented still.
1267 */
1268
1269 return (0);
1270 }
1271
1272 /*
1273 * The magic behind kern___getcwd() and vn_fullpath().
1274 */
1275 static int
1276 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1277 char *buf, char **retbuf, u_int buflen)
1278 {
1279 int error, slash_prefixed, vfslocked;
1280 #ifdef KDTRACE_HOOKS
1281 struct vnode *startvp = vp;
1282 #endif
1283 struct vnode *vp1;
1284
1285 buflen--;
1286 buf[buflen] = '\0';
1287 error = 0;
1288 slash_prefixed = 0;
1289
1290 SDT_PROBE1(vfs, namecache, fullpath, entry, vp);
1291 numfullpathcalls++;
1292 vref(vp);
1293 CACHE_RLOCK();
1294 if (vp->v_type != VDIR) {
1295 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1296 if (error)
1297 return (error);
1298 if (buflen == 0) {
1299 CACHE_RUNLOCK();
1300 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1301 vrele(vp);
1302 VFS_UNLOCK_GIANT(vfslocked);
1303 return (ENOMEM);
1304 }
1305 buf[--buflen] = '/';
1306 slash_prefixed = 1;
1307 }
1308 while (vp != rdir && vp != rootvnode) {
1309 if (vp->v_vflag & VV_ROOT) {
1310 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */
1311 CACHE_RUNLOCK();
1312 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1313 vrele(vp);
1314 VFS_UNLOCK_GIANT(vfslocked);
1315 error = ENOENT;
1316 SDT_PROBE3(vfs, namecache, fullpath, return,
1317 error, vp, NULL);
1318 break;
1319 }
1320 vp1 = vp->v_mount->mnt_vnodecovered;
1321 vref(vp1);
1322 CACHE_RUNLOCK();
1323 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1324 vrele(vp);
1325 VFS_UNLOCK_GIANT(vfslocked);
1326 vp = vp1;
1327 CACHE_RLOCK();
1328 continue;
1329 }
1330 if (vp->v_type != VDIR) {
1331 CACHE_RUNLOCK();
1332 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1333 vrele(vp);
1334 VFS_UNLOCK_GIANT(vfslocked);
1335 numfullpathfail1++;
1336 error = ENOTDIR;
1337 SDT_PROBE3(vfs, namecache, fullpath, return,
1338 error, vp, NULL);
1339 break;
1340 }
1341 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1342 if (error)
1343 break;
1344 if (buflen == 0) {
1345 CACHE_RUNLOCK();
1346 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1347 vrele(vp);
1348 VFS_UNLOCK_GIANT(vfslocked);
1349 error = ENOMEM;
1350 SDT_PROBE3(vfs, namecache, fullpath, return, error,
1351 startvp, NULL);
1352 break;
1353 }
1354 buf[--buflen] = '/';
1355 slash_prefixed = 1;
1356 }
1357 if (error)
1358 return (error);
1359 if (!slash_prefixed) {
1360 if (buflen == 0) {
1361 CACHE_RUNLOCK();
1362 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1363 vrele(vp);
1364 VFS_UNLOCK_GIANT(vfslocked);
1365 numfullpathfail4++;
1366 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM,
1367 startvp, NULL);
1368 return (ENOMEM);
1369 }
1370 buf[--buflen] = '/';
1371 }
1372 numfullpathfound++;
1373 CACHE_RUNLOCK();
1374 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1375 vrele(vp);
1376 VFS_UNLOCK_GIANT(vfslocked);
1377
1378 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen);
1379 *retbuf = buf + buflen;
1380 return (0);
1381 }
1382
1383 struct vnode *
1384 vn_dir_dd_ino(struct vnode *vp)
1385 {
1386 struct namecache *ncp;
1387 struct vnode *ddvp;
1388
1389 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
1390 CACHE_RLOCK();
1391 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
1392 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
1393 continue;
1394 ddvp = ncp->nc_dvp;
1395 VI_LOCK(ddvp);
1396 CACHE_RUNLOCK();
1397 if (vget(ddvp, LK_INTERLOCK | LK_SHARED | LK_NOWAIT, curthread))
1398 return (NULL);
1399 return (ddvp);
1400 }
1401 CACHE_RUNLOCK();
1402 return (NULL);
1403 }
1404
1405 int
1406 vn_commname(struct vnode *vp, char *buf, u_int buflen)
1407 {
1408 struct namecache *ncp;
1409 int l;
1410
1411 CACHE_RLOCK();
1412 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1413 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1414 break;
1415 if (ncp == NULL) {
1416 CACHE_RUNLOCK();
1417 return (ENOENT);
1418 }
1419 l = min(ncp->nc_nlen, buflen - 1);
1420 memcpy(buf, nc_get_name(ncp), l);
1421 CACHE_RUNLOCK();
1422 buf[l] = '\0';
1423 return (0);
1424 }
1425
1426 /* ABI compat shims for old kernel modules. */
1427 #undef cache_enter
1428 #undef cache_lookup
1429
1430 void cache_enter(struct vnode *dvp, struct vnode *vp,
1431 struct componentname *cnp);
1432 int cache_lookup(struct vnode *dvp, struct vnode **vpp,
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 int
1443 cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1444 {
1445
1446 return (cache_lookup_times(dvp, vpp, cnp, NULL, NULL));
1447 }
1448
1449 /*
1450 * This function updates path string to vnode's full global path
1451 * and checks the size of the new path string against the pathlen argument.
1452 *
1453 * Requires a locked, referenced vnode and GIANT lock held.
1454 * Vnode is re-locked on success or ENODEV, otherwise unlocked.
1455 *
1456 * If sysctl debug.disablefullpath is set, ENODEV is returned,
1457 * vnode is left locked and path remain untouched.
1458 *
1459 * If vp is a directory, the call to vn_fullpath_global() always succeeds
1460 * because it falls back to the ".." lookup if the namecache lookup fails.
1461 */
1462 int
1463 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
1464 u_int pathlen)
1465 {
1466 struct nameidata nd;
1467 struct vnode *vp1;
1468 char *rpath, *fbuf;
1469 int error, vfslocked;
1470
1471 VFS_ASSERT_GIANT(vp->v_mount);
1472 ASSERT_VOP_ELOCKED(vp, __func__);
1473
1474 /* Return ENODEV if sysctl debug.disablefullpath==1 */
1475 if (disablefullpath)
1476 return (ENODEV);
1477
1478 /* Construct global filesystem path from vp. */
1479 VOP_UNLOCK(vp, 0);
1480 error = vn_fullpath_global(td, vp, &rpath, &fbuf);
1481
1482 if (error != 0) {
1483 vrele(vp);
1484 return (error);
1485 }
1486
1487 if (strlen(rpath) >= pathlen) {
1488 vrele(vp);
1489 error = ENAMETOOLONG;
1490 goto out;
1491 }
1492
1493 /*
1494 * Re-lookup the vnode by path to detect a possible rename.
1495 * As a side effect, the vnode is relocked.
1496 * If vnode was renamed, return ENOENT.
1497 */
1498 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | MPSAFE | AUDITVNODE1,
1499 UIO_SYSSPACE, path, td);
1500 error = namei(&nd);
1501 if (error != 0) {
1502 vrele(vp);
1503 goto out;
1504 }
1505 vfslocked = NDHASGIANT(&nd);
1506 NDFREE(&nd, NDF_ONLY_PNBUF);
1507 vp1 = nd.ni_vp;
1508 vrele(vp);
1509 if (vp1 == vp)
1510 strcpy(path, rpath);
1511 else {
1512 vput(vp1);
1513 error = ENOENT;
1514 }
1515 VFS_UNLOCK_GIANT(vfslocked);
1516
1517 out:
1518 free(fbuf, M_TEMP);
1519 return (error);
1520 }
Cache object: db53fef04938acaf66c4eb6310cf50c8
|