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: releng/10.3/sys/kern/vfs_cache.c 289798 2015-10-23 07:40:43Z avg $");
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 struct nchashhead *ncpp;
328 struct namecache *ncp;
329 int i, error, n_nchash, *cntbuf;
330
331 retry:
332 n_nchash = nchash + 1; /* nchash is max index, not count */
333 if (!req->oldptr)
334 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
335 cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK);
336 CACHE_RLOCK();
337 if (n_nchash != nchash + 1) {
338 CACHE_RUNLOCK();
339 free(cntbuf, M_TEMP);
340 goto retry;
341 }
342 /* Scan hash tables counting entries */
343 for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++)
344 LIST_FOREACH(ncp, ncpp, nc_hash)
345 cntbuf[i]++;
346 CACHE_RUNLOCK();
347 for (error = 0, i = 0; i < n_nchash; i++)
348 if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0)
349 break;
350 free(cntbuf, M_TEMP);
351 return (error);
352 }
353 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
354 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
355 "nchash chain lengths");
356
357 static int
358 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
359 {
360 int error;
361 struct nchashhead *ncpp;
362 struct namecache *ncp;
363 int n_nchash;
364 int count, maxlength, used, pct;
365
366 if (!req->oldptr)
367 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
368
369 n_nchash = nchash + 1; /* nchash is max index, not count */
370 used = 0;
371 maxlength = 0;
372
373 /* Scan hash tables for applicable entries */
374 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
375 count = 0;
376 CACHE_RLOCK();
377 LIST_FOREACH(ncp, ncpp, nc_hash) {
378 count++;
379 }
380 CACHE_RUNLOCK();
381 if (count)
382 used++;
383 if (maxlength < count)
384 maxlength = count;
385 }
386 n_nchash = nchash + 1;
387 pct = (used * 100) / (n_nchash / 100);
388 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
389 if (error)
390 return (error);
391 error = SYSCTL_OUT(req, &used, sizeof(used));
392 if (error)
393 return (error);
394 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
395 if (error)
396 return (error);
397 error = SYSCTL_OUT(req, &pct, sizeof(pct));
398 if (error)
399 return (error);
400 return (0);
401 }
402 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
403 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
404 "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)");
405 #endif
406
407 /*
408 * cache_zap():
409 *
410 * Removes a namecache entry from cache, whether it contains an actual
411 * pointer to a vnode or if it is just a negative cache entry.
412 */
413 static void
414 cache_zap(ncp)
415 struct namecache *ncp;
416 {
417 struct vnode *vp;
418
419 rw_assert(&cache_lock, RA_WLOCKED);
420 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
421 #ifdef KDTRACE_HOOKS
422 if (ncp->nc_vp != NULL) {
423 SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp,
424 nc_get_name(ncp), ncp->nc_vp);
425 } else {
426 SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp,
427 nc_get_name(ncp));
428 }
429 #endif
430 vp = NULL;
431 LIST_REMOVE(ncp, nc_hash);
432 if (ncp->nc_flag & NCF_ISDOTDOT) {
433 if (ncp == ncp->nc_dvp->v_cache_dd)
434 ncp->nc_dvp->v_cache_dd = NULL;
435 } else {
436 LIST_REMOVE(ncp, nc_src);
437 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
438 vp = ncp->nc_dvp;
439 numcachehv--;
440 }
441 }
442 if (ncp->nc_vp) {
443 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
444 if (ncp == ncp->nc_vp->v_cache_dd)
445 ncp->nc_vp->v_cache_dd = NULL;
446 } else {
447 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
448 numneg--;
449 }
450 numcache--;
451 cache_free(ncp);
452 if (vp)
453 vdrop(vp);
454 }
455
456 /*
457 * Lookup an entry in the cache
458 *
459 * Lookup is called with dvp pointing to the directory to search,
460 * cnp pointing to the name of the entry being sought. If the lookup
461 * succeeds, the vnode is returned in *vpp, and a status of -1 is
462 * returned. If the lookup determines that the name does not exist
463 * (negative cacheing), a status of ENOENT is returned. If the lookup
464 * fails, a status of zero is returned. If the directory vnode is
465 * recycled out from under us due to a forced unmount, a status of
466 * ENOENT is returned.
467 *
468 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
469 * unlocked. If we're looking up . an extra ref is taken, but the lock is
470 * not recursively acquired.
471 */
472
473 int
474 cache_lookup(dvp, vpp, cnp, tsp, ticksp)
475 struct vnode *dvp;
476 struct vnode **vpp;
477 struct componentname *cnp;
478 struct timespec *tsp;
479 int *ticksp;
480 {
481 struct namecache *ncp;
482 uint32_t hash;
483 int error, ltype, wlocked;
484
485 if (!doingcache) {
486 cnp->cn_flags &= ~MAKEENTRY;
487 return (0);
488 }
489 retry:
490 CACHE_RLOCK();
491 wlocked = 0;
492 numcalls++;
493 error = 0;
494
495 retry_wlocked:
496 if (cnp->cn_nameptr[0] == '.') {
497 if (cnp->cn_namelen == 1) {
498 *vpp = dvp;
499 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
500 dvp, cnp->cn_nameptr);
501 dothits++;
502 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp);
503 if (tsp != NULL)
504 timespecclear(tsp);
505 if (ticksp != NULL)
506 *ticksp = ticks;
507 goto success;
508 }
509 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
510 dotdothits++;
511 if (dvp->v_cache_dd == NULL) {
512 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
513 "..", NULL);
514 goto unlock;
515 }
516 if ((cnp->cn_flags & MAKEENTRY) == 0) {
517 if (!wlocked && !CACHE_UPGRADE_LOCK())
518 goto wlock;
519 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
520 cache_zap(dvp->v_cache_dd);
521 dvp->v_cache_dd = NULL;
522 CACHE_WUNLOCK();
523 return (0);
524 }
525 ncp = dvp->v_cache_dd;
526 if (ncp->nc_flag & NCF_ISDOTDOT)
527 *vpp = ncp->nc_vp;
528 else
529 *vpp = ncp->nc_dvp;
530 /* Return failure if negative entry was found. */
531 if (*vpp == NULL)
532 goto negative_success;
533 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
534 dvp, cnp->cn_nameptr, *vpp);
535 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..",
536 *vpp);
537 cache_out_ts(ncp, tsp, ticksp);
538 if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
539 NCF_DTS && tsp != NULL)
540 *tsp = ((struct namecache_ts *)ncp)->
541 nc_dotdottime;
542 goto success;
543 }
544 }
545
546 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
547 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
548 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
549 numchecks++;
550 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
551 !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen))
552 break;
553 }
554
555 /* We failed to find an entry */
556 if (ncp == NULL) {
557 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
558 NULL);
559 if ((cnp->cn_flags & MAKEENTRY) == 0) {
560 nummisszap++;
561 } else {
562 nummiss++;
563 }
564 nchstats.ncs_miss++;
565 goto unlock;
566 }
567
568 /* We don't want to have an entry, so dump it */
569 if ((cnp->cn_flags & MAKEENTRY) == 0) {
570 numposzaps++;
571 nchstats.ncs_badhits++;
572 if (!wlocked && !CACHE_UPGRADE_LOCK())
573 goto wlock;
574 cache_zap(ncp);
575 CACHE_WUNLOCK();
576 return (0);
577 }
578
579 /* We found a "positive" match, return the vnode */
580 if (ncp->nc_vp) {
581 numposhits++;
582 nchstats.ncs_goodhits++;
583 *vpp = ncp->nc_vp;
584 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
585 dvp, cnp->cn_nameptr, *vpp, ncp);
586 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp),
587 *vpp);
588 cache_out_ts(ncp, tsp, ticksp);
589 goto success;
590 }
591
592 negative_success:
593 /* We found a negative match, and want to create it, so purge */
594 if (cnp->cn_nameiop == CREATE) {
595 numnegzaps++;
596 nchstats.ncs_badhits++;
597 if (!wlocked && !CACHE_UPGRADE_LOCK())
598 goto wlock;
599 cache_zap(ncp);
600 CACHE_WUNLOCK();
601 return (0);
602 }
603
604 if (!wlocked && !CACHE_UPGRADE_LOCK())
605 goto wlock;
606 numneghits++;
607 /*
608 * We found a "negative" match, so we shift it to the end of
609 * the "negative" cache entries queue to satisfy LRU. Also,
610 * check to see if the entry is a whiteout; indicate this to
611 * the componentname, if so.
612 */
613 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
614 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
615 nchstats.ncs_neghits++;
616 if (ncp->nc_flag & NCF_WHITE)
617 cnp->cn_flags |= ISWHITEOUT;
618 SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp,
619 nc_get_name(ncp));
620 cache_out_ts(ncp, tsp, ticksp);
621 CACHE_WUNLOCK();
622 return (ENOENT);
623
624 wlock:
625 /*
626 * We need to update the cache after our lookup, so upgrade to
627 * a write lock and retry the operation.
628 */
629 CACHE_RUNLOCK();
630 CACHE_WLOCK();
631 numupgrades++;
632 wlocked = 1;
633 goto retry_wlocked;
634
635 success:
636 /*
637 * On success we return a locked and ref'd vnode as per the lookup
638 * protocol.
639 */
640 if (dvp == *vpp) { /* lookup on "." */
641 VREF(*vpp);
642 if (wlocked)
643 CACHE_WUNLOCK();
644 else
645 CACHE_RUNLOCK();
646 /*
647 * When we lookup "." we still can be asked to lock it
648 * differently...
649 */
650 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
651 if (ltype != VOP_ISLOCKED(*vpp)) {
652 if (ltype == LK_EXCLUSIVE) {
653 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
654 if ((*vpp)->v_iflag & VI_DOOMED) {
655 /* forced unmount */
656 vrele(*vpp);
657 *vpp = NULL;
658 return (ENOENT);
659 }
660 } else
661 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
662 }
663 return (-1);
664 }
665 ltype = 0; /* silence gcc warning */
666 if (cnp->cn_flags & ISDOTDOT) {
667 ltype = VOP_ISLOCKED(dvp);
668 VOP_UNLOCK(dvp, 0);
669 }
670 VI_LOCK(*vpp);
671 if (wlocked)
672 CACHE_WUNLOCK();
673 else
674 CACHE_RUNLOCK();
675 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
676 if (cnp->cn_flags & ISDOTDOT) {
677 vn_lock(dvp, ltype | LK_RETRY);
678 if (dvp->v_iflag & VI_DOOMED) {
679 if (error == 0)
680 vput(*vpp);
681 *vpp = NULL;
682 return (ENOENT);
683 }
684 }
685 if (error) {
686 *vpp = NULL;
687 goto retry;
688 }
689 if ((cnp->cn_flags & ISLASTCN) &&
690 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
691 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
692 }
693 return (-1);
694
695 unlock:
696 if (wlocked)
697 CACHE_WUNLOCK();
698 else
699 CACHE_RUNLOCK();
700 return (0);
701 }
702
703 /*
704 * Add an entry to the cache.
705 */
706 void
707 cache_enter_time(dvp, vp, cnp, tsp, dtsp)
708 struct vnode *dvp;
709 struct vnode *vp;
710 struct componentname *cnp;
711 struct timespec *tsp;
712 struct timespec *dtsp;
713 {
714 struct namecache *ncp, *n2;
715 struct namecache_ts *n3;
716 struct nchashhead *ncpp;
717 uint32_t hash;
718 int flag;
719 int hold;
720 int zap;
721 int len;
722
723 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
724 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
725 ("cache_enter: Adding a doomed vnode"));
726 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
727 ("cache_enter: Doomed vnode used as src"));
728
729 if (!doingcache)
730 return;
731
732 /*
733 * Avoid blowout in namecache entries.
734 */
735 if (numcache >= desiredvnodes * ncsizefactor)
736 return;
737
738 flag = 0;
739 if (cnp->cn_nameptr[0] == '.') {
740 if (cnp->cn_namelen == 1)
741 return;
742 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
743 CACHE_WLOCK();
744 /*
745 * If dotdot entry already exists, just retarget it
746 * to new parent vnode, otherwise continue with new
747 * namecache entry allocation.
748 */
749 if ((ncp = dvp->v_cache_dd) != NULL &&
750 ncp->nc_flag & NCF_ISDOTDOT) {
751 KASSERT(ncp->nc_dvp == dvp,
752 ("wrong isdotdot parent"));
753 if (ncp->nc_vp != NULL) {
754 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
755 ncp, nc_dst);
756 } else {
757 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
758 numneg--;
759 }
760 if (vp != NULL) {
761 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
762 ncp, nc_dst);
763 } else {
764 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
765 numneg++;
766 }
767 ncp->nc_vp = vp;
768 CACHE_WUNLOCK();
769 return;
770 }
771 dvp->v_cache_dd = NULL;
772 SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp);
773 CACHE_WUNLOCK();
774 flag = NCF_ISDOTDOT;
775 }
776 }
777
778 hold = 0;
779 zap = 0;
780
781 /*
782 * Calculate the hash key and setup as much of the new
783 * namecache entry as possible before acquiring the lock.
784 */
785 ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
786 ncp->nc_vp = vp;
787 ncp->nc_dvp = dvp;
788 ncp->nc_flag = flag;
789 if (tsp != NULL) {
790 n3 = (struct namecache_ts *)ncp;
791 n3->nc_time = *tsp;
792 n3->nc_ticks = ticks;
793 n3->nc_flag |= NCF_TS;
794 if (dtsp != NULL) {
795 n3->nc_dotdottime = *dtsp;
796 n3->nc_flag |= NCF_DTS;
797 }
798 }
799 len = ncp->nc_nlen = cnp->cn_namelen;
800 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
801 strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1);
802 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
803 CACHE_WLOCK();
804
805 /*
806 * See if this vnode or negative entry is already in the cache
807 * with this name. This can happen with concurrent lookups of
808 * the same path name.
809 */
810 ncpp = NCHHASH(hash);
811 LIST_FOREACH(n2, ncpp, nc_hash) {
812 if (n2->nc_dvp == dvp &&
813 n2->nc_nlen == cnp->cn_namelen &&
814 !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) {
815 if (tsp != NULL) {
816 KASSERT((n2->nc_flag & NCF_TS) != 0,
817 ("no NCF_TS"));
818 n3 = (struct namecache_ts *)n2;
819 n3->nc_time =
820 ((struct namecache_ts *)ncp)->nc_time;
821 n3->nc_ticks =
822 ((struct namecache_ts *)ncp)->nc_ticks;
823 if (dtsp != NULL) {
824 n3->nc_dotdottime =
825 ((struct namecache_ts *)ncp)->
826 nc_dotdottime;
827 n3->nc_flag |= NCF_DTS;
828 }
829 }
830 CACHE_WUNLOCK();
831 cache_free(ncp);
832 return;
833 }
834 }
835
836 if (flag == NCF_ISDOTDOT) {
837 /*
838 * See if we are trying to add .. entry, but some other lookup
839 * has populated v_cache_dd pointer already.
840 */
841 if (dvp->v_cache_dd != NULL) {
842 CACHE_WUNLOCK();
843 cache_free(ncp);
844 return;
845 }
846 KASSERT(vp == NULL || vp->v_type == VDIR,
847 ("wrong vnode type %p", vp));
848 dvp->v_cache_dd = ncp;
849 }
850
851 numcache++;
852 if (!vp) {
853 numneg++;
854 if (cnp->cn_flags & ISWHITEOUT)
855 ncp->nc_flag |= NCF_WHITE;
856 } else if (vp->v_type == VDIR) {
857 if (flag != NCF_ISDOTDOT) {
858 /*
859 * For this case, the cache entry maps both the
860 * directory name in it and the name ".." for the
861 * directory's parent.
862 */
863 if ((n2 = vp->v_cache_dd) != NULL &&
864 (n2->nc_flag & NCF_ISDOTDOT) != 0)
865 cache_zap(n2);
866 vp->v_cache_dd = ncp;
867 }
868 } else {
869 vp->v_cache_dd = NULL;
870 }
871
872 /*
873 * Insert the new namecache entry into the appropriate chain
874 * within the cache entries table.
875 */
876 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
877 if (flag != NCF_ISDOTDOT) {
878 if (LIST_EMPTY(&dvp->v_cache_src)) {
879 hold = 1;
880 numcachehv++;
881 }
882 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
883 }
884
885 /*
886 * If the entry is "negative", we place it into the
887 * "negative" cache queue, otherwise, we place it into the
888 * destination vnode's cache entries queue.
889 */
890 if (vp) {
891 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
892 SDT_PROBE3(vfs, namecache, enter, done, dvp, nc_get_name(ncp),
893 vp);
894 } else {
895 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
896 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp,
897 nc_get_name(ncp));
898 }
899 if (numneg * ncnegfactor > numcache) {
900 ncp = TAILQ_FIRST(&ncneg);
901 KASSERT(ncp->nc_vp == NULL, ("ncp %p vp %p on ncneg",
902 ncp, ncp->nc_vp));
903 zap = 1;
904 }
905 if (hold)
906 vhold(dvp);
907 if (zap)
908 cache_zap(ncp);
909 CACHE_WUNLOCK();
910 }
911
912 /*
913 * Name cache initialization, from vfs_init() when we are booting
914 */
915 static void
916 nchinit(void *dummy __unused)
917 {
918
919 TAILQ_INIT(&ncneg);
920
921 cache_zone_small = uma_zcreate("S VFS Cache",
922 sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
923 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
924 cache_zone_small_ts = uma_zcreate("STS VFS Cache",
925 sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
926 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
927 cache_zone_large = uma_zcreate("L VFS Cache",
928 sizeof(struct namecache) + NAME_MAX + 1,
929 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
930 cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
931 sizeof(struct namecache_ts) + NAME_MAX + 1,
932 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
933
934 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
935 }
936 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
937
938 void
939 cache_changesize(int newmaxvnodes)
940 {
941 struct nchashhead *new_nchashtbl, *old_nchashtbl;
942 u_long new_nchash, old_nchash;
943 struct namecache *ncp;
944 uint32_t hash;
945 int i;
946
947 new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash);
948 /* If same hash table size, nothing to do */
949 if (nchash == new_nchash) {
950 free(new_nchashtbl, M_VFSCACHE);
951 return;
952 }
953 /*
954 * Move everything from the old hash table to the new table.
955 * None of the namecache entries in the table can be removed
956 * because to do so, they have to be removed from the hash table.
957 */
958 CACHE_WLOCK();
959 old_nchashtbl = nchashtbl;
960 old_nchash = nchash;
961 nchashtbl = new_nchashtbl;
962 nchash = new_nchash;
963 for (i = 0; i <= old_nchash; i++) {
964 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) {
965 hash = fnv_32_buf(nc_get_name(ncp), ncp->nc_nlen,
966 FNV1_32_INIT);
967 hash = fnv_32_buf(&ncp->nc_dvp, sizeof(ncp->nc_dvp),
968 hash);
969 LIST_REMOVE(ncp, nc_hash);
970 LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash);
971 }
972 }
973 CACHE_WUNLOCK();
974 free(old_nchashtbl, M_VFSCACHE);
975 }
976
977 /*
978 * Invalidate all entries to a particular vnode.
979 */
980 void
981 cache_purge(vp)
982 struct vnode *vp;
983 {
984
985 CTR1(KTR_VFS, "cache_purge(%p)", vp);
986 SDT_PROBE1(vfs, namecache, purge, done, vp);
987 CACHE_WLOCK();
988 while (!LIST_EMPTY(&vp->v_cache_src))
989 cache_zap(LIST_FIRST(&vp->v_cache_src));
990 while (!TAILQ_EMPTY(&vp->v_cache_dst))
991 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
992 if (vp->v_cache_dd != NULL) {
993 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
994 ("lost dotdot link"));
995 cache_zap(vp->v_cache_dd);
996 }
997 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
998 CACHE_WUNLOCK();
999 }
1000
1001 /*
1002 * Invalidate all negative entries for a particular directory vnode.
1003 */
1004 void
1005 cache_purge_negative(vp)
1006 struct vnode *vp;
1007 {
1008 struct namecache *cp, *ncp;
1009
1010 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
1011 SDT_PROBE1(vfs, namecache, purge_negative, done, vp);
1012 CACHE_WLOCK();
1013 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
1014 if (cp->nc_vp == NULL)
1015 cache_zap(cp);
1016 }
1017 CACHE_WUNLOCK();
1018 }
1019
1020 /*
1021 * Flush all entries referencing a particular filesystem.
1022 */
1023 void
1024 cache_purgevfs(mp)
1025 struct mount *mp;
1026 {
1027 struct nchashhead *ncpp;
1028 struct namecache *ncp, *nnp;
1029
1030 /* Scan hash tables for applicable entries */
1031 SDT_PROBE1(vfs, namecache, purgevfs, done, mp);
1032 CACHE_WLOCK();
1033 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
1034 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
1035 if (ncp->nc_dvp->v_mount == mp)
1036 cache_zap(ncp);
1037 }
1038 }
1039 CACHE_WUNLOCK();
1040 }
1041
1042 /*
1043 * Perform canonical checks and cache lookup and pass on to filesystem
1044 * through the vop_cachedlookup only if needed.
1045 */
1046
1047 int
1048 vfs_cache_lookup(ap)
1049 struct vop_lookup_args /* {
1050 struct vnode *a_dvp;
1051 struct vnode **a_vpp;
1052 struct componentname *a_cnp;
1053 } */ *ap;
1054 {
1055 struct vnode *dvp;
1056 int error;
1057 struct vnode **vpp = ap->a_vpp;
1058 struct componentname *cnp = ap->a_cnp;
1059 struct ucred *cred = cnp->cn_cred;
1060 int flags = cnp->cn_flags;
1061 struct thread *td = cnp->cn_thread;
1062
1063 *vpp = NULL;
1064 dvp = ap->a_dvp;
1065
1066 if (dvp->v_type != VDIR)
1067 return (ENOTDIR);
1068
1069 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1070 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1071 return (EROFS);
1072
1073 error = VOP_ACCESS(dvp, VEXEC, cred, td);
1074 if (error)
1075 return (error);
1076
1077 error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
1078 if (error == 0)
1079 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
1080 if (error == -1)
1081 return (0);
1082 return (error);
1083 }
1084
1085 /*
1086 * XXX All of these sysctls would probably be more productive dead.
1087 */
1088 static int disablecwd;
1089 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
1090 "Disable the getcwd syscall");
1091
1092 /* Implementation of the getcwd syscall. */
1093 int
1094 sys___getcwd(td, uap)
1095 struct thread *td;
1096 struct __getcwd_args *uap;
1097 {
1098
1099 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
1100 }
1101
1102 int
1103 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen)
1104 {
1105 char *bp, *tmpbuf;
1106 struct filedesc *fdp;
1107 struct vnode *cdir, *rdir;
1108 int error;
1109
1110 if (disablecwd)
1111 return (ENODEV);
1112 if (buflen < 2)
1113 return (EINVAL);
1114 if (buflen > MAXPATHLEN)
1115 buflen = MAXPATHLEN;
1116
1117 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
1118 fdp = td->td_proc->p_fd;
1119 FILEDESC_SLOCK(fdp);
1120 cdir = fdp->fd_cdir;
1121 VREF(cdir);
1122 rdir = fdp->fd_rdir;
1123 VREF(rdir);
1124 FILEDESC_SUNLOCK(fdp);
1125 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
1126 vrele(rdir);
1127 vrele(cdir);
1128
1129 if (!error) {
1130 if (bufseg == UIO_SYSSPACE)
1131 bcopy(bp, buf, strlen(bp) + 1);
1132 else
1133 error = copyout(bp, buf, strlen(bp) + 1);
1134 #ifdef KTRACE
1135 if (KTRPOINT(curthread, KTR_NAMEI))
1136 ktrnamei(bp);
1137 #endif
1138 }
1139 free(tmpbuf, M_TEMP);
1140 return (error);
1141 }
1142
1143 /*
1144 * Thus begins the fullpath magic.
1145 */
1146
1147 #undef STATNODE
1148 #define STATNODE(name, descr) \
1149 static u_int name; \
1150 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr)
1151
1152 static int disablefullpath;
1153 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
1154 "Disable the vn_fullpath function");
1155
1156 /* These count for kern___getcwd(), too. */
1157 STATNODE(numfullpathcalls, "Number of fullpath search calls");
1158 STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
1159 STATNODE(numfullpathfail2,
1160 "Number of fullpath search errors (VOP_VPTOCNP failures)");
1161 STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
1162 STATNODE(numfullpathfound, "Number of successful fullpath calls");
1163
1164 /*
1165 * Retrieve the full filesystem path that correspond to a vnode from the name
1166 * cache (if available)
1167 */
1168 int
1169 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
1170 {
1171 char *buf;
1172 struct filedesc *fdp;
1173 struct vnode *rdir;
1174 int error;
1175
1176 if (disablefullpath)
1177 return (ENODEV);
1178 if (vn == NULL)
1179 return (EINVAL);
1180
1181 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1182 fdp = td->td_proc->p_fd;
1183 FILEDESC_SLOCK(fdp);
1184 rdir = fdp->fd_rdir;
1185 VREF(rdir);
1186 FILEDESC_SUNLOCK(fdp);
1187 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1188 vrele(rdir);
1189
1190 if (!error)
1191 *freebuf = buf;
1192 else
1193 free(buf, M_TEMP);
1194 return (error);
1195 }
1196
1197 /*
1198 * This function is similar to vn_fullpath, but it attempts to lookup the
1199 * pathname relative to the global root mount point. This is required for the
1200 * auditing sub-system, as audited pathnames must be absolute, relative to the
1201 * global root mount point.
1202 */
1203 int
1204 vn_fullpath_global(struct thread *td, struct vnode *vn,
1205 char **retbuf, char **freebuf)
1206 {
1207 char *buf;
1208 int error;
1209
1210 if (disablefullpath)
1211 return (ENODEV);
1212 if (vn == NULL)
1213 return (EINVAL);
1214 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1215 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1216 if (!error)
1217 *freebuf = buf;
1218 else
1219 free(buf, M_TEMP);
1220 return (error);
1221 }
1222
1223 int
1224 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1225 {
1226 int error;
1227
1228 CACHE_RLOCK();
1229 error = vn_vptocnp_locked(vp, cred, buf, buflen);
1230 if (error == 0)
1231 CACHE_RUNLOCK();
1232 return (error);
1233 }
1234
1235 static int
1236 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1237 u_int *buflen)
1238 {
1239 struct vnode *dvp;
1240 struct namecache *ncp;
1241 int error;
1242
1243 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1244 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1245 break;
1246 }
1247 if (ncp != NULL) {
1248 if (*buflen < ncp->nc_nlen) {
1249 CACHE_RUNLOCK();
1250 vrele(*vp);
1251 numfullpathfail4++;
1252 error = ENOMEM;
1253 SDT_PROBE3(vfs, namecache, fullpath, return, error,
1254 vp, NULL);
1255 return (error);
1256 }
1257 *buflen -= ncp->nc_nlen;
1258 memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen);
1259 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1260 nc_get_name(ncp), vp);
1261 dvp = *vp;
1262 *vp = ncp->nc_dvp;
1263 vref(*vp);
1264 CACHE_RUNLOCK();
1265 vrele(dvp);
1266 CACHE_RLOCK();
1267 return (0);
1268 }
1269 SDT_PROBE1(vfs, namecache, fullpath, miss, vp);
1270
1271 CACHE_RUNLOCK();
1272 vn_lock(*vp, LK_SHARED | LK_RETRY);
1273 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1274 vput(*vp);
1275 if (error) {
1276 numfullpathfail2++;
1277 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
1278 return (error);
1279 }
1280
1281 *vp = dvp;
1282 CACHE_RLOCK();
1283 if (dvp->v_iflag & VI_DOOMED) {
1284 /* forced unmount */
1285 CACHE_RUNLOCK();
1286 vrele(dvp);
1287 error = ENOENT;
1288 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
1289 return (error);
1290 }
1291 /*
1292 * *vp has its use count incremented still.
1293 */
1294
1295 return (0);
1296 }
1297
1298 /*
1299 * The magic behind kern___getcwd() and vn_fullpath().
1300 */
1301 static int
1302 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1303 char *buf, char **retbuf, u_int buflen)
1304 {
1305 int error, slash_prefixed;
1306 #ifdef KDTRACE_HOOKS
1307 struct vnode *startvp = vp;
1308 #endif
1309 struct vnode *vp1;
1310
1311 buflen--;
1312 buf[buflen] = '\0';
1313 error = 0;
1314 slash_prefixed = 0;
1315
1316 SDT_PROBE1(vfs, namecache, fullpath, entry, vp);
1317 numfullpathcalls++;
1318 vref(vp);
1319 CACHE_RLOCK();
1320 if (vp->v_type != VDIR) {
1321 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1322 if (error)
1323 return (error);
1324 if (buflen == 0) {
1325 CACHE_RUNLOCK();
1326 vrele(vp);
1327 return (ENOMEM);
1328 }
1329 buf[--buflen] = '/';
1330 slash_prefixed = 1;
1331 }
1332 while (vp != rdir && vp != rootvnode) {
1333 if (vp->v_vflag & VV_ROOT) {
1334 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */
1335 CACHE_RUNLOCK();
1336 vrele(vp);
1337 error = ENOENT;
1338 SDT_PROBE3(vfs, namecache, fullpath, return,
1339 error, vp, NULL);
1340 break;
1341 }
1342 vp1 = vp->v_mount->mnt_vnodecovered;
1343 vref(vp1);
1344 CACHE_RUNLOCK();
1345 vrele(vp);
1346 vp = vp1;
1347 CACHE_RLOCK();
1348 continue;
1349 }
1350 if (vp->v_type != VDIR) {
1351 CACHE_RUNLOCK();
1352 vrele(vp);
1353 numfullpathfail1++;
1354 error = ENOTDIR;
1355 SDT_PROBE3(vfs, namecache, fullpath, return,
1356 error, vp, NULL);
1357 break;
1358 }
1359 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1360 if (error)
1361 break;
1362 if (buflen == 0) {
1363 CACHE_RUNLOCK();
1364 vrele(vp);
1365 error = ENOMEM;
1366 SDT_PROBE3(vfs, namecache, fullpath, return, error,
1367 startvp, NULL);
1368 break;
1369 }
1370 buf[--buflen] = '/';
1371 slash_prefixed = 1;
1372 }
1373 if (error)
1374 return (error);
1375 if (!slash_prefixed) {
1376 if (buflen == 0) {
1377 CACHE_RUNLOCK();
1378 vrele(vp);
1379 numfullpathfail4++;
1380 SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM,
1381 startvp, NULL);
1382 return (ENOMEM);
1383 }
1384 buf[--buflen] = '/';
1385 }
1386 numfullpathfound++;
1387 CACHE_RUNLOCK();
1388 vrele(vp);
1389
1390 SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen);
1391 *retbuf = buf + buflen;
1392 return (0);
1393 }
1394
1395 struct vnode *
1396 vn_dir_dd_ino(struct vnode *vp)
1397 {
1398 struct namecache *ncp;
1399 struct vnode *ddvp;
1400
1401 ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
1402 CACHE_RLOCK();
1403 TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
1404 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
1405 continue;
1406 ddvp = ncp->nc_dvp;
1407 VI_LOCK(ddvp);
1408 CACHE_RUNLOCK();
1409 if (vget(ddvp, LK_INTERLOCK | LK_SHARED | LK_NOWAIT, curthread))
1410 return (NULL);
1411 return (ddvp);
1412 }
1413 CACHE_RUNLOCK();
1414 return (NULL);
1415 }
1416
1417 int
1418 vn_commname(struct vnode *vp, char *buf, u_int buflen)
1419 {
1420 struct namecache *ncp;
1421 int l;
1422
1423 CACHE_RLOCK();
1424 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1425 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1426 break;
1427 if (ncp == NULL) {
1428 CACHE_RUNLOCK();
1429 return (ENOENT);
1430 }
1431 l = min(ncp->nc_nlen, buflen - 1);
1432 memcpy(buf, nc_get_name(ncp), l);
1433 CACHE_RUNLOCK();
1434 buf[l] = '\0';
1435 return (0);
1436 }
1437
1438 /* ABI compat shims for old kernel modules. */
1439 #undef cache_enter
1440
1441 void cache_enter(struct vnode *dvp, struct vnode *vp,
1442 struct componentname *cnp);
1443
1444 void
1445 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
1446 {
1447
1448 cache_enter_time(dvp, vp, cnp, NULL, NULL);
1449 }
1450
1451 /*
1452 * This function updates path string to vnode's full global path
1453 * and checks the size of the new path string against the pathlen argument.
1454 *
1455 * Requires a locked, referenced vnode and GIANT lock held.
1456 * Vnode is re-locked on success or ENODEV, otherwise unlocked.
1457 *
1458 * If sysctl debug.disablefullpath is set, ENODEV is returned,
1459 * vnode is left locked and path remain untouched.
1460 *
1461 * If vp is a directory, the call to vn_fullpath_global() always succeeds
1462 * because it falls back to the ".." lookup if the namecache lookup fails.
1463 */
1464 int
1465 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
1466 u_int pathlen)
1467 {
1468 struct nameidata nd;
1469 struct vnode *vp1;
1470 char *rpath, *fbuf;
1471 int error;
1472
1473 ASSERT_VOP_ELOCKED(vp, __func__);
1474
1475 /* Return ENODEV if sysctl debug.disablefullpath==1 */
1476 if (disablefullpath)
1477 return (ENODEV);
1478
1479 /* Construct global filesystem path from vp. */
1480 VOP_UNLOCK(vp, 0);
1481 error = vn_fullpath_global(td, vp, &rpath, &fbuf);
1482
1483 if (error != 0) {
1484 vrele(vp);
1485 return (error);
1486 }
1487
1488 if (strlen(rpath) >= pathlen) {
1489 vrele(vp);
1490 error = ENAMETOOLONG;
1491 goto out;
1492 }
1493
1494 /*
1495 * Re-lookup the vnode by path to detect a possible rename.
1496 * As a side effect, the vnode is relocked.
1497 * If vnode was renamed, return ENOENT.
1498 */
1499 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1500 UIO_SYSSPACE, path, td);
1501 error = namei(&nd);
1502 if (error != 0) {
1503 vrele(vp);
1504 goto out;
1505 }
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
1516 out:
1517 free(fbuf, M_TEMP);
1518 return (error);
1519 }
Cache object: 6994d809520096802064b02782c0777c
|