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>
| 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: head/sys/kern/vfs_cache.c 206894 2010-04-20 10:19:27Z kib $");
| 36__FBSDID("$FreeBSD: head/sys/kern/vfs_cache.c 209390 2010-06-21 09:55:56Z ed $");
|
37
38#include "opt_kdtrace.h"
39#include "opt_ktrace.h"
40
41#include <sys/param.h>
42#include <sys/filedesc.h>
43#include <sys/fnv_hash.h>
44#include <sys/kernel.h>
45#include <sys/lock.h>
46#include <sys/malloc.h>
47#include <sys/mount.h>
48#include <sys/namei.h>
49#include <sys/proc.h>
50#include <sys/rwlock.h>
51#include <sys/sdt.h>
52#include <sys/syscallsubr.h>
53#include <sys/sysctl.h>
54#include <sys/sysproto.h>
55#include <sys/systm.h>
56#include <sys/vnode.h>
57#ifdef KTRACE
58#include <sys/ktrace.h>
59#endif
60
61#include <vm/uma.h>
62
63SDT_PROVIDER_DECLARE(vfs);
64SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
65 "struct vnode *");
66SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
67 "char *");
68SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *");
69SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *",
70 "struct char *", "struct vnode *");
71SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *");
72SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", "struct vnode *",
73 "struct char *");
74SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *",
75 "struct vnode *");
76SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit_negative, "struct vnode *",
77 "char *");
78SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
79 "char *");
80SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *");
81SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *");
82SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *");
83SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *",
84 "struct vnode *");
85SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *",
86 "char *");
87
88/*
89 * This structure describes the elements in the cache of recent
90 * names looked up by namei.
91 */
92
93struct namecache {
94 LIST_ENTRY(namecache) nc_hash; /* hash chain */
95 LIST_ENTRY(namecache) nc_src; /* source vnode list */
96 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
97 struct vnode *nc_dvp; /* vnode of parent of name */
98 struct vnode *nc_vp; /* vnode the name refers to */
99 u_char nc_flag; /* flag bits */
100 u_char nc_nlen; /* length of name */
101 char nc_name[0]; /* segment name + nul */
102};
103
104/*
105 * Name caching works as follows:
106 *
107 * Names found by directory scans are retained in a cache
108 * for future reference. It is managed LRU, so frequently
109 * used names will hang around. Cache is indexed by hash value
110 * obtained from (vp, name) where vp refers to the directory
111 * containing name.
112 *
113 * If it is a "negative" entry, (i.e. for a name that is known NOT to
114 * exist) the vnode pointer will be NULL.
115 *
116 * Upon reaching the last segment of a path, if the reference
117 * is for DELETE, or NOCACHE is set (rewrite), and the
118 * name is located in the cache, it will be dropped.
119 */
120
121/*
122 * Structures associated with name cacheing.
123 */
124#define NCHHASH(hash) \
125 (&nchashtbl[(hash) & nchash])
126static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
127static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
128static u_long nchash; /* size of hash table */
129SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
130static u_long ncnegfactor = 16; /* ratio of negative entries */
131SYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
132static u_long numneg; /* number of cache entries allocated */
133SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
134static u_long numcache; /* number of cache entries allocated */
135SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
136static u_long numcachehv; /* number of cache entries with vnodes held */
137SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, "");
138#if 0
139static u_long numcachepl; /* number of cache purge for leaf entries */
140SYSCTL_ULONG(_debug, OID_AUTO, numcachepl, CTLFLAG_RD, &numcachepl, 0, "");
141#endif
142struct nchstats nchstats; /* cache effectiveness statistics */
143
144static struct rwlock cache_lock;
145RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
146
147#define CACHE_UPGRADE_LOCK() rw_try_upgrade(&cache_lock)
148#define CACHE_RLOCK() rw_rlock(&cache_lock)
149#define CACHE_RUNLOCK() rw_runlock(&cache_lock)
150#define CACHE_WLOCK() rw_wlock(&cache_lock)
151#define CACHE_WUNLOCK() rw_wunlock(&cache_lock)
152
153/*
154 * UMA zones for the VFS cache.
155 *
156 * The small cache is used for entries with short names, which are the
157 * most common. The large cache is used for entries which are too big to
158 * fit in the small cache.
159 */
160static uma_zone_t cache_zone_small;
161static uma_zone_t cache_zone_large;
162
163#define CACHE_PATH_CUTOFF 35
164#define CACHE_ZONE_SMALL (sizeof(struct namecache) + CACHE_PATH_CUTOFF \
165 + 1)
166#define CACHE_ZONE_LARGE (sizeof(struct namecache) + NAME_MAX + 1)
167
168#define cache_alloc(len) uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \
169 cache_zone_small : cache_zone_large, M_WAITOK)
170#define cache_free(ncp) do { \
171 if (ncp != NULL) \
172 uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \
173 cache_zone_small : cache_zone_large, (ncp)); \
174} while (0)
175
176static int doingcache = 1; /* 1 => enable the cache */
177SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
178
179/* Export size information to userland */
180SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0,
181 sizeof(struct namecache), "");
182
183/*
184 * The new name cache statistics
185 */
186static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
187#define STATNODE(mode, name, var) \
188 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
189STATNODE(CTLFLAG_RD, numneg, &numneg);
190STATNODE(CTLFLAG_RD, numcache, &numcache);
191static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
192static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
193static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
194static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
195static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
196static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
197static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
198static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
199static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
200static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
201static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades);
202
203SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
204 &nchstats, sizeof(nchstats), "LU", "VFS cache effectiveness statistics");
205
206
207
208static void cache_zap(struct namecache *ncp);
209static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
210 u_int *buflen);
211static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
212 char *buf, char **retbuf, u_int buflen);
213
214static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
215
216/*
217 * Flags in namecache.nc_flag
218 */
219#define NCF_WHITE 0x01
220#define NCF_ISDOTDOT 0x02
221
222#ifdef DIAGNOSTIC
223/*
224 * Grab an atomic snapshot of the name cache hash chain lengths
225 */
226SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
227
228static int
229sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
230{
231 int error;
232 struct nchashhead *ncpp;
233 struct namecache *ncp;
234 int n_nchash;
235 int count;
236
237 n_nchash = nchash + 1; /* nchash is max index, not count */
238 if (!req->oldptr)
239 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
240
241 /* Scan hash tables for applicable entries */
242 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
243 CACHE_RLOCK();
244 count = 0;
245 LIST_FOREACH(ncp, ncpp, nc_hash) {
246 count++;
247 }
248 CACHE_RUNLOCK();
249 error = SYSCTL_OUT(req, &count, sizeof(count));
250 if (error)
251 return (error);
252 }
253 return (0);
254}
255SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
256 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
257 "nchash chain lengths");
258
259static int
260sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
261{
262 int error;
263 struct nchashhead *ncpp;
264 struct namecache *ncp;
265 int n_nchash;
266 int count, maxlength, used, pct;
267
268 if (!req->oldptr)
269 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
270
271 n_nchash = nchash + 1; /* nchash is max index, not count */
272 used = 0;
273 maxlength = 0;
274
275 /* Scan hash tables for applicable entries */
276 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
277 count = 0;
278 CACHE_RLOCK();
279 LIST_FOREACH(ncp, ncpp, nc_hash) {
280 count++;
281 }
282 CACHE_RUNLOCK();
283 if (count)
284 used++;
285 if (maxlength < count)
286 maxlength = count;
287 }
288 n_nchash = nchash + 1;
289 pct = (used * 100 * 100) / n_nchash;
290 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
291 if (error)
292 return (error);
293 error = SYSCTL_OUT(req, &used, sizeof(used));
294 if (error)
295 return (error);
296 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
297 if (error)
298 return (error);
299 error = SYSCTL_OUT(req, &pct, sizeof(pct));
300 if (error)
301 return (error);
302 return (0);
303}
304SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
305 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
306 "nchash chain lengths");
307#endif
308
309/*
310 * cache_zap():
311 *
312 * Removes a namecache entry from cache, whether it contains an actual
313 * pointer to a vnode or if it is just a negative cache entry.
314 */
315static void
316cache_zap(ncp)
317 struct namecache *ncp;
318{
319 struct vnode *vp;
320
321 rw_assert(&cache_lock, RA_WLOCKED);
322 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
323#ifdef KDTRACE_HOOKS
324 if (ncp->nc_vp != NULL) {
325 SDT_PROBE(vfs, namecache, zap, done, ncp->nc_dvp,
326 ncp->nc_name, ncp->nc_vp, 0, 0);
327 } else {
328 SDT_PROBE(vfs, namecache, zap_negative, done, ncp->nc_dvp,
329 ncp->nc_name, 0, 0, 0);
330 }
331#endif
332 vp = NULL;
333 LIST_REMOVE(ncp, nc_hash);
334 if (ncp->nc_flag & NCF_ISDOTDOT) {
335 if (ncp == ncp->nc_dvp->v_cache_dd)
336 ncp->nc_dvp->v_cache_dd = NULL;
337 } else {
338 LIST_REMOVE(ncp, nc_src);
339 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
340 vp = ncp->nc_dvp;
341 numcachehv--;
342 }
343 }
344 if (ncp->nc_vp) {
345 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
346 if (ncp == ncp->nc_vp->v_cache_dd)
347 ncp->nc_vp->v_cache_dd = NULL;
348 } else {
349 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
350 numneg--;
351 }
352 numcache--;
353 cache_free(ncp);
354 if (vp)
355 vdrop(vp);
356}
357
358/*
359 * Lookup an entry in the cache
360 *
361 * Lookup is called with dvp pointing to the directory to search,
362 * cnp pointing to the name of the entry being sought. If the lookup
363 * succeeds, the vnode is returned in *vpp, and a status of -1 is
364 * returned. If the lookup determines that the name does not exist
365 * (negative cacheing), a status of ENOENT is returned. If the lookup
366 * fails, a status of zero is returned. If the directory vnode is
367 * recycled out from under us due to a forced unmount, a status of
368 * ENOENT is returned.
369 *
370 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
371 * unlocked. If we're looking up . an extra ref is taken, but the lock is
372 * not recursively acquired.
373 */
374
375int
376cache_lookup(dvp, vpp, cnp)
377 struct vnode *dvp;
378 struct vnode **vpp;
379 struct componentname *cnp;
380{
381 struct namecache *ncp;
| 37
38#include "opt_kdtrace.h"
39#include "opt_ktrace.h"
40
41#include <sys/param.h>
42#include <sys/filedesc.h>
43#include <sys/fnv_hash.h>
44#include <sys/kernel.h>
45#include <sys/lock.h>
46#include <sys/malloc.h>
47#include <sys/mount.h>
48#include <sys/namei.h>
49#include <sys/proc.h>
50#include <sys/rwlock.h>
51#include <sys/sdt.h>
52#include <sys/syscallsubr.h>
53#include <sys/sysctl.h>
54#include <sys/sysproto.h>
55#include <sys/systm.h>
56#include <sys/vnode.h>
57#ifdef KTRACE
58#include <sys/ktrace.h>
59#endif
60
61#include <vm/uma.h>
62
63SDT_PROVIDER_DECLARE(vfs);
64SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
65 "struct vnode *");
66SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
67 "char *");
68SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *");
69SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *",
70 "struct char *", "struct vnode *");
71SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *");
72SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", "struct vnode *",
73 "struct char *");
74SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *",
75 "struct vnode *");
76SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit_negative, "struct vnode *",
77 "char *");
78SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
79 "char *");
80SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *");
81SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *");
82SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *");
83SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *",
84 "struct vnode *");
85SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *",
86 "char *");
87
88/*
89 * This structure describes the elements in the cache of recent
90 * names looked up by namei.
91 */
92
93struct namecache {
94 LIST_ENTRY(namecache) nc_hash; /* hash chain */
95 LIST_ENTRY(namecache) nc_src; /* source vnode list */
96 TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */
97 struct vnode *nc_dvp; /* vnode of parent of name */
98 struct vnode *nc_vp; /* vnode the name refers to */
99 u_char nc_flag; /* flag bits */
100 u_char nc_nlen; /* length of name */
101 char nc_name[0]; /* segment name + nul */
102};
103
104/*
105 * Name caching works as follows:
106 *
107 * Names found by directory scans are retained in a cache
108 * for future reference. It is managed LRU, so frequently
109 * used names will hang around. Cache is indexed by hash value
110 * obtained from (vp, name) where vp refers to the directory
111 * containing name.
112 *
113 * If it is a "negative" entry, (i.e. for a name that is known NOT to
114 * exist) the vnode pointer will be NULL.
115 *
116 * Upon reaching the last segment of a path, if the reference
117 * is for DELETE, or NOCACHE is set (rewrite), and the
118 * name is located in the cache, it will be dropped.
119 */
120
121/*
122 * Structures associated with name cacheing.
123 */
124#define NCHHASH(hash) \
125 (&nchashtbl[(hash) & nchash])
126static LIST_HEAD(nchashhead, namecache) *nchashtbl; /* Hash Table */
127static TAILQ_HEAD(, namecache) ncneg; /* Hash Table */
128static u_long nchash; /* size of hash table */
129SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "");
130static u_long ncnegfactor = 16; /* ratio of negative entries */
131SYSCTL_ULONG(_debug, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "");
132static u_long numneg; /* number of cache entries allocated */
133SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "");
134static u_long numcache; /* number of cache entries allocated */
135SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "");
136static u_long numcachehv; /* number of cache entries with vnodes held */
137SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, "");
138#if 0
139static u_long numcachepl; /* number of cache purge for leaf entries */
140SYSCTL_ULONG(_debug, OID_AUTO, numcachepl, CTLFLAG_RD, &numcachepl, 0, "");
141#endif
142struct nchstats nchstats; /* cache effectiveness statistics */
143
144static struct rwlock cache_lock;
145RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
146
147#define CACHE_UPGRADE_LOCK() rw_try_upgrade(&cache_lock)
148#define CACHE_RLOCK() rw_rlock(&cache_lock)
149#define CACHE_RUNLOCK() rw_runlock(&cache_lock)
150#define CACHE_WLOCK() rw_wlock(&cache_lock)
151#define CACHE_WUNLOCK() rw_wunlock(&cache_lock)
152
153/*
154 * UMA zones for the VFS cache.
155 *
156 * The small cache is used for entries with short names, which are the
157 * most common. The large cache is used for entries which are too big to
158 * fit in the small cache.
159 */
160static uma_zone_t cache_zone_small;
161static uma_zone_t cache_zone_large;
162
163#define CACHE_PATH_CUTOFF 35
164#define CACHE_ZONE_SMALL (sizeof(struct namecache) + CACHE_PATH_CUTOFF \
165 + 1)
166#define CACHE_ZONE_LARGE (sizeof(struct namecache) + NAME_MAX + 1)
167
168#define cache_alloc(len) uma_zalloc(((len) <= CACHE_PATH_CUTOFF) ? \
169 cache_zone_small : cache_zone_large, M_WAITOK)
170#define cache_free(ncp) do { \
171 if (ncp != NULL) \
172 uma_zfree(((ncp)->nc_nlen <= CACHE_PATH_CUTOFF) ? \
173 cache_zone_small : cache_zone_large, (ncp)); \
174} while (0)
175
176static int doingcache = 1; /* 1 => enable the cache */
177SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "");
178
179/* Export size information to userland */
180SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, 0,
181 sizeof(struct namecache), "");
182
183/*
184 * The new name cache statistics
185 */
186static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics");
187#define STATNODE(mode, name, var) \
188 SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, "");
189STATNODE(CTLFLAG_RD, numneg, &numneg);
190STATNODE(CTLFLAG_RD, numcache, &numcache);
191static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls);
192static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits);
193static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits);
194static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks);
195static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss);
196static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap);
197static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps);
198static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits);
199static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps);
200static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits);
201static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades);
202
203SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
204 &nchstats, sizeof(nchstats), "LU", "VFS cache effectiveness statistics");
205
206
207
208static void cache_zap(struct namecache *ncp);
209static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
210 u_int *buflen);
211static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
212 char *buf, char **retbuf, u_int buflen);
213
214static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
215
216/*
217 * Flags in namecache.nc_flag
218 */
219#define NCF_WHITE 0x01
220#define NCF_ISDOTDOT 0x02
221
222#ifdef DIAGNOSTIC
223/*
224 * Grab an atomic snapshot of the name cache hash chain lengths
225 */
226SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats");
227
228static int
229sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
230{
231 int error;
232 struct nchashhead *ncpp;
233 struct namecache *ncp;
234 int n_nchash;
235 int count;
236
237 n_nchash = nchash + 1; /* nchash is max index, not count */
238 if (!req->oldptr)
239 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
240
241 /* Scan hash tables for applicable entries */
242 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
243 CACHE_RLOCK();
244 count = 0;
245 LIST_FOREACH(ncp, ncpp, nc_hash) {
246 count++;
247 }
248 CACHE_RUNLOCK();
249 error = SYSCTL_OUT(req, &count, sizeof(count));
250 if (error)
251 return (error);
252 }
253 return (0);
254}
255SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
256 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
257 "nchash chain lengths");
258
259static int
260sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
261{
262 int error;
263 struct nchashhead *ncpp;
264 struct namecache *ncp;
265 int n_nchash;
266 int count, maxlength, used, pct;
267
268 if (!req->oldptr)
269 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
270
271 n_nchash = nchash + 1; /* nchash is max index, not count */
272 used = 0;
273 maxlength = 0;
274
275 /* Scan hash tables for applicable entries */
276 for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
277 count = 0;
278 CACHE_RLOCK();
279 LIST_FOREACH(ncp, ncpp, nc_hash) {
280 count++;
281 }
282 CACHE_RUNLOCK();
283 if (count)
284 used++;
285 if (maxlength < count)
286 maxlength = count;
287 }
288 n_nchash = nchash + 1;
289 pct = (used * 100 * 100) / n_nchash;
290 error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
291 if (error)
292 return (error);
293 error = SYSCTL_OUT(req, &used, sizeof(used));
294 if (error)
295 return (error);
296 error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
297 if (error)
298 return (error);
299 error = SYSCTL_OUT(req, &pct, sizeof(pct));
300 if (error)
301 return (error);
302 return (0);
303}
304SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
305 CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
306 "nchash chain lengths");
307#endif
308
309/*
310 * cache_zap():
311 *
312 * Removes a namecache entry from cache, whether it contains an actual
313 * pointer to a vnode or if it is just a negative cache entry.
314 */
315static void
316cache_zap(ncp)
317 struct namecache *ncp;
318{
319 struct vnode *vp;
320
321 rw_assert(&cache_lock, RA_WLOCKED);
322 CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
323#ifdef KDTRACE_HOOKS
324 if (ncp->nc_vp != NULL) {
325 SDT_PROBE(vfs, namecache, zap, done, ncp->nc_dvp,
326 ncp->nc_name, ncp->nc_vp, 0, 0);
327 } else {
328 SDT_PROBE(vfs, namecache, zap_negative, done, ncp->nc_dvp,
329 ncp->nc_name, 0, 0, 0);
330 }
331#endif
332 vp = NULL;
333 LIST_REMOVE(ncp, nc_hash);
334 if (ncp->nc_flag & NCF_ISDOTDOT) {
335 if (ncp == ncp->nc_dvp->v_cache_dd)
336 ncp->nc_dvp->v_cache_dd = NULL;
337 } else {
338 LIST_REMOVE(ncp, nc_src);
339 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
340 vp = ncp->nc_dvp;
341 numcachehv--;
342 }
343 }
344 if (ncp->nc_vp) {
345 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
346 if (ncp == ncp->nc_vp->v_cache_dd)
347 ncp->nc_vp->v_cache_dd = NULL;
348 } else {
349 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
350 numneg--;
351 }
352 numcache--;
353 cache_free(ncp);
354 if (vp)
355 vdrop(vp);
356}
357
358/*
359 * Lookup an entry in the cache
360 *
361 * Lookup is called with dvp pointing to the directory to search,
362 * cnp pointing to the name of the entry being sought. If the lookup
363 * succeeds, the vnode is returned in *vpp, and a status of -1 is
364 * returned. If the lookup determines that the name does not exist
365 * (negative cacheing), a status of ENOENT is returned. If the lookup
366 * fails, a status of zero is returned. If the directory vnode is
367 * recycled out from under us due to a forced unmount, a status of
368 * ENOENT is returned.
369 *
370 * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is
371 * unlocked. If we're looking up . an extra ref is taken, but the lock is
372 * not recursively acquired.
373 */
374
375int
376cache_lookup(dvp, vpp, cnp)
377 struct vnode *dvp;
378 struct vnode **vpp;
379 struct componentname *cnp;
380{
381 struct namecache *ncp;
|
382 u_int32_t hash;
| 382 uint32_t hash;
|
383 int error, ltype, wlocked;
384
385 if (!doingcache) {
386 cnp->cn_flags &= ~MAKEENTRY;
387 return (0);
388 }
389retry:
390 CACHE_RLOCK();
391 wlocked = 0;
392 numcalls++;
393 error = 0;
394
395retry_wlocked:
396 if (cnp->cn_nameptr[0] == '.') {
397 if (cnp->cn_namelen == 1) {
398 *vpp = dvp;
399 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
400 dvp, cnp->cn_nameptr);
401 dothits++;
402 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ".",
403 *vpp, 0, 0);
404 goto success;
405 }
406 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
407 dotdothits++;
408 if (dvp->v_cache_dd == NULL) {
409 SDT_PROBE(vfs, namecache, lookup, miss, dvp,
410 "..", NULL, 0, 0);
411 goto unlock;
412 }
413 if ((cnp->cn_flags & MAKEENTRY) == 0) {
414 if (!wlocked && !CACHE_UPGRADE_LOCK())
415 goto wlock;
416 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
417 cache_zap(dvp->v_cache_dd);
418 dvp->v_cache_dd = NULL;
419 CACHE_WUNLOCK();
420 return (0);
421 }
422 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
423 *vpp = dvp->v_cache_dd->nc_vp;
424 else
425 *vpp = dvp->v_cache_dd->nc_dvp;
426 /* Return failure if negative entry was found. */
427 if (*vpp == NULL) {
428 ncp = dvp->v_cache_dd;
429 goto negative_success;
430 }
431 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
432 dvp, cnp->cn_nameptr, *vpp);
433 SDT_PROBE(vfs, namecache, lookup, hit, dvp, "..",
434 *vpp, 0, 0);
435 goto success;
436 }
437 }
438
439 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
440 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
441 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
442 numchecks++;
443 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
444 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
445 break;
446 }
447
448 /* We failed to find an entry */
449 if (ncp == NULL) {
450 SDT_PROBE(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
451 NULL, 0, 0);
452 if ((cnp->cn_flags & MAKEENTRY) == 0) {
453 nummisszap++;
454 } else {
455 nummiss++;
456 }
457 nchstats.ncs_miss++;
458 goto unlock;
459 }
460
461 /* We don't want to have an entry, so dump it */
462 if ((cnp->cn_flags & MAKEENTRY) == 0) {
463 numposzaps++;
464 nchstats.ncs_badhits++;
465 if (!wlocked && !CACHE_UPGRADE_LOCK())
466 goto wlock;
467 cache_zap(ncp);
468 CACHE_WUNLOCK();
469 return (0);
470 }
471
472 /* We found a "positive" match, return the vnode */
473 if (ncp->nc_vp) {
474 numposhits++;
475 nchstats.ncs_goodhits++;
476 *vpp = ncp->nc_vp;
477 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
478 dvp, cnp->cn_nameptr, *vpp, ncp);
479 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ncp->nc_name,
480 *vpp, 0, 0);
481 goto success;
482 }
483
484negative_success:
485 /* We found a negative match, and want to create it, so purge */
486 if (cnp->cn_nameiop == CREATE) {
487 numnegzaps++;
488 nchstats.ncs_badhits++;
489 if (!wlocked && !CACHE_UPGRADE_LOCK())
490 goto wlock;
491 cache_zap(ncp);
492 CACHE_WUNLOCK();
493 return (0);
494 }
495
496 if (!wlocked && !CACHE_UPGRADE_LOCK())
497 goto wlock;
498 numneghits++;
499 /*
500 * We found a "negative" match, so we shift it to the end of
501 * the "negative" cache entries queue to satisfy LRU. Also,
502 * check to see if the entry is a whiteout; indicate this to
503 * the componentname, if so.
504 */
505 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
506 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
507 nchstats.ncs_neghits++;
508 if (ncp->nc_flag & NCF_WHITE)
509 cnp->cn_flags |= ISWHITEOUT;
510 SDT_PROBE(vfs, namecache, lookup, hit_negative, dvp, ncp->nc_name,
511 0, 0, 0);
512 CACHE_WUNLOCK();
513 return (ENOENT);
514
515wlock:
516 /*
517 * We need to update the cache after our lookup, so upgrade to
518 * a write lock and retry the operation.
519 */
520 CACHE_RUNLOCK();
521 CACHE_WLOCK();
522 numupgrades++;
523 wlocked = 1;
524 goto retry_wlocked;
525
526success:
527 /*
528 * On success we return a locked and ref'd vnode as per the lookup
529 * protocol.
530 */
531 if (dvp == *vpp) { /* lookup on "." */
532 VREF(*vpp);
533 if (wlocked)
534 CACHE_WUNLOCK();
535 else
536 CACHE_RUNLOCK();
537 /*
538 * When we lookup "." we still can be asked to lock it
539 * differently...
540 */
541 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
542 if (ltype != VOP_ISLOCKED(*vpp)) {
543 if (ltype == LK_EXCLUSIVE) {
544 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
545 if ((*vpp)->v_iflag & VI_DOOMED) {
546 /* forced unmount */
547 vrele(*vpp);
548 *vpp = NULL;
549 return (ENOENT);
550 }
551 } else
552 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
553 }
554 return (-1);
555 }
556 ltype = 0; /* silence gcc warning */
557 if (cnp->cn_flags & ISDOTDOT) {
558 ltype = VOP_ISLOCKED(dvp);
559 VOP_UNLOCK(dvp, 0);
560 }
561 VI_LOCK(*vpp);
562 if (wlocked)
563 CACHE_WUNLOCK();
564 else
565 CACHE_RUNLOCK();
566 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
567 if (cnp->cn_flags & ISDOTDOT) {
568 vn_lock(dvp, ltype | LK_RETRY);
569 if (dvp->v_iflag & VI_DOOMED) {
570 if (error == 0)
571 vput(*vpp);
572 *vpp = NULL;
573 return (ENOENT);
574 }
575 }
576 if (error) {
577 *vpp = NULL;
578 goto retry;
579 }
580 if ((cnp->cn_flags & ISLASTCN) &&
581 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
582 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
583 }
584 return (-1);
585
586unlock:
587 if (wlocked)
588 CACHE_WUNLOCK();
589 else
590 CACHE_RUNLOCK();
591 return (0);
592}
593
594/*
595 * Add an entry to the cache.
596 */
597void
598cache_enter(dvp, vp, cnp)
599 struct vnode *dvp;
600 struct vnode *vp;
601 struct componentname *cnp;
602{
603 struct namecache *ncp, *n2;
604 struct nchashhead *ncpp;
| 383 int error, ltype, wlocked;
384
385 if (!doingcache) {
386 cnp->cn_flags &= ~MAKEENTRY;
387 return (0);
388 }
389retry:
390 CACHE_RLOCK();
391 wlocked = 0;
392 numcalls++;
393 error = 0;
394
395retry_wlocked:
396 if (cnp->cn_nameptr[0] == '.') {
397 if (cnp->cn_namelen == 1) {
398 *vpp = dvp;
399 CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
400 dvp, cnp->cn_nameptr);
401 dothits++;
402 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ".",
403 *vpp, 0, 0);
404 goto success;
405 }
406 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
407 dotdothits++;
408 if (dvp->v_cache_dd == NULL) {
409 SDT_PROBE(vfs, namecache, lookup, miss, dvp,
410 "..", NULL, 0, 0);
411 goto unlock;
412 }
413 if ((cnp->cn_flags & MAKEENTRY) == 0) {
414 if (!wlocked && !CACHE_UPGRADE_LOCK())
415 goto wlock;
416 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
417 cache_zap(dvp->v_cache_dd);
418 dvp->v_cache_dd = NULL;
419 CACHE_WUNLOCK();
420 return (0);
421 }
422 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
423 *vpp = dvp->v_cache_dd->nc_vp;
424 else
425 *vpp = dvp->v_cache_dd->nc_dvp;
426 /* Return failure if negative entry was found. */
427 if (*vpp == NULL) {
428 ncp = dvp->v_cache_dd;
429 goto negative_success;
430 }
431 CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
432 dvp, cnp->cn_nameptr, *vpp);
433 SDT_PROBE(vfs, namecache, lookup, hit, dvp, "..",
434 *vpp, 0, 0);
435 goto success;
436 }
437 }
438
439 hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
440 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
441 LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
442 numchecks++;
443 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
444 !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))
445 break;
446 }
447
448 /* We failed to find an entry */
449 if (ncp == NULL) {
450 SDT_PROBE(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
451 NULL, 0, 0);
452 if ((cnp->cn_flags & MAKEENTRY) == 0) {
453 nummisszap++;
454 } else {
455 nummiss++;
456 }
457 nchstats.ncs_miss++;
458 goto unlock;
459 }
460
461 /* We don't want to have an entry, so dump it */
462 if ((cnp->cn_flags & MAKEENTRY) == 0) {
463 numposzaps++;
464 nchstats.ncs_badhits++;
465 if (!wlocked && !CACHE_UPGRADE_LOCK())
466 goto wlock;
467 cache_zap(ncp);
468 CACHE_WUNLOCK();
469 return (0);
470 }
471
472 /* We found a "positive" match, return the vnode */
473 if (ncp->nc_vp) {
474 numposhits++;
475 nchstats.ncs_goodhits++;
476 *vpp = ncp->nc_vp;
477 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
478 dvp, cnp->cn_nameptr, *vpp, ncp);
479 SDT_PROBE(vfs, namecache, lookup, hit, dvp, ncp->nc_name,
480 *vpp, 0, 0);
481 goto success;
482 }
483
484negative_success:
485 /* We found a negative match, and want to create it, so purge */
486 if (cnp->cn_nameiop == CREATE) {
487 numnegzaps++;
488 nchstats.ncs_badhits++;
489 if (!wlocked && !CACHE_UPGRADE_LOCK())
490 goto wlock;
491 cache_zap(ncp);
492 CACHE_WUNLOCK();
493 return (0);
494 }
495
496 if (!wlocked && !CACHE_UPGRADE_LOCK())
497 goto wlock;
498 numneghits++;
499 /*
500 * We found a "negative" match, so we shift it to the end of
501 * the "negative" cache entries queue to satisfy LRU. Also,
502 * check to see if the entry is a whiteout; indicate this to
503 * the componentname, if so.
504 */
505 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
506 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
507 nchstats.ncs_neghits++;
508 if (ncp->nc_flag & NCF_WHITE)
509 cnp->cn_flags |= ISWHITEOUT;
510 SDT_PROBE(vfs, namecache, lookup, hit_negative, dvp, ncp->nc_name,
511 0, 0, 0);
512 CACHE_WUNLOCK();
513 return (ENOENT);
514
515wlock:
516 /*
517 * We need to update the cache after our lookup, so upgrade to
518 * a write lock and retry the operation.
519 */
520 CACHE_RUNLOCK();
521 CACHE_WLOCK();
522 numupgrades++;
523 wlocked = 1;
524 goto retry_wlocked;
525
526success:
527 /*
528 * On success we return a locked and ref'd vnode as per the lookup
529 * protocol.
530 */
531 if (dvp == *vpp) { /* lookup on "." */
532 VREF(*vpp);
533 if (wlocked)
534 CACHE_WUNLOCK();
535 else
536 CACHE_RUNLOCK();
537 /*
538 * When we lookup "." we still can be asked to lock it
539 * differently...
540 */
541 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
542 if (ltype != VOP_ISLOCKED(*vpp)) {
543 if (ltype == LK_EXCLUSIVE) {
544 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
545 if ((*vpp)->v_iflag & VI_DOOMED) {
546 /* forced unmount */
547 vrele(*vpp);
548 *vpp = NULL;
549 return (ENOENT);
550 }
551 } else
552 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
553 }
554 return (-1);
555 }
556 ltype = 0; /* silence gcc warning */
557 if (cnp->cn_flags & ISDOTDOT) {
558 ltype = VOP_ISLOCKED(dvp);
559 VOP_UNLOCK(dvp, 0);
560 }
561 VI_LOCK(*vpp);
562 if (wlocked)
563 CACHE_WUNLOCK();
564 else
565 CACHE_RUNLOCK();
566 error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
567 if (cnp->cn_flags & ISDOTDOT) {
568 vn_lock(dvp, ltype | LK_RETRY);
569 if (dvp->v_iflag & VI_DOOMED) {
570 if (error == 0)
571 vput(*vpp);
572 *vpp = NULL;
573 return (ENOENT);
574 }
575 }
576 if (error) {
577 *vpp = NULL;
578 goto retry;
579 }
580 if ((cnp->cn_flags & ISLASTCN) &&
581 (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
582 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
583 }
584 return (-1);
585
586unlock:
587 if (wlocked)
588 CACHE_WUNLOCK();
589 else
590 CACHE_RUNLOCK();
591 return (0);
592}
593
594/*
595 * Add an entry to the cache.
596 */
597void
598cache_enter(dvp, vp, cnp)
599 struct vnode *dvp;
600 struct vnode *vp;
601 struct componentname *cnp;
602{
603 struct namecache *ncp, *n2;
604 struct nchashhead *ncpp;
|
605 u_int32_t hash;
| 605 uint32_t hash;
|
606 int flag;
607 int hold;
608 int zap;
609 int len;
610
611 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
612 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
613 ("cache_enter: Adding a doomed vnode"));
614 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
615 ("cache_enter: Doomed vnode used as src"));
616
617 if (!doingcache)
618 return;
619
620 /*
621 * Avoid blowout in namecache entries.
622 */
623 if (numcache >= desiredvnodes * 2)
624 return;
625
626 flag = 0;
627 if (cnp->cn_nameptr[0] == '.') {
628 if (cnp->cn_namelen == 1)
629 return;
630 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
631 CACHE_WLOCK();
632 /*
633 * If dotdot entry already exists, just retarget it
634 * to new parent vnode, otherwise continue with new
635 * namecache entry allocation.
636 */
637 if ((ncp = dvp->v_cache_dd) != NULL &&
638 ncp->nc_flag & NCF_ISDOTDOT) {
639 KASSERT(ncp->nc_dvp == dvp,
640 ("wrong isdotdot parent"));
641 if (ncp->nc_vp != NULL)
642 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
643 ncp, nc_dst);
644 else
645 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
646 if (vp != NULL)
647 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
648 ncp, nc_dst);
649 else
650 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
651 ncp->nc_vp = vp;
652 CACHE_WUNLOCK();
653 return;
654 }
655 dvp->v_cache_dd = NULL;
656 SDT_PROBE(vfs, namecache, enter, done, dvp, "..", vp,
657 0, 0);
658 CACHE_WUNLOCK();
659 flag = NCF_ISDOTDOT;
660 }
661 }
662
663 hold = 0;
664 zap = 0;
665
666 /*
667 * Calculate the hash key and setup as much of the new
668 * namecache entry as possible before acquiring the lock.
669 */
670 ncp = cache_alloc(cnp->cn_namelen);
671 ncp->nc_vp = vp;
672 ncp->nc_dvp = dvp;
673 ncp->nc_flag = flag;
674 len = ncp->nc_nlen = cnp->cn_namelen;
675 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
676 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1);
677 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
678 CACHE_WLOCK();
679
680 /*
681 * See if this vnode or negative entry is already in the cache
682 * with this name. This can happen with concurrent lookups of
683 * the same path name.
684 */
685 ncpp = NCHHASH(hash);
686 LIST_FOREACH(n2, ncpp, nc_hash) {
687 if (n2->nc_dvp == dvp &&
688 n2->nc_nlen == cnp->cn_namelen &&
689 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
690 CACHE_WUNLOCK();
691 cache_free(ncp);
692 return;
693 }
694 }
695
696 if (flag == NCF_ISDOTDOT) {
697 /*
698 * See if we are trying to add .. entry, but some other lookup
699 * has populated v_cache_dd pointer already.
700 */
701 if (dvp->v_cache_dd != NULL) {
702 CACHE_WUNLOCK();
703 cache_free(ncp);
704 return;
705 }
706 KASSERT(vp == NULL || vp->v_type == VDIR,
707 ("wrong vnode type %p", vp));
708 dvp->v_cache_dd = ncp;
709 }
710
711 numcache++;
712 if (!vp) {
713 numneg++;
714 if (cnp->cn_flags & ISWHITEOUT)
715 ncp->nc_flag |= NCF_WHITE;
716 } else if (vp->v_type == VDIR) {
717 if (flag != NCF_ISDOTDOT) {
718 if ((n2 = vp->v_cache_dd) != NULL &&
719 (n2->nc_flag & NCF_ISDOTDOT) != 0)
720 cache_zap(n2);
721 vp->v_cache_dd = ncp;
722 }
723 } else {
724 vp->v_cache_dd = NULL;
725 }
726
727 /*
728 * Insert the new namecache entry into the appropriate chain
729 * within the cache entries table.
730 */
731 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
732 if (flag != NCF_ISDOTDOT) {
733 if (LIST_EMPTY(&dvp->v_cache_src)) {
734 hold = 1;
735 numcachehv++;
736 }
737 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
738 }
739
740 /*
741 * If the entry is "negative", we place it into the
742 * "negative" cache queue, otherwise, we place it into the
743 * destination vnode's cache entries queue.
744 */
745 if (vp) {
746 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
747 SDT_PROBE(vfs, namecache, enter, done, dvp, ncp->nc_name, vp,
748 0, 0);
749 } else {
750 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
751 SDT_PROBE(vfs, namecache, enter_negative, done, dvp,
752 ncp->nc_name, 0, 0, 0);
753 }
754 if (numneg * ncnegfactor > numcache) {
755 ncp = TAILQ_FIRST(&ncneg);
756 zap = 1;
757 }
758 if (hold)
759 vhold(dvp);
760 if (zap)
761 cache_zap(ncp);
762 CACHE_WUNLOCK();
763}
764
765/*
766 * Name cache initialization, from vfs_init() when we are booting
767 */
768static void
769nchinit(void *dummy __unused)
770{
771
772 TAILQ_INIT(&ncneg);
773
774 cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
775 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
776 cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
777 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
778
779 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
780}
781SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
782
783
784/*
785 * Invalidate all entries to a particular vnode.
786 */
787void
788cache_purge(vp)
789 struct vnode *vp;
790{
791
792 CTR1(KTR_VFS, "cache_purge(%p)", vp);
793 SDT_PROBE(vfs, namecache, purge, done, vp, 0, 0, 0, 0);
794 CACHE_WLOCK();
795 while (!LIST_EMPTY(&vp->v_cache_src))
796 cache_zap(LIST_FIRST(&vp->v_cache_src));
797 while (!TAILQ_EMPTY(&vp->v_cache_dst))
798 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
799 if (vp->v_cache_dd != NULL) {
800 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
801 ("lost dotdot link"));
802 cache_zap(vp->v_cache_dd);
803 }
804 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
805 CACHE_WUNLOCK();
806}
807
808/*
809 * Invalidate all negative entries for a particular directory vnode.
810 */
811void
812cache_purge_negative(vp)
813 struct vnode *vp;
814{
815 struct namecache *cp, *ncp;
816
817 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
818 SDT_PROBE(vfs, namecache, purge_negative, done, vp, 0, 0, 0, 0);
819 CACHE_WLOCK();
820 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
821 if (cp->nc_vp == NULL)
822 cache_zap(cp);
823 }
824 CACHE_WUNLOCK();
825}
826
827/*
828 * Flush all entries referencing a particular filesystem.
829 */
830void
831cache_purgevfs(mp)
832 struct mount *mp;
833{
834 struct nchashhead *ncpp;
835 struct namecache *ncp, *nnp;
836
837 /* Scan hash tables for applicable entries */
838 SDT_PROBE(vfs, namecache, purgevfs, done, mp, 0, 0, 0, 0);
839 CACHE_WLOCK();
840 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
841 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
842 if (ncp->nc_dvp->v_mount == mp)
843 cache_zap(ncp);
844 }
845 }
846 CACHE_WUNLOCK();
847}
848
849/*
850 * Perform canonical checks and cache lookup and pass on to filesystem
851 * through the vop_cachedlookup only if needed.
852 */
853
854int
855vfs_cache_lookup(ap)
856 struct vop_lookup_args /* {
857 struct vnode *a_dvp;
858 struct vnode **a_vpp;
859 struct componentname *a_cnp;
860 } */ *ap;
861{
862 struct vnode *dvp;
863 int error;
864 struct vnode **vpp = ap->a_vpp;
865 struct componentname *cnp = ap->a_cnp;
866 struct ucred *cred = cnp->cn_cred;
867 int flags = cnp->cn_flags;
868 struct thread *td = cnp->cn_thread;
869
870 *vpp = NULL;
871 dvp = ap->a_dvp;
872
873 if (dvp->v_type != VDIR)
874 return (ENOTDIR);
875
876 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
877 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
878 return (EROFS);
879
880 error = VOP_ACCESS(dvp, VEXEC, cred, td);
881 if (error)
882 return (error);
883
884 error = cache_lookup(dvp, vpp, cnp);
885 if (error == 0)
886 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
887 if (error == -1)
888 return (0);
889 return (error);
890}
891
892
893#ifndef _SYS_SYSPROTO_H_
894struct __getcwd_args {
895 u_char *buf;
896 u_int buflen;
897};
898#endif
899
900/*
901 * XXX All of these sysctls would probably be more productive dead.
902 */
903static int disablecwd;
904SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
905 "Disable the getcwd syscall");
906
907/* Implementation of the getcwd syscall. */
908int
909__getcwd(td, uap)
910 struct thread *td;
911 struct __getcwd_args *uap;
912{
913
914 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
915}
916
917int
918kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
919{
920 char *bp, *tmpbuf;
921 struct filedesc *fdp;
922 struct vnode *cdir, *rdir;
923 int error, vfslocked;
924
925 if (disablecwd)
926 return (ENODEV);
927 if (buflen < 2)
928 return (EINVAL);
929 if (buflen > MAXPATHLEN)
930 buflen = MAXPATHLEN;
931
932 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
933 fdp = td->td_proc->p_fd;
934 FILEDESC_SLOCK(fdp);
935 cdir = fdp->fd_cdir;
936 VREF(cdir);
937 rdir = fdp->fd_rdir;
938 VREF(rdir);
939 FILEDESC_SUNLOCK(fdp);
940 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
941 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
942 vrele(rdir);
943 VFS_UNLOCK_GIANT(vfslocked);
944 vfslocked = VFS_LOCK_GIANT(cdir->v_mount);
945 vrele(cdir);
946 VFS_UNLOCK_GIANT(vfslocked);
947
948 if (!error) {
949 if (bufseg == UIO_SYSSPACE)
950 bcopy(bp, buf, strlen(bp) + 1);
951 else
952 error = copyout(bp, buf, strlen(bp) + 1);
953#ifdef KTRACE
954 if (KTRPOINT(curthread, KTR_NAMEI))
955 ktrnamei(bp);
956#endif
957 }
958 free(tmpbuf, M_TEMP);
959 return (error);
960}
961
962/*
963 * Thus begins the fullpath magic.
964 */
965
966#undef STATNODE
967#define STATNODE(name) \
968 static u_int name; \
969 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
970
971static int disablefullpath;
972SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
973 "Disable the vn_fullpath function");
974
975/* These count for kern___getcwd(), too. */
976STATNODE(numfullpathcalls);
977STATNODE(numfullpathfail1);
978STATNODE(numfullpathfail2);
979STATNODE(numfullpathfail4);
980STATNODE(numfullpathfound);
981
982/*
983 * Retrieve the full filesystem path that correspond to a vnode from the name
984 * cache (if available)
985 */
986int
987vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
988{
989 char *buf;
990 struct filedesc *fdp;
991 struct vnode *rdir;
992 int error, vfslocked;
993
994 if (disablefullpath)
995 return (ENODEV);
996 if (vn == NULL)
997 return (EINVAL);
998
999 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1000 fdp = td->td_proc->p_fd;
1001 FILEDESC_SLOCK(fdp);
1002 rdir = fdp->fd_rdir;
1003 VREF(rdir);
1004 FILEDESC_SUNLOCK(fdp);
1005 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1006 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
1007 vrele(rdir);
1008 VFS_UNLOCK_GIANT(vfslocked);
1009
1010 if (!error)
1011 *freebuf = buf;
1012 else
1013 free(buf, M_TEMP);
1014 return (error);
1015}
1016
1017/*
1018 * This function is similar to vn_fullpath, but it attempts to lookup the
1019 * pathname relative to the global root mount point. This is required for the
1020 * auditing sub-system, as audited pathnames must be absolute, relative to the
1021 * global root mount point.
1022 */
1023int
1024vn_fullpath_global(struct thread *td, struct vnode *vn,
1025 char **retbuf, char **freebuf)
1026{
1027 char *buf;
1028 int error;
1029
1030 if (disablefullpath)
1031 return (ENODEV);
1032 if (vn == NULL)
1033 return (EINVAL);
1034 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1035 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1036 if (!error)
1037 *freebuf = buf;
1038 else
1039 free(buf, M_TEMP);
1040 return (error);
1041}
1042
1043int
1044vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1045{
1046 int error;
1047
1048 CACHE_RLOCK();
1049 error = vn_vptocnp_locked(vp, cred, buf, buflen);
1050 if (error == 0) {
1051 /*
1052 * vn_vptocnp_locked() dropped hold acquired by
1053 * VOP_VPTOCNP immediately after locking the
1054 * cache. Since we are going to drop the cache rlock,
1055 * re-hold the result.
1056 */
1057 vhold(*vp);
1058 CACHE_RUNLOCK();
1059 }
1060 return (error);
1061}
1062
1063static int
1064vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1065 u_int *buflen)
1066{
1067 struct vnode *dvp;
1068 struct namecache *ncp;
1069 int error, vfslocked;
1070
1071 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1072 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1073 break;
1074 }
1075 if (ncp != NULL) {
1076 if (*buflen < ncp->nc_nlen) {
1077 CACHE_RUNLOCK();
1078 numfullpathfail4++;
1079 error = ENOMEM;
1080 SDT_PROBE(vfs, namecache, fullpath, return, error,
1081 vp, NULL, 0, 0);
1082 return (error);
1083 }
1084 *buflen -= ncp->nc_nlen;
1085 memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen);
1086 SDT_PROBE(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1087 ncp->nc_name, vp, 0, 0);
1088 *vp = ncp->nc_dvp;
1089 return (0);
1090 }
1091 SDT_PROBE(vfs, namecache, fullpath, miss, vp, 0, 0, 0, 0);
1092
1093 vhold(*vp);
1094 CACHE_RUNLOCK();
1095 vfslocked = VFS_LOCK_GIANT((*vp)->v_mount);
1096 vn_lock(*vp, LK_SHARED | LK_RETRY);
1097 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1098 VOP_UNLOCK(*vp, 0);
1099 vdrop(*vp);
1100 VFS_UNLOCK_GIANT(vfslocked);
1101 if (error) {
1102 numfullpathfail2++;
1103 SDT_PROBE(vfs, namecache, fullpath, return, error, vp,
1104 NULL, 0, 0);
1105 return (error);
1106 }
1107
1108 *vp = dvp;
1109 CACHE_RLOCK();
1110 if ((*vp)->v_iflag & VI_DOOMED) {
1111 /* forced unmount */
1112 CACHE_RUNLOCK();
1113 vdrop(*vp);
1114 error = ENOENT;
1115 SDT_PROBE(vfs, namecache, fullpath, return, error, vp,
1116 NULL, 0, 0);
1117 return (error);
1118 }
1119 vdrop(*vp);
1120
1121 return (0);
1122}
1123
1124/*
1125 * The magic behind kern___getcwd() and vn_fullpath().
1126 */
1127static int
1128vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1129 char *buf, char **retbuf, u_int buflen)
1130{
1131 int error, slash_prefixed;
1132#ifdef KDTRACE_HOOKS
1133 struct vnode *startvp = vp;
1134#endif
1135
1136 buflen--;
1137 buf[buflen] = '\0';
1138 error = 0;
1139 slash_prefixed = 0;
1140
1141 SDT_PROBE(vfs, namecache, fullpath, entry, vp, 0, 0, 0, 0);
1142 numfullpathcalls++;
1143 CACHE_RLOCK();
1144 if (vp->v_type != VDIR) {
1145 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1146 if (error)
1147 return (error);
1148 if (buflen == 0) {
1149 CACHE_RUNLOCK();
1150 return (ENOMEM);
1151 }
1152 buf[--buflen] = '/';
1153 slash_prefixed = 1;
1154 }
1155 while (vp != rdir && vp != rootvnode) {
1156 if (vp->v_vflag & VV_ROOT) {
1157 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */
1158 CACHE_RUNLOCK();
1159 error = ENOENT;
1160 SDT_PROBE(vfs, namecache, fullpath, return,
1161 error, vp, NULL, 0, 0);
1162 break;
1163 }
1164 vp = vp->v_mount->mnt_vnodecovered;
1165 continue;
1166 }
1167 if (vp->v_type != VDIR) {
1168 CACHE_RUNLOCK();
1169 numfullpathfail1++;
1170 error = ENOTDIR;
1171 SDT_PROBE(vfs, namecache, fullpath, return,
1172 error, vp, NULL, 0, 0);
1173 break;
1174 }
1175 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1176 if (error)
1177 break;
1178 if (buflen == 0) {
1179 CACHE_RUNLOCK();
1180 error = ENOMEM;
1181 SDT_PROBE(vfs, namecache, fullpath, return, error,
1182 startvp, NULL, 0, 0);
1183 break;
1184 }
1185 buf[--buflen] = '/';
1186 slash_prefixed = 1;
1187 }
1188 if (error)
1189 return (error);
1190 if (!slash_prefixed) {
1191 if (buflen == 0) {
1192 CACHE_RUNLOCK();
1193 numfullpathfail4++;
1194 SDT_PROBE(vfs, namecache, fullpath, return, ENOMEM,
1195 startvp, NULL, 0, 0);
1196 return (ENOMEM);
1197 }
1198 buf[--buflen] = '/';
1199 }
1200 numfullpathfound++;
1201 CACHE_RUNLOCK();
1202
1203 SDT_PROBE(vfs, namecache, fullpath, return, 0, startvp, buf + buflen,
1204 0, 0);
1205 *retbuf = buf + buflen;
1206 return (0);
1207}
1208
1209int
1210vn_commname(struct vnode *vp, char *buf, u_int buflen)
1211{
1212 struct namecache *ncp;
1213 int l;
1214
1215 CACHE_RLOCK();
1216 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1217 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1218 break;
1219 if (ncp == NULL) {
1220 CACHE_RUNLOCK();
1221 return (ENOENT);
1222 }
1223 l = min(ncp->nc_nlen, buflen - 1);
1224 memcpy(buf, ncp->nc_name, l);
1225 CACHE_RUNLOCK();
1226 buf[l] = '\0';
1227 return (0);
1228}
| 606 int flag;
607 int hold;
608 int zap;
609 int len;
610
611 CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
612 VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
613 ("cache_enter: Adding a doomed vnode"));
614 VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
615 ("cache_enter: Doomed vnode used as src"));
616
617 if (!doingcache)
618 return;
619
620 /*
621 * Avoid blowout in namecache entries.
622 */
623 if (numcache >= desiredvnodes * 2)
624 return;
625
626 flag = 0;
627 if (cnp->cn_nameptr[0] == '.') {
628 if (cnp->cn_namelen == 1)
629 return;
630 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
631 CACHE_WLOCK();
632 /*
633 * If dotdot entry already exists, just retarget it
634 * to new parent vnode, otherwise continue with new
635 * namecache entry allocation.
636 */
637 if ((ncp = dvp->v_cache_dd) != NULL &&
638 ncp->nc_flag & NCF_ISDOTDOT) {
639 KASSERT(ncp->nc_dvp == dvp,
640 ("wrong isdotdot parent"));
641 if (ncp->nc_vp != NULL)
642 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
643 ncp, nc_dst);
644 else
645 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
646 if (vp != NULL)
647 TAILQ_INSERT_HEAD(&vp->v_cache_dst,
648 ncp, nc_dst);
649 else
650 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
651 ncp->nc_vp = vp;
652 CACHE_WUNLOCK();
653 return;
654 }
655 dvp->v_cache_dd = NULL;
656 SDT_PROBE(vfs, namecache, enter, done, dvp, "..", vp,
657 0, 0);
658 CACHE_WUNLOCK();
659 flag = NCF_ISDOTDOT;
660 }
661 }
662
663 hold = 0;
664 zap = 0;
665
666 /*
667 * Calculate the hash key and setup as much of the new
668 * namecache entry as possible before acquiring the lock.
669 */
670 ncp = cache_alloc(cnp->cn_namelen);
671 ncp->nc_vp = vp;
672 ncp->nc_dvp = dvp;
673 ncp->nc_flag = flag;
674 len = ncp->nc_nlen = cnp->cn_namelen;
675 hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
676 strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1);
677 hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
678 CACHE_WLOCK();
679
680 /*
681 * See if this vnode or negative entry is already in the cache
682 * with this name. This can happen with concurrent lookups of
683 * the same path name.
684 */
685 ncpp = NCHHASH(hash);
686 LIST_FOREACH(n2, ncpp, nc_hash) {
687 if (n2->nc_dvp == dvp &&
688 n2->nc_nlen == cnp->cn_namelen &&
689 !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) {
690 CACHE_WUNLOCK();
691 cache_free(ncp);
692 return;
693 }
694 }
695
696 if (flag == NCF_ISDOTDOT) {
697 /*
698 * See if we are trying to add .. entry, but some other lookup
699 * has populated v_cache_dd pointer already.
700 */
701 if (dvp->v_cache_dd != NULL) {
702 CACHE_WUNLOCK();
703 cache_free(ncp);
704 return;
705 }
706 KASSERT(vp == NULL || vp->v_type == VDIR,
707 ("wrong vnode type %p", vp));
708 dvp->v_cache_dd = ncp;
709 }
710
711 numcache++;
712 if (!vp) {
713 numneg++;
714 if (cnp->cn_flags & ISWHITEOUT)
715 ncp->nc_flag |= NCF_WHITE;
716 } else if (vp->v_type == VDIR) {
717 if (flag != NCF_ISDOTDOT) {
718 if ((n2 = vp->v_cache_dd) != NULL &&
719 (n2->nc_flag & NCF_ISDOTDOT) != 0)
720 cache_zap(n2);
721 vp->v_cache_dd = ncp;
722 }
723 } else {
724 vp->v_cache_dd = NULL;
725 }
726
727 /*
728 * Insert the new namecache entry into the appropriate chain
729 * within the cache entries table.
730 */
731 LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
732 if (flag != NCF_ISDOTDOT) {
733 if (LIST_EMPTY(&dvp->v_cache_src)) {
734 hold = 1;
735 numcachehv++;
736 }
737 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
738 }
739
740 /*
741 * If the entry is "negative", we place it into the
742 * "negative" cache queue, otherwise, we place it into the
743 * destination vnode's cache entries queue.
744 */
745 if (vp) {
746 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
747 SDT_PROBE(vfs, namecache, enter, done, dvp, ncp->nc_name, vp,
748 0, 0);
749 } else {
750 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
751 SDT_PROBE(vfs, namecache, enter_negative, done, dvp,
752 ncp->nc_name, 0, 0, 0);
753 }
754 if (numneg * ncnegfactor > numcache) {
755 ncp = TAILQ_FIRST(&ncneg);
756 zap = 1;
757 }
758 if (hold)
759 vhold(dvp);
760 if (zap)
761 cache_zap(ncp);
762 CACHE_WUNLOCK();
763}
764
765/*
766 * Name cache initialization, from vfs_init() when we are booting
767 */
768static void
769nchinit(void *dummy __unused)
770{
771
772 TAILQ_INIT(&ncneg);
773
774 cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL, NULL,
775 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
776 cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE, NULL,
777 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_ZINIT);
778
779 nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
780}
781SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
782
783
784/*
785 * Invalidate all entries to a particular vnode.
786 */
787void
788cache_purge(vp)
789 struct vnode *vp;
790{
791
792 CTR1(KTR_VFS, "cache_purge(%p)", vp);
793 SDT_PROBE(vfs, namecache, purge, done, vp, 0, 0, 0, 0);
794 CACHE_WLOCK();
795 while (!LIST_EMPTY(&vp->v_cache_src))
796 cache_zap(LIST_FIRST(&vp->v_cache_src));
797 while (!TAILQ_EMPTY(&vp->v_cache_dst))
798 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
799 if (vp->v_cache_dd != NULL) {
800 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
801 ("lost dotdot link"));
802 cache_zap(vp->v_cache_dd);
803 }
804 KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
805 CACHE_WUNLOCK();
806}
807
808/*
809 * Invalidate all negative entries for a particular directory vnode.
810 */
811void
812cache_purge_negative(vp)
813 struct vnode *vp;
814{
815 struct namecache *cp, *ncp;
816
817 CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
818 SDT_PROBE(vfs, namecache, purge_negative, done, vp, 0, 0, 0, 0);
819 CACHE_WLOCK();
820 LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
821 if (cp->nc_vp == NULL)
822 cache_zap(cp);
823 }
824 CACHE_WUNLOCK();
825}
826
827/*
828 * Flush all entries referencing a particular filesystem.
829 */
830void
831cache_purgevfs(mp)
832 struct mount *mp;
833{
834 struct nchashhead *ncpp;
835 struct namecache *ncp, *nnp;
836
837 /* Scan hash tables for applicable entries */
838 SDT_PROBE(vfs, namecache, purgevfs, done, mp, 0, 0, 0, 0);
839 CACHE_WLOCK();
840 for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
841 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
842 if (ncp->nc_dvp->v_mount == mp)
843 cache_zap(ncp);
844 }
845 }
846 CACHE_WUNLOCK();
847}
848
849/*
850 * Perform canonical checks and cache lookup and pass on to filesystem
851 * through the vop_cachedlookup only if needed.
852 */
853
854int
855vfs_cache_lookup(ap)
856 struct vop_lookup_args /* {
857 struct vnode *a_dvp;
858 struct vnode **a_vpp;
859 struct componentname *a_cnp;
860 } */ *ap;
861{
862 struct vnode *dvp;
863 int error;
864 struct vnode **vpp = ap->a_vpp;
865 struct componentname *cnp = ap->a_cnp;
866 struct ucred *cred = cnp->cn_cred;
867 int flags = cnp->cn_flags;
868 struct thread *td = cnp->cn_thread;
869
870 *vpp = NULL;
871 dvp = ap->a_dvp;
872
873 if (dvp->v_type != VDIR)
874 return (ENOTDIR);
875
876 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
877 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
878 return (EROFS);
879
880 error = VOP_ACCESS(dvp, VEXEC, cred, td);
881 if (error)
882 return (error);
883
884 error = cache_lookup(dvp, vpp, cnp);
885 if (error == 0)
886 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
887 if (error == -1)
888 return (0);
889 return (error);
890}
891
892
893#ifndef _SYS_SYSPROTO_H_
894struct __getcwd_args {
895 u_char *buf;
896 u_int buflen;
897};
898#endif
899
900/*
901 * XXX All of these sysctls would probably be more productive dead.
902 */
903static int disablecwd;
904SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
905 "Disable the getcwd syscall");
906
907/* Implementation of the getcwd syscall. */
908int
909__getcwd(td, uap)
910 struct thread *td;
911 struct __getcwd_args *uap;
912{
913
914 return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen));
915}
916
917int
918kern___getcwd(struct thread *td, u_char *buf, enum uio_seg bufseg, u_int buflen)
919{
920 char *bp, *tmpbuf;
921 struct filedesc *fdp;
922 struct vnode *cdir, *rdir;
923 int error, vfslocked;
924
925 if (disablecwd)
926 return (ENODEV);
927 if (buflen < 2)
928 return (EINVAL);
929 if (buflen > MAXPATHLEN)
930 buflen = MAXPATHLEN;
931
932 tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
933 fdp = td->td_proc->p_fd;
934 FILEDESC_SLOCK(fdp);
935 cdir = fdp->fd_cdir;
936 VREF(cdir);
937 rdir = fdp->fd_rdir;
938 VREF(rdir);
939 FILEDESC_SUNLOCK(fdp);
940 error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
941 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
942 vrele(rdir);
943 VFS_UNLOCK_GIANT(vfslocked);
944 vfslocked = VFS_LOCK_GIANT(cdir->v_mount);
945 vrele(cdir);
946 VFS_UNLOCK_GIANT(vfslocked);
947
948 if (!error) {
949 if (bufseg == UIO_SYSSPACE)
950 bcopy(bp, buf, strlen(bp) + 1);
951 else
952 error = copyout(bp, buf, strlen(bp) + 1);
953#ifdef KTRACE
954 if (KTRPOINT(curthread, KTR_NAMEI))
955 ktrnamei(bp);
956#endif
957 }
958 free(tmpbuf, M_TEMP);
959 return (error);
960}
961
962/*
963 * Thus begins the fullpath magic.
964 */
965
966#undef STATNODE
967#define STATNODE(name) \
968 static u_int name; \
969 SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, "")
970
971static int disablefullpath;
972SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
973 "Disable the vn_fullpath function");
974
975/* These count for kern___getcwd(), too. */
976STATNODE(numfullpathcalls);
977STATNODE(numfullpathfail1);
978STATNODE(numfullpathfail2);
979STATNODE(numfullpathfail4);
980STATNODE(numfullpathfound);
981
982/*
983 * Retrieve the full filesystem path that correspond to a vnode from the name
984 * cache (if available)
985 */
986int
987vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
988{
989 char *buf;
990 struct filedesc *fdp;
991 struct vnode *rdir;
992 int error, vfslocked;
993
994 if (disablefullpath)
995 return (ENODEV);
996 if (vn == NULL)
997 return (EINVAL);
998
999 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1000 fdp = td->td_proc->p_fd;
1001 FILEDESC_SLOCK(fdp);
1002 rdir = fdp->fd_rdir;
1003 VREF(rdir);
1004 FILEDESC_SUNLOCK(fdp);
1005 error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
1006 vfslocked = VFS_LOCK_GIANT(rdir->v_mount);
1007 vrele(rdir);
1008 VFS_UNLOCK_GIANT(vfslocked);
1009
1010 if (!error)
1011 *freebuf = buf;
1012 else
1013 free(buf, M_TEMP);
1014 return (error);
1015}
1016
1017/*
1018 * This function is similar to vn_fullpath, but it attempts to lookup the
1019 * pathname relative to the global root mount point. This is required for the
1020 * auditing sub-system, as audited pathnames must be absolute, relative to the
1021 * global root mount point.
1022 */
1023int
1024vn_fullpath_global(struct thread *td, struct vnode *vn,
1025 char **retbuf, char **freebuf)
1026{
1027 char *buf;
1028 int error;
1029
1030 if (disablefullpath)
1031 return (ENODEV);
1032 if (vn == NULL)
1033 return (EINVAL);
1034 buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1035 error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
1036 if (!error)
1037 *freebuf = buf;
1038 else
1039 free(buf, M_TEMP);
1040 return (error);
1041}
1042
1043int
1044vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
1045{
1046 int error;
1047
1048 CACHE_RLOCK();
1049 error = vn_vptocnp_locked(vp, cred, buf, buflen);
1050 if (error == 0) {
1051 /*
1052 * vn_vptocnp_locked() dropped hold acquired by
1053 * VOP_VPTOCNP immediately after locking the
1054 * cache. Since we are going to drop the cache rlock,
1055 * re-hold the result.
1056 */
1057 vhold(*vp);
1058 CACHE_RUNLOCK();
1059 }
1060 return (error);
1061}
1062
1063static int
1064vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
1065 u_int *buflen)
1066{
1067 struct vnode *dvp;
1068 struct namecache *ncp;
1069 int error, vfslocked;
1070
1071 TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
1072 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1073 break;
1074 }
1075 if (ncp != NULL) {
1076 if (*buflen < ncp->nc_nlen) {
1077 CACHE_RUNLOCK();
1078 numfullpathfail4++;
1079 error = ENOMEM;
1080 SDT_PROBE(vfs, namecache, fullpath, return, error,
1081 vp, NULL, 0, 0);
1082 return (error);
1083 }
1084 *buflen -= ncp->nc_nlen;
1085 memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen);
1086 SDT_PROBE(vfs, namecache, fullpath, hit, ncp->nc_dvp,
1087 ncp->nc_name, vp, 0, 0);
1088 *vp = ncp->nc_dvp;
1089 return (0);
1090 }
1091 SDT_PROBE(vfs, namecache, fullpath, miss, vp, 0, 0, 0, 0);
1092
1093 vhold(*vp);
1094 CACHE_RUNLOCK();
1095 vfslocked = VFS_LOCK_GIANT((*vp)->v_mount);
1096 vn_lock(*vp, LK_SHARED | LK_RETRY);
1097 error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
1098 VOP_UNLOCK(*vp, 0);
1099 vdrop(*vp);
1100 VFS_UNLOCK_GIANT(vfslocked);
1101 if (error) {
1102 numfullpathfail2++;
1103 SDT_PROBE(vfs, namecache, fullpath, return, error, vp,
1104 NULL, 0, 0);
1105 return (error);
1106 }
1107
1108 *vp = dvp;
1109 CACHE_RLOCK();
1110 if ((*vp)->v_iflag & VI_DOOMED) {
1111 /* forced unmount */
1112 CACHE_RUNLOCK();
1113 vdrop(*vp);
1114 error = ENOENT;
1115 SDT_PROBE(vfs, namecache, fullpath, return, error, vp,
1116 NULL, 0, 0);
1117 return (error);
1118 }
1119 vdrop(*vp);
1120
1121 return (0);
1122}
1123
1124/*
1125 * The magic behind kern___getcwd() and vn_fullpath().
1126 */
1127static int
1128vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
1129 char *buf, char **retbuf, u_int buflen)
1130{
1131 int error, slash_prefixed;
1132#ifdef KDTRACE_HOOKS
1133 struct vnode *startvp = vp;
1134#endif
1135
1136 buflen--;
1137 buf[buflen] = '\0';
1138 error = 0;
1139 slash_prefixed = 0;
1140
1141 SDT_PROBE(vfs, namecache, fullpath, entry, vp, 0, 0, 0, 0);
1142 numfullpathcalls++;
1143 CACHE_RLOCK();
1144 if (vp->v_type != VDIR) {
1145 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1146 if (error)
1147 return (error);
1148 if (buflen == 0) {
1149 CACHE_RUNLOCK();
1150 return (ENOMEM);
1151 }
1152 buf[--buflen] = '/';
1153 slash_prefixed = 1;
1154 }
1155 while (vp != rdir && vp != rootvnode) {
1156 if (vp->v_vflag & VV_ROOT) {
1157 if (vp->v_iflag & VI_DOOMED) { /* forced unmount */
1158 CACHE_RUNLOCK();
1159 error = ENOENT;
1160 SDT_PROBE(vfs, namecache, fullpath, return,
1161 error, vp, NULL, 0, 0);
1162 break;
1163 }
1164 vp = vp->v_mount->mnt_vnodecovered;
1165 continue;
1166 }
1167 if (vp->v_type != VDIR) {
1168 CACHE_RUNLOCK();
1169 numfullpathfail1++;
1170 error = ENOTDIR;
1171 SDT_PROBE(vfs, namecache, fullpath, return,
1172 error, vp, NULL, 0, 0);
1173 break;
1174 }
1175 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
1176 if (error)
1177 break;
1178 if (buflen == 0) {
1179 CACHE_RUNLOCK();
1180 error = ENOMEM;
1181 SDT_PROBE(vfs, namecache, fullpath, return, error,
1182 startvp, NULL, 0, 0);
1183 break;
1184 }
1185 buf[--buflen] = '/';
1186 slash_prefixed = 1;
1187 }
1188 if (error)
1189 return (error);
1190 if (!slash_prefixed) {
1191 if (buflen == 0) {
1192 CACHE_RUNLOCK();
1193 numfullpathfail4++;
1194 SDT_PROBE(vfs, namecache, fullpath, return, ENOMEM,
1195 startvp, NULL, 0, 0);
1196 return (ENOMEM);
1197 }
1198 buf[--buflen] = '/';
1199 }
1200 numfullpathfound++;
1201 CACHE_RUNLOCK();
1202
1203 SDT_PROBE(vfs, namecache, fullpath, return, 0, startvp, buf + buflen,
1204 0, 0);
1205 *retbuf = buf + buflen;
1206 return (0);
1207}
1208
1209int
1210vn_commname(struct vnode *vp, char *buf, u_int buflen)
1211{
1212 struct namecache *ncp;
1213 int l;
1214
1215 CACHE_RLOCK();
1216 TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
1217 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
1218 break;
1219 if (ncp == NULL) {
1220 CACHE_RUNLOCK();
1221 return (ENOENT);
1222 }
1223 l = min(ncp->nc_nlen, buflen - 1);
1224 memcpy(buf, ncp->nc_name, l);
1225 CACHE_RUNLOCK();
1226 buf[l] = '\0';
1227 return (0);
1228}
|