1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
26 /* All Rights Reserved */
27
28 /*
29 * University Copyright- Copyright (c) 1982, 1986, 1988
30 * The Regents of the University of California
31 * All Rights Reserved
32 *
33 * University Acknowledgment- Portions of this document are derived from
34 * software developed by the University of California, Berkeley, and its
35 * contributors.
36 */
37
38 #ifndef _SYS_DNLC_H
39 #define _SYS_DNLC_H
40
41 #ifdef __cplusplus
42 extern "C" {
43 #endif
44
45 #include <sys/kstat.h>
46
47 /*
48 * DNLC - Directory name lookup cache.
49 * There are now two sorts of name caching:
50 *
51 * Standard dnlc: This original cache holds recent mappings
52 * of <directory vnode, name> to vnode mappings.
53 *
54 * Directory caches: Entire large directories can be cached, subject to
55 * memory availability and tunables. A directory cache
56 * anchor point must be provided in the xxnode for
57 * a directory.
58 */
59
60
61 /*
62 * Standard dnlc
63 * =============
64 */
65
66 /*
67 * This structure describes the elements in the cache of recent
68 * names looked up.
69 *
70 * Note namlen is a uchar_t to conserve space
71 * and alignment padding. The max length of any
72 * pathname component is defined as MAXNAMELEN
73 * which is 256 (including the terminating null).
74 * So provided this doesn't change, we don't include the null,
75 * we always use bcmp to compare strings, and we don't start
76 * storing full names, then we are ok. The space savings are worth it.
77 */
78 typedef struct ncache {
79 struct ncache *hash_next; /* hash chain, MUST BE FIRST */
80 struct ncache *hash_prev;
81 struct vnode *vp; /* vnode the name refers to */
82 struct vnode *dp; /* vnode of parent of name */
83 int hash; /* hash signature */
84 uchar_t namlen; /* length of name */
85 char name[1]; /* segment name - null terminated */
86 } ncache_t;
87
88 /*
89 * Hash table bucket structure of name cache entries for fast lookup.
90 */
91 typedef struct nc_hash {
92 ncache_t *hash_next;
93 ncache_t *hash_prev;
94 kmutex_t hash_lock;
95 } nc_hash_t;
96
97 /*
98 * Statistics on name cache
99 * Not protected by locks
100 */
101 /*
102 * ncstats has been deprecated, due to the integer size of the counters
103 * which can easily overflow in the dnlc.
104 * It is maintained (at some expense) for compatability.
105 * The preferred interface is the kstat accessible nc_stats below, ehich
106 * is actually shared with directory caching.
107 */
108 struct ncstats {
109 int hits; /* hits that we can really use */
110 int misses; /* cache misses */
111 int enters; /* number of enters done */
112 int dbl_enters; /* number of enters tried when already cached */
113 int long_enter; /* deprecated, no longer accounted */
114 int long_look; /* deprecated, no longer accounted */
115 int move_to_front; /* entry moved to front of hash chain */
116 int purges; /* number of purges of cache */
117 };
118
119 struct nc_stats {
120 kstat_named_t ncs_hits; /* cache hits */
121 kstat_named_t ncs_misses; /* cache misses */
122 kstat_named_t ncs_neg_hits; /* negative cache hits */
123 kstat_named_t ncs_enters; /* enters */
124 kstat_named_t ncs_dbl_enters; /* enters when entry already cached */
125 kstat_named_t ncs_purge_total; /* total entries prurged */
126 kstat_named_t ncs_purge_all; /* dnlc_purge() calls */
127 kstat_named_t ncs_purge_vp; /* dnlc_purge_vp() calls */
128 kstat_named_t ncs_purge_vfs; /* dnlc_purge_vfs() calls */
129 kstat_named_t ncs_purge_fs1; /* dnlc_purge_fs1() calls */
130 kstat_named_t ncs_pick_free; /* found a free ncache */
131 kstat_named_t ncs_pick_heur; /* found ncache w/ NULL vpages */
132 kstat_named_t ncs_pick_last; /* found last ncache on chain */
133
134 /* directory caching stats */
135
136 kstat_named_t ncs_dir_hits; /* dir cache hits */
137 kstat_named_t ncs_dir_misses; /* dir cache misses */
138 kstat_named_t ncs_cur_dirs; /* current # directories cached */
139 kstat_named_t ncs_dir_num_ents; /* current # entries cached */
140 kstat_named_t ncs_dirs_cached; /* total # directories cached */
141 kstat_named_t ncs_dir_start_nm; /* dir start no memory */
142 kstat_named_t ncs_dir_add_nm; /* add entry/space - no memory */
143 kstat_named_t ncs_dir_addabort; /* add entry/space - abort */
144 kstat_named_t ncs_dir_add_max; /* add entry/space - max exceeded */
145 kstat_named_t ncs_dir_reme_fai; /* remove entry fail */
146 kstat_named_t ncs_dir_rems_fai; /* remove space fail */
147 kstat_named_t ncs_dir_upd_fail; /* update space fail */
148 kstat_named_t ncs_dir_finipurg; /* fini purges */
149 kstat_named_t ncs_dir_rec_last; /* reclaim last */
150 kstat_named_t ncs_dir_recl_any; /* reclaim any */
151 };
152
153 /*
154 * The dnlc hashing function.
155 * Although really a kernel macro we export it to allow validation
156 * of ncache_t entries by mdb. Note, mdb can handle the ASSERT.
157 *
158 * 'hash' and 'namlen' must be l-values. A check is made to ensure
159 * the name length fits into an unsigned char (see ncache_t).
160 */
161 #define DNLCHASH(name, dvp, hash, namlen) \
162 { \
163 char Xc; \
164 const char *Xcp; \
165 hash = (int)((uintptr_t)(dvp)) >> 8; \
166 for (Xcp = (name); (Xc = *Xcp) != 0; Xcp++) \
167 (hash) = ((hash) << 4) + (hash) + Xc; \
168 ASSERT((Xcp - (name)) <= ((1 << NBBY) - 1)); \
169 (namlen) = Xcp - (name); \
170 }
171
172 #if defined(_KERNEL)
173
174 #include <sys/vfs.h>
175 #include <sys/vnode.h>
176
177 extern int ncsize; /* set in param_init() # of dnlc entries */
178 extern vnode_t negative_cache_vnode;
179 #define DNLC_NO_VNODE &negative_cache_vnode
180
181 void dnlc_init(void);
182 void dnlc_enter(vnode_t *, const char *, vnode_t *);
183 void dnlc_update(vnode_t *, const char *, vnode_t *);
184 vnode_t *dnlc_lookup(vnode_t *, const char *);
185 void dnlc_purge(void);
186 void dnlc_purge_vp(vnode_t *);
187 int dnlc_purge_vfsp(vfs_t *, int);
188 void dnlc_remove(vnode_t *, const char *);
189 int dnlc_fs_purge1(struct vnodeops *);
190 void dnlc_reduce_cache(void *);
191
192 #endif /* defined(_KERNEL) */
193
194
195 /*
196 * Directory caching interfaces
197 * ============================
198 */
199
200 /*
201 * Typically for large directories, the file names will be the same or
202 * at least similar lengths. So there's no point in anything more elaborate
203 * than a simple unordered linked list of free space entries.
204 * For small directories the name length distribution doesn't really matter.
205 */
206 typedef struct dcfree {
207 uint64_t df_handle; /* fs supplied handle */
208 struct dcfree *df_next; /* link to next free entry in bucket */
209 uint_t df_len; /* length of free entry */
210 } dcfree_t;
211
212 typedef struct dcentry {
213 uint64_t de_handle; /* fs supplied and returned data */
214 struct dcentry *de_next; /* link to next name entry in bucket */
215 int de_hash; /* hash signature */
216 uchar_t de_namelen; /* length of name excluding null */
217 char de_name[1]; /* null terminated name */
218 } dcentry_t;
219
220 typedef struct dircache {
221 struct dircache *dc_next; /* chain - for purge purposes */
222 struct dircache *dc_prev; /* chain - for purge purposes */
223 int64_t dc_actime; /* dir access time, from lbolt64 */
224 dcentry_t **dc_namehash; /* entry hash table pointer */
225 dcfree_t **dc_freehash; /* free entry hash table pointer */
226 uint_t dc_num_entries; /* no of named entries */
227 uint_t dc_num_free; /* no of free space entries */
228 uint_t dc_nhash_mask; /* name hash table size - 1 */
229 uint_t dc_fhash_mask; /* free space hash table size - 1 */
230 struct dcanchor *dc_anchor; /* back ptr to anchor */
231 boolean_t dc_complete; /* cache complete boolean */
232 } dircache_t;
233
234 typedef struct dcanchor {
235 void *dca_dircache; /* pointer to directory cache */
236 kmutex_t dca_lock; /* protects the directory cache */
237 } dcanchor_t;
238
239 /*
240 * Head struct for doubly linked chain of dircache_t
241 * The next and prev fields must match those of a dircache_t
242 */
243 typedef struct {
244 dircache_t *dch_next; /* next in chain */
245 dircache_t *dch_prev; /* prev in chain */
246 kmutex_t dch_lock; /* lock for the chain */
247 } dchead_t;
248
249
250 #if defined(_KERNEL)
251
252 /*
253 * Status returns from the directory cache interfaces
254 */
255 typedef enum {
256 DOK, /* operation sucessful */
257 DNOCACHE, /* there is no cache */
258 DFOUND, /* entry found */
259 DNOENT, /* no entry found */
260 DTOOBIG, /* exceeds tunable dnlc_max_dir_cache */
261 DNOMEM /* no memory */
262 } dcret_t;
263
264 /*
265 * dnlc_dir_start() requests that a directory be cached.
266 * This must be called initially to enable caching on a directory.
267 * After a successful call, directory entries and free space can be
268 * added (see below) until the directory is marked complete.
269 * "num_entries" is an estimate of the current number of
270 * directory entries. The request is rejected with DNOCACHE
271 * if num_entries falls below the tunable dnlc_dir_min_size (see
272 * below), and rejected with DTOOBIG if it's above dnlc_dir_max_size.
273 * Returns DOK, DNOCACHE, DTOOBIG, DNOMEM.
274 *
275 * Due to memory shortages, directory caches can be purged at
276 * any time. If the last directory cache is purged due to memory
277 * shortage, then the directory cache is marked internally
278 * as "no memory". Future returns will all be DNOCACHE until
279 * the next dnlc_start_dir() which will return DNOMEM once.
280 * This memory shortage may only be transient. It's up to the
281 * file system as to what to do about this condition, but an
282 * attempt to immediately re-build the cache will very likely
283 * lead to the same shortage of memory and a thrashing situation.
284 *
285 * It's file system policy as to when and what size directories to cache.
286 */
287 dcret_t dnlc_dir_start(dcanchor_t *dcap, uint_t num_entries);
288
289 /*
290 * dnlc_dir_add_entry() adds an entry (name and handle) into a
291 * partial or complete cache. "handle" is a file system specific
292 * quantity that is returned on calls to dnlc_dir_lookup() - see below.
293 * For example, "handle" for ufs holds the inumber and a directory
294 * entry offset. Returns DOK, DNOCACHE, DTOOBIG.
295 */
296 dcret_t dnlc_dir_add_entry(dcanchor_t *dcap, const char *name, uint64_t handle);
297
298 /*
299 * dnlc_dir_add_space adds free space (length and file system specific
300 * handle) into a partial or complete cache. "handle" is a file
301 * system specific quantity that is returned on calls to
302 * dnlc_dir_rem_space_by_len(). For example, "handle" for ufs holds
303 * the directory entry offset. Returns DOK, DNOCACHE, DTOOBIG.
304 */
305 dcret_t dnlc_dir_add_space(dcanchor_t *dcap, uint_t len, uint64_t handle);
306
307 /*
308 * dnlc_dir_complete() indicates the previously partial cache is now complete.
309 */
310 void dnlc_dir_complete(dcanchor_t *dcap);
311
312 /*
313 * dnlc_dir_purge() deletes a partial or complete directory cache
314 */
315 void dnlc_dir_purge(dcanchor_t *dcap);
316
317 /*
318 * dnlc_dir_lookup() lookups a file name in a directory cache
319 * and returns the file system handle specified on dnlc_dir_add_entry()
320 * in "handlep". Returns DFOUND, DNOENT, DNOCACHE.
321 */
322 dcret_t dnlc_dir_lookup(dcanchor_t *dcap, const char *name, uint64_t *handlep);
323
324 /*
325 * dnlc_dir_update() amends the handle for an entry in a directory cache
326 * "handle" is the new file system specific handle for the file "name".
327 * Returns DFOUND, DNOENT, DNOCACHE.
328 */
329 dcret_t dnlc_dir_update(dcanchor_t *dcap, const char *name, uint64_t handle);
330
331 /*
332 * dnlc_dir_rem_entry() removes an entry form a directory cache.
333 * Returns the handle if "handlep" non null.
334 * Returns DFOUND, DNOENT, DNOCACHE.
335 */
336 dcret_t dnlc_dir_rem_entry(dcanchor_t *dcap, const char *name,
337 uint64_t *handlep);
338
339 /*
340 * dnlc_dir_rem_space_by_len() looks up and returns free space in a
341 * directory cache of at least the given "len". Returns in "handlep"
342 * the handle supplied when adding the free space in dnlc_dir_add_space().
343 * Returns DFOUND, DNOENT, DNOCACHE.
344 */
345 dcret_t dnlc_dir_rem_space_by_len(dcanchor_t *dcap, uint_t len,
346 uint64_t *handlep);
347
348 /*
349 * dnlc_dir_rem_space_by_handle() looks up and removes the free space in
350 * a directory cache with the given handle. Returns DFOUND, DNOENT, DNOCACHE.
351 */
352 dcret_t dnlc_dir_rem_space_by_handle(dcanchor_t *dcap, uint64_t handle);
353
354 /*
355 * dnlc_dir_init() initialises a directory anchor
356 */
357 #define dnlc_dir_init(dcap) { \
358 (dcap)->dca_dircache = NULL; \
359 mutex_init(&(dcap)->dca_lock, NULL, MUTEX_DEFAULT, NULL); }
360
361 /*
362 * dnlc_dir_fini() is called to indicate the anchor is no longer used.
363 * It ensures there's no directory cache and mutex_destroys the lock
364 */
365 void dnlc_dir_fini(dcanchor_t *dcap);
366
367 #endif /* defined(_KERNEL) */
368
369 #ifdef __cplusplus
370 }
371 #endif
372
373 #endif /* _SYS_DNLC_H */