vfs_dirhash.c revision 1.4 1 /* $NetBSD: vfs_dirhash.c,v 1.4 2008/10/30 17:19:18 reinoud Exp $ */
2
3 /*
4 * Copyright (c) 2008 Reinoud Zandijk
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 *
27 */
28
29
30 #include <sys/cdefs.h>
31 #ifndef lint
32 __KERNEL_RCSID(0, "$NetBSD: vfs_dirhash.c,v 1.4 2008/10/30 17:19:18 reinoud Exp $");
33 #endif /* not lint */
34
35 #if 1
36 # define DPRINTF(a) ;
37 #else
38 # define DPRINTF(a) printf(a);
39 #endif
40
41 /* CLEAN UP! */
42 #include <sys/sysctl.h>
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysctl.h>
46 #include <sys/namei.h>
47 #include <sys/proc.h>
48 #include <sys/kernel.h>
49 #include <sys/vnode.h>
50 #include <sys/mount.h>
51 #include <sys/buf.h>
52 #include <sys/file.h>
53 #include <sys/device.h>
54 #include <sys/disklabel.h>
55 #include <sys/ioctl.h>
56 #include <sys/malloc.h>
57 #include <sys/dirent.h>
58 #include <sys/stat.h>
59 #include <sys/conf.h>
60 #include <sys/kauth.h>
61 #include <dev/clock_subr.h>
62
63 #include <sys/dirhash.h>
64
65
66 static struct sysctllog *sysctl_log;
67 struct pool dirhash_pool;
68 struct pool dirhash_entry_pool;
69
70 static kmutex_t dirhashmutex;
71 static uint32_t maxdirhashsize = DIRHASH_SIZE;
72 static uint32_t dirhashsize = 0;
73 static TAILQ_HEAD(_dirhash, dirhash) dirhash_queue;
74
75
76 void
77 dirhash_init(void)
78 {
79 const struct sysctlnode *rnode, *cnode;
80 size_t size;
81 uint32_t max_entries;
82
83 /* initialise dirhash queue */
84 TAILQ_INIT(&dirhash_queue);
85
86 /* init dirhash pools */
87 size = sizeof(struct dirhash);
88 pool_init(&dirhash_pool, size, 0, 0, 0,
89 "dirhash_pool", NULL, IPL_NONE);
90
91 size = sizeof(struct dirhash_entry);
92 pool_init(&dirhash_entry_pool, size, 0, 0, 0,
93 "dirhash_entry_pool", NULL, IPL_NONE);
94
95 mutex_init(&dirhashmutex, MUTEX_DEFAULT, IPL_NONE);
96 max_entries = maxdirhashsize / size;
97 pool_sethiwat(&dirhash_entry_pool, max_entries);
98 dirhashsize = 0;
99
100 /* create sysctl knobs and dials */
101 sysctl_log = NULL;
102 sysctl_createv(&sysctl_log, 0, NULL, &rnode,
103 CTLFLAG_PERMANENT,
104 CTLTYPE_NODE, "dirhash", NULL,
105 NULL, 0, NULL, 0,
106 CTL_VFS, VFS_GENERIC, CTL_CREATE, CTL_EOL);
107 sysctl_createv(&sysctl_log, 0, &rnode, &cnode,
108 CTLFLAG_PERMANENT,
109 CTLTYPE_INT, "memused",
110 SYSCTL_DESCR("current dirhash memory usage"),
111 NULL, 0, &dirhashsize, 0,
112 CTL_CREATE, CTL_EOL);
113 sysctl_createv(&sysctl_log, 0, &rnode, &cnode,
114 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
115 CTLTYPE_INT, "maxmem",
116 SYSCTL_DESCR("maximum dirhash memory usage"),
117 NULL, 0, &maxdirhashsize, 0,
118 CTL_CREATE, CTL_EOL);
119 }
120
121
122 #if 0
123 void
124 dirhash_finish(void)
125 {
126 pool_destroy(&dirhash_pool);
127 pool_destroy(&dirhash_entry_pool);
128
129 mutex_destroy(&dirhashmutex);
130
131 /* sysctl_teardown(&sysctl_log); */
132 }
133 #endif
134
135
136 /*
137 * generic dirhash implementation
138 */
139
140 static uint32_t
141 dirhash_hash(const char *str, int namelen)
142 {
143 uint32_t hash = 5381;
144 int i, c;
145
146 for (i = 0; i < namelen; i++) {
147 c = *str++;
148 hash = ((hash << 5) + hash) + c; /* hash * 33 + c */
149 }
150 return hash;
151 }
152
153
154 void
155 dirhash_purge_entries(struct dirhash *dirh)
156 {
157 struct dirhash_entry *dirh_e;
158 uint32_t hashline;
159
160 if (dirh == NULL)
161 return;
162
163 if (dirh->size == 0)
164 return;
165
166 for (hashline = 0; hashline < DIRHASH_HASHSIZE; hashline++) {
167 dirh_e = LIST_FIRST(&dirh->entries[hashline]);
168 while (dirh_e) {
169 LIST_REMOVE(dirh_e, next);
170 pool_put(&dirhash_entry_pool, dirh_e);
171 dirh_e = LIST_FIRST(&dirh->entries[hashline]);
172 }
173 }
174 dirh_e = LIST_FIRST(&dirh->free_entries);
175
176 while (dirh_e) {
177 LIST_REMOVE(dirh_e, next);
178 pool_put(&dirhash_entry_pool, dirh_e);
179 dirh_e = LIST_FIRST(&dirh->entries[hashline]);
180 }
181
182 dirh->flags &= ~DIRH_COMPLETE;
183 dirh->flags |= DIRH_PURGED;
184
185 dirhashsize -= dirh->size;
186 dirh->size = 0;
187 }
188
189
190 void
191 dirhash_purge(struct dirhash **dirhp)
192 {
193 struct dirhash *dirh = *dirhp;
194
195 if (dirh == NULL)
196 return;
197
198 mutex_enter(&dirhashmutex);
199
200 dirhash_purge_entries(dirh);
201 TAILQ_REMOVE(&dirhash_queue, dirh, next);
202 pool_put(&dirhash_pool, dirh);
203
204 *dirhp = NULL;
205
206 mutex_exit(&dirhashmutex);
207 }
208
209
210 void
211 dirhash_get(struct dirhash **dirhp)
212 {
213 struct dirhash *dirh;
214 uint32_t hashline;
215
216 mutex_enter(&dirhashmutex);
217
218 dirh = *dirhp;
219 if (*dirhp == NULL) {
220 dirh = pool_get(&dirhash_pool, PR_WAITOK);
221 *dirhp = dirh;
222 memset(dirh, 0, sizeof(struct dirhash));
223 for (hashline = 0; hashline < DIRHASH_HASHSIZE; hashline++)
224 LIST_INIT(&dirh->entries[hashline]);
225 dirh->size = 0;
226 dirh->refcnt = 0;
227 dirh->flags = 0;
228 } else {
229 TAILQ_REMOVE(&dirhash_queue, dirh, next);
230 }
231
232 dirh->refcnt++;
233 TAILQ_INSERT_HEAD(&dirhash_queue, dirh, next);
234
235 mutex_exit(&dirhashmutex);
236 }
237
238
239 void
240 dirhash_put(struct dirhash *dirh)
241 {
242
243 mutex_enter(&dirhashmutex);
244 dirh->refcnt--;
245 mutex_exit(&dirhashmutex);
246 }
247
248
249 void
250 dirhash_enter(struct dirhash *dirh,
251 struct dirent *dirent, uint64_t offset, uint32_t entry_size, int new)
252 {
253 struct dirhash *del_dirh, *prev_dirh;
254 struct dirhash_entry *dirh_e;
255 uint32_t hashvalue, hashline;
256 int entrysize;
257
258 /* make sure we have a dirhash to work on */
259 KASSERT(dirh);
260 KASSERT(dirh->refcnt > 0);
261
262 /* are we trying to re-enter an entry? */
263 if (!new && (dirh->flags & DIRH_COMPLETE))
264 return;
265
266 /* calculate our hash */
267 hashvalue = dirhash_hash(dirent->d_name, dirent->d_namlen);
268 hashline = hashvalue & DIRHASH_HASHMASK;
269
270 /* lookup and insert entry if not there yet */
271 LIST_FOREACH(dirh_e, &dirh->entries[hashline], next) {
272 /* check for hash collision */
273 if (dirh_e->hashvalue != hashvalue)
274 continue;
275 if (dirh_e->offset != offset)
276 continue;
277 /* got it already */
278 KASSERT(dirh_e->d_namlen == dirent->d_namlen);
279 KASSERT(dirh_e->entry_size == entry_size);
280 return;
281 }
282
283 DPRINTF(("dirhash enter %"PRIu64", %d, %d for `%*.*s`\n",
284 offset, entry_size, dirent->d_namlen,
285 dirent->d_namlen, dirent->d_namlen, dirent->d_name));
286
287 /* check if entry is in free space list */
288 LIST_FOREACH(dirh_e, &dirh->free_entries, next) {
289 if (dirh_e->offset == offset) {
290 DPRINTF(("\tremoving free entry\n"));
291 LIST_REMOVE(dirh_e, next);
292 break;
293 }
294 }
295
296 /* ensure we are not passing the dirhash limit */
297 entrysize = sizeof(struct dirhash_entry);
298 if (dirhashsize + entrysize > maxdirhashsize) {
299 del_dirh = TAILQ_LAST(&dirhash_queue, _dirhash);
300 KASSERT(del_dirh);
301 while (dirhashsize + entrysize > maxdirhashsize) {
302 /* no use trying to delete myself */
303 if (del_dirh == dirh)
304 break;
305 prev_dirh = TAILQ_PREV(del_dirh, _dirhash, next);
306 if (del_dirh->refcnt == 0)
307 dirhash_purge_entries(del_dirh);
308 del_dirh = prev_dirh;
309 }
310 }
311
312 /* add to the hashline */
313 dirh_e = pool_get(&dirhash_entry_pool, PR_WAITOK);
314 memset(dirh_e, 0, sizeof(struct dirhash_entry));
315
316 dirh_e->hashvalue = hashvalue;
317 dirh_e->offset = offset;
318 dirh_e->d_namlen = dirent->d_namlen;
319 dirh_e->entry_size = entry_size;
320
321 dirh->size += sizeof(struct dirhash_entry);
322 dirhashsize += sizeof(struct dirhash_entry);
323 LIST_INSERT_HEAD(&dirh->entries[hashline], dirh_e, next);
324 }
325
326
327 void
328 dirhash_enter_freed(struct dirhash *dirh, uint64_t offset,
329 uint32_t entry_size)
330 {
331 struct dirhash_entry *dirh_e;
332
333 /* make sure we have a dirhash to work on */
334 KASSERT(dirh);
335 KASSERT(dirh->refcnt > 0);
336
337 #ifdef DEBUG
338 /* check for double entry of free space */
339 LIST_FOREACH(dirh_e, &dirh->free_entries, next)
340 KASSERT(dirh_e->offset != offset);
341 #endif
342
343 DPRINTF(("dirhash enter FREED %"PRIu64", %d\n",
344 offset, entry_size));
345 dirh_e = pool_get(&dirhash_entry_pool, PR_WAITOK);
346 memset(dirh_e, 0, sizeof(struct dirhash_entry));
347
348 dirh_e->hashvalue = 0; /* not relevant */
349 dirh_e->offset = offset;
350 dirh_e->d_namlen = 0; /* not relevant */
351 dirh_e->entry_size = entry_size;
352
353 /* XXX it might be preferable to append them at the tail */
354 LIST_INSERT_HEAD(&dirh->free_entries, dirh_e, next);
355 dirh->size += sizeof(struct dirhash_entry);
356 dirhashsize += sizeof(struct dirhash_entry);
357 }
358
359
360 void
361 dirhash_remove(struct dirhash *dirh, struct dirent *dirent,
362 uint64_t offset, uint32_t entry_size)
363 {
364 struct dirhash_entry *dirh_e;
365 uint32_t hashvalue, hashline;
366
367 DPRINTF(("dirhash remove %"PRIu64", %d for `%*.*s`\n",
368 offset, entry_size,
369 dirent->d_namlen, dirent->d_namlen, dirent->d_name));
370
371 /* make sure we have a dirhash to work on */
372 KASSERT(dirh);
373 KASSERT(dirh->refcnt > 0);
374
375 /* calculate our hash */
376 hashvalue = dirhash_hash(dirent->d_name, dirent->d_namlen);
377 hashline = hashvalue & DIRHASH_HASHMASK;
378
379 /* lookup entry */
380 LIST_FOREACH(dirh_e, &dirh->entries[hashline], next) {
381 /* check for hash collision */
382 if (dirh_e->hashvalue != hashvalue)
383 continue;
384 if (dirh_e->offset != offset)
385 continue;
386
387 /* got it! */
388 KASSERT(dirh_e->d_namlen == dirent->d_namlen);
389 KASSERT(dirh_e->entry_size == entry_size);
390 LIST_REMOVE(dirh_e, next);
391 dirh->size -= sizeof(struct dirhash_entry);
392 dirhashsize -= sizeof(struct dirhash_entry);
393
394 dirhash_enter_freed(dirh, offset, entry_size);
395 return;
396 }
397
398 /* not found! */
399 panic("dirhash_remove couldn't find entry in hash table\n");
400 }
401
402
403 /* BUGALERT: don't use result longer than needed, never past the node lock */
404 /* call with NULL *result initially and it will return nonzero if again */
405 int
406 dirhash_lookup(struct dirhash *dirh, const char *d_name, int d_namlen,
407 struct dirhash_entry **result)
408 {
409 struct dirhash_entry *dirh_e;
410 uint32_t hashvalue, hashline;
411
412 /* vnode should be locked */
413 //KASSERT(VOP_ISLOCKED(dirh->vnode));
414
415 /* make sure we have a dirhash to work on */
416 KASSERT(dirh);
417 KASSERT(dirh->refcnt > 0);
418
419 /* start where we were */
420 if (*result) {
421 dirh_e = *result;
422
423 /* retrieve information to avoid recalculation and advance */
424 hashvalue = dirh_e->hashvalue;
425 dirh_e = LIST_NEXT(*result, next);
426 } else {
427 /* calculate our hash and lookup all entries in hashline */
428 hashvalue = dirhash_hash(d_name, d_namlen);
429 hashline = hashvalue & DIRHASH_HASHMASK;
430 dirh_e = LIST_FIRST(&dirh->entries[hashline]);
431 }
432
433 for (; dirh_e; dirh_e = LIST_NEXT(dirh_e, next)) {
434 /* check for hash collision */
435 if (dirh_e->hashvalue != hashvalue)
436 continue;
437 if (dirh_e->d_namlen != d_namlen)
438 continue;
439 /* might have an entry in the cache */
440 *result = dirh_e;
441 return 1;
442 }
443
444 *result = NULL;
445 return 0;
446 }
447
448
449 /* BUGALERT: don't use result longer than needed, never past the node lock */
450 /* call with NULL *result initially and it will return nonzero if again */
451 int
452 dirhash_lookup_freed(struct dirhash *dirh, uint32_t min_entrysize,
453 struct dirhash_entry **result)
454 {
455 struct dirhash_entry *dirh_e;
456
457 //KASSERT(VOP_ISLOCKED(dirh->vnode));
458
459 /* make sure we have a dirhash to work on */
460 KASSERT(dirh);
461 KASSERT(dirh->refcnt > 0);
462
463 /* start where we were */
464 if (*result) {
465 dirh_e = LIST_NEXT(*result, next);
466 } else {
467 /* lookup all entries that match */
468 dirh_e = LIST_FIRST(&dirh->free_entries);
469 }
470
471 for (; dirh_e; dirh_e = LIST_NEXT(dirh_e, next)) {
472 /* check for minimum size */
473 if (dirh_e->entry_size < min_entrysize)
474 continue;
475 /* might be a candidate */
476 *result = dirh_e;
477 return 1;
478 }
479
480 *result = NULL;
481 return 0;
482 }
483
484
485