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 /*
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 /*
27 * Copyright 2019 Nexenta Systems, Inc.
28 */
29
30 #include <sys/systm.h>
31 #include <sys/cmn_err.h>
32 #include <sys/kmem.h>
33 #include <sys/disp.h>
34 #include <sys/id_space.h>
35 #include <rpc/rpc.h>
36 #include <nfs/nfs4.h>
37 #include <nfs/nfs4_db_impl.h>
38 #include <sys/sdt.h>
39
40 static int rfs4_reap_interval = RFS4_REAP_INTERVAL;
41
42 static void rfs4_dbe_reap(rfs4_table_t *, time_t, uint32_t);
43 static void rfs4_dbe_destroy(rfs4_dbe_t *);
44 static rfs4_dbe_t *rfs4_dbe_create(rfs4_table_t *, id_t, rfs4_entry_t);
45 static void rfs4_start_reaper(rfs4_table_t *);
46
47 /*
48 * t_lowat - integer percentage of table entries /etc/system only
49 * t_hiwat - integer percentage of table entries /etc/system only
50 * t_lreap - integer percentage of table reap time mdb or /etc/system
51 * t_hreap - integer percentage of table reap time mdb or /etc/system
52 */
53 uint32_t t_lowat = 50; /* reap at t_lreap when id's in use hit 50% */
54 uint32_t t_hiwat = 75; /* reap at t_hreap when id's in use hit 75% */
55 time_t t_lreap = 50; /* default to 50% of table's reap interval */
56 time_t t_hreap = 10; /* default to 10% of table's reap interval */
57
58 id_t
59 rfs4_dbe_getid(rfs4_dbe_t *entry)
60 {
61 return (entry->dbe_id);
62 }
63
64 void
65 rfs4_dbe_hold(rfs4_dbe_t *entry)
66 {
67 if (!MUTEX_HELD(entry->dbe_lock)) {
68 mutex_enter(entry->dbe_lock);
69 entry->dbe_refcnt++;
70 mutex_exit(entry->dbe_lock);
71 } else {
72 entry->dbe_refcnt++;
73 }
74 }
75
76 /*
77 * rfs4_dbe_rele_nolock only decrements the reference count of the entry.
78 */
79 void
80 rfs4_dbe_rele_nolock(rfs4_dbe_t *entry)
81 {
82 if (!MUTEX_HELD(entry->dbe_lock)) {
83 ASSERT(entry->dbe_refcnt > 0);
84 mutex_enter(entry->dbe_lock);
85 entry->dbe_refcnt--;
86 mutex_exit(entry->dbe_lock);
87 } else {
88 entry->dbe_refcnt--;
89 }
90 }
91
92
93 uint32_t
94 rfs4_dbe_refcnt(rfs4_dbe_t *entry)
95 {
96 return (entry->dbe_refcnt);
97 }
98
99 /*
100 * Mark an entry such that the dbsearch will skip it.
101 * Caller does not want this entry to be found any longer
102 */
103 void
104 rfs4_dbe_invalidate(rfs4_dbe_t *entry)
105 {
106 if (!MUTEX_HELD(entry->dbe_lock)) {
107 mutex_enter(entry->dbe_lock);
108 entry->dbe_invalid = TRUE;
109 entry->dbe_skipsearch = TRUE;
110 mutex_exit(entry->dbe_lock);
111 } else {
112 entry->dbe_invalid = TRUE;
113 entry->dbe_skipsearch = TRUE;
114 }
115 }
116
117 /*
118 * Is this entry invalid?
119 */
120 bool_t
121 rfs4_dbe_is_invalid(rfs4_dbe_t *entry)
122 {
123 return (entry->dbe_invalid);
124 }
125
126 time_t
127 rfs4_dbe_get_timerele(rfs4_dbe_t *entry)
128 {
129 return (entry->dbe_time_rele);
130 }
131
132 /*
133 * Use these to temporarily hide/unhide a db entry.
134 */
135 void
136 rfs4_dbe_hide(rfs4_dbe_t *entry)
137 {
138 rfs4_dbe_lock(entry);
139 entry->dbe_skipsearch = TRUE;
140 rfs4_dbe_unlock(entry);
141 }
142
143 void
144 rfs4_dbe_unhide(rfs4_dbe_t *entry)
145 {
146 rfs4_dbe_lock(entry);
147 entry->dbe_skipsearch = FALSE;
148 rfs4_dbe_unlock(entry);
149 }
150
151 void
152 rfs4_dbe_rele(rfs4_dbe_t *entry)
153 {
154 mutex_enter(entry->dbe_lock);
155 ASSERT(entry->dbe_refcnt > 1);
156 entry->dbe_refcnt--;
157 entry->dbe_time_rele = gethrestime_sec();
158 mutex_exit(entry->dbe_lock);
159 }
160
161 void
162 rfs4_dbe_lock(rfs4_dbe_t *entry)
163 {
164 mutex_enter(entry->dbe_lock);
165 }
166
167 void
168 rfs4_dbe_unlock(rfs4_dbe_t *entry)
169 {
170 mutex_exit(entry->dbe_lock);
171 }
172
173 bool_t
174 rfs4_dbe_islocked(rfs4_dbe_t *entry)
175 {
176 return (mutex_owned(entry->dbe_lock));
177 }
178
179 clock_t
180 rfs4_dbe_twait(rfs4_dbe_t *entry, clock_t timeout)
181 {
182 return (cv_timedwait(entry->dbe_cv, entry->dbe_lock, timeout));
183 }
184
185 void
186 rfs4_dbe_cv_broadcast(rfs4_dbe_t *entry)
187 {
188 cv_broadcast(entry->dbe_cv);
189 }
190
191 /* ARGSUSED */
192 static int
193 rfs4_dbe_kmem_constructor(void *obj, void *private, int kmflag)
194 {
195 rfs4_dbe_t *entry = obj;
196
197 mutex_init(entry->dbe_lock, NULL, MUTEX_DEFAULT, NULL);
198 cv_init(entry->dbe_cv, NULL, CV_DEFAULT, NULL);
199
200 return (0);
201 }
202
203 static void
204 rfs4_dbe_kmem_destructor(void *obj, void *private)
205 {
206 rfs4_dbe_t *entry = obj;
207 /*LINTED*/
208 rfs4_table_t *table = private;
209
210 mutex_destroy(entry->dbe_lock);
211 cv_destroy(entry->dbe_cv);
212 }
213
214 rfs4_database_t *
215 rfs4_database_create(uint32_t flags)
216 {
217 rfs4_database_t *db;
218
219 db = kmem_alloc(sizeof (rfs4_database_t), KM_SLEEP);
220 mutex_init(db->db_lock, NULL, MUTEX_DEFAULT, NULL);
221 db->db_tables = NULL;
222 db->db_debug_flags = flags;
223 db->db_shutdown_count = 0;
224 cv_init(&db->db_shutdown_wait, NULL, CV_DEFAULT, NULL);
225 return (db);
226 }
227
228
229 /*
230 * The reaper threads that have been created for the tables in this
231 * database must be stopped and the entries in the tables released.
232 * Each table will be marked as "shutdown" and the reaper threads
233 * poked and they will see that a shutdown is in progress and cleanup
234 * and exit. This function waits for all reaper threads to stop
235 * before returning to the caller.
236 */
237 void
238 rfs4_database_shutdown(rfs4_database_t *db)
239 {
240 rfs4_table_t *table;
241
242 mutex_enter(db->db_lock);
243 for (table = db->db_tables; table; table = table->dbt_tnext) {
244 mutex_enter(&table->dbt_reaper_cv_lock);
245 table->dbt_reaper_shutdown = TRUE;
246 cv_broadcast(&table->dbt_reaper_wait);
247 db->db_shutdown_count++;
248 mutex_exit(&table->dbt_reaper_cv_lock);
249 }
250 while (db->db_shutdown_count > 0) {
251 cv_wait(&db->db_shutdown_wait, db->db_lock);
252 }
253 mutex_exit(db->db_lock);
254 }
255
256 /*
257 * Given a database that has been "shutdown" by the function above all
258 * of the table tables are destroyed and then the database itself
259 * freed.
260 */
261 void
262 rfs4_database_destroy(rfs4_database_t *db)
263 {
264 rfs4_table_t *next, *tmp;
265
266 for (next = db->db_tables; next; ) {
267 tmp = next;
268 next = tmp->dbt_tnext;
269 rfs4_table_destroy(db, tmp);
270 }
271
272 mutex_destroy(db->db_lock);
273 kmem_free(db, sizeof (rfs4_database_t));
274 }
275
276 /*
277 * Used to get the correct kmem_cache database for the state table being
278 * created.
279 * Helper function for rfs4_table_create
280 */
281 static kmem_cache_t *
282 get_db_mem_cache(char *name)
283 {
284 int i;
285
286 for (i = 0; i < RFS4_DB_MEM_CACHE_NUM; i++) {
287 if (strcmp(name, rfs4_db_mem_cache_table[i].r_db_name) == 0)
288 return (rfs4_db_mem_cache_table[i].r_db_mem_cache);
289 }
290 /*
291 * There is no associated kmem cache for this NFS4 server state
292 * table name
293 */
294 return (NULL);
295 }
296
297 /*
298 * Used to initialize the global NFSv4 server state database.
299 * Helper funtion for rfs4_state_g_init and called when module is loaded.
300 */
301 kmem_cache_t *
302 /* CSTYLED */
303 nfs4_init_mem_cache(char *cache_name, uint32_t idxcnt, uint32_t size, uint32_t idx)
304 {
305 kmem_cache_t *mem_cache = kmem_cache_create(cache_name,
306 sizeof (rfs4_dbe_t) + idxcnt * sizeof (rfs4_link_t) + size,
307 0,
308 rfs4_dbe_kmem_constructor,
309 rfs4_dbe_kmem_destructor,
310 NULL,
311 NULL,
312 NULL,
313 0);
314 (void) strlcpy(rfs4_db_mem_cache_table[idx].r_db_name, cache_name,
315 strlen(cache_name) + 1);
316 rfs4_db_mem_cache_table[idx].r_db_mem_cache = mem_cache;
317 return (mem_cache);
318 }
319
320 rfs4_table_t *
321 rfs4_table_create(rfs4_database_t *db, char *tabname, time_t max_cache_time,
322 uint32_t idxcnt, bool_t (*create)(rfs4_entry_t, void *),
323 void (*destroy)(rfs4_entry_t),
324 bool_t (*expiry)(rfs4_entry_t),
325 uint32_t size, uint32_t hashsize,
326 uint32_t maxentries, id_t start)
327 {
328 rfs4_table_t *table;
329 int len;
330 char *cache_name;
331 char *id_name;
332
333 table = kmem_alloc(sizeof (rfs4_table_t), KM_SLEEP);
334 table->dbt_db = db;
335 rw_init(table->dbt_t_lock, NULL, RW_DEFAULT, NULL);
336 mutex_init(table->dbt_lock, NULL, MUTEX_DEFAULT, NULL);
337 mutex_init(&table->dbt_reaper_cv_lock, NULL, MUTEX_DEFAULT, NULL);
338 cv_init(&table->dbt_reaper_wait, NULL, CV_DEFAULT, NULL);
339
340 len = strlen(tabname);
341 table->dbt_name = kmem_alloc(len+1, KM_SLEEP);
342 cache_name = kmem_alloc(len + 12 /* "_entry_cache" */ + 1, KM_SLEEP);
343 (void) strcpy(table->dbt_name, tabname);
344 (void) sprintf(cache_name, "%s_entry_cache", table->dbt_name);
345 table->dbt_max_cache_time = max_cache_time;
346 table->dbt_usize = size;
347 table->dbt_len = hashsize;
348 table->dbt_count = 0;
349 table->dbt_idxcnt = 0;
350 table->dbt_ccnt = 0;
351 table->dbt_maxcnt = idxcnt;
352 table->dbt_indices = NULL;
353 table->dbt_id_space = NULL;
354 table->dbt_reaper_shutdown = FALSE;
355
356 if (start >= 0) {
357 if (maxentries + (uint32_t)start > (uint32_t)INT32_MAX)
358 maxentries = INT32_MAX - start;
359 id_name = kmem_alloc(len + 9 /* "_id_space" */ + 1, KM_SLEEP);
360 (void) sprintf(id_name, "%s_id_space", table->dbt_name);
361 table->dbt_id_space = id_space_create(id_name, start,
362 maxentries + start);
363 kmem_free(id_name, len + 10);
364 }
365 ASSERT(t_lowat != 0);
366 table->dbt_id_lwat = (maxentries * t_lowat) / 100;
367 ASSERT(t_hiwat != 0);
368 table->dbt_id_hwat = (maxentries * t_hiwat) / 100;
369 table->dbt_id_reap = MIN(rfs4_reap_interval, max_cache_time);
370 table->dbt_maxentries = maxentries;
371 table->dbt_create = create;
372 table->dbt_destroy = destroy;
373 table->dbt_expiry = expiry;
374
375 /*
376 * get the correct kmem_cache for this table type based on the name.
377 */
378 table->dbt_mem_cache = get_db_mem_cache(cache_name);
379
380 kmem_free(cache_name, len+13);
381
382 table->dbt_debug = db->db_debug_flags;
383
384 mutex_enter(db->db_lock);
385 table->dbt_tnext = db->db_tables;
386 db->db_tables = table;
387 mutex_exit(db->db_lock);
388
389 rfs4_start_reaper(table);
390
391 return (table);
392 }
393
394 void
395 rfs4_table_destroy(rfs4_database_t *db, rfs4_table_t *table)
396 {
397 rfs4_table_t *p;
398 rfs4_index_t *idx;
399
400 ASSERT(table->dbt_count == 0);
401
402 mutex_enter(db->db_lock);
403 if (table == db->db_tables)
404 db->db_tables = table->dbt_tnext;
405 else {
406 for (p = db->db_tables; p; p = p->dbt_tnext)
407 if (p->dbt_tnext == table) {
408 p->dbt_tnext = table->dbt_tnext;
409 table->dbt_tnext = NULL;
410 break;
411 }
412 ASSERT(p != NULL);
413 }
414 mutex_exit(db->db_lock);
415
416 /* Destroy indices */
417 while (table->dbt_indices) {
418 idx = table->dbt_indices;
419 table->dbt_indices = idx->dbi_inext;
420 rfs4_index_destroy(idx);
421 }
422
423 rw_destroy(table->dbt_t_lock);
424 mutex_destroy(table->dbt_lock);
425 mutex_destroy(&table->dbt_reaper_cv_lock);
426 cv_destroy(&table->dbt_reaper_wait);
427
428 kmem_free(table->dbt_name, strlen(table->dbt_name) + 1);
429 if (table->dbt_id_space)
430 id_space_destroy(table->dbt_id_space);
431 table->dbt_mem_cache = NULL;
432 kmem_free(table, sizeof (rfs4_table_t));
433 }
434
435 rfs4_index_t *
436 rfs4_index_create(rfs4_table_t *table, char *keyname,
437 uint32_t (*hash)(void *),
438 bool_t (compare)(rfs4_entry_t, void *),
439 void *(*mkkey)(rfs4_entry_t),
440 bool_t createable)
441 {
442 rfs4_index_t *idx;
443
444 ASSERT(table->dbt_idxcnt < table->dbt_maxcnt);
445
446 idx = kmem_alloc(sizeof (rfs4_index_t), KM_SLEEP);
447
448 idx->dbi_table = table;
449 idx->dbi_keyname = kmem_alloc(strlen(keyname) + 1, KM_SLEEP);
450 (void) strcpy(idx->dbi_keyname, keyname);
451 idx->dbi_hash = hash;
452 idx->dbi_compare = compare;
453 idx->dbi_mkkey = mkkey;
454 idx->dbi_tblidx = table->dbt_idxcnt;
455 table->dbt_idxcnt++;
456 if (createable) {
457 table->dbt_ccnt++;
458 if (table->dbt_ccnt > 1)
459 panic("Table %s currently can have only have one "
460 "index that will allow creation of entries",
461 table->dbt_name);
462 idx->dbi_createable = TRUE;
463 } else {
464 idx->dbi_createable = FALSE;
465 }
466
467 idx->dbi_inext = table->dbt_indices;
468 table->dbt_indices = idx;
469 idx->dbi_buckets = kmem_zalloc(sizeof (rfs4_bucket_t) * table->dbt_len,
470 KM_SLEEP);
471
472 return (idx);
473 }
474
475 void
476 rfs4_index_destroy(rfs4_index_t *idx)
477 {
478 kmem_free(idx->dbi_keyname, strlen(idx->dbi_keyname) + 1);
479 kmem_free(idx->dbi_buckets,
480 sizeof (rfs4_bucket_t) * idx->dbi_table->dbt_len);
481 kmem_free(idx, sizeof (rfs4_index_t));
482 }
483
484 static void
485 rfs4_dbe_destroy(rfs4_dbe_t *entry)
486 {
487 rfs4_index_t *idx;
488 void *key;
489 int i;
490 rfs4_bucket_t *bp;
491 rfs4_table_t *table = entry->dbe_table;
492 rfs4_link_t *l;
493
494 NFS4_DEBUG(table->dbt_debug & DESTROY_DEBUG,
495 (CE_NOTE, "Destroying entry %p from %s",
496 (void*)entry, table->dbt_name));
497
498 mutex_enter(entry->dbe_lock);
499 ASSERT(entry->dbe_refcnt == 0);
500 mutex_exit(entry->dbe_lock);
501
502 /* Unlink from all indices */
503 for (idx = table->dbt_indices; idx; idx = idx->dbi_inext) {
504 l = &entry->dbe_indices[idx->dbi_tblidx];
505 /* check and see if we were ever linked in to the index */
506 if (INVALID_LINK(l)) {
507 ASSERT(l->next == NULL && l->prev == NULL);
508 continue;
509 }
510 key = idx->dbi_mkkey(entry->dbe_data);
511 i = HASH(idx, key);
512 bp = &idx->dbi_buckets[i];
513 ASSERT(bp->dbk_head != NULL);
514 DEQUEUE_IDX(bp, &entry->dbe_indices[idx->dbi_tblidx]);
515 }
516
517 /* Destroy user data */
518 if (table->dbt_destroy)
519 (*table->dbt_destroy)(entry->dbe_data);
520
521 if (table->dbt_id_space)
522 id_free(table->dbt_id_space, entry->dbe_id);
523
524 mutex_enter(table->dbt_lock);
525 table->dbt_count--;
526 mutex_exit(table->dbt_lock);
527
528 /* Destroy the entry itself */
529 kmem_cache_free(table->dbt_mem_cache, entry);
530 }
531
532
533 static rfs4_dbe_t *
534 rfs4_dbe_create(rfs4_table_t *table, id_t id, rfs4_entry_t data)
535 {
536 rfs4_dbe_t *entry;
537 int i;
538
539 NFS4_DEBUG(table->dbt_debug & CREATE_DEBUG,
540 (CE_NOTE, "Creating entry in table %s", table->dbt_name));
541
542 entry = kmem_cache_alloc(table->dbt_mem_cache, KM_SLEEP);
543
544 entry->dbe_refcnt = 1;
545 entry->dbe_invalid = FALSE;
546 entry->dbe_skipsearch = FALSE;
547 entry->dbe_time_rele = 0;
548 entry->dbe_id = 0;
549
550 if (table->dbt_id_space)
551 entry->dbe_id = id;
552 entry->dbe_table = table;
553
554 for (i = 0; i < table->dbt_maxcnt; i++) {
555 entry->dbe_indices[i].next = entry->dbe_indices[i].prev = NULL;
556 entry->dbe_indices[i].entry = entry;
557 /*
558 * We mark the entry as not indexed by setting the low
559 * order bit, since address are word aligned. This has
560 * the advantage of causeing a trap if the address is
561 * used. After the entry is linked in to the
562 * corresponding index the bit will be cleared.
563 */
564 INVALIDATE_ADDR(entry->dbe_indices[i].entry);
565 }
566
567 entry->dbe_data = (rfs4_entry_t)&entry->dbe_indices[table->dbt_maxcnt];
568 bzero(entry->dbe_data, table->dbt_usize);
569 entry->dbe_data->dbe = entry;
570
571 if (!(*table->dbt_create)(entry->dbe_data, data)) {
572 kmem_cache_free(table->dbt_mem_cache, entry);
573 return (NULL);
574 }
575
576 mutex_enter(table->dbt_lock);
577 table->dbt_count++;
578 mutex_exit(table->dbt_lock);
579
580 return (entry);
581 }
582
583 static void
584 rfs4_dbe_tabreap_adjust(rfs4_table_t *table)
585 {
586 clock_t tabreap;
587 clock_t reap_int;
588 uint32_t in_use;
589
590 /*
591 * Adjust the table's reap interval based on the
592 * number of id's currently in use. Each table's
593 * default remains the same if id usage subsides.
594 */
595 ASSERT(MUTEX_HELD(&table->dbt_reaper_cv_lock));
596 tabreap = MIN(rfs4_reap_interval, table->dbt_max_cache_time);
597
598 in_use = table->dbt_count + 1; /* see rfs4_dbe_create */
599 if (in_use >= table->dbt_id_hwat) {
600 ASSERT(t_hreap != 0);
601 reap_int = (tabreap * t_hreap) / 100;
602 } else if (in_use >= table->dbt_id_lwat) {
603 ASSERT(t_lreap != 0);
604 reap_int = (tabreap * t_lreap) / 100;
605 } else {
606 reap_int = tabreap;
607 }
608 table->dbt_id_reap = reap_int;
609 DTRACE_PROBE2(table__reap__interval, char *,
610 table->dbt_name, time_t, table->dbt_id_reap);
611 }
612
613 rfs4_entry_t
614 rfs4_dbsearch(rfs4_index_t *idx, void *key, bool_t *create, void *arg,
615 rfs4_dbsearch_type_t dbsearch_type)
616 {
617 int already_done;
618 uint32_t i;
619 rfs4_table_t *table = idx->dbi_table;
620 rfs4_index_t *ip;
621 rfs4_bucket_t *bp;
622 rfs4_link_t *l;
623 rfs4_dbe_t *entry;
624 id_t id = -1;
625
626 i = HASH(idx, key);
627 bp = &idx->dbi_buckets[i];
628
629 NFS4_DEBUG(table->dbt_debug & SEARCH_DEBUG,
630 (CE_NOTE, "Searching for key %p in table %s by %s",
631 key, table->dbt_name, idx->dbi_keyname));
632
633 rw_enter(bp->dbk_lock, RW_READER);
634 retry:
635 for (l = bp->dbk_head; l; l = l->next) {
636 if (l->entry->dbe_refcnt > 0 &&
637 (l->entry->dbe_skipsearch == FALSE ||
638 (l->entry->dbe_skipsearch == TRUE &&
639 dbsearch_type == RFS4_DBS_INVALID)) &&
640 (*idx->dbi_compare)(l->entry->dbe_data, key)) {
641 mutex_enter(l->entry->dbe_lock);
642 if (l->entry->dbe_refcnt == 0) {
643 mutex_exit(l->entry->dbe_lock);
644 continue;
645 }
646
647 /* place an additional hold since we are returning */
648 rfs4_dbe_hold(l->entry);
649
650 mutex_exit(l->entry->dbe_lock);
651 rw_exit(bp->dbk_lock);
652
653 *create = FALSE;
654
655 NFS4_DEBUG((table->dbt_debug & SEARCH_DEBUG),
656 (CE_NOTE, "Found entry %p for %p in table %s",
657 (void *)l->entry, key, table->dbt_name));
658
659 if (id != -1)
660 id_free(table->dbt_id_space, id);
661 return (l->entry->dbe_data);
662 }
663 }
664
665 if (!*create || table->dbt_create == NULL || !idx->dbi_createable ||
666 table->dbt_maxentries == table->dbt_count) {
667 NFS4_DEBUG(table->dbt_debug & SEARCH_DEBUG,
668 (CE_NOTE, "Entry for %p in %s not found",
669 key, table->dbt_name));
670
671 rw_exit(bp->dbk_lock);
672 if (id != -1)
673 id_free(table->dbt_id_space, id);
674 return (NULL);
675 }
676
677 if (table->dbt_id_space && id == -1) {
678 rw_exit(bp->dbk_lock);
679
680 /* get an id, ok to sleep for it here */
681 id = id_alloc(table->dbt_id_space);
682 ASSERT(id != -1);
683
684 mutex_enter(&table->dbt_reaper_cv_lock);
685 rfs4_dbe_tabreap_adjust(table);
686 mutex_exit(&table->dbt_reaper_cv_lock);
687
688 rw_enter(bp->dbk_lock, RW_WRITER);
689 goto retry;
690 }
691
692 /* get an exclusive lock on the bucket */
693 if (rw_read_locked(bp->dbk_lock) && !rw_tryupgrade(bp->dbk_lock)) {
694 NFS4_DEBUG(table->dbt_debug & OTHER_DEBUG,
695 (CE_NOTE, "Trying to upgrade lock on "
696 "hash chain %d (%p) for %s by %s",
697 i, (void*)bp, table->dbt_name, idx->dbi_keyname));
698
699 rw_exit(bp->dbk_lock);
700 rw_enter(bp->dbk_lock, RW_WRITER);
701 goto retry;
702 }
703
704 /* create entry */
705 entry = rfs4_dbe_create(table, id, arg);
706 if (entry == NULL) {
707 rw_exit(bp->dbk_lock);
708 if (id != -1)
709 id_free(table->dbt_id_space, id);
710
711 NFS4_DEBUG(table->dbt_debug & CREATE_DEBUG,
712 (CE_NOTE, "Constructor for table %s failed",
713 table->dbt_name));
714 return (NULL);
715 }
716
717 /*
718 * Add one ref for entry into table's hash - only one
719 * reference added even though there may be multiple indices
720 */
721 rfs4_dbe_hold(entry);
722 ENQUEUE(bp->dbk_head, &entry->dbe_indices[idx->dbi_tblidx]);
723 VALIDATE_ADDR(entry->dbe_indices[idx->dbi_tblidx].entry);
724
725 already_done = idx->dbi_tblidx;
726 rw_exit(bp->dbk_lock);
727
728 for (ip = table->dbt_indices; ip; ip = ip->dbi_inext) {
729 if (ip->dbi_tblidx == already_done)
730 continue;
731 l = &entry->dbe_indices[ip->dbi_tblidx];
732 i = HASH(ip, ip->dbi_mkkey(entry->dbe_data));
733 ASSERT(i < ip->dbi_table->dbt_len);
734 bp = &ip->dbi_buckets[i];
735 ENQUEUE_IDX(bp, l);
736 }
737
738 NFS4_DEBUG(
739 table->dbt_debug & SEARCH_DEBUG || table->dbt_debug & CREATE_DEBUG,
740 (CE_NOTE, "Entry %p created for %s = %p in table %s",
741 (void*)entry, idx->dbi_keyname, (void*)key, table->dbt_name));
742
743 return (entry->dbe_data);
744 }
745
746 /*ARGSUSED*/
747 boolean_t
748 rfs4_cpr_callb(void *arg, int code)
749 {
750 rfs4_bucket_t *buckets, *bp;
751 rfs4_link_t *l;
752 rfs4_client_t *cp;
753 int i;
754
755 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
756 rfs4_table_t *table = nsrv4->rfs4_client_tab;
757
758 /*
759 * We get called for Suspend and Resume events.
760 * For the suspend case we simply don't care! Nor do we care if
761 * there are no clients.
762 */
763 if (code == CB_CODE_CPR_CHKPT || table == NULL) {
764 return (B_TRUE);
765 }
766
767 buckets = table->dbt_indices->dbi_buckets;
768
769 /*
770 * When we get this far we are in the process of
771 * resuming the system from a previous suspend.
772 *
773 * We are going to blast through and update the
774 * last_access time for all the clients and in
775 * doing so extend them by one lease period.
776 */
777 for (i = 0; i < table->dbt_len; i++) {
778 bp = &buckets[i];
779 for (l = bp->dbk_head; l; l = l->next) {
780 cp = (rfs4_client_t *)l->entry->dbe_data;
781 cp->rc_last_access = gethrestime_sec();
782 }
783 }
784
785 return (B_TRUE);
786 }
787
788 /*
789 * Given a table, lock each of the buckets and walk all entries (in
790 * turn locking those) and calling the provided "callout" function
791 * with the provided parameter. Obviously used to iterate across all
792 * entries in a particular table via the database locking hierarchy.
793 * Obviously the caller must not hold locks on any of the entries in
794 * the specified table.
795 */
796 void
797 rfs4_dbe_walk(rfs4_table_t *table,
798 void (*callout)(rfs4_entry_t, void *),
799 void *data)
800 {
801 rfs4_bucket_t *buckets = table->dbt_indices->dbi_buckets, *bp;
802 rfs4_link_t *l;
803 rfs4_dbe_t *entry;
804 int i;
805
806 NFS4_DEBUG(table->dbt_debug & WALK_DEBUG,
807 (CE_NOTE, "Walking entries in %s", table->dbt_name));
808
809 /* Walk the buckets looking for entries to release/destroy */
810 for (i = 0; i < table->dbt_len; i++) {
811 bp = &buckets[i];
812 rw_enter(bp->dbk_lock, RW_READER);
813 for (l = bp->dbk_head; l; l = l->next) {
814 entry = l->entry;
815 mutex_enter(entry->dbe_lock);
816 (*callout)(entry->dbe_data, data);
817 mutex_exit(entry->dbe_lock);
818 }
819 rw_exit(bp->dbk_lock);
820 }
821
822 NFS4_DEBUG(table->dbt_debug & WALK_DEBUG,
823 (CE_NOTE, "Walking entries complete %s", table->dbt_name));
824 }
825
826
827 static void
828 rfs4_dbe_reap(rfs4_table_t *table, time_t cache_time, uint32_t desired)
829 {
830 rfs4_index_t *idx = table->dbt_indices;
831 rfs4_bucket_t *buckets = idx->dbi_buckets, *bp;
832 rfs4_link_t *l, *t;
833 rfs4_dbe_t *entry;
834 bool_t found;
835 int i;
836 int count = 0;
837
838 NFS4_DEBUG(table->dbt_debug & REAP_DEBUG,
839 (CE_NOTE, "Reaping %d entries older than %ld seconds in table %s",
840 desired, cache_time, table->dbt_name));
841
842 /* Walk the buckets looking for entries to release/destroy */
843 for (i = 0; i < table->dbt_len; i++) {
844 int retries = 0;
845 bp = &buckets[i];
846 do {
847 found = FALSE;
848 rw_enter(bp->dbk_lock, RW_READER);
849 for (l = bp->dbk_head; l; l = l->next) {
850 entry = l->entry;
851 mutex_enter(entry->dbe_lock);
852 ASSERT(entry->dbe_refcnt != 0);
853 /*
854 * Examine an entry. Ref count of 1 means
855 * that the only reference is for the hash
856 * table reference.
857 */
858 if (entry->dbe_refcnt != 1) {
859 #ifdef DEBUG
860 rfs4_dbe_debug(entry);
861 #endif
862 mutex_exit(entry->dbe_lock);
863 continue;
864 }
865 if ((entry->dbe_refcnt == 1) &&
866 (table->dbt_reaper_shutdown ||
867 table->dbt_expiry == NULL ||
868 (*table->dbt_expiry)(entry->dbe_data))) {
869 rfs4_dbe_rele_nolock(entry);
870 count++;
871 found = TRUE;
872 }
873 mutex_exit(entry->dbe_lock);
874 }
875 if (found) {
876 if (!rw_tryupgrade(bp->dbk_lock)) {
877 rw_exit(bp->dbk_lock);
878 rw_enter(bp->dbk_lock, RW_WRITER);
879 }
880
881 l = bp->dbk_head;
882 while (l) {
883 t = l;
884 entry = t->entry;
885 l = l->next;
886 mutex_enter(entry->dbe_lock);
887 if (entry->dbe_refcnt == 0) {
888 DEQUEUE(bp->dbk_head, t);
889 mutex_exit(entry->dbe_lock);
890 t->next = NULL;
891 t->prev = NULL;
892 INVALIDATE_ADDR(t->entry);
893 rfs4_dbe_destroy(entry);
894 } else
895 mutex_exit(entry->dbe_lock);
896 }
897 }
898 rw_exit(bp->dbk_lock);
899 /*
900 * delay slightly if there is more work to do
901 * with the expectation that other reaper
902 * threads are freeing data structures as well
903 * and in turn will reduce ref counts on
904 * entries in this table allowing them to be
905 * released. This is only done in the
906 * instance that the tables are being shut down.
907 */
908 if (table->dbt_reaper_shutdown && bp->dbk_head != NULL) {
909 delay(hz/100);
910 retries++;
911 }
912 /*
913 * If this is a table shutdown, keep going until
914 * everything is gone
915 */
916 } while (table->dbt_reaper_shutdown && bp->dbk_head != NULL && retries < 5);
917
918 if (!table->dbt_reaper_shutdown && desired && count >= desired)
919 break;
920 }
921
922 NFS4_DEBUG(table->dbt_debug & REAP_DEBUG,
923 (CE_NOTE, "Reaped %d entries older than %ld seconds in table %s",
924 count, cache_time, table->dbt_name));
925 }
926
927 static void
928 reaper_thread(caddr_t *arg)
929 {
930 rfs4_table_t *table = (rfs4_table_t *)arg;
931 clock_t rc;
932
933 NFS4_DEBUG(table->dbt_debug,
934 (CE_NOTE, "rfs4_reaper_thread starting for %s", table->dbt_name));
935
936 CALLB_CPR_INIT(&table->dbt_reaper_cpr_info, &table->dbt_reaper_cv_lock,
937 callb_generic_cpr, "nfsv4Reaper");
938
939 mutex_enter(&table->dbt_reaper_cv_lock);
940 do {
941 CALLB_CPR_SAFE_BEGIN(&table->dbt_reaper_cpr_info);
942 rc = cv_reltimedwait_sig(&table->dbt_reaper_wait,
943 &table->dbt_reaper_cv_lock,
944 SEC_TO_TICK(table->dbt_id_reap), TR_CLOCK_TICK);
945 CALLB_CPR_SAFE_END(&table->dbt_reaper_cpr_info,
946 &table->dbt_reaper_cv_lock);
947 rfs4_dbe_reap(table, table->dbt_max_cache_time, 0);
948 } while (rc != 0 && table->dbt_reaper_shutdown == FALSE);
949
950 CALLB_CPR_EXIT(&table->dbt_reaper_cpr_info);
951
952 NFS4_DEBUG(table->dbt_debug,
953 (CE_NOTE, "rfs4_reaper_thread exiting for %s", table->dbt_name));
954
955 /* Notify the database shutdown processing that the table is shutdown */
956 mutex_enter(table->dbt_db->db_lock);
957 table->dbt_db->db_shutdown_count--;
958 cv_signal(&table->dbt_db->db_shutdown_wait);
959 mutex_exit(table->dbt_db->db_lock);
960 zthread_exit();
961 }
962
963 static void
964 rfs4_start_reaper(rfs4_table_t *table)
965 {
966 if (table->dbt_max_cache_time == 0)
967 return;
968
969 (void) zthread_create(NULL, 0, reaper_thread, table, 0,
970 minclsyspri);
971 }
972
973 #ifdef DEBUG
974 void
975 rfs4_dbe_debug(rfs4_dbe_t *entry)
976 {
977 cmn_err(CE_NOTE, "Entry %p from table %s",
978 (void *)entry, entry->dbe_table->dbt_name);
979 cmn_err(CE_CONT, "\trefcnt = %d id = %d",
980 entry->dbe_refcnt, entry->dbe_id);
981 }
982 #endif