1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2004 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 # define EXPORT_SYMTAB
25 #define DEBUG_SUBSYSTEM S_LOV
26 #include <linux/slab.h>
27 #include <linux/module.h>
28 #include <linux/init.h>
29 #include <linux/slab.h>
31 #include <linux/obd_support.h>
32 #include <linux/lustre_lib.h>
33 #include <linux/lustre_idl.h>
34 #include <linux/obd_class.h>
35 #include <linux/lustre_ucache.h>
38 * current ucache implementation is simply took from group hash code, almost
39 * without any change. it's very simple and have very limited functionality,
40 * and probably it's also only suitable for usage of group hash.
43 void upcall_cache_init_entry(struct upcall_cache *cache,
44 struct upcall_cache_entry *entry,
47 UC_CACHE_SET_NEW(entry);
48 INIT_LIST_HEAD(&entry->ue_hash);
49 atomic_set(&entry->ue_refcount, 0);
51 entry->ue_cache = cache;
52 init_waitqueue_head(&entry->ue_waitq);
54 EXPORT_SYMBOL(upcall_cache_init_entry);
56 static inline struct upcall_cache_entry *
57 alloc_entry(struct upcall_cache *cache, __u64 key)
59 LASSERT(cache->alloc_entry);
60 return cache->alloc_entry(cache, key);
63 static void free_entry(struct upcall_cache_entry *entry)
65 struct upcall_cache *cache = entry->ue_cache;
68 LASSERT(cache->free_entry);
69 LASSERT(atomic_read(&entry->ue_refcount) == 0);
71 CDEBUG(D_SEC, "destroy %s entry %p for key "LPU64"\n",
72 cache->uc_name, entry, entry->ue_key);
74 list_del(&entry->ue_hash);
75 cache->free_entry(cache, entry);
78 static inline void get_entry(struct upcall_cache_entry *entry)
80 atomic_inc(&entry->ue_refcount);
83 static inline void put_entry(struct upcall_cache_entry *entry)
85 if (atomic_dec_and_test(&entry->ue_refcount) &&
86 !UC_CACHE_IS_VALID(entry)) {
91 static inline int refresh_entry(struct upcall_cache_entry *entry)
93 struct upcall_cache *cache = entry->ue_cache;
96 LASSERT(cache->make_upcall);
98 return cache->make_upcall(cache, entry);
101 static int check_unlink_entry(struct upcall_cache_entry *entry)
103 /* upcall will be issued upon new entries immediately
104 * after they are created
106 LASSERT(!UC_CACHE_IS_NEW(entry));
108 if (UC_CACHE_IS_VALID(entry) &&
109 time_before(get_seconds(), entry->ue_expire))
112 if (UC_CACHE_IS_ACQUIRING(entry)) {
113 if (time_before(get_seconds(), entry->ue_acquire_expire))
116 UC_CACHE_SET_EXPIRED(entry);
117 wake_up_all(&entry->ue_waitq);
119 } else if (!UC_CACHE_IS_INVALID(entry)) {
120 UC_CACHE_SET_EXPIRED(entry);
123 list_del_init(&entry->ue_hash);
124 if (!atomic_read(&entry->ue_refcount))
130 * currently always use write_lock
132 static struct upcall_cache_entry *
133 __get_entry(struct upcall_cache *cache, unsigned int hash, __u64 key,
134 int create, int async)
136 struct list_head *head;
137 struct upcall_cache_entry *entry, *next, *new = NULL;
141 LASSERT(hash < cache->uc_hashsize);
143 head = &cache->uc_hashtable[hash];
146 write_lock(&cache->uc_hashlock);
147 list_for_each_entry_safe(entry, next, head, ue_hash) {
148 if (check_unlink_entry(entry))
150 if (entry->ue_key == key) {
160 write_unlock(&cache->uc_hashlock);
161 new = alloc_entry(cache, key);
163 CERROR("fail to alloc entry\n");
168 list_add(&new->ue_hash, head);
176 list_move(&entry->ue_hash, head);
180 /* as for this moment, we have found matched entry
181 * and hold a ref of it. if it's NEW (we created it),
182 * we must give it a push to refresh
184 if (UC_CACHE_IS_NEW(entry)) {
185 LASSERT(entry == new);
186 UC_CACHE_SET_ACQUIRING(entry);
187 UC_CACHE_CLEAR_NEW(entry);
188 entry->ue_acquire_expire = get_seconds() +
189 cache->uc_acquire_expire;
191 write_unlock(&cache->uc_hashlock);
192 rc = refresh_entry(entry);
193 write_lock(&cache->uc_hashlock);
195 UC_CACHE_CLEAR_ACQUIRING(entry);
196 UC_CACHE_SET_INVALID(entry);
200 /* caller don't want to wait */
202 write_unlock(&cache->uc_hashlock);
206 /* someone (and only one) is doing upcall upon
207 * this item, just wait it complete
209 if (UC_CACHE_IS_ACQUIRING(entry)) {
212 init_waitqueue_entry(&wait, current);
213 add_wait_queue(&entry->ue_waitq, &wait);
214 set_current_state(TASK_INTERRUPTIBLE);
215 write_unlock(&cache->uc_hashlock);
217 schedule_timeout(cache->uc_acquire_expire * HZ);
219 write_lock(&cache->uc_hashlock);
220 remove_wait_queue(&entry->ue_waitq, &wait);
221 if (UC_CACHE_IS_ACQUIRING(entry)) {
222 /* we're interrupted or upcall failed
225 CERROR("entry %p not refreshed: cur %lu, key "LPU64", "
226 "ref %d fl %u, ac %ld, ex %ld\n",
227 entry, get_seconds(), entry->ue_key,
228 atomic_read(&entry->ue_refcount),
229 entry->ue_flags, entry->ue_acquire_expire,
232 write_unlock(&cache->uc_hashlock);
233 CERROR("Interrupted? Or check whether %s is in place\n",
240 /* invalid means error, don't need to try again */
241 if (UC_CACHE_IS_INVALID(entry)) {
243 write_unlock(&cache->uc_hashlock);
248 * We can't refresh the existed one because some
249 * memory might be shared by multiple processes.
251 if (check_unlink_entry(entry)) {
252 /* if expired, try again. but if this entry is
253 * created by me and too quickly turn to expired
254 * without any error, should at least give a
255 * chance to use it once.
259 write_unlock(&cache->uc_hashlock);
265 /* Now we know it's good */
266 LASSERT(UC_CACHE_IS_VALID(entry));
267 write_unlock(&cache->uc_hashlock);
272 struct upcall_cache_entry *
273 upcall_cache_get_entry(struct upcall_cache *cache, __u64 key)
277 LASSERT(cache->hash);
279 hash = cache->hash(cache, key);
281 return __get_entry(cache, hash, key, 1, 0);
283 EXPORT_SYMBOL(upcall_cache_get_entry);
285 void upcall_cache_put_entry(struct upcall_cache_entry *entry)
287 struct upcall_cache *cache = entry->ue_cache;
289 write_lock(&cache->uc_hashlock);
290 LASSERTF(atomic_read(&entry->ue_refcount) > 0,
291 "entry %p: ref %d\n", entry, atomic_read(&entry->ue_refcount));
293 write_unlock(&cache->uc_hashlock);
295 EXPORT_SYMBOL(upcall_cache_put_entry);
297 int upcall_cache_downcall(struct upcall_cache *cache, __u64 key, void *args)
299 struct list_head *head;
300 struct upcall_cache_entry *entry;
305 hash = cache->hash(cache, key);
306 LASSERT(hash < cache->uc_hashsize);
308 head = &cache->uc_hashtable[hash];
310 write_lock(&cache->uc_hashlock);
311 list_for_each_entry(entry, head, ue_hash) {
312 if (entry->ue_key == key) {
318 /* haven't found, it's possible */
319 write_unlock(&cache->uc_hashlock);
320 CWARN("key "LPU64" entry dosen't found\n", key);
324 if (!UC_CACHE_IS_ACQUIRING(entry)) {
325 if (UC_CACHE_IS_VALID(entry)) {
326 /* This should not happen, just give a warning
329 CWARN("entry %p(key "LPU64", ac %ld, ex %ld): "
330 "already valid???\n", entry, entry->ue_key,
331 entry->ue_acquire_expire, entry->ue_expire);
335 CWARN("stale entry %p: cur %lu, key "LPU64", ref %d, "
336 "fl %u, ac %ld, ex %ld\n",
337 entry, get_seconds(), entry->ue_key,
338 atomic_read(&entry->ue_refcount), entry->ue_flags,
339 entry->ue_acquire_expire, entry->ue_expire);
340 GOTO(out, rc = -EINVAL);
343 if (!UC_CACHE_IS_ACQUIRING(entry) ||
344 UC_CACHE_IS_INVALID(entry) ||
345 UC_CACHE_IS_EXPIRED(entry)) {
346 CWARN("stale entry %p: cur %lu, key "LPU64", ref %d, "
347 "fl %u, ac %ld, ex %ld\n",
348 entry, get_seconds(), entry->ue_key,
349 atomic_read(&entry->ue_refcount), entry->ue_flags,
350 entry->ue_acquire_expire, entry->ue_expire);
351 GOTO(out, rc = -EINVAL);
354 atomic_inc(&entry->ue_refcount);
355 write_unlock(&cache->uc_hashlock);
356 rc = cache->parse_downcall(cache, entry, args);
357 write_lock(&cache->uc_hashlock);
358 atomic_dec(&entry->ue_refcount);
361 UC_CACHE_SET_INVALID(entry);
362 list_del_init(&entry->ue_hash);
364 } else if (rc == 0) {
365 entry->ue_expire = get_seconds() + cache->uc_entry_expire;
367 entry->ue_expire = get_seconds() + cache->uc_err_entry_expire;
370 UC_CACHE_SET_VALID(entry);
371 CDEBUG(D_SEC, "create ucache entry %p(key "LPU64")\n",
372 entry, entry->ue_key);
374 wake_up_all(&entry->ue_waitq);
375 write_unlock(&cache->uc_hashlock);
378 EXPORT_SYMBOL(upcall_cache_downcall);
380 void upcall_cache_flush_one(struct upcall_cache *cache, __u64 key)
382 struct list_head *head;
383 struct upcall_cache_entry *entry;
388 hash = cache->hash(cache, key);
389 LASSERT(hash < cache->uc_hashsize);
391 head = &cache->uc_hashtable[hash];
393 write_lock(&cache->uc_hashlock);
394 list_for_each_entry(entry, head, ue_hash) {
395 if (entry->ue_key == key) {
402 UC_CACHE_SET_EXPIRED(entry);
403 if (!atomic_read(&entry->ue_refcount))
406 write_unlock(&cache->uc_hashlock);
408 EXPORT_SYMBOL(upcall_cache_flush_one);
410 static void cache_flush(struct upcall_cache *cache, int force, int sync)
412 struct upcall_cache_entry *entry, *next;
416 write_lock(&cache->uc_hashlock);
417 for (i = 0; i < cache->uc_hashsize; i++) {
418 list_for_each_entry_safe(entry, next,
419 &cache->uc_hashtable[i], ue_hash) {
420 if (!force && atomic_read(&entry->ue_refcount)) {
421 UC_CACHE_SET_EXPIRED(entry);
424 LASSERT(!atomic_read(&entry->ue_refcount));
428 write_unlock(&cache->uc_hashlock);
432 void upcall_cache_flush_idle(struct upcall_cache *cache)
434 cache_flush(cache, 0, 0);
437 void upcall_cache_flush_all(struct upcall_cache *cache)
439 cache_flush(cache, 1, 0);
441 EXPORT_SYMBOL(upcall_cache_flush_idle);
442 EXPORT_SYMBOL(upcall_cache_flush_all);