Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf | |
19 | * | |
20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
21 | * CA 95054 USA or visit www.sun.com if you need additional information or | |
22 | * have any questions. | |
23 | * | |
24 | * GPL HEADER END | |
25 | */ | |
26 | /* | |
27 | * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. | |
28 | * Use is subject to license terms. | |
29 | * | |
30 | * Copyright (c) 2010, 2012, Intel Corporation. | |
31 | */ | |
32 | /* | |
33 | * This file is part of Lustre, http://www.lustre.org/ | |
34 | * Lustre is a trademark of Sun Microsystems, Inc. | |
35 | * | |
36 | * lustre/ldlm/ldlm_lock.c | |
37 | * | |
38 | * Author: Peter Braam <braam@clusterfs.com> | |
39 | * Author: Phil Schwan <phil@clusterfs.com> | |
40 | */ | |
41 | ||
42 | #define DEBUG_SUBSYSTEM S_LDLM | |
43 | ||
9fdaf8c0 | 44 | #include "../../include/linux/libcfs/libcfs.h" |
e27db149 GKH |
45 | #include "../include/linux/lustre_intent.h" |
46 | #include "../include/obd_class.h" | |
d7e09d03 PT |
47 | #include "ldlm_internal.h" |
48 | ||
49 | /* lock types */ | |
50 | char *ldlm_lockname[] = { | |
805e517a EG |
51 | [0] = "--", |
52 | [LCK_EX] = "EX", | |
53 | [LCK_PW] = "PW", | |
54 | [LCK_PR] = "PR", | |
55 | [LCK_CW] = "CW", | |
56 | [LCK_CR] = "CR", | |
57 | [LCK_NL] = "NL", | |
58 | [LCK_GROUP] = "GROUP", | |
59 | [LCK_COS] = "COS", | |
d7e09d03 PT |
60 | }; |
61 | EXPORT_SYMBOL(ldlm_lockname); | |
62 | ||
63 | char *ldlm_typename[] = { | |
805e517a EG |
64 | [LDLM_PLAIN] = "PLN", |
65 | [LDLM_EXTENT] = "EXT", | |
66 | [LDLM_FLOCK] = "FLK", | |
67 | [LDLM_IBITS] = "IBT", | |
d7e09d03 PT |
68 | }; |
69 | EXPORT_SYMBOL(ldlm_typename); | |
70 | ||
71 | static ldlm_policy_wire_to_local_t ldlm_policy_wire18_to_local[] = { | |
805e517a EG |
72 | [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_wire_to_local, |
73 | [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_wire_to_local, | |
74 | [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_wire18_to_local, | |
75 | [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local, | |
d7e09d03 PT |
76 | }; |
77 | ||
78 | static ldlm_policy_wire_to_local_t ldlm_policy_wire21_to_local[] = { | |
805e517a EG |
79 | [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_wire_to_local, |
80 | [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_wire_to_local, | |
81 | [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_wire21_to_local, | |
82 | [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local, | |
d7e09d03 PT |
83 | }; |
84 | ||
85 | static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire[] = { | |
805e517a EG |
86 | [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_local_to_wire, |
87 | [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_local_to_wire, | |
88 | [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_local_to_wire, | |
89 | [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_local_to_wire, | |
d7e09d03 PT |
90 | }; |
91 | ||
92 | /** | |
93 | * Converts lock policy from local format to on the wire lock_desc format | |
94 | */ | |
95 | void ldlm_convert_policy_to_wire(ldlm_type_t type, | |
96 | const ldlm_policy_data_t *lpolicy, | |
97 | ldlm_wire_policy_data_t *wpolicy) | |
98 | { | |
99 | ldlm_policy_local_to_wire_t convert; | |
100 | ||
101 | convert = ldlm_policy_local_to_wire[type - LDLM_MIN_TYPE]; | |
102 | ||
103 | convert(lpolicy, wpolicy); | |
104 | } | |
105 | ||
106 | /** | |
107 | * Converts lock policy from on the wire lock_desc format to local format | |
108 | */ | |
109 | void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type, | |
110 | const ldlm_wire_policy_data_t *wpolicy, | |
111 | ldlm_policy_data_t *lpolicy) | |
112 | { | |
113 | ldlm_policy_wire_to_local_t convert; | |
114 | int new_client; | |
115 | ||
116 | /** some badness for 2.0.0 clients, but 2.0.0 isn't supported */ | |
117 | new_client = (exp_connect_flags(exp) & OBD_CONNECT_FULL20) != 0; | |
118 | if (new_client) | |
119 | convert = ldlm_policy_wire21_to_local[type - LDLM_MIN_TYPE]; | |
120 | else | |
121 | convert = ldlm_policy_wire18_to_local[type - LDLM_MIN_TYPE]; | |
122 | ||
123 | convert(wpolicy, lpolicy); | |
124 | } | |
125 | ||
126 | char *ldlm_it2str(int it) | |
127 | { | |
128 | switch (it) { | |
129 | case IT_OPEN: | |
130 | return "open"; | |
131 | case IT_CREAT: | |
132 | return "creat"; | |
133 | case (IT_OPEN | IT_CREAT): | |
134 | return "open|creat"; | |
135 | case IT_READDIR: | |
136 | return "readdir"; | |
137 | case IT_GETATTR: | |
138 | return "getattr"; | |
139 | case IT_LOOKUP: | |
140 | return "lookup"; | |
141 | case IT_UNLINK: | |
142 | return "unlink"; | |
143 | case IT_GETXATTR: | |
144 | return "getxattr"; | |
145 | case IT_LAYOUT: | |
146 | return "layout"; | |
147 | default: | |
148 | CERROR("Unknown intent %d\n", it); | |
149 | return "UNKNOWN"; | |
150 | } | |
151 | } | |
152 | EXPORT_SYMBOL(ldlm_it2str); | |
153 | ||
154 | extern struct kmem_cache *ldlm_lock_slab; | |
155 | ||
156 | ||
157 | void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg) | |
158 | { | |
159 | ns->ns_policy = arg; | |
160 | } | |
161 | EXPORT_SYMBOL(ldlm_register_intent); | |
162 | ||
163 | /* | |
164 | * REFCOUNTED LOCK OBJECTS | |
165 | */ | |
166 | ||
167 | ||
168 | /** | |
169 | * Get a reference on a lock. | |
170 | * | |
171 | * Lock refcounts, during creation: | |
172 | * - one special one for allocation, dec'd only once in destroy | |
173 | * - one for being a lock that's in-use | |
174 | * - one for the addref associated with a new lock | |
175 | */ | |
176 | struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock) | |
177 | { | |
178 | atomic_inc(&lock->l_refc); | |
179 | return lock; | |
180 | } | |
181 | EXPORT_SYMBOL(ldlm_lock_get); | |
182 | ||
183 | /** | |
184 | * Release lock reference. | |
185 | * | |
186 | * Also frees the lock if it was last reference. | |
187 | */ | |
188 | void ldlm_lock_put(struct ldlm_lock *lock) | |
189 | { | |
d7e09d03 PT |
190 | LASSERT(lock->l_resource != LP_POISON); |
191 | LASSERT(atomic_read(&lock->l_refc) > 0); | |
192 | if (atomic_dec_and_test(&lock->l_refc)) { | |
193 | struct ldlm_resource *res; | |
194 | ||
195 | LDLM_DEBUG(lock, | |
196 | "final lock_put on destroyed lock, freeing it."); | |
197 | ||
198 | res = lock->l_resource; | |
f2145eae | 199 | LASSERT(lock->l_flags & LDLM_FL_DESTROYED); |
d7e09d03 PT |
200 | LASSERT(list_empty(&lock->l_res_link)); |
201 | LASSERT(list_empty(&lock->l_pending_chain)); | |
202 | ||
203 | lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats, | |
204 | LDLM_NSS_LOCKS); | |
205 | lu_ref_del(&res->lr_reference, "lock", lock); | |
206 | ldlm_resource_putref(res); | |
207 | lock->l_resource = NULL; | |
208 | if (lock->l_export) { | |
209 | class_export_lock_put(lock->l_export, lock); | |
210 | lock->l_export = NULL; | |
211 | } | |
212 | ||
213 | if (lock->l_lvb_data != NULL) | |
214 | OBD_FREE(lock->l_lvb_data, lock->l_lvb_len); | |
215 | ||
216 | ldlm_interval_free(ldlm_interval_detach(lock)); | |
217 | lu_ref_fini(&lock->l_reference); | |
218 | OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle); | |
219 | } | |
d7e09d03 PT |
220 | } |
221 | EXPORT_SYMBOL(ldlm_lock_put); | |
222 | ||
223 | /** | |
224 | * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked. | |
225 | */ | |
226 | int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock) | |
227 | { | |
228 | int rc = 0; | |
229 | if (!list_empty(&lock->l_lru)) { | |
230 | struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); | |
231 | ||
232 | LASSERT(lock->l_resource->lr_type != LDLM_FLOCK); | |
233 | list_del_init(&lock->l_lru); | |
d7e09d03 PT |
234 | LASSERT(ns->ns_nr_unused > 0); |
235 | ns->ns_nr_unused--; | |
236 | rc = 1; | |
237 | } | |
238 | return rc; | |
239 | } | |
240 | ||
241 | /** | |
242 | * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first. | |
243 | */ | |
244 | int ldlm_lock_remove_from_lru(struct ldlm_lock *lock) | |
245 | { | |
246 | struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); | |
247 | int rc; | |
248 | ||
f2145eae | 249 | if (lock->l_flags & LDLM_FL_NS_SRV) { |
d7e09d03 | 250 | LASSERT(list_empty(&lock->l_lru)); |
0a3bdb00 | 251 | return 0; |
d7e09d03 PT |
252 | } |
253 | ||
254 | spin_lock(&ns->ns_lock); | |
255 | rc = ldlm_lock_remove_from_lru_nolock(lock); | |
256 | spin_unlock(&ns->ns_lock); | |
d7e09d03 PT |
257 | return rc; |
258 | } | |
259 | ||
260 | /** | |
261 | * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked. | |
262 | */ | |
263 | void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock) | |
264 | { | |
265 | struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); | |
266 | ||
267 | lock->l_last_used = cfs_time_current(); | |
268 | LASSERT(list_empty(&lock->l_lru)); | |
269 | LASSERT(lock->l_resource->lr_type != LDLM_FLOCK); | |
270 | list_add_tail(&lock->l_lru, &ns->ns_unused_list); | |
88e4ff86 LX |
271 | if (lock->l_flags & LDLM_FL_SKIPPED) |
272 | lock->l_flags &= ~LDLM_FL_SKIPPED; | |
d7e09d03 PT |
273 | LASSERT(ns->ns_nr_unused >= 0); |
274 | ns->ns_nr_unused++; | |
275 | } | |
276 | ||
277 | /** | |
278 | * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks | |
279 | * first. | |
280 | */ | |
281 | void ldlm_lock_add_to_lru(struct ldlm_lock *lock) | |
282 | { | |
283 | struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); | |
284 | ||
d7e09d03 PT |
285 | spin_lock(&ns->ns_lock); |
286 | ldlm_lock_add_to_lru_nolock(lock); | |
287 | spin_unlock(&ns->ns_lock); | |
d7e09d03 PT |
288 | } |
289 | ||
290 | /** | |
291 | * Moves LDLM lock \a lock that is already in namespace LRU to the tail of | |
292 | * the LRU. Performs necessary LRU locking | |
293 | */ | |
294 | void ldlm_lock_touch_in_lru(struct ldlm_lock *lock) | |
295 | { | |
296 | struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); | |
297 | ||
f2145eae | 298 | if (lock->l_flags & LDLM_FL_NS_SRV) { |
d7e09d03 | 299 | LASSERT(list_empty(&lock->l_lru)); |
d7e09d03 PT |
300 | return; |
301 | } | |
302 | ||
303 | spin_lock(&ns->ns_lock); | |
304 | if (!list_empty(&lock->l_lru)) { | |
305 | ldlm_lock_remove_from_lru_nolock(lock); | |
306 | ldlm_lock_add_to_lru_nolock(lock); | |
307 | } | |
308 | spin_unlock(&ns->ns_lock); | |
d7e09d03 PT |
309 | } |
310 | ||
311 | /** | |
312 | * Helper to destroy a locked lock. | |
313 | * | |
314 | * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock | |
315 | * Must be called with l_lock and lr_lock held. | |
316 | * | |
317 | * Does not actually free the lock data, but rather marks the lock as | |
318 | * destroyed by setting l_destroyed field in the lock to 1. Destroys a | |
319 | * handle->lock association too, so that the lock can no longer be found | |
320 | * and removes the lock from LRU list. Actual lock freeing occurs when | |
321 | * last lock reference goes away. | |
322 | * | |
323 | * Original comment (of some historical value): | |
324 | * This used to have a 'strict' flag, which recovery would use to mark an | |
325 | * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I | |
326 | * shall explain why it's gone: with the new hash table scheme, once you call | |
327 | * ldlm_lock_destroy, you can never drop your final references on this lock. | |
328 | * Because it's not in the hash table anymore. -phil | |
329 | */ | |
330 | int ldlm_lock_destroy_internal(struct ldlm_lock *lock) | |
331 | { | |
d7e09d03 PT |
332 | if (lock->l_readers || lock->l_writers) { |
333 | LDLM_ERROR(lock, "lock still has references"); | |
334 | LBUG(); | |
335 | } | |
336 | ||
337 | if (!list_empty(&lock->l_res_link)) { | |
338 | LDLM_ERROR(lock, "lock still on resource"); | |
339 | LBUG(); | |
340 | } | |
341 | ||
f2145eae | 342 | if (lock->l_flags & LDLM_FL_DESTROYED) { |
d7e09d03 | 343 | LASSERT(list_empty(&lock->l_lru)); |
d7e09d03 PT |
344 | return 0; |
345 | } | |
f2145eae | 346 | lock->l_flags |= LDLM_FL_DESTROYED; |
d7e09d03 PT |
347 | |
348 | if (lock->l_export && lock->l_export->exp_lock_hash) { | |
349 | /* NB: it's safe to call cfs_hash_del() even lock isn't | |
350 | * in exp_lock_hash. */ | |
351 | /* In the function below, .hs_keycmp resolves to | |
352 | * ldlm_export_lock_keycmp() */ | |
353 | /* coverity[overrun-buffer-val] */ | |
354 | cfs_hash_del(lock->l_export->exp_lock_hash, | |
355 | &lock->l_remote_handle, &lock->l_exp_hash); | |
356 | } | |
357 | ||
358 | ldlm_lock_remove_from_lru(lock); | |
359 | class_handle_unhash(&lock->l_handle); | |
360 | ||
361 | #if 0 | |
362 | /* Wake anyone waiting for this lock */ | |
363 | /* FIXME: I should probably add yet another flag, instead of using | |
364 | * l_export to only call this on clients */ | |
365 | if (lock->l_export) | |
366 | class_export_put(lock->l_export); | |
367 | lock->l_export = NULL; | |
368 | if (lock->l_export && lock->l_completion_ast) | |
369 | lock->l_completion_ast(lock, 0); | |
370 | #endif | |
d7e09d03 PT |
371 | return 1; |
372 | } | |
373 | ||
374 | /** | |
375 | * Destroys a LDLM lock \a lock. Performs necessary locking first. | |
376 | */ | |
377 | void ldlm_lock_destroy(struct ldlm_lock *lock) | |
378 | { | |
379 | int first; | |
29aaf496 | 380 | |
d7e09d03 PT |
381 | lock_res_and_lock(lock); |
382 | first = ldlm_lock_destroy_internal(lock); | |
383 | unlock_res_and_lock(lock); | |
384 | ||
385 | /* drop reference from hashtable only for first destroy */ | |
386 | if (first) { | |
387 | lu_ref_del(&lock->l_reference, "hash", lock); | |
388 | LDLM_LOCK_RELEASE(lock); | |
389 | } | |
d7e09d03 PT |
390 | } |
391 | ||
392 | /** | |
393 | * Destroys a LDLM lock \a lock that is already locked. | |
394 | */ | |
395 | void ldlm_lock_destroy_nolock(struct ldlm_lock *lock) | |
396 | { | |
397 | int first; | |
29aaf496 | 398 | |
d7e09d03 PT |
399 | first = ldlm_lock_destroy_internal(lock); |
400 | /* drop reference from hashtable only for first destroy */ | |
401 | if (first) { | |
402 | lu_ref_del(&lock->l_reference, "hash", lock); | |
403 | LDLM_LOCK_RELEASE(lock); | |
404 | } | |
d7e09d03 PT |
405 | } |
406 | ||
407 | /* this is called by portals_handle2object with the handle lock taken */ | |
408 | static void lock_handle_addref(void *lock) | |
409 | { | |
410 | LDLM_LOCK_GET((struct ldlm_lock *)lock); | |
411 | } | |
412 | ||
413 | static void lock_handle_free(void *lock, int size) | |
414 | { | |
415 | LASSERT(size == sizeof(struct ldlm_lock)); | |
416 | OBD_SLAB_FREE(lock, ldlm_lock_slab, size); | |
417 | } | |
418 | ||
419 | struct portals_handle_ops lock_handle_ops = { | |
420 | .hop_addref = lock_handle_addref, | |
421 | .hop_free = lock_handle_free, | |
422 | }; | |
423 | ||
424 | /** | |
425 | * | |
426 | * Allocate and initialize new lock structure. | |
427 | * | |
428 | * usage: pass in a resource on which you have done ldlm_resource_get | |
429 | * new lock will take over the refcount. | |
430 | * returns: lock with refcount 2 - one for current caller and one for remote | |
431 | */ | |
432 | static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource) | |
433 | { | |
434 | struct ldlm_lock *lock; | |
d7e09d03 PT |
435 | |
436 | if (resource == NULL) | |
437 | LBUG(); | |
438 | ||
0be19afa | 439 | OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, GFP_NOFS); |
d7e09d03 | 440 | if (lock == NULL) |
0a3bdb00 | 441 | return NULL; |
d7e09d03 PT |
442 | |
443 | spin_lock_init(&lock->l_lock); | |
444 | lock->l_resource = resource; | |
445 | lu_ref_add(&resource->lr_reference, "lock", lock); | |
446 | ||
447 | atomic_set(&lock->l_refc, 2); | |
448 | INIT_LIST_HEAD(&lock->l_res_link); | |
449 | INIT_LIST_HEAD(&lock->l_lru); | |
450 | INIT_LIST_HEAD(&lock->l_pending_chain); | |
451 | INIT_LIST_HEAD(&lock->l_bl_ast); | |
452 | INIT_LIST_HEAD(&lock->l_cp_ast); | |
453 | INIT_LIST_HEAD(&lock->l_rk_ast); | |
454 | init_waitqueue_head(&lock->l_waitq); | |
455 | lock->l_blocking_lock = NULL; | |
456 | INIT_LIST_HEAD(&lock->l_sl_mode); | |
457 | INIT_LIST_HEAD(&lock->l_sl_policy); | |
458 | INIT_HLIST_NODE(&lock->l_exp_hash); | |
459 | INIT_HLIST_NODE(&lock->l_exp_flock_hash); | |
460 | ||
461 | lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats, | |
462 | LDLM_NSS_LOCKS); | |
463 | INIT_LIST_HEAD(&lock->l_handle.h_link); | |
464 | class_handle_hash(&lock->l_handle, &lock_handle_ops); | |
465 | ||
466 | lu_ref_init(&lock->l_reference); | |
467 | lu_ref_add(&lock->l_reference, "hash", lock); | |
468 | lock->l_callback_timeout = 0; | |
469 | ||
470 | #if LUSTRE_TRACKS_LOCK_EXP_REFS | |
471 | INIT_LIST_HEAD(&lock->l_exp_refs_link); | |
472 | lock->l_exp_refs_nr = 0; | |
473 | lock->l_exp_refs_target = NULL; | |
474 | #endif | |
475 | INIT_LIST_HEAD(&lock->l_exp_list); | |
476 | ||
0a3bdb00 | 477 | return lock; |
d7e09d03 PT |
478 | } |
479 | ||
480 | /** | |
481 | * Moves LDLM lock \a lock to another resource. | |
482 | * This is used on client when server returns some other lock than requested | |
483 | * (typically as a result of intent operation) | |
484 | */ | |
485 | int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock, | |
486 | const struct ldlm_res_id *new_resid) | |
487 | { | |
488 | struct ldlm_resource *oldres = lock->l_resource; | |
489 | struct ldlm_resource *newres; | |
490 | int type; | |
d7e09d03 PT |
491 | |
492 | LASSERT(ns_is_client(ns)); | |
493 | ||
494 | lock_res_and_lock(lock); | |
495 | if (memcmp(new_resid, &lock->l_resource->lr_name, | |
496 | sizeof(lock->l_resource->lr_name)) == 0) { | |
497 | /* Nothing to do */ | |
498 | unlock_res_and_lock(lock); | |
0a3bdb00 | 499 | return 0; |
d7e09d03 PT |
500 | } |
501 | ||
502 | LASSERT(new_resid->name[0] != 0); | |
503 | ||
504 | /* This function assumes that the lock isn't on any lists */ | |
505 | LASSERT(list_empty(&lock->l_res_link)); | |
506 | ||
507 | type = oldres->lr_type; | |
508 | unlock_res_and_lock(lock); | |
509 | ||
510 | newres = ldlm_resource_get(ns, NULL, new_resid, type, 1); | |
511 | if (newres == NULL) | |
0a3bdb00 | 512 | return -ENOMEM; |
d7e09d03 PT |
513 | |
514 | lu_ref_add(&newres->lr_reference, "lock", lock); | |
515 | /* | |
516 | * To flip the lock from the old to the new resource, lock, oldres and | |
517 | * newres have to be locked. Resource spin-locks are nested within | |
518 | * lock->l_lock, and are taken in the memory address order to avoid | |
519 | * dead-locks. | |
520 | */ | |
521 | spin_lock(&lock->l_lock); | |
522 | oldres = lock->l_resource; | |
523 | if (oldres < newres) { | |
524 | lock_res(oldres); | |
525 | lock_res_nested(newres, LRT_NEW); | |
526 | } else { | |
527 | lock_res(newres); | |
528 | lock_res_nested(oldres, LRT_NEW); | |
529 | } | |
530 | LASSERT(memcmp(new_resid, &oldres->lr_name, | |
ec83e611 | 531 | sizeof(oldres->lr_name)) != 0); |
d7e09d03 PT |
532 | lock->l_resource = newres; |
533 | unlock_res(oldres); | |
534 | unlock_res_and_lock(lock); | |
535 | ||
536 | /* ...and the flowers are still standing! */ | |
537 | lu_ref_del(&oldres->lr_reference, "lock", lock); | |
538 | ldlm_resource_putref(oldres); | |
539 | ||
0a3bdb00 | 540 | return 0; |
d7e09d03 PT |
541 | } |
542 | EXPORT_SYMBOL(ldlm_lock_change_resource); | |
543 | ||
544 | /** \defgroup ldlm_handles LDLM HANDLES | |
545 | * Ways to get hold of locks without any addresses. | |
546 | * @{ | |
547 | */ | |
548 | ||
549 | /** | |
550 | * Fills in handle for LDLM lock \a lock into supplied \a lockh | |
551 | * Does not take any references. | |
552 | */ | |
553 | void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh) | |
554 | { | |
555 | lockh->cookie = lock->l_handle.h_cookie; | |
556 | } | |
557 | EXPORT_SYMBOL(ldlm_lock2handle); | |
558 | ||
559 | /** | |
560 | * Obtain a lock reference by handle. | |
561 | * | |
562 | * if \a flags: atomically get the lock and set the flags. | |
563 | * Return NULL if flag already set | |
564 | */ | |
565 | struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle, | |
566 | __u64 flags) | |
567 | { | |
568 | struct ldlm_lock *lock; | |
d7e09d03 PT |
569 | |
570 | LASSERT(handle); | |
571 | ||
572 | lock = class_handle2object(handle->cookie); | |
573 | if (lock == NULL) | |
0a3bdb00 | 574 | return NULL; |
d7e09d03 PT |
575 | |
576 | /* It's unlikely but possible that someone marked the lock as | |
577 | * destroyed after we did handle2object on it */ | |
f2145eae | 578 | if (flags == 0 && ((lock->l_flags & LDLM_FL_DESTROYED)== 0)) { |
d7e09d03 | 579 | lu_ref_add(&lock->l_reference, "handle", current); |
0a3bdb00 | 580 | return lock; |
d7e09d03 PT |
581 | } |
582 | ||
583 | lock_res_and_lock(lock); | |
584 | ||
585 | LASSERT(lock->l_resource != NULL); | |
586 | ||
587 | lu_ref_add_atomic(&lock->l_reference, "handle", current); | |
f2145eae | 588 | if (unlikely(lock->l_flags & LDLM_FL_DESTROYED)) { |
d7e09d03 PT |
589 | unlock_res_and_lock(lock); |
590 | CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock); | |
591 | LDLM_LOCK_PUT(lock); | |
0a3bdb00 | 592 | return NULL; |
d7e09d03 PT |
593 | } |
594 | ||
595 | if (flags && (lock->l_flags & flags)) { | |
596 | unlock_res_and_lock(lock); | |
597 | LDLM_LOCK_PUT(lock); | |
0a3bdb00 | 598 | return NULL; |
d7e09d03 PT |
599 | } |
600 | ||
601 | if (flags) | |
602 | lock->l_flags |= flags; | |
603 | ||
604 | unlock_res_and_lock(lock); | |
0a3bdb00 | 605 | return lock; |
d7e09d03 PT |
606 | } |
607 | EXPORT_SYMBOL(__ldlm_handle2lock); | |
608 | /** @} ldlm_handles */ | |
609 | ||
610 | /** | |
611 | * Fill in "on the wire" representation for given LDLM lock into supplied | |
612 | * lock descriptor \a desc structure. | |
613 | */ | |
614 | void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc) | |
615 | { | |
f261f48a FY |
616 | ldlm_res2desc(lock->l_resource, &desc->l_resource); |
617 | desc->l_req_mode = lock->l_req_mode; | |
618 | desc->l_granted_mode = lock->l_granted_mode; | |
619 | ldlm_convert_policy_to_wire(lock->l_resource->lr_type, | |
620 | &lock->l_policy_data, | |
621 | &desc->l_policy_data); | |
d7e09d03 PT |
622 | } |
623 | EXPORT_SYMBOL(ldlm_lock2desc); | |
624 | ||
625 | /** | |
626 | * Add a lock to list of conflicting locks to send AST to. | |
627 | * | |
628 | * Only add if we have not sent a blocking AST to the lock yet. | |
629 | */ | |
630 | void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, | |
631 | struct list_head *work_list) | |
632 | { | |
633 | if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) { | |
634 | LDLM_DEBUG(lock, "lock incompatible; sending blocking AST."); | |
635 | lock->l_flags |= LDLM_FL_AST_SENT; | |
636 | /* If the enqueuing client said so, tell the AST recipient to | |
637 | * discard dirty data, rather than writing back. */ | |
f2145eae | 638 | if (new->l_flags & LDLM_FL_AST_DISCARD_DATA) |
d7e09d03 PT |
639 | lock->l_flags |= LDLM_FL_DISCARD_DATA; |
640 | LASSERT(list_empty(&lock->l_bl_ast)); | |
641 | list_add(&lock->l_bl_ast, work_list); | |
642 | LDLM_LOCK_GET(lock); | |
643 | LASSERT(lock->l_blocking_lock == NULL); | |
644 | lock->l_blocking_lock = LDLM_LOCK_GET(new); | |
645 | } | |
646 | } | |
647 | ||
648 | /** | |
649 | * Add a lock to list of just granted locks to send completion AST to. | |
650 | */ | |
651 | void ldlm_add_cp_work_item(struct ldlm_lock *lock, struct list_head *work_list) | |
652 | { | |
653 | if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) { | |
654 | lock->l_flags |= LDLM_FL_CP_REQD; | |
655 | LDLM_DEBUG(lock, "lock granted; sending completion AST."); | |
656 | LASSERT(list_empty(&lock->l_cp_ast)); | |
657 | list_add(&lock->l_cp_ast, work_list); | |
658 | LDLM_LOCK_GET(lock); | |
659 | } | |
660 | } | |
661 | ||
662 | /** | |
663 | * Aggregator function to add AST work items into a list. Determines | |
664 | * what sort of an AST work needs to be done and calls the proper | |
665 | * adding function. | |
666 | * Must be called with lr_lock held. | |
667 | */ | |
668 | void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, | |
669 | struct list_head *work_list) | |
670 | { | |
d7e09d03 PT |
671 | check_res_locked(lock->l_resource); |
672 | if (new) | |
673 | ldlm_add_bl_work_item(lock, new, work_list); | |
674 | else | |
675 | ldlm_add_cp_work_item(lock, work_list); | |
d7e09d03 PT |
676 | } |
677 | ||
678 | /** | |
679 | * Add specified reader/writer reference to LDLM lock with handle \a lockh. | |
680 | * r/w reference type is determined by \a mode | |
681 | * Calls ldlm_lock_addref_internal. | |
682 | */ | |
683 | void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode) | |
684 | { | |
685 | struct ldlm_lock *lock; | |
686 | ||
687 | lock = ldlm_handle2lock(lockh); | |
688 | LASSERT(lock != NULL); | |
689 | ldlm_lock_addref_internal(lock, mode); | |
690 | LDLM_LOCK_PUT(lock); | |
691 | } | |
692 | EXPORT_SYMBOL(ldlm_lock_addref); | |
693 | ||
694 | /** | |
695 | * Helper function. | |
696 | * Add specified reader/writer reference to LDLM lock \a lock. | |
697 | * r/w reference type is determined by \a mode | |
698 | * Removes lock from LRU if it is there. | |
699 | * Assumes the LDLM lock is already locked. | |
700 | */ | |
701 | void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode) | |
702 | { | |
703 | ldlm_lock_remove_from_lru(lock); | |
704 | if (mode & (LCK_NL | LCK_CR | LCK_PR)) { | |
705 | lock->l_readers++; | |
706 | lu_ref_add_atomic(&lock->l_reference, "reader", lock); | |
707 | } | |
708 | if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) { | |
709 | lock->l_writers++; | |
710 | lu_ref_add_atomic(&lock->l_reference, "writer", lock); | |
711 | } | |
712 | LDLM_LOCK_GET(lock); | |
713 | lu_ref_add_atomic(&lock->l_reference, "user", lock); | |
714 | LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]); | |
715 | } | |
716 | ||
717 | /** | |
718 | * Attempts to add reader/writer reference to a lock with handle \a lockh, and | |
719 | * fails if lock is already LDLM_FL_CBPENDING or destroyed. | |
720 | * | |
721 | * \retval 0 success, lock was addref-ed | |
722 | * | |
723 | * \retval -EAGAIN lock is being canceled. | |
724 | */ | |
725 | int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode) | |
726 | { | |
727 | struct ldlm_lock *lock; | |
728 | int result; | |
729 | ||
730 | result = -EAGAIN; | |
731 | lock = ldlm_handle2lock(lockh); | |
732 | if (lock != NULL) { | |
733 | lock_res_and_lock(lock); | |
734 | if (lock->l_readers != 0 || lock->l_writers != 0 || | |
735 | !(lock->l_flags & LDLM_FL_CBPENDING)) { | |
736 | ldlm_lock_addref_internal_nolock(lock, mode); | |
737 | result = 0; | |
738 | } | |
739 | unlock_res_and_lock(lock); | |
740 | LDLM_LOCK_PUT(lock); | |
741 | } | |
742 | return result; | |
743 | } | |
744 | EXPORT_SYMBOL(ldlm_lock_addref_try); | |
745 | ||
746 | /** | |
747 | * Add specified reader/writer reference to LDLM lock \a lock. | |
748 | * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work. | |
749 | * Only called for local locks. | |
750 | */ | |
751 | void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode) | |
752 | { | |
753 | lock_res_and_lock(lock); | |
754 | ldlm_lock_addref_internal_nolock(lock, mode); | |
755 | unlock_res_and_lock(lock); | |
756 | } | |
757 | ||
758 | /** | |
759 | * Removes reader/writer reference for LDLM lock \a lock. | |
760 | * Assumes LDLM lock is already locked. | |
761 | * only called in ldlm_flock_destroy and for local locks. | |
6e3dd654 | 762 | * Does NOT add lock to LRU if no r/w references left to accommodate flock locks |
d7e09d03 PT |
763 | * that cannot be placed in LRU. |
764 | */ | |
765 | void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode) | |
766 | { | |
767 | LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]); | |
768 | if (mode & (LCK_NL | LCK_CR | LCK_PR)) { | |
769 | LASSERT(lock->l_readers > 0); | |
770 | lu_ref_del(&lock->l_reference, "reader", lock); | |
771 | lock->l_readers--; | |
772 | } | |
773 | if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) { | |
774 | LASSERT(lock->l_writers > 0); | |
775 | lu_ref_del(&lock->l_reference, "writer", lock); | |
776 | lock->l_writers--; | |
777 | } | |
778 | ||
779 | lu_ref_del(&lock->l_reference, "user", lock); | |
780 | LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */ | |
781 | } | |
782 | ||
783 | /** | |
784 | * Removes reader/writer reference for LDLM lock \a lock. | |
785 | * Locks LDLM lock first. | |
786 | * If the lock is determined to be client lock on a client and r/w refcount | |
787 | * drops to zero and the lock is not blocked, the lock is added to LRU lock | |
788 | * on the namespace. | |
789 | * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called. | |
790 | */ | |
791 | void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode) | |
792 | { | |
793 | struct ldlm_namespace *ns; | |
d7e09d03 PT |
794 | |
795 | lock_res_and_lock(lock); | |
796 | ||
797 | ns = ldlm_lock_to_ns(lock); | |
798 | ||
799 | ldlm_lock_decref_internal_nolock(lock, mode); | |
800 | ||
d7e09d03 PT |
801 | if (lock->l_flags & LDLM_FL_LOCAL && |
802 | !lock->l_readers && !lock->l_writers) { | |
803 | /* If this is a local lock on a server namespace and this was | |
804 | * the last reference, cancel the lock. */ | |
805 | CDEBUG(D_INFO, "forcing cancel of local lock\n"); | |
806 | lock->l_flags |= LDLM_FL_CBPENDING; | |
807 | } | |
808 | ||
809 | if (!lock->l_readers && !lock->l_writers && | |
810 | (lock->l_flags & LDLM_FL_CBPENDING)) { | |
811 | /* If we received a blocked AST and this was the last reference, | |
812 | * run the callback. */ | |
f2145eae | 813 | if ((lock->l_flags & LDLM_FL_NS_SRV) && lock->l_export) |
d7e09d03 PT |
814 | CERROR("FL_CBPENDING set on non-local lock--just a " |
815 | "warning\n"); | |
816 | ||
817 | LDLM_DEBUG(lock, "final decref done on cbpending lock"); | |
818 | ||
819 | LDLM_LOCK_GET(lock); /* dropped by bl thread */ | |
820 | ldlm_lock_remove_from_lru(lock); | |
821 | unlock_res_and_lock(lock); | |
822 | ||
823 | if (lock->l_flags & LDLM_FL_FAIL_LOC) | |
824 | OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE); | |
825 | ||
826 | if ((lock->l_flags & LDLM_FL_ATOMIC_CB) || | |
827 | ldlm_bl_to_thread_lock(ns, NULL, lock) != 0) | |
828 | ldlm_handle_bl_callback(ns, NULL, lock); | |
829 | } else if (ns_is_client(ns) && | |
830 | !lock->l_readers && !lock->l_writers && | |
831 | !(lock->l_flags & LDLM_FL_NO_LRU) && | |
832 | !(lock->l_flags & LDLM_FL_BL_AST)) { | |
833 | ||
834 | LDLM_DEBUG(lock, "add lock into lru list"); | |
835 | ||
836 | /* If this is a client-side namespace and this was the last | |
837 | * reference, put it on the LRU. */ | |
838 | ldlm_lock_add_to_lru(lock); | |
839 | unlock_res_and_lock(lock); | |
840 | ||
841 | if (lock->l_flags & LDLM_FL_FAIL_LOC) | |
842 | OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE); | |
843 | ||
844 | /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE | |
845 | * are not supported by the server, otherwise, it is done on | |
846 | * enqueue. */ | |
847 | if (!exp_connect_cancelset(lock->l_conn_export) && | |
848 | !ns_connect_lru_resize(ns)) | |
849 | ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0); | |
850 | } else { | |
851 | LDLM_DEBUG(lock, "do not add lock into lru list"); | |
852 | unlock_res_and_lock(lock); | |
853 | } | |
d7e09d03 PT |
854 | } |
855 | ||
856 | /** | |
857 | * Decrease reader/writer refcount for LDLM lock with handle \a lockh | |
858 | */ | |
859 | void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode) | |
860 | { | |
861 | struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0); | |
55f5a824 | 862 | LASSERTF(lock != NULL, "Non-existing lock: %#llx\n", lockh->cookie); |
d7e09d03 PT |
863 | ldlm_lock_decref_internal(lock, mode); |
864 | LDLM_LOCK_PUT(lock); | |
865 | } | |
866 | EXPORT_SYMBOL(ldlm_lock_decref); | |
867 | ||
868 | /** | |
869 | * Decrease reader/writer refcount for LDLM lock with handle | |
870 | * \a lockh and mark it for subsequent cancellation once r/w refcount | |
871 | * drops to zero instead of putting into LRU. | |
872 | * | |
873 | * Typical usage is for GROUP locks which we cannot allow to be cached. | |
874 | */ | |
875 | void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode) | |
876 | { | |
877 | struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0); | |
d7e09d03 PT |
878 | |
879 | LASSERT(lock != NULL); | |
880 | ||
881 | LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]); | |
882 | lock_res_and_lock(lock); | |
883 | lock->l_flags |= LDLM_FL_CBPENDING; | |
884 | unlock_res_and_lock(lock); | |
885 | ldlm_lock_decref_internal(lock, mode); | |
886 | LDLM_LOCK_PUT(lock); | |
887 | } | |
888 | EXPORT_SYMBOL(ldlm_lock_decref_and_cancel); | |
889 | ||
890 | struct sl_insert_point { | |
891 | struct list_head *res_link; | |
892 | struct list_head *mode_link; | |
893 | struct list_head *policy_link; | |
894 | }; | |
895 | ||
896 | /** | |
897 | * Finds a position to insert the new lock into granted lock list. | |
898 | * | |
899 | * Used for locks eligible for skiplist optimization. | |
900 | * | |
901 | * Parameters: | |
902 | * queue [input]: the granted list where search acts on; | |
903 | * req [input]: the lock whose position to be located; | |
904 | * prev [output]: positions within 3 lists to insert @req to | |
905 | * Return Value: | |
906 | * filled @prev | |
907 | * NOTE: called by | |
908 | * - ldlm_grant_lock_with_skiplist | |
909 | */ | |
910 | static void search_granted_lock(struct list_head *queue, | |
911 | struct ldlm_lock *req, | |
912 | struct sl_insert_point *prev) | |
913 | { | |
914 | struct list_head *tmp; | |
915 | struct ldlm_lock *lock, *mode_end, *policy_end; | |
d7e09d03 PT |
916 | |
917 | list_for_each(tmp, queue) { | |
918 | lock = list_entry(tmp, struct ldlm_lock, l_res_link); | |
919 | ||
920 | mode_end = list_entry(lock->l_sl_mode.prev, | |
921 | struct ldlm_lock, l_sl_mode); | |
922 | ||
923 | if (lock->l_req_mode != req->l_req_mode) { | |
924 | /* jump to last lock of mode group */ | |
925 | tmp = &mode_end->l_res_link; | |
926 | continue; | |
927 | } | |
928 | ||
929 | /* suitable mode group is found */ | |
930 | if (lock->l_resource->lr_type == LDLM_PLAIN) { | |
931 | /* insert point is last lock of the mode group */ | |
932 | prev->res_link = &mode_end->l_res_link; | |
933 | prev->mode_link = &mode_end->l_sl_mode; | |
934 | prev->policy_link = &req->l_sl_policy; | |
d7e09d03 PT |
935 | return; |
936 | } else if (lock->l_resource->lr_type == LDLM_IBITS) { | |
937 | for (;;) { | |
938 | policy_end = | |
939 | list_entry(lock->l_sl_policy.prev, | |
940 | struct ldlm_lock, | |
941 | l_sl_policy); | |
942 | ||
943 | if (lock->l_policy_data.l_inodebits.bits == | |
944 | req->l_policy_data.l_inodebits.bits) { | |
945 | /* insert point is last lock of | |
946 | * the policy group */ | |
947 | prev->res_link = | |
948 | &policy_end->l_res_link; | |
949 | prev->mode_link = | |
950 | &policy_end->l_sl_mode; | |
951 | prev->policy_link = | |
952 | &policy_end->l_sl_policy; | |
d7e09d03 PT |
953 | return; |
954 | } | |
955 | ||
956 | if (policy_end == mode_end) | |
957 | /* done with mode group */ | |
958 | break; | |
959 | ||
960 | /* go to next policy group within mode group */ | |
961 | tmp = policy_end->l_res_link.next; | |
962 | lock = list_entry(tmp, struct ldlm_lock, | |
963 | l_res_link); | |
964 | } /* loop over policy groups within the mode group */ | |
965 | ||
966 | /* insert point is last lock of the mode group, | |
967 | * new policy group is started */ | |
968 | prev->res_link = &mode_end->l_res_link; | |
969 | prev->mode_link = &mode_end->l_sl_mode; | |
970 | prev->policy_link = &req->l_sl_policy; | |
d7e09d03 PT |
971 | return; |
972 | } else { | |
973 | LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock"); | |
974 | LBUG(); | |
975 | } | |
976 | } | |
977 | ||
978 | /* insert point is last lock on the queue, | |
979 | * new mode group and new policy group are started */ | |
980 | prev->res_link = queue->prev; | |
981 | prev->mode_link = &req->l_sl_mode; | |
982 | prev->policy_link = &req->l_sl_policy; | |
d7e09d03 PT |
983 | return; |
984 | } | |
985 | ||
986 | /** | |
987 | * Add a lock into resource granted list after a position described by | |
988 | * \a prev. | |
989 | */ | |
990 | static void ldlm_granted_list_add_lock(struct ldlm_lock *lock, | |
991 | struct sl_insert_point *prev) | |
992 | { | |
993 | struct ldlm_resource *res = lock->l_resource; | |
d7e09d03 PT |
994 | |
995 | check_res_locked(res); | |
996 | ||
997 | ldlm_resource_dump(D_INFO, res); | |
998 | LDLM_DEBUG(lock, "About to add lock:"); | |
999 | ||
f2145eae | 1000 | if (lock->l_flags & LDLM_FL_DESTROYED) { |
d7e09d03 PT |
1001 | CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n"); |
1002 | return; | |
1003 | } | |
1004 | ||
1005 | LASSERT(list_empty(&lock->l_res_link)); | |
1006 | LASSERT(list_empty(&lock->l_sl_mode)); | |
1007 | LASSERT(list_empty(&lock->l_sl_policy)); | |
1008 | ||
1009 | /* | |
1010 | * lock->link == prev->link means lock is first starting the group. | |
1011 | * Don't re-add to itself to suppress kernel warnings. | |
1012 | */ | |
1013 | if (&lock->l_res_link != prev->res_link) | |
1014 | list_add(&lock->l_res_link, prev->res_link); | |
1015 | if (&lock->l_sl_mode != prev->mode_link) | |
1016 | list_add(&lock->l_sl_mode, prev->mode_link); | |
1017 | if (&lock->l_sl_policy != prev->policy_link) | |
1018 | list_add(&lock->l_sl_policy, prev->policy_link); | |
d7e09d03 PT |
1019 | } |
1020 | ||
1021 | /** | |
1022 | * Add a lock to granted list on a resource maintaining skiplist | |
1023 | * correctness. | |
1024 | */ | |
1025 | static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock) | |
1026 | { | |
1027 | struct sl_insert_point prev; | |
d7e09d03 PT |
1028 | |
1029 | LASSERT(lock->l_req_mode == lock->l_granted_mode); | |
1030 | ||
1031 | search_granted_lock(&lock->l_resource->lr_granted, lock, &prev); | |
1032 | ldlm_granted_list_add_lock(lock, &prev); | |
d7e09d03 PT |
1033 | } |
1034 | ||
1035 | /** | |
1036 | * Perform lock granting bookkeeping. | |
1037 | * | |
1038 | * Includes putting the lock into granted list and updating lock mode. | |
1039 | * NOTE: called by | |
1040 | * - ldlm_lock_enqueue | |
1041 | * - ldlm_reprocess_queue | |
1042 | * - ldlm_lock_convert | |
1043 | * | |
1044 | * must be called with lr_lock held | |
1045 | */ | |
1046 | void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list) | |
1047 | { | |
1048 | struct ldlm_resource *res = lock->l_resource; | |
d7e09d03 PT |
1049 | |
1050 | check_res_locked(res); | |
1051 | ||
1052 | lock->l_granted_mode = lock->l_req_mode; | |
1053 | if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) | |
1054 | ldlm_grant_lock_with_skiplist(lock); | |
1055 | else if (res->lr_type == LDLM_EXTENT) | |
1056 | ldlm_extent_add_lock(res, lock); | |
1057 | else | |
1058 | ldlm_resource_add_lock(res, &res->lr_granted, lock); | |
1059 | ||
1060 | if (lock->l_granted_mode < res->lr_most_restr) | |
1061 | res->lr_most_restr = lock->l_granted_mode; | |
1062 | ||
1063 | if (work_list && lock->l_completion_ast != NULL) | |
1064 | ldlm_add_ast_work_item(lock, NULL, work_list); | |
1065 | ||
1066 | ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock); | |
d7e09d03 PT |
1067 | } |
1068 | ||
1069 | /** | |
1070 | * Search for a lock with given properties in a queue. | |
1071 | * | |
1072 | * \retval a referenced lock or NULL. See the flag descriptions below, in the | |
1073 | * comment above ldlm_lock_match | |
1074 | */ | |
1075 | static struct ldlm_lock *search_queue(struct list_head *queue, | |
1076 | ldlm_mode_t *mode, | |
1077 | ldlm_policy_data_t *policy, | |
1078 | struct ldlm_lock *old_lock, | |
1079 | __u64 flags, int unref) | |
1080 | { | |
1081 | struct ldlm_lock *lock; | |
1082 | struct list_head *tmp; | |
1083 | ||
1084 | list_for_each(tmp, queue) { | |
1085 | ldlm_mode_t match; | |
1086 | ||
1087 | lock = list_entry(tmp, struct ldlm_lock, l_res_link); | |
1088 | ||
1089 | if (lock == old_lock) | |
1090 | break; | |
1091 | ||
d3a8a4e2 JX |
1092 | /* Check if this lock can be matched. |
1093 | * Used by LU-2919(exclusive open) for open lease lock */ | |
1094 | if (ldlm_is_excl(lock)) | |
1095 | continue; | |
1096 | ||
d7e09d03 PT |
1097 | /* llite sometimes wants to match locks that will be |
1098 | * canceled when their users drop, but we allow it to match | |
1099 | * if it passes in CBPENDING and the lock still has users. | |
1100 | * this is generally only going to be used by children | |
1101 | * whose parents already hold a lock so forward progress | |
1102 | * can still happen. */ | |
1103 | if (lock->l_flags & LDLM_FL_CBPENDING && | |
1104 | !(flags & LDLM_FL_CBPENDING)) | |
1105 | continue; | |
1106 | if (!unref && lock->l_flags & LDLM_FL_CBPENDING && | |
1107 | lock->l_readers == 0 && lock->l_writers == 0) | |
1108 | continue; | |
1109 | ||
1110 | if (!(lock->l_req_mode & *mode)) | |
1111 | continue; | |
1112 | match = lock->l_req_mode; | |
1113 | ||
1114 | if (lock->l_resource->lr_type == LDLM_EXTENT && | |
1115 | (lock->l_policy_data.l_extent.start > | |
1116 | policy->l_extent.start || | |
1117 | lock->l_policy_data.l_extent.end < policy->l_extent.end)) | |
1118 | continue; | |
1119 | ||
1120 | if (unlikely(match == LCK_GROUP) && | |
1121 | lock->l_resource->lr_type == LDLM_EXTENT && | |
1122 | lock->l_policy_data.l_extent.gid != policy->l_extent.gid) | |
1123 | continue; | |
1124 | ||
1125 | /* We match if we have existing lock with same or wider set | |
1126 | of bits. */ | |
1127 | if (lock->l_resource->lr_type == LDLM_IBITS && | |
1128 | ((lock->l_policy_data.l_inodebits.bits & | |
1129 | policy->l_inodebits.bits) != | |
1130 | policy->l_inodebits.bits)) | |
1131 | continue; | |
1132 | ||
f2145eae | 1133 | if (!unref && (lock->l_flags & LDLM_FL_GONE_MASK)) |
d7e09d03 PT |
1134 | continue; |
1135 | ||
1136 | if ((flags & LDLM_FL_LOCAL_ONLY) && | |
1137 | !(lock->l_flags & LDLM_FL_LOCAL)) | |
1138 | continue; | |
1139 | ||
1140 | if (flags & LDLM_FL_TEST_LOCK) { | |
1141 | LDLM_LOCK_GET(lock); | |
1142 | ldlm_lock_touch_in_lru(lock); | |
1143 | } else { | |
1144 | ldlm_lock_addref_internal_nolock(lock, match); | |
1145 | } | |
1146 | *mode = match; | |
1147 | return lock; | |
1148 | } | |
1149 | ||
1150 | return NULL; | |
1151 | } | |
1152 | ||
1153 | void ldlm_lock_fail_match_locked(struct ldlm_lock *lock) | |
1154 | { | |
f2145eae BK |
1155 | if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) { |
1156 | lock->l_flags |= LDLM_FL_FAIL_NOTIFIED; | |
d7e09d03 PT |
1157 | wake_up_all(&lock->l_waitq); |
1158 | } | |
1159 | } | |
1160 | EXPORT_SYMBOL(ldlm_lock_fail_match_locked); | |
1161 | ||
1162 | void ldlm_lock_fail_match(struct ldlm_lock *lock) | |
1163 | { | |
1164 | lock_res_and_lock(lock); | |
1165 | ldlm_lock_fail_match_locked(lock); | |
1166 | unlock_res_and_lock(lock); | |
1167 | } | |
1168 | EXPORT_SYMBOL(ldlm_lock_fail_match); | |
1169 | ||
1170 | /** | |
1171 | * Mark lock as "matchable" by OST. | |
1172 | * | |
1173 | * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB | |
1174 | * is not yet valid. | |
1175 | * Assumes LDLM lock is already locked. | |
1176 | */ | |
1177 | void ldlm_lock_allow_match_locked(struct ldlm_lock *lock) | |
1178 | { | |
1179 | lock->l_flags |= LDLM_FL_LVB_READY; | |
1180 | wake_up_all(&lock->l_waitq); | |
1181 | } | |
1182 | EXPORT_SYMBOL(ldlm_lock_allow_match_locked); | |
1183 | ||
1184 | /** | |
1185 | * Mark lock as "matchable" by OST. | |
1186 | * Locks the lock and then \see ldlm_lock_allow_match_locked | |
1187 | */ | |
1188 | void ldlm_lock_allow_match(struct ldlm_lock *lock) | |
1189 | { | |
1190 | lock_res_and_lock(lock); | |
1191 | ldlm_lock_allow_match_locked(lock); | |
1192 | unlock_res_and_lock(lock); | |
1193 | } | |
1194 | EXPORT_SYMBOL(ldlm_lock_allow_match); | |
1195 | ||
1196 | /** | |
1197 | * Attempt to find a lock with specified properties. | |
1198 | * | |
1199 | * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is | |
1200 | * set in \a flags | |
1201 | * | |
1202 | * Can be called in two ways: | |
1203 | * | |
1204 | * If 'ns' is NULL, then lockh describes an existing lock that we want to look | |
1205 | * for a duplicate of. | |
1206 | * | |
1207 | * Otherwise, all of the fields must be filled in, to match against. | |
1208 | * | |
1209 | * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the | |
1210 | * server (ie, connh is NULL) | |
1211 | * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted | |
1212 | * list will be considered | |
1213 | * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked | |
1214 | * to be canceled can still be matched as long as they still have reader | |
6e3dd654 | 1215 | * or writer referneces |
d7e09d03 PT |
1216 | * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock, |
1217 | * just tell us if we would have matched. | |
1218 | * | |
1219 | * \retval 1 if it finds an already-existing lock that is compatible; in this | |
1220 | * case, lockh is filled in with a addref()ed lock | |
1221 | * | |
1222 | * We also check security context, and if that fails we simply return 0 (to | |
1223 | * keep caller code unchanged), the context failure will be discovered by | |
1224 | * caller sometime later. | |
1225 | */ | |
1226 | ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags, | |
1227 | const struct ldlm_res_id *res_id, ldlm_type_t type, | |
1228 | ldlm_policy_data_t *policy, ldlm_mode_t mode, | |
1229 | struct lustre_handle *lockh, int unref) | |
1230 | { | |
1231 | struct ldlm_resource *res; | |
1232 | struct ldlm_lock *lock, *old_lock = NULL; | |
1233 | int rc = 0; | |
d7e09d03 PT |
1234 | |
1235 | if (ns == NULL) { | |
1236 | old_lock = ldlm_handle2lock(lockh); | |
1237 | LASSERT(old_lock); | |
1238 | ||
1239 | ns = ldlm_lock_to_ns(old_lock); | |
1240 | res_id = &old_lock->l_resource->lr_name; | |
1241 | type = old_lock->l_resource->lr_type; | |
1242 | mode = old_lock->l_req_mode; | |
1243 | } | |
1244 | ||
1245 | res = ldlm_resource_get(ns, NULL, res_id, type, 0); | |
1246 | if (res == NULL) { | |
1247 | LASSERT(old_lock == NULL); | |
0a3bdb00 | 1248 | return 0; |
d7e09d03 PT |
1249 | } |
1250 | ||
1251 | LDLM_RESOURCE_ADDREF(res); | |
1252 | lock_res(res); | |
1253 | ||
1254 | lock = search_queue(&res->lr_granted, &mode, policy, old_lock, | |
1255 | flags, unref); | |
1256 | if (lock != NULL) | |
1257 | GOTO(out, rc = 1); | |
1258 | if (flags & LDLM_FL_BLOCK_GRANTED) | |
1259 | GOTO(out, rc = 0); | |
1260 | lock = search_queue(&res->lr_converting, &mode, policy, old_lock, | |
1261 | flags, unref); | |
1262 | if (lock != NULL) | |
1263 | GOTO(out, rc = 1); | |
1264 | lock = search_queue(&res->lr_waiting, &mode, policy, old_lock, | |
1265 | flags, unref); | |
1266 | if (lock != NULL) | |
1267 | GOTO(out, rc = 1); | |
1268 | ||
d7e09d03 PT |
1269 | out: |
1270 | unlock_res(res); | |
1271 | LDLM_RESOURCE_DELREF(res); | |
1272 | ldlm_resource_putref(res); | |
1273 | ||
1274 | if (lock) { | |
1275 | ldlm_lock2handle(lock, lockh); | |
1276 | if ((flags & LDLM_FL_LVB_READY) && | |
1277 | (!(lock->l_flags & LDLM_FL_LVB_READY))) { | |
f2145eae BK |
1278 | __u64 wait_flags = LDLM_FL_LVB_READY | |
1279 | LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED; | |
d7e09d03 PT |
1280 | struct l_wait_info lwi; |
1281 | if (lock->l_completion_ast) { | |
1282 | int err = lock->l_completion_ast(lock, | |
1283 | LDLM_FL_WAIT_NOREPROC, | |
1284 | NULL); | |
1285 | if (err) { | |
1286 | if (flags & LDLM_FL_TEST_LOCK) | |
1287 | LDLM_LOCK_RELEASE(lock); | |
1288 | else | |
1289 | ldlm_lock_decref_internal(lock, | |
1290 | mode); | |
1291 | rc = 0; | |
1292 | goto out2; | |
1293 | } | |
1294 | } | |
1295 | ||
1296 | lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout), | |
1297 | NULL, LWI_ON_SIGNAL_NOOP, NULL); | |
1298 | ||
1299 | /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */ | |
1300 | l_wait_event(lock->l_waitq, | |
f2145eae | 1301 | lock->l_flags & wait_flags, |
d7e09d03 PT |
1302 | &lwi); |
1303 | if (!(lock->l_flags & LDLM_FL_LVB_READY)) { | |
1304 | if (flags & LDLM_FL_TEST_LOCK) | |
1305 | LDLM_LOCK_RELEASE(lock); | |
1306 | else | |
1307 | ldlm_lock_decref_internal(lock, mode); | |
1308 | rc = 0; | |
1309 | } | |
1310 | } | |
1311 | } | |
1312 | out2: | |
1313 | if (rc) { | |
b0f5aad5 | 1314 | LDLM_DEBUG(lock, "matched (%llu %llu)", |
d7e09d03 PT |
1315 | (type == LDLM_PLAIN || type == LDLM_IBITS) ? |
1316 | res_id->name[2] : policy->l_extent.start, | |
1317 | (type == LDLM_PLAIN || type == LDLM_IBITS) ? | |
1318 | res_id->name[3] : policy->l_extent.end); | |
1319 | ||
1320 | /* check user's security context */ | |
1321 | if (lock->l_conn_export && | |
1322 | sptlrpc_import_check_ctx( | |
1323 | class_exp2cliimp(lock->l_conn_export))) { | |
1324 | if (!(flags & LDLM_FL_TEST_LOCK)) | |
1325 | ldlm_lock_decref_internal(lock, mode); | |
1326 | rc = 0; | |
1327 | } | |
1328 | ||
1329 | if (flags & LDLM_FL_TEST_LOCK) | |
1330 | LDLM_LOCK_RELEASE(lock); | |
1331 | ||
1332 | } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/ | |
b0f5aad5 GKH |
1333 | LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res %llu/%llu (%llu %llu)", |
1334 | ns, type, mode, res_id->name[0], res_id->name[1], | |
d7e09d03 PT |
1335 | (type == LDLM_PLAIN || type == LDLM_IBITS) ? |
1336 | res_id->name[2] :policy->l_extent.start, | |
1337 | (type == LDLM_PLAIN || type == LDLM_IBITS) ? | |
1338 | res_id->name[3] : policy->l_extent.end); | |
1339 | } | |
1340 | if (old_lock) | |
1341 | LDLM_LOCK_PUT(old_lock); | |
1342 | ||
1343 | return rc ? mode : 0; | |
1344 | } | |
1345 | EXPORT_SYMBOL(ldlm_lock_match); | |
1346 | ||
1347 | ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh, | |
1348 | __u64 *bits) | |
1349 | { | |
1350 | struct ldlm_lock *lock; | |
1351 | ldlm_mode_t mode = 0; | |
d7e09d03 PT |
1352 | |
1353 | lock = ldlm_handle2lock(lockh); | |
1354 | if (lock != NULL) { | |
1355 | lock_res_and_lock(lock); | |
f2145eae | 1356 | if (lock->l_flags & LDLM_FL_GONE_MASK) |
d7e09d03 PT |
1357 | GOTO(out, mode); |
1358 | ||
1359 | if (lock->l_flags & LDLM_FL_CBPENDING && | |
1360 | lock->l_readers == 0 && lock->l_writers == 0) | |
1361 | GOTO(out, mode); | |
1362 | ||
1363 | if (bits) | |
1364 | *bits = lock->l_policy_data.l_inodebits.bits; | |
1365 | mode = lock->l_granted_mode; | |
1366 | ldlm_lock_addref_internal_nolock(lock, mode); | |
1367 | } | |
1368 | ||
d7e09d03 PT |
1369 | out: |
1370 | if (lock != NULL) { | |
1371 | unlock_res_and_lock(lock); | |
1372 | LDLM_LOCK_PUT(lock); | |
1373 | } | |
1374 | return mode; | |
1375 | } | |
1376 | EXPORT_SYMBOL(ldlm_revalidate_lock_handle); | |
1377 | ||
1378 | /** The caller must guarantee that the buffer is large enough. */ | |
1379 | int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill, | |
1380 | enum req_location loc, void *data, int size) | |
1381 | { | |
1382 | void *lvb; | |
d7e09d03 PT |
1383 | |
1384 | LASSERT(data != NULL); | |
1385 | LASSERT(size >= 0); | |
1386 | ||
1387 | switch (lock->l_lvb_type) { | |
1388 | case LVB_T_OST: | |
1389 | if (size == sizeof(struct ost_lvb)) { | |
1390 | if (loc == RCL_CLIENT) | |
1391 | lvb = req_capsule_client_swab_get(pill, | |
1392 | &RMF_DLM_LVB, | |
1393 | lustre_swab_ost_lvb); | |
1394 | else | |
1395 | lvb = req_capsule_server_swab_get(pill, | |
1396 | &RMF_DLM_LVB, | |
1397 | lustre_swab_ost_lvb); | |
1398 | if (unlikely(lvb == NULL)) { | |
1399 | LDLM_ERROR(lock, "no LVB"); | |
0a3bdb00 | 1400 | return -EPROTO; |
d7e09d03 PT |
1401 | } |
1402 | ||
1403 | memcpy(data, lvb, size); | |
1404 | } else if (size == sizeof(struct ost_lvb_v1)) { | |
1405 | struct ost_lvb *olvb = data; | |
1406 | ||
1407 | if (loc == RCL_CLIENT) | |
1408 | lvb = req_capsule_client_swab_get(pill, | |
1409 | &RMF_DLM_LVB, | |
1410 | lustre_swab_ost_lvb_v1); | |
1411 | else | |
1412 | lvb = req_capsule_server_sized_swab_get(pill, | |
1413 | &RMF_DLM_LVB, size, | |
1414 | lustre_swab_ost_lvb_v1); | |
1415 | if (unlikely(lvb == NULL)) { | |
1416 | LDLM_ERROR(lock, "no LVB"); | |
0a3bdb00 | 1417 | return -EPROTO; |
d7e09d03 PT |
1418 | } |
1419 | ||
1420 | memcpy(data, lvb, size); | |
1421 | olvb->lvb_mtime_ns = 0; | |
1422 | olvb->lvb_atime_ns = 0; | |
1423 | olvb->lvb_ctime_ns = 0; | |
1424 | } else { | |
1425 | LDLM_ERROR(lock, "Replied unexpected ost LVB size %d", | |
1426 | size); | |
0a3bdb00 | 1427 | return -EINVAL; |
d7e09d03 PT |
1428 | } |
1429 | break; | |
1430 | case LVB_T_LQUOTA: | |
1431 | if (size == sizeof(struct lquota_lvb)) { | |
1432 | if (loc == RCL_CLIENT) | |
1433 | lvb = req_capsule_client_swab_get(pill, | |
1434 | &RMF_DLM_LVB, | |
1435 | lustre_swab_lquota_lvb); | |
1436 | else | |
1437 | lvb = req_capsule_server_swab_get(pill, | |
1438 | &RMF_DLM_LVB, | |
1439 | lustre_swab_lquota_lvb); | |
1440 | if (unlikely(lvb == NULL)) { | |
1441 | LDLM_ERROR(lock, "no LVB"); | |
0a3bdb00 | 1442 | return -EPROTO; |
d7e09d03 PT |
1443 | } |
1444 | ||
1445 | memcpy(data, lvb, size); | |
1446 | } else { | |
1447 | LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d", | |
1448 | size); | |
0a3bdb00 | 1449 | return -EINVAL; |
d7e09d03 PT |
1450 | } |
1451 | break; | |
1452 | case LVB_T_LAYOUT: | |
1453 | if (size == 0) | |
1454 | break; | |
1455 | ||
1456 | if (loc == RCL_CLIENT) | |
1457 | lvb = req_capsule_client_get(pill, &RMF_DLM_LVB); | |
1458 | else | |
1459 | lvb = req_capsule_server_get(pill, &RMF_DLM_LVB); | |
1460 | if (unlikely(lvb == NULL)) { | |
1461 | LDLM_ERROR(lock, "no LVB"); | |
0a3bdb00 | 1462 | return -EPROTO; |
d7e09d03 PT |
1463 | } |
1464 | ||
1465 | memcpy(data, lvb, size); | |
1466 | break; | |
1467 | default: | |
1468 | LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type); | |
5d4450c4 | 1469 | dump_stack(); |
0a3bdb00 | 1470 | return -EINVAL; |
d7e09d03 PT |
1471 | } |
1472 | ||
0a3bdb00 | 1473 | return 0; |
d7e09d03 PT |
1474 | } |
1475 | ||
1476 | /** | |
1477 | * Create and fill in new LDLM lock with specified properties. | |
1478 | * Returns a referenced lock | |
1479 | */ | |
1480 | struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns, | |
1481 | const struct ldlm_res_id *res_id, | |
1482 | ldlm_type_t type, | |
1483 | ldlm_mode_t mode, | |
1484 | const struct ldlm_callback_suite *cbs, | |
1485 | void *data, __u32 lvb_len, | |
1486 | enum lvb_type lvb_type) | |
1487 | { | |
1488 | struct ldlm_lock *lock; | |
1489 | struct ldlm_resource *res; | |
d7e09d03 PT |
1490 | |
1491 | res = ldlm_resource_get(ns, NULL, res_id, type, 1); | |
1492 | if (res == NULL) | |
0a3bdb00 | 1493 | return NULL; |
d7e09d03 PT |
1494 | |
1495 | lock = ldlm_lock_new(res); | |
1496 | ||
1497 | if (lock == NULL) | |
0a3bdb00 | 1498 | return NULL; |
d7e09d03 PT |
1499 | |
1500 | lock->l_req_mode = mode; | |
1501 | lock->l_ast_data = data; | |
1502 | lock->l_pid = current_pid(); | |
f2145eae BK |
1503 | if (ns_is_server(ns)) |
1504 | lock->l_flags |= LDLM_FL_NS_SRV; | |
d7e09d03 PT |
1505 | if (cbs) { |
1506 | lock->l_blocking_ast = cbs->lcs_blocking; | |
1507 | lock->l_completion_ast = cbs->lcs_completion; | |
1508 | lock->l_glimpse_ast = cbs->lcs_glimpse; | |
d7e09d03 PT |
1509 | } |
1510 | ||
1511 | lock->l_tree_node = NULL; | |
1512 | /* if this is the extent lock, allocate the interval tree node */ | |
1513 | if (type == LDLM_EXTENT) { | |
1514 | if (ldlm_interval_alloc(lock) == NULL) | |
1515 | GOTO(out, 0); | |
1516 | } | |
1517 | ||
1518 | if (lvb_len) { | |
1519 | lock->l_lvb_len = lvb_len; | |
1520 | OBD_ALLOC(lock->l_lvb_data, lvb_len); | |
1521 | if (lock->l_lvb_data == NULL) | |
1522 | GOTO(out, 0); | |
1523 | } | |
1524 | ||
1525 | lock->l_lvb_type = lvb_type; | |
1526 | if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK)) | |
1527 | GOTO(out, 0); | |
1528 | ||
0a3bdb00 | 1529 | return lock; |
d7e09d03 PT |
1530 | |
1531 | out: | |
1532 | ldlm_lock_destroy(lock); | |
1533 | LDLM_LOCK_RELEASE(lock); | |
1534 | return NULL; | |
1535 | } | |
1536 | ||
1537 | /** | |
1538 | * Enqueue (request) a lock. | |
1539 | * | |
1540 | * Does not block. As a result of enqueue the lock would be put | |
1541 | * into granted or waiting list. | |
1542 | * | |
1543 | * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag | |
1544 | * set, skip all the enqueueing and delegate lock processing to intent policy | |
1545 | * function. | |
1546 | */ | |
1547 | ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, | |
1548 | struct ldlm_lock **lockp, | |
1549 | void *cookie, __u64 *flags) | |
1550 | { | |
1551 | struct ldlm_lock *lock = *lockp; | |
1552 | struct ldlm_resource *res = lock->l_resource; | |
1553 | int local = ns_is_client(ldlm_res_to_ns(res)); | |
1554 | ldlm_error_t rc = ELDLM_OK; | |
1555 | struct ldlm_interval *node = NULL; | |
d7e09d03 | 1556 | |
7264b8a5 | 1557 | lock->l_last_activity = get_seconds(); |
d7e09d03 PT |
1558 | /* policies are not executed on the client or during replay */ |
1559 | if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT | |
1560 | && !local && ns->ns_policy) { | |
1561 | rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags, | |
1562 | NULL); | |
1563 | if (rc == ELDLM_LOCK_REPLACED) { | |
1564 | /* The lock that was returned has already been granted, | |
1565 | * and placed into lockp. If it's not the same as the | |
1566 | * one we passed in, then destroy the old one and our | |
1567 | * work here is done. */ | |
1568 | if (lock != *lockp) { | |
1569 | ldlm_lock_destroy(lock); | |
1570 | LDLM_LOCK_RELEASE(lock); | |
1571 | } | |
1572 | *flags |= LDLM_FL_LOCK_CHANGED; | |
0a3bdb00 | 1573 | return 0; |
d7e09d03 PT |
1574 | } else if (rc != ELDLM_OK || |
1575 | (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) { | |
1576 | ldlm_lock_destroy(lock); | |
0a3bdb00 | 1577 | return rc; |
d7e09d03 PT |
1578 | } |
1579 | } | |
1580 | ||
1581 | /* For a replaying lock, it might be already in granted list. So | |
1582 | * unlinking the lock will cause the interval node to be freed, we | |
1583 | * have to allocate the interval node early otherwise we can't regrant | |
1584 | * this lock in the future. - jay */ | |
1585 | if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT) | |
0be19afa | 1586 | OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS); |
d7e09d03 PT |
1587 | |
1588 | lock_res_and_lock(lock); | |
1589 | if (local && lock->l_req_mode == lock->l_granted_mode) { | |
1590 | /* The server returned a blocked lock, but it was granted | |
1591 | * before we got a chance to actually enqueue it. We don't | |
1592 | * need to do anything else. */ | |
1593 | *flags &= ~(LDLM_FL_BLOCK_GRANTED | | |
1594 | LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT); | |
1595 | GOTO(out, ELDLM_OK); | |
1596 | } | |
1597 | ||
1598 | ldlm_resource_unlink_lock(lock); | |
1599 | if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) { | |
1600 | if (node == NULL) { | |
1601 | ldlm_lock_destroy_nolock(lock); | |
1602 | GOTO(out, rc = -ENOMEM); | |
1603 | } | |
1604 | ||
1605 | INIT_LIST_HEAD(&node->li_group); | |
1606 | ldlm_interval_attach(node, lock); | |
1607 | node = NULL; | |
1608 | } | |
1609 | ||
1610 | /* Some flags from the enqueue want to make it into the AST, via the | |
1611 | * lock's l_flags. */ | |
f2145eae | 1612 | lock->l_flags |= *flags & LDLM_FL_AST_DISCARD_DATA; |
d7e09d03 PT |
1613 | |
1614 | /* This distinction between local lock trees is very important; a client | |
1615 | * namespace only has information about locks taken by that client, and | |
1616 | * thus doesn't have enough information to decide for itself if it can | |
1617 | * be granted (below). In this case, we do exactly what the server | |
1618 | * tells us to do, as dictated by the 'flags'. | |
1619 | * | |
1620 | * We do exactly the same thing during recovery, when the server is | |
1621 | * more or less trusting the clients not to lie. | |
1622 | * | |
1623 | * FIXME (bug 268): Detect obvious lies by checking compatibility in | |
1624 | * granted/converting queues. */ | |
1625 | if (local) { | |
1626 | if (*flags & LDLM_FL_BLOCK_CONV) | |
1627 | ldlm_resource_add_lock(res, &res->lr_converting, lock); | |
1628 | else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED)) | |
1629 | ldlm_resource_add_lock(res, &res->lr_waiting, lock); | |
1630 | else | |
1631 | ldlm_grant_lock(lock, NULL); | |
1632 | GOTO(out, ELDLM_OK); | |
1633 | } else { | |
1634 | CERROR("This is client-side-only module, cannot handle " | |
1635 | "LDLM_NAMESPACE_SERVER resource type lock.\n"); | |
1636 | LBUG(); | |
1637 | } | |
1638 | ||
1639 | out: | |
1640 | unlock_res_and_lock(lock); | |
1641 | if (node) | |
1642 | OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node)); | |
1643 | return rc; | |
1644 | } | |
1645 | ||
1646 | ||
1647 | /** | |
1648 | * Process a call to blocking AST callback for a lock in ast_work list | |
1649 | */ | |
1650 | static int | |
1651 | ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) | |
1652 | { | |
1653 | struct ldlm_cb_set_arg *arg = opaq; | |
1654 | struct ldlm_lock_desc d; | |
1655 | int rc; | |
1656 | struct ldlm_lock *lock; | |
d7e09d03 PT |
1657 | |
1658 | if (list_empty(arg->list)) | |
0a3bdb00 | 1659 | return -ENOENT; |
d7e09d03 PT |
1660 | |
1661 | lock = list_entry(arg->list->next, struct ldlm_lock, l_bl_ast); | |
1662 | ||
1663 | /* nobody should touch l_bl_ast */ | |
1664 | lock_res_and_lock(lock); | |
1665 | list_del_init(&lock->l_bl_ast); | |
1666 | ||
1667 | LASSERT(lock->l_flags & LDLM_FL_AST_SENT); | |
1668 | LASSERT(lock->l_bl_ast_run == 0); | |
1669 | LASSERT(lock->l_blocking_lock); | |
1670 | lock->l_bl_ast_run++; | |
1671 | unlock_res_and_lock(lock); | |
1672 | ||
1673 | ldlm_lock2desc(lock->l_blocking_lock, &d); | |
1674 | ||
1675 | rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING); | |
1676 | LDLM_LOCK_RELEASE(lock->l_blocking_lock); | |
1677 | lock->l_blocking_lock = NULL; | |
1678 | LDLM_LOCK_RELEASE(lock); | |
1679 | ||
0a3bdb00 | 1680 | return rc; |
d7e09d03 PT |
1681 | } |
1682 | ||
1683 | /** | |
1684 | * Process a call to completion AST callback for a lock in ast_work list | |
1685 | */ | |
1686 | static int | |
1687 | ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) | |
1688 | { | |
1689 | struct ldlm_cb_set_arg *arg = opaq; | |
1690 | int rc = 0; | |
1691 | struct ldlm_lock *lock; | |
1692 | ldlm_completion_callback completion_callback; | |
d7e09d03 PT |
1693 | |
1694 | if (list_empty(arg->list)) | |
0a3bdb00 | 1695 | return -ENOENT; |
d7e09d03 PT |
1696 | |
1697 | lock = list_entry(arg->list->next, struct ldlm_lock, l_cp_ast); | |
1698 | ||
1699 | /* It's possible to receive a completion AST before we've set | |
1700 | * the l_completion_ast pointer: either because the AST arrived | |
1701 | * before the reply, or simply because there's a small race | |
1702 | * window between receiving the reply and finishing the local | |
1703 | * enqueue. (bug 842) | |
1704 | * | |
1705 | * This can't happen with the blocking_ast, however, because we | |
1706 | * will never call the local blocking_ast until we drop our | |
1707 | * reader/writer reference, which we won't do until we get the | |
1708 | * reply and finish enqueueing. */ | |
1709 | ||
1710 | /* nobody should touch l_cp_ast */ | |
1711 | lock_res_and_lock(lock); | |
1712 | list_del_init(&lock->l_cp_ast); | |
1713 | LASSERT(lock->l_flags & LDLM_FL_CP_REQD); | |
1714 | /* save l_completion_ast since it can be changed by | |
1715 | * mds_intent_policy(), see bug 14225 */ | |
1716 | completion_callback = lock->l_completion_ast; | |
1717 | lock->l_flags &= ~LDLM_FL_CP_REQD; | |
1718 | unlock_res_and_lock(lock); | |
1719 | ||
1720 | if (completion_callback != NULL) | |
1721 | rc = completion_callback(lock, 0, (void *)arg); | |
1722 | LDLM_LOCK_RELEASE(lock); | |
1723 | ||
0a3bdb00 | 1724 | return rc; |
d7e09d03 PT |
1725 | } |
1726 | ||
1727 | /** | |
1728 | * Process a call to revocation AST callback for a lock in ast_work list | |
1729 | */ | |
1730 | static int | |
1731 | ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) | |
1732 | { | |
1733 | struct ldlm_cb_set_arg *arg = opaq; | |
1734 | struct ldlm_lock_desc desc; | |
1735 | int rc; | |
1736 | struct ldlm_lock *lock; | |
d7e09d03 PT |
1737 | |
1738 | if (list_empty(arg->list)) | |
0a3bdb00 | 1739 | return -ENOENT; |
d7e09d03 PT |
1740 | |
1741 | lock = list_entry(arg->list->next, struct ldlm_lock, l_rk_ast); | |
1742 | list_del_init(&lock->l_rk_ast); | |
1743 | ||
1744 | /* the desc just pretend to exclusive */ | |
1745 | ldlm_lock2desc(lock, &desc); | |
1746 | desc.l_req_mode = LCK_EX; | |
1747 | desc.l_granted_mode = 0; | |
1748 | ||
1749 | rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING); | |
1750 | LDLM_LOCK_RELEASE(lock); | |
1751 | ||
0a3bdb00 | 1752 | return rc; |
d7e09d03 PT |
1753 | } |
1754 | ||
1755 | /** | |
1756 | * Process a call to glimpse AST callback for a lock in ast_work list | |
1757 | */ | |
1758 | int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) | |
1759 | { | |
1760 | struct ldlm_cb_set_arg *arg = opaq; | |
1761 | struct ldlm_glimpse_work *gl_work; | |
1762 | struct ldlm_lock *lock; | |
1763 | int rc = 0; | |
d7e09d03 PT |
1764 | |
1765 | if (list_empty(arg->list)) | |
0a3bdb00 | 1766 | return -ENOENT; |
d7e09d03 PT |
1767 | |
1768 | gl_work = list_entry(arg->list->next, struct ldlm_glimpse_work, | |
1769 | gl_list); | |
1770 | list_del_init(&gl_work->gl_list); | |
1771 | ||
1772 | lock = gl_work->gl_lock; | |
1773 | ||
1774 | /* transfer the glimpse descriptor to ldlm_cb_set_arg */ | |
1775 | arg->gl_desc = gl_work->gl_desc; | |
1776 | ||
1777 | /* invoke the actual glimpse callback */ | |
1778 | if (lock->l_glimpse_ast(lock, (void*)arg) == 0) | |
1779 | rc = 1; | |
1780 | ||
1781 | LDLM_LOCK_RELEASE(lock); | |
1782 | ||
1783 | if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0) | |
1784 | OBD_FREE_PTR(gl_work); | |
1785 | ||
0a3bdb00 | 1786 | return rc; |
d7e09d03 PT |
1787 | } |
1788 | ||
1789 | /** | |
1790 | * Process list of locks in need of ASTs being sent. | |
1791 | * | |
1792 | * Used on server to send multiple ASTs together instead of sending one by | |
1793 | * one. | |
1794 | */ | |
1795 | int ldlm_run_ast_work(struct ldlm_namespace *ns, struct list_head *rpc_list, | |
1796 | ldlm_desc_ast_t ast_type) | |
1797 | { | |
1798 | struct ldlm_cb_set_arg *arg; | |
1799 | set_producer_func work_ast_lock; | |
1800 | int rc; | |
1801 | ||
1802 | if (list_empty(rpc_list)) | |
0a3bdb00 | 1803 | return 0; |
d7e09d03 PT |
1804 | |
1805 | OBD_ALLOC_PTR(arg); | |
1806 | if (arg == NULL) | |
0a3bdb00 | 1807 | return -ENOMEM; |
d7e09d03 PT |
1808 | |
1809 | atomic_set(&arg->restart, 0); | |
1810 | arg->list = rpc_list; | |
1811 | ||
1812 | switch (ast_type) { | |
1813 | case LDLM_WORK_BL_AST: | |
1814 | arg->type = LDLM_BL_CALLBACK; | |
1815 | work_ast_lock = ldlm_work_bl_ast_lock; | |
1816 | break; | |
1817 | case LDLM_WORK_CP_AST: | |
1818 | arg->type = LDLM_CP_CALLBACK; | |
1819 | work_ast_lock = ldlm_work_cp_ast_lock; | |
1820 | break; | |
1821 | case LDLM_WORK_REVOKE_AST: | |
1822 | arg->type = LDLM_BL_CALLBACK; | |
1823 | work_ast_lock = ldlm_work_revoke_ast_lock; | |
1824 | break; | |
1825 | case LDLM_WORK_GL_AST: | |
1826 | arg->type = LDLM_GL_CALLBACK; | |
1827 | work_ast_lock = ldlm_work_gl_ast_lock; | |
1828 | break; | |
1829 | default: | |
1830 | LBUG(); | |
1831 | } | |
1832 | ||
1833 | /* We create a ptlrpc request set with flow control extension. | |
1834 | * This request set will use the work_ast_lock function to produce new | |
1835 | * requests and will send a new request each time one completes in order | |
1836 | * to keep the number of requests in flight to ns_max_parallel_ast */ | |
1837 | arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX, | |
1838 | work_ast_lock, arg); | |
1839 | if (arg->set == NULL) | |
1840 | GOTO(out, rc = -ENOMEM); | |
1841 | ||
1842 | ptlrpc_set_wait(arg->set); | |
1843 | ptlrpc_set_destroy(arg->set); | |
1844 | ||
1845 | rc = atomic_read(&arg->restart) ? -ERESTART : 0; | |
1846 | GOTO(out, rc); | |
1847 | out: | |
1848 | OBD_FREE_PTR(arg); | |
1849 | return rc; | |
1850 | } | |
1851 | ||
1852 | static int reprocess_one_queue(struct ldlm_resource *res, void *closure) | |
1853 | { | |
1854 | ldlm_reprocess_all(res); | |
1855 | return LDLM_ITER_CONTINUE; | |
1856 | } | |
1857 | ||
6da6eabe | 1858 | static int ldlm_reprocess_res(struct cfs_hash *hs, struct cfs_hash_bd *bd, |
d7e09d03 PT |
1859 | struct hlist_node *hnode, void *arg) |
1860 | { | |
1861 | struct ldlm_resource *res = cfs_hash_object(hs, hnode); | |
1862 | int rc; | |
1863 | ||
1864 | rc = reprocess_one_queue(res, arg); | |
1865 | ||
1866 | return rc == LDLM_ITER_STOP; | |
1867 | } | |
1868 | ||
1869 | /** | |
1870 | * Iterate through all resources on a namespace attempting to grant waiting | |
1871 | * locks. | |
1872 | */ | |
1873 | void ldlm_reprocess_all_ns(struct ldlm_namespace *ns) | |
1874 | { | |
d7e09d03 PT |
1875 | if (ns != NULL) { |
1876 | cfs_hash_for_each_nolock(ns->ns_rs_hash, | |
1877 | ldlm_reprocess_res, NULL); | |
1878 | } | |
d7e09d03 PT |
1879 | } |
1880 | EXPORT_SYMBOL(ldlm_reprocess_all_ns); | |
1881 | ||
1882 | /** | |
1883 | * Try to grant all waiting locks on a resource. | |
1884 | * | |
1885 | * Calls ldlm_reprocess_queue on converting and waiting queues. | |
1886 | * | |
1887 | * Typically called after some resource locks are cancelled to see | |
1888 | * if anything could be granted as a result of the cancellation. | |
1889 | */ | |
1890 | void ldlm_reprocess_all(struct ldlm_resource *res) | |
1891 | { | |
1892 | LIST_HEAD(rpc_list); | |
1893 | ||
d7e09d03 PT |
1894 | if (!ns_is_client(ldlm_res_to_ns(res))) { |
1895 | CERROR("This is client-side-only module, cannot handle " | |
1896 | "LDLM_NAMESPACE_SERVER resource type lock.\n"); | |
1897 | LBUG(); | |
1898 | } | |
d7e09d03 PT |
1899 | } |
1900 | ||
1901 | /** | |
1902 | * Helper function to call blocking AST for LDLM lock \a lock in a | |
1903 | * "cancelling" mode. | |
1904 | */ | |
1905 | void ldlm_cancel_callback(struct ldlm_lock *lock) | |
1906 | { | |
1907 | check_res_locked(lock->l_resource); | |
1908 | if (!(lock->l_flags & LDLM_FL_CANCEL)) { | |
1909 | lock->l_flags |= LDLM_FL_CANCEL; | |
1910 | if (lock->l_blocking_ast) { | |
1911 | unlock_res_and_lock(lock); | |
1912 | lock->l_blocking_ast(lock, NULL, lock->l_ast_data, | |
1913 | LDLM_CB_CANCELING); | |
1914 | lock_res_and_lock(lock); | |
1915 | } else { | |
1916 | LDLM_DEBUG(lock, "no blocking ast"); | |
1917 | } | |
1918 | } | |
1919 | lock->l_flags |= LDLM_FL_BL_DONE; | |
1920 | } | |
1921 | ||
1922 | /** | |
1923 | * Remove skiplist-enabled LDLM lock \a req from granted list | |
1924 | */ | |
1925 | void ldlm_unlink_lock_skiplist(struct ldlm_lock *req) | |
1926 | { | |
1927 | if (req->l_resource->lr_type != LDLM_PLAIN && | |
1928 | req->l_resource->lr_type != LDLM_IBITS) | |
1929 | return; | |
1930 | ||
1931 | list_del_init(&req->l_sl_policy); | |
1932 | list_del_init(&req->l_sl_mode); | |
1933 | } | |
1934 | ||
1935 | /** | |
1936 | * Attempts to cancel LDLM lock \a lock that has no reader/writer references. | |
1937 | */ | |
1938 | void ldlm_lock_cancel(struct ldlm_lock *lock) | |
1939 | { | |
1940 | struct ldlm_resource *res; | |
1941 | struct ldlm_namespace *ns; | |
d7e09d03 PT |
1942 | |
1943 | lock_res_and_lock(lock); | |
1944 | ||
1945 | res = lock->l_resource; | |
1946 | ns = ldlm_res_to_ns(res); | |
1947 | ||
1948 | /* Please do not, no matter how tempting, remove this LBUG without | |
1949 | * talking to me first. -phik */ | |
1950 | if (lock->l_readers || lock->l_writers) { | |
1951 | LDLM_ERROR(lock, "lock still has references"); | |
1952 | LBUG(); | |
1953 | } | |
1954 | ||
f2145eae | 1955 | if (lock->l_flags & LDLM_FL_WAITED) |
d7e09d03 PT |
1956 | ldlm_del_waiting_lock(lock); |
1957 | ||
1958 | /* Releases cancel callback. */ | |
1959 | ldlm_cancel_callback(lock); | |
1960 | ||
1961 | /* Yes, second time, just in case it was added again while we were | |
f2145eae BK |
1962 | * running with no res lock in ldlm_cancel_callback */ |
1963 | if (lock->l_flags & LDLM_FL_WAITED) | |
d7e09d03 PT |
1964 | ldlm_del_waiting_lock(lock); |
1965 | ||
1966 | ldlm_resource_unlink_lock(lock); | |
1967 | ldlm_lock_destroy_nolock(lock); | |
1968 | ||
1969 | if (lock->l_granted_mode == lock->l_req_mode) | |
1970 | ldlm_pool_del(&ns->ns_pool, lock); | |
1971 | ||
1972 | /* Make sure we will not be called again for same lock what is possible | |
1973 | * if not to zero out lock->l_granted_mode */ | |
1974 | lock->l_granted_mode = LCK_MINMODE; | |
1975 | unlock_res_and_lock(lock); | |
d7e09d03 PT |
1976 | } |
1977 | EXPORT_SYMBOL(ldlm_lock_cancel); | |
1978 | ||
1979 | /** | |
1980 | * Set opaque data into the lock that only makes sense to upper layer. | |
1981 | */ | |
1982 | int ldlm_lock_set_data(struct lustre_handle *lockh, void *data) | |
1983 | { | |
1984 | struct ldlm_lock *lock = ldlm_handle2lock(lockh); | |
1985 | int rc = -EINVAL; | |
d7e09d03 PT |
1986 | |
1987 | if (lock) { | |
1988 | if (lock->l_ast_data == NULL) | |
1989 | lock->l_ast_data = data; | |
1990 | if (lock->l_ast_data == data) | |
1991 | rc = 0; | |
1992 | LDLM_LOCK_PUT(lock); | |
1993 | } | |
0a3bdb00 | 1994 | return rc; |
d7e09d03 PT |
1995 | } |
1996 | EXPORT_SYMBOL(ldlm_lock_set_data); | |
1997 | ||
1998 | struct export_cl_data { | |
1999 | struct obd_export *ecl_exp; | |
2000 | int ecl_loop; | |
2001 | }; | |
2002 | ||
2003 | /** | |
2004 | * Iterator function for ldlm_cancel_locks_for_export. | |
2005 | * Cancels passed locks. | |
2006 | */ | |
6da6eabe | 2007 | int ldlm_cancel_locks_for_export_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd, |
d7e09d03 PT |
2008 | struct hlist_node *hnode, void *data) |
2009 | ||
2010 | { | |
2011 | struct export_cl_data *ecl = (struct export_cl_data *)data; | |
2012 | struct obd_export *exp = ecl->ecl_exp; | |
2013 | struct ldlm_lock *lock = cfs_hash_object(hs, hnode); | |
2014 | struct ldlm_resource *res; | |
2015 | ||
2016 | res = ldlm_resource_getref(lock->l_resource); | |
2017 | LDLM_LOCK_GET(lock); | |
2018 | ||
2019 | LDLM_DEBUG(lock, "export %p", exp); | |
2020 | ldlm_res_lvbo_update(res, NULL, 1); | |
2021 | ldlm_lock_cancel(lock); | |
2022 | ldlm_reprocess_all(res); | |
2023 | ldlm_resource_putref(res); | |
2024 | LDLM_LOCK_RELEASE(lock); | |
2025 | ||
2026 | ecl->ecl_loop++; | |
2027 | if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) { | |
2028 | CDEBUG(D_INFO, | |
2029 | "Cancel lock %p for export %p (loop %d), still have " | |
2030 | "%d locks left on hash table.\n", | |
2031 | lock, exp, ecl->ecl_loop, | |
2032 | atomic_read(&hs->hs_count)); | |
2033 | } | |
2034 | ||
2035 | return 0; | |
2036 | } | |
2037 | ||
2038 | /** | |
2039 | * Cancel all locks for given export. | |
2040 | * | |
2041 | * Typically called on client disconnection/eviction | |
2042 | */ | |
2043 | void ldlm_cancel_locks_for_export(struct obd_export *exp) | |
2044 | { | |
2045 | struct export_cl_data ecl = { | |
2046 | .ecl_exp = exp, | |
2047 | .ecl_loop = 0, | |
2048 | }; | |
2049 | ||
2050 | cfs_hash_for_each_empty(exp->exp_lock_hash, | |
2051 | ldlm_cancel_locks_for_export_cb, &ecl); | |
2052 | } | |
2053 | ||
2054 | /** | |
2055 | * Downgrade an exclusive lock. | |
2056 | * | |
6e3dd654 MI |
2057 | * A fast variant of ldlm_lock_convert for conversion of exclusive |
2058 | * locks. The conversion is always successful. | |
d7e09d03 PT |
2059 | * Used by Commit on Sharing (COS) code. |
2060 | * | |
2061 | * \param lock A lock to convert | |
2062 | * \param new_mode new lock mode | |
2063 | */ | |
2064 | void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode) | |
2065 | { | |
d7e09d03 PT |
2066 | LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX)); |
2067 | LASSERT(new_mode == LCK_COS); | |
2068 | ||
2069 | lock_res_and_lock(lock); | |
2070 | ldlm_resource_unlink_lock(lock); | |
2071 | /* | |
2072 | * Remove the lock from pool as it will be added again in | |
2073 | * ldlm_grant_lock() called below. | |
2074 | */ | |
2075 | ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock); | |
2076 | ||
2077 | lock->l_req_mode = new_mode; | |
2078 | ldlm_grant_lock(lock, NULL); | |
2079 | unlock_res_and_lock(lock); | |
2080 | ldlm_reprocess_all(lock->l_resource); | |
d7e09d03 PT |
2081 | } |
2082 | EXPORT_SYMBOL(ldlm_lock_downgrade); | |
2083 | ||
2084 | /** | |
2085 | * Attempt to convert already granted lock to a different mode. | |
2086 | * | |
2087 | * While lock conversion is not currently used, future client-side | |
2088 | * optimizations could take advantage of it to avoid discarding cached | |
2089 | * pages on a file. | |
2090 | */ | |
2091 | struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, | |
2092 | __u32 *flags) | |
2093 | { | |
2094 | LIST_HEAD(rpc_list); | |
2095 | struct ldlm_resource *res; | |
2096 | struct ldlm_namespace *ns; | |
2097 | int granted = 0; | |
2098 | struct ldlm_interval *node; | |
d7e09d03 PT |
2099 | |
2100 | /* Just return if mode is unchanged. */ | |
2101 | if (new_mode == lock->l_granted_mode) { | |
2102 | *flags |= LDLM_FL_BLOCK_GRANTED; | |
0a3bdb00 | 2103 | return lock->l_resource; |
d7e09d03 PT |
2104 | } |
2105 | ||
2106 | /* I can't check the type of lock here because the bitlock of lock | |
2107 | * is not held here, so do the allocation blindly. -jay */ | |
0be19afa | 2108 | OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS); |
2d58de78 LW |
2109 | if (node == NULL) |
2110 | /* Actually, this causes EDEADLOCK to be returned */ | |
0a3bdb00 | 2111 | return NULL; |
d7e09d03 PT |
2112 | |
2113 | LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR), | |
2114 | "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode); | |
2115 | ||
2116 | lock_res_and_lock(lock); | |
2117 | ||
2118 | res = lock->l_resource; | |
2119 | ns = ldlm_res_to_ns(res); | |
2120 | ||
2121 | lock->l_req_mode = new_mode; | |
2122 | if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) { | |
2123 | ldlm_resource_unlink_lock(lock); | |
2124 | } else { | |
2125 | ldlm_resource_unlink_lock(lock); | |
2126 | if (res->lr_type == LDLM_EXTENT) { | |
2127 | /* FIXME: ugly code, I have to attach the lock to a | |
2128 | * interval node again since perhaps it will be granted | |
2129 | * soon */ | |
2130 | INIT_LIST_HEAD(&node->li_group); | |
2131 | ldlm_interval_attach(node, lock); | |
2132 | node = NULL; | |
2133 | } | |
2134 | } | |
2135 | ||
2136 | /* | |
2137 | * Remove old lock from the pool before adding the lock with new | |
2138 | * mode below in ->policy() | |
2139 | */ | |
2140 | ldlm_pool_del(&ns->ns_pool, lock); | |
2141 | ||
2142 | /* If this is a local resource, put it on the appropriate list. */ | |
2143 | if (ns_is_client(ldlm_res_to_ns(res))) { | |
2144 | if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) { | |
2145 | ldlm_resource_add_lock(res, &res->lr_converting, lock); | |
2146 | } else { | |
2147 | /* This should never happen, because of the way the | |
2148 | * server handles conversions. */ | |
2149 | LDLM_ERROR(lock, "Erroneous flags %x on local lock\n", | |
2150 | *flags); | |
2151 | LBUG(); | |
2152 | ||
2153 | ldlm_grant_lock(lock, &rpc_list); | |
2154 | granted = 1; | |
2155 | /* FIXME: completion handling not with lr_lock held ! */ | |
2156 | if (lock->l_completion_ast) | |
2157 | lock->l_completion_ast(lock, 0, NULL); | |
2158 | } | |
2159 | } else { | |
2160 | CERROR("This is client-side-only module, cannot handle " | |
2161 | "LDLM_NAMESPACE_SERVER resource type lock.\n"); | |
2162 | LBUG(); | |
2163 | } | |
2164 | unlock_res_and_lock(lock); | |
2165 | ||
2166 | if (granted) | |
2167 | ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST); | |
2168 | if (node) | |
2169 | OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node)); | |
0a3bdb00 | 2170 | return res; |
d7e09d03 PT |
2171 | } |
2172 | EXPORT_SYMBOL(ldlm_lock_convert); | |
2173 | ||
2174 | /** | |
2175 | * Print lock with lock handle \a lockh description into debug log. | |
2176 | * | |
2177 | * Used when printing all locks on a resource for debug purposes. | |
2178 | */ | |
2179 | void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh) | |
2180 | { | |
2181 | struct ldlm_lock *lock; | |
2182 | ||
2183 | if (!((libcfs_debug | D_ERROR) & level)) | |
2184 | return; | |
2185 | ||
2186 | lock = ldlm_handle2lock(lockh); | |
2187 | if (lock == NULL) | |
2188 | return; | |
2189 | ||
2190 | LDLM_DEBUG_LIMIT(level, lock, "###"); | |
2191 | ||
2192 | LDLM_LOCK_PUT(lock); | |
2193 | } | |
2194 | EXPORT_SYMBOL(ldlm_lock_dump_handle); | |
2195 | ||
2196 | /** | |
2197 | * Print lock information with custom message into debug log. | |
2198 | * Helper function. | |
2199 | */ | |
2200 | void _ldlm_lock_debug(struct ldlm_lock *lock, | |
2201 | struct libcfs_debug_msg_data *msgdata, | |
2202 | const char *fmt, ...) | |
2203 | { | |
2204 | va_list args; | |
2205 | struct obd_export *exp = lock->l_export; | |
2206 | struct ldlm_resource *resource = lock->l_resource; | |
2207 | char *nid = "local"; | |
2208 | ||
2209 | va_start(args, fmt); | |
2210 | ||
2211 | if (exp && exp->exp_connection) { | |
2212 | nid = libcfs_nid2str(exp->exp_connection->c_peer.nid); | |
2213 | } else if (exp && exp->exp_obd != NULL) { | |
2214 | struct obd_import *imp = exp->exp_obd->u.cli.cl_import; | |
2215 | nid = libcfs_nid2str(imp->imp_connection->c_peer.nid); | |
2216 | } | |
2217 | ||
2218 | if (resource == NULL) { | |
2219 | libcfs_debug_vmsg2(msgdata, fmt, args, | |
55f5a824 GKH |
2220 | " ns: \?\? lock: %p/%#llx lrc: %d/%d,%d mode: %s/%s " |
2221 | "res: \?\? rrc=\?\? type: \?\?\? flags: %#llx nid: %s " | |
2222 | "remote: %#llx expref: %d pid: %u timeout: %lu " | |
d7e09d03 PT |
2223 | "lvb_type: %d\n", |
2224 | lock, | |
2225 | lock->l_handle.h_cookie, atomic_read(&lock->l_refc), | |
2226 | lock->l_readers, lock->l_writers, | |
2227 | ldlm_lockname[lock->l_granted_mode], | |
2228 | ldlm_lockname[lock->l_req_mode], | |
2229 | lock->l_flags, nid, lock->l_remote_handle.cookie, | |
2230 | exp ? atomic_read(&exp->exp_refcount) : -99, | |
2231 | lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type); | |
2232 | va_end(args); | |
2233 | return; | |
2234 | } | |
2235 | ||
2236 | switch (resource->lr_type) { | |
2237 | case LDLM_EXTENT: | |
2238 | libcfs_debug_vmsg2(msgdata, fmt, args, | |
55f5a824 | 2239 | " ns: %s lock: %p/%#llx lrc: %d/%d,%d mode: %s/%s " |
b0f5aad5 | 2240 | "res: "DLDLMRES" rrc: %d type: %s [%llu->%llu] " |
55f5a824 GKH |
2241 | "(req %llu->%llu) flags: %#llx nid: %s remote: " |
2242 | "%#llx expref: %d pid: %u timeout: %lu lvb_type: %d\n", | |
ce74f92d AD |
2243 | ldlm_lock_to_ns_name(lock), lock, |
2244 | lock->l_handle.h_cookie, atomic_read(&lock->l_refc), | |
2245 | lock->l_readers, lock->l_writers, | |
2246 | ldlm_lockname[lock->l_granted_mode], | |
2247 | ldlm_lockname[lock->l_req_mode], | |
2248 | PLDLMRES(resource), | |
2249 | atomic_read(&resource->lr_refcount), | |
2250 | ldlm_typename[resource->lr_type], | |
2251 | lock->l_policy_data.l_extent.start, | |
2252 | lock->l_policy_data.l_extent.end, | |
2253 | lock->l_req_extent.start, lock->l_req_extent.end, | |
2254 | lock->l_flags, nid, lock->l_remote_handle.cookie, | |
2255 | exp ? atomic_read(&exp->exp_refcount) : -99, | |
2256 | lock->l_pid, lock->l_callback_timeout, | |
2257 | lock->l_lvb_type); | |
d7e09d03 PT |
2258 | break; |
2259 | ||
2260 | case LDLM_FLOCK: | |
2261 | libcfs_debug_vmsg2(msgdata, fmt, args, | |
55f5a824 | 2262 | " ns: %s lock: %p/%#llx lrc: %d/%d,%d mode: %s/%s " |
ce74f92d | 2263 | "res: "DLDLMRES" rrc: %d type: %s pid: %d " |
55f5a824 GKH |
2264 | "[%llu->%llu] flags: %#llx nid: %s " |
2265 | "remote: %#llx expref: %d pid: %u timeout: %lu\n", | |
ce74f92d AD |
2266 | ldlm_lock_to_ns_name(lock), lock, |
2267 | lock->l_handle.h_cookie, atomic_read(&lock->l_refc), | |
2268 | lock->l_readers, lock->l_writers, | |
2269 | ldlm_lockname[lock->l_granted_mode], | |
2270 | ldlm_lockname[lock->l_req_mode], | |
2271 | PLDLMRES(resource), | |
2272 | atomic_read(&resource->lr_refcount), | |
2273 | ldlm_typename[resource->lr_type], | |
2274 | lock->l_policy_data.l_flock.pid, | |
2275 | lock->l_policy_data.l_flock.start, | |
2276 | lock->l_policy_data.l_flock.end, | |
2277 | lock->l_flags, nid, lock->l_remote_handle.cookie, | |
2278 | exp ? atomic_read(&exp->exp_refcount) : -99, | |
2279 | lock->l_pid, lock->l_callback_timeout); | |
d7e09d03 PT |
2280 | break; |
2281 | ||
2282 | case LDLM_IBITS: | |
2283 | libcfs_debug_vmsg2(msgdata, fmt, args, | |
55f5a824 GKH |
2284 | " ns: %s lock: %p/%#llx lrc: %d/%d,%d mode: %s/%s " |
2285 | "res: "DLDLMRES" bits %#llx rrc: %d type: %s " | |
2286 | "flags: %#llx nid: %s remote: %#llx expref: %d " | |
ce74f92d AD |
2287 | "pid: %u timeout: %lu lvb_type: %d\n", |
2288 | ldlm_lock_to_ns_name(lock), | |
2289 | lock, lock->l_handle.h_cookie, | |
2290 | atomic_read(&lock->l_refc), | |
2291 | lock->l_readers, lock->l_writers, | |
2292 | ldlm_lockname[lock->l_granted_mode], | |
2293 | ldlm_lockname[lock->l_req_mode], | |
2294 | PLDLMRES(resource), | |
2295 | lock->l_policy_data.l_inodebits.bits, | |
2296 | atomic_read(&resource->lr_refcount), | |
2297 | ldlm_typename[resource->lr_type], | |
2298 | lock->l_flags, nid, lock->l_remote_handle.cookie, | |
2299 | exp ? atomic_read(&exp->exp_refcount) : -99, | |
2300 | lock->l_pid, lock->l_callback_timeout, | |
2301 | lock->l_lvb_type); | |
d7e09d03 PT |
2302 | break; |
2303 | ||
2304 | default: | |
2305 | libcfs_debug_vmsg2(msgdata, fmt, args, | |
55f5a824 GKH |
2306 | " ns: %s lock: %p/%#llx lrc: %d/%d,%d mode: %s/%s " |
2307 | "res: "DLDLMRES" rrc: %d type: %s flags: %#llx " | |
2308 | "nid: %s remote: %#llx expref: %d pid: %u " | |
ce74f92d AD |
2309 | "timeout: %lu lvb_type: %d\n", |
2310 | ldlm_lock_to_ns_name(lock), | |
2311 | lock, lock->l_handle.h_cookie, | |
2312 | atomic_read(&lock->l_refc), | |
2313 | lock->l_readers, lock->l_writers, | |
2314 | ldlm_lockname[lock->l_granted_mode], | |
2315 | ldlm_lockname[lock->l_req_mode], | |
2316 | PLDLMRES(resource), | |
2317 | atomic_read(&resource->lr_refcount), | |
2318 | ldlm_typename[resource->lr_type], | |
2319 | lock->l_flags, nid, lock->l_remote_handle.cookie, | |
2320 | exp ? atomic_read(&exp->exp_refcount) : -99, | |
2321 | lock->l_pid, lock->l_callback_timeout, | |
2322 | lock->l_lvb_type); | |
d7e09d03 PT |
2323 | break; |
2324 | } | |
2325 | va_end(args); | |
2326 | } | |
2327 | EXPORT_SYMBOL(_ldlm_lock_debug); |