4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/ldlm/ldlm_lock.c
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Phil Schwan <phil@clusterfs.com>
42 #define DEBUG_SUBSYSTEM S_LDLM
44 #include <libcfs/libcfs.h>
45 #include <obd_class.h>
46 #include "ldlm_internal.h"
49 char *ldlm_lockname[] = {
57 [LCK_GROUP] = "GROUP",
60 EXPORT_SYMBOL(ldlm_lockname);
62 char *ldlm_typename[] = {
64 [LDLM_EXTENT] = "EXT",
69 static ldlm_policy_wire_to_local_t ldlm_policy_wire_to_local[] = {
70 [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_wire_to_local,
71 [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_wire_to_local,
72 [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_wire_to_local,
73 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local,
76 static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire[] = {
77 [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_local_to_wire,
78 [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_local_to_wire,
79 [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_local_to_wire,
80 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_local_to_wire,
84 * Converts lock policy from local format to on the wire lock_desc format
86 void ldlm_convert_policy_to_wire(ldlm_type_t type,
87 const ldlm_policy_data_t *lpolicy,
88 ldlm_wire_policy_data_t *wpolicy)
90 ldlm_policy_local_to_wire_t convert;
92 convert = ldlm_policy_local_to_wire[type - LDLM_MIN_TYPE];
94 convert(lpolicy, wpolicy);
98 * Converts lock policy from on the wire lock_desc format to local format
100 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
101 const ldlm_wire_policy_data_t *wpolicy,
102 ldlm_policy_data_t *lpolicy)
104 ldlm_policy_wire_to_local_t convert;
106 convert = ldlm_policy_wire_to_local[type - LDLM_MIN_TYPE];
108 convert(wpolicy, lpolicy);
111 char *ldlm_it2str(int it)
118 case (IT_OPEN | IT_CREAT):
133 CERROR("Unknown intent %d\n", it);
137 EXPORT_SYMBOL(ldlm_it2str);
139 extern struct kmem_cache *ldlm_lock_slab;
141 #ifdef HAVE_SERVER_SUPPORT
142 static ldlm_processing_policy ldlm_processing_policy_table[] = {
143 [LDLM_PLAIN] = ldlm_process_plain_lock,
144 [LDLM_EXTENT] = ldlm_process_extent_lock,
145 [LDLM_FLOCK] = ldlm_process_flock_lock,
146 [LDLM_IBITS] = ldlm_process_inodebits_lock,
149 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
151 return ldlm_processing_policy_table[res->lr_type];
153 EXPORT_SYMBOL(ldlm_get_processing_policy);
154 #endif /* HAVE_SERVER_SUPPORT */
156 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
160 EXPORT_SYMBOL(ldlm_register_intent);
163 * REFCOUNTED LOCK OBJECTS
168 * Get a reference on a lock.
170 * Lock refcounts, during creation:
171 * - one special one for allocation, dec'd only once in destroy
172 * - one for being a lock that's in-use
173 * - one for the addref associated with a new lock
175 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
177 atomic_inc(&lock->l_refc);
180 EXPORT_SYMBOL(ldlm_lock_get);
183 * Release lock reference.
185 * Also frees the lock if it was last reference.
187 void ldlm_lock_put(struct ldlm_lock *lock)
191 LASSERT(lock->l_resource != LP_POISON);
192 LASSERT(atomic_read(&lock->l_refc) > 0);
193 if (atomic_dec_and_test(&lock->l_refc)) {
194 struct ldlm_resource *res;
197 "final lock_put on destroyed lock, freeing it.");
199 res = lock->l_resource;
200 LASSERT(ldlm_is_destroyed(lock));
201 LASSERT(list_empty(&lock->l_res_link));
202 LASSERT(list_empty(&lock->l_pending_chain));
204 lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats,
206 lu_ref_del(&res->lr_reference, "lock", lock);
207 ldlm_resource_putref(res);
208 lock->l_resource = NULL;
209 if (lock->l_export) {
210 class_export_lock_put(lock->l_export, lock);
211 lock->l_export = NULL;
214 if (lock->l_lvb_data != NULL)
215 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
217 ldlm_interval_free(ldlm_interval_detach(lock));
218 lu_ref_fini(&lock->l_reference);
219 OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle);
224 EXPORT_SYMBOL(ldlm_lock_put);
227 * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked.
229 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
232 if (!list_empty(&lock->l_lru)) {
233 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
235 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
236 list_del_init(&lock->l_lru);
237 LASSERT(ns->ns_nr_unused > 0);
245 * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first.
247 * If \a last_use is non-zero, it will remove the lock from LRU only if
248 * it matches lock's l_last_used.
250 * \retval 0 if \a last_use is set, the lock is not in LRU list or \a last_use
251 * doesn't match lock's l_last_used;
252 * otherwise, the lock hasn't been in the LRU list.
253 * \retval 1 the lock was in LRU list and removed.
255 int ldlm_lock_remove_from_lru_check(struct ldlm_lock *lock, cfs_time_t last_use)
257 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
261 if (ldlm_is_ns_srv(lock)) {
262 LASSERT(list_empty(&lock->l_lru));
266 spin_lock(&ns->ns_lock);
267 if (last_use == 0 || last_use == lock->l_last_used)
268 rc = ldlm_lock_remove_from_lru_nolock(lock);
269 spin_unlock(&ns->ns_lock);
275 * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked.
277 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
279 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
281 lock->l_last_used = cfs_time_current();
282 LASSERT(list_empty(&lock->l_lru));
283 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
284 list_add_tail(&lock->l_lru, &ns->ns_unused_list);
285 ldlm_clear_skipped(lock);
286 LASSERT(ns->ns_nr_unused >= 0);
291 * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks
294 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
296 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
299 spin_lock(&ns->ns_lock);
300 ldlm_lock_add_to_lru_nolock(lock);
301 spin_unlock(&ns->ns_lock);
306 * Moves LDLM lock \a lock that is already in namespace LRU to the tail of
307 * the LRU. Performs necessary LRU locking
309 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
311 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
314 if (ldlm_is_ns_srv(lock)) {
315 LASSERT(list_empty(&lock->l_lru));
320 spin_lock(&ns->ns_lock);
321 if (!list_empty(&lock->l_lru)) {
322 ldlm_lock_remove_from_lru_nolock(lock);
323 ldlm_lock_add_to_lru_nolock(lock);
325 spin_unlock(&ns->ns_lock);
330 * Helper to destroy a locked lock.
332 * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock
333 * Must be called with l_lock and lr_lock held.
335 * Does not actually free the lock data, but rather marks the lock as
336 * destroyed by setting l_destroyed field in the lock to 1. Destroys a
337 * handle->lock association too, so that the lock can no longer be found
338 * and removes the lock from LRU list. Actual lock freeing occurs when
339 * last lock reference goes away.
341 * Original comment (of some historical value):
342 * This used to have a 'strict' flag, which recovery would use to mark an
343 * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
344 * shall explain why it's gone: with the new hash table scheme, once you call
345 * ldlm_lock_destroy, you can never drop your final references on this lock.
346 * Because it's not in the hash table anymore. -phil
348 static int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
352 if (lock->l_readers || lock->l_writers) {
353 LDLM_ERROR(lock, "lock still has references");
357 if (!list_empty(&lock->l_res_link)) {
358 LDLM_ERROR(lock, "lock still on resource");
362 if (ldlm_is_destroyed(lock)) {
363 LASSERT(list_empty(&lock->l_lru));
367 ldlm_set_destroyed(lock);
369 if (lock->l_export && lock->l_export->exp_lock_hash) {
370 /* NB: it's safe to call cfs_hash_del() even lock isn't
371 * in exp_lock_hash. */
372 /* In the function below, .hs_keycmp resolves to
373 * ldlm_export_lock_keycmp() */
374 /* coverity[overrun-buffer-val] */
375 cfs_hash_del(lock->l_export->exp_lock_hash,
376 &lock->l_remote_handle, &lock->l_exp_hash);
379 ldlm_lock_remove_from_lru(lock);
380 class_handle_unhash(&lock->l_handle);
383 /* Wake anyone waiting for this lock */
384 /* FIXME: I should probably add yet another flag, instead of using
385 * l_export to only call this on clients */
387 class_export_put(lock->l_export);
388 lock->l_export = NULL;
389 if (lock->l_export && lock->l_completion_ast)
390 lock->l_completion_ast(lock, 0);
397 * Destroys a LDLM lock \a lock. Performs necessary locking first.
399 void ldlm_lock_destroy(struct ldlm_lock *lock)
403 lock_res_and_lock(lock);
404 first = ldlm_lock_destroy_internal(lock);
405 unlock_res_and_lock(lock);
407 /* drop reference from hashtable only for first destroy */
409 lu_ref_del(&lock->l_reference, "hash", lock);
410 LDLM_LOCK_RELEASE(lock);
416 * Destroys a LDLM lock \a lock that is already locked.
418 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
422 first = ldlm_lock_destroy_internal(lock);
423 /* drop reference from hashtable only for first destroy */
425 lu_ref_del(&lock->l_reference, "hash", lock);
426 LDLM_LOCK_RELEASE(lock);
431 /* this is called by portals_handle2object with the handle lock taken */
432 static void lock_handle_addref(void *lock)
434 LDLM_LOCK_GET((struct ldlm_lock *)lock);
437 static void lock_handle_free(void *lock, int size)
439 LASSERT(size == sizeof(struct ldlm_lock));
440 OBD_SLAB_FREE(lock, ldlm_lock_slab, size);
443 static struct portals_handle_ops lock_handle_ops = {
444 .hop_addref = lock_handle_addref,
445 .hop_free = lock_handle_free,
450 * Allocate and initialize new lock structure.
452 * usage: pass in a resource on which you have done ldlm_resource_get
453 * new lock will take over the refcount.
454 * returns: lock with refcount 2 - one for current caller and one for remote
456 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
458 struct ldlm_lock *lock;
461 if (resource == NULL)
464 OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, GFP_NOFS);
468 spin_lock_init(&lock->l_lock);
469 lock->l_resource = resource;
470 lu_ref_add(&resource->lr_reference, "lock", lock);
472 atomic_set(&lock->l_refc, 2);
473 INIT_LIST_HEAD(&lock->l_res_link);
474 INIT_LIST_HEAD(&lock->l_lru);
475 INIT_LIST_HEAD(&lock->l_pending_chain);
476 INIT_LIST_HEAD(&lock->l_bl_ast);
477 INIT_LIST_HEAD(&lock->l_cp_ast);
478 INIT_LIST_HEAD(&lock->l_rk_ast);
479 init_waitqueue_head(&lock->l_waitq);
480 lock->l_blocking_lock = NULL;
481 INIT_LIST_HEAD(&lock->l_sl_mode);
482 INIT_LIST_HEAD(&lock->l_sl_policy);
483 INIT_HLIST_NODE(&lock->l_exp_hash);
484 INIT_HLIST_NODE(&lock->l_exp_flock_hash);
486 lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats,
488 INIT_LIST_HEAD(&lock->l_handle.h_link);
489 class_handle_hash(&lock->l_handle, &lock_handle_ops);
491 lu_ref_init(&lock->l_reference);
492 lu_ref_add(&lock->l_reference, "hash", lock);
493 lock->l_callback_timeout = 0;
495 #if LUSTRE_TRACKS_LOCK_EXP_REFS
496 INIT_LIST_HEAD(&lock->l_exp_refs_link);
497 lock->l_exp_refs_nr = 0;
498 lock->l_exp_refs_target = NULL;
500 INIT_LIST_HEAD(&lock->l_exp_list);
506 * Moves LDLM lock \a lock to another resource.
507 * This is used on client when server returns some other lock than requested
508 * (typically as a result of intent operation)
510 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
511 const struct ldlm_res_id *new_resid)
513 struct ldlm_resource *oldres = lock->l_resource;
514 struct ldlm_resource *newres;
518 LASSERT(ns_is_client(ns));
520 lock_res_and_lock(lock);
521 if (memcmp(new_resid, &lock->l_resource->lr_name,
522 sizeof(lock->l_resource->lr_name)) == 0) {
524 unlock_res_and_lock(lock);
528 LASSERT(new_resid->name[0] != 0);
530 /* This function assumes that the lock isn't on any lists */
531 LASSERT(list_empty(&lock->l_res_link));
533 type = oldres->lr_type;
534 unlock_res_and_lock(lock);
536 newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
538 RETURN(PTR_ERR(newres));
540 lu_ref_add(&newres->lr_reference, "lock", lock);
542 * To flip the lock from the old to the new resource, lock, oldres and
543 * newres have to be locked. Resource spin-locks are nested within
544 * lock->l_lock, and are taken in the memory address order to avoid
547 spin_lock(&lock->l_lock);
548 oldres = lock->l_resource;
549 if (oldres < newres) {
551 lock_res_nested(newres, LRT_NEW);
554 lock_res_nested(oldres, LRT_NEW);
556 LASSERT(memcmp(new_resid, &oldres->lr_name,
557 sizeof oldres->lr_name) != 0);
558 lock->l_resource = newres;
560 unlock_res_and_lock(lock);
562 /* ...and the flowers are still standing! */
563 lu_ref_del(&oldres->lr_reference, "lock", lock);
564 ldlm_resource_putref(oldres);
569 /** \defgroup ldlm_handles LDLM HANDLES
570 * Ways to get hold of locks without any addresses.
575 * Fills in handle for LDLM lock \a lock into supplied \a lockh
576 * Does not take any references.
578 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
580 lockh->cookie = lock->l_handle.h_cookie;
582 EXPORT_SYMBOL(ldlm_lock2handle);
585 * Obtain a lock reference by handle.
587 * if \a flags: atomically get the lock and set the flags.
588 * Return NULL if flag already set
590 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
593 struct ldlm_lock *lock;
598 lock = class_handle2object(handle->cookie, NULL);
602 /* It's unlikely but possible that someone marked the lock as
603 * destroyed after we did handle2object on it */
604 if ((flags == 0) && !ldlm_is_destroyed(lock)) {
605 lu_ref_add(&lock->l_reference, "handle", current);
609 lock_res_and_lock(lock);
611 LASSERT(lock->l_resource != NULL);
613 lu_ref_add_atomic(&lock->l_reference, "handle", current);
614 if (unlikely(ldlm_is_destroyed(lock))) {
615 unlock_res_and_lock(lock);
616 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
621 /* If we're setting flags, make sure none of them are already set. */
623 if ((lock->l_flags & flags) != 0) {
624 unlock_res_and_lock(lock);
629 lock->l_flags |= flags;
632 unlock_res_and_lock(lock);
635 EXPORT_SYMBOL(__ldlm_handle2lock);
636 /** @} ldlm_handles */
639 * Fill in "on the wire" representation for given LDLM lock into supplied
640 * lock descriptor \a desc structure.
642 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
644 ldlm_res2desc(lock->l_resource, &desc->l_resource);
645 desc->l_req_mode = lock->l_req_mode;
646 desc->l_granted_mode = lock->l_granted_mode;
647 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
648 &lock->l_policy_data,
649 &desc->l_policy_data);
653 * Add a lock to list of conflicting locks to send AST to.
655 * Only add if we have not sent a blocking AST to the lock yet.
657 static void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
658 struct list_head *work_list)
660 if (!ldlm_is_ast_sent(lock)) {
661 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
662 ldlm_set_ast_sent(lock);
663 /* If the enqueuing client said so, tell the AST recipient to
664 * discard dirty data, rather than writing back. */
665 if (ldlm_is_ast_discard_data(new))
666 ldlm_set_discard_data(lock);
667 LASSERT(list_empty(&lock->l_bl_ast));
668 list_add(&lock->l_bl_ast, work_list);
670 LASSERT(lock->l_blocking_lock == NULL);
671 lock->l_blocking_lock = LDLM_LOCK_GET(new);
676 * Add a lock to list of just granted locks to send completion AST to.
678 static void ldlm_add_cp_work_item(struct ldlm_lock *lock,
679 struct list_head *work_list)
681 if (!ldlm_is_cp_reqd(lock)) {
682 ldlm_set_cp_reqd(lock);
683 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
684 LASSERT(list_empty(&lock->l_cp_ast));
685 list_add(&lock->l_cp_ast, work_list);
691 * Aggregator function to add AST work items into a list. Determines
692 * what sort of an AST work needs to be done and calls the proper
694 * Must be called with lr_lock held.
696 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
697 struct list_head *work_list)
700 check_res_locked(lock->l_resource);
702 ldlm_add_bl_work_item(lock, new, work_list);
704 ldlm_add_cp_work_item(lock, work_list);
709 * Add specified reader/writer reference to LDLM lock with handle \a lockh.
710 * r/w reference type is determined by \a mode
711 * Calls ldlm_lock_addref_internal.
713 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
715 struct ldlm_lock *lock;
717 lock = ldlm_handle2lock(lockh);
718 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
719 ldlm_lock_addref_internal(lock, mode);
722 EXPORT_SYMBOL(ldlm_lock_addref);
726 * Add specified reader/writer reference to LDLM lock \a lock.
727 * r/w reference type is determined by \a mode
728 * Removes lock from LRU if it is there.
729 * Assumes the LDLM lock is already locked.
731 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
733 ldlm_lock_remove_from_lru(lock);
734 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
736 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
738 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
740 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
743 lu_ref_add_atomic(&lock->l_reference, "user", lock);
744 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
748 * Attempts to add reader/writer reference to a lock with handle \a lockh, and
749 * fails if lock is already LDLM_FL_CBPENDING or destroyed.
751 * \retval 0 success, lock was addref-ed
753 * \retval -EAGAIN lock is being canceled.
755 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
757 struct ldlm_lock *lock;
761 lock = ldlm_handle2lock(lockh);
763 lock_res_and_lock(lock);
764 if (lock->l_readers != 0 || lock->l_writers != 0 ||
765 !ldlm_is_cbpending(lock)) {
766 ldlm_lock_addref_internal_nolock(lock, mode);
769 unlock_res_and_lock(lock);
774 EXPORT_SYMBOL(ldlm_lock_addref_try);
777 * Add specified reader/writer reference to LDLM lock \a lock.
778 * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work.
779 * Only called for local locks.
781 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
783 lock_res_and_lock(lock);
784 ldlm_lock_addref_internal_nolock(lock, mode);
785 unlock_res_and_lock(lock);
789 * Removes reader/writer reference for LDLM lock \a lock.
790 * Assumes LDLM lock is already locked.
791 * only called in ldlm_flock_destroy and for local locks.
792 * Does NOT add lock to LRU if no r/w references left to accomodate flock locks
793 * that cannot be placed in LRU.
795 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
797 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
798 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
799 LASSERT(lock->l_readers > 0);
800 lu_ref_del(&lock->l_reference, "reader", lock);
803 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
804 LASSERT(lock->l_writers > 0);
805 lu_ref_del(&lock->l_reference, "writer", lock);
809 lu_ref_del(&lock->l_reference, "user", lock);
810 LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */
814 * Removes reader/writer reference for LDLM lock \a lock.
815 * Locks LDLM lock first.
816 * If the lock is determined to be client lock on a client and r/w refcount
817 * drops to zero and the lock is not blocked, the lock is added to LRU lock
819 * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called.
821 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
823 struct ldlm_namespace *ns;
826 lock_res_and_lock(lock);
828 ns = ldlm_lock_to_ns(lock);
830 ldlm_lock_decref_internal_nolock(lock, mode);
832 if (ldlm_is_local(lock) &&
833 !lock->l_readers && !lock->l_writers) {
834 /* If this is a local lock on a server namespace and this was
835 * the last reference, cancel the lock. */
836 CDEBUG(D_INFO, "forcing cancel of local lock\n");
837 ldlm_set_cbpending(lock);
840 if (!lock->l_readers && !lock->l_writers &&
841 (ldlm_is_cbpending(lock) || lock->l_req_mode == LCK_GROUP)) {
842 /* If we received a blocked AST and this was the last reference,
844 * Group locks are special:
845 * They must not go in LRU, but they are not called back
846 * like non-group locks, instead they are manually released.
847 * They have an l_writers reference which they keep until
848 * they are manually released, so we remove them when they have
849 * no more reader or writer references. - LU-6368 */
850 if (ldlm_is_ns_srv(lock) && lock->l_export)
851 CERROR("FL_CBPENDING set on non-local lock--just a "
854 LDLM_DEBUG(lock, "final decref done on cbpending lock");
856 LDLM_LOCK_GET(lock); /* dropped by bl thread */
857 ldlm_lock_remove_from_lru(lock);
858 unlock_res_and_lock(lock);
860 if (ldlm_is_fail_loc(lock))
861 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
863 if (ldlm_is_atomic_cb(lock) ||
864 ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
865 ldlm_handle_bl_callback(ns, NULL, lock);
866 } else if (ns_is_client(ns) &&
867 !lock->l_readers && !lock->l_writers &&
868 !ldlm_is_no_lru(lock) &&
869 !ldlm_is_bl_ast(lock)) {
871 LDLM_DEBUG(lock, "add lock into lru list");
873 /* If this is a client-side namespace and this was the last
874 * reference, put it on the LRU. */
875 ldlm_lock_add_to_lru(lock);
876 unlock_res_and_lock(lock);
878 if (ldlm_is_fail_loc(lock))
879 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
881 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
882 * are not supported by the server, otherwise, it is done on
884 if (!exp_connect_cancelset(lock->l_conn_export) &&
885 !ns_connect_lru_resize(ns))
886 ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0);
888 LDLM_DEBUG(lock, "do not add lock into lru list");
889 unlock_res_and_lock(lock);
896 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
898 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
900 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
901 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
902 ldlm_lock_decref_internal(lock, mode);
905 EXPORT_SYMBOL(ldlm_lock_decref);
908 * Decrease reader/writer refcount for LDLM lock with handle
909 * \a lockh and mark it for subsequent cancellation once r/w refcount
910 * drops to zero instead of putting into LRU.
913 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
915 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
918 LASSERT(lock != NULL);
920 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
921 lock_res_and_lock(lock);
922 ldlm_set_cbpending(lock);
923 unlock_res_and_lock(lock);
924 ldlm_lock_decref_internal(lock, mode);
927 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
929 struct sl_insert_point {
930 struct list_head *res_link;
931 struct list_head *mode_link;
932 struct list_head *policy_link;
936 * Finds a position to insert the new lock into granted lock list.
938 * Used for locks eligible for skiplist optimization.
941 * queue [input]: the granted list where search acts on;
942 * req [input]: the lock whose position to be located;
943 * prev [output]: positions within 3 lists to insert @req to
947 * - ldlm_grant_lock_with_skiplist
949 static void search_granted_lock(struct list_head *queue,
950 struct ldlm_lock *req,
951 struct sl_insert_point *prev)
953 struct list_head *tmp;
954 struct ldlm_lock *lock, *mode_end, *policy_end;
957 list_for_each(tmp, queue) {
958 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
960 mode_end = list_entry(lock->l_sl_mode.prev,
961 struct ldlm_lock, l_sl_mode);
963 if (lock->l_req_mode != req->l_req_mode) {
964 /* jump to last lock of mode group */
965 tmp = &mode_end->l_res_link;
969 /* suitable mode group is found */
970 if (lock->l_resource->lr_type == LDLM_PLAIN) {
971 /* insert point is last lock of the mode group */
972 prev->res_link = &mode_end->l_res_link;
973 prev->mode_link = &mode_end->l_sl_mode;
974 prev->policy_link = &req->l_sl_policy;
977 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
980 list_entry(lock->l_sl_policy.prev,
984 if (lock->l_policy_data.l_inodebits.bits ==
985 req->l_policy_data.l_inodebits.bits) {
986 /* insert point is last lock of
987 * the policy group */
989 &policy_end->l_res_link;
991 &policy_end->l_sl_mode;
993 &policy_end->l_sl_policy;
998 if (policy_end == mode_end)
999 /* done with mode group */
1002 /* go to next policy group within mode group */
1003 tmp = policy_end->l_res_link.next;
1004 lock = list_entry(tmp, struct ldlm_lock,
1006 } /* loop over policy groups within the mode group */
1008 /* insert point is last lock of the mode group,
1009 * new policy group is started */
1010 prev->res_link = &mode_end->l_res_link;
1011 prev->mode_link = &mode_end->l_sl_mode;
1012 prev->policy_link = &req->l_sl_policy;
1016 LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
1021 /* insert point is last lock on the queue,
1022 * new mode group and new policy group are started */
1023 prev->res_link = queue->prev;
1024 prev->mode_link = &req->l_sl_mode;
1025 prev->policy_link = &req->l_sl_policy;
1031 * Add a lock into resource granted list after a position described by
1034 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
1035 struct sl_insert_point *prev)
1037 struct ldlm_resource *res = lock->l_resource;
1040 check_res_locked(res);
1042 ldlm_resource_dump(D_INFO, res);
1043 LDLM_DEBUG(lock, "About to add lock:");
1045 if (ldlm_is_destroyed(lock)) {
1046 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1050 LASSERT(list_empty(&lock->l_res_link));
1051 LASSERT(list_empty(&lock->l_sl_mode));
1052 LASSERT(list_empty(&lock->l_sl_policy));
1055 * lock->link == prev->link means lock is first starting the group.
1056 * Don't re-add to itself to suppress kernel warnings.
1058 if (&lock->l_res_link != prev->res_link)
1059 list_add(&lock->l_res_link, prev->res_link);
1060 if (&lock->l_sl_mode != prev->mode_link)
1061 list_add(&lock->l_sl_mode, prev->mode_link);
1062 if (&lock->l_sl_policy != prev->policy_link)
1063 list_add(&lock->l_sl_policy, prev->policy_link);
1069 * Add a lock to granted list on a resource maintaining skiplist
1072 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
1074 struct sl_insert_point prev;
1077 LASSERT(lock->l_req_mode == lock->l_granted_mode);
1079 search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
1080 ldlm_granted_list_add_lock(lock, &prev);
1085 * Perform lock granting bookkeeping.
1087 * Includes putting the lock into granted list and updating lock mode.
1089 * - ldlm_lock_enqueue
1090 * - ldlm_reprocess_queue
1091 * - ldlm_lock_convert
1093 * must be called with lr_lock held
1095 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
1097 struct ldlm_resource *res = lock->l_resource;
1100 check_res_locked(res);
1102 lock->l_granted_mode = lock->l_req_mode;
1104 if (work_list && lock->l_completion_ast != NULL)
1105 ldlm_add_ast_work_item(lock, NULL, work_list);
1107 /* We should not add locks to granted list in the following cases:
1108 * - this is an UNLOCK but not a real lock;
1109 * - this is a TEST lock;
1110 * - this is a F_CANCELLK lock (async flock has req_mode == 0)
1111 * - this is a deadlock (flock cannot be granted) */
1112 if (lock->l_req_mode == 0 ||
1113 lock->l_req_mode == LCK_NL ||
1114 ldlm_is_test_lock(lock) ||
1115 ldlm_is_flock_deadlock(lock))
1118 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
1119 ldlm_grant_lock_with_skiplist(lock);
1120 else if (res->lr_type == LDLM_EXTENT)
1121 ldlm_extent_add_lock(res, lock);
1123 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1125 ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1130 * Describe the overlap between two locks. itree_overlap_cb data.
1132 struct lock_match_data {
1133 struct ldlm_lock *lmd_old;
1134 struct ldlm_lock *lmd_lock;
1135 ldlm_mode_t *lmd_mode;
1136 ldlm_policy_data_t *lmd_policy;
1142 * Check if the given @lock meets the criteria for a match.
1143 * A reference on the lock is taken if matched.
1145 * \param lock test-against this lock
1146 * \param data parameters
1148 static int lock_matches(struct ldlm_lock *lock, struct lock_match_data *data)
1150 ldlm_policy_data_t *lpol = &lock->l_policy_data;
1153 if (lock == data->lmd_old)
1154 return INTERVAL_ITER_STOP;
1156 /* Check if this lock can be matched.
1157 * Used by LU-2919(exclusive open) for open lease lock */
1158 if (ldlm_is_excl(lock))
1159 return INTERVAL_ITER_CONT;
1161 /* llite sometimes wants to match locks that will be
1162 * canceled when their users drop, but we allow it to match
1163 * if it passes in CBPENDING and the lock still has users.
1164 * this is generally only going to be used by children
1165 * whose parents already hold a lock so forward progress
1166 * can still happen. */
1167 if (ldlm_is_cbpending(lock) &&
1168 !(data->lmd_flags & LDLM_FL_CBPENDING))
1169 return INTERVAL_ITER_CONT;
1170 if (!data->lmd_unref && ldlm_is_cbpending(lock) &&
1171 lock->l_readers == 0 && lock->l_writers == 0)
1172 return INTERVAL_ITER_CONT;
1174 if (!(lock->l_req_mode & *data->lmd_mode))
1175 return INTERVAL_ITER_CONT;
1176 match = lock->l_req_mode;
1178 switch (lock->l_resource->lr_type) {
1180 if (lpol->l_extent.start > data->lmd_policy->l_extent.start ||
1181 lpol->l_extent.end < data->lmd_policy->l_extent.end)
1182 return INTERVAL_ITER_CONT;
1184 if (unlikely(match == LCK_GROUP) &&
1185 data->lmd_policy->l_extent.gid != LDLM_GID_ANY &&
1186 lpol->l_extent.gid != data->lmd_policy->l_extent.gid)
1187 return INTERVAL_ITER_CONT;
1190 /* We match if we have existing lock with same or wider set
1192 if ((lpol->l_inodebits.bits &
1193 data->lmd_policy->l_inodebits.bits) !=
1194 data->lmd_policy->l_inodebits.bits)
1195 return INTERVAL_ITER_CONT;
1201 /* We match if we have existing lock with same or wider set
1203 if (!data->lmd_unref && LDLM_HAVE_MASK(lock, GONE))
1204 return INTERVAL_ITER_CONT;
1206 if ((data->lmd_flags & LDLM_FL_LOCAL_ONLY) &&
1207 !ldlm_is_local(lock))
1208 return INTERVAL_ITER_CONT;
1210 if (data->lmd_flags & LDLM_FL_TEST_LOCK) {
1211 LDLM_LOCK_GET(lock);
1212 ldlm_lock_touch_in_lru(lock);
1214 ldlm_lock_addref_internal_nolock(lock, match);
1217 *data->lmd_mode = match;
1218 data->lmd_lock = lock;
1220 return INTERVAL_ITER_STOP;
1223 static unsigned int itree_overlap_cb(struct interval_node *in, void *args)
1225 struct ldlm_interval *node = to_ldlm_interval(in);
1226 struct lock_match_data *data = args;
1227 struct ldlm_lock *lock;
1230 list_for_each_entry(lock, &node->li_group, l_sl_policy) {
1231 rc = lock_matches(lock, data);
1232 if (rc == INTERVAL_ITER_STOP)
1233 return INTERVAL_ITER_STOP;
1235 return INTERVAL_ITER_CONT;
1239 * Search for a lock with given parameters in interval trees.
1241 * \param res search for a lock in this resource
1242 * \param data parameters
1244 * \retval a referenced lock or NULL.
1246 static struct ldlm_lock *search_itree(struct ldlm_resource *res,
1247 struct lock_match_data *data)
1249 struct interval_node_extent ext = {
1250 .start = data->lmd_policy->l_extent.start,
1251 .end = data->lmd_policy->l_extent.end
1255 for (idx = 0; idx < LCK_MODE_NUM; idx++) {
1256 struct ldlm_interval_tree *tree = &res->lr_itree[idx];
1258 if (tree->lit_root == NULL)
1261 if (!(tree->lit_mode & *data->lmd_mode))
1264 interval_search(tree->lit_root, &ext,
1265 itree_overlap_cb, data);
1267 return data->lmd_lock;
1272 * Search for a lock with given properties in a queue.
1274 * \param queue search for a lock in this queue
1275 * \param data parameters
1277 * \retval a referenced lock or NULL.
1279 static struct ldlm_lock *search_queue(struct list_head *queue,
1280 struct lock_match_data *data)
1282 struct ldlm_lock *lock;
1285 list_for_each_entry(lock, queue, l_res_link) {
1286 rc = lock_matches(lock, data);
1287 if (rc == INTERVAL_ITER_STOP)
1288 return data->lmd_lock;
1293 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1295 if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) {
1296 lock->l_flags |= LDLM_FL_FAIL_NOTIFIED;
1297 wake_up_all(&lock->l_waitq);
1300 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1302 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1304 lock_res_and_lock(lock);
1305 ldlm_lock_fail_match_locked(lock);
1306 unlock_res_and_lock(lock);
1310 * Mark lock as "matchable" by OST.
1312 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1314 * Assumes LDLM lock is already locked.
1316 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1318 ldlm_set_lvb_ready(lock);
1319 wake_up_all(&lock->l_waitq);
1321 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1324 * Mark lock as "matchable" by OST.
1325 * Locks the lock and then \see ldlm_lock_allow_match_locked
1327 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1329 lock_res_and_lock(lock);
1330 ldlm_lock_allow_match_locked(lock);
1331 unlock_res_and_lock(lock);
1333 EXPORT_SYMBOL(ldlm_lock_allow_match);
1336 * Attempt to find a lock with specified properties.
1338 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1341 * Can be called in two ways:
1343 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1344 * for a duplicate of.
1346 * Otherwise, all of the fields must be filled in, to match against.
1348 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1349 * server (ie, connh is NULL)
1350 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1351 * list will be considered
1352 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1353 * to be canceled can still be matched as long as they still have reader
1354 * or writer refernces
1355 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1356 * just tell us if we would have matched.
1358 * \retval 1 if it finds an already-existing lock that is compatible; in this
1359 * case, lockh is filled in with a addref()ed lock
1361 * We also check security context, and if that fails we simply return 0 (to
1362 * keep caller code unchanged), the context failure will be discovered by
1363 * caller sometime later.
1365 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1366 const struct ldlm_res_id *res_id, ldlm_type_t type,
1367 ldlm_policy_data_t *policy, ldlm_mode_t mode,
1368 struct lustre_handle *lockh, int unref)
1370 struct lock_match_data data = {
1374 .lmd_policy = policy,
1378 struct ldlm_resource *res;
1379 struct ldlm_lock *lock;
1384 data.lmd_old = ldlm_handle2lock(lockh);
1385 LASSERT(data.lmd_old != NULL);
1387 ns = ldlm_lock_to_ns(data.lmd_old);
1388 res_id = &data.lmd_old->l_resource->lr_name;
1389 type = data.lmd_old->l_resource->lr_type;
1390 *data.lmd_mode = data.lmd_old->l_req_mode;
1393 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1395 LASSERT(data.lmd_old == NULL);
1399 LDLM_RESOURCE_ADDREF(res);
1402 if (res->lr_type == LDLM_EXTENT)
1403 lock = search_itree(res, &data);
1405 lock = search_queue(&res->lr_granted, &data);
1408 if (flags & LDLM_FL_BLOCK_GRANTED)
1410 lock = search_queue(&res->lr_converting, &data);
1413 lock = search_queue(&res->lr_waiting, &data);
1420 LDLM_RESOURCE_DELREF(res);
1421 ldlm_resource_putref(res);
1424 ldlm_lock2handle(lock, lockh);
1425 if ((flags & LDLM_FL_LVB_READY) &&
1426 (!ldlm_is_lvb_ready(lock))) {
1427 __u64 wait_flags = LDLM_FL_LVB_READY |
1428 LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED;
1429 struct l_wait_info lwi;
1430 if (lock->l_completion_ast) {
1431 int err = lock->l_completion_ast(lock,
1432 LDLM_FL_WAIT_NOREPROC,
1435 if (flags & LDLM_FL_TEST_LOCK)
1436 LDLM_LOCK_RELEASE(lock);
1438 ldlm_lock_decref_internal(lock,
1445 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1446 NULL, LWI_ON_SIGNAL_NOOP, NULL);
1448 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1449 l_wait_event(lock->l_waitq,
1450 lock->l_flags & wait_flags,
1452 if (!ldlm_is_lvb_ready(lock)) {
1453 if (flags & LDLM_FL_TEST_LOCK)
1454 LDLM_LOCK_RELEASE(lock);
1456 ldlm_lock_decref_internal(lock, mode);
1463 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1464 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1465 res_id->name[2] : policy->l_extent.start,
1466 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1467 res_id->name[3] : policy->l_extent.end);
1469 /* check user's security context */
1470 if (lock->l_conn_export &&
1471 sptlrpc_import_check_ctx(
1472 class_exp2cliimp(lock->l_conn_export))) {
1473 if (!(flags & LDLM_FL_TEST_LOCK))
1474 ldlm_lock_decref_internal(lock, mode);
1478 if (flags & LDLM_FL_TEST_LOCK)
1479 LDLM_LOCK_RELEASE(lock);
1481 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1482 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1483 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1484 type, mode, res_id->name[0], res_id->name[1],
1485 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1486 res_id->name[2] :policy->l_extent.start,
1487 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1488 res_id->name[3] : policy->l_extent.end);
1490 if (data.lmd_old != NULL)
1491 LDLM_LOCK_PUT(data.lmd_old);
1493 return rc ? mode : 0;
1495 EXPORT_SYMBOL(ldlm_lock_match);
1497 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1500 struct ldlm_lock *lock;
1501 ldlm_mode_t mode = 0;
1504 lock = ldlm_handle2lock(lockh);
1506 lock_res_and_lock(lock);
1507 if (LDLM_HAVE_MASK(lock, GONE))
1510 if (ldlm_is_cbpending(lock) &&
1511 lock->l_readers == 0 && lock->l_writers == 0)
1515 *bits = lock->l_policy_data.l_inodebits.bits;
1516 mode = lock->l_granted_mode;
1517 ldlm_lock_addref_internal_nolock(lock, mode);
1524 unlock_res_and_lock(lock);
1525 LDLM_LOCK_PUT(lock);
1529 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1531 /** The caller must guarantee that the buffer is large enough. */
1532 int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill,
1533 enum req_location loc, void *data, int size)
1538 LASSERT(data != NULL);
1541 switch (lock->l_lvb_type) {
1543 if (size == sizeof(struct ost_lvb)) {
1544 if (loc == RCL_CLIENT)
1545 lvb = req_capsule_client_swab_get(pill,
1547 lustre_swab_ost_lvb);
1549 lvb = req_capsule_server_swab_get(pill,
1551 lustre_swab_ost_lvb);
1552 if (unlikely(lvb == NULL)) {
1553 LDLM_ERROR(lock, "no LVB");
1557 memcpy(data, lvb, size);
1558 } else if (size == sizeof(struct ost_lvb_v1)) {
1559 struct ost_lvb *olvb = data;
1561 if (loc == RCL_CLIENT)
1562 lvb = req_capsule_client_swab_get(pill,
1564 lustre_swab_ost_lvb_v1);
1566 lvb = req_capsule_server_sized_swab_get(pill,
1568 lustre_swab_ost_lvb_v1);
1569 if (unlikely(lvb == NULL)) {
1570 LDLM_ERROR(lock, "no LVB");
1574 memcpy(data, lvb, size);
1575 olvb->lvb_mtime_ns = 0;
1576 olvb->lvb_atime_ns = 0;
1577 olvb->lvb_ctime_ns = 0;
1579 LDLM_ERROR(lock, "Replied unexpected ost LVB size %d",
1585 if (size == sizeof(struct lquota_lvb)) {
1586 if (loc == RCL_CLIENT)
1587 lvb = req_capsule_client_swab_get(pill,
1589 lustre_swab_lquota_lvb);
1591 lvb = req_capsule_server_swab_get(pill,
1593 lustre_swab_lquota_lvb);
1594 if (unlikely(lvb == NULL)) {
1595 LDLM_ERROR(lock, "no LVB");
1599 memcpy(data, lvb, size);
1601 LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d",
1610 if (loc == RCL_CLIENT)
1611 lvb = req_capsule_client_get(pill, &RMF_DLM_LVB);
1613 lvb = req_capsule_server_get(pill, &RMF_DLM_LVB);
1614 if (unlikely(lvb == NULL)) {
1615 LDLM_ERROR(lock, "no LVB");
1619 memcpy(data, lvb, size);
1622 LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type);
1623 libcfs_debug_dumpstack(NULL);
1631 * Create and fill in new LDLM lock with specified properties.
1632 * Returns a referenced lock
1634 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1635 const struct ldlm_res_id *res_id,
1638 const struct ldlm_callback_suite *cbs,
1639 void *data, __u32 lvb_len,
1640 enum lvb_type lvb_type)
1642 struct ldlm_lock *lock;
1643 struct ldlm_resource *res;
1647 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1649 RETURN(ERR_CAST(res));
1651 lock = ldlm_lock_new(res);
1653 RETURN(ERR_PTR(-ENOMEM));
1655 lock->l_req_mode = mode;
1656 lock->l_ast_data = data;
1657 lock->l_pid = current_pid();
1658 if (ns_is_server(ns))
1659 ldlm_set_ns_srv(lock);
1661 lock->l_blocking_ast = cbs->lcs_blocking;
1662 lock->l_completion_ast = cbs->lcs_completion;
1663 lock->l_glimpse_ast = cbs->lcs_glimpse;
1666 lock->l_tree_node = NULL;
1667 /* if this is the extent lock, allocate the interval tree node */
1668 if (type == LDLM_EXTENT)
1669 if (ldlm_interval_alloc(lock) == NULL)
1670 GOTO(out, rc = -ENOMEM);
1673 lock->l_lvb_len = lvb_len;
1674 OBD_ALLOC_LARGE(lock->l_lvb_data, lvb_len);
1675 if (lock->l_lvb_data == NULL)
1676 GOTO(out, rc = -ENOMEM);
1679 lock->l_lvb_type = lvb_type;
1680 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1681 GOTO(out, rc = -ENOENT);
1686 ldlm_lock_destroy(lock);
1687 LDLM_LOCK_RELEASE(lock);
1688 RETURN(ERR_PTR(rc));
1692 * Enqueue (request) a lock.
1694 * Does not block. As a result of enqueue the lock would be put
1695 * into granted or waiting list.
1697 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1698 * set, skip all the enqueueing and delegate lock processing to intent policy
1701 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1702 struct ldlm_lock **lockp,
1703 void *cookie, __u64 *flags)
1705 struct ldlm_lock *lock = *lockp;
1706 struct ldlm_resource *res = lock->l_resource;
1707 int local = ns_is_client(ldlm_res_to_ns(res));
1708 #ifdef HAVE_SERVER_SUPPORT
1709 ldlm_processing_policy policy;
1711 ldlm_error_t rc = ELDLM_OK;
1712 struct ldlm_interval *node = NULL;
1715 /* policies are not executed on the client or during replay */
1716 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1717 && !local && ns->ns_policy) {
1718 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1720 if (rc == ELDLM_LOCK_REPLACED) {
1721 /* The lock that was returned has already been granted,
1722 * and placed into lockp. If it's not the same as the
1723 * one we passed in, then destroy the old one and our
1724 * work here is done. */
1725 if (lock != *lockp) {
1726 ldlm_lock_destroy(lock);
1727 LDLM_LOCK_RELEASE(lock);
1729 *flags |= LDLM_FL_LOCK_CHANGED;
1731 } else if (rc != ELDLM_OK ||
1732 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1733 ldlm_lock_destroy(lock);
1738 if (*flags & LDLM_FL_RESENT) {
1739 /* Reconstruct LDLM_FL_SRV_ENQ_MASK @flags for reply.
1740 * Set LOCK_CHANGED always.
1741 * Check if the lock is granted for BLOCK_GRANTED.
1742 * Take NO_TIMEOUT from the lock as it is inherited through
1743 * LDLM_FL_INHERIT_MASK */
1744 *flags |= LDLM_FL_LOCK_CHANGED;
1745 if (lock->l_req_mode != lock->l_granted_mode)
1746 *flags |= LDLM_FL_BLOCK_GRANTED;
1747 *flags |= lock->l_flags & LDLM_FL_NO_TIMEOUT;
1751 /* For a replaying lock, it might be already in granted list. So
1752 * unlinking the lock will cause the interval node to be freed, we
1753 * have to allocate the interval node early otherwise we can't regrant
1754 * this lock in the future. - jay */
1755 if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1756 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
1758 lock_res_and_lock(lock);
1759 if (local && lock->l_req_mode == lock->l_granted_mode) {
1760 /* The server returned a blocked lock, but it was granted
1761 * before we got a chance to actually enqueue it. We don't
1762 * need to do anything else. */
1763 *flags &= ~LDLM_FL_BLOCKED_MASK;
1764 GOTO(out, rc = ELDLM_OK);
1767 ldlm_resource_unlink_lock(lock);
1768 if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1770 ldlm_lock_destroy_nolock(lock);
1771 GOTO(out, rc = -ENOMEM);
1774 INIT_LIST_HEAD(&node->li_group);
1775 ldlm_interval_attach(node, lock);
1779 /* Some flags from the enqueue want to make it into the AST, via the
1780 * lock's l_flags. */
1781 if (*flags & LDLM_FL_AST_DISCARD_DATA)
1782 ldlm_set_ast_discard_data(lock);
1783 if (*flags & LDLM_FL_TEST_LOCK)
1784 ldlm_set_test_lock(lock);
1786 /* This distinction between local lock trees is very important; a client
1787 * namespace only has information about locks taken by that client, and
1788 * thus doesn't have enough information to decide for itself if it can
1789 * be granted (below). In this case, we do exactly what the server
1790 * tells us to do, as dictated by the 'flags'.
1792 * We do exactly the same thing during recovery, when the server is
1793 * more or less trusting the clients not to lie.
1795 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1796 * granted/converting queues. */
1798 if (*flags & LDLM_FL_BLOCK_CONV)
1799 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1800 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1801 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1803 ldlm_grant_lock(lock, NULL);
1804 GOTO(out, rc = ELDLM_OK);
1805 #ifdef HAVE_SERVER_SUPPORT
1806 } else if (*flags & LDLM_FL_REPLAY) {
1807 if (*flags & LDLM_FL_BLOCK_CONV) {
1808 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1809 GOTO(out, rc = ELDLM_OK);
1810 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1811 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1812 GOTO(out, rc = ELDLM_OK);
1813 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1814 ldlm_grant_lock(lock, NULL);
1815 GOTO(out, rc = ELDLM_OK);
1817 /* If no flags, fall through to normal enqueue path. */
1820 policy = ldlm_processing_policy_table[res->lr_type];
1821 policy(lock, flags, 1, &rc, NULL);
1825 CERROR("This is client-side-only module, cannot handle "
1826 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1832 unlock_res_and_lock(lock);
1834 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1838 #ifdef HAVE_SERVER_SUPPORT
1840 * Iterate through all waiting locks on a given resource queue and attempt to
1843 * Must be called with resource lock held.
1845 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1846 struct list_head *work_list)
1848 struct list_head *tmp, *pos;
1849 ldlm_processing_policy policy;
1851 int rc = LDLM_ITER_CONTINUE;
1855 check_res_locked(res);
1857 policy = ldlm_processing_policy_table[res->lr_type];
1860 list_for_each_safe(tmp, pos, queue) {
1861 struct ldlm_lock *pending;
1862 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1864 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1867 rc = policy(pending, &flags, 0, &err, work_list);
1868 if (rc != LDLM_ITER_CONTINUE)
1877 * Process a call to blocking AST callback for a lock in ast_work list
1880 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1882 struct ldlm_cb_set_arg *arg = opaq;
1883 struct ldlm_lock_desc d;
1885 struct ldlm_lock *lock;
1888 if (list_empty(arg->list))
1891 lock = list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1893 /* nobody should touch l_bl_ast */
1894 lock_res_and_lock(lock);
1895 list_del_init(&lock->l_bl_ast);
1897 LASSERT(ldlm_is_ast_sent(lock));
1898 LASSERT(lock->l_bl_ast_run == 0);
1899 LASSERT(lock->l_blocking_lock);
1900 lock->l_bl_ast_run++;
1901 unlock_res_and_lock(lock);
1903 ldlm_lock2desc(lock->l_blocking_lock, &d);
1905 rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1906 LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1907 lock->l_blocking_lock = NULL;
1908 LDLM_LOCK_RELEASE(lock);
1914 * Process a call to completion AST callback for a lock in ast_work list
1917 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1919 struct ldlm_cb_set_arg *arg = opaq;
1921 struct ldlm_lock *lock;
1922 ldlm_completion_callback completion_callback;
1925 if (list_empty(arg->list))
1928 lock = list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1930 /* It's possible to receive a completion AST before we've set
1931 * the l_completion_ast pointer: either because the AST arrived
1932 * before the reply, or simply because there's a small race
1933 * window between receiving the reply and finishing the local
1934 * enqueue. (bug 842)
1936 * This can't happen with the blocking_ast, however, because we
1937 * will never call the local blocking_ast until we drop our
1938 * reader/writer reference, which we won't do until we get the
1939 * reply and finish enqueueing. */
1941 /* nobody should touch l_cp_ast */
1942 lock_res_and_lock(lock);
1943 list_del_init(&lock->l_cp_ast);
1944 LASSERT(ldlm_is_cp_reqd(lock));
1945 /* save l_completion_ast since it can be changed by
1946 * mds_intent_policy(), see bug 14225 */
1947 completion_callback = lock->l_completion_ast;
1948 ldlm_clear_cp_reqd(lock);
1949 unlock_res_and_lock(lock);
1951 if (completion_callback != NULL)
1952 rc = completion_callback(lock, 0, (void *)arg);
1953 LDLM_LOCK_RELEASE(lock);
1959 * Process a call to revocation AST callback for a lock in ast_work list
1962 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1964 struct ldlm_cb_set_arg *arg = opaq;
1965 struct ldlm_lock_desc desc;
1967 struct ldlm_lock *lock;
1970 if (list_empty(arg->list))
1973 lock = list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1974 list_del_init(&lock->l_rk_ast);
1976 /* the desc just pretend to exclusive */
1977 ldlm_lock2desc(lock, &desc);
1978 desc.l_req_mode = LCK_EX;
1979 desc.l_granted_mode = 0;
1981 rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1982 LDLM_LOCK_RELEASE(lock);
1988 * Process a call to glimpse AST callback for a lock in ast_work list
1990 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1992 struct ldlm_cb_set_arg *arg = opaq;
1993 struct ldlm_glimpse_work *gl_work;
1994 struct ldlm_lock *lock;
1998 if (list_empty(arg->list))
2001 gl_work = list_entry(arg->list->next, struct ldlm_glimpse_work,
2003 list_del_init(&gl_work->gl_list);
2005 lock = gl_work->gl_lock;
2007 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
2008 arg->gl_desc = gl_work->gl_desc;
2010 /* invoke the actual glimpse callback */
2011 if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
2014 LDLM_LOCK_RELEASE(lock);
2016 if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
2017 OBD_FREE_PTR(gl_work);
2023 * Process list of locks in need of ASTs being sent.
2025 * Used on server to send multiple ASTs together instead of sending one by
2028 int ldlm_run_ast_work(struct ldlm_namespace *ns, struct list_head *rpc_list,
2029 ldlm_desc_ast_t ast_type)
2031 struct ldlm_cb_set_arg *arg;
2032 set_producer_func work_ast_lock;
2035 if (list_empty(rpc_list))
2042 atomic_set(&arg->restart, 0);
2043 arg->list = rpc_list;
2046 case LDLM_WORK_BL_AST:
2047 arg->type = LDLM_BL_CALLBACK;
2048 work_ast_lock = ldlm_work_bl_ast_lock;
2050 case LDLM_WORK_CP_AST:
2051 arg->type = LDLM_CP_CALLBACK;
2052 work_ast_lock = ldlm_work_cp_ast_lock;
2054 case LDLM_WORK_REVOKE_AST:
2055 arg->type = LDLM_BL_CALLBACK;
2056 work_ast_lock = ldlm_work_revoke_ast_lock;
2058 case LDLM_WORK_GL_AST:
2059 arg->type = LDLM_GL_CALLBACK;
2060 work_ast_lock = ldlm_work_gl_ast_lock;
2066 /* We create a ptlrpc request set with flow control extension.
2067 * This request set will use the work_ast_lock function to produce new
2068 * requests and will send a new request each time one completes in order
2069 * to keep the number of requests in flight to ns_max_parallel_ast */
2070 arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
2071 work_ast_lock, arg);
2072 if (arg->set == NULL)
2073 GOTO(out, rc = -ENOMEM);
2075 ptlrpc_set_wait(arg->set);
2076 ptlrpc_set_destroy(arg->set);
2078 rc = atomic_read(&arg->restart) ? -ERESTART : 0;
2085 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
2087 ldlm_reprocess_all(res);
2088 return LDLM_ITER_CONTINUE;
2091 static int ldlm_reprocess_res(struct cfs_hash *hs, struct cfs_hash_bd *bd,
2092 struct hlist_node *hnode, void *arg)
2094 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2097 rc = reprocess_one_queue(res, arg);
2099 return rc == LDLM_ITER_STOP;
2103 * Iterate through all resources on a namespace attempting to grant waiting
2106 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
2111 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2112 ldlm_reprocess_res, NULL);
2118 * Try to grant all waiting locks on a resource.
2120 * Calls ldlm_reprocess_queue on converting and waiting queues.
2122 * Typically called after some resource locks are cancelled to see
2123 * if anything could be granted as a result of the cancellation.
2125 void ldlm_reprocess_all(struct ldlm_resource *res)
2127 struct list_head rpc_list;
2128 #ifdef HAVE_SERVER_SUPPORT
2132 INIT_LIST_HEAD(&rpc_list);
2133 /* Local lock trees don't get reprocessed. */
2134 if (ns_is_client(ldlm_res_to_ns(res))) {
2141 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
2142 if (rc == LDLM_ITER_CONTINUE)
2143 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
2146 rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
2148 if (rc == -ERESTART) {
2149 LASSERT(list_empty(&rpc_list));
2155 INIT_LIST_HEAD(&rpc_list);
2156 if (!ns_is_client(ldlm_res_to_ns(res))) {
2157 CERROR("This is client-side-only module, cannot handle "
2158 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2164 EXPORT_SYMBOL(ldlm_reprocess_all);
2167 * Helper function to call blocking AST for LDLM lock \a lock in a
2168 * "cancelling" mode.
2170 void ldlm_cancel_callback(struct ldlm_lock *lock)
2172 check_res_locked(lock->l_resource);
2173 if (!ldlm_is_cancel(lock)) {
2174 ldlm_set_cancel(lock);
2175 if (lock->l_blocking_ast) {
2176 unlock_res_and_lock(lock);
2177 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
2179 lock_res_and_lock(lock);
2181 LDLM_DEBUG(lock, "no blocking ast");
2184 ldlm_set_bl_done(lock);
2188 * Remove skiplist-enabled LDLM lock \a req from granted list
2190 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
2192 if (req->l_resource->lr_type != LDLM_PLAIN &&
2193 req->l_resource->lr_type != LDLM_IBITS)
2196 list_del_init(&req->l_sl_policy);
2197 list_del_init(&req->l_sl_mode);
2201 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2203 void ldlm_lock_cancel(struct ldlm_lock *lock)
2205 struct ldlm_resource *res;
2206 struct ldlm_namespace *ns;
2209 lock_res_and_lock(lock);
2211 res = lock->l_resource;
2212 ns = ldlm_res_to_ns(res);
2214 /* Please do not, no matter how tempting, remove this LBUG without
2215 * talking to me first. -phik */
2216 if (lock->l_readers || lock->l_writers) {
2217 LDLM_ERROR(lock, "lock still has references");
2221 if (ldlm_is_waited(lock))
2222 ldlm_del_waiting_lock(lock);
2224 /* Releases cancel callback. */
2225 ldlm_cancel_callback(lock);
2227 LASSERT(!ldlm_is_waited(lock));
2229 ldlm_resource_unlink_lock(lock);
2230 ldlm_lock_destroy_nolock(lock);
2232 if (lock->l_granted_mode == lock->l_req_mode)
2233 ldlm_pool_del(&ns->ns_pool, lock);
2235 /* Make sure we will not be called again for same lock what is possible
2236 * if not to zero out lock->l_granted_mode */
2237 lock->l_granted_mode = LCK_MINMODE;
2238 unlock_res_and_lock(lock);
2242 EXPORT_SYMBOL(ldlm_lock_cancel);
2245 * Set opaque data into the lock that only makes sense to upper layer.
2247 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
2249 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2254 if (lock->l_ast_data == NULL)
2255 lock->l_ast_data = data;
2256 if (lock->l_ast_data == data)
2258 LDLM_LOCK_PUT(lock);
2262 EXPORT_SYMBOL(ldlm_lock_set_data);
2264 struct export_cl_data {
2265 struct obd_export *ecl_exp;
2270 * Iterator function for ldlm_cancel_locks_for_export.
2271 * Cancels passed locks.
2274 ldlm_cancel_locks_for_export_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd,
2275 struct hlist_node *hnode, void *data)
2278 struct export_cl_data *ecl = (struct export_cl_data *)data;
2279 struct obd_export *exp = ecl->ecl_exp;
2280 struct ldlm_lock *lock = cfs_hash_object(hs, hnode);
2281 struct ldlm_resource *res;
2283 res = ldlm_resource_getref(lock->l_resource);
2284 LDLM_LOCK_GET(lock);
2286 LDLM_DEBUG(lock, "export %p", exp);
2287 ldlm_res_lvbo_update(res, NULL, 1);
2288 ldlm_lock_cancel(lock);
2289 ldlm_reprocess_all(res);
2290 ldlm_resource_putref(res);
2291 LDLM_LOCK_RELEASE(lock);
2294 if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
2296 "Cancel lock %p for export %p (loop %d), still have "
2297 "%d locks left on hash table.\n",
2298 lock, exp, ecl->ecl_loop,
2299 atomic_read(&hs->hs_count));
2306 * Cancel all locks for given export.
2308 * Typically called on client disconnection/eviction
2310 void ldlm_cancel_locks_for_export(struct obd_export *exp)
2312 struct export_cl_data ecl = {
2317 cfs_hash_for_each_empty(exp->exp_lock_hash,
2318 ldlm_cancel_locks_for_export_cb, &ecl);
2322 * Downgrade an exclusive lock.
2324 * A fast variant of ldlm_lock_convert for convertion of exclusive
2325 * locks. The convertion is always successful.
2326 * Used by Commit on Sharing (COS) code.
2328 * \param lock A lock to convert
2329 * \param new_mode new lock mode
2331 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
2335 LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
2336 LASSERT(new_mode == LCK_COS);
2338 lock_res_and_lock(lock);
2339 ldlm_resource_unlink_lock(lock);
2341 * Remove the lock from pool as it will be added again in
2342 * ldlm_grant_lock() called below.
2344 ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
2346 lock->l_req_mode = new_mode;
2347 ldlm_grant_lock(lock, NULL);
2348 unlock_res_and_lock(lock);
2349 ldlm_reprocess_all(lock->l_resource);
2353 EXPORT_SYMBOL(ldlm_lock_downgrade);
2356 * Attempt to convert already granted lock to a different mode.
2358 * While lock conversion is not currently used, future client-side
2359 * optimizations could take advantage of it to avoid discarding cached
2362 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
2365 struct list_head rpc_list;
2366 struct ldlm_resource *res;
2367 struct ldlm_namespace *ns;
2369 #ifdef HAVE_SERVER_SUPPORT
2371 struct sl_insert_point prev;
2373 struct ldlm_interval *node;
2376 INIT_LIST_HEAD(&rpc_list);
2377 /* Just return if mode is unchanged. */
2378 if (new_mode == lock->l_granted_mode) {
2379 *flags |= LDLM_FL_BLOCK_GRANTED;
2380 RETURN(lock->l_resource);
2383 /* I can't check the type of lock here because the bitlock of lock
2384 * is not held here, so do the allocation blindly. -jay */
2385 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
2386 if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */
2389 LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
2390 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2392 lock_res_and_lock(lock);
2394 res = lock->l_resource;
2395 ns = ldlm_res_to_ns(res);
2397 #ifdef HAVE_SERVER_SUPPORT
2398 old_mode = lock->l_req_mode;
2400 lock->l_req_mode = new_mode;
2401 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2402 #ifdef HAVE_SERVER_SUPPORT
2403 /* remember the lock position where the lock might be
2404 * added back to the granted list later and also
2405 * remember the join mode for skiplist fixing. */
2406 prev.res_link = lock->l_res_link.prev;
2407 prev.mode_link = lock->l_sl_mode.prev;
2408 prev.policy_link = lock->l_sl_policy.prev;
2410 ldlm_resource_unlink_lock(lock);
2412 ldlm_resource_unlink_lock(lock);
2413 if (res->lr_type == LDLM_EXTENT) {
2414 /* FIXME: ugly code, I have to attach the lock to a
2415 * interval node again since perhaps it will be granted
2417 INIT_LIST_HEAD(&node->li_group);
2418 ldlm_interval_attach(node, lock);
2424 * Remove old lock from the pool before adding the lock with new
2425 * mode below in ->policy()
2427 ldlm_pool_del(&ns->ns_pool, lock);
2429 /* If this is a local resource, put it on the appropriate list. */
2430 if (ns_is_client(ldlm_res_to_ns(res))) {
2431 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2432 ldlm_resource_add_lock(res, &res->lr_converting, lock);
2434 /* This should never happen, because of the way the
2435 * server handles conversions. */
2436 LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2440 ldlm_grant_lock(lock, &rpc_list);
2442 /* FIXME: completion handling not with lr_lock held ! */
2443 if (lock->l_completion_ast)
2444 lock->l_completion_ast(lock, 0, NULL);
2446 #ifdef HAVE_SERVER_SUPPORT
2451 ldlm_processing_policy policy;
2452 policy = ldlm_processing_policy_table[res->lr_type];
2453 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2454 if (rc == LDLM_ITER_STOP) {
2455 lock->l_req_mode = old_mode;
2456 if (res->lr_type == LDLM_EXTENT)
2457 ldlm_extent_add_lock(res, lock);
2459 ldlm_granted_list_add_lock(lock, &prev);
2463 *flags |= LDLM_FL_BLOCK_GRANTED;
2469 CERROR("This is client-side-only module, cannot handle "
2470 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2474 unlock_res_and_lock(lock);
2477 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2479 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2484 * Print lock with lock handle \a lockh description into debug log.
2486 * Used when printing all locks on a resource for debug purposes.
2488 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
2490 struct ldlm_lock *lock;
2492 if (!((libcfs_debug | D_ERROR) & level))
2495 lock = ldlm_handle2lock(lockh);
2499 LDLM_DEBUG_LIMIT(level, lock, "###");
2501 LDLM_LOCK_PUT(lock);
2503 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2506 * Print lock information with custom message into debug log.
2509 void _ldlm_lock_debug(struct ldlm_lock *lock,
2510 struct libcfs_debug_msg_data *msgdata,
2511 const char *fmt, ...)
2514 struct obd_export *exp = lock->l_export;
2515 struct ldlm_resource *resource = lock->l_resource;
2516 char *nid = "local";
2518 va_start(args, fmt);
2520 if (exp && exp->exp_connection) {
2521 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2522 } else if (exp && exp->exp_obd != NULL) {
2523 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2524 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2527 if (resource == NULL) {
2528 libcfs_debug_vmsg2(msgdata, fmt, args,
2529 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2530 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2531 "remote: "LPX64" expref: %d pid: %u timeout: %lu "
2534 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2535 lock->l_readers, lock->l_writers,
2536 ldlm_lockname[lock->l_granted_mode],
2537 ldlm_lockname[lock->l_req_mode],
2538 lock->l_flags, nid, lock->l_remote_handle.cookie,
2539 exp ? atomic_read(&exp->exp_refcount) : -99,
2540 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2545 switch (resource->lr_type) {
2547 libcfs_debug_vmsg2(msgdata, fmt, args,
2548 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2549 "res: "DLDLMRES" rrc: %d type: %s ["LPU64"->"LPU64"] "
2550 "(req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote: "
2551 LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2552 ldlm_lock_to_ns_name(lock), lock,
2553 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2554 lock->l_readers, lock->l_writers,
2555 ldlm_lockname[lock->l_granted_mode],
2556 ldlm_lockname[lock->l_req_mode],
2558 atomic_read(&resource->lr_refcount),
2559 ldlm_typename[resource->lr_type],
2560 lock->l_policy_data.l_extent.start,
2561 lock->l_policy_data.l_extent.end,
2562 lock->l_req_extent.start, lock->l_req_extent.end,
2563 lock->l_flags, nid, lock->l_remote_handle.cookie,
2564 exp ? atomic_read(&exp->exp_refcount) : -99,
2565 lock->l_pid, lock->l_callback_timeout,
2570 libcfs_debug_vmsg2(msgdata, fmt, args,
2571 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2572 "res: "DLDLMRES" rrc: %d type: %s pid: %d "
2573 "["LPU64"->"LPU64"] flags: "LPX64" nid: %s "
2574 "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2575 ldlm_lock_to_ns_name(lock), lock,
2576 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2577 lock->l_readers, lock->l_writers,
2578 ldlm_lockname[lock->l_granted_mode],
2579 ldlm_lockname[lock->l_req_mode],
2581 atomic_read(&resource->lr_refcount),
2582 ldlm_typename[resource->lr_type],
2583 lock->l_policy_data.l_flock.pid,
2584 lock->l_policy_data.l_flock.start,
2585 lock->l_policy_data.l_flock.end,
2586 lock->l_flags, nid, lock->l_remote_handle.cookie,
2587 exp ? atomic_read(&exp->exp_refcount) : -99,
2588 lock->l_pid, lock->l_callback_timeout);
2592 libcfs_debug_vmsg2(msgdata, fmt, args,
2593 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2594 "res: "DLDLMRES" bits "LPX64" rrc: %d type: %s "
2595 "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2596 "pid: %u timeout: %lu lvb_type: %d\n",
2597 ldlm_lock_to_ns_name(lock),
2598 lock, lock->l_handle.h_cookie,
2599 atomic_read(&lock->l_refc),
2600 lock->l_readers, lock->l_writers,
2601 ldlm_lockname[lock->l_granted_mode],
2602 ldlm_lockname[lock->l_req_mode],
2604 lock->l_policy_data.l_inodebits.bits,
2605 atomic_read(&resource->lr_refcount),
2606 ldlm_typename[resource->lr_type],
2607 lock->l_flags, nid, lock->l_remote_handle.cookie,
2608 exp ? atomic_read(&exp->exp_refcount) : -99,
2609 lock->l_pid, lock->l_callback_timeout,
2614 libcfs_debug_vmsg2(msgdata, fmt, args,
2615 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2616 "res: "DLDLMRES" rrc: %d type: %s flags: "LPX64" "
2617 "nid: %s remote: "LPX64" expref: %d pid: %u "
2618 "timeout: %lu lvb_type: %d\n",
2619 ldlm_lock_to_ns_name(lock),
2620 lock, lock->l_handle.h_cookie,
2621 atomic_read(&lock->l_refc),
2622 lock->l_readers, lock->l_writers,
2623 ldlm_lockname[lock->l_granted_mode],
2624 ldlm_lockname[lock->l_req_mode],
2626 atomic_read(&resource->lr_refcount),
2627 ldlm_typename[resource->lr_type],
2628 lock->l_flags, nid, lock->l_remote_handle.cookie,
2629 exp ? atomic_read(&exp->exp_refcount) : -99,
2630 lock->l_pid, lock->l_callback_timeout,
2636 EXPORT_SYMBOL(_ldlm_lock_debug);