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(enum ldlm_type type,
87 const union ldlm_policy_data *lpolicy,
88 union ldlm_wire_policy_data *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, enum ldlm_type type,
101 const union ldlm_wire_policy_data *wpolicy,
102 union ldlm_policy_data *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_exp_list));
202 LASSERT(list_empty(&lock->l_res_link));
203 LASSERT(list_empty(&lock->l_pending_chain));
205 lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats,
207 lu_ref_del(&res->lr_reference, "lock", lock);
208 ldlm_resource_putref(res);
209 lock->l_resource = NULL;
210 if (lock->l_export) {
211 class_export_lock_put(lock->l_export, lock);
212 lock->l_export = NULL;
215 if (lock->l_lvb_data != NULL)
216 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
218 ldlm_interval_free(ldlm_interval_detach(lock));
219 lu_ref_fini(&lock->l_reference);
220 OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle);
225 EXPORT_SYMBOL(ldlm_lock_put);
228 * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked.
230 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
233 if (!list_empty(&lock->l_lru)) {
234 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
236 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
237 list_del_init(&lock->l_lru);
238 LASSERT(ns->ns_nr_unused > 0);
246 * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first.
248 * If \a last_use is non-zero, it will remove the lock from LRU only if
249 * it matches lock's l_last_used.
251 * \retval 0 if \a last_use is set, the lock is not in LRU list or \a last_use
252 * doesn't match lock's l_last_used;
253 * otherwise, the lock hasn't been in the LRU list.
254 * \retval 1 the lock was in LRU list and removed.
256 int ldlm_lock_remove_from_lru_check(struct ldlm_lock *lock, cfs_time_t last_use)
258 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
262 if (ldlm_is_ns_srv(lock)) {
263 LASSERT(list_empty(&lock->l_lru));
267 spin_lock(&ns->ns_lock);
268 if (last_use == 0 || last_use == lock->l_last_used)
269 rc = ldlm_lock_remove_from_lru_nolock(lock);
270 spin_unlock(&ns->ns_lock);
276 * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked.
278 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
280 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
282 lock->l_last_used = cfs_time_current();
283 LASSERT(list_empty(&lock->l_lru));
284 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
285 list_add_tail(&lock->l_lru, &ns->ns_unused_list);
286 ldlm_clear_skipped(lock);
287 LASSERT(ns->ns_nr_unused >= 0);
292 * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks
295 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
297 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
300 spin_lock(&ns->ns_lock);
301 ldlm_lock_add_to_lru_nolock(lock);
302 spin_unlock(&ns->ns_lock);
307 * Moves LDLM lock \a lock that is already in namespace LRU to the tail of
308 * the LRU. Performs necessary LRU locking
310 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
312 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
315 if (ldlm_is_ns_srv(lock)) {
316 LASSERT(list_empty(&lock->l_lru));
321 spin_lock(&ns->ns_lock);
322 if (!list_empty(&lock->l_lru)) {
323 ldlm_lock_remove_from_lru_nolock(lock);
324 ldlm_lock_add_to_lru_nolock(lock);
326 spin_unlock(&ns->ns_lock);
331 * Helper to destroy a locked lock.
333 * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock
334 * Must be called with l_lock and lr_lock held.
336 * Does not actually free the lock data, but rather marks the lock as
337 * destroyed by setting l_destroyed field in the lock to 1. Destroys a
338 * handle->lock association too, so that the lock can no longer be found
339 * and removes the lock from LRU list. Actual lock freeing occurs when
340 * last lock reference goes away.
342 * Original comment (of some historical value):
343 * This used to have a 'strict' flag, which recovery would use to mark an
344 * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
345 * shall explain why it's gone: with the new hash table scheme, once you call
346 * ldlm_lock_destroy, you can never drop your final references on this lock.
347 * Because it's not in the hash table anymore. -phil
349 static int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
353 if (lock->l_readers || lock->l_writers) {
354 LDLM_ERROR(lock, "lock still has references");
358 if (!list_empty(&lock->l_res_link)) {
359 LDLM_ERROR(lock, "lock still on resource");
363 if (ldlm_is_destroyed(lock)) {
364 LASSERT(list_empty(&lock->l_lru));
368 ldlm_set_destroyed(lock);
370 if (lock->l_export && lock->l_export->exp_lock_hash) {
371 /* NB: it's safe to call cfs_hash_del() even lock isn't
372 * in exp_lock_hash. */
373 /* In the function below, .hs_keycmp resolves to
374 * ldlm_export_lock_keycmp() */
375 /* coverity[overrun-buffer-val] */
376 cfs_hash_del(lock->l_export->exp_lock_hash,
377 &lock->l_remote_handle, &lock->l_exp_hash);
380 ldlm_lock_remove_from_lru(lock);
381 class_handle_unhash(&lock->l_handle);
384 /* Wake anyone waiting for this lock */
385 /* FIXME: I should probably add yet another flag, instead of using
386 * l_export to only call this on clients */
388 class_export_put(lock->l_export);
389 lock->l_export = NULL;
390 if (lock->l_export && lock->l_completion_ast)
391 lock->l_completion_ast(lock, 0);
398 * Destroys a LDLM lock \a lock. Performs necessary locking first.
400 void ldlm_lock_destroy(struct ldlm_lock *lock)
404 lock_res_and_lock(lock);
405 first = ldlm_lock_destroy_internal(lock);
406 unlock_res_and_lock(lock);
408 /* drop reference from hashtable only for first destroy */
410 lu_ref_del(&lock->l_reference, "hash", lock);
411 LDLM_LOCK_RELEASE(lock);
417 * Destroys a LDLM lock \a lock that is already locked.
419 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
423 first = ldlm_lock_destroy_internal(lock);
424 /* drop reference from hashtable only for first destroy */
426 lu_ref_del(&lock->l_reference, "hash", lock);
427 LDLM_LOCK_RELEASE(lock);
432 /* this is called by portals_handle2object with the handle lock taken */
433 static void lock_handle_addref(void *lock)
435 LDLM_LOCK_GET((struct ldlm_lock *)lock);
438 static void lock_handle_free(void *lock, int size)
440 LASSERT(size == sizeof(struct ldlm_lock));
441 OBD_SLAB_FREE(lock, ldlm_lock_slab, size);
444 static struct portals_handle_ops lock_handle_ops = {
445 .hop_addref = lock_handle_addref,
446 .hop_free = lock_handle_free,
451 * Allocate and initialize new lock structure.
453 * usage: pass in a resource on which you have done ldlm_resource_get
454 * new lock will take over the refcount.
455 * returns: lock with refcount 2 - one for current caller and one for remote
457 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
459 struct ldlm_lock *lock;
462 if (resource == NULL)
465 OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, GFP_NOFS);
469 spin_lock_init(&lock->l_lock);
470 lock->l_resource = resource;
471 lu_ref_add(&resource->lr_reference, "lock", lock);
473 atomic_set(&lock->l_refc, 2);
474 INIT_LIST_HEAD(&lock->l_res_link);
475 INIT_LIST_HEAD(&lock->l_lru);
476 INIT_LIST_HEAD(&lock->l_pending_chain);
477 INIT_LIST_HEAD(&lock->l_bl_ast);
478 INIT_LIST_HEAD(&lock->l_cp_ast);
479 INIT_LIST_HEAD(&lock->l_rk_ast);
480 init_waitqueue_head(&lock->l_waitq);
481 lock->l_blocking_lock = NULL;
482 INIT_LIST_HEAD(&lock->l_sl_mode);
483 INIT_LIST_HEAD(&lock->l_sl_policy);
484 INIT_HLIST_NODE(&lock->l_exp_hash);
485 INIT_HLIST_NODE(&lock->l_exp_flock_hash);
487 lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats,
489 INIT_LIST_HEAD(&lock->l_handle.h_link);
490 class_handle_hash(&lock->l_handle, &lock_handle_ops);
492 lu_ref_init(&lock->l_reference);
493 lu_ref_add(&lock->l_reference, "hash", lock);
494 lock->l_callback_timeout = 0;
496 #if LUSTRE_TRACKS_LOCK_EXP_REFS
497 INIT_LIST_HEAD(&lock->l_exp_refs_link);
498 lock->l_exp_refs_nr = 0;
499 lock->l_exp_refs_target = NULL;
501 INIT_LIST_HEAD(&lock->l_exp_list);
507 * Moves LDLM lock \a lock to another resource.
508 * This is used on client when server returns some other lock than requested
509 * (typically as a result of intent operation)
511 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
512 const struct ldlm_res_id *new_resid)
514 struct ldlm_resource *oldres = lock->l_resource;
515 struct ldlm_resource *newres;
519 LASSERT(ns_is_client(ns));
521 lock_res_and_lock(lock);
522 if (memcmp(new_resid, &lock->l_resource->lr_name,
523 sizeof(lock->l_resource->lr_name)) == 0) {
525 unlock_res_and_lock(lock);
529 LASSERT(new_resid->name[0] != 0);
531 /* This function assumes that the lock isn't on any lists */
532 LASSERT(list_empty(&lock->l_res_link));
534 type = oldres->lr_type;
535 unlock_res_and_lock(lock);
537 newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
539 RETURN(PTR_ERR(newres));
541 lu_ref_add(&newres->lr_reference, "lock", lock);
543 * To flip the lock from the old to the new resource, lock, oldres and
544 * newres have to be locked. Resource spin-locks are nested within
545 * lock->l_lock, and are taken in the memory address order to avoid
548 spin_lock(&lock->l_lock);
549 oldres = lock->l_resource;
550 if (oldres < newres) {
552 lock_res_nested(newres, LRT_NEW);
555 lock_res_nested(oldres, LRT_NEW);
557 LASSERT(memcmp(new_resid, &oldres->lr_name,
558 sizeof oldres->lr_name) != 0);
559 lock->l_resource = newres;
561 unlock_res_and_lock(lock);
563 /* ...and the flowers are still standing! */
564 lu_ref_del(&oldres->lr_reference, "lock", lock);
565 ldlm_resource_putref(oldres);
570 /** \defgroup ldlm_handles LDLM HANDLES
571 * Ways to get hold of locks without any addresses.
576 * Fills in handle for LDLM lock \a lock into supplied \a lockh
577 * Does not take any references.
579 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
581 lockh->cookie = lock->l_handle.h_cookie;
583 EXPORT_SYMBOL(ldlm_lock2handle);
586 * Obtain a lock reference by handle.
588 * if \a flags: atomically get the lock and set the flags.
589 * Return NULL if flag already set
591 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
594 struct ldlm_lock *lock;
599 lock = class_handle2object(handle->cookie, NULL);
603 /* It's unlikely but possible that someone marked the lock as
604 * destroyed after we did handle2object on it */
605 if ((flags == 0) && !ldlm_is_destroyed(lock)) {
606 lu_ref_add(&lock->l_reference, "handle", current);
610 lock_res_and_lock(lock);
612 LASSERT(lock->l_resource != NULL);
614 lu_ref_add_atomic(&lock->l_reference, "handle", current);
615 if (unlikely(ldlm_is_destroyed(lock))) {
616 unlock_res_and_lock(lock);
617 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
622 /* If we're setting flags, make sure none of them are already set. */
624 if ((lock->l_flags & flags) != 0) {
625 unlock_res_and_lock(lock);
630 lock->l_flags |= flags;
633 unlock_res_and_lock(lock);
636 EXPORT_SYMBOL(__ldlm_handle2lock);
637 /** @} ldlm_handles */
640 * Fill in "on the wire" representation for given LDLM lock into supplied
641 * lock descriptor \a desc structure.
643 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
645 ldlm_res2desc(lock->l_resource, &desc->l_resource);
646 desc->l_req_mode = lock->l_req_mode;
647 desc->l_granted_mode = lock->l_granted_mode;
648 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
649 &lock->l_policy_data,
650 &desc->l_policy_data);
654 * Add a lock to list of conflicting locks to send AST to.
656 * Only add if we have not sent a blocking AST to the lock yet.
658 static void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
659 struct list_head *work_list)
661 if (!ldlm_is_ast_sent(lock)) {
662 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
663 ldlm_set_ast_sent(lock);
664 /* If the enqueuing client said so, tell the AST recipient to
665 * discard dirty data, rather than writing back. */
666 if (ldlm_is_ast_discard_data(new))
667 ldlm_set_discard_data(lock);
668 LASSERT(list_empty(&lock->l_bl_ast));
669 list_add(&lock->l_bl_ast, work_list);
671 LASSERT(lock->l_blocking_lock == NULL);
672 lock->l_blocking_lock = LDLM_LOCK_GET(new);
677 * Add a lock to list of just granted locks to send completion AST to.
679 static void ldlm_add_cp_work_item(struct ldlm_lock *lock,
680 struct list_head *work_list)
682 if (!ldlm_is_cp_reqd(lock)) {
683 ldlm_set_cp_reqd(lock);
684 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
685 LASSERT(list_empty(&lock->l_cp_ast));
686 list_add(&lock->l_cp_ast, work_list);
692 * Aggregator function to add AST work items into a list. Determines
693 * what sort of an AST work needs to be done and calls the proper
695 * Must be called with lr_lock held.
697 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
698 struct list_head *work_list)
701 check_res_locked(lock->l_resource);
703 ldlm_add_bl_work_item(lock, new, work_list);
705 ldlm_add_cp_work_item(lock, work_list);
710 * Add specified reader/writer reference to LDLM lock with handle \a lockh.
711 * r/w reference type is determined by \a mode
712 * Calls ldlm_lock_addref_internal.
714 void ldlm_lock_addref(struct lustre_handle *lockh, enum ldlm_mode mode)
716 struct ldlm_lock *lock;
718 lock = ldlm_handle2lock(lockh);
719 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
720 ldlm_lock_addref_internal(lock, mode);
723 EXPORT_SYMBOL(ldlm_lock_addref);
727 * Add specified reader/writer reference to LDLM lock \a lock.
728 * r/w reference type is determined by \a mode
729 * Removes lock from LRU if it is there.
730 * Assumes the LDLM lock is already locked.
732 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock,
735 ldlm_lock_remove_from_lru(lock);
736 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
738 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
740 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
742 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
745 lu_ref_add_atomic(&lock->l_reference, "user", lock);
746 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
750 * Attempts to add reader/writer reference to a lock with handle \a lockh, and
751 * fails if lock is already LDLM_FL_CBPENDING or destroyed.
753 * \retval 0 success, lock was addref-ed
755 * \retval -EAGAIN lock is being canceled.
757 int ldlm_lock_addref_try(struct lustre_handle *lockh, enum ldlm_mode mode)
759 struct ldlm_lock *lock;
763 lock = ldlm_handle2lock(lockh);
765 lock_res_and_lock(lock);
766 if (lock->l_readers != 0 || lock->l_writers != 0 ||
767 !ldlm_is_cbpending(lock)) {
768 ldlm_lock_addref_internal_nolock(lock, mode);
771 unlock_res_and_lock(lock);
776 EXPORT_SYMBOL(ldlm_lock_addref_try);
779 * Add specified reader/writer reference to LDLM lock \a lock.
780 * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work.
781 * Only called for local locks.
783 void ldlm_lock_addref_internal(struct ldlm_lock *lock, enum ldlm_mode mode)
785 lock_res_and_lock(lock);
786 ldlm_lock_addref_internal_nolock(lock, mode);
787 unlock_res_and_lock(lock);
791 * Removes reader/writer reference for LDLM lock \a lock.
792 * Assumes LDLM lock is already locked.
793 * only called in ldlm_flock_destroy and for local locks.
794 * Does NOT add lock to LRU if no r/w references left to accomodate flock locks
795 * that cannot be placed in LRU.
797 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock,
800 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
801 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
802 LASSERT(lock->l_readers > 0);
803 lu_ref_del(&lock->l_reference, "reader", lock);
806 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
807 LASSERT(lock->l_writers > 0);
808 lu_ref_del(&lock->l_reference, "writer", lock);
812 lu_ref_del(&lock->l_reference, "user", lock);
813 LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */
817 * Removes reader/writer reference for LDLM lock \a lock.
818 * Locks LDLM lock first.
819 * If the lock is determined to be client lock on a client and r/w refcount
820 * drops to zero and the lock is not blocked, the lock is added to LRU lock
822 * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called.
824 void ldlm_lock_decref_internal(struct ldlm_lock *lock, enum ldlm_mode mode)
826 struct ldlm_namespace *ns;
829 lock_res_and_lock(lock);
831 ns = ldlm_lock_to_ns(lock);
833 ldlm_lock_decref_internal_nolock(lock, mode);
835 if (ldlm_is_local(lock) &&
836 !lock->l_readers && !lock->l_writers) {
837 /* If this is a local lock on a server namespace and this was
838 * the last reference, cancel the lock. */
839 CDEBUG(D_INFO, "forcing cancel of local lock\n");
840 ldlm_set_cbpending(lock);
843 if (!lock->l_readers && !lock->l_writers &&
844 (ldlm_is_cbpending(lock) || lock->l_req_mode == LCK_GROUP)) {
845 /* If we received a blocked AST and this was the last reference,
847 * Group locks are special:
848 * They must not go in LRU, but they are not called back
849 * like non-group locks, instead they are manually released.
850 * They have an l_writers reference which they keep until
851 * they are manually released, so we remove them when they have
852 * no more reader or writer references. - LU-6368 */
853 if (ldlm_is_ns_srv(lock) && lock->l_export)
854 CERROR("FL_CBPENDING set on non-local lock--just a "
857 LDLM_DEBUG(lock, "final decref done on cbpending lock");
859 LDLM_LOCK_GET(lock); /* dropped by bl thread */
860 ldlm_lock_remove_from_lru(lock);
861 unlock_res_and_lock(lock);
863 if (ldlm_is_fail_loc(lock))
864 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
866 if (ldlm_is_atomic_cb(lock) ||
867 ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
868 ldlm_handle_bl_callback(ns, NULL, lock);
869 } else if (ns_is_client(ns) &&
870 !lock->l_readers && !lock->l_writers &&
871 !ldlm_is_no_lru(lock) &&
872 !ldlm_is_bl_ast(lock)) {
874 LDLM_DEBUG(lock, "add lock into lru list");
876 /* If this is a client-side namespace and this was the last
877 * reference, put it on the LRU. */
878 ldlm_lock_add_to_lru(lock);
879 unlock_res_and_lock(lock);
881 if (ldlm_is_fail_loc(lock))
882 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
884 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
885 * are not supported by the server, otherwise, it is done on
887 if (!exp_connect_cancelset(lock->l_conn_export) &&
888 !ns_connect_lru_resize(ns))
889 ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0);
891 LDLM_DEBUG(lock, "do not add lock into lru list");
892 unlock_res_and_lock(lock);
899 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
901 void ldlm_lock_decref(struct lustre_handle *lockh, enum ldlm_mode mode)
903 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
904 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
905 ldlm_lock_decref_internal(lock, mode);
908 EXPORT_SYMBOL(ldlm_lock_decref);
911 * Decrease reader/writer refcount for LDLM lock with handle
912 * \a lockh and mark it for subsequent cancellation once r/w refcount
913 * drops to zero instead of putting into LRU.
916 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh,
919 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
922 LASSERT(lock != NULL);
924 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
925 lock_res_and_lock(lock);
926 ldlm_set_cbpending(lock);
927 unlock_res_and_lock(lock);
928 ldlm_lock_decref_internal(lock, mode);
931 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
933 struct sl_insert_point {
934 struct list_head *res_link;
935 struct list_head *mode_link;
936 struct list_head *policy_link;
940 * Finds a position to insert the new lock into granted lock list.
942 * Used for locks eligible for skiplist optimization.
945 * queue [input]: the granted list where search acts on;
946 * req [input]: the lock whose position to be located;
947 * prev [output]: positions within 3 lists to insert @req to
951 * - ldlm_grant_lock_with_skiplist
953 static void search_granted_lock(struct list_head *queue,
954 struct ldlm_lock *req,
955 struct sl_insert_point *prev)
957 struct list_head *tmp;
958 struct ldlm_lock *lock, *mode_end, *policy_end;
961 list_for_each(tmp, queue) {
962 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
964 mode_end = list_entry(lock->l_sl_mode.prev,
965 struct ldlm_lock, l_sl_mode);
967 if (lock->l_req_mode != req->l_req_mode) {
968 /* jump to last lock of mode group */
969 tmp = &mode_end->l_res_link;
973 /* suitable mode group is found */
974 if (lock->l_resource->lr_type == LDLM_PLAIN) {
975 /* insert point is last lock of the mode group */
976 prev->res_link = &mode_end->l_res_link;
977 prev->mode_link = &mode_end->l_sl_mode;
978 prev->policy_link = &req->l_sl_policy;
981 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
984 list_entry(lock->l_sl_policy.prev,
988 if (lock->l_policy_data.l_inodebits.bits ==
989 req->l_policy_data.l_inodebits.bits) {
990 /* insert point is last lock of
991 * the policy group */
993 &policy_end->l_res_link;
995 &policy_end->l_sl_mode;
997 &policy_end->l_sl_policy;
1002 if (policy_end == mode_end)
1003 /* done with mode group */
1006 /* go to next policy group within mode group */
1007 tmp = policy_end->l_res_link.next;
1008 lock = list_entry(tmp, struct ldlm_lock,
1010 } /* loop over policy groups within the mode group */
1012 /* insert point is last lock of the mode group,
1013 * new policy group is started */
1014 prev->res_link = &mode_end->l_res_link;
1015 prev->mode_link = &mode_end->l_sl_mode;
1016 prev->policy_link = &req->l_sl_policy;
1020 LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
1025 /* insert point is last lock on the queue,
1026 * new mode group and new policy group are started */
1027 prev->res_link = queue->prev;
1028 prev->mode_link = &req->l_sl_mode;
1029 prev->policy_link = &req->l_sl_policy;
1035 * Add a lock into resource granted list after a position described by
1038 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
1039 struct sl_insert_point *prev)
1041 struct ldlm_resource *res = lock->l_resource;
1044 check_res_locked(res);
1046 ldlm_resource_dump(D_INFO, res);
1047 LDLM_DEBUG(lock, "About to add lock:");
1049 if (ldlm_is_destroyed(lock)) {
1050 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1054 LASSERT(list_empty(&lock->l_res_link));
1055 LASSERT(list_empty(&lock->l_sl_mode));
1056 LASSERT(list_empty(&lock->l_sl_policy));
1059 * lock->link == prev->link means lock is first starting the group.
1060 * Don't re-add to itself to suppress kernel warnings.
1062 if (&lock->l_res_link != prev->res_link)
1063 list_add(&lock->l_res_link, prev->res_link);
1064 if (&lock->l_sl_mode != prev->mode_link)
1065 list_add(&lock->l_sl_mode, prev->mode_link);
1066 if (&lock->l_sl_policy != prev->policy_link)
1067 list_add(&lock->l_sl_policy, prev->policy_link);
1073 * Add a lock to granted list on a resource maintaining skiplist
1076 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
1078 struct sl_insert_point prev;
1081 LASSERT(lock->l_req_mode == lock->l_granted_mode);
1083 search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
1084 ldlm_granted_list_add_lock(lock, &prev);
1089 * Perform lock granting bookkeeping.
1091 * Includes putting the lock into granted list and updating lock mode.
1093 * - ldlm_lock_enqueue
1094 * - ldlm_reprocess_queue
1095 * - ldlm_lock_convert
1097 * must be called with lr_lock held
1099 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
1101 struct ldlm_resource *res = lock->l_resource;
1104 check_res_locked(res);
1106 lock->l_granted_mode = lock->l_req_mode;
1108 if (work_list && lock->l_completion_ast != NULL)
1109 ldlm_add_ast_work_item(lock, NULL, work_list);
1111 /* We should not add locks to granted list in the following cases:
1112 * - this is an UNLOCK but not a real lock;
1113 * - this is a TEST lock;
1114 * - this is a F_CANCELLK lock (async flock has req_mode == 0)
1115 * - this is a deadlock (flock cannot be granted) */
1116 if (lock->l_req_mode == 0 ||
1117 lock->l_req_mode == LCK_NL ||
1118 ldlm_is_test_lock(lock) ||
1119 ldlm_is_flock_deadlock(lock))
1122 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
1123 ldlm_grant_lock_with_skiplist(lock);
1124 else if (res->lr_type == LDLM_EXTENT)
1125 ldlm_extent_add_lock(res, lock);
1127 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1129 ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1134 * Describe the overlap between two locks. itree_overlap_cb data.
1136 struct lock_match_data {
1137 struct ldlm_lock *lmd_old;
1138 struct ldlm_lock *lmd_lock;
1139 enum ldlm_mode *lmd_mode;
1140 union ldlm_policy_data *lmd_policy;
1146 * Check if the given @lock meets the criteria for a match.
1147 * A reference on the lock is taken if matched.
1149 * \param lock test-against this lock
1150 * \param data parameters
1152 static int lock_matches(struct ldlm_lock *lock, struct lock_match_data *data)
1154 union ldlm_policy_data *lpol = &lock->l_policy_data;
1155 enum ldlm_mode match;
1157 if (lock == data->lmd_old)
1158 return INTERVAL_ITER_STOP;
1160 /* Check if this lock can be matched.
1161 * Used by LU-2919(exclusive open) for open lease lock */
1162 if (ldlm_is_excl(lock))
1163 return INTERVAL_ITER_CONT;
1165 /* llite sometimes wants to match locks that will be
1166 * canceled when their users drop, but we allow it to match
1167 * if it passes in CBPENDING and the lock still has users.
1168 * this is generally only going to be used by children
1169 * whose parents already hold a lock so forward progress
1170 * can still happen. */
1171 if (ldlm_is_cbpending(lock) &&
1172 !(data->lmd_flags & LDLM_FL_CBPENDING))
1173 return INTERVAL_ITER_CONT;
1174 if (!data->lmd_unref && ldlm_is_cbpending(lock) &&
1175 lock->l_readers == 0 && lock->l_writers == 0)
1176 return INTERVAL_ITER_CONT;
1178 if (!(lock->l_req_mode & *data->lmd_mode))
1179 return INTERVAL_ITER_CONT;
1180 match = lock->l_req_mode;
1182 switch (lock->l_resource->lr_type) {
1184 if (lpol->l_extent.start > data->lmd_policy->l_extent.start ||
1185 lpol->l_extent.end < data->lmd_policy->l_extent.end)
1186 return INTERVAL_ITER_CONT;
1188 if (unlikely(match == LCK_GROUP) &&
1189 data->lmd_policy->l_extent.gid != LDLM_GID_ANY &&
1190 lpol->l_extent.gid != data->lmd_policy->l_extent.gid)
1191 return INTERVAL_ITER_CONT;
1194 /* We match if we have existing lock with same or wider set
1196 if ((lpol->l_inodebits.bits &
1197 data->lmd_policy->l_inodebits.bits) !=
1198 data->lmd_policy->l_inodebits.bits)
1199 return INTERVAL_ITER_CONT;
1205 /* We match if we have existing lock with same or wider set
1207 if (!data->lmd_unref && LDLM_HAVE_MASK(lock, GONE))
1208 return INTERVAL_ITER_CONT;
1210 if (!equi(data->lmd_flags & LDLM_FL_LOCAL_ONLY, ldlm_is_local(lock)))
1211 return INTERVAL_ITER_CONT;
1213 if (data->lmd_flags & LDLM_FL_TEST_LOCK) {
1214 LDLM_LOCK_GET(lock);
1215 ldlm_lock_touch_in_lru(lock);
1217 ldlm_lock_addref_internal_nolock(lock, match);
1220 *data->lmd_mode = match;
1221 data->lmd_lock = lock;
1223 return INTERVAL_ITER_STOP;
1226 static unsigned int itree_overlap_cb(struct interval_node *in, void *args)
1228 struct ldlm_interval *node = to_ldlm_interval(in);
1229 struct lock_match_data *data = args;
1230 struct ldlm_lock *lock;
1233 list_for_each_entry(lock, &node->li_group, l_sl_policy) {
1234 rc = lock_matches(lock, data);
1235 if (rc == INTERVAL_ITER_STOP)
1236 return INTERVAL_ITER_STOP;
1238 return INTERVAL_ITER_CONT;
1242 * Search for a lock with given parameters in interval trees.
1244 * \param res search for a lock in this resource
1245 * \param data parameters
1247 * \retval a referenced lock or NULL.
1249 static struct ldlm_lock *search_itree(struct ldlm_resource *res,
1250 struct lock_match_data *data)
1252 struct interval_node_extent ext = {
1253 .start = data->lmd_policy->l_extent.start,
1254 .end = data->lmd_policy->l_extent.end
1258 for (idx = 0; idx < LCK_MODE_NUM; idx++) {
1259 struct ldlm_interval_tree *tree = &res->lr_itree[idx];
1261 if (tree->lit_root == NULL)
1264 if (!(tree->lit_mode & *data->lmd_mode))
1267 interval_search(tree->lit_root, &ext,
1268 itree_overlap_cb, data);
1270 return data->lmd_lock;
1275 * Search for a lock with given properties in a queue.
1277 * \param queue search for a lock in this queue
1278 * \param data parameters
1280 * \retval a referenced lock or NULL.
1282 static struct ldlm_lock *search_queue(struct list_head *queue,
1283 struct lock_match_data *data)
1285 struct ldlm_lock *lock;
1288 list_for_each_entry(lock, queue, l_res_link) {
1289 rc = lock_matches(lock, data);
1290 if (rc == INTERVAL_ITER_STOP)
1291 return data->lmd_lock;
1296 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1298 if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) {
1299 lock->l_flags |= LDLM_FL_FAIL_NOTIFIED;
1300 wake_up_all(&lock->l_waitq);
1303 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1305 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1307 lock_res_and_lock(lock);
1308 ldlm_lock_fail_match_locked(lock);
1309 unlock_res_and_lock(lock);
1313 * Mark lock as "matchable" by OST.
1315 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1317 * Assumes LDLM lock is already locked.
1319 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1321 ldlm_set_lvb_ready(lock);
1322 wake_up_all(&lock->l_waitq);
1324 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1327 * Mark lock as "matchable" by OST.
1328 * Locks the lock and then \see ldlm_lock_allow_match_locked
1330 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1332 lock_res_and_lock(lock);
1333 ldlm_lock_allow_match_locked(lock);
1334 unlock_res_and_lock(lock);
1336 EXPORT_SYMBOL(ldlm_lock_allow_match);
1339 * Attempt to find a lock with specified properties.
1341 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1344 * Can be called in two ways:
1346 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1347 * for a duplicate of.
1349 * Otherwise, all of the fields must be filled in, to match against.
1351 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1352 * server (ie, connh is NULL)
1353 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1354 * list will be considered
1355 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1356 * to be canceled can still be matched as long as they still have reader
1357 * or writer refernces
1358 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1359 * just tell us if we would have matched.
1361 * \retval 1 if it finds an already-existing lock that is compatible; in this
1362 * case, lockh is filled in with a addref()ed lock
1364 * We also check security context, and if that fails we simply return 0 (to
1365 * keep caller code unchanged), the context failure will be discovered by
1366 * caller sometime later.
1368 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1369 const struct ldlm_res_id *res_id,
1370 enum ldlm_type type,
1371 union ldlm_policy_data *policy,
1372 enum ldlm_mode mode,
1373 struct lustre_handle *lockh, int unref)
1375 struct lock_match_data data = {
1379 .lmd_policy = policy,
1383 struct ldlm_resource *res;
1384 struct ldlm_lock *lock;
1389 data.lmd_old = ldlm_handle2lock(lockh);
1390 LASSERT(data.lmd_old != NULL);
1392 ns = ldlm_lock_to_ns(data.lmd_old);
1393 res_id = &data.lmd_old->l_resource->lr_name;
1394 type = data.lmd_old->l_resource->lr_type;
1395 *data.lmd_mode = data.lmd_old->l_req_mode;
1398 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1400 LASSERT(data.lmd_old == NULL);
1404 LDLM_RESOURCE_ADDREF(res);
1407 if (res->lr_type == LDLM_EXTENT)
1408 lock = search_itree(res, &data);
1410 lock = search_queue(&res->lr_granted, &data);
1413 if (flags & LDLM_FL_BLOCK_GRANTED)
1415 lock = search_queue(&res->lr_converting, &data);
1418 lock = search_queue(&res->lr_waiting, &data);
1425 LDLM_RESOURCE_DELREF(res);
1426 ldlm_resource_putref(res);
1429 ldlm_lock2handle(lock, lockh);
1430 if ((flags & LDLM_FL_LVB_READY) &&
1431 (!ldlm_is_lvb_ready(lock))) {
1432 __u64 wait_flags = LDLM_FL_LVB_READY |
1433 LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED;
1434 struct l_wait_info lwi;
1435 if (lock->l_completion_ast) {
1436 int err = lock->l_completion_ast(lock,
1437 LDLM_FL_WAIT_NOREPROC,
1440 if (flags & LDLM_FL_TEST_LOCK)
1441 LDLM_LOCK_RELEASE(lock);
1443 ldlm_lock_decref_internal(lock,
1450 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1451 NULL, LWI_ON_SIGNAL_NOOP, NULL);
1453 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1454 l_wait_event(lock->l_waitq,
1455 lock->l_flags & wait_flags,
1457 if (!ldlm_is_lvb_ready(lock)) {
1458 if (flags & LDLM_FL_TEST_LOCK)
1459 LDLM_LOCK_RELEASE(lock);
1461 ldlm_lock_decref_internal(lock, mode);
1468 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1469 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1470 res_id->name[2] : policy->l_extent.start,
1471 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1472 res_id->name[3] : policy->l_extent.end);
1474 /* check user's security context */
1475 if (lock->l_conn_export &&
1476 sptlrpc_import_check_ctx(
1477 class_exp2cliimp(lock->l_conn_export))) {
1478 if (!(flags & LDLM_FL_TEST_LOCK))
1479 ldlm_lock_decref_internal(lock, mode);
1483 if (flags & LDLM_FL_TEST_LOCK)
1484 LDLM_LOCK_RELEASE(lock);
1486 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1487 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1488 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1489 type, mode, res_id->name[0], res_id->name[1],
1490 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1491 res_id->name[2] :policy->l_extent.start,
1492 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1493 res_id->name[3] : policy->l_extent.end);
1495 if (data.lmd_old != NULL)
1496 LDLM_LOCK_PUT(data.lmd_old);
1498 return rc ? mode : 0;
1500 EXPORT_SYMBOL(ldlm_lock_match);
1502 enum ldlm_mode ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1505 struct ldlm_lock *lock;
1506 enum ldlm_mode mode = 0;
1509 lock = ldlm_handle2lock(lockh);
1511 lock_res_and_lock(lock);
1512 if (LDLM_HAVE_MASK(lock, GONE))
1515 if (ldlm_is_cbpending(lock) &&
1516 lock->l_readers == 0 && lock->l_writers == 0)
1520 *bits = lock->l_policy_data.l_inodebits.bits;
1521 mode = lock->l_granted_mode;
1522 ldlm_lock_addref_internal_nolock(lock, mode);
1529 unlock_res_and_lock(lock);
1530 LDLM_LOCK_PUT(lock);
1534 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1536 /** The caller must guarantee that the buffer is large enough. */
1537 int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill,
1538 enum req_location loc, void *data, int size)
1543 LASSERT(data != NULL);
1546 switch (lock->l_lvb_type) {
1548 if (size == sizeof(struct ost_lvb)) {
1549 if (loc == RCL_CLIENT)
1550 lvb = req_capsule_client_swab_get(pill,
1552 lustre_swab_ost_lvb);
1554 lvb = req_capsule_server_swab_get(pill,
1556 lustre_swab_ost_lvb);
1557 if (unlikely(lvb == NULL)) {
1558 LDLM_ERROR(lock, "no LVB");
1562 memcpy(data, lvb, size);
1563 } else if (size == sizeof(struct ost_lvb_v1)) {
1564 struct ost_lvb *olvb = data;
1566 if (loc == RCL_CLIENT)
1567 lvb = req_capsule_client_swab_get(pill,
1569 lustre_swab_ost_lvb_v1);
1571 lvb = req_capsule_server_sized_swab_get(pill,
1573 lustre_swab_ost_lvb_v1);
1574 if (unlikely(lvb == NULL)) {
1575 LDLM_ERROR(lock, "no LVB");
1579 memcpy(data, lvb, size);
1580 olvb->lvb_mtime_ns = 0;
1581 olvb->lvb_atime_ns = 0;
1582 olvb->lvb_ctime_ns = 0;
1584 LDLM_ERROR(lock, "Replied unexpected ost LVB size %d",
1590 if (size == sizeof(struct lquota_lvb)) {
1591 if (loc == RCL_CLIENT)
1592 lvb = req_capsule_client_swab_get(pill,
1594 lustre_swab_lquota_lvb);
1596 lvb = req_capsule_server_swab_get(pill,
1598 lustre_swab_lquota_lvb);
1599 if (unlikely(lvb == NULL)) {
1600 LDLM_ERROR(lock, "no LVB");
1604 memcpy(data, lvb, size);
1606 LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d",
1615 if (loc == RCL_CLIENT)
1616 lvb = req_capsule_client_get(pill, &RMF_DLM_LVB);
1618 lvb = req_capsule_server_get(pill, &RMF_DLM_LVB);
1619 if (unlikely(lvb == NULL)) {
1620 LDLM_ERROR(lock, "no LVB");
1624 memcpy(data, lvb, size);
1627 LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type);
1628 libcfs_debug_dumpstack(NULL);
1636 * Create and fill in new LDLM lock with specified properties.
1637 * Returns a referenced lock
1639 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1640 const struct ldlm_res_id *res_id,
1641 enum ldlm_type type,
1642 enum ldlm_mode mode,
1643 const struct ldlm_callback_suite *cbs,
1644 void *data, __u32 lvb_len,
1645 enum lvb_type lvb_type)
1647 struct ldlm_lock *lock;
1648 struct ldlm_resource *res;
1652 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1654 RETURN(ERR_CAST(res));
1656 lock = ldlm_lock_new(res);
1658 RETURN(ERR_PTR(-ENOMEM));
1660 lock->l_req_mode = mode;
1661 lock->l_ast_data = data;
1662 lock->l_pid = current_pid();
1663 if (ns_is_server(ns))
1664 ldlm_set_ns_srv(lock);
1666 lock->l_blocking_ast = cbs->lcs_blocking;
1667 lock->l_completion_ast = cbs->lcs_completion;
1668 lock->l_glimpse_ast = cbs->lcs_glimpse;
1671 lock->l_tree_node = NULL;
1672 /* if this is the extent lock, allocate the interval tree node */
1673 if (type == LDLM_EXTENT)
1674 if (ldlm_interval_alloc(lock) == NULL)
1675 GOTO(out, rc = -ENOMEM);
1678 lock->l_lvb_len = lvb_len;
1679 OBD_ALLOC_LARGE(lock->l_lvb_data, lvb_len);
1680 if (lock->l_lvb_data == NULL)
1681 GOTO(out, rc = -ENOMEM);
1684 lock->l_lvb_type = lvb_type;
1685 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1686 GOTO(out, rc = -ENOENT);
1691 ldlm_lock_destroy(lock);
1692 LDLM_LOCK_RELEASE(lock);
1693 RETURN(ERR_PTR(rc));
1697 * Enqueue (request) a lock.
1699 * Does not block. As a result of enqueue the lock would be put
1700 * into granted or waiting list.
1702 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1703 * set, skip all the enqueueing and delegate lock processing to intent policy
1706 enum ldlm_error ldlm_lock_enqueue(struct ldlm_namespace *ns,
1707 struct ldlm_lock **lockp,
1708 void *cookie, __u64 *flags)
1710 struct ldlm_lock *lock = *lockp;
1711 struct ldlm_resource *res = lock->l_resource;
1712 int local = ns_is_client(ldlm_res_to_ns(res));
1713 #ifdef HAVE_SERVER_SUPPORT
1714 ldlm_processing_policy policy;
1716 enum ldlm_error rc = ELDLM_OK;
1717 struct ldlm_interval *node = NULL;
1720 /* policies are not executed on the client or during replay */
1721 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1722 && !local && ns->ns_policy) {
1723 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1725 if (rc == ELDLM_LOCK_REPLACED) {
1726 /* The lock that was returned has already been granted,
1727 * and placed into lockp. If it's not the same as the
1728 * one we passed in, then destroy the old one and our
1729 * work here is done. */
1730 if (lock != *lockp) {
1731 ldlm_lock_destroy(lock);
1732 LDLM_LOCK_RELEASE(lock);
1734 *flags |= LDLM_FL_LOCK_CHANGED;
1736 } else if (rc != ELDLM_OK ||
1737 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1738 ldlm_lock_destroy(lock);
1743 if (*flags & LDLM_FL_RESENT) {
1744 /* Reconstruct LDLM_FL_SRV_ENQ_MASK @flags for reply.
1745 * Set LOCK_CHANGED always.
1746 * Check if the lock is granted for BLOCK_GRANTED.
1747 * Take NO_TIMEOUT from the lock as it is inherited through
1748 * LDLM_FL_INHERIT_MASK */
1749 *flags |= LDLM_FL_LOCK_CHANGED;
1750 if (lock->l_req_mode != lock->l_granted_mode)
1751 *flags |= LDLM_FL_BLOCK_GRANTED;
1752 *flags |= lock->l_flags & LDLM_FL_NO_TIMEOUT;
1756 /* For a replaying lock, it might be already in granted list. So
1757 * unlinking the lock will cause the interval node to be freed, we
1758 * have to allocate the interval node early otherwise we can't regrant
1759 * this lock in the future. - jay */
1760 if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1761 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
1763 lock_res_and_lock(lock);
1764 if (local && lock->l_req_mode == lock->l_granted_mode) {
1765 /* The server returned a blocked lock, but it was granted
1766 * before we got a chance to actually enqueue it. We don't
1767 * need to do anything else. */
1768 *flags &= ~LDLM_FL_BLOCKED_MASK;
1769 GOTO(out, rc = ELDLM_OK);
1772 ldlm_resource_unlink_lock(lock);
1773 if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1775 ldlm_lock_destroy_nolock(lock);
1776 GOTO(out, rc = -ENOMEM);
1779 INIT_LIST_HEAD(&node->li_group);
1780 ldlm_interval_attach(node, lock);
1784 /* Some flags from the enqueue want to make it into the AST, via the
1785 * lock's l_flags. */
1786 if (*flags & LDLM_FL_AST_DISCARD_DATA)
1787 ldlm_set_ast_discard_data(lock);
1788 if (*flags & LDLM_FL_TEST_LOCK)
1789 ldlm_set_test_lock(lock);
1791 /* This distinction between local lock trees is very important; a client
1792 * namespace only has information about locks taken by that client, and
1793 * thus doesn't have enough information to decide for itself if it can
1794 * be granted (below). In this case, we do exactly what the server
1795 * tells us to do, as dictated by the 'flags'.
1797 * We do exactly the same thing during recovery, when the server is
1798 * more or less trusting the clients not to lie.
1800 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1801 * granted/converting queues. */
1803 if (*flags & LDLM_FL_BLOCK_CONV)
1804 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1805 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1806 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1808 ldlm_grant_lock(lock, NULL);
1809 GOTO(out, rc = ELDLM_OK);
1810 #ifdef HAVE_SERVER_SUPPORT
1811 } else if (*flags & LDLM_FL_REPLAY) {
1812 if (*flags & LDLM_FL_BLOCK_CONV) {
1813 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1814 GOTO(out, rc = ELDLM_OK);
1815 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1816 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1817 GOTO(out, rc = ELDLM_OK);
1818 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1819 ldlm_grant_lock(lock, NULL);
1820 GOTO(out, rc = ELDLM_OK);
1822 /* If no flags, fall through to normal enqueue path. */
1825 policy = ldlm_processing_policy_table[res->lr_type];
1826 policy(lock, flags, 1, &rc, NULL);
1830 CERROR("This is client-side-only module, cannot handle "
1831 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1837 unlock_res_and_lock(lock);
1839 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1843 #ifdef HAVE_SERVER_SUPPORT
1845 * Iterate through all waiting locks on a given resource queue and attempt to
1848 * Must be called with resource lock held.
1850 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1851 struct list_head *work_list)
1853 struct list_head *tmp, *pos;
1854 ldlm_processing_policy policy;
1856 int rc = LDLM_ITER_CONTINUE;
1857 enum ldlm_error err;
1860 check_res_locked(res);
1862 policy = ldlm_processing_policy_table[res->lr_type];
1865 list_for_each_safe(tmp, pos, queue) {
1866 struct ldlm_lock *pending;
1868 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1870 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1873 rc = policy(pending, &flags, 0, &err, work_list);
1874 if (rc != LDLM_ITER_CONTINUE)
1883 * Process a call to blocking AST callback for a lock in ast_work list
1886 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1888 struct ldlm_cb_set_arg *arg = opaq;
1889 struct ldlm_lock_desc d;
1891 struct ldlm_lock *lock;
1894 if (list_empty(arg->list))
1897 lock = list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1899 /* nobody should touch l_bl_ast */
1900 lock_res_and_lock(lock);
1901 list_del_init(&lock->l_bl_ast);
1903 LASSERT(ldlm_is_ast_sent(lock));
1904 LASSERT(lock->l_bl_ast_run == 0);
1905 LASSERT(lock->l_blocking_lock);
1906 lock->l_bl_ast_run++;
1907 unlock_res_and_lock(lock);
1909 ldlm_lock2desc(lock->l_blocking_lock, &d);
1911 rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1912 LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1913 lock->l_blocking_lock = NULL;
1914 LDLM_LOCK_RELEASE(lock);
1920 * Process a call to completion AST callback for a lock in ast_work list
1923 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1925 struct ldlm_cb_set_arg *arg = opaq;
1927 struct ldlm_lock *lock;
1928 ldlm_completion_callback completion_callback;
1931 if (list_empty(arg->list))
1934 lock = list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1936 /* It's possible to receive a completion AST before we've set
1937 * the l_completion_ast pointer: either because the AST arrived
1938 * before the reply, or simply because there's a small race
1939 * window between receiving the reply and finishing the local
1940 * enqueue. (bug 842)
1942 * This can't happen with the blocking_ast, however, because we
1943 * will never call the local blocking_ast until we drop our
1944 * reader/writer reference, which we won't do until we get the
1945 * reply and finish enqueueing. */
1947 /* nobody should touch l_cp_ast */
1948 lock_res_and_lock(lock);
1949 list_del_init(&lock->l_cp_ast);
1950 LASSERT(ldlm_is_cp_reqd(lock));
1951 /* save l_completion_ast since it can be changed by
1952 * mds_intent_policy(), see bug 14225 */
1953 completion_callback = lock->l_completion_ast;
1954 ldlm_clear_cp_reqd(lock);
1955 unlock_res_and_lock(lock);
1957 if (completion_callback != NULL)
1958 rc = completion_callback(lock, 0, (void *)arg);
1959 LDLM_LOCK_RELEASE(lock);
1965 * Process a call to revocation AST callback for a lock in ast_work list
1968 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1970 struct ldlm_cb_set_arg *arg = opaq;
1971 struct ldlm_lock_desc desc;
1973 struct ldlm_lock *lock;
1976 if (list_empty(arg->list))
1979 lock = list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1980 list_del_init(&lock->l_rk_ast);
1982 /* the desc just pretend to exclusive */
1983 ldlm_lock2desc(lock, &desc);
1984 desc.l_req_mode = LCK_EX;
1985 desc.l_granted_mode = 0;
1987 rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1988 LDLM_LOCK_RELEASE(lock);
1994 * Process a call to glimpse AST callback for a lock in ast_work list
1996 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1998 struct ldlm_cb_set_arg *arg = opaq;
1999 struct ldlm_glimpse_work *gl_work;
2000 struct ldlm_lock *lock;
2004 if (list_empty(arg->list))
2007 gl_work = list_entry(arg->list->next, struct ldlm_glimpse_work,
2009 list_del_init(&gl_work->gl_list);
2011 lock = gl_work->gl_lock;
2013 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
2014 arg->gl_desc = gl_work->gl_desc;
2016 /* invoke the actual glimpse callback */
2017 if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
2020 LDLM_LOCK_RELEASE(lock);
2022 if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
2023 OBD_FREE_PTR(gl_work);
2029 * Process list of locks in need of ASTs being sent.
2031 * Used on server to send multiple ASTs together instead of sending one by
2034 int ldlm_run_ast_work(struct ldlm_namespace *ns, struct list_head *rpc_list,
2035 ldlm_desc_ast_t ast_type)
2037 struct ldlm_cb_set_arg *arg;
2038 set_producer_func work_ast_lock;
2041 if (list_empty(rpc_list))
2048 atomic_set(&arg->restart, 0);
2049 arg->list = rpc_list;
2052 case LDLM_WORK_BL_AST:
2053 arg->type = LDLM_BL_CALLBACK;
2054 work_ast_lock = ldlm_work_bl_ast_lock;
2056 case LDLM_WORK_CP_AST:
2057 arg->type = LDLM_CP_CALLBACK;
2058 work_ast_lock = ldlm_work_cp_ast_lock;
2060 case LDLM_WORK_REVOKE_AST:
2061 arg->type = LDLM_BL_CALLBACK;
2062 work_ast_lock = ldlm_work_revoke_ast_lock;
2064 case LDLM_WORK_GL_AST:
2065 arg->type = LDLM_GL_CALLBACK;
2066 work_ast_lock = ldlm_work_gl_ast_lock;
2072 /* We create a ptlrpc request set with flow control extension.
2073 * This request set will use the work_ast_lock function to produce new
2074 * requests and will send a new request each time one completes in order
2075 * to keep the number of requests in flight to ns_max_parallel_ast */
2076 arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
2077 work_ast_lock, arg);
2078 if (arg->set == NULL)
2079 GOTO(out, rc = -ENOMEM);
2081 ptlrpc_set_wait(arg->set);
2082 ptlrpc_set_destroy(arg->set);
2084 rc = atomic_read(&arg->restart) ? -ERESTART : 0;
2091 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
2093 ldlm_reprocess_all(res);
2094 return LDLM_ITER_CONTINUE;
2097 static int ldlm_reprocess_res(struct cfs_hash *hs, struct cfs_hash_bd *bd,
2098 struct hlist_node *hnode, void *arg)
2100 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2103 rc = reprocess_one_queue(res, arg);
2105 return rc == LDLM_ITER_STOP;
2109 * Iterate through all resources on a namespace attempting to grant waiting
2112 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
2117 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2118 ldlm_reprocess_res, NULL, 0);
2124 * Try to grant all waiting locks on a resource.
2126 * Calls ldlm_reprocess_queue on converting and waiting queues.
2128 * Typically called after some resource locks are cancelled to see
2129 * if anything could be granted as a result of the cancellation.
2131 void ldlm_reprocess_all(struct ldlm_resource *res)
2133 struct list_head rpc_list;
2134 #ifdef HAVE_SERVER_SUPPORT
2138 INIT_LIST_HEAD(&rpc_list);
2139 /* Local lock trees don't get reprocessed. */
2140 if (ns_is_client(ldlm_res_to_ns(res))) {
2147 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
2148 if (rc == LDLM_ITER_CONTINUE)
2149 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
2152 rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
2154 if (rc == -ERESTART) {
2155 LASSERT(list_empty(&rpc_list));
2161 INIT_LIST_HEAD(&rpc_list);
2162 if (!ns_is_client(ldlm_res_to_ns(res))) {
2163 CERROR("This is client-side-only module, cannot handle "
2164 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2170 EXPORT_SYMBOL(ldlm_reprocess_all);
2173 * Helper function to call blocking AST for LDLM lock \a lock in a
2174 * "cancelling" mode.
2176 void ldlm_cancel_callback(struct ldlm_lock *lock)
2178 check_res_locked(lock->l_resource);
2179 if (!ldlm_is_cancel(lock)) {
2180 ldlm_set_cancel(lock);
2181 if (lock->l_blocking_ast) {
2182 unlock_res_and_lock(lock);
2183 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
2185 lock_res_and_lock(lock);
2187 LDLM_DEBUG(lock, "no blocking ast");
2190 ldlm_set_bl_done(lock);
2194 * Remove skiplist-enabled LDLM lock \a req from granted list
2196 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
2198 if (req->l_resource->lr_type != LDLM_PLAIN &&
2199 req->l_resource->lr_type != LDLM_IBITS)
2202 list_del_init(&req->l_sl_policy);
2203 list_del_init(&req->l_sl_mode);
2207 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2209 void ldlm_lock_cancel(struct ldlm_lock *lock)
2211 struct ldlm_resource *res;
2212 struct ldlm_namespace *ns;
2215 lock_res_and_lock(lock);
2217 res = lock->l_resource;
2218 ns = ldlm_res_to_ns(res);
2220 /* Please do not, no matter how tempting, remove this LBUG without
2221 * talking to me first. -phik */
2222 if (lock->l_readers || lock->l_writers) {
2223 LDLM_ERROR(lock, "lock still has references");
2227 if (ldlm_is_waited(lock))
2228 ldlm_del_waiting_lock(lock);
2230 /* Releases cancel callback. */
2231 ldlm_cancel_callback(lock);
2233 LASSERT(!ldlm_is_waited(lock));
2235 ldlm_resource_unlink_lock(lock);
2236 ldlm_lock_destroy_nolock(lock);
2238 if (lock->l_granted_mode == lock->l_req_mode)
2239 ldlm_pool_del(&ns->ns_pool, lock);
2241 /* Make sure we will not be called again for same lock what is possible
2242 * if not to zero out lock->l_granted_mode */
2243 lock->l_granted_mode = LCK_MINMODE;
2244 unlock_res_and_lock(lock);
2248 EXPORT_SYMBOL(ldlm_lock_cancel);
2251 * Set opaque data into the lock that only makes sense to upper layer.
2253 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
2255 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2260 if (lock->l_ast_data == NULL)
2261 lock->l_ast_data = data;
2262 if (lock->l_ast_data == data)
2264 LDLM_LOCK_PUT(lock);
2268 EXPORT_SYMBOL(ldlm_lock_set_data);
2270 struct export_cl_data {
2271 struct obd_export *ecl_exp;
2275 static void ldlm_cancel_lock_for_export(struct obd_export *exp,
2276 struct ldlm_lock *lock,
2277 struct export_cl_data *ecl)
2279 struct ldlm_resource *res;
2281 res = ldlm_resource_getref(lock->l_resource);
2283 ldlm_res_lvbo_update(res, NULL, 1);
2284 ldlm_lock_cancel(lock);
2285 if (!exp->exp_obd->obd_stopping)
2286 ldlm_reprocess_all(res);
2287 ldlm_resource_putref(res);
2290 if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
2291 CDEBUG(D_INFO, "Export %p, %d locks cancelled.\n",
2292 exp, ecl->ecl_loop);
2297 * Iterator function for ldlm_export_cancel_locks.
2298 * Cancels passed locks.
2301 ldlm_cancel_locks_for_export_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd,
2302 struct hlist_node *hnode, void *data)
2305 struct export_cl_data *ecl = (struct export_cl_data *)data;
2306 struct obd_export *exp = ecl->ecl_exp;
2307 struct ldlm_lock *lock = cfs_hash_object(hs, hnode);
2309 LDLM_LOCK_GET(lock);
2310 ldlm_cancel_lock_for_export(exp, lock, ecl);
2311 LDLM_LOCK_RELEASE(lock);
2317 * Cancel all blocked locks for given export.
2319 * Typically called on client disconnection/eviction
2321 int ldlm_export_cancel_blocked_locks(struct obd_export *exp)
2323 struct export_cl_data ecl = {
2328 while (!list_empty(&exp->exp_bl_list)) {
2329 struct ldlm_lock *lock;
2331 spin_lock_bh(&exp->exp_bl_list_lock);
2332 if (!list_empty(&exp->exp_bl_list)) {
2333 lock = list_entry(exp->exp_bl_list.next,
2334 struct ldlm_lock, l_exp_list);
2335 LDLM_LOCK_GET(lock);
2336 list_del_init(&lock->l_exp_list);
2340 spin_unlock_bh(&exp->exp_bl_list_lock);
2345 ldlm_cancel_lock_for_export(exp, lock, &ecl);
2346 LDLM_LOCK_RELEASE(lock);
2349 CDEBUG(D_DLMTRACE, "Export %p, canceled %d locks, "
2350 "left on hash table %d.\n", exp, ecl.ecl_loop,
2351 atomic_read(&exp->exp_lock_hash->hs_count));
2353 return ecl.ecl_loop;
2357 * Cancel all locks for given export.
2359 * Typically called after client disconnection/eviction
2361 int ldlm_export_cancel_locks(struct obd_export *exp)
2363 struct export_cl_data ecl = {
2368 cfs_hash_for_each_empty(exp->exp_lock_hash,
2369 ldlm_cancel_locks_for_export_cb, &ecl);
2371 CDEBUG(D_DLMTRACE, "Export %p, canceled %d locks, "
2372 "left on hash table %d.\n", exp, ecl.ecl_loop,
2373 atomic_read(&exp->exp_lock_hash->hs_count));
2375 return ecl.ecl_loop;
2379 * Downgrade an exclusive lock.
2381 * A fast variant of ldlm_lock_convert for convertion of exclusive
2382 * locks. The convertion is always successful.
2383 * Used by Commit on Sharing (COS) code.
2385 * \param lock A lock to convert
2386 * \param new_mode new lock mode
2388 void ldlm_lock_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode)
2392 LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
2393 LASSERT(new_mode == LCK_COS);
2395 lock_res_and_lock(lock);
2396 ldlm_resource_unlink_lock(lock);
2398 * Remove the lock from pool as it will be added again in
2399 * ldlm_grant_lock() called below.
2401 ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
2403 lock->l_req_mode = new_mode;
2404 ldlm_grant_lock(lock, NULL);
2405 unlock_res_and_lock(lock);
2406 ldlm_reprocess_all(lock->l_resource);
2410 EXPORT_SYMBOL(ldlm_lock_downgrade);
2413 * Attempt to convert already granted lock to a different mode.
2415 * While lock conversion is not currently used, future client-side
2416 * optimizations could take advantage of it to avoid discarding cached
2419 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock,
2420 enum ldlm_mode new_mode, __u32 *flags)
2422 struct list_head rpc_list;
2423 struct ldlm_resource *res;
2424 struct ldlm_namespace *ns;
2426 #ifdef HAVE_SERVER_SUPPORT
2428 struct sl_insert_point prev;
2430 struct ldlm_interval *node;
2433 INIT_LIST_HEAD(&rpc_list);
2434 /* Just return if mode is unchanged. */
2435 if (new_mode == lock->l_granted_mode) {
2436 *flags |= LDLM_FL_BLOCK_GRANTED;
2437 RETURN(lock->l_resource);
2440 /* I can't check the type of lock here because the bitlock of lock
2441 * is not held here, so do the allocation blindly. -jay */
2442 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
2443 if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */
2446 LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
2447 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2449 lock_res_and_lock(lock);
2451 res = lock->l_resource;
2452 ns = ldlm_res_to_ns(res);
2454 #ifdef HAVE_SERVER_SUPPORT
2455 old_mode = lock->l_req_mode;
2457 lock->l_req_mode = new_mode;
2458 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2459 #ifdef HAVE_SERVER_SUPPORT
2460 /* remember the lock position where the lock might be
2461 * added back to the granted list later and also
2462 * remember the join mode for skiplist fixing. */
2463 prev.res_link = lock->l_res_link.prev;
2464 prev.mode_link = lock->l_sl_mode.prev;
2465 prev.policy_link = lock->l_sl_policy.prev;
2467 ldlm_resource_unlink_lock(lock);
2469 ldlm_resource_unlink_lock(lock);
2470 if (res->lr_type == LDLM_EXTENT) {
2471 /* FIXME: ugly code, I have to attach the lock to a
2472 * interval node again since perhaps it will be granted
2474 INIT_LIST_HEAD(&node->li_group);
2475 ldlm_interval_attach(node, lock);
2481 * Remove old lock from the pool before adding the lock with new
2482 * mode below in ->policy()
2484 ldlm_pool_del(&ns->ns_pool, lock);
2486 /* If this is a local resource, put it on the appropriate list. */
2487 if (ns_is_client(ldlm_res_to_ns(res))) {
2488 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2489 ldlm_resource_add_lock(res, &res->lr_converting, lock);
2491 /* This should never happen, because of the way the
2492 * server handles conversions. */
2493 LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2497 ldlm_grant_lock(lock, &rpc_list);
2499 /* FIXME: completion handling not with lr_lock held ! */
2500 if (lock->l_completion_ast)
2501 lock->l_completion_ast(lock, 0, NULL);
2503 #ifdef HAVE_SERVER_SUPPORT
2506 enum ldlm_error err;
2508 ldlm_processing_policy policy;
2510 policy = ldlm_processing_policy_table[res->lr_type];
2511 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2512 if (rc == LDLM_ITER_STOP) {
2513 lock->l_req_mode = old_mode;
2514 if (res->lr_type == LDLM_EXTENT)
2515 ldlm_extent_add_lock(res, lock);
2517 ldlm_granted_list_add_lock(lock, &prev);
2521 *flags |= LDLM_FL_BLOCK_GRANTED;
2527 CERROR("This is client-side-only module, cannot handle "
2528 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2532 unlock_res_and_lock(lock);
2535 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2537 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2542 * Print lock with lock handle \a lockh description into debug log.
2544 * Used when printing all locks on a resource for debug purposes.
2546 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
2548 struct ldlm_lock *lock;
2550 if (!((libcfs_debug | D_ERROR) & level))
2553 lock = ldlm_handle2lock(lockh);
2557 LDLM_DEBUG_LIMIT(level, lock, "###");
2559 LDLM_LOCK_PUT(lock);
2561 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2564 * Print lock information with custom message into debug log.
2567 void _ldlm_lock_debug(struct ldlm_lock *lock,
2568 struct libcfs_debug_msg_data *msgdata,
2569 const char *fmt, ...)
2572 struct obd_export *exp = lock->l_export;
2573 struct ldlm_resource *resource = lock->l_resource;
2574 char *nid = "local";
2576 va_start(args, fmt);
2578 if (exp && exp->exp_connection) {
2579 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2580 } else if (exp && exp->exp_obd != NULL) {
2581 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2582 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2585 if (resource == NULL) {
2586 libcfs_debug_vmsg2(msgdata, fmt, args,
2587 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2588 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2589 "remote: "LPX64" expref: %d pid: %u timeout: %lu "
2592 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2593 lock->l_readers, lock->l_writers,
2594 ldlm_lockname[lock->l_granted_mode],
2595 ldlm_lockname[lock->l_req_mode],
2596 lock->l_flags, nid, lock->l_remote_handle.cookie,
2597 exp ? atomic_read(&exp->exp_refcount) : -99,
2598 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2603 switch (resource->lr_type) {
2605 libcfs_debug_vmsg2(msgdata, fmt, args,
2606 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2607 "res: "DLDLMRES" rrc: %d type: %s ["LPU64"->"LPU64"] "
2608 "(req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote: "
2609 LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2610 ldlm_lock_to_ns_name(lock), lock,
2611 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2612 lock->l_readers, lock->l_writers,
2613 ldlm_lockname[lock->l_granted_mode],
2614 ldlm_lockname[lock->l_req_mode],
2616 atomic_read(&resource->lr_refcount),
2617 ldlm_typename[resource->lr_type],
2618 lock->l_policy_data.l_extent.start,
2619 lock->l_policy_data.l_extent.end,
2620 lock->l_req_extent.start, lock->l_req_extent.end,
2621 lock->l_flags, nid, lock->l_remote_handle.cookie,
2622 exp ? atomic_read(&exp->exp_refcount) : -99,
2623 lock->l_pid, lock->l_callback_timeout,
2628 libcfs_debug_vmsg2(msgdata, fmt, args,
2629 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2630 "res: "DLDLMRES" rrc: %d type: %s pid: %d "
2631 "["LPU64"->"LPU64"] flags: "LPX64" nid: %s "
2632 "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2633 ldlm_lock_to_ns_name(lock), lock,
2634 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2635 lock->l_readers, lock->l_writers,
2636 ldlm_lockname[lock->l_granted_mode],
2637 ldlm_lockname[lock->l_req_mode],
2639 atomic_read(&resource->lr_refcount),
2640 ldlm_typename[resource->lr_type],
2641 lock->l_policy_data.l_flock.pid,
2642 lock->l_policy_data.l_flock.start,
2643 lock->l_policy_data.l_flock.end,
2644 lock->l_flags, nid, lock->l_remote_handle.cookie,
2645 exp ? atomic_read(&exp->exp_refcount) : -99,
2646 lock->l_pid, lock->l_callback_timeout);
2650 libcfs_debug_vmsg2(msgdata, fmt, args,
2651 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2652 "res: "DLDLMRES" bits "LPX64" rrc: %d type: %s "
2653 "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2654 "pid: %u timeout: %lu lvb_type: %d\n",
2655 ldlm_lock_to_ns_name(lock),
2656 lock, lock->l_handle.h_cookie,
2657 atomic_read(&lock->l_refc),
2658 lock->l_readers, lock->l_writers,
2659 ldlm_lockname[lock->l_granted_mode],
2660 ldlm_lockname[lock->l_req_mode],
2662 lock->l_policy_data.l_inodebits.bits,
2663 atomic_read(&resource->lr_refcount),
2664 ldlm_typename[resource->lr_type],
2665 lock->l_flags, nid, lock->l_remote_handle.cookie,
2666 exp ? atomic_read(&exp->exp_refcount) : -99,
2667 lock->l_pid, lock->l_callback_timeout,
2672 libcfs_debug_vmsg2(msgdata, fmt, args,
2673 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2674 "res: "DLDLMRES" rrc: %d type: %s flags: "LPX64" "
2675 "nid: %s remote: "LPX64" expref: %d pid: %u "
2676 "timeout: %lu lvb_type: %d\n",
2677 ldlm_lock_to_ns_name(lock),
2678 lock, lock->l_handle.h_cookie,
2679 atomic_read(&lock->l_refc),
2680 lock->l_readers, lock->l_writers,
2681 ldlm_lockname[lock->l_granted_mode],
2682 ldlm_lockname[lock->l_req_mode],
2684 atomic_read(&resource->lr_refcount),
2685 ldlm_typename[resource->lr_type],
2686 lock->l_flags, nid, lock->l_remote_handle.cookie,
2687 exp ? atomic_read(&exp->exp_refcount) : -99,
2688 lock->l_pid, lock->l_callback_timeout,
2694 EXPORT_SYMBOL(_ldlm_lock_debug);