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_wire18_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_wire18_to_local,
73 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local,
76 static ldlm_policy_wire_to_local_t ldlm_policy_wire21_to_local[] = {
77 [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_wire_to_local,
78 [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_wire_to_local,
79 [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_wire21_to_local,
80 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local,
83 static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire[] = {
84 [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_local_to_wire,
85 [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_local_to_wire,
86 [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_local_to_wire,
87 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_local_to_wire,
91 * Converts lock policy from local format to on the wire lock_desc format
93 void ldlm_convert_policy_to_wire(ldlm_type_t type,
94 const ldlm_policy_data_t *lpolicy,
95 ldlm_wire_policy_data_t *wpolicy)
97 ldlm_policy_local_to_wire_t convert;
99 convert = ldlm_policy_local_to_wire[type - LDLM_MIN_TYPE];
101 convert(lpolicy, wpolicy);
105 * Converts lock policy from on the wire lock_desc format to local format
107 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
108 const ldlm_wire_policy_data_t *wpolicy,
109 ldlm_policy_data_t *lpolicy)
111 ldlm_policy_wire_to_local_t convert;
114 /** some badness for 2.0.0 clients, but 2.0.0 isn't supported */
115 new_client = (exp_connect_flags(exp) & OBD_CONNECT_FULL20) != 0;
117 convert = ldlm_policy_wire21_to_local[type - LDLM_MIN_TYPE];
119 convert = ldlm_policy_wire18_to_local[type - LDLM_MIN_TYPE];
121 convert(wpolicy, lpolicy);
124 char *ldlm_it2str(int it)
131 case (IT_OPEN | IT_CREAT):
146 CERROR("Unknown intent %d\n", it);
150 EXPORT_SYMBOL(ldlm_it2str);
152 extern struct kmem_cache *ldlm_lock_slab;
154 #ifdef HAVE_SERVER_SUPPORT
155 static ldlm_processing_policy ldlm_processing_policy_table[] = {
156 [LDLM_PLAIN] = ldlm_process_plain_lock,
157 [LDLM_EXTENT] = ldlm_process_extent_lock,
158 [LDLM_FLOCK] = ldlm_process_flock_lock,
159 [LDLM_IBITS] = ldlm_process_inodebits_lock,
162 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
164 return ldlm_processing_policy_table[res->lr_type];
166 EXPORT_SYMBOL(ldlm_get_processing_policy);
167 #endif /* HAVE_SERVER_SUPPORT */
169 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
173 EXPORT_SYMBOL(ldlm_register_intent);
176 * REFCOUNTED LOCK OBJECTS
181 * Get a reference on a lock.
183 * Lock refcounts, during creation:
184 * - one special one for allocation, dec'd only once in destroy
185 * - one for being a lock that's in-use
186 * - one for the addref associated with a new lock
188 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
190 atomic_inc(&lock->l_refc);
193 EXPORT_SYMBOL(ldlm_lock_get);
196 * Release lock reference.
198 * Also frees the lock if it was last reference.
200 void ldlm_lock_put(struct ldlm_lock *lock)
204 LASSERT(lock->l_resource != LP_POISON);
205 LASSERT(atomic_read(&lock->l_refc) > 0);
206 if (atomic_dec_and_test(&lock->l_refc)) {
207 struct ldlm_resource *res;
210 "final lock_put on destroyed lock, freeing it.");
212 res = lock->l_resource;
213 LASSERT(ldlm_is_destroyed(lock));
214 LASSERT(list_empty(&lock->l_res_link));
215 LASSERT(list_empty(&lock->l_pending_chain));
217 lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats,
219 lu_ref_del(&res->lr_reference, "lock", lock);
220 ldlm_resource_putref(res);
221 lock->l_resource = NULL;
222 if (lock->l_export) {
223 class_export_lock_put(lock->l_export, lock);
224 lock->l_export = NULL;
227 if (lock->l_lvb_data != NULL)
228 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
230 ldlm_interval_free(ldlm_interval_detach(lock));
231 lu_ref_fini(&lock->l_reference);
232 OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle);
237 EXPORT_SYMBOL(ldlm_lock_put);
240 * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked.
242 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
245 if (!list_empty(&lock->l_lru)) {
246 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
248 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
249 list_del_init(&lock->l_lru);
250 LASSERT(ns->ns_nr_unused > 0);
258 * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first.
260 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
262 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
266 if (ldlm_is_ns_srv(lock)) {
267 LASSERT(list_empty(&lock->l_lru));
271 spin_lock(&ns->ns_lock);
272 rc = ldlm_lock_remove_from_lru_nolock(lock);
273 spin_unlock(&ns->ns_lock);
279 * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked.
281 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
283 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
285 lock->l_last_used = cfs_time_current();
286 LASSERT(list_empty(&lock->l_lru));
287 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
288 list_add_tail(&lock->l_lru, &ns->ns_unused_list);
289 ldlm_clear_skipped(lock);
290 LASSERT(ns->ns_nr_unused >= 0);
295 * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks
298 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
300 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
303 spin_lock(&ns->ns_lock);
304 ldlm_lock_add_to_lru_nolock(lock);
305 spin_unlock(&ns->ns_lock);
310 * Moves LDLM lock \a lock that is already in namespace LRU to the tail of
311 * the LRU. Performs necessary LRU locking
313 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
315 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
318 if (ldlm_is_ns_srv(lock)) {
319 LASSERT(list_empty(&lock->l_lru));
324 spin_lock(&ns->ns_lock);
325 if (!list_empty(&lock->l_lru)) {
326 ldlm_lock_remove_from_lru_nolock(lock);
327 ldlm_lock_add_to_lru_nolock(lock);
329 spin_unlock(&ns->ns_lock);
334 * Helper to destroy a locked lock.
336 * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock
337 * Must be called with l_lock and lr_lock held.
339 * Does not actually free the lock data, but rather marks the lock as
340 * destroyed by setting l_destroyed field in the lock to 1. Destroys a
341 * handle->lock association too, so that the lock can no longer be found
342 * and removes the lock from LRU list. Actual lock freeing occurs when
343 * last lock reference goes away.
345 * Original comment (of some historical value):
346 * This used to have a 'strict' flag, which recovery would use to mark an
347 * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
348 * shall explain why it's gone: with the new hash table scheme, once you call
349 * ldlm_lock_destroy, you can never drop your final references on this lock.
350 * Because it's not in the hash table anymore. -phil
352 static int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
356 if (lock->l_readers || lock->l_writers) {
357 LDLM_ERROR(lock, "lock still has references");
361 if (!list_empty(&lock->l_res_link)) {
362 LDLM_ERROR(lock, "lock still on resource");
366 if (ldlm_is_destroyed(lock)) {
367 LASSERT(list_empty(&lock->l_lru));
371 ldlm_set_destroyed(lock);
373 if (lock->l_export && lock->l_export->exp_lock_hash) {
374 /* NB: it's safe to call cfs_hash_del() even lock isn't
375 * in exp_lock_hash. */
376 /* In the function below, .hs_keycmp resolves to
377 * ldlm_export_lock_keycmp() */
378 /* coverity[overrun-buffer-val] */
379 cfs_hash_del(lock->l_export->exp_lock_hash,
380 &lock->l_remote_handle, &lock->l_exp_hash);
383 ldlm_lock_remove_from_lru(lock);
384 class_handle_unhash(&lock->l_handle);
387 /* Wake anyone waiting for this lock */
388 /* FIXME: I should probably add yet another flag, instead of using
389 * l_export to only call this on clients */
391 class_export_put(lock->l_export);
392 lock->l_export = NULL;
393 if (lock->l_export && lock->l_completion_ast)
394 lock->l_completion_ast(lock, 0);
401 * Destroys a LDLM lock \a lock. Performs necessary locking first.
403 void ldlm_lock_destroy(struct ldlm_lock *lock)
407 lock_res_and_lock(lock);
408 first = ldlm_lock_destroy_internal(lock);
409 unlock_res_and_lock(lock);
411 /* drop reference from hashtable only for first destroy */
413 lu_ref_del(&lock->l_reference, "hash", lock);
414 LDLM_LOCK_RELEASE(lock);
420 * Destroys a LDLM lock \a lock that is already locked.
422 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
426 first = ldlm_lock_destroy_internal(lock);
427 /* drop reference from hashtable only for first destroy */
429 lu_ref_del(&lock->l_reference, "hash", lock);
430 LDLM_LOCK_RELEASE(lock);
435 /* this is called by portals_handle2object with the handle lock taken */
436 static void lock_handle_addref(void *lock)
438 LDLM_LOCK_GET((struct ldlm_lock *)lock);
441 static void lock_handle_free(void *lock, int size)
443 LASSERT(size == sizeof(struct ldlm_lock));
444 OBD_SLAB_FREE(lock, ldlm_lock_slab, size);
447 static struct portals_handle_ops lock_handle_ops = {
448 .hop_addref = lock_handle_addref,
449 .hop_free = lock_handle_free,
454 * Allocate and initialize new lock structure.
456 * usage: pass in a resource on which you have done ldlm_resource_get
457 * new lock will take over the refcount.
458 * returns: lock with refcount 2 - one for current caller and one for remote
460 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
462 struct ldlm_lock *lock;
465 if (resource == NULL)
468 OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, GFP_NOFS);
472 spin_lock_init(&lock->l_lock);
473 lock->l_resource = resource;
474 lu_ref_add(&resource->lr_reference, "lock", lock);
476 atomic_set(&lock->l_refc, 2);
477 INIT_LIST_HEAD(&lock->l_res_link);
478 INIT_LIST_HEAD(&lock->l_lru);
479 INIT_LIST_HEAD(&lock->l_pending_chain);
480 INIT_LIST_HEAD(&lock->l_bl_ast);
481 INIT_LIST_HEAD(&lock->l_cp_ast);
482 INIT_LIST_HEAD(&lock->l_rk_ast);
483 init_waitqueue_head(&lock->l_waitq);
484 lock->l_blocking_lock = NULL;
485 INIT_LIST_HEAD(&lock->l_sl_mode);
486 INIT_LIST_HEAD(&lock->l_sl_policy);
487 INIT_HLIST_NODE(&lock->l_exp_hash);
488 INIT_HLIST_NODE(&lock->l_exp_flock_hash);
490 lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats,
492 INIT_LIST_HEAD(&lock->l_handle.h_link);
493 class_handle_hash(&lock->l_handle, &lock_handle_ops);
495 lu_ref_init(&lock->l_reference);
496 lu_ref_add(&lock->l_reference, "hash", lock);
497 lock->l_callback_timeout = 0;
499 #if LUSTRE_TRACKS_LOCK_EXP_REFS
500 INIT_LIST_HEAD(&lock->l_exp_refs_link);
501 lock->l_exp_refs_nr = 0;
502 lock->l_exp_refs_target = NULL;
504 INIT_LIST_HEAD(&lock->l_exp_list);
510 * Moves LDLM lock \a lock to another resource.
511 * This is used on client when server returns some other lock than requested
512 * (typically as a result of intent operation)
514 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
515 const struct ldlm_res_id *new_resid)
517 struct ldlm_resource *oldres = lock->l_resource;
518 struct ldlm_resource *newres;
522 LASSERT(ns_is_client(ns));
524 lock_res_and_lock(lock);
525 if (memcmp(new_resid, &lock->l_resource->lr_name,
526 sizeof(lock->l_resource->lr_name)) == 0) {
528 unlock_res_and_lock(lock);
532 LASSERT(new_resid->name[0] != 0);
534 /* This function assumes that the lock isn't on any lists */
535 LASSERT(list_empty(&lock->l_res_link));
537 type = oldres->lr_type;
538 unlock_res_and_lock(lock);
540 newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
542 RETURN(PTR_ERR(newres));
544 lu_ref_add(&newres->lr_reference, "lock", lock);
546 * To flip the lock from the old to the new resource, lock, oldres and
547 * newres have to be locked. Resource spin-locks are nested within
548 * lock->l_lock, and are taken in the memory address order to avoid
551 spin_lock(&lock->l_lock);
552 oldres = lock->l_resource;
553 if (oldres < newres) {
555 lock_res_nested(newres, LRT_NEW);
558 lock_res_nested(oldres, LRT_NEW);
560 LASSERT(memcmp(new_resid, &oldres->lr_name,
561 sizeof oldres->lr_name) != 0);
562 lock->l_resource = newres;
564 unlock_res_and_lock(lock);
566 /* ...and the flowers are still standing! */
567 lu_ref_del(&oldres->lr_reference, "lock", lock);
568 ldlm_resource_putref(oldres);
573 /** \defgroup ldlm_handles LDLM HANDLES
574 * Ways to get hold of locks without any addresses.
579 * Fills in handle for LDLM lock \a lock into supplied \a lockh
580 * Does not take any references.
582 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
584 lockh->cookie = lock->l_handle.h_cookie;
586 EXPORT_SYMBOL(ldlm_lock2handle);
589 * Obtain a lock reference by handle.
591 * if \a flags: atomically get the lock and set the flags.
592 * Return NULL if flag already set
594 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
597 struct ldlm_lock *lock;
602 lock = class_handle2object(handle->cookie, NULL);
606 /* It's unlikely but possible that someone marked the lock as
607 * destroyed after we did handle2object on it */
608 if ((flags == 0) && !ldlm_is_destroyed(lock)) {
609 lu_ref_add(&lock->l_reference, "handle", current);
613 lock_res_and_lock(lock);
615 LASSERT(lock->l_resource != NULL);
617 lu_ref_add_atomic(&lock->l_reference, "handle", current);
618 if (unlikely(ldlm_is_destroyed(lock))) {
619 unlock_res_and_lock(lock);
620 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
625 /* If we're setting flags, make sure none of them are already set. */
627 if ((lock->l_flags & flags) != 0) {
628 unlock_res_and_lock(lock);
633 lock->l_flags |= flags;
636 unlock_res_and_lock(lock);
639 EXPORT_SYMBOL(__ldlm_handle2lock);
640 /** @} ldlm_handles */
643 * Fill in "on the wire" representation for given LDLM lock into supplied
644 * lock descriptor \a desc structure.
646 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
648 ldlm_res2desc(lock->l_resource, &desc->l_resource);
649 desc->l_req_mode = lock->l_req_mode;
650 desc->l_granted_mode = lock->l_granted_mode;
651 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
652 &lock->l_policy_data,
653 &desc->l_policy_data);
657 * Add a lock to list of conflicting locks to send AST to.
659 * Only add if we have not sent a blocking AST to the lock yet.
661 static void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
662 struct list_head *work_list)
664 if (!ldlm_is_ast_sent(lock)) {
665 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
666 ldlm_set_ast_sent(lock);
667 /* If the enqueuing client said so, tell the AST recipient to
668 * discard dirty data, rather than writing back. */
669 if (ldlm_is_ast_discard_data(new))
670 ldlm_set_discard_data(lock);
671 LASSERT(list_empty(&lock->l_bl_ast));
672 list_add(&lock->l_bl_ast, work_list);
674 LASSERT(lock->l_blocking_lock == NULL);
675 lock->l_blocking_lock = LDLM_LOCK_GET(new);
680 * Add a lock to list of just granted locks to send completion AST to.
682 static void ldlm_add_cp_work_item(struct ldlm_lock *lock,
683 struct list_head *work_list)
685 if (!ldlm_is_cp_reqd(lock)) {
686 ldlm_set_cp_reqd(lock);
687 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
688 LASSERT(list_empty(&lock->l_cp_ast));
689 list_add(&lock->l_cp_ast, work_list);
695 * Aggregator function to add AST work items into a list. Determines
696 * what sort of an AST work needs to be done and calls the proper
698 * Must be called with lr_lock held.
700 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
701 struct list_head *work_list)
704 check_res_locked(lock->l_resource);
706 ldlm_add_bl_work_item(lock, new, work_list);
708 ldlm_add_cp_work_item(lock, work_list);
713 * Add specified reader/writer reference to LDLM lock with handle \a lockh.
714 * r/w reference type is determined by \a mode
715 * Calls ldlm_lock_addref_internal.
717 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
719 struct ldlm_lock *lock;
721 lock = ldlm_handle2lock(lockh);
722 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
723 ldlm_lock_addref_internal(lock, mode);
726 EXPORT_SYMBOL(ldlm_lock_addref);
730 * Add specified reader/writer reference to LDLM lock \a lock.
731 * r/w reference type is determined by \a mode
732 * Removes lock from LRU if it is there.
733 * Assumes the LDLM lock is already locked.
735 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
737 ldlm_lock_remove_from_lru(lock);
738 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
740 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
742 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
744 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
747 lu_ref_add_atomic(&lock->l_reference, "user", lock);
748 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
752 * Attempts to add reader/writer reference to a lock with handle \a lockh, and
753 * fails if lock is already LDLM_FL_CBPENDING or destroyed.
755 * \retval 0 success, lock was addref-ed
757 * \retval -EAGAIN lock is being canceled.
759 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
761 struct ldlm_lock *lock;
765 lock = ldlm_handle2lock(lockh);
767 lock_res_and_lock(lock);
768 if (lock->l_readers != 0 || lock->l_writers != 0 ||
769 !ldlm_is_cbpending(lock)) {
770 ldlm_lock_addref_internal_nolock(lock, mode);
773 unlock_res_and_lock(lock);
778 EXPORT_SYMBOL(ldlm_lock_addref_try);
781 * Add specified reader/writer reference to LDLM lock \a lock.
782 * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work.
783 * Only called for local locks.
785 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
787 lock_res_and_lock(lock);
788 ldlm_lock_addref_internal_nolock(lock, mode);
789 unlock_res_and_lock(lock);
793 * Removes reader/writer reference for LDLM lock \a lock.
794 * Assumes LDLM lock is already locked.
795 * only called in ldlm_flock_destroy and for local locks.
796 * Does NOT add lock to LRU if no r/w references left to accomodate flock locks
797 * that cannot be placed in LRU.
799 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
801 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
802 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
803 LASSERT(lock->l_readers > 0);
804 lu_ref_del(&lock->l_reference, "reader", lock);
807 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
808 LASSERT(lock->l_writers > 0);
809 lu_ref_del(&lock->l_reference, "writer", lock);
813 lu_ref_del(&lock->l_reference, "user", lock);
814 LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */
818 * Removes reader/writer reference for LDLM lock \a lock.
819 * Locks LDLM lock first.
820 * If the lock is determined to be client lock on a client and r/w refcount
821 * drops to zero and the lock is not blocked, the lock is added to LRU lock
823 * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called.
825 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
827 struct ldlm_namespace *ns;
830 lock_res_and_lock(lock);
832 ns = ldlm_lock_to_ns(lock);
834 ldlm_lock_decref_internal_nolock(lock, mode);
836 if (ldlm_is_local(lock) &&
837 !lock->l_readers && !lock->l_writers) {
838 /* If this is a local lock on a server namespace and this was
839 * the last reference, cancel the lock. */
840 CDEBUG(D_INFO, "forcing cancel of local lock\n");
841 ldlm_set_cbpending(lock);
844 if (!lock->l_readers && !lock->l_writers &&
845 (ldlm_is_cbpending(lock) || lock->l_req_mode == LCK_GROUP)) {
846 /* If we received a blocked AST and this was the last reference,
848 * Group locks are special:
849 * They must not go in LRU, but they are not called back
850 * like non-group locks, instead they are manually released.
851 * They have an l_writers reference which they keep until
852 * they are manually released, so we remove them when they have
853 * no more reader or writer references. - LU-6368 */
854 if (ldlm_is_ns_srv(lock) && lock->l_export)
855 CERROR("FL_CBPENDING set on non-local lock--just a "
858 LDLM_DEBUG(lock, "final decref done on cbpending lock");
860 LDLM_LOCK_GET(lock); /* dropped by bl thread */
861 ldlm_lock_remove_from_lru(lock);
862 unlock_res_and_lock(lock);
864 if (ldlm_is_fail_loc(lock))
865 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
867 if (ldlm_is_atomic_cb(lock) ||
868 ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
869 ldlm_handle_bl_callback(ns, NULL, lock);
870 } else if (ns_is_client(ns) &&
871 !lock->l_readers && !lock->l_writers &&
872 !ldlm_is_no_lru(lock) &&
873 !ldlm_is_bl_ast(lock)) {
875 LDLM_DEBUG(lock, "add lock into lru list");
877 /* If this is a client-side namespace and this was the last
878 * reference, put it on the LRU. */
879 ldlm_lock_add_to_lru(lock);
880 unlock_res_and_lock(lock);
882 if (ldlm_is_fail_loc(lock))
883 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
885 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
886 * are not supported by the server, otherwise, it is done on
888 if (!exp_connect_cancelset(lock->l_conn_export) &&
889 !ns_connect_lru_resize(ns))
890 ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0);
892 LDLM_DEBUG(lock, "do not add lock into lru list");
893 unlock_res_and_lock(lock);
900 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
902 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
904 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
905 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
906 ldlm_lock_decref_internal(lock, mode);
909 EXPORT_SYMBOL(ldlm_lock_decref);
912 * Decrease reader/writer refcount for LDLM lock with handle
913 * \a lockh and mark it for subsequent cancellation once r/w refcount
914 * drops to zero instead of putting into LRU.
917 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
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 if (lock->l_granted_mode < res->lr_most_restr)
1130 res->lr_most_restr = lock->l_granted_mode;
1132 ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1137 * Search for a lock with given properties in a queue.
1139 * \retval a referenced lock or NULL. See the flag descriptions below, in the
1140 * comment above ldlm_lock_match
1142 static struct ldlm_lock *search_queue(struct list_head *queue,
1144 ldlm_policy_data_t *policy,
1145 struct ldlm_lock *old_lock,
1146 __u64 flags, int unref)
1148 struct ldlm_lock *lock;
1149 struct list_head *tmp;
1151 list_for_each(tmp, queue) {
1154 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
1156 if (lock == old_lock)
1159 /* Check if this lock can be matched.
1160 * Used by LU-2919(exclusive open) for open lease lock */
1161 if (ldlm_is_excl(lock))
1164 /* llite sometimes wants to match locks that will be
1165 * canceled when their users drop, but we allow it to match
1166 * if it passes in CBPENDING and the lock still has users.
1167 * this is generally only going to be used by children
1168 * whose parents already hold a lock so forward progress
1169 * can still happen. */
1170 if (ldlm_is_cbpending(lock) &&
1171 !(flags & LDLM_FL_CBPENDING))
1173 if (!unref && ldlm_is_cbpending(lock) &&
1174 lock->l_readers == 0 && lock->l_writers == 0)
1177 if (!(lock->l_req_mode & *mode))
1179 match = lock->l_req_mode;
1181 if (lock->l_resource->lr_type == LDLM_EXTENT &&
1182 (lock->l_policy_data.l_extent.start >
1183 policy->l_extent.start ||
1184 lock->l_policy_data.l_extent.end < policy->l_extent.end))
1187 if (unlikely(match == LCK_GROUP) &&
1188 lock->l_resource->lr_type == LDLM_EXTENT &&
1189 policy->l_extent.gid != LDLM_GID_ANY &&
1190 lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
1193 /* We match if we have existing lock with same or wider set
1195 if (lock->l_resource->lr_type == LDLM_IBITS &&
1196 ((lock->l_policy_data.l_inodebits.bits &
1197 policy->l_inodebits.bits) !=
1198 policy->l_inodebits.bits))
1201 if (!unref && LDLM_HAVE_MASK(lock, GONE))
1204 if ((flags & LDLM_FL_LOCAL_ONLY) &&
1205 !ldlm_is_local(lock))
1208 if (flags & LDLM_FL_TEST_LOCK) {
1209 LDLM_LOCK_GET(lock);
1210 ldlm_lock_touch_in_lru(lock);
1212 ldlm_lock_addref_internal_nolock(lock, match);
1221 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1223 if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) {
1224 lock->l_flags |= LDLM_FL_FAIL_NOTIFIED;
1225 wake_up_all(&lock->l_waitq);
1228 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1230 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1232 lock_res_and_lock(lock);
1233 ldlm_lock_fail_match_locked(lock);
1234 unlock_res_and_lock(lock);
1238 * Mark lock as "matchable" by OST.
1240 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1242 * Assumes LDLM lock is already locked.
1244 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1246 ldlm_set_lvb_ready(lock);
1247 wake_up_all(&lock->l_waitq);
1249 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1252 * Mark lock as "matchable" by OST.
1253 * Locks the lock and then \see ldlm_lock_allow_match_locked
1255 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1257 lock_res_and_lock(lock);
1258 ldlm_lock_allow_match_locked(lock);
1259 unlock_res_and_lock(lock);
1261 EXPORT_SYMBOL(ldlm_lock_allow_match);
1264 * Attempt to find a lock with specified properties.
1266 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1269 * Can be called in two ways:
1271 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1272 * for a duplicate of.
1274 * Otherwise, all of the fields must be filled in, to match against.
1276 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1277 * server (ie, connh is NULL)
1278 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1279 * list will be considered
1280 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1281 * to be canceled can still be matched as long as they still have reader
1282 * or writer refernces
1283 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1284 * just tell us if we would have matched.
1286 * \retval 1 if it finds an already-existing lock that is compatible; in this
1287 * case, lockh is filled in with a addref()ed lock
1289 * We also check security context, and if that fails we simply return 0 (to
1290 * keep caller code unchanged), the context failure will be discovered by
1291 * caller sometime later.
1293 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1294 const struct ldlm_res_id *res_id, ldlm_type_t type,
1295 ldlm_policy_data_t *policy, ldlm_mode_t mode,
1296 struct lustre_handle *lockh, int unref)
1298 struct ldlm_resource *res;
1299 struct ldlm_lock *lock, *old_lock = NULL;
1304 old_lock = ldlm_handle2lock(lockh);
1307 ns = ldlm_lock_to_ns(old_lock);
1308 res_id = &old_lock->l_resource->lr_name;
1309 type = old_lock->l_resource->lr_type;
1310 mode = old_lock->l_req_mode;
1313 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1315 LASSERT(old_lock == NULL);
1319 LDLM_RESOURCE_ADDREF(res);
1322 lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
1326 if (flags & LDLM_FL_BLOCK_GRANTED)
1328 lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
1332 lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
1340 LDLM_RESOURCE_DELREF(res);
1341 ldlm_resource_putref(res);
1344 ldlm_lock2handle(lock, lockh);
1345 if ((flags & LDLM_FL_LVB_READY) &&
1346 (!ldlm_is_lvb_ready(lock))) {
1347 __u64 wait_flags = LDLM_FL_LVB_READY |
1348 LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED;
1349 struct l_wait_info lwi;
1350 if (lock->l_completion_ast) {
1351 int err = lock->l_completion_ast(lock,
1352 LDLM_FL_WAIT_NOREPROC,
1355 if (flags & LDLM_FL_TEST_LOCK)
1356 LDLM_LOCK_RELEASE(lock);
1358 ldlm_lock_decref_internal(lock,
1365 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1366 NULL, LWI_ON_SIGNAL_NOOP, NULL);
1368 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1369 l_wait_event(lock->l_waitq,
1370 lock->l_flags & wait_flags,
1372 if (!ldlm_is_lvb_ready(lock)) {
1373 if (flags & LDLM_FL_TEST_LOCK)
1374 LDLM_LOCK_RELEASE(lock);
1376 ldlm_lock_decref_internal(lock, mode);
1383 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1384 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1385 res_id->name[2] : policy->l_extent.start,
1386 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1387 res_id->name[3] : policy->l_extent.end);
1389 /* check user's security context */
1390 if (lock->l_conn_export &&
1391 sptlrpc_import_check_ctx(
1392 class_exp2cliimp(lock->l_conn_export))) {
1393 if (!(flags & LDLM_FL_TEST_LOCK))
1394 ldlm_lock_decref_internal(lock, mode);
1398 if (flags & LDLM_FL_TEST_LOCK)
1399 LDLM_LOCK_RELEASE(lock);
1401 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1402 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1403 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1404 type, mode, res_id->name[0], res_id->name[1],
1405 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1406 res_id->name[2] :policy->l_extent.start,
1407 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1408 res_id->name[3] : policy->l_extent.end);
1411 LDLM_LOCK_PUT(old_lock);
1413 return rc ? mode : 0;
1415 EXPORT_SYMBOL(ldlm_lock_match);
1417 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1420 struct ldlm_lock *lock;
1421 ldlm_mode_t mode = 0;
1424 lock = ldlm_handle2lock(lockh);
1426 lock_res_and_lock(lock);
1427 if (LDLM_HAVE_MASK(lock, GONE))
1430 if (ldlm_is_cbpending(lock) &&
1431 lock->l_readers == 0 && lock->l_writers == 0)
1435 *bits = lock->l_policy_data.l_inodebits.bits;
1436 mode = lock->l_granted_mode;
1437 ldlm_lock_addref_internal_nolock(lock, mode);
1444 unlock_res_and_lock(lock);
1445 LDLM_LOCK_PUT(lock);
1449 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1451 /** The caller must guarantee that the buffer is large enough. */
1452 int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill,
1453 enum req_location loc, void *data, int size)
1458 LASSERT(data != NULL);
1461 switch (lock->l_lvb_type) {
1463 if (size == sizeof(struct ost_lvb)) {
1464 if (loc == RCL_CLIENT)
1465 lvb = req_capsule_client_swab_get(pill,
1467 lustre_swab_ost_lvb);
1469 lvb = req_capsule_server_swab_get(pill,
1471 lustre_swab_ost_lvb);
1472 if (unlikely(lvb == NULL)) {
1473 LDLM_ERROR(lock, "no LVB");
1477 memcpy(data, lvb, size);
1478 } else if (size == sizeof(struct ost_lvb_v1)) {
1479 struct ost_lvb *olvb = data;
1481 if (loc == RCL_CLIENT)
1482 lvb = req_capsule_client_swab_get(pill,
1484 lustre_swab_ost_lvb_v1);
1486 lvb = req_capsule_server_sized_swab_get(pill,
1488 lustre_swab_ost_lvb_v1);
1489 if (unlikely(lvb == NULL)) {
1490 LDLM_ERROR(lock, "no LVB");
1494 memcpy(data, lvb, size);
1495 olvb->lvb_mtime_ns = 0;
1496 olvb->lvb_atime_ns = 0;
1497 olvb->lvb_ctime_ns = 0;
1499 LDLM_ERROR(lock, "Replied unexpected ost LVB size %d",
1505 if (size == sizeof(struct lquota_lvb)) {
1506 if (loc == RCL_CLIENT)
1507 lvb = req_capsule_client_swab_get(pill,
1509 lustre_swab_lquota_lvb);
1511 lvb = req_capsule_server_swab_get(pill,
1513 lustre_swab_lquota_lvb);
1514 if (unlikely(lvb == NULL)) {
1515 LDLM_ERROR(lock, "no LVB");
1519 memcpy(data, lvb, size);
1521 LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d",
1530 if (loc == RCL_CLIENT)
1531 lvb = req_capsule_client_get(pill, &RMF_DLM_LVB);
1533 lvb = req_capsule_server_get(pill, &RMF_DLM_LVB);
1534 if (unlikely(lvb == NULL)) {
1535 LDLM_ERROR(lock, "no LVB");
1539 memcpy(data, lvb, size);
1542 LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type);
1543 libcfs_debug_dumpstack(NULL);
1551 * Create and fill in new LDLM lock with specified properties.
1552 * Returns a referenced lock
1554 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1555 const struct ldlm_res_id *res_id,
1558 const struct ldlm_callback_suite *cbs,
1559 void *data, __u32 lvb_len,
1560 enum lvb_type lvb_type)
1562 struct ldlm_lock *lock;
1563 struct ldlm_resource *res;
1567 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1569 RETURN(ERR_CAST(res));
1571 lock = ldlm_lock_new(res);
1573 RETURN(ERR_PTR(-ENOMEM));
1575 lock->l_req_mode = mode;
1576 lock->l_ast_data = data;
1577 lock->l_pid = current_pid();
1578 if (ns_is_server(ns))
1579 ldlm_set_ns_srv(lock);
1581 lock->l_blocking_ast = cbs->lcs_blocking;
1582 lock->l_completion_ast = cbs->lcs_completion;
1583 lock->l_glimpse_ast = cbs->lcs_glimpse;
1586 lock->l_tree_node = NULL;
1587 /* if this is the extent lock, allocate the interval tree node */
1588 if (type == LDLM_EXTENT)
1589 if (ldlm_interval_alloc(lock) == NULL)
1590 GOTO(out, rc = -ENOMEM);
1593 lock->l_lvb_len = lvb_len;
1594 OBD_ALLOC_LARGE(lock->l_lvb_data, lvb_len);
1595 if (lock->l_lvb_data == NULL)
1596 GOTO(out, rc = -ENOMEM);
1599 lock->l_lvb_type = lvb_type;
1600 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1601 GOTO(out, rc = -ENOENT);
1606 ldlm_lock_destroy(lock);
1607 LDLM_LOCK_RELEASE(lock);
1608 RETURN(ERR_PTR(rc));
1612 * Enqueue (request) a lock.
1614 * Does not block. As a result of enqueue the lock would be put
1615 * into granted or waiting list.
1617 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1618 * set, skip all the enqueueing and delegate lock processing to intent policy
1621 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1622 struct ldlm_lock **lockp,
1623 void *cookie, __u64 *flags)
1625 struct ldlm_lock *lock = *lockp;
1626 struct ldlm_resource *res = lock->l_resource;
1627 int local = ns_is_client(ldlm_res_to_ns(res));
1628 #ifdef HAVE_SERVER_SUPPORT
1629 ldlm_processing_policy policy;
1631 ldlm_error_t rc = ELDLM_OK;
1632 struct ldlm_interval *node = NULL;
1635 /* policies are not executed on the client or during replay */
1636 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1637 && !local && ns->ns_policy) {
1638 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1640 if (rc == ELDLM_LOCK_REPLACED) {
1641 /* The lock that was returned has already been granted,
1642 * and placed into lockp. If it's not the same as the
1643 * one we passed in, then destroy the old one and our
1644 * work here is done. */
1645 if (lock != *lockp) {
1646 ldlm_lock_destroy(lock);
1647 LDLM_LOCK_RELEASE(lock);
1649 *flags |= LDLM_FL_LOCK_CHANGED;
1651 } else if (rc != ELDLM_OK ||
1652 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1653 ldlm_lock_destroy(lock);
1658 if (*flags & LDLM_FL_RESENT) {
1659 /* Reconstruct LDLM_FL_SRV_ENQ_MASK @flags for reply.
1660 * Set LOCK_CHANGED always.
1661 * Check if the lock is granted for BLOCK_GRANTED.
1662 * Take NO_TIMEOUT from the lock as it is inherited through
1663 * LDLM_FL_INHERIT_MASK */
1664 *flags |= LDLM_FL_LOCK_CHANGED;
1665 if (lock->l_req_mode != lock->l_granted_mode)
1666 *flags |= LDLM_FL_BLOCK_GRANTED;
1667 *flags |= lock->l_flags & LDLM_FL_NO_TIMEOUT;
1671 /* For a replaying lock, it might be already in granted list. So
1672 * unlinking the lock will cause the interval node to be freed, we
1673 * have to allocate the interval node early otherwise we can't regrant
1674 * this lock in the future. - jay */
1675 if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1676 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
1678 lock_res_and_lock(lock);
1679 if (local && lock->l_req_mode == lock->l_granted_mode) {
1680 /* The server returned a blocked lock, but it was granted
1681 * before we got a chance to actually enqueue it. We don't
1682 * need to do anything else. */
1683 *flags &= ~LDLM_FL_BLOCKED_MASK;
1684 GOTO(out, rc = ELDLM_OK);
1687 ldlm_resource_unlink_lock(lock);
1688 if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1690 ldlm_lock_destroy_nolock(lock);
1691 GOTO(out, rc = -ENOMEM);
1694 INIT_LIST_HEAD(&node->li_group);
1695 ldlm_interval_attach(node, lock);
1699 /* Some flags from the enqueue want to make it into the AST, via the
1700 * lock's l_flags. */
1701 if (*flags & LDLM_FL_AST_DISCARD_DATA)
1702 ldlm_set_ast_discard_data(lock);
1703 if (*flags & LDLM_FL_TEST_LOCK)
1704 ldlm_set_test_lock(lock);
1706 /* This distinction between local lock trees is very important; a client
1707 * namespace only has information about locks taken by that client, and
1708 * thus doesn't have enough information to decide for itself if it can
1709 * be granted (below). In this case, we do exactly what the server
1710 * tells us to do, as dictated by the 'flags'.
1712 * We do exactly the same thing during recovery, when the server is
1713 * more or less trusting the clients not to lie.
1715 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1716 * granted/converting queues. */
1718 if (*flags & LDLM_FL_BLOCK_CONV)
1719 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1720 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1721 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1723 ldlm_grant_lock(lock, NULL);
1724 GOTO(out, rc = ELDLM_OK);
1725 #ifdef HAVE_SERVER_SUPPORT
1726 } else if (*flags & LDLM_FL_REPLAY) {
1727 if (*flags & LDLM_FL_BLOCK_CONV) {
1728 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1729 GOTO(out, rc = ELDLM_OK);
1730 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1731 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1732 GOTO(out, rc = ELDLM_OK);
1733 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1734 ldlm_grant_lock(lock, NULL);
1735 GOTO(out, rc = ELDLM_OK);
1737 /* If no flags, fall through to normal enqueue path. */
1740 policy = ldlm_processing_policy_table[res->lr_type];
1741 policy(lock, flags, 1, &rc, NULL);
1745 CERROR("This is client-side-only module, cannot handle "
1746 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1752 unlock_res_and_lock(lock);
1754 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1758 #ifdef HAVE_SERVER_SUPPORT
1760 * Iterate through all waiting locks on a given resource queue and attempt to
1763 * Must be called with resource lock held.
1765 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1766 struct list_head *work_list)
1768 struct list_head *tmp, *pos;
1769 ldlm_processing_policy policy;
1771 int rc = LDLM_ITER_CONTINUE;
1775 check_res_locked(res);
1777 policy = ldlm_processing_policy_table[res->lr_type];
1780 list_for_each_safe(tmp, pos, queue) {
1781 struct ldlm_lock *pending;
1782 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1784 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1787 rc = policy(pending, &flags, 0, &err, work_list);
1788 if (rc != LDLM_ITER_CONTINUE)
1797 * Process a call to blocking AST callback for a lock in ast_work list
1800 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1802 struct ldlm_cb_set_arg *arg = opaq;
1803 struct ldlm_lock_desc d;
1805 struct ldlm_lock *lock;
1808 if (list_empty(arg->list))
1811 lock = list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1813 /* nobody should touch l_bl_ast */
1814 lock_res_and_lock(lock);
1815 list_del_init(&lock->l_bl_ast);
1817 LASSERT(ldlm_is_ast_sent(lock));
1818 LASSERT(lock->l_bl_ast_run == 0);
1819 LASSERT(lock->l_blocking_lock);
1820 lock->l_bl_ast_run++;
1821 unlock_res_and_lock(lock);
1823 ldlm_lock2desc(lock->l_blocking_lock, &d);
1825 rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1826 LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1827 lock->l_blocking_lock = NULL;
1828 LDLM_LOCK_RELEASE(lock);
1834 * Process a call to completion AST callback for a lock in ast_work list
1837 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1839 struct ldlm_cb_set_arg *arg = opaq;
1841 struct ldlm_lock *lock;
1842 ldlm_completion_callback completion_callback;
1845 if (list_empty(arg->list))
1848 lock = list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1850 /* It's possible to receive a completion AST before we've set
1851 * the l_completion_ast pointer: either because the AST arrived
1852 * before the reply, or simply because there's a small race
1853 * window between receiving the reply and finishing the local
1854 * enqueue. (bug 842)
1856 * This can't happen with the blocking_ast, however, because we
1857 * will never call the local blocking_ast until we drop our
1858 * reader/writer reference, which we won't do until we get the
1859 * reply and finish enqueueing. */
1861 /* nobody should touch l_cp_ast */
1862 lock_res_and_lock(lock);
1863 list_del_init(&lock->l_cp_ast);
1864 LASSERT(ldlm_is_cp_reqd(lock));
1865 /* save l_completion_ast since it can be changed by
1866 * mds_intent_policy(), see bug 14225 */
1867 completion_callback = lock->l_completion_ast;
1868 ldlm_clear_cp_reqd(lock);
1869 unlock_res_and_lock(lock);
1871 if (completion_callback != NULL)
1872 rc = completion_callback(lock, 0, (void *)arg);
1873 LDLM_LOCK_RELEASE(lock);
1879 * Process a call to revocation AST callback for a lock in ast_work list
1882 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1884 struct ldlm_cb_set_arg *arg = opaq;
1885 struct ldlm_lock_desc desc;
1887 struct ldlm_lock *lock;
1890 if (list_empty(arg->list))
1893 lock = list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1894 list_del_init(&lock->l_rk_ast);
1896 /* the desc just pretend to exclusive */
1897 ldlm_lock2desc(lock, &desc);
1898 desc.l_req_mode = LCK_EX;
1899 desc.l_granted_mode = 0;
1901 rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1902 LDLM_LOCK_RELEASE(lock);
1908 * Process a call to glimpse AST callback for a lock in ast_work list
1910 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1912 struct ldlm_cb_set_arg *arg = opaq;
1913 struct ldlm_glimpse_work *gl_work;
1914 struct ldlm_lock *lock;
1918 if (list_empty(arg->list))
1921 gl_work = list_entry(arg->list->next, struct ldlm_glimpse_work,
1923 list_del_init(&gl_work->gl_list);
1925 lock = gl_work->gl_lock;
1927 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
1928 arg->gl_desc = gl_work->gl_desc;
1930 /* invoke the actual glimpse callback */
1931 if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
1934 LDLM_LOCK_RELEASE(lock);
1936 if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
1937 OBD_FREE_PTR(gl_work);
1943 * Process list of locks in need of ASTs being sent.
1945 * Used on server to send multiple ASTs together instead of sending one by
1948 int ldlm_run_ast_work(struct ldlm_namespace *ns, struct list_head *rpc_list,
1949 ldlm_desc_ast_t ast_type)
1951 struct ldlm_cb_set_arg *arg;
1952 set_producer_func work_ast_lock;
1955 if (list_empty(rpc_list))
1962 atomic_set(&arg->restart, 0);
1963 arg->list = rpc_list;
1966 case LDLM_WORK_BL_AST:
1967 arg->type = LDLM_BL_CALLBACK;
1968 work_ast_lock = ldlm_work_bl_ast_lock;
1970 case LDLM_WORK_CP_AST:
1971 arg->type = LDLM_CP_CALLBACK;
1972 work_ast_lock = ldlm_work_cp_ast_lock;
1974 case LDLM_WORK_REVOKE_AST:
1975 arg->type = LDLM_BL_CALLBACK;
1976 work_ast_lock = ldlm_work_revoke_ast_lock;
1978 case LDLM_WORK_GL_AST:
1979 arg->type = LDLM_GL_CALLBACK;
1980 work_ast_lock = ldlm_work_gl_ast_lock;
1986 /* We create a ptlrpc request set with flow control extension.
1987 * This request set will use the work_ast_lock function to produce new
1988 * requests and will send a new request each time one completes in order
1989 * to keep the number of requests in flight to ns_max_parallel_ast */
1990 arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
1991 work_ast_lock, arg);
1992 if (arg->set == NULL)
1993 GOTO(out, rc = -ENOMEM);
1995 ptlrpc_set_wait(arg->set);
1996 ptlrpc_set_destroy(arg->set);
1998 rc = atomic_read(&arg->restart) ? -ERESTART : 0;
2005 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
2007 ldlm_reprocess_all(res);
2008 return LDLM_ITER_CONTINUE;
2011 static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2012 struct hlist_node *hnode, void *arg)
2014 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2017 rc = reprocess_one_queue(res, arg);
2019 return rc == LDLM_ITER_STOP;
2023 * Iterate through all resources on a namespace attempting to grant waiting
2026 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
2031 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2032 ldlm_reprocess_res, NULL);
2038 * Try to grant all waiting locks on a resource.
2040 * Calls ldlm_reprocess_queue on converting and waiting queues.
2042 * Typically called after some resource locks are cancelled to see
2043 * if anything could be granted as a result of the cancellation.
2045 void ldlm_reprocess_all(struct ldlm_resource *res)
2047 struct list_head rpc_list;
2048 #ifdef HAVE_SERVER_SUPPORT
2052 INIT_LIST_HEAD(&rpc_list);
2053 /* Local lock trees don't get reprocessed. */
2054 if (ns_is_client(ldlm_res_to_ns(res))) {
2061 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
2062 if (rc == LDLM_ITER_CONTINUE)
2063 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
2066 rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
2068 if (rc == -ERESTART) {
2069 LASSERT(list_empty(&rpc_list));
2075 INIT_LIST_HEAD(&rpc_list);
2076 if (!ns_is_client(ldlm_res_to_ns(res))) {
2077 CERROR("This is client-side-only module, cannot handle "
2078 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2084 EXPORT_SYMBOL(ldlm_reprocess_all);
2087 * Helper function to call blocking AST for LDLM lock \a lock in a
2088 * "cancelling" mode.
2090 void ldlm_cancel_callback(struct ldlm_lock *lock)
2092 check_res_locked(lock->l_resource);
2093 if (!ldlm_is_cancel(lock)) {
2094 ldlm_set_cancel(lock);
2095 if (lock->l_blocking_ast) {
2096 unlock_res_and_lock(lock);
2097 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
2099 lock_res_and_lock(lock);
2101 LDLM_DEBUG(lock, "no blocking ast");
2104 ldlm_set_bl_done(lock);
2108 * Remove skiplist-enabled LDLM lock \a req from granted list
2110 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
2112 if (req->l_resource->lr_type != LDLM_PLAIN &&
2113 req->l_resource->lr_type != LDLM_IBITS)
2116 list_del_init(&req->l_sl_policy);
2117 list_del_init(&req->l_sl_mode);
2121 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2123 void ldlm_lock_cancel(struct ldlm_lock *lock)
2125 struct ldlm_resource *res;
2126 struct ldlm_namespace *ns;
2129 lock_res_and_lock(lock);
2131 res = lock->l_resource;
2132 ns = ldlm_res_to_ns(res);
2134 /* Please do not, no matter how tempting, remove this LBUG without
2135 * talking to me first. -phik */
2136 if (lock->l_readers || lock->l_writers) {
2137 LDLM_ERROR(lock, "lock still has references");
2141 if (ldlm_is_waited(lock))
2142 ldlm_del_waiting_lock(lock);
2144 /* Releases cancel callback. */
2145 ldlm_cancel_callback(lock);
2147 LASSERT(!ldlm_is_waited(lock));
2149 ldlm_resource_unlink_lock(lock);
2150 ldlm_lock_destroy_nolock(lock);
2152 if (lock->l_granted_mode == lock->l_req_mode)
2153 ldlm_pool_del(&ns->ns_pool, lock);
2155 /* Make sure we will not be called again for same lock what is possible
2156 * if not to zero out lock->l_granted_mode */
2157 lock->l_granted_mode = LCK_MINMODE;
2158 unlock_res_and_lock(lock);
2162 EXPORT_SYMBOL(ldlm_lock_cancel);
2165 * Set opaque data into the lock that only makes sense to upper layer.
2167 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
2169 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2174 if (lock->l_ast_data == NULL)
2175 lock->l_ast_data = data;
2176 if (lock->l_ast_data == data)
2178 LDLM_LOCK_PUT(lock);
2182 EXPORT_SYMBOL(ldlm_lock_set_data);
2184 struct export_cl_data {
2185 struct obd_export *ecl_exp;
2190 * Iterator function for ldlm_cancel_locks_for_export.
2191 * Cancels passed locks.
2193 static int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2194 struct hlist_node *hnode, void *data)
2197 struct export_cl_data *ecl = (struct export_cl_data *)data;
2198 struct obd_export *exp = ecl->ecl_exp;
2199 struct ldlm_lock *lock = cfs_hash_object(hs, hnode);
2200 struct ldlm_resource *res;
2202 res = ldlm_resource_getref(lock->l_resource);
2203 LDLM_LOCK_GET(lock);
2205 LDLM_DEBUG(lock, "export %p", exp);
2206 ldlm_res_lvbo_update(res, NULL, 1);
2207 ldlm_lock_cancel(lock);
2208 ldlm_reprocess_all(res);
2209 ldlm_resource_putref(res);
2210 LDLM_LOCK_RELEASE(lock);
2213 if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
2215 "Cancel lock %p for export %p (loop %d), still have "
2216 "%d locks left on hash table.\n",
2217 lock, exp, ecl->ecl_loop,
2218 atomic_read(&hs->hs_count));
2225 * Cancel all locks for given export.
2227 * Typically called on client disconnection/eviction
2229 void ldlm_cancel_locks_for_export(struct obd_export *exp)
2231 struct export_cl_data ecl = {
2236 cfs_hash_for_each_empty(exp->exp_lock_hash,
2237 ldlm_cancel_locks_for_export_cb, &ecl);
2241 * Downgrade an exclusive lock.
2243 * A fast variant of ldlm_lock_convert for convertion of exclusive
2244 * locks. The convertion is always successful.
2245 * Used by Commit on Sharing (COS) code.
2247 * \param lock A lock to convert
2248 * \param new_mode new lock mode
2250 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
2254 LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
2255 LASSERT(new_mode == LCK_COS);
2257 lock_res_and_lock(lock);
2258 ldlm_resource_unlink_lock(lock);
2260 * Remove the lock from pool as it will be added again in
2261 * ldlm_grant_lock() called below.
2263 ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
2265 lock->l_req_mode = new_mode;
2266 ldlm_grant_lock(lock, NULL);
2267 unlock_res_and_lock(lock);
2268 ldlm_reprocess_all(lock->l_resource);
2272 EXPORT_SYMBOL(ldlm_lock_downgrade);
2275 * Attempt to convert already granted lock to a different mode.
2277 * While lock conversion is not currently used, future client-side
2278 * optimizations could take advantage of it to avoid discarding cached
2281 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
2284 struct list_head rpc_list;
2285 struct ldlm_resource *res;
2286 struct ldlm_namespace *ns;
2288 #ifdef HAVE_SERVER_SUPPORT
2290 struct sl_insert_point prev;
2292 struct ldlm_interval *node;
2295 INIT_LIST_HEAD(&rpc_list);
2296 /* Just return if mode is unchanged. */
2297 if (new_mode == lock->l_granted_mode) {
2298 *flags |= LDLM_FL_BLOCK_GRANTED;
2299 RETURN(lock->l_resource);
2302 /* I can't check the type of lock here because the bitlock of lock
2303 * is not held here, so do the allocation blindly. -jay */
2304 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
2305 if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */
2308 LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
2309 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2311 lock_res_and_lock(lock);
2313 res = lock->l_resource;
2314 ns = ldlm_res_to_ns(res);
2316 #ifdef HAVE_SERVER_SUPPORT
2317 old_mode = lock->l_req_mode;
2319 lock->l_req_mode = new_mode;
2320 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2321 #ifdef HAVE_SERVER_SUPPORT
2322 /* remember the lock position where the lock might be
2323 * added back to the granted list later and also
2324 * remember the join mode for skiplist fixing. */
2325 prev.res_link = lock->l_res_link.prev;
2326 prev.mode_link = lock->l_sl_mode.prev;
2327 prev.policy_link = lock->l_sl_policy.prev;
2329 ldlm_resource_unlink_lock(lock);
2331 ldlm_resource_unlink_lock(lock);
2332 if (res->lr_type == LDLM_EXTENT) {
2333 /* FIXME: ugly code, I have to attach the lock to a
2334 * interval node again since perhaps it will be granted
2336 INIT_LIST_HEAD(&node->li_group);
2337 ldlm_interval_attach(node, lock);
2343 * Remove old lock from the pool before adding the lock with new
2344 * mode below in ->policy()
2346 ldlm_pool_del(&ns->ns_pool, lock);
2348 /* If this is a local resource, put it on the appropriate list. */
2349 if (ns_is_client(ldlm_res_to_ns(res))) {
2350 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2351 ldlm_resource_add_lock(res, &res->lr_converting, lock);
2353 /* This should never happen, because of the way the
2354 * server handles conversions. */
2355 LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2359 ldlm_grant_lock(lock, &rpc_list);
2361 /* FIXME: completion handling not with lr_lock held ! */
2362 if (lock->l_completion_ast)
2363 lock->l_completion_ast(lock, 0, NULL);
2365 #ifdef HAVE_SERVER_SUPPORT
2370 ldlm_processing_policy policy;
2371 policy = ldlm_processing_policy_table[res->lr_type];
2372 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2373 if (rc == LDLM_ITER_STOP) {
2374 lock->l_req_mode = old_mode;
2375 if (res->lr_type == LDLM_EXTENT)
2376 ldlm_extent_add_lock(res, lock);
2378 ldlm_granted_list_add_lock(lock, &prev);
2382 *flags |= LDLM_FL_BLOCK_GRANTED;
2388 CERROR("This is client-side-only module, cannot handle "
2389 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2393 unlock_res_and_lock(lock);
2396 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2398 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2403 * Print lock with lock handle \a lockh description into debug log.
2405 * Used when printing all locks on a resource for debug purposes.
2407 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
2409 struct ldlm_lock *lock;
2411 if (!((libcfs_debug | D_ERROR) & level))
2414 lock = ldlm_handle2lock(lockh);
2418 LDLM_DEBUG_LIMIT(level, lock, "###");
2420 LDLM_LOCK_PUT(lock);
2422 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2425 * Print lock information with custom message into debug log.
2428 void _ldlm_lock_debug(struct ldlm_lock *lock,
2429 struct libcfs_debug_msg_data *msgdata,
2430 const char *fmt, ...)
2433 struct obd_export *exp = lock->l_export;
2434 struct ldlm_resource *resource = lock->l_resource;
2435 char *nid = "local";
2437 va_start(args, fmt);
2439 if (exp && exp->exp_connection) {
2440 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2441 } else if (exp && exp->exp_obd != NULL) {
2442 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2443 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2446 if (resource == NULL) {
2447 libcfs_debug_vmsg2(msgdata, fmt, args,
2448 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2449 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2450 "remote: "LPX64" expref: %d pid: %u timeout: %lu "
2453 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2454 lock->l_readers, lock->l_writers,
2455 ldlm_lockname[lock->l_granted_mode],
2456 ldlm_lockname[lock->l_req_mode],
2457 lock->l_flags, nid, lock->l_remote_handle.cookie,
2458 exp ? atomic_read(&exp->exp_refcount) : -99,
2459 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2464 switch (resource->lr_type) {
2466 libcfs_debug_vmsg2(msgdata, fmt, args,
2467 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2468 "res: "DLDLMRES" rrc: %d type: %s ["LPU64"->"LPU64"] "
2469 "(req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote: "
2470 LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2471 ldlm_lock_to_ns_name(lock), lock,
2472 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2473 lock->l_readers, lock->l_writers,
2474 ldlm_lockname[lock->l_granted_mode],
2475 ldlm_lockname[lock->l_req_mode],
2477 atomic_read(&resource->lr_refcount),
2478 ldlm_typename[resource->lr_type],
2479 lock->l_policy_data.l_extent.start,
2480 lock->l_policy_data.l_extent.end,
2481 lock->l_req_extent.start, lock->l_req_extent.end,
2482 lock->l_flags, nid, lock->l_remote_handle.cookie,
2483 exp ? atomic_read(&exp->exp_refcount) : -99,
2484 lock->l_pid, lock->l_callback_timeout,
2489 libcfs_debug_vmsg2(msgdata, fmt, args,
2490 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2491 "res: "DLDLMRES" rrc: %d type: %s pid: %d "
2492 "["LPU64"->"LPU64"] flags: "LPX64" nid: %s "
2493 "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2494 ldlm_lock_to_ns_name(lock), lock,
2495 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2496 lock->l_readers, lock->l_writers,
2497 ldlm_lockname[lock->l_granted_mode],
2498 ldlm_lockname[lock->l_req_mode],
2500 atomic_read(&resource->lr_refcount),
2501 ldlm_typename[resource->lr_type],
2502 lock->l_policy_data.l_flock.pid,
2503 lock->l_policy_data.l_flock.start,
2504 lock->l_policy_data.l_flock.end,
2505 lock->l_flags, nid, lock->l_remote_handle.cookie,
2506 exp ? atomic_read(&exp->exp_refcount) : -99,
2507 lock->l_pid, lock->l_callback_timeout);
2511 libcfs_debug_vmsg2(msgdata, fmt, args,
2512 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2513 "res: "DLDLMRES" bits "LPX64" rrc: %d type: %s "
2514 "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2515 "pid: %u timeout: %lu lvb_type: %d\n",
2516 ldlm_lock_to_ns_name(lock),
2517 lock, lock->l_handle.h_cookie,
2518 atomic_read(&lock->l_refc),
2519 lock->l_readers, lock->l_writers,
2520 ldlm_lockname[lock->l_granted_mode],
2521 ldlm_lockname[lock->l_req_mode],
2523 lock->l_policy_data.l_inodebits.bits,
2524 atomic_read(&resource->lr_refcount),
2525 ldlm_typename[resource->lr_type],
2526 lock->l_flags, nid, lock->l_remote_handle.cookie,
2527 exp ? atomic_read(&exp->exp_refcount) : -99,
2528 lock->l_pid, lock->l_callback_timeout,
2533 libcfs_debug_vmsg2(msgdata, fmt, args,
2534 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2535 "res: "DLDLMRES" rrc: %d type: %s flags: "LPX64" "
2536 "nid: %s remote: "LPX64" expref: %d pid: %u "
2537 "timeout: %lu lvb_type: %d\n",
2538 ldlm_lock_to_ns_name(lock),
2539 lock, lock->l_handle.h_cookie,
2540 atomic_read(&lock->l_refc),
2541 lock->l_readers, lock->l_writers,
2542 ldlm_lockname[lock->l_granted_mode],
2543 ldlm_lockname[lock->l_req_mode],
2545 atomic_read(&resource->lr_refcount),
2546 ldlm_typename[resource->lr_type],
2547 lock->l_flags, nid, lock->l_remote_handle.cookie,
2548 exp ? atomic_read(&exp->exp_refcount) : -99,
2549 lock->l_pid, lock->l_callback_timeout,
2555 EXPORT_SYMBOL(_ldlm_lock_debug);