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, 2015, 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>
46 #include <lustre_swab.h>
47 #include <obd_class.h>
49 #include "ldlm_internal.h"
52 char *ldlm_lockname[] = {
60 [LCK_GROUP] = "GROUP",
63 EXPORT_SYMBOL(ldlm_lockname);
65 char *ldlm_typename[] = {
67 [LDLM_EXTENT] = "EXT",
72 static ldlm_policy_wire_to_local_t ldlm_policy_wire_to_local[] = {
73 [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_wire_to_local,
74 [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_wire_to_local,
75 [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_wire_to_local,
76 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local,
79 static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire[] = {
80 [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_local_to_wire,
81 [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_local_to_wire,
82 [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_local_to_wire,
83 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_local_to_wire,
87 * Converts lock policy from local format to on the wire lock_desc format
89 void ldlm_convert_policy_to_wire(enum ldlm_type type,
90 const union ldlm_policy_data *lpolicy,
91 union ldlm_wire_policy_data *wpolicy)
93 ldlm_policy_local_to_wire_t convert;
95 convert = ldlm_policy_local_to_wire[type - LDLM_MIN_TYPE];
97 convert(lpolicy, wpolicy);
101 * Converts lock policy from on the wire lock_desc format to local format
103 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
104 const union ldlm_wire_policy_data *wpolicy,
105 union ldlm_policy_data *lpolicy)
107 ldlm_policy_wire_to_local_t convert;
109 convert = ldlm_policy_wire_to_local[type - LDLM_MIN_TYPE];
111 convert(wpolicy, lpolicy);
114 const char *ldlm_it2str(enum ldlm_intent_flags it)
121 case (IT_OPEN | IT_CREAT):
136 CERROR("Unknown intent 0x%08x\n", it);
140 EXPORT_SYMBOL(ldlm_it2str);
142 extern struct kmem_cache *ldlm_lock_slab;
144 #ifdef HAVE_SERVER_SUPPORT
145 static ldlm_processing_policy ldlm_processing_policy_table[] = {
146 [LDLM_PLAIN] = ldlm_process_plain_lock,
147 [LDLM_EXTENT] = ldlm_process_extent_lock,
148 [LDLM_FLOCK] = ldlm_process_flock_lock,
149 [LDLM_IBITS] = ldlm_process_inodebits_lock,
152 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
154 return ldlm_processing_policy_table[res->lr_type];
156 EXPORT_SYMBOL(ldlm_get_processing_policy);
157 #endif /* HAVE_SERVER_SUPPORT */
159 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
163 EXPORT_SYMBOL(ldlm_register_intent);
166 * REFCOUNTED LOCK OBJECTS
171 * Get a reference on a lock.
173 * Lock refcounts, during creation:
174 * - one special one for allocation, dec'd only once in destroy
175 * - one for being a lock that's in-use
176 * - one for the addref associated with a new lock
178 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
180 atomic_inc(&lock->l_refc);
183 EXPORT_SYMBOL(ldlm_lock_get);
186 * Release lock reference.
188 * Also frees the lock if it was last reference.
190 void ldlm_lock_put(struct ldlm_lock *lock)
194 LASSERT(lock->l_resource != LP_POISON);
195 LASSERT(atomic_read(&lock->l_refc) > 0);
196 if (atomic_dec_and_test(&lock->l_refc)) {
197 struct ldlm_resource *res;
200 "final lock_put on destroyed lock, freeing it.");
202 res = lock->l_resource;
203 LASSERT(ldlm_is_destroyed(lock));
204 LASSERT(list_empty(&lock->l_exp_list));
205 LASSERT(list_empty(&lock->l_res_link));
206 LASSERT(list_empty(&lock->l_pending_chain));
208 lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats,
210 lu_ref_del(&res->lr_reference, "lock", lock);
211 ldlm_resource_putref(res);
212 lock->l_resource = NULL;
213 if (lock->l_export) {
214 class_export_lock_put(lock->l_export, lock);
215 lock->l_export = NULL;
218 if (lock->l_lvb_data != NULL)
219 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
221 ldlm_interval_free(ldlm_interval_detach(lock));
222 lu_ref_fini(&lock->l_reference);
223 OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle);
228 EXPORT_SYMBOL(ldlm_lock_put);
231 * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked.
233 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
236 if (!list_empty(&lock->l_lru)) {
237 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
239 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
240 list_del_init(&lock->l_lru);
241 LASSERT(ns->ns_nr_unused > 0);
249 * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first.
251 * If \a last_use is non-zero, it will remove the lock from LRU only if
252 * it matches lock's l_last_used.
254 * \retval 0 if \a last_use is set, the lock is not in LRU list or \a last_use
255 * doesn't match lock's l_last_used;
256 * otherwise, the lock hasn't been in the LRU list.
257 * \retval 1 the lock was in LRU list and removed.
259 int ldlm_lock_remove_from_lru_check(struct ldlm_lock *lock, cfs_time_t last_use)
261 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
265 if (ldlm_is_ns_srv(lock)) {
266 LASSERT(list_empty(&lock->l_lru));
270 spin_lock(&ns->ns_lock);
271 if (last_use == 0 || last_use == lock->l_last_used)
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 if (lock->l_export != NULL && lock->l_export->exp_failed) {
607 CDEBUG(D_INFO, "lock export failed: lock %p, exp %p\n",
608 lock, lock->l_export);
613 /* It's unlikely but possible that someone marked the lock as
614 * destroyed after we did handle2object on it */
615 if ((flags == 0) && !ldlm_is_destroyed(lock)) {
616 lu_ref_add(&lock->l_reference, "handle", current);
620 lock_res_and_lock(lock);
622 LASSERT(lock->l_resource != NULL);
624 lu_ref_add_atomic(&lock->l_reference, "handle", current);
625 if (unlikely(ldlm_is_destroyed(lock))) {
626 unlock_res_and_lock(lock);
627 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
632 /* If we're setting flags, make sure none of them are already set. */
634 if ((lock->l_flags & flags) != 0) {
635 unlock_res_and_lock(lock);
640 lock->l_flags |= flags;
643 unlock_res_and_lock(lock);
646 EXPORT_SYMBOL(__ldlm_handle2lock);
647 /** @} ldlm_handles */
650 * Fill in "on the wire" representation for given LDLM lock into supplied
651 * lock descriptor \a desc structure.
653 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
655 ldlm_res2desc(lock->l_resource, &desc->l_resource);
656 desc->l_req_mode = lock->l_req_mode;
657 desc->l_granted_mode = lock->l_granted_mode;
658 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
659 &lock->l_policy_data,
660 &desc->l_policy_data);
664 * Add a lock to list of conflicting locks to send AST to.
666 * Only add if we have not sent a blocking AST to the lock yet.
668 static void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
669 struct list_head *work_list)
671 if (!ldlm_is_ast_sent(lock)) {
672 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
673 ldlm_set_ast_sent(lock);
674 /* If the enqueuing client said so, tell the AST recipient to
675 * discard dirty data, rather than writing back. */
676 if (ldlm_is_ast_discard_data(new))
677 ldlm_set_discard_data(lock);
678 LASSERT(list_empty(&lock->l_bl_ast));
679 list_add(&lock->l_bl_ast, work_list);
681 LASSERT(lock->l_blocking_lock == NULL);
682 lock->l_blocking_lock = LDLM_LOCK_GET(new);
687 * Add a lock to list of just granted locks to send completion AST to.
689 static void ldlm_add_cp_work_item(struct ldlm_lock *lock,
690 struct list_head *work_list)
692 if (!ldlm_is_cp_reqd(lock)) {
693 ldlm_set_cp_reqd(lock);
694 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
695 LASSERT(list_empty(&lock->l_cp_ast));
696 list_add(&lock->l_cp_ast, work_list);
702 * Aggregator function to add AST work items into a list. Determines
703 * what sort of an AST work needs to be done and calls the proper
705 * Must be called with lr_lock held.
707 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
708 struct list_head *work_list)
711 check_res_locked(lock->l_resource);
713 ldlm_add_bl_work_item(lock, new, work_list);
715 ldlm_add_cp_work_item(lock, work_list);
720 * Add specified reader/writer reference to LDLM lock with handle \a lockh.
721 * r/w reference type is determined by \a mode
722 * Calls ldlm_lock_addref_internal.
724 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode)
726 struct ldlm_lock *lock;
728 lock = ldlm_handle2lock(lockh);
729 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
730 ldlm_lock_addref_internal(lock, mode);
733 EXPORT_SYMBOL(ldlm_lock_addref);
737 * Add specified reader/writer reference to LDLM lock \a lock.
738 * r/w reference type is determined by \a mode
739 * Removes lock from LRU if it is there.
740 * Assumes the LDLM lock is already locked.
742 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock,
745 ldlm_lock_remove_from_lru(lock);
746 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
748 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
750 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
752 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
755 lu_ref_add_atomic(&lock->l_reference, "user", lock);
756 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
760 * Attempts to add reader/writer reference to a lock with handle \a lockh, and
761 * fails if lock is already LDLM_FL_CBPENDING or destroyed.
763 * \retval 0 success, lock was addref-ed
765 * \retval -EAGAIN lock is being canceled.
767 int ldlm_lock_addref_try(const struct lustre_handle *lockh, enum ldlm_mode mode)
769 struct ldlm_lock *lock;
773 lock = ldlm_handle2lock(lockh);
775 lock_res_and_lock(lock);
776 if (lock->l_readers != 0 || lock->l_writers != 0 ||
777 !ldlm_is_cbpending(lock)) {
778 ldlm_lock_addref_internal_nolock(lock, mode);
781 unlock_res_and_lock(lock);
786 EXPORT_SYMBOL(ldlm_lock_addref_try);
789 * Add specified reader/writer reference to LDLM lock \a lock.
790 * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work.
791 * Only called for local locks.
793 void ldlm_lock_addref_internal(struct ldlm_lock *lock, enum ldlm_mode mode)
795 lock_res_and_lock(lock);
796 ldlm_lock_addref_internal_nolock(lock, mode);
797 unlock_res_and_lock(lock);
801 * Removes reader/writer reference for LDLM lock \a lock.
802 * Assumes LDLM lock is already locked.
803 * only called in ldlm_flock_destroy and for local locks.
804 * Does NOT add lock to LRU if no r/w references left to accomodate flock locks
805 * that cannot be placed in LRU.
807 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock,
810 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
811 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
812 LASSERT(lock->l_readers > 0);
813 lu_ref_del(&lock->l_reference, "reader", lock);
816 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
817 LASSERT(lock->l_writers > 0);
818 lu_ref_del(&lock->l_reference, "writer", lock);
822 lu_ref_del(&lock->l_reference, "user", lock);
823 LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */
827 * Removes reader/writer reference for LDLM lock \a lock.
828 * Locks LDLM lock first.
829 * If the lock is determined to be client lock on a client and r/w refcount
830 * drops to zero and the lock is not blocked, the lock is added to LRU lock
832 * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called.
834 void ldlm_lock_decref_internal(struct ldlm_lock *lock, enum ldlm_mode mode)
836 struct ldlm_namespace *ns;
839 lock_res_and_lock(lock);
841 ns = ldlm_lock_to_ns(lock);
843 ldlm_lock_decref_internal_nolock(lock, mode);
845 if ((ldlm_is_local(lock) || lock->l_req_mode == LCK_GROUP) &&
846 !lock->l_readers && !lock->l_writers) {
847 /* If this is a local lock on a server namespace and this was
848 * the last reference, cancel the lock.
850 * Group locks are special:
851 * They must not go in LRU, but they are not called back
852 * like non-group locks, instead they are manually released.
853 * They have an l_writers reference which they keep until
854 * they are manually released, so we remove them when they have
855 * no more reader or writer references. - LU-6368 */
856 ldlm_set_cbpending(lock);
859 if (!lock->l_readers && !lock->l_writers && ldlm_is_cbpending(lock)) {
860 /* If we received a blocked AST and this was the last reference,
861 * run the callback. */
862 if (ldlm_is_ns_srv(lock) && lock->l_export)
863 CERROR("FL_CBPENDING set on non-local lock--just a "
866 LDLM_DEBUG(lock, "final decref done on cbpending lock");
868 LDLM_LOCK_GET(lock); /* dropped by bl thread */
869 ldlm_lock_remove_from_lru(lock);
870 unlock_res_and_lock(lock);
872 if (ldlm_is_fail_loc(lock))
873 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
875 if (ldlm_is_atomic_cb(lock) ||
876 ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
877 ldlm_handle_bl_callback(ns, NULL, lock);
878 } else if (ns_is_client(ns) &&
879 !lock->l_readers && !lock->l_writers &&
880 !ldlm_is_no_lru(lock) &&
881 !ldlm_is_bl_ast(lock)) {
883 LDLM_DEBUG(lock, "add lock into lru list");
885 /* If this is a client-side namespace and this was the last
886 * reference, put it on the LRU. */
887 ldlm_lock_add_to_lru(lock);
888 unlock_res_and_lock(lock);
890 if (ldlm_is_fail_loc(lock))
891 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
893 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
894 * are not supported by the server, otherwise, it is done on
896 if (!exp_connect_cancelset(lock->l_conn_export) &&
897 !ns_connect_lru_resize(ns))
898 ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0);
900 LDLM_DEBUG(lock, "do not add lock into lru list");
901 unlock_res_and_lock(lock);
908 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
910 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode)
912 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
913 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
914 ldlm_lock_decref_internal(lock, mode);
917 EXPORT_SYMBOL(ldlm_lock_decref);
920 * Decrease reader/writer refcount for LDLM lock with handle
921 * \a lockh and mark it for subsequent cancellation once r/w refcount
922 * drops to zero instead of putting into LRU.
925 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
928 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
931 LASSERT(lock != NULL);
933 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
934 lock_res_and_lock(lock);
935 ldlm_set_cbpending(lock);
936 unlock_res_and_lock(lock);
937 ldlm_lock_decref_internal(lock, mode);
940 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
942 struct sl_insert_point {
943 struct list_head *res_link;
944 struct list_head *mode_link;
945 struct list_head *policy_link;
949 * Finds a position to insert the new lock into granted lock list.
951 * Used for locks eligible for skiplist optimization.
954 * queue [input]: the granted list where search acts on;
955 * req [input]: the lock whose position to be located;
956 * prev [output]: positions within 3 lists to insert @req to
960 * - ldlm_grant_lock_with_skiplist
962 static void search_granted_lock(struct list_head *queue,
963 struct ldlm_lock *req,
964 struct sl_insert_point *prev)
966 struct list_head *tmp;
967 struct ldlm_lock *lock, *mode_end, *policy_end;
970 list_for_each(tmp, queue) {
971 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
973 mode_end = list_entry(lock->l_sl_mode.prev,
974 struct ldlm_lock, l_sl_mode);
976 if (lock->l_req_mode != req->l_req_mode) {
977 /* jump to last lock of mode group */
978 tmp = &mode_end->l_res_link;
982 /* suitable mode group is found */
983 if (lock->l_resource->lr_type == LDLM_PLAIN) {
984 /* insert point is last lock of the mode group */
985 prev->res_link = &mode_end->l_res_link;
986 prev->mode_link = &mode_end->l_sl_mode;
987 prev->policy_link = &req->l_sl_policy;
990 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
993 list_entry(lock->l_sl_policy.prev,
997 if (lock->l_policy_data.l_inodebits.bits ==
998 req->l_policy_data.l_inodebits.bits) {
999 /* insert point is last lock of
1000 * the policy group */
1002 &policy_end->l_res_link;
1004 &policy_end->l_sl_mode;
1006 &policy_end->l_sl_policy;
1011 if (policy_end == mode_end)
1012 /* done with mode group */
1015 /* go to next policy group within mode group */
1016 tmp = policy_end->l_res_link.next;
1017 lock = list_entry(tmp, struct ldlm_lock,
1019 } /* loop over policy groups within the mode group */
1021 /* insert point is last lock of the mode group,
1022 * new policy group is started */
1023 prev->res_link = &mode_end->l_res_link;
1024 prev->mode_link = &mode_end->l_sl_mode;
1025 prev->policy_link = &req->l_sl_policy;
1029 LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
1034 /* insert point is last lock on the queue,
1035 * new mode group and new policy group are started */
1036 prev->res_link = queue->prev;
1037 prev->mode_link = &req->l_sl_mode;
1038 prev->policy_link = &req->l_sl_policy;
1044 * Add a lock into resource granted list after a position described by
1047 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
1048 struct sl_insert_point *prev)
1050 struct ldlm_resource *res = lock->l_resource;
1053 check_res_locked(res);
1055 ldlm_resource_dump(D_INFO, res);
1056 LDLM_DEBUG(lock, "About to add lock:");
1058 if (ldlm_is_destroyed(lock)) {
1059 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1063 LASSERT(list_empty(&lock->l_res_link));
1064 LASSERT(list_empty(&lock->l_sl_mode));
1065 LASSERT(list_empty(&lock->l_sl_policy));
1068 * lock->link == prev->link means lock is first starting the group.
1069 * Don't re-add to itself to suppress kernel warnings.
1071 if (&lock->l_res_link != prev->res_link)
1072 list_add(&lock->l_res_link, prev->res_link);
1073 if (&lock->l_sl_mode != prev->mode_link)
1074 list_add(&lock->l_sl_mode, prev->mode_link);
1075 if (&lock->l_sl_policy != prev->policy_link)
1076 list_add(&lock->l_sl_policy, prev->policy_link);
1082 * Add a lock to granted list on a resource maintaining skiplist
1085 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
1087 struct sl_insert_point prev;
1090 LASSERT(lock->l_req_mode == lock->l_granted_mode);
1092 search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
1093 ldlm_granted_list_add_lock(lock, &prev);
1098 * Perform lock granting bookkeeping.
1100 * Includes putting the lock into granted list and updating lock mode.
1102 * - ldlm_lock_enqueue
1103 * - ldlm_reprocess_queue
1104 * - ldlm_lock_convert
1106 * must be called with lr_lock held
1108 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
1110 struct ldlm_resource *res = lock->l_resource;
1113 check_res_locked(res);
1115 lock->l_granted_mode = lock->l_req_mode;
1117 if (work_list && lock->l_completion_ast != NULL)
1118 ldlm_add_ast_work_item(lock, NULL, work_list);
1120 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
1121 ldlm_grant_lock_with_skiplist(lock);
1122 else if (res->lr_type == LDLM_EXTENT)
1123 ldlm_extent_add_lock(res, lock);
1124 else if (res->lr_type == LDLM_FLOCK) {
1125 /* We should not add locks to granted list in the following
1127 * - this is an UNLOCK but not a real lock;
1128 * - this is a TEST lock;
1129 * - this is a F_CANCELLK lock (async flock has req_mode == 0)
1130 * - this is a deadlock (flock cannot be granted) */
1131 if (lock->l_req_mode == 0 ||
1132 lock->l_req_mode == LCK_NL ||
1133 ldlm_is_test_lock(lock) ||
1134 ldlm_is_flock_deadlock(lock))
1136 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1141 ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1146 * Describe the overlap between two locks. itree_overlap_cb data.
1148 struct lock_match_data {
1149 struct ldlm_lock *lmd_old;
1150 struct ldlm_lock *lmd_lock;
1151 enum ldlm_mode *lmd_mode;
1152 union ldlm_policy_data *lmd_policy;
1158 * Check if the given @lock meets the criteria for a match.
1159 * A reference on the lock is taken if matched.
1161 * \param lock test-against this lock
1162 * \param data parameters
1164 static int lock_matches(struct ldlm_lock *lock, struct lock_match_data *data)
1166 union ldlm_policy_data *lpol = &lock->l_policy_data;
1167 enum ldlm_mode match;
1169 if (lock == data->lmd_old)
1170 return INTERVAL_ITER_STOP;
1172 /* Check if this lock can be matched.
1173 * Used by LU-2919(exclusive open) for open lease lock */
1174 if (ldlm_is_excl(lock))
1175 return INTERVAL_ITER_CONT;
1177 /* llite sometimes wants to match locks that will be
1178 * canceled when their users drop, but we allow it to match
1179 * if it passes in CBPENDING and the lock still has users.
1180 * this is generally only going to be used by children
1181 * whose parents already hold a lock so forward progress
1182 * can still happen. */
1183 if (ldlm_is_cbpending(lock) &&
1184 !(data->lmd_flags & LDLM_FL_CBPENDING))
1185 return INTERVAL_ITER_CONT;
1186 if (!data->lmd_unref && ldlm_is_cbpending(lock) &&
1187 lock->l_readers == 0 && lock->l_writers == 0)
1188 return INTERVAL_ITER_CONT;
1190 if (!(lock->l_req_mode & *data->lmd_mode))
1191 return INTERVAL_ITER_CONT;
1192 match = lock->l_req_mode;
1194 switch (lock->l_resource->lr_type) {
1196 if (lpol->l_extent.start > data->lmd_policy->l_extent.start ||
1197 lpol->l_extent.end < data->lmd_policy->l_extent.end)
1198 return INTERVAL_ITER_CONT;
1200 if (unlikely(match == LCK_GROUP) &&
1201 data->lmd_policy->l_extent.gid != LDLM_GID_ANY &&
1202 lpol->l_extent.gid != data->lmd_policy->l_extent.gid)
1203 return INTERVAL_ITER_CONT;
1206 /* We match if we have existing lock with same or wider set
1208 if ((lpol->l_inodebits.bits &
1209 data->lmd_policy->l_inodebits.bits) !=
1210 data->lmd_policy->l_inodebits.bits)
1211 return INTERVAL_ITER_CONT;
1217 /* We match if we have existing lock with same or wider set
1219 if (!data->lmd_unref && LDLM_HAVE_MASK(lock, GONE))
1220 return INTERVAL_ITER_CONT;
1222 if (!equi(data->lmd_flags & LDLM_FL_LOCAL_ONLY, ldlm_is_local(lock)))
1223 return INTERVAL_ITER_CONT;
1225 if (data->lmd_flags & LDLM_FL_TEST_LOCK) {
1226 LDLM_LOCK_GET(lock);
1227 ldlm_lock_touch_in_lru(lock);
1229 ldlm_lock_addref_internal_nolock(lock, match);
1232 *data->lmd_mode = match;
1233 data->lmd_lock = lock;
1235 return INTERVAL_ITER_STOP;
1238 static unsigned int itree_overlap_cb(struct interval_node *in, void *args)
1240 struct ldlm_interval *node = to_ldlm_interval(in);
1241 struct lock_match_data *data = args;
1242 struct ldlm_lock *lock;
1245 list_for_each_entry(lock, &node->li_group, l_sl_policy) {
1246 rc = lock_matches(lock, data);
1247 if (rc == INTERVAL_ITER_STOP)
1248 return INTERVAL_ITER_STOP;
1250 return INTERVAL_ITER_CONT;
1254 * Search for a lock with given parameters in interval trees.
1256 * \param res search for a lock in this resource
1257 * \param data parameters
1259 * \retval a referenced lock or NULL.
1261 static struct ldlm_lock *search_itree(struct ldlm_resource *res,
1262 struct lock_match_data *data)
1264 struct interval_node_extent ext = {
1265 .start = data->lmd_policy->l_extent.start,
1266 .end = data->lmd_policy->l_extent.end
1270 for (idx = 0; idx < LCK_MODE_NUM; idx++) {
1271 struct ldlm_interval_tree *tree = &res->lr_itree[idx];
1273 if (tree->lit_root == NULL)
1276 if (!(tree->lit_mode & *data->lmd_mode))
1279 interval_search(tree->lit_root, &ext,
1280 itree_overlap_cb, data);
1282 return data->lmd_lock;
1287 * Search for a lock with given properties in a queue.
1289 * \param queue search for a lock in this queue
1290 * \param data parameters
1292 * \retval a referenced lock or NULL.
1294 static struct ldlm_lock *search_queue(struct list_head *queue,
1295 struct lock_match_data *data)
1297 struct ldlm_lock *lock;
1300 list_for_each_entry(lock, queue, l_res_link) {
1301 rc = lock_matches(lock, data);
1302 if (rc == INTERVAL_ITER_STOP)
1303 return data->lmd_lock;
1308 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1310 if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) {
1311 lock->l_flags |= LDLM_FL_FAIL_NOTIFIED;
1312 wake_up_all(&lock->l_waitq);
1315 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1317 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1319 lock_res_and_lock(lock);
1320 ldlm_lock_fail_match_locked(lock);
1321 unlock_res_and_lock(lock);
1325 * Mark lock as "matchable" by OST.
1327 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1329 * Assumes LDLM lock is already locked.
1331 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1333 ldlm_set_lvb_ready(lock);
1334 wake_up_all(&lock->l_waitq);
1336 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1339 * Mark lock as "matchable" by OST.
1340 * Locks the lock and then \see ldlm_lock_allow_match_locked
1342 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1344 lock_res_and_lock(lock);
1345 ldlm_lock_allow_match_locked(lock);
1346 unlock_res_and_lock(lock);
1348 EXPORT_SYMBOL(ldlm_lock_allow_match);
1351 * Attempt to find a lock with specified properties.
1353 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1356 * Can be called in two ways:
1358 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1359 * for a duplicate of.
1361 * Otherwise, all of the fields must be filled in, to match against.
1363 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1364 * server (ie, connh is NULL)
1365 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1366 * list will be considered
1367 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1368 * to be canceled can still be matched as long as they still have reader
1369 * or writer refernces
1370 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1371 * just tell us if we would have matched.
1373 * \retval 1 if it finds an already-existing lock that is compatible; in this
1374 * case, lockh is filled in with a addref()ed lock
1376 * We also check security context, and if that fails we simply return 0 (to
1377 * keep caller code unchanged), the context failure will be discovered by
1378 * caller sometime later.
1380 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1381 const struct ldlm_res_id *res_id,
1382 enum ldlm_type type,
1383 union ldlm_policy_data *policy,
1384 enum ldlm_mode mode,
1385 struct lustre_handle *lockh, int unref)
1387 struct lock_match_data data = {
1391 .lmd_policy = policy,
1395 struct ldlm_resource *res;
1396 struct ldlm_lock *lock;
1401 data.lmd_old = ldlm_handle2lock(lockh);
1402 LASSERT(data.lmd_old != NULL);
1404 ns = ldlm_lock_to_ns(data.lmd_old);
1405 res_id = &data.lmd_old->l_resource->lr_name;
1406 type = data.lmd_old->l_resource->lr_type;
1407 *data.lmd_mode = data.lmd_old->l_req_mode;
1410 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1412 LASSERT(data.lmd_old == NULL);
1416 LDLM_RESOURCE_ADDREF(res);
1419 if (res->lr_type == LDLM_EXTENT)
1420 lock = search_itree(res, &data);
1422 lock = search_queue(&res->lr_granted, &data);
1425 if (flags & LDLM_FL_BLOCK_GRANTED)
1427 lock = search_queue(&res->lr_converting, &data);
1430 lock = search_queue(&res->lr_waiting, &data);
1437 LDLM_RESOURCE_DELREF(res);
1438 ldlm_resource_putref(res);
1441 ldlm_lock2handle(lock, lockh);
1442 if ((flags & LDLM_FL_LVB_READY) &&
1443 (!ldlm_is_lvb_ready(lock))) {
1444 __u64 wait_flags = LDLM_FL_LVB_READY |
1445 LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED;
1446 struct l_wait_info lwi;
1447 if (lock->l_completion_ast) {
1448 int err = lock->l_completion_ast(lock,
1449 LDLM_FL_WAIT_NOREPROC,
1452 if (flags & LDLM_FL_TEST_LOCK)
1453 LDLM_LOCK_RELEASE(lock);
1455 ldlm_lock_decref_internal(lock,
1462 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1463 NULL, LWI_ON_SIGNAL_NOOP, NULL);
1465 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1466 l_wait_event(lock->l_waitq,
1467 lock->l_flags & wait_flags,
1469 if (!ldlm_is_lvb_ready(lock)) {
1470 if (flags & LDLM_FL_TEST_LOCK)
1471 LDLM_LOCK_RELEASE(lock);
1473 ldlm_lock_decref_internal(lock, mode);
1480 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1481 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1482 res_id->name[2] : policy->l_extent.start,
1483 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1484 res_id->name[3] : policy->l_extent.end);
1486 /* check user's security context */
1487 if (lock->l_conn_export &&
1488 sptlrpc_import_check_ctx(
1489 class_exp2cliimp(lock->l_conn_export))) {
1490 if (!(flags & LDLM_FL_TEST_LOCK))
1491 ldlm_lock_decref_internal(lock, mode);
1495 if (flags & LDLM_FL_TEST_LOCK)
1496 LDLM_LOCK_RELEASE(lock);
1498 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1499 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1500 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1501 type, mode, res_id->name[0], res_id->name[1],
1502 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1503 res_id->name[2] :policy->l_extent.start,
1504 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1505 res_id->name[3] : policy->l_extent.end);
1507 if (data.lmd_old != NULL)
1508 LDLM_LOCK_PUT(data.lmd_old);
1510 return rc ? mode : 0;
1512 EXPORT_SYMBOL(ldlm_lock_match);
1514 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1517 struct ldlm_lock *lock;
1518 enum ldlm_mode mode = 0;
1521 lock = ldlm_handle2lock(lockh);
1523 lock_res_and_lock(lock);
1524 if (LDLM_HAVE_MASK(lock, GONE))
1527 if (ldlm_is_cbpending(lock) &&
1528 lock->l_readers == 0 && lock->l_writers == 0)
1532 *bits = lock->l_policy_data.l_inodebits.bits;
1533 mode = lock->l_granted_mode;
1534 ldlm_lock_addref_internal_nolock(lock, mode);
1541 unlock_res_and_lock(lock);
1542 LDLM_LOCK_PUT(lock);
1546 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1548 /** The caller must guarantee that the buffer is large enough. */
1549 int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill,
1550 enum req_location loc, void *data, int size)
1555 LASSERT(data != NULL);
1558 switch (lock->l_lvb_type) {
1560 if (size == sizeof(struct ost_lvb)) {
1561 if (loc == RCL_CLIENT)
1562 lvb = req_capsule_client_swab_get(pill,
1564 lustre_swab_ost_lvb);
1566 lvb = req_capsule_server_swab_get(pill,
1568 lustre_swab_ost_lvb);
1569 if (unlikely(lvb == NULL)) {
1570 LDLM_ERROR(lock, "no LVB");
1574 memcpy(data, lvb, size);
1575 } else if (size == sizeof(struct ost_lvb_v1)) {
1576 struct ost_lvb *olvb = data;
1578 if (loc == RCL_CLIENT)
1579 lvb = req_capsule_client_swab_get(pill,
1581 lustre_swab_ost_lvb_v1);
1583 lvb = req_capsule_server_sized_swab_get(pill,
1585 lustre_swab_ost_lvb_v1);
1586 if (unlikely(lvb == NULL)) {
1587 LDLM_ERROR(lock, "no LVB");
1591 memcpy(data, lvb, size);
1592 olvb->lvb_mtime_ns = 0;
1593 olvb->lvb_atime_ns = 0;
1594 olvb->lvb_ctime_ns = 0;
1596 LDLM_ERROR(lock, "Replied unexpected ost LVB size %d",
1602 if (size == sizeof(struct lquota_lvb)) {
1603 if (loc == RCL_CLIENT)
1604 lvb = req_capsule_client_swab_get(pill,
1606 lustre_swab_lquota_lvb);
1608 lvb = req_capsule_server_swab_get(pill,
1610 lustre_swab_lquota_lvb);
1611 if (unlikely(lvb == NULL)) {
1612 LDLM_ERROR(lock, "no LVB");
1616 memcpy(data, lvb, size);
1618 LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d",
1627 if (loc == RCL_CLIENT)
1628 lvb = req_capsule_client_get(pill, &RMF_DLM_LVB);
1630 lvb = req_capsule_server_get(pill, &RMF_DLM_LVB);
1631 if (unlikely(lvb == NULL)) {
1632 LDLM_ERROR(lock, "no LVB");
1636 memcpy(data, lvb, size);
1639 LDLM_ERROR(lock, "Unknown LVB type: %d", lock->l_lvb_type);
1640 libcfs_debug_dumpstack(NULL);
1648 * Create and fill in new LDLM lock with specified properties.
1649 * Returns a referenced lock
1651 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1652 const struct ldlm_res_id *res_id,
1653 enum ldlm_type type,
1654 enum ldlm_mode mode,
1655 const struct ldlm_callback_suite *cbs,
1656 void *data, __u32 lvb_len,
1657 enum lvb_type lvb_type)
1659 struct ldlm_lock *lock;
1660 struct ldlm_resource *res;
1664 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1666 RETURN(ERR_CAST(res));
1668 lock = ldlm_lock_new(res);
1670 RETURN(ERR_PTR(-ENOMEM));
1672 lock->l_req_mode = mode;
1673 lock->l_ast_data = data;
1674 lock->l_pid = current_pid();
1675 if (ns_is_server(ns))
1676 ldlm_set_ns_srv(lock);
1678 lock->l_blocking_ast = cbs->lcs_blocking;
1679 lock->l_completion_ast = cbs->lcs_completion;
1680 lock->l_glimpse_ast = cbs->lcs_glimpse;
1683 lock->l_tree_node = NULL;
1684 /* if this is the extent lock, allocate the interval tree node */
1685 if (type == LDLM_EXTENT)
1686 if (ldlm_interval_alloc(lock) == NULL)
1687 GOTO(out, rc = -ENOMEM);
1690 lock->l_lvb_len = lvb_len;
1691 OBD_ALLOC_LARGE(lock->l_lvb_data, lvb_len);
1692 if (lock->l_lvb_data == NULL)
1693 GOTO(out, rc = -ENOMEM);
1696 lock->l_lvb_type = lvb_type;
1697 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1698 GOTO(out, rc = -ENOENT);
1703 ldlm_lock_destroy(lock);
1704 LDLM_LOCK_RELEASE(lock);
1705 RETURN(ERR_PTR(rc));
1709 * Enqueue (request) a lock.
1711 * Does not block. As a result of enqueue the lock would be put
1712 * into granted or waiting list.
1714 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1715 * set, skip all the enqueueing and delegate lock processing to intent policy
1718 enum ldlm_error ldlm_lock_enqueue(struct ldlm_namespace *ns,
1719 struct ldlm_lock **lockp,
1720 void *cookie, __u64 *flags)
1722 struct ldlm_lock *lock = *lockp;
1723 struct ldlm_resource *res = lock->l_resource;
1724 int local = ns_is_client(ldlm_res_to_ns(res));
1725 #ifdef HAVE_SERVER_SUPPORT
1726 ldlm_processing_policy policy;
1728 enum ldlm_error rc = ELDLM_OK;
1729 struct ldlm_interval *node = NULL;
1732 /* policies are not executed on the client or during replay */
1733 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1734 && !local && ns->ns_policy) {
1735 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1737 if (rc == ELDLM_LOCK_REPLACED) {
1738 /* The lock that was returned has already been granted,
1739 * and placed into lockp. If it's not the same as the
1740 * one we passed in, then destroy the old one and our
1741 * work here is done. */
1742 if (lock != *lockp) {
1743 ldlm_lock_destroy(lock);
1744 LDLM_LOCK_RELEASE(lock);
1746 *flags |= LDLM_FL_LOCK_CHANGED;
1748 } else if (rc != ELDLM_OK &&
1749 lock->l_req_mode == lock->l_granted_mode) {
1750 LASSERT(*flags & LDLM_FL_RESENT);
1751 /* It may happen that ns_policy returns an error in
1752 * resend case, object may be unlinked or just some
1753 * error occurs. It is unclear if lock reached the
1754 * client in the original reply, just leave the lock on
1755 * server, not returning it again to client. Due to
1756 * LU-6529, the server will not OOM. */
1758 } else if (rc != ELDLM_OK ||
1759 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1760 ldlm_lock_destroy(lock);
1765 if (*flags & LDLM_FL_RESENT) {
1766 /* Reconstruct LDLM_FL_SRV_ENQ_MASK @flags for reply.
1767 * Set LOCK_CHANGED always.
1768 * Check if the lock is granted for BLOCK_GRANTED.
1769 * Take NO_TIMEOUT from the lock as it is inherited through
1770 * LDLM_FL_INHERIT_MASK */
1771 *flags |= LDLM_FL_LOCK_CHANGED;
1772 if (lock->l_req_mode != lock->l_granted_mode)
1773 *flags |= LDLM_FL_BLOCK_GRANTED;
1774 *flags |= lock->l_flags & LDLM_FL_NO_TIMEOUT;
1778 /* For a replaying lock, it might be already in granted list. So
1779 * unlinking the lock will cause the interval node to be freed, we
1780 * have to allocate the interval node early otherwise we can't regrant
1781 * this lock in the future. - jay */
1782 if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1783 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
1785 lock_res_and_lock(lock);
1786 if (local && lock->l_req_mode == lock->l_granted_mode) {
1787 /* The server returned a blocked lock, but it was granted
1788 * before we got a chance to actually enqueue it. We don't
1789 * need to do anything else. */
1790 *flags &= ~LDLM_FL_BLOCKED_MASK;
1791 GOTO(out, rc = ELDLM_OK);
1794 ldlm_resource_unlink_lock(lock);
1795 if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1797 ldlm_lock_destroy_nolock(lock);
1798 GOTO(out, rc = -ENOMEM);
1801 INIT_LIST_HEAD(&node->li_group);
1802 ldlm_interval_attach(node, lock);
1806 /* Some flags from the enqueue want to make it into the AST, via the
1807 * lock's l_flags. */
1808 if (*flags & LDLM_FL_AST_DISCARD_DATA)
1809 ldlm_set_ast_discard_data(lock);
1810 if (*flags & LDLM_FL_TEST_LOCK)
1811 ldlm_set_test_lock(lock);
1812 if (*flags & LDLM_FL_COS_INCOMPAT)
1813 ldlm_set_cos_incompat(lock);
1814 if (*flags & LDLM_FL_COS_ENABLED)
1815 ldlm_set_cos_enabled(lock);
1817 /* This distinction between local lock trees is very important; a client
1818 * namespace only has information about locks taken by that client, and
1819 * thus doesn't have enough information to decide for itself if it can
1820 * be granted (below). In this case, we do exactly what the server
1821 * tells us to do, as dictated by the 'flags'.
1823 * We do exactly the same thing during recovery, when the server is
1824 * more or less trusting the clients not to lie.
1826 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1827 * granted/converting queues. */
1829 if (*flags & LDLM_FL_BLOCK_CONV)
1830 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1831 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1832 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1834 ldlm_grant_lock(lock, NULL);
1835 GOTO(out, rc = ELDLM_OK);
1836 #ifdef HAVE_SERVER_SUPPORT
1837 } else if (*flags & LDLM_FL_REPLAY) {
1838 if (*flags & LDLM_FL_BLOCK_CONV) {
1839 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1840 GOTO(out, rc = ELDLM_OK);
1841 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1842 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1843 GOTO(out, rc = ELDLM_OK);
1844 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1845 ldlm_grant_lock(lock, NULL);
1846 GOTO(out, rc = ELDLM_OK);
1848 /* If no flags, fall through to normal enqueue path. */
1851 policy = ldlm_processing_policy_table[res->lr_type];
1852 policy(lock, flags, 1, &rc, NULL);
1856 CERROR("This is client-side-only module, cannot handle "
1857 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1863 unlock_res_and_lock(lock);
1865 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1869 #ifdef HAVE_SERVER_SUPPORT
1871 * Iterate through all waiting locks on a given resource queue and attempt to
1874 * Must be called with resource lock held.
1876 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1877 struct list_head *work_list)
1879 struct list_head *tmp, *pos;
1880 ldlm_processing_policy policy;
1882 int rc = LDLM_ITER_CONTINUE;
1883 enum ldlm_error err;
1886 check_res_locked(res);
1888 policy = ldlm_processing_policy_table[res->lr_type];
1891 list_for_each_safe(tmp, pos, queue) {
1892 struct ldlm_lock *pending;
1894 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1896 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1899 rc = policy(pending, &flags, 0, &err, work_list);
1900 if (rc != LDLM_ITER_CONTINUE)
1909 * Process a call to blocking AST callback for a lock in ast_work list
1912 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1914 struct ldlm_cb_set_arg *arg = opaq;
1915 struct ldlm_lock_desc d;
1917 struct ldlm_lock *lock;
1920 if (list_empty(arg->list))
1923 lock = list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1925 /* nobody should touch l_bl_ast */
1926 lock_res_and_lock(lock);
1927 list_del_init(&lock->l_bl_ast);
1929 LASSERT(ldlm_is_ast_sent(lock));
1930 LASSERT(lock->l_bl_ast_run == 0);
1931 LASSERT(lock->l_blocking_lock);
1932 lock->l_bl_ast_run++;
1933 unlock_res_and_lock(lock);
1935 ldlm_lock2desc(lock->l_blocking_lock, &d);
1937 rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1938 LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1939 lock->l_blocking_lock = NULL;
1940 LDLM_LOCK_RELEASE(lock);
1946 * Process a call to completion AST callback for a lock in ast_work list
1949 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1951 struct ldlm_cb_set_arg *arg = opaq;
1953 struct ldlm_lock *lock;
1954 ldlm_completion_callback completion_callback;
1957 if (list_empty(arg->list))
1960 lock = list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1962 /* It's possible to receive a completion AST before we've set
1963 * the l_completion_ast pointer: either because the AST arrived
1964 * before the reply, or simply because there's a small race
1965 * window between receiving the reply and finishing the local
1966 * enqueue. (bug 842)
1968 * This can't happen with the blocking_ast, however, because we
1969 * will never call the local blocking_ast until we drop our
1970 * reader/writer reference, which we won't do until we get the
1971 * reply and finish enqueueing. */
1973 /* nobody should touch l_cp_ast */
1974 lock_res_and_lock(lock);
1975 list_del_init(&lock->l_cp_ast);
1976 LASSERT(ldlm_is_cp_reqd(lock));
1977 /* save l_completion_ast since it can be changed by
1978 * mds_intent_policy(), see bug 14225 */
1979 completion_callback = lock->l_completion_ast;
1980 ldlm_clear_cp_reqd(lock);
1981 unlock_res_and_lock(lock);
1983 if (completion_callback != NULL)
1984 rc = completion_callback(lock, 0, (void *)arg);
1985 LDLM_LOCK_RELEASE(lock);
1991 * Process a call to revocation AST callback for a lock in ast_work list
1994 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1996 struct ldlm_cb_set_arg *arg = opaq;
1997 struct ldlm_lock_desc desc;
1999 struct ldlm_lock *lock;
2002 if (list_empty(arg->list))
2005 lock = list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
2006 list_del_init(&lock->l_rk_ast);
2008 /* the desc just pretend to exclusive */
2009 ldlm_lock2desc(lock, &desc);
2010 desc.l_req_mode = LCK_EX;
2011 desc.l_granted_mode = 0;
2013 rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
2014 LDLM_LOCK_RELEASE(lock);
2020 * Process a call to glimpse AST callback for a lock in ast_work list
2022 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
2024 struct ldlm_cb_set_arg *arg = opaq;
2025 struct ldlm_glimpse_work *gl_work;
2026 struct ldlm_lock *lock;
2030 if (list_empty(arg->list))
2033 gl_work = list_entry(arg->list->next, struct ldlm_glimpse_work,
2035 list_del_init(&gl_work->gl_list);
2037 lock = gl_work->gl_lock;
2039 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
2040 arg->gl_desc = gl_work->gl_desc;
2042 /* invoke the actual glimpse callback */
2043 if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
2046 LDLM_LOCK_RELEASE(lock);
2048 if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
2049 OBD_FREE_PTR(gl_work);
2055 * Process list of locks in need of ASTs being sent.
2057 * Used on server to send multiple ASTs together instead of sending one by
2060 int ldlm_run_ast_work(struct ldlm_namespace *ns, struct list_head *rpc_list,
2061 ldlm_desc_ast_t ast_type)
2063 struct ldlm_cb_set_arg *arg;
2064 set_producer_func work_ast_lock;
2067 if (list_empty(rpc_list))
2074 atomic_set(&arg->restart, 0);
2075 arg->list = rpc_list;
2078 case LDLM_WORK_BL_AST:
2079 arg->type = LDLM_BL_CALLBACK;
2080 work_ast_lock = ldlm_work_bl_ast_lock;
2082 case LDLM_WORK_CP_AST:
2083 arg->type = LDLM_CP_CALLBACK;
2084 work_ast_lock = ldlm_work_cp_ast_lock;
2086 case LDLM_WORK_REVOKE_AST:
2087 arg->type = LDLM_BL_CALLBACK;
2088 work_ast_lock = ldlm_work_revoke_ast_lock;
2090 case LDLM_WORK_GL_AST:
2091 arg->type = LDLM_GL_CALLBACK;
2092 work_ast_lock = ldlm_work_gl_ast_lock;
2098 /* We create a ptlrpc request set with flow control extension.
2099 * This request set will use the work_ast_lock function to produce new
2100 * requests and will send a new request each time one completes in order
2101 * to keep the number of requests in flight to ns_max_parallel_ast */
2102 arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
2103 work_ast_lock, arg);
2104 if (arg->set == NULL)
2105 GOTO(out, rc = -ENOMEM);
2107 ptlrpc_set_wait(arg->set);
2108 ptlrpc_set_destroy(arg->set);
2110 rc = atomic_read(&arg->restart) ? -ERESTART : 0;
2117 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
2119 ldlm_reprocess_all(res);
2120 return LDLM_ITER_CONTINUE;
2123 static int ldlm_reprocess_res(struct cfs_hash *hs, struct cfs_hash_bd *bd,
2124 struct hlist_node *hnode, void *arg)
2126 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2129 rc = reprocess_one_queue(res, arg);
2131 return rc == LDLM_ITER_STOP;
2135 * Iterate through all resources on a namespace attempting to grant waiting
2138 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
2143 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2144 ldlm_reprocess_res, NULL, 0);
2150 * Try to grant all waiting locks on a resource.
2152 * Calls ldlm_reprocess_queue on converting and waiting queues.
2154 * Typically called after some resource locks are cancelled to see
2155 * if anything could be granted as a result of the cancellation.
2157 void ldlm_reprocess_all(struct ldlm_resource *res)
2159 struct list_head rpc_list;
2160 #ifdef HAVE_SERVER_SUPPORT
2164 INIT_LIST_HEAD(&rpc_list);
2165 /* Local lock trees don't get reprocessed. */
2166 if (ns_is_client(ldlm_res_to_ns(res))) {
2173 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
2174 if (rc == LDLM_ITER_CONTINUE)
2175 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
2178 rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
2180 if (rc == -ERESTART) {
2181 LASSERT(list_empty(&rpc_list));
2187 INIT_LIST_HEAD(&rpc_list);
2188 if (!ns_is_client(ldlm_res_to_ns(res))) {
2189 CERROR("This is client-side-only module, cannot handle "
2190 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2196 EXPORT_SYMBOL(ldlm_reprocess_all);
2198 static bool is_bl_done(struct ldlm_lock *lock)
2200 bool bl_done = true;
2202 if (!ldlm_is_bl_done(lock)) {
2203 lock_res_and_lock(lock);
2204 bl_done = ldlm_is_bl_done(lock);
2205 unlock_res_and_lock(lock);
2212 * Helper function to call blocking AST for LDLM lock \a lock in a
2213 * "cancelling" mode.
2215 void ldlm_cancel_callback(struct ldlm_lock *lock)
2217 check_res_locked(lock->l_resource);
2218 if (!ldlm_is_cancel(lock)) {
2219 ldlm_set_cancel(lock);
2220 if (lock->l_blocking_ast) {
2221 unlock_res_and_lock(lock);
2222 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
2224 lock_res_and_lock(lock);
2226 LDLM_DEBUG(lock, "no blocking ast");
2229 /* only canceller can set bl_done bit */
2230 ldlm_set_bl_done(lock);
2231 wake_up_all(&lock->l_waitq);
2232 } else if (!ldlm_is_bl_done(lock)) {
2233 struct l_wait_info lwi = { 0 };
2235 /* The lock is guaranteed to have been canceled once
2236 * returning from this function. */
2237 unlock_res_and_lock(lock);
2238 l_wait_event(lock->l_waitq, is_bl_done(lock), &lwi);
2239 lock_res_and_lock(lock);
2244 * Remove skiplist-enabled LDLM lock \a req from granted list
2246 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
2248 if (req->l_resource->lr_type != LDLM_PLAIN &&
2249 req->l_resource->lr_type != LDLM_IBITS)
2252 list_del_init(&req->l_sl_policy);
2253 list_del_init(&req->l_sl_mode);
2257 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2259 void ldlm_lock_cancel(struct ldlm_lock *lock)
2261 struct ldlm_resource *res;
2262 struct ldlm_namespace *ns;
2265 lock_res_and_lock(lock);
2267 res = lock->l_resource;
2268 ns = ldlm_res_to_ns(res);
2270 /* Please do not, no matter how tempting, remove this LBUG without
2271 * talking to me first. -phik */
2272 if (lock->l_readers || lock->l_writers) {
2273 LDLM_ERROR(lock, "lock still has references");
2277 if (ldlm_is_waited(lock))
2278 ldlm_del_waiting_lock(lock);
2280 /* Releases cancel callback. */
2281 ldlm_cancel_callback(lock);
2283 /* Yes, second time, just in case it was added again while we were
2284 * running with no res lock in ldlm_cancel_callback */
2285 if (ldlm_is_waited(lock))
2286 ldlm_del_waiting_lock(lock);
2288 ldlm_resource_unlink_lock(lock);
2289 ldlm_lock_destroy_nolock(lock);
2291 if (lock->l_granted_mode == lock->l_req_mode)
2292 ldlm_pool_del(&ns->ns_pool, lock);
2294 /* Make sure we will not be called again for same lock what is possible
2295 * if not to zero out lock->l_granted_mode */
2296 lock->l_granted_mode = LCK_MINMODE;
2297 unlock_res_and_lock(lock);
2301 EXPORT_SYMBOL(ldlm_lock_cancel);
2304 * Set opaque data into the lock that only makes sense to upper layer.
2306 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data)
2308 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2313 if (lock->l_ast_data == NULL)
2314 lock->l_ast_data = data;
2315 if (lock->l_ast_data == data)
2317 LDLM_LOCK_PUT(lock);
2321 EXPORT_SYMBOL(ldlm_lock_set_data);
2323 struct export_cl_data {
2324 struct obd_export *ecl_exp;
2328 static void ldlm_cancel_lock_for_export(struct obd_export *exp,
2329 struct ldlm_lock *lock,
2330 struct export_cl_data *ecl)
2332 struct ldlm_resource *res;
2334 res = ldlm_resource_getref(lock->l_resource);
2336 ldlm_res_lvbo_update(res, NULL, 1);
2337 ldlm_lock_cancel(lock);
2338 if (!exp->exp_obd->obd_stopping)
2339 ldlm_reprocess_all(res);
2340 ldlm_resource_putref(res);
2343 if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
2344 CDEBUG(D_INFO, "Export %p, %d locks cancelled.\n",
2345 exp, ecl->ecl_loop);
2350 * Iterator function for ldlm_export_cancel_locks.
2351 * Cancels passed locks.
2354 ldlm_cancel_locks_for_export_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd,
2355 struct hlist_node *hnode, void *data)
2358 struct export_cl_data *ecl = (struct export_cl_data *)data;
2359 struct obd_export *exp = ecl->ecl_exp;
2360 struct ldlm_lock *lock = cfs_hash_object(hs, hnode);
2362 LDLM_LOCK_GET(lock);
2363 ldlm_cancel_lock_for_export(exp, lock, ecl);
2364 LDLM_LOCK_RELEASE(lock);
2370 * Cancel all blocked locks for given export.
2372 * Typically called on client disconnection/eviction
2374 int ldlm_export_cancel_blocked_locks(struct obd_export *exp)
2376 struct export_cl_data ecl = {
2381 while (!list_empty(&exp->exp_bl_list)) {
2382 struct ldlm_lock *lock;
2384 spin_lock_bh(&exp->exp_bl_list_lock);
2385 if (!list_empty(&exp->exp_bl_list)) {
2386 lock = list_entry(exp->exp_bl_list.next,
2387 struct ldlm_lock, l_exp_list);
2388 LDLM_LOCK_GET(lock);
2389 list_del_init(&lock->l_exp_list);
2393 spin_unlock_bh(&exp->exp_bl_list_lock);
2398 ldlm_cancel_lock_for_export(exp, lock, &ecl);
2399 LDLM_LOCK_RELEASE(lock);
2402 CDEBUG(D_DLMTRACE, "Export %p, canceled %d locks, "
2403 "left on hash table %d.\n", exp, ecl.ecl_loop,
2404 atomic_read(&exp->exp_lock_hash->hs_count));
2406 return ecl.ecl_loop;
2410 * Cancel all locks for given export.
2412 * Typically called after client disconnection/eviction
2414 int ldlm_export_cancel_locks(struct obd_export *exp)
2416 struct export_cl_data ecl = {
2421 cfs_hash_for_each_empty(exp->exp_lock_hash,
2422 ldlm_cancel_locks_for_export_cb, &ecl);
2424 CDEBUG(D_DLMTRACE, "Export %p, canceled %d locks, "
2425 "left on hash table %d.\n", exp, ecl.ecl_loop,
2426 atomic_read(&exp->exp_lock_hash->hs_count));
2428 return ecl.ecl_loop;
2432 * Downgrade an exclusive lock.
2434 * A fast variant of ldlm_lock_convert for convertion of exclusive
2435 * locks. The convertion is always successful.
2436 * Used by Commit on Sharing (COS) code.
2438 * \param lock A lock to convert
2439 * \param new_mode new lock mode
2441 void ldlm_lock_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode)
2445 LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
2446 LASSERT(new_mode == LCK_COS);
2448 lock_res_and_lock(lock);
2449 ldlm_resource_unlink_lock(lock);
2451 * Remove the lock from pool as it will be added again in
2452 * ldlm_grant_lock() called below.
2454 ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
2456 lock->l_req_mode = new_mode;
2457 ldlm_grant_lock(lock, NULL);
2458 unlock_res_and_lock(lock);
2459 ldlm_reprocess_all(lock->l_resource);
2463 EXPORT_SYMBOL(ldlm_lock_downgrade);
2466 * Attempt to convert already granted lock to a different mode.
2468 * While lock conversion is not currently used, future client-side
2469 * optimizations could take advantage of it to avoid discarding cached
2472 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock,
2473 enum ldlm_mode new_mode, __u32 *flags)
2475 struct list_head rpc_list;
2476 struct ldlm_resource *res;
2477 struct ldlm_namespace *ns;
2479 #ifdef HAVE_SERVER_SUPPORT
2481 struct sl_insert_point prev;
2483 struct ldlm_interval *node;
2486 INIT_LIST_HEAD(&rpc_list);
2487 /* Just return if mode is unchanged. */
2488 if (new_mode == lock->l_granted_mode) {
2489 *flags |= LDLM_FL_BLOCK_GRANTED;
2490 RETURN(lock->l_resource);
2493 /* I can't check the type of lock here because the bitlock of lock
2494 * is not held here, so do the allocation blindly. -jay */
2495 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
2496 if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */
2499 LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
2500 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2502 lock_res_and_lock(lock);
2504 res = lock->l_resource;
2505 ns = ldlm_res_to_ns(res);
2507 #ifdef HAVE_SERVER_SUPPORT
2508 old_mode = lock->l_req_mode;
2510 lock->l_req_mode = new_mode;
2511 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2512 #ifdef HAVE_SERVER_SUPPORT
2513 /* remember the lock position where the lock might be
2514 * added back to the granted list later and also
2515 * remember the join mode for skiplist fixing. */
2516 prev.res_link = lock->l_res_link.prev;
2517 prev.mode_link = lock->l_sl_mode.prev;
2518 prev.policy_link = lock->l_sl_policy.prev;
2520 ldlm_resource_unlink_lock(lock);
2522 ldlm_resource_unlink_lock(lock);
2523 if (res->lr_type == LDLM_EXTENT) {
2524 /* FIXME: ugly code, I have to attach the lock to a
2525 * interval node again since perhaps it will be granted
2527 INIT_LIST_HEAD(&node->li_group);
2528 ldlm_interval_attach(node, lock);
2534 * Remove old lock from the pool before adding the lock with new
2535 * mode below in ->policy()
2537 ldlm_pool_del(&ns->ns_pool, lock);
2539 /* If this is a local resource, put it on the appropriate list. */
2540 if (ns_is_client(ldlm_res_to_ns(res))) {
2541 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2542 ldlm_resource_add_lock(res, &res->lr_converting, lock);
2544 /* This should never happen, because of the way the
2545 * server handles conversions. */
2546 LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2550 ldlm_grant_lock(lock, &rpc_list);
2552 /* FIXME: completion handling not with lr_lock held ! */
2553 if (lock->l_completion_ast)
2554 lock->l_completion_ast(lock, 0, NULL);
2556 #ifdef HAVE_SERVER_SUPPORT
2559 enum ldlm_error err;
2561 ldlm_processing_policy policy;
2563 policy = ldlm_processing_policy_table[res->lr_type];
2564 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2565 if (rc == LDLM_ITER_STOP) {
2566 lock->l_req_mode = old_mode;
2567 if (res->lr_type == LDLM_EXTENT)
2568 ldlm_extent_add_lock(res, lock);
2570 ldlm_granted_list_add_lock(lock, &prev);
2574 *flags |= LDLM_FL_BLOCK_GRANTED;
2580 CERROR("This is client-side-only module, cannot handle "
2581 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2585 unlock_res_and_lock(lock);
2588 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2590 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2595 * Print lock with lock handle \a lockh description into debug log.
2597 * Used when printing all locks on a resource for debug purposes.
2599 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh)
2601 struct ldlm_lock *lock;
2603 if (!((libcfs_debug | D_ERROR) & level))
2606 lock = ldlm_handle2lock(lockh);
2610 LDLM_DEBUG_LIMIT(level, lock, "###");
2612 LDLM_LOCK_PUT(lock);
2614 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2617 * Print lock information with custom message into debug log.
2620 void _ldlm_lock_debug(struct ldlm_lock *lock,
2621 struct libcfs_debug_msg_data *msgdata,
2622 const char *fmt, ...)
2625 struct obd_export *exp = lock->l_export;
2626 struct ldlm_resource *resource = lock->l_resource;
2627 char *nid = "local";
2629 va_start(args, fmt);
2631 if (exp && exp->exp_connection) {
2632 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2633 } else if (exp && exp->exp_obd != NULL) {
2634 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2635 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2638 if (resource == NULL) {
2639 libcfs_debug_vmsg2(msgdata, fmt, args,
2640 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2641 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2642 "remote: "LPX64" expref: %d pid: %u timeout: %lu "
2645 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2646 lock->l_readers, lock->l_writers,
2647 ldlm_lockname[lock->l_granted_mode],
2648 ldlm_lockname[lock->l_req_mode],
2649 lock->l_flags, nid, lock->l_remote_handle.cookie,
2650 exp ? atomic_read(&exp->exp_refcount) : -99,
2651 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2656 switch (resource->lr_type) {
2658 libcfs_debug_vmsg2(msgdata, fmt, args,
2659 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2660 "res: "DLDLMRES" rrc: %d type: %s ["LPU64"->"LPU64"] "
2661 "(req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote: "
2662 LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2663 ldlm_lock_to_ns_name(lock), lock,
2664 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2665 lock->l_readers, lock->l_writers,
2666 ldlm_lockname[lock->l_granted_mode],
2667 ldlm_lockname[lock->l_req_mode],
2669 atomic_read(&resource->lr_refcount),
2670 ldlm_typename[resource->lr_type],
2671 lock->l_policy_data.l_extent.start,
2672 lock->l_policy_data.l_extent.end,
2673 lock->l_req_extent.start, lock->l_req_extent.end,
2674 lock->l_flags, nid, lock->l_remote_handle.cookie,
2675 exp ? atomic_read(&exp->exp_refcount) : -99,
2676 lock->l_pid, lock->l_callback_timeout,
2681 libcfs_debug_vmsg2(msgdata, fmt, args,
2682 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2683 "res: "DLDLMRES" rrc: %d type: %s pid: %d "
2684 "["LPU64"->"LPU64"] flags: "LPX64" nid: %s "
2685 "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2686 ldlm_lock_to_ns_name(lock), lock,
2687 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2688 lock->l_readers, lock->l_writers,
2689 ldlm_lockname[lock->l_granted_mode],
2690 ldlm_lockname[lock->l_req_mode],
2692 atomic_read(&resource->lr_refcount),
2693 ldlm_typename[resource->lr_type],
2694 lock->l_policy_data.l_flock.pid,
2695 lock->l_policy_data.l_flock.start,
2696 lock->l_policy_data.l_flock.end,
2697 lock->l_flags, nid, lock->l_remote_handle.cookie,
2698 exp ? atomic_read(&exp->exp_refcount) : -99,
2699 lock->l_pid, lock->l_callback_timeout);
2703 libcfs_debug_vmsg2(msgdata, fmt, args,
2704 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2705 "res: "DLDLMRES" bits "LPX64" rrc: %d type: %s "
2706 "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2707 "pid: %u timeout: %lu lvb_type: %d\n",
2708 ldlm_lock_to_ns_name(lock),
2709 lock, lock->l_handle.h_cookie,
2710 atomic_read(&lock->l_refc),
2711 lock->l_readers, lock->l_writers,
2712 ldlm_lockname[lock->l_granted_mode],
2713 ldlm_lockname[lock->l_req_mode],
2715 lock->l_policy_data.l_inodebits.bits,
2716 atomic_read(&resource->lr_refcount),
2717 ldlm_typename[resource->lr_type],
2718 lock->l_flags, nid, lock->l_remote_handle.cookie,
2719 exp ? atomic_read(&exp->exp_refcount) : -99,
2720 lock->l_pid, lock->l_callback_timeout,
2725 libcfs_debug_vmsg2(msgdata, fmt, args,
2726 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2727 "res: "DLDLMRES" rrc: %d type: %s flags: "LPX64" "
2728 "nid: %s remote: "LPX64" expref: %d pid: %u "
2729 "timeout: %lu lvb_type: %d\n",
2730 ldlm_lock_to_ns_name(lock),
2731 lock, lock->l_handle.h_cookie,
2732 atomic_read(&lock->l_refc),
2733 lock->l_readers, lock->l_writers,
2734 ldlm_lockname[lock->l_granted_mode],
2735 ldlm_lockname[lock->l_req_mode],
2737 atomic_read(&resource->lr_refcount),
2738 ldlm_typename[resource->lr_type],
2739 lock->l_flags, nid, lock->l_remote_handle.cookie,
2740 exp ? atomic_read(&exp->exp_refcount) : -99,
2741 lock->l_pid, lock->l_callback_timeout,
2747 EXPORT_SYMBOL(_ldlm_lock_debug);