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, 2013, 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
45 # include <libcfs/libcfs.h>
46 # include <linux/lustre_intent.h>
48 # include <liblustre.h>
51 #include <obd_class.h>
52 #include "ldlm_internal.h"
55 char *ldlm_lockname[] = {
63 [LCK_GROUP] = "GROUP",
66 EXPORT_SYMBOL(ldlm_lockname);
68 char *ldlm_typename[] = {
70 [LDLM_EXTENT] = "EXT",
74 EXPORT_SYMBOL(ldlm_typename);
76 static ldlm_policy_wire_to_local_t ldlm_policy_wire18_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_wire18_to_local,
80 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local,
83 static ldlm_policy_wire_to_local_t ldlm_policy_wire21_to_local[] = {
84 [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_wire_to_local,
85 [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_wire_to_local,
86 [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_wire21_to_local,
87 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local,
90 static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire[] = {
91 [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_local_to_wire,
92 [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_local_to_wire,
93 [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_local_to_wire,
94 [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_local_to_wire,
98 * Converts lock policy from local format to on the wire lock_desc format
100 void ldlm_convert_policy_to_wire(ldlm_type_t type,
101 const ldlm_policy_data_t *lpolicy,
102 ldlm_wire_policy_data_t *wpolicy)
104 ldlm_policy_local_to_wire_t convert;
106 convert = ldlm_policy_local_to_wire[type - LDLM_MIN_TYPE];
108 convert(lpolicy, wpolicy);
112 * Converts lock policy from on the wire lock_desc format to local format
114 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
115 const ldlm_wire_policy_data_t *wpolicy,
116 ldlm_policy_data_t *lpolicy)
118 ldlm_policy_wire_to_local_t convert;
121 /** some badness for 2.0.0 clients, but 2.0.0 isn't supported */
122 new_client = (exp_connect_flags(exp) & OBD_CONNECT_FULL20) != 0;
124 convert = ldlm_policy_wire21_to_local[type - LDLM_MIN_TYPE];
126 convert = ldlm_policy_wire18_to_local[type - LDLM_MIN_TYPE];
128 convert(wpolicy, lpolicy);
131 char *ldlm_it2str(int it)
138 case (IT_OPEN | IT_CREAT):
153 CERROR("Unknown intent %d\n", it);
157 EXPORT_SYMBOL(ldlm_it2str);
159 extern struct kmem_cache *ldlm_lock_slab;
161 #ifdef HAVE_SERVER_SUPPORT
162 static ldlm_processing_policy ldlm_processing_policy_table[] = {
163 [LDLM_PLAIN] = ldlm_process_plain_lock,
164 [LDLM_EXTENT] = ldlm_process_extent_lock,
166 [LDLM_FLOCK] = ldlm_process_flock_lock,
168 [LDLM_IBITS] = ldlm_process_inodebits_lock,
171 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
173 return ldlm_processing_policy_table[res->lr_type];
175 EXPORT_SYMBOL(ldlm_get_processing_policy);
176 #endif /* HAVE_SERVER_SUPPORT */
178 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
182 EXPORT_SYMBOL(ldlm_register_intent);
185 * REFCOUNTED LOCK OBJECTS
190 * Get a reference on a lock.
192 * Lock refcounts, during creation:
193 * - one special one for allocation, dec'd only once in destroy
194 * - one for being a lock that's in-use
195 * - one for the addref associated with a new lock
197 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
199 atomic_inc(&lock->l_refc);
202 EXPORT_SYMBOL(ldlm_lock_get);
205 * Release lock reference.
207 * Also frees the lock if it was last reference.
209 void ldlm_lock_put(struct ldlm_lock *lock)
213 LASSERT(lock->l_resource != LP_POISON);
214 LASSERT(atomic_read(&lock->l_refc) > 0);
215 if (atomic_dec_and_test(&lock->l_refc)) {
216 struct ldlm_resource *res;
219 "final lock_put on destroyed lock, freeing it.");
221 res = lock->l_resource;
222 LASSERT(ldlm_is_destroyed(lock));
223 LASSERT(cfs_list_empty(&lock->l_res_link));
224 LASSERT(cfs_list_empty(&lock->l_pending_chain));
226 lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats,
228 lu_ref_del(&res->lr_reference, "lock", lock);
229 ldlm_resource_putref(res);
230 lock->l_resource = NULL;
231 if (lock->l_export) {
232 class_export_lock_put(lock->l_export, lock);
233 lock->l_export = NULL;
236 if (lock->l_lvb_data != NULL)
237 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
239 ldlm_interval_free(ldlm_interval_detach(lock));
240 lu_ref_fini(&lock->l_reference);
241 OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle);
246 EXPORT_SYMBOL(ldlm_lock_put);
249 * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked.
251 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
254 if (!cfs_list_empty(&lock->l_lru)) {
255 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
257 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
258 cfs_list_del_init(&lock->l_lru);
259 LASSERT(ns->ns_nr_unused > 0);
267 * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first.
269 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
271 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
275 if (ldlm_is_ns_srv(lock)) {
276 LASSERT(cfs_list_empty(&lock->l_lru));
280 spin_lock(&ns->ns_lock);
281 rc = ldlm_lock_remove_from_lru_nolock(lock);
282 spin_unlock(&ns->ns_lock);
288 * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked.
290 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
292 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
294 lock->l_last_used = cfs_time_current();
295 LASSERT(cfs_list_empty(&lock->l_lru));
296 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
297 cfs_list_add_tail(&lock->l_lru, &ns->ns_unused_list);
298 ldlm_clear_skipped(lock);
299 LASSERT(ns->ns_nr_unused >= 0);
304 * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks
307 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
309 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
312 spin_lock(&ns->ns_lock);
313 ldlm_lock_add_to_lru_nolock(lock);
314 spin_unlock(&ns->ns_lock);
319 * Moves LDLM lock \a lock that is already in namespace LRU to the tail of
320 * the LRU. Performs necessary LRU locking
322 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
324 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
327 if (ldlm_is_ns_srv(lock)) {
328 LASSERT(cfs_list_empty(&lock->l_lru));
333 spin_lock(&ns->ns_lock);
334 if (!cfs_list_empty(&lock->l_lru)) {
335 ldlm_lock_remove_from_lru_nolock(lock);
336 ldlm_lock_add_to_lru_nolock(lock);
338 spin_unlock(&ns->ns_lock);
343 * Helper to destroy a locked lock.
345 * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock
346 * Must be called with l_lock and lr_lock held.
348 * Does not actually free the lock data, but rather marks the lock as
349 * destroyed by setting l_destroyed field in the lock to 1. Destroys a
350 * handle->lock association too, so that the lock can no longer be found
351 * and removes the lock from LRU list. Actual lock freeing occurs when
352 * last lock reference goes away.
354 * Original comment (of some historical value):
355 * This used to have a 'strict' flag, which recovery would use to mark an
356 * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
357 * shall explain why it's gone: with the new hash table scheme, once you call
358 * ldlm_lock_destroy, you can never drop your final references on this lock.
359 * Because it's not in the hash table anymore. -phil
361 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
365 if (lock->l_readers || lock->l_writers) {
366 LDLM_ERROR(lock, "lock still has references");
370 if (!cfs_list_empty(&lock->l_res_link)) {
371 LDLM_ERROR(lock, "lock still on resource");
375 if (ldlm_is_destroyed(lock)) {
376 LASSERT(cfs_list_empty(&lock->l_lru));
380 ldlm_set_destroyed(lock);
382 if (lock->l_export && lock->l_export->exp_lock_hash) {
383 /* NB: it's safe to call cfs_hash_del() even lock isn't
384 * in exp_lock_hash. */
385 /* In the function below, .hs_keycmp resolves to
386 * ldlm_export_lock_keycmp() */
387 /* coverity[overrun-buffer-val] */
388 cfs_hash_del(lock->l_export->exp_lock_hash,
389 &lock->l_remote_handle, &lock->l_exp_hash);
392 ldlm_lock_remove_from_lru(lock);
393 class_handle_unhash(&lock->l_handle);
396 /* Wake anyone waiting for this lock */
397 /* FIXME: I should probably add yet another flag, instead of using
398 * l_export to only call this on clients */
400 class_export_put(lock->l_export);
401 lock->l_export = NULL;
402 if (lock->l_export && lock->l_completion_ast)
403 lock->l_completion_ast(lock, 0);
410 * Destroys a LDLM lock \a lock. Performs necessary locking first.
412 void ldlm_lock_destroy(struct ldlm_lock *lock)
416 lock_res_and_lock(lock);
417 first = ldlm_lock_destroy_internal(lock);
418 unlock_res_and_lock(lock);
420 /* drop reference from hashtable only for first destroy */
422 lu_ref_del(&lock->l_reference, "hash", lock);
423 LDLM_LOCK_RELEASE(lock);
429 * Destroys a LDLM lock \a lock that is already locked.
431 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
435 first = ldlm_lock_destroy_internal(lock);
436 /* drop reference from hashtable only for first destroy */
438 lu_ref_del(&lock->l_reference, "hash", lock);
439 LDLM_LOCK_RELEASE(lock);
444 /* this is called by portals_handle2object with the handle lock taken */
445 static void lock_handle_addref(void *lock)
447 LDLM_LOCK_GET((struct ldlm_lock *)lock);
450 static void lock_handle_free(void *lock, int size)
452 LASSERT(size == sizeof(struct ldlm_lock));
453 OBD_SLAB_FREE(lock, ldlm_lock_slab, size);
456 struct portals_handle_ops lock_handle_ops = {
457 .hop_addref = lock_handle_addref,
458 .hop_free = lock_handle_free,
463 * Allocate and initialize new lock structure.
465 * usage: pass in a resource on which you have done ldlm_resource_get
466 * new lock will take over the refcount.
467 * returns: lock with refcount 2 - one for current caller and one for remote
469 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
471 struct ldlm_lock *lock;
474 if (resource == NULL)
477 OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, GFP_NOFS);
481 spin_lock_init(&lock->l_lock);
482 lock->l_resource = resource;
483 lu_ref_add(&resource->lr_reference, "lock", lock);
485 atomic_set(&lock->l_refc, 2);
486 CFS_INIT_LIST_HEAD(&lock->l_res_link);
487 CFS_INIT_LIST_HEAD(&lock->l_lru);
488 CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
489 CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
490 CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
491 CFS_INIT_LIST_HEAD(&lock->l_rk_ast);
492 init_waitqueue_head(&lock->l_waitq);
493 lock->l_blocking_lock = NULL;
494 CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
495 CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
496 CFS_INIT_HLIST_NODE(&lock->l_exp_hash);
497 CFS_INIT_HLIST_NODE(&lock->l_exp_flock_hash);
499 lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats,
501 CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
502 class_handle_hash(&lock->l_handle, &lock_handle_ops);
504 lu_ref_init(&lock->l_reference);
505 lu_ref_add(&lock->l_reference, "hash", lock);
506 lock->l_callback_timeout = 0;
508 #if LUSTRE_TRACKS_LOCK_EXP_REFS
509 CFS_INIT_LIST_HEAD(&lock->l_exp_refs_link);
510 lock->l_exp_refs_nr = 0;
511 lock->l_exp_refs_target = NULL;
513 CFS_INIT_LIST_HEAD(&lock->l_exp_list);
519 * Moves LDLM lock \a lock to another resource.
520 * This is used on client when server returns some other lock than requested
521 * (typically as a result of intent operation)
523 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
524 const struct ldlm_res_id *new_resid)
526 struct ldlm_resource *oldres = lock->l_resource;
527 struct ldlm_resource *newres;
531 LASSERT(ns_is_client(ns));
533 lock_res_and_lock(lock);
534 if (memcmp(new_resid, &lock->l_resource->lr_name,
535 sizeof(lock->l_resource->lr_name)) == 0) {
537 unlock_res_and_lock(lock);
541 LASSERT(new_resid->name[0] != 0);
543 /* This function assumes that the lock isn't on any lists */
544 LASSERT(cfs_list_empty(&lock->l_res_link));
546 type = oldres->lr_type;
547 unlock_res_and_lock(lock);
549 newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
553 lu_ref_add(&newres->lr_reference, "lock", lock);
555 * To flip the lock from the old to the new resource, lock, oldres and
556 * newres have to be locked. Resource spin-locks are nested within
557 * lock->l_lock, and are taken in the memory address order to avoid
560 spin_lock(&lock->l_lock);
561 oldres = lock->l_resource;
562 if (oldres < newres) {
564 lock_res_nested(newres, LRT_NEW);
567 lock_res_nested(oldres, LRT_NEW);
569 LASSERT(memcmp(new_resid, &oldres->lr_name,
570 sizeof oldres->lr_name) != 0);
571 lock->l_resource = newres;
573 unlock_res_and_lock(lock);
575 /* ...and the flowers are still standing! */
576 lu_ref_del(&oldres->lr_reference, "lock", lock);
577 ldlm_resource_putref(oldres);
581 EXPORT_SYMBOL(ldlm_lock_change_resource);
583 /** \defgroup ldlm_handles LDLM HANDLES
584 * Ways to get hold of locks without any addresses.
589 * Fills in handle for LDLM lock \a lock into supplied \a lockh
590 * Does not take any references.
592 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
594 lockh->cookie = lock->l_handle.h_cookie;
596 EXPORT_SYMBOL(ldlm_lock2handle);
599 * Obtain a lock reference by handle.
601 * if \a flags: atomically get the lock and set the flags.
602 * Return NULL if flag already set
604 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
607 struct ldlm_lock *lock;
612 lock = class_handle2object(handle->cookie, NULL);
616 /* It's unlikely but possible that someone marked the lock as
617 * destroyed after we did handle2object on it */
618 if ((flags == 0) && !ldlm_is_destroyed(lock)) {
619 lu_ref_add(&lock->l_reference, "handle", current);
623 lock_res_and_lock(lock);
625 LASSERT(lock->l_resource != NULL);
627 lu_ref_add_atomic(&lock->l_reference, "handle", current);
628 if (unlikely(ldlm_is_destroyed(lock))) {
629 unlock_res_and_lock(lock);
630 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
635 /* If we're setting flags, make sure none of them are already set. */
637 if ((lock->l_flags & flags) != 0) {
638 unlock_res_and_lock(lock);
643 lock->l_flags |= flags;
646 unlock_res_and_lock(lock);
649 EXPORT_SYMBOL(__ldlm_handle2lock);
650 /** @} ldlm_handles */
653 * Fill in "on the wire" representation for given LDLM lock into supplied
654 * lock descriptor \a desc structure.
656 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
658 struct obd_export *exp = lock->l_export ?: lock->l_conn_export;
660 /* INODEBITS_INTEROP: If the other side does not support
661 * inodebits, reply with a plain lock descriptor. */
662 if ((lock->l_resource->lr_type == LDLM_IBITS) &&
663 (exp && !(exp_connect_flags(exp) & OBD_CONNECT_IBITS))) {
664 /* Make sure all the right bits are set in this lock we
665 are going to pass to client */
666 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
667 (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
668 MDS_INODELOCK_LAYOUT),
669 "Inappropriate inode lock bits during "
670 "conversion " LPU64 "\n",
671 lock->l_policy_data.l_inodebits.bits);
673 ldlm_res2desc(lock->l_resource, &desc->l_resource);
674 desc->l_resource.lr_type = LDLM_PLAIN;
676 /* Convert "new" lock mode to something old client can
678 if ((lock->l_req_mode == LCK_CR) ||
679 (lock->l_req_mode == LCK_CW))
680 desc->l_req_mode = LCK_PR;
682 desc->l_req_mode = lock->l_req_mode;
683 if ((lock->l_granted_mode == LCK_CR) ||
684 (lock->l_granted_mode == LCK_CW)) {
685 desc->l_granted_mode = LCK_PR;
687 /* We never grant PW/EX locks to clients */
688 LASSERT((lock->l_granted_mode != LCK_PW) &&
689 (lock->l_granted_mode != LCK_EX));
690 desc->l_granted_mode = lock->l_granted_mode;
693 /* We do not copy policy here, because there is no
694 policy for plain locks */
696 ldlm_res2desc(lock->l_resource, &desc->l_resource);
697 desc->l_req_mode = lock->l_req_mode;
698 desc->l_granted_mode = lock->l_granted_mode;
699 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
700 &lock->l_policy_data,
701 &desc->l_policy_data);
704 EXPORT_SYMBOL(ldlm_lock2desc);
707 * Add a lock to list of conflicting locks to send AST to.
709 * Only add if we have not sent a blocking AST to the lock yet.
711 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
712 cfs_list_t *work_list)
714 if (!ldlm_is_ast_sent(lock)) {
715 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
716 ldlm_set_ast_sent(lock);
717 /* If the enqueuing client said so, tell the AST recipient to
718 * discard dirty data, rather than writing back. */
719 if (ldlm_is_ast_discard_data(new))
720 ldlm_set_discard_data(lock);
721 LASSERT(cfs_list_empty(&lock->l_bl_ast));
722 cfs_list_add(&lock->l_bl_ast, work_list);
724 LASSERT(lock->l_blocking_lock == NULL);
725 lock->l_blocking_lock = LDLM_LOCK_GET(new);
730 * Add a lock to list of just granted locks to send completion AST to.
732 void ldlm_add_cp_work_item(struct ldlm_lock *lock, cfs_list_t *work_list)
734 if (!ldlm_is_cp_reqd(lock)) {
735 ldlm_set_cp_reqd(lock);
736 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
737 LASSERT(cfs_list_empty(&lock->l_cp_ast));
738 cfs_list_add(&lock->l_cp_ast, work_list);
744 * Aggregator function to add AST work items into a list. Determines
745 * what sort of an AST work needs to be done and calls the proper
747 * Must be called with lr_lock held.
749 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
750 cfs_list_t *work_list)
753 check_res_locked(lock->l_resource);
755 ldlm_add_bl_work_item(lock, new, work_list);
757 ldlm_add_cp_work_item(lock, work_list);
762 * Add specified reader/writer reference to LDLM lock with handle \a lockh.
763 * r/w reference type is determined by \a mode
764 * Calls ldlm_lock_addref_internal.
766 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
768 struct ldlm_lock *lock;
770 lock = ldlm_handle2lock(lockh);
771 LASSERT(lock != NULL);
772 ldlm_lock_addref_internal(lock, mode);
775 EXPORT_SYMBOL(ldlm_lock_addref);
779 * Add specified reader/writer reference to LDLM lock \a lock.
780 * r/w reference type is determined by \a mode
781 * Removes lock from LRU if it is there.
782 * Assumes the LDLM lock is already locked.
784 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
786 ldlm_lock_remove_from_lru(lock);
787 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
789 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
791 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
793 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
796 lu_ref_add_atomic(&lock->l_reference, "user", lock);
797 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
801 * Attempts to add reader/writer reference to a lock with handle \a lockh, and
802 * fails if lock is already LDLM_FL_CBPENDING or destroyed.
804 * \retval 0 success, lock was addref-ed
806 * \retval -EAGAIN lock is being canceled.
808 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
810 struct ldlm_lock *lock;
814 lock = ldlm_handle2lock(lockh);
816 lock_res_and_lock(lock);
817 if (lock->l_readers != 0 || lock->l_writers != 0 ||
818 !ldlm_is_cbpending(lock)) {
819 ldlm_lock_addref_internal_nolock(lock, mode);
822 unlock_res_and_lock(lock);
827 EXPORT_SYMBOL(ldlm_lock_addref_try);
830 * Add specified reader/writer reference to LDLM lock \a lock.
831 * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work.
832 * Only called for local locks.
834 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
836 lock_res_and_lock(lock);
837 ldlm_lock_addref_internal_nolock(lock, mode);
838 unlock_res_and_lock(lock);
842 * Removes reader/writer reference for LDLM lock \a lock.
843 * Assumes LDLM lock is already locked.
844 * only called in ldlm_flock_destroy and for local locks.
845 * Does NOT add lock to LRU if no r/w references left to accomodate flock locks
846 * that cannot be placed in LRU.
848 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
850 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
851 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
852 LASSERT(lock->l_readers > 0);
853 lu_ref_del(&lock->l_reference, "reader", lock);
856 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
857 LASSERT(lock->l_writers > 0);
858 lu_ref_del(&lock->l_reference, "writer", lock);
862 lu_ref_del(&lock->l_reference, "user", lock);
863 LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */
867 * Removes reader/writer reference for LDLM lock \a lock.
868 * Locks LDLM lock first.
869 * If the lock is determined to be client lock on a client and r/w refcount
870 * drops to zero and the lock is not blocked, the lock is added to LRU lock
872 * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called.
874 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
876 struct ldlm_namespace *ns;
879 lock_res_and_lock(lock);
881 ns = ldlm_lock_to_ns(lock);
883 ldlm_lock_decref_internal_nolock(lock, mode);
885 if (ldlm_is_local(lock) &&
886 !lock->l_readers && !lock->l_writers) {
887 /* If this is a local lock on a server namespace and this was
888 * the last reference, cancel the lock. */
889 CDEBUG(D_INFO, "forcing cancel of local lock\n");
890 ldlm_set_cbpending(lock);
893 if (!lock->l_readers && !lock->l_writers &&
894 ldlm_is_cbpending(lock)) {
895 /* If we received a blocked AST and this was the last reference,
896 * run the callback. */
897 if (ldlm_is_ns_srv(lock) && lock->l_export)
898 CERROR("FL_CBPENDING set on non-local lock--just a "
901 LDLM_DEBUG(lock, "final decref done on cbpending lock");
903 LDLM_LOCK_GET(lock); /* dropped by bl thread */
904 ldlm_lock_remove_from_lru(lock);
905 unlock_res_and_lock(lock);
907 if (ldlm_is_fail_loc(lock))
908 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
910 if (ldlm_is_atomic_cb(lock) ||
911 ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
912 ldlm_handle_bl_callback(ns, NULL, lock);
913 } else if (ns_is_client(ns) &&
914 !lock->l_readers && !lock->l_writers &&
915 !ldlm_is_no_lru(lock) &&
916 !ldlm_is_bl_ast(lock)) {
918 LDLM_DEBUG(lock, "add lock into lru list");
920 /* If this is a client-side namespace and this was the last
921 * reference, put it on the LRU. */
922 ldlm_lock_add_to_lru(lock);
923 unlock_res_and_lock(lock);
925 if (ldlm_is_fail_loc(lock))
926 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
928 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
929 * are not supported by the server, otherwise, it is done on
931 if (!exp_connect_cancelset(lock->l_conn_export) &&
932 !ns_connect_lru_resize(ns))
933 ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0);
935 LDLM_DEBUG(lock, "do not add lock into lru list");
936 unlock_res_and_lock(lock);
943 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
945 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
947 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
948 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
949 ldlm_lock_decref_internal(lock, mode);
952 EXPORT_SYMBOL(ldlm_lock_decref);
955 * Decrease reader/writer refcount for LDLM lock with handle
956 * \a lockh and mark it for subsequent cancellation once r/w refcount
957 * drops to zero instead of putting into LRU.
959 * Typical usage is for GROUP locks which we cannot allow to be cached.
961 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
963 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
966 LASSERT(lock != NULL);
968 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
969 lock_res_and_lock(lock);
970 ldlm_set_cbpending(lock);
971 unlock_res_and_lock(lock);
972 ldlm_lock_decref_internal(lock, mode);
975 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
977 struct sl_insert_point {
978 cfs_list_t *res_link;
979 cfs_list_t *mode_link;
980 cfs_list_t *policy_link;
984 * Finds a position to insert the new lock into granted lock list.
986 * Used for locks eligible for skiplist optimization.
989 * queue [input]: the granted list where search acts on;
990 * req [input]: the lock whose position to be located;
991 * prev [output]: positions within 3 lists to insert @req to
995 * - ldlm_grant_lock_with_skiplist
997 static void search_granted_lock(cfs_list_t *queue,
998 struct ldlm_lock *req,
999 struct sl_insert_point *prev)
1002 struct ldlm_lock *lock, *mode_end, *policy_end;
1005 cfs_list_for_each(tmp, queue) {
1006 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1008 mode_end = cfs_list_entry(lock->l_sl_mode.prev,
1009 struct ldlm_lock, l_sl_mode);
1011 if (lock->l_req_mode != req->l_req_mode) {
1012 /* jump to last lock of mode group */
1013 tmp = &mode_end->l_res_link;
1017 /* suitable mode group is found */
1018 if (lock->l_resource->lr_type == LDLM_PLAIN) {
1019 /* insert point is last lock of the mode group */
1020 prev->res_link = &mode_end->l_res_link;
1021 prev->mode_link = &mode_end->l_sl_mode;
1022 prev->policy_link = &req->l_sl_policy;
1025 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
1028 cfs_list_entry(lock->l_sl_policy.prev,
1032 if (lock->l_policy_data.l_inodebits.bits ==
1033 req->l_policy_data.l_inodebits.bits) {
1034 /* insert point is last lock of
1035 * the policy group */
1037 &policy_end->l_res_link;
1039 &policy_end->l_sl_mode;
1041 &policy_end->l_sl_policy;
1046 if (policy_end == mode_end)
1047 /* done with mode group */
1050 /* go to next policy group within mode group */
1051 tmp = policy_end->l_res_link.next;
1052 lock = cfs_list_entry(tmp, struct ldlm_lock,
1054 } /* loop over policy groups within the mode group */
1056 /* insert point is last lock of the mode group,
1057 * new policy group is started */
1058 prev->res_link = &mode_end->l_res_link;
1059 prev->mode_link = &mode_end->l_sl_mode;
1060 prev->policy_link = &req->l_sl_policy;
1064 LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
1069 /* insert point is last lock on the queue,
1070 * new mode group and new policy group are started */
1071 prev->res_link = queue->prev;
1072 prev->mode_link = &req->l_sl_mode;
1073 prev->policy_link = &req->l_sl_policy;
1079 * Add a lock into resource granted list after a position described by
1082 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
1083 struct sl_insert_point *prev)
1085 struct ldlm_resource *res = lock->l_resource;
1088 check_res_locked(res);
1090 ldlm_resource_dump(D_INFO, res);
1091 LDLM_DEBUG(lock, "About to add lock:");
1093 if (ldlm_is_destroyed(lock)) {
1094 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1098 LASSERT(cfs_list_empty(&lock->l_res_link));
1099 LASSERT(cfs_list_empty(&lock->l_sl_mode));
1100 LASSERT(cfs_list_empty(&lock->l_sl_policy));
1103 * lock->link == prev->link means lock is first starting the group.
1104 * Don't re-add to itself to suppress kernel warnings.
1106 if (&lock->l_res_link != prev->res_link)
1107 cfs_list_add(&lock->l_res_link, prev->res_link);
1108 if (&lock->l_sl_mode != prev->mode_link)
1109 cfs_list_add(&lock->l_sl_mode, prev->mode_link);
1110 if (&lock->l_sl_policy != prev->policy_link)
1111 cfs_list_add(&lock->l_sl_policy, prev->policy_link);
1117 * Add a lock to granted list on a resource maintaining skiplist
1120 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
1122 struct sl_insert_point prev;
1125 LASSERT(lock->l_req_mode == lock->l_granted_mode);
1127 search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
1128 ldlm_granted_list_add_lock(lock, &prev);
1133 * Perform lock granting bookkeeping.
1135 * Includes putting the lock into granted list and updating lock mode.
1137 * - ldlm_lock_enqueue
1138 * - ldlm_reprocess_queue
1139 * - ldlm_lock_convert
1141 * must be called with lr_lock held
1143 void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list)
1145 struct ldlm_resource *res = lock->l_resource;
1148 check_res_locked(res);
1150 lock->l_granted_mode = lock->l_req_mode;
1152 if (work_list && lock->l_completion_ast != NULL)
1153 ldlm_add_ast_work_item(lock, NULL, work_list);
1155 /* We should not add locks to granted list in the following cases:
1156 * - this is an UNLOCK but not a real lock;
1157 * - this is a TEST lock;
1158 * - this is a F_CANCELLK lock (async flock has req_mode == 0)
1159 * - this is a deadlock (flock cannot be granted) */
1160 if (lock->l_req_mode == 0 ||
1161 lock->l_req_mode == LCK_NL ||
1162 ldlm_is_test_lock(lock) ||
1163 ldlm_is_flock_deadlock(lock))
1166 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
1167 ldlm_grant_lock_with_skiplist(lock);
1168 else if (res->lr_type == LDLM_EXTENT)
1169 ldlm_extent_add_lock(res, lock);
1171 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1173 if (lock->l_granted_mode < res->lr_most_restr)
1174 res->lr_most_restr = lock->l_granted_mode;
1176 ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1181 * Search for a lock with given properties in a queue.
1183 * \retval a referenced lock or NULL. See the flag descriptions below, in the
1184 * comment above ldlm_lock_match
1186 static struct ldlm_lock *search_queue(cfs_list_t *queue,
1188 ldlm_policy_data_t *policy,
1189 struct ldlm_lock *old_lock,
1190 __u64 flags, int unref)
1192 struct ldlm_lock *lock;
1195 cfs_list_for_each(tmp, queue) {
1198 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1200 if (lock == old_lock)
1203 /* Check if this lock can be matched.
1204 * Used by LU-2919(exclusive open) for open lease lock */
1205 if (ldlm_is_excl(lock))
1208 /* llite sometimes wants to match locks that will be
1209 * canceled when their users drop, but we allow it to match
1210 * if it passes in CBPENDING and the lock still has users.
1211 * this is generally only going to be used by children
1212 * whose parents already hold a lock so forward progress
1213 * can still happen. */
1214 if (ldlm_is_cbpending(lock) &&
1215 !(flags & LDLM_FL_CBPENDING))
1217 if (!unref && ldlm_is_cbpending(lock) &&
1218 lock->l_readers == 0 && lock->l_writers == 0)
1221 if (!(lock->l_req_mode & *mode))
1223 match = lock->l_req_mode;
1225 if (lock->l_resource->lr_type == LDLM_EXTENT &&
1226 (lock->l_policy_data.l_extent.start >
1227 policy->l_extent.start ||
1228 lock->l_policy_data.l_extent.end < policy->l_extent.end))
1231 if (unlikely(match == LCK_GROUP) &&
1232 lock->l_resource->lr_type == LDLM_EXTENT &&
1233 lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
1236 /* We match if we have existing lock with same or wider set
1238 if (lock->l_resource->lr_type == LDLM_IBITS &&
1239 ((lock->l_policy_data.l_inodebits.bits &
1240 policy->l_inodebits.bits) !=
1241 policy->l_inodebits.bits))
1244 if (!unref && LDLM_HAVE_MASK(lock, GONE))
1247 if ((flags & LDLM_FL_LOCAL_ONLY) &&
1248 !ldlm_is_local(lock))
1251 if (flags & LDLM_FL_TEST_LOCK) {
1252 LDLM_LOCK_GET(lock);
1253 ldlm_lock_touch_in_lru(lock);
1255 ldlm_lock_addref_internal_nolock(lock, match);
1264 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1266 if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) {
1267 lock->l_flags |= LDLM_FL_FAIL_NOTIFIED;
1268 wake_up_all(&lock->l_waitq);
1271 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1273 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1275 lock_res_and_lock(lock);
1276 ldlm_lock_fail_match_locked(lock);
1277 unlock_res_and_lock(lock);
1279 EXPORT_SYMBOL(ldlm_lock_fail_match);
1282 * Mark lock as "matchable" by OST.
1284 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1286 * Assumes LDLM lock is already locked.
1288 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1290 ldlm_set_lvb_ready(lock);
1291 wake_up_all(&lock->l_waitq);
1293 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1296 * Mark lock as "matchable" by OST.
1297 * Locks the lock and then \see ldlm_lock_allow_match_locked
1299 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1301 lock_res_and_lock(lock);
1302 ldlm_lock_allow_match_locked(lock);
1303 unlock_res_and_lock(lock);
1305 EXPORT_SYMBOL(ldlm_lock_allow_match);
1308 * Attempt to find a lock with specified properties.
1310 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1313 * Can be called in two ways:
1315 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1316 * for a duplicate of.
1318 * Otherwise, all of the fields must be filled in, to match against.
1320 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1321 * server (ie, connh is NULL)
1322 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1323 * list will be considered
1324 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1325 * to be canceled can still be matched as long as they still have reader
1326 * or writer refernces
1327 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1328 * just tell us if we would have matched.
1330 * \retval 1 if it finds an already-existing lock that is compatible; in this
1331 * case, lockh is filled in with a addref()ed lock
1333 * We also check security context, and if that fails we simply return 0 (to
1334 * keep caller code unchanged), the context failure will be discovered by
1335 * caller sometime later.
1337 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1338 const struct ldlm_res_id *res_id, ldlm_type_t type,
1339 ldlm_policy_data_t *policy, ldlm_mode_t mode,
1340 struct lustre_handle *lockh, int unref)
1342 struct ldlm_resource *res;
1343 struct ldlm_lock *lock, *old_lock = NULL;
1348 old_lock = ldlm_handle2lock(lockh);
1351 ns = ldlm_lock_to_ns(old_lock);
1352 res_id = &old_lock->l_resource->lr_name;
1353 type = old_lock->l_resource->lr_type;
1354 mode = old_lock->l_req_mode;
1357 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1359 LASSERT(old_lock == NULL);
1363 LDLM_RESOURCE_ADDREF(res);
1366 lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
1370 if (flags & LDLM_FL_BLOCK_GRANTED)
1372 lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
1376 lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
1384 LDLM_RESOURCE_DELREF(res);
1385 ldlm_resource_putref(res);
1388 ldlm_lock2handle(lock, lockh);
1389 if ((flags & LDLM_FL_LVB_READY) &&
1390 (!ldlm_is_lvb_ready(lock))) {
1391 __u64 wait_flags = LDLM_FL_LVB_READY |
1392 LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED;
1393 struct l_wait_info lwi;
1394 if (lock->l_completion_ast) {
1395 int err = lock->l_completion_ast(lock,
1396 LDLM_FL_WAIT_NOREPROC,
1399 if (flags & LDLM_FL_TEST_LOCK)
1400 LDLM_LOCK_RELEASE(lock);
1402 ldlm_lock_decref_internal(lock,
1409 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1410 NULL, LWI_ON_SIGNAL_NOOP, NULL);
1412 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1413 l_wait_event(lock->l_waitq,
1414 lock->l_flags & wait_flags,
1416 if (!ldlm_is_lvb_ready(lock)) {
1417 if (flags & LDLM_FL_TEST_LOCK)
1418 LDLM_LOCK_RELEASE(lock);
1420 ldlm_lock_decref_internal(lock, mode);
1427 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1428 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1429 res_id->name[2] : policy->l_extent.start,
1430 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1431 res_id->name[3] : policy->l_extent.end);
1433 /* check user's security context */
1434 if (lock->l_conn_export &&
1435 sptlrpc_import_check_ctx(
1436 class_exp2cliimp(lock->l_conn_export))) {
1437 if (!(flags & LDLM_FL_TEST_LOCK))
1438 ldlm_lock_decref_internal(lock, mode);
1442 if (flags & LDLM_FL_TEST_LOCK)
1443 LDLM_LOCK_RELEASE(lock);
1445 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1446 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1447 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1448 type, mode, res_id->name[0], res_id->name[1],
1449 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1450 res_id->name[2] :policy->l_extent.start,
1451 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1452 res_id->name[3] : policy->l_extent.end);
1455 LDLM_LOCK_PUT(old_lock);
1457 return rc ? mode : 0;
1459 EXPORT_SYMBOL(ldlm_lock_match);
1461 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1464 struct ldlm_lock *lock;
1465 ldlm_mode_t mode = 0;
1468 lock = ldlm_handle2lock(lockh);
1470 lock_res_and_lock(lock);
1471 if (LDLM_HAVE_MASK(lock, GONE))
1474 if (ldlm_is_cbpending(lock) &&
1475 lock->l_readers == 0 && lock->l_writers == 0)
1479 *bits = lock->l_policy_data.l_inodebits.bits;
1480 mode = lock->l_granted_mode;
1481 ldlm_lock_addref_internal_nolock(lock, mode);
1488 unlock_res_and_lock(lock);
1489 LDLM_LOCK_PUT(lock);
1493 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1495 /** The caller must guarantee that the buffer is large enough. */
1496 int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill,
1497 enum req_location loc, void *data, int size)
1502 LASSERT(data != NULL);
1505 switch (lock->l_lvb_type) {
1507 if (size == sizeof(struct ost_lvb)) {
1508 if (loc == RCL_CLIENT)
1509 lvb = req_capsule_client_swab_get(pill,
1511 lustre_swab_ost_lvb);
1513 lvb = req_capsule_server_swab_get(pill,
1515 lustre_swab_ost_lvb);
1516 if (unlikely(lvb == NULL)) {
1517 LDLM_ERROR(lock, "no LVB");
1521 memcpy(data, lvb, size);
1522 } else if (size == sizeof(struct ost_lvb_v1)) {
1523 struct ost_lvb *olvb = data;
1525 if (loc == RCL_CLIENT)
1526 lvb = req_capsule_client_swab_get(pill,
1528 lustre_swab_ost_lvb_v1);
1530 lvb = req_capsule_server_sized_swab_get(pill,
1532 lustre_swab_ost_lvb_v1);
1533 if (unlikely(lvb == NULL)) {
1534 LDLM_ERROR(lock, "no LVB");
1538 memcpy(data, lvb, size);
1539 olvb->lvb_mtime_ns = 0;
1540 olvb->lvb_atime_ns = 0;
1541 olvb->lvb_ctime_ns = 0;
1543 LDLM_ERROR(lock, "Replied unexpected ost LVB size %d",
1549 if (size == sizeof(struct lquota_lvb)) {
1550 if (loc == RCL_CLIENT)
1551 lvb = req_capsule_client_swab_get(pill,
1553 lustre_swab_lquota_lvb);
1555 lvb = req_capsule_server_swab_get(pill,
1557 lustre_swab_lquota_lvb);
1558 if (unlikely(lvb == NULL)) {
1559 LDLM_ERROR(lock, "no LVB");
1563 memcpy(data, lvb, size);
1565 LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d",
1574 if (loc == RCL_CLIENT)
1575 lvb = req_capsule_client_get(pill, &RMF_DLM_LVB);
1577 lvb = req_capsule_server_get(pill, &RMF_DLM_LVB);
1578 if (unlikely(lvb == NULL)) {
1579 LDLM_ERROR(lock, "no LVB");
1583 memcpy(data, lvb, size);
1586 LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type);
1587 libcfs_debug_dumpstack(NULL);
1595 * Create and fill in new LDLM lock with specified properties.
1596 * Returns a referenced lock
1598 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1599 const struct ldlm_res_id *res_id,
1602 const struct ldlm_callback_suite *cbs,
1603 void *data, __u32 lvb_len,
1604 enum lvb_type lvb_type)
1606 struct ldlm_lock *lock;
1607 struct ldlm_resource *res;
1610 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1614 lock = ldlm_lock_new(res);
1619 lock->l_req_mode = mode;
1620 lock->l_ast_data = data;
1621 lock->l_pid = current_pid();
1622 if (ns_is_server(ns))
1623 ldlm_set_ns_srv(lock);
1625 lock->l_blocking_ast = cbs->lcs_blocking;
1626 lock->l_completion_ast = cbs->lcs_completion;
1627 lock->l_glimpse_ast = cbs->lcs_glimpse;
1630 lock->l_tree_node = NULL;
1631 /* if this is the extent lock, allocate the interval tree node */
1632 if (type == LDLM_EXTENT) {
1633 if (ldlm_interval_alloc(lock) == NULL)
1638 lock->l_lvb_len = lvb_len;
1639 OBD_ALLOC_LARGE(lock->l_lvb_data, lvb_len);
1640 if (lock->l_lvb_data == NULL)
1644 lock->l_lvb_type = lvb_type;
1645 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1651 ldlm_lock_destroy(lock);
1652 LDLM_LOCK_RELEASE(lock);
1657 * Enqueue (request) a lock.
1659 * Does not block. As a result of enqueue the lock would be put
1660 * into granted or waiting list.
1662 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1663 * set, skip all the enqueueing and delegate lock processing to intent policy
1666 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1667 struct ldlm_lock **lockp,
1668 void *cookie, __u64 *flags)
1670 struct ldlm_lock *lock = *lockp;
1671 struct ldlm_resource *res = lock->l_resource;
1672 int local = ns_is_client(ldlm_res_to_ns(res));
1673 #ifdef HAVE_SERVER_SUPPORT
1674 ldlm_processing_policy policy;
1676 ldlm_error_t rc = ELDLM_OK;
1677 struct ldlm_interval *node = NULL;
1680 lock->l_last_activity = cfs_time_current_sec();
1681 /* policies are not executed on the client or during replay */
1682 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1683 && !local && ns->ns_policy) {
1684 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1686 if (rc == ELDLM_LOCK_REPLACED) {
1687 /* The lock that was returned has already been granted,
1688 * and placed into lockp. If it's not the same as the
1689 * one we passed in, then destroy the old one and our
1690 * work here is done. */
1691 if (lock != *lockp) {
1692 ldlm_lock_destroy(lock);
1693 LDLM_LOCK_RELEASE(lock);
1695 *flags |= LDLM_FL_LOCK_CHANGED;
1697 } else if (rc != ELDLM_OK ||
1698 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1699 ldlm_lock_destroy(lock);
1704 /* For a replaying lock, it might be already in granted list. So
1705 * unlinking the lock will cause the interval node to be freed, we
1706 * have to allocate the interval node early otherwise we can't regrant
1707 * this lock in the future. - jay */
1708 if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1709 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
1711 lock_res_and_lock(lock);
1712 if (local && lock->l_req_mode == lock->l_granted_mode) {
1713 /* The server returned a blocked lock, but it was granted
1714 * before we got a chance to actually enqueue it. We don't
1715 * need to do anything else. */
1716 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1717 LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1718 GOTO(out, rc = ELDLM_OK);
1721 ldlm_resource_unlink_lock(lock);
1722 if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1724 ldlm_lock_destroy_nolock(lock);
1725 GOTO(out, rc = -ENOMEM);
1728 CFS_INIT_LIST_HEAD(&node->li_group);
1729 ldlm_interval_attach(node, lock);
1733 /* Some flags from the enqueue want to make it into the AST, via the
1734 * lock's l_flags. */
1735 if (*flags & LDLM_FL_AST_DISCARD_DATA)
1736 ldlm_set_ast_discard_data(lock);
1737 if (*flags & LDLM_FL_TEST_LOCK)
1738 ldlm_set_test_lock(lock);
1740 /* This distinction between local lock trees is very important; a client
1741 * namespace only has information about locks taken by that client, and
1742 * thus doesn't have enough information to decide for itself if it can
1743 * be granted (below). In this case, we do exactly what the server
1744 * tells us to do, as dictated by the 'flags'.
1746 * We do exactly the same thing during recovery, when the server is
1747 * more or less trusting the clients not to lie.
1749 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1750 * granted/converting queues. */
1752 if (*flags & LDLM_FL_BLOCK_CONV)
1753 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1754 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1755 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1757 ldlm_grant_lock(lock, NULL);
1758 GOTO(out, rc = ELDLM_OK);
1759 #ifdef HAVE_SERVER_SUPPORT
1760 } else if (*flags & LDLM_FL_REPLAY) {
1761 if (*flags & LDLM_FL_BLOCK_CONV) {
1762 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1763 GOTO(out, rc = ELDLM_OK);
1764 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1765 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1766 GOTO(out, rc = ELDLM_OK);
1767 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1768 ldlm_grant_lock(lock, NULL);
1769 GOTO(out, rc = ELDLM_OK);
1771 /* If no flags, fall through to normal enqueue path. */
1774 policy = ldlm_processing_policy_table[res->lr_type];
1775 policy(lock, flags, 1, &rc, NULL);
1779 CERROR("This is client-side-only module, cannot handle "
1780 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1786 unlock_res_and_lock(lock);
1788 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1792 #ifdef HAVE_SERVER_SUPPORT
1794 * Iterate through all waiting locks on a given resource queue and attempt to
1797 * Must be called with resource lock held.
1799 int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue,
1800 cfs_list_t *work_list)
1802 cfs_list_t *tmp, *pos;
1803 ldlm_processing_policy policy;
1805 int rc = LDLM_ITER_CONTINUE;
1809 check_res_locked(res);
1811 policy = ldlm_processing_policy_table[res->lr_type];
1814 cfs_list_for_each_safe(tmp, pos, queue) {
1815 struct ldlm_lock *pending;
1816 pending = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1818 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1821 rc = policy(pending, &flags, 0, &err, work_list);
1822 if (rc != LDLM_ITER_CONTINUE)
1831 * Process a call to blocking AST callback for a lock in ast_work list
1834 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1836 struct ldlm_cb_set_arg *arg = opaq;
1837 struct ldlm_lock_desc d;
1839 struct ldlm_lock *lock;
1842 if (cfs_list_empty(arg->list))
1845 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1847 /* nobody should touch l_bl_ast */
1848 lock_res_and_lock(lock);
1849 cfs_list_del_init(&lock->l_bl_ast);
1851 LASSERT(ldlm_is_ast_sent(lock));
1852 LASSERT(lock->l_bl_ast_run == 0);
1853 LASSERT(lock->l_blocking_lock);
1854 lock->l_bl_ast_run++;
1855 unlock_res_and_lock(lock);
1857 ldlm_lock2desc(lock->l_blocking_lock, &d);
1859 rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1860 LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1861 lock->l_blocking_lock = NULL;
1862 LDLM_LOCK_RELEASE(lock);
1868 * Process a call to completion AST callback for a lock in ast_work list
1871 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1873 struct ldlm_cb_set_arg *arg = opaq;
1875 struct ldlm_lock *lock;
1876 ldlm_completion_callback completion_callback;
1879 if (cfs_list_empty(arg->list))
1882 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1884 /* It's possible to receive a completion AST before we've set
1885 * the l_completion_ast pointer: either because the AST arrived
1886 * before the reply, or simply because there's a small race
1887 * window between receiving the reply and finishing the local
1888 * enqueue. (bug 842)
1890 * This can't happen with the blocking_ast, however, because we
1891 * will never call the local blocking_ast until we drop our
1892 * reader/writer reference, which we won't do until we get the
1893 * reply and finish enqueueing. */
1895 /* nobody should touch l_cp_ast */
1896 lock_res_and_lock(lock);
1897 cfs_list_del_init(&lock->l_cp_ast);
1898 LASSERT(ldlm_is_cp_reqd(lock));
1899 /* save l_completion_ast since it can be changed by
1900 * mds_intent_policy(), see bug 14225 */
1901 completion_callback = lock->l_completion_ast;
1902 ldlm_clear_cp_reqd(lock);
1903 unlock_res_and_lock(lock);
1905 if (completion_callback != NULL)
1906 rc = completion_callback(lock, 0, (void *)arg);
1907 LDLM_LOCK_RELEASE(lock);
1913 * Process a call to revocation AST callback for a lock in ast_work list
1916 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1918 struct ldlm_cb_set_arg *arg = opaq;
1919 struct ldlm_lock_desc desc;
1921 struct ldlm_lock *lock;
1924 if (cfs_list_empty(arg->list))
1927 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1928 cfs_list_del_init(&lock->l_rk_ast);
1930 /* the desc just pretend to exclusive */
1931 ldlm_lock2desc(lock, &desc);
1932 desc.l_req_mode = LCK_EX;
1933 desc.l_granted_mode = 0;
1935 rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1936 LDLM_LOCK_RELEASE(lock);
1942 * Process a call to glimpse AST callback for a lock in ast_work list
1944 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1946 struct ldlm_cb_set_arg *arg = opaq;
1947 struct ldlm_glimpse_work *gl_work;
1948 struct ldlm_lock *lock;
1952 if (cfs_list_empty(arg->list))
1955 gl_work = cfs_list_entry(arg->list->next, struct ldlm_glimpse_work,
1957 cfs_list_del_init(&gl_work->gl_list);
1959 lock = gl_work->gl_lock;
1961 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
1962 arg->gl_desc = gl_work->gl_desc;
1964 /* invoke the actual glimpse callback */
1965 if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
1968 LDLM_LOCK_RELEASE(lock);
1970 if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
1971 OBD_FREE_PTR(gl_work);
1977 * Process list of locks in need of ASTs being sent.
1979 * Used on server to send multiple ASTs together instead of sending one by
1982 int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list,
1983 ldlm_desc_ast_t ast_type)
1985 struct ldlm_cb_set_arg *arg;
1986 set_producer_func work_ast_lock;
1989 if (cfs_list_empty(rpc_list))
1996 atomic_set(&arg->restart, 0);
1997 arg->list = rpc_list;
2000 case LDLM_WORK_BL_AST:
2001 arg->type = LDLM_BL_CALLBACK;
2002 work_ast_lock = ldlm_work_bl_ast_lock;
2004 case LDLM_WORK_CP_AST:
2005 arg->type = LDLM_CP_CALLBACK;
2006 work_ast_lock = ldlm_work_cp_ast_lock;
2008 case LDLM_WORK_REVOKE_AST:
2009 arg->type = LDLM_BL_CALLBACK;
2010 work_ast_lock = ldlm_work_revoke_ast_lock;
2012 case LDLM_WORK_GL_AST:
2013 arg->type = LDLM_GL_CALLBACK;
2014 work_ast_lock = ldlm_work_gl_ast_lock;
2020 /* We create a ptlrpc request set with flow control extension.
2021 * This request set will use the work_ast_lock function to produce new
2022 * requests and will send a new request each time one completes in order
2023 * to keep the number of requests in flight to ns_max_parallel_ast */
2024 arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
2025 work_ast_lock, arg);
2026 if (arg->set == NULL)
2027 GOTO(out, rc = -ENOMEM);
2029 ptlrpc_set_wait(arg->set);
2030 ptlrpc_set_destroy(arg->set);
2032 rc = atomic_read(&arg->restart) ? -ERESTART : 0;
2039 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
2041 ldlm_reprocess_all(res);
2042 return LDLM_ITER_CONTINUE;
2045 static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2046 cfs_hlist_node_t *hnode, void *arg)
2048 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2051 rc = reprocess_one_queue(res, arg);
2053 return rc == LDLM_ITER_STOP;
2057 * Iterate through all resources on a namespace attempting to grant waiting
2060 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
2065 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2066 ldlm_reprocess_res, NULL);
2070 EXPORT_SYMBOL(ldlm_reprocess_all_ns);
2073 * Try to grant all waiting locks on a resource.
2075 * Calls ldlm_reprocess_queue on converting and waiting queues.
2077 * Typically called after some resource locks are cancelled to see
2078 * if anything could be granted as a result of the cancellation.
2080 void ldlm_reprocess_all(struct ldlm_resource *res)
2082 CFS_LIST_HEAD(rpc_list);
2084 #ifdef HAVE_SERVER_SUPPORT
2087 /* Local lock trees don't get reprocessed. */
2088 if (ns_is_client(ldlm_res_to_ns(res))) {
2095 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
2096 if (rc == LDLM_ITER_CONTINUE)
2097 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
2100 rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
2102 if (rc == -ERESTART) {
2103 LASSERT(cfs_list_empty(&rpc_list));
2108 if (!ns_is_client(ldlm_res_to_ns(res))) {
2109 CERROR("This is client-side-only module, cannot handle "
2110 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2116 EXPORT_SYMBOL(ldlm_reprocess_all);
2119 * Helper function to call blocking AST for LDLM lock \a lock in a
2120 * "cancelling" mode.
2122 void ldlm_cancel_callback(struct ldlm_lock *lock)
2124 check_res_locked(lock->l_resource);
2125 if (!ldlm_is_cancel(lock)) {
2126 ldlm_set_cancel(lock);
2127 if (lock->l_blocking_ast) {
2128 unlock_res_and_lock(lock);
2129 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
2131 lock_res_and_lock(lock);
2133 LDLM_DEBUG(lock, "no blocking ast");
2136 ldlm_set_bl_done(lock);
2140 * Remove skiplist-enabled LDLM lock \a req from granted list
2142 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
2144 if (req->l_resource->lr_type != LDLM_PLAIN &&
2145 req->l_resource->lr_type != LDLM_IBITS)
2148 cfs_list_del_init(&req->l_sl_policy);
2149 cfs_list_del_init(&req->l_sl_mode);
2153 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2155 void ldlm_lock_cancel(struct ldlm_lock *lock)
2157 struct ldlm_resource *res;
2158 struct ldlm_namespace *ns;
2161 lock_res_and_lock(lock);
2163 res = lock->l_resource;
2164 ns = ldlm_res_to_ns(res);
2166 /* Please do not, no matter how tempting, remove this LBUG without
2167 * talking to me first. -phik */
2168 if (lock->l_readers || lock->l_writers) {
2169 LDLM_ERROR(lock, "lock still has references");
2173 if (ldlm_is_waited(lock))
2174 ldlm_del_waiting_lock(lock);
2176 /* Releases cancel callback. */
2177 ldlm_cancel_callback(lock);
2179 /* Yes, second time, just in case it was added again while we were
2180 * running with no res lock in ldlm_cancel_callback */
2181 if (ldlm_is_waited(lock))
2182 ldlm_del_waiting_lock(lock);
2184 ldlm_resource_unlink_lock(lock);
2185 ldlm_lock_destroy_nolock(lock);
2187 if (lock->l_granted_mode == lock->l_req_mode)
2188 ldlm_pool_del(&ns->ns_pool, lock);
2190 /* Make sure we will not be called again for same lock what is possible
2191 * if not to zero out lock->l_granted_mode */
2192 lock->l_granted_mode = LCK_MINMODE;
2193 unlock_res_and_lock(lock);
2197 EXPORT_SYMBOL(ldlm_lock_cancel);
2200 * Set opaque data into the lock that only makes sense to upper layer.
2202 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
2204 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2209 if (lock->l_ast_data == NULL)
2210 lock->l_ast_data = data;
2211 if (lock->l_ast_data == data)
2213 LDLM_LOCK_PUT(lock);
2217 EXPORT_SYMBOL(ldlm_lock_set_data);
2219 struct export_cl_data {
2220 struct obd_export *ecl_exp;
2225 * Iterator function for ldlm_cancel_locks_for_export.
2226 * Cancels passed locks.
2228 int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2229 cfs_hlist_node_t *hnode, void *data)
2232 struct export_cl_data *ecl = (struct export_cl_data *)data;
2233 struct obd_export *exp = ecl->ecl_exp;
2234 struct ldlm_lock *lock = cfs_hash_object(hs, hnode);
2235 struct ldlm_resource *res;
2237 res = ldlm_resource_getref(lock->l_resource);
2238 LDLM_LOCK_GET(lock);
2240 LDLM_DEBUG(lock, "export %p", exp);
2241 ldlm_res_lvbo_update(res, NULL, 1);
2242 ldlm_lock_cancel(lock);
2243 ldlm_reprocess_all(res);
2244 ldlm_resource_putref(res);
2245 LDLM_LOCK_RELEASE(lock);
2248 if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
2250 "Cancel lock %p for export %p (loop %d), still have "
2251 "%d locks left on hash table.\n",
2252 lock, exp, ecl->ecl_loop,
2253 atomic_read(&hs->hs_count));
2260 * Cancel all locks for given export.
2262 * Typically called on client disconnection/eviction
2264 void ldlm_cancel_locks_for_export(struct obd_export *exp)
2266 struct export_cl_data ecl = {
2271 cfs_hash_for_each_empty(exp->exp_lock_hash,
2272 ldlm_cancel_locks_for_export_cb, &ecl);
2276 * Downgrade an exclusive lock.
2278 * A fast variant of ldlm_lock_convert for convertion of exclusive
2279 * locks. The convertion is always successful.
2280 * Used by Commit on Sharing (COS) code.
2282 * \param lock A lock to convert
2283 * \param new_mode new lock mode
2285 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
2289 LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
2290 LASSERT(new_mode == LCK_COS);
2292 lock_res_and_lock(lock);
2293 ldlm_resource_unlink_lock(lock);
2295 * Remove the lock from pool as it will be added again in
2296 * ldlm_grant_lock() called below.
2298 ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
2300 lock->l_req_mode = new_mode;
2301 ldlm_grant_lock(lock, NULL);
2302 unlock_res_and_lock(lock);
2303 ldlm_reprocess_all(lock->l_resource);
2307 EXPORT_SYMBOL(ldlm_lock_downgrade);
2310 * Attempt to convert already granted lock to a different mode.
2312 * While lock conversion is not currently used, future client-side
2313 * optimizations could take advantage of it to avoid discarding cached
2316 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
2319 CFS_LIST_HEAD(rpc_list);
2320 struct ldlm_resource *res;
2321 struct ldlm_namespace *ns;
2323 #ifdef HAVE_SERVER_SUPPORT
2325 struct sl_insert_point prev;
2327 struct ldlm_interval *node;
2330 /* Just return if mode is unchanged. */
2331 if (new_mode == lock->l_granted_mode) {
2332 *flags |= LDLM_FL_BLOCK_GRANTED;
2333 RETURN(lock->l_resource);
2336 /* I can't check the type of lock here because the bitlock of lock
2337 * is not held here, so do the allocation blindly. -jay */
2338 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
2339 if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */
2342 LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
2343 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2345 lock_res_and_lock(lock);
2347 res = lock->l_resource;
2348 ns = ldlm_res_to_ns(res);
2350 #ifdef HAVE_SERVER_SUPPORT
2351 old_mode = lock->l_req_mode;
2353 lock->l_req_mode = new_mode;
2354 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2355 #ifdef HAVE_SERVER_SUPPORT
2356 /* remember the lock position where the lock might be
2357 * added back to the granted list later and also
2358 * remember the join mode for skiplist fixing. */
2359 prev.res_link = lock->l_res_link.prev;
2360 prev.mode_link = lock->l_sl_mode.prev;
2361 prev.policy_link = lock->l_sl_policy.prev;
2363 ldlm_resource_unlink_lock(lock);
2365 ldlm_resource_unlink_lock(lock);
2366 if (res->lr_type == LDLM_EXTENT) {
2367 /* FIXME: ugly code, I have to attach the lock to a
2368 * interval node again since perhaps it will be granted
2370 CFS_INIT_LIST_HEAD(&node->li_group);
2371 ldlm_interval_attach(node, lock);
2377 * Remove old lock from the pool before adding the lock with new
2378 * mode below in ->policy()
2380 ldlm_pool_del(&ns->ns_pool, lock);
2382 /* If this is a local resource, put it on the appropriate list. */
2383 if (ns_is_client(ldlm_res_to_ns(res))) {
2384 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2385 ldlm_resource_add_lock(res, &res->lr_converting, lock);
2387 /* This should never happen, because of the way the
2388 * server handles conversions. */
2389 LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2393 ldlm_grant_lock(lock, &rpc_list);
2395 /* FIXME: completion handling not with lr_lock held ! */
2396 if (lock->l_completion_ast)
2397 lock->l_completion_ast(lock, 0, NULL);
2399 #ifdef HAVE_SERVER_SUPPORT
2404 ldlm_processing_policy policy;
2405 policy = ldlm_processing_policy_table[res->lr_type];
2406 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2407 if (rc == LDLM_ITER_STOP) {
2408 lock->l_req_mode = old_mode;
2409 if (res->lr_type == LDLM_EXTENT)
2410 ldlm_extent_add_lock(res, lock);
2412 ldlm_granted_list_add_lock(lock, &prev);
2416 *flags |= LDLM_FL_BLOCK_GRANTED;
2422 CERROR("This is client-side-only module, cannot handle "
2423 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2427 unlock_res_and_lock(lock);
2430 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2432 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2435 EXPORT_SYMBOL(ldlm_lock_convert);
2438 * Print lock with lock handle \a lockh description into debug log.
2440 * Used when printing all locks on a resource for debug purposes.
2442 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
2444 struct ldlm_lock *lock;
2446 if (!((libcfs_debug | D_ERROR) & level))
2449 lock = ldlm_handle2lock(lockh);
2453 LDLM_DEBUG_LIMIT(level, lock, "###");
2455 LDLM_LOCK_PUT(lock);
2457 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2460 * Print lock information with custom message into debug log.
2463 void _ldlm_lock_debug(struct ldlm_lock *lock,
2464 struct libcfs_debug_msg_data *msgdata,
2465 const char *fmt, ...)
2468 struct obd_export *exp = lock->l_export;
2469 struct ldlm_resource *resource = lock->l_resource;
2470 char *nid = "local";
2472 va_start(args, fmt);
2474 if (exp && exp->exp_connection) {
2475 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2476 } else if (exp && exp->exp_obd != NULL) {
2477 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2478 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2481 if (resource == NULL) {
2482 libcfs_debug_vmsg2(msgdata, fmt, args,
2483 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2484 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2485 "remote: "LPX64" expref: %d pid: %u timeout: %lu "
2488 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2489 lock->l_readers, lock->l_writers,
2490 ldlm_lockname[lock->l_granted_mode],
2491 ldlm_lockname[lock->l_req_mode],
2492 lock->l_flags, nid, lock->l_remote_handle.cookie,
2493 exp ? atomic_read(&exp->exp_refcount) : -99,
2494 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2499 switch (resource->lr_type) {
2501 libcfs_debug_vmsg2(msgdata, fmt, args,
2502 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2503 "res: "DLDLMRES" rrc: %d type: %s ["LPU64"->"LPU64"] "
2504 "(req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote: "
2505 LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2506 ldlm_lock_to_ns_name(lock), lock,
2507 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2508 lock->l_readers, lock->l_writers,
2509 ldlm_lockname[lock->l_granted_mode],
2510 ldlm_lockname[lock->l_req_mode],
2512 atomic_read(&resource->lr_refcount),
2513 ldlm_typename[resource->lr_type],
2514 lock->l_policy_data.l_extent.start,
2515 lock->l_policy_data.l_extent.end,
2516 lock->l_req_extent.start, lock->l_req_extent.end,
2517 lock->l_flags, nid, lock->l_remote_handle.cookie,
2518 exp ? atomic_read(&exp->exp_refcount) : -99,
2519 lock->l_pid, lock->l_callback_timeout,
2524 libcfs_debug_vmsg2(msgdata, fmt, args,
2525 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2526 "res: "DLDLMRES" rrc: %d type: %s pid: %d "
2527 "["LPU64"->"LPU64"] flags: "LPX64" nid: %s "
2528 "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2529 ldlm_lock_to_ns_name(lock), lock,
2530 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
2531 lock->l_readers, lock->l_writers,
2532 ldlm_lockname[lock->l_granted_mode],
2533 ldlm_lockname[lock->l_req_mode],
2535 atomic_read(&resource->lr_refcount),
2536 ldlm_typename[resource->lr_type],
2537 lock->l_policy_data.l_flock.pid,
2538 lock->l_policy_data.l_flock.start,
2539 lock->l_policy_data.l_flock.end,
2540 lock->l_flags, nid, lock->l_remote_handle.cookie,
2541 exp ? atomic_read(&exp->exp_refcount) : -99,
2542 lock->l_pid, lock->l_callback_timeout);
2546 libcfs_debug_vmsg2(msgdata, fmt, args,
2547 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2548 "res: "DLDLMRES" bits "LPX64" rrc: %d type: %s "
2549 "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2550 "pid: %u timeout: %lu lvb_type: %d\n",
2551 ldlm_lock_to_ns_name(lock),
2552 lock, lock->l_handle.h_cookie,
2553 atomic_read(&lock->l_refc),
2554 lock->l_readers, lock->l_writers,
2555 ldlm_lockname[lock->l_granted_mode],
2556 ldlm_lockname[lock->l_req_mode],
2558 lock->l_policy_data.l_inodebits.bits,
2559 atomic_read(&resource->lr_refcount),
2560 ldlm_typename[resource->lr_type],
2561 lock->l_flags, nid, lock->l_remote_handle.cookie,
2562 exp ? atomic_read(&exp->exp_refcount) : -99,
2563 lock->l_pid, lock->l_callback_timeout,
2568 libcfs_debug_vmsg2(msgdata, fmt, args,
2569 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2570 "res: "DLDLMRES" rrc: %d type: %s flags: "LPX64" "
2571 "nid: %s remote: "LPX64" expref: %d pid: %u "
2572 "timeout: %lu lvb_type: %d\n",
2573 ldlm_lock_to_ns_name(lock),
2574 lock, lock->l_handle.h_cookie,
2575 atomic_read(&lock->l_refc),
2576 lock->l_readers, lock->l_writers,
2577 ldlm_lockname[lock->l_granted_mode],
2578 ldlm_lockname[lock->l_req_mode],
2580 atomic_read(&resource->lr_refcount),
2581 ldlm_typename[resource->lr_type],
2582 lock->l_flags, nid, lock->l_remote_handle.cookie,
2583 exp ? atomic_read(&exp->exp_refcount) : -99,
2584 lock->l_pid, lock->l_callback_timeout,
2590 EXPORT_SYMBOL(_ldlm_lock_debug);