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[] = {
66 EXPORT_SYMBOL(ldlm_lockname);
68 char *ldlm_typename[] = {
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):
155 CERROR("Unknown intent %d\n", it);
159 EXPORT_SYMBOL(ldlm_it2str);
161 extern struct kmem_cache *ldlm_lock_slab;
163 #ifdef HAVE_SERVER_SUPPORT
164 static ldlm_processing_policy ldlm_processing_policy_table[] = {
165 [LDLM_PLAIN] ldlm_process_plain_lock,
166 [LDLM_EXTENT] ldlm_process_extent_lock,
168 [LDLM_FLOCK] ldlm_process_flock_lock,
170 [LDLM_IBITS] ldlm_process_inodebits_lock,
173 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
175 return ldlm_processing_policy_table[res->lr_type];
177 EXPORT_SYMBOL(ldlm_get_processing_policy);
178 #endif /* HAVE_SERVER_SUPPORT */
180 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
184 EXPORT_SYMBOL(ldlm_register_intent);
187 * REFCOUNTED LOCK OBJECTS
192 * Get a reference on a lock.
194 * Lock refcounts, during creation:
195 * - one special one for allocation, dec'd only once in destroy
196 * - one for being a lock that's in-use
197 * - one for the addref associated with a new lock
199 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
201 cfs_atomic_inc(&lock->l_refc);
204 EXPORT_SYMBOL(ldlm_lock_get);
207 * Release lock reference.
209 * Also frees the lock if it was last reference.
211 void ldlm_lock_put(struct ldlm_lock *lock)
215 LASSERT(lock->l_resource != LP_POISON);
216 LASSERT(cfs_atomic_read(&lock->l_refc) > 0);
217 if (cfs_atomic_dec_and_test(&lock->l_refc)) {
218 struct ldlm_resource *res;
221 "final lock_put on destroyed lock, freeing it.");
223 res = lock->l_resource;
224 LASSERT(lock->l_flags & LDLM_FL_DESTROYED);
225 LASSERT(cfs_list_empty(&lock->l_res_link));
226 LASSERT(cfs_list_empty(&lock->l_pending_chain));
228 lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats,
230 lu_ref_del(&res->lr_reference, "lock", lock);
231 ldlm_resource_putref(res);
232 lock->l_resource = NULL;
233 if (lock->l_export) {
234 class_export_lock_put(lock->l_export, lock);
235 lock->l_export = NULL;
238 if (lock->l_lvb_data != NULL)
239 OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
241 ldlm_interval_free(ldlm_interval_detach(lock));
242 lu_ref_fini(&lock->l_reference);
243 OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle);
248 EXPORT_SYMBOL(ldlm_lock_put);
251 * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked.
253 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
256 if (!cfs_list_empty(&lock->l_lru)) {
257 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
259 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
260 cfs_list_del_init(&lock->l_lru);
261 if (lock->l_flags & LDLM_FL_SKIPPED)
262 lock->l_flags &= ~LDLM_FL_SKIPPED;
263 LASSERT(ns->ns_nr_unused > 0);
271 * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first.
273 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
275 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
279 if (lock->l_flags & LDLM_FL_NS_SRV) {
280 LASSERT(cfs_list_empty(&lock->l_lru));
284 spin_lock(&ns->ns_lock);
285 rc = ldlm_lock_remove_from_lru_nolock(lock);
286 spin_unlock(&ns->ns_lock);
292 * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked.
294 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
296 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
298 lock->l_last_used = cfs_time_current();
299 LASSERT(cfs_list_empty(&lock->l_lru));
300 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
301 cfs_list_add_tail(&lock->l_lru, &ns->ns_unused_list);
302 LASSERT(ns->ns_nr_unused >= 0);
307 * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks
310 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
312 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
315 spin_lock(&ns->ns_lock);
316 ldlm_lock_add_to_lru_nolock(lock);
317 spin_unlock(&ns->ns_lock);
322 * Moves LDLM lock \a lock that is already in namespace LRU to the tail of
323 * the LRU. Performs necessary LRU locking
325 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
327 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
330 if (lock->l_flags & LDLM_FL_NS_SRV) {
331 LASSERT(cfs_list_empty(&lock->l_lru));
336 spin_lock(&ns->ns_lock);
337 if (!cfs_list_empty(&lock->l_lru)) {
338 ldlm_lock_remove_from_lru_nolock(lock);
339 ldlm_lock_add_to_lru_nolock(lock);
341 spin_unlock(&ns->ns_lock);
346 * Helper to destroy a locked lock.
348 * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock
349 * Must be called with l_lock and lr_lock held.
351 * Does not actually free the lock data, but rather marks the lock as
352 * destroyed by setting l_destroyed field in the lock to 1. Destroys a
353 * handle->lock association too, so that the lock can no longer be found
354 * and removes the lock from LRU list. Actual lock freeing occurs when
355 * last lock reference goes away.
357 * Original comment (of some historical value):
358 * This used to have a 'strict' flag, which recovery would use to mark an
359 * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
360 * shall explain why it's gone: with the new hash table scheme, once you call
361 * ldlm_lock_destroy, you can never drop your final references on this lock.
362 * Because it's not in the hash table anymore. -phil
364 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
368 if (lock->l_readers || lock->l_writers) {
369 LDLM_ERROR(lock, "lock still has references");
373 if (!cfs_list_empty(&lock->l_res_link)) {
374 LDLM_ERROR(lock, "lock still on resource");
378 if (lock->l_flags & LDLM_FL_DESTROYED) {
379 LASSERT(cfs_list_empty(&lock->l_lru));
383 lock->l_flags |= LDLM_FL_DESTROYED;
385 if (lock->l_export && lock->l_export->exp_lock_hash) {
386 /* NB: it's safe to call cfs_hash_del() even lock isn't
387 * in exp_lock_hash. */
388 /* In the function below, .hs_keycmp resolves to
389 * ldlm_export_lock_keycmp() */
390 /* coverity[overrun-buffer-val] */
391 cfs_hash_del(lock->l_export->exp_lock_hash,
392 &lock->l_remote_handle, &lock->l_exp_hash);
395 ldlm_lock_remove_from_lru(lock);
396 class_handle_unhash(&lock->l_handle);
399 /* Wake anyone waiting for this lock */
400 /* FIXME: I should probably add yet another flag, instead of using
401 * l_export to only call this on clients */
403 class_export_put(lock->l_export);
404 lock->l_export = NULL;
405 if (lock->l_export && lock->l_completion_ast)
406 lock->l_completion_ast(lock, 0);
413 * Destroys a LDLM lock \a lock. Performs necessary locking first.
415 void ldlm_lock_destroy(struct ldlm_lock *lock)
419 lock_res_and_lock(lock);
420 first = ldlm_lock_destroy_internal(lock);
421 unlock_res_and_lock(lock);
423 /* drop reference from hashtable only for first destroy */
425 lu_ref_del(&lock->l_reference, "hash", lock);
426 LDLM_LOCK_RELEASE(lock);
432 * Destroys a LDLM lock \a lock that is already locked.
434 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
438 first = ldlm_lock_destroy_internal(lock);
439 /* drop reference from hashtable only for first destroy */
441 lu_ref_del(&lock->l_reference, "hash", lock);
442 LDLM_LOCK_RELEASE(lock);
447 /* this is called by portals_handle2object with the handle lock taken */
448 static void lock_handle_addref(void *lock)
450 LDLM_LOCK_GET((struct ldlm_lock *)lock);
453 static void lock_handle_free(void *lock, int size)
455 LASSERT(size == sizeof(struct ldlm_lock));
456 OBD_SLAB_FREE(lock, ldlm_lock_slab, size);
459 struct portals_handle_ops lock_handle_ops = {
460 .hop_addref = lock_handle_addref,
461 .hop_free = lock_handle_free,
466 * Allocate and initialize new lock structure.
468 * usage: pass in a resource on which you have done ldlm_resource_get
469 * new lock will take over the refcount.
470 * returns: lock with refcount 2 - one for current caller and one for remote
472 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
474 struct ldlm_lock *lock;
477 if (resource == NULL)
480 OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, __GFP_IO);
484 spin_lock_init(&lock->l_lock);
485 lock->l_resource = resource;
486 lu_ref_add(&resource->lr_reference, "lock", lock);
488 cfs_atomic_set(&lock->l_refc, 2);
489 CFS_INIT_LIST_HEAD(&lock->l_res_link);
490 CFS_INIT_LIST_HEAD(&lock->l_lru);
491 CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
492 CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
493 CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
494 CFS_INIT_LIST_HEAD(&lock->l_rk_ast);
495 cfs_waitq_init(&lock->l_waitq);
496 lock->l_blocking_lock = NULL;
497 CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
498 CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
499 CFS_INIT_HLIST_NODE(&lock->l_exp_hash);
500 CFS_INIT_HLIST_NODE(&lock->l_exp_flock_hash);
502 lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats,
504 CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
505 class_handle_hash(&lock->l_handle, &lock_handle_ops);
507 lu_ref_init(&lock->l_reference);
508 lu_ref_add(&lock->l_reference, "hash", lock);
509 lock->l_callback_timeout = 0;
511 #if LUSTRE_TRACKS_LOCK_EXP_REFS
512 CFS_INIT_LIST_HEAD(&lock->l_exp_refs_link);
513 lock->l_exp_refs_nr = 0;
514 lock->l_exp_refs_target = NULL;
516 CFS_INIT_LIST_HEAD(&lock->l_exp_list);
522 * Moves LDLM lock \a lock to another resource.
523 * This is used on client when server returns some other lock than requested
524 * (typically as a result of intent operation)
526 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
527 const struct ldlm_res_id *new_resid)
529 struct ldlm_resource *oldres = lock->l_resource;
530 struct ldlm_resource *newres;
534 LASSERT(ns_is_client(ns));
536 lock_res_and_lock(lock);
537 if (memcmp(new_resid, &lock->l_resource->lr_name,
538 sizeof(lock->l_resource->lr_name)) == 0) {
540 unlock_res_and_lock(lock);
544 LASSERT(new_resid->name[0] != 0);
546 /* This function assumes that the lock isn't on any lists */
547 LASSERT(cfs_list_empty(&lock->l_res_link));
549 type = oldres->lr_type;
550 unlock_res_and_lock(lock);
552 newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
556 lu_ref_add(&newres->lr_reference, "lock", lock);
558 * To flip the lock from the old to the new resource, lock, oldres and
559 * newres have to be locked. Resource spin-locks are nested within
560 * lock->l_lock, and are taken in the memory address order to avoid
563 spin_lock(&lock->l_lock);
564 oldres = lock->l_resource;
565 if (oldres < newres) {
567 lock_res_nested(newres, LRT_NEW);
570 lock_res_nested(oldres, LRT_NEW);
572 LASSERT(memcmp(new_resid, &oldres->lr_name,
573 sizeof oldres->lr_name) != 0);
574 lock->l_resource = newres;
576 unlock_res_and_lock(lock);
578 /* ...and the flowers are still standing! */
579 lu_ref_del(&oldres->lr_reference, "lock", lock);
580 ldlm_resource_putref(oldres);
584 EXPORT_SYMBOL(ldlm_lock_change_resource);
586 /** \defgroup ldlm_handles LDLM HANDLES
587 * Ways to get hold of locks without any addresses.
592 * Fills in handle for LDLM lock \a lock into supplied \a lockh
593 * Does not take any references.
595 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
597 lockh->cookie = lock->l_handle.h_cookie;
599 EXPORT_SYMBOL(ldlm_lock2handle);
602 * Obtain a lock reference by handle.
604 * if \a flags: atomically get the lock and set the flags.
605 * Return NULL if flag already set
607 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
610 struct ldlm_lock *lock;
615 lock = class_handle2object(handle->cookie, NULL);
619 /* It's unlikely but possible that someone marked the lock as
620 * destroyed after we did handle2object on it */
621 if (flags == 0 && ((lock->l_flags & LDLM_FL_DESTROYED)== 0)) {
622 lu_ref_add(&lock->l_reference, "handle", cfs_current());
626 lock_res_and_lock(lock);
628 LASSERT(lock->l_resource != NULL);
630 lu_ref_add_atomic(&lock->l_reference, "handle", cfs_current());
631 if (unlikely(lock->l_flags & LDLM_FL_DESTROYED)) {
632 unlock_res_and_lock(lock);
633 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
638 if (flags && (lock->l_flags & flags)) {
639 unlock_res_and_lock(lock);
645 lock->l_flags |= flags;
647 unlock_res_and_lock(lock);
650 EXPORT_SYMBOL(__ldlm_handle2lock);
651 /** @} ldlm_handles */
654 * Fill in "on the wire" representation for given LDLM lock into supplied
655 * lock descriptor \a desc structure.
657 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
659 struct obd_export *exp = lock->l_export ?: lock->l_conn_export;
661 /* INODEBITS_INTEROP: If the other side does not support
662 * inodebits, reply with a plain lock descriptor. */
663 if ((lock->l_resource->lr_type == LDLM_IBITS) &&
664 (exp && !(exp_connect_flags(exp) & OBD_CONNECT_IBITS))) {
665 /* Make sure all the right bits are set in this lock we
666 are going to pass to client */
667 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
668 (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
669 MDS_INODELOCK_LAYOUT),
670 "Inappropriate inode lock bits during "
671 "conversion " LPU64 "\n",
672 lock->l_policy_data.l_inodebits.bits);
674 ldlm_res2desc(lock->l_resource, &desc->l_resource);
675 desc->l_resource.lr_type = LDLM_PLAIN;
677 /* Convert "new" lock mode to something old client can
679 if ((lock->l_req_mode == LCK_CR) ||
680 (lock->l_req_mode == LCK_CW))
681 desc->l_req_mode = LCK_PR;
683 desc->l_req_mode = lock->l_req_mode;
684 if ((lock->l_granted_mode == LCK_CR) ||
685 (lock->l_granted_mode == LCK_CW)) {
686 desc->l_granted_mode = LCK_PR;
688 /* We never grant PW/EX locks to clients */
689 LASSERT((lock->l_granted_mode != LCK_PW) &&
690 (lock->l_granted_mode != LCK_EX));
691 desc->l_granted_mode = lock->l_granted_mode;
694 /* We do not copy policy here, because there is no
695 policy for plain locks */
697 ldlm_res2desc(lock->l_resource, &desc->l_resource);
698 desc->l_req_mode = lock->l_req_mode;
699 desc->l_granted_mode = lock->l_granted_mode;
700 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
701 &lock->l_policy_data,
702 &desc->l_policy_data);
705 EXPORT_SYMBOL(ldlm_lock2desc);
708 * Add a lock to list of conflicting locks to send AST to.
710 * Only add if we have not sent a blocking AST to the lock yet.
712 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
713 cfs_list_t *work_list)
715 if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
716 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
717 lock->l_flags |= LDLM_FL_AST_SENT;
718 /* If the enqueuing client said so, tell the AST recipient to
719 * discard dirty data, rather than writing back. */
720 if (new->l_flags & LDLM_FL_AST_DISCARD_DATA)
721 lock->l_flags |= LDLM_FL_DISCARD_DATA;
722 LASSERT(cfs_list_empty(&lock->l_bl_ast));
723 cfs_list_add(&lock->l_bl_ast, work_list);
725 LASSERT(lock->l_blocking_lock == NULL);
726 lock->l_blocking_lock = LDLM_LOCK_GET(new);
731 * Add a lock to list of just granted locks to send completion AST to.
733 void ldlm_add_cp_work_item(struct ldlm_lock *lock, cfs_list_t *work_list)
735 if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
736 lock->l_flags |= LDLM_FL_CP_REQD;
737 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
738 LASSERT(cfs_list_empty(&lock->l_cp_ast));
739 cfs_list_add(&lock->l_cp_ast, work_list);
745 * Aggregator function to add AST work items into a list. Determines
746 * what sort of an AST work needs to be done and calls the proper
748 * Must be called with lr_lock held.
750 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
751 cfs_list_t *work_list)
754 check_res_locked(lock->l_resource);
756 ldlm_add_bl_work_item(lock, new, work_list);
758 ldlm_add_cp_work_item(lock, work_list);
763 * Add specified reader/writer reference to LDLM lock with handle \a lockh.
764 * r/w reference type is determined by \a mode
765 * Calls ldlm_lock_addref_internal.
767 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
769 struct ldlm_lock *lock;
771 lock = ldlm_handle2lock(lockh);
772 LASSERT(lock != NULL);
773 ldlm_lock_addref_internal(lock, mode);
776 EXPORT_SYMBOL(ldlm_lock_addref);
780 * Add specified reader/writer reference to LDLM lock \a lock.
781 * r/w reference type is determined by \a mode
782 * Removes lock from LRU if it is there.
783 * Assumes the LDLM lock is already locked.
785 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
787 ldlm_lock_remove_from_lru(lock);
788 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
790 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
792 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
794 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
797 lu_ref_add_atomic(&lock->l_reference, "user", lock);
798 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
802 * Attempts to add reader/writer reference to a lock with handle \a lockh, and
803 * fails if lock is already LDLM_FL_CBPENDING or destroyed.
805 * \retval 0 success, lock was addref-ed
807 * \retval -EAGAIN lock is being canceled.
809 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
811 struct ldlm_lock *lock;
815 lock = ldlm_handle2lock(lockh);
817 lock_res_and_lock(lock);
818 if (lock->l_readers != 0 || lock->l_writers != 0 ||
819 !(lock->l_flags & LDLM_FL_CBPENDING)) {
820 ldlm_lock_addref_internal_nolock(lock, mode);
823 unlock_res_and_lock(lock);
828 EXPORT_SYMBOL(ldlm_lock_addref_try);
831 * Add specified reader/writer reference to LDLM lock \a lock.
832 * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work.
833 * Only called for local locks.
835 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
837 lock_res_and_lock(lock);
838 ldlm_lock_addref_internal_nolock(lock, mode);
839 unlock_res_and_lock(lock);
843 * Removes reader/writer reference for LDLM lock \a lock.
844 * Assumes LDLM lock is already locked.
845 * only called in ldlm_flock_destroy and for local locks.
846 * Does NOT add lock to LRU if no r/w references left to accomodate flock locks
847 * that cannot be placed in LRU.
849 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
851 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
852 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
853 LASSERT(lock->l_readers > 0);
854 lu_ref_del(&lock->l_reference, "reader", lock);
857 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
858 LASSERT(lock->l_writers > 0);
859 lu_ref_del(&lock->l_reference, "writer", lock);
863 lu_ref_del(&lock->l_reference, "user", lock);
864 LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */
868 * Removes reader/writer reference for LDLM lock \a lock.
869 * Locks LDLM lock first.
870 * If the lock is determined to be client lock on a client and r/w refcount
871 * drops to zero and the lock is not blocked, the lock is added to LRU lock
873 * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called.
875 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
877 struct ldlm_namespace *ns;
880 lock_res_and_lock(lock);
882 ns = ldlm_lock_to_ns(lock);
884 ldlm_lock_decref_internal_nolock(lock, mode);
886 if (lock->l_flags & LDLM_FL_LOCAL &&
887 !lock->l_readers && !lock->l_writers) {
888 /* If this is a local lock on a server namespace and this was
889 * the last reference, cancel the lock. */
890 CDEBUG(D_INFO, "forcing cancel of local lock\n");
891 lock->l_flags |= LDLM_FL_CBPENDING;
894 if (!lock->l_readers && !lock->l_writers &&
895 (lock->l_flags & LDLM_FL_CBPENDING)) {
896 /* If we received a blocked AST and this was the last reference,
897 * run the callback. */
898 if ((lock->l_flags & LDLM_FL_NS_SRV) && lock->l_export)
899 CERROR("FL_CBPENDING set on non-local lock--just a "
902 LDLM_DEBUG(lock, "final decref done on cbpending lock");
904 LDLM_LOCK_GET(lock); /* dropped by bl thread */
905 ldlm_lock_remove_from_lru(lock);
906 unlock_res_and_lock(lock);
908 if (lock->l_flags & LDLM_FL_FAIL_LOC)
909 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
911 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
912 ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
913 ldlm_handle_bl_callback(ns, NULL, lock);
914 } else if (ns_is_client(ns) &&
915 !lock->l_readers && !lock->l_writers &&
916 !(lock->l_flags & LDLM_FL_NO_LRU) &&
917 !(lock->l_flags & LDLM_FL_BL_AST)) {
919 LDLM_DEBUG(lock, "add lock into lru list");
921 /* If this is a client-side namespace and this was the last
922 * reference, put it on the LRU. */
923 ldlm_lock_add_to_lru(lock);
924 unlock_res_and_lock(lock);
926 if (lock->l_flags & LDLM_FL_FAIL_LOC)
927 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
929 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
930 * are not supported by the server, otherwise, it is done on
932 if (!exp_connect_cancelset(lock->l_conn_export) &&
933 !ns_connect_lru_resize(ns))
934 ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0);
936 LDLM_DEBUG(lock, "do not add lock into lru list");
937 unlock_res_and_lock(lock);
944 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
946 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
948 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
949 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
950 ldlm_lock_decref_internal(lock, mode);
953 EXPORT_SYMBOL(ldlm_lock_decref);
956 * Decrease reader/writer refcount for LDLM lock with handle
957 * \a lockh and mark it for subsequent cancellation once r/w refcount
958 * drops to zero instead of putting into LRU.
960 * Typical usage is for GROUP locks which we cannot allow to be cached.
962 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
964 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
967 LASSERT(lock != NULL);
969 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
970 lock_res_and_lock(lock);
971 lock->l_flags |= LDLM_FL_CBPENDING;
972 unlock_res_and_lock(lock);
973 ldlm_lock_decref_internal(lock, mode);
976 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
978 struct sl_insert_point {
979 cfs_list_t *res_link;
980 cfs_list_t *mode_link;
981 cfs_list_t *policy_link;
985 * Finds a position to insert the new lock into granted lock list.
987 * Used for locks eligible for skiplist optimization.
990 * queue [input]: the granted list where search acts on;
991 * req [input]: the lock whose position to be located;
992 * prev [output]: positions within 3 lists to insert @req to
996 * - ldlm_grant_lock_with_skiplist
998 static void search_granted_lock(cfs_list_t *queue,
999 struct ldlm_lock *req,
1000 struct sl_insert_point *prev)
1003 struct ldlm_lock *lock, *mode_end, *policy_end;
1006 cfs_list_for_each(tmp, queue) {
1007 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1009 mode_end = cfs_list_entry(lock->l_sl_mode.prev,
1010 struct ldlm_lock, l_sl_mode);
1012 if (lock->l_req_mode != req->l_req_mode) {
1013 /* jump to last lock of mode group */
1014 tmp = &mode_end->l_res_link;
1018 /* suitable mode group is found */
1019 if (lock->l_resource->lr_type == LDLM_PLAIN) {
1020 /* insert point is last lock of the mode group */
1021 prev->res_link = &mode_end->l_res_link;
1022 prev->mode_link = &mode_end->l_sl_mode;
1023 prev->policy_link = &req->l_sl_policy;
1026 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
1029 cfs_list_entry(lock->l_sl_policy.prev,
1033 if (lock->l_policy_data.l_inodebits.bits ==
1034 req->l_policy_data.l_inodebits.bits) {
1035 /* insert point is last lock of
1036 * the policy group */
1038 &policy_end->l_res_link;
1040 &policy_end->l_sl_mode;
1042 &policy_end->l_sl_policy;
1047 if (policy_end == mode_end)
1048 /* done with mode group */
1051 /* go to next policy group within mode group */
1052 tmp = policy_end->l_res_link.next;
1053 lock = cfs_list_entry(tmp, struct ldlm_lock,
1055 } /* loop over policy groups within the mode group */
1057 /* insert point is last lock of the mode group,
1058 * new policy group is started */
1059 prev->res_link = &mode_end->l_res_link;
1060 prev->mode_link = &mode_end->l_sl_mode;
1061 prev->policy_link = &req->l_sl_policy;
1065 LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
1070 /* insert point is last lock on the queue,
1071 * new mode group and new policy group are started */
1072 prev->res_link = queue->prev;
1073 prev->mode_link = &req->l_sl_mode;
1074 prev->policy_link = &req->l_sl_policy;
1080 * Add a lock into resource granted list after a position described by
1083 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
1084 struct sl_insert_point *prev)
1086 struct ldlm_resource *res = lock->l_resource;
1089 check_res_locked(res);
1091 ldlm_resource_dump(D_INFO, res);
1092 LDLM_DEBUG(lock, "About to add lock:");
1094 if (lock->l_flags & LDLM_FL_DESTROYED) {
1095 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1099 LASSERT(cfs_list_empty(&lock->l_res_link));
1100 LASSERT(cfs_list_empty(&lock->l_sl_mode));
1101 LASSERT(cfs_list_empty(&lock->l_sl_policy));
1104 * lock->link == prev->link means lock is first starting the group.
1105 * Don't re-add to itself to suppress kernel warnings.
1107 if (&lock->l_res_link != prev->res_link)
1108 cfs_list_add(&lock->l_res_link, prev->res_link);
1109 if (&lock->l_sl_mode != prev->mode_link)
1110 cfs_list_add(&lock->l_sl_mode, prev->mode_link);
1111 if (&lock->l_sl_policy != prev->policy_link)
1112 cfs_list_add(&lock->l_sl_policy, prev->policy_link);
1118 * Add a lock to granted list on a resource maintaining skiplist
1121 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
1123 struct sl_insert_point prev;
1126 LASSERT(lock->l_req_mode == lock->l_granted_mode);
1128 search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
1129 ldlm_granted_list_add_lock(lock, &prev);
1134 * Perform lock granting bookkeeping.
1136 * Includes putting the lock into granted list and updating lock mode.
1138 * - ldlm_lock_enqueue
1139 * - ldlm_reprocess_queue
1140 * - ldlm_lock_convert
1142 * must be called with lr_lock held
1144 void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list)
1146 struct ldlm_resource *res = lock->l_resource;
1149 check_res_locked(res);
1151 lock->l_granted_mode = lock->l_req_mode;
1152 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
1153 ldlm_grant_lock_with_skiplist(lock);
1154 else if (res->lr_type == LDLM_EXTENT)
1155 ldlm_extent_add_lock(res, lock);
1157 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1159 if (lock->l_granted_mode < res->lr_most_restr)
1160 res->lr_most_restr = lock->l_granted_mode;
1162 if (work_list && lock->l_completion_ast != NULL)
1163 ldlm_add_ast_work_item(lock, NULL, work_list);
1165 ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1170 * Search for a lock with given properties in a queue.
1172 * \retval a referenced lock or NULL. See the flag descriptions below, in the
1173 * comment above ldlm_lock_match
1175 static struct ldlm_lock *search_queue(cfs_list_t *queue,
1177 ldlm_policy_data_t *policy,
1178 struct ldlm_lock *old_lock,
1179 __u64 flags, int unref)
1181 struct ldlm_lock *lock;
1184 cfs_list_for_each(tmp, queue) {
1187 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1189 if (lock == old_lock)
1192 /* Check if this lock can be matched.
1193 * Used by LU-2919(exclusive open) for open lease lock */
1194 if (ldlm_is_excl(lock))
1197 /* llite sometimes wants to match locks that will be
1198 * canceled when their users drop, but we allow it to match
1199 * if it passes in CBPENDING and the lock still has users.
1200 * this is generally only going to be used by children
1201 * whose parents already hold a lock so forward progress
1202 * can still happen. */
1203 if (lock->l_flags & LDLM_FL_CBPENDING &&
1204 !(flags & LDLM_FL_CBPENDING))
1206 if (!unref && lock->l_flags & LDLM_FL_CBPENDING &&
1207 lock->l_readers == 0 && lock->l_writers == 0)
1210 if (!(lock->l_req_mode & *mode))
1212 match = lock->l_req_mode;
1214 if (lock->l_resource->lr_type == LDLM_EXTENT &&
1215 (lock->l_policy_data.l_extent.start >
1216 policy->l_extent.start ||
1217 lock->l_policy_data.l_extent.end < policy->l_extent.end))
1220 if (unlikely(match == LCK_GROUP) &&
1221 lock->l_resource->lr_type == LDLM_EXTENT &&
1222 lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
1225 /* We match if we have existing lock with same or wider set
1227 if (lock->l_resource->lr_type == LDLM_IBITS &&
1228 ((lock->l_policy_data.l_inodebits.bits &
1229 policy->l_inodebits.bits) !=
1230 policy->l_inodebits.bits))
1233 if (!unref && (lock->l_flags & LDLM_FL_GONE_MASK))
1236 if ((flags & LDLM_FL_LOCAL_ONLY) &&
1237 !(lock->l_flags & LDLM_FL_LOCAL))
1240 if (flags & LDLM_FL_TEST_LOCK) {
1241 LDLM_LOCK_GET(lock);
1242 ldlm_lock_touch_in_lru(lock);
1244 ldlm_lock_addref_internal_nolock(lock, match);
1253 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1255 if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) {
1256 lock->l_flags |= LDLM_FL_FAIL_NOTIFIED;
1257 cfs_waitq_broadcast(&lock->l_waitq);
1260 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1262 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1264 lock_res_and_lock(lock);
1265 ldlm_lock_fail_match_locked(lock);
1266 unlock_res_and_lock(lock);
1268 EXPORT_SYMBOL(ldlm_lock_fail_match);
1271 * Mark lock as "matchable" by OST.
1273 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1275 * Assumes LDLM lock is already locked.
1277 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1279 lock->l_flags |= LDLM_FL_LVB_READY;
1280 cfs_waitq_broadcast(&lock->l_waitq);
1282 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1285 * Mark lock as "matchable" by OST.
1286 * Locks the lock and then \see ldlm_lock_allow_match_locked
1288 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1290 lock_res_and_lock(lock);
1291 ldlm_lock_allow_match_locked(lock);
1292 unlock_res_and_lock(lock);
1294 EXPORT_SYMBOL(ldlm_lock_allow_match);
1297 * Attempt to find a lock with specified properties.
1299 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1302 * Can be called in two ways:
1304 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1305 * for a duplicate of.
1307 * Otherwise, all of the fields must be filled in, to match against.
1309 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1310 * server (ie, connh is NULL)
1311 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1312 * list will be considered
1313 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1314 * to be canceled can still be matched as long as they still have reader
1315 * or writer refernces
1316 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1317 * just tell us if we would have matched.
1319 * \retval 1 if it finds an already-existing lock that is compatible; in this
1320 * case, lockh is filled in with a addref()ed lock
1322 * We also check security context, and if that fails we simply return 0 (to
1323 * keep caller code unchanged), the context failure will be discovered by
1324 * caller sometime later.
1326 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1327 const struct ldlm_res_id *res_id, ldlm_type_t type,
1328 ldlm_policy_data_t *policy, ldlm_mode_t mode,
1329 struct lustre_handle *lockh, int unref)
1331 struct ldlm_resource *res;
1332 struct ldlm_lock *lock, *old_lock = NULL;
1337 old_lock = ldlm_handle2lock(lockh);
1340 ns = ldlm_lock_to_ns(old_lock);
1341 res_id = &old_lock->l_resource->lr_name;
1342 type = old_lock->l_resource->lr_type;
1343 mode = old_lock->l_req_mode;
1346 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1348 LASSERT(old_lock == NULL);
1352 LDLM_RESOURCE_ADDREF(res);
1355 lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
1359 if (flags & LDLM_FL_BLOCK_GRANTED)
1361 lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
1365 lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
1373 LDLM_RESOURCE_DELREF(res);
1374 ldlm_resource_putref(res);
1377 ldlm_lock2handle(lock, lockh);
1378 if ((flags & LDLM_FL_LVB_READY) &&
1379 (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1380 __u64 wait_flags = LDLM_FL_LVB_READY |
1381 LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED;
1382 struct l_wait_info lwi;
1383 if (lock->l_completion_ast) {
1384 int err = lock->l_completion_ast(lock,
1385 LDLM_FL_WAIT_NOREPROC,
1388 if (flags & LDLM_FL_TEST_LOCK)
1389 LDLM_LOCK_RELEASE(lock);
1391 ldlm_lock_decref_internal(lock,
1398 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1399 NULL, LWI_ON_SIGNAL_NOOP, NULL);
1401 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1402 l_wait_event(lock->l_waitq,
1403 lock->l_flags & wait_flags,
1405 if (!(lock->l_flags & LDLM_FL_LVB_READY)) {
1406 if (flags & LDLM_FL_TEST_LOCK)
1407 LDLM_LOCK_RELEASE(lock);
1409 ldlm_lock_decref_internal(lock, mode);
1416 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1417 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1418 res_id->name[2] : policy->l_extent.start,
1419 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1420 res_id->name[3] : policy->l_extent.end);
1422 /* check user's security context */
1423 if (lock->l_conn_export &&
1424 sptlrpc_import_check_ctx(
1425 class_exp2cliimp(lock->l_conn_export))) {
1426 if (!(flags & LDLM_FL_TEST_LOCK))
1427 ldlm_lock_decref_internal(lock, mode);
1431 if (flags & LDLM_FL_TEST_LOCK)
1432 LDLM_LOCK_RELEASE(lock);
1434 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1435 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1436 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1437 type, mode, res_id->name[0], res_id->name[1],
1438 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1439 res_id->name[2] :policy->l_extent.start,
1440 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1441 res_id->name[3] : policy->l_extent.end);
1444 LDLM_LOCK_PUT(old_lock);
1446 return rc ? mode : 0;
1448 EXPORT_SYMBOL(ldlm_lock_match);
1450 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1453 struct ldlm_lock *lock;
1454 ldlm_mode_t mode = 0;
1457 lock = ldlm_handle2lock(lockh);
1459 lock_res_and_lock(lock);
1460 if (lock->l_flags & LDLM_FL_GONE_MASK)
1463 if (lock->l_flags & LDLM_FL_CBPENDING &&
1464 lock->l_readers == 0 && lock->l_writers == 0)
1468 *bits = lock->l_policy_data.l_inodebits.bits;
1469 mode = lock->l_granted_mode;
1470 ldlm_lock_addref_internal_nolock(lock, mode);
1477 unlock_res_and_lock(lock);
1478 LDLM_LOCK_PUT(lock);
1482 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1484 /** The caller must guarantee that the buffer is large enough. */
1485 int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill,
1486 enum req_location loc, void *data, int size)
1491 LASSERT(data != NULL);
1494 switch (lock->l_lvb_type) {
1496 if (size == sizeof(struct ost_lvb)) {
1497 if (loc == RCL_CLIENT)
1498 lvb = req_capsule_client_swab_get(pill,
1500 lustre_swab_ost_lvb);
1502 lvb = req_capsule_server_swab_get(pill,
1504 lustre_swab_ost_lvb);
1505 if (unlikely(lvb == NULL)) {
1506 LDLM_ERROR(lock, "no LVB");
1510 memcpy(data, lvb, size);
1511 } else if (size == sizeof(struct ost_lvb_v1)) {
1512 struct ost_lvb *olvb = data;
1514 if (loc == RCL_CLIENT)
1515 lvb = req_capsule_client_swab_get(pill,
1517 lustre_swab_ost_lvb_v1);
1519 lvb = req_capsule_server_sized_swab_get(pill,
1521 lustre_swab_ost_lvb_v1);
1522 if (unlikely(lvb == NULL)) {
1523 LDLM_ERROR(lock, "no LVB");
1527 memcpy(data, lvb, size);
1528 olvb->lvb_mtime_ns = 0;
1529 olvb->lvb_atime_ns = 0;
1530 olvb->lvb_ctime_ns = 0;
1532 LDLM_ERROR(lock, "Replied unexpected ost LVB size %d",
1538 if (size == sizeof(struct lquota_lvb)) {
1539 if (loc == RCL_CLIENT)
1540 lvb = req_capsule_client_swab_get(pill,
1542 lustre_swab_lquota_lvb);
1544 lvb = req_capsule_server_swab_get(pill,
1546 lustre_swab_lquota_lvb);
1547 if (unlikely(lvb == NULL)) {
1548 LDLM_ERROR(lock, "no LVB");
1552 memcpy(data, lvb, size);
1554 LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d",
1563 if (loc == RCL_CLIENT)
1564 lvb = req_capsule_client_get(pill, &RMF_DLM_LVB);
1566 lvb = req_capsule_server_get(pill, &RMF_DLM_LVB);
1567 if (unlikely(lvb == NULL)) {
1568 LDLM_ERROR(lock, "no LVB");
1572 memcpy(data, lvb, size);
1575 LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type);
1576 libcfs_debug_dumpstack(NULL);
1584 * Create and fill in new LDLM lock with specified properties.
1585 * Returns a referenced lock
1587 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1588 const struct ldlm_res_id *res_id,
1591 const struct ldlm_callback_suite *cbs,
1592 void *data, __u32 lvb_len,
1593 enum lvb_type lvb_type)
1595 struct ldlm_lock *lock;
1596 struct ldlm_resource *res;
1599 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1603 lock = ldlm_lock_new(res);
1608 lock->l_req_mode = mode;
1609 lock->l_ast_data = data;
1610 lock->l_pid = current_pid();
1611 if (ns_is_server(ns))
1612 lock->l_flags |= LDLM_FL_NS_SRV;
1614 lock->l_blocking_ast = cbs->lcs_blocking;
1615 lock->l_completion_ast = cbs->lcs_completion;
1616 lock->l_glimpse_ast = cbs->lcs_glimpse;
1619 lock->l_tree_node = NULL;
1620 /* if this is the extent lock, allocate the interval tree node */
1621 if (type == LDLM_EXTENT) {
1622 if (ldlm_interval_alloc(lock) == NULL)
1627 lock->l_lvb_len = lvb_len;
1628 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1629 if (lock->l_lvb_data == NULL)
1633 lock->l_lvb_type = lvb_type;
1634 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1640 ldlm_lock_destroy(lock);
1641 LDLM_LOCK_RELEASE(lock);
1646 * Enqueue (request) a lock.
1648 * Does not block. As a result of enqueue the lock would be put
1649 * into granted or waiting list.
1651 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1652 * set, skip all the enqueueing and delegate lock processing to intent policy
1655 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1656 struct ldlm_lock **lockp,
1657 void *cookie, __u64 *flags)
1659 struct ldlm_lock *lock = *lockp;
1660 struct ldlm_resource *res = lock->l_resource;
1661 int local = ns_is_client(ldlm_res_to_ns(res));
1662 #ifdef HAVE_SERVER_SUPPORT
1663 ldlm_processing_policy policy;
1665 ldlm_error_t rc = ELDLM_OK;
1666 struct ldlm_interval *node = NULL;
1669 lock->l_last_activity = cfs_time_current_sec();
1670 /* policies are not executed on the client or during replay */
1671 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1672 && !local && ns->ns_policy) {
1673 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1675 if (rc == ELDLM_LOCK_REPLACED) {
1676 /* The lock that was returned has already been granted,
1677 * and placed into lockp. If it's not the same as the
1678 * one we passed in, then destroy the old one and our
1679 * work here is done. */
1680 if (lock != *lockp) {
1681 ldlm_lock_destroy(lock);
1682 LDLM_LOCK_RELEASE(lock);
1684 *flags |= LDLM_FL_LOCK_CHANGED;
1686 } else if (rc != ELDLM_OK ||
1687 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1688 ldlm_lock_destroy(lock);
1693 /* For a replaying lock, it might be already in granted list. So
1694 * unlinking the lock will cause the interval node to be freed, we
1695 * have to allocate the interval node early otherwise we can't regrant
1696 * this lock in the future. - jay */
1697 if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1698 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, __GFP_IO);
1700 lock_res_and_lock(lock);
1701 if (local && lock->l_req_mode == lock->l_granted_mode) {
1702 /* The server returned a blocked lock, but it was granted
1703 * before we got a chance to actually enqueue it. We don't
1704 * need to do anything else. */
1705 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1706 LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1707 GOTO(out, ELDLM_OK);
1710 ldlm_resource_unlink_lock(lock);
1711 if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1713 ldlm_lock_destroy_nolock(lock);
1714 GOTO(out, rc = -ENOMEM);
1717 CFS_INIT_LIST_HEAD(&node->li_group);
1718 ldlm_interval_attach(node, lock);
1722 /* Some flags from the enqueue want to make it into the AST, via the
1723 * lock's l_flags. */
1724 lock->l_flags |= *flags & LDLM_FL_AST_DISCARD_DATA;
1726 /* This distinction between local lock trees is very important; a client
1727 * namespace only has information about locks taken by that client, and
1728 * thus doesn't have enough information to decide for itself if it can
1729 * be granted (below). In this case, we do exactly what the server
1730 * tells us to do, as dictated by the 'flags'.
1732 * We do exactly the same thing during recovery, when the server is
1733 * more or less trusting the clients not to lie.
1735 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1736 * granted/converting queues. */
1738 if (*flags & LDLM_FL_BLOCK_CONV)
1739 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1740 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1741 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1743 ldlm_grant_lock(lock, NULL);
1744 GOTO(out, ELDLM_OK);
1745 #ifdef HAVE_SERVER_SUPPORT
1746 } else if (*flags & LDLM_FL_REPLAY) {
1747 if (*flags & LDLM_FL_BLOCK_CONV) {
1748 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1749 GOTO(out, ELDLM_OK);
1750 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1751 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1752 GOTO(out, ELDLM_OK);
1753 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1754 ldlm_grant_lock(lock, NULL);
1755 GOTO(out, ELDLM_OK);
1757 /* If no flags, fall through to normal enqueue path. */
1760 policy = ldlm_processing_policy_table[res->lr_type];
1761 policy(lock, flags, 1, &rc, NULL);
1765 CERROR("This is client-side-only module, cannot handle "
1766 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1772 unlock_res_and_lock(lock);
1774 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1778 #ifdef HAVE_SERVER_SUPPORT
1780 * Iterate through all waiting locks on a given resource queue and attempt to
1783 * Must be called with resource lock held.
1785 int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue,
1786 cfs_list_t *work_list)
1788 cfs_list_t *tmp, *pos;
1789 ldlm_processing_policy policy;
1791 int rc = LDLM_ITER_CONTINUE;
1795 check_res_locked(res);
1797 policy = ldlm_processing_policy_table[res->lr_type];
1800 cfs_list_for_each_safe(tmp, pos, queue) {
1801 struct ldlm_lock *pending;
1802 pending = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1804 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1807 rc = policy(pending, &flags, 0, &err, work_list);
1808 if (rc != LDLM_ITER_CONTINUE)
1817 * Process a call to blocking AST callback for a lock in ast_work list
1820 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1822 struct ldlm_cb_set_arg *arg = opaq;
1823 struct ldlm_lock_desc d;
1825 struct ldlm_lock *lock;
1828 if (cfs_list_empty(arg->list))
1831 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1833 /* nobody should touch l_bl_ast */
1834 lock_res_and_lock(lock);
1835 cfs_list_del_init(&lock->l_bl_ast);
1837 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1838 LASSERT(lock->l_bl_ast_run == 0);
1839 LASSERT(lock->l_blocking_lock);
1840 lock->l_bl_ast_run++;
1841 unlock_res_and_lock(lock);
1843 ldlm_lock2desc(lock->l_blocking_lock, &d);
1845 rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1846 LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1847 lock->l_blocking_lock = NULL;
1848 LDLM_LOCK_RELEASE(lock);
1854 * Process a call to completion AST callback for a lock in ast_work list
1857 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1859 struct ldlm_cb_set_arg *arg = opaq;
1861 struct ldlm_lock *lock;
1862 ldlm_completion_callback completion_callback;
1865 if (cfs_list_empty(arg->list))
1868 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1870 /* It's possible to receive a completion AST before we've set
1871 * the l_completion_ast pointer: either because the AST arrived
1872 * before the reply, or simply because there's a small race
1873 * window between receiving the reply and finishing the local
1874 * enqueue. (bug 842)
1876 * This can't happen with the blocking_ast, however, because we
1877 * will never call the local blocking_ast until we drop our
1878 * reader/writer reference, which we won't do until we get the
1879 * reply and finish enqueueing. */
1881 /* nobody should touch l_cp_ast */
1882 lock_res_and_lock(lock);
1883 cfs_list_del_init(&lock->l_cp_ast);
1884 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1885 /* save l_completion_ast since it can be changed by
1886 * mds_intent_policy(), see bug 14225 */
1887 completion_callback = lock->l_completion_ast;
1888 lock->l_flags &= ~LDLM_FL_CP_REQD;
1889 unlock_res_and_lock(lock);
1891 if (completion_callback != NULL)
1892 rc = completion_callback(lock, 0, (void *)arg);
1893 LDLM_LOCK_RELEASE(lock);
1899 * Process a call to revocation AST callback for a lock in ast_work list
1902 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1904 struct ldlm_cb_set_arg *arg = opaq;
1905 struct ldlm_lock_desc desc;
1907 struct ldlm_lock *lock;
1910 if (cfs_list_empty(arg->list))
1913 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1914 cfs_list_del_init(&lock->l_rk_ast);
1916 /* the desc just pretend to exclusive */
1917 ldlm_lock2desc(lock, &desc);
1918 desc.l_req_mode = LCK_EX;
1919 desc.l_granted_mode = 0;
1921 rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1922 LDLM_LOCK_RELEASE(lock);
1928 * Process a call to glimpse AST callback for a lock in ast_work list
1930 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1932 struct ldlm_cb_set_arg *arg = opaq;
1933 struct ldlm_glimpse_work *gl_work;
1934 struct ldlm_lock *lock;
1938 if (cfs_list_empty(arg->list))
1941 gl_work = cfs_list_entry(arg->list->next, struct ldlm_glimpse_work,
1943 cfs_list_del_init(&gl_work->gl_list);
1945 lock = gl_work->gl_lock;
1947 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
1948 arg->gl_desc = gl_work->gl_desc;
1950 /* invoke the actual glimpse callback */
1951 if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
1954 LDLM_LOCK_RELEASE(lock);
1956 if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
1957 OBD_FREE_PTR(gl_work);
1963 * Process list of locks in need of ASTs being sent.
1965 * Used on server to send multiple ASTs together instead of sending one by
1968 int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list,
1969 ldlm_desc_ast_t ast_type)
1971 struct ldlm_cb_set_arg *arg;
1972 set_producer_func work_ast_lock;
1975 if (cfs_list_empty(rpc_list))
1982 cfs_atomic_set(&arg->restart, 0);
1983 arg->list = rpc_list;
1986 case LDLM_WORK_BL_AST:
1987 arg->type = LDLM_BL_CALLBACK;
1988 work_ast_lock = ldlm_work_bl_ast_lock;
1990 case LDLM_WORK_CP_AST:
1991 arg->type = LDLM_CP_CALLBACK;
1992 work_ast_lock = ldlm_work_cp_ast_lock;
1994 case LDLM_WORK_REVOKE_AST:
1995 arg->type = LDLM_BL_CALLBACK;
1996 work_ast_lock = ldlm_work_revoke_ast_lock;
1998 case LDLM_WORK_GL_AST:
1999 arg->type = LDLM_GL_CALLBACK;
2000 work_ast_lock = ldlm_work_gl_ast_lock;
2006 /* We create a ptlrpc request set with flow control extension.
2007 * This request set will use the work_ast_lock function to produce new
2008 * requests and will send a new request each time one completes in order
2009 * to keep the number of requests in flight to ns_max_parallel_ast */
2010 arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
2011 work_ast_lock, arg);
2012 if (arg->set == NULL)
2013 GOTO(out, rc = -ENOMEM);
2015 ptlrpc_set_wait(arg->set);
2016 ptlrpc_set_destroy(arg->set);
2018 rc = cfs_atomic_read(&arg->restart) ? -ERESTART : 0;
2025 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
2027 ldlm_reprocess_all(res);
2028 return LDLM_ITER_CONTINUE;
2031 static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2032 cfs_hlist_node_t *hnode, void *arg)
2034 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2037 rc = reprocess_one_queue(res, arg);
2039 return rc == LDLM_ITER_STOP;
2043 * Iterate through all resources on a namespace attempting to grant waiting
2046 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
2051 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2052 ldlm_reprocess_res, NULL);
2056 EXPORT_SYMBOL(ldlm_reprocess_all_ns);
2059 * Try to grant all waiting locks on a resource.
2061 * Calls ldlm_reprocess_queue on converting and waiting queues.
2063 * Typically called after some resource locks are cancelled to see
2064 * if anything could be granted as a result of the cancellation.
2066 void ldlm_reprocess_all(struct ldlm_resource *res)
2068 CFS_LIST_HEAD(rpc_list);
2070 #ifdef HAVE_SERVER_SUPPORT
2073 /* Local lock trees don't get reprocessed. */
2074 if (ns_is_client(ldlm_res_to_ns(res))) {
2081 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
2082 if (rc == LDLM_ITER_CONTINUE)
2083 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
2086 rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
2088 if (rc == -ERESTART) {
2089 LASSERT(cfs_list_empty(&rpc_list));
2094 if (!ns_is_client(ldlm_res_to_ns(res))) {
2095 CERROR("This is client-side-only module, cannot handle "
2096 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2104 * Helper function to call blocking AST for LDLM lock \a lock in a
2105 * "cancelling" mode.
2107 void ldlm_cancel_callback(struct ldlm_lock *lock)
2109 check_res_locked(lock->l_resource);
2110 if (!(lock->l_flags & LDLM_FL_CANCEL)) {
2111 lock->l_flags |= LDLM_FL_CANCEL;
2112 if (lock->l_blocking_ast) {
2113 unlock_res_and_lock(lock);
2114 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
2116 lock_res_and_lock(lock);
2118 LDLM_DEBUG(lock, "no blocking ast");
2121 lock->l_flags |= LDLM_FL_BL_DONE;
2125 * Remove skiplist-enabled LDLM lock \a req from granted list
2127 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
2129 if (req->l_resource->lr_type != LDLM_PLAIN &&
2130 req->l_resource->lr_type != LDLM_IBITS)
2133 cfs_list_del_init(&req->l_sl_policy);
2134 cfs_list_del_init(&req->l_sl_mode);
2138 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2140 void ldlm_lock_cancel(struct ldlm_lock *lock)
2142 struct ldlm_resource *res;
2143 struct ldlm_namespace *ns;
2146 lock_res_and_lock(lock);
2148 res = lock->l_resource;
2149 ns = ldlm_res_to_ns(res);
2151 /* Please do not, no matter how tempting, remove this LBUG without
2152 * talking to me first. -phik */
2153 if (lock->l_readers || lock->l_writers) {
2154 LDLM_ERROR(lock, "lock still has references");
2158 if (lock->l_flags & LDLM_FL_WAITED)
2159 ldlm_del_waiting_lock(lock);
2161 /* Releases cancel callback. */
2162 ldlm_cancel_callback(lock);
2164 /* Yes, second time, just in case it was added again while we were
2165 * running with no res lock in ldlm_cancel_callback */
2166 if (lock->l_flags & LDLM_FL_WAITED)
2167 ldlm_del_waiting_lock(lock);
2169 ldlm_resource_unlink_lock(lock);
2170 ldlm_lock_destroy_nolock(lock);
2172 if (lock->l_granted_mode == lock->l_req_mode)
2173 ldlm_pool_del(&ns->ns_pool, lock);
2175 /* Make sure we will not be called again for same lock what is possible
2176 * if not to zero out lock->l_granted_mode */
2177 lock->l_granted_mode = LCK_MINMODE;
2178 unlock_res_and_lock(lock);
2182 EXPORT_SYMBOL(ldlm_lock_cancel);
2185 * Set opaque data into the lock that only makes sense to upper layer.
2187 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
2189 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2194 if (lock->l_ast_data == NULL)
2195 lock->l_ast_data = data;
2196 if (lock->l_ast_data == data)
2198 LDLM_LOCK_PUT(lock);
2202 EXPORT_SYMBOL(ldlm_lock_set_data);
2204 struct export_cl_data {
2205 struct obd_export *ecl_exp;
2210 * Iterator function for ldlm_cancel_locks_for_export.
2211 * Cancels passed locks.
2213 int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2214 cfs_hlist_node_t *hnode, void *data)
2217 struct export_cl_data *ecl = (struct export_cl_data *)data;
2218 struct obd_export *exp = ecl->ecl_exp;
2219 struct ldlm_lock *lock = cfs_hash_object(hs, hnode);
2220 struct ldlm_resource *res;
2222 res = ldlm_resource_getref(lock->l_resource);
2223 LDLM_LOCK_GET(lock);
2225 LDLM_DEBUG(lock, "export %p", exp);
2226 ldlm_res_lvbo_update(res, NULL, 1);
2227 ldlm_lock_cancel(lock);
2228 ldlm_reprocess_all(res);
2229 ldlm_resource_putref(res);
2230 LDLM_LOCK_RELEASE(lock);
2233 if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
2235 "Cancel lock %p for export %p (loop %d), still have "
2236 "%d locks left on hash table.\n",
2237 lock, exp, ecl->ecl_loop,
2238 cfs_atomic_read(&hs->hs_count));
2245 * Cancel all locks for given export.
2247 * Typically called on client disconnection/eviction
2249 void ldlm_cancel_locks_for_export(struct obd_export *exp)
2251 struct export_cl_data ecl = {
2256 cfs_hash_for_each_empty(exp->exp_lock_hash,
2257 ldlm_cancel_locks_for_export_cb, &ecl);
2261 * Downgrade an exclusive lock.
2263 * A fast variant of ldlm_lock_convert for convertion of exclusive
2264 * locks. The convertion is always successful.
2265 * Used by Commit on Sharing (COS) code.
2267 * \param lock A lock to convert
2268 * \param new_mode new lock mode
2270 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
2274 LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
2275 LASSERT(new_mode == LCK_COS);
2277 lock_res_and_lock(lock);
2278 ldlm_resource_unlink_lock(lock);
2280 * Remove the lock from pool as it will be added again in
2281 * ldlm_grant_lock() called below.
2283 ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
2285 lock->l_req_mode = new_mode;
2286 ldlm_grant_lock(lock, NULL);
2287 unlock_res_and_lock(lock);
2288 ldlm_reprocess_all(lock->l_resource);
2292 EXPORT_SYMBOL(ldlm_lock_downgrade);
2295 * Attempt to convert already granted lock to a different mode.
2297 * While lock conversion is not currently used, future client-side
2298 * optimizations could take advantage of it to avoid discarding cached
2301 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
2304 CFS_LIST_HEAD(rpc_list);
2305 struct ldlm_resource *res;
2306 struct ldlm_namespace *ns;
2308 #ifdef HAVE_SERVER_SUPPORT
2310 struct sl_insert_point prev;
2312 struct ldlm_interval *node;
2315 /* Just return if mode is unchanged. */
2316 if (new_mode == lock->l_granted_mode) {
2317 *flags |= LDLM_FL_BLOCK_GRANTED;
2318 RETURN(lock->l_resource);
2321 /* I can't check the type of lock here because the bitlock of lock
2322 * is not held here, so do the allocation blindly. -jay */
2323 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, __GFP_IO);
2324 if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */
2327 LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
2328 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2330 lock_res_and_lock(lock);
2332 res = lock->l_resource;
2333 ns = ldlm_res_to_ns(res);
2335 #ifdef HAVE_SERVER_SUPPORT
2336 old_mode = lock->l_req_mode;
2338 lock->l_req_mode = new_mode;
2339 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2340 #ifdef HAVE_SERVER_SUPPORT
2341 /* remember the lock position where the lock might be
2342 * added back to the granted list later and also
2343 * remember the join mode for skiplist fixing. */
2344 prev.res_link = lock->l_res_link.prev;
2345 prev.mode_link = lock->l_sl_mode.prev;
2346 prev.policy_link = lock->l_sl_policy.prev;
2348 ldlm_resource_unlink_lock(lock);
2350 ldlm_resource_unlink_lock(lock);
2351 if (res->lr_type == LDLM_EXTENT) {
2352 /* FIXME: ugly code, I have to attach the lock to a
2353 * interval node again since perhaps it will be granted
2355 CFS_INIT_LIST_HEAD(&node->li_group);
2356 ldlm_interval_attach(node, lock);
2362 * Remove old lock from the pool before adding the lock with new
2363 * mode below in ->policy()
2365 ldlm_pool_del(&ns->ns_pool, lock);
2367 /* If this is a local resource, put it on the appropriate list. */
2368 if (ns_is_client(ldlm_res_to_ns(res))) {
2369 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2370 ldlm_resource_add_lock(res, &res->lr_converting, lock);
2372 /* This should never happen, because of the way the
2373 * server handles conversions. */
2374 LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2378 ldlm_grant_lock(lock, &rpc_list);
2380 /* FIXME: completion handling not with lr_lock held ! */
2381 if (lock->l_completion_ast)
2382 lock->l_completion_ast(lock, 0, NULL);
2384 #ifdef HAVE_SERVER_SUPPORT
2389 ldlm_processing_policy policy;
2390 policy = ldlm_processing_policy_table[res->lr_type];
2391 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2392 if (rc == LDLM_ITER_STOP) {
2393 lock->l_req_mode = old_mode;
2394 if (res->lr_type == LDLM_EXTENT)
2395 ldlm_extent_add_lock(res, lock);
2397 ldlm_granted_list_add_lock(lock, &prev);
2401 *flags |= LDLM_FL_BLOCK_GRANTED;
2407 CERROR("This is client-side-only module, cannot handle "
2408 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2412 unlock_res_and_lock(lock);
2415 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2417 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2420 EXPORT_SYMBOL(ldlm_lock_convert);
2423 * Print lock with lock handle \a lockh description into debug log.
2425 * Used when printing all locks on a resource for debug purposes.
2427 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
2429 struct ldlm_lock *lock;
2431 if (!((libcfs_debug | D_ERROR) & level))
2434 lock = ldlm_handle2lock(lockh);
2438 LDLM_DEBUG_LIMIT(level, lock, "###");
2440 LDLM_LOCK_PUT(lock);
2442 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2445 * Print lock information with custom message into debug log.
2448 void _ldlm_lock_debug(struct ldlm_lock *lock,
2449 struct libcfs_debug_msg_data *msgdata,
2450 const char *fmt, ...)
2453 struct obd_export *exp = lock->l_export;
2454 struct ldlm_resource *resource = lock->l_resource;
2455 char *nid = "local";
2457 va_start(args, fmt);
2459 if (exp && exp->exp_connection) {
2460 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2461 } else if (exp && exp->exp_obd != NULL) {
2462 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2463 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2466 if (resource == NULL) {
2467 libcfs_debug_vmsg2(msgdata, fmt, args,
2468 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2469 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2470 "remote: "LPX64" expref: %d pid: %u timeout: %lu "
2473 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2474 lock->l_readers, lock->l_writers,
2475 ldlm_lockname[lock->l_granted_mode],
2476 ldlm_lockname[lock->l_req_mode],
2477 lock->l_flags, nid, lock->l_remote_handle.cookie,
2478 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2479 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2484 switch (resource->lr_type) {
2486 libcfs_debug_vmsg2(msgdata, fmt, args,
2487 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2488 "res: "DLDLMRES" rrc: %d type: %s ["LPU64"->"LPU64"] "
2489 "(req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote: "
2490 LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2491 ldlm_lock_to_ns_name(lock), lock,
2492 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2493 lock->l_readers, lock->l_writers,
2494 ldlm_lockname[lock->l_granted_mode],
2495 ldlm_lockname[lock->l_req_mode],
2497 cfs_atomic_read(&resource->lr_refcount),
2498 ldlm_typename[resource->lr_type],
2499 lock->l_policy_data.l_extent.start,
2500 lock->l_policy_data.l_extent.end,
2501 lock->l_req_extent.start, lock->l_req_extent.end,
2502 lock->l_flags, nid, lock->l_remote_handle.cookie,
2503 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2504 lock->l_pid, lock->l_callback_timeout,
2509 libcfs_debug_vmsg2(msgdata, fmt, args,
2510 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2511 "res: "DLDLMRES" rrc: %d type: %s pid: %d "
2512 "["LPU64"->"LPU64"] flags: "LPX64" nid: %s "
2513 "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2514 ldlm_lock_to_ns_name(lock), lock,
2515 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2516 lock->l_readers, lock->l_writers,
2517 ldlm_lockname[lock->l_granted_mode],
2518 ldlm_lockname[lock->l_req_mode],
2520 cfs_atomic_read(&resource->lr_refcount),
2521 ldlm_typename[resource->lr_type],
2522 lock->l_policy_data.l_flock.pid,
2523 lock->l_policy_data.l_flock.start,
2524 lock->l_policy_data.l_flock.end,
2525 lock->l_flags, nid, lock->l_remote_handle.cookie,
2526 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2527 lock->l_pid, lock->l_callback_timeout);
2531 libcfs_debug_vmsg2(msgdata, fmt, args,
2532 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2533 "res: "DLDLMRES" bits "LPX64" rrc: %d type: %s "
2534 "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2535 "pid: %u timeout: %lu lvb_type: %d\n",
2536 ldlm_lock_to_ns_name(lock),
2537 lock, lock->l_handle.h_cookie,
2538 cfs_atomic_read(&lock->l_refc),
2539 lock->l_readers, lock->l_writers,
2540 ldlm_lockname[lock->l_granted_mode],
2541 ldlm_lockname[lock->l_req_mode],
2543 lock->l_policy_data.l_inodebits.bits,
2544 cfs_atomic_read(&resource->lr_refcount),
2545 ldlm_typename[resource->lr_type],
2546 lock->l_flags, nid, lock->l_remote_handle.cookie,
2547 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2548 lock->l_pid, lock->l_callback_timeout,
2553 libcfs_debug_vmsg2(msgdata, fmt, args,
2554 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2555 "res: "DLDLMRES" rrc: %d type: %s flags: "LPX64" "
2556 "nid: %s remote: "LPX64" expref: %d pid: %u "
2557 "timeout: %lu lvb_type: %d\n",
2558 ldlm_lock_to_ns_name(lock),
2559 lock, lock->l_handle.h_cookie,
2560 cfs_atomic_read(&lock->l_refc),
2561 lock->l_readers, lock->l_writers,
2562 ldlm_lockname[lock->l_granted_mode],
2563 ldlm_lockname[lock->l_req_mode],
2565 cfs_atomic_read(&resource->lr_refcount),
2566 ldlm_typename[resource->lr_type],
2567 lock->l_flags, nid, lock->l_remote_handle.cookie,
2568 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2569 lock->l_pid, lock->l_callback_timeout,
2575 EXPORT_SYMBOL(_ldlm_lock_debug);