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):
153 CERROR("Unknown intent %d\n", it);
157 EXPORT_SYMBOL(ldlm_it2str);
159 extern cfs_mem_cache_t *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 cfs_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(cfs_atomic_read(&lock->l_refc) > 0);
215 if (cfs_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(lock->l_destroyed);
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(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 if (lock->l_flags & LDLM_FL_SKIPPED)
260 lock->l_flags &= ~LDLM_FL_SKIPPED;
261 LASSERT(ns->ns_nr_unused > 0);
269 * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first.
271 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
273 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
277 if (lock->l_ns_srv) {
278 LASSERT(cfs_list_empty(&lock->l_lru));
282 spin_lock(&ns->ns_lock);
283 rc = ldlm_lock_remove_from_lru_nolock(lock);
284 spin_unlock(&ns->ns_lock);
290 * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked.
292 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
294 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
296 lock->l_last_used = cfs_time_current();
297 LASSERT(cfs_list_empty(&lock->l_lru));
298 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
299 cfs_list_add_tail(&lock->l_lru, &ns->ns_unused_list);
300 LASSERT(ns->ns_nr_unused >= 0);
305 * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks
308 void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
310 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
313 spin_lock(&ns->ns_lock);
314 ldlm_lock_add_to_lru_nolock(lock);
315 spin_unlock(&ns->ns_lock);
320 * Moves LDLM lock \a lock that is already in namespace LRU to the tail of
321 * the LRU. Performs necessary LRU locking
323 void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
325 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
328 if (lock->l_ns_srv) {
329 LASSERT(cfs_list_empty(&lock->l_lru));
334 spin_lock(&ns->ns_lock);
335 if (!cfs_list_empty(&lock->l_lru)) {
336 ldlm_lock_remove_from_lru_nolock(lock);
337 ldlm_lock_add_to_lru_nolock(lock);
339 spin_unlock(&ns->ns_lock);
344 * Helper to destroy a locked lock.
346 * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock
347 * Must be called with l_lock and lr_lock held.
349 * Does not actually free the lock data, but rather marks the lock as
350 * destroyed by setting l_destroyed field in the lock to 1. Destroys a
351 * handle->lock association too, so that the lock can no longer be found
352 * and removes the lock from LRU list. Actual lock freeing occurs when
353 * last lock reference goes away.
355 * Original comment (of some historical value):
356 * This used to have a 'strict' flag, which recovery would use to mark an
357 * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
358 * shall explain why it's gone: with the new hash table scheme, once you call
359 * ldlm_lock_destroy, you can never drop your final references on this lock.
360 * Because it's not in the hash table anymore. -phil
362 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
366 if (lock->l_readers || lock->l_writers) {
367 LDLM_ERROR(lock, "lock still has references");
371 if (!cfs_list_empty(&lock->l_res_link)) {
372 LDLM_ERROR(lock, "lock still on resource");
376 if (lock->l_destroyed) {
377 LASSERT(cfs_list_empty(&lock->l_lru));
381 lock->l_destroyed = 1;
383 if (lock->l_export && lock->l_export->exp_lock_hash) {
384 /* NB: it's safe to call cfs_hash_del() even lock isn't
385 * in exp_lock_hash. */
386 /* In the function below, .hs_keycmp resolves to
387 * ldlm_export_lock_keycmp() */
388 /* coverity[overrun-buffer-val] */
389 cfs_hash_del(lock->l_export->exp_lock_hash,
390 &lock->l_remote_handle, &lock->l_exp_hash);
393 ldlm_lock_remove_from_lru(lock);
394 class_handle_unhash(&lock->l_handle);
397 /* Wake anyone waiting for this lock */
398 /* FIXME: I should probably add yet another flag, instead of using
399 * l_export to only call this on clients */
401 class_export_put(lock->l_export);
402 lock->l_export = NULL;
403 if (lock->l_export && lock->l_completion_ast)
404 lock->l_completion_ast(lock, 0);
411 * Destroys a LDLM lock \a lock. Performs necessary locking first.
413 void ldlm_lock_destroy(struct ldlm_lock *lock)
417 lock_res_and_lock(lock);
418 first = ldlm_lock_destroy_internal(lock);
419 unlock_res_and_lock(lock);
421 /* drop reference from hashtable only for first destroy */
423 lu_ref_del(&lock->l_reference, "hash", lock);
424 LDLM_LOCK_RELEASE(lock);
430 * Destroys a LDLM lock \a lock that is already locked.
432 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
436 first = ldlm_lock_destroy_internal(lock);
437 /* drop reference from hashtable only for first destroy */
439 lu_ref_del(&lock->l_reference, "hash", lock);
440 LDLM_LOCK_RELEASE(lock);
445 /* this is called by portals_handle2object with the handle lock taken */
446 static void lock_handle_addref(void *lock)
448 LDLM_LOCK_GET((struct ldlm_lock *)lock);
451 static void lock_handle_free(void *lock, int size)
453 LASSERT(size == sizeof(struct ldlm_lock));
454 OBD_SLAB_FREE(lock, ldlm_lock_slab, size);
457 struct portals_handle_ops lock_handle_ops = {
458 .hop_addref = lock_handle_addref,
459 .hop_free = lock_handle_free,
464 * Allocate and initialize new lock structure.
466 * usage: pass in a resource on which you have done ldlm_resource_get
467 * new lock will take over the refcount.
468 * returns: lock with refcount 2 - one for current caller and one for remote
470 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
472 struct ldlm_lock *lock;
475 if (resource == NULL)
478 OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, CFS_ALLOC_IO);
482 spin_lock_init(&lock->l_lock);
483 lock->l_resource = resource;
484 lu_ref_add(&resource->lr_reference, "lock", lock);
486 cfs_atomic_set(&lock->l_refc, 2);
487 CFS_INIT_LIST_HEAD(&lock->l_res_link);
488 CFS_INIT_LIST_HEAD(&lock->l_lru);
489 CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
490 CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
491 CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
492 CFS_INIT_LIST_HEAD(&lock->l_rk_ast);
493 cfs_waitq_init(&lock->l_waitq);
494 lock->l_blocking_lock = NULL;
495 CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
496 CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
497 CFS_INIT_HLIST_NODE(&lock->l_exp_hash);
498 CFS_INIT_HLIST_NODE(&lock->l_exp_flock_hash);
500 lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats,
502 CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
503 class_handle_hash(&lock->l_handle, &lock_handle_ops);
505 lu_ref_init(&lock->l_reference);
506 lu_ref_add(&lock->l_reference, "hash", lock);
507 lock->l_callback_timeout = 0;
509 #if LUSTRE_TRACKS_LOCK_EXP_REFS
510 CFS_INIT_LIST_HEAD(&lock->l_exp_refs_link);
511 lock->l_exp_refs_nr = 0;
512 lock->l_exp_refs_target = NULL;
514 CFS_INIT_LIST_HEAD(&lock->l_exp_list);
520 * Moves LDLM lock \a lock to another resource.
521 * This is used on client when server returns some other lock than requested
522 * (typically as a result of intent operation)
524 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
525 const struct ldlm_res_id *new_resid)
527 struct ldlm_resource *oldres = lock->l_resource;
528 struct ldlm_resource *newres;
532 LASSERT(ns_is_client(ns));
534 lock_res_and_lock(lock);
535 if (memcmp(new_resid, &lock->l_resource->lr_name,
536 sizeof(lock->l_resource->lr_name)) == 0) {
538 unlock_res_and_lock(lock);
542 LASSERT(new_resid->name[0] != 0);
544 /* This function assumes that the lock isn't on any lists */
545 LASSERT(cfs_list_empty(&lock->l_res_link));
547 type = oldres->lr_type;
548 unlock_res_and_lock(lock);
550 newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
554 lu_ref_add(&newres->lr_reference, "lock", lock);
556 * To flip the lock from the old to the new resource, lock, oldres and
557 * newres have to be locked. Resource spin-locks are nested within
558 * lock->l_lock, and are taken in the memory address order to avoid
561 spin_lock(&lock->l_lock);
562 oldres = lock->l_resource;
563 if (oldres < newres) {
565 lock_res_nested(newres, LRT_NEW);
568 lock_res_nested(oldres, LRT_NEW);
570 LASSERT(memcmp(new_resid, &oldres->lr_name,
571 sizeof oldres->lr_name) != 0);
572 lock->l_resource = newres;
574 unlock_res_and_lock(lock);
576 /* ...and the flowers are still standing! */
577 lu_ref_del(&oldres->lr_reference, "lock", lock);
578 ldlm_resource_putref(oldres);
582 EXPORT_SYMBOL(ldlm_lock_change_resource);
584 /** \defgroup ldlm_handles LDLM HANDLES
585 * Ways to get hold of locks without any addresses.
590 * Fills in handle for LDLM lock \a lock into supplied \a lockh
591 * Does not take any references.
593 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
595 lockh->cookie = lock->l_handle.h_cookie;
597 EXPORT_SYMBOL(ldlm_lock2handle);
600 * Obtain a lock reference by handle.
602 * if \a flags: atomically get the lock and set the flags.
603 * Return NULL if flag already set
605 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
608 struct ldlm_lock *lock;
613 lock = class_handle2object(handle->cookie);
617 /* It's unlikely but possible that someone marked the lock as
618 * destroyed after we did handle2object on it */
619 if (flags == 0 && !lock->l_destroyed) {
620 lu_ref_add(&lock->l_reference, "handle", cfs_current());
624 lock_res_and_lock(lock);
626 LASSERT(lock->l_resource != NULL);
628 lu_ref_add_atomic(&lock->l_reference, "handle", cfs_current());
629 if (unlikely(lock->l_destroyed)) {
630 unlock_res_and_lock(lock);
631 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
636 if (flags && (lock->l_flags & flags)) {
637 unlock_res_and_lock(lock);
643 lock->l_flags |= flags;
645 unlock_res_and_lock(lock);
648 EXPORT_SYMBOL(__ldlm_handle2lock);
649 /** @} ldlm_handles */
652 * Fill in "on the wire" representation for given LDLM lock into supplied
653 * lock descriptor \a desc structure.
655 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
657 struct obd_export *exp = lock->l_export ?: lock->l_conn_export;
659 /* INODEBITS_INTEROP: If the other side does not support
660 * inodebits, reply with a plain lock descriptor. */
661 if ((lock->l_resource->lr_type == LDLM_IBITS) &&
662 (exp && !(exp_connect_flags(exp) & OBD_CONNECT_IBITS))) {
663 /* Make sure all the right bits are set in this lock we
664 are going to pass to client */
665 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
666 (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
667 MDS_INODELOCK_LAYOUT),
668 "Inappropriate inode lock bits during "
669 "conversion " LPU64 "\n",
670 lock->l_policy_data.l_inodebits.bits);
672 ldlm_res2desc(lock->l_resource, &desc->l_resource);
673 desc->l_resource.lr_type = LDLM_PLAIN;
675 /* Convert "new" lock mode to something old client can
677 if ((lock->l_req_mode == LCK_CR) ||
678 (lock->l_req_mode == LCK_CW))
679 desc->l_req_mode = LCK_PR;
681 desc->l_req_mode = lock->l_req_mode;
682 if ((lock->l_granted_mode == LCK_CR) ||
683 (lock->l_granted_mode == LCK_CW)) {
684 desc->l_granted_mode = LCK_PR;
686 /* We never grant PW/EX locks to clients */
687 LASSERT((lock->l_granted_mode != LCK_PW) &&
688 (lock->l_granted_mode != LCK_EX));
689 desc->l_granted_mode = lock->l_granted_mode;
692 /* We do not copy policy here, because there is no
693 policy for plain locks */
695 ldlm_res2desc(lock->l_resource, &desc->l_resource);
696 desc->l_req_mode = lock->l_req_mode;
697 desc->l_granted_mode = lock->l_granted_mode;
698 ldlm_convert_policy_to_wire(lock->l_resource->lr_type,
699 &lock->l_policy_data,
700 &desc->l_policy_data);
703 EXPORT_SYMBOL(ldlm_lock2desc);
706 * Add a lock to list of conflicting locks to send AST to.
708 * Only add if we have not sent a blocking AST to the lock yet.
710 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
711 cfs_list_t *work_list)
713 if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
714 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
715 lock->l_flags |= LDLM_FL_AST_SENT;
716 /* If the enqueuing client said so, tell the AST recipient to
717 * discard dirty data, rather than writing back. */
718 if (new->l_flags & LDLM_AST_DISCARD_DATA)
719 lock->l_flags |= LDLM_FL_DISCARD_DATA;
720 LASSERT(cfs_list_empty(&lock->l_bl_ast));
721 cfs_list_add(&lock->l_bl_ast, work_list);
723 LASSERT(lock->l_blocking_lock == NULL);
724 lock->l_blocking_lock = LDLM_LOCK_GET(new);
729 * Add a lock to list of just granted locks to send completion AST to.
731 void ldlm_add_cp_work_item(struct ldlm_lock *lock, cfs_list_t *work_list)
733 if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
734 lock->l_flags |= LDLM_FL_CP_REQD;
735 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
736 LASSERT(cfs_list_empty(&lock->l_cp_ast));
737 cfs_list_add(&lock->l_cp_ast, work_list);
743 * Aggregator function to add AST work items into a list. Determines
744 * what sort of an AST work needs to be done and calls the proper
746 * Must be called with lr_lock held.
748 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
749 cfs_list_t *work_list)
752 check_res_locked(lock->l_resource);
754 ldlm_add_bl_work_item(lock, new, work_list);
756 ldlm_add_cp_work_item(lock, work_list);
761 * Add specified reader/writer reference to LDLM lock with handle \a lockh.
762 * r/w reference type is determined by \a mode
763 * Calls ldlm_lock_addref_internal.
765 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
767 struct ldlm_lock *lock;
769 lock = ldlm_handle2lock(lockh);
770 LASSERT(lock != NULL);
771 ldlm_lock_addref_internal(lock, mode);
774 EXPORT_SYMBOL(ldlm_lock_addref);
778 * Add specified reader/writer reference to LDLM lock \a lock.
779 * r/w reference type is determined by \a mode
780 * Removes lock from LRU if it is there.
781 * Assumes the LDLM lock is already locked.
783 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
785 ldlm_lock_remove_from_lru(lock);
786 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
788 lu_ref_add_atomic(&lock->l_reference, "reader", lock);
790 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
792 lu_ref_add_atomic(&lock->l_reference, "writer", lock);
795 lu_ref_add_atomic(&lock->l_reference, "user", lock);
796 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
800 * Attempts to add reader/writer reference to a lock with handle \a lockh, and
801 * fails if lock is already LDLM_FL_CBPENDING or destroyed.
803 * \retval 0 success, lock was addref-ed
805 * \retval -EAGAIN lock is being canceled.
807 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
809 struct ldlm_lock *lock;
813 lock = ldlm_handle2lock(lockh);
815 lock_res_and_lock(lock);
816 if (lock->l_readers != 0 || lock->l_writers != 0 ||
817 !(lock->l_flags & LDLM_FL_CBPENDING)) {
818 ldlm_lock_addref_internal_nolock(lock, mode);
821 unlock_res_and_lock(lock);
826 EXPORT_SYMBOL(ldlm_lock_addref_try);
829 * Add specified reader/writer reference to LDLM lock \a lock.
830 * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work.
831 * Only called for local locks.
833 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
835 lock_res_and_lock(lock);
836 ldlm_lock_addref_internal_nolock(lock, mode);
837 unlock_res_and_lock(lock);
841 * Removes reader/writer reference for LDLM lock \a lock.
842 * Assumes LDLM lock is already locked.
843 * only called in ldlm_flock_destroy and for local locks.
844 * Does NOT add lock to LRU if no r/w references left to accomodate flock locks
845 * that cannot be placed in LRU.
847 void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
849 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
850 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
851 LASSERT(lock->l_readers > 0);
852 lu_ref_del(&lock->l_reference, "reader", lock);
855 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
856 LASSERT(lock->l_writers > 0);
857 lu_ref_del(&lock->l_reference, "writer", lock);
861 lu_ref_del(&lock->l_reference, "user", lock);
862 LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */
866 * Removes reader/writer reference for LDLM lock \a lock.
867 * Locks LDLM lock first.
868 * If the lock is determined to be client lock on a client and r/w refcount
869 * drops to zero and the lock is not blocked, the lock is added to LRU lock
871 * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called.
873 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
875 struct ldlm_namespace *ns;
878 lock_res_and_lock(lock);
880 ns = ldlm_lock_to_ns(lock);
882 ldlm_lock_decref_internal_nolock(lock, mode);
884 /* release lvb data for layout lock */
885 if (ns_is_client(ns) && !lock->l_readers && !lock->l_writers &&
886 ldlm_has_layout(lock) && lock->l_flags & LDLM_FL_LVB_READY) {
887 /* this is the last user of a layout lock and stripe has
888 * been set up, lvb is no longer used.
889 * This may be a large amount of memory, so we should free it
891 if (lock->l_lvb_data != NULL) {
892 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
893 lock->l_lvb_data = NULL;
898 if (lock->l_flags & LDLM_FL_LOCAL &&
899 !lock->l_readers && !lock->l_writers) {
900 /* If this is a local lock on a server namespace and this was
901 * the last reference, cancel the lock. */
902 CDEBUG(D_INFO, "forcing cancel of local lock\n");
903 lock->l_flags |= LDLM_FL_CBPENDING;
906 if (!lock->l_readers && !lock->l_writers &&
907 (lock->l_flags & LDLM_FL_CBPENDING)) {
908 /* If we received a blocked AST and this was the last reference,
909 * run the callback. */
910 if (lock->l_ns_srv && lock->l_export)
911 CERROR("FL_CBPENDING set on non-local lock--just a "
914 LDLM_DEBUG(lock, "final decref done on cbpending lock");
916 LDLM_LOCK_GET(lock); /* dropped by bl thread */
917 ldlm_lock_remove_from_lru(lock);
918 unlock_res_and_lock(lock);
920 if (lock->l_flags & LDLM_FL_FAIL_LOC)
921 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
923 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
924 ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
925 ldlm_handle_bl_callback(ns, NULL, lock);
926 } else if (ns_is_client(ns) &&
927 !lock->l_readers && !lock->l_writers &&
928 !(lock->l_flags & LDLM_FL_NO_LRU) &&
929 !(lock->l_flags & LDLM_FL_BL_AST)) {
931 LDLM_DEBUG(lock, "add lock into lru list");
933 /* If this is a client-side namespace and this was the last
934 * reference, put it on the LRU. */
935 ldlm_lock_add_to_lru(lock);
936 unlock_res_and_lock(lock);
938 if (lock->l_flags & LDLM_FL_FAIL_LOC)
939 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
941 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
942 * are not supported by the server, otherwise, it is done on
944 if (!exp_connect_cancelset(lock->l_conn_export) &&
945 !ns_connect_lru_resize(ns))
946 ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0);
948 LDLM_DEBUG(lock, "do not add lock into lru list");
949 unlock_res_and_lock(lock);
956 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
958 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
960 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
961 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
962 ldlm_lock_decref_internal(lock, mode);
965 EXPORT_SYMBOL(ldlm_lock_decref);
968 * Decrease reader/writer refcount for LDLM lock with handle
969 * \a lockh and mark it for subsequent cancellation once r/w refcount
970 * drops to zero instead of putting into LRU.
972 * Typical usage is for GROUP locks which we cannot allow to be cached.
974 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
976 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
979 LASSERT(lock != NULL);
981 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
982 lock_res_and_lock(lock);
983 lock->l_flags |= LDLM_FL_CBPENDING;
984 unlock_res_and_lock(lock);
985 ldlm_lock_decref_internal(lock, mode);
988 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
990 struct sl_insert_point {
991 cfs_list_t *res_link;
992 cfs_list_t *mode_link;
993 cfs_list_t *policy_link;
997 * Finds a position to insert the new lock into granted lock list.
999 * Used for locks eligible for skiplist optimization.
1002 * queue [input]: the granted list where search acts on;
1003 * req [input]: the lock whose position to be located;
1004 * prev [output]: positions within 3 lists to insert @req to
1008 * - ldlm_grant_lock_with_skiplist
1010 static void search_granted_lock(cfs_list_t *queue,
1011 struct ldlm_lock *req,
1012 struct sl_insert_point *prev)
1015 struct ldlm_lock *lock, *mode_end, *policy_end;
1018 cfs_list_for_each(tmp, queue) {
1019 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1021 mode_end = cfs_list_entry(lock->l_sl_mode.prev,
1022 struct ldlm_lock, l_sl_mode);
1024 if (lock->l_req_mode != req->l_req_mode) {
1025 /* jump to last lock of mode group */
1026 tmp = &mode_end->l_res_link;
1030 /* suitable mode group is found */
1031 if (lock->l_resource->lr_type == LDLM_PLAIN) {
1032 /* insert point is last lock of the mode group */
1033 prev->res_link = &mode_end->l_res_link;
1034 prev->mode_link = &mode_end->l_sl_mode;
1035 prev->policy_link = &req->l_sl_policy;
1038 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
1041 cfs_list_entry(lock->l_sl_policy.prev,
1045 if (lock->l_policy_data.l_inodebits.bits ==
1046 req->l_policy_data.l_inodebits.bits) {
1047 /* insert point is last lock of
1048 * the policy group */
1050 &policy_end->l_res_link;
1052 &policy_end->l_sl_mode;
1054 &policy_end->l_sl_policy;
1059 if (policy_end == mode_end)
1060 /* done with mode group */
1063 /* go to next policy group within mode group */
1064 tmp = policy_end->l_res_link.next;
1065 lock = cfs_list_entry(tmp, struct ldlm_lock,
1067 } /* loop over policy groups within the mode group */
1069 /* insert point is last lock of the mode group,
1070 * new policy group is started */
1071 prev->res_link = &mode_end->l_res_link;
1072 prev->mode_link = &mode_end->l_sl_mode;
1073 prev->policy_link = &req->l_sl_policy;
1077 LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
1082 /* insert point is last lock on the queue,
1083 * new mode group and new policy group are started */
1084 prev->res_link = queue->prev;
1085 prev->mode_link = &req->l_sl_mode;
1086 prev->policy_link = &req->l_sl_policy;
1092 * Add a lock into resource granted list after a position described by
1095 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
1096 struct sl_insert_point *prev)
1098 struct ldlm_resource *res = lock->l_resource;
1101 check_res_locked(res);
1103 ldlm_resource_dump(D_INFO, res);
1104 LDLM_DEBUG(lock, "About to add lock:");
1106 if (lock->l_destroyed) {
1107 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1111 LASSERT(cfs_list_empty(&lock->l_res_link));
1112 LASSERT(cfs_list_empty(&lock->l_sl_mode));
1113 LASSERT(cfs_list_empty(&lock->l_sl_policy));
1116 * lock->link == prev->link means lock is first starting the group.
1117 * Don't re-add to itself to suppress kernel warnings.
1119 if (&lock->l_res_link != prev->res_link)
1120 cfs_list_add(&lock->l_res_link, prev->res_link);
1121 if (&lock->l_sl_mode != prev->mode_link)
1122 cfs_list_add(&lock->l_sl_mode, prev->mode_link);
1123 if (&lock->l_sl_policy != prev->policy_link)
1124 cfs_list_add(&lock->l_sl_policy, prev->policy_link);
1130 * Add a lock to granted list on a resource maintaining skiplist
1133 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
1135 struct sl_insert_point prev;
1138 LASSERT(lock->l_req_mode == lock->l_granted_mode);
1140 search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
1141 ldlm_granted_list_add_lock(lock, &prev);
1146 * Perform lock granting bookkeeping.
1148 * Includes putting the lock into granted list and updating lock mode.
1150 * - ldlm_lock_enqueue
1151 * - ldlm_reprocess_queue
1152 * - ldlm_lock_convert
1154 * must be called with lr_lock held
1156 void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list)
1158 struct ldlm_resource *res = lock->l_resource;
1161 check_res_locked(res);
1163 lock->l_granted_mode = lock->l_req_mode;
1164 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
1165 ldlm_grant_lock_with_skiplist(lock);
1166 else if (res->lr_type == LDLM_EXTENT)
1167 ldlm_extent_add_lock(res, lock);
1169 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1171 if (lock->l_granted_mode < res->lr_most_restr)
1172 res->lr_most_restr = lock->l_granted_mode;
1174 if (work_list && lock->l_completion_ast != NULL)
1175 ldlm_add_ast_work_item(lock, NULL, work_list);
1177 ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1182 * Search for a lock with given properties in a queue.
1184 * \retval a referenced lock or NULL. See the flag descriptions below, in the
1185 * comment above ldlm_lock_match
1187 static struct ldlm_lock *search_queue(cfs_list_t *queue,
1189 ldlm_policy_data_t *policy,
1190 struct ldlm_lock *old_lock,
1191 __u64 flags, int unref)
1193 struct ldlm_lock *lock;
1196 cfs_list_for_each(tmp, queue) {
1199 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1201 if (lock == old_lock)
1204 /* llite sometimes wants to match locks that will be
1205 * canceled when their users drop, but we allow it to match
1206 * if it passes in CBPENDING and the lock still has users.
1207 * this is generally only going to be used by children
1208 * whose parents already hold a lock so forward progress
1209 * can still happen. */
1210 if (lock->l_flags & LDLM_FL_CBPENDING &&
1211 !(flags & LDLM_FL_CBPENDING))
1213 if (!unref && lock->l_flags & LDLM_FL_CBPENDING &&
1214 lock->l_readers == 0 && lock->l_writers == 0)
1217 if (!(lock->l_req_mode & *mode))
1219 match = lock->l_req_mode;
1221 if (lock->l_resource->lr_type == LDLM_EXTENT &&
1222 (lock->l_policy_data.l_extent.start >
1223 policy->l_extent.start ||
1224 lock->l_policy_data.l_extent.end < policy->l_extent.end))
1227 if (unlikely(match == LCK_GROUP) &&
1228 lock->l_resource->lr_type == LDLM_EXTENT &&
1229 lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
1232 /* We match if we have existing lock with same or wider set
1234 if (lock->l_resource->lr_type == LDLM_IBITS &&
1235 ((lock->l_policy_data.l_inodebits.bits &
1236 policy->l_inodebits.bits) !=
1237 policy->l_inodebits.bits))
1241 (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED ||
1245 if ((flags & LDLM_FL_LOCAL_ONLY) &&
1246 !(lock->l_flags & LDLM_FL_LOCAL))
1249 if (flags & LDLM_FL_TEST_LOCK) {
1250 LDLM_LOCK_GET(lock);
1251 ldlm_lock_touch_in_lru(lock);
1253 ldlm_lock_addref_internal_nolock(lock, match);
1262 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1264 if (!lock->l_failed) {
1266 cfs_waitq_broadcast(&lock->l_waitq);
1269 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1271 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1273 lock_res_and_lock(lock);
1274 ldlm_lock_fail_match_locked(lock);
1275 unlock_res_and_lock(lock);
1277 EXPORT_SYMBOL(ldlm_lock_fail_match);
1280 * Mark lock as "matchable" by OST.
1282 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1284 * Assumes LDLM lock is already locked.
1286 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1288 lock->l_flags |= LDLM_FL_LVB_READY;
1289 cfs_waitq_broadcast(&lock->l_waitq);
1291 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1294 * Mark lock as "matchable" by OST.
1295 * Locks the lock and then \see ldlm_lock_allow_match_locked
1297 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1299 lock_res_and_lock(lock);
1300 ldlm_lock_allow_match_locked(lock);
1301 unlock_res_and_lock(lock);
1303 EXPORT_SYMBOL(ldlm_lock_allow_match);
1306 * Attempt to find a lock with specified properties.
1308 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1311 * Can be called in two ways:
1313 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1314 * for a duplicate of.
1316 * Otherwise, all of the fields must be filled in, to match against.
1318 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1319 * server (ie, connh is NULL)
1320 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1321 * list will be considered
1322 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1323 * to be canceled can still be matched as long as they still have reader
1324 * or writer refernces
1325 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1326 * just tell us if we would have matched.
1328 * \retval 1 if it finds an already-existing lock that is compatible; in this
1329 * case, lockh is filled in with a addref()ed lock
1331 * We also check security context, and if that fails we simply return 0 (to
1332 * keep caller code unchanged), the context failure will be discovered by
1333 * caller sometime later.
1335 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1336 const struct ldlm_res_id *res_id, ldlm_type_t type,
1337 ldlm_policy_data_t *policy, ldlm_mode_t mode,
1338 struct lustre_handle *lockh, int unref)
1340 struct ldlm_resource *res;
1341 struct ldlm_lock *lock, *old_lock = NULL;
1346 old_lock = ldlm_handle2lock(lockh);
1349 ns = ldlm_lock_to_ns(old_lock);
1350 res_id = &old_lock->l_resource->lr_name;
1351 type = old_lock->l_resource->lr_type;
1352 mode = old_lock->l_req_mode;
1355 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1357 LASSERT(old_lock == NULL);
1361 LDLM_RESOURCE_ADDREF(res);
1364 lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
1368 if (flags & LDLM_FL_BLOCK_GRANTED)
1370 lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
1374 lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
1382 LDLM_RESOURCE_DELREF(res);
1383 ldlm_resource_putref(res);
1386 ldlm_lock2handle(lock, lockh);
1387 if ((flags & LDLM_FL_LVB_READY) &&
1388 (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1389 struct l_wait_info lwi;
1390 if (lock->l_completion_ast) {
1391 int err = lock->l_completion_ast(lock,
1392 LDLM_FL_WAIT_NOREPROC,
1395 if (flags & LDLM_FL_TEST_LOCK)
1396 LDLM_LOCK_RELEASE(lock);
1398 ldlm_lock_decref_internal(lock,
1405 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1406 NULL, LWI_ON_SIGNAL_NOOP, NULL);
1408 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1409 l_wait_event(lock->l_waitq,
1410 lock->l_flags & LDLM_FL_LVB_READY ||
1411 lock->l_destroyed || lock->l_failed,
1413 if (!(lock->l_flags & LDLM_FL_LVB_READY)) {
1414 if (flags & LDLM_FL_TEST_LOCK)
1415 LDLM_LOCK_RELEASE(lock);
1417 ldlm_lock_decref_internal(lock, mode);
1424 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1425 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1426 res_id->name[2] : policy->l_extent.start,
1427 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1428 res_id->name[3] : policy->l_extent.end);
1430 /* check user's security context */
1431 if (lock->l_conn_export &&
1432 sptlrpc_import_check_ctx(
1433 class_exp2cliimp(lock->l_conn_export))) {
1434 if (!(flags & LDLM_FL_TEST_LOCK))
1435 ldlm_lock_decref_internal(lock, mode);
1439 if (flags & LDLM_FL_TEST_LOCK)
1440 LDLM_LOCK_RELEASE(lock);
1442 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1443 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1444 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1445 type, mode, res_id->name[0], res_id->name[1],
1446 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1447 res_id->name[2] :policy->l_extent.start,
1448 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1449 res_id->name[3] : policy->l_extent.end);
1452 LDLM_LOCK_PUT(old_lock);
1454 return rc ? mode : 0;
1456 EXPORT_SYMBOL(ldlm_lock_match);
1458 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1461 struct ldlm_lock *lock;
1462 ldlm_mode_t mode = 0;
1465 lock = ldlm_handle2lock(lockh);
1467 lock_res_and_lock(lock);
1468 if (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED ||
1472 if (lock->l_flags & LDLM_FL_CBPENDING &&
1473 lock->l_readers == 0 && lock->l_writers == 0)
1477 *bits = lock->l_policy_data.l_inodebits.bits;
1478 mode = lock->l_granted_mode;
1479 ldlm_lock_addref_internal_nolock(lock, mode);
1486 unlock_res_and_lock(lock);
1487 LDLM_LOCK_PUT(lock);
1491 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1493 /** The caller must guarantee that the buffer is large enough. */
1494 int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill,
1495 enum req_location loc, void *data, int size)
1500 LASSERT(data != NULL);
1503 switch (lock->l_lvb_type) {
1505 if (size == sizeof(struct ost_lvb)) {
1506 if (loc == RCL_CLIENT)
1507 lvb = req_capsule_client_swab_get(pill,
1509 lustre_swab_ost_lvb);
1511 lvb = req_capsule_server_swab_get(pill,
1513 lustre_swab_ost_lvb);
1514 if (unlikely(lvb == NULL)) {
1515 LDLM_ERROR(lock, "no LVB");
1519 memcpy(data, lvb, size);
1520 } else if (size == sizeof(struct ost_lvb_v1)) {
1521 struct ost_lvb *olvb = data;
1523 if (loc == RCL_CLIENT)
1524 lvb = req_capsule_client_swab_get(pill,
1526 lustre_swab_ost_lvb_v1);
1528 lvb = req_capsule_server_sized_swab_get(pill,
1530 lustre_swab_ost_lvb_v1);
1531 if (unlikely(lvb == NULL)) {
1532 LDLM_ERROR(lock, "no LVB");
1536 memcpy(data, lvb, size);
1537 olvb->lvb_mtime_ns = 0;
1538 olvb->lvb_atime_ns = 0;
1539 olvb->lvb_ctime_ns = 0;
1541 LDLM_ERROR(lock, "Replied unexpected ost LVB size %d",
1547 if (size == sizeof(struct lquota_lvb)) {
1548 if (loc == RCL_CLIENT)
1549 lvb = req_capsule_client_swab_get(pill,
1551 lustre_swab_lquota_lvb);
1553 lvb = req_capsule_server_swab_get(pill,
1555 lustre_swab_lquota_lvb);
1556 if (unlikely(lvb == NULL)) {
1557 LDLM_ERROR(lock, "no LVB");
1561 memcpy(data, lvb, size);
1563 LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d",
1572 if (loc == RCL_CLIENT)
1573 lvb = req_capsule_client_get(pill, &RMF_DLM_LVB);
1575 lvb = req_capsule_server_get(pill, &RMF_DLM_LVB);
1576 if (unlikely(lvb == NULL)) {
1577 LDLM_ERROR(lock, "no LVB");
1581 memcpy(data, lvb, size);
1584 LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type);
1585 libcfs_debug_dumpstack(NULL);
1593 * Create and fill in new LDLM lock with specified properties.
1594 * Returns a referenced lock
1596 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1597 const struct ldlm_res_id *res_id,
1600 const struct ldlm_callback_suite *cbs,
1601 void *data, __u32 lvb_len,
1602 enum lvb_type lvb_type)
1604 struct ldlm_lock *lock;
1605 struct ldlm_resource *res;
1608 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1612 lock = ldlm_lock_new(res);
1617 lock->l_req_mode = mode;
1618 lock->l_ast_data = data;
1619 lock->l_pid = cfs_curproc_pid();
1620 lock->l_ns_srv = !!ns_is_server(ns);
1622 lock->l_blocking_ast = cbs->lcs_blocking;
1623 lock->l_completion_ast = cbs->lcs_completion;
1624 lock->l_glimpse_ast = cbs->lcs_glimpse;
1625 lock->l_weigh_ast = cbs->lcs_weigh;
1628 lock->l_tree_node = NULL;
1629 /* if this is the extent lock, allocate the interval tree node */
1630 if (type == LDLM_EXTENT) {
1631 if (ldlm_interval_alloc(lock) == NULL)
1636 lock->l_lvb_len = lvb_len;
1637 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1638 if (lock->l_lvb_data == NULL)
1642 lock->l_lvb_type = lvb_type;
1643 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1649 ldlm_lock_destroy(lock);
1650 LDLM_LOCK_RELEASE(lock);
1655 * Enqueue (request) a lock.
1657 * Does not block. As a result of enqueue the lock would be put
1658 * into granted or waiting list.
1660 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1661 * set, skip all the enqueueing and delegate lock processing to intent policy
1664 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1665 struct ldlm_lock **lockp,
1666 void *cookie, __u64 *flags)
1668 struct ldlm_lock *lock = *lockp;
1669 struct ldlm_resource *res = lock->l_resource;
1670 int local = ns_is_client(ldlm_res_to_ns(res));
1671 #ifdef HAVE_SERVER_SUPPORT
1672 ldlm_processing_policy policy;
1674 ldlm_error_t rc = ELDLM_OK;
1675 struct ldlm_interval *node = NULL;
1678 lock->l_last_activity = cfs_time_current_sec();
1679 /* policies are not executed on the client or during replay */
1680 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1681 && !local && ns->ns_policy) {
1682 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1684 if (rc == ELDLM_LOCK_REPLACED) {
1685 /* The lock that was returned has already been granted,
1686 * and placed into lockp. If it's not the same as the
1687 * one we passed in, then destroy the old one and our
1688 * work here is done. */
1689 if (lock != *lockp) {
1690 ldlm_lock_destroy(lock);
1691 LDLM_LOCK_RELEASE(lock);
1693 *flags |= LDLM_FL_LOCK_CHANGED;
1695 } else if (rc != ELDLM_OK ||
1696 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1697 ldlm_lock_destroy(lock);
1702 /* For a replaying lock, it might be already in granted list. So
1703 * unlinking the lock will cause the interval node to be freed, we
1704 * have to allocate the interval node early otherwise we can't regrant
1705 * this lock in the future. - jay */
1706 if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1707 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
1709 lock_res_and_lock(lock);
1710 if (local && lock->l_req_mode == lock->l_granted_mode) {
1711 /* The server returned a blocked lock, but it was granted
1712 * before we got a chance to actually enqueue it. We don't
1713 * need to do anything else. */
1714 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1715 LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1716 GOTO(out, ELDLM_OK);
1719 ldlm_resource_unlink_lock(lock);
1720 if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1722 ldlm_lock_destroy_nolock(lock);
1723 GOTO(out, rc = -ENOMEM);
1726 CFS_INIT_LIST_HEAD(&node->li_group);
1727 ldlm_interval_attach(node, lock);
1731 /* Some flags from the enqueue want to make it into the AST, via the
1732 * lock's l_flags. */
1733 lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1735 /* This distinction between local lock trees is very important; a client
1736 * namespace only has information about locks taken by that client, and
1737 * thus doesn't have enough information to decide for itself if it can
1738 * be granted (below). In this case, we do exactly what the server
1739 * tells us to do, as dictated by the 'flags'.
1741 * We do exactly the same thing during recovery, when the server is
1742 * more or less trusting the clients not to lie.
1744 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1745 * granted/converting queues. */
1747 if (*flags & LDLM_FL_BLOCK_CONV)
1748 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1749 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1750 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1752 ldlm_grant_lock(lock, NULL);
1753 GOTO(out, ELDLM_OK);
1754 #ifdef HAVE_SERVER_SUPPORT
1755 } else if (*flags & LDLM_FL_REPLAY) {
1756 if (*flags & LDLM_FL_BLOCK_CONV) {
1757 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1758 GOTO(out, ELDLM_OK);
1759 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1760 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1761 GOTO(out, ELDLM_OK);
1762 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1763 ldlm_grant_lock(lock, NULL);
1764 GOTO(out, ELDLM_OK);
1766 /* If no flags, fall through to normal enqueue path. */
1769 policy = ldlm_processing_policy_table[res->lr_type];
1770 policy(lock, flags, 1, &rc, NULL);
1774 CERROR("This is client-side-only module, cannot handle "
1775 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1781 unlock_res_and_lock(lock);
1783 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1787 #ifdef HAVE_SERVER_SUPPORT
1789 * Iterate through all waiting locks on a given resource queue and attempt to
1792 * Must be called with resource lock held.
1794 int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue,
1795 cfs_list_t *work_list)
1797 cfs_list_t *tmp, *pos;
1798 ldlm_processing_policy policy;
1800 int rc = LDLM_ITER_CONTINUE;
1804 check_res_locked(res);
1806 policy = ldlm_processing_policy_table[res->lr_type];
1809 cfs_list_for_each_safe(tmp, pos, queue) {
1810 struct ldlm_lock *pending;
1811 pending = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1813 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1816 rc = policy(pending, &flags, 0, &err, work_list);
1817 if (rc != LDLM_ITER_CONTINUE)
1826 * Process a call to blocking AST callback for a lock in ast_work list
1829 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1831 struct ldlm_cb_set_arg *arg = opaq;
1832 struct ldlm_lock_desc d;
1834 struct ldlm_lock *lock;
1837 if (cfs_list_empty(arg->list))
1840 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1842 /* nobody should touch l_bl_ast */
1843 lock_res_and_lock(lock);
1844 cfs_list_del_init(&lock->l_bl_ast);
1846 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1847 LASSERT(lock->l_bl_ast_run == 0);
1848 LASSERT(lock->l_blocking_lock);
1849 lock->l_bl_ast_run++;
1850 unlock_res_and_lock(lock);
1852 ldlm_lock2desc(lock->l_blocking_lock, &d);
1854 rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1855 LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1856 lock->l_blocking_lock = NULL;
1857 LDLM_LOCK_RELEASE(lock);
1863 * Process a call to completion AST callback for a lock in ast_work list
1866 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1868 struct ldlm_cb_set_arg *arg = opaq;
1870 struct ldlm_lock *lock;
1871 ldlm_completion_callback completion_callback;
1874 if (cfs_list_empty(arg->list))
1877 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1879 /* It's possible to receive a completion AST before we've set
1880 * the l_completion_ast pointer: either because the AST arrived
1881 * before the reply, or simply because there's a small race
1882 * window between receiving the reply and finishing the local
1883 * enqueue. (bug 842)
1885 * This can't happen with the blocking_ast, however, because we
1886 * will never call the local blocking_ast until we drop our
1887 * reader/writer reference, which we won't do until we get the
1888 * reply and finish enqueueing. */
1890 /* nobody should touch l_cp_ast */
1891 lock_res_and_lock(lock);
1892 cfs_list_del_init(&lock->l_cp_ast);
1893 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1894 /* save l_completion_ast since it can be changed by
1895 * mds_intent_policy(), see bug 14225 */
1896 completion_callback = lock->l_completion_ast;
1897 lock->l_flags &= ~LDLM_FL_CP_REQD;
1898 unlock_res_and_lock(lock);
1900 if (completion_callback != NULL)
1901 rc = completion_callback(lock, 0, (void *)arg);
1902 LDLM_LOCK_RELEASE(lock);
1908 * Process a call to revocation AST callback for a lock in ast_work list
1911 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1913 struct ldlm_cb_set_arg *arg = opaq;
1914 struct ldlm_lock_desc desc;
1916 struct ldlm_lock *lock;
1919 if (cfs_list_empty(arg->list))
1922 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1923 cfs_list_del_init(&lock->l_rk_ast);
1925 /* the desc just pretend to exclusive */
1926 ldlm_lock2desc(lock, &desc);
1927 desc.l_req_mode = LCK_EX;
1928 desc.l_granted_mode = 0;
1930 rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1931 LDLM_LOCK_RELEASE(lock);
1937 * Process a call to glimpse AST callback for a lock in ast_work list
1939 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1941 struct ldlm_cb_set_arg *arg = opaq;
1942 struct ldlm_glimpse_work *gl_work;
1943 struct ldlm_lock *lock;
1947 if (cfs_list_empty(arg->list))
1950 gl_work = cfs_list_entry(arg->list->next, struct ldlm_glimpse_work,
1952 cfs_list_del_init(&gl_work->gl_list);
1954 lock = gl_work->gl_lock;
1956 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
1957 arg->gl_desc = gl_work->gl_desc;
1959 /* invoke the actual glimpse callback */
1960 if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
1963 LDLM_LOCK_RELEASE(lock);
1965 if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
1966 OBD_FREE_PTR(gl_work);
1972 * Process list of locks in need of ASTs being sent.
1974 * Used on server to send multiple ASTs together instead of sending one by
1977 int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list,
1978 ldlm_desc_ast_t ast_type)
1980 struct ldlm_cb_set_arg *arg;
1981 set_producer_func work_ast_lock;
1984 if (cfs_list_empty(rpc_list))
1991 cfs_atomic_set(&arg->restart, 0);
1992 arg->list = rpc_list;
1995 case LDLM_WORK_BL_AST:
1996 arg->type = LDLM_BL_CALLBACK;
1997 work_ast_lock = ldlm_work_bl_ast_lock;
1999 case LDLM_WORK_CP_AST:
2000 arg->type = LDLM_CP_CALLBACK;
2001 work_ast_lock = ldlm_work_cp_ast_lock;
2003 case LDLM_WORK_REVOKE_AST:
2004 arg->type = LDLM_BL_CALLBACK;
2005 work_ast_lock = ldlm_work_revoke_ast_lock;
2007 case LDLM_WORK_GL_AST:
2008 arg->type = LDLM_GL_CALLBACK;
2009 work_ast_lock = ldlm_work_gl_ast_lock;
2015 /* We create a ptlrpc request set with flow control extension.
2016 * This request set will use the work_ast_lock function to produce new
2017 * requests and will send a new request each time one completes in order
2018 * to keep the number of requests in flight to ns_max_parallel_ast */
2019 arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
2020 work_ast_lock, arg);
2021 if (arg->set == NULL)
2022 GOTO(out, rc = -ENOMEM);
2024 ptlrpc_set_wait(arg->set);
2025 ptlrpc_set_destroy(arg->set);
2027 rc = cfs_atomic_read(&arg->restart) ? -ERESTART : 0;
2034 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
2036 ldlm_reprocess_all(res);
2037 return LDLM_ITER_CONTINUE;
2040 static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2041 cfs_hlist_node_t *hnode, void *arg)
2043 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2046 rc = reprocess_one_queue(res, arg);
2048 return rc == LDLM_ITER_STOP;
2052 * Iterate through all resources on a namespace attempting to grant waiting
2055 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
2060 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2061 ldlm_reprocess_res, NULL);
2065 EXPORT_SYMBOL(ldlm_reprocess_all_ns);
2068 * Try to grant all waiting locks on a resource.
2070 * Calls ldlm_reprocess_queue on converting and waiting queues.
2072 * Typically called after some resource locks are cancelled to see
2073 * if anything could be granted as a result of the cancellation.
2075 void ldlm_reprocess_all(struct ldlm_resource *res)
2077 CFS_LIST_HEAD(rpc_list);
2079 #ifdef HAVE_SERVER_SUPPORT
2082 /* Local lock trees don't get reprocessed. */
2083 if (ns_is_client(ldlm_res_to_ns(res))) {
2090 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
2091 if (rc == LDLM_ITER_CONTINUE)
2092 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
2095 rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
2097 if (rc == -ERESTART) {
2098 LASSERT(cfs_list_empty(&rpc_list));
2103 if (!ns_is_client(ldlm_res_to_ns(res))) {
2104 CERROR("This is client-side-only module, cannot handle "
2105 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2113 * Helper function to call blocking AST for LDLM lock \a lock in a
2114 * "cancelling" mode.
2116 void ldlm_cancel_callback(struct ldlm_lock *lock)
2118 check_res_locked(lock->l_resource);
2119 if (!(lock->l_flags & LDLM_FL_CANCEL)) {
2120 lock->l_flags |= LDLM_FL_CANCEL;
2121 if (lock->l_blocking_ast) {
2122 unlock_res_and_lock(lock);
2123 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
2125 lock_res_and_lock(lock);
2127 LDLM_DEBUG(lock, "no blocking ast");
2130 lock->l_flags |= LDLM_FL_BL_DONE;
2134 * Remove skiplist-enabled LDLM lock \a req from granted list
2136 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
2138 if (req->l_resource->lr_type != LDLM_PLAIN &&
2139 req->l_resource->lr_type != LDLM_IBITS)
2142 cfs_list_del_init(&req->l_sl_policy);
2143 cfs_list_del_init(&req->l_sl_mode);
2147 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2149 void ldlm_lock_cancel(struct ldlm_lock *lock)
2151 struct ldlm_resource *res;
2152 struct ldlm_namespace *ns;
2155 lock_res_and_lock(lock);
2157 res = lock->l_resource;
2158 ns = ldlm_res_to_ns(res);
2160 /* Please do not, no matter how tempting, remove this LBUG without
2161 * talking to me first. -phik */
2162 if (lock->l_readers || lock->l_writers) {
2163 LDLM_ERROR(lock, "lock still has references");
2168 ldlm_del_waiting_lock(lock);
2170 /* Releases cancel callback. */
2171 ldlm_cancel_callback(lock);
2173 /* Yes, second time, just in case it was added again while we were
2174 running with no res lock in ldlm_cancel_callback */
2176 ldlm_del_waiting_lock(lock);
2178 ldlm_resource_unlink_lock(lock);
2179 ldlm_lock_destroy_nolock(lock);
2181 if (lock->l_granted_mode == lock->l_req_mode)
2182 ldlm_pool_del(&ns->ns_pool, lock);
2184 /* Make sure we will not be called again for same lock what is possible
2185 * if not to zero out lock->l_granted_mode */
2186 lock->l_granted_mode = LCK_MINMODE;
2187 unlock_res_and_lock(lock);
2191 EXPORT_SYMBOL(ldlm_lock_cancel);
2194 * Set opaque data into the lock that only makes sense to upper layer.
2196 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
2198 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2203 if (lock->l_ast_data == NULL)
2204 lock->l_ast_data = data;
2205 if (lock->l_ast_data == data)
2207 LDLM_LOCK_PUT(lock);
2211 EXPORT_SYMBOL(ldlm_lock_set_data);
2213 struct export_cl_data {
2214 struct obd_export *ecl_exp;
2219 * Iterator function for ldlm_cancel_locks_for_export.
2220 * Cancels passed locks.
2222 int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2223 cfs_hlist_node_t *hnode, void *data)
2226 struct export_cl_data *ecl = (struct export_cl_data *)data;
2227 struct obd_export *exp = ecl->ecl_exp;
2228 struct ldlm_lock *lock = cfs_hash_object(hs, hnode);
2229 struct ldlm_resource *res;
2231 res = ldlm_resource_getref(lock->l_resource);
2232 LDLM_LOCK_GET(lock);
2234 LDLM_DEBUG(lock, "export %p", exp);
2235 ldlm_res_lvbo_update(res, NULL, 1);
2236 ldlm_lock_cancel(lock);
2237 ldlm_reprocess_all(res);
2238 ldlm_resource_putref(res);
2239 LDLM_LOCK_RELEASE(lock);
2242 if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
2244 "Cancel lock %p for export %p (loop %d), still have "
2245 "%d locks left on hash table.\n",
2246 lock, exp, ecl->ecl_loop,
2247 cfs_atomic_read(&hs->hs_count));
2254 * Cancel all locks for given export.
2256 * Typically called on client disconnection/eviction
2258 void ldlm_cancel_locks_for_export(struct obd_export *exp)
2260 struct export_cl_data ecl = {
2265 cfs_hash_for_each_empty(exp->exp_lock_hash,
2266 ldlm_cancel_locks_for_export_cb, &ecl);
2270 * Downgrade an exclusive lock.
2272 * A fast variant of ldlm_lock_convert for convertion of exclusive
2273 * locks. The convertion is always successful.
2274 * Used by Commit on Sharing (COS) code.
2276 * \param lock A lock to convert
2277 * \param new_mode new lock mode
2279 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
2283 LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
2284 LASSERT(new_mode == LCK_COS);
2286 lock_res_and_lock(lock);
2287 ldlm_resource_unlink_lock(lock);
2289 * Remove the lock from pool as it will be added again in
2290 * ldlm_grant_lock() called below.
2292 ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
2294 lock->l_req_mode = new_mode;
2295 ldlm_grant_lock(lock, NULL);
2296 unlock_res_and_lock(lock);
2297 ldlm_reprocess_all(lock->l_resource);
2301 EXPORT_SYMBOL(ldlm_lock_downgrade);
2304 * Attempt to convert already granted lock to a different mode.
2306 * While lock conversion is not currently used, future client-side
2307 * optimizations could take advantage of it to avoid discarding cached
2310 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
2313 CFS_LIST_HEAD(rpc_list);
2314 struct ldlm_resource *res;
2315 struct ldlm_namespace *ns;
2317 #ifdef HAVE_SERVER_SUPPORT
2319 struct sl_insert_point prev;
2321 struct ldlm_interval *node;
2324 /* Just return if mode is unchanged. */
2325 if (new_mode == lock->l_granted_mode) {
2326 *flags |= LDLM_FL_BLOCK_GRANTED;
2327 RETURN(lock->l_resource);
2330 /* I can't check the type of lock here because the bitlock of lock
2331 * is not held here, so do the allocation blindly. -jay */
2332 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
2333 if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */
2336 LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
2337 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2339 lock_res_and_lock(lock);
2341 res = lock->l_resource;
2342 ns = ldlm_res_to_ns(res);
2344 #ifdef HAVE_SERVER_SUPPORT
2345 old_mode = lock->l_req_mode;
2347 lock->l_req_mode = new_mode;
2348 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2349 #ifdef HAVE_SERVER_SUPPORT
2350 /* remember the lock position where the lock might be
2351 * added back to the granted list later and also
2352 * remember the join mode for skiplist fixing. */
2353 prev.res_link = lock->l_res_link.prev;
2354 prev.mode_link = lock->l_sl_mode.prev;
2355 prev.policy_link = lock->l_sl_policy.prev;
2357 ldlm_resource_unlink_lock(lock);
2359 ldlm_resource_unlink_lock(lock);
2360 if (res->lr_type == LDLM_EXTENT) {
2361 /* FIXME: ugly code, I have to attach the lock to a
2362 * interval node again since perhaps it will be granted
2364 CFS_INIT_LIST_HEAD(&node->li_group);
2365 ldlm_interval_attach(node, lock);
2371 * Remove old lock from the pool before adding the lock with new
2372 * mode below in ->policy()
2374 ldlm_pool_del(&ns->ns_pool, lock);
2376 /* If this is a local resource, put it on the appropriate list. */
2377 if (ns_is_client(ldlm_res_to_ns(res))) {
2378 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2379 ldlm_resource_add_lock(res, &res->lr_converting, lock);
2381 /* This should never happen, because of the way the
2382 * server handles conversions. */
2383 LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2387 ldlm_grant_lock(lock, &rpc_list);
2389 /* FIXME: completion handling not with lr_lock held ! */
2390 if (lock->l_completion_ast)
2391 lock->l_completion_ast(lock, 0, NULL);
2393 #ifdef HAVE_SERVER_SUPPORT
2398 ldlm_processing_policy policy;
2399 policy = ldlm_processing_policy_table[res->lr_type];
2400 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2401 if (rc == LDLM_ITER_STOP) {
2402 lock->l_req_mode = old_mode;
2403 if (res->lr_type == LDLM_EXTENT)
2404 ldlm_extent_add_lock(res, lock);
2406 ldlm_granted_list_add_lock(lock, &prev);
2410 *flags |= LDLM_FL_BLOCK_GRANTED;
2416 CERROR("This is client-side-only module, cannot handle "
2417 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2421 unlock_res_and_lock(lock);
2424 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2426 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2429 EXPORT_SYMBOL(ldlm_lock_convert);
2432 * Print lock with lock handle \a lockh description into debug log.
2434 * Used when printing all locks on a resource for debug purposes.
2436 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
2438 struct ldlm_lock *lock;
2440 if (!((libcfs_debug | D_ERROR) & level))
2443 lock = ldlm_handle2lock(lockh);
2447 LDLM_DEBUG_LIMIT(level, lock, "###");
2449 LDLM_LOCK_PUT(lock);
2451 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2454 * Print lock information with custom message into debug log.
2457 void _ldlm_lock_debug(struct ldlm_lock *lock,
2458 struct libcfs_debug_msg_data *msgdata,
2459 const char *fmt, ...)
2462 struct obd_export *exp = lock->l_export;
2463 struct ldlm_resource *resource = lock->l_resource;
2464 char *nid = "local";
2466 va_start(args, fmt);
2468 if (exp && exp->exp_connection) {
2469 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2470 } else if (exp && exp->exp_obd != NULL) {
2471 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2472 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2475 if (resource == NULL) {
2476 libcfs_debug_vmsg2(msgdata, fmt, args,
2477 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2478 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2479 "remote: "LPX64" expref: %d pid: %u timeout: %lu "
2482 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2483 lock->l_readers, lock->l_writers,
2484 ldlm_lockname[lock->l_granted_mode],
2485 ldlm_lockname[lock->l_req_mode],
2486 lock->l_flags, nid, lock->l_remote_handle.cookie,
2487 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2488 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2493 switch (resource->lr_type) {
2495 libcfs_debug_vmsg2(msgdata, fmt, args,
2496 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2497 "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
2498 "] (req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote:"
2499 " "LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2500 ldlm_lock_to_ns_name(lock), lock,
2501 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2502 lock->l_readers, lock->l_writers,
2503 ldlm_lockname[lock->l_granted_mode],
2504 ldlm_lockname[lock->l_req_mode],
2505 resource->lr_name.name[0],
2506 resource->lr_name.name[1],
2507 cfs_atomic_read(&resource->lr_refcount),
2508 ldlm_typename[resource->lr_type],
2509 lock->l_policy_data.l_extent.start,
2510 lock->l_policy_data.l_extent.end,
2511 lock->l_req_extent.start, lock->l_req_extent.end,
2512 lock->l_flags, nid, lock->l_remote_handle.cookie,
2513 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2514 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2518 libcfs_debug_vmsg2(msgdata, fmt, args,
2519 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2520 "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
2521 "["LPU64"->"LPU64"] flags: "LPX64" nid: %s remote: "LPX64
2522 " expref: %d pid: %u timeout: %lu\n",
2523 ldlm_lock_to_ns_name(lock), lock,
2524 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2525 lock->l_readers, lock->l_writers,
2526 ldlm_lockname[lock->l_granted_mode],
2527 ldlm_lockname[lock->l_req_mode],
2528 resource->lr_name.name[0],
2529 resource->lr_name.name[1],
2530 cfs_atomic_read(&resource->lr_refcount),
2531 ldlm_typename[resource->lr_type],
2532 lock->l_policy_data.l_flock.pid,
2533 lock->l_policy_data.l_flock.start,
2534 lock->l_policy_data.l_flock.end,
2535 lock->l_flags, nid, lock->l_remote_handle.cookie,
2536 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2537 lock->l_pid, lock->l_callback_timeout);
2541 libcfs_debug_vmsg2(msgdata, fmt, args,
2542 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2543 "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
2544 "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2545 "pid: %u timeout: %lu lvb_type: %d\n",
2546 ldlm_lock_to_ns_name(lock),
2547 lock, lock->l_handle.h_cookie,
2548 cfs_atomic_read (&lock->l_refc),
2549 lock->l_readers, lock->l_writers,
2550 ldlm_lockname[lock->l_granted_mode],
2551 ldlm_lockname[lock->l_req_mode],
2552 resource->lr_name.name[0],
2553 resource->lr_name.name[1],
2554 lock->l_policy_data.l_inodebits.bits,
2555 cfs_atomic_read(&resource->lr_refcount),
2556 ldlm_typename[resource->lr_type],
2557 lock->l_flags, nid, lock->l_remote_handle.cookie,
2558 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2559 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2563 libcfs_debug_vmsg2(msgdata, fmt, args,
2564 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2565 "res: "LPU64"/"LPU64" rrc: %d type: %s flags: "LPX64" "
2566 "nid: %s remote: "LPX64" expref: %d pid: %u timeout: %lu"
2568 ldlm_lock_to_ns_name(lock),
2569 lock, lock->l_handle.h_cookie,
2570 cfs_atomic_read (&lock->l_refc),
2571 lock->l_readers, lock->l_writers,
2572 ldlm_lockname[lock->l_granted_mode],
2573 ldlm_lockname[lock->l_req_mode],
2574 resource->lr_name.name[0],
2575 resource->lr_name.name[1],
2576 cfs_atomic_read(&resource->lr_refcount),
2577 ldlm_typename[resource->lr_type],
2578 lock->l_flags, nid, lock->l_remote_handle.cookie,
2579 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2580 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2585 EXPORT_SYMBOL(_ldlm_lock_debug);