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_flags & LDLM_FL_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_flags & LDLM_FL_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_flags & LDLM_FL_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_flags & LDLM_FL_DESTROYED) {
377 LASSERT(cfs_list_empty(&lock->l_lru));
381 lock->l_flags |= LDLM_FL_DESTROYED;
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_flags & LDLM_FL_DESTROYED)== 0)) {
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_flags & LDLM_FL_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_FL_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 if (lock->l_flags & LDLM_FL_LOCAL &&
885 !lock->l_readers && !lock->l_writers) {
886 /* If this is a local lock on a server namespace and this was
887 * the last reference, cancel the lock. */
888 CDEBUG(D_INFO, "forcing cancel of local lock\n");
889 lock->l_flags |= LDLM_FL_CBPENDING;
892 if (!lock->l_readers && !lock->l_writers &&
893 (lock->l_flags & LDLM_FL_CBPENDING)) {
894 /* If we received a blocked AST and this was the last reference,
895 * run the callback. */
896 if ((lock->l_flags & LDLM_FL_NS_SRV) && lock->l_export)
897 CERROR("FL_CBPENDING set on non-local lock--just a "
900 LDLM_DEBUG(lock, "final decref done on cbpending lock");
902 LDLM_LOCK_GET(lock); /* dropped by bl thread */
903 ldlm_lock_remove_from_lru(lock);
904 unlock_res_and_lock(lock);
906 if (lock->l_flags & LDLM_FL_FAIL_LOC)
907 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
909 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
910 ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
911 ldlm_handle_bl_callback(ns, NULL, lock);
912 } else if (ns_is_client(ns) &&
913 !lock->l_readers && !lock->l_writers &&
914 !(lock->l_flags & LDLM_FL_NO_LRU) &&
915 !(lock->l_flags & LDLM_FL_BL_AST)) {
917 LDLM_DEBUG(lock, "add lock into lru list");
919 /* If this is a client-side namespace and this was the last
920 * reference, put it on the LRU. */
921 ldlm_lock_add_to_lru(lock);
922 unlock_res_and_lock(lock);
924 if (lock->l_flags & LDLM_FL_FAIL_LOC)
925 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
927 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
928 * are not supported by the server, otherwise, it is done on
930 if (!exp_connect_cancelset(lock->l_conn_export) &&
931 !ns_connect_lru_resize(ns))
932 ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0);
934 LDLM_DEBUG(lock, "do not add lock into lru list");
935 unlock_res_and_lock(lock);
942 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
944 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
946 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
947 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
948 ldlm_lock_decref_internal(lock, mode);
951 EXPORT_SYMBOL(ldlm_lock_decref);
954 * Decrease reader/writer refcount for LDLM lock with handle
955 * \a lockh and mark it for subsequent cancellation once r/w refcount
956 * drops to zero instead of putting into LRU.
958 * Typical usage is for GROUP locks which we cannot allow to be cached.
960 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
962 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
965 LASSERT(lock != NULL);
967 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
968 lock_res_and_lock(lock);
969 lock->l_flags |= LDLM_FL_CBPENDING;
970 unlock_res_and_lock(lock);
971 ldlm_lock_decref_internal(lock, mode);
974 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel);
976 struct sl_insert_point {
977 cfs_list_t *res_link;
978 cfs_list_t *mode_link;
979 cfs_list_t *policy_link;
983 * Finds a position to insert the new lock into granted lock list.
985 * Used for locks eligible for skiplist optimization.
988 * queue [input]: the granted list where search acts on;
989 * req [input]: the lock whose position to be located;
990 * prev [output]: positions within 3 lists to insert @req to
994 * - ldlm_grant_lock_with_skiplist
996 static void search_granted_lock(cfs_list_t *queue,
997 struct ldlm_lock *req,
998 struct sl_insert_point *prev)
1001 struct ldlm_lock *lock, *mode_end, *policy_end;
1004 cfs_list_for_each(tmp, queue) {
1005 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1007 mode_end = cfs_list_entry(lock->l_sl_mode.prev,
1008 struct ldlm_lock, l_sl_mode);
1010 if (lock->l_req_mode != req->l_req_mode) {
1011 /* jump to last lock of mode group */
1012 tmp = &mode_end->l_res_link;
1016 /* suitable mode group is found */
1017 if (lock->l_resource->lr_type == LDLM_PLAIN) {
1018 /* insert point is last lock of the mode group */
1019 prev->res_link = &mode_end->l_res_link;
1020 prev->mode_link = &mode_end->l_sl_mode;
1021 prev->policy_link = &req->l_sl_policy;
1024 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
1027 cfs_list_entry(lock->l_sl_policy.prev,
1031 if (lock->l_policy_data.l_inodebits.bits ==
1032 req->l_policy_data.l_inodebits.bits) {
1033 /* insert point is last lock of
1034 * the policy group */
1036 &policy_end->l_res_link;
1038 &policy_end->l_sl_mode;
1040 &policy_end->l_sl_policy;
1045 if (policy_end == mode_end)
1046 /* done with mode group */
1049 /* go to next policy group within mode group */
1050 tmp = policy_end->l_res_link.next;
1051 lock = cfs_list_entry(tmp, struct ldlm_lock,
1053 } /* loop over policy groups within the mode group */
1055 /* insert point is last lock of the mode group,
1056 * new policy group is started */
1057 prev->res_link = &mode_end->l_res_link;
1058 prev->mode_link = &mode_end->l_sl_mode;
1059 prev->policy_link = &req->l_sl_policy;
1063 LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
1068 /* insert point is last lock on the queue,
1069 * new mode group and new policy group are started */
1070 prev->res_link = queue->prev;
1071 prev->mode_link = &req->l_sl_mode;
1072 prev->policy_link = &req->l_sl_policy;
1078 * Add a lock into resource granted list after a position described by
1081 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
1082 struct sl_insert_point *prev)
1084 struct ldlm_resource *res = lock->l_resource;
1087 check_res_locked(res);
1089 ldlm_resource_dump(D_INFO, res);
1090 LDLM_DEBUG(lock, "About to add lock:");
1092 if (lock->l_flags & LDLM_FL_DESTROYED) {
1093 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
1097 LASSERT(cfs_list_empty(&lock->l_res_link));
1098 LASSERT(cfs_list_empty(&lock->l_sl_mode));
1099 LASSERT(cfs_list_empty(&lock->l_sl_policy));
1102 * lock->link == prev->link means lock is first starting the group.
1103 * Don't re-add to itself to suppress kernel warnings.
1105 if (&lock->l_res_link != prev->res_link)
1106 cfs_list_add(&lock->l_res_link, prev->res_link);
1107 if (&lock->l_sl_mode != prev->mode_link)
1108 cfs_list_add(&lock->l_sl_mode, prev->mode_link);
1109 if (&lock->l_sl_policy != prev->policy_link)
1110 cfs_list_add(&lock->l_sl_policy, prev->policy_link);
1116 * Add a lock to granted list on a resource maintaining skiplist
1119 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
1121 struct sl_insert_point prev;
1124 LASSERT(lock->l_req_mode == lock->l_granted_mode);
1126 search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
1127 ldlm_granted_list_add_lock(lock, &prev);
1132 * Perform lock granting bookkeeping.
1134 * Includes putting the lock into granted list and updating lock mode.
1136 * - ldlm_lock_enqueue
1137 * - ldlm_reprocess_queue
1138 * - ldlm_lock_convert
1140 * must be called with lr_lock held
1142 void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list)
1144 struct ldlm_resource *res = lock->l_resource;
1147 check_res_locked(res);
1149 lock->l_granted_mode = lock->l_req_mode;
1150 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
1151 ldlm_grant_lock_with_skiplist(lock);
1152 else if (res->lr_type == LDLM_EXTENT)
1153 ldlm_extent_add_lock(res, lock);
1155 ldlm_resource_add_lock(res, &res->lr_granted, lock);
1157 if (lock->l_granted_mode < res->lr_most_restr)
1158 res->lr_most_restr = lock->l_granted_mode;
1160 if (work_list && lock->l_completion_ast != NULL)
1161 ldlm_add_ast_work_item(lock, NULL, work_list);
1163 ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock);
1168 * Search for a lock with given properties in a queue.
1170 * \retval a referenced lock or NULL. See the flag descriptions below, in the
1171 * comment above ldlm_lock_match
1173 static struct ldlm_lock *search_queue(cfs_list_t *queue,
1175 ldlm_policy_data_t *policy,
1176 struct ldlm_lock *old_lock,
1177 __u64 flags, int unref)
1179 struct ldlm_lock *lock;
1182 cfs_list_for_each(tmp, queue) {
1185 lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1187 if (lock == old_lock)
1190 /* llite sometimes wants to match locks that will be
1191 * canceled when their users drop, but we allow it to match
1192 * if it passes in CBPENDING and the lock still has users.
1193 * this is generally only going to be used by children
1194 * whose parents already hold a lock so forward progress
1195 * can still happen. */
1196 if (lock->l_flags & LDLM_FL_CBPENDING &&
1197 !(flags & LDLM_FL_CBPENDING))
1199 if (!unref && lock->l_flags & LDLM_FL_CBPENDING &&
1200 lock->l_readers == 0 && lock->l_writers == 0)
1203 if (!(lock->l_req_mode & *mode))
1205 match = lock->l_req_mode;
1207 if (lock->l_resource->lr_type == LDLM_EXTENT &&
1208 (lock->l_policy_data.l_extent.start >
1209 policy->l_extent.start ||
1210 lock->l_policy_data.l_extent.end < policy->l_extent.end))
1213 if (unlikely(match == LCK_GROUP) &&
1214 lock->l_resource->lr_type == LDLM_EXTENT &&
1215 lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
1218 /* We match if we have existing lock with same or wider set
1220 if (lock->l_resource->lr_type == LDLM_IBITS &&
1221 ((lock->l_policy_data.l_inodebits.bits &
1222 policy->l_inodebits.bits) !=
1223 policy->l_inodebits.bits))
1226 if (!unref && (lock->l_flags & LDLM_FL_GONE_MASK))
1229 if ((flags & LDLM_FL_LOCAL_ONLY) &&
1230 !(lock->l_flags & LDLM_FL_LOCAL))
1233 if (flags & LDLM_FL_TEST_LOCK) {
1234 LDLM_LOCK_GET(lock);
1235 ldlm_lock_touch_in_lru(lock);
1237 ldlm_lock_addref_internal_nolock(lock, match);
1246 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock)
1248 if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) {
1249 lock->l_flags |= LDLM_FL_FAIL_NOTIFIED;
1250 cfs_waitq_broadcast(&lock->l_waitq);
1253 EXPORT_SYMBOL(ldlm_lock_fail_match_locked);
1255 void ldlm_lock_fail_match(struct ldlm_lock *lock)
1257 lock_res_and_lock(lock);
1258 ldlm_lock_fail_match_locked(lock);
1259 unlock_res_and_lock(lock);
1261 EXPORT_SYMBOL(ldlm_lock_fail_match);
1264 * Mark lock as "matchable" by OST.
1266 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1268 * Assumes LDLM lock is already locked.
1270 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
1272 lock->l_flags |= LDLM_FL_LVB_READY;
1273 cfs_waitq_broadcast(&lock->l_waitq);
1275 EXPORT_SYMBOL(ldlm_lock_allow_match_locked);
1278 * Mark lock as "matchable" by OST.
1279 * Locks the lock and then \see ldlm_lock_allow_match_locked
1281 void ldlm_lock_allow_match(struct ldlm_lock *lock)
1283 lock_res_and_lock(lock);
1284 ldlm_lock_allow_match_locked(lock);
1285 unlock_res_and_lock(lock);
1287 EXPORT_SYMBOL(ldlm_lock_allow_match);
1290 * Attempt to find a lock with specified properties.
1292 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1295 * Can be called in two ways:
1297 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1298 * for a duplicate of.
1300 * Otherwise, all of the fields must be filled in, to match against.
1302 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1303 * server (ie, connh is NULL)
1304 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1305 * list will be considered
1306 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1307 * to be canceled can still be matched as long as they still have reader
1308 * or writer refernces
1309 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1310 * just tell us if we would have matched.
1312 * \retval 1 if it finds an already-existing lock that is compatible; in this
1313 * case, lockh is filled in with a addref()ed lock
1315 * We also check security context, and if that fails we simply return 0 (to
1316 * keep caller code unchanged), the context failure will be discovered by
1317 * caller sometime later.
1319 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1320 const struct ldlm_res_id *res_id, ldlm_type_t type,
1321 ldlm_policy_data_t *policy, ldlm_mode_t mode,
1322 struct lustre_handle *lockh, int unref)
1324 struct ldlm_resource *res;
1325 struct ldlm_lock *lock, *old_lock = NULL;
1330 old_lock = ldlm_handle2lock(lockh);
1333 ns = ldlm_lock_to_ns(old_lock);
1334 res_id = &old_lock->l_resource->lr_name;
1335 type = old_lock->l_resource->lr_type;
1336 mode = old_lock->l_req_mode;
1339 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
1341 LASSERT(old_lock == NULL);
1345 LDLM_RESOURCE_ADDREF(res);
1348 lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
1352 if (flags & LDLM_FL_BLOCK_GRANTED)
1354 lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
1358 lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
1366 LDLM_RESOURCE_DELREF(res);
1367 ldlm_resource_putref(res);
1370 ldlm_lock2handle(lock, lockh);
1371 if ((flags & LDLM_FL_LVB_READY) &&
1372 (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1373 __u64 wait_flags = LDLM_FL_LVB_READY |
1374 LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED;
1375 struct l_wait_info lwi;
1376 if (lock->l_completion_ast) {
1377 int err = lock->l_completion_ast(lock,
1378 LDLM_FL_WAIT_NOREPROC,
1381 if (flags & LDLM_FL_TEST_LOCK)
1382 LDLM_LOCK_RELEASE(lock);
1384 ldlm_lock_decref_internal(lock,
1391 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
1392 NULL, LWI_ON_SIGNAL_NOOP, NULL);
1394 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1395 l_wait_event(lock->l_waitq,
1396 lock->l_flags & wait_flags,
1398 if (!(lock->l_flags & LDLM_FL_LVB_READY)) {
1399 if (flags & LDLM_FL_TEST_LOCK)
1400 LDLM_LOCK_RELEASE(lock);
1402 ldlm_lock_decref_internal(lock, mode);
1409 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1410 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1411 res_id->name[2] : policy->l_extent.start,
1412 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1413 res_id->name[3] : policy->l_extent.end);
1415 /* check user's security context */
1416 if (lock->l_conn_export &&
1417 sptlrpc_import_check_ctx(
1418 class_exp2cliimp(lock->l_conn_export))) {
1419 if (!(flags & LDLM_FL_TEST_LOCK))
1420 ldlm_lock_decref_internal(lock, mode);
1424 if (flags & LDLM_FL_TEST_LOCK)
1425 LDLM_LOCK_RELEASE(lock);
1427 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1428 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1429 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1430 type, mode, res_id->name[0], res_id->name[1],
1431 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1432 res_id->name[2] :policy->l_extent.start,
1433 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1434 res_id->name[3] : policy->l_extent.end);
1437 LDLM_LOCK_PUT(old_lock);
1439 return rc ? mode : 0;
1441 EXPORT_SYMBOL(ldlm_lock_match);
1443 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1446 struct ldlm_lock *lock;
1447 ldlm_mode_t mode = 0;
1450 lock = ldlm_handle2lock(lockh);
1452 lock_res_and_lock(lock);
1453 if (lock->l_flags & LDLM_FL_GONE_MASK)
1456 if (lock->l_flags & LDLM_FL_CBPENDING &&
1457 lock->l_readers == 0 && lock->l_writers == 0)
1461 *bits = lock->l_policy_data.l_inodebits.bits;
1462 mode = lock->l_granted_mode;
1463 ldlm_lock_addref_internal_nolock(lock, mode);
1470 unlock_res_and_lock(lock);
1471 LDLM_LOCK_PUT(lock);
1475 EXPORT_SYMBOL(ldlm_revalidate_lock_handle);
1477 /** The caller must guarantee that the buffer is large enough. */
1478 int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill,
1479 enum req_location loc, void *data, int size)
1484 LASSERT(data != NULL);
1487 switch (lock->l_lvb_type) {
1489 if (size == sizeof(struct ost_lvb)) {
1490 if (loc == RCL_CLIENT)
1491 lvb = req_capsule_client_swab_get(pill,
1493 lustre_swab_ost_lvb);
1495 lvb = req_capsule_server_swab_get(pill,
1497 lustre_swab_ost_lvb);
1498 if (unlikely(lvb == NULL)) {
1499 LDLM_ERROR(lock, "no LVB");
1503 memcpy(data, lvb, size);
1504 } else if (size == sizeof(struct ost_lvb_v1)) {
1505 struct ost_lvb *olvb = data;
1507 if (loc == RCL_CLIENT)
1508 lvb = req_capsule_client_swab_get(pill,
1510 lustre_swab_ost_lvb_v1);
1512 lvb = req_capsule_server_sized_swab_get(pill,
1514 lustre_swab_ost_lvb_v1);
1515 if (unlikely(lvb == NULL)) {
1516 LDLM_ERROR(lock, "no LVB");
1520 memcpy(data, lvb, size);
1521 olvb->lvb_mtime_ns = 0;
1522 olvb->lvb_atime_ns = 0;
1523 olvb->lvb_ctime_ns = 0;
1525 LDLM_ERROR(lock, "Replied unexpected ost LVB size %d",
1531 if (size == sizeof(struct lquota_lvb)) {
1532 if (loc == RCL_CLIENT)
1533 lvb = req_capsule_client_swab_get(pill,
1535 lustre_swab_lquota_lvb);
1537 lvb = req_capsule_server_swab_get(pill,
1539 lustre_swab_lquota_lvb);
1540 if (unlikely(lvb == NULL)) {
1541 LDLM_ERROR(lock, "no LVB");
1545 memcpy(data, lvb, size);
1547 LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d",
1556 if (loc == RCL_CLIENT)
1557 lvb = req_capsule_client_get(pill, &RMF_DLM_LVB);
1559 lvb = req_capsule_server_get(pill, &RMF_DLM_LVB);
1560 if (unlikely(lvb == NULL)) {
1561 LDLM_ERROR(lock, "no LVB");
1565 memcpy(data, lvb, size);
1568 LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type);
1569 libcfs_debug_dumpstack(NULL);
1577 * Create and fill in new LDLM lock with specified properties.
1578 * Returns a referenced lock
1580 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1581 const struct ldlm_res_id *res_id,
1584 const struct ldlm_callback_suite *cbs,
1585 void *data, __u32 lvb_len,
1586 enum lvb_type lvb_type)
1588 struct ldlm_lock *lock;
1589 struct ldlm_resource *res;
1592 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1596 lock = ldlm_lock_new(res);
1601 lock->l_req_mode = mode;
1602 lock->l_ast_data = data;
1603 lock->l_pid = cfs_curproc_pid();
1604 if (ns_is_server(ns))
1605 lock->l_flags |= LDLM_FL_NS_SRV;
1607 lock->l_blocking_ast = cbs->lcs_blocking;
1608 lock->l_completion_ast = cbs->lcs_completion;
1609 lock->l_glimpse_ast = cbs->lcs_glimpse;
1612 lock->l_tree_node = NULL;
1613 /* if this is the extent lock, allocate the interval tree node */
1614 if (type == LDLM_EXTENT) {
1615 if (ldlm_interval_alloc(lock) == NULL)
1620 lock->l_lvb_len = lvb_len;
1621 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1622 if (lock->l_lvb_data == NULL)
1626 lock->l_lvb_type = lvb_type;
1627 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK))
1633 ldlm_lock_destroy(lock);
1634 LDLM_LOCK_RELEASE(lock);
1639 * Enqueue (request) a lock.
1641 * Does not block. As a result of enqueue the lock would be put
1642 * into granted or waiting list.
1644 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1645 * set, skip all the enqueueing and delegate lock processing to intent policy
1648 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1649 struct ldlm_lock **lockp,
1650 void *cookie, __u64 *flags)
1652 struct ldlm_lock *lock = *lockp;
1653 struct ldlm_resource *res = lock->l_resource;
1654 int local = ns_is_client(ldlm_res_to_ns(res));
1655 #ifdef HAVE_SERVER_SUPPORT
1656 ldlm_processing_policy policy;
1658 ldlm_error_t rc = ELDLM_OK;
1659 struct ldlm_interval *node = NULL;
1662 lock->l_last_activity = cfs_time_current_sec();
1663 /* policies are not executed on the client or during replay */
1664 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1665 && !local && ns->ns_policy) {
1666 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1668 if (rc == ELDLM_LOCK_REPLACED) {
1669 /* The lock that was returned has already been granted,
1670 * and placed into lockp. If it's not the same as the
1671 * one we passed in, then destroy the old one and our
1672 * work here is done. */
1673 if (lock != *lockp) {
1674 ldlm_lock_destroy(lock);
1675 LDLM_LOCK_RELEASE(lock);
1677 *flags |= LDLM_FL_LOCK_CHANGED;
1679 } else if (rc != ELDLM_OK ||
1680 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1681 ldlm_lock_destroy(lock);
1686 /* For a replaying lock, it might be already in granted list. So
1687 * unlinking the lock will cause the interval node to be freed, we
1688 * have to allocate the interval node early otherwise we can't regrant
1689 * this lock in the future. - jay */
1690 if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
1691 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
1693 lock_res_and_lock(lock);
1694 if (local && lock->l_req_mode == lock->l_granted_mode) {
1695 /* The server returned a blocked lock, but it was granted
1696 * before we got a chance to actually enqueue it. We don't
1697 * need to do anything else. */
1698 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1699 LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1700 GOTO(out, ELDLM_OK);
1703 ldlm_resource_unlink_lock(lock);
1704 if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
1706 ldlm_lock_destroy_nolock(lock);
1707 GOTO(out, rc = -ENOMEM);
1710 CFS_INIT_LIST_HEAD(&node->li_group);
1711 ldlm_interval_attach(node, lock);
1715 /* Some flags from the enqueue want to make it into the AST, via the
1716 * lock's l_flags. */
1717 lock->l_flags |= *flags & LDLM_FL_AST_DISCARD_DATA;
1719 /* This distinction between local lock trees is very important; a client
1720 * namespace only has information about locks taken by that client, and
1721 * thus doesn't have enough information to decide for itself if it can
1722 * be granted (below). In this case, we do exactly what the server
1723 * tells us to do, as dictated by the 'flags'.
1725 * We do exactly the same thing during recovery, when the server is
1726 * more or less trusting the clients not to lie.
1728 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1729 * granted/converting queues. */
1731 if (*flags & LDLM_FL_BLOCK_CONV)
1732 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1733 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1734 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1736 ldlm_grant_lock(lock, NULL);
1737 GOTO(out, ELDLM_OK);
1738 #ifdef HAVE_SERVER_SUPPORT
1739 } else if (*flags & LDLM_FL_REPLAY) {
1740 if (*flags & LDLM_FL_BLOCK_CONV) {
1741 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1742 GOTO(out, ELDLM_OK);
1743 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1744 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1745 GOTO(out, ELDLM_OK);
1746 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1747 ldlm_grant_lock(lock, NULL);
1748 GOTO(out, ELDLM_OK);
1750 /* If no flags, fall through to normal enqueue path. */
1753 policy = ldlm_processing_policy_table[res->lr_type];
1754 policy(lock, flags, 1, &rc, NULL);
1758 CERROR("This is client-side-only module, cannot handle "
1759 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1765 unlock_res_and_lock(lock);
1767 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
1771 #ifdef HAVE_SERVER_SUPPORT
1773 * Iterate through all waiting locks on a given resource queue and attempt to
1776 * Must be called with resource lock held.
1778 int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue,
1779 cfs_list_t *work_list)
1781 cfs_list_t *tmp, *pos;
1782 ldlm_processing_policy policy;
1784 int rc = LDLM_ITER_CONTINUE;
1788 check_res_locked(res);
1790 policy = ldlm_processing_policy_table[res->lr_type];
1793 cfs_list_for_each_safe(tmp, pos, queue) {
1794 struct ldlm_lock *pending;
1795 pending = cfs_list_entry(tmp, struct ldlm_lock, l_res_link);
1797 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1800 rc = policy(pending, &flags, 0, &err, work_list);
1801 if (rc != LDLM_ITER_CONTINUE)
1810 * Process a call to blocking AST callback for a lock in ast_work list
1813 ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1815 struct ldlm_cb_set_arg *arg = opaq;
1816 struct ldlm_lock_desc d;
1818 struct ldlm_lock *lock;
1821 if (cfs_list_empty(arg->list))
1824 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_bl_ast);
1826 /* nobody should touch l_bl_ast */
1827 lock_res_and_lock(lock);
1828 cfs_list_del_init(&lock->l_bl_ast);
1830 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1831 LASSERT(lock->l_bl_ast_run == 0);
1832 LASSERT(lock->l_blocking_lock);
1833 lock->l_bl_ast_run++;
1834 unlock_res_and_lock(lock);
1836 ldlm_lock2desc(lock->l_blocking_lock, &d);
1838 rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
1839 LDLM_LOCK_RELEASE(lock->l_blocking_lock);
1840 lock->l_blocking_lock = NULL;
1841 LDLM_LOCK_RELEASE(lock);
1847 * Process a call to completion AST callback for a lock in ast_work list
1850 ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1852 struct ldlm_cb_set_arg *arg = opaq;
1854 struct ldlm_lock *lock;
1855 ldlm_completion_callback completion_callback;
1858 if (cfs_list_empty(arg->list))
1861 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_cp_ast);
1863 /* It's possible to receive a completion AST before we've set
1864 * the l_completion_ast pointer: either because the AST arrived
1865 * before the reply, or simply because there's a small race
1866 * window between receiving the reply and finishing the local
1867 * enqueue. (bug 842)
1869 * This can't happen with the blocking_ast, however, because we
1870 * will never call the local blocking_ast until we drop our
1871 * reader/writer reference, which we won't do until we get the
1872 * reply and finish enqueueing. */
1874 /* nobody should touch l_cp_ast */
1875 lock_res_and_lock(lock);
1876 cfs_list_del_init(&lock->l_cp_ast);
1877 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1878 /* save l_completion_ast since it can be changed by
1879 * mds_intent_policy(), see bug 14225 */
1880 completion_callback = lock->l_completion_ast;
1881 lock->l_flags &= ~LDLM_FL_CP_REQD;
1882 unlock_res_and_lock(lock);
1884 if (completion_callback != NULL)
1885 rc = completion_callback(lock, 0, (void *)arg);
1886 LDLM_LOCK_RELEASE(lock);
1892 * Process a call to revocation AST callback for a lock in ast_work list
1895 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1897 struct ldlm_cb_set_arg *arg = opaq;
1898 struct ldlm_lock_desc desc;
1900 struct ldlm_lock *lock;
1903 if (cfs_list_empty(arg->list))
1906 lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_rk_ast);
1907 cfs_list_del_init(&lock->l_rk_ast);
1909 /* the desc just pretend to exclusive */
1910 ldlm_lock2desc(lock, &desc);
1911 desc.l_req_mode = LCK_EX;
1912 desc.l_granted_mode = 0;
1914 rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
1915 LDLM_LOCK_RELEASE(lock);
1921 * Process a call to glimpse AST callback for a lock in ast_work list
1923 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
1925 struct ldlm_cb_set_arg *arg = opaq;
1926 struct ldlm_glimpse_work *gl_work;
1927 struct ldlm_lock *lock;
1931 if (cfs_list_empty(arg->list))
1934 gl_work = cfs_list_entry(arg->list->next, struct ldlm_glimpse_work,
1936 cfs_list_del_init(&gl_work->gl_list);
1938 lock = gl_work->gl_lock;
1940 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
1941 arg->gl_desc = gl_work->gl_desc;
1943 /* invoke the actual glimpse callback */
1944 if (lock->l_glimpse_ast(lock, (void*)arg) == 0)
1947 LDLM_LOCK_RELEASE(lock);
1949 if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0)
1950 OBD_FREE_PTR(gl_work);
1956 * Process list of locks in need of ASTs being sent.
1958 * Used on server to send multiple ASTs together instead of sending one by
1961 int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list,
1962 ldlm_desc_ast_t ast_type)
1964 struct ldlm_cb_set_arg *arg;
1965 set_producer_func work_ast_lock;
1968 if (cfs_list_empty(rpc_list))
1975 cfs_atomic_set(&arg->restart, 0);
1976 arg->list = rpc_list;
1979 case LDLM_WORK_BL_AST:
1980 arg->type = LDLM_BL_CALLBACK;
1981 work_ast_lock = ldlm_work_bl_ast_lock;
1983 case LDLM_WORK_CP_AST:
1984 arg->type = LDLM_CP_CALLBACK;
1985 work_ast_lock = ldlm_work_cp_ast_lock;
1987 case LDLM_WORK_REVOKE_AST:
1988 arg->type = LDLM_BL_CALLBACK;
1989 work_ast_lock = ldlm_work_revoke_ast_lock;
1991 case LDLM_WORK_GL_AST:
1992 arg->type = LDLM_GL_CALLBACK;
1993 work_ast_lock = ldlm_work_gl_ast_lock;
1999 /* We create a ptlrpc request set with flow control extension.
2000 * This request set will use the work_ast_lock function to produce new
2001 * requests and will send a new request each time one completes in order
2002 * to keep the number of requests in flight to ns_max_parallel_ast */
2003 arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX,
2004 work_ast_lock, arg);
2005 if (arg->set == NULL)
2006 GOTO(out, rc = -ENOMEM);
2008 ptlrpc_set_wait(arg->set);
2009 ptlrpc_set_destroy(arg->set);
2011 rc = cfs_atomic_read(&arg->restart) ? -ERESTART : 0;
2018 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
2020 ldlm_reprocess_all(res);
2021 return LDLM_ITER_CONTINUE;
2024 static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2025 cfs_hlist_node_t *hnode, void *arg)
2027 struct ldlm_resource *res = cfs_hash_object(hs, hnode);
2030 rc = reprocess_one_queue(res, arg);
2032 return rc == LDLM_ITER_STOP;
2036 * Iterate through all resources on a namespace attempting to grant waiting
2039 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
2044 cfs_hash_for_each_nolock(ns->ns_rs_hash,
2045 ldlm_reprocess_res, NULL);
2049 EXPORT_SYMBOL(ldlm_reprocess_all_ns);
2052 * Try to grant all waiting locks on a resource.
2054 * Calls ldlm_reprocess_queue on converting and waiting queues.
2056 * Typically called after some resource locks are cancelled to see
2057 * if anything could be granted as a result of the cancellation.
2059 void ldlm_reprocess_all(struct ldlm_resource *res)
2061 CFS_LIST_HEAD(rpc_list);
2063 #ifdef HAVE_SERVER_SUPPORT
2066 /* Local lock trees don't get reprocessed. */
2067 if (ns_is_client(ldlm_res_to_ns(res))) {
2074 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
2075 if (rc == LDLM_ITER_CONTINUE)
2076 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
2079 rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
2081 if (rc == -ERESTART) {
2082 LASSERT(cfs_list_empty(&rpc_list));
2087 if (!ns_is_client(ldlm_res_to_ns(res))) {
2088 CERROR("This is client-side-only module, cannot handle "
2089 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2097 * Helper function to call blocking AST for LDLM lock \a lock in a
2098 * "cancelling" mode.
2100 void ldlm_cancel_callback(struct ldlm_lock *lock)
2102 check_res_locked(lock->l_resource);
2103 if (!(lock->l_flags & LDLM_FL_CANCEL)) {
2104 lock->l_flags |= LDLM_FL_CANCEL;
2105 if (lock->l_blocking_ast) {
2106 unlock_res_and_lock(lock);
2107 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
2109 lock_res_and_lock(lock);
2111 LDLM_DEBUG(lock, "no blocking ast");
2114 lock->l_flags |= LDLM_FL_BL_DONE;
2118 * Remove skiplist-enabled LDLM lock \a req from granted list
2120 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
2122 if (req->l_resource->lr_type != LDLM_PLAIN &&
2123 req->l_resource->lr_type != LDLM_IBITS)
2126 cfs_list_del_init(&req->l_sl_policy);
2127 cfs_list_del_init(&req->l_sl_mode);
2131 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2133 void ldlm_lock_cancel(struct ldlm_lock *lock)
2135 struct ldlm_resource *res;
2136 struct ldlm_namespace *ns;
2139 lock_res_and_lock(lock);
2141 res = lock->l_resource;
2142 ns = ldlm_res_to_ns(res);
2144 /* Please do not, no matter how tempting, remove this LBUG without
2145 * talking to me first. -phik */
2146 if (lock->l_readers || lock->l_writers) {
2147 LDLM_ERROR(lock, "lock still has references");
2151 if (lock->l_flags & LDLM_FL_WAITED)
2152 ldlm_del_waiting_lock(lock);
2154 /* Releases cancel callback. */
2155 ldlm_cancel_callback(lock);
2157 /* Yes, second time, just in case it was added again while we were
2158 * running with no res lock in ldlm_cancel_callback */
2159 if (lock->l_flags & LDLM_FL_WAITED)
2160 ldlm_del_waiting_lock(lock);
2162 ldlm_resource_unlink_lock(lock);
2163 ldlm_lock_destroy_nolock(lock);
2165 if (lock->l_granted_mode == lock->l_req_mode)
2166 ldlm_pool_del(&ns->ns_pool, lock);
2168 /* Make sure we will not be called again for same lock what is possible
2169 * if not to zero out lock->l_granted_mode */
2170 lock->l_granted_mode = LCK_MINMODE;
2171 unlock_res_and_lock(lock);
2175 EXPORT_SYMBOL(ldlm_lock_cancel);
2178 * Set opaque data into the lock that only makes sense to upper layer.
2180 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
2182 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
2187 if (lock->l_ast_data == NULL)
2188 lock->l_ast_data = data;
2189 if (lock->l_ast_data == data)
2191 LDLM_LOCK_PUT(lock);
2195 EXPORT_SYMBOL(ldlm_lock_set_data);
2197 struct export_cl_data {
2198 struct obd_export *ecl_exp;
2203 * Iterator function for ldlm_cancel_locks_for_export.
2204 * Cancels passed locks.
2206 int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
2207 cfs_hlist_node_t *hnode, void *data)
2210 struct export_cl_data *ecl = (struct export_cl_data *)data;
2211 struct obd_export *exp = ecl->ecl_exp;
2212 struct ldlm_lock *lock = cfs_hash_object(hs, hnode);
2213 struct ldlm_resource *res;
2215 res = ldlm_resource_getref(lock->l_resource);
2216 LDLM_LOCK_GET(lock);
2218 LDLM_DEBUG(lock, "export %p", exp);
2219 ldlm_res_lvbo_update(res, NULL, 1);
2220 ldlm_lock_cancel(lock);
2221 ldlm_reprocess_all(res);
2222 ldlm_resource_putref(res);
2223 LDLM_LOCK_RELEASE(lock);
2226 if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) {
2228 "Cancel lock %p for export %p (loop %d), still have "
2229 "%d locks left on hash table.\n",
2230 lock, exp, ecl->ecl_loop,
2231 cfs_atomic_read(&hs->hs_count));
2238 * Cancel all locks for given export.
2240 * Typically called on client disconnection/eviction
2242 void ldlm_cancel_locks_for_export(struct obd_export *exp)
2244 struct export_cl_data ecl = {
2249 cfs_hash_for_each_empty(exp->exp_lock_hash,
2250 ldlm_cancel_locks_for_export_cb, &ecl);
2254 * Downgrade an exclusive lock.
2256 * A fast variant of ldlm_lock_convert for convertion of exclusive
2257 * locks. The convertion is always successful.
2258 * Used by Commit on Sharing (COS) code.
2260 * \param lock A lock to convert
2261 * \param new_mode new lock mode
2263 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
2267 LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
2268 LASSERT(new_mode == LCK_COS);
2270 lock_res_and_lock(lock);
2271 ldlm_resource_unlink_lock(lock);
2273 * Remove the lock from pool as it will be added again in
2274 * ldlm_grant_lock() called below.
2276 ldlm_pool_del(&ldlm_lock_to_ns(lock)->ns_pool, lock);
2278 lock->l_req_mode = new_mode;
2279 ldlm_grant_lock(lock, NULL);
2280 unlock_res_and_lock(lock);
2281 ldlm_reprocess_all(lock->l_resource);
2285 EXPORT_SYMBOL(ldlm_lock_downgrade);
2288 * Attempt to convert already granted lock to a different mode.
2290 * While lock conversion is not currently used, future client-side
2291 * optimizations could take advantage of it to avoid discarding cached
2294 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
2297 CFS_LIST_HEAD(rpc_list);
2298 struct ldlm_resource *res;
2299 struct ldlm_namespace *ns;
2301 #ifdef HAVE_SERVER_SUPPORT
2303 struct sl_insert_point prev;
2305 struct ldlm_interval *node;
2308 /* Just return if mode is unchanged. */
2309 if (new_mode == lock->l_granted_mode) {
2310 *flags |= LDLM_FL_BLOCK_GRANTED;
2311 RETURN(lock->l_resource);
2314 /* I can't check the type of lock here because the bitlock of lock
2315 * is not held here, so do the allocation blindly. -jay */
2316 OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO);
2318 /* Actually, this causes LUSTRE_EDEADLK to be returned */
2321 LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
2322 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
2324 lock_res_and_lock(lock);
2326 res = lock->l_resource;
2327 ns = ldlm_res_to_ns(res);
2329 #ifdef HAVE_SERVER_SUPPORT
2330 old_mode = lock->l_req_mode;
2332 lock->l_req_mode = new_mode;
2333 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
2334 #ifdef HAVE_SERVER_SUPPORT
2335 /* remember the lock position where the lock might be
2336 * added back to the granted list later and also
2337 * remember the join mode for skiplist fixing. */
2338 prev.res_link = lock->l_res_link.prev;
2339 prev.mode_link = lock->l_sl_mode.prev;
2340 prev.policy_link = lock->l_sl_policy.prev;
2342 ldlm_resource_unlink_lock(lock);
2344 ldlm_resource_unlink_lock(lock);
2345 if (res->lr_type == LDLM_EXTENT) {
2346 /* FIXME: ugly code, I have to attach the lock to a
2347 * interval node again since perhaps it will be granted
2349 CFS_INIT_LIST_HEAD(&node->li_group);
2350 ldlm_interval_attach(node, lock);
2356 * Remove old lock from the pool before adding the lock with new
2357 * mode below in ->policy()
2359 ldlm_pool_del(&ns->ns_pool, lock);
2361 /* If this is a local resource, put it on the appropriate list. */
2362 if (ns_is_client(ldlm_res_to_ns(res))) {
2363 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
2364 ldlm_resource_add_lock(res, &res->lr_converting, lock);
2366 /* This should never happen, because of the way the
2367 * server handles conversions. */
2368 LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
2372 ldlm_grant_lock(lock, &rpc_list);
2374 /* FIXME: completion handling not with lr_lock held ! */
2375 if (lock->l_completion_ast)
2376 lock->l_completion_ast(lock, 0, NULL);
2378 #ifdef HAVE_SERVER_SUPPORT
2383 ldlm_processing_policy policy;
2384 policy = ldlm_processing_policy_table[res->lr_type];
2385 rc = policy(lock, &pflags, 0, &err, &rpc_list);
2386 if (rc == LDLM_ITER_STOP) {
2387 lock->l_req_mode = old_mode;
2388 if (res->lr_type == LDLM_EXTENT)
2389 ldlm_extent_add_lock(res, lock);
2391 ldlm_granted_list_add_lock(lock, &prev);
2395 *flags |= LDLM_FL_BLOCK_GRANTED;
2401 CERROR("This is client-side-only module, cannot handle "
2402 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2406 unlock_res_and_lock(lock);
2409 ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
2411 OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
2414 EXPORT_SYMBOL(ldlm_lock_convert);
2417 * Print lock with lock handle \a lockh description into debug log.
2419 * Used when printing all locks on a resource for debug purposes.
2421 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
2423 struct ldlm_lock *lock;
2425 if (!((libcfs_debug | D_ERROR) & level))
2428 lock = ldlm_handle2lock(lockh);
2432 LDLM_DEBUG_LIMIT(level, lock, "###");
2434 LDLM_LOCK_PUT(lock);
2436 EXPORT_SYMBOL(ldlm_lock_dump_handle);
2439 * Print lock information with custom message into debug log.
2442 void _ldlm_lock_debug(struct ldlm_lock *lock,
2443 struct libcfs_debug_msg_data *msgdata,
2444 const char *fmt, ...)
2447 struct obd_export *exp = lock->l_export;
2448 struct ldlm_resource *resource = lock->l_resource;
2449 char *nid = "local";
2451 va_start(args, fmt);
2453 if (exp && exp->exp_connection) {
2454 nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
2455 } else if (exp && exp->exp_obd != NULL) {
2456 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
2457 nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
2460 if (resource == NULL) {
2461 libcfs_debug_vmsg2(msgdata, fmt, args,
2462 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2463 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s "
2464 "remote: "LPX64" expref: %d pid: %u timeout: %lu "
2467 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2468 lock->l_readers, lock->l_writers,
2469 ldlm_lockname[lock->l_granted_mode],
2470 ldlm_lockname[lock->l_req_mode],
2471 lock->l_flags, nid, lock->l_remote_handle.cookie,
2472 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2473 lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type);
2478 switch (resource->lr_type) {
2480 libcfs_debug_vmsg2(msgdata, fmt, args,
2481 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2482 "res: "DLDLMRES" rrc: %d type: %s ["LPU64"->"LPU64"] "
2483 "(req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote: "
2484 LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2485 ldlm_lock_to_ns_name(lock), lock,
2486 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2487 lock->l_readers, lock->l_writers,
2488 ldlm_lockname[lock->l_granted_mode],
2489 ldlm_lockname[lock->l_req_mode],
2491 cfs_atomic_read(&resource->lr_refcount),
2492 ldlm_typename[resource->lr_type],
2493 lock->l_policy_data.l_extent.start,
2494 lock->l_policy_data.l_extent.end,
2495 lock->l_req_extent.start, lock->l_req_extent.end,
2496 lock->l_flags, nid, lock->l_remote_handle.cookie,
2497 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2498 lock->l_pid, lock->l_callback_timeout,
2503 libcfs_debug_vmsg2(msgdata, fmt, args,
2504 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2505 "res: "DLDLMRES" rrc: %d type: %s pid: %d "
2506 "["LPU64"->"LPU64"] flags: "LPX64" nid: %s "
2507 "remote: "LPX64" expref: %d pid: %u timeout: %lu\n",
2508 ldlm_lock_to_ns_name(lock), lock,
2509 lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc),
2510 lock->l_readers, lock->l_writers,
2511 ldlm_lockname[lock->l_granted_mode],
2512 ldlm_lockname[lock->l_req_mode],
2514 cfs_atomic_read(&resource->lr_refcount),
2515 ldlm_typename[resource->lr_type],
2516 lock->l_policy_data.l_flock.pid,
2517 lock->l_policy_data.l_flock.start,
2518 lock->l_policy_data.l_flock.end,
2519 lock->l_flags, nid, lock->l_remote_handle.cookie,
2520 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2521 lock->l_pid, lock->l_callback_timeout);
2525 libcfs_debug_vmsg2(msgdata, fmt, args,
2526 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2527 "res: "DLDLMRES" bits "LPX64" rrc: %d type: %s "
2528 "flags: "LPX64" nid: %s remote: "LPX64" expref: %d "
2529 "pid: %u timeout: %lu lvb_type: %d\n",
2530 ldlm_lock_to_ns_name(lock),
2531 lock, lock->l_handle.h_cookie,
2532 cfs_atomic_read(&lock->l_refc),
2533 lock->l_readers, lock->l_writers,
2534 ldlm_lockname[lock->l_granted_mode],
2535 ldlm_lockname[lock->l_req_mode],
2537 lock->l_policy_data.l_inodebits.bits,
2538 cfs_atomic_read(&resource->lr_refcount),
2539 ldlm_typename[resource->lr_type],
2540 lock->l_flags, nid, lock->l_remote_handle.cookie,
2541 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2542 lock->l_pid, lock->l_callback_timeout,
2547 libcfs_debug_vmsg2(msgdata, fmt, args,
2548 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
2549 "res: "DLDLMRES" rrc: %d type: %s flags: "LPX64" "
2550 "nid: %s remote: "LPX64" expref: %d pid: %u "
2551 "timeout: %lu lvb_type: %d\n",
2552 ldlm_lock_to_ns_name(lock),
2553 lock, lock->l_handle.h_cookie,
2554 cfs_atomic_read(&lock->l_refc),
2555 lock->l_readers, lock->l_writers,
2556 ldlm_lockname[lock->l_granted_mode],
2557 ldlm_lockname[lock->l_req_mode],
2559 cfs_atomic_read(&resource->lr_refcount),
2560 ldlm_typename[resource->lr_type],
2561 lock->l_flags, nid, lock->l_remote_handle.cookie,
2562 exp ? cfs_atomic_read(&exp->exp_refcount) : -99,
2563 lock->l_pid, lock->l_callback_timeout,
2569 EXPORT_SYMBOL(_ldlm_lock_debug);