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, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/ldlm/ldlm_lock.c
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Phil Schwan <phil@clusterfs.com>
42 #define DEBUG_SUBSYSTEM S_LDLM
44 # include <linux/libcfs/libcfs.h>
45 # include <linux/lustre_intent.h>
47 #include <obd_class.h>
48 #include "ldlm_internal.h"
51 char *ldlm_lockname
[] = {
62 EXPORT_SYMBOL(ldlm_lockname
);
64 char *ldlm_typename
[] = {
70 EXPORT_SYMBOL(ldlm_typename
);
72 static ldlm_policy_wire_to_local_t ldlm_policy_wire18_to_local
[] = {
73 [LDLM_PLAIN
- LDLM_MIN_TYPE
] ldlm_plain_policy_wire_to_local
,
74 [LDLM_EXTENT
- LDLM_MIN_TYPE
] ldlm_extent_policy_wire_to_local
,
75 [LDLM_FLOCK
- LDLM_MIN_TYPE
] ldlm_flock_policy_wire18_to_local
,
76 [LDLM_IBITS
- LDLM_MIN_TYPE
] ldlm_ibits_policy_wire_to_local
,
79 static ldlm_policy_wire_to_local_t ldlm_policy_wire21_to_local
[] = {
80 [LDLM_PLAIN
- LDLM_MIN_TYPE
] ldlm_plain_policy_wire_to_local
,
81 [LDLM_EXTENT
- LDLM_MIN_TYPE
] ldlm_extent_policy_wire_to_local
,
82 [LDLM_FLOCK
- LDLM_MIN_TYPE
] ldlm_flock_policy_wire21_to_local
,
83 [LDLM_IBITS
- LDLM_MIN_TYPE
] ldlm_ibits_policy_wire_to_local
,
86 static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire
[] = {
87 [LDLM_PLAIN
- LDLM_MIN_TYPE
] ldlm_plain_policy_local_to_wire
,
88 [LDLM_EXTENT
- LDLM_MIN_TYPE
] ldlm_extent_policy_local_to_wire
,
89 [LDLM_FLOCK
- LDLM_MIN_TYPE
] ldlm_flock_policy_local_to_wire
,
90 [LDLM_IBITS
- LDLM_MIN_TYPE
] ldlm_ibits_policy_local_to_wire
,
94 * Converts lock policy from local format to on the wire lock_desc format
96 void ldlm_convert_policy_to_wire(ldlm_type_t type
,
97 const ldlm_policy_data_t
*lpolicy
,
98 ldlm_wire_policy_data_t
*wpolicy
)
100 ldlm_policy_local_to_wire_t convert
;
102 convert
= ldlm_policy_local_to_wire
[type
- LDLM_MIN_TYPE
];
104 convert(lpolicy
, wpolicy
);
108 * Converts lock policy from on the wire lock_desc format to local format
110 void ldlm_convert_policy_to_local(struct obd_export
*exp
, ldlm_type_t type
,
111 const ldlm_wire_policy_data_t
*wpolicy
,
112 ldlm_policy_data_t
*lpolicy
)
114 ldlm_policy_wire_to_local_t convert
;
117 /** some badness for 2.0.0 clients, but 2.0.0 isn't supported */
118 new_client
= (exp_connect_flags(exp
) & OBD_CONNECT_FULL20
) != 0;
120 convert
= ldlm_policy_wire21_to_local
[type
- LDLM_MIN_TYPE
];
122 convert
= ldlm_policy_wire18_to_local
[type
- LDLM_MIN_TYPE
];
124 convert(wpolicy
, lpolicy
);
127 char *ldlm_it2str(int it
)
134 case (IT_OPEN
| IT_CREAT
):
149 CERROR("Unknown intent %d\n", it
);
153 EXPORT_SYMBOL(ldlm_it2str
);
155 extern struct kmem_cache
*ldlm_lock_slab
;
158 void ldlm_register_intent(struct ldlm_namespace
*ns
, ldlm_res_policy arg
)
162 EXPORT_SYMBOL(ldlm_register_intent
);
165 * REFCOUNTED LOCK OBJECTS
170 * Get a reference on a lock.
172 * Lock refcounts, during creation:
173 * - one special one for allocation, dec'd only once in destroy
174 * - one for being a lock that's in-use
175 * - one for the addref associated with a new lock
177 struct ldlm_lock
*ldlm_lock_get(struct ldlm_lock
*lock
)
179 atomic_inc(&lock
->l_refc
);
182 EXPORT_SYMBOL(ldlm_lock_get
);
185 * Release lock reference.
187 * Also frees the lock if it was last reference.
189 void ldlm_lock_put(struct ldlm_lock
*lock
)
193 LASSERT(lock
->l_resource
!= LP_POISON
);
194 LASSERT(atomic_read(&lock
->l_refc
) > 0);
195 if (atomic_dec_and_test(&lock
->l_refc
)) {
196 struct ldlm_resource
*res
;
199 "final lock_put on destroyed lock, freeing it.");
201 res
= lock
->l_resource
;
202 LASSERT(lock
->l_destroyed
);
203 LASSERT(list_empty(&lock
->l_res_link
));
204 LASSERT(list_empty(&lock
->l_pending_chain
));
206 lprocfs_counter_decr(ldlm_res_to_ns(res
)->ns_stats
,
208 lu_ref_del(&res
->lr_reference
, "lock", lock
);
209 ldlm_resource_putref(res
);
210 lock
->l_resource
= NULL
;
211 if (lock
->l_export
) {
212 class_export_lock_put(lock
->l_export
, lock
);
213 lock
->l_export
= NULL
;
216 if (lock
->l_lvb_data
!= NULL
)
217 OBD_FREE(lock
->l_lvb_data
, lock
->l_lvb_len
);
219 ldlm_interval_free(ldlm_interval_detach(lock
));
220 lu_ref_fini(&lock
->l_reference
);
221 OBD_FREE_RCU(lock
, sizeof(*lock
), &lock
->l_handle
);
226 EXPORT_SYMBOL(ldlm_lock_put
);
229 * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked.
231 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock
*lock
)
234 if (!list_empty(&lock
->l_lru
)) {
235 struct ldlm_namespace
*ns
= ldlm_lock_to_ns(lock
);
237 LASSERT(lock
->l_resource
->lr_type
!= LDLM_FLOCK
);
238 list_del_init(&lock
->l_lru
);
239 if (lock
->l_flags
& LDLM_FL_SKIPPED
)
240 lock
->l_flags
&= ~LDLM_FL_SKIPPED
;
241 LASSERT(ns
->ns_nr_unused
> 0);
249 * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first.
251 int ldlm_lock_remove_from_lru(struct ldlm_lock
*lock
)
253 struct ldlm_namespace
*ns
= ldlm_lock_to_ns(lock
);
257 if (lock
->l_ns_srv
) {
258 LASSERT(list_empty(&lock
->l_lru
));
262 spin_lock(&ns
->ns_lock
);
263 rc
= ldlm_lock_remove_from_lru_nolock(lock
);
264 spin_unlock(&ns
->ns_lock
);
270 * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked.
272 void ldlm_lock_add_to_lru_nolock(struct ldlm_lock
*lock
)
274 struct ldlm_namespace
*ns
= ldlm_lock_to_ns(lock
);
276 lock
->l_last_used
= cfs_time_current();
277 LASSERT(list_empty(&lock
->l_lru
));
278 LASSERT(lock
->l_resource
->lr_type
!= LDLM_FLOCK
);
279 list_add_tail(&lock
->l_lru
, &ns
->ns_unused_list
);
280 LASSERT(ns
->ns_nr_unused
>= 0);
285 * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks
288 void ldlm_lock_add_to_lru(struct ldlm_lock
*lock
)
290 struct ldlm_namespace
*ns
= ldlm_lock_to_ns(lock
);
293 spin_lock(&ns
->ns_lock
);
294 ldlm_lock_add_to_lru_nolock(lock
);
295 spin_unlock(&ns
->ns_lock
);
300 * Moves LDLM lock \a lock that is already in namespace LRU to the tail of
301 * the LRU. Performs necessary LRU locking
303 void ldlm_lock_touch_in_lru(struct ldlm_lock
*lock
)
305 struct ldlm_namespace
*ns
= ldlm_lock_to_ns(lock
);
308 if (lock
->l_ns_srv
) {
309 LASSERT(list_empty(&lock
->l_lru
));
314 spin_lock(&ns
->ns_lock
);
315 if (!list_empty(&lock
->l_lru
)) {
316 ldlm_lock_remove_from_lru_nolock(lock
);
317 ldlm_lock_add_to_lru_nolock(lock
);
319 spin_unlock(&ns
->ns_lock
);
324 * Helper to destroy a locked lock.
326 * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock
327 * Must be called with l_lock and lr_lock held.
329 * Does not actually free the lock data, but rather marks the lock as
330 * destroyed by setting l_destroyed field in the lock to 1. Destroys a
331 * handle->lock association too, so that the lock can no longer be found
332 * and removes the lock from LRU list. Actual lock freeing occurs when
333 * last lock reference goes away.
335 * Original comment (of some historical value):
336 * This used to have a 'strict' flag, which recovery would use to mark an
337 * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
338 * shall explain why it's gone: with the new hash table scheme, once you call
339 * ldlm_lock_destroy, you can never drop your final references on this lock.
340 * Because it's not in the hash table anymore. -phil
342 int ldlm_lock_destroy_internal(struct ldlm_lock
*lock
)
346 if (lock
->l_readers
|| lock
->l_writers
) {
347 LDLM_ERROR(lock
, "lock still has references");
351 if (!list_empty(&lock
->l_res_link
)) {
352 LDLM_ERROR(lock
, "lock still on resource");
356 if (lock
->l_destroyed
) {
357 LASSERT(list_empty(&lock
->l_lru
));
361 lock
->l_destroyed
= 1;
363 if (lock
->l_export
&& lock
->l_export
->exp_lock_hash
) {
364 /* NB: it's safe to call cfs_hash_del() even lock isn't
365 * in exp_lock_hash. */
366 /* In the function below, .hs_keycmp resolves to
367 * ldlm_export_lock_keycmp() */
368 /* coverity[overrun-buffer-val] */
369 cfs_hash_del(lock
->l_export
->exp_lock_hash
,
370 &lock
->l_remote_handle
, &lock
->l_exp_hash
);
373 ldlm_lock_remove_from_lru(lock
);
374 class_handle_unhash(&lock
->l_handle
);
377 /* Wake anyone waiting for this lock */
378 /* FIXME: I should probably add yet another flag, instead of using
379 * l_export to only call this on clients */
381 class_export_put(lock
->l_export
);
382 lock
->l_export
= NULL
;
383 if (lock
->l_export
&& lock
->l_completion_ast
)
384 lock
->l_completion_ast(lock
, 0);
391 * Destroys a LDLM lock \a lock. Performs necessary locking first.
393 void ldlm_lock_destroy(struct ldlm_lock
*lock
)
397 lock_res_and_lock(lock
);
398 first
= ldlm_lock_destroy_internal(lock
);
399 unlock_res_and_lock(lock
);
401 /* drop reference from hashtable only for first destroy */
403 lu_ref_del(&lock
->l_reference
, "hash", lock
);
404 LDLM_LOCK_RELEASE(lock
);
410 * Destroys a LDLM lock \a lock that is already locked.
412 void ldlm_lock_destroy_nolock(struct ldlm_lock
*lock
)
416 first
= ldlm_lock_destroy_internal(lock
);
417 /* drop reference from hashtable only for first destroy */
419 lu_ref_del(&lock
->l_reference
, "hash", lock
);
420 LDLM_LOCK_RELEASE(lock
);
425 /* this is called by portals_handle2object with the handle lock taken */
426 static void lock_handle_addref(void *lock
)
428 LDLM_LOCK_GET((struct ldlm_lock
*)lock
);
431 static void lock_handle_free(void *lock
, int size
)
433 LASSERT(size
== sizeof(struct ldlm_lock
));
434 OBD_SLAB_FREE(lock
, ldlm_lock_slab
, size
);
437 struct portals_handle_ops lock_handle_ops
= {
438 .hop_addref
= lock_handle_addref
,
439 .hop_free
= lock_handle_free
,
444 * Allocate and initialize new lock structure.
446 * usage: pass in a resource on which you have done ldlm_resource_get
447 * new lock will take over the refcount.
448 * returns: lock with refcount 2 - one for current caller and one for remote
450 static struct ldlm_lock
*ldlm_lock_new(struct ldlm_resource
*resource
)
452 struct ldlm_lock
*lock
;
455 if (resource
== NULL
)
458 OBD_SLAB_ALLOC_PTR_GFP(lock
, ldlm_lock_slab
, __GFP_IO
);
462 spin_lock_init(&lock
->l_lock
);
463 lock
->l_resource
= resource
;
464 lu_ref_add(&resource
->lr_reference
, "lock", lock
);
466 atomic_set(&lock
->l_refc
, 2);
467 INIT_LIST_HEAD(&lock
->l_res_link
);
468 INIT_LIST_HEAD(&lock
->l_lru
);
469 INIT_LIST_HEAD(&lock
->l_pending_chain
);
470 INIT_LIST_HEAD(&lock
->l_bl_ast
);
471 INIT_LIST_HEAD(&lock
->l_cp_ast
);
472 INIT_LIST_HEAD(&lock
->l_rk_ast
);
473 init_waitqueue_head(&lock
->l_waitq
);
474 lock
->l_blocking_lock
= NULL
;
475 INIT_LIST_HEAD(&lock
->l_sl_mode
);
476 INIT_LIST_HEAD(&lock
->l_sl_policy
);
477 INIT_HLIST_NODE(&lock
->l_exp_hash
);
478 INIT_HLIST_NODE(&lock
->l_exp_flock_hash
);
480 lprocfs_counter_incr(ldlm_res_to_ns(resource
)->ns_stats
,
482 INIT_LIST_HEAD(&lock
->l_handle
.h_link
);
483 class_handle_hash(&lock
->l_handle
, &lock_handle_ops
);
485 lu_ref_init(&lock
->l_reference
);
486 lu_ref_add(&lock
->l_reference
, "hash", lock
);
487 lock
->l_callback_timeout
= 0;
489 #if LUSTRE_TRACKS_LOCK_EXP_REFS
490 INIT_LIST_HEAD(&lock
->l_exp_refs_link
);
491 lock
->l_exp_refs_nr
= 0;
492 lock
->l_exp_refs_target
= NULL
;
494 INIT_LIST_HEAD(&lock
->l_exp_list
);
500 * Moves LDLM lock \a lock to another resource.
501 * This is used on client when server returns some other lock than requested
502 * (typically as a result of intent operation)
504 int ldlm_lock_change_resource(struct ldlm_namespace
*ns
, struct ldlm_lock
*lock
,
505 const struct ldlm_res_id
*new_resid
)
507 struct ldlm_resource
*oldres
= lock
->l_resource
;
508 struct ldlm_resource
*newres
;
512 LASSERT(ns_is_client(ns
));
514 lock_res_and_lock(lock
);
515 if (memcmp(new_resid
, &lock
->l_resource
->lr_name
,
516 sizeof(lock
->l_resource
->lr_name
)) == 0) {
518 unlock_res_and_lock(lock
);
522 LASSERT(new_resid
->name
[0] != 0);
524 /* This function assumes that the lock isn't on any lists */
525 LASSERT(list_empty(&lock
->l_res_link
));
527 type
= oldres
->lr_type
;
528 unlock_res_and_lock(lock
);
530 newres
= ldlm_resource_get(ns
, NULL
, new_resid
, type
, 1);
534 lu_ref_add(&newres
->lr_reference
, "lock", lock
);
536 * To flip the lock from the old to the new resource, lock, oldres and
537 * newres have to be locked. Resource spin-locks are nested within
538 * lock->l_lock, and are taken in the memory address order to avoid
541 spin_lock(&lock
->l_lock
);
542 oldres
= lock
->l_resource
;
543 if (oldres
< newres
) {
545 lock_res_nested(newres
, LRT_NEW
);
548 lock_res_nested(oldres
, LRT_NEW
);
550 LASSERT(memcmp(new_resid
, &oldres
->lr_name
,
551 sizeof oldres
->lr_name
) != 0);
552 lock
->l_resource
= newres
;
554 unlock_res_and_lock(lock
);
556 /* ...and the flowers are still standing! */
557 lu_ref_del(&oldres
->lr_reference
, "lock", lock
);
558 ldlm_resource_putref(oldres
);
562 EXPORT_SYMBOL(ldlm_lock_change_resource
);
564 /** \defgroup ldlm_handles LDLM HANDLES
565 * Ways to get hold of locks without any addresses.
570 * Fills in handle for LDLM lock \a lock into supplied \a lockh
571 * Does not take any references.
573 void ldlm_lock2handle(const struct ldlm_lock
*lock
, struct lustre_handle
*lockh
)
575 lockh
->cookie
= lock
->l_handle
.h_cookie
;
577 EXPORT_SYMBOL(ldlm_lock2handle
);
580 * Obtain a lock reference by handle.
582 * if \a flags: atomically get the lock and set the flags.
583 * Return NULL if flag already set
585 struct ldlm_lock
*__ldlm_handle2lock(const struct lustre_handle
*handle
,
588 struct ldlm_lock
*lock
;
593 lock
= class_handle2object(handle
->cookie
);
597 /* It's unlikely but possible that someone marked the lock as
598 * destroyed after we did handle2object on it */
599 if (flags
== 0 && !lock
->l_destroyed
) {
600 lu_ref_add(&lock
->l_reference
, "handle", current
);
604 lock_res_and_lock(lock
);
606 LASSERT(lock
->l_resource
!= NULL
);
608 lu_ref_add_atomic(&lock
->l_reference
, "handle", current
);
609 if (unlikely(lock
->l_destroyed
)) {
610 unlock_res_and_lock(lock
);
611 CDEBUG(D_INFO
, "lock already destroyed: lock %p\n", lock
);
616 if (flags
&& (lock
->l_flags
& flags
)) {
617 unlock_res_and_lock(lock
);
623 lock
->l_flags
|= flags
;
625 unlock_res_and_lock(lock
);
628 EXPORT_SYMBOL(__ldlm_handle2lock
);
629 /** @} ldlm_handles */
632 * Fill in "on the wire" representation for given LDLM lock into supplied
633 * lock descriptor \a desc structure.
635 void ldlm_lock2desc(struct ldlm_lock
*lock
, struct ldlm_lock_desc
*desc
)
637 struct obd_export
*exp
= lock
->l_export
?: lock
->l_conn_export
;
639 /* INODEBITS_INTEROP: If the other side does not support
640 * inodebits, reply with a plain lock descriptor. */
641 if ((lock
->l_resource
->lr_type
== LDLM_IBITS
) &&
642 (exp
&& !(exp_connect_flags(exp
) & OBD_CONNECT_IBITS
))) {
643 /* Make sure all the right bits are set in this lock we
644 are going to pass to client */
645 LASSERTF(lock
->l_policy_data
.l_inodebits
.bits
==
646 (MDS_INODELOCK_LOOKUP
| MDS_INODELOCK_UPDATE
|
647 MDS_INODELOCK_LAYOUT
),
648 "Inappropriate inode lock bits during "
649 "conversion " LPU64
"\n",
650 lock
->l_policy_data
.l_inodebits
.bits
);
652 ldlm_res2desc(lock
->l_resource
, &desc
->l_resource
);
653 desc
->l_resource
.lr_type
= LDLM_PLAIN
;
655 /* Convert "new" lock mode to something old client can
657 if ((lock
->l_req_mode
== LCK_CR
) ||
658 (lock
->l_req_mode
== LCK_CW
))
659 desc
->l_req_mode
= LCK_PR
;
661 desc
->l_req_mode
= lock
->l_req_mode
;
662 if ((lock
->l_granted_mode
== LCK_CR
) ||
663 (lock
->l_granted_mode
== LCK_CW
)) {
664 desc
->l_granted_mode
= LCK_PR
;
666 /* We never grant PW/EX locks to clients */
667 LASSERT((lock
->l_granted_mode
!= LCK_PW
) &&
668 (lock
->l_granted_mode
!= LCK_EX
));
669 desc
->l_granted_mode
= lock
->l_granted_mode
;
672 /* We do not copy policy here, because there is no
673 policy for plain locks */
675 ldlm_res2desc(lock
->l_resource
, &desc
->l_resource
);
676 desc
->l_req_mode
= lock
->l_req_mode
;
677 desc
->l_granted_mode
= lock
->l_granted_mode
;
678 ldlm_convert_policy_to_wire(lock
->l_resource
->lr_type
,
679 &lock
->l_policy_data
,
680 &desc
->l_policy_data
);
683 EXPORT_SYMBOL(ldlm_lock2desc
);
686 * Add a lock to list of conflicting locks to send AST to.
688 * Only add if we have not sent a blocking AST to the lock yet.
690 void ldlm_add_bl_work_item(struct ldlm_lock
*lock
, struct ldlm_lock
*new,
691 struct list_head
*work_list
)
693 if ((lock
->l_flags
& LDLM_FL_AST_SENT
) == 0) {
694 LDLM_DEBUG(lock
, "lock incompatible; sending blocking AST.");
695 lock
->l_flags
|= LDLM_FL_AST_SENT
;
696 /* If the enqueuing client said so, tell the AST recipient to
697 * discard dirty data, rather than writing back. */
698 if (new->l_flags
& LDLM_AST_DISCARD_DATA
)
699 lock
->l_flags
|= LDLM_FL_DISCARD_DATA
;
700 LASSERT(list_empty(&lock
->l_bl_ast
));
701 list_add(&lock
->l_bl_ast
, work_list
);
703 LASSERT(lock
->l_blocking_lock
== NULL
);
704 lock
->l_blocking_lock
= LDLM_LOCK_GET(new);
709 * Add a lock to list of just granted locks to send completion AST to.
711 void ldlm_add_cp_work_item(struct ldlm_lock
*lock
, struct list_head
*work_list
)
713 if ((lock
->l_flags
& LDLM_FL_CP_REQD
) == 0) {
714 lock
->l_flags
|= LDLM_FL_CP_REQD
;
715 LDLM_DEBUG(lock
, "lock granted; sending completion AST.");
716 LASSERT(list_empty(&lock
->l_cp_ast
));
717 list_add(&lock
->l_cp_ast
, work_list
);
723 * Aggregator function to add AST work items into a list. Determines
724 * what sort of an AST work needs to be done and calls the proper
726 * Must be called with lr_lock held.
728 void ldlm_add_ast_work_item(struct ldlm_lock
*lock
, struct ldlm_lock
*new,
729 struct list_head
*work_list
)
732 check_res_locked(lock
->l_resource
);
734 ldlm_add_bl_work_item(lock
, new, work_list
);
736 ldlm_add_cp_work_item(lock
, work_list
);
741 * Add specified reader/writer reference to LDLM lock with handle \a lockh.
742 * r/w reference type is determined by \a mode
743 * Calls ldlm_lock_addref_internal.
745 void ldlm_lock_addref(struct lustre_handle
*lockh
, __u32 mode
)
747 struct ldlm_lock
*lock
;
749 lock
= ldlm_handle2lock(lockh
);
750 LASSERT(lock
!= NULL
);
751 ldlm_lock_addref_internal(lock
, mode
);
754 EXPORT_SYMBOL(ldlm_lock_addref
);
758 * Add specified reader/writer reference to LDLM lock \a lock.
759 * r/w reference type is determined by \a mode
760 * Removes lock from LRU if it is there.
761 * Assumes the LDLM lock is already locked.
763 void ldlm_lock_addref_internal_nolock(struct ldlm_lock
*lock
, __u32 mode
)
765 ldlm_lock_remove_from_lru(lock
);
766 if (mode
& (LCK_NL
| LCK_CR
| LCK_PR
)) {
768 lu_ref_add_atomic(&lock
->l_reference
, "reader", lock
);
770 if (mode
& (LCK_EX
| LCK_CW
| LCK_PW
| LCK_GROUP
| LCK_COS
)) {
772 lu_ref_add_atomic(&lock
->l_reference
, "writer", lock
);
775 lu_ref_add_atomic(&lock
->l_reference
, "user", lock
);
776 LDLM_DEBUG(lock
, "ldlm_lock_addref(%s)", ldlm_lockname
[mode
]);
780 * Attempts to add reader/writer reference to a lock with handle \a lockh, and
781 * fails if lock is already LDLM_FL_CBPENDING or destroyed.
783 * \retval 0 success, lock was addref-ed
785 * \retval -EAGAIN lock is being canceled.
787 int ldlm_lock_addref_try(struct lustre_handle
*lockh
, __u32 mode
)
789 struct ldlm_lock
*lock
;
793 lock
= ldlm_handle2lock(lockh
);
795 lock_res_and_lock(lock
);
796 if (lock
->l_readers
!= 0 || lock
->l_writers
!= 0 ||
797 !(lock
->l_flags
& LDLM_FL_CBPENDING
)) {
798 ldlm_lock_addref_internal_nolock(lock
, mode
);
801 unlock_res_and_lock(lock
);
806 EXPORT_SYMBOL(ldlm_lock_addref_try
);
809 * Add specified reader/writer reference to LDLM lock \a lock.
810 * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work.
811 * Only called for local locks.
813 void ldlm_lock_addref_internal(struct ldlm_lock
*lock
, __u32 mode
)
815 lock_res_and_lock(lock
);
816 ldlm_lock_addref_internal_nolock(lock
, mode
);
817 unlock_res_and_lock(lock
);
821 * Removes reader/writer reference for LDLM lock \a lock.
822 * Assumes LDLM lock is already locked.
823 * only called in ldlm_flock_destroy and for local locks.
824 * Does NOT add lock to LRU if no r/w references left to accomodate flock locks
825 * that cannot be placed in LRU.
827 void ldlm_lock_decref_internal_nolock(struct ldlm_lock
*lock
, __u32 mode
)
829 LDLM_DEBUG(lock
, "ldlm_lock_decref(%s)", ldlm_lockname
[mode
]);
830 if (mode
& (LCK_NL
| LCK_CR
| LCK_PR
)) {
831 LASSERT(lock
->l_readers
> 0);
832 lu_ref_del(&lock
->l_reference
, "reader", lock
);
835 if (mode
& (LCK_EX
| LCK_CW
| LCK_PW
| LCK_GROUP
| LCK_COS
)) {
836 LASSERT(lock
->l_writers
> 0);
837 lu_ref_del(&lock
->l_reference
, "writer", lock
);
841 lu_ref_del(&lock
->l_reference
, "user", lock
);
842 LDLM_LOCK_RELEASE(lock
); /* matches the LDLM_LOCK_GET() in addref */
846 * Removes reader/writer reference for LDLM lock \a lock.
847 * Locks LDLM lock first.
848 * If the lock is determined to be client lock on a client and r/w refcount
849 * drops to zero and the lock is not blocked, the lock is added to LRU lock
851 * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called.
853 void ldlm_lock_decref_internal(struct ldlm_lock
*lock
, __u32 mode
)
855 struct ldlm_namespace
*ns
;
858 lock_res_and_lock(lock
);
860 ns
= ldlm_lock_to_ns(lock
);
862 ldlm_lock_decref_internal_nolock(lock
, mode
);
864 if (lock
->l_flags
& LDLM_FL_LOCAL
&&
865 !lock
->l_readers
&& !lock
->l_writers
) {
866 /* If this is a local lock on a server namespace and this was
867 * the last reference, cancel the lock. */
868 CDEBUG(D_INFO
, "forcing cancel of local lock\n");
869 lock
->l_flags
|= LDLM_FL_CBPENDING
;
872 if (!lock
->l_readers
&& !lock
->l_writers
&&
873 (lock
->l_flags
& LDLM_FL_CBPENDING
)) {
874 /* If we received a blocked AST and this was the last reference,
875 * run the callback. */
876 if (lock
->l_ns_srv
&& lock
->l_export
)
877 CERROR("FL_CBPENDING set on non-local lock--just a "
880 LDLM_DEBUG(lock
, "final decref done on cbpending lock");
882 LDLM_LOCK_GET(lock
); /* dropped by bl thread */
883 ldlm_lock_remove_from_lru(lock
);
884 unlock_res_and_lock(lock
);
886 if (lock
->l_flags
& LDLM_FL_FAIL_LOC
)
887 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE
);
889 if ((lock
->l_flags
& LDLM_FL_ATOMIC_CB
) ||
890 ldlm_bl_to_thread_lock(ns
, NULL
, lock
) != 0)
891 ldlm_handle_bl_callback(ns
, NULL
, lock
);
892 } else if (ns_is_client(ns
) &&
893 !lock
->l_readers
&& !lock
->l_writers
&&
894 !(lock
->l_flags
& LDLM_FL_NO_LRU
) &&
895 !(lock
->l_flags
& LDLM_FL_BL_AST
)) {
897 LDLM_DEBUG(lock
, "add lock into lru list");
899 /* If this is a client-side namespace and this was the last
900 * reference, put it on the LRU. */
901 ldlm_lock_add_to_lru(lock
);
902 unlock_res_and_lock(lock
);
904 if (lock
->l_flags
& LDLM_FL_FAIL_LOC
)
905 OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE
);
907 /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
908 * are not supported by the server, otherwise, it is done on
910 if (!exp_connect_cancelset(lock
->l_conn_export
) &&
911 !ns_connect_lru_resize(ns
))
912 ldlm_cancel_lru(ns
, 0, LCF_ASYNC
, 0);
914 LDLM_DEBUG(lock
, "do not add lock into lru list");
915 unlock_res_and_lock(lock
);
922 * Decrease reader/writer refcount for LDLM lock with handle \a lockh
924 void ldlm_lock_decref(struct lustre_handle
*lockh
, __u32 mode
)
926 struct ldlm_lock
*lock
= __ldlm_handle2lock(lockh
, 0);
927 LASSERTF(lock
!= NULL
, "Non-existing lock: "LPX64
"\n", lockh
->cookie
);
928 ldlm_lock_decref_internal(lock
, mode
);
931 EXPORT_SYMBOL(ldlm_lock_decref
);
934 * Decrease reader/writer refcount for LDLM lock with handle
935 * \a lockh and mark it for subsequent cancellation once r/w refcount
936 * drops to zero instead of putting into LRU.
938 * Typical usage is for GROUP locks which we cannot allow to be cached.
940 void ldlm_lock_decref_and_cancel(struct lustre_handle
*lockh
, __u32 mode
)
942 struct ldlm_lock
*lock
= __ldlm_handle2lock(lockh
, 0);
945 LASSERT(lock
!= NULL
);
947 LDLM_DEBUG(lock
, "ldlm_lock_decref(%s)", ldlm_lockname
[mode
]);
948 lock_res_and_lock(lock
);
949 lock
->l_flags
|= LDLM_FL_CBPENDING
;
950 unlock_res_and_lock(lock
);
951 ldlm_lock_decref_internal(lock
, mode
);
954 EXPORT_SYMBOL(ldlm_lock_decref_and_cancel
);
956 struct sl_insert_point
{
957 struct list_head
*res_link
;
958 struct list_head
*mode_link
;
959 struct list_head
*policy_link
;
963 * Finds a position to insert the new lock into granted lock list.
965 * Used for locks eligible for skiplist optimization.
968 * queue [input]: the granted list where search acts on;
969 * req [input]: the lock whose position to be located;
970 * prev [output]: positions within 3 lists to insert @req to
974 * - ldlm_grant_lock_with_skiplist
976 static void search_granted_lock(struct list_head
*queue
,
977 struct ldlm_lock
*req
,
978 struct sl_insert_point
*prev
)
980 struct list_head
*tmp
;
981 struct ldlm_lock
*lock
, *mode_end
, *policy_end
;
984 list_for_each(tmp
, queue
) {
985 lock
= list_entry(tmp
, struct ldlm_lock
, l_res_link
);
987 mode_end
= list_entry(lock
->l_sl_mode
.prev
,
988 struct ldlm_lock
, l_sl_mode
);
990 if (lock
->l_req_mode
!= req
->l_req_mode
) {
991 /* jump to last lock of mode group */
992 tmp
= &mode_end
->l_res_link
;
996 /* suitable mode group is found */
997 if (lock
->l_resource
->lr_type
== LDLM_PLAIN
) {
998 /* insert point is last lock of the mode group */
999 prev
->res_link
= &mode_end
->l_res_link
;
1000 prev
->mode_link
= &mode_end
->l_sl_mode
;
1001 prev
->policy_link
= &req
->l_sl_policy
;
1004 } else if (lock
->l_resource
->lr_type
== LDLM_IBITS
) {
1007 list_entry(lock
->l_sl_policy
.prev
,
1011 if (lock
->l_policy_data
.l_inodebits
.bits
==
1012 req
->l_policy_data
.l_inodebits
.bits
) {
1013 /* insert point is last lock of
1014 * the policy group */
1016 &policy_end
->l_res_link
;
1018 &policy_end
->l_sl_mode
;
1020 &policy_end
->l_sl_policy
;
1025 if (policy_end
== mode_end
)
1026 /* done with mode group */
1029 /* go to next policy group within mode group */
1030 tmp
= policy_end
->l_res_link
.next
;
1031 lock
= list_entry(tmp
, struct ldlm_lock
,
1033 } /* loop over policy groups within the mode group */
1035 /* insert point is last lock of the mode group,
1036 * new policy group is started */
1037 prev
->res_link
= &mode_end
->l_res_link
;
1038 prev
->mode_link
= &mode_end
->l_sl_mode
;
1039 prev
->policy_link
= &req
->l_sl_policy
;
1043 LDLM_ERROR(lock
,"is not LDLM_PLAIN or LDLM_IBITS lock");
1048 /* insert point is last lock on the queue,
1049 * new mode group and new policy group are started */
1050 prev
->res_link
= queue
->prev
;
1051 prev
->mode_link
= &req
->l_sl_mode
;
1052 prev
->policy_link
= &req
->l_sl_policy
;
1058 * Add a lock into resource granted list after a position described by
1061 static void ldlm_granted_list_add_lock(struct ldlm_lock
*lock
,
1062 struct sl_insert_point
*prev
)
1064 struct ldlm_resource
*res
= lock
->l_resource
;
1067 check_res_locked(res
);
1069 ldlm_resource_dump(D_INFO
, res
);
1070 LDLM_DEBUG(lock
, "About to add lock:");
1072 if (lock
->l_destroyed
) {
1073 CDEBUG(D_OTHER
, "Lock destroyed, not adding to resource\n");
1077 LASSERT(list_empty(&lock
->l_res_link
));
1078 LASSERT(list_empty(&lock
->l_sl_mode
));
1079 LASSERT(list_empty(&lock
->l_sl_policy
));
1082 * lock->link == prev->link means lock is first starting the group.
1083 * Don't re-add to itself to suppress kernel warnings.
1085 if (&lock
->l_res_link
!= prev
->res_link
)
1086 list_add(&lock
->l_res_link
, prev
->res_link
);
1087 if (&lock
->l_sl_mode
!= prev
->mode_link
)
1088 list_add(&lock
->l_sl_mode
, prev
->mode_link
);
1089 if (&lock
->l_sl_policy
!= prev
->policy_link
)
1090 list_add(&lock
->l_sl_policy
, prev
->policy_link
);
1096 * Add a lock to granted list on a resource maintaining skiplist
1099 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock
*lock
)
1101 struct sl_insert_point prev
;
1104 LASSERT(lock
->l_req_mode
== lock
->l_granted_mode
);
1106 search_granted_lock(&lock
->l_resource
->lr_granted
, lock
, &prev
);
1107 ldlm_granted_list_add_lock(lock
, &prev
);
1112 * Perform lock granting bookkeeping.
1114 * Includes putting the lock into granted list and updating lock mode.
1116 * - ldlm_lock_enqueue
1117 * - ldlm_reprocess_queue
1118 * - ldlm_lock_convert
1120 * must be called with lr_lock held
1122 void ldlm_grant_lock(struct ldlm_lock
*lock
, struct list_head
*work_list
)
1124 struct ldlm_resource
*res
= lock
->l_resource
;
1127 check_res_locked(res
);
1129 lock
->l_granted_mode
= lock
->l_req_mode
;
1130 if (res
->lr_type
== LDLM_PLAIN
|| res
->lr_type
== LDLM_IBITS
)
1131 ldlm_grant_lock_with_skiplist(lock
);
1132 else if (res
->lr_type
== LDLM_EXTENT
)
1133 ldlm_extent_add_lock(res
, lock
);
1135 ldlm_resource_add_lock(res
, &res
->lr_granted
, lock
);
1137 if (lock
->l_granted_mode
< res
->lr_most_restr
)
1138 res
->lr_most_restr
= lock
->l_granted_mode
;
1140 if (work_list
&& lock
->l_completion_ast
!= NULL
)
1141 ldlm_add_ast_work_item(lock
, NULL
, work_list
);
1143 ldlm_pool_add(&ldlm_res_to_ns(res
)->ns_pool
, lock
);
1148 * Search for a lock with given properties in a queue.
1150 * \retval a referenced lock or NULL. See the flag descriptions below, in the
1151 * comment above ldlm_lock_match
1153 static struct ldlm_lock
*search_queue(struct list_head
*queue
,
1155 ldlm_policy_data_t
*policy
,
1156 struct ldlm_lock
*old_lock
,
1157 __u64 flags
, int unref
)
1159 struct ldlm_lock
*lock
;
1160 struct list_head
*tmp
;
1162 list_for_each(tmp
, queue
) {
1165 lock
= list_entry(tmp
, struct ldlm_lock
, l_res_link
);
1167 if (lock
== old_lock
)
1170 /* llite sometimes wants to match locks that will be
1171 * canceled when their users drop, but we allow it to match
1172 * if it passes in CBPENDING and the lock still has users.
1173 * this is generally only going to be used by children
1174 * whose parents already hold a lock so forward progress
1175 * can still happen. */
1176 if (lock
->l_flags
& LDLM_FL_CBPENDING
&&
1177 !(flags
& LDLM_FL_CBPENDING
))
1179 if (!unref
&& lock
->l_flags
& LDLM_FL_CBPENDING
&&
1180 lock
->l_readers
== 0 && lock
->l_writers
== 0)
1183 if (!(lock
->l_req_mode
& *mode
))
1185 match
= lock
->l_req_mode
;
1187 if (lock
->l_resource
->lr_type
== LDLM_EXTENT
&&
1188 (lock
->l_policy_data
.l_extent
.start
>
1189 policy
->l_extent
.start
||
1190 lock
->l_policy_data
.l_extent
.end
< policy
->l_extent
.end
))
1193 if (unlikely(match
== LCK_GROUP
) &&
1194 lock
->l_resource
->lr_type
== LDLM_EXTENT
&&
1195 lock
->l_policy_data
.l_extent
.gid
!= policy
->l_extent
.gid
)
1198 /* We match if we have existing lock with same or wider set
1200 if (lock
->l_resource
->lr_type
== LDLM_IBITS
&&
1201 ((lock
->l_policy_data
.l_inodebits
.bits
&
1202 policy
->l_inodebits
.bits
) !=
1203 policy
->l_inodebits
.bits
))
1207 (lock
->l_destroyed
|| lock
->l_flags
& LDLM_FL_FAILED
||
1211 if ((flags
& LDLM_FL_LOCAL_ONLY
) &&
1212 !(lock
->l_flags
& LDLM_FL_LOCAL
))
1215 if (flags
& LDLM_FL_TEST_LOCK
) {
1216 LDLM_LOCK_GET(lock
);
1217 ldlm_lock_touch_in_lru(lock
);
1219 ldlm_lock_addref_internal_nolock(lock
, match
);
1228 void ldlm_lock_fail_match_locked(struct ldlm_lock
*lock
)
1230 if (!lock
->l_failed
) {
1232 wake_up_all(&lock
->l_waitq
);
1235 EXPORT_SYMBOL(ldlm_lock_fail_match_locked
);
1237 void ldlm_lock_fail_match(struct ldlm_lock
*lock
)
1239 lock_res_and_lock(lock
);
1240 ldlm_lock_fail_match_locked(lock
);
1241 unlock_res_and_lock(lock
);
1243 EXPORT_SYMBOL(ldlm_lock_fail_match
);
1246 * Mark lock as "matchable" by OST.
1248 * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB
1250 * Assumes LDLM lock is already locked.
1252 void ldlm_lock_allow_match_locked(struct ldlm_lock
*lock
)
1254 lock
->l_flags
|= LDLM_FL_LVB_READY
;
1255 wake_up_all(&lock
->l_waitq
);
1257 EXPORT_SYMBOL(ldlm_lock_allow_match_locked
);
1260 * Mark lock as "matchable" by OST.
1261 * Locks the lock and then \see ldlm_lock_allow_match_locked
1263 void ldlm_lock_allow_match(struct ldlm_lock
*lock
)
1265 lock_res_and_lock(lock
);
1266 ldlm_lock_allow_match_locked(lock
);
1267 unlock_res_and_lock(lock
);
1269 EXPORT_SYMBOL(ldlm_lock_allow_match
);
1272 * Attempt to find a lock with specified properties.
1274 * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is
1277 * Can be called in two ways:
1279 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
1280 * for a duplicate of.
1282 * Otherwise, all of the fields must be filled in, to match against.
1284 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
1285 * server (ie, connh is NULL)
1286 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
1287 * list will be considered
1288 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
1289 * to be canceled can still be matched as long as they still have reader
1290 * or writer refernces
1291 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
1292 * just tell us if we would have matched.
1294 * \retval 1 if it finds an already-existing lock that is compatible; in this
1295 * case, lockh is filled in with a addref()ed lock
1297 * We also check security context, and if that fails we simply return 0 (to
1298 * keep caller code unchanged), the context failure will be discovered by
1299 * caller sometime later.
1301 ldlm_mode_t
ldlm_lock_match(struct ldlm_namespace
*ns
, __u64 flags
,
1302 const struct ldlm_res_id
*res_id
, ldlm_type_t type
,
1303 ldlm_policy_data_t
*policy
, ldlm_mode_t mode
,
1304 struct lustre_handle
*lockh
, int unref
)
1306 struct ldlm_resource
*res
;
1307 struct ldlm_lock
*lock
, *old_lock
= NULL
;
1312 old_lock
= ldlm_handle2lock(lockh
);
1315 ns
= ldlm_lock_to_ns(old_lock
);
1316 res_id
= &old_lock
->l_resource
->lr_name
;
1317 type
= old_lock
->l_resource
->lr_type
;
1318 mode
= old_lock
->l_req_mode
;
1321 res
= ldlm_resource_get(ns
, NULL
, res_id
, type
, 0);
1323 LASSERT(old_lock
== NULL
);
1327 LDLM_RESOURCE_ADDREF(res
);
1330 lock
= search_queue(&res
->lr_granted
, &mode
, policy
, old_lock
,
1334 if (flags
& LDLM_FL_BLOCK_GRANTED
)
1336 lock
= search_queue(&res
->lr_converting
, &mode
, policy
, old_lock
,
1340 lock
= search_queue(&res
->lr_waiting
, &mode
, policy
, old_lock
,
1348 LDLM_RESOURCE_DELREF(res
);
1349 ldlm_resource_putref(res
);
1352 ldlm_lock2handle(lock
, lockh
);
1353 if ((flags
& LDLM_FL_LVB_READY
) &&
1354 (!(lock
->l_flags
& LDLM_FL_LVB_READY
))) {
1355 struct l_wait_info lwi
;
1356 if (lock
->l_completion_ast
) {
1357 int err
= lock
->l_completion_ast(lock
,
1358 LDLM_FL_WAIT_NOREPROC
,
1361 if (flags
& LDLM_FL_TEST_LOCK
)
1362 LDLM_LOCK_RELEASE(lock
);
1364 ldlm_lock_decref_internal(lock
,
1371 lwi
= LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout
),
1372 NULL
, LWI_ON_SIGNAL_NOOP
, NULL
);
1374 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1375 l_wait_event(lock
->l_waitq
,
1376 lock
->l_flags
& LDLM_FL_LVB_READY
||
1377 lock
->l_destroyed
|| lock
->l_failed
,
1379 if (!(lock
->l_flags
& LDLM_FL_LVB_READY
)) {
1380 if (flags
& LDLM_FL_TEST_LOCK
)
1381 LDLM_LOCK_RELEASE(lock
);
1383 ldlm_lock_decref_internal(lock
, mode
);
1390 LDLM_DEBUG(lock
, "matched ("LPU64
" "LPU64
")",
1391 (type
== LDLM_PLAIN
|| type
== LDLM_IBITS
) ?
1392 res_id
->name
[2] : policy
->l_extent
.start
,
1393 (type
== LDLM_PLAIN
|| type
== LDLM_IBITS
) ?
1394 res_id
->name
[3] : policy
->l_extent
.end
);
1396 /* check user's security context */
1397 if (lock
->l_conn_export
&&
1398 sptlrpc_import_check_ctx(
1399 class_exp2cliimp(lock
->l_conn_export
))) {
1400 if (!(flags
& LDLM_FL_TEST_LOCK
))
1401 ldlm_lock_decref_internal(lock
, mode
);
1405 if (flags
& LDLM_FL_TEST_LOCK
)
1406 LDLM_LOCK_RELEASE(lock
);
1408 } else if (!(flags
& LDLM_FL_TEST_LOCK
)) {/*less verbose for test-only*/
1409 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1410 LPU64
"/"LPU64
" ("LPU64
" "LPU64
")", ns
,
1411 type
, mode
, res_id
->name
[0], res_id
->name
[1],
1412 (type
== LDLM_PLAIN
|| type
== LDLM_IBITS
) ?
1413 res_id
->name
[2] :policy
->l_extent
.start
,
1414 (type
== LDLM_PLAIN
|| type
== LDLM_IBITS
) ?
1415 res_id
->name
[3] : policy
->l_extent
.end
);
1418 LDLM_LOCK_PUT(old_lock
);
1420 return rc
? mode
: 0;
1422 EXPORT_SYMBOL(ldlm_lock_match
);
1424 ldlm_mode_t
ldlm_revalidate_lock_handle(struct lustre_handle
*lockh
,
1427 struct ldlm_lock
*lock
;
1428 ldlm_mode_t mode
= 0;
1431 lock
= ldlm_handle2lock(lockh
);
1433 lock_res_and_lock(lock
);
1434 if (lock
->l_destroyed
|| lock
->l_flags
& LDLM_FL_FAILED
||
1438 if (lock
->l_flags
& LDLM_FL_CBPENDING
&&
1439 lock
->l_readers
== 0 && lock
->l_writers
== 0)
1443 *bits
= lock
->l_policy_data
.l_inodebits
.bits
;
1444 mode
= lock
->l_granted_mode
;
1445 ldlm_lock_addref_internal_nolock(lock
, mode
);
1452 unlock_res_and_lock(lock
);
1453 LDLM_LOCK_PUT(lock
);
1457 EXPORT_SYMBOL(ldlm_revalidate_lock_handle
);
1459 /** The caller must guarantee that the buffer is large enough. */
1460 int ldlm_fill_lvb(struct ldlm_lock
*lock
, struct req_capsule
*pill
,
1461 enum req_location loc
, void *data
, int size
)
1466 LASSERT(data
!= NULL
);
1469 switch (lock
->l_lvb_type
) {
1471 if (size
== sizeof(struct ost_lvb
)) {
1472 if (loc
== RCL_CLIENT
)
1473 lvb
= req_capsule_client_swab_get(pill
,
1475 lustre_swab_ost_lvb
);
1477 lvb
= req_capsule_server_swab_get(pill
,
1479 lustre_swab_ost_lvb
);
1480 if (unlikely(lvb
== NULL
)) {
1481 LDLM_ERROR(lock
, "no LVB");
1485 memcpy(data
, lvb
, size
);
1486 } else if (size
== sizeof(struct ost_lvb_v1
)) {
1487 struct ost_lvb
*olvb
= data
;
1489 if (loc
== RCL_CLIENT
)
1490 lvb
= req_capsule_client_swab_get(pill
,
1492 lustre_swab_ost_lvb_v1
);
1494 lvb
= req_capsule_server_sized_swab_get(pill
,
1496 lustre_swab_ost_lvb_v1
);
1497 if (unlikely(lvb
== NULL
)) {
1498 LDLM_ERROR(lock
, "no LVB");
1502 memcpy(data
, lvb
, size
);
1503 olvb
->lvb_mtime_ns
= 0;
1504 olvb
->lvb_atime_ns
= 0;
1505 olvb
->lvb_ctime_ns
= 0;
1507 LDLM_ERROR(lock
, "Replied unexpected ost LVB size %d",
1513 if (size
== sizeof(struct lquota_lvb
)) {
1514 if (loc
== RCL_CLIENT
)
1515 lvb
= req_capsule_client_swab_get(pill
,
1517 lustre_swab_lquota_lvb
);
1519 lvb
= req_capsule_server_swab_get(pill
,
1521 lustre_swab_lquota_lvb
);
1522 if (unlikely(lvb
== NULL
)) {
1523 LDLM_ERROR(lock
, "no LVB");
1527 memcpy(data
, lvb
, size
);
1529 LDLM_ERROR(lock
, "Replied unexpected lquota LVB size %d",
1538 if (loc
== RCL_CLIENT
)
1539 lvb
= req_capsule_client_get(pill
, &RMF_DLM_LVB
);
1541 lvb
= req_capsule_server_get(pill
, &RMF_DLM_LVB
);
1542 if (unlikely(lvb
== NULL
)) {
1543 LDLM_ERROR(lock
, "no LVB");
1547 memcpy(data
, lvb
, size
);
1550 LDLM_ERROR(lock
, "Unknown LVB type: %d\n", lock
->l_lvb_type
);
1551 libcfs_debug_dumpstack(NULL
);
1559 * Create and fill in new LDLM lock with specified properties.
1560 * Returns a referenced lock
1562 struct ldlm_lock
*ldlm_lock_create(struct ldlm_namespace
*ns
,
1563 const struct ldlm_res_id
*res_id
,
1566 const struct ldlm_callback_suite
*cbs
,
1567 void *data
, __u32 lvb_len
,
1568 enum lvb_type lvb_type
)
1570 struct ldlm_lock
*lock
;
1571 struct ldlm_resource
*res
;
1574 res
= ldlm_resource_get(ns
, NULL
, res_id
, type
, 1);
1578 lock
= ldlm_lock_new(res
);
1583 lock
->l_req_mode
= mode
;
1584 lock
->l_ast_data
= data
;
1585 lock
->l_pid
= current_pid();
1586 lock
->l_ns_srv
= !!ns_is_server(ns
);
1588 lock
->l_blocking_ast
= cbs
->lcs_blocking
;
1589 lock
->l_completion_ast
= cbs
->lcs_completion
;
1590 lock
->l_glimpse_ast
= cbs
->lcs_glimpse
;
1591 lock
->l_weigh_ast
= cbs
->lcs_weigh
;
1594 lock
->l_tree_node
= NULL
;
1595 /* if this is the extent lock, allocate the interval tree node */
1596 if (type
== LDLM_EXTENT
) {
1597 if (ldlm_interval_alloc(lock
) == NULL
)
1602 lock
->l_lvb_len
= lvb_len
;
1603 OBD_ALLOC(lock
->l_lvb_data
, lvb_len
);
1604 if (lock
->l_lvb_data
== NULL
)
1608 lock
->l_lvb_type
= lvb_type
;
1609 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK
))
1615 ldlm_lock_destroy(lock
);
1616 LDLM_LOCK_RELEASE(lock
);
1621 * Enqueue (request) a lock.
1623 * Does not block. As a result of enqueue the lock would be put
1624 * into granted or waiting list.
1626 * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag
1627 * set, skip all the enqueueing and delegate lock processing to intent policy
1630 ldlm_error_t
ldlm_lock_enqueue(struct ldlm_namespace
*ns
,
1631 struct ldlm_lock
**lockp
,
1632 void *cookie
, __u64
*flags
)
1634 struct ldlm_lock
*lock
= *lockp
;
1635 struct ldlm_resource
*res
= lock
->l_resource
;
1636 int local
= ns_is_client(ldlm_res_to_ns(res
));
1637 ldlm_error_t rc
= ELDLM_OK
;
1638 struct ldlm_interval
*node
= NULL
;
1641 lock
->l_last_activity
= cfs_time_current_sec();
1642 /* policies are not executed on the client or during replay */
1643 if ((*flags
& (LDLM_FL_HAS_INTENT
|LDLM_FL_REPLAY
)) == LDLM_FL_HAS_INTENT
1644 && !local
&& ns
->ns_policy
) {
1645 rc
= ns
->ns_policy(ns
, lockp
, cookie
, lock
->l_req_mode
, *flags
,
1647 if (rc
== ELDLM_LOCK_REPLACED
) {
1648 /* The lock that was returned has already been granted,
1649 * and placed into lockp. If it's not the same as the
1650 * one we passed in, then destroy the old one and our
1651 * work here is done. */
1652 if (lock
!= *lockp
) {
1653 ldlm_lock_destroy(lock
);
1654 LDLM_LOCK_RELEASE(lock
);
1656 *flags
|= LDLM_FL_LOCK_CHANGED
;
1658 } else if (rc
!= ELDLM_OK
||
1659 (rc
== ELDLM_OK
&& (*flags
& LDLM_FL_INTENT_ONLY
))) {
1660 ldlm_lock_destroy(lock
);
1665 /* For a replaying lock, it might be already in granted list. So
1666 * unlinking the lock will cause the interval node to be freed, we
1667 * have to allocate the interval node early otherwise we can't regrant
1668 * this lock in the future. - jay */
1669 if (!local
&& (*flags
& LDLM_FL_REPLAY
) && res
->lr_type
== LDLM_EXTENT
)
1670 OBD_SLAB_ALLOC_PTR_GFP(node
, ldlm_interval_slab
, __GFP_IO
);
1672 lock_res_and_lock(lock
);
1673 if (local
&& lock
->l_req_mode
== lock
->l_granted_mode
) {
1674 /* The server returned a blocked lock, but it was granted
1675 * before we got a chance to actually enqueue it. We don't
1676 * need to do anything else. */
1677 *flags
&= ~(LDLM_FL_BLOCK_GRANTED
|
1678 LDLM_FL_BLOCK_CONV
| LDLM_FL_BLOCK_WAIT
);
1679 GOTO(out
, ELDLM_OK
);
1682 ldlm_resource_unlink_lock(lock
);
1683 if (res
->lr_type
== LDLM_EXTENT
&& lock
->l_tree_node
== NULL
) {
1685 ldlm_lock_destroy_nolock(lock
);
1686 GOTO(out
, rc
= -ENOMEM
);
1689 INIT_LIST_HEAD(&node
->li_group
);
1690 ldlm_interval_attach(node
, lock
);
1694 /* Some flags from the enqueue want to make it into the AST, via the
1695 * lock's l_flags. */
1696 lock
->l_flags
|= *flags
& LDLM_AST_DISCARD_DATA
;
1698 /* This distinction between local lock trees is very important; a client
1699 * namespace only has information about locks taken by that client, and
1700 * thus doesn't have enough information to decide for itself if it can
1701 * be granted (below). In this case, we do exactly what the server
1702 * tells us to do, as dictated by the 'flags'.
1704 * We do exactly the same thing during recovery, when the server is
1705 * more or less trusting the clients not to lie.
1707 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1708 * granted/converting queues. */
1710 if (*flags
& LDLM_FL_BLOCK_CONV
)
1711 ldlm_resource_add_lock(res
, &res
->lr_converting
, lock
);
1712 else if (*flags
& (LDLM_FL_BLOCK_WAIT
| LDLM_FL_BLOCK_GRANTED
))
1713 ldlm_resource_add_lock(res
, &res
->lr_waiting
, lock
);
1715 ldlm_grant_lock(lock
, NULL
);
1716 GOTO(out
, ELDLM_OK
);
1718 CERROR("This is client-side-only module, cannot handle "
1719 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1724 unlock_res_and_lock(lock
);
1726 OBD_SLAB_FREE(node
, ldlm_interval_slab
, sizeof(*node
));
1732 * Process a call to blocking AST callback for a lock in ast_work list
1735 ldlm_work_bl_ast_lock(struct ptlrpc_request_set
*rqset
, void *opaq
)
1737 struct ldlm_cb_set_arg
*arg
= opaq
;
1738 struct ldlm_lock_desc d
;
1740 struct ldlm_lock
*lock
;
1743 if (list_empty(arg
->list
))
1746 lock
= list_entry(arg
->list
->next
, struct ldlm_lock
, l_bl_ast
);
1748 /* nobody should touch l_bl_ast */
1749 lock_res_and_lock(lock
);
1750 list_del_init(&lock
->l_bl_ast
);
1752 LASSERT(lock
->l_flags
& LDLM_FL_AST_SENT
);
1753 LASSERT(lock
->l_bl_ast_run
== 0);
1754 LASSERT(lock
->l_blocking_lock
);
1755 lock
->l_bl_ast_run
++;
1756 unlock_res_and_lock(lock
);
1758 ldlm_lock2desc(lock
->l_blocking_lock
, &d
);
1760 rc
= lock
->l_blocking_ast(lock
, &d
, (void *)arg
, LDLM_CB_BLOCKING
);
1761 LDLM_LOCK_RELEASE(lock
->l_blocking_lock
);
1762 lock
->l_blocking_lock
= NULL
;
1763 LDLM_LOCK_RELEASE(lock
);
1769 * Process a call to completion AST callback for a lock in ast_work list
1772 ldlm_work_cp_ast_lock(struct ptlrpc_request_set
*rqset
, void *opaq
)
1774 struct ldlm_cb_set_arg
*arg
= opaq
;
1776 struct ldlm_lock
*lock
;
1777 ldlm_completion_callback completion_callback
;
1780 if (list_empty(arg
->list
))
1783 lock
= list_entry(arg
->list
->next
, struct ldlm_lock
, l_cp_ast
);
1785 /* It's possible to receive a completion AST before we've set
1786 * the l_completion_ast pointer: either because the AST arrived
1787 * before the reply, or simply because there's a small race
1788 * window between receiving the reply and finishing the local
1789 * enqueue. (bug 842)
1791 * This can't happen with the blocking_ast, however, because we
1792 * will never call the local blocking_ast until we drop our
1793 * reader/writer reference, which we won't do until we get the
1794 * reply and finish enqueueing. */
1796 /* nobody should touch l_cp_ast */
1797 lock_res_and_lock(lock
);
1798 list_del_init(&lock
->l_cp_ast
);
1799 LASSERT(lock
->l_flags
& LDLM_FL_CP_REQD
);
1800 /* save l_completion_ast since it can be changed by
1801 * mds_intent_policy(), see bug 14225 */
1802 completion_callback
= lock
->l_completion_ast
;
1803 lock
->l_flags
&= ~LDLM_FL_CP_REQD
;
1804 unlock_res_and_lock(lock
);
1806 if (completion_callback
!= NULL
)
1807 rc
= completion_callback(lock
, 0, (void *)arg
);
1808 LDLM_LOCK_RELEASE(lock
);
1814 * Process a call to revocation AST callback for a lock in ast_work list
1817 ldlm_work_revoke_ast_lock(struct ptlrpc_request_set
*rqset
, void *opaq
)
1819 struct ldlm_cb_set_arg
*arg
= opaq
;
1820 struct ldlm_lock_desc desc
;
1822 struct ldlm_lock
*lock
;
1825 if (list_empty(arg
->list
))
1828 lock
= list_entry(arg
->list
->next
, struct ldlm_lock
, l_rk_ast
);
1829 list_del_init(&lock
->l_rk_ast
);
1831 /* the desc just pretend to exclusive */
1832 ldlm_lock2desc(lock
, &desc
);
1833 desc
.l_req_mode
= LCK_EX
;
1834 desc
.l_granted_mode
= 0;
1836 rc
= lock
->l_blocking_ast(lock
, &desc
, (void*)arg
, LDLM_CB_BLOCKING
);
1837 LDLM_LOCK_RELEASE(lock
);
1843 * Process a call to glimpse AST callback for a lock in ast_work list
1845 int ldlm_work_gl_ast_lock(struct ptlrpc_request_set
*rqset
, void *opaq
)
1847 struct ldlm_cb_set_arg
*arg
= opaq
;
1848 struct ldlm_glimpse_work
*gl_work
;
1849 struct ldlm_lock
*lock
;
1853 if (list_empty(arg
->list
))
1856 gl_work
= list_entry(arg
->list
->next
, struct ldlm_glimpse_work
,
1858 list_del_init(&gl_work
->gl_list
);
1860 lock
= gl_work
->gl_lock
;
1862 /* transfer the glimpse descriptor to ldlm_cb_set_arg */
1863 arg
->gl_desc
= gl_work
->gl_desc
;
1865 /* invoke the actual glimpse callback */
1866 if (lock
->l_glimpse_ast(lock
, (void*)arg
) == 0)
1869 LDLM_LOCK_RELEASE(lock
);
1871 if ((gl_work
->gl_flags
& LDLM_GL_WORK_NOFREE
) == 0)
1872 OBD_FREE_PTR(gl_work
);
1878 * Process list of locks in need of ASTs being sent.
1880 * Used on server to send multiple ASTs together instead of sending one by
1883 int ldlm_run_ast_work(struct ldlm_namespace
*ns
, struct list_head
*rpc_list
,
1884 ldlm_desc_ast_t ast_type
)
1886 struct ldlm_cb_set_arg
*arg
;
1887 set_producer_func work_ast_lock
;
1890 if (list_empty(rpc_list
))
1897 atomic_set(&arg
->restart
, 0);
1898 arg
->list
= rpc_list
;
1901 case LDLM_WORK_BL_AST
:
1902 arg
->type
= LDLM_BL_CALLBACK
;
1903 work_ast_lock
= ldlm_work_bl_ast_lock
;
1905 case LDLM_WORK_CP_AST
:
1906 arg
->type
= LDLM_CP_CALLBACK
;
1907 work_ast_lock
= ldlm_work_cp_ast_lock
;
1909 case LDLM_WORK_REVOKE_AST
:
1910 arg
->type
= LDLM_BL_CALLBACK
;
1911 work_ast_lock
= ldlm_work_revoke_ast_lock
;
1913 case LDLM_WORK_GL_AST
:
1914 arg
->type
= LDLM_GL_CALLBACK
;
1915 work_ast_lock
= ldlm_work_gl_ast_lock
;
1921 /* We create a ptlrpc request set with flow control extension.
1922 * This request set will use the work_ast_lock function to produce new
1923 * requests and will send a new request each time one completes in order
1924 * to keep the number of requests in flight to ns_max_parallel_ast */
1925 arg
->set
= ptlrpc_prep_fcset(ns
->ns_max_parallel_ast
? : UINT_MAX
,
1926 work_ast_lock
, arg
);
1927 if (arg
->set
== NULL
)
1928 GOTO(out
, rc
= -ENOMEM
);
1930 ptlrpc_set_wait(arg
->set
);
1931 ptlrpc_set_destroy(arg
->set
);
1933 rc
= atomic_read(&arg
->restart
) ? -ERESTART
: 0;
1940 static int reprocess_one_queue(struct ldlm_resource
*res
, void *closure
)
1942 ldlm_reprocess_all(res
);
1943 return LDLM_ITER_CONTINUE
;
1946 static int ldlm_reprocess_res(cfs_hash_t
*hs
, cfs_hash_bd_t
*bd
,
1947 struct hlist_node
*hnode
, void *arg
)
1949 struct ldlm_resource
*res
= cfs_hash_object(hs
, hnode
);
1952 rc
= reprocess_one_queue(res
, arg
);
1954 return rc
== LDLM_ITER_STOP
;
1958 * Iterate through all resources on a namespace attempting to grant waiting
1961 void ldlm_reprocess_all_ns(struct ldlm_namespace
*ns
)
1966 cfs_hash_for_each_nolock(ns
->ns_rs_hash
,
1967 ldlm_reprocess_res
, NULL
);
1971 EXPORT_SYMBOL(ldlm_reprocess_all_ns
);
1974 * Try to grant all waiting locks on a resource.
1976 * Calls ldlm_reprocess_queue on converting and waiting queues.
1978 * Typically called after some resource locks are cancelled to see
1979 * if anything could be granted as a result of the cancellation.
1981 void ldlm_reprocess_all(struct ldlm_resource
*res
)
1983 LIST_HEAD(rpc_list
);
1986 if (!ns_is_client(ldlm_res_to_ns(res
))) {
1987 CERROR("This is client-side-only module, cannot handle "
1988 "LDLM_NAMESPACE_SERVER resource type lock.\n");
1995 * Helper function to call blocking AST for LDLM lock \a lock in a
1996 * "cancelling" mode.
1998 void ldlm_cancel_callback(struct ldlm_lock
*lock
)
2000 check_res_locked(lock
->l_resource
);
2001 if (!(lock
->l_flags
& LDLM_FL_CANCEL
)) {
2002 lock
->l_flags
|= LDLM_FL_CANCEL
;
2003 if (lock
->l_blocking_ast
) {
2004 unlock_res_and_lock(lock
);
2005 lock
->l_blocking_ast(lock
, NULL
, lock
->l_ast_data
,
2007 lock_res_and_lock(lock
);
2009 LDLM_DEBUG(lock
, "no blocking ast");
2012 lock
->l_flags
|= LDLM_FL_BL_DONE
;
2016 * Remove skiplist-enabled LDLM lock \a req from granted list
2018 void ldlm_unlink_lock_skiplist(struct ldlm_lock
*req
)
2020 if (req
->l_resource
->lr_type
!= LDLM_PLAIN
&&
2021 req
->l_resource
->lr_type
!= LDLM_IBITS
)
2024 list_del_init(&req
->l_sl_policy
);
2025 list_del_init(&req
->l_sl_mode
);
2029 * Attempts to cancel LDLM lock \a lock that has no reader/writer references.
2031 void ldlm_lock_cancel(struct ldlm_lock
*lock
)
2033 struct ldlm_resource
*res
;
2034 struct ldlm_namespace
*ns
;
2037 lock_res_and_lock(lock
);
2039 res
= lock
->l_resource
;
2040 ns
= ldlm_res_to_ns(res
);
2042 /* Please do not, no matter how tempting, remove this LBUG without
2043 * talking to me first. -phik */
2044 if (lock
->l_readers
|| lock
->l_writers
) {
2045 LDLM_ERROR(lock
, "lock still has references");
2050 ldlm_del_waiting_lock(lock
);
2052 /* Releases cancel callback. */
2053 ldlm_cancel_callback(lock
);
2055 /* Yes, second time, just in case it was added again while we were
2056 running with no res lock in ldlm_cancel_callback */
2058 ldlm_del_waiting_lock(lock
);
2060 ldlm_resource_unlink_lock(lock
);
2061 ldlm_lock_destroy_nolock(lock
);
2063 if (lock
->l_granted_mode
== lock
->l_req_mode
)
2064 ldlm_pool_del(&ns
->ns_pool
, lock
);
2066 /* Make sure we will not be called again for same lock what is possible
2067 * if not to zero out lock->l_granted_mode */
2068 lock
->l_granted_mode
= LCK_MINMODE
;
2069 unlock_res_and_lock(lock
);
2073 EXPORT_SYMBOL(ldlm_lock_cancel
);
2076 * Set opaque data into the lock that only makes sense to upper layer.
2078 int ldlm_lock_set_data(struct lustre_handle
*lockh
, void *data
)
2080 struct ldlm_lock
*lock
= ldlm_handle2lock(lockh
);
2085 if (lock
->l_ast_data
== NULL
)
2086 lock
->l_ast_data
= data
;
2087 if (lock
->l_ast_data
== data
)
2089 LDLM_LOCK_PUT(lock
);
2093 EXPORT_SYMBOL(ldlm_lock_set_data
);
2095 struct export_cl_data
{
2096 struct obd_export
*ecl_exp
;
2101 * Iterator function for ldlm_cancel_locks_for_export.
2102 * Cancels passed locks.
2104 int ldlm_cancel_locks_for_export_cb(cfs_hash_t
*hs
, cfs_hash_bd_t
*bd
,
2105 struct hlist_node
*hnode
, void *data
)
2108 struct export_cl_data
*ecl
= (struct export_cl_data
*)data
;
2109 struct obd_export
*exp
= ecl
->ecl_exp
;
2110 struct ldlm_lock
*lock
= cfs_hash_object(hs
, hnode
);
2111 struct ldlm_resource
*res
;
2113 res
= ldlm_resource_getref(lock
->l_resource
);
2114 LDLM_LOCK_GET(lock
);
2116 LDLM_DEBUG(lock
, "export %p", exp
);
2117 ldlm_res_lvbo_update(res
, NULL
, 1);
2118 ldlm_lock_cancel(lock
);
2119 ldlm_reprocess_all(res
);
2120 ldlm_resource_putref(res
);
2121 LDLM_LOCK_RELEASE(lock
);
2124 if ((ecl
->ecl_loop
& -ecl
->ecl_loop
) == ecl
->ecl_loop
) {
2126 "Cancel lock %p for export %p (loop %d), still have "
2127 "%d locks left on hash table.\n",
2128 lock
, exp
, ecl
->ecl_loop
,
2129 atomic_read(&hs
->hs_count
));
2136 * Cancel all locks for given export.
2138 * Typically called on client disconnection/eviction
2140 void ldlm_cancel_locks_for_export(struct obd_export
*exp
)
2142 struct export_cl_data ecl
= {
2147 cfs_hash_for_each_empty(exp
->exp_lock_hash
,
2148 ldlm_cancel_locks_for_export_cb
, &ecl
);
2152 * Downgrade an exclusive lock.
2154 * A fast variant of ldlm_lock_convert for convertion of exclusive
2155 * locks. The convertion is always successful.
2156 * Used by Commit on Sharing (COS) code.
2158 * \param lock A lock to convert
2159 * \param new_mode new lock mode
2161 void ldlm_lock_downgrade(struct ldlm_lock
*lock
, int new_mode
)
2165 LASSERT(lock
->l_granted_mode
& (LCK_PW
| LCK_EX
));
2166 LASSERT(new_mode
== LCK_COS
);
2168 lock_res_and_lock(lock
);
2169 ldlm_resource_unlink_lock(lock
);
2171 * Remove the lock from pool as it will be added again in
2172 * ldlm_grant_lock() called below.
2174 ldlm_pool_del(&ldlm_lock_to_ns(lock
)->ns_pool
, lock
);
2176 lock
->l_req_mode
= new_mode
;
2177 ldlm_grant_lock(lock
, NULL
);
2178 unlock_res_and_lock(lock
);
2179 ldlm_reprocess_all(lock
->l_resource
);
2183 EXPORT_SYMBOL(ldlm_lock_downgrade
);
2186 * Attempt to convert already granted lock to a different mode.
2188 * While lock conversion is not currently used, future client-side
2189 * optimizations could take advantage of it to avoid discarding cached
2192 struct ldlm_resource
*ldlm_lock_convert(struct ldlm_lock
*lock
, int new_mode
,
2195 LIST_HEAD(rpc_list
);
2196 struct ldlm_resource
*res
;
2197 struct ldlm_namespace
*ns
;
2199 struct ldlm_interval
*node
;
2202 /* Just return if mode is unchanged. */
2203 if (new_mode
== lock
->l_granted_mode
) {
2204 *flags
|= LDLM_FL_BLOCK_GRANTED
;
2205 RETURN(lock
->l_resource
);
2208 /* I can't check the type of lock here because the bitlock of lock
2209 * is not held here, so do the allocation blindly. -jay */
2210 OBD_SLAB_ALLOC_PTR_GFP(node
, ldlm_interval_slab
, __GFP_IO
);
2211 if (node
== NULL
) /* Actually, this causes EDEADLOCK to be returned */
2214 LASSERTF((new_mode
== LCK_PW
&& lock
->l_granted_mode
== LCK_PR
),
2215 "new_mode %u, granted %u\n", new_mode
, lock
->l_granted_mode
);
2217 lock_res_and_lock(lock
);
2219 res
= lock
->l_resource
;
2220 ns
= ldlm_res_to_ns(res
);
2222 lock
->l_req_mode
= new_mode
;
2223 if (res
->lr_type
== LDLM_PLAIN
|| res
->lr_type
== LDLM_IBITS
) {
2224 ldlm_resource_unlink_lock(lock
);
2226 ldlm_resource_unlink_lock(lock
);
2227 if (res
->lr_type
== LDLM_EXTENT
) {
2228 /* FIXME: ugly code, I have to attach the lock to a
2229 * interval node again since perhaps it will be granted
2231 INIT_LIST_HEAD(&node
->li_group
);
2232 ldlm_interval_attach(node
, lock
);
2238 * Remove old lock from the pool before adding the lock with new
2239 * mode below in ->policy()
2241 ldlm_pool_del(&ns
->ns_pool
, lock
);
2243 /* If this is a local resource, put it on the appropriate list. */
2244 if (ns_is_client(ldlm_res_to_ns(res
))) {
2245 if (*flags
& (LDLM_FL_BLOCK_CONV
| LDLM_FL_BLOCK_GRANTED
)) {
2246 ldlm_resource_add_lock(res
, &res
->lr_converting
, lock
);
2248 /* This should never happen, because of the way the
2249 * server handles conversions. */
2250 LDLM_ERROR(lock
, "Erroneous flags %x on local lock\n",
2254 ldlm_grant_lock(lock
, &rpc_list
);
2256 /* FIXME: completion handling not with lr_lock held ! */
2257 if (lock
->l_completion_ast
)
2258 lock
->l_completion_ast(lock
, 0, NULL
);
2261 CERROR("This is client-side-only module, cannot handle "
2262 "LDLM_NAMESPACE_SERVER resource type lock.\n");
2265 unlock_res_and_lock(lock
);
2268 ldlm_run_ast_work(ns
, &rpc_list
, LDLM_WORK_CP_AST
);
2270 OBD_SLAB_FREE(node
, ldlm_interval_slab
, sizeof(*node
));
2273 EXPORT_SYMBOL(ldlm_lock_convert
);
2276 * Print lock with lock handle \a lockh description into debug log.
2278 * Used when printing all locks on a resource for debug purposes.
2280 void ldlm_lock_dump_handle(int level
, struct lustre_handle
*lockh
)
2282 struct ldlm_lock
*lock
;
2284 if (!((libcfs_debug
| D_ERROR
) & level
))
2287 lock
= ldlm_handle2lock(lockh
);
2291 LDLM_DEBUG_LIMIT(level
, lock
, "###");
2293 LDLM_LOCK_PUT(lock
);
2295 EXPORT_SYMBOL(ldlm_lock_dump_handle
);
2298 * Print lock information with custom message into debug log.
2301 void _ldlm_lock_debug(struct ldlm_lock
*lock
,
2302 struct libcfs_debug_msg_data
*msgdata
,
2303 const char *fmt
, ...)
2306 struct obd_export
*exp
= lock
->l_export
;
2307 struct ldlm_resource
*resource
= lock
->l_resource
;
2308 char *nid
= "local";
2310 va_start(args
, fmt
);
2312 if (exp
&& exp
->exp_connection
) {
2313 nid
= libcfs_nid2str(exp
->exp_connection
->c_peer
.nid
);
2314 } else if (exp
&& exp
->exp_obd
!= NULL
) {
2315 struct obd_import
*imp
= exp
->exp_obd
->u
.cli
.cl_import
;
2316 nid
= libcfs_nid2str(imp
->imp_connection
->c_peer
.nid
);
2319 if (resource
== NULL
) {
2320 libcfs_debug_vmsg2(msgdata
, fmt
, args
,
2321 " ns: \?\? lock: %p/"LPX64
" lrc: %d/%d,%d mode: %s/%s "
2322 "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64
" nid: %s "
2323 "remote: "LPX64
" expref: %d pid: %u timeout: %lu "
2326 lock
->l_handle
.h_cookie
, atomic_read(&lock
->l_refc
),
2327 lock
->l_readers
, lock
->l_writers
,
2328 ldlm_lockname
[lock
->l_granted_mode
],
2329 ldlm_lockname
[lock
->l_req_mode
],
2330 lock
->l_flags
, nid
, lock
->l_remote_handle
.cookie
,
2331 exp
? atomic_read(&exp
->exp_refcount
) : -99,
2332 lock
->l_pid
, lock
->l_callback_timeout
, lock
->l_lvb_type
);
2337 switch (resource
->lr_type
) {
2339 libcfs_debug_vmsg2(msgdata
, fmt
, args
,
2340 " ns: %s lock: %p/"LPX64
" lrc: %d/%d,%d mode: %s/%s "
2341 "res: "LPU64
"/"LPU64
" rrc: %d type: %s ["LPU64
"->"LPU64
2342 "] (req "LPU64
"->"LPU64
") flags: "LPX64
" nid: %s remote:"
2343 " "LPX64
" expref: %d pid: %u timeout: %lu lvb_type: %d\n",
2344 ldlm_lock_to_ns_name(lock
), lock
,
2345 lock
->l_handle
.h_cookie
, atomic_read(&lock
->l_refc
),
2346 lock
->l_readers
, lock
->l_writers
,
2347 ldlm_lockname
[lock
->l_granted_mode
],
2348 ldlm_lockname
[lock
->l_req_mode
],
2349 resource
->lr_name
.name
[0],
2350 resource
->lr_name
.name
[1],
2351 atomic_read(&resource
->lr_refcount
),
2352 ldlm_typename
[resource
->lr_type
],
2353 lock
->l_policy_data
.l_extent
.start
,
2354 lock
->l_policy_data
.l_extent
.end
,
2355 lock
->l_req_extent
.start
, lock
->l_req_extent
.end
,
2356 lock
->l_flags
, nid
, lock
->l_remote_handle
.cookie
,
2357 exp
? atomic_read(&exp
->exp_refcount
) : -99,
2358 lock
->l_pid
, lock
->l_callback_timeout
, lock
->l_lvb_type
);
2362 libcfs_debug_vmsg2(msgdata
, fmt
, args
,
2363 " ns: %s lock: %p/"LPX64
" lrc: %d/%d,%d mode: %s/%s "
2364 "res: "LPU64
"/"LPU64
" rrc: %d type: %s pid: %d "
2365 "["LPU64
"->"LPU64
"] flags: "LPX64
" nid: %s remote: "LPX64
2366 " expref: %d pid: %u timeout: %lu\n",
2367 ldlm_lock_to_ns_name(lock
), lock
,
2368 lock
->l_handle
.h_cookie
, atomic_read(&lock
->l_refc
),
2369 lock
->l_readers
, lock
->l_writers
,
2370 ldlm_lockname
[lock
->l_granted_mode
],
2371 ldlm_lockname
[lock
->l_req_mode
],
2372 resource
->lr_name
.name
[0],
2373 resource
->lr_name
.name
[1],
2374 atomic_read(&resource
->lr_refcount
),
2375 ldlm_typename
[resource
->lr_type
],
2376 lock
->l_policy_data
.l_flock
.pid
,
2377 lock
->l_policy_data
.l_flock
.start
,
2378 lock
->l_policy_data
.l_flock
.end
,
2379 lock
->l_flags
, nid
, lock
->l_remote_handle
.cookie
,
2380 exp
? atomic_read(&exp
->exp_refcount
) : -99,
2381 lock
->l_pid
, lock
->l_callback_timeout
);
2385 libcfs_debug_vmsg2(msgdata
, fmt
, args
,
2386 " ns: %s lock: %p/"LPX64
" lrc: %d/%d,%d mode: %s/%s "
2387 "res: "LPU64
"/"LPU64
" bits "LPX64
" rrc: %d type: %s "
2388 "flags: "LPX64
" nid: %s remote: "LPX64
" expref: %d "
2389 "pid: %u timeout: %lu lvb_type: %d\n",
2390 ldlm_lock_to_ns_name(lock
),
2391 lock
, lock
->l_handle
.h_cookie
,
2392 atomic_read (&lock
->l_refc
),
2393 lock
->l_readers
, lock
->l_writers
,
2394 ldlm_lockname
[lock
->l_granted_mode
],
2395 ldlm_lockname
[lock
->l_req_mode
],
2396 resource
->lr_name
.name
[0],
2397 resource
->lr_name
.name
[1],
2398 lock
->l_policy_data
.l_inodebits
.bits
,
2399 atomic_read(&resource
->lr_refcount
),
2400 ldlm_typename
[resource
->lr_type
],
2401 lock
->l_flags
, nid
, lock
->l_remote_handle
.cookie
,
2402 exp
? atomic_read(&exp
->exp_refcount
) : -99,
2403 lock
->l_pid
, lock
->l_callback_timeout
, lock
->l_lvb_type
);
2407 libcfs_debug_vmsg2(msgdata
, fmt
, args
,
2408 " ns: %s lock: %p/"LPX64
" lrc: %d/%d,%d mode: %s/%s "
2409 "res: "LPU64
"/"LPU64
" rrc: %d type: %s flags: "LPX64
" "
2410 "nid: %s remote: "LPX64
" expref: %d pid: %u timeout: %lu"
2412 ldlm_lock_to_ns_name(lock
),
2413 lock
, lock
->l_handle
.h_cookie
,
2414 atomic_read (&lock
->l_refc
),
2415 lock
->l_readers
, lock
->l_writers
,
2416 ldlm_lockname
[lock
->l_granted_mode
],
2417 ldlm_lockname
[lock
->l_req_mode
],
2418 resource
->lr_name
.name
[0],
2419 resource
->lr_name
.name
[1],
2420 atomic_read(&resource
->lr_refcount
),
2421 ldlm_typename
[resource
->lr_type
],
2422 lock
->l_flags
, nid
, lock
->l_remote_handle
.cookie
,
2423 exp
? atomic_read(&exp
->exp_refcount
) : -99,
2424 lock
->l_pid
, lock
->l_callback_timeout
, lock
->l_lvb_type
);
2429 EXPORT_SYMBOL(_ldlm_lock_debug
);