1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
5 * Author: Peter Braam <braam@clusterfs.com>
6 * Author: Phil Schwan <phil@clusterfs.com>
8 * This file is part of the Lustre file system, http://www.lustre.org
9 * Lustre is a trademark of Cluster File Systems, Inc.
11 * You may have signed or agreed to another license before downloading
12 * this software. If so, you are bound by the terms and conditions
13 * of that agreement, and the following does not apply to you. See the
14 * LICENSE file included with this distribution for more information.
16 * If you did not agree to a different license, then this copy of Lustre
17 * is open source software; you can redistribute it and/or modify it
18 * under the terms of version 2 of the GNU General Public License as
19 * published by the Free Software Foundation.
21 * In either case, Lustre is distributed in the hope that it will be
22 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
23 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
24 * license text for more details.
27 #define DEBUG_SUBSYSTEM S_LDLM
30 # include <libcfs/libcfs.h>
31 # include <linux/lustre_intent.h>
33 # include <liblustre.h>
34 # include <libcfs/kp30.h>
37 #include <obd_class.h>
38 #include "ldlm_internal.h"
40 /* lock's skip list pointers fix mode */
41 #define LDLM_JOIN_NONE 0
42 #define LDLM_MODE_JOIN_RIGHT 1
43 #define LDLM_MODE_JOIN_LEFT (1 << 1)
44 #define LDLM_POLICY_JOIN_RIGHT (1 << 2)
45 #define LDLM_POLICY_JOIN_LEFT (1 << 3)
48 char *ldlm_lockname[] = {
59 char *ldlm_typename[] = {
66 char *ldlm_it2str(int it)
73 case (IT_OPEN | IT_CREAT):
86 CERROR("Unknown intent %d\n", it);
91 extern cfs_mem_cache_t *ldlm_lock_slab;
93 static ldlm_processing_policy ldlm_processing_policy_table[] = {
94 [LDLM_PLAIN] ldlm_process_plain_lock,
95 [LDLM_EXTENT] ldlm_process_extent_lock,
97 [LDLM_FLOCK] ldlm_process_flock_lock,
99 [LDLM_IBITS] ldlm_process_inodebits_lock,
102 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res)
104 return ldlm_processing_policy_table[res->lr_type];
107 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg)
113 * REFCOUNTED LOCK OBJECTS
118 * Lock refcounts, during creation:
119 * - one special one for allocation, dec'd only once in destroy
120 * - one for being a lock that's in-use
121 * - one for the addref associated with a new lock
123 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock)
125 atomic_inc(&lock->l_refc);
129 static void ldlm_lock_free(struct ldlm_lock *lock, size_t size)
131 LASSERT(size == sizeof(*lock));
132 OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
135 void ldlm_lock_put(struct ldlm_lock *lock)
139 LASSERT(lock->l_resource != LP_POISON);
140 LASSERT(atomic_read(&lock->l_refc) > 0);
141 if (atomic_dec_and_test(&lock->l_refc)) {
142 struct ldlm_resource *res;
145 "final lock_put on destroyed lock, freeing it.");
147 res = lock->l_resource;
148 LASSERT(lock->l_destroyed);
149 LASSERT(list_empty(&lock->l_res_link));
150 LASSERT(list_empty(&lock->l_pending_chain));
152 atomic_dec(&res->lr_namespace->ns_locks);
153 ldlm_resource_putref(res);
154 lock->l_resource = NULL;
155 if (lock->l_export) {
156 class_export_put(lock->l_export);
157 lock->l_export = NULL;
160 if (lock->l_lvb_data != NULL)
161 OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
163 OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle,
170 int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
173 if (!list_empty(&lock->l_lru)) {
174 LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
175 list_del_init(&lock->l_lru);
176 lock->l_resource->lr_namespace->ns_nr_unused--;
177 LASSERT(lock->l_resource->lr_namespace->ns_nr_unused >= 0);
183 int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
187 spin_lock(&lock->l_resource->lr_namespace->ns_unused_lock);
188 rc = ldlm_lock_remove_from_lru_nolock(lock);
189 spin_unlock(&lock->l_resource->lr_namespace->ns_unused_lock);
194 /* This used to have a 'strict' flag, which recovery would use to mark an
195 * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
196 * shall explain why it's gone: with the new hash table scheme, once you call
197 * ldlm_lock_destroy, you can never drop your final references on this lock.
198 * Because it's not in the hash table anymore. -phil */
199 int ldlm_lock_destroy_internal(struct ldlm_lock *lock)
203 if (lock->l_readers || lock->l_writers) {
204 LDLM_ERROR(lock, "lock still has references");
205 ldlm_lock_dump(D_ERROR, lock, 0);
209 if (!list_empty(&lock->l_res_link)) {
210 LDLM_ERROR(lock, "lock still on resource");
211 ldlm_lock_dump(D_ERROR, lock, 0);
215 if (lock->l_destroyed) {
216 LASSERT(list_empty(&lock->l_lru));
220 lock->l_destroyed = 1;
223 spin_lock(&lock->l_export->exp_ldlm_data.led_lock);
224 list_del_init(&lock->l_export_chain);
226 spin_unlock(&lock->l_export->exp_ldlm_data.led_lock);
228 ldlm_lock_remove_from_lru(lock);
229 class_handle_unhash(&lock->l_handle);
232 /* Wake anyone waiting for this lock */
233 /* FIXME: I should probably add yet another flag, instead of using
234 * l_export to only call this on clients */
236 class_export_put(lock->l_export);
237 lock->l_export = NULL;
238 if (lock->l_export && lock->l_completion_ast)
239 lock->l_completion_ast(lock, 0);
245 void ldlm_lock_destroy(struct ldlm_lock *lock)
249 lock_res_and_lock(lock);
250 first = ldlm_lock_destroy_internal(lock);
251 unlock_res_and_lock(lock);
253 /* drop reference from hashtable only for first destroy */
259 void ldlm_lock_destroy_nolock(struct ldlm_lock *lock)
263 first = ldlm_lock_destroy_internal(lock);
264 /* drop reference from hashtable only for first destroy */
270 /* this is called by portals_handle2object with the handle lock taken */
271 static void lock_handle_addref(void *lock)
273 LDLM_LOCK_GET((struct ldlm_lock *)lock);
277 * usage: pass in a resource on which you have done ldlm_resource_get
278 * pass in a parent lock on which you have done a ldlm_lock_get
279 * after return, ldlm_*_put the resource and parent
280 * returns: lock with refcount 2 - one for current caller and one for remote
282 static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource)
284 struct ldlm_lock *lock;
287 if (resource == NULL)
290 OBD_SLAB_ALLOC(lock, ldlm_lock_slab, CFS_ALLOC_IO, sizeof(*lock));
294 lock->l_resource = ldlm_resource_getref(resource);
296 atomic_set(&lock->l_refc, 2);
297 CFS_INIT_LIST_HEAD(&lock->l_res_link);
298 CFS_INIT_LIST_HEAD(&lock->l_lru);
299 CFS_INIT_LIST_HEAD(&lock->l_export_chain);
300 CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
301 CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
302 CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
303 cfs_waitq_init(&lock->l_waitq);
304 lock->l_blocking_lock = NULL;
306 lock->l_sl_mode.prev = NULL;
307 lock->l_sl_mode.next = NULL;
308 lock->l_sl_policy.prev = NULL;
309 lock->l_sl_policy.next = NULL;
311 atomic_inc(&resource->lr_namespace->ns_locks);
312 CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
313 class_handle_hash(&lock->l_handle, lock_handle_addref);
318 int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock,
319 const struct ldlm_res_id *new_resid)
321 struct ldlm_resource *oldres = lock->l_resource;
322 struct ldlm_resource *newres;
326 LASSERT(ns->ns_client != 0);
328 lock_res_and_lock(lock);
329 if (memcmp(new_resid, &lock->l_resource->lr_name,
330 sizeof(lock->l_resource->lr_name)) == 0) {
332 unlock_res_and_lock(lock);
336 LASSERT(new_resid->name[0] != 0);
338 /* This function assumes that the lock isn't on any lists */
339 LASSERT(list_empty(&lock->l_res_link));
341 type = oldres->lr_type;
342 unlock_res_and_lock(lock);
344 newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
345 if (newres == NULL) {
350 lock_res_and_lock(lock);
351 LASSERT(memcmp(new_resid, &lock->l_resource->lr_name,
352 sizeof(lock->l_resource->lr_name)) != 0);
354 lock->l_resource = newres;
357 unlock_bitlock(lock);
359 /* ...and the flowers are still standing! */
360 ldlm_resource_putref(oldres);
369 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
371 lockh->cookie = lock->l_handle.h_cookie;
374 /* if flags: atomically get the lock and set the flags.
375 * Return NULL if flag already set
378 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
381 struct ldlm_namespace *ns;
382 struct ldlm_lock *lock, *retval = NULL;
387 lock = class_handle2object(handle->cookie);
391 LASSERT(lock->l_resource != NULL);
392 ns = lock->l_resource->lr_namespace;
395 lock_res_and_lock(lock);
397 /* It's unlikely but possible that someone marked the lock as
398 * destroyed after we did handle2object on it */
399 if (lock->l_destroyed) {
400 unlock_res_and_lock(lock);
401 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
406 if (flags && (lock->l_flags & flags)) {
407 unlock_res_and_lock(lock);
413 lock->l_flags |= flags;
415 unlock_res_and_lock(lock);
422 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *ns,
423 const struct lustre_handle *handle)
425 struct ldlm_lock *retval = NULL;
426 retval = __ldlm_handle2lock(handle, 0);
430 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
432 struct obd_export *exp = lock->l_export?:lock->l_conn_export;
433 /* INODEBITS_INTEROP: If the other side does not support
434 * inodebits, reply with a plain lock descriptor.
436 if ((lock->l_resource->lr_type == LDLM_IBITS) &&
437 (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) {
438 struct ldlm_resource res = *lock->l_resource;
440 /* Make sure all the right bits are set in this lock we
441 are going to pass to client */
442 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
443 (MDS_INODELOCK_LOOKUP|MDS_INODELOCK_UPDATE),
444 "Inappropriate inode lock bits during "
445 "conversion " LPU64 "\n",
446 lock->l_policy_data.l_inodebits.bits);
447 res.lr_type = LDLM_PLAIN;
448 ldlm_res2desc(&res, &desc->l_resource);
449 /* Convert "new" lock mode to something old client can
451 if ((lock->l_req_mode == LCK_CR) ||
452 (lock->l_req_mode == LCK_CW))
453 desc->l_req_mode = LCK_PR;
455 desc->l_req_mode = lock->l_req_mode;
456 if ((lock->l_granted_mode == LCK_CR) ||
457 (lock->l_granted_mode == LCK_CW)) {
458 desc->l_granted_mode = LCK_PR;
460 /* We never grant PW/EX locks to clients */
461 LASSERT((lock->l_granted_mode != LCK_PW) &&
462 (lock->l_granted_mode != LCK_EX));
463 desc->l_granted_mode = lock->l_granted_mode;
466 /* We do not copy policy here, because there is no
467 policy for plain locks */
469 ldlm_res2desc(lock->l_resource, &desc->l_resource);
470 desc->l_req_mode = lock->l_req_mode;
471 desc->l_granted_mode = lock->l_granted_mode;
472 desc->l_policy_data = lock->l_policy_data;
476 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
477 struct list_head *work_list)
479 if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
480 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
481 lock->l_flags |= LDLM_FL_AST_SENT;
482 /* If the enqueuing client said so, tell the AST recipient to
483 * discard dirty data, rather than writing back. */
484 if (new->l_flags & LDLM_AST_DISCARD_DATA)
485 lock->l_flags |= LDLM_FL_DISCARD_DATA;
486 LASSERT(list_empty(&lock->l_bl_ast));
487 list_add(&lock->l_bl_ast, work_list);
489 LASSERT(lock->l_blocking_lock == NULL);
490 lock->l_blocking_lock = LDLM_LOCK_GET(new);
494 void ldlm_add_cp_work_item(struct ldlm_lock *lock, struct list_head *work_list)
496 if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
497 lock->l_flags |= LDLM_FL_CP_REQD;
498 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
499 LASSERT(list_empty(&lock->l_cp_ast));
500 list_add(&lock->l_cp_ast, work_list);
505 /* must be called with lr_lock held */
506 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
507 struct list_head *work_list)
510 check_res_locked(lock->l_resource);
512 ldlm_add_bl_work_item(lock, new, work_list);
514 ldlm_add_cp_work_item(lock, work_list);
518 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
520 struct ldlm_lock *lock;
522 lock = ldlm_handle2lock(lockh);
523 LASSERT(lock != NULL);
524 ldlm_lock_addref_internal(lock, mode);
528 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
530 ldlm_lock_remove_from_lru(lock);
531 if (mode & (LCK_NL | LCK_CR | LCK_PR))
533 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP))
535 lock->l_last_used = cfs_time_current();
537 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
540 /* only called for local locks */
541 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
543 lock_res_and_lock(lock);
544 ldlm_lock_addref_internal_nolock(lock, mode);
545 unlock_res_and_lock(lock);
548 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
550 struct ldlm_namespace *ns;
553 lock_res_and_lock(lock);
555 ns = lock->l_resource->lr_namespace;
557 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
558 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
559 LASSERT(lock->l_readers > 0);
562 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP)) {
563 LASSERT(lock->l_writers > 0);
567 if (lock->l_flags & LDLM_FL_LOCAL &&
568 !lock->l_readers && !lock->l_writers) {
569 /* If this is a local lock on a server namespace and this was
570 * the last reference, cancel the lock. */
571 CDEBUG(D_INFO, "forcing cancel of local lock\n");
572 lock->l_flags |= LDLM_FL_CBPENDING;
575 if (!lock->l_readers && !lock->l_writers &&
576 (lock->l_flags & LDLM_FL_CBPENDING)) {
577 /* If we received a blocked AST and this was the last reference,
578 * run the callback. */
579 if (ns->ns_client == LDLM_NAMESPACE_SERVER && lock->l_export)
580 CERROR("FL_CBPENDING set on non-local lock--just a "
583 LDLM_DEBUG(lock, "final decref done on cbpending lock");
585 LDLM_LOCK_GET(lock); /* dropped by bl thread */
586 ldlm_lock_remove_from_lru(lock);
587 unlock_res_and_lock(lock);
588 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
589 ldlm_bl_to_thread(ns, NULL, lock, 0) != 0)
590 ldlm_handle_bl_callback(ns, NULL, lock);
591 } else if (ns->ns_client == LDLM_NAMESPACE_CLIENT &&
592 !lock->l_readers && !lock->l_writers &&
593 !(lock->l_flags & LDLM_FL_NO_LRU)) {
594 /* If this is a client-side namespace and this was the last
595 * reference, put it on the LRU. */
596 LASSERT(list_empty(&lock->l_lru));
597 LASSERT(ns->ns_nr_unused >= 0);
598 lock->l_last_used = cfs_time_current();
599 spin_lock(&ns->ns_unused_lock);
600 list_add_tail(&lock->l_lru, &ns->ns_unused_list);
602 spin_unlock(&ns->ns_unused_lock);
603 unlock_res_and_lock(lock);
604 /* Call ldlm_cancel_lru() only if EARLY_CANCEL is not supported
605 * by the server, otherwise, it is done on enqueue. */
606 if (!exp_connect_cancelset(lock->l_conn_export))
607 ldlm_cancel_lru(ns, LDLM_ASYNC);
609 unlock_res_and_lock(lock);
612 LDLM_LOCK_PUT(lock); /* matches the ldlm_lock_get in addref */
617 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
619 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
620 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
621 ldlm_lock_decref_internal(lock, mode);
625 /* This will drop a lock reference and mark it for destruction, but will not
626 * necessarily cancel the lock before returning. */
627 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
629 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
632 LASSERT(lock != NULL);
634 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
635 lock_res_and_lock(lock);
636 lock->l_flags |= LDLM_FL_CBPENDING;
637 unlock_res_and_lock(lock);
638 ldlm_lock_decref_internal(lock, mode);
643 * search_granted_lock
646 * Finds a position to insert the new lock.
648 * queue [input]: the granted list where search acts on;
649 * req [input]: the lock whose position to be located;
650 * lockp [output]: the position where the lock should be inserted before, or
651 * NULL indicating @req should be appended to @queue.
653 * Bit-masks combination of following values indicating in which way the
654 * lock need to be inserted.
655 * - LDLM_JOIN_NONE: noting about skip list needs to be fixed;
656 * - LDLM_MODE_JOIN_RIGHT: @req needs join right becoming the head of a
658 * - LDLM_POLICY_JOIN_RIGHT: @req needs join right becoming the head of
661 * - ldlm_grant_lock_with_skiplist
663 static int search_granted_lock(struct list_head *queue,
664 struct ldlm_lock *req,
665 struct ldlm_lock **lockp)
667 struct list_head *tmp, *tmp_tail;
668 struct ldlm_lock *lock, *mode_head_lock;
669 int rc = LDLM_JOIN_NONE;
672 list_for_each(tmp, queue) {
673 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
675 if (lock->l_req_mode != req->l_req_mode) {
676 if (LDLM_SL_HEAD(&lock->l_sl_mode))
677 tmp = &list_entry(lock->l_sl_mode.next,
679 l_sl_mode)->l_res_link;
683 /* found the same mode group */
684 if (lock->l_resource->lr_type == LDLM_PLAIN) {
686 rc = LDLM_MODE_JOIN_RIGHT;
688 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
690 if (LDLM_SL_HEAD(&lock->l_sl_mode))
691 tmp_tail = &list_entry(lock->l_sl_mode.next,
693 l_sl_mode)->l_res_link;
694 mode_head_lock = lock;
696 if (lock->l_policy_data.l_inodebits.bits ==
697 req->l_policy_data.l_inodebits.bits) {
698 /* matched policy lock is found */
700 rc |= LDLM_POLICY_JOIN_RIGHT;
702 /* if the policy group head is also a
703 * mode group head or a single mode
705 if (LDLM_SL_HEAD(&lock->l_sl_mode) ||
707 LDLM_SL_EMPTY(&lock->l_sl_mode)))
708 rc |= LDLM_MODE_JOIN_RIGHT;
712 if (LDLM_SL_HEAD(&lock->l_sl_policy))
713 tmp = &list_entry(lock->l_sl_policy.next,
715 l_sl_policy)->l_res_link;
721 lock = list_entry(tmp, struct ldlm_lock,
723 } /* for all locks in the matched mode group */
725 /* no matched policy group is found, insert before
726 * the mode group head lock */
727 *lockp = mode_head_lock;
728 rc = LDLM_MODE_JOIN_RIGHT;
731 LDLM_ERROR(lock, "is not LDLM_PLAIN or LDLM_IBITS lock");
736 /* no matched mode group is found, append to the end */
744 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
745 struct ldlm_lock *lockp,
748 struct ldlm_resource *res = lock->l_resource;
751 LASSERT(lockp || join == LDLM_JOIN_NONE);
753 check_res_locked(res);
755 ldlm_resource_dump(D_OTHER, res);
756 CDEBUG(D_OTHER, "About to add this lock:\n");
757 ldlm_lock_dump(D_OTHER, lock, 0);
759 if (lock->l_destroyed) {
760 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
764 LASSERT(list_empty(&lock->l_res_link));
767 list_add_tail(&lock->l_res_link, &lock->l_resource->lr_granted);
768 else if ((join & LDLM_MODE_JOIN_LEFT) || (join & LDLM_POLICY_JOIN_LEFT))
769 list_add(&lock->l_res_link, &lockp->l_res_link);
771 list_add_tail(&lock->l_res_link, &lockp->l_res_link);
774 if (join & LDLM_MODE_JOIN_RIGHT) {
775 LASSERT(! LDLM_SL_TAIL(&lockp->l_sl_mode));
776 if (LDLM_SL_EMPTY(&lockp->l_sl_mode)) {
777 lock->l_sl_mode.next = &lockp->l_sl_mode;
778 lockp->l_sl_mode.prev = &lock->l_sl_mode;
779 } else if (LDLM_SL_HEAD(&lockp->l_sl_mode)) {
780 lock->l_sl_mode.next = lockp->l_sl_mode.next;
781 lockp->l_sl_mode.next = NULL;
782 lock->l_sl_mode.next->prev = &lock->l_sl_mode;
784 } else if (join & LDLM_MODE_JOIN_LEFT) {
785 LASSERT(! LDLM_SL_HEAD(&lockp->l_sl_mode));
786 if (LDLM_SL_EMPTY(&lockp->l_sl_mode)) {
787 lock->l_sl_mode.prev = &lockp->l_sl_mode;
788 lockp->l_sl_mode.next = &lock->l_sl_mode;
789 } else if (LDLM_SL_TAIL(&lockp->l_sl_mode)) {
790 lock->l_sl_mode.prev = lockp->l_sl_mode.prev;
791 lockp->l_sl_mode.prev = NULL;
792 lock->l_sl_mode.prev->next = &lock->l_sl_mode;
796 if (join & LDLM_POLICY_JOIN_RIGHT) {
797 LASSERT(! LDLM_SL_TAIL(&lockp->l_sl_policy));
798 if (LDLM_SL_EMPTY(&lockp->l_sl_policy)) {
799 lock->l_sl_policy.next = &lockp->l_sl_policy;
800 lockp->l_sl_policy.prev = &lock->l_sl_policy;
801 } else if (LDLM_SL_HEAD(&lockp->l_sl_policy)) {
802 lock->l_sl_policy.next = lockp->l_sl_policy.next;
803 lockp->l_sl_policy.next = NULL;
804 lock->l_sl_policy.next->prev = &lock->l_sl_policy;
806 } else if (join & LDLM_POLICY_JOIN_LEFT) {
807 LASSERT(! LDLM_SL_HEAD(&lockp->l_sl_policy));
808 if (LDLM_SL_EMPTY(&lockp->l_sl_policy)) {
809 lock->l_sl_policy.prev = &lockp->l_sl_policy;
810 lockp->l_sl_policy.next = &lock->l_sl_policy;
811 } else if (LDLM_SL_TAIL(&lockp->l_sl_policy)) {
812 lock->l_sl_policy.prev = lockp->l_sl_policy.prev;
813 lockp->l_sl_policy.prev = NULL;
814 lock->l_sl_policy.prev->next = &lock->l_sl_policy;
821 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
823 int join = LDLM_JOIN_NONE;
824 struct ldlm_lock *lockp = NULL;
827 LASSERT(lock->l_req_mode == lock->l_granted_mode);
829 join = search_granted_lock(&lock->l_resource->lr_granted, lock, &lockp);
830 ldlm_granted_list_add_lock(lock, lockp, join);
835 * - ldlm_lock_enqueue
836 * - ldlm_reprocess_queue
837 * - ldlm_lock_convert
839 * must be called with lr_lock held
841 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
843 struct ldlm_resource *res = lock->l_resource;
846 check_res_locked(res);
848 lock->l_granted_mode = lock->l_req_mode;
849 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
850 ldlm_grant_lock_with_skiplist(lock);
852 ldlm_resource_add_lock(res, &res->lr_granted, lock);
854 if (lock->l_granted_mode < res->lr_most_restr)
855 res->lr_most_restr = lock->l_granted_mode;
857 if (work_list && lock->l_completion_ast != NULL)
858 ldlm_add_ast_work_item(lock, NULL, work_list);
863 /* returns a referenced lock or NULL. See the flag descriptions below, in the
864 * comment above ldlm_lock_match */
865 static struct ldlm_lock *search_queue(struct list_head *queue, ldlm_mode_t mode,
866 ldlm_policy_data_t *policy,
867 struct ldlm_lock *old_lock, int flags)
869 struct ldlm_lock *lock;
870 struct list_head *tmp;
872 list_for_each(tmp, queue) {
873 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
875 if (lock == old_lock)
878 /* llite sometimes wants to match locks that will be
879 * canceled when their users drop, but we allow it to match
880 * if it passes in CBPENDING and the lock still has users.
881 * this is generally only going to be used by children
882 * whose parents already hold a lock so forward progress
883 * can still happen. */
884 if (lock->l_flags & LDLM_FL_CBPENDING &&
885 !(flags & LDLM_FL_CBPENDING))
887 if (lock->l_flags & LDLM_FL_CBPENDING &&
888 lock->l_readers == 0 && lock->l_writers == 0)
891 if (!(lock->l_req_mode & mode))
894 if (lock->l_resource->lr_type == LDLM_EXTENT &&
895 (lock->l_policy_data.l_extent.start >
896 policy->l_extent.start ||
897 lock->l_policy_data.l_extent.end < policy->l_extent.end))
900 if (unlikely(mode == LCK_GROUP) &&
901 lock->l_resource->lr_type == LDLM_EXTENT &&
902 lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
905 /* We match if we have existing lock with same or wider set
907 if (lock->l_resource->lr_type == LDLM_IBITS &&
908 ((lock->l_policy_data.l_inodebits.bits &
909 policy->l_inodebits.bits) !=
910 policy->l_inodebits.bits))
913 if (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED))
916 if ((flags & LDLM_FL_LOCAL_ONLY) &&
917 !(lock->l_flags & LDLM_FL_LOCAL))
920 if (flags & LDLM_FL_TEST_LOCK)
923 ldlm_lock_addref_internal_nolock(lock, mode);
930 void ldlm_lock_allow_match(struct ldlm_lock *lock)
932 lock_res_and_lock(lock);
933 lock->l_flags |= LDLM_FL_LVB_READY;
934 cfs_waitq_signal(&lock->l_waitq);
935 unlock_res_and_lock(lock);
938 /* Can be called in two ways:
940 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
941 * for a duplicate of.
943 * Otherwise, all of the fields must be filled in, to match against.
945 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
946 * server (ie, connh is NULL)
947 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
948 * list will be considered
949 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
950 * to be canceled can still be matched as long as they still have reader
951 * or writer refernces
952 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
953 * just tell us if we would have matched.
955 * Returns 1 if it finds an already-existing lock that is compatible; in this
956 * case, lockh is filled in with a addref()ed lock
958 * we also check security context, if that failed we simply return 0 (to keep
959 * caller code unchanged), the context failure will be discovered by caller
962 int ldlm_lock_match(struct ldlm_namespace *ns, int flags,
963 const struct ldlm_res_id *res_id, ldlm_type_t type,
964 ldlm_policy_data_t *policy, ldlm_mode_t mode,
965 struct lustre_handle *lockh)
967 struct ldlm_resource *res;
968 struct ldlm_lock *lock, *old_lock = NULL;
973 old_lock = ldlm_handle2lock(lockh);
976 ns = old_lock->l_resource->lr_namespace;
977 res_id = &old_lock->l_resource->lr_name;
978 type = old_lock->l_resource->lr_type;
979 mode = old_lock->l_req_mode;
982 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
984 LASSERT(old_lock == NULL);
990 lock = search_queue(&res->lr_granted, mode, policy, old_lock, flags);
993 if (flags & LDLM_FL_BLOCK_GRANTED)
995 lock = search_queue(&res->lr_converting, mode, policy, old_lock, flags);
998 lock = search_queue(&res->lr_waiting, mode, policy, old_lock, flags);
1005 ldlm_resource_putref(res);
1008 ldlm_lock2handle(lock, lockh);
1009 if ((flags & LDLM_FL_LVB_READY) &&
1010 (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1011 struct l_wait_info lwi;
1012 if (lock->l_completion_ast) {
1013 int err = lock->l_completion_ast(lock,
1014 LDLM_FL_WAIT_NOREPROC,
1017 if (flags & LDLM_FL_TEST_LOCK)
1018 LDLM_LOCK_PUT(lock);
1020 ldlm_lock_decref_internal(lock, mode);
1026 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout), NULL,
1027 LWI_ON_SIGNAL_NOOP, NULL);
1029 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1030 l_wait_event(lock->l_waitq,
1031 (lock->l_flags & LDLM_FL_LVB_READY), &lwi);
1036 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1037 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1038 res_id->name[2] : policy->l_extent.start,
1039 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1040 res_id->name[3] : policy->l_extent.end);
1042 /* check user's security context */
1043 if (lock->l_conn_export &&
1044 sptlrpc_import_check_ctx(
1045 class_exp2cliimp(lock->l_conn_export))) {
1046 if (!(flags & LDLM_FL_TEST_LOCK))
1047 ldlm_lock_decref_internal(lock, mode);
1051 if (flags & LDLM_FL_TEST_LOCK)
1052 LDLM_LOCK_PUT(lock);
1054 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1055 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1056 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1057 type, mode, res_id->name[0], res_id->name[1],
1058 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1059 res_id->name[2] :policy->l_extent.start,
1060 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1061 res_id->name[3] : policy->l_extent.end);
1064 LDLM_LOCK_PUT(old_lock);
1069 /* Returns a referenced lock */
1070 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1071 const struct ldlm_res_id *res_id,
1074 ldlm_blocking_callback blocking,
1075 ldlm_completion_callback completion,
1076 ldlm_glimpse_callback glimpse,
1077 void *data, __u32 lvb_len)
1079 struct ldlm_lock *lock;
1080 struct ldlm_resource *res;
1083 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1087 lock = ldlm_lock_new(res);
1088 ldlm_resource_putref(res);
1093 lock->l_req_mode = mode;
1094 lock->l_ast_data = data;
1095 lock->l_blocking_ast = blocking;
1096 lock->l_completion_ast = completion;
1097 lock->l_glimpse_ast = glimpse;
1098 lock->l_pid = cfs_curproc_pid();
1101 lock->l_lvb_len = lvb_len;
1102 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1103 if (lock->l_lvb_data == NULL) {
1104 OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
1112 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1113 struct ldlm_lock **lockp,
1114 void *cookie, int *flags)
1116 struct ldlm_lock *lock = *lockp;
1117 struct ldlm_resource *res = lock->l_resource;
1118 int local = res->lr_namespace->ns_client;
1119 ldlm_processing_policy policy;
1120 ldlm_error_t rc = ELDLM_OK;
1123 do_gettimeofday(&lock->l_enqueued_time);
1124 /* policies are not executed on the client or during replay */
1125 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1126 && !local && ns->ns_policy) {
1127 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1129 if (rc == ELDLM_LOCK_REPLACED) {
1130 /* The lock that was returned has already been granted,
1131 * and placed into lockp. If it's not the same as the
1132 * one we passed in, then destroy the old one and our
1133 * work here is done. */
1134 if (lock != *lockp) {
1135 ldlm_lock_destroy(lock);
1136 LDLM_LOCK_PUT(lock);
1138 *flags |= LDLM_FL_LOCK_CHANGED;
1140 } else if (rc != ELDLM_OK ||
1141 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1142 ldlm_lock_destroy(lock);
1147 lock_res_and_lock(lock);
1148 if (local && lock->l_req_mode == lock->l_granted_mode) {
1149 /* The server returned a blocked lock, but it was granted before
1150 * we got a chance to actually enqueue it. We don't need to do
1152 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1153 LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1154 GOTO(out, ELDLM_OK);
1157 /* Some flags from the enqueue want to make it into the AST, via the
1158 * lock's l_flags. */
1159 lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1161 /* This distinction between local lock trees is very important; a client
1162 * namespace only has information about locks taken by that client, and
1163 * thus doesn't have enough information to decide for itself if it can
1164 * be granted (below). In this case, we do exactly what the server
1165 * tells us to do, as dictated by the 'flags'.
1167 * We do exactly the same thing during recovery, when the server is
1168 * more or less trusting the clients not to lie.
1170 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1171 * granted/converting queues. */
1172 ldlm_resource_unlink_lock(lock);
1174 if (*flags & LDLM_FL_BLOCK_CONV)
1175 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1176 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1177 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1179 ldlm_grant_lock(lock, NULL);
1180 GOTO(out, ELDLM_OK);
1181 } else if (*flags & LDLM_FL_REPLAY) {
1182 if (*flags & LDLM_FL_BLOCK_CONV) {
1183 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1184 GOTO(out, ELDLM_OK);
1185 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1186 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1187 GOTO(out, ELDLM_OK);
1188 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1189 ldlm_grant_lock(lock, NULL);
1190 GOTO(out, ELDLM_OK);
1192 /* If no flags, fall through to normal enqueue path. */
1195 policy = ldlm_processing_policy_table[res->lr_type];
1196 policy(lock, flags, 1, &rc, NULL);
1199 unlock_res_and_lock(lock);
1203 /* Must be called with namespace taken: queue is waiting or converting. */
1204 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1205 struct list_head *work_list)
1207 struct list_head *tmp, *pos;
1208 ldlm_processing_policy policy;
1210 int rc = LDLM_ITER_CONTINUE;
1214 check_res_locked(res);
1216 policy = ldlm_processing_policy_table[res->lr_type];
1219 list_for_each_safe(tmp, pos, queue) {
1220 struct ldlm_lock *pending;
1221 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1223 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1226 rc = policy(pending, &flags, 0, &err, work_list);
1227 if (rc != LDLM_ITER_CONTINUE)
1234 int ldlm_run_bl_ast_work(struct list_head *rpc_list)
1236 struct list_head *tmp, *pos;
1237 struct ldlm_lock_desc d;
1238 int rc = 0, retval = 0;
1241 list_for_each_safe(tmp, pos, rpc_list) {
1242 struct ldlm_lock *lock =
1243 list_entry(tmp, struct ldlm_lock, l_bl_ast);
1245 /* nobody should touch l_bl_ast */
1246 lock_res_and_lock(lock);
1247 list_del_init(&lock->l_bl_ast);
1249 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1250 LASSERT(lock->l_bl_ast_run == 0);
1251 LASSERT(lock->l_blocking_lock);
1252 lock->l_bl_ast_run++;
1253 unlock_res_and_lock(lock);
1255 ldlm_lock2desc(lock->l_blocking_lock, &d);
1257 LDLM_LOCK_PUT(lock->l_blocking_lock);
1258 lock->l_blocking_lock = NULL;
1259 rc = lock->l_blocking_ast(lock, &d, NULL, LDLM_CB_BLOCKING);
1261 if (rc == -ERESTART)
1264 CDEBUG(D_DLMTRACE, "Failed AST - should clean & "
1265 "disconnect client\n");
1266 LDLM_LOCK_PUT(lock);
1271 int ldlm_run_cp_ast_work(struct list_head *rpc_list)
1273 struct list_head *tmp, *pos;
1274 int rc = 0, retval = 0;
1277 /* It's possible to receive a completion AST before we've set
1278 * the l_completion_ast pointer: either because the AST arrived
1279 * before the reply, or simply because there's a small race
1280 * window between receiving the reply and finishing the local
1281 * enqueue. (bug 842)
1283 * This can't happen with the blocking_ast, however, because we
1284 * will never call the local blocking_ast until we drop our
1285 * reader/writer reference, which we won't do until we get the
1286 * reply and finish enqueueing. */
1288 list_for_each_safe(tmp, pos, rpc_list) {
1289 struct ldlm_lock *lock =
1290 list_entry(tmp, struct ldlm_lock, l_cp_ast);
1292 /* nobody should touch l_cp_ast */
1293 lock_res_and_lock(lock);
1294 list_del_init(&lock->l_cp_ast);
1295 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1296 lock->l_flags &= ~LDLM_FL_CP_REQD;
1297 unlock_res_and_lock(lock);
1299 if (lock->l_completion_ast != NULL)
1300 rc = lock->l_completion_ast(lock, 0, 0);
1301 if (rc == -ERESTART)
1304 CDEBUG(D_DLMTRACE, "Failed AST - should clean & "
1305 "disconnect client\n");
1306 LDLM_LOCK_PUT(lock);
1311 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
1313 ldlm_reprocess_all(res);
1314 return LDLM_ITER_CONTINUE;
1317 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
1319 struct list_head *tmp;
1326 spin_lock(&ns->ns_hash_lock);
1327 for (i = 0; i < RES_HASH_SIZE; i++) {
1328 tmp = ns->ns_hash[i].next;
1329 while (tmp != &(ns->ns_hash[i])) {
1330 struct ldlm_resource *res =
1331 list_entry(tmp, struct ldlm_resource, lr_hash);
1333 ldlm_resource_getref(res);
1334 spin_unlock(&ns->ns_hash_lock);
1336 rc = reprocess_one_queue(res, NULL);
1338 spin_lock(&ns->ns_hash_lock);
1340 ldlm_resource_putref_locked(res);
1342 if (rc == LDLM_ITER_STOP)
1347 spin_unlock(&ns->ns_hash_lock);
1351 void ldlm_reprocess_all(struct ldlm_resource *res)
1353 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1357 /* Local lock trees don't get reprocessed. */
1358 if (res->lr_namespace->ns_client) {
1365 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
1366 if (rc == LDLM_ITER_CONTINUE)
1367 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
1370 rc = ldlm_run_cp_ast_work(&rpc_list);
1371 if (rc == -ERESTART) {
1372 LASSERT(list_empty(&rpc_list));
1378 void ldlm_cancel_callback(struct ldlm_lock *lock)
1380 check_res_locked(lock->l_resource);
1381 if (!(lock->l_flags & LDLM_FL_CANCEL)) {
1382 lock->l_flags |= LDLM_FL_CANCEL;
1383 if (lock->l_blocking_ast) {
1384 // l_check_no_ns_lock(ns);
1385 unlock_res_and_lock(lock);
1386 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
1388 lock_res_and_lock(lock);
1390 LDLM_DEBUG(lock, "no blocking ast");
1395 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
1397 struct ldlm_lock *lock;
1399 if (req->l_resource->lr_type != LDLM_PLAIN &&
1400 req->l_resource->lr_type != LDLM_IBITS)
1403 if (LDLM_SL_HEAD(&req->l_sl_mode)) {
1404 lock = list_entry(req->l_res_link.next, struct ldlm_lock,
1406 if (req->l_sl_mode.next == &lock->l_sl_mode) {
1407 lock->l_sl_mode.prev = NULL;
1409 lock->l_sl_mode.next = req->l_sl_mode.next;
1410 lock->l_sl_mode.next->prev = &lock->l_sl_mode;
1412 req->l_sl_mode.next = NULL;
1413 } else if (LDLM_SL_TAIL(&req->l_sl_mode)) {
1414 lock = list_entry(req->l_res_link.prev, struct ldlm_lock,
1416 if (req->l_sl_mode.prev == &lock->l_sl_mode) {
1417 lock->l_sl_mode.next = NULL;
1419 lock->l_sl_mode.prev = req->l_sl_mode.prev;
1420 lock->l_sl_mode.prev->next = &lock->l_sl_mode;
1422 req->l_sl_mode.prev = NULL;
1425 if (LDLM_SL_HEAD(&req->l_sl_policy)) {
1426 lock = list_entry(req->l_res_link.next, struct ldlm_lock,
1428 if (req->l_sl_policy.next == &lock->l_sl_policy) {
1429 lock->l_sl_policy.prev = NULL;
1431 lock->l_sl_policy.next = req->l_sl_policy.next;
1432 lock->l_sl_policy.next->prev = &lock->l_sl_policy;
1434 req->l_sl_policy.next = NULL;
1435 } else if (LDLM_SL_TAIL(&req->l_sl_policy)) {
1436 lock = list_entry(req->l_res_link.prev, struct ldlm_lock,
1438 if (req->l_sl_policy.prev == &lock->l_sl_policy) {
1439 lock->l_sl_policy.next = NULL;
1441 lock->l_sl_policy.prev = req->l_sl_policy.prev;
1442 lock->l_sl_policy.prev->next = &lock->l_sl_policy;
1444 req->l_sl_policy.prev = NULL;
1448 void ldlm_lock_cancel(struct ldlm_lock *lock)
1450 struct ldlm_resource *res;
1451 struct ldlm_namespace *ns;
1454 lock_res_and_lock(lock);
1456 res = lock->l_resource;
1457 ns = res->lr_namespace;
1459 /* Please do not, no matter how tempting, remove this LBUG without
1460 * talking to me first. -phik */
1461 if (lock->l_readers || lock->l_writers) {
1462 LDLM_ERROR(lock, "lock still has references");
1466 ldlm_del_waiting_lock(lock);
1468 /* Releases cancel callback. */
1469 ldlm_cancel_callback(lock);
1471 /* Yes, second time, just in case it was added again while we were
1472 running with no res lock in ldlm_cancel_callback */
1473 ldlm_del_waiting_lock(lock);
1474 ldlm_resource_unlink_lock(lock);
1475 ldlm_lock_destroy_nolock(lock);
1476 unlock_res_and_lock(lock);
1481 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
1483 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1489 lock->l_ast_data = data;
1490 LDLM_LOCK_PUT(lock);
1494 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1496 struct ldlm_lock *lock;
1497 struct ldlm_resource *res;
1499 spin_lock(&exp->exp_ldlm_data.led_lock);
1500 while(!list_empty(&exp->exp_ldlm_data.led_held_locks)) {
1501 lock = list_entry(exp->exp_ldlm_data.led_held_locks.next,
1502 struct ldlm_lock, l_export_chain);
1503 res = ldlm_resource_getref(lock->l_resource);
1504 LDLM_LOCK_GET(lock);
1505 spin_unlock(&exp->exp_ldlm_data.led_lock);
1507 LDLM_DEBUG(lock, "export %p", exp);
1508 ldlm_lock_cancel(lock);
1509 ldlm_reprocess_all(res);
1511 ldlm_resource_putref(res);
1512 LDLM_LOCK_PUT(lock);
1513 spin_lock(&exp->exp_ldlm_data.led_lock);
1515 spin_unlock(&exp->exp_ldlm_data.led_lock);
1518 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1521 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1522 struct ldlm_resource *res;
1523 struct ldlm_namespace *ns;
1526 struct ldlm_lock *mark_lock = NULL;
1527 int join = LDLM_JOIN_NONE;
1531 if (new_mode == lock->l_granted_mode) { // No changes? Just return.
1532 *flags |= LDLM_FL_BLOCK_GRANTED;
1533 RETURN(lock->l_resource);
1536 LASSERTF(new_mode == LCK_PW && lock->l_granted_mode == LCK_PR,
1537 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
1539 lock_res_and_lock(lock);
1541 res = lock->l_resource;
1542 ns = res->lr_namespace;
1544 old_mode = lock->l_req_mode;
1545 lock->l_req_mode = new_mode;
1546 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
1547 /* remember the lock position where the lock might be
1548 * added back to the granted list later and also
1549 * remember the join mode for skiplist fixing. */
1550 if (LDLM_SL_HEAD(&lock->l_sl_mode))
1551 join = LDLM_MODE_JOIN_RIGHT;
1552 else if (LDLM_SL_TAIL(&lock->l_sl_mode))
1553 join = LDLM_MODE_JOIN_LEFT;
1554 if (LDLM_SL_HEAD(&lock->l_sl_policy))
1555 join |= LDLM_POLICY_JOIN_RIGHT;
1556 else if (LDLM_SL_TAIL(&lock->l_sl_policy))
1557 join |= LDLM_POLICY_JOIN_LEFT;
1559 LASSERT(!((join & LDLM_MODE_JOIN_RIGHT) &&
1560 (join & LDLM_POLICY_JOIN_LEFT)));
1561 LASSERT(!((join & LDLM_MODE_JOIN_LEFT) &&
1562 (join & LDLM_POLICY_JOIN_RIGHT)));
1564 if ((join & LDLM_MODE_JOIN_LEFT) ||
1565 (join & LDLM_POLICY_JOIN_LEFT))
1566 mark_lock = list_entry(lock->l_res_link.prev,
1567 struct ldlm_lock, l_res_link);
1568 else if (lock->l_res_link.next != &res->lr_granted)
1569 mark_lock = list_entry(lock->l_res_link.next,
1570 struct ldlm_lock, l_res_link);
1572 ldlm_resource_unlink_lock(lock);
1574 /* If this is a local resource, put it on the appropriate list. */
1575 if (res->lr_namespace->ns_client) {
1576 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
1577 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1579 /* This should never happen, because of the way the
1580 * server handles conversions. */
1581 LDLM_ERROR(lock, "Erroneous flags %d on local lock\n",
1585 ldlm_grant_lock(lock, &rpc_list);
1587 /* FIXME: completion handling not with ns_lock held ! */
1588 if (lock->l_completion_ast)
1589 lock->l_completion_ast(lock, 0, NULL);
1593 ldlm_processing_policy policy;
1594 policy = ldlm_processing_policy_table[res->lr_type];
1595 rc = policy(lock, &pflags, 0, &err, &rpc_list);
1596 if (rc == LDLM_ITER_STOP) {
1597 lock->l_req_mode = old_mode;
1598 ldlm_granted_list_add_lock(lock, mark_lock, join);
1601 *flags |= LDLM_FL_BLOCK_GRANTED;
1605 unlock_res_and_lock(lock);
1608 ldlm_run_cp_ast_work(&rpc_list);
1612 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos)
1614 struct obd_device *obd = NULL;
1616 if (!((libcfs_debug | D_ERROR) & level))
1620 CDEBUG(level, " NULL LDLM lock\n");
1624 CDEBUG(level," -- Lock dump: %p/"LPX64" (rc: %d) (pos: %d) (pid: %d)\n",
1625 lock, lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1627 if (lock->l_conn_export != NULL)
1628 obd = lock->l_conn_export->exp_obd;
1629 if (lock->l_export && lock->l_export->exp_connection) {
1630 CDEBUG(level, " Node: NID %s (rhandle: "LPX64")\n",
1631 libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid),
1632 lock->l_remote_handle.cookie);
1633 } else if (obd == NULL) {
1634 CDEBUG(level, " Node: local\n");
1636 struct obd_import *imp = obd->u.cli.cl_import;
1637 CDEBUG(level, " Node: NID %s (rhandle: "LPX64")\n",
1638 libcfs_nid2str(imp->imp_connection->c_peer.nid),
1639 lock->l_remote_handle.cookie);
1641 CDEBUG(level, " Resource: %p ("LPU64"/"LPU64"/"LPU64")\n",
1643 lock->l_resource->lr_name.name[0],
1644 lock->l_resource->lr_name.name[1],
1645 lock->l_resource->lr_name.name[2]);
1646 CDEBUG(level, " Req mode: %s, grant mode: %s, rc: %u, read: %d, "
1647 "write: %d flags: %#x\n", ldlm_lockname[lock->l_req_mode],
1648 ldlm_lockname[lock->l_granted_mode],
1649 atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers,
1651 if (lock->l_resource->lr_type == LDLM_EXTENT)
1652 CDEBUG(level, " Extent: "LPU64" -> "LPU64
1653 " (req "LPU64"-"LPU64")\n",
1654 lock->l_policy_data.l_extent.start,
1655 lock->l_policy_data.l_extent.end,
1656 lock->l_req_extent.start, lock->l_req_extent.end);
1657 else if (lock->l_resource->lr_type == LDLM_FLOCK)
1658 CDEBUG(level, " Pid: %d Extent: "LPU64" -> "LPU64"\n",
1659 lock->l_policy_data.l_flock.pid,
1660 lock->l_policy_data.l_flock.start,
1661 lock->l_policy_data.l_flock.end);
1662 else if (lock->l_resource->lr_type == LDLM_IBITS)
1663 CDEBUG(level, " Bits: "LPX64"\n",
1664 lock->l_policy_data.l_inodebits.bits);
1667 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1669 struct ldlm_lock *lock;
1671 if (!((libcfs_debug | D_ERROR) & level))
1674 lock = ldlm_handle2lock(lockh);
1678 ldlm_lock_dump(D_OTHER, lock, 0);
1680 LDLM_LOCK_PUT(lock);
1683 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level,
1684 struct libcfs_debug_msg_data *data, const char *fmt,
1688 cfs_debug_limit_state_t *cdls = data->msg_cdls;
1690 va_start(args, fmt);
1692 if (lock->l_resource == NULL) {
1693 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1694 data->msg_fn, data->msg_line, fmt, args,
1695 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1696 "res: \?\? rrc=\?\? type: \?\?\? flags: %x remote: "
1697 LPX64" expref: %d pid: %u\n", lock,
1698 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1699 lock->l_readers, lock->l_writers,
1700 ldlm_lockname[lock->l_granted_mode],
1701 ldlm_lockname[lock->l_req_mode],
1702 lock->l_flags, lock->l_remote_handle.cookie,
1704 atomic_read(&lock->l_export->exp_refcount) : -99,
1710 switch (lock->l_resource->lr_type) {
1712 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1713 data->msg_fn, data->msg_line, fmt, args,
1714 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1715 "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
1716 "] (req "LPU64"->"LPU64") flags: %x remote: "LPX64
1717 " expref: %d pid: %u\n",
1718 lock->l_resource->lr_namespace->ns_name, lock,
1719 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1720 lock->l_readers, lock->l_writers,
1721 ldlm_lockname[lock->l_granted_mode],
1722 ldlm_lockname[lock->l_req_mode],
1723 lock->l_resource->lr_name.name[0],
1724 lock->l_resource->lr_name.name[1],
1725 atomic_read(&lock->l_resource->lr_refcount),
1726 ldlm_typename[lock->l_resource->lr_type],
1727 lock->l_policy_data.l_extent.start,
1728 lock->l_policy_data.l_extent.end,
1729 lock->l_req_extent.start, lock->l_req_extent.end,
1730 lock->l_flags, lock->l_remote_handle.cookie,
1732 atomic_read(&lock->l_export->exp_refcount) : -99,
1737 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1738 data->msg_fn, data->msg_line, fmt, args,
1739 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1740 "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
1741 "["LPU64"->"LPU64"] flags: %x remote: "LPX64
1742 " expref: %d pid: %u\n",
1743 lock->l_resource->lr_namespace->ns_name, lock,
1744 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1745 lock->l_readers, lock->l_writers,
1746 ldlm_lockname[lock->l_granted_mode],
1747 ldlm_lockname[lock->l_req_mode],
1748 lock->l_resource->lr_name.name[0],
1749 lock->l_resource->lr_name.name[1],
1750 atomic_read(&lock->l_resource->lr_refcount),
1751 ldlm_typename[lock->l_resource->lr_type],
1752 lock->l_policy_data.l_flock.pid,
1753 lock->l_policy_data.l_flock.start,
1754 lock->l_policy_data.l_flock.end,
1755 lock->l_flags, lock->l_remote_handle.cookie,
1757 atomic_read(&lock->l_export->exp_refcount) : -99,
1762 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1763 data->msg_fn, data->msg_line, fmt, args,
1764 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1765 "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
1766 "flags: %x remote: "LPX64" expref: %d "
1768 lock->l_resource->lr_namespace->ns_name,
1769 lock, lock->l_handle.h_cookie,
1770 atomic_read (&lock->l_refc),
1771 lock->l_readers, lock->l_writers,
1772 ldlm_lockname[lock->l_granted_mode],
1773 ldlm_lockname[lock->l_req_mode],
1774 lock->l_resource->lr_name.name[0],
1775 lock->l_resource->lr_name.name[1],
1776 lock->l_policy_data.l_inodebits.bits,
1777 atomic_read(&lock->l_resource->lr_refcount),
1778 ldlm_typename[lock->l_resource->lr_type],
1779 lock->l_flags, lock->l_remote_handle.cookie,
1781 atomic_read(&lock->l_export->exp_refcount) : -99,
1786 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1787 data->msg_fn, data->msg_line, fmt, args,
1788 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1789 "res: "LPU64"/"LPU64" rrc: %d type: %s flags: %x "
1790 "remote: "LPX64" expref: %d pid: %u\n",
1791 lock->l_resource->lr_namespace->ns_name,
1792 lock, lock->l_handle.h_cookie,
1793 atomic_read (&lock->l_refc),
1794 lock->l_readers, lock->l_writers,
1795 ldlm_lockname[lock->l_granted_mode],
1796 ldlm_lockname[lock->l_req_mode],
1797 lock->l_resource->lr_name.name[0],
1798 lock->l_resource->lr_name.name[1],
1799 atomic_read(&lock->l_resource->lr_refcount),
1800 ldlm_typename[lock->l_resource->lr_type],
1801 lock->l_flags, lock->l_remote_handle.cookie,
1803 atomic_read(&lock->l_export->exp_refcount) : -99,
1809 EXPORT_SYMBOL(_ldlm_lock_debug);