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 spin_lock_init(&lock->l_lock);
295 lock->l_resource = ldlm_resource_getref(resource);
297 atomic_set(&lock->l_refc, 2);
298 CFS_INIT_LIST_HEAD(&lock->l_res_link);
299 CFS_INIT_LIST_HEAD(&lock->l_lru);
300 CFS_INIT_LIST_HEAD(&lock->l_export_chain);
301 CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
302 CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
303 CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
304 cfs_waitq_init(&lock->l_waitq);
305 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;
356 unlock_res_and_lock(lock);
358 /* ...and the flowers are still standing! */
359 ldlm_resource_putref(oldres);
368 void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh)
370 lockh->cookie = lock->l_handle.h_cookie;
373 /* if flags: atomically get the lock and set the flags.
374 * Return NULL if flag already set
377 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle,
380 struct ldlm_namespace *ns;
381 struct ldlm_lock *lock, *retval = NULL;
386 lock = class_handle2object(handle->cookie);
390 LASSERT(lock->l_resource != NULL);
391 ns = lock->l_resource->lr_namespace;
394 lock_res_and_lock(lock);
396 /* It's unlikely but possible that someone marked the lock as
397 * destroyed after we did handle2object on it */
398 if (lock->l_destroyed) {
399 unlock_res_and_lock(lock);
400 CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock);
405 if (flags && (lock->l_flags & flags)) {
406 unlock_res_and_lock(lock);
412 lock->l_flags |= flags;
414 unlock_res_and_lock(lock);
421 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *ns,
422 const struct lustre_handle *handle)
424 struct ldlm_lock *retval = NULL;
425 retval = __ldlm_handle2lock(handle, 0);
429 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
431 struct obd_export *exp = lock->l_export?:lock->l_conn_export;
432 /* INODEBITS_INTEROP: If the other side does not support
433 * inodebits, reply with a plain lock descriptor.
435 if ((lock->l_resource->lr_type == LDLM_IBITS) &&
436 (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) {
437 struct ldlm_resource res = *lock->l_resource;
439 /* Make sure all the right bits are set in this lock we
440 are going to pass to client */
441 LASSERTF(lock->l_policy_data.l_inodebits.bits ==
442 (MDS_INODELOCK_LOOKUP|MDS_INODELOCK_UPDATE),
443 "Inappropriate inode lock bits during "
444 "conversion " LPU64 "\n",
445 lock->l_policy_data.l_inodebits.bits);
446 res.lr_type = LDLM_PLAIN;
447 ldlm_res2desc(&res, &desc->l_resource);
448 /* Convert "new" lock mode to something old client can
450 if ((lock->l_req_mode == LCK_CR) ||
451 (lock->l_req_mode == LCK_CW))
452 desc->l_req_mode = LCK_PR;
454 desc->l_req_mode = lock->l_req_mode;
455 if ((lock->l_granted_mode == LCK_CR) ||
456 (lock->l_granted_mode == LCK_CW)) {
457 desc->l_granted_mode = LCK_PR;
459 /* We never grant PW/EX locks to clients */
460 LASSERT((lock->l_granted_mode != LCK_PW) &&
461 (lock->l_granted_mode != LCK_EX));
462 desc->l_granted_mode = lock->l_granted_mode;
465 /* We do not copy policy here, because there is no
466 policy for plain locks */
468 ldlm_res2desc(lock->l_resource, &desc->l_resource);
469 desc->l_req_mode = lock->l_req_mode;
470 desc->l_granted_mode = lock->l_granted_mode;
471 desc->l_policy_data = lock->l_policy_data;
475 void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
476 struct list_head *work_list)
478 if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) {
479 LDLM_DEBUG(lock, "lock incompatible; sending blocking AST.");
480 lock->l_flags |= LDLM_FL_AST_SENT;
481 /* If the enqueuing client said so, tell the AST recipient to
482 * discard dirty data, rather than writing back. */
483 if (new->l_flags & LDLM_AST_DISCARD_DATA)
484 lock->l_flags |= LDLM_FL_DISCARD_DATA;
485 LASSERT(list_empty(&lock->l_bl_ast));
486 list_add(&lock->l_bl_ast, work_list);
488 LASSERT(lock->l_blocking_lock == NULL);
489 lock->l_blocking_lock = LDLM_LOCK_GET(new);
493 void ldlm_add_cp_work_item(struct ldlm_lock *lock, struct list_head *work_list)
495 if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) {
496 lock->l_flags |= LDLM_FL_CP_REQD;
497 LDLM_DEBUG(lock, "lock granted; sending completion AST.");
498 LASSERT(list_empty(&lock->l_cp_ast));
499 list_add(&lock->l_cp_ast, work_list);
504 /* must be called with lr_lock held */
505 void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new,
506 struct list_head *work_list)
509 check_res_locked(lock->l_resource);
511 ldlm_add_bl_work_item(lock, new, work_list);
513 ldlm_add_cp_work_item(lock, work_list);
517 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode)
519 struct ldlm_lock *lock;
521 lock = ldlm_handle2lock(lockh);
522 LASSERT(lock != NULL);
523 ldlm_lock_addref_internal(lock, mode);
527 void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
529 ldlm_lock_remove_from_lru(lock);
530 if (mode & (LCK_NL | LCK_CR | LCK_PR))
532 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP))
534 lock->l_last_used = cfs_time_current();
536 LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
539 /* only called for local locks */
540 void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
542 lock_res_and_lock(lock);
543 ldlm_lock_addref_internal_nolock(lock, mode);
544 unlock_res_and_lock(lock);
547 void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
549 struct ldlm_namespace *ns;
552 lock_res_and_lock(lock);
554 ns = lock->l_resource->lr_namespace;
556 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
557 if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
558 LASSERT(lock->l_readers > 0);
561 if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP)) {
562 LASSERT(lock->l_writers > 0);
566 if (lock->l_flags & LDLM_FL_LOCAL &&
567 !lock->l_readers && !lock->l_writers) {
568 /* If this is a local lock on a server namespace and this was
569 * the last reference, cancel the lock. */
570 CDEBUG(D_INFO, "forcing cancel of local lock\n");
571 lock->l_flags |= LDLM_FL_CBPENDING;
574 if (!lock->l_readers && !lock->l_writers &&
575 (lock->l_flags & LDLM_FL_CBPENDING)) {
576 /* If we received a blocked AST and this was the last reference,
577 * run the callback. */
578 if (ns->ns_client == LDLM_NAMESPACE_SERVER && lock->l_export)
579 CERROR("FL_CBPENDING set on non-local lock--just a "
582 LDLM_DEBUG(lock, "final decref done on cbpending lock");
584 LDLM_LOCK_GET(lock); /* dropped by bl thread */
585 ldlm_lock_remove_from_lru(lock);
586 unlock_res_and_lock(lock);
587 if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
588 ldlm_bl_to_thread(ns, NULL, lock, 0) != 0)
589 ldlm_handle_bl_callback(ns, NULL, lock);
590 } else if (ns->ns_client == LDLM_NAMESPACE_CLIENT &&
591 !lock->l_readers && !lock->l_writers &&
592 !(lock->l_flags & LDLM_FL_NO_LRU)) {
593 /* If this is a client-side namespace and this was the last
594 * reference, put it on the LRU. */
595 LASSERT(list_empty(&lock->l_lru));
596 LASSERT(ns->ns_nr_unused >= 0);
597 lock->l_last_used = cfs_time_current();
598 spin_lock(&ns->ns_unused_lock);
599 list_add_tail(&lock->l_lru, &ns->ns_unused_list);
601 spin_unlock(&ns->ns_unused_lock);
602 unlock_res_and_lock(lock);
603 /* Call ldlm_cancel_lru() only if EARLY_CANCEL is not supported
604 * by the server, otherwise, it is done on enqueue. */
605 if (!exp_connect_cancelset(lock->l_conn_export))
606 ldlm_cancel_lru(ns, LDLM_ASYNC);
608 unlock_res_and_lock(lock);
611 LDLM_LOCK_PUT(lock); /* matches the ldlm_lock_get in addref */
616 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode)
618 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
619 LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie);
620 ldlm_lock_decref_internal(lock, mode);
624 /* This will drop a lock reference and mark it for destruction, but will not
625 * necessarily cancel the lock before returning. */
626 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode)
628 struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0);
631 LASSERT(lock != NULL);
633 LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
634 lock_res_and_lock(lock);
635 lock->l_flags |= LDLM_FL_CBPENDING;
636 unlock_res_and_lock(lock);
637 ldlm_lock_decref_internal(lock, mode);
642 * search_granted_lock
645 * Finds a position to insert the new lock.
647 * queue [input]: the granted list where search acts on;
648 * req [input]: the lock whose position to be located;
649 * lockp [output]: the position where the lock should be inserted before, or
650 * NULL indicating @req should be appended to @queue.
652 * Bit-masks combination of following values indicating in which way the
653 * lock need to be inserted.
654 * - LDLM_JOIN_NONE: noting about skip list needs to be fixed;
655 * - LDLM_MODE_JOIN_RIGHT: @req needs join right becoming the head of a
657 * - LDLM_POLICY_JOIN_RIGHT: @req needs join right becoming the head of
660 * - ldlm_grant_lock_with_skiplist
662 static int search_granted_lock(struct list_head *queue,
663 struct ldlm_lock *req,
664 struct ldlm_lock **lockp)
666 struct list_head *tmp, *tmp_tail;
667 struct ldlm_lock *lock, *mode_head_lock;
668 int rc = LDLM_JOIN_NONE;
671 list_for_each(tmp, queue) {
672 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
674 if (lock->l_req_mode != req->l_req_mode) {
675 if (LDLM_SL_HEAD(&lock->l_sl_mode))
676 tmp = &list_entry(lock->l_sl_mode.next,
678 l_sl_mode)->l_res_link;
682 /* found the same mode group */
683 if (lock->l_resource->lr_type == LDLM_PLAIN) {
685 rc = LDLM_MODE_JOIN_RIGHT;
687 } else if (lock->l_resource->lr_type == LDLM_IBITS) {
689 if (LDLM_SL_HEAD(&lock->l_sl_mode))
690 tmp_tail = &list_entry(lock->l_sl_mode.next,
692 l_sl_mode)->l_res_link;
693 mode_head_lock = lock;
695 if (lock->l_policy_data.l_inodebits.bits ==
696 req->l_policy_data.l_inodebits.bits) {
697 /* matched policy lock is found */
699 rc |= LDLM_POLICY_JOIN_RIGHT;
701 /* if the policy group head is also a
702 * mode group head or a single mode
704 if (LDLM_SL_HEAD(&lock->l_sl_mode) ||
706 LDLM_SL_EMPTY(&lock->l_sl_mode)))
707 rc |= LDLM_MODE_JOIN_RIGHT;
711 if (LDLM_SL_HEAD(&lock->l_sl_policy))
712 tmp = &list_entry(lock->l_sl_policy.next,
714 l_sl_policy)->l_res_link;
720 lock = list_entry(tmp, struct ldlm_lock,
722 } /* for all locks in the matched mode group */
724 /* no matched policy group is found, insert before
725 * the mode group head lock */
726 *lockp = mode_head_lock;
727 rc = LDLM_MODE_JOIN_RIGHT;
730 LDLM_ERROR(lock, "is not LDLM_PLAIN or LDLM_IBITS lock");
735 /* no matched mode group is found, append to the end */
743 static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
744 struct ldlm_lock *lockp,
747 struct ldlm_resource *res = lock->l_resource;
750 LASSERT(lockp || join == LDLM_JOIN_NONE);
752 check_res_locked(res);
754 ldlm_resource_dump(D_OTHER, res);
755 CDEBUG(D_OTHER, "About to add this lock:\n");
756 ldlm_lock_dump(D_OTHER, lock, 0);
758 if (lock->l_destroyed) {
759 CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
763 LASSERT(list_empty(&lock->l_res_link));
766 list_add_tail(&lock->l_res_link, &lock->l_resource->lr_granted);
767 else if ((join & LDLM_MODE_JOIN_LEFT) || (join & LDLM_POLICY_JOIN_LEFT))
768 list_add(&lock->l_res_link, &lockp->l_res_link);
770 list_add_tail(&lock->l_res_link, &lockp->l_res_link);
773 if (join & LDLM_MODE_JOIN_RIGHT) {
774 LASSERT(! LDLM_SL_TAIL(&lockp->l_sl_mode));
775 if (LDLM_SL_EMPTY(&lockp->l_sl_mode)) {
776 lock->l_sl_mode.next = &lockp->l_sl_mode;
777 lockp->l_sl_mode.prev = &lock->l_sl_mode;
778 } else if (LDLM_SL_HEAD(&lockp->l_sl_mode)) {
779 lock->l_sl_mode.next = lockp->l_sl_mode.next;
780 lockp->l_sl_mode.next = NULL;
781 lock->l_sl_mode.next->prev = &lock->l_sl_mode;
783 } else if (join & LDLM_MODE_JOIN_LEFT) {
784 LASSERT(! LDLM_SL_HEAD(&lockp->l_sl_mode));
785 if (LDLM_SL_EMPTY(&lockp->l_sl_mode)) {
786 lock->l_sl_mode.prev = &lockp->l_sl_mode;
787 lockp->l_sl_mode.next = &lock->l_sl_mode;
788 } else if (LDLM_SL_TAIL(&lockp->l_sl_mode)) {
789 lock->l_sl_mode.prev = lockp->l_sl_mode.prev;
790 lockp->l_sl_mode.prev = NULL;
791 lock->l_sl_mode.prev->next = &lock->l_sl_mode;
795 if (join & LDLM_POLICY_JOIN_RIGHT) {
796 LASSERT(! LDLM_SL_TAIL(&lockp->l_sl_policy));
797 if (LDLM_SL_EMPTY(&lockp->l_sl_policy)) {
798 lock->l_sl_policy.next = &lockp->l_sl_policy;
799 lockp->l_sl_policy.prev = &lock->l_sl_policy;
800 } else if (LDLM_SL_HEAD(&lockp->l_sl_policy)) {
801 lock->l_sl_policy.next = lockp->l_sl_policy.next;
802 lockp->l_sl_policy.next = NULL;
803 lock->l_sl_policy.next->prev = &lock->l_sl_policy;
805 } else if (join & LDLM_POLICY_JOIN_LEFT) {
806 LASSERT(! LDLM_SL_HEAD(&lockp->l_sl_policy));
807 if (LDLM_SL_EMPTY(&lockp->l_sl_policy)) {
808 lock->l_sl_policy.prev = &lockp->l_sl_policy;
809 lockp->l_sl_policy.next = &lock->l_sl_policy;
810 } else if (LDLM_SL_TAIL(&lockp->l_sl_policy)) {
811 lock->l_sl_policy.prev = lockp->l_sl_policy.prev;
812 lockp->l_sl_policy.prev = NULL;
813 lock->l_sl_policy.prev->next = &lock->l_sl_policy;
820 static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
822 int join = LDLM_JOIN_NONE;
823 struct ldlm_lock *lockp = NULL;
826 LASSERT(lock->l_req_mode == lock->l_granted_mode);
828 join = search_granted_lock(&lock->l_resource->lr_granted, lock, &lockp);
829 ldlm_granted_list_add_lock(lock, lockp, join);
834 * - ldlm_lock_enqueue
835 * - ldlm_reprocess_queue
836 * - ldlm_lock_convert
838 * must be called with lr_lock held
840 void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list)
842 struct ldlm_resource *res = lock->l_resource;
845 check_res_locked(res);
847 lock->l_granted_mode = lock->l_req_mode;
848 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
849 ldlm_grant_lock_with_skiplist(lock);
851 ldlm_resource_add_lock(res, &res->lr_granted, lock);
853 if (lock->l_granted_mode < res->lr_most_restr)
854 res->lr_most_restr = lock->l_granted_mode;
856 if (work_list && lock->l_completion_ast != NULL)
857 ldlm_add_ast_work_item(lock, NULL, work_list);
862 /* returns a referenced lock or NULL. See the flag descriptions below, in the
863 * comment above ldlm_lock_match */
864 static struct ldlm_lock *search_queue(struct list_head *queue, ldlm_mode_t mode,
865 ldlm_policy_data_t *policy,
866 struct ldlm_lock *old_lock, int flags)
868 struct ldlm_lock *lock;
869 struct list_head *tmp;
871 list_for_each(tmp, queue) {
872 lock = list_entry(tmp, struct ldlm_lock, l_res_link);
874 if (lock == old_lock)
877 /* llite sometimes wants to match locks that will be
878 * canceled when their users drop, but we allow it to match
879 * if it passes in CBPENDING and the lock still has users.
880 * this is generally only going to be used by children
881 * whose parents already hold a lock so forward progress
882 * can still happen. */
883 if (lock->l_flags & LDLM_FL_CBPENDING &&
884 !(flags & LDLM_FL_CBPENDING))
886 if (lock->l_flags & LDLM_FL_CBPENDING &&
887 lock->l_readers == 0 && lock->l_writers == 0)
890 if (!(lock->l_req_mode & mode))
893 if (lock->l_resource->lr_type == LDLM_EXTENT &&
894 (lock->l_policy_data.l_extent.start >
895 policy->l_extent.start ||
896 lock->l_policy_data.l_extent.end < policy->l_extent.end))
899 if (unlikely(mode == LCK_GROUP) &&
900 lock->l_resource->lr_type == LDLM_EXTENT &&
901 lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
904 /* We match if we have existing lock with same or wider set
906 if (lock->l_resource->lr_type == LDLM_IBITS &&
907 ((lock->l_policy_data.l_inodebits.bits &
908 policy->l_inodebits.bits) !=
909 policy->l_inodebits.bits))
912 if (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED))
915 if ((flags & LDLM_FL_LOCAL_ONLY) &&
916 !(lock->l_flags & LDLM_FL_LOCAL))
919 if (flags & LDLM_FL_TEST_LOCK)
922 ldlm_lock_addref_internal_nolock(lock, mode);
929 void ldlm_lock_allow_match(struct ldlm_lock *lock)
931 lock_res_and_lock(lock);
932 lock->l_flags |= LDLM_FL_LVB_READY;
933 cfs_waitq_signal(&lock->l_waitq);
934 unlock_res_and_lock(lock);
937 /* Can be called in two ways:
939 * If 'ns' is NULL, then lockh describes an existing lock that we want to look
940 * for a duplicate of.
942 * Otherwise, all of the fields must be filled in, to match against.
944 * If 'flags' contains LDLM_FL_LOCAL_ONLY, then only match local locks on the
945 * server (ie, connh is NULL)
946 * If 'flags' contains LDLM_FL_BLOCK_GRANTED, then only locks on the granted
947 * list will be considered
948 * If 'flags' contains LDLM_FL_CBPENDING, then locks that have been marked
949 * to be canceled can still be matched as long as they still have reader
950 * or writer refernces
951 * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock,
952 * just tell us if we would have matched.
954 * Returns 1 if it finds an already-existing lock that is compatible; in this
955 * case, lockh is filled in with a addref()ed lock
957 * we also check security context, if that failed we simply return 0 (to keep
958 * caller code unchanged), the context failure will be discovered by caller
961 int ldlm_lock_match(struct ldlm_namespace *ns, int flags,
962 const struct ldlm_res_id *res_id, ldlm_type_t type,
963 ldlm_policy_data_t *policy, ldlm_mode_t mode,
964 struct lustre_handle *lockh)
966 struct ldlm_resource *res;
967 struct ldlm_lock *lock, *old_lock = NULL;
972 old_lock = ldlm_handle2lock(lockh);
975 ns = old_lock->l_resource->lr_namespace;
976 res_id = &old_lock->l_resource->lr_name;
977 type = old_lock->l_resource->lr_type;
978 mode = old_lock->l_req_mode;
981 res = ldlm_resource_get(ns, NULL, res_id, type, 0);
983 LASSERT(old_lock == NULL);
989 lock = search_queue(&res->lr_granted, mode, policy, old_lock, flags);
992 if (flags & LDLM_FL_BLOCK_GRANTED)
994 lock = search_queue(&res->lr_converting, mode, policy, old_lock, flags);
997 lock = search_queue(&res->lr_waiting, mode, policy, old_lock, flags);
1004 ldlm_resource_putref(res);
1007 ldlm_lock2handle(lock, lockh);
1008 if ((flags & LDLM_FL_LVB_READY) &&
1009 (!(lock->l_flags & LDLM_FL_LVB_READY))) {
1010 struct l_wait_info lwi;
1011 if (lock->l_completion_ast) {
1012 int err = lock->l_completion_ast(lock,
1013 LDLM_FL_WAIT_NOREPROC,
1016 if (flags & LDLM_FL_TEST_LOCK)
1017 LDLM_LOCK_PUT(lock);
1019 ldlm_lock_decref_internal(lock, mode);
1025 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout), NULL,
1026 LWI_ON_SIGNAL_NOOP, NULL);
1028 /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
1029 l_wait_event(lock->l_waitq,
1030 (lock->l_flags & LDLM_FL_LVB_READY), &lwi);
1035 LDLM_DEBUG(lock, "matched ("LPU64" "LPU64")",
1036 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1037 res_id->name[2] : policy->l_extent.start,
1038 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1039 res_id->name[3] : policy->l_extent.end);
1041 /* check user's security context */
1042 if (lock->l_conn_export &&
1043 sptlrpc_import_check_ctx(
1044 class_exp2cliimp(lock->l_conn_export))) {
1045 if (!(flags & LDLM_FL_TEST_LOCK))
1046 ldlm_lock_decref_internal(lock, mode);
1050 if (flags & LDLM_FL_TEST_LOCK)
1051 LDLM_LOCK_PUT(lock);
1053 } else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
1054 LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
1055 LPU64"/"LPU64" ("LPU64" "LPU64")", ns,
1056 type, mode, res_id->name[0], res_id->name[1],
1057 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1058 res_id->name[2] :policy->l_extent.start,
1059 (type == LDLM_PLAIN || type == LDLM_IBITS) ?
1060 res_id->name[3] : policy->l_extent.end);
1063 LDLM_LOCK_PUT(old_lock);
1068 /* Returns a referenced lock */
1069 struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns,
1070 const struct ldlm_res_id *res_id,
1073 ldlm_blocking_callback blocking,
1074 ldlm_completion_callback completion,
1075 ldlm_glimpse_callback glimpse,
1076 void *data, __u32 lvb_len)
1078 struct ldlm_lock *lock;
1079 struct ldlm_resource *res;
1082 res = ldlm_resource_get(ns, NULL, res_id, type, 1);
1086 lock = ldlm_lock_new(res);
1087 ldlm_resource_putref(res);
1092 lock->l_req_mode = mode;
1093 lock->l_ast_data = data;
1094 lock->l_blocking_ast = blocking;
1095 lock->l_completion_ast = completion;
1096 lock->l_glimpse_ast = glimpse;
1097 lock->l_pid = cfs_curproc_pid();
1100 lock->l_lvb_len = lvb_len;
1101 OBD_ALLOC(lock->l_lvb_data, lvb_len);
1102 if (lock->l_lvb_data == NULL) {
1103 OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
1111 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
1112 struct ldlm_lock **lockp,
1113 void *cookie, int *flags)
1115 struct ldlm_lock *lock = *lockp;
1116 struct ldlm_resource *res = lock->l_resource;
1117 int local = res->lr_namespace->ns_client;
1118 ldlm_processing_policy policy;
1119 ldlm_error_t rc = ELDLM_OK;
1122 do_gettimeofday(&lock->l_enqueued_time);
1123 /* policies are not executed on the client or during replay */
1124 if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
1125 && !local && ns->ns_policy) {
1126 rc = ns->ns_policy(ns, lockp, cookie, lock->l_req_mode, *flags,
1128 if (rc == ELDLM_LOCK_REPLACED) {
1129 /* The lock that was returned has already been granted,
1130 * and placed into lockp. If it's not the same as the
1131 * one we passed in, then destroy the old one and our
1132 * work here is done. */
1133 if (lock != *lockp) {
1134 ldlm_lock_destroy(lock);
1135 LDLM_LOCK_PUT(lock);
1137 *flags |= LDLM_FL_LOCK_CHANGED;
1139 } else if (rc != ELDLM_OK ||
1140 (rc == ELDLM_OK && (*flags & LDLM_FL_INTENT_ONLY))) {
1141 ldlm_lock_destroy(lock);
1146 lock_res_and_lock(lock);
1147 if (local && lock->l_req_mode == lock->l_granted_mode) {
1148 /* The server returned a blocked lock, but it was granted before
1149 * we got a chance to actually enqueue it. We don't need to do
1151 *flags &= ~(LDLM_FL_BLOCK_GRANTED |
1152 LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
1153 GOTO(out, ELDLM_OK);
1156 /* Some flags from the enqueue want to make it into the AST, via the
1157 * lock's l_flags. */
1158 lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
1160 /* This distinction between local lock trees is very important; a client
1161 * namespace only has information about locks taken by that client, and
1162 * thus doesn't have enough information to decide for itself if it can
1163 * be granted (below). In this case, we do exactly what the server
1164 * tells us to do, as dictated by the 'flags'.
1166 * We do exactly the same thing during recovery, when the server is
1167 * more or less trusting the clients not to lie.
1169 * FIXME (bug 268): Detect obvious lies by checking compatibility in
1170 * granted/converting queues. */
1171 ldlm_resource_unlink_lock(lock);
1173 if (*flags & LDLM_FL_BLOCK_CONV)
1174 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1175 else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
1176 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1178 ldlm_grant_lock(lock, NULL);
1179 GOTO(out, ELDLM_OK);
1180 } else if (*flags & LDLM_FL_REPLAY) {
1181 if (*flags & LDLM_FL_BLOCK_CONV) {
1182 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1183 GOTO(out, ELDLM_OK);
1184 } else if (*flags & LDLM_FL_BLOCK_WAIT) {
1185 ldlm_resource_add_lock(res, &res->lr_waiting, lock);
1186 GOTO(out, ELDLM_OK);
1187 } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
1188 ldlm_grant_lock(lock, NULL);
1189 GOTO(out, ELDLM_OK);
1191 /* If no flags, fall through to normal enqueue path. */
1194 policy = ldlm_processing_policy_table[res->lr_type];
1195 policy(lock, flags, 1, &rc, NULL);
1198 unlock_res_and_lock(lock);
1202 /* Must be called with namespace taken: queue is waiting or converting. */
1203 int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
1204 struct list_head *work_list)
1206 struct list_head *tmp, *pos;
1207 ldlm_processing_policy policy;
1209 int rc = LDLM_ITER_CONTINUE;
1213 check_res_locked(res);
1215 policy = ldlm_processing_policy_table[res->lr_type];
1218 list_for_each_safe(tmp, pos, queue) {
1219 struct ldlm_lock *pending;
1220 pending = list_entry(tmp, struct ldlm_lock, l_res_link);
1222 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
1225 rc = policy(pending, &flags, 0, &err, work_list);
1226 if (rc != LDLM_ITER_CONTINUE)
1233 int ldlm_run_bl_ast_work(struct list_head *rpc_list)
1235 struct list_head *tmp, *pos;
1236 struct ldlm_lock_desc d;
1237 int rc = 0, retval = 0;
1240 list_for_each_safe(tmp, pos, rpc_list) {
1241 struct ldlm_lock *lock =
1242 list_entry(tmp, struct ldlm_lock, l_bl_ast);
1244 /* nobody should touch l_bl_ast */
1245 lock_res_and_lock(lock);
1246 list_del_init(&lock->l_bl_ast);
1248 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1249 LASSERT(lock->l_bl_ast_run == 0);
1250 LASSERT(lock->l_blocking_lock);
1251 lock->l_bl_ast_run++;
1252 unlock_res_and_lock(lock);
1254 ldlm_lock2desc(lock->l_blocking_lock, &d);
1256 LDLM_LOCK_PUT(lock->l_blocking_lock);
1257 lock->l_blocking_lock = NULL;
1258 rc = lock->l_blocking_ast(lock, &d, NULL, LDLM_CB_BLOCKING);
1260 if (rc == -ERESTART)
1263 CDEBUG(D_DLMTRACE, "Failed AST - should clean & "
1264 "disconnect client\n");
1265 LDLM_LOCK_PUT(lock);
1270 int ldlm_run_cp_ast_work(struct list_head *rpc_list)
1272 struct list_head *tmp, *pos;
1273 int rc = 0, retval = 0;
1276 /* It's possible to receive a completion AST before we've set
1277 * the l_completion_ast pointer: either because the AST arrived
1278 * before the reply, or simply because there's a small race
1279 * window between receiving the reply and finishing the local
1280 * enqueue. (bug 842)
1282 * This can't happen with the blocking_ast, however, because we
1283 * will never call the local blocking_ast until we drop our
1284 * reader/writer reference, which we won't do until we get the
1285 * reply and finish enqueueing. */
1287 list_for_each_safe(tmp, pos, rpc_list) {
1288 struct ldlm_lock *lock =
1289 list_entry(tmp, struct ldlm_lock, l_cp_ast);
1291 /* nobody should touch l_cp_ast */
1292 lock_res_and_lock(lock);
1293 list_del_init(&lock->l_cp_ast);
1294 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1295 lock->l_flags &= ~LDLM_FL_CP_REQD;
1296 unlock_res_and_lock(lock);
1298 if (lock->l_completion_ast != NULL)
1299 rc = lock->l_completion_ast(lock, 0, 0);
1300 if (rc == -ERESTART)
1303 CDEBUG(D_DLMTRACE, "Failed AST - should clean & "
1304 "disconnect client\n");
1305 LDLM_LOCK_PUT(lock);
1310 static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
1312 ldlm_reprocess_all(res);
1313 return LDLM_ITER_CONTINUE;
1316 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns)
1318 struct list_head *tmp;
1325 spin_lock(&ns->ns_hash_lock);
1326 for (i = 0; i < RES_HASH_SIZE; i++) {
1327 tmp = ns->ns_hash[i].next;
1328 while (tmp != &(ns->ns_hash[i])) {
1329 struct ldlm_resource *res =
1330 list_entry(tmp, struct ldlm_resource, lr_hash);
1332 ldlm_resource_getref(res);
1333 spin_unlock(&ns->ns_hash_lock);
1335 rc = reprocess_one_queue(res, NULL);
1337 spin_lock(&ns->ns_hash_lock);
1339 ldlm_resource_putref_locked(res);
1341 if (rc == LDLM_ITER_STOP)
1346 spin_unlock(&ns->ns_hash_lock);
1350 void ldlm_reprocess_all(struct ldlm_resource *res)
1352 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1356 /* Local lock trees don't get reprocessed. */
1357 if (res->lr_namespace->ns_client) {
1364 rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list);
1365 if (rc == LDLM_ITER_CONTINUE)
1366 ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
1369 rc = ldlm_run_cp_ast_work(&rpc_list);
1370 if (rc == -ERESTART) {
1371 LASSERT(list_empty(&rpc_list));
1377 void ldlm_cancel_callback(struct ldlm_lock *lock)
1379 check_res_locked(lock->l_resource);
1380 if (!(lock->l_flags & LDLM_FL_CANCEL)) {
1381 lock->l_flags |= LDLM_FL_CANCEL;
1382 if (lock->l_blocking_ast) {
1383 // l_check_no_ns_lock(ns);
1384 unlock_res_and_lock(lock);
1385 lock->l_blocking_ast(lock, NULL, lock->l_ast_data,
1387 lock_res_and_lock(lock);
1389 LDLM_DEBUG(lock, "no blocking ast");
1394 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
1396 struct ldlm_lock *lock;
1398 if (req->l_resource->lr_type != LDLM_PLAIN &&
1399 req->l_resource->lr_type != LDLM_IBITS)
1402 if (LDLM_SL_HEAD(&req->l_sl_mode)) {
1403 lock = list_entry(req->l_res_link.next, struct ldlm_lock,
1405 if (req->l_sl_mode.next == &lock->l_sl_mode) {
1406 lock->l_sl_mode.prev = NULL;
1408 lock->l_sl_mode.next = req->l_sl_mode.next;
1409 lock->l_sl_mode.next->prev = &lock->l_sl_mode;
1411 req->l_sl_mode.next = NULL;
1412 } else if (LDLM_SL_TAIL(&req->l_sl_mode)) {
1413 lock = list_entry(req->l_res_link.prev, struct ldlm_lock,
1415 if (req->l_sl_mode.prev == &lock->l_sl_mode) {
1416 lock->l_sl_mode.next = NULL;
1418 lock->l_sl_mode.prev = req->l_sl_mode.prev;
1419 lock->l_sl_mode.prev->next = &lock->l_sl_mode;
1421 req->l_sl_mode.prev = NULL;
1424 if (LDLM_SL_HEAD(&req->l_sl_policy)) {
1425 lock = list_entry(req->l_res_link.next, struct ldlm_lock,
1427 if (req->l_sl_policy.next == &lock->l_sl_policy) {
1428 lock->l_sl_policy.prev = NULL;
1430 lock->l_sl_policy.next = req->l_sl_policy.next;
1431 lock->l_sl_policy.next->prev = &lock->l_sl_policy;
1433 req->l_sl_policy.next = NULL;
1434 } else if (LDLM_SL_TAIL(&req->l_sl_policy)) {
1435 lock = list_entry(req->l_res_link.prev, struct ldlm_lock,
1437 if (req->l_sl_policy.prev == &lock->l_sl_policy) {
1438 lock->l_sl_policy.next = NULL;
1440 lock->l_sl_policy.prev = req->l_sl_policy.prev;
1441 lock->l_sl_policy.prev->next = &lock->l_sl_policy;
1443 req->l_sl_policy.prev = NULL;
1447 void ldlm_lock_cancel(struct ldlm_lock *lock)
1449 struct ldlm_resource *res;
1450 struct ldlm_namespace *ns;
1453 lock_res_and_lock(lock);
1455 res = lock->l_resource;
1456 ns = res->lr_namespace;
1458 /* Please do not, no matter how tempting, remove this LBUG without
1459 * talking to me first. -phik */
1460 if (lock->l_readers || lock->l_writers) {
1461 LDLM_ERROR(lock, "lock still has references");
1465 ldlm_del_waiting_lock(lock);
1467 /* Releases cancel callback. */
1468 ldlm_cancel_callback(lock);
1470 /* Yes, second time, just in case it was added again while we were
1471 running with no res lock in ldlm_cancel_callback */
1472 ldlm_del_waiting_lock(lock);
1473 ldlm_resource_unlink_lock(lock);
1474 ldlm_lock_destroy_nolock(lock);
1475 unlock_res_and_lock(lock);
1480 int ldlm_lock_set_data(struct lustre_handle *lockh, void *data)
1482 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
1488 lock->l_ast_data = data;
1489 LDLM_LOCK_PUT(lock);
1493 void ldlm_cancel_locks_for_export(struct obd_export *exp)
1495 struct ldlm_lock *lock;
1496 struct ldlm_resource *res;
1498 spin_lock(&exp->exp_ldlm_data.led_lock);
1499 while(!list_empty(&exp->exp_ldlm_data.led_held_locks)) {
1500 lock = list_entry(exp->exp_ldlm_data.led_held_locks.next,
1501 struct ldlm_lock, l_export_chain);
1502 res = ldlm_resource_getref(lock->l_resource);
1503 LDLM_LOCK_GET(lock);
1504 spin_unlock(&exp->exp_ldlm_data.led_lock);
1506 LDLM_DEBUG(lock, "export %p", exp);
1507 ldlm_lock_cancel(lock);
1508 ldlm_reprocess_all(res);
1510 ldlm_resource_putref(res);
1511 LDLM_LOCK_PUT(lock);
1512 spin_lock(&exp->exp_ldlm_data.led_lock);
1514 spin_unlock(&exp->exp_ldlm_data.led_lock);
1517 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1520 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1521 struct ldlm_resource *res;
1522 struct ldlm_namespace *ns;
1525 struct ldlm_lock *mark_lock = NULL;
1526 int join = LDLM_JOIN_NONE;
1530 if (new_mode == lock->l_granted_mode) { // No changes? Just return.
1531 *flags |= LDLM_FL_BLOCK_GRANTED;
1532 RETURN(lock->l_resource);
1535 LASSERTF(new_mode == LCK_PW && lock->l_granted_mode == LCK_PR,
1536 "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
1538 lock_res_and_lock(lock);
1540 res = lock->l_resource;
1541 ns = res->lr_namespace;
1543 old_mode = lock->l_req_mode;
1544 lock->l_req_mode = new_mode;
1545 if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
1546 /* remember the lock position where the lock might be
1547 * added back to the granted list later and also
1548 * remember the join mode for skiplist fixing. */
1549 if (LDLM_SL_HEAD(&lock->l_sl_mode))
1550 join = LDLM_MODE_JOIN_RIGHT;
1551 else if (LDLM_SL_TAIL(&lock->l_sl_mode))
1552 join = LDLM_MODE_JOIN_LEFT;
1553 if (LDLM_SL_HEAD(&lock->l_sl_policy))
1554 join |= LDLM_POLICY_JOIN_RIGHT;
1555 else if (LDLM_SL_TAIL(&lock->l_sl_policy))
1556 join |= LDLM_POLICY_JOIN_LEFT;
1558 LASSERT(!((join & LDLM_MODE_JOIN_RIGHT) &&
1559 (join & LDLM_POLICY_JOIN_LEFT)));
1560 LASSERT(!((join & LDLM_MODE_JOIN_LEFT) &&
1561 (join & LDLM_POLICY_JOIN_RIGHT)));
1563 if ((join & LDLM_MODE_JOIN_LEFT) ||
1564 (join & LDLM_POLICY_JOIN_LEFT))
1565 mark_lock = list_entry(lock->l_res_link.prev,
1566 struct ldlm_lock, l_res_link);
1567 else if (lock->l_res_link.next != &res->lr_granted)
1568 mark_lock = list_entry(lock->l_res_link.next,
1569 struct ldlm_lock, l_res_link);
1571 ldlm_resource_unlink_lock(lock);
1573 /* If this is a local resource, put it on the appropriate list. */
1574 if (res->lr_namespace->ns_client) {
1575 if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
1576 ldlm_resource_add_lock(res, &res->lr_converting, lock);
1578 /* This should never happen, because of the way the
1579 * server handles conversions. */
1580 LDLM_ERROR(lock, "Erroneous flags %d on local lock\n",
1584 ldlm_grant_lock(lock, &rpc_list);
1586 /* FIXME: completion handling not with ns_lock held ! */
1587 if (lock->l_completion_ast)
1588 lock->l_completion_ast(lock, 0, NULL);
1592 ldlm_processing_policy policy;
1593 policy = ldlm_processing_policy_table[res->lr_type];
1594 rc = policy(lock, &pflags, 0, &err, &rpc_list);
1595 if (rc == LDLM_ITER_STOP) {
1596 lock->l_req_mode = old_mode;
1597 ldlm_granted_list_add_lock(lock, mark_lock, join);
1600 *flags |= LDLM_FL_BLOCK_GRANTED;
1604 unlock_res_and_lock(lock);
1607 ldlm_run_cp_ast_work(&rpc_list);
1611 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos)
1613 struct obd_device *obd = NULL;
1615 if (!((libcfs_debug | D_ERROR) & level))
1619 CDEBUG(level, " NULL LDLM lock\n");
1623 CDEBUG(level," -- Lock dump: %p/"LPX64" (rc: %d) (pos: %d) (pid: %d)\n",
1624 lock, lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1626 if (lock->l_conn_export != NULL)
1627 obd = lock->l_conn_export->exp_obd;
1628 if (lock->l_export && lock->l_export->exp_connection) {
1629 CDEBUG(level, " Node: NID %s (rhandle: "LPX64")\n",
1630 libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid),
1631 lock->l_remote_handle.cookie);
1632 } else if (obd == NULL) {
1633 CDEBUG(level, " Node: local\n");
1635 struct obd_import *imp = obd->u.cli.cl_import;
1636 CDEBUG(level, " Node: NID %s (rhandle: "LPX64")\n",
1637 libcfs_nid2str(imp->imp_connection->c_peer.nid),
1638 lock->l_remote_handle.cookie);
1640 CDEBUG(level, " Resource: %p ("LPU64"/"LPU64"/"LPU64")\n",
1642 lock->l_resource->lr_name.name[0],
1643 lock->l_resource->lr_name.name[1],
1644 lock->l_resource->lr_name.name[2]);
1645 CDEBUG(level, " Req mode: %s, grant mode: %s, rc: %u, read: %d, "
1646 "write: %d flags: %#x\n", ldlm_lockname[lock->l_req_mode],
1647 ldlm_lockname[lock->l_granted_mode],
1648 atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers,
1650 if (lock->l_resource->lr_type == LDLM_EXTENT)
1651 CDEBUG(level, " Extent: "LPU64" -> "LPU64
1652 " (req "LPU64"-"LPU64")\n",
1653 lock->l_policy_data.l_extent.start,
1654 lock->l_policy_data.l_extent.end,
1655 lock->l_req_extent.start, lock->l_req_extent.end);
1656 else if (lock->l_resource->lr_type == LDLM_FLOCK)
1657 CDEBUG(level, " Pid: %d Extent: "LPU64" -> "LPU64"\n",
1658 lock->l_policy_data.l_flock.pid,
1659 lock->l_policy_data.l_flock.start,
1660 lock->l_policy_data.l_flock.end);
1661 else if (lock->l_resource->lr_type == LDLM_IBITS)
1662 CDEBUG(level, " Bits: "LPX64"\n",
1663 lock->l_policy_data.l_inodebits.bits);
1666 void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh)
1668 struct ldlm_lock *lock;
1670 if (!((libcfs_debug | D_ERROR) & level))
1673 lock = ldlm_handle2lock(lockh);
1677 ldlm_lock_dump(D_OTHER, lock, 0);
1679 LDLM_LOCK_PUT(lock);
1682 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level,
1683 struct libcfs_debug_msg_data *data, const char *fmt,
1687 cfs_debug_limit_state_t *cdls = data->msg_cdls;
1689 va_start(args, fmt);
1691 if (lock->l_resource == NULL) {
1692 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1693 data->msg_fn, data->msg_line, fmt, args,
1694 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1695 "res: \?\? rrc=\?\? type: \?\?\? flags: %x remote: "
1696 LPX64" expref: %d pid: %u\n", lock,
1697 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1698 lock->l_readers, lock->l_writers,
1699 ldlm_lockname[lock->l_granted_mode],
1700 ldlm_lockname[lock->l_req_mode],
1701 lock->l_flags, lock->l_remote_handle.cookie,
1703 atomic_read(&lock->l_export->exp_refcount) : -99,
1709 switch (lock->l_resource->lr_type) {
1711 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1712 data->msg_fn, data->msg_line, fmt, args,
1713 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1714 "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
1715 "] (req "LPU64"->"LPU64") flags: %x remote: "LPX64
1716 " expref: %d pid: %u\n",
1717 lock->l_resource->lr_namespace->ns_name, lock,
1718 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1719 lock->l_readers, lock->l_writers,
1720 ldlm_lockname[lock->l_granted_mode],
1721 ldlm_lockname[lock->l_req_mode],
1722 lock->l_resource->lr_name.name[0],
1723 lock->l_resource->lr_name.name[1],
1724 atomic_read(&lock->l_resource->lr_refcount),
1725 ldlm_typename[lock->l_resource->lr_type],
1726 lock->l_policy_data.l_extent.start,
1727 lock->l_policy_data.l_extent.end,
1728 lock->l_req_extent.start, lock->l_req_extent.end,
1729 lock->l_flags, lock->l_remote_handle.cookie,
1731 atomic_read(&lock->l_export->exp_refcount) : -99,
1736 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1737 data->msg_fn, data->msg_line, fmt, args,
1738 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1739 "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
1740 "["LPU64"->"LPU64"] flags: %x remote: "LPX64
1741 " expref: %d pid: %u\n",
1742 lock->l_resource->lr_namespace->ns_name, lock,
1743 lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
1744 lock->l_readers, lock->l_writers,
1745 ldlm_lockname[lock->l_granted_mode],
1746 ldlm_lockname[lock->l_req_mode],
1747 lock->l_resource->lr_name.name[0],
1748 lock->l_resource->lr_name.name[1],
1749 atomic_read(&lock->l_resource->lr_refcount),
1750 ldlm_typename[lock->l_resource->lr_type],
1751 lock->l_policy_data.l_flock.pid,
1752 lock->l_policy_data.l_flock.start,
1753 lock->l_policy_data.l_flock.end,
1754 lock->l_flags, lock->l_remote_handle.cookie,
1756 atomic_read(&lock->l_export->exp_refcount) : -99,
1761 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1762 data->msg_fn, data->msg_line, fmt, args,
1763 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1764 "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
1765 "flags: %x remote: "LPX64" expref: %d "
1767 lock->l_resource->lr_namespace->ns_name,
1768 lock, lock->l_handle.h_cookie,
1769 atomic_read (&lock->l_refc),
1770 lock->l_readers, lock->l_writers,
1771 ldlm_lockname[lock->l_granted_mode],
1772 ldlm_lockname[lock->l_req_mode],
1773 lock->l_resource->lr_name.name[0],
1774 lock->l_resource->lr_name.name[1],
1775 lock->l_policy_data.l_inodebits.bits,
1776 atomic_read(&lock->l_resource->lr_refcount),
1777 ldlm_typename[lock->l_resource->lr_type],
1778 lock->l_flags, lock->l_remote_handle.cookie,
1780 atomic_read(&lock->l_export->exp_refcount) : -99,
1785 libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
1786 data->msg_fn, data->msg_line, fmt, args,
1787 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
1788 "res: "LPU64"/"LPU64" rrc: %d type: %s flags: %x "
1789 "remote: "LPX64" expref: %d pid: %u\n",
1790 lock->l_resource->lr_namespace->ns_name,
1791 lock, lock->l_handle.h_cookie,
1792 atomic_read (&lock->l_refc),
1793 lock->l_readers, lock->l_writers,
1794 ldlm_lockname[lock->l_granted_mode],
1795 ldlm_lockname[lock->l_req_mode],
1796 lock->l_resource->lr_name.name[0],
1797 lock->l_resource->lr_name.name[1],
1798 atomic_read(&lock->l_resource->lr_refcount),
1799 ldlm_typename[lock->l_resource->lr_type],
1800 lock->l_flags, lock->l_remote_handle.cookie,
1802 atomic_read(&lock->l_export->exp_refcount) : -99,
1808 EXPORT_SYMBOL(_ldlm_lock_debug);