/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
- * Author: Peter Braam <braam@clusterfs.com>
- * Author: Phil Schwan <phil@clusterfs.com>
+ * GPL HEADER START
*
- * This file is part of the Lustre file system, http://www.lustre.org
- * Lustre is a trademark of Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * You may have signed or agreed to another license before downloading
- * this software. If so, you are bound by the terms and conditions
- * of that agreement, and the following does not apply to you. See the
- * LICENSE file included with this distribution for more information.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
*
- * If you did not agree to a different license, then this copy of Lustre
- * is open source software; you can redistribute it and/or modify it
- * under the terms of version 2 of the GNU General Public License as
- * published by the Free Software Foundation.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
*
- * In either case, Lustre is distributed in the hope that it will be
- * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * license text for more details.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/ldlm/ldlm_lock.c
+ *
+ * Author: Peter Braam <braam@clusterfs.com>
+ * Author: Phil Schwan <phil@clusterfs.com>
*/
#define DEBUG_SUBSYSTEM S_LDLM
#ifdef __KERNEL__
# include <libcfs/libcfs.h>
+# ifndef HAVE_VFS_INTENT_PATCHES
# include <linux/lustre_intent.h>
+# endif
#else
# include <liblustre.h>
-# include <libcfs/kp30.h>
#endif
#include <obd_class.h>
#include "ldlm_internal.h"
-/* lock's skip list pointers fix mode */
-#define LDLM_JOIN_NONE 0
-#define LDLM_MODE_JOIN_RIGHT 1
-#define LDLM_MODE_JOIN_LEFT (1 << 1)
-#define LDLM_POLICY_JOIN_RIGHT (1 << 2)
-#define LDLM_POLICY_JOIN_LEFT (1 << 3)
-
/* lock types */
char *ldlm_lockname[] = {
[0] "--",
[LCK_CW] "CW",
[LCK_CR] "CR",
[LCK_NL] "NL",
- [LCK_GROUP] "GROUP"
+ [LCK_GROUP] "GROUP",
+ [LCK_COS] "COS"
};
char *ldlm_typename[] = {
LASSERT(list_empty(&lock->l_pending_chain));
atomic_dec(&res->lr_namespace->ns_locks);
+ lu_ref_del(&res->lr_reference, "lock", lock);
ldlm_resource_putref(res);
lock->l_resource = NULL;
if (lock->l_export) {
if (lock->l_lvb_data != NULL)
OBD_FREE(lock->l_lvb_data, lock->l_lvb_len);
- OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle,
- ldlm_lock_free);
+ ldlm_interval_free(ldlm_interval_detach(lock));
+ lu_ref_fini(&lock->l_reference);
+ OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle,
+ ldlm_lock_free);
}
EXIT;
{
int rc = 0;
if (!list_empty(&lock->l_lru)) {
+ struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
list_del_init(&lock->l_lru);
- lock->l_resource->lr_namespace->ns_nr_unused--;
- LASSERT(lock->l_resource->lr_namespace->ns_nr_unused >= 0);
+ LASSERT(ns->ns_nr_unused > 0);
+ ns->ns_nr_unused--;
rc = 1;
}
return rc;
int ldlm_lock_remove_from_lru(struct ldlm_lock *lock)
{
+ struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
int rc;
ENTRY;
- spin_lock(&lock->l_resource->lr_namespace->ns_unused_lock);
+ spin_lock(&ns->ns_unused_lock);
rc = ldlm_lock_remove_from_lru_nolock(lock);
- spin_unlock(&lock->l_resource->lr_namespace->ns_unused_lock);
+ spin_unlock(&ns->ns_unused_lock);
EXIT;
return rc;
}
+void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
+{
+ struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
+ lock->l_last_used = cfs_time_current();
+ LASSERT(list_empty(&lock->l_lru));
+ LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
+ list_add_tail(&lock->l_lru, &ns->ns_unused_list);
+ LASSERT(ns->ns_nr_unused >= 0);
+ ns->ns_nr_unused++;
+}
+
+void ldlm_lock_add_to_lru(struct ldlm_lock *lock)
+{
+ struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
+ ENTRY;
+ spin_lock(&ns->ns_unused_lock);
+ ldlm_lock_add_to_lru_nolock(lock);
+ spin_unlock(&ns->ns_unused_lock);
+ EXIT;
+}
+
+void ldlm_lock_touch_in_lru(struct ldlm_lock *lock)
+{
+ struct ldlm_namespace *ns = lock->l_resource->lr_namespace;
+ ENTRY;
+ spin_lock(&ns->ns_unused_lock);
+ if (!list_empty(&lock->l_lru)) {
+ ldlm_lock_remove_from_lru_nolock(lock);
+ ldlm_lock_add_to_lru_nolock(lock);
+ }
+ spin_unlock(&ns->ns_unused_lock);
+ EXIT;
+}
+
/* This used to have a 'strict' flag, which recovery would use to mark an
* in-use lock as needing-to-die. Lest I am ever tempted to put it back, I
* shall explain why it's gone: with the new hash table scheme, once you call
}
lock->l_destroyed = 1;
- if (lock->l_export)
- spin_lock(&lock->l_export->exp_ldlm_data.led_lock);
- list_del_init(&lock->l_export_chain);
- if (lock->l_export)
- spin_unlock(&lock->l_export->exp_ldlm_data.led_lock);
+ if (lock->l_export && lock->l_export->exp_lock_hash &&
+ !hlist_unhashed(&lock->l_exp_hash))
+ lustre_hash_del(lock->l_export->exp_lock_hash,
+ &lock->l_remote_handle, &lock->l_exp_hash);
ldlm_lock_remove_from_lru(lock);
class_handle_unhash(&lock->l_handle);
unlock_res_and_lock(lock);
/* drop reference from hashtable only for first destroy */
- if (first)
- LDLM_LOCK_PUT(lock);
+ if (first) {
+ lu_ref_del(&lock->l_reference, "hash", lock);
+ LDLM_LOCK_RELEASE(lock);
+ }
EXIT;
}
ENTRY;
first = ldlm_lock_destroy_internal(lock);
/* drop reference from hashtable only for first destroy */
- if (first)
- LDLM_LOCK_PUT(lock);
+ if (first) {
+ lu_ref_del(&lock->l_reference, "hash", lock);
+ LDLM_LOCK_RELEASE(lock);
+ }
EXIT;
}
spin_lock_init(&lock->l_lock);
lock->l_resource = ldlm_resource_getref(resource);
+ lu_ref_add(&resource->lr_reference, "lock", lock);
atomic_set(&lock->l_refc, 2);
CFS_INIT_LIST_HEAD(&lock->l_res_link);
CFS_INIT_LIST_HEAD(&lock->l_lru);
- CFS_INIT_LIST_HEAD(&lock->l_export_chain);
CFS_INIT_LIST_HEAD(&lock->l_pending_chain);
CFS_INIT_LIST_HEAD(&lock->l_bl_ast);
CFS_INIT_LIST_HEAD(&lock->l_cp_ast);
+ CFS_INIT_LIST_HEAD(&lock->l_rk_ast);
cfs_waitq_init(&lock->l_waitq);
lock->l_blocking_lock = NULL;
- lock->l_sl_mode.prev = NULL;
- lock->l_sl_mode.next = NULL;
- lock->l_sl_policy.prev = NULL;
- lock->l_sl_policy.next = NULL;
+ CFS_INIT_LIST_HEAD(&lock->l_sl_mode);
+ CFS_INIT_LIST_HEAD(&lock->l_sl_policy);
+ CFS_INIT_HLIST_NODE(&lock->l_exp_hash);
atomic_inc(&resource->lr_namespace->ns_locks);
CFS_INIT_LIST_HEAD(&lock->l_handle.h_link);
class_handle_hash(&lock->l_handle, lock_handle_addref);
+ CFS_INIT_LIST_HEAD(&lock->l_extents_list);
+ spin_lock_init(&lock->l_extents_list_lock);
+ CFS_INIT_LIST_HEAD(&lock->l_cache_locks_list);
+ lu_ref_init(&lock->l_reference);
+ lu_ref_add(&lock->l_reference, "hash", lock);
+ lock->l_callback_timeout = 0;
+
RETURN(lock);
}
int type;
ENTRY;
- LASSERT(ns->ns_client != 0);
+ LASSERT(ns_is_client(ns));
lock_res_and_lock(lock);
if (memcmp(new_resid, &lock->l_resource->lr_name,
unlock_res_and_lock(lock);
newres = ldlm_resource_get(ns, NULL, new_resid, type, 1);
- if (newres == NULL) {
- LBUG();
+ lu_ref_add(&newres->lr_reference, "lock", lock);
+ if (newres == NULL)
RETURN(-ENOMEM);
+ /*
+ * To flip the lock from the old to the new resource, lock, oldres and
+ * newres have to be locked. Resource spin-locks are nested within
+ * lock->l_lock, and are taken in the memory address order to avoid
+ * dead-locks.
+ */
+ spin_lock(&lock->l_lock);
+ oldres = lock->l_resource;
+ if (oldres < newres) {
+ lock_res(oldres);
+ lock_res_nested(newres, LRT_NEW);
+ } else {
+ lock_res(newres);
+ lock_res_nested(oldres, LRT_NEW);
}
-
- lock_res_and_lock(lock);
- LASSERT(memcmp(new_resid, &lock->l_resource->lr_name,
- sizeof(lock->l_resource->lr_name)) != 0);
- lock_res(newres);
+ LASSERT(memcmp(new_resid, &oldres->lr_name,
+ sizeof oldres->lr_name) != 0);
lock->l_resource = newres;
unlock_res(oldres);
unlock_res_and_lock(lock);
/* ...and the flowers are still standing! */
+ lu_ref_del(&oldres->lr_reference, "lock", lock);
ldlm_resource_putref(oldres);
RETURN(0);
ns = lock->l_resource->lr_namespace;
LASSERT(ns != NULL);
+ lu_ref_add_atomic(&lock->l_reference, "handle", cfs_current());
lock_res_and_lock(lock);
/* It's unlikely but possible that someone marked the lock as
return retval;
}
-struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *ns,
- const struct lustre_handle *handle)
-{
- struct ldlm_lock *retval = NULL;
- retval = __ldlm_handle2lock(handle, 0);
- return retval;
-}
-
void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc)
{
struct obd_export *exp = lock->l_export?:lock->l_conn_export;
void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
{
ldlm_lock_remove_from_lru(lock);
- if (mode & (LCK_NL | LCK_CR | LCK_PR))
+ if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
lock->l_readers++;
- if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP))
+ lu_ref_add_atomic(&lock->l_reference, "reader", lock);
+ }
+ if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
lock->l_writers++;
- lock->l_last_used = cfs_time_current();
+ lu_ref_add_atomic(&lock->l_reference, "writer", lock);
+ }
LDLM_LOCK_GET(lock);
+ lu_ref_add_atomic(&lock->l_reference, "user", lock);
LDLM_DEBUG(lock, "ldlm_lock_addref(%s)", ldlm_lockname[mode]);
}
+/**
+ * Attempts to addref a lock, and fails if lock is already LDLM_FL_CBPENDING
+ * or destroyed.
+ *
+ * \retval 0 success, lock was addref-ed
+ *
+ * \retval -EAGAIN lock is being canceled.
+ */
+int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode)
+{
+ struct ldlm_lock *lock;
+ int result;
+
+ result = -EAGAIN;
+ lock = ldlm_handle2lock(lockh);
+ if (lock != NULL) {
+ lock_res_and_lock(lock);
+ if (lock->l_readers != 0 || lock->l_writers != 0 ||
+ !(lock->l_flags & LDLM_FL_CBPENDING)) {
+ ldlm_lock_addref_internal_nolock(lock, mode);
+ result = 0;
+ }
+ unlock_res_and_lock(lock);
+ LDLM_LOCK_PUT(lock);
+ }
+ return result;
+}
+
/* only called for local locks */
void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode)
{
unlock_res_and_lock(lock);
}
-void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
+/* only called in ldlm_flock_destroy and for local locks.
+ * * for LDLM_FLOCK type locks, l_blocking_ast is null, and
+ * * ldlm_lock_remove_from_lru() does nothing, it is safe
+ * * for ldlm_flock_destroy usage by dropping some code */
+void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode)
{
- struct ldlm_namespace *ns;
- ENTRY;
-
- lock_res_and_lock(lock);
-
- ns = lock->l_resource->lr_namespace;
-
LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]);
if (mode & (LCK_NL | LCK_CR | LCK_PR)) {
LASSERT(lock->l_readers > 0);
+ lu_ref_del(&lock->l_reference, "reader", lock);
lock->l_readers--;
}
- if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP)) {
+ if (mode & (LCK_EX | LCK_CW | LCK_PW | LCK_GROUP | LCK_COS)) {
LASSERT(lock->l_writers > 0);
+ lu_ref_del(&lock->l_reference, "writer", lock);
lock->l_writers--;
}
+ lu_ref_del(&lock->l_reference, "user", lock);
+ LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */
+}
+
+void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode)
+{
+ struct ldlm_namespace *ns;
+ ENTRY;
+
+ lock_res_and_lock(lock);
+
+ ns = lock->l_resource->lr_namespace;
+
+ ldlm_lock_decref_internal_nolock(lock, mode);
+
if (lock->l_flags & LDLM_FL_LOCAL &&
!lock->l_readers && !lock->l_writers) {
/* If this is a local lock on a server namespace and this was
(lock->l_flags & LDLM_FL_CBPENDING)) {
/* If we received a blocked AST and this was the last reference,
* run the callback. */
- if (ns->ns_client == LDLM_NAMESPACE_SERVER && lock->l_export)
+ if (ns_is_server(ns) && lock->l_export)
CERROR("FL_CBPENDING set on non-local lock--just a "
"warning\n");
LDLM_LOCK_GET(lock); /* dropped by bl thread */
ldlm_lock_remove_from_lru(lock);
unlock_res_and_lock(lock);
+
+ if (lock->l_flags & LDLM_FL_FAIL_LOC)
+ OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
+
if ((lock->l_flags & LDLM_FL_ATOMIC_CB) ||
- ldlm_bl_to_thread(ns, NULL, lock, 0) != 0)
+ ldlm_bl_to_thread_lock(ns, NULL, lock) != 0)
ldlm_handle_bl_callback(ns, NULL, lock);
- } else if (ns->ns_client == LDLM_NAMESPACE_CLIENT &&
+ } else if (ns_is_client(ns) &&
!lock->l_readers && !lock->l_writers &&
- !(lock->l_flags & LDLM_FL_NO_LRU)) {
+ !(lock->l_flags & LDLM_FL_BL_AST)) {
/* If this is a client-side namespace and this was the last
* reference, put it on the LRU. */
- LASSERT(list_empty(&lock->l_lru));
- LASSERT(ns->ns_nr_unused >= 0);
- lock->l_last_used = cfs_time_current();
- spin_lock(&ns->ns_unused_lock);
- list_add_tail(&lock->l_lru, &ns->ns_unused_list);
- ns->ns_nr_unused++;
- spin_unlock(&ns->ns_unused_lock);
+ ldlm_lock_add_to_lru(lock);
unlock_res_and_lock(lock);
- /* Call ldlm_cancel_lru() only if EARLY_CANCEL is not supported
- * by the server, otherwise, it is done on enqueue. */
- if (!exp_connect_cancelset(lock->l_conn_export))
- ldlm_cancel_lru(ns, LDLM_ASYNC);
+
+ if (lock->l_flags & LDLM_FL_FAIL_LOC)
+ OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE);
+
+ /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE
+ * are not supported by the server, otherwise, it is done on
+ * enqueue. */
+ if (!exp_connect_cancelset(lock->l_conn_export) &&
+ !ns_connect_lru_resize(ns))
+ ldlm_cancel_lru(ns, 0, LDLM_ASYNC, 0);
} else {
unlock_res_and_lock(lock);
}
- LDLM_LOCK_PUT(lock); /* matches the ldlm_lock_get in addref */
-
EXIT;
}
LDLM_LOCK_PUT(lock);
}
+struct sl_insert_point {
+ struct list_head *res_link;
+ struct list_head *mode_link;
+ struct list_head *policy_link;
+};
+
/*
* search_granted_lock
*
* Parameters:
* queue [input]: the granted list where search acts on;
* req [input]: the lock whose position to be located;
- * lockp [output]: the position where the lock should be inserted before, or
- * NULL indicating @req should be appended to @queue.
- * Return Values:
- * Bit-masks combination of following values indicating in which way the
- * lock need to be inserted.
- * - LDLM_JOIN_NONE: noting about skip list needs to be fixed;
- * - LDLM_MODE_JOIN_RIGHT: @req needs join right becoming the head of a
- * mode group;
- * - LDLM_POLICY_JOIN_RIGHT: @req needs join right becoming the head of
- * a policy group.
+ * prev [output]: positions within 3 lists to insert @req to
+ * Return Value:
+ * filled @prev
* NOTE: called by
* - ldlm_grant_lock_with_skiplist
*/
-static int search_granted_lock(struct list_head *queue,
- struct ldlm_lock *req,
- struct ldlm_lock **lockp)
+static void search_granted_lock(struct list_head *queue,
+ struct ldlm_lock *req,
+ struct sl_insert_point *prev)
{
- struct list_head *tmp, *tmp_tail;
- struct ldlm_lock *lock, *mode_head_lock;
- int rc = LDLM_JOIN_NONE;
+ struct list_head *tmp;
+ struct ldlm_lock *lock, *mode_end, *policy_end;
ENTRY;
list_for_each(tmp, queue) {
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
+ mode_end = list_entry(lock->l_sl_mode.prev, struct ldlm_lock,
+ l_sl_mode);
+
if (lock->l_req_mode != req->l_req_mode) {
- if (LDLM_SL_HEAD(&lock->l_sl_mode))
- tmp = &list_entry(lock->l_sl_mode.next,
- struct ldlm_lock,
- l_sl_mode)->l_res_link;
+ /* jump to last lock of mode group */
+ tmp = &mode_end->l_res_link;
continue;
}
-
- /* found the same mode group */
+
+ /* suitable mode group is found */
if (lock->l_resource->lr_type == LDLM_PLAIN) {
- *lockp = lock;
- rc = LDLM_MODE_JOIN_RIGHT;
- GOTO(out, rc);
+ /* insert point is last lock of the mode group */
+ prev->res_link = &mode_end->l_res_link;
+ prev->mode_link = &mode_end->l_sl_mode;
+ prev->policy_link = &req->l_sl_policy;
+ EXIT;
+ return;
} else if (lock->l_resource->lr_type == LDLM_IBITS) {
- tmp_tail = tmp;
- if (LDLM_SL_HEAD(&lock->l_sl_mode))
- tmp_tail = &list_entry(lock->l_sl_mode.next,
- struct ldlm_lock,
- l_sl_mode)->l_res_link;
- mode_head_lock = lock;
for (;;) {
+ policy_end = list_entry(lock->l_sl_policy.prev,
+ struct ldlm_lock,
+ l_sl_policy);
+
if (lock->l_policy_data.l_inodebits.bits ==
req->l_policy_data.l_inodebits.bits) {
- /* matched policy lock is found */
- *lockp = lock;
- rc |= LDLM_POLICY_JOIN_RIGHT;
-
- /* if the policy group head is also a
- * mode group head or a single mode
- * group lock */
- if (LDLM_SL_HEAD(&lock->l_sl_mode) ||
- (tmp == tmp_tail &&
- LDLM_SL_EMPTY(&lock->l_sl_mode)))
- rc |= LDLM_MODE_JOIN_RIGHT;
- GOTO(out, rc);
+ /* insert point is last lock of
+ * the policy group */
+ prev->res_link =
+ &policy_end->l_res_link;
+ prev->mode_link =
+ &policy_end->l_sl_mode;
+ prev->policy_link =
+ &policy_end->l_sl_policy;
+ EXIT;
+ return;
}
- if (LDLM_SL_HEAD(&lock->l_sl_policy))
- tmp = &list_entry(lock->l_sl_policy.next,
- struct ldlm_lock,
- l_sl_policy)->l_res_link;
-
- if (tmp == tmp_tail)
+ if (policy_end == mode_end)
+ /* done with mode group */
break;
- else
- tmp = tmp->next;
- lock = list_entry(tmp, struct ldlm_lock,
- l_res_link);
- } /* for all locks in the matched mode group */
- /* no matched policy group is found, insert before
- * the mode group head lock */
- *lockp = mode_head_lock;
- rc = LDLM_MODE_JOIN_RIGHT;
- GOTO(out, rc);
+ /* go to next policy group within mode group */
+ tmp = policy_end->l_res_link.next;
+ lock = list_entry(tmp, struct ldlm_lock,
+ l_res_link);
+ } /* loop over policy groups within the mode group */
+
+ /* insert point is last lock of the mode group,
+ * new policy group is started */
+ prev->res_link = &mode_end->l_res_link;
+ prev->mode_link = &mode_end->l_sl_mode;
+ prev->policy_link = &req->l_sl_policy;
+ EXIT;
+ return;
} else {
- LDLM_ERROR(lock, "is not LDLM_PLAIN or LDLM_IBITS lock");
+ LDLM_ERROR(lock,"is not LDLM_PLAIN or LDLM_IBITS lock");
LBUG();
}
}
- /* no matched mode group is found, append to the end */
- *lockp = NULL;
- rc = LDLM_JOIN_NONE;
+ /* insert point is last lock on the queue,
+ * new mode group and new policy group are started */
+ prev->res_link = queue->prev;
+ prev->mode_link = &req->l_sl_mode;
+ prev->policy_link = &req->l_sl_policy;
EXIT;
-out:
- return rc;
+ return;
}
-static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
- struct ldlm_lock *lockp,
- int join)
+static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
+ struct sl_insert_point *prev)
{
struct ldlm_resource *res = lock->l_resource;
ENTRY;
- LASSERT(lockp || join == LDLM_JOIN_NONE);
-
check_res_locked(res);
- ldlm_resource_dump(D_OTHER, res);
+ ldlm_resource_dump(D_INFO, res);
CDEBUG(D_OTHER, "About to add this lock:\n");
ldlm_lock_dump(D_OTHER, lock, 0);
}
LASSERT(list_empty(&lock->l_res_link));
+ LASSERT(list_empty(&lock->l_sl_mode));
+ LASSERT(list_empty(&lock->l_sl_policy));
- if (!lockp)
- list_add_tail(&lock->l_res_link, &lock->l_resource->lr_granted);
- else if ((join & LDLM_MODE_JOIN_LEFT) || (join & LDLM_POLICY_JOIN_LEFT))
- list_add(&lock->l_res_link, &lockp->l_res_link);
- else
- list_add_tail(&lock->l_res_link, &lockp->l_res_link);
-
- /* fix skip lists */
- if (join & LDLM_MODE_JOIN_RIGHT) {
- LASSERT(! LDLM_SL_TAIL(&lockp->l_sl_mode));
- if (LDLM_SL_EMPTY(&lockp->l_sl_mode)) {
- lock->l_sl_mode.next = &lockp->l_sl_mode;
- lockp->l_sl_mode.prev = &lock->l_sl_mode;
- } else if (LDLM_SL_HEAD(&lockp->l_sl_mode)) {
- lock->l_sl_mode.next = lockp->l_sl_mode.next;
- lockp->l_sl_mode.next = NULL;
- lock->l_sl_mode.next->prev = &lock->l_sl_mode;
- }
- } else if (join & LDLM_MODE_JOIN_LEFT) {
- LASSERT(! LDLM_SL_HEAD(&lockp->l_sl_mode));
- if (LDLM_SL_EMPTY(&lockp->l_sl_mode)) {
- lock->l_sl_mode.prev = &lockp->l_sl_mode;
- lockp->l_sl_mode.next = &lock->l_sl_mode;
- } else if (LDLM_SL_TAIL(&lockp->l_sl_mode)) {
- lock->l_sl_mode.prev = lockp->l_sl_mode.prev;
- lockp->l_sl_mode.prev = NULL;
- lock->l_sl_mode.prev->next = &lock->l_sl_mode;
- }
- }
-
- if (join & LDLM_POLICY_JOIN_RIGHT) {
- LASSERT(! LDLM_SL_TAIL(&lockp->l_sl_policy));
- if (LDLM_SL_EMPTY(&lockp->l_sl_policy)) {
- lock->l_sl_policy.next = &lockp->l_sl_policy;
- lockp->l_sl_policy.prev = &lock->l_sl_policy;
- } else if (LDLM_SL_HEAD(&lockp->l_sl_policy)) {
- lock->l_sl_policy.next = lockp->l_sl_policy.next;
- lockp->l_sl_policy.next = NULL;
- lock->l_sl_policy.next->prev = &lock->l_sl_policy;
- }
- } else if (join & LDLM_POLICY_JOIN_LEFT) {
- LASSERT(! LDLM_SL_HEAD(&lockp->l_sl_policy));
- if (LDLM_SL_EMPTY(&lockp->l_sl_policy)) {
- lock->l_sl_policy.prev = &lockp->l_sl_policy;
- lockp->l_sl_policy.next = &lock->l_sl_policy;
- } else if (LDLM_SL_TAIL(&lockp->l_sl_policy)) {
- lock->l_sl_policy.prev = lockp->l_sl_policy.prev;
- lockp->l_sl_policy.prev = NULL;
- lock->l_sl_policy.prev->next = &lock->l_sl_policy;
- }
- }
+ list_add(&lock->l_res_link, prev->res_link);
+ list_add(&lock->l_sl_mode, prev->mode_link);
+ list_add(&lock->l_sl_policy, prev->policy_link);
EXIT;
}
static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
{
- int join = LDLM_JOIN_NONE;
- struct ldlm_lock *lockp = NULL;
+ struct sl_insert_point prev;
ENTRY;
LASSERT(lock->l_req_mode == lock->l_granted_mode);
- join = search_granted_lock(&lock->l_resource->lr_granted, lock, &lockp);
- ldlm_granted_list_add_lock(lock, lockp, join);
+ search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
+ ldlm_granted_list_add_lock(lock, &prev);
EXIT;
}
lock->l_granted_mode = lock->l_req_mode;
if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS)
ldlm_grant_lock_with_skiplist(lock);
+ else if (res->lr_type == LDLM_EXTENT)
+ ldlm_extent_add_lock(res, lock);
else
ldlm_resource_add_lock(res, &res->lr_granted, lock);
if (work_list && lock->l_completion_ast != NULL)
ldlm_add_ast_work_item(lock, NULL, work_list);
+ ldlm_pool_add(&res->lr_namespace->ns_pool, lock);
EXIT;
}
/* returns a referenced lock or NULL. See the flag descriptions below, in the
* comment above ldlm_lock_match */
-static struct ldlm_lock *search_queue(struct list_head *queue, ldlm_mode_t mode,
+static struct ldlm_lock *search_queue(struct list_head *queue,
+ ldlm_mode_t *mode,
ldlm_policy_data_t *policy,
- struct ldlm_lock *old_lock, int flags)
+ struct ldlm_lock *old_lock,
+ int flags, int unref)
{
struct ldlm_lock *lock;
struct list_head *tmp;
list_for_each(tmp, queue) {
+ ldlm_mode_t match;
+
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
if (lock == old_lock)
if (lock->l_flags & LDLM_FL_CBPENDING &&
!(flags & LDLM_FL_CBPENDING))
continue;
- if (lock->l_flags & LDLM_FL_CBPENDING &&
+ if (!unref && lock->l_flags & LDLM_FL_CBPENDING &&
lock->l_readers == 0 && lock->l_writers == 0)
continue;
- if (!(lock->l_req_mode & mode))
+ if (!(lock->l_req_mode & *mode))
continue;
+ match = lock->l_req_mode;
if (lock->l_resource->lr_type == LDLM_EXTENT &&
(lock->l_policy_data.l_extent.start >
lock->l_policy_data.l_extent.end < policy->l_extent.end))
continue;
- if (unlikely(mode == LCK_GROUP) &&
+ if (unlikely(match == LCK_GROUP) &&
lock->l_resource->lr_type == LDLM_EXTENT &&
lock->l_policy_data.l_extent.gid != policy->l_extent.gid)
continue;
policy->l_inodebits.bits))
continue;
- if (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED))
+ if (!unref &&
+ (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED)))
continue;
if ((flags & LDLM_FL_LOCAL_ONLY) &&
!(lock->l_flags & LDLM_FL_LOCAL))
continue;
- if (flags & LDLM_FL_TEST_LOCK)
+ if (flags & LDLM_FL_TEST_LOCK) {
LDLM_LOCK_GET(lock);
- else
- ldlm_lock_addref_internal_nolock(lock, mode);
+ ldlm_lock_touch_in_lru(lock);
+ } else {
+ ldlm_lock_addref_internal_nolock(lock, match);
+ }
+ *mode = match;
return lock;
}
return NULL;
}
-void ldlm_lock_allow_match(struct ldlm_lock *lock)
+void ldlm_lock_allow_match_locked(struct ldlm_lock *lock)
{
- lock_res_and_lock(lock);
lock->l_flags |= LDLM_FL_LVB_READY;
cfs_waitq_signal(&lock->l_waitq);
+}
+
+void ldlm_lock_allow_match(struct ldlm_lock *lock)
+{
+ lock_res_and_lock(lock);
+ ldlm_lock_allow_match_locked(lock);
unlock_res_and_lock(lock);
}
* caller code unchanged), the context failure will be discovered by caller
* sometime later.
*/
-int ldlm_lock_match(struct ldlm_namespace *ns, int flags,
- const struct ldlm_res_id *res_id, ldlm_type_t type,
- ldlm_policy_data_t *policy, ldlm_mode_t mode,
- struct lustre_handle *lockh)
+ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
+ const struct ldlm_res_id *res_id, ldlm_type_t type,
+ ldlm_policy_data_t *policy, ldlm_mode_t mode,
+ struct lustre_handle *lockh, int unref)
{
struct ldlm_resource *res;
struct ldlm_lock *lock, *old_lock = NULL;
RETURN(0);
}
+ LDLM_RESOURCE_ADDREF(res);
lock_res(res);
- lock = search_queue(&res->lr_granted, mode, policy, old_lock, flags);
+ lock = search_queue(&res->lr_granted, &mode, policy, old_lock,
+ flags, unref);
if (lock != NULL)
GOTO(out, rc = 1);
if (flags & LDLM_FL_BLOCK_GRANTED)
GOTO(out, rc = 0);
- lock = search_queue(&res->lr_converting, mode, policy, old_lock, flags);
+ lock = search_queue(&res->lr_converting, &mode, policy, old_lock,
+ flags, unref);
if (lock != NULL)
GOTO(out, rc = 1);
- lock = search_queue(&res->lr_waiting, mode, policy, old_lock, flags);
+ lock = search_queue(&res->lr_waiting, &mode, policy, old_lock,
+ flags, unref);
if (lock != NULL)
GOTO(out, rc = 1);
EXIT;
out:
unlock_res(res);
+ LDLM_RESOURCE_DELREF(res);
ldlm_resource_putref(res);
if (lock) {
NULL);
if (err) {
if (flags & LDLM_FL_TEST_LOCK)
- LDLM_LOCK_PUT(lock);
+ LDLM_LOCK_RELEASE(lock);
else
- ldlm_lock_decref_internal(lock, mode);
+ ldlm_lock_decref_internal(lock,
+ mode);
rc = 0;
goto out2;
}
}
- lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout), NULL,
- LWI_ON_SIGNAL_NOOP, NULL);
+ lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout),
+ NULL, LWI_ON_SIGNAL_NOOP, NULL);
/* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */
l_wait_event(lock->l_waitq,
}
if (flags & LDLM_FL_TEST_LOCK)
- LDLM_LOCK_PUT(lock);
+ LDLM_LOCK_RELEASE(lock);
} else if (!(flags & LDLM_FL_TEST_LOCK)) {/*less verbose for test-only*/
LDLM_DEBUG_NOLOCK("not matched ns %p type %u mode %u res "
if (old_lock)
LDLM_LOCK_PUT(old_lock);
- return rc;
+ return rc ? mode : 0;
}
/* Returns a referenced lock */
const struct ldlm_res_id *res_id,
ldlm_type_t type,
ldlm_mode_t mode,
- ldlm_blocking_callback blocking,
- ldlm_completion_callback completion,
- ldlm_glimpse_callback glimpse,
+ const struct ldlm_callback_suite *cbs,
void *data, __u32 lvb_len)
{
struct ldlm_lock *lock;
lock->l_req_mode = mode;
lock->l_ast_data = data;
- lock->l_blocking_ast = blocking;
- lock->l_completion_ast = completion;
- lock->l_glimpse_ast = glimpse;
lock->l_pid = cfs_curproc_pid();
+ if (cbs) {
+ lock->l_blocking_ast = cbs->lcs_blocking;
+ lock->l_completion_ast = cbs->lcs_completion;
+ lock->l_glimpse_ast = cbs->lcs_glimpse;
+ lock->l_weigh_ast = cbs->lcs_weigh;
+ }
+
+ lock->l_tree_node = NULL;
+ /* if this is the extent lock, allocate the interval tree node */
+ if (type == LDLM_EXTENT) {
+ if (ldlm_interval_alloc(lock) == NULL)
+ GOTO(out, 0);
+ }
if (lvb_len) {
lock->l_lvb_len = lvb_len;
OBD_ALLOC(lock->l_lvb_data, lvb_len);
- if (lock->l_lvb_data == NULL) {
- OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
- RETURN(NULL);
- }
+ if (lock->l_lvb_data == NULL)
+ GOTO(out, 0);
}
RETURN(lock);
+
+out:
+ if (lock->l_lvb_data)
+ OBD_FREE(lock->l_lvb_data, lvb_len);
+ ldlm_interval_free(ldlm_interval_detach(lock));
+ OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock));
+ return NULL;
}
ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns,
{
struct ldlm_lock *lock = *lockp;
struct ldlm_resource *res = lock->l_resource;
- int local = res->lr_namespace->ns_client;
+ int local = ns_is_client(res->lr_namespace);
ldlm_processing_policy policy;
ldlm_error_t rc = ELDLM_OK;
+ struct ldlm_interval *node = NULL;
ENTRY;
- do_gettimeofday(&lock->l_enqueued_time);
+ lock->l_last_activity = cfs_time_current_sec();
/* policies are not executed on the client or during replay */
if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT
&& !local && ns->ns_policy) {
* work here is done. */
if (lock != *lockp) {
ldlm_lock_destroy(lock);
- LDLM_LOCK_PUT(lock);
+ LDLM_LOCK_RELEASE(lock);
}
*flags |= LDLM_FL_LOCK_CHANGED;
RETURN(0);
}
}
+ /* For a replaying lock, it might be already in granted list. So
+ * unlinking the lock will cause the interval node to be freed, we
+ * have to allocate the interval node early otherwise we can't regrant
+ * this lock in the future. - jay */
+ if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT)
+ OBD_SLAB_ALLOC(node, ldlm_interval_slab, CFS_ALLOC_IO,
+ sizeof(*node));
+
lock_res_and_lock(lock);
if (local && lock->l_req_mode == lock->l_granted_mode) {
- /* The server returned a blocked lock, but it was granted before
- * we got a chance to actually enqueue it. We don't need to do
- * anything else. */
+ /* The server returned a blocked lock, but it was granted
+ * before we got a chance to actually enqueue it. We don't
+ * need to do anything else. */
*flags &= ~(LDLM_FL_BLOCK_GRANTED |
LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT);
GOTO(out, ELDLM_OK);
}
+ ldlm_resource_unlink_lock(lock);
+ if (res->lr_type == LDLM_EXTENT && lock->l_tree_node == NULL) {
+ if (node == NULL) {
+ ldlm_lock_destroy_nolock(lock);
+ GOTO(out, rc = -ENOMEM);
+ }
+
+ CFS_INIT_LIST_HEAD(&node->li_group);
+ ldlm_interval_attach(node, lock);
+ node = NULL;
+ }
+
/* Some flags from the enqueue want to make it into the AST, via the
* lock's l_flags. */
lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA;
*
* FIXME (bug 268): Detect obvious lies by checking compatibility in
* granted/converting queues. */
- ldlm_resource_unlink_lock(lock);
if (local) {
if (*flags & LDLM_FL_BLOCK_CONV)
ldlm_resource_add_lock(res, &res->lr_converting, lock);
GOTO(out, rc);
out:
unlock_res_and_lock(lock);
+ if (node)
+ OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
return rc;
}
RETURN(rc);
}
-int ldlm_run_bl_ast_work(struct list_head *rpc_list)
+/* Helper function for ldlm_run_ast_work().
+ *
+ * Send an existing rpc set specified by @arg->set and then
+ * destroy it. Create new one if @do_create flag is set. */
+static void
+ldlm_send_and_maybe_create_set(struct ldlm_cb_set_arg *arg, int do_create)
{
- struct list_head *tmp, *pos;
- struct ldlm_lock_desc d;
- int rc = 0, retval = 0;
ENTRY;
- list_for_each_safe(tmp, pos, rpc_list) {
- struct ldlm_lock *lock =
- list_entry(tmp, struct ldlm_lock, l_bl_ast);
+ ptlrpc_set_wait(arg->set);
+ if (arg->type == LDLM_BL_CALLBACK)
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_GLIMPSE, 2);
+ ptlrpc_set_destroy(arg->set);
- /* nobody should touch l_bl_ast */
- lock_res_and_lock(lock);
- list_del_init(&lock->l_bl_ast);
+ if (do_create)
+ arg->set = ptlrpc_prep_set();
- LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
- LASSERT(lock->l_bl_ast_run == 0);
- LASSERT(lock->l_blocking_lock);
- lock->l_bl_ast_run++;
- unlock_res_and_lock(lock);
+ EXIT;
+}
- ldlm_lock2desc(lock->l_blocking_lock, &d);
+static int
+ldlm_work_bl_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
+{
+ struct ldlm_lock_desc d;
+ struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_bl_ast);
+ ENTRY;
- LDLM_LOCK_PUT(lock->l_blocking_lock);
- lock->l_blocking_lock = NULL;
- rc = lock->l_blocking_ast(lock, &d, NULL, LDLM_CB_BLOCKING);
+ /* nobody should touch l_bl_ast */
+ lock_res_and_lock(lock);
+ list_del_init(&lock->l_bl_ast);
- if (rc == -ERESTART)
- retval = rc;
- else if (rc)
- CDEBUG(D_DLMTRACE, "Failed AST - should clean & "
- "disconnect client\n");
- LDLM_LOCK_PUT(lock);
- }
- RETURN(retval);
+ LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
+ LASSERT(lock->l_bl_ast_run == 0);
+ LASSERT(lock->l_blocking_lock);
+ lock->l_bl_ast_run++;
+ unlock_res_and_lock(lock);
+
+ ldlm_lock2desc(lock->l_blocking_lock, &d);
+
+ lock->l_blocking_ast(lock, &d, (void *)arg,
+ LDLM_CB_BLOCKING);
+ LDLM_LOCK_RELEASE(lock->l_blocking_lock);
+ lock->l_blocking_lock = NULL;
+ LDLM_LOCK_RELEASE(lock);
+
+ RETURN(1);
}
-int ldlm_run_cp_ast_work(struct list_head *rpc_list)
+static int
+ldlm_work_cp_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
{
- struct list_head *tmp, *pos;
- int rc = 0, retval = 0;
+ struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_cp_ast);
+ ldlm_completion_callback completion_callback;
+ int rc = 0;
ENTRY;
/* It's possible to receive a completion AST before we've set
* reader/writer reference, which we won't do until we get the
* reply and finish enqueueing. */
- list_for_each_safe(tmp, pos, rpc_list) {
- struct ldlm_lock *lock =
- list_entry(tmp, struct ldlm_lock, l_cp_ast);
+ /* nobody should touch l_cp_ast */
+ lock_res_and_lock(lock);
+ list_del_init(&lock->l_cp_ast);
+ LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
+ /* save l_completion_ast since it can be changed by
+ * mds_intent_policy(), see bug 14225 */
+ completion_callback = lock->l_completion_ast;
+ lock->l_flags &= ~LDLM_FL_CP_REQD;
+ unlock_res_and_lock(lock);
- /* nobody should touch l_cp_ast */
- lock_res_and_lock(lock);
- list_del_init(&lock->l_cp_ast);
- LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
- lock->l_flags &= ~LDLM_FL_CP_REQD;
- unlock_res_and_lock(lock);
+ if (completion_callback != NULL) {
+ completion_callback(lock, 0, (void *)arg);
+ rc = 1;
+ }
+ LDLM_LOCK_RELEASE(lock);
- if (lock->l_completion_ast != NULL)
- rc = lock->l_completion_ast(lock, 0, 0);
- if (rc == -ERESTART)
- retval = rc;
- else if (rc)
- CDEBUG(D_DLMTRACE, "Failed AST - should clean & "
- "disconnect client\n");
- LDLM_LOCK_PUT(lock);
+ RETURN(rc);
+}
+
+static int
+ldlm_work_revoke_ast_lock(struct list_head *tmp, struct ldlm_cb_set_arg *arg)
+{
+ struct ldlm_lock_desc desc;
+ struct ldlm_lock *lock = list_entry(tmp, struct ldlm_lock, l_rk_ast);
+ ENTRY;
+
+ list_del_init(&lock->l_rk_ast);
+
+ /* the desc just pretend to exclusive */
+ ldlm_lock2desc(lock, &desc);
+ desc.l_req_mode = LCK_EX;
+ desc.l_granted_mode = 0;
+
+ lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING);
+ LDLM_LOCK_RELEASE(lock);
+
+ RETURN(1);
+}
+
+int ldlm_run_ast_work(struct list_head *rpc_list, ldlm_desc_ast_t ast_type)
+{
+ struct ldlm_cb_set_arg arg;
+ struct list_head *tmp, *pos;
+ int (*work_ast_lock)(struct list_head *tmp,struct ldlm_cb_set_arg *arg);
+ int ast_count;
+ ENTRY;
+
+ arg.set = ptlrpc_prep_set();
+ atomic_set(&arg.restart, 0);
+ switch (ast_type) {
+ case LDLM_WORK_BL_AST:
+ arg.type = LDLM_BL_CALLBACK;
+ work_ast_lock = ldlm_work_bl_ast_lock;
+ break;
+ case LDLM_WORK_CP_AST:
+ arg.type = LDLM_CP_CALLBACK;
+ work_ast_lock = ldlm_work_cp_ast_lock;
+ break;
+ case LDLM_WORK_REVOKE_AST:
+ arg.type = LDLM_BL_CALLBACK;
+ work_ast_lock = ldlm_work_revoke_ast_lock;
+ break;
+ default:
+ LBUG();
+ }
+
+ ast_count = 0;
+ list_for_each_safe(tmp, pos, rpc_list) {
+ ast_count += work_ast_lock(tmp, &arg);
+
+ /* Send the request set if it exceeds the PARALLEL_AST_LIMIT,
+ * and create a new set for requests that remained in
+ * @rpc_list */
+ if (unlikely(ast_count == PARALLEL_AST_LIMIT)) {
+ ldlm_send_and_maybe_create_set(&arg, 1);
+ ast_count = 0;
+ }
}
- RETURN(retval);
+
+ if (ast_count > 0)
+ ldlm_send_and_maybe_create_set(&arg, 0);
+ else
+ /* In case when number of ASTs is multiply of
+ * PARALLEL_AST_LIMIT or @rpc_list was initially empty,
+ * @arg.set must be destroyed here, otherwise we get
+ * write memory leaking. */
+ ptlrpc_set_destroy(arg.set);
+
+ RETURN(atomic_read(&arg.restart) ? -ERESTART : 0);
}
static int reprocess_one_queue(struct ldlm_resource *res, void *closure)
ldlm_resource_getref(res);
spin_unlock(&ns->ns_hash_lock);
+ LDLM_RESOURCE_ADDREF(res);
rc = reprocess_one_queue(res, NULL);
+ LDLM_RESOURCE_DELREF(res);
spin_lock(&ns->ns_hash_lock);
tmp = tmp->next;
ldlm_resource_putref_locked(res);
void ldlm_reprocess_all(struct ldlm_resource *res)
{
- struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
+ CFS_LIST_HEAD(rpc_list);
int rc;
ENTRY;
/* Local lock trees don't get reprocessed. */
- if (res->lr_namespace->ns_client) {
+ if (ns_is_client(res->lr_namespace)) {
EXIT;
return;
}
ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list);
unlock_res(res);
- rc = ldlm_run_cp_ast_work(&rpc_list);
+ rc = ldlm_run_ast_work(&rpc_list, LDLM_WORK_CP_AST);
if (rc == -ERESTART) {
LASSERT(list_empty(&rpc_list));
goto restart;
LDLM_DEBUG(lock, "no blocking ast");
}
}
+ lock->l_flags |= LDLM_FL_BL_DONE;
}
void ldlm_unlink_lock_skiplist(struct ldlm_lock *req)
{
- struct ldlm_lock *lock;
-
if (req->l_resource->lr_type != LDLM_PLAIN &&
req->l_resource->lr_type != LDLM_IBITS)
return;
-
- if (LDLM_SL_HEAD(&req->l_sl_mode)) {
- lock = list_entry(req->l_res_link.next, struct ldlm_lock,
- l_res_link);
- if (req->l_sl_mode.next == &lock->l_sl_mode) {
- lock->l_sl_mode.prev = NULL;
- } else {
- lock->l_sl_mode.next = req->l_sl_mode.next;
- lock->l_sl_mode.next->prev = &lock->l_sl_mode;
- }
- req->l_sl_mode.next = NULL;
- } else if (LDLM_SL_TAIL(&req->l_sl_mode)) {
- lock = list_entry(req->l_res_link.prev, struct ldlm_lock,
- l_res_link);
- if (req->l_sl_mode.prev == &lock->l_sl_mode) {
- lock->l_sl_mode.next = NULL;
- } else {
- lock->l_sl_mode.prev = req->l_sl_mode.prev;
- lock->l_sl_mode.prev->next = &lock->l_sl_mode;
- }
- req->l_sl_mode.prev = NULL;
- }
- if (LDLM_SL_HEAD(&req->l_sl_policy)) {
- lock = list_entry(req->l_res_link.next, struct ldlm_lock,
- l_res_link);
- if (req->l_sl_policy.next == &lock->l_sl_policy) {
- lock->l_sl_policy.prev = NULL;
- } else {
- lock->l_sl_policy.next = req->l_sl_policy.next;
- lock->l_sl_policy.next->prev = &lock->l_sl_policy;
- }
- req->l_sl_policy.next = NULL;
- } else if (LDLM_SL_TAIL(&req->l_sl_policy)) {
- lock = list_entry(req->l_res_link.prev, struct ldlm_lock,
- l_res_link);
- if (req->l_sl_policy.prev == &lock->l_sl_policy) {
- lock->l_sl_policy.next = NULL;
- } else {
- lock->l_sl_policy.prev = req->l_sl_policy.prev;
- lock->l_sl_policy.prev->next = &lock->l_sl_policy;
- }
- req->l_sl_policy.prev = NULL;
- }
+ list_del_init(&req->l_sl_policy);
+ list_del_init(&req->l_sl_mode);
}
void ldlm_lock_cancel(struct ldlm_lock *lock)
/* Yes, second time, just in case it was added again while we were
running with no res lock in ldlm_cancel_callback */
- ldlm_del_waiting_lock(lock);
+ ldlm_del_waiting_lock(lock);
ldlm_resource_unlink_lock(lock);
ldlm_lock_destroy_nolock(lock);
+
+ if (lock->l_granted_mode == lock->l_req_mode)
+ ldlm_pool_del(&ns->ns_pool, lock);
+
+ /* Make sure we will not be called again for same lock what is possible
+ * if not to zero out lock->l_granted_mode */
+ lock->l_granted_mode = LCK_MINMODE;
unlock_res_and_lock(lock);
EXIT;
RETURN(0);
}
-void ldlm_cancel_locks_for_export(struct obd_export *exp)
+void ldlm_cancel_locks_for_export_cb(void *obj, void *data)
{
- struct ldlm_lock *lock;
+ struct obd_export *exp = data;
+ struct ldlm_lock *lock = obj;
struct ldlm_resource *res;
- spin_lock(&exp->exp_ldlm_data.led_lock);
- while(!list_empty(&exp->exp_ldlm_data.led_held_locks)) {
- lock = list_entry(exp->exp_ldlm_data.led_held_locks.next,
- struct ldlm_lock, l_export_chain);
- res = ldlm_resource_getref(lock->l_resource);
- LDLM_LOCK_GET(lock);
- spin_unlock(&exp->exp_ldlm_data.led_lock);
+ res = ldlm_resource_getref(lock->l_resource);
+ LDLM_LOCK_GET(lock);
- LDLM_DEBUG(lock, "export %p", exp);
- ldlm_lock_cancel(lock);
- ldlm_reprocess_all(res);
+ LDLM_DEBUG(lock, "export %p", exp);
+ ldlm_res_lvbo_update(res, NULL, 0, 1);
+ ldlm_lock_cancel(lock);
+ ldlm_reprocess_all(res);
+ ldlm_resource_putref(res);
+ LDLM_LOCK_RELEASE(lock);
+}
- ldlm_resource_putref(res);
- LDLM_LOCK_PUT(lock);
- spin_lock(&exp->exp_ldlm_data.led_lock);
- }
- spin_unlock(&exp->exp_ldlm_data.led_lock);
+void ldlm_cancel_locks_for_export(struct obd_export *exp)
+{
+ lustre_hash_for_each_empty(exp->exp_lock_hash,
+ ldlm_cancel_locks_for_export_cb, exp);
+}
+
+/**
+ * Downgrade an exclusive lock.
+ *
+ * A fast variant of ldlm_lock_convert for convertion of exclusive
+ * locks. The convertion is always successful.
+ *
+ * \param lock A lock to convert
+ * \param new_mode new lock mode
+ */
+void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode)
+{
+ struct ldlm_namespace *ns;
+ ENTRY;
+
+ LASSERT(lock->l_granted_mode & (LCK_PW | LCK_EX));
+ LASSERT(new_mode == LCK_COS);
+
+ lock_res_and_lock(lock);
+ ldlm_resource_unlink_lock(lock);
+ /*
+ * Remove the lock from pool as it will be added again in
+ * ldlm_grant_lock() called below.
+ */
+ ns = lock->l_resource->lr_namespace;
+ ldlm_pool_del(&ns->ns_pool, lock);
+
+ lock->l_req_mode = new_mode;
+ ldlm_grant_lock(lock, NULL);
+ unlock_res_and_lock(lock);
+ ldlm_reprocess_all(lock->l_resource);
+
+ EXIT;
}
struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
- int *flags)
+ __u32 *flags)
{
- struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
+ CFS_LIST_HEAD(rpc_list);
struct ldlm_resource *res;
struct ldlm_namespace *ns;
int granted = 0;
int old_mode, rc;
- struct ldlm_lock *mark_lock = NULL;
- int join = LDLM_JOIN_NONE;
+ struct sl_insert_point prev;
ldlm_error_t err;
+ struct ldlm_interval *node;
ENTRY;
if (new_mode == lock->l_granted_mode) { // No changes? Just return.
RETURN(lock->l_resource);
}
- LASSERTF(new_mode == LCK_PW && lock->l_granted_mode == LCK_PR,
+ /* I can't check the type of lock here because the bitlock of lock
+ * is not held here, so do the allocation blindly. -jay */
+ OBD_SLAB_ALLOC(node, ldlm_interval_slab, CFS_ALLOC_IO, sizeof(*node));
+ if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */
+ RETURN(NULL);
+
+ LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
"new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
lock_res_and_lock(lock);
old_mode = lock->l_req_mode;
lock->l_req_mode = new_mode;
if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
- /* remember the lock position where the lock might be
- * added back to the granted list later and also
+ /* remember the lock position where the lock might be
+ * added back to the granted list later and also
* remember the join mode for skiplist fixing. */
- if (LDLM_SL_HEAD(&lock->l_sl_mode))
- join = LDLM_MODE_JOIN_RIGHT;
- else if (LDLM_SL_TAIL(&lock->l_sl_mode))
- join = LDLM_MODE_JOIN_LEFT;
- if (LDLM_SL_HEAD(&lock->l_sl_policy))
- join |= LDLM_POLICY_JOIN_RIGHT;
- else if (LDLM_SL_TAIL(&lock->l_sl_policy))
- join |= LDLM_POLICY_JOIN_LEFT;
-
- LASSERT(!((join & LDLM_MODE_JOIN_RIGHT) &&
- (join & LDLM_POLICY_JOIN_LEFT)));
- LASSERT(!((join & LDLM_MODE_JOIN_LEFT) &&
- (join & LDLM_POLICY_JOIN_RIGHT)));
-
- if ((join & LDLM_MODE_JOIN_LEFT) ||
- (join & LDLM_POLICY_JOIN_LEFT))
- mark_lock = list_entry(lock->l_res_link.prev,
- struct ldlm_lock, l_res_link);
- else if (lock->l_res_link.next != &res->lr_granted)
- mark_lock = list_entry(lock->l_res_link.next,
- struct ldlm_lock, l_res_link);
+ prev.res_link = lock->l_res_link.prev;
+ prev.mode_link = lock->l_sl_mode.prev;
+ prev.policy_link = lock->l_sl_policy.prev;
+ ldlm_resource_unlink_lock(lock);
+ } else {
+ ldlm_resource_unlink_lock(lock);
+ if (res->lr_type == LDLM_EXTENT) {
+ /* FIXME: ugly code, I have to attach the lock to a
+ * interval node again since perhaps it will be granted
+ * soon */
+ CFS_INIT_LIST_HEAD(&node->li_group);
+ ldlm_interval_attach(node, lock);
+ node = NULL;
+ }
}
- ldlm_resource_unlink_lock(lock);
+
+ /*
+ * Remove old lock from the pool before adding the lock with new
+ * mode below in ->policy()
+ */
+ ldlm_pool_del(&ns->ns_pool, lock);
/* If this is a local resource, put it on the appropriate list. */
- if (res->lr_namespace->ns_client) {
+ if (ns_is_client(res->lr_namespace)) {
if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
ldlm_resource_add_lock(res, &res->lr_converting, lock);
} else {
rc = policy(lock, &pflags, 0, &err, &rpc_list);
if (rc == LDLM_ITER_STOP) {
lock->l_req_mode = old_mode;
- ldlm_granted_list_add_lock(lock, mark_lock, join);
+ if (res->lr_type == LDLM_EXTENT)
+ ldlm_extent_add_lock(res, lock);
+ else
+ ldlm_granted_list_add_lock(lock, &prev);
+
res = NULL;
} else {
*flags |= LDLM_FL_BLOCK_GRANTED;
unlock_res_and_lock(lock);
if (granted)
- ldlm_run_cp_ast_work(&rpc_list);
+ ldlm_run_ast_work(&rpc_list, LDLM_WORK_CP_AST);
+ if (node)
+ OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
RETURN(res);
}
lock->l_resource->lr_name.name[1],
lock->l_resource->lr_name.name[2]);
CDEBUG(level, " Req mode: %s, grant mode: %s, rc: %u, read: %d, "
- "write: %d flags: %#x\n", ldlm_lockname[lock->l_req_mode],
+ "write: %d flags: "LPX64"\n", ldlm_lockname[lock->l_req_mode],
ldlm_lockname[lock->l_granted_mode],
atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers,
lock->l_flags);
{
va_list args;
cfs_debug_limit_state_t *cdls = data->msg_cdls;
-
+
va_start(args, fmt);
if (lock->l_resource == NULL) {
- libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
+ libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
data->msg_fn, data->msg_line, fmt, args,
" ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
- "res: \?\? rrc=\?\? type: \?\?\? flags: %x remote: "
- LPX64" expref: %d pid: %u\n", lock,
+ "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" remote: "
+ LPX64" expref: %d pid: %u timeout: %lu\n", lock,
lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
lock->l_readers, lock->l_writers,
ldlm_lockname[lock->l_granted_mode],
lock->l_flags, lock->l_remote_handle.cookie,
lock->l_export ?
atomic_read(&lock->l_export->exp_refcount) : -99,
- lock->l_pid);
+ lock->l_pid, lock->l_callback_timeout);
va_end(args);
return;
}
switch (lock->l_resource->lr_type) {
case LDLM_EXTENT:
- libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
+ libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
data->msg_fn, data->msg_line, fmt, args,
" ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
"res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64
- "] (req "LPU64"->"LPU64") flags: %x remote: "LPX64
- " expref: %d pid: %u\n",
+ "] (req "LPU64"->"LPU64") flags: "LPX64" remote: "LPX64
+ " expref: %d pid: %u timeout %lu\n",
lock->l_resource->lr_namespace->ns_name, lock,
lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
lock->l_readers, lock->l_writers,
lock->l_flags, lock->l_remote_handle.cookie,
lock->l_export ?
atomic_read(&lock->l_export->exp_refcount) : -99,
- lock->l_pid);
+ lock->l_pid, lock->l_callback_timeout);
break;
case LDLM_FLOCK:
- libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
+ libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
data->msg_fn, data->msg_line, fmt, args,
" ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
"res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d "
- "["LPU64"->"LPU64"] flags: %x remote: "LPX64
- " expref: %d pid: %u\n",
+ "["LPU64"->"LPU64"] flags: "LPX64" remote: "LPX64
+ " expref: %d pid: %u timeout: %lu\n",
lock->l_resource->lr_namespace->ns_name, lock,
lock->l_handle.h_cookie, atomic_read(&lock->l_refc),
lock->l_readers, lock->l_writers,
lock->l_flags, lock->l_remote_handle.cookie,
lock->l_export ?
atomic_read(&lock->l_export->exp_refcount) : -99,
- lock->l_pid);
+ lock->l_pid, lock->l_callback_timeout);
break;
case LDLM_IBITS:
- libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
+ libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
data->msg_fn, data->msg_line, fmt, args,
" ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
"res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s "
- "flags: %x remote: "LPX64" expref: %d "
- "pid %u\n",
+ "flags: "LPX64" remote: "LPX64" expref: %d "
+ "pid: %u timeout: %lu\n",
lock->l_resource->lr_namespace->ns_name,
lock, lock->l_handle.h_cookie,
atomic_read (&lock->l_refc),
lock->l_flags, lock->l_remote_handle.cookie,
lock->l_export ?
atomic_read(&lock->l_export->exp_refcount) : -99,
- lock->l_pid);
+ lock->l_pid, lock->l_callback_timeout);
break;
default:
- libcfs_debug_vmsg2(cdls, data->msg_subsys, level, data->msg_file,
+ libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file,
data->msg_fn, data->msg_line, fmt, args,
" ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "
- "res: "LPU64"/"LPU64" rrc: %d type: %s flags: %x "
- "remote: "LPX64" expref: %d pid: %u\n",
+ "res: "LPU64"/"LPU64" rrc: %d type: %s flags: "LPX64" "
+ "remote: "LPX64" expref: %d pid: %u timeout %lu\n",
lock->l_resource->lr_namespace->ns_name,
lock, lock->l_handle.h_cookie,
atomic_read (&lock->l_refc),
lock->l_flags, lock->l_remote_handle.cookie,
lock->l_export ?
atomic_read(&lock->l_export->exp_refcount) : -99,
- lock->l_pid);
+ lock->l_pid, lock->l_callback_timeout);
break;
}
va_end(args);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff