/* -*- 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: Phil Schwan <phil@clusterfs.com>
- * Author: Peter Braam <braam@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_resource.c
+ *
+ * Author: Phil Schwan <phil@clusterfs.com>
+ * Author: Peter Braam <braam@clusterfs.com>
*/
#define DEBUG_SUBSYSTEM S_LDLM
atomic_t ldlm_cli_namespace_nr = ATOMIC_INIT(0);
struct semaphore ldlm_srv_namespace_lock;
-struct list_head ldlm_srv_namespace_list =
- CFS_LIST_HEAD_INIT(ldlm_srv_namespace_list);
+CFS_LIST_HEAD(ldlm_srv_namespace_list);
struct semaphore ldlm_cli_namespace_lock;
-struct list_head ldlm_cli_namespace_list =
- CFS_LIST_HEAD_INIT(ldlm_cli_namespace_list);
+CFS_LIST_HEAD(ldlm_cli_namespace_list);
cfs_proc_dir_entry_t *ldlm_type_proc_dir = NULL;
cfs_proc_dir_entry_t *ldlm_ns_proc_dir = NULL;
if (copy_from_user(dummy, buffer, MAX_STRING_SIZE))
return -EFAULT;
- if (count == 6 && memcmp(dummy, "clear", 5) == 0) {
+ if (strncmp(dummy, "clear", 5) == 0) {
CDEBUG(D_DLMTRACE,
"dropping all unused locks from namespace %s\n",
ns->ns_name);
if (ns_connect_lru_resize(ns)) {
int canceled, unused = ns->ns_nr_unused;
-
+
/* Try to cancel all @ns_nr_unused locks. */
- canceled = ldlm_cancel_lru(ns, unused, LDLM_SYNC,
+ canceled = ldlm_cancel_lru(ns, unused, LDLM_SYNC,
LDLM_CANCEL_PASSED);
if (canceled < unused) {
CERROR("not all requested locks are canceled, "
- "requested: %d, canceled: %d\n", unused,
+ "requested: %d, canceled: %d\n", unused,
canceled);
return -EINVAL;
}
return -EINVAL;
}
lru_resize = (tmp == 0);
-
+
if (ns_connect_lru_resize(ns)) {
if (!lru_resize)
ns->ns_max_unused = (unsigned int)tmp;
-
+
if (tmp > ns->ns_nr_unused)
tmp = ns->ns_nr_unused;
tmp = ns->ns_nr_unused - tmp;
-
- CDEBUG(D_DLMTRACE, "changing namespace %s unused locks from %u to %u\n",
+
+ CDEBUG(D_DLMTRACE,
+ "changing namespace %s unused locks from %u to %u\n",
ns->ns_name, ns->ns_nr_unused, (unsigned int)tmp);
- ldlm_cancel_lru(ns, (unsigned int)tmp, LDLM_ASYNC, LDLM_CANCEL_PASSED);
-
+ ldlm_cancel_lru(ns, tmp, LDLM_ASYNC, LDLM_CANCEL_PASSED);
+
if (!lru_resize) {
- CDEBUG(D_DLMTRACE, "disable lru_resize for namespace %s\n",
+ CDEBUG(D_DLMTRACE,
+ "disable lru_resize for namespace %s\n",
ns->ns_name);
ns->ns_connect_flags &= ~OBD_CONNECT_LRU_RESIZE;
}
} else {
- CDEBUG(D_DLMTRACE, "changing namespace %s max_unused from %u to %u\n",
+ CDEBUG(D_DLMTRACE,
+ "changing namespace %s max_unused from %u to %u\n",
ns->ns_name, ns->ns_max_unused, (unsigned int)tmp);
ns->ns_max_unused = (unsigned int)tmp;
ldlm_cancel_lru(ns, 0, LDLM_ASYNC, LDLM_CANCEL_PASSED);
-
- /* Make sure that originally lru resize was supported before
+
+ /* Make sure that originally lru resize was supported before
* turning it on here. */
- if (lru_resize &&
+ if (lru_resize &&
(ns->ns_orig_connect_flags & OBD_CONNECT_LRU_RESIZE)) {
- CDEBUG(D_DLMTRACE, "enable lru_resize for namespace %s\n",
+ CDEBUG(D_DLMTRACE,
+ "enable lru_resize for namespace %s\n",
ns->ns_name);
ns->ns_connect_flags |= OBD_CONNECT_LRU_RESIZE;
}
lock_vars[0].write_fptr = lprocfs_wr_lru_size;
lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
- snprintf(lock_name, MAX_STRING_SIZE, "%s/shrink_thumb",
- ns->ns_name);
- lock_vars[0].data = ns;
- lock_vars[0].read_fptr = lprocfs_rd_uint;
- lock_vars[0].write_fptr = lprocfs_wr_uint;
- lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
-
snprintf(lock_name, MAX_STRING_SIZE, "%s/lru_max_age",
ns->ns_name);
lock_vars[0].data = &ns->ns_max_age;
lock_vars[0].write_fptr = lprocfs_wr_uint;
lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
+ snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_timeouts",
+ ns->ns_name);
+ lock_vars[0].data = &ns->ns_timeouts;
+ lock_vars[0].read_fptr = lprocfs_rd_uint;
+ lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
+
snprintf(lock_name, MAX_STRING_SIZE, "%s/max_nolock_bytes",
ns->ns_name);
lock_vars[0].data = &ns->ns_max_nolock_size;
#define ldlm_proc_namespace(ns) do {} while (0)
#endif /* LPROCFS */
-struct ldlm_namespace *ldlm_namespace_new(char *name, ldlm_side_t client,
- ldlm_appetite_t apt)
+struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
+ ldlm_side_t client, ldlm_appetite_t apt)
{
struct ldlm_namespace *ns = NULL;
struct list_head *bucket;
if (!ns->ns_hash)
GOTO(out_ns, NULL);
- ns->ns_shrink_thumb = LDLM_LOCK_SHRINK_THUMB;
ns->ns_appetite = apt;
+
+ LASSERT(obd != NULL);
+ ns->ns_obd = obd;
+
namelen = strlen(name);
OBD_ALLOC(ns->ns_name, namelen + 1);
if (!ns->ns_name)
strcpy(ns->ns_name, name);
CFS_INIT_LIST_HEAD(&ns->ns_root_list);
+ CFS_INIT_LIST_HEAD(&ns->ns_list_chain);
ns->ns_refcount = 0;
ns->ns_client = client;
spin_lock_init(&ns->ns_hash_lock);
ns->ns_max_unused = LDLM_DEFAULT_LRU_SIZE;
ns->ns_max_age = LDLM_DEFAULT_MAX_ALIVE;
ns->ns_ctime_age_limit = LDLM_CTIME_AGE_LIMIT;
+ ns->ns_timeouts = 0;
spin_lock_init(&ns->ns_unused_lock);
ns->ns_orig_connect_flags = 0;
ns->ns_connect_flags = 0;
GOTO(out_proc, rc);
}
- mutex_down(ldlm_namespace_lock(client));
- list_add(&ns->ns_list_chain, ldlm_namespace_list(client));
- atomic_inc(ldlm_namespace_nr(client));
- mutex_up(ldlm_namespace_lock(client));
+ at_init(&ns->ns_at_estimate, ldlm_enqueue_min, 0);
+ ldlm_namespace_register(ns, client);
RETURN(ns);
out_proc:
ldlm_namespace_cleanup(ns, 0);
out_ns:
OBD_FREE_PTR(ns);
out_ref:
- ldlm_put_ref(0);
+ ldlm_put_ref();
RETURN(NULL);
}
LDLM_DEBUG(lock, "setting FL_LOCAL_ONLY");
if (lock->l_completion_ast)
lock->l_completion_ast(lock, 0, NULL);
- LDLM_LOCK_PUT(lock);
+ LDLM_LOCK_RELEASE(lock);
continue;
}
"client node");
ldlm_lock_destroy(lock);
}
- LDLM_LOCK_PUT(lock);
+ LDLM_LOCK_RELEASE(lock);
} while (1);
EXIT;
res = list_entry(tmp, struct ldlm_resource, lr_hash);
ldlm_resource_getref(res);
spin_unlock(&ns->ns_hash_lock);
+ LDLM_RESOURCE_ADDREF(res);
cleanup_resource(res, &res->lr_granted, flags);
cleanup_resource(res, &res->lr_converting, flags);
* client gets blocking ast when lock gets distracted by
* server. This is 1_4 branch solution, let's see how
* will it behave. */
+ LDLM_RESOURCE_DELREF(res);
if (!ldlm_resource_putref_locked(res))
CDEBUG(D_INFO,
"Namespace %s resource refcount nonzero "
return ELDLM_OK;
}
-/* Cleanup, but also free, the namespace */
-int ldlm_namespace_free_prior(struct ldlm_namespace *ns)
+static int __ldlm_namespace_free(struct ldlm_namespace *ns, int force)
{
ENTRY;
- if (!ns)
- RETURN(ELDLM_OK);
-
- mutex_down(ldlm_namespace_lock(ns->ns_client));
- /*
- * Some asserts and possibly other parts of code still using
- * list_empty(&ns->ns_list_chain). This is why it is important
- * to use list_del_init() here.
- */
- list_del_init(&ns->ns_list_chain);
- atomic_dec(ldlm_namespace_nr(ns->ns_client));
- ldlm_pool_fini(&ns->ns_pool);
- mutex_up(ldlm_namespace_lock(ns->ns_client));
/* At shutdown time, don't call the cancellation callback */
- ldlm_namespace_cleanup(ns, 0);
+ ldlm_namespace_cleanup(ns, force ? LDLM_FL_LOCAL_ONLY : 0);
if (ns->ns_refcount > 0) {
struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
CDEBUG(D_DLMTRACE,
"dlm namespace %s free waiting on refcount %d\n",
ns->ns_name, ns->ns_refcount);
+force_wait:
+ if (force)
+ lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
+
rc = l_wait_event(ns->ns_waitq,
ns->ns_refcount == 0, &lwi);
- if (ns->ns_refcount)
- LCONSOLE_ERROR_MSG(0x139, "Lock manager: wait for %s "
- "namespace cleanup aborted with %d "
- "resources in use. (%d)\nI'm going "
- "to try to clean up anyway, but I "
- "might need a reboot of this node.\n",
- ns->ns_name, (int) ns->ns_refcount,
- rc);
+
+ /* Forced cleanups should be able to reclaim all references,
+ * so it's safe to wait forever... we can't leak locks... */
+ if (force && rc == -ETIMEDOUT) {
+ LCONSOLE_ERROR("Forced cleanup waiting for %s "
+ "namespace with %d resources in use, "
+ "(rc=%d)\n", ns->ns_name,
+ ns->ns_refcount, rc);
+ GOTO(force_wait, rc);
+ }
+
+ if (ns->ns_refcount) {
+ LCONSOLE_ERROR("Cleanup waiting for %s namespace "
+ "with %d resources in use, (rc=%d)\n",
+ ns->ns_name,
+ ns->ns_refcount, rc);
+ RETURN(ELDLM_NAMESPACE_EXISTS);
+ }
CDEBUG(D_DLMTRACE,
"dlm namespace %s free done waiting\n", ns->ns_name);
}
RETURN(ELDLM_OK);
}
-int ldlm_namespace_free_post(struct ldlm_namespace *ns, int force)
+/**
+ * Performs various cleanups for passed \a ns to make it drop refc and be ready
+ * for freeing. Waits for refc == 0.
+ *
+ * The following is done:
+ * (0) Unregister \a ns from its list to make inaccessible for potential users
+ * like pools thread and others;
+ * (1) Clear all locks in \a ns.
+ */
+void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
+ struct obd_import *imp,
+ int force)
{
+ int rc;
ENTRY;
- if (!ns)
- RETURN(ELDLM_OK);
+ if (!ns) {
+ EXIT;
+ return;
+ }
+
+ /*
+ * Make sure that nobody can find this ns in its list.
+ */
+ ldlm_namespace_unregister(ns, ns->ns_client);
+
+ /*
+ * Can fail with -EINTR when force == 0 in which case try harder.
+ */
+ rc = __ldlm_namespace_free(ns, force);
+ if (rc != ELDLM_OK) {
+ if (imp) {
+ ptlrpc_disconnect_import(imp, 0);
+ ptlrpc_invalidate_import(imp);
+ }
+
+ /*
+ * With all requests dropped and the import inactive
+ * we are gaurenteed all reference will be dropped.
+ */
+ rc = __ldlm_namespace_free(ns, 1);
+ LASSERT(rc == 0);
+ }
+ EXIT;
+}
+
+/**
+ * Performs freeing memory structures related to \a ns. This is only done when
+ * ldlm_namespce_free_prior() successfully removed all resources referencing
+ * \a ns and its refc == 0.
+ */
+void ldlm_namespace_free_post(struct ldlm_namespace *ns)
+{
+ ENTRY;
+ if (!ns) {
+ EXIT;
+ return;
+ }
+
+ /*
+ * Fini pool _before_ parent proc dir is removed. This is important as
+ * ldlm_pool_fini() removes own proc dir which is child to @dir. Removing
+ * it after @dir may cause oops.
+ */
+ ldlm_pool_fini(&ns->ns_pool);
#ifdef LPROCFS
{
OBD_VFREE(ns->ns_hash, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
OBD_FREE(ns->ns_name, strlen(ns->ns_name) + 1);
- /*
- * @ns should be not on list in this time, otherwise this will cause
- * issues realted to using freed @ns in pools thread.
+
+ /*
+ * Namespace \a ns should be not on list in this time, otherwise this
+ * will cause issues realted to using freed \a ns in pools thread.
*/
LASSERT(list_empty(&ns->ns_list_chain));
OBD_FREE_PTR(ns);
- ldlm_put_ref(force);
- RETURN(ELDLM_OK);
+ ldlm_put_ref();
+ EXIT;
}
* lprocfs entries, and then free memory. It will be called w/o cli->cl_sem
* held.
*/
-int ldlm_namespace_free(struct ldlm_namespace *ns, int force)
+void ldlm_namespace_free(struct ldlm_namespace *ns,
+ struct obd_import *imp,
+ int force)
{
- ldlm_namespace_free_prior(ns);
- ldlm_namespace_free_post(ns, force);
- return ELDLM_OK;
+ ldlm_namespace_free_prior(ns, imp, force);
+ ldlm_namespace_free_post(ns);
}
-void ldlm_namespace_get_nolock(struct ldlm_namespace *ns)
+void ldlm_namespace_get_locked(struct ldlm_namespace *ns)
{
- LASSERT(ns->ns_refcount >= 0);
ns->ns_refcount++;
}
void ldlm_namespace_get(struct ldlm_namespace *ns)
{
spin_lock(&ns->ns_hash_lock);
- ldlm_namespace_get_nolock(ns);
+ ldlm_namespace_get_locked(ns);
spin_unlock(&ns->ns_hash_lock);
}
-void ldlm_namespace_put_nolock(struct ldlm_namespace *ns, int wakeup)
+void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup)
{
LASSERT(ns->ns_refcount > 0);
ns->ns_refcount--;
void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup)
{
spin_lock(&ns->ns_hash_lock);
- ldlm_namespace_put_nolock(ns, wakeup);
+ ldlm_namespace_put_locked(ns, wakeup);
spin_unlock(&ns->ns_hash_lock);
}
+/* Register @ns in the list of namespaces */
+void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client)
+{
+ mutex_down(ldlm_namespace_lock(client));
+ LASSERT(list_empty(&ns->ns_list_chain));
+ list_add(&ns->ns_list_chain, ldlm_namespace_list(client));
+ atomic_inc(ldlm_namespace_nr(client));
+ mutex_up(ldlm_namespace_lock(client));
+}
+
+/* Unregister @ns from the list of namespaces */
+void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client)
+{
+ mutex_down(ldlm_namespace_lock(client));
+ LASSERT(!list_empty(&ns->ns_list_chain));
+ /*
+ * Some asserts and possibly other parts of code still using
+ * list_empty(&ns->ns_list_chain). This is why it is important
+ * to use list_del_init() here.
+ */
+ list_del_init(&ns->ns_list_chain);
+ atomic_dec(ldlm_namespace_nr(client));
+ mutex_up(ldlm_namespace_lock(client));
+}
+
/* Should be called under ldlm_namespace_lock(client) taken */
-void ldlm_namespace_move(struct ldlm_namespace *ns, ldlm_side_t client)
+void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client)
{
LASSERT(!list_empty(&ns->ns_list_chain));
LASSERT_SEM_LOCKED(ldlm_namespace_lock(client));
}
/* Should be called under ldlm_namespace_lock(client) taken */
-struct ldlm_namespace *ldlm_namespace_first(ldlm_side_t client)
+struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client)
{
LASSERT_SEM_LOCKED(ldlm_namespace_lock(client));
LASSERT(!list_empty(ldlm_namespace_list(client)));
- return container_of(ldlm_namespace_list(client)->next,
+ return container_of(ldlm_namespace_list(client)->next,
struct ldlm_namespace, ns_list_chain);
}
static __u32 ldlm_hash_fn(struct ldlm_resource *parent,
struct ldlm_resource *res;
int idx;
- OBD_SLAB_ALLOC(res, ldlm_resource_slab, CFS_ALLOC_IO, sizeof *res);
+ OBD_SLAB_ALLOC_PTR_GFP(res, ldlm_resource_slab, CFS_ALLOC_IO);
if (res == NULL)
return NULL;
atomic_set(&res->lr_refcount, 1);
spin_lock_init(&res->lr_lock);
+ lu_ref_init(&res->lr_reference);
/* one who creates the resource must unlock
* the semaphore after lvb initialization */
bucket = ns->ns_hash + hash;
list_add(&res->lr_hash, bucket);
ns->ns_resources++;
- ldlm_namespace_get_nolock(ns);
+ ldlm_namespace_get_locked(ns);
if (parent == NULL) {
list_add(&res->lr_childof, &ns->ns_root_list);
LBUG();
}
- /* Pass 0 here to not wake ->ns_waitq up yet, we will do it few
+ /* Pass 0 here to not wake ->ns_waitq up yet, we will do it few
* lines below when all children are freed. */
- ldlm_namespace_put_nolock(ns, 0);
+ ldlm_namespace_put_locked(ns, 0);
list_del_init(&res->lr_hash);
list_del_init(&res->lr_childof);
+ lu_ref_fini(&res->lr_reference);
ns->ns_resources--;
if (ns->ns_resources == 0)
{
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);
check_res_locked(res);
- ldlm_resource_dump(D_OTHER, res);
+ ldlm_resource_dump(D_INFO, res);
CDEBUG(D_OTHER, "About to insert this lock after %p:\n", original);
ldlm_lock_dump(D_OTHER, new, 0);
if (!((libcfs_debug | D_ERROR) & level))
return;
- CDEBUG(level, "--- Namespace: %s (rc: %d, side: %s)\n",
- ns->ns_name, ns->ns_refcount,
+ CDEBUG(level, "--- Namespace: %s (rc: %d, side: %s)\n",
+ ns->ns_name, ns->ns_refcount,
ns_is_client(ns) ? "client" : "server");
if (cfs_time_before(cfs_time_current(), ns->ns_next_dump))
ldlm_resource_getref(res);
spin_unlock(&ns->ns_hash_lock);
+ LDLM_RESOURCE_ADDREF(res);
lock_res(res);
ldlm_resource_dump(level, res);
unlock_res(res);
+ LDLM_RESOURCE_DELREF(res);
spin_lock(&ns->ns_hash_lock);
tmp = tmp->next;
ldlm_resource_putref_locked(res);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff