/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (C) 2002 Cluster File Systems, Inc.
+ * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
+ * Author: Phil Schwan <phil@clusterfs.com>
+ * Author: Peter Braam <braam@clusterfs.com>
*
- * This code is issued under the GNU General Public License.
- * See the file COPYING in this distribution
+ * This file is part of Lustre, http://www.lustre.org.
*
- * by Cluster File Systems, Inc.
+ * Lustre is free 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.
+ *
+ * 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
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Lustre; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
#define DEBUG_SUBSYSTEM S_LDLM
-
+#ifdef __KERNEL__
#include <linux/lustre_dlm.h>
+#else
+#include <liblustre.h>
+#endif
+
#include <linux/obd_class.h>
kmem_cache_t *ldlm_resource_slab, *ldlm_lock_slab;
int ldlm_proc_setup(struct obd_device *obd)
{
+ int rc;
ENTRY;
-
- if (obd->obd_proc_entry == NULL)
- RETURN(-EINVAL);
-
- ldlm_ns_proc_dir = proc_mkdir("namespaces", obd->obd_proc_entry);
- if (ldlm_ns_proc_dir == NULL) {
- CERROR("Couldn't create /proc/lustre/ldlm/namespaces\n");
- RETURN(-EPERM);
+ LASSERT(ldlm_ns_proc_dir == NULL);
+ LASSERT(obd != NULL);
+ rc = lprocfs_obd_attach(obd, 0);
+ if (rc) {
+ CERROR("LProcFS failed in ldlm-init\n");
+ RETURN(rc);
}
+ ldlm_ns_proc_dir = obd->obd_proc_entry;
RETURN(0);
}
void ldlm_proc_cleanup(struct obd_device *obd)
{
- proc_lustre_remove_obd_entry("namespaces", obd);
+ if (ldlm_ns_proc_dir) {
+ lprocfs_obd_detach(obd);
+ ldlm_ns_proc_dir = NULL;
+ }
}
-/* FIXME: This can go away when we start to really use lprocfs */
-static int lprocfs_ll_rd(char *page, char **start, off_t off,
- int count, int *eof, void *data)
+#ifdef __KERNEL__
+static int lprocfs_uint_rd(char *page, char **start, off_t off,
+ int count, int *eof, void *data)
{
- int len;
- __u64 *temp = (__u64 *)data;
+ unsigned int *temp = (unsigned int *)data;
+ return snprintf(page, count, "%u\n", *temp);
+}
+
+
+#define MAX_STRING_SIZE 128
+void ldlm_proc_namespace(struct ldlm_namespace *ns)
+{
+ struct lprocfs_vars lock_vars[2];
+ char lock_name[MAX_STRING_SIZE + 1];
+
+ LASSERT(ns != NULL);
+ LASSERT(ns->ns_name != NULL);
+
+ lock_name[MAX_STRING_SIZE] = '\0';
+
+ memset(lock_vars, 0, sizeof(lock_vars));
+ lock_vars[0].read_fptr = lprocfs_rd_u64;
+
+ lock_vars[0].name = lock_name;
+
+ snprintf(lock_name, MAX_STRING_SIZE, "%s/resource_count", ns->ns_name);
+
+ lock_vars[0].data = &ns->ns_resources;
+ lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
- len = snprintf(page, count, "%Lu\n", *temp);
+ snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_count", ns->ns_name);
- return len;
+ lock_vars[0].data = &ns->ns_locks;
+ lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
+
+ snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_unused_count",
+ ns->ns_name);
+ lock_vars[0].data = &ns->ns_nr_unused;
+ lock_vars[0].read_fptr = lprocfs_uint_rd;
+ lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
}
+#endif
+#undef MAX_STRING_SIZE
+#define LDLM_MAX_UNUSED 20
struct ldlm_namespace *ldlm_namespace_new(char *name, __u32 client)
{
struct ldlm_namespace *ns = NULL;
struct list_head *bucket;
- struct proc_dir_entry *proc_entry;
+ ENTRY;
OBD_ALLOC(ns, sizeof(*ns));
- if (!ns) {
- LBUG();
- GOTO(out, NULL);
- }
+ if (!ns)
+ RETURN(NULL);
ns->ns_hash = vmalloc(sizeof(*ns->ns_hash) * RES_HASH_SIZE);
- if (!ns->ns_hash) {
- LBUG();
- GOTO(out, ns);
- }
- obd_memory += sizeof(*ns->ns_hash) * RES_HASH_SIZE;
+ if (!ns->ns_hash)
+ GOTO(out_ns, NULL);
+
+ atomic_add(sizeof(*ns->ns_hash) * RES_HASH_SIZE, &obd_memory);
OBD_ALLOC(ns->ns_name, strlen(name) + 1);
- if (!ns->ns_name) {
- LBUG();
- GOTO(out, ns);
- }
+ if (!ns->ns_name)
+ GOTO(out_hash, NULL);
+
strcpy(ns->ns_name, name);
INIT_LIST_HEAD(&ns->ns_root_list);
bucket--)
INIT_LIST_HEAD(bucket);
+ INIT_LIST_HEAD(&ns->ns_unused_list);
+ ns->ns_nr_unused = 0;
+ ns->ns_max_unused = LDLM_MAX_UNUSED;
+
spin_lock(&ldlm_namespace_lock);
list_add(&ns->ns_list_chain, &ldlm_namespace_list);
spin_unlock(&ldlm_namespace_lock);
-
- ns->ns_proc_dir = proc_mkdir(ns->ns_name, ldlm_ns_proc_dir);
- if (ns->ns_proc_dir == NULL)
- CERROR("Unable to create proc directory for namespace.\n");
- proc_entry = create_proc_entry("resource_count", 0444, ns->ns_proc_dir);
- proc_entry->read_proc = lprocfs_ll_rd;
- proc_entry->data = &ns->ns_resources;
- proc_entry = create_proc_entry("lock_count", 0444, ns->ns_proc_dir);
- proc_entry->read_proc = lprocfs_ll_rd;
- proc_entry->data = &ns->ns_locks;
-
+#ifdef __KERNEL__
+ ldlm_proc_namespace(ns);
+#endif
RETURN(ns);
- out:
- if (ns && ns->ns_hash) {
- vfree(ns->ns_hash);
- obd_memory -= sizeof(*ns->ns_hash) * RES_HASH_SIZE;
- }
- if (ns && ns->ns_name)
- OBD_FREE(ns->ns_name, strlen(name) + 1);
- if (ns)
- OBD_FREE(ns, sizeof(*ns));
+out_hash:
+ POISON(ns->ns_hash, 0x5a, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
+ vfree(ns->ns_hash);
+ atomic_sub(sizeof(*ns->ns_hash) * RES_HASH_SIZE, &obd_memory);
+out_ns:
+ OBD_FREE(ns, sizeof(*ns));
return NULL;
}
extern struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
-/* If 'local' is true, don't try to tell the server, just cleanup. */
+/* If 'local_only' is true, don't try to tell the server, just cleanup.
+ * This is currently only used for recovery, and we make certain assumptions
+ * as a result--notably, that we shouldn't cancel locks with refs. -phil
+ *
+ * Called with the ns_lock held. */
static void cleanup_resource(struct ldlm_resource *res, struct list_head *q,
- int local)
+ int local_only)
{
struct list_head *tmp, *pos;
int rc = 0, client = res->lr_namespace->ns_client;
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
LDLM_LOCK_GET(lock);
+ if (local_only && (lock->l_readers || lock->l_writers)) {
+ /* This is a little bit gross, but much better than the
+ * alternative: pretend that we got a blocking AST from
+ * the server, so that when the lock is decref'd, it
+ * will go away ... */
+ lock->l_flags |= LDLM_FL_CBPENDING;
+ /* ... without sending a CANCEL message. */
+ lock->l_flags |= LDLM_FL_LOCAL_ONLY;
+ /* ... and without calling the cancellation callback */
+ lock->l_flags |= LDLM_FL_CANCEL;
+ LDLM_LOCK_PUT(lock);
+ continue;
+ }
+
+ /* At shutdown time, don't call the cancellation callback */
+ lock->l_flags |= LDLM_FL_CANCEL;
+
if (client) {
struct lustre_handle lockh;
ldlm_lock2handle(lock, &lockh);
- if (!local) {
+ if (!local_only) {
rc = ldlm_cli_cancel(&lockh);
if (rc)
CERROR("ldlm_cli_cancel: %d\n", rc);
}
/* Force local cleanup on errors, too. */
- if (local || rc != ELDLM_OK)
+ if (local_only || rc != ELDLM_OK)
ldlm_lock_cancel(lock);
} else {
LDLM_DEBUG(lock, "Freeing a lock still held by a "
- "client node.\n");
+ "client node");
ldlm_resource_unlink_lock(lock);
ldlm_lock_destroy(lock);
}
LDLM_LOCK_PUT(lock);
}
+ EXIT;
}
-int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int local)
+int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int local_only)
{
int i;
+ if (ns == NULL) {
+ CDEBUG(D_INFO, "NULL ns, skipping cleanup\n");
+ return ELDLM_OK;
+ }
+
l_lock(&ns->ns_lock);
for (i = 0; i < RES_HASH_SIZE; i++) {
struct list_head *tmp, *pos;
res = list_entry(tmp, struct ldlm_resource, lr_hash);
ldlm_resource_getref(res);
- cleanup_resource(res, &res->lr_granted, local);
- cleanup_resource(res, &res->lr_converting, local);
- cleanup_resource(res, &res->lr_waiting, local);
+ cleanup_resource(res, &res->lr_granted, local_only);
+ cleanup_resource(res, &res->lr_converting, local_only);
+ cleanup_resource(res, &res->lr_waiting, local_only);
- if (!ldlm_resource_put(res)) {
+ /* XXX what a mess: don't force cleanup if we're
+ * local_only (which is only used by recovery). In that
+ * case, we probably still have outstanding lock refs
+ * which reference these resources. -phil */
+ if (!ldlm_resource_putref(res) && !local_only) {
CERROR("Resource refcount nonzero (%d) after "
"lock cleanup; forcing cleanup.\n",
atomic_read(&res->lr_refcount));
ldlm_resource_dump(res);
atomic_set(&res->lr_refcount, 1);
- ldlm_resource_put(res);
+ ldlm_resource_putref(res);
}
}
}
spin_lock(&ldlm_namespace_lock);
list_del(&ns->ns_list_chain);
- remove_proc_entry("resource_count", ns->ns_proc_dir);
- remove_proc_entry("lock_count", ns->ns_proc_dir);
- remove_proc_entry(ns->ns_name, ldlm_ns_proc_dir);
+
spin_unlock(&ldlm_namespace_lock);
ldlm_namespace_cleanup(ns, 0);
+ POISON(ns->ns_hash, 0x5a, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
vfree(ns->ns_hash /* , sizeof(*ns->ns_hash) * RES_HASH_SIZE */);
- obd_memory -= sizeof(*ns->ns_hash) * RES_HASH_SIZE;
+ atomic_sub(sizeof(*ns->ns_hash) * RES_HASH_SIZE, &obd_memory);
OBD_FREE(ns->ns_name, strlen(ns->ns_name) + 1);
OBD_FREE(ns, sizeof(*ns));
RETURN(0);
}
-static __u32 ldlm_hash_fn(struct ldlm_resource *parent, __u64 *name)
+static __u32 ldlm_hash_fn(struct ldlm_resource *parent, struct ldlm_res_id name)
{
__u32 hash = 0;
int i;
for (i = 0; i < RES_NAME_SIZE; i++)
- hash += name[i];
+ hash += name.name[i];
hash += (__u32)((unsigned long)parent >> 4);
/* Args: locked namespace
* Returns: newly-allocated, referenced, unlocked resource */
-static struct ldlm_resource *ldlm_resource_add(struct ldlm_namespace *ns,
- struct ldlm_resource *parent,
- __u64 *name, __u32 type)
+static struct ldlm_resource *
+ldlm_resource_add(struct ldlm_namespace *ns, struct ldlm_resource *parent,
+ struct ldlm_res_id name, __u32 type)
{
struct list_head *bucket;
struct ldlm_resource *res;
ns->ns_resources++;
spin_unlock(&ns->ns_counter_lock);
- memcpy(res->lr_name, name, sizeof(res->lr_name));
+ l_lock(&ns->ns_lock);
+ memcpy(&res->lr_name, &name, sizeof(res->lr_name));
res->lr_namespace = ns;
ns->ns_refcount++;
bucket = ns->ns_hash + ldlm_hash_fn(parent, name);
list_add(&res->lr_hash, bucket);
- if (parent == NULL)
+ if (parent == NULL) {
list_add(&res->lr_childof, &ns->ns_root_list);
- else {
+ } else {
res->lr_parent = parent;
list_add(&res->lr_childof, &parent->lr_children);
}
+ l_unlock(&ns->ns_lock);
RETURN(res);
}
/* Args: unlocked namespace
* Locks: takes and releases ns->ns_lock and res->lr_lock
* Returns: referenced, unlocked ldlm_resource or NULL */
-struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
- struct ldlm_resource *parent,
- __u64 *name, __u32 type, int create)
+struct ldlm_resource *
+ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent,
+ struct ldlm_res_id name, __u32 type, int create)
{
- struct list_head *bucket;
- struct list_head *tmp = bucket;
+ struct list_head *bucket, *tmp;
struct ldlm_resource *res = NULL;
ENTRY;
- if (ns == NULL || ns->ns_hash == NULL) {
- LBUG();
- RETURN(NULL);
- }
+ LASSERT(ns != NULL);
+ LASSERT(ns->ns_hash != NULL);
l_lock(&ns->ns_lock);
bucket = ns->ns_hash + ldlm_hash_fn(parent, name);
list_for_each(tmp, bucket) {
- struct ldlm_resource *chk;
- chk = list_entry(tmp, struct ldlm_resource, lr_hash);
-
- if (memcmp(chk->lr_name, name, sizeof(chk->lr_name)) == 0) {
- res = chk;
- atomic_inc(&res->lr_refcount);
- EXIT;
- break;
+ res = list_entry(tmp, struct ldlm_resource, lr_hash);
+
+ if (memcmp(&res->lr_name, &name, sizeof(res->lr_name)) == 0) {
+ ldlm_resource_getref(res);
+ l_unlock(&ns->ns_lock);
+ RETURN(res);
}
}
- if (res == NULL && create)
+ if (create)
res = ldlm_resource_add(ns, parent, name, type);
+ else
+ res = NULL;
+
l_unlock(&ns->ns_lock);
RETURN(res);
struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res)
{
atomic_inc(&res->lr_refcount);
+ CDEBUG(D_INFO, "getref res: %p count: %d\n", res,
+ atomic_read(&res->lr_refcount));
return res;
}
/* Returns 1 if the resource was freed, 0 if it remains. */
-int ldlm_resource_put(struct ldlm_resource *res)
+int ldlm_resource_putref(struct ldlm_resource *res)
{
int rc = 0;
+ ENTRY;
+
+ CDEBUG(D_INFO, "putref res: %p count: %d\n", res,
+ atomic_read(&res->lr_refcount) - 1);
+ LASSERT(atomic_read(&res->lr_refcount) > 0);
+ LASSERT(atomic_read(&res->lr_refcount) < 0x5a5a5a5a);
if (atomic_dec_and_test(&res->lr_refcount)) {
struct ldlm_namespace *ns = res->lr_namespace;
if (atomic_read(&res->lr_refcount) != 0) {
/* We lost the race. */
l_unlock(&ns->ns_lock);
- goto out;
+ RETURN(rc);
}
- if (!list_empty(&res->lr_granted))
+ if (!list_empty(&res->lr_granted)) {
+ ldlm_resource_dump(res);
LBUG();
+ }
- if (!list_empty(&res->lr_converting))
+ if (!list_empty(&res->lr_converting)) {
+ ldlm_resource_dump(res);
LBUG();
+ }
- if (!list_empty(&res->lr_waiting))
+ if (!list_empty(&res->lr_waiting)) {
+ ldlm_resource_dump(res);
LBUG();
+ }
- if (!list_empty(&res->lr_children))
+ if (!list_empty(&res->lr_children)) {
+ ldlm_resource_dump(res);
LBUG();
+ }
ns->ns_refcount--;
- list_del(&res->lr_hash);
- list_del(&res->lr_childof);
+ list_del_init(&res->lr_hash);
+ list_del_init(&res->lr_childof);
+ POISON(res, 0x5a, sizeof(*res));
kmem_cache_free(ldlm_resource_slab, res);
l_unlock(&ns->ns_lock);
spin_unlock(&ns->ns_counter_lock);
rc = 1;
- } else {
- ENTRY;
- out:
- if (atomic_read(&res->lr_refcount) < 0)
- LBUG();
+ EXIT;
}
RETURN(rc);
l_lock(&res->lr_namespace->ns_lock);
ldlm_resource_dump(res);
- ldlm_lock_dump(lock);
+ CDEBUG(D_OTHER, "About to add this lock:\n");
+ ldlm_lock_dump(D_OTHER, lock);
- if (!list_empty(&lock->l_res_link))
- LBUG();
+ if (lock->l_destroyed) {
+ CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
+ return;
+ }
+
+ LASSERT(list_empty(&lock->l_res_link));
- list_add(&lock->l_res_link, head);
+ list_add_tail(&lock->l_res_link, head);
l_unlock(&res->lr_namespace->ns_lock);
}
void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc)
{
desc->lr_type = res->lr_type;
- memcpy(desc->lr_name, res->lr_name, sizeof(desc->lr_name));
+ memcpy(&desc->lr_name, &res->lr_name, sizeof(desc->lr_name));
memcpy(desc->lr_version, res->lr_version, sizeof(desc->lr_version));
}
LBUG();
snprintf(name, sizeof(name), "%Lx %Lx %Lx",
- (unsigned long long)res->lr_name[0],
- (unsigned long long)res->lr_name[1],
- (unsigned long long)res->lr_name[2]);
+ (unsigned long long)res->lr_name.name[0],
+ (unsigned long long)res->lr_name.name[1],
+ (unsigned long long)res->lr_name.name[2]);
CDEBUG(D_OTHER, "--- Resource: %p (%s) (rc: %d)\n", res, name,
atomic_read(&res->lr_refcount));
list_for_each(tmp, &res->lr_granted) {
struct ldlm_lock *lock;
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
- ldlm_lock_dump(lock);
+ ldlm_lock_dump(D_OTHER, lock);
}
CDEBUG(D_OTHER, "Converting locks:\n");
list_for_each(tmp, &res->lr_converting) {
struct ldlm_lock *lock;
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
- ldlm_lock_dump(lock);
+ ldlm_lock_dump(D_OTHER, lock);
}
CDEBUG(D_OTHER, "Waiting locks:\n");
list_for_each(tmp, &res->lr_waiting) {
struct ldlm_lock *lock;
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
- ldlm_lock_dump(lock);
+ ldlm_lock_dump(D_OTHER, lock);
}
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff