/* -*- 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 (c) 2002, 2010, Oracle and/or its affiliates. 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
# include <liblustre.h>
#endif
+#include <lustre_fid.h>
#include <obd_class.h>
#include "ldlm_internal.h"
cfs_mem_cache_t *ldlm_resource_slab, *ldlm_lock_slab;
-struct semaphore ldlm_namespace_lock;
-struct list_head ldlm_namespace_list = CFS_LIST_HEAD_INIT(ldlm_namespace_list);
+cfs_atomic_t ldlm_srv_namespace_nr = CFS_ATOMIC_INIT(0);
+cfs_atomic_t ldlm_cli_namespace_nr = CFS_ATOMIC_INIT(0);
+
+cfs_semaphore_t ldlm_srv_namespace_lock;
+CFS_LIST_HEAD(ldlm_srv_namespace_list);
+
+cfs_semaphore_t ldlm_cli_namespace_lock;
+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;
cfs_proc_dir_entry_t *ldlm_svc_proc_dir = NULL;
+extern unsigned int ldlm_cancel_unused_locks_before_replay;
+
#ifdef LPROCFS
static int ldlm_proc_dump_ns(struct file *file, const char *buffer,
unsigned long count, void *data)
{
- ldlm_dump_all_namespaces(D_DLMTRACE);
+ ldlm_dump_all_namespaces(LDLM_NAMESPACE_SERVER, D_DLMTRACE);
+ ldlm_dump_all_namespaces(LDLM_NAMESPACE_CLIENT, D_DLMTRACE);
RETURN(count);
}
int rc;
struct lprocfs_vars list[] = {
{ "dump_namespaces", NULL, ldlm_proc_dump_ns, NULL },
+ { "cancel_unused_locks_before_replay",
+ lprocfs_rd_uint, lprocfs_wr_uint,
+ &ldlm_cancel_unused_locks_before_replay, NULL },
{ NULL }};
ENTRY;
LASSERT(ldlm_ns_proc_dir == NULL);
void ldlm_proc_cleanup(void)
{
- if (ldlm_svc_proc_dir)
+ if (ldlm_svc_proc_dir)
lprocfs_remove(&ldlm_svc_proc_dir);
- if (ldlm_ns_proc_dir)
+ if (ldlm_ns_proc_dir)
lprocfs_remove(&ldlm_ns_proc_dir);
if (ldlm_type_proc_dir)
lprocfs_remove(&ldlm_type_proc_dir);
}
-static int lprocfs_uint_rd(char *page, char **start, off_t off,
- int count, int *eof, void *data)
+static int lprocfs_rd_ns_resources(char *page, char **start, off_t off,
+ int count, int *eof, void *data)
+{
+ struct ldlm_namespace *ns = data;
+ __u64 res = 0;
+ cfs_hash_bd_t bd;
+ int i;
+
+ /* result is not strictly consistant */
+ cfs_hash_for_each_bucket(ns->ns_rs_hash, &bd, i)
+ res += cfs_hash_bd_count_get(&bd);
+ return lprocfs_rd_u64(page, start, off, count, eof, &res);
+}
+
+static int lprocfs_rd_ns_locks(char *page, char **start, off_t off,
+ int count, int *eof, void *data)
{
- unsigned int *temp = (unsigned int *)data;
- return snprintf(page, count, "%u\n", *temp);
+ struct ldlm_namespace *ns = data;
+ __u64 locks;
+
+ locks = lprocfs_stats_collector(ns->ns_stats, LDLM_NSS_LOCKS,
+ LPROCFS_FIELDS_FLAGS_SUM);
+ return lprocfs_rd_u64(page, start, off, count, eof, &locks);
}
-static int lprocfs_read_lru_size(char *page, char **start, off_t off,
- int count, int *eof, void *data)
+static int lprocfs_rd_lru_size(char *page, char **start, off_t off,
+ int count, int *eof, void *data)
{
struct ldlm_namespace *ns = data;
- return lprocfs_uint_rd(page, start, off, count, eof,
- &ns->ns_max_unused);
+ __u32 *nr = &ns->ns_max_unused;
+
+ if (ns_connect_lru_resize(ns))
+ nr = &ns->ns_nr_unused;
+ return lprocfs_rd_uint(page, start, off, count, eof, nr);
}
-#define MAX_STRING_SIZE 128
-static int lprocfs_write_lru_size(struct file *file, const char *buffer,
- unsigned long count, void *data)
+static int lprocfs_wr_lru_size(struct file *file, const char *buffer,
+ unsigned long count, void *data)
{
struct ldlm_namespace *ns = data;
char dummy[MAX_STRING_SIZE + 1], *end;
unsigned long tmp;
+ int lru_resize;
dummy[MAX_STRING_SIZE] = '\0';
- if (copy_from_user(dummy, buffer, MAX_STRING_SIZE))
+ if (cfs_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);
- tmp = ns->ns_max_unused;
- ns->ns_max_unused = 0;
- ldlm_cancel_lru(ns, LDLM_SYNC);
- ns->ns_max_unused = tmp;
+ ldlm_ns_name(ns));
+ 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,
+ LDLM_CANCEL_PASSED);
+ if (canceled < unused) {
+ CDEBUG(D_DLMTRACE,
+ "not all requested locks are canceled, "
+ "requested: %d, canceled: %d\n", unused,
+ canceled);
+ return -EINVAL;
+ }
+ } else {
+ tmp = ns->ns_max_unused;
+ ns->ns_max_unused = 0;
+ ldlm_cancel_lru(ns, 0, LDLM_SYNC, LDLM_CANCEL_PASSED);
+ ns->ns_max_unused = tmp;
+ }
return count;
}
tmp = simple_strtoul(dummy, &end, 0);
- if (tmp == 0 && *end) {
+ if (dummy == end) {
CERROR("invalid value written\n");
return -EINVAL;
}
+ lru_resize = (tmp == 0);
+
+ if (ns_connect_lru_resize(ns)) {
+ if (!lru_resize)
+ ns->ns_max_unused = (unsigned int)tmp;
- 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;
+ if (tmp > ns->ns_nr_unused)
+ tmp = ns->ns_nr_unused;
+ tmp = ns->ns_nr_unused - tmp;
- ldlm_cancel_lru(ns, LDLM_ASYNC);
+ CDEBUG(D_DLMTRACE,
+ "changing namespace %s unused locks from %u to %u\n",
+ ldlm_ns_name(ns), ns->ns_nr_unused,
+ (unsigned int)tmp);
+ ldlm_cancel_lru(ns, tmp, LDLM_ASYNC, LDLM_CANCEL_PASSED);
+
+ if (!lru_resize) {
+ CDEBUG(D_DLMTRACE,
+ "disable lru_resize for namespace %s\n",
+ ldlm_ns_name(ns));
+ ns->ns_connect_flags &= ~OBD_CONNECT_LRU_RESIZE;
+ }
+ } else {
+ CDEBUG(D_DLMTRACE,
+ "changing namespace %s max_unused from %u to %u\n",
+ ldlm_ns_name(ns), 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
+ * turning it on here. */
+ if (lru_resize &&
+ (ns->ns_orig_connect_flags & OBD_CONNECT_LRU_RESIZE)) {
+ CDEBUG(D_DLMTRACE,
+ "enable lru_resize for namespace %s\n",
+ ldlm_ns_name(ns));
+ ns->ns_connect_flags |= OBD_CONNECT_LRU_RESIZE;
+ }
+ }
return count;
}
-void ldlm_proc_namespace(struct ldlm_namespace *ns)
+void ldlm_namespace_proc_unregister(struct ldlm_namespace *ns)
+{
+ struct proc_dir_entry *dir;
+
+ dir = lprocfs_srch(ldlm_ns_proc_dir, ldlm_ns_name(ns));
+ if (dir == NULL) {
+ CERROR("dlm namespace %s has no procfs dir?\n",
+ ldlm_ns_name(ns));
+ } else {
+ lprocfs_remove(&dir);
+ }
+
+ if (ns->ns_stats != NULL)
+ lprocfs_free_stats(&ns->ns_stats);
+}
+
+int ldlm_namespace_proc_register(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);
+ LASSERT(ns->ns_rs_hash != NULL);
+
+ ns->ns_stats = lprocfs_alloc_stats(LDLM_NSS_LAST, 0);
+ if (ns->ns_stats == NULL)
+ return -ENOMEM;
+
+ lprocfs_counter_init(ns->ns_stats, LDLM_NSS_LOCKS,
+ LPROCFS_CNTR_AVGMINMAX, "locks", "locks");
lock_name[MAX_STRING_SIZE] = '\0';
memset(lock_vars, 0, sizeof(lock_vars));
lock_vars[0].name = lock_name;
- snprintf(lock_name, MAX_STRING_SIZE, "%s/resource_count", ns->ns_name);
- lock_vars[0].data = &ns->ns_refcount;
- lock_vars[0].read_fptr = lprocfs_rd_atomic;
+ snprintf(lock_name, MAX_STRING_SIZE, "%s/resource_count",
+ ldlm_ns_name(ns));
+ lock_vars[0].data = ns;
+ lock_vars[0].read_fptr = lprocfs_rd_ns_resources;
lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
- snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_count", ns->ns_name);
- lock_vars[0].data = &ns->ns_locks;
- lock_vars[0].read_fptr = lprocfs_rd_u64;
+ snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_count",
+ ldlm_ns_name(ns));
+ lock_vars[0].data = ns;
+ lock_vars[0].read_fptr = lprocfs_rd_ns_locks;
lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
- if (ns->ns_client) {
+ if (ns_is_client(ns)) {
snprintf(lock_name, MAX_STRING_SIZE, "%s/lock_unused_count",
- ns->ns_name);
+ ldlm_ns_name(ns));
lock_vars[0].data = &ns->ns_nr_unused;
- lock_vars[0].read_fptr = lprocfs_uint_rd;
+ 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/lru_size",
- ns->ns_name);
+ ldlm_ns_name(ns));
lock_vars[0].data = ns;
- lock_vars[0].read_fptr = lprocfs_read_lru_size;
- lock_vars[0].write_fptr = lprocfs_write_lru_size;
+ lock_vars[0].read_fptr = lprocfs_rd_lru_size;
+ 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/lru_max_age",
+ ldlm_ns_name(ns));
+ lock_vars[0].data = &ns->ns_max_age;
+ 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);
+ } else {
+ snprintf(lock_name, MAX_STRING_SIZE, "%s/ctime_age_limit",
+ ldlm_ns_name(ns));
+ lock_vars[0].data = &ns->ns_ctime_age_limit;
+ 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/lock_timeouts",
+ ldlm_ns_name(ns));
+ 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",
+ ldlm_ns_name(ns));
+ lock_vars[0].data = &ns->ns_max_nolock_size;
+ 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/contention_seconds",
+ ldlm_ns_name(ns));
+ lock_vars[0].data = &ns->ns_contention_time;
+ 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/contended_locks",
+ ldlm_ns_name(ns));
+ lock_vars[0].data = &ns->ns_contended_locks;
+ 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/max_parallel_ast",
+ ldlm_ns_name(ns));
+ lock_vars[0].data = &ns->ns_max_parallel_ast;
+ 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);
}
+ return 0;
}
#undef MAX_STRING_SIZE
-#else
-#define ldlm_proc_namespace(ns) do {} while (0)
+#else /* LPROCFS */
+
+#define ldlm_namespace_proc_unregister(ns) ({;})
+#define ldlm_namespace_proc_register(ns) ({0;})
+
#endif /* LPROCFS */
-struct ldlm_namespace *ldlm_namespace_new(char *name, __u32 client)
+static unsigned ldlm_res_hop_hash(cfs_hash_t *hs,
+ const void *key, unsigned mask)
+{
+ const struct ldlm_res_id *id = key;
+ unsigned val = 0;
+ unsigned i;
+
+ for (i = 0; i < RES_NAME_SIZE; i++)
+ val += id->name[i];
+ return val & mask;
+}
+
+static unsigned ldlm_res_hop_fid_hash(cfs_hash_t *hs,
+ const void *key, unsigned mask)
+{
+ const struct ldlm_res_id *id = key;
+ struct lu_fid fid;
+ __u32 hash;
+ __u32 val;
+
+ fid.f_seq = id->name[LUSTRE_RES_ID_SEQ_OFF];
+ fid.f_oid = (__u32)id->name[LUSTRE_RES_ID_OID_OFF];
+ fid.f_ver = (__u32)id->name[LUSTRE_RES_ID_VER_OFF];
+
+ hash = fid_flatten32(&fid);
+ hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
+ if (id->name[LUSTRE_RES_ID_HSH_OFF] != 0) {
+ val = id->name[LUSTRE_RES_ID_HSH_OFF];
+ hash += (val >> 5) + (val << 11);
+ } else {
+ val = fid_oid(&fid);
+ }
+ hash = cfs_hash_long(hash, hs->hs_bkt_bits);
+ /* give me another random factor */
+ hash -= cfs_hash_long((unsigned long)hs, val % 11 + 3);
+
+ hash <<= hs->hs_cur_bits - hs->hs_bkt_bits;
+ hash |= ldlm_res_hop_hash(hs, key, CFS_HASH_NBKT(hs) - 1);
+
+ return hash & mask;
+}
+
+static void *ldlm_res_hop_key(cfs_hlist_node_t *hnode)
+{
+ struct ldlm_resource *res;
+
+ res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
+ return &res->lr_name;
+}
+
+static int ldlm_res_hop_keycmp(const void *key, cfs_hlist_node_t *hnode)
+{
+ struct ldlm_resource *res;
+
+ res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
+ return ldlm_res_eq((const struct ldlm_res_id *)key,
+ (const struct ldlm_res_id *)&res->lr_name);
+}
+
+static void *ldlm_res_hop_object(cfs_hlist_node_t *hnode)
+{
+ return cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
+}
+
+static void ldlm_res_hop_get_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+{
+ struct ldlm_resource *res;
+
+ res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
+ ldlm_resource_getref(res);
+ LDLM_RESOURCE_ADDREF(res);
+}
+
+static void ldlm_res_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+{
+ struct ldlm_resource *res;
+
+ res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
+ /* cfs_hash_for_each_nolock is the only chance we call it */
+ LDLM_RESOURCE_DELREF(res);
+ ldlm_resource_putref_locked(res);
+}
+
+static void ldlm_res_hop_put(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+{
+ struct ldlm_resource *res;
+
+ res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
+ LDLM_RESOURCE_DELREF(res);
+ ldlm_resource_putref(res);
+}
+
+cfs_hash_ops_t ldlm_ns_hash_ops = {
+ .hs_hash = ldlm_res_hop_hash,
+ .hs_key = ldlm_res_hop_key,
+ .hs_keycmp = ldlm_res_hop_keycmp,
+ .hs_keycpy = NULL,
+ .hs_object = ldlm_res_hop_object,
+ .hs_get = ldlm_res_hop_get_locked,
+ .hs_put_locked = ldlm_res_hop_put_locked,
+ .hs_put = ldlm_res_hop_put
+};
+
+cfs_hash_ops_t ldlm_ns_fid_hash_ops = {
+ .hs_hash = ldlm_res_hop_fid_hash,
+ .hs_key = ldlm_res_hop_key,
+ .hs_keycmp = ldlm_res_hop_keycmp,
+ .hs_keycpy = NULL,
+ .hs_object = ldlm_res_hop_object,
+ .hs_get = ldlm_res_hop_get_locked,
+ .hs_put_locked = ldlm_res_hop_put_locked,
+ .hs_put = ldlm_res_hop_put
+};
+
+typedef struct {
+ ldlm_ns_type_t nsd_type;
+ /** hash bucket bits */
+ unsigned nsd_bkt_bits;
+ /** hash bits */
+ unsigned nsd_all_bits;
+ /** hash operations */
+ cfs_hash_ops_t *nsd_hops;
+} ldlm_ns_hash_def_t;
+
+ldlm_ns_hash_def_t ldlm_ns_hash_defs[] =
+{
+ {
+ .nsd_type = LDLM_NS_TYPE_MDC,
+ .nsd_bkt_bits = 11,
+ .nsd_all_bits = 16,
+ .nsd_hops = &ldlm_ns_fid_hash_ops,
+ },
+ {
+ .nsd_type = LDLM_NS_TYPE_MDT,
+ .nsd_bkt_bits = 14,
+ .nsd_all_bits = 21,
+ .nsd_hops = &ldlm_ns_fid_hash_ops,
+ },
+ {
+ .nsd_type = LDLM_NS_TYPE_OSC,
+ .nsd_bkt_bits = 8,
+ .nsd_all_bits = 12,
+ .nsd_hops = &ldlm_ns_hash_ops,
+ },
+ {
+ .nsd_type = LDLM_NS_TYPE_OST,
+ .nsd_bkt_bits = 11,
+ .nsd_all_bits = 17,
+ .nsd_hops = &ldlm_ns_hash_ops,
+ },
+ {
+ .nsd_type = LDLM_NS_TYPE_MGC,
+ .nsd_bkt_bits = 4,
+ .nsd_all_bits = 4,
+ .nsd_hops = &ldlm_ns_hash_ops,
+ },
+ {
+ .nsd_type = LDLM_NS_TYPE_MGT,
+ .nsd_bkt_bits = 4,
+ .nsd_all_bits = 4,
+ .nsd_hops = &ldlm_ns_hash_ops,
+ },
+ {
+ .nsd_type = LDLM_NS_TYPE_UNKNOWN,
+ },
+};
+
+struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
+ ldlm_side_t client,
+ ldlm_appetite_t apt,
+ ldlm_ns_type_t ns_type)
{
struct ldlm_namespace *ns = NULL;
- struct list_head *bucket;
- int rc;
+ struct ldlm_ns_bucket *nsb;
+ ldlm_ns_hash_def_t *nsd;
+ cfs_hash_bd_t bd;
+ int idx;
+ int rc;
ENTRY;
+ LASSERT(obd != NULL);
+
rc = ldlm_get_ref();
if (rc) {
CERROR("ldlm_get_ref failed: %d\n", rc);
RETURN(NULL);
}
- OBD_ALLOC(ns, sizeof(*ns));
+ for (idx = 0;;idx++) {
+ nsd = &ldlm_ns_hash_defs[idx];
+ if (nsd->nsd_type == LDLM_NS_TYPE_UNKNOWN) {
+ CERROR("Unknown type %d for ns %s\n", ns_type, name);
+ GOTO(out_ref, NULL);
+ }
+
+ if (nsd->nsd_type == ns_type)
+ break;
+ }
+
+ OBD_ALLOC_PTR(ns);
if (!ns)
GOTO(out_ref, NULL);
- OBD_VMALLOC(ns->ns_hash, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
- if (!ns->ns_hash)
+ ns->ns_rs_hash = cfs_hash_create(name,
+ nsd->nsd_all_bits, nsd->nsd_all_bits,
+ nsd->nsd_bkt_bits, sizeof(*nsb),
+ CFS_HASH_MIN_THETA,
+ CFS_HASH_MAX_THETA,
+ nsd->nsd_hops,
+ CFS_HASH_DEPTH |
+ CFS_HASH_BIGNAME |
+ CFS_HASH_SPIN_BKTLOCK |
+ CFS_HASH_NO_ITEMREF);
+ if (ns->ns_rs_hash == NULL)
GOTO(out_ns, NULL);
- OBD_ALLOC(ns->ns_name, strlen(name) + 1);
- if (!ns->ns_name)
- GOTO(out_hash, NULL);
+ cfs_hash_for_each_bucket(ns->ns_rs_hash, &bd, idx) {
+ nsb = cfs_hash_bd_extra_get(ns->ns_rs_hash, &bd);
+ at_init(&nsb->nsb_at_estimate, ldlm_enqueue_min, 0);
+ nsb->nsb_namespace = ns;
+ }
- strcpy(ns->ns_name, name);
+ ns->ns_obd = obd;
+ ns->ns_appetite = apt;
+ ns->ns_client = client;
- CFS_INIT_LIST_HEAD(&ns->ns_root_list);
- ns->ns_refcount = 0;
- ns->ns_client = client;
- spin_lock_init(&ns->ns_hash_lock);
- atomic_set(&ns->ns_locks, 0);
- ns->ns_resources = 0;
+ CFS_INIT_LIST_HEAD(&ns->ns_list_chain);
+ CFS_INIT_LIST_HEAD(&ns->ns_unused_list);
+ cfs_spin_lock_init(&ns->ns_lock);
+ cfs_atomic_set(&ns->ns_bref, 0);
cfs_waitq_init(&ns->ns_waitq);
- for (bucket = ns->ns_hash + RES_HASH_SIZE - 1; bucket >= ns->ns_hash;
- bucket--)
- CFS_INIT_LIST_HEAD(bucket);
+ ns->ns_max_nolock_size = NS_DEFAULT_MAX_NOLOCK_BYTES;
+ ns->ns_contention_time = NS_DEFAULT_CONTENTION_SECONDS;
+ ns->ns_contended_locks = NS_DEFAULT_CONTENDED_LOCKS;
+
+ ns->ns_max_parallel_ast = LDLM_DEFAULT_PARALLEL_AST_LIMIT;
+ ns->ns_nr_unused = 0;
+ 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;
+ ns->ns_orig_connect_flags = 0;
+ ns->ns_connect_flags = 0;
+ ns->ns_stopping = 0;
+ rc = ldlm_namespace_proc_register(ns);
+ if (rc != 0) {
+ CERROR("Can't initialize ns proc, rc %d\n", rc);
+ GOTO(out_hash, rc);
+ }
- CFS_INIT_LIST_HEAD(&ns->ns_unused_list);
- ns->ns_nr_unused = 0;
- ns->ns_max_unused = LDLM_DEFAULT_LRU_SIZE;
- spin_lock_init(&ns->ns_unused_lock);
-
- mutex_down(&ldlm_namespace_lock);
- list_add(&ns->ns_list_chain, &ldlm_namespace_list);
- mutex_up(&ldlm_namespace_lock);
- ldlm_proc_namespace(ns);
- RETURN(ns);
+ idx = cfs_atomic_read(ldlm_namespace_nr(client));
+ rc = ldlm_pool_init(&ns->ns_pool, ns, idx, client);
+ if (rc) {
+ CERROR("Can't initialize lock pool, rc %d\n", rc);
+ GOTO(out_proc, rc);
+ }
+ ldlm_namespace_register(ns, client);
+ RETURN(ns);
+out_proc:
+ ldlm_namespace_proc_unregister(ns);
+ ldlm_namespace_cleanup(ns, 0);
out_hash:
- POISON(ns->ns_hash, 0x5a, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
- OBD_VFREE(ns->ns_hash, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
+ cfs_hash_putref(ns->ns_rs_hash);
out_ns:
- OBD_FREE(ns, sizeof(*ns));
+ OBD_FREE_PTR(ns);
out_ref:
- ldlm_put_ref(0);
+ ldlm_put_ref();
RETURN(NULL);
}
/* If flags contains FL_LOCAL_ONLY, 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,
+ * as a result--notably, that we shouldn't cancel locks with refs. -phil */
+static void cleanup_resource(struct ldlm_resource *res, cfs_list_t *q,
int flags)
{
- struct list_head *tmp;
- int rc = 0, client = res->lr_namespace->ns_client;
+ cfs_list_t *tmp;
+ int rc = 0, client = ns_is_client(ldlm_res_to_ns(res));
int local_only = (flags & LDLM_FL_LOCAL_ONLY);
- ENTRY;
-
do {
struct ldlm_lock *lock = NULL;
-
+
/* first, we look for non-cleaned-yet lock
* all cleaned locks are marked by CLEANED flag */
lock_res(res);
- list_for_each(tmp, q) {
- lock = list_entry(tmp, struct ldlm_lock, l_res_link);
+ cfs_list_for_each(tmp, q) {
+ lock = cfs_list_entry(tmp, struct ldlm_lock,
+ l_res_link);
if (lock->l_flags & LDLM_FL_CLEANED) {
lock = NULL;
continue;
lock->l_flags |= LDLM_FL_CLEANED;
break;
}
-
+
if (lock == NULL) {
unlock_res(res);
break;
lock->l_flags |= LDLM_FL_FAILED;
lock->l_flags |= flags;
+ /* ... without sending a CANCEL message for local_only. */
+ if (local_only)
+ lock->l_flags |= LDLM_FL_LOCAL_ONLY;
+
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 ... */
- /* ... without sending a CANCEL message. */
- lock->l_flags |= LDLM_FL_LOCAL_ONLY;
unlock_res(res);
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;
}
unlock_res(res);
ldlm_lock2handle(lock, &lockh);
- 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_only || rc != ELDLM_OK)
- ldlm_lock_cancel(lock);
+ rc = ldlm_cli_cancel(&lockh);
+ if (rc)
+ CERROR("ldlm_cli_cancel: %d\n", rc);
} else {
ldlm_resource_unlink_lock(lock);
unlock_res(res);
"client node");
ldlm_lock_destroy(lock);
}
- LDLM_LOCK_PUT(lock);
+ LDLM_LOCK_RELEASE(lock);
} while (1);
+}
- EXIT;
+static int ldlm_resource_clean(cfs_hash_t *hs, cfs_hash_bd_t *bd,
+ cfs_hlist_node_t *hnode, void *arg)
+{
+ struct ldlm_resource *res = cfs_hash_object(hs, hnode);
+ int flags = (int)(unsigned long)arg;
+
+ cleanup_resource(res, &res->lr_granted, flags);
+ cleanup_resource(res, &res->lr_converting, flags);
+ cleanup_resource(res, &res->lr_waiting, flags);
+
+ return 0;
}
-int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags)
+static int ldlm_resource_complain(cfs_hash_t *hs, cfs_hash_bd_t *bd,
+ cfs_hlist_node_t *hnode, void *arg)
{
- struct list_head *tmp;
- int i;
+ struct ldlm_resource *res = cfs_hash_object(hs, hnode);
+
+ CERROR("Namespace %s resource refcount nonzero "
+ "(%d) after lock cleanup; forcing "
+ "cleanup.\n",
+ ldlm_ns_name(ldlm_res_to_ns(res)),
+ cfs_atomic_read(&res->lr_refcount) - 1);
+ CERROR("Resource: %p ("LPU64"/"LPU64"/"LPU64"/"
+ LPU64") (rc: %d)\n", res,
+ res->lr_name.name[0], res->lr_name.name[1],
+ res->lr_name.name[2], res->lr_name.name[3],
+ cfs_atomic_read(&res->lr_refcount) - 1);
+ return 0;
+}
+
+int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags)
+{
if (ns == NULL) {
CDEBUG(D_INFO, "NULL ns, skipping cleanup\n");
return ELDLM_OK;
}
- for (i = 0; i < RES_HASH_SIZE; i++) {
- spin_lock(&ns->ns_hash_lock);
- tmp = ns->ns_hash[i].next;
- while (tmp != &(ns->ns_hash[i])) {
- struct ldlm_resource *res;
- res = list_entry(tmp, struct ldlm_resource, lr_hash);
- ldlm_resource_getref(res);
- spin_unlock(&ns->ns_hash_lock);
-
- cleanup_resource(res, &res->lr_granted, flags);
- cleanup_resource(res, &res->lr_converting, flags);
- cleanup_resource(res, &res->lr_waiting, flags);
-
- spin_lock(&ns->ns_hash_lock);
- tmp = tmp->next;
-
- /* XXX: former stuff caused issues in case of race
- * between ldlm_namespace_cleanup() and lockd() when
- * client gets blocking ast when lock gets distracted by
- * server. This is 1_4 branch solution, let's see how
- * will it behave. */
- if (!ldlm_resource_putref_locked(res))
- CDEBUG(D_INFO,
- "Namespace %s resource refcount nonzero "
- "(%d) after lock cleanup; forcing cleanup.\n",
- ns->ns_name, atomic_read(&res->lr_refcount));
- }
- spin_unlock(&ns->ns_hash_lock);
- }
-
+ cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_clean,
+ (void *)(unsigned long)flags);
+ cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_complain, NULL);
return ELDLM_OK;
}
-/* Cleanup, but also free, the namespace */
-int ldlm_namespace_free(struct ldlm_namespace *ns, int force)
+static int __ldlm_namespace_free(struct ldlm_namespace *ns, int force)
{
ENTRY;
- if (!ns)
- RETURN(ELDLM_OK);
-
- mutex_down(&ldlm_namespace_lock);
- list_del(&ns->ns_list_chain);
- mutex_up(&ldlm_namespace_lock);
/* At shutdown time, don't call the cancellation callback */
- ldlm_namespace_cleanup(ns, 0);
+ ldlm_namespace_cleanup(ns, force ? LDLM_FL_LOCAL_ONLY : 0);
-#ifdef LPROCFS
- {
- struct proc_dir_entry *dir;
- dir = lprocfs_srch(ldlm_ns_proc_dir, ns->ns_name);
- if (dir == NULL) {
- CERROR("dlm namespace %s has no procfs dir?\n",
- ns->ns_name);
- } else {
- lprocfs_remove(&dir);
- }
- }
-#endif
-
- if (ns->ns_refcount > 0) {
+ if (cfs_atomic_read(&ns->ns_bref) > 0) {
struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
int rc;
CDEBUG(D_DLMTRACE,
"dlm namespace %s free waiting on refcount %d\n",
- ns->ns_name, ns->ns_refcount);
+ ldlm_ns_name(ns), cfs_atomic_read(&ns->ns_bref));
+force_wait:
+ if (force)
+ lwi = LWI_TIMEOUT(obd_timeout * CFS_HZ / 4, NULL, NULL);
+
rc = l_wait_event(ns->ns_waitq,
- ns->ns_refcount == 0, &lwi);
- if (ns->ns_refcount)
- LCONSOLE_ERROR("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);
- CDEBUG(D_DLMTRACE,
- "dlm namespace %s free done waiting\n", ns->ns_name);
+ cfs_atomic_read(&ns->ns_bref) == 0, &lwi);
+
+ /* 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", ldlm_ns_name(ns),
+ cfs_atomic_read(&ns->ns_bref), rc);
+ GOTO(force_wait, rc);
+ }
+
+ if (cfs_atomic_read(&ns->ns_bref)) {
+ LCONSOLE_ERROR("Cleanup waiting for %s namespace "
+ "with %d resources in use, (rc=%d)\n",
+ ldlm_ns_name(ns),
+ cfs_atomic_read(&ns->ns_bref), rc);
+ RETURN(ELDLM_NAMESPACE_EXISTS);
+ }
+ CDEBUG(D_DLMTRACE, "dlm namespace %s free done waiting\n",
+ ldlm_ns_name(ns));
}
- POISON(ns->ns_hash, 0x5a, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
- OBD_VFREE(ns->ns_hash, sizeof(*ns->ns_hash) * RES_HASH_SIZE);
- OBD_FREE(ns->ns_name, strlen(ns->ns_name) + 1);
- OBD_FREE(ns, sizeof(*ns));
+ RETURN(ELDLM_OK);
+}
- ldlm_put_ref(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) {
+ EXIT;
+ return;
+ }
- return ELDLM_OK;
+ cfs_spin_lock(&ns->ns_lock);
+ ns->ns_stopping = 1;
+ cfs_spin_unlock(&ns->ns_lock);
+
+ /*
+ * 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;
}
-static __u32 ldlm_hash_fn(struct ldlm_resource *parent, struct ldlm_res_id name)
+/**
+ * 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)
{
- __u32 hash = 0;
- int i;
+ ENTRY;
+ if (!ns) {
+ EXIT;
+ return;
+ }
- for (i = 0; i < RES_NAME_SIZE; i++)
- hash += name.name[i];
- hash += (__u32)((unsigned long)parent >> 4);
+ /*
+ * Make sure that nobody can find this ns in its list.
+ */
+ ldlm_namespace_unregister(ns, ns->ns_client);
+ /*
+ * 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);
+
+ ldlm_namespace_proc_unregister(ns);
+ cfs_hash_putref(ns->ns_rs_hash);
+ /*
+ * 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(cfs_list_empty(&ns->ns_list_chain));
+ OBD_FREE_PTR(ns);
+ ldlm_put_ref();
+ EXIT;
+}
- return (hash & RES_HASH_MASK);
+
+/* Cleanup the resource, and free namespace.
+ * bug 12864:
+ * Deadlock issue:
+ * proc1: destroy import
+ * class_disconnect_export(grab cl_sem) ->
+ * -> ldlm_namespace_free ->
+ * -> lprocfs_remove(grab _lprocfs_lock).
+ * proc2: read proc info
+ * lprocfs_fops_read(grab _lprocfs_lock) ->
+ * -> osc_rd_active, etc(grab cl_sem).
+ *
+ * So that I have to split the ldlm_namespace_free into two parts - the first
+ * part ldlm_namespace_free_prior is used to cleanup the resource which is
+ * being used; the 2nd part ldlm_namespace_free_post is used to unregister the
+ * lprocfs entries, and then free memory. It will be called w/o cli->cl_sem
+ * held.
+ */
+void ldlm_namespace_free(struct ldlm_namespace *ns,
+ struct obd_import *imp,
+ int force)
+{
+ ldlm_namespace_free_prior(ns, imp, force);
+ ldlm_namespace_free_post(ns);
+}
+
+void ldlm_namespace_get(struct ldlm_namespace *ns)
+{
+ cfs_atomic_inc(&ns->ns_bref);
+}
+
+void ldlm_namespace_put(struct ldlm_namespace *ns)
+{
+ if (cfs_atomic_dec_and_lock(&ns->ns_bref, &ns->ns_lock)) {
+ cfs_waitq_signal(&ns->ns_waitq);
+ cfs_spin_unlock(&ns->ns_lock);
+ }
+}
+
+/* Register @ns in the list of namespaces */
+void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client)
+{
+ cfs_mutex_down(ldlm_namespace_lock(client));
+ LASSERT(cfs_list_empty(&ns->ns_list_chain));
+ cfs_list_add(&ns->ns_list_chain, ldlm_namespace_list(client));
+ cfs_atomic_inc(ldlm_namespace_nr(client));
+ cfs_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)
+{
+ cfs_mutex_down(ldlm_namespace_lock(client));
+ LASSERT(!cfs_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.
+ */
+ cfs_list_del_init(&ns->ns_list_chain);
+ cfs_atomic_dec(ldlm_namespace_nr(client));
+ cfs_mutex_up(ldlm_namespace_lock(client));
+}
+
+/* Should be called under ldlm_namespace_lock(client) taken */
+void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client)
+{
+ LASSERT(!cfs_list_empty(&ns->ns_list_chain));
+ LASSERT_SEM_LOCKED(ldlm_namespace_lock(client));
+ cfs_list_move_tail(&ns->ns_list_chain, ldlm_namespace_list(client));
+}
+
+/* Should be called under ldlm_namespace_lock(client) taken */
+struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client)
+{
+ LASSERT_SEM_LOCKED(ldlm_namespace_lock(client));
+ LASSERT(!cfs_list_empty(ldlm_namespace_list(client)));
+ return container_of(ldlm_namespace_list(client)->next,
+ struct ldlm_namespace, ns_list_chain);
}
static struct ldlm_resource *ldlm_resource_new(void)
{
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;
- memset(res, 0, sizeof(*res));
-
- CFS_INIT_LIST_HEAD(&res->lr_children);
- CFS_INIT_LIST_HEAD(&res->lr_childof);
CFS_INIT_LIST_HEAD(&res->lr_granted);
CFS_INIT_LIST_HEAD(&res->lr_converting);
CFS_INIT_LIST_HEAD(&res->lr_waiting);
- atomic_set(&res->lr_refcount, 1);
- spin_lock_init(&res->lr_lock);
+
+ /* initialize interval trees for each lock mode*/
+ for (idx = 0; idx < LCK_MODE_NUM; idx++) {
+ res->lr_itree[idx].lit_size = 0;
+ res->lr_itree[idx].lit_mode = 1 << idx;
+ res->lr_itree[idx].lit_root = NULL;
+ }
+
+ cfs_atomic_set(&res->lr_refcount, 1);
+ cfs_spin_lock_init(&res->lr_lock);
+ lu_ref_init(&res->lr_reference);
/* one who creates the resource must unlock
* the semaphore after lvb initialization */
- init_MUTEX_LOCKED(&res->lr_lvb_sem);
+ cfs_init_mutex_locked(&res->lr_lvb_sem);
return res;
}
-/* must be called with hash lock held */
-static struct ldlm_resource *
-ldlm_resource_find(struct ldlm_namespace *ns, struct ldlm_res_id name, __u32 hash)
+/* Args: unlocked namespace
+ * * Locks: takes and releases NS hash-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,
+ const struct ldlm_res_id *name, ldlm_type_t type, int create)
{
- struct list_head *bucket, *tmp;
+ cfs_hlist_node_t *hnode;
struct ldlm_resource *res;
+ cfs_hash_bd_t bd;
+ __u64 version;
- LASSERT_SPIN_LOCKED(&ns->ns_hash_lock);
- bucket = ns->ns_hash + hash;
-
- list_for_each(tmp, bucket) {
- res = list_entry(tmp, struct ldlm_resource, lr_hash);
- if (memcmp(&res->lr_name, &name, sizeof(res->lr_name)) == 0)
- return res;
+ LASSERT(ns != NULL);
+ LASSERT(parent == NULL);
+ LASSERT(ns->ns_rs_hash != NULL);
+ LASSERT(name->name[0] != 0);
+
+ cfs_hash_bd_get_and_lock(ns->ns_rs_hash, (void *)name, &bd, 0);
+ hnode = cfs_hash_bd_lookup_locked(ns->ns_rs_hash, &bd, (void *)name);
+ if (hnode != NULL) {
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 0);
+ res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
+ /* synchronize WRT resource creation */
+ if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
+ cfs_down(&res->lr_lvb_sem);
+ cfs_up(&res->lr_lvb_sem);
+ }
+ return res;
}
- return NULL;
-}
+ version = cfs_hash_bd_version_get(&bd);
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 0);
-/* Args: locked namespace
- * Returns: newly-allocated, referenced, unlocked resource */
-static struct ldlm_resource *
-ldlm_resource_add(struct ldlm_namespace *ns, struct ldlm_resource *parent,
- struct ldlm_res_id name, __u32 hash, ldlm_type_t type)
-{
- struct list_head *bucket;
- struct ldlm_resource *res, *old_res;
- ENTRY;
+ if (create == 0)
+ return NULL;
LASSERTF(type >= LDLM_MIN_TYPE && type < LDLM_MAX_TYPE,
"type: %d\n", type);
-
res = ldlm_resource_new();
if (!res)
- RETURN(NULL);
+ return NULL;
- res->lr_name = name;
- res->lr_namespace = ns;
- res->lr_type = type;
+ res->lr_ns_bucket = cfs_hash_bd_extra_get(ns->ns_rs_hash, &bd);
+ res->lr_name = *name;
+ res->lr_type = type;
res->lr_most_restr = LCK_NL;
- spin_lock(&ns->ns_hash_lock);
- old_res = ldlm_resource_find(ns, name, hash);
- if (old_res) {
+ cfs_hash_bd_lock(ns->ns_rs_hash, &bd, 1);
+ hnode = (version == cfs_hash_bd_version_get(&bd)) ? NULL :
+ cfs_hash_bd_lookup_locked(ns->ns_rs_hash, &bd, (void *)name);
+
+ if (hnode != NULL) {
/* someone won the race and added the resource before */
- ldlm_resource_getref(old_res);
- spin_unlock(&ns->ns_hash_lock);
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
+ /* clean lu_ref for failed resource */
+ lu_ref_fini(&res->lr_reference);
OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
+
+ res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash);
/* synchronize WRT resource creation */
if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
- down(&old_res->lr_lvb_sem);
- up(&old_res->lr_lvb_sem);
+ cfs_down(&res->lr_lvb_sem);
+ cfs_up(&res->lr_lvb_sem);
}
- RETURN(old_res);
+ return res;
}
-
/* we won! let's add the resource */
- bucket = ns->ns_hash + hash;
- list_add(&res->lr_hash, bucket);
- ns->ns_resources++;
- ns->ns_refcount++;
-
- if (parent == NULL) {
- list_add(&res->lr_childof, &ns->ns_root_list);
- } else {
- res->lr_parent = parent;
- list_add(&res->lr_childof, &parent->lr_children);
- }
- spin_unlock(&ns->ns_hash_lock);
+ cfs_hash_bd_add_locked(ns->ns_rs_hash, &bd, &res->lr_hash);
+ if (cfs_hash_bd_count_get(&bd) == 1)
+ ldlm_namespace_get(ns);
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
int rc;
rc = ns->ns_lvbo->lvbo_init(res);
if (rc)
CERROR("lvbo_init failed for resource "
- LPU64": rc %d\n", name.name[0], rc);
+ LPU64": rc %d\n", name->name[0], rc);
/* we create resource with locked lr_lvb_sem */
- up(&res->lr_lvb_sem);
- }
-
- 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,
- struct ldlm_res_id name, ldlm_type_t type, int create)
-{
- __u32 hash = ldlm_hash_fn(parent, name);
- struct ldlm_resource *res = NULL;
- ENTRY;
-
- LASSERT(ns != NULL);
- LASSERT(ns->ns_hash != NULL);
- LASSERT(name.name[0] != 0);
-
- spin_lock(&ns->ns_hash_lock);
- res = ldlm_resource_find(ns, name, hash);
- if (res) {
- ldlm_resource_getref(res);
- spin_unlock(&ns->ns_hash_lock);
- /* synchronize WRT resource creation */
- if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
- down(&res->lr_lvb_sem);
- up(&res->lr_lvb_sem);
- }
- RETURN(res);
+ cfs_up(&res->lr_lvb_sem);
}
- spin_unlock(&ns->ns_hash_lock);
-
- if (create == 0)
- RETURN(NULL);
- res = ldlm_resource_add(ns, parent, name, hash, type);
- RETURN(res);
+ return res;
}
struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res)
{
LASSERT(res != NULL);
LASSERT(res != LP_POISON);
- atomic_inc(&res->lr_refcount);
+ cfs_atomic_inc(&res->lr_refcount);
CDEBUG(D_INFO, "getref res: %p count: %d\n", res,
- atomic_read(&res->lr_refcount));
+ cfs_atomic_read(&res->lr_refcount));
return res;
}
-void __ldlm_resource_putref_final(struct ldlm_resource *res)
+static void __ldlm_resource_putref_final(cfs_hash_bd_t *bd,
+ struct ldlm_resource *res)
{
- struct ldlm_namespace *ns = res->lr_namespace;
-
- LASSERT_SPIN_LOCKED(&ns->ns_hash_lock);
+ struct ldlm_ns_bucket *nsb = res->lr_ns_bucket;
- if (!list_empty(&res->lr_granted)) {
+ if (!cfs_list_empty(&res->lr_granted)) {
ldlm_resource_dump(D_ERROR, res);
LBUG();
}
- if (!list_empty(&res->lr_converting)) {
+ if (!cfs_list_empty(&res->lr_converting)) {
ldlm_resource_dump(D_ERROR, res);
LBUG();
}
- if (!list_empty(&res->lr_waiting)) {
+ if (!cfs_list_empty(&res->lr_waiting)) {
ldlm_resource_dump(D_ERROR, res);
LBUG();
}
- if (!list_empty(&res->lr_children)) {
- ldlm_resource_dump(D_ERROR, res);
- LBUG();
- }
-
- ns->ns_refcount--;
- list_del_init(&res->lr_hash);
- list_del_init(&res->lr_childof);
-
- ns->ns_resources--;
- if (ns->ns_resources == 0)
- wake_up(&ns->ns_waitq);
+ cfs_hash_bd_del_locked(nsb->nsb_namespace->ns_rs_hash,
+ bd, &res->lr_hash);
+ lu_ref_fini(&res->lr_reference);
+ if (cfs_hash_bd_count_get(bd) == 0)
+ ldlm_namespace_put(nsb->nsb_namespace);
}
/* Returns 1 if the resource was freed, 0 if it remains. */
int ldlm_resource_putref(struct ldlm_resource *res)
{
- struct ldlm_namespace *ns = res->lr_namespace;
- int rc = 0;
- ENTRY;
-
- CDEBUG(D_INFO, "putref res: %p count: %d\n", res,
- atomic_read(&res->lr_refcount) - 1);
- LASSERTF(atomic_read(&res->lr_refcount) > 0, "%d",
- atomic_read(&res->lr_refcount));
- LASSERTF(atomic_read(&res->lr_refcount) < LI_POISON, "%d",
- atomic_read(&res->lr_refcount));
-
- if (atomic_dec_and_lock(&res->lr_refcount, &ns->ns_hash_lock)) {
- __ldlm_resource_putref_final(res);
- spin_unlock(&ns->ns_hash_lock);
- if (res->lr_lvb_data)
- OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
+ struct ldlm_namespace *ns = ldlm_res_to_ns(res);
+ cfs_hash_bd_t bd;
+
+ LASSERT_ATOMIC_GT_LT(&res->lr_refcount, 0, LI_POISON);
+ CDEBUG(D_INFO, "putref res: %p count: %d\n",
+ res, cfs_atomic_read(&res->lr_refcount) - 1);
+
+ cfs_hash_bd_get(ns->ns_rs_hash, &res->lr_name, &bd);
+ if (cfs_hash_bd_dec_and_lock(ns->ns_rs_hash, &bd, &res->lr_refcount)) {
+ __ldlm_resource_putref_final(&bd, res);
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
+ if (ns->ns_lvbo && ns->ns_lvbo->lvbo_free)
+ ns->ns_lvbo->lvbo_free(res);
OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
- rc = 1;
+ return 1;
}
-
- RETURN(rc);
+ return 0;
}
/* Returns 1 if the resource was freed, 0 if it remains. */
int ldlm_resource_putref_locked(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) < LI_POISON);
-
- LASSERT(atomic_read(&res->lr_refcount) >= 0);
- if (atomic_dec_and_test(&res->lr_refcount)) {
- __ldlm_resource_putref_final(res);
- if (res->lr_lvb_data)
- OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
+ struct ldlm_namespace *ns = ldlm_res_to_ns(res);
+
+ LASSERT_ATOMIC_GT_LT(&res->lr_refcount, 0, LI_POISON);
+ CDEBUG(D_INFO, "putref res: %p count: %d\n",
+ res, cfs_atomic_read(&res->lr_refcount) - 1);
+
+ if (cfs_atomic_dec_and_test(&res->lr_refcount)) {
+ cfs_hash_bd_t bd;
+
+ cfs_hash_bd_get(ldlm_res_to_ns(res)->ns_rs_hash,
+ &res->lr_name, &bd);
+ __ldlm_resource_putref_final(&bd, res);
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
+ /* NB: ns_rs_hash is created with CFS_HASH_NO_ITEMREF,
+ * so we should never be here while calling cfs_hash_del,
+ * cfs_hash_for_each_nolock is the only case we can get
+ * here, which is safe to release cfs_hash_bd_lock.
+ */
+ if (ns->ns_lvbo && ns->ns_lvbo->lvbo_free)
+ ns->ns_lvbo->lvbo_free(res);
OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
- rc = 1;
- }
- RETURN(rc);
+ cfs_hash_bd_lock(ns->ns_rs_hash, &bd, 1);
+ return 1;
+ }
+ return 0;
}
-void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
+void ldlm_resource_add_lock(struct ldlm_resource *res, cfs_list_t *head,
struct ldlm_lock *lock)
{
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);
return;
}
- LASSERT(list_empty(&lock->l_res_link));
+ LASSERT(cfs_list_empty(&lock->l_res_link));
- list_add_tail(&lock->l_res_link, head);
+ cfs_list_add_tail(&lock->l_res_link, head);
}
void ldlm_resource_insert_lock_after(struct ldlm_lock *original,
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);
goto out;
}
- LASSERT(list_empty(&new->l_res_link));
+ LASSERT(cfs_list_empty(&new->l_res_link));
- list_add(&new->l_res_link, &original->l_res_link);
+ cfs_list_add(&new->l_res_link, &original->l_res_link);
out:;
}
void ldlm_resource_unlink_lock(struct ldlm_lock *lock)
{
+ int type = lock->l_resource->lr_type;
+
check_res_locked(lock->l_resource);
- list_del_init(&lock->l_res_link);
+ if (type == LDLM_IBITS || type == LDLM_PLAIN)
+ ldlm_unlink_lock_skiplist(lock);
+ else if (type == LDLM_EXTENT)
+ ldlm_extent_unlink_lock(lock);
+ cfs_list_del_init(&lock->l_res_link);
}
void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc)
desc->lr_name = res->lr_name;
}
-void ldlm_dump_all_namespaces(int level)
+void ldlm_dump_all_namespaces(ldlm_side_t client, int level)
{
- struct list_head *tmp;
+ cfs_list_t *tmp;
if (!((libcfs_debug | D_ERROR) & level))
return;
- mutex_down(&ldlm_namespace_lock);
+ cfs_mutex_down(ldlm_namespace_lock(client));
- list_for_each(tmp, &ldlm_namespace_list) {
+ cfs_list_for_each(tmp, ldlm_namespace_list(client)) {
struct ldlm_namespace *ns;
- ns = list_entry(tmp, struct ldlm_namespace, ns_list_chain);
+ ns = cfs_list_entry(tmp, struct ldlm_namespace, ns_list_chain);
ldlm_namespace_dump(level, ns);
}
- mutex_up(&ldlm_namespace_lock);
+ cfs_mutex_up(ldlm_namespace_lock(client));
}
-void ldlm_namespace_dump(int level, struct ldlm_namespace *ns)
+static int ldlm_res_hash_dump(cfs_hash_t *hs, cfs_hash_bd_t *bd,
+ cfs_hlist_node_t *hnode, void *arg)
{
- struct list_head *tmp;
+ struct ldlm_resource *res = cfs_hash_object(hs, hnode);
+ int level = (int)(unsigned long)arg;
+
+ lock_res(res);
+ ldlm_resource_dump(level, res);
+ unlock_res(res);
+ return 0;
+}
+
+void ldlm_namespace_dump(int level, struct ldlm_namespace *ns)
+{
if (!((libcfs_debug | D_ERROR) & level))
return;
- CDEBUG(level, "--- Namespace: %s (rc: %d, client: %d)\n",
- ns->ns_name, ns->ns_refcount, ns->ns_client);
+ CDEBUG(level, "--- Namespace: %s (rc: %d, side: %s)\n",
+ ldlm_ns_name(ns), cfs_atomic_read(&ns->ns_bref),
+ ns_is_client(ns) ? "client" : "server");
if (cfs_time_before(cfs_time_current(), ns->ns_next_dump))
return;
- spin_lock(&ns->ns_hash_lock);
- tmp = ns->ns_root_list.next;
- while (tmp != &ns->ns_root_list) {
- struct ldlm_resource *res;
- res = list_entry(tmp, struct ldlm_resource, lr_childof);
-
- ldlm_resource_getref(res);
- spin_unlock(&ns->ns_hash_lock);
-
- lock_res(res);
- ldlm_resource_dump(level, res);
- unlock_res(res);
-
- spin_lock(&ns->ns_hash_lock);
- tmp = tmp->next;
- ldlm_resource_putref_locked(res);
- }
+ cfs_hash_for_each_nolock(ns->ns_rs_hash,
+ ldlm_res_hash_dump,
+ (void *)(unsigned long)level);
+ cfs_spin_lock(&ns->ns_lock);
ns->ns_next_dump = cfs_time_shift(10);
- spin_unlock(&ns->ns_hash_lock);
+ cfs_spin_unlock(&ns->ns_lock);
}
void ldlm_resource_dump(int level, struct ldlm_resource *res)
{
- struct list_head *tmp;
+ cfs_list_t *tmp;
int pos;
CLASSERT(RES_NAME_SIZE == 4);
CDEBUG(level, "--- Resource: %p ("LPU64"/"LPU64"/"LPU64"/"LPU64
") (rc: %d)\n", res, res->lr_name.name[0], res->lr_name.name[1],
res->lr_name.name[2], res->lr_name.name[3],
- atomic_read(&res->lr_refcount));
+ cfs_atomic_read(&res->lr_refcount));
- if (!list_empty(&res->lr_granted)) {
+ if (!cfs_list_empty(&res->lr_granted)) {
pos = 0;
CDEBUG(level, "Granted locks:\n");
- list_for_each(tmp, &res->lr_granted) {
+ cfs_list_for_each(tmp, &res->lr_granted) {
struct ldlm_lock *lock;
- lock = list_entry(tmp, struct ldlm_lock, l_res_link);
+ lock = cfs_list_entry(tmp, struct ldlm_lock,
+ l_res_link);
ldlm_lock_dump(level, lock, ++pos);
}
}
- if (!list_empty(&res->lr_converting)) {
+ if (!cfs_list_empty(&res->lr_converting)) {
pos = 0;
CDEBUG(level, "Converting locks:\n");
- list_for_each(tmp, &res->lr_converting) {
+ cfs_list_for_each(tmp, &res->lr_converting) {
struct ldlm_lock *lock;
- lock = list_entry(tmp, struct ldlm_lock, l_res_link);
+ lock = cfs_list_entry(tmp, struct ldlm_lock,
+ l_res_link);
ldlm_lock_dump(level, lock, ++pos);
}
}
- if (!list_empty(&res->lr_waiting)) {
+ if (!cfs_list_empty(&res->lr_waiting)) {
pos = 0;
CDEBUG(level, "Waiting locks:\n");
- list_for_each(tmp, &res->lr_waiting) {
+ cfs_list_for_each(tmp, &res->lr_waiting) {
struct ldlm_lock *lock;
- lock = list_entry(tmp, struct ldlm_lock, l_res_link);
+ lock = cfs_list_entry(tmp, struct ldlm_lock,
+ l_res_link);
ldlm_lock_dump(level, lock, ++pos);
}
}