-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
*
- * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
- * Author: Phil Schwan <phil@clusterfs.com>
- * Author: Peter Braam <braam@clusterfs.com>
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * This file is part of the Lustre file system, http://www.lustre.org
- * Lustre is a trademark of Cluster File Systems, Inc.
+ * 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.
*
- * 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 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).
*
- * 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.
+ * 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
*
- * 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.
+ * 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.
+ *
+ * Copyright (c) 2010, 2013, Intel Corporation.
+ */
+/*
+ * 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 kmem_cache *ldlm_resource_slab, *ldlm_lock_slab;
+
+int ldlm_srv_namespace_nr = 0;
+int ldlm_cli_namespace_nr = 0;
+
+struct mutex ldlm_srv_namespace_lock;
+CFS_LIST_HEAD(ldlm_srv_namespace_list);
-atomic_t ldlm_srv_namespace_nr = ATOMIC_INIT(0);
-atomic_t ldlm_cli_namespace_nr = ATOMIC_INIT(0);
+struct mutex ldlm_cli_namespace_lock;
+/* Client Namespaces that have active resources in them.
+ * Once all resources go away, ldlm_poold moves such namespaces to the
+ * inactive list */
+CFS_LIST_HEAD(ldlm_cli_active_namespace_list);
+/* Client namespaces that don't have any locks in them */
+CFS_LIST_HEAD(ldlm_cli_inactive_namespace_list);
-struct semaphore ldlm_srv_namespace_lock;
-struct list_head ldlm_srv_namespace_list =
- CFS_LIST_HEAD_INIT(ldlm_srv_namespace_list);
+struct proc_dir_entry *ldlm_type_proc_dir = NULL;
+struct proc_dir_entry *ldlm_ns_proc_dir = NULL;
+struct proc_dir_entry *ldlm_svc_proc_dir = NULL;
-struct semaphore ldlm_cli_namespace_lock;
-struct list_head ldlm_cli_namespace_list =
- CFS_LIST_HEAD_INIT(ldlm_cli_namespace_list);
+extern unsigned int ldlm_cancel_unused_locks_before_replay;
-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;
+/* during debug dump certain amount of granted locks for one resource to avoid
+ * DDOS. */
+unsigned int ldlm_dump_granted_max = 256;
#ifdef LPROCFS
-static int ldlm_proc_dump_ns(struct file *file, const char *buffer,
- unsigned long count, void *data)
+static ssize_t
+lprocfs_dump_ns_seq_write(struct file *file, const char *buffer,
+ size_t count, loff_t *off)
{
- ldlm_dump_all_namespaces(LDLM_NAMESPACE_SERVER, D_DLMTRACE);
- ldlm_dump_all_namespaces(LDLM_NAMESPACE_CLIENT, D_DLMTRACE);
- RETURN(count);
+ ldlm_dump_all_namespaces(LDLM_NAMESPACE_SERVER, D_DLMTRACE);
+ ldlm_dump_all_namespaces(LDLM_NAMESPACE_CLIENT, D_DLMTRACE);
+ RETURN(count);
}
+LPROC_SEQ_FOPS_WO_TYPE(ldlm, dump_ns);
+
+LPROC_SEQ_FOPS_RW_TYPE(ldlm_rw, uint);
+LPROC_SEQ_FOPS_RO_TYPE(ldlm, uint);
int ldlm_proc_setup(void)
{
- int rc;
- struct lprocfs_vars list[] = {
- { "dump_namespaces", NULL, ldlm_proc_dump_ns, NULL },
- { NULL }};
- ENTRY;
- LASSERT(ldlm_ns_proc_dir == NULL);
-
- ldlm_type_proc_dir = lprocfs_register(OBD_LDLM_DEVICENAME,
- proc_lustre_root,
- NULL, NULL);
- if (IS_ERR(ldlm_type_proc_dir)) {
- CERROR("LProcFS failed in ldlm-init\n");
- rc = PTR_ERR(ldlm_type_proc_dir);
- GOTO(err, rc);
- }
-
- ldlm_ns_proc_dir = lprocfs_register("namespaces",
- ldlm_type_proc_dir,
- NULL, NULL);
- if (IS_ERR(ldlm_ns_proc_dir)) {
- CERROR("LProcFS failed in ldlm-init\n");
- rc = PTR_ERR(ldlm_ns_proc_dir);
- GOTO(err_type, rc);
- }
-
- ldlm_svc_proc_dir = lprocfs_register("services",
- ldlm_type_proc_dir,
- NULL, NULL);
- if (IS_ERR(ldlm_svc_proc_dir)) {
- CERROR("LProcFS failed in ldlm-init\n");
- rc = PTR_ERR(ldlm_svc_proc_dir);
- GOTO(err_ns, rc);
- }
-
- rc = lprocfs_add_vars(ldlm_type_proc_dir, list, NULL);
-
- RETURN(0);
-
+ int rc;
+ struct lprocfs_seq_vars list[] = {
+ { .name = "dump_namespaces",
+ .fops = &ldlm_dump_ns_fops,
+ .proc_mode = 0222 },
+ { .name = "dump_granted_max",
+ .fops = &ldlm_rw_uint_fops,
+ .data = &ldlm_dump_granted_max },
+ { .name = "cancel_unused_locks_before_replay",
+ .fops = &ldlm_rw_uint_fops,
+ .data = &ldlm_cancel_unused_locks_before_replay },
+ { NULL }};
+ ENTRY;
+ LASSERT(ldlm_ns_proc_dir == NULL);
+
+ ldlm_type_proc_dir = lprocfs_seq_register(OBD_LDLM_DEVICENAME,
+ proc_lustre_root,
+ NULL, NULL);
+ if (IS_ERR(ldlm_type_proc_dir)) {
+ CERROR("LProcFS failed in ldlm-init\n");
+ rc = PTR_ERR(ldlm_type_proc_dir);
+ GOTO(err, rc);
+ }
+
+ ldlm_ns_proc_dir = lprocfs_seq_register("namespaces",
+ ldlm_type_proc_dir,
+ NULL, NULL);
+ if (IS_ERR(ldlm_ns_proc_dir)) {
+ CERROR("LProcFS failed in ldlm-init\n");
+ rc = PTR_ERR(ldlm_ns_proc_dir);
+ GOTO(err_type, rc);
+ }
+
+ ldlm_svc_proc_dir = lprocfs_seq_register("services",
+ ldlm_type_proc_dir,
+ NULL, NULL);
+ if (IS_ERR(ldlm_svc_proc_dir)) {
+ CERROR("LProcFS failed in ldlm-init\n");
+ rc = PTR_ERR(ldlm_svc_proc_dir);
+ GOTO(err_ns, rc);
+ }
+
+ rc = lprocfs_seq_add_vars(ldlm_type_proc_dir, list, NULL);
+ if (rc != 0) {
+ CERROR("LProcFS failed in ldlm-init\n");
+ GOTO(err_svc, rc);
+ }
+
+ RETURN(0);
+
+err_svc:
+ lprocfs_remove(&ldlm_svc_proc_dir);
err_ns:
lprocfs_remove(&ldlm_ns_proc_dir);
err_type:
lprocfs_remove(&ldlm_type_proc_dir);
}
-static int lprocfs_rd_lru_size(char *page, char **start, off_t off,
- int count, int *eof, void *data)
+static int lprocfs_ns_resources_seq_show(struct seq_file *m, void *v)
+{
+ struct ldlm_namespace *ns = m->private;
+ __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_u64_seq_show(m, &res);
+}
+LPROC_SEQ_FOPS_RO(lprocfs_ns_resources);
+
+static int lprocfs_ns_locks_seq_show(struct seq_file *m, void *v)
+{
+ struct ldlm_namespace *ns = m->private;
+ __u64 locks;
+
+ locks = lprocfs_stats_collector(ns->ns_stats, LDLM_NSS_LOCKS,
+ LPROCFS_FIELDS_FLAGS_SUM);
+ return lprocfs_u64_seq_show(m, &locks);
+}
+LPROC_SEQ_FOPS_RO(lprocfs_ns_locks);
+
+static int lprocfs_lru_size_seq_show(struct seq_file *m, void *v)
{
- struct ldlm_namespace *ns = data;
- __u32 *nr = &ns->ns_max_unused;
+ struct ldlm_namespace *ns = m->private;
+ __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);
+ if (ns_connect_lru_resize(ns))
+ nr = &ns->ns_nr_unused;
+ return lprocfs_uint_seq_show(m, nr);
}
-static int lprocfs_wr_lru_size(struct file *file, const char *buffer,
- unsigned long count, void *data)
+static ssize_t lprocfs_lru_size_seq_write(struct file *file, const char *buffer,
+ size_t count, loff_t *off)
{
- struct ldlm_namespace *ns = data;
+ struct ldlm_namespace *ns = ((struct seq_file *)file->private_data)->private;
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 (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);
+ 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);
+ canceled = ldlm_cancel_lru(ns, unused, 0,
+ LDLM_CANCEL_PASSED);
if (canceled < unused) {
- CERROR("not all requested locks are canceled, "
- "requested: %d, canceled: %d\n", 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_lru(ns, 0, 0, LDLM_CANCEL_PASSED);
ns->ns_max_unused = tmp;
}
return count;
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;
+
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",
- ns->ns_name, ns->ns_nr_unused, (unsigned int)tmp);
- ldlm_cancel_lru(ns, (unsigned int)tmp, 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, LCF_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",
- ns->ns_name, ns->ns_max_unused, (unsigned int)tmp);
+ 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_lru(ns, 0, LCF_ASYNC, LDLM_CANCEL_PASSED);
+
+ /* Make sure that LRU resize was originally 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;
}
+LPROC_SEQ_FOPS(lprocfs_lru_size);
+
+static int lprocfs_elc_seq_show(struct seq_file *m, void *v)
+{
+ struct ldlm_namespace *ns = m->private;
+ unsigned int supp = ns_connect_cancelset(ns);
+
+ return lprocfs_uint_seq_show(m, &supp);
+}
+
+static ssize_t lprocfs_elc_seq_write(struct file *file, const char *buffer,
+ size_t count, loff_t *off)
+{
+ struct ldlm_namespace *ns = ((struct seq_file *)file->private_data)->private;
+ unsigned int supp = -1;
+ int rc;
+
+ rc = lprocfs_wr_uint(file, buffer, count, &supp);
+ if (rc < 0)
+ return rc;
+
+ if (supp == 0)
+ ns->ns_connect_flags &= ~OBD_CONNECT_CANCELSET;
+ else if (ns->ns_orig_connect_flags & OBD_CONNECT_CANCELSET)
+ ns->ns_connect_flags |= OBD_CONNECT_CANCELSET;
+ return count;
+}
+LPROC_SEQ_FOPS(lprocfs_elc);
-void ldlm_proc_namespace(struct ldlm_namespace *ns)
+void ldlm_namespace_proc_unregister(struct ldlm_namespace *ns)
{
- struct lprocfs_vars lock_vars[2];
+ if (ns->ns_proc_dir_entry == NULL)
+ CERROR("dlm namespace %s has no procfs dir?\n",
+ ldlm_ns_name(ns));
+ else
+ lprocfs_remove(&ns->ns_proc_dir_entry);
+
+ if (ns->ns_stats != NULL)
+ lprocfs_free_stats(&ns->ns_stats);
+}
+
+int ldlm_namespace_proc_register(struct ldlm_namespace *ns)
+{
+ struct lprocfs_seq_vars lock_vars[2];
char lock_name[MAX_STRING_SIZE + 1];
+ struct proc_dir_entry *ns_pde;
LASSERT(ns != NULL);
- LASSERT(ns->ns_name != NULL);
+ LASSERT(ns->ns_rs_hash != NULL);
+
+ if (ns->ns_proc_dir_entry != NULL) {
+ ns_pde = ns->ns_proc_dir_entry;
+ } else {
+ ns_pde = proc_mkdir(ldlm_ns_name(ns), ldlm_ns_proc_dir);
+ if (ns_pde == NULL)
+ return -ENOMEM;
+ ns->ns_proc_dir_entry = ns_pde;
+ }
+
+ 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;
- 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_atomic;
- lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
-
- if (ns->ns_client) {
- 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_rd_uint;
- lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0);
-
- snprintf(lock_name, MAX_STRING_SIZE, "%s/lru_size",
- ns->ns_name);
- lock_vars[0].data = ns;
- 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",
- ns->ns_name);
- 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);
-
- }
+ ldlm_add_var(&lock_vars[0], ns_pde, "resource_count", ns,
+ &lprocfs_ns_resources_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "lock_count", ns,
+ &lprocfs_ns_locks_fops);
+
+ if (ns_is_client(ns)) {
+ ldlm_add_var(&lock_vars[0], ns_pde, "lock_unused_count",
+ &ns->ns_nr_unused, &ldlm_uint_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "lru_size", ns,
+ &lprocfs_lru_size_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "lru_max_age",
+ &ns->ns_max_age, &ldlm_rw_uint_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "early_lock_cancel",
+ ns, &lprocfs_elc_fops);
+ } else {
+ ldlm_add_var(&lock_vars[0], ns_pde, "ctime_age_limit",
+ &ns->ns_ctime_age_limit, &ldlm_rw_uint_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "lock_timeouts",
+ &ns->ns_timeouts, &ldlm_uint_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "max_nolock_bytes",
+ &ns->ns_max_nolock_size, &ldlm_rw_uint_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "contention_seconds",
+ &ns->ns_contention_time, &ldlm_rw_uint_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "contended_locks",
+ &ns->ns_contended_locks, &ldlm_rw_uint_fops);
+ ldlm_add_var(&lock_vars[0], ns_pde, "max_parallel_ast",
+ &ns->ns_max_parallel_ast, &ldlm_rw_uint_fops);
+ }
+ 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, ldlm_side_t client,
- ldlm_appetite_t apt)
+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_VER_OID_OFF];
+ fid.f_ver = (__u32)(id->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32);
+
+ 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 = hash_long(hash, hs->hs_bkt_bits);
+ /* give me another random factor */
+ hash -= 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);
+}
+
+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_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_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,
+ },
+};
+
+/**
+ * Create and initialize new empty namespace.
+ */
+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, idx;
+ struct ldlm_ns_bucket *nsb;
+ ldlm_ns_hash_def_t *nsd;
+ cfs_hash_bd_t bd;
+ int idx;
+ int rc;
ENTRY;
- rc = ldlm_get_ref(client);
+ LASSERT(obd != NULL);
+
+ rc = ldlm_get_ref();
if (rc) {
CERROR("ldlm_get_ref failed: %d\n", rc);
RETURN(NULL);
}
+ 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);
+ 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;
+ }
+
+ ns->ns_obd = obd;
ns->ns_appetite = apt;
- OBD_ALLOC(ns->ns_name, strlen(name) + 1);
- if (!ns->ns_name)
- GOTO(out_hash, NULL);
-
- strcpy(ns->ns_name, name);
-
- 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_waitq_init(&ns->ns_waitq);
-
- for (bucket = ns->ns_hash + RES_HASH_SIZE - 1; bucket >= ns->ns_hash;
- bucket--)
- CFS_INIT_LIST_HEAD(bucket);
-
- CFS_INIT_LIST_HEAD(&ns->ns_unused_list);
- ns->ns_nr_unused = 0;
- ns->ns_max_unused = LDLM_DEFAULT_LRU_SIZE;
- ns->ns_max_age = LDLM_DEFAULT_MAX_ALIVE;
- spin_lock_init(&ns->ns_unused_lock);
-
- ns->ns_connect_flags = 0;
- mutex_down(ldlm_namespace_lock(client));
- list_add(&ns->ns_list_chain, ldlm_namespace_list(client));
- idx = atomic_read(ldlm_namespace_nr(client));
- atomic_inc(ldlm_namespace_nr(client));
- mutex_up(ldlm_namespace_lock(client));
- ldlm_proc_namespace(ns);
-
+ ns->ns_client = client;
+
+ CFS_INIT_LIST_HEAD(&ns->ns_list_chain);
+ CFS_INIT_LIST_HEAD(&ns->ns_unused_list);
+ spin_lock_init(&ns->ns_lock);
+ atomic_set(&ns->ns_bref, 0);
+ init_waitqueue_head(&ns->ns_waitq);
+
+ 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);
+ }
+
+ idx = ldlm_namespace_nr_read(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_del, rc);
+ CERROR("Can't initialize lock pool, rc %d\n", rc);
+ GOTO(out_proc, rc);
}
- RETURN(ns);
-out_del:
- mutex_down(ldlm_namespace_lock(client));
- list_del(&ns->ns_list_chain);
- atomic_dec(ldlm_namespace_nr(client));
- mutex_up(ldlm_namespace_lock(client));
+ 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_PTR(ns);
out_ref:
- ldlm_put_ref(client, 0);
+ ldlm_put_ref();
RETURN(NULL);
}
+EXPORT_SYMBOL(ldlm_namespace_new);
extern struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
-/* 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
+/**
+ * Cancel and destroy all locks on a resource.
*
- * Called with the ns_lock held. */
-static void cleanup_resource(struct ldlm_resource *res, struct list_head *q,
- int flags)
+ * If flags contains FL_LOCAL_ONLY, don't try to tell the server, just
+ * clean up. This is currently only used for recovery, and we make
+ * certain assumptions as a result--notably, that we shouldn't cancel
+ * locks with refs.
+ */
+static void cleanup_resource(struct ldlm_resource *res, cfs_list_t *q,
+ __u64 flags)
{
- struct list_head *tmp;
- int rc = 0, client = res->lr_namespace->ns_client;
- int local_only = (flags & LDLM_FL_LOCAL_ONLY);
- ENTRY;
-
+ cfs_list_t *tmp;
+ int rc = 0, client = ns_is_client(ldlm_res_to_ns(res));
+ bool local_only = !!(flags & LDLM_FL_LOCAL_ONLY);
do {
struct ldlm_lock *lock = NULL;
- /* first, we look for non-cleaned-yet lock
- * all cleaned locks are marked by CLEANED flag */
+ /* 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);
- if (lock->l_flags & LDLM_FL_CLEANED) {
+ cfs_list_for_each(tmp, q) {
+ lock = cfs_list_entry(tmp, struct ldlm_lock,
+ l_res_link);
+ if (ldlm_is_cleaned(lock)) {
lock = NULL;
continue;
}
LDLM_LOCK_GET(lock);
- lock->l_flags |= LDLM_FL_CLEANED;
+ ldlm_set_cleaned(lock);
break;
}
}
/* Set CBPENDING so nothing in the cancellation path
- * can match this lock */
- lock->l_flags |= LDLM_FL_CBPENDING;
- lock->l_flags |= LDLM_FL_FAILED;
+ * can match this lock. */
+ ldlm_set_cbpending(lock);
+ ldlm_set_failed(lock);
lock->l_flags |= flags;
/* ... without sending a CANCEL message for local_only. */
if (local_only)
- lock->l_flags |= LDLM_FL_LOCAL_ONLY;
+ ldlm_set_local_only(lock);
if (local_only && (lock->l_readers || lock->l_writers)) {
/* This is a little bit gross, but much better than the
* will go away ... */
unlock_res(res);
LDLM_DEBUG(lock, "setting FL_LOCAL_ONLY");
+ if (lock->l_flags & LDLM_FL_FAIL_LOC) {
+ schedule_timeout_and_set_state(
+ TASK_UNINTERRUPTIBLE,
+ cfs_time_seconds(4));
+ set_current_state(TASK_RUNNING);
+ }
if (lock->l_completion_ast)
- lock->l_completion_ast(lock, 0, NULL);
- LDLM_LOCK_PUT(lock);
+ lock->l_completion_ast(lock,
+ LDLM_FL_FAILED, NULL);
+ LDLM_LOCK_RELEASE(lock);
continue;
}
unlock_res(res);
ldlm_lock2handle(lock, &lockh);
- rc = ldlm_cli_cancel(&lockh);
+ rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
if (rc)
CERROR("ldlm_cli_cancel: %d\n", rc);
} else {
"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);
+ __u64 flags = *(__u64 *)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);
+
+ lock_res(res);
+ CERROR("%s: namespace resource "DLDLMRES" (%p) refcount nonzero "
+ "(%d) after lock cleanup; forcing cleanup.\n",
+ ldlm_ns_name(ldlm_res_to_ns(res)), PLDLMRES(res), res,
+ atomic_read(&res->lr_refcount) - 1);
+
+ ldlm_resource_dump(D_ERROR, res);
+ unlock_res(res);
+ return 0;
+}
+/**
+ * Cancel and destroy all locks in the namespace.
+ *
+ * Typically used during evictions when server notified client that it was
+ * evicted and all of its state needs to be destroyed.
+ * Also used during shutdown.
+ */
+int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 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, &flags);
+ cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_complain, NULL);
return ELDLM_OK;
}
+EXPORT_SYMBOL(ldlm_namespace_cleanup);
-/* Cleanup, but also free, the namespace */
-int ldlm_namespace_free(struct ldlm_namespace *ns, int force)
+/**
+ * Attempts to free namespace.
+ *
+ * Only used when namespace goes away, like during an unmount.
+ */
+static int __ldlm_namespace_free(struct ldlm_namespace *ns, int force)
{
- ldlm_side_t client;
- ENTRY;
- if (!ns)
- RETURN(ELDLM_OK);
-
- client = ns->ns_client;
- mutex_down(ldlm_namespace_lock(client));
- list_del(&ns->ns_list_chain);
- atomic_dec(ldlm_namespace_nr(ns->ns_client));
- ldlm_pool_fini(&ns->ns_pool);
- mutex_up(ldlm_namespace_lock(client));
+ ENTRY;
+
+ /* At shutdown time, don't call the cancellation callback */
+ ldlm_namespace_cleanup(ns, force ? LDLM_FL_LOCAL_ONLY : 0);
+
+ if (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",
+ ldlm_ns_name(ns), atomic_read(&ns->ns_bref));
+force_wait:
+ if (force)
+ lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
+
+ rc = l_wait_event(ns->ns_waitq,
+ 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),
+ atomic_read(&ns->ns_bref), rc);
+ GOTO(force_wait, rc);
+ }
+
+ if (atomic_read(&ns->ns_bref)) {
+ LCONSOLE_ERROR("Cleanup waiting for %s namespace "
+ "with %d resources in use, (rc=%d)\n",
+ ldlm_ns_name(ns),
+ atomic_read(&ns->ns_bref), rc);
+ RETURN(ELDLM_NAMESPACE_EXISTS);
+ }
+ CDEBUG(D_DLMTRACE, "dlm namespace %s free done waiting\n",
+ ldlm_ns_name(ns));
+ }
+
+ RETURN(ELDLM_OK);
+}
- /* At shutdown time, don't call the cancellation callback */
- ldlm_namespace_cleanup(ns, 0);
+/**
+ * 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;
+ }
-#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);
+ spin_lock(&ns->ns_lock);
+ ns->ns_stopping = 1;
+ 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);
}
- }
-#endif
- if (ns->ns_refcount > 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);
- 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);
- CDEBUG(D_DLMTRACE,
- "dlm namespace %s free done waiting\n", ns->ns_name);
+ /*
+ * 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;
+}
- 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_PTR(ns);
-
- ldlm_put_ref(client, force);
+/**
+ * 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;
+ }
- RETURN(ELDLM_OK);
+ /* 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 at this time, otherwise
+ * this will cause issues related to using freed \a ns in poold
+ * thread. */
+ LASSERT(cfs_list_empty(&ns->ns_list_chain));
+ OBD_FREE_PTR(ns);
+ ldlm_put_ref();
+ EXIT;
}
-void ldlm_namespace_get_nolock(struct ldlm_namespace *ns)
+/**
+ * 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)
{
- LASSERT(ns->ns_refcount >= 0);
- ns->ns_refcount++;
+ ldlm_namespace_free_prior(ns, imp, force);
+ ldlm_namespace_free_post(ns);
}
+EXPORT_SYMBOL(ldlm_namespace_free);
void ldlm_namespace_get(struct ldlm_namespace *ns)
{
- spin_lock(&ns->ns_hash_lock);
- ldlm_namespace_get_nolock(ns);
- spin_unlock(&ns->ns_hash_lock);
+ atomic_inc(&ns->ns_bref);
}
+EXPORT_SYMBOL(ldlm_namespace_get);
-void ldlm_namespace_put_nolock(struct ldlm_namespace *ns, int wakeup)
+/* This is only for callers that care about refcount */
+int ldlm_namespace_get_return(struct ldlm_namespace *ns)
{
- LASSERT(ns->ns_refcount > 0);
- ns->ns_refcount--;
- if (ns->ns_refcount == 0 && wakeup)
- wake_up(&ns->ns_waitq);
+ return atomic_inc_return(&ns->ns_bref);
}
-void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup)
+void ldlm_namespace_put(struct ldlm_namespace *ns)
{
- spin_lock(&ns->ns_hash_lock);
- ldlm_namespace_put_nolock(ns, wakeup);
- spin_unlock(&ns->ns_hash_lock);
+ if (atomic_dec_and_lock(&ns->ns_bref, &ns->ns_lock)) {
+ wake_up(&ns->ns_waitq);
+ spin_unlock(&ns->ns_lock);
+ }
}
+EXPORT_SYMBOL(ldlm_namespace_put);
-/* Should be called under ldlm_namespace_lock(client) taken */
-void ldlm_namespace_move(struct ldlm_namespace *ns, ldlm_side_t client)
+/** Register \a ns in the list of namespaces */
+void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client)
{
- LASSERT(!list_empty(&ns->ns_list_chain));
- LASSERT_SEM_LOCKED(ldlm_namespace_lock(client));
- list_move_tail(&ns->ns_list_chain, ldlm_namespace_list(client));
+ mutex_lock(ldlm_namespace_lock(client));
+ LASSERT(cfs_list_empty(&ns->ns_list_chain));
+ cfs_list_add(&ns->ns_list_chain, ldlm_namespace_inactive_list(client));
+ ldlm_namespace_nr_inc(client);
+ mutex_unlock(ldlm_namespace_lock(client));
}
-/* Should be called under ldlm_namespace_lock(client) taken */
-struct ldlm_namespace *ldlm_namespace_first(ldlm_side_t client)
+/** Unregister \a ns from the list of namespaces. */
+void ldlm_namespace_unregister(struct ldlm_namespace *ns, 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,
- struct ldlm_namespace, ns_list_chain);
+ mutex_lock(ldlm_namespace_lock(client));
+ LASSERT(!cfs_list_empty(&ns->ns_list_chain));
+ /* Some asserts and possibly other parts of the code are 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);
+ ldlm_namespace_nr_dec(client);
+ mutex_unlock(ldlm_namespace_lock(client));
}
-static __u32 ldlm_hash_fn(struct ldlm_resource *parent,
- const struct ldlm_res_id *name)
-{
- __u32 hash = 0;
- int i;
- for (i = 0; i < RES_NAME_SIZE; i++)
- hash += name->name[i];
-
- hash += (__u32)((unsigned long)parent >> 4);
-
- return (hash & RES_HASH_MASK);
+/** Should be called with ldlm_namespace_lock(client) taken. */
+void ldlm_namespace_move_to_active_locked(struct ldlm_namespace *ns,
+ ldlm_side_t client)
+{
+ LASSERT(!cfs_list_empty(&ns->ns_list_chain));
+ LASSERT(mutex_is_locked(ldlm_namespace_lock(client)));
+ cfs_list_move_tail(&ns->ns_list_chain, ldlm_namespace_list(client));
}
-static struct ldlm_resource *ldlm_resource_new(void)
+/** Should be called with ldlm_namespace_lock(client) taken. */
+void ldlm_namespace_move_to_inactive_locked(struct ldlm_namespace *ns,
+ ldlm_side_t client)
{
- struct ldlm_resource *res;
-
- OBD_SLAB_ALLOC(res, ldlm_resource_slab, CFS_ALLOC_IO, sizeof *res);
- 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);
-
- /* one who creates the resource must unlock
- * the semaphore after lvb initialization */
- init_MUTEX_LOCKED(&res->lr_lvb_sem);
-
- return res;
+ LASSERT(!cfs_list_empty(&ns->ns_list_chain));
+ LASSERT(mutex_is_locked(ldlm_namespace_lock(client)));
+ cfs_list_move_tail(&ns->ns_list_chain,
+ ldlm_namespace_inactive_list(client));
}
-/* must be called with hash lock held */
-static struct ldlm_resource *
-ldlm_resource_find(struct ldlm_namespace *ns, const struct ldlm_res_id *name,
- __u32 hash)
+/** Should be called with ldlm_namespace_lock(client) taken. */
+struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client)
{
- struct list_head *bucket, *tmp;
- struct ldlm_resource *res;
-
- 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;
- }
-
- return NULL;
+ LASSERT(mutex_is_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);
}
-/* Args: locked namespace
- * Returns: newly-allocated, referenced, unlocked resource */
-static struct ldlm_resource *
-ldlm_resource_add(struct ldlm_namespace *ns, struct ldlm_resource *parent,
- const struct ldlm_res_id *name, __u32 hash, ldlm_type_t type)
+/** Create and initialize new resource. */
+static struct ldlm_resource *ldlm_resource_new(void)
{
- struct list_head *bucket;
- struct ldlm_resource *res, *old_res;
- ENTRY;
-
- LASSERTF(type >= LDLM_MIN_TYPE && type < LDLM_MAX_TYPE,
- "type: %d\n", type);
-
- res = ldlm_resource_new();
- if (!res)
- RETURN(NULL);
-
- res->lr_name = *name;
- res->lr_namespace = ns;
- 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) {
- /* someone won the race and added the resource before */
- ldlm_resource_getref(old_res);
- spin_unlock(&ns->ns_hash_lock);
- OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
- /* 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);
- }
- RETURN(old_res);
- }
-
- /* we won! let's add the resource */
- bucket = ns->ns_hash + hash;
- list_add(&res->lr_hash, bucket);
- ns->ns_resources++;
- ldlm_namespace_get_nolock(ns);
-
- 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);
-
- if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
- int rc;
-
- OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CREATE_RESOURCE, 2);
- rc = ns->ns_lvbo->lvbo_init(res);
- if (rc)
- CERROR("lvbo_init failed for resource "
- LPU64": rc %d\n", name->name[0], rc);
- /* we create resource with locked lr_lvb_sem */
- up(&res->lr_lvb_sem);
- }
-
- RETURN(res);
+ struct ldlm_resource *res;
+ int idx;
+
+ OBD_SLAB_ALLOC_PTR_GFP(res, ldlm_resource_slab, GFP_NOFS);
+ if (res == NULL)
+ return NULL;
+
+ CFS_INIT_LIST_HEAD(&res->lr_granted);
+ CFS_INIT_LIST_HEAD(&res->lr_converting);
+ CFS_INIT_LIST_HEAD(&res->lr_waiting);
+
+ /* 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;
+ }
+
+ atomic_set(&res->lr_refcount, 1);
+ spin_lock_init(&res->lr_lock);
+ lu_ref_init(&res->lr_reference);
+
+ /* The creator of the resource must unlock the mutex after LVB
+ * initialization. */
+ mutex_init(&res->lr_lvb_mutex);
+ mutex_lock(&res->lr_lvb_mutex);
+
+ return res;
}
-/* Args: unlocked namespace
- * Locks: takes and releases ns->ns_lock and res->lr_lock
- * Returns: referenced, unlocked ldlm_resource or NULL */
+/**
+ * Return a reference to resource with given name, creating it if necessary.
+ * Args: namespace with ns_lock unlocked
+ * 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)
{
- __u32 hash = ldlm_hash_fn(parent, name);
- struct ldlm_resource *res = NULL;
- ENTRY;
+ struct hlist_node *hnode;
+ struct ldlm_resource *res = NULL;
+ cfs_hash_bd_t bd;
+ __u64 version;
+ int ns_refcount = 0;
+ int rc;
LASSERT(ns != NULL);
- LASSERT(ns->ns_hash != NULL);
+ LASSERT(parent == NULL);
+ LASSERT(ns->ns_rs_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);
- }
- spin_unlock(&ns->ns_hash_lock);
-
- if (create == 0)
- RETURN(NULL);
-
- res = ldlm_resource_add(ns, parent, name, hash, type);
- RETURN(res);
+ 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);
+ GOTO(lvbo_init, res);
+ }
+
+ version = cfs_hash_bd_version_get(&bd);
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 0);
+
+ if (create == 0)
+ return ERR_PTR(-ENOENT);
+
+ LASSERTF(type >= LDLM_MIN_TYPE && type < LDLM_MAX_TYPE,
+ "type: %d\n", type);
+ res = ldlm_resource_new();
+ if (res == NULL)
+ return ERR_PTR(-ENOMEM);
+
+ 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;
+
+ 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 already added the resource. */
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
+ /* Clean lu_ref for failed resource. */
+ lu_ref_fini(&res->lr_reference);
+ /* We have taken lr_lvb_mutex. Drop it. */
+ mutex_unlock(&res->lr_lvb_mutex);
+ OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
+lvbo_init:
+ res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
+ /* Synchronize with regard to resource creation. */
+ if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
+ mutex_lock(&res->lr_lvb_mutex);
+ mutex_unlock(&res->lr_lvb_mutex);
+ }
+
+ if (unlikely(res->lr_lvb_len < 0)) {
+ rc = res->lr_lvb_len;
+ ldlm_resource_putref(res);
+ res = ERR_PTR(rc);
+ }
+ return res;
+ }
+ /* We won! Let's add the resource. */
+ cfs_hash_bd_add_locked(ns->ns_rs_hash, &bd, &res->lr_hash);
+ if (cfs_hash_bd_count_get(&bd) == 1)
+ ns_refcount = ldlm_namespace_get_return(ns);
+
+ cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1);
+ if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) {
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CREATE_RESOURCE, 2);
+ rc = ns->ns_lvbo->lvbo_init(res);
+ if (rc < 0) {
+ CERROR("%s: lvbo_init failed for resource "LPX64":"
+ LPX64": rc = %d\n", ns->ns_obd->obd_name,
+ name->name[0], name->name[1], rc);
+ if (res->lr_lvb_data) {
+ OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
+ res->lr_lvb_data = NULL;
+ }
+ res->lr_lvb_len = rc;
+ mutex_unlock(&res->lr_lvb_mutex);
+ ldlm_resource_putref(res);
+ return ERR_PTR(rc);
+ }
+ }
+
+ /* We create resource with locked lr_lvb_mutex. */
+ mutex_unlock(&res->lr_lvb_mutex);
+
+ /* Let's see if we happened to be the very first resource in this
+ * namespace. If so, and this is a client namespace, we need to move
+ * the namespace into the active namespaces list to be patrolled by
+ * the ldlm_poold. */
+ if (ns_is_client(ns) && ns_refcount == 1) {
+ mutex_lock(ldlm_namespace_lock(LDLM_NAMESPACE_CLIENT));
+ ldlm_namespace_move_to_active_locked(ns, LDLM_NAMESPACE_CLIENT);
+ mutex_unlock(ldlm_namespace_lock(LDLM_NAMESPACE_CLIENT));
+ }
+
+ return res;
}
+EXPORT_SYMBOL(ldlm_resource_get);
struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res)
{
- LASSERT(res != NULL);
- LASSERT(res != LP_POISON);
- atomic_inc(&res->lr_refcount);
- CDEBUG(D_INFO, "getref res: %p count: %d\n", res,
- atomic_read(&res->lr_refcount));
- return res;
+ LASSERT(res != NULL);
+ LASSERT(res != LP_POISON);
+ atomic_inc(&res->lr_refcount);
+ CDEBUG(D_INFO, "getref res: %p count: %d\n", res,
+ 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;
+ struct ldlm_ns_bucket *nsb = res->lr_ns_bucket;
- LASSERT_SPIN_LOCKED(&ns->ns_hash_lock);
-
- 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();
- }
-
- /* 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);
- 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);
- OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
- rc = 1;
- }
-
- RETURN(rc);
+ 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, 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);
+ return 1;
+ }
+ return 0;
}
+EXPORT_SYMBOL(ldlm_resource_putref);
/* 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);
- OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res);
- rc = 1;
- }
-
- RETURN(rc);
+ 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, atomic_read(&res->lr_refcount) - 1);
+
+ if (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);
+
+ 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,
+/**
+ * Add a lock into a given resource into specified lock list.
+ */
+void ldlm_resource_add_lock(struct ldlm_resource *res, cfs_list_t *head,
struct ldlm_lock *lock)
{
- check_res_locked(res);
+ check_res_locked(res);
- ldlm_resource_dump(D_OTHER, res);
- CDEBUG(D_OTHER, "About to add this lock:\n");
- ldlm_lock_dump(D_OTHER, lock, 0);
+ LDLM_DEBUG(lock, "About to add this lock:\n");
- if (lock->l_destroyed) {
- CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
- return;
- }
+ if (ldlm_is_destroyed(lock)) {
+ CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
+ 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);
}
+/**
+ * Insert a lock into resource after specified lock.
+ *
+ * Obtain resource description from the lock we are inserting after.
+ */
void ldlm_resource_insert_lock_after(struct ldlm_lock *original,
struct ldlm_lock *new)
{
check_res_locked(res);
- ldlm_resource_dump(D_OTHER, res);
- CDEBUG(D_OTHER, "About to insert this lock after %p:\n", original);
- ldlm_lock_dump(D_OTHER, new, 0);
+ ldlm_resource_dump(D_INFO, res);
+ LDLM_DEBUG(new, "About to insert this lock after %p:\n", original);
- if (new->l_destroyed) {
- CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
- goto out;
- }
+ if (ldlm_is_destroyed(new)) {
+ CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
+ 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);
- ldlm_unlink_lock_skiplist(lock);
- 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);
}
+EXPORT_SYMBOL(ldlm_resource_unlink_lock);
void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc)
{
desc->lr_name = res->lr_name;
}
+/**
+ * Print information about all locks in all namespaces on this node to debug
+ * log.
+ */
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(client));
+ mutex_lock(ldlm_namespace_lock(client));
- list_for_each(tmp, ldlm_namespace_list(client)) {
+ 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(client));
+ mutex_unlock(ldlm_namespace_lock(client));
}
+EXPORT_SYMBOL(ldlm_dump_all_namespaces);
-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;
-
- 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);
-
- 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);
+ struct ldlm_resource *res = cfs_hash_object(hs, hnode);
+ int level = (int)(unsigned long)arg;
- ldlm_resource_getref(res);
- spin_unlock(&ns->ns_hash_lock);
+ lock_res(res);
+ ldlm_resource_dump(level, res);
+ unlock_res(res);
- lock_res(res);
- ldlm_resource_dump(level, res);
- unlock_res(res);
+ return 0;
+}
- spin_lock(&ns->ns_hash_lock);
- tmp = tmp->next;
- ldlm_resource_putref_locked(res);
- }
- ns->ns_next_dump = cfs_time_shift(10);
- spin_unlock(&ns->ns_hash_lock);
+/**
+ * Print information about all locks in this namespace on this node to debug
+ * log.
+ */
+void ldlm_namespace_dump(int level, struct ldlm_namespace *ns)
+{
+ if (!((libcfs_debug | D_ERROR) & level))
+ return;
+
+ CDEBUG(level, "--- Namespace: %s (rc: %d, side: %s)\n",
+ ldlm_ns_name(ns), atomic_read(&ns->ns_bref),
+ ns_is_client(ns) ? "client" : "server");
+
+ if (cfs_time_before(cfs_time_current(), ns->ns_next_dump))
+ return;
+
+ cfs_hash_for_each_nolock(ns->ns_rs_hash,
+ ldlm_res_hash_dump,
+ (void *)(unsigned long)level);
+ spin_lock(&ns->ns_lock);
+ ns->ns_next_dump = cfs_time_shift(10);
+ spin_unlock(&ns->ns_lock);
}
+EXPORT_SYMBOL(ldlm_namespace_dump);
+/**
+ * Print information about all locks in this resource to debug log.
+ */
void ldlm_resource_dump(int level, struct ldlm_resource *res)
{
- struct list_head *tmp;
- int pos;
-
- CLASSERT(RES_NAME_SIZE == 4);
-
- if (!((libcfs_debug | D_ERROR) & level))
- return;
-
- 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));
-
- if (!list_empty(&res->lr_granted)) {
- pos = 0;
- CDEBUG(level, "Granted locks:\n");
- list_for_each(tmp, &res->lr_granted) {
- struct ldlm_lock *lock;
- lock = list_entry(tmp, struct ldlm_lock, l_res_link);
- ldlm_lock_dump(level, lock, ++pos);
+ struct ldlm_lock *lock;
+ unsigned int granted = 0;
+
+ CLASSERT(RES_NAME_SIZE == 4);
+
+ if (!((libcfs_debug | D_ERROR) & level))
+ return;
+
+ CDEBUG(level, "--- Resource: "DLDLMRES" (%p) refcount = %d\n",
+ PLDLMRES(res), res, atomic_read(&res->lr_refcount));
+
+ if (!cfs_list_empty(&res->lr_granted)) {
+ CDEBUG(level, "Granted locks (in reverse order):\n");
+ cfs_list_for_each_entry_reverse(lock, &res->lr_granted,
+ l_res_link) {
+ LDLM_DEBUG_LIMIT(level, lock, "###");
+ if (!(level & D_CANTMASK) &&
+ ++granted > ldlm_dump_granted_max) {
+ CDEBUG(level, "only dump %d granted locks to "
+ "avoid DDOS.\n", granted);
+ break;
+ }
}
}
- if (!list_empty(&res->lr_converting)) {
- pos = 0;
+ if (!cfs_list_empty(&res->lr_converting)) {
CDEBUG(level, "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(level, lock, ++pos);
- }
+ cfs_list_for_each_entry(lock, &res->lr_converting, l_res_link)
+ LDLM_DEBUG_LIMIT(level, lock, "###");
}
- if (!list_empty(&res->lr_waiting)) {
- pos = 0;
+ if (!cfs_list_empty(&res->lr_waiting)) {
CDEBUG(level, "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(level, lock, ++pos);
- }
+ cfs_list_for_each_entry(lock, &res->lr_waiting, l_res_link)
+ LDLM_DEBUG_LIMIT(level, lock, "###");
}
}