These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / staging / lustre / lustre / ldlm / ldlm_resource.c
index f750d42..c0a54bf 100644 (file)
@@ -47,8 +47,8 @@
 
 struct kmem_cache *ldlm_resource_slab, *ldlm_lock_slab;
 
-int ldlm_srv_namespace_nr = 0;
-int ldlm_cli_namespace_nr = 0;
+int ldlm_srv_namespace_nr;
+int ldlm_cli_namespace_nr;
 
 struct mutex ldlm_srv_namespace_lock;
 LIST_HEAD(ldlm_srv_namespace_list);
@@ -59,104 +59,104 @@ struct mutex ldlm_cli_namespace_lock;
  * inactive list */
 LIST_HEAD(ldlm_cli_active_namespace_list);
 /* Client namespaces that don't have any locks in them */
-LIST_HEAD(ldlm_cli_inactive_namespace_list);
+static LIST_HEAD(ldlm_cli_inactive_namespace_list);
 
-struct proc_dir_entry *ldlm_type_proc_dir = NULL;
-static struct proc_dir_entry *ldlm_ns_proc_dir = NULL;
-struct proc_dir_entry *ldlm_svc_proc_dir = NULL;
-
-extern unsigned int ldlm_cancel_unused_locks_before_replay;
+static struct dentry *ldlm_debugfs_dir;
+static struct dentry *ldlm_ns_debugfs_dir;
+struct dentry *ldlm_svc_debugfs_dir;
 
 /* during debug dump certain amount of granted locks for one resource to avoid
  * DDOS. */
-unsigned int ldlm_dump_granted_max = 256;
+static unsigned int ldlm_dump_granted_max = 256;
 
-#if defined(CONFIG_PROC_FS)
-static ssize_t lprocfs_wr_dump_ns(struct file *file, const char __user *buffer,
-                                 size_t count, loff_t *off)
+static ssize_t
+lprocfs_wr_dump_ns(struct file *file, const char __user *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;
 }
+
 LPROC_SEQ_FOPS_WR_ONLY(ldlm, dump_ns);
 
 LPROC_SEQ_FOPS_RW_TYPE(ldlm_rw, uint);
-LPROC_SEQ_FOPS_RO_TYPE(ldlm, uint);
 
-int ldlm_proc_setup(void)
+static struct lprocfs_vars ldlm_debugfs_list[] = {
+       { "dump_namespaces", &ldlm_dump_ns_fops, NULL, 0222 },
+       { "dump_granted_max", &ldlm_rw_uint_fops, &ldlm_dump_granted_max },
+       { NULL }
+};
+
+int ldlm_debugfs_setup(void)
 {
        int rc;
-       struct lprocfs_vars list[] = {
-               { "dump_namespaces", &ldlm_dump_ns_fops, NULL, 0222 },
-               { "dump_granted_max", &ldlm_rw_uint_fops,
-                 &ldlm_dump_granted_max },
-               { "cancel_unused_locks_before_replay", &ldlm_rw_uint_fops,
-                 &ldlm_cancel_unused_locks_before_replay },
-               { NULL } };
-       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)) {
+
+       ldlm_debugfs_dir = ldebugfs_register(OBD_LDLM_DEVICENAME,
+                                            debugfs_lustre_root,
+                                            NULL, NULL);
+       if (IS_ERR_OR_NULL(ldlm_debugfs_dir)) {
                CERROR("LProcFS failed in ldlm-init\n");
-               rc = PTR_ERR(ldlm_type_proc_dir);
+               rc = ldlm_debugfs_dir ? PTR_ERR(ldlm_debugfs_dir) : -ENOMEM;
                goto err;
        }
 
-       ldlm_ns_proc_dir = lprocfs_register("namespaces",
-                                           ldlm_type_proc_dir,
-                                           NULL, NULL);
-       if (IS_ERR(ldlm_ns_proc_dir)) {
+       ldlm_ns_debugfs_dir = ldebugfs_register("namespaces",
+                                               ldlm_debugfs_dir,
+                                               NULL, NULL);
+       if (IS_ERR_OR_NULL(ldlm_ns_debugfs_dir)) {
                CERROR("LProcFS failed in ldlm-init\n");
-               rc = PTR_ERR(ldlm_ns_proc_dir);
+               rc = ldlm_ns_debugfs_dir ? PTR_ERR(ldlm_ns_debugfs_dir)
+                                        : -ENOMEM;
                goto err_type;
        }
 
-       ldlm_svc_proc_dir = lprocfs_register("services",
-                                           ldlm_type_proc_dir,
-                                           NULL, NULL);
-       if (IS_ERR(ldlm_svc_proc_dir)) {
+       ldlm_svc_debugfs_dir = ldebugfs_register("services",
+                                                ldlm_debugfs_dir,
+                                                NULL, NULL);
+       if (IS_ERR_OR_NULL(ldlm_svc_debugfs_dir)) {
                CERROR("LProcFS failed in ldlm-init\n");
-               rc = PTR_ERR(ldlm_svc_proc_dir);
+               rc = ldlm_svc_debugfs_dir ? PTR_ERR(ldlm_svc_debugfs_dir)
+                                         : -ENOMEM;
                goto err_ns;
        }
 
-       rc = lprocfs_add_vars(ldlm_type_proc_dir, list, NULL);
+       rc = ldebugfs_add_vars(ldlm_debugfs_dir, ldlm_debugfs_list, NULL);
 
        return 0;
 
 err_ns:
-       lprocfs_remove(&ldlm_ns_proc_dir);
+       ldebugfs_remove(&ldlm_ns_debugfs_dir);
 err_type:
-       lprocfs_remove(&ldlm_type_proc_dir);
+       ldebugfs_remove(&ldlm_debugfs_dir);
 err:
-       ldlm_svc_proc_dir = NULL;
-       ldlm_type_proc_dir = NULL;
-       ldlm_ns_proc_dir = NULL;
+       ldlm_svc_debugfs_dir = NULL;
+       ldlm_ns_debugfs_dir = NULL;
+       ldlm_debugfs_dir = NULL;
        return rc;
 }
 
-void ldlm_proc_cleanup(void)
+void ldlm_debugfs_cleanup(void)
 {
-       if (ldlm_svc_proc_dir)
-               lprocfs_remove(&ldlm_svc_proc_dir);
+       if (!IS_ERR_OR_NULL(ldlm_svc_debugfs_dir))
+               ldebugfs_remove(&ldlm_svc_debugfs_dir);
 
-       if (ldlm_ns_proc_dir)
-               lprocfs_remove(&ldlm_ns_proc_dir);
+       if (!IS_ERR_OR_NULL(ldlm_ns_debugfs_dir))
+               ldebugfs_remove(&ldlm_ns_debugfs_dir);
 
-       if (ldlm_type_proc_dir)
-               lprocfs_remove(&ldlm_type_proc_dir);
+       if (!IS_ERR_OR_NULL(ldlm_debugfs_dir))
+               ldebugfs_remove(&ldlm_debugfs_dir);
 
-       ldlm_svc_proc_dir = NULL;
-       ldlm_type_proc_dir = NULL;
-       ldlm_ns_proc_dir = NULL;
+       ldlm_svc_debugfs_dir = NULL;
+       ldlm_ns_debugfs_dir = NULL;
+       ldlm_debugfs_dir = NULL;
 }
 
-static int lprocfs_ns_resources_seq_show(struct seq_file *m, void *v)
+static ssize_t resource_count_show(struct kobject *kobj, struct attribute *attr,
+                                  char *buf)
 {
-       struct ldlm_namespace *ns  = m->private;
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
        __u64             res = 0;
        struct cfs_hash_bd        bd;
        int                 i;
@@ -164,46 +164,56 @@ static int lprocfs_ns_resources_seq_show(struct seq_file *m, void *v)
        /* result is not strictly consistent */
        cfs_hash_for_each_bucket(ns->ns_rs_hash, &bd, i)
                res += cfs_hash_bd_count_get(&bd);
-       return lprocfs_rd_u64(m, &res);
+       return sprintf(buf, "%lld\n", res);
 }
-LPROC_SEQ_FOPS_RO(lprocfs_ns_resources);
+LUSTRE_RO_ATTR(resource_count);
 
-static int lprocfs_ns_locks_seq_show(struct seq_file *m, void *v)
+static ssize_t lock_count_show(struct kobject *kobj, struct attribute *attr,
+                              char *buf)
 {
-       struct ldlm_namespace *ns = m->private;
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
        __u64             locks;
 
        locks = lprocfs_stats_collector(ns->ns_stats, LDLM_NSS_LOCKS,
                                        LPROCFS_FIELDS_FLAGS_SUM);
-       return lprocfs_rd_u64(m, &locks);
+       return sprintf(buf, "%lld\n", locks);
+}
+LUSTRE_RO_ATTR(lock_count);
+
+static ssize_t lock_unused_count_show(struct kobject *kobj,
+                                     struct attribute *attr,
+                                     char *buf)
+{
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
+
+       return sprintf(buf, "%d\n", ns->ns_nr_unused);
 }
-LPROC_SEQ_FOPS_RO(lprocfs_ns_locks);
+LUSTRE_RO_ATTR(lock_unused_count);
 
-static int lprocfs_lru_size_seq_show(struct seq_file *m, void *v)
+static ssize_t lru_size_show(struct kobject *kobj, struct attribute *attr,
+                            char *buf)
 {
-       struct ldlm_namespace *ns = m->private;
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
        __u32 *nr = &ns->ns_max_unused;
 
        if (ns_connect_lru_resize(ns))
                nr = &ns->ns_nr_unused;
-       return lprocfs_rd_uint(m, nr);
+       return sprintf(buf, "%u", *nr);
 }
 
-static ssize_t lprocfs_lru_size_seq_write(struct file *file,
-                                       const char __user *buffer,
-                                       size_t count, loff_t *off)
+static ssize_t lru_size_store(struct kobject *kobj, struct attribute *attr,
+                             const char *buffer, size_t count)
 {
-       struct ldlm_namespace *ns = ((struct seq_file *)file->private_data)->private;
-       char dummy[MAX_STRING_SIZE + 1];
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
        unsigned long tmp;
        int lru_resize;
        int err;
 
-       dummy[MAX_STRING_SIZE] = '\0';
-       if (copy_from_user(dummy, buffer, MAX_STRING_SIZE))
-               return -EFAULT;
-
-       if (strncmp(dummy, "clear", 5) == 0) {
+       if (strncmp(buffer, "clear", 5) == 0) {
                CDEBUG(D_DLMTRACE,
                       "dropping all unused locks from namespace %s\n",
                       ldlm_ns_name(ns));
@@ -229,9 +239,9 @@ static ssize_t lprocfs_lru_size_seq_write(struct file *file,
                return count;
        }
 
-       err = kstrtoul(dummy, 10, &tmp);
+       err = kstrtoul(buffer, 10, &tmp);
        if (err != 0) {
-               CERROR("invalid value written\n");
+               CERROR("lru_size: invalid value written\n");
                return -EINVAL;
        }
        lru_resize = (tmp == 0);
@@ -277,25 +287,56 @@ static ssize_t lprocfs_lru_size_seq_write(struct file *file,
 
        return count;
 }
-LPROC_SEQ_FOPS(lprocfs_lru_size);
+LUSTRE_RW_ATTR(lru_size);
+
+static ssize_t lru_max_age_show(struct kobject *kobj, struct attribute *attr,
+                               char *buf)
+{
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
+
+       return sprintf(buf, "%u", ns->ns_max_age);
+}
+
+static ssize_t lru_max_age_store(struct kobject *kobj, struct attribute *attr,
+                                const char *buffer, size_t count)
+{
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
+       unsigned long tmp;
+       int err;
+
+       err = kstrtoul(buffer, 10, &tmp);
+       if (err != 0)
+               return -EINVAL;
+
+       ns->ns_max_age = tmp;
+
+       return count;
+}
+LUSTRE_RW_ATTR(lru_max_age);
 
-static int lprocfs_elc_seq_show(struct seq_file *m, void *v)
+static ssize_t early_lock_cancel_show(struct kobject *kobj,
+                                     struct attribute *attr,
+                                     char *buf)
 {
-       struct ldlm_namespace *ns = m->private;
-       unsigned int supp = ns_connect_cancelset(ns);
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
 
-       return lprocfs_rd_uint(m, &supp);
+       return sprintf(buf, "%d\n", ns_connect_cancelset(ns));
 }
 
-static ssize_t lprocfs_elc_seq_write(struct file *file,
-                               const char __user *buffer,
-                               size_t count, loff_t *off)
+static ssize_t early_lock_cancel_store(struct kobject *kobj,
+                                      struct attribute *attr,
+                                      const char *buffer,
+                                      size_t count)
 {
-       struct ldlm_namespace *ns = ((struct seq_file *)file->private_data)->private;
-       unsigned int supp = -1;
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
+       unsigned long supp = -1;
        int rc;
 
-       rc = lprocfs_wr_uint(file, buffer, count, &supp);
+       rc = kstrtoul(buffer, 10, &supp);
        if (rc < 0)
                return rc;
 
@@ -305,91 +346,99 @@ static ssize_t lprocfs_elc_seq_write(struct file *file,
                ns->ns_connect_flags |= OBD_CONNECT_CANCELSET;
        return count;
 }
-LPROC_SEQ_FOPS(lprocfs_elc);
+LUSTRE_RW_ATTR(early_lock_cancel);
+
+/* These are for namespaces in /sys/fs/lustre/ldlm/namespaces/ */
+static struct attribute *ldlm_ns_attrs[] = {
+       &lustre_attr_resource_count.attr,
+       &lustre_attr_lock_count.attr,
+       &lustre_attr_lock_unused_count.attr,
+       &lustre_attr_lru_size.attr,
+       &lustre_attr_lru_max_age.attr,
+       &lustre_attr_early_lock_cancel.attr,
+       NULL,
+};
+
+static void ldlm_ns_release(struct kobject *kobj)
+{
+       struct ldlm_namespace *ns = container_of(kobj, struct ldlm_namespace,
+                                                ns_kobj);
+       complete(&ns->ns_kobj_unregister);
+}
+
+static struct kobj_type ldlm_ns_ktype = {
+       .default_attrs  = ldlm_ns_attrs,
+       .sysfs_ops      = &lustre_sysfs_ops,
+       .release        = ldlm_ns_release,
+};
 
-void ldlm_namespace_proc_unregister(struct ldlm_namespace *ns)
+static void ldlm_namespace_debugfs_unregister(struct ldlm_namespace *ns)
 {
-       if (ns->ns_proc_dir_entry == NULL)
+       if (IS_ERR_OR_NULL(ns->ns_debugfs_entry))
                CERROR("dlm namespace %s has no procfs dir?\n",
                       ldlm_ns_name(ns));
        else
-               lprocfs_remove(&ns->ns_proc_dir_entry);
+               ldebugfs_remove(&ns->ns_debugfs_entry);
 
        if (ns->ns_stats != NULL)
                lprocfs_free_stats(&ns->ns_stats);
 }
 
-#define LDLM_NS_ADD_VAR(name, var, ops)                                \
-       do {                                                    \
-               snprintf(lock_name, MAX_STRING_SIZE, name);     \
-               lock_vars[0].data = var;                        \
-               lock_vars[0].fops = ops;                        \
-               lprocfs_add_vars(ns_pde, lock_vars, NULL);      \
-       } while (0)
+static void ldlm_namespace_sysfs_unregister(struct ldlm_namespace *ns)
+{
+       kobject_put(&ns->ns_kobj);
+       wait_for_completion(&ns->ns_kobj_unregister);
+}
 
-int ldlm_namespace_proc_register(struct ldlm_namespace *ns)
+static int ldlm_namespace_sysfs_register(struct ldlm_namespace *ns)
 {
-       struct lprocfs_vars lock_vars[2];
-       char lock_name[MAX_STRING_SIZE + 1];
-       struct proc_dir_entry *ns_pde;
+       int err;
 
-       LASSERT(ns != 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_kobj.kset = ldlm_ns_kset;
+       init_completion(&ns->ns_kobj_unregister);
+       err = kobject_init_and_add(&ns->ns_kobj, &ldlm_ns_ktype, NULL,
+                                  "%s", ldlm_ns_name(ns));
 
        ns->ns_stats = lprocfs_alloc_stats(LDLM_NSS_LAST, 0);
-       if (ns->ns_stats == NULL)
+       if (ns->ns_stats == NULL) {
+               kobject_put(&ns->ns_kobj);
                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;
+       return err;
+}
 
-       LDLM_NS_ADD_VAR("resource_count", ns, &lprocfs_ns_resources_fops);
-       LDLM_NS_ADD_VAR("lock_count", ns, &lprocfs_ns_locks_fops);
+static int ldlm_namespace_debugfs_register(struct ldlm_namespace *ns)
+{
+       struct dentry *ns_entry;
 
-       if (ns_is_client(ns)) {
-               LDLM_NS_ADD_VAR("lock_unused_count", &ns->ns_nr_unused,
-                               &ldlm_uint_fops);
-               LDLM_NS_ADD_VAR("lru_size", ns, &lprocfs_lru_size_fops);
-               LDLM_NS_ADD_VAR("lru_max_age", &ns->ns_max_age,
-                               &ldlm_rw_uint_fops);
-               LDLM_NS_ADD_VAR("early_lock_cancel", ns, &lprocfs_elc_fops);
+       if (!IS_ERR_OR_NULL(ns->ns_debugfs_entry)) {
+               ns_entry = ns->ns_debugfs_entry;
        } else {
-               LDLM_NS_ADD_VAR("ctime_age_limit", &ns->ns_ctime_age_limit,
-                               &ldlm_rw_uint_fops);
-               LDLM_NS_ADD_VAR("lock_timeouts", &ns->ns_timeouts,
-                               &ldlm_uint_fops);
-               LDLM_NS_ADD_VAR("max_nolock_bytes", &ns->ns_max_nolock_size,
-                               &ldlm_rw_uint_fops);
-               LDLM_NS_ADD_VAR("contention_seconds", &ns->ns_contention_time,
-                               &ldlm_rw_uint_fops);
-               LDLM_NS_ADD_VAR("contended_locks", &ns->ns_contended_locks,
-                               &ldlm_rw_uint_fops);
-               LDLM_NS_ADD_VAR("max_parallel_ast", &ns->ns_max_parallel_ast,
-                               &ldlm_rw_uint_fops);
+               ns_entry = debugfs_create_dir(ldlm_ns_name(ns),
+                                             ldlm_ns_debugfs_dir);
+               if (ns_entry == NULL)
+                       return -ENOMEM;
+               ns->ns_debugfs_entry = ns_entry;
        }
+
        return 0;
 }
-#undef MAX_STRING_SIZE
-#else /* CONFIG_PROC_FS */
 
-#define ldlm_namespace_proc_unregister(ns)      ({; })
-#define ldlm_namespace_proc_register(ns)       ({0; })
+#undef MAX_STRING_SIZE
 
-#endif /* CONFIG_PROC_FS */
+static struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res)
+{
+       LASSERT(res);
+       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;
+}
 
 static unsigned ldlm_res_hop_hash(struct cfs_hash *hs,
                                  const void *key, unsigned mask)
@@ -482,26 +531,26 @@ static void ldlm_res_hop_put(struct cfs_hash *hs, struct hlist_node *hnode)
        ldlm_resource_putref(res);
 }
 
-cfs_hash_ops_t ldlm_ns_hash_ops = {
+static struct cfs_hash_ops ldlm_ns_hash_ops = {
        .hs_hash        = ldlm_res_hop_hash,
-       .hs_key  = ldlm_res_hop_key,
+       .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_get         = ldlm_res_hop_get_locked,
        .hs_put_locked  = ldlm_res_hop_put_locked,
-       .hs_put  = ldlm_res_hop_put
+       .hs_put         = ldlm_res_hop_put
 };
 
-cfs_hash_ops_t ldlm_ns_fid_hash_ops = {
+static struct cfs_hash_ops ldlm_ns_fid_hash_ops = {
        .hs_hash        = ldlm_res_hop_fid_hash,
-       .hs_key  = ldlm_res_hop_key,
+       .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_get         = ldlm_res_hop_get_locked,
        .hs_put_locked  = ldlm_res_hop_put_locked,
-       .hs_put  = ldlm_res_hop_put
+       .hs_put         = ldlm_res_hop_put
 };
 
 struct ldlm_ns_hash_def {
@@ -511,10 +560,10 @@ struct ldlm_ns_hash_def {
        /** hash bits */
        unsigned        nsd_all_bits;
        /** hash operations */
-       cfs_hash_ops_t *nsd_hops;
+       struct cfs_hash_ops *nsd_hops;
 };
 
-struct ldlm_ns_hash_def ldlm_ns_hash_defs[] = {
+static struct ldlm_ns_hash_def ldlm_ns_hash_defs[] = {
        {
                .nsd_type       = LDLM_NS_TYPE_MDC,
                .nsd_bkt_bits   = 11,
@@ -556,6 +605,17 @@ struct ldlm_ns_hash_def ldlm_ns_hash_defs[] = {
        },
 };
 
+/** Register \a ns in the list of namespaces */
+static void ldlm_namespace_register(struct ldlm_namespace *ns,
+                                   ldlm_side_t client)
+{
+       mutex_lock(ldlm_namespace_lock(client));
+       LASSERT(list_empty(&ns->ns_list_chain));
+       list_add(&ns->ns_list_chain, &ldlm_cli_inactive_namespace_list);
+       ldlm_namespace_nr_inc(client);
+       mutex_unlock(ldlm_namespace_lock(client));
+}
+
 /**
  * Create and initialize new empty namespace.
  */
@@ -590,7 +650,7 @@ struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
                        break;
        }
 
-       OBD_ALLOC_PTR(ns);
+       ns = kzalloc(sizeof(*ns), GFP_NOFS);
        if (!ns)
                goto out_ref;
 
@@ -623,25 +683,26 @@ struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
        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);
+
+       rc = ldlm_namespace_sysfs_register(ns);
        if (rc != 0) {
-               CERROR("Can't initialize ns proc, rc %d\n", rc);
+               CERROR("Can't initialize ns sysfs, rc %d\n", rc);
                goto out_hash;
        }
 
+       rc = ldlm_namespace_debugfs_register(ns);
+       if (rc != 0) {
+               CERROR("Can't initialize ns proc, rc %d\n", rc);
+               goto out_sysfs;
+       }
+
        idx = ldlm_namespace_nr_read(client);
        rc = ldlm_pool_init(&ns->ns_pool, ns, idx, client);
        if (rc) {
@@ -652,12 +713,14 @@ struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
        ldlm_namespace_register(ns, client);
        return ns;
 out_proc:
-       ldlm_namespace_proc_unregister(ns);
+       ldlm_namespace_debugfs_unregister(ns);
+out_sysfs:
+       ldlm_namespace_sysfs_unregister(ns);
        ldlm_namespace_cleanup(ns, 0);
 out_hash:
        cfs_hash_putref(ns->ns_rs_hash);
 out_ns:
-       OBD_FREE_PTR(ns);
+       kfree(ns);
 out_ref:
        ldlm_put_ref();
        return NULL;
@@ -678,11 +741,12 @@ static void cleanup_resource(struct ldlm_resource *res, struct list_head *q,
                             __u64 flags)
 {
        struct list_head *tmp;
-       int rc = 0, client = ns_is_client(ldlm_res_to_ns(res));
+       int rc = 0;
        bool local_only = !!(flags & LDLM_FL_LOCAL_ONLY);
 
        do {
                struct ldlm_lock *lock = NULL;
+               struct lustre_handle lockh;
 
                /* First, we look for non-cleaned-yet lock
                 * all cleaned locks are marked by CLEANED flag. */
@@ -727,20 +791,11 @@ static void cleanup_resource(struct ldlm_resource *res, struct list_head *q,
                        continue;
                }
 
-               if (client) {
-                       struct lustre_handle lockh;
-
-                       unlock_res(res);
-                       ldlm_lock2handle(lock, &lockh);
-                       rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
-                       if (rc)
-                               CERROR("ldlm_cli_cancel: %d\n", rc);
-               } else {
-                       ldlm_resource_unlink_lock(lock);
-                       unlock_res(res);
-                       LDLM_DEBUG(lock, "Freeing a lock still held by a client node");
-                       ldlm_lock_destroy(lock);
-               }
+               unlock_res(res);
+               ldlm_lock2handle(lock, &lockh);
+               rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
+               if (rc)
+                       CERROR("ldlm_cli_cancel: %d\n", rc);
                LDLM_LOCK_RELEASE(lock);
        } while (1);
 }
@@ -752,7 +807,6 @@ static int ldlm_resource_clean(struct cfs_hash *hs, struct cfs_hash_bd *bd,
        __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;
@@ -881,6 +935,20 @@ void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
        }
 }
 
+/** Unregister \a ns from the list of namespaces. */
+static void ldlm_namespace_unregister(struct ldlm_namespace *ns,
+                                     ldlm_side_t client)
+{
+       mutex_lock(ldlm_namespace_lock(client));
+       LASSERT(!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. */
+       list_del_init(&ns->ns_list_chain);
+       ldlm_namespace_nr_dec(client);
+       mutex_unlock(ldlm_namespace_lock(client));
+}
+
 /**
  * Performs freeing memory structures related to \a ns. This is only done
  * when ldlm_namespce_free_prior() successfully removed all resources
@@ -898,43 +966,17 @@ void ldlm_namespace_free_post(struct ldlm_namespace *ns)
         * Removing it after @dir may cause oops. */
        ldlm_pool_fini(&ns->ns_pool);
 
-       ldlm_namespace_proc_unregister(ns);
+       ldlm_namespace_debugfs_unregister(ns);
+       ldlm_namespace_sysfs_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(list_empty(&ns->ns_list_chain));
-       OBD_FREE_PTR(ns);
+       kfree(ns);
        ldlm_put_ref();
 }
 
-/**
- * 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);
-}
-EXPORT_SYMBOL(ldlm_namespace_free);
-
 void ldlm_namespace_get(struct ldlm_namespace *ns)
 {
        atomic_inc(&ns->ns_bref);
@@ -942,7 +984,7 @@ void ldlm_namespace_get(struct ldlm_namespace *ns)
 EXPORT_SYMBOL(ldlm_namespace_get);
 
 /* This is only for callers that care about refcount */
-int ldlm_namespace_get_return(struct ldlm_namespace *ns)
+static int ldlm_namespace_get_return(struct ldlm_namespace *ns)
 {
        return atomic_inc_return(&ns->ns_bref);
 }
@@ -956,29 +998,6 @@ void ldlm_namespace_put(struct ldlm_namespace *ns)
 }
 EXPORT_SYMBOL(ldlm_namespace_put);
 
-/** Register \a ns in the list of namespaces */
-void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client)
-{
-       mutex_lock(ldlm_namespace_lock(client));
-       LASSERT(list_empty(&ns->ns_list_chain));
-       list_add(&ns->ns_list_chain, ldlm_namespace_inactive_list(client));
-       ldlm_namespace_nr_inc(client);
-       mutex_unlock(ldlm_namespace_lock(client));
-}
-
-/** Unregister \a ns from the list of namespaces. */
-void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client)
-{
-       mutex_lock(ldlm_namespace_lock(client));
-       LASSERT(!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. */
-       list_del_init(&ns->ns_list_chain);
-       ldlm_namespace_nr_dec(client);
-       mutex_unlock(ldlm_namespace_lock(client));
-}
-
 /** Should be called with ldlm_namespace_lock(client) taken. */
 void ldlm_namespace_move_to_active_locked(struct ldlm_namespace *ns,
                                          ldlm_side_t client)
@@ -994,8 +1013,7 @@ void ldlm_namespace_move_to_inactive_locked(struct ldlm_namespace *ns,
 {
        LASSERT(!list_empty(&ns->ns_list_chain));
        LASSERT(mutex_is_locked(ldlm_namespace_lock(client)));
-       list_move_tail(&ns->ns_list_chain,
-                      ldlm_namespace_inactive_list(client));
+       list_move_tail(&ns->ns_list_chain, &ldlm_cli_inactive_namespace_list);
 }
 
 /** Should be called with ldlm_namespace_lock(client) taken. */
@@ -1013,12 +1031,11 @@ static struct ldlm_resource *ldlm_resource_new(void)
        struct ldlm_resource *res;
        int idx;
 
-       OBD_SLAB_ALLOC_PTR_GFP(res, ldlm_resource_slab, GFP_NOFS);
+       res = kmem_cache_alloc(ldlm_resource_slab, GFP_NOFS | __GFP_ZERO);
        if (res == NULL)
                return NULL;
 
        INIT_LIST_HEAD(&res->lr_granted);
-       INIT_LIST_HEAD(&res->lr_converting);
        INIT_LIST_HEAD(&res->lr_waiting);
 
        /* Initialize interval trees for each lock mode. */
@@ -1107,7 +1124,7 @@ ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent,
                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));
+               kmem_cache_free(ldlm_resource_slab, res);
 
                res = hlist_entry(hnode, struct ldlm_resource, lr_hash);
                /* Synchronize with regard to resource creation. */
@@ -1137,10 +1154,8 @@ ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent,
                        CERROR("%s: lvbo_init failed for resource %#llx:%#llx: 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;
-                       }
+                       kfree(res->lr_lvb_data);
+                       res->lr_lvb_data = NULL;
                        res->lr_lvb_len = rc;
                        mutex_unlock(&res->lr_lvb_mutex);
                        ldlm_resource_putref(res);
@@ -1155,7 +1170,7 @@ ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent,
         * 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) {
+       if (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));
@@ -1165,16 +1180,6 @@ ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent,
 }
 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;
-}
-
 static void __ldlm_resource_putref_final(struct cfs_hash_bd *bd,
                                         struct ldlm_resource *res)
 {
@@ -1185,11 +1190,6 @@ static void __ldlm_resource_putref_final(struct cfs_hash_bd *bd,
                LBUG();
        }
 
-       if (!list_empty(&res->lr_converting)) {
-               ldlm_resource_dump(D_ERROR, res);
-               LBUG();
-       }
-
        if (!list_empty(&res->lr_waiting)) {
                ldlm_resource_dump(D_ERROR, res);
                LBUG();
@@ -1218,7 +1218,7 @@ int ldlm_resource_putref(struct ldlm_resource *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));
+               kmem_cache_free(ldlm_resource_slab, res);
                return 1;
        }
        return 0;
@@ -1248,7 +1248,7 @@ int ldlm_resource_putref_locked(struct ldlm_resource *res)
                 */
                if (ns->ns_lvbo && ns->ns_lvbo->lvbo_free)
                        ns->ns_lvbo->lvbo_free(res);
-               OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof(*res));
+               kmem_cache_free(ldlm_resource_slab, res);
 
                cfs_hash_bd_lock(ns->ns_rs_hash, &bd, 1);
                return 1;
@@ -1276,32 +1276,6 @@ void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
        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)
-{
-       struct ldlm_resource *res = original->l_resource;
-
-       check_res_locked(res);
-
-       ldlm_resource_dump(D_INFO, res);
-       LDLM_DEBUG(new, "About to insert this lock after %p:\n", original);
-
-       if (new->l_flags & LDLM_FL_DESTROYED) {
-               CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n");
-               goto out;
-       }
-
-       LASSERT(list_empty(&new->l_res_link));
-
-       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;
@@ -1367,9 +1341,8 @@ 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");
+       CDEBUG(level, "--- Namespace: %s (rc: %d, side: client)\n",
+              ldlm_ns_name(ns), atomic_read(&ns->ns_bref));
 
        if (time_before(cfs_time_current(), ns->ns_next_dump))
                return;
@@ -1412,11 +1385,6 @@ void ldlm_resource_dump(int level, struct ldlm_resource *res)
                        }
                }
        }
-       if (!list_empty(&res->lr_converting)) {
-               CDEBUG(level, "Converting locks:\n");
-               list_for_each_entry(lock, &res->lr_converting, l_res_link)
-                       LDLM_DEBUG_LIMIT(level, lock, "###");
-       }
        if (!list_empty(&res->lr_waiting)) {
                CDEBUG(level, "Waiting locks:\n");
                list_for_each_entry(lock, &res->lr_waiting, l_res_link)