#define DEBUG_SUBSYSTEM S_LDLM
+#include <linux/kthread.h>
#include <libcfs/libcfs.h>
#include <lustre_dlm.h>
#include <obd_class.h>
};
struct ldlm_bl_work_item {
- struct list_head blwi_entry;
- struct ldlm_namespace *blwi_ns;
- struct ldlm_lock_desc blwi_ld;
- struct ldlm_lock *blwi_lock;
- struct list_head blwi_head;
- int blwi_count;
- struct completion blwi_comp;
- ldlm_cancel_flags_t blwi_flags;
- int blwi_mem_pressure;
+ struct list_head blwi_entry;
+ struct ldlm_namespace *blwi_ns;
+ struct ldlm_lock_desc blwi_ld;
+ struct ldlm_lock *blwi_lock;
+ struct list_head blwi_head;
+ int blwi_count;
+ struct completion blwi_comp;
+ enum ldlm_cancel_flags blwi_flags;
+ int blwi_mem_pressure;
};
#ifdef HAVE_SERVER_SUPPORT
return 1;
}
+static void ldlm_add_blocked_lock(struct ldlm_lock *lock)
+{
+ spin_lock_bh(&lock->l_export->exp_bl_list_lock);
+ if (list_empty(&lock->l_exp_list)) {
+ if (lock->l_granted_mode != lock->l_req_mode)
+ list_add_tail(&lock->l_exp_list,
+ &lock->l_export->exp_bl_list);
+ else
+ list_add(&lock->l_exp_list,
+ &lock->l_export->exp_bl_list);
+ }
+ spin_unlock_bh(&lock->l_export->exp_bl_list_lock);
+
+ /* A blocked lock is added. Adjust the position in
+ * the stale list if the export is in the list.
+ * If export is stale and not in the list - it is being
+ * processed and will be placed on the right position
+ * on obd_stale_export_put(). */
+ if (!list_empty(&lock->l_export->exp_stale_list))
+ obd_stale_export_adjust(lock->l_export);
+}
+
static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
{
int ret;
/* NB: must be called with hold of lock_res_and_lock() */
LASSERT(ldlm_is_res_locked(lock));
- ldlm_set_waited(lock);
-
LASSERT(!ldlm_is_cancel_on_block(lock));
+ /* Do not put cross-MDT lock in the waiting list, since we
+ * will not evict it due to timeout for now */
+ if (lock->l_export != NULL &&
+ (exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS))
+ return 0;
+
spin_lock_bh(&waiting_locks_spinlock);
+ if (ldlm_is_cancel(lock)) {
+ spin_unlock_bh(&waiting_locks_spinlock);
+ return 0;
+ }
+
if (ldlm_is_destroyed(lock)) {
static cfs_time_t next;
+
spin_unlock_bh(&waiting_locks_spinlock);
LDLM_ERROR(lock, "not waiting on destroyed lock (bug 5653)");
if (cfs_time_after(cfs_time_current(), next)) {
return 0;
}
+ ldlm_set_waited(lock);
lock->l_last_activity = cfs_time_current_sec();
ret = __ldlm_add_waiting_lock(lock, timeout);
if (ret) {
}
spin_unlock_bh(&waiting_locks_spinlock);
- if (ret) {
- spin_lock_bh(&lock->l_export->exp_bl_list_lock);
- if (list_empty(&lock->l_exp_list))
- list_add(&lock->l_exp_list,
- &lock->l_export->exp_bl_list);
- spin_unlock_bh(&lock->l_export->exp_bl_list_lock);
- }
+ if (ret)
+ ldlm_add_blocked_lock(lock);
LDLM_DEBUG(lock, "%sadding to wait list(timeout: %d, AT: %s)",
ret == 0 ? "not re-" : "", timeout,
spin_lock_bh(&waiting_locks_spinlock);
ret = __ldlm_del_waiting_lock(lock);
+ ldlm_clear_waited(lock);
spin_unlock_bh(&waiting_locks_spinlock);
/* remove the lock out of export blocking list */
return 0;
}
+ if (exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) {
+ /* We don't have a "waiting locks list" on OSP. */
+ LDLM_DEBUG(lock, "MDS-MDS lock: no-op");
+ return 0;
+ }
+
spin_lock_bh(&waiting_locks_spinlock);
if (list_empty(&lock->l_pending_chain)) {
req->rq_interpret_reply = ldlm_cb_interpret;
lock_res_and_lock(lock);
- if (lock->l_granted_mode != lock->l_req_mode) {
- /* this blocking AST will be communicated as part of the
- * completion AST instead */
+ if (ldlm_is_destroyed(lock)) {
+ /* What's the point? */
unlock_res_and_lock(lock);
-
ptlrpc_req_finished(req);
- LDLM_DEBUG(lock, "lock not granted, not sending blocking AST");
RETURN(0);
}
- if (ldlm_is_destroyed(lock)) {
- /* What's the point? */
+ if (lock->l_granted_mode != lock->l_req_mode) {
+ /* this blocking AST will be communicated as part of the
+ * completion AST instead */
+ ldlm_add_blocked_lock(lock);
+ ldlm_set_waited(lock);
unlock_res_and_lock(lock);
+
ptlrpc_req_finished(req);
+ LDLM_DEBUG(lock, "lock not granted, not sending blocking AST");
RETURN(0);
}
* service threads to carry out client lock enqueueing requests.
*/
int ldlm_handle_enqueue0(struct ldlm_namespace *ns,
- struct ptlrpc_request *req,
- const struct ldlm_request *dlm_req,
- const struct ldlm_callback_suite *cbs)
+ struct ptlrpc_request *req,
+ const struct ldlm_request *dlm_req,
+ const struct ldlm_callback_suite *cbs)
{
- struct ldlm_reply *dlm_rep;
+ struct ldlm_reply *dlm_rep;
__u64 flags;
- ldlm_error_t err = ELDLM_OK;
- struct ldlm_lock *lock = NULL;
- void *cookie = NULL;
- int rc = 0;
+ enum ldlm_error err = ELDLM_OK;
+ struct ldlm_lock *lock = NULL;
+ void *cookie = NULL;
+ int rc = 0;
struct ldlm_resource *res = NULL;
- ENTRY;
+ ENTRY;
- LDLM_DEBUG_NOLOCK("server-side enqueue handler START");
+ LDLM_DEBUG_NOLOCK("server-side enqueue handler START");
ldlm_request_cancel(req, dlm_req, LDLM_ENQUEUE_CANCEL_OFF, LATF_SKIP);
flags = ldlm_flags_from_wire(dlm_req->lock_flags);
- LASSERT(req->rq_export);
+ LASSERT(req->rq_export);
if (ptlrpc_req2svc(req)->srv_stats != NULL)
ldlm_svc_get_eopc(dlm_req, ptlrpc_req2svc(req)->srv_stats);
flags |= LDLM_FL_RESENT;
GOTO(existing_lock, rc = 0);
}
- }
+ } else {
+ if (ldlm_reclaim_full()) {
+ DEBUG_REQ(D_DLMTRACE, req, "Too many granted locks, "
+ "reject current enqueue request and let the "
+ "client retry later.\n");
+ GOTO(out, rc = -EINPROGRESS);
+ }
+ }
/* The lock's callback data might be set in the policy function */
lock = ldlm_lock_create(ns, &dlm_req->lock_desc.l_resource.lr_name,
ldlm_lock2desc(lock, &dlm_rep->lock_desc);
ldlm_lock2handle(lock, &dlm_rep->lock_handle);
+ if (lock && lock->l_resource->lr_type == LDLM_EXTENT)
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_BL_EVICT, 6);
+
/* We never send a blocking AST until the lock is granted, but
* we can tell it right now */
lock_res_and_lock(lock);
}
if (rc != 0) {
- lock_res_and_lock(lock);
- ldlm_resource_unlink_lock(lock);
- ldlm_lock_destroy_nolock(lock);
- unlock_res_and_lock(lock);
- }
+ if (lock->l_export) {
+ ldlm_lock_cancel(lock);
+ } else {
+ lock_res_and_lock(lock);
+ ldlm_resource_unlink_lock(lock);
+ ldlm_lock_destroy_nolock(lock);
+ unlock_res_and_lock(lock);
+
+ }
+ }
if (!err && dlm_req->lock_desc.l_resource.lr_type != LDLM_FLOCK)
ldlm_reprocess_all(lock->l_resource);
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE)) {
int to = cfs_time_seconds(1);
while (to > 0) {
- schedule_timeout_and_set_state(
- TASK_INTERRUPTIBLE, to);
+ set_current_state(TASK_INTERRUPTIBLE);
+ schedule_timeout(to);
if (lock->l_granted_mode == lock->l_req_mode ||
ldlm_is_destroyed(lock))
break;
lock->l_lvb_len, lvb_len);
GOTO(out, rc = -EINVAL);
}
- } else if (ldlm_has_layout(lock)) { /* for layout lock, lvb has
- * variable length */
- void *lvb_data;
-
- OBD_ALLOC_LARGE(lvb_data, lvb_len);
- if (lvb_data == NULL) {
- LDLM_ERROR(lock, "No memory: %d.\n", lvb_len);
- GOTO(out, rc = -ENOMEM);
- }
-
- lock_res_and_lock(lock);
- LASSERT(lock->l_lvb_data == NULL);
- lock->l_lvb_type = LVB_T_LAYOUT;
- lock->l_lvb_data = lvb_data;
- lock->l_lvb_len = lvb_len;
- unlock_res_and_lock(lock);
}
}
}
static int __ldlm_bl_to_thread(struct ldlm_bl_work_item *blwi,
- ldlm_cancel_flags_t cancel_flags)
+ enum ldlm_cancel_flags cancel_flags)
{
struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool;
ENTRY;
struct ldlm_lock_desc *ld,
struct list_head *cancels, int count,
struct ldlm_lock *lock,
- ldlm_cancel_flags_t cancel_flags)
+ enum ldlm_cancel_flags cancel_flags)
{
init_completion(&blwi->blwi_comp);
INIT_LIST_HEAD(&blwi->blwi_head);
struct ldlm_lock_desc *ld,
struct ldlm_lock *lock,
struct list_head *cancels, int count,
- ldlm_cancel_flags_t cancel_flags)
+ enum ldlm_cancel_flags cancel_flags)
{
ENTRY;
int ldlm_bl_to_thread_list(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld,
struct list_head *cancels, int count,
- ldlm_cancel_flags_t cancel_flags)
+ enum ldlm_cancel_flags cancel_flags)
{
return ldlm_bl_to_thread(ns, ld, NULL, cancels, count, cancel_flags);
}
+int ldlm_bl_thread_wakeup(void)
+{
+ wake_up(&ldlm_state->ldlm_bl_pool->blp_waitq);
+ return 0;
+}
+
/* Setinfo coming from Server (eg MDT) to Client (eg MDC)! */
static int ldlm_handle_setinfo(struct ptlrpc_request *req)
{
CWARN("Send reply failed, maybe cause bug 21636.\n");
}
-static int ldlm_handle_qc_callback(struct ptlrpc_request *req)
-{
- struct obd_quotactl *oqctl;
- struct client_obd *cli = &req->rq_export->exp_obd->u.cli;
-
- oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
- if (oqctl == NULL) {
- CERROR("Can't unpack obd_quotactl\n");
- RETURN(-EPROTO);
- }
-
- oqctl->qc_stat = ptlrpc_status_ntoh(oqctl->qc_stat);
-
- cli->cl_qchk_stat = oqctl->qc_stat;
- return 0;
-}
-
/* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
static int ldlm_callback_handler(struct ptlrpc_request *req)
{
rc = llog_origin_handle_close(req);
ldlm_callback_reply(req, rc);
RETURN(0);
- case OBD_QC_CALLBACK:
- req_capsule_set(&req->rq_pill, &RQF_QC_CALLBACK);
- if (OBD_FAIL_CHECK(OBD_FAIL_OBD_QC_CALLBACK_NET))
- RETURN(0);
- rc = ldlm_handle_qc_callback(req);
- ldlm_callback_reply(req, rc);
- RETURN(0);
default:
CERROR("unknown opcode %u\n",
lustre_msg_get_opc(req->rq_reqmsg));
req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
CDEBUG(D_INODE, "cancel\n");
if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_NET) ||
- CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_CANCEL_RESEND))
+ CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_CANCEL_RESEND) ||
+ CFS_FAIL_CHECK(OBD_FAIL_LDLM_BL_EVICT))
RETURN(0);
rc = ldlm_handle_cancel(req);
if (rc)
RETURN(0);
}
-static int ldlm_revoke_lock_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
+static int ldlm_revoke_lock_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd,
struct hlist_node *hnode, void *data)
{
EXPORT_SYMBOL(ldlm_revoke_export_locks);
#endif /* HAVE_SERVER_SUPPORT */
-static struct ldlm_bl_work_item *ldlm_bl_get_work(struct ldlm_bl_pool *blp)
+static int ldlm_bl_get_work(struct ldlm_bl_pool *blp,
+ struct ldlm_bl_work_item **p_blwi,
+ struct obd_export **p_exp)
{
struct ldlm_bl_work_item *blwi = NULL;
static unsigned int num_bl = 0;
+ static unsigned int num_stale;
+ int num_th = atomic_read(&blp->blp_num_threads);
+
+ *p_exp = obd_stale_export_get();
spin_lock(&blp->blp_lock);
+ if (*p_exp != NULL) {
+ if (num_th == 1 || ++num_stale < num_th) {
+ spin_unlock(&blp->blp_lock);
+ return 1;
+ } else {
+ num_stale = 0;
+ }
+ }
+
/* process a request from the blp_list at least every blp_num_threads */
if (!list_empty(&blp->blp_list) &&
(list_empty(&blp->blp_prio_list) || num_bl == 0))
blwi_entry);
if (blwi) {
- if (++num_bl >= atomic_read(&blp->blp_num_threads))
+ if (++num_bl >= num_th)
num_bl = 0;
list_del(&blwi->blwi_entry);
}
spin_unlock(&blp->blp_lock);
+ *p_blwi = blwi;
+
+ if (*p_exp != NULL && *p_blwi != NULL) {
+ obd_stale_export_put(*p_exp);
+ *p_exp = NULL;
+ }
- return blwi;
+ return (*p_blwi != NULL || *p_exp != NULL) ? 1 : 0;
}
/* This only contains temporary data until the thread starts */
return 0;
}
+/* Not fatal if racy and have a few too many threads */
+static int ldlm_bl_thread_need_create(struct ldlm_bl_pool *blp,
+ struct ldlm_bl_work_item *blwi)
+{
+ int busy = atomic_read(&blp->blp_busy_threads);
+
+ if (busy >= blp->blp_max_threads)
+ return 0;
+
+ if (busy < atomic_read(&blp->blp_num_threads))
+ return 0;
+
+ if (blwi != NULL && (blwi->blwi_ns == NULL ||
+ blwi->blwi_mem_pressure))
+ return 0;
+
+ return 1;
+}
+
+static int ldlm_bl_thread_blwi(struct ldlm_bl_pool *blp,
+ struct ldlm_bl_work_item *blwi)
+{
+ ENTRY;
+
+ if (blwi->blwi_ns == NULL)
+ /* added by ldlm_cleanup() */
+ RETURN(LDLM_ITER_STOP);
+
+ if (blwi->blwi_mem_pressure)
+ memory_pressure_set();
+
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_PAUSE_CANCEL2, 4);
+
+ if (blwi->blwi_count) {
+ int count;
+ /* The special case when we cancel locks in lru
+ * asynchronously, we pass the list of locks here.
+ * Thus locks are marked LDLM_FL_CANCELING, but NOT
+ * canceled locally yet. */
+ count = ldlm_cli_cancel_list_local(&blwi->blwi_head,
+ blwi->blwi_count,
+ LCF_BL_AST);
+ ldlm_cli_cancel_list(&blwi->blwi_head, count, NULL,
+ blwi->blwi_flags);
+ } else {
+ ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld,
+ blwi->blwi_lock);
+ }
+ if (blwi->blwi_mem_pressure)
+ memory_pressure_clr();
+
+ if (blwi->blwi_flags & LCF_ASYNC)
+ OBD_FREE(blwi, sizeof(*blwi));
+ else
+ complete(&blwi->blwi_comp);
+
+ RETURN(0);
+}
+
+/**
+ * Cancel stale locks on export. Cancel blocked locks first.
+ * If the given export has blocked locks, the next in the list may have
+ * them too, thus cancel not blocked locks only if the current export has
+ * no blocked locks.
+ **/
+static int ldlm_bl_thread_exports(struct ldlm_bl_pool *blp,
+ struct obd_export *exp)
+{
+ int num;
+ ENTRY;
+
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_BL_EVICT, 4);
+
+ num = ldlm_export_cancel_blocked_locks(exp);
+ if (num == 0)
+ ldlm_export_cancel_locks(exp);
+
+ obd_stale_export_put(exp);
+
+ RETURN(0);
+}
+
+
/**
* Main blocking requests processing thread.
*
while (1) {
struct l_wait_info lwi = { 0 };
struct ldlm_bl_work_item *blwi = NULL;
- int busy;
+ struct obd_export *exp = NULL;
+ int rc;
- blwi = ldlm_bl_get_work(blp);
+ rc = ldlm_bl_get_work(blp, &blwi, &exp);
- if (blwi == NULL) {
+ if (rc == 0) {
atomic_dec(&blp->blp_busy_threads);
l_wait_event_exclusive(blp->blp_waitq,
- (blwi = ldlm_bl_get_work(blp)) != NULL,
- &lwi);
- busy = atomic_inc_return(&blp->blp_busy_threads);
- } else {
- busy = atomic_read(&blp->blp_busy_threads);
+ ldlm_bl_get_work(blp, &blwi,
+ &exp),
+ &lwi);
+ atomic_inc(&blp->blp_busy_threads);
}
- if (blwi->blwi_ns == NULL)
- /* added by ldlm_cleanup() */
- break;
-
- /* Not fatal if racy and have a few too many threads */
- if (unlikely(busy < blp->blp_max_threads &&
- busy >= atomic_read(&blp->blp_num_threads) &&
- !blwi->blwi_mem_pressure))
+ if (ldlm_bl_thread_need_create(blp, blwi))
/* discard the return value, we tried */
ldlm_bl_thread_start(blp);
- if (blwi->blwi_mem_pressure)
- memory_pressure_set();
-
- OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_PAUSE_CANCEL2, 4);
-
- if (blwi->blwi_count) {
- int count;
- /* The special case when we cancel locks in LRU
- * asynchronously, we pass the list of locks here.
- * Thus locks are marked LDLM_FL_CANCELING, but NOT
- * canceled locally yet. */
- count = ldlm_cli_cancel_list_local(&blwi->blwi_head,
- blwi->blwi_count,
- LCF_BL_AST);
- ldlm_cli_cancel_list(&blwi->blwi_head, count, NULL,
- blwi->blwi_flags);
- } else {
- ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld,
- blwi->blwi_lock);
- }
- if (blwi->blwi_mem_pressure)
- memory_pressure_clr();
+ if (exp)
+ rc = ldlm_bl_thread_exports(blp, exp);
+ else if (blwi)
+ rc = ldlm_bl_thread_blwi(blp, blwi);
- if (blwi->blwi_flags & LCF_ASYNC)
- OBD_FREE(blwi, sizeof(*blwi));
- else
- complete(&blwi->blwi_comp);
+ if (rc == LDLM_ITER_STOP)
+ break;
}
atomic_dec(&blp->blp_busy_threads);
* Export handle<->lock hash operations.
*/
static unsigned
-ldlm_export_lock_hash(cfs_hash_t *hs, const void *key, unsigned mask)
+ldlm_export_lock_hash(struct cfs_hash *hs, const void *key, unsigned mask)
{
return cfs_hash_u64_hash(((struct lustre_handle *)key)->cookie, mask);
}
}
static void
-ldlm_export_lock_get(cfs_hash_t *hs, struct hlist_node *hnode)
+ldlm_export_lock_get(struct cfs_hash *hs, struct hlist_node *hnode)
{
struct ldlm_lock *lock;
}
static void
-ldlm_export_lock_put(cfs_hash_t *hs, struct hlist_node *hnode)
+ldlm_export_lock_put(struct cfs_hash *hs, struct hlist_node *hnode)
{
struct ldlm_lock *lock;
LDLM_LOCK_RELEASE(lock);
}
-static cfs_hash_ops_t ldlm_export_lock_ops = {
+static struct cfs_hash_ops ldlm_export_lock_ops = {
.hs_hash = ldlm_export_lock_hash,
.hs_key = ldlm_export_lock_key,
.hs_keycmp = ldlm_export_lock_keycmp,
CERROR("Failed to initialize LDLM pools: %d\n", rc);
GOTO(out, rc);
}
+
+ rc = ldlm_reclaim_setup();
+ if (rc) {
+ CERROR("Failed to setup reclaim thread: rc = %d\n", rc);
+ GOTO(out, rc);
+ }
RETURN(0);
out:
RETURN(-EBUSY);
}
- ldlm_pools_fini();
+ ldlm_reclaim_cleanup();
+ ldlm_pools_fini();
if (ldlm_state->ldlm_bl_pool != NULL) {
struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool;
ldlm_lock_slab = kmem_cache_create("ldlm_locks",
sizeof(struct ldlm_lock), 0,
SLAB_HWCACHE_ALIGN | SLAB_DESTROY_BY_RCU, NULL);
- if (ldlm_lock_slab == NULL) {
- kmem_cache_destroy(ldlm_resource_slab);
- return -ENOMEM;
- }
+ if (ldlm_lock_slab == NULL)
+ goto out_resource;
ldlm_interval_slab = kmem_cache_create("interval_node",
sizeof(struct ldlm_interval),
0, SLAB_HWCACHE_ALIGN, NULL);
- if (ldlm_interval_slab == NULL) {
- kmem_cache_destroy(ldlm_resource_slab);
- kmem_cache_destroy(ldlm_lock_slab);
- return -ENOMEM;
- }
+ if (ldlm_interval_slab == NULL)
+ goto out_lock;
+
+ ldlm_interval_tree_slab = kmem_cache_create("interval_tree",
+ sizeof(struct ldlm_interval_tree) * LCK_MODE_NUM,
+ 0, SLAB_HWCACHE_ALIGN, NULL);
+ if (ldlm_interval_tree_slab == NULL)
+ goto out_interval;
+
#if LUSTRE_TRACKS_LOCK_EXP_REFS
- class_export_dump_hook = ldlm_dump_export_locks;
+ class_export_dump_hook = ldlm_dump_export_locks;
#endif
- return 0;
+ return 0;
+
+out_interval:
+ kmem_cache_destroy(ldlm_interval_slab);
+out_lock:
+ kmem_cache_destroy(ldlm_lock_slab);
+out_resource:
+ kmem_cache_destroy(ldlm_resource_slab);
+
+ return -ENOMEM;
}
void ldlm_exit(void)
synchronize_rcu();
kmem_cache_destroy(ldlm_lock_slab);
kmem_cache_destroy(ldlm_interval_slab);
+ kmem_cache_destroy(ldlm_interval_tree_slab);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff