Whamcloud - gitweb
LU-5496 ldlm: reconstruct proper flags on enqueue resend
[fs/lustre-release.git] / lustre / ldlm / ldlm_lockd.c
index 6580b8d..5d1bc0d 100644 (file)
 
 #define DEBUG_SUBSYSTEM S_LDLM
 
-#ifdef __KERNEL__
-# include <libcfs/libcfs.h>
-#else
-# include <liblustre.h>
-#endif
-
+#include <libcfs/libcfs.h>
 #include <lustre_dlm.h>
 #include <obd_class.h>
 #include <libcfs/list.h>
@@ -100,13 +95,13 @@ struct ldlm_bl_pool {
         * as a priority. It is used for LDLM_FL_DISCARD_DATA requests.
         * see bug 13843
         */
-       cfs_list_t              blp_prio_list;
+       struct list_head              blp_prio_list;
 
        /*
         * blp_list is used for all other callbacks which are likely
         * to take longer to process.
         */
-       cfs_list_t              blp_list;
+       struct list_head              blp_list;
 
        wait_queue_head_t       blp_waitq;
        struct completion       blp_comp;
@@ -117,18 +112,18 @@ struct ldlm_bl_pool {
 };
 
 struct ldlm_bl_work_item {
-        cfs_list_t              blwi_entry;
+       struct list_head              blwi_entry;
         struct ldlm_namespace  *blwi_ns;
         struct ldlm_lock_desc   blwi_ld;
         struct ldlm_lock       *blwi_lock;
-        cfs_list_t              blwi_head;
+       struct list_head              blwi_head;
         int                     blwi_count;
        struct completion        blwi_comp;
        ldlm_cancel_flags_t     blwi_flags;
         int                     blwi_mem_pressure;
 };
 
-#if defined(HAVE_SERVER_SUPPORT) && defined(__KERNEL__)
+#ifdef HAVE_SERVER_SUPPORT
 
 /**
  * Protects both waiting_locks_list and expired_lock_thread.
@@ -146,14 +141,14 @@ static spinlock_t waiting_locks_spinlock;   /* BH lock (timer) */
  *
  * All access to it should be under waiting_locks_spinlock.
  */
-static cfs_list_t waiting_locks_list;
+static struct list_head waiting_locks_list;
 static struct timer_list waiting_locks_timer;
 
 static struct expired_lock_thread {
        wait_queue_head_t       elt_waitq;
        int                     elt_state;
        int                     elt_dump;
-       cfs_list_t              elt_expired_locks;
+       struct list_head                elt_expired_locks;
 } expired_lock_thread;
 
 static inline int have_expired_locks(void)
@@ -162,7 +157,7 @@ static inline int have_expired_locks(void)
 
        ENTRY;
        spin_lock_bh(&waiting_locks_spinlock);
-       need_to_run = !cfs_list_empty(&expired_lock_thread.elt_expired_locks);
+       need_to_run = !list_empty(&expired_lock_thread.elt_expired_locks);
        spin_unlock_bh(&waiting_locks_spinlock);
 
        RETURN(need_to_run);
@@ -173,7 +168,7 @@ static inline int have_expired_locks(void)
  */
 static int expired_lock_main(void *arg)
 {
-       cfs_list_t *expired = &expired_lock_thread.elt_expired_locks;
+       struct list_head *expired = &expired_lock_thread.elt_expired_locks;
        struct l_wait_info lwi = { 0 };
        int do_dump;
 
@@ -206,11 +201,11 @@ static int expired_lock_main(void *arg)
 
                do_dump = 0;
 
-               while (!cfs_list_empty(expired)) {
+               while (!list_empty(expired)) {
                        struct obd_export *export;
                        struct ldlm_lock *lock;
 
-                       lock = cfs_list_entry(expired->next, struct ldlm_lock,
+                       lock = list_entry(expired->next, struct ldlm_lock,
                                          l_pending_chain);
                        if ((void *)lock < LP_POISON + PAGE_CACHE_SIZE &&
                            (void *)lock >= LP_POISON) {
@@ -218,7 +213,7 @@ static int expired_lock_main(void *arg)
                                CERROR("free lock on elt list %p\n", lock);
                                LBUG();
                        }
-                       cfs_list_del_init(&lock->l_pending_chain);
+                       list_del_init(&lock->l_pending_chain);
                        if ((void *)lock->l_export <
                             LP_POISON + PAGE_CACHE_SIZE &&
                            (void *)lock->l_export >= LP_POISON) {
@@ -285,7 +280,7 @@ static int ldlm_lock_busy(struct ldlm_lock *lock)
                return 0;
 
        spin_lock_bh(&lock->l_export->exp_rpc_lock);
-       cfs_list_for_each_entry(req, &lock->l_export->exp_hp_rpcs,
+       list_for_each_entry(req, &lock->l_export->exp_hp_rpcs,
                                rq_exp_list) {
                if (req->rq_ops->hpreq_lock_match) {
                        match = req->rq_ops->hpreq_lock_match(req, lock);
@@ -304,8 +299,8 @@ static void waiting_locks_callback(unsigned long unused)
        int                     need_dump = 0;
 
        spin_lock_bh(&waiting_locks_spinlock);
-        while (!cfs_list_empty(&waiting_locks_list)) {
-                lock = cfs_list_entry(waiting_locks_list.next, struct ldlm_lock,
+       while (!list_empty(&waiting_locks_list)) {
+               lock = list_entry(waiting_locks_list.next, struct ldlm_lock,
                                       l_pending_chain);
                 if (cfs_time_after(lock->l_callback_timeout,
                                    cfs_time_current()) ||
@@ -339,20 +334,20 @@ static void waiting_locks_callback(unsigned long unused)
                 ldlm_lock_to_ns(lock)->ns_timeouts++;
                 LDLM_ERROR(lock, "lock callback timer expired after %lds: "
                            "evicting client at %s ",
-                           cfs_time_current_sec()- lock->l_last_activity,
+                           cfs_time_current_sec() - lock->l_last_activity,
                            libcfs_nid2str(
                                    lock->l_export->exp_connection->c_peer.nid));
 
                 /* no needs to take an extra ref on the lock since it was in
                  * the waiting_locks_list and ldlm_add_waiting_lock()
                  * already grabbed a ref */
-                cfs_list_del(&lock->l_pending_chain);
-                cfs_list_add(&lock->l_pending_chain,
+               list_del(&lock->l_pending_chain);
+               list_add(&lock->l_pending_chain,
                              &expired_lock_thread.elt_expired_locks);
                need_dump = 1;
        }
 
-       if (!cfs_list_empty(&expired_lock_thread.elt_expired_locks)) {
+       if (!list_empty(&expired_lock_thread.elt_expired_locks)) {
                if (obd_dump_on_timeout && need_dump)
                        expired_lock_thread.elt_dump = __LINE__;
 
@@ -363,9 +358,9 @@ static void waiting_locks_callback(unsigned long unused)
          * Make sure the timer will fire again if we have any locks
          * left.
          */
-        if (!cfs_list_empty(&waiting_locks_list)) {
+       if (!list_empty(&waiting_locks_list)) {
                 cfs_time_t timeout_rounded;
-                lock = cfs_list_entry(waiting_locks_list.next, struct ldlm_lock,
+               lock = list_entry(waiting_locks_list.next, struct ldlm_lock,
                                       l_pending_chain);
                 timeout_rounded = (cfs_time_t)round_timeout(lock->l_callback_timeout);
                 cfs_timer_arm(&waiting_locks_timer, timeout_rounded);
@@ -390,7 +385,7 @@ static int __ldlm_add_waiting_lock(struct ldlm_lock *lock, int seconds)
         cfs_time_t timeout;
         cfs_time_t timeout_rounded;
 
-        if (!cfs_list_empty(&lock->l_pending_chain))
+       if (!list_empty(&lock->l_pending_chain))
                 return 0;
 
         if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_HPREQ_NOTIMEOUT) ||
@@ -411,7 +406,7 @@ static int __ldlm_add_waiting_lock(struct ldlm_lock *lock, int seconds)
         /* if the new lock has a shorter timeout than something earlier on
            the list, we'll wait the longer amount of time; no big deal. */
         /* FIFO */
-        cfs_list_add_tail(&lock->l_pending_chain, &waiting_locks_list);
+       list_add_tail(&lock->l_pending_chain, &waiting_locks_list);
         return 1;
 }
 
@@ -448,8 +443,8 @@ static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
 
        if (ret) {
                spin_lock_bh(&lock->l_export->exp_bl_list_lock);
-               if (cfs_list_empty(&lock->l_exp_list))
-                       cfs_list_add(&lock->l_exp_list,
+               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);
        }
@@ -471,9 +466,9 @@ static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
  */
 static int __ldlm_del_waiting_lock(struct ldlm_lock *lock)
 {
-        cfs_list_t *list_next;
+       struct list_head *list_next;
 
-        if (cfs_list_empty(&lock->l_pending_chain))
+       if (list_empty(&lock->l_pending_chain))
                 return 0;
 
         list_next = lock->l_pending_chain.next;
@@ -484,13 +479,13 @@ static int __ldlm_del_waiting_lock(struct ldlm_lock *lock)
                         cfs_timer_disarm(&waiting_locks_timer);
                 } else {
                         struct ldlm_lock *next;
-                        next = cfs_list_entry(list_next, struct ldlm_lock,
+                       next = list_entry(list_next, struct ldlm_lock,
                                               l_pending_chain);
                         cfs_timer_arm(&waiting_locks_timer,
                                       round_timeout(next->l_callback_timeout));
                 }
         }
-        cfs_list_del_init(&lock->l_pending_chain);
+       list_del_init(&lock->l_pending_chain);
 
         return 1;
 }
@@ -511,7 +506,7 @@ int ldlm_del_waiting_lock(struct ldlm_lock *lock)
 
        /* remove the lock out of export blocking list */
        spin_lock_bh(&lock->l_export->exp_bl_list_lock);
-       cfs_list_del_init(&lock->l_exp_list);
+       list_del_init(&lock->l_exp_list);
        spin_unlock_bh(&lock->l_export->exp_bl_list_lock);
 
         if (ret) {
@@ -540,7 +535,7 @@ int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout)
 
        spin_lock_bh(&waiting_locks_spinlock);
 
-       if (cfs_list_empty(&lock->l_pending_chain)) {
+       if (list_empty(&lock->l_pending_chain)) {
                spin_unlock_bh(&waiting_locks_spinlock);
                LDLM_DEBUG(lock, "wasn't waiting");
                return 0;
@@ -557,7 +552,7 @@ int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout)
 }
 EXPORT_SYMBOL(ldlm_refresh_waiting_lock);
 
-#else /* !HAVE_SERVER_SUPPORT ||  !__KERNEL__ */
+#else /* HAVE_SERVER_SUPPORT */
 
 int ldlm_del_waiting_lock(struct ldlm_lock *lock)
 {
@@ -569,15 +564,7 @@ int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout)
         RETURN(0);
 }
 
-# ifdef HAVE_SERVER_SUPPORT
-static int ldlm_add_waiting_lock(struct ldlm_lock *lock)
-{
-       LASSERT(ldlm_is_res_locked(lock) && !ldlm_is_cancel_on_block(lock));
-       RETURN(1);
-}
-
-# endif
-#endif /* HAVE_SERVER_SUPPORT && __KERNEL__ */
+#endif /* !HAVE_SERVER_SUPPORT */
 
 #ifdef HAVE_SERVER_SUPPORT
 
@@ -594,19 +581,15 @@ static void ldlm_failed_ast(struct ldlm_lock *lock, int rc,
 
         if (obd_dump_on_timeout)
                 libcfs_debug_dumplog();
-#ifdef __KERNEL__
        spin_lock_bh(&waiting_locks_spinlock);
        if (__ldlm_del_waiting_lock(lock) == 0)
                /* the lock was not in any list, grab an extra ref before adding
                 * the lock to the expired list */
                LDLM_LOCK_GET(lock);
-       cfs_list_add(&lock->l_pending_chain,
+       list_add(&lock->l_pending_chain,
                     &expired_lock_thread.elt_expired_locks);
        wake_up(&expired_lock_thread.elt_waitq);
        spin_unlock_bh(&waiting_locks_spinlock);
-#else
-       class_fail_export(lock->l_export);
-#endif
 }
 
 /**
@@ -758,8 +741,8 @@ static void ldlm_lock_reorder_req(struct ldlm_lock *lock)
        }
 
        spin_lock_bh(&lock->l_export->exp_rpc_lock);
-       cfs_list_for_each_entry(req, &lock->l_export->exp_hp_rpcs,
-                               rq_exp_list) {
+       list_for_each_entry(req, &lock->l_export->exp_hp_rpcs,
+                           rq_exp_list) {
                /* Do not process requests that were not yet added to there
                 * incoming queue or were already removed from there for
                 * processing. We evaluate ptlrpc_nrs_req_can_move() without
@@ -862,6 +845,8 @@ int ldlm_server_blocking_ast(struct ldlm_lock *lock,
         if (AT_OFF)
                 req->rq_timeout = ldlm_get_rq_timeout();
 
+       lock->l_last_activity = cfs_time_current_sec();
+
         if (lock->l_export && lock->l_export->exp_nid_stats &&
             lock->l_export->exp_nid_stats->nid_ldlm_stats)
                 lprocfs_counter_incr(lock->l_export->exp_nid_stats->nid_ldlm_stats,
@@ -898,6 +883,11 @@ int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data)
         total_enqueue_wait = cfs_time_sub(cfs_time_current_sec(),
                                           lock->l_last_activity);
 
+       if (OBD_FAIL_PRECHECK(OBD_FAIL_OST_LDLM_REPLY_NET)) {
+               LDLM_DEBUG(lock, "dropping CP AST");
+               RETURN(0);
+       }
+
         req = ptlrpc_request_alloc(lock->l_export->exp_imp_reverse,
                                     &RQF_LDLM_CP_CALLBACK);
         if (req == NULL)
@@ -954,6 +944,8 @@ int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data)
         LDLM_DEBUG(lock, "server preparing completion AST (after %lds wait)",
                    total_enqueue_wait);
 
+       lock->l_last_activity = cfs_time_current_sec();
+
         /* Server-side enqueue wait time estimate, used in
             __ldlm_add_waiting_lock to set future enqueue timers */
         if (total_enqueue_wait < ldlm_get_enq_timeout(lock))
@@ -1070,6 +1062,8 @@ int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data)
         if (AT_OFF)
                 req->rq_timeout = ldlm_get_rq_timeout();
 
+       lock->l_last_activity = cfs_time_current_sec();
+
        req->rq_interpret_reply = ldlm_cb_interpret;
 
         if (lock->l_export && lock->l_export->exp_nid_stats &&
@@ -1083,7 +1077,8 @@ int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data)
 }
 EXPORT_SYMBOL(ldlm_server_glimpse_ast);
 
-int ldlm_glimpse_locks(struct ldlm_resource *res, cfs_list_t *gl_work_list)
+int ldlm_glimpse_locks(struct ldlm_resource *res,
+                      struct list_head *gl_work_list)
 {
        int     rc;
        ENTRY;
@@ -1162,6 +1157,7 @@ int ldlm_handle_enqueue0(struct ldlm_namespace *ns,
         struct ldlm_lock *lock = NULL;
         void *cookie = NULL;
         int rc = 0;
+       struct ldlm_resource *res = NULL;
         ENTRY;
 
         LDLM_DEBUG_NOLOCK("server-side enqueue handler START");
@@ -1226,7 +1222,8 @@ int ldlm_handle_enqueue0(struct ldlm_namespace *ns,
         }
 #endif
 
-        if (unlikely(flags & LDLM_FL_REPLAY)) {
+       if (unlikely((flags & LDLM_FL_REPLAY) ||
+                    (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))) {
                 /* Find an existing lock in the per-export lock hash */
                /* In the function below, .hs_keycmp resolves to
                 * ldlm_export_lock_keycmp() */
@@ -1236,22 +1233,41 @@ int ldlm_handle_enqueue0(struct ldlm_namespace *ns,
                 if (lock != NULL) {
                         DEBUG_REQ(D_DLMTRACE, req, "found existing lock cookie "
                                   LPX64, lock->l_handle.h_cookie);
+                       flags |= LDLM_FL_RESENT;
                         GOTO(existing_lock, rc = 0);
-                }
+               }
         }
 
-        /* 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,
-                                dlm_req->lock_desc.l_resource.lr_type,
-                                dlm_req->lock_desc.l_req_mode,
+       /* 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,
+                               dlm_req->lock_desc.l_resource.lr_type,
+                               dlm_req->lock_desc.l_req_mode,
                                cbs, NULL, 0, LVB_T_NONE);
-        if (!lock)
-                GOTO(out, rc = -ENOMEM);
+       if (IS_ERR(lock)) {
+               rc = PTR_ERR(lock);
+               lock = NULL;
+               GOTO(out, rc);
+       }
 
         lock->l_last_activity = cfs_time_current_sec();
         lock->l_remote_handle = dlm_req->lock_handle[0];
         LDLM_DEBUG(lock, "server-side enqueue handler, new lock created");
 
+       /* Initialize resource lvb but not for a lock being replayed since
+        * Client already got lvb sent in this case.
+        * This must occur early since some policy methods assume resource
+        * lvb is available (lr_lvb_data != NULL).
+        */
+       res = lock->l_resource;
+       if (!(flags & LDLM_FL_REPLAY)) {
+               /* non-replayed lock, delayed lvb init may need to be done */
+               rc = ldlm_lvbo_init(res);
+               if (rc < 0) {
+                       LDLM_ERROR(lock, "delayed lvb init failed (rc %d)", rc);
+                       GOTO(out, rc);
+               }
+       }
+
         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_BLOCKED, obd_timeout * 2);
         /* Don't enqueue a lock onto the export if it is been disonnected
          * due to eviction (bug 3822) or server umount (bug 24324).
@@ -1268,6 +1284,11 @@ int ldlm_handle_enqueue0(struct ldlm_namespace *ns,
                              &lock->l_remote_handle,
                              &lock->l_exp_hash);
 
+       /* Inherit the enqueue flags before the operation, because we do not
+        * keep the res lock on return and next operations (BL AST) may proceed
+        * without them. */
+       lock->l_flags |= ldlm_flags_from_wire(dlm_req->lock_flags &
+                                             LDLM_FL_INHERIT_MASK);
 existing_lock:
 
         if (flags & LDLM_FL_HAS_INTENT) {
@@ -1305,7 +1326,6 @@ existing_lock:
        }
 
         dlm_rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
-       dlm_rep->lock_flags = ldlm_flags_to_wire(flags);
 
         ldlm_lock2desc(lock, &dlm_rep->lock_desc);
         ldlm_lock2handle(lock, &dlm_rep->lock_handle);
@@ -1316,9 +1336,8 @@ existing_lock:
 
         /* Now take into account flags to be inherited from original lock
            request both in reply to client and in our own lock flags. */
-       dlm_rep->lock_flags |= dlm_req->lock_flags & LDLM_FL_INHERIT_MASK;
-       lock->l_flags |= ldlm_flags_from_wire(dlm_req->lock_flags &
-                                             LDLM_FL_INHERIT_MASK);
+       dlm_rep->lock_flags = ldlm_flags_to_wire(flags);
+       lock->l_flags |= flags & LDLM_FL_INHERIT_MASK;
 
         /* Don't move a pending lock onto the export if it has already been
          * disconnected due to eviction (bug 5683) or server umount (bug 24324).
@@ -1383,11 +1402,11 @@ existing_lock:
 
         /* The LOCK_CHANGED code in ldlm_lock_enqueue depends on this
          * ldlm_reprocess_all.  If this moves, revisit that code. -phil */
-        if (lock) {
-                LDLM_DEBUG(lock, "server-side enqueue handler, sending reply"
-                           "(err=%d, rc=%d)", err, rc);
+       if (lock != NULL) {
+               LDLM_DEBUG(lock, "server-side enqueue handler, sending reply"
+                          "(err=%d, rc=%d)", err, rc);
 
-                if (rc == 0) {
+               if (rc == 0) {
                        if (req_capsule_has_field(&req->rq_pill, &RMF_DLM_LVB,
                                                  RCL_SERVER) &&
                            ldlm_lvbo_size(lock) > 0) {
@@ -1400,7 +1419,9 @@ existing_lock:
                                         req, lock);
                                buflen = req_capsule_get_size(&req->rq_pill,
                                                &RMF_DLM_LVB, RCL_SERVER);
-                               if (buflen > 0) {
+                               /* non-replayed lock, delayed lvb init may
+                                * need to be occur now */
+                               if ((buflen > 0) && !(flags & LDLM_FL_REPLAY)) {
                                        buflen = ldlm_lvbo_fill(lock, buf,
                                                                buflen);
                                        if (buflen >= 0)
@@ -1410,11 +1431,22 @@ existing_lock:
                                                        buflen, RCL_SERVER);
                                        else
                                                rc = buflen;
+                               } else if (flags & LDLM_FL_REPLAY) {
+                                       /* no LVB resend upon replay */
+                                       if (buflen > 0)
+                                               req_capsule_shrink(
+                                                       &req->rq_pill,
+                                                       &RMF_DLM_LVB,
+                                                       0, RCL_SERVER);
+                                       else
+                                               rc = buflen;
                                } else {
                                        rc = buflen;
                                }
                        }
-                } else {
+               }
+
+               if (rc != 0) {
                         lock_res_and_lock(lock);
                         ldlm_resource_unlink_lock(lock);
                         ldlm_lock_destroy_nolock(lock);
@@ -1688,13 +1720,14 @@ static void ldlm_handle_cp_callback(struct ptlrpc_request *req,
                                     struct ldlm_request *dlm_req,
                                     struct ldlm_lock *lock)
 {
+       struct list_head ast_list;
        int lvb_len;
-       CFS_LIST_HEAD(ast_list);
        int rc = 0;
        ENTRY;
 
        LDLM_DEBUG(lock, "client completion callback handler START");
 
+       INIT_LIST_HEAD(&ast_list);
        if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE)) {
                int to = cfs_time_seconds(1);
                while (to > 0) {
@@ -1882,7 +1915,6 @@ static int ldlm_callback_reply(struct ptlrpc_request *req, int rc)
         return ptlrpc_reply(req);
 }
 
-#ifdef __KERNEL__
 static int __ldlm_bl_to_thread(struct ldlm_bl_work_item *blwi,
                               ldlm_cancel_flags_t cancel_flags)
 {
@@ -1893,10 +1925,10 @@ static int __ldlm_bl_to_thread(struct ldlm_bl_work_item *blwi,
        if (blwi->blwi_lock &&
            ldlm_is_discard_data(blwi->blwi_lock)) {
                /* add LDLM_FL_DISCARD_DATA requests to the priority list */
-               cfs_list_add_tail(&blwi->blwi_entry, &blp->blp_prio_list);
+               list_add_tail(&blwi->blwi_entry, &blp->blp_prio_list);
        } else {
                /* other blocking callbacks are added to the regular list */
-               cfs_list_add_tail(&blwi->blwi_entry, &blp->blp_list);
+               list_add_tail(&blwi->blwi_entry, &blp->blp_list);
        }
        spin_unlock(&blp->blp_lock);
 
@@ -1913,12 +1945,12 @@ static int __ldlm_bl_to_thread(struct ldlm_bl_work_item *blwi,
 static inline void init_blwi(struct ldlm_bl_work_item *blwi,
                             struct ldlm_namespace *ns,
                             struct ldlm_lock_desc *ld,
-                            cfs_list_t *cancels, int count,
+                            struct list_head *cancels, int count,
                             struct ldlm_lock *lock,
                             ldlm_cancel_flags_t cancel_flags)
 {
        init_completion(&blwi->blwi_comp);
-        CFS_INIT_LIST_HEAD(&blwi->blwi_head);
+       INIT_LIST_HEAD(&blwi->blwi_head);
 
        if (memory_pressure_get())
                 blwi->blwi_mem_pressure = 1;
@@ -1928,8 +1960,8 @@ static inline void init_blwi(struct ldlm_bl_work_item *blwi,
         if (ld != NULL)
                 blwi->blwi_ld = *ld;
         if (count) {
-                cfs_list_add(&blwi->blwi_head, cancels);
-                cfs_list_del_init(cancels);
+               list_add(&blwi->blwi_head, cancels);
+               list_del_init(cancels);
                 blwi->blwi_count = count;
         } else {
                 blwi->blwi_lock = lock;
@@ -1948,7 +1980,7 @@ static inline void init_blwi(struct ldlm_bl_work_item *blwi,
 static int ldlm_bl_to_thread(struct ldlm_namespace *ns,
                             struct ldlm_lock_desc *ld,
                             struct ldlm_lock *lock,
-                            cfs_list_t *cancels, int count,
+                            struct list_head *cancels, int count,
                             ldlm_cancel_flags_t cancel_flags)
 {
        ENTRY;
@@ -1977,27 +2009,18 @@ static int ldlm_bl_to_thread(struct ldlm_namespace *ns,
        }
 }
 
-#endif
 
 int ldlm_bl_to_thread_lock(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld,
                           struct ldlm_lock *lock)
 {
-#ifdef __KERNEL__
        return ldlm_bl_to_thread(ns, ld, lock, NULL, 0, LCF_ASYNC);
-#else
-       return -ENOSYS;
-#endif
 }
 
 int ldlm_bl_to_thread_list(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld,
-                          cfs_list_t *cancels, int count,
+                          struct list_head *cancels, int count,
                           ldlm_cancel_flags_t cancel_flags)
 {
-#ifdef __KERNEL__
        return ldlm_bl_to_thread(ns, ld, NULL, cancels, count, cancel_flags);
-#else
-       return -ENOSYS;
-#endif
 }
 
 /* Setinfo coming from Server (eg MDT) to Client (eg MDC)! */
@@ -2418,10 +2441,10 @@ static int ldlm_hpreq_handler(struct ptlrpc_request *req)
 }
 
 int ldlm_revoke_lock_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
-                        cfs_hlist_node_t *hnode, void *data)
+                       struct hlist_node *hnode, void *data)
 
 {
-        cfs_list_t         *rpc_list = data;
+       struct list_head         *rpc_list = data;
         struct ldlm_lock   *lock = cfs_hash_object(hs, hnode);
 
         lock_res_and_lock(lock);
@@ -2457,7 +2480,7 @@ int ldlm_revoke_lock_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
                             &lock->l_remote_handle, &lock->l_exp_hash);
        }
 
-        cfs_list_add_tail(&lock->l_rk_ast, rpc_list);
+       list_add_tail(&lock->l_rk_ast, rpc_list);
         LDLM_LOCK_GET(lock);
 
         unlock_res_and_lock(lock);
@@ -2466,10 +2489,10 @@ int ldlm_revoke_lock_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 
 void ldlm_revoke_export_locks(struct obd_export *exp)
 {
-        cfs_list_t  rpc_list;
+       struct list_head  rpc_list;
         ENTRY;
 
-        CFS_INIT_LIST_HEAD(&rpc_list);
+       INIT_LIST_HEAD(&rpc_list);
         cfs_hash_for_each_empty(exp->exp_lock_hash,
                                 ldlm_revoke_lock_cb, &rpc_list);
         ldlm_run_ast_work(exp->exp_obd->obd_namespace, &rpc_list,
@@ -2480,7 +2503,6 @@ void ldlm_revoke_export_locks(struct obd_export *exp)
 EXPORT_SYMBOL(ldlm_revoke_export_locks);
 #endif /* HAVE_SERVER_SUPPORT */
 
-#ifdef __KERNEL__
 static struct ldlm_bl_work_item *ldlm_bl_get_work(struct ldlm_bl_pool *blp)
 {
        struct ldlm_bl_work_item *blwi = NULL;
@@ -2488,20 +2510,20 @@ static struct ldlm_bl_work_item *ldlm_bl_get_work(struct ldlm_bl_pool *blp)
 
        spin_lock(&blp->blp_lock);
        /* process a request from the blp_list at least every blp_num_threads */
-       if (!cfs_list_empty(&blp->blp_list) &&
-           (cfs_list_empty(&blp->blp_prio_list) || num_bl == 0))
-               blwi = cfs_list_entry(blp->blp_list.next,
-                                     struct ldlm_bl_work_item, blwi_entry);
+       if (!list_empty(&blp->blp_list) &&
+           (list_empty(&blp->blp_prio_list) || num_bl == 0))
+               blwi = list_entry(blp->blp_list.next,
+                                 struct ldlm_bl_work_item, blwi_entry);
        else
-               if (!cfs_list_empty(&blp->blp_prio_list))
-                       blwi = cfs_list_entry(blp->blp_prio_list.next,
-                                             struct ldlm_bl_work_item,
-                                             blwi_entry);
+               if (!list_empty(&blp->blp_prio_list))
+                       blwi = list_entry(blp->blp_prio_list.next,
+                                         struct ldlm_bl_work_item,
+                                         blwi_entry);
 
        if (blwi) {
                if (++num_bl >= atomic_read(&blp->blp_num_threads))
                        num_bl = 0;
-               cfs_list_del(&blwi->blwi_entry);
+               list_del(&blwi->blwi_entry);
        }
        spin_unlock(&blp->blp_lock);
 
@@ -2590,6 +2612,8 @@ static int ldlm_bl_thread_main(void *arg)
                 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
@@ -2620,7 +2644,6 @@ static int ldlm_bl_thread_main(void *arg)
        RETURN(0);
 }
 
-#endif
 
 static int ldlm_setup(void);
 static int ldlm_cleanup(void);
@@ -2670,50 +2693,50 @@ ldlm_export_lock_hash(cfs_hash_t *hs, const void *key, unsigned mask)
 }
 
 static void *
-ldlm_export_lock_key(cfs_hlist_node_t *hnode)
+ldlm_export_lock_key(struct hlist_node *hnode)
 {
         struct ldlm_lock *lock;
 
-        lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
+       lock = hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
         return &lock->l_remote_handle;
 }
 
 static void
-ldlm_export_lock_keycpy(cfs_hlist_node_t *hnode, void *key)
+ldlm_export_lock_keycpy(struct hlist_node *hnode, void *key)
 {
         struct ldlm_lock     *lock;
 
-        lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
+       lock = hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
         lock->l_remote_handle = *(struct lustre_handle *)key;
 }
 
 static int
-ldlm_export_lock_keycmp(const void *key, cfs_hlist_node_t *hnode)
+ldlm_export_lock_keycmp(const void *key, struct hlist_node *hnode)
 {
         return lustre_handle_equal(ldlm_export_lock_key(hnode), key);
 }
 
 static void *
-ldlm_export_lock_object(cfs_hlist_node_t *hnode)
+ldlm_export_lock_object(struct hlist_node *hnode)
 {
-        return cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
+       return hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
 }
 
 static void
-ldlm_export_lock_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+ldlm_export_lock_get(cfs_hash_t *hs, struct hlist_node *hnode)
 {
         struct ldlm_lock *lock;
 
-        lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
+       lock = hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
         LDLM_LOCK_GET(lock);
 }
 
 static void
-ldlm_export_lock_put(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
+ldlm_export_lock_put(cfs_hash_t *hs, struct hlist_node *hnode)
 {
         struct ldlm_lock *lock;
 
-        lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
+       lock = hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
         LDLM_LOCK_RELEASE(lock);
 }
 
@@ -2771,11 +2794,13 @@ EXPORT_SYMBOL(ldlm_destroy_export);
 static int ldlm_setup(void)
 {
        static struct ptlrpc_service_conf       conf;
-       struct ldlm_bl_pool                     *blp = NULL;
-        int rc = 0;
-#ifdef __KERNEL__
-        int i;
-#endif
+       struct ldlm_bl_pool                    *blp = NULL;
+#ifdef HAVE_SERVER_SUPPORT
+       struct task_struct *task;
+#endif /* HAVE_SERVER_SUPPORT */
+       int i;
+       int rc = 0;
+
         ENTRY;
 
         if (ldlm_state != NULL)
@@ -2789,7 +2814,7 @@ static int ldlm_setup(void)
         rc = ldlm_proc_setup();
         if (rc != 0)
                GOTO(out, rc);
-#endif
+#endif /* LPROCFS */
 
        memset(&conf, 0, sizeof(conf));
        conf = (typeof(conf)) {
@@ -2871,7 +2896,7 @@ static int ldlm_setup(void)
                ldlm_state->ldlm_cancel_service = NULL;
                GOTO(out, rc);
        }
-#endif
+#endif /* HAVE_SERVER_SUPPORT */
 
        OBD_ALLOC(blp, sizeof(*blp));
        if (blp == NULL)
@@ -2879,13 +2904,12 @@ static int ldlm_setup(void)
        ldlm_state->ldlm_bl_pool = blp;
 
        spin_lock_init(&blp->blp_lock);
-       CFS_INIT_LIST_HEAD(&blp->blp_list);
-       CFS_INIT_LIST_HEAD(&blp->blp_prio_list);
+       INIT_LIST_HEAD(&blp->blp_list);
+       INIT_LIST_HEAD(&blp->blp_prio_list);
        init_waitqueue_head(&blp->blp_waitq);
        atomic_set(&blp->blp_num_threads, 0);
        atomic_set(&blp->blp_busy_threads, 0);
 
-#ifdef __KERNEL__
        if (ldlm_num_threads == 0) {
                blp->blp_min_threads = LDLM_NTHRS_INIT;
                blp->blp_max_threads = LDLM_NTHRS_MAX;
@@ -2901,31 +2925,31 @@ static int ldlm_setup(void)
                        GOTO(out, rc);
        }
 
-# ifdef HAVE_SERVER_SUPPORT
-       CFS_INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks);
+#ifdef HAVE_SERVER_SUPPORT
+       INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks);
        expired_lock_thread.elt_state = ELT_STOPPED;
        init_waitqueue_head(&expired_lock_thread.elt_waitq);
 
-       CFS_INIT_LIST_HEAD(&waiting_locks_list);
+       INIT_LIST_HEAD(&waiting_locks_list);
        spin_lock_init(&waiting_locks_spinlock);
        cfs_timer_init(&waiting_locks_timer, waiting_locks_callback, 0);
 
-       rc = PTR_ERR(kthread_run(expired_lock_main, NULL, "ldlm_elt"));
-       if (IS_ERR_VALUE(rc)) {
+       task = kthread_run(expired_lock_main, NULL, "ldlm_elt");
+       if (IS_ERR(task)) {
+               rc = PTR_ERR(task);
                CERROR("Cannot start ldlm expired-lock thread: %d\n", rc);
                GOTO(out, rc);
        }
 
        wait_event(expired_lock_thread.elt_waitq,
                       expired_lock_thread.elt_state == ELT_READY);
-# endif /* HAVE_SERVER_SUPPORT */
+#endif /* HAVE_SERVER_SUPPORT */
 
        rc = ldlm_pools_init();
        if (rc) {
                CERROR("Failed to initialize LDLM pools: %d\n", rc);
                GOTO(out, rc);
        }
-#endif
        RETURN(0);
 
  out:
@@ -2937,15 +2961,14 @@ static int ldlm_cleanup(void)
 {
         ENTRY;
 
-        if (!cfs_list_empty(ldlm_namespace_list(LDLM_NAMESPACE_SERVER)) ||
-            !cfs_list_empty(ldlm_namespace_list(LDLM_NAMESPACE_CLIENT))) {
+       if (!list_empty(ldlm_namespace_list(LDLM_NAMESPACE_SERVER)) ||
+           !list_empty(ldlm_namespace_list(LDLM_NAMESPACE_CLIENT))) {
                 CERROR("ldlm still has namespaces; clean these up first.\n");
                 ldlm_dump_all_namespaces(LDLM_NAMESPACE_SERVER, D_DLMTRACE);
                 ldlm_dump_all_namespaces(LDLM_NAMESPACE_CLIENT, D_DLMTRACE);
                 RETURN(-EBUSY);
         }
 
-#ifdef __KERNEL__
         ldlm_pools_fini();
 
        if (ldlm_state->ldlm_bl_pool != NULL) {
@@ -2957,7 +2980,7 @@ static int ldlm_cleanup(void)
                        init_completion(&blp->blp_comp);
 
                        spin_lock(&blp->blp_lock);
-                       cfs_list_add_tail(&blwi.blwi_entry, &blp->blp_list);
+                       list_add_tail(&blwi.blwi_entry, &blp->blp_list);
                        wake_up(&blp->blp_waitq);
                        spin_unlock(&blp->blp_lock);
 
@@ -2966,27 +2989,24 @@ static int ldlm_cleanup(void)
 
                OBD_FREE(blp, sizeof(*blp));
        }
-#endif /* __KERNEL__ */
 
        if (ldlm_state->ldlm_cb_service != NULL)
                ptlrpc_unregister_service(ldlm_state->ldlm_cb_service);
-# ifdef HAVE_SERVER_SUPPORT
+#ifdef HAVE_SERVER_SUPPORT
        if (ldlm_state->ldlm_cancel_service != NULL)
                ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service);
-# endif
+#endif
 
-#ifdef __KERNEL__
        ldlm_proc_cleanup();
 
-# ifdef HAVE_SERVER_SUPPORT
+#ifdef HAVE_SERVER_SUPPORT
        if (expired_lock_thread.elt_state != ELT_STOPPED) {
                expired_lock_thread.elt_state = ELT_TERMINATE;
                wake_up(&expired_lock_thread.elt_waitq);
                wait_event(expired_lock_thread.elt_waitq,
                               expired_lock_thread.elt_state == ELT_STOPPED);
        }
-# endif
-#endif /* __KERNEL__ */
+#endif
 
         OBD_FREE(ldlm_state, sizeof(*ldlm_state));
         ldlm_state = NULL;
@@ -2999,6 +3019,11 @@ int ldlm_init(void)
        mutex_init(&ldlm_ref_mutex);
        mutex_init(ldlm_namespace_lock(LDLM_NAMESPACE_SERVER));
        mutex_init(ldlm_namespace_lock(LDLM_NAMESPACE_CLIENT));
+
+       INIT_LIST_HEAD(&ldlm_srv_namespace_list);
+       INIT_LIST_HEAD(&ldlm_cli_active_namespace_list);
+       INIT_LIST_HEAD(&ldlm_cli_inactive_namespace_list);
+
        ldlm_resource_slab = kmem_cache_create("ldlm_resources",
                                               sizeof(struct ldlm_resource), 0,
                                               SLAB_HWCACHE_ALIGN, NULL);
@@ -3032,12 +3057,10 @@ void ldlm_exit(void)
        if (ldlm_refcount)
                CERROR("ldlm_refcount is %d in ldlm_exit!\n", ldlm_refcount);
        kmem_cache_destroy(ldlm_resource_slab);
-#ifdef __KERNEL__
        /* ldlm_lock_put() use RCU to call ldlm_lock_free, so need call
         * synchronize_rcu() to wait a grace period elapsed, so that
         * ldlm_lock_free() get a chance to be called. */
        synchronize_rcu();
-#endif
        kmem_cache_destroy(ldlm_lock_slab);
        kmem_cache_destroy(ldlm_interval_slab);
 }