LU-17705 ptlrpc: replace synchronize_rcu() with rcu_barrier()

[fs/lustre-release.git] / lustre / ldlm / ldlm_lockd.c

/*
 * GPL HEADER START
 *
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * 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.
 *
 * 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).
 *
 * You should have received a copy of the GNU General Public License
 * version 2 along with this program; If not, see
 * http://www.gnu.org/licenses/gpl-2.0.html
 *
 * GPL HEADER END
 */
/*
 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright (c) 2010, 2017, Intel Corporation.
 */
/*
 * This file is part of Lustre, http://www.lustre.org/
 *
 * lustre/ldlm/ldlm_lockd.c
 *
 * Author: Peter Braam <braam@clusterfs.com>
 * Author: Phil Schwan <phil@clusterfs.com>
 */

#define DEBUG_SUBSYSTEM S_LDLM

#include <linux/kthread.h>
#include <linux/list.h>
#include <libcfs/libcfs.h>
#include <libcfs/linux/linux-mem.h>
#include <lustre_errno.h>
#include <lustre_dlm.h>
#include <obd_class.h>
#include "ldlm_internal.h"

static int ldlm_num_threads;
module_param(ldlm_num_threads, int, 0444);
MODULE_PARM_DESC(ldlm_num_threads, "number of DLM service threads to start");

static unsigned int ldlm_cpu_bind = 1;
module_param(ldlm_cpu_bind, uint, 0444);
MODULE_PARM_DESC(ldlm_cpu_bind,
                 "bind DLM service threads to particular CPU partitions");

static char *ldlm_cpts;
module_param(ldlm_cpts, charp, 0444);
MODULE_PARM_DESC(ldlm_cpts, "CPU partitions ldlm threads should run on");

static DEFINE_MUTEX(ldlm_ref_mutex);
static int ldlm_refcount;

struct kobject *ldlm_kobj;
struct kset *ldlm_ns_kset;
struct kset *ldlm_svc_kset;

/* LDLM state */

static struct ldlm_state *ldlm_state;

/*
 * timeout for initial callback (AST) reply (bz10399)
 * Due to having to send a 32 bit time value over the
 * wire return it as timeout_t instead of time64_t
 */
static inline timeout_t ldlm_get_rq_timeout(void)
{
        /* Non-AT value */
        timeout_t timeout = min(ldlm_timeout, obd_timeout / 3);

        return timeout < 1 ? 1 : timeout;
}

struct ldlm_bl_pool {
        spinlock_t              blp_lock;

        /*
         * blp_prio_list is used for callbacks that should be handled
         * as a priority. It is used for LDLM_FL_DISCARD_DATA requests.
         * see b=13843
         */
        struct list_head        blp_prio_list;

        /*
         * blp_list is used for all other callbacks which are likely
         * to take longer to process.
         */
        struct list_head        blp_list;

        wait_queue_head_t       blp_waitq;
        struct completion       blp_comp;
        atomic_t                blp_num_threads;
        atomic_t                blp_busy_threads;
        int                     blp_min_threads;
        int                     blp_max_threads;
        int                     blp_total_locks;
        int                     blp_total_blwis;
};

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;
        enum ldlm_cancel_flags  blwi_flags;
        int                     blwi_mem_pressure;
};

#ifdef HAVE_SERVER_SUPPORT

/**
 * Protects both waiting_locks_list and expired_lock_thread.
 */
static DEFINE_SPINLOCK(waiting_locks_spinlock); /* BH lock (timer) */

/**
 * List for contended locks.
 *
 * As soon as a lock is contended, it gets placed on this list and
 * expected time to get a response is filled in the lock. A special
 * thread walks the list looking for locks that should be released and
 * schedules client evictions for those that have not been released in
 * time.
 *
 * All access to it should be under waiting_locks_spinlock.
 */
static LIST_HEAD(waiting_locks_list);
static void waiting_locks_callback(TIMER_DATA_TYPE unused);
static CFS_DEFINE_TIMER(waiting_locks_timer, waiting_locks_callback, 0, 0);

enum elt_state {
        ELT_STOPPED,
        ELT_READY,
        ELT_TERMINATE,
};

static DECLARE_WAIT_QUEUE_HEAD(expired_lock_wait_queue);
static enum elt_state expired_lock_thread_state = ELT_STOPPED;
static int expired_lock_dump;
static LIST_HEAD(expired_lock_list);

static int ldlm_lock_busy(struct ldlm_lock *lock);
static int ldlm_add_waiting_lock(struct ldlm_lock *lock, timeout_t timeout);
static int __ldlm_add_waiting_lock(struct ldlm_lock *lock, timeout_t timeout);

static inline int have_expired_locks(void)
{
        int need_to_run;

        ENTRY;
        spin_lock_bh(&waiting_locks_spinlock);
        need_to_run = !list_empty(&expired_lock_list);
        spin_unlock_bh(&waiting_locks_spinlock);

        RETURN(need_to_run);
}

/**
 * Check expired lock list for expired locks and time them out.
 */
static int expired_lock_main(void *arg)
{
        struct list_head *expired = &expired_lock_list;
        int do_dump;

        ENTRY;

        expired_lock_thread_state = ELT_READY;
        wake_up(&expired_lock_wait_queue);

        while (1) {
                wait_event_idle(expired_lock_wait_queue,
                                have_expired_locks() ||
                                expired_lock_thread_state == ELT_TERMINATE);

                spin_lock_bh(&waiting_locks_spinlock);
                if (expired_lock_dump) {
                        spin_unlock_bh(&waiting_locks_spinlock);

                        /* from waiting_locks_callback, but not in timer */
                        libcfs_debug_dumplog();

                        spin_lock_bh(&waiting_locks_spinlock);
                        expired_lock_dump = 0;
                }

                do_dump = 0;

                while (!list_empty(expired)) {
                        struct obd_export *export;
                        struct ldlm_lock *lock;

                        lock = list_first_entry(expired, struct ldlm_lock,
                                                l_pending_chain);
                        if ((void *)lock < LP_POISON + PAGE_SIZE &&
                            (void *)lock >= LP_POISON) {
                                spin_unlock_bh(&waiting_locks_spinlock);
                                CERROR("free lock on elt list %p\n", lock);
                                LBUG();
                        }
                        list_del_init(&lock->l_pending_chain);
                        if ((void *)lock->l_export <
                             LP_POISON + PAGE_SIZE &&
                            (void *)lock->l_export >= LP_POISON) {
                                CERROR("lock with free export on elt list %p\n",
                                       lock->l_export);
                                lock->l_export = NULL;
                                LDLM_ERROR(lock, "free export");
                                /*
                                 * release extra ref grabbed by
                                 * ldlm_add_waiting_lock() or
                                 * ldlm_failed_ast()
                                 */
                                LDLM_LOCK_RELEASE(lock);
                                continue;
                        }

                        if (ldlm_is_destroyed(lock)) {
                                /*
                                 * release the lock refcount where
                                 * waiting_locks_callback() founds
                                 */
                                LDLM_LOCK_RELEASE(lock);
                                continue;
                        }
                        export = class_export_lock_get(lock->l_export, lock);
                        spin_unlock_bh(&waiting_locks_spinlock);

                        /* Check if we need to prolong timeout */
                        if (!CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_HPREQ_TIMEOUT) &&
                            lock->l_callback_timestamp != 0 && /* not AST err */
                            ldlm_lock_busy(lock)) {
                                LDLM_DEBUG(lock, "prolong the busy lock");
                                lock_res_and_lock(lock);
                                ldlm_add_waiting_lock(lock,
                                                ldlm_bl_timeout(lock) >> 1);
                                unlock_res_and_lock(lock);
                        } else {
                                spin_lock_bh(&export->exp_bl_list_lock);
                                list_del_init(&lock->l_exp_list);
                                spin_unlock_bh(&export->exp_bl_list_lock);

                                LDLM_ERROR(lock,
                                           "lock callback timer expired after %llds: evicting client at %s ",
                                           ktime_get_seconds() -
                                           lock->l_blast_sent,
                                           obd_export_nid2str(export));
                                ldlm_lock_to_ns(lock)->ns_timeouts++;
                                if (do_dump_on_eviction(export->exp_obd))
                                        do_dump++;
                                class_fail_export(export);
                        }
                        class_export_lock_put(export, lock);
                        /*
                         * release extra ref grabbed by ldlm_add_waiting_lock()
                         * or ldlm_failed_ast()
                         */
                        LDLM_LOCK_RELEASE(lock);

                        spin_lock_bh(&waiting_locks_spinlock);
                }
                spin_unlock_bh(&waiting_locks_spinlock);

                if (do_dump) {
                        CERROR("dump the log upon eviction\n");
                        libcfs_debug_dumplog();
                }

                if (expired_lock_thread_state == ELT_TERMINATE)
                        break;
        }

        expired_lock_thread_state = ELT_STOPPED;
        wake_up(&expired_lock_wait_queue);
        RETURN(0);
}

/**
 * Check if there is a request in the export request list
 * which prevents the lock canceling.
 */
static int ldlm_lock_busy(struct ldlm_lock *lock)
{
        struct ptlrpc_request *req;
        int match = 0;

        ENTRY;

        if (lock->l_export == NULL)
                return 0;

        spin_lock(&lock->l_export->exp_rpc_lock);
        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);
                        if (match)
                                break;
                }
        }
        spin_unlock(&lock->l_export->exp_rpc_lock);
        RETURN(match);
}

/* This is called from within a timer interrupt and cannot schedule */
static void waiting_locks_callback(TIMER_DATA_TYPE unused)
{
        struct ldlm_lock *lock;
        int need_dump = 0;

        spin_lock_bh(&waiting_locks_spinlock);
        while (!list_empty(&waiting_locks_list)) {
                lock = list_first_entry(&waiting_locks_list, struct ldlm_lock,
                                        l_pending_chain);
                if (lock->l_callback_timestamp > ktime_get_seconds() ||
                    lock->l_req_mode == LCK_GROUP)
                        break;

                /*
                 * 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
                 */
                list_move(&lock->l_pending_chain, &expired_lock_list);
                need_dump = 1;
        }

        if (!list_empty(&expired_lock_list)) {
                if (obd_dump_on_timeout && need_dump)
                        expired_lock_dump = __LINE__;

                wake_up(&expired_lock_wait_queue);
        }

        /*
         * Make sure the timer will fire again if we have any locks
         * left.
         */
        if (!list_empty(&waiting_locks_list)) {
                time64_t now = ktime_get_seconds();
                timeout_t delta = 0;

                lock = list_first_entry(&waiting_locks_list, struct ldlm_lock,
                                        l_pending_chain);
                if (lock->l_callback_timestamp - now > 0)
                        delta = lock->l_callback_timestamp - now;
                mod_timer(&waiting_locks_timer,
                          jiffies + cfs_time_seconds(delta));
        }
        spin_unlock_bh(&waiting_locks_spinlock);
}

/**
 * Add lock to the list of contended locks.
 *
 * Indicate that we're waiting for a client to call us back cancelling a given
 * lock.  We add it to the pending-callback chain, and schedule the lock-timeout
 * timer to fire appropriately.  (We round up to the next second, to avoid
 * floods of timer firings during periods of high lock contention and traffic).
 * As done by ldlm_add_waiting_lock(), the caller must grab a lock reference
 * if it has been added to the waiting list (1 is returned).
 *
 * Called with the namespace lock held.
 */
static int __ldlm_add_waiting_lock(struct ldlm_lock *lock, timeout_t delay)
{
        unsigned long timeout_jiffies = jiffies;
        time64_t deadline;
        timeout_t timeout;

        lock->l_blast_sent = ktime_get_seconds();
        if (!list_empty(&lock->l_pending_chain))
                return 0;

        if (CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_HPREQ_NOTIMEOUT) ||
            CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_HPREQ_TIMEOUT))
                delay = 1;

        deadline = lock->l_blast_sent + delay;
        if (likely(deadline > lock->l_callback_timestamp))
                lock->l_callback_timestamp = deadline;

        timeout = clamp_t(timeout_t,
                          lock->l_callback_timestamp - lock->l_blast_sent,
                          0, delay);
        timeout_jiffies += cfs_time_seconds(timeout);

        if (time_before(timeout_jiffies, waiting_locks_timer.expires) ||
            !timer_pending(&waiting_locks_timer))
                mod_timer(&waiting_locks_timer, timeout_jiffies);

        /*
         * 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 */
        list_add_tail(&lock->l_pending_chain, &waiting_locks_list);
        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 (!ldlm_is_granted(lock))
                        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, timeout_t timeout)
{
        struct obd_device *obd = NULL;
        int at_off, ret;

        /* NB: must be called with hold of lock_res_and_lock() */
        LASSERT(ldlm_is_res_locked(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) {
                obd = lock->l_export->exp_obd;

                if (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 time64_t next;

                spin_unlock_bh(&waiting_locks_spinlock);
                LDLM_ERROR(lock, "not waiting on destroyed lock (b=5653)");
                if (ktime_get_seconds() > next) {
                        next = ktime_get_seconds() + 14400;
                        dump_stack();
                }
                return 0;
        }

        ldlm_set_waited(lock);
        ret = __ldlm_add_waiting_lock(lock, timeout);
        if (ret) {
                /*
                 * grab ref on the lock if it has been added to the
                 * waiting list
                 */
                LDLM_LOCK_GET(lock);
        }
        spin_unlock_bh(&waiting_locks_spinlock);

        if (ret)
                ldlm_add_blocked_lock(lock);

        at_off = obd_at_off(obd);
        LDLM_DEBUG(lock, "%sadding to wait list(timeout: %d, AT: %s)",
                   ret == 0 ? "not re-" : "", timeout,
                   at_off ? "off" : "on");
        return ret;
}

/**
 * Remove a lock from the pending list, likely because it had its cancellation
 * callback arrive without incident.  This adjusts the lock-timeout timer if
 * needed.  Returns 0 if the lock wasn't pending after all, 1 if it was.
 * As done by ldlm_del_waiting_lock(), the caller must release the lock
 * reference when the lock is removed from any list (1 is returned).
 *
 * Called with namespace lock held.
 */
static int __ldlm_del_waiting_lock(struct ldlm_lock *lock)
{
        struct list_head *list_next;

        if (list_empty(&lock->l_pending_chain))
                return 0;

        list_next = lock->l_pending_chain.next;
        if (lock->l_pending_chain.prev == &waiting_locks_list) {
                /* Removing the head of the list, adjust timer. */
                if (list_next == &waiting_locks_list) {
                        /* No more, just cancel. */
                        timer_delete(&waiting_locks_timer);
                } else {
                        time64_t now = ktime_get_seconds();
                        struct ldlm_lock *next;
                        timeout_t delta = 0;

                        next = list_entry(list_next, struct ldlm_lock,
                                          l_pending_chain);
                        if (next->l_callback_timestamp - now > 0)
                                delta = lock->l_callback_timestamp - now;

                        mod_timer(&waiting_locks_timer,
                                  jiffies + cfs_time_seconds(delta));
                }
        }
        list_del_init(&lock->l_pending_chain);

        return 1;
}

int ldlm_del_waiting_lock(struct ldlm_lock *lock)
{
        int ret;

        if (lock->l_export == NULL) {
                /* We don't have a "waiting locks list" on clients. */
                CDEBUG(D_DLMTRACE, "Client lock %p : no-op\n", lock);
                return 0;
        }

        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 */
        spin_lock_bh(&lock->l_export->exp_bl_list_lock);
        list_del_init(&lock->l_exp_list);
        spin_unlock_bh(&lock->l_export->exp_bl_list_lock);

        if (ret) {
                /*
                 * release lock ref if it has indeed been removed
                 * from a list
                 */
                LDLM_LOCK_RELEASE(lock);
        }

        LDLM_DEBUG(lock, "%s", ret == 0 ? "wasn't waiting" : "removed");
        return ret;
}

/**
 * Prolong the contended lock waiting time.
 *
 * Called with namespace lock held.
 */
int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, timeout_t timeout)
{
        if (lock->l_export == NULL) {
                /* We don't have a "waiting locks list" on clients. */
                LDLM_DEBUG(lock, "client lock: no-op");
                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)) {
                spin_unlock_bh(&waiting_locks_spinlock);
                LDLM_DEBUG(lock, "wasn't waiting");
                return 0;
        }

        /*
         * we remove/add the lock to the waiting list, so no needs to
         * release/take a lock reference
         */
        __ldlm_del_waiting_lock(lock);
        __ldlm_add_waiting_lock(lock, timeout);
        spin_unlock_bh(&waiting_locks_spinlock);

        LDLM_DEBUG(lock, "refreshed to %ds", timeout);
        return 1;
}
EXPORT_SYMBOL(ldlm_refresh_waiting_lock);

#else /* HAVE_SERVER_SUPPORT */

int ldlm_del_waiting_lock(struct ldlm_lock *lock)
{
        RETURN(0);
}

int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, timeout_t timeout)
{
        RETURN(0);
}

#endif /* !HAVE_SERVER_SUPPORT */

#ifdef HAVE_SERVER_SUPPORT

/**
 * Calculate the per-export Blocking timeout (covering BL AST, data flush,
 * lock cancel, and their replies). Used for lock callback timeout and AST
 * re-send period.
 *
 * \param[in] lock        lock which is getting the blocking callback
 *
 * \retval            timeout in seconds to wait for the client reply
 */
timeout_t ldlm_bl_timeout(struct ldlm_lock *lock)
{
        timeout_t timeout;
        struct obd_device *obd = lock->l_export->exp_obd;

        if (obd_at_off(obd))
                return obd_timeout / 2;

        /*
         * Since these are non-updating timeouts, we should be conservative.
         * Take more than usually, 150%
         * It would be nice to have some kind of "early reply" mechanism for
         * lock callbacks too...
         */
        timeout = obd_at_get(obd, &lock->l_export->exp_bl_lock_at);
        return max_t(timeout_t, timeout + (timeout >> 1),
                     (timeout_t)obd_get_ldlm_enqueue_min(obd));
}
EXPORT_SYMBOL(ldlm_bl_timeout);

/**
 * Calculate the per-export Blocking timeout by the given RPC (covering the
 * reply to this RPC and the next RPC). The next RPC could be still not CANCEL,
 * but having the lock refresh mechanism it is enough.
 *
 * Used for lock refresh timeout when we are in the middle of the process -
 * BL AST is sent, CANCEL is ahead - it is still 1 reply for the current RPC
 * and at least 1 RPC (which will trigger another refresh if it will be not
 * CANCEL) - but more accurate than ldlm_bl_timeout as the timeout is taken
 * from the RPC (i.e. the view of the client on the current AT) is taken into
 * account.
 *
 * \param[in] req     req which export needs the timeout calculation
 *
 * \retval            timeout in seconds to wait for the next client's RPC
 */
timeout_t ldlm_bl_timeout_by_rpc(struct ptlrpc_request *req)
{
        struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
        timeout_t timeout, req_timeout, at_timeout, netl;
        struct obd_device *obd = req->rq_export->exp_obd;

        if (obd_at_off(obd))
                return obd_timeout / 2;

        /* A blocked lock means somebody in the cluster is waiting, and we
         * should not consider the worst ever case, consisting of a chain of
         * failures on each step, however this timeout should survive a
         * recovery of at least 1 failure, let this one to be the worst one:
         * in case a server NID is dead first re-connect is done through the
         * same router and also times out.
         *
         * Either this on the next RPC times out, take the max.
         * Considering the current RPC, take just the left time.
         */
        netl = obd_at_get(obd,
                          &req->rq_export->exp_imp_reverse->imp_at.iat_net_latency);
        req_timeout = req->rq_deadline - ktime_get_real_seconds() + netl;
        at_timeout = at_est2timeout(obd_at_get(obd, &svcpt->scp_at_estimate))
                                    + netl;
        req_timeout = max(req_timeout, at_timeout);

        /* Take 1 re-connect failure and 1 re-connect success into account. */
        timeout = at_timeout + INITIAL_CONNECT_TIMEOUT + netl + req_timeout;

        /* Client's timeout is calculated as at_est2timeout(), let's be a bit
         * more conservative than client
         */
        return max(timeout + (timeout >> 4),
                   (timeout_t)obd_get_ldlm_enqueue_min(obd));
}
EXPORT_SYMBOL(ldlm_bl_timeout_by_rpc);

/**
 * Perform lock cleanup if AST sending failed.
 */
static void ldlm_failed_ast(struct ldlm_lock *lock, int rc,
                            const char *ast_type)
{
        LCONSOLE_ERROR_MSG(0x138,
                           "%s: A client on nid %s was evicted due to a lock %s callback time out: rc %d\n",
                           lock->l_export->exp_obd->obd_name,
                           obd_export_nid2str(lock->l_export), ast_type, rc);

        if (obd_dump_on_timeout)
                libcfs_debug_dumplog();
        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);
        /* differentiate it from expired locks */
        lock->l_callback_timestamp = 0;
        list_add(&lock->l_pending_chain, &expired_lock_list);
        wake_up(&expired_lock_wait_queue);
        spin_unlock_bh(&waiting_locks_spinlock);
}

/**
 * Perform lock cleanup if AST reply came with error.
 */
static int ldlm_handle_ast_error(struct ldlm_lock *lock,
                                 struct ptlrpc_request *req, int rc,
                                 const char *ast_type)
{
        struct lnet_processid *peer = &req->rq_import->imp_connection->c_peer;

        if (!req->rq_replied || (rc && rc != -EINVAL)) {
                if (ldlm_is_cancel(lock)) {
                        LDLM_DEBUG(lock,
                                   "%s AST (req@%p x%llu) timeout from nid %s, but cancel was received (AST reply lost?)",
                                   ast_type, req, req->rq_xid,
                                   libcfs_nidstr(&peer->nid));
                        ldlm_lock_cancel(lock);
                        rc = -ERESTART;
                } else if (rc == -ENODEV || rc == -ESHUTDOWN ||
                           (rc == -EIO &&
                            req->rq_import->imp_state == LUSTRE_IMP_CLOSED)) {
                        /*
                         * Upon umount process the AST fails because cannot be
                         * sent. This shouldn't lead to the client eviction.
                         * -ENODEV error is returned by ptl_send_rpc() for
                         *  new request in such import.
                         * -SHUTDOWN is returned by ptlrpc_import_delay_req()
                         *  if imp_invalid is set or obd_no_recov.
                         * Meanwhile there is also check for LUSTRE_IMP_CLOSED
                         * in ptlrpc_import_delay_req() as well with -EIO code.
                         * In all such cases errors are ignored.
                         */
                        LDLM_DEBUG(lock,
                                   "%s AST can't be sent due to a server %s failure or umount process: rc = %d",
                                    ast_type,
                                     req->rq_import->imp_obd->obd_name, rc);
                } else {
                        LDLM_ERROR(lock,
                                   "client (nid %s) %s %s AST (req@%p x%llu status %d rc %d), evict it",
                                   libcfs_nidstr(&peer->nid),
                                   req->rq_replied ? "returned error from" :
                                   "failed to reply to",
                                   ast_type, req, req->rq_xid,
                                   (req->rq_repmsg != NULL) ?
                                   lustre_msg_get_status(req->rq_repmsg) : 0,
                                   rc);
                        ldlm_failed_ast(lock, rc, ast_type);
                }
                return rc;
        }

        if (rc == -EINVAL) {
                struct ldlm_resource *res = lock->l_resource;

                LDLM_DEBUG(lock,
                           "client (nid %s) returned %d from %s AST (req@%p x%llu) - normal race",
                           libcfs_nidstr(&peer->nid),
                           req->rq_repmsg ?
                           lustre_msg_get_status(req->rq_repmsg) : -1,
                           ast_type, req, req->rq_xid);
                if (res) {
                        /*
                         * update lvbo to return proper attributes.
                         * see b=23174
                         */
                        ldlm_resource_getref(res);
                        ldlm_lvbo_update(res, lock, NULL, 1);
                        ldlm_resource_putref(res);
                }
                ldlm_lock_cancel(lock);
                rc = -ERESTART;
        }

        return rc;
}

static int ldlm_cb_interpret(const struct lu_env *env,
                             struct ptlrpc_request *req, void *args, int rc)
{
        struct ldlm_cb_async_args *ca = args;
        struct ldlm_lock *lock = ca->ca_lock;
        struct ldlm_cb_set_arg *arg  = ca->ca_set_arg;

        ENTRY;

        LASSERT(lock != NULL);

        switch (arg->type) {
        case LDLM_GL_CALLBACK:
                /*
                 * Update the LVB from disk if the AST failed
                 * (this is a legal race)
                 *
                 * - Glimpse callback of local lock just returns
                 *   -ELDLM_NO_LOCK_DATA.
                 * - Glimpse callback of remote lock might return
                 *   -ELDLM_NO_LOCK_DATA when inode is cleared. LU-274
                 */
                if (unlikely(arg->gl_interpret_reply)) {
                        rc = arg->gl_interpret_reply(NULL, req, args, rc);
                } else if (rc == -ELDLM_NO_LOCK_DATA) {
                        LDLM_DEBUG(lock,
                                   "lost race - client has a lock but no inode");
                        ldlm_lvbo_update(lock->l_resource, lock, NULL, 1);
                } else if (rc != 0) {
                        rc = ldlm_handle_ast_error(lock, req, rc, "glimpse");
                } else {
                        rc = ldlm_lvbo_update(lock->l_resource,
                                              lock, req, 1);
                }
                break;
        case LDLM_BL_CALLBACK:
                if (rc != 0)
                        rc = ldlm_handle_ast_error(lock, req, rc, "blocking");
                break;
        case LDLM_CP_CALLBACK:
                if (rc != 0)
                        rc = ldlm_handle_ast_error(lock, req, rc, "completion");
                break;
        default:
                LDLM_ERROR(lock, "invalid opcode for lock callback %d",
                           arg->type);
                LBUG();
        }

        /* release extra reference taken in ldlm_ast_fini() */
        LDLM_LOCK_RELEASE(lock);

        if (rc == -ERESTART)
                atomic_inc(&arg->restart);

        RETURN(0);
}

static void ldlm_update_resend(struct ptlrpc_request *req, void *data)
{
        struct ldlm_cb_async_args *ca = data;
        struct ldlm_lock *lock = ca->ca_lock;

        ldlm_refresh_waiting_lock(lock, ldlm_bl_timeout(lock));
}

static inline int ldlm_ast_fini(struct ptlrpc_request *req,
                                struct ldlm_cb_set_arg *arg,
                                struct ldlm_lock *lock,
                                int instant_cancel)
{
        int rc = 0;

        ENTRY;

        if (unlikely(instant_cancel)) {
                rc = ptl_send_rpc(req, 1);
                ptlrpc_req_finished(req);
                if (rc == 0)
                        atomic_inc(&arg->restart);
        } else {
                LDLM_LOCK_GET(lock);
                ptlrpc_set_add_req(arg->set, req);
        }

        RETURN(rc);
}

/**
 * Check if there are requests in the export request list which prevent
 * the lock canceling and make these requests high priority ones.
 */
static void ldlm_lock_reorder_req(struct ldlm_lock *lock)
{
        struct ptlrpc_request *req;

        ENTRY;

        if (lock->l_export == NULL) {
                LDLM_DEBUG(lock, "client lock: no-op");
                RETURN_EXIT;
        }

        spin_lock(&lock->l_export->exp_rpc_lock);
        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
                 * holding svcpt->scp_req_lock, and then redo the check with
                 * the lock held once we need to obtain a reliable result.
                 */
                if (ptlrpc_nrs_req_can_move(req) &&
                    req->rq_ops->hpreq_lock_match &&
                    req->rq_ops->hpreq_lock_match(req, lock))
                        ptlrpc_nrs_req_hp_move(req);
        }
        spin_unlock(&lock->l_export->exp_rpc_lock);
        EXIT;
}

/**
 * ->l_blocking_ast() method for server-side locks. This is invoked when newly
 * enqueued server lock conflicts with given one.
 *
 * Sends blocking AST RPC to the client owning that lock; arms timeout timer
 * to wait for client response.
 */
int ldlm_server_blocking_ast(struct ldlm_lock *lock,
                             struct ldlm_lock_desc *desc,
                             void *data, int flag)
{
        struct ldlm_cb_async_args *ca;
        struct ldlm_cb_set_arg *arg = data;
        struct ldlm_request *body;
        struct ptlrpc_request  *req;
        int instant_cancel = 0;
        int rc = 0;
        struct obd_device *obd;

        ENTRY;

        if (flag == LDLM_CB_CANCELING)
                /* Don't need to do anything here. */
                RETURN(0);

        if (CFS_FAIL_PRECHECK(OBD_FAIL_LDLM_SRV_BL_AST)) {
                LDLM_DEBUG(lock, "dropping BL AST");
                RETURN(0);
        }

        LASSERT(lock);
        LASSERT(data != NULL);

        obd = lock->l_export->exp_obd;
        if (obd->obd_recovering != 0)
                LDLM_ERROR(lock, "BUG 6063: lock collide during recovery");

        ldlm_lock_reorder_req(lock);

        req = ptlrpc_request_alloc_pack(lock->l_export->exp_imp_reverse,
                                        &RQF_LDLM_BL_CALLBACK,
                                        LUSTRE_DLM_VERSION, LDLM_BL_CALLBACK);
        if (req == NULL)
                RETURN(-ENOMEM);

        ca = ptlrpc_req_async_args(ca, req);
        ca->ca_set_arg = arg;
        ca->ca_lock = lock;

        req->rq_interpret_reply = ldlm_cb_interpret;

        lock_res_and_lock(lock);
        if (ldlm_is_destroyed(lock)) {
                /* What's the point? */
                unlock_res_and_lock(lock);
                ptlrpc_req_finished(req);
                RETURN(0);
        }

        if (!ldlm_is_granted(lock)) {
                /*
                 * 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);
        }

        if (ldlm_is_cancel_on_block(lock))
                instant_cancel = 1;

        body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
        body->lock_handle[0] = lock->l_remote_handle;
        body->lock_handle[1].cookie = lock->l_handle.h_cookie;
        body->lock_desc = *desc;
        body->lock_flags |= ldlm_flags_to_wire(lock->l_flags &
                                               LDLM_FL_AST_MASK);

        LDLM_DEBUG(lock, "server preparing blocking AST");

        ptlrpc_request_set_replen(req);
        ldlm_set_cbpending(lock);
        if (instant_cancel) {
                unlock_res_and_lock(lock);
                ldlm_lock_cancel(lock);

                req->rq_no_resend = 1;
        } else {
                LASSERT(ldlm_is_granted(lock));
                ldlm_add_waiting_lock(lock, ldlm_bl_timeout(lock));
                unlock_res_and_lock(lock);

                /* Do not resend after lock callback timeout */
                req->rq_delay_limit = ldlm_bl_timeout(lock);
                req->rq_resend_cb = ldlm_update_resend;
        }

        req->rq_send_state = LUSTRE_IMP_FULL;
        /* ptlrpc_request_alloc_pack already set timeout */
        if (obd_at_off(obd))
                req->rq_timeout = ldlm_get_rq_timeout();

        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,
                                     LDLM_BL_CALLBACK - LDLM_FIRST_OPC);

        rc = ldlm_ast_fini(req, arg, lock, instant_cancel);

        RETURN(rc);
}

/**
 * ->l_completion_ast callback for a remote lock in server namespace.
 *
 *  Sends AST to the client notifying it of lock granting.  If initial
 *  lock response was not sent yet, instead of sending another RPC, just
 *  mark the lock as granted and client will understand
 */
int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data)
{
        struct ldlm_cb_set_arg *arg = data;
        struct ldlm_request *body;
        struct ptlrpc_request *req;
        struct ldlm_cb_async_args *ca;
        int instant_cancel = 0;
        int rc = 0;
        int lvb_len;
        struct obd_device *obd;

        ENTRY;

        LASSERT(lock != NULL);
        LASSERT(data != NULL);

        if (CFS_FAIL_PRECHECK(OBD_FAIL_LDLM_SRV_CP_AST)) {
                LDLM_DEBUG(lock, "dropping CP AST");
                RETURN(0);
        }

        obd = lock->l_export->exp_obd;
        req = ptlrpc_request_alloc(lock->l_export->exp_imp_reverse,
                                   &RQF_LDLM_CP_CALLBACK);
        if (req == NULL)
                RETURN(-ENOMEM);

        /* server namespace, doesn't need lock */
        lvb_len = ldlm_lvbo_size(lock);
        /*
         * LU-3124 & LU-2187: to not return layout in completion AST because
         * it may deadlock for LU-2187, or client may not have enough space
         * for large layout. The layout will be returned to client with an
         * extra RPC to fetch xattr.lov
         */
        if (ldlm_has_layout(lock))
                lvb_len = 0;

        req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_CLIENT, lvb_len);
        rc = ptlrpc_request_pack(req, LUSTRE_DLM_VERSION, LDLM_CP_CALLBACK);
        if (rc) {
                ptlrpc_request_free(req);
                RETURN(rc);
        }

        ca = ptlrpc_req_async_args(ca, req);
        ca->ca_set_arg = arg;
        ca->ca_lock = lock;

        req->rq_interpret_reply = ldlm_cb_interpret;
        body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);

        body->lock_handle[0] = lock->l_remote_handle;
        body->lock_handle[1].cookie = lock->l_handle.h_cookie;
        body->lock_flags = ldlm_flags_to_wire(flags);
        ldlm_lock2desc(lock, &body->lock_desc);
        if (lvb_len > 0) {
                void *lvb = req_capsule_client_get(&req->rq_pill, &RMF_DLM_LVB);

                lvb_len = ldlm_lvbo_fill(lock, lvb, &lvb_len);
                if (lvb_len < 0) {
                        /*
                         * We still need to send the RPC to wake up the blocked
                         * enqueue thread on the client.
                         *
                         * Consider old client, there is no better way to notify
                         * the failure, just zero-sized the LVB, then the client
                         * will fail out as "-EPROTO".
                         */
                        req_capsule_shrink(&req->rq_pill, &RMF_DLM_LVB, 0,
                                           RCL_CLIENT);
                        instant_cancel = 1;
                } else {
                        req_capsule_shrink(&req->rq_pill, &RMF_DLM_LVB, lvb_len,
                                           RCL_CLIENT);
                }
        }

        LDLM_DEBUG(lock, "server preparing completion AST");

        ptlrpc_request_set_replen(req);

        req->rq_send_state = LUSTRE_IMP_FULL;
        /* ptlrpc_request_pack already set timeout */
        if (obd_at_off(obd))
                req->rq_timeout = ldlm_get_rq_timeout();

        /* We only send real blocking ASTs after the lock is granted */
        lock_res_and_lock(lock);
        if (ldlm_is_ast_sent(lock)) {
                body->lock_flags |= ldlm_flags_to_wire(LDLM_FL_AST_SENT);
                /* Copy AST flags like LDLM_FL_DISCARD_DATA. */
                body->lock_flags |= ldlm_flags_to_wire(lock->l_flags &
                                                       LDLM_FL_AST_MASK);

                /*
                 * We might get here prior to ldlm_handle_enqueue setting
                 * LDLM_FL_CANCEL_ON_BLOCK flag. Then we will put this lock
                 * into waiting list, but this is safe and similar code in
                 * ldlm_handle_enqueue will call ldlm_lock_cancel() still,
                 * that would not only cancel the lock, but will also remove
                 * it from waiting list
                 */
                if (ldlm_is_cancel_on_block(lock)) {
                        unlock_res_and_lock(lock);
                        ldlm_lock_cancel(lock);

                        instant_cancel = 1;
                        req->rq_no_resend = 1;

                        lock_res_and_lock(lock);
                } else {
                        /* start the lock-timeout clock */
                        ldlm_add_waiting_lock(lock, ldlm_bl_timeout(lock));
                        /* Do not resend after lock callback timeout */
                        req->rq_delay_limit = ldlm_bl_timeout(lock);
                        req->rq_resend_cb = ldlm_update_resend;
                }
        }
        unlock_res_and_lock(lock);

        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,
                                     LDLM_CP_CALLBACK - LDLM_FIRST_OPC);

        rc = ldlm_ast_fini(req, arg, lock, instant_cancel);

        RETURN(lvb_len < 0 ? lvb_len : rc);
}
EXPORT_SYMBOL(ldlm_server_completion_ast);

/**
 * Server side ->l_glimpse_ast handler for client locks.
 *
 * Sends glimpse AST to the client and waits for reply. Then updates
 * lvbo with the result.
 */
int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data)
{
        struct ldlm_cb_set_arg *arg = data;
        struct ldlm_request *body;
        struct ptlrpc_request *req;
        struct ldlm_cb_async_args *ca;
        int rc;
        struct req_format *req_fmt;
        struct obd_device *obd = lock->l_export->exp_obd;

        ENTRY;

        LASSERT(lock != NULL);

        if (arg->gl_desc != NULL)
                /* There is a glimpse descriptor to pack */
                req_fmt = &RQF_LDLM_GL_CALLBACK_DESC;
        else
                req_fmt = &RQF_LDLM_GL_CALLBACK;

        req = ptlrpc_request_alloc_pack(lock->l_export->exp_imp_reverse,
                                        req_fmt, LUSTRE_DLM_VERSION,
                                        LDLM_GL_CALLBACK);

        if (req == NULL)
                RETURN(-ENOMEM);

        if (arg->gl_desc != NULL) {
                /* copy the GL descriptor */
                union ldlm_gl_desc      *desc;

                desc = req_capsule_client_get(&req->rq_pill, &RMF_DLM_GL_DESC);
                *desc = *arg->gl_desc;
        }

        body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
        body->lock_handle[0] = lock->l_remote_handle;
        ldlm_lock2desc(lock, &body->lock_desc);

        ca = ptlrpc_req_async_args(ca, req);
        ca->ca_set_arg = arg;
        ca->ca_lock = lock;

        /* server namespace, doesn't need lock */
        req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
                             ldlm_lvbo_size(lock));
        ptlrpc_request_set_replen(req);

        req->rq_send_state = LUSTRE_IMP_FULL;
        /* ptlrpc_request_alloc_pack already set timeout */
        if (obd_at_off(obd))
                req->rq_timeout = ldlm_get_rq_timeout();

        req->rq_interpret_reply = ldlm_cb_interpret;

        if (lock->l_export && lock->l_export->exp_nid_stats) {
                struct nid_stat *nid_stats = lock->l_export->exp_nid_stats;

                lprocfs_counter_incr(nid_stats->nid_ldlm_stats,
                                     LDLM_GL_CALLBACK - LDLM_FIRST_OPC);
        }

        rc = ldlm_ast_fini(req, arg, lock, 0);

        RETURN(rc);
}
EXPORT_SYMBOL(ldlm_server_glimpse_ast);

int ldlm_glimpse_locks(struct ldlm_resource *res,
                       struct list_head *gl_work_list)
{
        int rc;

        ENTRY;

        rc = ldlm_run_ast_work(ldlm_res_to_ns(res), gl_work_list,
                               LDLM_WORK_GL_AST);
        if (rc == -ERESTART)
                ldlm_reprocess_all(res, 0);

        RETURN(rc);
}
EXPORT_SYMBOL(ldlm_glimpse_locks);

/* return LDLM lock associated with a lock callback request */
struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req)
{
        struct ldlm_cb_async_args *ca;
        struct ldlm_lock *lock;

        ENTRY;

        ca = ptlrpc_req_async_args(ca, req);
        lock = ca->ca_lock;
        if (lock == NULL)
                RETURN(ERR_PTR(-EFAULT));

        RETURN(lock);
}
EXPORT_SYMBOL(ldlm_request_lock);

/**
 * Main server-side entry point into LDLM for enqueue. This is called by ptlrpc
 * service threads to carry out client lock enqueueing requests.
 */
int ldlm_handle_enqueue(struct ldlm_namespace *ns,
                        struct req_capsule *pill,
                        const struct ldlm_request *dlm_req,
                        const struct ldlm_callback_suite *cbs)
{
        struct ldlm_reply *dlm_rep;
        __u64 flags;
        enum ldlm_error err = ELDLM_OK;
        struct ldlm_lock *lock = NULL;
        void *cookie = NULL;
        int rc = 0;
        struct ldlm_resource *res = NULL;
        struct ptlrpc_request *req = pill->rc_req;
        const struct lu_env *env = req->rq_svc_thread->t_env;

        ENTRY;

        LDLM_DEBUG_NOLOCK("server-side enqueue handler START");

        LASSERT(req && req->rq_export);

        if (req_capsule_ptlreq(pill))
                ldlm_request_cancel(req, dlm_req, LDLM_ENQUEUE_CANCEL_OFF,
                                    LATF_SKIP);

        flags = ldlm_flags_from_wire(dlm_req->lock_flags);

        /* for intent enqueue the stat will be updated inside intent policy */
        if (ptlrpc_req2svc(req)->srv_stats != NULL &&
            !(dlm_req->lock_flags & LDLM_FL_HAS_INTENT))
                ldlm_svc_get_eopc(dlm_req, ptlrpc_req2svc(req)->srv_stats);

        if (req->rq_export->exp_nid_stats &&
            req->rq_export->exp_nid_stats->nid_ldlm_stats)
                lprocfs_counter_incr(req->rq_export->exp_nid_stats->nid_ldlm_stats,
                                     LDLM_ENQUEUE - LDLM_FIRST_OPC);

        if (unlikely(dlm_req->lock_desc.l_resource.lr_type < LDLM_MIN_TYPE ||
                     dlm_req->lock_desc.l_resource.lr_type >= LDLM_MAX_TYPE)) {
                DEBUG_REQ(D_ERROR, req, "invalid lock request type %d",
                          dlm_req->lock_desc.l_resource.lr_type);
                GOTO(out, rc = -EFAULT);
        }

        if (unlikely(dlm_req->lock_desc.l_req_mode <= LCK_MINMODE ||
                     dlm_req->lock_desc.l_req_mode >= LCK_MAXMODE ||
                     dlm_req->lock_desc.l_req_mode &
                     (dlm_req->lock_desc.l_req_mode-1))) {
                DEBUG_REQ(D_ERROR, req, "invalid lock request mode %d",
                          dlm_req->lock_desc.l_req_mode);
                GOTO(out, rc = -EFAULT);
        }

        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()
                 */
                lock = cfs_hash_lookup(req->rq_export->exp_lock_hash,
                                       (void *)&dlm_req->lock_handle[0]);
                if (lock != NULL) {
                        DEBUG_REQ(D_DLMTRACE, req,
                                  "found existing lock cookie %#llx",
                                  lock->l_handle.h_cookie);
                        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");
                        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,
                                dlm_req->lock_desc.l_resource.lr_type,
                                dlm_req->lock_desc.l_req_mode,
                                cbs, NULL, 0, LVB_T_NONE);
        if (IS_ERR(lock)) {
                rc = PTR_ERR(lock);
                lock = NULL;
                GOTO(out, rc);
        }

        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_DEBUG(lock, "delayed lvb init failed (rc %d)", rc);
                        GOTO(out, rc);
                }
        }

        CFS_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 (b=3822) or server umount (b=24324).
         * Cancel it now instead.
         */
        if (req->rq_export->exp_disconnected) {
                LDLM_ERROR(lock, "lock on disconnected export %p",
                           req->rq_export);
                GOTO(out, rc = -ENOTCONN);
        }

        lock->l_export = class_export_lock_get(req->rq_export, lock);
        if (lock->l_export->exp_lock_hash)
                cfs_hash_add(lock->l_export->exp_lock_hash,
                             &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);

        ldlm_convert_policy_to_local(req->rq_export,
                                     dlm_req->lock_desc.l_resource.lr_type,
                                     &dlm_req->lock_desc.l_policy_data,
                                     &lock->l_policy_data);
        if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT) {
                lock->l_req_extent = lock->l_policy_data.l_extent;
        } else if (dlm_req->lock_desc.l_resource.lr_type == LDLM_IBITS) {
                lock->l_policy_data.l_inodebits.try_bits =
                        dlm_req->lock_desc.l_policy_data.l_inodebits.try_bits;
                lock->l_policy_data.l_inodebits.li_gid =
                        dlm_req->lock_desc.l_policy_data.l_inodebits.li_gid;
        }

existing_lock:
        cookie = req;
        if (!(flags & LDLM_FL_HAS_INTENT)) {
                /* based on the assumption that lvb size never changes during
                 * resource life time otherwise it need resource->lr_lock's
                 * protection
                 */
                req_capsule_set_size(pill, &RMF_DLM_LVB,
                                     RCL_SERVER, ldlm_lvbo_size(lock));

                if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR))
                        GOTO(out, rc = -ENOMEM);

                rc = req_capsule_server_pack(pill);
                if (rc)
                        GOTO(out, rc);
        }

        err = ldlm_lock_enqueue(env, ns, &lock, cookie, &flags);
        if (err) {
                if ((int)err < 0)
                        rc = (int)err;
                GOTO(out, err);
        }

        dlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);

        ldlm_lock2desc(lock, &dlm_rep->lock_desc);
        ldlm_lock2handle(lock, &dlm_rep->lock_handle);

        if (lock->l_resource->lr_type == LDLM_EXTENT)
                CFS_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);

        /*
         * 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 = 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 (b=5683) or server umount (b=24324).
         * Cancel it now instead.
         */
        if (unlikely(req->rq_export->exp_disconnected ||
                     CFS_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT))) {
                LDLM_ERROR(lock, "lock on destroyed export %p", req->rq_export);
                rc = -ENOTCONN;
        } else if (ldlm_is_ast_sent(lock)) {
                /* fill lock desc for possible lock convert */
                if (lock->l_blocking_lock &&
                    lock->l_resource->lr_type == LDLM_IBITS) {
                        struct ldlm_lock *bl_lock = lock->l_blocking_lock;
                        struct ldlm_lock_desc *rep_desc = &dlm_rep->lock_desc;

                        LDLM_DEBUG(lock,
                                   "save blocking bits %llx in granted lock",
                                   bl_lock->l_policy_data.l_inodebits.bits);
                        /*
                         * If lock is blocked then save blocking ibits
                         * in returned lock policy for the possible lock
                         * convert on a client.
                         */
                        rep_desc->l_policy_data.l_inodebits.cancel_bits =
                                bl_lock->l_policy_data.l_inodebits.bits;
                }
                dlm_rep->lock_flags |= ldlm_flags_to_wire(LDLM_FL_AST_SENT);
                if (ldlm_is_granted(lock)) {
                        /*
                         * Only cancel lock if it was granted, because it would
                         * be destroyed immediately and would never be granted
                         * in the future, causing timeouts on client.  Not
                         * granted lock will be cancelled immediately after
                         * sending completion AST.
                         */
                        if (ldlm_is_cancel_on_block(lock)) {
                                unlock_res_and_lock(lock);
                                ldlm_lock_cancel(lock);
                                lock_res_and_lock(lock);
                        } else {
                                ldlm_add_waiting_lock(lock,
                                                      ldlm_bl_timeout(lock));
                        }
                }
        }
        unlock_res_and_lock(lock);

        EXIT;
out:
        if (req_capsule_ptlreq(pill)) {
                req->rq_status = rc ?: err; /* return either error - b=11190 */
                if (!req->rq_packed_final) {
                        int rc1 = lustre_pack_reply(req, 1, NULL, NULL);

                        if (rc == 0)
                                rc = rc1;
                }
        }

        /*
         * The LOCK_CHANGED code in ldlm_lock_enqueue depends on this
         * ldlm_reprocess_all.  If this moves, revisit that code. -phil
         */
        if (lock != NULL) {
                LDLM_DEBUG(lock,
                           "server-side enqueue handler, sending reply (err=%d, rc=%d)",
                           err, rc);

                if (rc == 0 &&
                    req_capsule_has_field(pill, &RMF_DLM_LVB,
                                          RCL_SERVER) &&
                    ldlm_lvbo_size(lock) > 0) {
                        void *buf;
                        int buflen;

retry:
                        buf = req_capsule_server_get(pill, &RMF_DLM_LVB);
                        LASSERTF(buf != NULL, "req %px, lock %px\n", req, lock);
                        buflen = req_capsule_get_size(pill, &RMF_DLM_LVB,
                                                      RCL_SERVER);
                        /*
                         * non-replayed lock, delayed lvb init may
                         * need to be occur now
                         */
                        if ((buflen > 0) && !(flags & LDLM_FL_REPLAY)) {
                                int rc2;

                                rc2 = ldlm_lvbo_fill(lock, buf, &buflen);
                                if (rc2 >= 0) {
                                        req_capsule_shrink(pill, &RMF_DLM_LVB,
                                                           rc2, RCL_SERVER);
                                } else if (rc2 == -ERANGE) {
                                        rc2 = req_capsule_server_grow(
                                                        pill, &RMF_DLM_LVB,
                                                        buflen);
                                        if (!rc2) {
                                                goto retry;
                                        } else {
                                                /*
                                                 * if we can't grow the buffer,
                                                 * it's ok to return empty lvb
                                                 * to client.
                                                 */
                                                req_capsule_shrink(
                                                        pill, &RMF_DLM_LVB, 0,
                                                        RCL_SERVER);
                                        }
                                } else {
                                        rc = rc2;
                                }
                        } else if (flags & LDLM_FL_REPLAY) {
                                /* no LVB resend upon replay */
                                if (buflen > 0)
                                        req_capsule_shrink(pill, &RMF_DLM_LVB,
                                                           0, RCL_SERVER);
                                else
                                        rc = buflen;
                        } else {
                                rc = buflen;
                        }
                }

                if (rc != 0 && !(flags & LDLM_FL_RESENT)) {
                        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);
                        }
                        ldlm_reprocess_all(lock->l_resource,
                                           lock->l_policy_data.l_inodebits.bits);
                }

                if (!err && !ldlm_is_cbpending(lock) &&
                    dlm_req->lock_desc.l_resource.lr_type != LDLM_FLOCK)
                        ldlm_reprocess_all(lock->l_resource,
                                           lock->l_policy_data.l_inodebits.bits);

                LDLM_LOCK_RELEASE(lock);
        }

        LDLM_DEBUG_NOLOCK("server-side enqueue handler END (lock %p, rc %d)",
                          lock, rc);

        return rc;
}
EXPORT_SYMBOL(ldlm_handle_enqueue);

/*
 * Clear the blocking lock, the race is possible between ldlm_handle_convert0()
 * and ldlm_work_bl_ast_lock(), so this is done under lock with check for NULL.
 */
void ldlm_clear_blocking_lock(struct ldlm_lock *lock)
{
        if (lock->l_blocking_lock) {
                LDLM_LOCK_RELEASE(lock->l_blocking_lock);
                lock->l_blocking_lock = NULL;
        }
}

/* A lock can be converted to new ibits or mode and should be considered
 * as new lock. Clear all states related to a previous blocking AST
 * processing so new conflicts will cause new blocking ASTs.
 *
 * This is used during lock convert below and lock downgrade to COS mode in
 * ldlm_lock_mode_downgrade().
 */
void ldlm_clear_blocking_data(struct ldlm_lock *lock)
{
        ldlm_clear_ast_sent(lock);
        lock->l_bl_ast_run = 0;
        ldlm_clear_blocking_lock(lock);
}

/* Main LDLM entry point for server code to process lock conversion requests */
int ldlm_handle_convert0(struct ptlrpc_request *req,
                         const struct ldlm_request *dlm_req)
{
        struct obd_export *exp = req->rq_export;
        struct ldlm_reply *dlm_rep;
        struct ldlm_lock *lock;
        __u64 bits;
        __u64 new_bits;
        int rc;

        ENTRY;

        if (exp && exp->exp_nid_stats && exp->exp_nid_stats->nid_ldlm_stats)
                lprocfs_counter_incr(exp->exp_nid_stats->nid_ldlm_stats,
                                     LDLM_CONVERT - LDLM_FIRST_OPC);

        rc = req_capsule_server_pack(&req->rq_pill);
        if (rc)
                RETURN(rc);

        dlm_rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
        dlm_rep->lock_flags = dlm_req->lock_flags;

        lock = ldlm_handle2lock(&dlm_req->lock_handle[0]);
        if (!lock) {
                LDLM_DEBUG_NOLOCK("server lock is canceled already");
                req->rq_status = ELDLM_NO_LOCK_DATA;
                RETURN(0);
        }

        LDLM_DEBUG(lock, "server-side convert handler START");

        lock_res_and_lock(lock);
        bits = lock->l_policy_data.l_inodebits.bits;
        new_bits = dlm_req->lock_desc.l_policy_data.l_inodebits.bits;

        if (ldlm_is_cancel(lock)) {
                LDLM_DEBUG(lock, "convert on canceled lock!");
                unlock_res_and_lock(lock);
                GOTO(out_put, rc = ELDLM_NO_LOCK_DATA);
        }

        if (dlm_req->lock_desc.l_req_mode != lock->l_granted_mode) {
                LDLM_ERROR(lock, "lock mode differs!");
                unlock_res_and_lock(lock);
                GOTO(out_put, rc = -EPROTO);
        }

        if (bits == new_bits) {
                /*
                 * This can be valid situation if CONVERT RPCs are
                 * re-ordered. Just finish silently
                 */
                LDLM_DEBUG(lock, "lock is converted already!");
                unlock_res_and_lock(lock);
        } else {
                if (ldlm_is_waited(lock))
                        ldlm_del_waiting_lock(lock);

                ldlm_clear_cbpending(lock);
                lock->l_policy_data.l_inodebits.cancel_bits = 0;
                ldlm_inodebits_drop(lock, bits & ~new_bits);

                ldlm_clear_blocking_data(lock);
                unlock_res_and_lock(lock);

                /* All old bits should be reprocessed to send new BL AST if
                 * it wasn't sent earlier due to LDLM_FL_AST_SENT bit set.
                 */
                ldlm_reprocess_all(lock->l_resource, bits);
        }

        dlm_rep->lock_handle = lock->l_remote_handle;
        ldlm_ibits_policy_local_to_wire(&lock->l_policy_data,
                                        &dlm_rep->lock_desc.l_policy_data);
        rc = ELDLM_OK;
        EXIT;
out_put:
        LDLM_DEBUG(lock, "server-side convert handler END, rc = %d", rc);
        LDLM_LOCK_PUT(lock);
        req->rq_status = rc;
        return 0;
}

/**
 * Cancel all the locks whose handles are packed into ldlm_request
 *
 * Called by server code expecting such combined cancel activity
 * requests.
 */
int ldlm_request_cancel(struct ptlrpc_request *req,
                        const struct ldlm_request *dlm_req,
                        int first, enum lustre_at_flags flags)
{
        struct ldlm_resource *res, *pres = NULL;
        struct ldlm_lock *lock;
        int i, count, done = 0;
        unsigned int size;

        ENTRY;

        size = req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT);
        if (size <= offsetof(struct ldlm_request, lock_handle) ||
            (size - offsetof(struct ldlm_request, lock_handle)) /
             sizeof(struct lustre_handle) < dlm_req->lock_count)
                RETURN(0);

        count = dlm_req->lock_count ? dlm_req->lock_count : 1;
        if (first >= count)
                RETURN(0);

        if (count == 1 && dlm_req->lock_handle[0].cookie == 0)
                RETURN(0);

        /*
         * There is no lock on the server at the replay time,
         * skip lock cancelling to make replay tests to pass.
         */
        if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
                RETURN(0);

        LDLM_DEBUG_NOLOCK("server-side cancel handler START: %d locks, starting at %d",
                          count, first);

        for (i = first; i < count; i++) {
                lock = ldlm_handle2lock(&dlm_req->lock_handle[i]);
                if (!lock) {
                        /* below message checked in replay-single.sh test_36 */
                        LDLM_DEBUG_NOLOCK("server-side cancel handler stale lock (cookie %llu)",
                                          dlm_req->lock_handle[i].cookie);
                        continue;
                }

                res = lock->l_resource;
                done++;

                /*
                 * This code is an optimization to only attempt lock
                 * granting on the resource (that could be CPU-expensive)
                 * after we are done cancelling lock in that resource.
                 */
                if (res != pres) {
                        if (pres != NULL) {
                                ldlm_reprocess_all(pres, 0);
                                LDLM_RESOURCE_DELREF(pres);
                                ldlm_resource_putref(pres);
                        }
                        if (res != NULL) {
                                ldlm_resource_getref(res);
                                LDLM_RESOURCE_ADDREF(res);

                                if (!ldlm_is_discard_data(lock))
                                        ldlm_lvbo_update(res, lock,
                                                         NULL, 1);
                        }
                        pres = res;
                }

                if ((flags & LATF_STATS) && ldlm_is_ast_sent(lock) &&
                    lock->l_blast_sent != 0) {
                        timeout_t delay = 0;

                        if (ktime_get_seconds() > lock->l_blast_sent)
                                delay = ktime_get_seconds() -
                                        lock->l_blast_sent;
                        LDLM_DEBUG(lock,
                                   "server cancels blocked lock after %ds",
                                   delay);
                        obd_at_measure(lock->l_export->exp_obd,
                                       &lock->l_export->exp_bl_lock_at,
                                       delay);
                }
                ldlm_lock_cancel(lock);
                LDLM_LOCK_PUT(lock);
        }
        if (pres != NULL) {
                ldlm_reprocess_all(pres, 0);
                LDLM_RESOURCE_DELREF(pres);
                ldlm_resource_putref(pres);
        }
        LDLM_DEBUG_NOLOCK("server-side cancel handler END");
        RETURN(done);
}
EXPORT_SYMBOL(ldlm_request_cancel);

/**
 * Main LDLM entry point for server code to cancel locks.
 *
 * Typically gets called from service handler on LDLM_CANCEL opc.
 */
int ldlm_handle_cancel(struct ptlrpc_request *req)
{
        struct ldlm_request *dlm_req;
        int rc;

        ENTRY;

        dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
        if (dlm_req == NULL) {
                CDEBUG(D_INFO, "bad request buffer for cancel\n");
                RETURN(-EFAULT);
        }

        if (req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT) <
            offsetof(struct ldlm_request, lock_handle[1]))
                RETURN(-EPROTO);

        if (req->rq_export && req->rq_export->exp_nid_stats &&
            req->rq_export->exp_nid_stats->nid_ldlm_stats)
                lprocfs_counter_incr(req->rq_export->exp_nid_stats->nid_ldlm_stats,
                                     LDLM_CANCEL - LDLM_FIRST_OPC);

        rc = req_capsule_server_pack(&req->rq_pill);
        if (rc)
                RETURN(rc);

        if (!ldlm_request_cancel(req, dlm_req, 0, LATF_STATS))
                req->rq_status = LUSTRE_ESTALE;

        RETURN(ptlrpc_reply(req));
}
#endif /* HAVE_SERVER_SUPPORT */

/**
 * Server may pass additional information about blocking lock.
 * For IBITS locks it is conflicting bits which can be used for
 * lock convert instead of cancel.
 */
void ldlm_bl_desc2lock(const struct ldlm_lock_desc *ld, struct ldlm_lock *lock)
{
        struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);

        check_res_locked(lock->l_resource);
        if (ns_is_client(ns) && ld &&
            (lock->l_resource->lr_type == LDLM_IBITS)) {
                /*
                 * Lock description contains policy of blocking lock,
                 * and its cancel_bits is used to pass conflicting bits.
                 * NOTE: ld can be NULL or can be not NULL but zeroed if
                 * passed from ldlm_bl_thread_blwi(), check below used bits
                 * in ld to make sure it is valid description.
                 *
                 * If server may replace lock resource keeping the same cookie,
                 * never use cancel bits from different resource, full cancel
                 * is to be used.
                 */
                if (ld->l_policy_data.l_inodebits.cancel_bits &&
                    ldlm_res_eq(&ld->l_resource.lr_name,
                                &lock->l_resource->lr_name) &&
                    !(ldlm_is_cbpending(lock) &&
                      lock->l_policy_data.l_inodebits.cancel_bits == 0)) {
                        /* always combine conflicting ibits */
                        lock->l_policy_data.l_inodebits.cancel_bits |=
                                ld->l_policy_data.l_inodebits.cancel_bits;
                } else {
                        /* If cancel_bits are not obtained or
                         * if the lock is already CBPENDING and
                         * has no cancel_bits set
                         * - the full lock is to be cancelled
                         */
                        lock->l_policy_data.l_inodebits.cancel_bits = 0;
                }
        }
}

/**
 * Callback handler for receiving incoming blocking ASTs.
 *
 * This can only happen on client side.
 */
void ldlm_handle_bl_callback(struct ldlm_namespace *ns,
                             struct ldlm_lock_desc *ld, struct ldlm_lock *lock)
{
        int do_ast;

        ENTRY;

        LDLM_DEBUG(lock, "client blocking AST callback handler");

        lock_res_and_lock(lock);

        /* get extra information from desc if any */
        ldlm_bl_desc2lock(ld, lock);
        ldlm_set_cbpending(lock);

        do_ast = (!lock->l_readers && !lock->l_writers);
        unlock_res_and_lock(lock);

        if (do_ast) {
                CDEBUG(D_DLMTRACE,
                       "Lock %p already unused, calling callback (%p)\n",
                       lock, lock->l_blocking_ast);
                if (lock->l_blocking_ast != NULL)
                        lock->l_blocking_ast(lock, ld, lock->l_ast_data,
                                             LDLM_CB_BLOCKING);
        } else {
                CDEBUG(D_DLMTRACE,
                       "Lock %p is referenced, will be cancelled later\n",
                       lock);
        }

        LDLM_DEBUG(lock, "client blocking callback handler END");
        LDLM_LOCK_RELEASE(lock);
        EXIT;
}

static int ldlm_callback_reply(struct ptlrpc_request *req, int rc)
{
        if (req->rq_no_reply)
                return 0;

        req->rq_status = rc;
        if (!req->rq_packed_final) {
                rc = lustre_pack_reply(req, 1, NULL, NULL);
                if (rc)
                        return rc;
        }
        return ptlrpc_reply(req);
}

/**
 * Callback handler for receiving incoming completion ASTs.
 *
 * This only can happen on client side.
 */
static int ldlm_handle_cp_callback(struct ptlrpc_request *req,
                                   struct ldlm_namespace *ns,
                                   struct ldlm_request *dlm_req,
                                   struct ldlm_lock *lock)
{
        LIST_HEAD(ast_list);
        int lvb_len;
        int rc = 0;

        ENTRY;

        LDLM_DEBUG(lock, "client completion callback handler START");

        if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE)) {
                long to = cfs_time_seconds(1);

                ldlm_callback_reply(req, 0);

                while (to > 0) {
                        to = schedule_timeout_interruptible(to);
                        if (ldlm_is_granted(lock) ||
                            ldlm_is_destroyed(lock))
                                break;
                }
        }

        lvb_len = req_capsule_get_size(&req->rq_pill, &RMF_DLM_LVB, RCL_CLIENT);
        if (lvb_len < 0) {
                LDLM_ERROR(lock, "Fail to get lvb_len, rc = %d", lvb_len);
                GOTO(out, rc = lvb_len);
        } else if (lvb_len > 0) {
                if (lock->l_lvb_len > 0) {
                        /* for extent lock, lvb contains ost_lvb{}. */
                        LASSERT(lock->l_lvb_data != NULL);

                        if (unlikely(lock->l_lvb_len < lvb_len)) {
                                LDLM_ERROR(lock,
                                           "Replied LVB is larger than expectation, expected = %d, replied = %d",
                                           lock->l_lvb_len, lvb_len);
                                GOTO(out, rc = -EINVAL);
                        }
                }
        }

        lock_res_and_lock(lock);

        if (!ldlm_res_eq(&dlm_req->lock_desc.l_resource.lr_name,
                         &lock->l_resource->lr_name)) {
                ldlm_resource_unlink_lock(lock);
                unlock_res_and_lock(lock);
                rc = ldlm_lock_change_resource(ns, lock,
                                &dlm_req->lock_desc.l_resource.lr_name);
                if (rc < 0) {
                        LDLM_ERROR(lock, "Failed to allocate resource");
                        GOTO(out, rc);
                }
                LDLM_DEBUG(lock, "completion AST, new resource");
                lock_res_and_lock(lock);
        }

        if (ldlm_is_failed(lock)) {
                unlock_res_and_lock(lock);
                LDLM_LOCK_RELEASE(lock);
                RETURN(-EINVAL);
        }

        if (ldlm_is_destroyed(lock) ||
            ldlm_is_granted(lock)) {
                /* b=11300: the lock has already been granted */
                unlock_res_and_lock(lock);
                LDLM_DEBUG(lock, "Double grant race happened");
                GOTO(out, rc = 0);
        }

        /*
         * If we receive the completion AST before the actual enqueue returned,
         * then we might need to switch lock modes, resources, or extents.
         */
        if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
                lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
                LDLM_DEBUG(lock, "completion AST, new lock mode");
        }

        if (lock->l_resource->lr_type != LDLM_PLAIN) {
                ldlm_convert_policy_to_local(req->rq_export,
                                          dlm_req->lock_desc.l_resource.lr_type,
                                          &dlm_req->lock_desc.l_policy_data,
                                          &lock->l_policy_data);
                LDLM_DEBUG(lock, "completion AST, new policy data");
        }

        ldlm_resource_unlink_lock(lock);

        if (dlm_req->lock_flags & LDLM_FL_AST_SENT) {
                /*
                 * BL_AST locks are not needed in LRU.
                 * Let ldlm_cancel_lru() be fast.
                 */
                ldlm_lock_remove_from_lru(lock);
                ldlm_bl_desc2lock(&dlm_req->lock_desc, lock);
                lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_BL_AST;
                LDLM_DEBUG(lock, "completion AST includes blocking AST");
        }

        if (lock->l_lvb_len > 0) {
                rc = ldlm_fill_lvb(lock, &req->rq_pill, RCL_CLIENT,
                                   lock->l_lvb_data, lvb_len);
                if (rc < 0) {
                        unlock_res_and_lock(lock);
                        GOTO(out, rc);
                }
        }

        ldlm_grant_lock(lock, &ast_list);
        unlock_res_and_lock(lock);

        LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");

        /*
         * Let Enqueue to call osc_lock_upcall() and initialize
         * l_ast_data
         */
        CFS_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 2);

        ldlm_run_ast_work(ns, &ast_list, LDLM_WORK_CP_AST);

        LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
                          lock);
        GOTO(out, rc);

out:
        if (rc < 0) {
                lock_res_and_lock(lock);
                ldlm_set_failed(lock);
                unlock_res_and_lock(lock);
                wake_up(&lock->l_waitq);
        }
        LDLM_LOCK_RELEASE(lock);

        return 0;
}

/**
 * Callback handler for receiving incoming glimpse ASTs.
 *
 * This only can happen on client side.  After handling the glimpse AST
 * we also consider dropping the lock here if it is unused locally for a
 * long time.
 */
static void ldlm_handle_gl_callback(struct ptlrpc_request *req,
                                    struct ldlm_namespace *ns,
                                    struct ldlm_request *dlm_req,
                                    struct ldlm_lock *lock)
{
        struct ldlm_lock_desc *ld = &dlm_req->lock_desc;
        int rc = -ENOSYS;

        ENTRY;

        LDLM_DEBUG(lock, "client glimpse AST callback handler");

        if (lock->l_glimpse_ast != NULL)
                rc = lock->l_glimpse_ast(lock, req);

        if (req->rq_repmsg != NULL) {
                ptlrpc_reply(req);
        } else {
                req->rq_status = rc;
                ptlrpc_error(req);
        }

        lock_res_and_lock(lock);
        if (lock->l_granted_mode == LCK_PW &&
            !lock->l_readers && !lock->l_writers &&
            ktime_after(ktime_get(),
                        ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
                unlock_res_and_lock(lock);

                /* For MDS glimpse it is always DOM lock, set corresponding
                 * cancel_bits to perform lock convert if needed
                 */
                if (lock->l_resource->lr_type == LDLM_IBITS)
                        ld->l_policy_data.l_inodebits.cancel_bits =
                                                        MDS_INODELOCK_DOM;
                if (ldlm_bl_to_thread_lock(ns, ld, lock))
                        ldlm_handle_bl_callback(ns, ld, lock);

                EXIT;
                return;
        }
        unlock_res_and_lock(lock);
        LDLM_LOCK_RELEASE(lock);
        EXIT;
}

static int __ldlm_bl_to_thread(struct ldlm_bl_work_item *blwi,
                               enum ldlm_cancel_flags cancel_flags)
{
        struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool;
        char *prio = "regular";
        int count;

        ENTRY;

        spin_lock(&blp->blp_lock);
        /* cannot access blwi after added to list and lock is dropped */
        count = blwi->blwi_lock ? 1 : blwi->blwi_count;

        /* if the server is waiting on a lock to be cancelled (bl_ast), this is
         * an urgent request and should go in the priority queue so it doesn't
         * get stuck behind non-priority work (eg, lru size management)
         *
         * We also prioritize discard_data, which is for eviction handling
         */
        if (blwi->blwi_lock &&
            (ldlm_is_discard_data(blwi->blwi_lock) ||
             ldlm_is_bl_ast(blwi->blwi_lock))) {
                list_add_tail(&blwi->blwi_entry, &blp->blp_prio_list);
                prio = "priority";
        } else {
                /* other blocking callbacks are added to the regular list */
                list_add_tail(&blwi->blwi_entry, &blp->blp_list);
        }
        blp->blp_total_locks += count;
        blp->blp_total_blwis++;
        spin_unlock(&blp->blp_lock);

        wake_up(&blp->blp_waitq);

        /* unlocked read of blp values is intentional - OK for debug */
        CDEBUG(D_DLMTRACE,
               "added %d/%d locks to %s blp list, %d blwis in pool\n",
               count, blp->blp_total_locks, prio, blp->blp_total_blwis);

        /*
         * can not check blwi->blwi_flags as blwi could be already freed in
         * LCF_ASYNC mode
         */
        if (!(cancel_flags & LCF_ASYNC))
                wait_for_completion(&blwi->blwi_comp);

        RETURN(0);
}

static inline void init_blwi(struct ldlm_bl_work_item *blwi,
                             struct ldlm_namespace *ns,
                             struct ldlm_lock_desc *ld,
                             struct list_head *cancels, int count,
                             struct ldlm_lock *lock,
                             enum ldlm_cancel_flags cancel_flags)
{
        init_completion(&blwi->blwi_comp);
        INIT_LIST_HEAD(&blwi->blwi_head);

        if (current->flags & PF_MEMALLOC)
                blwi->blwi_mem_pressure = 1;

        blwi->blwi_ns = ns;
        blwi->blwi_flags = cancel_flags;
        if (ld != NULL)
                blwi->blwi_ld = *ld;
        if (count) {
                list_splice_init(cancels, &blwi->blwi_head);
                blwi->blwi_count = count;
        } else {
                blwi->blwi_lock = lock;
        }
}

/**
 * Queues a list of locks \a cancels containing \a count locks
 * for later processing by a blocking thread.  If \a count is zero,
 * then the lock referenced as \a lock is queued instead.
 *
 * The blocking thread would then call ->l_blocking_ast callback in the lock.
 * If list addition fails an error is returned and caller is supposed to
 * call ->l_blocking_ast itself.
 */
static int ldlm_bl_to_thread(struct ldlm_namespace *ns,
                             struct ldlm_lock_desc *ld,
                             struct ldlm_lock *lock,
                             struct list_head *cancels, int count,
                             enum ldlm_cancel_flags cancel_flags)
{
        ENTRY;

        if (cancels && count == 0)
                RETURN(0);

        if (cancel_flags & LCF_ASYNC) {
                struct ldlm_bl_work_item *blwi;

                OBD_ALLOC(blwi, sizeof(*blwi));
                if (blwi == NULL)
                        RETURN(-ENOMEM);
                init_blwi(blwi, ns, ld, cancels, count, lock, cancel_flags);

                RETURN(__ldlm_bl_to_thread(blwi, cancel_flags));
        } else {
                /*
                 * if it is synchronous call do minimum mem alloc, as it could
                 * be triggered from kernel shrinker
                 */
                struct ldlm_bl_work_item blwi;

                memset(&blwi, 0, sizeof(blwi));
                init_blwi(&blwi, ns, ld, cancels, count, lock, cancel_flags);
                RETURN(__ldlm_bl_to_thread(&blwi, cancel_flags));
        }
}


int ldlm_bl_to_thread_lock(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld,
                           struct ldlm_lock *lock)
{
        return ldlm_bl_to_thread(ns, ld, lock, NULL, 0, LCF_ASYNC);
}

int ldlm_bl_to_thread_list(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld,
                           struct list_head *cancels, int count,
                           enum ldlm_cancel_flags cancel_flags)
{
        return ldlm_bl_to_thread(ns, ld, NULL, cancels, count, cancel_flags);
}

int ldlm_bl_to_thread_ns(struct ldlm_namespace *ns)
{
        return ldlm_bl_to_thread(ns, NULL, NULL, NULL, 0, LCF_ASYNC);
}

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)
{
        struct obd_device *obd = req->rq_export->exp_obd;
        char *key;
        void *val;
        int keylen, vallen;
        int rc = -ENOSYS;

        ENTRY;

        DEBUG_REQ(D_HSM, req, "%s: handle setinfo", obd->obd_name);

        req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);

        key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
        if (key == NULL) {
                DEBUG_REQ(D_IOCTL, req, "no set_info key");
                RETURN(-EFAULT);
        }
        keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
                                      RCL_CLIENT);
        val = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
        if (val == NULL) {
                DEBUG_REQ(D_IOCTL, req, "no set_info val");
                RETURN(-EFAULT);
        }
        vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
                                      RCL_CLIENT);

        /* We are responsible for swabbing contents of val */

        if (KEY_IS(KEY_HSM_COPYTOOL_SEND))
                /* Pass it on to mdc (the "export" in this case) */
                rc = obd_set_info_async(req->rq_svc_thread->t_env,
                                        req->rq_export,
                                        sizeof(KEY_HSM_COPYTOOL_SEND),
                                        KEY_HSM_COPYTOOL_SEND,
                                        vallen, val, NULL);
        else
                DEBUG_REQ(D_WARNING, req, "ignoring unknown key '%s'", key);

        return rc;
}

static inline void ldlm_callback_errmsg(struct ptlrpc_request *req,
                                        const char *msg, int rc,
                                        const struct lustre_handle *handle)
{
        DEBUG_REQ((req->rq_no_reply || rc) ? D_WARNING : D_DLMTRACE, req,
                  "%s, NID=%s lock=%#llx: rc = %d",
                  msg, libcfs_idstr(&req->rq_peer),
                  handle ? handle->cookie : 0, rc);
        if (req->rq_no_reply)
                CWARN("No reply was sent, maybe cause b=21636.\n");
        else if (rc)
                CWARN("Send reply failed, maybe cause b=21636.\n");
}

/* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
static int ldlm_callback_handler(struct ptlrpc_request *req)
{
        struct ldlm_namespace *ns;
        struct ldlm_request *dlm_req;
        struct ldlm_lock *lock;
        int rc;

        ENTRY;

        /*
         * Requests arrive in sender's byte order.  The ptlrpc service
         * handler has already checked and, if necessary, byte-swapped the
         * incoming request message body, but I am responsible for the
         * message buffers.
         */

        /* do nothing for sec context finalize */
        if (lustre_msg_get_opc(req->rq_reqmsg) == SEC_CTX_FINI)
                RETURN(0);

        req_capsule_init(&req->rq_pill, req, RCL_SERVER);

        if (req->rq_export == NULL) {
                rc = ldlm_callback_reply(req, -ENOTCONN);
                ldlm_callback_errmsg(req, "Operate on unconnected server",
                                     rc, NULL);
                RETURN(0);
        }

        LASSERT(req->rq_export != NULL);
        LASSERT(req->rq_export->exp_obd != NULL);

        switch (lustre_msg_get_opc(req->rq_reqmsg)) {
        case LDLM_BL_CALLBACK:
                if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET)) {
                        if (cfs_fail_err)
                                ldlm_callback_reply(req, -(int)cfs_fail_err);
                        RETURN(0);
                }
                break;
        case LDLM_CP_CALLBACK:
                if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_CP_CALLBACK_NET))
                        RETURN(0);
                break;
        case LDLM_GL_CALLBACK:
                if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_GL_CALLBACK_NET))
                        RETURN(0);
                break;
        case LDLM_SET_INFO:
                rc = ldlm_handle_setinfo(req);
                ldlm_callback_reply(req, rc);
                RETURN(0);
        default:
                CERROR("unknown opcode %u\n",
                       lustre_msg_get_opc(req->rq_reqmsg));
                ldlm_callback_reply(req, -EPROTO);
                RETURN(0);
        }

        ns = req->rq_export->exp_obd->obd_namespace;
        LASSERT(ns != NULL);

        req_capsule_set(&req->rq_pill, &RQF_LDLM_CALLBACK);

        dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
        if (dlm_req == NULL) {
                rc = ldlm_callback_reply(req, -EPROTO);
                ldlm_callback_errmsg(req, "Operate without parameter", rc,
                                     NULL);
                RETURN(0);
        }

        /*
         * Force a known safe race, send a cancel to the server for a lock
         * which the server has already started a blocking callback on.
         */
        if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE) &&
            lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) {
                rc = ldlm_cli_cancel(&dlm_req->lock_handle[0], 0);
                if (rc < 0)
                        CERROR("ldlm_cli_cancel: %d\n", rc);
        }

        lock = ldlm_handle2lock_long(&dlm_req->lock_handle[0], 0);
        if (!lock) {
                CDEBUG(D_DLMTRACE,
                       "callback on lock %#llx - lock disappeared\n",
                       dlm_req->lock_handle[0].cookie);
                rc = ldlm_callback_reply(req, -EINVAL);
                ldlm_callback_errmsg(req, "Operate with invalid parameter", rc,
                                     &dlm_req->lock_handle[0]);
                RETURN(0);
        }

        if (ldlm_is_fail_loc(lock) &&
            lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK)
                CFS_RACE(OBD_FAIL_LDLM_CP_BL_RACE);

        /* Copy hints/flags (e.g. LDLM_FL_DISCARD_DATA) from AST. */
        lock_res_and_lock(lock);
        lock->l_flags |= ldlm_flags_from_wire(dlm_req->lock_flags &
                                              LDLM_FL_AST_MASK);
        if (lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) {
                /*
                 * If somebody cancels lock and cache is already dropped,
                 * or lock is failed before cp_ast received on client,
                 * we can tell the server we have no lock. Otherwise, we
                 * should send cancel after dropping the cache.
                 */
                if ((ldlm_is_canceling(lock) && ldlm_is_bl_done(lock)) ||
                     ldlm_is_failed(lock)) {
                        LDLM_DEBUG(lock,
                                   "callback on lock %llx - lock disappeared",
                                   dlm_req->lock_handle[0].cookie);
                        unlock_res_and_lock(lock);
                        LDLM_LOCK_RELEASE(lock);
                        rc = ldlm_callback_reply(req, -EINVAL);
                        ldlm_callback_errmsg(req, "Operate on stale lock", rc,
                                             &dlm_req->lock_handle[0]);
                        RETURN(0);
                }
                /*
                 * BL_AST locks are not needed in LRU.
                 * Let ldlm_cancel_lru() be fast.
                 */
                ldlm_lock_remove_from_lru(lock);
                ldlm_set_bl_ast(lock);
        }
        if (lock->l_remote_handle.cookie == 0)
                lock->l_remote_handle = dlm_req->lock_handle[1];
        unlock_res_and_lock(lock);

        /*
         * We want the ost thread to get this reply so that it can respond
         * to ost requests (write cache writeback) that might be triggered
         * in the callback.
         *
         * But we'd also like to be able to indicate in the reply that we're
         * cancelling right now, because it's unused, or have an intent result
         * in the reply, so we might have to push the responsibility for sending
         * the reply down into the AST handlers, alas.
         */

        switch (lustre_msg_get_opc(req->rq_reqmsg)) {
        case LDLM_BL_CALLBACK:
                LDLM_DEBUG(lock, "blocking ast");
                req_capsule_extend(&req->rq_pill, &RQF_LDLM_BL_CALLBACK);
                if (!ldlm_is_cancel_on_block(lock)) {
                        rc = ldlm_callback_reply(req, 0);
                        if (req->rq_no_reply || rc)
                                ldlm_callback_errmsg(req, "Normal process", rc,
                                                     &dlm_req->lock_handle[0]);
                }
                if (ldlm_bl_to_thread_lock(ns, &dlm_req->lock_desc, lock))
                        ldlm_handle_bl_callback(ns, &dlm_req->lock_desc, lock);
                break;
        case LDLM_CP_CALLBACK:
                LDLM_DEBUG(lock, "completion ast");
                req_capsule_extend(&req->rq_pill, &RQF_LDLM_CP_CALLBACK);
                rc = ldlm_handle_cp_callback(req, ns, dlm_req, lock);
                if (!CFS_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE))
                        ldlm_callback_reply(req, rc);
                break;
        case LDLM_GL_CALLBACK:
                LDLM_DEBUG(lock, "glimpse ast");
                req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
                ldlm_handle_gl_callback(req, ns, dlm_req, lock);
                break;
        default:
                LBUG(); /* checked above */
        }

        RETURN(0);
}

#ifdef HAVE_SERVER_SUPPORT
/**
 * Main handler for canceld thread.
 *
 * Separated into its own thread to avoid deadlocks.
 */
static int ldlm_cancel_handler(struct ptlrpc_request *req)
{
        int rc;

        ENTRY;

        /*
         * Requests arrive in sender's byte order.  The ptlrpc service
         * handler has already checked and, if necessary, byte-swapped the
         * incoming request message body, but I am responsible for the
         * message buffers.
         */

        req_capsule_init(&req->rq_pill, req, RCL_SERVER);

        if (req->rq_export == NULL) {
                struct ldlm_request *dlm_req;

                CERROR("%s from %s arrived at %llu with bad export cookie %llu\n",
                       ll_opcode2str(lustre_msg_get_opc(req->rq_reqmsg)),
                       libcfs_nidstr(&req->rq_peer.nid),
                       (unsigned long long)req->rq_arrival_time.tv_sec,
                       lustre_msg_get_handle(req->rq_reqmsg)->cookie);

                if (lustre_msg_get_opc(req->rq_reqmsg) == LDLM_CANCEL) {
                        req_capsule_set(&req->rq_pill, &RQF_LDLM_CALLBACK);
                        dlm_req = req_capsule_client_get(&req->rq_pill,
                                                         &RMF_DLM_REQ);
                        if (dlm_req != NULL)
                                ldlm_lock_dump_handle(D_ERROR,
                                                      &dlm_req->lock_handle[0]);
                }
                ldlm_callback_reply(req, -ENOTCONN);
                RETURN(0);
        }

        switch (lustre_msg_get_opc(req->rq_reqmsg)) {
        /* XXX FIXME move this back to mds/handler.c, b=249 */
        case LDLM_CANCEL:
                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_LDLM_BL_EVICT))
                        RETURN(0);
                rc = ldlm_handle_cancel(req);
                break;
        case LDLM_CONVERT:
        {
                struct ldlm_request *dlm_req;

                req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
                CDEBUG(D_INODE, "convert\n");

                dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
                if (dlm_req == NULL) {
                        CDEBUG(D_INFO, "bad request buffer for cancel\n");
                        rc = ldlm_callback_reply(req, -EPROTO);
                } else {
                        req->rq_status = ldlm_handle_convert0(req, dlm_req);
                        rc = ptlrpc_reply(req);
                }
                break;
        }
        default:
                CERROR("invalid opcode %d\n",
                       lustre_msg_get_opc(req->rq_reqmsg));
                req_capsule_set(&req->rq_pill, &RQF_LDLM_CALLBACK);
                rc = ldlm_callback_reply(req, -EINVAL);
        }

        RETURN(rc);
}

static int ldlm_cancel_hpreq_lock_match(struct ptlrpc_request *req,
                                        struct ldlm_lock *lock)
{
        struct ldlm_request *dlm_req;
        struct lustre_handle lockh;
        int rc = 0;
        int i;

        ENTRY;

        dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
        if (dlm_req == NULL)
                RETURN(0);

        ldlm_lock2handle(lock, &lockh);
        for (i = 0; i < dlm_req->lock_count; i++) {
                if (lustre_handle_equal(&dlm_req->lock_handle[i],
                                        &lockh)) {
                        DEBUG_REQ(D_RPCTRACE, req,
                                  "Prio raised by lock %#llx", lockh.cookie);
                        rc = 1;
                        break;
                }
        }

        RETURN(rc);
}

static int ldlm_cancel_hpreq_check(struct ptlrpc_request *req)
{
        struct ldlm_request *dlm_req;
        int rc = 0;
        int i;
        unsigned int size;

        ENTRY;

        /* no prolong in recovery */
        if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
                RETURN(0);

        dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
        if (dlm_req == NULL)
                RETURN(-EFAULT);

        size = req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT);
        if (size <= offsetof(struct ldlm_request, lock_handle) ||
            (size - offsetof(struct ldlm_request, lock_handle)) /
             sizeof(struct lustre_handle) < dlm_req->lock_count)
                RETURN(-EPROTO);

        for (i = 0; i < dlm_req->lock_count; i++) {
                struct ldlm_lock *lock;

                lock = ldlm_handle2lock(&dlm_req->lock_handle[i]);
                if (lock == NULL)
                        continue;

                rc = ldlm_is_ast_sent(lock) ? 1 : 0;
                if (rc)
                        LDLM_DEBUG(lock, "hpreq cancel/convert lock");
                LDLM_LOCK_PUT(lock);

                if (rc)
                        break;
        }

        RETURN(rc);
}

static struct ptlrpc_hpreq_ops ldlm_cancel_hpreq_ops = {
        .hpreq_lock_match = ldlm_cancel_hpreq_lock_match,
        .hpreq_check      = ldlm_cancel_hpreq_check,
        .hpreq_fini       = NULL,
};

static int ldlm_hpreq_handler(struct ptlrpc_request *req)
{
        ENTRY;

        req_capsule_init(&req->rq_pill, req, RCL_SERVER);

        if (req->rq_export == NULL)
                RETURN(0);

        if (LDLM_CANCEL == lustre_msg_get_opc(req->rq_reqmsg)) {
                req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
                req->rq_ops = &ldlm_cancel_hpreq_ops;
        } else if (LDLM_CONVERT == lustre_msg_get_opc(req->rq_reqmsg)) {
                req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
                req->rq_ops = &ldlm_cancel_hpreq_ops;
        }
        RETURN(0);
}

static int ldlm_revoke_lock_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd,
                               struct hlist_node *hnode, void *data)

{
        struct list_head *rpc_list = data;
        struct ldlm_lock *lock = cfs_hash_object(hs, hnode);

        lock_res_and_lock(lock);

        if (!ldlm_is_granted(lock)) {
                unlock_res_and_lock(lock);
                return 0;
        }

        LASSERT(lock->l_resource);
        if (lock->l_resource->lr_type != LDLM_IBITS &&
            lock->l_resource->lr_type != LDLM_PLAIN) {
                unlock_res_and_lock(lock);
                return 0;
        }

        if (ldlm_is_ast_sent(lock)) {
                unlock_res_and_lock(lock);
                return 0;
        }

        LASSERT(lock->l_blocking_ast);
        LASSERT(!lock->l_blocking_lock);

        ldlm_set_ast_sent(lock);
        if (lock->l_export && lock->l_export->exp_lock_hash) {
                /*
                 * NB: it's safe to call cfs_hash_del() even lock isn't
                 * in exp_lock_hash.
                 */
                /*
                 * In the function below, .hs_keycmp resolves to
                 * ldlm_export_lock_keycmp()
                 */
                cfs_hash_del(lock->l_export->exp_lock_hash,
                             &lock->l_remote_handle, &lock->l_exp_hash);
        }

        list_add_tail(&lock->l_rk_ast, rpc_list);
        LDLM_LOCK_GET(lock);

        unlock_res_and_lock(lock);
        return 0;
}

void ldlm_revoke_export_locks(struct obd_export *exp)
{
        int rc;
        LIST_HEAD(rpc_list);

        ENTRY;

        cfs_hash_for_each_nolock(exp->exp_lock_hash,
                                 ldlm_revoke_lock_cb, &rpc_list, 0);
        rc = ldlm_run_ast_work(exp->exp_obd->obd_namespace, &rpc_list,
                          LDLM_WORK_REVOKE_AST);

        if (rc == -ERESTART)
                ldlm_reprocess_recovery_done(exp->exp_obd->obd_namespace);

        EXIT;
}
EXPORT_SYMBOL(ldlm_revoke_export_locks);
#endif /* HAVE_SERVER_SUPPORT */

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;
        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;
                }
                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 = list_first_entry(&blp->blp_list,
                                        struct ldlm_bl_work_item, blwi_entry);
        else
                if (!list_empty(&blp->blp_prio_list))
                        blwi = list_first_entry(&blp->blp_prio_list,
                                                struct ldlm_bl_work_item,
                                                blwi_entry);

        if (blwi) {
                if (++num_bl >= num_th)
                        num_bl = 0;
                list_del(&blwi->blwi_entry);
                blp->blp_total_locks -= blwi->blwi_lock ? 1 : blwi->blwi_count;
                blp->blp_total_blwis--;
        }
        spin_unlock(&blp->blp_lock);
        *p_blwi = blwi;

        /* intentional unlocked read of blp values - OK for debug */
        if (blwi) {
                CDEBUG(D_DLMTRACE,
                       "Got %d locks of %d total in blp.  (%d blwis in pool)\n",
                       blwi->blwi_lock ? 1 : blwi->blwi_count,
                       blp->blp_total_locks, blp->blp_total_blwis);
        } else {
                CDEBUG(D_DLMTRACE,
                       "No blwi found in queue (no bl locks in queue)\n");
        }

        if (*p_exp != NULL && *p_blwi != NULL) {
                obd_stale_export_put(*p_exp);
                *p_exp = NULL;
        }

        return (*p_blwi != NULL || *p_exp != NULL) ? 1 : 0;
}

/* This only contains temporary data until the thread starts */
struct ldlm_bl_thread_data {
        struct ldlm_bl_pool     *bltd_blp;
        struct completion       bltd_comp;
        int                     bltd_num;
};

static int ldlm_bl_thread_main(void *arg);

static int ldlm_bl_thread_start(struct ldlm_bl_pool *blp, bool check_busy)
{
        struct ldlm_bl_thread_data bltd = { .bltd_blp = blp };
        struct task_struct *task;

        init_completion(&bltd.bltd_comp);

        bltd.bltd_num = atomic_inc_return(&blp->blp_num_threads);
        if (bltd.bltd_num >= blp->blp_max_threads) {
                atomic_dec(&blp->blp_num_threads);
                return 0;
        }

        LASSERTF(bltd.bltd_num > 0, "thread num:%d\n", bltd.bltd_num);
        if (check_busy &&
            atomic_read(&blp->blp_busy_threads) < (bltd.bltd_num - 1)) {
                atomic_dec(&blp->blp_num_threads);
                return 0;
        }

        task = kthread_run(ldlm_bl_thread_main, &bltd, "ldlm_bl_%02d",
                           bltd.bltd_num);
        if (IS_ERR(task)) {
                CERROR("cannot start LDLM thread ldlm_bl_%02d: rc %ld\n",
                       bltd.bltd_num, PTR_ERR(task));
                atomic_dec(&blp->blp_num_threads);
                return PTR_ERR(task);
        }
        wait_for_completion(&bltd.bltd_comp);

        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)
{
        if (atomic_read(&blp->blp_num_threads) >= blp->blp_max_threads)
                return 0;

        if (atomic_read(&blp->blp_busy_threads) <
            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)
{
        /* '1' for consistency with code that checks !mpflag to restore */
        unsigned int mpflags = 1;

        ENTRY;

        if (blwi->blwi_ns == NULL)
                /* added by ldlm_cleanup() */
                RETURN(LDLM_ITER_STOP);

        if (blwi->blwi_mem_pressure)
                mpflags = memalloc_noreclaim_save();

        CFS_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 if (blwi->blwi_lock) {
                ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld,
                                        blwi->blwi_lock);
        } else {
                ldlm_pool_recalc(&blwi->blwi_ns->ns_pool, true);
                spin_lock(&blwi->blwi_ns->ns_lock);
                blwi->blwi_ns->ns_rpc_recalc = 0;
                spin_unlock(&blwi->blwi_ns->ns_lock);
                ldlm_namespace_put(blwi->blwi_ns);
        }

        if (blwi->blwi_mem_pressure)
                memalloc_noreclaim_restore(mpflags);

        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;

        CFS_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.
 *
 * Callers put locks into its queue by calling ldlm_bl_to_thread.
 * This thread in the end ends up doing actual call to ->l_blocking_ast
 * for queued locks.
 */
static int ldlm_bl_thread_main(void *arg)
{
        struct lu_env *env;
        struct ldlm_bl_pool *blp;
        struct ldlm_bl_thread_data *bltd = arg;
        int rc;

        ENTRY;

        OBD_ALLOC_PTR(env);
        if (!env)
                RETURN(-ENOMEM);
        rc = lu_env_init(env, LCT_DT_THREAD);
        if (rc)
                GOTO(out_env, rc);
        rc = lu_env_add(env);
        if (rc)
                GOTO(out_env_fini, rc);

        blp = bltd->bltd_blp;

        complete(&bltd->bltd_comp);
        /* cannot use bltd after this, it is only on caller's stack */

        while (1) {
                struct ldlm_bl_work_item *blwi = NULL;
                struct obd_export *exp = NULL;
                int rc;

                rc = ldlm_bl_get_work(blp, &blwi, &exp);

                if (rc == 0)
                        wait_event_idle_exclusive(blp->blp_waitq,
                                                  ldlm_bl_get_work(blp, &blwi,
                                                                   &exp));
                atomic_inc(&blp->blp_busy_threads);

                if (ldlm_bl_thread_need_create(blp, blwi))
                        /* discard the return value, we tried */
                        ldlm_bl_thread_start(blp, true);

                if (exp)
                        rc = ldlm_bl_thread_exports(blp, exp);
                else if (blwi)
                        rc = ldlm_bl_thread_blwi(blp, blwi);

                atomic_dec(&blp->blp_busy_threads);

                if (rc == LDLM_ITER_STOP)
                        break;

                /*
                 * If there are many namespaces, we will not sleep waiting for
                 * work, and must do a cond_resched to avoid holding the CPU
                 * for too long
                 */
                cond_resched();
        }

        atomic_dec(&blp->blp_num_threads);
        complete(&blp->blp_comp);

        lu_env_remove(env);
out_env_fini:
        lu_env_fini(env);
out_env:
        OBD_FREE_PTR(env);
        RETURN(rc);
}


static int ldlm_setup(void);
static int ldlm_cleanup(void);

int ldlm_get_ref(void)
{
        int rc = 0;

        ENTRY;
        mutex_lock(&ldlm_ref_mutex);
        if (++ldlm_refcount == 1) {
                rc = ldlm_setup();
                if (rc)
                        ldlm_refcount--;
        }
        mutex_unlock(&ldlm_ref_mutex);

        RETURN(rc);
}

void ldlm_put_ref(void)
{
        ENTRY;
        mutex_lock(&ldlm_ref_mutex);
        if (ldlm_refcount == 1) {
                int rc = ldlm_cleanup();

                if (rc)
                        CERROR("ldlm_cleanup failed: %d\n", rc);
                else
                        ldlm_refcount--;
        } else {
                ldlm_refcount--;
        }
        mutex_unlock(&ldlm_ref_mutex);

        EXIT;
}

/* Export handle<->lock hash operations. */
static unsigned
ldlm_export_lock_hash(struct cfs_hash *hs, const void *key,
                      const unsigned int bits)
{
        return cfs_hash_64(((struct lustre_handle *)key)->cookie, bits);
}

static void *
ldlm_export_lock_key(struct hlist_node *hnode)
{
        struct ldlm_lock *lock;

        lock = hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
        return &lock->l_remote_handle;
}

static void
ldlm_export_lock_keycpy(struct hlist_node *hnode, void *key)
{
        struct ldlm_lock     *lock;

        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, struct hlist_node *hnode)
{
        return lustre_handle_equal(ldlm_export_lock_key(hnode), key);
}

static void *
ldlm_export_lock_object(struct hlist_node *hnode)
{
        return hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
}

static void
ldlm_export_lock_get(struct cfs_hash *hs, struct hlist_node *hnode)
{
        struct ldlm_lock *lock;

        lock = hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
        LDLM_LOCK_GET(lock);
}

static void
ldlm_export_lock_put(struct cfs_hash *hs, struct hlist_node *hnode)
{
        struct ldlm_lock *lock;

        lock = hlist_entry(hnode, struct ldlm_lock, l_exp_hash);
        LDLM_LOCK_RELEASE(lock);
}

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,
        .hs_keycpy      = ldlm_export_lock_keycpy,
        .hs_object      = ldlm_export_lock_object,
        .hs_get         = ldlm_export_lock_get,
        .hs_put         = ldlm_export_lock_put,
        .hs_put_locked  = ldlm_export_lock_put,
};

int ldlm_init_export(struct obd_export *exp)
{
        int rc;

        ENTRY;

        exp->exp_lock_hash =
                cfs_hash_create(obd_uuid2str(&exp->exp_client_uuid),
                                HASH_EXP_LOCK_CUR_BITS,
                                HASH_EXP_LOCK_MAX_BITS,
                                HASH_EXP_LOCK_BKT_BITS, 0,
                                CFS_HASH_MIN_THETA, CFS_HASH_MAX_THETA,
                                &ldlm_export_lock_ops,
                                CFS_HASH_DEFAULT | CFS_HASH_REHASH_KEY |
                                CFS_HASH_NBLK_CHANGE);

        if (!exp->exp_lock_hash)
                RETURN(-ENOMEM);

        rc = ldlm_init_flock_export(exp);
        if (rc)
                GOTO(err, rc);

        RETURN(0);
err:
        ldlm_destroy_export(exp);
        RETURN(rc);
}
EXPORT_SYMBOL(ldlm_init_export);

void ldlm_destroy_export(struct obd_export *exp)
{
        ENTRY;
        cfs_hash_putref(exp->exp_lock_hash);
        exp->exp_lock_hash = NULL;

        ldlm_destroy_flock_export(exp);
        EXIT;
}
EXPORT_SYMBOL(ldlm_destroy_export);

static ssize_t cancel_unused_locks_before_replay_show(struct kobject *kobj,
                                                      struct attribute *attr,
                                                      char *buf)
{
        return sprintf(buf, "%d\n", ldlm_cancel_unused_locks_before_replay);
}

static ssize_t cancel_unused_locks_before_replay_store(struct kobject *kobj,
                                                       struct attribute *attr,
                                                       const char *buffer,
                                                       size_t count)
{
        int rc;
        unsigned long val;

        rc = kstrtoul(buffer, 10, &val);
        if (rc)
                return rc;

        ldlm_cancel_unused_locks_before_replay = val;

        return count;
}
LUSTRE_RW_ATTR(cancel_unused_locks_before_replay);

static struct attribute *ldlm_attrs[] = {
        &lustre_attr_cancel_unused_locks_before_replay.attr,
        NULL,
};

static struct attribute_group ldlm_attr_group = {
        .attrs = ldlm_attrs,
};

static int ldlm_setup(void)
{
        static struct ptlrpc_service_conf       conf;
        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)
                RETURN(-EALREADY);

        OBD_ALLOC(ldlm_state, sizeof(*ldlm_state));
        if (ldlm_state == NULL)
                RETURN(-ENOMEM);

        ldlm_kobj = kobject_create_and_add("ldlm", &lustre_kset->kobj);
        if (!ldlm_kobj)
                GOTO(out, -ENOMEM);

        rc = sysfs_create_group(ldlm_kobj, &ldlm_attr_group);
        if (rc)
                GOTO(out, rc);

        ldlm_ns_kset = kset_create_and_add("namespaces", NULL, ldlm_kobj);
        if (!ldlm_ns_kset)
                GOTO(out, -ENOMEM);

        ldlm_svc_kset = kset_create_and_add("services", NULL, ldlm_kobj);
        if (!ldlm_svc_kset)
                GOTO(out, -ENOMEM);

        rc = ldlm_debugfs_setup();
        if (rc != 0)
                GOTO(out, rc);

        memset(&conf, 0, sizeof(conf));
        conf = (typeof(conf)) {
                .psc_name               = "ldlm_cbd",
                .psc_watchdog_factor    = 2,
                .psc_buf                = {
                        .bc_nbufs               = LDLM_CLIENT_NBUFS,
                        .bc_buf_size            = LDLM_BUFSIZE,
                        .bc_req_max_size        = LDLM_MAXREQSIZE,
                        .bc_rep_max_size        = LDLM_MAXREPSIZE,
                        .bc_req_portal          = LDLM_CB_REQUEST_PORTAL,
                        .bc_rep_portal          = LDLM_CB_REPLY_PORTAL,
                },
                .psc_thr                = {
                        .tc_thr_name            = "ldlm_cb",
                        .tc_thr_factor          = LDLM_THR_FACTOR,
                        .tc_nthrs_init          = LDLM_NTHRS_INIT,
                        .tc_nthrs_base          = LDLM_NTHRS_BASE,
                        .tc_nthrs_max           = LDLM_NTHRS_MAX,
                        .tc_nthrs_user          = ldlm_num_threads,
                        .tc_cpu_bind            = ldlm_cpu_bind,
                        .tc_ctx_tags            = LCT_MD_THREAD | LCT_DT_THREAD,
                },
                .psc_cpt                = {
                        .cc_pattern             = ldlm_cpts,
                        .cc_affinity            = true,
                },
                .psc_ops                = {
                        .so_req_handler         = ldlm_callback_handler,
                },
        };
        ldlm_state->ldlm_cb_service = ptlrpc_register_service(&conf,
                                                              ldlm_svc_kset,
                                                              ldlm_svc_debugfs_dir);
        if (IS_ERR(ldlm_state->ldlm_cb_service)) {
                CERROR("failed to start service\n");
                rc = PTR_ERR(ldlm_state->ldlm_cb_service);
                ldlm_state->ldlm_cb_service = NULL;
                GOTO(out, rc);
        }

#ifdef HAVE_SERVER_SUPPORT
        memset(&conf, 0, sizeof(conf));
        conf = (typeof(conf)) {
                .psc_name               = "ldlm_canceld",
                .psc_watchdog_factor    = 6,
                .psc_buf                = {
                        .bc_nbufs               = LDLM_SERVER_NBUFS,
                        .bc_buf_size            = LDLM_BUFSIZE,
                        .bc_req_max_size        = LDLM_MAXREQSIZE,
                        .bc_rep_max_size        = LDLM_MAXREPSIZE,
                        .bc_req_portal          = LDLM_CANCEL_REQUEST_PORTAL,
                        .bc_rep_portal          = LDLM_CANCEL_REPLY_PORTAL,

                },
                .psc_thr                = {
                        .tc_thr_name            = "ldlm_cn",
                        .tc_thr_factor          = LDLM_THR_FACTOR,
                        .tc_nthrs_init          = LDLM_NTHRS_INIT,
                        .tc_nthrs_base          = LDLM_NTHRS_BASE,
                        .tc_nthrs_max           = LDLM_NTHRS_MAX,
                        .tc_nthrs_user          = ldlm_num_threads,
                        .tc_cpu_bind            = ldlm_cpu_bind,
                        .tc_ctx_tags            = LCT_MD_THREAD |
                                                  LCT_DT_THREAD |
                                                  LCT_CL_THREAD,
                },
                .psc_cpt                = {
                        .cc_pattern             = ldlm_cpts,
                        .cc_affinity            = true,
                },
                .psc_ops                = {
                        .so_req_handler         = ldlm_cancel_handler,
                        .so_hpreq_handler       = ldlm_hpreq_handler,
                },
        };
        ldlm_state->ldlm_cancel_service =
                        ptlrpc_register_service(&conf, ldlm_svc_kset,
                                                ldlm_svc_debugfs_dir);
        if (IS_ERR(ldlm_state->ldlm_cancel_service)) {
                CERROR("failed to start service\n");
                rc = PTR_ERR(ldlm_state->ldlm_cancel_service);
                ldlm_state->ldlm_cancel_service = NULL;
                GOTO(out, rc);
        }
#endif /* HAVE_SERVER_SUPPORT */

        OBD_ALLOC(blp, sizeof(*blp));
        if (blp == NULL)
                GOTO(out, rc = -ENOMEM);
        ldlm_state->ldlm_bl_pool = blp;

        spin_lock_init(&blp->blp_lock);
        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);
        blp->blp_total_locks = 0;
        blp->blp_total_blwis = 0;

        if (ldlm_num_threads == 0) {
                blp->blp_min_threads = LDLM_NTHRS_INIT;
                blp->blp_max_threads = LDLM_NTHRS_MAX;
        } else {
                blp->blp_min_threads = blp->blp_max_threads =
                        min_t(int, LDLM_NTHRS_MAX, max_t(int, LDLM_NTHRS_INIT,
                                                         ldlm_num_threads));
        }

        for (i = 0; i < blp->blp_min_threads; i++) {
                rc = ldlm_bl_thread_start(blp, false);
                if (rc < 0)
                        GOTO(out, rc);
        }

#ifdef HAVE_SERVER_SUPPORT
        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_wait_queue,
                   expired_lock_thread_state == ELT_READY);
#endif /* HAVE_SERVER_SUPPORT */

        rc = ldlm_pools_init();
        if (rc) {
                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:
        ldlm_cleanup();
        RETURN(rc);
}

static int ldlm_cleanup(void)
{
        ENTRY;

        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);
        }

        ldlm_reclaim_cleanup();
        ldlm_pools_fini();

        if (ldlm_state->ldlm_bl_pool != NULL) {
                struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool;

                while (atomic_read(&blp->blp_num_threads) > 0) {
                        struct ldlm_bl_work_item blwi = { .blwi_ns = NULL };

                        init_completion(&blp->blp_comp);

                        spin_lock(&blp->blp_lock);
                        list_add_tail(&blwi.blwi_entry, &blp->blp_list);
                        wake_up(&blp->blp_waitq);
                        spin_unlock(&blp->blp_lock);

                        wait_for_completion(&blp->blp_comp);
                }

                OBD_FREE(blp, sizeof(*blp));
        }

        if (ldlm_state->ldlm_cb_service != NULL)
                ptlrpc_unregister_service(ldlm_state->ldlm_cb_service);
#ifdef HAVE_SERVER_SUPPORT
        if (ldlm_state->ldlm_cancel_service != NULL)
                ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service);
#endif

        if (ldlm_ns_kset)
                kset_unregister(ldlm_ns_kset);
        if (ldlm_svc_kset)
                kset_unregister(ldlm_svc_kset);
        if (ldlm_kobj) {
                sysfs_remove_group(ldlm_kobj, &ldlm_attr_group);
                kobject_put(ldlm_kobj);
        }

        ldlm_debugfs_cleanup();

#ifdef HAVE_SERVER_SUPPORT
        if (expired_lock_thread_state != ELT_STOPPED) {
                expired_lock_thread_state = ELT_TERMINATE;
                wake_up(&expired_lock_wait_queue);
                wait_event(expired_lock_wait_queue,
                           expired_lock_thread_state == ELT_STOPPED);
        }
#endif

        OBD_FREE(ldlm_state, sizeof(*ldlm_state));
        ldlm_state = NULL;

        RETURN(0);
}

int ldlm_init(void)
{
        ldlm_resource_slab = kmem_cache_create("ldlm_resources",
                                               sizeof(struct ldlm_resource), 0,
                                               SLAB_HWCACHE_ALIGN, NULL);
        if (ldlm_resource_slab == NULL)
                return -ENOMEM;

        ldlm_lock_slab = kmem_cache_create("ldlm_locks",
                              sizeof(struct ldlm_lock), 0,
                              SLAB_HWCACHE_ALIGN, NULL);
        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)
                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;

#ifdef HAVE_SERVER_SUPPORT
        ldlm_inodebits_slab = kmem_cache_create("ldlm_ibits_node",
                                                sizeof(struct ldlm_ibits_node),
                                                0, SLAB_HWCACHE_ALIGN, NULL);
        if (ldlm_inodebits_slab == NULL)
                goto out_interval_tree;

        ldlm_glimpse_work_kmem = kmem_cache_create("ldlm_glimpse_work_kmem",
                                        sizeof(struct ldlm_glimpse_work),
                                        0, 0, NULL);
        if (ldlm_glimpse_work_kmem == NULL)
                goto out_inodebits;
#endif

#if LUSTRE_TRACKS_LOCK_EXP_REFS
        class_export_dump_hook = ldlm_dump_export_locks;
#endif
        return 0;
#ifdef HAVE_SERVER_SUPPORT
out_inodebits:
        kmem_cache_destroy(ldlm_inodebits_slab);
out_interval_tree:
        kmem_cache_destroy(ldlm_interval_tree_slab);
#endif
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)
{
        if (ldlm_refcount)
                CERROR("ldlm_refcount is %d in %s\n", ldlm_refcount, __func__);
        rcu_barrier();
        kmem_cache_destroy(ldlm_resource_slab);
        /*
         * ldlm_lock_put() use RCU to call ldlm_lock_free, so need call
         * rcu_barrier() to wait all outstanding RCU callbacks to complete,
         * so that ldlm_lock_free() get a chance to be called.
         */
        rcu_barrier();
        kmem_cache_destroy(ldlm_lock_slab);
        kmem_cache_destroy(ldlm_interval_slab);
        kmem_cache_destroy(ldlm_interval_tree_slab);
#ifdef HAVE_SERVER_SUPPORT
        kmem_cache_destroy(ldlm_inodebits_slab);
        kmem_cache_destroy(ldlm_glimpse_work_kmem);
#endif
}