Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / ocfs2 / dlm / dlmunlock.c
diff --git a/kernel/fs/ocfs2/dlm/dlmunlock.c b/kernel/fs/ocfs2/dlm/dlmunlock.c
new file mode 100644 (file)
index 0000000..2e3c9db
--- /dev/null
@@ -0,0 +1,698 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmunlock.c
+ *
+ * underlying calls for unlocking locks
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * 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 for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/spinlock.h>
+#include <linux/delay.h>
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+
+#define MLOG_MASK_PREFIX ML_DLM
+#include "cluster/masklog.h"
+
+#define DLM_UNLOCK_FREE_LOCK           0x00000001
+#define DLM_UNLOCK_CALL_AST            0x00000002
+#define DLM_UNLOCK_REMOVE_LOCK         0x00000004
+#define DLM_UNLOCK_REGRANT_LOCK        0x00000008
+#define DLM_UNLOCK_CLEAR_CONVERT_TYPE  0x00000010
+
+
+static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm,
+                                             struct dlm_lock_resource *res,
+                                             struct dlm_lock *lock,
+                                             struct dlm_lockstatus *lksb,
+                                             int *actions);
+static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm,
+                                             struct dlm_lock_resource *res,
+                                             struct dlm_lock *lock,
+                                             struct dlm_lockstatus *lksb,
+                                             int *actions);
+
+static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm,
+                                                struct dlm_lock_resource *res,
+                                                struct dlm_lock *lock,
+                                                struct dlm_lockstatus *lksb,
+                                                int flags,
+                                                u8 owner);
+
+
+/*
+ * according to the spec:
+ * http://opendlm.sourceforge.net/cvsmirror/opendlm/docs/dlmbook_final.pdf
+ *
+ *  flags & LKM_CANCEL != 0: must be converting or blocked
+ *  flags & LKM_CANCEL == 0: must be granted
+ *
+ * So to unlock a converting lock, you must first cancel the
+ * convert (passing LKM_CANCEL in flags), then call the unlock
+ * again (with no LKM_CANCEL in flags).
+ */
+
+
+/*
+ * locking:
+ *   caller needs:  none
+ *   taken:         res->spinlock and lock->spinlock taken and dropped
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_NOLOCKMGR, status from network
+ * all callers should have taken an extra ref on lock coming in
+ */
+static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm,
+                                       struct dlm_lock_resource *res,
+                                       struct dlm_lock *lock,
+                                       struct dlm_lockstatus *lksb,
+                                       int flags, int *call_ast,
+                                       int master_node)
+{
+       enum dlm_status status;
+       int actions = 0;
+       int in_use;
+        u8 owner;
+
+       mlog(0, "master_node = %d, valblk = %d\n", master_node,
+            flags & LKM_VALBLK);
+
+       if (master_node)
+               BUG_ON(res->owner != dlm->node_num);
+       else
+               BUG_ON(res->owner == dlm->node_num);
+
+       spin_lock(&dlm->ast_lock);
+       /* We want to be sure that we're not freeing a lock
+        * that still has AST's pending... */
+       in_use = !list_empty(&lock->ast_list);
+       spin_unlock(&dlm->ast_lock);
+       if (in_use && !(flags & LKM_CANCEL)) {
+              mlog(ML_ERROR, "lockres %.*s: Someone is calling dlmunlock "
+                   "while waiting for an ast!", res->lockname.len,
+                   res->lockname.name);
+               return DLM_BADPARAM;
+       }
+
+       spin_lock(&res->spinlock);
+       if (res->state & DLM_LOCK_RES_IN_PROGRESS) {
+               if (master_node && !(flags & LKM_CANCEL)) {
+                       mlog(ML_ERROR, "lockres in progress!\n");
+                       spin_unlock(&res->spinlock);
+                       return DLM_FORWARD;
+               }
+               /* ok for this to sleep if not in a network handler */
+               __dlm_wait_on_lockres(res);
+               res->state |= DLM_LOCK_RES_IN_PROGRESS;
+       }
+       spin_lock(&lock->spinlock);
+
+       if (res->state & DLM_LOCK_RES_RECOVERING) {
+               status = DLM_RECOVERING;
+               goto leave;
+       }
+
+       if (res->state & DLM_LOCK_RES_MIGRATING) {
+               status = DLM_MIGRATING;
+               goto leave;
+       }
+
+       /* see above for what the spec says about
+        * LKM_CANCEL and the lock queue state */
+       if (flags & LKM_CANCEL)
+               status = dlm_get_cancel_actions(dlm, res, lock, lksb, &actions);
+       else
+               status = dlm_get_unlock_actions(dlm, res, lock, lksb, &actions);
+
+       if (status != DLM_NORMAL && (status != DLM_CANCELGRANT || !master_node))
+               goto leave;
+
+       /* By now this has been masked out of cancel requests. */
+       if (flags & LKM_VALBLK) {
+               /* make the final update to the lvb */
+               if (master_node)
+                       memcpy(res->lvb, lksb->lvb, DLM_LVB_LEN);
+               else
+                       flags |= LKM_PUT_LVB; /* let the send function
+                                              * handle it. */
+       }
+
+       if (!master_node) {
+               owner = res->owner;
+               /* drop locks and send message */
+               if (flags & LKM_CANCEL)
+                       lock->cancel_pending = 1;
+               else
+                       lock->unlock_pending = 1;
+               spin_unlock(&lock->spinlock);
+               spin_unlock(&res->spinlock);
+               status = dlm_send_remote_unlock_request(dlm, res, lock, lksb,
+                                                       flags, owner);
+               spin_lock(&res->spinlock);
+               spin_lock(&lock->spinlock);
+               /* if the master told us the lock was already granted,
+                * let the ast handle all of these actions */
+               if (status == DLM_CANCELGRANT) {
+                       actions &= ~(DLM_UNLOCK_REMOVE_LOCK|
+                                    DLM_UNLOCK_REGRANT_LOCK|
+                                    DLM_UNLOCK_CLEAR_CONVERT_TYPE);
+               } else if (status == DLM_RECOVERING ||
+                          status == DLM_MIGRATING ||
+                          status == DLM_FORWARD ||
+                          status == DLM_NOLOCKMGR
+                          ) {
+                       /* must clear the actions because this unlock
+                        * is about to be retried.  cannot free or do
+                        * any list manipulation. */
+                       mlog(0, "%s:%.*s: clearing actions, %s\n",
+                            dlm->name, res->lockname.len,
+                            res->lockname.name,
+                            status==DLM_RECOVERING?"recovering":
+                            (status==DLM_MIGRATING?"migrating":
+                               (status == DLM_FORWARD ? "forward" :
+                                               "nolockmanager")));
+                       actions = 0;
+               }
+               if (flags & LKM_CANCEL)
+                       lock->cancel_pending = 0;
+               else
+                       lock->unlock_pending = 0;
+
+       }
+
+       /* get an extra ref on lock.  if we are just switching
+        * lists here, we dont want the lock to go away. */
+       dlm_lock_get(lock);
+
+       if (actions & DLM_UNLOCK_REMOVE_LOCK) {
+               list_del_init(&lock->list);
+               dlm_lock_put(lock);
+       }
+       if (actions & DLM_UNLOCK_REGRANT_LOCK) {
+               dlm_lock_get(lock);
+               list_add_tail(&lock->list, &res->granted);
+       }
+       if (actions & DLM_UNLOCK_CLEAR_CONVERT_TYPE) {
+               mlog(0, "clearing convert_type at %smaster node\n",
+                    master_node ? "" : "non-");
+               lock->ml.convert_type = LKM_IVMODE;
+       }
+
+       /* remove the extra ref on lock */
+       dlm_lock_put(lock);
+
+leave:
+       res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+       if (!dlm_lock_on_list(&res->converting, lock))
+               BUG_ON(lock->ml.convert_type != LKM_IVMODE);
+       else
+               BUG_ON(lock->ml.convert_type == LKM_IVMODE);
+       spin_unlock(&lock->spinlock);
+       spin_unlock(&res->spinlock);
+       wake_up(&res->wq);
+
+       /* let the caller's final dlm_lock_put handle the actual kfree */
+       if (actions & DLM_UNLOCK_FREE_LOCK) {
+               /* this should always be coupled with list removal */
+               BUG_ON(!(actions & DLM_UNLOCK_REMOVE_LOCK));
+               mlog(0, "lock %u:%llu should be gone now! refs=%d\n",
+                    dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+                    dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+                    atomic_read(&lock->lock_refs.refcount)-1);
+               dlm_lock_put(lock);
+       }
+       if (actions & DLM_UNLOCK_CALL_AST)
+               *call_ast = 1;
+
+       /* if cancel or unlock succeeded, lvb work is done */
+       if (status == DLM_NORMAL)
+               lksb->flags &= ~(DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB);
+
+       return status;
+}
+
+void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
+                              struct dlm_lock *lock)
+{
+       /* leave DLM_LKSB_PUT_LVB on the lksb so any final
+        * update of the lvb will be sent to the new master */
+       list_del_init(&lock->list);
+}
+
+void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
+                              struct dlm_lock *lock)
+{
+       list_move_tail(&lock->list, &res->granted);
+       lock->ml.convert_type = LKM_IVMODE;
+}
+
+
+static inline enum dlm_status dlmunlock_master(struct dlm_ctxt *dlm,
+                                         struct dlm_lock_resource *res,
+                                         struct dlm_lock *lock,
+                                         struct dlm_lockstatus *lksb,
+                                         int flags,
+                                         int *call_ast)
+{
+       return dlmunlock_common(dlm, res, lock, lksb, flags, call_ast, 1);
+}
+
+static inline enum dlm_status dlmunlock_remote(struct dlm_ctxt *dlm,
+                                         struct dlm_lock_resource *res,
+                                         struct dlm_lock *lock,
+                                         struct dlm_lockstatus *lksb,
+                                         int flags, int *call_ast)
+{
+       return dlmunlock_common(dlm, res, lock, lksb, flags, call_ast, 0);
+}
+
+/*
+ * locking:
+ *   caller needs:  none
+ *   taken:         none
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_NOLOCKMGR, status from network
+ */
+static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm,
+                                                struct dlm_lock_resource *res,
+                                                struct dlm_lock *lock,
+                                                struct dlm_lockstatus *lksb,
+                                                int flags,
+                                                u8 owner)
+{
+       struct dlm_unlock_lock unlock;
+       int tmpret;
+       enum dlm_status ret;
+       int status = 0;
+       struct kvec vec[2];
+       size_t veclen = 1;
+
+       mlog(0, "%.*s\n", res->lockname.len, res->lockname.name);
+
+       if (owner == dlm->node_num) {
+               /* ended up trying to contact ourself.  this means
+                * that the lockres had been remote but became local
+                * via a migration.  just retry it, now as local */
+               mlog(0, "%s:%.*s: this node became the master due to a "
+                    "migration, re-evaluate now\n", dlm->name,
+                    res->lockname.len, res->lockname.name);
+               return DLM_FORWARD;
+       }
+
+       memset(&unlock, 0, sizeof(unlock));
+       unlock.node_idx = dlm->node_num;
+       unlock.flags = cpu_to_be32(flags);
+       unlock.cookie = lock->ml.cookie;
+       unlock.namelen = res->lockname.len;
+       memcpy(unlock.name, res->lockname.name, unlock.namelen);
+
+       vec[0].iov_len = sizeof(struct dlm_unlock_lock);
+       vec[0].iov_base = &unlock;
+
+       if (flags & LKM_PUT_LVB) {
+               /* extra data to send if we are updating lvb */
+               vec[1].iov_len = DLM_LVB_LEN;
+               vec[1].iov_base = lock->lksb->lvb;
+               veclen++;
+       }
+
+       tmpret = o2net_send_message_vec(DLM_UNLOCK_LOCK_MSG, dlm->key,
+                                       vec, veclen, owner, &status);
+       if (tmpret >= 0) {
+               // successfully sent and received
+               if (status == DLM_FORWARD)
+                       mlog(0, "master was in-progress.  retry\n");
+               ret = status;
+       } else {
+               mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
+                    "node %u\n", tmpret, DLM_UNLOCK_LOCK_MSG, dlm->key, owner);
+               if (dlm_is_host_down(tmpret)) {
+                       /* NOTE: this seems strange, but it is what we want.
+                        * when the master goes down during a cancel or
+                        * unlock, the recovery code completes the operation
+                        * as if the master had not died, then passes the
+                        * updated state to the recovery master.  this thread
+                        * just needs to finish out the operation and call
+                        * the unlockast. */
+                       if (dlm_is_node_dead(dlm, owner))
+                               ret = DLM_NORMAL;
+                       else
+                               ret = DLM_NOLOCKMGR;
+               } else {
+                       /* something bad.  this will BUG in ocfs2 */
+                       ret = dlm_err_to_dlm_status(tmpret);
+               }
+       }
+
+       return ret;
+}
+
+/*
+ * locking:
+ *   caller needs:  none
+ *   taken:         takes and drops res->spinlock
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_BADARGS, DLM_IVLOCKID,
+ *          return value from dlmunlock_master
+ */
+int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
+                           void **ret_data)
+{
+       struct dlm_ctxt *dlm = data;
+       struct dlm_unlock_lock *unlock = (struct dlm_unlock_lock *)msg->buf;
+       struct dlm_lock_resource *res = NULL;
+       struct dlm_lock *lock = NULL;
+       enum dlm_status status = DLM_NORMAL;
+       int found = 0, i;
+       struct dlm_lockstatus *lksb = NULL;
+       int ignore;
+       u32 flags;
+       struct list_head *queue;
+
+       flags = be32_to_cpu(unlock->flags);
+
+       if (flags & LKM_GET_LVB) {
+               mlog(ML_ERROR, "bad args!  GET_LVB specified on unlock!\n");
+               return DLM_BADARGS;
+       }
+
+       if ((flags & (LKM_PUT_LVB|LKM_CANCEL)) == (LKM_PUT_LVB|LKM_CANCEL)) {
+               mlog(ML_ERROR, "bad args!  cannot modify lvb on a CANCEL "
+                    "request!\n");
+               return DLM_BADARGS;
+       }
+
+       if (unlock->namelen > DLM_LOCKID_NAME_MAX) {
+               mlog(ML_ERROR, "Invalid name length in unlock handler!\n");
+               return DLM_IVBUFLEN;
+       }
+
+       if (!dlm_grab(dlm))
+               return DLM_REJECTED;
+
+       mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
+                       "Domain %s not fully joined!\n", dlm->name);
+
+       mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" : "none");
+
+       res = dlm_lookup_lockres(dlm, unlock->name, unlock->namelen);
+       if (!res) {
+               /* We assume here that a no lock resource simply means
+                * it was migrated away and destroyed before the other
+                * node could detect it. */
+               mlog(0, "returning DLM_FORWARD -- res no longer exists\n");
+               status = DLM_FORWARD;
+               goto not_found;
+       }
+
+       queue=&res->granted;
+       found = 0;
+       spin_lock(&res->spinlock);
+       if (res->state & DLM_LOCK_RES_RECOVERING) {
+               spin_unlock(&res->spinlock);
+               mlog(0, "returning DLM_RECOVERING\n");
+               status = DLM_RECOVERING;
+               goto leave;
+       }
+
+       if (res->state & DLM_LOCK_RES_MIGRATING) {
+               spin_unlock(&res->spinlock);
+               mlog(0, "returning DLM_MIGRATING\n");
+               status = DLM_MIGRATING;
+               goto leave;
+       }
+
+       if (res->owner != dlm->node_num) {
+               spin_unlock(&res->spinlock);
+               mlog(0, "returning DLM_FORWARD -- not master\n");
+               status = DLM_FORWARD;
+               goto leave;
+       }
+
+       for (i=0; i<3; i++) {
+               list_for_each_entry(lock, queue, list) {
+                       if (lock->ml.cookie == unlock->cookie &&
+                           lock->ml.node == unlock->node_idx) {
+                               dlm_lock_get(lock);
+                               found = 1;
+                               break;
+                       }
+               }
+               if (found)
+                       break;
+               /* scan granted -> converting -> blocked queues */
+               queue++;
+       }
+       spin_unlock(&res->spinlock);
+       if (!found) {
+               status = DLM_IVLOCKID;
+               goto not_found;
+       }
+
+       /* lock was found on queue */
+       lksb = lock->lksb;
+       if (flags & (LKM_VALBLK|LKM_PUT_LVB) &&
+           lock->ml.type != LKM_EXMODE)
+               flags &= ~(LKM_VALBLK|LKM_PUT_LVB);
+
+       /* unlockast only called on originating node */
+       if (flags & LKM_PUT_LVB) {
+               lksb->flags |= DLM_LKSB_PUT_LVB;
+               memcpy(&lksb->lvb[0], &unlock->lvb[0], DLM_LVB_LEN);
+       }
+
+       /* if this is in-progress, propagate the DLM_FORWARD
+        * all the way back out */
+       status = dlmunlock_master(dlm, res, lock, lksb, flags, &ignore);
+       if (status == DLM_FORWARD)
+               mlog(0, "lockres is in progress\n");
+
+       if (flags & LKM_PUT_LVB)
+               lksb->flags &= ~DLM_LKSB_PUT_LVB;
+
+       dlm_lockres_calc_usage(dlm, res);
+       dlm_kick_thread(dlm, res);
+
+not_found:
+       if (!found)
+               mlog(ML_ERROR, "failed to find lock to unlock! "
+                              "cookie=%u:%llu\n",
+                    dlm_get_lock_cookie_node(be64_to_cpu(unlock->cookie)),
+                    dlm_get_lock_cookie_seq(be64_to_cpu(unlock->cookie)));
+       else
+               dlm_lock_put(lock);
+
+leave:
+       if (res)
+               dlm_lockres_put(res);
+
+       dlm_put(dlm);
+
+       return status;
+}
+
+
+static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm,
+                                             struct dlm_lock_resource *res,
+                                             struct dlm_lock *lock,
+                                             struct dlm_lockstatus *lksb,
+                                             int *actions)
+{
+       enum dlm_status status;
+
+       if (dlm_lock_on_list(&res->blocked, lock)) {
+               /* cancel this outright */
+               status = DLM_NORMAL;
+               *actions = (DLM_UNLOCK_CALL_AST |
+                           DLM_UNLOCK_REMOVE_LOCK);
+       } else if (dlm_lock_on_list(&res->converting, lock)) {
+               /* cancel the request, put back on granted */
+               status = DLM_NORMAL;
+               *actions = (DLM_UNLOCK_CALL_AST |
+                           DLM_UNLOCK_REMOVE_LOCK |
+                           DLM_UNLOCK_REGRANT_LOCK |
+                           DLM_UNLOCK_CLEAR_CONVERT_TYPE);
+       } else if (dlm_lock_on_list(&res->granted, lock)) {
+               /* too late, already granted. */
+               status = DLM_CANCELGRANT;
+               *actions = DLM_UNLOCK_CALL_AST;
+       } else {
+               mlog(ML_ERROR, "lock to cancel is not on any list!\n");
+               status = DLM_IVLOCKID;
+               *actions = 0;
+       }
+       return status;
+}
+
+static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm,
+                                             struct dlm_lock_resource *res,
+                                             struct dlm_lock *lock,
+                                             struct dlm_lockstatus *lksb,
+                                             int *actions)
+{
+       enum dlm_status status;
+
+       /* unlock request */
+       if (!dlm_lock_on_list(&res->granted, lock)) {
+               status = DLM_DENIED;
+               dlm_error(status);
+               *actions = 0;
+       } else {
+               /* unlock granted lock */
+               status = DLM_NORMAL;
+               *actions = (DLM_UNLOCK_FREE_LOCK |
+                           DLM_UNLOCK_CALL_AST |
+                           DLM_UNLOCK_REMOVE_LOCK);
+       }
+       return status;
+}
+
+/* there seems to be no point in doing this async
+ * since (even for the remote case) there is really
+ * no work to queue up... so just do it and fire the
+ * unlockast by hand when done... */
+enum dlm_status dlmunlock(struct dlm_ctxt *dlm, struct dlm_lockstatus *lksb,
+                         int flags, dlm_astunlockfunc_t *unlockast, void *data)
+{
+       enum dlm_status status;
+       struct dlm_lock_resource *res;
+       struct dlm_lock *lock = NULL;
+       int call_ast, is_master;
+
+       if (!lksb) {
+               dlm_error(DLM_BADARGS);
+               return DLM_BADARGS;
+       }
+
+       if (flags & ~(LKM_CANCEL | LKM_VALBLK | LKM_INVVALBLK)) {
+               dlm_error(DLM_BADPARAM);
+               return DLM_BADPARAM;
+       }
+
+       if ((flags & (LKM_VALBLK | LKM_CANCEL)) == (LKM_VALBLK | LKM_CANCEL)) {
+               mlog(0, "VALBLK given with CANCEL: ignoring VALBLK\n");
+               flags &= ~LKM_VALBLK;
+       }
+
+       if (!lksb->lockid || !lksb->lockid->lockres) {
+               dlm_error(DLM_BADPARAM);
+               return DLM_BADPARAM;
+       }
+
+       lock = lksb->lockid;
+       BUG_ON(!lock);
+       dlm_lock_get(lock);
+
+       res = lock->lockres;
+       BUG_ON(!res);
+       dlm_lockres_get(res);
+retry:
+       call_ast = 0;
+       /* need to retry up here because owner may have changed */
+       mlog(0, "lock=%p res=%p\n", lock, res);
+
+       spin_lock(&res->spinlock);
+       is_master = (res->owner == dlm->node_num);
+       if (flags & LKM_VALBLK && lock->ml.type != LKM_EXMODE)
+               flags &= ~LKM_VALBLK;
+       spin_unlock(&res->spinlock);
+
+       if (is_master) {
+               status = dlmunlock_master(dlm, res, lock, lksb, flags,
+                                         &call_ast);
+               mlog(0, "done calling dlmunlock_master: returned %d, "
+                    "call_ast is %d\n", status, call_ast);
+       } else {
+               status = dlmunlock_remote(dlm, res, lock, lksb, flags,
+                                         &call_ast);
+               mlog(0, "done calling dlmunlock_remote: returned %d, "
+                    "call_ast is %d\n", status, call_ast);
+       }
+
+       if (status == DLM_RECOVERING ||
+           status == DLM_MIGRATING ||
+           status == DLM_FORWARD ||
+           status == DLM_NOLOCKMGR) {
+
+               /* We want to go away for a tiny bit to allow recovery
+                * / migration to complete on this resource. I don't
+                * know of any wait queue we could sleep on as this
+                * may be happening on another node. Perhaps the
+                * proper solution is to queue up requests on the
+                * other end? */
+
+               /* do we want to yield(); ?? */
+               msleep(50);
+
+               mlog(0, "retrying unlock due to pending recovery/"
+                    "migration/in-progress/reconnect\n");
+               goto retry;
+       }
+
+       if (call_ast) {
+               mlog(0, "calling unlockast(%p, %d)\n", data, status);
+               if (is_master) {
+                       /* it is possible that there is one last bast
+                        * pending.  make sure it is flushed, then
+                        * call the unlockast.
+                        * not an issue if this is a mastered remotely,
+                        * since this lock has been removed from the
+                        * lockres queues and cannot be found. */
+                       dlm_kick_thread(dlm, NULL);
+                       wait_event(dlm->ast_wq,
+                                  dlm_lock_basts_flushed(dlm, lock));
+               }
+               (*unlockast)(data, status);
+       }
+
+       if (status == DLM_CANCELGRANT)
+               status = DLM_NORMAL;
+
+       if (status == DLM_NORMAL) {
+               mlog(0, "kicking the thread\n");
+               dlm_kick_thread(dlm, res);
+       } else
+               dlm_error(status);
+
+       dlm_lockres_calc_usage(dlm, res);
+       dlm_lockres_put(res);
+       dlm_lock_put(lock);
+
+       mlog(0, "returning status=%d!\n", status);
+       return status;
+}
+EXPORT_SYMBOL_GPL(dlmunlock);
+