[linux-block.git] / fs / ocfs2 / dlm / dlmunlock.c

/* -*- 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) {
			/* 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":
			      "forward"));
			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_entry("%.*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_errno(tmpret);
		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. */
			ret = DLM_NORMAL;
		} 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 list_head *iter;
	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(iter, queue) {
			lock = list_entry(iter, struct dlm_lock, 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;

	mlog_entry_void();

	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) {
		/* 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\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);
Commit	Line	Data
6714d8e8 KH	1	/* -- mode: c; c-basic-offset: 8; --
	2	* vim: noexpandtab sw=8 ts=8 sts=0:
	3	*
	4	* dlmunlock.c
	5	*
	6	* underlying calls for unlocking locks
	7	*
	8	* Copyright (C) 2004 Oracle. All rights reserved.
	9	*
	10	* This program is free software; you can redistribute it and/or
	11	* modify it under the terms of the GNU General Public
	12	* License as published by the Free Software Foundation; either
	13	* version 2 of the License, or (at your option) any later version.
	14	*
	15	* This program is distributed in the hope that it will be useful,
	16	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	17	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	18	* General Public License for more details.
	19	*
	20	* You should have received a copy of the GNU General Public
	21	* License along with this program; if not, write to the
	22	* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
	23	* Boston, MA 021110-1307, USA.
	24	*
	25	*/
	26
	27
	28	#include <linux/module.h>
	29	#include <linux/fs.h>
	30	#include <linux/types.h>
6714d8e8	31	#include <linux/highmem.h>
6714d8e8 KH	32	#include <linux/init.h>
	33	#include <linux/sysctl.h>
	34	#include <linux/random.h>
	35	#include <linux/blkdev.h>
	36	#include <linux/socket.h>
	37	#include <linux/inet.h>
	38	#include <linux/spinlock.h>
	39	#include <linux/delay.h>
	40
	41	#include "cluster/heartbeat.h"
	42	#include "cluster/nodemanager.h"
	43	#include "cluster/tcp.h"
	44
	45	#include "dlmapi.h"
	46	#include "dlmcommon.h"
	47
	48	#define MLOG_MASK_PREFIX ML_DLM
	49	#include "cluster/masklog.h"
	50
	51	#define DLM_UNLOCK_FREE_LOCK 0x00000001
	52	#define DLM_UNLOCK_CALL_AST 0x00000002
	53	#define DLM_UNLOCK_REMOVE_LOCK 0x00000004
	54	#define DLM_UNLOCK_REGRANT_LOCK 0x00000008
	55	#define DLM_UNLOCK_CLEAR_CONVERT_TYPE 0x00000010
	56
	57
	58	static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm,
	59	struct dlm_lock_resource *res,
	60	struct dlm_lock *lock,
	61	struct dlm_lockstatus *lksb,
	62	int *actions);
	63	static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm,
	64	struct dlm_lock_resource *res,
	65	struct dlm_lock *lock,
	66	struct dlm_lockstatus *lksb,
	67	int *actions);
	68
	69	static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm,
	70	struct dlm_lock_resource *res,
	71	struct dlm_lock *lock,
	72	struct dlm_lockstatus *lksb,
	73	int flags,
	74	u8 owner);
	75
	76
	77	/*
	78	* according to the spec:
	79	* http://opendlm.sourceforge.net/cvsmirror/opendlm/docs/dlmbook_final.pdf
	80	*
	81	* flags & LKM_CANCEL != 0: must be converting or blocked
	82	* flags & LKM_CANCEL == 0: must be granted
	83	*
	84	* So to unlock a converting lock, you must first cancel the
	85	* convert (passing LKM_CANCEL in flags), then call the unlock
	86	* again (with no LKM_CANCEL in flags).
	87	*/
	88
	89
	90	/*
	91	* locking:
	92	* caller needs: none
	93	* taken: res->spinlock and lock->spinlock taken and dropped
	94	* held on exit: none
	95	* returns: DLM_NORMAL, DLM_NOLOCKMGR, status from network
96	* all callers should have taken an extra ref on lock coming in
97	*/
98	static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm,
99	struct dlm_lock_resource *res,
100	struct dlm_lock *lock,
101	struct dlm_lockstatus *lksb,
102	int flags, int *call_ast,
103	int master_node)
104	{
105	enum dlm_status status;
106	int actions = 0;
107	int in_use;
108	u8 owner;
109
110	mlog(0, "master_node = %d, valblk = %d\n", master_node,
111	flags & LKM_VALBLK);
112
113	if (master_node)
114	BUG_ON(res->owner != dlm->node_num);
115	else
116	BUG_ON(res->owner == dlm->node_num);
117
dabc47de	118	spin_lock(&dlm->ast_lock);
6714d8e8 KH	119	/* We want to be sure that we're not freeing a lock
	120	* that still has AST's pending... */
	121	in_use = !list_empty(&lock->ast_list);
dabc47de	122	spin_unlock(&dlm->ast_lock);
c795b33b	123	if (in_use && !(flags & LKM_CANCEL)) {
6714d8e8 KH	124	mlog(ML_ERROR, "lockres %.*s: Someone is calling dlmunlock "
	125	"while waiting for an ast!", res->lockname.len,
	126	res->lockname.name);
	127	return DLM_BADPARAM;
	128	}
	129
	130	spin_lock(&res->spinlock);
	131	if (res->state & DLM_LOCK_RES_IN_PROGRESS) {
c795b33b	132	if (master_node && !(flags & LKM_CANCEL)) {
6714d8e8 KH	133	mlog(ML_ERROR, "lockres in progress!\n");
	134	spin_unlock(&res->spinlock);
	135	return DLM_FORWARD;
	136	}
	137	/* ok for this to sleep if not in a network handler */
	138	__dlm_wait_on_lockres(res);
	139	res->state \|= DLM_LOCK_RES_IN_PROGRESS;
	140	}
	141	spin_lock(&lock->spinlock);
	142
	143	if (res->state & DLM_LOCK_RES_RECOVERING) {
	144	status = DLM_RECOVERING;
	145	goto leave;
	146	}
	147
71ac1062 KH	148	if (res->state & DLM_LOCK_RES_MIGRATING) {
	149	status = DLM_MIGRATING;
	150	goto leave;
	151	}
6714d8e8 KH	152
	153	/* see above for what the spec says about
	154	* LKM_CANCEL and the lock queue state */
	155	if (flags & LKM_CANCEL)
	156	status = dlm_get_cancel_actions(dlm, res, lock, lksb, &actions);
	157	else
	158	status = dlm_get_unlock_actions(dlm, res, lock, lksb, &actions);
	159
34e3d180	160	if (status != DLM_NORMAL && (status != DLM_CANCELGRANT \|\| !master_node))
6714d8e8 KH	161	goto leave;
	162
	163	/* By now this has been masked out of cancel requests. */
	164	if (flags & LKM_VALBLK) {
	165	/* make the final update to the lvb */
	166	if (master_node)
	167	memcpy(res->lvb, lksb->lvb, DLM_LVB_LEN);
	168	else
	169	flags \|= LKM_PUT_LVB; /* let the send function
	170	* handle it. */
	171	}
	172
	173	if (!master_node) {
	174	owner = res->owner;
	175	/* drop locks and send message */
	176	if (flags & LKM_CANCEL)
	177	lock->cancel_pending = 1;
	178	else
	179	lock->unlock_pending = 1;
	180	spin_unlock(&lock->spinlock);
	181	spin_unlock(&res->spinlock);
	182	status = dlm_send_remote_unlock_request(dlm, res, lock, lksb,
	183	flags, owner);
	184	spin_lock(&res->spinlock);
	185	spin_lock(&lock->spinlock);
	186	/* if the master told us the lock was already granted,
	187	* let the ast handle all of these actions */
a23eac99	188	if (status == DLM_CANCELGRANT) {
6714d8e8 KH	189	actions &= ~(DLM_UNLOCK_REMOVE_LOCK\|
	190	DLM_UNLOCK_REGRANT_LOCK\|
	191	DLM_UNLOCK_CLEAR_CONVERT_TYPE);
2bd63216 SM	192	} else if (status == DLM_RECOVERING \|\|
2bd63216 SM	193	status == DLM_MIGRATING \|\|
e2faea4c KH	194	status == DLM_FORWARD) {
	195	/* must clear the actions because this unlock
	196	* is about to be retried. cannot free or do
	197	* any list manipulation. */
	198	mlog(0, "%s:%.*s: clearing actions, %s\n",
	199	dlm->name, res->lockname.len,
	200	res->lockname.name,
	201	status==DLM_RECOVERING?"recovering":
	202	(status==DLM_MIGRATING?"migrating":
	203	"forward"));
	204	actions = 0;
6714d8e8 KH	205	}
	206	if (flags & LKM_CANCEL)
	207	lock->cancel_pending = 0;
	208	else
	209	lock->unlock_pending = 0;
	210
	211	}
	212
	213	/* get an extra ref on lock. if we are just switching
	214	* lists here, we dont want the lock to go away. */
	215	dlm_lock_get(lock);
	216
	217	if (actions & DLM_UNLOCK_REMOVE_LOCK) {
	218	list_del_init(&lock->list);
	219	dlm_lock_put(lock);
	220	}
	221	if (actions & DLM_UNLOCK_REGRANT_LOCK) {
	222	dlm_lock_get(lock);
	223	list_add_tail(&lock->list, &res->granted);
	224	}
	225	if (actions & DLM_UNLOCK_CLEAR_CONVERT_TYPE) {
	226	mlog(0, "clearing convert_type at %smaster node\n",
	227	master_node ? "" : "non-");
	228	lock->ml.convert_type = LKM_IVMODE;
	229	}
	230
	231	/* remove the extra ref on lock */
	232	dlm_lock_put(lock);
	233
	234	leave:
	235	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
	236	if (!dlm_lock_on_list(&res->converting, lock))
	237	BUG_ON(lock->ml.convert_type != LKM_IVMODE);
	238	else
	239	BUG_ON(lock->ml.convert_type == LKM_IVMODE);
	240	spin_unlock(&lock->spinlock);
	241	spin_unlock(&res->spinlock);
	242	wake_up(&res->wq);
	243
	244	/* let the caller's final dlm_lock_put handle the actual kfree */
	245	if (actions & DLM_UNLOCK_FREE_LOCK) {
	246	/* this should always be coupled with list removal */
	247	BUG_ON(!(actions & DLM_UNLOCK_REMOVE_LOCK));
29004858	248	mlog(0, "lock %u:%llu should be gone now! refs=%d\n",
74aa2585 KH	249	dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
74aa2585 KH	250	dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
29004858	251	atomic_read(&lock->lock_refs.refcount)-1);
6714d8e8 KH	252	dlm_lock_put(lock);
	253	}
	254	if (actions & DLM_UNLOCK_CALL_AST)
	255	*call_ast = 1;
	256
	257	/* if cancel or unlock succeeded, lvb work is done */
	258	if (status == DLM_NORMAL)
	259	lksb->flags &= ~(DLM_LKSB_PUT_LVB\|DLM_LKSB_GET_LVB);
	260
	261	return status;
	262	}
	263
	264	void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
	265	struct dlm_lock *lock)
	266	{
	267	/* leave DLM_LKSB_PUT_LVB on the lksb so any final
	268	* update of the lvb will be sent to the new master */
	269	list_del_init(&lock->list);
	270	}
	271
	272	void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
	273	struct dlm_lock *lock)
	274	{
f116629d	275	list_move_tail(&lock->list, &res->granted);
6714d8e8 KH	276	lock->ml.convert_type = LKM_IVMODE;
	277	}
	278
	279
	280	static inline enum dlm_status dlmunlock_master(struct dlm_ctxt *dlm,
	281	struct dlm_lock_resource *res,
	282	struct dlm_lock *lock,
	283	struct dlm_lockstatus *lksb,
	284	int flags,
	285	int *call_ast)
	286	{
	287	return dlmunlock_common(dlm, res, lock, lksb, flags, call_ast, 1);
	288	}
	289
	290	static inline enum dlm_status dlmunlock_remote(struct dlm_ctxt *dlm,
	291	struct dlm_lock_resource *res,
	292	struct dlm_lock *lock,
	293	struct dlm_lockstatus *lksb,
	294	int flags, int *call_ast)
	295	{
	296	return dlmunlock_common(dlm, res, lock, lksb, flags, call_ast, 0);
	297	}
	298
	299	/*
	300	* locking:
	301	* caller needs: none
	302	* taken: none
	303	* held on exit: none
	304	* returns: DLM_NORMAL, DLM_NOLOCKMGR, status from network
	305	*/
	306	static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm,
	307	struct dlm_lock_resource *res,
	308	struct dlm_lock *lock,
	309	struct dlm_lockstatus *lksb,
	310	int flags,
	311	u8 owner)
	312	{
	313	struct dlm_unlock_lock unlock;
	314	int tmpret;
	315	enum dlm_status ret;
	316	int status = 0;
	317	struct kvec vec[2];
	318	size_t veclen = 1;
	319
	320	mlog_entry("%.*s\n", res->lockname.len, res->lockname.name);
0032abd6	321
2580a580 KH	322	if (owner == dlm->node_num) {
	323	/* ended up trying to contact ourself. this means
	324	* that the lockres had been remote but became local
	325	* via a migration. just retry it, now as local */
	326	mlog(0, "%s:%.*s: this node became the master due to a "
	327	"migration, re-evaluate now\n", dlm->name,
	328	res->lockname.len, res->lockname.name);
	329	return DLM_FORWARD;
	330	}
6714d8e8 KH	331
	332	memset(&unlock, 0, sizeof(unlock));
	333	unlock.node_idx = dlm->node_num;
	334	unlock.flags = cpu_to_be32(flags);
	335	unlock.cookie = lock->ml.cookie;
	336	unlock.namelen = res->lockname.len;
	337	memcpy(unlock.name, res->lockname.name, unlock.namelen);
	338
	339	vec[0].iov_len = sizeof(struct dlm_unlock_lock);
	340	vec[0].iov_base = &unlock;
	341
	342	if (flags & LKM_PUT_LVB) {
	343	/* extra data to send if we are updating lvb */
	344	vec[1].iov_len = DLM_LVB_LEN;
	345	vec[1].iov_base = lock->lksb->lvb;
	346	veclen++;
	347	}
	348
	349	tmpret = o2net_send_message_vec(DLM_UNLOCK_LOCK_MSG, dlm->key,
	350	vec, veclen, owner, &status);
	351	if (tmpret >= 0) {
	352	// successfully sent and received
a23eac99	353	if (status == DLM_FORWARD)
6714d8e8	354	mlog(0, "master was in-progress. retry\n");
a23eac99	355	ret = status;
6714d8e8 KH	356	} else {
	357	mlog_errno(tmpret);
	358	if (dlm_is_host_down(tmpret)) {
	359	/* NOTE: this seems strange, but it is what we want.
	360	* when the master goes down during a cancel or
	361	* unlock, the recovery code completes the operation
	362	* as if the master had not died, then passes the
	363	* updated state to the recovery master. this thread
	364	* just needs to finish out the operation and call
	365	* the unlockast. */
	366	ret = DLM_NORMAL;
	367	} else {
	368	/* something bad. this will BUG in ocfs2 */
	369	ret = dlm_err_to_dlm_status(tmpret);
	370	}
6714d8e8 KH	371	}
	372
	373	return ret;
	374	}
	375
	376	/*
	377	* locking:
	378	* caller needs: none
	379	* taken: takes and drops res->spinlock
	380	* held on exit: none
	381	* returns: DLM_NORMAL, DLM_BADARGS, DLM_IVLOCKID,
	382	* return value from dlmunlock_master
	383	*/
d74c9803 KH	384	int dlm_unlock_lock_handler(struct o2net_msg msg, u32 len, void data,
d74c9803 KH	385	void **ret_data)
6714d8e8 KH	386	{
	387	struct dlm_ctxt *dlm = data;
	388	struct dlm_unlock_lock unlock = (struct dlm_unlock_lock )msg->buf;
	389	struct dlm_lock_resource *res = NULL;
	390	struct list_head *iter;
	391	struct dlm_lock *lock = NULL;
	392	enum dlm_status status = DLM_NORMAL;
	393	int found = 0, i;
	394	struct dlm_lockstatus *lksb = NULL;
	395	int ignore;
	396	u32 flags;
	397	struct list_head *queue;
	398
	399	flags = be32_to_cpu(unlock->flags);
	400
	401	if (flags & LKM_GET_LVB) {
	402	mlog(ML_ERROR, "bad args! GET_LVB specified on unlock!\n");
	403	return DLM_BADARGS;
	404	}
	405
	406	if ((flags & (LKM_PUT_LVB\|LKM_CANCEL)) == (LKM_PUT_LVB\|LKM_CANCEL)) {
	407	mlog(ML_ERROR, "bad args! cannot modify lvb on a CANCEL "
	408	"request!\n");
	409	return DLM_BADARGS;
	410	}
	411
	412	if (unlock->namelen > DLM_LOCKID_NAME_MAX) {
	413	mlog(ML_ERROR, "Invalid name length in unlock handler!\n");
	414	return DLM_IVBUFLEN;
	415	}
	416
	417	if (!dlm_grab(dlm))
	418	return DLM_REJECTED;
	419
	420	mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
	421	"Domain %s not fully joined!\n", dlm->name);
	422
	423	mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" : "none");
	424
	425	res = dlm_lookup_lockres(dlm, unlock->name, unlock->namelen);
	426	if (!res) {
	427	/* We assume here that a no lock resource simply means
	428	* it was migrated away and destroyed before the other
	429	* node could detect it. */
	430	mlog(0, "returning DLM_FORWARD -- res no longer exists\n");
	431	status = DLM_FORWARD;
	432	goto not_found;
	433	}
	434
	435	queue=&res->granted;
	436	found = 0;
	437	spin_lock(&res->spinlock);
	438	if (res->state & DLM_LOCK_RES_RECOVERING) {
	439	spin_unlock(&res->spinlock);
	440	mlog(0, "returning DLM_RECOVERING\n");
	441	status = DLM_RECOVERING;
	442	goto leave;
	443	}
	444
	445	if (res->state & DLM_LOCK_RES_MIGRATING) {
	446	spin_unlock(&res->spinlock);
	447	mlog(0, "returning DLM_MIGRATING\n");
	448	status = DLM_MIGRATING;
	449	goto leave;
450	}
451
452	if (res->owner != dlm->node_num) {
453	spin_unlock(&res->spinlock);
454	mlog(0, "returning DLM_FORWARD -- not master\n");
455	status = DLM_FORWARD;
456	goto leave;
457	}
458
459	for (i=0; i<3; i++) {
460	list_for_each(iter, queue) {
461	lock = list_entry(iter, struct dlm_lock, list);
462	if (lock->ml.cookie == unlock->cookie &&
463	lock->ml.node == unlock->node_idx) {
464	dlm_lock_get(lock);
465	found = 1;
466	break;
467	}
468	}
469	if (found)
470	break;
471	/* scan granted -> converting -> blocked queues */
472	queue++;
473	}
474	spin_unlock(&res->spinlock);
475	if (!found) {
476	status = DLM_IVLOCKID;
477	goto not_found;
478	}
479
480	/* lock was found on queue */
481	lksb = lock->lksb;
4b1af774 KH	482	if (flags & (LKM_VALBLK\|LKM_PUT_LVB) &&
	483	lock->ml.type != LKM_EXMODE)
	484	flags &= ~(LKM_VALBLK\|LKM_PUT_LVB);
	485
6714d8e8 KH	486	/* unlockast only called on originating node */
	487	if (flags & LKM_PUT_LVB) {
	488	lksb->flags \|= DLM_LKSB_PUT_LVB;
	489	memcpy(&lksb->lvb[0], &unlock->lvb[0], DLM_LVB_LEN);
	490	}
	491
	492	/* if this is in-progress, propagate the DLM_FORWARD
	493	* all the way back out */
	494	status = dlmunlock_master(dlm, res, lock, lksb, flags, &ignore);
	495	if (status == DLM_FORWARD)
	496	mlog(0, "lockres is in progress\n");
	497
	498	if (flags & LKM_PUT_LVB)
	499	lksb->flags &= ~DLM_LKSB_PUT_LVB;
	500
	501	dlm_lockres_calc_usage(dlm, res);
	502	dlm_kick_thread(dlm, res);
	503
	504	not_found:
	505	if (!found)
	506	mlog(ML_ERROR, "failed to find lock to unlock! "
29004858	507	"cookie=%u:%llu\n",
74aa2585 KH	508	dlm_get_lock_cookie_node(be64_to_cpu(unlock->cookie)),
74aa2585 KH	509	dlm_get_lock_cookie_seq(be64_to_cpu(unlock->cookie)));
a23eac99	510	else
6714d8e8	511	dlm_lock_put(lock);
6714d8e8 KH	512
	513	leave:
	514	if (res)
	515	dlm_lockres_put(res);
	516
	517	dlm_put(dlm);
	518
	519	return status;
	520	}
	521
	522
	523	static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm,
	524	struct dlm_lock_resource *res,
	525	struct dlm_lock *lock,
	526	struct dlm_lockstatus *lksb,
	527	int *actions)
	528	{
	529	enum dlm_status status;
	530
	531	if (dlm_lock_on_list(&res->blocked, lock)) {
	532	/* cancel this outright */
6714d8e8 KH	533	status = DLM_NORMAL;
	534	*actions = (DLM_UNLOCK_CALL_AST \|
	535	DLM_UNLOCK_REMOVE_LOCK);
	536	} else if (dlm_lock_on_list(&res->converting, lock)) {
	537	/* cancel the request, put back on granted */
6714d8e8 KH	538	status = DLM_NORMAL;
	539	*actions = (DLM_UNLOCK_CALL_AST \|
	540	DLM_UNLOCK_REMOVE_LOCK \|
	541	DLM_UNLOCK_REGRANT_LOCK \|
	542	DLM_UNLOCK_CLEAR_CONVERT_TYPE);
	543	} else if (dlm_lock_on_list(&res->granted, lock)) {
a23eac99 KH	544	/* too late, already granted. */
a23eac99 KH	545	status = DLM_CANCELGRANT;
6714d8e8 KH	546	*actions = DLM_UNLOCK_CALL_AST;
	547	} else {
	548	mlog(ML_ERROR, "lock to cancel is not on any list!\n");
6714d8e8 KH	549	status = DLM_IVLOCKID;
	550	*actions = 0;
	551	}
	552	return status;
	553	}
	554
	555	static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm,
	556	struct dlm_lock_resource *res,
	557	struct dlm_lock *lock,
	558	struct dlm_lockstatus *lksb,
	559	int *actions)
	560	{
	561	enum dlm_status status;
	562
	563	/* unlock request */
	564	if (!dlm_lock_on_list(&res->granted, lock)) {
6714d8e8 KH	565	status = DLM_DENIED;
	566	dlm_error(status);
	567	*actions = 0;
	568	} else {
	569	/* unlock granted lock */
6714d8e8 KH	570	status = DLM_NORMAL;
	571	*actions = (DLM_UNLOCK_FREE_LOCK \|
	572	DLM_UNLOCK_CALL_AST \|
	573	DLM_UNLOCK_REMOVE_LOCK);
	574	}
	575	return status;
	576	}
	577
	578	/* there seems to be no point in doing this async
	579	* since (even for the remote case) there is really
	580	* no work to queue up... so just do it and fire the
	581	* unlockast by hand when done... */
	582	enum dlm_status dlmunlock(struct dlm_ctxt dlm, struct dlm_lockstatus lksb,
	583	int flags, dlm_astunlockfunc_t unlockast, void data)
	584	{
	585	enum dlm_status status;
	586	struct dlm_lock_resource *res;
	587	struct dlm_lock *lock = NULL;
	588	int call_ast, is_master;
	589
	590	mlog_entry_void();
	591
	592	if (!lksb) {
	593	dlm_error(DLM_BADARGS);
	594	return DLM_BADARGS;
	595	}
	596
	597	if (flags & ~(LKM_CANCEL \| LKM_VALBLK \| LKM_INVVALBLK)) {
	598	dlm_error(DLM_BADPARAM);
	599	return DLM_BADPARAM;
	600	}
	601
	602	if ((flags & (LKM_VALBLK \| LKM_CANCEL)) == (LKM_VALBLK \| LKM_CANCEL)) {
	603	mlog(0, "VALBLK given with CANCEL: ignoring VALBLK\n");
	604	flags &= ~LKM_VALBLK;
	605	}
	606
	607	if (!lksb->lockid \|\| !lksb->lockid->lockres) {
	608	dlm_error(DLM_BADPARAM);
	609	return DLM_BADPARAM;
	610	}
	611
	612	lock = lksb->lockid;
	613	BUG_ON(!lock);
	614	dlm_lock_get(lock);
	615
	616	res = lock->lockres;
	617	BUG_ON(!res);
	618	dlm_lockres_get(res);
	619	retry:
	620	call_ast = 0;
	621	/* need to retry up here because owner may have changed */
	622	mlog(0, "lock=%p res=%p\n", lock, res);
	623
	624	spin_lock(&res->spinlock);
	625	is_master = (res->owner == dlm->node_num);
4b1af774 KH	626	if (flags & LKM_VALBLK && lock->ml.type != LKM_EXMODE)
4b1af774 KH	627	flags &= ~LKM_VALBLK;
6714d8e8 KH	628	spin_unlock(&res->spinlock);
	629
	630	if (is_master) {
	631	status = dlmunlock_master(dlm, res, lock, lksb, flags,
	632	&call_ast);
	633	mlog(0, "done calling dlmunlock_master: returned %d, "
	634	"call_ast is %d\n", status, call_ast);
	635	} else {
	636	status = dlmunlock_remote(dlm, res, lock, lksb, flags,
	637	&call_ast);
	638	mlog(0, "done calling dlmunlock_remote: returned %d, "
	639	"call_ast is %d\n", status, call_ast);
	640	}
	641
	642	if (status == DLM_RECOVERING \|\|
	643	status == DLM_MIGRATING \|\|
	644	status == DLM_FORWARD) {
	645	/* We want to go away for a tiny bit to allow recovery
	646	* / migration to complete on this resource. I don't
	647	* know of any wait queue we could sleep on as this
	648	* may be happening on another node. Perhaps the
	649	* proper solution is to queue up requests on the
	650	* other end? */
	651
	652	/* do we want to yield(); ?? */
	653	msleep(50);
	654
	655	mlog(0, "retrying unlock due to pending recovery/"
	656	"migration/in-progress\n");
	657	goto retry;
	658	}
	659
	660	if (call_ast) {
a23eac99	661	mlog(0, "calling unlockast(%p, %d)\n", data, status);
6714d8e8	662	if (is_master) {
2bd63216	663	/* it is possible that there is one last bast
6714d8e8 KH	664	* pending. make sure it is flushed, then
	665	* call the unlockast.
	666	* not an issue if this is a mastered remotely,
	667	* since this lock has been removed from the
	668	* lockres queues and cannot be found. */
	669	dlm_kick_thread(dlm, NULL);
2bd63216	670	wait_event(dlm->ast_wq,
6714d8e8 KH	671	dlm_lock_basts_flushed(dlm, lock));
6714d8e8 KH	672	}
a23eac99	673	(*unlockast)(data, status);
6714d8e8 KH	674	}
6714d8e8 KH	675
a23eac99 KH	676	if (status == DLM_CANCELGRANT)
	677	status = DLM_NORMAL;
	678
6714d8e8 KH	679	if (status == DLM_NORMAL) {
	680	mlog(0, "kicking the thread\n");
	681	dlm_kick_thread(dlm, res);
	682	} else
	683	dlm_error(status);
	684
	685	dlm_lockres_calc_usage(dlm, res);
	686	dlm_lockres_put(res);
	687	dlm_lock_put(lock);
	688
	689	mlog(0, "returning status=%d!\n", status);
	690	return status;
	691	}
	692	EXPORT_SYMBOL_GPL(dlmunlock);
	693