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

// SPDX-License-Identifier: GPL-2.0-or-later
/* -*- 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.
 */


#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)),
		     kref_read(&lock->lock_refs)-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_FORWARD;

	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);
Commit	Line	Data
328970de	1	// SPDX-License-Identifier: GPL-2.0-or-later
6714d8e8 KH	2	/* -- mode: c; c-basic-offset: 8; --
	3	* vim: noexpandtab sw=8 ts=8 sts=0:
	4	*
	5	* dlmunlock.c
	6	*
	7	* underlying calls for unlocking locks
	8	*
	9	* Copyright (C) 2004 Oracle. All rights reserved.
6714d8e8 KH	10	*/
	11
	12
	13	#include <linux/module.h>
	14	#include <linux/fs.h>
	15	#include <linux/types.h>
6714d8e8	16	#include <linux/highmem.h>
6714d8e8 KH	17	#include <linux/init.h>
	18	#include <linux/sysctl.h>
	19	#include <linux/random.h>
	20	#include <linux/blkdev.h>
	21	#include <linux/socket.h>
	22	#include <linux/inet.h>
	23	#include <linux/spinlock.h>
	24	#include <linux/delay.h>
	25
	26	#include "cluster/heartbeat.h"
	27	#include "cluster/nodemanager.h"
	28	#include "cluster/tcp.h"
	29
	30	#include "dlmapi.h"
	31	#include "dlmcommon.h"
	32
	33	#define MLOG_MASK_PREFIX ML_DLM
	34	#include "cluster/masklog.h"
	35
	36	#define DLM_UNLOCK_FREE_LOCK 0x00000001
	37	#define DLM_UNLOCK_CALL_AST 0x00000002
	38	#define DLM_UNLOCK_REMOVE_LOCK 0x00000004
	39	#define DLM_UNLOCK_REGRANT_LOCK 0x00000008
	40	#define DLM_UNLOCK_CLEAR_CONVERT_TYPE 0x00000010
	41
	42
	43	static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm,
	44	struct dlm_lock_resource *res,
	45	struct dlm_lock *lock,
	46	struct dlm_lockstatus *lksb,
	47	int *actions);
	48	static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm,
	49	struct dlm_lock_resource *res,
	50	struct dlm_lock *lock,
	51	struct dlm_lockstatus *lksb,
	52	int *actions);
	53
	54	static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm,
	55	struct dlm_lock_resource *res,
	56	struct dlm_lock *lock,
	57	struct dlm_lockstatus *lksb,
	58	int flags,
	59	u8 owner);
	60
	61
	62	/*
	63	* according to the spec:
	64	* http://opendlm.sourceforge.net/cvsmirror/opendlm/docs/dlmbook_final.pdf
	65	*
	66	* flags & LKM_CANCEL != 0: must be converting or blocked
	67	* flags & LKM_CANCEL == 0: must be granted
	68	*
	69	* So to unlock a converting lock, you must first cancel the
	70	* convert (passing LKM_CANCEL in flags), then call the unlock
	71	* again (with no LKM_CANCEL in flags).
	72	*/
	73
	74
	75	/*
	76	* locking:
	77	* caller needs: none
	78	* taken: res->spinlock and lock->spinlock taken and dropped
	79	* held on exit: none
	80	* returns: DLM_NORMAL, DLM_NOLOCKMGR, status from network
81	* all callers should have taken an extra ref on lock coming in
82	*/
83	static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm,
84	struct dlm_lock_resource *res,
85	struct dlm_lock *lock,
86	struct dlm_lockstatus *lksb,
87	int flags, int *call_ast,
88	int master_node)
89	{
90	enum dlm_status status;
91	int actions = 0;
92	int in_use;
93	u8 owner;
94
95	mlog(0, "master_node = %d, valblk = %d\n", master_node,
96	flags & LKM_VALBLK);
97
98	if (master_node)
99	BUG_ON(res->owner != dlm->node_num);
100	else
101	BUG_ON(res->owner == dlm->node_num);
102
dabc47de	103	spin_lock(&dlm->ast_lock);
6714d8e8 KH	104	/* We want to be sure that we're not freeing a lock
	105	* that still has AST's pending... */
	106	in_use = !list_empty(&lock->ast_list);
dabc47de	107	spin_unlock(&dlm->ast_lock);
c795b33b	108	if (in_use && !(flags & LKM_CANCEL)) {
6714d8e8 KH	109	mlog(ML_ERROR, "lockres %.*s: Someone is calling dlmunlock "
	110	"while waiting for an ast!", res->lockname.len,
	111	res->lockname.name);
	112	return DLM_BADPARAM;
	113	}
	114
	115	spin_lock(&res->spinlock);
	116	if (res->state & DLM_LOCK_RES_IN_PROGRESS) {
c795b33b	117	if (master_node && !(flags & LKM_CANCEL)) {
6714d8e8 KH	118	mlog(ML_ERROR, "lockres in progress!\n");
	119	spin_unlock(&res->spinlock);
	120	return DLM_FORWARD;
	121	}
	122	/* ok for this to sleep if not in a network handler */
	123	__dlm_wait_on_lockres(res);
	124	res->state \|= DLM_LOCK_RES_IN_PROGRESS;
	125	}
	126	spin_lock(&lock->spinlock);
	127
	128	if (res->state & DLM_LOCK_RES_RECOVERING) {
	129	status = DLM_RECOVERING;
	130	goto leave;
	131	}
	132
71ac1062 KH	133	if (res->state & DLM_LOCK_RES_MIGRATING) {
	134	status = DLM_MIGRATING;
	135	goto leave;
	136	}
6714d8e8 KH	137
	138	/* see above for what the spec says about
	139	* LKM_CANCEL and the lock queue state */
	140	if (flags & LKM_CANCEL)
	141	status = dlm_get_cancel_actions(dlm, res, lock, lksb, &actions);
	142	else
	143	status = dlm_get_unlock_actions(dlm, res, lock, lksb, &actions);
	144
34e3d180	145	if (status != DLM_NORMAL && (status != DLM_CANCELGRANT \|\| !master_node))
6714d8e8 KH	146	goto leave;
	147
	148	/* By now this has been masked out of cancel requests. */
	149	if (flags & LKM_VALBLK) {
	150	/* make the final update to the lvb */
	151	if (master_node)
	152	memcpy(res->lvb, lksb->lvb, DLM_LVB_LEN);
	153	else
	154	flags \|= LKM_PUT_LVB; /* let the send function
	155	* handle it. */
	156	}
	157
	158	if (!master_node) {
	159	owner = res->owner;
	160	/* drop locks and send message */
	161	if (flags & LKM_CANCEL)
	162	lock->cancel_pending = 1;
	163	else
	164	lock->unlock_pending = 1;
	165	spin_unlock(&lock->spinlock);
	166	spin_unlock(&res->spinlock);
	167	status = dlm_send_remote_unlock_request(dlm, res, lock, lksb,
	168	flags, owner);
	169	spin_lock(&res->spinlock);
	170	spin_lock(&lock->spinlock);
	171	/* if the master told us the lock was already granted,
	172	* let the ast handle all of these actions */
a23eac99	173	if (status == DLM_CANCELGRANT) {
6714d8e8 KH	174	actions &= ~(DLM_UNLOCK_REMOVE_LOCK\|
	175	DLM_UNLOCK_REGRANT_LOCK\|
	176	DLM_UNLOCK_CLEAR_CONVERT_TYPE);
2bd63216 SM	177	} else if (status == DLM_RECOVERING \|\|
2bd63216 SM	178	status == DLM_MIGRATING \|\|
27bf6305 TS	179	status == DLM_FORWARD \|\|
	180	status == DLM_NOLOCKMGR
	181	) {
e2faea4c KH	182	/* must clear the actions because this unlock
	183	* is about to be retried. cannot free or do
	184	* any list manipulation. */
	185	mlog(0, "%s:%.*s: clearing actions, %s\n",
	186	dlm->name, res->lockname.len,
	187	res->lockname.name,
	188	status==DLM_RECOVERING?"recovering":
	189	(status==DLM_MIGRATING?"migrating":
27bf6305 TS	190	(status == DLM_FORWARD ? "forward" :
27bf6305 TS	191	"nolockmanager")));
e2faea4c	192	actions = 0;
6714d8e8 KH	193	}
	194	if (flags & LKM_CANCEL)
	195	lock->cancel_pending = 0;
	196	else
	197	lock->unlock_pending = 0;
	198
	199	}
	200
	201	/* get an extra ref on lock. if we are just switching
	202	* lists here, we dont want the lock to go away. */
	203	dlm_lock_get(lock);
	204
	205	if (actions & DLM_UNLOCK_REMOVE_LOCK) {
	206	list_del_init(&lock->list);
	207	dlm_lock_put(lock);
	208	}
	209	if (actions & DLM_UNLOCK_REGRANT_LOCK) {
	210	dlm_lock_get(lock);
	211	list_add_tail(&lock->list, &res->granted);
	212	}
	213	if (actions & DLM_UNLOCK_CLEAR_CONVERT_TYPE) {
	214	mlog(0, "clearing convert_type at %smaster node\n",
	215	master_node ? "" : "non-");
	216	lock->ml.convert_type = LKM_IVMODE;
	217	}
	218
	219	/* remove the extra ref on lock */
	220	dlm_lock_put(lock);
	221
	222	leave:
	223	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
	224	if (!dlm_lock_on_list(&res->converting, lock))
	225	BUG_ON(lock->ml.convert_type != LKM_IVMODE);
	226	else
	227	BUG_ON(lock->ml.convert_type == LKM_IVMODE);
	228	spin_unlock(&lock->spinlock);
	229	spin_unlock(&res->spinlock);
	230	wake_up(&res->wq);
	231
	232	/* let the caller's final dlm_lock_put handle the actual kfree */
	233	if (actions & DLM_UNLOCK_FREE_LOCK) {
	234	/* this should always be coupled with list removal */
	235	BUG_ON(!(actions & DLM_UNLOCK_REMOVE_LOCK));
29004858	236	mlog(0, "lock %u:%llu should be gone now! refs=%d\n",
74aa2585 KH	237	dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
74aa2585 KH	238	dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
2c935bc5	239	kref_read(&lock->lock_refs)-1);
6714d8e8 KH	240	dlm_lock_put(lock);
	241	}
	242	if (actions & DLM_UNLOCK_CALL_AST)
	243	*call_ast = 1;
	244
	245	/* if cancel or unlock succeeded, lvb work is done */
	246	if (status == DLM_NORMAL)
	247	lksb->flags &= ~(DLM_LKSB_PUT_LVB\|DLM_LKSB_GET_LVB);
	248
	249	return status;
	250	}
	251
	252	void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
	253	struct dlm_lock *lock)
	254	{
	255	/* leave DLM_LKSB_PUT_LVB on the lksb so any final
	256	* update of the lvb will be sent to the new master */
	257	list_del_init(&lock->list);
	258	}
	259
	260	void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
	261	struct dlm_lock *lock)
	262	{
f116629d	263	list_move_tail(&lock->list, &res->granted);
6714d8e8 KH	264	lock->ml.convert_type = LKM_IVMODE;
	265	}
	266
	267
	268	static inline enum dlm_status dlmunlock_master(struct dlm_ctxt *dlm,
	269	struct dlm_lock_resource *res,
	270	struct dlm_lock *lock,
	271	struct dlm_lockstatus *lksb,
	272	int flags,
	273	int *call_ast)
	274	{
	275	return dlmunlock_common(dlm, res, lock, lksb, flags, call_ast, 1);
	276	}
	277
	278	static inline enum dlm_status dlmunlock_remote(struct dlm_ctxt *dlm,
	279	struct dlm_lock_resource *res,
	280	struct dlm_lock *lock,
	281	struct dlm_lockstatus *lksb,
	282	int flags, int *call_ast)
	283	{
	284	return dlmunlock_common(dlm, res, lock, lksb, flags, call_ast, 0);
	285	}
	286
	287	/*
	288	* locking:
	289	* caller needs: none
	290	* taken: none
	291	* held on exit: none
	292	* returns: DLM_NORMAL, DLM_NOLOCKMGR, status from network
	293	*/
	294	static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm,
	295	struct dlm_lock_resource *res,
	296	struct dlm_lock *lock,
	297	struct dlm_lockstatus *lksb,
	298	int flags,
	299	u8 owner)
	300	{
	301	struct dlm_unlock_lock unlock;
	302	int tmpret;
	303	enum dlm_status ret;
	304	int status = 0;
	305	struct kvec vec[2];
	306	size_t veclen = 1;
	307
ef6b689b	308	mlog(0, "%.*s\n", res->lockname.len, res->lockname.name);
0032abd6	309
2580a580 KH	310	if (owner == dlm->node_num) {
	311	/* ended up trying to contact ourself. this means
	312	* that the lockres had been remote but became local
	313	* via a migration. just retry it, now as local */
	314	mlog(0, "%s:%.*s: this node became the master due to a "
	315	"migration, re-evaluate now\n", dlm->name,
	316	res->lockname.len, res->lockname.name);
	317	return DLM_FORWARD;
	318	}
6714d8e8 KH	319
	320	memset(&unlock, 0, sizeof(unlock));
	321	unlock.node_idx = dlm->node_num;
	322	unlock.flags = cpu_to_be32(flags);
	323	unlock.cookie = lock->ml.cookie;
	324	unlock.namelen = res->lockname.len;
	325	memcpy(unlock.name, res->lockname.name, unlock.namelen);
	326
	327	vec[0].iov_len = sizeof(struct dlm_unlock_lock);
	328	vec[0].iov_base = &unlock;
	329
	330	if (flags & LKM_PUT_LVB) {
	331	/* extra data to send if we are updating lvb */
	332	vec[1].iov_len = DLM_LVB_LEN;
	333	vec[1].iov_base = lock->lksb->lvb;
	334	veclen++;
	335	}
	336
	337	tmpret = o2net_send_message_vec(DLM_UNLOCK_LOCK_MSG, dlm->key,
	338	vec, veclen, owner, &status);
	339	if (tmpret >= 0) {
	340	// successfully sent and received
a23eac99	341	if (status == DLM_FORWARD)
6714d8e8	342	mlog(0, "master was in-progress. retry\n");
a23eac99	343	ret = status;
6714d8e8	344	} else {
a5196ec5 WW	345	mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
a5196ec5 WW	346	"node %u\n", tmpret, DLM_UNLOCK_LOCK_MSG, dlm->key, owner);
6714d8e8 KH	347	if (dlm_is_host_down(tmpret)) {
	348	/* NOTE: this seems strange, but it is what we want.
	349	* when the master goes down during a cancel or
	350	* unlock, the recovery code completes the operation
	351	* as if the master had not died, then passes the
	352	* updated state to the recovery master. this thread
	353	* just needs to finish out the operation and call
	354	* the unlockast. */
27bf6305 TS	355	if (dlm_is_node_dead(dlm, owner))
	356	ret = DLM_NORMAL;
	357	else
	358	ret = DLM_NOLOCKMGR;
6714d8e8 KH	359	} else {
	360	/* something bad. this will BUG in ocfs2 */
	361	ret = dlm_err_to_dlm_status(tmpret);
	362	}
6714d8e8 KH	363	}
	364
	365	return ret;
	366	}
	367
	368	/*
	369	* locking:
	370	* caller needs: none
	371	* taken: takes and drops res->spinlock
	372	* held on exit: none
	373	* returns: DLM_NORMAL, DLM_BADARGS, DLM_IVLOCKID,
	374	* return value from dlmunlock_master
	375	*/
d74c9803 KH	376	int dlm_unlock_lock_handler(struct o2net_msg msg, u32 len, void data,
d74c9803 KH	377	void **ret_data)
6714d8e8 KH	378	{
	379	struct dlm_ctxt *dlm = data;
	380	struct dlm_unlock_lock unlock = (struct dlm_unlock_lock )msg->buf;
	381	struct dlm_lock_resource *res = NULL;
6714d8e8 KH	382	struct dlm_lock *lock = NULL;
	383	enum dlm_status status = DLM_NORMAL;
	384	int found = 0, i;
	385	struct dlm_lockstatus *lksb = NULL;
	386	int ignore;
	387	u32 flags;
	388	struct list_head *queue;
	389
	390	flags = be32_to_cpu(unlock->flags);
	391
	392	if (flags & LKM_GET_LVB) {
	393	mlog(ML_ERROR, "bad args! GET_LVB specified on unlock!\n");
	394	return DLM_BADARGS;
	395	}
	396
	397	if ((flags & (LKM_PUT_LVB\|LKM_CANCEL)) == (LKM_PUT_LVB\|LKM_CANCEL)) {
	398	mlog(ML_ERROR, "bad args! cannot modify lvb on a CANCEL "
	399	"request!\n");
	400	return DLM_BADARGS;
	401	}
	402
	403	if (unlock->namelen > DLM_LOCKID_NAME_MAX) {
	404	mlog(ML_ERROR, "Invalid name length in unlock handler!\n");
	405	return DLM_IVBUFLEN;
	406	}
	407
	408	if (!dlm_grab(dlm))
c372f219	409	return DLM_FORWARD;
6714d8e8 KH	410
	411	mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
	412	"Domain %s not fully joined!\n", dlm->name);
	413
	414	mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" : "none");
	415
	416	res = dlm_lookup_lockres(dlm, unlock->name, unlock->namelen);
	417	if (!res) {
	418	/* We assume here that a no lock resource simply means
	419	* it was migrated away and destroyed before the other
	420	* node could detect it. */
	421	mlog(0, "returning DLM_FORWARD -- res no longer exists\n");
	422	status = DLM_FORWARD;
	423	goto not_found;
	424	}
	425
	426	queue=&res->granted;
	427	found = 0;
	428	spin_lock(&res->spinlock);
	429	if (res->state & DLM_LOCK_RES_RECOVERING) {
	430	spin_unlock(&res->spinlock);
	431	mlog(0, "returning DLM_RECOVERING\n");
	432	status = DLM_RECOVERING;
	433	goto leave;
	434	}
	435
	436	if (res->state & DLM_LOCK_RES_MIGRATING) {
	437	spin_unlock(&res->spinlock);
	438	mlog(0, "returning DLM_MIGRATING\n");
	439	status = DLM_MIGRATING;
	440	goto leave;
	441	}
	442
	443	if (res->owner != dlm->node_num) {
	444	spin_unlock(&res->spinlock);
	445	mlog(0, "returning DLM_FORWARD -- not master\n");
	446	status = DLM_FORWARD;
	447	goto leave;
	448	}
	449
	450	for (i=0; i<3; i++) {
df53cd3b	451	list_for_each_entry(lock, queue, list) {
6714d8e8 KH	452	if (lock->ml.cookie == unlock->cookie &&
	453	lock->ml.node == unlock->node_idx) {
	454	dlm_lock_get(lock);
	455	found = 1;
	456	break;
	457	}
	458	}
	459	if (found)
	460	break;
	461	/* scan granted -> converting -> blocked queues */
	462	queue++;
	463	}
	464	spin_unlock(&res->spinlock);
	465	if (!found) {
	466	status = DLM_IVLOCKID;
	467	goto not_found;
	468	}
	469
	470	/* lock was found on queue */
	471	lksb = lock->lksb;
4b1af774 KH	472	if (flags & (LKM_VALBLK\|LKM_PUT_LVB) &&
	473	lock->ml.type != LKM_EXMODE)
	474	flags &= ~(LKM_VALBLK\|LKM_PUT_LVB);
	475
6714d8e8 KH	476	/* unlockast only called on originating node */
	477	if (flags & LKM_PUT_LVB) {
	478	lksb->flags \|= DLM_LKSB_PUT_LVB;
	479	memcpy(&lksb->lvb[0], &unlock->lvb[0], DLM_LVB_LEN);
	480	}
	481
	482	/* if this is in-progress, propagate the DLM_FORWARD
	483	* all the way back out */
	484	status = dlmunlock_master(dlm, res, lock, lksb, flags, &ignore);
	485	if (status == DLM_FORWARD)
	486	mlog(0, "lockres is in progress\n");
	487
	488	if (flags & LKM_PUT_LVB)
	489	lksb->flags &= ~DLM_LKSB_PUT_LVB;
	490
	491	dlm_lockres_calc_usage(dlm, res);
	492	dlm_kick_thread(dlm, res);
	493
	494	not_found:
	495	if (!found)
	496	mlog(ML_ERROR, "failed to find lock to unlock! "
29004858	497	"cookie=%u:%llu\n",
74aa2585 KH	498	dlm_get_lock_cookie_node(be64_to_cpu(unlock->cookie)),
74aa2585 KH	499	dlm_get_lock_cookie_seq(be64_to_cpu(unlock->cookie)));
a23eac99	500	else
6714d8e8	501	dlm_lock_put(lock);
6714d8e8 KH	502
	503	leave:
	504	if (res)
	505	dlm_lockres_put(res);
	506
	507	dlm_put(dlm);
	508
	509	return status;
	510	}
	511
	512
	513	static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm,
	514	struct dlm_lock_resource *res,
	515	struct dlm_lock *lock,
	516	struct dlm_lockstatus *lksb,
	517	int *actions)
	518	{
	519	enum dlm_status status;
	520
	521	if (dlm_lock_on_list(&res->blocked, lock)) {
	522	/* cancel this outright */
6714d8e8 KH	523	status = DLM_NORMAL;
	524	*actions = (DLM_UNLOCK_CALL_AST \|
	525	DLM_UNLOCK_REMOVE_LOCK);
	526	} else if (dlm_lock_on_list(&res->converting, lock)) {
	527	/* cancel the request, put back on granted */
6714d8e8 KH	528	status = DLM_NORMAL;
	529	*actions = (DLM_UNLOCK_CALL_AST \|
	530	DLM_UNLOCK_REMOVE_LOCK \|
	531	DLM_UNLOCK_REGRANT_LOCK \|
	532	DLM_UNLOCK_CLEAR_CONVERT_TYPE);
	533	} else if (dlm_lock_on_list(&res->granted, lock)) {
a23eac99 KH	534	/* too late, already granted. */
a23eac99 KH	535	status = DLM_CANCELGRANT;
6714d8e8 KH	536	*actions = DLM_UNLOCK_CALL_AST;
	537	} else {
	538	mlog(ML_ERROR, "lock to cancel is not on any list!\n");
6714d8e8 KH	539	status = DLM_IVLOCKID;
	540	*actions = 0;
	541	}
	542	return status;
	543	}
	544
	545	static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm,
	546	struct dlm_lock_resource *res,
	547	struct dlm_lock *lock,
	548	struct dlm_lockstatus *lksb,
	549	int *actions)
	550	{
	551	enum dlm_status status;
	552
	553	/* unlock request */
	554	if (!dlm_lock_on_list(&res->granted, lock)) {
6714d8e8 KH	555	status = DLM_DENIED;
	556	dlm_error(status);
	557	*actions = 0;
	558	} else {
	559	/* unlock granted lock */
6714d8e8 KH	560	status = DLM_NORMAL;
	561	*actions = (DLM_UNLOCK_FREE_LOCK \|
	562	DLM_UNLOCK_CALL_AST \|
	563	DLM_UNLOCK_REMOVE_LOCK);
	564	}
	565	return status;
	566	}
	567
	568	/* there seems to be no point in doing this async
	569	* since (even for the remote case) there is really
	570	* no work to queue up... so just do it and fire the
	571	* unlockast by hand when done... */
	572	enum dlm_status dlmunlock(struct dlm_ctxt dlm, struct dlm_lockstatus lksb,
	573	int flags, dlm_astunlockfunc_t unlockast, void data)
	574	{
	575	enum dlm_status status;
	576	struct dlm_lock_resource *res;
	577	struct dlm_lock *lock = NULL;
	578	int call_ast, is_master;
	579
6714d8e8 KH	580	if (!lksb) {
	581	dlm_error(DLM_BADARGS);
	582	return DLM_BADARGS;
	583	}
	584
	585	if (flags & ~(LKM_CANCEL \| LKM_VALBLK \| LKM_INVVALBLK)) {
	586	dlm_error(DLM_BADPARAM);
	587	return DLM_BADPARAM;
	588	}
	589
	590	if ((flags & (LKM_VALBLK \| LKM_CANCEL)) == (LKM_VALBLK \| LKM_CANCEL)) {
	591	mlog(0, "VALBLK given with CANCEL: ignoring VALBLK\n");
	592	flags &= ~LKM_VALBLK;
	593	}
	594
	595	if (!lksb->lockid \|\| !lksb->lockid->lockres) {
	596	dlm_error(DLM_BADPARAM);
	597	return DLM_BADPARAM;
	598	}
	599
	600	lock = lksb->lockid;
	601	BUG_ON(!lock);
	602	dlm_lock_get(lock);
	603
	604	res = lock->lockres;
	605	BUG_ON(!res);
	606	dlm_lockres_get(res);
	607	retry:
	608	call_ast = 0;
	609	/* need to retry up here because owner may have changed */
	610	mlog(0, "lock=%p res=%p\n", lock, res);
	611
	612	spin_lock(&res->spinlock);
	613	is_master = (res->owner == dlm->node_num);
4b1af774 KH	614	if (flags & LKM_VALBLK && lock->ml.type != LKM_EXMODE)
4b1af774 KH	615	flags &= ~LKM_VALBLK;
6714d8e8 KH	616	spin_unlock(&res->spinlock);
	617
	618	if (is_master) {
	619	status = dlmunlock_master(dlm, res, lock, lksb, flags,
	620	&call_ast);
	621	mlog(0, "done calling dlmunlock_master: returned %d, "
	622	"call_ast is %d\n", status, call_ast);
	623	} else {
	624	status = dlmunlock_remote(dlm, res, lock, lksb, flags,
	625	&call_ast);
	626	mlog(0, "done calling dlmunlock_remote: returned %d, "
	627	"call_ast is %d\n", status, call_ast);
	628	}
	629
	630	if (status == DLM_RECOVERING \|\|
	631	status == DLM_MIGRATING \|\|
27bf6305 TS	632	status == DLM_FORWARD \|\|
	633	status == DLM_NOLOCKMGR) {
	634
6714d8e8 KH	635	/* We want to go away for a tiny bit to allow recovery
	636	* / migration to complete on this resource. I don't
	637	* know of any wait queue we could sleep on as this
	638	* may be happening on another node. Perhaps the
	639	* proper solution is to queue up requests on the
	640	* other end? */
	641
	642	/* do we want to yield(); ?? */
	643	msleep(50);
	644
	645	mlog(0, "retrying unlock due to pending recovery/"
27bf6305	646	"migration/in-progress/reconnect\n");
6714d8e8 KH	647	goto retry;
	648	}
	649
	650	if (call_ast) {
a23eac99	651	mlog(0, "calling unlockast(%p, %d)\n", data, status);
6714d8e8	652	if (is_master) {
2bd63216	653	/* it is possible that there is one last bast
6714d8e8 KH	654	* pending. make sure it is flushed, then
	655	* call the unlockast.
	656	* not an issue if this is a mastered remotely,
	657	* since this lock has been removed from the
	658	* lockres queues and cannot be found. */
	659	dlm_kick_thread(dlm, NULL);
2bd63216	660	wait_event(dlm->ast_wq,
6714d8e8 KH	661	dlm_lock_basts_flushed(dlm, lock));
6714d8e8 KH	662	}
a23eac99	663	(*unlockast)(data, status);
6714d8e8 KH	664	}
6714d8e8 KH	665
a23eac99 KH	666	if (status == DLM_CANCELGRANT)
	667	status = DLM_NORMAL;
	668
6714d8e8 KH	669	if (status == DLM_NORMAL) {
	670	mlog(0, "kicking the thread\n");
	671	dlm_kick_thread(dlm, res);
	672	} else
	673	dlm_error(status);
	674
	675	dlm_lockres_calc_usage(dlm, res);
	676	dlm_lockres_put(res);
	677	dlm_lock_put(lock);
	678
	679	mlog(0, "returning status=%d!\n", status);
	680	return status;
	681	}
	682	EXPORT_SYMBOL_GPL(dlmunlock);
	683