Merge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw

* git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw: (34 commits)
  [GFS2] Uncomment sprintf_symbol calling code
  [DLM] lowcomms style
  [GFS2] printk warning fixes
  [GFS2] Patch to fix mmap of stuffed files
  [GFS2] use lib/parser for parsing mount options
  [DLM] Lowcomms nodeid range & initialisation fixes
  [DLM] Fix dlm_lowcoms_stop hang
  [DLM] fix mode munging
  [GFS2] lockdump improvements
  [GFS2] Patch to detect corrupt number of dir entries in leaf and/or inode blocks
  [GFS2] bz 236008: Kernel gpf doing cat /debugfs/gfs2/xxx (lock dump)
  [DLM] fs/dlm/ast.c should #include "ast.h"
  [DLM] Consolidate transport protocols
  [DLM] Remove redundant assignment
  [GFS2] Fix bz 234168 (ignoring rgrp flags)
  [DLM] change lkid format
  [DLM] interface for purge (2/2)
  [DLM] add orphan purging code (1/2)
  [DLM] split create_message function
  [GFS2] Set drop_count to 0 (off) by default
  ...

12 files changed, 1508 insertions, 1676 deletions

diff --git a/fs/dlm/Kconfig b/fs/dlm/Kconfig
index 6fa7b0d5c04..69a94690e49 100644
--- a/fs/dlm/Kconfig
+++ b/fs/dlm/Kconfig
@@ -3,36 +3,19 @@ menu "Distributed Lock Manager"
 
 config DLM
 	tristate "Distributed Lock Manager (DLM)"
-	depends on SYSFS && (IPV6 || IPV6=n)
+	depends on IPV6 || IPV6=n
 	select CONFIGFS_FS
-	select IP_SCTP if DLM_SCTP
+	select IP_SCTP
 	help
-	  A general purpose distributed lock manager for kernel or userspace
-	  applications.
-
-choice
-	prompt "Select DLM communications protocol"
-	depends on DLM
-	default DLM_TCP
-	help
-	  The DLM Can use TCP or SCTP for it's network communications.
-	  SCTP supports multi-homed operations whereas TCP doesn't.
-	  However, SCTP seems to have stability problems at the moment.
-
-config DLM_TCP
-	bool "TCP/IP"
-
-config DLM_SCTP
-	bool "SCTP"
-
-endchoice
+	A general purpose distributed lock manager for kernel or userspace
+	applications.
 
 config DLM_DEBUG
 	bool "DLM debugging"
 	depends on DLM
 	help
-	  Under the debugfs mount point, the name of each lockspace will
-	  appear as a file in the "dlm" directory.  The output is the
-	  list of resource and locks the local node knows about.
+	Under the debugfs mount point, the name of each lockspace will
+	appear as a file in the "dlm" directory.  The output is the
+	list of resource and locks the local node knows about.
 
 endmenu
diff --git a/fs/dlm/Makefile b/fs/dlm/Makefile
index 65388944eba..604cf7dc5f3 100644
--- a/fs/dlm/Makefile
+++ b/fs/dlm/Makefile
@@ -8,14 +8,12 @@ dlm-y :=			ast.o \
 				member.o \
 				memory.o \
 				midcomms.o \
+				lowcomms.o \
 				rcom.o \
 				recover.o \
 				recoverd.o \
 				requestqueue.o \
 				user.o \
-				util.o
+				util.o 
 dlm-$(CONFIG_DLM_DEBUG) +=	debug_fs.o
 
-dlm-$(CONFIG_DLM_TCP)   += lowcomms-tcp.o
-
-dlm-$(CONFIG_DLM_SCTP)  += lowcomms-sctp.o
\ No newline at end of file
diff --git a/fs/dlm/ast.c b/fs/dlm/ast.c
index f91d39cb1e0..6308122890c 100644
--- a/fs/dlm/ast.c
+++ b/fs/dlm/ast.c
@@ -14,6 +14,7 @@
 #include "dlm_internal.h"
 #include "lock.h"
 #include "user.h"
+#include "ast.h"
 
 #define WAKE_ASTS  0
 
diff --git a/fs/dlm/config.c b/fs/dlm/config.c
index 8665c88e5af..822abdcd143 100644
--- a/fs/dlm/config.c
+++ b/fs/dlm/config.c
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -89,6 +89,7 @@ struct cluster {
 	unsigned int cl_toss_secs;
 	unsigned int cl_scan_secs;
 	unsigned int cl_log_debug;
+	unsigned int cl_protocol;
 };
 
 enum {
@@ -101,6 +102,7 @@ enum {
 	CLUSTER_ATTR_TOSS_SECS,
 	CLUSTER_ATTR_SCAN_SECS,
 	CLUSTER_ATTR_LOG_DEBUG,
+	CLUSTER_ATTR_PROTOCOL,
 };
 
 struct cluster_attribute {
@@ -159,6 +161,7 @@ CLUSTER_ATTR(recover_timer, 1);
 CLUSTER_ATTR(toss_secs, 1);
 CLUSTER_ATTR(scan_secs, 1);
 CLUSTER_ATTR(log_debug, 0);
+CLUSTER_ATTR(protocol, 0);
 
 static struct configfs_attribute *cluster_attrs[] = {
 	[CLUSTER_ATTR_TCP_PORT] = &cluster_attr_tcp_port.attr,
@@ -170,6 +173,7 @@ static struct configfs_attribute *cluster_attrs[] = {
 	[CLUSTER_ATTR_TOSS_SECS] = &cluster_attr_toss_secs.attr,
 	[CLUSTER_ATTR_SCAN_SECS] = &cluster_attr_scan_secs.attr,
 	[CLUSTER_ATTR_LOG_DEBUG] = &cluster_attr_log_debug.attr,
+	[CLUSTER_ATTR_PROTOCOL] = &cluster_attr_protocol.attr,
 	NULL,
 };
 
@@ -904,6 +908,7 @@ int dlm_our_addr(struct sockaddr_storage *addr, int num)
 #define DEFAULT_TOSS_SECS         10
 #define DEFAULT_SCAN_SECS          5
 #define DEFAULT_LOG_DEBUG          0
+#define DEFAULT_PROTOCOL           0
 
 struct dlm_config_info dlm_config = {
 	.ci_tcp_port = DEFAULT_TCP_PORT,
@@ -914,6 +919,7 @@ struct dlm_config_info dlm_config = {
 	.ci_recover_timer = DEFAULT_RECOVER_TIMER,
 	.ci_toss_secs = DEFAULT_TOSS_SECS,
 	.ci_scan_secs = DEFAULT_SCAN_SECS,
-	.ci_log_debug = DEFAULT_LOG_DEBUG
+	.ci_log_debug = DEFAULT_LOG_DEBUG,
+	.ci_protocol = DEFAULT_PROTOCOL
 };
 
diff --git a/fs/dlm/config.h b/fs/dlm/config.h
index 1e978611a96..967cc3d72e5 100644
--- a/fs/dlm/config.h
+++ b/fs/dlm/config.h
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -26,6 +26,7 @@ struct dlm_config_info {
 	int ci_toss_secs;
 	int ci_scan_secs;
 	int ci_log_debug;
+	int ci_protocol;
 };
 
 extern struct dlm_config_info dlm_config;
diff --git a/fs/dlm/dlm_internal.h b/fs/dlm/dlm_internal.h
index 61d93201e1b..30994d68f6a 100644
--- a/fs/dlm/dlm_internal.h
+++ b/fs/dlm/dlm_internal.h
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -210,6 +210,9 @@ struct dlm_args {
 #define DLM_IFL_MSTCPY		0x00010000
 #define DLM_IFL_RESEND		0x00020000
 #define DLM_IFL_DEAD		0x00040000
+#define DLM_IFL_OVERLAP_UNLOCK  0x00080000
+#define DLM_IFL_OVERLAP_CANCEL  0x00100000
+#define DLM_IFL_ENDOFLIFE	0x00200000
 #define DLM_IFL_USER		0x00000001
 #define DLM_IFL_ORPHAN		0x00000002
 
@@ -230,8 +233,8 @@ struct dlm_lkb {
 	int8_t			lkb_grmode;	/* granted lock mode */
 	int8_t			lkb_bastmode;	/* requested mode */
 	int8_t			lkb_highbast;	/* highest mode bast sent for */
-
 	int8_t			lkb_wait_type;	/* type of reply waiting for */
+	int8_t			lkb_wait_count;
 	int8_t			lkb_ast_type;	/* type of ast queued for */
 
 	struct list_head	lkb_idtbl_list;	/* lockspace lkbtbl */
@@ -339,6 +342,7 @@ struct dlm_header {
 #define DLM_MSG_LOOKUP		11
 #define DLM_MSG_REMOVE		12
 #define DLM_MSG_LOOKUP_REPLY	13
+#define DLM_MSG_PURGE		14
 
 struct dlm_message {
 	struct dlm_header	m_header;
@@ -440,6 +444,9 @@ struct dlm_ls {
 	struct mutex		ls_waiters_mutex;
 	struct list_head	ls_waiters;	/* lkbs needing a reply */
 
+	struct mutex		ls_orphans_mutex;
+	struct list_head	ls_orphans;
+
 	struct list_head	ls_nodes;	/* current nodes in ls */
 	struct list_head	ls_nodes_gone;	/* dead node list, recovery */
 	int			ls_num_nodes;	/* number of nodes in ls */
diff --git a/fs/dlm/lock.c b/fs/dlm/lock.c
index e725005fafd..d8d6e729f96 100644
--- a/fs/dlm/lock.c
+++ b/fs/dlm/lock.c
@@ -1,7 +1,7 @@
 /******************************************************************************
 *******************************************************************************
 **
-**  Copyright (C) 2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2005-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -85,6 +85,7 @@ static int _request_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
 				    struct dlm_message *ms);
 static int receive_extralen(struct dlm_message *ms);
+static void do_purge(struct dlm_ls *ls, int nodeid, int pid);
 
 /*
  * Lock compatibilty matrix - thanks Steve
@@ -223,6 +224,16 @@ static inline int is_demoted(struct dlm_lkb *lkb)
 	return (lkb->lkb_sbflags & DLM_SBF_DEMOTED);
 }
 
+static inline int is_altmode(struct dlm_lkb *lkb)
+{
+	return (lkb->lkb_sbflags & DLM_SBF_ALTMODE);
+}
+
+static inline int is_granted(struct dlm_lkb *lkb)
+{
+	return (lkb->lkb_status == DLM_LKSTS_GRANTED);
+}
+
 static inline int is_remote(struct dlm_rsb *r)
 {
 	DLM_ASSERT(r->res_nodeid >= 0, dlm_print_rsb(r););
@@ -254,6 +265,22 @@ static inline int down_conversion(struct dlm_lkb *lkb)
 	return (!middle_conversion(lkb) && lkb->lkb_rqmode < lkb->lkb_grmode);
 }
 
+static inline int is_overlap_unlock(struct dlm_lkb *lkb)
+{
+	return lkb->lkb_flags & DLM_IFL_OVERLAP_UNLOCK;
+}
+
+static inline int is_overlap_cancel(struct dlm_lkb *lkb)
+{
+	return lkb->lkb_flags & DLM_IFL_OVERLAP_CANCEL;
+}
+
+static inline int is_overlap(struct dlm_lkb *lkb)
+{
+	return (lkb->lkb_flags & (DLM_IFL_OVERLAP_UNLOCK |
+				  DLM_IFL_OVERLAP_CANCEL));
+}
+
 static void queue_cast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
 {
 	if (is_master_copy(lkb))
@@ -267,6 +294,12 @@ static void queue_cast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rv)
 	dlm_add_ast(lkb, AST_COMP);
 }
 
+static inline void queue_cast_overlap(struct dlm_rsb *r, struct dlm_lkb *lkb)
+{
+	queue_cast(r, lkb,
+		   is_overlap_unlock(lkb) ? -DLM_EUNLOCK : -DLM_ECANCEL);
+}
+
 static void queue_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rqmode)
 {
 	if (is_master_copy(lkb))
@@ -547,6 +580,7 @@ static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
 	lkb->lkb_grmode = DLM_LOCK_IV;
 	kref_init(&lkb->lkb_ref);
 	INIT_LIST_HEAD(&lkb->lkb_ownqueue);
+	INIT_LIST_HEAD(&lkb->lkb_rsb_lookup);
 
 	get_random_bytes(&bucket, sizeof(bucket));
 	bucket &= (ls->ls_lkbtbl_size - 1);
@@ -556,7 +590,7 @@ static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
 	/* counter can roll over so we must verify lkid is not in use */
 
 	while (lkid == 0) {
-		lkid = bucket | (ls->ls_lkbtbl[bucket].counter++ << 16);
+		lkid = (bucket << 16) | ls->ls_lkbtbl[bucket].counter++;
 
 		list_for_each_entry(tmp, &ls->ls_lkbtbl[bucket].list,
 				    lkb_idtbl_list) {
@@ -577,8 +611,8 @@ static int create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
 
 static struct dlm_lkb *__find_lkb(struct dlm_ls *ls, uint32_t lkid)
 {
-	uint16_t bucket = lkid & 0xFFFF;
 	struct dlm_lkb *lkb;
+	uint16_t bucket = (lkid >> 16);
 
 	list_for_each_entry(lkb, &ls->ls_lkbtbl[bucket].list, lkb_idtbl_list) {
 		if (lkb->lkb_id == lkid)
@@ -590,7 +624,7 @@ static struct dlm_lkb *__find_lkb(struct dlm_ls *ls, uint32_t lkid)
 static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret)
 {
 	struct dlm_lkb *lkb;
-	uint16_t bucket = lkid & 0xFFFF;
+	uint16_t bucket = (lkid >> 16);
 
 	if (bucket >= ls->ls_lkbtbl_size)
 		return -EBADSLT;
@@ -620,7 +654,7 @@ static void kill_lkb(struct kref *kref)
 
 static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb)
 {
-	uint16_t bucket = lkb->lkb_id & 0xFFFF;
+	uint16_t bucket = (lkb->lkb_id >> 16);
 
 	write_lock(&ls->ls_lkbtbl[bucket].lock);
 	if (kref_put(&lkb->lkb_ref, kill_lkb)) {
@@ -735,23 +769,75 @@ static void move_lkb(struct dlm_rsb *r, struct dlm_lkb *lkb, int sts)
 	unhold_lkb(lkb);
 }
 
+static int msg_reply_type(int mstype)
+{
+	switch (mstype) {
+	case DLM_MSG_REQUEST:
+		return DLM_MSG_REQUEST_REPLY;
+	case DLM_MSG_CONVERT:
+		return DLM_MSG_CONVERT_REPLY;
+	case DLM_MSG_UNLOCK:
+		return DLM_MSG_UNLOCK_REPLY;
+	case DLM_MSG_CANCEL:
+		return DLM_MSG_CANCEL_REPLY;
+	case DLM_MSG_LOOKUP:
+		return DLM_MSG_LOOKUP_REPLY;
+	}
+	return -1;
+}
+
 /* add/remove lkb from global waiters list of lkb's waiting for
    a reply from a remote node */
 
-static void add_to_waiters(struct dlm_lkb *lkb, int mstype)
+static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
 {
 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
+	int error = 0;
 
 	mutex_lock(&ls->ls_waiters_mutex);
-	if (lkb->lkb_wait_type) {
-		log_print("add_to_waiters error %d", lkb->lkb_wait_type);
+
+	if (is_overlap_unlock(lkb) ||
+	    (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
+		error = -EINVAL;
+		goto out;
+	}
+
+	if (lkb->lkb_wait_type || is_overlap_cancel(lkb)) {
+		switch (mstype) {
+		case DLM_MSG_UNLOCK:
+			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
+			break;
+		case DLM_MSG_CANCEL:
+			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
+			break;
+		default:
+			error = -EBUSY;
+			goto out;
+		}
+		lkb->lkb_wait_count++;
+		hold_lkb(lkb);
+
+		log_debug(ls, "add overlap %x cur %d new %d count %d flags %x",
+			  lkb->lkb_id, lkb->lkb_wait_type, mstype,
+			  lkb->lkb_wait_count, lkb->lkb_flags);
 		goto out;
 	}
+
+	DLM_ASSERT(!lkb->lkb_wait_count,
+		   dlm_print_lkb(lkb);
+		   printk("wait_count %d\n", lkb->lkb_wait_count););
+
+	lkb->lkb_wait_count++;
 	lkb->lkb_wait_type = mstype;
-	kref_get(&lkb->lkb_ref);
+	hold_lkb(lkb);
 	list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
  out:
+	if (error)
+		log_error(ls, "add_to_waiters %x error %d flags %x %d %d %s",
+			  lkb->lkb_id, error, lkb->lkb_flags, mstype,
+			  lkb->lkb_wait_type, lkb->lkb_resource->res_name);
 	mutex_unlock(&ls->ls_waiters_mutex);
+	return error;
 }
 
 /* We clear the RESEND flag because we might be taking an lkb off the waiters
@@ -759,34 +845,85 @@ static void add_to_waiters(struct dlm_lkb *lkb, int mstype)
    request reply on the requestqueue) between dlm_recover_waiters_pre() which
    set RESEND and dlm_recover_waiters_post() */
 
-static int _remove_from_waiters(struct dlm_lkb *lkb)
+static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype)
 {
-	int error = 0;
+	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
+	int overlap_done = 0;
 
-	if (!lkb->lkb_wait_type) {
-		log_print("remove_from_waiters error");
-		error = -EINVAL;
-		goto out;
+	if (is_overlap_unlock(lkb) && (mstype == DLM_MSG_UNLOCK_REPLY)) {
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
+		overlap_done = 1;
+		goto out_del;
+	}
+
+	if (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL_REPLY)) {
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
+		overlap_done = 1;
+		goto out_del;
+	}
+
+	/* N.B. type of reply may not always correspond to type of original
+	   msg due to lookup->request optimization, verify others? */
+
+	if (lkb->lkb_wait_type) {
+		lkb->lkb_wait_type = 0;
+		goto out_del;
+	}
+
+	log_error(ls, "remove_from_waiters lkid %x flags %x types %d %d",
+		  lkb->lkb_id, lkb->lkb_flags, mstype, lkb->lkb_wait_type);
+	return -1;
+
+ out_del:
+	/* the force-unlock/cancel has completed and we haven't recvd a reply
+	   to the op that was in progress prior to the unlock/cancel; we
+	   give up on any reply to the earlier op.  FIXME: not sure when/how
+	   this would happen */
+
+	if (overlap_done && lkb->lkb_wait_type) {
+		log_error(ls, "remove_from_waiters %x reply %d give up on %d",
+			  lkb->lkb_id, mstype, lkb->lkb_wait_type);
+		lkb->lkb_wait_count--;
+		lkb->lkb_wait_type = 0;
 	}
-	lkb->lkb_wait_type = 0;
+
+	DLM_ASSERT(lkb->lkb_wait_count, dlm_print_lkb(lkb););
+
 	lkb->lkb_flags &= ~DLM_IFL_RESEND;
-	list_del(&lkb->lkb_wait_reply);
+	lkb->lkb_wait_count--;
+	if (!lkb->lkb_wait_count)
+		list_del_init(&lkb->lkb_wait_reply);
 	unhold_lkb(lkb);
- out:
-	return error;
+	return 0;
 }
 
-static int remove_from_waiters(struct dlm_lkb *lkb)
+static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
 {
 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
 	int error;
 
 	mutex_lock(&ls->ls_waiters_mutex);
-	error = _remove_from_waiters(lkb);
+	error = _remove_from_waiters(lkb, mstype);
 	mutex_unlock(&ls->ls_waiters_mutex);
 	return error;
 }
 
+/* Handles situations where we might be processing a "fake" or "stub" reply in
+   which we can't try to take waiters_mutex again. */
+
+static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms)
+{
+	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
+	int error;
+
+	if (ms != &ls->ls_stub_ms)
+		mutex_lock(&ls->ls_waiters_mutex);
+	error = _remove_from_waiters(lkb, ms->m_type);
+	if (ms != &ls->ls_stub_ms)
+		mutex_unlock(&ls->ls_waiters_mutex);
+	return error;
+}
+
 static void dir_remove(struct dlm_rsb *r)
 {
 	int to_nodeid;
@@ -988,8 +1125,14 @@ static void remove_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
 	_remove_lock(r, lkb);
 }
 
-static void revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
+/* returns: 0 did nothing
+	    1 moved lock to granted
+	   -1 removed lock */
+
+static int revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
 {
+	int rv = 0;
+
 	lkb->lkb_rqmode = DLM_LOCK_IV;
 
 	switch (lkb->lkb_status) {
@@ -997,6 +1140,7 @@ static void revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
 		break;
 	case DLM_LKSTS_CONVERT:
 		move_lkb(r, lkb, DLM_LKSTS_GRANTED);
+		rv = 1;
 		break;
 	case DLM_LKSTS_WAITING:
 		del_lkb(r, lkb);
@@ -1004,15 +1148,17 @@ static void revert_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
 		/* this unhold undoes the original ref from create_lkb()
 		   so this leads to the lkb being freed */
 		unhold_lkb(lkb);
+		rv = -1;
 		break;
 	default:
 		log_print("invalid status for revert %d", lkb->lkb_status);
 	}
+	return rv;
 }
 
-static void revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
+static int revert_lock_pc(struct dlm_rsb *r, struct dlm_lkb *lkb)
 {
-	revert_lock(r, lkb);
+	return revert_lock(r, lkb);
 }
 
 static void _grant_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
@@ -1055,6 +1201,50 @@ static void grant_lock_pending(struct dlm_rsb *r, struct dlm_lkb *lkb)
 		queue_cast(r, lkb, 0);
 }
 
+/* The special CONVDEADLK, ALTPR and ALTCW flags allow the master to
+   change the granted/requested modes.  We're munging things accordingly in
+   the process copy.
+   CONVDEADLK: our grmode may have been forced down to NL to resolve a
+   conversion deadlock
+   ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
+   compatible with other granted locks */
+
+static void munge_demoted(struct dlm_lkb *lkb, struct dlm_message *ms)
+{
+	if (ms->m_type != DLM_MSG_CONVERT_REPLY) {
+		log_print("munge_demoted %x invalid reply type %d",
+			  lkb->lkb_id, ms->m_type);
+		return;
+	}
+
+	if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
+		log_print("munge_demoted %x invalid modes gr %d rq %d",
+			  lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
+		return;
+	}
+
+	lkb->lkb_grmode = DLM_LOCK_NL;
+}
+
+static void munge_altmode(struct dlm_lkb *lkb, struct dlm_message *ms)
+{
+	if (ms->m_type != DLM_MSG_REQUEST_REPLY &&
+	    ms->m_type != DLM_MSG_GRANT) {
+		log_print("munge_altmode %x invalid reply type %d",
+			  lkb->lkb_id, ms->m_type);
+		return;
+	}
+
+	if (lkb->lkb_exflags & DLM_LKF_ALTPR)
+		lkb->lkb_rqmode = DLM_LOCK_PR;
+	else if (lkb->lkb_exflags & DLM_LKF_ALTCW)
+		lkb->lkb_rqmode = DLM_LOCK_CW;
+	else {
+		log_print("munge_altmode invalid exflags %x", lkb->lkb_exflags);
+		dlm_print_lkb(lkb);
+	}
+}
+
 static inline int first_in_list(struct dlm_lkb *lkb, struct list_head *head)
 {
 	struct dlm_lkb *first = list_entry(head->next, struct dlm_lkb,
@@ -1499,7 +1689,7 @@ static void process_lookup_list(struct dlm_rsb *r)
 	struct dlm_lkb *lkb, *safe;
 
 	list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
-		list_del(&lkb->lkb_rsb_lookup);
+		list_del_init(&lkb->lkb_rsb_lookup);
 		_request_lock(r, lkb);
 		schedule();
 	}
@@ -1530,7 +1720,7 @@ static void confirm_master(struct dlm_rsb *r, int error)
 		if (!list_empty(&r->res_lookup)) {
 			lkb = list_entry(r->res_lookup.next, struct dlm_lkb,
 					 lkb_rsb_lookup);
-			list_del(&lkb->lkb_rsb_lookup);
+			list_del_init(&lkb->lkb_rsb_lookup);
 			r->res_first_lkid = lkb->lkb_id;
 			_request_lock(r, lkb);
 		} else
@@ -1614,6 +1804,9 @@ static int set_unlock_args(uint32_t flags, void *astarg, struct dlm_args *args)
  		      DLM_LKF_FORCEUNLOCK))
 		return -EINVAL;
 
+	if (flags & DLM_LKF_CANCEL && flags & DLM_LKF_FORCEUNLOCK)
+		return -EINVAL;
+
 	args->flags = flags;
 	args->astparam = (long) astarg;
 	return 0;
@@ -1638,6 +1831,9 @@ static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
 
 		if (lkb->lkb_wait_type)
 			goto out;
+
+		if (is_overlap(lkb))
+			goto out;
 	}
 
 	lkb->lkb_exflags = args->flags;
@@ -1654,35 +1850,126 @@ static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
 	return rv;
 }
 
+/* when dlm_unlock() sees -EBUSY with CANCEL/FORCEUNLOCK it returns 0
+   for success */
+
+/* note: it's valid for lkb_nodeid/res_nodeid to be -1 when we get here
+   because there may be a lookup in progress and it's valid to do
+   cancel/unlockf on it */
+
 static int validate_unlock_args(struct dlm_lkb *lkb, struct dlm_args *args)
 {
+	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
 	int rv = -EINVAL;
 
-	if (lkb->lkb_flags & DLM_IFL_MSTCPY)
+	if (lkb->lkb_flags & DLM_IFL_MSTCPY) {
+		log_error(ls, "unlock on MSTCPY %x", lkb->lkb_id);
+		dlm_print_lkb(lkb);
 		goto out;
+	}
 
-	if (args->flags & DLM_LKF_FORCEUNLOCK)
-		goto out_ok;
+	/* an lkb may still exist even though the lock is EOL'ed due to a
+	   cancel, unlock or failed noqueue request; an app can't use these
+	   locks; return same error as if the lkid had not been found at all */
 
-	if (args->flags & DLM_LKF_CANCEL &&
-	    lkb->lkb_status == DLM_LKSTS_GRANTED)
+	if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
+		log_debug(ls, "unlock on ENDOFLIFE %x", lkb->lkb_id);
+		rv = -ENOENT;
 		goto out;
+	}
 
-	if (!(args->flags & DLM_LKF_CANCEL) &&
-	    lkb->lkb_status != DLM_LKSTS_GRANTED)
-		goto out;
+	/* an lkb may be waiting for an rsb lookup to complete where the
+	   lookup was initiated by another lock */
+
+	if (args->flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)) {
+		if (!list_empty(&lkb->lkb_rsb_lookup)) {
+			log_debug(ls, "unlock on rsb_lookup %x", lkb->lkb_id);
+			list_del_init(&lkb->lkb_rsb_lookup);
+			queue_cast(lkb->lkb_resource, lkb,
+				   args->flags & DLM_LKF_CANCEL ?
+				   -DLM_ECANCEL : -DLM_EUNLOCK);
+			unhold_lkb(lkb); /* undoes create_lkb() */
+			rv = -EBUSY;
+			goto out;
+		}
+	}
+
+	/* cancel not allowed with another cancel/unlock in progress */
+
+	if (args->flags & DLM_LKF_CANCEL) {
+		if (lkb->lkb_exflags & DLM_LKF_CANCEL)
+			goto out;
+
+		if (is_overlap(lkb))
+			goto out;
+
+		if (lkb->lkb_flags & DLM_IFL_RESEND) {
+			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
+			rv = -EBUSY;
+			goto out;
+		}
+
+		switch (lkb->lkb_wait_type) {
+		case DLM_MSG_LOOKUP:
+		case DLM_MSG_REQUEST:
+			lkb->lkb_flags |= DLM_IFL_OVERLAP_CANCEL;
+			rv = -EBUSY;
+			goto out;
+		case DLM_MSG_UNLOCK:
+		case DLM_MSG_CANCEL:
+			goto out;
+		}
+		/* add_to_waiters() will set OVERLAP_CANCEL */
+		goto out_ok;
+	}
+
+	/* do we need to allow a force-unlock if there's a normal unlock
+	   already in progress?  in what conditions could the normal unlock
+	   fail such that we'd want to send a force-unlock to be sure? */
+
+	if (args->flags & DLM_LKF_FORCEUNLOCK) {
+		if (lkb->lkb_exflags & DLM_LKF_FORCEUNLOCK)
+			goto out;
+
+		if (is_overlap_unlock(lkb))
+			goto out;
 
+		if (lkb->lkb_flags & DLM_IFL_RESEND) {
+			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
+			rv = -EBUSY;
+			goto out;
+		}
+
+		switch (lkb->lkb_wait_type) {
+		case DLM_MSG_LOOKUP:
+		case DLM_MSG_REQUEST:
+			lkb->lkb_flags |= DLM_IFL_OVERLAP_UNLOCK;
+			rv = -EBUSY;
+			goto out;
+		case DLM_MSG_UNLOCK:
+			goto out;
+		}
+		/* add_to_waiters() will set OVERLAP_UNLOCK */
+		goto out_ok;
+	}
+
+	/* normal unlock not allowed if there's any op in progress */
 	rv = -EBUSY;
-	if (lkb->lkb_wait_type)
+	if (lkb->lkb_wait_type || lkb->lkb_wait_count)
 		goto out;
 
  out_ok:
-	lkb->lkb_exflags = args->flags;
+	/* an overlapping op shouldn't blow away exflags from other op */
+	lkb->lkb_exflags |= args->flags;
 	lkb->lkb_sbflags = 0;
 	lkb->lkb_astparam = args->astparam;
-
 	rv = 0;
  out:
+	if (rv)
+		log_debug(ls, "validate_unlock_args %d %x %x %x %x %d %s", rv,
+			  lkb->lkb_id, lkb->lkb_flags, lkb->lkb_exflags,
+			  args->flags, lkb->lkb_wait_type,
+			  lkb->lkb_resource->res_name);
 	return rv;
 }
 
@@ -1732,9 +2019,24 @@ static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
 		goto out;
 	}
 
-	if (can_be_queued(lkb)) {
-		if (is_demoted(lkb))
+	/* is_demoted() means the can_be_granted() above set the grmode
+	   to NL, and left us on the granted queue.  This auto-demotion
+	   (due to CONVDEADLK) might mean other locks, and/or this lock, are
+	   now grantable.  We have to try to grant other converting locks
+	   before we try again to grant this one. */
+
+	if (is_demoted(lkb)) {
+		grant_pending_convert(r, DLM_LOCK_IV);
+		if (_can_be_granted(r, lkb, 1)) {
+			grant_lock(r, lkb);
+			queue_cast(r, lkb, 0);
 			grant_pending_locks(r);
+			goto out;
+		}
+		/* else fall through and move to convert queue */
+	}
+
+	if (can_be_queued(lkb)) {
 		error = -EINPROGRESS;
 		del_lkb(r, lkb);
 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
@@ -1759,17 +2061,19 @@ static int do_unlock(struct dlm_rsb *r, struct dlm_lkb *lkb)
 	return -DLM_EUNLOCK;
 }
 
-/* FIXME: if revert_lock() finds that the lkb is granted, we should
-   skip the queue_cast(ECANCEL).  It indicates that the request/convert
-   completed (and queued a normal ast) just before the cancel; we don't
-   want to clobber the sb_result for the normal ast with ECANCEL. */
+/* returns: 0 did nothing, -DLM_ECANCEL canceled lock */
  
 static int do_cancel(struct dlm_rsb *r, struct dlm_lkb *lkb)
 {
-	revert_lock(r, lkb);
-	queue_cast(r, lkb, -DLM_ECANCEL);
-	grant_pending_locks(r);
-	return -DLM_ECANCEL;
+	int error;
+
+	error = revert_lock(r, lkb);
+	if (error) {
+		queue_cast(r, lkb, -DLM_ECANCEL);
+		grant_pending_locks(r);
+		return -DLM_ECANCEL;
+	}
+	return 0;
 }
 
 /*
@@ -2035,6 +2339,8 @@ int dlm_unlock(dlm_lockspace_t *lockspace,
 
 	if (error == -DLM_EUNLOCK || error == -DLM_ECANCEL)
 		error = 0;
+	if (error == -EBUSY && (flags & (DLM_LKF_CANCEL | DLM_LKF_FORCEUNLOCK)))
+		error = 0;
  out_put:
 	dlm_put_lkb(lkb);
  out:
@@ -2065,31 +2371,14 @@ int dlm_unlock(dlm_lockspace_t *lockspace,
  * receive_lookup_reply		send_lookup_reply
  */
 
-static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
-			  int to_nodeid, int mstype,
-			  struct dlm_message **ms_ret,
-			  struct dlm_mhandle **mh_ret)
+static int _create_message(struct dlm_ls *ls, int mb_len,
+			   int to_nodeid, int mstype,
+			   struct dlm_message **ms_ret,
+			   struct dlm_mhandle **mh_ret)
 {
 	struct dlm_message *ms;
 	struct dlm_mhandle *mh;
 	char *mb;
-	int mb_len = sizeof(struct dlm_message);
-
-	switch (mstype) {
-	case DLM_MSG_REQUEST:
-	case DLM_MSG_LOOKUP:
-	case DLM_MSG_REMOVE:
-		mb_len += r->res_length;
-		break;
-	case DLM_MSG_CONVERT:
-	case DLM_MSG_UNLOCK:
-	case DLM_MSG_REQUEST_REPLY:
-	case DLM_MSG_CONVERT_REPLY:
-	case DLM_MSG_GRANT:
-		if (lkb && lkb->lkb_lvbptr)
-			mb_len += r->res_ls->ls_lvblen;
-		break;
-	}
 
 	/* get_buffer gives us a message handle (mh) that we need to
 	   pass into lowcomms_commit and a message buffer (mb) that we
@@ -2104,7 +2393,7 @@ static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
 	ms = (struct dlm_message *) mb;
 
 	ms->m_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
-	ms->m_header.h_lockspace = r->res_ls->ls_global_id;
+	ms->m_header.h_lockspace = ls->ls_global_id;
 	ms->m_header.h_nodeid = dlm_our_nodeid();
 	ms->m_header.h_length = mb_len;
 	ms->m_header.h_cmd = DLM_MSG;
@@ -2116,6 +2405,33 @@ static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
 	return 0;
 }
 
+static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
+			  int to_nodeid, int mstype,
+			  struct dlm_message **ms_ret,
+			  struct dlm_mhandle **mh_ret)
+{
+	int mb_len = sizeof(struct dlm_message);
+
+	switch (mstype) {
+	case DLM_MSG_REQUEST:
+	case DLM_MSG_LOOKUP:
+	case DLM_MSG_REMOVE:
+		mb_len += r->res_length;
+		break;
+	case DLM_MSG_CONVERT:
+	case DLM_MSG_UNLOCK:
+	case DLM_MSG_REQUEST_REPLY:
+	case DLM_MSG_CONVERT_REPLY:
+	case DLM_MSG_GRANT:
+		if (lkb && lkb->lkb_lvbptr)
+			mb_len += r->res_ls->ls_lvblen;
+		break;
+	}
+
+	return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
+			       ms_ret, mh_ret);
+}
+
 /* further lowcomms enhancements or alternate implementations may make
    the return value from this function useful at some point */
 
@@ -2176,7 +2492,9 @@ static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
 	struct dlm_mhandle *mh;
 	int to_nodeid, error;
 
-	add_to_waiters(lkb, mstype);
+	error = add_to_waiters(lkb, mstype);
+	if (error)
+		return error;
 
 	to_nodeid = r->res_nodeid;
 
@@ -2192,7 +2510,7 @@ static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
 	return 0;
 
  fail:
-	remove_from_waiters(lkb);
+	remove_from_waiters(lkb, msg_reply_type(mstype));
 	return error;
 }
 
@@ -2209,7 +2527,8 @@ static int send_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
 
 	/* down conversions go without a reply from the master */
 	if (!error && down_conversion(lkb)) {
-		remove_from_waiters(lkb);
+		remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
+		r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
 		r->res_ls->ls_stub_ms.m_result = 0;
 		r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
 		__receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
@@ -2280,7 +2599,9 @@ static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
 	struct dlm_mhandle *mh;
 	int to_nodeid, error;
 
-	add_to_waiters(lkb, DLM_MSG_LOOKUP);
+	error = add_to_waiters(lkb, DLM_MSG_LOOKUP);
+	if (error)
+		return error;
 
 	to_nodeid = dlm_dir_nodeid(r);
 
@@ -2296,7 +2617,7 @@ static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
 	return 0;
 
  fail:
-	remove_from_waiters(lkb);
+	remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
 	return error;
 }
 
@@ -2656,6 +2977,8 @@ static void receive_grant(struct dlm_ls *ls, struct dlm_message *ms)
 	lock_rsb(r);
 
 	receive_flags_reply(lkb, ms);
+	if (is_altmode(lkb))
+		munge_altmode(lkb, ms);
 	grant_lock_pc(r, lkb, ms);
 	queue_cast(r, lkb, 0);
 
@@ -2736,11 +3059,16 @@ static void receive_remove(struct dlm_ls *ls, struct dlm_message *ms)
 	dlm_dir_remove_entry(ls, from_nodeid, ms->m_extra, len);
 }
 
+static void receive_purge(struct dlm_ls *ls, struct dlm_message *ms)
+{
+	do_purge(ls, ms->m_nodeid, ms->m_pid);
+}
+
 static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
 {
 	struct dlm_lkb *lkb;
 	struct dlm_rsb *r;
-	int error, mstype;
+	int error, mstype, result;
 
 	error = find_lkb(ls, ms->m_remid, &lkb);
 	if (error) {
@@ -2749,20 +3077,15 @@ static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
 	}
 	DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
 
-	mstype = lkb->lkb_wait_type;
-	error = remove_from_waiters(lkb);
-	if (error) {
-		log_error(ls, "receive_request_reply not on waiters");
-		goto out;
-	}
-
-	/* this is the value returned from do_request() on the master */
-	error = ms->m_result;
-
 	r = lkb->lkb_resource;
 	hold_rsb(r);
 	lock_rsb(r);
 
+	mstype = lkb->lkb_wait_type;
+	error = remove_from_waiters(lkb, DLM_MSG_REQUEST_REPLY);
+	if (error)
+		goto out;
+
 	/* Optimization: the dir node was also the master, so it took our
 	   lookup as a request and sent request reply instead of lookup reply */
 	if (mstype == DLM_MSG_LOOKUP) {
@@ -2770,14 +3093,15 @@ static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
 		lkb->lkb_nodeid = r->res_nodeid;
 	}
 
-	switch (error) {
+	/* this is the value returned from do_request() on the master */
+	result = ms->m_result;
+
+	switch (result) {
 	case -EAGAIN:
-		/* request would block (be queued) on remote master;
-		   the unhold undoes the original ref from create_lkb()
-		   so it leads to the lkb being freed */
+		/* request would block (be queued) on remote master */
 		queue_cast(r, lkb, -EAGAIN);
 		confirm_master(r, -EAGAIN);
-		unhold_lkb(lkb);
+		unhold_lkb(lkb); /* undoes create_lkb() */
 		break;
 
 	case -EINPROGRESS:
@@ -2785,41 +3109,64 @@ static void receive_request_reply(struct dlm_ls *ls, struct dlm_message *ms)
 		/* request was queued or granted on remote master */
 		receive_flags_reply(lkb, ms);
 		lkb->lkb_remid = ms->m_lkid;
-		if (error)
+		if (is_altmode(lkb))
+			munge_altmode(lkb, ms);
+		if (result)
 			add_lkb(r, lkb, DLM_LKSTS_WAITING);
 		else {
 			grant_lock_pc(r, lkb, ms);
 			queue_cast(r, lkb, 0);
 		}
-		confirm_master(r, error);
+		confirm_master(r, result);
 		break;
 
 	case -EBADR:
 	case -ENOTBLK:
 		/* find_rsb failed to find rsb or rsb wasn't master */
+		log_debug(ls, "receive_request_reply %x %x master diff %d %d",
+			  lkb->lkb_id, lkb->lkb_flags, r->res_nodeid, result);
 		r->res_nodeid = -1;
 		lkb->lkb_nodeid = -1;
-		_request_lock(r, lkb);
+
+		if (is_overlap(lkb)) {
+			/* we'll ignore error in cancel/unlock reply */
+			queue_cast_overlap(r, lkb);
+			unhold_lkb(lkb); /* undoes create_lkb() */
+		} else
+			_request_lock(r, lkb);
 		break;
 
 	default:
-		log_error(ls, "receive_request_reply error %d", error);
+		log_error(ls, "receive_request_reply %x error %d",
+			  lkb->lkb_id, result);
 	}
 
+	if (is_overlap_unlock(lkb) && (result == 0 || result == -EINPROGRESS)) {
+		log_debug(ls, "receive_request_reply %x result %d unlock",
+			  lkb->lkb_id, result);
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
+		send_unlock(r, lkb);
+	} else if (is_overlap_cancel(lkb) && (result == -EINPROGRESS)) {
+		log_debug(ls, "receive_request_reply %x cancel", lkb->lkb_id);
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
+		send_cancel(r, lkb);
+	} else {
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
+	}
+ out:
 	unlock_rsb(r);
 	put_rsb(r);
- out:
 	dlm_put_lkb(lkb);
 }
 
 static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
 				    struct dlm_message *ms)
 {
-	int error = ms->m_result;
-
 	/* this is the value returned from do_convert() on the master */
-
-	switch (error) {
+	switch (ms->m_result) {
 	case -EAGAIN:
 		/* convert would block (be queued) on remote master */
 		queue_cast(r, lkb, -EAGAIN);
@@ -2827,6 +3174,9 @@ static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
 
 	case -EINPROGRESS:
 		/* convert was queued on remote master */
+		receive_flags_reply(lkb, ms);
+		if (is_demoted(lkb))
+			munge_demoted(lkb, ms);
 		del_lkb(r, lkb);
 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
 		break;
@@ -2834,24 +3184,33 @@ static void __receive_convert_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
 	case 0:
 		/* convert was granted on remote master */
 		receive_flags_reply(lkb, ms);
+		if (is_demoted(lkb))
+			munge_demoted(lkb, ms);
 		grant_lock_pc(r, lkb, ms);
 		queue_cast(r, lkb, 0);
 		break;
 
 	default:
-		log_error(r->res_ls, "receive_convert_reply error %d", error);
+		log_error(r->res_ls, "receive_convert_reply %x error %d",
+			  lkb->lkb_id, ms->m_result);
 	}
 }
 
 static void _receive_convert_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
 {
 	struct dlm_rsb *r = lkb->lkb_resource;
+	int error;
 
 	hold_rsb(r);
 	lock_rsb(r);
 
-	__receive_convert_reply(r, lkb, ms);
+	/* stub reply can happen with waiters_mutex held */
+	error = remove_from_waiters_ms(lkb, ms);
+	if (error)
+		goto out;
 
+	__receive_convert_reply(r, lkb, ms);
+ out:
 	unlock_rsb(r);
 	put_rsb(r);
 }
@@ -2868,37 +3227,38 @@ static void receive_convert_reply(struct dlm_ls *ls, struct dlm_message *ms)
 	}
 	DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
 
-	error = remove_from_waiters(lkb);
-	if (error) {
-		log_error(ls, "receive_convert_reply not on waiters");
-		goto out;
-	}
-
 	_receive_convert_reply(lkb, ms);
- out:
 	dlm_put_lkb(lkb);
 }
 
 static void _receive_unlock_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
 {
 	struct dlm_rsb *r = lkb->lkb_resource;
-	int error = ms->m_result;
+	int error;
 
 	hold_rsb(r);
 	lock_rsb(r);
 
+	/* stub reply can happen with waiters_mutex held */
+	error = remove_from_waiters_ms(lkb, ms);
+	if (error)
+		goto out;
+
 	/* this is the value returned from do_unlock() on the master */
 
-	switch (error) {
+	switch (ms->m_result) {
 	case -DLM_EUNLOCK:
 		receive_flags_reply(lkb, ms);
 		remove_lock_pc(r, lkb);
 		queue_cast(r, lkb, -DLM_EUNLOCK);
 		break;
+	case -ENOENT:
+		break;
 	default:
-		log_error(r->res_ls, "receive_unlock_reply error %d", error);
+		log_error(r->res_ls, "receive_unlock_reply %x error %d",
+			  lkb->lkb_id, ms->m_result);
 	}
-
+ out:
 	unlock_rsb(r);
 	put_rsb(r);
 }
@@ -2915,37 +3275,39 @@ static void receive_unlock_reply(struct dlm_ls *ls, struct dlm_message *ms)
 	}
 	DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
 
-	error = remove_from_waiters(lkb);
-	if (error) {
-		log_error(ls, "receive_unlock_reply not on waiters");
-		goto out;
-	}
-
 	_receive_unlock_reply(lkb, ms);
- out:
 	dlm_put_lkb(lkb);
 }
 
 static void _receive_cancel_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
 {
 	struct dlm_rsb *r = lkb->lkb_resource;
-	int error = ms->m_result;
+	int error;
 
 	hold_rsb(r);
 	lock_rsb(r);
 
+	/* stub reply can happen with waiters_mutex held */
+	error = remove_from_waiters_ms(lkb, ms);
+	if (error)
+		goto out;
+
 	/* this is the value returned from do_cancel() on the master */
 
-	switch (error) {
+	switch (ms->m_result) {
 	case -DLM_ECANCEL:
 		receive_flags_reply(lkb, ms);
 		revert_lock_pc(r, lkb);
-		queue_cast(r, lkb, -DLM_ECANCEL);
+		if (ms->m_result)
+			queue_cast(r, lkb, -DLM_ECANCEL);
+		break;
+	case 0:
 		break;
 	default:
-		log_error(r->res_ls, "receive_cancel_reply error %d", error);
+		log_error(r->res_ls, "receive_cancel_reply %x error %d",
+			  lkb->lkb_id, ms->m_result);
 	}
-
+ out:
 	unlock_rsb(r);
 	put_rsb(r);
 }
@@ -2962,14 +3324,7 @@ static void receive_cancel_reply(struct dlm_ls *ls, struct dlm_message *ms)
 	}
 	DLM_ASSERT(is_process_copy(lkb), dlm_print_lkb(lkb););
 
-	error = remove_from_waiters(lkb);
-	if (error) {
-		log_error(ls, "receive_cancel_reply not on waiters");
-		goto out;
-	}
-
 	_receive_cancel_reply(lkb, ms);
- out:
 	dlm_put_lkb(lkb);
 }
 
@@ -2985,20 +3340,17 @@ static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
 		return;
 	}
 
-	error = remove_from_waiters(lkb);
-	if (error) {
-		log_error(ls, "receive_lookup_reply not on waiters");
-		goto out;
-	}
-
-	/* this is the value returned by dlm_dir_lookup on dir node
+	/* ms->m_result is the value returned by dlm_dir_lookup on dir node
 	   FIXME: will a non-zero error ever be returned? */
-	error = ms->m_result;
 
 	r = lkb->lkb_resource;
 	hold_rsb(r);
 	lock_rsb(r);
 
+	error = remove_from_waiters(lkb, DLM_MSG_LOOKUP_REPLY);
+	if (error)
+		goto out;
+
 	ret_nodeid = ms->m_nodeid;
 	if (ret_nodeid == dlm_our_nodeid()) {
 		r->res_nodeid = 0;
@@ -3009,14 +3361,22 @@ static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
 		r->res_nodeid = ret_nodeid;
 	}
 
+	if (is_overlap(lkb)) {
+		log_debug(ls, "receive_lookup_reply %x unlock %x",
+			  lkb->lkb_id, lkb->lkb_flags);
+		queue_cast_overlap(r, lkb);
+		unhold_lkb(lkb); /* undoes create_lkb() */
+		goto out_list;
+	}
+
 	_request_lock(r, lkb);
 
+ out_list:
 	if (!ret_nodeid)
 		process_lookup_list(r);
-
+ out:
 	unlock_rsb(r);
 	put_rsb(r);
- out:
 	dlm_put_lkb(lkb);
 }
 
@@ -3133,6 +3493,12 @@ int dlm_receive_message(struct dlm_header *hd, int nodeid, int recovery)
 		receive_lookup_reply(ls, ms);
 		break;
 
+	/* other messages */
+
+	case DLM_MSG_PURGE:
+		receive_purge(ls, ms);
+		break;
+
 	default:
 		log_error(ls, "unknown message type %d", ms->m_type);
 	}
@@ -3153,9 +3519,9 @@ static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
 {
 	if (middle_conversion(lkb)) {
 		hold_lkb(lkb);
+		ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
 		ls->ls_stub_ms.m_result = -EINPROGRESS;
 		ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-		_remove_from_waiters(lkb);
 		_receive_convert_reply(lkb, &ls->ls_stub_ms);
 
 		/* Same special case as in receive_rcom_lock_args() */
@@ -3227,18 +3593,18 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls)
 
 		case DLM_MSG_UNLOCK:
 			hold_lkb(lkb);
+			ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
 			ls->ls_stub_ms.m_result = -DLM_EUNLOCK;
 			ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-			_remove_from_waiters(lkb);
 			_receive_unlock_reply(lkb, &ls->ls_stub_ms);
 			dlm_put_lkb(lkb);
 			break;
 
 		case DLM_MSG_CANCEL:
 			hold_lkb(lkb);
+			ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
 			ls->ls_stub_ms.m_result = -DLM_ECANCEL;
 			ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-			_remove_from_waiters(lkb);
 			_receive_cancel_reply(lkb, &ls->ls_stub_ms);
 			dlm_put_lkb(lkb);
 			break;
@@ -3252,37 +3618,47 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls)
 	mutex_unlock(&ls->ls_waiters_mutex);
 }
 
-static int remove_resend_waiter(struct dlm_ls *ls, struct dlm_lkb **lkb_ret)
+static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
 {
 	struct dlm_lkb *lkb;
-	int rv = 0;
+	int found = 0;
 
 	mutex_lock(&ls->ls_waiters_mutex);
 	list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
 		if (lkb->lkb_flags & DLM_IFL_RESEND) {
-			rv = lkb->lkb_wait_type;
-			_remove_from_waiters(lkb);
-			lkb->lkb_flags &= ~DLM_IFL_RESEND;
+			hold_lkb(lkb);
+			found = 1;
 			break;
 		}
 	}
 	mutex_unlock(&ls->ls_waiters_mutex);
 
-	if (!rv)
+	if (!found)
 		lkb = NULL;
-	*lkb_ret = lkb;
-	return rv;
+	return lkb;
 }
 
 /* Deal with lookups and lkb's marked RESEND from _pre.  We may now be the
    master or dir-node for r.  Processing the lkb may result in it being placed
    back on waiters. */
 
+/* We do this after normal locking has been enabled and any saved messages
+   (in requestqueue) have been processed.  We should be confident that at
+   this point we won't get or process a reply to any of these waiting
+   operations.  But, new ops may be coming in on the rsbs/locks here from
+   userspace or remotely. */
+
+/* there may have been an overlap unlock/cancel prior to recovery or after
+   recovery.  if before, the lkb may still have a pos wait_count; if after, the
+   overlap flag would just have been set and nothing new sent.  we can be
+   confident here than any replies to either the initial op or overlap ops
+   prior to recovery have been received. */
+
 int dlm_recover_waiters_post(struct dlm_ls *ls)
 {
 	struct dlm_lkb *lkb;
 	struct dlm_rsb *r;
-	int error = 0, mstype;
+	int error = 0, mstype, err, oc, ou;
 
 	while (1) {
 		if (dlm_locking_stopped(ls)) {
@@ -3291,48 +3667,78 @@ int dlm_recover_waiters_post(struct dlm_ls *ls)
 			break;
 		}
 
-		mstype = remove_resend_waiter(ls, &lkb);
-		if (!mstype)
+		lkb = find_resend_waiter(ls);
+		if (!lkb)
 			break;
 
 		r = lkb->lkb_resource;
+		hold_rsb(r);
+		lock_rsb(r);
+
+		mstype = lkb->lkb_wait_type;
+		oc = is_overlap_cancel(lkb);
+		ou = is_overlap_unlock(lkb);
+		err = 0;
 
 		log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
 			  lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
 
-		switch (mstype) {
-
-		case DLM_MSG_LOOKUP:
-			hold_rsb(r);
-			lock_rsb(r);
-			_request_lock(r, lkb);
-			if (is_master(r))
-				confirm_master(r, 0);
-			unlock_rsb(r);
-			put_rsb(r);
-			break;
-
-		case DLM_MSG_REQUEST:
-			hold_rsb(r);
-			lock_rsb(r);
-			_request_lock(r, lkb);
-			if (is_master(r))
-				confirm_master(r, 0);
-			unlock_rsb(r);
-			put_rsb(r);
-			break;
-
-		case DLM_MSG_CONVERT:
-			hold_rsb(r);
-			lock_rsb(r);
-			_convert_lock(r, lkb);
-			unlock_rsb(r);
-			put_rsb(r);
-			break;
-
-		default:
-			log_error(ls, "recover_waiters_post type %d", mstype);
+		/* At this point we assume that we won't get a reply to any
+		   previous op or overlap op on this lock.  First, do a big
+		   remove_from_waiters() for all previous ops. */
+
+		lkb->lkb_flags &= ~DLM_IFL_RESEND;
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
+		lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
+		lkb->lkb_wait_type = 0;
+		lkb->lkb_wait_count = 0;
+		mutex_lock(&ls->ls_waiters_mutex);
+		list_del_init(&lkb->lkb_wait_reply);
+		mutex_unlock(&ls->ls_waiters_mutex);
+		unhold_lkb(lkb); /* for waiters list */
+
+		if (oc || ou) {
+			/* do an unlock or cancel instead of resending */
+			switch (mstype) {
+			case DLM_MSG_LOOKUP:
+			case DLM_MSG_REQUEST:
+				queue_cast(r, lkb, ou ? -DLM_EUNLOCK :
+							-DLM_ECANCEL);
+				unhold_lkb(lkb); /* undoes create_lkb() */
+				break;
+			case DLM_MSG_CONVERT:
+				if (oc) {
+					queue_cast(r, lkb, -DLM_ECANCEL);
+				} else {
+					lkb->lkb_exflags |= DLM_LKF_FORCEUNLOCK;
+					_unlock_lock(r, lkb);
+				}
+				break;
+			default:
+				err = 1;
+			}
+		} else {
+			switch (mstype) {
+			case DLM_MSG_LOOKUP:
+			case DLM_MSG_REQUEST:
+				_request_lock(r, lkb);
+				if (is_master(r))
+					confirm_master(r, 0);
+				break;
+			case DLM_MSG_CONVERT:
+				_convert_lock(r, lkb);
+				break;
+			default:
+				err = 1;
+			}
 		}
+
+		if (err)
+			log_error(ls, "recover_waiters_post %x %d %x %d %d",
+			  	  lkb->lkb_id, mstype, lkb->lkb_flags, oc, ou);
+		unlock_rsb(r);
+		put_rsb(r);
+		dlm_put_lkb(lkb);
 	}
 
 	return error;
@@ -3684,7 +4090,7 @@ int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
 
 	/* add this new lkb to the per-process list of locks */
 	spin_lock(&ua->proc->locks_spin);
-	kref_get(&lkb->lkb_ref);
+	hold_lkb(lkb);
 	list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
 	spin_unlock(&ua->proc->locks_spin);
  out:
@@ -3774,6 +4180,9 @@ int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 
 	if (error == -DLM_EUNLOCK)
 		error = 0;
+	/* from validate_unlock_args() */
+	if (error == -EBUSY && (flags & DLM_LKF_FORCEUNLOCK))
+		error = 0;
 	if (error)
 		goto out_put;
 
@@ -3786,6 +4195,7 @@ int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 	dlm_put_lkb(lkb);
  out:
 	unlock_recovery(ls);
+	kfree(ua_tmp);
 	return error;
 }
 
@@ -3815,33 +4225,37 @@ int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 
 	if (error == -DLM_ECANCEL)
 		error = 0;
-	if (error)
-		goto out_put;
-
-	/* this lkb was removed from the WAITING queue */
-	if (lkb->lkb_grmode == DLM_LOCK_IV) {
-		spin_lock(&ua->proc->locks_spin);
-		list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
-		spin_unlock(&ua->proc->locks_spin);
-	}
+	/* from validate_unlock_args() */
+	if (error == -EBUSY)
+		error = 0;
  out_put:
 	dlm_put_lkb(lkb);
  out:
 	unlock_recovery(ls);
+	kfree(ua_tmp);
 	return error;
 }
 
+/* lkb's that are removed from the waiters list by revert are just left on the
+   orphans list with the granted orphan locks, to be freed by purge */
+
 static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
 {
 	struct dlm_user_args *ua = (struct dlm_user_args *)lkb->lkb_astparam;
+	struct dlm_args args;
+	int error;
 
-	if (ua->lksb.sb_lvbptr)
-		kfree(ua->lksb.sb_lvbptr);
-	kfree(ua);
-	lkb->lkb_astparam = (long)NULL;
+	hold_lkb(lkb);
+	mutex_lock(&ls->ls_orphans_mutex);
+	list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
+	mutex_unlock(&ls->ls_orphans_mutex);
 
-	/* TODO: propogate to master if needed */
-	return 0;
+	set_unlock_args(0, ua, &args);
+
+	error = cancel_lock(ls, lkb, &args);
+	if (error == -DLM_ECANCEL)
+		error = 0;
+	return error;
 }
 
 /* The force flag allows the unlock to go ahead even if the lkb isn't granted.
@@ -3853,10 +4267,6 @@ static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
 	struct dlm_args args;
 	int error;
 
-	/* FIXME: we need to handle the case where the lkb is in limbo
-	   while the rsb is being looked up, currently we assert in
-	   _unlock_lock/is_remote because rsb nodeid is -1. */
-
 	set_unlock_args(DLM_LKF_FORCEUNLOCK, ua, &args);
 
 	error = unlock_lock(ls, lkb, &args);
@@ -3865,6 +4275,31 @@ static int unlock_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
 	return error;
 }
 
+/* We have to release clear_proc_locks mutex before calling unlock_proc_lock()
+   (which does lock_rsb) due to deadlock with receiving a message that does
+   lock_rsb followed by dlm_user_add_ast() */
+
+static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
+				     struct dlm_user_proc *proc)
+{
+	struct dlm_lkb *lkb = NULL;
+
+	mutex_lock(&ls->ls_clear_proc_locks);
+	if (list_empty(&proc->locks))
+		goto out;
+
+	lkb = list_entry(proc->locks.next, struct dlm_lkb, lkb_ownqueue);
+	list_del_init(&lkb->lkb_ownqueue);
+
+	if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
+		lkb->lkb_flags |= DLM_IFL_ORPHAN;
+	else
+		lkb->lkb_flags |= DLM_IFL_DEAD;
+ out:
+	mutex_unlock(&ls->ls_clear_proc_locks);
+	return lkb;
+}
+
 /* The ls_clear_proc_locks mutex protects against dlm_user_add_asts() which
    1) references lkb->ua which we free here and 2) adds lkbs to proc->asts,
    which we clear here. */
@@ -3880,18 +4315,15 @@ void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 	struct dlm_lkb *lkb, *safe;
 
 	lock_recovery(ls);
-	mutex_lock(&ls->ls_clear_proc_locks);
 
-	list_for_each_entry_safe(lkb, safe, &proc->locks, lkb_ownqueue) {
-		list_del_init(&lkb->lkb_ownqueue);
-
-		if (lkb->lkb_exflags & DLM_LKF_PERSISTENT) {
-			lkb->lkb_flags |= DLM_IFL_ORPHAN;
+	while (1) {
+		lkb = del_proc_lock(ls, proc);
+		if (!lkb)
+			break;
+		if (lkb->lkb_exflags & DLM_LKF_PERSISTENT)
 			orphan_proc_lock(ls, lkb);
-		} else {
-			lkb->lkb_flags |= DLM_IFL_DEAD;
+		else
 			unlock_proc_lock(ls, lkb);
-		}
 
 		/* this removes the reference for the proc->locks list
 		   added by dlm_user_request, it may result in the lkb
@@ -3900,6 +4332,8 @@ void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 		dlm_put_lkb(lkb);
 	}
 
+	mutex_lock(&ls->ls_clear_proc_locks);
+
 	/* in-progress unlocks */
 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
 		list_del_init(&lkb->lkb_ownqueue);
@@ -3916,3 +4350,92 @@ void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 	unlock_recovery(ls);
 }
 
+static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
+{
+	struct dlm_lkb *lkb, *safe;
+
+	while (1) {
+		lkb = NULL;
+		spin_lock(&proc->locks_spin);
+		if (!list_empty(&proc->locks)) {
+			lkb = list_entry(proc->locks.next, struct dlm_lkb,
+					 lkb_ownqueue);
+			list_del_init(&lkb->lkb_ownqueue);
+		}
+		spin_unlock(&proc->locks_spin);
+
+		if (!lkb)
+			break;
+
+		lkb->lkb_flags |= DLM_IFL_DEAD;
+		unlock_proc_lock(ls, lkb);
+		dlm_put_lkb(lkb); /* ref from proc->locks list */
+	}
+
+	spin_lock(&proc->locks_spin);
+	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
+		list_del_init(&lkb->lkb_ownqueue);
+		lkb->lkb_flags |= DLM_IFL_DEAD;
+		dlm_put_lkb(lkb);
+	}
+	spin_unlock(&proc->locks_spin);
+
+	spin_lock(&proc->asts_spin);
+	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_astqueue) {
+		list_del(&lkb->lkb_astqueue);
+		dlm_put_lkb(lkb);
+	}
+	spin_unlock(&proc->asts_spin);
+}
+
+/* pid of 0 means purge all orphans */
+
+static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
+{
+	struct dlm_lkb *lkb, *safe;
+
+	mutex_lock(&ls->ls_orphans_mutex);
+	list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
+		if (pid && lkb->lkb_ownpid != pid)
+			continue;
+		unlock_proc_lock(ls, lkb);
+		list_del_init(&lkb->lkb_ownqueue);
+		dlm_put_lkb(lkb);
+	}
+	mutex_unlock(&ls->ls_orphans_mutex);
+}
+
+static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
+{
+	struct dlm_message *ms;
+	struct dlm_mhandle *mh;
+	int error;
+
+	error = _create_message(ls, sizeof(struct dlm_message), nodeid,
+				DLM_MSG_PURGE, &ms, &mh);
+	if (error)
+		return error;
+	ms->m_nodeid = nodeid;
+	ms->m_pid = pid;
+
+	return send_message(mh, ms);
+}
+
+int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
+		   int nodeid, int pid)
+{
+	int error = 0;
+
+	if (nodeid != dlm_our_nodeid()) {
+		error = send_purge(ls, nodeid, pid);
+	} else {
+		lock_recovery(ls);
+		if (pid == current->pid)
+			purge_proc_locks(ls, proc);
+		else
+			do_purge(ls, nodeid, pid);
+		unlock_recovery(ls);
+	}
+	return error;
+}
+
diff --git a/fs/dlm/lock.h b/fs/dlm/lock.h
index 0843a3073ec..64fc4ec4066 100644
--- a/fs/dlm/lock.h
+++ b/fs/dlm/lock.h
@@ -41,6 +41,8 @@ int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 	uint32_t flags, uint32_t lkid, char *lvb_in);
 int dlm_user_cancel(struct dlm_ls *ls,  struct dlm_user_args *ua_tmp,
 	uint32_t flags, uint32_t lkid);
+int dlm_user_purge(struct dlm_ls *ls, struct dlm_user_proc *proc,
+	int nodeid, int pid);
 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc);
 
 static inline int is_master(struct dlm_rsb *r)
diff --git a/fs/dlm/lockspace.c b/fs/dlm/lockspace.c
index 8e6ad760023..a677b2a5eed 100644
--- a/fs/dlm/lockspace.c
+++ b/fs/dlm/lockspace.c
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -459,6 +459,8 @@ static int new_lockspace(char *name, int namelen, void **lockspace,
 
 	INIT_LIST_HEAD(&ls->ls_waiters);
 	mutex_init(&ls->ls_waiters_mutex);
+	INIT_LIST_HEAD(&ls->ls_orphans);
+	mutex_init(&ls->ls_orphans_mutex);
 
 	INIT_LIST_HEAD(&ls->ls_nodes);
 	INIT_LIST_HEAD(&ls->ls_nodes_gone);
diff --git a/fs/dlm/lowcomms-sctp.c b/fs/dlm/lowcomms-sctp.c
deleted file mode 100644
index dc83a9d979b..00000000000
--- a/fs/dlm/lowcomms-sctp.c
+++ /dev/null
@@ -1,1210 +0,0 @@
-/******************************************************************************
-*******************************************************************************
-**
-**  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
-**
-**  This copyrighted material is made available to anyone wishing to use,
-**  modify, copy, or redistribute it subject to the terms and conditions
-**  of the GNU General Public License v.2.
-**
-*******************************************************************************
-******************************************************************************/
-
-/*
- * lowcomms.c
- *
- * This is the "low-level" comms layer.
- *
- * It is responsible for sending/receiving messages
- * from other nodes in the cluster.
- *
- * Cluster nodes are referred to by their nodeids. nodeids are
- * simply 32 bit numbers to the locking module - if they need to
- * be expanded for the cluster infrastructure then that is it's
- * responsibility. It is this layer's
- * responsibility to resolve these into IP address or
- * whatever it needs for inter-node communication.
- *
- * The comms level is two kernel threads that deal mainly with
- * the receiving of messages from other nodes and passing them
- * up to the mid-level comms layer (which understands the
- * message format) for execution by the locking core, and
- * a send thread which does all the setting up of connections
- * to remote nodes and the sending of data. Threads are not allowed
- * to send their own data because it may cause them to wait in times
- * of high load. Also, this way, the sending thread can collect together
- * messages bound for one node and send them in one block.
- *
- * I don't see any problem with the recv thread executing the locking
- * code on behalf of remote processes as the locking code is
- * short, efficient and never (well, hardly ever) waits.
- *
- */
-
-#include <asm/ioctls.h>
-#include <net/sock.h>
-#include <net/tcp.h>
-#include <net/sctp/user.h>
-#include <linux/pagemap.h>
-#include <linux/socket.h>
-#include <linux/idr.h>
-
-#include "dlm_internal.h"
-#include "lowcomms.h"
-#include "config.h"
-#include "midcomms.h"
-
-static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
-static int			dlm_local_count;
-static int			dlm_local_nodeid;
-
-/* One of these per connected node */
-
-#define NI_INIT_PENDING 1
-#define NI_WRITE_PENDING 2
-
-struct nodeinfo {
-	spinlock_t		lock;
-	sctp_assoc_t		assoc_id;
-	unsigned long		flags;
-	struct list_head	write_list; /* nodes with pending writes */
-	struct list_head	writequeue; /* outgoing writequeue_entries */
-	spinlock_t		writequeue_lock;
-	int			nodeid;
-	struct work_struct      swork; /* Send workqueue */
-	struct work_struct      lwork; /* Locking workqueue */
-};
-
-static DEFINE_IDR(nodeinfo_idr);
-static DECLARE_RWSEM(nodeinfo_lock);
-static int max_nodeid;
-
-struct cbuf {
-	unsigned int base;
-	unsigned int len;
-	unsigned int mask;
-};
-
-/* Just the one of these, now. But this struct keeps
-   the connection-specific variables together */
-
-#define CF_READ_PENDING 1
-
-struct connection {
-	struct socket           *sock;
-	unsigned long		flags;
-	struct page             *rx_page;
-	atomic_t		waiting_requests;
-	struct cbuf		cb;
-	int                     eagain_flag;
-	struct work_struct      work; /* Send workqueue */
-};
-
-/* An entry waiting to be sent */
-
-struct writequeue_entry {
-	struct list_head	list;
-	struct page             *page;
-	int			offset;
-	int			len;
-	int			end;
-	int			users;
-	struct nodeinfo         *ni;
-};
-
-static void cbuf_add(struct cbuf *cb, int n)
-{
-	cb->len += n;
-}
-
-static int cbuf_data(struct cbuf *cb)
-{
-	return ((cb->base + cb->len) & cb->mask);
-}
-
-static void cbuf_init(struct cbuf *cb, int size)
-{
-	cb->base = cb->len = 0;
-	cb->mask = size-1;
-}
-
-static void cbuf_eat(struct cbuf *cb, int n)
-{
-	cb->len  -= n;
-	cb->base += n;
-	cb->base &= cb->mask;
-}
-
-/* List of nodes which have writes pending */
-static LIST_HEAD(write_nodes);
-static DEFINE_SPINLOCK(write_nodes_lock);
-
-
-/* Maximum number of incoming messages to process before
- * doing a schedule()
- */
-#define MAX_RX_MSG_COUNT 25
-
-/* Work queues */
-static struct workqueue_struct *recv_workqueue;
-static struct workqueue_struct *send_workqueue;
-static struct workqueue_struct *lock_workqueue;
-
-/* The SCTP connection */
-static struct connection sctp_con;
-
-static void process_send_sockets(struct work_struct *work);
-static void process_recv_sockets(struct work_struct *work);
-static void process_lock_request(struct work_struct *work);
-
-static int nodeid_to_addr(int nodeid, struct sockaddr *retaddr)
-{
-	struct sockaddr_storage addr;
-	int error;
-
-	if (!dlm_local_count)
-		return -1;
-
-	error = dlm_nodeid_to_addr(nodeid, &addr);
-	if (error)
-		return error;
-
-	if (dlm_local_addr[0]->ss_family == AF_INET) {
-		struct sockaddr_in *in4  = (struct sockaddr_in *) &addr;
-		struct sockaddr_in *ret4 = (struct sockaddr_in *) retaddr;
-		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
-	} else {
-		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &addr;
-		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) retaddr;
-		memcpy(&ret6->sin6_addr, &in6->sin6_addr,
-		       sizeof(in6->sin6_addr));
-	}
-
-	return 0;
-}
-
-/* If alloc is 0 here we will not attempt to allocate a new
-   nodeinfo struct */
-static struct nodeinfo *nodeid2nodeinfo(int nodeid, gfp_t alloc)
-{
-	struct nodeinfo *ni;
-	int r;
-	int n;
-
-	down_read(&nodeinfo_lock);
-	ni = idr_find(&nodeinfo_idr, nodeid);
-	up_read(&nodeinfo_lock);
-
-	if (ni || !alloc)
-		return ni;
-
-	down_write(&nodeinfo_lock);
-
-	ni = idr_find(&nodeinfo_idr, nodeid);
-	if (ni)
-		goto out_up;
-
-	r = idr_pre_get(&nodeinfo_idr, alloc);
-	if (!r)
-		goto out_up;
-
-	ni = kmalloc(sizeof(struct nodeinfo), alloc);
-	if (!ni)
-		goto out_up;
-
-	r = idr_get_new_above(&nodeinfo_idr, ni, nodeid, &n);
-	if (r) {
-		kfree(ni);
-		ni = NULL;
-		goto out_up;
-	}
-	if (n != nodeid) {
-		idr_remove(&nodeinfo_idr, n);
-		kfree(ni);
-		ni = NULL;
-		goto out_up;
-	}
-	memset(ni, 0, sizeof(struct nodeinfo));
-	spin_lock_init(&ni->lock);
-	INIT_LIST_HEAD(&ni->writequeue);
-	spin_lock_init(&ni->writequeue_lock);
-	INIT_WORK(&ni->lwork, process_lock_request);
-	INIT_WORK(&ni->swork, process_send_sockets);
-	ni->nodeid = nodeid;
-
-	if (nodeid > max_nodeid)
-		max_nodeid = nodeid;
-out_up:
-	up_write(&nodeinfo_lock);
-
-	return ni;
-}
-
-/* Don't call this too often... */
-static struct nodeinfo *assoc2nodeinfo(sctp_assoc_t assoc)
-{
-	int i;
-	struct nodeinfo *ni;
-
-	for (i=1; i<=max_nodeid; i++) {
-		ni = nodeid2nodeinfo(i, 0);
-		if (ni && ni->assoc_id == assoc)
-			return ni;
-	}
-	return NULL;
-}
-
-/* Data or notification available on socket */
-static void lowcomms_data_ready(struct sock *sk, int count_unused)
-{
-	if (test_and_set_bit(CF_READ_PENDING, &sctp_con.flags))
-		queue_work(recv_workqueue, &sctp_con.work);
-}
-
-
-/* Add the port number to an IP6 or 4 sockaddr and return the address length.
-   Also padd out the struct with zeros to make comparisons meaningful */
-
-static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
-			  int *addr_len)
-{
-	struct sockaddr_in *local4_addr;
-	struct sockaddr_in6 *local6_addr;
-
-	if (!dlm_local_count)
-		return;
-
-	if (!port) {
-		if (dlm_local_addr[0]->ss_family == AF_INET) {
-			local4_addr = (struct sockaddr_in *)dlm_local_addr[0];
-			port = be16_to_cpu(local4_addr->sin_port);
-		} else {
-			local6_addr = (struct sockaddr_in6 *)dlm_local_addr[0];
-			port = be16_to_cpu(local6_addr->sin6_port);
-		}
-	}
-
-	saddr->ss_family = dlm_local_addr[0]->ss_family;
-	if (dlm_local_addr[0]->ss_family == AF_INET) {
-		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
-		in4_addr->sin_port = cpu_to_be16(port);
-		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
-		memset(in4_addr+1, 0, sizeof(struct sockaddr_storage) -
-		       sizeof(struct sockaddr_in));
-		*addr_len = sizeof(struct sockaddr_in);
-	} else {
-		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
-		in6_addr->sin6_port = cpu_to_be16(port);
-		memset(in6_addr+1, 0, sizeof(struct sockaddr_storage) -
-		       sizeof(struct sockaddr_in6));
-		*addr_len = sizeof(struct sockaddr_in6);
-	}
-}
-
-/* Close the connection and tidy up */
-static void close_connection(void)
-{
-	if (sctp_con.sock) {
-		sock_release(sctp_con.sock);
-		sctp_con.sock = NULL;
-	}
-
-	if (sctp_con.rx_page) {
-		__free_page(sctp_con.rx_page);
-		sctp_con.rx_page = NULL;
-	}
-}
-
-/* We only send shutdown messages to nodes that are not part of the cluster */
-static void send_shutdown(sctp_assoc_t associd)
-{
-	static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
-	struct msghdr outmessage;
-	struct cmsghdr *cmsg;
-	struct sctp_sndrcvinfo *sinfo;
-	int ret;
-
-	outmessage.msg_name = NULL;
-	outmessage.msg_namelen = 0;
-	outmessage.msg_control = outcmsg;
-	outmessage.msg_controllen = sizeof(outcmsg);
-	outmessage.msg_flags = MSG_EOR;
-
-	cmsg = CMSG_FIRSTHDR(&outmessage);
-	cmsg->cmsg_level = IPPROTO_SCTP;
-	cmsg->cmsg_type = SCTP_SNDRCV;
-	cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
-	outmessage.msg_controllen = cmsg->cmsg_len;
-	sinfo = CMSG_DATA(cmsg);
-	memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
-
-	sinfo->sinfo_flags |= MSG_EOF;
-	sinfo->sinfo_assoc_id = associd;
-
-	ret = kernel_sendmsg(sctp_con.sock, &outmessage, NULL, 0, 0);
-
-	if (ret != 0)
-		log_print("send EOF to node failed: %d", ret);
-}
-
-
-/* INIT failed but we don't know which node...
-   restart INIT on all pending nodes */
-static void init_failed(void)
-{
-	int i;
-	struct nodeinfo *ni;
-
-	for (i=1; i<=max_nodeid; i++) {
-		ni = nodeid2nodeinfo(i, 0);
-		if (!ni)
-			continue;
-
-		if (test_and_clear_bit(NI_INIT_PENDING, &ni->flags)) {
-			ni->assoc_id = 0;
-			if (!test_and_set_bit(NI_WRITE_PENDING, &ni->flags)) {
-				spin_lock_bh(&write_nodes_lock);
-				list_add_tail(&ni->write_list, &write_nodes);
-				spin_unlock_bh(&write_nodes_lock);
-				queue_work(send_workqueue, &ni->swork);
-			}
-		}
-	}
-}
-
-/* Something happened to an association */
-static void process_sctp_notification(struct msghdr *msg, char *buf)
-{
-	union sctp_notification *sn = (union sctp_notification *)buf;
-
-	if (sn->sn_header.sn_type == SCTP_ASSOC_CHANGE) {
-		switch (sn->sn_assoc_change.sac_state) {
-
-		case SCTP_COMM_UP:
-		case SCTP_RESTART:
-		{
-			/* Check that the new node is in the lockspace */
-			struct sctp_prim prim;
-			mm_segment_t fs;
-			int nodeid;
-			int prim_len, ret;
-			int addr_len;
-			struct nodeinfo *ni;
-
-			/* This seems to happen when we received a connection
-			 * too early... or something...  anyway, it happens but
-			 * we always seem to get a real message too, see
-			 * receive_from_sock */
-
-			if ((int)sn->sn_assoc_change.sac_assoc_id <= 0) {
-				log_print("COMM_UP for invalid assoc ID %d",
-					  (int)sn->sn_assoc_change.sac_assoc_id);
-				init_failed();
-				return;
-			}
-			memset(&prim, 0, sizeof(struct sctp_prim));
-			prim_len = sizeof(struct sctp_prim);
-			prim.ssp_assoc_id = sn->sn_assoc_change.sac_assoc_id;
-
-			fs = get_fs();
-			set_fs(get_ds());
-			ret = sctp_con.sock->ops->getsockopt(sctp_con.sock,
-							     IPPROTO_SCTP,
-							     SCTP_PRIMARY_ADDR,
-							     (char*)&prim,
-							     &prim_len);
-			set_fs(fs);
-			if (ret < 0) {
-				struct nodeinfo *ni;
-
-				log_print("getsockopt/sctp_primary_addr on "
-					  "new assoc %d failed : %d",
-					  (int)sn->sn_assoc_change.sac_assoc_id,
-					  ret);
-
-				/* Retry INIT later */
-				ni = assoc2nodeinfo(sn->sn_assoc_change.sac_assoc_id);
-				if (ni)
-					clear_bit(NI_INIT_PENDING, &ni->flags);
-				return;
-			}
-			make_sockaddr(&prim.ssp_addr, 0, &addr_len);
-			if (dlm_addr_to_nodeid(&prim.ssp_addr, &nodeid)) {
-				log_print("reject connect from unknown addr");
-				send_shutdown(prim.ssp_assoc_id);
-				return;
-			}
-
-			ni = nodeid2nodeinfo(nodeid, GFP_KERNEL);
-			if (!ni)
-				return;
-
-			/* Save the assoc ID */
-			ni->assoc_id = sn->sn_assoc_change.sac_assoc_id;
-
-			log_print("got new/restarted association %d nodeid %d",
-				  (int)sn->sn_assoc_change.sac_assoc_id, nodeid);
-
-			/* Send any pending writes */
-			clear_bit(NI_INIT_PENDING, &ni->flags);
-			if (!test_and_set_bit(NI_WRITE_PENDING, &ni->flags)) {
-				spin_lock_bh(&write_nodes_lock);
-				list_add_tail(&ni->write_list, &write_nodes);
-				spin_unlock_bh(&write_nodes_lock);
-				queue_work(send_workqueue, &ni->swork);
-			}
-		}
-		break;
-
-		case SCTP_COMM_LOST:
-		case SCTP_SHUTDOWN_COMP:
-		{
-			struct nodeinfo *ni;
-
-			ni = assoc2nodeinfo(sn->sn_assoc_change.sac_assoc_id);
-			if (ni) {
-				spin_lock(&ni->lock);
-				ni->assoc_id = 0;
-				spin_unlock(&ni->lock);
-			}
-		}
-		break;
-
-		/* We don't know which INIT failed, so clear the PENDING flags
-		 * on them all.  if assoc_id is zero then it will then try
-		 * again */
-
-		case SCTP_CANT_STR_ASSOC:
-		{
-			log_print("Can't start SCTP association - retrying");
-			init_failed();
-		}
-		break;
-
-		default:
-			log_print("unexpected SCTP assoc change id=%d state=%d",
-				  (int)sn->sn_assoc_change.sac_assoc_id,
-				  sn->sn_assoc_change.sac_state);
-		}
-	}
-}
-
-/* Data received from remote end */
-static int receive_from_sock(void)
-{
-	int ret = 0;
-	struct msghdr msg;
-	struct kvec iov[2];
-	unsigned len;
-	int r;
-	struct sctp_sndrcvinfo *sinfo;
-	struct cmsghdr *cmsg;
-	struct nodeinfo *ni;
-
-	/* These two are marginally too big for stack allocation, but this
-	 * function is (currently) only called by dlm_recvd so static should be
-	 * OK.
-	 */
-	static struct sockaddr_storage msgname;
-	static char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
-
-	if (sctp_con.sock == NULL)
-		goto out;
-
-	if (sctp_con.rx_page == NULL) {
-		/*
-		 * This doesn't need to be atomic, but I think it should
-		 * improve performance if it is.
-		 */
-		sctp_con.rx_page = alloc_page(GFP_ATOMIC);
-		if (sctp_con.rx_page == NULL)
-			goto out_resched;
-		cbuf_init(&sctp_con.cb, PAGE_CACHE_SIZE);
-	}
-
-	memset(&incmsg, 0, sizeof(incmsg));
-	memset(&msgname, 0, sizeof(msgname));
-
-	msg.msg_name = &msgname;
-	msg.msg_namelen = sizeof(msgname);
-	msg.msg_flags = 0;
-	msg.msg_control = incmsg;
-	msg.msg_controllen = sizeof(incmsg);
-	msg.msg_iovlen = 1;
-
-	/* I don't see why this circular buffer stuff is necessary for SCTP
-	 * which is a packet-based protocol, but the whole thing breaks under
-	 * load without it! The overhead is minimal (and is in the TCP lowcomms
-	 * anyway, of course) so I'll leave it in until I can figure out what's
-	 * really happening.
-	 */
-
-	/*
-	 * iov[0] is the bit of the circular buffer between the current end
-	 * point (cb.base + cb.len) and the end of the buffer.
-	 */
-	iov[0].iov_len = sctp_con.cb.base - cbuf_data(&sctp_con.cb);
-	iov[0].iov_base = page_address(sctp_con.rx_page) +
-		cbuf_data(&sctp_con.cb);
-	iov[1].iov_len = 0;
-
-	/*
-	 * iov[1] is the bit of the circular buffer between the start of the
-	 * buffer and the start of the currently used section (cb.base)
-	 */
-	if (cbuf_data(&sctp_con.cb) >= sctp_con.cb.base) {
-		iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&sctp_con.cb);
-		iov[1].iov_len = sctp_con.cb.base;
-		iov[1].iov_base = page_address(sctp_con.rx_page);
-		msg.msg_iovlen = 2;
-	}
-	len = iov[0].iov_len + iov[1].iov_len;
-
-	r = ret = kernel_recvmsg(sctp_con.sock, &msg, iov, msg.msg_iovlen, len,
-				 MSG_NOSIGNAL | MSG_DONTWAIT);
-	if (ret <= 0)
-		goto out_close;
-
-	msg.msg_control = incmsg;
-	msg.msg_controllen = sizeof(incmsg);
-	cmsg = CMSG_FIRSTHDR(&msg);
-	sinfo = CMSG_DATA(cmsg);
-
-	if (msg.msg_flags & MSG_NOTIFICATION) {
-		process_sctp_notification(&msg, page_address(sctp_con.rx_page));
-		return 0;
-	}
-
-	/* Is this a new association ? */
-	ni = nodeid2nodeinfo(le32_to_cpu(sinfo->sinfo_ppid), GFP_KERNEL);
-	if (ni) {
-		ni->assoc_id = sinfo->sinfo_assoc_id;
-		if (test_and_clear_bit(NI_INIT_PENDING, &ni->flags)) {
-
-			if (!test_and_set_bit(NI_WRITE_PENDING, &ni->flags)) {
-				spin_lock_bh(&write_nodes_lock);
-				list_add_tail(&ni->write_list, &write_nodes);
-				spin_unlock_bh(&write_nodes_lock);
-				queue_work(send_workqueue, &ni->swork);
-			}
-		}
-	}
-
-	/* INIT sends a message with length of 1 - ignore it */
-	if (r == 1)
-		return 0;
-
-	cbuf_add(&sctp_con.cb, ret);
-	// PJC: TODO: Add to node's workqueue....can we ??
-	ret = dlm_process_incoming_buffer(cpu_to_le32(sinfo->sinfo_ppid),
-					  page_address(sctp_con.rx_page),
-					  sctp_con.cb.base, sctp_con.cb.len,
-					  PAGE_CACHE_SIZE);
-	if (ret < 0)
-		goto out_close;
-	cbuf_eat(&sctp_con.cb, ret);
-
-out:
-	ret = 0;
-	goto out_ret;
-
-out_resched:
-	lowcomms_data_ready(sctp_con.sock->sk, 0);
-	ret = 0;
-	cond_resched();
-	goto out_ret;
-
-out_close:
-	if (ret != -EAGAIN)
-		log_print("error reading from sctp socket: %d", ret);
-out_ret:
-	return ret;
-}
-
-/* Bind to an IP address. SCTP allows multiple address so it can do multi-homing */
-static int add_bind_addr(struct sockaddr_storage *addr, int addr_len, int num)
-{
-	mm_segment_t fs;
-	int result = 0;
-
-	fs = get_fs();
-	set_fs(get_ds());
-	if (num == 1)
-		result = sctp_con.sock->ops->bind(sctp_con.sock,
-						  (struct sockaddr *) addr,
-						  addr_len);
-	else
-		result = sctp_con.sock->ops->setsockopt(sctp_con.sock, SOL_SCTP,
-							SCTP_SOCKOPT_BINDX_ADD,
-							(char *)addr, addr_len);
-	set_fs(fs);
-
-	if (result < 0)
-		log_print("Can't bind to port %d addr number %d",
-			  dlm_config.ci_tcp_port, num);
-
-	return result;
-}
-
-static void init_local(void)
-{
-	struct sockaddr_storage sas, *addr;
-	int i;
-
-	dlm_local_nodeid = dlm_our_nodeid();
-
-	for (i = 0; i < DLM_MAX_ADDR_COUNT - 1; i++) {
-		if (dlm_our_addr(&sas, i))
-			break;
-
-		addr = kmalloc(sizeof(*addr), GFP_KERNEL);
-		if (!addr)
-			break;
-		memcpy(addr, &sas, sizeof(*addr));
-		dlm_local_addr[dlm_local_count++] = addr;
-	}
-}
-
-/* Initialise SCTP socket and bind to all interfaces */
-static int init_sock(void)
-{
-	mm_segment_t fs;
-	struct socket *sock = NULL;
-	struct sockaddr_storage localaddr;
-	struct sctp_event_subscribe subscribe;
-	int result = -EINVAL, num = 1, i, addr_len;
-
-	if (!dlm_local_count) {
-		init_local();
-		if (!dlm_local_count) {
-			log_print("no local IP address has been set");
-			goto out;
-		}
-	}
-
-	result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_SEQPACKET,
-				  IPPROTO_SCTP, &sock);
-	if (result < 0) {
-		log_print("Can't create comms socket, check SCTP is loaded");
-		goto out;
-	}
-
-	/* Listen for events */
-	memset(&subscribe, 0, sizeof(subscribe));
-	subscribe.sctp_data_io_event = 1;
-	subscribe.sctp_association_event = 1;
-	subscribe.sctp_send_failure_event = 1;
-	subscribe.sctp_shutdown_event = 1;
-	subscribe.sctp_partial_delivery_event = 1;
-
-	fs = get_fs();
-	set_fs(get_ds());
-	result = sock->ops->setsockopt(sock, SOL_SCTP, SCTP_EVENTS,
-				       (char *)&subscribe, sizeof(subscribe));
-	set_fs(fs);
-
-	if (result < 0) {
-		log_print("Failed to set SCTP_EVENTS on socket: result=%d",
-			  result);
-		goto create_delsock;
-	}
-
-	/* Init con struct */
-	sock->sk->sk_user_data = &sctp_con;
-	sctp_con.sock = sock;
-	sctp_con.sock->sk->sk_data_ready = lowcomms_data_ready;
-
-	/* Bind to all interfaces. */
-	for (i = 0; i < dlm_local_count; i++) {
-		memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
-		make_sockaddr(&localaddr, dlm_config.ci_tcp_port, &addr_len);
-
-		result = add_bind_addr(&localaddr, addr_len, num);
-		if (result)
-			goto create_delsock;
-		++num;
-	}
-
-	result = sock->ops->listen(sock, 5);
-	if (result < 0) {
-		log_print("Can't set socket listening");
-		goto create_delsock;
-	}
-
-	return 0;
-
-create_delsock:
-	sock_release(sock);
-	sctp_con.sock = NULL;
-out:
-	return result;
-}
-
-
-static struct writequeue_entry *new_writequeue_entry(gfp_t allocation)
-{
-	struct writequeue_entry *entry;
-
-	entry = kmalloc(sizeof(struct writequeue_entry), allocation);
-	if (!entry)
-		return NULL;
-
-	entry->page = alloc_page(allocation);
-	if (!entry->page) {
-		kfree(entry);
-		return NULL;
-	}
-
-	entry->offset = 0;
-	entry->len = 0;
-	entry->end = 0;
-	entry->users = 0;
-
-	return entry;
-}
-
-void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
-{
-	struct writequeue_entry *e;
-	int offset = 0;
-	int users = 0;
-	struct nodeinfo *ni;
-
-	ni = nodeid2nodeinfo(nodeid, allocation);
-	if (!ni)
-		return NULL;
-
-	spin_lock(&ni->writequeue_lock);
-	e = list_entry(ni->writequeue.prev, struct writequeue_entry, list);
-	if ((&e->list == &ni->writequeue) ||
-	    (PAGE_CACHE_SIZE - e->end < len)) {
-		e = NULL;
-	} else {
-		offset = e->end;
-		e->end += len;
-		users = e->users++;
-	}
-	spin_unlock(&ni->writequeue_lock);
-
-	if (e) {
-	got_one:
-		if (users == 0)
-			kmap(e->page);
-		*ppc = page_address(e->page) + offset;
-		return e;
-	}
-
-	e = new_writequeue_entry(allocation);
-	if (e) {
-		spin_lock(&ni->writequeue_lock);
-		offset = e->end;
-		e->end += len;
-		e->ni = ni;
-		users = e->users++;
-		list_add_tail(&e->list, &ni->writequeue);
-		spin_unlock(&ni->writequeue_lock);
-		goto got_one;
-	}
-	return NULL;
-}
-
-void dlm_lowcomms_commit_buffer(void *arg)
-{
-	struct writequeue_entry *e = (struct writequeue_entry *) arg;
-	int users;
-	struct nodeinfo *ni = e->ni;
-
-	spin_lock(&ni->writequeue_lock);
-	users = --e->users;
-	if (users)
-		goto out;
-	e->len = e->end - e->offset;
-	kunmap(e->page);
-	spin_unlock(&ni->writequeue_lock);
-
-	if (!test_and_set_bit(NI_WRITE_PENDING, &ni->flags)) {
-		spin_lock_bh(&write_nodes_lock);
-		list_add_tail(&ni->write_list, &write_nodes);
-		spin_unlock_bh(&write_nodes_lock);
-
-		queue_work(send_workqueue, &ni->swork);
-	}
-	return;
-
-out:
-	spin_unlock(&ni->writequeue_lock);
-	return;
-}
-
-static void free_entry(struct writequeue_entry *e)
-{
-	__free_page(e->page);
-	kfree(e);
-}
-
-/* Initiate an SCTP association. In theory we could just use sendmsg() on
-   the first IP address and it should work, but this allows us to set up the
-   association before sending any valuable data that we can't afford to lose.
-   It also keeps the send path clean as it can now always use the association ID */
-static void initiate_association(int nodeid)
-{
-	struct sockaddr_storage rem_addr;
-	static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
-	struct msghdr outmessage;
-	struct cmsghdr *cmsg;
-	struct sctp_sndrcvinfo *sinfo;
-	int ret;
-	int addrlen;
-	char buf[1];
-	struct kvec iov[1];
-	struct nodeinfo *ni;
-
-	log_print("Initiating association with node %d", nodeid);
-
-	ni = nodeid2nodeinfo(nodeid, GFP_KERNEL);
-	if (!ni)
-		return;
-
-	if (nodeid_to_addr(nodeid, (struct sockaddr *)&rem_addr)) {
-		log_print("no address for nodeid %d", nodeid);
-		return;
-	}
-
-	make_sockaddr(&rem_addr, dlm_config.ci_tcp_port, &addrlen);
-
-	outmessage.msg_name = &rem_addr;
-	outmessage.msg_namelen = addrlen;
-	outmessage.msg_control = outcmsg;
-	outmessage.msg_controllen = sizeof(outcmsg);
-	outmessage.msg_flags = MSG_EOR;
-
-	iov[0].iov_base = buf;
-	iov[0].iov_len = 1;
-
-	/* Real INIT messages seem to cause trouble. Just send a 1 byte message
-	   we can afford to lose */
-	cmsg = CMSG_FIRSTHDR(&outmessage);
-	cmsg->cmsg_level = IPPROTO_SCTP;
-	cmsg->cmsg_type = SCTP_SNDRCV;
-	cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
-	sinfo = CMSG_DATA(cmsg);
-	memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
-	sinfo->sinfo_ppid = cpu_to_le32(dlm_local_nodeid);
-
-	outmessage.msg_controllen = cmsg->cmsg_len;
-	ret = kernel_sendmsg(sctp_con.sock, &outmessage, iov, 1, 1);
-	if (ret < 0) {
-		log_print("send INIT to node failed: %d", ret);
-		/* Try again later */
-		clear_bit(NI_INIT_PENDING, &ni->flags);
-	}
-}
-
-/* Send a message */
-static void send_to_sock(struct nodeinfo *ni)
-{
-	int ret = 0;
-	struct writequeue_entry *e;
-	int len, offset;
-	struct msghdr outmsg;
-	static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
-	struct cmsghdr *cmsg;
-	struct sctp_sndrcvinfo *sinfo;
-	struct kvec iov;
-
-	/* See if we need to init an association before we start
-	   sending precious messages */
-	spin_lock(&ni->lock);
-	if (!ni->assoc_id && !test_and_set_bit(NI_INIT_PENDING, &ni->flags)) {
-		spin_unlock(&ni->lock);
-		initiate_association(ni->nodeid);
-		return;
-	}
-	spin_unlock(&ni->lock);
-
-	outmsg.msg_name = NULL; /* We use assoc_id */
-	outmsg.msg_namelen = 0;
-	outmsg.msg_control = outcmsg;
-	outmsg.msg_controllen = sizeof(outcmsg);
-	outmsg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL | MSG_EOR;
-
-	cmsg = CMSG_FIRSTHDR(&outmsg);
-	cmsg->cmsg_level = IPPROTO_SCTP;
-	cmsg->cmsg_type = SCTP_SNDRCV;
-	cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
-	sinfo = CMSG_DATA(cmsg);
-	memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
-	sinfo->sinfo_ppid = cpu_to_le32(dlm_local_nodeid);
-	sinfo->sinfo_assoc_id = ni->assoc_id;
-	outmsg.msg_controllen = cmsg->cmsg_len;
-
-	spin_lock(&ni->writequeue_lock);
-	for (;;) {
-		if (list_empty(&ni->writequeue))
-			break;
-		e = list_entry(ni->writequeue.next, struct writequeue_entry,
-			       list);
-		len = e->len;
-		offset = e->offset;
-		BUG_ON(len == 0 && e->users == 0);
-		spin_unlock(&ni->writequeue_lock);
-		kmap(e->page);
-
-		ret = 0;
-		if (len) {
-			iov.iov_base = page_address(e->page)+offset;
-			iov.iov_len = len;
-
-			ret = kernel_sendmsg(sctp_con.sock, &outmsg, &iov, 1,
-					     len);
-			if (ret == -EAGAIN) {
-				sctp_con.eagain_flag = 1;
-				goto out;
-			} else if (ret < 0)
-				goto send_error;
-		} else {
-			/* Don't starve people filling buffers */
-			cond_resched();
-		}
-
-		spin_lock(&ni->writequeue_lock);
-		e->offset += ret;
-		e->len -= ret;
-
-		if (e->len == 0 && e->users == 0) {
-			list_del(&e->list);
-			kunmap(e->page);
-			free_entry(e);
-			continue;
-		}
-	}
-	spin_unlock(&ni->writequeue_lock);
-out:
-	return;
-
-send_error:
-	log_print("Error sending to node %d %d", ni->nodeid, ret);
-	spin_lock(&ni->lock);
-	if (!test_and_set_bit(NI_INIT_PENDING, &ni->flags)) {
-		ni->assoc_id = 0;
-		spin_unlock(&ni->lock);
-		initiate_association(ni->nodeid);
-	} else
-		spin_unlock(&ni->lock);
-
-	return;
-}
-
-/* Try to send any messages that are pending */
-static void process_output_queue(void)
-{
-	struct list_head *list;
-	struct list_head *temp;
-
-	spin_lock_bh(&write_nodes_lock);
-	list_for_each_safe(list, temp, &write_nodes) {
-		struct nodeinfo *ni =
-			list_entry(list, struct nodeinfo, write_list);
-		clear_bit(NI_WRITE_PENDING, &ni->flags);
-		list_del(&ni->write_list);
-
-		spin_unlock_bh(&write_nodes_lock);
-
-		send_to_sock(ni);
-		spin_lock_bh(&write_nodes_lock);
-	}
-	spin_unlock_bh(&write_nodes_lock);
-}
-
-/* Called after we've had -EAGAIN and been woken up */
-static void refill_write_queue(void)
-{
-	int i;
-
-	for (i=1; i<=max_nodeid; i++) {
-		struct nodeinfo *ni = nodeid2nodeinfo(i, 0);
-
-		if (ni) {
-			if (!test_and_set_bit(NI_WRITE_PENDING, &ni->flags)) {
-				spin_lock_bh(&write_nodes_lock);
-				list_add_tail(&ni->write_list, &write_nodes);
-				spin_unlock_bh(&write_nodes_lock);
-			}
-		}
-	}
-}
-
-static void clean_one_writequeue(struct nodeinfo *ni)
-{
-	struct list_head *list;
-	struct list_head *temp;
-
-	spin_lock(&ni->writequeue_lock);
-	list_for_each_safe(list, temp, &ni->writequeue) {
-		struct writequeue_entry *e =
-			list_entry(list, struct writequeue_entry, list);
-		list_del(&e->list);
-		free_entry(e);
-	}
-	spin_unlock(&ni->writequeue_lock);
-}
-
-static void clean_writequeues(void)
-{
-	int i;
-
-	for (i=1; i<=max_nodeid; i++) {
-		struct nodeinfo *ni = nodeid2nodeinfo(i, 0);
-		if (ni)
-			clean_one_writequeue(ni);
-	}
-}
-
-
-static void dealloc_nodeinfo(void)
-{
-	int i;
-
-	for (i=1; i<=max_nodeid; i++) {
-		struct nodeinfo *ni = nodeid2nodeinfo(i, 0);
-		if (ni) {
-			idr_remove(&nodeinfo_idr, i);
-			kfree(ni);
-		}
-	}
-}
-
-int dlm_lowcomms_close(int nodeid)
-{
-	struct nodeinfo *ni;
-
-	ni = nodeid2nodeinfo(nodeid, 0);
-	if (!ni)
-		return -1;
-
-	spin_lock(&ni->lock);
-	if (ni->assoc_id) {
-		ni->assoc_id = 0;
-		/* Don't send shutdown here, sctp will just queue it
-		   till the node comes back up! */
-	}
-	spin_unlock(&ni->lock);
-
-	clean_one_writequeue(ni);
-	clear_bit(NI_INIT_PENDING, &ni->flags);
-	return 0;
-}
-
-// PJC: The work queue function for receiving.
-static void process_recv_sockets(struct work_struct *work)
-{
-	if (test_and_clear_bit(CF_READ_PENDING, &sctp_con.flags)) {
-		int ret;
-		int count = 0;
-
-		do {
-			ret = receive_from_sock();
-
-			/* Don't starve out everyone else */
-			if (++count >= MAX_RX_MSG_COUNT) {
-				cond_resched();
-				count = 0;
-			}
-		} while (!kthread_should_stop() && ret >=0);
-	}
-	cond_resched();
-}
-
-// PJC: the work queue function for sending
-static void process_send_sockets(struct work_struct *work)
-{
-	if (sctp_con.eagain_flag) {
-		sctp_con.eagain_flag = 0;
-		refill_write_queue();
-	}
-	process_output_queue();
-}
-
-// PJC: Process lock requests from a particular node.
-// TODO: can we optimise this out on UP ??
-static void process_lock_request(struct work_struct *work)
-{
-}
-
-static void daemons_stop(void)
-{
-	destroy_workqueue(recv_workqueue);
-	destroy_workqueue(send_workqueue);
-	destroy_workqueue(lock_workqueue);
-}
-
-static int daemons_start(void)
-{
-	int error;
-	recv_workqueue = create_workqueue("dlm_recv");
-	error = IS_ERR(recv_workqueue);
-	if (error) {
-		log_print("can't start dlm_recv %d", error);
-		return error;
-	}
-
-	send_workqueue = create_singlethread_workqueue("dlm_send");
-	error = IS_ERR(send_workqueue);
-	if (error) {
-		log_print("can't start dlm_send %d", error);
-		destroy_workqueue(recv_workqueue);
-		return error;
-	}
-
-	lock_workqueue = create_workqueue("dlm_rlock");
-	error = IS_ERR(lock_workqueue);
-	if (error) {
-		log_print("can't start dlm_rlock %d", error);
-		destroy_workqueue(send_workqueue);
-		destroy_workqueue(recv_workqueue);
-		return error;
-	}
-
-	return 0;
-}
-
-/*
- * This is quite likely to sleep...
- */
-int dlm_lowcomms_start(void)
-{
-	int error;
-
-	INIT_WORK(&sctp_con.work, process_recv_sockets);
-
-	error = init_sock();
-	if (error)
-		goto fail_sock;
-	error = daemons_start();
-	if (error)
-		goto fail_sock;
-	return 0;
-
-fail_sock:
-	close_connection();
-	return error;
-}
-
-void dlm_lowcomms_stop(void)
-{
-	int i;
-
-	sctp_con.flags = 0x7;
-	daemons_stop();
-	clean_writequeues();
-	close_connection();
-	dealloc_nodeinfo();
-	max_nodeid = 0;
-
-	dlm_local_count = 0;
-	dlm_local_nodeid = 0;
-
-	for (i = 0; i < dlm_local_count; i++)
-		kfree(dlm_local_addr[i]);
-}
diff --git a/fs/dlm/lowcomms-tcp.c b/fs/dlm/lowcomms.c
index 07e0a122c32..27970a58d29 100644
--- a/fs/dlm/lowcomms-tcp.c
+++ b/fs/dlm/lowcomms.c
@@ -36,30 +36,36 @@
  * of high load. Also, this way, the sending thread can collect together
  * messages bound for one node and send them in one block.
  *
- * I don't see any problem with the recv thread executing the locking
- * code on behalf of remote processes as the locking code is
- * short, efficient and never waits.
+ * lowcomms will choose to use wither TCP or SCTP as its transport layer
+ * depending on the configuration variable 'protocol'. This should be set
+ * to 0 (default) for TCP or 1 for SCTP. It shouldbe configured using a
+ * cluster-wide mechanism as it must be the same on all nodes of the cluster
+ * for the DLM to function.
  *
  */
 
-
 #include <asm/ioctls.h>
 #include <net/sock.h>
 #include <net/tcp.h>
 #include <linux/pagemap.h>
+#include <linux/idr.h>
+#include <linux/file.h>
+#include <linux/sctp.h>
+#include <net/sctp/user.h>
 
 #include "dlm_internal.h"
 #include "lowcomms.h"
 #include "midcomms.h"
 #include "config.h"
 
+#define NEEDED_RMEM (4*1024*1024)
+
 struct cbuf {
 	unsigned int base;
 	unsigned int len;
 	unsigned int mask;
 };
 
-#define NODE_INCREMENT 32
 static void cbuf_add(struct cbuf *cb, int n)
 {
 	cb->len += n;
@@ -88,28 +94,25 @@ static bool cbuf_empty(struct cbuf *cb)
 	return cb->len == 0;
 }
 
-/* Maximum number of incoming messages to process before
-   doing a cond_resched()
-*/
-#define MAX_RX_MSG_COUNT 25
-
 struct connection {
 	struct socket *sock;	/* NULL if not connected */
 	uint32_t nodeid;	/* So we know who we are in the list */
 	struct mutex sock_mutex;
-	unsigned long flags;	/* bit 1,2 = We are on the read/write lists */
+	unsigned long flags;
 #define CF_READ_PENDING 1
 #define CF_WRITE_PENDING 2
 #define CF_CONNECT_PENDING 3
-#define CF_IS_OTHERCON 4
+#define CF_INIT_PENDING 4
+#define CF_IS_OTHERCON 5
 	struct list_head writequeue;  /* List of outgoing writequeue_entries */
-	struct list_head listenlist;  /* List of allocated listening sockets */
 	spinlock_t writequeue_lock;
 	int (*rx_action) (struct connection *);	/* What to do when active */
+	void (*connect_action) (struct connection *);	/* What to do to connect */
 	struct page *rx_page;
 	struct cbuf cb;
 	int retries;
 #define MAX_CONNECT_RETRIES 3
+	int sctp_assoc;
 	struct connection *othercon;
 	struct work_struct rwork; /* Receive workqueue */
 	struct work_struct swork; /* Send workqueue */
@@ -127,68 +130,136 @@ struct writequeue_entry {
 	struct connection *con;
 };
 
-static struct sockaddr_storage dlm_local_addr;
+static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
+static int dlm_local_count;
 
 /* Work queues */
 static struct workqueue_struct *recv_workqueue;
 static struct workqueue_struct *send_workqueue;
 
-/* An array of pointers to connections, indexed by NODEID */
-static struct connection **connections;
+static DEFINE_IDR(connections_idr);
 static DECLARE_MUTEX(connections_lock);
+static int max_nodeid;
 static struct kmem_cache *con_cache;
-static int conn_array_size;
 
 static void process_recv_sockets(struct work_struct *work);
 static void process_send_sockets(struct work_struct *work);
 
-static struct connection *nodeid2con(int nodeid, gfp_t allocation)
+/*
+ * If 'allocation' is zero then we don't attempt to create a new
+ * connection structure for this node.
+ */
+static struct connection *__nodeid2con(int nodeid, gfp_t alloc)
 {
 	struct connection *con = NULL;
+	int r;
+	int n;
 
-	down(&connections_lock);
-	if (nodeid >= conn_array_size) {
-		int new_size = nodeid + NODE_INCREMENT;
-		struct connection **new_conns;
+	con = idr_find(&connections_idr, nodeid);
+	if (con || !alloc)
+		return con;
 
-		new_conns = kzalloc(sizeof(struct connection *) *
-				    new_size, allocation);
-		if (!new_conns)
-			goto finish;
+	r = idr_pre_get(&connections_idr, alloc);
+	if (!r)
+		return NULL;
+
+	con = kmem_cache_zalloc(con_cache, alloc);
+	if (!con)
+		return NULL;
 
-		memcpy(new_conns, connections,  sizeof(struct connection *) * conn_array_size);
-		conn_array_size = new_size;
-		kfree(connections);
-		connections = new_conns;
+	r = idr_get_new_above(&connections_idr, con, nodeid, &n);
+	if (r) {
+		kmem_cache_free(con_cache, con);
+		return NULL;
+	}
 
+	if (n != nodeid) {
+		idr_remove(&connections_idr, n);
+		kmem_cache_free(con_cache, con);
+		return NULL;
 	}
 
-	con = connections[nodeid];
-	if (con == NULL && allocation) {
-		con = kmem_cache_zalloc(con_cache, allocation);
-		if (!con)
-			goto finish;
+	con->nodeid = nodeid;
+	mutex_init(&con->sock_mutex);
+	INIT_LIST_HEAD(&con->writequeue);
+	spin_lock_init(&con->writequeue_lock);
+	INIT_WORK(&con->swork, process_send_sockets);
+	INIT_WORK(&con->rwork, process_recv_sockets);
 
-		con->nodeid = nodeid;
-		mutex_init(&con->sock_mutex);
-		INIT_LIST_HEAD(&con->writequeue);
-		spin_lock_init(&con->writequeue_lock);
-		INIT_WORK(&con->swork, process_send_sockets);
-		INIT_WORK(&con->rwork, process_recv_sockets);
+	/* Setup action pointers for child sockets */
+	if (con->nodeid) {
+		struct connection *zerocon = idr_find(&connections_idr, 0);
 
-		connections[nodeid] = con;
+		con->connect_action = zerocon->connect_action;
+		if (!con->rx_action)
+			con->rx_action = zerocon->rx_action;
 	}
 
-finish:
+	if (nodeid > max_nodeid)
+		max_nodeid = nodeid;
+
+	return con;
+}
+
+static struct connection *nodeid2con(int nodeid, gfp_t allocation)
+{
+	struct connection *con;
+
+	down(&connections_lock);
+	con = __nodeid2con(nodeid, allocation);
 	up(&connections_lock);
+
 	return con;
 }
 
+/* This is a bit drastic, but only called when things go wrong */
+static struct connection *assoc2con(int assoc_id)
+{
+	int i;
+	struct connection *con;
+
+	down(&connections_lock);
+	for (i=0; i<=max_nodeid; i++) {
+		con = __nodeid2con(i, 0);
+		if (con && con->sctp_assoc == assoc_id) {
+			up(&connections_lock);
+			return con;
+		}
+	}
+	up(&connections_lock);
+	return NULL;
+}
+
+static int nodeid_to_addr(int nodeid, struct sockaddr *retaddr)
+{
+	struct sockaddr_storage addr;
+	int error;
+
+	if (!dlm_local_count)
+		return -1;
+
+	error = dlm_nodeid_to_addr(nodeid, &addr);
+	if (error)
+		return error;
+
+	if (dlm_local_addr[0]->ss_family == AF_INET) {
+		struct sockaddr_in *in4  = (struct sockaddr_in *) &addr;
+		struct sockaddr_in *ret4 = (struct sockaddr_in *) retaddr;
+		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
+	} else {
+		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &addr;
+		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) retaddr;
+		memcpy(&ret6->sin6_addr, &in6->sin6_addr,
+		       sizeof(in6->sin6_addr));
+	}
+
+	return 0;
+}
+
 /* Data available on socket or listen socket received a connect */
 static void lowcomms_data_ready(struct sock *sk, int count_unused)
 {
 	struct connection *con = sock2con(sk);
-
 	if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 		queue_work(recv_workqueue, &con->rwork);
 }
@@ -222,20 +293,21 @@ static int add_sock(struct socket *sock, struct connection *con)
 	con->sock->sk->sk_data_ready = lowcomms_data_ready;
 	con->sock->sk->sk_write_space = lowcomms_write_space;
 	con->sock->sk->sk_state_change = lowcomms_state_change;
-
+	con->sock->sk->sk_user_data = con;
 	return 0;
 }
 
-/* Add the port number to an IP6 or 4 sockaddr and return the address
+/* Add the port number to an IPv6 or 4 sockaddr and return the address
    length */
 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 			  int *addr_len)
 {
-	saddr->ss_family =  dlm_local_addr.ss_family;
+	saddr->ss_family =  dlm_local_addr[0]->ss_family;
 	if (saddr->ss_family == AF_INET) {
 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 		in4_addr->sin_port = cpu_to_be16(port);
 		*addr_len = sizeof(struct sockaddr_in);
+		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 	} else {
 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 		in6_addr->sin6_port = cpu_to_be16(port);
@@ -264,6 +336,193 @@ static void close_connection(struct connection *con, bool and_other)
 	mutex_unlock(&con->sock_mutex);
 }
 
+/* We only send shutdown messages to nodes that are not part of the cluster */
+static void sctp_send_shutdown(sctp_assoc_t associd)
+{
+	static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
+	struct msghdr outmessage;
+	struct cmsghdr *cmsg;
+	struct sctp_sndrcvinfo *sinfo;
+	int ret;
+	struct connection *con;
+
+	con = nodeid2con(0,0);
+	BUG_ON(con == NULL);
+
+	outmessage.msg_name = NULL;
+	outmessage.msg_namelen = 0;
+	outmessage.msg_control = outcmsg;
+	outmessage.msg_controllen = sizeof(outcmsg);
+	outmessage.msg_flags = MSG_EOR;
+
+	cmsg = CMSG_FIRSTHDR(&outmessage);
+	cmsg->cmsg_level = IPPROTO_SCTP;
+	cmsg->cmsg_type = SCTP_SNDRCV;
+	cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
+	outmessage.msg_controllen = cmsg->cmsg_len;
+	sinfo = CMSG_DATA(cmsg);
+	memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
+
+	sinfo->sinfo_flags |= MSG_EOF;
+	sinfo->sinfo_assoc_id = associd;
+
+	ret = kernel_sendmsg(con->sock, &outmessage, NULL, 0, 0);
+
+	if (ret != 0)
+		log_print("send EOF to node failed: %d", ret);
+}
+
+/* INIT failed but we don't know which node...
+   restart INIT on all pending nodes */
+static void sctp_init_failed(void)
+{
+	int i;
+	struct connection *con;
+
+	down(&connections_lock);
+	for (i=1; i<=max_nodeid; i++) {
+		con = __nodeid2con(i, 0);
+		if (!con)
+			continue;
+		con->sctp_assoc = 0;
+		if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) {
+			if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) {
+				queue_work(send_workqueue, &con->swork);
+			}
+		}
+	}
+	up(&connections_lock);
+}
+
+/* Something happened to an association */
+static void process_sctp_notification(struct connection *con,
+				      struct msghdr *msg, char *buf)
+{
+	union sctp_notification *sn = (union sctp_notification *)buf;
+
+	if (sn->sn_header.sn_type == SCTP_ASSOC_CHANGE) {
+		switch (sn->sn_assoc_change.sac_state) {
+
+		case SCTP_COMM_UP:
+		case SCTP_RESTART:
+		{
+			/* Check that the new node is in the lockspace */
+			struct sctp_prim prim;
+			int nodeid;
+			int prim_len, ret;
+			int addr_len;
+			struct connection *new_con;
+			struct file *file;
+			sctp_peeloff_arg_t parg;
+			int parglen = sizeof(parg);
+
+			/*
+			 * We get this before any data for an association.
+			 * We verify that the node is in the cluster and
+			 * then peel off a socket for it.
+			 */
+			if ((int)sn->sn_assoc_change.sac_assoc_id <= 0) {
+				log_print("COMM_UP for invalid assoc ID %d",
+					 (int)sn->sn_assoc_change.sac_assoc_id);
+				sctp_init_failed();
+				return;
+			}
+			memset(&prim, 0, sizeof(struct sctp_prim));
+			prim_len = sizeof(struct sctp_prim);
+			prim.ssp_assoc_id = sn->sn_assoc_change.sac_assoc_id;
+
+			ret = kernel_getsockopt(con->sock,
+						IPPROTO_SCTP,
+						SCTP_PRIMARY_ADDR,
+						(char*)&prim,
+						&prim_len);
+			if (ret < 0) {
+				log_print("getsockopt/sctp_primary_addr on "
+					  "new assoc %d failed : %d",
+					  (int)sn->sn_assoc_change.sac_assoc_id,
+					  ret);
+
+				/* Retry INIT later */
+				new_con = assoc2con(sn->sn_assoc_change.sac_assoc_id);
+				if (new_con)
+					clear_bit(CF_CONNECT_PENDING, &con->flags);
+				return;
+			}
+			make_sockaddr(&prim.ssp_addr, 0, &addr_len);
+			if (dlm_addr_to_nodeid(&prim.ssp_addr, &nodeid)) {
+				int i;
+				unsigned char *b=(unsigned char *)&prim.ssp_addr;
+				log_print("reject connect from unknown addr");
+				for (i=0; i<sizeof(struct sockaddr_storage);i++)
+					printk("%02x ", b[i]);
+				printk("\n");
+				sctp_send_shutdown(prim.ssp_assoc_id);
+				return;
+			}
+
+			new_con = nodeid2con(nodeid, GFP_KERNEL);
+			if (!new_con)
+				return;
+
+			/* Peel off a new sock */
+			parg.associd = sn->sn_assoc_change.sac_assoc_id;
+			ret = kernel_getsockopt(con->sock, IPPROTO_SCTP,
+						SCTP_SOCKOPT_PEELOFF,
+						(void *)&parg, &parglen);
+			if (ret) {
+				log_print("Can't peel off a socket for "
+					  "connection %d to node %d: err=%d\n",
+					  parg.associd, nodeid, ret);
+			}
+			file = fget(parg.sd);
+			new_con->sock = SOCKET_I(file->f_dentry->d_inode);
+			add_sock(new_con->sock, new_con);
+			fput(file);
+			put_unused_fd(parg.sd);
+
+			log_print("got new/restarted association %d nodeid %d",
+				 (int)sn->sn_assoc_change.sac_assoc_id, nodeid);
+
+			/* Send any pending writes */
+			clear_bit(CF_CONNECT_PENDING, &new_con->flags);
+			clear_bit(CF_INIT_PENDING, &con->flags);
+			if (!test_and_set_bit(CF_WRITE_PENDING, &new_con->flags)) {
+				queue_work(send_workqueue, &new_con->swork);
+			}
+			if (!test_and_set_bit(CF_READ_PENDING, &new_con->flags))
+				queue_work(recv_workqueue, &new_con->rwork);
+		}
+		break;
+
+		case SCTP_COMM_LOST:
+		case SCTP_SHUTDOWN_COMP:
+		{
+			con = assoc2con(sn->sn_assoc_change.sac_assoc_id);
+			if (con) {
+				con->sctp_assoc = 0;
+			}
+		}
+		break;
+
+		/* We don't know which INIT failed, so clear the PENDING flags
+		 * on them all.  if assoc_id is zero then it will then try
+		 * again */
+
+		case SCTP_CANT_STR_ASSOC:
+		{
+			log_print("Can't start SCTP association - retrying");
+			sctp_init_failed();
+		}
+		break;
+
+		default:
+			log_print("unexpected SCTP assoc change id=%d state=%d",
+				  (int)sn->sn_assoc_change.sac_assoc_id,
+				  sn->sn_assoc_change.sac_state);
+		}
+	}
+}
+
 /* Data received from remote end */
 static int receive_from_sock(struct connection *con)
 {
@@ -274,6 +533,7 @@ static int receive_from_sock(struct connection *con)
 	int r;
 	int call_again_soon = 0;
 	int nvec;
+	char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
 
 	mutex_lock(&con->sock_mutex);
 
@@ -293,12 +553,18 @@ static int receive_from_sock(struct connection *con)
 		cbuf_init(&con->cb, PAGE_CACHE_SIZE);
 	}
 
+	/* Only SCTP needs these really */
+	memset(&incmsg, 0, sizeof(incmsg));
+	msg.msg_control = incmsg;
+	msg.msg_controllen = sizeof(incmsg);
+
 	/*
 	 * iov[0] is the bit of the circular buffer between the current end
 	 * point (cb.base + cb.len) and the end of the buffer.
 	 */
 	iov[0].iov_len = con->cb.base - cbuf_data(&con->cb);
 	iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb);
+	iov[1].iov_len = 0;
 	nvec = 1;
 
 	/*
@@ -315,11 +581,20 @@ static int receive_from_sock(struct connection *con)
 
 	r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len,
 			       MSG_DONTWAIT | MSG_NOSIGNAL);
-
 	if (ret <= 0)
 		goto out_close;
-	if (ret == -EAGAIN)
-		goto out_resched;
+
+	/* Process SCTP notifications */
+	if (msg.msg_flags & MSG_NOTIFICATION) {
+		msg.msg_control = incmsg;
+		msg.msg_controllen = sizeof(incmsg);
+
+		process_sctp_notification(con, &msg,
+				page_address(con->rx_page) + con->cb.base);
+		mutex_unlock(&con->sock_mutex);
+		return 0;
+	}
+	BUG_ON(con->nodeid == 0);
 
 	if (ret == len)
 		call_again_soon = 1;
@@ -329,10 +604,10 @@ static int receive_from_sock(struct connection *con)
 					  con->cb.base, con->cb.len,
 					  PAGE_CACHE_SIZE);
 	if (ret == -EBADMSG) {
-		printk(KERN_INFO "dlm: lowcomms: addr=%p, base=%u, len=%u, "
-		       "iov_len=%u, iov_base[0]=%p, read=%d\n",
-		       page_address(con->rx_page), con->cb.base, con->cb.len,
-		       len, iov[0].iov_base, r);
+		log_print("lowcomms: addr=%p, base=%u, len=%u, "
+			  "iov_len=%u, iov_base[0]=%p, read=%d",
+			  page_address(con->rx_page), con->cb.base, con->cb.len,
+			  len, iov[0].iov_base, r);
 	}
 	if (ret < 0)
 		goto out_close;
@@ -368,7 +643,7 @@ out_close:
 }
 
 /* Listening socket is busy, accept a connection */
-static int accept_from_sock(struct connection *con)
+static int tcp_accept_from_sock(struct connection *con)
 {
 	int result;
 	struct sockaddr_storage peeraddr;
@@ -379,7 +654,7 @@ static int accept_from_sock(struct connection *con)
 	struct connection *addcon;
 
 	memset(&peeraddr, 0, sizeof(peeraddr));
-	result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM,
+	result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
 				  IPPROTO_TCP, &newsock);
 	if (result < 0)
 		return -ENOMEM;
@@ -408,7 +683,7 @@ static int accept_from_sock(struct connection *con)
 	/* Get the new node's NODEID */
 	make_sockaddr(&peeraddr, 0, &len);
 	if (dlm_addr_to_nodeid(&peeraddr, &nodeid)) {
-		printk("dlm: connect from non cluster node\n");
+		log_print("connect from non cluster node");
 		sock_release(newsock);
 		mutex_unlock(&con->sock_mutex);
 		return -1;
@@ -419,7 +694,6 @@ static int accept_from_sock(struct connection *con)
 	/*  Check to see if we already have a connection to this node. This
 	 *  could happen if the two nodes initiate a connection at roughly
 	 *  the same time and the connections cross on the wire.
-	 * TEMPORARY FIX:
 	 *  In this case we store the incoming one in "othercon"
 	 */
 	newcon = nodeid2con(nodeid, GFP_KERNEL);
@@ -434,7 +708,7 @@ static int accept_from_sock(struct connection *con)
 		if (!othercon) {
 			othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL);
 			if (!othercon) {
-				printk("dlm: failed to allocate incoming socket\n");
+				log_print("failed to allocate incoming socket");
 				mutex_unlock(&newcon->sock_mutex);
 				result = -ENOMEM;
 				goto accept_err;
@@ -477,12 +751,107 @@ accept_err:
 	sock_release(newsock);
 
 	if (result != -EAGAIN)
-		printk("dlm: error accepting connection from node: %d\n", result);
+		log_print("error accepting connection from node: %d", result);
 	return result;
 }
 
+static void free_entry(struct writequeue_entry *e)
+{
+	__free_page(e->page);
+	kfree(e);
+}
+
+/* Initiate an SCTP association.
+   This is a special case of send_to_sock() in that we don't yet have a
+   peeled-off socket for this association, so we use the listening socket
+   and add the primary IP address of the remote node.
+ */
+static void sctp_init_assoc(struct connection *con)
+{
+	struct sockaddr_storage rem_addr;
+	char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
+	struct msghdr outmessage;
+	struct cmsghdr *cmsg;
+	struct sctp_sndrcvinfo *sinfo;
+	struct connection *base_con;
+	struct writequeue_entry *e;
+	int len, offset;
+	int ret;
+	int addrlen;
+	struct kvec iov[1];
+
+	if (test_and_set_bit(CF_INIT_PENDING, &con->flags))
+		return;
+
+	if (con->retries++ > MAX_CONNECT_RETRIES)
+		return;
+
+	log_print("Initiating association with node %d", con->nodeid);
+
+	if (nodeid_to_addr(con->nodeid, (struct sockaddr *)&rem_addr)) {
+		log_print("no address for nodeid %d", con->nodeid);
+		return;
+	}
+	base_con = nodeid2con(0, 0);
+	BUG_ON(base_con == NULL);
+
+	make_sockaddr(&rem_addr, dlm_config.ci_tcp_port, &addrlen);
+
+	outmessage.msg_name = &rem_addr;
+	outmessage.msg_namelen = addrlen;
+	outmessage.msg_control = outcmsg;
+	outmessage.msg_controllen = sizeof(outcmsg);
+	outmessage.msg_flags = MSG_EOR;
+
+	spin_lock(&con->writequeue_lock);
+	e = list_entry(con->writequeue.next, struct writequeue_entry,
+		       list);
+
+	BUG_ON((struct list_head *) e == &con->writequeue);
+
+	len = e->len;
+	offset = e->offset;
+	spin_unlock(&con->writequeue_lock);
+	kmap(e->page);
+
+	/* Send the first block off the write queue */
+	iov[0].iov_base = page_address(e->page)+offset;
+	iov[0].iov_len = len;
+
+	cmsg = CMSG_FIRSTHDR(&outmessage);
+	cmsg->cmsg_level = IPPROTO_SCTP;
+	cmsg->cmsg_type = SCTP_SNDRCV;
+	cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
+	sinfo = CMSG_DATA(cmsg);
+	memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
+	sinfo->sinfo_ppid = cpu_to_le32(dlm_our_nodeid());
+	outmessage.msg_controllen = cmsg->cmsg_len;
+
+	ret = kernel_sendmsg(base_con->sock, &outmessage, iov, 1, len);
+	if (ret < 0) {
+		log_print("Send first packet to node %d failed: %d",
+			  con->nodeid, ret);
+
+		/* Try again later */
+		clear_bit(CF_CONNECT_PENDING, &con->flags);
+		clear_bit(CF_INIT_PENDING, &con->flags);
+	}
+	else {
+		spin_lock(&con->writequeue_lock);
+		e->offset += ret;
+		e->len -= ret;
+
+		if (e->len == 0 && e->users == 0) {
+			list_del(&e->list);
+			kunmap(e->page);
+			free_entry(e);
+		}
+		spin_unlock(&con->writequeue_lock);
+	}
+}
+
 /* Connect a new socket to its peer */
-static void connect_to_sock(struct connection *con)
+static void tcp_connect_to_sock(struct connection *con)
 {
 	int result = -EHOSTUNREACH;
 	struct sockaddr_storage saddr;
@@ -505,7 +874,7 @@ static void connect_to_sock(struct connection *con)
 	}
 
 	/* Create a socket to communicate with */
-	result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM,
+	result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
 				  IPPROTO_TCP, &sock);
 	if (result < 0)
 		goto out_err;
@@ -516,11 +885,11 @@ static void connect_to_sock(struct connection *con)
 
 	sock->sk->sk_user_data = con;
 	con->rx_action = receive_from_sock;
+	con->connect_action = tcp_connect_to_sock;
+	add_sock(sock, con);
 
 	make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
 
-	add_sock(sock, con);
-
 	log_print("connecting to %d", con->nodeid);
 	result =
 		sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
@@ -550,64 +919,57 @@ out:
 	return;
 }
 
-static struct socket *create_listen_sock(struct connection *con,
-					 struct sockaddr_storage *saddr)
+static struct socket *tcp_create_listen_sock(struct connection *con,
+					     struct sockaddr_storage *saddr)
 {
 	struct socket *sock = NULL;
-	mm_segment_t fs;
 	int result = 0;
 	int one = 1;
 	int addr_len;
 
-	if (dlm_local_addr.ss_family == AF_INET)
+	if (dlm_local_addr[0]->ss_family == AF_INET)
 		addr_len = sizeof(struct sockaddr_in);
 	else
 		addr_len = sizeof(struct sockaddr_in6);
 
 	/* Create a socket to communicate with */
-	result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM, IPPROTO_TCP, &sock);
+	result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
+				  IPPROTO_TCP, &sock);
 	if (result < 0) {
-		printk("dlm: Can't create listening comms socket\n");
+		log_print("Can't create listening comms socket");
 		goto create_out;
 	}
 
-	fs = get_fs();
-	set_fs(get_ds());
-	result = sock_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
-				 (char *)&one, sizeof(one));
-	set_fs(fs);
+	result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
+				   (char *)&one, sizeof(one));
+
 	if (result < 0) {
-		printk("dlm: Failed to set SO_REUSEADDR on socket: result=%d\n",
-		       result);
+		log_print("Failed to set SO_REUSEADDR on socket: %d", result);
 	}
 	sock->sk->sk_user_data = con;
-	con->rx_action = accept_from_sock;
+	con->rx_action = tcp_accept_from_sock;
+	con->connect_action = tcp_connect_to_sock;
 	con->sock = sock;
 
 	/* Bind to our port */
 	make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
 	result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
 	if (result < 0) {
-		printk("dlm: Can't bind to port %d\n", dlm_config.ci_tcp_port);
+		log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
 		sock_release(sock);
 		sock = NULL;
 		con->sock = NULL;
 		goto create_out;
 	}
-
-	fs = get_fs();
-	set_fs(get_ds());
-
-	result = sock_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
+	result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
 				 (char *)&one, sizeof(one));
-	set_fs(fs);
 	if (result < 0) {
-		printk("dlm: Set keepalive failed: %d\n", result);
+		log_print("Set keepalive failed: %d", result);
 	}
 
 	result = sock->ops->listen(sock, 5);
 	if (result < 0) {
-		printk("dlm: Can't listen on port %d\n", dlm_config.ci_tcp_port);
+		log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
 		sock_release(sock);
 		sock = NULL;
 		goto create_out;
@@ -617,18 +979,146 @@ create_out:
 	return sock;
 }
 
+/* Get local addresses */
+static void init_local(void)
+{
+	struct sockaddr_storage sas, *addr;
+	int i;
+
+	dlm_local_count = 0;
+	for (i = 0; i < DLM_MAX_ADDR_COUNT - 1; i++) {
+		if (dlm_our_addr(&sas, i))
+			break;
+
+		addr = kmalloc(sizeof(*addr), GFP_KERNEL);
+		if (!addr)
+			break;
+		memcpy(addr, &sas, sizeof(*addr));
+		dlm_local_addr[dlm_local_count++] = addr;
+	}
+}
+
+/* Bind to an IP address. SCTP allows multiple address so it can do
+   multi-homing */
+static int add_sctp_bind_addr(struct connection *sctp_con,
+			      struct sockaddr_storage *addr,
+			      int addr_len, int num)
+{
+	int result = 0;
+
+	if (num == 1)
+		result = kernel_bind(sctp_con->sock,
+				     (struct sockaddr *) addr,
+				     addr_len);
+	else
+		result = kernel_setsockopt(sctp_con->sock, SOL_SCTP,
+					   SCTP_SOCKOPT_BINDX_ADD,
+					   (char *)addr, addr_len);
+
+	if (result < 0)
+		log_print("Can't bind to port %d addr number %d",
+			  dlm_config.ci_tcp_port, num);
+
+	return result;
+}
 
-/* Listen on all interfaces */
-static int listen_for_all(void)
+/* Initialise SCTP socket and bind to all interfaces */
+static int sctp_listen_for_all(void)
+{
+	struct socket *sock = NULL;
+	struct sockaddr_storage localaddr;
+	struct sctp_event_subscribe subscribe;
+	int result = -EINVAL, num = 1, i, addr_len;
+	struct connection *con = nodeid2con(0, GFP_KERNEL);
+	int bufsize = NEEDED_RMEM;
+
+	if (!con)
+		return -ENOMEM;
+
+	log_print("Using SCTP for communications");
+
+	result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_SEQPACKET,
+				  IPPROTO_SCTP, &sock);
+	if (result < 0) {
+		log_print("Can't create comms socket, check SCTP is loaded");
+		goto out;
+	}
+
+	/* Listen for events */
+	memset(&subscribe, 0, sizeof(subscribe));
+	subscribe.sctp_data_io_event = 1;
+	subscribe.sctp_association_event = 1;
+	subscribe.sctp_send_failure_event = 1;
+	subscribe.sctp_shutdown_event = 1;
+	subscribe.sctp_partial_delivery_event = 1;
+
+	result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
+				 (char *)&bufsize, sizeof(bufsize));
+	if (result)
+		log_print("Error increasing buffer space on socket %d", result);
+
+	result = kernel_setsockopt(sock, SOL_SCTP, SCTP_EVENTS,
+				   (char *)&subscribe, sizeof(subscribe));
+	if (result < 0) {
+		log_print("Failed to set SCTP_EVENTS on socket: result=%d",
+			  result);
+		goto create_delsock;
+	}
+
+	/* Init con struct */
+	sock->sk->sk_user_data = con;
+	con->sock = sock;
+	con->sock->sk->sk_data_ready = lowcomms_data_ready;
+	con->rx_action = receive_from_sock;
+	con->connect_action = sctp_init_assoc;
+
+	/* Bind to all interfaces. */
+	for (i = 0; i < dlm_local_count; i++) {
+		memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
+		make_sockaddr(&localaddr, dlm_config.ci_tcp_port, &addr_len);
+
+		result = add_sctp_bind_addr(con, &localaddr, addr_len, num);
+		if (result)
+			goto create_delsock;
+		++num;
+	}
+
+	result = sock->ops->listen(sock, 5);
+	if (result < 0) {
+		log_print("Can't set socket listening");
+		goto create_delsock;
+	}
+
+	return 0;
+
+create_delsock:
+	sock_release(sock);
+	con->sock = NULL;
+out:
+	return result;
+}
+
+static int tcp_listen_for_all(void)
 {
 	struct socket *sock = NULL;
 	struct connection *con = nodeid2con(0, GFP_KERNEL);
 	int result = -EINVAL;
 
+	if (!con)
+		return -ENOMEM;
+
 	/* We don't support multi-homed hosts */
+	if (dlm_local_addr[1] != NULL) {
+		log_print("TCP protocol can't handle multi-homed hosts, "
+			  "try SCTP");
+		return -EINVAL;
+	}
+
+	log_print("Using TCP for communications");
+
 	set_bit(CF_IS_OTHERCON, &con->flags);
 
-	sock = create_listen_sock(con, &dlm_local_addr);
+	sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
 	if (sock) {
 		add_sock(sock, con);
 		result = 0;
@@ -666,8 +1156,7 @@ static struct writequeue_entry *new_writequeue_entry(struct connection *con,
 	return entry;
 }
 
-void *dlm_lowcomms_get_buffer(int nodeid, int len,
-			      gfp_t allocation, char **ppc)
+void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
 {
 	struct connection *con;
 	struct writequeue_entry *e;
@@ -735,12 +1224,6 @@ out:
 	return;
 }
 
-static void free_entry(struct writequeue_entry *e)
-{
-	__free_page(e->page);
-	kfree(e);
-}
-
 /* Send a message */
 static void send_to_sock(struct connection *con)
 {
@@ -777,8 +1260,7 @@ static void send_to_sock(struct connection *con)
 				goto out;
 			if (ret <= 0)
 				goto send_error;
-		}
-		else {
+		} else {
 			/* Don't starve people filling buffers */
 			cond_resched();
 		}
@@ -807,7 +1289,8 @@ send_error:
 
 out_connect:
 	mutex_unlock(&con->sock_mutex);
-	connect_to_sock(con);
+	if (!test_bit(CF_INIT_PENDING, &con->flags))
+		lowcomms_connect_sock(con);
 	return;
 }
 
@@ -832,9 +1315,6 @@ int dlm_lowcomms_close(int nodeid)
 {
 	struct connection *con;
 
-	if (!connections)
-		goto out;
-
 	log_print("closing connection to node %d", nodeid);
 	con = nodeid2con(nodeid, 0);
 	if (con) {
@@ -842,12 +1322,9 @@ int dlm_lowcomms_close(int nodeid)
 		close_connection(con, true);
 	}
 	return 0;
-
-out:
-	return -1;
 }
 
-/* Look for activity on active sockets */
+/* Receive workqueue function */
 static void process_recv_sockets(struct work_struct *work)
 {
 	struct connection *con = container_of(work, struct connection, rwork);
@@ -859,15 +1336,14 @@ static void process_recv_sockets(struct work_struct *work)
 	} while (!err);
 }
 
-
+/* Send workqueue function */
 static void process_send_sockets(struct work_struct *work)
 {
 	struct connection *con = container_of(work, struct connection, swork);
 
 	if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) {
-		connect_to_sock(con);
+		con->connect_action(con);
 	}
-
 	clear_bit(CF_WRITE_PENDING, &con->flags);
 	send_to_sock(con);
 }
@@ -878,8 +1354,8 @@ static void clean_writequeues(void)
 {
 	int nodeid;
 
-	for (nodeid = 1; nodeid < conn_array_size; nodeid++) {
-		struct connection *con = nodeid2con(nodeid, 0);
+	for (nodeid = 1; nodeid <= max_nodeid; nodeid++) {
+		struct connection *con = __nodeid2con(nodeid, 0);
 
 		if (con)
 			clean_one_writequeue(con);
@@ -916,64 +1392,67 @@ static int work_start(void)
 void dlm_lowcomms_stop(void)
 {
 	int i;
+	struct connection *con;
 
 	/* Set all the flags to prevent any
 	   socket activity.
 	*/
-	for (i = 0; i < conn_array_size; i++) {
-		if (connections[i])
-			connections[i]->flags |= 0xFF;
+	down(&connections_lock);
+	for (i = 0; i <= max_nodeid; i++) {
+		con = __nodeid2con(i, 0);
+		if (con)
+			con->flags |= 0xFF;
 	}
+	up(&connections_lock);
 
 	work_stop();
+
+	down(&connections_lock);
 	clean_writequeues();
 
-	for (i = 0; i < conn_array_size; i++) {
-		if (connections[i]) {
-			close_connection(connections[i], true);
-			if (connections[i]->othercon)
-				kmem_cache_free(con_cache, connections[i]->othercon);
-			kmem_cache_free(con_cache, connections[i]);
+	for (i = 0; i <= max_nodeid; i++) {
+		con = __nodeid2con(i, 0);
+		if (con) {
+			close_connection(con, true);
+			if (con->othercon)
+				kmem_cache_free(con_cache, con->othercon);
+			kmem_cache_free(con_cache, con);
 		}
 	}
-
-	kfree(connections);
-	connections = NULL;
-
+	max_nodeid = 0;
+	up(&connections_lock);
 	kmem_cache_destroy(con_cache);
+	idr_init(&connections_idr);
 }
 
-/* This is quite likely to sleep... */
 int dlm_lowcomms_start(void)
 {
-	int error = 0;
-
-	error = -ENOMEM;
-	connections = kzalloc(sizeof(struct connection *) *
-			      NODE_INCREMENT, GFP_KERNEL);
-	if (!connections)
-		goto out;
-
-	conn_array_size = NODE_INCREMENT;
+	int error = -EINVAL;
+	struct connection *con;
 
-	if (dlm_our_addr(&dlm_local_addr, 0)) {
+	init_local();
+	if (!dlm_local_count) {
+		error = -ENOTCONN;
 		log_print("no local IP address has been set");
-		goto fail_free_conn;
-	}
-	if (!dlm_our_addr(&dlm_local_addr, 1)) {
-		log_print("This dlm comms module does not support multi-homed clustering");
-		goto fail_free_conn;
+		goto out;
 	}
 
+	error = -ENOMEM;
 	con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
 				      __alignof__(struct connection), 0,
 				      NULL, NULL);
 	if (!con_cache)
-		goto fail_free_conn;
+		goto out;
 
+	/* Set some sysctl minima */
+	if (sysctl_rmem_max < NEEDED_RMEM)
+		sysctl_rmem_max = NEEDED_RMEM;
 
 	/* Start listening */
-	error = listen_for_all();
+	if (dlm_config.ci_protocol == 0)
+		error = tcp_listen_for_all();
+	else
+		error = sctp_listen_for_all();
 	if (error)
 		goto fail_unlisten;
 
@@ -984,24 +1463,13 @@ int dlm_lowcomms_start(void)
 	return 0;
 
 fail_unlisten:
-	close_connection(connections[0], false);
-	kmem_cache_free(con_cache, connections[0]);
+	con = nodeid2con(0,0);
+	if (con) {
+		close_connection(con, false);
+		kmem_cache_free(con_cache, con);
+	}
 	kmem_cache_destroy(con_cache);
 
-fail_free_conn:
-	kfree(connections);
-
 out:
 	return error;
 }
-
-/*
- * Overrides for Emacs so that we follow Linus's tabbing style.
- * Emacs will notice this stuff at the end of the file and automatically
- * adjust the settings for this buffer only.  This must remain at the end
- * of the file.
- * ---------------------------------------------------------------------------
- * Local variables:
- * c-file-style: "linux"
- * End:
- */
diff --git a/fs/dlm/user.c b/fs/dlm/user.c
index 3870150b83a..b0201ec325a 100644
--- a/fs/dlm/user.c
+++ b/fs/dlm/user.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
+ * Copyright (C) 2006-2007 Red Hat, Inc.  All rights reserved.
  *
  * This copyrighted material is made available to anyone wishing to use,
  * modify, copy, or redistribute it subject to the terms and conditions
@@ -56,6 +56,7 @@ struct dlm_write_request32 {
 	union  {
 		struct dlm_lock_params32 lock;
 		struct dlm_lspace_params lspace;
+		struct dlm_purge_params purge;
 	} i;
 };
 
@@ -92,6 +93,9 @@ static void compat_input(struct dlm_write_request *kb,
 		kb->i.lspace.flags = kb32->i.lspace.flags;
 		kb->i.lspace.minor = kb32->i.lspace.minor;
 		strcpy(kb->i.lspace.name, kb32->i.lspace.name);
+	} else if (kb->cmd == DLM_USER_PURGE) {
+		kb->i.purge.nodeid = kb32->i.purge.nodeid;
+		kb->i.purge.pid = kb32->i.purge.pid;
 	} else {
 		kb->i.lock.mode = kb32->i.lock.mode;
 		kb->i.lock.namelen = kb32->i.lock.namelen;
@@ -111,8 +115,6 @@ static void compat_input(struct dlm_write_request *kb,
 static void compat_output(struct dlm_lock_result *res,
 			  struct dlm_lock_result32 *res32)
 {
-	res32->length = res->length - (sizeof(struct dlm_lock_result) -
-				       sizeof(struct dlm_lock_result32));
 	res32->user_astaddr = (__u32)(long)res->user_astaddr;
 	res32->user_astparam = (__u32)(long)res->user_astparam;
 	res32->user_lksb = (__u32)(long)res->user_lksb;
@@ -128,35 +130,30 @@ static void compat_output(struct dlm_lock_result *res,
 }
 #endif
 
+/* we could possibly check if the cancel of an orphan has resulted in the lkb
+   being removed and then remove that lkb from the orphans list and free it */
 
 void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
 {
 	struct dlm_ls *ls;
 	struct dlm_user_args *ua;
 	struct dlm_user_proc *proc;
-	int remove_ownqueue = 0;
+	int eol = 0, ast_type;
 
-	/* dlm_clear_proc_locks() sets ORPHAN/DEAD flag on each
-	   lkb before dealing with it.  We need to check this
-	   flag before taking ls_clear_proc_locks mutex because if
-	   it's set, dlm_clear_proc_locks() holds the mutex. */
-
-	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) {
-		/* log_print("user_add_ast skip1 %x", lkb->lkb_flags); */
+	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
 		return;
-	}
 
 	ls = lkb->lkb_resource->res_ls;
 	mutex_lock(&ls->ls_clear_proc_locks);
 
 	/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
 	   can't be delivered.  For ORPHAN's, dlm_clear_proc_locks() freed
-	   lkb->ua so we can't try to use it. */
+	   lkb->ua so we can't try to use it.  This second check is necessary
+	   for cases where a completion ast is received for an operation that
+	   began before clear_proc_locks did its cancel/unlock. */
 
-	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) {
-		/* log_print("user_add_ast skip2 %x", lkb->lkb_flags); */
+	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
 		goto out;
-	}
 
 	DLM_ASSERT(lkb->lkb_astparam, dlm_print_lkb(lkb););
 	ua = (struct dlm_user_args *)lkb->lkb_astparam;
@@ -166,28 +163,42 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
 		goto out;
 
 	spin_lock(&proc->asts_spin);
-	if (!(lkb->lkb_ast_type & (AST_COMP | AST_BAST))) {
+
+	ast_type = lkb->lkb_ast_type;
+	lkb->lkb_ast_type |= type;
+
+	if (!ast_type) {
 		kref_get(&lkb->lkb_ref);
 		list_add_tail(&lkb->lkb_astqueue, &proc->asts);
-		lkb->lkb_ast_type |= type;
 		wake_up_interruptible(&proc->wait);
 	}
-
-	/* noqueue requests that fail may need to be removed from the
-	   proc's locks list, there should be a better way of detecting
-	   this situation than checking all these things... */
-
-	if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV &&
-	    ua->lksb.sb_status == -EAGAIN && !list_empty(&lkb->lkb_ownqueue))
-		remove_ownqueue = 1;
-
-	/* unlocks or cancels of waiting requests need to be removed from the
-	   proc's unlocking list, again there must be a better way...  */
-
-	if (ua->lksb.sb_status == -DLM_EUNLOCK ||
+	if (type == AST_COMP && (ast_type & AST_COMP))
+		log_debug(ls, "ast overlap %x status %x %x",
+			  lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
+
+	/* Figure out if this lock is at the end of its life and no longer
+	   available for the application to use.  The lkb still exists until
+	   the final ast is read.  A lock becomes EOL in three situations:
+	     1. a noqueue request fails with EAGAIN
+	     2. an unlock completes with EUNLOCK
+	     3. a cancel of a waiting request completes with ECANCEL
+	   An EOL lock needs to be removed from the process's list of locks.
+	   And we can't allow any new operation on an EOL lock.  This is
+	   not related to the lifetime of the lkb struct which is managed
+	   entirely by refcount. */
+
+	if (type == AST_COMP &&
+	    lkb->lkb_grmode == DLM_LOCK_IV &&
+	    ua->lksb.sb_status == -EAGAIN)
+		eol = 1;
+	else if (ua->lksb.sb_status == -DLM_EUNLOCK ||
 	    (ua->lksb.sb_status == -DLM_ECANCEL &&
 	     lkb->lkb_grmode == DLM_LOCK_IV))
-		remove_ownqueue = 1;
+		eol = 1;
+	if (eol) {
+		lkb->lkb_ast_type &= ~AST_BAST;
+		lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
+	}
 
 	/* We want to copy the lvb to userspace when the completion
 	   ast is read if the status is 0, the lock has an lvb and
@@ -204,11 +215,13 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
 
 	spin_unlock(&proc->asts_spin);
 
-	if (remove_ownqueue) {
+	if (eol) {
 		spin_lock(&ua->proc->locks_spin);
-		list_del_init(&lkb->lkb_ownqueue);
+		if (!list_empty(&lkb->lkb_ownqueue)) {
+			list_del_init(&lkb->lkb_ownqueue);
+			dlm_put_lkb(lkb);
+		}
 		spin_unlock(&ua->proc->locks_spin);
-		dlm_put_lkb(lkb);
 	}
  out:
 	mutex_unlock(&ls->ls_clear_proc_locks);
@@ -286,47 +299,71 @@ static int device_user_unlock(struct dlm_user_proc *proc,
 	return error;
 }
 
-static int device_create_lockspace(struct dlm_lspace_params *params)
+static int create_misc_device(struct dlm_ls *ls, char *name)
 {
-	dlm_lockspace_t *lockspace;
-	struct dlm_ls *ls;
 	int error, len;
 
-	if (!capable(CAP_SYS_ADMIN))
-		return -EPERM;
-
-	error = dlm_new_lockspace(params->name, strlen(params->name),
-				  &lockspace, 0, DLM_USER_LVB_LEN);
-	if (error)
-		return error;
-
-	ls = dlm_find_lockspace_local(lockspace);
-	if (!ls)
-		return -ENOENT;
-
 	error = -ENOMEM;
-	len = strlen(params->name) + strlen(name_prefix) + 2;
+	len = strlen(name) + strlen(name_prefix) + 2;
 	ls->ls_device.name = kzalloc(len, GFP_KERNEL);
 	if (!ls->ls_device.name)
 		goto fail;
+
 	snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
-		 params->name);
+		 name);
 	ls->ls_device.fops = &device_fops;
 	ls->ls_device.minor = MISC_DYNAMIC_MINOR;
 
 	error = misc_register(&ls->ls_device);
 	if (error) {
 		kfree(ls->ls_device.name);
-		goto fail;
 	}
+fail:
+	return error;
+}
+
+static int device_user_purge(struct dlm_user_proc *proc,
+			     struct dlm_purge_params *params)
+{
+	struct dlm_ls *ls;
+	int error;
+
+	ls = dlm_find_lockspace_local(proc->lockspace);
+	if (!ls)
+		return -ENOENT;
+
+	error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
 
-	error = ls->ls_device.minor;
 	dlm_put_lockspace(ls);
 	return error;
+}
+
+static int device_create_lockspace(struct dlm_lspace_params *params)
+{
+	dlm_lockspace_t *lockspace;
+	struct dlm_ls *ls;
+	int error;
 
- fail:
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	error = dlm_new_lockspace(params->name, strlen(params->name),
+				  &lockspace, 0, DLM_USER_LVB_LEN);
+	if (error)
+		return error;
+
+	ls = dlm_find_lockspace_local(lockspace);
+	if (!ls)
+		return -ENOENT;
+
+	error = create_misc_device(ls, params->name);
 	dlm_put_lockspace(ls);
-	dlm_release_lockspace(lockspace, 0);
+
+	if (error)
+		dlm_release_lockspace(lockspace, 0);
+	else
+		error = ls->ls_device.minor;
+
 	return error;
 }
 
@@ -343,6 +380,10 @@ static int device_remove_lockspace(struct dlm_lspace_params *params)
 	if (!ls)
 		return -ENOENT;
 
+	/* Deregister the misc device first, so we don't have
+	 * a device that's not attached to a lockspace. If
+	 * dlm_release_lockspace fails then we can recreate it
+	 */
 	error = misc_deregister(&ls->ls_device);
 	if (error) {
 		dlm_put_lockspace(ls);
@@ -361,6 +402,8 @@ static int device_remove_lockspace(struct dlm_lspace_params *params)
 
 	dlm_put_lockspace(ls);
 	error = dlm_release_lockspace(lockspace, force);
+	if (error)
+		create_misc_device(ls, ls->ls_name);
  out:
 	return error;
 }
@@ -497,6 +540,14 @@ static ssize_t device_write(struct file *file, const char __user *buf,
 		error = device_remove_lockspace(&kbuf->i.lspace);
 		break;
 
+	case DLM_USER_PURGE:
+		if (!proc) {
+			log_print("no locking on control device");
+			goto out_sig;
+		}
+		error = device_user_purge(proc, &kbuf->i.purge);
+		break;
+
 	default:
 		log_print("Unknown command passed to DLM device : %d\n",
 			  kbuf->cmd);