[DLM] Consolidate transport protocols
authorPatrick Caulfield <pcaulfie@redhat.com>
Tue, 17 Apr 2007 14:39:57 +0000 (15:39 +0100)
committerSteven Whitehouse <swhiteho@redhat.com>
Tue, 1 May 2007 08:11:23 +0000 (09:11 +0100)
This patch consolidates the TCP & SCTP protocols for the DLM into a single file
and makes it switchable at run-time (well, at least before the DLM actually
starts up!)

For RHEL5 this patch requires Neil Horman's patch that expands the in-kernel
socket API but that has already been twice ACKed so it should be OK.

The patch adds a new lowcomms.c file that replaces the existing lowcomms-sctp.c
& lowcomms-tcp.c files.

Signed-off-By: Patrick Caulfield <pcaulfie@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
fs/dlm/Kconfig
fs/dlm/Makefile
fs/dlm/config.c
fs/dlm/config.h
fs/dlm/lowcomms-sctp.c [deleted file]
fs/dlm/lowcomms-tcp.c [deleted file]
fs/dlm/lowcomms.c [new file with mode: 0644]

index 6fa7b0d5c04387b6ab7cf9b3f90f7dc21e21831a..69a94690e493504259fe0d4c04bdd152c56e71ee 100644 (file)
@@ -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
index 65388944eba0e8a5507842ffe23bcde8d0d39866..604cf7dc5f39856e558b8f8362867e1aaf12323f 100644 (file)
@@ -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
index 8665c88e5af26c674bc9e066355193e99f5c7061..822abdcd1434ad94638b6ec652e5720bddd74532 100644 (file)
@@ -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
 };
 
index 1e978611a96e7165b480d09ac3a129f1839b0950..967cc3d72e5e844a893eb2aab1bf7e23ef966fb3 100644 (file)
@@ -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/lowcomms-sctp.c b/fs/dlm/lowcomms-sctp.c
deleted file mode 100644 (file)
index dc83a9d..0000000
+++ /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-tcp.c
deleted file mode 100644 (file)
index 919e92a..0000000
+++ /dev/null
@@ -1,1006 +0,0 @@
-/******************************************************************************
-*******************************************************************************
-**
-**  Copyright (C) Sistina Software, Inc.  1997-2003  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
-**  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 waits.
- *
- */
-
-
-#include <asm/ioctls.h>
-#include <net/sock.h>
-#include <net/tcp.h>
-#include <linux/pagemap.h>
-
-#include "dlm_internal.h"
-#include "lowcomms.h"
-#include "midcomms.h"
-#include "config.h"
-
-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;
-}
-
-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;
-}
-
-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 */
-#define CF_READ_PENDING 1
-#define CF_WRITE_PENDING 2
-#define CF_CONNECT_PENDING 3
-#define CF_IS_OTHERCON 4
-       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 */
-       struct page *rx_page;
-       struct cbuf cb;
-       int retries;
-#define MAX_CONNECT_RETRIES 3
-       struct connection *othercon;
-       struct work_struct rwork; /* Receive workqueue */
-       struct work_struct swork; /* Send workqueue */
-};
-#define sock2con(x) ((struct connection *)(x)->sk_user_data)
-
-/* 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 connection *con;
-};
-
-static struct sockaddr_storage dlm_local_addr;
-
-/* 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 DECLARE_MUTEX(connections_lock);
-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)
-{
-       struct connection *con = NULL;
-
-       down(&connections_lock);
-       if (nodeid >= conn_array_size) {
-               int new_size = nodeid + NODE_INCREMENT;
-               struct connection **new_conns;
-
-               new_conns = kzalloc(sizeof(struct connection *) *
-                                   new_size, allocation);
-               if (!new_conns)
-                       goto finish;
-
-               memcpy(new_conns, connections,  sizeof(struct connection *) * conn_array_size);
-               conn_array_size = new_size;
-               kfree(connections);
-               connections = new_conns;
-
-       }
-
-       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);
-
-               connections[nodeid] = con;
-       }
-
-finish:
-       up(&connections_lock);
-       return con;
-}
-
-/* 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);
-}
-
-static void lowcomms_write_space(struct sock *sk)
-{
-       struct connection *con = sock2con(sk);
-
-       if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags))
-               queue_work(send_workqueue, &con->swork);
-}
-
-static inline void lowcomms_connect_sock(struct connection *con)
-{
-       if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags))
-               queue_work(send_workqueue, &con->swork);
-}
-
-static void lowcomms_state_change(struct sock *sk)
-{
-       if (sk->sk_state == TCP_ESTABLISHED)
-               lowcomms_write_space(sk);
-}
-
-/* Make a socket active */
-static int add_sock(struct socket *sock, struct connection *con)
-{
-       con->sock = sock;
-
-       /* Install a data_ready callback */
-       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;
-
-       return 0;
-}
-
-/* Add the port number to an IP6 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;
-       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);
-       } else {
-               struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
-               in6_addr->sin6_port = cpu_to_be16(port);
-               *addr_len = sizeof(struct sockaddr_in6);
-       }
-}
-
-/* Close a remote connection and tidy up */
-static void close_connection(struct connection *con, bool and_other)
-{
-       mutex_lock(&con->sock_mutex);
-
-       if (con->sock) {
-               sock_release(con->sock);
-               con->sock = NULL;
-       }
-       if (con->othercon && and_other) {
-               /* Will only re-enter once. */
-               close_connection(con->othercon, false);
-       }
-       if (con->rx_page) {
-               __free_page(con->rx_page);
-               con->rx_page = NULL;
-       }
-       con->retries = 0;
-       mutex_unlock(&con->sock_mutex);
-}
-
-/* Data received from remote end */
-static int receive_from_sock(struct connection *con)
-{
-       int ret = 0;
-       struct msghdr msg = {};
-       struct kvec iov[2];
-       unsigned len;
-       int r;
-       int call_again_soon = 0;
-       int nvec;
-
-       mutex_lock(&con->sock_mutex);
-
-       if (con->sock == NULL) {
-               ret = -EAGAIN;
-               goto out_close;
-       }
-
-       if (con->rx_page == NULL) {
-               /*
-                * This doesn't need to be atomic, but I think it should
-                * improve performance if it is.
-                */
-               con->rx_page = alloc_page(GFP_ATOMIC);
-               if (con->rx_page == NULL)
-                       goto out_resched;
-               cbuf_init(&con->cb, PAGE_CACHE_SIZE);
-       }
-
-       /*
-        * 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;
-
-       /*
-        * 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(&con->cb) >= con->cb.base) {
-               iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb);
-               iov[1].iov_len = con->cb.base;
-               iov[1].iov_base = page_address(con->rx_page);
-               nvec = 2;
-       }
-       len = iov[0].iov_len + iov[1].iov_len;
-
-       r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len,
-                              MSG_DONTWAIT | MSG_NOSIGNAL);
-
-       if (ret <= 0)
-               goto out_close;
-
-       if (ret == len)
-               call_again_soon = 1;
-       cbuf_add(&con->cb, ret);
-       ret = dlm_process_incoming_buffer(con->nodeid,
-                                         page_address(con->rx_page),
-                                         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);
-       }
-       if (ret < 0)
-               goto out_close;
-       cbuf_eat(&con->cb, ret);
-
-       if (cbuf_empty(&con->cb) && !call_again_soon) {
-               __free_page(con->rx_page);
-               con->rx_page = NULL;
-       }
-
-       if (call_again_soon)
-               goto out_resched;
-       mutex_unlock(&con->sock_mutex);
-       return 0;
-
-out_resched:
-       if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
-               queue_work(recv_workqueue, &con->rwork);
-       mutex_unlock(&con->sock_mutex);
-       return -EAGAIN;
-
-out_close:
-       mutex_unlock(&con->sock_mutex);
-       if (ret != -EAGAIN && !test_bit(CF_IS_OTHERCON, &con->flags)) {
-               close_connection(con, false);
-               /* Reconnect when there is something to send */
-       }
-       /* Don't return success if we really got EOF */
-       if (ret == 0)
-               ret = -EAGAIN;
-
-       return ret;
-}
-
-/* Listening socket is busy, accept a connection */
-static int accept_from_sock(struct connection *con)
-{
-       int result;
-       struct sockaddr_storage peeraddr;
-       struct socket *newsock;
-       int len;
-       int nodeid;
-       struct connection *newcon;
-       struct connection *addcon;
-
-       memset(&peeraddr, 0, sizeof(peeraddr));
-       result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM,
-                                 IPPROTO_TCP, &newsock);
-       if (result < 0)
-               return -ENOMEM;
-
-       mutex_lock_nested(&con->sock_mutex, 0);
-
-       result = -ENOTCONN;
-       if (con->sock == NULL)
-               goto accept_err;
-
-       newsock->type = con->sock->type;
-       newsock->ops = con->sock->ops;
-
-       result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK);
-       if (result < 0)
-               goto accept_err;
-
-       /* Get the connected socket's peer */
-       memset(&peeraddr, 0, sizeof(peeraddr));
-       if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr,
-                                 &len, 2)) {
-               result = -ECONNABORTED;
-               goto accept_err;
-       }
-
-       /* 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");
-               sock_release(newsock);
-               mutex_unlock(&con->sock_mutex);
-               return -1;
-       }
-
-       log_print("got connection from %d", nodeid);
-
-       /*  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);
-       if (!newcon) {
-               result = -ENOMEM;
-               goto accept_err;
-       }
-       mutex_lock_nested(&newcon->sock_mutex, 1);
-       if (newcon->sock) {
-               struct connection *othercon = newcon->othercon;
-
-               if (!othercon) {
-                       othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL);
-                       if (!othercon) {
-                               printk("dlm: failed to allocate incoming socket\n");
-                               mutex_unlock(&newcon->sock_mutex);
-                               result = -ENOMEM;
-                               goto accept_err;
-                       }
-                       othercon->nodeid = nodeid;
-                       othercon->rx_action = receive_from_sock;
-                       mutex_init(&othercon->sock_mutex);
-                       INIT_WORK(&othercon->swork, process_send_sockets);
-                       INIT_WORK(&othercon->rwork, process_recv_sockets);
-                       set_bit(CF_IS_OTHERCON, &othercon->flags);
-                       newcon->othercon = othercon;
-               }
-               othercon->sock = newsock;
-               newsock->sk->sk_user_data = othercon;
-               add_sock(newsock, othercon);
-               addcon = othercon;
-       }
-       else {
-               newsock->sk->sk_user_data = newcon;
-               newcon->rx_action = receive_from_sock;
-               add_sock(newsock, newcon);
-               addcon = newcon;
-       }
-
-       mutex_unlock(&newcon->sock_mutex);
-
-       /*
-        * Add it to the active queue in case we got data
-        * beween processing the accept adding the socket
-        * to the read_sockets list
-        */
-       if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
-               queue_work(recv_workqueue, &addcon->rwork);
-       mutex_unlock(&con->sock_mutex);
-
-       return 0;
-
-accept_err:
-       mutex_unlock(&con->sock_mutex);
-       sock_release(newsock);
-
-       if (result != -EAGAIN)
-               printk("dlm: error accepting connection from node: %d\n", result);
-       return result;
-}
-
-/* Connect a new socket to its peer */
-static void connect_to_sock(struct connection *con)
-{
-       int result = -EHOSTUNREACH;
-       struct sockaddr_storage saddr;
-       int addr_len;
-       struct socket *sock;
-
-       if (con->nodeid == 0) {
-               log_print("attempt to connect sock 0 foiled");
-               return;
-       }
-
-       mutex_lock(&con->sock_mutex);
-       if (con->retries++ > MAX_CONNECT_RETRIES)
-               goto out;
-
-       /* Some odd races can cause double-connects, ignore them */
-       if (con->sock) {
-               result = 0;
-               goto out;
-       }
-
-       /* Create a socket to communicate with */
-       result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM,
-                                 IPPROTO_TCP, &sock);
-       if (result < 0)
-               goto out_err;
-
-       memset(&saddr, 0, sizeof(saddr));
-       if (dlm_nodeid_to_addr(con->nodeid, &saddr))
-               goto out_err;
-
-       sock->sk->sk_user_data = con;
-       con->rx_action = receive_from_sock;
-
-       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,
-                                  O_NONBLOCK);
-       if (result == -EINPROGRESS)
-               result = 0;
-       if (result == 0)
-               goto out;
-
-out_err:
-       if (con->sock) {
-               sock_release(con->sock);
-               con->sock = NULL;
-       }
-       /*
-        * Some errors are fatal and this list might need adjusting. For other
-        * errors we try again until the max number of retries is reached.
-        */
-       if (result != -EHOSTUNREACH && result != -ENETUNREACH &&
-           result != -ENETDOWN && result != EINVAL
-           && result != -EPROTONOSUPPORT) {
-               lowcomms_connect_sock(con);
-               result = 0;
-       }
-out:
-       mutex_unlock(&con->sock_mutex);
-       return;
-}
-
-static struct socket *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)
-               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);
-       if (result < 0) {
-               printk("dlm: Can't create listening comms socket\n");
-               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);
-       if (result < 0) {
-               printk("dlm: Failed to set SO_REUSEADDR on socket: result=%d\n",
-                      result);
-       }
-       sock->sk->sk_user_data = con;
-       con->rx_action = accept_from_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);
-               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,
-                                (char *)&one, sizeof(one));
-       set_fs(fs);
-       if (result < 0) {
-               printk("dlm: Set keepalive failed: %d\n", result);
-       }
-
-       result = sock->ops->listen(sock, 5);
-       if (result < 0) {
-               printk("dlm: Can't listen on port %d\n", dlm_config.ci_tcp_port);
-               sock_release(sock);
-               sock = NULL;
-               goto create_out;
-       }
-
-create_out:
-       return sock;
-}
-
-
-/* Listen on all interfaces */
-static int listen_for_all(void)
-{
-       struct socket *sock = NULL;
-       struct connection *con = nodeid2con(0, GFP_KERNEL);
-       int result = -EINVAL;
-
-       /* We don't support multi-homed hosts */
-       set_bit(CF_IS_OTHERCON, &con->flags);
-
-       sock = create_listen_sock(con, &dlm_local_addr);
-       if (sock) {
-               add_sock(sock, con);
-               result = 0;
-       }
-       else {
-               result = -EADDRINUSE;
-       }
-
-       return result;
-}
-
-
-
-static struct writequeue_entry *new_writequeue_entry(struct connection *con,
-                                                    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;
-       entry->con = con;
-
-       return entry;
-}
-
-void *dlm_lowcomms_get_buffer(int nodeid, int len,
-                             gfp_t allocation, char **ppc)
-{
-       struct connection *con;
-       struct writequeue_entry *e;
-       int offset = 0;
-       int users = 0;
-
-       con = nodeid2con(nodeid, allocation);
-       if (!con)
-               return NULL;
-
-       spin_lock(&con->writequeue_lock);
-       e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
-       if ((&e->list == &con->writequeue) ||
-           (PAGE_CACHE_SIZE - e->end < len)) {
-               e = NULL;
-       } else {
-               offset = e->end;
-               e->end += len;
-               users = e->users++;
-       }
-       spin_unlock(&con->writequeue_lock);
-
-       if (e) {
-       got_one:
-               if (users == 0)
-                       kmap(e->page);
-               *ppc = page_address(e->page) + offset;
-               return e;
-       }
-
-       e = new_writequeue_entry(con, allocation);
-       if (e) {
-               spin_lock(&con->writequeue_lock);
-               offset = e->end;
-               e->end += len;
-               users = e->users++;
-               list_add_tail(&e->list, &con->writequeue);
-               spin_unlock(&con->writequeue_lock);
-               goto got_one;
-       }
-       return NULL;
-}
-
-void dlm_lowcomms_commit_buffer(void *mh)
-{
-       struct writequeue_entry *e = (struct writequeue_entry *)mh;
-       struct connection *con = e->con;
-       int users;
-
-       spin_lock(&con->writequeue_lock);
-       users = --e->users;
-       if (users)
-               goto out;
-       e->len = e->end - e->offset;
-       kunmap(e->page);
-       spin_unlock(&con->writequeue_lock);
-
-       if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) {
-               queue_work(send_workqueue, &con->swork);
-       }
-       return;
-
-out:
-       spin_unlock(&con->writequeue_lock);
-       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)
-{
-       int ret = 0;
-       ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int);
-       const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
-       struct writequeue_entry *e;
-       int len, offset;
-
-       mutex_lock(&con->sock_mutex);
-       if (con->sock == NULL)
-               goto out_connect;
-
-       sendpage = con->sock->ops->sendpage;
-
-       spin_lock(&con->writequeue_lock);
-       for (;;) {
-               e = list_entry(con->writequeue.next, struct writequeue_entry,
-                              list);
-               if ((struct list_head *) e == &con->writequeue)
-                       break;
-
-               len = e->len;
-               offset = e->offset;
-               BUG_ON(len == 0 && e->users == 0);
-               spin_unlock(&con->writequeue_lock);
-               kmap(e->page);
-
-               ret = 0;
-               if (len) {
-                       ret = sendpage(con->sock, e->page, offset, len,
-                                      msg_flags);
-                       if (ret == -EAGAIN || ret == 0)
-                               goto out;
-                       if (ret <= 0)
-                               goto send_error;
-               }
-               else {
-                       /* Don't starve people filling buffers */
-                       cond_resched();
-               }
-
-               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);
-                       continue;
-               }
-       }
-       spin_unlock(&con->writequeue_lock);
-out:
-       mutex_unlock(&con->sock_mutex);
-       return;
-
-send_error:
-       mutex_unlock(&con->sock_mutex);
-       close_connection(con, false);
-       lowcomms_connect_sock(con);
-       return;
-
-out_connect:
-       mutex_unlock(&con->sock_mutex);
-       connect_to_sock(con);
-       return;
-}
-
-static void clean_one_writequeue(struct connection *con)
-{
-       struct list_head *list;
-       struct list_head *temp;
-
-       spin_lock(&con->writequeue_lock);
-       list_for_each_safe(list, temp, &con->writequeue) {
-               struct writequeue_entry *e =
-                       list_entry(list, struct writequeue_entry, list);
-               list_del(&e->list);
-               free_entry(e);
-       }
-       spin_unlock(&con->writequeue_lock);
-}
-
-/* Called from recovery when it knows that a node has
-   left the cluster */
-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) {
-               clean_one_writequeue(con);
-               close_connection(con, true);
-       }
-       return 0;
-
-out:
-       return -1;
-}
-
-/* Look for activity on active sockets */
-static void process_recv_sockets(struct work_struct *work)
-{
-       struct connection *con = container_of(work, struct connection, rwork);
-       int err;
-
-       clear_bit(CF_READ_PENDING, &con->flags);
-       do {
-               err = con->rx_action(con);
-       } while (!err);
-}
-
-
-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);
-       }
-
-       clear_bit(CF_WRITE_PENDING, &con->flags);
-       send_to_sock(con);
-}
-
-
-/* Discard all entries on the write queues */
-static void clean_writequeues(void)
-{
-       int nodeid;
-
-       for (nodeid = 1; nodeid < conn_array_size; nodeid++) {
-               struct connection *con = nodeid2con(nodeid, 0);
-
-               if (con)
-                       clean_one_writequeue(con);
-       }
-}
-
-static void work_stop(void)
-{
-       destroy_workqueue(recv_workqueue);
-       destroy_workqueue(send_workqueue);
-}
-
-static int work_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;
-       }
-
-       return 0;
-}
-
-void dlm_lowcomms_stop(void)
-{
-       int i;
-
-       /* Set all the flags to prevent any
-          socket activity.
-       */
-       for (i = 0; i < conn_array_size; i++) {
-               if (connections[i])
-                       connections[i]->flags |= 0xFF;
-       }
-
-       work_stop();
-       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]);
-               }
-       }
-
-       kfree(connections);
-       connections = NULL;
-
-       kmem_cache_destroy(con_cache);
-}
-
-/* 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;
-
-       if (dlm_our_addr(&dlm_local_addr, 0)) {
-               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;
-       }
-
-       con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
-                                     __alignof__(struct connection), 0,
-                                     NULL, NULL);
-       if (!con_cache)
-               goto fail_free_conn;
-
-
-       /* Start listening */
-       error = listen_for_all();
-       if (error)
-               goto fail_unlisten;
-
-       error = work_start();
-       if (error)
-               goto fail_unlisten;
-
-       return 0;
-
-fail_unlisten:
-       close_connection(connections[0], false);
-       kmem_cache_free(con_cache, connections[0]);
-       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/lowcomms.c b/fs/dlm/lowcomms.c
new file mode 100644 (file)
index 0000000..76399b7
--- /dev/null
@@ -0,0 +1,1468 @@
+/******************************************************************************
+*******************************************************************************
+**
+**  Copyright (C) Sistina Software, Inc.  1997-2003  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
+**  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.
+ *
+ * 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;
+};
+
+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;
+}
+
+static bool cbuf_empty(struct cbuf *cb)
+{
+       return cb->len == 0;
+}
+
+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;
+#define CF_READ_PENDING 1
+#define CF_WRITE_PENDING 2
+#define CF_CONNECT_PENDING 3
+#define CF_INIT_PENDING 4
+#define CF_IS_OTHERCON 5
+       struct list_head writequeue;  /* List of outgoing writequeue_entries */
+       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 */
+};
+#define sock2con(x) ((struct connection *)(x)->sk_user_data)
+
+/* 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 connection *con;
+};
+
+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;
+
+static DEFINE_IDR(connections_idr);
+static DECLARE_MUTEX(connections_lock);
+static int max_nodeid;
+static struct kmem_cache *con_cache;
+
+static void process_recv_sockets(struct work_struct *work);
+static void process_send_sockets(struct work_struct *work);
+
+/*
+ * 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;
+
+       con = idr_find(&connections_idr, nodeid);
+       if (con || !alloc)
+               return con;
+
+       r = idr_pre_get(&connections_idr, alloc);
+       if (!r)
+               return NULL;
+
+       con = kmem_cache_zalloc(con_cache, alloc);
+       if (!con)
+               return NULL;
+
+       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->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);
+
+               con->connect_action = zerocon->connect_action;
+               if (!con->rx_action)
+                       con->rx_action = zerocon->rx_action;
+       }
+
+       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);
+}
+
+static void lowcomms_write_space(struct sock *sk)
+{
+       struct connection *con = sock2con(sk);
+
+       if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags))
+               queue_work(send_workqueue, &con->swork);
+}
+
+static inline void lowcomms_connect_sock(struct connection *con)
+{
+       if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags))
+               queue_work(send_workqueue, &con->swork);
+}
+
+static void lowcomms_state_change(struct sock *sk)
+{
+       if (sk->sk_state == TCP_ESTABLISHED)
+               lowcomms_write_space(sk);
+}
+
+/* Make a socket active */
+static int add_sock(struct socket *sock, struct connection *con)
+{
+       con->sock = sock;
+
+       /* Install a data_ready callback */
+       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 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[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);
+               *addr_len = sizeof(struct sockaddr_in6);
+       }
+}
+
+/* Close a remote connection and tidy up */
+static void close_connection(struct connection *con, bool and_other)
+{
+       mutex_lock(&con->sock_mutex);
+
+       if (con->sock) {
+               sock_release(con->sock);
+               con->sock = NULL;
+       }
+       if (con->othercon && and_other) {
+               /* Will only re-enter once. */
+               close_connection(con->othercon, false);
+       }
+       if (con->rx_page) {
+               __free_page(con->rx_page);
+               con->rx_page = NULL;
+       }
+       con->retries = 0;
+       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 < 0) {
+                               log_print("Can't peel off a socket for connection %d to node %d: err=%d\n",
+                                         parg.associd, nodeid, ret);
+                               return;
+                       }
+
+                       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)
+{
+       int ret = 0;
+       struct msghdr msg = {};
+       struct kvec iov[2];
+       unsigned len;
+       int r;
+       int call_again_soon = 0;
+       int nvec;
+       char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
+
+       mutex_lock(&con->sock_mutex);
+
+       if (con->sock == NULL) {
+               ret = -EAGAIN;
+               goto out_close;
+       }
+
+       if (con->rx_page == NULL) {
+               /*
+                * This doesn't need to be atomic, but I think it should
+                * improve performance if it is.
+                */
+               con->rx_page = alloc_page(GFP_ATOMIC);
+               if (con->rx_page == NULL)
+                       goto out_resched;
+               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;
+
+       /*
+        * 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(&con->cb) >= con->cb.base) {
+               iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb);
+               iov[1].iov_len = con->cb.base;
+               iov[1].iov_base = page_address(con->rx_page);
+               nvec = 2;
+       }
+       len = iov[0].iov_len + iov[1].iov_len;
+
+       r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len,
+                              MSG_DONTWAIT | MSG_NOSIGNAL);
+       if (ret <= 0)
+               goto out_close;
+
+       /* Process SCTP notifications */
+       if (msg.msg_flags & MSG_NOTIFICATION) {
+               BUG_ON(con->nodeid != 0);
+               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;
+       cbuf_add(&con->cb, ret);
+       ret = dlm_process_incoming_buffer(con->nodeid,
+                                         page_address(con->rx_page),
+                                         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);
+       }
+       if (ret < 0)
+               goto out_close;
+       cbuf_eat(&con->cb, ret);
+
+       if (cbuf_empty(&con->cb) && !call_again_soon) {
+               __free_page(con->rx_page);
+               con->rx_page = NULL;
+       }
+
+       if (call_again_soon)
+               goto out_resched;
+       mutex_unlock(&con->sock_mutex);
+       return 0;
+
+out_resched:
+       if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
+               queue_work(recv_workqueue, &con->rwork);
+       mutex_unlock(&con->sock_mutex);
+       return -EAGAIN;
+
+out_close:
+       mutex_unlock(&con->sock_mutex);
+       if (ret != -EAGAIN && !test_bit(CF_IS_OTHERCON, &con->flags)) {
+               close_connection(con, false);
+               /* Reconnect when there is something to send */
+       }
+       /* Don't return success if we really got EOF */
+       if (ret == 0)
+               ret = -EAGAIN;
+
+       return ret;
+}
+
+/* Listening socket is busy, accept a connection */
+static int tcp_accept_from_sock(struct connection *con)
+{
+       int result;
+       struct sockaddr_storage peeraddr;
+       struct socket *newsock;
+       int len;
+       int nodeid;
+       struct connection *newcon;
+       struct connection *addcon;
+
+       memset(&peeraddr, 0, sizeof(peeraddr));
+       result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
+                                 IPPROTO_TCP, &newsock);
+       if (result < 0)
+               return -ENOMEM;
+
+       mutex_lock_nested(&con->sock_mutex, 0);
+
+       result = -ENOTCONN;
+       if (con->sock == NULL)
+               goto accept_err;
+
+       newsock->type = con->sock->type;
+       newsock->ops = con->sock->ops;
+
+       result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK);
+       if (result < 0)
+               goto accept_err;
+
+       /* Get the connected socket's peer */
+       memset(&peeraddr, 0, sizeof(peeraddr));
+       if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr,
+                                 &len, 2)) {
+               result = -ECONNABORTED;
+               goto accept_err;
+       }
+
+       /* 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");
+               sock_release(newsock);
+               mutex_unlock(&con->sock_mutex);
+               return -1;
+       }
+
+       log_print("got connection from %d", nodeid);
+
+       /*  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.
+        *  In this case we store the incoming one in "othercon"
+        */
+       newcon = nodeid2con(nodeid, GFP_KERNEL);
+       if (!newcon) {
+               result = -ENOMEM;
+               goto accept_err;
+       }
+       mutex_lock_nested(&newcon->sock_mutex, 1);
+       if (newcon->sock) {
+               struct connection *othercon = newcon->othercon;
+
+               if (!othercon) {
+                       othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL);
+                       if (!othercon) {
+                               printk("dlm: failed to allocate incoming socket\n");
+                               mutex_unlock(&newcon->sock_mutex);
+                               result = -ENOMEM;
+                               goto accept_err;
+                       }
+                       othercon->nodeid = nodeid;
+                       othercon->rx_action = receive_from_sock;
+                       mutex_init(&othercon->sock_mutex);
+                       INIT_WORK(&othercon->swork, process_send_sockets);
+                       INIT_WORK(&othercon->rwork, process_recv_sockets);
+                       set_bit(CF_IS_OTHERCON, &othercon->flags);
+                       newcon->othercon = othercon;
+               }
+               othercon->sock = newsock;
+               newsock->sk->sk_user_data = othercon;
+               add_sock(newsock, othercon);
+               addcon = othercon;
+       }
+       else {
+               newsock->sk->sk_user_data = newcon;
+               newcon->rx_action = receive_from_sock;
+               add_sock(newsock, newcon);
+               addcon = newcon;
+       }
+
+       mutex_unlock(&newcon->sock_mutex);
+
+       /*
+        * Add it to the active queue in case we got data
+        * beween processing the accept adding the socket
+        * to the read_sockets list
+        */
+       if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
+               queue_work(recv_workqueue, &addcon->rwork);
+       mutex_unlock(&con->sock_mutex);
+
+       return 0;
+
+accept_err:
+       mutex_unlock(&con->sock_mutex);
+       sock_release(newsock);
+
+       if (result != -EAGAIN)
+               printk("dlm: error accepting connection from node: %d\n", 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 tcp_connect_to_sock(struct connection *con)
+{
+       int result = -EHOSTUNREACH;
+       struct sockaddr_storage saddr;
+       int addr_len;
+       struct socket *sock;
+
+       if (con->nodeid == 0) {
+               log_print("attempt to connect sock 0 foiled");
+               return;
+       }
+
+       mutex_lock(&con->sock_mutex);
+       if (con->retries++ > MAX_CONNECT_RETRIES)
+               goto out;
+
+       /* Some odd races can cause double-connects, ignore them */
+       if (con->sock) {
+               result = 0;
+               goto out;
+       }
+
+       /* Create a socket to communicate with */
+       result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
+                                 IPPROTO_TCP, &sock);
+       if (result < 0)
+               goto out_err;
+
+       memset(&saddr, 0, sizeof(saddr));
+       if (dlm_nodeid_to_addr(con->nodeid, &saddr))
+               goto out_err;
+
+       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);
+
+       log_print("connecting to %d", con->nodeid);
+       result =
+               sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
+                                  O_NONBLOCK);
+       if (result == -EINPROGRESS)
+               result = 0;
+       if (result == 0)
+               goto out;
+
+out_err:
+       if (con->sock) {
+               sock_release(con->sock);
+               con->sock = NULL;
+       }
+       /*
+        * Some errors are fatal and this list might need adjusting. For other
+        * errors we try again until the max number of retries is reached.
+        */
+       if (result != -EHOSTUNREACH && result != -ENETUNREACH &&
+           result != -ENETDOWN && result != EINVAL
+           && result != -EPROTONOSUPPORT) {
+               lowcomms_connect_sock(con);
+               result = 0;
+       }
+out:
+       mutex_unlock(&con->sock_mutex);
+       return;
+}
+
+static struct socket *tcp_create_listen_sock(struct connection *con,
+                                            struct sockaddr_storage *saddr)
+{
+       struct socket *sock = NULL;
+       int result = 0;
+       int one = 1;
+       int addr_len;
+
+       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[0]->ss_family, SOCK_STREAM, IPPROTO_TCP, &sock);
+       if (result < 0) {
+               printk("dlm: Can't create listening comms socket\n");
+               goto create_out;
+       }
+
+       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);
+       }
+       sock->sk->sk_user_data = con;
+       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);
+               sock_release(sock);
+               sock = NULL;
+               con->sock = NULL;
+               goto create_out;
+       }
+       result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
+                                (char *)&one, sizeof(one));
+       if (result < 0) {
+               printk("dlm: Set keepalive failed: %d\n", result);
+       }
+
+       result = sock->ops->listen(sock, 5);
+       if (result < 0) {
+               printk("dlm: Can't listen on port %d\n", dlm_config.ci_tcp_port);
+               sock_release(sock);
+               sock = NULL;
+               goto create_out;
+       }
+
+create_out:
+       return sock;
+}
+
+/* Get local addresses */
+static void init_local(void)
+{
+       struct sockaddr_storage sas, *addr;
+       int i;
+
+       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;
+}
+
+/* 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 = tcp_create_listen_sock(con, dlm_local_addr[0]);
+       if (sock) {
+               add_sock(sock, con);
+               result = 0;
+       }
+       else {
+               result = -EADDRINUSE;
+       }
+
+       return result;
+}
+
+
+
+static struct writequeue_entry *new_writequeue_entry(struct connection *con,
+                                                    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;
+       entry->con = con;
+
+       return entry;
+}
+
+void *dlm_lowcomms_get_buffer(int nodeid, int len,
+                             gfp_t allocation, char **ppc)
+{
+       struct connection *con;
+       struct writequeue_entry *e;
+       int offset = 0;
+       int users = 0;
+
+       con = nodeid2con(nodeid, allocation);
+       if (!con)
+               return NULL;
+
+       spin_lock(&con->writequeue_lock);
+       e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
+       if ((&e->list == &con->writequeue) ||
+           (PAGE_CACHE_SIZE - e->end < len)) {
+               e = NULL;
+       } else {
+               offset = e->end;
+               e->end += len;
+               users = e->users++;
+       }
+       spin_unlock(&con->writequeue_lock);
+
+       if (e) {
+       got_one:
+               if (users == 0)
+                       kmap(e->page);
+               *ppc = page_address(e->page) + offset;
+               return e;
+       }
+
+       e = new_writequeue_entry(con, allocation);
+       if (e) {
+               spin_lock(&con->writequeue_lock);
+               offset = e->end;
+               e->end += len;
+               users = e->users++;
+               list_add_tail(&e->list, &con->writequeue);
+               spin_unlock(&con->writequeue_lock);
+               goto got_one;
+       }
+       return NULL;
+}
+
+void dlm_lowcomms_commit_buffer(void *mh)
+{
+       struct writequeue_entry *e = (struct writequeue_entry *)mh;
+       struct connection *con = e->con;
+       int users;
+
+       spin_lock(&con->writequeue_lock);
+       users = --e->users;
+       if (users)
+               goto out;
+       e->len = e->end - e->offset;
+       kunmap(e->page);
+       spin_unlock(&con->writequeue_lock);
+
+       if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) {
+               queue_work(send_workqueue, &con->swork);
+       }
+       return;
+
+out:
+       spin_unlock(&con->writequeue_lock);
+       return;
+}
+
+/* Send a message */
+static void send_to_sock(struct connection *con)
+{
+       int ret = 0;
+       ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int);
+       const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
+       struct writequeue_entry *e;
+       int len, offset;
+
+       mutex_lock(&con->sock_mutex);
+       if (con->sock == NULL)
+               goto out_connect;
+
+       sendpage = con->sock->ops->sendpage;
+
+       spin_lock(&con->writequeue_lock);
+       for (;;) {
+               e = list_entry(con->writequeue.next, struct writequeue_entry,
+                              list);
+               if ((struct list_head *) e == &con->writequeue)
+                       break;
+
+               len = e->len;
+               offset = e->offset;
+               BUG_ON(len == 0 && e->users == 0);
+               spin_unlock(&con->writequeue_lock);
+               kmap(e->page);
+
+               ret = 0;
+               if (len) {
+                       ret = sendpage(con->sock, e->page, offset, len,
+                                      msg_flags);
+                       if (ret == -EAGAIN || ret == 0)
+                               goto out;
+                       if (ret <= 0)
+                               goto send_error;
+               }
+               else {
+                       /* Don't starve people filling buffers */
+                       cond_resched();
+               }
+
+               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);
+                       continue;
+               }
+       }
+       spin_unlock(&con->writequeue_lock);
+out:
+       mutex_unlock(&con->sock_mutex);
+       return;
+
+send_error:
+       mutex_unlock(&con->sock_mutex);
+       close_connection(con, false);
+       lowcomms_connect_sock(con);
+       return;
+
+out_connect:
+       mutex_unlock(&con->sock_mutex);
+       if (!test_bit(CF_INIT_PENDING, &con->flags))
+               lowcomms_connect_sock(con);
+       return;
+}
+
+static void clean_one_writequeue(struct connection *con)
+{
+       struct list_head *list;
+       struct list_head *temp;
+
+       spin_lock(&con->writequeue_lock);
+       list_for_each_safe(list, temp, &con->writequeue) {
+               struct writequeue_entry *e =
+                       list_entry(list, struct writequeue_entry, list);
+               list_del(&e->list);
+               free_entry(e);
+       }
+       spin_unlock(&con->writequeue_lock);
+}
+
+/* Called from recovery when it knows that a node has
+   left the cluster */
+int dlm_lowcomms_close(int nodeid)
+{
+       struct connection *con;
+
+       log_print("closing connection to node %d", nodeid);
+       con = nodeid2con(nodeid, 0);
+       if (con) {
+               clean_one_writequeue(con);
+               close_connection(con, true);
+       }
+       return 0;
+}
+
+/* Receive workqueue function */
+static void process_recv_sockets(struct work_struct *work)
+{
+       struct connection *con = container_of(work, struct connection, rwork);
+       int err;
+
+       clear_bit(CF_READ_PENDING, &con->flags);
+       do {
+               err = con->rx_action(con);
+       } 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)) {
+               con->connect_action(con);
+       }
+       clear_bit(CF_WRITE_PENDING, &con->flags);
+       send_to_sock(con);
+}
+
+
+/* Discard all entries on the write queues */
+static void clean_writequeues(void)
+{
+       int nodeid;
+
+       for (nodeid = 1; nodeid < max_nodeid; nodeid++) {
+               struct connection *con = nodeid2con(nodeid, 0);
+
+               if (con)
+                       clean_one_writequeue(con);
+       }
+}
+
+static void work_stop(void)
+{
+       destroy_workqueue(recv_workqueue);
+       destroy_workqueue(send_workqueue);
+}
+
+static int work_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;
+       }
+
+       return 0;
+}
+
+void dlm_lowcomms_stop(void)
+{
+       int i;
+       struct connection *con;
+
+       /* Set all the flags to prevent any
+          socket activity.
+       */
+       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 < 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);
+               }
+       }
+       up(&connections_lock);
+       kmem_cache_destroy(con_cache);
+}
+
+int dlm_lowcomms_start(void)
+{
+       int error = -EINVAL;
+       struct connection *con;
+
+       init_local();
+       if (!dlm_local_count) {
+               log_print("no local IP address has been set");
+               goto out;
+       }
+
+       error = -ENOMEM;
+       con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
+                                     __alignof__(struct connection), 0,
+                                     NULL, NULL);
+       if (!con_cache)
+               goto out;
+
+       /* Set some sysctl minima */
+       if (sysctl_rmem_max < NEEDED_RMEM)
+               sysctl_rmem_max = NEEDED_RMEM;
+
+       /* Start listening */
+       if (dlm_config.ci_protocol == 0)
+               error = tcp_listen_for_all();
+       else
+               error = sctp_listen_for_all();
+       if (error)
+               goto fail_unlisten;
+
+       error = work_start();
+       if (error)
+               goto fail_unlisten;
+
+       return 0;
+
+fail_unlisten:
+       con = nodeid2con(0,0);
+       if (con) {
+               close_connection(con, false);
+               kmem_cache_free(con_cache, con);
+       }
+       kmem_cache_destroy(con_cache);
+
+out:
+       return error;
+}