16ed7257bSPatrick Caulfield /****************************************************************************** 26ed7257bSPatrick Caulfield ******************************************************************************* 36ed7257bSPatrick Caulfield ** 46ed7257bSPatrick Caulfield ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 56ed7257bSPatrick Caulfield ** Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. 66ed7257bSPatrick Caulfield ** 76ed7257bSPatrick Caulfield ** This copyrighted material is made available to anyone wishing to use, 86ed7257bSPatrick Caulfield ** modify, copy, or redistribute it subject to the terms and conditions 96ed7257bSPatrick Caulfield ** of the GNU General Public License v.2. 106ed7257bSPatrick Caulfield ** 116ed7257bSPatrick Caulfield ******************************************************************************* 126ed7257bSPatrick Caulfield ******************************************************************************/ 136ed7257bSPatrick Caulfield 146ed7257bSPatrick Caulfield /* 156ed7257bSPatrick Caulfield * lowcomms.c 166ed7257bSPatrick Caulfield * 176ed7257bSPatrick Caulfield * This is the "low-level" comms layer. 186ed7257bSPatrick Caulfield * 196ed7257bSPatrick Caulfield * It is responsible for sending/receiving messages 206ed7257bSPatrick Caulfield * from other nodes in the cluster. 216ed7257bSPatrick Caulfield * 226ed7257bSPatrick Caulfield * Cluster nodes are referred to by their nodeids. nodeids are 236ed7257bSPatrick Caulfield * simply 32 bit numbers to the locking module - if they need to 246ed7257bSPatrick Caulfield * be expanded for the cluster infrastructure then that is it's 256ed7257bSPatrick Caulfield * responsibility. It is this layer's 266ed7257bSPatrick Caulfield * responsibility to resolve these into IP address or 276ed7257bSPatrick Caulfield * whatever it needs for inter-node communication. 286ed7257bSPatrick Caulfield * 296ed7257bSPatrick Caulfield * The comms level is two kernel threads that deal mainly with 306ed7257bSPatrick Caulfield * the receiving of messages from other nodes and passing them 316ed7257bSPatrick Caulfield * up to the mid-level comms layer (which understands the 326ed7257bSPatrick Caulfield * message format) for execution by the locking core, and 336ed7257bSPatrick Caulfield * a send thread which does all the setting up of connections 346ed7257bSPatrick Caulfield * to remote nodes and the sending of data. Threads are not allowed 356ed7257bSPatrick Caulfield * to send their own data because it may cause them to wait in times 366ed7257bSPatrick Caulfield * of high load. Also, this way, the sending thread can collect together 376ed7257bSPatrick Caulfield * messages bound for one node and send them in one block. 386ed7257bSPatrick Caulfield * 396ed7257bSPatrick Caulfield * lowcomms will choose to use wither TCP or SCTP as its transport layer 406ed7257bSPatrick Caulfield * depending on the configuration variable 'protocol'. This should be set 416ed7257bSPatrick Caulfield * to 0 (default) for TCP or 1 for SCTP. It shouldbe configured using a 426ed7257bSPatrick Caulfield * cluster-wide mechanism as it must be the same on all nodes of the cluster 436ed7257bSPatrick Caulfield * for the DLM to function. 446ed7257bSPatrick Caulfield * 456ed7257bSPatrick Caulfield */ 466ed7257bSPatrick Caulfield 476ed7257bSPatrick Caulfield #include <asm/ioctls.h> 486ed7257bSPatrick Caulfield #include <net/sock.h> 496ed7257bSPatrick Caulfield #include <net/tcp.h> 506ed7257bSPatrick Caulfield #include <linux/pagemap.h> 516ed7257bSPatrick Caulfield #include <linux/idr.h> 526ed7257bSPatrick Caulfield #include <linux/file.h> 536ed7257bSPatrick Caulfield #include <linux/sctp.h> 546ed7257bSPatrick Caulfield #include <net/sctp/user.h> 556ed7257bSPatrick Caulfield 566ed7257bSPatrick Caulfield #include "dlm_internal.h" 576ed7257bSPatrick Caulfield #include "lowcomms.h" 586ed7257bSPatrick Caulfield #include "midcomms.h" 596ed7257bSPatrick Caulfield #include "config.h" 606ed7257bSPatrick Caulfield 616ed7257bSPatrick Caulfield #define NEEDED_RMEM (4*1024*1024) 626ed7257bSPatrick Caulfield 636ed7257bSPatrick Caulfield struct cbuf { 646ed7257bSPatrick Caulfield unsigned int base; 656ed7257bSPatrick Caulfield unsigned int len; 666ed7257bSPatrick Caulfield unsigned int mask; 676ed7257bSPatrick Caulfield }; 686ed7257bSPatrick Caulfield 696ed7257bSPatrick Caulfield static void cbuf_add(struct cbuf *cb, int n) 706ed7257bSPatrick Caulfield { 716ed7257bSPatrick Caulfield cb->len += n; 726ed7257bSPatrick Caulfield } 736ed7257bSPatrick Caulfield 746ed7257bSPatrick Caulfield static int cbuf_data(struct cbuf *cb) 756ed7257bSPatrick Caulfield { 766ed7257bSPatrick Caulfield return ((cb->base + cb->len) & cb->mask); 776ed7257bSPatrick Caulfield } 786ed7257bSPatrick Caulfield 796ed7257bSPatrick Caulfield static void cbuf_init(struct cbuf *cb, int size) 806ed7257bSPatrick Caulfield { 816ed7257bSPatrick Caulfield cb->base = cb->len = 0; 826ed7257bSPatrick Caulfield cb->mask = size-1; 836ed7257bSPatrick Caulfield } 846ed7257bSPatrick Caulfield 856ed7257bSPatrick Caulfield static void cbuf_eat(struct cbuf *cb, int n) 866ed7257bSPatrick Caulfield { 876ed7257bSPatrick Caulfield cb->len -= n; 886ed7257bSPatrick Caulfield cb->base += n; 896ed7257bSPatrick Caulfield cb->base &= cb->mask; 906ed7257bSPatrick Caulfield } 916ed7257bSPatrick Caulfield 926ed7257bSPatrick Caulfield static bool cbuf_empty(struct cbuf *cb) 936ed7257bSPatrick Caulfield { 946ed7257bSPatrick Caulfield return cb->len == 0; 956ed7257bSPatrick Caulfield } 966ed7257bSPatrick Caulfield 976ed7257bSPatrick Caulfield struct connection { 986ed7257bSPatrick Caulfield struct socket *sock; /* NULL if not connected */ 996ed7257bSPatrick Caulfield uint32_t nodeid; /* So we know who we are in the list */ 1006ed7257bSPatrick Caulfield struct mutex sock_mutex; 1016ed7257bSPatrick Caulfield unsigned long flags; 1026ed7257bSPatrick Caulfield #define CF_READ_PENDING 1 1036ed7257bSPatrick Caulfield #define CF_WRITE_PENDING 2 1046ed7257bSPatrick Caulfield #define CF_CONNECT_PENDING 3 1056ed7257bSPatrick Caulfield #define CF_INIT_PENDING 4 1066ed7257bSPatrick Caulfield #define CF_IS_OTHERCON 5 1076ed7257bSPatrick Caulfield struct list_head writequeue; /* List of outgoing writequeue_entries */ 1086ed7257bSPatrick Caulfield spinlock_t writequeue_lock; 1096ed7257bSPatrick Caulfield int (*rx_action) (struct connection *); /* What to do when active */ 1106ed7257bSPatrick Caulfield void (*connect_action) (struct connection *); /* What to do to connect */ 1116ed7257bSPatrick Caulfield struct page *rx_page; 1126ed7257bSPatrick Caulfield struct cbuf cb; 1136ed7257bSPatrick Caulfield int retries; 1146ed7257bSPatrick Caulfield #define MAX_CONNECT_RETRIES 3 1156ed7257bSPatrick Caulfield int sctp_assoc; 1166ed7257bSPatrick Caulfield struct connection *othercon; 1176ed7257bSPatrick Caulfield struct work_struct rwork; /* Receive workqueue */ 1186ed7257bSPatrick Caulfield struct work_struct swork; /* Send workqueue */ 1196ed7257bSPatrick Caulfield }; 1206ed7257bSPatrick Caulfield #define sock2con(x) ((struct connection *)(x)->sk_user_data) 1216ed7257bSPatrick Caulfield 1226ed7257bSPatrick Caulfield /* An entry waiting to be sent */ 1236ed7257bSPatrick Caulfield struct writequeue_entry { 1246ed7257bSPatrick Caulfield struct list_head list; 1256ed7257bSPatrick Caulfield struct page *page; 1266ed7257bSPatrick Caulfield int offset; 1276ed7257bSPatrick Caulfield int len; 1286ed7257bSPatrick Caulfield int end; 1296ed7257bSPatrick Caulfield int users; 1306ed7257bSPatrick Caulfield struct connection *con; 1316ed7257bSPatrick Caulfield }; 1326ed7257bSPatrick Caulfield 1336ed7257bSPatrick Caulfield static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; 1346ed7257bSPatrick Caulfield static int dlm_local_count; 1356ed7257bSPatrick Caulfield 1366ed7257bSPatrick Caulfield /* Work queues */ 1376ed7257bSPatrick Caulfield static struct workqueue_struct *recv_workqueue; 1386ed7257bSPatrick Caulfield static struct workqueue_struct *send_workqueue; 1396ed7257bSPatrick Caulfield 1406ed7257bSPatrick Caulfield static DEFINE_IDR(connections_idr); 1416ed7257bSPatrick Caulfield static DECLARE_MUTEX(connections_lock); 1426ed7257bSPatrick Caulfield static int max_nodeid; 1436ed7257bSPatrick Caulfield static struct kmem_cache *con_cache; 1446ed7257bSPatrick Caulfield 1456ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work); 1466ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work); 1476ed7257bSPatrick Caulfield 1486ed7257bSPatrick Caulfield /* 1496ed7257bSPatrick Caulfield * If 'allocation' is zero then we don't attempt to create a new 1506ed7257bSPatrick Caulfield * connection structure for this node. 1516ed7257bSPatrick Caulfield */ 1526ed7257bSPatrick Caulfield static struct connection *__nodeid2con(int nodeid, gfp_t alloc) 1536ed7257bSPatrick Caulfield { 1546ed7257bSPatrick Caulfield struct connection *con = NULL; 1556ed7257bSPatrick Caulfield int r; 1566ed7257bSPatrick Caulfield int n; 1576ed7257bSPatrick Caulfield 1586ed7257bSPatrick Caulfield con = idr_find(&connections_idr, nodeid); 1596ed7257bSPatrick Caulfield if (con || !alloc) 1606ed7257bSPatrick Caulfield return con; 1616ed7257bSPatrick Caulfield 1626ed7257bSPatrick Caulfield r = idr_pre_get(&connections_idr, alloc); 1636ed7257bSPatrick Caulfield if (!r) 1646ed7257bSPatrick Caulfield return NULL; 1656ed7257bSPatrick Caulfield 1666ed7257bSPatrick Caulfield con = kmem_cache_zalloc(con_cache, alloc); 1676ed7257bSPatrick Caulfield if (!con) 1686ed7257bSPatrick Caulfield return NULL; 1696ed7257bSPatrick Caulfield 1706ed7257bSPatrick Caulfield r = idr_get_new_above(&connections_idr, con, nodeid, &n); 1716ed7257bSPatrick Caulfield if (r) { 1726ed7257bSPatrick Caulfield kmem_cache_free(con_cache, con); 1736ed7257bSPatrick Caulfield return NULL; 1746ed7257bSPatrick Caulfield } 1756ed7257bSPatrick Caulfield 1766ed7257bSPatrick Caulfield if (n != nodeid) { 1776ed7257bSPatrick Caulfield idr_remove(&connections_idr, n); 1786ed7257bSPatrick Caulfield kmem_cache_free(con_cache, con); 1796ed7257bSPatrick Caulfield return NULL; 1806ed7257bSPatrick Caulfield } 1816ed7257bSPatrick Caulfield 1826ed7257bSPatrick Caulfield con->nodeid = nodeid; 1836ed7257bSPatrick Caulfield mutex_init(&con->sock_mutex); 1846ed7257bSPatrick Caulfield INIT_LIST_HEAD(&con->writequeue); 1856ed7257bSPatrick Caulfield spin_lock_init(&con->writequeue_lock); 1866ed7257bSPatrick Caulfield INIT_WORK(&con->swork, process_send_sockets); 1876ed7257bSPatrick Caulfield INIT_WORK(&con->rwork, process_recv_sockets); 1886ed7257bSPatrick Caulfield 1896ed7257bSPatrick Caulfield /* Setup action pointers for child sockets */ 1906ed7257bSPatrick Caulfield if (con->nodeid) { 1916ed7257bSPatrick Caulfield struct connection *zerocon = idr_find(&connections_idr, 0); 1926ed7257bSPatrick Caulfield 1936ed7257bSPatrick Caulfield con->connect_action = zerocon->connect_action; 1946ed7257bSPatrick Caulfield if (!con->rx_action) 1956ed7257bSPatrick Caulfield con->rx_action = zerocon->rx_action; 1966ed7257bSPatrick Caulfield } 1976ed7257bSPatrick Caulfield 1986ed7257bSPatrick Caulfield if (nodeid > max_nodeid) 1996ed7257bSPatrick Caulfield max_nodeid = nodeid; 2006ed7257bSPatrick Caulfield 2016ed7257bSPatrick Caulfield return con; 2026ed7257bSPatrick Caulfield } 2036ed7257bSPatrick Caulfield 2046ed7257bSPatrick Caulfield static struct connection *nodeid2con(int nodeid, gfp_t allocation) 2056ed7257bSPatrick Caulfield { 2066ed7257bSPatrick Caulfield struct connection *con; 2076ed7257bSPatrick Caulfield 2086ed7257bSPatrick Caulfield down(&connections_lock); 2096ed7257bSPatrick Caulfield con = __nodeid2con(nodeid, allocation); 2106ed7257bSPatrick Caulfield up(&connections_lock); 2116ed7257bSPatrick Caulfield 2126ed7257bSPatrick Caulfield return con; 2136ed7257bSPatrick Caulfield } 2146ed7257bSPatrick Caulfield 2156ed7257bSPatrick Caulfield /* This is a bit drastic, but only called when things go wrong */ 2166ed7257bSPatrick Caulfield static struct connection *assoc2con(int assoc_id) 2176ed7257bSPatrick Caulfield { 2186ed7257bSPatrick Caulfield int i; 2196ed7257bSPatrick Caulfield struct connection *con; 2206ed7257bSPatrick Caulfield 2216ed7257bSPatrick Caulfield down(&connections_lock); 22230d3a237SPatrick Caulfield for (i=0; i<=max_nodeid; i++) { 2236ed7257bSPatrick Caulfield con = __nodeid2con(i, 0); 2246ed7257bSPatrick Caulfield if (con && con->sctp_assoc == assoc_id) { 2256ed7257bSPatrick Caulfield up(&connections_lock); 2266ed7257bSPatrick Caulfield return con; 2276ed7257bSPatrick Caulfield } 2286ed7257bSPatrick Caulfield } 2296ed7257bSPatrick Caulfield up(&connections_lock); 2306ed7257bSPatrick Caulfield return NULL; 2316ed7257bSPatrick Caulfield } 2326ed7257bSPatrick Caulfield 2336ed7257bSPatrick Caulfield static int nodeid_to_addr(int nodeid, struct sockaddr *retaddr) 2346ed7257bSPatrick Caulfield { 2356ed7257bSPatrick Caulfield struct sockaddr_storage addr; 2366ed7257bSPatrick Caulfield int error; 2376ed7257bSPatrick Caulfield 2386ed7257bSPatrick Caulfield if (!dlm_local_count) 2396ed7257bSPatrick Caulfield return -1; 2406ed7257bSPatrick Caulfield 2416ed7257bSPatrick Caulfield error = dlm_nodeid_to_addr(nodeid, &addr); 2426ed7257bSPatrick Caulfield if (error) 2436ed7257bSPatrick Caulfield return error; 2446ed7257bSPatrick Caulfield 2456ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) { 2466ed7257bSPatrick Caulfield struct sockaddr_in *in4 = (struct sockaddr_in *) &addr; 2476ed7257bSPatrick Caulfield struct sockaddr_in *ret4 = (struct sockaddr_in *) retaddr; 2486ed7257bSPatrick Caulfield ret4->sin_addr.s_addr = in4->sin_addr.s_addr; 2496ed7257bSPatrick Caulfield } else { 2506ed7257bSPatrick Caulfield struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &addr; 2516ed7257bSPatrick Caulfield struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) retaddr; 2526ed7257bSPatrick Caulfield memcpy(&ret6->sin6_addr, &in6->sin6_addr, 2536ed7257bSPatrick Caulfield sizeof(in6->sin6_addr)); 2546ed7257bSPatrick Caulfield } 2556ed7257bSPatrick Caulfield 2566ed7257bSPatrick Caulfield return 0; 2576ed7257bSPatrick Caulfield } 2586ed7257bSPatrick Caulfield 2596ed7257bSPatrick Caulfield /* Data available on socket or listen socket received a connect */ 2606ed7257bSPatrick Caulfield static void lowcomms_data_ready(struct sock *sk, int count_unused) 2616ed7257bSPatrick Caulfield { 2626ed7257bSPatrick Caulfield struct connection *con = sock2con(sk); 2636ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) 2646ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 2656ed7257bSPatrick Caulfield } 2666ed7257bSPatrick Caulfield 2676ed7257bSPatrick Caulfield static void lowcomms_write_space(struct sock *sk) 2686ed7257bSPatrick Caulfield { 2696ed7257bSPatrick Caulfield struct connection *con = sock2con(sk); 2706ed7257bSPatrick Caulfield 2716ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) 2726ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 2736ed7257bSPatrick Caulfield } 2746ed7257bSPatrick Caulfield 2756ed7257bSPatrick Caulfield static inline void lowcomms_connect_sock(struct connection *con) 2766ed7257bSPatrick Caulfield { 2776ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags)) 2786ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 2796ed7257bSPatrick Caulfield } 2806ed7257bSPatrick Caulfield 2816ed7257bSPatrick Caulfield static void lowcomms_state_change(struct sock *sk) 2826ed7257bSPatrick Caulfield { 2836ed7257bSPatrick Caulfield if (sk->sk_state == TCP_ESTABLISHED) 2846ed7257bSPatrick Caulfield lowcomms_write_space(sk); 2856ed7257bSPatrick Caulfield } 2866ed7257bSPatrick Caulfield 2876ed7257bSPatrick Caulfield /* Make a socket active */ 2886ed7257bSPatrick Caulfield static int add_sock(struct socket *sock, struct connection *con) 2896ed7257bSPatrick Caulfield { 2906ed7257bSPatrick Caulfield con->sock = sock; 2916ed7257bSPatrick Caulfield 2926ed7257bSPatrick Caulfield /* Install a data_ready callback */ 2936ed7257bSPatrick Caulfield con->sock->sk->sk_data_ready = lowcomms_data_ready; 2946ed7257bSPatrick Caulfield con->sock->sk->sk_write_space = lowcomms_write_space; 2956ed7257bSPatrick Caulfield con->sock->sk->sk_state_change = lowcomms_state_change; 2966ed7257bSPatrick Caulfield con->sock->sk->sk_user_data = con; 2976ed7257bSPatrick Caulfield return 0; 2986ed7257bSPatrick Caulfield } 2996ed7257bSPatrick Caulfield 3006ed7257bSPatrick Caulfield /* Add the port number to an IPv6 or 4 sockaddr and return the address 3016ed7257bSPatrick Caulfield length */ 3026ed7257bSPatrick Caulfield static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, 3036ed7257bSPatrick Caulfield int *addr_len) 3046ed7257bSPatrick Caulfield { 3056ed7257bSPatrick Caulfield saddr->ss_family = dlm_local_addr[0]->ss_family; 3066ed7257bSPatrick Caulfield if (saddr->ss_family == AF_INET) { 3076ed7257bSPatrick Caulfield struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; 3086ed7257bSPatrick Caulfield in4_addr->sin_port = cpu_to_be16(port); 3096ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in); 3106ed7257bSPatrick Caulfield memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); 3116ed7257bSPatrick Caulfield } else { 3126ed7257bSPatrick Caulfield struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; 3136ed7257bSPatrick Caulfield in6_addr->sin6_port = cpu_to_be16(port); 3146ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in6); 3156ed7257bSPatrick Caulfield } 3166ed7257bSPatrick Caulfield } 3176ed7257bSPatrick Caulfield 3186ed7257bSPatrick Caulfield /* Close a remote connection and tidy up */ 3196ed7257bSPatrick Caulfield static void close_connection(struct connection *con, bool and_other) 3206ed7257bSPatrick Caulfield { 3216ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 3226ed7257bSPatrick Caulfield 3236ed7257bSPatrick Caulfield if (con->sock) { 3246ed7257bSPatrick Caulfield sock_release(con->sock); 3256ed7257bSPatrick Caulfield con->sock = NULL; 3266ed7257bSPatrick Caulfield } 3276ed7257bSPatrick Caulfield if (con->othercon && and_other) { 3286ed7257bSPatrick Caulfield /* Will only re-enter once. */ 3296ed7257bSPatrick Caulfield close_connection(con->othercon, false); 3306ed7257bSPatrick Caulfield } 3316ed7257bSPatrick Caulfield if (con->rx_page) { 3326ed7257bSPatrick Caulfield __free_page(con->rx_page); 3336ed7257bSPatrick Caulfield con->rx_page = NULL; 3346ed7257bSPatrick Caulfield } 3356ed7257bSPatrick Caulfield con->retries = 0; 3366ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 3376ed7257bSPatrick Caulfield } 3386ed7257bSPatrick Caulfield 3396ed7257bSPatrick Caulfield /* We only send shutdown messages to nodes that are not part of the cluster */ 3406ed7257bSPatrick Caulfield static void sctp_send_shutdown(sctp_assoc_t associd) 3416ed7257bSPatrick Caulfield { 3426ed7257bSPatrick Caulfield static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; 3436ed7257bSPatrick Caulfield struct msghdr outmessage; 3446ed7257bSPatrick Caulfield struct cmsghdr *cmsg; 3456ed7257bSPatrick Caulfield struct sctp_sndrcvinfo *sinfo; 3466ed7257bSPatrick Caulfield int ret; 3476ed7257bSPatrick Caulfield struct connection *con; 3486ed7257bSPatrick Caulfield 3496ed7257bSPatrick Caulfield con = nodeid2con(0,0); 3506ed7257bSPatrick Caulfield BUG_ON(con == NULL); 3516ed7257bSPatrick Caulfield 3526ed7257bSPatrick Caulfield outmessage.msg_name = NULL; 3536ed7257bSPatrick Caulfield outmessage.msg_namelen = 0; 3546ed7257bSPatrick Caulfield outmessage.msg_control = outcmsg; 3556ed7257bSPatrick Caulfield outmessage.msg_controllen = sizeof(outcmsg); 3566ed7257bSPatrick Caulfield outmessage.msg_flags = MSG_EOR; 3576ed7257bSPatrick Caulfield 3586ed7257bSPatrick Caulfield cmsg = CMSG_FIRSTHDR(&outmessage); 3596ed7257bSPatrick Caulfield cmsg->cmsg_level = IPPROTO_SCTP; 3606ed7257bSPatrick Caulfield cmsg->cmsg_type = SCTP_SNDRCV; 3616ed7257bSPatrick Caulfield cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo)); 3626ed7257bSPatrick Caulfield outmessage.msg_controllen = cmsg->cmsg_len; 3636ed7257bSPatrick Caulfield sinfo = CMSG_DATA(cmsg); 3646ed7257bSPatrick Caulfield memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo)); 3656ed7257bSPatrick Caulfield 3666ed7257bSPatrick Caulfield sinfo->sinfo_flags |= MSG_EOF; 3676ed7257bSPatrick Caulfield sinfo->sinfo_assoc_id = associd; 3686ed7257bSPatrick Caulfield 3696ed7257bSPatrick Caulfield ret = kernel_sendmsg(con->sock, &outmessage, NULL, 0, 0); 3706ed7257bSPatrick Caulfield 3716ed7257bSPatrick Caulfield if (ret != 0) 3726ed7257bSPatrick Caulfield log_print("send EOF to node failed: %d", ret); 3736ed7257bSPatrick Caulfield } 3746ed7257bSPatrick Caulfield 3756ed7257bSPatrick Caulfield /* INIT failed but we don't know which node... 3766ed7257bSPatrick Caulfield restart INIT on all pending nodes */ 3776ed7257bSPatrick Caulfield static void sctp_init_failed(void) 3786ed7257bSPatrick Caulfield { 3796ed7257bSPatrick Caulfield int i; 3806ed7257bSPatrick Caulfield struct connection *con; 3816ed7257bSPatrick Caulfield 3826ed7257bSPatrick Caulfield down(&connections_lock); 3836ed7257bSPatrick Caulfield for (i=1; i<=max_nodeid; i++) { 3846ed7257bSPatrick Caulfield con = __nodeid2con(i, 0); 3856ed7257bSPatrick Caulfield if (!con) 3866ed7257bSPatrick Caulfield continue; 3876ed7257bSPatrick Caulfield con->sctp_assoc = 0; 3886ed7257bSPatrick Caulfield if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { 3896ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { 3906ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 3916ed7257bSPatrick Caulfield } 3926ed7257bSPatrick Caulfield } 3936ed7257bSPatrick Caulfield } 3946ed7257bSPatrick Caulfield up(&connections_lock); 3956ed7257bSPatrick Caulfield } 3966ed7257bSPatrick Caulfield 3976ed7257bSPatrick Caulfield /* Something happened to an association */ 3986ed7257bSPatrick Caulfield static void process_sctp_notification(struct connection *con, struct msghdr *msg, char *buf) 3996ed7257bSPatrick Caulfield { 4006ed7257bSPatrick Caulfield union sctp_notification *sn = (union sctp_notification *)buf; 4016ed7257bSPatrick Caulfield 4026ed7257bSPatrick Caulfield if (sn->sn_header.sn_type == SCTP_ASSOC_CHANGE) { 4036ed7257bSPatrick Caulfield switch (sn->sn_assoc_change.sac_state) { 4046ed7257bSPatrick Caulfield 4056ed7257bSPatrick Caulfield case SCTP_COMM_UP: 4066ed7257bSPatrick Caulfield case SCTP_RESTART: 4076ed7257bSPatrick Caulfield { 4086ed7257bSPatrick Caulfield /* Check that the new node is in the lockspace */ 4096ed7257bSPatrick Caulfield struct sctp_prim prim; 4106ed7257bSPatrick Caulfield int nodeid; 4116ed7257bSPatrick Caulfield int prim_len, ret; 4126ed7257bSPatrick Caulfield int addr_len; 4136ed7257bSPatrick Caulfield struct connection *new_con; 4146ed7257bSPatrick Caulfield struct file *file; 4156ed7257bSPatrick Caulfield sctp_peeloff_arg_t parg; 4166ed7257bSPatrick Caulfield int parglen = sizeof(parg); 4176ed7257bSPatrick Caulfield 4186ed7257bSPatrick Caulfield /* 4196ed7257bSPatrick Caulfield * We get this before any data for an association. 4206ed7257bSPatrick Caulfield * We verify that the node is in the cluster and 4216ed7257bSPatrick Caulfield * then peel off a socket for it. 4226ed7257bSPatrick Caulfield */ 4236ed7257bSPatrick Caulfield if ((int)sn->sn_assoc_change.sac_assoc_id <= 0) { 4246ed7257bSPatrick Caulfield log_print("COMM_UP for invalid assoc ID %d", 4256ed7257bSPatrick Caulfield (int)sn->sn_assoc_change.sac_assoc_id); 4266ed7257bSPatrick Caulfield sctp_init_failed(); 4276ed7257bSPatrick Caulfield return; 4286ed7257bSPatrick Caulfield } 4296ed7257bSPatrick Caulfield memset(&prim, 0, sizeof(struct sctp_prim)); 4306ed7257bSPatrick Caulfield prim_len = sizeof(struct sctp_prim); 4316ed7257bSPatrick Caulfield prim.ssp_assoc_id = sn->sn_assoc_change.sac_assoc_id; 4326ed7257bSPatrick Caulfield 4336ed7257bSPatrick Caulfield ret = kernel_getsockopt(con->sock, 4346ed7257bSPatrick Caulfield IPPROTO_SCTP, 4356ed7257bSPatrick Caulfield SCTP_PRIMARY_ADDR, 4366ed7257bSPatrick Caulfield (char*)&prim, 4376ed7257bSPatrick Caulfield &prim_len); 4386ed7257bSPatrick Caulfield if (ret < 0) { 4396ed7257bSPatrick Caulfield log_print("getsockopt/sctp_primary_addr on " 4406ed7257bSPatrick Caulfield "new assoc %d failed : %d", 4416ed7257bSPatrick Caulfield (int)sn->sn_assoc_change.sac_assoc_id, 4426ed7257bSPatrick Caulfield ret); 4436ed7257bSPatrick Caulfield 4446ed7257bSPatrick Caulfield /* Retry INIT later */ 4456ed7257bSPatrick Caulfield new_con = assoc2con(sn->sn_assoc_change.sac_assoc_id); 4466ed7257bSPatrick Caulfield if (new_con) 4476ed7257bSPatrick Caulfield clear_bit(CF_CONNECT_PENDING, &con->flags); 4486ed7257bSPatrick Caulfield return; 4496ed7257bSPatrick Caulfield } 4506ed7257bSPatrick Caulfield make_sockaddr(&prim.ssp_addr, 0, &addr_len); 4516ed7257bSPatrick Caulfield if (dlm_addr_to_nodeid(&prim.ssp_addr, &nodeid)) { 4526ed7257bSPatrick Caulfield int i; 4536ed7257bSPatrick Caulfield unsigned char *b=(unsigned char *)&prim.ssp_addr; 4546ed7257bSPatrick Caulfield log_print("reject connect from unknown addr"); 4556ed7257bSPatrick Caulfield for (i=0; i<sizeof(struct sockaddr_storage);i++) 4566ed7257bSPatrick Caulfield printk("%02x ", b[i]); 4576ed7257bSPatrick Caulfield printk("\n"); 4586ed7257bSPatrick Caulfield sctp_send_shutdown(prim.ssp_assoc_id); 4596ed7257bSPatrick Caulfield return; 4606ed7257bSPatrick Caulfield } 4616ed7257bSPatrick Caulfield 4626ed7257bSPatrick Caulfield new_con = nodeid2con(nodeid, GFP_KERNEL); 4636ed7257bSPatrick Caulfield if (!new_con) 4646ed7257bSPatrick Caulfield return; 4656ed7257bSPatrick Caulfield 4666ed7257bSPatrick Caulfield /* Peel off a new sock */ 4676ed7257bSPatrick Caulfield parg.associd = sn->sn_assoc_change.sac_assoc_id; 4686ed7257bSPatrick Caulfield ret = kernel_getsockopt(con->sock, IPPROTO_SCTP, SCTP_SOCKOPT_PEELOFF, 4696ed7257bSPatrick Caulfield (void *)&parg, &parglen); 47030d3a237SPatrick Caulfield if (ret) { 4716ed7257bSPatrick Caulfield log_print("Can't peel off a socket for connection %d to node %d: err=%d\n", 4726ed7257bSPatrick Caulfield parg.associd, nodeid, ret); 4736ed7257bSPatrick Caulfield } 4746ed7257bSPatrick Caulfield file = fget(parg.sd); 4756ed7257bSPatrick Caulfield new_con->sock = SOCKET_I(file->f_dentry->d_inode); 4766ed7257bSPatrick Caulfield add_sock(new_con->sock, new_con); 4776ed7257bSPatrick Caulfield fput(file); 4786ed7257bSPatrick Caulfield put_unused_fd(parg.sd); 4796ed7257bSPatrick Caulfield 4806ed7257bSPatrick Caulfield log_print("got new/restarted association %d nodeid %d", 4816ed7257bSPatrick Caulfield (int)sn->sn_assoc_change.sac_assoc_id, nodeid); 4826ed7257bSPatrick Caulfield 4836ed7257bSPatrick Caulfield /* Send any pending writes */ 4846ed7257bSPatrick Caulfield clear_bit(CF_CONNECT_PENDING, &new_con->flags); 4856ed7257bSPatrick Caulfield clear_bit(CF_INIT_PENDING, &con->flags); 4866ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_WRITE_PENDING, &new_con->flags)) { 4876ed7257bSPatrick Caulfield queue_work(send_workqueue, &new_con->swork); 4886ed7257bSPatrick Caulfield } 4896ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &new_con->flags)) 4906ed7257bSPatrick Caulfield queue_work(recv_workqueue, &new_con->rwork); 4916ed7257bSPatrick Caulfield } 4926ed7257bSPatrick Caulfield break; 4936ed7257bSPatrick Caulfield 4946ed7257bSPatrick Caulfield case SCTP_COMM_LOST: 4956ed7257bSPatrick Caulfield case SCTP_SHUTDOWN_COMP: 4966ed7257bSPatrick Caulfield { 4976ed7257bSPatrick Caulfield con = assoc2con(sn->sn_assoc_change.sac_assoc_id); 4986ed7257bSPatrick Caulfield if (con) { 4996ed7257bSPatrick Caulfield con->sctp_assoc = 0; 5006ed7257bSPatrick Caulfield } 5016ed7257bSPatrick Caulfield } 5026ed7257bSPatrick Caulfield break; 5036ed7257bSPatrick Caulfield 5046ed7257bSPatrick Caulfield /* We don't know which INIT failed, so clear the PENDING flags 5056ed7257bSPatrick Caulfield * on them all. if assoc_id is zero then it will then try 5066ed7257bSPatrick Caulfield * again */ 5076ed7257bSPatrick Caulfield 5086ed7257bSPatrick Caulfield case SCTP_CANT_STR_ASSOC: 5096ed7257bSPatrick Caulfield { 5106ed7257bSPatrick Caulfield log_print("Can't start SCTP association - retrying"); 5116ed7257bSPatrick Caulfield sctp_init_failed(); 5126ed7257bSPatrick Caulfield } 5136ed7257bSPatrick Caulfield break; 5146ed7257bSPatrick Caulfield 5156ed7257bSPatrick Caulfield default: 5166ed7257bSPatrick Caulfield log_print("unexpected SCTP assoc change id=%d state=%d", 5176ed7257bSPatrick Caulfield (int)sn->sn_assoc_change.sac_assoc_id, 5186ed7257bSPatrick Caulfield sn->sn_assoc_change.sac_state); 5196ed7257bSPatrick Caulfield } 5206ed7257bSPatrick Caulfield } 5216ed7257bSPatrick Caulfield } 5226ed7257bSPatrick Caulfield 5236ed7257bSPatrick Caulfield /* Data received from remote end */ 5246ed7257bSPatrick Caulfield static int receive_from_sock(struct connection *con) 5256ed7257bSPatrick Caulfield { 5266ed7257bSPatrick Caulfield int ret = 0; 5276ed7257bSPatrick Caulfield struct msghdr msg = {}; 5286ed7257bSPatrick Caulfield struct kvec iov[2]; 5296ed7257bSPatrick Caulfield unsigned len; 5306ed7257bSPatrick Caulfield int r; 5316ed7257bSPatrick Caulfield int call_again_soon = 0; 5326ed7257bSPatrick Caulfield int nvec; 5336ed7257bSPatrick Caulfield char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; 5346ed7257bSPatrick Caulfield 5356ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 5366ed7257bSPatrick Caulfield 5376ed7257bSPatrick Caulfield if (con->sock == NULL) { 5386ed7257bSPatrick Caulfield ret = -EAGAIN; 5396ed7257bSPatrick Caulfield goto out_close; 5406ed7257bSPatrick Caulfield } 5416ed7257bSPatrick Caulfield 5426ed7257bSPatrick Caulfield if (con->rx_page == NULL) { 5436ed7257bSPatrick Caulfield /* 5446ed7257bSPatrick Caulfield * This doesn't need to be atomic, but I think it should 5456ed7257bSPatrick Caulfield * improve performance if it is. 5466ed7257bSPatrick Caulfield */ 5476ed7257bSPatrick Caulfield con->rx_page = alloc_page(GFP_ATOMIC); 5486ed7257bSPatrick Caulfield if (con->rx_page == NULL) 5496ed7257bSPatrick Caulfield goto out_resched; 5506ed7257bSPatrick Caulfield cbuf_init(&con->cb, PAGE_CACHE_SIZE); 5516ed7257bSPatrick Caulfield } 5526ed7257bSPatrick Caulfield 5536ed7257bSPatrick Caulfield /* Only SCTP needs these really */ 5546ed7257bSPatrick Caulfield memset(&incmsg, 0, sizeof(incmsg)); 5556ed7257bSPatrick Caulfield msg.msg_control = incmsg; 5566ed7257bSPatrick Caulfield msg.msg_controllen = sizeof(incmsg); 5576ed7257bSPatrick Caulfield 5586ed7257bSPatrick Caulfield /* 5596ed7257bSPatrick Caulfield * iov[0] is the bit of the circular buffer between the current end 5606ed7257bSPatrick Caulfield * point (cb.base + cb.len) and the end of the buffer. 5616ed7257bSPatrick Caulfield */ 5626ed7257bSPatrick Caulfield iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); 5636ed7257bSPatrick Caulfield iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); 5646ed7257bSPatrick Caulfield iov[1].iov_len = 0; 5656ed7257bSPatrick Caulfield nvec = 1; 5666ed7257bSPatrick Caulfield 5676ed7257bSPatrick Caulfield /* 5686ed7257bSPatrick Caulfield * iov[1] is the bit of the circular buffer between the start of the 5696ed7257bSPatrick Caulfield * buffer and the start of the currently used section (cb.base) 5706ed7257bSPatrick Caulfield */ 5716ed7257bSPatrick Caulfield if (cbuf_data(&con->cb) >= con->cb.base) { 5726ed7257bSPatrick Caulfield iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb); 5736ed7257bSPatrick Caulfield iov[1].iov_len = con->cb.base; 5746ed7257bSPatrick Caulfield iov[1].iov_base = page_address(con->rx_page); 5756ed7257bSPatrick Caulfield nvec = 2; 5766ed7257bSPatrick Caulfield } 5776ed7257bSPatrick Caulfield len = iov[0].iov_len + iov[1].iov_len; 5786ed7257bSPatrick Caulfield 5796ed7257bSPatrick Caulfield r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len, 5806ed7257bSPatrick Caulfield MSG_DONTWAIT | MSG_NOSIGNAL); 5816ed7257bSPatrick Caulfield if (ret <= 0) 5826ed7257bSPatrick Caulfield goto out_close; 5836ed7257bSPatrick Caulfield 5846ed7257bSPatrick Caulfield /* Process SCTP notifications */ 5856ed7257bSPatrick Caulfield if (msg.msg_flags & MSG_NOTIFICATION) { 5866ed7257bSPatrick Caulfield msg.msg_control = incmsg; 5876ed7257bSPatrick Caulfield msg.msg_controllen = sizeof(incmsg); 5886ed7257bSPatrick Caulfield 5896ed7257bSPatrick Caulfield process_sctp_notification(con, &msg, 5906ed7257bSPatrick Caulfield page_address(con->rx_page) + con->cb.base); 5916ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 5926ed7257bSPatrick Caulfield return 0; 5936ed7257bSPatrick Caulfield } 5946ed7257bSPatrick Caulfield BUG_ON(con->nodeid == 0); 5956ed7257bSPatrick Caulfield 5966ed7257bSPatrick Caulfield if (ret == len) 5976ed7257bSPatrick Caulfield call_again_soon = 1; 5986ed7257bSPatrick Caulfield cbuf_add(&con->cb, ret); 5996ed7257bSPatrick Caulfield ret = dlm_process_incoming_buffer(con->nodeid, 6006ed7257bSPatrick Caulfield page_address(con->rx_page), 6016ed7257bSPatrick Caulfield con->cb.base, con->cb.len, 6026ed7257bSPatrick Caulfield PAGE_CACHE_SIZE); 6036ed7257bSPatrick Caulfield if (ret == -EBADMSG) { 6046ed7257bSPatrick Caulfield printk(KERN_INFO "dlm: lowcomms: addr=%p, base=%u, len=%u, " 6056ed7257bSPatrick Caulfield "iov_len=%u, iov_base[0]=%p, read=%d\n", 6066ed7257bSPatrick Caulfield page_address(con->rx_page), con->cb.base, con->cb.len, 6076ed7257bSPatrick Caulfield len, iov[0].iov_base, r); 6086ed7257bSPatrick Caulfield } 6096ed7257bSPatrick Caulfield if (ret < 0) 6106ed7257bSPatrick Caulfield goto out_close; 6116ed7257bSPatrick Caulfield cbuf_eat(&con->cb, ret); 6126ed7257bSPatrick Caulfield 6136ed7257bSPatrick Caulfield if (cbuf_empty(&con->cb) && !call_again_soon) { 6146ed7257bSPatrick Caulfield __free_page(con->rx_page); 6156ed7257bSPatrick Caulfield con->rx_page = NULL; 6166ed7257bSPatrick Caulfield } 6176ed7257bSPatrick Caulfield 6186ed7257bSPatrick Caulfield if (call_again_soon) 6196ed7257bSPatrick Caulfield goto out_resched; 6206ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 6216ed7257bSPatrick Caulfield return 0; 6226ed7257bSPatrick Caulfield 6236ed7257bSPatrick Caulfield out_resched: 6246ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) 6256ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 6266ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 6276ed7257bSPatrick Caulfield return -EAGAIN; 6286ed7257bSPatrick Caulfield 6296ed7257bSPatrick Caulfield out_close: 6306ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 6316ed7257bSPatrick Caulfield if (ret != -EAGAIN && !test_bit(CF_IS_OTHERCON, &con->flags)) { 6326ed7257bSPatrick Caulfield close_connection(con, false); 6336ed7257bSPatrick Caulfield /* Reconnect when there is something to send */ 6346ed7257bSPatrick Caulfield } 6356ed7257bSPatrick Caulfield /* Don't return success if we really got EOF */ 6366ed7257bSPatrick Caulfield if (ret == 0) 6376ed7257bSPatrick Caulfield ret = -EAGAIN; 6386ed7257bSPatrick Caulfield 6396ed7257bSPatrick Caulfield return ret; 6406ed7257bSPatrick Caulfield } 6416ed7257bSPatrick Caulfield 6426ed7257bSPatrick Caulfield /* Listening socket is busy, accept a connection */ 6436ed7257bSPatrick Caulfield static int tcp_accept_from_sock(struct connection *con) 6446ed7257bSPatrick Caulfield { 6456ed7257bSPatrick Caulfield int result; 6466ed7257bSPatrick Caulfield struct sockaddr_storage peeraddr; 6476ed7257bSPatrick Caulfield struct socket *newsock; 6486ed7257bSPatrick Caulfield int len; 6496ed7257bSPatrick Caulfield int nodeid; 6506ed7257bSPatrick Caulfield struct connection *newcon; 6516ed7257bSPatrick Caulfield struct connection *addcon; 6526ed7257bSPatrick Caulfield 6536ed7257bSPatrick Caulfield memset(&peeraddr, 0, sizeof(peeraddr)); 6546ed7257bSPatrick Caulfield result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, 6556ed7257bSPatrick Caulfield IPPROTO_TCP, &newsock); 6566ed7257bSPatrick Caulfield if (result < 0) 6576ed7257bSPatrick Caulfield return -ENOMEM; 6586ed7257bSPatrick Caulfield 6596ed7257bSPatrick Caulfield mutex_lock_nested(&con->sock_mutex, 0); 6606ed7257bSPatrick Caulfield 6616ed7257bSPatrick Caulfield result = -ENOTCONN; 6626ed7257bSPatrick Caulfield if (con->sock == NULL) 6636ed7257bSPatrick Caulfield goto accept_err; 6646ed7257bSPatrick Caulfield 6656ed7257bSPatrick Caulfield newsock->type = con->sock->type; 6666ed7257bSPatrick Caulfield newsock->ops = con->sock->ops; 6676ed7257bSPatrick Caulfield 6686ed7257bSPatrick Caulfield result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK); 6696ed7257bSPatrick Caulfield if (result < 0) 6706ed7257bSPatrick Caulfield goto accept_err; 6716ed7257bSPatrick Caulfield 6726ed7257bSPatrick Caulfield /* Get the connected socket's peer */ 6736ed7257bSPatrick Caulfield memset(&peeraddr, 0, sizeof(peeraddr)); 6746ed7257bSPatrick Caulfield if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 6756ed7257bSPatrick Caulfield &len, 2)) { 6766ed7257bSPatrick Caulfield result = -ECONNABORTED; 6776ed7257bSPatrick Caulfield goto accept_err; 6786ed7257bSPatrick Caulfield } 6796ed7257bSPatrick Caulfield 6806ed7257bSPatrick Caulfield /* Get the new node's NODEID */ 6816ed7257bSPatrick Caulfield make_sockaddr(&peeraddr, 0, &len); 6826ed7257bSPatrick Caulfield if (dlm_addr_to_nodeid(&peeraddr, &nodeid)) { 6836ed7257bSPatrick Caulfield printk("dlm: connect from non cluster node\n"); 6846ed7257bSPatrick Caulfield sock_release(newsock); 6856ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 6866ed7257bSPatrick Caulfield return -1; 6876ed7257bSPatrick Caulfield } 6886ed7257bSPatrick Caulfield 6896ed7257bSPatrick Caulfield log_print("got connection from %d", nodeid); 6906ed7257bSPatrick Caulfield 6916ed7257bSPatrick Caulfield /* Check to see if we already have a connection to this node. This 6926ed7257bSPatrick Caulfield * could happen if the two nodes initiate a connection at roughly 6936ed7257bSPatrick Caulfield * the same time and the connections cross on the wire. 6946ed7257bSPatrick Caulfield * In this case we store the incoming one in "othercon" 6956ed7257bSPatrick Caulfield */ 6966ed7257bSPatrick Caulfield newcon = nodeid2con(nodeid, GFP_KERNEL); 6976ed7257bSPatrick Caulfield if (!newcon) { 6986ed7257bSPatrick Caulfield result = -ENOMEM; 6996ed7257bSPatrick Caulfield goto accept_err; 7006ed7257bSPatrick Caulfield } 7016ed7257bSPatrick Caulfield mutex_lock_nested(&newcon->sock_mutex, 1); 7026ed7257bSPatrick Caulfield if (newcon->sock) { 7036ed7257bSPatrick Caulfield struct connection *othercon = newcon->othercon; 7046ed7257bSPatrick Caulfield 7056ed7257bSPatrick Caulfield if (!othercon) { 7066ed7257bSPatrick Caulfield othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL); 7076ed7257bSPatrick Caulfield if (!othercon) { 7086ed7257bSPatrick Caulfield printk("dlm: failed to allocate incoming socket\n"); 7096ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 7106ed7257bSPatrick Caulfield result = -ENOMEM; 7116ed7257bSPatrick Caulfield goto accept_err; 7126ed7257bSPatrick Caulfield } 7136ed7257bSPatrick Caulfield othercon->nodeid = nodeid; 7146ed7257bSPatrick Caulfield othercon->rx_action = receive_from_sock; 7156ed7257bSPatrick Caulfield mutex_init(&othercon->sock_mutex); 7166ed7257bSPatrick Caulfield INIT_WORK(&othercon->swork, process_send_sockets); 7176ed7257bSPatrick Caulfield INIT_WORK(&othercon->rwork, process_recv_sockets); 7186ed7257bSPatrick Caulfield set_bit(CF_IS_OTHERCON, &othercon->flags); 7196ed7257bSPatrick Caulfield newcon->othercon = othercon; 7206ed7257bSPatrick Caulfield } 7216ed7257bSPatrick Caulfield othercon->sock = newsock; 7226ed7257bSPatrick Caulfield newsock->sk->sk_user_data = othercon; 7236ed7257bSPatrick Caulfield add_sock(newsock, othercon); 7246ed7257bSPatrick Caulfield addcon = othercon; 7256ed7257bSPatrick Caulfield } 7266ed7257bSPatrick Caulfield else { 7276ed7257bSPatrick Caulfield newsock->sk->sk_user_data = newcon; 7286ed7257bSPatrick Caulfield newcon->rx_action = receive_from_sock; 7296ed7257bSPatrick Caulfield add_sock(newsock, newcon); 7306ed7257bSPatrick Caulfield addcon = newcon; 7316ed7257bSPatrick Caulfield } 7326ed7257bSPatrick Caulfield 7336ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 7346ed7257bSPatrick Caulfield 7356ed7257bSPatrick Caulfield /* 7366ed7257bSPatrick Caulfield * Add it to the active queue in case we got data 7376ed7257bSPatrick Caulfield * beween processing the accept adding the socket 7386ed7257bSPatrick Caulfield * to the read_sockets list 7396ed7257bSPatrick Caulfield */ 7406ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) 7416ed7257bSPatrick Caulfield queue_work(recv_workqueue, &addcon->rwork); 7426ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7436ed7257bSPatrick Caulfield 7446ed7257bSPatrick Caulfield return 0; 7456ed7257bSPatrick Caulfield 7466ed7257bSPatrick Caulfield accept_err: 7476ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7486ed7257bSPatrick Caulfield sock_release(newsock); 7496ed7257bSPatrick Caulfield 7506ed7257bSPatrick Caulfield if (result != -EAGAIN) 7516ed7257bSPatrick Caulfield printk("dlm: error accepting connection from node: %d\n", result); 7526ed7257bSPatrick Caulfield return result; 7536ed7257bSPatrick Caulfield } 7546ed7257bSPatrick Caulfield 7556ed7257bSPatrick Caulfield static void free_entry(struct writequeue_entry *e) 7566ed7257bSPatrick Caulfield { 7576ed7257bSPatrick Caulfield __free_page(e->page); 7586ed7257bSPatrick Caulfield kfree(e); 7596ed7257bSPatrick Caulfield } 7606ed7257bSPatrick Caulfield 7616ed7257bSPatrick Caulfield /* Initiate an SCTP association. 7626ed7257bSPatrick Caulfield This is a special case of send_to_sock() in that we don't yet have a 7636ed7257bSPatrick Caulfield peeled-off socket for this association, so we use the listening socket 7646ed7257bSPatrick Caulfield and add the primary IP address of the remote node. 7656ed7257bSPatrick Caulfield */ 7666ed7257bSPatrick Caulfield static void sctp_init_assoc(struct connection *con) 7676ed7257bSPatrick Caulfield { 7686ed7257bSPatrick Caulfield struct sockaddr_storage rem_addr; 7696ed7257bSPatrick Caulfield char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; 7706ed7257bSPatrick Caulfield struct msghdr outmessage; 7716ed7257bSPatrick Caulfield struct cmsghdr *cmsg; 7726ed7257bSPatrick Caulfield struct sctp_sndrcvinfo *sinfo; 7736ed7257bSPatrick Caulfield struct connection *base_con; 7746ed7257bSPatrick Caulfield struct writequeue_entry *e; 7756ed7257bSPatrick Caulfield int len, offset; 7766ed7257bSPatrick Caulfield int ret; 7776ed7257bSPatrick Caulfield int addrlen; 7786ed7257bSPatrick Caulfield struct kvec iov[1]; 7796ed7257bSPatrick Caulfield 7806ed7257bSPatrick Caulfield if (test_and_set_bit(CF_INIT_PENDING, &con->flags)) 7816ed7257bSPatrick Caulfield return; 7826ed7257bSPatrick Caulfield 7836ed7257bSPatrick Caulfield if (con->retries++ > MAX_CONNECT_RETRIES) 7846ed7257bSPatrick Caulfield return; 7856ed7257bSPatrick Caulfield 7866ed7257bSPatrick Caulfield log_print("Initiating association with node %d", con->nodeid); 7876ed7257bSPatrick Caulfield 7886ed7257bSPatrick Caulfield if (nodeid_to_addr(con->nodeid, (struct sockaddr *)&rem_addr)) { 7896ed7257bSPatrick Caulfield log_print("no address for nodeid %d", con->nodeid); 7906ed7257bSPatrick Caulfield return; 7916ed7257bSPatrick Caulfield } 7926ed7257bSPatrick Caulfield base_con = nodeid2con(0, 0); 7936ed7257bSPatrick Caulfield BUG_ON(base_con == NULL); 7946ed7257bSPatrick Caulfield 7956ed7257bSPatrick Caulfield make_sockaddr(&rem_addr, dlm_config.ci_tcp_port, &addrlen); 7966ed7257bSPatrick Caulfield 7976ed7257bSPatrick Caulfield outmessage.msg_name = &rem_addr; 7986ed7257bSPatrick Caulfield outmessage.msg_namelen = addrlen; 7996ed7257bSPatrick Caulfield outmessage.msg_control = outcmsg; 8006ed7257bSPatrick Caulfield outmessage.msg_controllen = sizeof(outcmsg); 8016ed7257bSPatrick Caulfield outmessage.msg_flags = MSG_EOR; 8026ed7257bSPatrick Caulfield 8036ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 8046ed7257bSPatrick Caulfield e = list_entry(con->writequeue.next, struct writequeue_entry, 8056ed7257bSPatrick Caulfield list); 8066ed7257bSPatrick Caulfield 8076ed7257bSPatrick Caulfield BUG_ON((struct list_head *) e == &con->writequeue); 8086ed7257bSPatrick Caulfield 8096ed7257bSPatrick Caulfield len = e->len; 8106ed7257bSPatrick Caulfield offset = e->offset; 8116ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 8126ed7257bSPatrick Caulfield kmap(e->page); 8136ed7257bSPatrick Caulfield 8146ed7257bSPatrick Caulfield /* Send the first block off the write queue */ 8156ed7257bSPatrick Caulfield iov[0].iov_base = page_address(e->page)+offset; 8166ed7257bSPatrick Caulfield iov[0].iov_len = len; 8176ed7257bSPatrick Caulfield 8186ed7257bSPatrick Caulfield cmsg = CMSG_FIRSTHDR(&outmessage); 8196ed7257bSPatrick Caulfield cmsg->cmsg_level = IPPROTO_SCTP; 8206ed7257bSPatrick Caulfield cmsg->cmsg_type = SCTP_SNDRCV; 8216ed7257bSPatrick Caulfield cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo)); 8226ed7257bSPatrick Caulfield sinfo = CMSG_DATA(cmsg); 8236ed7257bSPatrick Caulfield memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo)); 8246ed7257bSPatrick Caulfield sinfo->sinfo_ppid = cpu_to_le32(dlm_our_nodeid()); 8256ed7257bSPatrick Caulfield outmessage.msg_controllen = cmsg->cmsg_len; 8266ed7257bSPatrick Caulfield 8276ed7257bSPatrick Caulfield ret = kernel_sendmsg(base_con->sock, &outmessage, iov, 1, len); 8286ed7257bSPatrick Caulfield if (ret < 0) { 8296ed7257bSPatrick Caulfield log_print("Send first packet to node %d failed: %d", con->nodeid, ret); 8306ed7257bSPatrick Caulfield 8316ed7257bSPatrick Caulfield /* Try again later */ 8326ed7257bSPatrick Caulfield clear_bit(CF_CONNECT_PENDING, &con->flags); 8336ed7257bSPatrick Caulfield clear_bit(CF_INIT_PENDING, &con->flags); 8346ed7257bSPatrick Caulfield } 8356ed7257bSPatrick Caulfield else { 8366ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 8376ed7257bSPatrick Caulfield e->offset += ret; 8386ed7257bSPatrick Caulfield e->len -= ret; 8396ed7257bSPatrick Caulfield 8406ed7257bSPatrick Caulfield if (e->len == 0 && e->users == 0) { 8416ed7257bSPatrick Caulfield list_del(&e->list); 8426ed7257bSPatrick Caulfield kunmap(e->page); 8436ed7257bSPatrick Caulfield free_entry(e); 8446ed7257bSPatrick Caulfield } 8456ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 8466ed7257bSPatrick Caulfield } 8476ed7257bSPatrick Caulfield } 8486ed7257bSPatrick Caulfield 8496ed7257bSPatrick Caulfield /* Connect a new socket to its peer */ 8506ed7257bSPatrick Caulfield static void tcp_connect_to_sock(struct connection *con) 8516ed7257bSPatrick Caulfield { 8526ed7257bSPatrick Caulfield int result = -EHOSTUNREACH; 8536ed7257bSPatrick Caulfield struct sockaddr_storage saddr; 8546ed7257bSPatrick Caulfield int addr_len; 8556ed7257bSPatrick Caulfield struct socket *sock; 8566ed7257bSPatrick Caulfield 8576ed7257bSPatrick Caulfield if (con->nodeid == 0) { 8586ed7257bSPatrick Caulfield log_print("attempt to connect sock 0 foiled"); 8596ed7257bSPatrick Caulfield return; 8606ed7257bSPatrick Caulfield } 8616ed7257bSPatrick Caulfield 8626ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 8636ed7257bSPatrick Caulfield if (con->retries++ > MAX_CONNECT_RETRIES) 8646ed7257bSPatrick Caulfield goto out; 8656ed7257bSPatrick Caulfield 8666ed7257bSPatrick Caulfield /* Some odd races can cause double-connects, ignore them */ 8676ed7257bSPatrick Caulfield if (con->sock) { 8686ed7257bSPatrick Caulfield result = 0; 8696ed7257bSPatrick Caulfield goto out; 8706ed7257bSPatrick Caulfield } 8716ed7257bSPatrick Caulfield 8726ed7257bSPatrick Caulfield /* Create a socket to communicate with */ 8736ed7257bSPatrick Caulfield result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, 8746ed7257bSPatrick Caulfield IPPROTO_TCP, &sock); 8756ed7257bSPatrick Caulfield if (result < 0) 8766ed7257bSPatrick Caulfield goto out_err; 8776ed7257bSPatrick Caulfield 8786ed7257bSPatrick Caulfield memset(&saddr, 0, sizeof(saddr)); 8796ed7257bSPatrick Caulfield if (dlm_nodeid_to_addr(con->nodeid, &saddr)) 8806ed7257bSPatrick Caulfield goto out_err; 8816ed7257bSPatrick Caulfield 8826ed7257bSPatrick Caulfield sock->sk->sk_user_data = con; 8836ed7257bSPatrick Caulfield con->rx_action = receive_from_sock; 8846ed7257bSPatrick Caulfield con->connect_action = tcp_connect_to_sock; 8856ed7257bSPatrick Caulfield add_sock(sock, con); 8866ed7257bSPatrick Caulfield 8876ed7257bSPatrick Caulfield make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); 8886ed7257bSPatrick Caulfield 8896ed7257bSPatrick Caulfield log_print("connecting to %d", con->nodeid); 8906ed7257bSPatrick Caulfield result = 8916ed7257bSPatrick Caulfield sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, 8926ed7257bSPatrick Caulfield O_NONBLOCK); 8936ed7257bSPatrick Caulfield if (result == -EINPROGRESS) 8946ed7257bSPatrick Caulfield result = 0; 8956ed7257bSPatrick Caulfield if (result == 0) 8966ed7257bSPatrick Caulfield goto out; 8976ed7257bSPatrick Caulfield 8986ed7257bSPatrick Caulfield out_err: 8996ed7257bSPatrick Caulfield if (con->sock) { 9006ed7257bSPatrick Caulfield sock_release(con->sock); 9016ed7257bSPatrick Caulfield con->sock = NULL; 9026ed7257bSPatrick Caulfield } 9036ed7257bSPatrick Caulfield /* 9046ed7257bSPatrick Caulfield * Some errors are fatal and this list might need adjusting. For other 9056ed7257bSPatrick Caulfield * errors we try again until the max number of retries is reached. 9066ed7257bSPatrick Caulfield */ 9076ed7257bSPatrick Caulfield if (result != -EHOSTUNREACH && result != -ENETUNREACH && 9086ed7257bSPatrick Caulfield result != -ENETDOWN && result != EINVAL 9096ed7257bSPatrick Caulfield && result != -EPROTONOSUPPORT) { 9106ed7257bSPatrick Caulfield lowcomms_connect_sock(con); 9116ed7257bSPatrick Caulfield result = 0; 9126ed7257bSPatrick Caulfield } 9136ed7257bSPatrick Caulfield out: 9146ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 9156ed7257bSPatrick Caulfield return; 9166ed7257bSPatrick Caulfield } 9176ed7257bSPatrick Caulfield 9186ed7257bSPatrick Caulfield static struct socket *tcp_create_listen_sock(struct connection *con, 9196ed7257bSPatrick Caulfield struct sockaddr_storage *saddr) 9206ed7257bSPatrick Caulfield { 9216ed7257bSPatrick Caulfield struct socket *sock = NULL; 9226ed7257bSPatrick Caulfield int result = 0; 9236ed7257bSPatrick Caulfield int one = 1; 9246ed7257bSPatrick Caulfield int addr_len; 9256ed7257bSPatrick Caulfield 9266ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) 9276ed7257bSPatrick Caulfield addr_len = sizeof(struct sockaddr_in); 9286ed7257bSPatrick Caulfield else 9296ed7257bSPatrick Caulfield addr_len = sizeof(struct sockaddr_in6); 9306ed7257bSPatrick Caulfield 9316ed7257bSPatrick Caulfield /* Create a socket to communicate with */ 9326ed7257bSPatrick Caulfield result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, IPPROTO_TCP, &sock); 9336ed7257bSPatrick Caulfield if (result < 0) { 9346ed7257bSPatrick Caulfield printk("dlm: Can't create listening comms socket\n"); 9356ed7257bSPatrick Caulfield goto create_out; 9366ed7257bSPatrick Caulfield } 9376ed7257bSPatrick Caulfield 9386ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, 9396ed7257bSPatrick Caulfield (char *)&one, sizeof(one)); 9406ed7257bSPatrick Caulfield 9416ed7257bSPatrick Caulfield if (result < 0) { 9426ed7257bSPatrick Caulfield printk("dlm: Failed to set SO_REUSEADDR on socket: result=%d\n", 9436ed7257bSPatrick Caulfield result); 9446ed7257bSPatrick Caulfield } 9456ed7257bSPatrick Caulfield sock->sk->sk_user_data = con; 9466ed7257bSPatrick Caulfield con->rx_action = tcp_accept_from_sock; 9476ed7257bSPatrick Caulfield con->connect_action = tcp_connect_to_sock; 9486ed7257bSPatrick Caulfield con->sock = sock; 9496ed7257bSPatrick Caulfield 9506ed7257bSPatrick Caulfield /* Bind to our port */ 9516ed7257bSPatrick Caulfield make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); 9526ed7257bSPatrick Caulfield result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); 9536ed7257bSPatrick Caulfield if (result < 0) { 9546ed7257bSPatrick Caulfield printk("dlm: Can't bind to port %d\n", dlm_config.ci_tcp_port); 9556ed7257bSPatrick Caulfield sock_release(sock); 9566ed7257bSPatrick Caulfield sock = NULL; 9576ed7257bSPatrick Caulfield con->sock = NULL; 9586ed7257bSPatrick Caulfield goto create_out; 9596ed7257bSPatrick Caulfield } 9606ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 9616ed7257bSPatrick Caulfield (char *)&one, sizeof(one)); 9626ed7257bSPatrick Caulfield if (result < 0) { 9636ed7257bSPatrick Caulfield printk("dlm: Set keepalive failed: %d\n", result); 9646ed7257bSPatrick Caulfield } 9656ed7257bSPatrick Caulfield 9666ed7257bSPatrick Caulfield result = sock->ops->listen(sock, 5); 9676ed7257bSPatrick Caulfield if (result < 0) { 9686ed7257bSPatrick Caulfield printk("dlm: Can't listen on port %d\n", dlm_config.ci_tcp_port); 9696ed7257bSPatrick Caulfield sock_release(sock); 9706ed7257bSPatrick Caulfield sock = NULL; 9716ed7257bSPatrick Caulfield goto create_out; 9726ed7257bSPatrick Caulfield } 9736ed7257bSPatrick Caulfield 9746ed7257bSPatrick Caulfield create_out: 9756ed7257bSPatrick Caulfield return sock; 9766ed7257bSPatrick Caulfield } 9776ed7257bSPatrick Caulfield 9786ed7257bSPatrick Caulfield /* Get local addresses */ 9796ed7257bSPatrick Caulfield static void init_local(void) 9806ed7257bSPatrick Caulfield { 9816ed7257bSPatrick Caulfield struct sockaddr_storage sas, *addr; 9826ed7257bSPatrick Caulfield int i; 9836ed7257bSPatrick Caulfield 98430d3a237SPatrick Caulfield dlm_local_count = 0; 9856ed7257bSPatrick Caulfield for (i = 0; i < DLM_MAX_ADDR_COUNT - 1; i++) { 9866ed7257bSPatrick Caulfield if (dlm_our_addr(&sas, i)) 9876ed7257bSPatrick Caulfield break; 9886ed7257bSPatrick Caulfield 9896ed7257bSPatrick Caulfield addr = kmalloc(sizeof(*addr), GFP_KERNEL); 9906ed7257bSPatrick Caulfield if (!addr) 9916ed7257bSPatrick Caulfield break; 9926ed7257bSPatrick Caulfield memcpy(addr, &sas, sizeof(*addr)); 9936ed7257bSPatrick Caulfield dlm_local_addr[dlm_local_count++] = addr; 9946ed7257bSPatrick Caulfield } 9956ed7257bSPatrick Caulfield } 9966ed7257bSPatrick Caulfield 9976ed7257bSPatrick Caulfield /* Bind to an IP address. SCTP allows multiple address so it can do multi-homing */ 9986ed7257bSPatrick Caulfield static int add_sctp_bind_addr(struct connection *sctp_con, struct sockaddr_storage *addr, int addr_len, int num) 9996ed7257bSPatrick Caulfield { 10006ed7257bSPatrick Caulfield int result = 0; 10016ed7257bSPatrick Caulfield 10026ed7257bSPatrick Caulfield if (num == 1) 10036ed7257bSPatrick Caulfield result = kernel_bind(sctp_con->sock, 10046ed7257bSPatrick Caulfield (struct sockaddr *) addr, 10056ed7257bSPatrick Caulfield addr_len); 10066ed7257bSPatrick Caulfield else 10076ed7257bSPatrick Caulfield result = kernel_setsockopt(sctp_con->sock, SOL_SCTP, 10086ed7257bSPatrick Caulfield SCTP_SOCKOPT_BINDX_ADD, 10096ed7257bSPatrick Caulfield (char *)addr, addr_len); 10106ed7257bSPatrick Caulfield 10116ed7257bSPatrick Caulfield if (result < 0) 10126ed7257bSPatrick Caulfield log_print("Can't bind to port %d addr number %d", 10136ed7257bSPatrick Caulfield dlm_config.ci_tcp_port, num); 10146ed7257bSPatrick Caulfield 10156ed7257bSPatrick Caulfield return result; 10166ed7257bSPatrick Caulfield } 10176ed7257bSPatrick Caulfield 10186ed7257bSPatrick Caulfield /* Initialise SCTP socket and bind to all interfaces */ 10196ed7257bSPatrick Caulfield static int sctp_listen_for_all(void) 10206ed7257bSPatrick Caulfield { 10216ed7257bSPatrick Caulfield struct socket *sock = NULL; 10226ed7257bSPatrick Caulfield struct sockaddr_storage localaddr; 10236ed7257bSPatrick Caulfield struct sctp_event_subscribe subscribe; 10246ed7257bSPatrick Caulfield int result = -EINVAL, num = 1, i, addr_len; 10256ed7257bSPatrick Caulfield struct connection *con = nodeid2con(0, GFP_KERNEL); 10266ed7257bSPatrick Caulfield int bufsize = NEEDED_RMEM; 10276ed7257bSPatrick Caulfield 10286ed7257bSPatrick Caulfield if (!con) 10296ed7257bSPatrick Caulfield return -ENOMEM; 10306ed7257bSPatrick Caulfield 10316ed7257bSPatrick Caulfield log_print("Using SCTP for communications"); 10326ed7257bSPatrick Caulfield 10336ed7257bSPatrick Caulfield result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_SEQPACKET, 10346ed7257bSPatrick Caulfield IPPROTO_SCTP, &sock); 10356ed7257bSPatrick Caulfield if (result < 0) { 10366ed7257bSPatrick Caulfield log_print("Can't create comms socket, check SCTP is loaded"); 10376ed7257bSPatrick Caulfield goto out; 10386ed7257bSPatrick Caulfield } 10396ed7257bSPatrick Caulfield 10406ed7257bSPatrick Caulfield /* Listen for events */ 10416ed7257bSPatrick Caulfield memset(&subscribe, 0, sizeof(subscribe)); 10426ed7257bSPatrick Caulfield subscribe.sctp_data_io_event = 1; 10436ed7257bSPatrick Caulfield subscribe.sctp_association_event = 1; 10446ed7257bSPatrick Caulfield subscribe.sctp_send_failure_event = 1; 10456ed7257bSPatrick Caulfield subscribe.sctp_shutdown_event = 1; 10466ed7257bSPatrick Caulfield subscribe.sctp_partial_delivery_event = 1; 10476ed7257bSPatrick Caulfield 10486ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, 10496ed7257bSPatrick Caulfield (char *)&bufsize, sizeof(bufsize)); 10506ed7257bSPatrick Caulfield if (result) 10516ed7257bSPatrick Caulfield log_print("Error increasing buffer space on socket: %d", result); 10526ed7257bSPatrick Caulfield 10536ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SCTP, SCTP_EVENTS, 10546ed7257bSPatrick Caulfield (char *)&subscribe, sizeof(subscribe)); 10556ed7257bSPatrick Caulfield if (result < 0) { 10566ed7257bSPatrick Caulfield log_print("Failed to set SCTP_EVENTS on socket: result=%d", 10576ed7257bSPatrick Caulfield result); 10586ed7257bSPatrick Caulfield goto create_delsock; 10596ed7257bSPatrick Caulfield } 10606ed7257bSPatrick Caulfield 10616ed7257bSPatrick Caulfield /* Init con struct */ 10626ed7257bSPatrick Caulfield sock->sk->sk_user_data = con; 10636ed7257bSPatrick Caulfield con->sock = sock; 10646ed7257bSPatrick Caulfield con->sock->sk->sk_data_ready = lowcomms_data_ready; 10656ed7257bSPatrick Caulfield con->rx_action = receive_from_sock; 10666ed7257bSPatrick Caulfield con->connect_action = sctp_init_assoc; 10676ed7257bSPatrick Caulfield 10686ed7257bSPatrick Caulfield /* Bind to all interfaces. */ 10696ed7257bSPatrick Caulfield for (i = 0; i < dlm_local_count; i++) { 10706ed7257bSPatrick Caulfield memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); 10716ed7257bSPatrick Caulfield make_sockaddr(&localaddr, dlm_config.ci_tcp_port, &addr_len); 10726ed7257bSPatrick Caulfield 10736ed7257bSPatrick Caulfield result = add_sctp_bind_addr(con, &localaddr, addr_len, num); 10746ed7257bSPatrick Caulfield if (result) 10756ed7257bSPatrick Caulfield goto create_delsock; 10766ed7257bSPatrick Caulfield ++num; 10776ed7257bSPatrick Caulfield } 10786ed7257bSPatrick Caulfield 10796ed7257bSPatrick Caulfield result = sock->ops->listen(sock, 5); 10806ed7257bSPatrick Caulfield if (result < 0) { 10816ed7257bSPatrick Caulfield log_print("Can't set socket listening"); 10826ed7257bSPatrick Caulfield goto create_delsock; 10836ed7257bSPatrick Caulfield } 10846ed7257bSPatrick Caulfield 10856ed7257bSPatrick Caulfield return 0; 10866ed7257bSPatrick Caulfield 10876ed7257bSPatrick Caulfield create_delsock: 10886ed7257bSPatrick Caulfield sock_release(sock); 10896ed7257bSPatrick Caulfield con->sock = NULL; 10906ed7257bSPatrick Caulfield out: 10916ed7257bSPatrick Caulfield return result; 10926ed7257bSPatrick Caulfield } 10936ed7257bSPatrick Caulfield 10946ed7257bSPatrick Caulfield static int tcp_listen_for_all(void) 10956ed7257bSPatrick Caulfield { 10966ed7257bSPatrick Caulfield struct socket *sock = NULL; 10976ed7257bSPatrick Caulfield struct connection *con = nodeid2con(0, GFP_KERNEL); 10986ed7257bSPatrick Caulfield int result = -EINVAL; 10996ed7257bSPatrick Caulfield 11006ed7257bSPatrick Caulfield if (!con) 11016ed7257bSPatrick Caulfield return -ENOMEM; 11026ed7257bSPatrick Caulfield 11036ed7257bSPatrick Caulfield /* We don't support multi-homed hosts */ 11046ed7257bSPatrick Caulfield if (dlm_local_addr[1] != NULL) { 11056ed7257bSPatrick Caulfield log_print("TCP protocol can't handle multi-homed hosts, try SCTP"); 11066ed7257bSPatrick Caulfield return -EINVAL; 11076ed7257bSPatrick Caulfield } 11086ed7257bSPatrick Caulfield 11096ed7257bSPatrick Caulfield log_print("Using TCP for communications"); 11106ed7257bSPatrick Caulfield 11116ed7257bSPatrick Caulfield set_bit(CF_IS_OTHERCON, &con->flags); 11126ed7257bSPatrick Caulfield 11136ed7257bSPatrick Caulfield sock = tcp_create_listen_sock(con, dlm_local_addr[0]); 11146ed7257bSPatrick Caulfield if (sock) { 11156ed7257bSPatrick Caulfield add_sock(sock, con); 11166ed7257bSPatrick Caulfield result = 0; 11176ed7257bSPatrick Caulfield } 11186ed7257bSPatrick Caulfield else { 11196ed7257bSPatrick Caulfield result = -EADDRINUSE; 11206ed7257bSPatrick Caulfield } 11216ed7257bSPatrick Caulfield 11226ed7257bSPatrick Caulfield return result; 11236ed7257bSPatrick Caulfield } 11246ed7257bSPatrick Caulfield 11256ed7257bSPatrick Caulfield 11266ed7257bSPatrick Caulfield 11276ed7257bSPatrick Caulfield static struct writequeue_entry *new_writequeue_entry(struct connection *con, 11286ed7257bSPatrick Caulfield gfp_t allocation) 11296ed7257bSPatrick Caulfield { 11306ed7257bSPatrick Caulfield struct writequeue_entry *entry; 11316ed7257bSPatrick Caulfield 11326ed7257bSPatrick Caulfield entry = kmalloc(sizeof(struct writequeue_entry), allocation); 11336ed7257bSPatrick Caulfield if (!entry) 11346ed7257bSPatrick Caulfield return NULL; 11356ed7257bSPatrick Caulfield 11366ed7257bSPatrick Caulfield entry->page = alloc_page(allocation); 11376ed7257bSPatrick Caulfield if (!entry->page) { 11386ed7257bSPatrick Caulfield kfree(entry); 11396ed7257bSPatrick Caulfield return NULL; 11406ed7257bSPatrick Caulfield } 11416ed7257bSPatrick Caulfield 11426ed7257bSPatrick Caulfield entry->offset = 0; 11436ed7257bSPatrick Caulfield entry->len = 0; 11446ed7257bSPatrick Caulfield entry->end = 0; 11456ed7257bSPatrick Caulfield entry->users = 0; 11466ed7257bSPatrick Caulfield entry->con = con; 11476ed7257bSPatrick Caulfield 11486ed7257bSPatrick Caulfield return entry; 11496ed7257bSPatrick Caulfield } 11506ed7257bSPatrick Caulfield 11516ed7257bSPatrick Caulfield void *dlm_lowcomms_get_buffer(int nodeid, int len, 11526ed7257bSPatrick Caulfield gfp_t allocation, char **ppc) 11536ed7257bSPatrick Caulfield { 11546ed7257bSPatrick Caulfield struct connection *con; 11556ed7257bSPatrick Caulfield struct writequeue_entry *e; 11566ed7257bSPatrick Caulfield int offset = 0; 11576ed7257bSPatrick Caulfield int users = 0; 11586ed7257bSPatrick Caulfield 11596ed7257bSPatrick Caulfield con = nodeid2con(nodeid, allocation); 11606ed7257bSPatrick Caulfield if (!con) 11616ed7257bSPatrick Caulfield return NULL; 11626ed7257bSPatrick Caulfield 11636ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 11646ed7257bSPatrick Caulfield e = list_entry(con->writequeue.prev, struct writequeue_entry, list); 11656ed7257bSPatrick Caulfield if ((&e->list == &con->writequeue) || 11666ed7257bSPatrick Caulfield (PAGE_CACHE_SIZE - e->end < len)) { 11676ed7257bSPatrick Caulfield e = NULL; 11686ed7257bSPatrick Caulfield } else { 11696ed7257bSPatrick Caulfield offset = e->end; 11706ed7257bSPatrick Caulfield e->end += len; 11716ed7257bSPatrick Caulfield users = e->users++; 11726ed7257bSPatrick Caulfield } 11736ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 11746ed7257bSPatrick Caulfield 11756ed7257bSPatrick Caulfield if (e) { 11766ed7257bSPatrick Caulfield got_one: 11776ed7257bSPatrick Caulfield if (users == 0) 11786ed7257bSPatrick Caulfield kmap(e->page); 11796ed7257bSPatrick Caulfield *ppc = page_address(e->page) + offset; 11806ed7257bSPatrick Caulfield return e; 11816ed7257bSPatrick Caulfield } 11826ed7257bSPatrick Caulfield 11836ed7257bSPatrick Caulfield e = new_writequeue_entry(con, allocation); 11846ed7257bSPatrick Caulfield if (e) { 11856ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 11866ed7257bSPatrick Caulfield offset = e->end; 11876ed7257bSPatrick Caulfield e->end += len; 11886ed7257bSPatrick Caulfield users = e->users++; 11896ed7257bSPatrick Caulfield list_add_tail(&e->list, &con->writequeue); 11906ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 11916ed7257bSPatrick Caulfield goto got_one; 11926ed7257bSPatrick Caulfield } 11936ed7257bSPatrick Caulfield return NULL; 11946ed7257bSPatrick Caulfield } 11956ed7257bSPatrick Caulfield 11966ed7257bSPatrick Caulfield void dlm_lowcomms_commit_buffer(void *mh) 11976ed7257bSPatrick Caulfield { 11986ed7257bSPatrick Caulfield struct writequeue_entry *e = (struct writequeue_entry *)mh; 11996ed7257bSPatrick Caulfield struct connection *con = e->con; 12006ed7257bSPatrick Caulfield int users; 12016ed7257bSPatrick Caulfield 12026ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 12036ed7257bSPatrick Caulfield users = --e->users; 12046ed7257bSPatrick Caulfield if (users) 12056ed7257bSPatrick Caulfield goto out; 12066ed7257bSPatrick Caulfield e->len = e->end - e->offset; 12076ed7257bSPatrick Caulfield kunmap(e->page); 12086ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 12096ed7257bSPatrick Caulfield 12106ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { 12116ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 12126ed7257bSPatrick Caulfield } 12136ed7257bSPatrick Caulfield return; 12146ed7257bSPatrick Caulfield 12156ed7257bSPatrick Caulfield out: 12166ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 12176ed7257bSPatrick Caulfield return; 12186ed7257bSPatrick Caulfield } 12196ed7257bSPatrick Caulfield 12206ed7257bSPatrick Caulfield /* Send a message */ 12216ed7257bSPatrick Caulfield static void send_to_sock(struct connection *con) 12226ed7257bSPatrick Caulfield { 12236ed7257bSPatrick Caulfield int ret = 0; 12246ed7257bSPatrick Caulfield ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int); 12256ed7257bSPatrick Caulfield const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; 12266ed7257bSPatrick Caulfield struct writequeue_entry *e; 12276ed7257bSPatrick Caulfield int len, offset; 12286ed7257bSPatrick Caulfield 12296ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 12306ed7257bSPatrick Caulfield if (con->sock == NULL) 12316ed7257bSPatrick Caulfield goto out_connect; 12326ed7257bSPatrick Caulfield 12336ed7257bSPatrick Caulfield sendpage = con->sock->ops->sendpage; 12346ed7257bSPatrick Caulfield 12356ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 12366ed7257bSPatrick Caulfield for (;;) { 12376ed7257bSPatrick Caulfield e = list_entry(con->writequeue.next, struct writequeue_entry, 12386ed7257bSPatrick Caulfield list); 12396ed7257bSPatrick Caulfield if ((struct list_head *) e == &con->writequeue) 12406ed7257bSPatrick Caulfield break; 12416ed7257bSPatrick Caulfield 12426ed7257bSPatrick Caulfield len = e->len; 12436ed7257bSPatrick Caulfield offset = e->offset; 12446ed7257bSPatrick Caulfield BUG_ON(len == 0 && e->users == 0); 12456ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 12466ed7257bSPatrick Caulfield kmap(e->page); 12476ed7257bSPatrick Caulfield 12486ed7257bSPatrick Caulfield ret = 0; 12496ed7257bSPatrick Caulfield if (len) { 12506ed7257bSPatrick Caulfield ret = sendpage(con->sock, e->page, offset, len, 12516ed7257bSPatrick Caulfield msg_flags); 12526ed7257bSPatrick Caulfield if (ret == -EAGAIN || ret == 0) 12536ed7257bSPatrick Caulfield goto out; 12546ed7257bSPatrick Caulfield if (ret <= 0) 12556ed7257bSPatrick Caulfield goto send_error; 12566ed7257bSPatrick Caulfield } 12576ed7257bSPatrick Caulfield else { 12586ed7257bSPatrick Caulfield /* Don't starve people filling buffers */ 12596ed7257bSPatrick Caulfield cond_resched(); 12606ed7257bSPatrick Caulfield } 12616ed7257bSPatrick Caulfield 12626ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 12636ed7257bSPatrick Caulfield e->offset += ret; 12646ed7257bSPatrick Caulfield e->len -= ret; 12656ed7257bSPatrick Caulfield 12666ed7257bSPatrick Caulfield if (e->len == 0 && e->users == 0) { 12676ed7257bSPatrick Caulfield list_del(&e->list); 12686ed7257bSPatrick Caulfield kunmap(e->page); 12696ed7257bSPatrick Caulfield free_entry(e); 12706ed7257bSPatrick Caulfield continue; 12716ed7257bSPatrick Caulfield } 12726ed7257bSPatrick Caulfield } 12736ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 12746ed7257bSPatrick Caulfield out: 12756ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 12766ed7257bSPatrick Caulfield return; 12776ed7257bSPatrick Caulfield 12786ed7257bSPatrick Caulfield send_error: 12796ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 12806ed7257bSPatrick Caulfield close_connection(con, false); 12816ed7257bSPatrick Caulfield lowcomms_connect_sock(con); 12826ed7257bSPatrick Caulfield return; 12836ed7257bSPatrick Caulfield 12846ed7257bSPatrick Caulfield out_connect: 12856ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 12866ed7257bSPatrick Caulfield if (!test_bit(CF_INIT_PENDING, &con->flags)) 12876ed7257bSPatrick Caulfield lowcomms_connect_sock(con); 12886ed7257bSPatrick Caulfield return; 12896ed7257bSPatrick Caulfield } 12906ed7257bSPatrick Caulfield 12916ed7257bSPatrick Caulfield static void clean_one_writequeue(struct connection *con) 12926ed7257bSPatrick Caulfield { 12936ed7257bSPatrick Caulfield struct list_head *list; 12946ed7257bSPatrick Caulfield struct list_head *temp; 12956ed7257bSPatrick Caulfield 12966ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 12976ed7257bSPatrick Caulfield list_for_each_safe(list, temp, &con->writequeue) { 12986ed7257bSPatrick Caulfield struct writequeue_entry *e = 12996ed7257bSPatrick Caulfield list_entry(list, struct writequeue_entry, list); 13006ed7257bSPatrick Caulfield list_del(&e->list); 13016ed7257bSPatrick Caulfield free_entry(e); 13026ed7257bSPatrick Caulfield } 13036ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 13046ed7257bSPatrick Caulfield } 13056ed7257bSPatrick Caulfield 13066ed7257bSPatrick Caulfield /* Called from recovery when it knows that a node has 13076ed7257bSPatrick Caulfield left the cluster */ 13086ed7257bSPatrick Caulfield int dlm_lowcomms_close(int nodeid) 13096ed7257bSPatrick Caulfield { 13106ed7257bSPatrick Caulfield struct connection *con; 13116ed7257bSPatrick Caulfield 13126ed7257bSPatrick Caulfield log_print("closing connection to node %d", nodeid); 13136ed7257bSPatrick Caulfield con = nodeid2con(nodeid, 0); 13146ed7257bSPatrick Caulfield if (con) { 13156ed7257bSPatrick Caulfield clean_one_writequeue(con); 13166ed7257bSPatrick Caulfield close_connection(con, true); 13176ed7257bSPatrick Caulfield } 13186ed7257bSPatrick Caulfield return 0; 13196ed7257bSPatrick Caulfield } 13206ed7257bSPatrick Caulfield 13216ed7257bSPatrick Caulfield /* Receive workqueue function */ 13226ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work) 13236ed7257bSPatrick Caulfield { 13246ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, rwork); 13256ed7257bSPatrick Caulfield int err; 13266ed7257bSPatrick Caulfield 13276ed7257bSPatrick Caulfield clear_bit(CF_READ_PENDING, &con->flags); 13286ed7257bSPatrick Caulfield do { 13296ed7257bSPatrick Caulfield err = con->rx_action(con); 13306ed7257bSPatrick Caulfield } while (!err); 13316ed7257bSPatrick Caulfield } 13326ed7257bSPatrick Caulfield 13336ed7257bSPatrick Caulfield /* Send workqueue function */ 13346ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work) 13356ed7257bSPatrick Caulfield { 13366ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, swork); 13376ed7257bSPatrick Caulfield 13386ed7257bSPatrick Caulfield if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { 13396ed7257bSPatrick Caulfield con->connect_action(con); 13406ed7257bSPatrick Caulfield } 13416ed7257bSPatrick Caulfield clear_bit(CF_WRITE_PENDING, &con->flags); 13426ed7257bSPatrick Caulfield send_to_sock(con); 13436ed7257bSPatrick Caulfield } 13446ed7257bSPatrick Caulfield 13456ed7257bSPatrick Caulfield 13466ed7257bSPatrick Caulfield /* Discard all entries on the write queues */ 13476ed7257bSPatrick Caulfield static void clean_writequeues(void) 13486ed7257bSPatrick Caulfield { 13496ed7257bSPatrick Caulfield int nodeid; 13506ed7257bSPatrick Caulfield 135130d3a237SPatrick Caulfield for (nodeid = 1; nodeid <= max_nodeid; nodeid++) { 135230d3a237SPatrick Caulfield struct connection *con = __nodeid2con(nodeid, 0); 13536ed7257bSPatrick Caulfield 13546ed7257bSPatrick Caulfield if (con) 13556ed7257bSPatrick Caulfield clean_one_writequeue(con); 13566ed7257bSPatrick Caulfield } 13576ed7257bSPatrick Caulfield } 13586ed7257bSPatrick Caulfield 13596ed7257bSPatrick Caulfield static void work_stop(void) 13606ed7257bSPatrick Caulfield { 13616ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 13626ed7257bSPatrick Caulfield destroy_workqueue(send_workqueue); 13636ed7257bSPatrick Caulfield } 13646ed7257bSPatrick Caulfield 13656ed7257bSPatrick Caulfield static int work_start(void) 13666ed7257bSPatrick Caulfield { 13676ed7257bSPatrick Caulfield int error; 13686ed7257bSPatrick Caulfield recv_workqueue = create_workqueue("dlm_recv"); 13696ed7257bSPatrick Caulfield error = IS_ERR(recv_workqueue); 13706ed7257bSPatrick Caulfield if (error) { 13716ed7257bSPatrick Caulfield log_print("can't start dlm_recv %d", error); 13726ed7257bSPatrick Caulfield return error; 13736ed7257bSPatrick Caulfield } 13746ed7257bSPatrick Caulfield 13756ed7257bSPatrick Caulfield send_workqueue = create_singlethread_workqueue("dlm_send"); 13766ed7257bSPatrick Caulfield error = IS_ERR(send_workqueue); 13776ed7257bSPatrick Caulfield if (error) { 13786ed7257bSPatrick Caulfield log_print("can't start dlm_send %d", error); 13796ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 13806ed7257bSPatrick Caulfield return error; 13816ed7257bSPatrick Caulfield } 13826ed7257bSPatrick Caulfield 13836ed7257bSPatrick Caulfield return 0; 13846ed7257bSPatrick Caulfield } 13856ed7257bSPatrick Caulfield 13866ed7257bSPatrick Caulfield void dlm_lowcomms_stop(void) 13876ed7257bSPatrick Caulfield { 13886ed7257bSPatrick Caulfield int i; 13896ed7257bSPatrick Caulfield struct connection *con; 13906ed7257bSPatrick Caulfield 13916ed7257bSPatrick Caulfield /* Set all the flags to prevent any 13926ed7257bSPatrick Caulfield socket activity. 13936ed7257bSPatrick Caulfield */ 13946ed7257bSPatrick Caulfield down(&connections_lock); 139530d3a237SPatrick Caulfield for (i = 0; i <= max_nodeid; i++) { 13966ed7257bSPatrick Caulfield con = __nodeid2con(i, 0); 13976ed7257bSPatrick Caulfield if (con) 13986ed7257bSPatrick Caulfield con->flags |= 0xFF; 13996ed7257bSPatrick Caulfield } 14006ed7257bSPatrick Caulfield up(&connections_lock); 14016ed7257bSPatrick Caulfield 14026ed7257bSPatrick Caulfield work_stop(); 14036ed7257bSPatrick Caulfield 14046ed7257bSPatrick Caulfield down(&connections_lock); 14056ed7257bSPatrick Caulfield clean_writequeues(); 14066ed7257bSPatrick Caulfield 140730d3a237SPatrick Caulfield for (i = 0; i <= max_nodeid; i++) { 14082439fe50SJosef Bacik con = __nodeid2con(i, 0); 14096ed7257bSPatrick Caulfield if (con) { 14106ed7257bSPatrick Caulfield close_connection(con, true); 14116ed7257bSPatrick Caulfield if (con->othercon) 14126ed7257bSPatrick Caulfield kmem_cache_free(con_cache, con->othercon); 14136ed7257bSPatrick Caulfield kmem_cache_free(con_cache, con); 14146ed7257bSPatrick Caulfield } 14156ed7257bSPatrick Caulfield } 141630d3a237SPatrick Caulfield max_nodeid = 0; 14176ed7257bSPatrick Caulfield up(&connections_lock); 14186ed7257bSPatrick Caulfield kmem_cache_destroy(con_cache); 141930d3a237SPatrick Caulfield idr_init(&connections_idr); 14206ed7257bSPatrick Caulfield } 14216ed7257bSPatrick Caulfield 14226ed7257bSPatrick Caulfield int dlm_lowcomms_start(void) 14236ed7257bSPatrick Caulfield { 14246ed7257bSPatrick Caulfield int error = -EINVAL; 14256ed7257bSPatrick Caulfield struct connection *con; 14266ed7257bSPatrick Caulfield 14276ed7257bSPatrick Caulfield init_local(); 14286ed7257bSPatrick Caulfield if (!dlm_local_count) { 14296ed7257bSPatrick Caulfield log_print("no local IP address has been set"); 14306ed7257bSPatrick Caulfield goto out; 14316ed7257bSPatrick Caulfield } 14326ed7257bSPatrick Caulfield 14336ed7257bSPatrick Caulfield error = -ENOMEM; 14346ed7257bSPatrick Caulfield con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), 14356ed7257bSPatrick Caulfield __alignof__(struct connection), 0, 14366ed7257bSPatrick Caulfield NULL, NULL); 14376ed7257bSPatrick Caulfield if (!con_cache) 14386ed7257bSPatrick Caulfield goto out; 14396ed7257bSPatrick Caulfield 14406ed7257bSPatrick Caulfield /* Set some sysctl minima */ 14416ed7257bSPatrick Caulfield if (sysctl_rmem_max < NEEDED_RMEM) 14426ed7257bSPatrick Caulfield sysctl_rmem_max = NEEDED_RMEM; 14436ed7257bSPatrick Caulfield 14446ed7257bSPatrick Caulfield /* Start listening */ 14456ed7257bSPatrick Caulfield if (dlm_config.ci_protocol == 0) 14466ed7257bSPatrick Caulfield error = tcp_listen_for_all(); 14476ed7257bSPatrick Caulfield else 14486ed7257bSPatrick Caulfield error = sctp_listen_for_all(); 14496ed7257bSPatrick Caulfield if (error) 14506ed7257bSPatrick Caulfield goto fail_unlisten; 14516ed7257bSPatrick Caulfield 14526ed7257bSPatrick Caulfield error = work_start(); 14536ed7257bSPatrick Caulfield if (error) 14546ed7257bSPatrick Caulfield goto fail_unlisten; 14556ed7257bSPatrick Caulfield 14566ed7257bSPatrick Caulfield return 0; 14576ed7257bSPatrick Caulfield 14586ed7257bSPatrick Caulfield fail_unlisten: 14596ed7257bSPatrick Caulfield con = nodeid2con(0,0); 14606ed7257bSPatrick Caulfield if (con) { 14616ed7257bSPatrick Caulfield close_connection(con, false); 14626ed7257bSPatrick Caulfield kmem_cache_free(con_cache, con); 14636ed7257bSPatrick Caulfield } 14646ed7257bSPatrick Caulfield kmem_cache_destroy(con_cache); 14656ed7257bSPatrick Caulfield 14666ed7257bSPatrick Caulfield out: 14676ed7257bSPatrick Caulfield return error; 14686ed7257bSPatrick Caulfield } 1469