16ed7257bSPatrick Caulfield /****************************************************************************** 26ed7257bSPatrick Caulfield ******************************************************************************* 36ed7257bSPatrick Caulfield ** 46ed7257bSPatrick Caulfield ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 55e9ccc37SChristine Caulfield ** Copyright (C) 2004-2009 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 242cf12c0bSJoe Perches * be expanded for the cluster infrastructure then that is its 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 * 392cf12c0bSJoe Perches * lowcomms will choose to use either 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 should be 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/file.h> 527a936ce7SMatthias Kaehlcke #include <linux/mutex.h> 536ed7257bSPatrick Caulfield #include <linux/sctp.h> 545a0e3ad6STejun Heo #include <linux/slab.h> 552f2d76ccSBenjamin Poirier #include <net/sctp/sctp.h> 5644ad532bSJoe Perches #include <net/ipv6.h> 576ed7257bSPatrick Caulfield 586ed7257bSPatrick Caulfield #include "dlm_internal.h" 596ed7257bSPatrick Caulfield #include "lowcomms.h" 606ed7257bSPatrick Caulfield #include "midcomms.h" 616ed7257bSPatrick Caulfield #include "config.h" 626ed7257bSPatrick Caulfield 636ed7257bSPatrick Caulfield #define NEEDED_RMEM (4*1024*1024) 645e9ccc37SChristine Caulfield #define CONN_HASH_SIZE 32 656ed7257bSPatrick Caulfield 66f92c8dd7SBob Peterson /* Number of messages to send before rescheduling */ 67f92c8dd7SBob Peterson #define MAX_SEND_MSG_COUNT 25 68f92c8dd7SBob Peterson 696ed7257bSPatrick Caulfield struct cbuf { 706ed7257bSPatrick Caulfield unsigned int base; 716ed7257bSPatrick Caulfield unsigned int len; 726ed7257bSPatrick Caulfield unsigned int mask; 736ed7257bSPatrick Caulfield }; 746ed7257bSPatrick Caulfield 756ed7257bSPatrick Caulfield static void cbuf_add(struct cbuf *cb, int n) 766ed7257bSPatrick Caulfield { 776ed7257bSPatrick Caulfield cb->len += n; 786ed7257bSPatrick Caulfield } 796ed7257bSPatrick Caulfield 806ed7257bSPatrick Caulfield static int cbuf_data(struct cbuf *cb) 816ed7257bSPatrick Caulfield { 826ed7257bSPatrick Caulfield return ((cb->base + cb->len) & cb->mask); 836ed7257bSPatrick Caulfield } 846ed7257bSPatrick Caulfield 856ed7257bSPatrick Caulfield static void cbuf_init(struct cbuf *cb, int size) 866ed7257bSPatrick Caulfield { 876ed7257bSPatrick Caulfield cb->base = cb->len = 0; 886ed7257bSPatrick Caulfield cb->mask = size-1; 896ed7257bSPatrick Caulfield } 906ed7257bSPatrick Caulfield 916ed7257bSPatrick Caulfield static void cbuf_eat(struct cbuf *cb, int n) 926ed7257bSPatrick Caulfield { 936ed7257bSPatrick Caulfield cb->len -= n; 946ed7257bSPatrick Caulfield cb->base += n; 956ed7257bSPatrick Caulfield cb->base &= cb->mask; 966ed7257bSPatrick Caulfield } 976ed7257bSPatrick Caulfield 986ed7257bSPatrick Caulfield static bool cbuf_empty(struct cbuf *cb) 996ed7257bSPatrick Caulfield { 1006ed7257bSPatrick Caulfield return cb->len == 0; 1016ed7257bSPatrick Caulfield } 1026ed7257bSPatrick Caulfield 1036ed7257bSPatrick Caulfield struct connection { 1046ed7257bSPatrick Caulfield struct socket *sock; /* NULL if not connected */ 1056ed7257bSPatrick Caulfield uint32_t nodeid; /* So we know who we are in the list */ 1066ed7257bSPatrick Caulfield struct mutex sock_mutex; 1076ed7257bSPatrick Caulfield unsigned long flags; 1086ed7257bSPatrick Caulfield #define CF_READ_PENDING 1 1098a4abb08Stsutomu.owa@toshiba.co.jp #define CF_WRITE_PENDING 2 1106ed7257bSPatrick Caulfield #define CF_INIT_PENDING 4 1116ed7257bSPatrick Caulfield #define CF_IS_OTHERCON 5 112063c4c99SLars Marowsky-Bree #define CF_CLOSE 6 113b36930ddSDavid Miller #define CF_APP_LIMITED 7 114b2a66629Stsutomu.owa@toshiba.co.jp #define CF_CLOSING 8 1156ed7257bSPatrick Caulfield struct list_head writequeue; /* List of outgoing writequeue_entries */ 1166ed7257bSPatrick Caulfield spinlock_t writequeue_lock; 1176ed7257bSPatrick Caulfield int (*rx_action) (struct connection *); /* What to do when active */ 1186ed7257bSPatrick Caulfield void (*connect_action) (struct connection *); /* What to do to connect */ 1196ed7257bSPatrick Caulfield struct page *rx_page; 1206ed7257bSPatrick Caulfield struct cbuf cb; 1216ed7257bSPatrick Caulfield int retries; 1226ed7257bSPatrick Caulfield #define MAX_CONNECT_RETRIES 3 1235e9ccc37SChristine Caulfield struct hlist_node list; 1246ed7257bSPatrick Caulfield struct connection *othercon; 1256ed7257bSPatrick Caulfield struct work_struct rwork; /* Receive workqueue */ 1266ed7257bSPatrick Caulfield struct work_struct swork; /* Send workqueue */ 1276ed7257bSPatrick Caulfield }; 1286ed7257bSPatrick Caulfield #define sock2con(x) ((struct connection *)(x)->sk_user_data) 1296ed7257bSPatrick Caulfield 1306ed7257bSPatrick Caulfield /* An entry waiting to be sent */ 1316ed7257bSPatrick Caulfield struct writequeue_entry { 1326ed7257bSPatrick Caulfield struct list_head list; 1336ed7257bSPatrick Caulfield struct page *page; 1346ed7257bSPatrick Caulfield int offset; 1356ed7257bSPatrick Caulfield int len; 1366ed7257bSPatrick Caulfield int end; 1376ed7257bSPatrick Caulfield int users; 1386ed7257bSPatrick Caulfield struct connection *con; 1396ed7257bSPatrick Caulfield }; 1406ed7257bSPatrick Caulfield 14136b71a8bSDavid Teigland struct dlm_node_addr { 14236b71a8bSDavid Teigland struct list_head list; 14336b71a8bSDavid Teigland int nodeid; 14436b71a8bSDavid Teigland int addr_count; 14598e1b60eSMike Christie int curr_addr_index; 14636b71a8bSDavid Teigland struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT]; 14736b71a8bSDavid Teigland }; 14836b71a8bSDavid Teigland 149cc661fc9SBob Peterson static struct listen_sock_callbacks { 150cc661fc9SBob Peterson void (*sk_error_report)(struct sock *); 151cc661fc9SBob Peterson void (*sk_data_ready)(struct sock *); 152cc661fc9SBob Peterson void (*sk_state_change)(struct sock *); 153cc661fc9SBob Peterson void (*sk_write_space)(struct sock *); 154cc661fc9SBob Peterson } listen_sock; 155cc661fc9SBob Peterson 15636b71a8bSDavid Teigland static LIST_HEAD(dlm_node_addrs); 15736b71a8bSDavid Teigland static DEFINE_SPINLOCK(dlm_node_addrs_spin); 15836b71a8bSDavid Teigland 1596ed7257bSPatrick Caulfield static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; 1606ed7257bSPatrick Caulfield static int dlm_local_count; 161513ef596SDavid Teigland static int dlm_allow_conn; 1626ed7257bSPatrick Caulfield 1636ed7257bSPatrick Caulfield /* Work queues */ 1646ed7257bSPatrick Caulfield static struct workqueue_struct *recv_workqueue; 1656ed7257bSPatrick Caulfield static struct workqueue_struct *send_workqueue; 1666ed7257bSPatrick Caulfield 1675e9ccc37SChristine Caulfield static struct hlist_head connection_hash[CONN_HASH_SIZE]; 1687a936ce7SMatthias Kaehlcke static DEFINE_MUTEX(connections_lock); 1696ed7257bSPatrick Caulfield static struct kmem_cache *con_cache; 1706ed7257bSPatrick Caulfield 1716ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work); 1726ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work); 1736ed7257bSPatrick Caulfield 1745e9ccc37SChristine Caulfield 1755e9ccc37SChristine Caulfield /* This is deliberately very simple because most clusters have simple 1765e9ccc37SChristine Caulfield sequential nodeids, so we should be able to go straight to a connection 1775e9ccc37SChristine Caulfield struct in the array */ 1785e9ccc37SChristine Caulfield static inline int nodeid_hash(int nodeid) 1795e9ccc37SChristine Caulfield { 1805e9ccc37SChristine Caulfield return nodeid & (CONN_HASH_SIZE-1); 1815e9ccc37SChristine Caulfield } 1825e9ccc37SChristine Caulfield 1835e9ccc37SChristine Caulfield static struct connection *__find_con(int nodeid) 1845e9ccc37SChristine Caulfield { 1855e9ccc37SChristine Caulfield int r; 1865e9ccc37SChristine Caulfield struct connection *con; 1875e9ccc37SChristine Caulfield 1885e9ccc37SChristine Caulfield r = nodeid_hash(nodeid); 1895e9ccc37SChristine Caulfield 190b67bfe0dSSasha Levin hlist_for_each_entry(con, &connection_hash[r], list) { 1915e9ccc37SChristine Caulfield if (con->nodeid == nodeid) 1925e9ccc37SChristine Caulfield return con; 1935e9ccc37SChristine Caulfield } 1945e9ccc37SChristine Caulfield return NULL; 1955e9ccc37SChristine Caulfield } 1965e9ccc37SChristine Caulfield 1976ed7257bSPatrick Caulfield /* 1986ed7257bSPatrick Caulfield * If 'allocation' is zero then we don't attempt to create a new 1996ed7257bSPatrick Caulfield * connection structure for this node. 2006ed7257bSPatrick Caulfield */ 2016ed7257bSPatrick Caulfield static struct connection *__nodeid2con(int nodeid, gfp_t alloc) 2026ed7257bSPatrick Caulfield { 2036ed7257bSPatrick Caulfield struct connection *con = NULL; 2046ed7257bSPatrick Caulfield int r; 2056ed7257bSPatrick Caulfield 2065e9ccc37SChristine Caulfield con = __find_con(nodeid); 2076ed7257bSPatrick Caulfield if (con || !alloc) 2086ed7257bSPatrick Caulfield return con; 2096ed7257bSPatrick Caulfield 2106ed7257bSPatrick Caulfield con = kmem_cache_zalloc(con_cache, alloc); 2116ed7257bSPatrick Caulfield if (!con) 2126ed7257bSPatrick Caulfield return NULL; 2136ed7257bSPatrick Caulfield 2145e9ccc37SChristine Caulfield r = nodeid_hash(nodeid); 2155e9ccc37SChristine Caulfield hlist_add_head(&con->list, &connection_hash[r]); 2166ed7257bSPatrick Caulfield 2176ed7257bSPatrick Caulfield con->nodeid = nodeid; 2186ed7257bSPatrick Caulfield mutex_init(&con->sock_mutex); 2196ed7257bSPatrick Caulfield INIT_LIST_HEAD(&con->writequeue); 2206ed7257bSPatrick Caulfield spin_lock_init(&con->writequeue_lock); 2216ed7257bSPatrick Caulfield INIT_WORK(&con->swork, process_send_sockets); 2226ed7257bSPatrick Caulfield INIT_WORK(&con->rwork, process_recv_sockets); 2236ed7257bSPatrick Caulfield 2246ed7257bSPatrick Caulfield /* Setup action pointers for child sockets */ 2256ed7257bSPatrick Caulfield if (con->nodeid) { 2265e9ccc37SChristine Caulfield struct connection *zerocon = __find_con(0); 2276ed7257bSPatrick Caulfield 2286ed7257bSPatrick Caulfield con->connect_action = zerocon->connect_action; 2296ed7257bSPatrick Caulfield if (!con->rx_action) 2306ed7257bSPatrick Caulfield con->rx_action = zerocon->rx_action; 2316ed7257bSPatrick Caulfield } 2326ed7257bSPatrick Caulfield 2336ed7257bSPatrick Caulfield return con; 2346ed7257bSPatrick Caulfield } 2356ed7257bSPatrick Caulfield 2365e9ccc37SChristine Caulfield /* Loop round all connections */ 2375e9ccc37SChristine Caulfield static void foreach_conn(void (*conn_func)(struct connection *c)) 2385e9ccc37SChristine Caulfield { 2395e9ccc37SChristine Caulfield int i; 240b67bfe0dSSasha Levin struct hlist_node *n; 2415e9ccc37SChristine Caulfield struct connection *con; 2425e9ccc37SChristine Caulfield 2435e9ccc37SChristine Caulfield for (i = 0; i < CONN_HASH_SIZE; i++) { 244b67bfe0dSSasha Levin hlist_for_each_entry_safe(con, n, &connection_hash[i], list) 2455e9ccc37SChristine Caulfield conn_func(con); 2465e9ccc37SChristine Caulfield } 2475e9ccc37SChristine Caulfield } 2485e9ccc37SChristine Caulfield 2496ed7257bSPatrick Caulfield static struct connection *nodeid2con(int nodeid, gfp_t allocation) 2506ed7257bSPatrick Caulfield { 2516ed7257bSPatrick Caulfield struct connection *con; 2526ed7257bSPatrick Caulfield 2537a936ce7SMatthias Kaehlcke mutex_lock(&connections_lock); 2546ed7257bSPatrick Caulfield con = __nodeid2con(nodeid, allocation); 2557a936ce7SMatthias Kaehlcke mutex_unlock(&connections_lock); 2566ed7257bSPatrick Caulfield 2576ed7257bSPatrick Caulfield return con; 2586ed7257bSPatrick Caulfield } 2596ed7257bSPatrick Caulfield 26036b71a8bSDavid Teigland static struct dlm_node_addr *find_node_addr(int nodeid) 2616ed7257bSPatrick Caulfield { 26236b71a8bSDavid Teigland struct dlm_node_addr *na; 26336b71a8bSDavid Teigland 26436b71a8bSDavid Teigland list_for_each_entry(na, &dlm_node_addrs, list) { 26536b71a8bSDavid Teigland if (na->nodeid == nodeid) 26636b71a8bSDavid Teigland return na; 26736b71a8bSDavid Teigland } 26836b71a8bSDavid Teigland return NULL; 26936b71a8bSDavid Teigland } 27036b71a8bSDavid Teigland 27136b71a8bSDavid Teigland static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y) 27236b71a8bSDavid Teigland { 27336b71a8bSDavid Teigland switch (x->ss_family) { 27436b71a8bSDavid Teigland case AF_INET: { 27536b71a8bSDavid Teigland struct sockaddr_in *sinx = (struct sockaddr_in *)x; 27636b71a8bSDavid Teigland struct sockaddr_in *siny = (struct sockaddr_in *)y; 27736b71a8bSDavid Teigland if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr) 27836b71a8bSDavid Teigland return 0; 27936b71a8bSDavid Teigland if (sinx->sin_port != siny->sin_port) 28036b71a8bSDavid Teigland return 0; 28136b71a8bSDavid Teigland break; 28236b71a8bSDavid Teigland } 28336b71a8bSDavid Teigland case AF_INET6: { 28436b71a8bSDavid Teigland struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x; 28536b71a8bSDavid Teigland struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y; 28636b71a8bSDavid Teigland if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr)) 28736b71a8bSDavid Teigland return 0; 28836b71a8bSDavid Teigland if (sinx->sin6_port != siny->sin6_port) 28936b71a8bSDavid Teigland return 0; 29036b71a8bSDavid Teigland break; 29136b71a8bSDavid Teigland } 29236b71a8bSDavid Teigland default: 29336b71a8bSDavid Teigland return 0; 29436b71a8bSDavid Teigland } 29536b71a8bSDavid Teigland return 1; 29636b71a8bSDavid Teigland } 29736b71a8bSDavid Teigland 29836b71a8bSDavid Teigland static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out, 29998e1b60eSMike Christie struct sockaddr *sa_out, bool try_new_addr) 30036b71a8bSDavid Teigland { 30136b71a8bSDavid Teigland struct sockaddr_storage sas; 30236b71a8bSDavid Teigland struct dlm_node_addr *na; 3036ed7257bSPatrick Caulfield 3046ed7257bSPatrick Caulfield if (!dlm_local_count) 3056ed7257bSPatrick Caulfield return -1; 3066ed7257bSPatrick Caulfield 30736b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 30836b71a8bSDavid Teigland na = find_node_addr(nodeid); 30998e1b60eSMike Christie if (na && na->addr_count) { 310ee44b4bcSMarcelo Ricardo Leitner memcpy(&sas, na->addr[na->curr_addr_index], 311ee44b4bcSMarcelo Ricardo Leitner sizeof(struct sockaddr_storage)); 312ee44b4bcSMarcelo Ricardo Leitner 31398e1b60eSMike Christie if (try_new_addr) { 31498e1b60eSMike Christie na->curr_addr_index++; 31598e1b60eSMike Christie if (na->curr_addr_index == na->addr_count) 31698e1b60eSMike Christie na->curr_addr_index = 0; 31798e1b60eSMike Christie } 31898e1b60eSMike Christie } 31936b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 32036b71a8bSDavid Teigland 32136b71a8bSDavid Teigland if (!na) 32236b71a8bSDavid Teigland return -EEXIST; 32336b71a8bSDavid Teigland 32436b71a8bSDavid Teigland if (!na->addr_count) 32536b71a8bSDavid Teigland return -ENOENT; 32636b71a8bSDavid Teigland 32736b71a8bSDavid Teigland if (sas_out) 32836b71a8bSDavid Teigland memcpy(sas_out, &sas, sizeof(struct sockaddr_storage)); 32936b71a8bSDavid Teigland 33036b71a8bSDavid Teigland if (!sa_out) 33136b71a8bSDavid Teigland return 0; 3326ed7257bSPatrick Caulfield 3336ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) { 33436b71a8bSDavid Teigland struct sockaddr_in *in4 = (struct sockaddr_in *) &sas; 33536b71a8bSDavid Teigland struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out; 3366ed7257bSPatrick Caulfield ret4->sin_addr.s_addr = in4->sin_addr.s_addr; 3376ed7257bSPatrick Caulfield } else { 33836b71a8bSDavid Teigland struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &sas; 33936b71a8bSDavid Teigland struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out; 3404e3fd7a0SAlexey Dobriyan ret6->sin6_addr = in6->sin6_addr; 3416ed7257bSPatrick Caulfield } 3426ed7257bSPatrick Caulfield 3436ed7257bSPatrick Caulfield return 0; 3446ed7257bSPatrick Caulfield } 3456ed7257bSPatrick Caulfield 34636b71a8bSDavid Teigland static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid) 34736b71a8bSDavid Teigland { 34836b71a8bSDavid Teigland struct dlm_node_addr *na; 34936b71a8bSDavid Teigland int rv = -EEXIST; 35098e1b60eSMike Christie int addr_i; 35136b71a8bSDavid Teigland 35236b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 35336b71a8bSDavid Teigland list_for_each_entry(na, &dlm_node_addrs, list) { 35436b71a8bSDavid Teigland if (!na->addr_count) 35536b71a8bSDavid Teigland continue; 35636b71a8bSDavid Teigland 35798e1b60eSMike Christie for (addr_i = 0; addr_i < na->addr_count; addr_i++) { 35898e1b60eSMike Christie if (addr_compare(na->addr[addr_i], addr)) { 35936b71a8bSDavid Teigland *nodeid = na->nodeid; 36036b71a8bSDavid Teigland rv = 0; 36198e1b60eSMike Christie goto unlock; 36236b71a8bSDavid Teigland } 36398e1b60eSMike Christie } 36498e1b60eSMike Christie } 36598e1b60eSMike Christie unlock: 36636b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 36736b71a8bSDavid Teigland return rv; 36836b71a8bSDavid Teigland } 36936b71a8bSDavid Teigland 37036b71a8bSDavid Teigland int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len) 37136b71a8bSDavid Teigland { 37236b71a8bSDavid Teigland struct sockaddr_storage *new_addr; 37336b71a8bSDavid Teigland struct dlm_node_addr *new_node, *na; 37436b71a8bSDavid Teigland 37536b71a8bSDavid Teigland new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS); 37636b71a8bSDavid Teigland if (!new_node) 37736b71a8bSDavid Teigland return -ENOMEM; 37836b71a8bSDavid Teigland 37936b71a8bSDavid Teigland new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS); 38036b71a8bSDavid Teigland if (!new_addr) { 38136b71a8bSDavid Teigland kfree(new_node); 38236b71a8bSDavid Teigland return -ENOMEM; 38336b71a8bSDavid Teigland } 38436b71a8bSDavid Teigland 38536b71a8bSDavid Teigland memcpy(new_addr, addr, len); 38636b71a8bSDavid Teigland 38736b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 38836b71a8bSDavid Teigland na = find_node_addr(nodeid); 38936b71a8bSDavid Teigland if (!na) { 39036b71a8bSDavid Teigland new_node->nodeid = nodeid; 39136b71a8bSDavid Teigland new_node->addr[0] = new_addr; 39236b71a8bSDavid Teigland new_node->addr_count = 1; 39336b71a8bSDavid Teigland list_add(&new_node->list, &dlm_node_addrs); 39436b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 39536b71a8bSDavid Teigland return 0; 39636b71a8bSDavid Teigland } 39736b71a8bSDavid Teigland 39836b71a8bSDavid Teigland if (na->addr_count >= DLM_MAX_ADDR_COUNT) { 39936b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 40036b71a8bSDavid Teigland kfree(new_addr); 40136b71a8bSDavid Teigland kfree(new_node); 40236b71a8bSDavid Teigland return -ENOSPC; 40336b71a8bSDavid Teigland } 40436b71a8bSDavid Teigland 40536b71a8bSDavid Teigland na->addr[na->addr_count++] = new_addr; 40636b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 40736b71a8bSDavid Teigland kfree(new_node); 40836b71a8bSDavid Teigland return 0; 40936b71a8bSDavid Teigland } 41036b71a8bSDavid Teigland 4116ed7257bSPatrick Caulfield /* Data available on socket or listen socket received a connect */ 412676d2369SDavid S. Miller static void lowcomms_data_ready(struct sock *sk) 4136ed7257bSPatrick Caulfield { 41493eaadebStsutomu.owa@toshiba.co.jp struct connection *con; 41593eaadebStsutomu.owa@toshiba.co.jp 41693eaadebStsutomu.owa@toshiba.co.jp read_lock_bh(&sk->sk_callback_lock); 41793eaadebStsutomu.owa@toshiba.co.jp con = sock2con(sk); 418afb853fbSPatrick Caulfield if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) 4196ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 42093eaadebStsutomu.owa@toshiba.co.jp read_unlock_bh(&sk->sk_callback_lock); 4216ed7257bSPatrick Caulfield } 4226ed7257bSPatrick Caulfield 4236ed7257bSPatrick Caulfield static void lowcomms_write_space(struct sock *sk) 4246ed7257bSPatrick Caulfield { 42593eaadebStsutomu.owa@toshiba.co.jp struct connection *con; 4266ed7257bSPatrick Caulfield 42793eaadebStsutomu.owa@toshiba.co.jp read_lock_bh(&sk->sk_callback_lock); 42893eaadebStsutomu.owa@toshiba.co.jp con = sock2con(sk); 429b36930ddSDavid Miller if (!con) 43093eaadebStsutomu.owa@toshiba.co.jp goto out; 431b36930ddSDavid Miller 432b36930ddSDavid Miller clear_bit(SOCK_NOSPACE, &con->sock->flags); 433b36930ddSDavid Miller 434b36930ddSDavid Miller if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) { 435b36930ddSDavid Miller con->sock->sk->sk_write_pending--; 4369cd3e072SEric Dumazet clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags); 437b36930ddSDavid Miller } 438b36930ddSDavid Miller 4396ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 44093eaadebStsutomu.owa@toshiba.co.jp out: 44193eaadebStsutomu.owa@toshiba.co.jp read_unlock_bh(&sk->sk_callback_lock); 4426ed7257bSPatrick Caulfield } 4436ed7257bSPatrick Caulfield 4446ed7257bSPatrick Caulfield static inline void lowcomms_connect_sock(struct connection *con) 4456ed7257bSPatrick Caulfield { 446063c4c99SLars Marowsky-Bree if (test_bit(CF_CLOSE, &con->flags)) 447063c4c99SLars Marowsky-Bree return; 4486ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 44961d9102bSBob Peterson cond_resched(); 4506ed7257bSPatrick Caulfield } 4516ed7257bSPatrick Caulfield 4526ed7257bSPatrick Caulfield static void lowcomms_state_change(struct sock *sk) 4536ed7257bSPatrick Caulfield { 454ee44b4bcSMarcelo Ricardo Leitner /* SCTP layer is not calling sk_data_ready when the connection 455ee44b4bcSMarcelo Ricardo Leitner * is done, so we catch the signal through here. Also, it 456ee44b4bcSMarcelo Ricardo Leitner * doesn't switch socket state when entering shutdown, so we 457ee44b4bcSMarcelo Ricardo Leitner * skip the write in that case. 458ee44b4bcSMarcelo Ricardo Leitner */ 459ee44b4bcSMarcelo Ricardo Leitner if (sk->sk_shutdown) { 460ee44b4bcSMarcelo Ricardo Leitner if (sk->sk_shutdown == RCV_SHUTDOWN) 461ee44b4bcSMarcelo Ricardo Leitner lowcomms_data_ready(sk); 462ee44b4bcSMarcelo Ricardo Leitner } else if (sk->sk_state == TCP_ESTABLISHED) { 4636ed7257bSPatrick Caulfield lowcomms_write_space(sk); 4646ed7257bSPatrick Caulfield } 465ee44b4bcSMarcelo Ricardo Leitner } 4666ed7257bSPatrick Caulfield 467391fbdc5SChristine Caulfield int dlm_lowcomms_connect_node(int nodeid) 468391fbdc5SChristine Caulfield { 469391fbdc5SChristine Caulfield struct connection *con; 470391fbdc5SChristine Caulfield 471391fbdc5SChristine Caulfield if (nodeid == dlm_our_nodeid()) 472391fbdc5SChristine Caulfield return 0; 473391fbdc5SChristine Caulfield 474391fbdc5SChristine Caulfield con = nodeid2con(nodeid, GFP_NOFS); 475391fbdc5SChristine Caulfield if (!con) 476391fbdc5SChristine Caulfield return -ENOMEM; 477391fbdc5SChristine Caulfield lowcomms_connect_sock(con); 478391fbdc5SChristine Caulfield return 0; 479391fbdc5SChristine Caulfield } 480391fbdc5SChristine Caulfield 481b3a5bbfdSBob Peterson static void lowcomms_error_report(struct sock *sk) 482b3a5bbfdSBob Peterson { 483b81171cbSBob Peterson struct connection *con; 484b3a5bbfdSBob Peterson struct sockaddr_storage saddr; 485b81171cbSBob Peterson void (*orig_report)(struct sock *) = NULL; 486b3a5bbfdSBob Peterson 487b81171cbSBob Peterson read_lock_bh(&sk->sk_callback_lock); 488b81171cbSBob Peterson con = sock2con(sk); 489b81171cbSBob Peterson if (con == NULL) 490b81171cbSBob Peterson goto out; 491b81171cbSBob Peterson 492cc661fc9SBob Peterson orig_report = listen_sock.sk_error_report; 4931a31833dSBob Peterson if (con->sock == NULL || 4949b2c45d4SDenys Vlasenko kernel_getpeername(con->sock, (struct sockaddr *)&saddr) < 0) { 495b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 496b3a5bbfdSBob Peterson "sending to node %d, port %d, " 497b3a5bbfdSBob Peterson "sk_err=%d/%d\n", dlm_our_nodeid(), 498b3a5bbfdSBob Peterson con->nodeid, dlm_config.ci_tcp_port, 499b3a5bbfdSBob Peterson sk->sk_err, sk->sk_err_soft); 500b3a5bbfdSBob Peterson } else if (saddr.ss_family == AF_INET) { 501b3a5bbfdSBob Peterson struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr; 502b3a5bbfdSBob Peterson 503b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 504b3a5bbfdSBob Peterson "sending to node %d at %pI4, port %d, " 505b3a5bbfdSBob Peterson "sk_err=%d/%d\n", dlm_our_nodeid(), 506b3a5bbfdSBob Peterson con->nodeid, &sin4->sin_addr.s_addr, 507b3a5bbfdSBob Peterson dlm_config.ci_tcp_port, sk->sk_err, 508b3a5bbfdSBob Peterson sk->sk_err_soft); 509b3a5bbfdSBob Peterson } else { 510b3a5bbfdSBob Peterson struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr; 511b3a5bbfdSBob Peterson 512b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 513b3a5bbfdSBob Peterson "sending to node %d at %u.%u.%u.%u, " 514b3a5bbfdSBob Peterson "port %d, sk_err=%d/%d\n", dlm_our_nodeid(), 515b3a5bbfdSBob Peterson con->nodeid, sin6->sin6_addr.s6_addr32[0], 516b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[1], 517b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[2], 518b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[3], 519b3a5bbfdSBob Peterson dlm_config.ci_tcp_port, sk->sk_err, 520b3a5bbfdSBob Peterson sk->sk_err_soft); 521b3a5bbfdSBob Peterson } 522b81171cbSBob Peterson out: 523b81171cbSBob Peterson read_unlock_bh(&sk->sk_callback_lock); 524b81171cbSBob Peterson if (orig_report) 525b81171cbSBob Peterson orig_report(sk); 526b81171cbSBob Peterson } 527b81171cbSBob Peterson 528b81171cbSBob Peterson /* Note: sk_callback_lock must be locked before calling this function. */ 529cc661fc9SBob Peterson static void save_listen_callbacks(struct socket *sock) 530b81171cbSBob Peterson { 531cc661fc9SBob Peterson struct sock *sk = sock->sk; 532cc661fc9SBob Peterson 533cc661fc9SBob Peterson listen_sock.sk_data_ready = sk->sk_data_ready; 534cc661fc9SBob Peterson listen_sock.sk_state_change = sk->sk_state_change; 535cc661fc9SBob Peterson listen_sock.sk_write_space = sk->sk_write_space; 536cc661fc9SBob Peterson listen_sock.sk_error_report = sk->sk_error_report; 537b81171cbSBob Peterson } 538b81171cbSBob Peterson 539cc661fc9SBob Peterson static void restore_callbacks(struct socket *sock) 540b81171cbSBob Peterson { 541cc661fc9SBob Peterson struct sock *sk = sock->sk; 542cc661fc9SBob Peterson 543b81171cbSBob Peterson write_lock_bh(&sk->sk_callback_lock); 544b81171cbSBob Peterson sk->sk_user_data = NULL; 545cc661fc9SBob Peterson sk->sk_data_ready = listen_sock.sk_data_ready; 546cc661fc9SBob Peterson sk->sk_state_change = listen_sock.sk_state_change; 547cc661fc9SBob Peterson sk->sk_write_space = listen_sock.sk_write_space; 548cc661fc9SBob Peterson sk->sk_error_report = listen_sock.sk_error_report; 549b81171cbSBob Peterson write_unlock_bh(&sk->sk_callback_lock); 550b3a5bbfdSBob Peterson } 551b3a5bbfdSBob Peterson 5526ed7257bSPatrick Caulfield /* Make a socket active */ 553988419a9Stsutomu.owa@toshiba.co.jp static void add_sock(struct socket *sock, struct connection *con) 5546ed7257bSPatrick Caulfield { 555b81171cbSBob Peterson struct sock *sk = sock->sk; 556b81171cbSBob Peterson 557b81171cbSBob Peterson write_lock_bh(&sk->sk_callback_lock); 5586ed7257bSPatrick Caulfield con->sock = sock; 5596ed7257bSPatrick Caulfield 560b81171cbSBob Peterson sk->sk_user_data = con; 5616ed7257bSPatrick Caulfield /* Install a data_ready callback */ 562b81171cbSBob Peterson sk->sk_data_ready = lowcomms_data_ready; 563b81171cbSBob Peterson sk->sk_write_space = lowcomms_write_space; 564b81171cbSBob Peterson sk->sk_state_change = lowcomms_state_change; 565b81171cbSBob Peterson sk->sk_allocation = GFP_NOFS; 566b81171cbSBob Peterson sk->sk_error_report = lowcomms_error_report; 567b81171cbSBob Peterson write_unlock_bh(&sk->sk_callback_lock); 5686ed7257bSPatrick Caulfield } 5696ed7257bSPatrick Caulfield 5706ed7257bSPatrick Caulfield /* Add the port number to an IPv6 or 4 sockaddr and return the address 5716ed7257bSPatrick Caulfield length */ 5726ed7257bSPatrick Caulfield static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, 5736ed7257bSPatrick Caulfield int *addr_len) 5746ed7257bSPatrick Caulfield { 5756ed7257bSPatrick Caulfield saddr->ss_family = dlm_local_addr[0]->ss_family; 5766ed7257bSPatrick Caulfield if (saddr->ss_family == AF_INET) { 5776ed7257bSPatrick Caulfield struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; 5786ed7257bSPatrick Caulfield in4_addr->sin_port = cpu_to_be16(port); 5796ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in); 5806ed7257bSPatrick Caulfield memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); 5816ed7257bSPatrick Caulfield } else { 5826ed7257bSPatrick Caulfield struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; 5836ed7257bSPatrick Caulfield in6_addr->sin6_port = cpu_to_be16(port); 5846ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in6); 5856ed7257bSPatrick Caulfield } 58601c8cab2SPatrick Caulfield memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); 5876ed7257bSPatrick Caulfield } 5886ed7257bSPatrick Caulfield 5896ed7257bSPatrick Caulfield /* Close a remote connection and tidy up */ 5900d737a8cSMarcelo Ricardo Leitner static void close_connection(struct connection *con, bool and_other, 5910d737a8cSMarcelo Ricardo Leitner bool tx, bool rx) 5926ed7257bSPatrick Caulfield { 593b2a66629Stsutomu.owa@toshiba.co.jp bool closing = test_and_set_bit(CF_CLOSING, &con->flags); 594b2a66629Stsutomu.owa@toshiba.co.jp 5950aa18464Stsutomu.owa@toshiba.co.jp if (tx && !closing && cancel_work_sync(&con->swork)) { 5960d737a8cSMarcelo Ricardo Leitner log_print("canceled swork for node %d", con->nodeid); 5970aa18464Stsutomu.owa@toshiba.co.jp clear_bit(CF_WRITE_PENDING, &con->flags); 5980aa18464Stsutomu.owa@toshiba.co.jp } 5990aa18464Stsutomu.owa@toshiba.co.jp if (rx && !closing && cancel_work_sync(&con->rwork)) { 6000d737a8cSMarcelo Ricardo Leitner log_print("canceled rwork for node %d", con->nodeid); 6010aa18464Stsutomu.owa@toshiba.co.jp clear_bit(CF_READ_PENDING, &con->flags); 6020aa18464Stsutomu.owa@toshiba.co.jp } 6036ed7257bSPatrick Caulfield 6040d737a8cSMarcelo Ricardo Leitner mutex_lock(&con->sock_mutex); 6056ed7257bSPatrick Caulfield if (con->sock) { 606cc661fc9SBob Peterson restore_callbacks(con->sock); 6076ed7257bSPatrick Caulfield sock_release(con->sock); 6086ed7257bSPatrick Caulfield con->sock = NULL; 6096ed7257bSPatrick Caulfield } 6106ed7257bSPatrick Caulfield if (con->othercon && and_other) { 6116ed7257bSPatrick Caulfield /* Will only re-enter once. */ 6120d737a8cSMarcelo Ricardo Leitner close_connection(con->othercon, false, true, true); 6136ed7257bSPatrick Caulfield } 6146ed7257bSPatrick Caulfield if (con->rx_page) { 6156ed7257bSPatrick Caulfield __free_page(con->rx_page); 6166ed7257bSPatrick Caulfield con->rx_page = NULL; 6176ed7257bSPatrick Caulfield } 6189e5f2825SPatrick Caulfield 6196ed7257bSPatrick Caulfield con->retries = 0; 6206ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 621b2a66629Stsutomu.owa@toshiba.co.jp clear_bit(CF_CLOSING, &con->flags); 6226ed7257bSPatrick Caulfield } 6236ed7257bSPatrick Caulfield 6246ed7257bSPatrick Caulfield /* Data received from remote end */ 6256ed7257bSPatrick Caulfield static int receive_from_sock(struct connection *con) 6266ed7257bSPatrick Caulfield { 6276ed7257bSPatrick Caulfield int ret = 0; 6286ed7257bSPatrick Caulfield struct msghdr msg = {}; 6296ed7257bSPatrick Caulfield struct kvec iov[2]; 6306ed7257bSPatrick Caulfield unsigned len; 6316ed7257bSPatrick Caulfield int r; 6326ed7257bSPatrick Caulfield int call_again_soon = 0; 6336ed7257bSPatrick Caulfield int nvec; 6346ed7257bSPatrick Caulfield 6356ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 6366ed7257bSPatrick Caulfield 6376ed7257bSPatrick Caulfield if (con->sock == NULL) { 6386ed7257bSPatrick Caulfield ret = -EAGAIN; 6396ed7257bSPatrick Caulfield goto out_close; 6406ed7257bSPatrick Caulfield } 641acee4e52SMarcelo Ricardo Leitner if (con->nodeid == 0) { 642acee4e52SMarcelo Ricardo Leitner ret = -EINVAL; 643acee4e52SMarcelo Ricardo Leitner goto out_close; 644acee4e52SMarcelo Ricardo Leitner } 6456ed7257bSPatrick Caulfield 6466ed7257bSPatrick Caulfield if (con->rx_page == NULL) { 6476ed7257bSPatrick Caulfield /* 6486ed7257bSPatrick Caulfield * This doesn't need to be atomic, but I think it should 6496ed7257bSPatrick Caulfield * improve performance if it is. 6506ed7257bSPatrick Caulfield */ 6516ed7257bSPatrick Caulfield con->rx_page = alloc_page(GFP_ATOMIC); 6526ed7257bSPatrick Caulfield if (con->rx_page == NULL) 6536ed7257bSPatrick Caulfield goto out_resched; 65409cbfeafSKirill A. Shutemov cbuf_init(&con->cb, PAGE_SIZE); 6556ed7257bSPatrick Caulfield } 6566ed7257bSPatrick Caulfield 6576ed7257bSPatrick Caulfield /* 6586ed7257bSPatrick Caulfield * iov[0] is the bit of the circular buffer between the current end 6596ed7257bSPatrick Caulfield * point (cb.base + cb.len) and the end of the buffer. 6606ed7257bSPatrick Caulfield */ 6616ed7257bSPatrick Caulfield iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); 6626ed7257bSPatrick Caulfield iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); 6636ed7257bSPatrick Caulfield iov[1].iov_len = 0; 6646ed7257bSPatrick Caulfield nvec = 1; 6656ed7257bSPatrick Caulfield 6666ed7257bSPatrick Caulfield /* 6676ed7257bSPatrick Caulfield * iov[1] is the bit of the circular buffer between the start of the 6686ed7257bSPatrick Caulfield * buffer and the start of the currently used section (cb.base) 6696ed7257bSPatrick Caulfield */ 6706ed7257bSPatrick Caulfield if (cbuf_data(&con->cb) >= con->cb.base) { 67109cbfeafSKirill A. Shutemov iov[0].iov_len = PAGE_SIZE - cbuf_data(&con->cb); 6726ed7257bSPatrick Caulfield iov[1].iov_len = con->cb.base; 6736ed7257bSPatrick Caulfield iov[1].iov_base = page_address(con->rx_page); 6746ed7257bSPatrick Caulfield nvec = 2; 6756ed7257bSPatrick Caulfield } 6766ed7257bSPatrick Caulfield len = iov[0].iov_len + iov[1].iov_len; 677c8c7840eSAl Viro iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, iov, nvec, len); 6786ed7257bSPatrick Caulfield 679c8c7840eSAl Viro r = ret = sock_recvmsg(con->sock, &msg, MSG_DONTWAIT | MSG_NOSIGNAL); 6806ed7257bSPatrick Caulfield if (ret <= 0) 6816ed7257bSPatrick Caulfield goto out_close; 682ee44b4bcSMarcelo Ricardo Leitner else if (ret == len) 683ee44b4bcSMarcelo Ricardo Leitner call_again_soon = 1; 6846ed7257bSPatrick Caulfield 6856ed7257bSPatrick Caulfield cbuf_add(&con->cb, ret); 6866ed7257bSPatrick Caulfield ret = dlm_process_incoming_buffer(con->nodeid, 6876ed7257bSPatrick Caulfield page_address(con->rx_page), 6886ed7257bSPatrick Caulfield con->cb.base, con->cb.len, 68909cbfeafSKirill A. Shutemov PAGE_SIZE); 6906ed7257bSPatrick Caulfield if (ret == -EBADMSG) { 691ee44b4bcSMarcelo Ricardo Leitner log_print("lowcomms: addr=%p, base=%u, len=%u, read=%d", 692ee44b4bcSMarcelo Ricardo Leitner page_address(con->rx_page), con->cb.base, 693ee44b4bcSMarcelo Ricardo Leitner con->cb.len, r); 6946ed7257bSPatrick Caulfield } 6956ed7257bSPatrick Caulfield if (ret < 0) 6966ed7257bSPatrick Caulfield goto out_close; 6976ed7257bSPatrick Caulfield cbuf_eat(&con->cb, ret); 6986ed7257bSPatrick Caulfield 6996ed7257bSPatrick Caulfield if (cbuf_empty(&con->cb) && !call_again_soon) { 7006ed7257bSPatrick Caulfield __free_page(con->rx_page); 7016ed7257bSPatrick Caulfield con->rx_page = NULL; 7026ed7257bSPatrick Caulfield } 7036ed7257bSPatrick Caulfield 7046ed7257bSPatrick Caulfield if (call_again_soon) 7056ed7257bSPatrick Caulfield goto out_resched; 7066ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7076ed7257bSPatrick Caulfield return 0; 7086ed7257bSPatrick Caulfield 7096ed7257bSPatrick Caulfield out_resched: 7106ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) 7116ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 7126ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7136ed7257bSPatrick Caulfield return -EAGAIN; 7146ed7257bSPatrick Caulfield 7156ed7257bSPatrick Caulfield out_close: 7166ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7179e5f2825SPatrick Caulfield if (ret != -EAGAIN) { 718c553e173Stsutomu.owa@toshiba.co.jp close_connection(con, true, true, false); 7196ed7257bSPatrick Caulfield /* Reconnect when there is something to send */ 7206ed7257bSPatrick Caulfield } 7216ed7257bSPatrick Caulfield /* Don't return success if we really got EOF */ 7226ed7257bSPatrick Caulfield if (ret == 0) 7236ed7257bSPatrick Caulfield ret = -EAGAIN; 7246ed7257bSPatrick Caulfield 7256ed7257bSPatrick Caulfield return ret; 7266ed7257bSPatrick Caulfield } 7276ed7257bSPatrick Caulfield 7286ed7257bSPatrick Caulfield /* Listening socket is busy, accept a connection */ 7296ed7257bSPatrick Caulfield static int tcp_accept_from_sock(struct connection *con) 7306ed7257bSPatrick Caulfield { 7316ed7257bSPatrick Caulfield int result; 7326ed7257bSPatrick Caulfield struct sockaddr_storage peeraddr; 7336ed7257bSPatrick Caulfield struct socket *newsock; 7346ed7257bSPatrick Caulfield int len; 7356ed7257bSPatrick Caulfield int nodeid; 7366ed7257bSPatrick Caulfield struct connection *newcon; 7376ed7257bSPatrick Caulfield struct connection *addcon; 7386ed7257bSPatrick Caulfield 739513ef596SDavid Teigland mutex_lock(&connections_lock); 740513ef596SDavid Teigland if (!dlm_allow_conn) { 741513ef596SDavid Teigland mutex_unlock(&connections_lock); 742513ef596SDavid Teigland return -1; 743513ef596SDavid Teigland } 744513ef596SDavid Teigland mutex_unlock(&connections_lock); 745513ef596SDavid Teigland 7466ed7257bSPatrick Caulfield mutex_lock_nested(&con->sock_mutex, 0); 7476ed7257bSPatrick Caulfield 7483421fb15Stsutomu.owa@toshiba.co.jp if (!con->sock) { 7493421fb15Stsutomu.owa@toshiba.co.jp mutex_unlock(&con->sock_mutex); 7503421fb15Stsutomu.owa@toshiba.co.jp return -ENOTCONN; 7513421fb15Stsutomu.owa@toshiba.co.jp } 7526ed7257bSPatrick Caulfield 7533421fb15Stsutomu.owa@toshiba.co.jp result = kernel_accept(con->sock, &newsock, O_NONBLOCK); 7546ed7257bSPatrick Caulfield if (result < 0) 7556ed7257bSPatrick Caulfield goto accept_err; 7566ed7257bSPatrick Caulfield 7576ed7257bSPatrick Caulfield /* Get the connected socket's peer */ 7586ed7257bSPatrick Caulfield memset(&peeraddr, 0, sizeof(peeraddr)); 7599b2c45d4SDenys Vlasenko len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2); 7609b2c45d4SDenys Vlasenko if (len < 0) { 7616ed7257bSPatrick Caulfield result = -ECONNABORTED; 7626ed7257bSPatrick Caulfield goto accept_err; 7636ed7257bSPatrick Caulfield } 7646ed7257bSPatrick Caulfield 7656ed7257bSPatrick Caulfield /* Get the new node's NODEID */ 7666ed7257bSPatrick Caulfield make_sockaddr(&peeraddr, 0, &len); 76736b71a8bSDavid Teigland if (addr_to_nodeid(&peeraddr, &nodeid)) { 768bcaadf5cSMasatake YAMATO unsigned char *b=(unsigned char *)&peeraddr; 769617e82e1SDavid Teigland log_print("connect from non cluster node"); 770bcaadf5cSMasatake YAMATO print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, 771bcaadf5cSMasatake YAMATO b, sizeof(struct sockaddr_storage)); 7726ed7257bSPatrick Caulfield sock_release(newsock); 7736ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7746ed7257bSPatrick Caulfield return -1; 7756ed7257bSPatrick Caulfield } 7766ed7257bSPatrick Caulfield 7776ed7257bSPatrick Caulfield log_print("got connection from %d", nodeid); 7786ed7257bSPatrick Caulfield 7796ed7257bSPatrick Caulfield /* Check to see if we already have a connection to this node. This 7806ed7257bSPatrick Caulfield * could happen if the two nodes initiate a connection at roughly 7816ed7257bSPatrick Caulfield * the same time and the connections cross on the wire. 7826ed7257bSPatrick Caulfield * In this case we store the incoming one in "othercon" 7836ed7257bSPatrick Caulfield */ 784748285ccSDavid Teigland newcon = nodeid2con(nodeid, GFP_NOFS); 7856ed7257bSPatrick Caulfield if (!newcon) { 7866ed7257bSPatrick Caulfield result = -ENOMEM; 7876ed7257bSPatrick Caulfield goto accept_err; 7886ed7257bSPatrick Caulfield } 7896ed7257bSPatrick Caulfield mutex_lock_nested(&newcon->sock_mutex, 1); 7906ed7257bSPatrick Caulfield if (newcon->sock) { 7916ed7257bSPatrick Caulfield struct connection *othercon = newcon->othercon; 7926ed7257bSPatrick Caulfield 7936ed7257bSPatrick Caulfield if (!othercon) { 794748285ccSDavid Teigland othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); 7956ed7257bSPatrick Caulfield if (!othercon) { 796617e82e1SDavid Teigland log_print("failed to allocate incoming socket"); 7976ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 7986ed7257bSPatrick Caulfield result = -ENOMEM; 7996ed7257bSPatrick Caulfield goto accept_err; 8006ed7257bSPatrick Caulfield } 8016ed7257bSPatrick Caulfield othercon->nodeid = nodeid; 8026ed7257bSPatrick Caulfield othercon->rx_action = receive_from_sock; 8036ed7257bSPatrick Caulfield mutex_init(&othercon->sock_mutex); 80426b41099Stsutomu.owa@toshiba.co.jp INIT_LIST_HEAD(&othercon->writequeue); 80526b41099Stsutomu.owa@toshiba.co.jp spin_lock_init(&othercon->writequeue_lock); 8066ed7257bSPatrick Caulfield INIT_WORK(&othercon->swork, process_send_sockets); 8076ed7257bSPatrick Caulfield INIT_WORK(&othercon->rwork, process_recv_sockets); 8086ed7257bSPatrick Caulfield set_bit(CF_IS_OTHERCON, &othercon->flags); 80961d96be0SPatrick Caulfield } 810c7355827Stsutomu.owa@toshiba.co.jp mutex_lock_nested(&othercon->sock_mutex, 2); 81161d96be0SPatrick Caulfield if (!othercon->sock) { 8126ed7257bSPatrick Caulfield newcon->othercon = othercon; 813988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, othercon); 8146ed7257bSPatrick Caulfield addcon = othercon; 815c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 8166ed7257bSPatrick Caulfield } 8176ed7257bSPatrick Caulfield else { 81897d84836SPatrick Caulfield printk("Extra connection from node %d attempted\n", nodeid); 81997d84836SPatrick Caulfield result = -EAGAIN; 820c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 821f4fadb23Sakpm@linux-foundation.org mutex_unlock(&newcon->sock_mutex); 82297d84836SPatrick Caulfield goto accept_err; 82397d84836SPatrick Caulfield } 82497d84836SPatrick Caulfield } 82597d84836SPatrick Caulfield else { 8266ed7257bSPatrick Caulfield newcon->rx_action = receive_from_sock; 8273735b4b9SBob Peterson /* accept copies the sk after we've saved the callbacks, so we 8283735b4b9SBob Peterson don't want to save them a second time or comm errors will 8293735b4b9SBob Peterson result in calling sk_error_report recursively. */ 830988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, newcon); 8316ed7257bSPatrick Caulfield addcon = newcon; 8326ed7257bSPatrick Caulfield } 8336ed7257bSPatrick Caulfield 8346ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 8356ed7257bSPatrick Caulfield 8366ed7257bSPatrick Caulfield /* 8376ed7257bSPatrick Caulfield * Add it to the active queue in case we got data 83825985edcSLucas De Marchi * between processing the accept adding the socket 8396ed7257bSPatrick Caulfield * to the read_sockets list 8406ed7257bSPatrick Caulfield */ 8416ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) 8426ed7257bSPatrick Caulfield queue_work(recv_workqueue, &addcon->rwork); 8436ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 8446ed7257bSPatrick Caulfield 8456ed7257bSPatrick Caulfield return 0; 8466ed7257bSPatrick Caulfield 8476ed7257bSPatrick Caulfield accept_err: 8486ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 8493421fb15Stsutomu.owa@toshiba.co.jp if (newsock) 8506ed7257bSPatrick Caulfield sock_release(newsock); 8516ed7257bSPatrick Caulfield 8526ed7257bSPatrick Caulfield if (result != -EAGAIN) 853617e82e1SDavid Teigland log_print("error accepting connection from node: %d", result); 8546ed7257bSPatrick Caulfield return result; 8556ed7257bSPatrick Caulfield } 8566ed7257bSPatrick Caulfield 85718df8a87Skbuild test robot static int sctp_accept_from_sock(struct connection *con) 858ee44b4bcSMarcelo Ricardo Leitner { 859ee44b4bcSMarcelo Ricardo Leitner /* Check that the new node is in the lockspace */ 860ee44b4bcSMarcelo Ricardo Leitner struct sctp_prim prim; 861ee44b4bcSMarcelo Ricardo Leitner int nodeid; 862ee44b4bcSMarcelo Ricardo Leitner int prim_len, ret; 863ee44b4bcSMarcelo Ricardo Leitner int addr_len; 864ee44b4bcSMarcelo Ricardo Leitner struct connection *newcon; 865ee44b4bcSMarcelo Ricardo Leitner struct connection *addcon; 866ee44b4bcSMarcelo Ricardo Leitner struct socket *newsock; 867ee44b4bcSMarcelo Ricardo Leitner 868ee44b4bcSMarcelo Ricardo Leitner mutex_lock(&connections_lock); 869ee44b4bcSMarcelo Ricardo Leitner if (!dlm_allow_conn) { 870ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&connections_lock); 871ee44b4bcSMarcelo Ricardo Leitner return -1; 872ee44b4bcSMarcelo Ricardo Leitner } 873ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&connections_lock); 874ee44b4bcSMarcelo Ricardo Leitner 875ee44b4bcSMarcelo Ricardo Leitner mutex_lock_nested(&con->sock_mutex, 0); 876ee44b4bcSMarcelo Ricardo Leitner 877ee44b4bcSMarcelo Ricardo Leitner ret = kernel_accept(con->sock, &newsock, O_NONBLOCK); 878ee44b4bcSMarcelo Ricardo Leitner if (ret < 0) 879ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 880ee44b4bcSMarcelo Ricardo Leitner 881ee44b4bcSMarcelo Ricardo Leitner memset(&prim, 0, sizeof(struct sctp_prim)); 882ee44b4bcSMarcelo Ricardo Leitner prim_len = sizeof(struct sctp_prim); 883ee44b4bcSMarcelo Ricardo Leitner 884ee44b4bcSMarcelo Ricardo Leitner ret = kernel_getsockopt(newsock, IPPROTO_SCTP, SCTP_PRIMARY_ADDR, 885ee44b4bcSMarcelo Ricardo Leitner (char *)&prim, &prim_len); 886ee44b4bcSMarcelo Ricardo Leitner if (ret < 0) { 887ee44b4bcSMarcelo Ricardo Leitner log_print("getsockopt/sctp_primary_addr failed: %d", ret); 888ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 889ee44b4bcSMarcelo Ricardo Leitner } 890ee44b4bcSMarcelo Ricardo Leitner 891ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&prim.ssp_addr, 0, &addr_len); 89226c1ec2fSWei Yongjun ret = addr_to_nodeid(&prim.ssp_addr, &nodeid); 89326c1ec2fSWei Yongjun if (ret) { 894ee44b4bcSMarcelo Ricardo Leitner unsigned char *b = (unsigned char *)&prim.ssp_addr; 895ee44b4bcSMarcelo Ricardo Leitner 896ee44b4bcSMarcelo Ricardo Leitner log_print("reject connect from unknown addr"); 897ee44b4bcSMarcelo Ricardo Leitner print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, 898ee44b4bcSMarcelo Ricardo Leitner b, sizeof(struct sockaddr_storage)); 899ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 900ee44b4bcSMarcelo Ricardo Leitner } 901ee44b4bcSMarcelo Ricardo Leitner 902ee44b4bcSMarcelo Ricardo Leitner newcon = nodeid2con(nodeid, GFP_NOFS); 903ee44b4bcSMarcelo Ricardo Leitner if (!newcon) { 904ee44b4bcSMarcelo Ricardo Leitner ret = -ENOMEM; 905ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 906ee44b4bcSMarcelo Ricardo Leitner } 907ee44b4bcSMarcelo Ricardo Leitner 908ee44b4bcSMarcelo Ricardo Leitner mutex_lock_nested(&newcon->sock_mutex, 1); 909ee44b4bcSMarcelo Ricardo Leitner 910ee44b4bcSMarcelo Ricardo Leitner if (newcon->sock) { 911ee44b4bcSMarcelo Ricardo Leitner struct connection *othercon = newcon->othercon; 912ee44b4bcSMarcelo Ricardo Leitner 913ee44b4bcSMarcelo Ricardo Leitner if (!othercon) { 914ee44b4bcSMarcelo Ricardo Leitner othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); 915ee44b4bcSMarcelo Ricardo Leitner if (!othercon) { 916ee44b4bcSMarcelo Ricardo Leitner log_print("failed to allocate incoming socket"); 917ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 918ee44b4bcSMarcelo Ricardo Leitner ret = -ENOMEM; 919ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 920ee44b4bcSMarcelo Ricardo Leitner } 921ee44b4bcSMarcelo Ricardo Leitner othercon->nodeid = nodeid; 922ee44b4bcSMarcelo Ricardo Leitner othercon->rx_action = receive_from_sock; 923ee44b4bcSMarcelo Ricardo Leitner mutex_init(&othercon->sock_mutex); 92426b41099Stsutomu.owa@toshiba.co.jp INIT_LIST_HEAD(&othercon->writequeue); 92526b41099Stsutomu.owa@toshiba.co.jp spin_lock_init(&othercon->writequeue_lock); 926ee44b4bcSMarcelo Ricardo Leitner INIT_WORK(&othercon->swork, process_send_sockets); 927ee44b4bcSMarcelo Ricardo Leitner INIT_WORK(&othercon->rwork, process_recv_sockets); 928ee44b4bcSMarcelo Ricardo Leitner set_bit(CF_IS_OTHERCON, &othercon->flags); 929ee44b4bcSMarcelo Ricardo Leitner } 930c7355827Stsutomu.owa@toshiba.co.jp mutex_lock_nested(&othercon->sock_mutex, 2); 931ee44b4bcSMarcelo Ricardo Leitner if (!othercon->sock) { 932ee44b4bcSMarcelo Ricardo Leitner newcon->othercon = othercon; 933988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, othercon); 934ee44b4bcSMarcelo Ricardo Leitner addcon = othercon; 935c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 936ee44b4bcSMarcelo Ricardo Leitner } else { 937ee44b4bcSMarcelo Ricardo Leitner printk("Extra connection from node %d attempted\n", nodeid); 938ee44b4bcSMarcelo Ricardo Leitner ret = -EAGAIN; 939c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 940ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 941ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 942ee44b4bcSMarcelo Ricardo Leitner } 943ee44b4bcSMarcelo Ricardo Leitner } else { 944ee44b4bcSMarcelo Ricardo Leitner newcon->rx_action = receive_from_sock; 945988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, newcon); 946ee44b4bcSMarcelo Ricardo Leitner addcon = newcon; 947ee44b4bcSMarcelo Ricardo Leitner } 948ee44b4bcSMarcelo Ricardo Leitner 949ee44b4bcSMarcelo Ricardo Leitner log_print("connected to %d", nodeid); 950ee44b4bcSMarcelo Ricardo Leitner 951ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 952ee44b4bcSMarcelo Ricardo Leitner 953ee44b4bcSMarcelo Ricardo Leitner /* 954ee44b4bcSMarcelo Ricardo Leitner * Add it to the active queue in case we got data 955ee44b4bcSMarcelo Ricardo Leitner * between processing the accept adding the socket 956ee44b4bcSMarcelo Ricardo Leitner * to the read_sockets list 957ee44b4bcSMarcelo Ricardo Leitner */ 958ee44b4bcSMarcelo Ricardo Leitner if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) 959ee44b4bcSMarcelo Ricardo Leitner queue_work(recv_workqueue, &addcon->rwork); 960ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 961ee44b4bcSMarcelo Ricardo Leitner 962ee44b4bcSMarcelo Ricardo Leitner return 0; 963ee44b4bcSMarcelo Ricardo Leitner 964ee44b4bcSMarcelo Ricardo Leitner accept_err: 965ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 966ee44b4bcSMarcelo Ricardo Leitner if (newsock) 967ee44b4bcSMarcelo Ricardo Leitner sock_release(newsock); 968ee44b4bcSMarcelo Ricardo Leitner if (ret != -EAGAIN) 969ee44b4bcSMarcelo Ricardo Leitner log_print("error accepting connection from node: %d", ret); 970ee44b4bcSMarcelo Ricardo Leitner 971ee44b4bcSMarcelo Ricardo Leitner return ret; 972ee44b4bcSMarcelo Ricardo Leitner } 973ee44b4bcSMarcelo Ricardo Leitner 9746ed7257bSPatrick Caulfield static void free_entry(struct writequeue_entry *e) 9756ed7257bSPatrick Caulfield { 9766ed7257bSPatrick Caulfield __free_page(e->page); 9776ed7257bSPatrick Caulfield kfree(e); 9786ed7257bSPatrick Caulfield } 9796ed7257bSPatrick Caulfield 9805d689871SMike Christie /* 9815d689871SMike Christie * writequeue_entry_complete - try to delete and free write queue entry 9825d689871SMike Christie * @e: write queue entry to try to delete 9835d689871SMike Christie * @completed: bytes completed 9845d689871SMike Christie * 9855d689871SMike Christie * writequeue_lock must be held. 9865d689871SMike Christie */ 9875d689871SMike Christie static void writequeue_entry_complete(struct writequeue_entry *e, int completed) 9885d689871SMike Christie { 9895d689871SMike Christie e->offset += completed; 9905d689871SMike Christie e->len -= completed; 9915d689871SMike Christie 9925d689871SMike Christie if (e->len == 0 && e->users == 0) { 9935d689871SMike Christie list_del(&e->list); 9945d689871SMike Christie free_entry(e); 9955d689871SMike Christie } 9965d689871SMike Christie } 9975d689871SMike Christie 998ee44b4bcSMarcelo Ricardo Leitner /* 999ee44b4bcSMarcelo Ricardo Leitner * sctp_bind_addrs - bind a SCTP socket to all our addresses 1000ee44b4bcSMarcelo Ricardo Leitner */ 1001ee44b4bcSMarcelo Ricardo Leitner static int sctp_bind_addrs(struct connection *con, uint16_t port) 1002ee44b4bcSMarcelo Ricardo Leitner { 1003ee44b4bcSMarcelo Ricardo Leitner struct sockaddr_storage localaddr; 1004ee44b4bcSMarcelo Ricardo Leitner int i, addr_len, result = 0; 1005ee44b4bcSMarcelo Ricardo Leitner 1006ee44b4bcSMarcelo Ricardo Leitner for (i = 0; i < dlm_local_count; i++) { 1007ee44b4bcSMarcelo Ricardo Leitner memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); 1008ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&localaddr, port, &addr_len); 1009ee44b4bcSMarcelo Ricardo Leitner 1010ee44b4bcSMarcelo Ricardo Leitner if (!i) 1011ee44b4bcSMarcelo Ricardo Leitner result = kernel_bind(con->sock, 1012ee44b4bcSMarcelo Ricardo Leitner (struct sockaddr *)&localaddr, 1013ee44b4bcSMarcelo Ricardo Leitner addr_len); 1014ee44b4bcSMarcelo Ricardo Leitner else 1015ee44b4bcSMarcelo Ricardo Leitner result = kernel_setsockopt(con->sock, SOL_SCTP, 1016ee44b4bcSMarcelo Ricardo Leitner SCTP_SOCKOPT_BINDX_ADD, 1017ee44b4bcSMarcelo Ricardo Leitner (char *)&localaddr, addr_len); 1018ee44b4bcSMarcelo Ricardo Leitner 1019ee44b4bcSMarcelo Ricardo Leitner if (result < 0) { 1020ee44b4bcSMarcelo Ricardo Leitner log_print("Can't bind to %d addr number %d, %d.\n", 1021ee44b4bcSMarcelo Ricardo Leitner port, i + 1, result); 1022ee44b4bcSMarcelo Ricardo Leitner break; 1023ee44b4bcSMarcelo Ricardo Leitner } 1024ee44b4bcSMarcelo Ricardo Leitner } 1025ee44b4bcSMarcelo Ricardo Leitner return result; 1026ee44b4bcSMarcelo Ricardo Leitner } 1027ee44b4bcSMarcelo Ricardo Leitner 10286ed7257bSPatrick Caulfield /* Initiate an SCTP association. 10296ed7257bSPatrick Caulfield This is a special case of send_to_sock() in that we don't yet have a 10306ed7257bSPatrick Caulfield peeled-off socket for this association, so we use the listening socket 10316ed7257bSPatrick Caulfield and add the primary IP address of the remote node. 10326ed7257bSPatrick Caulfield */ 1033ee44b4bcSMarcelo Ricardo Leitner static void sctp_connect_to_sock(struct connection *con) 10346ed7257bSPatrick Caulfield { 1035ee44b4bcSMarcelo Ricardo Leitner struct sockaddr_storage daddr; 1036ee44b4bcSMarcelo Ricardo Leitner int one = 1; 1037ee44b4bcSMarcelo Ricardo Leitner int result; 1038ee44b4bcSMarcelo Ricardo Leitner int addr_len; 1039ee44b4bcSMarcelo Ricardo Leitner struct socket *sock; 1040f706d830SGang He struct timeval tv = { .tv_sec = 5, .tv_usec = 0 }; 1041ee44b4bcSMarcelo Ricardo Leitner 1042ee44b4bcSMarcelo Ricardo Leitner if (con->nodeid == 0) { 1043ee44b4bcSMarcelo Ricardo Leitner log_print("attempt to connect sock 0 foiled"); 1044ee44b4bcSMarcelo Ricardo Leitner return; 1045ee44b4bcSMarcelo Ricardo Leitner } 10466ed7257bSPatrick Caulfield 10475d689871SMike Christie mutex_lock(&con->sock_mutex); 10486ed7257bSPatrick Caulfield 1049ee44b4bcSMarcelo Ricardo Leitner /* Some odd races can cause double-connects, ignore them */ 1050ee44b4bcSMarcelo Ricardo Leitner if (con->retries++ > MAX_CONNECT_RETRIES) 1051ee44b4bcSMarcelo Ricardo Leitner goto out; 1052ee44b4bcSMarcelo Ricardo Leitner 1053ee44b4bcSMarcelo Ricardo Leitner if (con->sock) { 1054ee44b4bcSMarcelo Ricardo Leitner log_print("node %d already connected.", con->nodeid); 1055ee44b4bcSMarcelo Ricardo Leitner goto out; 1056ee44b4bcSMarcelo Ricardo Leitner } 1057ee44b4bcSMarcelo Ricardo Leitner 1058ee44b4bcSMarcelo Ricardo Leitner memset(&daddr, 0, sizeof(daddr)); 1059ee44b4bcSMarcelo Ricardo Leitner result = nodeid_to_addr(con->nodeid, &daddr, NULL, true); 1060ee44b4bcSMarcelo Ricardo Leitner if (result < 0) { 10616ed7257bSPatrick Caulfield log_print("no address for nodeid %d", con->nodeid); 1062ee44b4bcSMarcelo Ricardo Leitner goto out; 106304bedd79SDavid Teigland } 10646ed7257bSPatrick Caulfield 1065ee44b4bcSMarcelo Ricardo Leitner /* Create a socket to communicate with */ 1066ee44b4bcSMarcelo Ricardo Leitner result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1067ee44b4bcSMarcelo Ricardo Leitner SOCK_STREAM, IPPROTO_SCTP, &sock); 1068ee44b4bcSMarcelo Ricardo Leitner if (result < 0) 1069ee44b4bcSMarcelo Ricardo Leitner goto socket_err; 10706ed7257bSPatrick Caulfield 1071ee44b4bcSMarcelo Ricardo Leitner con->rx_action = receive_from_sock; 1072ee44b4bcSMarcelo Ricardo Leitner con->connect_action = sctp_connect_to_sock; 1073988419a9Stsutomu.owa@toshiba.co.jp add_sock(sock, con); 10746ed7257bSPatrick Caulfield 1075ee44b4bcSMarcelo Ricardo Leitner /* Bind to all addresses. */ 1076ee44b4bcSMarcelo Ricardo Leitner if (sctp_bind_addrs(con, 0)) 1077ee44b4bcSMarcelo Ricardo Leitner goto bind_err; 107898e1b60eSMike Christie 1079ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len); 10806ed7257bSPatrick Caulfield 1081ee44b4bcSMarcelo Ricardo Leitner log_print("connecting to %d", con->nodeid); 10826ed7257bSPatrick Caulfield 1083ee44b4bcSMarcelo Ricardo Leitner /* Turn off Nagle's algorithm */ 1084b09c603cSGang He kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one, 1085ee44b4bcSMarcelo Ricardo Leitner sizeof(one)); 1086ee44b4bcSMarcelo Ricardo Leitner 1087f706d830SGang He /* 1088f706d830SGang He * Make sock->ops->connect() function return in specified time, 1089f706d830SGang He * since O_NONBLOCK argument in connect() function does not work here, 1090f706d830SGang He * then, we should restore the default value of this attribute. 1091f706d830SGang He */ 1092f706d830SGang He kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO, (char *)&tv, 1093f706d830SGang He sizeof(tv)); 1094ee44b4bcSMarcelo Ricardo Leitner result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len, 1095da3627c3SGang He 0); 1096f706d830SGang He memset(&tv, 0, sizeof(tv)); 1097f706d830SGang He kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO, (char *)&tv, 1098f706d830SGang He sizeof(tv)); 1099f706d830SGang He 1100ee44b4bcSMarcelo Ricardo Leitner if (result == -EINPROGRESS) 1101ee44b4bcSMarcelo Ricardo Leitner result = 0; 1102ee44b4bcSMarcelo Ricardo Leitner if (result == 0) 1103ee44b4bcSMarcelo Ricardo Leitner goto out; 1104ee44b4bcSMarcelo Ricardo Leitner 1105ee44b4bcSMarcelo Ricardo Leitner bind_err: 1106ee44b4bcSMarcelo Ricardo Leitner con->sock = NULL; 1107ee44b4bcSMarcelo Ricardo Leitner sock_release(sock); 1108ee44b4bcSMarcelo Ricardo Leitner 1109ee44b4bcSMarcelo Ricardo Leitner socket_err: 1110ee44b4bcSMarcelo Ricardo Leitner /* 1111ee44b4bcSMarcelo Ricardo Leitner * Some errors are fatal and this list might need adjusting. For other 1112ee44b4bcSMarcelo Ricardo Leitner * errors we try again until the max number of retries is reached. 1113ee44b4bcSMarcelo Ricardo Leitner */ 1114ee44b4bcSMarcelo Ricardo Leitner if (result != -EHOSTUNREACH && 1115ee44b4bcSMarcelo Ricardo Leitner result != -ENETUNREACH && 1116ee44b4bcSMarcelo Ricardo Leitner result != -ENETDOWN && 1117ee44b4bcSMarcelo Ricardo Leitner result != -EINVAL && 1118ee44b4bcSMarcelo Ricardo Leitner result != -EPROTONOSUPPORT) { 1119ee44b4bcSMarcelo Ricardo Leitner log_print("connect %d try %d error %d", con->nodeid, 1120ee44b4bcSMarcelo Ricardo Leitner con->retries, result); 1121ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 1122ee44b4bcSMarcelo Ricardo Leitner msleep(1000); 1123ee44b4bcSMarcelo Ricardo Leitner lowcomms_connect_sock(con); 1124ee44b4bcSMarcelo Ricardo Leitner return; 11256ed7257bSPatrick Caulfield } 11265d689871SMike Christie 1127ee44b4bcSMarcelo Ricardo Leitner out: 11285d689871SMike Christie mutex_unlock(&con->sock_mutex); 11296ed7257bSPatrick Caulfield } 11306ed7257bSPatrick Caulfield 11316ed7257bSPatrick Caulfield /* Connect a new socket to its peer */ 11326ed7257bSPatrick Caulfield static void tcp_connect_to_sock(struct connection *con) 11336ed7257bSPatrick Caulfield { 11346bd8fedaSLon Hohberger struct sockaddr_storage saddr, src_addr; 11356ed7257bSPatrick Caulfield int addr_len; 1136a89d63a1SCasey Dahlin struct socket *sock = NULL; 1137cb2d45daSDavid Teigland int one = 1; 113836b71a8bSDavid Teigland int result; 11396ed7257bSPatrick Caulfield 11406ed7257bSPatrick Caulfield if (con->nodeid == 0) { 11416ed7257bSPatrick Caulfield log_print("attempt to connect sock 0 foiled"); 11426ed7257bSPatrick Caulfield return; 11436ed7257bSPatrick Caulfield } 11446ed7257bSPatrick Caulfield 11456ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 11466ed7257bSPatrick Caulfield if (con->retries++ > MAX_CONNECT_RETRIES) 11476ed7257bSPatrick Caulfield goto out; 11486ed7257bSPatrick Caulfield 11496ed7257bSPatrick Caulfield /* Some odd races can cause double-connects, ignore them */ 115036b71a8bSDavid Teigland if (con->sock) 11516ed7257bSPatrick Caulfield goto out; 11526ed7257bSPatrick Caulfield 11536ed7257bSPatrick Caulfield /* Create a socket to communicate with */ 1154eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1155eeb1bd5cSEric W. Biederman SOCK_STREAM, IPPROTO_TCP, &sock); 11566ed7257bSPatrick Caulfield if (result < 0) 11576ed7257bSPatrick Caulfield goto out_err; 11586ed7257bSPatrick Caulfield 11596ed7257bSPatrick Caulfield memset(&saddr, 0, sizeof(saddr)); 116098e1b60eSMike Christie result = nodeid_to_addr(con->nodeid, &saddr, NULL, false); 116136b71a8bSDavid Teigland if (result < 0) { 116236b71a8bSDavid Teigland log_print("no address for nodeid %d", con->nodeid); 11636ed7257bSPatrick Caulfield goto out_err; 116436b71a8bSDavid Teigland } 11656ed7257bSPatrick Caulfield 11666ed7257bSPatrick Caulfield con->rx_action = receive_from_sock; 11676ed7257bSPatrick Caulfield con->connect_action = tcp_connect_to_sock; 1168988419a9Stsutomu.owa@toshiba.co.jp add_sock(sock, con); 11696ed7257bSPatrick Caulfield 11706bd8fedaSLon Hohberger /* Bind to our cluster-known address connecting to avoid 11716bd8fedaSLon Hohberger routing problems */ 11726bd8fedaSLon Hohberger memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); 11736bd8fedaSLon Hohberger make_sockaddr(&src_addr, 0, &addr_len); 11746bd8fedaSLon Hohberger result = sock->ops->bind(sock, (struct sockaddr *) &src_addr, 11756bd8fedaSLon Hohberger addr_len); 11766bd8fedaSLon Hohberger if (result < 0) { 11776bd8fedaSLon Hohberger log_print("could not bind for connect: %d", result); 11786bd8fedaSLon Hohberger /* This *may* not indicate a critical error */ 11796bd8fedaSLon Hohberger } 11806bd8fedaSLon Hohberger 11816ed7257bSPatrick Caulfield make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); 11826ed7257bSPatrick Caulfield 11836ed7257bSPatrick Caulfield log_print("connecting to %d", con->nodeid); 1184cb2d45daSDavid Teigland 1185cb2d45daSDavid Teigland /* Turn off Nagle's algorithm */ 1186cb2d45daSDavid Teigland kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, 1187cb2d45daSDavid Teigland sizeof(one)); 1188cb2d45daSDavid Teigland 118936b71a8bSDavid Teigland result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, 11906ed7257bSPatrick Caulfield O_NONBLOCK); 11916ed7257bSPatrick Caulfield if (result == -EINPROGRESS) 11926ed7257bSPatrick Caulfield result = 0; 11936ed7257bSPatrick Caulfield if (result == 0) 11946ed7257bSPatrick Caulfield goto out; 11956ed7257bSPatrick Caulfield 11966ed7257bSPatrick Caulfield out_err: 11976ed7257bSPatrick Caulfield if (con->sock) { 11986ed7257bSPatrick Caulfield sock_release(con->sock); 11996ed7257bSPatrick Caulfield con->sock = NULL; 1200a89d63a1SCasey Dahlin } else if (sock) { 1201a89d63a1SCasey Dahlin sock_release(sock); 12026ed7257bSPatrick Caulfield } 12036ed7257bSPatrick Caulfield /* 12046ed7257bSPatrick Caulfield * Some errors are fatal and this list might need adjusting. For other 12056ed7257bSPatrick Caulfield * errors we try again until the max number of retries is reached. 12066ed7257bSPatrick Caulfield */ 120736b71a8bSDavid Teigland if (result != -EHOSTUNREACH && 120836b71a8bSDavid Teigland result != -ENETUNREACH && 120936b71a8bSDavid Teigland result != -ENETDOWN && 121036b71a8bSDavid Teigland result != -EINVAL && 121136b71a8bSDavid Teigland result != -EPROTONOSUPPORT) { 121236b71a8bSDavid Teigland log_print("connect %d try %d error %d", con->nodeid, 121336b71a8bSDavid Teigland con->retries, result); 121436b71a8bSDavid Teigland mutex_unlock(&con->sock_mutex); 121536b71a8bSDavid Teigland msleep(1000); 12166ed7257bSPatrick Caulfield lowcomms_connect_sock(con); 121736b71a8bSDavid Teigland return; 12186ed7257bSPatrick Caulfield } 12196ed7257bSPatrick Caulfield out: 12206ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 12216ed7257bSPatrick Caulfield return; 12226ed7257bSPatrick Caulfield } 12236ed7257bSPatrick Caulfield 12246ed7257bSPatrick Caulfield static struct socket *tcp_create_listen_sock(struct connection *con, 12256ed7257bSPatrick Caulfield struct sockaddr_storage *saddr) 12266ed7257bSPatrick Caulfield { 12276ed7257bSPatrick Caulfield struct socket *sock = NULL; 12286ed7257bSPatrick Caulfield int result = 0; 12296ed7257bSPatrick Caulfield int one = 1; 12306ed7257bSPatrick Caulfield int addr_len; 12316ed7257bSPatrick Caulfield 12326ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) 12336ed7257bSPatrick Caulfield addr_len = sizeof(struct sockaddr_in); 12346ed7257bSPatrick Caulfield else 12356ed7257bSPatrick Caulfield addr_len = sizeof(struct sockaddr_in6); 12366ed7257bSPatrick Caulfield 12376ed7257bSPatrick Caulfield /* Create a socket to communicate with */ 1238eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1239eeb1bd5cSEric W. Biederman SOCK_STREAM, IPPROTO_TCP, &sock); 12406ed7257bSPatrick Caulfield if (result < 0) { 1241617e82e1SDavid Teigland log_print("Can't create listening comms socket"); 12426ed7257bSPatrick Caulfield goto create_out; 12436ed7257bSPatrick Caulfield } 12446ed7257bSPatrick Caulfield 1245cb2d45daSDavid Teigland /* Turn off Nagle's algorithm */ 1246cb2d45daSDavid Teigland kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, 1247cb2d45daSDavid Teigland sizeof(one)); 1248cb2d45daSDavid Teigland 12496ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, 12506ed7257bSPatrick Caulfield (char *)&one, sizeof(one)); 12516ed7257bSPatrick Caulfield 12526ed7257bSPatrick Caulfield if (result < 0) { 1253617e82e1SDavid Teigland log_print("Failed to set SO_REUSEADDR on socket: %d", result); 12546ed7257bSPatrick Caulfield } 125593eaadebStsutomu.owa@toshiba.co.jp write_lock_bh(&sock->sk->sk_callback_lock); 1256b81171cbSBob Peterson sock->sk->sk_user_data = con; 1257cc661fc9SBob Peterson save_listen_callbacks(sock); 12586ed7257bSPatrick Caulfield con->rx_action = tcp_accept_from_sock; 12596ed7257bSPatrick Caulfield con->connect_action = tcp_connect_to_sock; 126093eaadebStsutomu.owa@toshiba.co.jp write_unlock_bh(&sock->sk->sk_callback_lock); 12616ed7257bSPatrick Caulfield 12626ed7257bSPatrick Caulfield /* Bind to our port */ 12636ed7257bSPatrick Caulfield make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); 12646ed7257bSPatrick Caulfield result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); 12656ed7257bSPatrick Caulfield if (result < 0) { 1266617e82e1SDavid Teigland log_print("Can't bind to port %d", dlm_config.ci_tcp_port); 12676ed7257bSPatrick Caulfield sock_release(sock); 12686ed7257bSPatrick Caulfield sock = NULL; 12696ed7257bSPatrick Caulfield con->sock = NULL; 12706ed7257bSPatrick Caulfield goto create_out; 12716ed7257bSPatrick Caulfield } 12726ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 12736ed7257bSPatrick Caulfield (char *)&one, sizeof(one)); 12746ed7257bSPatrick Caulfield if (result < 0) { 1275617e82e1SDavid Teigland log_print("Set keepalive failed: %d", result); 12766ed7257bSPatrick Caulfield } 12776ed7257bSPatrick Caulfield 12786ed7257bSPatrick Caulfield result = sock->ops->listen(sock, 5); 12796ed7257bSPatrick Caulfield if (result < 0) { 1280617e82e1SDavid Teigland log_print("Can't listen on port %d", dlm_config.ci_tcp_port); 12816ed7257bSPatrick Caulfield sock_release(sock); 12826ed7257bSPatrick Caulfield sock = NULL; 12836ed7257bSPatrick Caulfield goto create_out; 12846ed7257bSPatrick Caulfield } 12856ed7257bSPatrick Caulfield 12866ed7257bSPatrick Caulfield create_out: 12876ed7257bSPatrick Caulfield return sock; 12886ed7257bSPatrick Caulfield } 12896ed7257bSPatrick Caulfield 12906ed7257bSPatrick Caulfield /* Get local addresses */ 12916ed7257bSPatrick Caulfield static void init_local(void) 12926ed7257bSPatrick Caulfield { 12936ed7257bSPatrick Caulfield struct sockaddr_storage sas, *addr; 12946ed7257bSPatrick Caulfield int i; 12956ed7257bSPatrick Caulfield 129630d3a237SPatrick Caulfield dlm_local_count = 0; 12971b189b88SDavid Teigland for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) { 12986ed7257bSPatrick Caulfield if (dlm_our_addr(&sas, i)) 12996ed7257bSPatrick Caulfield break; 13006ed7257bSPatrick Caulfield 13015c93f56fSAmitoj Kaur Chawla addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS); 13026ed7257bSPatrick Caulfield if (!addr) 13036ed7257bSPatrick Caulfield break; 13046ed7257bSPatrick Caulfield dlm_local_addr[dlm_local_count++] = addr; 13056ed7257bSPatrick Caulfield } 13066ed7257bSPatrick Caulfield } 13076ed7257bSPatrick Caulfield 13086ed7257bSPatrick Caulfield /* Initialise SCTP socket and bind to all interfaces */ 13096ed7257bSPatrick Caulfield static int sctp_listen_for_all(void) 13106ed7257bSPatrick Caulfield { 13116ed7257bSPatrick Caulfield struct socket *sock = NULL; 1312ee44b4bcSMarcelo Ricardo Leitner int result = -EINVAL; 1313573c24c4SDavid Teigland struct connection *con = nodeid2con(0, GFP_NOFS); 13146ed7257bSPatrick Caulfield int bufsize = NEEDED_RMEM; 131586e92ad2SMike Christie int one = 1; 13166ed7257bSPatrick Caulfield 13176ed7257bSPatrick Caulfield if (!con) 13186ed7257bSPatrick Caulfield return -ENOMEM; 13196ed7257bSPatrick Caulfield 13206ed7257bSPatrick Caulfield log_print("Using SCTP for communications"); 13216ed7257bSPatrick Caulfield 1322eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1323ee44b4bcSMarcelo Ricardo Leitner SOCK_STREAM, IPPROTO_SCTP, &sock); 13246ed7257bSPatrick Caulfield if (result < 0) { 13256ed7257bSPatrick Caulfield log_print("Can't create comms socket, check SCTP is loaded"); 13266ed7257bSPatrick Caulfield goto out; 13276ed7257bSPatrick Caulfield } 13286ed7257bSPatrick Caulfield 1329df61c952SDavid S. Miller result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, 13306ed7257bSPatrick Caulfield (char *)&bufsize, sizeof(bufsize)); 13316ed7257bSPatrick Caulfield if (result) 1332617e82e1SDavid Teigland log_print("Error increasing buffer space on socket %d", result); 13336ed7257bSPatrick Caulfield 133486e92ad2SMike Christie result = kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one, 133586e92ad2SMike Christie sizeof(one)); 133686e92ad2SMike Christie if (result < 0) 133786e92ad2SMike Christie log_print("Could not set SCTP NODELAY error %d\n", result); 133886e92ad2SMike Christie 1339b81171cbSBob Peterson write_lock_bh(&sock->sk->sk_callback_lock); 13406ed7257bSPatrick Caulfield /* Init con struct */ 13416ed7257bSPatrick Caulfield sock->sk->sk_user_data = con; 1342cc661fc9SBob Peterson save_listen_callbacks(sock); 13436ed7257bSPatrick Caulfield con->sock = sock; 13446ed7257bSPatrick Caulfield con->sock->sk->sk_data_ready = lowcomms_data_ready; 1345ee44b4bcSMarcelo Ricardo Leitner con->rx_action = sctp_accept_from_sock; 1346ee44b4bcSMarcelo Ricardo Leitner con->connect_action = sctp_connect_to_sock; 13476ed7257bSPatrick Caulfield 1348b81171cbSBob Peterson write_unlock_bh(&sock->sk->sk_callback_lock); 1349b81171cbSBob Peterson 1350ee44b4bcSMarcelo Ricardo Leitner /* Bind to all addresses. */ 1351ee44b4bcSMarcelo Ricardo Leitner if (sctp_bind_addrs(con, dlm_config.ci_tcp_port)) 13526ed7257bSPatrick Caulfield goto create_delsock; 13536ed7257bSPatrick Caulfield 13546ed7257bSPatrick Caulfield result = sock->ops->listen(sock, 5); 13556ed7257bSPatrick Caulfield if (result < 0) { 13566ed7257bSPatrick Caulfield log_print("Can't set socket listening"); 13576ed7257bSPatrick Caulfield goto create_delsock; 13586ed7257bSPatrick Caulfield } 13596ed7257bSPatrick Caulfield 13606ed7257bSPatrick Caulfield return 0; 13616ed7257bSPatrick Caulfield 13626ed7257bSPatrick Caulfield create_delsock: 13636ed7257bSPatrick Caulfield sock_release(sock); 13646ed7257bSPatrick Caulfield con->sock = NULL; 13656ed7257bSPatrick Caulfield out: 13666ed7257bSPatrick Caulfield return result; 13676ed7257bSPatrick Caulfield } 13686ed7257bSPatrick Caulfield 13696ed7257bSPatrick Caulfield static int tcp_listen_for_all(void) 13706ed7257bSPatrick Caulfield { 13716ed7257bSPatrick Caulfield struct socket *sock = NULL; 1372573c24c4SDavid Teigland struct connection *con = nodeid2con(0, GFP_NOFS); 13736ed7257bSPatrick Caulfield int result = -EINVAL; 13746ed7257bSPatrick Caulfield 13756ed7257bSPatrick Caulfield if (!con) 13766ed7257bSPatrick Caulfield return -ENOMEM; 13776ed7257bSPatrick Caulfield 13786ed7257bSPatrick Caulfield /* We don't support multi-homed hosts */ 13796ed7257bSPatrick Caulfield if (dlm_local_addr[1] != NULL) { 1380617e82e1SDavid Teigland log_print("TCP protocol can't handle multi-homed hosts, " 1381617e82e1SDavid Teigland "try SCTP"); 13826ed7257bSPatrick Caulfield return -EINVAL; 13836ed7257bSPatrick Caulfield } 13846ed7257bSPatrick Caulfield 13856ed7257bSPatrick Caulfield log_print("Using TCP for communications"); 13866ed7257bSPatrick Caulfield 13876ed7257bSPatrick Caulfield sock = tcp_create_listen_sock(con, dlm_local_addr[0]); 13886ed7257bSPatrick Caulfield if (sock) { 1389988419a9Stsutomu.owa@toshiba.co.jp add_sock(sock, con); 13906ed7257bSPatrick Caulfield result = 0; 13916ed7257bSPatrick Caulfield } 13926ed7257bSPatrick Caulfield else { 13936ed7257bSPatrick Caulfield result = -EADDRINUSE; 13946ed7257bSPatrick Caulfield } 13956ed7257bSPatrick Caulfield 13966ed7257bSPatrick Caulfield return result; 13976ed7257bSPatrick Caulfield } 13986ed7257bSPatrick Caulfield 13996ed7257bSPatrick Caulfield 14006ed7257bSPatrick Caulfield 14016ed7257bSPatrick Caulfield static struct writequeue_entry *new_writequeue_entry(struct connection *con, 14026ed7257bSPatrick Caulfield gfp_t allocation) 14036ed7257bSPatrick Caulfield { 14046ed7257bSPatrick Caulfield struct writequeue_entry *entry; 14056ed7257bSPatrick Caulfield 14066ed7257bSPatrick Caulfield entry = kmalloc(sizeof(struct writequeue_entry), allocation); 14076ed7257bSPatrick Caulfield if (!entry) 14086ed7257bSPatrick Caulfield return NULL; 14096ed7257bSPatrick Caulfield 14106ed7257bSPatrick Caulfield entry->page = alloc_page(allocation); 14116ed7257bSPatrick Caulfield if (!entry->page) { 14126ed7257bSPatrick Caulfield kfree(entry); 14136ed7257bSPatrick Caulfield return NULL; 14146ed7257bSPatrick Caulfield } 14156ed7257bSPatrick Caulfield 14166ed7257bSPatrick Caulfield entry->offset = 0; 14176ed7257bSPatrick Caulfield entry->len = 0; 14186ed7257bSPatrick Caulfield entry->end = 0; 14196ed7257bSPatrick Caulfield entry->users = 0; 14206ed7257bSPatrick Caulfield entry->con = con; 14216ed7257bSPatrick Caulfield 14226ed7257bSPatrick Caulfield return entry; 14236ed7257bSPatrick Caulfield } 14246ed7257bSPatrick Caulfield 1425617e82e1SDavid Teigland void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) 14266ed7257bSPatrick Caulfield { 14276ed7257bSPatrick Caulfield struct connection *con; 14286ed7257bSPatrick Caulfield struct writequeue_entry *e; 14296ed7257bSPatrick Caulfield int offset = 0; 14306ed7257bSPatrick Caulfield 14316ed7257bSPatrick Caulfield con = nodeid2con(nodeid, allocation); 14326ed7257bSPatrick Caulfield if (!con) 14336ed7257bSPatrick Caulfield return NULL; 14346ed7257bSPatrick Caulfield 14356ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14366ed7257bSPatrick Caulfield e = list_entry(con->writequeue.prev, struct writequeue_entry, list); 14376ed7257bSPatrick Caulfield if ((&e->list == &con->writequeue) || 143809cbfeafSKirill A. Shutemov (PAGE_SIZE - e->end < len)) { 14396ed7257bSPatrick Caulfield e = NULL; 14406ed7257bSPatrick Caulfield } else { 14416ed7257bSPatrick Caulfield offset = e->end; 14426ed7257bSPatrick Caulfield e->end += len; 1443eeee2b5fSWei Yongjun e->users++; 14446ed7257bSPatrick Caulfield } 14456ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14466ed7257bSPatrick Caulfield 14476ed7257bSPatrick Caulfield if (e) { 14486ed7257bSPatrick Caulfield got_one: 14496ed7257bSPatrick Caulfield *ppc = page_address(e->page) + offset; 14506ed7257bSPatrick Caulfield return e; 14516ed7257bSPatrick Caulfield } 14526ed7257bSPatrick Caulfield 14536ed7257bSPatrick Caulfield e = new_writequeue_entry(con, allocation); 14546ed7257bSPatrick Caulfield if (e) { 14556ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14566ed7257bSPatrick Caulfield offset = e->end; 14576ed7257bSPatrick Caulfield e->end += len; 1458eeee2b5fSWei Yongjun e->users++; 14596ed7257bSPatrick Caulfield list_add_tail(&e->list, &con->writequeue); 14606ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14616ed7257bSPatrick Caulfield goto got_one; 14626ed7257bSPatrick Caulfield } 14636ed7257bSPatrick Caulfield return NULL; 14646ed7257bSPatrick Caulfield } 14656ed7257bSPatrick Caulfield 14666ed7257bSPatrick Caulfield void dlm_lowcomms_commit_buffer(void *mh) 14676ed7257bSPatrick Caulfield { 14686ed7257bSPatrick Caulfield struct writequeue_entry *e = (struct writequeue_entry *)mh; 14696ed7257bSPatrick Caulfield struct connection *con = e->con; 14706ed7257bSPatrick Caulfield int users; 14716ed7257bSPatrick Caulfield 14726ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14736ed7257bSPatrick Caulfield users = --e->users; 14746ed7257bSPatrick Caulfield if (users) 14756ed7257bSPatrick Caulfield goto out; 14766ed7257bSPatrick Caulfield e->len = e->end - e->offset; 14776ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14786ed7257bSPatrick Caulfield 14796ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 14806ed7257bSPatrick Caulfield return; 14816ed7257bSPatrick Caulfield 14826ed7257bSPatrick Caulfield out: 14836ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14846ed7257bSPatrick Caulfield return; 14856ed7257bSPatrick Caulfield } 14866ed7257bSPatrick Caulfield 14876ed7257bSPatrick Caulfield /* Send a message */ 14886ed7257bSPatrick Caulfield static void send_to_sock(struct connection *con) 14896ed7257bSPatrick Caulfield { 14906ed7257bSPatrick Caulfield int ret = 0; 14916ed7257bSPatrick Caulfield const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; 14926ed7257bSPatrick Caulfield struct writequeue_entry *e; 14936ed7257bSPatrick Caulfield int len, offset; 1494f92c8dd7SBob Peterson int count = 0; 14956ed7257bSPatrick Caulfield 14966ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 14976ed7257bSPatrick Caulfield if (con->sock == NULL) 14986ed7257bSPatrick Caulfield goto out_connect; 14996ed7257bSPatrick Caulfield 15006ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 15016ed7257bSPatrick Caulfield for (;;) { 15026ed7257bSPatrick Caulfield e = list_entry(con->writequeue.next, struct writequeue_entry, 15036ed7257bSPatrick Caulfield list); 15046ed7257bSPatrick Caulfield if ((struct list_head *) e == &con->writequeue) 15056ed7257bSPatrick Caulfield break; 15066ed7257bSPatrick Caulfield 15076ed7257bSPatrick Caulfield len = e->len; 15086ed7257bSPatrick Caulfield offset = e->offset; 15096ed7257bSPatrick Caulfield BUG_ON(len == 0 && e->users == 0); 15106ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15116ed7257bSPatrick Caulfield 15126ed7257bSPatrick Caulfield ret = 0; 15136ed7257bSPatrick Caulfield if (len) { 15141329e3f2SPaolo Bonzini ret = kernel_sendpage(con->sock, e->page, offset, len, 15156ed7257bSPatrick Caulfield msg_flags); 1516d66f8277SPatrick Caulfield if (ret == -EAGAIN || ret == 0) { 1517b36930ddSDavid Miller if (ret == -EAGAIN && 15189cd3e072SEric Dumazet test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) && 1519b36930ddSDavid Miller !test_and_set_bit(CF_APP_LIMITED, &con->flags)) { 1520b36930ddSDavid Miller /* Notify TCP that we're limited by the 1521b36930ddSDavid Miller * application window size. 1522b36930ddSDavid Miller */ 1523b36930ddSDavid Miller set_bit(SOCK_NOSPACE, &con->sock->flags); 1524b36930ddSDavid Miller con->sock->sk->sk_write_pending++; 1525b36930ddSDavid Miller } 1526d66f8277SPatrick Caulfield cond_resched(); 15276ed7257bSPatrick Caulfield goto out; 15289c5bef58SYing Xue } else if (ret < 0) 15296ed7257bSPatrick Caulfield goto send_error; 1530d66f8277SPatrick Caulfield } 1531f92c8dd7SBob Peterson 15326ed7257bSPatrick Caulfield /* Don't starve people filling buffers */ 1533f92c8dd7SBob Peterson if (++count >= MAX_SEND_MSG_COUNT) { 15346ed7257bSPatrick Caulfield cond_resched(); 1535f92c8dd7SBob Peterson count = 0; 1536f92c8dd7SBob Peterson } 15376ed7257bSPatrick Caulfield 15386ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 15395d689871SMike Christie writequeue_entry_complete(e, ret); 15406ed7257bSPatrick Caulfield } 15416ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15426ed7257bSPatrick Caulfield out: 15436ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 15446ed7257bSPatrick Caulfield return; 15456ed7257bSPatrick Caulfield 15466ed7257bSPatrick Caulfield send_error: 15476ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 1548c553e173Stsutomu.owa@toshiba.co.jp close_connection(con, true, false, true); 154901da24d3SBob Peterson /* Requeue the send work. When the work daemon runs again, it will try 155001da24d3SBob Peterson a new connection, then call this function again. */ 155101da24d3SBob Peterson queue_work(send_workqueue, &con->swork); 15526ed7257bSPatrick Caulfield return; 15536ed7257bSPatrick Caulfield 15546ed7257bSPatrick Caulfield out_connect: 15556ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 155601da24d3SBob Peterson queue_work(send_workqueue, &con->swork); 155701da24d3SBob Peterson cond_resched(); 15586ed7257bSPatrick Caulfield } 15596ed7257bSPatrick Caulfield 15606ed7257bSPatrick Caulfield static void clean_one_writequeue(struct connection *con) 15616ed7257bSPatrick Caulfield { 15625e9ccc37SChristine Caulfield struct writequeue_entry *e, *safe; 15636ed7257bSPatrick Caulfield 15646ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 15655e9ccc37SChristine Caulfield list_for_each_entry_safe(e, safe, &con->writequeue, list) { 15666ed7257bSPatrick Caulfield list_del(&e->list); 15676ed7257bSPatrick Caulfield free_entry(e); 15686ed7257bSPatrick Caulfield } 15696ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15706ed7257bSPatrick Caulfield } 15716ed7257bSPatrick Caulfield 15726ed7257bSPatrick Caulfield /* Called from recovery when it knows that a node has 15736ed7257bSPatrick Caulfield left the cluster */ 15746ed7257bSPatrick Caulfield int dlm_lowcomms_close(int nodeid) 15756ed7257bSPatrick Caulfield { 15766ed7257bSPatrick Caulfield struct connection *con; 157736b71a8bSDavid Teigland struct dlm_node_addr *na; 15786ed7257bSPatrick Caulfield 15796ed7257bSPatrick Caulfield log_print("closing connection to node %d", nodeid); 15806ed7257bSPatrick Caulfield con = nodeid2con(nodeid, 0); 15816ed7257bSPatrick Caulfield if (con) { 1582063c4c99SLars Marowsky-Bree set_bit(CF_CLOSE, &con->flags); 15830d737a8cSMarcelo Ricardo Leitner close_connection(con, true, true, true); 15846ed7257bSPatrick Caulfield clean_one_writequeue(con); 15856ed7257bSPatrick Caulfield } 158636b71a8bSDavid Teigland 158736b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 158836b71a8bSDavid Teigland na = find_node_addr(nodeid); 158936b71a8bSDavid Teigland if (na) { 159036b71a8bSDavid Teigland list_del(&na->list); 159136b71a8bSDavid Teigland while (na->addr_count--) 159236b71a8bSDavid Teigland kfree(na->addr[na->addr_count]); 159336b71a8bSDavid Teigland kfree(na); 159436b71a8bSDavid Teigland } 159536b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 159636b71a8bSDavid Teigland 15976ed7257bSPatrick Caulfield return 0; 15986ed7257bSPatrick Caulfield } 15996ed7257bSPatrick Caulfield 16006ed7257bSPatrick Caulfield /* Receive workqueue function */ 16016ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work) 16026ed7257bSPatrick Caulfield { 16036ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, rwork); 16046ed7257bSPatrick Caulfield int err; 16056ed7257bSPatrick Caulfield 16066ed7257bSPatrick Caulfield clear_bit(CF_READ_PENDING, &con->flags); 16076ed7257bSPatrick Caulfield do { 16086ed7257bSPatrick Caulfield err = con->rx_action(con); 16096ed7257bSPatrick Caulfield } while (!err); 16106ed7257bSPatrick Caulfield } 16116ed7257bSPatrick Caulfield 16126ed7257bSPatrick Caulfield /* Send workqueue function */ 16136ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work) 16146ed7257bSPatrick Caulfield { 16156ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, swork); 16166ed7257bSPatrick Caulfield 16178a4abb08Stsutomu.owa@toshiba.co.jp clear_bit(CF_WRITE_PENDING, &con->flags); 161861d9102bSBob Peterson if (con->sock == NULL) /* not mutex protected so check it inside too */ 16196ed7257bSPatrick Caulfield con->connect_action(con); 162001da24d3SBob Peterson if (!list_empty(&con->writequeue)) 16216ed7257bSPatrick Caulfield send_to_sock(con); 16226ed7257bSPatrick Caulfield } 16236ed7257bSPatrick Caulfield 16246ed7257bSPatrick Caulfield 16256ed7257bSPatrick Caulfield /* Discard all entries on the write queues */ 16266ed7257bSPatrick Caulfield static void clean_writequeues(void) 16276ed7257bSPatrick Caulfield { 16285e9ccc37SChristine Caulfield foreach_conn(clean_one_writequeue); 16296ed7257bSPatrick Caulfield } 16306ed7257bSPatrick Caulfield 16316ed7257bSPatrick Caulfield static void work_stop(void) 16326ed7257bSPatrick Caulfield { 16336ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 16346ed7257bSPatrick Caulfield destroy_workqueue(send_workqueue); 16356ed7257bSPatrick Caulfield } 16366ed7257bSPatrick Caulfield 16376ed7257bSPatrick Caulfield static int work_start(void) 16386ed7257bSPatrick Caulfield { 1639e43f055aSDavid Teigland recv_workqueue = alloc_workqueue("dlm_recv", 1640e43f055aSDavid Teigland WQ_UNBOUND | WQ_MEM_RECLAIM, 1); 1641b9d41052SNamhyung Kim if (!recv_workqueue) { 1642b9d41052SNamhyung Kim log_print("can't start dlm_recv"); 1643b9d41052SNamhyung Kim return -ENOMEM; 16446ed7257bSPatrick Caulfield } 16456ed7257bSPatrick Caulfield 1646e43f055aSDavid Teigland send_workqueue = alloc_workqueue("dlm_send", 1647e43f055aSDavid Teigland WQ_UNBOUND | WQ_MEM_RECLAIM, 1); 1648b9d41052SNamhyung Kim if (!send_workqueue) { 1649b9d41052SNamhyung Kim log_print("can't start dlm_send"); 16506ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 1651b9d41052SNamhyung Kim return -ENOMEM; 16526ed7257bSPatrick Caulfield } 16536ed7257bSPatrick Caulfield 16546ed7257bSPatrick Caulfield return 0; 16556ed7257bSPatrick Caulfield } 16566ed7257bSPatrick Caulfield 1657f0fb83cbStsutomu.owa@toshiba.co.jp static void _stop_conn(struct connection *con, bool and_other) 16586ed7257bSPatrick Caulfield { 1659f0fb83cbStsutomu.owa@toshiba.co.jp mutex_lock(&con->sock_mutex); 1660173a31feStsutomu.owa@toshiba.co.jp set_bit(CF_CLOSE, &con->flags); 1661f0fb83cbStsutomu.owa@toshiba.co.jp set_bit(CF_READ_PENDING, &con->flags); 16628a4abb08Stsutomu.owa@toshiba.co.jp set_bit(CF_WRITE_PENDING, &con->flags); 166393eaadebStsutomu.owa@toshiba.co.jp if (con->sock && con->sock->sk) { 166493eaadebStsutomu.owa@toshiba.co.jp write_lock_bh(&con->sock->sk->sk_callback_lock); 1665afb853fbSPatrick Caulfield con->sock->sk->sk_user_data = NULL; 166693eaadebStsutomu.owa@toshiba.co.jp write_unlock_bh(&con->sock->sk->sk_callback_lock); 166793eaadebStsutomu.owa@toshiba.co.jp } 1668f0fb83cbStsutomu.owa@toshiba.co.jp if (con->othercon && and_other) 1669f0fb83cbStsutomu.owa@toshiba.co.jp _stop_conn(con->othercon, false); 1670f0fb83cbStsutomu.owa@toshiba.co.jp mutex_unlock(&con->sock_mutex); 1671f0fb83cbStsutomu.owa@toshiba.co.jp } 1672f0fb83cbStsutomu.owa@toshiba.co.jp 1673f0fb83cbStsutomu.owa@toshiba.co.jp static void stop_conn(struct connection *con) 1674f0fb83cbStsutomu.owa@toshiba.co.jp { 1675f0fb83cbStsutomu.owa@toshiba.co.jp _stop_conn(con, true); 1676afb853fbSPatrick Caulfield } 16775e9ccc37SChristine Caulfield 16785e9ccc37SChristine Caulfield static void free_conn(struct connection *con) 16795e9ccc37SChristine Caulfield { 16800d737a8cSMarcelo Ricardo Leitner close_connection(con, true, true, true); 16815e9ccc37SChristine Caulfield if (con->othercon) 16825e9ccc37SChristine Caulfield kmem_cache_free(con_cache, con->othercon); 16835e9ccc37SChristine Caulfield hlist_del(&con->list); 16845e9ccc37SChristine Caulfield kmem_cache_free(con_cache, con); 16856ed7257bSPatrick Caulfield } 16865e9ccc37SChristine Caulfield 1687f0fb83cbStsutomu.owa@toshiba.co.jp static void work_flush(void) 1688f0fb83cbStsutomu.owa@toshiba.co.jp { 1689f0fb83cbStsutomu.owa@toshiba.co.jp int ok; 1690f0fb83cbStsutomu.owa@toshiba.co.jp int i; 1691f0fb83cbStsutomu.owa@toshiba.co.jp struct hlist_node *n; 1692f0fb83cbStsutomu.owa@toshiba.co.jp struct connection *con; 1693f0fb83cbStsutomu.owa@toshiba.co.jp 1694f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(recv_workqueue); 1695f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(send_workqueue); 1696f0fb83cbStsutomu.owa@toshiba.co.jp do { 1697f0fb83cbStsutomu.owa@toshiba.co.jp ok = 1; 1698f0fb83cbStsutomu.owa@toshiba.co.jp foreach_conn(stop_conn); 1699f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(recv_workqueue); 1700f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(send_workqueue); 1701f0fb83cbStsutomu.owa@toshiba.co.jp for (i = 0; i < CONN_HASH_SIZE && ok; i++) { 1702f0fb83cbStsutomu.owa@toshiba.co.jp hlist_for_each_entry_safe(con, n, 1703f0fb83cbStsutomu.owa@toshiba.co.jp &connection_hash[i], list) { 1704f0fb83cbStsutomu.owa@toshiba.co.jp ok &= test_bit(CF_READ_PENDING, &con->flags); 17058a4abb08Stsutomu.owa@toshiba.co.jp ok &= test_bit(CF_WRITE_PENDING, &con->flags); 17068a4abb08Stsutomu.owa@toshiba.co.jp if (con->othercon) { 1707f0fb83cbStsutomu.owa@toshiba.co.jp ok &= test_bit(CF_READ_PENDING, 1708f0fb83cbStsutomu.owa@toshiba.co.jp &con->othercon->flags); 17098a4abb08Stsutomu.owa@toshiba.co.jp ok &= test_bit(CF_WRITE_PENDING, 17108a4abb08Stsutomu.owa@toshiba.co.jp &con->othercon->flags); 17118a4abb08Stsutomu.owa@toshiba.co.jp } 1712f0fb83cbStsutomu.owa@toshiba.co.jp } 1713f0fb83cbStsutomu.owa@toshiba.co.jp } 1714f0fb83cbStsutomu.owa@toshiba.co.jp } while (!ok); 1715f0fb83cbStsutomu.owa@toshiba.co.jp } 1716f0fb83cbStsutomu.owa@toshiba.co.jp 17175e9ccc37SChristine Caulfield void dlm_lowcomms_stop(void) 17185e9ccc37SChristine Caulfield { 17195e9ccc37SChristine Caulfield /* Set all the flags to prevent any 17205e9ccc37SChristine Caulfield socket activity. 17215e9ccc37SChristine Caulfield */ 17225e9ccc37SChristine Caulfield mutex_lock(&connections_lock); 1723513ef596SDavid Teigland dlm_allow_conn = 0; 1724f0fb83cbStsutomu.owa@toshiba.co.jp mutex_unlock(&connections_lock); 1725f0fb83cbStsutomu.owa@toshiba.co.jp work_flush(); 17263a8db798SMarcelo Ricardo Leitner clean_writequeues(); 17273a8db798SMarcelo Ricardo Leitner foreach_conn(free_conn); 17286ed7257bSPatrick Caulfield work_stop(); 17296ed7257bSPatrick Caulfield 17306ed7257bSPatrick Caulfield kmem_cache_destroy(con_cache); 17316ed7257bSPatrick Caulfield } 17326ed7257bSPatrick Caulfield 17336ed7257bSPatrick Caulfield int dlm_lowcomms_start(void) 17346ed7257bSPatrick Caulfield { 17356ed7257bSPatrick Caulfield int error = -EINVAL; 17366ed7257bSPatrick Caulfield struct connection *con; 17375e9ccc37SChristine Caulfield int i; 17385e9ccc37SChristine Caulfield 17395e9ccc37SChristine Caulfield for (i = 0; i < CONN_HASH_SIZE; i++) 17405e9ccc37SChristine Caulfield INIT_HLIST_HEAD(&connection_hash[i]); 17416ed7257bSPatrick Caulfield 17426ed7257bSPatrick Caulfield init_local(); 17436ed7257bSPatrick Caulfield if (!dlm_local_count) { 1744617e82e1SDavid Teigland error = -ENOTCONN; 17456ed7257bSPatrick Caulfield log_print("no local IP address has been set"); 1746513ef596SDavid Teigland goto fail; 17476ed7257bSPatrick Caulfield } 17486ed7257bSPatrick Caulfield 17496ed7257bSPatrick Caulfield error = -ENOMEM; 17506ed7257bSPatrick Caulfield con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), 17516ed7257bSPatrick Caulfield __alignof__(struct connection), 0, 175220c2df83SPaul Mundt NULL); 17536ed7257bSPatrick Caulfield if (!con_cache) 1754513ef596SDavid Teigland goto fail; 1755513ef596SDavid Teigland 1756513ef596SDavid Teigland error = work_start(); 1757513ef596SDavid Teigland if (error) 1758513ef596SDavid Teigland goto fail_destroy; 1759513ef596SDavid Teigland 1760513ef596SDavid Teigland dlm_allow_conn = 1; 17616ed7257bSPatrick Caulfield 17626ed7257bSPatrick Caulfield /* Start listening */ 17636ed7257bSPatrick Caulfield if (dlm_config.ci_protocol == 0) 17646ed7257bSPatrick Caulfield error = tcp_listen_for_all(); 17656ed7257bSPatrick Caulfield else 17666ed7257bSPatrick Caulfield error = sctp_listen_for_all(); 17676ed7257bSPatrick Caulfield if (error) 17686ed7257bSPatrick Caulfield goto fail_unlisten; 17696ed7257bSPatrick Caulfield 17706ed7257bSPatrick Caulfield return 0; 17716ed7257bSPatrick Caulfield 17726ed7257bSPatrick Caulfield fail_unlisten: 1773513ef596SDavid Teigland dlm_allow_conn = 0; 17746ed7257bSPatrick Caulfield con = nodeid2con(0,0); 17756ed7257bSPatrick Caulfield if (con) { 17760d737a8cSMarcelo Ricardo Leitner close_connection(con, false, true, true); 17776ed7257bSPatrick Caulfield kmem_cache_free(con_cache, con); 17786ed7257bSPatrick Caulfield } 1779513ef596SDavid Teigland fail_destroy: 17806ed7257bSPatrick Caulfield kmem_cache_destroy(con_cache); 1781513ef596SDavid Teigland fail: 17826ed7257bSPatrick Caulfield return error; 17836ed7257bSPatrick Caulfield } 178436b71a8bSDavid Teigland 178536b71a8bSDavid Teigland void dlm_lowcomms_exit(void) 178636b71a8bSDavid Teigland { 178736b71a8bSDavid Teigland struct dlm_node_addr *na, *safe; 178836b71a8bSDavid Teigland 178936b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 179036b71a8bSDavid Teigland list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) { 179136b71a8bSDavid Teigland list_del(&na->list); 179236b71a8bSDavid Teigland while (na->addr_count--) 179336b71a8bSDavid Teigland kfree(na->addr[na->addr_count]); 179436b71a8bSDavid Teigland kfree(na); 179536b71a8bSDavid Teigland } 179636b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 179736b71a8bSDavid Teigland } 1798