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 1096ed7257bSPatrick Caulfield #define CF_WRITE_PENDING 2 1106ed7257bSPatrick Caulfield #define CF_CONNECT_PENDING 3 1116ed7257bSPatrick Caulfield #define CF_INIT_PENDING 4 1126ed7257bSPatrick Caulfield #define CF_IS_OTHERCON 5 113063c4c99SLars Marowsky-Bree #define CF_CLOSE 6 114b36930ddSDavid Miller #define CF_APP_LIMITED 7 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 */ 127b81171cbSBob Peterson void (*orig_error_report)(struct sock *); 128b81171cbSBob Peterson void (*orig_data_ready)(struct sock *); 129b81171cbSBob Peterson void (*orig_state_change)(struct sock *); 130b81171cbSBob Peterson void (*orig_write_space)(struct sock *); 1316ed7257bSPatrick Caulfield }; 1326ed7257bSPatrick Caulfield #define sock2con(x) ((struct connection *)(x)->sk_user_data) 1336ed7257bSPatrick Caulfield 1346ed7257bSPatrick Caulfield /* An entry waiting to be sent */ 1356ed7257bSPatrick Caulfield struct writequeue_entry { 1366ed7257bSPatrick Caulfield struct list_head list; 1376ed7257bSPatrick Caulfield struct page *page; 1386ed7257bSPatrick Caulfield int offset; 1396ed7257bSPatrick Caulfield int len; 1406ed7257bSPatrick Caulfield int end; 1416ed7257bSPatrick Caulfield int users; 1426ed7257bSPatrick Caulfield struct connection *con; 1436ed7257bSPatrick Caulfield }; 1446ed7257bSPatrick Caulfield 14536b71a8bSDavid Teigland struct dlm_node_addr { 14636b71a8bSDavid Teigland struct list_head list; 14736b71a8bSDavid Teigland int nodeid; 14836b71a8bSDavid Teigland int addr_count; 14998e1b60eSMike Christie int curr_addr_index; 15036b71a8bSDavid Teigland struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT]; 15136b71a8bSDavid Teigland }; 15236b71a8bSDavid Teigland 15336b71a8bSDavid Teigland static LIST_HEAD(dlm_node_addrs); 15436b71a8bSDavid Teigland static DEFINE_SPINLOCK(dlm_node_addrs_spin); 15536b71a8bSDavid Teigland 1566ed7257bSPatrick Caulfield static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; 1576ed7257bSPatrick Caulfield static int dlm_local_count; 158513ef596SDavid Teigland static int dlm_allow_conn; 1596ed7257bSPatrick Caulfield 1606ed7257bSPatrick Caulfield /* Work queues */ 1616ed7257bSPatrick Caulfield static struct workqueue_struct *recv_workqueue; 1626ed7257bSPatrick Caulfield static struct workqueue_struct *send_workqueue; 1636ed7257bSPatrick Caulfield 1645e9ccc37SChristine Caulfield static struct hlist_head connection_hash[CONN_HASH_SIZE]; 1657a936ce7SMatthias Kaehlcke static DEFINE_MUTEX(connections_lock); 1666ed7257bSPatrick Caulfield static struct kmem_cache *con_cache; 1676ed7257bSPatrick Caulfield 1686ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work); 1696ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work); 1706ed7257bSPatrick Caulfield 1715e9ccc37SChristine Caulfield 1725e9ccc37SChristine Caulfield /* This is deliberately very simple because most clusters have simple 1735e9ccc37SChristine Caulfield sequential nodeids, so we should be able to go straight to a connection 1745e9ccc37SChristine Caulfield struct in the array */ 1755e9ccc37SChristine Caulfield static inline int nodeid_hash(int nodeid) 1765e9ccc37SChristine Caulfield { 1775e9ccc37SChristine Caulfield return nodeid & (CONN_HASH_SIZE-1); 1785e9ccc37SChristine Caulfield } 1795e9ccc37SChristine Caulfield 1805e9ccc37SChristine Caulfield static struct connection *__find_con(int nodeid) 1815e9ccc37SChristine Caulfield { 1825e9ccc37SChristine Caulfield int r; 1835e9ccc37SChristine Caulfield struct connection *con; 1845e9ccc37SChristine Caulfield 1855e9ccc37SChristine Caulfield r = nodeid_hash(nodeid); 1865e9ccc37SChristine Caulfield 187b67bfe0dSSasha Levin hlist_for_each_entry(con, &connection_hash[r], list) { 1885e9ccc37SChristine Caulfield if (con->nodeid == nodeid) 1895e9ccc37SChristine Caulfield return con; 1905e9ccc37SChristine Caulfield } 1915e9ccc37SChristine Caulfield return NULL; 1925e9ccc37SChristine Caulfield } 1935e9ccc37SChristine Caulfield 1946ed7257bSPatrick Caulfield /* 1956ed7257bSPatrick Caulfield * If 'allocation' is zero then we don't attempt to create a new 1966ed7257bSPatrick Caulfield * connection structure for this node. 1976ed7257bSPatrick Caulfield */ 1986ed7257bSPatrick Caulfield static struct connection *__nodeid2con(int nodeid, gfp_t alloc) 1996ed7257bSPatrick Caulfield { 2006ed7257bSPatrick Caulfield struct connection *con = NULL; 2016ed7257bSPatrick Caulfield int r; 2026ed7257bSPatrick Caulfield 2035e9ccc37SChristine Caulfield con = __find_con(nodeid); 2046ed7257bSPatrick Caulfield if (con || !alloc) 2056ed7257bSPatrick Caulfield return con; 2066ed7257bSPatrick Caulfield 2076ed7257bSPatrick Caulfield con = kmem_cache_zalloc(con_cache, alloc); 2086ed7257bSPatrick Caulfield if (!con) 2096ed7257bSPatrick Caulfield return NULL; 2106ed7257bSPatrick Caulfield 2115e9ccc37SChristine Caulfield r = nodeid_hash(nodeid); 2125e9ccc37SChristine Caulfield hlist_add_head(&con->list, &connection_hash[r]); 2136ed7257bSPatrick Caulfield 2146ed7257bSPatrick Caulfield con->nodeid = nodeid; 2156ed7257bSPatrick Caulfield mutex_init(&con->sock_mutex); 2166ed7257bSPatrick Caulfield INIT_LIST_HEAD(&con->writequeue); 2176ed7257bSPatrick Caulfield spin_lock_init(&con->writequeue_lock); 2186ed7257bSPatrick Caulfield INIT_WORK(&con->swork, process_send_sockets); 2196ed7257bSPatrick Caulfield INIT_WORK(&con->rwork, process_recv_sockets); 2206ed7257bSPatrick Caulfield 2216ed7257bSPatrick Caulfield /* Setup action pointers for child sockets */ 2226ed7257bSPatrick Caulfield if (con->nodeid) { 2235e9ccc37SChristine Caulfield struct connection *zerocon = __find_con(0); 2246ed7257bSPatrick Caulfield 2256ed7257bSPatrick Caulfield con->connect_action = zerocon->connect_action; 2266ed7257bSPatrick Caulfield if (!con->rx_action) 2276ed7257bSPatrick Caulfield con->rx_action = zerocon->rx_action; 2286ed7257bSPatrick Caulfield } 2296ed7257bSPatrick Caulfield 2306ed7257bSPatrick Caulfield return con; 2316ed7257bSPatrick Caulfield } 2326ed7257bSPatrick Caulfield 2335e9ccc37SChristine Caulfield /* Loop round all connections */ 2345e9ccc37SChristine Caulfield static void foreach_conn(void (*conn_func)(struct connection *c)) 2355e9ccc37SChristine Caulfield { 2365e9ccc37SChristine Caulfield int i; 237b67bfe0dSSasha Levin struct hlist_node *n; 2385e9ccc37SChristine Caulfield struct connection *con; 2395e9ccc37SChristine Caulfield 2405e9ccc37SChristine Caulfield for (i = 0; i < CONN_HASH_SIZE; i++) { 241b67bfe0dSSasha Levin hlist_for_each_entry_safe(con, n, &connection_hash[i], list) 2425e9ccc37SChristine Caulfield conn_func(con); 2435e9ccc37SChristine Caulfield } 2445e9ccc37SChristine Caulfield } 2455e9ccc37SChristine Caulfield 2466ed7257bSPatrick Caulfield static struct connection *nodeid2con(int nodeid, gfp_t allocation) 2476ed7257bSPatrick Caulfield { 2486ed7257bSPatrick Caulfield struct connection *con; 2496ed7257bSPatrick Caulfield 2507a936ce7SMatthias Kaehlcke mutex_lock(&connections_lock); 2516ed7257bSPatrick Caulfield con = __nodeid2con(nodeid, allocation); 2527a936ce7SMatthias Kaehlcke mutex_unlock(&connections_lock); 2536ed7257bSPatrick Caulfield 2546ed7257bSPatrick Caulfield return con; 2556ed7257bSPatrick Caulfield } 2566ed7257bSPatrick Caulfield 25736b71a8bSDavid Teigland static struct dlm_node_addr *find_node_addr(int nodeid) 2586ed7257bSPatrick Caulfield { 25936b71a8bSDavid Teigland struct dlm_node_addr *na; 26036b71a8bSDavid Teigland 26136b71a8bSDavid Teigland list_for_each_entry(na, &dlm_node_addrs, list) { 26236b71a8bSDavid Teigland if (na->nodeid == nodeid) 26336b71a8bSDavid Teigland return na; 26436b71a8bSDavid Teigland } 26536b71a8bSDavid Teigland return NULL; 26636b71a8bSDavid Teigland } 26736b71a8bSDavid Teigland 26836b71a8bSDavid Teigland static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y) 26936b71a8bSDavid Teigland { 27036b71a8bSDavid Teigland switch (x->ss_family) { 27136b71a8bSDavid Teigland case AF_INET: { 27236b71a8bSDavid Teigland struct sockaddr_in *sinx = (struct sockaddr_in *)x; 27336b71a8bSDavid Teigland struct sockaddr_in *siny = (struct sockaddr_in *)y; 27436b71a8bSDavid Teigland if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr) 27536b71a8bSDavid Teigland return 0; 27636b71a8bSDavid Teigland if (sinx->sin_port != siny->sin_port) 27736b71a8bSDavid Teigland return 0; 27836b71a8bSDavid Teigland break; 27936b71a8bSDavid Teigland } 28036b71a8bSDavid Teigland case AF_INET6: { 28136b71a8bSDavid Teigland struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x; 28236b71a8bSDavid Teigland struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y; 28336b71a8bSDavid Teigland if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr)) 28436b71a8bSDavid Teigland return 0; 28536b71a8bSDavid Teigland if (sinx->sin6_port != siny->sin6_port) 28636b71a8bSDavid Teigland return 0; 28736b71a8bSDavid Teigland break; 28836b71a8bSDavid Teigland } 28936b71a8bSDavid Teigland default: 29036b71a8bSDavid Teigland return 0; 29136b71a8bSDavid Teigland } 29236b71a8bSDavid Teigland return 1; 29336b71a8bSDavid Teigland } 29436b71a8bSDavid Teigland 29536b71a8bSDavid Teigland static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out, 29698e1b60eSMike Christie struct sockaddr *sa_out, bool try_new_addr) 29736b71a8bSDavid Teigland { 29836b71a8bSDavid Teigland struct sockaddr_storage sas; 29936b71a8bSDavid Teigland struct dlm_node_addr *na; 3006ed7257bSPatrick Caulfield 3016ed7257bSPatrick Caulfield if (!dlm_local_count) 3026ed7257bSPatrick Caulfield return -1; 3036ed7257bSPatrick Caulfield 30436b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 30536b71a8bSDavid Teigland na = find_node_addr(nodeid); 30698e1b60eSMike Christie if (na && na->addr_count) { 307ee44b4bcSMarcelo Ricardo Leitner memcpy(&sas, na->addr[na->curr_addr_index], 308ee44b4bcSMarcelo Ricardo Leitner sizeof(struct sockaddr_storage)); 309ee44b4bcSMarcelo Ricardo Leitner 31098e1b60eSMike Christie if (try_new_addr) { 31198e1b60eSMike Christie na->curr_addr_index++; 31298e1b60eSMike Christie if (na->curr_addr_index == na->addr_count) 31398e1b60eSMike Christie na->curr_addr_index = 0; 31498e1b60eSMike Christie } 31598e1b60eSMike Christie } 31636b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 31736b71a8bSDavid Teigland 31836b71a8bSDavid Teigland if (!na) 31936b71a8bSDavid Teigland return -EEXIST; 32036b71a8bSDavid Teigland 32136b71a8bSDavid Teigland if (!na->addr_count) 32236b71a8bSDavid Teigland return -ENOENT; 32336b71a8bSDavid Teigland 32436b71a8bSDavid Teigland if (sas_out) 32536b71a8bSDavid Teigland memcpy(sas_out, &sas, sizeof(struct sockaddr_storage)); 32636b71a8bSDavid Teigland 32736b71a8bSDavid Teigland if (!sa_out) 32836b71a8bSDavid Teigland return 0; 3296ed7257bSPatrick Caulfield 3306ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) { 33136b71a8bSDavid Teigland struct sockaddr_in *in4 = (struct sockaddr_in *) &sas; 33236b71a8bSDavid Teigland struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out; 3336ed7257bSPatrick Caulfield ret4->sin_addr.s_addr = in4->sin_addr.s_addr; 3346ed7257bSPatrick Caulfield } else { 33536b71a8bSDavid Teigland struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &sas; 33636b71a8bSDavid Teigland struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out; 3374e3fd7a0SAlexey Dobriyan ret6->sin6_addr = in6->sin6_addr; 3386ed7257bSPatrick Caulfield } 3396ed7257bSPatrick Caulfield 3406ed7257bSPatrick Caulfield return 0; 3416ed7257bSPatrick Caulfield } 3426ed7257bSPatrick Caulfield 34336b71a8bSDavid Teigland static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid) 34436b71a8bSDavid Teigland { 34536b71a8bSDavid Teigland struct dlm_node_addr *na; 34636b71a8bSDavid Teigland int rv = -EEXIST; 34798e1b60eSMike Christie int addr_i; 34836b71a8bSDavid Teigland 34936b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 35036b71a8bSDavid Teigland list_for_each_entry(na, &dlm_node_addrs, list) { 35136b71a8bSDavid Teigland if (!na->addr_count) 35236b71a8bSDavid Teigland continue; 35336b71a8bSDavid Teigland 35498e1b60eSMike Christie for (addr_i = 0; addr_i < na->addr_count; addr_i++) { 35598e1b60eSMike Christie if (addr_compare(na->addr[addr_i], addr)) { 35636b71a8bSDavid Teigland *nodeid = na->nodeid; 35736b71a8bSDavid Teigland rv = 0; 35898e1b60eSMike Christie goto unlock; 35936b71a8bSDavid Teigland } 36098e1b60eSMike Christie } 36198e1b60eSMike Christie } 36298e1b60eSMike Christie unlock: 36336b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 36436b71a8bSDavid Teigland return rv; 36536b71a8bSDavid Teigland } 36636b71a8bSDavid Teigland 36736b71a8bSDavid Teigland int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len) 36836b71a8bSDavid Teigland { 36936b71a8bSDavid Teigland struct sockaddr_storage *new_addr; 37036b71a8bSDavid Teigland struct dlm_node_addr *new_node, *na; 37136b71a8bSDavid Teigland 37236b71a8bSDavid Teigland new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS); 37336b71a8bSDavid Teigland if (!new_node) 37436b71a8bSDavid Teigland return -ENOMEM; 37536b71a8bSDavid Teigland 37636b71a8bSDavid Teigland new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS); 37736b71a8bSDavid Teigland if (!new_addr) { 37836b71a8bSDavid Teigland kfree(new_node); 37936b71a8bSDavid Teigland return -ENOMEM; 38036b71a8bSDavid Teigland } 38136b71a8bSDavid Teigland 38236b71a8bSDavid Teigland memcpy(new_addr, addr, len); 38336b71a8bSDavid Teigland 38436b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 38536b71a8bSDavid Teigland na = find_node_addr(nodeid); 38636b71a8bSDavid Teigland if (!na) { 38736b71a8bSDavid Teigland new_node->nodeid = nodeid; 38836b71a8bSDavid Teigland new_node->addr[0] = new_addr; 38936b71a8bSDavid Teigland new_node->addr_count = 1; 39036b71a8bSDavid Teigland list_add(&new_node->list, &dlm_node_addrs); 39136b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 39236b71a8bSDavid Teigland return 0; 39336b71a8bSDavid Teigland } 39436b71a8bSDavid Teigland 39536b71a8bSDavid Teigland if (na->addr_count >= DLM_MAX_ADDR_COUNT) { 39636b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 39736b71a8bSDavid Teigland kfree(new_addr); 39836b71a8bSDavid Teigland kfree(new_node); 39936b71a8bSDavid Teigland return -ENOSPC; 40036b71a8bSDavid Teigland } 40136b71a8bSDavid Teigland 40236b71a8bSDavid Teigland na->addr[na->addr_count++] = new_addr; 40336b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 40436b71a8bSDavid Teigland kfree(new_node); 40536b71a8bSDavid Teigland return 0; 40636b71a8bSDavid Teigland } 40736b71a8bSDavid Teigland 4086ed7257bSPatrick Caulfield /* Data available on socket or listen socket received a connect */ 409676d2369SDavid S. Miller static void lowcomms_data_ready(struct sock *sk) 4106ed7257bSPatrick Caulfield { 4116ed7257bSPatrick Caulfield struct connection *con = sock2con(sk); 412afb853fbSPatrick Caulfield if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) 4136ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 4146ed7257bSPatrick Caulfield } 4156ed7257bSPatrick Caulfield 4166ed7257bSPatrick Caulfield static void lowcomms_write_space(struct sock *sk) 4176ed7257bSPatrick Caulfield { 4186ed7257bSPatrick Caulfield struct connection *con = sock2con(sk); 4196ed7257bSPatrick Caulfield 420b36930ddSDavid Miller if (!con) 421b36930ddSDavid Miller return; 422b36930ddSDavid Miller 423b36930ddSDavid Miller clear_bit(SOCK_NOSPACE, &con->sock->flags); 424b36930ddSDavid Miller 425b36930ddSDavid Miller if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) { 426b36930ddSDavid Miller con->sock->sk->sk_write_pending--; 4279cd3e072SEric Dumazet clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags); 428b36930ddSDavid Miller } 429b36930ddSDavid Miller 430b36930ddSDavid Miller if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) 4316ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 4326ed7257bSPatrick Caulfield } 4336ed7257bSPatrick Caulfield 4346ed7257bSPatrick Caulfield static inline void lowcomms_connect_sock(struct connection *con) 4356ed7257bSPatrick Caulfield { 436063c4c99SLars Marowsky-Bree if (test_bit(CF_CLOSE, &con->flags)) 437063c4c99SLars Marowsky-Bree return; 4386ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags)) 4396ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 4406ed7257bSPatrick Caulfield } 4416ed7257bSPatrick Caulfield 4426ed7257bSPatrick Caulfield static void lowcomms_state_change(struct sock *sk) 4436ed7257bSPatrick Caulfield { 444ee44b4bcSMarcelo Ricardo Leitner /* SCTP layer is not calling sk_data_ready when the connection 445ee44b4bcSMarcelo Ricardo Leitner * is done, so we catch the signal through here. Also, it 446ee44b4bcSMarcelo Ricardo Leitner * doesn't switch socket state when entering shutdown, so we 447ee44b4bcSMarcelo Ricardo Leitner * skip the write in that case. 448ee44b4bcSMarcelo Ricardo Leitner */ 449ee44b4bcSMarcelo Ricardo Leitner if (sk->sk_shutdown) { 450ee44b4bcSMarcelo Ricardo Leitner if (sk->sk_shutdown == RCV_SHUTDOWN) 451ee44b4bcSMarcelo Ricardo Leitner lowcomms_data_ready(sk); 452ee44b4bcSMarcelo Ricardo Leitner } else if (sk->sk_state == TCP_ESTABLISHED) { 4536ed7257bSPatrick Caulfield lowcomms_write_space(sk); 4546ed7257bSPatrick Caulfield } 455ee44b4bcSMarcelo Ricardo Leitner } 4566ed7257bSPatrick Caulfield 457391fbdc5SChristine Caulfield int dlm_lowcomms_connect_node(int nodeid) 458391fbdc5SChristine Caulfield { 459391fbdc5SChristine Caulfield struct connection *con; 460391fbdc5SChristine Caulfield 461391fbdc5SChristine Caulfield if (nodeid == dlm_our_nodeid()) 462391fbdc5SChristine Caulfield return 0; 463391fbdc5SChristine Caulfield 464391fbdc5SChristine Caulfield con = nodeid2con(nodeid, GFP_NOFS); 465391fbdc5SChristine Caulfield if (!con) 466391fbdc5SChristine Caulfield return -ENOMEM; 467391fbdc5SChristine Caulfield lowcomms_connect_sock(con); 468391fbdc5SChristine Caulfield return 0; 469391fbdc5SChristine Caulfield } 470391fbdc5SChristine Caulfield 471b3a5bbfdSBob Peterson static void lowcomms_error_report(struct sock *sk) 472b3a5bbfdSBob Peterson { 473b81171cbSBob Peterson struct connection *con; 474b3a5bbfdSBob Peterson struct sockaddr_storage saddr; 4751a31833dSBob Peterson int buflen; 476b81171cbSBob Peterson void (*orig_report)(struct sock *) = NULL; 477b3a5bbfdSBob Peterson 478b81171cbSBob Peterson read_lock_bh(&sk->sk_callback_lock); 479b81171cbSBob Peterson con = sock2con(sk); 480b81171cbSBob Peterson if (con == NULL) 481b81171cbSBob Peterson goto out; 482b81171cbSBob Peterson 483b81171cbSBob Peterson orig_report = con->orig_error_report; 4841a31833dSBob Peterson if (con->sock == NULL || 4851a31833dSBob Peterson kernel_getpeername(con->sock, (struct sockaddr *)&saddr, &buflen)) { 486b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 487b3a5bbfdSBob Peterson "sending to node %d, port %d, " 488b3a5bbfdSBob Peterson "sk_err=%d/%d\n", dlm_our_nodeid(), 489b3a5bbfdSBob Peterson con->nodeid, dlm_config.ci_tcp_port, 490b3a5bbfdSBob Peterson sk->sk_err, sk->sk_err_soft); 491b3a5bbfdSBob Peterson } else if (saddr.ss_family == AF_INET) { 492b3a5bbfdSBob Peterson struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr; 493b3a5bbfdSBob Peterson 494b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 495b3a5bbfdSBob Peterson "sending to node %d at %pI4, port %d, " 496b3a5bbfdSBob Peterson "sk_err=%d/%d\n", dlm_our_nodeid(), 497b3a5bbfdSBob Peterson con->nodeid, &sin4->sin_addr.s_addr, 498b3a5bbfdSBob Peterson dlm_config.ci_tcp_port, sk->sk_err, 499b3a5bbfdSBob Peterson sk->sk_err_soft); 500b3a5bbfdSBob Peterson } else { 501b3a5bbfdSBob Peterson struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr; 502b3a5bbfdSBob Peterson 503b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 504b3a5bbfdSBob Peterson "sending to node %d at %u.%u.%u.%u, " 505b3a5bbfdSBob Peterson "port %d, sk_err=%d/%d\n", dlm_our_nodeid(), 506b3a5bbfdSBob Peterson con->nodeid, sin6->sin6_addr.s6_addr32[0], 507b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[1], 508b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[2], 509b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[3], 510b3a5bbfdSBob Peterson dlm_config.ci_tcp_port, sk->sk_err, 511b3a5bbfdSBob Peterson sk->sk_err_soft); 512b3a5bbfdSBob Peterson } 513b81171cbSBob Peterson out: 514b81171cbSBob Peterson read_unlock_bh(&sk->sk_callback_lock); 515b81171cbSBob Peterson if (orig_report) 516b81171cbSBob Peterson orig_report(sk); 517b81171cbSBob Peterson } 518b81171cbSBob Peterson 519b81171cbSBob Peterson /* Note: sk_callback_lock must be locked before calling this function. */ 520b81171cbSBob Peterson static void save_callbacks(struct connection *con, struct sock *sk) 521b81171cbSBob Peterson { 522b81171cbSBob Peterson con->orig_data_ready = sk->sk_data_ready; 523b81171cbSBob Peterson con->orig_state_change = sk->sk_state_change; 524b81171cbSBob Peterson con->orig_write_space = sk->sk_write_space; 525b81171cbSBob Peterson con->orig_error_report = sk->sk_error_report; 526b81171cbSBob Peterson } 527b81171cbSBob Peterson 528b81171cbSBob Peterson static void restore_callbacks(struct connection *con, struct sock *sk) 529b81171cbSBob Peterson { 530b81171cbSBob Peterson write_lock_bh(&sk->sk_callback_lock); 531b81171cbSBob Peterson sk->sk_user_data = NULL; 532b81171cbSBob Peterson sk->sk_data_ready = con->orig_data_ready; 533b81171cbSBob Peterson sk->sk_state_change = con->orig_state_change; 534b81171cbSBob Peterson sk->sk_write_space = con->orig_write_space; 535b81171cbSBob Peterson sk->sk_error_report = con->orig_error_report; 536b81171cbSBob Peterson write_unlock_bh(&sk->sk_callback_lock); 537b3a5bbfdSBob Peterson } 538b3a5bbfdSBob Peterson 5396ed7257bSPatrick Caulfield /* Make a socket active */ 5403735b4b9SBob Peterson static void add_sock(struct socket *sock, struct connection *con, bool save_cb) 5416ed7257bSPatrick Caulfield { 542b81171cbSBob Peterson struct sock *sk = sock->sk; 543b81171cbSBob Peterson 544b81171cbSBob Peterson write_lock_bh(&sk->sk_callback_lock); 5456ed7257bSPatrick Caulfield con->sock = sock; 5466ed7257bSPatrick Caulfield 547b81171cbSBob Peterson sk->sk_user_data = con; 5483735b4b9SBob Peterson if (save_cb) 549b81171cbSBob Peterson save_callbacks(con, sk); 5506ed7257bSPatrick Caulfield /* Install a data_ready callback */ 551b81171cbSBob Peterson sk->sk_data_ready = lowcomms_data_ready; 552b81171cbSBob Peterson sk->sk_write_space = lowcomms_write_space; 553b81171cbSBob Peterson sk->sk_state_change = lowcomms_state_change; 554b81171cbSBob Peterson sk->sk_allocation = GFP_NOFS; 555b81171cbSBob Peterson sk->sk_error_report = lowcomms_error_report; 556b81171cbSBob Peterson write_unlock_bh(&sk->sk_callback_lock); 5576ed7257bSPatrick Caulfield } 5586ed7257bSPatrick Caulfield 5596ed7257bSPatrick Caulfield /* Add the port number to an IPv6 or 4 sockaddr and return the address 5606ed7257bSPatrick Caulfield length */ 5616ed7257bSPatrick Caulfield static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, 5626ed7257bSPatrick Caulfield int *addr_len) 5636ed7257bSPatrick Caulfield { 5646ed7257bSPatrick Caulfield saddr->ss_family = dlm_local_addr[0]->ss_family; 5656ed7257bSPatrick Caulfield if (saddr->ss_family == AF_INET) { 5666ed7257bSPatrick Caulfield struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; 5676ed7257bSPatrick Caulfield in4_addr->sin_port = cpu_to_be16(port); 5686ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in); 5696ed7257bSPatrick Caulfield memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); 5706ed7257bSPatrick Caulfield } else { 5716ed7257bSPatrick Caulfield struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; 5726ed7257bSPatrick Caulfield in6_addr->sin6_port = cpu_to_be16(port); 5736ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in6); 5746ed7257bSPatrick Caulfield } 57501c8cab2SPatrick Caulfield memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); 5766ed7257bSPatrick Caulfield } 5776ed7257bSPatrick Caulfield 5786ed7257bSPatrick Caulfield /* Close a remote connection and tidy up */ 5790d737a8cSMarcelo Ricardo Leitner static void close_connection(struct connection *con, bool and_other, 5800d737a8cSMarcelo Ricardo Leitner bool tx, bool rx) 5816ed7257bSPatrick Caulfield { 5820d737a8cSMarcelo Ricardo Leitner clear_bit(CF_CONNECT_PENDING, &con->flags); 5830d737a8cSMarcelo Ricardo Leitner clear_bit(CF_WRITE_PENDING, &con->flags); 5840d737a8cSMarcelo Ricardo Leitner if (tx && cancel_work_sync(&con->swork)) 5850d737a8cSMarcelo Ricardo Leitner log_print("canceled swork for node %d", con->nodeid); 5860d737a8cSMarcelo Ricardo Leitner if (rx && cancel_work_sync(&con->rwork)) 5870d737a8cSMarcelo Ricardo Leitner log_print("canceled rwork for node %d", con->nodeid); 5886ed7257bSPatrick Caulfield 5890d737a8cSMarcelo Ricardo Leitner mutex_lock(&con->sock_mutex); 5906ed7257bSPatrick Caulfield if (con->sock) { 591b81171cbSBob Peterson if (!test_bit(CF_IS_OTHERCON, &con->flags)) 592b81171cbSBob Peterson restore_callbacks(con, con->sock->sk); 5936ed7257bSPatrick Caulfield sock_release(con->sock); 5946ed7257bSPatrick Caulfield con->sock = NULL; 5956ed7257bSPatrick Caulfield } 5966ed7257bSPatrick Caulfield if (con->othercon && and_other) { 5976ed7257bSPatrick Caulfield /* Will only re-enter once. */ 5980d737a8cSMarcelo Ricardo Leitner close_connection(con->othercon, false, true, true); 5996ed7257bSPatrick Caulfield } 6006ed7257bSPatrick Caulfield if (con->rx_page) { 6016ed7257bSPatrick Caulfield __free_page(con->rx_page); 6026ed7257bSPatrick Caulfield con->rx_page = NULL; 6036ed7257bSPatrick Caulfield } 6049e5f2825SPatrick Caulfield 6056ed7257bSPatrick Caulfield con->retries = 0; 6066ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 6076ed7257bSPatrick Caulfield } 6086ed7257bSPatrick Caulfield 6096ed7257bSPatrick Caulfield /* Data received from remote end */ 6106ed7257bSPatrick Caulfield static int receive_from_sock(struct connection *con) 6116ed7257bSPatrick Caulfield { 6126ed7257bSPatrick Caulfield int ret = 0; 6136ed7257bSPatrick Caulfield struct msghdr msg = {}; 6146ed7257bSPatrick Caulfield struct kvec iov[2]; 6156ed7257bSPatrick Caulfield unsigned len; 6166ed7257bSPatrick Caulfield int r; 6176ed7257bSPatrick Caulfield int call_again_soon = 0; 6186ed7257bSPatrick Caulfield int nvec; 6196ed7257bSPatrick Caulfield 6206ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 6216ed7257bSPatrick Caulfield 6226ed7257bSPatrick Caulfield if (con->sock == NULL) { 6236ed7257bSPatrick Caulfield ret = -EAGAIN; 6246ed7257bSPatrick Caulfield goto out_close; 6256ed7257bSPatrick Caulfield } 626acee4e52SMarcelo Ricardo Leitner if (con->nodeid == 0) { 627acee4e52SMarcelo Ricardo Leitner ret = -EINVAL; 628acee4e52SMarcelo Ricardo Leitner goto out_close; 629acee4e52SMarcelo Ricardo Leitner } 6306ed7257bSPatrick Caulfield 6316ed7257bSPatrick Caulfield if (con->rx_page == NULL) { 6326ed7257bSPatrick Caulfield /* 6336ed7257bSPatrick Caulfield * This doesn't need to be atomic, but I think it should 6346ed7257bSPatrick Caulfield * improve performance if it is. 6356ed7257bSPatrick Caulfield */ 6366ed7257bSPatrick Caulfield con->rx_page = alloc_page(GFP_ATOMIC); 6376ed7257bSPatrick Caulfield if (con->rx_page == NULL) 6386ed7257bSPatrick Caulfield goto out_resched; 63909cbfeafSKirill A. Shutemov cbuf_init(&con->cb, PAGE_SIZE); 6406ed7257bSPatrick Caulfield } 6416ed7257bSPatrick Caulfield 6426ed7257bSPatrick Caulfield /* 6436ed7257bSPatrick Caulfield * iov[0] is the bit of the circular buffer between the current end 6446ed7257bSPatrick Caulfield * point (cb.base + cb.len) and the end of the buffer. 6456ed7257bSPatrick Caulfield */ 6466ed7257bSPatrick Caulfield iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); 6476ed7257bSPatrick Caulfield iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); 6486ed7257bSPatrick Caulfield iov[1].iov_len = 0; 6496ed7257bSPatrick Caulfield nvec = 1; 6506ed7257bSPatrick Caulfield 6516ed7257bSPatrick Caulfield /* 6526ed7257bSPatrick Caulfield * iov[1] is the bit of the circular buffer between the start of the 6536ed7257bSPatrick Caulfield * buffer and the start of the currently used section (cb.base) 6546ed7257bSPatrick Caulfield */ 6556ed7257bSPatrick Caulfield if (cbuf_data(&con->cb) >= con->cb.base) { 65609cbfeafSKirill A. Shutemov iov[0].iov_len = PAGE_SIZE - cbuf_data(&con->cb); 6576ed7257bSPatrick Caulfield iov[1].iov_len = con->cb.base; 6586ed7257bSPatrick Caulfield iov[1].iov_base = page_address(con->rx_page); 6596ed7257bSPatrick Caulfield nvec = 2; 6606ed7257bSPatrick Caulfield } 6616ed7257bSPatrick Caulfield len = iov[0].iov_len + iov[1].iov_len; 6626ed7257bSPatrick Caulfield 6636ed7257bSPatrick Caulfield r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len, 6646ed7257bSPatrick Caulfield MSG_DONTWAIT | MSG_NOSIGNAL); 6656ed7257bSPatrick Caulfield if (ret <= 0) 6666ed7257bSPatrick Caulfield goto out_close; 667ee44b4bcSMarcelo Ricardo Leitner else if (ret == len) 668ee44b4bcSMarcelo Ricardo Leitner call_again_soon = 1; 6696ed7257bSPatrick Caulfield 6706ed7257bSPatrick Caulfield cbuf_add(&con->cb, ret); 6716ed7257bSPatrick Caulfield ret = dlm_process_incoming_buffer(con->nodeid, 6726ed7257bSPatrick Caulfield page_address(con->rx_page), 6736ed7257bSPatrick Caulfield con->cb.base, con->cb.len, 67409cbfeafSKirill A. Shutemov PAGE_SIZE); 6756ed7257bSPatrick Caulfield if (ret == -EBADMSG) { 676ee44b4bcSMarcelo Ricardo Leitner log_print("lowcomms: addr=%p, base=%u, len=%u, read=%d", 677ee44b4bcSMarcelo Ricardo Leitner page_address(con->rx_page), con->cb.base, 678ee44b4bcSMarcelo Ricardo Leitner con->cb.len, r); 6796ed7257bSPatrick Caulfield } 6806ed7257bSPatrick Caulfield if (ret < 0) 6816ed7257bSPatrick Caulfield goto out_close; 6826ed7257bSPatrick Caulfield cbuf_eat(&con->cb, ret); 6836ed7257bSPatrick Caulfield 6846ed7257bSPatrick Caulfield if (cbuf_empty(&con->cb) && !call_again_soon) { 6856ed7257bSPatrick Caulfield __free_page(con->rx_page); 6866ed7257bSPatrick Caulfield con->rx_page = NULL; 6876ed7257bSPatrick Caulfield } 6886ed7257bSPatrick Caulfield 6896ed7257bSPatrick Caulfield if (call_again_soon) 6906ed7257bSPatrick Caulfield goto out_resched; 6916ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 6926ed7257bSPatrick Caulfield return 0; 6936ed7257bSPatrick Caulfield 6946ed7257bSPatrick Caulfield out_resched: 6956ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) 6966ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 6976ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 6986ed7257bSPatrick Caulfield return -EAGAIN; 6996ed7257bSPatrick Caulfield 7006ed7257bSPatrick Caulfield out_close: 7016ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7029e5f2825SPatrick Caulfield if (ret != -EAGAIN) { 7030d737a8cSMarcelo Ricardo Leitner close_connection(con, false, true, false); 7046ed7257bSPatrick Caulfield /* Reconnect when there is something to send */ 7056ed7257bSPatrick Caulfield } 7066ed7257bSPatrick Caulfield /* Don't return success if we really got EOF */ 7076ed7257bSPatrick Caulfield if (ret == 0) 7086ed7257bSPatrick Caulfield ret = -EAGAIN; 7096ed7257bSPatrick Caulfield 7106ed7257bSPatrick Caulfield return ret; 7116ed7257bSPatrick Caulfield } 7126ed7257bSPatrick Caulfield 7136ed7257bSPatrick Caulfield /* Listening socket is busy, accept a connection */ 7146ed7257bSPatrick Caulfield static int tcp_accept_from_sock(struct connection *con) 7156ed7257bSPatrick Caulfield { 7166ed7257bSPatrick Caulfield int result; 7176ed7257bSPatrick Caulfield struct sockaddr_storage peeraddr; 7186ed7257bSPatrick Caulfield struct socket *newsock; 7196ed7257bSPatrick Caulfield int len; 7206ed7257bSPatrick Caulfield int nodeid; 7216ed7257bSPatrick Caulfield struct connection *newcon; 7226ed7257bSPatrick Caulfield struct connection *addcon; 7236ed7257bSPatrick Caulfield 724513ef596SDavid Teigland mutex_lock(&connections_lock); 725513ef596SDavid Teigland if (!dlm_allow_conn) { 726513ef596SDavid Teigland mutex_unlock(&connections_lock); 727513ef596SDavid Teigland return -1; 728513ef596SDavid Teigland } 729513ef596SDavid Teigland mutex_unlock(&connections_lock); 730513ef596SDavid Teigland 7316ed7257bSPatrick Caulfield memset(&peeraddr, 0, sizeof(peeraddr)); 732eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 733eeb1bd5cSEric W. Biederman SOCK_STREAM, IPPROTO_TCP, &newsock); 7346ed7257bSPatrick Caulfield if (result < 0) 7356ed7257bSPatrick Caulfield return -ENOMEM; 7366ed7257bSPatrick Caulfield 7376ed7257bSPatrick Caulfield mutex_lock_nested(&con->sock_mutex, 0); 7386ed7257bSPatrick Caulfield 7396ed7257bSPatrick Caulfield result = -ENOTCONN; 7406ed7257bSPatrick Caulfield if (con->sock == NULL) 7416ed7257bSPatrick Caulfield goto accept_err; 7426ed7257bSPatrick Caulfield 7436ed7257bSPatrick Caulfield newsock->type = con->sock->type; 7446ed7257bSPatrick Caulfield newsock->ops = con->sock->ops; 7456ed7257bSPatrick Caulfield 746cdfbabfbSDavid Howells result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK, true); 7476ed7257bSPatrick Caulfield if (result < 0) 7486ed7257bSPatrick Caulfield goto accept_err; 7496ed7257bSPatrick Caulfield 7506ed7257bSPatrick Caulfield /* Get the connected socket's peer */ 7516ed7257bSPatrick Caulfield memset(&peeraddr, 0, sizeof(peeraddr)); 7526ed7257bSPatrick Caulfield if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 7536ed7257bSPatrick Caulfield &len, 2)) { 7546ed7257bSPatrick Caulfield result = -ECONNABORTED; 7556ed7257bSPatrick Caulfield goto accept_err; 7566ed7257bSPatrick Caulfield } 7576ed7257bSPatrick Caulfield 7586ed7257bSPatrick Caulfield /* Get the new node's NODEID */ 7596ed7257bSPatrick Caulfield make_sockaddr(&peeraddr, 0, &len); 76036b71a8bSDavid Teigland if (addr_to_nodeid(&peeraddr, &nodeid)) { 761bcaadf5cSMasatake YAMATO unsigned char *b=(unsigned char *)&peeraddr; 762617e82e1SDavid Teigland log_print("connect from non cluster node"); 763bcaadf5cSMasatake YAMATO print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, 764bcaadf5cSMasatake YAMATO b, sizeof(struct sockaddr_storage)); 7656ed7257bSPatrick Caulfield sock_release(newsock); 7666ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7676ed7257bSPatrick Caulfield return -1; 7686ed7257bSPatrick Caulfield } 7696ed7257bSPatrick Caulfield 7706ed7257bSPatrick Caulfield log_print("got connection from %d", nodeid); 7716ed7257bSPatrick Caulfield 7726ed7257bSPatrick Caulfield /* Check to see if we already have a connection to this node. This 7736ed7257bSPatrick Caulfield * could happen if the two nodes initiate a connection at roughly 7746ed7257bSPatrick Caulfield * the same time and the connections cross on the wire. 7756ed7257bSPatrick Caulfield * In this case we store the incoming one in "othercon" 7766ed7257bSPatrick Caulfield */ 777748285ccSDavid Teigland newcon = nodeid2con(nodeid, GFP_NOFS); 7786ed7257bSPatrick Caulfield if (!newcon) { 7796ed7257bSPatrick Caulfield result = -ENOMEM; 7806ed7257bSPatrick Caulfield goto accept_err; 7816ed7257bSPatrick Caulfield } 7826ed7257bSPatrick Caulfield mutex_lock_nested(&newcon->sock_mutex, 1); 7836ed7257bSPatrick Caulfield if (newcon->sock) { 7846ed7257bSPatrick Caulfield struct connection *othercon = newcon->othercon; 7856ed7257bSPatrick Caulfield 7866ed7257bSPatrick Caulfield if (!othercon) { 787748285ccSDavid Teigland othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); 7886ed7257bSPatrick Caulfield if (!othercon) { 789617e82e1SDavid Teigland log_print("failed to allocate incoming socket"); 7906ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 7916ed7257bSPatrick Caulfield result = -ENOMEM; 7926ed7257bSPatrick Caulfield goto accept_err; 7936ed7257bSPatrick Caulfield } 7946ed7257bSPatrick Caulfield othercon->nodeid = nodeid; 7956ed7257bSPatrick Caulfield othercon->rx_action = receive_from_sock; 7966ed7257bSPatrick Caulfield mutex_init(&othercon->sock_mutex); 7976ed7257bSPatrick Caulfield INIT_WORK(&othercon->swork, process_send_sockets); 7986ed7257bSPatrick Caulfield INIT_WORK(&othercon->rwork, process_recv_sockets); 7996ed7257bSPatrick Caulfield set_bit(CF_IS_OTHERCON, &othercon->flags); 80061d96be0SPatrick Caulfield } 80161d96be0SPatrick Caulfield if (!othercon->sock) { 8026ed7257bSPatrick Caulfield newcon->othercon = othercon; 8036ed7257bSPatrick Caulfield othercon->sock = newsock; 8046ed7257bSPatrick Caulfield newsock->sk->sk_user_data = othercon; 8053735b4b9SBob Peterson add_sock(newsock, othercon, false); 8066ed7257bSPatrick Caulfield addcon = othercon; 8076ed7257bSPatrick Caulfield } 8086ed7257bSPatrick Caulfield else { 80997d84836SPatrick Caulfield printk("Extra connection from node %d attempted\n", nodeid); 81097d84836SPatrick Caulfield result = -EAGAIN; 811f4fadb23Sakpm@linux-foundation.org mutex_unlock(&newcon->sock_mutex); 81297d84836SPatrick Caulfield goto accept_err; 81397d84836SPatrick Caulfield } 81497d84836SPatrick Caulfield } 81597d84836SPatrick Caulfield else { 8166ed7257bSPatrick Caulfield newsock->sk->sk_user_data = newcon; 8176ed7257bSPatrick Caulfield newcon->rx_action = receive_from_sock; 8183735b4b9SBob Peterson /* accept copies the sk after we've saved the callbacks, so we 8193735b4b9SBob Peterson don't want to save them a second time or comm errors will 8203735b4b9SBob Peterson result in calling sk_error_report recursively. */ 8213735b4b9SBob Peterson add_sock(newsock, newcon, false); 8226ed7257bSPatrick Caulfield addcon = newcon; 8236ed7257bSPatrick Caulfield } 8246ed7257bSPatrick Caulfield 8256ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 8266ed7257bSPatrick Caulfield 8276ed7257bSPatrick Caulfield /* 8286ed7257bSPatrick Caulfield * Add it to the active queue in case we got data 82925985edcSLucas De Marchi * between processing the accept adding the socket 8306ed7257bSPatrick Caulfield * to the read_sockets list 8316ed7257bSPatrick Caulfield */ 8326ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) 8336ed7257bSPatrick Caulfield queue_work(recv_workqueue, &addcon->rwork); 8346ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 8356ed7257bSPatrick Caulfield 8366ed7257bSPatrick Caulfield return 0; 8376ed7257bSPatrick Caulfield 8386ed7257bSPatrick Caulfield accept_err: 8396ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 8406ed7257bSPatrick Caulfield sock_release(newsock); 8416ed7257bSPatrick Caulfield 8426ed7257bSPatrick Caulfield if (result != -EAGAIN) 843617e82e1SDavid Teigland log_print("error accepting connection from node: %d", result); 8446ed7257bSPatrick Caulfield return result; 8456ed7257bSPatrick Caulfield } 8466ed7257bSPatrick Caulfield 84718df8a87Skbuild test robot static int sctp_accept_from_sock(struct connection *con) 848ee44b4bcSMarcelo Ricardo Leitner { 849ee44b4bcSMarcelo Ricardo Leitner /* Check that the new node is in the lockspace */ 850ee44b4bcSMarcelo Ricardo Leitner struct sctp_prim prim; 851ee44b4bcSMarcelo Ricardo Leitner int nodeid; 852ee44b4bcSMarcelo Ricardo Leitner int prim_len, ret; 853ee44b4bcSMarcelo Ricardo Leitner int addr_len; 854ee44b4bcSMarcelo Ricardo Leitner struct connection *newcon; 855ee44b4bcSMarcelo Ricardo Leitner struct connection *addcon; 856ee44b4bcSMarcelo Ricardo Leitner struct socket *newsock; 857ee44b4bcSMarcelo Ricardo Leitner 858ee44b4bcSMarcelo Ricardo Leitner mutex_lock(&connections_lock); 859ee44b4bcSMarcelo Ricardo Leitner if (!dlm_allow_conn) { 860ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&connections_lock); 861ee44b4bcSMarcelo Ricardo Leitner return -1; 862ee44b4bcSMarcelo Ricardo Leitner } 863ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&connections_lock); 864ee44b4bcSMarcelo Ricardo Leitner 865ee44b4bcSMarcelo Ricardo Leitner mutex_lock_nested(&con->sock_mutex, 0); 866ee44b4bcSMarcelo Ricardo Leitner 867ee44b4bcSMarcelo Ricardo Leitner ret = kernel_accept(con->sock, &newsock, O_NONBLOCK); 868ee44b4bcSMarcelo Ricardo Leitner if (ret < 0) 869ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 870ee44b4bcSMarcelo Ricardo Leitner 871ee44b4bcSMarcelo Ricardo Leitner memset(&prim, 0, sizeof(struct sctp_prim)); 872ee44b4bcSMarcelo Ricardo Leitner prim_len = sizeof(struct sctp_prim); 873ee44b4bcSMarcelo Ricardo Leitner 874ee44b4bcSMarcelo Ricardo Leitner ret = kernel_getsockopt(newsock, IPPROTO_SCTP, SCTP_PRIMARY_ADDR, 875ee44b4bcSMarcelo Ricardo Leitner (char *)&prim, &prim_len); 876ee44b4bcSMarcelo Ricardo Leitner if (ret < 0) { 877ee44b4bcSMarcelo Ricardo Leitner log_print("getsockopt/sctp_primary_addr failed: %d", ret); 878ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 879ee44b4bcSMarcelo Ricardo Leitner } 880ee44b4bcSMarcelo Ricardo Leitner 881ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&prim.ssp_addr, 0, &addr_len); 88226c1ec2fSWei Yongjun ret = addr_to_nodeid(&prim.ssp_addr, &nodeid); 88326c1ec2fSWei Yongjun if (ret) { 884ee44b4bcSMarcelo Ricardo Leitner unsigned char *b = (unsigned char *)&prim.ssp_addr; 885ee44b4bcSMarcelo Ricardo Leitner 886ee44b4bcSMarcelo Ricardo Leitner log_print("reject connect from unknown addr"); 887ee44b4bcSMarcelo Ricardo Leitner print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, 888ee44b4bcSMarcelo Ricardo Leitner b, sizeof(struct sockaddr_storage)); 889ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 890ee44b4bcSMarcelo Ricardo Leitner } 891ee44b4bcSMarcelo Ricardo Leitner 892ee44b4bcSMarcelo Ricardo Leitner newcon = nodeid2con(nodeid, GFP_NOFS); 893ee44b4bcSMarcelo Ricardo Leitner if (!newcon) { 894ee44b4bcSMarcelo Ricardo Leitner ret = -ENOMEM; 895ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 896ee44b4bcSMarcelo Ricardo Leitner } 897ee44b4bcSMarcelo Ricardo Leitner 898ee44b4bcSMarcelo Ricardo Leitner mutex_lock_nested(&newcon->sock_mutex, 1); 899ee44b4bcSMarcelo Ricardo Leitner 900ee44b4bcSMarcelo Ricardo Leitner if (newcon->sock) { 901ee44b4bcSMarcelo Ricardo Leitner struct connection *othercon = newcon->othercon; 902ee44b4bcSMarcelo Ricardo Leitner 903ee44b4bcSMarcelo Ricardo Leitner if (!othercon) { 904ee44b4bcSMarcelo Ricardo Leitner othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); 905ee44b4bcSMarcelo Ricardo Leitner if (!othercon) { 906ee44b4bcSMarcelo Ricardo Leitner log_print("failed to allocate incoming socket"); 907ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 908ee44b4bcSMarcelo Ricardo Leitner ret = -ENOMEM; 909ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 910ee44b4bcSMarcelo Ricardo Leitner } 911ee44b4bcSMarcelo Ricardo Leitner othercon->nodeid = nodeid; 912ee44b4bcSMarcelo Ricardo Leitner othercon->rx_action = receive_from_sock; 913ee44b4bcSMarcelo Ricardo Leitner mutex_init(&othercon->sock_mutex); 914ee44b4bcSMarcelo Ricardo Leitner INIT_WORK(&othercon->swork, process_send_sockets); 915ee44b4bcSMarcelo Ricardo Leitner INIT_WORK(&othercon->rwork, process_recv_sockets); 916ee44b4bcSMarcelo Ricardo Leitner set_bit(CF_IS_OTHERCON, &othercon->flags); 917ee44b4bcSMarcelo Ricardo Leitner } 918ee44b4bcSMarcelo Ricardo Leitner if (!othercon->sock) { 919ee44b4bcSMarcelo Ricardo Leitner newcon->othercon = othercon; 920ee44b4bcSMarcelo Ricardo Leitner othercon->sock = newsock; 921ee44b4bcSMarcelo Ricardo Leitner newsock->sk->sk_user_data = othercon; 9223735b4b9SBob Peterson add_sock(newsock, othercon, false); 923ee44b4bcSMarcelo Ricardo Leitner addcon = othercon; 924ee44b4bcSMarcelo Ricardo Leitner } else { 925ee44b4bcSMarcelo Ricardo Leitner printk("Extra connection from node %d attempted\n", nodeid); 926ee44b4bcSMarcelo Ricardo Leitner ret = -EAGAIN; 927ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 928ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 929ee44b4bcSMarcelo Ricardo Leitner } 930ee44b4bcSMarcelo Ricardo Leitner } else { 931ee44b4bcSMarcelo Ricardo Leitner newsock->sk->sk_user_data = newcon; 932ee44b4bcSMarcelo Ricardo Leitner newcon->rx_action = receive_from_sock; 9333735b4b9SBob Peterson add_sock(newsock, newcon, false); 934ee44b4bcSMarcelo Ricardo Leitner addcon = newcon; 935ee44b4bcSMarcelo Ricardo Leitner } 936ee44b4bcSMarcelo Ricardo Leitner 937ee44b4bcSMarcelo Ricardo Leitner log_print("connected to %d", nodeid); 938ee44b4bcSMarcelo Ricardo Leitner 939ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 940ee44b4bcSMarcelo Ricardo Leitner 941ee44b4bcSMarcelo Ricardo Leitner /* 942ee44b4bcSMarcelo Ricardo Leitner * Add it to the active queue in case we got data 943ee44b4bcSMarcelo Ricardo Leitner * between processing the accept adding the socket 944ee44b4bcSMarcelo Ricardo Leitner * to the read_sockets list 945ee44b4bcSMarcelo Ricardo Leitner */ 946ee44b4bcSMarcelo Ricardo Leitner if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) 947ee44b4bcSMarcelo Ricardo Leitner queue_work(recv_workqueue, &addcon->rwork); 948ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 949ee44b4bcSMarcelo Ricardo Leitner 950ee44b4bcSMarcelo Ricardo Leitner return 0; 951ee44b4bcSMarcelo Ricardo Leitner 952ee44b4bcSMarcelo Ricardo Leitner accept_err: 953ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 954ee44b4bcSMarcelo Ricardo Leitner if (newsock) 955ee44b4bcSMarcelo Ricardo Leitner sock_release(newsock); 956ee44b4bcSMarcelo Ricardo Leitner if (ret != -EAGAIN) 957ee44b4bcSMarcelo Ricardo Leitner log_print("error accepting connection from node: %d", ret); 958ee44b4bcSMarcelo Ricardo Leitner 959ee44b4bcSMarcelo Ricardo Leitner return ret; 960ee44b4bcSMarcelo Ricardo Leitner } 961ee44b4bcSMarcelo Ricardo Leitner 9626ed7257bSPatrick Caulfield static void free_entry(struct writequeue_entry *e) 9636ed7257bSPatrick Caulfield { 9646ed7257bSPatrick Caulfield __free_page(e->page); 9656ed7257bSPatrick Caulfield kfree(e); 9666ed7257bSPatrick Caulfield } 9676ed7257bSPatrick Caulfield 9685d689871SMike Christie /* 9695d689871SMike Christie * writequeue_entry_complete - try to delete and free write queue entry 9705d689871SMike Christie * @e: write queue entry to try to delete 9715d689871SMike Christie * @completed: bytes completed 9725d689871SMike Christie * 9735d689871SMike Christie * writequeue_lock must be held. 9745d689871SMike Christie */ 9755d689871SMike Christie static void writequeue_entry_complete(struct writequeue_entry *e, int completed) 9765d689871SMike Christie { 9775d689871SMike Christie e->offset += completed; 9785d689871SMike Christie e->len -= completed; 9795d689871SMike Christie 9805d689871SMike Christie if (e->len == 0 && e->users == 0) { 9815d689871SMike Christie list_del(&e->list); 9825d689871SMike Christie free_entry(e); 9835d689871SMike Christie } 9845d689871SMike Christie } 9855d689871SMike Christie 986ee44b4bcSMarcelo Ricardo Leitner /* 987ee44b4bcSMarcelo Ricardo Leitner * sctp_bind_addrs - bind a SCTP socket to all our addresses 988ee44b4bcSMarcelo Ricardo Leitner */ 989ee44b4bcSMarcelo Ricardo Leitner static int sctp_bind_addrs(struct connection *con, uint16_t port) 990ee44b4bcSMarcelo Ricardo Leitner { 991ee44b4bcSMarcelo Ricardo Leitner struct sockaddr_storage localaddr; 992ee44b4bcSMarcelo Ricardo Leitner int i, addr_len, result = 0; 993ee44b4bcSMarcelo Ricardo Leitner 994ee44b4bcSMarcelo Ricardo Leitner for (i = 0; i < dlm_local_count; i++) { 995ee44b4bcSMarcelo Ricardo Leitner memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); 996ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&localaddr, port, &addr_len); 997ee44b4bcSMarcelo Ricardo Leitner 998ee44b4bcSMarcelo Ricardo Leitner if (!i) 999ee44b4bcSMarcelo Ricardo Leitner result = kernel_bind(con->sock, 1000ee44b4bcSMarcelo Ricardo Leitner (struct sockaddr *)&localaddr, 1001ee44b4bcSMarcelo Ricardo Leitner addr_len); 1002ee44b4bcSMarcelo Ricardo Leitner else 1003ee44b4bcSMarcelo Ricardo Leitner result = kernel_setsockopt(con->sock, SOL_SCTP, 1004ee44b4bcSMarcelo Ricardo Leitner SCTP_SOCKOPT_BINDX_ADD, 1005ee44b4bcSMarcelo Ricardo Leitner (char *)&localaddr, addr_len); 1006ee44b4bcSMarcelo Ricardo Leitner 1007ee44b4bcSMarcelo Ricardo Leitner if (result < 0) { 1008ee44b4bcSMarcelo Ricardo Leitner log_print("Can't bind to %d addr number %d, %d.\n", 1009ee44b4bcSMarcelo Ricardo Leitner port, i + 1, result); 1010ee44b4bcSMarcelo Ricardo Leitner break; 1011ee44b4bcSMarcelo Ricardo Leitner } 1012ee44b4bcSMarcelo Ricardo Leitner } 1013ee44b4bcSMarcelo Ricardo Leitner return result; 1014ee44b4bcSMarcelo Ricardo Leitner } 1015ee44b4bcSMarcelo Ricardo Leitner 10166ed7257bSPatrick Caulfield /* Initiate an SCTP association. 10176ed7257bSPatrick Caulfield This is a special case of send_to_sock() in that we don't yet have a 10186ed7257bSPatrick Caulfield peeled-off socket for this association, so we use the listening socket 10196ed7257bSPatrick Caulfield and add the primary IP address of the remote node. 10206ed7257bSPatrick Caulfield */ 1021ee44b4bcSMarcelo Ricardo Leitner static void sctp_connect_to_sock(struct connection *con) 10226ed7257bSPatrick Caulfield { 1023ee44b4bcSMarcelo Ricardo Leitner struct sockaddr_storage daddr; 1024ee44b4bcSMarcelo Ricardo Leitner int one = 1; 1025ee44b4bcSMarcelo Ricardo Leitner int result; 1026ee44b4bcSMarcelo Ricardo Leitner int addr_len; 1027ee44b4bcSMarcelo Ricardo Leitner struct socket *sock; 1028ee44b4bcSMarcelo Ricardo Leitner 1029ee44b4bcSMarcelo Ricardo Leitner if (con->nodeid == 0) { 1030ee44b4bcSMarcelo Ricardo Leitner log_print("attempt to connect sock 0 foiled"); 1031ee44b4bcSMarcelo Ricardo Leitner return; 1032ee44b4bcSMarcelo Ricardo Leitner } 10336ed7257bSPatrick Caulfield 10345d689871SMike Christie mutex_lock(&con->sock_mutex); 10356ed7257bSPatrick Caulfield 1036ee44b4bcSMarcelo Ricardo Leitner /* Some odd races can cause double-connects, ignore them */ 1037ee44b4bcSMarcelo Ricardo Leitner if (con->retries++ > MAX_CONNECT_RETRIES) 1038ee44b4bcSMarcelo Ricardo Leitner goto out; 1039ee44b4bcSMarcelo Ricardo Leitner 1040ee44b4bcSMarcelo Ricardo Leitner if (con->sock) { 1041ee44b4bcSMarcelo Ricardo Leitner log_print("node %d already connected.", con->nodeid); 1042ee44b4bcSMarcelo Ricardo Leitner goto out; 1043ee44b4bcSMarcelo Ricardo Leitner } 1044ee44b4bcSMarcelo Ricardo Leitner 1045ee44b4bcSMarcelo Ricardo Leitner memset(&daddr, 0, sizeof(daddr)); 1046ee44b4bcSMarcelo Ricardo Leitner result = nodeid_to_addr(con->nodeid, &daddr, NULL, true); 1047ee44b4bcSMarcelo Ricardo Leitner if (result < 0) { 10486ed7257bSPatrick Caulfield log_print("no address for nodeid %d", con->nodeid); 1049ee44b4bcSMarcelo Ricardo Leitner goto out; 105004bedd79SDavid Teigland } 10516ed7257bSPatrick Caulfield 1052ee44b4bcSMarcelo Ricardo Leitner /* Create a socket to communicate with */ 1053ee44b4bcSMarcelo Ricardo Leitner result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1054ee44b4bcSMarcelo Ricardo Leitner SOCK_STREAM, IPPROTO_SCTP, &sock); 1055ee44b4bcSMarcelo Ricardo Leitner if (result < 0) 1056ee44b4bcSMarcelo Ricardo Leitner goto socket_err; 10576ed7257bSPatrick Caulfield 1058ee44b4bcSMarcelo Ricardo Leitner sock->sk->sk_user_data = con; 1059ee44b4bcSMarcelo Ricardo Leitner con->rx_action = receive_from_sock; 1060ee44b4bcSMarcelo Ricardo Leitner con->connect_action = sctp_connect_to_sock; 10613735b4b9SBob Peterson add_sock(sock, con, true); 10626ed7257bSPatrick Caulfield 1063ee44b4bcSMarcelo Ricardo Leitner /* Bind to all addresses. */ 1064ee44b4bcSMarcelo Ricardo Leitner if (sctp_bind_addrs(con, 0)) 1065ee44b4bcSMarcelo Ricardo Leitner goto bind_err; 106698e1b60eSMike Christie 1067ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len); 10686ed7257bSPatrick Caulfield 1069ee44b4bcSMarcelo Ricardo Leitner log_print("connecting to %d", con->nodeid); 10706ed7257bSPatrick Caulfield 1071ee44b4bcSMarcelo Ricardo Leitner /* Turn off Nagle's algorithm */ 1072ee44b4bcSMarcelo Ricardo Leitner kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, 1073ee44b4bcSMarcelo Ricardo Leitner sizeof(one)); 1074ee44b4bcSMarcelo Ricardo Leitner 1075ee44b4bcSMarcelo Ricardo Leitner result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len, 1076ee44b4bcSMarcelo Ricardo Leitner O_NONBLOCK); 1077ee44b4bcSMarcelo Ricardo Leitner if (result == -EINPROGRESS) 1078ee44b4bcSMarcelo Ricardo Leitner result = 0; 1079ee44b4bcSMarcelo Ricardo Leitner if (result == 0) 1080ee44b4bcSMarcelo Ricardo Leitner goto out; 1081ee44b4bcSMarcelo Ricardo Leitner 1082ee44b4bcSMarcelo Ricardo Leitner 1083ee44b4bcSMarcelo Ricardo Leitner bind_err: 1084ee44b4bcSMarcelo Ricardo Leitner con->sock = NULL; 1085ee44b4bcSMarcelo Ricardo Leitner sock_release(sock); 1086ee44b4bcSMarcelo Ricardo Leitner 1087ee44b4bcSMarcelo Ricardo Leitner socket_err: 1088ee44b4bcSMarcelo Ricardo Leitner /* 1089ee44b4bcSMarcelo Ricardo Leitner * Some errors are fatal and this list might need adjusting. For other 1090ee44b4bcSMarcelo Ricardo Leitner * errors we try again until the max number of retries is reached. 1091ee44b4bcSMarcelo Ricardo Leitner */ 1092ee44b4bcSMarcelo Ricardo Leitner if (result != -EHOSTUNREACH && 1093ee44b4bcSMarcelo Ricardo Leitner result != -ENETUNREACH && 1094ee44b4bcSMarcelo Ricardo Leitner result != -ENETDOWN && 1095ee44b4bcSMarcelo Ricardo Leitner result != -EINVAL && 1096ee44b4bcSMarcelo Ricardo Leitner result != -EPROTONOSUPPORT) { 1097ee44b4bcSMarcelo Ricardo Leitner log_print("connect %d try %d error %d", con->nodeid, 1098ee44b4bcSMarcelo Ricardo Leitner con->retries, result); 1099ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 1100ee44b4bcSMarcelo Ricardo Leitner msleep(1000); 11016ed7257bSPatrick Caulfield clear_bit(CF_CONNECT_PENDING, &con->flags); 1102ee44b4bcSMarcelo Ricardo Leitner lowcomms_connect_sock(con); 1103ee44b4bcSMarcelo Ricardo Leitner return; 11046ed7257bSPatrick Caulfield } 11055d689871SMike Christie 1106ee44b4bcSMarcelo Ricardo Leitner out: 11075d689871SMike Christie mutex_unlock(&con->sock_mutex); 110800dcffaeSMarcelo Ricardo Leitner set_bit(CF_WRITE_PENDING, &con->flags); 11096ed7257bSPatrick Caulfield } 11106ed7257bSPatrick Caulfield 11116ed7257bSPatrick Caulfield /* Connect a new socket to its peer */ 11126ed7257bSPatrick Caulfield static void tcp_connect_to_sock(struct connection *con) 11136ed7257bSPatrick Caulfield { 11146bd8fedaSLon Hohberger struct sockaddr_storage saddr, src_addr; 11156ed7257bSPatrick Caulfield int addr_len; 1116a89d63a1SCasey Dahlin struct socket *sock = NULL; 1117cb2d45daSDavid Teigland int one = 1; 111836b71a8bSDavid Teigland int result; 11196ed7257bSPatrick Caulfield 11206ed7257bSPatrick Caulfield if (con->nodeid == 0) { 11216ed7257bSPatrick Caulfield log_print("attempt to connect sock 0 foiled"); 11226ed7257bSPatrick Caulfield return; 11236ed7257bSPatrick Caulfield } 11246ed7257bSPatrick Caulfield 11256ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 11266ed7257bSPatrick Caulfield if (con->retries++ > MAX_CONNECT_RETRIES) 11276ed7257bSPatrick Caulfield goto out; 11286ed7257bSPatrick Caulfield 11296ed7257bSPatrick Caulfield /* Some odd races can cause double-connects, ignore them */ 113036b71a8bSDavid Teigland if (con->sock) 11316ed7257bSPatrick Caulfield goto out; 11326ed7257bSPatrick Caulfield 11336ed7257bSPatrick Caulfield /* Create a socket to communicate with */ 1134eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1135eeb1bd5cSEric W. Biederman SOCK_STREAM, IPPROTO_TCP, &sock); 11366ed7257bSPatrick Caulfield if (result < 0) 11376ed7257bSPatrick Caulfield goto out_err; 11386ed7257bSPatrick Caulfield 11396ed7257bSPatrick Caulfield memset(&saddr, 0, sizeof(saddr)); 114098e1b60eSMike Christie result = nodeid_to_addr(con->nodeid, &saddr, NULL, false); 114136b71a8bSDavid Teigland if (result < 0) { 114236b71a8bSDavid Teigland log_print("no address for nodeid %d", con->nodeid); 11436ed7257bSPatrick Caulfield goto out_err; 114436b71a8bSDavid Teigland } 11456ed7257bSPatrick Caulfield 11466ed7257bSPatrick Caulfield sock->sk->sk_user_data = con; 11476ed7257bSPatrick Caulfield con->rx_action = receive_from_sock; 11486ed7257bSPatrick Caulfield con->connect_action = tcp_connect_to_sock; 11493735b4b9SBob Peterson add_sock(sock, con, true); 11506ed7257bSPatrick Caulfield 11516bd8fedaSLon Hohberger /* Bind to our cluster-known address connecting to avoid 11526bd8fedaSLon Hohberger routing problems */ 11536bd8fedaSLon Hohberger memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); 11546bd8fedaSLon Hohberger make_sockaddr(&src_addr, 0, &addr_len); 11556bd8fedaSLon Hohberger result = sock->ops->bind(sock, (struct sockaddr *) &src_addr, 11566bd8fedaSLon Hohberger addr_len); 11576bd8fedaSLon Hohberger if (result < 0) { 11586bd8fedaSLon Hohberger log_print("could not bind for connect: %d", result); 11596bd8fedaSLon Hohberger /* This *may* not indicate a critical error */ 11606bd8fedaSLon Hohberger } 11616bd8fedaSLon Hohberger 11626ed7257bSPatrick Caulfield make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); 11636ed7257bSPatrick Caulfield 11646ed7257bSPatrick Caulfield log_print("connecting to %d", con->nodeid); 1165cb2d45daSDavid Teigland 1166cb2d45daSDavid Teigland /* Turn off Nagle's algorithm */ 1167cb2d45daSDavid Teigland kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, 1168cb2d45daSDavid Teigland sizeof(one)); 1169cb2d45daSDavid Teigland 117036b71a8bSDavid Teigland result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, 11716ed7257bSPatrick Caulfield O_NONBLOCK); 11726ed7257bSPatrick Caulfield if (result == -EINPROGRESS) 11736ed7257bSPatrick Caulfield result = 0; 11746ed7257bSPatrick Caulfield if (result == 0) 11756ed7257bSPatrick Caulfield goto out; 11766ed7257bSPatrick Caulfield 11776ed7257bSPatrick Caulfield out_err: 11786ed7257bSPatrick Caulfield if (con->sock) { 11796ed7257bSPatrick Caulfield sock_release(con->sock); 11806ed7257bSPatrick Caulfield con->sock = NULL; 1181a89d63a1SCasey Dahlin } else if (sock) { 1182a89d63a1SCasey Dahlin sock_release(sock); 11836ed7257bSPatrick Caulfield } 11846ed7257bSPatrick Caulfield /* 11856ed7257bSPatrick Caulfield * Some errors are fatal and this list might need adjusting. For other 11866ed7257bSPatrick Caulfield * errors we try again until the max number of retries is reached. 11876ed7257bSPatrick Caulfield */ 118836b71a8bSDavid Teigland if (result != -EHOSTUNREACH && 118936b71a8bSDavid Teigland result != -ENETUNREACH && 119036b71a8bSDavid Teigland result != -ENETDOWN && 119136b71a8bSDavid Teigland result != -EINVAL && 119236b71a8bSDavid Teigland result != -EPROTONOSUPPORT) { 119336b71a8bSDavid Teigland log_print("connect %d try %d error %d", con->nodeid, 119436b71a8bSDavid Teigland con->retries, result); 119536b71a8bSDavid Teigland mutex_unlock(&con->sock_mutex); 119636b71a8bSDavid Teigland msleep(1000); 1197356344c4SMarcelo Ricardo Leitner clear_bit(CF_CONNECT_PENDING, &con->flags); 11986ed7257bSPatrick Caulfield lowcomms_connect_sock(con); 119936b71a8bSDavid Teigland return; 12006ed7257bSPatrick Caulfield } 12016ed7257bSPatrick Caulfield out: 12026ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 120300dcffaeSMarcelo Ricardo Leitner set_bit(CF_WRITE_PENDING, &con->flags); 12046ed7257bSPatrick Caulfield return; 12056ed7257bSPatrick Caulfield } 12066ed7257bSPatrick Caulfield 12076ed7257bSPatrick Caulfield static struct socket *tcp_create_listen_sock(struct connection *con, 12086ed7257bSPatrick Caulfield struct sockaddr_storage *saddr) 12096ed7257bSPatrick Caulfield { 12106ed7257bSPatrick Caulfield struct socket *sock = NULL; 12116ed7257bSPatrick Caulfield int result = 0; 12126ed7257bSPatrick Caulfield int one = 1; 12136ed7257bSPatrick Caulfield int addr_len; 12146ed7257bSPatrick Caulfield 12156ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) 12166ed7257bSPatrick Caulfield addr_len = sizeof(struct sockaddr_in); 12176ed7257bSPatrick Caulfield else 12186ed7257bSPatrick Caulfield addr_len = sizeof(struct sockaddr_in6); 12196ed7257bSPatrick Caulfield 12206ed7257bSPatrick Caulfield /* Create a socket to communicate with */ 1221eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1222eeb1bd5cSEric W. Biederman SOCK_STREAM, IPPROTO_TCP, &sock); 12236ed7257bSPatrick Caulfield if (result < 0) { 1224617e82e1SDavid Teigland log_print("Can't create listening comms socket"); 12256ed7257bSPatrick Caulfield goto create_out; 12266ed7257bSPatrick Caulfield } 12276ed7257bSPatrick Caulfield 1228cb2d45daSDavid Teigland /* Turn off Nagle's algorithm */ 1229cb2d45daSDavid Teigland kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, 1230cb2d45daSDavid Teigland sizeof(one)); 1231cb2d45daSDavid Teigland 12326ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, 12336ed7257bSPatrick Caulfield (char *)&one, sizeof(one)); 12346ed7257bSPatrick Caulfield 12356ed7257bSPatrick Caulfield if (result < 0) { 1236617e82e1SDavid Teigland log_print("Failed to set SO_REUSEADDR on socket: %d", result); 12376ed7257bSPatrick Caulfield } 1238b81171cbSBob Peterson sock->sk->sk_user_data = con; 1239b81171cbSBob Peterson 12406ed7257bSPatrick Caulfield con->rx_action = tcp_accept_from_sock; 12416ed7257bSPatrick Caulfield con->connect_action = tcp_connect_to_sock; 12426ed7257bSPatrick Caulfield 12436ed7257bSPatrick Caulfield /* Bind to our port */ 12446ed7257bSPatrick Caulfield make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); 12456ed7257bSPatrick Caulfield result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); 12466ed7257bSPatrick Caulfield if (result < 0) { 1247617e82e1SDavid Teigland log_print("Can't bind to port %d", dlm_config.ci_tcp_port); 12486ed7257bSPatrick Caulfield sock_release(sock); 12496ed7257bSPatrick Caulfield sock = NULL; 12506ed7257bSPatrick Caulfield con->sock = NULL; 12516ed7257bSPatrick Caulfield goto create_out; 12526ed7257bSPatrick Caulfield } 12536ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 12546ed7257bSPatrick Caulfield (char *)&one, sizeof(one)); 12556ed7257bSPatrick Caulfield if (result < 0) { 1256617e82e1SDavid Teigland log_print("Set keepalive failed: %d", result); 12576ed7257bSPatrick Caulfield } 12586ed7257bSPatrick Caulfield 12596ed7257bSPatrick Caulfield result = sock->ops->listen(sock, 5); 12606ed7257bSPatrick Caulfield if (result < 0) { 1261617e82e1SDavid Teigland log_print("Can't listen on port %d", dlm_config.ci_tcp_port); 12626ed7257bSPatrick Caulfield sock_release(sock); 12636ed7257bSPatrick Caulfield sock = NULL; 12646ed7257bSPatrick Caulfield goto create_out; 12656ed7257bSPatrick Caulfield } 12666ed7257bSPatrick Caulfield 12676ed7257bSPatrick Caulfield create_out: 12686ed7257bSPatrick Caulfield return sock; 12696ed7257bSPatrick Caulfield } 12706ed7257bSPatrick Caulfield 12716ed7257bSPatrick Caulfield /* Get local addresses */ 12726ed7257bSPatrick Caulfield static void init_local(void) 12736ed7257bSPatrick Caulfield { 12746ed7257bSPatrick Caulfield struct sockaddr_storage sas, *addr; 12756ed7257bSPatrick Caulfield int i; 12766ed7257bSPatrick Caulfield 127730d3a237SPatrick Caulfield dlm_local_count = 0; 12781b189b88SDavid Teigland for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) { 12796ed7257bSPatrick Caulfield if (dlm_our_addr(&sas, i)) 12806ed7257bSPatrick Caulfield break; 12816ed7257bSPatrick Caulfield 12825c93f56fSAmitoj Kaur Chawla addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS); 12836ed7257bSPatrick Caulfield if (!addr) 12846ed7257bSPatrick Caulfield break; 12856ed7257bSPatrick Caulfield dlm_local_addr[dlm_local_count++] = addr; 12866ed7257bSPatrick Caulfield } 12876ed7257bSPatrick Caulfield } 12886ed7257bSPatrick Caulfield 12896ed7257bSPatrick Caulfield /* Initialise SCTP socket and bind to all interfaces */ 12906ed7257bSPatrick Caulfield static int sctp_listen_for_all(void) 12916ed7257bSPatrick Caulfield { 12926ed7257bSPatrick Caulfield struct socket *sock = NULL; 1293ee44b4bcSMarcelo Ricardo Leitner int result = -EINVAL; 1294573c24c4SDavid Teigland struct connection *con = nodeid2con(0, GFP_NOFS); 12956ed7257bSPatrick Caulfield int bufsize = NEEDED_RMEM; 129686e92ad2SMike Christie int one = 1; 12976ed7257bSPatrick Caulfield 12986ed7257bSPatrick Caulfield if (!con) 12996ed7257bSPatrick Caulfield return -ENOMEM; 13006ed7257bSPatrick Caulfield 13016ed7257bSPatrick Caulfield log_print("Using SCTP for communications"); 13026ed7257bSPatrick Caulfield 1303eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1304ee44b4bcSMarcelo Ricardo Leitner SOCK_STREAM, IPPROTO_SCTP, &sock); 13056ed7257bSPatrick Caulfield if (result < 0) { 13066ed7257bSPatrick Caulfield log_print("Can't create comms socket, check SCTP is loaded"); 13076ed7257bSPatrick Caulfield goto out; 13086ed7257bSPatrick Caulfield } 13096ed7257bSPatrick Caulfield 1310df61c952SDavid S. Miller result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, 13116ed7257bSPatrick Caulfield (char *)&bufsize, sizeof(bufsize)); 13126ed7257bSPatrick Caulfield if (result) 1313617e82e1SDavid Teigland log_print("Error increasing buffer space on socket %d", result); 13146ed7257bSPatrick Caulfield 131586e92ad2SMike Christie result = kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one, 131686e92ad2SMike Christie sizeof(one)); 131786e92ad2SMike Christie if (result < 0) 131886e92ad2SMike Christie log_print("Could not set SCTP NODELAY error %d\n", result); 131986e92ad2SMike Christie 1320b81171cbSBob Peterson write_lock_bh(&sock->sk->sk_callback_lock); 13216ed7257bSPatrick Caulfield /* Init con struct */ 13226ed7257bSPatrick Caulfield sock->sk->sk_user_data = con; 13236ed7257bSPatrick Caulfield con->sock = sock; 13246ed7257bSPatrick Caulfield con->sock->sk->sk_data_ready = lowcomms_data_ready; 1325ee44b4bcSMarcelo Ricardo Leitner con->rx_action = sctp_accept_from_sock; 1326ee44b4bcSMarcelo Ricardo Leitner con->connect_action = sctp_connect_to_sock; 13276ed7257bSPatrick Caulfield 1328b81171cbSBob Peterson write_unlock_bh(&sock->sk->sk_callback_lock); 1329b81171cbSBob Peterson 1330ee44b4bcSMarcelo Ricardo Leitner /* Bind to all addresses. */ 1331ee44b4bcSMarcelo Ricardo Leitner if (sctp_bind_addrs(con, dlm_config.ci_tcp_port)) 13326ed7257bSPatrick Caulfield goto create_delsock; 13336ed7257bSPatrick Caulfield 13346ed7257bSPatrick Caulfield result = sock->ops->listen(sock, 5); 13356ed7257bSPatrick Caulfield if (result < 0) { 13366ed7257bSPatrick Caulfield log_print("Can't set socket listening"); 13376ed7257bSPatrick Caulfield goto create_delsock; 13386ed7257bSPatrick Caulfield } 13396ed7257bSPatrick Caulfield 13406ed7257bSPatrick Caulfield return 0; 13416ed7257bSPatrick Caulfield 13426ed7257bSPatrick Caulfield create_delsock: 13436ed7257bSPatrick Caulfield sock_release(sock); 13446ed7257bSPatrick Caulfield con->sock = NULL; 13456ed7257bSPatrick Caulfield out: 13466ed7257bSPatrick Caulfield return result; 13476ed7257bSPatrick Caulfield } 13486ed7257bSPatrick Caulfield 13496ed7257bSPatrick Caulfield static int tcp_listen_for_all(void) 13506ed7257bSPatrick Caulfield { 13516ed7257bSPatrick Caulfield struct socket *sock = NULL; 1352573c24c4SDavid Teigland struct connection *con = nodeid2con(0, GFP_NOFS); 13536ed7257bSPatrick Caulfield int result = -EINVAL; 13546ed7257bSPatrick Caulfield 13556ed7257bSPatrick Caulfield if (!con) 13566ed7257bSPatrick Caulfield return -ENOMEM; 13576ed7257bSPatrick Caulfield 13586ed7257bSPatrick Caulfield /* We don't support multi-homed hosts */ 13596ed7257bSPatrick Caulfield if (dlm_local_addr[1] != NULL) { 1360617e82e1SDavid Teigland log_print("TCP protocol can't handle multi-homed hosts, " 1361617e82e1SDavid Teigland "try SCTP"); 13626ed7257bSPatrick Caulfield return -EINVAL; 13636ed7257bSPatrick Caulfield } 13646ed7257bSPatrick Caulfield 13656ed7257bSPatrick Caulfield log_print("Using TCP for communications"); 13666ed7257bSPatrick Caulfield 13676ed7257bSPatrick Caulfield sock = tcp_create_listen_sock(con, dlm_local_addr[0]); 13686ed7257bSPatrick Caulfield if (sock) { 13693735b4b9SBob Peterson add_sock(sock, con, true); 13706ed7257bSPatrick Caulfield result = 0; 13716ed7257bSPatrick Caulfield } 13726ed7257bSPatrick Caulfield else { 13736ed7257bSPatrick Caulfield result = -EADDRINUSE; 13746ed7257bSPatrick Caulfield } 13756ed7257bSPatrick Caulfield 13766ed7257bSPatrick Caulfield return result; 13776ed7257bSPatrick Caulfield } 13786ed7257bSPatrick Caulfield 13796ed7257bSPatrick Caulfield 13806ed7257bSPatrick Caulfield 13816ed7257bSPatrick Caulfield static struct writequeue_entry *new_writequeue_entry(struct connection *con, 13826ed7257bSPatrick Caulfield gfp_t allocation) 13836ed7257bSPatrick Caulfield { 13846ed7257bSPatrick Caulfield struct writequeue_entry *entry; 13856ed7257bSPatrick Caulfield 13866ed7257bSPatrick Caulfield entry = kmalloc(sizeof(struct writequeue_entry), allocation); 13876ed7257bSPatrick Caulfield if (!entry) 13886ed7257bSPatrick Caulfield return NULL; 13896ed7257bSPatrick Caulfield 13906ed7257bSPatrick Caulfield entry->page = alloc_page(allocation); 13916ed7257bSPatrick Caulfield if (!entry->page) { 13926ed7257bSPatrick Caulfield kfree(entry); 13936ed7257bSPatrick Caulfield return NULL; 13946ed7257bSPatrick Caulfield } 13956ed7257bSPatrick Caulfield 13966ed7257bSPatrick Caulfield entry->offset = 0; 13976ed7257bSPatrick Caulfield entry->len = 0; 13986ed7257bSPatrick Caulfield entry->end = 0; 13996ed7257bSPatrick Caulfield entry->users = 0; 14006ed7257bSPatrick Caulfield entry->con = con; 14016ed7257bSPatrick Caulfield 14026ed7257bSPatrick Caulfield return entry; 14036ed7257bSPatrick Caulfield } 14046ed7257bSPatrick Caulfield 1405617e82e1SDavid Teigland void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) 14066ed7257bSPatrick Caulfield { 14076ed7257bSPatrick Caulfield struct connection *con; 14086ed7257bSPatrick Caulfield struct writequeue_entry *e; 14096ed7257bSPatrick Caulfield int offset = 0; 14106ed7257bSPatrick Caulfield 14116ed7257bSPatrick Caulfield con = nodeid2con(nodeid, allocation); 14126ed7257bSPatrick Caulfield if (!con) 14136ed7257bSPatrick Caulfield return NULL; 14146ed7257bSPatrick Caulfield 14156ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14166ed7257bSPatrick Caulfield e = list_entry(con->writequeue.prev, struct writequeue_entry, list); 14176ed7257bSPatrick Caulfield if ((&e->list == &con->writequeue) || 141809cbfeafSKirill A. Shutemov (PAGE_SIZE - e->end < len)) { 14196ed7257bSPatrick Caulfield e = NULL; 14206ed7257bSPatrick Caulfield } else { 14216ed7257bSPatrick Caulfield offset = e->end; 14226ed7257bSPatrick Caulfield e->end += len; 1423eeee2b5fSWei Yongjun e->users++; 14246ed7257bSPatrick Caulfield } 14256ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14266ed7257bSPatrick Caulfield 14276ed7257bSPatrick Caulfield if (e) { 14286ed7257bSPatrick Caulfield got_one: 14296ed7257bSPatrick Caulfield *ppc = page_address(e->page) + offset; 14306ed7257bSPatrick Caulfield return e; 14316ed7257bSPatrick Caulfield } 14326ed7257bSPatrick Caulfield 14336ed7257bSPatrick Caulfield e = new_writequeue_entry(con, allocation); 14346ed7257bSPatrick Caulfield if (e) { 14356ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14366ed7257bSPatrick Caulfield offset = e->end; 14376ed7257bSPatrick Caulfield e->end += len; 1438eeee2b5fSWei Yongjun e->users++; 14396ed7257bSPatrick Caulfield list_add_tail(&e->list, &con->writequeue); 14406ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14416ed7257bSPatrick Caulfield goto got_one; 14426ed7257bSPatrick Caulfield } 14436ed7257bSPatrick Caulfield return NULL; 14446ed7257bSPatrick Caulfield } 14456ed7257bSPatrick Caulfield 14466ed7257bSPatrick Caulfield void dlm_lowcomms_commit_buffer(void *mh) 14476ed7257bSPatrick Caulfield { 14486ed7257bSPatrick Caulfield struct writequeue_entry *e = (struct writequeue_entry *)mh; 14496ed7257bSPatrick Caulfield struct connection *con = e->con; 14506ed7257bSPatrick Caulfield int users; 14516ed7257bSPatrick Caulfield 14526ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14536ed7257bSPatrick Caulfield users = --e->users; 14546ed7257bSPatrick Caulfield if (users) 14556ed7257bSPatrick Caulfield goto out; 14566ed7257bSPatrick Caulfield e->len = e->end - e->offset; 14576ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14586ed7257bSPatrick Caulfield 14596ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { 14606ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 14616ed7257bSPatrick Caulfield } 14626ed7257bSPatrick Caulfield return; 14636ed7257bSPatrick Caulfield 14646ed7257bSPatrick Caulfield out: 14656ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14666ed7257bSPatrick Caulfield return; 14676ed7257bSPatrick Caulfield } 14686ed7257bSPatrick Caulfield 14696ed7257bSPatrick Caulfield /* Send a message */ 14706ed7257bSPatrick Caulfield static void send_to_sock(struct connection *con) 14716ed7257bSPatrick Caulfield { 14726ed7257bSPatrick Caulfield int ret = 0; 14736ed7257bSPatrick Caulfield const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; 14746ed7257bSPatrick Caulfield struct writequeue_entry *e; 14756ed7257bSPatrick Caulfield int len, offset; 1476f92c8dd7SBob Peterson int count = 0; 14776ed7257bSPatrick Caulfield 14786ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 14796ed7257bSPatrick Caulfield if (con->sock == NULL) 14806ed7257bSPatrick Caulfield goto out_connect; 14816ed7257bSPatrick Caulfield 14826ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14836ed7257bSPatrick Caulfield for (;;) { 14846ed7257bSPatrick Caulfield e = list_entry(con->writequeue.next, struct writequeue_entry, 14856ed7257bSPatrick Caulfield list); 14866ed7257bSPatrick Caulfield if ((struct list_head *) e == &con->writequeue) 14876ed7257bSPatrick Caulfield break; 14886ed7257bSPatrick Caulfield 14896ed7257bSPatrick Caulfield len = e->len; 14906ed7257bSPatrick Caulfield offset = e->offset; 14916ed7257bSPatrick Caulfield BUG_ON(len == 0 && e->users == 0); 14926ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14936ed7257bSPatrick Caulfield 14946ed7257bSPatrick Caulfield ret = 0; 14956ed7257bSPatrick Caulfield if (len) { 14961329e3f2SPaolo Bonzini ret = kernel_sendpage(con->sock, e->page, offset, len, 14976ed7257bSPatrick Caulfield msg_flags); 1498d66f8277SPatrick Caulfield if (ret == -EAGAIN || ret == 0) { 1499b36930ddSDavid Miller if (ret == -EAGAIN && 15009cd3e072SEric Dumazet test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) && 1501b36930ddSDavid Miller !test_and_set_bit(CF_APP_LIMITED, &con->flags)) { 1502b36930ddSDavid Miller /* Notify TCP that we're limited by the 1503b36930ddSDavid Miller * application window size. 1504b36930ddSDavid Miller */ 1505b36930ddSDavid Miller set_bit(SOCK_NOSPACE, &con->sock->flags); 1506b36930ddSDavid Miller con->sock->sk->sk_write_pending++; 1507b36930ddSDavid Miller } 1508d66f8277SPatrick Caulfield cond_resched(); 15096ed7257bSPatrick Caulfield goto out; 15109c5bef58SYing Xue } else if (ret < 0) 15116ed7257bSPatrick Caulfield goto send_error; 1512d66f8277SPatrick Caulfield } 1513f92c8dd7SBob Peterson 15146ed7257bSPatrick Caulfield /* Don't starve people filling buffers */ 1515f92c8dd7SBob Peterson if (++count >= MAX_SEND_MSG_COUNT) { 15166ed7257bSPatrick Caulfield cond_resched(); 1517f92c8dd7SBob Peterson count = 0; 1518f92c8dd7SBob Peterson } 15196ed7257bSPatrick Caulfield 15206ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 15215d689871SMike Christie writequeue_entry_complete(e, ret); 15226ed7257bSPatrick Caulfield } 15236ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15246ed7257bSPatrick Caulfield out: 15256ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 15266ed7257bSPatrick Caulfield return; 15276ed7257bSPatrick Caulfield 15286ed7257bSPatrick Caulfield send_error: 15296ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 15300d737a8cSMarcelo Ricardo Leitner close_connection(con, false, false, true); 15316ed7257bSPatrick Caulfield lowcomms_connect_sock(con); 15326ed7257bSPatrick Caulfield return; 15336ed7257bSPatrick Caulfield 15346ed7257bSPatrick Caulfield out_connect: 15356ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 15366ed7257bSPatrick Caulfield lowcomms_connect_sock(con); 15376ed7257bSPatrick Caulfield } 15386ed7257bSPatrick Caulfield 15396ed7257bSPatrick Caulfield static void clean_one_writequeue(struct connection *con) 15406ed7257bSPatrick Caulfield { 15415e9ccc37SChristine Caulfield struct writequeue_entry *e, *safe; 15426ed7257bSPatrick Caulfield 15436ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 15445e9ccc37SChristine Caulfield list_for_each_entry_safe(e, safe, &con->writequeue, list) { 15456ed7257bSPatrick Caulfield list_del(&e->list); 15466ed7257bSPatrick Caulfield free_entry(e); 15476ed7257bSPatrick Caulfield } 15486ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15496ed7257bSPatrick Caulfield } 15506ed7257bSPatrick Caulfield 15516ed7257bSPatrick Caulfield /* Called from recovery when it knows that a node has 15526ed7257bSPatrick Caulfield left the cluster */ 15536ed7257bSPatrick Caulfield int dlm_lowcomms_close(int nodeid) 15546ed7257bSPatrick Caulfield { 15556ed7257bSPatrick Caulfield struct connection *con; 155636b71a8bSDavid Teigland struct dlm_node_addr *na; 15576ed7257bSPatrick Caulfield 15586ed7257bSPatrick Caulfield log_print("closing connection to node %d", nodeid); 15596ed7257bSPatrick Caulfield con = nodeid2con(nodeid, 0); 15606ed7257bSPatrick Caulfield if (con) { 1561063c4c99SLars Marowsky-Bree set_bit(CF_CLOSE, &con->flags); 15620d737a8cSMarcelo Ricardo Leitner close_connection(con, true, true, true); 15636ed7257bSPatrick Caulfield clean_one_writequeue(con); 15646ed7257bSPatrick Caulfield } 156536b71a8bSDavid Teigland 156636b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 156736b71a8bSDavid Teigland na = find_node_addr(nodeid); 156836b71a8bSDavid Teigland if (na) { 156936b71a8bSDavid Teigland list_del(&na->list); 157036b71a8bSDavid Teigland while (na->addr_count--) 157136b71a8bSDavid Teigland kfree(na->addr[na->addr_count]); 157236b71a8bSDavid Teigland kfree(na); 157336b71a8bSDavid Teigland } 157436b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 157536b71a8bSDavid Teigland 15766ed7257bSPatrick Caulfield return 0; 15776ed7257bSPatrick Caulfield } 15786ed7257bSPatrick Caulfield 15796ed7257bSPatrick Caulfield /* Receive workqueue function */ 15806ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work) 15816ed7257bSPatrick Caulfield { 15826ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, rwork); 15836ed7257bSPatrick Caulfield int err; 15846ed7257bSPatrick Caulfield 15856ed7257bSPatrick Caulfield clear_bit(CF_READ_PENDING, &con->flags); 15866ed7257bSPatrick Caulfield do { 15876ed7257bSPatrick Caulfield err = con->rx_action(con); 15886ed7257bSPatrick Caulfield } while (!err); 15896ed7257bSPatrick Caulfield } 15906ed7257bSPatrick Caulfield 15916ed7257bSPatrick Caulfield /* Send workqueue function */ 15926ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work) 15936ed7257bSPatrick Caulfield { 15946ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, swork); 15956ed7257bSPatrick Caulfield 159600dcffaeSMarcelo Ricardo Leitner if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) 15976ed7257bSPatrick Caulfield con->connect_action(con); 1598063c4c99SLars Marowsky-Bree if (test_and_clear_bit(CF_WRITE_PENDING, &con->flags)) 15996ed7257bSPatrick Caulfield send_to_sock(con); 16006ed7257bSPatrick Caulfield } 16016ed7257bSPatrick Caulfield 16026ed7257bSPatrick Caulfield 16036ed7257bSPatrick Caulfield /* Discard all entries on the write queues */ 16046ed7257bSPatrick Caulfield static void clean_writequeues(void) 16056ed7257bSPatrick Caulfield { 16065e9ccc37SChristine Caulfield foreach_conn(clean_one_writequeue); 16076ed7257bSPatrick Caulfield } 16086ed7257bSPatrick Caulfield 16096ed7257bSPatrick Caulfield static void work_stop(void) 16106ed7257bSPatrick Caulfield { 16116ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 16126ed7257bSPatrick Caulfield destroy_workqueue(send_workqueue); 16136ed7257bSPatrick Caulfield } 16146ed7257bSPatrick Caulfield 16156ed7257bSPatrick Caulfield static int work_start(void) 16166ed7257bSPatrick Caulfield { 1617e43f055aSDavid Teigland recv_workqueue = alloc_workqueue("dlm_recv", 1618e43f055aSDavid Teigland WQ_UNBOUND | WQ_MEM_RECLAIM, 1); 1619b9d41052SNamhyung Kim if (!recv_workqueue) { 1620b9d41052SNamhyung Kim log_print("can't start dlm_recv"); 1621b9d41052SNamhyung Kim return -ENOMEM; 16226ed7257bSPatrick Caulfield } 16236ed7257bSPatrick Caulfield 1624e43f055aSDavid Teigland send_workqueue = alloc_workqueue("dlm_send", 1625e43f055aSDavid Teigland WQ_UNBOUND | WQ_MEM_RECLAIM, 1); 1626b9d41052SNamhyung Kim if (!send_workqueue) { 1627b9d41052SNamhyung Kim log_print("can't start dlm_send"); 16286ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 1629b9d41052SNamhyung Kim return -ENOMEM; 16306ed7257bSPatrick Caulfield } 16316ed7257bSPatrick Caulfield 16326ed7257bSPatrick Caulfield return 0; 16336ed7257bSPatrick Caulfield } 16346ed7257bSPatrick Caulfield 16355e9ccc37SChristine Caulfield static void stop_conn(struct connection *con) 16366ed7257bSPatrick Caulfield { 16379e5f2825SPatrick Caulfield con->flags |= 0x0F; 1638391fbdc5SChristine Caulfield if (con->sock && con->sock->sk) 1639afb853fbSPatrick Caulfield con->sock->sk->sk_user_data = NULL; 1640afb853fbSPatrick Caulfield } 16415e9ccc37SChristine Caulfield 16425e9ccc37SChristine Caulfield static void free_conn(struct connection *con) 16435e9ccc37SChristine Caulfield { 16440d737a8cSMarcelo Ricardo Leitner close_connection(con, true, true, true); 16455e9ccc37SChristine Caulfield if (con->othercon) 16465e9ccc37SChristine Caulfield kmem_cache_free(con_cache, con->othercon); 16475e9ccc37SChristine Caulfield hlist_del(&con->list); 16485e9ccc37SChristine Caulfield kmem_cache_free(con_cache, con); 16496ed7257bSPatrick Caulfield } 16505e9ccc37SChristine Caulfield 16515e9ccc37SChristine Caulfield void dlm_lowcomms_stop(void) 16525e9ccc37SChristine Caulfield { 16535e9ccc37SChristine Caulfield /* Set all the flags to prevent any 16545e9ccc37SChristine Caulfield socket activity. 16555e9ccc37SChristine Caulfield */ 16565e9ccc37SChristine Caulfield mutex_lock(&connections_lock); 1657513ef596SDavid Teigland dlm_allow_conn = 0; 16585e9ccc37SChristine Caulfield foreach_conn(stop_conn); 16593a8db798SMarcelo Ricardo Leitner clean_writequeues(); 16603a8db798SMarcelo Ricardo Leitner foreach_conn(free_conn); 16617a936ce7SMatthias Kaehlcke mutex_unlock(&connections_lock); 16626ed7257bSPatrick Caulfield 16636ed7257bSPatrick Caulfield work_stop(); 16646ed7257bSPatrick Caulfield 16656ed7257bSPatrick Caulfield kmem_cache_destroy(con_cache); 16666ed7257bSPatrick Caulfield } 16676ed7257bSPatrick Caulfield 16686ed7257bSPatrick Caulfield int dlm_lowcomms_start(void) 16696ed7257bSPatrick Caulfield { 16706ed7257bSPatrick Caulfield int error = -EINVAL; 16716ed7257bSPatrick Caulfield struct connection *con; 16725e9ccc37SChristine Caulfield int i; 16735e9ccc37SChristine Caulfield 16745e9ccc37SChristine Caulfield for (i = 0; i < CONN_HASH_SIZE; i++) 16755e9ccc37SChristine Caulfield INIT_HLIST_HEAD(&connection_hash[i]); 16766ed7257bSPatrick Caulfield 16776ed7257bSPatrick Caulfield init_local(); 16786ed7257bSPatrick Caulfield if (!dlm_local_count) { 1679617e82e1SDavid Teigland error = -ENOTCONN; 16806ed7257bSPatrick Caulfield log_print("no local IP address has been set"); 1681513ef596SDavid Teigland goto fail; 16826ed7257bSPatrick Caulfield } 16836ed7257bSPatrick Caulfield 16846ed7257bSPatrick Caulfield error = -ENOMEM; 16856ed7257bSPatrick Caulfield con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), 16866ed7257bSPatrick Caulfield __alignof__(struct connection), 0, 168720c2df83SPaul Mundt NULL); 16886ed7257bSPatrick Caulfield if (!con_cache) 1689513ef596SDavid Teigland goto fail; 1690513ef596SDavid Teigland 1691513ef596SDavid Teigland error = work_start(); 1692513ef596SDavid Teigland if (error) 1693513ef596SDavid Teigland goto fail_destroy; 1694513ef596SDavid Teigland 1695513ef596SDavid Teigland dlm_allow_conn = 1; 16966ed7257bSPatrick Caulfield 16976ed7257bSPatrick Caulfield /* Start listening */ 16986ed7257bSPatrick Caulfield if (dlm_config.ci_protocol == 0) 16996ed7257bSPatrick Caulfield error = tcp_listen_for_all(); 17006ed7257bSPatrick Caulfield else 17016ed7257bSPatrick Caulfield error = sctp_listen_for_all(); 17026ed7257bSPatrick Caulfield if (error) 17036ed7257bSPatrick Caulfield goto fail_unlisten; 17046ed7257bSPatrick Caulfield 17056ed7257bSPatrick Caulfield return 0; 17066ed7257bSPatrick Caulfield 17076ed7257bSPatrick Caulfield fail_unlisten: 1708513ef596SDavid Teigland dlm_allow_conn = 0; 17096ed7257bSPatrick Caulfield con = nodeid2con(0,0); 17106ed7257bSPatrick Caulfield if (con) { 17110d737a8cSMarcelo Ricardo Leitner close_connection(con, false, true, true); 17126ed7257bSPatrick Caulfield kmem_cache_free(con_cache, con); 17136ed7257bSPatrick Caulfield } 1714513ef596SDavid Teigland fail_destroy: 17156ed7257bSPatrick Caulfield kmem_cache_destroy(con_cache); 1716513ef596SDavid Teigland fail: 17176ed7257bSPatrick Caulfield return error; 17186ed7257bSPatrick Caulfield } 171936b71a8bSDavid Teigland 172036b71a8bSDavid Teigland void dlm_lowcomms_exit(void) 172136b71a8bSDavid Teigland { 172236b71a8bSDavid Teigland struct dlm_node_addr *na, *safe; 172336b71a8bSDavid Teigland 172436b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 172536b71a8bSDavid Teigland list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) { 172636b71a8bSDavid Teigland list_del(&na->list); 172736b71a8bSDavid Teigland while (na->addr_count--) 172836b71a8bSDavid Teigland kfree(na->addr[na->addr_count]); 172936b71a8bSDavid Teigland kfree(na); 173036b71a8bSDavid Teigland } 173136b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 173236b71a8bSDavid Teigland } 1733