12522fe45SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 26ed7257bSPatrick Caulfield /****************************************************************************** 36ed7257bSPatrick Caulfield ******************************************************************************* 46ed7257bSPatrick Caulfield ** 56ed7257bSPatrick Caulfield ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 65e9ccc37SChristine Caulfield ** Copyright (C) 2004-2009 Red Hat, Inc. All rights reserved. 76ed7257bSPatrick Caulfield ** 86ed7257bSPatrick Caulfield ** 96ed7257bSPatrick Caulfield ******************************************************************************* 106ed7257bSPatrick Caulfield ******************************************************************************/ 116ed7257bSPatrick Caulfield 126ed7257bSPatrick Caulfield /* 136ed7257bSPatrick Caulfield * lowcomms.c 146ed7257bSPatrick Caulfield * 156ed7257bSPatrick Caulfield * This is the "low-level" comms layer. 166ed7257bSPatrick Caulfield * 176ed7257bSPatrick Caulfield * It is responsible for sending/receiving messages 186ed7257bSPatrick Caulfield * from other nodes in the cluster. 196ed7257bSPatrick Caulfield * 206ed7257bSPatrick Caulfield * Cluster nodes are referred to by their nodeids. nodeids are 216ed7257bSPatrick Caulfield * simply 32 bit numbers to the locking module - if they need to 222cf12c0bSJoe Perches * be expanded for the cluster infrastructure then that is its 236ed7257bSPatrick Caulfield * responsibility. It is this layer's 246ed7257bSPatrick Caulfield * responsibility to resolve these into IP address or 256ed7257bSPatrick Caulfield * whatever it needs for inter-node communication. 266ed7257bSPatrick Caulfield * 276ed7257bSPatrick Caulfield * The comms level is two kernel threads that deal mainly with 286ed7257bSPatrick Caulfield * the receiving of messages from other nodes and passing them 296ed7257bSPatrick Caulfield * up to the mid-level comms layer (which understands the 306ed7257bSPatrick Caulfield * message format) for execution by the locking core, and 316ed7257bSPatrick Caulfield * a send thread which does all the setting up of connections 326ed7257bSPatrick Caulfield * to remote nodes and the sending of data. Threads are not allowed 336ed7257bSPatrick Caulfield * to send their own data because it may cause them to wait in times 346ed7257bSPatrick Caulfield * of high load. Also, this way, the sending thread can collect together 356ed7257bSPatrick Caulfield * messages bound for one node and send them in one block. 366ed7257bSPatrick Caulfield * 372cf12c0bSJoe Perches * lowcomms will choose to use either TCP or SCTP as its transport layer 386ed7257bSPatrick Caulfield * depending on the configuration variable 'protocol'. This should be set 396ed7257bSPatrick Caulfield * to 0 (default) for TCP or 1 for SCTP. It should be configured using a 406ed7257bSPatrick Caulfield * cluster-wide mechanism as it must be the same on all nodes of the cluster 416ed7257bSPatrick Caulfield * for the DLM to function. 426ed7257bSPatrick Caulfield * 436ed7257bSPatrick Caulfield */ 446ed7257bSPatrick Caulfield 456ed7257bSPatrick Caulfield #include <asm/ioctls.h> 466ed7257bSPatrick Caulfield #include <net/sock.h> 476ed7257bSPatrick Caulfield #include <net/tcp.h> 486ed7257bSPatrick Caulfield #include <linux/pagemap.h> 496ed7257bSPatrick Caulfield #include <linux/file.h> 507a936ce7SMatthias Kaehlcke #include <linux/mutex.h> 516ed7257bSPatrick Caulfield #include <linux/sctp.h> 525a0e3ad6STejun Heo #include <linux/slab.h> 532f2d76ccSBenjamin Poirier #include <net/sctp/sctp.h> 5444ad532bSJoe Perches #include <net/ipv6.h> 556ed7257bSPatrick Caulfield 566ed7257bSPatrick Caulfield #include "dlm_internal.h" 576ed7257bSPatrick Caulfield #include "lowcomms.h" 586ed7257bSPatrick Caulfield #include "midcomms.h" 596ed7257bSPatrick Caulfield #include "config.h" 606ed7257bSPatrick Caulfield 616ed7257bSPatrick Caulfield #define NEEDED_RMEM (4*1024*1024) 625e9ccc37SChristine Caulfield #define CONN_HASH_SIZE 32 636ed7257bSPatrick Caulfield 64f92c8dd7SBob Peterson /* Number of messages to send before rescheduling */ 65f92c8dd7SBob Peterson #define MAX_SEND_MSG_COUNT 25 66f92c8dd7SBob Peterson 676ed7257bSPatrick Caulfield struct cbuf { 686ed7257bSPatrick Caulfield unsigned int base; 696ed7257bSPatrick Caulfield unsigned int len; 706ed7257bSPatrick Caulfield unsigned int mask; 716ed7257bSPatrick Caulfield }; 726ed7257bSPatrick Caulfield 736ed7257bSPatrick Caulfield static void cbuf_add(struct cbuf *cb, int n) 746ed7257bSPatrick Caulfield { 756ed7257bSPatrick Caulfield cb->len += n; 766ed7257bSPatrick Caulfield } 776ed7257bSPatrick Caulfield 786ed7257bSPatrick Caulfield static int cbuf_data(struct cbuf *cb) 796ed7257bSPatrick Caulfield { 806ed7257bSPatrick Caulfield return ((cb->base + cb->len) & cb->mask); 816ed7257bSPatrick Caulfield } 826ed7257bSPatrick Caulfield 836ed7257bSPatrick Caulfield static void cbuf_init(struct cbuf *cb, int size) 846ed7257bSPatrick Caulfield { 856ed7257bSPatrick Caulfield cb->base = cb->len = 0; 866ed7257bSPatrick Caulfield cb->mask = size-1; 876ed7257bSPatrick Caulfield } 886ed7257bSPatrick Caulfield 896ed7257bSPatrick Caulfield static void cbuf_eat(struct cbuf *cb, int n) 906ed7257bSPatrick Caulfield { 916ed7257bSPatrick Caulfield cb->len -= n; 926ed7257bSPatrick Caulfield cb->base += n; 936ed7257bSPatrick Caulfield cb->base &= cb->mask; 946ed7257bSPatrick Caulfield } 956ed7257bSPatrick Caulfield 966ed7257bSPatrick Caulfield static bool cbuf_empty(struct cbuf *cb) 976ed7257bSPatrick Caulfield { 986ed7257bSPatrick Caulfield return cb->len == 0; 996ed7257bSPatrick Caulfield } 1006ed7257bSPatrick Caulfield 1016ed7257bSPatrick Caulfield struct connection { 1026ed7257bSPatrick Caulfield struct socket *sock; /* NULL if not connected */ 1036ed7257bSPatrick Caulfield uint32_t nodeid; /* So we know who we are in the list */ 1046ed7257bSPatrick Caulfield struct mutex sock_mutex; 1056ed7257bSPatrick Caulfield unsigned long flags; 1066ed7257bSPatrick Caulfield #define CF_READ_PENDING 1 1078a4abb08Stsutomu.owa@toshiba.co.jp #define CF_WRITE_PENDING 2 1086ed7257bSPatrick Caulfield #define CF_INIT_PENDING 4 1096ed7257bSPatrick Caulfield #define CF_IS_OTHERCON 5 110063c4c99SLars Marowsky-Bree #define CF_CLOSE 6 111b36930ddSDavid Miller #define CF_APP_LIMITED 7 112b2a66629Stsutomu.owa@toshiba.co.jp #define CF_CLOSING 8 1136ed7257bSPatrick Caulfield struct list_head writequeue; /* List of outgoing writequeue_entries */ 1146ed7257bSPatrick Caulfield spinlock_t writequeue_lock; 1156ed7257bSPatrick Caulfield int (*rx_action) (struct connection *); /* What to do when active */ 1166ed7257bSPatrick Caulfield void (*connect_action) (struct connection *); /* What to do to connect */ 1176ed7257bSPatrick Caulfield struct page *rx_page; 1186ed7257bSPatrick Caulfield struct cbuf cb; 1196ed7257bSPatrick Caulfield int retries; 1206ed7257bSPatrick Caulfield #define MAX_CONNECT_RETRIES 3 1215e9ccc37SChristine Caulfield struct hlist_node list; 1226ed7257bSPatrick Caulfield struct connection *othercon; 1236ed7257bSPatrick Caulfield struct work_struct rwork; /* Receive workqueue */ 1246ed7257bSPatrick Caulfield struct work_struct swork; /* Send workqueue */ 1256ed7257bSPatrick Caulfield }; 1266ed7257bSPatrick Caulfield #define sock2con(x) ((struct connection *)(x)->sk_user_data) 1276ed7257bSPatrick Caulfield 1286ed7257bSPatrick Caulfield /* An entry waiting to be sent */ 1296ed7257bSPatrick Caulfield struct writequeue_entry { 1306ed7257bSPatrick Caulfield struct list_head list; 1316ed7257bSPatrick Caulfield struct page *page; 1326ed7257bSPatrick Caulfield int offset; 1336ed7257bSPatrick Caulfield int len; 1346ed7257bSPatrick Caulfield int end; 1356ed7257bSPatrick Caulfield int users; 1366ed7257bSPatrick Caulfield struct connection *con; 1376ed7257bSPatrick Caulfield }; 1386ed7257bSPatrick Caulfield 13936b71a8bSDavid Teigland struct dlm_node_addr { 14036b71a8bSDavid Teigland struct list_head list; 14136b71a8bSDavid Teigland int nodeid; 14236b71a8bSDavid Teigland int addr_count; 14398e1b60eSMike Christie int curr_addr_index; 14436b71a8bSDavid Teigland struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT]; 14536b71a8bSDavid Teigland }; 14636b71a8bSDavid Teigland 147cc661fc9SBob Peterson static struct listen_sock_callbacks { 148cc661fc9SBob Peterson void (*sk_error_report)(struct sock *); 149cc661fc9SBob Peterson void (*sk_data_ready)(struct sock *); 150cc661fc9SBob Peterson void (*sk_state_change)(struct sock *); 151cc661fc9SBob Peterson void (*sk_write_space)(struct sock *); 152cc661fc9SBob Peterson } listen_sock; 153cc661fc9SBob Peterson 15436b71a8bSDavid Teigland static LIST_HEAD(dlm_node_addrs); 15536b71a8bSDavid Teigland static DEFINE_SPINLOCK(dlm_node_addrs_spin); 15636b71a8bSDavid Teigland 1576ed7257bSPatrick Caulfield static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; 1586ed7257bSPatrick Caulfield static int dlm_local_count; 159513ef596SDavid Teigland static int dlm_allow_conn; 1606ed7257bSPatrick Caulfield 1616ed7257bSPatrick Caulfield /* Work queues */ 1626ed7257bSPatrick Caulfield static struct workqueue_struct *recv_workqueue; 1636ed7257bSPatrick Caulfield static struct workqueue_struct *send_workqueue; 1646ed7257bSPatrick Caulfield 1655e9ccc37SChristine Caulfield static struct hlist_head connection_hash[CONN_HASH_SIZE]; 1667a936ce7SMatthias Kaehlcke static DEFINE_MUTEX(connections_lock); 1676ed7257bSPatrick Caulfield static struct kmem_cache *con_cache; 1686ed7257bSPatrick Caulfield 1696ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work); 1706ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work); 1716ed7257bSPatrick Caulfield 1725e9ccc37SChristine Caulfield 1735e9ccc37SChristine Caulfield /* This is deliberately very simple because most clusters have simple 1745e9ccc37SChristine Caulfield sequential nodeids, so we should be able to go straight to a connection 1755e9ccc37SChristine Caulfield struct in the array */ 1765e9ccc37SChristine Caulfield static inline int nodeid_hash(int nodeid) 1775e9ccc37SChristine Caulfield { 1785e9ccc37SChristine Caulfield return nodeid & (CONN_HASH_SIZE-1); 1795e9ccc37SChristine Caulfield } 1805e9ccc37SChristine Caulfield 1815e9ccc37SChristine Caulfield static struct connection *__find_con(int nodeid) 1825e9ccc37SChristine Caulfield { 1835e9ccc37SChristine Caulfield int r; 1845e9ccc37SChristine Caulfield struct connection *con; 1855e9ccc37SChristine Caulfield 1865e9ccc37SChristine Caulfield r = nodeid_hash(nodeid); 1875e9ccc37SChristine Caulfield 188b67bfe0dSSasha Levin hlist_for_each_entry(con, &connection_hash[r], list) { 1895e9ccc37SChristine Caulfield if (con->nodeid == nodeid) 1905e9ccc37SChristine Caulfield return con; 1915e9ccc37SChristine Caulfield } 1925e9ccc37SChristine Caulfield return NULL; 1935e9ccc37SChristine Caulfield } 1945e9ccc37SChristine Caulfield 1956ed7257bSPatrick Caulfield /* 1966ed7257bSPatrick Caulfield * If 'allocation' is zero then we don't attempt to create a new 1976ed7257bSPatrick Caulfield * connection structure for this node. 1986ed7257bSPatrick Caulfield */ 1996ed7257bSPatrick Caulfield static struct connection *__nodeid2con(int nodeid, gfp_t alloc) 2006ed7257bSPatrick Caulfield { 2016ed7257bSPatrick Caulfield struct connection *con = NULL; 2026ed7257bSPatrick Caulfield int r; 2036ed7257bSPatrick Caulfield 2045e9ccc37SChristine Caulfield con = __find_con(nodeid); 2056ed7257bSPatrick Caulfield if (con || !alloc) 2066ed7257bSPatrick Caulfield return con; 2076ed7257bSPatrick Caulfield 2086ed7257bSPatrick Caulfield con = kmem_cache_zalloc(con_cache, alloc); 2096ed7257bSPatrick Caulfield if (!con) 2106ed7257bSPatrick Caulfield return NULL; 2116ed7257bSPatrick Caulfield 2125e9ccc37SChristine Caulfield r = nodeid_hash(nodeid); 2135e9ccc37SChristine Caulfield hlist_add_head(&con->list, &connection_hash[r]); 2146ed7257bSPatrick Caulfield 2156ed7257bSPatrick Caulfield con->nodeid = nodeid; 2166ed7257bSPatrick Caulfield mutex_init(&con->sock_mutex); 2176ed7257bSPatrick Caulfield INIT_LIST_HEAD(&con->writequeue); 2186ed7257bSPatrick Caulfield spin_lock_init(&con->writequeue_lock); 2196ed7257bSPatrick Caulfield INIT_WORK(&con->swork, process_send_sockets); 2206ed7257bSPatrick Caulfield INIT_WORK(&con->rwork, process_recv_sockets); 2216ed7257bSPatrick Caulfield 2226ed7257bSPatrick Caulfield /* Setup action pointers for child sockets */ 2236ed7257bSPatrick Caulfield if (con->nodeid) { 2245e9ccc37SChristine Caulfield struct connection *zerocon = __find_con(0); 2256ed7257bSPatrick Caulfield 2266ed7257bSPatrick Caulfield con->connect_action = zerocon->connect_action; 2276ed7257bSPatrick Caulfield if (!con->rx_action) 2286ed7257bSPatrick Caulfield con->rx_action = zerocon->rx_action; 2296ed7257bSPatrick Caulfield } 2306ed7257bSPatrick Caulfield 2316ed7257bSPatrick Caulfield return con; 2326ed7257bSPatrick Caulfield } 2336ed7257bSPatrick Caulfield 2345e9ccc37SChristine Caulfield /* Loop round all connections */ 2355e9ccc37SChristine Caulfield static void foreach_conn(void (*conn_func)(struct connection *c)) 2365e9ccc37SChristine Caulfield { 2375e9ccc37SChristine Caulfield int i; 238b67bfe0dSSasha Levin struct hlist_node *n; 2395e9ccc37SChristine Caulfield struct connection *con; 2405e9ccc37SChristine Caulfield 2415e9ccc37SChristine Caulfield for (i = 0; i < CONN_HASH_SIZE; i++) { 242b67bfe0dSSasha Levin hlist_for_each_entry_safe(con, n, &connection_hash[i], list) 2435e9ccc37SChristine Caulfield conn_func(con); 2445e9ccc37SChristine Caulfield } 2455e9ccc37SChristine Caulfield } 2465e9ccc37SChristine Caulfield 2476ed7257bSPatrick Caulfield static struct connection *nodeid2con(int nodeid, gfp_t allocation) 2486ed7257bSPatrick Caulfield { 2496ed7257bSPatrick Caulfield struct connection *con; 2506ed7257bSPatrick Caulfield 2517a936ce7SMatthias Kaehlcke mutex_lock(&connections_lock); 2526ed7257bSPatrick Caulfield con = __nodeid2con(nodeid, allocation); 2537a936ce7SMatthias Kaehlcke mutex_unlock(&connections_lock); 2546ed7257bSPatrick Caulfield 2556ed7257bSPatrick Caulfield return con; 2566ed7257bSPatrick Caulfield } 2576ed7257bSPatrick Caulfield 25836b71a8bSDavid Teigland static struct dlm_node_addr *find_node_addr(int nodeid) 2596ed7257bSPatrick Caulfield { 26036b71a8bSDavid Teigland struct dlm_node_addr *na; 26136b71a8bSDavid Teigland 26236b71a8bSDavid Teigland list_for_each_entry(na, &dlm_node_addrs, list) { 26336b71a8bSDavid Teigland if (na->nodeid == nodeid) 26436b71a8bSDavid Teigland return na; 26536b71a8bSDavid Teigland } 26636b71a8bSDavid Teigland return NULL; 26736b71a8bSDavid Teigland } 26836b71a8bSDavid Teigland 26936b71a8bSDavid Teigland static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y) 27036b71a8bSDavid Teigland { 27136b71a8bSDavid Teigland switch (x->ss_family) { 27236b71a8bSDavid Teigland case AF_INET: { 27336b71a8bSDavid Teigland struct sockaddr_in *sinx = (struct sockaddr_in *)x; 27436b71a8bSDavid Teigland struct sockaddr_in *siny = (struct sockaddr_in *)y; 27536b71a8bSDavid Teigland if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr) 27636b71a8bSDavid Teigland return 0; 27736b71a8bSDavid Teigland if (sinx->sin_port != siny->sin_port) 27836b71a8bSDavid Teigland return 0; 27936b71a8bSDavid Teigland break; 28036b71a8bSDavid Teigland } 28136b71a8bSDavid Teigland case AF_INET6: { 28236b71a8bSDavid Teigland struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x; 28336b71a8bSDavid Teigland struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y; 28436b71a8bSDavid Teigland if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr)) 28536b71a8bSDavid Teigland return 0; 28636b71a8bSDavid Teigland if (sinx->sin6_port != siny->sin6_port) 28736b71a8bSDavid Teigland return 0; 28836b71a8bSDavid Teigland break; 28936b71a8bSDavid Teigland } 29036b71a8bSDavid Teigland default: 29136b71a8bSDavid Teigland return 0; 29236b71a8bSDavid Teigland } 29336b71a8bSDavid Teigland return 1; 29436b71a8bSDavid Teigland } 29536b71a8bSDavid Teigland 29636b71a8bSDavid Teigland static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out, 29798e1b60eSMike Christie struct sockaddr *sa_out, bool try_new_addr) 29836b71a8bSDavid Teigland { 29936b71a8bSDavid Teigland struct sockaddr_storage sas; 30036b71a8bSDavid Teigland struct dlm_node_addr *na; 3016ed7257bSPatrick Caulfield 3026ed7257bSPatrick Caulfield if (!dlm_local_count) 3036ed7257bSPatrick Caulfield return -1; 3046ed7257bSPatrick Caulfield 30536b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 30636b71a8bSDavid Teigland na = find_node_addr(nodeid); 30798e1b60eSMike Christie if (na && na->addr_count) { 308ee44b4bcSMarcelo Ricardo Leitner memcpy(&sas, na->addr[na->curr_addr_index], 309ee44b4bcSMarcelo Ricardo Leitner sizeof(struct sockaddr_storage)); 310ee44b4bcSMarcelo Ricardo Leitner 31198e1b60eSMike Christie if (try_new_addr) { 31298e1b60eSMike Christie na->curr_addr_index++; 31398e1b60eSMike Christie if (na->curr_addr_index == na->addr_count) 31498e1b60eSMike Christie na->curr_addr_index = 0; 31598e1b60eSMike Christie } 31698e1b60eSMike Christie } 31736b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 31836b71a8bSDavid Teigland 31936b71a8bSDavid Teigland if (!na) 32036b71a8bSDavid Teigland return -EEXIST; 32136b71a8bSDavid Teigland 32236b71a8bSDavid Teigland if (!na->addr_count) 32336b71a8bSDavid Teigland return -ENOENT; 32436b71a8bSDavid Teigland 32536b71a8bSDavid Teigland if (sas_out) 32636b71a8bSDavid Teigland memcpy(sas_out, &sas, sizeof(struct sockaddr_storage)); 32736b71a8bSDavid Teigland 32836b71a8bSDavid Teigland if (!sa_out) 32936b71a8bSDavid Teigland return 0; 3306ed7257bSPatrick Caulfield 3316ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) { 33236b71a8bSDavid Teigland struct sockaddr_in *in4 = (struct sockaddr_in *) &sas; 33336b71a8bSDavid Teigland struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out; 3346ed7257bSPatrick Caulfield ret4->sin_addr.s_addr = in4->sin_addr.s_addr; 3356ed7257bSPatrick Caulfield } else { 33636b71a8bSDavid Teigland struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &sas; 33736b71a8bSDavid Teigland struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out; 3384e3fd7a0SAlexey Dobriyan ret6->sin6_addr = in6->sin6_addr; 3396ed7257bSPatrick Caulfield } 3406ed7257bSPatrick Caulfield 3416ed7257bSPatrick Caulfield return 0; 3426ed7257bSPatrick Caulfield } 3436ed7257bSPatrick Caulfield 34436b71a8bSDavid Teigland static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid) 34536b71a8bSDavid Teigland { 34636b71a8bSDavid Teigland struct dlm_node_addr *na; 34736b71a8bSDavid Teigland int rv = -EEXIST; 34898e1b60eSMike Christie int addr_i; 34936b71a8bSDavid Teigland 35036b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 35136b71a8bSDavid Teigland list_for_each_entry(na, &dlm_node_addrs, list) { 35236b71a8bSDavid Teigland if (!na->addr_count) 35336b71a8bSDavid Teigland continue; 35436b71a8bSDavid Teigland 35598e1b60eSMike Christie for (addr_i = 0; addr_i < na->addr_count; addr_i++) { 35698e1b60eSMike Christie if (addr_compare(na->addr[addr_i], addr)) { 35736b71a8bSDavid Teigland *nodeid = na->nodeid; 35836b71a8bSDavid Teigland rv = 0; 35998e1b60eSMike Christie goto unlock; 36036b71a8bSDavid Teigland } 36198e1b60eSMike Christie } 36298e1b60eSMike Christie } 36398e1b60eSMike Christie unlock: 36436b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 36536b71a8bSDavid Teigland return rv; 36636b71a8bSDavid Teigland } 36736b71a8bSDavid Teigland 36836b71a8bSDavid Teigland int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len) 36936b71a8bSDavid Teigland { 37036b71a8bSDavid Teigland struct sockaddr_storage *new_addr; 37136b71a8bSDavid Teigland struct dlm_node_addr *new_node, *na; 37236b71a8bSDavid Teigland 37336b71a8bSDavid Teigland new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS); 37436b71a8bSDavid Teigland if (!new_node) 37536b71a8bSDavid Teigland return -ENOMEM; 37636b71a8bSDavid Teigland 37736b71a8bSDavid Teigland new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS); 37836b71a8bSDavid Teigland if (!new_addr) { 37936b71a8bSDavid Teigland kfree(new_node); 38036b71a8bSDavid Teigland return -ENOMEM; 38136b71a8bSDavid Teigland } 38236b71a8bSDavid Teigland 38336b71a8bSDavid Teigland memcpy(new_addr, addr, len); 38436b71a8bSDavid Teigland 38536b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 38636b71a8bSDavid Teigland na = find_node_addr(nodeid); 38736b71a8bSDavid Teigland if (!na) { 38836b71a8bSDavid Teigland new_node->nodeid = nodeid; 38936b71a8bSDavid Teigland new_node->addr[0] = new_addr; 39036b71a8bSDavid Teigland new_node->addr_count = 1; 39136b71a8bSDavid Teigland list_add(&new_node->list, &dlm_node_addrs); 39236b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 39336b71a8bSDavid Teigland return 0; 39436b71a8bSDavid Teigland } 39536b71a8bSDavid Teigland 39636b71a8bSDavid Teigland if (na->addr_count >= DLM_MAX_ADDR_COUNT) { 39736b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 39836b71a8bSDavid Teigland kfree(new_addr); 39936b71a8bSDavid Teigland kfree(new_node); 40036b71a8bSDavid Teigland return -ENOSPC; 40136b71a8bSDavid Teigland } 40236b71a8bSDavid Teigland 40336b71a8bSDavid Teigland na->addr[na->addr_count++] = new_addr; 40436b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 40536b71a8bSDavid Teigland kfree(new_node); 40636b71a8bSDavid Teigland return 0; 40736b71a8bSDavid Teigland } 40836b71a8bSDavid Teigland 4096ed7257bSPatrick Caulfield /* Data available on socket or listen socket received a connect */ 410676d2369SDavid S. Miller static void lowcomms_data_ready(struct sock *sk) 4116ed7257bSPatrick Caulfield { 41293eaadebStsutomu.owa@toshiba.co.jp struct connection *con; 41393eaadebStsutomu.owa@toshiba.co.jp 41493eaadebStsutomu.owa@toshiba.co.jp read_lock_bh(&sk->sk_callback_lock); 41593eaadebStsutomu.owa@toshiba.co.jp con = sock2con(sk); 416afb853fbSPatrick Caulfield if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) 4176ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 41893eaadebStsutomu.owa@toshiba.co.jp read_unlock_bh(&sk->sk_callback_lock); 4196ed7257bSPatrick Caulfield } 4206ed7257bSPatrick Caulfield 4216ed7257bSPatrick Caulfield static void lowcomms_write_space(struct sock *sk) 4226ed7257bSPatrick Caulfield { 42393eaadebStsutomu.owa@toshiba.co.jp struct connection *con; 4246ed7257bSPatrick Caulfield 42593eaadebStsutomu.owa@toshiba.co.jp read_lock_bh(&sk->sk_callback_lock); 42693eaadebStsutomu.owa@toshiba.co.jp con = sock2con(sk); 427b36930ddSDavid Miller if (!con) 42893eaadebStsutomu.owa@toshiba.co.jp goto out; 429b36930ddSDavid Miller 430b36930ddSDavid Miller clear_bit(SOCK_NOSPACE, &con->sock->flags); 431b36930ddSDavid Miller 432b36930ddSDavid Miller if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) { 433b36930ddSDavid Miller con->sock->sk->sk_write_pending--; 4349cd3e072SEric Dumazet clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags); 435b36930ddSDavid Miller } 436b36930ddSDavid Miller 4376ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 43893eaadebStsutomu.owa@toshiba.co.jp out: 43993eaadebStsutomu.owa@toshiba.co.jp read_unlock_bh(&sk->sk_callback_lock); 4406ed7257bSPatrick Caulfield } 4416ed7257bSPatrick Caulfield 4426ed7257bSPatrick Caulfield static inline void lowcomms_connect_sock(struct connection *con) 4436ed7257bSPatrick Caulfield { 444063c4c99SLars Marowsky-Bree if (test_bit(CF_CLOSE, &con->flags)) 445063c4c99SLars Marowsky-Bree return; 4466ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 44761d9102bSBob Peterson cond_resched(); 4486ed7257bSPatrick Caulfield } 4496ed7257bSPatrick Caulfield 4506ed7257bSPatrick Caulfield static void lowcomms_state_change(struct sock *sk) 4516ed7257bSPatrick Caulfield { 452ee44b4bcSMarcelo Ricardo Leitner /* SCTP layer is not calling sk_data_ready when the connection 453ee44b4bcSMarcelo Ricardo Leitner * is done, so we catch the signal through here. Also, it 454ee44b4bcSMarcelo Ricardo Leitner * doesn't switch socket state when entering shutdown, so we 455ee44b4bcSMarcelo Ricardo Leitner * skip the write in that case. 456ee44b4bcSMarcelo Ricardo Leitner */ 457ee44b4bcSMarcelo Ricardo Leitner if (sk->sk_shutdown) { 458ee44b4bcSMarcelo Ricardo Leitner if (sk->sk_shutdown == RCV_SHUTDOWN) 459ee44b4bcSMarcelo Ricardo Leitner lowcomms_data_ready(sk); 460ee44b4bcSMarcelo Ricardo Leitner } else if (sk->sk_state == TCP_ESTABLISHED) { 4616ed7257bSPatrick Caulfield lowcomms_write_space(sk); 4626ed7257bSPatrick Caulfield } 463ee44b4bcSMarcelo Ricardo Leitner } 4646ed7257bSPatrick Caulfield 465391fbdc5SChristine Caulfield int dlm_lowcomms_connect_node(int nodeid) 466391fbdc5SChristine Caulfield { 467391fbdc5SChristine Caulfield struct connection *con; 468391fbdc5SChristine Caulfield 469391fbdc5SChristine Caulfield if (nodeid == dlm_our_nodeid()) 470391fbdc5SChristine Caulfield return 0; 471391fbdc5SChristine Caulfield 472391fbdc5SChristine Caulfield con = nodeid2con(nodeid, GFP_NOFS); 473391fbdc5SChristine Caulfield if (!con) 474391fbdc5SChristine Caulfield return -ENOMEM; 475391fbdc5SChristine Caulfield lowcomms_connect_sock(con); 476391fbdc5SChristine Caulfield return 0; 477391fbdc5SChristine Caulfield } 478391fbdc5SChristine Caulfield 479b3a5bbfdSBob Peterson static void lowcomms_error_report(struct sock *sk) 480b3a5bbfdSBob Peterson { 481b81171cbSBob Peterson struct connection *con; 482b3a5bbfdSBob Peterson struct sockaddr_storage saddr; 483b81171cbSBob Peterson void (*orig_report)(struct sock *) = NULL; 484b3a5bbfdSBob Peterson 485b81171cbSBob Peterson read_lock_bh(&sk->sk_callback_lock); 486b81171cbSBob Peterson con = sock2con(sk); 487b81171cbSBob Peterson if (con == NULL) 488b81171cbSBob Peterson goto out; 489b81171cbSBob Peterson 490cc661fc9SBob Peterson orig_report = listen_sock.sk_error_report; 4911a31833dSBob Peterson if (con->sock == NULL || 4929b2c45d4SDenys Vlasenko kernel_getpeername(con->sock, (struct sockaddr *)&saddr) < 0) { 493b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 494b3a5bbfdSBob Peterson "sending to node %d, port %d, " 495b3a5bbfdSBob Peterson "sk_err=%d/%d\n", dlm_our_nodeid(), 496b3a5bbfdSBob Peterson con->nodeid, dlm_config.ci_tcp_port, 497b3a5bbfdSBob Peterson sk->sk_err, sk->sk_err_soft); 498b3a5bbfdSBob Peterson } else if (saddr.ss_family == AF_INET) { 499b3a5bbfdSBob Peterson struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr; 500b3a5bbfdSBob Peterson 501b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 502b3a5bbfdSBob Peterson "sending to node %d at %pI4, port %d, " 503b3a5bbfdSBob Peterson "sk_err=%d/%d\n", dlm_our_nodeid(), 504b3a5bbfdSBob Peterson con->nodeid, &sin4->sin_addr.s_addr, 505b3a5bbfdSBob Peterson dlm_config.ci_tcp_port, sk->sk_err, 506b3a5bbfdSBob Peterson sk->sk_err_soft); 507b3a5bbfdSBob Peterson } else { 508b3a5bbfdSBob Peterson struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr; 509b3a5bbfdSBob Peterson 510b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 511b3a5bbfdSBob Peterson "sending to node %d at %u.%u.%u.%u, " 512b3a5bbfdSBob Peterson "port %d, sk_err=%d/%d\n", dlm_our_nodeid(), 513b3a5bbfdSBob Peterson con->nodeid, sin6->sin6_addr.s6_addr32[0], 514b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[1], 515b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[2], 516b3a5bbfdSBob Peterson sin6->sin6_addr.s6_addr32[3], 517b3a5bbfdSBob Peterson dlm_config.ci_tcp_port, sk->sk_err, 518b3a5bbfdSBob Peterson sk->sk_err_soft); 519b3a5bbfdSBob Peterson } 520b81171cbSBob Peterson out: 521b81171cbSBob Peterson read_unlock_bh(&sk->sk_callback_lock); 522b81171cbSBob Peterson if (orig_report) 523b81171cbSBob Peterson orig_report(sk); 524b81171cbSBob Peterson } 525b81171cbSBob Peterson 526b81171cbSBob Peterson /* Note: sk_callback_lock must be locked before calling this function. */ 527cc661fc9SBob Peterson static void save_listen_callbacks(struct socket *sock) 528b81171cbSBob Peterson { 529cc661fc9SBob Peterson struct sock *sk = sock->sk; 530cc661fc9SBob Peterson 531cc661fc9SBob Peterson listen_sock.sk_data_ready = sk->sk_data_ready; 532cc661fc9SBob Peterson listen_sock.sk_state_change = sk->sk_state_change; 533cc661fc9SBob Peterson listen_sock.sk_write_space = sk->sk_write_space; 534cc661fc9SBob Peterson listen_sock.sk_error_report = sk->sk_error_report; 535b81171cbSBob Peterson } 536b81171cbSBob Peterson 537cc661fc9SBob Peterson static void restore_callbacks(struct socket *sock) 538b81171cbSBob Peterson { 539cc661fc9SBob Peterson struct sock *sk = sock->sk; 540cc661fc9SBob Peterson 541b81171cbSBob Peterson write_lock_bh(&sk->sk_callback_lock); 542b81171cbSBob Peterson sk->sk_user_data = NULL; 543cc661fc9SBob Peterson sk->sk_data_ready = listen_sock.sk_data_ready; 544cc661fc9SBob Peterson sk->sk_state_change = listen_sock.sk_state_change; 545cc661fc9SBob Peterson sk->sk_write_space = listen_sock.sk_write_space; 546cc661fc9SBob Peterson sk->sk_error_report = listen_sock.sk_error_report; 547b81171cbSBob Peterson write_unlock_bh(&sk->sk_callback_lock); 548b3a5bbfdSBob Peterson } 549b3a5bbfdSBob Peterson 5506ed7257bSPatrick Caulfield /* Make a socket active */ 551988419a9Stsutomu.owa@toshiba.co.jp static void add_sock(struct socket *sock, struct connection *con) 5526ed7257bSPatrick Caulfield { 553b81171cbSBob Peterson struct sock *sk = sock->sk; 554b81171cbSBob Peterson 555b81171cbSBob Peterson write_lock_bh(&sk->sk_callback_lock); 5566ed7257bSPatrick Caulfield con->sock = sock; 5576ed7257bSPatrick Caulfield 558b81171cbSBob Peterson sk->sk_user_data = con; 5596ed7257bSPatrick Caulfield /* Install a data_ready callback */ 560b81171cbSBob Peterson sk->sk_data_ready = lowcomms_data_ready; 561b81171cbSBob Peterson sk->sk_write_space = lowcomms_write_space; 562b81171cbSBob Peterson sk->sk_state_change = lowcomms_state_change; 563b81171cbSBob Peterson sk->sk_allocation = GFP_NOFS; 564b81171cbSBob Peterson sk->sk_error_report = lowcomms_error_report; 565b81171cbSBob Peterson write_unlock_bh(&sk->sk_callback_lock); 5666ed7257bSPatrick Caulfield } 5676ed7257bSPatrick Caulfield 5686ed7257bSPatrick Caulfield /* Add the port number to an IPv6 or 4 sockaddr and return the address 5696ed7257bSPatrick Caulfield length */ 5706ed7257bSPatrick Caulfield static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, 5716ed7257bSPatrick Caulfield int *addr_len) 5726ed7257bSPatrick Caulfield { 5736ed7257bSPatrick Caulfield saddr->ss_family = dlm_local_addr[0]->ss_family; 5746ed7257bSPatrick Caulfield if (saddr->ss_family == AF_INET) { 5756ed7257bSPatrick Caulfield struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; 5766ed7257bSPatrick Caulfield in4_addr->sin_port = cpu_to_be16(port); 5776ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in); 5786ed7257bSPatrick Caulfield memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); 5796ed7257bSPatrick Caulfield } else { 5806ed7257bSPatrick Caulfield struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; 5816ed7257bSPatrick Caulfield in6_addr->sin6_port = cpu_to_be16(port); 5826ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in6); 5836ed7257bSPatrick Caulfield } 58401c8cab2SPatrick Caulfield memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); 5856ed7257bSPatrick Caulfield } 5866ed7257bSPatrick Caulfield 5876ed7257bSPatrick Caulfield /* Close a remote connection and tidy up */ 5880d737a8cSMarcelo Ricardo Leitner static void close_connection(struct connection *con, bool and_other, 5890d737a8cSMarcelo Ricardo Leitner bool tx, bool rx) 5906ed7257bSPatrick Caulfield { 591b2a66629Stsutomu.owa@toshiba.co.jp bool closing = test_and_set_bit(CF_CLOSING, &con->flags); 592b2a66629Stsutomu.owa@toshiba.co.jp 5930aa18464Stsutomu.owa@toshiba.co.jp if (tx && !closing && cancel_work_sync(&con->swork)) { 5940d737a8cSMarcelo Ricardo Leitner log_print("canceled swork for node %d", con->nodeid); 5950aa18464Stsutomu.owa@toshiba.co.jp clear_bit(CF_WRITE_PENDING, &con->flags); 5960aa18464Stsutomu.owa@toshiba.co.jp } 5970aa18464Stsutomu.owa@toshiba.co.jp if (rx && !closing && cancel_work_sync(&con->rwork)) { 5980d737a8cSMarcelo Ricardo Leitner log_print("canceled rwork for node %d", con->nodeid); 5990aa18464Stsutomu.owa@toshiba.co.jp clear_bit(CF_READ_PENDING, &con->flags); 6000aa18464Stsutomu.owa@toshiba.co.jp } 6016ed7257bSPatrick Caulfield 6020d737a8cSMarcelo Ricardo Leitner mutex_lock(&con->sock_mutex); 6036ed7257bSPatrick Caulfield if (con->sock) { 604cc661fc9SBob Peterson restore_callbacks(con->sock); 6056ed7257bSPatrick Caulfield sock_release(con->sock); 6066ed7257bSPatrick Caulfield con->sock = NULL; 6076ed7257bSPatrick Caulfield } 6086ed7257bSPatrick Caulfield if (con->othercon && and_other) { 6096ed7257bSPatrick Caulfield /* Will only re-enter once. */ 6100d737a8cSMarcelo Ricardo Leitner close_connection(con->othercon, false, true, true); 6116ed7257bSPatrick Caulfield } 6126ed7257bSPatrick Caulfield if (con->rx_page) { 6136ed7257bSPatrick Caulfield __free_page(con->rx_page); 6146ed7257bSPatrick Caulfield con->rx_page = NULL; 6156ed7257bSPatrick Caulfield } 6169e5f2825SPatrick Caulfield 6176ed7257bSPatrick Caulfield con->retries = 0; 6186ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 619b2a66629Stsutomu.owa@toshiba.co.jp clear_bit(CF_CLOSING, &con->flags); 6206ed7257bSPatrick Caulfield } 6216ed7257bSPatrick Caulfield 6226ed7257bSPatrick Caulfield /* Data received from remote end */ 6236ed7257bSPatrick Caulfield static int receive_from_sock(struct connection *con) 6246ed7257bSPatrick Caulfield { 6256ed7257bSPatrick Caulfield int ret = 0; 6266ed7257bSPatrick Caulfield struct msghdr msg = {}; 6276ed7257bSPatrick Caulfield struct kvec iov[2]; 6286ed7257bSPatrick Caulfield unsigned len; 6296ed7257bSPatrick Caulfield int r; 6306ed7257bSPatrick Caulfield int call_again_soon = 0; 6316ed7257bSPatrick Caulfield int nvec; 6326ed7257bSPatrick Caulfield 6336ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 6346ed7257bSPatrick Caulfield 6356ed7257bSPatrick Caulfield if (con->sock == NULL) { 6366ed7257bSPatrick Caulfield ret = -EAGAIN; 6376ed7257bSPatrick Caulfield goto out_close; 6386ed7257bSPatrick Caulfield } 639acee4e52SMarcelo Ricardo Leitner if (con->nodeid == 0) { 640acee4e52SMarcelo Ricardo Leitner ret = -EINVAL; 641acee4e52SMarcelo Ricardo Leitner goto out_close; 642acee4e52SMarcelo Ricardo Leitner } 6436ed7257bSPatrick Caulfield 6446ed7257bSPatrick Caulfield if (con->rx_page == NULL) { 6456ed7257bSPatrick Caulfield /* 6466ed7257bSPatrick Caulfield * This doesn't need to be atomic, but I think it should 6476ed7257bSPatrick Caulfield * improve performance if it is. 6486ed7257bSPatrick Caulfield */ 6496ed7257bSPatrick Caulfield con->rx_page = alloc_page(GFP_ATOMIC); 6506ed7257bSPatrick Caulfield if (con->rx_page == NULL) 6516ed7257bSPatrick Caulfield goto out_resched; 65209cbfeafSKirill A. Shutemov cbuf_init(&con->cb, PAGE_SIZE); 6536ed7257bSPatrick Caulfield } 6546ed7257bSPatrick Caulfield 6556ed7257bSPatrick Caulfield /* 6566ed7257bSPatrick Caulfield * iov[0] is the bit of the circular buffer between the current end 6576ed7257bSPatrick Caulfield * point (cb.base + cb.len) and the end of the buffer. 6586ed7257bSPatrick Caulfield */ 6596ed7257bSPatrick Caulfield iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); 6606ed7257bSPatrick Caulfield iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); 6616ed7257bSPatrick Caulfield iov[1].iov_len = 0; 6626ed7257bSPatrick Caulfield nvec = 1; 6636ed7257bSPatrick Caulfield 6646ed7257bSPatrick Caulfield /* 6656ed7257bSPatrick Caulfield * iov[1] is the bit of the circular buffer between the start of the 6666ed7257bSPatrick Caulfield * buffer and the start of the currently used section (cb.base) 6676ed7257bSPatrick Caulfield */ 6686ed7257bSPatrick Caulfield if (cbuf_data(&con->cb) >= con->cb.base) { 66909cbfeafSKirill A. Shutemov iov[0].iov_len = PAGE_SIZE - cbuf_data(&con->cb); 6706ed7257bSPatrick Caulfield iov[1].iov_len = con->cb.base; 6716ed7257bSPatrick Caulfield iov[1].iov_base = page_address(con->rx_page); 6726ed7257bSPatrick Caulfield nvec = 2; 6736ed7257bSPatrick Caulfield } 6746ed7257bSPatrick Caulfield len = iov[0].iov_len + iov[1].iov_len; 675aa563d7bSDavid Howells iov_iter_kvec(&msg.msg_iter, READ, iov, nvec, len); 6766ed7257bSPatrick Caulfield 677c8c7840eSAl Viro r = ret = sock_recvmsg(con->sock, &msg, MSG_DONTWAIT | MSG_NOSIGNAL); 6786ed7257bSPatrick Caulfield if (ret <= 0) 6796ed7257bSPatrick Caulfield goto out_close; 680ee44b4bcSMarcelo Ricardo Leitner else if (ret == len) 681ee44b4bcSMarcelo Ricardo Leitner call_again_soon = 1; 6826ed7257bSPatrick Caulfield 6836ed7257bSPatrick Caulfield cbuf_add(&con->cb, ret); 6846ed7257bSPatrick Caulfield ret = dlm_process_incoming_buffer(con->nodeid, 6856ed7257bSPatrick Caulfield page_address(con->rx_page), 6866ed7257bSPatrick Caulfield con->cb.base, con->cb.len, 68709cbfeafSKirill A. Shutemov PAGE_SIZE); 6886ed7257bSPatrick Caulfield if (ret == -EBADMSG) { 689ee44b4bcSMarcelo Ricardo Leitner log_print("lowcomms: addr=%p, base=%u, len=%u, read=%d", 690ee44b4bcSMarcelo Ricardo Leitner page_address(con->rx_page), con->cb.base, 691ee44b4bcSMarcelo Ricardo Leitner con->cb.len, r); 6926ed7257bSPatrick Caulfield } 6936ed7257bSPatrick Caulfield if (ret < 0) 6946ed7257bSPatrick Caulfield goto out_close; 6956ed7257bSPatrick Caulfield cbuf_eat(&con->cb, ret); 6966ed7257bSPatrick Caulfield 6976ed7257bSPatrick Caulfield if (cbuf_empty(&con->cb) && !call_again_soon) { 6986ed7257bSPatrick Caulfield __free_page(con->rx_page); 6996ed7257bSPatrick Caulfield con->rx_page = NULL; 7006ed7257bSPatrick Caulfield } 7016ed7257bSPatrick Caulfield 7026ed7257bSPatrick Caulfield if (call_again_soon) 7036ed7257bSPatrick Caulfield goto out_resched; 7046ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7056ed7257bSPatrick Caulfield return 0; 7066ed7257bSPatrick Caulfield 7076ed7257bSPatrick Caulfield out_resched: 7086ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) 7096ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 7106ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7116ed7257bSPatrick Caulfield return -EAGAIN; 7126ed7257bSPatrick Caulfield 7136ed7257bSPatrick Caulfield out_close: 7146ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7159e5f2825SPatrick Caulfield if (ret != -EAGAIN) { 716c553e173Stsutomu.owa@toshiba.co.jp close_connection(con, true, true, false); 7176ed7257bSPatrick Caulfield /* Reconnect when there is something to send */ 7186ed7257bSPatrick Caulfield } 7196ed7257bSPatrick Caulfield /* Don't return success if we really got EOF */ 7206ed7257bSPatrick Caulfield if (ret == 0) 7216ed7257bSPatrick Caulfield ret = -EAGAIN; 7226ed7257bSPatrick Caulfield 7236ed7257bSPatrick Caulfield return ret; 7246ed7257bSPatrick Caulfield } 7256ed7257bSPatrick Caulfield 7266ed7257bSPatrick Caulfield /* Listening socket is busy, accept a connection */ 7276ed7257bSPatrick Caulfield static int tcp_accept_from_sock(struct connection *con) 7286ed7257bSPatrick Caulfield { 7296ed7257bSPatrick Caulfield int result; 7306ed7257bSPatrick Caulfield struct sockaddr_storage peeraddr; 7316ed7257bSPatrick Caulfield struct socket *newsock; 7326ed7257bSPatrick Caulfield int len; 7336ed7257bSPatrick Caulfield int nodeid; 7346ed7257bSPatrick Caulfield struct connection *newcon; 7356ed7257bSPatrick Caulfield struct connection *addcon; 7366ed7257bSPatrick Caulfield 737513ef596SDavid Teigland mutex_lock(&connections_lock); 738513ef596SDavid Teigland if (!dlm_allow_conn) { 739513ef596SDavid Teigland mutex_unlock(&connections_lock); 740513ef596SDavid Teigland return -1; 741513ef596SDavid Teigland } 742513ef596SDavid Teigland mutex_unlock(&connections_lock); 743513ef596SDavid Teigland 7446ed7257bSPatrick Caulfield mutex_lock_nested(&con->sock_mutex, 0); 7456ed7257bSPatrick Caulfield 7463421fb15Stsutomu.owa@toshiba.co.jp if (!con->sock) { 7473421fb15Stsutomu.owa@toshiba.co.jp mutex_unlock(&con->sock_mutex); 7483421fb15Stsutomu.owa@toshiba.co.jp return -ENOTCONN; 7493421fb15Stsutomu.owa@toshiba.co.jp } 7506ed7257bSPatrick Caulfield 7513421fb15Stsutomu.owa@toshiba.co.jp result = kernel_accept(con->sock, &newsock, O_NONBLOCK); 7526ed7257bSPatrick Caulfield if (result < 0) 7536ed7257bSPatrick Caulfield goto accept_err; 7546ed7257bSPatrick Caulfield 7556ed7257bSPatrick Caulfield /* Get the connected socket's peer */ 7566ed7257bSPatrick Caulfield memset(&peeraddr, 0, sizeof(peeraddr)); 7579b2c45d4SDenys Vlasenko len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2); 7589b2c45d4SDenys Vlasenko if (len < 0) { 7596ed7257bSPatrick Caulfield result = -ECONNABORTED; 7606ed7257bSPatrick Caulfield goto accept_err; 7616ed7257bSPatrick Caulfield } 7626ed7257bSPatrick Caulfield 7636ed7257bSPatrick Caulfield /* Get the new node's NODEID */ 7646ed7257bSPatrick Caulfield make_sockaddr(&peeraddr, 0, &len); 76536b71a8bSDavid Teigland if (addr_to_nodeid(&peeraddr, &nodeid)) { 766bcaadf5cSMasatake YAMATO unsigned char *b=(unsigned char *)&peeraddr; 767617e82e1SDavid Teigland log_print("connect from non cluster node"); 768bcaadf5cSMasatake YAMATO print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, 769bcaadf5cSMasatake YAMATO b, sizeof(struct sockaddr_storage)); 7706ed7257bSPatrick Caulfield sock_release(newsock); 7716ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 7726ed7257bSPatrick Caulfield return -1; 7736ed7257bSPatrick Caulfield } 7746ed7257bSPatrick Caulfield 7756ed7257bSPatrick Caulfield log_print("got connection from %d", nodeid); 7766ed7257bSPatrick Caulfield 7776ed7257bSPatrick Caulfield /* Check to see if we already have a connection to this node. This 7786ed7257bSPatrick Caulfield * could happen if the two nodes initiate a connection at roughly 7796ed7257bSPatrick Caulfield * the same time and the connections cross on the wire. 7806ed7257bSPatrick Caulfield * In this case we store the incoming one in "othercon" 7816ed7257bSPatrick Caulfield */ 782748285ccSDavid Teigland newcon = nodeid2con(nodeid, GFP_NOFS); 7836ed7257bSPatrick Caulfield if (!newcon) { 7846ed7257bSPatrick Caulfield result = -ENOMEM; 7856ed7257bSPatrick Caulfield goto accept_err; 7866ed7257bSPatrick Caulfield } 7876ed7257bSPatrick Caulfield mutex_lock_nested(&newcon->sock_mutex, 1); 7886ed7257bSPatrick Caulfield if (newcon->sock) { 7896ed7257bSPatrick Caulfield struct connection *othercon = newcon->othercon; 7906ed7257bSPatrick Caulfield 7916ed7257bSPatrick Caulfield if (!othercon) { 792748285ccSDavid Teigland othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); 7936ed7257bSPatrick Caulfield if (!othercon) { 794617e82e1SDavid Teigland log_print("failed to allocate incoming socket"); 7956ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 7966ed7257bSPatrick Caulfield result = -ENOMEM; 7976ed7257bSPatrick Caulfield goto accept_err; 7986ed7257bSPatrick Caulfield } 7996ed7257bSPatrick Caulfield othercon->nodeid = nodeid; 8006ed7257bSPatrick Caulfield othercon->rx_action = receive_from_sock; 8016ed7257bSPatrick Caulfield mutex_init(&othercon->sock_mutex); 80226b41099Stsutomu.owa@toshiba.co.jp INIT_LIST_HEAD(&othercon->writequeue); 80326b41099Stsutomu.owa@toshiba.co.jp spin_lock_init(&othercon->writequeue_lock); 8046ed7257bSPatrick Caulfield INIT_WORK(&othercon->swork, process_send_sockets); 8056ed7257bSPatrick Caulfield INIT_WORK(&othercon->rwork, process_recv_sockets); 8066ed7257bSPatrick Caulfield set_bit(CF_IS_OTHERCON, &othercon->flags); 80761d96be0SPatrick Caulfield } 808c7355827Stsutomu.owa@toshiba.co.jp mutex_lock_nested(&othercon->sock_mutex, 2); 80961d96be0SPatrick Caulfield if (!othercon->sock) { 8106ed7257bSPatrick Caulfield newcon->othercon = othercon; 811988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, othercon); 8126ed7257bSPatrick Caulfield addcon = othercon; 813c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 8146ed7257bSPatrick Caulfield } 8156ed7257bSPatrick Caulfield else { 81697d84836SPatrick Caulfield printk("Extra connection from node %d attempted\n", nodeid); 81797d84836SPatrick Caulfield result = -EAGAIN; 818c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 819f4fadb23Sakpm@linux-foundation.org mutex_unlock(&newcon->sock_mutex); 82097d84836SPatrick Caulfield goto accept_err; 82197d84836SPatrick Caulfield } 82297d84836SPatrick Caulfield } 82397d84836SPatrick Caulfield else { 8246ed7257bSPatrick Caulfield newcon->rx_action = receive_from_sock; 8253735b4b9SBob Peterson /* accept copies the sk after we've saved the callbacks, so we 8263735b4b9SBob Peterson don't want to save them a second time or comm errors will 8273735b4b9SBob Peterson result in calling sk_error_report recursively. */ 828988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, newcon); 8296ed7257bSPatrick Caulfield addcon = newcon; 8306ed7257bSPatrick Caulfield } 8316ed7257bSPatrick Caulfield 8326ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 8336ed7257bSPatrick Caulfield 8346ed7257bSPatrick Caulfield /* 8356ed7257bSPatrick Caulfield * Add it to the active queue in case we got data 83625985edcSLucas De Marchi * between processing the accept adding the socket 8376ed7257bSPatrick Caulfield * to the read_sockets list 8386ed7257bSPatrick Caulfield */ 8396ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) 8406ed7257bSPatrick Caulfield queue_work(recv_workqueue, &addcon->rwork); 8416ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 8426ed7257bSPatrick Caulfield 8436ed7257bSPatrick Caulfield return 0; 8446ed7257bSPatrick Caulfield 8456ed7257bSPatrick Caulfield accept_err: 8466ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 8473421fb15Stsutomu.owa@toshiba.co.jp if (newsock) 8486ed7257bSPatrick Caulfield sock_release(newsock); 8496ed7257bSPatrick Caulfield 8506ed7257bSPatrick Caulfield if (result != -EAGAIN) 851617e82e1SDavid Teigland log_print("error accepting connection from node: %d", result); 8526ed7257bSPatrick Caulfield return result; 8536ed7257bSPatrick Caulfield } 8546ed7257bSPatrick Caulfield 85518df8a87Skbuild test robot static int sctp_accept_from_sock(struct connection *con) 856ee44b4bcSMarcelo Ricardo Leitner { 857ee44b4bcSMarcelo Ricardo Leitner /* Check that the new node is in the lockspace */ 858ee44b4bcSMarcelo Ricardo Leitner struct sctp_prim prim; 859ee44b4bcSMarcelo Ricardo Leitner int nodeid; 860ee44b4bcSMarcelo Ricardo Leitner int prim_len, ret; 861ee44b4bcSMarcelo Ricardo Leitner int addr_len; 862ee44b4bcSMarcelo Ricardo Leitner struct connection *newcon; 863ee44b4bcSMarcelo Ricardo Leitner struct connection *addcon; 864ee44b4bcSMarcelo Ricardo Leitner struct socket *newsock; 865ee44b4bcSMarcelo Ricardo Leitner 866ee44b4bcSMarcelo Ricardo Leitner mutex_lock(&connections_lock); 867ee44b4bcSMarcelo Ricardo Leitner if (!dlm_allow_conn) { 868ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&connections_lock); 869ee44b4bcSMarcelo Ricardo Leitner return -1; 870ee44b4bcSMarcelo Ricardo Leitner } 871ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&connections_lock); 872ee44b4bcSMarcelo Ricardo Leitner 873ee44b4bcSMarcelo Ricardo Leitner mutex_lock_nested(&con->sock_mutex, 0); 874ee44b4bcSMarcelo Ricardo Leitner 875ee44b4bcSMarcelo Ricardo Leitner ret = kernel_accept(con->sock, &newsock, O_NONBLOCK); 876ee44b4bcSMarcelo Ricardo Leitner if (ret < 0) 877ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 878ee44b4bcSMarcelo Ricardo Leitner 879ee44b4bcSMarcelo Ricardo Leitner memset(&prim, 0, sizeof(struct sctp_prim)); 880ee44b4bcSMarcelo Ricardo Leitner prim_len = sizeof(struct sctp_prim); 881ee44b4bcSMarcelo Ricardo Leitner 882ee44b4bcSMarcelo Ricardo Leitner ret = kernel_getsockopt(newsock, IPPROTO_SCTP, SCTP_PRIMARY_ADDR, 883ee44b4bcSMarcelo Ricardo Leitner (char *)&prim, &prim_len); 884ee44b4bcSMarcelo Ricardo Leitner if (ret < 0) { 885ee44b4bcSMarcelo Ricardo Leitner log_print("getsockopt/sctp_primary_addr failed: %d", ret); 886ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 887ee44b4bcSMarcelo Ricardo Leitner } 888ee44b4bcSMarcelo Ricardo Leitner 889ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&prim.ssp_addr, 0, &addr_len); 89026c1ec2fSWei Yongjun ret = addr_to_nodeid(&prim.ssp_addr, &nodeid); 89126c1ec2fSWei Yongjun if (ret) { 892ee44b4bcSMarcelo Ricardo Leitner unsigned char *b = (unsigned char *)&prim.ssp_addr; 893ee44b4bcSMarcelo Ricardo Leitner 894ee44b4bcSMarcelo Ricardo Leitner log_print("reject connect from unknown addr"); 895ee44b4bcSMarcelo Ricardo Leitner print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, 896ee44b4bcSMarcelo Ricardo Leitner b, sizeof(struct sockaddr_storage)); 897ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 898ee44b4bcSMarcelo Ricardo Leitner } 899ee44b4bcSMarcelo Ricardo Leitner 900ee44b4bcSMarcelo Ricardo Leitner newcon = nodeid2con(nodeid, GFP_NOFS); 901ee44b4bcSMarcelo Ricardo Leitner if (!newcon) { 902ee44b4bcSMarcelo Ricardo Leitner ret = -ENOMEM; 903ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 904ee44b4bcSMarcelo Ricardo Leitner } 905ee44b4bcSMarcelo Ricardo Leitner 906ee44b4bcSMarcelo Ricardo Leitner mutex_lock_nested(&newcon->sock_mutex, 1); 907ee44b4bcSMarcelo Ricardo Leitner 908ee44b4bcSMarcelo Ricardo Leitner if (newcon->sock) { 909ee44b4bcSMarcelo Ricardo Leitner struct connection *othercon = newcon->othercon; 910ee44b4bcSMarcelo Ricardo Leitner 911ee44b4bcSMarcelo Ricardo Leitner if (!othercon) { 912ee44b4bcSMarcelo Ricardo Leitner othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); 913ee44b4bcSMarcelo Ricardo Leitner if (!othercon) { 914ee44b4bcSMarcelo Ricardo Leitner log_print("failed to allocate incoming socket"); 915ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 916ee44b4bcSMarcelo Ricardo Leitner ret = -ENOMEM; 917ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 918ee44b4bcSMarcelo Ricardo Leitner } 919ee44b4bcSMarcelo Ricardo Leitner othercon->nodeid = nodeid; 920ee44b4bcSMarcelo Ricardo Leitner othercon->rx_action = receive_from_sock; 921ee44b4bcSMarcelo Ricardo Leitner mutex_init(&othercon->sock_mutex); 92226b41099Stsutomu.owa@toshiba.co.jp INIT_LIST_HEAD(&othercon->writequeue); 92326b41099Stsutomu.owa@toshiba.co.jp spin_lock_init(&othercon->writequeue_lock); 924ee44b4bcSMarcelo Ricardo Leitner INIT_WORK(&othercon->swork, process_send_sockets); 925ee44b4bcSMarcelo Ricardo Leitner INIT_WORK(&othercon->rwork, process_recv_sockets); 926ee44b4bcSMarcelo Ricardo Leitner set_bit(CF_IS_OTHERCON, &othercon->flags); 927ee44b4bcSMarcelo Ricardo Leitner } 928c7355827Stsutomu.owa@toshiba.co.jp mutex_lock_nested(&othercon->sock_mutex, 2); 929ee44b4bcSMarcelo Ricardo Leitner if (!othercon->sock) { 930ee44b4bcSMarcelo Ricardo Leitner newcon->othercon = othercon; 931988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, othercon); 932ee44b4bcSMarcelo Ricardo Leitner addcon = othercon; 933c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 934ee44b4bcSMarcelo Ricardo Leitner } else { 935ee44b4bcSMarcelo Ricardo Leitner printk("Extra connection from node %d attempted\n", nodeid); 936ee44b4bcSMarcelo Ricardo Leitner ret = -EAGAIN; 937c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 938ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 939ee44b4bcSMarcelo Ricardo Leitner goto accept_err; 940ee44b4bcSMarcelo Ricardo Leitner } 941ee44b4bcSMarcelo Ricardo Leitner } else { 942ee44b4bcSMarcelo Ricardo Leitner newcon->rx_action = receive_from_sock; 943988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, newcon); 944ee44b4bcSMarcelo Ricardo Leitner addcon = newcon; 945ee44b4bcSMarcelo Ricardo Leitner } 946ee44b4bcSMarcelo Ricardo Leitner 947ee44b4bcSMarcelo Ricardo Leitner log_print("connected to %d", nodeid); 948ee44b4bcSMarcelo Ricardo Leitner 949ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&newcon->sock_mutex); 950ee44b4bcSMarcelo Ricardo Leitner 951ee44b4bcSMarcelo Ricardo Leitner /* 952ee44b4bcSMarcelo Ricardo Leitner * Add it to the active queue in case we got data 953ee44b4bcSMarcelo Ricardo Leitner * between processing the accept adding the socket 954ee44b4bcSMarcelo Ricardo Leitner * to the read_sockets list 955ee44b4bcSMarcelo Ricardo Leitner */ 956ee44b4bcSMarcelo Ricardo Leitner if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) 957ee44b4bcSMarcelo Ricardo Leitner queue_work(recv_workqueue, &addcon->rwork); 958ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 959ee44b4bcSMarcelo Ricardo Leitner 960ee44b4bcSMarcelo Ricardo Leitner return 0; 961ee44b4bcSMarcelo Ricardo Leitner 962ee44b4bcSMarcelo Ricardo Leitner accept_err: 963ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 964ee44b4bcSMarcelo Ricardo Leitner if (newsock) 965ee44b4bcSMarcelo Ricardo Leitner sock_release(newsock); 966ee44b4bcSMarcelo Ricardo Leitner if (ret != -EAGAIN) 967ee44b4bcSMarcelo Ricardo Leitner log_print("error accepting connection from node: %d", ret); 968ee44b4bcSMarcelo Ricardo Leitner 969ee44b4bcSMarcelo Ricardo Leitner return ret; 970ee44b4bcSMarcelo Ricardo Leitner } 971ee44b4bcSMarcelo Ricardo Leitner 9726ed7257bSPatrick Caulfield static void free_entry(struct writequeue_entry *e) 9736ed7257bSPatrick Caulfield { 9746ed7257bSPatrick Caulfield __free_page(e->page); 9756ed7257bSPatrick Caulfield kfree(e); 9766ed7257bSPatrick Caulfield } 9776ed7257bSPatrick Caulfield 9785d689871SMike Christie /* 9795d689871SMike Christie * writequeue_entry_complete - try to delete and free write queue entry 9805d689871SMike Christie * @e: write queue entry to try to delete 9815d689871SMike Christie * @completed: bytes completed 9825d689871SMike Christie * 9835d689871SMike Christie * writequeue_lock must be held. 9845d689871SMike Christie */ 9855d689871SMike Christie static void writequeue_entry_complete(struct writequeue_entry *e, int completed) 9865d689871SMike Christie { 9875d689871SMike Christie e->offset += completed; 9885d689871SMike Christie e->len -= completed; 9895d689871SMike Christie 9905d689871SMike Christie if (e->len == 0 && e->users == 0) { 9915d689871SMike Christie list_del(&e->list); 9925d689871SMike Christie free_entry(e); 9935d689871SMike Christie } 9945d689871SMike Christie } 9955d689871SMike Christie 996ee44b4bcSMarcelo Ricardo Leitner /* 997ee44b4bcSMarcelo Ricardo Leitner * sctp_bind_addrs - bind a SCTP socket to all our addresses 998ee44b4bcSMarcelo Ricardo Leitner */ 999ee44b4bcSMarcelo Ricardo Leitner static int sctp_bind_addrs(struct connection *con, uint16_t port) 1000ee44b4bcSMarcelo Ricardo Leitner { 1001ee44b4bcSMarcelo Ricardo Leitner struct sockaddr_storage localaddr; 1002ee44b4bcSMarcelo Ricardo Leitner int i, addr_len, result = 0; 1003ee44b4bcSMarcelo Ricardo Leitner 1004ee44b4bcSMarcelo Ricardo Leitner for (i = 0; i < dlm_local_count; i++) { 1005ee44b4bcSMarcelo Ricardo Leitner memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); 1006ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&localaddr, port, &addr_len); 1007ee44b4bcSMarcelo Ricardo Leitner 1008ee44b4bcSMarcelo Ricardo Leitner if (!i) 1009ee44b4bcSMarcelo Ricardo Leitner result = kernel_bind(con->sock, 1010ee44b4bcSMarcelo Ricardo Leitner (struct sockaddr *)&localaddr, 1011ee44b4bcSMarcelo Ricardo Leitner addr_len); 1012ee44b4bcSMarcelo Ricardo Leitner else 1013ee44b4bcSMarcelo Ricardo Leitner result = kernel_setsockopt(con->sock, SOL_SCTP, 1014ee44b4bcSMarcelo Ricardo Leitner SCTP_SOCKOPT_BINDX_ADD, 1015ee44b4bcSMarcelo Ricardo Leitner (char *)&localaddr, addr_len); 1016ee44b4bcSMarcelo Ricardo Leitner 1017ee44b4bcSMarcelo Ricardo Leitner if (result < 0) { 1018ee44b4bcSMarcelo Ricardo Leitner log_print("Can't bind to %d addr number %d, %d.\n", 1019ee44b4bcSMarcelo Ricardo Leitner port, i + 1, result); 1020ee44b4bcSMarcelo Ricardo Leitner break; 1021ee44b4bcSMarcelo Ricardo Leitner } 1022ee44b4bcSMarcelo Ricardo Leitner } 1023ee44b4bcSMarcelo Ricardo Leitner return result; 1024ee44b4bcSMarcelo Ricardo Leitner } 1025ee44b4bcSMarcelo Ricardo Leitner 10266ed7257bSPatrick Caulfield /* Initiate an SCTP association. 10276ed7257bSPatrick Caulfield This is a special case of send_to_sock() in that we don't yet have a 10286ed7257bSPatrick Caulfield peeled-off socket for this association, so we use the listening socket 10296ed7257bSPatrick Caulfield and add the primary IP address of the remote node. 10306ed7257bSPatrick Caulfield */ 1031ee44b4bcSMarcelo Ricardo Leitner static void sctp_connect_to_sock(struct connection *con) 10326ed7257bSPatrick Caulfield { 1033ee44b4bcSMarcelo Ricardo Leitner struct sockaddr_storage daddr; 1034ee44b4bcSMarcelo Ricardo Leitner int one = 1; 1035ee44b4bcSMarcelo Ricardo Leitner int result; 1036ee44b4bcSMarcelo Ricardo Leitner int addr_len; 1037ee44b4bcSMarcelo Ricardo Leitner struct socket *sock; 1038f706d830SGang He struct timeval tv = { .tv_sec = 5, .tv_usec = 0 }; 1039ee44b4bcSMarcelo Ricardo Leitner 1040ee44b4bcSMarcelo Ricardo Leitner if (con->nodeid == 0) { 1041ee44b4bcSMarcelo Ricardo Leitner log_print("attempt to connect sock 0 foiled"); 1042ee44b4bcSMarcelo Ricardo Leitner return; 1043ee44b4bcSMarcelo Ricardo Leitner } 10446ed7257bSPatrick Caulfield 10455d689871SMike Christie mutex_lock(&con->sock_mutex); 10466ed7257bSPatrick Caulfield 1047ee44b4bcSMarcelo Ricardo Leitner /* Some odd races can cause double-connects, ignore them */ 1048ee44b4bcSMarcelo Ricardo Leitner if (con->retries++ > MAX_CONNECT_RETRIES) 1049ee44b4bcSMarcelo Ricardo Leitner goto out; 1050ee44b4bcSMarcelo Ricardo Leitner 1051ee44b4bcSMarcelo Ricardo Leitner if (con->sock) { 1052ee44b4bcSMarcelo Ricardo Leitner log_print("node %d already connected.", con->nodeid); 1053ee44b4bcSMarcelo Ricardo Leitner goto out; 1054ee44b4bcSMarcelo Ricardo Leitner } 1055ee44b4bcSMarcelo Ricardo Leitner 1056ee44b4bcSMarcelo Ricardo Leitner memset(&daddr, 0, sizeof(daddr)); 1057ee44b4bcSMarcelo Ricardo Leitner result = nodeid_to_addr(con->nodeid, &daddr, NULL, true); 1058ee44b4bcSMarcelo Ricardo Leitner if (result < 0) { 10596ed7257bSPatrick Caulfield log_print("no address for nodeid %d", con->nodeid); 1060ee44b4bcSMarcelo Ricardo Leitner goto out; 106104bedd79SDavid Teigland } 10626ed7257bSPatrick Caulfield 1063ee44b4bcSMarcelo Ricardo Leitner /* Create a socket to communicate with */ 1064ee44b4bcSMarcelo Ricardo Leitner result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1065ee44b4bcSMarcelo Ricardo Leitner SOCK_STREAM, IPPROTO_SCTP, &sock); 1066ee44b4bcSMarcelo Ricardo Leitner if (result < 0) 1067ee44b4bcSMarcelo Ricardo Leitner goto socket_err; 10686ed7257bSPatrick Caulfield 1069ee44b4bcSMarcelo Ricardo Leitner con->rx_action = receive_from_sock; 1070ee44b4bcSMarcelo Ricardo Leitner con->connect_action = sctp_connect_to_sock; 1071988419a9Stsutomu.owa@toshiba.co.jp add_sock(sock, con); 10726ed7257bSPatrick Caulfield 1073ee44b4bcSMarcelo Ricardo Leitner /* Bind to all addresses. */ 1074ee44b4bcSMarcelo Ricardo Leitner if (sctp_bind_addrs(con, 0)) 1075ee44b4bcSMarcelo Ricardo Leitner goto bind_err; 107698e1b60eSMike Christie 1077ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len); 10786ed7257bSPatrick Caulfield 1079ee44b4bcSMarcelo Ricardo Leitner log_print("connecting to %d", con->nodeid); 10806ed7257bSPatrick Caulfield 1081ee44b4bcSMarcelo Ricardo Leitner /* Turn off Nagle's algorithm */ 1082b09c603cSGang He kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one, 1083ee44b4bcSMarcelo Ricardo Leitner sizeof(one)); 1084ee44b4bcSMarcelo Ricardo Leitner 1085f706d830SGang He /* 1086f706d830SGang He * Make sock->ops->connect() function return in specified time, 1087f706d830SGang He * since O_NONBLOCK argument in connect() function does not work here, 1088f706d830SGang He * then, we should restore the default value of this attribute. 1089f706d830SGang He */ 109045bdc661SDeepa Dinamani kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO_OLD, (char *)&tv, 1091f706d830SGang He sizeof(tv)); 1092ee44b4bcSMarcelo Ricardo Leitner result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len, 1093da3627c3SGang He 0); 1094f706d830SGang He memset(&tv, 0, sizeof(tv)); 109545bdc661SDeepa Dinamani kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO_OLD, (char *)&tv, 1096f706d830SGang He sizeof(tv)); 1097f706d830SGang He 1098ee44b4bcSMarcelo Ricardo Leitner if (result == -EINPROGRESS) 1099ee44b4bcSMarcelo Ricardo Leitner result = 0; 1100ee44b4bcSMarcelo Ricardo Leitner if (result == 0) 1101ee44b4bcSMarcelo Ricardo Leitner goto out; 1102ee44b4bcSMarcelo Ricardo Leitner 1103ee44b4bcSMarcelo Ricardo Leitner bind_err: 1104ee44b4bcSMarcelo Ricardo Leitner con->sock = NULL; 1105ee44b4bcSMarcelo Ricardo Leitner sock_release(sock); 1106ee44b4bcSMarcelo Ricardo Leitner 1107ee44b4bcSMarcelo Ricardo Leitner socket_err: 1108ee44b4bcSMarcelo Ricardo Leitner /* 1109ee44b4bcSMarcelo Ricardo Leitner * Some errors are fatal and this list might need adjusting. For other 1110ee44b4bcSMarcelo Ricardo Leitner * errors we try again until the max number of retries is reached. 1111ee44b4bcSMarcelo Ricardo Leitner */ 1112ee44b4bcSMarcelo Ricardo Leitner if (result != -EHOSTUNREACH && 1113ee44b4bcSMarcelo Ricardo Leitner result != -ENETUNREACH && 1114ee44b4bcSMarcelo Ricardo Leitner result != -ENETDOWN && 1115ee44b4bcSMarcelo Ricardo Leitner result != -EINVAL && 1116ee44b4bcSMarcelo Ricardo Leitner result != -EPROTONOSUPPORT) { 1117ee44b4bcSMarcelo Ricardo Leitner log_print("connect %d try %d error %d", con->nodeid, 1118ee44b4bcSMarcelo Ricardo Leitner con->retries, result); 1119ee44b4bcSMarcelo Ricardo Leitner mutex_unlock(&con->sock_mutex); 1120ee44b4bcSMarcelo Ricardo Leitner msleep(1000); 1121ee44b4bcSMarcelo Ricardo Leitner lowcomms_connect_sock(con); 1122ee44b4bcSMarcelo Ricardo Leitner return; 11236ed7257bSPatrick Caulfield } 11245d689871SMike Christie 1125ee44b4bcSMarcelo Ricardo Leitner out: 11265d689871SMike Christie mutex_unlock(&con->sock_mutex); 11276ed7257bSPatrick Caulfield } 11286ed7257bSPatrick Caulfield 11296ed7257bSPatrick Caulfield /* Connect a new socket to its peer */ 11306ed7257bSPatrick Caulfield static void tcp_connect_to_sock(struct connection *con) 11316ed7257bSPatrick Caulfield { 11326bd8fedaSLon Hohberger struct sockaddr_storage saddr, src_addr; 11336ed7257bSPatrick Caulfield int addr_len; 1134a89d63a1SCasey Dahlin struct socket *sock = NULL; 1135cb2d45daSDavid Teigland int one = 1; 113636b71a8bSDavid Teigland int result; 11376ed7257bSPatrick Caulfield 11386ed7257bSPatrick Caulfield if (con->nodeid == 0) { 11396ed7257bSPatrick Caulfield log_print("attempt to connect sock 0 foiled"); 11406ed7257bSPatrick Caulfield return; 11416ed7257bSPatrick Caulfield } 11426ed7257bSPatrick Caulfield 11436ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 11446ed7257bSPatrick Caulfield if (con->retries++ > MAX_CONNECT_RETRIES) 11456ed7257bSPatrick Caulfield goto out; 11466ed7257bSPatrick Caulfield 11476ed7257bSPatrick Caulfield /* Some odd races can cause double-connects, ignore them */ 114836b71a8bSDavid Teigland if (con->sock) 11496ed7257bSPatrick Caulfield goto out; 11506ed7257bSPatrick Caulfield 11516ed7257bSPatrick Caulfield /* Create a socket to communicate with */ 1152eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1153eeb1bd5cSEric W. Biederman SOCK_STREAM, IPPROTO_TCP, &sock); 11546ed7257bSPatrick Caulfield if (result < 0) 11556ed7257bSPatrick Caulfield goto out_err; 11566ed7257bSPatrick Caulfield 11576ed7257bSPatrick Caulfield memset(&saddr, 0, sizeof(saddr)); 115898e1b60eSMike Christie result = nodeid_to_addr(con->nodeid, &saddr, NULL, false); 115936b71a8bSDavid Teigland if (result < 0) { 116036b71a8bSDavid Teigland log_print("no address for nodeid %d", con->nodeid); 11616ed7257bSPatrick Caulfield goto out_err; 116236b71a8bSDavid Teigland } 11636ed7257bSPatrick Caulfield 11646ed7257bSPatrick Caulfield con->rx_action = receive_from_sock; 11656ed7257bSPatrick Caulfield con->connect_action = tcp_connect_to_sock; 1166988419a9Stsutomu.owa@toshiba.co.jp add_sock(sock, con); 11676ed7257bSPatrick Caulfield 11686bd8fedaSLon Hohberger /* Bind to our cluster-known address connecting to avoid 11696bd8fedaSLon Hohberger routing problems */ 11706bd8fedaSLon Hohberger memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); 11716bd8fedaSLon Hohberger make_sockaddr(&src_addr, 0, &addr_len); 11726bd8fedaSLon Hohberger result = sock->ops->bind(sock, (struct sockaddr *) &src_addr, 11736bd8fedaSLon Hohberger addr_len); 11746bd8fedaSLon Hohberger if (result < 0) { 11756bd8fedaSLon Hohberger log_print("could not bind for connect: %d", result); 11766bd8fedaSLon Hohberger /* This *may* not indicate a critical error */ 11776bd8fedaSLon Hohberger } 11786bd8fedaSLon Hohberger 11796ed7257bSPatrick Caulfield make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); 11806ed7257bSPatrick Caulfield 11816ed7257bSPatrick Caulfield log_print("connecting to %d", con->nodeid); 1182cb2d45daSDavid Teigland 1183cb2d45daSDavid Teigland /* Turn off Nagle's algorithm */ 1184cb2d45daSDavid Teigland kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, 1185cb2d45daSDavid Teigland sizeof(one)); 1186cb2d45daSDavid Teigland 118736b71a8bSDavid Teigland result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, 11886ed7257bSPatrick Caulfield O_NONBLOCK); 11896ed7257bSPatrick Caulfield if (result == -EINPROGRESS) 11906ed7257bSPatrick Caulfield result = 0; 11916ed7257bSPatrick Caulfield if (result == 0) 11926ed7257bSPatrick Caulfield goto out; 11936ed7257bSPatrick Caulfield 11946ed7257bSPatrick Caulfield out_err: 11956ed7257bSPatrick Caulfield if (con->sock) { 11966ed7257bSPatrick Caulfield sock_release(con->sock); 11976ed7257bSPatrick Caulfield con->sock = NULL; 1198a89d63a1SCasey Dahlin } else if (sock) { 1199a89d63a1SCasey Dahlin sock_release(sock); 12006ed7257bSPatrick Caulfield } 12016ed7257bSPatrick Caulfield /* 12026ed7257bSPatrick Caulfield * Some errors are fatal and this list might need adjusting. For other 12036ed7257bSPatrick Caulfield * errors we try again until the max number of retries is reached. 12046ed7257bSPatrick Caulfield */ 120536b71a8bSDavid Teigland if (result != -EHOSTUNREACH && 120636b71a8bSDavid Teigland result != -ENETUNREACH && 120736b71a8bSDavid Teigland result != -ENETDOWN && 120836b71a8bSDavid Teigland result != -EINVAL && 120936b71a8bSDavid Teigland result != -EPROTONOSUPPORT) { 121036b71a8bSDavid Teigland log_print("connect %d try %d error %d", con->nodeid, 121136b71a8bSDavid Teigland con->retries, result); 121236b71a8bSDavid Teigland mutex_unlock(&con->sock_mutex); 121336b71a8bSDavid Teigland msleep(1000); 12146ed7257bSPatrick Caulfield lowcomms_connect_sock(con); 121536b71a8bSDavid Teigland return; 12166ed7257bSPatrick Caulfield } 12176ed7257bSPatrick Caulfield out: 12186ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 12196ed7257bSPatrick Caulfield return; 12206ed7257bSPatrick Caulfield } 12216ed7257bSPatrick Caulfield 12226ed7257bSPatrick Caulfield static struct socket *tcp_create_listen_sock(struct connection *con, 12236ed7257bSPatrick Caulfield struct sockaddr_storage *saddr) 12246ed7257bSPatrick Caulfield { 12256ed7257bSPatrick Caulfield struct socket *sock = NULL; 12266ed7257bSPatrick Caulfield int result = 0; 12276ed7257bSPatrick Caulfield int one = 1; 12286ed7257bSPatrick Caulfield int addr_len; 12296ed7257bSPatrick Caulfield 12306ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) 12316ed7257bSPatrick Caulfield addr_len = sizeof(struct sockaddr_in); 12326ed7257bSPatrick Caulfield else 12336ed7257bSPatrick Caulfield addr_len = sizeof(struct sockaddr_in6); 12346ed7257bSPatrick Caulfield 12356ed7257bSPatrick Caulfield /* Create a socket to communicate with */ 1236eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1237eeb1bd5cSEric W. Biederman SOCK_STREAM, IPPROTO_TCP, &sock); 12386ed7257bSPatrick Caulfield if (result < 0) { 1239617e82e1SDavid Teigland log_print("Can't create listening comms socket"); 12406ed7257bSPatrick Caulfield goto create_out; 12416ed7257bSPatrick Caulfield } 12426ed7257bSPatrick Caulfield 1243cb2d45daSDavid Teigland /* Turn off Nagle's algorithm */ 1244cb2d45daSDavid Teigland kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, 1245cb2d45daSDavid Teigland sizeof(one)); 1246cb2d45daSDavid Teigland 12476ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, 12486ed7257bSPatrick Caulfield (char *)&one, sizeof(one)); 12496ed7257bSPatrick Caulfield 12506ed7257bSPatrick Caulfield if (result < 0) { 1251617e82e1SDavid Teigland log_print("Failed to set SO_REUSEADDR on socket: %d", result); 12526ed7257bSPatrick Caulfield } 125393eaadebStsutomu.owa@toshiba.co.jp write_lock_bh(&sock->sk->sk_callback_lock); 1254b81171cbSBob Peterson sock->sk->sk_user_data = con; 1255cc661fc9SBob Peterson save_listen_callbacks(sock); 12566ed7257bSPatrick Caulfield con->rx_action = tcp_accept_from_sock; 12576ed7257bSPatrick Caulfield con->connect_action = tcp_connect_to_sock; 125893eaadebStsutomu.owa@toshiba.co.jp write_unlock_bh(&sock->sk->sk_callback_lock); 12596ed7257bSPatrick Caulfield 12606ed7257bSPatrick Caulfield /* Bind to our port */ 12616ed7257bSPatrick Caulfield make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); 12626ed7257bSPatrick Caulfield result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); 12636ed7257bSPatrick Caulfield if (result < 0) { 1264617e82e1SDavid Teigland log_print("Can't bind to port %d", dlm_config.ci_tcp_port); 12656ed7257bSPatrick Caulfield sock_release(sock); 12666ed7257bSPatrick Caulfield sock = NULL; 12676ed7257bSPatrick Caulfield con->sock = NULL; 12686ed7257bSPatrick Caulfield goto create_out; 12696ed7257bSPatrick Caulfield } 12706ed7257bSPatrick Caulfield result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, 12716ed7257bSPatrick Caulfield (char *)&one, sizeof(one)); 12726ed7257bSPatrick Caulfield if (result < 0) { 1273617e82e1SDavid Teigland log_print("Set keepalive failed: %d", result); 12746ed7257bSPatrick Caulfield } 12756ed7257bSPatrick Caulfield 12766ed7257bSPatrick Caulfield result = sock->ops->listen(sock, 5); 12776ed7257bSPatrick Caulfield if (result < 0) { 1278617e82e1SDavid Teigland log_print("Can't listen on port %d", dlm_config.ci_tcp_port); 12796ed7257bSPatrick Caulfield sock_release(sock); 12806ed7257bSPatrick Caulfield sock = NULL; 12816ed7257bSPatrick Caulfield goto create_out; 12826ed7257bSPatrick Caulfield } 12836ed7257bSPatrick Caulfield 12846ed7257bSPatrick Caulfield create_out: 12856ed7257bSPatrick Caulfield return sock; 12866ed7257bSPatrick Caulfield } 12876ed7257bSPatrick Caulfield 12886ed7257bSPatrick Caulfield /* Get local addresses */ 12896ed7257bSPatrick Caulfield static void init_local(void) 12906ed7257bSPatrick Caulfield { 12916ed7257bSPatrick Caulfield struct sockaddr_storage sas, *addr; 12926ed7257bSPatrick Caulfield int i; 12936ed7257bSPatrick Caulfield 129430d3a237SPatrick Caulfield dlm_local_count = 0; 12951b189b88SDavid Teigland for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) { 12966ed7257bSPatrick Caulfield if (dlm_our_addr(&sas, i)) 12976ed7257bSPatrick Caulfield break; 12986ed7257bSPatrick Caulfield 12995c93f56fSAmitoj Kaur Chawla addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS); 13006ed7257bSPatrick Caulfield if (!addr) 13016ed7257bSPatrick Caulfield break; 13026ed7257bSPatrick Caulfield dlm_local_addr[dlm_local_count++] = addr; 13036ed7257bSPatrick Caulfield } 13046ed7257bSPatrick Caulfield } 13056ed7257bSPatrick Caulfield 13066ed7257bSPatrick Caulfield /* Initialise SCTP socket and bind to all interfaces */ 13076ed7257bSPatrick Caulfield static int sctp_listen_for_all(void) 13086ed7257bSPatrick Caulfield { 13096ed7257bSPatrick Caulfield struct socket *sock = NULL; 1310ee44b4bcSMarcelo Ricardo Leitner int result = -EINVAL; 1311573c24c4SDavid Teigland struct connection *con = nodeid2con(0, GFP_NOFS); 13126ed7257bSPatrick Caulfield int bufsize = NEEDED_RMEM; 131386e92ad2SMike Christie int one = 1; 13146ed7257bSPatrick Caulfield 13156ed7257bSPatrick Caulfield if (!con) 13166ed7257bSPatrick Caulfield return -ENOMEM; 13176ed7257bSPatrick Caulfield 13186ed7257bSPatrick Caulfield log_print("Using SCTP for communications"); 13196ed7257bSPatrick Caulfield 1320eeb1bd5cSEric W. Biederman result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1321ee44b4bcSMarcelo Ricardo Leitner SOCK_STREAM, IPPROTO_SCTP, &sock); 13226ed7257bSPatrick Caulfield if (result < 0) { 13236ed7257bSPatrick Caulfield log_print("Can't create comms socket, check SCTP is loaded"); 13246ed7257bSPatrick Caulfield goto out; 13256ed7257bSPatrick Caulfield } 13266ed7257bSPatrick Caulfield 1327df61c952SDavid S. Miller result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, 13286ed7257bSPatrick Caulfield (char *)&bufsize, sizeof(bufsize)); 13296ed7257bSPatrick Caulfield if (result) 1330617e82e1SDavid Teigland log_print("Error increasing buffer space on socket %d", result); 13316ed7257bSPatrick Caulfield 133286e92ad2SMike Christie result = kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one, 133386e92ad2SMike Christie sizeof(one)); 133486e92ad2SMike Christie if (result < 0) 133586e92ad2SMike Christie log_print("Could not set SCTP NODELAY error %d\n", result); 133686e92ad2SMike Christie 1337b81171cbSBob Peterson write_lock_bh(&sock->sk->sk_callback_lock); 13386ed7257bSPatrick Caulfield /* Init con struct */ 13396ed7257bSPatrick Caulfield sock->sk->sk_user_data = con; 1340cc661fc9SBob Peterson save_listen_callbacks(sock); 13416ed7257bSPatrick Caulfield con->sock = sock; 13426ed7257bSPatrick Caulfield con->sock->sk->sk_data_ready = lowcomms_data_ready; 1343ee44b4bcSMarcelo Ricardo Leitner con->rx_action = sctp_accept_from_sock; 1344ee44b4bcSMarcelo Ricardo Leitner con->connect_action = sctp_connect_to_sock; 13456ed7257bSPatrick Caulfield 1346b81171cbSBob Peterson write_unlock_bh(&sock->sk->sk_callback_lock); 1347b81171cbSBob Peterson 1348ee44b4bcSMarcelo Ricardo Leitner /* Bind to all addresses. */ 1349ee44b4bcSMarcelo Ricardo Leitner if (sctp_bind_addrs(con, dlm_config.ci_tcp_port)) 13506ed7257bSPatrick Caulfield goto create_delsock; 13516ed7257bSPatrick Caulfield 13526ed7257bSPatrick Caulfield result = sock->ops->listen(sock, 5); 13536ed7257bSPatrick Caulfield if (result < 0) { 13546ed7257bSPatrick Caulfield log_print("Can't set socket listening"); 13556ed7257bSPatrick Caulfield goto create_delsock; 13566ed7257bSPatrick Caulfield } 13576ed7257bSPatrick Caulfield 13586ed7257bSPatrick Caulfield return 0; 13596ed7257bSPatrick Caulfield 13606ed7257bSPatrick Caulfield create_delsock: 13616ed7257bSPatrick Caulfield sock_release(sock); 13626ed7257bSPatrick Caulfield con->sock = NULL; 13636ed7257bSPatrick Caulfield out: 13646ed7257bSPatrick Caulfield return result; 13656ed7257bSPatrick Caulfield } 13666ed7257bSPatrick Caulfield 13676ed7257bSPatrick Caulfield static int tcp_listen_for_all(void) 13686ed7257bSPatrick Caulfield { 13696ed7257bSPatrick Caulfield struct socket *sock = NULL; 1370573c24c4SDavid Teigland struct connection *con = nodeid2con(0, GFP_NOFS); 13716ed7257bSPatrick Caulfield int result = -EINVAL; 13726ed7257bSPatrick Caulfield 13736ed7257bSPatrick Caulfield if (!con) 13746ed7257bSPatrick Caulfield return -ENOMEM; 13756ed7257bSPatrick Caulfield 13766ed7257bSPatrick Caulfield /* We don't support multi-homed hosts */ 13776ed7257bSPatrick Caulfield if (dlm_local_addr[1] != NULL) { 1378617e82e1SDavid Teigland log_print("TCP protocol can't handle multi-homed hosts, " 1379617e82e1SDavid Teigland "try SCTP"); 13806ed7257bSPatrick Caulfield return -EINVAL; 13816ed7257bSPatrick Caulfield } 13826ed7257bSPatrick Caulfield 13836ed7257bSPatrick Caulfield log_print("Using TCP for communications"); 13846ed7257bSPatrick Caulfield 13856ed7257bSPatrick Caulfield sock = tcp_create_listen_sock(con, dlm_local_addr[0]); 13866ed7257bSPatrick Caulfield if (sock) { 1387988419a9Stsutomu.owa@toshiba.co.jp add_sock(sock, con); 13886ed7257bSPatrick Caulfield result = 0; 13896ed7257bSPatrick Caulfield } 13906ed7257bSPatrick Caulfield else { 13916ed7257bSPatrick Caulfield result = -EADDRINUSE; 13926ed7257bSPatrick Caulfield } 13936ed7257bSPatrick Caulfield 13946ed7257bSPatrick Caulfield return result; 13956ed7257bSPatrick Caulfield } 13966ed7257bSPatrick Caulfield 13976ed7257bSPatrick Caulfield 13986ed7257bSPatrick Caulfield 13996ed7257bSPatrick Caulfield static struct writequeue_entry *new_writequeue_entry(struct connection *con, 14006ed7257bSPatrick Caulfield gfp_t allocation) 14016ed7257bSPatrick Caulfield { 14026ed7257bSPatrick Caulfield struct writequeue_entry *entry; 14036ed7257bSPatrick Caulfield 14046ed7257bSPatrick Caulfield entry = kmalloc(sizeof(struct writequeue_entry), allocation); 14056ed7257bSPatrick Caulfield if (!entry) 14066ed7257bSPatrick Caulfield return NULL; 14076ed7257bSPatrick Caulfield 14086ed7257bSPatrick Caulfield entry->page = alloc_page(allocation); 14096ed7257bSPatrick Caulfield if (!entry->page) { 14106ed7257bSPatrick Caulfield kfree(entry); 14116ed7257bSPatrick Caulfield return NULL; 14126ed7257bSPatrick Caulfield } 14136ed7257bSPatrick Caulfield 14146ed7257bSPatrick Caulfield entry->offset = 0; 14156ed7257bSPatrick Caulfield entry->len = 0; 14166ed7257bSPatrick Caulfield entry->end = 0; 14176ed7257bSPatrick Caulfield entry->users = 0; 14186ed7257bSPatrick Caulfield entry->con = con; 14196ed7257bSPatrick Caulfield 14206ed7257bSPatrick Caulfield return entry; 14216ed7257bSPatrick Caulfield } 14226ed7257bSPatrick Caulfield 1423617e82e1SDavid Teigland void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) 14246ed7257bSPatrick Caulfield { 14256ed7257bSPatrick Caulfield struct connection *con; 14266ed7257bSPatrick Caulfield struct writequeue_entry *e; 14276ed7257bSPatrick Caulfield int offset = 0; 14286ed7257bSPatrick Caulfield 14296ed7257bSPatrick Caulfield con = nodeid2con(nodeid, allocation); 14306ed7257bSPatrick Caulfield if (!con) 14316ed7257bSPatrick Caulfield return NULL; 14326ed7257bSPatrick Caulfield 14336ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14346ed7257bSPatrick Caulfield e = list_entry(con->writequeue.prev, struct writequeue_entry, list); 14356ed7257bSPatrick Caulfield if ((&e->list == &con->writequeue) || 143609cbfeafSKirill A. Shutemov (PAGE_SIZE - e->end < len)) { 14376ed7257bSPatrick Caulfield e = NULL; 14386ed7257bSPatrick Caulfield } else { 14396ed7257bSPatrick Caulfield offset = e->end; 14406ed7257bSPatrick Caulfield e->end += len; 1441eeee2b5fSWei Yongjun e->users++; 14426ed7257bSPatrick Caulfield } 14436ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14446ed7257bSPatrick Caulfield 14456ed7257bSPatrick Caulfield if (e) { 14466ed7257bSPatrick Caulfield got_one: 14476ed7257bSPatrick Caulfield *ppc = page_address(e->page) + offset; 14486ed7257bSPatrick Caulfield return e; 14496ed7257bSPatrick Caulfield } 14506ed7257bSPatrick Caulfield 14516ed7257bSPatrick Caulfield e = new_writequeue_entry(con, allocation); 14526ed7257bSPatrick Caulfield if (e) { 14536ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14546ed7257bSPatrick Caulfield offset = e->end; 14556ed7257bSPatrick Caulfield e->end += len; 1456eeee2b5fSWei Yongjun e->users++; 14576ed7257bSPatrick Caulfield list_add_tail(&e->list, &con->writequeue); 14586ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14596ed7257bSPatrick Caulfield goto got_one; 14606ed7257bSPatrick Caulfield } 14616ed7257bSPatrick Caulfield return NULL; 14626ed7257bSPatrick Caulfield } 14636ed7257bSPatrick Caulfield 14646ed7257bSPatrick Caulfield void dlm_lowcomms_commit_buffer(void *mh) 14656ed7257bSPatrick Caulfield { 14666ed7257bSPatrick Caulfield struct writequeue_entry *e = (struct writequeue_entry *)mh; 14676ed7257bSPatrick Caulfield struct connection *con = e->con; 14686ed7257bSPatrick Caulfield int users; 14696ed7257bSPatrick Caulfield 14706ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14716ed7257bSPatrick Caulfield users = --e->users; 14726ed7257bSPatrick Caulfield if (users) 14736ed7257bSPatrick Caulfield goto out; 14746ed7257bSPatrick Caulfield e->len = e->end - e->offset; 14756ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14766ed7257bSPatrick Caulfield 14776ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 14786ed7257bSPatrick Caulfield return; 14796ed7257bSPatrick Caulfield 14806ed7257bSPatrick Caulfield out: 14816ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14826ed7257bSPatrick Caulfield return; 14836ed7257bSPatrick Caulfield } 14846ed7257bSPatrick Caulfield 14856ed7257bSPatrick Caulfield /* Send a message */ 14866ed7257bSPatrick Caulfield static void send_to_sock(struct connection *con) 14876ed7257bSPatrick Caulfield { 14886ed7257bSPatrick Caulfield int ret = 0; 14896ed7257bSPatrick Caulfield const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; 14906ed7257bSPatrick Caulfield struct writequeue_entry *e; 14916ed7257bSPatrick Caulfield int len, offset; 1492f92c8dd7SBob Peterson int count = 0; 14936ed7257bSPatrick Caulfield 14946ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 14956ed7257bSPatrick Caulfield if (con->sock == NULL) 14966ed7257bSPatrick Caulfield goto out_connect; 14976ed7257bSPatrick Caulfield 14986ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14996ed7257bSPatrick Caulfield for (;;) { 15006ed7257bSPatrick Caulfield e = list_entry(con->writequeue.next, struct writequeue_entry, 15016ed7257bSPatrick Caulfield list); 15026ed7257bSPatrick Caulfield if ((struct list_head *) e == &con->writequeue) 15036ed7257bSPatrick Caulfield break; 15046ed7257bSPatrick Caulfield 15056ed7257bSPatrick Caulfield len = e->len; 15066ed7257bSPatrick Caulfield offset = e->offset; 15076ed7257bSPatrick Caulfield BUG_ON(len == 0 && e->users == 0); 15086ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15096ed7257bSPatrick Caulfield 15106ed7257bSPatrick Caulfield ret = 0; 15116ed7257bSPatrick Caulfield if (len) { 15121329e3f2SPaolo Bonzini ret = kernel_sendpage(con->sock, e->page, offset, len, 15136ed7257bSPatrick Caulfield msg_flags); 1514d66f8277SPatrick Caulfield if (ret == -EAGAIN || ret == 0) { 1515b36930ddSDavid Miller if (ret == -EAGAIN && 15169cd3e072SEric Dumazet test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) && 1517b36930ddSDavid Miller !test_and_set_bit(CF_APP_LIMITED, &con->flags)) { 1518b36930ddSDavid Miller /* Notify TCP that we're limited by the 1519b36930ddSDavid Miller * application window size. 1520b36930ddSDavid Miller */ 1521b36930ddSDavid Miller set_bit(SOCK_NOSPACE, &con->sock->flags); 1522b36930ddSDavid Miller con->sock->sk->sk_write_pending++; 1523b36930ddSDavid Miller } 1524d66f8277SPatrick Caulfield cond_resched(); 15256ed7257bSPatrick Caulfield goto out; 15269c5bef58SYing Xue } else if (ret < 0) 15276ed7257bSPatrick Caulfield goto send_error; 1528d66f8277SPatrick Caulfield } 1529f92c8dd7SBob Peterson 15306ed7257bSPatrick Caulfield /* Don't starve people filling buffers */ 1531f92c8dd7SBob Peterson if (++count >= MAX_SEND_MSG_COUNT) { 15326ed7257bSPatrick Caulfield cond_resched(); 1533f92c8dd7SBob Peterson count = 0; 1534f92c8dd7SBob Peterson } 15356ed7257bSPatrick Caulfield 15366ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 15375d689871SMike Christie writequeue_entry_complete(e, ret); 15386ed7257bSPatrick Caulfield } 15396ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15406ed7257bSPatrick Caulfield out: 15416ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 15426ed7257bSPatrick Caulfield return; 15436ed7257bSPatrick Caulfield 15446ed7257bSPatrick Caulfield send_error: 15456ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 1546c553e173Stsutomu.owa@toshiba.co.jp close_connection(con, true, false, true); 154701da24d3SBob Peterson /* Requeue the send work. When the work daemon runs again, it will try 154801da24d3SBob Peterson a new connection, then call this function again. */ 154901da24d3SBob Peterson queue_work(send_workqueue, &con->swork); 15506ed7257bSPatrick Caulfield return; 15516ed7257bSPatrick Caulfield 15526ed7257bSPatrick Caulfield out_connect: 15536ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 155401da24d3SBob Peterson queue_work(send_workqueue, &con->swork); 155501da24d3SBob Peterson cond_resched(); 15566ed7257bSPatrick Caulfield } 15576ed7257bSPatrick Caulfield 15586ed7257bSPatrick Caulfield static void clean_one_writequeue(struct connection *con) 15596ed7257bSPatrick Caulfield { 15605e9ccc37SChristine Caulfield struct writequeue_entry *e, *safe; 15616ed7257bSPatrick Caulfield 15626ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 15635e9ccc37SChristine Caulfield list_for_each_entry_safe(e, safe, &con->writequeue, list) { 15646ed7257bSPatrick Caulfield list_del(&e->list); 15656ed7257bSPatrick Caulfield free_entry(e); 15666ed7257bSPatrick Caulfield } 15676ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15686ed7257bSPatrick Caulfield } 15696ed7257bSPatrick Caulfield 15706ed7257bSPatrick Caulfield /* Called from recovery when it knows that a node has 15716ed7257bSPatrick Caulfield left the cluster */ 15726ed7257bSPatrick Caulfield int dlm_lowcomms_close(int nodeid) 15736ed7257bSPatrick Caulfield { 15746ed7257bSPatrick Caulfield struct connection *con; 157536b71a8bSDavid Teigland struct dlm_node_addr *na; 15766ed7257bSPatrick Caulfield 15776ed7257bSPatrick Caulfield log_print("closing connection to node %d", nodeid); 15786ed7257bSPatrick Caulfield con = nodeid2con(nodeid, 0); 15796ed7257bSPatrick Caulfield if (con) { 1580063c4c99SLars Marowsky-Bree set_bit(CF_CLOSE, &con->flags); 15810d737a8cSMarcelo Ricardo Leitner close_connection(con, true, true, true); 15826ed7257bSPatrick Caulfield clean_one_writequeue(con); 15836ed7257bSPatrick Caulfield } 158436b71a8bSDavid Teigland 158536b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 158636b71a8bSDavid Teigland na = find_node_addr(nodeid); 158736b71a8bSDavid Teigland if (na) { 158836b71a8bSDavid Teigland list_del(&na->list); 158936b71a8bSDavid Teigland while (na->addr_count--) 159036b71a8bSDavid Teigland kfree(na->addr[na->addr_count]); 159136b71a8bSDavid Teigland kfree(na); 159236b71a8bSDavid Teigland } 159336b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 159436b71a8bSDavid Teigland 15956ed7257bSPatrick Caulfield return 0; 15966ed7257bSPatrick Caulfield } 15976ed7257bSPatrick Caulfield 15986ed7257bSPatrick Caulfield /* Receive workqueue function */ 15996ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work) 16006ed7257bSPatrick Caulfield { 16016ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, rwork); 16026ed7257bSPatrick Caulfield int err; 16036ed7257bSPatrick Caulfield 16046ed7257bSPatrick Caulfield clear_bit(CF_READ_PENDING, &con->flags); 16056ed7257bSPatrick Caulfield do { 16066ed7257bSPatrick Caulfield err = con->rx_action(con); 16076ed7257bSPatrick Caulfield } while (!err); 16086ed7257bSPatrick Caulfield } 16096ed7257bSPatrick Caulfield 16106ed7257bSPatrick Caulfield /* Send workqueue function */ 16116ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work) 16126ed7257bSPatrick Caulfield { 16136ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, swork); 16146ed7257bSPatrick Caulfield 16158a4abb08Stsutomu.owa@toshiba.co.jp clear_bit(CF_WRITE_PENDING, &con->flags); 161661d9102bSBob Peterson if (con->sock == NULL) /* not mutex protected so check it inside too */ 16176ed7257bSPatrick Caulfield con->connect_action(con); 161801da24d3SBob Peterson if (!list_empty(&con->writequeue)) 16196ed7257bSPatrick Caulfield send_to_sock(con); 16206ed7257bSPatrick Caulfield } 16216ed7257bSPatrick Caulfield 16226ed7257bSPatrick Caulfield 16236ed7257bSPatrick Caulfield /* Discard all entries on the write queues */ 16246ed7257bSPatrick Caulfield static void clean_writequeues(void) 16256ed7257bSPatrick Caulfield { 16265e9ccc37SChristine Caulfield foreach_conn(clean_one_writequeue); 16276ed7257bSPatrick Caulfield } 16286ed7257bSPatrick Caulfield 16296ed7257bSPatrick Caulfield static void work_stop(void) 16306ed7257bSPatrick Caulfield { 16316ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 16326ed7257bSPatrick Caulfield destroy_workqueue(send_workqueue); 16336ed7257bSPatrick Caulfield } 16346ed7257bSPatrick Caulfield 16356ed7257bSPatrick Caulfield static int work_start(void) 16366ed7257bSPatrick Caulfield { 1637e43f055aSDavid Teigland recv_workqueue = alloc_workqueue("dlm_recv", 1638e43f055aSDavid Teigland WQ_UNBOUND | WQ_MEM_RECLAIM, 1); 1639b9d41052SNamhyung Kim if (!recv_workqueue) { 1640b9d41052SNamhyung Kim log_print("can't start dlm_recv"); 1641b9d41052SNamhyung Kim return -ENOMEM; 16426ed7257bSPatrick Caulfield } 16436ed7257bSPatrick Caulfield 1644e43f055aSDavid Teigland send_workqueue = alloc_workqueue("dlm_send", 1645e43f055aSDavid Teigland WQ_UNBOUND | WQ_MEM_RECLAIM, 1); 1646b9d41052SNamhyung Kim if (!send_workqueue) { 1647b9d41052SNamhyung Kim log_print("can't start dlm_send"); 16486ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 1649b9d41052SNamhyung Kim return -ENOMEM; 16506ed7257bSPatrick Caulfield } 16516ed7257bSPatrick Caulfield 16526ed7257bSPatrick Caulfield return 0; 16536ed7257bSPatrick Caulfield } 16546ed7257bSPatrick Caulfield 1655f0fb83cbStsutomu.owa@toshiba.co.jp static void _stop_conn(struct connection *con, bool and_other) 16566ed7257bSPatrick Caulfield { 1657f0fb83cbStsutomu.owa@toshiba.co.jp mutex_lock(&con->sock_mutex); 1658173a31feStsutomu.owa@toshiba.co.jp set_bit(CF_CLOSE, &con->flags); 1659f0fb83cbStsutomu.owa@toshiba.co.jp set_bit(CF_READ_PENDING, &con->flags); 16608a4abb08Stsutomu.owa@toshiba.co.jp set_bit(CF_WRITE_PENDING, &con->flags); 166193eaadebStsutomu.owa@toshiba.co.jp if (con->sock && con->sock->sk) { 166293eaadebStsutomu.owa@toshiba.co.jp write_lock_bh(&con->sock->sk->sk_callback_lock); 1663afb853fbSPatrick Caulfield con->sock->sk->sk_user_data = NULL; 166493eaadebStsutomu.owa@toshiba.co.jp write_unlock_bh(&con->sock->sk->sk_callback_lock); 166593eaadebStsutomu.owa@toshiba.co.jp } 1666f0fb83cbStsutomu.owa@toshiba.co.jp if (con->othercon && and_other) 1667f0fb83cbStsutomu.owa@toshiba.co.jp _stop_conn(con->othercon, false); 1668f0fb83cbStsutomu.owa@toshiba.co.jp mutex_unlock(&con->sock_mutex); 1669f0fb83cbStsutomu.owa@toshiba.co.jp } 1670f0fb83cbStsutomu.owa@toshiba.co.jp 1671f0fb83cbStsutomu.owa@toshiba.co.jp static void stop_conn(struct connection *con) 1672f0fb83cbStsutomu.owa@toshiba.co.jp { 1673f0fb83cbStsutomu.owa@toshiba.co.jp _stop_conn(con, true); 1674afb853fbSPatrick Caulfield } 16755e9ccc37SChristine Caulfield 16765e9ccc37SChristine Caulfield static void free_conn(struct connection *con) 16775e9ccc37SChristine Caulfield { 16780d737a8cSMarcelo Ricardo Leitner close_connection(con, true, true, true); 16795e9ccc37SChristine Caulfield if (con->othercon) 16805e9ccc37SChristine Caulfield kmem_cache_free(con_cache, con->othercon); 16815e9ccc37SChristine Caulfield hlist_del(&con->list); 16825e9ccc37SChristine Caulfield kmem_cache_free(con_cache, con); 16836ed7257bSPatrick Caulfield } 16845e9ccc37SChristine Caulfield 1685f0fb83cbStsutomu.owa@toshiba.co.jp static void work_flush(void) 1686f0fb83cbStsutomu.owa@toshiba.co.jp { 1687f0fb83cbStsutomu.owa@toshiba.co.jp int ok; 1688f0fb83cbStsutomu.owa@toshiba.co.jp int i; 1689f0fb83cbStsutomu.owa@toshiba.co.jp struct hlist_node *n; 1690f0fb83cbStsutomu.owa@toshiba.co.jp struct connection *con; 1691f0fb83cbStsutomu.owa@toshiba.co.jp 1692f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(recv_workqueue); 1693f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(send_workqueue); 1694f0fb83cbStsutomu.owa@toshiba.co.jp do { 1695f0fb83cbStsutomu.owa@toshiba.co.jp ok = 1; 1696f0fb83cbStsutomu.owa@toshiba.co.jp foreach_conn(stop_conn); 1697f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(recv_workqueue); 1698f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(send_workqueue); 1699f0fb83cbStsutomu.owa@toshiba.co.jp for (i = 0; i < CONN_HASH_SIZE && ok; i++) { 1700f0fb83cbStsutomu.owa@toshiba.co.jp hlist_for_each_entry_safe(con, n, 1701f0fb83cbStsutomu.owa@toshiba.co.jp &connection_hash[i], list) { 1702f0fb83cbStsutomu.owa@toshiba.co.jp ok &= test_bit(CF_READ_PENDING, &con->flags); 17038a4abb08Stsutomu.owa@toshiba.co.jp ok &= test_bit(CF_WRITE_PENDING, &con->flags); 17048a4abb08Stsutomu.owa@toshiba.co.jp if (con->othercon) { 1705f0fb83cbStsutomu.owa@toshiba.co.jp ok &= test_bit(CF_READ_PENDING, 1706f0fb83cbStsutomu.owa@toshiba.co.jp &con->othercon->flags); 17078a4abb08Stsutomu.owa@toshiba.co.jp ok &= test_bit(CF_WRITE_PENDING, 17088a4abb08Stsutomu.owa@toshiba.co.jp &con->othercon->flags); 17098a4abb08Stsutomu.owa@toshiba.co.jp } 1710f0fb83cbStsutomu.owa@toshiba.co.jp } 1711f0fb83cbStsutomu.owa@toshiba.co.jp } 1712f0fb83cbStsutomu.owa@toshiba.co.jp } while (!ok); 1713f0fb83cbStsutomu.owa@toshiba.co.jp } 1714f0fb83cbStsutomu.owa@toshiba.co.jp 17155e9ccc37SChristine Caulfield void dlm_lowcomms_stop(void) 17165e9ccc37SChristine Caulfield { 17175e9ccc37SChristine Caulfield /* Set all the flags to prevent any 17185e9ccc37SChristine Caulfield socket activity. 17195e9ccc37SChristine Caulfield */ 17205e9ccc37SChristine Caulfield mutex_lock(&connections_lock); 1721513ef596SDavid Teigland dlm_allow_conn = 0; 1722f0fb83cbStsutomu.owa@toshiba.co.jp mutex_unlock(&connections_lock); 1723f0fb83cbStsutomu.owa@toshiba.co.jp work_flush(); 17243a8db798SMarcelo Ricardo Leitner clean_writequeues(); 17253a8db798SMarcelo Ricardo Leitner foreach_conn(free_conn); 17266ed7257bSPatrick Caulfield work_stop(); 17276ed7257bSPatrick Caulfield 17286ed7257bSPatrick Caulfield kmem_cache_destroy(con_cache); 17296ed7257bSPatrick Caulfield } 17306ed7257bSPatrick Caulfield 17316ed7257bSPatrick Caulfield int dlm_lowcomms_start(void) 17326ed7257bSPatrick Caulfield { 17336ed7257bSPatrick Caulfield int error = -EINVAL; 17346ed7257bSPatrick Caulfield struct connection *con; 17355e9ccc37SChristine Caulfield int i; 17365e9ccc37SChristine Caulfield 17375e9ccc37SChristine Caulfield for (i = 0; i < CONN_HASH_SIZE; i++) 17385e9ccc37SChristine Caulfield INIT_HLIST_HEAD(&connection_hash[i]); 17396ed7257bSPatrick Caulfield 17406ed7257bSPatrick Caulfield init_local(); 17416ed7257bSPatrick Caulfield if (!dlm_local_count) { 1742617e82e1SDavid Teigland error = -ENOTCONN; 17436ed7257bSPatrick Caulfield log_print("no local IP address has been set"); 1744513ef596SDavid Teigland goto fail; 17456ed7257bSPatrick Caulfield } 17466ed7257bSPatrick Caulfield 17476ed7257bSPatrick Caulfield error = -ENOMEM; 17486ed7257bSPatrick Caulfield con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), 17496ed7257bSPatrick Caulfield __alignof__(struct connection), 0, 175020c2df83SPaul Mundt NULL); 17516ed7257bSPatrick Caulfield if (!con_cache) 1752513ef596SDavid Teigland goto fail; 1753513ef596SDavid Teigland 1754513ef596SDavid Teigland error = work_start(); 1755513ef596SDavid Teigland if (error) 1756513ef596SDavid Teigland goto fail_destroy; 1757513ef596SDavid Teigland 1758513ef596SDavid Teigland dlm_allow_conn = 1; 17596ed7257bSPatrick Caulfield 17606ed7257bSPatrick Caulfield /* Start listening */ 17616ed7257bSPatrick Caulfield if (dlm_config.ci_protocol == 0) 17626ed7257bSPatrick Caulfield error = tcp_listen_for_all(); 17636ed7257bSPatrick Caulfield else 17646ed7257bSPatrick Caulfield error = sctp_listen_for_all(); 17656ed7257bSPatrick Caulfield if (error) 17666ed7257bSPatrick Caulfield goto fail_unlisten; 17676ed7257bSPatrick Caulfield 17686ed7257bSPatrick Caulfield return 0; 17696ed7257bSPatrick Caulfield 17706ed7257bSPatrick Caulfield fail_unlisten: 1771513ef596SDavid Teigland dlm_allow_conn = 0; 17726ed7257bSPatrick Caulfield con = nodeid2con(0,0); 17736ed7257bSPatrick Caulfield if (con) { 17740d737a8cSMarcelo Ricardo Leitner close_connection(con, false, true, true); 17756ed7257bSPatrick Caulfield kmem_cache_free(con_cache, con); 17766ed7257bSPatrick Caulfield } 1777513ef596SDavid Teigland fail_destroy: 17786ed7257bSPatrick Caulfield kmem_cache_destroy(con_cache); 1779513ef596SDavid Teigland fail: 17806ed7257bSPatrick Caulfield return error; 17816ed7257bSPatrick Caulfield } 178236b71a8bSDavid Teigland 178336b71a8bSDavid Teigland void dlm_lowcomms_exit(void) 178436b71a8bSDavid Teigland { 178536b71a8bSDavid Teigland struct dlm_node_addr *na, *safe; 178636b71a8bSDavid Teigland 178736b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 178836b71a8bSDavid Teigland list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) { 178936b71a8bSDavid Teigland list_del(&na->list); 179036b71a8bSDavid Teigland while (na->addr_count--) 179136b71a8bSDavid Teigland kfree(na->addr[na->addr_count]); 179236b71a8bSDavid Teigland kfree(na); 179336b71a8bSDavid Teigland } 179436b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 179536b71a8bSDavid Teigland } 1796