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 5692732376SAlexander Aring #include <trace/events/dlm.h> 5792732376SAlexander Aring 586ed7257bSPatrick Caulfield #include "dlm_internal.h" 596ed7257bSPatrick Caulfield #include "lowcomms.h" 606ed7257bSPatrick Caulfield #include "midcomms.h" 613af2326cSAlexander Aring #include "memory.h" 626ed7257bSPatrick Caulfield #include "config.h" 636ed7257bSPatrick Caulfield 646ed7257bSPatrick Caulfield #define NEEDED_RMEM (4*1024*1024) 656ed7257bSPatrick Caulfield 66f92c8dd7SBob Peterson /* Number of messages to send before rescheduling */ 67f92c8dd7SBob Peterson #define MAX_SEND_MSG_COUNT 25 68055923bfSAlexander Aring #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(10000) 69f92c8dd7SBob Peterson 706ed7257bSPatrick Caulfield struct connection { 716ed7257bSPatrick Caulfield struct socket *sock; /* NULL if not connected */ 726ed7257bSPatrick Caulfield uint32_t nodeid; /* So we know who we are in the list */ 736ed7257bSPatrick Caulfield struct mutex sock_mutex; 746ed7257bSPatrick Caulfield unsigned long flags; 756ed7257bSPatrick Caulfield #define CF_READ_PENDING 1 768a4abb08Stsutomu.owa@toshiba.co.jp #define CF_WRITE_PENDING 2 776ed7257bSPatrick Caulfield #define CF_INIT_PENDING 4 786ed7257bSPatrick Caulfield #define CF_IS_OTHERCON 5 79063c4c99SLars Marowsky-Bree #define CF_CLOSE 6 80b36930ddSDavid Miller #define CF_APP_LIMITED 7 81b2a66629Stsutomu.owa@toshiba.co.jp #define CF_CLOSING 8 82055923bfSAlexander Aring #define CF_SHUTDOWN 9 8319633c7eSAlexander Aring #define CF_CONNECTED 10 84ba868d9dSAlexander Aring #define CF_RECONNECT 11 85ba868d9dSAlexander Aring #define CF_DELAY_CONNECT 12 868aa31cbfSAlexander Aring #define CF_EOF 13 876ed7257bSPatrick Caulfield struct list_head writequeue; /* List of outgoing writequeue_entries */ 886ed7257bSPatrick Caulfield spinlock_t writequeue_lock; 898aa31cbfSAlexander Aring atomic_t writequeue_cnt; 906ed7257bSPatrick Caulfield int retries; 916ed7257bSPatrick Caulfield #define MAX_CONNECT_RETRIES 3 925e9ccc37SChristine Caulfield struct hlist_node list; 936ed7257bSPatrick Caulfield struct connection *othercon; 94ba868d9dSAlexander Aring struct connection *sendcon; 956ed7257bSPatrick Caulfield struct work_struct rwork; /* Receive workqueue */ 966ed7257bSPatrick Caulfield struct work_struct swork; /* Send workqueue */ 97055923bfSAlexander Aring wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */ 984798cbbfSAlexander Aring unsigned char *rx_buf; 994798cbbfSAlexander Aring int rx_buflen; 1004798cbbfSAlexander Aring int rx_leftover; 101a47666ebSAlexander Aring struct rcu_head rcu; 1026ed7257bSPatrick Caulfield }; 1036ed7257bSPatrick Caulfield #define sock2con(x) ((struct connection *)(x)->sk_user_data) 1046ed7257bSPatrick Caulfield 105d11ccd45SAlexander Aring struct listen_connection { 106d11ccd45SAlexander Aring struct socket *sock; 107d11ccd45SAlexander Aring struct work_struct rwork; 108d11ccd45SAlexander Aring }; 109d11ccd45SAlexander Aring 110f0747ebfSAlexander Aring #define DLM_WQ_REMAIN_BYTES(e) (PAGE_SIZE - e->end) 111f0747ebfSAlexander Aring #define DLM_WQ_LENGTH_BYTES(e) (e->end - e->offset) 112f0747ebfSAlexander Aring 1136ed7257bSPatrick Caulfield /* An entry waiting to be sent */ 1146ed7257bSPatrick Caulfield struct writequeue_entry { 1156ed7257bSPatrick Caulfield struct list_head list; 1166ed7257bSPatrick Caulfield struct page *page; 1176ed7257bSPatrick Caulfield int offset; 1186ed7257bSPatrick Caulfield int len; 1196ed7257bSPatrick Caulfield int end; 1206ed7257bSPatrick Caulfield int users; 121706474fbSAlexander Aring bool dirty; 1226ed7257bSPatrick Caulfield struct connection *con; 1238f2dc78dSAlexander Aring struct list_head msgs; 1248f2dc78dSAlexander Aring struct kref ref; 1258f2dc78dSAlexander Aring }; 1268f2dc78dSAlexander Aring 1278f2dc78dSAlexander Aring struct dlm_msg { 1288f2dc78dSAlexander Aring struct writequeue_entry *entry; 1292874d1a6SAlexander Aring struct dlm_msg *orig_msg; 1302874d1a6SAlexander Aring bool retransmit; 1318f2dc78dSAlexander Aring void *ppc; 1328f2dc78dSAlexander Aring int len; 1338f2dc78dSAlexander Aring int idx; /* new()/commit() idx exchange */ 1348f2dc78dSAlexander Aring 1358f2dc78dSAlexander Aring struct list_head list; 1368f2dc78dSAlexander Aring struct kref ref; 1376ed7257bSPatrick Caulfield }; 1386ed7257bSPatrick Caulfield 13936b71a8bSDavid Teigland struct dlm_node_addr { 14036b71a8bSDavid Teigland struct list_head list; 14136b71a8bSDavid Teigland int nodeid; 142e125fbebSAlexander Aring int mark; 14336b71a8bSDavid Teigland int addr_count; 14498e1b60eSMike Christie int curr_addr_index; 14536b71a8bSDavid Teigland struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT]; 14636b71a8bSDavid Teigland }; 14736b71a8bSDavid Teigland 148a66c008cSAlexander Aring struct dlm_proto_ops { 1498728a455SAlexander Aring bool try_new_addr; 1502dc6b115SAlexander Aring const char *name; 1512dc6b115SAlexander Aring int proto; 1522dc6b115SAlexander Aring 1538728a455SAlexander Aring int (*connect)(struct connection *con, struct socket *sock, 1548728a455SAlexander Aring struct sockaddr *addr, int addr_len); 1558728a455SAlexander Aring void (*sockopts)(struct socket *sock); 1568728a455SAlexander Aring int (*bind)(struct socket *sock); 1572dc6b115SAlexander Aring int (*listen_validate)(void); 1582dc6b115SAlexander Aring void (*listen_sockopts)(struct socket *sock); 1592dc6b115SAlexander Aring int (*listen_bind)(struct socket *sock); 160a66c008cSAlexander Aring /* What to do to shutdown */ 161a66c008cSAlexander Aring void (*shutdown_action)(struct connection *con); 162a66c008cSAlexander Aring /* What to do to eof check */ 163a66c008cSAlexander Aring bool (*eof_condition)(struct connection *con); 164a66c008cSAlexander Aring }; 165a66c008cSAlexander Aring 166cc661fc9SBob Peterson static struct listen_sock_callbacks { 167cc661fc9SBob Peterson void (*sk_error_report)(struct sock *); 168cc661fc9SBob Peterson void (*sk_data_ready)(struct sock *); 169cc661fc9SBob Peterson void (*sk_state_change)(struct sock *); 170cc661fc9SBob Peterson void (*sk_write_space)(struct sock *); 171cc661fc9SBob Peterson } listen_sock; 172cc661fc9SBob Peterson 17336b71a8bSDavid Teigland static LIST_HEAD(dlm_node_addrs); 17436b71a8bSDavid Teigland static DEFINE_SPINLOCK(dlm_node_addrs_spin); 17536b71a8bSDavid Teigland 176d11ccd45SAlexander Aring static struct listen_connection listen_con; 1776ed7257bSPatrick Caulfield static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; 1786ed7257bSPatrick Caulfield static int dlm_local_count; 17951746163SAlexander Aring int dlm_allow_conn; 1806ed7257bSPatrick Caulfield 1816ed7257bSPatrick Caulfield /* Work queues */ 1826ed7257bSPatrick Caulfield static struct workqueue_struct *recv_workqueue; 1836ed7257bSPatrick Caulfield static struct workqueue_struct *send_workqueue; 1846ed7257bSPatrick Caulfield 1855e9ccc37SChristine Caulfield static struct hlist_head connection_hash[CONN_HASH_SIZE]; 186a47666ebSAlexander Aring static DEFINE_SPINLOCK(connections_lock); 187a47666ebSAlexander Aring DEFINE_STATIC_SRCU(connections_srcu); 1886ed7257bSPatrick Caulfield 189a66c008cSAlexander Aring static const struct dlm_proto_ops *dlm_proto_ops; 190a66c008cSAlexander Aring 1916ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work); 1926ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work); 1936ed7257bSPatrick Caulfield 1943af2326cSAlexander Aring static void writequeue_entry_ctor(void *data) 1953af2326cSAlexander Aring { 1963af2326cSAlexander Aring struct writequeue_entry *entry = data; 1973af2326cSAlexander Aring 1983af2326cSAlexander Aring INIT_LIST_HEAD(&entry->msgs); 1993af2326cSAlexander Aring } 2003af2326cSAlexander Aring 2013af2326cSAlexander Aring struct kmem_cache *dlm_lowcomms_writequeue_cache_create(void) 2023af2326cSAlexander Aring { 2033af2326cSAlexander Aring return kmem_cache_create("dlm_writequeue", sizeof(struct writequeue_entry), 2043af2326cSAlexander Aring 0, 0, writequeue_entry_ctor); 2053af2326cSAlexander Aring } 2063af2326cSAlexander Aring 207e4dc81edSAlexander Aring struct kmem_cache *dlm_lowcomms_msg_cache_create(void) 208e4dc81edSAlexander Aring { 209e4dc81edSAlexander Aring return kmem_cache_create("dlm_msg", sizeof(struct dlm_msg), 0, 0, NULL); 210e4dc81edSAlexander Aring } 211e4dc81edSAlexander Aring 21266d5955aSAlexander Aring /* need to held writequeue_lock */ 21366d5955aSAlexander Aring static struct writequeue_entry *con_next_wq(struct connection *con) 21466d5955aSAlexander Aring { 21566d5955aSAlexander Aring struct writequeue_entry *e; 21666d5955aSAlexander Aring 21766d5955aSAlexander Aring if (list_empty(&con->writequeue)) 21866d5955aSAlexander Aring return NULL; 21966d5955aSAlexander Aring 22066d5955aSAlexander Aring e = list_first_entry(&con->writequeue, struct writequeue_entry, 22166d5955aSAlexander Aring list); 222bcbfea41SAlexander Aring /* if len is zero nothing is to send, if there are users filling 223bcbfea41SAlexander Aring * buffers we wait until the users are done so we can send more. 224bcbfea41SAlexander Aring */ 225bcbfea41SAlexander Aring if (e->users || e->len == 0) 22666d5955aSAlexander Aring return NULL; 22766d5955aSAlexander Aring 22866d5955aSAlexander Aring return e; 22966d5955aSAlexander Aring } 23066d5955aSAlexander Aring 231b38bc9c2SAlexander Aring static struct connection *__find_con(int nodeid, int r) 2325e9ccc37SChristine Caulfield { 2335e9ccc37SChristine Caulfield struct connection *con; 2345e9ccc37SChristine Caulfield 235a47666ebSAlexander Aring hlist_for_each_entry_rcu(con, &connection_hash[r], list) { 236b38bc9c2SAlexander Aring if (con->nodeid == nodeid) 2375e9ccc37SChristine Caulfield return con; 2385e9ccc37SChristine Caulfield } 239a47666ebSAlexander Aring 2405e9ccc37SChristine Caulfield return NULL; 2415e9ccc37SChristine Caulfield } 2425e9ccc37SChristine Caulfield 2438aa31cbfSAlexander Aring static bool tcp_eof_condition(struct connection *con) 2448aa31cbfSAlexander Aring { 2458aa31cbfSAlexander Aring return atomic_read(&con->writequeue_cnt); 2468aa31cbfSAlexander Aring } 2478aa31cbfSAlexander Aring 2486cde210aSAlexander Aring static int dlm_con_init(struct connection *con, int nodeid) 2496ed7257bSPatrick Caulfield { 2504798cbbfSAlexander Aring con->rx_buflen = dlm_config.ci_buffer_size; 2514798cbbfSAlexander Aring con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS); 2526cde210aSAlexander Aring if (!con->rx_buf) 2536cde210aSAlexander Aring return -ENOMEM; 2544798cbbfSAlexander Aring 2556ed7257bSPatrick Caulfield con->nodeid = nodeid; 2566ed7257bSPatrick Caulfield mutex_init(&con->sock_mutex); 2576ed7257bSPatrick Caulfield INIT_LIST_HEAD(&con->writequeue); 2586ed7257bSPatrick Caulfield spin_lock_init(&con->writequeue_lock); 2598aa31cbfSAlexander Aring atomic_set(&con->writequeue_cnt, 0); 2606ed7257bSPatrick Caulfield INIT_WORK(&con->swork, process_send_sockets); 2616ed7257bSPatrick Caulfield INIT_WORK(&con->rwork, process_recv_sockets); 262055923bfSAlexander Aring init_waitqueue_head(&con->shutdown_wait); 2636ed7257bSPatrick Caulfield 2646cde210aSAlexander Aring return 0; 2656cde210aSAlexander Aring } 2666cde210aSAlexander Aring 2676cde210aSAlexander Aring /* 2686cde210aSAlexander Aring * If 'allocation' is zero then we don't attempt to create a new 2696cde210aSAlexander Aring * connection structure for this node. 2706cde210aSAlexander Aring */ 2716cde210aSAlexander Aring static struct connection *nodeid2con(int nodeid, gfp_t alloc) 2726cde210aSAlexander Aring { 2736cde210aSAlexander Aring struct connection *con, *tmp; 2746cde210aSAlexander Aring int r, ret; 2756cde210aSAlexander Aring 276b38bc9c2SAlexander Aring r = nodeid_hash(nodeid); 277b38bc9c2SAlexander Aring con = __find_con(nodeid, r); 2786cde210aSAlexander Aring if (con || !alloc) 2796cde210aSAlexander Aring return con; 2806cde210aSAlexander Aring 2816cde210aSAlexander Aring con = kzalloc(sizeof(*con), alloc); 2826cde210aSAlexander Aring if (!con) 2836cde210aSAlexander Aring return NULL; 2846cde210aSAlexander Aring 2856cde210aSAlexander Aring ret = dlm_con_init(con, nodeid); 2866cde210aSAlexander Aring if (ret) { 2876cde210aSAlexander Aring kfree(con); 2886cde210aSAlexander Aring return NULL; 2896cde210aSAlexander Aring } 2906cde210aSAlexander Aring 291a47666ebSAlexander Aring spin_lock(&connections_lock); 2924f2b30fdSAlexander Aring /* Because multiple workqueues/threads calls this function it can 2934f2b30fdSAlexander Aring * race on multiple cpu's. Instead of locking hot path __find_con() 2944f2b30fdSAlexander Aring * we just check in rare cases of recently added nodes again 2954f2b30fdSAlexander Aring * under protection of connections_lock. If this is the case we 2964f2b30fdSAlexander Aring * abort our connection creation and return the existing connection. 2974f2b30fdSAlexander Aring */ 298b38bc9c2SAlexander Aring tmp = __find_con(nodeid, r); 2994f2b30fdSAlexander Aring if (tmp) { 3004f2b30fdSAlexander Aring spin_unlock(&connections_lock); 3014f2b30fdSAlexander Aring kfree(con->rx_buf); 3024f2b30fdSAlexander Aring kfree(con); 3034f2b30fdSAlexander Aring return tmp; 3044f2b30fdSAlexander Aring } 3054f2b30fdSAlexander Aring 306a47666ebSAlexander Aring hlist_add_head_rcu(&con->list, &connection_hash[r]); 307a47666ebSAlexander Aring spin_unlock(&connections_lock); 308a47666ebSAlexander Aring 3096ed7257bSPatrick Caulfield return con; 3106ed7257bSPatrick Caulfield } 3116ed7257bSPatrick Caulfield 3125e9ccc37SChristine Caulfield /* Loop round all connections */ 3135e9ccc37SChristine Caulfield static void foreach_conn(void (*conn_func)(struct connection *c)) 3145e9ccc37SChristine Caulfield { 315b38bc9c2SAlexander Aring int i; 3165e9ccc37SChristine Caulfield struct connection *con; 3175e9ccc37SChristine Caulfield 3185e9ccc37SChristine Caulfield for (i = 0; i < CONN_HASH_SIZE; i++) { 319a47666ebSAlexander Aring hlist_for_each_entry_rcu(con, &connection_hash[i], list) 3205e9ccc37SChristine Caulfield conn_func(con); 3215e9ccc37SChristine Caulfield } 3226ed7257bSPatrick Caulfield } 3236ed7257bSPatrick Caulfield 32436b71a8bSDavid Teigland static struct dlm_node_addr *find_node_addr(int nodeid) 3256ed7257bSPatrick Caulfield { 32636b71a8bSDavid Teigland struct dlm_node_addr *na; 32736b71a8bSDavid Teigland 32836b71a8bSDavid Teigland list_for_each_entry(na, &dlm_node_addrs, list) { 32936b71a8bSDavid Teigland if (na->nodeid == nodeid) 33036b71a8bSDavid Teigland return na; 33136b71a8bSDavid Teigland } 33236b71a8bSDavid Teigland return NULL; 33336b71a8bSDavid Teigland } 33436b71a8bSDavid Teigland 33540c6b83eSAlexander Aring static int addr_compare(const struct sockaddr_storage *x, 33640c6b83eSAlexander Aring const struct sockaddr_storage *y) 33736b71a8bSDavid Teigland { 33836b71a8bSDavid Teigland switch (x->ss_family) { 33936b71a8bSDavid Teigland case AF_INET: { 34036b71a8bSDavid Teigland struct sockaddr_in *sinx = (struct sockaddr_in *)x; 34136b71a8bSDavid Teigland struct sockaddr_in *siny = (struct sockaddr_in *)y; 34236b71a8bSDavid Teigland if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr) 34336b71a8bSDavid Teigland return 0; 34436b71a8bSDavid Teigland if (sinx->sin_port != siny->sin_port) 34536b71a8bSDavid Teigland return 0; 34636b71a8bSDavid Teigland break; 34736b71a8bSDavid Teigland } 34836b71a8bSDavid Teigland case AF_INET6: { 34936b71a8bSDavid Teigland struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x; 35036b71a8bSDavid Teigland struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y; 35136b71a8bSDavid Teigland if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr)) 35236b71a8bSDavid Teigland return 0; 35336b71a8bSDavid Teigland if (sinx->sin6_port != siny->sin6_port) 35436b71a8bSDavid Teigland return 0; 35536b71a8bSDavid Teigland break; 35636b71a8bSDavid Teigland } 35736b71a8bSDavid Teigland default: 35836b71a8bSDavid Teigland return 0; 35936b71a8bSDavid Teigland } 36036b71a8bSDavid Teigland return 1; 36136b71a8bSDavid Teigland } 36236b71a8bSDavid Teigland 36336b71a8bSDavid Teigland static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out, 364e125fbebSAlexander Aring struct sockaddr *sa_out, bool try_new_addr, 365e125fbebSAlexander Aring unsigned int *mark) 36636b71a8bSDavid Teigland { 36736b71a8bSDavid Teigland struct sockaddr_storage sas; 36836b71a8bSDavid Teigland struct dlm_node_addr *na; 3696ed7257bSPatrick Caulfield 3706ed7257bSPatrick Caulfield if (!dlm_local_count) 3716ed7257bSPatrick Caulfield return -1; 3726ed7257bSPatrick Caulfield 37336b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 37436b71a8bSDavid Teigland na = find_node_addr(nodeid); 37598e1b60eSMike Christie if (na && na->addr_count) { 376ee44b4bcSMarcelo Ricardo Leitner memcpy(&sas, na->addr[na->curr_addr_index], 377ee44b4bcSMarcelo Ricardo Leitner sizeof(struct sockaddr_storage)); 378ee44b4bcSMarcelo Ricardo Leitner 37998e1b60eSMike Christie if (try_new_addr) { 38098e1b60eSMike Christie na->curr_addr_index++; 38198e1b60eSMike Christie if (na->curr_addr_index == na->addr_count) 38298e1b60eSMike Christie na->curr_addr_index = 0; 38398e1b60eSMike Christie } 38498e1b60eSMike Christie } 38536b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 38636b71a8bSDavid Teigland 38736b71a8bSDavid Teigland if (!na) 38836b71a8bSDavid Teigland return -EEXIST; 38936b71a8bSDavid Teigland 39036b71a8bSDavid Teigland if (!na->addr_count) 39136b71a8bSDavid Teigland return -ENOENT; 39236b71a8bSDavid Teigland 393e125fbebSAlexander Aring *mark = na->mark; 394e125fbebSAlexander Aring 39536b71a8bSDavid Teigland if (sas_out) 39636b71a8bSDavid Teigland memcpy(sas_out, &sas, sizeof(struct sockaddr_storage)); 39736b71a8bSDavid Teigland 39836b71a8bSDavid Teigland if (!sa_out) 39936b71a8bSDavid Teigland return 0; 4006ed7257bSPatrick Caulfield 4016ed7257bSPatrick Caulfield if (dlm_local_addr[0]->ss_family == AF_INET) { 40236b71a8bSDavid Teigland struct sockaddr_in *in4 = (struct sockaddr_in *) &sas; 40336b71a8bSDavid Teigland struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out; 4046ed7257bSPatrick Caulfield ret4->sin_addr.s_addr = in4->sin_addr.s_addr; 4056ed7257bSPatrick Caulfield } else { 40636b71a8bSDavid Teigland struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &sas; 40736b71a8bSDavid Teigland struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out; 4084e3fd7a0SAlexey Dobriyan ret6->sin6_addr = in6->sin6_addr; 4096ed7257bSPatrick Caulfield } 4106ed7257bSPatrick Caulfield 4116ed7257bSPatrick Caulfield return 0; 4126ed7257bSPatrick Caulfield } 4136ed7257bSPatrick Caulfield 414e125fbebSAlexander Aring static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid, 415e125fbebSAlexander Aring unsigned int *mark) 41636b71a8bSDavid Teigland { 41736b71a8bSDavid Teigland struct dlm_node_addr *na; 41836b71a8bSDavid Teigland int rv = -EEXIST; 41998e1b60eSMike Christie int addr_i; 42036b71a8bSDavid Teigland 42136b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 42236b71a8bSDavid Teigland list_for_each_entry(na, &dlm_node_addrs, list) { 42336b71a8bSDavid Teigland if (!na->addr_count) 42436b71a8bSDavid Teigland continue; 42536b71a8bSDavid Teigland 42698e1b60eSMike Christie for (addr_i = 0; addr_i < na->addr_count; addr_i++) { 42798e1b60eSMike Christie if (addr_compare(na->addr[addr_i], addr)) { 42836b71a8bSDavid Teigland *nodeid = na->nodeid; 429e125fbebSAlexander Aring *mark = na->mark; 43036b71a8bSDavid Teigland rv = 0; 43198e1b60eSMike Christie goto unlock; 43236b71a8bSDavid Teigland } 43398e1b60eSMike Christie } 43498e1b60eSMike Christie } 43598e1b60eSMike Christie unlock: 43636b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 43736b71a8bSDavid Teigland return rv; 43836b71a8bSDavid Teigland } 43936b71a8bSDavid Teigland 4404f19d071SAlexander Aring /* caller need to held dlm_node_addrs_spin lock */ 4414f19d071SAlexander Aring static bool dlm_lowcomms_na_has_addr(const struct dlm_node_addr *na, 4424f19d071SAlexander Aring const struct sockaddr_storage *addr) 4434f19d071SAlexander Aring { 4444f19d071SAlexander Aring int i; 4454f19d071SAlexander Aring 4464f19d071SAlexander Aring for (i = 0; i < na->addr_count; i++) { 4474f19d071SAlexander Aring if (addr_compare(na->addr[i], addr)) 4484f19d071SAlexander Aring return true; 4494f19d071SAlexander Aring } 4504f19d071SAlexander Aring 4514f19d071SAlexander Aring return false; 4524f19d071SAlexander Aring } 4534f19d071SAlexander Aring 45436b71a8bSDavid Teigland int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len) 45536b71a8bSDavid Teigland { 45636b71a8bSDavid Teigland struct sockaddr_storage *new_addr; 45736b71a8bSDavid Teigland struct dlm_node_addr *new_node, *na; 4584f19d071SAlexander Aring bool ret; 45936b71a8bSDavid Teigland 46036b71a8bSDavid Teigland new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS); 46136b71a8bSDavid Teigland if (!new_node) 46236b71a8bSDavid Teigland return -ENOMEM; 46336b71a8bSDavid Teigland 46436b71a8bSDavid Teigland new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS); 46536b71a8bSDavid Teigland if (!new_addr) { 46636b71a8bSDavid Teigland kfree(new_node); 46736b71a8bSDavid Teigland return -ENOMEM; 46836b71a8bSDavid Teigland } 46936b71a8bSDavid Teigland 47036b71a8bSDavid Teigland memcpy(new_addr, addr, len); 47136b71a8bSDavid Teigland 47236b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 47336b71a8bSDavid Teigland na = find_node_addr(nodeid); 47436b71a8bSDavid Teigland if (!na) { 47536b71a8bSDavid Teigland new_node->nodeid = nodeid; 47636b71a8bSDavid Teigland new_node->addr[0] = new_addr; 47736b71a8bSDavid Teigland new_node->addr_count = 1; 478e125fbebSAlexander Aring new_node->mark = dlm_config.ci_mark; 47936b71a8bSDavid Teigland list_add(&new_node->list, &dlm_node_addrs); 48036b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 48136b71a8bSDavid Teigland return 0; 48236b71a8bSDavid Teigland } 48336b71a8bSDavid Teigland 4844f19d071SAlexander Aring ret = dlm_lowcomms_na_has_addr(na, addr); 4854f19d071SAlexander Aring if (ret) { 4864f19d071SAlexander Aring spin_unlock(&dlm_node_addrs_spin); 4874f19d071SAlexander Aring kfree(new_addr); 4884f19d071SAlexander Aring kfree(new_node); 4894f19d071SAlexander Aring return -EEXIST; 4904f19d071SAlexander Aring } 4914f19d071SAlexander Aring 49236b71a8bSDavid Teigland if (na->addr_count >= DLM_MAX_ADDR_COUNT) { 49336b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 49436b71a8bSDavid Teigland kfree(new_addr); 49536b71a8bSDavid Teigland kfree(new_node); 49636b71a8bSDavid Teigland return -ENOSPC; 49736b71a8bSDavid Teigland } 49836b71a8bSDavid Teigland 49936b71a8bSDavid Teigland na->addr[na->addr_count++] = new_addr; 50036b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 50136b71a8bSDavid Teigland kfree(new_node); 50236b71a8bSDavid Teigland return 0; 50336b71a8bSDavid Teigland } 50436b71a8bSDavid Teigland 5056ed7257bSPatrick Caulfield /* Data available on socket or listen socket received a connect */ 506676d2369SDavid S. Miller static void lowcomms_data_ready(struct sock *sk) 5076ed7257bSPatrick Caulfield { 50893eaadebStsutomu.owa@toshiba.co.jp struct connection *con; 50993eaadebStsutomu.owa@toshiba.co.jp 51093eaadebStsutomu.owa@toshiba.co.jp con = sock2con(sk); 511afb853fbSPatrick Caulfield if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) 5126ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 5136ed7257bSPatrick Caulfield } 5146ed7257bSPatrick Caulfield 515d11ccd45SAlexander Aring static void lowcomms_listen_data_ready(struct sock *sk) 516d11ccd45SAlexander Aring { 5179a4139a7SAlexander Aring if (!dlm_allow_conn) 5189a4139a7SAlexander Aring return; 5199a4139a7SAlexander Aring 520d11ccd45SAlexander Aring queue_work(recv_workqueue, &listen_con.rwork); 521d11ccd45SAlexander Aring } 522d11ccd45SAlexander Aring 5236ed7257bSPatrick Caulfield static void lowcomms_write_space(struct sock *sk) 5246ed7257bSPatrick Caulfield { 52593eaadebStsutomu.owa@toshiba.co.jp struct connection *con; 5266ed7257bSPatrick Caulfield 52793eaadebStsutomu.owa@toshiba.co.jp con = sock2con(sk); 528b36930ddSDavid Miller if (!con) 52992c44605SAlexander Aring return; 530b36930ddSDavid Miller 53119633c7eSAlexander Aring if (!test_and_set_bit(CF_CONNECTED, &con->flags)) { 532dfc020f3SAlexander Aring log_print("connected to node %d", con->nodeid); 53319633c7eSAlexander Aring queue_work(send_workqueue, &con->swork); 53492c44605SAlexander Aring return; 53519633c7eSAlexander Aring } 53619633c7eSAlexander Aring 537b36930ddSDavid Miller clear_bit(SOCK_NOSPACE, &con->sock->flags); 538b36930ddSDavid Miller 539b36930ddSDavid Miller if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) { 540b36930ddSDavid Miller con->sock->sk->sk_write_pending--; 5419cd3e072SEric Dumazet clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags); 542b36930ddSDavid Miller } 543b36930ddSDavid Miller 5446ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 5456ed7257bSPatrick Caulfield } 5466ed7257bSPatrick Caulfield 5476ed7257bSPatrick Caulfield static inline void lowcomms_connect_sock(struct connection *con) 5486ed7257bSPatrick Caulfield { 549063c4c99SLars Marowsky-Bree if (test_bit(CF_CLOSE, &con->flags)) 550063c4c99SLars Marowsky-Bree return; 5516ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 55261d9102bSBob Peterson cond_resched(); 5536ed7257bSPatrick Caulfield } 5546ed7257bSPatrick Caulfield 5556ed7257bSPatrick Caulfield static void lowcomms_state_change(struct sock *sk) 5566ed7257bSPatrick Caulfield { 557ee44b4bcSMarcelo Ricardo Leitner /* SCTP layer is not calling sk_data_ready when the connection 558ee44b4bcSMarcelo Ricardo Leitner * is done, so we catch the signal through here. Also, it 559ee44b4bcSMarcelo Ricardo Leitner * doesn't switch socket state when entering shutdown, so we 560ee44b4bcSMarcelo Ricardo Leitner * skip the write in that case. 561ee44b4bcSMarcelo Ricardo Leitner */ 562ee44b4bcSMarcelo Ricardo Leitner if (sk->sk_shutdown) { 563ee44b4bcSMarcelo Ricardo Leitner if (sk->sk_shutdown == RCV_SHUTDOWN) 564ee44b4bcSMarcelo Ricardo Leitner lowcomms_data_ready(sk); 565ee44b4bcSMarcelo Ricardo Leitner } else if (sk->sk_state == TCP_ESTABLISHED) { 5666ed7257bSPatrick Caulfield lowcomms_write_space(sk); 5676ed7257bSPatrick Caulfield } 568ee44b4bcSMarcelo Ricardo Leitner } 5696ed7257bSPatrick Caulfield 570391fbdc5SChristine Caulfield int dlm_lowcomms_connect_node(int nodeid) 571391fbdc5SChristine Caulfield { 572391fbdc5SChristine Caulfield struct connection *con; 573b38bc9c2SAlexander Aring int idx; 574391fbdc5SChristine Caulfield 575391fbdc5SChristine Caulfield if (nodeid == dlm_our_nodeid()) 576391fbdc5SChristine Caulfield return 0; 577391fbdc5SChristine Caulfield 578b38bc9c2SAlexander Aring idx = srcu_read_lock(&connections_srcu); 579391fbdc5SChristine Caulfield con = nodeid2con(nodeid, GFP_NOFS); 580b38bc9c2SAlexander Aring if (!con) { 581b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 582391fbdc5SChristine Caulfield return -ENOMEM; 583b38bc9c2SAlexander Aring } 584b38bc9c2SAlexander Aring 585391fbdc5SChristine Caulfield lowcomms_connect_sock(con); 586b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 587b38bc9c2SAlexander Aring 588391fbdc5SChristine Caulfield return 0; 589391fbdc5SChristine Caulfield } 590391fbdc5SChristine Caulfield 591e125fbebSAlexander Aring int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark) 592e125fbebSAlexander Aring { 593e125fbebSAlexander Aring struct dlm_node_addr *na; 594e125fbebSAlexander Aring 595e125fbebSAlexander Aring spin_lock(&dlm_node_addrs_spin); 596e125fbebSAlexander Aring na = find_node_addr(nodeid); 597e125fbebSAlexander Aring if (!na) { 598e125fbebSAlexander Aring spin_unlock(&dlm_node_addrs_spin); 599e125fbebSAlexander Aring return -ENOENT; 600e125fbebSAlexander Aring } 601e125fbebSAlexander Aring 602e125fbebSAlexander Aring na->mark = mark; 603e125fbebSAlexander Aring spin_unlock(&dlm_node_addrs_spin); 604e125fbebSAlexander Aring 605e125fbebSAlexander Aring return 0; 606e125fbebSAlexander Aring } 607e125fbebSAlexander Aring 608b3a5bbfdSBob Peterson static void lowcomms_error_report(struct sock *sk) 609b3a5bbfdSBob Peterson { 610b81171cbSBob Peterson struct connection *con; 611b81171cbSBob Peterson void (*orig_report)(struct sock *) = NULL; 6124c3d9057SAlexander Aring struct inet_sock *inet; 613b3a5bbfdSBob Peterson 614b81171cbSBob Peterson con = sock2con(sk); 615b81171cbSBob Peterson if (con == NULL) 616b81171cbSBob Peterson goto out; 617b81171cbSBob Peterson 618cc661fc9SBob Peterson orig_report = listen_sock.sk_error_report; 619b3a5bbfdSBob Peterson 6204c3d9057SAlexander Aring inet = inet_sk(sk); 6214c3d9057SAlexander Aring switch (sk->sk_family) { 6224c3d9057SAlexander Aring case AF_INET: 623b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 6244c3d9057SAlexander Aring "sending to node %d at %pI4, dport %d, " 625b3a5bbfdSBob Peterson "sk_err=%d/%d\n", dlm_our_nodeid(), 6264c3d9057SAlexander Aring con->nodeid, &inet->inet_daddr, 6274c3d9057SAlexander Aring ntohs(inet->inet_dport), sk->sk_err, 628b3a5bbfdSBob Peterson sk->sk_err_soft); 6294c3d9057SAlexander Aring break; 6301b9beda8SAlexander Aring #if IS_ENABLED(CONFIG_IPV6) 6314c3d9057SAlexander Aring case AF_INET6: 632b3a5bbfdSBob Peterson printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 6334c3d9057SAlexander Aring "sending to node %d at %pI6c, " 6344c3d9057SAlexander Aring "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(), 6354c3d9057SAlexander Aring con->nodeid, &sk->sk_v6_daddr, 6364c3d9057SAlexander Aring ntohs(inet->inet_dport), sk->sk_err, 637b3a5bbfdSBob Peterson sk->sk_err_soft); 6384c3d9057SAlexander Aring break; 6391b9beda8SAlexander Aring #endif 6404c3d9057SAlexander Aring default: 6414c3d9057SAlexander Aring printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 6424c3d9057SAlexander Aring "invalid socket family %d set, " 6434c3d9057SAlexander Aring "sk_err=%d/%d\n", dlm_our_nodeid(), 6444c3d9057SAlexander Aring sk->sk_family, sk->sk_err, sk->sk_err_soft); 6454c3d9057SAlexander Aring goto out; 646b3a5bbfdSBob Peterson } 647ba868d9dSAlexander Aring 648ba868d9dSAlexander Aring /* below sendcon only handling */ 649ba868d9dSAlexander Aring if (test_bit(CF_IS_OTHERCON, &con->flags)) 650ba868d9dSAlexander Aring con = con->sendcon; 651ba868d9dSAlexander Aring 652ba868d9dSAlexander Aring switch (sk->sk_err) { 653ba868d9dSAlexander Aring case ECONNREFUSED: 654ba868d9dSAlexander Aring set_bit(CF_DELAY_CONNECT, &con->flags); 655ba868d9dSAlexander Aring break; 656ba868d9dSAlexander Aring default: 657ba868d9dSAlexander Aring break; 658ba868d9dSAlexander Aring } 659ba868d9dSAlexander Aring 660ba868d9dSAlexander Aring if (!test_and_set_bit(CF_RECONNECT, &con->flags)) 661ba868d9dSAlexander Aring queue_work(send_workqueue, &con->swork); 662ba868d9dSAlexander Aring 663b81171cbSBob Peterson out: 664b81171cbSBob Peterson if (orig_report) 665b81171cbSBob Peterson orig_report(sk); 666b81171cbSBob Peterson } 667b81171cbSBob Peterson 668b81171cbSBob Peterson /* Note: sk_callback_lock must be locked before calling this function. */ 669cc661fc9SBob Peterson static void save_listen_callbacks(struct socket *sock) 670b81171cbSBob Peterson { 671cc661fc9SBob Peterson struct sock *sk = sock->sk; 672cc661fc9SBob Peterson 673cc661fc9SBob Peterson listen_sock.sk_data_ready = sk->sk_data_ready; 674cc661fc9SBob Peterson listen_sock.sk_state_change = sk->sk_state_change; 675cc661fc9SBob Peterson listen_sock.sk_write_space = sk->sk_write_space; 676cc661fc9SBob Peterson listen_sock.sk_error_report = sk->sk_error_report; 677b81171cbSBob Peterson } 678b81171cbSBob Peterson 679cc661fc9SBob Peterson static void restore_callbacks(struct socket *sock) 680b81171cbSBob Peterson { 681cc661fc9SBob Peterson struct sock *sk = sock->sk; 682cc661fc9SBob Peterson 68392c44605SAlexander Aring lock_sock(sk); 684b81171cbSBob Peterson sk->sk_user_data = NULL; 685cc661fc9SBob Peterson sk->sk_data_ready = listen_sock.sk_data_ready; 686cc661fc9SBob Peterson sk->sk_state_change = listen_sock.sk_state_change; 687cc661fc9SBob Peterson sk->sk_write_space = listen_sock.sk_write_space; 688cc661fc9SBob Peterson sk->sk_error_report = listen_sock.sk_error_report; 68992c44605SAlexander Aring release_sock(sk); 690b3a5bbfdSBob Peterson } 691b3a5bbfdSBob Peterson 692d11ccd45SAlexander Aring static void add_listen_sock(struct socket *sock, struct listen_connection *con) 693d11ccd45SAlexander Aring { 694d11ccd45SAlexander Aring struct sock *sk = sock->sk; 695d11ccd45SAlexander Aring 69692c44605SAlexander Aring lock_sock(sk); 697d11ccd45SAlexander Aring save_listen_callbacks(sock); 698d11ccd45SAlexander Aring con->sock = sock; 699d11ccd45SAlexander Aring 700d11ccd45SAlexander Aring sk->sk_user_data = con; 701d11ccd45SAlexander Aring sk->sk_allocation = GFP_NOFS; 702d11ccd45SAlexander Aring /* Install a data_ready callback */ 703d11ccd45SAlexander Aring sk->sk_data_ready = lowcomms_listen_data_ready; 70492c44605SAlexander Aring release_sock(sk); 705d11ccd45SAlexander Aring } 706d11ccd45SAlexander Aring 7076ed7257bSPatrick Caulfield /* Make a socket active */ 708988419a9Stsutomu.owa@toshiba.co.jp static void add_sock(struct socket *sock, struct connection *con) 7096ed7257bSPatrick Caulfield { 710b81171cbSBob Peterson struct sock *sk = sock->sk; 711b81171cbSBob Peterson 71292c44605SAlexander Aring lock_sock(sk); 7136ed7257bSPatrick Caulfield con->sock = sock; 7146ed7257bSPatrick Caulfield 715b81171cbSBob Peterson sk->sk_user_data = con; 7166ed7257bSPatrick Caulfield /* Install a data_ready callback */ 717b81171cbSBob Peterson sk->sk_data_ready = lowcomms_data_ready; 718b81171cbSBob Peterson sk->sk_write_space = lowcomms_write_space; 719b81171cbSBob Peterson sk->sk_state_change = lowcomms_state_change; 720b81171cbSBob Peterson sk->sk_allocation = GFP_NOFS; 721b81171cbSBob Peterson sk->sk_error_report = lowcomms_error_report; 72292c44605SAlexander Aring release_sock(sk); 7236ed7257bSPatrick Caulfield } 7246ed7257bSPatrick Caulfield 7256ed7257bSPatrick Caulfield /* Add the port number to an IPv6 or 4 sockaddr and return the address 7266ed7257bSPatrick Caulfield length */ 7276ed7257bSPatrick Caulfield static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, 7286ed7257bSPatrick Caulfield int *addr_len) 7296ed7257bSPatrick Caulfield { 7306ed7257bSPatrick Caulfield saddr->ss_family = dlm_local_addr[0]->ss_family; 7316ed7257bSPatrick Caulfield if (saddr->ss_family == AF_INET) { 7326ed7257bSPatrick Caulfield struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; 7336ed7257bSPatrick Caulfield in4_addr->sin_port = cpu_to_be16(port); 7346ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in); 7356ed7257bSPatrick Caulfield memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); 7366ed7257bSPatrick Caulfield } else { 7376ed7257bSPatrick Caulfield struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; 7386ed7257bSPatrick Caulfield in6_addr->sin6_port = cpu_to_be16(port); 7396ed7257bSPatrick Caulfield *addr_len = sizeof(struct sockaddr_in6); 7406ed7257bSPatrick Caulfield } 74101c8cab2SPatrick Caulfield memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); 7426ed7257bSPatrick Caulfield } 7436ed7257bSPatrick Caulfield 744706474fbSAlexander Aring static void dlm_page_release(struct kref *kref) 745706474fbSAlexander Aring { 746706474fbSAlexander Aring struct writequeue_entry *e = container_of(kref, struct writequeue_entry, 747706474fbSAlexander Aring ref); 748706474fbSAlexander Aring 749706474fbSAlexander Aring __free_page(e->page); 7503af2326cSAlexander Aring dlm_free_writequeue(e); 751706474fbSAlexander Aring } 752706474fbSAlexander Aring 753706474fbSAlexander Aring static void dlm_msg_release(struct kref *kref) 754706474fbSAlexander Aring { 755706474fbSAlexander Aring struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref); 756706474fbSAlexander Aring 757706474fbSAlexander Aring kref_put(&msg->entry->ref, dlm_page_release); 758e4dc81edSAlexander Aring dlm_free_msg(msg); 759706474fbSAlexander Aring } 760706474fbSAlexander Aring 761706474fbSAlexander Aring static void free_entry(struct writequeue_entry *e) 762706474fbSAlexander Aring { 763706474fbSAlexander Aring struct dlm_msg *msg, *tmp; 764706474fbSAlexander Aring 765706474fbSAlexander Aring list_for_each_entry_safe(msg, tmp, &e->msgs, list) { 766706474fbSAlexander Aring if (msg->orig_msg) { 767706474fbSAlexander Aring msg->orig_msg->retransmit = false; 768706474fbSAlexander Aring kref_put(&msg->orig_msg->ref, dlm_msg_release); 769706474fbSAlexander Aring } 770706474fbSAlexander Aring 771706474fbSAlexander Aring list_del(&msg->list); 772706474fbSAlexander Aring kref_put(&msg->ref, dlm_msg_release); 773706474fbSAlexander Aring } 774706474fbSAlexander Aring 775706474fbSAlexander Aring list_del(&e->list); 776706474fbSAlexander Aring atomic_dec(&e->con->writequeue_cnt); 777706474fbSAlexander Aring kref_put(&e->ref, dlm_page_release); 778706474fbSAlexander Aring } 779706474fbSAlexander Aring 780d11ccd45SAlexander Aring static void dlm_close_sock(struct socket **sock) 781d11ccd45SAlexander Aring { 782d11ccd45SAlexander Aring if (*sock) { 783d11ccd45SAlexander Aring restore_callbacks(*sock); 784d11ccd45SAlexander Aring sock_release(*sock); 785d11ccd45SAlexander Aring *sock = NULL; 786d11ccd45SAlexander Aring } 787d11ccd45SAlexander Aring } 788d11ccd45SAlexander Aring 7896ed7257bSPatrick Caulfield /* Close a remote connection and tidy up */ 7900d737a8cSMarcelo Ricardo Leitner static void close_connection(struct connection *con, bool and_other, 7910d737a8cSMarcelo Ricardo Leitner bool tx, bool rx) 7926ed7257bSPatrick Caulfield { 793b2a66629Stsutomu.owa@toshiba.co.jp bool closing = test_and_set_bit(CF_CLOSING, &con->flags); 794706474fbSAlexander Aring struct writequeue_entry *e; 795b2a66629Stsutomu.owa@toshiba.co.jp 7960aa18464Stsutomu.owa@toshiba.co.jp if (tx && !closing && cancel_work_sync(&con->swork)) { 7970d737a8cSMarcelo Ricardo Leitner log_print("canceled swork for node %d", con->nodeid); 7980aa18464Stsutomu.owa@toshiba.co.jp clear_bit(CF_WRITE_PENDING, &con->flags); 7990aa18464Stsutomu.owa@toshiba.co.jp } 8000aa18464Stsutomu.owa@toshiba.co.jp if (rx && !closing && cancel_work_sync(&con->rwork)) { 8010d737a8cSMarcelo Ricardo Leitner log_print("canceled rwork for node %d", con->nodeid); 8020aa18464Stsutomu.owa@toshiba.co.jp clear_bit(CF_READ_PENDING, &con->flags); 8030aa18464Stsutomu.owa@toshiba.co.jp } 8046ed7257bSPatrick Caulfield 8050d737a8cSMarcelo Ricardo Leitner mutex_lock(&con->sock_mutex); 806d11ccd45SAlexander Aring dlm_close_sock(&con->sock); 807d11ccd45SAlexander Aring 8086ed7257bSPatrick Caulfield if (con->othercon && and_other) { 8096ed7257bSPatrick Caulfield /* Will only re-enter once. */ 810c6aa00e3SAlexander Aring close_connection(con->othercon, false, tx, rx); 8116ed7257bSPatrick Caulfield } 8129e5f2825SPatrick Caulfield 813706474fbSAlexander Aring /* if we send a writequeue entry only a half way, we drop the 814706474fbSAlexander Aring * whole entry because reconnection and that we not start of the 815706474fbSAlexander Aring * middle of a msg which will confuse the other end. 816706474fbSAlexander Aring * 817706474fbSAlexander Aring * we can always drop messages because retransmits, but what we 818706474fbSAlexander Aring * cannot allow is to transmit half messages which may be processed 819706474fbSAlexander Aring * at the other side. 820706474fbSAlexander Aring * 821706474fbSAlexander Aring * our policy is to start on a clean state when disconnects, we don't 822706474fbSAlexander Aring * know what's send/received on transport layer in this case. 823706474fbSAlexander Aring */ 824706474fbSAlexander Aring spin_lock(&con->writequeue_lock); 825706474fbSAlexander Aring if (!list_empty(&con->writequeue)) { 826706474fbSAlexander Aring e = list_first_entry(&con->writequeue, struct writequeue_entry, 827706474fbSAlexander Aring list); 828706474fbSAlexander Aring if (e->dirty) 829706474fbSAlexander Aring free_entry(e); 830706474fbSAlexander Aring } 831706474fbSAlexander Aring spin_unlock(&con->writequeue_lock); 832706474fbSAlexander Aring 8334798cbbfSAlexander Aring con->rx_leftover = 0; 8346ed7257bSPatrick Caulfield con->retries = 0; 835052849beSAlexander Aring clear_bit(CF_APP_LIMITED, &con->flags); 83619633c7eSAlexander Aring clear_bit(CF_CONNECTED, &con->flags); 837ba868d9dSAlexander Aring clear_bit(CF_DELAY_CONNECT, &con->flags); 838ba868d9dSAlexander Aring clear_bit(CF_RECONNECT, &con->flags); 8398aa31cbfSAlexander Aring clear_bit(CF_EOF, &con->flags); 8406ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 841b2a66629Stsutomu.owa@toshiba.co.jp clear_bit(CF_CLOSING, &con->flags); 8426ed7257bSPatrick Caulfield } 8436ed7257bSPatrick Caulfield 844055923bfSAlexander Aring static void shutdown_connection(struct connection *con) 845055923bfSAlexander Aring { 846055923bfSAlexander Aring int ret; 847055923bfSAlexander Aring 848eec054b5SAlexander Aring flush_work(&con->swork); 849055923bfSAlexander Aring 850055923bfSAlexander Aring mutex_lock(&con->sock_mutex); 851055923bfSAlexander Aring /* nothing to shutdown */ 852055923bfSAlexander Aring if (!con->sock) { 853055923bfSAlexander Aring mutex_unlock(&con->sock_mutex); 854055923bfSAlexander Aring return; 855055923bfSAlexander Aring } 856055923bfSAlexander Aring 857055923bfSAlexander Aring set_bit(CF_SHUTDOWN, &con->flags); 858055923bfSAlexander Aring ret = kernel_sock_shutdown(con->sock, SHUT_WR); 859055923bfSAlexander Aring mutex_unlock(&con->sock_mutex); 860055923bfSAlexander Aring if (ret) { 861055923bfSAlexander Aring log_print("Connection %p failed to shutdown: %d will force close", 862055923bfSAlexander Aring con, ret); 863055923bfSAlexander Aring goto force_close; 864055923bfSAlexander Aring } else { 865055923bfSAlexander Aring ret = wait_event_timeout(con->shutdown_wait, 866055923bfSAlexander Aring !test_bit(CF_SHUTDOWN, &con->flags), 867055923bfSAlexander Aring DLM_SHUTDOWN_WAIT_TIMEOUT); 868055923bfSAlexander Aring if (ret == 0) { 869055923bfSAlexander Aring log_print("Connection %p shutdown timed out, will force close", 870055923bfSAlexander Aring con); 871055923bfSAlexander Aring goto force_close; 872055923bfSAlexander Aring } 873055923bfSAlexander Aring } 874055923bfSAlexander Aring 875055923bfSAlexander Aring return; 876055923bfSAlexander Aring 877055923bfSAlexander Aring force_close: 878055923bfSAlexander Aring clear_bit(CF_SHUTDOWN, &con->flags); 879055923bfSAlexander Aring close_connection(con, false, true, true); 880055923bfSAlexander Aring } 881055923bfSAlexander Aring 882055923bfSAlexander Aring static void dlm_tcp_shutdown(struct connection *con) 883055923bfSAlexander Aring { 884055923bfSAlexander Aring if (con->othercon) 885055923bfSAlexander Aring shutdown_connection(con->othercon); 886055923bfSAlexander Aring shutdown_connection(con); 887055923bfSAlexander Aring } 888055923bfSAlexander Aring 8894798cbbfSAlexander Aring static int con_realloc_receive_buf(struct connection *con, int newlen) 8904798cbbfSAlexander Aring { 8914798cbbfSAlexander Aring unsigned char *newbuf; 8924798cbbfSAlexander Aring 8934798cbbfSAlexander Aring newbuf = kmalloc(newlen, GFP_NOFS); 8944798cbbfSAlexander Aring if (!newbuf) 8954798cbbfSAlexander Aring return -ENOMEM; 8964798cbbfSAlexander Aring 8974798cbbfSAlexander Aring /* copy any leftover from last receive */ 8984798cbbfSAlexander Aring if (con->rx_leftover) 8994798cbbfSAlexander Aring memmove(newbuf, con->rx_buf, con->rx_leftover); 9004798cbbfSAlexander Aring 9014798cbbfSAlexander Aring /* swap to new buffer space */ 9024798cbbfSAlexander Aring kfree(con->rx_buf); 9034798cbbfSAlexander Aring con->rx_buflen = newlen; 9044798cbbfSAlexander Aring con->rx_buf = newbuf; 9054798cbbfSAlexander Aring 9064798cbbfSAlexander Aring return 0; 9074798cbbfSAlexander Aring } 9084798cbbfSAlexander Aring 9096ed7257bSPatrick Caulfield /* Data received from remote end */ 9106ed7257bSPatrick Caulfield static int receive_from_sock(struct connection *con) 9116ed7257bSPatrick Caulfield { 9124798cbbfSAlexander Aring struct msghdr msg; 9134798cbbfSAlexander Aring struct kvec iov; 9144798cbbfSAlexander Aring int ret, buflen; 9156ed7257bSPatrick Caulfield 9166ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 9176ed7257bSPatrick Caulfield 9186ed7257bSPatrick Caulfield if (con->sock == NULL) { 9196ed7257bSPatrick Caulfield ret = -EAGAIN; 9206ed7257bSPatrick Caulfield goto out_close; 9216ed7257bSPatrick Caulfield } 9224798cbbfSAlexander Aring 9234798cbbfSAlexander Aring /* realloc if we get new buffer size to read out */ 9244798cbbfSAlexander Aring buflen = dlm_config.ci_buffer_size; 9254798cbbfSAlexander Aring if (con->rx_buflen != buflen && con->rx_leftover <= buflen) { 9264798cbbfSAlexander Aring ret = con_realloc_receive_buf(con, buflen); 9274798cbbfSAlexander Aring if (ret < 0) 9286ed7257bSPatrick Caulfield goto out_resched; 9296ed7257bSPatrick Caulfield } 9306ed7257bSPatrick Caulfield 93162699b3fSAlexander Aring for (;;) { 9324798cbbfSAlexander Aring /* calculate new buffer parameter regarding last receive and 9334798cbbfSAlexander Aring * possible leftover bytes 9346ed7257bSPatrick Caulfield */ 9354798cbbfSAlexander Aring iov.iov_base = con->rx_buf + con->rx_leftover; 9364798cbbfSAlexander Aring iov.iov_len = con->rx_buflen - con->rx_leftover; 9376ed7257bSPatrick Caulfield 9384798cbbfSAlexander Aring memset(&msg, 0, sizeof(msg)); 9394798cbbfSAlexander Aring msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; 9404798cbbfSAlexander Aring ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len, 9414798cbbfSAlexander Aring msg.msg_flags); 94292732376SAlexander Aring trace_dlm_recv(con->nodeid, ret); 94362699b3fSAlexander Aring if (ret == -EAGAIN) 94462699b3fSAlexander Aring break; 94562699b3fSAlexander Aring else if (ret <= 0) 9466ed7257bSPatrick Caulfield goto out_close; 9476ed7257bSPatrick Caulfield 9484798cbbfSAlexander Aring /* new buflen according readed bytes and leftover from last receive */ 9494798cbbfSAlexander Aring buflen = ret + con->rx_leftover; 9504798cbbfSAlexander Aring ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen); 9514798cbbfSAlexander Aring if (ret < 0) 9524798cbbfSAlexander Aring goto out_close; 9536ed7257bSPatrick Caulfield 9544798cbbfSAlexander Aring /* calculate leftover bytes from process and put it into begin of 9554798cbbfSAlexander Aring * the receive buffer, so next receive we have the full message 9564798cbbfSAlexander Aring * at the start address of the receive buffer. 9574798cbbfSAlexander Aring */ 9584798cbbfSAlexander Aring con->rx_leftover = buflen - ret; 9594798cbbfSAlexander Aring if (con->rx_leftover) { 9604798cbbfSAlexander Aring memmove(con->rx_buf, con->rx_buf + ret, 9614798cbbfSAlexander Aring con->rx_leftover); 9626ed7257bSPatrick Caulfield } 96362699b3fSAlexander Aring } 9644798cbbfSAlexander Aring 965b97f8525SAlexander Aring dlm_midcomms_receive_done(con->nodeid); 9666ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 9676ed7257bSPatrick Caulfield return 0; 9686ed7257bSPatrick Caulfield 9696ed7257bSPatrick Caulfield out_resched: 9706ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) 9716ed7257bSPatrick Caulfield queue_work(recv_workqueue, &con->rwork); 9726ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 9736ed7257bSPatrick Caulfield return -EAGAIN; 9746ed7257bSPatrick Caulfield 9756ed7257bSPatrick Caulfield out_close: 976055923bfSAlexander Aring if (ret == 0) { 977055923bfSAlexander Aring log_print("connection %p got EOF from %d", 978055923bfSAlexander Aring con, con->nodeid); 9798aa31cbfSAlexander Aring 980a66c008cSAlexander Aring if (dlm_proto_ops->eof_condition && 981a66c008cSAlexander Aring dlm_proto_ops->eof_condition(con)) { 9828aa31cbfSAlexander Aring set_bit(CF_EOF, &con->flags); 9838aa31cbfSAlexander Aring mutex_unlock(&con->sock_mutex); 9848aa31cbfSAlexander Aring } else { 9858aa31cbfSAlexander Aring mutex_unlock(&con->sock_mutex); 9868aa31cbfSAlexander Aring close_connection(con, false, true, false); 9878aa31cbfSAlexander Aring 988055923bfSAlexander Aring /* handling for tcp shutdown */ 989055923bfSAlexander Aring clear_bit(CF_SHUTDOWN, &con->flags); 990055923bfSAlexander Aring wake_up(&con->shutdown_wait); 9918aa31cbfSAlexander Aring } 9928aa31cbfSAlexander Aring 993055923bfSAlexander Aring /* signal to breaking receive worker */ 994055923bfSAlexander Aring ret = -1; 9958aa31cbfSAlexander Aring } else { 9968aa31cbfSAlexander Aring mutex_unlock(&con->sock_mutex); 9976ed7257bSPatrick Caulfield } 9986ed7257bSPatrick Caulfield return ret; 9996ed7257bSPatrick Caulfield } 10006ed7257bSPatrick Caulfield 10016ed7257bSPatrick Caulfield /* Listening socket is busy, accept a connection */ 1002d11ccd45SAlexander Aring static int accept_from_sock(struct listen_connection *con) 10036ed7257bSPatrick Caulfield { 10046ed7257bSPatrick Caulfield int result; 10056ed7257bSPatrick Caulfield struct sockaddr_storage peeraddr; 10066ed7257bSPatrick Caulfield struct socket *newsock; 1007b38bc9c2SAlexander Aring int len, idx; 10086ed7257bSPatrick Caulfield int nodeid; 10096ed7257bSPatrick Caulfield struct connection *newcon; 10106ed7257bSPatrick Caulfield struct connection *addcon; 10113f78cd7dSAlexander Aring unsigned int mark; 10126ed7257bSPatrick Caulfield 1013d11ccd45SAlexander Aring if (!con->sock) 10143421fb15Stsutomu.owa@toshiba.co.jp return -ENOTCONN; 10156ed7257bSPatrick Caulfield 10163421fb15Stsutomu.owa@toshiba.co.jp result = kernel_accept(con->sock, &newsock, O_NONBLOCK); 10176ed7257bSPatrick Caulfield if (result < 0) 10186ed7257bSPatrick Caulfield goto accept_err; 10196ed7257bSPatrick Caulfield 10206ed7257bSPatrick Caulfield /* Get the connected socket's peer */ 10216ed7257bSPatrick Caulfield memset(&peeraddr, 0, sizeof(peeraddr)); 10229b2c45d4SDenys Vlasenko len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2); 10239b2c45d4SDenys Vlasenko if (len < 0) { 10246ed7257bSPatrick Caulfield result = -ECONNABORTED; 10256ed7257bSPatrick Caulfield goto accept_err; 10266ed7257bSPatrick Caulfield } 10276ed7257bSPatrick Caulfield 10286ed7257bSPatrick Caulfield /* Get the new node's NODEID */ 10296ed7257bSPatrick Caulfield make_sockaddr(&peeraddr, 0, &len); 1030e125fbebSAlexander Aring if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) { 1031feae43f8SAlexander Aring switch (peeraddr.ss_family) { 1032feae43f8SAlexander Aring case AF_INET: { 1033feae43f8SAlexander Aring struct sockaddr_in *sin = (struct sockaddr_in *)&peeraddr; 1034feae43f8SAlexander Aring 1035feae43f8SAlexander Aring log_print("connect from non cluster IPv4 node %pI4", 1036feae43f8SAlexander Aring &sin->sin_addr); 1037feae43f8SAlexander Aring break; 1038feae43f8SAlexander Aring } 1039feae43f8SAlexander Aring #if IS_ENABLED(CONFIG_IPV6) 1040feae43f8SAlexander Aring case AF_INET6: { 1041feae43f8SAlexander Aring struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&peeraddr; 1042feae43f8SAlexander Aring 1043feae43f8SAlexander Aring log_print("connect from non cluster IPv6 node %pI6c", 1044feae43f8SAlexander Aring &sin6->sin6_addr); 1045feae43f8SAlexander Aring break; 1046feae43f8SAlexander Aring } 1047feae43f8SAlexander Aring #endif 1048feae43f8SAlexander Aring default: 1049feae43f8SAlexander Aring log_print("invalid family from non cluster node"); 1050feae43f8SAlexander Aring break; 1051feae43f8SAlexander Aring } 1052feae43f8SAlexander Aring 10536ed7257bSPatrick Caulfield sock_release(newsock); 10546ed7257bSPatrick Caulfield return -1; 10556ed7257bSPatrick Caulfield } 10566ed7257bSPatrick Caulfield 10576ed7257bSPatrick Caulfield log_print("got connection from %d", nodeid); 10586ed7257bSPatrick Caulfield 10596ed7257bSPatrick Caulfield /* Check to see if we already have a connection to this node. This 10606ed7257bSPatrick Caulfield * could happen if the two nodes initiate a connection at roughly 10616ed7257bSPatrick Caulfield * the same time and the connections cross on the wire. 10626ed7257bSPatrick Caulfield * In this case we store the incoming one in "othercon" 10636ed7257bSPatrick Caulfield */ 1064b38bc9c2SAlexander Aring idx = srcu_read_lock(&connections_srcu); 1065748285ccSDavid Teigland newcon = nodeid2con(nodeid, GFP_NOFS); 10666ed7257bSPatrick Caulfield if (!newcon) { 1067b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 10686ed7257bSPatrick Caulfield result = -ENOMEM; 10696ed7257bSPatrick Caulfield goto accept_err; 10706ed7257bSPatrick Caulfield } 1071d11ccd45SAlexander Aring 1072e125fbebSAlexander Aring sock_set_mark(newsock->sk, mark); 1073e125fbebSAlexander Aring 1074d11ccd45SAlexander Aring mutex_lock(&newcon->sock_mutex); 10756ed7257bSPatrick Caulfield if (newcon->sock) { 10766ed7257bSPatrick Caulfield struct connection *othercon = newcon->othercon; 10776ed7257bSPatrick Caulfield 10786ed7257bSPatrick Caulfield if (!othercon) { 1079a47666ebSAlexander Aring othercon = kzalloc(sizeof(*othercon), GFP_NOFS); 10806ed7257bSPatrick Caulfield if (!othercon) { 1081617e82e1SDavid Teigland log_print("failed to allocate incoming socket"); 10826ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 1083b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 10846ed7257bSPatrick Caulfield result = -ENOMEM; 10856ed7257bSPatrick Caulfield goto accept_err; 10866ed7257bSPatrick Caulfield } 10874798cbbfSAlexander Aring 10886cde210aSAlexander Aring result = dlm_con_init(othercon, nodeid); 10896cde210aSAlexander Aring if (result < 0) { 10904798cbbfSAlexander Aring kfree(othercon); 10912fd8db2dSYang Yingliang mutex_unlock(&newcon->sock_mutex); 1092b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 10934798cbbfSAlexander Aring goto accept_err; 10944798cbbfSAlexander Aring } 10954798cbbfSAlexander Aring 1096e9a470acSAlexander Aring lockdep_set_subclass(&othercon->sock_mutex, 1); 10977443bc96SAlexander Aring set_bit(CF_IS_OTHERCON, &othercon->flags); 10986cde210aSAlexander Aring newcon->othercon = othercon; 1099ba868d9dSAlexander Aring othercon->sendcon = newcon; 1100ba3ab3caSAlexander Aring } else { 1101ba3ab3caSAlexander Aring /* close other sock con if we have something new */ 1102ba3ab3caSAlexander Aring close_connection(othercon, false, true, false); 110361d96be0SPatrick Caulfield } 1104ba3ab3caSAlexander Aring 1105e9a470acSAlexander Aring mutex_lock(&othercon->sock_mutex); 1106988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, othercon); 11076ed7257bSPatrick Caulfield addcon = othercon; 1108c7355827Stsutomu.owa@toshiba.co.jp mutex_unlock(&othercon->sock_mutex); 11096ed7257bSPatrick Caulfield } 11106ed7257bSPatrick Caulfield else { 11113735b4b9SBob Peterson /* accept copies the sk after we've saved the callbacks, so we 11123735b4b9SBob Peterson don't want to save them a second time or comm errors will 11133735b4b9SBob Peterson result in calling sk_error_report recursively. */ 1114988419a9Stsutomu.owa@toshiba.co.jp add_sock(newsock, newcon); 11156ed7257bSPatrick Caulfield addcon = newcon; 11166ed7257bSPatrick Caulfield } 11176ed7257bSPatrick Caulfield 1118b30a624fSAlexander Aring set_bit(CF_CONNECTED, &addcon->flags); 11196ed7257bSPatrick Caulfield mutex_unlock(&newcon->sock_mutex); 11206ed7257bSPatrick Caulfield 11216ed7257bSPatrick Caulfield /* 11226ed7257bSPatrick Caulfield * Add it to the active queue in case we got data 112325985edcSLucas De Marchi * between processing the accept adding the socket 11246ed7257bSPatrick Caulfield * to the read_sockets list 11256ed7257bSPatrick Caulfield */ 11266ed7257bSPatrick Caulfield if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) 11276ed7257bSPatrick Caulfield queue_work(recv_workqueue, &addcon->rwork); 11286ed7257bSPatrick Caulfield 1129b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 1130b38bc9c2SAlexander Aring 11316ed7257bSPatrick Caulfield return 0; 11326ed7257bSPatrick Caulfield 11336ed7257bSPatrick Caulfield accept_err: 11343421fb15Stsutomu.owa@toshiba.co.jp if (newsock) 11356ed7257bSPatrick Caulfield sock_release(newsock); 11366ed7257bSPatrick Caulfield 11376ed7257bSPatrick Caulfield if (result != -EAGAIN) 1138617e82e1SDavid Teigland log_print("error accepting connection from node: %d", result); 11396ed7257bSPatrick Caulfield return result; 11406ed7257bSPatrick Caulfield } 11416ed7257bSPatrick Caulfield 11425d689871SMike Christie /* 11435d689871SMike Christie * writequeue_entry_complete - try to delete and free write queue entry 11445d689871SMike Christie * @e: write queue entry to try to delete 11455d689871SMike Christie * @completed: bytes completed 11465d689871SMike Christie * 11475d689871SMike Christie * writequeue_lock must be held. 11485d689871SMike Christie */ 11495d689871SMike Christie static void writequeue_entry_complete(struct writequeue_entry *e, int completed) 11505d689871SMike Christie { 11515d689871SMike Christie e->offset += completed; 11525d689871SMike Christie e->len -= completed; 1153706474fbSAlexander Aring /* signal that page was half way transmitted */ 1154706474fbSAlexander Aring e->dirty = true; 11555d689871SMike Christie 11568f2dc78dSAlexander Aring if (e->len == 0 && e->users == 0) 11575d689871SMike Christie free_entry(e); 11585d689871SMike Christie } 11595d689871SMike Christie 1160ee44b4bcSMarcelo Ricardo Leitner /* 1161ee44b4bcSMarcelo Ricardo Leitner * sctp_bind_addrs - bind a SCTP socket to all our addresses 1162ee44b4bcSMarcelo Ricardo Leitner */ 116313004e8aSAlexander Aring static int sctp_bind_addrs(struct socket *sock, uint16_t port) 1164ee44b4bcSMarcelo Ricardo Leitner { 1165ee44b4bcSMarcelo Ricardo Leitner struct sockaddr_storage localaddr; 1166c0425a42SChristoph Hellwig struct sockaddr *addr = (struct sockaddr *)&localaddr; 1167ee44b4bcSMarcelo Ricardo Leitner int i, addr_len, result = 0; 1168ee44b4bcSMarcelo Ricardo Leitner 1169ee44b4bcSMarcelo Ricardo Leitner for (i = 0; i < dlm_local_count; i++) { 1170ee44b4bcSMarcelo Ricardo Leitner memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); 1171ee44b4bcSMarcelo Ricardo Leitner make_sockaddr(&localaddr, port, &addr_len); 1172ee44b4bcSMarcelo Ricardo Leitner 1173ee44b4bcSMarcelo Ricardo Leitner if (!i) 117413004e8aSAlexander Aring result = kernel_bind(sock, addr, addr_len); 1175ee44b4bcSMarcelo Ricardo Leitner else 117613004e8aSAlexander Aring result = sock_bind_add(sock->sk, addr, addr_len); 1177ee44b4bcSMarcelo Ricardo Leitner 1178ee44b4bcSMarcelo Ricardo Leitner if (result < 0) { 1179ee44b4bcSMarcelo Ricardo Leitner log_print("Can't bind to %d addr number %d, %d.\n", 1180ee44b4bcSMarcelo Ricardo Leitner port, i + 1, result); 1181ee44b4bcSMarcelo Ricardo Leitner break; 1182ee44b4bcSMarcelo Ricardo Leitner } 1183ee44b4bcSMarcelo Ricardo Leitner } 1184ee44b4bcSMarcelo Ricardo Leitner return result; 1185ee44b4bcSMarcelo Ricardo Leitner } 1186ee44b4bcSMarcelo Ricardo Leitner 11876ed7257bSPatrick Caulfield /* Get local addresses */ 11886ed7257bSPatrick Caulfield static void init_local(void) 11896ed7257bSPatrick Caulfield { 11906ed7257bSPatrick Caulfield struct sockaddr_storage sas, *addr; 11916ed7257bSPatrick Caulfield int i; 11926ed7257bSPatrick Caulfield 119330d3a237SPatrick Caulfield dlm_local_count = 0; 11941b189b88SDavid Teigland for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) { 11956ed7257bSPatrick Caulfield if (dlm_our_addr(&sas, i)) 11966ed7257bSPatrick Caulfield break; 11976ed7257bSPatrick Caulfield 11985c93f56fSAmitoj Kaur Chawla addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS); 11996ed7257bSPatrick Caulfield if (!addr) 12006ed7257bSPatrick Caulfield break; 12016ed7257bSPatrick Caulfield dlm_local_addr[dlm_local_count++] = addr; 12026ed7257bSPatrick Caulfield } 12036ed7257bSPatrick Caulfield } 12046ed7257bSPatrick Caulfield 1205043697f0SAlexander Aring static void deinit_local(void) 1206043697f0SAlexander Aring { 1207043697f0SAlexander Aring int i; 1208043697f0SAlexander Aring 1209043697f0SAlexander Aring for (i = 0; i < dlm_local_count; i++) 1210043697f0SAlexander Aring kfree(dlm_local_addr[i]); 1211043697f0SAlexander Aring } 1212043697f0SAlexander Aring 1213be3b0400SAlexander Aring static struct writequeue_entry *new_writequeue_entry(struct connection *con) 12146ed7257bSPatrick Caulfield { 12156ed7257bSPatrick Caulfield struct writequeue_entry *entry; 12166ed7257bSPatrick Caulfield 12173af2326cSAlexander Aring entry = dlm_allocate_writequeue(); 12186ed7257bSPatrick Caulfield if (!entry) 12196ed7257bSPatrick Caulfield return NULL; 12206ed7257bSPatrick Caulfield 1221be3b0400SAlexander Aring entry->page = alloc_page(GFP_ATOMIC | __GFP_ZERO); 12226ed7257bSPatrick Caulfield if (!entry->page) { 12233af2326cSAlexander Aring dlm_free_writequeue(entry); 12246ed7257bSPatrick Caulfield return NULL; 12256ed7257bSPatrick Caulfield } 12266ed7257bSPatrick Caulfield 12273af2326cSAlexander Aring entry->offset = 0; 12283af2326cSAlexander Aring entry->len = 0; 12293af2326cSAlexander Aring entry->end = 0; 12303af2326cSAlexander Aring entry->dirty = false; 12316ed7257bSPatrick Caulfield entry->con = con; 1232f0747ebfSAlexander Aring entry->users = 1; 12338f2dc78dSAlexander Aring kref_init(&entry->ref); 12346ed7257bSPatrick Caulfield return entry; 12356ed7257bSPatrick Caulfield } 12366ed7257bSPatrick Caulfield 1237f0747ebfSAlexander Aring static struct writequeue_entry *new_wq_entry(struct connection *con, int len, 1238be3b0400SAlexander Aring char **ppc, void (*cb)(void *data), 1239be3b0400SAlexander Aring void *data) 1240f0747ebfSAlexander Aring { 1241f0747ebfSAlexander Aring struct writequeue_entry *e; 1242f0747ebfSAlexander Aring 1243f0747ebfSAlexander Aring spin_lock(&con->writequeue_lock); 1244f0747ebfSAlexander Aring if (!list_empty(&con->writequeue)) { 1245f0747ebfSAlexander Aring e = list_last_entry(&con->writequeue, struct writequeue_entry, list); 1246f0747ebfSAlexander Aring if (DLM_WQ_REMAIN_BYTES(e) >= len) { 12478f2dc78dSAlexander Aring kref_get(&e->ref); 12488f2dc78dSAlexander Aring 1249f0747ebfSAlexander Aring *ppc = page_address(e->page) + e->end; 12508f2dc78dSAlexander Aring if (cb) 12515c16febbSAlexander Aring cb(data); 12528f2dc78dSAlexander Aring 1253f0747ebfSAlexander Aring e->end += len; 1254f0747ebfSAlexander Aring e->users++; 1255be3b0400SAlexander Aring goto out; 1256f0747ebfSAlexander Aring } 1257f0747ebfSAlexander Aring } 1258f0747ebfSAlexander Aring 1259be3b0400SAlexander Aring e = new_writequeue_entry(con); 1260f0747ebfSAlexander Aring if (!e) 1261be3b0400SAlexander Aring goto out; 1262f0747ebfSAlexander Aring 12638f2dc78dSAlexander Aring kref_get(&e->ref); 1264f0747ebfSAlexander Aring *ppc = page_address(e->page); 1265f0747ebfSAlexander Aring e->end += len; 12668aa31cbfSAlexander Aring atomic_inc(&con->writequeue_cnt); 12678f2dc78dSAlexander Aring if (cb) 12685c16febbSAlexander Aring cb(data); 12698f2dc78dSAlexander Aring 1270f0747ebfSAlexander Aring list_add_tail(&e->list, &con->writequeue); 1271f0747ebfSAlexander Aring 1272be3b0400SAlexander Aring out: 1273be3b0400SAlexander Aring spin_unlock(&con->writequeue_lock); 1274f0747ebfSAlexander Aring return e; 1275f0747ebfSAlexander Aring }; 1276f0747ebfSAlexander Aring 12772874d1a6SAlexander Aring static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len, 12782874d1a6SAlexander Aring gfp_t allocation, char **ppc, 12795c16febbSAlexander Aring void (*cb)(void *data), 12805c16febbSAlexander Aring void *data) 12812874d1a6SAlexander Aring { 12822874d1a6SAlexander Aring struct writequeue_entry *e; 12832874d1a6SAlexander Aring struct dlm_msg *msg; 12842874d1a6SAlexander Aring 1285e4dc81edSAlexander Aring msg = dlm_allocate_msg(allocation); 12862874d1a6SAlexander Aring if (!msg) 12872874d1a6SAlexander Aring return NULL; 12882874d1a6SAlexander Aring 12892874d1a6SAlexander Aring kref_init(&msg->ref); 12902874d1a6SAlexander Aring 1291be3b0400SAlexander Aring e = new_wq_entry(con, len, ppc, cb, data); 12922874d1a6SAlexander Aring if (!e) { 1293e4dc81edSAlexander Aring dlm_free_msg(msg); 12942874d1a6SAlexander Aring return NULL; 12952874d1a6SAlexander Aring } 12962874d1a6SAlexander Aring 1297e4dc81edSAlexander Aring msg->retransmit = false; 1298e4dc81edSAlexander Aring msg->orig_msg = NULL; 12992874d1a6SAlexander Aring msg->ppc = *ppc; 13002874d1a6SAlexander Aring msg->len = len; 13012874d1a6SAlexander Aring msg->entry = e; 13022874d1a6SAlexander Aring 13032874d1a6SAlexander Aring return msg; 13042874d1a6SAlexander Aring } 13052874d1a6SAlexander Aring 1306a8449f23SAlexander Aring /* avoid false positive for nodes_srcu, unlock happens in 1307a8449f23SAlexander Aring * dlm_lowcomms_commit_msg which is a must call if success 1308a8449f23SAlexander Aring */ 1309a8449f23SAlexander Aring #ifndef __CHECKER__ 13108f2dc78dSAlexander Aring struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation, 13115c16febbSAlexander Aring char **ppc, void (*cb)(void *data), 13125c16febbSAlexander Aring void *data) 13136ed7257bSPatrick Caulfield { 13146ed7257bSPatrick Caulfield struct connection *con; 13158f2dc78dSAlexander Aring struct dlm_msg *msg; 1316b38bc9c2SAlexander Aring int idx; 13176ed7257bSPatrick Caulfield 1318d10a0b88SAlexander Aring if (len > DLM_MAX_SOCKET_BUFSIZE || 1319c45674fbSAlexander Aring len < sizeof(struct dlm_header)) { 1320d10a0b88SAlexander Aring BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE); 1321692f51c8SAlexander Aring log_print("failed to allocate a buffer of size %d", len); 1322c45674fbSAlexander Aring WARN_ON(1); 1323692f51c8SAlexander Aring return NULL; 1324692f51c8SAlexander Aring } 1325692f51c8SAlexander Aring 1326b38bc9c2SAlexander Aring idx = srcu_read_lock(&connections_srcu); 13276ed7257bSPatrick Caulfield con = nodeid2con(nodeid, allocation); 1328b38bc9c2SAlexander Aring if (!con) { 1329b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 13306ed7257bSPatrick Caulfield return NULL; 1331b38bc9c2SAlexander Aring } 13326ed7257bSPatrick Caulfield 13335c16febbSAlexander Aring msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, data); 13348f2dc78dSAlexander Aring if (!msg) { 1335b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 1336b38bc9c2SAlexander Aring return NULL; 1337b38bc9c2SAlexander Aring } 1338b38bc9c2SAlexander Aring 133930ea3257SAlexander Aring /* for dlm_lowcomms_commit_msg() */ 134030ea3257SAlexander Aring kref_get(&msg->ref); 13418f2dc78dSAlexander Aring /* we assume if successful commit must called */ 13428f2dc78dSAlexander Aring msg->idx = idx; 13438f2dc78dSAlexander Aring return msg; 13448f2dc78dSAlexander Aring } 1345a8449f23SAlexander Aring #endif 13468f2dc78dSAlexander Aring 13472874d1a6SAlexander Aring static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg) 13486ed7257bSPatrick Caulfield { 13498f2dc78dSAlexander Aring struct writequeue_entry *e = msg->entry; 13506ed7257bSPatrick Caulfield struct connection *con = e->con; 13516ed7257bSPatrick Caulfield int users; 13526ed7257bSPatrick Caulfield 13536ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 13548f2dc78dSAlexander Aring kref_get(&msg->ref); 13558f2dc78dSAlexander Aring list_add(&msg->list, &e->msgs); 13568f2dc78dSAlexander Aring 13576ed7257bSPatrick Caulfield users = --e->users; 13586ed7257bSPatrick Caulfield if (users) 13596ed7257bSPatrick Caulfield goto out; 1360f0747ebfSAlexander Aring 1361f0747ebfSAlexander Aring e->len = DLM_WQ_LENGTH_BYTES(e); 13626ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 13636ed7257bSPatrick Caulfield 13646ed7257bSPatrick Caulfield queue_work(send_workqueue, &con->swork); 13656ed7257bSPatrick Caulfield return; 13666ed7257bSPatrick Caulfield 13676ed7257bSPatrick Caulfield out: 13686ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 13696ed7257bSPatrick Caulfield return; 13706ed7257bSPatrick Caulfield } 13716ed7257bSPatrick Caulfield 1372a8449f23SAlexander Aring /* avoid false positive for nodes_srcu, lock was happen in 1373a8449f23SAlexander Aring * dlm_lowcomms_new_msg 1374a8449f23SAlexander Aring */ 1375a8449f23SAlexander Aring #ifndef __CHECKER__ 13762874d1a6SAlexander Aring void dlm_lowcomms_commit_msg(struct dlm_msg *msg) 13772874d1a6SAlexander Aring { 13782874d1a6SAlexander Aring _dlm_lowcomms_commit_msg(msg); 13792874d1a6SAlexander Aring srcu_read_unlock(&connections_srcu, msg->idx); 138030ea3257SAlexander Aring /* because dlm_lowcomms_new_msg() */ 138130ea3257SAlexander Aring kref_put(&msg->ref, dlm_msg_release); 13822874d1a6SAlexander Aring } 1383a8449f23SAlexander Aring #endif 13842874d1a6SAlexander Aring 13858f2dc78dSAlexander Aring void dlm_lowcomms_put_msg(struct dlm_msg *msg) 13868f2dc78dSAlexander Aring { 13878f2dc78dSAlexander Aring kref_put(&msg->ref, dlm_msg_release); 13888f2dc78dSAlexander Aring } 13898f2dc78dSAlexander Aring 13902874d1a6SAlexander Aring /* does not held connections_srcu, usage workqueue only */ 13912874d1a6SAlexander Aring int dlm_lowcomms_resend_msg(struct dlm_msg *msg) 13922874d1a6SAlexander Aring { 13932874d1a6SAlexander Aring struct dlm_msg *msg_resend; 13942874d1a6SAlexander Aring char *ppc; 13952874d1a6SAlexander Aring 13962874d1a6SAlexander Aring if (msg->retransmit) 13972874d1a6SAlexander Aring return 1; 13982874d1a6SAlexander Aring 13992874d1a6SAlexander Aring msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len, 14002874d1a6SAlexander Aring GFP_ATOMIC, &ppc, NULL, NULL); 14012874d1a6SAlexander Aring if (!msg_resend) 14022874d1a6SAlexander Aring return -ENOMEM; 14032874d1a6SAlexander Aring 14042874d1a6SAlexander Aring msg->retransmit = true; 14052874d1a6SAlexander Aring kref_get(&msg->ref); 14062874d1a6SAlexander Aring msg_resend->orig_msg = msg; 14072874d1a6SAlexander Aring 14082874d1a6SAlexander Aring memcpy(ppc, msg->ppc, msg->len); 14092874d1a6SAlexander Aring _dlm_lowcomms_commit_msg(msg_resend); 14102874d1a6SAlexander Aring dlm_lowcomms_put_msg(msg_resend); 14112874d1a6SAlexander Aring 14122874d1a6SAlexander Aring return 0; 14132874d1a6SAlexander Aring } 14142874d1a6SAlexander Aring 14156ed7257bSPatrick Caulfield /* Send a message */ 14166ed7257bSPatrick Caulfield static void send_to_sock(struct connection *con) 14176ed7257bSPatrick Caulfield { 14186ed7257bSPatrick Caulfield const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; 14196ed7257bSPatrick Caulfield struct writequeue_entry *e; 142066d5955aSAlexander Aring int len, offset, ret; 1421f92c8dd7SBob Peterson int count = 0; 14226ed7257bSPatrick Caulfield 14236ed7257bSPatrick Caulfield mutex_lock(&con->sock_mutex); 14246ed7257bSPatrick Caulfield if (con->sock == NULL) 14256ed7257bSPatrick Caulfield goto out_connect; 14266ed7257bSPatrick Caulfield 14276ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14286ed7257bSPatrick Caulfield for (;;) { 142966d5955aSAlexander Aring e = con_next_wq(con); 143066d5955aSAlexander Aring if (!e) 14316ed7257bSPatrick Caulfield break; 14326ed7257bSPatrick Caulfield 14336ed7257bSPatrick Caulfield len = e->len; 14346ed7257bSPatrick Caulfield offset = e->offset; 14356ed7257bSPatrick Caulfield BUG_ON(len == 0 && e->users == 0); 14366ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14376ed7257bSPatrick Caulfield 14381329e3f2SPaolo Bonzini ret = kernel_sendpage(con->sock, e->page, offset, len, 14396ed7257bSPatrick Caulfield msg_flags); 144092732376SAlexander Aring trace_dlm_send(con->nodeid, ret); 1441d66f8277SPatrick Caulfield if (ret == -EAGAIN || ret == 0) { 1442b36930ddSDavid Miller if (ret == -EAGAIN && 14439cd3e072SEric Dumazet test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) && 1444b36930ddSDavid Miller !test_and_set_bit(CF_APP_LIMITED, &con->flags)) { 1445b36930ddSDavid Miller /* Notify TCP that we're limited by the 1446b36930ddSDavid Miller * application window size. 1447b36930ddSDavid Miller */ 1448b36930ddSDavid Miller set_bit(SOCK_NOSPACE, &con->sock->flags); 1449b36930ddSDavid Miller con->sock->sk->sk_write_pending++; 1450b36930ddSDavid Miller } 1451d66f8277SPatrick Caulfield cond_resched(); 14526ed7257bSPatrick Caulfield goto out; 14539c5bef58SYing Xue } else if (ret < 0) 1454ba868d9dSAlexander Aring goto out; 1455f92c8dd7SBob Peterson 14566ed7257bSPatrick Caulfield /* Don't starve people filling buffers */ 1457f92c8dd7SBob Peterson if (++count >= MAX_SEND_MSG_COUNT) { 14586ed7257bSPatrick Caulfield cond_resched(); 1459f92c8dd7SBob Peterson count = 0; 1460f92c8dd7SBob Peterson } 14616ed7257bSPatrick Caulfield 14626ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14635d689871SMike Christie writequeue_entry_complete(e, ret); 14646ed7257bSPatrick Caulfield } 14656ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 14668aa31cbfSAlexander Aring 14678aa31cbfSAlexander Aring /* close if we got EOF */ 14688aa31cbfSAlexander Aring if (test_and_clear_bit(CF_EOF, &con->flags)) { 14698aa31cbfSAlexander Aring mutex_unlock(&con->sock_mutex); 14708aa31cbfSAlexander Aring close_connection(con, false, false, true); 14718aa31cbfSAlexander Aring 14728aa31cbfSAlexander Aring /* handling for tcp shutdown */ 14738aa31cbfSAlexander Aring clear_bit(CF_SHUTDOWN, &con->flags); 14748aa31cbfSAlexander Aring wake_up(&con->shutdown_wait); 14758aa31cbfSAlexander Aring } else { 14768aa31cbfSAlexander Aring mutex_unlock(&con->sock_mutex); 14778aa31cbfSAlexander Aring } 14788aa31cbfSAlexander Aring 14798aa31cbfSAlexander Aring return; 14808aa31cbfSAlexander Aring 14816ed7257bSPatrick Caulfield out: 14826ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 14836ed7257bSPatrick Caulfield return; 14846ed7257bSPatrick Caulfield 14856ed7257bSPatrick Caulfield out_connect: 14866ed7257bSPatrick Caulfield mutex_unlock(&con->sock_mutex); 148701da24d3SBob Peterson queue_work(send_workqueue, &con->swork); 148801da24d3SBob Peterson cond_resched(); 14896ed7257bSPatrick Caulfield } 14906ed7257bSPatrick Caulfield 14916ed7257bSPatrick Caulfield static void clean_one_writequeue(struct connection *con) 14926ed7257bSPatrick Caulfield { 14935e9ccc37SChristine Caulfield struct writequeue_entry *e, *safe; 14946ed7257bSPatrick Caulfield 14956ed7257bSPatrick Caulfield spin_lock(&con->writequeue_lock); 14965e9ccc37SChristine Caulfield list_for_each_entry_safe(e, safe, &con->writequeue, list) { 14976ed7257bSPatrick Caulfield free_entry(e); 14986ed7257bSPatrick Caulfield } 14996ed7257bSPatrick Caulfield spin_unlock(&con->writequeue_lock); 15006ed7257bSPatrick Caulfield } 15016ed7257bSPatrick Caulfield 15026ed7257bSPatrick Caulfield /* Called from recovery when it knows that a node has 15036ed7257bSPatrick Caulfield left the cluster */ 15046ed7257bSPatrick Caulfield int dlm_lowcomms_close(int nodeid) 15056ed7257bSPatrick Caulfield { 15066ed7257bSPatrick Caulfield struct connection *con; 150736b71a8bSDavid Teigland struct dlm_node_addr *na; 1508b38bc9c2SAlexander Aring int idx; 15096ed7257bSPatrick Caulfield 15106ed7257bSPatrick Caulfield log_print("closing connection to node %d", nodeid); 1511b38bc9c2SAlexander Aring idx = srcu_read_lock(&connections_srcu); 15126ed7257bSPatrick Caulfield con = nodeid2con(nodeid, 0); 15136ed7257bSPatrick Caulfield if (con) { 1514063c4c99SLars Marowsky-Bree set_bit(CF_CLOSE, &con->flags); 15150d737a8cSMarcelo Ricardo Leitner close_connection(con, true, true, true); 15166ed7257bSPatrick Caulfield clean_one_writequeue(con); 151753a5edaaSAlexander Aring if (con->othercon) 151853a5edaaSAlexander Aring clean_one_writequeue(con->othercon); 15196ed7257bSPatrick Caulfield } 1520b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 152136b71a8bSDavid Teigland 152236b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 152336b71a8bSDavid Teigland na = find_node_addr(nodeid); 152436b71a8bSDavid Teigland if (na) { 152536b71a8bSDavid Teigland list_del(&na->list); 152636b71a8bSDavid Teigland while (na->addr_count--) 152736b71a8bSDavid Teigland kfree(na->addr[na->addr_count]); 152836b71a8bSDavid Teigland kfree(na); 152936b71a8bSDavid Teigland } 153036b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 153136b71a8bSDavid Teigland 15326ed7257bSPatrick Caulfield return 0; 15336ed7257bSPatrick Caulfield } 15346ed7257bSPatrick Caulfield 15356ed7257bSPatrick Caulfield /* Receive workqueue function */ 15366ed7257bSPatrick Caulfield static void process_recv_sockets(struct work_struct *work) 15376ed7257bSPatrick Caulfield { 15386ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, rwork); 15396ed7257bSPatrick Caulfield 15406ed7257bSPatrick Caulfield clear_bit(CF_READ_PENDING, &con->flags); 154162699b3fSAlexander Aring receive_from_sock(con); 15426ed7257bSPatrick Caulfield } 15436ed7257bSPatrick Caulfield 1544d11ccd45SAlexander Aring static void process_listen_recv_socket(struct work_struct *work) 1545d11ccd45SAlexander Aring { 1546d11ccd45SAlexander Aring accept_from_sock(&listen_con); 1547d11ccd45SAlexander Aring } 1548d11ccd45SAlexander Aring 15498728a455SAlexander Aring static void dlm_connect(struct connection *con) 15508728a455SAlexander Aring { 15518728a455SAlexander Aring struct sockaddr_storage addr; 15528728a455SAlexander Aring int result, addr_len; 15538728a455SAlexander Aring struct socket *sock; 15548728a455SAlexander Aring unsigned int mark; 15558728a455SAlexander Aring 15568728a455SAlexander Aring /* Some odd races can cause double-connects, ignore them */ 15578728a455SAlexander Aring if (con->retries++ > MAX_CONNECT_RETRIES) 15588728a455SAlexander Aring return; 15598728a455SAlexander Aring 15608728a455SAlexander Aring if (con->sock) { 15618728a455SAlexander Aring log_print("node %d already connected.", con->nodeid); 15628728a455SAlexander Aring return; 15638728a455SAlexander Aring } 15648728a455SAlexander Aring 15658728a455SAlexander Aring memset(&addr, 0, sizeof(addr)); 15668728a455SAlexander Aring result = nodeid_to_addr(con->nodeid, &addr, NULL, 15678728a455SAlexander Aring dlm_proto_ops->try_new_addr, &mark); 15688728a455SAlexander Aring if (result < 0) { 15698728a455SAlexander Aring log_print("no address for nodeid %d", con->nodeid); 15708728a455SAlexander Aring return; 15718728a455SAlexander Aring } 15728728a455SAlexander Aring 15738728a455SAlexander Aring /* Create a socket to communicate with */ 15748728a455SAlexander Aring result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 15758728a455SAlexander Aring SOCK_STREAM, dlm_proto_ops->proto, &sock); 15768728a455SAlexander Aring if (result < 0) 15778728a455SAlexander Aring goto socket_err; 15788728a455SAlexander Aring 15798728a455SAlexander Aring sock_set_mark(sock->sk, mark); 15808728a455SAlexander Aring dlm_proto_ops->sockopts(sock); 15818728a455SAlexander Aring 15828728a455SAlexander Aring add_sock(sock, con); 15838728a455SAlexander Aring 15848728a455SAlexander Aring result = dlm_proto_ops->bind(sock); 15858728a455SAlexander Aring if (result < 0) 15868728a455SAlexander Aring goto add_sock_err; 15878728a455SAlexander Aring 15888728a455SAlexander Aring log_print_ratelimited("connecting to %d", con->nodeid); 15898728a455SAlexander Aring make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len); 15908728a455SAlexander Aring result = dlm_proto_ops->connect(con, sock, (struct sockaddr *)&addr, 15918728a455SAlexander Aring addr_len); 15928728a455SAlexander Aring if (result < 0) 15938728a455SAlexander Aring goto add_sock_err; 15948728a455SAlexander Aring 15958728a455SAlexander Aring return; 15968728a455SAlexander Aring 15978728a455SAlexander Aring add_sock_err: 15988728a455SAlexander Aring dlm_close_sock(&con->sock); 15998728a455SAlexander Aring 16008728a455SAlexander Aring socket_err: 16018728a455SAlexander Aring /* 16028728a455SAlexander Aring * Some errors are fatal and this list might need adjusting. For other 16038728a455SAlexander Aring * errors we try again until the max number of retries is reached. 16048728a455SAlexander Aring */ 16058728a455SAlexander Aring if (result != -EHOSTUNREACH && 16068728a455SAlexander Aring result != -ENETUNREACH && 16078728a455SAlexander Aring result != -ENETDOWN && 16088728a455SAlexander Aring result != -EINVAL && 16098728a455SAlexander Aring result != -EPROTONOSUPPORT) { 16108728a455SAlexander Aring log_print("connect %d try %d error %d", con->nodeid, 16118728a455SAlexander Aring con->retries, result); 16128728a455SAlexander Aring msleep(1000); 16138728a455SAlexander Aring lowcomms_connect_sock(con); 16148728a455SAlexander Aring } 16158728a455SAlexander Aring } 16168728a455SAlexander Aring 16176ed7257bSPatrick Caulfield /* Send workqueue function */ 16186ed7257bSPatrick Caulfield static void process_send_sockets(struct work_struct *work) 16196ed7257bSPatrick Caulfield { 16206ed7257bSPatrick Caulfield struct connection *con = container_of(work, struct connection, swork); 16216ed7257bSPatrick Caulfield 16227443bc96SAlexander Aring WARN_ON(test_bit(CF_IS_OTHERCON, &con->flags)); 16237443bc96SAlexander Aring 16248a4abb08Stsutomu.owa@toshiba.co.jp clear_bit(CF_WRITE_PENDING, &con->flags); 1625ba868d9dSAlexander Aring 1626489d8e55SAlexander Aring if (test_and_clear_bit(CF_RECONNECT, &con->flags)) { 1627ba868d9dSAlexander Aring close_connection(con, false, false, true); 1628489d8e55SAlexander Aring dlm_midcomms_unack_msg_resend(con->nodeid); 1629489d8e55SAlexander Aring } 1630ba868d9dSAlexander Aring 16318728a455SAlexander Aring if (con->sock == NULL) { 1632ba868d9dSAlexander Aring if (test_and_clear_bit(CF_DELAY_CONNECT, &con->flags)) 1633ba868d9dSAlexander Aring msleep(1000); 16348728a455SAlexander Aring 16358728a455SAlexander Aring mutex_lock(&con->sock_mutex); 16368728a455SAlexander Aring dlm_connect(con); 16378728a455SAlexander Aring mutex_unlock(&con->sock_mutex); 1638ba868d9dSAlexander Aring } 16398728a455SAlexander Aring 164001da24d3SBob Peterson if (!list_empty(&con->writequeue)) 16416ed7257bSPatrick Caulfield send_to_sock(con); 16426ed7257bSPatrick Caulfield } 16436ed7257bSPatrick Caulfield 16446ed7257bSPatrick Caulfield static void work_stop(void) 16456ed7257bSPatrick Caulfield { 1646fcef0e6cSAlexander Aring if (recv_workqueue) { 16476ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 1648fcef0e6cSAlexander Aring recv_workqueue = NULL; 1649fcef0e6cSAlexander Aring } 1650fcef0e6cSAlexander Aring 1651fcef0e6cSAlexander Aring if (send_workqueue) { 16526ed7257bSPatrick Caulfield destroy_workqueue(send_workqueue); 1653fcef0e6cSAlexander Aring send_workqueue = NULL; 1654fcef0e6cSAlexander Aring } 16556ed7257bSPatrick Caulfield } 16566ed7257bSPatrick Caulfield 16576ed7257bSPatrick Caulfield static int work_start(void) 16586ed7257bSPatrick Caulfield { 16596c6a1cc6SAlexander Aring recv_workqueue = alloc_ordered_workqueue("dlm_recv", WQ_MEM_RECLAIM); 1660b9d41052SNamhyung Kim if (!recv_workqueue) { 1661b9d41052SNamhyung Kim log_print("can't start dlm_recv"); 1662b9d41052SNamhyung Kim return -ENOMEM; 16636ed7257bSPatrick Caulfield } 16646ed7257bSPatrick Caulfield 16656c6a1cc6SAlexander Aring send_workqueue = alloc_ordered_workqueue("dlm_send", WQ_MEM_RECLAIM); 1666b9d41052SNamhyung Kim if (!send_workqueue) { 1667b9d41052SNamhyung Kim log_print("can't start dlm_send"); 16686ed7257bSPatrick Caulfield destroy_workqueue(recv_workqueue); 1669fcef0e6cSAlexander Aring recv_workqueue = NULL; 1670b9d41052SNamhyung Kim return -ENOMEM; 16716ed7257bSPatrick Caulfield } 16726ed7257bSPatrick Caulfield 16736ed7257bSPatrick Caulfield return 0; 16746ed7257bSPatrick Caulfield } 16756ed7257bSPatrick Caulfield 16769d232469SAlexander Aring static void shutdown_conn(struct connection *con) 16779d232469SAlexander Aring { 1678a66c008cSAlexander Aring if (dlm_proto_ops->shutdown_action) 1679a66c008cSAlexander Aring dlm_proto_ops->shutdown_action(con); 16809d232469SAlexander Aring } 16819d232469SAlexander Aring 16829d232469SAlexander Aring void dlm_lowcomms_shutdown(void) 16839d232469SAlexander Aring { 1684b38bc9c2SAlexander Aring int idx; 1685b38bc9c2SAlexander Aring 16869d232469SAlexander Aring /* Set all the flags to prevent any 16879d232469SAlexander Aring * socket activity. 16889d232469SAlexander Aring */ 16899d232469SAlexander Aring dlm_allow_conn = 0; 16909d232469SAlexander Aring 16919d232469SAlexander Aring if (recv_workqueue) 16929d232469SAlexander Aring flush_workqueue(recv_workqueue); 16939d232469SAlexander Aring if (send_workqueue) 16949d232469SAlexander Aring flush_workqueue(send_workqueue); 16959d232469SAlexander Aring 16969d232469SAlexander Aring dlm_close_sock(&listen_con.sock); 16979d232469SAlexander Aring 1698b38bc9c2SAlexander Aring idx = srcu_read_lock(&connections_srcu); 16999d232469SAlexander Aring foreach_conn(shutdown_conn); 1700b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 17019d232469SAlexander Aring } 17029d232469SAlexander Aring 1703f0fb83cbStsutomu.owa@toshiba.co.jp static void _stop_conn(struct connection *con, bool and_other) 17046ed7257bSPatrick Caulfield { 1705f0fb83cbStsutomu.owa@toshiba.co.jp mutex_lock(&con->sock_mutex); 1706173a31feStsutomu.owa@toshiba.co.jp set_bit(CF_CLOSE, &con->flags); 1707f0fb83cbStsutomu.owa@toshiba.co.jp set_bit(CF_READ_PENDING, &con->flags); 17088a4abb08Stsutomu.owa@toshiba.co.jp set_bit(CF_WRITE_PENDING, &con->flags); 170993eaadebStsutomu.owa@toshiba.co.jp if (con->sock && con->sock->sk) { 171092c44605SAlexander Aring lock_sock(con->sock->sk); 1711afb853fbSPatrick Caulfield con->sock->sk->sk_user_data = NULL; 171292c44605SAlexander Aring release_sock(con->sock->sk); 171393eaadebStsutomu.owa@toshiba.co.jp } 1714f0fb83cbStsutomu.owa@toshiba.co.jp if (con->othercon && and_other) 1715f0fb83cbStsutomu.owa@toshiba.co.jp _stop_conn(con->othercon, false); 1716f0fb83cbStsutomu.owa@toshiba.co.jp mutex_unlock(&con->sock_mutex); 1717f0fb83cbStsutomu.owa@toshiba.co.jp } 1718f0fb83cbStsutomu.owa@toshiba.co.jp 1719f0fb83cbStsutomu.owa@toshiba.co.jp static void stop_conn(struct connection *con) 1720f0fb83cbStsutomu.owa@toshiba.co.jp { 1721f0fb83cbStsutomu.owa@toshiba.co.jp _stop_conn(con, true); 1722afb853fbSPatrick Caulfield } 17235e9ccc37SChristine Caulfield 17244798cbbfSAlexander Aring static void connection_release(struct rcu_head *rcu) 17254798cbbfSAlexander Aring { 17264798cbbfSAlexander Aring struct connection *con = container_of(rcu, struct connection, rcu); 17274798cbbfSAlexander Aring 17284798cbbfSAlexander Aring kfree(con->rx_buf); 17294798cbbfSAlexander Aring kfree(con); 17304798cbbfSAlexander Aring } 17314798cbbfSAlexander Aring 17325e9ccc37SChristine Caulfield static void free_conn(struct connection *con) 17335e9ccc37SChristine Caulfield { 17340d737a8cSMarcelo Ricardo Leitner close_connection(con, true, true, true); 1735a47666ebSAlexander Aring spin_lock(&connections_lock); 1736a47666ebSAlexander Aring hlist_del_rcu(&con->list); 1737a47666ebSAlexander Aring spin_unlock(&connections_lock); 1738948c47e9SAlexander Aring if (con->othercon) { 1739948c47e9SAlexander Aring clean_one_writequeue(con->othercon); 17405cbec208SAlexander Aring call_srcu(&connections_srcu, &con->othercon->rcu, 17415cbec208SAlexander Aring connection_release); 1742948c47e9SAlexander Aring } 17430de98432SAlexander Aring clean_one_writequeue(con); 17445cbec208SAlexander Aring call_srcu(&connections_srcu, &con->rcu, connection_release); 17456ed7257bSPatrick Caulfield } 17465e9ccc37SChristine Caulfield 1747f0fb83cbStsutomu.owa@toshiba.co.jp static void work_flush(void) 1748f0fb83cbStsutomu.owa@toshiba.co.jp { 1749b38bc9c2SAlexander Aring int ok; 1750f0fb83cbStsutomu.owa@toshiba.co.jp int i; 1751f0fb83cbStsutomu.owa@toshiba.co.jp struct connection *con; 1752f0fb83cbStsutomu.owa@toshiba.co.jp 1753f0fb83cbStsutomu.owa@toshiba.co.jp do { 1754f0fb83cbStsutomu.owa@toshiba.co.jp ok = 1; 1755f0fb83cbStsutomu.owa@toshiba.co.jp foreach_conn(stop_conn); 1756b355516fSDavid Windsor if (recv_workqueue) 1757f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(recv_workqueue); 1758b355516fSDavid Windsor if (send_workqueue) 1759f0fb83cbStsutomu.owa@toshiba.co.jp flush_workqueue(send_workqueue); 1760f0fb83cbStsutomu.owa@toshiba.co.jp for (i = 0; i < CONN_HASH_SIZE && ok; i++) { 1761a47666ebSAlexander Aring hlist_for_each_entry_rcu(con, &connection_hash[i], 1762a47666ebSAlexander Aring list) { 1763f0fb83cbStsutomu.owa@toshiba.co.jp ok &= test_bit(CF_READ_PENDING, &con->flags); 17648a4abb08Stsutomu.owa@toshiba.co.jp ok &= test_bit(CF_WRITE_PENDING, &con->flags); 17658a4abb08Stsutomu.owa@toshiba.co.jp if (con->othercon) { 1766f0fb83cbStsutomu.owa@toshiba.co.jp ok &= test_bit(CF_READ_PENDING, 1767f0fb83cbStsutomu.owa@toshiba.co.jp &con->othercon->flags); 17688a4abb08Stsutomu.owa@toshiba.co.jp ok &= test_bit(CF_WRITE_PENDING, 17698a4abb08Stsutomu.owa@toshiba.co.jp &con->othercon->flags); 17708a4abb08Stsutomu.owa@toshiba.co.jp } 1771f0fb83cbStsutomu.owa@toshiba.co.jp } 1772f0fb83cbStsutomu.owa@toshiba.co.jp } 1773f0fb83cbStsutomu.owa@toshiba.co.jp } while (!ok); 1774f0fb83cbStsutomu.owa@toshiba.co.jp } 1775f0fb83cbStsutomu.owa@toshiba.co.jp 17765e9ccc37SChristine Caulfield void dlm_lowcomms_stop(void) 17775e9ccc37SChristine Caulfield { 1778b38bc9c2SAlexander Aring int idx; 1779b38bc9c2SAlexander Aring 1780b38bc9c2SAlexander Aring idx = srcu_read_lock(&connections_srcu); 1781f0fb83cbStsutomu.owa@toshiba.co.jp work_flush(); 17823a8db798SMarcelo Ricardo Leitner foreach_conn(free_conn); 1783b38bc9c2SAlexander Aring srcu_read_unlock(&connections_srcu, idx); 17846ed7257bSPatrick Caulfield work_stop(); 1785043697f0SAlexander Aring deinit_local(); 1786a66c008cSAlexander Aring 1787a66c008cSAlexander Aring dlm_proto_ops = NULL; 17886ed7257bSPatrick Caulfield } 17896ed7257bSPatrick Caulfield 17902dc6b115SAlexander Aring static int dlm_listen_for_all(void) 17912dc6b115SAlexander Aring { 17922dc6b115SAlexander Aring struct socket *sock; 17932dc6b115SAlexander Aring int result; 17942dc6b115SAlexander Aring 17952dc6b115SAlexander Aring log_print("Using %s for communications", 17962dc6b115SAlexander Aring dlm_proto_ops->name); 17972dc6b115SAlexander Aring 17982dc6b115SAlexander Aring result = dlm_proto_ops->listen_validate(); 17992dc6b115SAlexander Aring if (result < 0) 18002dc6b115SAlexander Aring return result; 18012dc6b115SAlexander Aring 18022dc6b115SAlexander Aring result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 18032dc6b115SAlexander Aring SOCK_STREAM, dlm_proto_ops->proto, &sock); 18042dc6b115SAlexander Aring if (result < 0) { 1805fe933675SAlexander Aring log_print("Can't create comms socket: %d", result); 18061f4f1084SDan Carpenter return result; 18072dc6b115SAlexander Aring } 18082dc6b115SAlexander Aring 18092dc6b115SAlexander Aring sock_set_mark(sock->sk, dlm_config.ci_mark); 18102dc6b115SAlexander Aring dlm_proto_ops->listen_sockopts(sock); 18112dc6b115SAlexander Aring 18122dc6b115SAlexander Aring result = dlm_proto_ops->listen_bind(sock); 18132dc6b115SAlexander Aring if (result < 0) 18142dc6b115SAlexander Aring goto out; 18152dc6b115SAlexander Aring 18162dc6b115SAlexander Aring save_listen_callbacks(sock); 18172dc6b115SAlexander Aring add_listen_sock(sock, &listen_con); 18182dc6b115SAlexander Aring 18192dc6b115SAlexander Aring INIT_WORK(&listen_con.rwork, process_listen_recv_socket); 18202dc6b115SAlexander Aring result = sock->ops->listen(sock, 5); 18212dc6b115SAlexander Aring if (result < 0) { 18222dc6b115SAlexander Aring dlm_close_sock(&listen_con.sock); 1823*08ae0547SAlexander Aring return result; 18242dc6b115SAlexander Aring } 18252dc6b115SAlexander Aring 18262dc6b115SAlexander Aring return 0; 18272dc6b115SAlexander Aring 18282dc6b115SAlexander Aring out: 18292dc6b115SAlexander Aring sock_release(sock); 18302dc6b115SAlexander Aring return result; 18312dc6b115SAlexander Aring } 18322dc6b115SAlexander Aring 18338728a455SAlexander Aring static int dlm_tcp_bind(struct socket *sock) 18348728a455SAlexander Aring { 18358728a455SAlexander Aring struct sockaddr_storage src_addr; 18368728a455SAlexander Aring int result, addr_len; 18378728a455SAlexander Aring 18388728a455SAlexander Aring /* Bind to our cluster-known address connecting to avoid 18398728a455SAlexander Aring * routing problems. 18408728a455SAlexander Aring */ 18418728a455SAlexander Aring memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); 18428728a455SAlexander Aring make_sockaddr(&src_addr, 0, &addr_len); 18438728a455SAlexander Aring 18448728a455SAlexander Aring result = sock->ops->bind(sock, (struct sockaddr *)&src_addr, 18458728a455SAlexander Aring addr_len); 18468728a455SAlexander Aring if (result < 0) { 18478728a455SAlexander Aring /* This *may* not indicate a critical error */ 18488728a455SAlexander Aring log_print("could not bind for connect: %d", result); 18498728a455SAlexander Aring } 18508728a455SAlexander Aring 18518728a455SAlexander Aring return 0; 18528728a455SAlexander Aring } 18538728a455SAlexander Aring 18548728a455SAlexander Aring static int dlm_tcp_connect(struct connection *con, struct socket *sock, 18558728a455SAlexander Aring struct sockaddr *addr, int addr_len) 18568728a455SAlexander Aring { 18578728a455SAlexander Aring int ret; 18588728a455SAlexander Aring 18598728a455SAlexander Aring ret = sock->ops->connect(sock, addr, addr_len, O_NONBLOCK); 18608728a455SAlexander Aring switch (ret) { 18618728a455SAlexander Aring case -EINPROGRESS: 18628728a455SAlexander Aring fallthrough; 18638728a455SAlexander Aring case 0: 18648728a455SAlexander Aring return 0; 18658728a455SAlexander Aring } 18668728a455SAlexander Aring 18678728a455SAlexander Aring return ret; 18688728a455SAlexander Aring } 18698728a455SAlexander Aring 18702dc6b115SAlexander Aring static int dlm_tcp_listen_validate(void) 18712dc6b115SAlexander Aring { 18722dc6b115SAlexander Aring /* We don't support multi-homed hosts */ 18732dc6b115SAlexander Aring if (dlm_local_count > 1) { 18742dc6b115SAlexander Aring log_print("TCP protocol can't handle multi-homed hosts, try SCTP"); 18752dc6b115SAlexander Aring return -EINVAL; 18762dc6b115SAlexander Aring } 18772dc6b115SAlexander Aring 18782dc6b115SAlexander Aring return 0; 18792dc6b115SAlexander Aring } 18802dc6b115SAlexander Aring 18812dc6b115SAlexander Aring static void dlm_tcp_sockopts(struct socket *sock) 18822dc6b115SAlexander Aring { 18832dc6b115SAlexander Aring /* Turn off Nagle's algorithm */ 18842dc6b115SAlexander Aring tcp_sock_set_nodelay(sock->sk); 18852dc6b115SAlexander Aring } 18862dc6b115SAlexander Aring 18872dc6b115SAlexander Aring static void dlm_tcp_listen_sockopts(struct socket *sock) 18882dc6b115SAlexander Aring { 18892dc6b115SAlexander Aring dlm_tcp_sockopts(sock); 18902dc6b115SAlexander Aring sock_set_reuseaddr(sock->sk); 18912dc6b115SAlexander Aring } 18922dc6b115SAlexander Aring 18932dc6b115SAlexander Aring static int dlm_tcp_listen_bind(struct socket *sock) 18942dc6b115SAlexander Aring { 18952dc6b115SAlexander Aring int addr_len; 18962dc6b115SAlexander Aring 18972dc6b115SAlexander Aring /* Bind to our port */ 18982dc6b115SAlexander Aring make_sockaddr(dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len); 18992dc6b115SAlexander Aring return sock->ops->bind(sock, (struct sockaddr *)dlm_local_addr[0], 19002dc6b115SAlexander Aring addr_len); 19012dc6b115SAlexander Aring } 19022dc6b115SAlexander Aring 1903a66c008cSAlexander Aring static const struct dlm_proto_ops dlm_tcp_ops = { 19042dc6b115SAlexander Aring .name = "TCP", 19052dc6b115SAlexander Aring .proto = IPPROTO_TCP, 19068728a455SAlexander Aring .connect = dlm_tcp_connect, 19078728a455SAlexander Aring .sockopts = dlm_tcp_sockopts, 19088728a455SAlexander Aring .bind = dlm_tcp_bind, 19092dc6b115SAlexander Aring .listen_validate = dlm_tcp_listen_validate, 19102dc6b115SAlexander Aring .listen_sockopts = dlm_tcp_listen_sockopts, 19112dc6b115SAlexander Aring .listen_bind = dlm_tcp_listen_bind, 1912a66c008cSAlexander Aring .shutdown_action = dlm_tcp_shutdown, 1913a66c008cSAlexander Aring .eof_condition = tcp_eof_condition, 1914a66c008cSAlexander Aring }; 1915a66c008cSAlexander Aring 19168728a455SAlexander Aring static int dlm_sctp_bind(struct socket *sock) 19178728a455SAlexander Aring { 19188728a455SAlexander Aring return sctp_bind_addrs(sock, 0); 19198728a455SAlexander Aring } 19208728a455SAlexander Aring 19218728a455SAlexander Aring static int dlm_sctp_connect(struct connection *con, struct socket *sock, 19228728a455SAlexander Aring struct sockaddr *addr, int addr_len) 19238728a455SAlexander Aring { 19248728a455SAlexander Aring int ret; 19258728a455SAlexander Aring 19268728a455SAlexander Aring /* 19278728a455SAlexander Aring * Make sock->ops->connect() function return in specified time, 19288728a455SAlexander Aring * since O_NONBLOCK argument in connect() function does not work here, 19298728a455SAlexander Aring * then, we should restore the default value of this attribute. 19308728a455SAlexander Aring */ 19318728a455SAlexander Aring sock_set_sndtimeo(sock->sk, 5); 19328728a455SAlexander Aring ret = sock->ops->connect(sock, addr, addr_len, 0); 19338728a455SAlexander Aring sock_set_sndtimeo(sock->sk, 0); 19348728a455SAlexander Aring if (ret < 0) 19358728a455SAlexander Aring return ret; 19368728a455SAlexander Aring 19378728a455SAlexander Aring if (!test_and_set_bit(CF_CONNECTED, &con->flags)) 1938dfc020f3SAlexander Aring log_print("connected to node %d", con->nodeid); 19398728a455SAlexander Aring 19408728a455SAlexander Aring return 0; 19418728a455SAlexander Aring } 19428728a455SAlexander Aring 194390d21fc0SAlexander Aring static int dlm_sctp_listen_validate(void) 194490d21fc0SAlexander Aring { 194590d21fc0SAlexander Aring if (!IS_ENABLED(CONFIG_IP_SCTP)) { 194690d21fc0SAlexander Aring log_print("SCTP is not enabled by this kernel"); 194790d21fc0SAlexander Aring return -EOPNOTSUPP; 194890d21fc0SAlexander Aring } 194990d21fc0SAlexander Aring 195090d21fc0SAlexander Aring request_module("sctp"); 195190d21fc0SAlexander Aring return 0; 195290d21fc0SAlexander Aring } 195390d21fc0SAlexander Aring 19542dc6b115SAlexander Aring static int dlm_sctp_bind_listen(struct socket *sock) 19552dc6b115SAlexander Aring { 19562dc6b115SAlexander Aring return sctp_bind_addrs(sock, dlm_config.ci_tcp_port); 19572dc6b115SAlexander Aring } 19582dc6b115SAlexander Aring 19592dc6b115SAlexander Aring static void dlm_sctp_sockopts(struct socket *sock) 19602dc6b115SAlexander Aring { 19612dc6b115SAlexander Aring /* Turn off Nagle's algorithm */ 19622dc6b115SAlexander Aring sctp_sock_set_nodelay(sock->sk); 19632dc6b115SAlexander Aring sock_set_rcvbuf(sock->sk, NEEDED_RMEM); 19642dc6b115SAlexander Aring } 19652dc6b115SAlexander Aring 1966a66c008cSAlexander Aring static const struct dlm_proto_ops dlm_sctp_ops = { 19672dc6b115SAlexander Aring .name = "SCTP", 19682dc6b115SAlexander Aring .proto = IPPROTO_SCTP, 19698728a455SAlexander Aring .try_new_addr = true, 19708728a455SAlexander Aring .connect = dlm_sctp_connect, 19718728a455SAlexander Aring .sockopts = dlm_sctp_sockopts, 19728728a455SAlexander Aring .bind = dlm_sctp_bind, 197390d21fc0SAlexander Aring .listen_validate = dlm_sctp_listen_validate, 19742dc6b115SAlexander Aring .listen_sockopts = dlm_sctp_sockopts, 19752dc6b115SAlexander Aring .listen_bind = dlm_sctp_bind_listen, 1976a66c008cSAlexander Aring }; 1977a66c008cSAlexander Aring 19786ed7257bSPatrick Caulfield int dlm_lowcomms_start(void) 19796ed7257bSPatrick Caulfield { 19806ed7257bSPatrick Caulfield int error = -EINVAL; 19815e9ccc37SChristine Caulfield int i; 19825e9ccc37SChristine Caulfield 19835e9ccc37SChristine Caulfield for (i = 0; i < CONN_HASH_SIZE; i++) 19845e9ccc37SChristine Caulfield INIT_HLIST_HEAD(&connection_hash[i]); 19856ed7257bSPatrick Caulfield 19866ed7257bSPatrick Caulfield init_local(); 19876ed7257bSPatrick Caulfield if (!dlm_local_count) { 1988617e82e1SDavid Teigland error = -ENOTCONN; 19896ed7257bSPatrick Caulfield log_print("no local IP address has been set"); 1990513ef596SDavid Teigland goto fail; 19916ed7257bSPatrick Caulfield } 19926ed7257bSPatrick Caulfield 1993d11ccd45SAlexander Aring INIT_WORK(&listen_con.rwork, process_listen_recv_socket); 1994d11ccd45SAlexander Aring 1995513ef596SDavid Teigland error = work_start(); 1996513ef596SDavid Teigland if (error) 1997fcef0e6cSAlexander Aring goto fail_local; 1998513ef596SDavid Teigland 1999513ef596SDavid Teigland dlm_allow_conn = 1; 20006ed7257bSPatrick Caulfield 20016ed7257bSPatrick Caulfield /* Start listening */ 2002ac7d5d03SAlexander Aring switch (dlm_config.ci_protocol) { 2003ac7d5d03SAlexander Aring case DLM_PROTO_TCP: 2004a66c008cSAlexander Aring dlm_proto_ops = &dlm_tcp_ops; 2005ac7d5d03SAlexander Aring break; 2006ac7d5d03SAlexander Aring case DLM_PROTO_SCTP: 2007a66c008cSAlexander Aring dlm_proto_ops = &dlm_sctp_ops; 2008ac7d5d03SAlexander Aring break; 2009ac7d5d03SAlexander Aring default: 2010ac7d5d03SAlexander Aring log_print("Invalid protocol identifier %d set", 2011ac7d5d03SAlexander Aring dlm_config.ci_protocol); 2012ac7d5d03SAlexander Aring error = -EINVAL; 20132dc6b115SAlexander Aring goto fail_proto_ops; 2014ac7d5d03SAlexander Aring } 20152dc6b115SAlexander Aring 20162dc6b115SAlexander Aring error = dlm_listen_for_all(); 20176ed7257bSPatrick Caulfield if (error) 20182dc6b115SAlexander Aring goto fail_listen; 20196ed7257bSPatrick Caulfield 20206ed7257bSPatrick Caulfield return 0; 20216ed7257bSPatrick Caulfield 20222dc6b115SAlexander Aring fail_listen: 20232dc6b115SAlexander Aring dlm_proto_ops = NULL; 20242dc6b115SAlexander Aring fail_proto_ops: 2025513ef596SDavid Teigland dlm_allow_conn = 0; 2026fcef0e6cSAlexander Aring work_stop(); 2027fcef0e6cSAlexander Aring fail_local: 2028fcef0e6cSAlexander Aring deinit_local(); 2029513ef596SDavid Teigland fail: 20306ed7257bSPatrick Caulfield return error; 20316ed7257bSPatrick Caulfield } 203236b71a8bSDavid Teigland 203336b71a8bSDavid Teigland void dlm_lowcomms_exit(void) 203436b71a8bSDavid Teigland { 203536b71a8bSDavid Teigland struct dlm_node_addr *na, *safe; 203636b71a8bSDavid Teigland 203736b71a8bSDavid Teigland spin_lock(&dlm_node_addrs_spin); 203836b71a8bSDavid Teigland list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) { 203936b71a8bSDavid Teigland list_del(&na->list); 204036b71a8bSDavid Teigland while (na->addr_count--) 204136b71a8bSDavid Teigland kfree(na->addr[na->addr_count]); 204236b71a8bSDavid Teigland kfree(na); 204336b71a8bSDavid Teigland } 204436b71a8bSDavid Teigland spin_unlock(&dlm_node_addrs_spin); 204536b71a8bSDavid Teigland } 2046