fs/dlm/lowcomms.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /******************************************************************************
   3 *******************************************************************************
   4 **
   5 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   6 **  Copyright (C) 2004-2009 Red Hat, Inc.  All rights reserved.
   7 **
   8 **
   9 *******************************************************************************
  10 ******************************************************************************/
  11
  12 /*
  13  * lowcomms.c
  14  *
  15  * This is the "low-level" comms layer.
  16  *
  17  * It is responsible for sending/receiving messages
  18  * from other nodes in the cluster.
  19  *
  20  * Cluster nodes are referred to by their nodeids. nodeids are
  21  * simply 32 bit numbers to the locking module - if they need to
  22  * be expanded for the cluster infrastructure then that is its
  23  * responsibility. It is this layer's
  24  * responsibility to resolve these into IP address or
  25  * whatever it needs for inter-node communication.
  26  *
  27  * The comms level is two kernel threads that deal mainly with
  28  * the receiving of messages from other nodes and passing them
  29  * up to the mid-level comms layer (which understands the
  30  * message format) for execution by the locking core, and
  31  * a send thread which does all the setting up of connections
  32  * to remote nodes and the sending of data. Threads are not allowed
  33  * to send their own data because it may cause them to wait in times
  34  * of high load. Also, this way, the sending thread can collect together
  35  * messages bound for one node and send them in one block.
  36  *
  37  * lowcomms will choose to use either TCP or SCTP as its transport layer
  38  * depending on the configuration variable 'protocol'. This should be set
  39  * to 0 (default) for TCP or 1 for SCTP. It should be configured using a
  40  * cluster-wide mechanism as it must be the same on all nodes of the cluster
  41  * for the DLM to function.
  42  *
  43  */
  44
  45 #include <asm/ioctls.h>
  46 #include <net/sock.h>
  47 #include <net/tcp.h>
  48 #include <linux/pagemap.h>
  49 #include <linux/file.h>
  50 #include <linux/mutex.h>
  51 #include <linux/sctp.h>
  52 #include <linux/slab.h>
  53 #include <net/sctp/sctp.h>
  54 #include <net/ipv6.h>
  55
  56 #include "dlm_internal.h"
  57 #include "lowcomms.h"
  58 #include "midcomms.h"
  59 #include "config.h"
  60
  61 #define NEEDED_RMEM (4*1024*1024)
  62
  63 /* Number of messages to send before rescheduling */
  64 #define MAX_SEND_MSG_COUNT 25
  65 #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(10000)
  66
  67 struct connection {
  68         struct socket *sock;    /* NULL if not connected */
  69         uint32_t nodeid;        /* So we know who we are in the list */
  70         struct mutex sock_mutex;
  71         unsigned long flags;
  72 #define CF_READ_PENDING 1
  73 #define CF_WRITE_PENDING 2
  74 #define CF_INIT_PENDING 4
  75 #define CF_IS_OTHERCON 5
  76 #define CF_CLOSE 6
  77 #define CF_APP_LIMITED 7
  78 #define CF_CLOSING 8
  79 #define CF_SHUTDOWN 9
  80 #define CF_CONNECTED 10
  81 #define CF_RECONNECT 11
  82 #define CF_DELAY_CONNECT 12
  83 #define CF_EOF 13
  84         struct list_head writequeue;  /* List of outgoing writequeue_entries */
  85         spinlock_t writequeue_lock;
  86         atomic_t writequeue_cnt;
  87         void (*connect_action) (struct connection *);   /* What to do to connect */
  88         void (*shutdown_action)(struct connection *con); /* What to do to shutdown */
  89         bool (*eof_condition)(struct connection *con); /* What to do to eof check */
  90         int retries;
  91 #define MAX_CONNECT_RETRIES 3
  92         struct hlist_node list;
  93         struct connection *othercon;
  94         struct connection *sendcon;
  95         struct work_struct rwork; /* Receive workqueue */
  96         struct work_struct swork; /* Send workqueue */
  97         wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */
  98         unsigned char *rx_buf;
  99         int rx_buflen;
 100         int rx_leftover;
 101         struct rcu_head rcu;
 102 };
 103 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 104
 105 struct listen_connection {
 106         struct socket *sock;
 107         struct work_struct rwork;
 108 };
 109
 110 #define DLM_WQ_REMAIN_BYTES(e) (PAGE_SIZE - e->end)
 111 #define DLM_WQ_LENGTH_BYTES(e) (e->end - e->offset)
 112
 113 /* An entry waiting to be sent */
 114 struct writequeue_entry {
 115         struct list_head list;
 116         struct page *page;
 117         int offset;
 118         int len;
 119         int end;
 120         int users;
 121         bool dirty;
 122         struct connection *con;
 123         struct list_head msgs;
 124         struct kref ref;
 125 };
 126
 127 struct dlm_msg {
 128         struct writequeue_entry *entry;
 129         struct dlm_msg *orig_msg;
 130         bool retransmit;
 131         void *ppc;
 132         int len;
 133         int idx; /* new()/commit() idx exchange */
 134
 135         struct list_head list;
 136         struct kref ref;
 137 };
 138
 139 struct dlm_node_addr {
 140         struct list_head list;
 141         int nodeid;
 142         int mark;
 143         int addr_count;
 144         int curr_addr_index;
 145         struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
 146 };
 147
 148 static struct listen_sock_callbacks {
 149         void (*sk_error_report)(struct sock *);
 150         void (*sk_data_ready)(struct sock *);
 151         void (*sk_state_change)(struct sock *);
 152         void (*sk_write_space)(struct sock *);
 153 } listen_sock;
 154
 155 static LIST_HEAD(dlm_node_addrs);
 156 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
 157
 158 static struct listen_connection listen_con;
 159 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 160 static int dlm_local_count;
 161 int dlm_allow_conn;
 162
 163 /* Work queues */
 164 static struct workqueue_struct *recv_workqueue;
 165 static struct workqueue_struct *send_workqueue;
 166
 167 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 168 static DEFINE_SPINLOCK(connections_lock);
 169 DEFINE_STATIC_SRCU(connections_srcu);
 170
 171 static void process_recv_sockets(struct work_struct *work);
 172 static void process_send_sockets(struct work_struct *work);
 173
 174 static void sctp_connect_to_sock(struct connection *con);
 175 static void tcp_connect_to_sock(struct connection *con);
 176 static void dlm_tcp_shutdown(struct connection *con);
 177
 178 static struct connection *__find_con(int nodeid, int r)
 179 {
 180         struct connection *con;
 181
 182         hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
 183                 if (con->nodeid == nodeid)
 184                         return con;
 185         }
 186
 187         return NULL;
 188 }
 189
 190 static bool tcp_eof_condition(struct connection *con)
 191 {
 192         return atomic_read(&con->writequeue_cnt);
 193 }
 194
 195 static int dlm_con_init(struct connection *con, int nodeid)
 196 {
 197         con->rx_buflen = dlm_config.ci_buffer_size;
 198         con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS);
 199         if (!con->rx_buf)
 200                 return -ENOMEM;
 201
 202         con->nodeid = nodeid;
 203         mutex_init(&con->sock_mutex);
 204         INIT_LIST_HEAD(&con->writequeue);
 205         spin_lock_init(&con->writequeue_lock);
 206         atomic_set(&con->writequeue_cnt, 0);
 207         INIT_WORK(&con->swork, process_send_sockets);
 208         INIT_WORK(&con->rwork, process_recv_sockets);
 209         init_waitqueue_head(&con->shutdown_wait);
 210
 211         switch (dlm_config.ci_protocol) {
 212         case DLM_PROTO_TCP:
 213                 con->connect_action = tcp_connect_to_sock;
 214                 con->shutdown_action = dlm_tcp_shutdown;
 215                 con->eof_condition = tcp_eof_condition;
 216                 break;
 217         case DLM_PROTO_SCTP:
 218                 con->connect_action = sctp_connect_to_sock;
 219                 break;
 220         default:
 221                 kfree(con->rx_buf);
 222                 return -EINVAL;
 223         }
 224
 225         return 0;
 226 }
 227
 228 /*
 229  * If 'allocation' is zero then we don't attempt to create a new
 230  * connection structure for this node.
 231  */
 232 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
 233 {
 234         struct connection *con, *tmp;
 235         int r, ret;
 236
 237         r = nodeid_hash(nodeid);
 238         con = __find_con(nodeid, r);
 239         if (con || !alloc)
 240                 return con;
 241
 242         con = kzalloc(sizeof(*con), alloc);
 243         if (!con)
 244                 return NULL;
 245
 246         ret = dlm_con_init(con, nodeid);
 247         if (ret) {
 248                 kfree(con);
 249                 return NULL;
 250         }
 251
 252         spin_lock(&connections_lock);
 253         /* Because multiple workqueues/threads calls this function it can
 254          * race on multiple cpu's. Instead of locking hot path __find_con()
 255          * we just check in rare cases of recently added nodes again
 256          * under protection of connections_lock. If this is the case we
 257          * abort our connection creation and return the existing connection.
 258          */
 259         tmp = __find_con(nodeid, r);
 260         if (tmp) {
 261                 spin_unlock(&connections_lock);
 262                 kfree(con->rx_buf);
 263                 kfree(con);
 264                 return tmp;
 265         }
 266
 267         hlist_add_head_rcu(&con->list, &connection_hash[r]);
 268         spin_unlock(&connections_lock);
 269
 270         return con;
 271 }
 272
 273 /* Loop round all connections */
 274 static void foreach_conn(void (*conn_func)(struct connection *c))
 275 {
 276         int i;
 277         struct connection *con;
 278
 279         for (i = 0; i < CONN_HASH_SIZE; i++) {
 280                 hlist_for_each_entry_rcu(con, &connection_hash[i], list)
 281                         conn_func(con);
 282         }
 283 }
 284
 285 static struct dlm_node_addr *find_node_addr(int nodeid)
 286 {
 287         struct dlm_node_addr *na;
 288
 289         list_for_each_entry(na, &dlm_node_addrs, list) {
 290                 if (na->nodeid == nodeid)
 291                         return na;
 292         }
 293         return NULL;
 294 }
 295
 296 static int addr_compare(const struct sockaddr_storage *x,
 297                         const struct sockaddr_storage *y)
 298 {
 299         switch (x->ss_family) {
 300         case AF_INET: {
 301                 struct sockaddr_in *sinx = (struct sockaddr_in *)x;
 302                 struct sockaddr_in *siny = (struct sockaddr_in *)y;
 303                 if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
 304                         return 0;
 305                 if (sinx->sin_port != siny->sin_port)
 306                         return 0;
 307                 break;
 308         }
 309         case AF_INET6: {
 310                 struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
 311                 struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
 312                 if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
 313                         return 0;
 314                 if (sinx->sin6_port != siny->sin6_port)
 315                         return 0;
 316                 break;
 317         }
 318         default:
 319                 return 0;
 320         }
 321         return 1;
 322 }
 323
 324 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
 325                           struct sockaddr *sa_out, bool try_new_addr,
 326                           unsigned int *mark)
 327 {
 328         struct sockaddr_storage sas;
 329         struct dlm_node_addr *na;
 330
 331         if (!dlm_local_count)
 332                 return -1;
 333
 334         spin_lock(&dlm_node_addrs_spin);
 335         na = find_node_addr(nodeid);
 336         if (na && na->addr_count) {
 337                 memcpy(&sas, na->addr[na->curr_addr_index],
 338                        sizeof(struct sockaddr_storage));
 339
 340                 if (try_new_addr) {
 341                         na->curr_addr_index++;
 342                         if (na->curr_addr_index == na->addr_count)
 343                                 na->curr_addr_index = 0;
 344                 }
 345         }
 346         spin_unlock(&dlm_node_addrs_spin);
 347
 348         if (!na)
 349                 return -EEXIST;
 350
 351         if (!na->addr_count)
 352                 return -ENOENT;
 353
 354         *mark = na->mark;
 355
 356         if (sas_out)
 357                 memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
 358
 359         if (!sa_out)
 360                 return 0;
 361
 362         if (dlm_local_addr[0]->ss_family == AF_INET) {
 363                 struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
 364                 struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
 365                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 366         } else {
 367                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
 368                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
 369                 ret6->sin6_addr = in6->sin6_addr;
 370         }
 371
 372         return 0;
 373 }
 374
 375 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
 376                           unsigned int *mark)
 377 {
 378         struct dlm_node_addr *na;
 379         int rv = -EEXIST;
 380         int addr_i;
 381
 382         spin_lock(&dlm_node_addrs_spin);
 383         list_for_each_entry(na, &dlm_node_addrs, list) {
 384                 if (!na->addr_count)
 385                         continue;
 386
 387                 for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
 388                         if (addr_compare(na->addr[addr_i], addr)) {
 389                                 *nodeid = na->nodeid;
 390                                 *mark = na->mark;
 391                                 rv = 0;
 392                                 goto unlock;
 393                         }
 394                 }
 395         }
 396 unlock:
 397         spin_unlock(&dlm_node_addrs_spin);
 398         return rv;
 399 }
 400
 401 /* caller need to held dlm_node_addrs_spin lock */
 402 static bool dlm_lowcomms_na_has_addr(const struct dlm_node_addr *na,
 403                                      const struct sockaddr_storage *addr)
 404 {
 405         int i;
 406
 407         for (i = 0; i < na->addr_count; i++) {
 408                 if (addr_compare(na->addr[i], addr))
 409                         return true;
 410         }
 411
 412         return false;
 413 }
 414
 415 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 416 {
 417         struct sockaddr_storage *new_addr;
 418         struct dlm_node_addr *new_node, *na;
 419         bool ret;
 420
 421         new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
 422         if (!new_node)
 423                 return -ENOMEM;
 424
 425         new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
 426         if (!new_addr) {
 427                 kfree(new_node);
 428                 return -ENOMEM;
 429         }
 430
 431         memcpy(new_addr, addr, len);
 432
 433         spin_lock(&dlm_node_addrs_spin);
 434         na = find_node_addr(nodeid);
 435         if (!na) {
 436                 new_node->nodeid = nodeid;
 437                 new_node->addr[0] = new_addr;
 438                 new_node->addr_count = 1;
 439                 new_node->mark = dlm_config.ci_mark;
 440                 list_add(&new_node->list, &dlm_node_addrs);
 441                 spin_unlock(&dlm_node_addrs_spin);
 442                 return 0;
 443         }
 444
 445         ret = dlm_lowcomms_na_has_addr(na, addr);
 446         if (ret) {
 447                 spin_unlock(&dlm_node_addrs_spin);
 448                 kfree(new_addr);
 449                 kfree(new_node);
 450                 return -EEXIST;
 451         }
 452
 453         if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
 454                 spin_unlock(&dlm_node_addrs_spin);
 455                 kfree(new_addr);
 456                 kfree(new_node);
 457                 return -ENOSPC;
 458         }
 459
 460         na->addr[na->addr_count++] = new_addr;
 461         spin_unlock(&dlm_node_addrs_spin);
 462         kfree(new_node);
 463         return 0;
 464 }
 465
 466 /* Data available on socket or listen socket received a connect */
 467 static void lowcomms_data_ready(struct sock *sk)
 468 {
 469         struct connection *con;
 470
 471         read_lock_bh(&sk->sk_callback_lock);
 472         con = sock2con(sk);
 473         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 474                 queue_work(recv_workqueue, &con->rwork);
 475         read_unlock_bh(&sk->sk_callback_lock);
 476 }
 477
 478 static void lowcomms_listen_data_ready(struct sock *sk)
 479 {
 480         if (!dlm_allow_conn)
 481                 return;
 482
 483         queue_work(recv_workqueue, &listen_con.rwork);
 484 }
 485
 486 static void lowcomms_write_space(struct sock *sk)
 487 {
 488         struct connection *con;
 489
 490         read_lock_bh(&sk->sk_callback_lock);
 491         con = sock2con(sk);
 492         if (!con)
 493                 goto out;
 494
 495         if (!test_and_set_bit(CF_CONNECTED, &con->flags)) {
 496                 log_print("successful connected to node %d", con->nodeid);
 497                 queue_work(send_workqueue, &con->swork);
 498                 goto out;
 499         }
 500
 501         clear_bit(SOCK_NOSPACE, &con->sock->flags);
 502
 503         if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
 504                 con->sock->sk->sk_write_pending--;
 505                 clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
 506         }
 507
 508         queue_work(send_workqueue, &con->swork);
 509 out:
 510         read_unlock_bh(&sk->sk_callback_lock);
 511 }
 512
 513 static inline void lowcomms_connect_sock(struct connection *con)
 514 {
 515         if (test_bit(CF_CLOSE, &con->flags))
 516                 return;
 517         queue_work(send_workqueue, &con->swork);
 518         cond_resched();
 519 }
 520
 521 static void lowcomms_state_change(struct sock *sk)
 522 {
 523         /* SCTP layer is not calling sk_data_ready when the connection
 524          * is done, so we catch the signal through here. Also, it
 525          * doesn't switch socket state when entering shutdown, so we
 526          * skip the write in that case.
 527          */
 528         if (sk->sk_shutdown) {
 529                 if (sk->sk_shutdown == RCV_SHUTDOWN)
 530                         lowcomms_data_ready(sk);
 531         } else if (sk->sk_state == TCP_ESTABLISHED) {
 532                 lowcomms_write_space(sk);
 533         }
 534 }
 535
 536 int dlm_lowcomms_connect_node(int nodeid)
 537 {
 538         struct connection *con;
 539         int idx;
 540
 541         if (nodeid == dlm_our_nodeid())
 542                 return 0;
 543
 544         idx = srcu_read_lock(&connections_srcu);
 545         con = nodeid2con(nodeid, GFP_NOFS);
 546         if (!con) {
 547                 srcu_read_unlock(&connections_srcu, idx);
 548                 return -ENOMEM;
 549         }
 550
 551         lowcomms_connect_sock(con);
 552         srcu_read_unlock(&connections_srcu, idx);
 553
 554         return 0;
 555 }
 556
 557 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
 558 {
 559         struct dlm_node_addr *na;
 560
 561         spin_lock(&dlm_node_addrs_spin);
 562         na = find_node_addr(nodeid);
 563         if (!na) {
 564                 spin_unlock(&dlm_node_addrs_spin);
 565                 return -ENOENT;
 566         }
 567
 568         na->mark = mark;
 569         spin_unlock(&dlm_node_addrs_spin);
 570
 571         return 0;
 572 }
 573
 574 static void lowcomms_error_report(struct sock *sk)
 575 {
 576         struct connection *con;
 577         struct sockaddr_storage saddr;
 578         void (*orig_report)(struct sock *) = NULL;
 579
 580         read_lock_bh(&sk->sk_callback_lock);
 581         con = sock2con(sk);
 582         if (con == NULL)
 583                 goto out;
 584
 585         orig_report = listen_sock.sk_error_report;
 586         if (con->sock == NULL ||
 587             kernel_getpeername(con->sock, (struct sockaddr *)&saddr) < 0) {
 588                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 589                                    "sending to node %d, port %d, "
 590                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 591                                    con->nodeid, dlm_config.ci_tcp_port,
 592                                    sk->sk_err, sk->sk_err_soft);
 593         } else if (saddr.ss_family == AF_INET) {
 594                 struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr;
 595
 596                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 597                                    "sending to node %d at %pI4, port %d, "
 598                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 599                                    con->nodeid, &sin4->sin_addr.s_addr,
 600                                    dlm_config.ci_tcp_port, sk->sk_err,
 601                                    sk->sk_err_soft);
 602         } else {
 603                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr;
 604
 605                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 606                                    "sending to node %d at %u.%u.%u.%u, "
 607                                    "port %d, sk_err=%d/%d\n", dlm_our_nodeid(),
 608                                    con->nodeid, sin6->sin6_addr.s6_addr32[0],
 609                                    sin6->sin6_addr.s6_addr32[1],
 610                                    sin6->sin6_addr.s6_addr32[2],
 611                                    sin6->sin6_addr.s6_addr32[3],
 612                                    dlm_config.ci_tcp_port, sk->sk_err,
 613                                    sk->sk_err_soft);
 614         }
 615
 616         /* below sendcon only handling */
 617         if (test_bit(CF_IS_OTHERCON, &con->flags))
 618                 con = con->sendcon;
 619
 620         switch (sk->sk_err) {
 621         case ECONNREFUSED:
 622                 set_bit(CF_DELAY_CONNECT, &con->flags);
 623                 break;
 624         default:
 625                 break;
 626         }
 627
 628         if (!test_and_set_bit(CF_RECONNECT, &con->flags))
 629                 queue_work(send_workqueue, &con->swork);
 630
 631 out:
 632         read_unlock_bh(&sk->sk_callback_lock);
 633         if (orig_report)
 634                 orig_report(sk);
 635 }
 636
 637 /* Note: sk_callback_lock must be locked before calling this function. */
 638 static void save_listen_callbacks(struct socket *sock)
 639 {
 640         struct sock *sk = sock->sk;
 641
 642         listen_sock.sk_data_ready = sk->sk_data_ready;
 643         listen_sock.sk_state_change = sk->sk_state_change;
 644         listen_sock.sk_write_space = sk->sk_write_space;
 645         listen_sock.sk_error_report = sk->sk_error_report;
 646 }
 647
 648 static void restore_callbacks(struct socket *sock)
 649 {
 650         struct sock *sk = sock->sk;
 651
 652         write_lock_bh(&sk->sk_callback_lock);
 653         sk->sk_user_data = NULL;
 654         sk->sk_data_ready = listen_sock.sk_data_ready;
 655         sk->sk_state_change = listen_sock.sk_state_change;
 656         sk->sk_write_space = listen_sock.sk_write_space;
 657         sk->sk_error_report = listen_sock.sk_error_report;
 658         write_unlock_bh(&sk->sk_callback_lock);
 659 }
 660
 661 static void add_listen_sock(struct socket *sock, struct listen_connection *con)
 662 {
 663         struct sock *sk = sock->sk;
 664
 665         write_lock_bh(&sk->sk_callback_lock);
 666         save_listen_callbacks(sock);
 667         con->sock = sock;
 668
 669         sk->sk_user_data = con;
 670         sk->sk_allocation = GFP_NOFS;
 671         /* Install a data_ready callback */
 672         sk->sk_data_ready = lowcomms_listen_data_ready;
 673         write_unlock_bh(&sk->sk_callback_lock);
 674 }
 675
 676 /* Make a socket active */
 677 static void add_sock(struct socket *sock, struct connection *con)
 678 {
 679         struct sock *sk = sock->sk;
 680
 681         write_lock_bh(&sk->sk_callback_lock);
 682         con->sock = sock;
 683
 684         sk->sk_user_data = con;
 685         /* Install a data_ready callback */
 686         sk->sk_data_ready = lowcomms_data_ready;
 687         sk->sk_write_space = lowcomms_write_space;
 688         sk->sk_state_change = lowcomms_state_change;
 689         sk->sk_allocation = GFP_NOFS;
 690         sk->sk_error_report = lowcomms_error_report;
 691         write_unlock_bh(&sk->sk_callback_lock);
 692 }
 693
 694 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 695    length */
 696 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 697                           int *addr_len)
 698 {
 699         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 700         if (saddr->ss_family == AF_INET) {
 701                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 702                 in4_addr->sin_port = cpu_to_be16(port);
 703                 *addr_len = sizeof(struct sockaddr_in);
 704                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 705         } else {
 706                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 707                 in6_addr->sin6_port = cpu_to_be16(port);
 708                 *addr_len = sizeof(struct sockaddr_in6);
 709         }
 710         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 711 }
 712
 713 static void dlm_page_release(struct kref *kref)
 714 {
 715         struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
 716                                                   ref);
 717
 718         __free_page(e->page);
 719         kfree(e);
 720 }
 721
 722 static void dlm_msg_release(struct kref *kref)
 723 {
 724         struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
 725
 726         kref_put(&msg->entry->ref, dlm_page_release);
 727         kfree(msg);
 728 }
 729
 730 static void free_entry(struct writequeue_entry *e)
 731 {
 732         struct dlm_msg *msg, *tmp;
 733
 734         list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
 735                 if (msg->orig_msg) {
 736                         msg->orig_msg->retransmit = false;
 737                         kref_put(&msg->orig_msg->ref, dlm_msg_release);
 738                 }
 739
 740                 list_del(&msg->list);
 741                 kref_put(&msg->ref, dlm_msg_release);
 742         }
 743
 744         list_del(&e->list);
 745         atomic_dec(&e->con->writequeue_cnt);
 746         kref_put(&e->ref, dlm_page_release);
 747 }
 748
 749 static void dlm_close_sock(struct socket **sock)
 750 {
 751         if (*sock) {
 752                 restore_callbacks(*sock);
 753                 sock_release(*sock);
 754                 *sock = NULL;
 755         }
 756 }
 757
 758 /* Close a remote connection and tidy up */
 759 static void close_connection(struct connection *con, bool and_other,
 760                              bool tx, bool rx)
 761 {
 762         bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
 763         struct writequeue_entry *e;
 764
 765         if (tx && !closing && cancel_work_sync(&con->swork)) {
 766                 log_print("canceled swork for node %d", con->nodeid);
 767                 clear_bit(CF_WRITE_PENDING, &con->flags);
 768         }
 769         if (rx && !closing && cancel_work_sync(&con->rwork)) {
 770                 log_print("canceled rwork for node %d", con->nodeid);
 771                 clear_bit(CF_READ_PENDING, &con->flags);
 772         }
 773
 774         mutex_lock(&con->sock_mutex);
 775         dlm_close_sock(&con->sock);
 776
 777         if (con->othercon && and_other) {
 778                 /* Will only re-enter once. */
 779                 close_connection(con->othercon, false, tx, rx);
 780         }
 781
 782         /* if we send a writequeue entry only a half way, we drop the
 783          * whole entry because reconnection and that we not start of the
 784          * middle of a msg which will confuse the other end.
 785          *
 786          * we can always drop messages because retransmits, but what we
 787          * cannot allow is to transmit half messages which may be processed
 788          * at the other side.
 789          *
 790          * our policy is to start on a clean state when disconnects, we don't
 791          * know what's send/received on transport layer in this case.
 792          */
 793         spin_lock(&con->writequeue_lock);
 794         if (!list_empty(&con->writequeue)) {
 795                 e = list_first_entry(&con->writequeue, struct writequeue_entry,
 796                                      list);
 797                 if (e->dirty)
 798                         free_entry(e);
 799         }
 800         spin_unlock(&con->writequeue_lock);
 801
 802         con->rx_leftover = 0;
 803         con->retries = 0;
 804         clear_bit(CF_CONNECTED, &con->flags);
 805         clear_bit(CF_DELAY_CONNECT, &con->flags);
 806         clear_bit(CF_RECONNECT, &con->flags);
 807         clear_bit(CF_EOF, &con->flags);
 808         mutex_unlock(&con->sock_mutex);
 809         clear_bit(CF_CLOSING, &con->flags);
 810 }
 811
 812 static void shutdown_connection(struct connection *con)
 813 {
 814         int ret;
 815
 816         flush_work(&con->swork);
 817
 818         mutex_lock(&con->sock_mutex);
 819         /* nothing to shutdown */
 820         if (!con->sock) {
 821                 mutex_unlock(&con->sock_mutex);
 822                 return;
 823         }
 824
 825         set_bit(CF_SHUTDOWN, &con->flags);
 826         ret = kernel_sock_shutdown(con->sock, SHUT_WR);
 827         mutex_unlock(&con->sock_mutex);
 828         if (ret) {
 829                 log_print("Connection %p failed to shutdown: %d will force close",
 830                           con, ret);
 831                 goto force_close;
 832         } else {
 833                 ret = wait_event_timeout(con->shutdown_wait,
 834                                          !test_bit(CF_SHUTDOWN, &con->flags),
 835                                          DLM_SHUTDOWN_WAIT_TIMEOUT);
 836                 if (ret == 0) {
 837                         log_print("Connection %p shutdown timed out, will force close",
 838                                   con);
 839                         goto force_close;
 840                 }
 841         }
 842
 843         return;
 844
 845 force_close:
 846         clear_bit(CF_SHUTDOWN, &con->flags);
 847         close_connection(con, false, true, true);
 848 }
 849
 850 static void dlm_tcp_shutdown(struct connection *con)
 851 {
 852         if (con->othercon)
 853                 shutdown_connection(con->othercon);
 854         shutdown_connection(con);
 855 }
 856
 857 static int con_realloc_receive_buf(struct connection *con, int newlen)
 858 {
 859         unsigned char *newbuf;
 860
 861         newbuf = kmalloc(newlen, GFP_NOFS);
 862         if (!newbuf)
 863                 return -ENOMEM;
 864
 865         /* copy any leftover from last receive */
 866         if (con->rx_leftover)
 867                 memmove(newbuf, con->rx_buf, con->rx_leftover);
 868
 869         /* swap to new buffer space */
 870         kfree(con->rx_buf);
 871         con->rx_buflen = newlen;
 872         con->rx_buf = newbuf;
 873
 874         return 0;
 875 }
 876
 877 /* Data received from remote end */
 878 static int receive_from_sock(struct connection *con)
 879 {
 880         int call_again_soon = 0;
 881         struct msghdr msg;
 882         struct kvec iov;
 883         int ret, buflen;
 884
 885         mutex_lock(&con->sock_mutex);
 886
 887         if (con->sock == NULL) {
 888                 ret = -EAGAIN;
 889                 goto out_close;
 890         }
 891
 892         /* realloc if we get new buffer size to read out */
 893         buflen = dlm_config.ci_buffer_size;
 894         if (con->rx_buflen != buflen && con->rx_leftover <= buflen) {
 895                 ret = con_realloc_receive_buf(con, buflen);
 896                 if (ret < 0)
 897                         goto out_resched;
 898         }
 899
 900         /* calculate new buffer parameter regarding last receive and
 901          * possible leftover bytes
 902          */
 903         iov.iov_base = con->rx_buf + con->rx_leftover;
 904         iov.iov_len = con->rx_buflen - con->rx_leftover;
 905
 906         memset(&msg, 0, sizeof(msg));
 907         msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
 908         ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
 909                              msg.msg_flags);
 910         if (ret <= 0)
 911                 goto out_close;
 912         else if (ret == iov.iov_len)
 913                 call_again_soon = 1;
 914
 915         /* new buflen according readed bytes and leftover from last receive */
 916         buflen = ret + con->rx_leftover;
 917         ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen);
 918         if (ret < 0)
 919                 goto out_close;
 920
 921         /* calculate leftover bytes from process and put it into begin of
 922          * the receive buffer, so next receive we have the full message
 923          * at the start address of the receive buffer.
 924          */
 925         con->rx_leftover = buflen - ret;
 926         if (con->rx_leftover) {
 927                 memmove(con->rx_buf, con->rx_buf + ret,
 928                         con->rx_leftover);
 929                 call_again_soon = true;
 930         }
 931
 932         if (call_again_soon)
 933                 goto out_resched;
 934
 935         mutex_unlock(&con->sock_mutex);
 936         return 0;
 937
 938 out_resched:
 939         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 940                 queue_work(recv_workqueue, &con->rwork);
 941         mutex_unlock(&con->sock_mutex);
 942         return -EAGAIN;
 943
 944 out_close:
 945         if (ret == 0) {
 946                 log_print("connection %p got EOF from %d",
 947                           con, con->nodeid);
 948
 949                 if (con->eof_condition && con->eof_condition(con)) {
 950                         set_bit(CF_EOF, &con->flags);
 951                         mutex_unlock(&con->sock_mutex);
 952                 } else {
 953                         mutex_unlock(&con->sock_mutex);
 954                         close_connection(con, false, true, false);
 955
 956                         /* handling for tcp shutdown */
 957                         clear_bit(CF_SHUTDOWN, &con->flags);
 958                         wake_up(&con->shutdown_wait);
 959                 }
 960
 961                 /* signal to breaking receive worker */
 962                 ret = -1;
 963         } else {
 964                 mutex_unlock(&con->sock_mutex);
 965         }
 966         return ret;
 967 }
 968
 969 /* Listening socket is busy, accept a connection */
 970 static int accept_from_sock(struct listen_connection *con)
 971 {
 972         int result;
 973         struct sockaddr_storage peeraddr;
 974         struct socket *newsock;
 975         int len, idx;
 976         int nodeid;
 977         struct connection *newcon;
 978         struct connection *addcon;
 979         unsigned int mark;
 980
 981         if (!con->sock)
 982                 return -ENOTCONN;
 983
 984         result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
 985         if (result < 0)
 986                 goto accept_err;
 987
 988         /* Get the connected socket's peer */
 989         memset(&peeraddr, 0, sizeof(peeraddr));
 990         len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
 991         if (len < 0) {
 992                 result = -ECONNABORTED;
 993                 goto accept_err;
 994         }
 995
 996         /* Get the new node's NODEID */
 997         make_sockaddr(&peeraddr, 0, &len);
 998         if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
 999                 unsigned char *b=(unsigned char *)&peeraddr;
1000                 log_print("connect from non cluster node");
1001                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
1002                                      b, sizeof(struct sockaddr_storage));
1003                 sock_release(newsock);
1004                 return -1;
1005         }
1006
1007         log_print("got connection from %d", nodeid);
1008
1009         /*  Check to see if we already have a connection to this node. This
1010          *  could happen if the two nodes initiate a connection at roughly
1011          *  the same time and the connections cross on the wire.
1012          *  In this case we store the incoming one in "othercon"
1013          */
1014         idx = srcu_read_lock(&connections_srcu);
1015         newcon = nodeid2con(nodeid, GFP_NOFS);
1016         if (!newcon) {
1017                 srcu_read_unlock(&connections_srcu, idx);
1018                 result = -ENOMEM;
1019                 goto accept_err;
1020         }
1021
1022         sock_set_mark(newsock->sk, mark);
1023
1024         mutex_lock(&newcon->sock_mutex);
1025         if (newcon->sock) {
1026                 struct connection *othercon = newcon->othercon;
1027
1028                 if (!othercon) {
1029                         othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1030                         if (!othercon) {
1031                                 log_print("failed to allocate incoming socket");
1032                                 mutex_unlock(&newcon->sock_mutex);
1033                                 srcu_read_unlock(&connections_srcu, idx);
1034                                 result = -ENOMEM;
1035                                 goto accept_err;
1036                         }
1037
1038                         result = dlm_con_init(othercon, nodeid);
1039                         if (result < 0) {
1040                                 kfree(othercon);
1041                                 mutex_unlock(&newcon->sock_mutex);
1042                                 srcu_read_unlock(&connections_srcu, idx);
1043                                 goto accept_err;
1044                         }
1045
1046                         lockdep_set_subclass(&othercon->sock_mutex, 1);
1047                         set_bit(CF_IS_OTHERCON, &othercon->flags);
1048                         newcon->othercon = othercon;
1049                         othercon->sendcon = newcon;
1050                 } else {
1051                         /* close other sock con if we have something new */
1052                         close_connection(othercon, false, true, false);
1053                 }
1054
1055                 mutex_lock(&othercon->sock_mutex);
1056                 add_sock(newsock, othercon);
1057                 addcon = othercon;
1058                 mutex_unlock(&othercon->sock_mutex);
1059         }
1060         else {
1061                 /* accept copies the sk after we've saved the callbacks, so we
1062                    don't want to save them a second time or comm errors will
1063                    result in calling sk_error_report recursively. */
1064                 add_sock(newsock, newcon);
1065                 addcon = newcon;
1066         }
1067
1068         set_bit(CF_CONNECTED, &addcon->flags);
1069         mutex_unlock(&newcon->sock_mutex);
1070
1071         /*
1072          * Add it to the active queue in case we got data
1073          * between processing the accept adding the socket
1074          * to the read_sockets list
1075          */
1076         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
1077                 queue_work(recv_workqueue, &addcon->rwork);
1078
1079         srcu_read_unlock(&connections_srcu, idx);
1080
1081         return 0;
1082
1083 accept_err:
1084         if (newsock)
1085                 sock_release(newsock);
1086
1087         if (result != -EAGAIN)
1088                 log_print("error accepting connection from node: %d", result);
1089         return result;
1090 }
1091
1092 /*
1093  * writequeue_entry_complete - try to delete and free write queue entry
1094  * @e: write queue entry to try to delete
1095  * @completed: bytes completed
1096  *
1097  * writequeue_lock must be held.
1098  */
1099 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1100 {
1101         e->offset += completed;
1102         e->len -= completed;
1103         /* signal that page was half way transmitted */
1104         e->dirty = true;
1105
1106         if (e->len == 0 && e->users == 0)
1107                 free_entry(e);
1108 }
1109
1110 /*
1111  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1112  */
1113 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1114 {
1115         struct sockaddr_storage localaddr;
1116         struct sockaddr *addr = (struct sockaddr *)&localaddr;
1117         int i, addr_len, result = 0;
1118
1119         for (i = 0; i < dlm_local_count; i++) {
1120                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
1121                 make_sockaddr(&localaddr, port, &addr_len);
1122
1123                 if (!i)
1124                         result = kernel_bind(sock, addr, addr_len);
1125                 else
1126                         result = sock_bind_add(sock->sk, addr, addr_len);
1127
1128                 if (result < 0) {
1129                         log_print("Can't bind to %d addr number %d, %d.\n",
1130                                   port, i + 1, result);
1131                         break;
1132                 }
1133         }
1134         return result;
1135 }
1136
1137 /* Initiate an SCTP association.
1138    This is a special case of send_to_sock() in that we don't yet have a
1139    peeled-off socket for this association, so we use the listening socket
1140    and add the primary IP address of the remote node.
1141  */
1142 static void sctp_connect_to_sock(struct connection *con)
1143 {
1144         struct sockaddr_storage daddr;
1145         int result;
1146         int addr_len;
1147         struct socket *sock;
1148         unsigned int mark;
1149
1150         mutex_lock(&con->sock_mutex);
1151
1152         /* Some odd races can cause double-connects, ignore them */
1153         if (con->retries++ > MAX_CONNECT_RETRIES)
1154                 goto out;
1155
1156         if (con->sock) {
1157                 log_print("node %d already connected.", con->nodeid);
1158                 goto out;
1159         }
1160
1161         memset(&daddr, 0, sizeof(daddr));
1162         result = nodeid_to_addr(con->nodeid, &daddr, NULL, true, &mark);
1163         if (result < 0) {
1164                 log_print("no address for nodeid %d", con->nodeid);
1165                 goto out;
1166         }
1167
1168         /* Create a socket to communicate with */
1169         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1170                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1171         if (result < 0)
1172                 goto socket_err;
1173
1174         sock_set_mark(sock->sk, mark);
1175
1176         add_sock(sock, con);
1177
1178         /* Bind to all addresses. */
1179         if (sctp_bind_addrs(con->sock, 0))
1180                 goto bind_err;
1181
1182         make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len);
1183
1184         log_print_ratelimited("connecting to %d", con->nodeid);
1185
1186         /* Turn off Nagle's algorithm */
1187         sctp_sock_set_nodelay(sock->sk);
1188
1189         /*
1190          * Make sock->ops->connect() function return in specified time,
1191          * since O_NONBLOCK argument in connect() function does not work here,
1192          * then, we should restore the default value of this attribute.
1193          */
1194         sock_set_sndtimeo(sock->sk, 5);
1195         result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len,
1196                                    0);
1197         sock_set_sndtimeo(sock->sk, 0);
1198
1199         if (result == -EINPROGRESS)
1200                 result = 0;
1201         if (result == 0) {
1202                 if (!test_and_set_bit(CF_CONNECTED, &con->flags))
1203                         log_print("successful connected to node %d", con->nodeid);
1204                 goto out;
1205         }
1206
1207 bind_err:
1208         con->sock = NULL;
1209         sock_release(sock);
1210
1211 socket_err:
1212         /*
1213          * Some errors are fatal and this list might need adjusting. For other
1214          * errors we try again until the max number of retries is reached.
1215          */
1216         if (result != -EHOSTUNREACH &&
1217             result != -ENETUNREACH &&
1218             result != -ENETDOWN &&
1219             result != -EINVAL &&
1220             result != -EPROTONOSUPPORT) {
1221                 log_print("connect %d try %d error %d", con->nodeid,
1222                           con->retries, result);
1223                 mutex_unlock(&con->sock_mutex);
1224                 msleep(1000);
1225                 lowcomms_connect_sock(con);
1226                 return;
1227         }
1228
1229 out:
1230         mutex_unlock(&con->sock_mutex);
1231 }
1232
1233 /* Connect a new socket to its peer */
1234 static void tcp_connect_to_sock(struct connection *con)
1235 {
1236         struct sockaddr_storage saddr, src_addr;
1237         unsigned int mark;
1238         int addr_len;
1239         struct socket *sock = NULL;
1240         int result;
1241
1242         mutex_lock(&con->sock_mutex);
1243         if (con->retries++ > MAX_CONNECT_RETRIES)
1244                 goto out;
1245
1246         /* Some odd races can cause double-connects, ignore them */
1247         if (con->sock)
1248                 goto out;
1249
1250         /* Create a socket to communicate with */
1251         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1252                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1253         if (result < 0)
1254                 goto out_err;
1255
1256         memset(&saddr, 0, sizeof(saddr));
1257         result = nodeid_to_addr(con->nodeid, &saddr, NULL, false, &mark);
1258         if (result < 0) {
1259                 log_print("no address for nodeid %d", con->nodeid);
1260                 goto out_err;
1261         }
1262
1263         sock_set_mark(sock->sk, mark);
1264
1265         add_sock(sock, con);
1266
1267         /* Bind to our cluster-known address connecting to avoid
1268            routing problems */
1269         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1270         make_sockaddr(&src_addr, 0, &addr_len);
1271         result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1272                                  addr_len);
1273         if (result < 0) {
1274                 log_print("could not bind for connect: %d", result);
1275                 /* This *may* not indicate a critical error */
1276         }
1277
1278         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1279
1280         log_print_ratelimited("connecting to %d", con->nodeid);
1281
1282         /* Turn off Nagle's algorithm */
1283         tcp_sock_set_nodelay(sock->sk);
1284
1285         result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1286                                    O_NONBLOCK);
1287         if (result == -EINPROGRESS)
1288                 result = 0;
1289         if (result == 0)
1290                 goto out;
1291
1292 out_err:
1293         if (con->sock) {
1294                 sock_release(con->sock);
1295                 con->sock = NULL;
1296         } else if (sock) {
1297                 sock_release(sock);
1298         }
1299         /*
1300          * Some errors are fatal and this list might need adjusting. For other
1301          * errors we try again until the max number of retries is reached.
1302          */
1303         if (result != -EHOSTUNREACH &&
1304             result != -ENETUNREACH &&
1305             result != -ENETDOWN &&
1306             result != -EINVAL &&
1307             result != -EPROTONOSUPPORT) {
1308                 log_print("connect %d try %d error %d", con->nodeid,
1309                           con->retries, result);
1310                 mutex_unlock(&con->sock_mutex);
1311                 msleep(1000);
1312                 lowcomms_connect_sock(con);
1313                 return;
1314         }
1315 out:
1316         mutex_unlock(&con->sock_mutex);
1317         return;
1318 }
1319
1320 /* On error caller must run dlm_close_sock() for the
1321  * listen connection socket.
1322  */
1323 static int tcp_create_listen_sock(struct listen_connection *con,
1324                                   struct sockaddr_storage *saddr)
1325 {
1326         struct socket *sock = NULL;
1327         int result = 0;
1328         int addr_len;
1329
1330         if (dlm_local_addr[0]->ss_family == AF_INET)
1331                 addr_len = sizeof(struct sockaddr_in);
1332         else
1333                 addr_len = sizeof(struct sockaddr_in6);
1334
1335         /* Create a socket to communicate with */
1336         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1337                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1338         if (result < 0) {
1339                 log_print("Can't create listening comms socket");
1340                 goto create_out;
1341         }
1342
1343         sock_set_mark(sock->sk, dlm_config.ci_mark);
1344
1345         /* Turn off Nagle's algorithm */
1346         tcp_sock_set_nodelay(sock->sk);
1347
1348         sock_set_reuseaddr(sock->sk);
1349
1350         add_listen_sock(sock, con);
1351
1352         /* Bind to our port */
1353         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1354         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1355         if (result < 0) {
1356                 log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1357                 goto create_out;
1358         }
1359         sock_set_keepalive(sock->sk);
1360
1361         result = sock->ops->listen(sock, 5);
1362         if (result < 0) {
1363                 log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1364                 goto create_out;
1365         }
1366
1367         return 0;
1368
1369 create_out:
1370         return result;
1371 }
1372
1373 /* Get local addresses */
1374 static void init_local(void)
1375 {
1376         struct sockaddr_storage sas, *addr;
1377         int i;
1378
1379         dlm_local_count = 0;
1380         for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1381                 if (dlm_our_addr(&sas, i))
1382                         break;
1383
1384                 addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1385                 if (!addr)
1386                         break;
1387                 dlm_local_addr[dlm_local_count++] = addr;
1388         }
1389 }
1390
1391 static void deinit_local(void)
1392 {
1393         int i;
1394
1395         for (i = 0; i < dlm_local_count; i++)
1396                 kfree(dlm_local_addr[i]);
1397 }
1398
1399 /* Initialise SCTP socket and bind to all interfaces
1400  * On error caller must run dlm_close_sock() for the
1401  * listen connection socket.
1402  */
1403 static int sctp_listen_for_all(struct listen_connection *con)
1404 {
1405         struct socket *sock = NULL;
1406         int result = -EINVAL;
1407
1408         log_print("Using SCTP for communications");
1409
1410         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1411                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1412         if (result < 0) {
1413                 log_print("Can't create comms socket, check SCTP is loaded");
1414                 goto out;
1415         }
1416
1417         sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1418         sock_set_mark(sock->sk, dlm_config.ci_mark);
1419         sctp_sock_set_nodelay(sock->sk);
1420
1421         add_listen_sock(sock, con);
1422
1423         /* Bind to all addresses. */
1424         result = sctp_bind_addrs(con->sock, dlm_config.ci_tcp_port);
1425         if (result < 0)
1426                 goto out;
1427
1428         result = sock->ops->listen(sock, 5);
1429         if (result < 0) {
1430                 log_print("Can't set socket listening");
1431                 goto out;
1432         }
1433
1434         return 0;
1435
1436 out:
1437         return result;
1438 }
1439
1440 static int tcp_listen_for_all(void)
1441 {
1442         /* We don't support multi-homed hosts */
1443         if (dlm_local_count > 1) {
1444                 log_print("TCP protocol can't handle multi-homed hosts, "
1445                           "try SCTP");
1446                 return -EINVAL;
1447         }
1448
1449         log_print("Using TCP for communications");
1450
1451         return tcp_create_listen_sock(&listen_con, dlm_local_addr[0]);
1452 }
1453
1454
1455
1456 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1457                                                      gfp_t allocation)
1458 {
1459         struct writequeue_entry *entry;
1460
1461         entry = kzalloc(sizeof(*entry), allocation);
1462         if (!entry)
1463                 return NULL;
1464
1465         entry->page = alloc_page(allocation | __GFP_ZERO);
1466         if (!entry->page) {
1467                 kfree(entry);
1468                 return NULL;
1469         }
1470
1471         entry->con = con;
1472         entry->users = 1;
1473         kref_init(&entry->ref);
1474         INIT_LIST_HEAD(&entry->msgs);
1475
1476         return entry;
1477 }
1478
1479 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1480                                              gfp_t allocation, char **ppc,
1481                                              void (*cb)(struct dlm_mhandle *mh),
1482                                              struct dlm_mhandle *mh)
1483 {
1484         struct writequeue_entry *e;
1485
1486         spin_lock(&con->writequeue_lock);
1487         if (!list_empty(&con->writequeue)) {
1488                 e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1489                 if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1490                         kref_get(&e->ref);
1491
1492                         *ppc = page_address(e->page) + e->end;
1493                         if (cb)
1494                                 cb(mh);
1495
1496                         e->end += len;
1497                         e->users++;
1498                         spin_unlock(&con->writequeue_lock);
1499
1500                         return e;
1501                 }
1502         }
1503         spin_unlock(&con->writequeue_lock);
1504
1505         e = new_writequeue_entry(con, allocation);
1506         if (!e)
1507                 return NULL;
1508
1509         kref_get(&e->ref);
1510         *ppc = page_address(e->page);
1511         e->end += len;
1512         atomic_inc(&con->writequeue_cnt);
1513
1514         spin_lock(&con->writequeue_lock);
1515         if (cb)
1516                 cb(mh);
1517
1518         list_add_tail(&e->list, &con->writequeue);
1519         spin_unlock(&con->writequeue_lock);
1520
1521         return e;
1522 };
1523
1524 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1525                                                 gfp_t allocation, char **ppc,
1526                                                 void (*cb)(struct dlm_mhandle *mh),
1527                                                 struct dlm_mhandle *mh)
1528 {
1529         struct writequeue_entry *e;
1530         struct dlm_msg *msg;
1531
1532         msg = kzalloc(sizeof(*msg), allocation);
1533         if (!msg)
1534                 return NULL;
1535
1536         kref_init(&msg->ref);
1537
1538         e = new_wq_entry(con, len, allocation, ppc, cb, mh);
1539         if (!e) {
1540                 kfree(msg);
1541                 return NULL;
1542         }
1543
1544         msg->ppc = *ppc;
1545         msg->len = len;
1546         msg->entry = e;
1547
1548         return msg;
1549 }
1550
1551 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
1552                                      char **ppc, void (*cb)(struct dlm_mhandle *mh),
1553                                      struct dlm_mhandle *mh)
1554 {
1555         struct connection *con;
1556         struct dlm_msg *msg;
1557         int idx;
1558
1559         if (len > DLM_MAX_SOCKET_BUFSIZE ||
1560             len < sizeof(struct dlm_header)) {
1561                 BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1562                 log_print("failed to allocate a buffer of size %d", len);
1563                 WARN_ON(1);
1564                 return NULL;
1565         }
1566
1567         idx = srcu_read_lock(&connections_srcu);
1568         con = nodeid2con(nodeid, allocation);
1569         if (!con) {
1570                 srcu_read_unlock(&connections_srcu, idx);
1571                 return NULL;
1572         }
1573
1574         msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, mh);
1575         if (!msg) {
1576                 srcu_read_unlock(&connections_srcu, idx);
1577                 return NULL;
1578         }
1579
1580         /* we assume if successful commit must called */
1581         msg->idx = idx;
1582         return msg;
1583 }
1584
1585 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1586 {
1587         struct writequeue_entry *e = msg->entry;
1588         struct connection *con = e->con;
1589         int users;
1590
1591         spin_lock(&con->writequeue_lock);
1592         kref_get(&msg->ref);
1593         list_add(&msg->list, &e->msgs);
1594
1595         users = --e->users;
1596         if (users)
1597                 goto out;
1598
1599         e->len = DLM_WQ_LENGTH_BYTES(e);
1600         spin_unlock(&con->writequeue_lock);
1601
1602         queue_work(send_workqueue, &con->swork);
1603         return;
1604
1605 out:
1606         spin_unlock(&con->writequeue_lock);
1607         return;
1608 }
1609
1610 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1611 {
1612         _dlm_lowcomms_commit_msg(msg);
1613         srcu_read_unlock(&connections_srcu, msg->idx);
1614 }
1615
1616 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1617 {
1618         kref_put(&msg->ref, dlm_msg_release);
1619 }
1620
1621 /* does not held connections_srcu, usage workqueue only */
1622 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1623 {
1624         struct dlm_msg *msg_resend;
1625         char *ppc;
1626
1627         if (msg->retransmit)
1628                 return 1;
1629
1630         msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
1631                                               GFP_ATOMIC, &ppc, NULL, NULL);
1632         if (!msg_resend)
1633                 return -ENOMEM;
1634
1635         msg->retransmit = true;
1636         kref_get(&msg->ref);
1637         msg_resend->orig_msg = msg;
1638
1639         memcpy(ppc, msg->ppc, msg->len);
1640         _dlm_lowcomms_commit_msg(msg_resend);
1641         dlm_lowcomms_put_msg(msg_resend);
1642
1643         return 0;
1644 }
1645
1646 /* Send a message */
1647 static void send_to_sock(struct connection *con)
1648 {
1649         int ret = 0;
1650         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1651         struct writequeue_entry *e;
1652         int len, offset;
1653         int count = 0;
1654
1655         mutex_lock(&con->sock_mutex);
1656         if (con->sock == NULL)
1657                 goto out_connect;
1658
1659         spin_lock(&con->writequeue_lock);
1660         for (;;) {
1661                 if (list_empty(&con->writequeue))
1662                         break;
1663
1664                 e = list_first_entry(&con->writequeue, struct writequeue_entry, list);
1665                 len = e->len;
1666                 offset = e->offset;
1667                 BUG_ON(len == 0 && e->users == 0);
1668                 spin_unlock(&con->writequeue_lock);
1669
1670                 ret = 0;
1671                 if (len) {
1672                         ret = kernel_sendpage(con->sock, e->page, offset, len,
1673                                               msg_flags);
1674                         if (ret == -EAGAIN || ret == 0) {
1675                                 if (ret == -EAGAIN &&
1676                                     test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1677                                     !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1678                                         /* Notify TCP that we're limited by the
1679                                          * application window size.
1680                                          */
1681                                         set_bit(SOCK_NOSPACE, &con->sock->flags);
1682                                         con->sock->sk->sk_write_pending++;
1683                                 }
1684                                 cond_resched();
1685                                 goto out;
1686                         } else if (ret < 0)
1687                                 goto out;
1688                 }
1689
1690                 /* Don't starve people filling buffers */
1691                 if (++count >= MAX_SEND_MSG_COUNT) {
1692                         cond_resched();
1693                         count = 0;
1694                 }
1695
1696                 spin_lock(&con->writequeue_lock);
1697                 writequeue_entry_complete(e, ret);
1698         }
1699         spin_unlock(&con->writequeue_lock);
1700
1701         /* close if we got EOF */
1702         if (test_and_clear_bit(CF_EOF, &con->flags)) {
1703                 mutex_unlock(&con->sock_mutex);
1704                 close_connection(con, false, false, true);
1705
1706                 /* handling for tcp shutdown */
1707                 clear_bit(CF_SHUTDOWN, &con->flags);
1708                 wake_up(&con->shutdown_wait);
1709         } else {
1710                 mutex_unlock(&con->sock_mutex);
1711         }
1712
1713         return;
1714
1715 out:
1716         mutex_unlock(&con->sock_mutex);
1717         return;
1718
1719 out_connect:
1720         mutex_unlock(&con->sock_mutex);
1721         queue_work(send_workqueue, &con->swork);
1722         cond_resched();
1723 }
1724
1725 static void clean_one_writequeue(struct connection *con)
1726 {
1727         struct writequeue_entry *e, *safe;
1728
1729         spin_lock(&con->writequeue_lock);
1730         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1731                 free_entry(e);
1732         }
1733         spin_unlock(&con->writequeue_lock);
1734 }
1735
1736 /* Called from recovery when it knows that a node has
1737    left the cluster */
1738 int dlm_lowcomms_close(int nodeid)
1739 {
1740         struct connection *con;
1741         struct dlm_node_addr *na;
1742         int idx;
1743
1744         log_print("closing connection to node %d", nodeid);
1745         idx = srcu_read_lock(&connections_srcu);
1746         con = nodeid2con(nodeid, 0);
1747         if (con) {
1748                 set_bit(CF_CLOSE, &con->flags);
1749                 close_connection(con, true, true, true);
1750                 clean_one_writequeue(con);
1751                 if (con->othercon)
1752                         clean_one_writequeue(con->othercon);
1753         }
1754         srcu_read_unlock(&connections_srcu, idx);
1755
1756         spin_lock(&dlm_node_addrs_spin);
1757         na = find_node_addr(nodeid);
1758         if (na) {
1759                 list_del(&na->list);
1760                 while (na->addr_count--)
1761                         kfree(na->addr[na->addr_count]);
1762                 kfree(na);
1763         }
1764         spin_unlock(&dlm_node_addrs_spin);
1765
1766         return 0;
1767 }
1768
1769 /* Receive workqueue function */
1770 static void process_recv_sockets(struct work_struct *work)
1771 {
1772         struct connection *con = container_of(work, struct connection, rwork);
1773         int err;
1774
1775         clear_bit(CF_READ_PENDING, &con->flags);
1776         do {
1777                 err = receive_from_sock(con);
1778         } while (!err);
1779 }
1780
1781 static void process_listen_recv_socket(struct work_struct *work)
1782 {
1783         accept_from_sock(&listen_con);
1784 }
1785
1786 /* Send workqueue function */
1787 static void process_send_sockets(struct work_struct *work)
1788 {
1789         struct connection *con = container_of(work, struct connection, swork);
1790
1791         WARN_ON(test_bit(CF_IS_OTHERCON, &con->flags));
1792
1793         clear_bit(CF_WRITE_PENDING, &con->flags);
1794
1795         if (test_and_clear_bit(CF_RECONNECT, &con->flags)) {
1796                 close_connection(con, false, false, true);
1797                 dlm_midcomms_unack_msg_resend(con->nodeid);
1798         }
1799
1800         if (con->sock == NULL) { /* not mutex protected so check it inside too */
1801                 if (test_and_clear_bit(CF_DELAY_CONNECT, &con->flags))
1802                         msleep(1000);
1803                 con->connect_action(con);
1804         }
1805         if (!list_empty(&con->writequeue))
1806                 send_to_sock(con);
1807 }
1808
1809 static void work_stop(void)
1810 {
1811         if (recv_workqueue) {
1812                 destroy_workqueue(recv_workqueue);
1813                 recv_workqueue = NULL;
1814         }
1815
1816         if (send_workqueue) {
1817                 destroy_workqueue(send_workqueue);
1818                 send_workqueue = NULL;
1819         }
1820 }
1821
1822 static int work_start(void)
1823 {
1824         recv_workqueue = alloc_ordered_workqueue("dlm_recv", WQ_MEM_RECLAIM);
1825         if (!recv_workqueue) {
1826                 log_print("can't start dlm_recv");
1827                 return -ENOMEM;
1828         }
1829
1830         send_workqueue = alloc_ordered_workqueue("dlm_send", WQ_MEM_RECLAIM);
1831         if (!send_workqueue) {
1832                 log_print("can't start dlm_send");
1833                 destroy_workqueue(recv_workqueue);
1834                 recv_workqueue = NULL;
1835                 return -ENOMEM;
1836         }
1837
1838         return 0;
1839 }
1840
1841 static void shutdown_conn(struct connection *con)
1842 {
1843         if (con->shutdown_action)
1844                 con->shutdown_action(con);
1845 }
1846
1847 void dlm_lowcomms_shutdown(void)
1848 {
1849         int idx;
1850
1851         /* Set all the flags to prevent any
1852          * socket activity.
1853          */
1854         dlm_allow_conn = 0;
1855
1856         if (recv_workqueue)
1857                 flush_workqueue(recv_workqueue);
1858         if (send_workqueue)
1859                 flush_workqueue(send_workqueue);
1860
1861         dlm_close_sock(&listen_con.sock);
1862
1863         idx = srcu_read_lock(&connections_srcu);
1864         foreach_conn(shutdown_conn);
1865         srcu_read_unlock(&connections_srcu, idx);
1866 }
1867
1868 static void _stop_conn(struct connection *con, bool and_other)
1869 {
1870         mutex_lock(&con->sock_mutex);
1871         set_bit(CF_CLOSE, &con->flags);
1872         set_bit(CF_READ_PENDING, &con->flags);
1873         set_bit(CF_WRITE_PENDING, &con->flags);
1874         if (con->sock && con->sock->sk) {
1875                 write_lock_bh(&con->sock->sk->sk_callback_lock);
1876                 con->sock->sk->sk_user_data = NULL;
1877                 write_unlock_bh(&con->sock->sk->sk_callback_lock);
1878         }
1879         if (con->othercon && and_other)
1880                 _stop_conn(con->othercon, false);
1881         mutex_unlock(&con->sock_mutex);
1882 }
1883
1884 static void stop_conn(struct connection *con)
1885 {
1886         _stop_conn(con, true);
1887 }
1888
1889 static void connection_release(struct rcu_head *rcu)
1890 {
1891         struct connection *con = container_of(rcu, struct connection, rcu);
1892
1893         kfree(con->rx_buf);
1894         kfree(con);
1895 }
1896
1897 static void free_conn(struct connection *con)
1898 {
1899         close_connection(con, true, true, true);
1900         spin_lock(&connections_lock);
1901         hlist_del_rcu(&con->list);
1902         spin_unlock(&connections_lock);
1903         if (con->othercon) {
1904                 clean_one_writequeue(con->othercon);
1905                 call_srcu(&connections_srcu, &con->othercon->rcu,
1906                           connection_release);
1907         }
1908         clean_one_writequeue(con);
1909         call_srcu(&connections_srcu, &con->rcu, connection_release);
1910 }
1911
1912 static void work_flush(void)
1913 {
1914         int ok;
1915         int i;
1916         struct connection *con;
1917
1918         do {
1919                 ok = 1;
1920                 foreach_conn(stop_conn);
1921                 if (recv_workqueue)
1922                         flush_workqueue(recv_workqueue);
1923                 if (send_workqueue)
1924                         flush_workqueue(send_workqueue);
1925                 for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1926                         hlist_for_each_entry_rcu(con, &connection_hash[i],
1927                                                  list) {
1928                                 ok &= test_bit(CF_READ_PENDING, &con->flags);
1929                                 ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1930                                 if (con->othercon) {
1931                                         ok &= test_bit(CF_READ_PENDING,
1932                                                        &con->othercon->flags);
1933                                         ok &= test_bit(CF_WRITE_PENDING,
1934                                                        &con->othercon->flags);
1935                                 }
1936                         }
1937                 }
1938         } while (!ok);
1939 }
1940
1941 void dlm_lowcomms_stop(void)
1942 {
1943         int idx;
1944
1945         idx = srcu_read_lock(&connections_srcu);
1946         work_flush();
1947         foreach_conn(free_conn);
1948         srcu_read_unlock(&connections_srcu, idx);
1949         work_stop();
1950         deinit_local();
1951 }
1952
1953 int dlm_lowcomms_start(void)
1954 {
1955         int error = -EINVAL;
1956         int i;
1957
1958         for (i = 0; i < CONN_HASH_SIZE; i++)
1959                 INIT_HLIST_HEAD(&connection_hash[i]);
1960
1961         init_local();
1962         if (!dlm_local_count) {
1963                 error = -ENOTCONN;
1964                 log_print("no local IP address has been set");
1965                 goto fail;
1966         }
1967
1968         INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1969
1970         error = work_start();
1971         if (error)
1972                 goto fail_local;
1973
1974         dlm_allow_conn = 1;
1975
1976         /* Start listening */
1977         switch (dlm_config.ci_protocol) {
1978         case DLM_PROTO_TCP:
1979                 error = tcp_listen_for_all();
1980                 break;
1981         case DLM_PROTO_SCTP:
1982                 error = sctp_listen_for_all(&listen_con);
1983                 break;
1984         default:
1985                 log_print("Invalid protocol identifier %d set",
1986                           dlm_config.ci_protocol);
1987                 error = -EINVAL;
1988                 break;
1989         }
1990         if (error)
1991                 goto fail_unlisten;
1992
1993         return 0;
1994
1995 fail_unlisten:
1996         dlm_allow_conn = 0;
1997         dlm_close_sock(&listen_con.sock);
1998         work_stop();
1999 fail_local:
2000         deinit_local();
2001 fail:
2002         return error;
2003 }
2004
2005 void dlm_lowcomms_exit(void)
2006 {
2007         struct dlm_node_addr *na, *safe;
2008
2009         spin_lock(&dlm_node_addrs_spin);
2010         list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
2011                 list_del(&na->list);
2012                 while (na->addr_count--)
2013                         kfree(na->addr[na->addr_count]);
2014                 kfree(na);
2015         }
2016         spin_unlock(&dlm_node_addrs_spin);
2017 }