fs/dlm/lowcomms.c

   1 /******************************************************************************
   2 *******************************************************************************
   3 **
   4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5 **  Copyright (C) 2004-2009 Red Hat, Inc.  All rights reserved.
   6 **
   7 **  This copyrighted material is made available to anyone wishing to use,
   8 **  modify, copy, or redistribute it subject to the terms and conditions
   9 **  of the GNU General Public License v.2.
  10 **
  11 *******************************************************************************
  12 ******************************************************************************/
  13
  14 /*
  15  * lowcomms.c
  16  *
  17  * This is the "low-level" comms layer.
  18  *
  19  * It is responsible for sending/receiving messages
  20  * from other nodes in the cluster.
  21  *
  22  * Cluster nodes are referred to by their nodeids. nodeids are
  23  * simply 32 bit numbers to the locking module - if they need to
  24  * be expanded for the cluster infrastructure then that is its
  25  * responsibility. It is this layer's
  26  * responsibility to resolve these into IP address or
  27  * whatever it needs for inter-node communication.
  28  *
  29  * The comms level is two kernel threads that deal mainly with
  30  * the receiving of messages from other nodes and passing them
  31  * up to the mid-level comms layer (which understands the
  32  * message format) for execution by the locking core, and
  33  * a send thread which does all the setting up of connections
  34  * to remote nodes and the sending of data. Threads are not allowed
  35  * to send their own data because it may cause them to wait in times
  36  * of high load. Also, this way, the sending thread can collect together
  37  * messages bound for one node and send them in one block.
  38  *
  39  * lowcomms will choose to use either TCP or SCTP as its transport layer
  40  * depending on the configuration variable 'protocol'. This should be set
  41  * to 0 (default) for TCP or 1 for SCTP. It should be configured using a
  42  * cluster-wide mechanism as it must be the same on all nodes of the cluster
  43  * for the DLM to function.
  44  *
  45  */
  46
  47 #include <asm/ioctls.h>
  48 #include <net/sock.h>
  49 #include <net/tcp.h>
  50 #include <linux/pagemap.h>
  51 #include <linux/file.h>
  52 #include <linux/mutex.h>
  53 #include <linux/sctp.h>
  54 #include <linux/slab.h>
  55 #include <linux/sctp.h>
  56 #include <net/sctp/sctp.h>
  57 #include <net/ipv6.h>
  58
  59 #include "dlm_internal.h"
  60 #include "lowcomms.h"
  61 #include "midcomms.h"
  62 #include "config.h"
  63
  64 #define NEEDED_RMEM (4*1024*1024)
  65 #define CONN_HASH_SIZE 32
  66
  67 /* Number of messages to send before rescheduling */
  68 #define MAX_SEND_MSG_COUNT 25
  69
  70 struct cbuf {
  71         unsigned int base;
  72         unsigned int len;
  73         unsigned int mask;
  74 };
  75
  76 static void cbuf_add(struct cbuf *cb, int n)
  77 {
  78         cb->len += n;
  79 }
  80
  81 static int cbuf_data(struct cbuf *cb)
  82 {
  83         return ((cb->base + cb->len) & cb->mask);
  84 }
  85
  86 static void cbuf_init(struct cbuf *cb, int size)
  87 {
  88         cb->base = cb->len = 0;
  89         cb->mask = size-1;
  90 }
  91
  92 static void cbuf_eat(struct cbuf *cb, int n)
  93 {
  94         cb->len  -= n;
  95         cb->base += n;
  96         cb->base &= cb->mask;
  97 }
  98
  99 static bool cbuf_empty(struct cbuf *cb)
 100 {
 101         return cb->len == 0;
 102 }
 103
 104 struct connection {
 105         struct socket *sock;    /* NULL if not connected */
 106         uint32_t nodeid;        /* So we know who we are in the list */
 107         struct mutex sock_mutex;
 108         unsigned long flags;
 109 #define CF_READ_PENDING 1
 110 #define CF_WRITE_PENDING 2
 111 #define CF_CONNECT_PENDING 3
 112 #define CF_INIT_PENDING 4
 113 #define CF_IS_OTHERCON 5
 114 #define CF_CLOSE 6
 115 #define CF_APP_LIMITED 7
 116         struct list_head writequeue;  /* List of outgoing writequeue_entries */
 117         spinlock_t writequeue_lock;
 118         int (*rx_action) (struct connection *); /* What to do when active */
 119         void (*connect_action) (struct connection *);   /* What to do to connect */
 120         struct page *rx_page;
 121         struct cbuf cb;
 122         int retries;
 123 #define MAX_CONNECT_RETRIES 3
 124         int sctp_assoc;
 125         struct hlist_node list;
 126         struct connection *othercon;
 127         struct work_struct rwork; /* Receive workqueue */
 128         struct work_struct swork; /* Send workqueue */
 129         bool try_new_addr;
 130 };
 131 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 132
 133 /* An entry waiting to be sent */
 134 struct writequeue_entry {
 135         struct list_head list;
 136         struct page *page;
 137         int offset;
 138         int len;
 139         int end;
 140         int users;
 141         struct connection *con;
 142 };
 143
 144 struct dlm_node_addr {
 145         struct list_head list;
 146         int nodeid;
 147         int addr_count;
 148         int curr_addr_index;
 149         struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
 150 };
 151
 152 static LIST_HEAD(dlm_node_addrs);
 153 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
 154
 155 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 156 static int dlm_local_count;
 157 static int dlm_allow_conn;
 158
 159 /* Work queues */
 160 static struct workqueue_struct *recv_workqueue;
 161 static struct workqueue_struct *send_workqueue;
 162
 163 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 164 static DEFINE_MUTEX(connections_lock);
 165 static struct kmem_cache *con_cache;
 166
 167 static void process_recv_sockets(struct work_struct *work);
 168 static void process_send_sockets(struct work_struct *work);
 169
 170
 171 /* This is deliberately very simple because most clusters have simple
 172    sequential nodeids, so we should be able to go straight to a connection
 173    struct in the array */
 174 static inline int nodeid_hash(int nodeid)
 175 {
 176         return nodeid & (CONN_HASH_SIZE-1);
 177 }
 178
 179 static struct connection *__find_con(int nodeid)
 180 {
 181         int r;
 182         struct connection *con;
 183
 184         r = nodeid_hash(nodeid);
 185
 186         hlist_for_each_entry(con, &connection_hash[r], list) {
 187                 if (con->nodeid == nodeid)
 188                         return con;
 189         }
 190         return NULL;
 191 }
 192
 193 /*
 194  * If 'allocation' is zero then we don't attempt to create a new
 195  * connection structure for this node.
 196  */
 197 static struct connection *__nodeid2con(int nodeid, gfp_t alloc)
 198 {
 199         struct connection *con = NULL;
 200         int r;
 201
 202         con = __find_con(nodeid);
 203         if (con || !alloc)
 204                 return con;
 205
 206         con = kmem_cache_zalloc(con_cache, alloc);
 207         if (!con)
 208                 return NULL;
 209
 210         r = nodeid_hash(nodeid);
 211         hlist_add_head(&con->list, &connection_hash[r]);
 212
 213         con->nodeid = nodeid;
 214         mutex_init(&con->sock_mutex);
 215         INIT_LIST_HEAD(&con->writequeue);
 216         spin_lock_init(&con->writequeue_lock);
 217         INIT_WORK(&con->swork, process_send_sockets);
 218         INIT_WORK(&con->rwork, process_recv_sockets);
 219
 220         /* Setup action pointers for child sockets */
 221         if (con->nodeid) {
 222                 struct connection *zerocon = __find_con(0);
 223
 224                 con->connect_action = zerocon->connect_action;
 225                 if (!con->rx_action)
 226                         con->rx_action = zerocon->rx_action;
 227         }
 228
 229         return con;
 230 }
 231
 232 /* Loop round all connections */
 233 static void foreach_conn(void (*conn_func)(struct connection *c))
 234 {
 235         int i;
 236         struct hlist_node *n;
 237         struct connection *con;
 238
 239         for (i = 0; i < CONN_HASH_SIZE; i++) {
 240                 hlist_for_each_entry_safe(con, n, &connection_hash[i], list)
 241                         conn_func(con);
 242         }
 243 }
 244
 245 static struct connection *nodeid2con(int nodeid, gfp_t allocation)
 246 {
 247         struct connection *con;
 248
 249         mutex_lock(&connections_lock);
 250         con = __nodeid2con(nodeid, allocation);
 251         mutex_unlock(&connections_lock);
 252
 253         return con;
 254 }
 255
 256 /* This is a bit drastic, but only called when things go wrong */
 257 static struct connection *assoc2con(int assoc_id)
 258 {
 259         int i;
 260         struct connection *con;
 261
 262         mutex_lock(&connections_lock);
 263
 264         for (i = 0 ; i < CONN_HASH_SIZE; i++) {
 265                 hlist_for_each_entry(con, &connection_hash[i], list) {
 266                         if (con->sctp_assoc == assoc_id) {
 267                                 mutex_unlock(&connections_lock);
 268                                 return con;
 269                         }
 270                 }
 271         }
 272         mutex_unlock(&connections_lock);
 273         return NULL;
 274 }
 275
 276 static struct dlm_node_addr *find_node_addr(int nodeid)
 277 {
 278         struct dlm_node_addr *na;
 279
 280         list_for_each_entry(na, &dlm_node_addrs, list) {
 281                 if (na->nodeid == nodeid)
 282                         return na;
 283         }
 284         return NULL;
 285 }
 286
 287 static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y)
 288 {
 289         switch (x->ss_family) {
 290         case AF_INET: {
 291                 struct sockaddr_in *sinx = (struct sockaddr_in *)x;
 292                 struct sockaddr_in *siny = (struct sockaddr_in *)y;
 293                 if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
 294                         return 0;
 295                 if (sinx->sin_port != siny->sin_port)
 296                         return 0;
 297                 break;
 298         }
 299         case AF_INET6: {
 300                 struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
 301                 struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
 302                 if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
 303                         return 0;
 304                 if (sinx->sin6_port != siny->sin6_port)
 305                         return 0;
 306                 break;
 307         }
 308         default:
 309                 return 0;
 310         }
 311         return 1;
 312 }
 313
 314 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
 315                           struct sockaddr *sa_out, bool try_new_addr)
 316 {
 317         struct sockaddr_storage sas;
 318         struct dlm_node_addr *na;
 319
 320         if (!dlm_local_count)
 321                 return -1;
 322
 323         spin_lock(&dlm_node_addrs_spin);
 324         na = find_node_addr(nodeid);
 325         if (na && na->addr_count) {
 326                 if (try_new_addr) {
 327                         na->curr_addr_index++;
 328                         if (na->curr_addr_index == na->addr_count)
 329                                 na->curr_addr_index = 0;
 330                 }
 331
 332                 memcpy(&sas, na->addr[na->curr_addr_index ],
 333                         sizeof(struct sockaddr_storage));
 334         }
 335         spin_unlock(&dlm_node_addrs_spin);
 336
 337         if (!na)
 338                 return -EEXIST;
 339
 340         if (!na->addr_count)
 341                 return -ENOENT;
 342
 343         if (sas_out)
 344                 memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
 345
 346         if (!sa_out)
 347                 return 0;
 348
 349         if (dlm_local_addr[0]->ss_family == AF_INET) {
 350                 struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
 351                 struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
 352                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 353         } else {
 354                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
 355                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
 356                 ret6->sin6_addr = in6->sin6_addr;
 357         }
 358
 359         return 0;
 360 }
 361
 362 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid)
 363 {
 364         struct dlm_node_addr *na;
 365         int rv = -EEXIST;
 366         int addr_i;
 367
 368         spin_lock(&dlm_node_addrs_spin);
 369         list_for_each_entry(na, &dlm_node_addrs, list) {
 370                 if (!na->addr_count)
 371                         continue;
 372
 373                 for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
 374                         if (addr_compare(na->addr[addr_i], addr)) {
 375                                 *nodeid = na->nodeid;
 376                                 rv = 0;
 377                                 goto unlock;
 378                         }
 379                 }
 380         }
 381 unlock:
 382         spin_unlock(&dlm_node_addrs_spin);
 383         return rv;
 384 }
 385
 386 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 387 {
 388         struct sockaddr_storage *new_addr;
 389         struct dlm_node_addr *new_node, *na;
 390
 391         new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
 392         if (!new_node)
 393                 return -ENOMEM;
 394
 395         new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
 396         if (!new_addr) {
 397                 kfree(new_node);
 398                 return -ENOMEM;
 399         }
 400
 401         memcpy(new_addr, addr, len);
 402
 403         spin_lock(&dlm_node_addrs_spin);
 404         na = find_node_addr(nodeid);
 405         if (!na) {
 406                 new_node->nodeid = nodeid;
 407                 new_node->addr[0] = new_addr;
 408                 new_node->addr_count = 1;
 409                 list_add(&new_node->list, &dlm_node_addrs);
 410                 spin_unlock(&dlm_node_addrs_spin);
 411                 return 0;
 412         }
 413
 414         if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
 415                 spin_unlock(&dlm_node_addrs_spin);
 416                 kfree(new_addr);
 417                 kfree(new_node);
 418                 return -ENOSPC;
 419         }
 420
 421         na->addr[na->addr_count++] = new_addr;
 422         spin_unlock(&dlm_node_addrs_spin);
 423         kfree(new_node);
 424         return 0;
 425 }
 426
 427 /* Data available on socket or listen socket received a connect */
 428 static void lowcomms_data_ready(struct sock *sk, int count_unused)
 429 {
 430         struct connection *con = sock2con(sk);
 431         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 432                 queue_work(recv_workqueue, &con->rwork);
 433 }
 434
 435 static void lowcomms_write_space(struct sock *sk)
 436 {
 437         struct connection *con = sock2con(sk);
 438
 439         if (!con)
 440                 return;
 441
 442         clear_bit(SOCK_NOSPACE, &con->sock->flags);
 443
 444         if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
 445                 con->sock->sk->sk_write_pending--;
 446                 clear_bit(SOCK_ASYNC_NOSPACE, &con->sock->flags);
 447         }
 448
 449         if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags))
 450                 queue_work(send_workqueue, &con->swork);
 451 }
 452
 453 static inline void lowcomms_connect_sock(struct connection *con)
 454 {
 455         if (test_bit(CF_CLOSE, &con->flags))
 456                 return;
 457         if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags))
 458                 queue_work(send_workqueue, &con->swork);
 459 }
 460
 461 static void lowcomms_state_change(struct sock *sk)
 462 {
 463         if (sk->sk_state == TCP_ESTABLISHED)
 464                 lowcomms_write_space(sk);
 465 }
 466
 467 int dlm_lowcomms_connect_node(int nodeid)
 468 {
 469         struct connection *con;
 470
 471         /* with sctp there's no connecting without sending */
 472         if (dlm_config.ci_protocol != 0)
 473                 return 0;
 474
 475         if (nodeid == dlm_our_nodeid())
 476                 return 0;
 477
 478         con = nodeid2con(nodeid, GFP_NOFS);
 479         if (!con)
 480                 return -ENOMEM;
 481         lowcomms_connect_sock(con);
 482         return 0;
 483 }
 484
 485 /* Make a socket active */
 486 static void add_sock(struct socket *sock, struct connection *con)
 487 {
 488         con->sock = sock;
 489
 490         /* Install a data_ready callback */
 491         con->sock->sk->sk_data_ready = lowcomms_data_ready;
 492         con->sock->sk->sk_write_space = lowcomms_write_space;
 493         con->sock->sk->sk_state_change = lowcomms_state_change;
 494         con->sock->sk->sk_user_data = con;
 495         con->sock->sk->sk_allocation = GFP_NOFS;
 496 }
 497
 498 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 499    length */
 500 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 501                           int *addr_len)
 502 {
 503         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 504         if (saddr->ss_family == AF_INET) {
 505                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 506                 in4_addr->sin_port = cpu_to_be16(port);
 507                 *addr_len = sizeof(struct sockaddr_in);
 508                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 509         } else {
 510                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 511                 in6_addr->sin6_port = cpu_to_be16(port);
 512                 *addr_len = sizeof(struct sockaddr_in6);
 513         }
 514         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 515 }
 516
 517 /* Close a remote connection and tidy up */
 518 static void close_connection(struct connection *con, bool and_other)
 519 {
 520         mutex_lock(&con->sock_mutex);
 521
 522         if (con->sock) {
 523                 sock_release(con->sock);
 524                 con->sock = NULL;
 525         }
 526         if (con->othercon && and_other) {
 527                 /* Will only re-enter once. */
 528                 close_connection(con->othercon, false);
 529         }
 530         if (con->rx_page) {
 531                 __free_page(con->rx_page);
 532                 con->rx_page = NULL;
 533         }
 534
 535         con->retries = 0;
 536         mutex_unlock(&con->sock_mutex);
 537 }
 538
 539 /* We only send shutdown messages to nodes that are not part of the cluster */
 540 static void sctp_send_shutdown(sctp_assoc_t associd)
 541 {
 542         static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
 543         struct msghdr outmessage;
 544         struct cmsghdr *cmsg;
 545         struct sctp_sndrcvinfo *sinfo;
 546         int ret;
 547         struct connection *con;
 548
 549         con = nodeid2con(0,0);
 550         BUG_ON(con == NULL);
 551
 552         outmessage.msg_name = NULL;
 553         outmessage.msg_namelen = 0;
 554         outmessage.msg_control = outcmsg;
 555         outmessage.msg_controllen = sizeof(outcmsg);
 556         outmessage.msg_flags = MSG_EOR;
 557
 558         cmsg = CMSG_FIRSTHDR(&outmessage);
 559         cmsg->cmsg_level = IPPROTO_SCTP;
 560         cmsg->cmsg_type = SCTP_SNDRCV;
 561         cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
 562         outmessage.msg_controllen = cmsg->cmsg_len;
 563         sinfo = CMSG_DATA(cmsg);
 564         memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
 565
 566         sinfo->sinfo_flags |= MSG_EOF;
 567         sinfo->sinfo_assoc_id = associd;
 568
 569         ret = kernel_sendmsg(con->sock, &outmessage, NULL, 0, 0);
 570
 571         if (ret != 0)
 572                 log_print("send EOF to node failed: %d", ret);
 573 }
 574
 575 static void sctp_init_failed_foreach(struct connection *con)
 576 {
 577
 578         /*
 579          * Don't try to recover base con and handle race where the
 580          * other node's assoc init creates a assoc and we get that
 581          * notification, then we get a notification that our attempt
 582          * failed due. This happens when we are still trying the primary
 583          * address, but the other node has already tried secondary addrs
 584          * and found one that worked.
 585          */
 586         if (!con->nodeid || con->sctp_assoc)
 587                 return;
 588
 589         log_print("Retrying SCTP association init for node %d\n", con->nodeid);
 590
 591         con->try_new_addr = true;
 592         con->sctp_assoc = 0;
 593         if (test_and_clear_bit(CF_INIT_PENDING, &con->flags)) {
 594                 if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags))
 595                         queue_work(send_workqueue, &con->swork);
 596         }
 597 }
 598
 599 /* INIT failed but we don't know which node...
 600    restart INIT on all pending nodes */
 601 static void sctp_init_failed(void)
 602 {
 603         mutex_lock(&connections_lock);
 604
 605         foreach_conn(sctp_init_failed_foreach);
 606
 607         mutex_unlock(&connections_lock);
 608 }
 609
 610 /* Something happened to an association */
 611 static void process_sctp_notification(struct connection *con,
 612                                       struct msghdr *msg, char *buf)
 613 {
 614         union sctp_notification *sn = (union sctp_notification *)buf;
 615
 616         if (sn->sn_header.sn_type == SCTP_ASSOC_CHANGE) {
 617                 switch (sn->sn_assoc_change.sac_state) {
 618
 619                 case SCTP_COMM_UP:
 620                 case SCTP_RESTART:
 621                 {
 622                         /* Check that the new node is in the lockspace */
 623                         struct sctp_prim prim;
 624                         int nodeid;
 625                         int prim_len, ret;
 626                         int addr_len;
 627                         struct connection *new_con;
 628
 629                         /*
 630                          * We get this before any data for an association.
 631                          * We verify that the node is in the cluster and
 632                          * then peel off a socket for it.
 633                          */
 634                         if ((int)sn->sn_assoc_change.sac_assoc_id <= 0) {
 635                                 log_print("COMM_UP for invalid assoc ID %d",
 636                                          (int)sn->sn_assoc_change.sac_assoc_id);
 637                                 sctp_init_failed();
 638                                 return;
 639                         }
 640                         memset(&prim, 0, sizeof(struct sctp_prim));
 641                         prim_len = sizeof(struct sctp_prim);
 642                         prim.ssp_assoc_id = sn->sn_assoc_change.sac_assoc_id;
 643
 644                         ret = kernel_getsockopt(con->sock,
 645                                                 IPPROTO_SCTP,
 646                                                 SCTP_PRIMARY_ADDR,
 647                                                 (char*)&prim,
 648                                                 &prim_len);
 649                         if (ret < 0) {
 650                                 log_print("getsockopt/sctp_primary_addr on "
 651                                           "new assoc %d failed : %d",
 652                                           (int)sn->sn_assoc_change.sac_assoc_id,
 653                                           ret);
 654
 655                                 /* Retry INIT later */
 656                                 new_con = assoc2con(sn->sn_assoc_change.sac_assoc_id);
 657                                 if (new_con)
 658                                         clear_bit(CF_CONNECT_PENDING, &con->flags);
 659                                 return;
 660                         }
 661                         make_sockaddr(&prim.ssp_addr, 0, &addr_len);
 662                         if (addr_to_nodeid(&prim.ssp_addr, &nodeid)) {
 663                                 unsigned char *b=(unsigned char *)&prim.ssp_addr;
 664                                 log_print("reject connect from unknown addr");
 665                                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
 666                                                      b, sizeof(struct sockaddr_storage));
 667                                 sctp_send_shutdown(prim.ssp_assoc_id);
 668                                 return;
 669                         }
 670
 671                         new_con = nodeid2con(nodeid, GFP_NOFS);
 672                         if (!new_con)
 673                                 return;
 674
 675                         /* Peel off a new sock */
 676                         sctp_lock_sock(con->sock->sk);
 677                         ret = sctp_do_peeloff(con->sock->sk,
 678                                 sn->sn_assoc_change.sac_assoc_id,
 679                                 &new_con->sock);
 680                         sctp_release_sock(con->sock->sk);
 681                         if (ret < 0) {
 682                                 log_print("Can't peel off a socket for "
 683                                           "connection %d to node %d: err=%d",
 684                                           (int)sn->sn_assoc_change.sac_assoc_id,
 685                                           nodeid, ret);
 686                                 return;
 687                         }
 688                         add_sock(new_con->sock, new_con);
 689
 690                         log_print("connecting to %d sctp association %d",
 691                                  nodeid, (int)sn->sn_assoc_change.sac_assoc_id);
 692
 693                         new_con->sctp_assoc = sn->sn_assoc_change.sac_assoc_id;
 694                         new_con->try_new_addr = false;
 695                         /* Send any pending writes */
 696                         clear_bit(CF_CONNECT_PENDING, &new_con->flags);
 697                         clear_bit(CF_INIT_PENDING, &new_con->flags);
 698                         if (!test_and_set_bit(CF_WRITE_PENDING, &new_con->flags)) {
 699                                 queue_work(send_workqueue, &new_con->swork);
 700                         }
 701                         if (!test_and_set_bit(CF_READ_PENDING, &new_con->flags))
 702                                 queue_work(recv_workqueue, &new_con->rwork);
 703                 }
 704                 break;
 705
 706                 case SCTP_COMM_LOST:
 707                 case SCTP_SHUTDOWN_COMP:
 708                 {
 709                         con = assoc2con(sn->sn_assoc_change.sac_assoc_id);
 710                         if (con) {
 711                                 con->sctp_assoc = 0;
 712                         }
 713                 }
 714                 break;
 715
 716                 /* We don't know which INIT failed, so clear the PENDING flags
 717                  * on them all.  if assoc_id is zero then it will then try
 718                  * again */
 719
 720                 case SCTP_CANT_STR_ASSOC:
 721                 {
 722                         log_print("Can't start SCTP association - retrying");
 723                         sctp_init_failed();
 724                 }
 725                 break;
 726
 727                 default:
 728                         log_print("unexpected SCTP assoc change id=%d state=%d",
 729                                   (int)sn->sn_assoc_change.sac_assoc_id,
 730                                   sn->sn_assoc_change.sac_state);
 731                 }
 732         }
 733 }
 734
 735 /* Data received from remote end */
 736 static int receive_from_sock(struct connection *con)
 737 {
 738         int ret = 0;
 739         struct msghdr msg = {};
 740         struct kvec iov[2];
 741         unsigned len;
 742         int r;
 743         int call_again_soon = 0;
 744         int nvec;
 745         char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
 746
 747         mutex_lock(&con->sock_mutex);
 748
 749         if (con->sock == NULL) {
 750                 ret = -EAGAIN;
 751                 goto out_close;
 752         }
 753
 754         if (con->rx_page == NULL) {
 755                 /*
 756                  * This doesn't need to be atomic, but I think it should
 757                  * improve performance if it is.
 758                  */
 759                 con->rx_page = alloc_page(GFP_ATOMIC);
 760                 if (con->rx_page == NULL)
 761                         goto out_resched;
 762                 cbuf_init(&con->cb, PAGE_CACHE_SIZE);
 763         }
 764
 765         /* Only SCTP needs these really */
 766         memset(&incmsg, 0, sizeof(incmsg));
 767         msg.msg_control = incmsg;
 768         msg.msg_controllen = sizeof(incmsg);
 769
 770         /*
 771          * iov[0] is the bit of the circular buffer between the current end
 772          * point (cb.base + cb.len) and the end of the buffer.
 773          */
 774         iov[0].iov_len = con->cb.base - cbuf_data(&con->cb);
 775         iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb);
 776         iov[1].iov_len = 0;
 777         nvec = 1;
 778
 779         /*
 780          * iov[1] is the bit of the circular buffer between the start of the
 781          * buffer and the start of the currently used section (cb.base)
 782          */
 783         if (cbuf_data(&con->cb) >= con->cb.base) {
 784                 iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb);
 785                 iov[1].iov_len = con->cb.base;
 786                 iov[1].iov_base = page_address(con->rx_page);
 787                 nvec = 2;
 788         }
 789         len = iov[0].iov_len + iov[1].iov_len;
 790
 791         r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len,
 792                                MSG_DONTWAIT | MSG_NOSIGNAL);
 793         if (ret <= 0)
 794                 goto out_close;
 795
 796         /* Process SCTP notifications */
 797         if (msg.msg_flags & MSG_NOTIFICATION) {
 798                 msg.msg_control = incmsg;
 799                 msg.msg_controllen = sizeof(incmsg);
 800
 801                 process_sctp_notification(con, &msg,
 802                                 page_address(con->rx_page) + con->cb.base);
 803                 mutex_unlock(&con->sock_mutex);
 804                 return 0;
 805         }
 806         BUG_ON(con->nodeid == 0);
 807
 808         if (ret == len)
 809                 call_again_soon = 1;
 810         cbuf_add(&con->cb, ret);
 811         ret = dlm_process_incoming_buffer(con->nodeid,
 812                                           page_address(con->rx_page),
 813                                           con->cb.base, con->cb.len,
 814                                           PAGE_CACHE_SIZE);
 815         if (ret == -EBADMSG) {
 816                 log_print("lowcomms: addr=%p, base=%u, len=%u, "
 817                           "iov_len=%u, iov_base[0]=%p, read=%d",
 818                           page_address(con->rx_page), con->cb.base, con->cb.len,
 819                           len, iov[0].iov_base, r);
 820         }
 821         if (ret < 0)
 822                 goto out_close;
 823         cbuf_eat(&con->cb, ret);
 824
 825         if (cbuf_empty(&con->cb) && !call_again_soon) {
 826                 __free_page(con->rx_page);
 827                 con->rx_page = NULL;
 828         }
 829
 830         if (call_again_soon)
 831                 goto out_resched;
 832         mutex_unlock(&con->sock_mutex);
 833         return 0;
 834
 835 out_resched:
 836         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 837                 queue_work(recv_workqueue, &con->rwork);
 838         mutex_unlock(&con->sock_mutex);
 839         return -EAGAIN;
 840
 841 out_close:
 842         mutex_unlock(&con->sock_mutex);
 843         if (ret != -EAGAIN) {
 844                 close_connection(con, false);
 845                 /* Reconnect when there is something to send */
 846         }
 847         /* Don't return success if we really got EOF */
 848         if (ret == 0)
 849                 ret = -EAGAIN;
 850
 851         return ret;
 852 }
 853
 854 /* Listening socket is busy, accept a connection */
 855 static int tcp_accept_from_sock(struct connection *con)
 856 {
 857         int result;
 858         struct sockaddr_storage peeraddr;
 859         struct socket *newsock;
 860         int len;
 861         int nodeid;
 862         struct connection *newcon;
 863         struct connection *addcon;
 864
 865         mutex_lock(&connections_lock);
 866         if (!dlm_allow_conn) {
 867                 mutex_unlock(&connections_lock);
 868                 return -1;
 869         }
 870         mutex_unlock(&connections_lock);
 871
 872         memset(&peeraddr, 0, sizeof(peeraddr));
 873         result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
 874                                   IPPROTO_TCP, &newsock);
 875         if (result < 0)
 876                 return -ENOMEM;
 877
 878         mutex_lock_nested(&con->sock_mutex, 0);
 879
 880         result = -ENOTCONN;
 881         if (con->sock == NULL)
 882                 goto accept_err;
 883
 884         newsock->type = con->sock->type;
 885         newsock->ops = con->sock->ops;
 886
 887         result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK);
 888         if (result < 0)
 889                 goto accept_err;
 890
 891         /* Get the connected socket's peer */
 892         memset(&peeraddr, 0, sizeof(peeraddr));
 893         if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr,
 894                                   &len, 2)) {
 895                 result = -ECONNABORTED;
 896                 goto accept_err;
 897         }
 898
 899         /* Get the new node's NODEID */
 900         make_sockaddr(&peeraddr, 0, &len);
 901         if (addr_to_nodeid(&peeraddr, &nodeid)) {
 902                 unsigned char *b=(unsigned char *)&peeraddr;
 903                 log_print("connect from non cluster node");
 904                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
 905                                      b, sizeof(struct sockaddr_storage));
 906                 sock_release(newsock);
 907                 mutex_unlock(&con->sock_mutex);
 908                 return -1;
 909         }
 910
 911         log_print("got connection from %d", nodeid);
 912
 913         /*  Check to see if we already have a connection to this node. This
 914          *  could happen if the two nodes initiate a connection at roughly
 915          *  the same time and the connections cross on the wire.
 916          *  In this case we store the incoming one in "othercon"
 917          */
 918         newcon = nodeid2con(nodeid, GFP_NOFS);
 919         if (!newcon) {
 920                 result = -ENOMEM;
 921                 goto accept_err;
 922         }
 923         mutex_lock_nested(&newcon->sock_mutex, 1);
 924         if (newcon->sock) {
 925                 struct connection *othercon = newcon->othercon;
 926
 927                 if (!othercon) {
 928                         othercon = kmem_cache_zalloc(con_cache, GFP_NOFS);
 929                         if (!othercon) {
 930                                 log_print("failed to allocate incoming socket");
 931                                 mutex_unlock(&newcon->sock_mutex);
 932                                 result = -ENOMEM;
 933                                 goto accept_err;
 934                         }
 935                         othercon->nodeid = nodeid;
 936                         othercon->rx_action = receive_from_sock;
 937                         mutex_init(&othercon->sock_mutex);
 938                         INIT_WORK(&othercon->swork, process_send_sockets);
 939                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 940                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 941                 }
 942                 if (!othercon->sock) {
 943                         newcon->othercon = othercon;
 944                         othercon->sock = newsock;
 945                         newsock->sk->sk_user_data = othercon;
 946                         add_sock(newsock, othercon);
 947                         addcon = othercon;
 948                 }
 949                 else {
 950                         printk("Extra connection from node %d attempted\n", nodeid);
 951                         result = -EAGAIN;
 952                         mutex_unlock(&newcon->sock_mutex);
 953                         goto accept_err;
 954                 }
 955         }
 956         else {
 957                 newsock->sk->sk_user_data = newcon;
 958                 newcon->rx_action = receive_from_sock;
 959                 add_sock(newsock, newcon);
 960                 addcon = newcon;
 961         }
 962
 963         mutex_unlock(&newcon->sock_mutex);
 964
 965         /*
 966          * Add it to the active queue in case we got data
 967          * between processing the accept adding the socket
 968          * to the read_sockets list
 969          */
 970         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 971                 queue_work(recv_workqueue, &addcon->rwork);
 972         mutex_unlock(&con->sock_mutex);
 973
 974         return 0;
 975
 976 accept_err:
 977         mutex_unlock(&con->sock_mutex);
 978         sock_release(newsock);
 979
 980         if (result != -EAGAIN)
 981                 log_print("error accepting connection from node: %d", result);
 982         return result;
 983 }
 984
 985 static void free_entry(struct writequeue_entry *e)
 986 {
 987         __free_page(e->page);
 988         kfree(e);
 989 }
 990
 991 /* Initiate an SCTP association.
 992    This is a special case of send_to_sock() in that we don't yet have a
 993    peeled-off socket for this association, so we use the listening socket
 994    and add the primary IP address of the remote node.
 995  */
 996 static void sctp_init_assoc(struct connection *con)
 997 {
 998         struct sockaddr_storage rem_addr;
 999         char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
1000         struct msghdr outmessage;
1001         struct cmsghdr *cmsg;
1002         struct sctp_sndrcvinfo *sinfo;
1003         struct connection *base_con;
1004         struct writequeue_entry *e;
1005         int len, offset;
1006         int ret;
1007         int addrlen;
1008         struct kvec iov[1];
1009
1010         if (test_and_set_bit(CF_INIT_PENDING, &con->flags))
1011                 return;
1012
1013         if (con->retries++ > MAX_CONNECT_RETRIES)
1014                 return;
1015
1016         if (nodeid_to_addr(con->nodeid, NULL, (struct sockaddr *)&rem_addr,
1017                            con->try_new_addr)) {
1018                 log_print("no address for nodeid %d", con->nodeid);
1019                 return;
1020         }
1021         base_con = nodeid2con(0, 0);
1022         BUG_ON(base_con == NULL);
1023
1024         make_sockaddr(&rem_addr, dlm_config.ci_tcp_port, &addrlen);
1025
1026         outmessage.msg_name = &rem_addr;
1027         outmessage.msg_namelen = addrlen;
1028         outmessage.msg_control = outcmsg;
1029         outmessage.msg_controllen = sizeof(outcmsg);
1030         outmessage.msg_flags = MSG_EOR;
1031
1032         spin_lock(&con->writequeue_lock);
1033
1034         if (list_empty(&con->writequeue)) {
1035                 spin_unlock(&con->writequeue_lock);
1036                 log_print("writequeue empty for nodeid %d", con->nodeid);
1037                 return;
1038         }
1039
1040         e = list_first_entry(&con->writequeue, struct writequeue_entry, list);
1041         len = e->len;
1042         offset = e->offset;
1043         spin_unlock(&con->writequeue_lock);
1044
1045         /* Send the first block off the write queue */
1046         iov[0].iov_base = page_address(e->page)+offset;
1047         iov[0].iov_len = len;
1048
1049         if (rem_addr.ss_family == AF_INET) {
1050                 struct sockaddr_in *sin = (struct sockaddr_in *)&rem_addr;
1051                 log_print("Trying to connect to %pI4", &sin->sin_addr.s_addr);
1052         } else {
1053                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&rem_addr;
1054                 log_print("Trying to connect to %pI6", &sin6->sin6_addr);
1055         }
1056
1057         cmsg = CMSG_FIRSTHDR(&outmessage);
1058         cmsg->cmsg_level = IPPROTO_SCTP;
1059         cmsg->cmsg_type = SCTP_SNDRCV;
1060         cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
1061         sinfo = CMSG_DATA(cmsg);
1062         memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
1063         sinfo->sinfo_ppid = cpu_to_le32(dlm_our_nodeid());
1064         outmessage.msg_controllen = cmsg->cmsg_len;
1065         sinfo->sinfo_flags |= SCTP_ADDR_OVER;
1066
1067         ret = kernel_sendmsg(base_con->sock, &outmessage, iov, 1, len);
1068         if (ret < 0) {
1069                 log_print("Send first packet to node %d failed: %d",
1070                           con->nodeid, ret);
1071
1072                 /* Try again later */
1073                 clear_bit(CF_CONNECT_PENDING, &con->flags);
1074                 clear_bit(CF_INIT_PENDING, &con->flags);
1075         }
1076         else {
1077                 spin_lock(&con->writequeue_lock);
1078                 e->offset += ret;
1079                 e->len -= ret;
1080
1081                 if (e->len == 0 && e->users == 0) {
1082                         list_del(&e->list);
1083                         free_entry(e);
1084                 }
1085                 spin_unlock(&con->writequeue_lock);
1086         }
1087 }
1088
1089 /* Connect a new socket to its peer */
1090 static void tcp_connect_to_sock(struct connection *con)
1091 {
1092         struct sockaddr_storage saddr, src_addr;
1093         int addr_len;
1094         struct socket *sock = NULL;
1095         int one = 1;
1096         int result;
1097
1098         if (con->nodeid == 0) {
1099                 log_print("attempt to connect sock 0 foiled");
1100                 return;
1101         }
1102
1103         mutex_lock(&con->sock_mutex);
1104         if (con->retries++ > MAX_CONNECT_RETRIES)
1105                 goto out;
1106
1107         /* Some odd races can cause double-connects, ignore them */
1108         if (con->sock)
1109                 goto out;
1110
1111         /* Create a socket to communicate with */
1112         result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
1113                                   IPPROTO_TCP, &sock);
1114         if (result < 0)
1115                 goto out_err;
1116
1117         memset(&saddr, 0, sizeof(saddr));
1118         result = nodeid_to_addr(con->nodeid, &saddr, NULL, false);
1119         if (result < 0) {
1120                 log_print("no address for nodeid %d", con->nodeid);
1121                 goto out_err;
1122         }
1123
1124         sock->sk->sk_user_data = con;
1125         con->rx_action = receive_from_sock;
1126         con->connect_action = tcp_connect_to_sock;
1127         add_sock(sock, con);
1128
1129         /* Bind to our cluster-known address connecting to avoid
1130            routing problems */
1131         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1132         make_sockaddr(&src_addr, 0, &addr_len);
1133         result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1134                                  addr_len);
1135         if (result < 0) {
1136                 log_print("could not bind for connect: %d", result);
1137                 /* This *may* not indicate a critical error */
1138         }
1139
1140         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1141
1142         log_print("connecting to %d", con->nodeid);
1143
1144         /* Turn off Nagle's algorithm */
1145         kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
1146                           sizeof(one));
1147
1148         result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1149                                    O_NONBLOCK);
1150         if (result == -EINPROGRESS)
1151                 result = 0;
1152         if (result == 0)
1153                 goto out;
1154
1155 out_err:
1156         if (con->sock) {
1157                 sock_release(con->sock);
1158                 con->sock = NULL;
1159         } else if (sock) {
1160                 sock_release(sock);
1161         }
1162         /*
1163          * Some errors are fatal and this list might need adjusting. For other
1164          * errors we try again until the max number of retries is reached.
1165          */
1166         if (result != -EHOSTUNREACH &&
1167             result != -ENETUNREACH &&
1168             result != -ENETDOWN &&
1169             result != -EINVAL &&
1170             result != -EPROTONOSUPPORT) {
1171                 log_print("connect %d try %d error %d", con->nodeid,
1172                           con->retries, result);
1173                 mutex_unlock(&con->sock_mutex);
1174                 msleep(1000);
1175                 lowcomms_connect_sock(con);
1176                 return;
1177         }
1178 out:
1179         mutex_unlock(&con->sock_mutex);
1180         return;
1181 }
1182
1183 static struct socket *tcp_create_listen_sock(struct connection *con,
1184                                              struct sockaddr_storage *saddr)
1185 {
1186         struct socket *sock = NULL;
1187         int result = 0;
1188         int one = 1;
1189         int addr_len;
1190
1191         if (dlm_local_addr[0]->ss_family == AF_INET)
1192                 addr_len = sizeof(struct sockaddr_in);
1193         else
1194                 addr_len = sizeof(struct sockaddr_in6);
1195
1196         /* Create a socket to communicate with */
1197         result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
1198                                   IPPROTO_TCP, &sock);
1199         if (result < 0) {
1200                 log_print("Can't create listening comms socket");
1201                 goto create_out;
1202         }
1203
1204         /* Turn off Nagle's algorithm */
1205         kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
1206                           sizeof(one));
1207
1208         result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
1209                                    (char *)&one, sizeof(one));
1210
1211         if (result < 0) {
1212                 log_print("Failed to set SO_REUSEADDR on socket: %d", result);
1213         }
1214         con->rx_action = tcp_accept_from_sock;
1215         con->connect_action = tcp_connect_to_sock;
1216
1217         /* Bind to our port */
1218         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1219         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1220         if (result < 0) {
1221                 log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1222                 sock_release(sock);
1223                 sock = NULL;
1224                 con->sock = NULL;
1225                 goto create_out;
1226         }
1227         result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
1228                                  (char *)&one, sizeof(one));
1229         if (result < 0) {
1230                 log_print("Set keepalive failed: %d", result);
1231         }
1232
1233         result = sock->ops->listen(sock, 5);
1234         if (result < 0) {
1235                 log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1236                 sock_release(sock);
1237                 sock = NULL;
1238                 goto create_out;
1239         }
1240
1241 create_out:
1242         return sock;
1243 }
1244
1245 /* Get local addresses */
1246 static void init_local(void)
1247 {
1248         struct sockaddr_storage sas, *addr;
1249         int i;
1250
1251         dlm_local_count = 0;
1252         for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1253                 if (dlm_our_addr(&sas, i))
1254                         break;
1255
1256                 addr = kmalloc(sizeof(*addr), GFP_NOFS);
1257                 if (!addr)
1258                         break;
1259                 memcpy(addr, &sas, sizeof(*addr));
1260                 dlm_local_addr[dlm_local_count++] = addr;
1261         }
1262 }
1263
1264 /* Bind to an IP address. SCTP allows multiple address so it can do
1265    multi-homing */
1266 static int add_sctp_bind_addr(struct connection *sctp_con,
1267                               struct sockaddr_storage *addr,
1268                               int addr_len, int num)
1269 {
1270         int result = 0;
1271
1272         if (num == 1)
1273                 result = kernel_bind(sctp_con->sock,
1274                                      (struct sockaddr *) addr,
1275                                      addr_len);
1276         else
1277                 result = kernel_setsockopt(sctp_con->sock, SOL_SCTP,
1278                                            SCTP_SOCKOPT_BINDX_ADD,
1279                                            (char *)addr, addr_len);
1280
1281         if (result < 0)
1282                 log_print("Can't bind to port %d addr number %d",
1283                           dlm_config.ci_tcp_port, num);
1284
1285         return result;
1286 }
1287
1288 /* Initialise SCTP socket and bind to all interfaces */
1289 static int sctp_listen_for_all(void)
1290 {
1291         struct socket *sock = NULL;
1292         struct sockaddr_storage localaddr;
1293         struct sctp_event_subscribe subscribe;
1294         int result = -EINVAL, num = 1, i, addr_len;
1295         struct connection *con = nodeid2con(0, GFP_NOFS);
1296         int bufsize = NEEDED_RMEM;
1297
1298         if (!con)
1299                 return -ENOMEM;
1300
1301         log_print("Using SCTP for communications");
1302
1303         result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_SEQPACKET,
1304                                   IPPROTO_SCTP, &sock);
1305         if (result < 0) {
1306                 log_print("Can't create comms socket, check SCTP is loaded");
1307                 goto out;
1308         }
1309
1310         /* Listen for events */
1311         memset(&subscribe, 0, sizeof(subscribe));
1312         subscribe.sctp_data_io_event = 1;
1313         subscribe.sctp_association_event = 1;
1314         subscribe.sctp_send_failure_event = 1;
1315         subscribe.sctp_shutdown_event = 1;
1316         subscribe.sctp_partial_delivery_event = 1;
1317
1318         result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE,
1319                                  (char *)&bufsize, sizeof(bufsize));
1320         if (result)
1321                 log_print("Error increasing buffer space on socket %d", result);
1322
1323         result = kernel_setsockopt(sock, SOL_SCTP, SCTP_EVENTS,
1324                                    (char *)&subscribe, sizeof(subscribe));
1325         if (result < 0) {
1326                 log_print("Failed to set SCTP_EVENTS on socket: result=%d",
1327                           result);
1328                 goto create_delsock;
1329         }
1330
1331         /* Init con struct */
1332         sock->sk->sk_user_data = con;
1333         con->sock = sock;
1334         con->sock->sk->sk_data_ready = lowcomms_data_ready;
1335         con->rx_action = receive_from_sock;
1336         con->connect_action = sctp_init_assoc;
1337
1338         /* Bind to all interfaces. */
1339         for (i = 0; i < dlm_local_count; i++) {
1340                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
1341                 make_sockaddr(&localaddr, dlm_config.ci_tcp_port, &addr_len);
1342
1343                 result = add_sctp_bind_addr(con, &localaddr, addr_len, num);
1344                 if (result)
1345                         goto create_delsock;
1346                 ++num;
1347         }
1348
1349         result = sock->ops->listen(sock, 5);
1350         if (result < 0) {
1351                 log_print("Can't set socket listening");
1352                 goto create_delsock;
1353         }
1354
1355         return 0;
1356
1357 create_delsock:
1358         sock_release(sock);
1359         con->sock = NULL;
1360 out:
1361         return result;
1362 }
1363
1364 static int tcp_listen_for_all(void)
1365 {
1366         struct socket *sock = NULL;
1367         struct connection *con = nodeid2con(0, GFP_NOFS);
1368         int result = -EINVAL;
1369
1370         if (!con)
1371                 return -ENOMEM;
1372
1373         /* We don't support multi-homed hosts */
1374         if (dlm_local_addr[1] != NULL) {
1375                 log_print("TCP protocol can't handle multi-homed hosts, "
1376                           "try SCTP");
1377                 return -EINVAL;
1378         }
1379
1380         log_print("Using TCP for communications");
1381
1382         sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
1383         if (sock) {
1384                 add_sock(sock, con);
1385                 result = 0;
1386         }
1387         else {
1388                 result = -EADDRINUSE;
1389         }
1390
1391         return result;
1392 }
1393
1394
1395
1396 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1397                                                      gfp_t allocation)
1398 {
1399         struct writequeue_entry *entry;
1400
1401         entry = kmalloc(sizeof(struct writequeue_entry), allocation);
1402         if (!entry)
1403                 return NULL;
1404
1405         entry->page = alloc_page(allocation);
1406         if (!entry->page) {
1407                 kfree(entry);
1408                 return NULL;
1409         }
1410
1411         entry->offset = 0;
1412         entry->len = 0;
1413         entry->end = 0;
1414         entry->users = 0;
1415         entry->con = con;
1416
1417         return entry;
1418 }
1419
1420 void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
1421 {
1422         struct connection *con;
1423         struct writequeue_entry *e;
1424         int offset = 0;
1425
1426         con = nodeid2con(nodeid, allocation);
1427         if (!con)
1428                 return NULL;
1429
1430         spin_lock(&con->writequeue_lock);
1431         e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
1432         if ((&e->list == &con->writequeue) ||
1433             (PAGE_CACHE_SIZE - e->end < len)) {
1434                 e = NULL;
1435         } else {
1436                 offset = e->end;
1437                 e->end += len;
1438                 e->users++;
1439         }
1440         spin_unlock(&con->writequeue_lock);
1441
1442         if (e) {
1443         got_one:
1444                 *ppc = page_address(e->page) + offset;
1445                 return e;
1446         }
1447
1448         e = new_writequeue_entry(con, allocation);
1449         if (e) {
1450                 spin_lock(&con->writequeue_lock);
1451                 offset = e->end;
1452                 e->end += len;
1453                 e->users++;
1454                 list_add_tail(&e->list, &con->writequeue);
1455                 spin_unlock(&con->writequeue_lock);
1456                 goto got_one;
1457         }
1458         return NULL;
1459 }
1460
1461 void dlm_lowcomms_commit_buffer(void *mh)
1462 {
1463         struct writequeue_entry *e = (struct writequeue_entry *)mh;
1464         struct connection *con = e->con;
1465         int users;
1466
1467         spin_lock(&con->writequeue_lock);
1468         users = --e->users;
1469         if (users)
1470                 goto out;
1471         e->len = e->end - e->offset;
1472         spin_unlock(&con->writequeue_lock);
1473
1474         if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) {
1475                 queue_work(send_workqueue, &con->swork);
1476         }
1477         return;
1478
1479 out:
1480         spin_unlock(&con->writequeue_lock);
1481         return;
1482 }
1483
1484 /* Send a message */
1485 static void send_to_sock(struct connection *con)
1486 {
1487         int ret = 0;
1488         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1489         struct writequeue_entry *e;
1490         int len, offset;
1491         int count = 0;
1492
1493         mutex_lock(&con->sock_mutex);
1494         if (con->sock == NULL)
1495                 goto out_connect;
1496
1497         spin_lock(&con->writequeue_lock);
1498         for (;;) {
1499                 e = list_entry(con->writequeue.next, struct writequeue_entry,
1500                                list);
1501                 if ((struct list_head *) e == &con->writequeue)
1502                         break;
1503
1504                 len = e->len;
1505                 offset = e->offset;
1506                 BUG_ON(len == 0 && e->users == 0);
1507                 spin_unlock(&con->writequeue_lock);
1508
1509                 ret = 0;
1510                 if (len) {
1511                         ret = kernel_sendpage(con->sock, e->page, offset, len,
1512                                               msg_flags);
1513                         if (ret == -EAGAIN || ret == 0) {
1514                                 if (ret == -EAGAIN &&
1515                                     test_bit(SOCK_ASYNC_NOSPACE, &con->sock->flags) &&
1516                                     !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1517                                         /* Notify TCP that we're limited by the
1518                                          * application window size.
1519                                          */
1520                                         set_bit(SOCK_NOSPACE, &con->sock->flags);
1521                                         con->sock->sk->sk_write_pending++;
1522                                 }
1523                                 cond_resched();
1524                                 goto out;
1525                         } else if (ret < 0)
1526                                 goto send_error;
1527                 }
1528
1529                 /* Don't starve people filling buffers */
1530                 if (++count >= MAX_SEND_MSG_COUNT) {
1531                         cond_resched();
1532                         count = 0;
1533                 }
1534
1535                 spin_lock(&con->writequeue_lock);
1536                 e->offset += ret;
1537                 e->len -= ret;
1538
1539                 if (e->len == 0 && e->users == 0) {
1540                         list_del(&e->list);
1541                         free_entry(e);
1542                 }
1543         }
1544         spin_unlock(&con->writequeue_lock);
1545 out:
1546         mutex_unlock(&con->sock_mutex);
1547         return;
1548
1549 send_error:
1550         mutex_unlock(&con->sock_mutex);
1551         close_connection(con, false);
1552         lowcomms_connect_sock(con);
1553         return;
1554
1555 out_connect:
1556         mutex_unlock(&con->sock_mutex);
1557         if (!test_bit(CF_INIT_PENDING, &con->flags))
1558                 lowcomms_connect_sock(con);
1559 }
1560
1561 static void clean_one_writequeue(struct connection *con)
1562 {
1563         struct writequeue_entry *e, *safe;
1564
1565         spin_lock(&con->writequeue_lock);
1566         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1567                 list_del(&e->list);
1568                 free_entry(e);
1569         }
1570         spin_unlock(&con->writequeue_lock);
1571 }
1572
1573 /* Called from recovery when it knows that a node has
1574    left the cluster */
1575 int dlm_lowcomms_close(int nodeid)
1576 {
1577         struct connection *con;
1578         struct dlm_node_addr *na;
1579
1580         log_print("closing connection to node %d", nodeid);
1581         con = nodeid2con(nodeid, 0);
1582         if (con) {
1583                 clear_bit(CF_CONNECT_PENDING, &con->flags);
1584                 clear_bit(CF_WRITE_PENDING, &con->flags);
1585                 set_bit(CF_CLOSE, &con->flags);
1586                 if (cancel_work_sync(&con->swork))
1587                         log_print("canceled swork for node %d", nodeid);
1588                 if (cancel_work_sync(&con->rwork))
1589                         log_print("canceled rwork for node %d", nodeid);
1590                 clean_one_writequeue(con);
1591                 close_connection(con, true);
1592         }
1593
1594         spin_lock(&dlm_node_addrs_spin);
1595         na = find_node_addr(nodeid);
1596         if (na) {
1597                 list_del(&na->list);
1598                 while (na->addr_count--)
1599                         kfree(na->addr[na->addr_count]);
1600                 kfree(na);
1601         }
1602         spin_unlock(&dlm_node_addrs_spin);
1603
1604         return 0;
1605 }
1606
1607 /* Receive workqueue function */
1608 static void process_recv_sockets(struct work_struct *work)
1609 {
1610         struct connection *con = container_of(work, struct connection, rwork);
1611         int err;
1612
1613         clear_bit(CF_READ_PENDING, &con->flags);
1614         do {
1615                 err = con->rx_action(con);
1616         } while (!err);
1617 }
1618
1619 /* Send workqueue function */
1620 static void process_send_sockets(struct work_struct *work)
1621 {
1622         struct connection *con = container_of(work, struct connection, swork);
1623
1624         if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) {
1625                 con->connect_action(con);
1626                 set_bit(CF_WRITE_PENDING, &con->flags);
1627         }
1628         if (test_and_clear_bit(CF_WRITE_PENDING, &con->flags))
1629                 send_to_sock(con);
1630 }
1631
1632
1633 /* Discard all entries on the write queues */
1634 static void clean_writequeues(void)
1635 {
1636         foreach_conn(clean_one_writequeue);
1637 }
1638
1639 static void work_stop(void)
1640 {
1641         destroy_workqueue(recv_workqueue);
1642         destroy_workqueue(send_workqueue);
1643 }
1644
1645 static int work_start(void)
1646 {
1647         recv_workqueue = alloc_workqueue("dlm_recv",
1648                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1649         if (!recv_workqueue) {
1650                 log_print("can't start dlm_recv");
1651                 return -ENOMEM;
1652         }
1653
1654         send_workqueue = alloc_workqueue("dlm_send",
1655                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1656         if (!send_workqueue) {
1657                 log_print("can't start dlm_send");
1658                 destroy_workqueue(recv_workqueue);
1659                 return -ENOMEM;
1660         }
1661
1662         return 0;
1663 }
1664
1665 static void stop_conn(struct connection *con)
1666 {
1667         con->flags |= 0x0F;
1668         if (con->sock && con->sock->sk)
1669                 con->sock->sk->sk_user_data = NULL;
1670 }
1671
1672 static void free_conn(struct connection *con)
1673 {
1674         close_connection(con, true);
1675         if (con->othercon)
1676                 kmem_cache_free(con_cache, con->othercon);
1677         hlist_del(&con->list);
1678         kmem_cache_free(con_cache, con);
1679 }
1680
1681 void dlm_lowcomms_stop(void)
1682 {
1683         /* Set all the flags to prevent any
1684            socket activity.
1685         */
1686         mutex_lock(&connections_lock);
1687         dlm_allow_conn = 0;
1688         foreach_conn(stop_conn);
1689         mutex_unlock(&connections_lock);
1690
1691         work_stop();
1692
1693         mutex_lock(&connections_lock);
1694         clean_writequeues();
1695
1696         foreach_conn(free_conn);
1697
1698         mutex_unlock(&connections_lock);
1699         kmem_cache_destroy(con_cache);
1700 }
1701
1702 int dlm_lowcomms_start(void)
1703 {
1704         int error = -EINVAL;
1705         struct connection *con;
1706         int i;
1707
1708         for (i = 0; i < CONN_HASH_SIZE; i++)
1709                 INIT_HLIST_HEAD(&connection_hash[i]);
1710
1711         init_local();
1712         if (!dlm_local_count) {
1713                 error = -ENOTCONN;
1714                 log_print("no local IP address has been set");
1715                 goto fail;
1716         }
1717
1718         error = -ENOMEM;
1719         con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
1720                                       __alignof__(struct connection), 0,
1721                                       NULL);
1722         if (!con_cache)
1723                 goto fail;
1724
1725         error = work_start();
1726         if (error)
1727                 goto fail_destroy;
1728
1729         dlm_allow_conn = 1;
1730
1731         /* Start listening */
1732         if (dlm_config.ci_protocol == 0)
1733                 error = tcp_listen_for_all();
1734         else
1735                 error = sctp_listen_for_all();
1736         if (error)
1737                 goto fail_unlisten;
1738
1739         return 0;
1740
1741 fail_unlisten:
1742         dlm_allow_conn = 0;
1743         con = nodeid2con(0,0);
1744         if (con) {
1745                 close_connection(con, false);
1746                 kmem_cache_free(con_cache, con);
1747         }
1748 fail_destroy:
1749         kmem_cache_destroy(con_cache);
1750 fail:
1751         return error;
1752 }
1753
1754 void dlm_lowcomms_exit(void)
1755 {
1756         struct dlm_node_addr *na, *safe;
1757
1758         spin_lock(&dlm_node_addrs_spin);
1759         list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
1760                 list_del(&na->list);
1761                 while (na->addr_count--)
1762                         kfree(na->addr[na->addr_count]);
1763                 kfree(na);
1764         }
1765         spin_unlock(&dlm_node_addrs_spin);
1766 }