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