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