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 <net/sctp/user.h>
  55 #include <net/ipv6.h>
  56
  57 #include "dlm_internal.h"
  58 #include "lowcomms.h"
  59 #include "midcomms.h"
  60 #include "config.h"
  61
  62 #define NEEDED_RMEM (4*1024*1024)
  63 #define CONN_HASH_SIZE 32
  64
  65 struct cbuf {
  66         unsigned int base;
  67         unsigned int len;
  68         unsigned int mask;
  69 };
  70
  71 static void cbuf_add(struct cbuf *cb, int n)
  72 {
  73         cb->len += n;
  74 }
  75
  76 static int cbuf_data(struct cbuf *cb)
  77 {
  78         return ((cb->base + cb->len) & cb->mask);
  79 }
  80
  81 static void cbuf_init(struct cbuf *cb, int size)
  82 {
  83         cb->base = cb->len = 0;
  84         cb->mask = size-1;
  85 }
  86
  87 static void cbuf_eat(struct cbuf *cb, int n)
  88 {
  89         cb->len  -= n;
  90         cb->base += n;
  91         cb->base &= cb->mask;
  92 }
  93
  94 static bool cbuf_empty(struct cbuf *cb)
  95 {
  96         return cb->len == 0;
  97 }
  98
  99 struct connection {
 100         struct socket *sock;    /* NULL if not connected */
 101         uint32_t nodeid;        /* So we know who we are in the list */
 102         struct mutex sock_mutex;
 103         unsigned long flags;
 104 #define CF_READ_PENDING 1
 105 #define CF_WRITE_PENDING 2
 106 #define CF_CONNECT_PENDING 3
 107 #define CF_INIT_PENDING 4
 108 #define CF_IS_OTHERCON 5
 109         struct list_head writequeue;  /* List of outgoing writequeue_entries */
 110         spinlock_t writequeue_lock;
 111         int (*rx_action) (struct connection *); /* What to do when active */
 112         void (*connect_action) (struct connection *);   /* What to do to connect */
 113         struct page *rx_page;
 114         struct cbuf cb;
 115         int retries;
 116 #define MAX_CONNECT_RETRIES 3
 117         int sctp_assoc;
 118         struct hlist_node list;
 119         struct connection *othercon;
 120         struct work_struct rwork; /* Receive workqueue */
 121         struct work_struct swork; /* Send workqueue */
 122 };
 123 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 124
 125 /* An entry waiting to be sent */
 126 struct writequeue_entry {
 127         struct list_head list;
 128         struct page *page;
 129         int offset;
 130         int len;
 131         int end;
 132         int users;
 133         struct connection *con;
 134 };
 135
 136 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 137 static int dlm_local_count;
 138
 139 /* Work queues */
 140 static struct workqueue_struct *recv_workqueue;
 141 static struct workqueue_struct *send_workqueue;
 142
 143 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 144 static DEFINE_MUTEX(connections_lock);
 145 static struct kmem_cache *con_cache;
 146
 147 static void process_recv_sockets(struct work_struct *work);
 148 static void process_send_sockets(struct work_struct *work);
 149
 150
 151 /* This is deliberately very simple because most clusters have simple
 152    sequential nodeids, so we should be able to go straight to a connection
 153    struct in the array */
 154 static inline int nodeid_hash(int nodeid)
 155 {
 156         return nodeid & (CONN_HASH_SIZE-1);
 157 }
 158
 159 static struct connection *__find_con(int nodeid)
 160 {
 161         int r;
 162         struct hlist_node *h;
 163         struct connection *con;
 164
 165         r = nodeid_hash(nodeid);
 166
 167         hlist_for_each_entry(con, h, &connection_hash[r], list) {
 168                 if (con->nodeid == nodeid)
 169                         return con;
 170         }
 171         return NULL;
 172 }
 173
 174 /*
 175  * If 'allocation' is zero then we don't attempt to create a new
 176  * connection structure for this node.
 177  */
 178 static struct connection *__nodeid2con(int nodeid, gfp_t alloc)
 179 {
 180         struct connection *con = NULL;
 181         int r;
 182
 183         con = __find_con(nodeid);
 184         if (con || !alloc)
 185                 return con;
 186
 187         con = kmem_cache_zalloc(con_cache, alloc);
 188         if (!con)
 189                 return NULL;
 190
 191         r = nodeid_hash(nodeid);
 192         hlist_add_head(&con->list, &connection_hash[r]);
 193
 194         con->nodeid = nodeid;
 195         mutex_init(&con->sock_mutex);
 196         INIT_LIST_HEAD(&con->writequeue);
 197         spin_lock_init(&con->writequeue_lock);
 198         INIT_WORK(&con->swork, process_send_sockets);
 199         INIT_WORK(&con->rwork, process_recv_sockets);
 200
 201         /* Setup action pointers for child sockets */
 202         if (con->nodeid) {
 203                 struct connection *zerocon = __find_con(0);
 204
 205                 con->connect_action = zerocon->connect_action;
 206                 if (!con->rx_action)
 207                         con->rx_action = zerocon->rx_action;
 208         }
 209
 210         return con;
 211 }
 212
 213 /* Loop round all connections */
 214 static void foreach_conn(void (*conn_func)(struct connection *c))
 215 {
 216         int i;
 217         struct hlist_node *h, *n;
 218         struct connection *con;
 219
 220         for (i = 0; i < CONN_HASH_SIZE; i++) {
 221                 hlist_for_each_entry_safe(con, h, n, &connection_hash[i], list){
 222                         conn_func(con);
 223                 }
 224         }
 225 }
 226
 227 static struct connection *nodeid2con(int nodeid, gfp_t allocation)
 228 {
 229         struct connection *con;
 230
 231         mutex_lock(&connections_lock);
 232         con = __nodeid2con(nodeid, allocation);
 233         mutex_unlock(&connections_lock);
 234
 235         return con;
 236 }
 237
 238 /* This is a bit drastic, but only called when things go wrong */
 239 static struct connection *assoc2con(int assoc_id)
 240 {
 241         int i;
 242         struct hlist_node *h;
 243         struct connection *con;
 244
 245         mutex_lock(&connections_lock);
 246
 247         for (i = 0 ; i < CONN_HASH_SIZE; i++) {
 248                 hlist_for_each_entry(con, h, &connection_hash[i], list) {
 249                         if (con && con->sctp_assoc == assoc_id) {
 250                                 mutex_unlock(&connections_lock);
 251                                 return con;
 252                         }
 253                 }
 254         }
 255         mutex_unlock(&connections_lock);
 256         return NULL;
 257 }
 258
 259 static int nodeid_to_addr(int nodeid, struct sockaddr *retaddr)
 260 {
 261         struct sockaddr_storage addr;
 262         int error;
 263
 264         if (!dlm_local_count)
 265                 return -1;
 266
 267         error = dlm_nodeid_to_addr(nodeid, &addr);
 268         if (error)
 269                 return error;
 270
 271         if (dlm_local_addr[0]->ss_family == AF_INET) {
 272                 struct sockaddr_in *in4  = (struct sockaddr_in *) &addr;
 273                 struct sockaddr_in *ret4 = (struct sockaddr_in *) retaddr;
 274                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 275         } else {
 276                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &addr;
 277                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) retaddr;
 278                 ipv6_addr_copy(&ret6->sin6_addr, &in6->sin6_addr);
 279         }
 280
 281         return 0;
 282 }
 283
 284 /* Data available on socket or listen socket received a connect */
 285 static void lowcomms_data_ready(struct sock *sk, int count_unused)
 286 {
 287         struct connection *con = sock2con(sk);
 288         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 289                 queue_work(recv_workqueue, &con->rwork);
 290 }
 291
 292 static void lowcomms_write_space(struct sock *sk)
 293 {
 294         struct connection *con = sock2con(sk);
 295
 296         if (con && !test_and_set_bit(CF_WRITE_PENDING, &con->flags))
 297                 queue_work(send_workqueue, &con->swork);
 298 }
 299
 300 static inline void lowcomms_connect_sock(struct connection *con)
 301 {
 302         if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags))
 303                 queue_work(send_workqueue, &con->swork);
 304 }
 305
 306 static void lowcomms_state_change(struct sock *sk)
 307 {
 308         if (sk->sk_state == TCP_ESTABLISHED)
 309                 lowcomms_write_space(sk);
 310 }
 311
 312 /* Make a socket active */
 313 static int add_sock(struct socket *sock, struct connection *con)
 314 {
 315         con->sock = sock;
 316
 317         /* Install a data_ready callback */
 318         con->sock->sk->sk_data_ready = lowcomms_data_ready;
 319         con->sock->sk->sk_write_space = lowcomms_write_space;
 320         con->sock->sk->sk_state_change = lowcomms_state_change;
 321         con->sock->sk->sk_user_data = con;
 322         con->sock->sk->sk_allocation = GFP_NOFS;
 323         return 0;
 324 }
 325
 326 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 327    length */
 328 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 329                           int *addr_len)
 330 {
 331         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 332         if (saddr->ss_family == AF_INET) {
 333                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 334                 in4_addr->sin_port = cpu_to_be16(port);
 335                 *addr_len = sizeof(struct sockaddr_in);
 336                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 337         } else {
 338                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 339                 in6_addr->sin6_port = cpu_to_be16(port);
 340                 *addr_len = sizeof(struct sockaddr_in6);
 341         }
 342         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 343 }
 344
 345 /* Close a remote connection and tidy up */
 346 static void close_connection(struct connection *con, bool and_other)
 347 {
 348         mutex_lock(&con->sock_mutex);
 349
 350         if (con->sock) {
 351                 sock_release(con->sock);
 352                 con->sock = NULL;
 353         }
 354         if (con->othercon && and_other) {
 355                 /* Will only re-enter once. */
 356                 close_connection(con->othercon, false);
 357         }
 358         if (con->rx_page) {
 359                 __free_page(con->rx_page);
 360                 con->rx_page = NULL;
 361         }
 362
 363         con->retries = 0;
 364         mutex_unlock(&con->sock_mutex);
 365 }
 366
 367 /* We only send shutdown messages to nodes that are not part of the cluster */
 368 static void sctp_send_shutdown(sctp_assoc_t associd)
 369 {
 370         static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
 371         struct msghdr outmessage;
 372         struct cmsghdr *cmsg;
 373         struct sctp_sndrcvinfo *sinfo;
 374         int ret;
 375         struct connection *con;
 376
 377         con = nodeid2con(0,0);
 378         BUG_ON(con == NULL);
 379
 380         outmessage.msg_name = NULL;
 381         outmessage.msg_namelen = 0;
 382         outmessage.msg_control = outcmsg;
 383         outmessage.msg_controllen = sizeof(outcmsg);
 384         outmessage.msg_flags = MSG_EOR;
 385
 386         cmsg = CMSG_FIRSTHDR(&outmessage);
 387         cmsg->cmsg_level = IPPROTO_SCTP;
 388         cmsg->cmsg_type = SCTP_SNDRCV;
 389         cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
 390         outmessage.msg_controllen = cmsg->cmsg_len;
 391         sinfo = CMSG_DATA(cmsg);
 392         memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
 393
 394         sinfo->sinfo_flags |= MSG_EOF;
 395         sinfo->sinfo_assoc_id = associd;
 396
 397         ret = kernel_sendmsg(con->sock, &outmessage, NULL, 0, 0);
 398
 399         if (ret != 0)
 400                 log_print("send EOF to node failed: %d", ret);
 401 }
 402
 403 static void sctp_init_failed_foreach(struct connection *con)
 404 {
 405         con->sctp_assoc = 0;
 406         if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) {
 407                 if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags))
 408                         queue_work(send_workqueue, &con->swork);
 409         }
 410 }
 411
 412 /* INIT failed but we don't know which node...
 413    restart INIT on all pending nodes */
 414 static void sctp_init_failed(void)
 415 {
 416         mutex_lock(&connections_lock);
 417
 418         foreach_conn(sctp_init_failed_foreach);
 419
 420         mutex_unlock(&connections_lock);
 421 }
 422
 423 /* Something happened to an association */
 424 static void process_sctp_notification(struct connection *con,
 425                                       struct msghdr *msg, char *buf)
 426 {
 427         union sctp_notification *sn = (union sctp_notification *)buf;
 428
 429         if (sn->sn_header.sn_type == SCTP_ASSOC_CHANGE) {
 430                 switch (sn->sn_assoc_change.sac_state) {
 431
 432                 case SCTP_COMM_UP:
 433                 case SCTP_RESTART:
 434                 {
 435                         /* Check that the new node is in the lockspace */
 436                         struct sctp_prim prim;
 437                         int nodeid;
 438                         int prim_len, ret;
 439                         int addr_len;
 440                         struct connection *new_con;
 441                         struct file *file;
 442                         sctp_peeloff_arg_t parg;
 443                         int parglen = sizeof(parg);
 444
 445                         /*
 446                          * We get this before any data for an association.
 447                          * We verify that the node is in the cluster and
 448                          * then peel off a socket for it.
 449                          */
 450                         if ((int)sn->sn_assoc_change.sac_assoc_id <= 0) {
 451                                 log_print("COMM_UP for invalid assoc ID %d",
 452                                          (int)sn->sn_assoc_change.sac_assoc_id);
 453                                 sctp_init_failed();
 454                                 return;
 455                         }
 456                         memset(&prim, 0, sizeof(struct sctp_prim));
 457                         prim_len = sizeof(struct sctp_prim);
 458                         prim.ssp_assoc_id = sn->sn_assoc_change.sac_assoc_id;
 459
 460                         ret = kernel_getsockopt(con->sock,
 461                                                 IPPROTO_SCTP,
 462                                                 SCTP_PRIMARY_ADDR,
 463                                                 (char*)&prim,
 464                                                 &prim_len);
 465                         if (ret < 0) {
 466                                 log_print("getsockopt/sctp_primary_addr on "
 467                                           "new assoc %d failed : %d",
 468                                           (int)sn->sn_assoc_change.sac_assoc_id,
 469                                           ret);
 470
 471                                 /* Retry INIT later */
 472                                 new_con = assoc2con(sn->sn_assoc_change.sac_assoc_id);
 473                                 if (new_con)
 474                                         clear_bit(CF_CONNECT_PENDING, &con->flags);
 475                                 return;
 476                         }
 477                         make_sockaddr(&prim.ssp_addr, 0, &addr_len);
 478                         if (dlm_addr_to_nodeid(&prim.ssp_addr, &nodeid)) {
 479                                 int i;
 480                                 unsigned char *b=(unsigned char *)&prim.ssp_addr;
 481                                 log_print("reject connect from unknown addr");
 482                                 for (i=0; i<sizeof(struct sockaddr_storage);i++)
 483                                         printk("%02x ", b[i]);
 484                                 printk("\n");
 485                                 sctp_send_shutdown(prim.ssp_assoc_id);
 486                                 return;
 487                         }
 488
 489                         new_con = nodeid2con(nodeid, GFP_KERNEL);
 490                         if (!new_con)
 491                                 return;
 492
 493                         /* Peel off a new sock */
 494                         parg.associd = sn->sn_assoc_change.sac_assoc_id;
 495                         ret = kernel_getsockopt(con->sock, IPPROTO_SCTP,
 496                                                 SCTP_SOCKOPT_PEELOFF,
 497                                                 (void *)&parg, &parglen);
 498                         if (ret) {
 499                                 log_print("Can't peel off a socket for "
 500                                           "connection %d to node %d: err=%d\n",
 501                                           parg.associd, nodeid, ret);
 502                         }
 503                         file = fget(parg.sd);
 504                         new_con->sock = SOCKET_I(file->f_dentry->d_inode);
 505                         add_sock(new_con->sock, new_con);
 506                         fput(file);
 507                         put_unused_fd(parg.sd);
 508
 509                         log_print("got new/restarted association %d nodeid %d",
 510                                  (int)sn->sn_assoc_change.sac_assoc_id, nodeid);
 511
 512                         /* Send any pending writes */
 513                         clear_bit(CF_CONNECT_PENDING, &new_con->flags);
 514                         clear_bit(CF_INIT_PENDING, &con->flags);
 515                         if (!test_and_set_bit(CF_WRITE_PENDING, &new_con->flags)) {
 516                                 queue_work(send_workqueue, &new_con->swork);
 517                         }
 518                         if (!test_and_set_bit(CF_READ_PENDING, &new_con->flags))
 519                                 queue_work(recv_workqueue, &new_con->rwork);
 520                 }
 521                 break;
 522
 523                 case SCTP_COMM_LOST:
 524                 case SCTP_SHUTDOWN_COMP:
 525                 {
 526                         con = assoc2con(sn->sn_assoc_change.sac_assoc_id);
 527                         if (con) {
 528                                 con->sctp_assoc = 0;
 529                         }
 530                 }
 531                 break;
 532
 533                 /* We don't know which INIT failed, so clear the PENDING flags
 534                  * on them all.  if assoc_id is zero then it will then try
 535                  * again */
 536
 537                 case SCTP_CANT_STR_ASSOC:
 538                 {
 539                         log_print("Can't start SCTP association - retrying");
 540                         sctp_init_failed();
 541                 }
 542                 break;
 543
 544                 default:
 545                         log_print("unexpected SCTP assoc change id=%d state=%d",
 546                                   (int)sn->sn_assoc_change.sac_assoc_id,
 547                                   sn->sn_assoc_change.sac_state);
 548                 }
 549         }
 550 }
 551
 552 /* Data received from remote end */
 553 static int receive_from_sock(struct connection *con)
 554 {
 555         int ret = 0;
 556         struct msghdr msg = {};
 557         struct kvec iov[2];
 558         unsigned len;
 559         int r;
 560         int call_again_soon = 0;
 561         int nvec;
 562         char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
 563
 564         mutex_lock(&con->sock_mutex);
 565
 566         if (con->sock == NULL) {
 567                 ret = -EAGAIN;
 568                 goto out_close;
 569         }
 570
 571         if (con->rx_page == NULL) {
 572                 /*
 573                  * This doesn't need to be atomic, but I think it should
 574                  * improve performance if it is.
 575                  */
 576                 con->rx_page = alloc_page(GFP_ATOMIC);
 577                 if (con->rx_page == NULL)
 578                         goto out_resched;
 579                 cbuf_init(&con->cb, PAGE_CACHE_SIZE);
 580         }
 581
 582         /* Only SCTP needs these really */
 583         memset(&incmsg, 0, sizeof(incmsg));
 584         msg.msg_control = incmsg;
 585         msg.msg_controllen = sizeof(incmsg);
 586
 587         /*
 588          * iov[0] is the bit of the circular buffer between the current end
 589          * point (cb.base + cb.len) and the end of the buffer.
 590          */
 591         iov[0].iov_len = con->cb.base - cbuf_data(&con->cb);
 592         iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb);
 593         iov[1].iov_len = 0;
 594         nvec = 1;
 595
 596         /*
 597          * iov[1] is the bit of the circular buffer between the start of the
 598          * buffer and the start of the currently used section (cb.base)
 599          */
 600         if (cbuf_data(&con->cb) >= con->cb.base) {
 601                 iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb);
 602                 iov[1].iov_len = con->cb.base;
 603                 iov[1].iov_base = page_address(con->rx_page);
 604                 nvec = 2;
 605         }
 606         len = iov[0].iov_len + iov[1].iov_len;
 607
 608         r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len,
 609                                MSG_DONTWAIT | MSG_NOSIGNAL);
 610         if (ret <= 0)
 611                 goto out_close;
 612
 613         /* Process SCTP notifications */
 614         if (msg.msg_flags & MSG_NOTIFICATION) {
 615                 msg.msg_control = incmsg;
 616                 msg.msg_controllen = sizeof(incmsg);
 617
 618                 process_sctp_notification(con, &msg,
 619                                 page_address(con->rx_page) + con->cb.base);
 620                 mutex_unlock(&con->sock_mutex);
 621                 return 0;
 622         }
 623         BUG_ON(con->nodeid == 0);
 624
 625         if (ret == len)
 626                 call_again_soon = 1;
 627         cbuf_add(&con->cb, ret);
 628         ret = dlm_process_incoming_buffer(con->nodeid,
 629                                           page_address(con->rx_page),
 630                                           con->cb.base, con->cb.len,
 631                                           PAGE_CACHE_SIZE);
 632         if (ret == -EBADMSG) {
 633                 log_print("lowcomms: addr=%p, base=%u, len=%u, "
 634                           "iov_len=%u, iov_base[0]=%p, read=%d",
 635                           page_address(con->rx_page), con->cb.base, con->cb.len,
 636                           len, iov[0].iov_base, r);
 637         }
 638         if (ret < 0)
 639                 goto out_close;
 640         cbuf_eat(&con->cb, ret);
 641
 642         if (cbuf_empty(&con->cb) && !call_again_soon) {
 643                 __free_page(con->rx_page);
 644                 con->rx_page = NULL;
 645         }
 646
 647         if (call_again_soon)
 648                 goto out_resched;
 649         mutex_unlock(&con->sock_mutex);
 650         return 0;
 651
 652 out_resched:
 653         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 654                 queue_work(recv_workqueue, &con->rwork);
 655         mutex_unlock(&con->sock_mutex);
 656         return -EAGAIN;
 657
 658 out_close:
 659         mutex_unlock(&con->sock_mutex);
 660         if (ret != -EAGAIN) {
 661                 close_connection(con, false);
 662                 /* Reconnect when there is something to send */
 663         }
 664         /* Don't return success if we really got EOF */
 665         if (ret == 0)
 666                 ret = -EAGAIN;
 667
 668         return ret;
 669 }
 670
 671 /* Listening socket is busy, accept a connection */
 672 static int tcp_accept_from_sock(struct connection *con)
 673 {
 674         int result;
 675         struct sockaddr_storage peeraddr;
 676         struct socket *newsock;
 677         int len;
 678         int nodeid;
 679         struct connection *newcon;
 680         struct connection *addcon;
 681
 682         memset(&peeraddr, 0, sizeof(peeraddr));
 683         result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
 684                                   IPPROTO_TCP, &newsock);
 685         if (result < 0)
 686                 return -ENOMEM;
 687
 688         mutex_lock_nested(&con->sock_mutex, 0);
 689
 690         result = -ENOTCONN;
 691         if (con->sock == NULL)
 692                 goto accept_err;
 693
 694         newsock->type = con->sock->type;
 695         newsock->ops = con->sock->ops;
 696
 697         result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK);
 698         if (result < 0)
 699                 goto accept_err;
 700
 701         /* Get the connected socket's peer */
 702         memset(&peeraddr, 0, sizeof(peeraddr));
 703         if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr,
 704                                   &len, 2)) {
 705                 result = -ECONNABORTED;
 706                 goto accept_err;
 707         }
 708
 709         /* Get the new node's NODEID */
 710         make_sockaddr(&peeraddr, 0, &len);
 711         if (dlm_addr_to_nodeid(&peeraddr, &nodeid)) {
 712                 log_print("connect from non cluster node");
 713                 sock_release(newsock);
 714                 mutex_unlock(&con->sock_mutex);
 715                 return -1;
 716         }
 717
 718         log_print("got connection from %d", nodeid);
 719
 720         /*  Check to see if we already have a connection to this node. This
 721          *  could happen if the two nodes initiate a connection at roughly
 722          *  the same time and the connections cross on the wire.
 723          *  In this case we store the incoming one in "othercon"
 724          */
 725         newcon = nodeid2con(nodeid, GFP_KERNEL);
 726         if (!newcon) {
 727                 result = -ENOMEM;
 728                 goto accept_err;
 729         }
 730         mutex_lock_nested(&newcon->sock_mutex, 1);
 731         if (newcon->sock) {
 732                 struct connection *othercon = newcon->othercon;
 733
 734                 if (!othercon) {
 735                         othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL);
 736                         if (!othercon) {
 737                                 log_print("failed to allocate incoming socket");
 738                                 mutex_unlock(&newcon->sock_mutex);
 739                                 result = -ENOMEM;
 740                                 goto accept_err;
 741                         }
 742                         othercon->nodeid = nodeid;
 743                         othercon->rx_action = receive_from_sock;
 744                         mutex_init(&othercon->sock_mutex);
 745                         INIT_WORK(&othercon->swork, process_send_sockets);
 746                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 747                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 748                 }
 749                 if (!othercon->sock) {
 750                         newcon->othercon = othercon;
 751                         othercon->sock = newsock;
 752                         newsock->sk->sk_user_data = othercon;
 753                         add_sock(newsock, othercon);
 754                         addcon = othercon;
 755                 }
 756                 else {
 757                         printk("Extra connection from node %d attempted\n", nodeid);
 758                         result = -EAGAIN;
 759                         mutex_unlock(&newcon->sock_mutex);
 760                         goto accept_err;
 761                 }
 762         }
 763         else {
 764                 newsock->sk->sk_user_data = newcon;
 765                 newcon->rx_action = receive_from_sock;
 766                 add_sock(newsock, newcon);
 767                 addcon = newcon;
 768         }
 769
 770         mutex_unlock(&newcon->sock_mutex);
 771
 772         /*
 773          * Add it to the active queue in case we got data
 774          * beween processing the accept adding the socket
 775          * to the read_sockets list
 776          */
 777         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 778                 queue_work(recv_workqueue, &addcon->rwork);
 779         mutex_unlock(&con->sock_mutex);
 780
 781         return 0;
 782
 783 accept_err:
 784         mutex_unlock(&con->sock_mutex);
 785         sock_release(newsock);
 786
 787         if (result != -EAGAIN)
 788                 log_print("error accepting connection from node: %d", result);
 789         return result;
 790 }
 791
 792 static void free_entry(struct writequeue_entry *e)
 793 {
 794         __free_page(e->page);
 795         kfree(e);
 796 }
 797
 798 /* Initiate an SCTP association.
 799    This is a special case of send_to_sock() in that we don't yet have a
 800    peeled-off socket for this association, so we use the listening socket
 801    and add the primary IP address of the remote node.
 802  */
 803 static void sctp_init_assoc(struct connection *con)
 804 {
 805         struct sockaddr_storage rem_addr;
 806         char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))];
 807         struct msghdr outmessage;
 808         struct cmsghdr *cmsg;
 809         struct sctp_sndrcvinfo *sinfo;
 810         struct connection *base_con;
 811         struct writequeue_entry *e;
 812         int len, offset;
 813         int ret;
 814         int addrlen;
 815         struct kvec iov[1];
 816
 817         if (test_and_set_bit(CF_INIT_PENDING, &con->flags))
 818                 return;
 819
 820         if (con->retries++ > MAX_CONNECT_RETRIES)
 821                 return;
 822
 823         log_print("Initiating association with node %d", con->nodeid);
 824
 825         if (nodeid_to_addr(con->nodeid, (struct sockaddr *)&rem_addr)) {
 826                 log_print("no address for nodeid %d", con->nodeid);
 827                 return;
 828         }
 829         base_con = nodeid2con(0, 0);
 830         BUG_ON(base_con == NULL);
 831
 832         make_sockaddr(&rem_addr, dlm_config.ci_tcp_port, &addrlen);
 833
 834         outmessage.msg_name = &rem_addr;
 835         outmessage.msg_namelen = addrlen;
 836         outmessage.msg_control = outcmsg;
 837         outmessage.msg_controllen = sizeof(outcmsg);
 838         outmessage.msg_flags = MSG_EOR;
 839
 840         spin_lock(&con->writequeue_lock);
 841         e = list_entry(con->writequeue.next, struct writequeue_entry,
 842                        list);
 843
 844         BUG_ON((struct list_head *) e == &con->writequeue);
 845
 846         len = e->len;
 847         offset = e->offset;
 848         spin_unlock(&con->writequeue_lock);
 849
 850         /* Send the first block off the write queue */
 851         iov[0].iov_base = page_address(e->page)+offset;
 852         iov[0].iov_len = len;
 853
 854         cmsg = CMSG_FIRSTHDR(&outmessage);
 855         cmsg->cmsg_level = IPPROTO_SCTP;
 856         cmsg->cmsg_type = SCTP_SNDRCV;
 857         cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
 858         sinfo = CMSG_DATA(cmsg);
 859         memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo));
 860         sinfo->sinfo_ppid = cpu_to_le32(dlm_our_nodeid());
 861         outmessage.msg_controllen = cmsg->cmsg_len;
 862
 863         ret = kernel_sendmsg(base_con->sock, &outmessage, iov, 1, len);
 864         if (ret < 0) {
 865                 log_print("Send first packet to node %d failed: %d",
 866                           con->nodeid, ret);
 867
 868                 /* Try again later */
 869                 clear_bit(CF_CONNECT_PENDING, &con->flags);
 870                 clear_bit(CF_INIT_PENDING, &con->flags);
 871         }
 872         else {
 873                 spin_lock(&con->writequeue_lock);
 874                 e->offset += ret;
 875                 e->len -= ret;
 876
 877                 if (e->len == 0 && e->users == 0) {
 878                         list_del(&e->list);
 879                         free_entry(e);
 880                 }
 881                 spin_unlock(&con->writequeue_lock);
 882         }
 883 }
 884
 885 /* Connect a new socket to its peer */
 886 static void tcp_connect_to_sock(struct connection *con)
 887 {
 888         int result = -EHOSTUNREACH;
 889         struct sockaddr_storage saddr, src_addr;
 890         int addr_len;
 891         struct socket *sock;
 892
 893         if (con->nodeid == 0) {
 894                 log_print("attempt to connect sock 0 foiled");
 895                 return;
 896         }
 897
 898         mutex_lock(&con->sock_mutex);
 899         if (con->retries++ > MAX_CONNECT_RETRIES)
 900                 goto out;
 901
 902         /* Some odd races can cause double-connects, ignore them */
 903         if (con->sock) {
 904                 result = 0;
 905                 goto out;
 906         }
 907
 908         /* Create a socket to communicate with */
 909         result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
 910                                   IPPROTO_TCP, &sock);
 911         if (result < 0)
 912                 goto out_err;
 913
 914         memset(&saddr, 0, sizeof(saddr));
 915         if (dlm_nodeid_to_addr(con->nodeid, &saddr)) {
 916                 sock_release(sock);
 917                 goto out_err;
 918         }
 919
 920         sock->sk->sk_user_data = con;
 921         con->rx_action = receive_from_sock;
 922         con->connect_action = tcp_connect_to_sock;
 923         add_sock(sock, con);
 924
 925         /* Bind to our cluster-known address connecting to avoid
 926            routing problems */
 927         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
 928         make_sockaddr(&src_addr, 0, &addr_len);
 929         result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
 930                                  addr_len);
 931         if (result < 0) {
 932                 log_print("could not bind for connect: %d", result);
 933                 /* This *may* not indicate a critical error */
 934         }
 935
 936         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
 937
 938         log_print("connecting to %d", con->nodeid);
 939         result =
 940                 sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
 941                                    O_NONBLOCK);
 942         if (result == -EINPROGRESS)
 943                 result = 0;
 944         if (result == 0)
 945                 goto out;
 946
 947 out_err:
 948         if (con->sock) {
 949                 sock_release(con->sock);
 950                 con->sock = NULL;
 951         }
 952         /*
 953          * Some errors are fatal and this list might need adjusting. For other
 954          * errors we try again until the max number of retries is reached.
 955          */
 956         if (result != -EHOSTUNREACH && result != -ENETUNREACH &&
 957             result != -ENETDOWN && result != -EINVAL
 958             && result != -EPROTONOSUPPORT) {
 959                 lowcomms_connect_sock(con);
 960                 result = 0;
 961         }
 962 out:
 963         mutex_unlock(&con->sock_mutex);
 964         return;
 965 }
 966
 967 static struct socket *tcp_create_listen_sock(struct connection *con,
 968                                              struct sockaddr_storage *saddr)
 969 {
 970         struct socket *sock = NULL;
 971         int result = 0;
 972         int one = 1;
 973         int addr_len;
 974
 975         if (dlm_local_addr[0]->ss_family == AF_INET)
 976                 addr_len = sizeof(struct sockaddr_in);
 977         else
 978                 addr_len = sizeof(struct sockaddr_in6);
 979
 980         /* Create a socket to communicate with */
 981         result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM,
 982                                   IPPROTO_TCP, &sock);
 983         if (result < 0) {
 984                 log_print("Can't create listening comms socket");
 985                 goto create_out;
 986         }
 987
 988         result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
 989                                    (char *)&one, sizeof(one));
 990
 991         if (result < 0) {
 992                 log_print("Failed to set SO_REUSEADDR on socket: %d", result);
 993         }
 994         sock->sk->sk_user_data = con;
 995         con->rx_action = tcp_accept_from_sock;
 996         con->connect_action = tcp_connect_to_sock;
 997         con->sock = sock;
 998
 999         /* Bind to our port */
1000         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1001         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1002         if (result < 0) {
1003                 log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1004                 sock_release(sock);
1005                 sock = NULL;
1006                 con->sock = NULL;
1007                 goto create_out;
1008         }
1009         result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
1010                                  (char *)&one, sizeof(one));
1011         if (result < 0) {
1012                 log_print("Set keepalive failed: %d", result);
1013         }
1014
1015         result = sock->ops->listen(sock, 5);
1016         if (result < 0) {
1017                 log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1018                 sock_release(sock);
1019                 sock = NULL;
1020                 goto create_out;
1021         }
1022
1023 create_out:
1024         return sock;
1025 }
1026
1027 /* Get local addresses */
1028 static void init_local(void)
1029 {
1030         struct sockaddr_storage sas, *addr;
1031         int i;
1032
1033         dlm_local_count = 0;
1034         for (i = 0; i < DLM_MAX_ADDR_COUNT - 1; i++) {
1035                 if (dlm_our_addr(&sas, i))
1036                         break;
1037
1038                 addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1039                 if (!addr)
1040                         break;
1041                 memcpy(addr, &sas, sizeof(*addr));
1042                 dlm_local_addr[dlm_local_count++] = addr;
1043         }
1044 }
1045
1046 /* Bind to an IP address. SCTP allows multiple address so it can do
1047    multi-homing */
1048 static int add_sctp_bind_addr(struct connection *sctp_con,
1049                               struct sockaddr_storage *addr,
1050                               int addr_len, int num)
1051 {
1052         int result = 0;
1053
1054         if (num == 1)
1055                 result = kernel_bind(sctp_con->sock,
1056                                      (struct sockaddr *) addr,
1057                                      addr_len);
1058         else
1059                 result = kernel_setsockopt(sctp_con->sock, SOL_SCTP,
1060                                            SCTP_SOCKOPT_BINDX_ADD,
1061                                            (char *)addr, addr_len);
1062
1063         if (result < 0)
1064                 log_print("Can't bind to port %d addr number %d",
1065                           dlm_config.ci_tcp_port, num);
1066
1067         return result;
1068 }
1069
1070 /* Initialise SCTP socket and bind to all interfaces */
1071 static int sctp_listen_for_all(void)
1072 {
1073         struct socket *sock = NULL;
1074         struct sockaddr_storage localaddr;
1075         struct sctp_event_subscribe subscribe;
1076         int result = -EINVAL, num = 1, i, addr_len;
1077         struct connection *con = nodeid2con(0, GFP_KERNEL);
1078         int bufsize = NEEDED_RMEM;
1079
1080         if (!con)
1081                 return -ENOMEM;
1082
1083         log_print("Using SCTP for communications");
1084
1085         result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_SEQPACKET,
1086                                   IPPROTO_SCTP, &sock);
1087         if (result < 0) {
1088                 log_print("Can't create comms socket, check SCTP is loaded");
1089                 goto out;
1090         }
1091
1092         /* Listen for events */
1093         memset(&subscribe, 0, sizeof(subscribe));
1094         subscribe.sctp_data_io_event = 1;
1095         subscribe.sctp_association_event = 1;
1096         subscribe.sctp_send_failure_event = 1;
1097         subscribe.sctp_shutdown_event = 1;
1098         subscribe.sctp_partial_delivery_event = 1;
1099
1100         result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE,
1101                                  (char *)&bufsize, sizeof(bufsize));
1102         if (result)
1103                 log_print("Error increasing buffer space on socket %d", result);
1104
1105         result = kernel_setsockopt(sock, SOL_SCTP, SCTP_EVENTS,
1106                                    (char *)&subscribe, sizeof(subscribe));
1107         if (result < 0) {
1108                 log_print("Failed to set SCTP_EVENTS on socket: result=%d",
1109                           result);
1110                 goto create_delsock;
1111         }
1112
1113         /* Init con struct */
1114         sock->sk->sk_user_data = con;
1115         con->sock = sock;
1116         con->sock->sk->sk_data_ready = lowcomms_data_ready;
1117         con->rx_action = receive_from_sock;
1118         con->connect_action = sctp_init_assoc;
1119
1120         /* Bind to all interfaces. */
1121         for (i = 0; i < dlm_local_count; i++) {
1122                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
1123                 make_sockaddr(&localaddr, dlm_config.ci_tcp_port, &addr_len);
1124
1125                 result = add_sctp_bind_addr(con, &localaddr, addr_len, num);
1126                 if (result)
1127                         goto create_delsock;
1128                 ++num;
1129         }
1130
1131         result = sock->ops->listen(sock, 5);
1132         if (result < 0) {
1133                 log_print("Can't set socket listening");
1134                 goto create_delsock;
1135         }
1136
1137         return 0;
1138
1139 create_delsock:
1140         sock_release(sock);
1141         con->sock = NULL;
1142 out:
1143         return result;
1144 }
1145
1146 static int tcp_listen_for_all(void)
1147 {
1148         struct socket *sock = NULL;
1149         struct connection *con = nodeid2con(0, GFP_KERNEL);
1150         int result = -EINVAL;
1151
1152         if (!con)
1153                 return -ENOMEM;
1154
1155         /* We don't support multi-homed hosts */
1156         if (dlm_local_addr[1] != NULL) {
1157                 log_print("TCP protocol can't handle multi-homed hosts, "
1158                           "try SCTP");
1159                 return -EINVAL;
1160         }
1161
1162         log_print("Using TCP for communications");
1163
1164         sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
1165         if (sock) {
1166                 add_sock(sock, con);
1167                 result = 0;
1168         }
1169         else {
1170                 result = -EADDRINUSE;
1171         }
1172
1173         return result;
1174 }
1175
1176
1177
1178 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1179                                                      gfp_t allocation)
1180 {
1181         struct writequeue_entry *entry;
1182
1183         entry = kmalloc(sizeof(struct writequeue_entry), allocation);
1184         if (!entry)
1185                 return NULL;
1186
1187         entry->page = alloc_page(allocation);
1188         if (!entry->page) {
1189                 kfree(entry);
1190                 return NULL;
1191         }
1192
1193         entry->offset = 0;
1194         entry->len = 0;
1195         entry->end = 0;
1196         entry->users = 0;
1197         entry->con = con;
1198
1199         return entry;
1200 }
1201
1202 void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
1203 {
1204         struct connection *con;
1205         struct writequeue_entry *e;
1206         int offset = 0;
1207         int users = 0;
1208
1209         con = nodeid2con(nodeid, allocation);
1210         if (!con)
1211                 return NULL;
1212
1213         spin_lock(&con->writequeue_lock);
1214         e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
1215         if ((&e->list == &con->writequeue) ||
1216             (PAGE_CACHE_SIZE - e->end < len)) {
1217                 e = NULL;
1218         } else {
1219                 offset = e->end;
1220                 e->end += len;
1221                 users = e->users++;
1222         }
1223         spin_unlock(&con->writequeue_lock);
1224
1225         if (e) {
1226         got_one:
1227                 *ppc = page_address(e->page) + offset;
1228                 return e;
1229         }
1230
1231         e = new_writequeue_entry(con, allocation);
1232         if (e) {
1233                 spin_lock(&con->writequeue_lock);
1234                 offset = e->end;
1235                 e->end += len;
1236                 users = e->users++;
1237                 list_add_tail(&e->list, &con->writequeue);
1238                 spin_unlock(&con->writequeue_lock);
1239                 goto got_one;
1240         }
1241         return NULL;
1242 }
1243
1244 void dlm_lowcomms_commit_buffer(void *mh)
1245 {
1246         struct writequeue_entry *e = (struct writequeue_entry *)mh;
1247         struct connection *con = e->con;
1248         int users;
1249
1250         spin_lock(&con->writequeue_lock);
1251         users = --e->users;
1252         if (users)
1253                 goto out;
1254         e->len = e->end - e->offset;
1255         spin_unlock(&con->writequeue_lock);
1256
1257         if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) {
1258                 queue_work(send_workqueue, &con->swork);
1259         }
1260         return;
1261
1262 out:
1263         spin_unlock(&con->writequeue_lock);
1264         return;
1265 }
1266
1267 /* Send a message */
1268 static void send_to_sock(struct connection *con)
1269 {
1270         int ret = 0;
1271         ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int);
1272         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1273         struct writequeue_entry *e;
1274         int len, offset;
1275
1276         mutex_lock(&con->sock_mutex);
1277         if (con->sock == NULL)
1278                 goto out_connect;
1279
1280         sendpage = con->sock->ops->sendpage;
1281
1282         spin_lock(&con->writequeue_lock);
1283         for (;;) {
1284                 e = list_entry(con->writequeue.next, struct writequeue_entry,
1285                                list);
1286                 if ((struct list_head *) e == &con->writequeue)
1287                         break;
1288
1289                 len = e->len;
1290                 offset = e->offset;
1291                 BUG_ON(len == 0 && e->users == 0);
1292                 spin_unlock(&con->writequeue_lock);
1293
1294                 ret = 0;
1295                 if (len) {
1296                         ret = sendpage(con->sock, e->page, offset, len,
1297                                        msg_flags);
1298                         if (ret == -EAGAIN || ret == 0) {
1299                                 cond_resched();
1300                                 goto out;
1301                         }
1302                         if (ret <= 0)
1303                                 goto send_error;
1304                 }
1305                         /* Don't starve people filling buffers */
1306                         cond_resched();
1307
1308                 spin_lock(&con->writequeue_lock);
1309                 e->offset += ret;
1310                 e->len -= ret;
1311
1312                 if (e->len == 0 && e->users == 0) {
1313                         list_del(&e->list);
1314                         free_entry(e);
1315                         continue;
1316                 }
1317         }
1318         spin_unlock(&con->writequeue_lock);
1319 out:
1320         mutex_unlock(&con->sock_mutex);
1321         return;
1322
1323 send_error:
1324         mutex_unlock(&con->sock_mutex);
1325         close_connection(con, false);
1326         lowcomms_connect_sock(con);
1327         return;
1328
1329 out_connect:
1330         mutex_unlock(&con->sock_mutex);
1331         if (!test_bit(CF_INIT_PENDING, &con->flags))
1332                 lowcomms_connect_sock(con);
1333         return;
1334 }
1335
1336 static void clean_one_writequeue(struct connection *con)
1337 {
1338         struct writequeue_entry *e, *safe;
1339
1340         spin_lock(&con->writequeue_lock);
1341         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1342                 list_del(&e->list);
1343                 free_entry(e);
1344         }
1345         spin_unlock(&con->writequeue_lock);
1346 }
1347
1348 /* Called from recovery when it knows that a node has
1349    left the cluster */
1350 int dlm_lowcomms_close(int nodeid)
1351 {
1352         struct connection *con;
1353
1354         log_print("closing connection to node %d", nodeid);
1355         con = nodeid2con(nodeid, 0);
1356         if (con) {
1357                 clean_one_writequeue(con);
1358                 close_connection(con, true);
1359         }
1360         return 0;
1361 }
1362
1363 /* Receive workqueue function */
1364 static void process_recv_sockets(struct work_struct *work)
1365 {
1366         struct connection *con = container_of(work, struct connection, rwork);
1367         int err;
1368
1369         clear_bit(CF_READ_PENDING, &con->flags);
1370         do {
1371                 err = con->rx_action(con);
1372         } while (!err);
1373 }
1374
1375 /* Send workqueue function */
1376 static void process_send_sockets(struct work_struct *work)
1377 {
1378         struct connection *con = container_of(work, struct connection, swork);
1379
1380         if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) {
1381                 con->connect_action(con);
1382         }
1383         clear_bit(CF_WRITE_PENDING, &con->flags);
1384         send_to_sock(con);
1385 }
1386
1387
1388 /* Discard all entries on the write queues */
1389 static void clean_writequeues(void)
1390 {
1391         foreach_conn(clean_one_writequeue);
1392 }
1393
1394 static void work_stop(void)
1395 {
1396         destroy_workqueue(recv_workqueue);
1397         destroy_workqueue(send_workqueue);
1398 }
1399
1400 static int work_start(void)
1401 {
1402         int error;
1403         recv_workqueue = create_workqueue("dlm_recv");
1404         error = IS_ERR(recv_workqueue);
1405         if (error) {
1406                 log_print("can't start dlm_recv %d", error);
1407                 return error;
1408         }
1409
1410         send_workqueue = create_singlethread_workqueue("dlm_send");
1411         error = IS_ERR(send_workqueue);
1412         if (error) {
1413                 log_print("can't start dlm_send %d", error);
1414                 destroy_workqueue(recv_workqueue);
1415                 return error;
1416         }
1417
1418         return 0;
1419 }
1420
1421 static void stop_conn(struct connection *con)
1422 {
1423         con->flags |= 0x0F;
1424         if (con->sock)
1425                 con->sock->sk->sk_user_data = NULL;
1426 }
1427
1428 static void free_conn(struct connection *con)
1429 {
1430         close_connection(con, true);
1431         if (con->othercon)
1432                 kmem_cache_free(con_cache, con->othercon);
1433         hlist_del(&con->list);
1434         kmem_cache_free(con_cache, con);
1435 }
1436
1437 void dlm_lowcomms_stop(void)
1438 {
1439         /* Set all the flags to prevent any
1440            socket activity.
1441         */
1442         mutex_lock(&connections_lock);
1443         foreach_conn(stop_conn);
1444         mutex_unlock(&connections_lock);
1445
1446         work_stop();
1447
1448         mutex_lock(&connections_lock);
1449         clean_writequeues();
1450
1451         foreach_conn(free_conn);
1452
1453         mutex_unlock(&connections_lock);
1454         kmem_cache_destroy(con_cache);
1455 }
1456
1457 int dlm_lowcomms_start(void)
1458 {
1459         int error = -EINVAL;
1460         struct connection *con;
1461         int i;
1462
1463         for (i = 0; i < CONN_HASH_SIZE; i++)
1464                 INIT_HLIST_HEAD(&connection_hash[i]);
1465
1466         init_local();
1467         if (!dlm_local_count) {
1468                 error = -ENOTCONN;
1469                 log_print("no local IP address has been set");
1470                 goto out;
1471         }
1472
1473         error = -ENOMEM;
1474         con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
1475                                       __alignof__(struct connection), 0,
1476                                       NULL);
1477         if (!con_cache)
1478                 goto out;
1479
1480         /* Start listening */
1481         if (dlm_config.ci_protocol == 0)
1482                 error = tcp_listen_for_all();
1483         else
1484                 error = sctp_listen_for_all();
1485         if (error)
1486                 goto fail_unlisten;
1487
1488         error = work_start();
1489         if (error)
1490                 goto fail_unlisten;
1491
1492         return 0;
1493
1494 fail_unlisten:
1495         con = nodeid2con(0,0);
1496         if (con) {
1497                 close_connection(con, false);
1498                 kmem_cache_free(con_cache, con);
1499         }
1500         kmem_cache_destroy(con_cache);
1501
1502 out:
1503         return error;
1504 }