fs/dlm/lowcomms.c

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