fs/dlm/lowcomms.c

   1 /******************************************************************************
   2 *******************************************************************************
   3 **
   4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5 **  Copyright (C) 2004-2009 Red Hat, Inc.  All rights reserved.
   6 **
   7 **  This copyrighted material is made available to anyone wishing to use,
   8 **  modify, copy, or redistribute it subject to the terms and conditions
   9 **  of the GNU General Public License v.2.
  10 **
  11 *******************************************************************************
  12 ******************************************************************************/
  13
  14 /*
  15  * lowcomms.c
  16  *
  17  * This is the "low-level" comms layer.
  18  *
  19  * It is responsible for sending/receiving messages
  20  * from other nodes in the cluster.
  21  *
  22  * Cluster nodes are referred to by their nodeids. nodeids are
  23  * simply 32 bit numbers to the locking module - if they need to
  24  * be expanded for the cluster infrastructure then that is its
  25  * responsibility. It is this layer's
  26  * responsibility to resolve these into IP address or
  27  * whatever it needs for inter-node communication.
  28  *
  29  * The comms level is two kernel threads that deal mainly with
  30  * the receiving of messages from other nodes and passing them
  31  * up to the mid-level comms layer (which understands the
  32  * message format) for execution by the locking core, and
  33  * a send thread which does all the setting up of connections
  34  * to remote nodes and the sending of data. Threads are not allowed
  35  * to send their own data because it may cause them to wait in times
  36  * of high load. Also, this way, the sending thread can collect together
  37  * messages bound for one node and send them in one block.
  38  *
  39  * lowcomms will choose to use either TCP or SCTP as its transport layer
  40  * depending on the configuration variable 'protocol'. This should be set
  41  * to 0 (default) for TCP or 1 for SCTP. It should be configured using a
  42  * cluster-wide mechanism as it must be the same on all nodes of the cluster
  43  * for the DLM to function.
  44  *
  45  */
  46
  47 #include <asm/ioctls.h>
  48 #include <net/sock.h>
  49 #include <net/tcp.h>
  50 #include <linux/pagemap.h>
  51 #include <linux/file.h>
  52 #include <linux/mutex.h>
  53 #include <linux/sctp.h>
  54 #include <linux/slab.h>
  55 #include <net/sctp/sctp.h>
  56 #include <net/ipv6.h>
  57
  58 #include "dlm_internal.h"
  59 #include "lowcomms.h"
  60 #include "midcomms.h"
  61 #include "config.h"
  62
  63 #define NEEDED_RMEM (4*1024*1024)
  64 #define CONN_HASH_SIZE 32
  65
  66 /* Number of messages to send before rescheduling */
  67 #define MAX_SEND_MSG_COUNT 25
  68
  69 struct cbuf {
  70         unsigned int base;
  71         unsigned int len;
  72         unsigned int mask;
  73 };
  74
  75 static void cbuf_add(struct cbuf *cb, int n)
  76 {
  77         cb->len += n;
  78 }
  79
  80 static int cbuf_data(struct cbuf *cb)
  81 {
  82         return ((cb->base + cb->len) & cb->mask);
  83 }
  84
  85 static void cbuf_init(struct cbuf *cb, int size)
  86 {
  87         cb->base = cb->len = 0;
  88         cb->mask = size-1;
  89 }
  90
  91 static void cbuf_eat(struct cbuf *cb, int n)
  92 {
  93         cb->len  -= n;
  94         cb->base += n;
  95         cb->base &= cb->mask;
  96 }
  97
  98 static bool cbuf_empty(struct cbuf *cb)
  99 {
 100         return cb->len == 0;
 101 }
 102
 103 struct connection {
 104         struct socket *sock;    /* NULL if not connected */
 105         uint32_t nodeid;        /* So we know who we are in the list */
 106         struct mutex sock_mutex;
 107         unsigned long flags;
 108 #define CF_READ_PENDING 1
 109 #define CF_WRITE_PENDING 2
 110 #define CF_CONNECT_PENDING 3
 111 #define CF_INIT_PENDING 4
 112 #define CF_IS_OTHERCON 5
 113 #define CF_CLOSE 6
 114 #define CF_APP_LIMITED 7
 115         struct list_head writequeue;  /* List of outgoing writequeue_entries */
 116         spinlock_t writequeue_lock;
 117         int (*rx_action) (struct connection *); /* What to do when active */
 118         void (*connect_action) (struct connection *);   /* What to do to connect */
 119         struct page *rx_page;
 120         struct cbuf cb;
 121         int retries;
 122 #define MAX_CONNECT_RETRIES 3
 123         struct hlist_node list;
 124         struct connection *othercon;
 125         struct work_struct rwork; /* Receive workqueue */
 126         struct work_struct swork; /* Send workqueue */
 127         void (*orig_error_report)(struct sock *);
 128         void (*orig_data_ready)(struct sock *);
 129         void (*orig_state_change)(struct sock *);
 130         void (*orig_write_space)(struct sock *);
 131 };
 132 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 133
 134 /* An entry waiting to be sent */
 135 struct writequeue_entry {
 136         struct list_head list;
 137         struct page *page;
 138         int offset;
 139         int len;
 140         int end;
 141         int users;
 142         struct connection *con;
 143 };
 144
 145 struct dlm_node_addr {
 146         struct list_head list;
 147         int nodeid;
 148         int addr_count;
 149         int curr_addr_index;
 150         struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
 151 };
 152
 153 static LIST_HEAD(dlm_node_addrs);
 154 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
 155
 156 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 157 static int dlm_local_count;
 158 static int dlm_allow_conn;
 159
 160 /* Work queues */
 161 static struct workqueue_struct *recv_workqueue;
 162 static struct workqueue_struct *send_workqueue;
 163
 164 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 165 static DEFINE_MUTEX(connections_lock);
 166 static struct kmem_cache *con_cache;
 167
 168 static void process_recv_sockets(struct work_struct *work);
 169 static void process_send_sockets(struct work_struct *work);
 170
 171
 172 /* This is deliberately very simple because most clusters have simple
 173    sequential nodeids, so we should be able to go straight to a connection
 174    struct in the array */
 175 static inline int nodeid_hash(int nodeid)
 176 {
 177         return nodeid & (CONN_HASH_SIZE-1);
 178 }
 179
 180 static struct connection *__find_con(int nodeid)
 181 {
 182         int r;
 183         struct connection *con;
 184
 185         r = nodeid_hash(nodeid);
 186
 187         hlist_for_each_entry(con, &connection_hash[r], list) {
 188                 if (con->nodeid == nodeid)
 189                         return con;
 190         }
 191         return NULL;
 192 }
 193
 194 /*
 195  * If 'allocation' is zero then we don't attempt to create a new
 196  * connection structure for this node.
 197  */
 198 static struct connection *__nodeid2con(int nodeid, gfp_t alloc)
 199 {
 200         struct connection *con = NULL;
 201         int r;
 202
 203         con = __find_con(nodeid);
 204         if (con || !alloc)
 205                 return con;
 206
 207         con = kmem_cache_zalloc(con_cache, alloc);
 208         if (!con)
 209                 return NULL;
 210
 211         r = nodeid_hash(nodeid);
 212         hlist_add_head(&con->list, &connection_hash[r]);
 213
 214         con->nodeid = nodeid;
 215         mutex_init(&con->sock_mutex);
 216         INIT_LIST_HEAD(&con->writequeue);
 217         spin_lock_init(&con->writequeue_lock);
 218         INIT_WORK(&con->swork, process_send_sockets);
 219         INIT_WORK(&con->rwork, process_recv_sockets);
 220
 221         /* Setup action pointers for child sockets */
 222         if (con->nodeid) {
 223                 struct connection *zerocon = __find_con(0);
 224
 225                 con->connect_action = zerocon->connect_action;
 226                 if (!con->rx_action)
 227                         con->rx_action = zerocon->rx_action;
 228         }
 229
 230         return con;
 231 }
 232
 233 /* Loop round all connections */
 234 static void foreach_conn(void (*conn_func)(struct connection *c))
 235 {
 236         int i;
 237         struct hlist_node *n;
 238         struct connection *con;
 239
 240         for (i = 0; i < CONN_HASH_SIZE; i++) {
 241                 hlist_for_each_entry_safe(con, n, &connection_hash[i], list)
 242                         conn_func(con);
 243         }
 244 }
 245
 246 static struct connection *nodeid2con(int nodeid, gfp_t allocation)
 247 {
 248         struct connection *con;
 249
 250         mutex_lock(&connections_lock);
 251         con = __nodeid2con(nodeid, allocation);
 252         mutex_unlock(&connections_lock);
 253
 254         return con;
 255 }
 256
 257 static struct dlm_node_addr *find_node_addr(int nodeid)
 258 {
 259         struct dlm_node_addr *na;
 260
 261         list_for_each_entry(na, &dlm_node_addrs, list) {
 262                 if (na->nodeid == nodeid)
 263                         return na;
 264         }
 265         return NULL;
 266 }
 267
 268 static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y)
 269 {
 270         switch (x->ss_family) {
 271         case AF_INET: {
 272                 struct sockaddr_in *sinx = (struct sockaddr_in *)x;
 273                 struct sockaddr_in *siny = (struct sockaddr_in *)y;
 274                 if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
 275                         return 0;
 276                 if (sinx->sin_port != siny->sin_port)
 277                         return 0;
 278                 break;
 279         }
 280         case AF_INET6: {
 281                 struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
 282                 struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
 283                 if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
 284                         return 0;
 285                 if (sinx->sin6_port != siny->sin6_port)
 286                         return 0;
 287                 break;
 288         }
 289         default:
 290                 return 0;
 291         }
 292         return 1;
 293 }
 294
 295 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
 296                           struct sockaddr *sa_out, bool try_new_addr)
 297 {
 298         struct sockaddr_storage sas;
 299         struct dlm_node_addr *na;
 300
 301         if (!dlm_local_count)
 302                 return -1;
 303
 304         spin_lock(&dlm_node_addrs_spin);
 305         na = find_node_addr(nodeid);
 306         if (na && na->addr_count) {
 307                 memcpy(&sas, na->addr[na->curr_addr_index],
 308                        sizeof(struct sockaddr_storage));
 309
 310                 if (try_new_addr) {
 311                         na->curr_addr_index++;
 312                         if (na->curr_addr_index == na->addr_count)
 313                                 na->curr_addr_index = 0;
 314                 }
 315         }
 316         spin_unlock(&dlm_node_addrs_spin);
 317
 318         if (!na)
 319                 return -EEXIST;
 320
 321         if (!na->addr_count)
 322                 return -ENOENT;
 323
 324         if (sas_out)
 325                 memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
 326
 327         if (!sa_out)
 328                 return 0;
 329
 330         if (dlm_local_addr[0]->ss_family == AF_INET) {
 331                 struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
 332                 struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
 333                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 334         } else {
 335                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
 336                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
 337                 ret6->sin6_addr = in6->sin6_addr;
 338         }
 339
 340         return 0;
 341 }
 342
 343 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid)
 344 {
 345         struct dlm_node_addr *na;
 346         int rv = -EEXIST;
 347         int addr_i;
 348
 349         spin_lock(&dlm_node_addrs_spin);
 350         list_for_each_entry(na, &dlm_node_addrs, list) {
 351                 if (!na->addr_count)
 352                         continue;
 353
 354                 for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
 355                         if (addr_compare(na->addr[addr_i], addr)) {
 356                                 *nodeid = na->nodeid;
 357                                 rv = 0;
 358                                 goto unlock;
 359                         }
 360                 }
 361         }
 362 unlock:
 363         spin_unlock(&dlm_node_addrs_spin);
 364         return rv;
 365 }
 366
 367 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 368 {
 369         struct sockaddr_storage *new_addr;
 370         struct dlm_node_addr *new_node, *na;
 371
 372         new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
 373         if (!new_node)
 374                 return -ENOMEM;
 375
 376         new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
 377         if (!new_addr) {
 378                 kfree(new_node);
 379                 return -ENOMEM;
 380         }
 381
 382         memcpy(new_addr, addr, len);
 383
 384         spin_lock(&dlm_node_addrs_spin);
 385         na = find_node_addr(nodeid);
 386         if (!na) {
 387                 new_node->nodeid = nodeid;
 388                 new_node->addr[0] = new_addr;
 389                 new_node->addr_count = 1;
 390                 list_add(&new_node->list, &dlm_node_addrs);
 391                 spin_unlock(&dlm_node_addrs_spin);
 392                 return 0;
 393         }
 394
 395         if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
 396                 spin_unlock(&dlm_node_addrs_spin);
 397                 kfree(new_addr);
 398                 kfree(new_node);
 399                 return -ENOSPC;
 400         }
 401
 402         na->addr[na->addr_count++] = new_addr;
 403         spin_unlock(&dlm_node_addrs_spin);
 404         kfree(new_node);
 405         return 0;
 406 }
 407
 408 /* Data available on socket or listen socket received a connect */
 409 static void lowcomms_data_ready(struct sock *sk)
 410 {
 411         struct connection *con = sock2con(sk);
 412         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 413                 queue_work(recv_workqueue, &con->rwork);
 414 }
 415
 416 static void lowcomms_write_space(struct sock *sk)
 417 {
 418         struct connection *con = sock2con(sk);
 419
 420         if (!con)
 421                 return;
 422
 423         clear_bit(SOCK_NOSPACE, &con->sock->flags);
 424
 425         if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
 426                 con->sock->sk->sk_write_pending--;
 427                 clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
 428         }
 429
 430         if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags))
 431                 queue_work(send_workqueue, &con->swork);
 432 }
 433
 434 static inline void lowcomms_connect_sock(struct connection *con)
 435 {
 436         if (test_bit(CF_CLOSE, &con->flags))
 437                 return;
 438         if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags))
 439                 queue_work(send_workqueue, &con->swork);
 440 }
 441
 442 static void lowcomms_state_change(struct sock *sk)
 443 {
 444         /* SCTP layer is not calling sk_data_ready when the connection
 445          * is done, so we catch the signal through here. Also, it
 446          * doesn't switch socket state when entering shutdown, so we
 447          * skip the write in that case.
 448          */
 449         if (sk->sk_shutdown) {
 450                 if (sk->sk_shutdown == RCV_SHUTDOWN)
 451                         lowcomms_data_ready(sk);
 452         } else if (sk->sk_state == TCP_ESTABLISHED) {
 453                 lowcomms_write_space(sk);
 454         }
 455 }
 456
 457 int dlm_lowcomms_connect_node(int nodeid)
 458 {
 459         struct connection *con;
 460
 461         if (nodeid == dlm_our_nodeid())
 462                 return 0;
 463
 464         con = nodeid2con(nodeid, GFP_NOFS);
 465         if (!con)
 466                 return -ENOMEM;
 467         lowcomms_connect_sock(con);
 468         return 0;
 469 }
 470
 471 static void lowcomms_error_report(struct sock *sk)
 472 {
 473         struct connection *con;
 474         struct sockaddr_storage saddr;
 475         int buflen;
 476         void (*orig_report)(struct sock *) = NULL;
 477
 478         read_lock_bh(&sk->sk_callback_lock);
 479         con = sock2con(sk);
 480         if (con == NULL)
 481                 goto out;
 482
 483         orig_report = con->orig_error_report;
 484         if (con->sock == NULL ||
 485             kernel_getpeername(con->sock, (struct sockaddr *)&saddr, &buflen)) {
 486                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 487                                    "sending to node %d, port %d, "
 488                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 489                                    con->nodeid, dlm_config.ci_tcp_port,
 490                                    sk->sk_err, sk->sk_err_soft);
 491         } else if (saddr.ss_family == AF_INET) {
 492                 struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr;
 493
 494                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 495                                    "sending to node %d at %pI4, port %d, "
 496                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 497                                    con->nodeid, &sin4->sin_addr.s_addr,
 498                                    dlm_config.ci_tcp_port, sk->sk_err,
 499                                    sk->sk_err_soft);
 500         } else {
 501                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr;
 502
 503                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 504                                    "sending to node %d at %u.%u.%u.%u, "
 505                                    "port %d, sk_err=%d/%d\n", dlm_our_nodeid(),
 506                                    con->nodeid, sin6->sin6_addr.s6_addr32[0],
 507                                    sin6->sin6_addr.s6_addr32[1],
 508                                    sin6->sin6_addr.s6_addr32[2],
 509                                    sin6->sin6_addr.s6_addr32[3],
 510                                    dlm_config.ci_tcp_port, sk->sk_err,
 511                                    sk->sk_err_soft);
 512         }
 513 out:
 514         read_unlock_bh(&sk->sk_callback_lock);
 515         if (orig_report)
 516                 orig_report(sk);
 517 }
 518
 519 /* Note: sk_callback_lock must be locked before calling this function. */
 520 static void save_callbacks(struct connection *con, struct sock *sk)
 521 {
 522         con->orig_data_ready = sk->sk_data_ready;
 523         con->orig_state_change = sk->sk_state_change;
 524         con->orig_write_space = sk->sk_write_space;
 525         con->orig_error_report = sk->sk_error_report;
 526 }
 527
 528 static void restore_callbacks(struct connection *con, struct sock *sk)
 529 {
 530         write_lock_bh(&sk->sk_callback_lock);
 531         sk->sk_user_data = NULL;
 532         sk->sk_data_ready = con->orig_data_ready;
 533         sk->sk_state_change = con->orig_state_change;
 534         sk->sk_write_space = con->orig_write_space;
 535         sk->sk_error_report = con->orig_error_report;
 536         write_unlock_bh(&sk->sk_callback_lock);
 537 }
 538
 539 /* Make a socket active */
 540 static void add_sock(struct socket *sock, struct connection *con, bool save_cb)
 541 {
 542         struct sock *sk = sock->sk;
 543
 544         write_lock_bh(&sk->sk_callback_lock);
 545         con->sock = sock;
 546
 547         sk->sk_user_data = con;
 548         if (save_cb)
 549                 save_callbacks(con, sk);
 550         /* Install a data_ready callback */
 551         sk->sk_data_ready = lowcomms_data_ready;
 552         sk->sk_write_space = lowcomms_write_space;
 553         sk->sk_state_change = lowcomms_state_change;
 554         sk->sk_allocation = GFP_NOFS;
 555         sk->sk_error_report = lowcomms_error_report;
 556         write_unlock_bh(&sk->sk_callback_lock);
 557 }
 558
 559 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 560    length */
 561 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 562                           int *addr_len)
 563 {
 564         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 565         if (saddr->ss_family == AF_INET) {
 566                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 567                 in4_addr->sin_port = cpu_to_be16(port);
 568                 *addr_len = sizeof(struct sockaddr_in);
 569                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 570         } else {
 571                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 572                 in6_addr->sin6_port = cpu_to_be16(port);
 573                 *addr_len = sizeof(struct sockaddr_in6);
 574         }
 575         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 576 }
 577
 578 /* Close a remote connection and tidy up */
 579 static void close_connection(struct connection *con, bool and_other,
 580                              bool tx, bool rx)
 581 {
 582         clear_bit(CF_CONNECT_PENDING, &con->flags);
 583         clear_bit(CF_WRITE_PENDING, &con->flags);
 584         if (tx && cancel_work_sync(&con->swork))
 585                 log_print("canceled swork for node %d", con->nodeid);
 586         if (rx && cancel_work_sync(&con->rwork))
 587                 log_print("canceled rwork for node %d", con->nodeid);
 588
 589         mutex_lock(&con->sock_mutex);
 590         if (con->sock) {
 591                 if (!test_bit(CF_IS_OTHERCON, &con->flags))
 592                         restore_callbacks(con, con->sock->sk);
 593                 sock_release(con->sock);
 594                 con->sock = NULL;
 595         }
 596         if (con->othercon && and_other) {
 597                 /* Will only re-enter once. */
 598                 close_connection(con->othercon, false, true, true);
 599         }
 600         if (con->rx_page) {
 601                 __free_page(con->rx_page);
 602                 con->rx_page = NULL;
 603         }
 604
 605         con->retries = 0;
 606         mutex_unlock(&con->sock_mutex);
 607 }
 608
 609 /* Data received from remote end */
 610 static int receive_from_sock(struct connection *con)
 611 {
 612         int ret = 0;
 613         struct msghdr msg = {};
 614         struct kvec iov[2];
 615         unsigned len;
 616         int r;
 617         int call_again_soon = 0;
 618         int nvec;
 619
 620         mutex_lock(&con->sock_mutex);
 621
 622         if (con->sock == NULL) {
 623                 ret = -EAGAIN;
 624                 goto out_close;
 625         }
 626         if (con->nodeid == 0) {
 627                 ret = -EINVAL;
 628                 goto out_close;
 629         }
 630
 631         if (con->rx_page == NULL) {
 632                 /*
 633                  * This doesn't need to be atomic, but I think it should
 634                  * improve performance if it is.
 635                  */
 636                 con->rx_page = alloc_page(GFP_ATOMIC);
 637                 if (con->rx_page == NULL)
 638                         goto out_resched;
 639                 cbuf_init(&con->cb, PAGE_SIZE);
 640         }
 641
 642         /*
 643          * iov[0] is the bit of the circular buffer between the current end
 644          * point (cb.base + cb.len) and the end of the buffer.
 645          */
 646         iov[0].iov_len = con->cb.base - cbuf_data(&con->cb);
 647         iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb);
 648         iov[1].iov_len = 0;
 649         nvec = 1;
 650
 651         /*
 652          * iov[1] is the bit of the circular buffer between the start of the
 653          * buffer and the start of the currently used section (cb.base)
 654          */
 655         if (cbuf_data(&con->cb) >= con->cb.base) {
 656                 iov[0].iov_len = PAGE_SIZE - cbuf_data(&con->cb);
 657                 iov[1].iov_len = con->cb.base;
 658                 iov[1].iov_base = page_address(con->rx_page);
 659                 nvec = 2;
 660         }
 661         len = iov[0].iov_len + iov[1].iov_len;
 662
 663         r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len,
 664                                MSG_DONTWAIT | MSG_NOSIGNAL);
 665         if (ret <= 0)
 666                 goto out_close;
 667         else if (ret == len)
 668                 call_again_soon = 1;
 669
 670         cbuf_add(&con->cb, ret);
 671         ret = dlm_process_incoming_buffer(con->nodeid,
 672                                           page_address(con->rx_page),
 673                                           con->cb.base, con->cb.len,
 674                                           PAGE_SIZE);
 675         if (ret == -EBADMSG) {
 676                 log_print("lowcomms: addr=%p, base=%u, len=%u, read=%d",
 677                           page_address(con->rx_page), con->cb.base,
 678                           con->cb.len, r);
 679         }
 680         if (ret < 0)
 681                 goto out_close;
 682         cbuf_eat(&con->cb, ret);
 683
 684         if (cbuf_empty(&con->cb) && !call_again_soon) {
 685                 __free_page(con->rx_page);
 686                 con->rx_page = NULL;
 687         }
 688
 689         if (call_again_soon)
 690                 goto out_resched;
 691         mutex_unlock(&con->sock_mutex);
 692         return 0;
 693
 694 out_resched:
 695         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 696                 queue_work(recv_workqueue, &con->rwork);
 697         mutex_unlock(&con->sock_mutex);
 698         return -EAGAIN;
 699
 700 out_close:
 701         mutex_unlock(&con->sock_mutex);
 702         if (ret != -EAGAIN) {
 703                 close_connection(con, false, true, false);
 704                 /* Reconnect when there is something to send */
 705         }
 706         /* Don't return success if we really got EOF */
 707         if (ret == 0)
 708                 ret = -EAGAIN;
 709
 710         return ret;
 711 }
 712
 713 /* Listening socket is busy, accept a connection */
 714 static int tcp_accept_from_sock(struct connection *con)
 715 {
 716         int result;
 717         struct sockaddr_storage peeraddr;
 718         struct socket *newsock;
 719         int len;
 720         int nodeid;
 721         struct connection *newcon;
 722         struct connection *addcon;
 723
 724         mutex_lock(&connections_lock);
 725         if (!dlm_allow_conn) {
 726                 mutex_unlock(&connections_lock);
 727                 return -1;
 728         }
 729         mutex_unlock(&connections_lock);
 730
 731         memset(&peeraddr, 0, sizeof(peeraddr));
 732         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
 733                                   SOCK_STREAM, IPPROTO_TCP, &newsock);
 734         if (result < 0)
 735                 return -ENOMEM;
 736
 737         mutex_lock_nested(&con->sock_mutex, 0);
 738
 739         result = -ENOTCONN;
 740         if (con->sock == NULL)
 741                 goto accept_err;
 742
 743         newsock->type = con->sock->type;
 744         newsock->ops = con->sock->ops;
 745
 746         result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK, true);
 747         if (result < 0)
 748                 goto accept_err;
 749
 750         /* Get the connected socket's peer */
 751         memset(&peeraddr, 0, sizeof(peeraddr));
 752         if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr,
 753                                   &len, 2)) {
 754                 result = -ECONNABORTED;
 755                 goto accept_err;
 756         }
 757
 758         /* Get the new node's NODEID */
 759         make_sockaddr(&peeraddr, 0, &len);
 760         if (addr_to_nodeid(&peeraddr, &nodeid)) {
 761                 unsigned char *b=(unsigned char *)&peeraddr;
 762                 log_print("connect from non cluster node");
 763                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
 764                                      b, sizeof(struct sockaddr_storage));
 765                 sock_release(newsock);
 766                 mutex_unlock(&con->sock_mutex);
 767                 return -1;
 768         }
 769
 770         log_print("got connection from %d", nodeid);
 771
 772         /*  Check to see if we already have a connection to this node. This
 773          *  could happen if the two nodes initiate a connection at roughly
 774          *  the same time and the connections cross on the wire.
 775          *  In this case we store the incoming one in "othercon"
 776          */
 777         newcon = nodeid2con(nodeid, GFP_NOFS);
 778         if (!newcon) {
 779                 result = -ENOMEM;
 780                 goto accept_err;
 781         }
 782         mutex_lock_nested(&newcon->sock_mutex, 1);
 783         if (newcon->sock) {
 784                 struct connection *othercon = newcon->othercon;
 785
 786                 if (!othercon) {
 787                         othercon = kmem_cache_zalloc(con_cache, GFP_NOFS);
 788                         if (!othercon) {
 789                                 log_print("failed to allocate incoming socket");
 790                                 mutex_unlock(&newcon->sock_mutex);
 791                                 result = -ENOMEM;
 792                                 goto accept_err;
 793                         }
 794                         othercon->nodeid = nodeid;
 795                         othercon->rx_action = receive_from_sock;
 796                         mutex_init(&othercon->sock_mutex);
 797                         INIT_WORK(&othercon->swork, process_send_sockets);
 798                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 799                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 800                 }
 801                 if (!othercon->sock) {
 802                         newcon->othercon = othercon;
 803                         othercon->sock = newsock;
 804                         newsock->sk->sk_user_data = othercon;
 805                         add_sock(newsock, othercon, false);
 806                         addcon = othercon;
 807                 }
 808                 else {
 809                         printk("Extra connection from node %d attempted\n", nodeid);
 810                         result = -EAGAIN;
 811                         mutex_unlock(&newcon->sock_mutex);
 812                         goto accept_err;
 813                 }
 814         }
 815         else {
 816                 newsock->sk->sk_user_data = newcon;
 817                 newcon->rx_action = receive_from_sock;
 818                 /* accept copies the sk after we've saved the callbacks, so we
 819                    don't want to save them a second time or comm errors will
 820                    result in calling sk_error_report recursively. */
 821                 add_sock(newsock, newcon, false);
 822                 addcon = newcon;
 823         }
 824
 825         mutex_unlock(&newcon->sock_mutex);
 826
 827         /*
 828          * Add it to the active queue in case we got data
 829          * between processing the accept adding the socket
 830          * to the read_sockets list
 831          */
 832         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 833                 queue_work(recv_workqueue, &addcon->rwork);
 834         mutex_unlock(&con->sock_mutex);
 835
 836         return 0;
 837
 838 accept_err:
 839         mutex_unlock(&con->sock_mutex);
 840         sock_release(newsock);
 841
 842         if (result != -EAGAIN)
 843                 log_print("error accepting connection from node: %d", result);
 844         return result;
 845 }
 846
 847 static int sctp_accept_from_sock(struct connection *con)
 848 {
 849         /* Check that the new node is in the lockspace */
 850         struct sctp_prim prim;
 851         int nodeid;
 852         int prim_len, ret;
 853         int addr_len;
 854         struct connection *newcon;
 855         struct connection *addcon;
 856         struct socket *newsock;
 857
 858         mutex_lock(&connections_lock);
 859         if (!dlm_allow_conn) {
 860                 mutex_unlock(&connections_lock);
 861                 return -1;
 862         }
 863         mutex_unlock(&connections_lock);
 864
 865         mutex_lock_nested(&con->sock_mutex, 0);
 866
 867         ret = kernel_accept(con->sock, &newsock, O_NONBLOCK);
 868         if (ret < 0)
 869                 goto accept_err;
 870
 871         memset(&prim, 0, sizeof(struct sctp_prim));
 872         prim_len = sizeof(struct sctp_prim);
 873
 874         ret = kernel_getsockopt(newsock, IPPROTO_SCTP, SCTP_PRIMARY_ADDR,
 875                                 (char *)&prim, &prim_len);
 876         if (ret < 0) {
 877                 log_print("getsockopt/sctp_primary_addr failed: %d", ret);
 878                 goto accept_err;
 879         }
 880
 881         make_sockaddr(&prim.ssp_addr, 0, &addr_len);
 882         ret = addr_to_nodeid(&prim.ssp_addr, &nodeid);
 883         if (ret) {
 884                 unsigned char *b = (unsigned char *)&prim.ssp_addr;
 885
 886                 log_print("reject connect from unknown addr");
 887                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
 888                                      b, sizeof(struct sockaddr_storage));
 889                 goto accept_err;
 890         }
 891
 892         newcon = nodeid2con(nodeid, GFP_NOFS);
 893         if (!newcon) {
 894                 ret = -ENOMEM;
 895                 goto accept_err;
 896         }
 897
 898         mutex_lock_nested(&newcon->sock_mutex, 1);
 899
 900         if (newcon->sock) {
 901                 struct connection *othercon = newcon->othercon;
 902
 903                 if (!othercon) {
 904                         othercon = kmem_cache_zalloc(con_cache, GFP_NOFS);
 905                         if (!othercon) {
 906                                 log_print("failed to allocate incoming socket");
 907                                 mutex_unlock(&newcon->sock_mutex);
 908                                 ret = -ENOMEM;
 909                                 goto accept_err;
 910                         }
 911                         othercon->nodeid = nodeid;
 912                         othercon->rx_action = receive_from_sock;
 913                         mutex_init(&othercon->sock_mutex);
 914                         INIT_WORK(&othercon->swork, process_send_sockets);
 915                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 916                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 917                 }
 918                 if (!othercon->sock) {
 919                         newcon->othercon = othercon;
 920                         othercon->sock = newsock;
 921                         newsock->sk->sk_user_data = othercon;
 922                         add_sock(newsock, othercon, false);
 923                         addcon = othercon;
 924                 } else {
 925                         printk("Extra connection from node %d attempted\n", nodeid);
 926                         ret = -EAGAIN;
 927                         mutex_unlock(&newcon->sock_mutex);
 928                         goto accept_err;
 929                 }
 930         } else {
 931                 newsock->sk->sk_user_data = newcon;
 932                 newcon->rx_action = receive_from_sock;
 933                 add_sock(newsock, newcon, false);
 934                 addcon = newcon;
 935         }
 936
 937         log_print("connected to %d", nodeid);
 938
 939         mutex_unlock(&newcon->sock_mutex);
 940
 941         /*
 942          * Add it to the active queue in case we got data
 943          * between processing the accept adding the socket
 944          * to the read_sockets list
 945          */
 946         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 947                 queue_work(recv_workqueue, &addcon->rwork);
 948         mutex_unlock(&con->sock_mutex);
 949
 950         return 0;
 951
 952 accept_err:
 953         mutex_unlock(&con->sock_mutex);
 954         if (newsock)
 955                 sock_release(newsock);
 956         if (ret != -EAGAIN)
 957                 log_print("error accepting connection from node: %d", ret);
 958
 959         return ret;
 960 }
 961
 962 static void free_entry(struct writequeue_entry *e)
 963 {
 964         __free_page(e->page);
 965         kfree(e);
 966 }
 967
 968 /*
 969  * writequeue_entry_complete - try to delete and free write queue entry
 970  * @e: write queue entry to try to delete
 971  * @completed: bytes completed
 972  *
 973  * writequeue_lock must be held.
 974  */
 975 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
 976 {
 977         e->offset += completed;
 978         e->len -= completed;
 979
 980         if (e->len == 0 && e->users == 0) {
 981                 list_del(&e->list);
 982                 free_entry(e);
 983         }
 984 }
 985
 986 /*
 987  * sctp_bind_addrs - bind a SCTP socket to all our addresses
 988  */
 989 static int sctp_bind_addrs(struct connection *con, uint16_t port)
 990 {
 991         struct sockaddr_storage localaddr;
 992         int i, addr_len, result = 0;
 993
 994         for (i = 0; i < dlm_local_count; i++) {
 995                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
 996                 make_sockaddr(&localaddr, port, &addr_len);
 997
 998                 if (!i)
 999                         result = kernel_bind(con->sock,
1000                                              (struct sockaddr *)&localaddr,
1001                                              addr_len);
1002                 else
1003                         result = kernel_setsockopt(con->sock, SOL_SCTP,
1004                                                    SCTP_SOCKOPT_BINDX_ADD,
1005                                                    (char *)&localaddr, addr_len);
1006
1007                 if (result < 0) {
1008                         log_print("Can't bind to %d addr number %d, %d.\n",
1009                                   port, i + 1, result);
1010                         break;
1011                 }
1012         }
1013         return result;
1014 }
1015
1016 /* Initiate an SCTP association.
1017    This is a special case of send_to_sock() in that we don't yet have a
1018    peeled-off socket for this association, so we use the listening socket
1019    and add the primary IP address of the remote node.
1020  */
1021 static void sctp_connect_to_sock(struct connection *con)
1022 {
1023         struct sockaddr_storage daddr;
1024         int one = 1;
1025         int result;
1026         int addr_len;
1027         struct socket *sock;
1028
1029         if (con->nodeid == 0) {
1030                 log_print("attempt to connect sock 0 foiled");
1031                 return;
1032         }
1033
1034         mutex_lock(&con->sock_mutex);
1035
1036         /* Some odd races can cause double-connects, ignore them */
1037         if (con->retries++ > MAX_CONNECT_RETRIES)
1038                 goto out;
1039
1040         if (con->sock) {
1041                 log_print("node %d already connected.", con->nodeid);
1042                 goto out;
1043         }
1044
1045         memset(&daddr, 0, sizeof(daddr));
1046         result = nodeid_to_addr(con->nodeid, &daddr, NULL, true);
1047         if (result < 0) {
1048                 log_print("no address for nodeid %d", con->nodeid);
1049                 goto out;
1050         }
1051
1052         /* Create a socket to communicate with */
1053         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1054                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1055         if (result < 0)
1056                 goto socket_err;
1057
1058         sock->sk->sk_user_data = con;
1059         con->rx_action = receive_from_sock;
1060         con->connect_action = sctp_connect_to_sock;
1061         add_sock(sock, con, true);
1062
1063         /* Bind to all addresses. */
1064         if (sctp_bind_addrs(con, 0))
1065                 goto bind_err;
1066
1067         make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len);
1068
1069         log_print("connecting to %d", con->nodeid);
1070
1071         /* Turn off Nagle's algorithm */
1072         kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
1073                           sizeof(one));
1074
1075         result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len,
1076                                    O_NONBLOCK);
1077         if (result == -EINPROGRESS)
1078                 result = 0;
1079         if (result == 0)
1080                 goto out;
1081
1082
1083 bind_err:
1084         con->sock = NULL;
1085         sock_release(sock);
1086
1087 socket_err:
1088         /*
1089          * Some errors are fatal and this list might need adjusting. For other
1090          * errors we try again until the max number of retries is reached.
1091          */
1092         if (result != -EHOSTUNREACH &&
1093             result != -ENETUNREACH &&
1094             result != -ENETDOWN &&
1095             result != -EINVAL &&
1096             result != -EPROTONOSUPPORT) {
1097                 log_print("connect %d try %d error %d", con->nodeid,
1098                           con->retries, result);
1099                 mutex_unlock(&con->sock_mutex);
1100                 msleep(1000);
1101                 clear_bit(CF_CONNECT_PENDING, &con->flags);
1102                 lowcomms_connect_sock(con);
1103                 return;
1104         }
1105
1106 out:
1107         mutex_unlock(&con->sock_mutex);
1108         set_bit(CF_WRITE_PENDING, &con->flags);
1109 }
1110
1111 /* Connect a new socket to its peer */
1112 static void tcp_connect_to_sock(struct connection *con)
1113 {
1114         struct sockaddr_storage saddr, src_addr;
1115         int addr_len;
1116         struct socket *sock = NULL;
1117         int one = 1;
1118         int result;
1119
1120         if (con->nodeid == 0) {
1121                 log_print("attempt to connect sock 0 foiled");
1122                 return;
1123         }
1124
1125         mutex_lock(&con->sock_mutex);
1126         if (con->retries++ > MAX_CONNECT_RETRIES)
1127                 goto out;
1128
1129         /* Some odd races can cause double-connects, ignore them */
1130         if (con->sock)
1131                 goto out;
1132
1133         /* Create a socket to communicate with */
1134         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1135                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1136         if (result < 0)
1137                 goto out_err;
1138
1139         memset(&saddr, 0, sizeof(saddr));
1140         result = nodeid_to_addr(con->nodeid, &saddr, NULL, false);
1141         if (result < 0) {
1142                 log_print("no address for nodeid %d", con->nodeid);
1143                 goto out_err;
1144         }
1145
1146         sock->sk->sk_user_data = con;
1147         con->rx_action = receive_from_sock;
1148         con->connect_action = tcp_connect_to_sock;
1149         add_sock(sock, con, true);
1150
1151         /* Bind to our cluster-known address connecting to avoid
1152            routing problems */
1153         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1154         make_sockaddr(&src_addr, 0, &addr_len);
1155         result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1156                                  addr_len);
1157         if (result < 0) {
1158                 log_print("could not bind for connect: %d", result);
1159                 /* This *may* not indicate a critical error */
1160         }
1161
1162         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1163
1164         log_print("connecting to %d", con->nodeid);
1165
1166         /* Turn off Nagle's algorithm */
1167         kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
1168                           sizeof(one));
1169
1170         result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1171                                    O_NONBLOCK);
1172         if (result == -EINPROGRESS)
1173                 result = 0;
1174         if (result == 0)
1175                 goto out;
1176
1177 out_err:
1178         if (con->sock) {
1179                 sock_release(con->sock);
1180                 con->sock = NULL;
1181         } else if (sock) {
1182                 sock_release(sock);
1183         }
1184         /*
1185          * Some errors are fatal and this list might need adjusting. For other
1186          * errors we try again until the max number of retries is reached.
1187          */
1188         if (result != -EHOSTUNREACH &&
1189             result != -ENETUNREACH &&
1190             result != -ENETDOWN &&
1191             result != -EINVAL &&
1192             result != -EPROTONOSUPPORT) {
1193                 log_print("connect %d try %d error %d", con->nodeid,
1194                           con->retries, result);
1195                 mutex_unlock(&con->sock_mutex);
1196                 msleep(1000);
1197                 clear_bit(CF_CONNECT_PENDING, &con->flags);
1198                 lowcomms_connect_sock(con);
1199                 return;
1200         }
1201 out:
1202         mutex_unlock(&con->sock_mutex);
1203         set_bit(CF_WRITE_PENDING, &con->flags);
1204         return;
1205 }
1206
1207 static struct socket *tcp_create_listen_sock(struct connection *con,
1208                                              struct sockaddr_storage *saddr)
1209 {
1210         struct socket *sock = NULL;
1211         int result = 0;
1212         int one = 1;
1213         int addr_len;
1214
1215         if (dlm_local_addr[0]->ss_family == AF_INET)
1216                 addr_len = sizeof(struct sockaddr_in);
1217         else
1218                 addr_len = sizeof(struct sockaddr_in6);
1219
1220         /* Create a socket to communicate with */
1221         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1222                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1223         if (result < 0) {
1224                 log_print("Can't create listening comms socket");
1225                 goto create_out;
1226         }
1227
1228         /* Turn off Nagle's algorithm */
1229         kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
1230                           sizeof(one));
1231
1232         result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
1233                                    (char *)&one, sizeof(one));
1234
1235         if (result < 0) {
1236                 log_print("Failed to set SO_REUSEADDR on socket: %d", result);
1237         }
1238         sock->sk->sk_user_data = con;
1239
1240         con->rx_action = tcp_accept_from_sock;
1241         con->connect_action = tcp_connect_to_sock;
1242
1243         /* Bind to our port */
1244         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1245         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1246         if (result < 0) {
1247                 log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1248                 sock_release(sock);
1249                 sock = NULL;
1250                 con->sock = NULL;
1251                 goto create_out;
1252         }
1253         result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
1254                                  (char *)&one, sizeof(one));
1255         if (result < 0) {
1256                 log_print("Set keepalive failed: %d", result);
1257         }
1258
1259         result = sock->ops->listen(sock, 5);
1260         if (result < 0) {
1261                 log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1262                 sock_release(sock);
1263                 sock = NULL;
1264                 goto create_out;
1265         }
1266
1267 create_out:
1268         return sock;
1269 }
1270
1271 /* Get local addresses */
1272 static void init_local(void)
1273 {
1274         struct sockaddr_storage sas, *addr;
1275         int i;
1276
1277         dlm_local_count = 0;
1278         for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1279                 if (dlm_our_addr(&sas, i))
1280                         break;
1281
1282                 addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1283                 if (!addr)
1284                         break;
1285                 dlm_local_addr[dlm_local_count++] = addr;
1286         }
1287 }
1288
1289 /* Initialise SCTP socket and bind to all interfaces */
1290 static int sctp_listen_for_all(void)
1291 {
1292         struct socket *sock = NULL;
1293         int result = -EINVAL;
1294         struct connection *con = nodeid2con(0, GFP_NOFS);
1295         int bufsize = NEEDED_RMEM;
1296         int one = 1;
1297
1298         if (!con)
1299                 return -ENOMEM;
1300
1301         log_print("Using SCTP for communications");
1302
1303         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1304                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1305         if (result < 0) {
1306                 log_print("Can't create comms socket, check SCTP is loaded");
1307                 goto out;
1308         }
1309
1310         result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE,
1311                                  (char *)&bufsize, sizeof(bufsize));
1312         if (result)
1313                 log_print("Error increasing buffer space on socket %d", result);
1314
1315         result = kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one,
1316                                    sizeof(one));
1317         if (result < 0)
1318                 log_print("Could not set SCTP NODELAY error %d\n", result);
1319
1320         write_lock_bh(&sock->sk->sk_callback_lock);
1321         /* Init con struct */
1322         sock->sk->sk_user_data = con;
1323         con->sock = sock;
1324         con->sock->sk->sk_data_ready = lowcomms_data_ready;
1325         con->rx_action = sctp_accept_from_sock;
1326         con->connect_action = sctp_connect_to_sock;
1327
1328         write_unlock_bh(&sock->sk->sk_callback_lock);
1329
1330         /* Bind to all addresses. */
1331         if (sctp_bind_addrs(con, dlm_config.ci_tcp_port))
1332                 goto create_delsock;
1333
1334         result = sock->ops->listen(sock, 5);
1335         if (result < 0) {
1336                 log_print("Can't set socket listening");
1337                 goto create_delsock;
1338         }
1339
1340         return 0;
1341
1342 create_delsock:
1343         sock_release(sock);
1344         con->sock = NULL;
1345 out:
1346         return result;
1347 }
1348
1349 static int tcp_listen_for_all(void)
1350 {
1351         struct socket *sock = NULL;
1352         struct connection *con = nodeid2con(0, GFP_NOFS);
1353         int result = -EINVAL;
1354
1355         if (!con)
1356                 return -ENOMEM;
1357
1358         /* We don't support multi-homed hosts */
1359         if (dlm_local_addr[1] != NULL) {
1360                 log_print("TCP protocol can't handle multi-homed hosts, "
1361                           "try SCTP");
1362                 return -EINVAL;
1363         }
1364
1365         log_print("Using TCP for communications");
1366
1367         sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
1368         if (sock) {
1369                 add_sock(sock, con, true);
1370                 result = 0;
1371         }
1372         else {
1373                 result = -EADDRINUSE;
1374         }
1375
1376         return result;
1377 }
1378
1379
1380
1381 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1382                                                      gfp_t allocation)
1383 {
1384         struct writequeue_entry *entry;
1385
1386         entry = kmalloc(sizeof(struct writequeue_entry), allocation);
1387         if (!entry)
1388                 return NULL;
1389
1390         entry->page = alloc_page(allocation);
1391         if (!entry->page) {
1392                 kfree(entry);
1393                 return NULL;
1394         }
1395
1396         entry->offset = 0;
1397         entry->len = 0;
1398         entry->end = 0;
1399         entry->users = 0;
1400         entry->con = con;
1401
1402         return entry;
1403 }
1404
1405 void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
1406 {
1407         struct connection *con;
1408         struct writequeue_entry *e;
1409         int offset = 0;
1410
1411         con = nodeid2con(nodeid, allocation);
1412         if (!con)
1413                 return NULL;
1414
1415         spin_lock(&con->writequeue_lock);
1416         e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
1417         if ((&e->list == &con->writequeue) ||
1418             (PAGE_SIZE - e->end < len)) {
1419                 e = NULL;
1420         } else {
1421                 offset = e->end;
1422                 e->end += len;
1423                 e->users++;
1424         }
1425         spin_unlock(&con->writequeue_lock);
1426
1427         if (e) {
1428         got_one:
1429                 *ppc = page_address(e->page) + offset;
1430                 return e;
1431         }
1432
1433         e = new_writequeue_entry(con, allocation);
1434         if (e) {
1435                 spin_lock(&con->writequeue_lock);
1436                 offset = e->end;
1437                 e->end += len;
1438                 e->users++;
1439                 list_add_tail(&e->list, &con->writequeue);
1440                 spin_unlock(&con->writequeue_lock);
1441                 goto got_one;
1442         }
1443         return NULL;
1444 }
1445
1446 void dlm_lowcomms_commit_buffer(void *mh)
1447 {
1448         struct writequeue_entry *e = (struct writequeue_entry *)mh;
1449         struct connection *con = e->con;
1450         int users;
1451
1452         spin_lock(&con->writequeue_lock);
1453         users = --e->users;
1454         if (users)
1455                 goto out;
1456         e->len = e->end - e->offset;
1457         spin_unlock(&con->writequeue_lock);
1458
1459         if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) {
1460                 queue_work(send_workqueue, &con->swork);
1461         }
1462         return;
1463
1464 out:
1465         spin_unlock(&con->writequeue_lock);
1466         return;
1467 }
1468
1469 /* Send a message */
1470 static void send_to_sock(struct connection *con)
1471 {
1472         int ret = 0;
1473         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1474         struct writequeue_entry *e;
1475         int len, offset;
1476         int count = 0;
1477
1478         mutex_lock(&con->sock_mutex);
1479         if (con->sock == NULL)
1480                 goto out_connect;
1481
1482         spin_lock(&con->writequeue_lock);
1483         for (;;) {
1484                 e = list_entry(con->writequeue.next, struct writequeue_entry,
1485                                list);
1486                 if ((struct list_head *) e == &con->writequeue)
1487                         break;
1488
1489                 len = e->len;
1490                 offset = e->offset;
1491                 BUG_ON(len == 0 && e->users == 0);
1492                 spin_unlock(&con->writequeue_lock);
1493
1494                 ret = 0;
1495                 if (len) {
1496                         ret = kernel_sendpage(con->sock, e->page, offset, len,
1497                                               msg_flags);
1498                         if (ret == -EAGAIN || ret == 0) {
1499                                 if (ret == -EAGAIN &&
1500                                     test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1501                                     !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1502                                         /* Notify TCP that we're limited by the
1503                                          * application window size.
1504                                          */
1505                                         set_bit(SOCK_NOSPACE, &con->sock->flags);
1506                                         con->sock->sk->sk_write_pending++;
1507                                 }
1508                                 cond_resched();
1509                                 goto out;
1510                         } else if (ret < 0)
1511                                 goto send_error;
1512                 }
1513
1514                 /* Don't starve people filling buffers */
1515                 if (++count >= MAX_SEND_MSG_COUNT) {
1516                         cond_resched();
1517                         count = 0;
1518                 }
1519
1520                 spin_lock(&con->writequeue_lock);
1521                 writequeue_entry_complete(e, ret);
1522         }
1523         spin_unlock(&con->writequeue_lock);
1524 out:
1525         mutex_unlock(&con->sock_mutex);
1526         return;
1527
1528 send_error:
1529         mutex_unlock(&con->sock_mutex);
1530         close_connection(con, false, false, true);
1531         lowcomms_connect_sock(con);
1532         return;
1533
1534 out_connect:
1535         mutex_unlock(&con->sock_mutex);
1536         lowcomms_connect_sock(con);
1537 }
1538
1539 static void clean_one_writequeue(struct connection *con)
1540 {
1541         struct writequeue_entry *e, *safe;
1542
1543         spin_lock(&con->writequeue_lock);
1544         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1545                 list_del(&e->list);
1546                 free_entry(e);
1547         }
1548         spin_unlock(&con->writequeue_lock);
1549 }
1550
1551 /* Called from recovery when it knows that a node has
1552    left the cluster */
1553 int dlm_lowcomms_close(int nodeid)
1554 {
1555         struct connection *con;
1556         struct dlm_node_addr *na;
1557
1558         log_print("closing connection to node %d", nodeid);
1559         con = nodeid2con(nodeid, 0);
1560         if (con) {
1561                 set_bit(CF_CLOSE, &con->flags);
1562                 close_connection(con, true, true, true);
1563                 clean_one_writequeue(con);
1564         }
1565
1566         spin_lock(&dlm_node_addrs_spin);
1567         na = find_node_addr(nodeid);
1568         if (na) {
1569                 list_del(&na->list);
1570                 while (na->addr_count--)
1571                         kfree(na->addr[na->addr_count]);
1572                 kfree(na);
1573         }
1574         spin_unlock(&dlm_node_addrs_spin);
1575
1576         return 0;
1577 }
1578
1579 /* Receive workqueue function */
1580 static void process_recv_sockets(struct work_struct *work)
1581 {
1582         struct connection *con = container_of(work, struct connection, rwork);
1583         int err;
1584
1585         clear_bit(CF_READ_PENDING, &con->flags);
1586         do {
1587                 err = con->rx_action(con);
1588         } while (!err);
1589 }
1590
1591 /* Send workqueue function */
1592 static void process_send_sockets(struct work_struct *work)
1593 {
1594         struct connection *con = container_of(work, struct connection, swork);
1595
1596         if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags))
1597                 con->connect_action(con);
1598         if (test_and_clear_bit(CF_WRITE_PENDING, &con->flags))
1599                 send_to_sock(con);
1600 }
1601
1602
1603 /* Discard all entries on the write queues */
1604 static void clean_writequeues(void)
1605 {
1606         foreach_conn(clean_one_writequeue);
1607 }
1608
1609 static void work_stop(void)
1610 {
1611         destroy_workqueue(recv_workqueue);
1612         destroy_workqueue(send_workqueue);
1613 }
1614
1615 static int work_start(void)
1616 {
1617         recv_workqueue = alloc_workqueue("dlm_recv",
1618                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1619         if (!recv_workqueue) {
1620                 log_print("can't start dlm_recv");
1621                 return -ENOMEM;
1622         }
1623
1624         send_workqueue = alloc_workqueue("dlm_send",
1625                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1626         if (!send_workqueue) {
1627                 log_print("can't start dlm_send");
1628                 destroy_workqueue(recv_workqueue);
1629                 return -ENOMEM;
1630         }
1631
1632         return 0;
1633 }
1634
1635 static void stop_conn(struct connection *con)
1636 {
1637         con->flags |= 0x0F;
1638         if (con->sock && con->sock->sk)
1639                 con->sock->sk->sk_user_data = NULL;
1640 }
1641
1642 static void free_conn(struct connection *con)
1643 {
1644         close_connection(con, true, true, true);
1645         if (con->othercon)
1646                 kmem_cache_free(con_cache, con->othercon);
1647         hlist_del(&con->list);
1648         kmem_cache_free(con_cache, con);
1649 }
1650
1651 void dlm_lowcomms_stop(void)
1652 {
1653         /* Set all the flags to prevent any
1654            socket activity.
1655         */
1656         mutex_lock(&connections_lock);
1657         dlm_allow_conn = 0;
1658         foreach_conn(stop_conn);
1659         clean_writequeues();
1660         foreach_conn(free_conn);
1661         mutex_unlock(&connections_lock);
1662
1663         work_stop();
1664
1665         kmem_cache_destroy(con_cache);
1666 }
1667
1668 int dlm_lowcomms_start(void)
1669 {
1670         int error = -EINVAL;
1671         struct connection *con;
1672         int i;
1673
1674         for (i = 0; i < CONN_HASH_SIZE; i++)
1675                 INIT_HLIST_HEAD(&connection_hash[i]);
1676
1677         init_local();
1678         if (!dlm_local_count) {
1679                 error = -ENOTCONN;
1680                 log_print("no local IP address has been set");
1681                 goto fail;
1682         }
1683
1684         error = -ENOMEM;
1685         con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
1686                                       __alignof__(struct connection), 0,
1687                                       NULL);
1688         if (!con_cache)
1689                 goto fail;
1690
1691         error = work_start();
1692         if (error)
1693                 goto fail_destroy;
1694
1695         dlm_allow_conn = 1;
1696
1697         /* Start listening */
1698         if (dlm_config.ci_protocol == 0)
1699                 error = tcp_listen_for_all();
1700         else
1701                 error = sctp_listen_for_all();
1702         if (error)
1703                 goto fail_unlisten;
1704
1705         return 0;
1706
1707 fail_unlisten:
1708         dlm_allow_conn = 0;
1709         con = nodeid2con(0,0);
1710         if (con) {
1711                 close_connection(con, false, true, true);
1712                 kmem_cache_free(con_cache, con);
1713         }
1714 fail_destroy:
1715         kmem_cache_destroy(con_cache);
1716 fail:
1717         return error;
1718 }
1719
1720 void dlm_lowcomms_exit(void)
1721 {
1722         struct dlm_node_addr *na, *safe;
1723
1724         spin_lock(&dlm_node_addrs_spin);
1725         list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
1726                 list_del(&na->list);
1727                 while (na->addr_count--)
1728                         kfree(na->addr[na->addr_count]);
1729                 kfree(na);
1730         }
1731         spin_unlock(&dlm_node_addrs_spin);
1732 }