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_INIT_PENDING 4
 110 #define CF_IS_OTHERCON 5
 111 #define CF_CLOSE 6
 112 #define CF_APP_LIMITED 7
 113         struct list_head writequeue;  /* List of outgoing writequeue_entries */
 114         spinlock_t writequeue_lock;
 115         int (*rx_action) (struct connection *); /* What to do when active */
 116         void (*connect_action) (struct connection *);   /* What to do to connect */
 117         struct page *rx_page;
 118         struct cbuf cb;
 119         int retries;
 120 #define MAX_CONNECT_RETRIES 3
 121         struct hlist_node list;
 122         struct connection *othercon;
 123         struct work_struct rwork; /* Receive workqueue */
 124         struct work_struct swork; /* Send workqueue */
 125 };
 126 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 127
 128 /* An entry waiting to be sent */
 129 struct writequeue_entry {
 130         struct list_head list;
 131         struct page *page;
 132         int offset;
 133         int len;
 134         int end;
 135         int users;
 136         struct connection *con;
 137 };
 138
 139 struct dlm_node_addr {
 140         struct list_head list;
 141         int nodeid;
 142         int addr_count;
 143         int curr_addr_index;
 144         struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
 145 };
 146
 147 static struct listen_sock_callbacks {
 148         void (*sk_error_report)(struct sock *);
 149         void (*sk_data_ready)(struct sock *);
 150         void (*sk_state_change)(struct sock *);
 151         void (*sk_write_space)(struct sock *);
 152 } listen_sock;
 153
 154 static LIST_HEAD(dlm_node_addrs);
 155 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
 156
 157 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 158 static int dlm_local_count;
 159 static int dlm_allow_conn;
 160
 161 /* Work queues */
 162 static struct workqueue_struct *recv_workqueue;
 163 static struct workqueue_struct *send_workqueue;
 164
 165 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 166 static DEFINE_MUTEX(connections_lock);
 167 static struct kmem_cache *con_cache;
 168
 169 static void process_recv_sockets(struct work_struct *work);
 170 static void process_send_sockets(struct work_struct *work);
 171
 172
 173 /* This is deliberately very simple because most clusters have simple
 174    sequential nodeids, so we should be able to go straight to a connection
 175    struct in the array */
 176 static inline int nodeid_hash(int nodeid)
 177 {
 178         return nodeid & (CONN_HASH_SIZE-1);
 179 }
 180
 181 static struct connection *__find_con(int nodeid)
 182 {
 183         int r;
 184         struct connection *con;
 185
 186         r = nodeid_hash(nodeid);
 187
 188         hlist_for_each_entry(con, &connection_hash[r], list) {
 189                 if (con->nodeid == nodeid)
 190                         return con;
 191         }
 192         return NULL;
 193 }
 194
 195 /*
 196  * If 'allocation' is zero then we don't attempt to create a new
 197  * connection structure for this node.
 198  */
 199 static struct connection *__nodeid2con(int nodeid, gfp_t alloc)
 200 {
 201         struct connection *con = NULL;
 202         int r;
 203
 204         con = __find_con(nodeid);
 205         if (con || !alloc)
 206                 return con;
 207
 208         con = kmem_cache_zalloc(con_cache, alloc);
 209         if (!con)
 210                 return NULL;
 211
 212         r = nodeid_hash(nodeid);
 213         hlist_add_head(&con->list, &connection_hash[r]);
 214
 215         con->nodeid = nodeid;
 216         mutex_init(&con->sock_mutex);
 217         INIT_LIST_HEAD(&con->writequeue);
 218         spin_lock_init(&con->writequeue_lock);
 219         INIT_WORK(&con->swork, process_send_sockets);
 220         INIT_WORK(&con->rwork, process_recv_sockets);
 221
 222         /* Setup action pointers for child sockets */
 223         if (con->nodeid) {
 224                 struct connection *zerocon = __find_con(0);
 225
 226                 con->connect_action = zerocon->connect_action;
 227                 if (!con->rx_action)
 228                         con->rx_action = zerocon->rx_action;
 229         }
 230
 231         return con;
 232 }
 233
 234 /* Loop round all connections */
 235 static void foreach_conn(void (*conn_func)(struct connection *c))
 236 {
 237         int i;
 238         struct hlist_node *n;
 239         struct connection *con;
 240
 241         for (i = 0; i < CONN_HASH_SIZE; i++) {
 242                 hlist_for_each_entry_safe(con, n, &connection_hash[i], list)
 243                         conn_func(con);
 244         }
 245 }
 246
 247 static struct connection *nodeid2con(int nodeid, gfp_t allocation)
 248 {
 249         struct connection *con;
 250
 251         mutex_lock(&connections_lock);
 252         con = __nodeid2con(nodeid, allocation);
 253         mutex_unlock(&connections_lock);
 254
 255         return con;
 256 }
 257
 258 static struct dlm_node_addr *find_node_addr(int nodeid)
 259 {
 260         struct dlm_node_addr *na;
 261
 262         list_for_each_entry(na, &dlm_node_addrs, list) {
 263                 if (na->nodeid == nodeid)
 264                         return na;
 265         }
 266         return NULL;
 267 }
 268
 269 static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y)
 270 {
 271         switch (x->ss_family) {
 272         case AF_INET: {
 273                 struct sockaddr_in *sinx = (struct sockaddr_in *)x;
 274                 struct sockaddr_in *siny = (struct sockaddr_in *)y;
 275                 if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
 276                         return 0;
 277                 if (sinx->sin_port != siny->sin_port)
 278                         return 0;
 279                 break;
 280         }
 281         case AF_INET6: {
 282                 struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
 283                 struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
 284                 if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
 285                         return 0;
 286                 if (sinx->sin6_port != siny->sin6_port)
 287                         return 0;
 288                 break;
 289         }
 290         default:
 291                 return 0;
 292         }
 293         return 1;
 294 }
 295
 296 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
 297                           struct sockaddr *sa_out, bool try_new_addr)
 298 {
 299         struct sockaddr_storage sas;
 300         struct dlm_node_addr *na;
 301
 302         if (!dlm_local_count)
 303                 return -1;
 304
 305         spin_lock(&dlm_node_addrs_spin);
 306         na = find_node_addr(nodeid);
 307         if (na && na->addr_count) {
 308                 memcpy(&sas, na->addr[na->curr_addr_index],
 309                        sizeof(struct sockaddr_storage));
 310
 311                 if (try_new_addr) {
 312                         na->curr_addr_index++;
 313                         if (na->curr_addr_index == na->addr_count)
 314                                 na->curr_addr_index = 0;
 315                 }
 316         }
 317         spin_unlock(&dlm_node_addrs_spin);
 318
 319         if (!na)
 320                 return -EEXIST;
 321
 322         if (!na->addr_count)
 323                 return -ENOENT;
 324
 325         if (sas_out)
 326                 memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
 327
 328         if (!sa_out)
 329                 return 0;
 330
 331         if (dlm_local_addr[0]->ss_family == AF_INET) {
 332                 struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
 333                 struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
 334                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 335         } else {
 336                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
 337                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
 338                 ret6->sin6_addr = in6->sin6_addr;
 339         }
 340
 341         return 0;
 342 }
 343
 344 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid)
 345 {
 346         struct dlm_node_addr *na;
 347         int rv = -EEXIST;
 348         int addr_i;
 349
 350         spin_lock(&dlm_node_addrs_spin);
 351         list_for_each_entry(na, &dlm_node_addrs, list) {
 352                 if (!na->addr_count)
 353                         continue;
 354
 355                 for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
 356                         if (addr_compare(na->addr[addr_i], addr)) {
 357                                 *nodeid = na->nodeid;
 358                                 rv = 0;
 359                                 goto unlock;
 360                         }
 361                 }
 362         }
 363 unlock:
 364         spin_unlock(&dlm_node_addrs_spin);
 365         return rv;
 366 }
 367
 368 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 369 {
 370         struct sockaddr_storage *new_addr;
 371         struct dlm_node_addr *new_node, *na;
 372
 373         new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
 374         if (!new_node)
 375                 return -ENOMEM;
 376
 377         new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
 378         if (!new_addr) {
 379                 kfree(new_node);
 380                 return -ENOMEM;
 381         }
 382
 383         memcpy(new_addr, addr, len);
 384
 385         spin_lock(&dlm_node_addrs_spin);
 386         na = find_node_addr(nodeid);
 387         if (!na) {
 388                 new_node->nodeid = nodeid;
 389                 new_node->addr[0] = new_addr;
 390                 new_node->addr_count = 1;
 391                 list_add(&new_node->list, &dlm_node_addrs);
 392                 spin_unlock(&dlm_node_addrs_spin);
 393                 return 0;
 394         }
 395
 396         if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
 397                 spin_unlock(&dlm_node_addrs_spin);
 398                 kfree(new_addr);
 399                 kfree(new_node);
 400                 return -ENOSPC;
 401         }
 402
 403         na->addr[na->addr_count++] = new_addr;
 404         spin_unlock(&dlm_node_addrs_spin);
 405         kfree(new_node);
 406         return 0;
 407 }
 408
 409 /* Data available on socket or listen socket received a connect */
 410 static void lowcomms_data_ready(struct sock *sk)
 411 {
 412         struct connection *con = sock2con(sk);
 413         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 414                 queue_work(recv_workqueue, &con->rwork);
 415 }
 416
 417 static void lowcomms_write_space(struct sock *sk)
 418 {
 419         struct connection *con = sock2con(sk);
 420
 421         if (!con)
 422                 return;
 423
 424         clear_bit(SOCK_NOSPACE, &con->sock->flags);
 425
 426         if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
 427                 con->sock->sk->sk_write_pending--;
 428                 clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
 429         }
 430
 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         queue_work(send_workqueue, &con->swork);
 439         cond_resched();
 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 = listen_sock.sk_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_listen_callbacks(struct socket *sock)
 521 {
 522         struct sock *sk = sock->sk;
 523
 524         listen_sock.sk_data_ready = sk->sk_data_ready;
 525         listen_sock.sk_state_change = sk->sk_state_change;
 526         listen_sock.sk_write_space = sk->sk_write_space;
 527         listen_sock.sk_error_report = sk->sk_error_report;
 528 }
 529
 530 static void restore_callbacks(struct socket *sock)
 531 {
 532         struct sock *sk = sock->sk;
 533
 534         write_lock_bh(&sk->sk_callback_lock);
 535         sk->sk_user_data = NULL;
 536         sk->sk_data_ready = listen_sock.sk_data_ready;
 537         sk->sk_state_change = listen_sock.sk_state_change;
 538         sk->sk_write_space = listen_sock.sk_write_space;
 539         sk->sk_error_report = listen_sock.sk_error_report;
 540         write_unlock_bh(&sk->sk_callback_lock);
 541 }
 542
 543 /* Make a socket active */
 544 static void add_sock(struct socket *sock, struct connection *con)
 545 {
 546         struct sock *sk = sock->sk;
 547
 548         write_lock_bh(&sk->sk_callback_lock);
 549         con->sock = sock;
 550
 551         sk->sk_user_data = con;
 552         /* Install a data_ready callback */
 553         sk->sk_data_ready = lowcomms_data_ready;
 554         sk->sk_write_space = lowcomms_write_space;
 555         sk->sk_state_change = lowcomms_state_change;
 556         sk->sk_allocation = GFP_NOFS;
 557         sk->sk_error_report = lowcomms_error_report;
 558         write_unlock_bh(&sk->sk_callback_lock);
 559 }
 560
 561 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 562    length */
 563 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 564                           int *addr_len)
 565 {
 566         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 567         if (saddr->ss_family == AF_INET) {
 568                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 569                 in4_addr->sin_port = cpu_to_be16(port);
 570                 *addr_len = sizeof(struct sockaddr_in);
 571                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 572         } else {
 573                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 574                 in6_addr->sin6_port = cpu_to_be16(port);
 575                 *addr_len = sizeof(struct sockaddr_in6);
 576         }
 577         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 578 }
 579
 580 /* Close a remote connection and tidy up */
 581 static void close_connection(struct connection *con, bool and_other,
 582                              bool tx, bool rx)
 583 {
 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                 restore_callbacks(con->sock);
 592                 sock_release(con->sock);
 593                 con->sock = NULL;
 594         }
 595         if (con->othercon && and_other) {
 596                 /* Will only re-enter once. */
 597                 close_connection(con->othercon, false, true, true);
 598         }
 599         if (con->rx_page) {
 600                 __free_page(con->rx_page);
 601                 con->rx_page = NULL;
 602         }
 603
 604         con->retries = 0;
 605         mutex_unlock(&con->sock_mutex);
 606 }
 607
 608 /* Data received from remote end */
 609 static int receive_from_sock(struct connection *con)
 610 {
 611         int ret = 0;
 612         struct msghdr msg = {};
 613         struct kvec iov[2];
 614         unsigned len;
 615         int r;
 616         int call_again_soon = 0;
 617         int nvec;
 618
 619         mutex_lock(&con->sock_mutex);
 620
 621         if (con->sock == NULL) {
 622                 ret = -EAGAIN;
 623                 goto out_close;
 624         }
 625         if (con->nodeid == 0) {
 626                 ret = -EINVAL;
 627                 goto out_close;
 628         }
 629
 630         if (con->rx_page == NULL) {
 631                 /*
 632                  * This doesn't need to be atomic, but I think it should
 633                  * improve performance if it is.
 634                  */
 635                 con->rx_page = alloc_page(GFP_ATOMIC);
 636                 if (con->rx_page == NULL)
 637                         goto out_resched;
 638                 cbuf_init(&con->cb, PAGE_SIZE);
 639         }
 640
 641         /*
 642          * iov[0] is the bit of the circular buffer between the current end
 643          * point (cb.base + cb.len) and the end of the buffer.
 644          */
 645         iov[0].iov_len = con->cb.base - cbuf_data(&con->cb);
 646         iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb);
 647         iov[1].iov_len = 0;
 648         nvec = 1;
 649
 650         /*
 651          * iov[1] is the bit of the circular buffer between the start of the
 652          * buffer and the start of the currently used section (cb.base)
 653          */
 654         if (cbuf_data(&con->cb) >= con->cb.base) {
 655                 iov[0].iov_len = PAGE_SIZE - cbuf_data(&con->cb);
 656                 iov[1].iov_len = con->cb.base;
 657                 iov[1].iov_base = page_address(con->rx_page);
 658                 nvec = 2;
 659         }
 660         len = iov[0].iov_len + iov[1].iov_len;
 661
 662         r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len,
 663                                MSG_DONTWAIT | MSG_NOSIGNAL);
 664         if (ret <= 0)
 665                 goto out_close;
 666         else if (ret == len)
 667                 call_again_soon = 1;
 668
 669         cbuf_add(&con->cb, ret);
 670         ret = dlm_process_incoming_buffer(con->nodeid,
 671                                           page_address(con->rx_page),
 672                                           con->cb.base, con->cb.len,
 673                                           PAGE_SIZE);
 674         if (ret == -EBADMSG) {
 675                 log_print("lowcomms: addr=%p, base=%u, len=%u, read=%d",
 676                           page_address(con->rx_page), con->cb.base,
 677                           con->cb.len, r);
 678         }
 679         if (ret < 0)
 680                 goto out_close;
 681         cbuf_eat(&con->cb, ret);
 682
 683         if (cbuf_empty(&con->cb) && !call_again_soon) {
 684                 __free_page(con->rx_page);
 685                 con->rx_page = NULL;
 686         }
 687
 688         if (call_again_soon)
 689                 goto out_resched;
 690         mutex_unlock(&con->sock_mutex);
 691         return 0;
 692
 693 out_resched:
 694         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 695                 queue_work(recv_workqueue, &con->rwork);
 696         mutex_unlock(&con->sock_mutex);
 697         return -EAGAIN;
 698
 699 out_close:
 700         mutex_unlock(&con->sock_mutex);
 701         if (ret != -EAGAIN) {
 702                 close_connection(con, false, true, false);
 703                 /* Reconnect when there is something to send */
 704         }
 705         /* Don't return success if we really got EOF */
 706         if (ret == 0)
 707                 ret = -EAGAIN;
 708
 709         return ret;
 710 }
 711
 712 /* Listening socket is busy, accept a connection */
 713 static int tcp_accept_from_sock(struct connection *con)
 714 {
 715         int result;
 716         struct sockaddr_storage peeraddr;
 717         struct socket *newsock;
 718         int len;
 719         int nodeid;
 720         struct connection *newcon;
 721         struct connection *addcon;
 722
 723         mutex_lock(&connections_lock);
 724         if (!dlm_allow_conn) {
 725                 mutex_unlock(&connections_lock);
 726                 return -1;
 727         }
 728         mutex_unlock(&connections_lock);
 729
 730         memset(&peeraddr, 0, sizeof(peeraddr));
 731         result = sock_create_lite(dlm_local_addr[0]->ss_family,
 732                                   SOCK_STREAM, IPPROTO_TCP, &newsock);
 733         if (result < 0)
 734                 return -ENOMEM;
 735
 736         mutex_lock_nested(&con->sock_mutex, 0);
 737
 738         result = -ENOTCONN;
 739         if (con->sock == NULL)
 740                 goto accept_err;
 741
 742         newsock->type = con->sock->type;
 743         newsock->ops = con->sock->ops;
 744
 745         result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK, true);
 746         if (result < 0)
 747                 goto accept_err;
 748
 749         /* Get the connected socket's peer */
 750         memset(&peeraddr, 0, sizeof(peeraddr));
 751         if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr,
 752                                   &len, 2)) {
 753                 result = -ECONNABORTED;
 754                 goto accept_err;
 755         }
 756
 757         /* Get the new node's NODEID */
 758         make_sockaddr(&peeraddr, 0, &len);
 759         if (addr_to_nodeid(&peeraddr, &nodeid)) {
 760                 unsigned char *b=(unsigned char *)&peeraddr;
 761                 log_print("connect from non cluster node");
 762                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
 763                                      b, sizeof(struct sockaddr_storage));
 764                 sock_release(newsock);
 765                 mutex_unlock(&con->sock_mutex);
 766                 return -1;
 767         }
 768
 769         log_print("got connection from %d", nodeid);
 770
 771         /*  Check to see if we already have a connection to this node. This
 772          *  could happen if the two nodes initiate a connection at roughly
 773          *  the same time and the connections cross on the wire.
 774          *  In this case we store the incoming one in "othercon"
 775          */
 776         newcon = nodeid2con(nodeid, GFP_NOFS);
 777         if (!newcon) {
 778                 result = -ENOMEM;
 779                 goto accept_err;
 780         }
 781         mutex_lock_nested(&newcon->sock_mutex, 1);
 782         if (newcon->sock) {
 783                 struct connection *othercon = newcon->othercon;
 784
 785                 if (!othercon) {
 786                         othercon = kmem_cache_zalloc(con_cache, GFP_NOFS);
 787                         if (!othercon) {
 788                                 log_print("failed to allocate incoming socket");
 789                                 mutex_unlock(&newcon->sock_mutex);
 790                                 result = -ENOMEM;
 791                                 goto accept_err;
 792                         }
 793                         othercon->nodeid = nodeid;
 794                         othercon->rx_action = receive_from_sock;
 795                         mutex_init(&othercon->sock_mutex);
 796                         INIT_WORK(&othercon->swork, process_send_sockets);
 797                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 798                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 799                 }
 800                 if (!othercon->sock) {
 801                         newcon->othercon = othercon;
 802                         othercon->sock = newsock;
 803                         newsock->sk->sk_user_data = othercon;
 804                         add_sock(newsock, othercon);
 805                         addcon = othercon;
 806                 }
 807                 else {
 808                         printk("Extra connection from node %d attempted\n", nodeid);
 809                         result = -EAGAIN;
 810                         mutex_unlock(&newcon->sock_mutex);
 811                         goto accept_err;
 812                 }
 813         }
 814         else {
 815                 newsock->sk->sk_user_data = newcon;
 816                 newcon->rx_action = receive_from_sock;
 817                 /* accept copies the sk after we've saved the callbacks, so we
 818                    don't want to save them a second time or comm errors will
 819                    result in calling sk_error_report recursively. */
 820                 add_sock(newsock, newcon);
 821                 addcon = newcon;
 822         }
 823
 824         mutex_unlock(&newcon->sock_mutex);
 825
 826         /*
 827          * Add it to the active queue in case we got data
 828          * between processing the accept adding the socket
 829          * to the read_sockets list
 830          */
 831         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 832                 queue_work(recv_workqueue, &addcon->rwork);
 833         mutex_unlock(&con->sock_mutex);
 834
 835         return 0;
 836
 837 accept_err:
 838         mutex_unlock(&con->sock_mutex);
 839         sock_release(newsock);
 840
 841         if (result != -EAGAIN)
 842                 log_print("error accepting connection from node: %d", result);
 843         return result;
 844 }
 845
 846 static int sctp_accept_from_sock(struct connection *con)
 847 {
 848         /* Check that the new node is in the lockspace */
 849         struct sctp_prim prim;
 850         int nodeid;
 851         int prim_len, ret;
 852         int addr_len;
 853         struct connection *newcon;
 854         struct connection *addcon;
 855         struct socket *newsock;
 856
 857         mutex_lock(&connections_lock);
 858         if (!dlm_allow_conn) {
 859                 mutex_unlock(&connections_lock);
 860                 return -1;
 861         }
 862         mutex_unlock(&connections_lock);
 863
 864         mutex_lock_nested(&con->sock_mutex, 0);
 865
 866         ret = kernel_accept(con->sock, &newsock, O_NONBLOCK);
 867         if (ret < 0)
 868                 goto accept_err;
 869
 870         memset(&prim, 0, sizeof(struct sctp_prim));
 871         prim_len = sizeof(struct sctp_prim);
 872
 873         ret = kernel_getsockopt(newsock, IPPROTO_SCTP, SCTP_PRIMARY_ADDR,
 874                                 (char *)&prim, &prim_len);
 875         if (ret < 0) {
 876                 log_print("getsockopt/sctp_primary_addr failed: %d", ret);
 877                 goto accept_err;
 878         }
 879
 880         make_sockaddr(&prim.ssp_addr, 0, &addr_len);
 881         ret = addr_to_nodeid(&prim.ssp_addr, &nodeid);
 882         if (ret) {
 883                 unsigned char *b = (unsigned char *)&prim.ssp_addr;
 884
 885                 log_print("reject connect from unknown addr");
 886                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
 887                                      b, sizeof(struct sockaddr_storage));
 888                 goto accept_err;
 889         }
 890
 891         newcon = nodeid2con(nodeid, GFP_NOFS);
 892         if (!newcon) {
 893                 ret = -ENOMEM;
 894                 goto accept_err;
 895         }
 896
 897         mutex_lock_nested(&newcon->sock_mutex, 1);
 898
 899         if (newcon->sock) {
 900                 struct connection *othercon = newcon->othercon;
 901
 902                 if (!othercon) {
 903                         othercon = kmem_cache_zalloc(con_cache, GFP_NOFS);
 904                         if (!othercon) {
 905                                 log_print("failed to allocate incoming socket");
 906                                 mutex_unlock(&newcon->sock_mutex);
 907                                 ret = -ENOMEM;
 908                                 goto accept_err;
 909                         }
 910                         othercon->nodeid = nodeid;
 911                         othercon->rx_action = receive_from_sock;
 912                         mutex_init(&othercon->sock_mutex);
 913                         INIT_WORK(&othercon->swork, process_send_sockets);
 914                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 915                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 916                 }
 917                 if (!othercon->sock) {
 918                         newcon->othercon = othercon;
 919                         othercon->sock = newsock;
 920                         newsock->sk->sk_user_data = othercon;
 921                         add_sock(newsock, othercon);
 922                         addcon = othercon;
 923                 } else {
 924                         printk("Extra connection from node %d attempted\n", nodeid);
 925                         ret = -EAGAIN;
 926                         mutex_unlock(&newcon->sock_mutex);
 927                         goto accept_err;
 928                 }
 929         } else {
 930                 newsock->sk->sk_user_data = newcon;
 931                 newcon->rx_action = receive_from_sock;
 932                 add_sock(newsock, newcon);
 933                 addcon = newcon;
 934         }
 935
 936         log_print("connected to %d", nodeid);
 937
 938         mutex_unlock(&newcon->sock_mutex);
 939
 940         /*
 941          * Add it to the active queue in case we got data
 942          * between processing the accept adding the socket
 943          * to the read_sockets list
 944          */
 945         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 946                 queue_work(recv_workqueue, &addcon->rwork);
 947         mutex_unlock(&con->sock_mutex);
 948
 949         return 0;
 950
 951 accept_err:
 952         mutex_unlock(&con->sock_mutex);
 953         if (newsock)
 954                 sock_release(newsock);
 955         if (ret != -EAGAIN)
 956                 log_print("error accepting connection from node: %d", ret);
 957
 958         return ret;
 959 }
 960
 961 static void free_entry(struct writequeue_entry *e)
 962 {
 963         __free_page(e->page);
 964         kfree(e);
 965 }
 966
 967 /*
 968  * writequeue_entry_complete - try to delete and free write queue entry
 969  * @e: write queue entry to try to delete
 970  * @completed: bytes completed
 971  *
 972  * writequeue_lock must be held.
 973  */
 974 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
 975 {
 976         e->offset += completed;
 977         e->len -= completed;
 978
 979         if (e->len == 0 && e->users == 0) {
 980                 list_del(&e->list);
 981                 free_entry(e);
 982         }
 983 }
 984
 985 /*
 986  * sctp_bind_addrs - bind a SCTP socket to all our addresses
 987  */
 988 static int sctp_bind_addrs(struct connection *con, uint16_t port)
 989 {
 990         struct sockaddr_storage localaddr;
 991         int i, addr_len, result = 0;
 992
 993         for (i = 0; i < dlm_local_count; i++) {
 994                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
 995                 make_sockaddr(&localaddr, port, &addr_len);
 996
 997                 if (!i)
 998                         result = kernel_bind(con->sock,
 999                                              (struct sockaddr *)&localaddr,
1000                                              addr_len);
1001                 else
1002                         result = kernel_setsockopt(con->sock, SOL_SCTP,
1003                                                    SCTP_SOCKOPT_BINDX_ADD,
1004                                                    (char *)&localaddr, addr_len);
1005
1006                 if (result < 0) {
1007                         log_print("Can't bind to %d addr number %d, %d.\n",
1008                                   port, i + 1, result);
1009                         break;
1010                 }
1011         }
1012         return result;
1013 }
1014
1015 /* Initiate an SCTP association.
1016    This is a special case of send_to_sock() in that we don't yet have a
1017    peeled-off socket for this association, so we use the listening socket
1018    and add the primary IP address of the remote node.
1019  */
1020 static void sctp_connect_to_sock(struct connection *con)
1021 {
1022         struct sockaddr_storage daddr;
1023         int one = 1;
1024         int result;
1025         int addr_len;
1026         struct socket *sock;
1027
1028         if (con->nodeid == 0) {
1029                 log_print("attempt to connect sock 0 foiled");
1030                 return;
1031         }
1032
1033         mutex_lock(&con->sock_mutex);
1034
1035         /* Some odd races can cause double-connects, ignore them */
1036         if (con->retries++ > MAX_CONNECT_RETRIES)
1037                 goto out;
1038
1039         if (con->sock) {
1040                 log_print("node %d already connected.", con->nodeid);
1041                 goto out;
1042         }
1043
1044         memset(&daddr, 0, sizeof(daddr));
1045         result = nodeid_to_addr(con->nodeid, &daddr, NULL, true);
1046         if (result < 0) {
1047                 log_print("no address for nodeid %d", con->nodeid);
1048                 goto out;
1049         }
1050
1051         /* Create a socket to communicate with */
1052         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1053                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1054         if (result < 0)
1055                 goto socket_err;
1056
1057         sock->sk->sk_user_data = con;
1058         con->rx_action = receive_from_sock;
1059         con->connect_action = sctp_connect_to_sock;
1060         add_sock(sock, con);
1061
1062         /* Bind to all addresses. */
1063         if (sctp_bind_addrs(con, 0))
1064                 goto bind_err;
1065
1066         make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len);
1067
1068         log_print("connecting to %d", con->nodeid);
1069
1070         /* Turn off Nagle's algorithm */
1071         kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
1072                           sizeof(one));
1073
1074         result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len,
1075                                    O_NONBLOCK);
1076         if (result == -EINPROGRESS)
1077                 result = 0;
1078         if (result == 0)
1079                 goto out;
1080
1081 bind_err:
1082         con->sock = NULL;
1083         sock_release(sock);
1084
1085 socket_err:
1086         /*
1087          * Some errors are fatal and this list might need adjusting. For other
1088          * errors we try again until the max number of retries is reached.
1089          */
1090         if (result != -EHOSTUNREACH &&
1091             result != -ENETUNREACH &&
1092             result != -ENETDOWN &&
1093             result != -EINVAL &&
1094             result != -EPROTONOSUPPORT) {
1095                 log_print("connect %d try %d error %d", con->nodeid,
1096                           con->retries, result);
1097                 mutex_unlock(&con->sock_mutex);
1098                 msleep(1000);
1099                 lowcomms_connect_sock(con);
1100                 return;
1101         }
1102
1103 out:
1104         mutex_unlock(&con->sock_mutex);
1105 }
1106
1107 /* Connect a new socket to its peer */
1108 static void tcp_connect_to_sock(struct connection *con)
1109 {
1110         struct sockaddr_storage saddr, src_addr;
1111         int addr_len;
1112         struct socket *sock = NULL;
1113         int one = 1;
1114         int result;
1115
1116         if (con->nodeid == 0) {
1117                 log_print("attempt to connect sock 0 foiled");
1118                 return;
1119         }
1120
1121         mutex_lock(&con->sock_mutex);
1122         if (con->retries++ > MAX_CONNECT_RETRIES)
1123                 goto out;
1124
1125         /* Some odd races can cause double-connects, ignore them */
1126         if (con->sock)
1127                 goto out;
1128
1129         /* Create a socket to communicate with */
1130         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1131                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1132         if (result < 0)
1133                 goto out_err;
1134
1135         memset(&saddr, 0, sizeof(saddr));
1136         result = nodeid_to_addr(con->nodeid, &saddr, NULL, false);
1137         if (result < 0) {
1138                 log_print("no address for nodeid %d", con->nodeid);
1139                 goto out_err;
1140         }
1141
1142         sock->sk->sk_user_data = con;
1143         con->rx_action = receive_from_sock;
1144         con->connect_action = tcp_connect_to_sock;
1145         add_sock(sock, con);
1146
1147         /* Bind to our cluster-known address connecting to avoid
1148            routing problems */
1149         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1150         make_sockaddr(&src_addr, 0, &addr_len);
1151         result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1152                                  addr_len);
1153         if (result < 0) {
1154                 log_print("could not bind for connect: %d", result);
1155                 /* This *may* not indicate a critical error */
1156         }
1157
1158         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1159
1160         log_print("connecting to %d", con->nodeid);
1161
1162         /* Turn off Nagle's algorithm */
1163         kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
1164                           sizeof(one));
1165
1166         result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1167                                    O_NONBLOCK);
1168         if (result == -EINPROGRESS)
1169                 result = 0;
1170         if (result == 0)
1171                 goto out;
1172
1173 out_err:
1174         if (con->sock) {
1175                 sock_release(con->sock);
1176                 con->sock = NULL;
1177         } else if (sock) {
1178                 sock_release(sock);
1179         }
1180         /*
1181          * Some errors are fatal and this list might need adjusting. For other
1182          * errors we try again until the max number of retries is reached.
1183          */
1184         if (result != -EHOSTUNREACH &&
1185             result != -ENETUNREACH &&
1186             result != -ENETDOWN &&
1187             result != -EINVAL &&
1188             result != -EPROTONOSUPPORT) {
1189                 log_print("connect %d try %d error %d", con->nodeid,
1190                           con->retries, result);
1191                 mutex_unlock(&con->sock_mutex);
1192                 msleep(1000);
1193                 lowcomms_connect_sock(con);
1194                 return;
1195         }
1196 out:
1197         mutex_unlock(&con->sock_mutex);
1198         return;
1199 }
1200
1201 static struct socket *tcp_create_listen_sock(struct connection *con,
1202                                              struct sockaddr_storage *saddr)
1203 {
1204         struct socket *sock = NULL;
1205         int result = 0;
1206         int one = 1;
1207         int addr_len;
1208
1209         if (dlm_local_addr[0]->ss_family == AF_INET)
1210                 addr_len = sizeof(struct sockaddr_in);
1211         else
1212                 addr_len = sizeof(struct sockaddr_in6);
1213
1214         /* Create a socket to communicate with */
1215         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1216                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1217         if (result < 0) {
1218                 log_print("Can't create listening comms socket");
1219                 goto create_out;
1220         }
1221
1222         /* Turn off Nagle's algorithm */
1223         kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
1224                           sizeof(one));
1225
1226         result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
1227                                    (char *)&one, sizeof(one));
1228
1229         if (result < 0) {
1230                 log_print("Failed to set SO_REUSEADDR on socket: %d", result);
1231         }
1232         sock->sk->sk_user_data = con;
1233         save_listen_callbacks(sock);
1234         con->rx_action = tcp_accept_from_sock;
1235         con->connect_action = tcp_connect_to_sock;
1236
1237         /* Bind to our port */
1238         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1239         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1240         if (result < 0) {
1241                 log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1242                 sock_release(sock);
1243                 sock = NULL;
1244                 con->sock = NULL;
1245                 goto create_out;
1246         }
1247         result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
1248                                  (char *)&one, sizeof(one));
1249         if (result < 0) {
1250                 log_print("Set keepalive failed: %d", result);
1251         }
1252
1253         result = sock->ops->listen(sock, 5);
1254         if (result < 0) {
1255                 log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1256                 sock_release(sock);
1257                 sock = NULL;
1258                 goto create_out;
1259         }
1260
1261 create_out:
1262         return sock;
1263 }
1264
1265 /* Get local addresses */
1266 static void init_local(void)
1267 {
1268         struct sockaddr_storage sas, *addr;
1269         int i;
1270
1271         dlm_local_count = 0;
1272         for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1273                 if (dlm_our_addr(&sas, i))
1274                         break;
1275
1276                 addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1277                 if (!addr)
1278                         break;
1279                 dlm_local_addr[dlm_local_count++] = addr;
1280         }
1281 }
1282
1283 /* Initialise SCTP socket and bind to all interfaces */
1284 static int sctp_listen_for_all(void)
1285 {
1286         struct socket *sock = NULL;
1287         int result = -EINVAL;
1288         struct connection *con = nodeid2con(0, GFP_NOFS);
1289         int bufsize = NEEDED_RMEM;
1290         int one = 1;
1291
1292         if (!con)
1293                 return -ENOMEM;
1294
1295         log_print("Using SCTP for communications");
1296
1297         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1298                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1299         if (result < 0) {
1300                 log_print("Can't create comms socket, check SCTP is loaded");
1301                 goto out;
1302         }
1303
1304         result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE,
1305                                  (char *)&bufsize, sizeof(bufsize));
1306         if (result)
1307                 log_print("Error increasing buffer space on socket %d", result);
1308
1309         result = kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one,
1310                                    sizeof(one));
1311         if (result < 0)
1312                 log_print("Could not set SCTP NODELAY error %d\n", result);
1313
1314         write_lock_bh(&sock->sk->sk_callback_lock);
1315         /* Init con struct */
1316         sock->sk->sk_user_data = con;
1317         save_listen_callbacks(sock);
1318         con->sock = sock;
1319         con->sock->sk->sk_data_ready = lowcomms_data_ready;
1320         con->rx_action = sctp_accept_from_sock;
1321         con->connect_action = sctp_connect_to_sock;
1322
1323         write_unlock_bh(&sock->sk->sk_callback_lock);
1324
1325         /* Bind to all addresses. */
1326         if (sctp_bind_addrs(con, dlm_config.ci_tcp_port))
1327                 goto create_delsock;
1328
1329         result = sock->ops->listen(sock, 5);
1330         if (result < 0) {
1331                 log_print("Can't set socket listening");
1332                 goto create_delsock;
1333         }
1334
1335         return 0;
1336
1337 create_delsock:
1338         sock_release(sock);
1339         con->sock = NULL;
1340 out:
1341         return result;
1342 }
1343
1344 static int tcp_listen_for_all(void)
1345 {
1346         struct socket *sock = NULL;
1347         struct connection *con = nodeid2con(0, GFP_NOFS);
1348         int result = -EINVAL;
1349
1350         if (!con)
1351                 return -ENOMEM;
1352
1353         /* We don't support multi-homed hosts */
1354         if (dlm_local_addr[1] != NULL) {
1355                 log_print("TCP protocol can't handle multi-homed hosts, "
1356                           "try SCTP");
1357                 return -EINVAL;
1358         }
1359
1360         log_print("Using TCP for communications");
1361
1362         sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
1363         if (sock) {
1364                 add_sock(sock, con);
1365                 result = 0;
1366         }
1367         else {
1368                 result = -EADDRINUSE;
1369         }
1370
1371         return result;
1372 }
1373
1374
1375
1376 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1377                                                      gfp_t allocation)
1378 {
1379         struct writequeue_entry *entry;
1380
1381         entry = kmalloc(sizeof(struct writequeue_entry), allocation);
1382         if (!entry)
1383                 return NULL;
1384
1385         entry->page = alloc_page(allocation);
1386         if (!entry->page) {
1387                 kfree(entry);
1388                 return NULL;
1389         }
1390
1391         entry->offset = 0;
1392         entry->len = 0;
1393         entry->end = 0;
1394         entry->users = 0;
1395         entry->con = con;
1396
1397         return entry;
1398 }
1399
1400 void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
1401 {
1402         struct connection *con;
1403         struct writequeue_entry *e;
1404         int offset = 0;
1405
1406         con = nodeid2con(nodeid, allocation);
1407         if (!con)
1408                 return NULL;
1409
1410         spin_lock(&con->writequeue_lock);
1411         e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
1412         if ((&e->list == &con->writequeue) ||
1413             (PAGE_SIZE - e->end < len)) {
1414                 e = NULL;
1415         } else {
1416                 offset = e->end;
1417                 e->end += len;
1418                 e->users++;
1419         }
1420         spin_unlock(&con->writequeue_lock);
1421
1422         if (e) {
1423         got_one:
1424                 *ppc = page_address(e->page) + offset;
1425                 return e;
1426         }
1427
1428         e = new_writequeue_entry(con, allocation);
1429         if (e) {
1430                 spin_lock(&con->writequeue_lock);
1431                 offset = e->end;
1432                 e->end += len;
1433                 e->users++;
1434                 list_add_tail(&e->list, &con->writequeue);
1435                 spin_unlock(&con->writequeue_lock);
1436                 goto got_one;
1437         }
1438         return NULL;
1439 }
1440
1441 void dlm_lowcomms_commit_buffer(void *mh)
1442 {
1443         struct writequeue_entry *e = (struct writequeue_entry *)mh;
1444         struct connection *con = e->con;
1445         int users;
1446
1447         spin_lock(&con->writequeue_lock);
1448         users = --e->users;
1449         if (users)
1450                 goto out;
1451         e->len = e->end - e->offset;
1452         spin_unlock(&con->writequeue_lock);
1453
1454         queue_work(send_workqueue, &con->swork);
1455         return;
1456
1457 out:
1458         spin_unlock(&con->writequeue_lock);
1459         return;
1460 }
1461
1462 /* Send a message */
1463 static void send_to_sock(struct connection *con)
1464 {
1465         int ret = 0;
1466         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1467         struct writequeue_entry *e;
1468         int len, offset;
1469         int count = 0;
1470
1471         mutex_lock(&con->sock_mutex);
1472         if (con->sock == NULL)
1473                 goto out_connect;
1474
1475         spin_lock(&con->writequeue_lock);
1476         for (;;) {
1477                 e = list_entry(con->writequeue.next, struct writequeue_entry,
1478                                list);
1479                 if ((struct list_head *) e == &con->writequeue)
1480                         break;
1481
1482                 len = e->len;
1483                 offset = e->offset;
1484                 BUG_ON(len == 0 && e->users == 0);
1485                 spin_unlock(&con->writequeue_lock);
1486
1487                 ret = 0;
1488                 if (len) {
1489                         ret = kernel_sendpage(con->sock, e->page, offset, len,
1490                                               msg_flags);
1491                         if (ret == -EAGAIN || ret == 0) {
1492                                 if (ret == -EAGAIN &&
1493                                     test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1494                                     !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1495                                         /* Notify TCP that we're limited by the
1496                                          * application window size.
1497                                          */
1498                                         set_bit(SOCK_NOSPACE, &con->sock->flags);
1499                                         con->sock->sk->sk_write_pending++;
1500                                 }
1501                                 cond_resched();
1502                                 goto out;
1503                         } else if (ret < 0)
1504                                 goto send_error;
1505                 }
1506
1507                 /* Don't starve people filling buffers */
1508                 if (++count >= MAX_SEND_MSG_COUNT) {
1509                         cond_resched();
1510                         count = 0;
1511                 }
1512
1513                 spin_lock(&con->writequeue_lock);
1514                 writequeue_entry_complete(e, ret);
1515         }
1516         spin_unlock(&con->writequeue_lock);
1517 out:
1518         mutex_unlock(&con->sock_mutex);
1519         return;
1520
1521 send_error:
1522         mutex_unlock(&con->sock_mutex);
1523         close_connection(con, false, false, true);
1524         /* Requeue the send work. When the work daemon runs again, it will try
1525            a new connection, then call this function again. */
1526         queue_work(send_workqueue, &con->swork);
1527         return;
1528
1529 out_connect:
1530         mutex_unlock(&con->sock_mutex);
1531         queue_work(send_workqueue, &con->swork);
1532         cond_resched();
1533 }
1534
1535 static void clean_one_writequeue(struct connection *con)
1536 {
1537         struct writequeue_entry *e, *safe;
1538
1539         spin_lock(&con->writequeue_lock);
1540         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1541                 list_del(&e->list);
1542                 free_entry(e);
1543         }
1544         spin_unlock(&con->writequeue_lock);
1545 }
1546
1547 /* Called from recovery when it knows that a node has
1548    left the cluster */
1549 int dlm_lowcomms_close(int nodeid)
1550 {
1551         struct connection *con;
1552         struct dlm_node_addr *na;
1553
1554         log_print("closing connection to node %d", nodeid);
1555         con = nodeid2con(nodeid, 0);
1556         if (con) {
1557                 set_bit(CF_CLOSE, &con->flags);
1558                 close_connection(con, true, true, true);
1559                 clean_one_writequeue(con);
1560         }
1561
1562         spin_lock(&dlm_node_addrs_spin);
1563         na = find_node_addr(nodeid);
1564         if (na) {
1565                 list_del(&na->list);
1566                 while (na->addr_count--)
1567                         kfree(na->addr[na->addr_count]);
1568                 kfree(na);
1569         }
1570         spin_unlock(&dlm_node_addrs_spin);
1571
1572         return 0;
1573 }
1574
1575 /* Receive workqueue function */
1576 static void process_recv_sockets(struct work_struct *work)
1577 {
1578         struct connection *con = container_of(work, struct connection, rwork);
1579         int err;
1580
1581         clear_bit(CF_READ_PENDING, &con->flags);
1582         do {
1583                 err = con->rx_action(con);
1584         } while (!err);
1585 }
1586
1587 /* Send workqueue function */
1588 static void process_send_sockets(struct work_struct *work)
1589 {
1590         struct connection *con = container_of(work, struct connection, swork);
1591
1592         if (con->sock == NULL) /* not mutex protected so check it inside too */
1593                 con->connect_action(con);
1594         if (!list_empty(&con->writequeue))
1595                 send_to_sock(con);
1596 }
1597
1598
1599 /* Discard all entries on the write queues */
1600 static void clean_writequeues(void)
1601 {
1602         foreach_conn(clean_one_writequeue);
1603 }
1604
1605 static void work_stop(void)
1606 {
1607         destroy_workqueue(recv_workqueue);
1608         destroy_workqueue(send_workqueue);
1609 }
1610
1611 static int work_start(void)
1612 {
1613         recv_workqueue = alloc_workqueue("dlm_recv",
1614                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1615         if (!recv_workqueue) {
1616                 log_print("can't start dlm_recv");
1617                 return -ENOMEM;
1618         }
1619
1620         send_workqueue = alloc_workqueue("dlm_send",
1621                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1622         if (!send_workqueue) {
1623                 log_print("can't start dlm_send");
1624                 destroy_workqueue(recv_workqueue);
1625                 return -ENOMEM;
1626         }
1627
1628         return 0;
1629 }
1630
1631 static void _stop_conn(struct connection *con, bool and_other)
1632 {
1633         mutex_lock(&con->sock_mutex);
1634         set_bit(CF_READ_PENDING, &con->flags);
1635         if (con->sock && con->sock->sk)
1636                 con->sock->sk->sk_user_data = NULL;
1637         if (con->othercon && and_other)
1638                 _stop_conn(con->othercon, false);
1639         mutex_unlock(&con->sock_mutex);
1640 }
1641
1642 static void stop_conn(struct connection *con)
1643 {
1644         _stop_conn(con, true);
1645 }
1646
1647 static void free_conn(struct connection *con)
1648 {
1649         close_connection(con, true, true, true);
1650         if (con->othercon)
1651                 kmem_cache_free(con_cache, con->othercon);
1652         hlist_del(&con->list);
1653         kmem_cache_free(con_cache, con);
1654 }
1655
1656 static void work_flush(void)
1657 {
1658         int ok;
1659         int i;
1660         struct hlist_node *n;
1661         struct connection *con;
1662
1663         flush_workqueue(recv_workqueue);
1664         flush_workqueue(send_workqueue);
1665         do {
1666                 ok = 1;
1667                 foreach_conn(stop_conn);
1668                 flush_workqueue(recv_workqueue);
1669                 flush_workqueue(send_workqueue);
1670                 for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1671                         hlist_for_each_entry_safe(con, n,
1672                                                   &connection_hash[i], list) {
1673                                 ok &= test_bit(CF_READ_PENDING, &con->flags);
1674                                 if (con->othercon)
1675                                         ok &= test_bit(CF_READ_PENDING,
1676                                                        &con->othercon->flags);
1677                         }
1678                 }
1679         } while (!ok);
1680 }
1681
1682 void dlm_lowcomms_stop(void)
1683 {
1684         /* Set all the flags to prevent any
1685            socket activity.
1686         */
1687         mutex_lock(&connections_lock);
1688         dlm_allow_conn = 0;
1689         mutex_unlock(&connections_lock);
1690         work_flush();
1691         clean_writequeues();
1692         foreach_conn(free_conn);
1693         work_stop();
1694
1695         kmem_cache_destroy(con_cache);
1696 }
1697
1698 int dlm_lowcomms_start(void)
1699 {
1700         int error = -EINVAL;
1701         struct connection *con;
1702         int i;
1703
1704         for (i = 0; i < CONN_HASH_SIZE; i++)
1705                 INIT_HLIST_HEAD(&connection_hash[i]);
1706
1707         init_local();
1708         if (!dlm_local_count) {
1709                 error = -ENOTCONN;
1710                 log_print("no local IP address has been set");
1711                 goto fail;
1712         }
1713
1714         error = -ENOMEM;
1715         con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
1716                                       __alignof__(struct connection), 0,
1717                                       NULL);
1718         if (!con_cache)
1719                 goto fail;
1720
1721         error = work_start();
1722         if (error)
1723                 goto fail_destroy;
1724
1725         dlm_allow_conn = 1;
1726
1727         /* Start listening */
1728         if (dlm_config.ci_protocol == 0)
1729                 error = tcp_listen_for_all();
1730         else
1731                 error = sctp_listen_for_all();
1732         if (error)
1733                 goto fail_unlisten;
1734
1735         return 0;
1736
1737 fail_unlisten:
1738         dlm_allow_conn = 0;
1739         con = nodeid2con(0,0);
1740         if (con) {
1741                 close_connection(con, false, true, true);
1742                 kmem_cache_free(con_cache, con);
1743         }
1744 fail_destroy:
1745         kmem_cache_destroy(con_cache);
1746 fail:
1747         return error;
1748 }
1749
1750 void dlm_lowcomms_exit(void)
1751 {
1752         struct dlm_node_addr *na, *safe;
1753
1754         spin_lock(&dlm_node_addrs_spin);
1755         list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
1756                 list_del(&na->list);
1757                 while (na->addr_count--)
1758                         kfree(na->addr[na->addr_count]);
1759                 kfree(na);
1760         }
1761         spin_unlock(&dlm_node_addrs_spin);
1762 }