fs/dlm/lowcomms.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /******************************************************************************
   3 *******************************************************************************
   4 **
   5 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   6 **  Copyright (C) 2004-2009 Red Hat, Inc.  All rights reserved.
   7 **
   8 **
   9 *******************************************************************************
  10 ******************************************************************************/
  11
  12 /*
  13  * lowcomms.c
  14  *
  15  * This is the "low-level" comms layer.
  16  *
  17  * It is responsible for sending/receiving messages
  18  * from other nodes in the cluster.
  19  *
  20  * Cluster nodes are referred to by their nodeids. nodeids are
  21  * simply 32 bit numbers to the locking module - if they need to
  22  * be expanded for the cluster infrastructure then that is its
  23  * responsibility. It is this layer's
  24  * responsibility to resolve these into IP address or
  25  * whatever it needs for inter-node communication.
  26  *
  27  * The comms level is two kernel threads that deal mainly with
  28  * the receiving of messages from other nodes and passing them
  29  * up to the mid-level comms layer (which understands the
  30  * message format) for execution by the locking core, and
  31  * a send thread which does all the setting up of connections
  32  * to remote nodes and the sending of data. Threads are not allowed
  33  * to send their own data because it may cause them to wait in times
  34  * of high load. Also, this way, the sending thread can collect together
  35  * messages bound for one node and send them in one block.
  36  *
  37  * lowcomms will choose to use either TCP or SCTP as its transport layer
  38  * depending on the configuration variable 'protocol'. This should be set
  39  * to 0 (default) for TCP or 1 for SCTP. It should be configured using a
  40  * cluster-wide mechanism as it must be the same on all nodes of the cluster
  41  * for the DLM to function.
  42  *
  43  */
  44
  45 #include <asm/ioctls.h>
  46 #include <net/sock.h>
  47 #include <net/tcp.h>
  48 #include <linux/pagemap.h>
  49 #include <linux/file.h>
  50 #include <linux/mutex.h>
  51 #include <linux/sctp.h>
  52 #include <linux/slab.h>
  53 #include <net/sctp/sctp.h>
  54 #include <net/ipv6.h>
  55
  56 #include "dlm_internal.h"
  57 #include "lowcomms.h"
  58 #include "midcomms.h"
  59 #include "config.h"
  60
  61 #define NEEDED_RMEM (4*1024*1024)
  62 #define CONN_HASH_SIZE 32
  63
  64 /* Number of messages to send before rescheduling */
  65 #define MAX_SEND_MSG_COUNT 25
  66 #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(10000)
  67
  68 struct cbuf {
  69         unsigned int base;
  70         unsigned int len;
  71         unsigned int mask;
  72 };
  73
  74 static void cbuf_add(struct cbuf *cb, int n)
  75 {
  76         cb->len += n;
  77 }
  78
  79 static int cbuf_data(struct cbuf *cb)
  80 {
  81         return ((cb->base + cb->len) & cb->mask);
  82 }
  83
  84 static void cbuf_init(struct cbuf *cb, int size)
  85 {
  86         cb->base = cb->len = 0;
  87         cb->mask = size-1;
  88 }
  89
  90 static void cbuf_eat(struct cbuf *cb, int n)
  91 {
  92         cb->len  -= n;
  93         cb->base += n;
  94         cb->base &= cb->mask;
  95 }
  96
  97 static bool cbuf_empty(struct cbuf *cb)
  98 {
  99         return cb->len == 0;
 100 }
 101
 102 struct connection {
 103         struct socket *sock;    /* NULL if not connected */
 104         uint32_t nodeid;        /* So we know who we are in the list */
 105         struct mutex sock_mutex;
 106         unsigned long flags;
 107 #define CF_READ_PENDING 1
 108 #define CF_WRITE_PENDING 2
 109 #define CF_INIT_PENDING 4
 110 #define CF_IS_OTHERCON 5
 111 #define CF_CLOSE 6
 112 #define CF_APP_LIMITED 7
 113 #define CF_CLOSING 8
 114 #define CF_SHUTDOWN 9
 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         void (*shutdown_action)(struct connection *con); /* What to do to shutdown */
 120         struct page *rx_page;
 121         struct cbuf cb;
 122         int retries;
 123 #define MAX_CONNECT_RETRIES 3
 124         struct hlist_node list;
 125         struct connection *othercon;
 126         struct work_struct rwork; /* Receive workqueue */
 127         struct work_struct swork; /* Send workqueue */
 128         wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */
 129 };
 130 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 131
 132 /* An entry waiting to be sent */
 133 struct writequeue_entry {
 134         struct list_head list;
 135         struct page *page;
 136         int offset;
 137         int len;
 138         int end;
 139         int users;
 140         struct connection *con;
 141 };
 142
 143 struct dlm_node_addr {
 144         struct list_head list;
 145         int nodeid;
 146         int addr_count;
 147         int curr_addr_index;
 148         struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
 149 };
 150
 151 static struct listen_sock_callbacks {
 152         void (*sk_error_report)(struct sock *);
 153         void (*sk_data_ready)(struct sock *);
 154         void (*sk_state_change)(struct sock *);
 155         void (*sk_write_space)(struct sock *);
 156 } listen_sock;
 157
 158 static LIST_HEAD(dlm_node_addrs);
 159 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
 160
 161 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 162 static int dlm_local_count;
 163 static int dlm_allow_conn;
 164
 165 /* Work queues */
 166 static struct workqueue_struct *recv_workqueue;
 167 static struct workqueue_struct *send_workqueue;
 168
 169 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 170 static DEFINE_MUTEX(connections_lock);
 171 static struct kmem_cache *con_cache;
 172
 173 static void process_recv_sockets(struct work_struct *work);
 174 static void process_send_sockets(struct work_struct *work);
 175
 176
 177 /* This is deliberately very simple because most clusters have simple
 178    sequential nodeids, so we should be able to go straight to a connection
 179    struct in the array */
 180 static inline int nodeid_hash(int nodeid)
 181 {
 182         return nodeid & (CONN_HASH_SIZE-1);
 183 }
 184
 185 static struct connection *__find_con(int nodeid)
 186 {
 187         int r;
 188         struct connection *con;
 189
 190         r = nodeid_hash(nodeid);
 191
 192         hlist_for_each_entry(con, &connection_hash[r], list) {
 193                 if (con->nodeid == nodeid)
 194                         return con;
 195         }
 196         return NULL;
 197 }
 198
 199 /*
 200  * If 'allocation' is zero then we don't attempt to create a new
 201  * connection structure for this node.
 202  */
 203 static struct connection *__nodeid2con(int nodeid, gfp_t alloc)
 204 {
 205         struct connection *con = NULL;
 206         int r;
 207
 208         con = __find_con(nodeid);
 209         if (con || !alloc)
 210                 return con;
 211
 212         con = kmem_cache_zalloc(con_cache, alloc);
 213         if (!con)
 214                 return NULL;
 215
 216         r = nodeid_hash(nodeid);
 217         hlist_add_head(&con->list, &connection_hash[r]);
 218
 219         con->nodeid = nodeid;
 220         mutex_init(&con->sock_mutex);
 221         INIT_LIST_HEAD(&con->writequeue);
 222         spin_lock_init(&con->writequeue_lock);
 223         INIT_WORK(&con->swork, process_send_sockets);
 224         INIT_WORK(&con->rwork, process_recv_sockets);
 225         init_waitqueue_head(&con->shutdown_wait);
 226
 227         /* Setup action pointers for child sockets */
 228         if (con->nodeid) {
 229                 struct connection *zerocon = __find_con(0);
 230
 231                 con->connect_action = zerocon->connect_action;
 232                 if (!con->rx_action)
 233                         con->rx_action = zerocon->rx_action;
 234         }
 235
 236         return con;
 237 }
 238
 239 /* Loop round all connections */
 240 static void foreach_conn(void (*conn_func)(struct connection *c))
 241 {
 242         int i;
 243         struct hlist_node *n;
 244         struct connection *con;
 245
 246         for (i = 0; i < CONN_HASH_SIZE; i++) {
 247                 hlist_for_each_entry_safe(con, n, &connection_hash[i], list)
 248                         conn_func(con);
 249         }
 250 }
 251
 252 static struct connection *nodeid2con(int nodeid, gfp_t allocation)
 253 {
 254         struct connection *con;
 255
 256         mutex_lock(&connections_lock);
 257         con = __nodeid2con(nodeid, allocation);
 258         mutex_unlock(&connections_lock);
 259
 260         return con;
 261 }
 262
 263 static struct dlm_node_addr *find_node_addr(int nodeid)
 264 {
 265         struct dlm_node_addr *na;
 266
 267         list_for_each_entry(na, &dlm_node_addrs, list) {
 268                 if (na->nodeid == nodeid)
 269                         return na;
 270         }
 271         return NULL;
 272 }
 273
 274 static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y)
 275 {
 276         switch (x->ss_family) {
 277         case AF_INET: {
 278                 struct sockaddr_in *sinx = (struct sockaddr_in *)x;
 279                 struct sockaddr_in *siny = (struct sockaddr_in *)y;
 280                 if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
 281                         return 0;
 282                 if (sinx->sin_port != siny->sin_port)
 283                         return 0;
 284                 break;
 285         }
 286         case AF_INET6: {
 287                 struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
 288                 struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
 289                 if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
 290                         return 0;
 291                 if (sinx->sin6_port != siny->sin6_port)
 292                         return 0;
 293                 break;
 294         }
 295         default:
 296                 return 0;
 297         }
 298         return 1;
 299 }
 300
 301 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
 302                           struct sockaddr *sa_out, bool try_new_addr)
 303 {
 304         struct sockaddr_storage sas;
 305         struct dlm_node_addr *na;
 306
 307         if (!dlm_local_count)
 308                 return -1;
 309
 310         spin_lock(&dlm_node_addrs_spin);
 311         na = find_node_addr(nodeid);
 312         if (na && na->addr_count) {
 313                 memcpy(&sas, na->addr[na->curr_addr_index],
 314                        sizeof(struct sockaddr_storage));
 315
 316                 if (try_new_addr) {
 317                         na->curr_addr_index++;
 318                         if (na->curr_addr_index == na->addr_count)
 319                                 na->curr_addr_index = 0;
 320                 }
 321         }
 322         spin_unlock(&dlm_node_addrs_spin);
 323
 324         if (!na)
 325                 return -EEXIST;
 326
 327         if (!na->addr_count)
 328                 return -ENOENT;
 329
 330         if (sas_out)
 331                 memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
 332
 333         if (!sa_out)
 334                 return 0;
 335
 336         if (dlm_local_addr[0]->ss_family == AF_INET) {
 337                 struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
 338                 struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
 339                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 340         } else {
 341                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
 342                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
 343                 ret6->sin6_addr = in6->sin6_addr;
 344         }
 345
 346         return 0;
 347 }
 348
 349 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid)
 350 {
 351         struct dlm_node_addr *na;
 352         int rv = -EEXIST;
 353         int addr_i;
 354
 355         spin_lock(&dlm_node_addrs_spin);
 356         list_for_each_entry(na, &dlm_node_addrs, list) {
 357                 if (!na->addr_count)
 358                         continue;
 359
 360                 for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
 361                         if (addr_compare(na->addr[addr_i], addr)) {
 362                                 *nodeid = na->nodeid;
 363                                 rv = 0;
 364                                 goto unlock;
 365                         }
 366                 }
 367         }
 368 unlock:
 369         spin_unlock(&dlm_node_addrs_spin);
 370         return rv;
 371 }
 372
 373 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 374 {
 375         struct sockaddr_storage *new_addr;
 376         struct dlm_node_addr *new_node, *na;
 377
 378         new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
 379         if (!new_node)
 380                 return -ENOMEM;
 381
 382         new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
 383         if (!new_addr) {
 384                 kfree(new_node);
 385                 return -ENOMEM;
 386         }
 387
 388         memcpy(new_addr, addr, len);
 389
 390         spin_lock(&dlm_node_addrs_spin);
 391         na = find_node_addr(nodeid);
 392         if (!na) {
 393                 new_node->nodeid = nodeid;
 394                 new_node->addr[0] = new_addr;
 395                 new_node->addr_count = 1;
 396                 list_add(&new_node->list, &dlm_node_addrs);
 397                 spin_unlock(&dlm_node_addrs_spin);
 398                 return 0;
 399         }
 400
 401         if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
 402                 spin_unlock(&dlm_node_addrs_spin);
 403                 kfree(new_addr);
 404                 kfree(new_node);
 405                 return -ENOSPC;
 406         }
 407
 408         na->addr[na->addr_count++] = new_addr;
 409         spin_unlock(&dlm_node_addrs_spin);
 410         kfree(new_node);
 411         return 0;
 412 }
 413
 414 /* Data available on socket or listen socket received a connect */
 415 static void lowcomms_data_ready(struct sock *sk)
 416 {
 417         struct connection *con;
 418
 419         read_lock_bh(&sk->sk_callback_lock);
 420         con = sock2con(sk);
 421         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 422                 queue_work(recv_workqueue, &con->rwork);
 423         read_unlock_bh(&sk->sk_callback_lock);
 424 }
 425
 426 static void lowcomms_write_space(struct sock *sk)
 427 {
 428         struct connection *con;
 429
 430         read_lock_bh(&sk->sk_callback_lock);
 431         con = sock2con(sk);
 432         if (!con)
 433                 goto out;
 434
 435         clear_bit(SOCK_NOSPACE, &con->sock->flags);
 436
 437         if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
 438                 con->sock->sk->sk_write_pending--;
 439                 clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
 440         }
 441
 442         queue_work(send_workqueue, &con->swork);
 443 out:
 444         read_unlock_bh(&sk->sk_callback_lock);
 445 }
 446
 447 static inline void lowcomms_connect_sock(struct connection *con)
 448 {
 449         if (test_bit(CF_CLOSE, &con->flags))
 450                 return;
 451         queue_work(send_workqueue, &con->swork);
 452         cond_resched();
 453 }
 454
 455 static void lowcomms_state_change(struct sock *sk)
 456 {
 457         /* SCTP layer is not calling sk_data_ready when the connection
 458          * is done, so we catch the signal through here. Also, it
 459          * doesn't switch socket state when entering shutdown, so we
 460          * skip the write in that case.
 461          */
 462         if (sk->sk_shutdown) {
 463                 if (sk->sk_shutdown == RCV_SHUTDOWN)
 464                         lowcomms_data_ready(sk);
 465         } else if (sk->sk_state == TCP_ESTABLISHED) {
 466                 lowcomms_write_space(sk);
 467         }
 468 }
 469
 470 int dlm_lowcomms_connect_node(int nodeid)
 471 {
 472         struct connection *con;
 473
 474         if (nodeid == dlm_our_nodeid())
 475                 return 0;
 476
 477         con = nodeid2con(nodeid, GFP_NOFS);
 478         if (!con)
 479                 return -ENOMEM;
 480         lowcomms_connect_sock(con);
 481         return 0;
 482 }
 483
 484 static void lowcomms_error_report(struct sock *sk)
 485 {
 486         struct connection *con;
 487         struct sockaddr_storage saddr;
 488         void (*orig_report)(struct sock *) = NULL;
 489
 490         read_lock_bh(&sk->sk_callback_lock);
 491         con = sock2con(sk);
 492         if (con == NULL)
 493                 goto out;
 494
 495         orig_report = listen_sock.sk_error_report;
 496         if (con->sock == NULL ||
 497             kernel_getpeername(con->sock, (struct sockaddr *)&saddr) < 0) {
 498                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 499                                    "sending to node %d, port %d, "
 500                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 501                                    con->nodeid, dlm_config.ci_tcp_port,
 502                                    sk->sk_err, sk->sk_err_soft);
 503         } else if (saddr.ss_family == AF_INET) {
 504                 struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr;
 505
 506                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 507                                    "sending to node %d at %pI4, port %d, "
 508                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 509                                    con->nodeid, &sin4->sin_addr.s_addr,
 510                                    dlm_config.ci_tcp_port, sk->sk_err,
 511                                    sk->sk_err_soft);
 512         } else {
 513                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr;
 514
 515                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 516                                    "sending to node %d at %u.%u.%u.%u, "
 517                                    "port %d, sk_err=%d/%d\n", dlm_our_nodeid(),
 518                                    con->nodeid, sin6->sin6_addr.s6_addr32[0],
 519                                    sin6->sin6_addr.s6_addr32[1],
 520                                    sin6->sin6_addr.s6_addr32[2],
 521                                    sin6->sin6_addr.s6_addr32[3],
 522                                    dlm_config.ci_tcp_port, sk->sk_err,
 523                                    sk->sk_err_soft);
 524         }
 525 out:
 526         read_unlock_bh(&sk->sk_callback_lock);
 527         if (orig_report)
 528                 orig_report(sk);
 529 }
 530
 531 /* Note: sk_callback_lock must be locked before calling this function. */
 532 static void save_listen_callbacks(struct socket *sock)
 533 {
 534         struct sock *sk = sock->sk;
 535
 536         listen_sock.sk_data_ready = sk->sk_data_ready;
 537         listen_sock.sk_state_change = sk->sk_state_change;
 538         listen_sock.sk_write_space = sk->sk_write_space;
 539         listen_sock.sk_error_report = sk->sk_error_report;
 540 }
 541
 542 static void restore_callbacks(struct socket *sock)
 543 {
 544         struct sock *sk = sock->sk;
 545
 546         write_lock_bh(&sk->sk_callback_lock);
 547         sk->sk_user_data = NULL;
 548         sk->sk_data_ready = listen_sock.sk_data_ready;
 549         sk->sk_state_change = listen_sock.sk_state_change;
 550         sk->sk_write_space = listen_sock.sk_write_space;
 551         sk->sk_error_report = listen_sock.sk_error_report;
 552         write_unlock_bh(&sk->sk_callback_lock);
 553 }
 554
 555 /* Make a socket active */
 556 static void add_sock(struct socket *sock, struct connection *con)
 557 {
 558         struct sock *sk = sock->sk;
 559
 560         write_lock_bh(&sk->sk_callback_lock);
 561         con->sock = sock;
 562
 563         sk->sk_user_data = con;
 564         /* Install a data_ready callback */
 565         sk->sk_data_ready = lowcomms_data_ready;
 566         sk->sk_write_space = lowcomms_write_space;
 567         sk->sk_state_change = lowcomms_state_change;
 568         sk->sk_allocation = GFP_NOFS;
 569         sk->sk_error_report = lowcomms_error_report;
 570         write_unlock_bh(&sk->sk_callback_lock);
 571 }
 572
 573 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 574    length */
 575 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 576                           int *addr_len)
 577 {
 578         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 579         if (saddr->ss_family == AF_INET) {
 580                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 581                 in4_addr->sin_port = cpu_to_be16(port);
 582                 *addr_len = sizeof(struct sockaddr_in);
 583                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 584         } else {
 585                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 586                 in6_addr->sin6_port = cpu_to_be16(port);
 587                 *addr_len = sizeof(struct sockaddr_in6);
 588         }
 589         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 590 }
 591
 592 /* Close a remote connection and tidy up */
 593 static void close_connection(struct connection *con, bool and_other,
 594                              bool tx, bool rx)
 595 {
 596         bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
 597
 598         if (tx && !closing && cancel_work_sync(&con->swork)) {
 599                 log_print("canceled swork for node %d", con->nodeid);
 600                 clear_bit(CF_WRITE_PENDING, &con->flags);
 601         }
 602         if (rx && !closing && cancel_work_sync(&con->rwork)) {
 603                 log_print("canceled rwork for node %d", con->nodeid);
 604                 clear_bit(CF_READ_PENDING, &con->flags);
 605         }
 606
 607         mutex_lock(&con->sock_mutex);
 608         if (con->sock) {
 609                 restore_callbacks(con->sock);
 610                 sock_release(con->sock);
 611                 con->sock = NULL;
 612         }
 613         if (con->othercon && and_other) {
 614                 /* Will only re-enter once. */
 615                 close_connection(con->othercon, false, true, true);
 616         }
 617         if (con->rx_page) {
 618                 __free_page(con->rx_page);
 619                 con->rx_page = NULL;
 620         }
 621
 622         con->retries = 0;
 623         mutex_unlock(&con->sock_mutex);
 624         clear_bit(CF_CLOSING, &con->flags);
 625 }
 626
 627 static void shutdown_connection(struct connection *con)
 628 {
 629         int ret;
 630
 631         if (cancel_work_sync(&con->swork)) {
 632                 log_print("canceled swork for node %d", con->nodeid);
 633                 clear_bit(CF_WRITE_PENDING, &con->flags);
 634         }
 635
 636         mutex_lock(&con->sock_mutex);
 637         /* nothing to shutdown */
 638         if (!con->sock) {
 639                 mutex_unlock(&con->sock_mutex);
 640                 return;
 641         }
 642
 643         set_bit(CF_SHUTDOWN, &con->flags);
 644         ret = kernel_sock_shutdown(con->sock, SHUT_WR);
 645         mutex_unlock(&con->sock_mutex);
 646         if (ret) {
 647                 log_print("Connection %p failed to shutdown: %d will force close",
 648                           con, ret);
 649                 goto force_close;
 650         } else {
 651                 ret = wait_event_timeout(con->shutdown_wait,
 652                                          !test_bit(CF_SHUTDOWN, &con->flags),
 653                                          DLM_SHUTDOWN_WAIT_TIMEOUT);
 654                 if (ret == 0) {
 655                         log_print("Connection %p shutdown timed out, will force close",
 656                                   con);
 657                         goto force_close;
 658                 }
 659         }
 660
 661         return;
 662
 663 force_close:
 664         clear_bit(CF_SHUTDOWN, &con->flags);
 665         close_connection(con, false, true, true);
 666 }
 667
 668 static void dlm_tcp_shutdown(struct connection *con)
 669 {
 670         if (con->othercon)
 671                 shutdown_connection(con->othercon);
 672         shutdown_connection(con);
 673 }
 674
 675 /* Data received from remote end */
 676 static int receive_from_sock(struct connection *con)
 677 {
 678         int ret = 0;
 679         struct msghdr msg = {};
 680         struct kvec iov[2];
 681         unsigned len;
 682         int r;
 683         int call_again_soon = 0;
 684         int nvec;
 685
 686         mutex_lock(&con->sock_mutex);
 687
 688         if (con->sock == NULL) {
 689                 ret = -EAGAIN;
 690                 goto out_close;
 691         }
 692         if (con->nodeid == 0) {
 693                 ret = -EINVAL;
 694                 goto out_close;
 695         }
 696
 697         if (con->rx_page == NULL) {
 698                 /*
 699                  * This doesn't need to be atomic, but I think it should
 700                  * improve performance if it is.
 701                  */
 702                 con->rx_page = alloc_page(GFP_ATOMIC);
 703                 if (con->rx_page == NULL)
 704                         goto out_resched;
 705                 cbuf_init(&con->cb, PAGE_SIZE);
 706         }
 707
 708         /*
 709          * iov[0] is the bit of the circular buffer between the current end
 710          * point (cb.base + cb.len) and the end of the buffer.
 711          */
 712         iov[0].iov_len = con->cb.base - cbuf_data(&con->cb);
 713         iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb);
 714         iov[1].iov_len = 0;
 715         nvec = 1;
 716
 717         /*
 718          * iov[1] is the bit of the circular buffer between the start of the
 719          * buffer and the start of the currently used section (cb.base)
 720          */
 721         if (cbuf_data(&con->cb) >= con->cb.base) {
 722                 iov[0].iov_len = PAGE_SIZE - cbuf_data(&con->cb);
 723                 iov[1].iov_len = con->cb.base;
 724                 iov[1].iov_base = page_address(con->rx_page);
 725                 nvec = 2;
 726         }
 727         len = iov[0].iov_len + iov[1].iov_len;
 728         iov_iter_kvec(&msg.msg_iter, READ, iov, nvec, len);
 729
 730         r = ret = sock_recvmsg(con->sock, &msg, MSG_DONTWAIT | MSG_NOSIGNAL);
 731         if (ret <= 0)
 732                 goto out_close;
 733         else if (ret == len)
 734                 call_again_soon = 1;
 735
 736         cbuf_add(&con->cb, ret);
 737         ret = dlm_process_incoming_buffer(con->nodeid,
 738                                           page_address(con->rx_page),
 739                                           con->cb.base, con->cb.len,
 740                                           PAGE_SIZE);
 741         if (ret < 0) {
 742                 log_print("lowcomms err %d: addr=%p, base=%u, len=%u, read=%d",
 743                           ret, page_address(con->rx_page), con->cb.base,
 744                           con->cb.len, r);
 745                 cbuf_eat(&con->cb, r);
 746         } else {
 747                 cbuf_eat(&con->cb, ret);
 748         }
 749
 750         if (cbuf_empty(&con->cb) && !call_again_soon) {
 751                 __free_page(con->rx_page);
 752                 con->rx_page = NULL;
 753         }
 754
 755         if (call_again_soon)
 756                 goto out_resched;
 757         mutex_unlock(&con->sock_mutex);
 758         return 0;
 759
 760 out_resched:
 761         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 762                 queue_work(recv_workqueue, &con->rwork);
 763         mutex_unlock(&con->sock_mutex);
 764         return -EAGAIN;
 765
 766 out_close:
 767         mutex_unlock(&con->sock_mutex);
 768         if (ret != -EAGAIN) {
 769                 /* Reconnect when there is something to send */
 770                 close_connection(con, false, true, false);
 771                 if (ret == 0) {
 772                         log_print("connection %p got EOF from %d",
 773                                   con, con->nodeid);
 774                         /* handling for tcp shutdown */
 775                         clear_bit(CF_SHUTDOWN, &con->flags);
 776                         wake_up(&con->shutdown_wait);
 777                         /* signal to breaking receive worker */
 778                         ret = -1;
 779                 }
 780         }
 781         return ret;
 782 }
 783
 784 /* Listening socket is busy, accept a connection */
 785 static int accept_from_sock(struct connection *con)
 786 {
 787         int result;
 788         struct sockaddr_storage peeraddr;
 789         struct socket *newsock;
 790         int len;
 791         int nodeid;
 792         struct connection *newcon;
 793         struct connection *addcon;
 794
 795         mutex_lock(&connections_lock);
 796         if (!dlm_allow_conn) {
 797                 mutex_unlock(&connections_lock);
 798                 return -1;
 799         }
 800         mutex_unlock(&connections_lock);
 801
 802         mutex_lock_nested(&con->sock_mutex, 0);
 803
 804         if (!con->sock) {
 805                 mutex_unlock(&con->sock_mutex);
 806                 return -ENOTCONN;
 807         }
 808
 809         result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
 810         if (result < 0)
 811                 goto accept_err;
 812
 813         /* Get the connected socket's peer */
 814         memset(&peeraddr, 0, sizeof(peeraddr));
 815         len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
 816         if (len < 0) {
 817                 result = -ECONNABORTED;
 818                 goto accept_err;
 819         }
 820
 821         /* Get the new node's NODEID */
 822         make_sockaddr(&peeraddr, 0, &len);
 823         if (addr_to_nodeid(&peeraddr, &nodeid)) {
 824                 unsigned char *b=(unsigned char *)&peeraddr;
 825                 log_print("connect from non cluster node");
 826                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
 827                                      b, sizeof(struct sockaddr_storage));
 828                 sock_release(newsock);
 829                 mutex_unlock(&con->sock_mutex);
 830                 return -1;
 831         }
 832
 833         log_print("got connection from %d", nodeid);
 834
 835         /*  Check to see if we already have a connection to this node. This
 836          *  could happen if the two nodes initiate a connection at roughly
 837          *  the same time and the connections cross on the wire.
 838          *  In this case we store the incoming one in "othercon"
 839          */
 840         newcon = nodeid2con(nodeid, GFP_NOFS);
 841         if (!newcon) {
 842                 result = -ENOMEM;
 843                 goto accept_err;
 844         }
 845         mutex_lock_nested(&newcon->sock_mutex, 1);
 846         if (newcon->sock) {
 847                 struct connection *othercon = newcon->othercon;
 848
 849                 if (!othercon) {
 850                         othercon = kmem_cache_zalloc(con_cache, GFP_NOFS);
 851                         if (!othercon) {
 852                                 log_print("failed to allocate incoming socket");
 853                                 mutex_unlock(&newcon->sock_mutex);
 854                                 result = -ENOMEM;
 855                                 goto accept_err;
 856                         }
 857                         othercon->nodeid = nodeid;
 858                         othercon->rx_action = receive_from_sock;
 859                         mutex_init(&othercon->sock_mutex);
 860                         INIT_LIST_HEAD(&othercon->writequeue);
 861                         spin_lock_init(&othercon->writequeue_lock);
 862                         INIT_WORK(&othercon->swork, process_send_sockets);
 863                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 864                         init_waitqueue_head(&othercon->shutdown_wait);
 865                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 866                 } else {
 867                         /* close other sock con if we have something new */
 868                         close_connection(othercon, false, true, false);
 869                 }
 870
 871                 mutex_lock_nested(&othercon->sock_mutex, 2);
 872                 newcon->othercon = othercon;
 873                 add_sock(newsock, othercon);
 874                 addcon = othercon;
 875                 mutex_unlock(&othercon->sock_mutex);
 876         }
 877         else {
 878                 newcon->rx_action = receive_from_sock;
 879                 /* accept copies the sk after we've saved the callbacks, so we
 880                    don't want to save them a second time or comm errors will
 881                    result in calling sk_error_report recursively. */
 882                 add_sock(newsock, newcon);
 883                 addcon = newcon;
 884         }
 885
 886         mutex_unlock(&newcon->sock_mutex);
 887
 888         /*
 889          * Add it to the active queue in case we got data
 890          * between processing the accept adding the socket
 891          * to the read_sockets list
 892          */
 893         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 894                 queue_work(recv_workqueue, &addcon->rwork);
 895         mutex_unlock(&con->sock_mutex);
 896
 897         return 0;
 898
 899 accept_err:
 900         mutex_unlock(&con->sock_mutex);
 901         if (newsock)
 902                 sock_release(newsock);
 903
 904         if (result != -EAGAIN)
 905                 log_print("error accepting connection from node: %d", result);
 906         return result;
 907 }
 908
 909 static void free_entry(struct writequeue_entry *e)
 910 {
 911         __free_page(e->page);
 912         kfree(e);
 913 }
 914
 915 /*
 916  * writequeue_entry_complete - try to delete and free write queue entry
 917  * @e: write queue entry to try to delete
 918  * @completed: bytes completed
 919  *
 920  * writequeue_lock must be held.
 921  */
 922 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
 923 {
 924         e->offset += completed;
 925         e->len -= completed;
 926
 927         if (e->len == 0 && e->users == 0) {
 928                 list_del(&e->list);
 929                 free_entry(e);
 930         }
 931 }
 932
 933 /*
 934  * sctp_bind_addrs - bind a SCTP socket to all our addresses
 935  */
 936 static int sctp_bind_addrs(struct connection *con, uint16_t port)
 937 {
 938         struct sockaddr_storage localaddr;
 939         struct sockaddr *addr = (struct sockaddr *)&localaddr;
 940         int i, addr_len, result = 0;
 941
 942         for (i = 0; i < dlm_local_count; i++) {
 943                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
 944                 make_sockaddr(&localaddr, port, &addr_len);
 945
 946                 if (!i)
 947                         result = kernel_bind(con->sock, addr, addr_len);
 948                 else
 949                         result = sock_bind_add(con->sock->sk, addr, addr_len);
 950
 951                 if (result < 0) {
 952                         log_print("Can't bind to %d addr number %d, %d.\n",
 953                                   port, i + 1, result);
 954                         break;
 955                 }
 956         }
 957         return result;
 958 }
 959
 960 /* Initiate an SCTP association.
 961    This is a special case of send_to_sock() in that we don't yet have a
 962    peeled-off socket for this association, so we use the listening socket
 963    and add the primary IP address of the remote node.
 964  */
 965 static void sctp_connect_to_sock(struct connection *con)
 966 {
 967         struct sockaddr_storage daddr;
 968         int result;
 969         int addr_len;
 970         struct socket *sock;
 971         unsigned int mark;
 972
 973         if (con->nodeid == 0) {
 974                 log_print("attempt to connect sock 0 foiled");
 975                 return;
 976         }
 977
 978         mutex_lock(&con->sock_mutex);
 979
 980         /* Some odd races can cause double-connects, ignore them */
 981         if (con->retries++ > MAX_CONNECT_RETRIES)
 982                 goto out;
 983
 984         if (con->sock) {
 985                 log_print("node %d already connected.", con->nodeid);
 986                 goto out;
 987         }
 988
 989         memset(&daddr, 0, sizeof(daddr));
 990         result = nodeid_to_addr(con->nodeid, &daddr, NULL, true);
 991         if (result < 0) {
 992                 log_print("no address for nodeid %d", con->nodeid);
 993                 goto out;
 994         }
 995
 996         /* Create a socket to communicate with */
 997         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
 998                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
 999         if (result < 0)
1000                 goto socket_err;
1001
1002         /* set skb mark */
1003         result = dlm_comm_mark(con->nodeid, &mark);
1004         if (result < 0)
1005                 goto bind_err;
1006
1007         sock_set_mark(sock->sk, mark);
1008
1009         con->rx_action = receive_from_sock;
1010         con->connect_action = sctp_connect_to_sock;
1011         add_sock(sock, con);
1012
1013         /* Bind to all addresses. */
1014         if (sctp_bind_addrs(con, 0))
1015                 goto bind_err;
1016
1017         make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len);
1018
1019         log_print("connecting to %d", con->nodeid);
1020
1021         /* Turn off Nagle's algorithm */
1022         sctp_sock_set_nodelay(sock->sk);
1023
1024         /*
1025          * Make sock->ops->connect() function return in specified time,
1026          * since O_NONBLOCK argument in connect() function does not work here,
1027          * then, we should restore the default value of this attribute.
1028          */
1029         sock_set_sndtimeo(sock->sk, 5);
1030         result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len,
1031                                    0);
1032         sock_set_sndtimeo(sock->sk, 0);
1033
1034         if (result == -EINPROGRESS)
1035                 result = 0;
1036         if (result == 0)
1037                 goto out;
1038
1039 bind_err:
1040         con->sock = NULL;
1041         sock_release(sock);
1042
1043 socket_err:
1044         /*
1045          * Some errors are fatal and this list might need adjusting. For other
1046          * errors we try again until the max number of retries is reached.
1047          */
1048         if (result != -EHOSTUNREACH &&
1049             result != -ENETUNREACH &&
1050             result != -ENETDOWN &&
1051             result != -EINVAL &&
1052             result != -EPROTONOSUPPORT) {
1053                 log_print("connect %d try %d error %d", con->nodeid,
1054                           con->retries, result);
1055                 mutex_unlock(&con->sock_mutex);
1056                 msleep(1000);
1057                 lowcomms_connect_sock(con);
1058                 return;
1059         }
1060
1061 out:
1062         mutex_unlock(&con->sock_mutex);
1063 }
1064
1065 /* Connect a new socket to its peer */
1066 static void tcp_connect_to_sock(struct connection *con)
1067 {
1068         struct sockaddr_storage saddr, src_addr;
1069         int addr_len;
1070         struct socket *sock = NULL;
1071         unsigned int mark;
1072         int result;
1073
1074         if (con->nodeid == 0) {
1075                 log_print("attempt to connect sock 0 foiled");
1076                 return;
1077         }
1078
1079         mutex_lock(&con->sock_mutex);
1080         if (con->retries++ > MAX_CONNECT_RETRIES)
1081                 goto out;
1082
1083         /* Some odd races can cause double-connects, ignore them */
1084         if (con->sock)
1085                 goto out;
1086
1087         /* Create a socket to communicate with */
1088         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1089                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1090         if (result < 0)
1091                 goto out_err;
1092
1093         /* set skb mark */
1094         result = dlm_comm_mark(con->nodeid, &mark);
1095         if (result < 0)
1096                 goto out_err;
1097
1098         sock_set_mark(sock->sk, mark);
1099
1100         memset(&saddr, 0, sizeof(saddr));
1101         result = nodeid_to_addr(con->nodeid, &saddr, NULL, false);
1102         if (result < 0) {
1103                 log_print("no address for nodeid %d", con->nodeid);
1104                 goto out_err;
1105         }
1106
1107         con->rx_action = receive_from_sock;
1108         con->connect_action = tcp_connect_to_sock;
1109         con->shutdown_action = dlm_tcp_shutdown;
1110         add_sock(sock, con);
1111
1112         /* Bind to our cluster-known address connecting to avoid
1113            routing problems */
1114         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1115         make_sockaddr(&src_addr, 0, &addr_len);
1116         result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1117                                  addr_len);
1118         if (result < 0) {
1119                 log_print("could not bind for connect: %d", result);
1120                 /* This *may* not indicate a critical error */
1121         }
1122
1123         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1124
1125         log_print("connecting to %d", con->nodeid);
1126
1127         /* Turn off Nagle's algorithm */
1128         tcp_sock_set_nodelay(sock->sk);
1129
1130         result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1131                                    O_NONBLOCK);
1132         if (result == -EINPROGRESS)
1133                 result = 0;
1134         if (result == 0)
1135                 goto out;
1136
1137 out_err:
1138         if (con->sock) {
1139                 sock_release(con->sock);
1140                 con->sock = NULL;
1141         } else if (sock) {
1142                 sock_release(sock);
1143         }
1144         /*
1145          * Some errors are fatal and this list might need adjusting. For other
1146          * errors we try again until the max number of retries is reached.
1147          */
1148         if (result != -EHOSTUNREACH &&
1149             result != -ENETUNREACH &&
1150             result != -ENETDOWN &&
1151             result != -EINVAL &&
1152             result != -EPROTONOSUPPORT) {
1153                 log_print("connect %d try %d error %d", con->nodeid,
1154                           con->retries, result);
1155                 mutex_unlock(&con->sock_mutex);
1156                 msleep(1000);
1157                 lowcomms_connect_sock(con);
1158                 return;
1159         }
1160 out:
1161         mutex_unlock(&con->sock_mutex);
1162         return;
1163 }
1164
1165 static struct socket *tcp_create_listen_sock(struct connection *con,
1166                                              struct sockaddr_storage *saddr)
1167 {
1168         struct socket *sock = NULL;
1169         int result = 0;
1170         int addr_len;
1171
1172         if (dlm_local_addr[0]->ss_family == AF_INET)
1173                 addr_len = sizeof(struct sockaddr_in);
1174         else
1175                 addr_len = sizeof(struct sockaddr_in6);
1176
1177         /* Create a socket to communicate with */
1178         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1179                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1180         if (result < 0) {
1181                 log_print("Can't create listening comms socket");
1182                 goto create_out;
1183         }
1184
1185         sock_set_mark(sock->sk, dlm_config.ci_mark);
1186
1187         /* Turn off Nagle's algorithm */
1188         tcp_sock_set_nodelay(sock->sk);
1189
1190         sock_set_reuseaddr(sock->sk);
1191
1192         write_lock_bh(&sock->sk->sk_callback_lock);
1193         sock->sk->sk_user_data = con;
1194         save_listen_callbacks(sock);
1195         con->rx_action = accept_from_sock;
1196         con->connect_action = tcp_connect_to_sock;
1197         write_unlock_bh(&sock->sk->sk_callback_lock);
1198
1199         /* Bind to our port */
1200         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1201         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1202         if (result < 0) {
1203                 log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1204                 sock_release(sock);
1205                 sock = NULL;
1206                 con->sock = NULL;
1207                 goto create_out;
1208         }
1209         sock_set_keepalive(sock->sk);
1210
1211         result = sock->ops->listen(sock, 5);
1212         if (result < 0) {
1213                 log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1214                 sock_release(sock);
1215                 sock = NULL;
1216                 goto create_out;
1217         }
1218
1219 create_out:
1220         return sock;
1221 }
1222
1223 /* Get local addresses */
1224 static void init_local(void)
1225 {
1226         struct sockaddr_storage sas, *addr;
1227         int i;
1228
1229         dlm_local_count = 0;
1230         for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1231                 if (dlm_our_addr(&sas, i))
1232                         break;
1233
1234                 addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1235                 if (!addr)
1236                         break;
1237                 dlm_local_addr[dlm_local_count++] = addr;
1238         }
1239 }
1240
1241 /* Initialise SCTP socket and bind to all interfaces */
1242 static int sctp_listen_for_all(void)
1243 {
1244         struct socket *sock = NULL;
1245         int result = -EINVAL;
1246         struct connection *con = nodeid2con(0, GFP_NOFS);
1247
1248         if (!con)
1249                 return -ENOMEM;
1250
1251         log_print("Using SCTP for communications");
1252
1253         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1254                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1255         if (result < 0) {
1256                 log_print("Can't create comms socket, check SCTP is loaded");
1257                 goto out;
1258         }
1259
1260         sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1261         sock_set_mark(sock->sk, dlm_config.ci_mark);
1262         sctp_sock_set_nodelay(sock->sk);
1263
1264         write_lock_bh(&sock->sk->sk_callback_lock);
1265         /* Init con struct */
1266         sock->sk->sk_user_data = con;
1267         save_listen_callbacks(sock);
1268         con->sock = sock;
1269         con->sock->sk->sk_data_ready = lowcomms_data_ready;
1270         con->rx_action = accept_from_sock;
1271         con->connect_action = sctp_connect_to_sock;
1272
1273         write_unlock_bh(&sock->sk->sk_callback_lock);
1274
1275         /* Bind to all addresses. */
1276         if (sctp_bind_addrs(con, dlm_config.ci_tcp_port))
1277                 goto create_delsock;
1278
1279         result = sock->ops->listen(sock, 5);
1280         if (result < 0) {
1281                 log_print("Can't set socket listening");
1282                 goto create_delsock;
1283         }
1284
1285         return 0;
1286
1287 create_delsock:
1288         sock_release(sock);
1289         con->sock = NULL;
1290 out:
1291         return result;
1292 }
1293
1294 static int tcp_listen_for_all(void)
1295 {
1296         struct socket *sock = NULL;
1297         struct connection *con = nodeid2con(0, GFP_NOFS);
1298         int result = -EINVAL;
1299
1300         if (!con)
1301                 return -ENOMEM;
1302
1303         /* We don't support multi-homed hosts */
1304         if (dlm_local_addr[1] != NULL) {
1305                 log_print("TCP protocol can't handle multi-homed hosts, "
1306                           "try SCTP");
1307                 return -EINVAL;
1308         }
1309
1310         log_print("Using TCP for communications");
1311
1312         sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
1313         if (sock) {
1314                 add_sock(sock, con);
1315                 result = 0;
1316         }
1317         else {
1318                 result = -EADDRINUSE;
1319         }
1320
1321         return result;
1322 }
1323
1324
1325
1326 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1327                                                      gfp_t allocation)
1328 {
1329         struct writequeue_entry *entry;
1330
1331         entry = kmalloc(sizeof(struct writequeue_entry), allocation);
1332         if (!entry)
1333                 return NULL;
1334
1335         entry->page = alloc_page(allocation);
1336         if (!entry->page) {
1337                 kfree(entry);
1338                 return NULL;
1339         }
1340
1341         entry->offset = 0;
1342         entry->len = 0;
1343         entry->end = 0;
1344         entry->users = 0;
1345         entry->con = con;
1346
1347         return entry;
1348 }
1349
1350 void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
1351 {
1352         struct connection *con;
1353         struct writequeue_entry *e;
1354         int offset = 0;
1355
1356         con = nodeid2con(nodeid, allocation);
1357         if (!con)
1358                 return NULL;
1359
1360         spin_lock(&con->writequeue_lock);
1361         e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
1362         if ((&e->list == &con->writequeue) ||
1363             (PAGE_SIZE - e->end < len)) {
1364                 e = NULL;
1365         } else {
1366                 offset = e->end;
1367                 e->end += len;
1368                 e->users++;
1369         }
1370         spin_unlock(&con->writequeue_lock);
1371
1372         if (e) {
1373         got_one:
1374                 *ppc = page_address(e->page) + offset;
1375                 return e;
1376         }
1377
1378         e = new_writequeue_entry(con, allocation);
1379         if (e) {
1380                 spin_lock(&con->writequeue_lock);
1381                 offset = e->end;
1382                 e->end += len;
1383                 e->users++;
1384                 list_add_tail(&e->list, &con->writequeue);
1385                 spin_unlock(&con->writequeue_lock);
1386                 goto got_one;
1387         }
1388         return NULL;
1389 }
1390
1391 void dlm_lowcomms_commit_buffer(void *mh)
1392 {
1393         struct writequeue_entry *e = (struct writequeue_entry *)mh;
1394         struct connection *con = e->con;
1395         int users;
1396
1397         spin_lock(&con->writequeue_lock);
1398         users = --e->users;
1399         if (users)
1400                 goto out;
1401         e->len = e->end - e->offset;
1402         spin_unlock(&con->writequeue_lock);
1403
1404         queue_work(send_workqueue, &con->swork);
1405         return;
1406
1407 out:
1408         spin_unlock(&con->writequeue_lock);
1409         return;
1410 }
1411
1412 /* Send a message */
1413 static void send_to_sock(struct connection *con)
1414 {
1415         int ret = 0;
1416         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1417         struct writequeue_entry *e;
1418         int len, offset;
1419         int count = 0;
1420
1421         mutex_lock(&con->sock_mutex);
1422         if (con->sock == NULL)
1423                 goto out_connect;
1424
1425         spin_lock(&con->writequeue_lock);
1426         for (;;) {
1427                 e = list_entry(con->writequeue.next, struct writequeue_entry,
1428                                list);
1429                 if ((struct list_head *) e == &con->writequeue)
1430                         break;
1431
1432                 len = e->len;
1433                 offset = e->offset;
1434                 BUG_ON(len == 0 && e->users == 0);
1435                 spin_unlock(&con->writequeue_lock);
1436
1437                 ret = 0;
1438                 if (len) {
1439                         ret = kernel_sendpage(con->sock, e->page, offset, len,
1440                                               msg_flags);
1441                         if (ret == -EAGAIN || ret == 0) {
1442                                 if (ret == -EAGAIN &&
1443                                     test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1444                                     !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1445                                         /* Notify TCP that we're limited by the
1446                                          * application window size.
1447                                          */
1448                                         set_bit(SOCK_NOSPACE, &con->sock->flags);
1449                                         con->sock->sk->sk_write_pending++;
1450                                 }
1451                                 cond_resched();
1452                                 goto out;
1453                         } else if (ret < 0)
1454                                 goto send_error;
1455                 }
1456
1457                 /* Don't starve people filling buffers */
1458                 if (++count >= MAX_SEND_MSG_COUNT) {
1459                         cond_resched();
1460                         count = 0;
1461                 }
1462
1463                 spin_lock(&con->writequeue_lock);
1464                 writequeue_entry_complete(e, ret);
1465         }
1466         spin_unlock(&con->writequeue_lock);
1467 out:
1468         mutex_unlock(&con->sock_mutex);
1469         return;
1470
1471 send_error:
1472         mutex_unlock(&con->sock_mutex);
1473         close_connection(con, false, false, true);
1474         /* Requeue the send work. When the work daemon runs again, it will try
1475            a new connection, then call this function again. */
1476         queue_work(send_workqueue, &con->swork);
1477         return;
1478
1479 out_connect:
1480         mutex_unlock(&con->sock_mutex);
1481         queue_work(send_workqueue, &con->swork);
1482         cond_resched();
1483 }
1484
1485 static void clean_one_writequeue(struct connection *con)
1486 {
1487         struct writequeue_entry *e, *safe;
1488
1489         spin_lock(&con->writequeue_lock);
1490         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1491                 list_del(&e->list);
1492                 free_entry(e);
1493         }
1494         spin_unlock(&con->writequeue_lock);
1495 }
1496
1497 /* Called from recovery when it knows that a node has
1498    left the cluster */
1499 int dlm_lowcomms_close(int nodeid)
1500 {
1501         struct connection *con;
1502         struct dlm_node_addr *na;
1503
1504         log_print("closing connection to node %d", nodeid);
1505         con = nodeid2con(nodeid, 0);
1506         if (con) {
1507                 set_bit(CF_CLOSE, &con->flags);
1508                 close_connection(con, true, true, true);
1509                 clean_one_writequeue(con);
1510         }
1511
1512         spin_lock(&dlm_node_addrs_spin);
1513         na = find_node_addr(nodeid);
1514         if (na) {
1515                 list_del(&na->list);
1516                 while (na->addr_count--)
1517                         kfree(na->addr[na->addr_count]);
1518                 kfree(na);
1519         }
1520         spin_unlock(&dlm_node_addrs_spin);
1521
1522         return 0;
1523 }
1524
1525 /* Receive workqueue function */
1526 static void process_recv_sockets(struct work_struct *work)
1527 {
1528         struct connection *con = container_of(work, struct connection, rwork);
1529         int err;
1530
1531         clear_bit(CF_READ_PENDING, &con->flags);
1532         do {
1533                 err = con->rx_action(con);
1534         } while (!err);
1535 }
1536
1537 /* Send workqueue function */
1538 static void process_send_sockets(struct work_struct *work)
1539 {
1540         struct connection *con = container_of(work, struct connection, swork);
1541
1542         clear_bit(CF_WRITE_PENDING, &con->flags);
1543         if (con->sock == NULL) /* not mutex protected so check it inside too */
1544                 con->connect_action(con);
1545         if (!list_empty(&con->writequeue))
1546                 send_to_sock(con);
1547 }
1548
1549
1550 /* Discard all entries on the write queues */
1551 static void clean_writequeues(void)
1552 {
1553         foreach_conn(clean_one_writequeue);
1554 }
1555
1556 static void work_stop(void)
1557 {
1558         if (recv_workqueue)
1559                 destroy_workqueue(recv_workqueue);
1560         if (send_workqueue)
1561                 destroy_workqueue(send_workqueue);
1562 }
1563
1564 static int work_start(void)
1565 {
1566         recv_workqueue = alloc_workqueue("dlm_recv",
1567                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1568         if (!recv_workqueue) {
1569                 log_print("can't start dlm_recv");
1570                 return -ENOMEM;
1571         }
1572
1573         send_workqueue = alloc_workqueue("dlm_send",
1574                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1575         if (!send_workqueue) {
1576                 log_print("can't start dlm_send");
1577                 destroy_workqueue(recv_workqueue);
1578                 return -ENOMEM;
1579         }
1580
1581         return 0;
1582 }
1583
1584 static void _stop_conn(struct connection *con, bool and_other)
1585 {
1586         mutex_lock(&con->sock_mutex);
1587         set_bit(CF_CLOSE, &con->flags);
1588         set_bit(CF_READ_PENDING, &con->flags);
1589         set_bit(CF_WRITE_PENDING, &con->flags);
1590         if (con->sock && con->sock->sk) {
1591                 write_lock_bh(&con->sock->sk->sk_callback_lock);
1592                 con->sock->sk->sk_user_data = NULL;
1593                 write_unlock_bh(&con->sock->sk->sk_callback_lock);
1594         }
1595         if (con->othercon && and_other)
1596                 _stop_conn(con->othercon, false);
1597         mutex_unlock(&con->sock_mutex);
1598 }
1599
1600 static void stop_conn(struct connection *con)
1601 {
1602         _stop_conn(con, true);
1603 }
1604
1605 static void shutdown_conn(struct connection *con)
1606 {
1607         if (con->shutdown_action)
1608                 con->shutdown_action(con);
1609 }
1610
1611 static void free_conn(struct connection *con)
1612 {
1613         close_connection(con, true, true, true);
1614         if (con->othercon)
1615                 kmem_cache_free(con_cache, con->othercon);
1616         hlist_del(&con->list);
1617         kmem_cache_free(con_cache, con);
1618 }
1619
1620 static void work_flush(void)
1621 {
1622         int ok;
1623         int i;
1624         struct hlist_node *n;
1625         struct connection *con;
1626
1627         if (recv_workqueue)
1628                 flush_workqueue(recv_workqueue);
1629         if (send_workqueue)
1630                 flush_workqueue(send_workqueue);
1631         do {
1632                 ok = 1;
1633                 foreach_conn(stop_conn);
1634                 if (recv_workqueue)
1635                         flush_workqueue(recv_workqueue);
1636                 if (send_workqueue)
1637                         flush_workqueue(send_workqueue);
1638                 for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1639                         hlist_for_each_entry_safe(con, n,
1640                                                   &connection_hash[i], list) {
1641                                 ok &= test_bit(CF_READ_PENDING, &con->flags);
1642                                 ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1643                                 if (con->othercon) {
1644                                         ok &= test_bit(CF_READ_PENDING,
1645                                                        &con->othercon->flags);
1646                                         ok &= test_bit(CF_WRITE_PENDING,
1647                                                        &con->othercon->flags);
1648                                 }
1649                         }
1650                 }
1651         } while (!ok);
1652 }
1653
1654 void dlm_lowcomms_stop(void)
1655 {
1656         /* Set all the flags to prevent any
1657            socket activity.
1658         */
1659         mutex_lock(&connections_lock);
1660         dlm_allow_conn = 0;
1661         mutex_unlock(&connections_lock);
1662         foreach_conn(shutdown_conn);
1663         work_flush();
1664         clean_writequeues();
1665         foreach_conn(free_conn);
1666         work_stop();
1667
1668         kmem_cache_destroy(con_cache);
1669 }
1670
1671 int dlm_lowcomms_start(void)
1672 {
1673         int error = -EINVAL;
1674         struct connection *con;
1675         int i;
1676
1677         for (i = 0; i < CONN_HASH_SIZE; i++)
1678                 INIT_HLIST_HEAD(&connection_hash[i]);
1679
1680         init_local();
1681         if (!dlm_local_count) {
1682                 error = -ENOTCONN;
1683                 log_print("no local IP address has been set");
1684                 goto fail;
1685         }
1686
1687         error = -ENOMEM;
1688         con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
1689                                       __alignof__(struct connection), 0,
1690                                       NULL);
1691         if (!con_cache)
1692                 goto fail;
1693
1694         error = work_start();
1695         if (error)
1696                 goto fail_destroy;
1697
1698         dlm_allow_conn = 1;
1699
1700         /* Start listening */
1701         if (dlm_config.ci_protocol == 0)
1702                 error = tcp_listen_for_all();
1703         else
1704                 error = sctp_listen_for_all();
1705         if (error)
1706                 goto fail_unlisten;
1707
1708         return 0;
1709
1710 fail_unlisten:
1711         dlm_allow_conn = 0;
1712         con = nodeid2con(0,0);
1713         if (con) {
1714                 close_connection(con, false, true, true);
1715                 kmem_cache_free(con_cache, con);
1716         }
1717 fail_destroy:
1718         kmem_cache_destroy(con_cache);
1719 fail:
1720         return error;
1721 }
1722
1723 void dlm_lowcomms_exit(void)
1724 {
1725         struct dlm_node_addr *na, *safe;
1726
1727         spin_lock(&dlm_node_addrs_spin);
1728         list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
1729                 list_del(&na->list);
1730                 while (na->addr_count--)
1731                         kfree(na->addr[na->addr_count]);
1732                 kfree(na);
1733         }
1734         spin_unlock(&dlm_node_addrs_spin);
1735 }