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 connection {
  69         struct socket *sock;    /* NULL if not connected */
  70         uint32_t nodeid;        /* So we know who we are in the list */
  71         struct mutex sock_mutex;
  72         unsigned long flags;
  73 #define CF_READ_PENDING 1
  74 #define CF_WRITE_PENDING 2
  75 #define CF_INIT_PENDING 4
  76 #define CF_IS_OTHERCON 5
  77 #define CF_CLOSE 6
  78 #define CF_APP_LIMITED 7
  79 #define CF_CLOSING 8
  80 #define CF_SHUTDOWN 9
  81         struct list_head writequeue;  /* List of outgoing writequeue_entries */
  82         spinlock_t writequeue_lock;
  83         int (*rx_action) (struct connection *); /* What to do when active */
  84         void (*connect_action) (struct connection *);   /* What to do to connect */
  85         void (*shutdown_action)(struct connection *con); /* What to do to shutdown */
  86         int retries;
  87 #define MAX_CONNECT_RETRIES 3
  88         struct hlist_node list;
  89         struct connection *othercon;
  90         struct work_struct rwork; /* Receive workqueue */
  91         struct work_struct swork; /* Send workqueue */
  92         wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */
  93         unsigned char *rx_buf;
  94         int rx_buflen;
  95         int rx_leftover;
  96         struct rcu_head rcu;
  97 };
  98 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
  99
 100 /* An entry waiting to be sent */
 101 struct writequeue_entry {
 102         struct list_head list;
 103         struct page *page;
 104         int offset;
 105         int len;
 106         int end;
 107         int users;
 108         struct connection *con;
 109 };
 110
 111 struct dlm_node_addr {
 112         struct list_head list;
 113         int nodeid;
 114         int addr_count;
 115         int curr_addr_index;
 116         struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
 117 };
 118
 119 static struct listen_sock_callbacks {
 120         void (*sk_error_report)(struct sock *);
 121         void (*sk_data_ready)(struct sock *);
 122         void (*sk_state_change)(struct sock *);
 123         void (*sk_write_space)(struct sock *);
 124 } listen_sock;
 125
 126 static LIST_HEAD(dlm_node_addrs);
 127 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
 128
 129 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 130 static int dlm_local_count;
 131 static int dlm_allow_conn;
 132
 133 /* Work queues */
 134 static struct workqueue_struct *recv_workqueue;
 135 static struct workqueue_struct *send_workqueue;
 136
 137 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 138 static DEFINE_SPINLOCK(connections_lock);
 139 DEFINE_STATIC_SRCU(connections_srcu);
 140
 141 static void process_recv_sockets(struct work_struct *work);
 142 static void process_send_sockets(struct work_struct *work);
 143
 144
 145 /* This is deliberately very simple because most clusters have simple
 146    sequential nodeids, so we should be able to go straight to a connection
 147    struct in the array */
 148 static inline int nodeid_hash(int nodeid)
 149 {
 150         return nodeid & (CONN_HASH_SIZE-1);
 151 }
 152
 153 static struct connection *__find_con(int nodeid)
 154 {
 155         int r, idx;
 156         struct connection *con;
 157
 158         r = nodeid_hash(nodeid);
 159
 160         idx = srcu_read_lock(&connections_srcu);
 161         hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
 162                 if (con->nodeid == nodeid) {
 163                         srcu_read_unlock(&connections_srcu, idx);
 164                         return con;
 165                 }
 166         }
 167         srcu_read_unlock(&connections_srcu, idx);
 168
 169         return NULL;
 170 }
 171
 172 /*
 173  * If 'allocation' is zero then we don't attempt to create a new
 174  * connection structure for this node.
 175  */
 176 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
 177 {
 178         struct connection *con, *tmp;
 179         int r;
 180
 181         con = __find_con(nodeid);
 182         if (con || !alloc)
 183                 return con;
 184
 185         con = kzalloc(sizeof(*con), alloc);
 186         if (!con)
 187                 return NULL;
 188
 189         con->rx_buflen = dlm_config.ci_buffer_size;
 190         con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS);
 191         if (!con->rx_buf) {
 192                 kfree(con);
 193                 return NULL;
 194         }
 195
 196         con->nodeid = nodeid;
 197         mutex_init(&con->sock_mutex);
 198         INIT_LIST_HEAD(&con->writequeue);
 199         spin_lock_init(&con->writequeue_lock);
 200         INIT_WORK(&con->swork, process_send_sockets);
 201         INIT_WORK(&con->rwork, process_recv_sockets);
 202         init_waitqueue_head(&con->shutdown_wait);
 203
 204         /* Setup action pointers for child sockets */
 205         if (con->nodeid) {
 206                 struct connection *zerocon = __find_con(0);
 207
 208                 con->connect_action = zerocon->connect_action;
 209                 if (!con->rx_action)
 210                         con->rx_action = zerocon->rx_action;
 211         }
 212
 213         r = nodeid_hash(nodeid);
 214
 215         spin_lock(&connections_lock);
 216         /* Because multiple workqueues/threads calls this function it can
 217          * race on multiple cpu's. Instead of locking hot path __find_con()
 218          * we just check in rare cases of recently added nodes again
 219          * under protection of connections_lock. If this is the case we
 220          * abort our connection creation and return the existing connection.
 221          */
 222         tmp = __find_con(nodeid);
 223         if (tmp) {
 224                 spin_unlock(&connections_lock);
 225                 kfree(con->rx_buf);
 226                 kfree(con);
 227                 return tmp;
 228         }
 229
 230         hlist_add_head_rcu(&con->list, &connection_hash[r]);
 231         spin_unlock(&connections_lock);
 232
 233         return con;
 234 }
 235
 236 /* Loop round all connections */
 237 static void foreach_conn(void (*conn_func)(struct connection *c))
 238 {
 239         int i, idx;
 240         struct connection *con;
 241
 242         idx = srcu_read_lock(&connections_srcu);
 243         for (i = 0; i < CONN_HASH_SIZE; i++) {
 244                 hlist_for_each_entry_rcu(con, &connection_hash[i], list)
 245                         conn_func(con);
 246         }
 247         srcu_read_unlock(&connections_srcu, idx);
 248 }
 249
 250 static struct dlm_node_addr *find_node_addr(int nodeid)
 251 {
 252         struct dlm_node_addr *na;
 253
 254         list_for_each_entry(na, &dlm_node_addrs, list) {
 255                 if (na->nodeid == nodeid)
 256                         return na;
 257         }
 258         return NULL;
 259 }
 260
 261 static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y)
 262 {
 263         switch (x->ss_family) {
 264         case AF_INET: {
 265                 struct sockaddr_in *sinx = (struct sockaddr_in *)x;
 266                 struct sockaddr_in *siny = (struct sockaddr_in *)y;
 267                 if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
 268                         return 0;
 269                 if (sinx->sin_port != siny->sin_port)
 270                         return 0;
 271                 break;
 272         }
 273         case AF_INET6: {
 274                 struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
 275                 struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
 276                 if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
 277                         return 0;
 278                 if (sinx->sin6_port != siny->sin6_port)
 279                         return 0;
 280                 break;
 281         }
 282         default:
 283                 return 0;
 284         }
 285         return 1;
 286 }
 287
 288 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
 289                           struct sockaddr *sa_out, bool try_new_addr)
 290 {
 291         struct sockaddr_storage sas;
 292         struct dlm_node_addr *na;
 293
 294         if (!dlm_local_count)
 295                 return -1;
 296
 297         spin_lock(&dlm_node_addrs_spin);
 298         na = find_node_addr(nodeid);
 299         if (na && na->addr_count) {
 300                 memcpy(&sas, na->addr[na->curr_addr_index],
 301                        sizeof(struct sockaddr_storage));
 302
 303                 if (try_new_addr) {
 304                         na->curr_addr_index++;
 305                         if (na->curr_addr_index == na->addr_count)
 306                                 na->curr_addr_index = 0;
 307                 }
 308         }
 309         spin_unlock(&dlm_node_addrs_spin);
 310
 311         if (!na)
 312                 return -EEXIST;
 313
 314         if (!na->addr_count)
 315                 return -ENOENT;
 316
 317         if (sas_out)
 318                 memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
 319
 320         if (!sa_out)
 321                 return 0;
 322
 323         if (dlm_local_addr[0]->ss_family == AF_INET) {
 324                 struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
 325                 struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
 326                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 327         } else {
 328                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
 329                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
 330                 ret6->sin6_addr = in6->sin6_addr;
 331         }
 332
 333         return 0;
 334 }
 335
 336 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid)
 337 {
 338         struct dlm_node_addr *na;
 339         int rv = -EEXIST;
 340         int addr_i;
 341
 342         spin_lock(&dlm_node_addrs_spin);
 343         list_for_each_entry(na, &dlm_node_addrs, list) {
 344                 if (!na->addr_count)
 345                         continue;
 346
 347                 for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
 348                         if (addr_compare(na->addr[addr_i], addr)) {
 349                                 *nodeid = na->nodeid;
 350                                 rv = 0;
 351                                 goto unlock;
 352                         }
 353                 }
 354         }
 355 unlock:
 356         spin_unlock(&dlm_node_addrs_spin);
 357         return rv;
 358 }
 359
 360 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 361 {
 362         struct sockaddr_storage *new_addr;
 363         struct dlm_node_addr *new_node, *na;
 364
 365         new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
 366         if (!new_node)
 367                 return -ENOMEM;
 368
 369         new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
 370         if (!new_addr) {
 371                 kfree(new_node);
 372                 return -ENOMEM;
 373         }
 374
 375         memcpy(new_addr, addr, len);
 376
 377         spin_lock(&dlm_node_addrs_spin);
 378         na = find_node_addr(nodeid);
 379         if (!na) {
 380                 new_node->nodeid = nodeid;
 381                 new_node->addr[0] = new_addr;
 382                 new_node->addr_count = 1;
 383                 list_add(&new_node->list, &dlm_node_addrs);
 384                 spin_unlock(&dlm_node_addrs_spin);
 385                 return 0;
 386         }
 387
 388         if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
 389                 spin_unlock(&dlm_node_addrs_spin);
 390                 kfree(new_addr);
 391                 kfree(new_node);
 392                 return -ENOSPC;
 393         }
 394
 395         na->addr[na->addr_count++] = new_addr;
 396         spin_unlock(&dlm_node_addrs_spin);
 397         kfree(new_node);
 398         return 0;
 399 }
 400
 401 /* Data available on socket or listen socket received a connect */
 402 static void lowcomms_data_ready(struct sock *sk)
 403 {
 404         struct connection *con;
 405
 406         read_lock_bh(&sk->sk_callback_lock);
 407         con = sock2con(sk);
 408         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 409                 queue_work(recv_workqueue, &con->rwork);
 410         read_unlock_bh(&sk->sk_callback_lock);
 411 }
 412
 413 static void lowcomms_write_space(struct sock *sk)
 414 {
 415         struct connection *con;
 416
 417         read_lock_bh(&sk->sk_callback_lock);
 418         con = sock2con(sk);
 419         if (!con)
 420                 goto out;
 421
 422         clear_bit(SOCK_NOSPACE, &con->sock->flags);
 423
 424         if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
 425                 con->sock->sk->sk_write_pending--;
 426                 clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
 427         }
 428
 429         queue_work(send_workqueue, &con->swork);
 430 out:
 431         read_unlock_bh(&sk->sk_callback_lock);
 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         void (*orig_report)(struct sock *) = NULL;
 475         struct inet_sock *inet;
 476
 477         read_lock_bh(&sk->sk_callback_lock);
 478         con = sock2con(sk);
 479         if (con == NULL)
 480                 goto out;
 481
 482         orig_report = listen_sock.sk_error_report;
 483
 484         inet = inet_sk(sk);
 485         switch (sk->sk_family) {
 486         case AF_INET:
 487                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 488                                    "sending to node %d at %pI4, dport %d, "
 489                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 490                                    con->nodeid, &inet->inet_daddr,
 491                                    ntohs(inet->inet_dport), sk->sk_err,
 492                                    sk->sk_err_soft);
 493                 break;
 494 #if IS_ENABLED(CONFIG_IPV6)
 495         case AF_INET6:
 496                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 497                                    "sending to node %d at %pI6c, "
 498                                    "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(),
 499                                    con->nodeid, &sk->sk_v6_daddr,
 500                                    ntohs(inet->inet_dport), sk->sk_err,
 501                                    sk->sk_err_soft);
 502                 break;
 503 #endif
 504         default:
 505                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 506                                    "invalid socket family %d set, "
 507                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 508                                    sk->sk_family, sk->sk_err, sk->sk_err_soft);
 509                 goto out;
 510         }
 511 out:
 512         read_unlock_bh(&sk->sk_callback_lock);
 513         if (orig_report)
 514                 orig_report(sk);
 515 }
 516
 517 /* Note: sk_callback_lock must be locked before calling this function. */
 518 static void save_listen_callbacks(struct socket *sock)
 519 {
 520         struct sock *sk = sock->sk;
 521
 522         listen_sock.sk_data_ready = sk->sk_data_ready;
 523         listen_sock.sk_state_change = sk->sk_state_change;
 524         listen_sock.sk_write_space = sk->sk_write_space;
 525         listen_sock.sk_error_report = sk->sk_error_report;
 526 }
 527
 528 static void restore_callbacks(struct socket *sock)
 529 {
 530         struct sock *sk = sock->sk;
 531
 532         write_lock_bh(&sk->sk_callback_lock);
 533         sk->sk_user_data = NULL;
 534         sk->sk_data_ready = listen_sock.sk_data_ready;
 535         sk->sk_state_change = listen_sock.sk_state_change;
 536         sk->sk_write_space = listen_sock.sk_write_space;
 537         sk->sk_error_report = listen_sock.sk_error_report;
 538         write_unlock_bh(&sk->sk_callback_lock);
 539 }
 540
 541 /* Make a socket active */
 542 static void add_sock(struct socket *sock, struct connection *con)
 543 {
 544         struct sock *sk = sock->sk;
 545
 546         write_lock_bh(&sk->sk_callback_lock);
 547         con->sock = sock;
 548
 549         sk->sk_user_data = con;
 550         /* Install a data_ready callback */
 551         sk->sk_data_ready = lowcomms_data_ready;
 552         sk->sk_write_space = lowcomms_write_space;
 553         sk->sk_state_change = lowcomms_state_change;
 554         sk->sk_allocation = GFP_NOFS;
 555         sk->sk_error_report = lowcomms_error_report;
 556         write_unlock_bh(&sk->sk_callback_lock);
 557 }
 558
 559 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 560    length */
 561 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 562                           int *addr_len)
 563 {
 564         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 565         if (saddr->ss_family == AF_INET) {
 566                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 567                 in4_addr->sin_port = cpu_to_be16(port);
 568                 *addr_len = sizeof(struct sockaddr_in);
 569                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 570         } else {
 571                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 572                 in6_addr->sin6_port = cpu_to_be16(port);
 573                 *addr_len = sizeof(struct sockaddr_in6);
 574         }
 575         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 576 }
 577
 578 /* Close a remote connection and tidy up */
 579 static void close_connection(struct connection *con, bool and_other,
 580                              bool tx, bool rx)
 581 {
 582         bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
 583
 584         if (tx && !closing && cancel_work_sync(&con->swork)) {
 585                 log_print("canceled swork for node %d", con->nodeid);
 586                 clear_bit(CF_WRITE_PENDING, &con->flags);
 587         }
 588         if (rx && !closing && cancel_work_sync(&con->rwork)) {
 589                 log_print("canceled rwork for node %d", con->nodeid);
 590                 clear_bit(CF_READ_PENDING, &con->flags);
 591         }
 592
 593         mutex_lock(&con->sock_mutex);
 594         if (con->sock) {
 595                 restore_callbacks(con->sock);
 596                 sock_release(con->sock);
 597                 con->sock = NULL;
 598         }
 599         if (con->othercon && and_other) {
 600                 /* Will only re-enter once. */
 601                 close_connection(con->othercon, false, tx, rx);
 602         }
 603
 604         con->rx_leftover = 0;
 605         con->retries = 0;
 606         mutex_unlock(&con->sock_mutex);
 607         clear_bit(CF_CLOSING, &con->flags);
 608 }
 609
 610 static void shutdown_connection(struct connection *con)
 611 {
 612         int ret;
 613
 614         flush_work(&con->swork);
 615
 616         mutex_lock(&con->sock_mutex);
 617         /* nothing to shutdown */
 618         if (!con->sock) {
 619                 mutex_unlock(&con->sock_mutex);
 620                 return;
 621         }
 622
 623         set_bit(CF_SHUTDOWN, &con->flags);
 624         ret = kernel_sock_shutdown(con->sock, SHUT_WR);
 625         mutex_unlock(&con->sock_mutex);
 626         if (ret) {
 627                 log_print("Connection %p failed to shutdown: %d will force close",
 628                           con, ret);
 629                 goto force_close;
 630         } else {
 631                 ret = wait_event_timeout(con->shutdown_wait,
 632                                          !test_bit(CF_SHUTDOWN, &con->flags),
 633                                          DLM_SHUTDOWN_WAIT_TIMEOUT);
 634                 if (ret == 0) {
 635                         log_print("Connection %p shutdown timed out, will force close",
 636                                   con);
 637                         goto force_close;
 638                 }
 639         }
 640
 641         return;
 642
 643 force_close:
 644         clear_bit(CF_SHUTDOWN, &con->flags);
 645         close_connection(con, false, true, true);
 646 }
 647
 648 static void dlm_tcp_shutdown(struct connection *con)
 649 {
 650         if (con->othercon)
 651                 shutdown_connection(con->othercon);
 652         shutdown_connection(con);
 653 }
 654
 655 static int con_realloc_receive_buf(struct connection *con, int newlen)
 656 {
 657         unsigned char *newbuf;
 658
 659         newbuf = kmalloc(newlen, GFP_NOFS);
 660         if (!newbuf)
 661                 return -ENOMEM;
 662
 663         /* copy any leftover from last receive */
 664         if (con->rx_leftover)
 665                 memmove(newbuf, con->rx_buf, con->rx_leftover);
 666
 667         /* swap to new buffer space */
 668         kfree(con->rx_buf);
 669         con->rx_buflen = newlen;
 670         con->rx_buf = newbuf;
 671
 672         return 0;
 673 }
 674
 675 /* Data received from remote end */
 676 static int receive_from_sock(struct connection *con)
 677 {
 678         int call_again_soon = 0;
 679         struct msghdr msg;
 680         struct kvec iov;
 681         int ret, buflen;
 682
 683         mutex_lock(&con->sock_mutex);
 684
 685         if (con->sock == NULL) {
 686                 ret = -EAGAIN;
 687                 goto out_close;
 688         }
 689
 690         if (con->nodeid == 0) {
 691                 ret = -EINVAL;
 692                 goto out_close;
 693         }
 694
 695         /* realloc if we get new buffer size to read out */
 696         buflen = dlm_config.ci_buffer_size;
 697         if (con->rx_buflen != buflen && con->rx_leftover <= buflen) {
 698                 ret = con_realloc_receive_buf(con, buflen);
 699                 if (ret < 0)
 700                         goto out_resched;
 701         }
 702
 703         /* calculate new buffer parameter regarding last receive and
 704          * possible leftover bytes
 705          */
 706         iov.iov_base = con->rx_buf + con->rx_leftover;
 707         iov.iov_len = con->rx_buflen - con->rx_leftover;
 708
 709         memset(&msg, 0, sizeof(msg));
 710         msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
 711         ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
 712                              msg.msg_flags);
 713         if (ret <= 0)
 714                 goto out_close;
 715         else if (ret == iov.iov_len)
 716                 call_again_soon = 1;
 717
 718         /* new buflen according readed bytes and leftover from last receive */
 719         buflen = ret + con->rx_leftover;
 720         ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen);
 721         if (ret < 0)
 722                 goto out_close;
 723
 724         /* calculate leftover bytes from process and put it into begin of
 725          * the receive buffer, so next receive we have the full message
 726          * at the start address of the receive buffer.
 727          */
 728         con->rx_leftover = buflen - ret;
 729         if (con->rx_leftover) {
 730                 memmove(con->rx_buf, con->rx_buf + ret,
 731                         con->rx_leftover);
 732                 call_again_soon = true;
 733         }
 734
 735         if (call_again_soon)
 736                 goto out_resched;
 737
 738         mutex_unlock(&con->sock_mutex);
 739         return 0;
 740
 741 out_resched:
 742         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 743                 queue_work(recv_workqueue, &con->rwork);
 744         mutex_unlock(&con->sock_mutex);
 745         return -EAGAIN;
 746
 747 out_close:
 748         mutex_unlock(&con->sock_mutex);
 749         if (ret != -EAGAIN) {
 750                 /* Reconnect when there is something to send */
 751                 close_connection(con, false, true, false);
 752                 if (ret == 0) {
 753                         log_print("connection %p got EOF from %d",
 754                                   con, con->nodeid);
 755                         /* handling for tcp shutdown */
 756                         clear_bit(CF_SHUTDOWN, &con->flags);
 757                         wake_up(&con->shutdown_wait);
 758                         /* signal to breaking receive worker */
 759                         ret = -1;
 760                 }
 761         }
 762         return ret;
 763 }
 764
 765 /* Listening socket is busy, accept a connection */
 766 static int accept_from_sock(struct connection *con)
 767 {
 768         int result;
 769         struct sockaddr_storage peeraddr;
 770         struct socket *newsock;
 771         int len;
 772         int nodeid;
 773         struct connection *newcon;
 774         struct connection *addcon;
 775         unsigned int mark;
 776
 777         if (!dlm_allow_conn) {
 778                 return -1;
 779         }
 780
 781         mutex_lock_nested(&con->sock_mutex, 0);
 782
 783         if (!con->sock) {
 784                 mutex_unlock(&con->sock_mutex);
 785                 return -ENOTCONN;
 786         }
 787
 788         result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
 789         if (result < 0)
 790                 goto accept_err;
 791
 792         /* Get the connected socket's peer */
 793         memset(&peeraddr, 0, sizeof(peeraddr));
 794         len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
 795         if (len < 0) {
 796                 result = -ECONNABORTED;
 797                 goto accept_err;
 798         }
 799
 800         /* Get the new node's NODEID */
 801         make_sockaddr(&peeraddr, 0, &len);
 802         if (addr_to_nodeid(&peeraddr, &nodeid)) {
 803                 unsigned char *b=(unsigned char *)&peeraddr;
 804                 log_print("connect from non cluster node");
 805                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
 806                                      b, sizeof(struct sockaddr_storage));
 807                 sock_release(newsock);
 808                 mutex_unlock(&con->sock_mutex);
 809                 return -1;
 810         }
 811
 812         dlm_comm_mark(nodeid, &mark);
 813         sock_set_mark(newsock->sk, mark);
 814
 815         log_print("got connection from %d", nodeid);
 816
 817         /*  Check to see if we already have a connection to this node. This
 818          *  could happen if the two nodes initiate a connection at roughly
 819          *  the same time and the connections cross on the wire.
 820          *  In this case we store the incoming one in "othercon"
 821          */
 822         newcon = nodeid2con(nodeid, GFP_NOFS);
 823         if (!newcon) {
 824                 result = -ENOMEM;
 825                 goto accept_err;
 826         }
 827         mutex_lock_nested(&newcon->sock_mutex, 1);
 828         if (newcon->sock) {
 829                 struct connection *othercon = newcon->othercon;
 830
 831                 if (!othercon) {
 832                         othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
 833                         if (!othercon) {
 834                                 log_print("failed to allocate incoming socket");
 835                                 mutex_unlock(&newcon->sock_mutex);
 836                                 result = -ENOMEM;
 837                                 goto accept_err;
 838                         }
 839
 840                         othercon->rx_buflen = dlm_config.ci_buffer_size;
 841                         othercon->rx_buf = kmalloc(othercon->rx_buflen, GFP_NOFS);
 842                         if (!othercon->rx_buf) {
 843                                 mutex_unlock(&newcon->sock_mutex);
 844                                 kfree(othercon);
 845                                 log_print("failed to allocate incoming socket receive buffer");
 846                                 result = -ENOMEM;
 847                                 goto accept_err;
 848                         }
 849
 850                         othercon->nodeid = nodeid;
 851                         othercon->rx_action = receive_from_sock;
 852                         mutex_init(&othercon->sock_mutex);
 853                         INIT_LIST_HEAD(&othercon->writequeue);
 854                         spin_lock_init(&othercon->writequeue_lock);
 855                         INIT_WORK(&othercon->swork, process_send_sockets);
 856                         INIT_WORK(&othercon->rwork, process_recv_sockets);
 857                         init_waitqueue_head(&othercon->shutdown_wait);
 858                         set_bit(CF_IS_OTHERCON, &othercon->flags);
 859                 } else {
 860                         /* close other sock con if we have something new */
 861                         close_connection(othercon, false, true, false);
 862                 }
 863
 864                 mutex_lock_nested(&othercon->sock_mutex, 2);
 865                 newcon->othercon = othercon;
 866                 add_sock(newsock, othercon);
 867                 addcon = othercon;
 868                 mutex_unlock(&othercon->sock_mutex);
 869         }
 870         else {
 871                 newcon->rx_action = receive_from_sock;
 872                 /* accept copies the sk after we've saved the callbacks, so we
 873                    don't want to save them a second time or comm errors will
 874                    result in calling sk_error_report recursively. */
 875                 add_sock(newsock, newcon);
 876                 addcon = newcon;
 877         }
 878
 879         mutex_unlock(&newcon->sock_mutex);
 880
 881         /*
 882          * Add it to the active queue in case we got data
 883          * between processing the accept adding the socket
 884          * to the read_sockets list
 885          */
 886         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
 887                 queue_work(recv_workqueue, &addcon->rwork);
 888         mutex_unlock(&con->sock_mutex);
 889
 890         return 0;
 891
 892 accept_err:
 893         mutex_unlock(&con->sock_mutex);
 894         if (newsock)
 895                 sock_release(newsock);
 896
 897         if (result != -EAGAIN)
 898                 log_print("error accepting connection from node: %d", result);
 899         return result;
 900 }
 901
 902 static void free_entry(struct writequeue_entry *e)
 903 {
 904         __free_page(e->page);
 905         kfree(e);
 906 }
 907
 908 /*
 909  * writequeue_entry_complete - try to delete and free write queue entry
 910  * @e: write queue entry to try to delete
 911  * @completed: bytes completed
 912  *
 913  * writequeue_lock must be held.
 914  */
 915 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
 916 {
 917         e->offset += completed;
 918         e->len -= completed;
 919
 920         if (e->len == 0 && e->users == 0) {
 921                 list_del(&e->list);
 922                 free_entry(e);
 923         }
 924 }
 925
 926 /*
 927  * sctp_bind_addrs - bind a SCTP socket to all our addresses
 928  */
 929 static int sctp_bind_addrs(struct connection *con, uint16_t port)
 930 {
 931         struct sockaddr_storage localaddr;
 932         struct sockaddr *addr = (struct sockaddr *)&localaddr;
 933         int i, addr_len, result = 0;
 934
 935         for (i = 0; i < dlm_local_count; i++) {
 936                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
 937                 make_sockaddr(&localaddr, port, &addr_len);
 938
 939                 if (!i)
 940                         result = kernel_bind(con->sock, addr, addr_len);
 941                 else
 942                         result = sock_bind_add(con->sock->sk, addr, addr_len);
 943
 944                 if (result < 0) {
 945                         log_print("Can't bind to %d addr number %d, %d.\n",
 946                                   port, i + 1, result);
 947                         break;
 948                 }
 949         }
 950         return result;
 951 }
 952
 953 /* Initiate an SCTP association.
 954    This is a special case of send_to_sock() in that we don't yet have a
 955    peeled-off socket for this association, so we use the listening socket
 956    and add the primary IP address of the remote node.
 957  */
 958 static void sctp_connect_to_sock(struct connection *con)
 959 {
 960         struct sockaddr_storage daddr;
 961         int result;
 962         int addr_len;
 963         struct socket *sock;
 964         unsigned int mark;
 965
 966         if (con->nodeid == 0) {
 967                 log_print("attempt to connect sock 0 foiled");
 968                 return;
 969         }
 970
 971         dlm_comm_mark(con->nodeid, &mark);
 972
 973         mutex_lock(&con->sock_mutex);
 974
 975         /* Some odd races can cause double-connects, ignore them */
 976         if (con->retries++ > MAX_CONNECT_RETRIES)
 977                 goto out;
 978
 979         if (con->sock) {
 980                 log_print("node %d already connected.", con->nodeid);
 981                 goto out;
 982         }
 983
 984         memset(&daddr, 0, sizeof(daddr));
 985         result = nodeid_to_addr(con->nodeid, &daddr, NULL, true);
 986         if (result < 0) {
 987                 log_print("no address for nodeid %d", con->nodeid);
 988                 goto out;
 989         }
 990
 991         /* Create a socket to communicate with */
 992         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
 993                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
 994         if (result < 0)
 995                 goto socket_err;
 996
 997         sock_set_mark(sock->sk, mark);
 998
 999         con->rx_action = receive_from_sock;
1000         con->connect_action = sctp_connect_to_sock;
1001         add_sock(sock, con);
1002
1003         /* Bind to all addresses. */
1004         if (sctp_bind_addrs(con, 0))
1005                 goto bind_err;
1006
1007         make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len);
1008
1009         log_print("connecting to %d", con->nodeid);
1010
1011         /* Turn off Nagle's algorithm */
1012         sctp_sock_set_nodelay(sock->sk);
1013
1014         /*
1015          * Make sock->ops->connect() function return in specified time,
1016          * since O_NONBLOCK argument in connect() function does not work here,
1017          * then, we should restore the default value of this attribute.
1018          */
1019         sock_set_sndtimeo(sock->sk, 5);
1020         result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len,
1021                                    0);
1022         sock_set_sndtimeo(sock->sk, 0);
1023
1024         if (result == -EINPROGRESS)
1025                 result = 0;
1026         if (result == 0)
1027                 goto out;
1028
1029 bind_err:
1030         con->sock = NULL;
1031         sock_release(sock);
1032
1033 socket_err:
1034         /*
1035          * Some errors are fatal and this list might need adjusting. For other
1036          * errors we try again until the max number of retries is reached.
1037          */
1038         if (result != -EHOSTUNREACH &&
1039             result != -ENETUNREACH &&
1040             result != -ENETDOWN &&
1041             result != -EINVAL &&
1042             result != -EPROTONOSUPPORT) {
1043                 log_print("connect %d try %d error %d", con->nodeid,
1044                           con->retries, result);
1045                 mutex_unlock(&con->sock_mutex);
1046                 msleep(1000);
1047                 lowcomms_connect_sock(con);
1048                 return;
1049         }
1050
1051 out:
1052         mutex_unlock(&con->sock_mutex);
1053 }
1054
1055 /* Connect a new socket to its peer */
1056 static void tcp_connect_to_sock(struct connection *con)
1057 {
1058         struct sockaddr_storage saddr, src_addr;
1059         int addr_len;
1060         struct socket *sock = NULL;
1061         unsigned int mark;
1062         int result;
1063
1064         if (con->nodeid == 0) {
1065                 log_print("attempt to connect sock 0 foiled");
1066                 return;
1067         }
1068
1069         dlm_comm_mark(con->nodeid, &mark);
1070
1071         mutex_lock(&con->sock_mutex);
1072         if (con->retries++ > MAX_CONNECT_RETRIES)
1073                 goto out;
1074
1075         /* Some odd races can cause double-connects, ignore them */
1076         if (con->sock)
1077                 goto out;
1078
1079         /* Create a socket to communicate with */
1080         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1081                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1082         if (result < 0)
1083                 goto out_err;
1084
1085         sock_set_mark(sock->sk, mark);
1086
1087         memset(&saddr, 0, sizeof(saddr));
1088         result = nodeid_to_addr(con->nodeid, &saddr, NULL, false);
1089         if (result < 0) {
1090                 log_print("no address for nodeid %d", con->nodeid);
1091                 goto out_err;
1092         }
1093
1094         con->rx_action = receive_from_sock;
1095         con->connect_action = tcp_connect_to_sock;
1096         con->shutdown_action = dlm_tcp_shutdown;
1097         add_sock(sock, con);
1098
1099         /* Bind to our cluster-known address connecting to avoid
1100            routing problems */
1101         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1102         make_sockaddr(&src_addr, 0, &addr_len);
1103         result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1104                                  addr_len);
1105         if (result < 0) {
1106                 log_print("could not bind for connect: %d", result);
1107                 /* This *may* not indicate a critical error */
1108         }
1109
1110         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1111
1112         log_print("connecting to %d", con->nodeid);
1113
1114         /* Turn off Nagle's algorithm */
1115         tcp_sock_set_nodelay(sock->sk);
1116
1117         result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1118                                    O_NONBLOCK);
1119         if (result == -EINPROGRESS)
1120                 result = 0;
1121         if (result == 0)
1122                 goto out;
1123
1124 out_err:
1125         if (con->sock) {
1126                 sock_release(con->sock);
1127                 con->sock = NULL;
1128         } else if (sock) {
1129                 sock_release(sock);
1130         }
1131         /*
1132          * Some errors are fatal and this list might need adjusting. For other
1133          * errors we try again until the max number of retries is reached.
1134          */
1135         if (result != -EHOSTUNREACH &&
1136             result != -ENETUNREACH &&
1137             result != -ENETDOWN &&
1138             result != -EINVAL &&
1139             result != -EPROTONOSUPPORT) {
1140                 log_print("connect %d try %d error %d", con->nodeid,
1141                           con->retries, result);
1142                 mutex_unlock(&con->sock_mutex);
1143                 msleep(1000);
1144                 lowcomms_connect_sock(con);
1145                 return;
1146         }
1147 out:
1148         mutex_unlock(&con->sock_mutex);
1149         return;
1150 }
1151
1152 static struct socket *tcp_create_listen_sock(struct connection *con,
1153                                              struct sockaddr_storage *saddr)
1154 {
1155         struct socket *sock = NULL;
1156         int result = 0;
1157         int addr_len;
1158
1159         if (dlm_local_addr[0]->ss_family == AF_INET)
1160                 addr_len = sizeof(struct sockaddr_in);
1161         else
1162                 addr_len = sizeof(struct sockaddr_in6);
1163
1164         /* Create a socket to communicate with */
1165         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1166                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1167         if (result < 0) {
1168                 log_print("Can't create listening comms socket");
1169                 goto create_out;
1170         }
1171
1172         sock_set_mark(sock->sk, dlm_config.ci_mark);
1173
1174         /* Turn off Nagle's algorithm */
1175         tcp_sock_set_nodelay(sock->sk);
1176
1177         sock_set_reuseaddr(sock->sk);
1178
1179         write_lock_bh(&sock->sk->sk_callback_lock);
1180         sock->sk->sk_user_data = con;
1181         save_listen_callbacks(sock);
1182         con->rx_action = accept_from_sock;
1183         con->connect_action = tcp_connect_to_sock;
1184         write_unlock_bh(&sock->sk->sk_callback_lock);
1185
1186         /* Bind to our port */
1187         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1188         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1189         if (result < 0) {
1190                 log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1191                 sock_release(sock);
1192                 sock = NULL;
1193                 con->sock = NULL;
1194                 goto create_out;
1195         }
1196         sock_set_keepalive(sock->sk);
1197
1198         result = sock->ops->listen(sock, 5);
1199         if (result < 0) {
1200                 log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1201                 sock_release(sock);
1202                 sock = NULL;
1203                 goto create_out;
1204         }
1205
1206 create_out:
1207         return sock;
1208 }
1209
1210 /* Get local addresses */
1211 static void init_local(void)
1212 {
1213         struct sockaddr_storage sas, *addr;
1214         int i;
1215
1216         dlm_local_count = 0;
1217         for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1218                 if (dlm_our_addr(&sas, i))
1219                         break;
1220
1221                 addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1222                 if (!addr)
1223                         break;
1224                 dlm_local_addr[dlm_local_count++] = addr;
1225         }
1226 }
1227
1228 static void deinit_local(void)
1229 {
1230         int i;
1231
1232         for (i = 0; i < dlm_local_count; i++)
1233                 kfree(dlm_local_addr[i]);
1234 }
1235
1236 /* Initialise SCTP socket and bind to all interfaces */
1237 static int sctp_listen_for_all(void)
1238 {
1239         struct socket *sock = NULL;
1240         int result = -EINVAL;
1241         struct connection *con = nodeid2con(0, GFP_NOFS);
1242
1243         if (!con)
1244                 return -ENOMEM;
1245
1246         log_print("Using SCTP for communications");
1247
1248         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1249                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1250         if (result < 0) {
1251                 log_print("Can't create comms socket, check SCTP is loaded");
1252                 goto out;
1253         }
1254
1255         sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1256         sock_set_mark(sock->sk, dlm_config.ci_mark);
1257         sctp_sock_set_nodelay(sock->sk);
1258
1259         write_lock_bh(&sock->sk->sk_callback_lock);
1260         /* Init con struct */
1261         sock->sk->sk_user_data = con;
1262         save_listen_callbacks(sock);
1263         con->sock = sock;
1264         con->sock->sk->sk_data_ready = lowcomms_data_ready;
1265         con->rx_action = accept_from_sock;
1266         con->connect_action = sctp_connect_to_sock;
1267
1268         write_unlock_bh(&sock->sk->sk_callback_lock);
1269
1270         /* Bind to all addresses. */
1271         if (sctp_bind_addrs(con, dlm_config.ci_tcp_port))
1272                 goto create_delsock;
1273
1274         result = sock->ops->listen(sock, 5);
1275         if (result < 0) {
1276                 log_print("Can't set socket listening");
1277                 goto create_delsock;
1278         }
1279
1280         return 0;
1281
1282 create_delsock:
1283         sock_release(sock);
1284         con->sock = NULL;
1285 out:
1286         return result;
1287 }
1288
1289 static int tcp_listen_for_all(void)
1290 {
1291         struct socket *sock = NULL;
1292         struct connection *con = nodeid2con(0, GFP_NOFS);
1293         int result = -EINVAL;
1294
1295         if (!con)
1296                 return -ENOMEM;
1297
1298         /* We don't support multi-homed hosts */
1299         if (dlm_local_addr[1] != NULL) {
1300                 log_print("TCP protocol can't handle multi-homed hosts, "
1301                           "try SCTP");
1302                 return -EINVAL;
1303         }
1304
1305         log_print("Using TCP for communications");
1306
1307         sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
1308         if (sock) {
1309                 add_sock(sock, con);
1310                 result = 0;
1311         }
1312         else {
1313                 result = -EADDRINUSE;
1314         }
1315
1316         return result;
1317 }
1318
1319
1320
1321 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1322                                                      gfp_t allocation)
1323 {
1324         struct writequeue_entry *entry;
1325
1326         entry = kmalloc(sizeof(struct writequeue_entry), allocation);
1327         if (!entry)
1328                 return NULL;
1329
1330         entry->page = alloc_page(allocation);
1331         if (!entry->page) {
1332                 kfree(entry);
1333                 return NULL;
1334         }
1335
1336         entry->offset = 0;
1337         entry->len = 0;
1338         entry->end = 0;
1339         entry->users = 0;
1340         entry->con = con;
1341
1342         return entry;
1343 }
1344
1345 void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
1346 {
1347         struct connection *con;
1348         struct writequeue_entry *e;
1349         int offset = 0;
1350
1351         con = nodeid2con(nodeid, allocation);
1352         if (!con)
1353                 return NULL;
1354
1355         spin_lock(&con->writequeue_lock);
1356         e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
1357         if ((&e->list == &con->writequeue) ||
1358             (PAGE_SIZE - e->end < len)) {
1359                 e = NULL;
1360         } else {
1361                 offset = e->end;
1362                 e->end += len;
1363                 e->users++;
1364         }
1365         spin_unlock(&con->writequeue_lock);
1366
1367         if (e) {
1368         got_one:
1369                 *ppc = page_address(e->page) + offset;
1370                 return e;
1371         }
1372
1373         e = new_writequeue_entry(con, allocation);
1374         if (e) {
1375                 spin_lock(&con->writequeue_lock);
1376                 offset = e->end;
1377                 e->end += len;
1378                 e->users++;
1379                 list_add_tail(&e->list, &con->writequeue);
1380                 spin_unlock(&con->writequeue_lock);
1381                 goto got_one;
1382         }
1383         return NULL;
1384 }
1385
1386 void dlm_lowcomms_commit_buffer(void *mh)
1387 {
1388         struct writequeue_entry *e = (struct writequeue_entry *)mh;
1389         struct connection *con = e->con;
1390         int users;
1391
1392         spin_lock(&con->writequeue_lock);
1393         users = --e->users;
1394         if (users)
1395                 goto out;
1396         e->len = e->end - e->offset;
1397         spin_unlock(&con->writequeue_lock);
1398
1399         queue_work(send_workqueue, &con->swork);
1400         return;
1401
1402 out:
1403         spin_unlock(&con->writequeue_lock);
1404         return;
1405 }
1406
1407 /* Send a message */
1408 static void send_to_sock(struct connection *con)
1409 {
1410         int ret = 0;
1411         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1412         struct writequeue_entry *e;
1413         int len, offset;
1414         int count = 0;
1415
1416         mutex_lock(&con->sock_mutex);
1417         if (con->sock == NULL)
1418                 goto out_connect;
1419
1420         spin_lock(&con->writequeue_lock);
1421         for (;;) {
1422                 e = list_entry(con->writequeue.next, struct writequeue_entry,
1423                                list);
1424                 if ((struct list_head *) e == &con->writequeue)
1425                         break;
1426
1427                 len = e->len;
1428                 offset = e->offset;
1429                 BUG_ON(len == 0 && e->users == 0);
1430                 spin_unlock(&con->writequeue_lock);
1431
1432                 ret = 0;
1433                 if (len) {
1434                         ret = kernel_sendpage(con->sock, e->page, offset, len,
1435                                               msg_flags);
1436                         if (ret == -EAGAIN || ret == 0) {
1437                                 if (ret == -EAGAIN &&
1438                                     test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1439                                     !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1440                                         /* Notify TCP that we're limited by the
1441                                          * application window size.
1442                                          */
1443                                         set_bit(SOCK_NOSPACE, &con->sock->flags);
1444                                         con->sock->sk->sk_write_pending++;
1445                                 }
1446                                 cond_resched();
1447                                 goto out;
1448                         } else if (ret < 0)
1449                                 goto send_error;
1450                 }
1451
1452                 /* Don't starve people filling buffers */
1453                 if (++count >= MAX_SEND_MSG_COUNT) {
1454                         cond_resched();
1455                         count = 0;
1456                 }
1457
1458                 spin_lock(&con->writequeue_lock);
1459                 writequeue_entry_complete(e, ret);
1460         }
1461         spin_unlock(&con->writequeue_lock);
1462 out:
1463         mutex_unlock(&con->sock_mutex);
1464         return;
1465
1466 send_error:
1467         mutex_unlock(&con->sock_mutex);
1468         close_connection(con, false, false, true);
1469         /* Requeue the send work. When the work daemon runs again, it will try
1470            a new connection, then call this function again. */
1471         queue_work(send_workqueue, &con->swork);
1472         return;
1473
1474 out_connect:
1475         mutex_unlock(&con->sock_mutex);
1476         queue_work(send_workqueue, &con->swork);
1477         cond_resched();
1478 }
1479
1480 static void clean_one_writequeue(struct connection *con)
1481 {
1482         struct writequeue_entry *e, *safe;
1483
1484         spin_lock(&con->writequeue_lock);
1485         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1486                 list_del(&e->list);
1487                 free_entry(e);
1488         }
1489         spin_unlock(&con->writequeue_lock);
1490 }
1491
1492 /* Called from recovery when it knows that a node has
1493    left the cluster */
1494 int dlm_lowcomms_close(int nodeid)
1495 {
1496         struct connection *con;
1497         struct dlm_node_addr *na;
1498
1499         log_print("closing connection to node %d", nodeid);
1500         con = nodeid2con(nodeid, 0);
1501         if (con) {
1502                 set_bit(CF_CLOSE, &con->flags);
1503                 close_connection(con, true, true, true);
1504                 clean_one_writequeue(con);
1505         }
1506
1507         spin_lock(&dlm_node_addrs_spin);
1508         na = find_node_addr(nodeid);
1509         if (na) {
1510                 list_del(&na->list);
1511                 while (na->addr_count--)
1512                         kfree(na->addr[na->addr_count]);
1513                 kfree(na);
1514         }
1515         spin_unlock(&dlm_node_addrs_spin);
1516
1517         return 0;
1518 }
1519
1520 /* Receive workqueue function */
1521 static void process_recv_sockets(struct work_struct *work)
1522 {
1523         struct connection *con = container_of(work, struct connection, rwork);
1524         int err;
1525
1526         clear_bit(CF_READ_PENDING, &con->flags);
1527         do {
1528                 err = con->rx_action(con);
1529         } while (!err);
1530 }
1531
1532 /* Send workqueue function */
1533 static void process_send_sockets(struct work_struct *work)
1534 {
1535         struct connection *con = container_of(work, struct connection, swork);
1536
1537         clear_bit(CF_WRITE_PENDING, &con->flags);
1538         if (con->sock == NULL) /* not mutex protected so check it inside too */
1539                 con->connect_action(con);
1540         if (!list_empty(&con->writequeue))
1541                 send_to_sock(con);
1542 }
1543
1544 static void work_stop(void)
1545 {
1546         if (recv_workqueue)
1547                 destroy_workqueue(recv_workqueue);
1548         if (send_workqueue)
1549                 destroy_workqueue(send_workqueue);
1550 }
1551
1552 static int work_start(void)
1553 {
1554         recv_workqueue = alloc_workqueue("dlm_recv",
1555                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1556         if (!recv_workqueue) {
1557                 log_print("can't start dlm_recv");
1558                 return -ENOMEM;
1559         }
1560
1561         send_workqueue = alloc_workqueue("dlm_send",
1562                                          WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1563         if (!send_workqueue) {
1564                 log_print("can't start dlm_send");
1565                 destroy_workqueue(recv_workqueue);
1566                 return -ENOMEM;
1567         }
1568
1569         return 0;
1570 }
1571
1572 static void _stop_conn(struct connection *con, bool and_other)
1573 {
1574         mutex_lock(&con->sock_mutex);
1575         set_bit(CF_CLOSE, &con->flags);
1576         set_bit(CF_READ_PENDING, &con->flags);
1577         set_bit(CF_WRITE_PENDING, &con->flags);
1578         if (con->sock && con->sock->sk) {
1579                 write_lock_bh(&con->sock->sk->sk_callback_lock);
1580                 con->sock->sk->sk_user_data = NULL;
1581                 write_unlock_bh(&con->sock->sk->sk_callback_lock);
1582         }
1583         if (con->othercon && and_other)
1584                 _stop_conn(con->othercon, false);
1585         mutex_unlock(&con->sock_mutex);
1586 }
1587
1588 static void stop_conn(struct connection *con)
1589 {
1590         _stop_conn(con, true);
1591 }
1592
1593 static void shutdown_conn(struct connection *con)
1594 {
1595         if (con->shutdown_action)
1596                 con->shutdown_action(con);
1597 }
1598
1599 static void connection_release(struct rcu_head *rcu)
1600 {
1601         struct connection *con = container_of(rcu, struct connection, rcu);
1602
1603         kfree(con->rx_buf);
1604         kfree(con);
1605 }
1606
1607 static void free_conn(struct connection *con)
1608 {
1609         close_connection(con, true, true, true);
1610         spin_lock(&connections_lock);
1611         hlist_del_rcu(&con->list);
1612         spin_unlock(&connections_lock);
1613         if (con->othercon) {
1614                 clean_one_writequeue(con->othercon);
1615                 call_rcu(&con->othercon->rcu, connection_release);
1616         }
1617         clean_one_writequeue(con);
1618         call_rcu(&con->rcu, connection_release);
1619 }
1620
1621 static void work_flush(void)
1622 {
1623         int ok, idx;
1624         int i;
1625         struct connection *con;
1626
1627         do {
1628                 ok = 1;
1629                 foreach_conn(stop_conn);
1630                 if (recv_workqueue)
1631                         flush_workqueue(recv_workqueue);
1632                 if (send_workqueue)
1633                         flush_workqueue(send_workqueue);
1634                 idx = srcu_read_lock(&connections_srcu);
1635                 for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1636                         hlist_for_each_entry_rcu(con, &connection_hash[i],
1637                                                  list) {
1638                                 ok &= test_bit(CF_READ_PENDING, &con->flags);
1639                                 ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1640                                 if (con->othercon) {
1641                                         ok &= test_bit(CF_READ_PENDING,
1642                                                        &con->othercon->flags);
1643                                         ok &= test_bit(CF_WRITE_PENDING,
1644                                                        &con->othercon->flags);
1645                                 }
1646                         }
1647                 }
1648                 srcu_read_unlock(&connections_srcu, idx);
1649         } while (!ok);
1650 }
1651
1652 void dlm_lowcomms_stop(void)
1653 {
1654         /* Set all the flags to prevent any
1655            socket activity.
1656         */
1657         dlm_allow_conn = 0;
1658
1659         if (recv_workqueue)
1660                 flush_workqueue(recv_workqueue);
1661         if (send_workqueue)
1662                 flush_workqueue(send_workqueue);
1663
1664         foreach_conn(shutdown_conn);
1665         work_flush();
1666         foreach_conn(free_conn);
1667         work_stop();
1668         deinit_local();
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 = work_start();
1688         if (error)
1689                 goto fail;
1690
1691         dlm_allow_conn = 1;
1692
1693         /* Start listening */
1694         if (dlm_config.ci_protocol == 0)
1695                 error = tcp_listen_for_all();
1696         else
1697                 error = sctp_listen_for_all();
1698         if (error)
1699                 goto fail_unlisten;
1700
1701         return 0;
1702
1703 fail_unlisten:
1704         dlm_allow_conn = 0;
1705         con = nodeid2con(0,0);
1706         if (con)
1707                 free_conn(con);
1708 fail:
1709         return error;
1710 }
1711
1712 void dlm_lowcomms_exit(void)
1713 {
1714         struct dlm_node_addr *na, *safe;
1715
1716         spin_lock(&dlm_node_addrs_spin);
1717         list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
1718                 list_del(&na->list);
1719                 while (na->addr_count--)
1720                         kfree(na->addr[na->addr_count]);
1721                 kfree(na);
1722         }
1723         spin_unlock(&dlm_node_addrs_spin);
1724 }