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
  63 /* Number of messages to send before rescheduling */
  64 #define MAX_SEND_MSG_COUNT 25
  65 #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(10000)
  66
  67 struct connection {
  68         struct socket *sock;    /* NULL if not connected */
  69         uint32_t nodeid;        /* So we know who we are in the list */
  70         struct mutex sock_mutex;
  71         unsigned long flags;
  72 #define CF_READ_PENDING 1
  73 #define CF_WRITE_PENDING 2
  74 #define CF_INIT_PENDING 4
  75 #define CF_IS_OTHERCON 5
  76 #define CF_CLOSE 6
  77 #define CF_APP_LIMITED 7
  78 #define CF_CLOSING 8
  79 #define CF_SHUTDOWN 9
  80 #define CF_CONNECTED 10
  81 #define CF_RECONNECT 11
  82 #define CF_DELAY_CONNECT 12
  83 #define CF_EOF 13
  84         struct list_head writequeue;  /* List of outgoing writequeue_entries */
  85         spinlock_t writequeue_lock;
  86         atomic_t writequeue_cnt;
  87         void (*connect_action) (struct connection *);   /* What to do to connect */
  88         void (*shutdown_action)(struct connection *con); /* What to do to shutdown */
  89         bool (*eof_condition)(struct connection *con); /* What to do to eof check */
  90         int retries;
  91 #define MAX_CONNECT_RETRIES 3
  92         struct hlist_node list;
  93         struct connection *othercon;
  94         struct connection *sendcon;
  95         struct work_struct rwork; /* Receive workqueue */
  96         struct work_struct swork; /* Send workqueue */
  97         wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */
  98         unsigned char *rx_buf;
  99         int rx_buflen;
 100         int rx_leftover;
 101         struct rcu_head rcu;
 102 };
 103 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 104
 105 struct listen_connection {
 106         struct socket *sock;
 107         struct work_struct rwork;
 108 };
 109
 110 #define DLM_WQ_REMAIN_BYTES(e) (PAGE_SIZE - e->end)
 111 #define DLM_WQ_LENGTH_BYTES(e) (e->end - e->offset)
 112
 113 /* An entry waiting to be sent */
 114 struct writequeue_entry {
 115         struct list_head list;
 116         struct page *page;
 117         int offset;
 118         int len;
 119         int end;
 120         int users;
 121         bool dirty;
 122         struct connection *con;
 123         struct list_head msgs;
 124         struct kref ref;
 125 };
 126
 127 struct dlm_msg {
 128         struct writequeue_entry *entry;
 129         struct dlm_msg *orig_msg;
 130         bool retransmit;
 131         void *ppc;
 132         int len;
 133         int idx; /* new()/commit() idx exchange */
 134
 135         struct list_head list;
 136         struct kref ref;
 137 };
 138
 139 struct dlm_node_addr {
 140         struct list_head list;
 141         int nodeid;
 142         int mark;
 143         int addr_count;
 144         int curr_addr_index;
 145         struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
 146 };
 147
 148 static struct listen_sock_callbacks {
 149         void (*sk_error_report)(struct sock *);
 150         void (*sk_data_ready)(struct sock *);
 151         void (*sk_state_change)(struct sock *);
 152         void (*sk_write_space)(struct sock *);
 153 } listen_sock;
 154
 155 static LIST_HEAD(dlm_node_addrs);
 156 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
 157
 158 static struct listen_connection listen_con;
 159 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 160 static int dlm_local_count;
 161 int dlm_allow_conn;
 162
 163 /* Work queues */
 164 static struct workqueue_struct *recv_workqueue;
 165 static struct workqueue_struct *send_workqueue;
 166
 167 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 168 static DEFINE_SPINLOCK(connections_lock);
 169 DEFINE_STATIC_SRCU(connections_srcu);
 170
 171 static void process_recv_sockets(struct work_struct *work);
 172 static void process_send_sockets(struct work_struct *work);
 173
 174 static void sctp_connect_to_sock(struct connection *con);
 175 static void tcp_connect_to_sock(struct connection *con);
 176 static void dlm_tcp_shutdown(struct connection *con);
 177
 178 /* need to held writequeue_lock */
 179 static struct writequeue_entry *con_next_wq(struct connection *con)
 180 {
 181         struct writequeue_entry *e;
 182
 183         if (list_empty(&con->writequeue))
 184                 return NULL;
 185
 186         e = list_first_entry(&con->writequeue, struct writequeue_entry,
 187                              list);
 188         if (e->len == 0)
 189                 return NULL;
 190
 191         return e;
 192 }
 193
 194 static struct connection *__find_con(int nodeid, int r)
 195 {
 196         struct connection *con;
 197
 198         hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
 199                 if (con->nodeid == nodeid)
 200                         return con;
 201         }
 202
 203         return NULL;
 204 }
 205
 206 static bool tcp_eof_condition(struct connection *con)
 207 {
 208         return atomic_read(&con->writequeue_cnt);
 209 }
 210
 211 static int dlm_con_init(struct connection *con, int nodeid)
 212 {
 213         con->rx_buflen = dlm_config.ci_buffer_size;
 214         con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS);
 215         if (!con->rx_buf)
 216                 return -ENOMEM;
 217
 218         con->nodeid = nodeid;
 219         mutex_init(&con->sock_mutex);
 220         INIT_LIST_HEAD(&con->writequeue);
 221         spin_lock_init(&con->writequeue_lock);
 222         atomic_set(&con->writequeue_cnt, 0);
 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         switch (dlm_config.ci_protocol) {
 228         case DLM_PROTO_TCP:
 229                 con->connect_action = tcp_connect_to_sock;
 230                 con->shutdown_action = dlm_tcp_shutdown;
 231                 con->eof_condition = tcp_eof_condition;
 232                 break;
 233         case DLM_PROTO_SCTP:
 234                 con->connect_action = sctp_connect_to_sock;
 235                 break;
 236         default:
 237                 kfree(con->rx_buf);
 238                 return -EINVAL;
 239         }
 240
 241         return 0;
 242 }
 243
 244 /*
 245  * If 'allocation' is zero then we don't attempt to create a new
 246  * connection structure for this node.
 247  */
 248 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
 249 {
 250         struct connection *con, *tmp;
 251         int r, ret;
 252
 253         r = nodeid_hash(nodeid);
 254         con = __find_con(nodeid, r);
 255         if (con || !alloc)
 256                 return con;
 257
 258         con = kzalloc(sizeof(*con), alloc);
 259         if (!con)
 260                 return NULL;
 261
 262         ret = dlm_con_init(con, nodeid);
 263         if (ret) {
 264                 kfree(con);
 265                 return NULL;
 266         }
 267
 268         spin_lock(&connections_lock);
 269         /* Because multiple workqueues/threads calls this function it can
 270          * race on multiple cpu's. Instead of locking hot path __find_con()
 271          * we just check in rare cases of recently added nodes again
 272          * under protection of connections_lock. If this is the case we
 273          * abort our connection creation and return the existing connection.
 274          */
 275         tmp = __find_con(nodeid, r);
 276         if (tmp) {
 277                 spin_unlock(&connections_lock);
 278                 kfree(con->rx_buf);
 279                 kfree(con);
 280                 return tmp;
 281         }
 282
 283         hlist_add_head_rcu(&con->list, &connection_hash[r]);
 284         spin_unlock(&connections_lock);
 285
 286         return con;
 287 }
 288
 289 /* Loop round all connections */
 290 static void foreach_conn(void (*conn_func)(struct connection *c))
 291 {
 292         int i;
 293         struct connection *con;
 294
 295         for (i = 0; i < CONN_HASH_SIZE; i++) {
 296                 hlist_for_each_entry_rcu(con, &connection_hash[i], list)
 297                         conn_func(con);
 298         }
 299 }
 300
 301 static struct dlm_node_addr *find_node_addr(int nodeid)
 302 {
 303         struct dlm_node_addr *na;
 304
 305         list_for_each_entry(na, &dlm_node_addrs, list) {
 306                 if (na->nodeid == nodeid)
 307                         return na;
 308         }
 309         return NULL;
 310 }
 311
 312 static int addr_compare(const struct sockaddr_storage *x,
 313                         const struct sockaddr_storage *y)
 314 {
 315         switch (x->ss_family) {
 316         case AF_INET: {
 317                 struct sockaddr_in *sinx = (struct sockaddr_in *)x;
 318                 struct sockaddr_in *siny = (struct sockaddr_in *)y;
 319                 if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
 320                         return 0;
 321                 if (sinx->sin_port != siny->sin_port)
 322                         return 0;
 323                 break;
 324         }
 325         case AF_INET6: {
 326                 struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
 327                 struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
 328                 if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
 329                         return 0;
 330                 if (sinx->sin6_port != siny->sin6_port)
 331                         return 0;
 332                 break;
 333         }
 334         default:
 335                 return 0;
 336         }
 337         return 1;
 338 }
 339
 340 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
 341                           struct sockaddr *sa_out, bool try_new_addr,
 342                           unsigned int *mark)
 343 {
 344         struct sockaddr_storage sas;
 345         struct dlm_node_addr *na;
 346
 347         if (!dlm_local_count)
 348                 return -1;
 349
 350         spin_lock(&dlm_node_addrs_spin);
 351         na = find_node_addr(nodeid);
 352         if (na && na->addr_count) {
 353                 memcpy(&sas, na->addr[na->curr_addr_index],
 354                        sizeof(struct sockaddr_storage));
 355
 356                 if (try_new_addr) {
 357                         na->curr_addr_index++;
 358                         if (na->curr_addr_index == na->addr_count)
 359                                 na->curr_addr_index = 0;
 360                 }
 361         }
 362         spin_unlock(&dlm_node_addrs_spin);
 363
 364         if (!na)
 365                 return -EEXIST;
 366
 367         if (!na->addr_count)
 368                 return -ENOENT;
 369
 370         *mark = na->mark;
 371
 372         if (sas_out)
 373                 memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
 374
 375         if (!sa_out)
 376                 return 0;
 377
 378         if (dlm_local_addr[0]->ss_family == AF_INET) {
 379                 struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
 380                 struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
 381                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 382         } else {
 383                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
 384                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
 385                 ret6->sin6_addr = in6->sin6_addr;
 386         }
 387
 388         return 0;
 389 }
 390
 391 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
 392                           unsigned int *mark)
 393 {
 394         struct dlm_node_addr *na;
 395         int rv = -EEXIST;
 396         int addr_i;
 397
 398         spin_lock(&dlm_node_addrs_spin);
 399         list_for_each_entry(na, &dlm_node_addrs, list) {
 400                 if (!na->addr_count)
 401                         continue;
 402
 403                 for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
 404                         if (addr_compare(na->addr[addr_i], addr)) {
 405                                 *nodeid = na->nodeid;
 406                                 *mark = na->mark;
 407                                 rv = 0;
 408                                 goto unlock;
 409                         }
 410                 }
 411         }
 412 unlock:
 413         spin_unlock(&dlm_node_addrs_spin);
 414         return rv;
 415 }
 416
 417 /* caller need to held dlm_node_addrs_spin lock */
 418 static bool dlm_lowcomms_na_has_addr(const struct dlm_node_addr *na,
 419                                      const struct sockaddr_storage *addr)
 420 {
 421         int i;
 422
 423         for (i = 0; i < na->addr_count; i++) {
 424                 if (addr_compare(na->addr[i], addr))
 425                         return true;
 426         }
 427
 428         return false;
 429 }
 430
 431 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 432 {
 433         struct sockaddr_storage *new_addr;
 434         struct dlm_node_addr *new_node, *na;
 435         bool ret;
 436
 437         new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
 438         if (!new_node)
 439                 return -ENOMEM;
 440
 441         new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
 442         if (!new_addr) {
 443                 kfree(new_node);
 444                 return -ENOMEM;
 445         }
 446
 447         memcpy(new_addr, addr, len);
 448
 449         spin_lock(&dlm_node_addrs_spin);
 450         na = find_node_addr(nodeid);
 451         if (!na) {
 452                 new_node->nodeid = nodeid;
 453                 new_node->addr[0] = new_addr;
 454                 new_node->addr_count = 1;
 455                 new_node->mark = dlm_config.ci_mark;
 456                 list_add(&new_node->list, &dlm_node_addrs);
 457                 spin_unlock(&dlm_node_addrs_spin);
 458                 return 0;
 459         }
 460
 461         ret = dlm_lowcomms_na_has_addr(na, addr);
 462         if (ret) {
 463                 spin_unlock(&dlm_node_addrs_spin);
 464                 kfree(new_addr);
 465                 kfree(new_node);
 466                 return -EEXIST;
 467         }
 468
 469         if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
 470                 spin_unlock(&dlm_node_addrs_spin);
 471                 kfree(new_addr);
 472                 kfree(new_node);
 473                 return -ENOSPC;
 474         }
 475
 476         na->addr[na->addr_count++] = new_addr;
 477         spin_unlock(&dlm_node_addrs_spin);
 478         kfree(new_node);
 479         return 0;
 480 }
 481
 482 /* Data available on socket or listen socket received a connect */
 483 static void lowcomms_data_ready(struct sock *sk)
 484 {
 485         struct connection *con;
 486
 487         read_lock_bh(&sk->sk_callback_lock);
 488         con = sock2con(sk);
 489         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 490                 queue_work(recv_workqueue, &con->rwork);
 491         read_unlock_bh(&sk->sk_callback_lock);
 492 }
 493
 494 static void lowcomms_listen_data_ready(struct sock *sk)
 495 {
 496         if (!dlm_allow_conn)
 497                 return;
 498
 499         queue_work(recv_workqueue, &listen_con.rwork);
 500 }
 501
 502 static void lowcomms_write_space(struct sock *sk)
 503 {
 504         struct connection *con;
 505
 506         read_lock_bh(&sk->sk_callback_lock);
 507         con = sock2con(sk);
 508         if (!con)
 509                 goto out;
 510
 511         if (!test_and_set_bit(CF_CONNECTED, &con->flags)) {
 512                 log_print("successful connected to node %d", con->nodeid);
 513                 queue_work(send_workqueue, &con->swork);
 514                 goto out;
 515         }
 516
 517         clear_bit(SOCK_NOSPACE, &con->sock->flags);
 518
 519         if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
 520                 con->sock->sk->sk_write_pending--;
 521                 clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
 522         }
 523
 524         queue_work(send_workqueue, &con->swork);
 525 out:
 526         read_unlock_bh(&sk->sk_callback_lock);
 527 }
 528
 529 static inline void lowcomms_connect_sock(struct connection *con)
 530 {
 531         if (test_bit(CF_CLOSE, &con->flags))
 532                 return;
 533         queue_work(send_workqueue, &con->swork);
 534         cond_resched();
 535 }
 536
 537 static void lowcomms_state_change(struct sock *sk)
 538 {
 539         /* SCTP layer is not calling sk_data_ready when the connection
 540          * is done, so we catch the signal through here. Also, it
 541          * doesn't switch socket state when entering shutdown, so we
 542          * skip the write in that case.
 543          */
 544         if (sk->sk_shutdown) {
 545                 if (sk->sk_shutdown == RCV_SHUTDOWN)
 546                         lowcomms_data_ready(sk);
 547         } else if (sk->sk_state == TCP_ESTABLISHED) {
 548                 lowcomms_write_space(sk);
 549         }
 550 }
 551
 552 int dlm_lowcomms_connect_node(int nodeid)
 553 {
 554         struct connection *con;
 555         int idx;
 556
 557         if (nodeid == dlm_our_nodeid())
 558                 return 0;
 559
 560         idx = srcu_read_lock(&connections_srcu);
 561         con = nodeid2con(nodeid, GFP_NOFS);
 562         if (!con) {
 563                 srcu_read_unlock(&connections_srcu, idx);
 564                 return -ENOMEM;
 565         }
 566
 567         lowcomms_connect_sock(con);
 568         srcu_read_unlock(&connections_srcu, idx);
 569
 570         return 0;
 571 }
 572
 573 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
 574 {
 575         struct dlm_node_addr *na;
 576
 577         spin_lock(&dlm_node_addrs_spin);
 578         na = find_node_addr(nodeid);
 579         if (!na) {
 580                 spin_unlock(&dlm_node_addrs_spin);
 581                 return -ENOENT;
 582         }
 583
 584         na->mark = mark;
 585         spin_unlock(&dlm_node_addrs_spin);
 586
 587         return 0;
 588 }
 589
 590 static void lowcomms_error_report(struct sock *sk)
 591 {
 592         struct connection *con;
 593         struct sockaddr_storage saddr;
 594         void (*orig_report)(struct sock *) = NULL;
 595
 596         read_lock_bh(&sk->sk_callback_lock);
 597         con = sock2con(sk);
 598         if (con == NULL)
 599                 goto out;
 600
 601         orig_report = listen_sock.sk_error_report;
 602         if (kernel_getpeername(sk->sk_socket, (struct sockaddr *)&saddr) < 0) {
 603                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 604                                    "sending to node %d, port %d, "
 605                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 606                                    con->nodeid, dlm_config.ci_tcp_port,
 607                                    sk->sk_err, sk->sk_err_soft);
 608         } else if (saddr.ss_family == AF_INET) {
 609                 struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr;
 610
 611                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 612                                    "sending to node %d at %pI4, port %d, "
 613                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 614                                    con->nodeid, &sin4->sin_addr.s_addr,
 615                                    dlm_config.ci_tcp_port, sk->sk_err,
 616                                    sk->sk_err_soft);
 617         } else {
 618                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr;
 619
 620                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 621                                    "sending to node %d at %u.%u.%u.%u, "
 622                                    "port %d, sk_err=%d/%d\n", dlm_our_nodeid(),
 623                                    con->nodeid, sin6->sin6_addr.s6_addr32[0],
 624                                    sin6->sin6_addr.s6_addr32[1],
 625                                    sin6->sin6_addr.s6_addr32[2],
 626                                    sin6->sin6_addr.s6_addr32[3],
 627                                    dlm_config.ci_tcp_port, sk->sk_err,
 628                                    sk->sk_err_soft);
 629         }
 630
 631         /* below sendcon only handling */
 632         if (test_bit(CF_IS_OTHERCON, &con->flags))
 633                 con = con->sendcon;
 634
 635         switch (sk->sk_err) {
 636         case ECONNREFUSED:
 637                 set_bit(CF_DELAY_CONNECT, &con->flags);
 638                 break;
 639         default:
 640                 break;
 641         }
 642
 643         if (!test_and_set_bit(CF_RECONNECT, &con->flags))
 644                 queue_work(send_workqueue, &con->swork);
 645
 646 out:
 647         read_unlock_bh(&sk->sk_callback_lock);
 648         if (orig_report)
 649                 orig_report(sk);
 650 }
 651
 652 /* Note: sk_callback_lock must be locked before calling this function. */
 653 static void save_listen_callbacks(struct socket *sock)
 654 {
 655         struct sock *sk = sock->sk;
 656
 657         listen_sock.sk_data_ready = sk->sk_data_ready;
 658         listen_sock.sk_state_change = sk->sk_state_change;
 659         listen_sock.sk_write_space = sk->sk_write_space;
 660         listen_sock.sk_error_report = sk->sk_error_report;
 661 }
 662
 663 static void restore_callbacks(struct socket *sock)
 664 {
 665         struct sock *sk = sock->sk;
 666
 667         write_lock_bh(&sk->sk_callback_lock);
 668         sk->sk_user_data = NULL;
 669         sk->sk_data_ready = listen_sock.sk_data_ready;
 670         sk->sk_state_change = listen_sock.sk_state_change;
 671         sk->sk_write_space = listen_sock.sk_write_space;
 672         sk->sk_error_report = listen_sock.sk_error_report;
 673         write_unlock_bh(&sk->sk_callback_lock);
 674 }
 675
 676 static void add_listen_sock(struct socket *sock, struct listen_connection *con)
 677 {
 678         struct sock *sk = sock->sk;
 679
 680         write_lock_bh(&sk->sk_callback_lock);
 681         save_listen_callbacks(sock);
 682         con->sock = sock;
 683
 684         sk->sk_user_data = con;
 685         sk->sk_allocation = GFP_NOFS;
 686         /* Install a data_ready callback */
 687         sk->sk_data_ready = lowcomms_listen_data_ready;
 688         write_unlock_bh(&sk->sk_callback_lock);
 689 }
 690
 691 /* Make a socket active */
 692 static void add_sock(struct socket *sock, struct connection *con)
 693 {
 694         struct sock *sk = sock->sk;
 695
 696         write_lock_bh(&sk->sk_callback_lock);
 697         con->sock = sock;
 698
 699         sk->sk_user_data = con;
 700         /* Install a data_ready callback */
 701         sk->sk_data_ready = lowcomms_data_ready;
 702         sk->sk_write_space = lowcomms_write_space;
 703         sk->sk_state_change = lowcomms_state_change;
 704         sk->sk_allocation = GFP_NOFS;
 705         sk->sk_error_report = lowcomms_error_report;
 706         write_unlock_bh(&sk->sk_callback_lock);
 707 }
 708
 709 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 710    length */
 711 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 712                           int *addr_len)
 713 {
 714         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 715         if (saddr->ss_family == AF_INET) {
 716                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 717                 in4_addr->sin_port = cpu_to_be16(port);
 718                 *addr_len = sizeof(struct sockaddr_in);
 719                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 720         } else {
 721                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 722                 in6_addr->sin6_port = cpu_to_be16(port);
 723                 *addr_len = sizeof(struct sockaddr_in6);
 724         }
 725         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 726 }
 727
 728 static void dlm_page_release(struct kref *kref)
 729 {
 730         struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
 731                                                   ref);
 732
 733         __free_page(e->page);
 734         kfree(e);
 735 }
 736
 737 static void dlm_msg_release(struct kref *kref)
 738 {
 739         struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
 740
 741         kref_put(&msg->entry->ref, dlm_page_release);
 742         kfree(msg);
 743 }
 744
 745 static void free_entry(struct writequeue_entry *e)
 746 {
 747         struct dlm_msg *msg, *tmp;
 748
 749         list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
 750                 if (msg->orig_msg) {
 751                         msg->orig_msg->retransmit = false;
 752                         kref_put(&msg->orig_msg->ref, dlm_msg_release);
 753                 }
 754
 755                 list_del(&msg->list);
 756                 kref_put(&msg->ref, dlm_msg_release);
 757         }
 758
 759         list_del(&e->list);
 760         atomic_dec(&e->con->writequeue_cnt);
 761         kref_put(&e->ref, dlm_page_release);
 762 }
 763
 764 static void dlm_close_sock(struct socket **sock)
 765 {
 766         if (*sock) {
 767                 restore_callbacks(*sock);
 768                 sock_release(*sock);
 769                 *sock = NULL;
 770         }
 771 }
 772
 773 /* Close a remote connection and tidy up */
 774 static void close_connection(struct connection *con, bool and_other,
 775                              bool tx, bool rx)
 776 {
 777         bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
 778         struct writequeue_entry *e;
 779
 780         if (tx && !closing && cancel_work_sync(&con->swork)) {
 781                 log_print("canceled swork for node %d", con->nodeid);
 782                 clear_bit(CF_WRITE_PENDING, &con->flags);
 783         }
 784         if (rx && !closing && cancel_work_sync(&con->rwork)) {
 785                 log_print("canceled rwork for node %d", con->nodeid);
 786                 clear_bit(CF_READ_PENDING, &con->flags);
 787         }
 788
 789         mutex_lock(&con->sock_mutex);
 790         dlm_close_sock(&con->sock);
 791
 792         if (con->othercon && and_other) {
 793                 /* Will only re-enter once. */
 794                 close_connection(con->othercon, false, tx, rx);
 795         }
 796
 797         /* if we send a writequeue entry only a half way, we drop the
 798          * whole entry because reconnection and that we not start of the
 799          * middle of a msg which will confuse the other end.
 800          *
 801          * we can always drop messages because retransmits, but what we
 802          * cannot allow is to transmit half messages which may be processed
 803          * at the other side.
 804          *
 805          * our policy is to start on a clean state when disconnects, we don't
 806          * know what's send/received on transport layer in this case.
 807          */
 808         spin_lock(&con->writequeue_lock);
 809         if (!list_empty(&con->writequeue)) {
 810                 e = list_first_entry(&con->writequeue, struct writequeue_entry,
 811                                      list);
 812                 if (e->dirty)
 813                         free_entry(e);
 814         }
 815         spin_unlock(&con->writequeue_lock);
 816
 817         con->rx_leftover = 0;
 818         con->retries = 0;
 819         clear_bit(CF_APP_LIMITED, &con->flags);
 820         clear_bit(CF_CONNECTED, &con->flags);
 821         clear_bit(CF_DELAY_CONNECT, &con->flags);
 822         clear_bit(CF_RECONNECT, &con->flags);
 823         clear_bit(CF_EOF, &con->flags);
 824         mutex_unlock(&con->sock_mutex);
 825         clear_bit(CF_CLOSING, &con->flags);
 826 }
 827
 828 static void shutdown_connection(struct connection *con)
 829 {
 830         int ret;
 831
 832         flush_work(&con->swork);
 833
 834         mutex_lock(&con->sock_mutex);
 835         /* nothing to shutdown */
 836         if (!con->sock) {
 837                 mutex_unlock(&con->sock_mutex);
 838                 return;
 839         }
 840
 841         set_bit(CF_SHUTDOWN, &con->flags);
 842         ret = kernel_sock_shutdown(con->sock, SHUT_WR);
 843         mutex_unlock(&con->sock_mutex);
 844         if (ret) {
 845                 log_print("Connection %p failed to shutdown: %d will force close",
 846                           con, ret);
 847                 goto force_close;
 848         } else {
 849                 ret = wait_event_timeout(con->shutdown_wait,
 850                                          !test_bit(CF_SHUTDOWN, &con->flags),
 851                                          DLM_SHUTDOWN_WAIT_TIMEOUT);
 852                 if (ret == 0) {
 853                         log_print("Connection %p shutdown timed out, will force close",
 854                                   con);
 855                         goto force_close;
 856                 }
 857         }
 858
 859         return;
 860
 861 force_close:
 862         clear_bit(CF_SHUTDOWN, &con->flags);
 863         close_connection(con, false, true, true);
 864 }
 865
 866 static void dlm_tcp_shutdown(struct connection *con)
 867 {
 868         if (con->othercon)
 869                 shutdown_connection(con->othercon);
 870         shutdown_connection(con);
 871 }
 872
 873 static int con_realloc_receive_buf(struct connection *con, int newlen)
 874 {
 875         unsigned char *newbuf;
 876
 877         newbuf = kmalloc(newlen, GFP_NOFS);
 878         if (!newbuf)
 879                 return -ENOMEM;
 880
 881         /* copy any leftover from last receive */
 882         if (con->rx_leftover)
 883                 memmove(newbuf, con->rx_buf, con->rx_leftover);
 884
 885         /* swap to new buffer space */
 886         kfree(con->rx_buf);
 887         con->rx_buflen = newlen;
 888         con->rx_buf = newbuf;
 889
 890         return 0;
 891 }
 892
 893 /* Data received from remote end */
 894 static int receive_from_sock(struct connection *con)
 895 {
 896         int call_again_soon = 0;
 897         struct msghdr msg;
 898         struct kvec iov;
 899         int ret, buflen;
 900
 901         mutex_lock(&con->sock_mutex);
 902
 903         if (con->sock == NULL) {
 904                 ret = -EAGAIN;
 905                 goto out_close;
 906         }
 907
 908         /* realloc if we get new buffer size to read out */
 909         buflen = dlm_config.ci_buffer_size;
 910         if (con->rx_buflen != buflen && con->rx_leftover <= buflen) {
 911                 ret = con_realloc_receive_buf(con, buflen);
 912                 if (ret < 0)
 913                         goto out_resched;
 914         }
 915
 916         /* calculate new buffer parameter regarding last receive and
 917          * possible leftover bytes
 918          */
 919         iov.iov_base = con->rx_buf + con->rx_leftover;
 920         iov.iov_len = con->rx_buflen - con->rx_leftover;
 921
 922         memset(&msg, 0, sizeof(msg));
 923         msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
 924         ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
 925                              msg.msg_flags);
 926         if (ret <= 0)
 927                 goto out_close;
 928         else if (ret == iov.iov_len)
 929                 call_again_soon = 1;
 930
 931         /* new buflen according readed bytes and leftover from last receive */
 932         buflen = ret + con->rx_leftover;
 933         ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen);
 934         if (ret < 0)
 935                 goto out_close;
 936
 937         /* calculate leftover bytes from process and put it into begin of
 938          * the receive buffer, so next receive we have the full message
 939          * at the start address of the receive buffer.
 940          */
 941         con->rx_leftover = buflen - ret;
 942         if (con->rx_leftover) {
 943                 memmove(con->rx_buf, con->rx_buf + ret,
 944                         con->rx_leftover);
 945                 call_again_soon = true;
 946         }
 947
 948         if (call_again_soon)
 949                 goto out_resched;
 950
 951         mutex_unlock(&con->sock_mutex);
 952         return 0;
 953
 954 out_resched:
 955         if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
 956                 queue_work(recv_workqueue, &con->rwork);
 957         mutex_unlock(&con->sock_mutex);
 958         return -EAGAIN;
 959
 960 out_close:
 961         if (ret == 0) {
 962                 log_print("connection %p got EOF from %d",
 963                           con, con->nodeid);
 964
 965                 if (con->eof_condition && con->eof_condition(con)) {
 966                         set_bit(CF_EOF, &con->flags);
 967                         mutex_unlock(&con->sock_mutex);
 968                 } else {
 969                         mutex_unlock(&con->sock_mutex);
 970                         close_connection(con, false, true, false);
 971
 972                         /* handling for tcp shutdown */
 973                         clear_bit(CF_SHUTDOWN, &con->flags);
 974                         wake_up(&con->shutdown_wait);
 975                 }
 976
 977                 /* signal to breaking receive worker */
 978                 ret = -1;
 979         } else {
 980                 mutex_unlock(&con->sock_mutex);
 981         }
 982         return ret;
 983 }
 984
 985 /* Listening socket is busy, accept a connection */
 986 static int accept_from_sock(struct listen_connection *con)
 987 {
 988         int result;
 989         struct sockaddr_storage peeraddr;
 990         struct socket *newsock;
 991         int len, idx;
 992         int nodeid;
 993         struct connection *newcon;
 994         struct connection *addcon;
 995         unsigned int mark;
 996
 997         if (!con->sock)
 998                 return -ENOTCONN;
 999
1000         result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
1001         if (result < 0)
1002                 goto accept_err;
1003
1004         /* Get the connected socket's peer */
1005         memset(&peeraddr, 0, sizeof(peeraddr));
1006         len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
1007         if (len < 0) {
1008                 result = -ECONNABORTED;
1009                 goto accept_err;
1010         }
1011
1012         /* Get the new node's NODEID */
1013         make_sockaddr(&peeraddr, 0, &len);
1014         if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
1015                 unsigned char *b=(unsigned char *)&peeraddr;
1016                 log_print("connect from non cluster node");
1017                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
1018                                      b, sizeof(struct sockaddr_storage));
1019                 sock_release(newsock);
1020                 return -1;
1021         }
1022
1023         log_print("got connection from %d", nodeid);
1024
1025         /*  Check to see if we already have a connection to this node. This
1026          *  could happen if the two nodes initiate a connection at roughly
1027          *  the same time and the connections cross on the wire.
1028          *  In this case we store the incoming one in "othercon"
1029          */
1030         idx = srcu_read_lock(&connections_srcu);
1031         newcon = nodeid2con(nodeid, GFP_NOFS);
1032         if (!newcon) {
1033                 srcu_read_unlock(&connections_srcu, idx);
1034                 result = -ENOMEM;
1035                 goto accept_err;
1036         }
1037
1038         sock_set_mark(newsock->sk, mark);
1039
1040         mutex_lock(&newcon->sock_mutex);
1041         if (newcon->sock) {
1042                 struct connection *othercon = newcon->othercon;
1043
1044                 if (!othercon) {
1045                         othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1046                         if (!othercon) {
1047                                 log_print("failed to allocate incoming socket");
1048                                 mutex_unlock(&newcon->sock_mutex);
1049                                 srcu_read_unlock(&connections_srcu, idx);
1050                                 result = -ENOMEM;
1051                                 goto accept_err;
1052                         }
1053
1054                         result = dlm_con_init(othercon, nodeid);
1055                         if (result < 0) {
1056                                 kfree(othercon);
1057                                 mutex_unlock(&newcon->sock_mutex);
1058                                 srcu_read_unlock(&connections_srcu, idx);
1059                                 goto accept_err;
1060                         }
1061
1062                         lockdep_set_subclass(&othercon->sock_mutex, 1);
1063                         set_bit(CF_IS_OTHERCON, &othercon->flags);
1064                         newcon->othercon = othercon;
1065                         othercon->sendcon = newcon;
1066                 } else {
1067                         /* close other sock con if we have something new */
1068                         close_connection(othercon, false, true, false);
1069                 }
1070
1071                 mutex_lock(&othercon->sock_mutex);
1072                 add_sock(newsock, othercon);
1073                 addcon = othercon;
1074                 mutex_unlock(&othercon->sock_mutex);
1075         }
1076         else {
1077                 /* accept copies the sk after we've saved the callbacks, so we
1078                    don't want to save them a second time or comm errors will
1079                    result in calling sk_error_report recursively. */
1080                 add_sock(newsock, newcon);
1081                 addcon = newcon;
1082         }
1083
1084         set_bit(CF_CONNECTED, &addcon->flags);
1085         mutex_unlock(&newcon->sock_mutex);
1086
1087         /*
1088          * Add it to the active queue in case we got data
1089          * between processing the accept adding the socket
1090          * to the read_sockets list
1091          */
1092         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
1093                 queue_work(recv_workqueue, &addcon->rwork);
1094
1095         srcu_read_unlock(&connections_srcu, idx);
1096
1097         return 0;
1098
1099 accept_err:
1100         if (newsock)
1101                 sock_release(newsock);
1102
1103         if (result != -EAGAIN)
1104                 log_print("error accepting connection from node: %d", result);
1105         return result;
1106 }
1107
1108 /*
1109  * writequeue_entry_complete - try to delete and free write queue entry
1110  * @e: write queue entry to try to delete
1111  * @completed: bytes completed
1112  *
1113  * writequeue_lock must be held.
1114  */
1115 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1116 {
1117         e->offset += completed;
1118         e->len -= completed;
1119         /* signal that page was half way transmitted */
1120         e->dirty = true;
1121
1122         if (e->len == 0 && e->users == 0)
1123                 free_entry(e);
1124 }
1125
1126 /*
1127  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1128  */
1129 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1130 {
1131         struct sockaddr_storage localaddr;
1132         struct sockaddr *addr = (struct sockaddr *)&localaddr;
1133         int i, addr_len, result = 0;
1134
1135         for (i = 0; i < dlm_local_count; i++) {
1136                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
1137                 make_sockaddr(&localaddr, port, &addr_len);
1138
1139                 if (!i)
1140                         result = kernel_bind(sock, addr, addr_len);
1141                 else
1142                         result = sock_bind_add(sock->sk, addr, addr_len);
1143
1144                 if (result < 0) {
1145                         log_print("Can't bind to %d addr number %d, %d.\n",
1146                                   port, i + 1, result);
1147                         break;
1148                 }
1149         }
1150         return result;
1151 }
1152
1153 /* Initiate an SCTP association.
1154    This is a special case of send_to_sock() in that we don't yet have a
1155    peeled-off socket for this association, so we use the listening socket
1156    and add the primary IP address of the remote node.
1157  */
1158 static void sctp_connect_to_sock(struct connection *con)
1159 {
1160         struct sockaddr_storage daddr;
1161         int result;
1162         int addr_len;
1163         struct socket *sock;
1164         unsigned int mark;
1165
1166         mutex_lock(&con->sock_mutex);
1167
1168         /* Some odd races can cause double-connects, ignore them */
1169         if (con->retries++ > MAX_CONNECT_RETRIES)
1170                 goto out;
1171
1172         if (con->sock) {
1173                 log_print("node %d already connected.", con->nodeid);
1174                 goto out;
1175         }
1176
1177         memset(&daddr, 0, sizeof(daddr));
1178         result = nodeid_to_addr(con->nodeid, &daddr, NULL, true, &mark);
1179         if (result < 0) {
1180                 log_print("no address for nodeid %d", con->nodeid);
1181                 goto out;
1182         }
1183
1184         /* Create a socket to communicate with */
1185         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1186                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1187         if (result < 0)
1188                 goto socket_err;
1189
1190         sock_set_mark(sock->sk, mark);
1191
1192         add_sock(sock, con);
1193
1194         /* Bind to all addresses. */
1195         if (sctp_bind_addrs(con->sock, 0))
1196                 goto bind_err;
1197
1198         make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len);
1199
1200         log_print_ratelimited("connecting to %d", con->nodeid);
1201
1202         /* Turn off Nagle's algorithm */
1203         sctp_sock_set_nodelay(sock->sk);
1204
1205         /*
1206          * Make sock->ops->connect() function return in specified time,
1207          * since O_NONBLOCK argument in connect() function does not work here,
1208          * then, we should restore the default value of this attribute.
1209          */
1210         sock_set_sndtimeo(sock->sk, 5);
1211         result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len,
1212                                    0);
1213         sock_set_sndtimeo(sock->sk, 0);
1214
1215         if (result == -EINPROGRESS)
1216                 result = 0;
1217         if (result == 0) {
1218                 if (!test_and_set_bit(CF_CONNECTED, &con->flags))
1219                         log_print("successful connected to node %d", con->nodeid);
1220                 goto out;
1221         }
1222
1223 bind_err:
1224         con->sock = NULL;
1225         sock_release(sock);
1226
1227 socket_err:
1228         /*
1229          * Some errors are fatal and this list might need adjusting. For other
1230          * errors we try again until the max number of retries is reached.
1231          */
1232         if (result != -EHOSTUNREACH &&
1233             result != -ENETUNREACH &&
1234             result != -ENETDOWN &&
1235             result != -EINVAL &&
1236             result != -EPROTONOSUPPORT) {
1237                 log_print("connect %d try %d error %d", con->nodeid,
1238                           con->retries, result);
1239                 mutex_unlock(&con->sock_mutex);
1240                 msleep(1000);
1241                 lowcomms_connect_sock(con);
1242                 return;
1243         }
1244
1245 out:
1246         mutex_unlock(&con->sock_mutex);
1247 }
1248
1249 /* Connect a new socket to its peer */
1250 static void tcp_connect_to_sock(struct connection *con)
1251 {
1252         struct sockaddr_storage saddr, src_addr;
1253         unsigned int mark;
1254         int addr_len;
1255         struct socket *sock = NULL;
1256         int result;
1257
1258         mutex_lock(&con->sock_mutex);
1259         if (con->retries++ > MAX_CONNECT_RETRIES)
1260                 goto out;
1261
1262         /* Some odd races can cause double-connects, ignore them */
1263         if (con->sock)
1264                 goto out;
1265
1266         /* Create a socket to communicate with */
1267         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1268                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1269         if (result < 0)
1270                 goto out_err;
1271
1272         memset(&saddr, 0, sizeof(saddr));
1273         result = nodeid_to_addr(con->nodeid, &saddr, NULL, false, &mark);
1274         if (result < 0) {
1275                 log_print("no address for nodeid %d", con->nodeid);
1276                 goto out_err;
1277         }
1278
1279         sock_set_mark(sock->sk, mark);
1280
1281         add_sock(sock, con);
1282
1283         /* Bind to our cluster-known address connecting to avoid
1284            routing problems */
1285         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1286         make_sockaddr(&src_addr, 0, &addr_len);
1287         result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1288                                  addr_len);
1289         if (result < 0) {
1290                 log_print("could not bind for connect: %d", result);
1291                 /* This *may* not indicate a critical error */
1292         }
1293
1294         make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1295
1296         log_print_ratelimited("connecting to %d", con->nodeid);
1297
1298         /* Turn off Nagle's algorithm */
1299         tcp_sock_set_nodelay(sock->sk);
1300
1301         result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1302                                    O_NONBLOCK);
1303         if (result == -EINPROGRESS)
1304                 result = 0;
1305         if (result == 0)
1306                 goto out;
1307
1308 out_err:
1309         if (con->sock) {
1310                 sock_release(con->sock);
1311                 con->sock = NULL;
1312         } else if (sock) {
1313                 sock_release(sock);
1314         }
1315         /*
1316          * Some errors are fatal and this list might need adjusting. For other
1317          * errors we try again until the max number of retries is reached.
1318          */
1319         if (result != -EHOSTUNREACH &&
1320             result != -ENETUNREACH &&
1321             result != -ENETDOWN &&
1322             result != -EINVAL &&
1323             result != -EPROTONOSUPPORT) {
1324                 log_print("connect %d try %d error %d", con->nodeid,
1325                           con->retries, result);
1326                 mutex_unlock(&con->sock_mutex);
1327                 msleep(1000);
1328                 lowcomms_connect_sock(con);
1329                 return;
1330         }
1331 out:
1332         mutex_unlock(&con->sock_mutex);
1333         return;
1334 }
1335
1336 /* On error caller must run dlm_close_sock() for the
1337  * listen connection socket.
1338  */
1339 static int tcp_create_listen_sock(struct listen_connection *con,
1340                                   struct sockaddr_storage *saddr)
1341 {
1342         struct socket *sock = NULL;
1343         int result = 0;
1344         int addr_len;
1345
1346         if (dlm_local_addr[0]->ss_family == AF_INET)
1347                 addr_len = sizeof(struct sockaddr_in);
1348         else
1349                 addr_len = sizeof(struct sockaddr_in6);
1350
1351         /* Create a socket to communicate with */
1352         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1353                                   SOCK_STREAM, IPPROTO_TCP, &sock);
1354         if (result < 0) {
1355                 log_print("Can't create listening comms socket");
1356                 goto create_out;
1357         }
1358
1359         sock_set_mark(sock->sk, dlm_config.ci_mark);
1360
1361         /* Turn off Nagle's algorithm */
1362         tcp_sock_set_nodelay(sock->sk);
1363
1364         sock_set_reuseaddr(sock->sk);
1365
1366         add_listen_sock(sock, con);
1367
1368         /* Bind to our port */
1369         make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1370         result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1371         if (result < 0) {
1372                 log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1373                 goto create_out;
1374         }
1375         sock_set_keepalive(sock->sk);
1376
1377         result = sock->ops->listen(sock, 5);
1378         if (result < 0) {
1379                 log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1380                 goto create_out;
1381         }
1382
1383         return 0;
1384
1385 create_out:
1386         return result;
1387 }
1388
1389 /* Get local addresses */
1390 static void init_local(void)
1391 {
1392         struct sockaddr_storage sas, *addr;
1393         int i;
1394
1395         dlm_local_count = 0;
1396         for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1397                 if (dlm_our_addr(&sas, i))
1398                         break;
1399
1400                 addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1401                 if (!addr)
1402                         break;
1403                 dlm_local_addr[dlm_local_count++] = addr;
1404         }
1405 }
1406
1407 static void deinit_local(void)
1408 {
1409         int i;
1410
1411         for (i = 0; i < dlm_local_count; i++)
1412                 kfree(dlm_local_addr[i]);
1413 }
1414
1415 /* Initialise SCTP socket and bind to all interfaces
1416  * On error caller must run dlm_close_sock() for the
1417  * listen connection socket.
1418  */
1419 static int sctp_listen_for_all(struct listen_connection *con)
1420 {
1421         struct socket *sock = NULL;
1422         int result = -EINVAL;
1423
1424         log_print("Using SCTP for communications");
1425
1426         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1427                                   SOCK_STREAM, IPPROTO_SCTP, &sock);
1428         if (result < 0) {
1429                 log_print("Can't create comms socket, check SCTP is loaded");
1430                 goto out;
1431         }
1432
1433         sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1434         sock_set_mark(sock->sk, dlm_config.ci_mark);
1435         sctp_sock_set_nodelay(sock->sk);
1436
1437         add_listen_sock(sock, con);
1438
1439         /* Bind to all addresses. */
1440         result = sctp_bind_addrs(con->sock, dlm_config.ci_tcp_port);
1441         if (result < 0)
1442                 goto out;
1443
1444         result = sock->ops->listen(sock, 5);
1445         if (result < 0) {
1446                 log_print("Can't set socket listening");
1447                 goto out;
1448         }
1449
1450         return 0;
1451
1452 out:
1453         return result;
1454 }
1455
1456 static int tcp_listen_for_all(void)
1457 {
1458         /* We don't support multi-homed hosts */
1459         if (dlm_local_count > 1) {
1460                 log_print("TCP protocol can't handle multi-homed hosts, "
1461                           "try SCTP");
1462                 return -EINVAL;
1463         }
1464
1465         log_print("Using TCP for communications");
1466
1467         return tcp_create_listen_sock(&listen_con, dlm_local_addr[0]);
1468 }
1469
1470
1471
1472 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1473                                                      gfp_t allocation)
1474 {
1475         struct writequeue_entry *entry;
1476
1477         entry = kzalloc(sizeof(*entry), allocation);
1478         if (!entry)
1479                 return NULL;
1480
1481         entry->page = alloc_page(allocation | __GFP_ZERO);
1482         if (!entry->page) {
1483                 kfree(entry);
1484                 return NULL;
1485         }
1486
1487         entry->con = con;
1488         entry->users = 1;
1489         kref_init(&entry->ref);
1490         INIT_LIST_HEAD(&entry->msgs);
1491
1492         return entry;
1493 }
1494
1495 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1496                                              gfp_t allocation, char **ppc,
1497                                              void (*cb)(struct dlm_mhandle *mh),
1498                                              struct dlm_mhandle *mh)
1499 {
1500         struct writequeue_entry *e;
1501
1502         spin_lock(&con->writequeue_lock);
1503         if (!list_empty(&con->writequeue)) {
1504                 e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1505                 if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1506                         kref_get(&e->ref);
1507
1508                         *ppc = page_address(e->page) + e->end;
1509                         if (cb)
1510                                 cb(mh);
1511
1512                         e->end += len;
1513                         e->users++;
1514                         spin_unlock(&con->writequeue_lock);
1515
1516                         return e;
1517                 }
1518         }
1519         spin_unlock(&con->writequeue_lock);
1520
1521         e = new_writequeue_entry(con, allocation);
1522         if (!e)
1523                 return NULL;
1524
1525         kref_get(&e->ref);
1526         *ppc = page_address(e->page);
1527         e->end += len;
1528         atomic_inc(&con->writequeue_cnt);
1529
1530         spin_lock(&con->writequeue_lock);
1531         if (cb)
1532                 cb(mh);
1533
1534         list_add_tail(&e->list, &con->writequeue);
1535         spin_unlock(&con->writequeue_lock);
1536
1537         return e;
1538 };
1539
1540 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1541                                                 gfp_t allocation, char **ppc,
1542                                                 void (*cb)(struct dlm_mhandle *mh),
1543                                                 struct dlm_mhandle *mh)
1544 {
1545         struct writequeue_entry *e;
1546         struct dlm_msg *msg;
1547
1548         msg = kzalloc(sizeof(*msg), allocation);
1549         if (!msg)
1550                 return NULL;
1551
1552         kref_init(&msg->ref);
1553
1554         e = new_wq_entry(con, len, allocation, ppc, cb, mh);
1555         if (!e) {
1556                 kfree(msg);
1557                 return NULL;
1558         }
1559
1560         msg->ppc = *ppc;
1561         msg->len = len;
1562         msg->entry = e;
1563
1564         return msg;
1565 }
1566
1567 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
1568                                      char **ppc, void (*cb)(struct dlm_mhandle *mh),
1569                                      struct dlm_mhandle *mh)
1570 {
1571         struct connection *con;
1572         struct dlm_msg *msg;
1573         int idx;
1574
1575         if (len > DLM_MAX_SOCKET_BUFSIZE ||
1576             len < sizeof(struct dlm_header)) {
1577                 BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1578                 log_print("failed to allocate a buffer of size %d", len);
1579                 WARN_ON(1);
1580                 return NULL;
1581         }
1582
1583         idx = srcu_read_lock(&connections_srcu);
1584         con = nodeid2con(nodeid, allocation);
1585         if (!con) {
1586                 srcu_read_unlock(&connections_srcu, idx);
1587                 return NULL;
1588         }
1589
1590         msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, mh);
1591         if (!msg) {
1592                 srcu_read_unlock(&connections_srcu, idx);
1593                 return NULL;
1594         }
1595
1596         /* we assume if successful commit must called */
1597         msg->idx = idx;
1598         return msg;
1599 }
1600
1601 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1602 {
1603         struct writequeue_entry *e = msg->entry;
1604         struct connection *con = e->con;
1605         int users;
1606
1607         spin_lock(&con->writequeue_lock);
1608         kref_get(&msg->ref);
1609         list_add(&msg->list, &e->msgs);
1610
1611         users = --e->users;
1612         if (users)
1613                 goto out;
1614
1615         e->len = DLM_WQ_LENGTH_BYTES(e);
1616         spin_unlock(&con->writequeue_lock);
1617
1618         queue_work(send_workqueue, &con->swork);
1619         return;
1620
1621 out:
1622         spin_unlock(&con->writequeue_lock);
1623         return;
1624 }
1625
1626 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1627 {
1628         _dlm_lowcomms_commit_msg(msg);
1629         srcu_read_unlock(&connections_srcu, msg->idx);
1630 }
1631
1632 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1633 {
1634         kref_put(&msg->ref, dlm_msg_release);
1635 }
1636
1637 /* does not held connections_srcu, usage workqueue only */
1638 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1639 {
1640         struct dlm_msg *msg_resend;
1641         char *ppc;
1642
1643         if (msg->retransmit)
1644                 return 1;
1645
1646         msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
1647                                               GFP_ATOMIC, &ppc, NULL, NULL);
1648         if (!msg_resend)
1649                 return -ENOMEM;
1650
1651         msg->retransmit = true;
1652         kref_get(&msg->ref);
1653         msg_resend->orig_msg = msg;
1654
1655         memcpy(ppc, msg->ppc, msg->len);
1656         _dlm_lowcomms_commit_msg(msg_resend);
1657         dlm_lowcomms_put_msg(msg_resend);
1658
1659         return 0;
1660 }
1661
1662 /* Send a message */
1663 static void send_to_sock(struct connection *con)
1664 {
1665         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1666         struct writequeue_entry *e;
1667         int len, offset, ret;
1668         int count = 0;
1669
1670         mutex_lock(&con->sock_mutex);
1671         if (con->sock == NULL)
1672                 goto out_connect;
1673
1674         spin_lock(&con->writequeue_lock);
1675         for (;;) {
1676                 e = con_next_wq(con);
1677                 if (!e)
1678                         break;
1679
1680                 e = list_first_entry(&con->writequeue, struct writequeue_entry, list);
1681                 len = e->len;
1682                 offset = e->offset;
1683                 BUG_ON(len == 0 && e->users == 0);
1684                 spin_unlock(&con->writequeue_lock);
1685
1686                 ret = kernel_sendpage(con->sock, e->page, offset, len,
1687                                       msg_flags);
1688                 if (ret == -EAGAIN || ret == 0) {
1689                         if (ret == -EAGAIN &&
1690                             test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1691                             !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1692                                 /* Notify TCP that we're limited by the
1693                                  * application window size.
1694                                  */
1695                                 set_bit(SOCK_NOSPACE, &con->sock->flags);
1696                                 con->sock->sk->sk_write_pending++;
1697                         }
1698                         cond_resched();
1699                         goto out;
1700                 } else if (ret < 0)
1701                         goto out;
1702
1703                 /* Don't starve people filling buffers */
1704                 if (++count >= MAX_SEND_MSG_COUNT) {
1705                         cond_resched();
1706                         count = 0;
1707                 }
1708
1709                 spin_lock(&con->writequeue_lock);
1710                 writequeue_entry_complete(e, ret);
1711         }
1712         spin_unlock(&con->writequeue_lock);
1713
1714         /* close if we got EOF */
1715         if (test_and_clear_bit(CF_EOF, &con->flags)) {
1716                 mutex_unlock(&con->sock_mutex);
1717                 close_connection(con, false, false, true);
1718
1719                 /* handling for tcp shutdown */
1720                 clear_bit(CF_SHUTDOWN, &con->flags);
1721                 wake_up(&con->shutdown_wait);
1722         } else {
1723                 mutex_unlock(&con->sock_mutex);
1724         }
1725
1726         return;
1727
1728 out:
1729         mutex_unlock(&con->sock_mutex);
1730         return;
1731
1732 out_connect:
1733         mutex_unlock(&con->sock_mutex);
1734         queue_work(send_workqueue, &con->swork);
1735         cond_resched();
1736 }
1737
1738 static void clean_one_writequeue(struct connection *con)
1739 {
1740         struct writequeue_entry *e, *safe;
1741
1742         spin_lock(&con->writequeue_lock);
1743         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1744                 free_entry(e);
1745         }
1746         spin_unlock(&con->writequeue_lock);
1747 }
1748
1749 /* Called from recovery when it knows that a node has
1750    left the cluster */
1751 int dlm_lowcomms_close(int nodeid)
1752 {
1753         struct connection *con;
1754         struct dlm_node_addr *na;
1755         int idx;
1756
1757         log_print("closing connection to node %d", nodeid);
1758         idx = srcu_read_lock(&connections_srcu);
1759         con = nodeid2con(nodeid, 0);
1760         if (con) {
1761                 set_bit(CF_CLOSE, &con->flags);
1762                 close_connection(con, true, true, true);
1763                 clean_one_writequeue(con);
1764                 if (con->othercon)
1765                         clean_one_writequeue(con->othercon);
1766         }
1767         srcu_read_unlock(&connections_srcu, idx);
1768
1769         spin_lock(&dlm_node_addrs_spin);
1770         na = find_node_addr(nodeid);
1771         if (na) {
1772                 list_del(&na->list);
1773                 while (na->addr_count--)
1774                         kfree(na->addr[na->addr_count]);
1775                 kfree(na);
1776         }
1777         spin_unlock(&dlm_node_addrs_spin);
1778
1779         return 0;
1780 }
1781
1782 /* Receive workqueue function */
1783 static void process_recv_sockets(struct work_struct *work)
1784 {
1785         struct connection *con = container_of(work, struct connection, rwork);
1786         int err;
1787
1788         clear_bit(CF_READ_PENDING, &con->flags);
1789         do {
1790                 err = receive_from_sock(con);
1791         } while (!err);
1792 }
1793
1794 static void process_listen_recv_socket(struct work_struct *work)
1795 {
1796         accept_from_sock(&listen_con);
1797 }
1798
1799 /* Send workqueue function */
1800 static void process_send_sockets(struct work_struct *work)
1801 {
1802         struct connection *con = container_of(work, struct connection, swork);
1803
1804         WARN_ON(test_bit(CF_IS_OTHERCON, &con->flags));
1805
1806         clear_bit(CF_WRITE_PENDING, &con->flags);
1807
1808         if (test_and_clear_bit(CF_RECONNECT, &con->flags)) {
1809                 close_connection(con, false, false, true);
1810                 dlm_midcomms_unack_msg_resend(con->nodeid);
1811         }
1812
1813         if (con->sock == NULL) { /* not mutex protected so check it inside too */
1814                 if (test_and_clear_bit(CF_DELAY_CONNECT, &con->flags))
1815                         msleep(1000);
1816                 con->connect_action(con);
1817         }
1818         if (!list_empty(&con->writequeue))
1819                 send_to_sock(con);
1820 }
1821
1822 static void work_stop(void)
1823 {
1824         if (recv_workqueue) {
1825                 destroy_workqueue(recv_workqueue);
1826                 recv_workqueue = NULL;
1827         }
1828
1829         if (send_workqueue) {
1830                 destroy_workqueue(send_workqueue);
1831                 send_workqueue = NULL;
1832         }
1833 }
1834
1835 static int work_start(void)
1836 {
1837         recv_workqueue = alloc_ordered_workqueue("dlm_recv", WQ_MEM_RECLAIM);
1838         if (!recv_workqueue) {
1839                 log_print("can't start dlm_recv");
1840                 return -ENOMEM;
1841         }
1842
1843         send_workqueue = alloc_ordered_workqueue("dlm_send", WQ_MEM_RECLAIM);
1844         if (!send_workqueue) {
1845                 log_print("can't start dlm_send");
1846                 destroy_workqueue(recv_workqueue);
1847                 recv_workqueue = NULL;
1848                 return -ENOMEM;
1849         }
1850
1851         return 0;
1852 }
1853
1854 static void shutdown_conn(struct connection *con)
1855 {
1856         if (con->shutdown_action)
1857                 con->shutdown_action(con);
1858 }
1859
1860 void dlm_lowcomms_shutdown(void)
1861 {
1862         int idx;
1863
1864         /* Set all the flags to prevent any
1865          * socket activity.
1866          */
1867         dlm_allow_conn = 0;
1868
1869         if (recv_workqueue)
1870                 flush_workqueue(recv_workqueue);
1871         if (send_workqueue)
1872                 flush_workqueue(send_workqueue);
1873
1874         dlm_close_sock(&listen_con.sock);
1875
1876         idx = srcu_read_lock(&connections_srcu);
1877         foreach_conn(shutdown_conn);
1878         srcu_read_unlock(&connections_srcu, idx);
1879 }
1880
1881 static void _stop_conn(struct connection *con, bool and_other)
1882 {
1883         mutex_lock(&con->sock_mutex);
1884         set_bit(CF_CLOSE, &con->flags);
1885         set_bit(CF_READ_PENDING, &con->flags);
1886         set_bit(CF_WRITE_PENDING, &con->flags);
1887         if (con->sock && con->sock->sk) {
1888                 write_lock_bh(&con->sock->sk->sk_callback_lock);
1889                 con->sock->sk->sk_user_data = NULL;
1890                 write_unlock_bh(&con->sock->sk->sk_callback_lock);
1891         }
1892         if (con->othercon && and_other)
1893                 _stop_conn(con->othercon, false);
1894         mutex_unlock(&con->sock_mutex);
1895 }
1896
1897 static void stop_conn(struct connection *con)
1898 {
1899         _stop_conn(con, true);
1900 }
1901
1902 static void connection_release(struct rcu_head *rcu)
1903 {
1904         struct connection *con = container_of(rcu, struct connection, rcu);
1905
1906         kfree(con->rx_buf);
1907         kfree(con);
1908 }
1909
1910 static void free_conn(struct connection *con)
1911 {
1912         close_connection(con, true, true, true);
1913         spin_lock(&connections_lock);
1914         hlist_del_rcu(&con->list);
1915         spin_unlock(&connections_lock);
1916         if (con->othercon) {
1917                 clean_one_writequeue(con->othercon);
1918                 call_srcu(&connections_srcu, &con->othercon->rcu,
1919                           connection_release);
1920         }
1921         clean_one_writequeue(con);
1922         call_srcu(&connections_srcu, &con->rcu, connection_release);
1923 }
1924
1925 static void work_flush(void)
1926 {
1927         int ok;
1928         int i;
1929         struct connection *con;
1930
1931         do {
1932                 ok = 1;
1933                 foreach_conn(stop_conn);
1934                 if (recv_workqueue)
1935                         flush_workqueue(recv_workqueue);
1936                 if (send_workqueue)
1937                         flush_workqueue(send_workqueue);
1938                 for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1939                         hlist_for_each_entry_rcu(con, &connection_hash[i],
1940                                                  list) {
1941                                 ok &= test_bit(CF_READ_PENDING, &con->flags);
1942                                 ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1943                                 if (con->othercon) {
1944                                         ok &= test_bit(CF_READ_PENDING,
1945                                                        &con->othercon->flags);
1946                                         ok &= test_bit(CF_WRITE_PENDING,
1947                                                        &con->othercon->flags);
1948                                 }
1949                         }
1950                 }
1951         } while (!ok);
1952 }
1953
1954 void dlm_lowcomms_stop(void)
1955 {
1956         int idx;
1957
1958         idx = srcu_read_lock(&connections_srcu);
1959         work_flush();
1960         foreach_conn(free_conn);
1961         srcu_read_unlock(&connections_srcu, idx);
1962         work_stop();
1963         deinit_local();
1964 }
1965
1966 int dlm_lowcomms_start(void)
1967 {
1968         int error = -EINVAL;
1969         int i;
1970
1971         for (i = 0; i < CONN_HASH_SIZE; i++)
1972                 INIT_HLIST_HEAD(&connection_hash[i]);
1973
1974         init_local();
1975         if (!dlm_local_count) {
1976                 error = -ENOTCONN;
1977                 log_print("no local IP address has been set");
1978                 goto fail;
1979         }
1980
1981         INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1982
1983         error = work_start();
1984         if (error)
1985                 goto fail_local;
1986
1987         dlm_allow_conn = 1;
1988
1989         /* Start listening */
1990         switch (dlm_config.ci_protocol) {
1991         case DLM_PROTO_TCP:
1992                 error = tcp_listen_for_all();
1993                 break;
1994         case DLM_PROTO_SCTP:
1995                 error = sctp_listen_for_all(&listen_con);
1996                 break;
1997         default:
1998                 log_print("Invalid protocol identifier %d set",
1999                           dlm_config.ci_protocol);
2000                 error = -EINVAL;
2001                 break;
2002         }
2003         if (error)
2004                 goto fail_unlisten;
2005
2006         return 0;
2007
2008 fail_unlisten:
2009         dlm_allow_conn = 0;
2010         dlm_close_sock(&listen_con.sock);
2011         work_stop();
2012 fail_local:
2013         deinit_local();
2014 fail:
2015         return error;
2016 }
2017
2018 void dlm_lowcomms_exit(void)
2019 {
2020         struct dlm_node_addr *na, *safe;
2021
2022         spin_lock(&dlm_node_addrs_spin);
2023         list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
2024                 list_del(&na->list);
2025                 while (na->addr_count--)
2026                         kfree(na->addr[na->addr_count]);
2027                 kfree(na);
2028         }
2029         spin_unlock(&dlm_node_addrs_spin);
2030 }