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         struct mutex wq_alloc;
  88         int retries;
  89 #define MAX_CONNECT_RETRIES 3
  90         struct hlist_node list;
  91         struct connection *othercon;
  92         struct connection *sendcon;
  93         struct work_struct rwork; /* Receive workqueue */
  94         struct work_struct swork; /* Send workqueue */
  95         wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */
  96         unsigned char *rx_buf;
  97         int rx_buflen;
  98         int rx_leftover;
  99         struct rcu_head rcu;
 100 };
 101 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
 102
 103 struct listen_connection {
 104         struct socket *sock;
 105         struct work_struct rwork;
 106 };
 107
 108 #define DLM_WQ_REMAIN_BYTES(e) (PAGE_SIZE - e->end)
 109 #define DLM_WQ_LENGTH_BYTES(e) (e->end - e->offset)
 110
 111 /* An entry waiting to be sent */
 112 struct writequeue_entry {
 113         struct list_head list;
 114         struct page *page;
 115         int offset;
 116         int len;
 117         int end;
 118         int users;
 119         bool dirty;
 120         struct connection *con;
 121         struct list_head msgs;
 122         struct kref ref;
 123 };
 124
 125 struct dlm_msg {
 126         struct writequeue_entry *entry;
 127         struct dlm_msg *orig_msg;
 128         bool retransmit;
 129         void *ppc;
 130         int len;
 131         int idx; /* new()/commit() idx exchange */
 132
 133         struct list_head list;
 134         struct kref ref;
 135 };
 136
 137 struct dlm_node_addr {
 138         struct list_head list;
 139         int nodeid;
 140         int mark;
 141         int addr_count;
 142         int curr_addr_index;
 143         struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
 144 };
 145
 146 struct dlm_proto_ops {
 147         bool try_new_addr;
 148         const char *name;
 149         int proto;
 150
 151         int (*connect)(struct connection *con, struct socket *sock,
 152                        struct sockaddr *addr, int addr_len);
 153         void (*sockopts)(struct socket *sock);
 154         int (*bind)(struct socket *sock);
 155         int (*listen_validate)(void);
 156         void (*listen_sockopts)(struct socket *sock);
 157         int (*listen_bind)(struct socket *sock);
 158         /* What to do to shutdown */
 159         void (*shutdown_action)(struct connection *con);
 160         /* What to do to eof check */
 161         bool (*eof_condition)(struct connection *con);
 162 };
 163
 164 static struct listen_sock_callbacks {
 165         void (*sk_error_report)(struct sock *);
 166         void (*sk_data_ready)(struct sock *);
 167         void (*sk_state_change)(struct sock *);
 168         void (*sk_write_space)(struct sock *);
 169 } listen_sock;
 170
 171 static LIST_HEAD(dlm_node_addrs);
 172 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
 173
 174 static struct listen_connection listen_con;
 175 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
 176 static int dlm_local_count;
 177 int dlm_allow_conn;
 178
 179 /* Work queues */
 180 static struct workqueue_struct *recv_workqueue;
 181 static struct workqueue_struct *send_workqueue;
 182
 183 static struct hlist_head connection_hash[CONN_HASH_SIZE];
 184 static DEFINE_SPINLOCK(connections_lock);
 185 DEFINE_STATIC_SRCU(connections_srcu);
 186
 187 static const struct dlm_proto_ops *dlm_proto_ops;
 188
 189 static void process_recv_sockets(struct work_struct *work);
 190 static void process_send_sockets(struct work_struct *work);
 191
 192 /* need to held writequeue_lock */
 193 static struct writequeue_entry *con_next_wq(struct connection *con)
 194 {
 195         struct writequeue_entry *e;
 196
 197         if (list_empty(&con->writequeue))
 198                 return NULL;
 199
 200         e = list_first_entry(&con->writequeue, struct writequeue_entry,
 201                              list);
 202         if (e->len == 0)
 203                 return NULL;
 204
 205         return e;
 206 }
 207
 208 static struct connection *__find_con(int nodeid, int r)
 209 {
 210         struct connection *con;
 211
 212         hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
 213                 if (con->nodeid == nodeid)
 214                         return con;
 215         }
 216
 217         return NULL;
 218 }
 219
 220 static bool tcp_eof_condition(struct connection *con)
 221 {
 222         return atomic_read(&con->writequeue_cnt);
 223 }
 224
 225 static int dlm_con_init(struct connection *con, int nodeid)
 226 {
 227         con->rx_buflen = dlm_config.ci_buffer_size;
 228         con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS);
 229         if (!con->rx_buf)
 230                 return -ENOMEM;
 231
 232         con->nodeid = nodeid;
 233         mutex_init(&con->sock_mutex);
 234         INIT_LIST_HEAD(&con->writequeue);
 235         spin_lock_init(&con->writequeue_lock);
 236         atomic_set(&con->writequeue_cnt, 0);
 237         INIT_WORK(&con->swork, process_send_sockets);
 238         INIT_WORK(&con->rwork, process_recv_sockets);
 239         init_waitqueue_head(&con->shutdown_wait);
 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         mutex_init(&con->wq_alloc);
 269
 270         spin_lock(&connections_lock);
 271         /* Because multiple workqueues/threads calls this function it can
 272          * race on multiple cpu's. Instead of locking hot path __find_con()
 273          * we just check in rare cases of recently added nodes again
 274          * under protection of connections_lock. If this is the case we
 275          * abort our connection creation and return the existing connection.
 276          */
 277         tmp = __find_con(nodeid, r);
 278         if (tmp) {
 279                 spin_unlock(&connections_lock);
 280                 kfree(con->rx_buf);
 281                 kfree(con);
 282                 return tmp;
 283         }
 284
 285         hlist_add_head_rcu(&con->list, &connection_hash[r]);
 286         spin_unlock(&connections_lock);
 287
 288         return con;
 289 }
 290
 291 /* Loop round all connections */
 292 static void foreach_conn(void (*conn_func)(struct connection *c))
 293 {
 294         int i;
 295         struct connection *con;
 296
 297         for (i = 0; i < CONN_HASH_SIZE; i++) {
 298                 hlist_for_each_entry_rcu(con, &connection_hash[i], list)
 299                         conn_func(con);
 300         }
 301 }
 302
 303 static struct dlm_node_addr *find_node_addr(int nodeid)
 304 {
 305         struct dlm_node_addr *na;
 306
 307         list_for_each_entry(na, &dlm_node_addrs, list) {
 308                 if (na->nodeid == nodeid)
 309                         return na;
 310         }
 311         return NULL;
 312 }
 313
 314 static int addr_compare(const struct sockaddr_storage *x,
 315                         const struct sockaddr_storage *y)
 316 {
 317         switch (x->ss_family) {
 318         case AF_INET: {
 319                 struct sockaddr_in *sinx = (struct sockaddr_in *)x;
 320                 struct sockaddr_in *siny = (struct sockaddr_in *)y;
 321                 if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
 322                         return 0;
 323                 if (sinx->sin_port != siny->sin_port)
 324                         return 0;
 325                 break;
 326         }
 327         case AF_INET6: {
 328                 struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
 329                 struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
 330                 if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
 331                         return 0;
 332                 if (sinx->sin6_port != siny->sin6_port)
 333                         return 0;
 334                 break;
 335         }
 336         default:
 337                 return 0;
 338         }
 339         return 1;
 340 }
 341
 342 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
 343                           struct sockaddr *sa_out, bool try_new_addr,
 344                           unsigned int *mark)
 345 {
 346         struct sockaddr_storage sas;
 347         struct dlm_node_addr *na;
 348
 349         if (!dlm_local_count)
 350                 return -1;
 351
 352         spin_lock(&dlm_node_addrs_spin);
 353         na = find_node_addr(nodeid);
 354         if (na && na->addr_count) {
 355                 memcpy(&sas, na->addr[na->curr_addr_index],
 356                        sizeof(struct sockaddr_storage));
 357
 358                 if (try_new_addr) {
 359                         na->curr_addr_index++;
 360                         if (na->curr_addr_index == na->addr_count)
 361                                 na->curr_addr_index = 0;
 362                 }
 363         }
 364         spin_unlock(&dlm_node_addrs_spin);
 365
 366         if (!na)
 367                 return -EEXIST;
 368
 369         if (!na->addr_count)
 370                 return -ENOENT;
 371
 372         *mark = na->mark;
 373
 374         if (sas_out)
 375                 memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
 376
 377         if (!sa_out)
 378                 return 0;
 379
 380         if (dlm_local_addr[0]->ss_family == AF_INET) {
 381                 struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
 382                 struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
 383                 ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
 384         } else {
 385                 struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
 386                 struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
 387                 ret6->sin6_addr = in6->sin6_addr;
 388         }
 389
 390         return 0;
 391 }
 392
 393 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
 394                           unsigned int *mark)
 395 {
 396         struct dlm_node_addr *na;
 397         int rv = -EEXIST;
 398         int addr_i;
 399
 400         spin_lock(&dlm_node_addrs_spin);
 401         list_for_each_entry(na, &dlm_node_addrs, list) {
 402                 if (!na->addr_count)
 403                         continue;
 404
 405                 for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
 406                         if (addr_compare(na->addr[addr_i], addr)) {
 407                                 *nodeid = na->nodeid;
 408                                 *mark = na->mark;
 409                                 rv = 0;
 410                                 goto unlock;
 411                         }
 412                 }
 413         }
 414 unlock:
 415         spin_unlock(&dlm_node_addrs_spin);
 416         return rv;
 417 }
 418
 419 /* caller need to held dlm_node_addrs_spin lock */
 420 static bool dlm_lowcomms_na_has_addr(const struct dlm_node_addr *na,
 421                                      const struct sockaddr_storage *addr)
 422 {
 423         int i;
 424
 425         for (i = 0; i < na->addr_count; i++) {
 426                 if (addr_compare(na->addr[i], addr))
 427                         return true;
 428         }
 429
 430         return false;
 431 }
 432
 433 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 434 {
 435         struct sockaddr_storage *new_addr;
 436         struct dlm_node_addr *new_node, *na;
 437         bool ret;
 438
 439         new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
 440         if (!new_node)
 441                 return -ENOMEM;
 442
 443         new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
 444         if (!new_addr) {
 445                 kfree(new_node);
 446                 return -ENOMEM;
 447         }
 448
 449         memcpy(new_addr, addr, len);
 450
 451         spin_lock(&dlm_node_addrs_spin);
 452         na = find_node_addr(nodeid);
 453         if (!na) {
 454                 new_node->nodeid = nodeid;
 455                 new_node->addr[0] = new_addr;
 456                 new_node->addr_count = 1;
 457                 new_node->mark = dlm_config.ci_mark;
 458                 list_add(&new_node->list, &dlm_node_addrs);
 459                 spin_unlock(&dlm_node_addrs_spin);
 460                 return 0;
 461         }
 462
 463         ret = dlm_lowcomms_na_has_addr(na, addr);
 464         if (ret) {
 465                 spin_unlock(&dlm_node_addrs_spin);
 466                 kfree(new_addr);
 467                 kfree(new_node);
 468                 return -EEXIST;
 469         }
 470
 471         if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
 472                 spin_unlock(&dlm_node_addrs_spin);
 473                 kfree(new_addr);
 474                 kfree(new_node);
 475                 return -ENOSPC;
 476         }
 477
 478         na->addr[na->addr_count++] = new_addr;
 479         spin_unlock(&dlm_node_addrs_spin);
 480         kfree(new_node);
 481         return 0;
 482 }
 483
 484 /* Data available on socket or listen socket received a connect */
 485 static void lowcomms_data_ready(struct sock *sk)
 486 {
 487         struct connection *con;
 488
 489         read_lock_bh(&sk->sk_callback_lock);
 490         con = sock2con(sk);
 491         if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
 492                 queue_work(recv_workqueue, &con->rwork);
 493         read_unlock_bh(&sk->sk_callback_lock);
 494 }
 495
 496 static void lowcomms_listen_data_ready(struct sock *sk)
 497 {
 498         if (!dlm_allow_conn)
 499                 return;
 500
 501         queue_work(recv_workqueue, &listen_con.rwork);
 502 }
 503
 504 static void lowcomms_write_space(struct sock *sk)
 505 {
 506         struct connection *con;
 507
 508         read_lock_bh(&sk->sk_callback_lock);
 509         con = sock2con(sk);
 510         if (!con)
 511                 goto out;
 512
 513         if (!test_and_set_bit(CF_CONNECTED, &con->flags)) {
 514                 log_print("successful connected to node %d", con->nodeid);
 515                 queue_work(send_workqueue, &con->swork);
 516                 goto out;
 517         }
 518
 519         clear_bit(SOCK_NOSPACE, &con->sock->flags);
 520
 521         if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
 522                 con->sock->sk->sk_write_pending--;
 523                 clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
 524         }
 525
 526         queue_work(send_workqueue, &con->swork);
 527 out:
 528         read_unlock_bh(&sk->sk_callback_lock);
 529 }
 530
 531 static inline void lowcomms_connect_sock(struct connection *con)
 532 {
 533         if (test_bit(CF_CLOSE, &con->flags))
 534                 return;
 535         queue_work(send_workqueue, &con->swork);
 536         cond_resched();
 537 }
 538
 539 static void lowcomms_state_change(struct sock *sk)
 540 {
 541         /* SCTP layer is not calling sk_data_ready when the connection
 542          * is done, so we catch the signal through here. Also, it
 543          * doesn't switch socket state when entering shutdown, so we
 544          * skip the write in that case.
 545          */
 546         if (sk->sk_shutdown) {
 547                 if (sk->sk_shutdown == RCV_SHUTDOWN)
 548                         lowcomms_data_ready(sk);
 549         } else if (sk->sk_state == TCP_ESTABLISHED) {
 550                 lowcomms_write_space(sk);
 551         }
 552 }
 553
 554 int dlm_lowcomms_connect_node(int nodeid)
 555 {
 556         struct connection *con;
 557         int idx;
 558
 559         if (nodeid == dlm_our_nodeid())
 560                 return 0;
 561
 562         idx = srcu_read_lock(&connections_srcu);
 563         con = nodeid2con(nodeid, GFP_NOFS);
 564         if (!con) {
 565                 srcu_read_unlock(&connections_srcu, idx);
 566                 return -ENOMEM;
 567         }
 568
 569         lowcomms_connect_sock(con);
 570         srcu_read_unlock(&connections_srcu, idx);
 571
 572         return 0;
 573 }
 574
 575 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
 576 {
 577         struct dlm_node_addr *na;
 578
 579         spin_lock(&dlm_node_addrs_spin);
 580         na = find_node_addr(nodeid);
 581         if (!na) {
 582                 spin_unlock(&dlm_node_addrs_spin);
 583                 return -ENOENT;
 584         }
 585
 586         na->mark = mark;
 587         spin_unlock(&dlm_node_addrs_spin);
 588
 589         return 0;
 590 }
 591
 592 static void lowcomms_error_report(struct sock *sk)
 593 {
 594         struct connection *con;
 595         void (*orig_report)(struct sock *) = NULL;
 596         struct inet_sock *inet;
 597
 598         read_lock_bh(&sk->sk_callback_lock);
 599         con = sock2con(sk);
 600         if (con == NULL)
 601                 goto out;
 602
 603         orig_report = listen_sock.sk_error_report;
 604
 605         inet = inet_sk(sk);
 606         switch (sk->sk_family) {
 607         case AF_INET:
 608                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 609                                    "sending to node %d at %pI4, dport %d, "
 610                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 611                                    con->nodeid, &inet->inet_daddr,
 612                                    ntohs(inet->inet_dport), sk->sk_err,
 613                                    sk->sk_err_soft);
 614                 break;
 615 #if IS_ENABLED(CONFIG_IPV6)
 616         case AF_INET6:
 617                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 618                                    "sending to node %d at %pI6c, "
 619                                    "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(),
 620                                    con->nodeid, &sk->sk_v6_daddr,
 621                                    ntohs(inet->inet_dport), sk->sk_err,
 622                                    sk->sk_err_soft);
 623                 break;
 624 #endif
 625         default:
 626                 printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
 627                                    "invalid socket family %d set, "
 628                                    "sk_err=%d/%d\n", dlm_our_nodeid(),
 629                                    sk->sk_family, sk->sk_err, sk->sk_err_soft);
 630                 goto out;
 631         }
 632
 633         /* below sendcon only handling */
 634         if (test_bit(CF_IS_OTHERCON, &con->flags))
 635                 con = con->sendcon;
 636
 637         switch (sk->sk_err) {
 638         case ECONNREFUSED:
 639                 set_bit(CF_DELAY_CONNECT, &con->flags);
 640                 break;
 641         default:
 642                 break;
 643         }
 644
 645         if (!test_and_set_bit(CF_RECONNECT, &con->flags))
 646                 queue_work(send_workqueue, &con->swork);
 647
 648 out:
 649         read_unlock_bh(&sk->sk_callback_lock);
 650         if (orig_report)
 651                 orig_report(sk);
 652 }
 653
 654 /* Note: sk_callback_lock must be locked before calling this function. */
 655 static void save_listen_callbacks(struct socket *sock)
 656 {
 657         struct sock *sk = sock->sk;
 658
 659         listen_sock.sk_data_ready = sk->sk_data_ready;
 660         listen_sock.sk_state_change = sk->sk_state_change;
 661         listen_sock.sk_write_space = sk->sk_write_space;
 662         listen_sock.sk_error_report = sk->sk_error_report;
 663 }
 664
 665 static void restore_callbacks(struct socket *sock)
 666 {
 667         struct sock *sk = sock->sk;
 668
 669         write_lock_bh(&sk->sk_callback_lock);
 670         sk->sk_user_data = NULL;
 671         sk->sk_data_ready = listen_sock.sk_data_ready;
 672         sk->sk_state_change = listen_sock.sk_state_change;
 673         sk->sk_write_space = listen_sock.sk_write_space;
 674         sk->sk_error_report = listen_sock.sk_error_report;
 675         write_unlock_bh(&sk->sk_callback_lock);
 676 }
 677
 678 static void add_listen_sock(struct socket *sock, struct listen_connection *con)
 679 {
 680         struct sock *sk = sock->sk;
 681
 682         write_lock_bh(&sk->sk_callback_lock);
 683         save_listen_callbacks(sock);
 684         con->sock = sock;
 685
 686         sk->sk_user_data = con;
 687         sk->sk_allocation = GFP_NOFS;
 688         /* Install a data_ready callback */
 689         sk->sk_data_ready = lowcomms_listen_data_ready;
 690         write_unlock_bh(&sk->sk_callback_lock);
 691 }
 692
 693 /* Make a socket active */
 694 static void add_sock(struct socket *sock, struct connection *con)
 695 {
 696         struct sock *sk = sock->sk;
 697
 698         write_lock_bh(&sk->sk_callback_lock);
 699         con->sock = sock;
 700
 701         sk->sk_user_data = con;
 702         /* Install a data_ready callback */
 703         sk->sk_data_ready = lowcomms_data_ready;
 704         sk->sk_write_space = lowcomms_write_space;
 705         sk->sk_state_change = lowcomms_state_change;
 706         sk->sk_allocation = GFP_NOFS;
 707         sk->sk_error_report = lowcomms_error_report;
 708         write_unlock_bh(&sk->sk_callback_lock);
 709 }
 710
 711 /* Add the port number to an IPv6 or 4 sockaddr and return the address
 712    length */
 713 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 714                           int *addr_len)
 715 {
 716         saddr->ss_family =  dlm_local_addr[0]->ss_family;
 717         if (saddr->ss_family == AF_INET) {
 718                 struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
 719                 in4_addr->sin_port = cpu_to_be16(port);
 720                 *addr_len = sizeof(struct sockaddr_in);
 721                 memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
 722         } else {
 723                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
 724                 in6_addr->sin6_port = cpu_to_be16(port);
 725                 *addr_len = sizeof(struct sockaddr_in6);
 726         }
 727         memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 728 }
 729
 730 static void dlm_page_release(struct kref *kref)
 731 {
 732         struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
 733                                                   ref);
 734
 735         __free_page(e->page);
 736         kfree(e);
 737 }
 738
 739 static void dlm_msg_release(struct kref *kref)
 740 {
 741         struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
 742
 743         kref_put(&msg->entry->ref, dlm_page_release);
 744         kfree(msg);
 745 }
 746
 747 static void free_entry(struct writequeue_entry *e)
 748 {
 749         struct dlm_msg *msg, *tmp;
 750
 751         list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
 752                 if (msg->orig_msg) {
 753                         msg->orig_msg->retransmit = false;
 754                         kref_put(&msg->orig_msg->ref, dlm_msg_release);
 755                 }
 756
 757                 list_del(&msg->list);
 758                 kref_put(&msg->ref, dlm_msg_release);
 759         }
 760
 761         list_del(&e->list);
 762         atomic_dec(&e->con->writequeue_cnt);
 763         kref_put(&e->ref, dlm_page_release);
 764 }
 765
 766 static void dlm_close_sock(struct socket **sock)
 767 {
 768         if (*sock) {
 769                 restore_callbacks(*sock);
 770                 sock_release(*sock);
 771                 *sock = NULL;
 772         }
 773 }
 774
 775 /* Close a remote connection and tidy up */
 776 static void close_connection(struct connection *con, bool and_other,
 777                              bool tx, bool rx)
 778 {
 779         bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
 780         struct writequeue_entry *e;
 781
 782         if (tx && !closing && cancel_work_sync(&con->swork)) {
 783                 log_print("canceled swork for node %d", con->nodeid);
 784                 clear_bit(CF_WRITE_PENDING, &con->flags);
 785         }
 786         if (rx && !closing && cancel_work_sync(&con->rwork)) {
 787                 log_print("canceled rwork for node %d", con->nodeid);
 788                 clear_bit(CF_READ_PENDING, &con->flags);
 789         }
 790
 791         mutex_lock(&con->sock_mutex);
 792         dlm_close_sock(&con->sock);
 793
 794         if (con->othercon && and_other) {
 795                 /* Will only re-enter once. */
 796                 close_connection(con->othercon, false, tx, rx);
 797         }
 798
 799         /* if we send a writequeue entry only a half way, we drop the
 800          * whole entry because reconnection and that we not start of the
 801          * middle of a msg which will confuse the other end.
 802          *
 803          * we can always drop messages because retransmits, but what we
 804          * cannot allow is to transmit half messages which may be processed
 805          * at the other side.
 806          *
 807          * our policy is to start on a clean state when disconnects, we don't
 808          * know what's send/received on transport layer in this case.
 809          */
 810         spin_lock(&con->writequeue_lock);
 811         if (!list_empty(&con->writequeue)) {
 812                 e = list_first_entry(&con->writequeue, struct writequeue_entry,
 813                                      list);
 814                 if (e->dirty)
 815                         free_entry(e);
 816         }
 817         spin_unlock(&con->writequeue_lock);
 818
 819         con->rx_leftover = 0;
 820         con->retries = 0;
 821         clear_bit(CF_APP_LIMITED, &con->flags);
 822         clear_bit(CF_CONNECTED, &con->flags);
 823         clear_bit(CF_DELAY_CONNECT, &con->flags);
 824         clear_bit(CF_RECONNECT, &con->flags);
 825         clear_bit(CF_EOF, &con->flags);
 826         mutex_unlock(&con->sock_mutex);
 827         clear_bit(CF_CLOSING, &con->flags);
 828 }
 829
 830 static void shutdown_connection(struct connection *con)
 831 {
 832         int ret;
 833
 834         flush_work(&con->swork);
 835
 836         mutex_lock(&con->sock_mutex);
 837         /* nothing to shutdown */
 838         if (!con->sock) {
 839                 mutex_unlock(&con->sock_mutex);
 840                 return;
 841         }
 842
 843         set_bit(CF_SHUTDOWN, &con->flags);
 844         ret = kernel_sock_shutdown(con->sock, SHUT_WR);
 845         mutex_unlock(&con->sock_mutex);
 846         if (ret) {
 847                 log_print("Connection %p failed to shutdown: %d will force close",
 848                           con, ret);
 849                 goto force_close;
 850         } else {
 851                 ret = wait_event_timeout(con->shutdown_wait,
 852                                          !test_bit(CF_SHUTDOWN, &con->flags),
 853                                          DLM_SHUTDOWN_WAIT_TIMEOUT);
 854                 if (ret == 0) {
 855                         log_print("Connection %p shutdown timed out, will force close",
 856                                   con);
 857                         goto force_close;
 858                 }
 859         }
 860
 861         return;
 862
 863 force_close:
 864         clear_bit(CF_SHUTDOWN, &con->flags);
 865         close_connection(con, false, true, true);
 866 }
 867
 868 static void dlm_tcp_shutdown(struct connection *con)
 869 {
 870         if (con->othercon)
 871                 shutdown_connection(con->othercon);
 872         shutdown_connection(con);
 873 }
 874
 875 static int con_realloc_receive_buf(struct connection *con, int newlen)
 876 {
 877         unsigned char *newbuf;
 878
 879         newbuf = kmalloc(newlen, GFP_NOFS);
 880         if (!newbuf)
 881                 return -ENOMEM;
 882
 883         /* copy any leftover from last receive */
 884         if (con->rx_leftover)
 885                 memmove(newbuf, con->rx_buf, con->rx_leftover);
 886
 887         /* swap to new buffer space */
 888         kfree(con->rx_buf);
 889         con->rx_buflen = newlen;
 890         con->rx_buf = newbuf;
 891
 892         return 0;
 893 }
 894
 895 /* Data received from remote end */
 896 static int receive_from_sock(struct connection *con)
 897 {
 898         struct msghdr msg;
 899         struct kvec iov;
 900         int ret, buflen;
 901
 902         mutex_lock(&con->sock_mutex);
 903
 904         if (con->sock == NULL) {
 905                 ret = -EAGAIN;
 906                 goto out_close;
 907         }
 908
 909         /* realloc if we get new buffer size to read out */
 910         buflen = dlm_config.ci_buffer_size;
 911         if (con->rx_buflen != buflen && con->rx_leftover <= buflen) {
 912                 ret = con_realloc_receive_buf(con, buflen);
 913                 if (ret < 0)
 914                         goto out_resched;
 915         }
 916
 917         for (;;) {
 918                 /* calculate new buffer parameter regarding last receive and
 919                  * possible leftover bytes
 920                  */
 921                 iov.iov_base = con->rx_buf + con->rx_leftover;
 922                 iov.iov_len = con->rx_buflen - con->rx_leftover;
 923
 924                 memset(&msg, 0, sizeof(msg));
 925                 msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
 926                 ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
 927                                      msg.msg_flags);
 928                 if (ret == -EAGAIN)
 929                         break;
 930                 else if (ret <= 0)
 931                         goto out_close;
 932
 933                 /* new buflen according readed bytes and leftover from last receive */
 934                 buflen = ret + con->rx_leftover;
 935                 ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen);
 936                 if (ret < 0)
 937                         goto out_close;
 938
 939                 /* calculate leftover bytes from process and put it into begin of
 940                  * the receive buffer, so next receive we have the full message
 941                  * at the start address of the receive buffer.
 942                  */
 943                 con->rx_leftover = buflen - ret;
 944                 if (con->rx_leftover) {
 945                         memmove(con->rx_buf, con->rx_buf + ret,
 946                                 con->rx_leftover);
 947                 }
 948         }
 949
 950         dlm_midcomms_receive_done(con->nodeid);
 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 (dlm_proto_ops->eof_condition &&
 966                     dlm_proto_ops->eof_condition(con)) {
 967                         set_bit(CF_EOF, &con->flags);
 968                         mutex_unlock(&con->sock_mutex);
 969                 } else {
 970                         mutex_unlock(&con->sock_mutex);
 971                         close_connection(con, false, true, false);
 972
 973                         /* handling for tcp shutdown */
 974                         clear_bit(CF_SHUTDOWN, &con->flags);
 975                         wake_up(&con->shutdown_wait);
 976                 }
 977
 978                 /* signal to breaking receive worker */
 979                 ret = -1;
 980         } else {
 981                 mutex_unlock(&con->sock_mutex);
 982         }
 983         return ret;
 984 }
 985
 986 /* Listening socket is busy, accept a connection */
 987 static int accept_from_sock(struct listen_connection *con)
 988 {
 989         int result;
 990         struct sockaddr_storage peeraddr;
 991         struct socket *newsock;
 992         int len, idx;
 993         int nodeid;
 994         struct connection *newcon;
 995         struct connection *addcon;
 996         unsigned int mark;
 997
 998         if (!con->sock)
 999                 return -ENOTCONN;
1000
1001         result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
1002         if (result < 0)
1003                 goto accept_err;
1004
1005         /* Get the connected socket's peer */
1006         memset(&peeraddr, 0, sizeof(peeraddr));
1007         len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
1008         if (len < 0) {
1009                 result = -ECONNABORTED;
1010                 goto accept_err;
1011         }
1012
1013         /* Get the new node's NODEID */
1014         make_sockaddr(&peeraddr, 0, &len);
1015         if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
1016                 unsigned char *b=(unsigned char *)&peeraddr;
1017                 log_print("connect from non cluster node");
1018                 print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
1019                                      b, sizeof(struct sockaddr_storage));
1020                 sock_release(newsock);
1021                 return -1;
1022         }
1023
1024         log_print("got connection from %d", nodeid);
1025
1026         /*  Check to see if we already have a connection to this node. This
1027          *  could happen if the two nodes initiate a connection at roughly
1028          *  the same time and the connections cross on the wire.
1029          *  In this case we store the incoming one in "othercon"
1030          */
1031         idx = srcu_read_lock(&connections_srcu);
1032         newcon = nodeid2con(nodeid, GFP_NOFS);
1033         if (!newcon) {
1034                 srcu_read_unlock(&connections_srcu, idx);
1035                 result = -ENOMEM;
1036                 goto accept_err;
1037         }
1038
1039         sock_set_mark(newsock->sk, mark);
1040
1041         mutex_lock(&newcon->sock_mutex);
1042         if (newcon->sock) {
1043                 struct connection *othercon = newcon->othercon;
1044
1045                 if (!othercon) {
1046                         othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1047                         if (!othercon) {
1048                                 log_print("failed to allocate incoming socket");
1049                                 mutex_unlock(&newcon->sock_mutex);
1050                                 srcu_read_unlock(&connections_srcu, idx);
1051                                 result = -ENOMEM;
1052                                 goto accept_err;
1053                         }
1054
1055                         result = dlm_con_init(othercon, nodeid);
1056                         if (result < 0) {
1057                                 kfree(othercon);
1058                                 mutex_unlock(&newcon->sock_mutex);
1059                                 srcu_read_unlock(&connections_srcu, idx);
1060                                 goto accept_err;
1061                         }
1062
1063                         lockdep_set_subclass(&othercon->sock_mutex, 1);
1064                         set_bit(CF_IS_OTHERCON, &othercon->flags);
1065                         newcon->othercon = othercon;
1066                         othercon->sendcon = newcon;
1067                 } else {
1068                         /* close other sock con if we have something new */
1069                         close_connection(othercon, false, true, false);
1070                 }
1071
1072                 mutex_lock(&othercon->sock_mutex);
1073                 add_sock(newsock, othercon);
1074                 addcon = othercon;
1075                 mutex_unlock(&othercon->sock_mutex);
1076         }
1077         else {
1078                 /* accept copies the sk after we've saved the callbacks, so we
1079                    don't want to save them a second time or comm errors will
1080                    result in calling sk_error_report recursively. */
1081                 add_sock(newsock, newcon);
1082                 addcon = newcon;
1083         }
1084
1085         set_bit(CF_CONNECTED, &addcon->flags);
1086         mutex_unlock(&newcon->sock_mutex);
1087
1088         /*
1089          * Add it to the active queue in case we got data
1090          * between processing the accept adding the socket
1091          * to the read_sockets list
1092          */
1093         if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
1094                 queue_work(recv_workqueue, &addcon->rwork);
1095
1096         srcu_read_unlock(&connections_srcu, idx);
1097
1098         return 0;
1099
1100 accept_err:
1101         if (newsock)
1102                 sock_release(newsock);
1103
1104         if (result != -EAGAIN)
1105                 log_print("error accepting connection from node: %d", result);
1106         return result;
1107 }
1108
1109 /*
1110  * writequeue_entry_complete - try to delete and free write queue entry
1111  * @e: write queue entry to try to delete
1112  * @completed: bytes completed
1113  *
1114  * writequeue_lock must be held.
1115  */
1116 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1117 {
1118         e->offset += completed;
1119         e->len -= completed;
1120         /* signal that page was half way transmitted */
1121         e->dirty = true;
1122
1123         if (e->len == 0 && e->users == 0)
1124                 free_entry(e);
1125 }
1126
1127 /*
1128  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1129  */
1130 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1131 {
1132         struct sockaddr_storage localaddr;
1133         struct sockaddr *addr = (struct sockaddr *)&localaddr;
1134         int i, addr_len, result = 0;
1135
1136         for (i = 0; i < dlm_local_count; i++) {
1137                 memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
1138                 make_sockaddr(&localaddr, port, &addr_len);
1139
1140                 if (!i)
1141                         result = kernel_bind(sock, addr, addr_len);
1142                 else
1143                         result = sock_bind_add(sock->sk, addr, addr_len);
1144
1145                 if (result < 0) {
1146                         log_print("Can't bind to %d addr number %d, %d.\n",
1147                                   port, i + 1, result);
1148                         break;
1149                 }
1150         }
1151         return result;
1152 }
1153
1154 /* Get local addresses */
1155 static void init_local(void)
1156 {
1157         struct sockaddr_storage sas, *addr;
1158         int i;
1159
1160         dlm_local_count = 0;
1161         for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1162                 if (dlm_our_addr(&sas, i))
1163                         break;
1164
1165                 addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1166                 if (!addr)
1167                         break;
1168                 dlm_local_addr[dlm_local_count++] = addr;
1169         }
1170 }
1171
1172 static void deinit_local(void)
1173 {
1174         int i;
1175
1176         for (i = 0; i < dlm_local_count; i++)
1177                 kfree(dlm_local_addr[i]);
1178 }
1179
1180 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1181                                                      gfp_t allocation)
1182 {
1183         struct writequeue_entry *entry;
1184
1185         entry = kzalloc(sizeof(*entry), allocation);
1186         if (!entry)
1187                 return NULL;
1188
1189         entry->page = alloc_page(allocation | __GFP_ZERO);
1190         if (!entry->page) {
1191                 kfree(entry);
1192                 return NULL;
1193         }
1194
1195         entry->con = con;
1196         entry->users = 1;
1197         kref_init(&entry->ref);
1198         INIT_LIST_HEAD(&entry->msgs);
1199
1200         return entry;
1201 }
1202
1203 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1204                                              gfp_t allocation, char **ppc,
1205                                              void (*cb)(struct dlm_mhandle *mh),
1206                                              struct dlm_mhandle *mh)
1207 {
1208         struct writequeue_entry *e;
1209
1210         spin_lock(&con->writequeue_lock);
1211         if (!list_empty(&con->writequeue)) {
1212                 e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1213                 if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1214                         kref_get(&e->ref);
1215
1216                         *ppc = page_address(e->page) + e->end;
1217                         if (cb)
1218                                 cb(mh);
1219
1220                         e->end += len;
1221                         e->users++;
1222                         spin_unlock(&con->writequeue_lock);
1223
1224                         return e;
1225                 }
1226         }
1227         spin_unlock(&con->writequeue_lock);
1228
1229         e = new_writequeue_entry(con, allocation);
1230         if (!e)
1231                 return NULL;
1232
1233         kref_get(&e->ref);
1234         *ppc = page_address(e->page);
1235         e->end += len;
1236         atomic_inc(&con->writequeue_cnt);
1237
1238         spin_lock(&con->writequeue_lock);
1239         if (cb)
1240                 cb(mh);
1241
1242         list_add_tail(&e->list, &con->writequeue);
1243         spin_unlock(&con->writequeue_lock);
1244
1245         return e;
1246 };
1247
1248 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1249                                                 gfp_t allocation, char **ppc,
1250                                                 void (*cb)(struct dlm_mhandle *mh),
1251                                                 struct dlm_mhandle *mh)
1252 {
1253         struct writequeue_entry *e;
1254         struct dlm_msg *msg;
1255         bool sleepable;
1256
1257         msg = kzalloc(sizeof(*msg), allocation);
1258         if (!msg)
1259                 return NULL;
1260
1261         /* this mutex is being used as a wait to avoid multiple "fast"
1262          * new writequeue page list entry allocs in new_wq_entry in
1263          * normal operation which is sleepable context. Without it
1264          * we could end in multiple writequeue entries with one
1265          * dlm message because multiple callers were waiting at
1266          * the writequeue_lock in new_wq_entry().
1267          */
1268         sleepable = gfpflags_normal_context(allocation);
1269         if (sleepable)
1270                 mutex_lock(&con->wq_alloc);
1271
1272         kref_init(&msg->ref);
1273
1274         e = new_wq_entry(con, len, allocation, ppc, cb, mh);
1275         if (!e) {
1276                 if (sleepable)
1277                         mutex_unlock(&con->wq_alloc);
1278
1279                 kfree(msg);
1280                 return NULL;
1281         }
1282
1283         if (sleepable)
1284                 mutex_unlock(&con->wq_alloc);
1285
1286         msg->ppc = *ppc;
1287         msg->len = len;
1288         msg->entry = e;
1289
1290         return msg;
1291 }
1292
1293 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
1294                                      char **ppc, void (*cb)(struct dlm_mhandle *mh),
1295                                      struct dlm_mhandle *mh)
1296 {
1297         struct connection *con;
1298         struct dlm_msg *msg;
1299         int idx;
1300
1301         if (len > DLM_MAX_SOCKET_BUFSIZE ||
1302             len < sizeof(struct dlm_header)) {
1303                 BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1304                 log_print("failed to allocate a buffer of size %d", len);
1305                 WARN_ON(1);
1306                 return NULL;
1307         }
1308
1309         idx = srcu_read_lock(&connections_srcu);
1310         con = nodeid2con(nodeid, allocation);
1311         if (!con) {
1312                 srcu_read_unlock(&connections_srcu, idx);
1313                 return NULL;
1314         }
1315
1316         msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, mh);
1317         if (!msg) {
1318                 srcu_read_unlock(&connections_srcu, idx);
1319                 return NULL;
1320         }
1321
1322         /* for dlm_lowcomms_commit_msg() */
1323         kref_get(&msg->ref);
1324         /* we assume if successful commit must called */
1325         msg->idx = idx;
1326         return msg;
1327 }
1328
1329 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1330 {
1331         struct writequeue_entry *e = msg->entry;
1332         struct connection *con = e->con;
1333         int users;
1334
1335         spin_lock(&con->writequeue_lock);
1336         kref_get(&msg->ref);
1337         list_add(&msg->list, &e->msgs);
1338
1339         users = --e->users;
1340         if (users)
1341                 goto out;
1342
1343         e->len = DLM_WQ_LENGTH_BYTES(e);
1344         spin_unlock(&con->writequeue_lock);
1345
1346         queue_work(send_workqueue, &con->swork);
1347         return;
1348
1349 out:
1350         spin_unlock(&con->writequeue_lock);
1351         return;
1352 }
1353
1354 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1355 {
1356         _dlm_lowcomms_commit_msg(msg);
1357         srcu_read_unlock(&connections_srcu, msg->idx);
1358         /* because dlm_lowcomms_new_msg() */
1359         kref_put(&msg->ref, dlm_msg_release);
1360 }
1361
1362 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1363 {
1364         kref_put(&msg->ref, dlm_msg_release);
1365 }
1366
1367 /* does not held connections_srcu, usage workqueue only */
1368 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1369 {
1370         struct dlm_msg *msg_resend;
1371         char *ppc;
1372
1373         if (msg->retransmit)
1374                 return 1;
1375
1376         msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
1377                                               GFP_ATOMIC, &ppc, NULL, NULL);
1378         if (!msg_resend)
1379                 return -ENOMEM;
1380
1381         msg->retransmit = true;
1382         kref_get(&msg->ref);
1383         msg_resend->orig_msg = msg;
1384
1385         memcpy(ppc, msg->ppc, msg->len);
1386         _dlm_lowcomms_commit_msg(msg_resend);
1387         dlm_lowcomms_put_msg(msg_resend);
1388
1389         return 0;
1390 }
1391
1392 /* Send a message */
1393 static void send_to_sock(struct connection *con)
1394 {
1395         const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1396         struct writequeue_entry *e;
1397         int len, offset, ret;
1398         int count = 0;
1399
1400         mutex_lock(&con->sock_mutex);
1401         if (con->sock == NULL)
1402                 goto out_connect;
1403
1404         spin_lock(&con->writequeue_lock);
1405         for (;;) {
1406                 e = con_next_wq(con);
1407                 if (!e)
1408                         break;
1409
1410                 e = list_first_entry(&con->writequeue, struct writequeue_entry, list);
1411                 len = e->len;
1412                 offset = e->offset;
1413                 BUG_ON(len == 0 && e->users == 0);
1414                 spin_unlock(&con->writequeue_lock);
1415
1416                 ret = kernel_sendpage(con->sock, e->page, offset, len,
1417                                       msg_flags);
1418                 if (ret == -EAGAIN || ret == 0) {
1419                         if (ret == -EAGAIN &&
1420                             test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1421                             !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1422                                 /* Notify TCP that we're limited by the
1423                                  * application window size.
1424                                  */
1425                                 set_bit(SOCK_NOSPACE, &con->sock->flags);
1426                                 con->sock->sk->sk_write_pending++;
1427                         }
1428                         cond_resched();
1429                         goto out;
1430                 } else if (ret < 0)
1431                         goto out;
1432
1433                 /* Don't starve people filling buffers */
1434                 if (++count >= MAX_SEND_MSG_COUNT) {
1435                         cond_resched();
1436                         count = 0;
1437                 }
1438
1439                 spin_lock(&con->writequeue_lock);
1440                 writequeue_entry_complete(e, ret);
1441         }
1442         spin_unlock(&con->writequeue_lock);
1443
1444         /* close if we got EOF */
1445         if (test_and_clear_bit(CF_EOF, &con->flags)) {
1446                 mutex_unlock(&con->sock_mutex);
1447                 close_connection(con, false, false, true);
1448
1449                 /* handling for tcp shutdown */
1450                 clear_bit(CF_SHUTDOWN, &con->flags);
1451                 wake_up(&con->shutdown_wait);
1452         } else {
1453                 mutex_unlock(&con->sock_mutex);
1454         }
1455
1456         return;
1457
1458 out:
1459         mutex_unlock(&con->sock_mutex);
1460         return;
1461
1462 out_connect:
1463         mutex_unlock(&con->sock_mutex);
1464         queue_work(send_workqueue, &con->swork);
1465         cond_resched();
1466 }
1467
1468 static void clean_one_writequeue(struct connection *con)
1469 {
1470         struct writequeue_entry *e, *safe;
1471
1472         spin_lock(&con->writequeue_lock);
1473         list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1474                 free_entry(e);
1475         }
1476         spin_unlock(&con->writequeue_lock);
1477 }
1478
1479 /* Called from recovery when it knows that a node has
1480    left the cluster */
1481 int dlm_lowcomms_close(int nodeid)
1482 {
1483         struct connection *con;
1484         struct dlm_node_addr *na;
1485         int idx;
1486
1487         log_print("closing connection to node %d", nodeid);
1488         idx = srcu_read_lock(&connections_srcu);
1489         con = nodeid2con(nodeid, 0);
1490         if (con) {
1491                 set_bit(CF_CLOSE, &con->flags);
1492                 close_connection(con, true, true, true);
1493                 clean_one_writequeue(con);
1494                 if (con->othercon)
1495                         clean_one_writequeue(con->othercon);
1496         }
1497         srcu_read_unlock(&connections_srcu, idx);
1498
1499         spin_lock(&dlm_node_addrs_spin);
1500         na = find_node_addr(nodeid);
1501         if (na) {
1502                 list_del(&na->list);
1503                 while (na->addr_count--)
1504                         kfree(na->addr[na->addr_count]);
1505                 kfree(na);
1506         }
1507         spin_unlock(&dlm_node_addrs_spin);
1508
1509         return 0;
1510 }
1511
1512 /* Receive workqueue function */
1513 static void process_recv_sockets(struct work_struct *work)
1514 {
1515         struct connection *con = container_of(work, struct connection, rwork);
1516
1517         clear_bit(CF_READ_PENDING, &con->flags);
1518         receive_from_sock(con);
1519 }
1520
1521 static void process_listen_recv_socket(struct work_struct *work)
1522 {
1523         int ret;
1524
1525         do {
1526                 ret = accept_from_sock(&listen_con);
1527         } while (!ret);
1528 }
1529
1530 static void dlm_connect(struct connection *con)
1531 {
1532         struct sockaddr_storage addr;
1533         int result, addr_len;
1534         struct socket *sock;
1535         unsigned int mark;
1536
1537         /* Some odd races can cause double-connects, ignore them */
1538         if (con->retries++ > MAX_CONNECT_RETRIES)
1539                 return;
1540
1541         if (con->sock) {
1542                 log_print("node %d already connected.", con->nodeid);
1543                 return;
1544         }
1545
1546         memset(&addr, 0, sizeof(addr));
1547         result = nodeid_to_addr(con->nodeid, &addr, NULL,
1548                                 dlm_proto_ops->try_new_addr, &mark);
1549         if (result < 0) {
1550                 log_print("no address for nodeid %d", con->nodeid);
1551                 return;
1552         }
1553
1554         /* Create a socket to communicate with */
1555         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1556                                   SOCK_STREAM, dlm_proto_ops->proto, &sock);
1557         if (result < 0)
1558                 goto socket_err;
1559
1560         sock_set_mark(sock->sk, mark);
1561         dlm_proto_ops->sockopts(sock);
1562
1563         add_sock(sock, con);
1564
1565         result = dlm_proto_ops->bind(sock);
1566         if (result < 0)
1567                 goto add_sock_err;
1568
1569         log_print_ratelimited("connecting to %d", con->nodeid);
1570         make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len);
1571         result = dlm_proto_ops->connect(con, sock, (struct sockaddr *)&addr,
1572                                         addr_len);
1573         if (result < 0)
1574                 goto add_sock_err;
1575
1576         return;
1577
1578 add_sock_err:
1579         dlm_close_sock(&con->sock);
1580
1581 socket_err:
1582         /*
1583          * Some errors are fatal and this list might need adjusting. For other
1584          * errors we try again until the max number of retries is reached.
1585          */
1586         if (result != -EHOSTUNREACH &&
1587             result != -ENETUNREACH &&
1588             result != -ENETDOWN &&
1589             result != -EINVAL &&
1590             result != -EPROTONOSUPPORT) {
1591                 log_print("connect %d try %d error %d", con->nodeid,
1592                           con->retries, result);
1593                 msleep(1000);
1594                 lowcomms_connect_sock(con);
1595         }
1596 }
1597
1598 /* Send workqueue function */
1599 static void process_send_sockets(struct work_struct *work)
1600 {
1601         struct connection *con = container_of(work, struct connection, swork);
1602
1603         WARN_ON(test_bit(CF_IS_OTHERCON, &con->flags));
1604
1605         clear_bit(CF_WRITE_PENDING, &con->flags);
1606
1607         if (test_and_clear_bit(CF_RECONNECT, &con->flags)) {
1608                 close_connection(con, false, false, true);
1609                 dlm_midcomms_unack_msg_resend(con->nodeid);
1610         }
1611
1612         if (con->sock == NULL) {
1613                 if (test_and_clear_bit(CF_DELAY_CONNECT, &con->flags))
1614                         msleep(1000);
1615
1616                 mutex_lock(&con->sock_mutex);
1617                 dlm_connect(con);
1618                 mutex_unlock(&con->sock_mutex);
1619         }
1620
1621         if (!list_empty(&con->writequeue))
1622                 send_to_sock(con);
1623 }
1624
1625 static void work_stop(void)
1626 {
1627         if (recv_workqueue) {
1628                 destroy_workqueue(recv_workqueue);
1629                 recv_workqueue = NULL;
1630         }
1631
1632         if (send_workqueue) {
1633                 destroy_workqueue(send_workqueue);
1634                 send_workqueue = NULL;
1635         }
1636 }
1637
1638 static int work_start(void)
1639 {
1640         recv_workqueue = alloc_ordered_workqueue("dlm_recv", WQ_MEM_RECLAIM);
1641         if (!recv_workqueue) {
1642                 log_print("can't start dlm_recv");
1643                 return -ENOMEM;
1644         }
1645
1646         send_workqueue = alloc_ordered_workqueue("dlm_send", WQ_MEM_RECLAIM);
1647         if (!send_workqueue) {
1648                 log_print("can't start dlm_send");
1649                 destroy_workqueue(recv_workqueue);
1650                 recv_workqueue = NULL;
1651                 return -ENOMEM;
1652         }
1653
1654         return 0;
1655 }
1656
1657 static void shutdown_conn(struct connection *con)
1658 {
1659         if (dlm_proto_ops->shutdown_action)
1660                 dlm_proto_ops->shutdown_action(con);
1661 }
1662
1663 void dlm_lowcomms_shutdown(void)
1664 {
1665         int idx;
1666
1667         /* Set all the flags to prevent any
1668          * socket activity.
1669          */
1670         dlm_allow_conn = 0;
1671
1672         if (recv_workqueue)
1673                 flush_workqueue(recv_workqueue);
1674         if (send_workqueue)
1675                 flush_workqueue(send_workqueue);
1676
1677         dlm_close_sock(&listen_con.sock);
1678
1679         idx = srcu_read_lock(&connections_srcu);
1680         foreach_conn(shutdown_conn);
1681         srcu_read_unlock(&connections_srcu, idx);
1682 }
1683
1684 static void _stop_conn(struct connection *con, bool and_other)
1685 {
1686         mutex_lock(&con->sock_mutex);
1687         set_bit(CF_CLOSE, &con->flags);
1688         set_bit(CF_READ_PENDING, &con->flags);
1689         set_bit(CF_WRITE_PENDING, &con->flags);
1690         if (con->sock && con->sock->sk) {
1691                 write_lock_bh(&con->sock->sk->sk_callback_lock);
1692                 con->sock->sk->sk_user_data = NULL;
1693                 write_unlock_bh(&con->sock->sk->sk_callback_lock);
1694         }
1695         if (con->othercon && and_other)
1696                 _stop_conn(con->othercon, false);
1697         mutex_unlock(&con->sock_mutex);
1698 }
1699
1700 static void stop_conn(struct connection *con)
1701 {
1702         _stop_conn(con, true);
1703 }
1704
1705 static void connection_release(struct rcu_head *rcu)
1706 {
1707         struct connection *con = container_of(rcu, struct connection, rcu);
1708
1709         kfree(con->rx_buf);
1710         kfree(con);
1711 }
1712
1713 static void free_conn(struct connection *con)
1714 {
1715         close_connection(con, true, true, true);
1716         spin_lock(&connections_lock);
1717         hlist_del_rcu(&con->list);
1718         spin_unlock(&connections_lock);
1719         if (con->othercon) {
1720                 clean_one_writequeue(con->othercon);
1721                 call_srcu(&connections_srcu, &con->othercon->rcu,
1722                           connection_release);
1723         }
1724         clean_one_writequeue(con);
1725         call_srcu(&connections_srcu, &con->rcu, connection_release);
1726 }
1727
1728 static void work_flush(void)
1729 {
1730         int ok;
1731         int i;
1732         struct connection *con;
1733
1734         do {
1735                 ok = 1;
1736                 foreach_conn(stop_conn);
1737                 if (recv_workqueue)
1738                         flush_workqueue(recv_workqueue);
1739                 if (send_workqueue)
1740                         flush_workqueue(send_workqueue);
1741                 for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1742                         hlist_for_each_entry_rcu(con, &connection_hash[i],
1743                                                  list) {
1744                                 ok &= test_bit(CF_READ_PENDING, &con->flags);
1745                                 ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1746                                 if (con->othercon) {
1747                                         ok &= test_bit(CF_READ_PENDING,
1748                                                        &con->othercon->flags);
1749                                         ok &= test_bit(CF_WRITE_PENDING,
1750                                                        &con->othercon->flags);
1751                                 }
1752                         }
1753                 }
1754         } while (!ok);
1755 }
1756
1757 void dlm_lowcomms_stop(void)
1758 {
1759         int idx;
1760
1761         idx = srcu_read_lock(&connections_srcu);
1762         work_flush();
1763         foreach_conn(free_conn);
1764         srcu_read_unlock(&connections_srcu, idx);
1765         work_stop();
1766         deinit_local();
1767
1768         dlm_proto_ops = NULL;
1769 }
1770
1771 static int dlm_listen_for_all(void)
1772 {
1773         struct socket *sock;
1774         int result;
1775
1776         log_print("Using %s for communications",
1777                   dlm_proto_ops->name);
1778
1779         result = dlm_proto_ops->listen_validate();
1780         if (result < 0)
1781                 return result;
1782
1783         result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1784                                   SOCK_STREAM, dlm_proto_ops->proto, &sock);
1785         if (result < 0) {
1786                 log_print("Can't create comms socket, check SCTP is loaded");
1787                 return result;
1788         }
1789
1790         sock_set_mark(sock->sk, dlm_config.ci_mark);
1791         dlm_proto_ops->listen_sockopts(sock);
1792
1793         result = dlm_proto_ops->listen_bind(sock);
1794         if (result < 0)
1795                 goto out;
1796
1797         save_listen_callbacks(sock);
1798         add_listen_sock(sock, &listen_con);
1799
1800         INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1801         result = sock->ops->listen(sock, 5);
1802         if (result < 0) {
1803                 dlm_close_sock(&listen_con.sock);
1804                 return result;
1805         }
1806
1807         return 0;
1808
1809 out:
1810         sock_release(sock);
1811         return result;
1812 }
1813
1814 static int dlm_tcp_bind(struct socket *sock)
1815 {
1816         struct sockaddr_storage src_addr;
1817         int result, addr_len;
1818
1819         /* Bind to our cluster-known address connecting to avoid
1820          * routing problems.
1821          */
1822         memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1823         make_sockaddr(&src_addr, 0, &addr_len);
1824
1825         result = sock->ops->bind(sock, (struct sockaddr *)&src_addr,
1826                                  addr_len);
1827         if (result < 0) {
1828                 /* This *may* not indicate a critical error */
1829                 log_print("could not bind for connect: %d", result);
1830         }
1831
1832         return 0;
1833 }
1834
1835 static int dlm_tcp_connect(struct connection *con, struct socket *sock,
1836                            struct sockaddr *addr, int addr_len)
1837 {
1838         int ret;
1839
1840         ret = sock->ops->connect(sock, addr, addr_len, O_NONBLOCK);
1841         switch (ret) {
1842         case -EINPROGRESS:
1843                 fallthrough;
1844         case 0:
1845                 return 0;
1846         }
1847
1848         return ret;
1849 }
1850
1851 static int dlm_tcp_listen_validate(void)
1852 {
1853         /* We don't support multi-homed hosts */
1854         if (dlm_local_count > 1) {
1855                 log_print("TCP protocol can't handle multi-homed hosts, try SCTP");
1856                 return -EINVAL;
1857         }
1858
1859         return 0;
1860 }
1861
1862 static void dlm_tcp_sockopts(struct socket *sock)
1863 {
1864         /* Turn off Nagle's algorithm */
1865         tcp_sock_set_nodelay(sock->sk);
1866 }
1867
1868 static void dlm_tcp_listen_sockopts(struct socket *sock)
1869 {
1870         dlm_tcp_sockopts(sock);
1871         sock_set_reuseaddr(sock->sk);
1872 }
1873
1874 static int dlm_tcp_listen_bind(struct socket *sock)
1875 {
1876         int addr_len;
1877
1878         /* Bind to our port */
1879         make_sockaddr(dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len);
1880         return sock->ops->bind(sock, (struct sockaddr *)dlm_local_addr[0],
1881                                addr_len);
1882 }
1883
1884 static const struct dlm_proto_ops dlm_tcp_ops = {
1885         .name = "TCP",
1886         .proto = IPPROTO_TCP,
1887         .connect = dlm_tcp_connect,
1888         .sockopts = dlm_tcp_sockopts,
1889         .bind = dlm_tcp_bind,
1890         .listen_validate = dlm_tcp_listen_validate,
1891         .listen_sockopts = dlm_tcp_listen_sockopts,
1892         .listen_bind = dlm_tcp_listen_bind,
1893         .shutdown_action = dlm_tcp_shutdown,
1894         .eof_condition = tcp_eof_condition,
1895 };
1896
1897 static int dlm_sctp_bind(struct socket *sock)
1898 {
1899         return sctp_bind_addrs(sock, 0);
1900 }
1901
1902 static int dlm_sctp_connect(struct connection *con, struct socket *sock,
1903                             struct sockaddr *addr, int addr_len)
1904 {
1905         int ret;
1906
1907         /*
1908          * Make sock->ops->connect() function return in specified time,
1909          * since O_NONBLOCK argument in connect() function does not work here,
1910          * then, we should restore the default value of this attribute.
1911          */
1912         sock_set_sndtimeo(sock->sk, 5);
1913         ret = sock->ops->connect(sock, addr, addr_len, 0);
1914         sock_set_sndtimeo(sock->sk, 0);
1915         if (ret < 0)
1916                 return ret;
1917
1918         if (!test_and_set_bit(CF_CONNECTED, &con->flags))
1919                 log_print("successful connected to node %d", con->nodeid);
1920
1921         return 0;
1922 }
1923
1924 static int dlm_sctp_listen_validate(void)
1925 {
1926         if (!IS_ENABLED(CONFIG_IP_SCTP)) {
1927                 log_print("SCTP is not enabled by this kernel");
1928                 return -EOPNOTSUPP;
1929         }
1930
1931         request_module("sctp");
1932         return 0;
1933 }
1934
1935 static int dlm_sctp_bind_listen(struct socket *sock)
1936 {
1937         return sctp_bind_addrs(sock, dlm_config.ci_tcp_port);
1938 }
1939
1940 static void dlm_sctp_sockopts(struct socket *sock)
1941 {
1942         /* Turn off Nagle's algorithm */
1943         sctp_sock_set_nodelay(sock->sk);
1944         sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1945 }
1946
1947 static const struct dlm_proto_ops dlm_sctp_ops = {
1948         .name = "SCTP",
1949         .proto = IPPROTO_SCTP,
1950         .try_new_addr = true,
1951         .connect = dlm_sctp_connect,
1952         .sockopts = dlm_sctp_sockopts,
1953         .bind = dlm_sctp_bind,
1954         .listen_validate = dlm_sctp_listen_validate,
1955         .listen_sockopts = dlm_sctp_sockopts,
1956         .listen_bind = dlm_sctp_bind_listen,
1957 };
1958
1959 int dlm_lowcomms_start(void)
1960 {
1961         int error = -EINVAL;
1962         int i;
1963
1964         for (i = 0; i < CONN_HASH_SIZE; i++)
1965                 INIT_HLIST_HEAD(&connection_hash[i]);
1966
1967         init_local();
1968         if (!dlm_local_count) {
1969                 error = -ENOTCONN;
1970                 log_print("no local IP address has been set");
1971                 goto fail;
1972         }
1973
1974         INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1975
1976         error = work_start();
1977         if (error)
1978                 goto fail_local;
1979
1980         dlm_allow_conn = 1;
1981
1982         /* Start listening */
1983         switch (dlm_config.ci_protocol) {
1984         case DLM_PROTO_TCP:
1985                 dlm_proto_ops = &dlm_tcp_ops;
1986                 break;
1987         case DLM_PROTO_SCTP:
1988                 dlm_proto_ops = &dlm_sctp_ops;
1989                 break;
1990         default:
1991                 log_print("Invalid protocol identifier %d set",
1992                           dlm_config.ci_protocol);
1993                 error = -EINVAL;
1994                 goto fail_proto_ops;
1995         }
1996
1997         error = dlm_listen_for_all();
1998         if (error)
1999                 goto fail_listen;
2000
2001         return 0;
2002
2003 fail_listen:
2004         dlm_proto_ops = NULL;
2005 fail_proto_ops:
2006         dlm_allow_conn = 0;
2007         work_stop();
2008 fail_local:
2009         deinit_local();
2010 fail:
2011         return error;
2012 }
2013
2014 void dlm_lowcomms_exit(void)
2015 {
2016         struct dlm_node_addr *na, *safe;
2017
2018         spin_lock(&dlm_node_addrs_spin);
2019         list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
2020                 list_del(&na->list);
2021                 while (na->addr_count--)
2022                         kfree(na->addr[na->addr_count]);
2023                 kfree(na);
2024         }
2025         spin_unlock(&dlm_node_addrs_spin);
2026 }