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 struct sockaddr_storage saddr; 596 void (*orig_report)(struct sock *) = NULL; 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 if (kernel_getpeername(sk->sk_socket, (struct sockaddr *)&saddr) < 0) { 605 printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 606 "sending to node %d, port %d, " 607 "sk_err=%d/%d\n", dlm_our_nodeid(), 608 con->nodeid, dlm_config.ci_tcp_port, 609 sk->sk_err, sk->sk_err_soft); 610 } else if (saddr.ss_family == AF_INET) { 611 struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr; 612 613 printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 614 "sending to node %d at %pI4, port %d, " 615 "sk_err=%d/%d\n", dlm_our_nodeid(), 616 con->nodeid, &sin4->sin_addr.s_addr, 617 dlm_config.ci_tcp_port, sk->sk_err, 618 sk->sk_err_soft); 619 } else { 620 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr; 621 622 printk_ratelimited(KERN_ERR "dlm: node %d: socket error " 623 "sending to node %d at %u.%u.%u.%u, " 624 "port %d, sk_err=%d/%d\n", dlm_our_nodeid(), 625 con->nodeid, sin6->sin6_addr.s6_addr32[0], 626 sin6->sin6_addr.s6_addr32[1], 627 sin6->sin6_addr.s6_addr32[2], 628 sin6->sin6_addr.s6_addr32[3], 629 dlm_config.ci_tcp_port, sk->sk_err, 630 sk->sk_err_soft); 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 /* we assume if successful commit must called */ 1323 msg->idx = idx; 1324 return msg; 1325 } 1326 1327 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg) 1328 { 1329 struct writequeue_entry *e = msg->entry; 1330 struct connection *con = e->con; 1331 int users; 1332 1333 spin_lock(&con->writequeue_lock); 1334 kref_get(&msg->ref); 1335 list_add(&msg->list, &e->msgs); 1336 1337 users = --e->users; 1338 if (users) 1339 goto out; 1340 1341 e->len = DLM_WQ_LENGTH_BYTES(e); 1342 spin_unlock(&con->writequeue_lock); 1343 1344 queue_work(send_workqueue, &con->swork); 1345 return; 1346 1347 out: 1348 spin_unlock(&con->writequeue_lock); 1349 return; 1350 } 1351 1352 void dlm_lowcomms_commit_msg(struct dlm_msg *msg) 1353 { 1354 _dlm_lowcomms_commit_msg(msg); 1355 srcu_read_unlock(&connections_srcu, msg->idx); 1356 } 1357 1358 void dlm_lowcomms_put_msg(struct dlm_msg *msg) 1359 { 1360 kref_put(&msg->ref, dlm_msg_release); 1361 } 1362 1363 /* does not held connections_srcu, usage workqueue only */ 1364 int dlm_lowcomms_resend_msg(struct dlm_msg *msg) 1365 { 1366 struct dlm_msg *msg_resend; 1367 char *ppc; 1368 1369 if (msg->retransmit) 1370 return 1; 1371 1372 msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len, 1373 GFP_ATOMIC, &ppc, NULL, NULL); 1374 if (!msg_resend) 1375 return -ENOMEM; 1376 1377 msg->retransmit = true; 1378 kref_get(&msg->ref); 1379 msg_resend->orig_msg = msg; 1380 1381 memcpy(ppc, msg->ppc, msg->len); 1382 _dlm_lowcomms_commit_msg(msg_resend); 1383 dlm_lowcomms_put_msg(msg_resend); 1384 1385 return 0; 1386 } 1387 1388 /* Send a message */ 1389 static void send_to_sock(struct connection *con) 1390 { 1391 const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; 1392 struct writequeue_entry *e; 1393 int len, offset, ret; 1394 int count = 0; 1395 1396 mutex_lock(&con->sock_mutex); 1397 if (con->sock == NULL) 1398 goto out_connect; 1399 1400 spin_lock(&con->writequeue_lock); 1401 for (;;) { 1402 e = con_next_wq(con); 1403 if (!e) 1404 break; 1405 1406 e = list_first_entry(&con->writequeue, struct writequeue_entry, list); 1407 len = e->len; 1408 offset = e->offset; 1409 BUG_ON(len == 0 && e->users == 0); 1410 spin_unlock(&con->writequeue_lock); 1411 1412 ret = kernel_sendpage(con->sock, e->page, offset, len, 1413 msg_flags); 1414 if (ret == -EAGAIN || ret == 0) { 1415 if (ret == -EAGAIN && 1416 test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) && 1417 !test_and_set_bit(CF_APP_LIMITED, &con->flags)) { 1418 /* Notify TCP that we're limited by the 1419 * application window size. 1420 */ 1421 set_bit(SOCK_NOSPACE, &con->sock->flags); 1422 con->sock->sk->sk_write_pending++; 1423 } 1424 cond_resched(); 1425 goto out; 1426 } else if (ret < 0) 1427 goto out; 1428 1429 /* Don't starve people filling buffers */ 1430 if (++count >= MAX_SEND_MSG_COUNT) { 1431 cond_resched(); 1432 count = 0; 1433 } 1434 1435 spin_lock(&con->writequeue_lock); 1436 writequeue_entry_complete(e, ret); 1437 } 1438 spin_unlock(&con->writequeue_lock); 1439 1440 /* close if we got EOF */ 1441 if (test_and_clear_bit(CF_EOF, &con->flags)) { 1442 mutex_unlock(&con->sock_mutex); 1443 close_connection(con, false, false, true); 1444 1445 /* handling for tcp shutdown */ 1446 clear_bit(CF_SHUTDOWN, &con->flags); 1447 wake_up(&con->shutdown_wait); 1448 } else { 1449 mutex_unlock(&con->sock_mutex); 1450 } 1451 1452 return; 1453 1454 out: 1455 mutex_unlock(&con->sock_mutex); 1456 return; 1457 1458 out_connect: 1459 mutex_unlock(&con->sock_mutex); 1460 queue_work(send_workqueue, &con->swork); 1461 cond_resched(); 1462 } 1463 1464 static void clean_one_writequeue(struct connection *con) 1465 { 1466 struct writequeue_entry *e, *safe; 1467 1468 spin_lock(&con->writequeue_lock); 1469 list_for_each_entry_safe(e, safe, &con->writequeue, list) { 1470 free_entry(e); 1471 } 1472 spin_unlock(&con->writequeue_lock); 1473 } 1474 1475 /* Called from recovery when it knows that a node has 1476 left the cluster */ 1477 int dlm_lowcomms_close(int nodeid) 1478 { 1479 struct connection *con; 1480 struct dlm_node_addr *na; 1481 int idx; 1482 1483 log_print("closing connection to node %d", nodeid); 1484 idx = srcu_read_lock(&connections_srcu); 1485 con = nodeid2con(nodeid, 0); 1486 if (con) { 1487 set_bit(CF_CLOSE, &con->flags); 1488 close_connection(con, true, true, true); 1489 clean_one_writequeue(con); 1490 if (con->othercon) 1491 clean_one_writequeue(con->othercon); 1492 } 1493 srcu_read_unlock(&connections_srcu, idx); 1494 1495 spin_lock(&dlm_node_addrs_spin); 1496 na = find_node_addr(nodeid); 1497 if (na) { 1498 list_del(&na->list); 1499 while (na->addr_count--) 1500 kfree(na->addr[na->addr_count]); 1501 kfree(na); 1502 } 1503 spin_unlock(&dlm_node_addrs_spin); 1504 1505 return 0; 1506 } 1507 1508 /* Receive workqueue function */ 1509 static void process_recv_sockets(struct work_struct *work) 1510 { 1511 struct connection *con = container_of(work, struct connection, rwork); 1512 1513 clear_bit(CF_READ_PENDING, &con->flags); 1514 receive_from_sock(con); 1515 } 1516 1517 static void process_listen_recv_socket(struct work_struct *work) 1518 { 1519 accept_from_sock(&listen_con); 1520 } 1521 1522 static void dlm_connect(struct connection *con) 1523 { 1524 struct sockaddr_storage addr; 1525 int result, addr_len; 1526 struct socket *sock; 1527 unsigned int mark; 1528 1529 /* Some odd races can cause double-connects, ignore them */ 1530 if (con->retries++ > MAX_CONNECT_RETRIES) 1531 return; 1532 1533 if (con->sock) { 1534 log_print("node %d already connected.", con->nodeid); 1535 return; 1536 } 1537 1538 memset(&addr, 0, sizeof(addr)); 1539 result = nodeid_to_addr(con->nodeid, &addr, NULL, 1540 dlm_proto_ops->try_new_addr, &mark); 1541 if (result < 0) { 1542 log_print("no address for nodeid %d", con->nodeid); 1543 return; 1544 } 1545 1546 /* Create a socket to communicate with */ 1547 result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1548 SOCK_STREAM, dlm_proto_ops->proto, &sock); 1549 if (result < 0) 1550 goto socket_err; 1551 1552 sock_set_mark(sock->sk, mark); 1553 dlm_proto_ops->sockopts(sock); 1554 1555 add_sock(sock, con); 1556 1557 result = dlm_proto_ops->bind(sock); 1558 if (result < 0) 1559 goto add_sock_err; 1560 1561 log_print_ratelimited("connecting to %d", con->nodeid); 1562 make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len); 1563 result = dlm_proto_ops->connect(con, sock, (struct sockaddr *)&addr, 1564 addr_len); 1565 if (result < 0) 1566 goto add_sock_err; 1567 1568 return; 1569 1570 add_sock_err: 1571 dlm_close_sock(&con->sock); 1572 1573 socket_err: 1574 /* 1575 * Some errors are fatal and this list might need adjusting. For other 1576 * errors we try again until the max number of retries is reached. 1577 */ 1578 if (result != -EHOSTUNREACH && 1579 result != -ENETUNREACH && 1580 result != -ENETDOWN && 1581 result != -EINVAL && 1582 result != -EPROTONOSUPPORT) { 1583 log_print("connect %d try %d error %d", con->nodeid, 1584 con->retries, result); 1585 msleep(1000); 1586 lowcomms_connect_sock(con); 1587 } 1588 } 1589 1590 /* Send workqueue function */ 1591 static void process_send_sockets(struct work_struct *work) 1592 { 1593 struct connection *con = container_of(work, struct connection, swork); 1594 1595 WARN_ON(test_bit(CF_IS_OTHERCON, &con->flags)); 1596 1597 clear_bit(CF_WRITE_PENDING, &con->flags); 1598 1599 if (test_and_clear_bit(CF_RECONNECT, &con->flags)) { 1600 close_connection(con, false, false, true); 1601 dlm_midcomms_unack_msg_resend(con->nodeid); 1602 } 1603 1604 if (con->sock == NULL) { 1605 if (test_and_clear_bit(CF_DELAY_CONNECT, &con->flags)) 1606 msleep(1000); 1607 1608 mutex_lock(&con->sock_mutex); 1609 dlm_connect(con); 1610 mutex_unlock(&con->sock_mutex); 1611 } 1612 1613 if (!list_empty(&con->writequeue)) 1614 send_to_sock(con); 1615 } 1616 1617 static void work_stop(void) 1618 { 1619 if (recv_workqueue) { 1620 destroy_workqueue(recv_workqueue); 1621 recv_workqueue = NULL; 1622 } 1623 1624 if (send_workqueue) { 1625 destroy_workqueue(send_workqueue); 1626 send_workqueue = NULL; 1627 } 1628 } 1629 1630 static int work_start(void) 1631 { 1632 recv_workqueue = alloc_ordered_workqueue("dlm_recv", WQ_MEM_RECLAIM); 1633 if (!recv_workqueue) { 1634 log_print("can't start dlm_recv"); 1635 return -ENOMEM; 1636 } 1637 1638 send_workqueue = alloc_ordered_workqueue("dlm_send", WQ_MEM_RECLAIM); 1639 if (!send_workqueue) { 1640 log_print("can't start dlm_send"); 1641 destroy_workqueue(recv_workqueue); 1642 recv_workqueue = NULL; 1643 return -ENOMEM; 1644 } 1645 1646 return 0; 1647 } 1648 1649 static void shutdown_conn(struct connection *con) 1650 { 1651 if (dlm_proto_ops->shutdown_action) 1652 dlm_proto_ops->shutdown_action(con); 1653 } 1654 1655 void dlm_lowcomms_shutdown(void) 1656 { 1657 int idx; 1658 1659 /* Set all the flags to prevent any 1660 * socket activity. 1661 */ 1662 dlm_allow_conn = 0; 1663 1664 if (recv_workqueue) 1665 flush_workqueue(recv_workqueue); 1666 if (send_workqueue) 1667 flush_workqueue(send_workqueue); 1668 1669 dlm_close_sock(&listen_con.sock); 1670 1671 idx = srcu_read_lock(&connections_srcu); 1672 foreach_conn(shutdown_conn); 1673 srcu_read_unlock(&connections_srcu, idx); 1674 } 1675 1676 static void _stop_conn(struct connection *con, bool and_other) 1677 { 1678 mutex_lock(&con->sock_mutex); 1679 set_bit(CF_CLOSE, &con->flags); 1680 set_bit(CF_READ_PENDING, &con->flags); 1681 set_bit(CF_WRITE_PENDING, &con->flags); 1682 if (con->sock && con->sock->sk) { 1683 write_lock_bh(&con->sock->sk->sk_callback_lock); 1684 con->sock->sk->sk_user_data = NULL; 1685 write_unlock_bh(&con->sock->sk->sk_callback_lock); 1686 } 1687 if (con->othercon && and_other) 1688 _stop_conn(con->othercon, false); 1689 mutex_unlock(&con->sock_mutex); 1690 } 1691 1692 static void stop_conn(struct connection *con) 1693 { 1694 _stop_conn(con, true); 1695 } 1696 1697 static void connection_release(struct rcu_head *rcu) 1698 { 1699 struct connection *con = container_of(rcu, struct connection, rcu); 1700 1701 kfree(con->rx_buf); 1702 kfree(con); 1703 } 1704 1705 static void free_conn(struct connection *con) 1706 { 1707 close_connection(con, true, true, true); 1708 spin_lock(&connections_lock); 1709 hlist_del_rcu(&con->list); 1710 spin_unlock(&connections_lock); 1711 if (con->othercon) { 1712 clean_one_writequeue(con->othercon); 1713 call_srcu(&connections_srcu, &con->othercon->rcu, 1714 connection_release); 1715 } 1716 clean_one_writequeue(con); 1717 call_srcu(&connections_srcu, &con->rcu, connection_release); 1718 } 1719 1720 static void work_flush(void) 1721 { 1722 int ok; 1723 int i; 1724 struct connection *con; 1725 1726 do { 1727 ok = 1; 1728 foreach_conn(stop_conn); 1729 if (recv_workqueue) 1730 flush_workqueue(recv_workqueue); 1731 if (send_workqueue) 1732 flush_workqueue(send_workqueue); 1733 for (i = 0; i < CONN_HASH_SIZE && ok; i++) { 1734 hlist_for_each_entry_rcu(con, &connection_hash[i], 1735 list) { 1736 ok &= test_bit(CF_READ_PENDING, &con->flags); 1737 ok &= test_bit(CF_WRITE_PENDING, &con->flags); 1738 if (con->othercon) { 1739 ok &= test_bit(CF_READ_PENDING, 1740 &con->othercon->flags); 1741 ok &= test_bit(CF_WRITE_PENDING, 1742 &con->othercon->flags); 1743 } 1744 } 1745 } 1746 } while (!ok); 1747 } 1748 1749 void dlm_lowcomms_stop(void) 1750 { 1751 int idx; 1752 1753 idx = srcu_read_lock(&connections_srcu); 1754 work_flush(); 1755 foreach_conn(free_conn); 1756 srcu_read_unlock(&connections_srcu, idx); 1757 work_stop(); 1758 deinit_local(); 1759 1760 dlm_proto_ops = NULL; 1761 } 1762 1763 static int dlm_listen_for_all(void) 1764 { 1765 struct socket *sock; 1766 int result; 1767 1768 log_print("Using %s for communications", 1769 dlm_proto_ops->name); 1770 1771 result = dlm_proto_ops->listen_validate(); 1772 if (result < 0) 1773 return result; 1774 1775 result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, 1776 SOCK_STREAM, dlm_proto_ops->proto, &sock); 1777 if (result < 0) { 1778 log_print("Can't create comms socket, check SCTP is loaded"); 1779 goto out; 1780 } 1781 1782 sock_set_mark(sock->sk, dlm_config.ci_mark); 1783 dlm_proto_ops->listen_sockopts(sock); 1784 1785 result = dlm_proto_ops->listen_bind(sock); 1786 if (result < 0) 1787 goto out; 1788 1789 save_listen_callbacks(sock); 1790 add_listen_sock(sock, &listen_con); 1791 1792 INIT_WORK(&listen_con.rwork, process_listen_recv_socket); 1793 result = sock->ops->listen(sock, 5); 1794 if (result < 0) { 1795 dlm_close_sock(&listen_con.sock); 1796 goto out; 1797 } 1798 1799 return 0; 1800 1801 out: 1802 sock_release(sock); 1803 return result; 1804 } 1805 1806 static int dlm_tcp_bind(struct socket *sock) 1807 { 1808 struct sockaddr_storage src_addr; 1809 int result, addr_len; 1810 1811 /* Bind to our cluster-known address connecting to avoid 1812 * routing problems. 1813 */ 1814 memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); 1815 make_sockaddr(&src_addr, 0, &addr_len); 1816 1817 result = sock->ops->bind(sock, (struct sockaddr *)&src_addr, 1818 addr_len); 1819 if (result < 0) { 1820 /* This *may* not indicate a critical error */ 1821 log_print("could not bind for connect: %d", result); 1822 } 1823 1824 return 0; 1825 } 1826 1827 static int dlm_tcp_connect(struct connection *con, struct socket *sock, 1828 struct sockaddr *addr, int addr_len) 1829 { 1830 int ret; 1831 1832 ret = sock->ops->connect(sock, addr, addr_len, O_NONBLOCK); 1833 switch (ret) { 1834 case -EINPROGRESS: 1835 fallthrough; 1836 case 0: 1837 return 0; 1838 } 1839 1840 return ret; 1841 } 1842 1843 static int dlm_tcp_listen_validate(void) 1844 { 1845 /* We don't support multi-homed hosts */ 1846 if (dlm_local_count > 1) { 1847 log_print("TCP protocol can't handle multi-homed hosts, try SCTP"); 1848 return -EINVAL; 1849 } 1850 1851 return 0; 1852 } 1853 1854 static void dlm_tcp_sockopts(struct socket *sock) 1855 { 1856 /* Turn off Nagle's algorithm */ 1857 tcp_sock_set_nodelay(sock->sk); 1858 } 1859 1860 static void dlm_tcp_listen_sockopts(struct socket *sock) 1861 { 1862 dlm_tcp_sockopts(sock); 1863 sock_set_reuseaddr(sock->sk); 1864 } 1865 1866 static int dlm_tcp_listen_bind(struct socket *sock) 1867 { 1868 int addr_len; 1869 1870 /* Bind to our port */ 1871 make_sockaddr(dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len); 1872 return sock->ops->bind(sock, (struct sockaddr *)dlm_local_addr[0], 1873 addr_len); 1874 } 1875 1876 static const struct dlm_proto_ops dlm_tcp_ops = { 1877 .name = "TCP", 1878 .proto = IPPROTO_TCP, 1879 .connect = dlm_tcp_connect, 1880 .sockopts = dlm_tcp_sockopts, 1881 .bind = dlm_tcp_bind, 1882 .listen_validate = dlm_tcp_listen_validate, 1883 .listen_sockopts = dlm_tcp_listen_sockopts, 1884 .listen_bind = dlm_tcp_listen_bind, 1885 .shutdown_action = dlm_tcp_shutdown, 1886 .eof_condition = tcp_eof_condition, 1887 }; 1888 1889 static int dlm_sctp_bind(struct socket *sock) 1890 { 1891 return sctp_bind_addrs(sock, 0); 1892 } 1893 1894 static int dlm_sctp_connect(struct connection *con, struct socket *sock, 1895 struct sockaddr *addr, int addr_len) 1896 { 1897 int ret; 1898 1899 /* 1900 * Make sock->ops->connect() function return in specified time, 1901 * since O_NONBLOCK argument in connect() function does not work here, 1902 * then, we should restore the default value of this attribute. 1903 */ 1904 sock_set_sndtimeo(sock->sk, 5); 1905 ret = sock->ops->connect(sock, addr, addr_len, 0); 1906 sock_set_sndtimeo(sock->sk, 0); 1907 if (ret < 0) 1908 return ret; 1909 1910 if (!test_and_set_bit(CF_CONNECTED, &con->flags)) 1911 log_print("successful connected to node %d", con->nodeid); 1912 1913 return 0; 1914 } 1915 1916 static int dlm_sctp_listen_validate(void) 1917 { 1918 if (!IS_ENABLED(CONFIG_IP_SCTP)) { 1919 log_print("SCTP is not enabled by this kernel"); 1920 return -EOPNOTSUPP; 1921 } 1922 1923 request_module("sctp"); 1924 return 0; 1925 } 1926 1927 static int dlm_sctp_bind_listen(struct socket *sock) 1928 { 1929 return sctp_bind_addrs(sock, dlm_config.ci_tcp_port); 1930 } 1931 1932 static void dlm_sctp_sockopts(struct socket *sock) 1933 { 1934 /* Turn off Nagle's algorithm */ 1935 sctp_sock_set_nodelay(sock->sk); 1936 sock_set_rcvbuf(sock->sk, NEEDED_RMEM); 1937 } 1938 1939 static const struct dlm_proto_ops dlm_sctp_ops = { 1940 .name = "SCTP", 1941 .proto = IPPROTO_SCTP, 1942 .try_new_addr = true, 1943 .connect = dlm_sctp_connect, 1944 .sockopts = dlm_sctp_sockopts, 1945 .bind = dlm_sctp_bind, 1946 .listen_validate = dlm_sctp_listen_validate, 1947 .listen_sockopts = dlm_sctp_sockopts, 1948 .listen_bind = dlm_sctp_bind_listen, 1949 }; 1950 1951 int dlm_lowcomms_start(void) 1952 { 1953 int error = -EINVAL; 1954 int i; 1955 1956 for (i = 0; i < CONN_HASH_SIZE; i++) 1957 INIT_HLIST_HEAD(&connection_hash[i]); 1958 1959 init_local(); 1960 if (!dlm_local_count) { 1961 error = -ENOTCONN; 1962 log_print("no local IP address has been set"); 1963 goto fail; 1964 } 1965 1966 INIT_WORK(&listen_con.rwork, process_listen_recv_socket); 1967 1968 error = work_start(); 1969 if (error) 1970 goto fail_local; 1971 1972 dlm_allow_conn = 1; 1973 1974 /* Start listening */ 1975 switch (dlm_config.ci_protocol) { 1976 case DLM_PROTO_TCP: 1977 dlm_proto_ops = &dlm_tcp_ops; 1978 break; 1979 case DLM_PROTO_SCTP: 1980 dlm_proto_ops = &dlm_sctp_ops; 1981 break; 1982 default: 1983 log_print("Invalid protocol identifier %d set", 1984 dlm_config.ci_protocol); 1985 error = -EINVAL; 1986 goto fail_proto_ops; 1987 } 1988 1989 error = dlm_listen_for_all(); 1990 if (error) 1991 goto fail_listen; 1992 1993 return 0; 1994 1995 fail_listen: 1996 dlm_proto_ops = NULL; 1997 fail_proto_ops: 1998 dlm_allow_conn = 0; 1999 dlm_close_sock(&listen_con.sock); 2000 work_stop(); 2001 fail_local: 2002 deinit_local(); 2003 fail: 2004 return error; 2005 } 2006 2007 void dlm_lowcomms_exit(void) 2008 { 2009 struct dlm_node_addr *na, *safe; 2010 2011 spin_lock(&dlm_node_addrs_spin); 2012 list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) { 2013 list_del(&na->list); 2014 while (na->addr_count--) 2015 kfree(na->addr[na->addr_count]); 2016 kfree(na); 2017 } 2018 spin_unlock(&dlm_node_addrs_spin); 2019 } 2020