net/tipc/bcast.c

   1 /*
   2  * net/tipc/bcast.c: TIPC broadcast code
   3  *
   4  * Copyright (c) 2004-2006, 2014, Ericsson AB
   5  * Copyright (c) 2004, Intel Corporation.
   6  * Copyright (c) 2005, 2010-2011, Wind River Systems
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions are met:
  11  *
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. Neither the names of the copyright holders nor the names of its
  18  *    contributors may be used to endorse or promote products derived from
  19  *    this software without specific prior written permission.
  20  *
  21  * Alternatively, this software may be distributed under the terms of the
  22  * GNU General Public License ("GPL") version 2 as published by the Free
  23  * Software Foundation.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  26  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  29  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  35  * POSSIBILITY OF SUCH DAMAGE.
  36  */
  37
  38 #include "core.h"
  39 #include "link.h"
  40 #include "socket.h"
  41 #include "msg.h"
  42 #include "bcast.h"
  43 #include "name_distr.h"
  44
  45 #define MAX_PKT_DEFAULT_MCAST   1500    /* bcast link max packet size (fixed) */
  46 #define BCLINK_WIN_DEFAULT      20      /* bcast link window size (default) */
  47 #define BCBEARER                MAX_BEARERS
  48
  49 /**
  50  * struct tipc_bcbearer_pair - a pair of bearers used by broadcast link
  51  * @primary: pointer to primary bearer
  52  * @secondary: pointer to secondary bearer
  53  *
  54  * Bearers must have same priority and same set of reachable destinations
  55  * to be paired.
  56  */
  57
  58 struct tipc_bcbearer_pair {
  59         struct tipc_bearer *primary;
  60         struct tipc_bearer *secondary;
  61 };
  62
  63 /**
  64  * struct tipc_bcbearer - bearer used by broadcast link
  65  * @bearer: (non-standard) broadcast bearer structure
  66  * @media: (non-standard) broadcast media structure
  67  * @bpairs: array of bearer pairs
  68  * @bpairs_temp: temporary array of bearer pairs used by tipc_bcbearer_sort()
  69  * @remains: temporary node map used by tipc_bcbearer_send()
  70  * @remains_new: temporary node map used tipc_bcbearer_send()
  71  *
  72  * Note: The fields labelled "temporary" are incorporated into the bearer
  73  * to avoid consuming potentially limited stack space through the use of
  74  * large local variables within multicast routines.  Concurrent access is
  75  * prevented through use of the spinlock "bclink_lock".
  76  */
  77 struct tipc_bcbearer {
  78         struct tipc_bearer bearer;
  79         struct tipc_media media;
  80         struct tipc_bcbearer_pair bpairs[MAX_BEARERS];
  81         struct tipc_bcbearer_pair bpairs_temp[TIPC_MAX_LINK_PRI + 1];
  82         struct tipc_node_map remains;
  83         struct tipc_node_map remains_new;
  84 };
  85
  86 /**
  87  * struct tipc_bclink - link used for broadcast messages
  88  * @lock: spinlock governing access to structure
  89  * @link: (non-standard) broadcast link structure
  90  * @node: (non-standard) node structure representing b'cast link's peer node
  91  * @flags: represent bclink states
  92  * @bcast_nodes: map of broadcast-capable nodes
  93  * @retransmit_to: node that most recently requested a retransmit
  94  *
  95  * Handles sequence numbering, fragmentation, bundling, etc.
  96  */
  97 struct tipc_bclink {
  98         spinlock_t lock;
  99         struct tipc_link link;
 100         struct tipc_node node;
 101         unsigned int flags;
 102         struct tipc_node_map bcast_nodes;
 103         struct tipc_node *retransmit_to;
 104 };
 105
 106 static struct tipc_bcbearer *bcbearer;
 107 static struct tipc_bclink *bclink;
 108 static struct tipc_link *bcl;
 109
 110 const char tipc_bclink_name[] = "broadcast-link";
 111
 112 static void tipc_nmap_diff(struct tipc_node_map *nm_a,
 113                            struct tipc_node_map *nm_b,
 114                            struct tipc_node_map *nm_diff);
 115 static void tipc_nmap_add(struct tipc_node_map *nm_ptr, u32 node);
 116 static void tipc_nmap_remove(struct tipc_node_map *nm_ptr, u32 node);
 117
 118 static void tipc_bclink_lock(void)
 119 {
 120         spin_lock_bh(&bclink->lock);
 121 }
 122
 123 static void tipc_bclink_unlock(void)
 124 {
 125         struct tipc_node *node = NULL;
 126
 127         if (likely(!bclink->flags)) {
 128                 spin_unlock_bh(&bclink->lock);
 129                 return;
 130         }
 131
 132         if (bclink->flags & TIPC_BCLINK_RESET) {
 133                 bclink->flags &= ~TIPC_BCLINK_RESET;
 134                 node = tipc_bclink_retransmit_to();
 135         }
 136         spin_unlock_bh(&bclink->lock);
 137
 138         if (node)
 139                 tipc_link_reset_all(node);
 140 }
 141
 142 uint  tipc_bclink_get_mtu(void)
 143 {
 144         return MAX_PKT_DEFAULT_MCAST;
 145 }
 146
 147 void tipc_bclink_set_flags(unsigned int flags)
 148 {
 149         bclink->flags |= flags;
 150 }
 151
 152 static u32 bcbuf_acks(struct sk_buff *buf)
 153 {
 154         return (u32)(unsigned long)TIPC_SKB_CB(buf)->handle;
 155 }
 156
 157 static void bcbuf_set_acks(struct sk_buff *buf, u32 acks)
 158 {
 159         TIPC_SKB_CB(buf)->handle = (void *)(unsigned long)acks;
 160 }
 161
 162 static void bcbuf_decr_acks(struct sk_buff *buf)
 163 {
 164         bcbuf_set_acks(buf, bcbuf_acks(buf) - 1);
 165 }
 166
 167 void tipc_bclink_add_node(u32 addr)
 168 {
 169         tipc_bclink_lock();
 170         tipc_nmap_add(&bclink->bcast_nodes, addr);
 171         tipc_bclink_unlock();
 172 }
 173
 174 void tipc_bclink_remove_node(u32 addr)
 175 {
 176         tipc_bclink_lock();
 177         tipc_nmap_remove(&bclink->bcast_nodes, addr);
 178         tipc_bclink_unlock();
 179 }
 180
 181 static void bclink_set_last_sent(void)
 182 {
 183         if (bcl->next_out)
 184                 bcl->fsm_msg_cnt = mod(buf_seqno(bcl->next_out) - 1);
 185         else
 186                 bcl->fsm_msg_cnt = mod(bcl->next_out_no - 1);
 187 }
 188
 189 u32 tipc_bclink_get_last_sent(void)
 190 {
 191         return bcl->fsm_msg_cnt;
 192 }
 193
 194 static void bclink_update_last_sent(struct tipc_node *node, u32 seqno)
 195 {
 196         node->bclink.last_sent = less_eq(node->bclink.last_sent, seqno) ?
 197                                                 seqno : node->bclink.last_sent;
 198 }
 199
 200
 201 /**
 202  * tipc_bclink_retransmit_to - get most recent node to request retransmission
 203  *
 204  * Called with bclink_lock locked
 205  */
 206 struct tipc_node *tipc_bclink_retransmit_to(void)
 207 {
 208         return bclink->retransmit_to;
 209 }
 210
 211 /**
 212  * bclink_retransmit_pkt - retransmit broadcast packets
 213  * @after: sequence number of last packet to *not* retransmit
 214  * @to: sequence number of last packet to retransmit
 215  *
 216  * Called with bclink_lock locked
 217  */
 218 static void bclink_retransmit_pkt(u32 after, u32 to)
 219 {
 220         struct sk_buff *skb;
 221
 222         skb_queue_walk(&bcl->outqueue, skb) {
 223                 if (more(buf_seqno(skb), after))
 224                         break;
 225         }
 226         tipc_link_retransmit(bcl, skb, mod(to - after));
 227 }
 228
 229 /**
 230  * tipc_bclink_wakeup_users - wake up pending users
 231  *
 232  * Called with no locks taken
 233  */
 234 void tipc_bclink_wakeup_users(void)
 235 {
 236         while (skb_queue_len(&bclink->link.waiting_sks))
 237                 tipc_sk_rcv(skb_dequeue(&bclink->link.waiting_sks));
 238 }
 239
 240 /**
 241  * tipc_bclink_acknowledge - handle acknowledgement of broadcast packets
 242  * @n_ptr: node that sent acknowledgement info
 243  * @acked: broadcast sequence # that has been acknowledged
 244  *
 245  * Node is locked, bclink_lock unlocked.
 246  */
 247 void tipc_bclink_acknowledge(struct tipc_node *n_ptr, u32 acked)
 248 {
 249         struct sk_buff *skb, *tmp;
 250         struct sk_buff *next;
 251         unsigned int released = 0;
 252
 253         tipc_bclink_lock();
 254         /* Bail out if tx queue is empty (no clean up is required) */
 255         skb = skb_peek(&bcl->outqueue);
 256         if (!skb)
 257                 goto exit;
 258
 259         /* Determine which messages need to be acknowledged */
 260         if (acked == INVALID_LINK_SEQ) {
 261                 /*
 262                  * Contact with specified node has been lost, so need to
 263                  * acknowledge sent messages only (if other nodes still exist)
 264                  * or both sent and unsent messages (otherwise)
 265                  */
 266                 if (bclink->bcast_nodes.count)
 267                         acked = bcl->fsm_msg_cnt;
 268                 else
 269                         acked = bcl->next_out_no;
 270         } else {
 271                 /*
 272                  * Bail out if specified sequence number does not correspond
 273                  * to a message that has been sent and not yet acknowledged
 274                  */
 275                 if (less(acked, buf_seqno(skb)) ||
 276                     less(bcl->fsm_msg_cnt, acked) ||
 277                     less_eq(acked, n_ptr->bclink.acked))
 278                         goto exit;
 279         }
 280
 281         /* Skip over packets that node has previously acknowledged */
 282         skb_queue_walk(&bcl->outqueue, skb) {
 283                 if (more(buf_seqno(skb), n_ptr->bclink.acked))
 284                         break;
 285         }
 286
 287         /* Update packets that node is now acknowledging */
 288         skb_queue_walk_from_safe(&bcl->outqueue, skb, tmp) {
 289                 if (more(buf_seqno(skb), acked))
 290                         break;
 291
 292                 next = tipc_skb_queue_next(&bcl->outqueue, skb);
 293                 if (skb != bcl->next_out) {
 294                         bcbuf_decr_acks(skb);
 295                 } else {
 296                         bcbuf_set_acks(skb, 0);
 297                         bcl->next_out = next;
 298                         bclink_set_last_sent();
 299                 }
 300
 301                 if (bcbuf_acks(skb) == 0) {
 302                         __skb_unlink(skb, &bcl->outqueue);
 303                         kfree_skb(skb);
 304                         released = 1;
 305                 }
 306         }
 307         n_ptr->bclink.acked = acked;
 308
 309         /* Try resolving broadcast link congestion, if necessary */
 310         if (unlikely(bcl->next_out)) {
 311                 tipc_link_push_packets(bcl);
 312                 bclink_set_last_sent();
 313         }
 314         if (unlikely(released && !skb_queue_empty(&bcl->waiting_sks)))
 315                 n_ptr->action_flags |= TIPC_WAKEUP_BCAST_USERS;
 316
 317 exit:
 318         tipc_bclink_unlock();
 319 }
 320
 321 /**
 322  * tipc_bclink_update_link_state - update broadcast link state
 323  *
 324  * RCU and node lock set
 325  */
 326 void tipc_bclink_update_link_state(struct tipc_node *n_ptr, u32 last_sent)
 327 {
 328         struct sk_buff *buf;
 329
 330         /* Ignore "stale" link state info */
 331         if (less_eq(last_sent, n_ptr->bclink.last_in))
 332                 return;
 333
 334         /* Update link synchronization state; quit if in sync */
 335         bclink_update_last_sent(n_ptr, last_sent);
 336
 337         if (n_ptr->bclink.last_sent == n_ptr->bclink.last_in)
 338                 return;
 339
 340         /* Update out-of-sync state; quit if loss is still unconfirmed */
 341         if ((++n_ptr->bclink.oos_state) == 1) {
 342                 if (n_ptr->bclink.deferred_size < (TIPC_MIN_LINK_WIN / 2))
 343                         return;
 344                 n_ptr->bclink.oos_state++;
 345         }
 346
 347         /* Don't NACK if one has been recently sent (or seen) */
 348         if (n_ptr->bclink.oos_state & 0x1)
 349                 return;
 350
 351         /* Send NACK */
 352         buf = tipc_buf_acquire(INT_H_SIZE);
 353         if (buf) {
 354                 struct tipc_msg *msg = buf_msg(buf);
 355                 struct sk_buff *skb = skb_peek(&n_ptr->bclink.deferred_queue);
 356                 u32 to = skb ? buf_seqno(skb) - 1 : n_ptr->bclink.last_sent;
 357
 358                 tipc_msg_init(msg, BCAST_PROTOCOL, STATE_MSG,
 359                               INT_H_SIZE, n_ptr->addr);
 360                 msg_set_non_seq(msg, 1);
 361                 msg_set_mc_netid(msg, tipc_net_id);
 362                 msg_set_bcast_ack(msg, n_ptr->bclink.last_in);
 363                 msg_set_bcgap_after(msg, n_ptr->bclink.last_in);
 364                 msg_set_bcgap_to(msg, to);
 365
 366                 tipc_bclink_lock();
 367                 tipc_bearer_send(MAX_BEARERS, buf, NULL);
 368                 bcl->stats.sent_nacks++;
 369                 tipc_bclink_unlock();
 370                 kfree_skb(buf);
 371
 372                 n_ptr->bclink.oos_state++;
 373         }
 374 }
 375
 376 /**
 377  * bclink_peek_nack - monitor retransmission requests sent by other nodes
 378  *
 379  * Delay any upcoming NACK by this node if another node has already
 380  * requested the first message this node is going to ask for.
 381  */
 382 static void bclink_peek_nack(struct tipc_msg *msg)
 383 {
 384         struct tipc_node *n_ptr = tipc_node_find(msg_destnode(msg));
 385
 386         if (unlikely(!n_ptr))
 387                 return;
 388
 389         tipc_node_lock(n_ptr);
 390
 391         if (n_ptr->bclink.recv_permitted &&
 392             (n_ptr->bclink.last_in != n_ptr->bclink.last_sent) &&
 393             (n_ptr->bclink.last_in == msg_bcgap_after(msg)))
 394                 n_ptr->bclink.oos_state = 2;
 395
 396         tipc_node_unlock(n_ptr);
 397 }
 398
 399 /* tipc_bclink_xmit - broadcast buffer chain to all nodes in cluster
 400  *                    and to identified node local sockets
 401  * @list: chain of buffers containing message
 402  * Consumes the buffer chain, except when returning -ELINKCONG
 403  * Returns 0 if success, otherwise errno: -ELINKCONG,-EHOSTUNREACH,-EMSGSIZE
 404  */
 405 int tipc_bclink_xmit(struct sk_buff_head *list)
 406 {
 407         int rc = 0;
 408         int bc = 0;
 409         struct sk_buff *skb;
 410
 411         /* Prepare clone of message for local node */
 412         skb = tipc_msg_reassemble(list);
 413         if (unlikely(!skb)) {
 414                 __skb_queue_purge(list);
 415                 return -EHOSTUNREACH;
 416         }
 417
 418         /* Broadcast to all other nodes */
 419         if (likely(bclink)) {
 420                 tipc_bclink_lock();
 421                 if (likely(bclink->bcast_nodes.count)) {
 422                         rc = __tipc_link_xmit(bcl, list);
 423                         if (likely(!rc)) {
 424                                 u32 len = skb_queue_len(&bcl->outqueue);
 425
 426                                 bclink_set_last_sent();
 427                                 bcl->stats.queue_sz_counts++;
 428                                 bcl->stats.accu_queue_sz += len;
 429                         }
 430                         bc = 1;
 431                 }
 432                 tipc_bclink_unlock();
 433         }
 434
 435         if (unlikely(!bc))
 436                 __skb_queue_purge(list);
 437
 438         /* Deliver message clone */
 439         if (likely(!rc))
 440                 tipc_sk_mcast_rcv(skb);
 441         else
 442                 kfree_skb(skb);
 443
 444         return rc;
 445 }
 446
 447 /**
 448  * bclink_accept_pkt - accept an incoming, in-sequence broadcast packet
 449  *
 450  * Called with both sending node's lock and bclink_lock taken.
 451  */
 452 static void bclink_accept_pkt(struct tipc_node *node, u32 seqno)
 453 {
 454         bclink_update_last_sent(node, seqno);
 455         node->bclink.last_in = seqno;
 456         node->bclink.oos_state = 0;
 457         bcl->stats.recv_info++;
 458
 459         /*
 460          * Unicast an ACK periodically, ensuring that
 461          * all nodes in the cluster don't ACK at the same time
 462          */
 463         if (((seqno - tipc_own_addr) % TIPC_MIN_LINK_WIN) == 0) {
 464                 tipc_link_proto_xmit(node->active_links[node->addr & 1],
 465                                      STATE_MSG, 0, 0, 0, 0, 0);
 466                 bcl->stats.sent_acks++;
 467         }
 468 }
 469
 470 /**
 471  * tipc_bclink_rcv - receive a broadcast packet, and deliver upwards
 472  *
 473  * RCU is locked, no other locks set
 474  */
 475 void tipc_bclink_rcv(struct sk_buff *buf)
 476 {
 477         struct tipc_msg *msg = buf_msg(buf);
 478         struct tipc_node *node;
 479         u32 next_in;
 480         u32 seqno;
 481         int deferred = 0;
 482
 483         /* Screen out unwanted broadcast messages */
 484         if (msg_mc_netid(msg) != tipc_net_id)
 485                 goto exit;
 486
 487         node = tipc_node_find(msg_prevnode(msg));
 488         if (unlikely(!node))
 489                 goto exit;
 490
 491         tipc_node_lock(node);
 492         if (unlikely(!node->bclink.recv_permitted))
 493                 goto unlock;
 494
 495         /* Handle broadcast protocol message */
 496         if (unlikely(msg_user(msg) == BCAST_PROTOCOL)) {
 497                 if (msg_type(msg) != STATE_MSG)
 498                         goto unlock;
 499                 if (msg_destnode(msg) == tipc_own_addr) {
 500                         tipc_bclink_acknowledge(node, msg_bcast_ack(msg));
 501                         tipc_node_unlock(node);
 502                         tipc_bclink_lock();
 503                         bcl->stats.recv_nacks++;
 504                         bclink->retransmit_to = node;
 505                         bclink_retransmit_pkt(msg_bcgap_after(msg),
 506                                               msg_bcgap_to(msg));
 507                         tipc_bclink_unlock();
 508                 } else {
 509                         tipc_node_unlock(node);
 510                         bclink_peek_nack(msg);
 511                 }
 512                 goto exit;
 513         }
 514
 515         /* Handle in-sequence broadcast message */
 516         seqno = msg_seqno(msg);
 517         next_in = mod(node->bclink.last_in + 1);
 518
 519         if (likely(seqno == next_in)) {
 520 receive:
 521                 /* Deliver message to destination */
 522                 if (likely(msg_isdata(msg))) {
 523                         tipc_bclink_lock();
 524                         bclink_accept_pkt(node, seqno);
 525                         tipc_bclink_unlock();
 526                         tipc_node_unlock(node);
 527                         if (likely(msg_mcast(msg)))
 528                                 tipc_sk_mcast_rcv(buf);
 529                         else
 530                                 kfree_skb(buf);
 531                 } else if (msg_user(msg) == MSG_BUNDLER) {
 532                         tipc_bclink_lock();
 533                         bclink_accept_pkt(node, seqno);
 534                         bcl->stats.recv_bundles++;
 535                         bcl->stats.recv_bundled += msg_msgcnt(msg);
 536                         tipc_bclink_unlock();
 537                         tipc_node_unlock(node);
 538                         tipc_link_bundle_rcv(buf);
 539                 } else if (msg_user(msg) == MSG_FRAGMENTER) {
 540                         tipc_buf_append(&node->bclink.reasm_buf, &buf);
 541                         if (unlikely(!buf && !node->bclink.reasm_buf))
 542                                 goto unlock;
 543                         tipc_bclink_lock();
 544                         bclink_accept_pkt(node, seqno);
 545                         bcl->stats.recv_fragments++;
 546                         if (buf) {
 547                                 bcl->stats.recv_fragmented++;
 548                                 msg = buf_msg(buf);
 549                                 tipc_bclink_unlock();
 550                                 goto receive;
 551                         }
 552                         tipc_bclink_unlock();
 553                         tipc_node_unlock(node);
 554                 } else if (msg_user(msg) == NAME_DISTRIBUTOR) {
 555                         tipc_bclink_lock();
 556                         bclink_accept_pkt(node, seqno);
 557                         tipc_bclink_unlock();
 558                         tipc_node_unlock(node);
 559                         tipc_named_rcv(buf);
 560                 } else {
 561                         tipc_bclink_lock();
 562                         bclink_accept_pkt(node, seqno);
 563                         tipc_bclink_unlock();
 564                         tipc_node_unlock(node);
 565                         kfree_skb(buf);
 566                 }
 567                 buf = NULL;
 568
 569                 /* Determine new synchronization state */
 570                 tipc_node_lock(node);
 571                 if (unlikely(!tipc_node_is_up(node)))
 572                         goto unlock;
 573
 574                 if (node->bclink.last_in == node->bclink.last_sent)
 575                         goto unlock;
 576
 577                 if (skb_queue_empty(&node->bclink.deferred_queue)) {
 578                         node->bclink.oos_state = 1;
 579                         goto unlock;
 580                 }
 581
 582                 msg = buf_msg(skb_peek(&node->bclink.deferred_queue));
 583                 seqno = msg_seqno(msg);
 584                 next_in = mod(next_in + 1);
 585                 if (seqno != next_in)
 586                         goto unlock;
 587
 588                 /* Take in-sequence message from deferred queue & deliver it */
 589                 buf = __skb_dequeue(&node->bclink.deferred_queue);
 590                 goto receive;
 591         }
 592
 593         /* Handle out-of-sequence broadcast message */
 594         if (less(next_in, seqno)) {
 595                 deferred = tipc_link_defer_pkt(&node->bclink.deferred_queue,
 596                                                buf);
 597                 bclink_update_last_sent(node, seqno);
 598                 buf = NULL;
 599         }
 600
 601         tipc_bclink_lock();
 602
 603         if (deferred)
 604                 bcl->stats.deferred_recv++;
 605         else
 606                 bcl->stats.duplicates++;
 607
 608         tipc_bclink_unlock();
 609
 610 unlock:
 611         tipc_node_unlock(node);
 612 exit:
 613         kfree_skb(buf);
 614 }
 615
 616 u32 tipc_bclink_acks_missing(struct tipc_node *n_ptr)
 617 {
 618         return (n_ptr->bclink.recv_permitted &&
 619                 (tipc_bclink_get_last_sent() != n_ptr->bclink.acked));
 620 }
 621
 622
 623 /**
 624  * tipc_bcbearer_send - send a packet through the broadcast pseudo-bearer
 625  *
 626  * Send packet over as many bearers as necessary to reach all nodes
 627  * that have joined the broadcast link.
 628  *
 629  * Returns 0 (packet sent successfully) under all circumstances,
 630  * since the broadcast link's pseudo-bearer never blocks
 631  */
 632 static int tipc_bcbearer_send(struct sk_buff *buf, struct tipc_bearer *unused1,
 633                               struct tipc_media_addr *unused2)
 634 {
 635         int bp_index;
 636         struct tipc_msg *msg = buf_msg(buf);
 637
 638         /* Prepare broadcast link message for reliable transmission,
 639          * if first time trying to send it;
 640          * preparation is skipped for broadcast link protocol messages
 641          * since they are sent in an unreliable manner and don't need it
 642          */
 643         if (likely(!msg_non_seq(buf_msg(buf)))) {
 644                 bcbuf_set_acks(buf, bclink->bcast_nodes.count);
 645                 msg_set_non_seq(msg, 1);
 646                 msg_set_mc_netid(msg, tipc_net_id);
 647                 bcl->stats.sent_info++;
 648
 649                 if (WARN_ON(!bclink->bcast_nodes.count)) {
 650                         dump_stack();
 651                         return 0;
 652                 }
 653         }
 654
 655         /* Send buffer over bearers until all targets reached */
 656         bcbearer->remains = bclink->bcast_nodes;
 657
 658         for (bp_index = 0; bp_index < MAX_BEARERS; bp_index++) {
 659                 struct tipc_bearer *p = bcbearer->bpairs[bp_index].primary;
 660                 struct tipc_bearer *s = bcbearer->bpairs[bp_index].secondary;
 661                 struct tipc_bearer *bp[2] = {p, s};
 662                 struct tipc_bearer *b = bp[msg_link_selector(msg)];
 663                 struct sk_buff *tbuf;
 664
 665                 if (!p)
 666                         break; /* No more bearers to try */
 667                 if (!b)
 668                         b = p;
 669                 tipc_nmap_diff(&bcbearer->remains, &b->nodes,
 670                                &bcbearer->remains_new);
 671                 if (bcbearer->remains_new.count == bcbearer->remains.count)
 672                         continue; /* Nothing added by bearer pair */
 673
 674                 if (bp_index == 0) {
 675                         /* Use original buffer for first bearer */
 676                         tipc_bearer_send(b->identity, buf, &b->bcast_addr);
 677                 } else {
 678                         /* Avoid concurrent buffer access */
 679                         tbuf = pskb_copy_for_clone(buf, GFP_ATOMIC);
 680                         if (!tbuf)
 681                                 break;
 682                         tipc_bearer_send(b->identity, tbuf, &b->bcast_addr);
 683                         kfree_skb(tbuf); /* Bearer keeps a clone */
 684                 }
 685                 if (bcbearer->remains_new.count == 0)
 686                         break; /* All targets reached */
 687
 688                 bcbearer->remains = bcbearer->remains_new;
 689         }
 690
 691         return 0;
 692 }
 693
 694 /**
 695  * tipc_bcbearer_sort - create sets of bearer pairs used by broadcast bearer
 696  */
 697 void tipc_bcbearer_sort(struct tipc_node_map *nm_ptr, u32 node, bool action)
 698 {
 699         struct tipc_bcbearer_pair *bp_temp = bcbearer->bpairs_temp;
 700         struct tipc_bcbearer_pair *bp_curr;
 701         struct tipc_bearer *b;
 702         int b_index;
 703         int pri;
 704
 705         tipc_bclink_lock();
 706
 707         if (action)
 708                 tipc_nmap_add(nm_ptr, node);
 709         else
 710                 tipc_nmap_remove(nm_ptr, node);
 711
 712         /* Group bearers by priority (can assume max of two per priority) */
 713         memset(bp_temp, 0, sizeof(bcbearer->bpairs_temp));
 714
 715         rcu_read_lock();
 716         for (b_index = 0; b_index < MAX_BEARERS; b_index++) {
 717                 b = rcu_dereference_rtnl(bearer_list[b_index]);
 718                 if (!b || !b->nodes.count)
 719                         continue;
 720
 721                 if (!bp_temp[b->priority].primary)
 722                         bp_temp[b->priority].primary = b;
 723                 else
 724                         bp_temp[b->priority].secondary = b;
 725         }
 726         rcu_read_unlock();
 727
 728         /* Create array of bearer pairs for broadcasting */
 729         bp_curr = bcbearer->bpairs;
 730         memset(bcbearer->bpairs, 0, sizeof(bcbearer->bpairs));
 731
 732         for (pri = TIPC_MAX_LINK_PRI; pri >= 0; pri--) {
 733
 734                 if (!bp_temp[pri].primary)
 735                         continue;
 736
 737                 bp_curr->primary = bp_temp[pri].primary;
 738
 739                 if (bp_temp[pri].secondary) {
 740                         if (tipc_nmap_equal(&bp_temp[pri].primary->nodes,
 741                                             &bp_temp[pri].secondary->nodes)) {
 742                                 bp_curr->secondary = bp_temp[pri].secondary;
 743                         } else {
 744                                 bp_curr++;
 745                                 bp_curr->primary = bp_temp[pri].secondary;
 746                         }
 747                 }
 748
 749                 bp_curr++;
 750         }
 751
 752         tipc_bclink_unlock();
 753 }
 754
 755 static int __tipc_nl_add_bc_link_stat(struct sk_buff *skb,
 756                                       struct tipc_stats *stats)
 757 {
 758         int i;
 759         struct nlattr *nest;
 760
 761         struct nla_map {
 762                 __u32 key;
 763                 __u32 val;
 764         };
 765
 766         struct nla_map map[] = {
 767                 {TIPC_NLA_STATS_RX_INFO, stats->recv_info},
 768                 {TIPC_NLA_STATS_RX_FRAGMENTS, stats->recv_fragments},
 769                 {TIPC_NLA_STATS_RX_FRAGMENTED, stats->recv_fragmented},
 770                 {TIPC_NLA_STATS_RX_BUNDLES, stats->recv_bundles},
 771                 {TIPC_NLA_STATS_RX_BUNDLED, stats->recv_bundled},
 772                 {TIPC_NLA_STATS_TX_INFO, stats->sent_info},
 773                 {TIPC_NLA_STATS_TX_FRAGMENTS, stats->sent_fragments},
 774                 {TIPC_NLA_STATS_TX_FRAGMENTED, stats->sent_fragmented},
 775                 {TIPC_NLA_STATS_TX_BUNDLES, stats->sent_bundles},
 776                 {TIPC_NLA_STATS_TX_BUNDLED, stats->sent_bundled},
 777                 {TIPC_NLA_STATS_RX_NACKS, stats->recv_nacks},
 778                 {TIPC_NLA_STATS_RX_DEFERRED, stats->deferred_recv},
 779                 {TIPC_NLA_STATS_TX_NACKS, stats->sent_nacks},
 780                 {TIPC_NLA_STATS_TX_ACKS, stats->sent_acks},
 781                 {TIPC_NLA_STATS_RETRANSMITTED, stats->retransmitted},
 782                 {TIPC_NLA_STATS_DUPLICATES, stats->duplicates},
 783                 {TIPC_NLA_STATS_LINK_CONGS, stats->link_congs},
 784                 {TIPC_NLA_STATS_MAX_QUEUE, stats->max_queue_sz},
 785                 {TIPC_NLA_STATS_AVG_QUEUE, stats->queue_sz_counts ?
 786                         (stats->accu_queue_sz / stats->queue_sz_counts) : 0}
 787         };
 788
 789         nest = nla_nest_start(skb, TIPC_NLA_LINK_STATS);
 790         if (!nest)
 791                 return -EMSGSIZE;
 792
 793         for (i = 0; i <  ARRAY_SIZE(map); i++)
 794                 if (nla_put_u32(skb, map[i].key, map[i].val))
 795                         goto msg_full;
 796
 797         nla_nest_end(skb, nest);
 798
 799         return 0;
 800 msg_full:
 801         nla_nest_cancel(skb, nest);
 802
 803         return -EMSGSIZE;
 804 }
 805
 806 int tipc_nl_add_bc_link(struct tipc_nl_msg *msg)
 807 {
 808         int err;
 809         void *hdr;
 810         struct nlattr *attrs;
 811         struct nlattr *prop;
 812
 813         if (!bcl)
 814                 return 0;
 815
 816         tipc_bclink_lock();
 817
 818         hdr = genlmsg_put(msg->skb, msg->portid, msg->seq, &tipc_genl_v2_family,
 819                           NLM_F_MULTI, TIPC_NL_LINK_GET);
 820         if (!hdr)
 821                 return -EMSGSIZE;
 822
 823         attrs = nla_nest_start(msg->skb, TIPC_NLA_LINK);
 824         if (!attrs)
 825                 goto msg_full;
 826
 827         /* The broadcast link is always up */
 828         if (nla_put_flag(msg->skb, TIPC_NLA_LINK_UP))
 829                 goto attr_msg_full;
 830
 831         if (nla_put_flag(msg->skb, TIPC_NLA_LINK_BROADCAST))
 832                 goto attr_msg_full;
 833         if (nla_put_string(msg->skb, TIPC_NLA_LINK_NAME, bcl->name))
 834                 goto attr_msg_full;
 835         if (nla_put_u32(msg->skb, TIPC_NLA_LINK_RX, bcl->next_in_no))
 836                 goto attr_msg_full;
 837         if (nla_put_u32(msg->skb, TIPC_NLA_LINK_TX, bcl->next_out_no))
 838                 goto attr_msg_full;
 839
 840         prop = nla_nest_start(msg->skb, TIPC_NLA_LINK_PROP);
 841         if (!prop)
 842                 goto attr_msg_full;
 843         if (nla_put_u32(msg->skb, TIPC_NLA_PROP_WIN, bcl->queue_limit[0]))
 844                 goto prop_msg_full;
 845         nla_nest_end(msg->skb, prop);
 846
 847         err = __tipc_nl_add_bc_link_stat(msg->skb, &bcl->stats);
 848         if (err)
 849                 goto attr_msg_full;
 850
 851         tipc_bclink_unlock();
 852         nla_nest_end(msg->skb, attrs);
 853         genlmsg_end(msg->skb, hdr);
 854
 855         return 0;
 856
 857 prop_msg_full:
 858         nla_nest_cancel(msg->skb, prop);
 859 attr_msg_full:
 860         nla_nest_cancel(msg->skb, attrs);
 861 msg_full:
 862         tipc_bclink_unlock();
 863         genlmsg_cancel(msg->skb, hdr);
 864
 865         return -EMSGSIZE;
 866 }
 867
 868 int tipc_bclink_stats(char *buf, const u32 buf_size)
 869 {
 870         int ret;
 871         struct tipc_stats *s;
 872
 873         if (!bcl)
 874                 return 0;
 875
 876         tipc_bclink_lock();
 877
 878         s = &bcl->stats;
 879
 880         ret = tipc_snprintf(buf, buf_size, "Link <%s>\n"
 881                             "  Window:%u packets\n",
 882                             bcl->name, bcl->queue_limit[0]);
 883         ret += tipc_snprintf(buf + ret, buf_size - ret,
 884                              "  RX packets:%u fragments:%u/%u bundles:%u/%u\n",
 885                              s->recv_info, s->recv_fragments,
 886                              s->recv_fragmented, s->recv_bundles,
 887                              s->recv_bundled);
 888         ret += tipc_snprintf(buf + ret, buf_size - ret,
 889                              "  TX packets:%u fragments:%u/%u bundles:%u/%u\n",
 890                              s->sent_info, s->sent_fragments,
 891                              s->sent_fragmented, s->sent_bundles,
 892                              s->sent_bundled);
 893         ret += tipc_snprintf(buf + ret, buf_size - ret,
 894                              "  RX naks:%u defs:%u dups:%u\n",
 895                              s->recv_nacks, s->deferred_recv, s->duplicates);
 896         ret += tipc_snprintf(buf + ret, buf_size - ret,
 897                              "  TX naks:%u acks:%u dups:%u\n",
 898                              s->sent_nacks, s->sent_acks, s->retransmitted);
 899         ret += tipc_snprintf(buf + ret, buf_size - ret,
 900                              "  Congestion link:%u  Send queue max:%u avg:%u\n",
 901                              s->link_congs, s->max_queue_sz,
 902                              s->queue_sz_counts ?
 903                              (s->accu_queue_sz / s->queue_sz_counts) : 0);
 904
 905         tipc_bclink_unlock();
 906         return ret;
 907 }
 908
 909 int tipc_bclink_reset_stats(void)
 910 {
 911         if (!bcl)
 912                 return -ENOPROTOOPT;
 913
 914         tipc_bclink_lock();
 915         memset(&bcl->stats, 0, sizeof(bcl->stats));
 916         tipc_bclink_unlock();
 917         return 0;
 918 }
 919
 920 int tipc_bclink_set_queue_limits(u32 limit)
 921 {
 922         if (!bcl)
 923                 return -ENOPROTOOPT;
 924         if ((limit < TIPC_MIN_LINK_WIN) || (limit > TIPC_MAX_LINK_WIN))
 925                 return -EINVAL;
 926
 927         tipc_bclink_lock();
 928         tipc_link_set_queue_limits(bcl, limit);
 929         tipc_bclink_unlock();
 930         return 0;
 931 }
 932
 933 int tipc_bclink_init(void)
 934 {
 935         bcbearer = kzalloc(sizeof(*bcbearer), GFP_ATOMIC);
 936         if (!bcbearer)
 937                 return -ENOMEM;
 938
 939         bclink = kzalloc(sizeof(*bclink), GFP_ATOMIC);
 940         if (!bclink) {
 941                 kfree(bcbearer);
 942                 return -ENOMEM;
 943         }
 944
 945         bcl = &bclink->link;
 946         bcbearer->bearer.media = &bcbearer->media;
 947         bcbearer->media.send_msg = tipc_bcbearer_send;
 948         sprintf(bcbearer->media.name, "tipc-broadcast");
 949
 950         spin_lock_init(&bclink->lock);
 951         __skb_queue_head_init(&bcl->outqueue);
 952         __skb_queue_head_init(&bcl->deferred_queue);
 953         __skb_queue_head_init(&bcl->waiting_sks);
 954         bcl->next_out_no = 1;
 955         spin_lock_init(&bclink->node.lock);
 956         __skb_queue_head_init(&bclink->node.waiting_sks);
 957         bcl->owner = &bclink->node;
 958         bcl->max_pkt = MAX_PKT_DEFAULT_MCAST;
 959         tipc_link_set_queue_limits(bcl, BCLINK_WIN_DEFAULT);
 960         bcl->bearer_id = MAX_BEARERS;
 961         rcu_assign_pointer(bearer_list[MAX_BEARERS], &bcbearer->bearer);
 962         bcl->state = WORKING_WORKING;
 963         strlcpy(bcl->name, tipc_bclink_name, TIPC_MAX_LINK_NAME);
 964         return 0;
 965 }
 966
 967 void tipc_bclink_stop(void)
 968 {
 969         tipc_bclink_lock();
 970         tipc_link_purge_queues(bcl);
 971         tipc_bclink_unlock();
 972
 973         RCU_INIT_POINTER(bearer_list[BCBEARER], NULL);
 974         synchronize_net();
 975         kfree(bcbearer);
 976         kfree(bclink);
 977 }
 978
 979 /**
 980  * tipc_nmap_add - add a node to a node map
 981  */
 982 static void tipc_nmap_add(struct tipc_node_map *nm_ptr, u32 node)
 983 {
 984         int n = tipc_node(node);
 985         int w = n / WSIZE;
 986         u32 mask = (1 << (n % WSIZE));
 987
 988         if ((nm_ptr->map[w] & mask) == 0) {
 989                 nm_ptr->count++;
 990                 nm_ptr->map[w] |= mask;
 991         }
 992 }
 993
 994 /**
 995  * tipc_nmap_remove - remove a node from a node map
 996  */
 997 static void tipc_nmap_remove(struct tipc_node_map *nm_ptr, u32 node)
 998 {
 999         int n = tipc_node(node);
1000         int w = n / WSIZE;
1001         u32 mask = (1 << (n % WSIZE));
1002
1003         if ((nm_ptr->map[w] & mask) != 0) {
1004                 nm_ptr->map[w] &= ~mask;
1005                 nm_ptr->count--;
1006         }
1007 }
1008
1009 /**
1010  * tipc_nmap_diff - find differences between node maps
1011  * @nm_a: input node map A
1012  * @nm_b: input node map B
1013  * @nm_diff: output node map A-B (i.e. nodes of A that are not in B)
1014  */
1015 static void tipc_nmap_diff(struct tipc_node_map *nm_a,
1016                            struct tipc_node_map *nm_b,
1017                            struct tipc_node_map *nm_diff)
1018 {
1019         int stop = ARRAY_SIZE(nm_a->map);
1020         int w;
1021         int b;
1022         u32 map;
1023
1024         memset(nm_diff, 0, sizeof(*nm_diff));
1025         for (w = 0; w < stop; w++) {
1026                 map = nm_a->map[w] ^ (nm_a->map[w] & nm_b->map[w]);
1027                 nm_diff->map[w] = map;
1028                 if (map != 0) {
1029                         for (b = 0 ; b < WSIZE; b++) {
1030                                 if (map & (1 << b))
1031                                         nm_diff->count++;
1032                         }
1033                 }
1034         }
1035 }
1036
1037 /**
1038  * tipc_port_list_add - add a port to a port list, ensuring no duplicates
1039  */
1040 void tipc_port_list_add(struct tipc_port_list *pl_ptr, u32 port)
1041 {
1042         struct tipc_port_list *item = pl_ptr;
1043         int i;
1044         int item_sz = PLSIZE;
1045         int cnt = pl_ptr->count;
1046
1047         for (; ; cnt -= item_sz, item = item->next) {
1048                 if (cnt < PLSIZE)
1049                         item_sz = cnt;
1050                 for (i = 0; i < item_sz; i++)
1051                         if (item->ports[i] == port)
1052                                 return;
1053                 if (i < PLSIZE) {
1054                         item->ports[i] = port;
1055                         pl_ptr->count++;
1056                         return;
1057                 }
1058                 if (!item->next) {
1059                         item->next = kmalloc(sizeof(*item), GFP_ATOMIC);
1060                         if (!item->next) {
1061                                 pr_warn("Incomplete multicast delivery, no memory\n");
1062                                 return;
1063                         }
1064                         item->next->next = NULL;
1065                 }
1066         }
1067 }
1068
1069 /**
1070  * tipc_port_list_free - free dynamically created entries in port_list chain
1071  *
1072  */
1073 void tipc_port_list_free(struct tipc_port_list *pl_ptr)
1074 {
1075         struct tipc_port_list *item;
1076         struct tipc_port_list *next;
1077
1078         for (item = pl_ptr->next; item; item = next) {
1079                 next = item->next;
1080                 kfree(item);
1081         }
1082 }