net/tipc/bcast.c

   1 /*
   2  * net/tipc/bcast.c: TIPC broadcast code
   3  *
   4  * Copyright (c) 2004-2006, 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 "port.h"
  41 #include "bcast.h"
  42 #include "name_distr.h"
  43
  44 #define MAX_PKT_DEFAULT_MCAST   1500    /* bcast link max packet size (fixed) */
  45 #define BCLINK_WIN_DEFAULT      20      /* bcast link window size (default) */
  46 #define BCBEARER                MAX_BEARERS
  47
  48 /**
  49  * struct tipc_bcbearer_pair - a pair of bearers used by broadcast link
  50  * @primary: pointer to primary bearer
  51  * @secondary: pointer to secondary bearer
  52  *
  53  * Bearers must have same priority and same set of reachable destinations
  54  * to be paired.
  55  */
  56
  57 struct tipc_bcbearer_pair {
  58         struct tipc_bearer *primary;
  59         struct tipc_bearer *secondary;
  60 };
  61
  62 /**
  63  * struct tipc_bcbearer - bearer used by broadcast link
  64  * @bearer: (non-standard) broadcast bearer structure
  65  * @media: (non-standard) broadcast media structure
  66  * @bpairs: array of bearer pairs
  67  * @bpairs_temp: temporary array of bearer pairs used by tipc_bcbearer_sort()
  68  * @remains: temporary node map used by tipc_bcbearer_send()
  69  * @remains_new: temporary node map used tipc_bcbearer_send()
  70  *
  71  * Note: The fields labelled "temporary" are incorporated into the bearer
  72  * to avoid consuming potentially limited stack space through the use of
  73  * large local variables within multicast routines.  Concurrent access is
  74  * prevented through use of the spinlock "bc_lock".
  75  */
  76 struct tipc_bcbearer {
  77         struct tipc_bearer bearer;
  78         struct tipc_media media;
  79         struct tipc_bcbearer_pair bpairs[MAX_BEARERS];
  80         struct tipc_bcbearer_pair bpairs_temp[TIPC_MAX_LINK_PRI + 1];
  81         struct tipc_node_map remains;
  82         struct tipc_node_map remains_new;
  83 };
  84
  85 /**
  86  * struct tipc_bclink - link used for broadcast messages
  87  * @link: (non-standard) broadcast link structure
  88  * @node: (non-standard) node structure representing b'cast link's peer node
  89  * @bcast_nodes: map of broadcast-capable nodes
  90  * @retransmit_to: node that most recently requested a retransmit
  91  *
  92  * Handles sequence numbering, fragmentation, bundling, etc.
  93  */
  94 struct tipc_bclink {
  95         struct tipc_link link;
  96         struct tipc_node node;
  97         struct tipc_node_map bcast_nodes;
  98         struct tipc_node *retransmit_to;
  99 };
 100
 101 static struct tipc_bcbearer bcast_bearer;
 102 static struct tipc_bclink bcast_link;
 103
 104 static struct tipc_bcbearer *bcbearer = &bcast_bearer;
 105 static struct tipc_bclink *bclink = &bcast_link;
 106 static struct tipc_link *bcl = &bcast_link.link;
 107
 108 static DEFINE_SPINLOCK(bc_lock);
 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 u32 bcbuf_acks(struct sk_buff *buf)
 119 {
 120         return (u32)(unsigned long)TIPC_SKB_CB(buf)->handle;
 121 }
 122
 123 static void bcbuf_set_acks(struct sk_buff *buf, u32 acks)
 124 {
 125         TIPC_SKB_CB(buf)->handle = (void *)(unsigned long)acks;
 126 }
 127
 128 static void bcbuf_decr_acks(struct sk_buff *buf)
 129 {
 130         bcbuf_set_acks(buf, bcbuf_acks(buf) - 1);
 131 }
 132
 133 void tipc_bclink_add_node(u32 addr)
 134 {
 135         spin_lock_bh(&bc_lock);
 136         tipc_nmap_add(&bclink->bcast_nodes, addr);
 137         spin_unlock_bh(&bc_lock);
 138 }
 139
 140 void tipc_bclink_remove_node(u32 addr)
 141 {
 142         spin_lock_bh(&bc_lock);
 143         tipc_nmap_remove(&bclink->bcast_nodes, addr);
 144         spin_unlock_bh(&bc_lock);
 145 }
 146
 147 static void bclink_set_last_sent(void)
 148 {
 149         if (bcl->next_out)
 150                 bcl->fsm_msg_cnt = mod(buf_seqno(bcl->next_out) - 1);
 151         else
 152                 bcl->fsm_msg_cnt = mod(bcl->next_out_no - 1);
 153 }
 154
 155 u32 tipc_bclink_get_last_sent(void)
 156 {
 157         return bcl->fsm_msg_cnt;
 158 }
 159
 160 static void bclink_update_last_sent(struct tipc_node *node, u32 seqno)
 161 {
 162         node->bclink.last_sent = less_eq(node->bclink.last_sent, seqno) ?
 163                                                 seqno : node->bclink.last_sent;
 164 }
 165
 166
 167 /**
 168  * tipc_bclink_retransmit_to - get most recent node to request retransmission
 169  *
 170  * Called with bc_lock locked
 171  */
 172 struct tipc_node *tipc_bclink_retransmit_to(void)
 173 {
 174         return bclink->retransmit_to;
 175 }
 176
 177 /**
 178  * bclink_retransmit_pkt - retransmit broadcast packets
 179  * @after: sequence number of last packet to *not* retransmit
 180  * @to: sequence number of last packet to retransmit
 181  *
 182  * Called with bc_lock locked
 183  */
 184 static void bclink_retransmit_pkt(u32 after, u32 to)
 185 {
 186         struct sk_buff *buf;
 187
 188         buf = bcl->first_out;
 189         while (buf && less_eq(buf_seqno(buf), after))
 190                 buf = buf->next;
 191         tipc_link_retransmit(bcl, buf, mod(to - after));
 192 }
 193
 194 /**
 195  * tipc_bclink_acknowledge - handle acknowledgement of broadcast packets
 196  * @n_ptr: node that sent acknowledgement info
 197  * @acked: broadcast sequence # that has been acknowledged
 198  *
 199  * Node is locked, bc_lock unlocked.
 200  */
 201 void tipc_bclink_acknowledge(struct tipc_node *n_ptr, u32 acked)
 202 {
 203         struct sk_buff *crs;
 204         struct sk_buff *next;
 205         unsigned int released = 0;
 206
 207         spin_lock_bh(&bc_lock);
 208
 209         /* Bail out if tx queue is empty (no clean up is required) */
 210         crs = bcl->first_out;
 211         if (!crs)
 212                 goto exit;
 213
 214         /* Determine which messages need to be acknowledged */
 215         if (acked == INVALID_LINK_SEQ) {
 216                 /*
 217                  * Contact with specified node has been lost, so need to
 218                  * acknowledge sent messages only (if other nodes still exist)
 219                  * or both sent and unsent messages (otherwise)
 220                  */
 221                 if (bclink->bcast_nodes.count)
 222                         acked = bcl->fsm_msg_cnt;
 223                 else
 224                         acked = bcl->next_out_no;
 225         } else {
 226                 /*
 227                  * Bail out if specified sequence number does not correspond
 228                  * to a message that has been sent and not yet acknowledged
 229                  */
 230                 if (less(acked, buf_seqno(crs)) ||
 231                     less(bcl->fsm_msg_cnt, acked) ||
 232                     less_eq(acked, n_ptr->bclink.acked))
 233                         goto exit;
 234         }
 235
 236         /* Skip over packets that node has previously acknowledged */
 237         while (crs && less_eq(buf_seqno(crs), n_ptr->bclink.acked))
 238                 crs = crs->next;
 239
 240         /* Update packets that node is now acknowledging */
 241
 242         while (crs && less_eq(buf_seqno(crs), acked)) {
 243                 next = crs->next;
 244
 245                 if (crs != bcl->next_out)
 246                         bcbuf_decr_acks(crs);
 247                 else {
 248                         bcbuf_set_acks(crs, 0);
 249                         bcl->next_out = next;
 250                         bclink_set_last_sent();
 251                 }
 252
 253                 if (bcbuf_acks(crs) == 0) {
 254                         bcl->first_out = next;
 255                         bcl->out_queue_size--;
 256                         kfree_skb(crs);
 257                         released = 1;
 258                 }
 259                 crs = next;
 260         }
 261         n_ptr->bclink.acked = acked;
 262
 263         /* Try resolving broadcast link congestion, if necessary */
 264
 265         if (unlikely(bcl->next_out)) {
 266                 tipc_link_push_queue(bcl);
 267                 bclink_set_last_sent();
 268         }
 269         if (unlikely(released && !list_empty(&bcl->waiting_ports)))
 270                 tipc_link_wakeup_ports(bcl, 0);
 271 exit:
 272         spin_unlock_bh(&bc_lock);
 273 }
 274
 275 /**
 276  * tipc_bclink_update_link_state - update broadcast link state
 277  *
 278  * RCU and node lock set
 279  */
 280 void tipc_bclink_update_link_state(struct tipc_node *n_ptr, u32 last_sent)
 281 {
 282         struct sk_buff *buf;
 283
 284         /* Ignore "stale" link state info */
 285
 286         if (less_eq(last_sent, n_ptr->bclink.last_in))
 287                 return;
 288
 289         /* Update link synchronization state; quit if in sync */
 290
 291         bclink_update_last_sent(n_ptr, last_sent);
 292
 293         if (n_ptr->bclink.last_sent == n_ptr->bclink.last_in)
 294                 return;
 295
 296         /* Update out-of-sync state; quit if loss is still unconfirmed */
 297
 298         if ((++n_ptr->bclink.oos_state) == 1) {
 299                 if (n_ptr->bclink.deferred_size < (TIPC_MIN_LINK_WIN / 2))
 300                         return;
 301                 n_ptr->bclink.oos_state++;
 302         }
 303
 304         /* Don't NACK if one has been recently sent (or seen) */
 305
 306         if (n_ptr->bclink.oos_state & 0x1)
 307                 return;
 308
 309         /* Send NACK */
 310
 311         buf = tipc_buf_acquire(INT_H_SIZE);
 312         if (buf) {
 313                 struct tipc_msg *msg = buf_msg(buf);
 314
 315                 tipc_msg_init(msg, BCAST_PROTOCOL, STATE_MSG,
 316                               INT_H_SIZE, n_ptr->addr);
 317                 msg_set_non_seq(msg, 1);
 318                 msg_set_mc_netid(msg, tipc_net_id);
 319                 msg_set_bcast_ack(msg, n_ptr->bclink.last_in);
 320                 msg_set_bcgap_after(msg, n_ptr->bclink.last_in);
 321                 msg_set_bcgap_to(msg, n_ptr->bclink.deferred_head
 322                                  ? buf_seqno(n_ptr->bclink.deferred_head) - 1
 323                                  : n_ptr->bclink.last_sent);
 324
 325                 spin_lock_bh(&bc_lock);
 326                 tipc_bearer_send(MAX_BEARERS, buf, NULL);
 327                 bcl->stats.sent_nacks++;
 328                 spin_unlock_bh(&bc_lock);
 329                 kfree_skb(buf);
 330
 331                 n_ptr->bclink.oos_state++;
 332         }
 333 }
 334
 335 /**
 336  * bclink_peek_nack - monitor retransmission requests sent by other nodes
 337  *
 338  * Delay any upcoming NACK by this node if another node has already
 339  * requested the first message this node is going to ask for.
 340  */
 341 static void bclink_peek_nack(struct tipc_msg *msg)
 342 {
 343         struct tipc_node *n_ptr = tipc_node_find(msg_destnode(msg));
 344
 345         if (unlikely(!n_ptr))
 346                 return;
 347
 348         tipc_node_lock(n_ptr);
 349
 350         if (n_ptr->bclink.recv_permitted &&
 351             (n_ptr->bclink.last_in != n_ptr->bclink.last_sent) &&
 352             (n_ptr->bclink.last_in == msg_bcgap_after(msg)))
 353                 n_ptr->bclink.oos_state = 2;
 354
 355         tipc_node_unlock(n_ptr);
 356 }
 357
 358 /*
 359  * tipc_bclink_xmit - broadcast a packet to all nodes in cluster
 360  */
 361 int tipc_bclink_xmit(struct sk_buff *buf)
 362 {
 363         int res;
 364
 365         spin_lock_bh(&bc_lock);
 366
 367         if (!bclink->bcast_nodes.count) {
 368                 res = msg_data_sz(buf_msg(buf));
 369                 kfree_skb(buf);
 370                 goto exit;
 371         }
 372
 373         res = __tipc_link_xmit(bcl, buf);
 374         if (likely(res >= 0)) {
 375                 bclink_set_last_sent();
 376                 bcl->stats.queue_sz_counts++;
 377                 bcl->stats.accu_queue_sz += bcl->out_queue_size;
 378         }
 379 exit:
 380         spin_unlock_bh(&bc_lock);
 381         return res;
 382 }
 383
 384 /**
 385  * bclink_accept_pkt - accept an incoming, in-sequence broadcast packet
 386  *
 387  * Called with both sending node's lock and bc_lock taken.
 388  */
 389 static void bclink_accept_pkt(struct tipc_node *node, u32 seqno)
 390 {
 391         bclink_update_last_sent(node, seqno);
 392         node->bclink.last_in = seqno;
 393         node->bclink.oos_state = 0;
 394         bcl->stats.recv_info++;
 395
 396         /*
 397          * Unicast an ACK periodically, ensuring that
 398          * all nodes in the cluster don't ACK at the same time
 399          */
 400
 401         if (((seqno - tipc_own_addr) % TIPC_MIN_LINK_WIN) == 0) {
 402                 tipc_link_proto_xmit(node->active_links[node->addr & 1],
 403                                      STATE_MSG, 0, 0, 0, 0, 0);
 404                 bcl->stats.sent_acks++;
 405         }
 406 }
 407
 408 /**
 409  * tipc_bclink_rcv - receive a broadcast packet, and deliver upwards
 410  *
 411  * RCU is locked, no other locks set
 412  */
 413 void tipc_bclink_rcv(struct sk_buff *buf)
 414 {
 415         struct tipc_msg *msg = buf_msg(buf);
 416         struct tipc_node *node;
 417         u32 next_in;
 418         u32 seqno;
 419         int deferred;
 420
 421         /* Screen out unwanted broadcast messages */
 422
 423         if (msg_mc_netid(msg) != tipc_net_id)
 424                 goto exit;
 425
 426         node = tipc_node_find(msg_prevnode(msg));
 427         if (unlikely(!node))
 428                 goto exit;
 429
 430         tipc_node_lock(node);
 431         if (unlikely(!node->bclink.recv_permitted))
 432                 goto unlock;
 433
 434         /* Handle broadcast protocol message */
 435
 436         if (unlikely(msg_user(msg) == BCAST_PROTOCOL)) {
 437                 if (msg_type(msg) != STATE_MSG)
 438                         goto unlock;
 439                 if (msg_destnode(msg) == tipc_own_addr) {
 440                         tipc_bclink_acknowledge(node, msg_bcast_ack(msg));
 441                         tipc_node_unlock(node);
 442                         spin_lock_bh(&bc_lock);
 443                         bcl->stats.recv_nacks++;
 444                         bclink->retransmit_to = node;
 445                         bclink_retransmit_pkt(msg_bcgap_after(msg),
 446                                               msg_bcgap_to(msg));
 447                         spin_unlock_bh(&bc_lock);
 448                 } else {
 449                         tipc_node_unlock(node);
 450                         bclink_peek_nack(msg);
 451                 }
 452                 goto exit;
 453         }
 454
 455         /* Handle in-sequence broadcast message */
 456
 457         seqno = msg_seqno(msg);
 458         next_in = mod(node->bclink.last_in + 1);
 459
 460         if (likely(seqno == next_in)) {
 461 receive:
 462                 /* Deliver message to destination */
 463
 464                 if (likely(msg_isdata(msg))) {
 465                         spin_lock_bh(&bc_lock);
 466                         bclink_accept_pkt(node, seqno);
 467                         spin_unlock_bh(&bc_lock);
 468                         tipc_node_unlock(node);
 469                         if (likely(msg_mcast(msg)))
 470                                 tipc_port_mcast_rcv(buf, NULL);
 471                         else
 472                                 kfree_skb(buf);
 473                 } else if (msg_user(msg) == MSG_BUNDLER) {
 474                         spin_lock_bh(&bc_lock);
 475                         bclink_accept_pkt(node, seqno);
 476                         bcl->stats.recv_bundles++;
 477                         bcl->stats.recv_bundled += msg_msgcnt(msg);
 478                         spin_unlock_bh(&bc_lock);
 479                         tipc_node_unlock(node);
 480                         tipc_link_bundle_rcv(buf);
 481                 } else if (msg_user(msg) == MSG_FRAGMENTER) {
 482                         int ret;
 483                         ret = tipc_link_frag_rcv(&node->bclink.reasm_head,
 484                                                  &node->bclink.reasm_tail,
 485                                                  &buf);
 486                         if (ret == LINK_REASM_ERROR)
 487                                 goto unlock;
 488                         spin_lock_bh(&bc_lock);
 489                         bclink_accept_pkt(node, seqno);
 490                         bcl->stats.recv_fragments++;
 491                         if (ret == LINK_REASM_COMPLETE) {
 492                                 bcl->stats.recv_fragmented++;
 493                                 /* Point msg to inner header */
 494                                 msg = buf_msg(buf);
 495                                 spin_unlock_bh(&bc_lock);
 496                                 goto receive;
 497                         }
 498                         spin_unlock_bh(&bc_lock);
 499                         tipc_node_unlock(node);
 500                 } else if (msg_user(msg) == NAME_DISTRIBUTOR) {
 501                         spin_lock_bh(&bc_lock);
 502                         bclink_accept_pkt(node, seqno);
 503                         spin_unlock_bh(&bc_lock);
 504                         tipc_node_unlock(node);
 505                         tipc_named_rcv(buf);
 506                 } else {
 507                         spin_lock_bh(&bc_lock);
 508                         bclink_accept_pkt(node, seqno);
 509                         spin_unlock_bh(&bc_lock);
 510                         tipc_node_unlock(node);
 511                         kfree_skb(buf);
 512                 }
 513                 buf = NULL;
 514
 515                 /* Determine new synchronization state */
 516
 517                 tipc_node_lock(node);
 518                 if (unlikely(!tipc_node_is_up(node)))
 519                         goto unlock;
 520
 521                 if (node->bclink.last_in == node->bclink.last_sent)
 522                         goto unlock;
 523
 524                 if (!node->bclink.deferred_head) {
 525                         node->bclink.oos_state = 1;
 526                         goto unlock;
 527                 }
 528
 529                 msg = buf_msg(node->bclink.deferred_head);
 530                 seqno = msg_seqno(msg);
 531                 next_in = mod(next_in + 1);
 532                 if (seqno != next_in)
 533                         goto unlock;
 534
 535                 /* Take in-sequence message from deferred queue & deliver it */
 536
 537                 buf = node->bclink.deferred_head;
 538                 node->bclink.deferred_head = buf->next;
 539                 node->bclink.deferred_size--;
 540                 goto receive;
 541         }
 542
 543         /* Handle out-of-sequence broadcast message */
 544
 545         if (less(next_in, seqno)) {
 546                 deferred = tipc_link_defer_pkt(&node->bclink.deferred_head,
 547                                                &node->bclink.deferred_tail,
 548                                                buf);
 549                 node->bclink.deferred_size += deferred;
 550                 bclink_update_last_sent(node, seqno);
 551                 buf = NULL;
 552         } else
 553                 deferred = 0;
 554
 555         spin_lock_bh(&bc_lock);
 556
 557         if (deferred)
 558                 bcl->stats.deferred_recv++;
 559         else
 560                 bcl->stats.duplicates++;
 561
 562         spin_unlock_bh(&bc_lock);
 563
 564 unlock:
 565         tipc_node_unlock(node);
 566 exit:
 567         kfree_skb(buf);
 568 }
 569
 570 u32 tipc_bclink_acks_missing(struct tipc_node *n_ptr)
 571 {
 572         return (n_ptr->bclink.recv_permitted &&
 573                 (tipc_bclink_get_last_sent() != n_ptr->bclink.acked));
 574 }
 575
 576
 577 /**
 578  * tipc_bcbearer_send - send a packet through the broadcast pseudo-bearer
 579  *
 580  * Send packet over as many bearers as necessary to reach all nodes
 581  * that have joined the broadcast link.
 582  *
 583  * Returns 0 (packet sent successfully) under all circumstances,
 584  * since the broadcast link's pseudo-bearer never blocks
 585  */
 586 static int tipc_bcbearer_send(struct sk_buff *buf, struct tipc_bearer *unused1,
 587                               struct tipc_media_addr *unused2)
 588 {
 589         int bp_index;
 590
 591         /* Prepare broadcast link message for reliable transmission,
 592          * if first time trying to send it;
 593          * preparation is skipped for broadcast link protocol messages
 594          * since they are sent in an unreliable manner and don't need it
 595          */
 596         if (likely(!msg_non_seq(buf_msg(buf)))) {
 597                 struct tipc_msg *msg;
 598
 599                 bcbuf_set_acks(buf, bclink->bcast_nodes.count);
 600                 msg = buf_msg(buf);
 601                 msg_set_non_seq(msg, 1);
 602                 msg_set_mc_netid(msg, tipc_net_id);
 603                 bcl->stats.sent_info++;
 604
 605                 if (WARN_ON(!bclink->bcast_nodes.count)) {
 606                         dump_stack();
 607                         return 0;
 608                 }
 609         }
 610
 611         /* Send buffer over bearers until all targets reached */
 612         bcbearer->remains = bclink->bcast_nodes;
 613
 614         for (bp_index = 0; bp_index < MAX_BEARERS; bp_index++) {
 615                 struct tipc_bearer *p = bcbearer->bpairs[bp_index].primary;
 616                 struct tipc_bearer *s = bcbearer->bpairs[bp_index].secondary;
 617                 struct tipc_bearer *b = p;
 618                 struct sk_buff *tbuf;
 619
 620                 if (!p)
 621                         break; /* No more bearers to try */
 622
 623                 tipc_nmap_diff(&bcbearer->remains, &b->nodes,
 624                                &bcbearer->remains_new);
 625                 if (bcbearer->remains_new.count == bcbearer->remains.count)
 626                         continue; /* Nothing added by bearer pair */
 627
 628                 if (bp_index == 0) {
 629                         /* Use original buffer for first bearer */
 630                         tipc_bearer_send(b->identity, buf, &b->bcast_addr);
 631                 } else {
 632                         /* Avoid concurrent buffer access */
 633                         tbuf = pskb_copy(buf, GFP_ATOMIC);
 634                         if (!tbuf)
 635                                 break;
 636                         tipc_bearer_send(b->identity, tbuf, &b->bcast_addr);
 637                         kfree_skb(tbuf); /* Bearer keeps a clone */
 638                 }
 639
 640                 /* Swap bearers for next packet */
 641                 if (s) {
 642                         bcbearer->bpairs[bp_index].primary = s;
 643                         bcbearer->bpairs[bp_index].secondary = p;
 644                 }
 645
 646                 if (bcbearer->remains_new.count == 0)
 647                         break; /* All targets reached */
 648
 649                 bcbearer->remains = bcbearer->remains_new;
 650         }
 651
 652         return 0;
 653 }
 654
 655 /**
 656  * tipc_bcbearer_sort - create sets of bearer pairs used by broadcast bearer
 657  */
 658 void tipc_bcbearer_sort(struct tipc_node_map *nm_ptr, u32 node, bool action)
 659 {
 660         struct tipc_bcbearer_pair *bp_temp = bcbearer->bpairs_temp;
 661         struct tipc_bcbearer_pair *bp_curr;
 662         struct tipc_bearer *b;
 663         int b_index;
 664         int pri;
 665
 666         spin_lock_bh(&bc_lock);
 667
 668         if (action)
 669                 tipc_nmap_add(nm_ptr, node);
 670         else
 671                 tipc_nmap_remove(nm_ptr, node);
 672
 673         /* Group bearers by priority (can assume max of two per priority) */
 674         memset(bp_temp, 0, sizeof(bcbearer->bpairs_temp));
 675
 676         rcu_read_lock();
 677         for (b_index = 0; b_index < MAX_BEARERS; b_index++) {
 678                 b = rcu_dereference_rtnl(bearer_list[b_index]);
 679                 if (!b || !b->nodes.count)
 680                         continue;
 681
 682                 if (!bp_temp[b->priority].primary)
 683                         bp_temp[b->priority].primary = b;
 684                 else
 685                         bp_temp[b->priority].secondary = b;
 686         }
 687         rcu_read_unlock();
 688
 689         /* Create array of bearer pairs for broadcasting */
 690         bp_curr = bcbearer->bpairs;
 691         memset(bcbearer->bpairs, 0, sizeof(bcbearer->bpairs));
 692
 693         for (pri = TIPC_MAX_LINK_PRI; pri >= 0; pri--) {
 694
 695                 if (!bp_temp[pri].primary)
 696                         continue;
 697
 698                 bp_curr->primary = bp_temp[pri].primary;
 699
 700                 if (bp_temp[pri].secondary) {
 701                         if (tipc_nmap_equal(&bp_temp[pri].primary->nodes,
 702                                             &bp_temp[pri].secondary->nodes)) {
 703                                 bp_curr->secondary = bp_temp[pri].secondary;
 704                         } else {
 705                                 bp_curr++;
 706                                 bp_curr->primary = bp_temp[pri].secondary;
 707                         }
 708                 }
 709
 710                 bp_curr++;
 711         }
 712
 713         spin_unlock_bh(&bc_lock);
 714 }
 715
 716
 717 int tipc_bclink_stats(char *buf, const u32 buf_size)
 718 {
 719         int ret;
 720         struct tipc_stats *s;
 721
 722         if (!bcl)
 723                 return 0;
 724
 725         spin_lock_bh(&bc_lock);
 726
 727         s = &bcl->stats;
 728
 729         ret = tipc_snprintf(buf, buf_size, "Link <%s>\n"
 730                             "  Window:%u packets\n",
 731                             bcl->name, bcl->queue_limit[0]);
 732         ret += tipc_snprintf(buf + ret, buf_size - ret,
 733                              "  RX packets:%u fragments:%u/%u bundles:%u/%u\n",
 734                              s->recv_info, s->recv_fragments,
 735                              s->recv_fragmented, s->recv_bundles,
 736                              s->recv_bundled);
 737         ret += tipc_snprintf(buf + ret, buf_size - ret,
 738                              "  TX packets:%u fragments:%u/%u bundles:%u/%u\n",
 739                              s->sent_info, s->sent_fragments,
 740                              s->sent_fragmented, s->sent_bundles,
 741                              s->sent_bundled);
 742         ret += tipc_snprintf(buf + ret, buf_size - ret,
 743                              "  RX naks:%u defs:%u dups:%u\n",
 744                              s->recv_nacks, s->deferred_recv, s->duplicates);
 745         ret += tipc_snprintf(buf + ret, buf_size - ret,
 746                              "  TX naks:%u acks:%u dups:%u\n",
 747                              s->sent_nacks, s->sent_acks, s->retransmitted);
 748         ret += tipc_snprintf(buf + ret, buf_size - ret,
 749                              "  Congestion link:%u  Send queue max:%u avg:%u\n",
 750                              s->link_congs, s->max_queue_sz,
 751                              s->queue_sz_counts ?
 752                              (s->accu_queue_sz / s->queue_sz_counts) : 0);
 753
 754         spin_unlock_bh(&bc_lock);
 755         return ret;
 756 }
 757
 758 int tipc_bclink_reset_stats(void)
 759 {
 760         if (!bcl)
 761                 return -ENOPROTOOPT;
 762
 763         spin_lock_bh(&bc_lock);
 764         memset(&bcl->stats, 0, sizeof(bcl->stats));
 765         spin_unlock_bh(&bc_lock);
 766         return 0;
 767 }
 768
 769 int tipc_bclink_set_queue_limits(u32 limit)
 770 {
 771         if (!bcl)
 772                 return -ENOPROTOOPT;
 773         if ((limit < TIPC_MIN_LINK_WIN) || (limit > TIPC_MAX_LINK_WIN))
 774                 return -EINVAL;
 775
 776         spin_lock_bh(&bc_lock);
 777         tipc_link_set_queue_limits(bcl, limit);
 778         spin_unlock_bh(&bc_lock);
 779         return 0;
 780 }
 781
 782 void tipc_bclink_init(void)
 783 {
 784         bcbearer->bearer.media = &bcbearer->media;
 785         bcbearer->media.send_msg = tipc_bcbearer_send;
 786         sprintf(bcbearer->media.name, "tipc-broadcast");
 787
 788         INIT_LIST_HEAD(&bcl->waiting_ports);
 789         bcl->next_out_no = 1;
 790         spin_lock_init(&bclink->node.lock);
 791         bcl->owner = &bclink->node;
 792         bcl->max_pkt = MAX_PKT_DEFAULT_MCAST;
 793         tipc_link_set_queue_limits(bcl, BCLINK_WIN_DEFAULT);
 794         bcl->bearer_id = MAX_BEARERS;
 795         rcu_assign_pointer(bearer_list[MAX_BEARERS], &bcbearer->bearer);
 796         bcl->state = WORKING_WORKING;
 797         strlcpy(bcl->name, tipc_bclink_name, TIPC_MAX_LINK_NAME);
 798 }
 799
 800 void tipc_bclink_stop(void)
 801 {
 802         spin_lock_bh(&bc_lock);
 803         tipc_link_purge_queues(bcl);
 804         spin_unlock_bh(&bc_lock);
 805
 806         RCU_INIT_POINTER(bearer_list[BCBEARER], NULL);
 807         memset(bclink, 0, sizeof(*bclink));
 808         memset(bcbearer, 0, sizeof(*bcbearer));
 809 }
 810
 811 /**
 812  * tipc_nmap_add - add a node to a node map
 813  */
 814 static void tipc_nmap_add(struct tipc_node_map *nm_ptr, u32 node)
 815 {
 816         int n = tipc_node(node);
 817         int w = n / WSIZE;
 818         u32 mask = (1 << (n % WSIZE));
 819
 820         if ((nm_ptr->map[w] & mask) == 0) {
 821                 nm_ptr->count++;
 822                 nm_ptr->map[w] |= mask;
 823         }
 824 }
 825
 826 /**
 827  * tipc_nmap_remove - remove a node from a node map
 828  */
 829 static void tipc_nmap_remove(struct tipc_node_map *nm_ptr, u32 node)
 830 {
 831         int n = tipc_node(node);
 832         int w = n / WSIZE;
 833         u32 mask = (1 << (n % WSIZE));
 834
 835         if ((nm_ptr->map[w] & mask) != 0) {
 836                 nm_ptr->map[w] &= ~mask;
 837                 nm_ptr->count--;
 838         }
 839 }
 840
 841 /**
 842  * tipc_nmap_diff - find differences between node maps
 843  * @nm_a: input node map A
 844  * @nm_b: input node map B
 845  * @nm_diff: output node map A-B (i.e. nodes of A that are not in B)
 846  */
 847 static void tipc_nmap_diff(struct tipc_node_map *nm_a,
 848                            struct tipc_node_map *nm_b,
 849                            struct tipc_node_map *nm_diff)
 850 {
 851         int stop = ARRAY_SIZE(nm_a->map);
 852         int w;
 853         int b;
 854         u32 map;
 855
 856         memset(nm_diff, 0, sizeof(*nm_diff));
 857         for (w = 0; w < stop; w++) {
 858                 map = nm_a->map[w] ^ (nm_a->map[w] & nm_b->map[w]);
 859                 nm_diff->map[w] = map;
 860                 if (map != 0) {
 861                         for (b = 0 ; b < WSIZE; b++) {
 862                                 if (map & (1 << b))
 863                                         nm_diff->count++;
 864                         }
 865                 }
 866         }
 867 }
 868
 869 /**
 870  * tipc_port_list_add - add a port to a port list, ensuring no duplicates
 871  */
 872 void tipc_port_list_add(struct tipc_port_list *pl_ptr, u32 port)
 873 {
 874         struct tipc_port_list *item = pl_ptr;
 875         int i;
 876         int item_sz = PLSIZE;
 877         int cnt = pl_ptr->count;
 878
 879         for (; ; cnt -= item_sz, item = item->next) {
 880                 if (cnt < PLSIZE)
 881                         item_sz = cnt;
 882                 for (i = 0; i < item_sz; i++)
 883                         if (item->ports[i] == port)
 884                                 return;
 885                 if (i < PLSIZE) {
 886                         item->ports[i] = port;
 887                         pl_ptr->count++;
 888                         return;
 889                 }
 890                 if (!item->next) {
 891                         item->next = kmalloc(sizeof(*item), GFP_ATOMIC);
 892                         if (!item->next) {
 893                                 pr_warn("Incomplete multicast delivery, no memory\n");
 894                                 return;
 895                         }
 896                         item->next->next = NULL;
 897                 }
 898         }
 899 }
 900
 901 /**
 902  * tipc_port_list_free - free dynamically created entries in port_list chain
 903  *
 904  */
 905 void tipc_port_list_free(struct tipc_port_list *pl_ptr)
 906 {
 907         struct tipc_port_list *item;
 908         struct tipc_port_list *next;
 909
 910         for (item = pl_ptr->next; item; item = next) {
 911                 next = item->next;
 912                 kfree(item);
 913         }
 914 }