2 * net/tipc/bcast.c: TIPC broadcast code
4 * Copyright (c) 2004-2006, Ericsson AB
5 * Copyright (c) 2004, Intel Corporation.
6 * Copyright (c) 2005, Wind River Systems
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions are met:
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.
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.
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.
46 #include "node_subscr.h"
47 #include "name_distr.h"
49 #include "name_table.h"
53 #define MAX_PKT_DEFAULT_MCAST 1500 /* bcast link max packet size (fixed) */
55 #define BCLINK_WIN_DEFAULT 20 /* bcast link window size (default) */
57 #define BCLINK_LOG_BUF_SIZE 0
60 * struct bcbearer_pair - a pair of bearers used by broadcast link
61 * @primary: pointer to primary bearer
62 * @secondary: pointer to secondary bearer
64 * Bearers must have same priority and same set of reachable destinations
68 struct bcbearer_pair {
69 struct bearer *primary;
70 struct bearer *secondary;
74 * struct bcbearer - bearer used by broadcast link
75 * @bearer: (non-standard) broadcast bearer structure
76 * @media: (non-standard) broadcast media structure
77 * @bpairs: array of bearer pairs
78 * @bpairs_temp: array of bearer pairs used during creation of "bpairs"
84 struct bcbearer_pair bpairs[MAX_BEARERS];
85 struct bcbearer_pair bpairs_temp[TIPC_NUM_LINK_PRI];
89 * struct bclink - link used for broadcast messages
90 * @link: (non-standard) broadcast link structure
91 * @node: (non-standard) node structure representing b'cast link's peer node
93 * Handles sequence numbering, fragmentation, bundling, etc.
102 static struct bcbearer *bcbearer = NULL;
103 static struct bclink *bclink = NULL;
104 static struct link *bcl = NULL;
105 static spinlock_t bc_lock = SPIN_LOCK_UNLOCKED;
107 char bc_link_name[] = "multicast-link";
110 static inline u32 buf_seqno(struct sk_buff *buf)
112 return msg_seqno(buf_msg(buf));
115 static inline u32 bcbuf_acks(struct sk_buff *buf)
117 return (u32)(unsigned long)TIPC_SKB_CB(buf)->handle;
120 static inline void bcbuf_set_acks(struct sk_buff *buf, u32 acks)
122 TIPC_SKB_CB(buf)->handle = (void *)(unsigned long)acks;
125 static inline void bcbuf_decr_acks(struct sk_buff *buf)
127 bcbuf_set_acks(buf, bcbuf_acks(buf) - 1);
132 * bclink_set_gap - set gap according to contents of current deferred pkt queue
134 * Called with 'node' locked, bc_lock unlocked
137 static inline void bclink_set_gap(struct node *n_ptr)
139 struct sk_buff *buf = n_ptr->bclink.deferred_head;
141 n_ptr->bclink.gap_after = n_ptr->bclink.gap_to =
142 mod(n_ptr->bclink.last_in);
143 if (unlikely(buf != NULL))
144 n_ptr->bclink.gap_to = mod(buf_seqno(buf) - 1);
148 * bclink_ack_allowed - test if ACK or NACK message can be sent at this moment
150 * This mechanism endeavours to prevent all nodes in network from trying
151 * to ACK or NACK at the same time.
153 * Note: TIPC uses a different trigger to distribute ACKs than it does to
154 * distribute NACKs, but tries to use the same spacing (divide by 16).
157 static inline int bclink_ack_allowed(u32 n)
159 return((n % TIPC_MIN_LINK_WIN) == tipc_own_tag);
164 * bclink_retransmit_pkt - retransmit broadcast packets
165 * @after: sequence number of last packet to *not* retransmit
166 * @to: sequence number of last packet to retransmit
168 * Called with 'node' locked, bc_lock unlocked
171 static void bclink_retransmit_pkt(u32 after, u32 to)
175 spin_lock_bh(&bc_lock);
176 buf = bcl->first_out;
177 while (buf && less_eq(buf_seqno(buf), after)) {
181 link_retransmit(bcl, buf, mod(to - after));
182 spin_unlock_bh(&bc_lock);
186 * bclink_acknowledge - handle acknowledgement of broadcast packets
187 * @n_ptr: node that sent acknowledgement info
188 * @acked: broadcast sequence # that has been acknowledged
190 * Node is locked, bc_lock unlocked.
193 void bclink_acknowledge(struct node *n_ptr, u32 acked)
196 struct sk_buff *next;
197 unsigned int released = 0;
199 if (less_eq(acked, n_ptr->bclink.acked))
202 spin_lock_bh(&bc_lock);
204 /* Skip over packets that node has previously acknowledged */
206 crs = bcl->first_out;
207 while (crs && less_eq(buf_seqno(crs), n_ptr->bclink.acked)) {
211 /* Update packets that node is now acknowledging */
213 while (crs && less_eq(buf_seqno(crs), acked)) {
215 bcbuf_decr_acks(crs);
216 if (bcbuf_acks(crs) == 0) {
217 bcl->first_out = next;
218 bcl->out_queue_size--;
224 n_ptr->bclink.acked = acked;
226 /* Try resolving broadcast link congestion, if necessary */
228 if (unlikely(bcl->next_out))
229 link_push_queue(bcl);
230 if (unlikely(released && !list_empty(&bcl->waiting_ports)))
231 link_wakeup_ports(bcl, 0);
232 spin_unlock_bh(&bc_lock);
236 * bclink_send_ack - unicast an ACK msg
238 * net_lock and node lock set
241 static void bclink_send_ack(struct node *n_ptr)
243 struct link *l_ptr = n_ptr->active_links[n_ptr->addr & 1];
246 link_send_proto_msg(l_ptr, STATE_MSG, 0, 0, 0, 0, 0);
250 * bclink_send_nack- broadcast a NACK msg
252 * net_lock and node lock set
255 static void bclink_send_nack(struct node *n_ptr)
258 struct tipc_msg *msg;
260 if (!less(n_ptr->bclink.gap_after, n_ptr->bclink.gap_to))
263 buf = buf_acquire(INT_H_SIZE);
266 msg_init(msg, BCAST_PROTOCOL, STATE_MSG,
267 TIPC_OK, INT_H_SIZE, n_ptr->addr);
268 msg_set_mc_netid(msg, tipc_net_id);
269 msg_set_bcast_ack(msg, mod(n_ptr->bclink.last_in));
270 msg_set_bcgap_after(msg, n_ptr->bclink.gap_after);
271 msg_set_bcgap_to(msg, n_ptr->bclink.gap_to);
272 msg_set_bcast_tag(msg, tipc_own_tag);
274 if (bearer_send(&bcbearer->bearer, buf, 0)) {
275 bcl->stats.sent_nacks++;
278 bearer_schedule(bcl->b_ptr, bcl);
279 bcl->proto_msg_queue = buf;
280 bcl->stats.bearer_congs++;
284 * Ensure we doesn't send another NACK msg to the node
285 * until 16 more deferred messages arrive from it
286 * (i.e. helps prevent all nodes from NACK'ing at same time)
289 n_ptr->bclink.nack_sync = tipc_own_tag;
294 * bclink_check_gap - send a NACK if a sequence gap exists
296 * net_lock and node lock set
299 void bclink_check_gap(struct node *n_ptr, u32 last_sent)
301 if (!n_ptr->bclink.supported ||
302 less_eq(last_sent, mod(n_ptr->bclink.last_in)))
305 bclink_set_gap(n_ptr);
306 if (n_ptr->bclink.gap_after == n_ptr->bclink.gap_to)
307 n_ptr->bclink.gap_to = last_sent;
308 bclink_send_nack(n_ptr);
312 * bclink_peek_nack - process a NACK msg meant for another node
317 void bclink_peek_nack(u32 dest, u32 sender_tag, u32 gap_after, u32 gap_to)
319 struct node *n_ptr = node_find(dest);
322 if (unlikely(!n_ptr || !node_is_up(n_ptr)))
326 * Modify gap to suppress unnecessary NACKs from this node
328 my_after = n_ptr->bclink.gap_after;
329 my_to = n_ptr->bclink.gap_to;
331 if (less_eq(gap_after, my_after)) {
332 if (less(my_after, gap_to) && less(gap_to, my_to))
333 n_ptr->bclink.gap_after = gap_to;
334 else if (less_eq(my_to, gap_to))
335 n_ptr->bclink.gap_to = n_ptr->bclink.gap_after;
336 } else if (less_eq(gap_after, my_to)) {
337 if (less_eq(my_to, gap_to))
338 n_ptr->bclink.gap_to = gap_after;
341 * Expand gap if missing bufs not in deferred queue:
343 struct sk_buff *buf = n_ptr->bclink.deferred_head;
344 u32 prev = n_ptr->bclink.gap_to;
346 for (; buf; buf = buf->next) {
347 u32 seqno = buf_seqno(buf);
349 if (mod(seqno - prev) != 1)
351 if (seqno == gap_after)
356 n_ptr->bclink.gap_to = gap_after;
359 * Some nodes may send a complementary NACK now:
361 if (bclink_ack_allowed(sender_tag + 1)) {
362 if (n_ptr->bclink.gap_to != n_ptr->bclink.gap_after) {
363 bclink_send_nack(n_ptr);
364 bclink_set_gap(n_ptr);
371 * bclink_send_msg - broadcast a packet to all nodes in cluster
374 int bclink_send_msg(struct sk_buff *buf)
378 spin_lock_bh(&bc_lock);
380 res = link_send_buf(bcl, buf);
381 if (unlikely(res == -ELINKCONG))
384 bcl->stats.sent_info++;
386 if (bcl->out_queue_size > bcl->stats.max_queue_sz)
387 bcl->stats.max_queue_sz = bcl->out_queue_size;
388 bcl->stats.queue_sz_counts++;
389 bcl->stats.accu_queue_sz += bcl->out_queue_size;
391 spin_unlock_bh(&bc_lock);
396 * bclink_recv_pkt - receive a broadcast packet, and deliver upwards
398 * net_lock is read_locked, no other locks set
401 void bclink_recv_pkt(struct sk_buff *buf)
403 struct tipc_msg *msg = buf_msg(buf);
404 struct node* node = node_find(msg_prevnode(msg));
407 struct sk_buff *deferred;
409 msg_dbg(msg, "<BC<<<");
411 if (unlikely(!node || !node_is_up(node) || !node->bclink.supported ||
412 (msg_mc_netid(msg) != tipc_net_id))) {
417 if (unlikely(msg_user(msg) == BCAST_PROTOCOL)) {
418 msg_dbg(msg, "<BCNACK<<<");
419 if (msg_destnode(msg) == tipc_own_addr) {
421 bclink_acknowledge(node, msg_bcast_ack(msg));
423 bcl->stats.recv_nacks++;
424 bclink_retransmit_pkt(msg_bcgap_after(msg),
427 bclink_peek_nack(msg_destnode(msg),
429 msg_bcgap_after(msg),
438 deferred = node->bclink.deferred_head;
439 next_in = mod(node->bclink.last_in + 1);
440 seqno = msg_seqno(msg);
442 if (likely(seqno == next_in)) {
443 bcl->stats.recv_info++;
444 node->bclink.last_in++;
445 bclink_set_gap(node);
446 if (unlikely(bclink_ack_allowed(seqno))) {
447 bclink_send_ack(node);
448 bcl->stats.sent_acks++;
450 if (likely(msg_isdata(msg))) {
452 port_recv_mcast(buf, NULL);
453 } else if (msg_user(msg) == MSG_BUNDLER) {
454 bcl->stats.recv_bundles++;
455 bcl->stats.recv_bundled += msg_msgcnt(msg);
457 link_recv_bundle(buf);
458 } else if (msg_user(msg) == MSG_FRAGMENTER) {
459 bcl->stats.recv_fragments++;
460 if (link_recv_fragment(&node->bclink.defragm,
462 bcl->stats.recv_fragmented++;
469 if (deferred && (buf_seqno(deferred) == mod(next_in + 1))) {
473 node->bclink.deferred_head = deferred->next;
477 } else if (less(next_in, seqno)) {
478 u32 gap_after = node->bclink.gap_after;
479 u32 gap_to = node->bclink.gap_to;
481 if (link_defer_pkt(&node->bclink.deferred_head,
482 &node->bclink.deferred_tail,
484 node->bclink.nack_sync++;
485 bcl->stats.deferred_recv++;
486 if (seqno == mod(gap_after + 1))
487 node->bclink.gap_after = seqno;
488 else if (less(gap_after, seqno) && less(seqno, gap_to))
489 node->bclink.gap_to = seqno;
491 if (bclink_ack_allowed(node->bclink.nack_sync)) {
492 if (gap_to != gap_after)
493 bclink_send_nack(node);
494 bclink_set_gap(node);
497 bcl->stats.duplicates++;
503 u32 bclink_get_last_sent(void)
505 u32 last_sent = mod(bcl->next_out_no - 1);
508 last_sent = mod(buf_seqno(bcl->next_out) - 1);
512 u32 bclink_acks_missing(struct node *n_ptr)
514 return (n_ptr->bclink.supported &&
515 (bclink_get_last_sent() != n_ptr->bclink.acked));
520 * bcbearer_send - send a packet through the broadcast pseudo-bearer
522 * Send through as many bearers as necessary to reach all nodes
523 * that support TIPC multicasting.
525 * Returns 0 if packet sent successfully, non-zero if not
528 int bcbearer_send(struct sk_buff *buf,
529 struct tipc_bearer *unused1,
530 struct tipc_media_addr *unused2)
532 static int send_count = 0;
534 struct node_map remains;
535 struct node_map remains_new;
539 /* Prepare buffer for broadcasting (if first time trying to send it) */
541 if (likely(!msg_non_seq(buf_msg(buf)))) {
542 struct tipc_msg *msg;
544 assert(cluster_bcast_nodes.count != 0);
545 bcbuf_set_acks(buf, cluster_bcast_nodes.count);
547 msg_set_non_seq(msg);
548 msg_set_mc_netid(msg, tipc_net_id);
551 /* Determine if bearer pairs should be swapped following this attempt */
553 if ((swap_time = (++send_count >= 10)))
556 /* Send buffer over bearers until all targets reached */
558 remains = cluster_bcast_nodes;
560 for (bp_index = 0; bp_index < MAX_BEARERS; bp_index++) {
561 struct bearer *p = bcbearer->bpairs[bp_index].primary;
562 struct bearer *s = bcbearer->bpairs[bp_index].secondary;
565 break; /* no more bearers to try */
567 nmap_diff(&remains, &p->nodes, &remains_new);
568 if (remains_new.count == remains.count)
569 continue; /* bearer pair doesn't add anything */
571 if (!p->publ.blocked &&
572 !p->media->send_msg(buf, &p->publ, &p->media->bcast_addr)) {
573 if (swap_time && s && !s->publ.blocked)
579 if (!s || s->publ.blocked ||
580 s->media->send_msg(buf, &s->publ, &s->media->bcast_addr))
581 continue; /* unable to send using bearer pair */
583 bcbearer->bpairs[bp_index].primary = s;
584 bcbearer->bpairs[bp_index].secondary = p;
586 if (remains_new.count == 0)
589 remains = remains_new;
592 /* Unable to reach all targets */
594 bcbearer->bearer.publ.blocked = 1;
595 bcl->stats.bearer_congs++;
600 * bcbearer_sort - create sets of bearer pairs used by broadcast bearer
603 void bcbearer_sort(void)
605 struct bcbearer_pair *bp_temp = bcbearer->bpairs_temp;
606 struct bcbearer_pair *bp_curr;
610 spin_lock_bh(&bc_lock);
612 /* Group bearers by priority (can assume max of two per priority) */
614 memset(bp_temp, 0, sizeof(bcbearer->bpairs_temp));
616 for (b_index = 0; b_index < MAX_BEARERS; b_index++) {
617 struct bearer *b = &bearers[b_index];
619 if (!b->active || !b->nodes.count)
622 if (!bp_temp[b->priority].primary)
623 bp_temp[b->priority].primary = b;
625 bp_temp[b->priority].secondary = b;
628 /* Create array of bearer pairs for broadcasting */
630 bp_curr = bcbearer->bpairs;
631 memset(bcbearer->bpairs, 0, sizeof(bcbearer->bpairs));
633 for (pri = (TIPC_NUM_LINK_PRI - 1); pri >= 0; pri--) {
635 if (!bp_temp[pri].primary)
638 bp_curr->primary = bp_temp[pri].primary;
640 if (bp_temp[pri].secondary) {
641 if (nmap_equal(&bp_temp[pri].primary->nodes,
642 &bp_temp[pri].secondary->nodes)) {
643 bp_curr->secondary = bp_temp[pri].secondary;
646 bp_curr->primary = bp_temp[pri].secondary;
653 spin_unlock_bh(&bc_lock);
657 * bcbearer_push - resolve bearer congestion
659 * Forces bclink to push out any unsent packets, until all packets are gone
660 * or congestion reoccurs.
661 * No locks set when function called
664 void bcbearer_push(void)
666 struct bearer *b_ptr;
668 spin_lock_bh(&bc_lock);
669 b_ptr = &bcbearer->bearer;
670 if (b_ptr->publ.blocked) {
671 b_ptr->publ.blocked = 0;
672 bearer_lock_push(b_ptr);
674 spin_unlock_bh(&bc_lock);
678 int bclink_stats(char *buf, const u32 buf_size)
685 printbuf_init(&pb, buf, buf_size);
687 spin_lock_bh(&bc_lock);
689 tipc_printf(&pb, "Link <%s>\n"
690 " Window:%u packets\n",
691 bcl->name, bcl->queue_limit[0]);
692 tipc_printf(&pb, " RX packets:%u fragments:%u/%u bundles:%u/%u\n",
693 bcl->stats.recv_info,
694 bcl->stats.recv_fragments,
695 bcl->stats.recv_fragmented,
696 bcl->stats.recv_bundles,
697 bcl->stats.recv_bundled);
698 tipc_printf(&pb, " TX packets:%u fragments:%u/%u bundles:%u/%u\n",
699 bcl->stats.sent_info,
700 bcl->stats.sent_fragments,
701 bcl->stats.sent_fragmented,
702 bcl->stats.sent_bundles,
703 bcl->stats.sent_bundled);
704 tipc_printf(&pb, " RX naks:%u defs:%u dups:%u\n",
705 bcl->stats.recv_nacks,
706 bcl->stats.deferred_recv,
707 bcl->stats.duplicates);
708 tipc_printf(&pb, " TX naks:%u acks:%u dups:%u\n",
709 bcl->stats.sent_nacks,
710 bcl->stats.sent_acks,
711 bcl->stats.retransmitted);
712 tipc_printf(&pb, " Congestion bearer:%u link:%u Send queue max:%u avg:%u\n",
713 bcl->stats.bearer_congs,
714 bcl->stats.link_congs,
715 bcl->stats.max_queue_sz,
716 bcl->stats.queue_sz_counts
717 ? (bcl->stats.accu_queue_sz / bcl->stats.queue_sz_counts)
720 spin_unlock_bh(&bc_lock);
721 return printbuf_validate(&pb);
724 int bclink_reset_stats(void)
729 spin_lock_bh(&bc_lock);
730 memset(&bcl->stats, 0, sizeof(bcl->stats));
731 spin_unlock_bh(&bc_lock);
735 int bclink_set_queue_limits(u32 limit)
739 if ((limit < TIPC_MIN_LINK_WIN) || (limit > TIPC_MAX_LINK_WIN))
742 spin_lock_bh(&bc_lock);
743 link_set_queue_limits(bcl, limit);
744 spin_unlock_bh(&bc_lock);
748 int bclink_init(void)
750 bcbearer = kmalloc(sizeof(*bcbearer), GFP_ATOMIC);
751 bclink = kmalloc(sizeof(*bclink), GFP_ATOMIC);
752 if (!bcbearer || !bclink) {
754 warn("Memory squeeze; Failed to create multicast link\n");
762 memset(bcbearer, 0, sizeof(struct bcbearer));
763 INIT_LIST_HEAD(&bcbearer->bearer.cong_links);
764 bcbearer->bearer.media = &bcbearer->media;
765 bcbearer->media.send_msg = bcbearer_send;
766 sprintf(bcbearer->media.name, "tipc-multicast");
769 memset(bclink, 0, sizeof(struct bclink));
770 INIT_LIST_HEAD(&bcl->waiting_ports);
771 bcl->next_out_no = 1;
772 bclink->node.lock = SPIN_LOCK_UNLOCKED;
773 bcl->owner = &bclink->node;
774 bcl->max_pkt = MAX_PKT_DEFAULT_MCAST;
775 link_set_queue_limits(bcl, BCLINK_WIN_DEFAULT);
776 bcl->b_ptr = &bcbearer->bearer;
777 bcl->state = WORKING_WORKING;
778 sprintf(bcl->name, bc_link_name);
780 if (BCLINK_LOG_BUF_SIZE) {
781 char *pb = kmalloc(BCLINK_LOG_BUF_SIZE, GFP_ATOMIC);
785 printbuf_init(&bcl->print_buf, pb, BCLINK_LOG_BUF_SIZE);
791 void bclink_stop(void)
793 spin_lock_bh(&bc_lock);
796 if (BCLINK_LOG_BUF_SIZE)
797 kfree(bcl->print_buf.buf);
804 spin_unlock_bh(&bc_lock);