2 * net/tipc/link.c: TIPC link code
4 * Copyright (c) 2003-2005, Ericsson Research Canada
5 * Copyright (c) 2004-2005, Wind River Systems
6 * Copyright (c) 2005-2006, Ericsson AB
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.
45 #include "node_subscr.h"
46 #include "name_distr.h"
48 #include "name_table.h"
55 * Limit for deferred reception queue:
58 #define DEF_QUEUE_LIMIT 256u
64 #define STARTING_EVT 856384768 /* link processing trigger */
65 #define TRAFFIC_MSG_EVT 560815u /* rx'd ??? */
66 #define TIMEOUT_EVT 560817u /* link timer expired */
69 * The following two 'message types' is really just implementation
70 * data conveniently stored in the message header.
71 * They must not be considered part of the protocol
77 * State value stored in 'exp_msg_count'
80 #define START_CHANGEOVER 100000u
83 * struct link_name - deconstructed link name
84 * @addr_local: network address of node at this end
85 * @if_local: name of interface at this end
86 * @addr_peer: network address of node at far end
87 * @if_peer: name of interface at far end
92 char if_local[TIPC_MAX_IF_NAME];
94 char if_peer[TIPC_MAX_IF_NAME];
99 /* LINK EVENT CODE IS NOT SUPPORTED AT PRESENT */
102 * struct link_event - link up/down event notification
108 void (*fcn)(u32, char *, int);
109 char name[TIPC_MAX_LINK_NAME];
114 static void link_handle_out_of_seq_msg(struct link *l_ptr,
115 struct sk_buff *buf);
116 static void link_recv_proto_msg(struct link *l_ptr, struct sk_buff *buf);
117 static int link_recv_changeover_msg(struct link **l_ptr, struct sk_buff **buf);
118 static void link_set_supervision_props(struct link *l_ptr, u32 tolerance);
119 static int link_send_sections_long(struct port *sender,
120 struct iovec const *msg_sect,
121 u32 num_sect, u32 destnode);
122 static void link_check_defragm_bufs(struct link *l_ptr);
123 static void link_state_event(struct link *l_ptr, u32 event);
124 static void link_reset_statistics(struct link *l_ptr);
125 static void link_print(struct link *l_ptr, struct print_buf *buf,
129 * Debugging code used by link routines only
131 * When debugging link problems on a system that has multiple links,
132 * the standard TIPC debugging routines may not be useful since they
133 * allow the output from multiple links to be intermixed. For this reason
134 * routines of the form "dbg_link_XXX()" have been created that will capture
135 * debug info into a link's personal print buffer, which can then be dumped
136 * into the TIPC system log (LOG) upon request.
138 * To enable per-link debugging, use LINK_LOG_BUF_SIZE to specify the size
139 * of the print buffer used by each link. If LINK_LOG_BUF_SIZE is set to 0,
140 * the dbg_link_XXX() routines simply send their output to the standard
141 * debug print buffer (DBG_OUTPUT), if it has been defined; this can be useful
142 * when there is only a single link in the system being debugged.
145 * - When enabled, LINK_LOG_BUF_SIZE should be set to at least 1000 (bytes)
146 * - "l_ptr" must be valid when using dbg_link_XXX() macros
149 #define LINK_LOG_BUF_SIZE 0
151 #define dbg_link(fmt, arg...) do {if (LINK_LOG_BUF_SIZE) tipc_printf(&l_ptr->print_buf, fmt, ## arg); } while(0)
152 #define dbg_link_msg(msg, txt) do {if (LINK_LOG_BUF_SIZE) msg_print(&l_ptr->print_buf, msg, txt); } while(0)
153 #define dbg_link_state(txt) do {if (LINK_LOG_BUF_SIZE) link_print(l_ptr, &l_ptr->print_buf, txt); } while(0)
154 #define dbg_link_dump() do { \
155 if (LINK_LOG_BUF_SIZE) { \
156 tipc_printf(LOG, "\n\nDumping link <%s>:\n", l_ptr->name); \
157 printbuf_move(LOG, &l_ptr->print_buf); \
161 static inline void dbg_print_link(struct link *l_ptr, const char *str)
164 link_print(l_ptr, DBG_OUTPUT, str);
167 static inline void dbg_print_buf_chain(struct sk_buff *root_buf)
170 struct sk_buff *buf = root_buf;
173 msg_dbg(buf_msg(buf), "In chain: ");
180 * Simple inlined link routines
183 static inline unsigned int align(unsigned int i)
185 return (i + 3) & ~3u;
188 static inline int link_working_working(struct link *l_ptr)
190 return (l_ptr->state == WORKING_WORKING);
193 static inline int link_working_unknown(struct link *l_ptr)
195 return (l_ptr->state == WORKING_UNKNOWN);
198 static inline int link_reset_unknown(struct link *l_ptr)
200 return (l_ptr->state == RESET_UNKNOWN);
203 static inline int link_reset_reset(struct link *l_ptr)
205 return (l_ptr->state == RESET_RESET);
208 static inline int link_blocked(struct link *l_ptr)
210 return (l_ptr->exp_msg_count || l_ptr->blocked);
213 static inline int link_congested(struct link *l_ptr)
215 return (l_ptr->out_queue_size >= l_ptr->queue_limit[0]);
218 static inline u32 link_max_pkt(struct link *l_ptr)
220 return l_ptr->max_pkt;
223 static inline void link_init_max_pkt(struct link *l_ptr)
227 max_pkt = (l_ptr->b_ptr->publ.mtu & ~3);
228 if (max_pkt > MAX_MSG_SIZE)
229 max_pkt = MAX_MSG_SIZE;
231 l_ptr->max_pkt_target = max_pkt;
232 if (l_ptr->max_pkt_target < MAX_PKT_DEFAULT)
233 l_ptr->max_pkt = l_ptr->max_pkt_target;
235 l_ptr->max_pkt = MAX_PKT_DEFAULT;
237 l_ptr->max_pkt_probes = 0;
240 static inline u32 link_next_sent(struct link *l_ptr)
243 return msg_seqno(buf_msg(l_ptr->next_out));
244 return mod(l_ptr->next_out_no);
247 static inline u32 link_last_sent(struct link *l_ptr)
249 return mod(link_next_sent(l_ptr) - 1);
253 * Simple non-inlined link routines (i.e. referenced outside this file)
256 int link_is_up(struct link *l_ptr)
260 return (link_working_working(l_ptr) || link_working_unknown(l_ptr));
263 int link_is_active(struct link *l_ptr)
265 return ((l_ptr->owner->active_links[0] == l_ptr) ||
266 (l_ptr->owner->active_links[1] == l_ptr));
270 * link_name_validate - validate & (optionally) deconstruct link name
271 * @name - ptr to link name string
272 * @name_parts - ptr to area for link name components (or NULL if not needed)
274 * Returns 1 if link name is valid, otherwise 0.
277 static int link_name_validate(const char *name, struct link_name *name_parts)
279 char name_copy[TIPC_MAX_LINK_NAME];
285 u32 z_local, c_local, n_local;
286 u32 z_peer, c_peer, n_peer;
290 /* copy link name & ensure length is OK */
292 name_copy[TIPC_MAX_LINK_NAME - 1] = 0;
293 /* need above in case non-Posix strncpy() doesn't pad with nulls */
294 strncpy(name_copy, name, TIPC_MAX_LINK_NAME);
295 if (name_copy[TIPC_MAX_LINK_NAME - 1] != 0)
298 /* ensure all component parts of link name are present */
300 addr_local = name_copy;
301 if ((if_local = strchr(addr_local, ':')) == NULL)
304 if ((addr_peer = strchr(if_local, '-')) == NULL)
307 if_local_len = addr_peer - if_local;
308 if ((if_peer = strchr(addr_peer, ':')) == NULL)
311 if_peer_len = strlen(if_peer) + 1;
313 /* validate component parts of link name */
315 if ((sscanf(addr_local, "%u.%u.%u%c",
316 &z_local, &c_local, &n_local, &dummy) != 3) ||
317 (sscanf(addr_peer, "%u.%u.%u%c",
318 &z_peer, &c_peer, &n_peer, &dummy) != 3) ||
319 (z_local > 255) || (c_local > 4095) || (n_local > 4095) ||
320 (z_peer > 255) || (c_peer > 4095) || (n_peer > 4095) ||
321 (if_local_len <= 1) || (if_local_len > TIPC_MAX_IF_NAME) ||
322 (if_peer_len <= 1) || (if_peer_len > TIPC_MAX_IF_NAME) ||
323 (strspn(if_local, tipc_alphabet) != (if_local_len - 1)) ||
324 (strspn(if_peer, tipc_alphabet) != (if_peer_len - 1)))
327 /* return link name components, if necessary */
330 name_parts->addr_local = tipc_addr(z_local, c_local, n_local);
331 strcpy(name_parts->if_local, if_local);
332 name_parts->addr_peer = tipc_addr(z_peer, c_peer, n_peer);
333 strcpy(name_parts->if_peer, if_peer);
339 * link_timeout - handle expiration of link timer
340 * @l_ptr: pointer to link
342 * This routine must not grab "net_lock" to avoid a potential deadlock conflict
343 * with link_delete(). (There is no risk that the node will be deleted by
344 * another thread because link_delete() always cancels the link timer before
345 * node_delete() is called.)
348 static void link_timeout(struct link *l_ptr)
350 node_lock(l_ptr->owner);
352 /* update counters used in statistical profiling of send traffic */
354 l_ptr->stats.accu_queue_sz += l_ptr->out_queue_size;
355 l_ptr->stats.queue_sz_counts++;
357 if (l_ptr->out_queue_size > l_ptr->stats.max_queue_sz)
358 l_ptr->stats.max_queue_sz = l_ptr->out_queue_size;
360 if (l_ptr->first_out) {
361 struct tipc_msg *msg = buf_msg(l_ptr->first_out);
362 u32 length = msg_size(msg);
364 if ((msg_user(msg) == MSG_FRAGMENTER)
365 && (msg_type(msg) == FIRST_FRAGMENT)) {
366 length = msg_size(msg_get_wrapped(msg));
369 l_ptr->stats.msg_lengths_total += length;
370 l_ptr->stats.msg_length_counts++;
372 l_ptr->stats.msg_length_profile[0]++;
373 else if (length <= 256)
374 l_ptr->stats.msg_length_profile[1]++;
375 else if (length <= 1024)
376 l_ptr->stats.msg_length_profile[2]++;
377 else if (length <= 4096)
378 l_ptr->stats.msg_length_profile[3]++;
379 else if (length <= 16384)
380 l_ptr->stats.msg_length_profile[4]++;
381 else if (length <= 32768)
382 l_ptr->stats.msg_length_profile[5]++;
384 l_ptr->stats.msg_length_profile[6]++;
388 /* do all other link processing performed on a periodic basis */
390 link_check_defragm_bufs(l_ptr);
392 link_state_event(l_ptr, TIMEOUT_EVT);
395 link_push_queue(l_ptr);
397 node_unlock(l_ptr->owner);
400 static inline void link_set_timer(struct link *l_ptr, u32 time)
402 k_start_timer(&l_ptr->timer, time);
406 * link_create - create a new link
407 * @b_ptr: pointer to associated bearer
408 * @peer: network address of node at other end of link
409 * @media_addr: media address to use when sending messages over link
411 * Returns pointer to link.
414 struct link *link_create(struct bearer *b_ptr, const u32 peer,
415 const struct tipc_media_addr *media_addr)
418 struct tipc_msg *msg;
421 l_ptr = (struct link *)kmalloc(sizeof(*l_ptr), GFP_ATOMIC);
423 warn("Memory squeeze; Failed to create link\n");
426 memset(l_ptr, 0, sizeof(*l_ptr));
429 if_name = strchr(b_ptr->publ.name, ':') + 1;
430 sprintf(l_ptr->name, "%u.%u.%u:%s-%u.%u.%u:",
431 tipc_zone(tipc_own_addr), tipc_cluster(tipc_own_addr),
432 tipc_node(tipc_own_addr),
434 tipc_zone(peer), tipc_cluster(peer), tipc_node(peer));
435 /* note: peer i/f is appended to link name by reset/activate */
436 memcpy(&l_ptr->media_addr, media_addr, sizeof(*media_addr));
437 k_init_timer(&l_ptr->timer, (Handler)link_timeout, (unsigned long)l_ptr);
438 list_add_tail(&l_ptr->link_list, &b_ptr->links);
439 l_ptr->checkpoint = 1;
440 l_ptr->b_ptr = b_ptr;
441 link_set_supervision_props(l_ptr, b_ptr->media->tolerance);
442 l_ptr->state = RESET_UNKNOWN;
444 l_ptr->pmsg = (struct tipc_msg *)&l_ptr->proto_msg;
446 msg_init(msg, LINK_PROTOCOL, RESET_MSG, TIPC_OK, INT_H_SIZE, l_ptr->addr);
447 msg_set_size(msg, sizeof(l_ptr->proto_msg));
448 msg_set_session(msg, tipc_random);
449 msg_set_bearer_id(msg, b_ptr->identity);
450 strcpy((char *)msg_data(msg), if_name);
452 l_ptr->priority = b_ptr->priority;
453 link_set_queue_limits(l_ptr, b_ptr->media->window);
455 link_init_max_pkt(l_ptr);
457 l_ptr->next_out_no = 1;
458 INIT_LIST_HEAD(&l_ptr->waiting_ports);
460 link_reset_statistics(l_ptr);
462 l_ptr->owner = node_attach_link(l_ptr);
468 if (LINK_LOG_BUF_SIZE) {
469 char *pb = kmalloc(LINK_LOG_BUF_SIZE, GFP_ATOMIC);
473 warn("Memory squeeze; Failed to create link\n");
476 printbuf_init(&l_ptr->print_buf, pb, LINK_LOG_BUF_SIZE);
479 k_signal((Handler)link_start, (unsigned long)l_ptr);
481 dbg("link_create(): tolerance = %u,cont intv = %u, abort_limit = %u\n",
482 l_ptr->tolerance, l_ptr->continuity_interval, l_ptr->abort_limit);
488 * link_delete - delete a link
489 * @l_ptr: pointer to link
491 * Note: 'net_lock' is write_locked, bearer is locked.
492 * This routine must not grab the node lock until after link timer cancellation
493 * to avoid a potential deadlock situation.
496 void link_delete(struct link *l_ptr)
499 err("Attempt to delete non-existent link\n");
503 dbg("link_delete()\n");
505 k_cancel_timer(&l_ptr->timer);
507 node_lock(l_ptr->owner);
509 node_detach_link(l_ptr->owner, l_ptr);
511 list_del_init(&l_ptr->link_list);
512 if (LINK_LOG_BUF_SIZE)
513 kfree(l_ptr->print_buf.buf);
514 node_unlock(l_ptr->owner);
515 k_term_timer(&l_ptr->timer);
519 void link_start(struct link *l_ptr)
521 dbg("link_start %x\n", l_ptr);
522 link_state_event(l_ptr, STARTING_EVT);
526 * link_schedule_port - schedule port for deferred sending
527 * @l_ptr: pointer to link
528 * @origport: reference to sending port
529 * @sz: amount of data to be sent
531 * Schedules port for renewed sending of messages after link congestion
535 static int link_schedule_port(struct link *l_ptr, u32 origport, u32 sz)
539 spin_lock_bh(&port_list_lock);
540 p_ptr = port_lock(origport);
544 if (!list_empty(&p_ptr->wait_list))
546 p_ptr->congested_link = l_ptr;
547 p_ptr->publ.congested = 1;
548 p_ptr->waiting_pkts = 1 + ((sz - 1) / link_max_pkt(l_ptr));
549 list_add_tail(&p_ptr->wait_list, &l_ptr->waiting_ports);
550 l_ptr->stats.link_congs++;
554 spin_unlock_bh(&port_list_lock);
558 void link_wakeup_ports(struct link *l_ptr, int all)
561 struct port *temp_p_ptr;
562 int win = l_ptr->queue_limit[0] - l_ptr->out_queue_size;
568 if (!spin_trylock_bh(&port_list_lock))
570 if (link_congested(l_ptr))
572 list_for_each_entry_safe(p_ptr, temp_p_ptr, &l_ptr->waiting_ports,
576 list_del_init(&p_ptr->wait_list);
577 p_ptr->congested_link = 0;
578 assert(p_ptr->wakeup);
579 spin_lock_bh(p_ptr->publ.lock);
580 p_ptr->publ.congested = 0;
581 p_ptr->wakeup(&p_ptr->publ);
582 win -= p_ptr->waiting_pkts;
583 spin_unlock_bh(p_ptr->publ.lock);
587 spin_unlock_bh(&port_list_lock);
591 * link_release_outqueue - purge link's outbound message queue
592 * @l_ptr: pointer to link
595 static void link_release_outqueue(struct link *l_ptr)
597 struct sk_buff *buf = l_ptr->first_out;
598 struct sk_buff *next;
605 l_ptr->first_out = NULL;
606 l_ptr->out_queue_size = 0;
610 * link_reset_fragments - purge link's inbound message fragments queue
611 * @l_ptr: pointer to link
614 void link_reset_fragments(struct link *l_ptr)
616 struct sk_buff *buf = l_ptr->defragm_buf;
617 struct sk_buff *next;
624 l_ptr->defragm_buf = NULL;
628 * link_stop - purge all inbound and outbound messages associated with link
629 * @l_ptr: pointer to link
632 void link_stop(struct link *l_ptr)
635 struct sk_buff *next;
637 buf = l_ptr->oldest_deferred_in;
644 buf = l_ptr->first_out;
651 link_reset_fragments(l_ptr);
653 buf_discard(l_ptr->proto_msg_queue);
654 l_ptr->proto_msg_queue = NULL;
659 /* LINK EVENT CODE IS NOT SUPPORTED AT PRESENT */
661 static void link_recv_event(struct link_event *ev)
663 ev->fcn(ev->addr, ev->name, ev->up);
667 static void link_send_event(void (*fcn)(u32 a, char *n, int up),
668 struct link *l_ptr, int up)
670 struct link_event *ev;
672 ev = kmalloc(sizeof(*ev), GFP_ATOMIC);
674 warn("Link event allocation failure\n");
677 ev->addr = l_ptr->addr;
680 memcpy(ev->name, l_ptr->name, TIPC_MAX_LINK_NAME);
681 k_signal((Handler)link_recv_event, (unsigned long)ev);
686 #define link_send_event(fcn, l_ptr, up) do { } while (0)
690 void link_reset(struct link *l_ptr)
693 u32 prev_state = l_ptr->state;
694 u32 checkpoint = l_ptr->next_in_no;
696 msg_set_session(l_ptr->pmsg, msg_session(l_ptr->pmsg) + 1);
698 /* Link is down, accept any session: */
699 l_ptr->peer_session = 0;
701 /* Prepare for max packet size negotiation */
702 link_init_max_pkt(l_ptr);
704 l_ptr->state = RESET_UNKNOWN;
705 dbg_link_state("Resetting Link\n");
707 if ((prev_state == RESET_UNKNOWN) || (prev_state == RESET_RESET))
710 node_link_down(l_ptr->owner, l_ptr);
711 bearer_remove_dest(l_ptr->b_ptr, l_ptr->addr);
713 tipc_printf(CONS, "\nReset link <%s>\n", l_ptr->name);
716 if (node_has_active_links(l_ptr->owner) &&
717 l_ptr->owner->permit_changeover) {
718 l_ptr->reset_checkpoint = checkpoint;
719 l_ptr->exp_msg_count = START_CHANGEOVER;
722 /* Clean up all queues: */
724 link_release_outqueue(l_ptr);
725 buf_discard(l_ptr->proto_msg_queue);
726 l_ptr->proto_msg_queue = NULL;
727 buf = l_ptr->oldest_deferred_in;
729 struct sk_buff *next = buf->next;
733 if (!list_empty(&l_ptr->waiting_ports))
734 link_wakeup_ports(l_ptr, 1);
736 l_ptr->retransm_queue_head = 0;
737 l_ptr->retransm_queue_size = 0;
738 l_ptr->last_out = NULL;
739 l_ptr->first_out = NULL;
740 l_ptr->next_out = NULL;
741 l_ptr->unacked_window = 0;
742 l_ptr->checkpoint = 1;
743 l_ptr->next_out_no = 1;
744 l_ptr->deferred_inqueue_sz = 0;
745 l_ptr->oldest_deferred_in = NULL;
746 l_ptr->newest_deferred_in = NULL;
747 l_ptr->fsm_msg_cnt = 0;
748 l_ptr->stale_count = 0;
749 link_reset_statistics(l_ptr);
751 link_send_event(cfg_link_event, l_ptr, 0);
752 if (!in_own_cluster(l_ptr->addr))
753 link_send_event(disc_link_event, l_ptr, 0);
757 static void link_activate(struct link *l_ptr)
759 l_ptr->next_in_no = 1;
760 node_link_up(l_ptr->owner, l_ptr);
761 bearer_add_dest(l_ptr->b_ptr, l_ptr->addr);
762 link_send_event(cfg_link_event, l_ptr, 1);
763 if (!in_own_cluster(l_ptr->addr))
764 link_send_event(disc_link_event, l_ptr, 1);
768 * link_state_event - link finite state machine
769 * @l_ptr: pointer to link
770 * @event: state machine event to process
773 static void link_state_event(struct link *l_ptr, unsigned event)
776 u32 cont_intv = l_ptr->continuity_interval;
778 if (!l_ptr->started && (event != STARTING_EVT))
779 return; /* Not yet. */
781 if (link_blocked(l_ptr)) {
782 if (event == TIMEOUT_EVT) {
783 link_set_timer(l_ptr, cont_intv);
785 return; /* Changeover going on */
787 dbg_link("STATE_EV: <%s> ", l_ptr->name);
789 switch (l_ptr->state) {
790 case WORKING_WORKING:
793 case TRAFFIC_MSG_EVT:
801 if (l_ptr->next_in_no != l_ptr->checkpoint) {
802 l_ptr->checkpoint = l_ptr->next_in_no;
803 if (bclink_acks_missing(l_ptr->owner)) {
804 link_send_proto_msg(l_ptr, STATE_MSG,
806 l_ptr->fsm_msg_cnt++;
807 } else if (l_ptr->max_pkt < l_ptr->max_pkt_target) {
808 link_send_proto_msg(l_ptr, STATE_MSG,
810 l_ptr->fsm_msg_cnt++;
812 link_set_timer(l_ptr, cont_intv);
815 dbg_link(" -> WU\n");
816 l_ptr->state = WORKING_UNKNOWN;
817 l_ptr->fsm_msg_cnt = 0;
818 link_send_proto_msg(l_ptr, STATE_MSG, 1, 0, 0, 0, 0);
819 l_ptr->fsm_msg_cnt++;
820 link_set_timer(l_ptr, cont_intv / 4);
823 dbg_link("RES -> RR\n");
825 l_ptr->state = RESET_RESET;
826 l_ptr->fsm_msg_cnt = 0;
827 link_send_proto_msg(l_ptr, ACTIVATE_MSG, 0, 0, 0, 0, 0);
828 l_ptr->fsm_msg_cnt++;
829 link_set_timer(l_ptr, cont_intv);
832 err("Unknown link event %u in WW state\n", event);
835 case WORKING_UNKNOWN:
838 case TRAFFIC_MSG_EVT:
841 dbg_link("ACT -> WW\n");
842 l_ptr->state = WORKING_WORKING;
843 l_ptr->fsm_msg_cnt = 0;
844 link_set_timer(l_ptr, cont_intv);
847 dbg_link("RES -> RR\n");
849 l_ptr->state = RESET_RESET;
850 l_ptr->fsm_msg_cnt = 0;
851 link_send_proto_msg(l_ptr, ACTIVATE_MSG, 0, 0, 0, 0, 0);
852 l_ptr->fsm_msg_cnt++;
853 link_set_timer(l_ptr, cont_intv);
857 if (l_ptr->next_in_no != l_ptr->checkpoint) {
858 dbg_link("-> WW \n");
859 l_ptr->state = WORKING_WORKING;
860 l_ptr->fsm_msg_cnt = 0;
861 l_ptr->checkpoint = l_ptr->next_in_no;
862 if (bclink_acks_missing(l_ptr->owner)) {
863 link_send_proto_msg(l_ptr, STATE_MSG,
865 l_ptr->fsm_msg_cnt++;
867 link_set_timer(l_ptr, cont_intv);
868 } else if (l_ptr->fsm_msg_cnt < l_ptr->abort_limit) {
869 dbg_link("Probing %u/%u,timer = %u ms)\n",
870 l_ptr->fsm_msg_cnt, l_ptr->abort_limit,
872 link_send_proto_msg(l_ptr, STATE_MSG,
874 l_ptr->fsm_msg_cnt++;
875 link_set_timer(l_ptr, cont_intv / 4);
876 } else { /* Link has failed */
877 dbg_link("-> RU (%u probes unanswered)\n",
880 l_ptr->state = RESET_UNKNOWN;
881 l_ptr->fsm_msg_cnt = 0;
882 link_send_proto_msg(l_ptr, RESET_MSG,
884 l_ptr->fsm_msg_cnt++;
885 link_set_timer(l_ptr, cont_intv);
889 err("Unknown link event %u in WU state\n", event);
895 case TRAFFIC_MSG_EVT:
899 other = l_ptr->owner->active_links[0];
900 if (other && link_working_unknown(other)) {
904 dbg_link("ACT -> WW\n");
905 l_ptr->state = WORKING_WORKING;
906 l_ptr->fsm_msg_cnt = 0;
907 link_activate(l_ptr);
908 link_send_proto_msg(l_ptr, STATE_MSG, 1, 0, 0, 0, 0);
909 l_ptr->fsm_msg_cnt++;
910 link_set_timer(l_ptr, cont_intv);
914 dbg_link(" -> RR\n");
915 l_ptr->state = RESET_RESET;
916 l_ptr->fsm_msg_cnt = 0;
917 link_send_proto_msg(l_ptr, ACTIVATE_MSG, 1, 0, 0, 0, 0);
918 l_ptr->fsm_msg_cnt++;
919 link_set_timer(l_ptr, cont_intv);
927 link_send_proto_msg(l_ptr, RESET_MSG, 0, 0, 0, 0, 0);
928 l_ptr->fsm_msg_cnt++;
929 link_set_timer(l_ptr, cont_intv);
932 err("Unknown link event %u in RU state\n", event);
938 case TRAFFIC_MSG_EVT:
942 other = l_ptr->owner->active_links[0];
943 if (other && link_working_unknown(other)) {
947 dbg_link("ACT -> WW\n");
948 l_ptr->state = WORKING_WORKING;
949 l_ptr->fsm_msg_cnt = 0;
950 link_activate(l_ptr);
951 link_send_proto_msg(l_ptr, STATE_MSG, 1, 0, 0, 0, 0);
952 l_ptr->fsm_msg_cnt++;
953 link_set_timer(l_ptr, cont_intv);
960 link_send_proto_msg(l_ptr, ACTIVATE_MSG, 0, 0, 0, 0, 0);
961 l_ptr->fsm_msg_cnt++;
962 link_set_timer(l_ptr, cont_intv);
963 dbg_link("fsm_msg_cnt %u\n", l_ptr->fsm_msg_cnt);
966 err("Unknown link event %u in RR state\n", event);
970 err("Unknown link state %u/%u\n", l_ptr->state, event);
975 * link_bundle_buf(): Append contents of a buffer to
976 * the tail of an existing one.
979 static int link_bundle_buf(struct link *l_ptr,
980 struct sk_buff *bundler,
983 struct tipc_msg *bundler_msg = buf_msg(bundler);
984 struct tipc_msg *msg = buf_msg(buf);
985 u32 size = msg_size(msg);
986 u32 to_pos = align(msg_size(bundler_msg));
987 u32 rest = link_max_pkt(l_ptr) - to_pos;
989 if (msg_user(bundler_msg) != MSG_BUNDLER)
991 if (msg_type(bundler_msg) != OPEN_MSG)
993 if (rest < align(size))
996 skb_put(bundler, (to_pos - msg_size(bundler_msg)) + size);
997 memcpy(bundler->data + to_pos, buf->data, size);
998 msg_set_size(bundler_msg, to_pos + size);
999 msg_set_msgcnt(bundler_msg, msg_msgcnt(bundler_msg) + 1);
1000 dbg("Packed msg # %u(%u octets) into pos %u in buf(#%u)\n",
1001 msg_msgcnt(bundler_msg), size, to_pos, msg_seqno(bundler_msg));
1002 msg_dbg(msg, "PACKD:");
1004 l_ptr->stats.sent_bundled++;
1008 static inline void link_add_to_outqueue(struct link *l_ptr,
1009 struct sk_buff *buf,
1010 struct tipc_msg *msg)
1012 u32 ack = mod(l_ptr->next_in_no - 1);
1013 u32 seqno = mod(l_ptr->next_out_no++);
1015 msg_set_word(msg, 2, ((ack << 16) | seqno));
1016 msg_set_bcast_ack(msg, l_ptr->owner->bclink.last_in);
1018 if (l_ptr->first_out) {
1019 l_ptr->last_out->next = buf;
1020 l_ptr->last_out = buf;
1022 l_ptr->first_out = l_ptr->last_out = buf;
1023 l_ptr->out_queue_size++;
1027 * link_send_buf() is the 'full path' for messages, called from
1028 * inside TIPC when the 'fast path' in tipc_send_buf
1029 * has failed, and from link_send()
1032 int link_send_buf(struct link *l_ptr, struct sk_buff *buf)
1034 struct tipc_msg *msg = buf_msg(buf);
1035 u32 size = msg_size(msg);
1036 u32 dsz = msg_data_sz(msg);
1037 u32 queue_size = l_ptr->out_queue_size;
1038 u32 imp = msg_tot_importance(msg);
1039 u32 queue_limit = l_ptr->queue_limit[imp];
1040 u32 max_packet = link_max_pkt(l_ptr);
1042 msg_set_prevnode(msg, tipc_own_addr); /* If routed message */
1044 /* Match msg importance against queue limits: */
1046 if (unlikely(queue_size >= queue_limit)) {
1047 if (imp <= TIPC_CRITICAL_IMPORTANCE) {
1048 return link_schedule_port(l_ptr, msg_origport(msg),
1051 msg_dbg(msg, "TIPC: Congestion, throwing away\n");
1053 if (imp > CONN_MANAGER) {
1054 warn("Resetting <%s>, send queue full", l_ptr->name);
1060 /* Fragmentation needed ? */
1062 if (size > max_packet)
1063 return link_send_long_buf(l_ptr, buf);
1065 /* Packet can be queued or sent: */
1067 if (queue_size > l_ptr->stats.max_queue_sz)
1068 l_ptr->stats.max_queue_sz = queue_size;
1070 if (likely(!bearer_congested(l_ptr->b_ptr, l_ptr) &&
1071 !link_congested(l_ptr))) {
1072 link_add_to_outqueue(l_ptr, buf, msg);
1074 if (likely(bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr))) {
1075 l_ptr->unacked_window = 0;
1077 bearer_schedule(l_ptr->b_ptr, l_ptr);
1078 l_ptr->stats.bearer_congs++;
1079 l_ptr->next_out = buf;
1083 /* Congestion: can message be bundled ?: */
1085 if ((msg_user(msg) != CHANGEOVER_PROTOCOL) &&
1086 (msg_user(msg) != MSG_FRAGMENTER)) {
1088 /* Try adding message to an existing bundle */
1090 if (l_ptr->next_out &&
1091 link_bundle_buf(l_ptr, l_ptr->last_out, buf)) {
1092 bearer_resolve_congestion(l_ptr->b_ptr, l_ptr);
1096 /* Try creating a new bundle */
1098 if (size <= max_packet * 2 / 3) {
1099 struct sk_buff *bundler = buf_acquire(max_packet);
1100 struct tipc_msg bundler_hdr;
1103 msg_init(&bundler_hdr, MSG_BUNDLER, OPEN_MSG,
1104 TIPC_OK, INT_H_SIZE, l_ptr->addr);
1105 memcpy(bundler->data, (unchar *)&bundler_hdr,
1107 skb_trim(bundler, INT_H_SIZE);
1108 link_bundle_buf(l_ptr, bundler, buf);
1111 l_ptr->stats.sent_bundles++;
1115 if (!l_ptr->next_out)
1116 l_ptr->next_out = buf;
1117 link_add_to_outqueue(l_ptr, buf, msg);
1118 bearer_resolve_congestion(l_ptr->b_ptr, l_ptr);
1123 * link_send(): same as link_send_buf(), but the link to use has
1124 * not been selected yet, and the the owner node is not locked
1125 * Called by TIPC internal users, e.g. the name distributor
1128 int link_send(struct sk_buff *buf, u32 dest, u32 selector)
1132 int res = -ELINKCONG;
1134 read_lock_bh(&net_lock);
1135 n_ptr = node_select(dest, selector);
1138 l_ptr = n_ptr->active_links[selector & 1];
1139 dbg("link_send: found link %x for dest %x\n", l_ptr, dest);
1141 res = link_send_buf(l_ptr, buf);
1145 dbg("Attempt to send msg to unknown node:\n");
1146 msg_dbg(buf_msg(buf),">>>");
1149 read_unlock_bh(&net_lock);
1154 * link_send_buf_fast: Entry for data messages where the
1155 * destination link is known and the header is complete,
1156 * inclusive total message length. Very time critical.
1157 * Link is locked. Returns user data length.
1160 static inline int link_send_buf_fast(struct link *l_ptr, struct sk_buff *buf,
1163 struct tipc_msg *msg = buf_msg(buf);
1164 int res = msg_data_sz(msg);
1166 if (likely(!link_congested(l_ptr))) {
1167 if (likely(msg_size(msg) <= link_max_pkt(l_ptr))) {
1168 if (likely(list_empty(&l_ptr->b_ptr->cong_links))) {
1169 link_add_to_outqueue(l_ptr, buf, msg);
1170 if (likely(bearer_send(l_ptr->b_ptr, buf,
1171 &l_ptr->media_addr))) {
1172 l_ptr->unacked_window = 0;
1173 msg_dbg(msg,"SENT_FAST:");
1176 dbg("failed sent fast...\n");
1177 bearer_schedule(l_ptr->b_ptr, l_ptr);
1178 l_ptr->stats.bearer_congs++;
1179 l_ptr->next_out = buf;
1184 *used_max_pkt = link_max_pkt(l_ptr);
1186 return link_send_buf(l_ptr, buf); /* All other cases */
1190 * tipc_send_buf_fast: Entry for data messages where the
1191 * destination node is known and the header is complete,
1192 * inclusive total message length.
1193 * Returns user data length.
1195 int tipc_send_buf_fast(struct sk_buff *buf, u32 destnode)
1200 u32 selector = msg_origport(buf_msg(buf)) & 1;
1203 if (destnode == tipc_own_addr)
1204 return port_recv_msg(buf);
1206 read_lock_bh(&net_lock);
1207 n_ptr = node_select(destnode, selector);
1208 if (likely(n_ptr)) {
1210 l_ptr = n_ptr->active_links[selector];
1211 dbg("send_fast: buf %x selected %x, destnode = %x\n",
1212 buf, l_ptr, destnode);
1213 if (likely(l_ptr)) {
1214 res = link_send_buf_fast(l_ptr, buf, &dummy);
1216 read_unlock_bh(&net_lock);
1221 read_unlock_bh(&net_lock);
1222 res = msg_data_sz(buf_msg(buf));
1223 tipc_reject_msg(buf, TIPC_ERR_NO_NODE);
1229 * link_send_sections_fast: Entry for messages where the
1230 * destination processor is known and the header is complete,
1231 * except for total message length.
1232 * Returns user data length or errno.
1234 int link_send_sections_fast(struct port *sender,
1235 struct iovec const *msg_sect,
1239 struct tipc_msg *hdr = &sender->publ.phdr;
1241 struct sk_buff *buf;
1244 u32 selector = msg_origport(hdr) & 1;
1246 assert(destaddr != tipc_own_addr);
1250 * Try building message using port's max_pkt hint.
1251 * (Must not hold any locks while building message.)
1254 res = msg_build(hdr, msg_sect, num_sect, sender->max_pkt,
1255 !sender->user_port, &buf);
1257 read_lock_bh(&net_lock);
1258 node = node_select(destaddr, selector);
1261 l_ptr = node->active_links[selector];
1262 if (likely(l_ptr)) {
1264 res = link_send_buf_fast(l_ptr, buf,
1266 if (unlikely(res < 0))
1270 read_unlock_bh(&net_lock);
1274 /* Exit if build request was invalid */
1276 if (unlikely(res < 0))
1279 /* Exit if link (or bearer) is congested */
1281 if (link_congested(l_ptr) ||
1282 !list_empty(&l_ptr->b_ptr->cong_links)) {
1283 res = link_schedule_port(l_ptr,
1284 sender->publ.ref, res);
1289 * Message size exceeds max_pkt hint; update hint,
1290 * then re-try fast path or fragment the message
1293 sender->max_pkt = link_max_pkt(l_ptr);
1295 read_unlock_bh(&net_lock);
1298 if ((msg_hdr_sz(hdr) + res) <= sender->max_pkt)
1301 return link_send_sections_long(sender, msg_sect,
1302 num_sect, destaddr);
1306 read_unlock_bh(&net_lock);
1308 /* Couldn't find a link to the destination node */
1311 return tipc_reject_msg(buf, TIPC_ERR_NO_NODE);
1313 return port_reject_sections(sender, hdr, msg_sect, num_sect,
1319 * link_send_sections_long(): Entry for long messages where the
1320 * destination node is known and the header is complete,
1321 * inclusive total message length.
1322 * Link and bearer congestion status have been checked to be ok,
1323 * and are ignored if they change.
1325 * Note that fragments do not use the full link MTU so that they won't have
1326 * to undergo refragmentation if link changeover causes them to be sent
1327 * over another link with an additional tunnel header added as prefix.
1328 * (Refragmentation will still occur if the other link has a smaller MTU.)
1330 * Returns user data length or errno.
1332 static int link_send_sections_long(struct port *sender,
1333 struct iovec const *msg_sect,
1339 struct tipc_msg *hdr = &sender->publ.phdr;
1340 u32 dsz = msg_data_sz(hdr);
1341 u32 max_pkt,fragm_sz,rest;
1342 struct tipc_msg fragm_hdr;
1343 struct sk_buff *buf,*buf_chain,*prev;
1344 u32 fragm_crs,fragm_rest,hsz,sect_rest;
1345 const unchar *sect_crs;
1351 max_pkt = sender->max_pkt - INT_H_SIZE;
1352 /* leave room for tunnel header in case of link changeover */
1353 fragm_sz = max_pkt - INT_H_SIZE;
1354 /* leave room for fragmentation header in each fragment */
1362 /* Prepare reusable fragment header: */
1364 msg_dbg(hdr, ">FRAGMENTING>");
1365 msg_init(&fragm_hdr, MSG_FRAGMENTER, FIRST_FRAGMENT,
1366 TIPC_OK, INT_H_SIZE, msg_destnode(hdr));
1367 msg_set_link_selector(&fragm_hdr, sender->publ.ref);
1368 msg_set_size(&fragm_hdr, max_pkt);
1369 msg_set_fragm_no(&fragm_hdr, 1);
1371 /* Prepare header of first fragment: */
1373 buf_chain = buf = buf_acquire(max_pkt);
1377 memcpy(buf->data, (unchar *)&fragm_hdr, INT_H_SIZE);
1378 hsz = msg_hdr_sz(hdr);
1379 memcpy(buf->data + INT_H_SIZE, (unchar *)hdr, hsz);
1380 msg_dbg(buf_msg(buf), ">BUILD>");
1382 /* Chop up message: */
1384 fragm_crs = INT_H_SIZE + hsz;
1385 fragm_rest = fragm_sz - hsz;
1387 do { /* For all sections */
1391 sect_rest = msg_sect[++curr_sect].iov_len;
1392 sect_crs = (const unchar *)msg_sect[curr_sect].iov_base;
1395 if (sect_rest < fragm_rest)
1400 if (likely(!sender->user_port)) {
1401 if (copy_from_user(buf->data + fragm_crs, sect_crs, sz)) {
1403 for (; buf_chain; buf_chain = buf) {
1404 buf = buf_chain->next;
1405 buf_discard(buf_chain);
1410 memcpy(buf->data + fragm_crs, sect_crs, sz);
1418 if (!fragm_rest && rest) {
1420 /* Initiate new fragment: */
1421 if (rest <= fragm_sz) {
1423 msg_set_type(&fragm_hdr,LAST_FRAGMENT);
1425 msg_set_type(&fragm_hdr, FRAGMENT);
1427 msg_set_size(&fragm_hdr, fragm_sz + INT_H_SIZE);
1428 msg_set_fragm_no(&fragm_hdr, ++fragm_no);
1430 buf = buf_acquire(fragm_sz + INT_H_SIZE);
1436 memcpy(buf->data, (unchar *)&fragm_hdr, INT_H_SIZE);
1437 fragm_crs = INT_H_SIZE;
1438 fragm_rest = fragm_sz;
1439 msg_dbg(buf_msg(buf)," >BUILD>");
1445 * Now we have a buffer chain. Select a link and check
1446 * that packet size is still OK
1448 node = node_select(destaddr, sender->publ.ref & 1);
1451 l_ptr = node->active_links[sender->publ.ref & 1];
1456 if (link_max_pkt(l_ptr) < max_pkt) {
1457 sender->max_pkt = link_max_pkt(l_ptr);
1459 for (; buf_chain; buf_chain = buf) {
1460 buf = buf_chain->next;
1461 buf_discard(buf_chain);
1467 for (; buf_chain; buf_chain = buf) {
1468 buf = buf_chain->next;
1469 buf_discard(buf_chain);
1471 return port_reject_sections(sender, hdr, msg_sect, num_sect,
1475 /* Append whole chain to send queue: */
1478 l_ptr->long_msg_seq_no = mod(l_ptr->long_msg_seq_no + 1);
1479 if (!l_ptr->next_out)
1480 l_ptr->next_out = buf_chain;
1481 l_ptr->stats.sent_fragmented++;
1483 struct sk_buff *next = buf->next;
1484 struct tipc_msg *msg = buf_msg(buf);
1486 l_ptr->stats.sent_fragments++;
1487 msg_set_long_msgno(msg, l_ptr->long_msg_seq_no);
1488 link_add_to_outqueue(l_ptr, buf, msg);
1489 msg_dbg(msg, ">ADD>");
1493 /* Send it, if possible: */
1495 link_push_queue(l_ptr);
1501 * link_push_packet: Push one unsent packet to the media
1503 u32 link_push_packet(struct link *l_ptr)
1505 struct sk_buff *buf = l_ptr->first_out;
1506 u32 r_q_size = l_ptr->retransm_queue_size;
1507 u32 r_q_head = l_ptr->retransm_queue_head;
1509 /* Step to position where retransmission failed, if any, */
1510 /* consider that buffers may have been released in meantime */
1512 if (r_q_size && buf) {
1513 u32 last = lesser(mod(r_q_head + r_q_size),
1514 link_last_sent(l_ptr));
1515 u32 first = msg_seqno(buf_msg(buf));
1517 while (buf && less(first, r_q_head)) {
1518 first = mod(first + 1);
1521 l_ptr->retransm_queue_head = r_q_head = first;
1522 l_ptr->retransm_queue_size = r_q_size = mod(last - first);
1525 /* Continue retransmission now, if there is anything: */
1527 if (r_q_size && buf && !skb_cloned(buf)) {
1528 msg_set_ack(buf_msg(buf), mod(l_ptr->next_in_no - 1));
1529 msg_set_bcast_ack(buf_msg(buf), l_ptr->owner->bclink.last_in);
1530 if (bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
1531 msg_dbg(buf_msg(buf), ">DEF-RETR>");
1532 l_ptr->retransm_queue_head = mod(++r_q_head);
1533 l_ptr->retransm_queue_size = --r_q_size;
1534 l_ptr->stats.retransmitted++;
1537 l_ptr->stats.bearer_congs++;
1538 msg_dbg(buf_msg(buf), "|>DEF-RETR>");
1543 /* Send deferred protocol message, if any: */
1545 buf = l_ptr->proto_msg_queue;
1547 msg_set_ack(buf_msg(buf), mod(l_ptr->next_in_no - 1));
1548 msg_set_bcast_ack(buf_msg(buf),l_ptr->owner->bclink.last_in);
1549 if (bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
1550 msg_dbg(buf_msg(buf), ">DEF-PROT>");
1551 l_ptr->unacked_window = 0;
1553 l_ptr->proto_msg_queue = 0;
1556 msg_dbg(buf_msg(buf), "|>DEF-PROT>");
1557 l_ptr->stats.bearer_congs++;
1562 /* Send one deferred data message, if send window not full: */
1564 buf = l_ptr->next_out;
1566 struct tipc_msg *msg = buf_msg(buf);
1567 u32 next = msg_seqno(msg);
1568 u32 first = msg_seqno(buf_msg(l_ptr->first_out));
1570 if (mod(next - first) < l_ptr->queue_limit[0]) {
1571 msg_set_ack(msg, mod(l_ptr->next_in_no - 1));
1572 msg_set_bcast_ack(msg, l_ptr->owner->bclink.last_in);
1573 if (bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
1574 if (msg_user(msg) == MSG_BUNDLER)
1575 msg_set_type(msg, CLOSED_MSG);
1576 msg_dbg(msg, ">PUSH-DATA>");
1577 l_ptr->next_out = buf->next;
1580 msg_dbg(msg, "|PUSH-DATA|");
1581 l_ptr->stats.bearer_congs++;
1586 return PUSH_FINISHED;
1590 * push_queue(): push out the unsent messages of a link where
1591 * congestion has abated. Node is locked
1593 void link_push_queue(struct link *l_ptr)
1597 if (bearer_congested(l_ptr->b_ptr, l_ptr))
1601 res = link_push_packet(l_ptr);
1603 while (res == TIPC_OK);
1604 if (res == PUSH_FAILED)
1605 bearer_schedule(l_ptr->b_ptr, l_ptr);
1608 void link_retransmit(struct link *l_ptr, struct sk_buff *buf,
1611 struct tipc_msg *msg;
1613 dbg("Retransmitting %u in link %x\n", retransmits, l_ptr);
1615 if (bearer_congested(l_ptr->b_ptr, l_ptr) && buf && !skb_cloned(buf)) {
1616 msg_dbg(buf_msg(buf), ">NO_RETR->BCONG>");
1617 dbg_print_link(l_ptr, " ");
1618 l_ptr->retransm_queue_head = msg_seqno(buf_msg(buf));
1619 l_ptr->retransm_queue_size = retransmits;
1622 while (retransmits && (buf != l_ptr->next_out) && buf && !skb_cloned(buf)) {
1624 msg_set_ack(msg, mod(l_ptr->next_in_no - 1));
1625 msg_set_bcast_ack(msg, l_ptr->owner->bclink.last_in);
1626 if (bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
1627 /* Catch if retransmissions fail repeatedly: */
1628 if (l_ptr->last_retransmitted == msg_seqno(msg)) {
1629 if (++l_ptr->stale_count > 100) {
1630 msg_print(CONS, buf_msg(buf), ">RETR>");
1631 info("...Retransmitted %u times\n",
1632 l_ptr->stale_count);
1633 link_print(l_ptr, CONS, "Resetting Link\n");;
1638 l_ptr->stale_count = 0;
1640 l_ptr->last_retransmitted = msg_seqno(msg);
1642 msg_dbg(buf_msg(buf), ">RETR>");
1645 l_ptr->stats.retransmitted++;
1647 bearer_schedule(l_ptr->b_ptr, l_ptr);
1648 l_ptr->stats.bearer_congs++;
1649 l_ptr->retransm_queue_head = msg_seqno(buf_msg(buf));
1650 l_ptr->retransm_queue_size = retransmits;
1654 l_ptr->retransm_queue_head = l_ptr->retransm_queue_size = 0;
1658 * link_recv_non_seq: Receive packets which are outside
1659 * the link sequence flow
1662 static void link_recv_non_seq(struct sk_buff *buf)
1664 struct tipc_msg *msg = buf_msg(buf);
1666 if (msg_user(msg) == LINK_CONFIG)
1669 bclink_recv_pkt(buf);
1673 * link_insert_deferred_queue - insert deferred messages back into receive chain
1676 static struct sk_buff *link_insert_deferred_queue(struct link *l_ptr,
1677 struct sk_buff *buf)
1681 if (l_ptr->oldest_deferred_in == NULL)
1684 seq_no = msg_seqno(buf_msg(l_ptr->oldest_deferred_in));
1685 if (seq_no == mod(l_ptr->next_in_no)) {
1686 l_ptr->newest_deferred_in->next = buf;
1687 buf = l_ptr->oldest_deferred_in;
1688 l_ptr->oldest_deferred_in = NULL;
1689 l_ptr->deferred_inqueue_sz = 0;
1694 void tipc_recv_msg(struct sk_buff *head, struct tipc_bearer *tb_ptr)
1696 read_lock_bh(&net_lock);
1698 struct bearer *b_ptr;
1701 struct sk_buff *crs;
1702 struct sk_buff *buf = head;
1703 struct tipc_msg *msg = buf_msg(buf);
1704 u32 seq_no = msg_seqno(msg);
1705 u32 ackd = msg_ack(msg);
1709 b_ptr = (struct bearer *)tb_ptr;
1710 TIPC_SKB_CB(buf)->handle = b_ptr;
1713 if (unlikely(msg_version(msg) != TIPC_VERSION))
1716 if (msg_user(msg) != LINK_PROTOCOL)
1718 msg_dbg(msg,"<REC<");
1720 if (unlikely(msg_non_seq(msg))) {
1721 link_recv_non_seq(buf);
1724 n_ptr = node_find(msg_prevnode(msg));
1725 if (unlikely(!n_ptr))
1729 l_ptr = n_ptr->links[b_ptr->identity];
1730 if (unlikely(!l_ptr)) {
1735 * Release acked messages
1737 if (less(n_ptr->bclink.acked, msg_bcast_ack(msg))) {
1738 if (node_is_up(n_ptr) && n_ptr->bclink.supported)
1739 bclink_acknowledge(n_ptr, msg_bcast_ack(msg));
1742 crs = l_ptr->first_out;
1743 while ((crs != l_ptr->next_out) &&
1744 less_eq(msg_seqno(buf_msg(crs)), ackd)) {
1745 struct sk_buff *next = crs->next;
1752 l_ptr->first_out = crs;
1753 l_ptr->out_queue_size -= released;
1755 if (unlikely(l_ptr->next_out))
1756 link_push_queue(l_ptr);
1757 if (unlikely(!list_empty(&l_ptr->waiting_ports)))
1758 link_wakeup_ports(l_ptr, 0);
1759 if (unlikely(++l_ptr->unacked_window >= TIPC_MIN_LINK_WIN)) {
1760 l_ptr->stats.sent_acks++;
1761 link_send_proto_msg(l_ptr, STATE_MSG, 0, 0, 0, 0, 0);
1765 if (likely(link_working_working(l_ptr))) {
1766 if (likely(seq_no == mod(l_ptr->next_in_no))) {
1767 l_ptr->next_in_no++;
1768 if (unlikely(l_ptr->oldest_deferred_in))
1769 head = link_insert_deferred_queue(l_ptr,
1771 if (likely(msg_is_dest(msg, tipc_own_addr))) {
1773 if (likely(msg_isdata(msg))) {
1778 switch (msg_user(msg)) {
1780 l_ptr->stats.recv_bundles++;
1781 l_ptr->stats.recv_bundled +=
1784 link_recv_bundle(buf);
1786 case ROUTE_DISTRIBUTOR:
1788 cluster_recv_routing_table(buf);
1790 case NAME_DISTRIBUTOR:
1796 port_recv_proto_msg(buf);
1798 case MSG_FRAGMENTER:
1799 l_ptr->stats.recv_fragments++;
1800 if (link_recv_fragment(
1801 &l_ptr->defragm_buf,
1803 l_ptr->stats.recv_fragmented++;
1807 case CHANGEOVER_PROTOCOL:
1808 type = msg_type(msg);
1809 if (link_recv_changeover_msg(
1812 seq_no = msg_seqno(msg);
1813 TIPC_SKB_CB(buf)->handle
1815 if (type == ORIGINAL_MSG)
1817 goto protocol_check;
1826 link_handle_out_of_seq_msg(l_ptr, buf);
1827 head = link_insert_deferred_queue(l_ptr, head);
1832 if (msg_user(msg) == LINK_PROTOCOL) {
1833 link_recv_proto_msg(l_ptr, buf);
1834 head = link_insert_deferred_queue(l_ptr, head);
1838 msg_dbg(msg,"NSEQ<REC<");
1839 link_state_event(l_ptr, TRAFFIC_MSG_EVT);
1841 if (link_working_working(l_ptr)) {
1842 /* Re-insert in front of queue */
1843 msg_dbg(msg,"RECV-REINS:");
1853 read_unlock_bh(&net_lock);
1857 * link_defer_buf(): Sort a received out-of-sequence packet
1858 * into the deferred reception queue.
1859 * Returns the increase of the queue length,i.e. 0 or 1
1862 u32 link_defer_pkt(struct sk_buff **head,
1863 struct sk_buff **tail,
1864 struct sk_buff *buf)
1866 struct sk_buff *prev = 0;
1867 struct sk_buff *crs = *head;
1868 u32 seq_no = msg_seqno(buf_msg(buf));
1873 if (*head == NULL) {
1874 *head = *tail = buf;
1879 if (less(msg_seqno(buf_msg(*tail)), seq_no)) {
1880 (*tail)->next = buf;
1885 /* Scan through queue and sort it in */
1887 struct tipc_msg *msg = buf_msg(crs);
1889 if (less(seq_no, msg_seqno(msg))) {
1897 if (seq_no == msg_seqno(msg)) {
1905 /* Message is a duplicate of an existing message */
1912 * link_handle_out_of_seq_msg - handle arrival of out-of-sequence packet
1915 static void link_handle_out_of_seq_msg(struct link *l_ptr,
1916 struct sk_buff *buf)
1918 u32 seq_no = msg_seqno(buf_msg(buf));
1920 if (likely(msg_user(buf_msg(buf)) == LINK_PROTOCOL)) {
1921 link_recv_proto_msg(l_ptr, buf);
1925 dbg("rx OOS msg: seq_no %u, expecting %u (%u)\n",
1926 seq_no, mod(l_ptr->next_in_no), l_ptr->next_in_no);
1928 /* Record OOS packet arrival (force mismatch on next timeout) */
1930 l_ptr->checkpoint--;
1933 * Discard packet if a duplicate; otherwise add it to deferred queue
1934 * and notify peer of gap as per protocol specification
1937 if (less(seq_no, mod(l_ptr->next_in_no))) {
1938 l_ptr->stats.duplicates++;
1943 if (link_defer_pkt(&l_ptr->oldest_deferred_in,
1944 &l_ptr->newest_deferred_in, buf)) {
1945 l_ptr->deferred_inqueue_sz++;
1946 l_ptr->stats.deferred_recv++;
1947 if ((l_ptr->deferred_inqueue_sz % 16) == 1)
1948 link_send_proto_msg(l_ptr, STATE_MSG, 0, 0, 0, 0, 0);
1950 l_ptr->stats.duplicates++;
1954 * Send protocol message to the other endpoint.
1956 void link_send_proto_msg(struct link *l_ptr, u32 msg_typ, int probe_msg,
1957 u32 gap, u32 tolerance, u32 priority, u32 ack_mtu)
1959 struct sk_buff *buf = 0;
1960 struct tipc_msg *msg = l_ptr->pmsg;
1961 u32 msg_size = sizeof(l_ptr->proto_msg);
1963 if (link_blocked(l_ptr))
1965 msg_set_type(msg, msg_typ);
1966 msg_set_net_plane(msg, l_ptr->b_ptr->net_plane);
1967 msg_set_bcast_ack(msg, mod(l_ptr->owner->bclink.last_in));
1968 msg_set_last_bcast(msg, bclink_get_last_sent());
1970 if (msg_typ == STATE_MSG) {
1971 u32 next_sent = mod(l_ptr->next_out_no);
1973 if (!link_is_up(l_ptr))
1975 if (l_ptr->next_out)
1976 next_sent = msg_seqno(buf_msg(l_ptr->next_out));
1977 msg_set_next_sent(msg, next_sent);
1978 if (l_ptr->oldest_deferred_in) {
1979 u32 rec = msg_seqno(buf_msg(l_ptr->oldest_deferred_in));
1980 gap = mod(rec - mod(l_ptr->next_in_no));
1982 msg_set_seq_gap(msg, gap);
1984 l_ptr->stats.sent_nacks++;
1985 msg_set_link_tolerance(msg, tolerance);
1986 msg_set_linkprio(msg, priority);
1987 msg_set_max_pkt(msg, ack_mtu);
1988 msg_set_ack(msg, mod(l_ptr->next_in_no - 1));
1989 msg_set_probe(msg, probe_msg != 0);
1991 u32 mtu = l_ptr->max_pkt;
1993 if ((mtu < l_ptr->max_pkt_target) &&
1994 link_working_working(l_ptr) &&
1995 l_ptr->fsm_msg_cnt) {
1996 msg_size = (mtu + (l_ptr->max_pkt_target - mtu)/2 + 2) & ~3;
1997 if (l_ptr->max_pkt_probes == 10) {
1998 l_ptr->max_pkt_target = (msg_size - 4);
1999 l_ptr->max_pkt_probes = 0;
2000 msg_size = (mtu + (l_ptr->max_pkt_target - mtu)/2 + 2) & ~3;
2002 l_ptr->max_pkt_probes++;
2005 l_ptr->stats.sent_probes++;
2007 l_ptr->stats.sent_states++;
2008 } else { /* RESET_MSG or ACTIVATE_MSG */
2009 msg_set_ack(msg, mod(l_ptr->reset_checkpoint - 1));
2010 msg_set_seq_gap(msg, 0);
2011 msg_set_next_sent(msg, 1);
2012 msg_set_link_tolerance(msg, l_ptr->tolerance);
2013 msg_set_linkprio(msg, l_ptr->priority);
2014 msg_set_max_pkt(msg, l_ptr->max_pkt_target);
2017 if (node_has_redundant_links(l_ptr->owner)) {
2018 msg_set_redundant_link(msg);
2020 msg_clear_redundant_link(msg);
2022 msg_set_linkprio(msg, l_ptr->priority);
2024 /* Ensure sequence number will not fit : */
2026 msg_set_seqno(msg, mod(l_ptr->next_out_no + (0xffff/2)));
2030 if (bearer_congested(l_ptr->b_ptr, l_ptr)) {
2031 if (!l_ptr->proto_msg_queue) {
2032 l_ptr->proto_msg_queue =
2033 buf_acquire(sizeof(l_ptr->proto_msg));
2035 buf = l_ptr->proto_msg_queue;
2038 memcpy(buf->data, (unchar *)msg, sizeof(l_ptr->proto_msg));
2041 msg_set_timestamp(msg, jiffies_to_msecs(jiffies));
2043 /* Message can be sent */
2047 buf = buf_acquire(msg_size);
2051 memcpy(buf->data, (unchar *)msg, sizeof(l_ptr->proto_msg));
2052 msg_set_size(buf_msg(buf), msg_size);
2054 if (bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
2055 l_ptr->unacked_window = 0;
2060 /* New congestion */
2061 bearer_schedule(l_ptr->b_ptr, l_ptr);
2062 l_ptr->proto_msg_queue = buf;
2063 l_ptr->stats.bearer_congs++;
2067 * Receive protocol message :
2068 * Note that network plane id propagates through the network, and may
2069 * change at any time. The node with lowest address rules
2072 static void link_recv_proto_msg(struct link *l_ptr, struct sk_buff *buf)
2078 struct tipc_msg *msg = buf_msg(buf);
2080 dbg("AT(%u):", jiffies_to_msecs(jiffies));
2082 if (link_blocked(l_ptr))
2085 /* record unnumbered packet arrival (force mismatch on next timeout) */
2087 l_ptr->checkpoint--;
2089 if (l_ptr->b_ptr->net_plane != msg_net_plane(msg))
2090 if (tipc_own_addr > msg_prevnode(msg))
2091 l_ptr->b_ptr->net_plane = msg_net_plane(msg);
2093 l_ptr->owner->permit_changeover = msg_redundant_link(msg);
2095 switch (msg_type(msg)) {
2098 if (!link_working_unknown(l_ptr) && l_ptr->peer_session) {
2099 if (msg_session(msg) == l_ptr->peer_session) {
2100 dbg("Duplicate RESET: %u<->%u\n",
2101 msg_session(msg), l_ptr->peer_session);
2102 break; /* duplicate: ignore */
2107 /* Update link settings according other endpoint's values */
2109 strcpy((strrchr(l_ptr->name, ':') + 1), (char *)msg_data(msg));
2111 if ((msg_tol = msg_link_tolerance(msg)) &&
2112 (msg_tol > l_ptr->tolerance))
2113 link_set_supervision_props(l_ptr, msg_tol);
2115 if (msg_linkprio(msg) > l_ptr->priority)
2116 l_ptr->priority = msg_linkprio(msg);
2118 max_pkt_info = msg_max_pkt(msg);
2120 if (max_pkt_info < l_ptr->max_pkt_target)
2121 l_ptr->max_pkt_target = max_pkt_info;
2122 if (l_ptr->max_pkt > l_ptr->max_pkt_target)
2123 l_ptr->max_pkt = l_ptr->max_pkt_target;
2125 l_ptr->max_pkt = l_ptr->max_pkt_target;
2127 l_ptr->owner->bclink.supported = (max_pkt_info != 0);
2129 link_state_event(l_ptr, msg_type(msg));
2131 l_ptr->peer_session = msg_session(msg);
2132 l_ptr->peer_bearer_id = msg_bearer_id(msg);
2134 /* Synchronize broadcast sequence numbers */
2135 if (!node_has_redundant_links(l_ptr->owner)) {
2136 l_ptr->owner->bclink.last_in = mod(msg_last_bcast(msg));
2141 if ((msg_tol = msg_link_tolerance(msg)))
2142 link_set_supervision_props(l_ptr, msg_tol);
2144 if (msg_linkprio(msg) &&
2145 (msg_linkprio(msg) != l_ptr->priority)) {
2146 warn("Changing prio <%s>: %u->%u\n",
2147 l_ptr->name, l_ptr->priority, msg_linkprio(msg));
2148 l_ptr->priority = msg_linkprio(msg);
2149 link_reset(l_ptr); /* Enforce change to take effect */
2152 link_state_event(l_ptr, TRAFFIC_MSG_EVT);
2153 l_ptr->stats.recv_states++;
2154 if (link_reset_unknown(l_ptr))
2157 if (less_eq(mod(l_ptr->next_in_no), msg_next_sent(msg))) {
2158 rec_gap = mod(msg_next_sent(msg) -
2159 mod(l_ptr->next_in_no));
2162 max_pkt_ack = msg_max_pkt(msg);
2163 if (max_pkt_ack > l_ptr->max_pkt) {
2164 dbg("Link <%s> updated MTU %u -> %u\n",
2165 l_ptr->name, l_ptr->max_pkt, max_pkt_ack);
2166 l_ptr->max_pkt = max_pkt_ack;
2167 l_ptr->max_pkt_probes = 0;
2171 if (msg_probe(msg)) {
2172 l_ptr->stats.recv_probes++;
2173 if (msg_size(msg) > sizeof(l_ptr->proto_msg)) {
2174 max_pkt_ack = msg_size(msg);
2178 /* Protocol message before retransmits, reduce loss risk */
2180 bclink_check_gap(l_ptr->owner, msg_last_bcast(msg));
2182 if (rec_gap || (msg_probe(msg))) {
2183 link_send_proto_msg(l_ptr, STATE_MSG,
2184 0, rec_gap, 0, 0, max_pkt_ack);
2186 if (msg_seq_gap(msg)) {
2187 msg_dbg(msg, "With Gap:");
2188 l_ptr->stats.recv_nacks++;
2189 link_retransmit(l_ptr, l_ptr->first_out,
2194 msg_dbg(buf_msg(buf), "<DISCARDING UNKNOWN<");
2202 * link_tunnel(): Send one message via a link belonging to
2203 * another bearer. Owner node is locked.
2205 void link_tunnel(struct link *l_ptr,
2206 struct tipc_msg *tunnel_hdr,
2207 struct tipc_msg *msg,
2210 struct link *tunnel;
2211 struct sk_buff *buf;
2212 u32 length = msg_size(msg);
2214 tunnel = l_ptr->owner->active_links[selector & 1];
2215 if (!link_is_up(tunnel))
2217 msg_set_size(tunnel_hdr, length + INT_H_SIZE);
2218 buf = buf_acquire(length + INT_H_SIZE);
2221 memcpy(buf->data, (unchar *)tunnel_hdr, INT_H_SIZE);
2222 memcpy(buf->data + INT_H_SIZE, (unchar *)msg, length);
2223 dbg("%c->%c:", l_ptr->b_ptr->net_plane, tunnel->b_ptr->net_plane);
2224 msg_dbg(buf_msg(buf), ">SEND>");
2226 link_send_buf(tunnel, buf);
2232 * changeover(): Send whole message queue via the remaining link
2233 * Owner node is locked.
2236 void link_changeover(struct link *l_ptr)
2238 u32 msgcount = l_ptr->out_queue_size;
2239 struct sk_buff *crs = l_ptr->first_out;
2240 struct link *tunnel = l_ptr->owner->active_links[0];
2241 int split_bundles = node_has_redundant_links(l_ptr->owner);
2242 struct tipc_msg tunnel_hdr;
2247 if (!l_ptr->owner->permit_changeover)
2250 msg_init(&tunnel_hdr, CHANGEOVER_PROTOCOL,
2251 ORIGINAL_MSG, TIPC_OK, INT_H_SIZE, l_ptr->addr);
2252 msg_set_bearer_id(&tunnel_hdr, l_ptr->peer_bearer_id);
2253 msg_set_msgcnt(&tunnel_hdr, msgcount);
2254 if (!l_ptr->first_out) {
2255 struct sk_buff *buf;
2258 buf = buf_acquire(INT_H_SIZE);
2260 memcpy(buf->data, (unchar *)&tunnel_hdr, INT_H_SIZE);
2261 msg_set_size(&tunnel_hdr, INT_H_SIZE);
2262 dbg("%c->%c:", l_ptr->b_ptr->net_plane,
2263 tunnel->b_ptr->net_plane);
2264 msg_dbg(&tunnel_hdr, "EMPTY>SEND>");
2265 link_send_buf(tunnel, buf);
2267 warn("Memory squeeze; link changeover failed\n");
2272 struct tipc_msg *msg = buf_msg(crs);
2274 if ((msg_user(msg) == MSG_BUNDLER) && split_bundles) {
2275 u32 msgcount = msg_msgcnt(msg);
2276 struct tipc_msg *m = msg_get_wrapped(msg);
2277 unchar* pos = (unchar*)m;
2279 while (msgcount--) {
2280 msg_set_seqno(m,msg_seqno(msg));
2281 link_tunnel(l_ptr, &tunnel_hdr, m,
2282 msg_link_selector(m));
2283 pos += align(msg_size(m));
2284 m = (struct tipc_msg *)pos;
2287 link_tunnel(l_ptr, &tunnel_hdr, msg,
2288 msg_link_selector(msg));
2294 void link_send_duplicate(struct link *l_ptr, struct link *tunnel)
2296 struct sk_buff *iter;
2297 struct tipc_msg tunnel_hdr;
2299 msg_init(&tunnel_hdr, CHANGEOVER_PROTOCOL,
2300 DUPLICATE_MSG, TIPC_OK, INT_H_SIZE, l_ptr->addr);
2301 msg_set_msgcnt(&tunnel_hdr, l_ptr->out_queue_size);
2302 msg_set_bearer_id(&tunnel_hdr, l_ptr->peer_bearer_id);
2303 iter = l_ptr->first_out;
2305 struct sk_buff *outbuf;
2306 struct tipc_msg *msg = buf_msg(iter);
2307 u32 length = msg_size(msg);
2309 if (msg_user(msg) == MSG_BUNDLER)
2310 msg_set_type(msg, CLOSED_MSG);
2311 msg_set_ack(msg, mod(l_ptr->next_in_no - 1)); /* Update */
2312 msg_set_bcast_ack(msg, l_ptr->owner->bclink.last_in);
2313 msg_set_size(&tunnel_hdr, length + INT_H_SIZE);
2314 outbuf = buf_acquire(length + INT_H_SIZE);
2315 if (outbuf == NULL) {
2316 warn("Memory squeeze; buffer duplication failed\n");
2319 memcpy(outbuf->data, (unchar *)&tunnel_hdr, INT_H_SIZE);
2320 memcpy(outbuf->data + INT_H_SIZE, iter->data, length);
2321 dbg("%c->%c:", l_ptr->b_ptr->net_plane,
2322 tunnel->b_ptr->net_plane);
2323 msg_dbg(buf_msg(outbuf), ">SEND>");
2324 link_send_buf(tunnel, outbuf);
2325 if (!link_is_up(l_ptr))
2334 * buf_extract - extracts embedded TIPC message from another message
2335 * @skb: encapsulating message buffer
2336 * @from_pos: offset to extract from
2338 * Returns a new message buffer containing an embedded message. The
2339 * encapsulating message itself is left unchanged.
2342 static struct sk_buff *buf_extract(struct sk_buff *skb, u32 from_pos)
2344 struct tipc_msg *msg = (struct tipc_msg *)(skb->data + from_pos);
2345 u32 size = msg_size(msg);
2348 eb = buf_acquire(size);
2350 memcpy(eb->data, (unchar *)msg, size);
2355 * link_recv_changeover_msg(): Receive tunneled packet sent
2356 * via other link. Node is locked. Return extracted buffer.
2359 static int link_recv_changeover_msg(struct link **l_ptr,
2360 struct sk_buff **buf)
2362 struct sk_buff *tunnel_buf = *buf;
2363 struct link *dest_link;
2364 struct tipc_msg *msg;
2365 struct tipc_msg *tunnel_msg = buf_msg(tunnel_buf);
2366 u32 msg_typ = msg_type(tunnel_msg);
2367 u32 msg_count = msg_msgcnt(tunnel_msg);
2369 dest_link = (*l_ptr)->owner->links[msg_bearer_id(tunnel_msg)];
2370 assert(dest_link != *l_ptr);
2372 msg_dbg(tunnel_msg, "NOLINK/<REC<");
2375 dbg("%c<-%c:", dest_link->b_ptr->net_plane,
2376 (*l_ptr)->b_ptr->net_plane);
2378 msg = msg_get_wrapped(tunnel_msg);
2380 if (msg_typ == DUPLICATE_MSG) {
2381 if (less(msg_seqno(msg), mod(dest_link->next_in_no))) {
2382 msg_dbg(tunnel_msg, "DROP/<REC<");
2385 *buf = buf_extract(tunnel_buf,INT_H_SIZE);
2387 warn("Memory squeeze; failed to extract msg\n");
2390 msg_dbg(tunnel_msg, "TNL<REC<");
2391 buf_discard(tunnel_buf);
2395 /* First original message ?: */
2397 if (link_is_up(dest_link)) {
2398 msg_dbg(tunnel_msg, "UP/FIRST/<REC<");
2399 link_reset(dest_link);
2400 dest_link->exp_msg_count = msg_count;
2403 } else if (dest_link->exp_msg_count == START_CHANGEOVER) {
2404 msg_dbg(tunnel_msg, "BLK/FIRST/<REC<");
2405 dest_link->exp_msg_count = msg_count;
2410 /* Receive original message */
2412 if (dest_link->exp_msg_count == 0) {
2413 msg_dbg(tunnel_msg, "OVERDUE/DROP/<REC<");
2414 dbg_print_link(dest_link, "LINK:");
2417 dest_link->exp_msg_count--;
2418 if (less(msg_seqno(msg), dest_link->reset_checkpoint)) {
2419 msg_dbg(tunnel_msg, "DROP/DUPL/<REC<");
2422 *buf = buf_extract(tunnel_buf, INT_H_SIZE);
2424 msg_dbg(tunnel_msg, "TNL<REC<");
2425 buf_discard(tunnel_buf);
2428 warn("Memory squeeze; dropped incoming msg\n");
2433 buf_discard(tunnel_buf);
2438 * Bundler functionality:
2440 void link_recv_bundle(struct sk_buff *buf)
2442 u32 msgcount = msg_msgcnt(buf_msg(buf));
2443 u32 pos = INT_H_SIZE;
2444 struct sk_buff *obuf;
2446 msg_dbg(buf_msg(buf), "<BNDL<: ");
2447 while (msgcount--) {
2448 obuf = buf_extract(buf, pos);
2450 char addr_string[16];
2452 warn("Buffer allocation failure;\n");
2453 warn(" incoming message(s) from %s lost\n",
2454 addr_string_fill(addr_string,
2455 msg_orignode(buf_msg(buf))));
2458 pos += align(msg_size(buf_msg(obuf)));
2459 msg_dbg(buf_msg(obuf), " /");
2460 net_route_msg(obuf);
2466 * Fragmentation/defragmentation:
2471 * link_send_long_buf: Entry for buffers needing fragmentation.
2472 * The buffer is complete, inclusive total message length.
2473 * Returns user data length.
2475 int link_send_long_buf(struct link *l_ptr, struct sk_buff *buf)
2477 struct tipc_msg *inmsg = buf_msg(buf);
2478 struct tipc_msg fragm_hdr;
2479 u32 insize = msg_size(inmsg);
2480 u32 dsz = msg_data_sz(inmsg);
2481 unchar *crs = buf->data;
2483 u32 pack_sz = link_max_pkt(l_ptr);
2484 u32 fragm_sz = pack_sz - INT_H_SIZE;
2486 u32 destaddr = msg_destnode(inmsg);
2488 if (msg_short(inmsg))
2489 destaddr = l_ptr->addr;
2491 if (msg_routed(inmsg))
2492 msg_set_prevnode(inmsg, tipc_own_addr);
2494 /* Prepare reusable fragment header: */
2496 msg_init(&fragm_hdr, MSG_FRAGMENTER, FIRST_FRAGMENT,
2497 TIPC_OK, INT_H_SIZE, destaddr);
2498 msg_set_link_selector(&fragm_hdr, msg_link_selector(inmsg));
2499 msg_set_long_msgno(&fragm_hdr, mod(l_ptr->long_msg_seq_no++));
2500 msg_set_fragm_no(&fragm_hdr, fragm_no);
2501 l_ptr->stats.sent_fragmented++;
2503 /* Chop up message: */
2506 struct sk_buff *fragm;
2508 if (rest <= fragm_sz) {
2510 msg_set_type(&fragm_hdr, LAST_FRAGMENT);
2512 fragm = buf_acquire(fragm_sz + INT_H_SIZE);
2513 if (fragm == NULL) {
2514 warn("Memory squeeze; failed to fragment msg\n");
2518 msg_set_size(&fragm_hdr, fragm_sz + INT_H_SIZE);
2519 memcpy(fragm->data, (unchar *)&fragm_hdr, INT_H_SIZE);
2520 memcpy(fragm->data + INT_H_SIZE, crs, fragm_sz);
2522 /* Send queued messages first, if any: */
2524 l_ptr->stats.sent_fragments++;
2525 link_send_buf(l_ptr, fragm);
2526 if (!link_is_up(l_ptr))
2528 msg_set_fragm_no(&fragm_hdr, ++fragm_no);
2531 msg_set_type(&fragm_hdr, FRAGMENT);
2539 * A pending message being re-assembled must store certain values
2540 * to handle subsequent fragments correctly. The following functions
2541 * help storing these values in unused, available fields in the
2542 * pending message. This makes dynamic memory allocation unecessary.
2545 static inline u32 get_long_msg_seqno(struct sk_buff *buf)
2547 return msg_seqno(buf_msg(buf));
2550 static inline void set_long_msg_seqno(struct sk_buff *buf, u32 seqno)
2552 msg_set_seqno(buf_msg(buf), seqno);
2555 static inline u32 get_fragm_size(struct sk_buff *buf)
2557 return msg_ack(buf_msg(buf));
2560 static inline void set_fragm_size(struct sk_buff *buf, u32 sz)
2562 msg_set_ack(buf_msg(buf), sz);
2565 static inline u32 get_expected_frags(struct sk_buff *buf)
2567 return msg_bcast_ack(buf_msg(buf));
2570 static inline void set_expected_frags(struct sk_buff *buf, u32 exp)
2572 msg_set_bcast_ack(buf_msg(buf), exp);
2575 static inline u32 get_timer_cnt(struct sk_buff *buf)
2577 return msg_reroute_cnt(buf_msg(buf));
2580 static inline void incr_timer_cnt(struct sk_buff *buf)
2582 msg_incr_reroute_cnt(buf_msg(buf));
2586 * link_recv_fragment(): Called with node lock on. Returns
2587 * the reassembled buffer if message is complete.
2589 int link_recv_fragment(struct sk_buff **pending, struct sk_buff **fb,
2590 struct tipc_msg **m)
2592 struct sk_buff *prev = 0;
2593 struct sk_buff *fbuf = *fb;
2594 struct tipc_msg *fragm = buf_msg(fbuf);
2595 struct sk_buff *pbuf = *pending;
2596 u32 long_msg_seq_no = msg_long_msgno(fragm);
2599 msg_dbg(fragm,"FRG<REC<");
2601 /* Is there an incomplete message waiting for this fragment? */
2603 while (pbuf && ((msg_seqno(buf_msg(pbuf)) != long_msg_seq_no)
2604 || (msg_orignode(fragm) != msg_orignode(buf_msg(pbuf))))) {
2609 if (!pbuf && (msg_type(fragm) == FIRST_FRAGMENT)) {
2610 struct tipc_msg *imsg = (struct tipc_msg *)msg_data(fragm);
2611 u32 msg_sz = msg_size(imsg);
2612 u32 fragm_sz = msg_data_sz(fragm);
2613 u32 exp_fragm_cnt = msg_sz/fragm_sz + !!(msg_sz % fragm_sz);
2614 u32 max = TIPC_MAX_USER_MSG_SIZE + LONG_H_SIZE;
2615 if (msg_type(imsg) == TIPC_MCAST_MSG)
2616 max = TIPC_MAX_USER_MSG_SIZE + MCAST_H_SIZE;
2617 if (msg_size(imsg) > max) {
2618 msg_dbg(fragm,"<REC<Oversized: ");
2622 pbuf = buf_acquire(msg_size(imsg));
2624 pbuf->next = *pending;
2626 memcpy(pbuf->data, (unchar *)imsg, msg_data_sz(fragm));
2628 /* Prepare buffer for subsequent fragments. */
2630 set_long_msg_seqno(pbuf, long_msg_seq_no);
2631 set_fragm_size(pbuf,fragm_sz);
2632 set_expected_frags(pbuf,exp_fragm_cnt - 1);
2634 warn("Memory squeeze; got no defragmenting buffer\n");
2638 } else if (pbuf && (msg_type(fragm) != FIRST_FRAGMENT)) {
2639 u32 dsz = msg_data_sz(fragm);
2640 u32 fsz = get_fragm_size(pbuf);
2641 u32 crs = ((msg_fragm_no(fragm) - 1) * fsz);
2642 u32 exp_frags = get_expected_frags(pbuf) - 1;
2643 memcpy(pbuf->data + crs, msg_data(fragm), dsz);
2646 /* Is message complete? */
2648 if (exp_frags == 0) {
2650 prev->next = pbuf->next;
2652 *pending = pbuf->next;
2653 msg_reset_reroute_cnt(buf_msg(pbuf));
2658 set_expected_frags(pbuf,exp_frags);
2661 dbg(" Discarding orphan fragment %x\n",fbuf);
2662 msg_dbg(fragm,"ORPHAN:");
2663 dbg("Pending long buffers:\n");
2664 dbg_print_buf_chain(*pending);
2670 * link_check_defragm_bufs - flush stale incoming message fragments
2671 * @l_ptr: pointer to link
2674 static void link_check_defragm_bufs(struct link *l_ptr)
2676 struct sk_buff *prev = 0;
2677 struct sk_buff *next = 0;
2678 struct sk_buff *buf = l_ptr->defragm_buf;
2682 if (!link_working_working(l_ptr))
2685 u32 cnt = get_timer_cnt(buf);
2689 incr_timer_cnt(buf);
2692 dbg(" Discarding incomplete long buffer\n");
2693 msg_dbg(buf_msg(buf), "LONG:");
2694 dbg_print_link(l_ptr, "curr:");
2695 dbg("Pending long buffers:\n");
2696 dbg_print_buf_chain(l_ptr->defragm_buf);
2698 prev->next = buf->next;
2700 l_ptr->defragm_buf = buf->next;
2709 static void link_set_supervision_props(struct link *l_ptr, u32 tolerance)
2711 l_ptr->tolerance = tolerance;
2712 l_ptr->continuity_interval =
2713 ((tolerance / 4) > 500) ? 500 : tolerance / 4;
2714 l_ptr->abort_limit = tolerance / (l_ptr->continuity_interval / 4);
2718 void link_set_queue_limits(struct link *l_ptr, u32 window)
2720 /* Data messages from this node, inclusive FIRST_FRAGM */
2721 l_ptr->queue_limit[DATA_LOW] = window;
2722 l_ptr->queue_limit[DATA_MEDIUM] = (window / 3) * 4;
2723 l_ptr->queue_limit[DATA_HIGH] = (window / 3) * 5;
2724 l_ptr->queue_limit[DATA_CRITICAL] = (window / 3) * 6;
2725 /* Transiting data messages,inclusive FIRST_FRAGM */
2726 l_ptr->queue_limit[DATA_LOW + 4] = 300;
2727 l_ptr->queue_limit[DATA_MEDIUM + 4] = 600;
2728 l_ptr->queue_limit[DATA_HIGH + 4] = 900;
2729 l_ptr->queue_limit[DATA_CRITICAL + 4] = 1200;
2730 l_ptr->queue_limit[CONN_MANAGER] = 1200;
2731 l_ptr->queue_limit[ROUTE_DISTRIBUTOR] = 1200;
2732 l_ptr->queue_limit[CHANGEOVER_PROTOCOL] = 2500;
2733 l_ptr->queue_limit[NAME_DISTRIBUTOR] = 3000;
2734 /* FRAGMENT and LAST_FRAGMENT packets */
2735 l_ptr->queue_limit[MSG_FRAGMENTER] = 4000;
2739 * link_find_link - locate link by name
2740 * @name - ptr to link name string
2741 * @node - ptr to area to be filled with ptr to associated node
2743 * Caller must hold 'net_lock' to ensure node and bearer are not deleted;
2744 * this also prevents link deletion.
2746 * Returns pointer to link (or 0 if invalid link name).
2749 static struct link *link_find_link(const char *name, struct node **node)
2751 struct link_name link_name_parts;
2752 struct bearer *b_ptr;
2755 if (!link_name_validate(name, &link_name_parts))
2758 b_ptr = bearer_find_interface(link_name_parts.if_local);
2762 *node = node_find(link_name_parts.addr_peer);
2766 l_ptr = (*node)->links[b_ptr->identity];
2767 if (!l_ptr || strcmp(l_ptr->name, name))
2773 struct sk_buff *link_cmd_config(const void *req_tlv_area, int req_tlv_space,
2776 struct tipc_link_config *args;
2782 if (!TLV_CHECK(req_tlv_area, req_tlv_space, TIPC_TLV_LINK_CONFIG))
2783 return cfg_reply_error_string(TIPC_CFG_TLV_ERROR);
2785 args = (struct tipc_link_config *)TLV_DATA(req_tlv_area);
2786 new_value = ntohl(args->value);
2788 if (!strcmp(args->name, bc_link_name)) {
2789 if ((cmd == TIPC_CMD_SET_LINK_WINDOW) &&
2790 (bclink_set_queue_limits(new_value) == 0))
2791 return cfg_reply_none();
2792 return cfg_reply_error_string(TIPC_CFG_NOT_SUPPORTED
2793 " (cannot change setting on broadcast link)");
2796 read_lock_bh(&net_lock);
2797 l_ptr = link_find_link(args->name, &node);
2799 read_unlock_bh(&net_lock);
2800 return cfg_reply_error_string("link not found");
2806 case TIPC_CMD_SET_LINK_TOL:
2807 if ((new_value >= TIPC_MIN_LINK_TOL) &&
2808 (new_value <= TIPC_MAX_LINK_TOL)) {
2809 link_set_supervision_props(l_ptr, new_value);
2810 link_send_proto_msg(l_ptr, STATE_MSG,
2811 0, 0, new_value, 0, 0);
2815 case TIPC_CMD_SET_LINK_PRI:
2816 if (new_value < TIPC_NUM_LINK_PRI) {
2817 l_ptr->priority = new_value;
2818 link_send_proto_msg(l_ptr, STATE_MSG,
2819 0, 0, 0, new_value, 0);
2823 case TIPC_CMD_SET_LINK_WINDOW:
2824 if ((new_value >= TIPC_MIN_LINK_WIN) &&
2825 (new_value <= TIPC_MAX_LINK_WIN)) {
2826 link_set_queue_limits(l_ptr, new_value);
2833 read_unlock_bh(&net_lock);
2835 return cfg_reply_error_string("cannot change link setting");
2837 return cfg_reply_none();
2841 * link_reset_statistics - reset link statistics
2842 * @l_ptr: pointer to link
2845 static void link_reset_statistics(struct link *l_ptr)
2847 memset(&l_ptr->stats, 0, sizeof(l_ptr->stats));
2848 l_ptr->stats.sent_info = l_ptr->next_out_no;
2849 l_ptr->stats.recv_info = l_ptr->next_in_no;
2852 struct sk_buff *link_cmd_reset_stats(const void *req_tlv_area, int req_tlv_space)
2858 if (!TLV_CHECK(req_tlv_area, req_tlv_space, TIPC_TLV_LINK_NAME))
2859 return cfg_reply_error_string(TIPC_CFG_TLV_ERROR);
2861 link_name = (char *)TLV_DATA(req_tlv_area);
2862 if (!strcmp(link_name, bc_link_name)) {
2863 if (bclink_reset_stats())
2864 return cfg_reply_error_string("link not found");
2865 return cfg_reply_none();
2868 read_lock_bh(&net_lock);
2869 l_ptr = link_find_link(link_name, &node);
2871 read_unlock_bh(&net_lock);
2872 return cfg_reply_error_string("link not found");
2876 link_reset_statistics(l_ptr);
2878 read_unlock_bh(&net_lock);
2879 return cfg_reply_none();
2883 * percent - convert count to a percentage of total (rounding up or down)
2886 static u32 percent(u32 count, u32 total)
2888 return (count * 100 + (total / 2)) / total;
2892 * link_stats - print link statistics
2894 * @buf: print buffer area
2895 * @buf_size: size of print buffer area
2897 * Returns length of print buffer data string (or 0 if error)
2900 static int link_stats(const char *name, char *buf, const u32 buf_size)
2902 struct print_buf pb;
2906 u32 profile_total = 0;
2908 if (!strcmp(name, bc_link_name))
2909 return bclink_stats(buf, buf_size);
2911 printbuf_init(&pb, buf, buf_size);
2913 read_lock_bh(&net_lock);
2914 l_ptr = link_find_link(name, &node);
2916 read_unlock_bh(&net_lock);
2921 if (link_is_active(l_ptr))
2923 else if (link_is_up(l_ptr))
2927 tipc_printf(&pb, "Link <%s>\n"
2928 " %s MTU:%u Priority:%u Tolerance:%u ms"
2929 " Window:%u packets\n",
2930 l_ptr->name, status, link_max_pkt(l_ptr),
2931 l_ptr->priority, l_ptr->tolerance, l_ptr->queue_limit[0]);
2932 tipc_printf(&pb, " RX packets:%u fragments:%u/%u bundles:%u/%u\n",
2933 l_ptr->next_in_no - l_ptr->stats.recv_info,
2934 l_ptr->stats.recv_fragments,
2935 l_ptr->stats.recv_fragmented,
2936 l_ptr->stats.recv_bundles,
2937 l_ptr->stats.recv_bundled);
2938 tipc_printf(&pb, " TX packets:%u fragments:%u/%u bundles:%u/%u\n",
2939 l_ptr->next_out_no - l_ptr->stats.sent_info,
2940 l_ptr->stats.sent_fragments,
2941 l_ptr->stats.sent_fragmented,
2942 l_ptr->stats.sent_bundles,
2943 l_ptr->stats.sent_bundled);
2944 profile_total = l_ptr->stats.msg_length_counts;
2947 tipc_printf(&pb, " TX profile sample:%u packets average:%u octets\n"
2948 " 0-64:%u%% -256:%u%% -1024:%u%% -4096:%u%% "
2949 "-16354:%u%% -32768:%u%% -66000:%u%%\n",
2950 l_ptr->stats.msg_length_counts,
2951 l_ptr->stats.msg_lengths_total / profile_total,
2952 percent(l_ptr->stats.msg_length_profile[0], profile_total),
2953 percent(l_ptr->stats.msg_length_profile[1], profile_total),
2954 percent(l_ptr->stats.msg_length_profile[2], profile_total),
2955 percent(l_ptr->stats.msg_length_profile[3], profile_total),
2956 percent(l_ptr->stats.msg_length_profile[4], profile_total),
2957 percent(l_ptr->stats.msg_length_profile[5], profile_total),
2958 percent(l_ptr->stats.msg_length_profile[6], profile_total));
2959 tipc_printf(&pb, " RX states:%u probes:%u naks:%u defs:%u dups:%u\n",
2960 l_ptr->stats.recv_states,
2961 l_ptr->stats.recv_probes,
2962 l_ptr->stats.recv_nacks,
2963 l_ptr->stats.deferred_recv,
2964 l_ptr->stats.duplicates);
2965 tipc_printf(&pb, " TX states:%u probes:%u naks:%u acks:%u dups:%u\n",
2966 l_ptr->stats.sent_states,
2967 l_ptr->stats.sent_probes,
2968 l_ptr->stats.sent_nacks,
2969 l_ptr->stats.sent_acks,
2970 l_ptr->stats.retransmitted);
2971 tipc_printf(&pb, " Congestion bearer:%u link:%u Send queue max:%u avg:%u\n",
2972 l_ptr->stats.bearer_congs,
2973 l_ptr->stats.link_congs,
2974 l_ptr->stats.max_queue_sz,
2975 l_ptr->stats.queue_sz_counts
2976 ? (l_ptr->stats.accu_queue_sz / l_ptr->stats.queue_sz_counts)
2980 read_unlock_bh(&net_lock);
2981 return printbuf_validate(&pb);
2984 #define MAX_LINK_STATS_INFO 2000
2986 struct sk_buff *link_cmd_show_stats(const void *req_tlv_area, int req_tlv_space)
2988 struct sk_buff *buf;
2989 struct tlv_desc *rep_tlv;
2992 if (!TLV_CHECK(req_tlv_area, req_tlv_space, TIPC_TLV_LINK_NAME))
2993 return cfg_reply_error_string(TIPC_CFG_TLV_ERROR);
2995 buf = cfg_reply_alloc(TLV_SPACE(MAX_LINK_STATS_INFO));
2999 rep_tlv = (struct tlv_desc *)buf->data;
3001 str_len = link_stats((char *)TLV_DATA(req_tlv_area),
3002 (char *)TLV_DATA(rep_tlv), MAX_LINK_STATS_INFO);
3005 return cfg_reply_error_string("link not found");
3008 skb_put(buf, TLV_SPACE(str_len));
3009 TLV_SET(rep_tlv, TIPC_TLV_ULTRA_STRING, NULL, str_len);
3015 int link_control(const char *name, u32 op, u32 val)
3023 a = link_name2addr(name, &bearer_id);
3024 read_lock_bh(&net_lock);
3025 node = node_find(a);
3028 l_ptr = node->links[bearer_id];
3030 if (op == TIPC_REMOVE_LINK) {
3031 struct bearer *b_ptr = l_ptr->b_ptr;
3032 spin_lock_bh(&b_ptr->publ.lock);
3034 spin_unlock_bh(&b_ptr->publ.lock);
3036 if (op == TIPC_CMD_BLOCK_LINK) {
3040 if (op == TIPC_CMD_UNBLOCK_LINK) {
3047 read_unlock_bh(&net_lock);
3053 * link_get_max_pkt - get maximum packet size to use when sending to destination
3054 * @dest: network address of destination node
3055 * @selector: used to select from set of active links
3057 * If no active link can be found, uses default maximum packet size.
3060 u32 link_get_max_pkt(u32 dest, u32 selector)
3064 u32 res = MAX_PKT_DEFAULT;
3066 if (dest == tipc_own_addr)
3067 return MAX_MSG_SIZE;
3069 read_lock_bh(&net_lock);
3070 n_ptr = node_select(dest, selector);
3073 l_ptr = n_ptr->active_links[selector & 1];
3075 res = link_max_pkt(l_ptr);
3078 read_unlock_bh(&net_lock);
3083 static void link_dump_rec_queue(struct link *l_ptr)
3085 struct sk_buff *crs;
3087 if (!l_ptr->oldest_deferred_in) {
3088 info("Reception queue empty\n");
3091 info("Contents of Reception queue:\n");
3092 crs = l_ptr->oldest_deferred_in;
3094 if (crs->data == (void *)0x0000a3a3) {
3095 info("buffer %x invalid\n", crs);
3098 msg_dbg(buf_msg(crs), "In rec queue: \n");
3104 static void link_dump_send_queue(struct link *l_ptr)
3106 if (l_ptr->next_out) {
3107 info("\nContents of unsent queue:\n");
3108 dbg_print_buf_chain(l_ptr->next_out);
3110 info("\nContents of send queue:\n");
3111 if (l_ptr->first_out) {
3112 dbg_print_buf_chain(l_ptr->first_out);
3114 info("Empty send queue\n");
3117 static void link_print(struct link *l_ptr, struct print_buf *buf,
3120 tipc_printf(buf, str);
3121 if (link_reset_reset(l_ptr) || link_reset_unknown(l_ptr))
3123 tipc_printf(buf, "Link %x<%s>:",
3124 l_ptr->addr, l_ptr->b_ptr->publ.name);
3125 tipc_printf(buf, ": NXO(%u):", mod(l_ptr->next_out_no));
3126 tipc_printf(buf, "NXI(%u):", mod(l_ptr->next_in_no));
3127 tipc_printf(buf, "SQUE");
3128 if (l_ptr->first_out) {
3129 tipc_printf(buf, "[%u..", msg_seqno(buf_msg(l_ptr->first_out)));
3130 if (l_ptr->next_out)
3131 tipc_printf(buf, "%u..",
3132 msg_seqno(buf_msg(l_ptr->next_out)));
3133 tipc_printf(buf, "%u]",
3135 (l_ptr->last_out)), l_ptr->out_queue_size);
3136 if ((mod(msg_seqno(buf_msg(l_ptr->last_out)) -
3137 msg_seqno(buf_msg(l_ptr->first_out)))
3138 != (l_ptr->out_queue_size - 1))
3139 || (l_ptr->last_out->next != 0)) {
3140 tipc_printf(buf, "\nSend queue inconsistency\n");
3141 tipc_printf(buf, "first_out= %x ", l_ptr->first_out);
3142 tipc_printf(buf, "next_out= %x ", l_ptr->next_out);
3143 tipc_printf(buf, "last_out= %x ", l_ptr->last_out);
3144 link_dump_send_queue(l_ptr);
3147 tipc_printf(buf, "[]");
3148 tipc_printf(buf, "SQSIZ(%u)", l_ptr->out_queue_size);
3149 if (l_ptr->oldest_deferred_in) {
3150 u32 o = msg_seqno(buf_msg(l_ptr->oldest_deferred_in));
3151 u32 n = msg_seqno(buf_msg(l_ptr->newest_deferred_in));
3152 tipc_printf(buf, ":RQUE[%u..%u]", o, n);
3153 if (l_ptr->deferred_inqueue_sz != mod((n + 1) - o)) {
3154 tipc_printf(buf, ":RQSIZ(%u)",
3155 l_ptr->deferred_inqueue_sz);
3158 if (link_working_unknown(l_ptr))
3159 tipc_printf(buf, ":WU");
3160 if (link_reset_reset(l_ptr))
3161 tipc_printf(buf, ":RR");
3162 if (link_reset_unknown(l_ptr))
3163 tipc_printf(buf, ":RU");
3164 if (link_working_working(l_ptr))
3165 tipc_printf(buf, ":WW");
3166 tipc_printf(buf, "\n");