Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next
[pandora-kernel.git] / net / tipc / link.c
1 /*
2  * net/tipc/link.c: TIPC link code
3  *
4  * Copyright (c) 1996-2007, Ericsson AB
5  * Copyright (c) 2004-2007, 2010-2011, Wind River Systems
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the names of the copyright holders nor the names of its
17  *    contributors may be used to endorse or promote products derived from
18  *    this software without specific prior written permission.
19  *
20  * Alternatively, this software may be distributed under the terms of the
21  * GNU General Public License ("GPL") version 2 as published by the Free
22  * Software Foundation.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
28  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
31  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
32  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
33  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
34  * POSSIBILITY OF SUCH DAMAGE.
35  */
36
37 #include "core.h"
38 #include "link.h"
39 #include "port.h"
40 #include "name_distr.h"
41 #include "discover.h"
42 #include "config.h"
43
44
45 /*
46  * Out-of-range value for link session numbers
47  */
48
49 #define INVALID_SESSION 0x10000
50
51 /*
52  * Link state events:
53  */
54
55 #define  STARTING_EVT    856384768      /* link processing trigger */
56 #define  TRAFFIC_MSG_EVT 560815u        /* rx'd ??? */
57 #define  TIMEOUT_EVT     560817u        /* link timer expired */
58
59 /*
60  * The following two 'message types' is really just implementation
61  * data conveniently stored in the message header.
62  * They must not be considered part of the protocol
63  */
64 #define OPEN_MSG   0
65 #define CLOSED_MSG 1
66
67 /*
68  * State value stored in 'exp_msg_count'
69  */
70
71 #define START_CHANGEOVER 100000u
72
73 /**
74  * struct link_name - deconstructed link name
75  * @addr_local: network address of node at this end
76  * @if_local: name of interface at this end
77  * @addr_peer: network address of node at far end
78  * @if_peer: name of interface at far end
79  */
80
81 struct link_name {
82         u32 addr_local;
83         char if_local[TIPC_MAX_IF_NAME];
84         u32 addr_peer;
85         char if_peer[TIPC_MAX_IF_NAME];
86 };
87
88 static void link_handle_out_of_seq_msg(struct link *l_ptr,
89                                        struct sk_buff *buf);
90 static void link_recv_proto_msg(struct link *l_ptr, struct sk_buff *buf);
91 static int  link_recv_changeover_msg(struct link **l_ptr, struct sk_buff **buf);
92 static void link_set_supervision_props(struct link *l_ptr, u32 tolerance);
93 static int  link_send_sections_long(struct tipc_port *sender,
94                                     struct iovec const *msg_sect,
95                                     u32 num_sect, unsigned int total_len,
96                                     u32 destnode);
97 static void link_check_defragm_bufs(struct link *l_ptr);
98 static void link_state_event(struct link *l_ptr, u32 event);
99 static void link_reset_statistics(struct link *l_ptr);
100 static void link_print(struct link *l_ptr, const char *str);
101 static void link_start(struct link *l_ptr);
102 static int link_send_long_buf(struct link *l_ptr, struct sk_buff *buf);
103
104 /*
105  *  Simple link routines
106  */
107
108 static unsigned int align(unsigned int i)
109 {
110         return (i + 3) & ~3u;
111 }
112
113 static void link_init_max_pkt(struct link *l_ptr)
114 {
115         u32 max_pkt;
116
117         max_pkt = (l_ptr->b_ptr->mtu & ~3);
118         if (max_pkt > MAX_MSG_SIZE)
119                 max_pkt = MAX_MSG_SIZE;
120
121         l_ptr->max_pkt_target = max_pkt;
122         if (l_ptr->max_pkt_target < MAX_PKT_DEFAULT)
123                 l_ptr->max_pkt = l_ptr->max_pkt_target;
124         else
125                 l_ptr->max_pkt = MAX_PKT_DEFAULT;
126
127         l_ptr->max_pkt_probes = 0;
128 }
129
130 static u32 link_next_sent(struct link *l_ptr)
131 {
132         if (l_ptr->next_out)
133                 return msg_seqno(buf_msg(l_ptr->next_out));
134         return mod(l_ptr->next_out_no);
135 }
136
137 static u32 link_last_sent(struct link *l_ptr)
138 {
139         return mod(link_next_sent(l_ptr) - 1);
140 }
141
142 /*
143  *  Simple non-static link routines (i.e. referenced outside this file)
144  */
145
146 int tipc_link_is_up(struct link *l_ptr)
147 {
148         if (!l_ptr)
149                 return 0;
150         return link_working_working(l_ptr) || link_working_unknown(l_ptr);
151 }
152
153 int tipc_link_is_active(struct link *l_ptr)
154 {
155         return  (l_ptr->owner->active_links[0] == l_ptr) ||
156                 (l_ptr->owner->active_links[1] == l_ptr);
157 }
158
159 /**
160  * link_name_validate - validate & (optionally) deconstruct link name
161  * @name - ptr to link name string
162  * @name_parts - ptr to area for link name components (or NULL if not needed)
163  *
164  * Returns 1 if link name is valid, otherwise 0.
165  */
166
167 static int link_name_validate(const char *name, struct link_name *name_parts)
168 {
169         char name_copy[TIPC_MAX_LINK_NAME];
170         char *addr_local;
171         char *if_local;
172         char *addr_peer;
173         char *if_peer;
174         char dummy;
175         u32 z_local, c_local, n_local;
176         u32 z_peer, c_peer, n_peer;
177         u32 if_local_len;
178         u32 if_peer_len;
179
180         /* copy link name & ensure length is OK */
181
182         name_copy[TIPC_MAX_LINK_NAME - 1] = 0;
183         /* need above in case non-Posix strncpy() doesn't pad with nulls */
184         strncpy(name_copy, name, TIPC_MAX_LINK_NAME);
185         if (name_copy[TIPC_MAX_LINK_NAME - 1] != 0)
186                 return 0;
187
188         /* ensure all component parts of link name are present */
189
190         addr_local = name_copy;
191         if_local = strchr(addr_local, ':');
192         if (if_local == NULL)
193                 return 0;
194         *(if_local++) = 0;
195         addr_peer = strchr(if_local, '-');
196         if (addr_peer == NULL)
197                 return 0;
198         *(addr_peer++) = 0;
199         if_local_len = addr_peer - if_local;
200         if_peer = strchr(addr_peer, ':');
201         if (if_peer == NULL)
202                 return 0;
203         *(if_peer++) = 0;
204         if_peer_len = strlen(if_peer) + 1;
205
206         /* validate component parts of link name */
207
208         if ((sscanf(addr_local, "%u.%u.%u%c",
209                     &z_local, &c_local, &n_local, &dummy) != 3) ||
210             (sscanf(addr_peer, "%u.%u.%u%c",
211                     &z_peer, &c_peer, &n_peer, &dummy) != 3) ||
212             (z_local > 255) || (c_local > 4095) || (n_local > 4095) ||
213             (z_peer  > 255) || (c_peer  > 4095) || (n_peer  > 4095) ||
214             (if_local_len <= 1) || (if_local_len > TIPC_MAX_IF_NAME) ||
215             (if_peer_len  <= 1) || (if_peer_len  > TIPC_MAX_IF_NAME) ||
216             (strspn(if_local, tipc_alphabet) != (if_local_len - 1)) ||
217             (strspn(if_peer, tipc_alphabet) != (if_peer_len - 1)))
218                 return 0;
219
220         /* return link name components, if necessary */
221
222         if (name_parts) {
223                 name_parts->addr_local = tipc_addr(z_local, c_local, n_local);
224                 strcpy(name_parts->if_local, if_local);
225                 name_parts->addr_peer = tipc_addr(z_peer, c_peer, n_peer);
226                 strcpy(name_parts->if_peer, if_peer);
227         }
228         return 1;
229 }
230
231 /**
232  * link_timeout - handle expiration of link timer
233  * @l_ptr: pointer to link
234  *
235  * This routine must not grab "tipc_net_lock" to avoid a potential deadlock conflict
236  * with tipc_link_delete().  (There is no risk that the node will be deleted by
237  * another thread because tipc_link_delete() always cancels the link timer before
238  * tipc_node_delete() is called.)
239  */
240
241 static void link_timeout(struct link *l_ptr)
242 {
243         tipc_node_lock(l_ptr->owner);
244
245         /* update counters used in statistical profiling of send traffic */
246
247         l_ptr->stats.accu_queue_sz += l_ptr->out_queue_size;
248         l_ptr->stats.queue_sz_counts++;
249
250         if (l_ptr->first_out) {
251                 struct tipc_msg *msg = buf_msg(l_ptr->first_out);
252                 u32 length = msg_size(msg);
253
254                 if ((msg_user(msg) == MSG_FRAGMENTER) &&
255                     (msg_type(msg) == FIRST_FRAGMENT)) {
256                         length = msg_size(msg_get_wrapped(msg));
257                 }
258                 if (length) {
259                         l_ptr->stats.msg_lengths_total += length;
260                         l_ptr->stats.msg_length_counts++;
261                         if (length <= 64)
262                                 l_ptr->stats.msg_length_profile[0]++;
263                         else if (length <= 256)
264                                 l_ptr->stats.msg_length_profile[1]++;
265                         else if (length <= 1024)
266                                 l_ptr->stats.msg_length_profile[2]++;
267                         else if (length <= 4096)
268                                 l_ptr->stats.msg_length_profile[3]++;
269                         else if (length <= 16384)
270                                 l_ptr->stats.msg_length_profile[4]++;
271                         else if (length <= 32768)
272                                 l_ptr->stats.msg_length_profile[5]++;
273                         else
274                                 l_ptr->stats.msg_length_profile[6]++;
275                 }
276         }
277
278         /* do all other link processing performed on a periodic basis */
279
280         link_check_defragm_bufs(l_ptr);
281
282         link_state_event(l_ptr, TIMEOUT_EVT);
283
284         if (l_ptr->next_out)
285                 tipc_link_push_queue(l_ptr);
286
287         tipc_node_unlock(l_ptr->owner);
288 }
289
290 static void link_set_timer(struct link *l_ptr, u32 time)
291 {
292         k_start_timer(&l_ptr->timer, time);
293 }
294
295 /**
296  * tipc_link_create - create a new link
297  * @n_ptr: pointer to associated node
298  * @b_ptr: pointer to associated bearer
299  * @media_addr: media address to use when sending messages over link
300  *
301  * Returns pointer to link.
302  */
303
304 struct link *tipc_link_create(struct tipc_node *n_ptr,
305                               struct tipc_bearer *b_ptr,
306                               const struct tipc_media_addr *media_addr)
307 {
308         struct link *l_ptr;
309         struct tipc_msg *msg;
310         char *if_name;
311         char addr_string[16];
312         u32 peer = n_ptr->addr;
313
314         if (n_ptr->link_cnt >= 2) {
315                 tipc_addr_string_fill(addr_string, n_ptr->addr);
316                 err("Attempt to establish third link to %s\n", addr_string);
317                 return NULL;
318         }
319
320         if (n_ptr->links[b_ptr->identity]) {
321                 tipc_addr_string_fill(addr_string, n_ptr->addr);
322                 err("Attempt to establish second link on <%s> to %s\n",
323                     b_ptr->name, addr_string);
324                 return NULL;
325         }
326
327         l_ptr = kzalloc(sizeof(*l_ptr), GFP_ATOMIC);
328         if (!l_ptr) {
329                 warn("Link creation failed, no memory\n");
330                 return NULL;
331         }
332
333         l_ptr->addr = peer;
334         if_name = strchr(b_ptr->name, ':') + 1;
335         sprintf(l_ptr->name, "%u.%u.%u:%s-%u.%u.%u:",
336                 tipc_zone(tipc_own_addr), tipc_cluster(tipc_own_addr),
337                 tipc_node(tipc_own_addr),
338                 if_name,
339                 tipc_zone(peer), tipc_cluster(peer), tipc_node(peer));
340                 /* note: peer i/f is appended to link name by reset/activate */
341         memcpy(&l_ptr->media_addr, media_addr, sizeof(*media_addr));
342         l_ptr->owner = n_ptr;
343         l_ptr->checkpoint = 1;
344         l_ptr->b_ptr = b_ptr;
345         link_set_supervision_props(l_ptr, b_ptr->media->tolerance);
346         l_ptr->state = RESET_UNKNOWN;
347
348         l_ptr->pmsg = (struct tipc_msg *)&l_ptr->proto_msg;
349         msg = l_ptr->pmsg;
350         tipc_msg_init(msg, LINK_PROTOCOL, RESET_MSG, INT_H_SIZE, l_ptr->addr);
351         msg_set_size(msg, sizeof(l_ptr->proto_msg));
352         msg_set_session(msg, (tipc_random & 0xffff));
353         msg_set_bearer_id(msg, b_ptr->identity);
354         strcpy((char *)msg_data(msg), if_name);
355
356         l_ptr->priority = b_ptr->priority;
357         tipc_link_set_queue_limits(l_ptr, b_ptr->media->window);
358
359         link_init_max_pkt(l_ptr);
360
361         l_ptr->next_out_no = 1;
362         INIT_LIST_HEAD(&l_ptr->waiting_ports);
363
364         link_reset_statistics(l_ptr);
365
366         tipc_node_attach_link(n_ptr, l_ptr);
367
368         k_init_timer(&l_ptr->timer, (Handler)link_timeout, (unsigned long)l_ptr);
369         list_add_tail(&l_ptr->link_list, &b_ptr->links);
370         tipc_k_signal((Handler)link_start, (unsigned long)l_ptr);
371
372         return l_ptr;
373 }
374
375 /**
376  * tipc_link_delete - delete a link
377  * @l_ptr: pointer to link
378  *
379  * Note: 'tipc_net_lock' is write_locked, bearer is locked.
380  * This routine must not grab the node lock until after link timer cancellation
381  * to avoid a potential deadlock situation.
382  */
383
384 void tipc_link_delete(struct link *l_ptr)
385 {
386         if (!l_ptr) {
387                 err("Attempt to delete non-existent link\n");
388                 return;
389         }
390
391         k_cancel_timer(&l_ptr->timer);
392
393         tipc_node_lock(l_ptr->owner);
394         tipc_link_reset(l_ptr);
395         tipc_node_detach_link(l_ptr->owner, l_ptr);
396         tipc_link_stop(l_ptr);
397         list_del_init(&l_ptr->link_list);
398         tipc_node_unlock(l_ptr->owner);
399         k_term_timer(&l_ptr->timer);
400         kfree(l_ptr);
401 }
402
403 static void link_start(struct link *l_ptr)
404 {
405         tipc_node_lock(l_ptr->owner);
406         link_state_event(l_ptr, STARTING_EVT);
407         tipc_node_unlock(l_ptr->owner);
408 }
409
410 /**
411  * link_schedule_port - schedule port for deferred sending
412  * @l_ptr: pointer to link
413  * @origport: reference to sending port
414  * @sz: amount of data to be sent
415  *
416  * Schedules port for renewed sending of messages after link congestion
417  * has abated.
418  */
419
420 static int link_schedule_port(struct link *l_ptr, u32 origport, u32 sz)
421 {
422         struct tipc_port *p_ptr;
423
424         spin_lock_bh(&tipc_port_list_lock);
425         p_ptr = tipc_port_lock(origport);
426         if (p_ptr) {
427                 if (!p_ptr->wakeup)
428                         goto exit;
429                 if (!list_empty(&p_ptr->wait_list))
430                         goto exit;
431                 p_ptr->congested = 1;
432                 p_ptr->waiting_pkts = 1 + ((sz - 1) / l_ptr->max_pkt);
433                 list_add_tail(&p_ptr->wait_list, &l_ptr->waiting_ports);
434                 l_ptr->stats.link_congs++;
435 exit:
436                 tipc_port_unlock(p_ptr);
437         }
438         spin_unlock_bh(&tipc_port_list_lock);
439         return -ELINKCONG;
440 }
441
442 void tipc_link_wakeup_ports(struct link *l_ptr, int all)
443 {
444         struct tipc_port *p_ptr;
445         struct tipc_port *temp_p_ptr;
446         int win = l_ptr->queue_limit[0] - l_ptr->out_queue_size;
447
448         if (all)
449                 win = 100000;
450         if (win <= 0)
451                 return;
452         if (!spin_trylock_bh(&tipc_port_list_lock))
453                 return;
454         if (link_congested(l_ptr))
455                 goto exit;
456         list_for_each_entry_safe(p_ptr, temp_p_ptr, &l_ptr->waiting_ports,
457                                  wait_list) {
458                 if (win <= 0)
459                         break;
460                 list_del_init(&p_ptr->wait_list);
461                 spin_lock_bh(p_ptr->lock);
462                 p_ptr->congested = 0;
463                 p_ptr->wakeup(p_ptr);
464                 win -= p_ptr->waiting_pkts;
465                 spin_unlock_bh(p_ptr->lock);
466         }
467
468 exit:
469         spin_unlock_bh(&tipc_port_list_lock);
470 }
471
472 /**
473  * link_release_outqueue - purge link's outbound message queue
474  * @l_ptr: pointer to link
475  */
476
477 static void link_release_outqueue(struct link *l_ptr)
478 {
479         struct sk_buff *buf = l_ptr->first_out;
480         struct sk_buff *next;
481
482         while (buf) {
483                 next = buf->next;
484                 buf_discard(buf);
485                 buf = next;
486         }
487         l_ptr->first_out = NULL;
488         l_ptr->out_queue_size = 0;
489 }
490
491 /**
492  * tipc_link_reset_fragments - purge link's inbound message fragments queue
493  * @l_ptr: pointer to link
494  */
495
496 void tipc_link_reset_fragments(struct link *l_ptr)
497 {
498         struct sk_buff *buf = l_ptr->defragm_buf;
499         struct sk_buff *next;
500
501         while (buf) {
502                 next = buf->next;
503                 buf_discard(buf);
504                 buf = next;
505         }
506         l_ptr->defragm_buf = NULL;
507 }
508
509 /**
510  * tipc_link_stop - purge all inbound and outbound messages associated with link
511  * @l_ptr: pointer to link
512  */
513
514 void tipc_link_stop(struct link *l_ptr)
515 {
516         struct sk_buff *buf;
517         struct sk_buff *next;
518
519         buf = l_ptr->oldest_deferred_in;
520         while (buf) {
521                 next = buf->next;
522                 buf_discard(buf);
523                 buf = next;
524         }
525
526         buf = l_ptr->first_out;
527         while (buf) {
528                 next = buf->next;
529                 buf_discard(buf);
530                 buf = next;
531         }
532
533         tipc_link_reset_fragments(l_ptr);
534
535         buf_discard(l_ptr->proto_msg_queue);
536         l_ptr->proto_msg_queue = NULL;
537 }
538
539 /* LINK EVENT CODE IS NOT SUPPORTED AT PRESENT */
540 #define link_send_event(fcn, l_ptr, up) do { } while (0)
541
542 void tipc_link_reset(struct link *l_ptr)
543 {
544         struct sk_buff *buf;
545         u32 prev_state = l_ptr->state;
546         u32 checkpoint = l_ptr->next_in_no;
547         int was_active_link = tipc_link_is_active(l_ptr);
548
549         msg_set_session(l_ptr->pmsg, ((msg_session(l_ptr->pmsg) + 1) & 0xffff));
550
551         /* Link is down, accept any session */
552         l_ptr->peer_session = INVALID_SESSION;
553
554         /* Prepare for max packet size negotiation */
555         link_init_max_pkt(l_ptr);
556
557         l_ptr->state = RESET_UNKNOWN;
558
559         if ((prev_state == RESET_UNKNOWN) || (prev_state == RESET_RESET))
560                 return;
561
562         tipc_node_link_down(l_ptr->owner, l_ptr);
563         tipc_bearer_remove_dest(l_ptr->b_ptr, l_ptr->addr);
564
565         if (was_active_link && tipc_node_active_links(l_ptr->owner) &&
566             l_ptr->owner->permit_changeover) {
567                 l_ptr->reset_checkpoint = checkpoint;
568                 l_ptr->exp_msg_count = START_CHANGEOVER;
569         }
570
571         /* Clean up all queues: */
572
573         link_release_outqueue(l_ptr);
574         buf_discard(l_ptr->proto_msg_queue);
575         l_ptr->proto_msg_queue = NULL;
576         buf = l_ptr->oldest_deferred_in;
577         while (buf) {
578                 struct sk_buff *next = buf->next;
579                 buf_discard(buf);
580                 buf = next;
581         }
582         if (!list_empty(&l_ptr->waiting_ports))
583                 tipc_link_wakeup_ports(l_ptr, 1);
584
585         l_ptr->retransm_queue_head = 0;
586         l_ptr->retransm_queue_size = 0;
587         l_ptr->last_out = NULL;
588         l_ptr->first_out = NULL;
589         l_ptr->next_out = NULL;
590         l_ptr->unacked_window = 0;
591         l_ptr->checkpoint = 1;
592         l_ptr->next_out_no = 1;
593         l_ptr->deferred_inqueue_sz = 0;
594         l_ptr->oldest_deferred_in = NULL;
595         l_ptr->newest_deferred_in = NULL;
596         l_ptr->fsm_msg_cnt = 0;
597         l_ptr->stale_count = 0;
598         link_reset_statistics(l_ptr);
599
600         link_send_event(tipc_cfg_link_event, l_ptr, 0);
601         if (!in_own_cluster(l_ptr->addr))
602                 link_send_event(tipc_disc_link_event, l_ptr, 0);
603 }
604
605
606 static void link_activate(struct link *l_ptr)
607 {
608         l_ptr->next_in_no = l_ptr->stats.recv_info = 1;
609         tipc_node_link_up(l_ptr->owner, l_ptr);
610         tipc_bearer_add_dest(l_ptr->b_ptr, l_ptr->addr);
611         link_send_event(tipc_cfg_link_event, l_ptr, 1);
612         if (!in_own_cluster(l_ptr->addr))
613                 link_send_event(tipc_disc_link_event, l_ptr, 1);
614 }
615
616 /**
617  * link_state_event - link finite state machine
618  * @l_ptr: pointer to link
619  * @event: state machine event to process
620  */
621
622 static void link_state_event(struct link *l_ptr, unsigned event)
623 {
624         struct link *other;
625         u32 cont_intv = l_ptr->continuity_interval;
626
627         if (!l_ptr->started && (event != STARTING_EVT))
628                 return;         /* Not yet. */
629
630         if (link_blocked(l_ptr)) {
631                 if (event == TIMEOUT_EVT)
632                         link_set_timer(l_ptr, cont_intv);
633                 return;   /* Changeover going on */
634         }
635
636         switch (l_ptr->state) {
637         case WORKING_WORKING:
638                 switch (event) {
639                 case TRAFFIC_MSG_EVT:
640                 case ACTIVATE_MSG:
641                         break;
642                 case TIMEOUT_EVT:
643                         if (l_ptr->next_in_no != l_ptr->checkpoint) {
644                                 l_ptr->checkpoint = l_ptr->next_in_no;
645                                 if (tipc_bclink_acks_missing(l_ptr->owner)) {
646                                         tipc_link_send_proto_msg(l_ptr, STATE_MSG,
647                                                                  0, 0, 0, 0, 0);
648                                         l_ptr->fsm_msg_cnt++;
649                                 } else if (l_ptr->max_pkt < l_ptr->max_pkt_target) {
650                                         tipc_link_send_proto_msg(l_ptr, STATE_MSG,
651                                                                  1, 0, 0, 0, 0);
652                                         l_ptr->fsm_msg_cnt++;
653                                 }
654                                 link_set_timer(l_ptr, cont_intv);
655                                 break;
656                         }
657                         l_ptr->state = WORKING_UNKNOWN;
658                         l_ptr->fsm_msg_cnt = 0;
659                         tipc_link_send_proto_msg(l_ptr, STATE_MSG, 1, 0, 0, 0, 0);
660                         l_ptr->fsm_msg_cnt++;
661                         link_set_timer(l_ptr, cont_intv / 4);
662                         break;
663                 case RESET_MSG:
664                         info("Resetting link <%s>, requested by peer\n",
665                              l_ptr->name);
666                         tipc_link_reset(l_ptr);
667                         l_ptr->state = RESET_RESET;
668                         l_ptr->fsm_msg_cnt = 0;
669                         tipc_link_send_proto_msg(l_ptr, ACTIVATE_MSG, 0, 0, 0, 0, 0);
670                         l_ptr->fsm_msg_cnt++;
671                         link_set_timer(l_ptr, cont_intv);
672                         break;
673                 default:
674                         err("Unknown link event %u in WW state\n", event);
675                 }
676                 break;
677         case WORKING_UNKNOWN:
678                 switch (event) {
679                 case TRAFFIC_MSG_EVT:
680                 case ACTIVATE_MSG:
681                         l_ptr->state = WORKING_WORKING;
682                         l_ptr->fsm_msg_cnt = 0;
683                         link_set_timer(l_ptr, cont_intv);
684                         break;
685                 case RESET_MSG:
686                         info("Resetting link <%s>, requested by peer "
687                              "while probing\n", l_ptr->name);
688                         tipc_link_reset(l_ptr);
689                         l_ptr->state = RESET_RESET;
690                         l_ptr->fsm_msg_cnt = 0;
691                         tipc_link_send_proto_msg(l_ptr, ACTIVATE_MSG, 0, 0, 0, 0, 0);
692                         l_ptr->fsm_msg_cnt++;
693                         link_set_timer(l_ptr, cont_intv);
694                         break;
695                 case TIMEOUT_EVT:
696                         if (l_ptr->next_in_no != l_ptr->checkpoint) {
697                                 l_ptr->state = WORKING_WORKING;
698                                 l_ptr->fsm_msg_cnt = 0;
699                                 l_ptr->checkpoint = l_ptr->next_in_no;
700                                 if (tipc_bclink_acks_missing(l_ptr->owner)) {
701                                         tipc_link_send_proto_msg(l_ptr, STATE_MSG,
702                                                                  0, 0, 0, 0, 0);
703                                         l_ptr->fsm_msg_cnt++;
704                                 }
705                                 link_set_timer(l_ptr, cont_intv);
706                         } else if (l_ptr->fsm_msg_cnt < l_ptr->abort_limit) {
707                                 tipc_link_send_proto_msg(l_ptr, STATE_MSG,
708                                                          1, 0, 0, 0, 0);
709                                 l_ptr->fsm_msg_cnt++;
710                                 link_set_timer(l_ptr, cont_intv / 4);
711                         } else {        /* Link has failed */
712                                 warn("Resetting link <%s>, peer not responding\n",
713                                      l_ptr->name);
714                                 tipc_link_reset(l_ptr);
715                                 l_ptr->state = RESET_UNKNOWN;
716                                 l_ptr->fsm_msg_cnt = 0;
717                                 tipc_link_send_proto_msg(l_ptr, RESET_MSG,
718                                                          0, 0, 0, 0, 0);
719                                 l_ptr->fsm_msg_cnt++;
720                                 link_set_timer(l_ptr, cont_intv);
721                         }
722                         break;
723                 default:
724                         err("Unknown link event %u in WU state\n", event);
725                 }
726                 break;
727         case RESET_UNKNOWN:
728                 switch (event) {
729                 case TRAFFIC_MSG_EVT:
730                         break;
731                 case ACTIVATE_MSG:
732                         other = l_ptr->owner->active_links[0];
733                         if (other && link_working_unknown(other))
734                                 break;
735                         l_ptr->state = WORKING_WORKING;
736                         l_ptr->fsm_msg_cnt = 0;
737                         link_activate(l_ptr);
738                         tipc_link_send_proto_msg(l_ptr, STATE_MSG, 1, 0, 0, 0, 0);
739                         l_ptr->fsm_msg_cnt++;
740                         link_set_timer(l_ptr, cont_intv);
741                         break;
742                 case RESET_MSG:
743                         l_ptr->state = RESET_RESET;
744                         l_ptr->fsm_msg_cnt = 0;
745                         tipc_link_send_proto_msg(l_ptr, ACTIVATE_MSG, 1, 0, 0, 0, 0);
746                         l_ptr->fsm_msg_cnt++;
747                         link_set_timer(l_ptr, cont_intv);
748                         break;
749                 case STARTING_EVT:
750                         l_ptr->started = 1;
751                         /* fall through */
752                 case TIMEOUT_EVT:
753                         tipc_link_send_proto_msg(l_ptr, RESET_MSG, 0, 0, 0, 0, 0);
754                         l_ptr->fsm_msg_cnt++;
755                         link_set_timer(l_ptr, cont_intv);
756                         break;
757                 default:
758                         err("Unknown link event %u in RU state\n", event);
759                 }
760                 break;
761         case RESET_RESET:
762                 switch (event) {
763                 case TRAFFIC_MSG_EVT:
764                 case ACTIVATE_MSG:
765                         other = l_ptr->owner->active_links[0];
766                         if (other && link_working_unknown(other))
767                                 break;
768                         l_ptr->state = WORKING_WORKING;
769                         l_ptr->fsm_msg_cnt = 0;
770                         link_activate(l_ptr);
771                         tipc_link_send_proto_msg(l_ptr, STATE_MSG, 1, 0, 0, 0, 0);
772                         l_ptr->fsm_msg_cnt++;
773                         link_set_timer(l_ptr, cont_intv);
774                         break;
775                 case RESET_MSG:
776                         break;
777                 case TIMEOUT_EVT:
778                         tipc_link_send_proto_msg(l_ptr, ACTIVATE_MSG, 0, 0, 0, 0, 0);
779                         l_ptr->fsm_msg_cnt++;
780                         link_set_timer(l_ptr, cont_intv);
781                         break;
782                 default:
783                         err("Unknown link event %u in RR state\n", event);
784                 }
785                 break;
786         default:
787                 err("Unknown link state %u/%u\n", l_ptr->state, event);
788         }
789 }
790
791 /*
792  * link_bundle_buf(): Append contents of a buffer to
793  * the tail of an existing one.
794  */
795
796 static int link_bundle_buf(struct link *l_ptr,
797                            struct sk_buff *bundler,
798                            struct sk_buff *buf)
799 {
800         struct tipc_msg *bundler_msg = buf_msg(bundler);
801         struct tipc_msg *msg = buf_msg(buf);
802         u32 size = msg_size(msg);
803         u32 bundle_size = msg_size(bundler_msg);
804         u32 to_pos = align(bundle_size);
805         u32 pad = to_pos - bundle_size;
806
807         if (msg_user(bundler_msg) != MSG_BUNDLER)
808                 return 0;
809         if (msg_type(bundler_msg) != OPEN_MSG)
810                 return 0;
811         if (skb_tailroom(bundler) < (pad + size))
812                 return 0;
813         if (l_ptr->max_pkt < (to_pos + size))
814                 return 0;
815
816         skb_put(bundler, pad + size);
817         skb_copy_to_linear_data_offset(bundler, to_pos, buf->data, size);
818         msg_set_size(bundler_msg, to_pos + size);
819         msg_set_msgcnt(bundler_msg, msg_msgcnt(bundler_msg) + 1);
820         buf_discard(buf);
821         l_ptr->stats.sent_bundled++;
822         return 1;
823 }
824
825 static void link_add_to_outqueue(struct link *l_ptr,
826                                  struct sk_buff *buf,
827                                  struct tipc_msg *msg)
828 {
829         u32 ack = mod(l_ptr->next_in_no - 1);
830         u32 seqno = mod(l_ptr->next_out_no++);
831
832         msg_set_word(msg, 2, ((ack << 16) | seqno));
833         msg_set_bcast_ack(msg, l_ptr->owner->bclink.last_in);
834         buf->next = NULL;
835         if (l_ptr->first_out) {
836                 l_ptr->last_out->next = buf;
837                 l_ptr->last_out = buf;
838         } else
839                 l_ptr->first_out = l_ptr->last_out = buf;
840
841         l_ptr->out_queue_size++;
842         if (l_ptr->out_queue_size > l_ptr->stats.max_queue_sz)
843                 l_ptr->stats.max_queue_sz = l_ptr->out_queue_size;
844 }
845
846 static void link_add_chain_to_outqueue(struct link *l_ptr,
847                                        struct sk_buff *buf_chain,
848                                        u32 long_msgno)
849 {
850         struct sk_buff *buf;
851         struct tipc_msg *msg;
852
853         if (!l_ptr->next_out)
854                 l_ptr->next_out = buf_chain;
855         while (buf_chain) {
856                 buf = buf_chain;
857                 buf_chain = buf_chain->next;
858
859                 msg = buf_msg(buf);
860                 msg_set_long_msgno(msg, long_msgno);
861                 link_add_to_outqueue(l_ptr, buf, msg);
862         }
863 }
864
865 /*
866  * tipc_link_send_buf() is the 'full path' for messages, called from
867  * inside TIPC when the 'fast path' in tipc_send_buf
868  * has failed, and from link_send()
869  */
870
871 int tipc_link_send_buf(struct link *l_ptr, struct sk_buff *buf)
872 {
873         struct tipc_msg *msg = buf_msg(buf);
874         u32 size = msg_size(msg);
875         u32 dsz = msg_data_sz(msg);
876         u32 queue_size = l_ptr->out_queue_size;
877         u32 imp = tipc_msg_tot_importance(msg);
878         u32 queue_limit = l_ptr->queue_limit[imp];
879         u32 max_packet = l_ptr->max_pkt;
880
881         msg_set_prevnode(msg, tipc_own_addr);   /* If routed message */
882
883         /* Match msg importance against queue limits: */
884
885         if (unlikely(queue_size >= queue_limit)) {
886                 if (imp <= TIPC_CRITICAL_IMPORTANCE) {
887                         link_schedule_port(l_ptr, msg_origport(msg), size);
888                         buf_discard(buf);
889                         return -ELINKCONG;
890                 }
891                 buf_discard(buf);
892                 if (imp > CONN_MANAGER) {
893                         warn("Resetting link <%s>, send queue full", l_ptr->name);
894                         tipc_link_reset(l_ptr);
895                 }
896                 return dsz;
897         }
898
899         /* Fragmentation needed ? */
900
901         if (size > max_packet)
902                 return link_send_long_buf(l_ptr, buf);
903
904         /* Packet can be queued or sent: */
905
906         if (likely(!tipc_bearer_congested(l_ptr->b_ptr, l_ptr) &&
907                    !link_congested(l_ptr))) {
908                 link_add_to_outqueue(l_ptr, buf, msg);
909
910                 if (likely(tipc_bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr))) {
911                         l_ptr->unacked_window = 0;
912                 } else {
913                         tipc_bearer_schedule(l_ptr->b_ptr, l_ptr);
914                         l_ptr->stats.bearer_congs++;
915                         l_ptr->next_out = buf;
916                 }
917                 return dsz;
918         }
919         /* Congestion: can message be bundled ?: */
920
921         if ((msg_user(msg) != CHANGEOVER_PROTOCOL) &&
922             (msg_user(msg) != MSG_FRAGMENTER)) {
923
924                 /* Try adding message to an existing bundle */
925
926                 if (l_ptr->next_out &&
927                     link_bundle_buf(l_ptr, l_ptr->last_out, buf)) {
928                         tipc_bearer_resolve_congestion(l_ptr->b_ptr, l_ptr);
929                         return dsz;
930                 }
931
932                 /* Try creating a new bundle */
933
934                 if (size <= max_packet * 2 / 3) {
935                         struct sk_buff *bundler = tipc_buf_acquire(max_packet);
936                         struct tipc_msg bundler_hdr;
937
938                         if (bundler) {
939                                 tipc_msg_init(&bundler_hdr, MSG_BUNDLER, OPEN_MSG,
940                                          INT_H_SIZE, l_ptr->addr);
941                                 skb_copy_to_linear_data(bundler, &bundler_hdr,
942                                                         INT_H_SIZE);
943                                 skb_trim(bundler, INT_H_SIZE);
944                                 link_bundle_buf(l_ptr, bundler, buf);
945                                 buf = bundler;
946                                 msg = buf_msg(buf);
947                                 l_ptr->stats.sent_bundles++;
948                         }
949                 }
950         }
951         if (!l_ptr->next_out)
952                 l_ptr->next_out = buf;
953         link_add_to_outqueue(l_ptr, buf, msg);
954         tipc_bearer_resolve_congestion(l_ptr->b_ptr, l_ptr);
955         return dsz;
956 }
957
958 /*
959  * tipc_link_send(): same as tipc_link_send_buf(), but the link to use has
960  * not been selected yet, and the the owner node is not locked
961  * Called by TIPC internal users, e.g. the name distributor
962  */
963
964 int tipc_link_send(struct sk_buff *buf, u32 dest, u32 selector)
965 {
966         struct link *l_ptr;
967         struct tipc_node *n_ptr;
968         int res = -ELINKCONG;
969
970         read_lock_bh(&tipc_net_lock);
971         n_ptr = tipc_node_find(dest);
972         if (n_ptr) {
973                 tipc_node_lock(n_ptr);
974                 l_ptr = n_ptr->active_links[selector & 1];
975                 if (l_ptr)
976                         res = tipc_link_send_buf(l_ptr, buf);
977                 else
978                         buf_discard(buf);
979                 tipc_node_unlock(n_ptr);
980         } else {
981                 buf_discard(buf);
982         }
983         read_unlock_bh(&tipc_net_lock);
984         return res;
985 }
986
987 /*
988  * link_send_buf_fast: Entry for data messages where the
989  * destination link is known and the header is complete,
990  * inclusive total message length. Very time critical.
991  * Link is locked. Returns user data length.
992  */
993
994 static int link_send_buf_fast(struct link *l_ptr, struct sk_buff *buf,
995                               u32 *used_max_pkt)
996 {
997         struct tipc_msg *msg = buf_msg(buf);
998         int res = msg_data_sz(msg);
999
1000         if (likely(!link_congested(l_ptr))) {
1001                 if (likely(msg_size(msg) <= l_ptr->max_pkt)) {
1002                         if (likely(list_empty(&l_ptr->b_ptr->cong_links))) {
1003                                 link_add_to_outqueue(l_ptr, buf, msg);
1004                                 if (likely(tipc_bearer_send(l_ptr->b_ptr, buf,
1005                                                             &l_ptr->media_addr))) {
1006                                         l_ptr->unacked_window = 0;
1007                                         return res;
1008                                 }
1009                                 tipc_bearer_schedule(l_ptr->b_ptr, l_ptr);
1010                                 l_ptr->stats.bearer_congs++;
1011                                 l_ptr->next_out = buf;
1012                                 return res;
1013                         }
1014                 } else
1015                         *used_max_pkt = l_ptr->max_pkt;
1016         }
1017         return tipc_link_send_buf(l_ptr, buf);  /* All other cases */
1018 }
1019
1020 /*
1021  * tipc_send_buf_fast: Entry for data messages where the
1022  * destination node is known and the header is complete,
1023  * inclusive total message length.
1024  * Returns user data length.
1025  */
1026 int tipc_send_buf_fast(struct sk_buff *buf, u32 destnode)
1027 {
1028         struct link *l_ptr;
1029         struct tipc_node *n_ptr;
1030         int res;
1031         u32 selector = msg_origport(buf_msg(buf)) & 1;
1032         u32 dummy;
1033
1034         if (destnode == tipc_own_addr)
1035                 return tipc_port_recv_msg(buf);
1036
1037         read_lock_bh(&tipc_net_lock);
1038         n_ptr = tipc_node_find(destnode);
1039         if (likely(n_ptr)) {
1040                 tipc_node_lock(n_ptr);
1041                 l_ptr = n_ptr->active_links[selector];
1042                 if (likely(l_ptr)) {
1043                         res = link_send_buf_fast(l_ptr, buf, &dummy);
1044                         tipc_node_unlock(n_ptr);
1045                         read_unlock_bh(&tipc_net_lock);
1046                         return res;
1047                 }
1048                 tipc_node_unlock(n_ptr);
1049         }
1050         read_unlock_bh(&tipc_net_lock);
1051         res = msg_data_sz(buf_msg(buf));
1052         tipc_reject_msg(buf, TIPC_ERR_NO_NODE);
1053         return res;
1054 }
1055
1056
1057 /*
1058  * tipc_link_send_sections_fast: Entry for messages where the
1059  * destination processor is known and the header is complete,
1060  * except for total message length.
1061  * Returns user data length or errno.
1062  */
1063 int tipc_link_send_sections_fast(struct tipc_port *sender,
1064                                  struct iovec const *msg_sect,
1065                                  const u32 num_sect,
1066                                  unsigned int total_len,
1067                                  u32 destaddr)
1068 {
1069         struct tipc_msg *hdr = &sender->phdr;
1070         struct link *l_ptr;
1071         struct sk_buff *buf;
1072         struct tipc_node *node;
1073         int res;
1074         u32 selector = msg_origport(hdr) & 1;
1075
1076 again:
1077         /*
1078          * Try building message using port's max_pkt hint.
1079          * (Must not hold any locks while building message.)
1080          */
1081
1082         res = tipc_msg_build(hdr, msg_sect, num_sect, total_len,
1083                              sender->max_pkt, !sender->user_port, &buf);
1084
1085         read_lock_bh(&tipc_net_lock);
1086         node = tipc_node_find(destaddr);
1087         if (likely(node)) {
1088                 tipc_node_lock(node);
1089                 l_ptr = node->active_links[selector];
1090                 if (likely(l_ptr)) {
1091                         if (likely(buf)) {
1092                                 res = link_send_buf_fast(l_ptr, buf,
1093                                                          &sender->max_pkt);
1094 exit:
1095                                 tipc_node_unlock(node);
1096                                 read_unlock_bh(&tipc_net_lock);
1097                                 return res;
1098                         }
1099
1100                         /* Exit if build request was invalid */
1101
1102                         if (unlikely(res < 0))
1103                                 goto exit;
1104
1105                         /* Exit if link (or bearer) is congested */
1106
1107                         if (link_congested(l_ptr) ||
1108                             !list_empty(&l_ptr->b_ptr->cong_links)) {
1109                                 res = link_schedule_port(l_ptr,
1110                                                          sender->ref, res);
1111                                 goto exit;
1112                         }
1113
1114                         /*
1115                          * Message size exceeds max_pkt hint; update hint,
1116                          * then re-try fast path or fragment the message
1117                          */
1118
1119                         sender->max_pkt = l_ptr->max_pkt;
1120                         tipc_node_unlock(node);
1121                         read_unlock_bh(&tipc_net_lock);
1122
1123
1124                         if ((msg_hdr_sz(hdr) + res) <= sender->max_pkt)
1125                                 goto again;
1126
1127                         return link_send_sections_long(sender, msg_sect,
1128                                                        num_sect, total_len,
1129                                                        destaddr);
1130                 }
1131                 tipc_node_unlock(node);
1132         }
1133         read_unlock_bh(&tipc_net_lock);
1134
1135         /* Couldn't find a link to the destination node */
1136
1137         if (buf)
1138                 return tipc_reject_msg(buf, TIPC_ERR_NO_NODE);
1139         if (res >= 0)
1140                 return tipc_port_reject_sections(sender, hdr, msg_sect, num_sect,
1141                                                  total_len, TIPC_ERR_NO_NODE);
1142         return res;
1143 }
1144
1145 /*
1146  * link_send_sections_long(): Entry for long messages where the
1147  * destination node is known and the header is complete,
1148  * inclusive total message length.
1149  * Link and bearer congestion status have been checked to be ok,
1150  * and are ignored if they change.
1151  *
1152  * Note that fragments do not use the full link MTU so that they won't have
1153  * to undergo refragmentation if link changeover causes them to be sent
1154  * over another link with an additional tunnel header added as prefix.
1155  * (Refragmentation will still occur if the other link has a smaller MTU.)
1156  *
1157  * Returns user data length or errno.
1158  */
1159 static int link_send_sections_long(struct tipc_port *sender,
1160                                    struct iovec const *msg_sect,
1161                                    u32 num_sect,
1162                                    unsigned int total_len,
1163                                    u32 destaddr)
1164 {
1165         struct link *l_ptr;
1166         struct tipc_node *node;
1167         struct tipc_msg *hdr = &sender->phdr;
1168         u32 dsz = total_len;
1169         u32 max_pkt, fragm_sz, rest;
1170         struct tipc_msg fragm_hdr;
1171         struct sk_buff *buf, *buf_chain, *prev;
1172         u32 fragm_crs, fragm_rest, hsz, sect_rest;
1173         const unchar *sect_crs;
1174         int curr_sect;
1175         u32 fragm_no;
1176
1177 again:
1178         fragm_no = 1;
1179         max_pkt = sender->max_pkt - INT_H_SIZE;
1180                 /* leave room for tunnel header in case of link changeover */
1181         fragm_sz = max_pkt - INT_H_SIZE;
1182                 /* leave room for fragmentation header in each fragment */
1183         rest = dsz;
1184         fragm_crs = 0;
1185         fragm_rest = 0;
1186         sect_rest = 0;
1187         sect_crs = NULL;
1188         curr_sect = -1;
1189
1190         /* Prepare reusable fragment header: */
1191
1192         tipc_msg_init(&fragm_hdr, MSG_FRAGMENTER, FIRST_FRAGMENT,
1193                  INT_H_SIZE, msg_destnode(hdr));
1194         msg_set_size(&fragm_hdr, max_pkt);
1195         msg_set_fragm_no(&fragm_hdr, 1);
1196
1197         /* Prepare header of first fragment: */
1198
1199         buf_chain = buf = tipc_buf_acquire(max_pkt);
1200         if (!buf)
1201                 return -ENOMEM;
1202         buf->next = NULL;
1203         skb_copy_to_linear_data(buf, &fragm_hdr, INT_H_SIZE);
1204         hsz = msg_hdr_sz(hdr);
1205         skb_copy_to_linear_data_offset(buf, INT_H_SIZE, hdr, hsz);
1206
1207         /* Chop up message: */
1208
1209         fragm_crs = INT_H_SIZE + hsz;
1210         fragm_rest = fragm_sz - hsz;
1211
1212         do {            /* For all sections */
1213                 u32 sz;
1214
1215                 if (!sect_rest) {
1216                         sect_rest = msg_sect[++curr_sect].iov_len;
1217                         sect_crs = (const unchar *)msg_sect[curr_sect].iov_base;
1218                 }
1219
1220                 if (sect_rest < fragm_rest)
1221                         sz = sect_rest;
1222                 else
1223                         sz = fragm_rest;
1224
1225                 if (likely(!sender->user_port)) {
1226                         if (copy_from_user(buf->data + fragm_crs, sect_crs, sz)) {
1227 error:
1228                                 for (; buf_chain; buf_chain = buf) {
1229                                         buf = buf_chain->next;
1230                                         buf_discard(buf_chain);
1231                                 }
1232                                 return -EFAULT;
1233                         }
1234                 } else
1235                         skb_copy_to_linear_data_offset(buf, fragm_crs,
1236                                                        sect_crs, sz);
1237                 sect_crs += sz;
1238                 sect_rest -= sz;
1239                 fragm_crs += sz;
1240                 fragm_rest -= sz;
1241                 rest -= sz;
1242
1243                 if (!fragm_rest && rest) {
1244
1245                         /* Initiate new fragment: */
1246                         if (rest <= fragm_sz) {
1247                                 fragm_sz = rest;
1248                                 msg_set_type(&fragm_hdr, LAST_FRAGMENT);
1249                         } else {
1250                                 msg_set_type(&fragm_hdr, FRAGMENT);
1251                         }
1252                         msg_set_size(&fragm_hdr, fragm_sz + INT_H_SIZE);
1253                         msg_set_fragm_no(&fragm_hdr, ++fragm_no);
1254                         prev = buf;
1255                         buf = tipc_buf_acquire(fragm_sz + INT_H_SIZE);
1256                         if (!buf)
1257                                 goto error;
1258
1259                         buf->next = NULL;
1260                         prev->next = buf;
1261                         skb_copy_to_linear_data(buf, &fragm_hdr, INT_H_SIZE);
1262                         fragm_crs = INT_H_SIZE;
1263                         fragm_rest = fragm_sz;
1264                 }
1265         } while (rest > 0);
1266
1267         /*
1268          * Now we have a buffer chain. Select a link and check
1269          * that packet size is still OK
1270          */
1271         node = tipc_node_find(destaddr);
1272         if (likely(node)) {
1273                 tipc_node_lock(node);
1274                 l_ptr = node->active_links[sender->ref & 1];
1275                 if (!l_ptr) {
1276                         tipc_node_unlock(node);
1277                         goto reject;
1278                 }
1279                 if (l_ptr->max_pkt < max_pkt) {
1280                         sender->max_pkt = l_ptr->max_pkt;
1281                         tipc_node_unlock(node);
1282                         for (; buf_chain; buf_chain = buf) {
1283                                 buf = buf_chain->next;
1284                                 buf_discard(buf_chain);
1285                         }
1286                         goto again;
1287                 }
1288         } else {
1289 reject:
1290                 for (; buf_chain; buf_chain = buf) {
1291                         buf = buf_chain->next;
1292                         buf_discard(buf_chain);
1293                 }
1294                 return tipc_port_reject_sections(sender, hdr, msg_sect, num_sect,
1295                                                  total_len, TIPC_ERR_NO_NODE);
1296         }
1297
1298         /* Append chain of fragments to send queue & send them */
1299
1300         l_ptr->long_msg_seq_no++;
1301         link_add_chain_to_outqueue(l_ptr, buf_chain, l_ptr->long_msg_seq_no);
1302         l_ptr->stats.sent_fragments += fragm_no;
1303         l_ptr->stats.sent_fragmented++;
1304         tipc_link_push_queue(l_ptr);
1305         tipc_node_unlock(node);
1306         return dsz;
1307 }
1308
1309 /*
1310  * tipc_link_push_packet: Push one unsent packet to the media
1311  */
1312 u32 tipc_link_push_packet(struct link *l_ptr)
1313 {
1314         struct sk_buff *buf = l_ptr->first_out;
1315         u32 r_q_size = l_ptr->retransm_queue_size;
1316         u32 r_q_head = l_ptr->retransm_queue_head;
1317
1318         /* Step to position where retransmission failed, if any,    */
1319         /* consider that buffers may have been released in meantime */
1320
1321         if (r_q_size && buf) {
1322                 u32 last = lesser(mod(r_q_head + r_q_size),
1323                                   link_last_sent(l_ptr));
1324                 u32 first = msg_seqno(buf_msg(buf));
1325
1326                 while (buf && less(first, r_q_head)) {
1327                         first = mod(first + 1);
1328                         buf = buf->next;
1329                 }
1330                 l_ptr->retransm_queue_head = r_q_head = first;
1331                 l_ptr->retransm_queue_size = r_q_size = mod(last - first);
1332         }
1333
1334         /* Continue retransmission now, if there is anything: */
1335
1336         if (r_q_size && buf) {
1337                 msg_set_ack(buf_msg(buf), mod(l_ptr->next_in_no - 1));
1338                 msg_set_bcast_ack(buf_msg(buf), l_ptr->owner->bclink.last_in);
1339                 if (tipc_bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
1340                         l_ptr->retransm_queue_head = mod(++r_q_head);
1341                         l_ptr->retransm_queue_size = --r_q_size;
1342                         l_ptr->stats.retransmitted++;
1343                         return 0;
1344                 } else {
1345                         l_ptr->stats.bearer_congs++;
1346                         return PUSH_FAILED;
1347                 }
1348         }
1349
1350         /* Send deferred protocol message, if any: */
1351
1352         buf = l_ptr->proto_msg_queue;
1353         if (buf) {
1354                 msg_set_ack(buf_msg(buf), mod(l_ptr->next_in_no - 1));
1355                 msg_set_bcast_ack(buf_msg(buf), l_ptr->owner->bclink.last_in);
1356                 if (tipc_bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
1357                         l_ptr->unacked_window = 0;
1358                         buf_discard(buf);
1359                         l_ptr->proto_msg_queue = NULL;
1360                         return 0;
1361                 } else {
1362                         l_ptr->stats.bearer_congs++;
1363                         return PUSH_FAILED;
1364                 }
1365         }
1366
1367         /* Send one deferred data message, if send window not full: */
1368
1369         buf = l_ptr->next_out;
1370         if (buf) {
1371                 struct tipc_msg *msg = buf_msg(buf);
1372                 u32 next = msg_seqno(msg);
1373                 u32 first = msg_seqno(buf_msg(l_ptr->first_out));
1374
1375                 if (mod(next - first) < l_ptr->queue_limit[0]) {
1376                         msg_set_ack(msg, mod(l_ptr->next_in_no - 1));
1377                         msg_set_bcast_ack(msg, l_ptr->owner->bclink.last_in);
1378                         if (tipc_bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
1379                                 if (msg_user(msg) == MSG_BUNDLER)
1380                                         msg_set_type(msg, CLOSED_MSG);
1381                                 l_ptr->next_out = buf->next;
1382                                 return 0;
1383                         } else {
1384                                 l_ptr->stats.bearer_congs++;
1385                                 return PUSH_FAILED;
1386                         }
1387                 }
1388         }
1389         return PUSH_FINISHED;
1390 }
1391
1392 /*
1393  * push_queue(): push out the unsent messages of a link where
1394  *               congestion has abated. Node is locked
1395  */
1396 void tipc_link_push_queue(struct link *l_ptr)
1397 {
1398         u32 res;
1399
1400         if (tipc_bearer_congested(l_ptr->b_ptr, l_ptr))
1401                 return;
1402
1403         do {
1404                 res = tipc_link_push_packet(l_ptr);
1405         } while (!res);
1406
1407         if (res == PUSH_FAILED)
1408                 tipc_bearer_schedule(l_ptr->b_ptr, l_ptr);
1409 }
1410
1411 static void link_reset_all(unsigned long addr)
1412 {
1413         struct tipc_node *n_ptr;
1414         char addr_string[16];
1415         u32 i;
1416
1417         read_lock_bh(&tipc_net_lock);
1418         n_ptr = tipc_node_find((u32)addr);
1419         if (!n_ptr) {
1420                 read_unlock_bh(&tipc_net_lock);
1421                 return; /* node no longer exists */
1422         }
1423
1424         tipc_node_lock(n_ptr);
1425
1426         warn("Resetting all links to %s\n",
1427              tipc_addr_string_fill(addr_string, n_ptr->addr));
1428
1429         for (i = 0; i < MAX_BEARERS; i++) {
1430                 if (n_ptr->links[i]) {
1431                         link_print(n_ptr->links[i], "Resetting link\n");
1432                         tipc_link_reset(n_ptr->links[i]);
1433                 }
1434         }
1435
1436         tipc_node_unlock(n_ptr);
1437         read_unlock_bh(&tipc_net_lock);
1438 }
1439
1440 static void link_retransmit_failure(struct link *l_ptr, struct sk_buff *buf)
1441 {
1442         struct tipc_msg *msg = buf_msg(buf);
1443
1444         warn("Retransmission failure on link <%s>\n", l_ptr->name);
1445
1446         if (l_ptr->addr) {
1447
1448                 /* Handle failure on standard link */
1449
1450                 link_print(l_ptr, "Resetting link\n");
1451                 tipc_link_reset(l_ptr);
1452
1453         } else {
1454
1455                 /* Handle failure on broadcast link */
1456
1457                 struct tipc_node *n_ptr;
1458                 char addr_string[16];
1459
1460                 info("Msg seq number: %u,  ", msg_seqno(msg));
1461                 info("Outstanding acks: %lu\n",
1462                      (unsigned long) TIPC_SKB_CB(buf)->handle);
1463
1464                 n_ptr = tipc_bclink_retransmit_to();
1465                 tipc_node_lock(n_ptr);
1466
1467                 tipc_addr_string_fill(addr_string, n_ptr->addr);
1468                 info("Multicast link info for %s\n", addr_string);
1469                 info("Supported: %d,  ", n_ptr->bclink.supported);
1470                 info("Acked: %u\n", n_ptr->bclink.acked);
1471                 info("Last in: %u,  ", n_ptr->bclink.last_in);
1472                 info("Gap after: %u,  ", n_ptr->bclink.gap_after);
1473                 info("Gap to: %u\n", n_ptr->bclink.gap_to);
1474                 info("Nack sync: %u\n\n", n_ptr->bclink.nack_sync);
1475
1476                 tipc_k_signal((Handler)link_reset_all, (unsigned long)n_ptr->addr);
1477
1478                 tipc_node_unlock(n_ptr);
1479
1480                 l_ptr->stale_count = 0;
1481         }
1482 }
1483
1484 void tipc_link_retransmit(struct link *l_ptr, struct sk_buff *buf,
1485                           u32 retransmits)
1486 {
1487         struct tipc_msg *msg;
1488
1489         if (!buf)
1490                 return;
1491
1492         msg = buf_msg(buf);
1493
1494         if (tipc_bearer_congested(l_ptr->b_ptr, l_ptr)) {
1495                 if (l_ptr->retransm_queue_size == 0) {
1496                         l_ptr->retransm_queue_head = msg_seqno(msg);
1497                         l_ptr->retransm_queue_size = retransmits;
1498                 } else {
1499                         err("Unexpected retransmit on link %s (qsize=%d)\n",
1500                             l_ptr->name, l_ptr->retransm_queue_size);
1501                 }
1502                 return;
1503         } else {
1504                 /* Detect repeated retransmit failures on uncongested bearer */
1505
1506                 if (l_ptr->last_retransmitted == msg_seqno(msg)) {
1507                         if (++l_ptr->stale_count > 100) {
1508                                 link_retransmit_failure(l_ptr, buf);
1509                                 return;
1510                         }
1511                 } else {
1512                         l_ptr->last_retransmitted = msg_seqno(msg);
1513                         l_ptr->stale_count = 1;
1514                 }
1515         }
1516
1517         while (retransmits && (buf != l_ptr->next_out) && buf) {
1518                 msg = buf_msg(buf);
1519                 msg_set_ack(msg, mod(l_ptr->next_in_no - 1));
1520                 msg_set_bcast_ack(msg, l_ptr->owner->bclink.last_in);
1521                 if (tipc_bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
1522                         buf = buf->next;
1523                         retransmits--;
1524                         l_ptr->stats.retransmitted++;
1525                 } else {
1526                         tipc_bearer_schedule(l_ptr->b_ptr, l_ptr);
1527                         l_ptr->stats.bearer_congs++;
1528                         l_ptr->retransm_queue_head = msg_seqno(buf_msg(buf));
1529                         l_ptr->retransm_queue_size = retransmits;
1530                         return;
1531                 }
1532         }
1533
1534         l_ptr->retransm_queue_head = l_ptr->retransm_queue_size = 0;
1535 }
1536
1537 /**
1538  * link_insert_deferred_queue - insert deferred messages back into receive chain
1539  */
1540
1541 static struct sk_buff *link_insert_deferred_queue(struct link *l_ptr,
1542                                                   struct sk_buff *buf)
1543 {
1544         u32 seq_no;
1545
1546         if (l_ptr->oldest_deferred_in == NULL)
1547                 return buf;
1548
1549         seq_no = msg_seqno(buf_msg(l_ptr->oldest_deferred_in));
1550         if (seq_no == mod(l_ptr->next_in_no)) {
1551                 l_ptr->newest_deferred_in->next = buf;
1552                 buf = l_ptr->oldest_deferred_in;
1553                 l_ptr->oldest_deferred_in = NULL;
1554                 l_ptr->deferred_inqueue_sz = 0;
1555         }
1556         return buf;
1557 }
1558
1559 /**
1560  * link_recv_buf_validate - validate basic format of received message
1561  *
1562  * This routine ensures a TIPC message has an acceptable header, and at least
1563  * as much data as the header indicates it should.  The routine also ensures
1564  * that the entire message header is stored in the main fragment of the message
1565  * buffer, to simplify future access to message header fields.
1566  *
1567  * Note: Having extra info present in the message header or data areas is OK.
1568  * TIPC will ignore the excess, under the assumption that it is optional info
1569  * introduced by a later release of the protocol.
1570  */
1571
1572 static int link_recv_buf_validate(struct sk_buff *buf)
1573 {
1574         static u32 min_data_hdr_size[8] = {
1575                 SHORT_H_SIZE, MCAST_H_SIZE, NAMED_H_SIZE, BASIC_H_SIZE,
1576                 MAX_H_SIZE, MAX_H_SIZE, MAX_H_SIZE, MAX_H_SIZE
1577                 };
1578
1579         struct tipc_msg *msg;
1580         u32 tipc_hdr[2];
1581         u32 size;
1582         u32 hdr_size;
1583         u32 min_hdr_size;
1584
1585         if (unlikely(buf->len < MIN_H_SIZE))
1586                 return 0;
1587
1588         msg = skb_header_pointer(buf, 0, sizeof(tipc_hdr), tipc_hdr);
1589         if (msg == NULL)
1590                 return 0;
1591
1592         if (unlikely(msg_version(msg) != TIPC_VERSION))
1593                 return 0;
1594
1595         size = msg_size(msg);
1596         hdr_size = msg_hdr_sz(msg);
1597         min_hdr_size = msg_isdata(msg) ?
1598                 min_data_hdr_size[msg_type(msg)] : INT_H_SIZE;
1599
1600         if (unlikely((hdr_size < min_hdr_size) ||
1601                      (size < hdr_size) ||
1602                      (buf->len < size) ||
1603                      (size - hdr_size > TIPC_MAX_USER_MSG_SIZE)))
1604                 return 0;
1605
1606         return pskb_may_pull(buf, hdr_size);
1607 }
1608
1609 /**
1610  * tipc_recv_msg - process TIPC messages arriving from off-node
1611  * @head: pointer to message buffer chain
1612  * @tb_ptr: pointer to bearer message arrived on
1613  *
1614  * Invoked with no locks held.  Bearer pointer must point to a valid bearer
1615  * structure (i.e. cannot be NULL), but bearer can be inactive.
1616  */
1617
1618 void tipc_recv_msg(struct sk_buff *head, struct tipc_bearer *b_ptr)
1619 {
1620         read_lock_bh(&tipc_net_lock);
1621         while (head) {
1622                 struct tipc_node *n_ptr;
1623                 struct link *l_ptr;
1624                 struct sk_buff *crs;
1625                 struct sk_buff *buf = head;
1626                 struct tipc_msg *msg;
1627                 u32 seq_no;
1628                 u32 ackd;
1629                 u32 released = 0;
1630                 int type;
1631
1632                 head = head->next;
1633
1634                 /* Ensure bearer is still enabled */
1635
1636                 if (unlikely(!b_ptr->active))
1637                         goto cont;
1638
1639                 /* Ensure message is well-formed */
1640
1641                 if (unlikely(!link_recv_buf_validate(buf)))
1642                         goto cont;
1643
1644                 /* Ensure message data is a single contiguous unit */
1645
1646                 if (unlikely(buf_linearize(buf)))
1647                         goto cont;
1648
1649                 /* Handle arrival of a non-unicast link message */
1650
1651                 msg = buf_msg(buf);
1652
1653                 if (unlikely(msg_non_seq(msg))) {
1654                         if (msg_user(msg) ==  LINK_CONFIG)
1655                                 tipc_disc_recv_msg(buf, b_ptr);
1656                         else
1657                                 tipc_bclink_recv_pkt(buf);
1658                         continue;
1659                 }
1660
1661                 if (unlikely(!msg_short(msg) &&
1662                              (msg_destnode(msg) != tipc_own_addr)))
1663                         goto cont;
1664
1665                 /* Discard non-routeable messages destined for another node */
1666
1667                 if (unlikely(!msg_isdata(msg) &&
1668                              (msg_destnode(msg) != tipc_own_addr))) {
1669                         if ((msg_user(msg) != CONN_MANAGER) &&
1670                             (msg_user(msg) != MSG_FRAGMENTER))
1671                                 goto cont;
1672                 }
1673
1674                 /* Locate neighboring node that sent message */
1675
1676                 n_ptr = tipc_node_find(msg_prevnode(msg));
1677                 if (unlikely(!n_ptr))
1678                         goto cont;
1679                 tipc_node_lock(n_ptr);
1680
1681                 /* Don't talk to neighbor during cleanup after last session */
1682
1683                 if (n_ptr->cleanup_required) {
1684                         tipc_node_unlock(n_ptr);
1685                         goto cont;
1686                 }
1687
1688                 /* Locate unicast link endpoint that should handle message */
1689
1690                 l_ptr = n_ptr->links[b_ptr->identity];
1691                 if (unlikely(!l_ptr)) {
1692                         tipc_node_unlock(n_ptr);
1693                         goto cont;
1694                 }
1695
1696                 /* Validate message sequence number info */
1697
1698                 seq_no = msg_seqno(msg);
1699                 ackd = msg_ack(msg);
1700
1701                 /* Release acked messages */
1702
1703                 if (less(n_ptr->bclink.acked, msg_bcast_ack(msg))) {
1704                         if (tipc_node_is_up(n_ptr) && n_ptr->bclink.supported)
1705                                 tipc_bclink_acknowledge(n_ptr, msg_bcast_ack(msg));
1706                 }
1707
1708                 crs = l_ptr->first_out;
1709                 while ((crs != l_ptr->next_out) &&
1710                        less_eq(msg_seqno(buf_msg(crs)), ackd)) {
1711                         struct sk_buff *next = crs->next;
1712
1713                         buf_discard(crs);
1714                         crs = next;
1715                         released++;
1716                 }
1717                 if (released) {
1718                         l_ptr->first_out = crs;
1719                         l_ptr->out_queue_size -= released;
1720                 }
1721
1722                 /* Try sending any messages link endpoint has pending */
1723
1724                 if (unlikely(l_ptr->next_out))
1725                         tipc_link_push_queue(l_ptr);
1726                 if (unlikely(!list_empty(&l_ptr->waiting_ports)))
1727                         tipc_link_wakeup_ports(l_ptr, 0);
1728                 if (unlikely(++l_ptr->unacked_window >= TIPC_MIN_LINK_WIN)) {
1729                         l_ptr->stats.sent_acks++;
1730                         tipc_link_send_proto_msg(l_ptr, STATE_MSG, 0, 0, 0, 0, 0);
1731                 }
1732
1733                 /* Now (finally!) process the incoming message */
1734
1735 protocol_check:
1736                 if (likely(link_working_working(l_ptr))) {
1737                         if (likely(seq_no == mod(l_ptr->next_in_no))) {
1738                                 l_ptr->next_in_no++;
1739                                 if (unlikely(l_ptr->oldest_deferred_in))
1740                                         head = link_insert_deferred_queue(l_ptr,
1741                                                                           head);
1742                                 if (likely(msg_is_dest(msg, tipc_own_addr))) {
1743 deliver:
1744                                         if (likely(msg_isdata(msg))) {
1745                                                 tipc_node_unlock(n_ptr);
1746                                                 tipc_port_recv_msg(buf);
1747                                                 continue;
1748                                         }
1749                                         switch (msg_user(msg)) {
1750                                         case MSG_BUNDLER:
1751                                                 l_ptr->stats.recv_bundles++;
1752                                                 l_ptr->stats.recv_bundled +=
1753                                                         msg_msgcnt(msg);
1754                                                 tipc_node_unlock(n_ptr);
1755                                                 tipc_link_recv_bundle(buf);
1756                                                 continue;
1757                                         case NAME_DISTRIBUTOR:
1758                                                 tipc_node_unlock(n_ptr);
1759                                                 tipc_named_recv(buf);
1760                                                 continue;
1761                                         case CONN_MANAGER:
1762                                                 tipc_node_unlock(n_ptr);
1763                                                 tipc_port_recv_proto_msg(buf);
1764                                                 continue;
1765                                         case MSG_FRAGMENTER:
1766                                                 l_ptr->stats.recv_fragments++;
1767                                                 if (tipc_link_recv_fragment(&l_ptr->defragm_buf,
1768                                                                             &buf, &msg)) {
1769                                                         l_ptr->stats.recv_fragmented++;
1770                                                         goto deliver;
1771                                                 }
1772                                                 break;
1773                                         case CHANGEOVER_PROTOCOL:
1774                                                 type = msg_type(msg);
1775                                                 if (link_recv_changeover_msg(&l_ptr, &buf)) {
1776                                                         msg = buf_msg(buf);
1777                                                         seq_no = msg_seqno(msg);
1778                                                         if (type == ORIGINAL_MSG)
1779                                                                 goto deliver;
1780                                                         goto protocol_check;
1781                                                 }
1782                                                 break;
1783                                         default:
1784                                                 buf_discard(buf);
1785                                                 buf = NULL;
1786                                                 break;
1787                                         }
1788                                 }
1789                                 tipc_node_unlock(n_ptr);
1790                                 tipc_net_route_msg(buf);
1791                                 continue;
1792                         }
1793                         link_handle_out_of_seq_msg(l_ptr, buf);
1794                         head = link_insert_deferred_queue(l_ptr, head);
1795                         tipc_node_unlock(n_ptr);
1796                         continue;
1797                 }
1798
1799                 if (msg_user(msg) == LINK_PROTOCOL) {
1800                         link_recv_proto_msg(l_ptr, buf);
1801                         head = link_insert_deferred_queue(l_ptr, head);
1802                         tipc_node_unlock(n_ptr);
1803                         continue;
1804                 }
1805                 link_state_event(l_ptr, TRAFFIC_MSG_EVT);
1806
1807                 if (link_working_working(l_ptr)) {
1808                         /* Re-insert in front of queue */
1809                         buf->next = head;
1810                         head = buf;
1811                         tipc_node_unlock(n_ptr);
1812                         continue;
1813                 }
1814                 tipc_node_unlock(n_ptr);
1815 cont:
1816                 buf_discard(buf);
1817         }
1818         read_unlock_bh(&tipc_net_lock);
1819 }
1820
1821 /*
1822  * link_defer_buf(): Sort a received out-of-sequence packet
1823  *                   into the deferred reception queue.
1824  * Returns the increase of the queue length,i.e. 0 or 1
1825  */
1826
1827 u32 tipc_link_defer_pkt(struct sk_buff **head,
1828                         struct sk_buff **tail,
1829                         struct sk_buff *buf)
1830 {
1831         struct sk_buff *prev = NULL;
1832         struct sk_buff *crs = *head;
1833         u32 seq_no = msg_seqno(buf_msg(buf));
1834
1835         buf->next = NULL;
1836
1837         /* Empty queue ? */
1838         if (*head == NULL) {
1839                 *head = *tail = buf;
1840                 return 1;
1841         }
1842
1843         /* Last ? */
1844         if (less(msg_seqno(buf_msg(*tail)), seq_no)) {
1845                 (*tail)->next = buf;
1846                 *tail = buf;
1847                 return 1;
1848         }
1849
1850         /* Scan through queue and sort it in */
1851         do {
1852                 struct tipc_msg *msg = buf_msg(crs);
1853
1854                 if (less(seq_no, msg_seqno(msg))) {
1855                         buf->next = crs;
1856                         if (prev)
1857                                 prev->next = buf;
1858                         else
1859                                 *head = buf;
1860                         return 1;
1861                 }
1862                 if (seq_no == msg_seqno(msg))
1863                         break;
1864                 prev = crs;
1865                 crs = crs->next;
1866         } while (crs);
1867
1868         /* Message is a duplicate of an existing message */
1869
1870         buf_discard(buf);
1871         return 0;
1872 }
1873
1874 /**
1875  * link_handle_out_of_seq_msg - handle arrival of out-of-sequence packet
1876  */
1877
1878 static void link_handle_out_of_seq_msg(struct link *l_ptr,
1879                                        struct sk_buff *buf)
1880 {
1881         u32 seq_no = msg_seqno(buf_msg(buf));
1882
1883         if (likely(msg_user(buf_msg(buf)) == LINK_PROTOCOL)) {
1884                 link_recv_proto_msg(l_ptr, buf);
1885                 return;
1886         }
1887
1888         /* Record OOS packet arrival (force mismatch on next timeout) */
1889
1890         l_ptr->checkpoint--;
1891
1892         /*
1893          * Discard packet if a duplicate; otherwise add it to deferred queue
1894          * and notify peer of gap as per protocol specification
1895          */
1896
1897         if (less(seq_no, mod(l_ptr->next_in_no))) {
1898                 l_ptr->stats.duplicates++;
1899                 buf_discard(buf);
1900                 return;
1901         }
1902
1903         if (tipc_link_defer_pkt(&l_ptr->oldest_deferred_in,
1904                                 &l_ptr->newest_deferred_in, buf)) {
1905                 l_ptr->deferred_inqueue_sz++;
1906                 l_ptr->stats.deferred_recv++;
1907                 if ((l_ptr->deferred_inqueue_sz % 16) == 1)
1908                         tipc_link_send_proto_msg(l_ptr, STATE_MSG, 0, 0, 0, 0, 0);
1909         } else
1910                 l_ptr->stats.duplicates++;
1911 }
1912
1913 /*
1914  * Send protocol message to the other endpoint.
1915  */
1916 void tipc_link_send_proto_msg(struct link *l_ptr, u32 msg_typ, int probe_msg,
1917                               u32 gap, u32 tolerance, u32 priority, u32 ack_mtu)
1918 {
1919         struct sk_buff *buf = NULL;
1920         struct tipc_msg *msg = l_ptr->pmsg;
1921         u32 msg_size = sizeof(l_ptr->proto_msg);
1922         int r_flag;
1923
1924         if (link_blocked(l_ptr))
1925                 return;
1926         msg_set_type(msg, msg_typ);
1927         msg_set_net_plane(msg, l_ptr->b_ptr->net_plane);
1928         msg_set_bcast_ack(msg, mod(l_ptr->owner->bclink.last_in));
1929         msg_set_last_bcast(msg, tipc_bclink_get_last_sent());
1930
1931         if (msg_typ == STATE_MSG) {
1932                 u32 next_sent = mod(l_ptr->next_out_no);
1933
1934                 if (!tipc_link_is_up(l_ptr))
1935                         return;
1936                 if (l_ptr->next_out)
1937                         next_sent = msg_seqno(buf_msg(l_ptr->next_out));
1938                 msg_set_next_sent(msg, next_sent);
1939                 if (l_ptr->oldest_deferred_in) {
1940                         u32 rec = msg_seqno(buf_msg(l_ptr->oldest_deferred_in));
1941                         gap = mod(rec - mod(l_ptr->next_in_no));
1942                 }
1943                 msg_set_seq_gap(msg, gap);
1944                 if (gap)
1945                         l_ptr->stats.sent_nacks++;
1946                 msg_set_link_tolerance(msg, tolerance);
1947                 msg_set_linkprio(msg, priority);
1948                 msg_set_max_pkt(msg, ack_mtu);
1949                 msg_set_ack(msg, mod(l_ptr->next_in_no - 1));
1950                 msg_set_probe(msg, probe_msg != 0);
1951                 if (probe_msg) {
1952                         u32 mtu = l_ptr->max_pkt;
1953
1954                         if ((mtu < l_ptr->max_pkt_target) &&
1955                             link_working_working(l_ptr) &&
1956                             l_ptr->fsm_msg_cnt) {
1957                                 msg_size = (mtu + (l_ptr->max_pkt_target - mtu)/2 + 2) & ~3;
1958                                 if (l_ptr->max_pkt_probes == 10) {
1959                                         l_ptr->max_pkt_target = (msg_size - 4);
1960                                         l_ptr->max_pkt_probes = 0;
1961                                         msg_size = (mtu + (l_ptr->max_pkt_target - mtu)/2 + 2) & ~3;
1962                                 }
1963                                 l_ptr->max_pkt_probes++;
1964                         }
1965
1966                         l_ptr->stats.sent_probes++;
1967                 }
1968                 l_ptr->stats.sent_states++;
1969         } else {                /* RESET_MSG or ACTIVATE_MSG */
1970                 msg_set_ack(msg, mod(l_ptr->reset_checkpoint - 1));
1971                 msg_set_seq_gap(msg, 0);
1972                 msg_set_next_sent(msg, 1);
1973                 msg_set_probe(msg, 0);
1974                 msg_set_link_tolerance(msg, l_ptr->tolerance);
1975                 msg_set_linkprio(msg, l_ptr->priority);
1976                 msg_set_max_pkt(msg, l_ptr->max_pkt_target);
1977         }
1978
1979         r_flag = (l_ptr->owner->working_links > tipc_link_is_up(l_ptr));
1980         msg_set_redundant_link(msg, r_flag);
1981         msg_set_linkprio(msg, l_ptr->priority);
1982
1983         /* Ensure sequence number will not fit : */
1984
1985         msg_set_seqno(msg, mod(l_ptr->next_out_no + (0xffff/2)));
1986
1987         /* Congestion? */
1988
1989         if (tipc_bearer_congested(l_ptr->b_ptr, l_ptr)) {
1990                 if (!l_ptr->proto_msg_queue) {
1991                         l_ptr->proto_msg_queue =
1992                                 tipc_buf_acquire(sizeof(l_ptr->proto_msg));
1993                 }
1994                 buf = l_ptr->proto_msg_queue;
1995                 if (!buf)
1996                         return;
1997                 skb_copy_to_linear_data(buf, msg, sizeof(l_ptr->proto_msg));
1998                 return;
1999         }
2000
2001         /* Message can be sent */
2002
2003         buf = tipc_buf_acquire(msg_size);
2004         if (!buf)
2005                 return;
2006
2007         skb_copy_to_linear_data(buf, msg, sizeof(l_ptr->proto_msg));
2008         msg_set_size(buf_msg(buf), msg_size);
2009
2010         if (tipc_bearer_send(l_ptr->b_ptr, buf, &l_ptr->media_addr)) {
2011                 l_ptr->unacked_window = 0;
2012                 buf_discard(buf);
2013                 return;
2014         }
2015
2016         /* New congestion */
2017         tipc_bearer_schedule(l_ptr->b_ptr, l_ptr);
2018         l_ptr->proto_msg_queue = buf;
2019         l_ptr->stats.bearer_congs++;
2020 }
2021
2022 /*
2023  * Receive protocol message :
2024  * Note that network plane id propagates through the network, and may
2025  * change at any time. The node with lowest address rules
2026  */
2027
2028 static void link_recv_proto_msg(struct link *l_ptr, struct sk_buff *buf)
2029 {
2030         u32 rec_gap = 0;
2031         u32 max_pkt_info;
2032         u32 max_pkt_ack;
2033         u32 msg_tol;
2034         struct tipc_msg *msg = buf_msg(buf);
2035
2036         if (link_blocked(l_ptr))
2037                 goto exit;
2038
2039         /* record unnumbered packet arrival (force mismatch on next timeout) */
2040
2041         l_ptr->checkpoint--;
2042
2043         if (l_ptr->b_ptr->net_plane != msg_net_plane(msg))
2044                 if (tipc_own_addr > msg_prevnode(msg))
2045                         l_ptr->b_ptr->net_plane = msg_net_plane(msg);
2046
2047         l_ptr->owner->permit_changeover = msg_redundant_link(msg);
2048
2049         switch (msg_type(msg)) {
2050
2051         case RESET_MSG:
2052                 if (!link_working_unknown(l_ptr) &&
2053                     (l_ptr->peer_session != INVALID_SESSION)) {
2054                         if (msg_session(msg) == l_ptr->peer_session)
2055                                 break; /* duplicate: ignore */
2056                 }
2057                 /* fall thru' */
2058         case ACTIVATE_MSG:
2059                 /* Update link settings according other endpoint's values */
2060
2061                 strcpy((strrchr(l_ptr->name, ':') + 1), (char *)msg_data(msg));
2062
2063                 msg_tol = msg_link_tolerance(msg);
2064                 if (msg_tol > l_ptr->tolerance)
2065                         link_set_supervision_props(l_ptr, msg_tol);
2066
2067                 if (msg_linkprio(msg) > l_ptr->priority)
2068                         l_ptr->priority = msg_linkprio(msg);
2069
2070                 max_pkt_info = msg_max_pkt(msg);
2071                 if (max_pkt_info) {
2072                         if (max_pkt_info < l_ptr->max_pkt_target)
2073                                 l_ptr->max_pkt_target = max_pkt_info;
2074                         if (l_ptr->max_pkt > l_ptr->max_pkt_target)
2075                                 l_ptr->max_pkt = l_ptr->max_pkt_target;
2076                 } else {
2077                         l_ptr->max_pkt = l_ptr->max_pkt_target;
2078                 }
2079                 l_ptr->owner->bclink.supported = (max_pkt_info != 0);
2080
2081                 link_state_event(l_ptr, msg_type(msg));
2082
2083                 l_ptr->peer_session = msg_session(msg);
2084                 l_ptr->peer_bearer_id = msg_bearer_id(msg);
2085
2086                 /* Synchronize broadcast sequence numbers */
2087                 if (!tipc_node_redundant_links(l_ptr->owner))
2088                         l_ptr->owner->bclink.last_in = mod(msg_last_bcast(msg));
2089                 break;
2090         case STATE_MSG:
2091
2092                 msg_tol = msg_link_tolerance(msg);
2093                 if (msg_tol)
2094                         link_set_supervision_props(l_ptr, msg_tol);
2095
2096                 if (msg_linkprio(msg) &&
2097                     (msg_linkprio(msg) != l_ptr->priority)) {
2098                         warn("Resetting link <%s>, priority change %u->%u\n",
2099                              l_ptr->name, l_ptr->priority, msg_linkprio(msg));
2100                         l_ptr->priority = msg_linkprio(msg);
2101                         tipc_link_reset(l_ptr); /* Enforce change to take effect */
2102                         break;
2103                 }
2104                 link_state_event(l_ptr, TRAFFIC_MSG_EVT);
2105                 l_ptr->stats.recv_states++;
2106                 if (link_reset_unknown(l_ptr))
2107                         break;
2108
2109                 if (less_eq(mod(l_ptr->next_in_no), msg_next_sent(msg))) {
2110                         rec_gap = mod(msg_next_sent(msg) -
2111                                       mod(l_ptr->next_in_no));
2112                 }
2113
2114                 max_pkt_ack = msg_max_pkt(msg);
2115                 if (max_pkt_ack > l_ptr->max_pkt) {
2116                         l_ptr->max_pkt = max_pkt_ack;
2117                         l_ptr->max_pkt_probes = 0;
2118                 }
2119
2120                 max_pkt_ack = 0;
2121                 if (msg_probe(msg)) {
2122                         l_ptr->stats.recv_probes++;
2123                         if (msg_size(msg) > sizeof(l_ptr->proto_msg))
2124                                 max_pkt_ack = msg_size(msg);
2125                 }
2126
2127                 /* Protocol message before retransmits, reduce loss risk */
2128
2129                 tipc_bclink_check_gap(l_ptr->owner, msg_last_bcast(msg));
2130
2131                 if (rec_gap || (msg_probe(msg))) {
2132                         tipc_link_send_proto_msg(l_ptr, STATE_MSG,
2133                                                  0, rec_gap, 0, 0, max_pkt_ack);
2134                 }
2135                 if (msg_seq_gap(msg)) {
2136                         l_ptr->stats.recv_nacks++;
2137                         tipc_link_retransmit(l_ptr, l_ptr->first_out,
2138                                              msg_seq_gap(msg));
2139                 }
2140                 break;
2141         }
2142 exit:
2143         buf_discard(buf);
2144 }
2145
2146
2147 /*
2148  * tipc_link_tunnel(): Send one message via a link belonging to
2149  * another bearer. Owner node is locked.
2150  */
2151 static void tipc_link_tunnel(struct link *l_ptr,
2152                              struct tipc_msg *tunnel_hdr,
2153                              struct tipc_msg  *msg,
2154                              u32 selector)
2155 {
2156         struct link *tunnel;
2157         struct sk_buff *buf;
2158         u32 length = msg_size(msg);
2159
2160         tunnel = l_ptr->owner->active_links[selector & 1];
2161         if (!tipc_link_is_up(tunnel)) {
2162                 warn("Link changeover error, "
2163                      "tunnel link no longer available\n");
2164                 return;
2165         }
2166         msg_set_size(tunnel_hdr, length + INT_H_SIZE);
2167         buf = tipc_buf_acquire(length + INT_H_SIZE);
2168         if (!buf) {
2169                 warn("Link changeover error, "
2170                      "unable to send tunnel msg\n");
2171                 return;
2172         }
2173         skb_copy_to_linear_data(buf, tunnel_hdr, INT_H_SIZE);
2174         skb_copy_to_linear_data_offset(buf, INT_H_SIZE, msg, length);
2175         tipc_link_send_buf(tunnel, buf);
2176 }
2177
2178
2179
2180 /*
2181  * changeover(): Send whole message queue via the remaining link
2182  *               Owner node is locked.
2183  */
2184
2185 void tipc_link_changeover(struct link *l_ptr)
2186 {
2187         u32 msgcount = l_ptr->out_queue_size;
2188         struct sk_buff *crs = l_ptr->first_out;
2189         struct link *tunnel = l_ptr->owner->active_links[0];
2190         struct tipc_msg tunnel_hdr;
2191         int split_bundles;
2192
2193         if (!tunnel)
2194                 return;
2195
2196         if (!l_ptr->owner->permit_changeover) {
2197                 warn("Link changeover error, "
2198                      "peer did not permit changeover\n");
2199                 return;
2200         }
2201
2202         tipc_msg_init(&tunnel_hdr, CHANGEOVER_PROTOCOL,
2203                  ORIGINAL_MSG, INT_H_SIZE, l_ptr->addr);
2204         msg_set_bearer_id(&tunnel_hdr, l_ptr->peer_bearer_id);
2205         msg_set_msgcnt(&tunnel_hdr, msgcount);
2206
2207         if (!l_ptr->first_out) {
2208                 struct sk_buff *buf;
2209
2210                 buf = tipc_buf_acquire(INT_H_SIZE);
2211                 if (buf) {
2212                         skb_copy_to_linear_data(buf, &tunnel_hdr, INT_H_SIZE);
2213                         msg_set_size(&tunnel_hdr, INT_H_SIZE);
2214                         tipc_link_send_buf(tunnel, buf);
2215                 } else {
2216                         warn("Link changeover error, "
2217                              "unable to send changeover msg\n");
2218                 }
2219                 return;
2220         }
2221
2222         split_bundles = (l_ptr->owner->active_links[0] !=
2223                          l_ptr->owner->active_links[1]);
2224
2225         while (crs) {
2226                 struct tipc_msg *msg = buf_msg(crs);
2227
2228                 if ((msg_user(msg) == MSG_BUNDLER) && split_bundles) {
2229                         struct tipc_msg *m = msg_get_wrapped(msg);
2230                         unchar *pos = (unchar *)m;
2231
2232                         msgcount = msg_msgcnt(msg);
2233                         while (msgcount--) {
2234                                 msg_set_seqno(m, msg_seqno(msg));
2235                                 tipc_link_tunnel(l_ptr, &tunnel_hdr, m,
2236                                                  msg_link_selector(m));
2237                                 pos += align(msg_size(m));
2238                                 m = (struct tipc_msg *)pos;
2239                         }
2240                 } else {
2241                         tipc_link_tunnel(l_ptr, &tunnel_hdr, msg,
2242                                          msg_link_selector(msg));
2243                 }
2244                 crs = crs->next;
2245         }
2246 }
2247
2248 void tipc_link_send_duplicate(struct link *l_ptr, struct link *tunnel)
2249 {
2250         struct sk_buff *iter;
2251         struct tipc_msg tunnel_hdr;
2252
2253         tipc_msg_init(&tunnel_hdr, CHANGEOVER_PROTOCOL,
2254                  DUPLICATE_MSG, INT_H_SIZE, l_ptr->addr);
2255         msg_set_msgcnt(&tunnel_hdr, l_ptr->out_queue_size);
2256         msg_set_bearer_id(&tunnel_hdr, l_ptr->peer_bearer_id);
2257         iter = l_ptr->first_out;
2258         while (iter) {
2259                 struct sk_buff *outbuf;
2260                 struct tipc_msg *msg = buf_msg(iter);
2261                 u32 length = msg_size(msg);
2262
2263                 if (msg_user(msg) == MSG_BUNDLER)
2264                         msg_set_type(msg, CLOSED_MSG);
2265                 msg_set_ack(msg, mod(l_ptr->next_in_no - 1));   /* Update */
2266                 msg_set_bcast_ack(msg, l_ptr->owner->bclink.last_in);
2267                 msg_set_size(&tunnel_hdr, length + INT_H_SIZE);
2268                 outbuf = tipc_buf_acquire(length + INT_H_SIZE);
2269                 if (outbuf == NULL) {
2270                         warn("Link changeover error, "
2271                              "unable to send duplicate msg\n");
2272                         return;
2273                 }
2274                 skb_copy_to_linear_data(outbuf, &tunnel_hdr, INT_H_SIZE);
2275                 skb_copy_to_linear_data_offset(outbuf, INT_H_SIZE, iter->data,
2276                                                length);
2277                 tipc_link_send_buf(tunnel, outbuf);
2278                 if (!tipc_link_is_up(l_ptr))
2279                         return;
2280                 iter = iter->next;
2281         }
2282 }
2283
2284
2285
2286 /**
2287  * buf_extract - extracts embedded TIPC message from another message
2288  * @skb: encapsulating message buffer
2289  * @from_pos: offset to extract from
2290  *
2291  * Returns a new message buffer containing an embedded message.  The
2292  * encapsulating message itself is left unchanged.
2293  */
2294
2295 static struct sk_buff *buf_extract(struct sk_buff *skb, u32 from_pos)
2296 {
2297         struct tipc_msg *msg = (struct tipc_msg *)(skb->data + from_pos);
2298         u32 size = msg_size(msg);
2299         struct sk_buff *eb;
2300
2301         eb = tipc_buf_acquire(size);
2302         if (eb)
2303                 skb_copy_to_linear_data(eb, msg, size);
2304         return eb;
2305 }
2306
2307 /*
2308  *  link_recv_changeover_msg(): Receive tunneled packet sent
2309  *  via other link. Node is locked. Return extracted buffer.
2310  */
2311
2312 static int link_recv_changeover_msg(struct link **l_ptr,
2313                                     struct sk_buff **buf)
2314 {
2315         struct sk_buff *tunnel_buf = *buf;
2316         struct link *dest_link;
2317         struct tipc_msg *msg;
2318         struct tipc_msg *tunnel_msg = buf_msg(tunnel_buf);
2319         u32 msg_typ = msg_type(tunnel_msg);
2320         u32 msg_count = msg_msgcnt(tunnel_msg);
2321
2322         dest_link = (*l_ptr)->owner->links[msg_bearer_id(tunnel_msg)];
2323         if (!dest_link)
2324                 goto exit;
2325         if (dest_link == *l_ptr) {
2326                 err("Unexpected changeover message on link <%s>\n",
2327                     (*l_ptr)->name);
2328                 goto exit;
2329         }
2330         *l_ptr = dest_link;
2331         msg = msg_get_wrapped(tunnel_msg);
2332
2333         if (msg_typ == DUPLICATE_MSG) {
2334                 if (less(msg_seqno(msg), mod(dest_link->next_in_no)))
2335                         goto exit;
2336                 *buf = buf_extract(tunnel_buf, INT_H_SIZE);
2337                 if (*buf == NULL) {
2338                         warn("Link changeover error, duplicate msg dropped\n");
2339                         goto exit;
2340                 }
2341                 buf_discard(tunnel_buf);
2342                 return 1;
2343         }
2344
2345         /* First original message ?: */
2346
2347         if (tipc_link_is_up(dest_link)) {
2348                 info("Resetting link <%s>, changeover initiated by peer\n",
2349                      dest_link->name);
2350                 tipc_link_reset(dest_link);
2351                 dest_link->exp_msg_count = msg_count;
2352                 if (!msg_count)
2353                         goto exit;
2354         } else if (dest_link->exp_msg_count == START_CHANGEOVER) {
2355                 dest_link->exp_msg_count = msg_count;
2356                 if (!msg_count)
2357                         goto exit;
2358         }
2359
2360         /* Receive original message */
2361
2362         if (dest_link->exp_msg_count == 0) {
2363                 warn("Link switchover error, "
2364                      "got too many tunnelled messages\n");
2365                 goto exit;
2366         }
2367         dest_link->exp_msg_count--;
2368         if (less(msg_seqno(msg), dest_link->reset_checkpoint)) {
2369                 goto exit;
2370         } else {
2371                 *buf = buf_extract(tunnel_buf, INT_H_SIZE);
2372                 if (*buf != NULL) {
2373                         buf_discard(tunnel_buf);
2374                         return 1;
2375                 } else {
2376                         warn("Link changeover error, original msg dropped\n");
2377                 }
2378         }
2379 exit:
2380         *buf = NULL;
2381         buf_discard(tunnel_buf);
2382         return 0;
2383 }
2384
2385 /*
2386  *  Bundler functionality:
2387  */
2388 void tipc_link_recv_bundle(struct sk_buff *buf)
2389 {
2390         u32 msgcount = msg_msgcnt(buf_msg(buf));
2391         u32 pos = INT_H_SIZE;
2392         struct sk_buff *obuf;
2393
2394         while (msgcount--) {
2395                 obuf = buf_extract(buf, pos);
2396                 if (obuf == NULL) {
2397                         warn("Link unable to unbundle message(s)\n");
2398                         break;
2399                 }
2400                 pos += align(msg_size(buf_msg(obuf)));
2401                 tipc_net_route_msg(obuf);
2402         }
2403         buf_discard(buf);
2404 }
2405
2406 /*
2407  *  Fragmentation/defragmentation:
2408  */
2409
2410
2411 /*
2412  * link_send_long_buf: Entry for buffers needing fragmentation.
2413  * The buffer is complete, inclusive total message length.
2414  * Returns user data length.
2415  */
2416 static int link_send_long_buf(struct link *l_ptr, struct sk_buff *buf)
2417 {
2418         struct sk_buff *buf_chain = NULL;
2419         struct sk_buff *buf_chain_tail = (struct sk_buff *)&buf_chain;
2420         struct tipc_msg *inmsg = buf_msg(buf);
2421         struct tipc_msg fragm_hdr;
2422         u32 insize = msg_size(inmsg);
2423         u32 dsz = msg_data_sz(inmsg);
2424         unchar *crs = buf->data;
2425         u32 rest = insize;
2426         u32 pack_sz = l_ptr->max_pkt;
2427         u32 fragm_sz = pack_sz - INT_H_SIZE;
2428         u32 fragm_no = 0;
2429         u32 destaddr;
2430
2431         if (msg_short(inmsg))
2432                 destaddr = l_ptr->addr;
2433         else
2434                 destaddr = msg_destnode(inmsg);
2435
2436         /* Prepare reusable fragment header: */
2437
2438         tipc_msg_init(&fragm_hdr, MSG_FRAGMENTER, FIRST_FRAGMENT,
2439                  INT_H_SIZE, destaddr);
2440
2441         /* Chop up message: */
2442
2443         while (rest > 0) {
2444                 struct sk_buff *fragm;
2445
2446                 if (rest <= fragm_sz) {
2447                         fragm_sz = rest;
2448                         msg_set_type(&fragm_hdr, LAST_FRAGMENT);
2449                 }
2450                 fragm = tipc_buf_acquire(fragm_sz + INT_H_SIZE);
2451                 if (fragm == NULL) {
2452                         buf_discard(buf);
2453                         while (buf_chain) {
2454                                 buf = buf_chain;
2455                                 buf_chain = buf_chain->next;
2456                                 buf_discard(buf);
2457                         }
2458                         return -ENOMEM;
2459                 }
2460                 msg_set_size(&fragm_hdr, fragm_sz + INT_H_SIZE);
2461                 fragm_no++;
2462                 msg_set_fragm_no(&fragm_hdr, fragm_no);
2463                 skb_copy_to_linear_data(fragm, &fragm_hdr, INT_H_SIZE);
2464                 skb_copy_to_linear_data_offset(fragm, INT_H_SIZE, crs,
2465                                                fragm_sz);
2466                 buf_chain_tail->next = fragm;
2467                 buf_chain_tail = fragm;
2468
2469                 rest -= fragm_sz;
2470                 crs += fragm_sz;
2471                 msg_set_type(&fragm_hdr, FRAGMENT);
2472         }
2473         buf_discard(buf);
2474
2475         /* Append chain of fragments to send queue & send them */
2476
2477         l_ptr->long_msg_seq_no++;
2478         link_add_chain_to_outqueue(l_ptr, buf_chain, l_ptr->long_msg_seq_no);
2479         l_ptr->stats.sent_fragments += fragm_no;
2480         l_ptr->stats.sent_fragmented++;
2481         tipc_link_push_queue(l_ptr);
2482
2483         return dsz;
2484 }
2485
2486 /*
2487  * A pending message being re-assembled must store certain values
2488  * to handle subsequent fragments correctly. The following functions
2489  * help storing these values in unused, available fields in the
2490  * pending message. This makes dynamic memory allocation unnecessary.
2491  */
2492
2493 static void set_long_msg_seqno(struct sk_buff *buf, u32 seqno)
2494 {
2495         msg_set_seqno(buf_msg(buf), seqno);
2496 }
2497
2498 static u32 get_fragm_size(struct sk_buff *buf)
2499 {
2500         return msg_ack(buf_msg(buf));
2501 }
2502
2503 static void set_fragm_size(struct sk_buff *buf, u32 sz)
2504 {
2505         msg_set_ack(buf_msg(buf), sz);
2506 }
2507
2508 static u32 get_expected_frags(struct sk_buff *buf)
2509 {
2510         return msg_bcast_ack(buf_msg(buf));
2511 }
2512
2513 static void set_expected_frags(struct sk_buff *buf, u32 exp)
2514 {
2515         msg_set_bcast_ack(buf_msg(buf), exp);
2516 }
2517
2518 static u32 get_timer_cnt(struct sk_buff *buf)
2519 {
2520         return msg_reroute_cnt(buf_msg(buf));
2521 }
2522
2523 static void incr_timer_cnt(struct sk_buff *buf)
2524 {
2525         msg_incr_reroute_cnt(buf_msg(buf));
2526 }
2527
2528 /*
2529  * tipc_link_recv_fragment(): Called with node lock on. Returns
2530  * the reassembled buffer if message is complete.
2531  */
2532 int tipc_link_recv_fragment(struct sk_buff **pending, struct sk_buff **fb,
2533                             struct tipc_msg **m)
2534 {
2535         struct sk_buff *prev = NULL;
2536         struct sk_buff *fbuf = *fb;
2537         struct tipc_msg *fragm = buf_msg(fbuf);
2538         struct sk_buff *pbuf = *pending;
2539         u32 long_msg_seq_no = msg_long_msgno(fragm);
2540
2541         *fb = NULL;
2542
2543         /* Is there an incomplete message waiting for this fragment? */
2544
2545         while (pbuf && ((msg_seqno(buf_msg(pbuf)) != long_msg_seq_no) ||
2546                         (msg_orignode(fragm) != msg_orignode(buf_msg(pbuf))))) {
2547                 prev = pbuf;
2548                 pbuf = pbuf->next;
2549         }
2550
2551         if (!pbuf && (msg_type(fragm) == FIRST_FRAGMENT)) {
2552                 struct tipc_msg *imsg = (struct tipc_msg *)msg_data(fragm);
2553                 u32 msg_sz = msg_size(imsg);
2554                 u32 fragm_sz = msg_data_sz(fragm);
2555                 u32 exp_fragm_cnt = msg_sz/fragm_sz + !!(msg_sz % fragm_sz);
2556                 u32 max =  TIPC_MAX_USER_MSG_SIZE + NAMED_H_SIZE;
2557                 if (msg_type(imsg) == TIPC_MCAST_MSG)
2558                         max = TIPC_MAX_USER_MSG_SIZE + MCAST_H_SIZE;
2559                 if (msg_size(imsg) > max) {
2560                         buf_discard(fbuf);
2561                         return 0;
2562                 }
2563                 pbuf = tipc_buf_acquire(msg_size(imsg));
2564                 if (pbuf != NULL) {
2565                         pbuf->next = *pending;
2566                         *pending = pbuf;
2567                         skb_copy_to_linear_data(pbuf, imsg,
2568                                                 msg_data_sz(fragm));
2569                         /*  Prepare buffer for subsequent fragments. */
2570
2571                         set_long_msg_seqno(pbuf, long_msg_seq_no);
2572                         set_fragm_size(pbuf, fragm_sz);
2573                         set_expected_frags(pbuf, exp_fragm_cnt - 1);
2574                 } else {
2575                         warn("Link unable to reassemble fragmented message\n");
2576                 }
2577                 buf_discard(fbuf);
2578                 return 0;
2579         } else if (pbuf && (msg_type(fragm) != FIRST_FRAGMENT)) {
2580                 u32 dsz = msg_data_sz(fragm);
2581                 u32 fsz = get_fragm_size(pbuf);
2582                 u32 crs = ((msg_fragm_no(fragm) - 1) * fsz);
2583                 u32 exp_frags = get_expected_frags(pbuf) - 1;
2584                 skb_copy_to_linear_data_offset(pbuf, crs,
2585                                                msg_data(fragm), dsz);
2586                 buf_discard(fbuf);
2587
2588                 /* Is message complete? */
2589
2590                 if (exp_frags == 0) {
2591                         if (prev)
2592                                 prev->next = pbuf->next;
2593                         else
2594                                 *pending = pbuf->next;
2595                         msg_reset_reroute_cnt(buf_msg(pbuf));
2596                         *fb = pbuf;
2597                         *m = buf_msg(pbuf);
2598                         return 1;
2599                 }
2600                 set_expected_frags(pbuf, exp_frags);
2601                 return 0;
2602         }
2603         buf_discard(fbuf);
2604         return 0;
2605 }
2606
2607 /**
2608  * link_check_defragm_bufs - flush stale incoming message fragments
2609  * @l_ptr: pointer to link
2610  */
2611
2612 static void link_check_defragm_bufs(struct link *l_ptr)
2613 {
2614         struct sk_buff *prev = NULL;
2615         struct sk_buff *next = NULL;
2616         struct sk_buff *buf = l_ptr->defragm_buf;
2617
2618         if (!buf)
2619                 return;
2620         if (!link_working_working(l_ptr))
2621                 return;
2622         while (buf) {
2623                 u32 cnt = get_timer_cnt(buf);
2624
2625                 next = buf->next;
2626                 if (cnt < 4) {
2627                         incr_timer_cnt(buf);
2628                         prev = buf;
2629                 } else {
2630                         if (prev)
2631                                 prev->next = buf->next;
2632                         else
2633                                 l_ptr->defragm_buf = buf->next;
2634                         buf_discard(buf);
2635                 }
2636                 buf = next;
2637         }
2638 }
2639
2640
2641
2642 static void link_set_supervision_props(struct link *l_ptr, u32 tolerance)
2643 {
2644         if ((tolerance < TIPC_MIN_LINK_TOL) || (tolerance > TIPC_MAX_LINK_TOL))
2645                 return;
2646
2647         l_ptr->tolerance = tolerance;
2648         l_ptr->continuity_interval =
2649                 ((tolerance / 4) > 500) ? 500 : tolerance / 4;
2650         l_ptr->abort_limit = tolerance / (l_ptr->continuity_interval / 4);
2651 }
2652
2653
2654 void tipc_link_set_queue_limits(struct link *l_ptr, u32 window)
2655 {
2656         /* Data messages from this node, inclusive FIRST_FRAGM */
2657         l_ptr->queue_limit[TIPC_LOW_IMPORTANCE] = window;
2658         l_ptr->queue_limit[TIPC_MEDIUM_IMPORTANCE] = (window / 3) * 4;
2659         l_ptr->queue_limit[TIPC_HIGH_IMPORTANCE] = (window / 3) * 5;
2660         l_ptr->queue_limit[TIPC_CRITICAL_IMPORTANCE] = (window / 3) * 6;
2661         /* Transiting data messages,inclusive FIRST_FRAGM */
2662         l_ptr->queue_limit[TIPC_LOW_IMPORTANCE + 4] = 300;
2663         l_ptr->queue_limit[TIPC_MEDIUM_IMPORTANCE + 4] = 600;
2664         l_ptr->queue_limit[TIPC_HIGH_IMPORTANCE + 4] = 900;
2665         l_ptr->queue_limit[TIPC_CRITICAL_IMPORTANCE + 4] = 1200;
2666         l_ptr->queue_limit[CONN_MANAGER] = 1200;
2667         l_ptr->queue_limit[CHANGEOVER_PROTOCOL] = 2500;
2668         l_ptr->queue_limit[NAME_DISTRIBUTOR] = 3000;
2669         /* FRAGMENT and LAST_FRAGMENT packets */
2670         l_ptr->queue_limit[MSG_FRAGMENTER] = 4000;
2671 }
2672
2673 /**
2674  * link_find_link - locate link by name
2675  * @name - ptr to link name string
2676  * @node - ptr to area to be filled with ptr to associated node
2677  *
2678  * Caller must hold 'tipc_net_lock' to ensure node and bearer are not deleted;
2679  * this also prevents link deletion.
2680  *
2681  * Returns pointer to link (or 0 if invalid link name).
2682  */
2683
2684 static struct link *link_find_link(const char *name, struct tipc_node **node)
2685 {
2686         struct link_name link_name_parts;
2687         struct tipc_bearer *b_ptr;
2688         struct link *l_ptr;
2689
2690         if (!link_name_validate(name, &link_name_parts))
2691                 return NULL;
2692
2693         b_ptr = tipc_bearer_find_interface(link_name_parts.if_local);
2694         if (!b_ptr)
2695                 return NULL;
2696
2697         *node = tipc_node_find(link_name_parts.addr_peer);
2698         if (!*node)
2699                 return NULL;
2700
2701         l_ptr = (*node)->links[b_ptr->identity];
2702         if (!l_ptr || strcmp(l_ptr->name, name))
2703                 return NULL;
2704
2705         return l_ptr;
2706 }
2707
2708 struct sk_buff *tipc_link_cmd_config(const void *req_tlv_area, int req_tlv_space,
2709                                      u16 cmd)
2710 {
2711         struct tipc_link_config *args;
2712         u32 new_value;
2713         struct link *l_ptr;
2714         struct tipc_node *node;
2715         int res;
2716
2717         if (!TLV_CHECK(req_tlv_area, req_tlv_space, TIPC_TLV_LINK_CONFIG))
2718                 return tipc_cfg_reply_error_string(TIPC_CFG_TLV_ERROR);
2719
2720         args = (struct tipc_link_config *)TLV_DATA(req_tlv_area);
2721         new_value = ntohl(args->value);
2722
2723         if (!strcmp(args->name, tipc_bclink_name)) {
2724                 if ((cmd == TIPC_CMD_SET_LINK_WINDOW) &&
2725                     (tipc_bclink_set_queue_limits(new_value) == 0))
2726                         return tipc_cfg_reply_none();
2727                 return tipc_cfg_reply_error_string(TIPC_CFG_NOT_SUPPORTED
2728                                                    " (cannot change setting on broadcast link)");
2729         }
2730
2731         read_lock_bh(&tipc_net_lock);
2732         l_ptr = link_find_link(args->name, &node);
2733         if (!l_ptr) {
2734                 read_unlock_bh(&tipc_net_lock);
2735                 return tipc_cfg_reply_error_string("link not found");
2736         }
2737
2738         tipc_node_lock(node);
2739         res = -EINVAL;
2740         switch (cmd) {
2741         case TIPC_CMD_SET_LINK_TOL:
2742                 if ((new_value >= TIPC_MIN_LINK_TOL) &&
2743                     (new_value <= TIPC_MAX_LINK_TOL)) {
2744                         link_set_supervision_props(l_ptr, new_value);
2745                         tipc_link_send_proto_msg(l_ptr, STATE_MSG,
2746                                                  0, 0, new_value, 0, 0);
2747                         res = 0;
2748                 }
2749                 break;
2750         case TIPC_CMD_SET_LINK_PRI:
2751                 if ((new_value >= TIPC_MIN_LINK_PRI) &&
2752                     (new_value <= TIPC_MAX_LINK_PRI)) {
2753                         l_ptr->priority = new_value;
2754                         tipc_link_send_proto_msg(l_ptr, STATE_MSG,
2755                                                  0, 0, 0, new_value, 0);
2756                         res = 0;
2757                 }
2758                 break;
2759         case TIPC_CMD_SET_LINK_WINDOW:
2760                 if ((new_value >= TIPC_MIN_LINK_WIN) &&
2761                     (new_value <= TIPC_MAX_LINK_WIN)) {
2762                         tipc_link_set_queue_limits(l_ptr, new_value);
2763                         res = 0;
2764                 }
2765                 break;
2766         }
2767         tipc_node_unlock(node);
2768
2769         read_unlock_bh(&tipc_net_lock);
2770         if (res)
2771                 return tipc_cfg_reply_error_string("cannot change link setting");
2772
2773         return tipc_cfg_reply_none();
2774 }
2775
2776 /**
2777  * link_reset_statistics - reset link statistics
2778  * @l_ptr: pointer to link
2779  */
2780
2781 static void link_reset_statistics(struct link *l_ptr)
2782 {
2783         memset(&l_ptr->stats, 0, sizeof(l_ptr->stats));
2784         l_ptr->stats.sent_info = l_ptr->next_out_no;
2785         l_ptr->stats.recv_info = l_ptr->next_in_no;
2786 }
2787
2788 struct sk_buff *tipc_link_cmd_reset_stats(const void *req_tlv_area, int req_tlv_space)
2789 {
2790         char *link_name;
2791         struct link *l_ptr;
2792         struct tipc_node *node;
2793
2794         if (!TLV_CHECK(req_tlv_area, req_tlv_space, TIPC_TLV_LINK_NAME))
2795                 return tipc_cfg_reply_error_string(TIPC_CFG_TLV_ERROR);
2796
2797         link_name = (char *)TLV_DATA(req_tlv_area);
2798         if (!strcmp(link_name, tipc_bclink_name)) {
2799                 if (tipc_bclink_reset_stats())
2800                         return tipc_cfg_reply_error_string("link not found");
2801                 return tipc_cfg_reply_none();
2802         }
2803
2804         read_lock_bh(&tipc_net_lock);
2805         l_ptr = link_find_link(link_name, &node);
2806         if (!l_ptr) {
2807                 read_unlock_bh(&tipc_net_lock);
2808                 return tipc_cfg_reply_error_string("link not found");
2809         }
2810
2811         tipc_node_lock(node);
2812         link_reset_statistics(l_ptr);
2813         tipc_node_unlock(node);
2814         read_unlock_bh(&tipc_net_lock);
2815         return tipc_cfg_reply_none();
2816 }
2817
2818 /**
2819  * percent - convert count to a percentage of total (rounding up or down)
2820  */
2821
2822 static u32 percent(u32 count, u32 total)
2823 {
2824         return (count * 100 + (total / 2)) / total;
2825 }
2826
2827 /**
2828  * tipc_link_stats - print link statistics
2829  * @name: link name
2830  * @buf: print buffer area
2831  * @buf_size: size of print buffer area
2832  *
2833  * Returns length of print buffer data string (or 0 if error)
2834  */
2835
2836 static int tipc_link_stats(const char *name, char *buf, const u32 buf_size)
2837 {
2838         struct print_buf pb;
2839         struct link *l_ptr;
2840         struct tipc_node *node;
2841         char *status;
2842         u32 profile_total = 0;
2843
2844         if (!strcmp(name, tipc_bclink_name))
2845                 return tipc_bclink_stats(buf, buf_size);
2846
2847         tipc_printbuf_init(&pb, buf, buf_size);
2848
2849         read_lock_bh(&tipc_net_lock);
2850         l_ptr = link_find_link(name, &node);
2851         if (!l_ptr) {
2852                 read_unlock_bh(&tipc_net_lock);
2853                 return 0;
2854         }
2855         tipc_node_lock(node);
2856
2857         if (tipc_link_is_active(l_ptr))
2858                 status = "ACTIVE";
2859         else if (tipc_link_is_up(l_ptr))
2860                 status = "STANDBY";
2861         else
2862                 status = "DEFUNCT";
2863         tipc_printf(&pb, "Link <%s>\n"
2864                          "  %s  MTU:%u  Priority:%u  Tolerance:%u ms"
2865                          "  Window:%u packets\n",
2866                     l_ptr->name, status, l_ptr->max_pkt,
2867                     l_ptr->priority, l_ptr->tolerance, l_ptr->queue_limit[0]);
2868         tipc_printf(&pb, "  RX packets:%u fragments:%u/%u bundles:%u/%u\n",
2869                     l_ptr->next_in_no - l_ptr->stats.recv_info,
2870                     l_ptr->stats.recv_fragments,
2871                     l_ptr->stats.recv_fragmented,
2872                     l_ptr->stats.recv_bundles,
2873                     l_ptr->stats.recv_bundled);
2874         tipc_printf(&pb, "  TX packets:%u fragments:%u/%u bundles:%u/%u\n",
2875                     l_ptr->next_out_no - l_ptr->stats.sent_info,
2876                     l_ptr->stats.sent_fragments,
2877                     l_ptr->stats.sent_fragmented,
2878                     l_ptr->stats.sent_bundles,
2879                     l_ptr->stats.sent_bundled);
2880         profile_total = l_ptr->stats.msg_length_counts;
2881         if (!profile_total)
2882                 profile_total = 1;
2883         tipc_printf(&pb, "  TX profile sample:%u packets  average:%u octets\n"
2884                          "  0-64:%u%% -256:%u%% -1024:%u%% -4096:%u%% "
2885                          "-16384:%u%% -32768:%u%% -66000:%u%%\n",
2886                     l_ptr->stats.msg_length_counts,
2887                     l_ptr->stats.msg_lengths_total / profile_total,
2888                     percent(l_ptr->stats.msg_length_profile[0], profile_total),
2889                     percent(l_ptr->stats.msg_length_profile[1], profile_total),
2890                     percent(l_ptr->stats.msg_length_profile[2], profile_total),
2891                     percent(l_ptr->stats.msg_length_profile[3], profile_total),
2892                     percent(l_ptr->stats.msg_length_profile[4], profile_total),
2893                     percent(l_ptr->stats.msg_length_profile[5], profile_total),
2894                     percent(l_ptr->stats.msg_length_profile[6], profile_total));
2895         tipc_printf(&pb, "  RX states:%u probes:%u naks:%u defs:%u dups:%u\n",
2896                     l_ptr->stats.recv_states,
2897                     l_ptr->stats.recv_probes,
2898                     l_ptr->stats.recv_nacks,
2899                     l_ptr->stats.deferred_recv,
2900                     l_ptr->stats.duplicates);
2901         tipc_printf(&pb, "  TX states:%u probes:%u naks:%u acks:%u dups:%u\n",
2902                     l_ptr->stats.sent_states,
2903                     l_ptr->stats.sent_probes,
2904                     l_ptr->stats.sent_nacks,
2905                     l_ptr->stats.sent_acks,
2906                     l_ptr->stats.retransmitted);
2907         tipc_printf(&pb, "  Congestion bearer:%u link:%u  Send queue max:%u avg:%u\n",
2908                     l_ptr->stats.bearer_congs,
2909                     l_ptr->stats.link_congs,
2910                     l_ptr->stats.max_queue_sz,
2911                     l_ptr->stats.queue_sz_counts
2912                     ? (l_ptr->stats.accu_queue_sz / l_ptr->stats.queue_sz_counts)
2913                     : 0);
2914
2915         tipc_node_unlock(node);
2916         read_unlock_bh(&tipc_net_lock);
2917         return tipc_printbuf_validate(&pb);
2918 }
2919
2920 #define MAX_LINK_STATS_INFO 2000
2921
2922 struct sk_buff *tipc_link_cmd_show_stats(const void *req_tlv_area, int req_tlv_space)
2923 {
2924         struct sk_buff *buf;
2925         struct tlv_desc *rep_tlv;
2926         int str_len;
2927
2928         if (!TLV_CHECK(req_tlv_area, req_tlv_space, TIPC_TLV_LINK_NAME))
2929                 return tipc_cfg_reply_error_string(TIPC_CFG_TLV_ERROR);
2930
2931         buf = tipc_cfg_reply_alloc(TLV_SPACE(MAX_LINK_STATS_INFO));
2932         if (!buf)
2933                 return NULL;
2934
2935         rep_tlv = (struct tlv_desc *)buf->data;
2936
2937         str_len = tipc_link_stats((char *)TLV_DATA(req_tlv_area),
2938                                   (char *)TLV_DATA(rep_tlv), MAX_LINK_STATS_INFO);
2939         if (!str_len) {
2940                 buf_discard(buf);
2941                 return tipc_cfg_reply_error_string("link not found");
2942         }
2943
2944         skb_put(buf, TLV_SPACE(str_len));
2945         TLV_SET(rep_tlv, TIPC_TLV_ULTRA_STRING, NULL, str_len);
2946
2947         return buf;
2948 }
2949
2950 /**
2951  * tipc_link_get_max_pkt - get maximum packet size to use when sending to destination
2952  * @dest: network address of destination node
2953  * @selector: used to select from set of active links
2954  *
2955  * If no active link can be found, uses default maximum packet size.
2956  */
2957
2958 u32 tipc_link_get_max_pkt(u32 dest, u32 selector)
2959 {
2960         struct tipc_node *n_ptr;
2961         struct link *l_ptr;
2962         u32 res = MAX_PKT_DEFAULT;
2963
2964         if (dest == tipc_own_addr)
2965                 return MAX_MSG_SIZE;
2966
2967         read_lock_bh(&tipc_net_lock);
2968         n_ptr = tipc_node_find(dest);
2969         if (n_ptr) {
2970                 tipc_node_lock(n_ptr);
2971                 l_ptr = n_ptr->active_links[selector & 1];
2972                 if (l_ptr)
2973                         res = l_ptr->max_pkt;
2974                 tipc_node_unlock(n_ptr);
2975         }
2976         read_unlock_bh(&tipc_net_lock);
2977         return res;
2978 }
2979
2980 static void link_print(struct link *l_ptr, const char *str)
2981 {
2982         char print_area[256];
2983         struct print_buf pb;
2984         struct print_buf *buf = &pb;
2985
2986         tipc_printbuf_init(buf, print_area, sizeof(print_area));
2987
2988         tipc_printf(buf, str);
2989         tipc_printf(buf, "Link %x<%s>:",
2990                     l_ptr->addr, l_ptr->b_ptr->name);
2991
2992 #ifdef CONFIG_TIPC_DEBUG
2993         if (link_reset_reset(l_ptr) || link_reset_unknown(l_ptr))
2994                 goto print_state;
2995
2996         tipc_printf(buf, ": NXO(%u):", mod(l_ptr->next_out_no));
2997         tipc_printf(buf, "NXI(%u):", mod(l_ptr->next_in_no));
2998         tipc_printf(buf, "SQUE");
2999         if (l_ptr->first_out) {
3000                 tipc_printf(buf, "[%u..", msg_seqno(buf_msg(l_ptr->first_out)));
3001                 if (l_ptr->next_out)
3002                         tipc_printf(buf, "%u..",
3003                                     msg_seqno(buf_msg(l_ptr->next_out)));
3004                 tipc_printf(buf, "%u]", msg_seqno(buf_msg(l_ptr->last_out)));
3005                 if ((mod(msg_seqno(buf_msg(l_ptr->last_out)) -
3006                          msg_seqno(buf_msg(l_ptr->first_out)))
3007                      != (l_ptr->out_queue_size - 1)) ||
3008                     (l_ptr->last_out->next != NULL)) {
3009                         tipc_printf(buf, "\nSend queue inconsistency\n");
3010                         tipc_printf(buf, "first_out= %p ", l_ptr->first_out);
3011                         tipc_printf(buf, "next_out= %p ", l_ptr->next_out);
3012                         tipc_printf(buf, "last_out= %p ", l_ptr->last_out);
3013                 }
3014         } else
3015                 tipc_printf(buf, "[]");
3016         tipc_printf(buf, "SQSIZ(%u)", l_ptr->out_queue_size);
3017         if (l_ptr->oldest_deferred_in) {
3018                 u32 o = msg_seqno(buf_msg(l_ptr->oldest_deferred_in));
3019                 u32 n = msg_seqno(buf_msg(l_ptr->newest_deferred_in));
3020                 tipc_printf(buf, ":RQUE[%u..%u]", o, n);
3021                 if (l_ptr->deferred_inqueue_sz != mod((n + 1) - o)) {
3022                         tipc_printf(buf, ":RQSIZ(%u)",
3023                                     l_ptr->deferred_inqueue_sz);
3024                 }
3025         }
3026 print_state:
3027 #endif
3028
3029         if (link_working_unknown(l_ptr))
3030                 tipc_printf(buf, ":WU");
3031         else if (link_reset_reset(l_ptr))
3032                 tipc_printf(buf, ":RR");
3033         else if (link_reset_unknown(l_ptr))
3034                 tipc_printf(buf, ":RU");
3035         else if (link_working_working(l_ptr))
3036                 tipc_printf(buf, ":WW");
3037         tipc_printf(buf, "\n");
3038
3039         tipc_printbuf_validate(buf);
3040         info("%s", print_area);
3041 }
3042