2 * Connection oriented routing
3 * Copyright (C) 2007-2010 Michael Blizek
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; either version 2
8 * of the License, or (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
24 * Splited packet data format:
25 * announce proto version [4]
26 * is 0, may be increased if format changes
28 * starts with 0, increments every time the data field changes
30 * total data size of all merged packets
32 * used to determine the order when merging the split packet
35 * commulative checksum [8] (not yet)
36 * chunk 1 contains the checksum of the data in chunk 1
37 * chunk 2 contains the checksum of the data in chunk 1+2
40 * Data format of the announce packet "data" field:
41 * min_announce_proto_version [4]
42 * max_announce_proto_version [4]
43 * min_cor_proto_version [4]
44 * max_cor_proto_version [4]
45 * versions which are understood
49 * commanddata [commandlength]
54 #define NEIGHCMD_ADDADDR 1
59 * addrtype [addrtypelen]
65 DEFINE_MUTEX(neighbor_operation_lock
);
67 char *addrtype
= "id";
73 struct kmem_cache
*nb_slab
;
75 LIST_HEAD(announce_out_list
);
77 struct notifier_block netdev_notify
;
80 #define ADDRTYPE_UNKNOWN 0
83 static int get_addrtype(__u32 addrtypelen
, char *addrtype
)
85 if (addrtypelen
== 2 &&
86 (addrtype
[0] == 'i' || addrtype
[0] == 'I') &&
87 (addrtype
[1] == 'd' || addrtype
[1] == 'D'))
90 return ADDRTYPE_UNKNOWN
;
93 void neighbor_free(struct kref
*ref
)
95 struct neighbor
*nb
= container_of(ref
, struct neighbor
, ref
);
96 printk(KERN_ERR
"neighbor free");
97 BUG_ON(nb
->nb_list
.next
!= LIST_POISON1
);
98 BUG_ON(nb
->nb_list
.prev
!= LIST_POISON2
);
105 kmem_cache_free(nb_slab
, nb
);
108 static struct neighbor
*alloc_neighbor(gfp_t allocflags
)
110 struct neighbor
*nb
= kmem_cache_alloc(nb_slab
, allocflags
);
113 if (unlikely(nb
== 0))
116 memset(nb
, 0, sizeof(struct neighbor
));
118 kref_init(&(nb
->ref
));
119 mutex_init(&(nb
->cmsg_lock
));
120 INIT_LIST_HEAD(&(nb
->control_msgs_out
));
121 INIT_LIST_HEAD(&(nb
->ucontrol_msgs_out
));
122 nb
->last_ping_time
= jiffies
;
123 atomic_set(&(nb
->ooo_packets
), 0);
124 spin_lock_init(&(nb
->credits_lock
));
125 nb
->jiffies_credit_update
= nb
->last_ping_time
;
126 get_random_bytes((char *) &seqno
, sizeof(seqno
));
127 mutex_init(&(nb
->pingcookie_lock
));
128 atomic_set(&(nb
->latency
), 1000000);
129 atomic_set(&(nb
->max_remote_cmsg_delay
), 1000000);
130 spin_lock_init(&(nb
->state_lock
));
131 atomic_set(&(nb
->kpacket_seqno
), seqno
);
132 mutex_init(&(nb
->conn_list_lock
));
133 INIT_LIST_HEAD(&(nb
->rcv_conn_list
));
134 INIT_LIST_HEAD(&(nb
->snd_conn_list
));
135 spin_lock_init(&(nb
->retrans_lock
));
136 INIT_LIST_HEAD(&(nb
->retrans_list
));
137 INIT_LIST_HEAD(&(nb
->retrans_list_conn
));
142 struct neighbor
*get_neigh_by_mac(struct sk_buff
*skb
)
144 struct list_head
*currlh
;
145 struct neighbor
*ret
= 0;
148 char source_hw
[MAX_ADDR_LEN
];
149 memset(source_hw
, 0, MAX_ADDR_LEN
);
150 if (skb
->dev
->header_ops
!= 0 &&
151 skb
->dev
->header_ops
->parse
!= 0)
152 skb
->dev
->header_ops
->parse(skb
, source_hw
);
154 mutex_lock(&(neighbor_operation_lock
));
156 currlh
= nb_list
.next
;
158 while (currlh
!= &nb_list
) {
159 struct neighbor
*curr
= container_of(currlh
, struct neighbor
,
162 if (memcmp(curr
->mac
, source_hw
, MAX_ADDR_LEN
) == 0) {
164 kref_get(&(ret
->ref
));
168 currlh
= currlh
->next
;
171 mutex_unlock(&(neighbor_operation_lock
));
176 struct neighbor
*find_neigh(__u16 addrtypelen
, __u8
*addrtype
,
177 __u16 addrlen
, __u8
*addr
)
179 struct list_head
*currlh
;
180 struct neighbor
*ret
= 0;
182 if (get_addrtype(addrtypelen
, addrtype
) != ADDRTYPE_ID
)
185 mutex_lock(&(neighbor_operation_lock
));
187 currlh
= nb_list
.next
;
189 while (currlh
!= &nb_list
) {
190 struct neighbor
*curr
= container_of(currlh
, struct neighbor
,
193 if (curr
->addrlen
== addrlen
&& memcmp(curr
->addr
, addr
,
196 kref_get(&(ret
->ref
));
201 currlh
= currlh
->next
;
205 mutex_unlock(&(neighbor_operation_lock
));
215 * credit exchange factor + unstable flag
216 * throughput bound conns: throughput,credits/msecs
217 * latency bound conns: latency (ms), credits/byte
220 __u32
generate_neigh_list(char *buf
, __u32 buflen
, __u32 limit
, __u32 offset
)
222 struct list_head
*currlh
;
229 __u32 buf_offset
= 8;
230 __u32 headoffset
= 0;
237 mutex_lock(&(neighbor_operation_lock
));
239 currlh
= nb_list
.next
;
241 while (currlh
!= &nb_list
) {
242 struct neighbor
*curr
= container_of(currlh
, struct neighbor
,
245 unsigned long iflags
;
247 /* get_neigh_state not used here because it would deadlock */
248 spin_lock_irqsave( &(curr
->state_lock
), iflags
);
250 spin_unlock_irqrestore( &(curr
->state_lock
), iflags
);
252 if (state
!= NEIGHBOR_STATE_ACTIVE
)
258 if (unlikely(buflen
< buf_offset
+ 4 + 4 + 2 + 4 +
266 rc
= encode_len(buf
+ buf_offset
, buflen
- buf_offset
, 1);
271 rc
= encode_len(buf
+ buf_offset
, buflen
- buf_offset
, 2);
276 rc
= encode_len(buf
+ buf_offset
, buflen
- buf_offset
,
281 buf
[buf_offset
] = 'i'; /* addrtype */
283 buf
[buf_offset
] = 'd';
285 BUG_ON(curr
->addrlen
> buflen
- buf_offset
);
286 memcpy(buf
+ buf_offset
, curr
->addr
, curr
->addrlen
); /* addr */
287 buf_offset
+= curr
->addrlen
;
289 BUG_ON(buf_offset
> buflen
);
296 currlh
= currlh
->next
;
299 mutex_unlock(&(neighbor_operation_lock
));
301 rc
= encode_len(buf
, 4, total
);
305 rc
= encode_len(buf
+ headoffset
, 4, cnt
);
309 if (likely(headoffset
< 8))
310 memmove(buf
+headoffset
, buf
+8, buf_offset
);
312 return buf_offset
+ headoffset
- 8;
315 void set_last_routdtrip(struct neighbor
*nb
, unsigned long time
)
317 unsigned long iflags
;
321 spin_lock_irqsave( &(nb
->state_lock
), iflags
);
323 if(likely(nb
->state
== NEIGHBOR_STATE_ACTIVE
) && time_after(time
,
324 nb
->state_time
.last_roundtrip
))
325 nb
->state_time
.last_roundtrip
= time
;
327 spin_unlock_irqrestore( &(nb
->state_lock
), iflags
);
330 static void _refresh_initial_debitsrate(struct net_device
*dev
,
334 struct list_head
*currlh
;
336 currlh
= nb_list
.next
;
338 while (currlh
!= &nb_list
) {
339 struct neighbor
*curr
= container_of(currlh
, struct neighbor
,
342 if (curr
->dev
== dev
)
345 currlh
= currlh
->next
;
348 currlh
= nb_list
.next
;
350 while (currlh
!= &nb_list
) {
351 struct neighbor
*curr
= container_of(currlh
, struct neighbor
,
354 if (curr
->dev
== dev
)
355 set_debitrate_initial(curr
,
356 debitsrate
/neighbors
);
358 currlh
= currlh
->next
;
362 /* neighbor operation lock has to be held while calling this */
363 static void refresh_initial_debitsrate(void)
365 struct list_head
*currlh1
;
368 currlh1
= nb_list
.next
;
370 while (currlh1
!= &nb_list
) {
371 struct neighbor
*curr1
= container_of(currlh1
, struct neighbor
,
374 struct list_head
*currlh2
;
375 currlh2
= nb_list
.next
;
376 while (currlh2
!= currlh1
) {
377 struct neighbor
*curr2
= container_of(currlh2
,
378 struct neighbor
, nb_list
);
379 if (curr1
->dev
== curr2
->dev
)
387 currlh1
= currlh1
->next
;
390 currlh1
= nb_list
.next
;
392 while (currlh1
!= &nb_list
) {
393 struct neighbor
*curr1
= container_of(currlh1
, struct neighbor
,
396 struct list_head
*currlh2
;
397 currlh2
= nb_list
.next
;
398 while (currlh2
!= currlh1
) {
399 struct neighbor
*curr2
= container_of(currlh2
,
400 struct neighbor
, nb_list
);
401 if (curr1
->dev
== curr2
->dev
)
405 _refresh_initial_debitsrate(curr1
->dev
,
406 CREDIT_RATE_INITIAL
/ifcnt
);
410 currlh1
= currlh1
->next
;
414 static void reset_all_conns(struct neighbor
*nb
)
419 mutex_lock(&(nb
->conn_list_lock
));
421 if (list_empty(&(nb
->snd_conn_list
))) {
422 BUG_ON(nb
->num_send_conns
!= 0);
423 mutex_unlock(&(nb
->conn_list_lock
));
427 sconn
= container_of(nb
->snd_conn_list
.next
, struct conn
,
429 BUG_ON(sconn
->targettype
!= TARGET_OUT
);
432 * reset_conn must not be called with conn_list_lock
435 mutex_unlock(&(nb
->conn_list_lock
));
440 static void stall_timer(struct work_struct
*work
)
442 struct neighbor
*nb
= container_of(to_delayed_work(work
),
443 struct neighbor
, stalltimeout_timer
);
448 unsigned long iflags
;
450 spin_lock_irqsave( &(nb
->state_lock
), iflags
);
451 stall_time_ms
= jiffies_to_msecs(jiffies
-
452 nb
->state_time
.last_roundtrip
);
454 if (unlikely(nbstate
!= NEIGHBOR_STATE_STALLED
))
455 nb
->str_timer_pending
= 0;
457 spin_unlock_irqrestore( &(nb
->state_lock
), iflags
);
459 if (unlikely(nbstate
!= NEIGHBOR_STATE_STALLED
)) {
460 kref_put(&(nb
->ref
), neighbor_free
);
464 if (stall_time_ms
< NB_KILL_TIME_MS
) {
465 INIT_DELAYED_WORK(&(nb
->stalltimeout_timer
), stall_timer
);
466 schedule_delayed_work(&(nb
->stalltimeout_timer
),
467 msecs_to_jiffies(NB_KILL_TIME_MS
-
472 printk(KERN_ERR
"reset_all");
476 spin_lock_irqsave( &(nb
->state_lock
), iflags
);
477 nb
->state
= NEIGHBOR_STATE_KILLED
;
478 spin_unlock_irqrestore( &(nb
->state_lock
), iflags
);
480 mutex_lock(&neighbor_operation_lock
);
481 list_del(&(nb
->nb_list
));
482 refresh_initial_debitsrate();
483 mutex_unlock(&neighbor_operation_lock
);
485 kref_put(&(nb
->ref
), neighbor_free
); /* nb_list */
486 kref_put(&(nb
->ref
), neighbor_free
); /* stall_timer */
489 int get_neigh_state(struct neighbor
*nb
)
492 unsigned long iflags
;
498 spin_lock_irqsave( &(nb
->state_lock
), iflags
);
500 if (unlikely(likely(nb
->state
== NEIGHBOR_STATE_ACTIVE
) && unlikely(
501 time_after_eq(jiffies
, nb
->state_time
.last_roundtrip
+
502 msecs_to_jiffies(NB_STALL_TIME_MS
)) && (
503 nb
->ping_intransit
>= NB_STALL_MINPINGS
||
504 nb
->ping_intransit
>= PING_COOKIES_PER_NEIGH
)))) {
505 nb
->state
= NEIGHBOR_STATE_STALLED
;
506 starttimer
= (nb
->str_timer_pending
== 0);
507 stall_time_ms
= jiffies
- nb
->state_time
.last_roundtrip
;
508 nb
->str_timer_pending
= 1;
509 printk(KERN_ERR
"switched to stalled");
510 BUG_ON(nb
->ping_intransit
> PING_COOKIES_PER_NEIGH
);
515 spin_unlock_irqrestore( &(nb
->state_lock
), iflags
);
518 if (unlikely(starttimer
)) {
519 kref_get(&(nb
->ref
));
520 INIT_DELAYED_WORK(&(nb
->stalltimeout_timer
),
522 schedule_delayed_work(&(nb
->stalltimeout_timer
),
523 NB_KILL_TIME_MS
- stall_time_ms
);
529 static struct ping_cookie
*find_cookie(struct neighbor
*nb
, __u32 cookie
)
533 for(i
=0;i
<PING_COOKIES_PER_NEIGH
;i
++) {
534 if (nb
->cookies
[i
].cookie
== cookie
)
535 return &(nb
->cookies
[i
]);
540 void ping_resp(struct neighbor
*nb
, __u32 cookie
, __u32 respdelay
)
542 struct ping_cookie
*c
;
545 unsigned long cookie_sendtime
;
548 unsigned long iflags
;
550 mutex_lock(&(nb
->pingcookie_lock
));
552 c
= find_cookie(nb
, cookie
);
554 if (unlikely(c
== 0))
557 cookie_sendtime
= c
->time
;
559 newlatency
= ((((__s64
) ((__u32
)atomic_read(&(nb
->latency
)))) * 15 +
560 jiffies_to_usecs(jiffies
- c
->time
) - respdelay
) / 16);
561 if (unlikely(newlatency
< 0))
563 if (unlikely(newlatency
> (((__s64
)256)*256*256*256 - 1)))
564 newlatency
= ((__s64
)256)*256*256*256 - 1;
566 atomic_set(&(nb
->latency
), (__u32
) newlatency
);
569 nb
->ping_intransit
--;
571 for(i
=0;i
<PING_COOKIES_PER_NEIGH
;i
++) {
572 if (nb
->cookies
[i
].cookie
!= 0 &&
573 time_before(nb
->cookies
[i
].time
, c
->time
)) {
574 nb
->cookies
[i
].pongs
++;
575 if (nb
->cookies
[i
].pongs
>= PING_PONGLIMIT
) {
576 nb
->cookies
[i
].cookie
= 0;
577 nb
->cookies
[i
].pongs
= 0;
578 nb
->ping_intransit
--;
583 spin_lock_irqsave( &(nb
->state_lock
), iflags
);
585 if (unlikely(nb
->state
== NEIGHBOR_STATE_INITIAL
||
586 nb
->state
== NEIGHBOR_STATE_STALLED
)) {
589 if (nb
->state
== NEIGHBOR_STATE_INITIAL
) {
590 __u64 jiffies64
= get_jiffies_64();
591 if (nb
->state_time
.last_state_change
== 0)
592 nb
->state_time
.last_state_change
= jiffies64
;
593 if (jiffies64
<= (nb
->state_time
.last_state_change
+
594 msecs_to_jiffies(INITIAL_TIME_MS
)))
598 if (nb
->ping_success
>= PING_SUCCESS_CNT
) {
599 /*if (nb->state == NEIGHBOR_STATE_INITIAL)
600 printk(KERN_ERR "switched from initial to active");
602 printk(KERN_ERR "switched from stalled to active");
604 nb
->state
= NEIGHBOR_STATE_ACTIVE
;
605 nb
->ping_success
= 0;
606 nb
->state_time
.last_roundtrip
= jiffies
;
609 nb
->state_time
.last_roundtrip
= cookie_sendtime
;
613 spin_unlock_irqrestore( &(nb
->state_lock
), iflags
);
616 mutex_unlock(&(nb
->pingcookie_lock
));
619 __u32
add_ping_req(struct neighbor
*nb
)
621 struct ping_cookie
*c
;
626 mutex_lock(&(nb
->pingcookie_lock
));
628 for (i
=0;i
<PING_COOKIES_PER_NEIGH
;i
++) {
629 if (nb
->cookies
[i
].cookie
== 0)
633 get_random_bytes((char *) &i
, sizeof(i
));
634 i
= (i
% (PING_COOKIES_PER_NEIGH
- PING_COOKIES_FIFO
)) +
638 c
= &(nb
->cookies
[i
]);
642 if (unlikely(nb
->lastcookie
== 0))
644 c
->cookie
= nb
->lastcookie
;
646 nb
->ping_intransit
++;
650 nb
->last_ping_time
= jiffies
;
652 mutex_unlock(&(nb
->pingcookie_lock
));
657 void unadd_ping_req(struct neighbor
*nb
, __u32 cookie
)
664 mutex_lock(&(nb
->pingcookie_lock
));
666 for (i
=0;i
<PING_COOKIES_PER_NEIGH
;i
++) {
667 if (nb
->cookies
[i
].cookie
== cookie
) {
668 nb
->cookies
[i
].cookie
= 0;
669 nb
->ping_intransit
--;
674 mutex_unlock(&(nb
->pingcookie_lock
));
677 static int neighbor_idle(struct neighbor
*nb
)
680 mutex_lock(&(nb
->conn_list_lock
));
681 ret
= (list_empty(&(nb
->rcv_conn_list
)) &&
682 list_empty(&(nb
->snd_conn_list
)));
683 BUG_ON(list_empty(&(nb
->snd_conn_list
)) && nb
->num_send_conns
!= 0);
684 mutex_unlock(&(nb
->conn_list_lock
));
689 * Check additional to the checks and timings already done in kpacket_gen.c
690 * This is primarily to make sure that we do not invalidate other ping cookies
691 * which might still receive responses. It does this by requiring a certain
692 * mimimum delay between pings, depending on how many pings are already in
695 int time_to_send_ping(struct neighbor
*nb
)
699 int state
= get_neigh_state(nb
);
700 int idle
= (state
!= NEIGHBOR_STATE_ACTIVE
? 0 :
704 mutex_lock(&(nb
->pingcookie_lock
));
705 if (nb
->ping_intransit
>= PING_COOKIES_NOTHROTTLE
) {
706 __u32 mindelay
= (( ((__u32
) atomic_read(&(nb
->latency
))) +
707 ((__u32
) atomic_read(
708 &(nb
->max_remote_cmsg_delay
))) )/1000) <<
709 (nb
->ping_intransit
+ 1 -
710 PING_COOKIES_NOTHROTTLE
);
712 if (mindelay
> PING_THROTTLE_LIMIT_MS
)
713 mindelay
= PING_THROTTLE_LIMIT_MS
;
715 if (jiffies_to_msecs(jiffies
- nb
->last_ping_time
) < mindelay
)
719 if (unlikely(state
!= NEIGHBOR_STATE_ACTIVE
) ||
720 nb
->ping_intransit
!= 0)
721 forcetime
= PING_FORCETIME_MS
;
723 forcetime
= PING_FORCETIME_ACTIVEIDLE_MS
;
725 forcetime
= PING_FORCETIME_ACTIVE_MS
;
727 if (jiffies_to_msecs(jiffies
- nb
->last_ping_time
) < (forcetime
/2))
729 else if (jiffies_to_msecs(jiffies
- nb
->last_ping_time
) >= forcetime
)
732 mutex_unlock(&(nb
->pingcookie_lock
));
737 static void add_neighbor(struct neighbor
*nb
)
739 struct list_head
*currlh
= nb_list
.next
;
741 BUG_ON((nb
->addr
== 0) != (nb
->addrlen
== 0));
743 while (currlh
!= &nb_list
) {
744 struct neighbor
*curr
= container_of(currlh
, struct neighbor
,
747 if (curr
->addrlen
== nb
->addrlen
&& memcmp(curr
->addr
, nb
->addr
,
749 goto already_present
;
751 currlh
= currlh
->next
;
754 /* kref_get not needed here, because the caller leaves its ref to us */
755 printk(KERN_ERR
"add_neigh");
757 list_add_tail(&(nb
->nb_list
), &nb_list
);
758 refresh_initial_debitsrate();
759 schedule_controlmsg_timerfunc(nb
);
760 INIT_DELAYED_WORK(&(nb
->retrans_timer
), retransmit_timerfunc
);
761 INIT_DELAYED_WORK(&(nb
->retrans_timer_conn
), retransmit_conn_timerfunc
);
765 kmem_cache_free(nb_slab
, nb
);
769 static __u32
pull_u32(struct sk_buff
*skb
, int convbo
)
771 char *ptr
= cor_pull_skb(skb
, 4);
777 ((char *)&ret
)[0] = ptr
[0];
778 ((char *)&ret
)[1] = ptr
[1];
779 ((char *)&ret
)[2] = ptr
[2];
780 ((char *)&ret
)[3] = ptr
[3];
783 return be32_to_cpu(ret
);
787 static int apply_announce_addaddr(struct neighbor
*nb
, __u32 cmd
, __u32 len
,
795 BUG_ON((nb
->addr
== 0) != (nb
->addrlen
== 0));
803 addrtypelen
= be16_to_cpu(*((__u16
*) cmddata
));
810 addrlen
= be16_to_cpu(*((__u16
*) cmddata
));
815 cmddata
+= addrtypelen
;
825 if (get_addrtype(addrtypelen
, addrtype
) != ADDRTYPE_ID
)
828 nb
->addr
= kmalloc(addrlen
, GFP_KERNEL
);
829 if (unlikely(nb
->addr
== 0))
832 memcpy(nb
->addr
, addr
, addrlen
);
833 nb
->addrlen
= addrlen
;
838 static void apply_announce_cmd(struct neighbor
*nb
, __u32 cmd
, __u32 len
,
841 if (cmd
== NEIGHCMD_ADDADDR
) {
842 apply_announce_addaddr(nb
, cmd
, len
, cmddata
);
844 /* ignore unknown cmds */
848 static void apply_announce_cmds(char *msg
, __u32 len
, struct net_device
*dev
,
851 struct neighbor
*nb
= alloc_neighbor(GFP_KERNEL
);
853 if (unlikely(nb
== 0))
860 cmd
= be32_to_cpu(*((__u32
*) msg
));
863 cmdlen
= be32_to_cpu(*((__u32
*) msg
));
867 BUG_ON(cmdlen
> len
);
869 apply_announce_cmd(nb
, cmd
, cmdlen
, msg
);
877 memcpy(nb
->mac
, source_hw
, MAX_ADDR_LEN
);
884 static int check_announce_cmds(char *msg
, __u32 len
)
890 cmd
= be32_to_cpu(*((__u32
*) msg
));
893 cmdlen
= be32_to_cpu(*((__u32
*) msg
));
897 /* malformated packet */
898 if (unlikely(cmdlen
> len
))
905 if (unlikely(len
!= 0))
911 static void parse_announce(char *msg
, __u32 len
, struct net_device
*dev
,
914 __u32 min_announce_version
;
915 __u32 max_announce_version
;
916 __u32 min_cor_version
;
917 __u32 max_cor_version
;
919 if (unlikely(len
< 16))
922 min_announce_version
= be32_to_cpu(*((__u32
*) msg
));
925 max_announce_version
= be32_to_cpu(*((__u32
*) msg
));
928 min_cor_version
= be32_to_cpu(*((__u32
*) msg
));
931 max_cor_version
= be32_to_cpu(*((__u32
*) msg
));
935 if (min_announce_version
!= 0)
937 if (min_cor_version
!= 0)
939 if (check_announce_cmds(msg
, len
)) {
942 apply_announce_cmds(msg
, len
, dev
, source_hw
);
946 /* lh has to be first */
948 struct sk_buff_head skbs
; /* sorted by offset */
949 struct net_device
*dev
;
950 char source_hw
[MAX_ADDR_LEN
];
951 __u32 announce_proto_version
;
952 __u32 packet_version
;
955 __u64 last_received_packet
;
958 LIST_HEAD(announce_list
);
960 struct kmem_cache
*announce_in_slab
;
962 static void merge_announce(struct announce_in
*ann
)
964 char *msg
= kmalloc(ann
->total_size
, GFP_KERNEL
);
968 /* try again when next packet arrives */
972 while (copy
!= ann
->total_size
) {
976 struct skb_procstate
*ps
;
978 if (unlikely(skb_queue_empty(&(ann
->skbs
)))) {
979 printk(KERN_ERR
"net/cor/neighbor.c: sk_head ran "
980 "empty while merging packets\n");
984 skb
= skb_dequeue(&(ann
->skbs
));
985 ps
= skb_pstate(skb
);
988 if (unlikely(ps
->funcstate
.announce
.offset
> copy
)) {
989 printk(KERN_ERR
"net/cor/neighbor.c: invalid offset"
994 if (unlikely(ps
->funcstate
.announce
.offset
< copy
)) {
995 offset
= copy
- ps
->funcstate
.announce
.offset
;
999 if (unlikely(currcpy
+ copy
> ann
->total_size
))
1002 memcpy(msg
+ copy
, skb
->data
+ offset
, currcpy
);
1007 parse_announce(msg
, ann
->total_size
, ann
->dev
, ann
->source_hw
);
1014 list_del(&(ann
->lh
));
1015 kmem_cache_free(announce_in_slab
, ann
);
1018 static int _rcv_announce(struct sk_buff
*skb
, struct announce_in
*ann
)
1020 struct skb_procstate
*ps
= skb_pstate(skb
);
1022 __u32 offset
= ps
->funcstate
.announce
.offset
;
1023 __u32 len
= skb
->len
;
1025 __u32 curroffset
= 0;
1026 __u32 prevoffset
= 0;
1029 struct sk_buff
*curr
= ann
->skbs
.next
;
1031 if (unlikely(len
+ offset
> ann
->total_size
)) {
1032 /* invalid header */
1038 * Try to find the right place to insert in the sorted list. This
1039 * means to process the list until we find a skb which has a greater
1040 * offset, so we can insert before it to keep the sort order. However,
1041 * this is complicated by the fact that the new skb must not be inserted
1042 * between 2 skbs if there is no data missing in between. So the loop
1043 * runs has to keep running until there is either a gap to insert or
1044 * we see that this data has already been received.
1046 while ((void *) curr
!= (void *) &(ann
->skbs
)) {
1047 struct skb_procstate
*currps
= skb_pstate(skb
);
1049 curroffset
= currps
->funcstate
.announce
.offset
;
1051 if (curroffset
> offset
&& (prevoffset
+ prevlen
) < curroffset
)
1054 prevoffset
= curroffset
;
1055 prevlen
= curr
->len
;
1058 if ((offset
+len
) <= (prevoffset
+prevlen
)) {
1059 /* we already have this data */
1066 * Calculate how much data was really received, by substracting
1067 * the bytes we already have.
1069 if (unlikely(prevoffset
+ prevlen
> offset
)) {
1070 len
-= (prevoffset
+ prevlen
) - offset
;
1071 offset
= prevoffset
+ prevlen
;
1074 if (unlikely((void *) curr
!= (void *) &(ann
->skbs
) &&
1075 (offset
+ len
) > curroffset
))
1076 len
= curroffset
- offset
;
1078 ann
->received_size
+= len
;
1079 BUG_ON(ann
->received_size
> ann
->total_size
);
1080 __skb_queue_before(&(ann
->skbs
), curr
, skb
);
1081 ann
->last_received_packet
= get_jiffies_64();
1083 if (ann
->received_size
== ann
->total_size
)
1084 merge_announce(ann
);
1085 else if (unlikely(ann
->skbs
.qlen
>= 16))
1091 void rcv_announce(struct sk_buff
*skb
)
1093 struct skb_procstate
*ps
= skb_pstate(skb
);
1094 struct announce_in
*curr
= 0;
1095 struct announce_in
*leastactive
= 0;
1096 __u32 list_size
= 0;
1098 __u32 announce_proto_version
= pull_u32(skb
, 1);
1099 __u32 packet_version
= pull_u32(skb
, 1);
1100 __u32 total_size
= pull_u32(skb
, 1);
1102 char source_hw
[MAX_ADDR_LEN
];
1103 memset(source_hw
, 0, MAX_ADDR_LEN
);
1104 if (skb
->dev
->header_ops
!= 0 &&
1105 skb
->dev
->header_ops
->parse
!= 0)
1106 skb
->dev
->header_ops
->parse(skb
, source_hw
);
1108 ps
->funcstate
.announce
.offset
= pull_u32(skb
, 1);
1110 if (total_size
> 8192)
1113 mutex_lock(&(neighbor_operation_lock
));
1115 if (announce_proto_version
!= 0)
1118 curr
= (struct announce_in
*) announce_list
.next
;
1120 while (((struct list_head
*) curr
) != &(announce_list
)) {
1122 if (curr
->dev
== skb
->dev
&& memcmp(curr
->source_hw
, source_hw
,
1123 MAX_ADDR_LEN
) == 0 &&
1124 curr
->announce_proto_version
==
1125 announce_proto_version
&&
1126 curr
->packet_version
== packet_version
&&
1127 curr
->total_size
== total_size
)
1130 if (leastactive
== 0 || curr
->last_received_packet
<
1131 leastactive
->last_received_packet
)
1134 curr
= (struct announce_in
*) curr
->lh
.next
;
1137 if (list_size
>= 128) {
1138 BUG_ON(leastactive
== 0);
1141 curr
->last_received_packet
= get_jiffies_64();
1143 while (!skb_queue_empty(&(curr
->skbs
))) {
1144 struct sk_buff
*skb2
= skb_dequeue(&(curr
->skbs
));
1150 curr
= kmem_cache_alloc(announce_in_slab
,
1155 skb_queue_head_init(&(curr
->skbs
));
1156 list_add_tail((struct list_head
*) curr
, &announce_list
);
1159 curr
->packet_version
= packet_version
;
1160 curr
->total_size
= total_size
;
1161 curr
->received_size
= 0;
1162 curr
->announce_proto_version
= announce_proto_version
;
1163 curr
->dev
= skb
->dev
;
1164 dev_hold(curr
->dev
);
1165 memcpy(curr
->source_hw
, source_hw
, MAX_ADDR_LEN
);
1168 if (_rcv_announce(skb
, curr
)) {
1169 list_del((struct list_head
*) curr
);
1171 kmem_cache_free(announce_in_slab
, curr
);
1179 mutex_unlock(&(neighbor_operation_lock
));
1185 __u32 packet_version
;
1187 __u32 announce_msg_len
;
1190 struct announce
*last_announce
;
1192 static int send_announce_chunk(struct announce_data
*ann
)
1194 struct sk_buff
*skb
;
1195 __u32 packet_size
= 256;
1196 __u32 remainingdata
= ann
->ann
->announce_msg_len
-
1197 ann
->curr_announce_msg_offset
;
1198 __u32 headroom
= LL_ALLOCATED_SPACE(ann
->dev
);
1199 __u32 overhead
= 17 + headroom
;
1204 if (remainingdata
< packet_size
)
1205 packet_size
= remainingdata
;
1207 skb
= alloc_skb(packet_size
+ overhead
, GFP_KERNEL
);
1208 if (unlikely(skb
== 0))
1211 skb
->protocol
= htons(ETH_P_COR
);
1212 skb
->dev
= ann
->dev
;
1213 skb_reserve(skb
, headroom
);
1215 if(unlikely(dev_hard_header(skb
, ann
->dev
, ETH_P_COR
,
1216 ann
->dev
->broadcast
, ann
->dev
->dev_addr
, skb
->len
) < 0))
1219 skb_reset_network_header(skb
);
1221 header
= skb_put(skb
, 17);
1222 if (unlikely(header
== 0))
1225 header
[0] = PACKET_TYPE_ANNOUNCE
;
1227 put_u32(header
+ 1, 0, 1); /* announce proto version */
1228 put_u32(header
+ 5, ann
->ann
->packet_version
, 1); /* packet version */
1229 put_u32(header
+ 9, ann
->ann
->announce_msg_len
, 1); /* total size */
1230 put_u32(header
+ 13, ann
->curr_announce_msg_offset
, 1); /* offset */
1232 ptr
= skb_put(skb
, packet_size
);
1233 if (unlikely(ptr
== 0))
1236 memcpy(ptr
, ann
->ann
->announce_msg
+ ann
->curr_announce_msg_offset
,
1239 rc
= dev_queue_xmit(skb
);
1242 ann
->curr_announce_msg_offset
+= packet_size
;
1244 if (ann
->curr_announce_msg_offset
== ann
->ann
->announce_msg_len
)
1245 ann
->curr_announce_msg_offset
= 0;
1257 int send_announce_qos(struct announce_data
*ann
)
1260 mutex_lock(&(neighbor_operation_lock
));
1261 rc
= send_announce_chunk(ann
);
1262 mutex_unlock(&(neighbor_operation_lock
));
1266 static void announce_free(struct kref
*ref
)
1268 struct announce
*ann
= container_of(ref
, struct announce
, ref
);
1269 kfree(&(ann
->announce_msg
));
1273 void announce_data_free(struct kref
*ref
)
1275 struct announce_data
*ann
= container_of(ref
, struct announce_data
,
1278 kref_put(&(ann
->ann
->ref
), announce_free
);
1282 static void send_announce(struct work_struct
*work
)
1284 struct announce_data
*ann
= container_of(to_delayed_work(work
),
1285 struct announce_data
, announce_work
);
1289 mutex_lock(&(neighbor_operation_lock
));
1291 if (unlikely(ann
->dev
== 0))
1295 if (unlikely(ann
->ann
== 0 && last_announce
== 0))
1297 if (ann
->curr_announce_msg_offset
== 0 &&
1298 unlikely(ann
->ann
!= last_announce
)) {
1300 kref_put(&(ann
->ann
->ref
), announce_free
);
1301 ann
->ann
= last_announce
;
1302 kref_get(&(ann
->ann
->ref
));
1305 rc
= send_announce_chunk(ann
);
1308 mutex_unlock(&(neighbor_operation_lock
));
1311 qos_enqueue(ann
->dev
, &(ann
->rb
), QOS_CALLER_ANNOUNCE
);
1313 if (unlikely(reschedule
== 0)) {
1314 kref_put(&(ann
->ref
), announce_data_free
);
1316 __u64 jiffies
= get_jiffies_64();
1319 ann
->scheduled_announce_timer
+= msecs_to_jiffies(
1320 ANNOUNCE_SEND_PACKETINTELVAL_MS
);
1322 delay
= ann
->scheduled_announce_timer
- jiffies
;
1326 INIT_DELAYED_WORK(&(ann
->announce_work
), send_announce
);
1327 schedule_delayed_work(&(ann
->announce_work
), delay
);
1331 static struct announce_data
*get_announce_by_netdev(struct net_device
*dev
)
1333 struct list_head
*lh
= announce_out_list
.next
;
1335 while (lh
!= &announce_out_list
) {
1336 struct announce_data
*curr
= (struct announce_data
*)(
1338 offsetof(struct announce_data
, lh
));
1340 if (curr
->dev
== dev
)
1347 static void announce_send_adddev(struct net_device
*dev
)
1349 struct announce_data
*ann
;
1351 ann
= kmalloc(sizeof(struct announce_data
), GFP_KERNEL
);
1353 if (unlikely(ann
== 0)) {
1354 printk(KERN_ERR
"cor cannot allocate memory for sending "
1359 memset(ann
, 0, sizeof(struct announce_data
));
1361 kref_init(&(ann
->ref
));
1366 mutex_lock(&(neighbor_operation_lock
));
1367 list_add_tail(&(ann
->lh
), &announce_out_list
);
1368 mutex_unlock(&(neighbor_operation_lock
));
1370 ann
->scheduled_announce_timer
= get_jiffies_64();
1371 INIT_DELAYED_WORK(&(ann
->announce_work
), send_announce
);
1372 schedule_delayed_work(&(ann
->announce_work
), 1);
1375 static void announce_send_rmdev(struct net_device
*dev
)
1377 struct announce_data
*ann
;
1379 mutex_lock(&(neighbor_operation_lock
));
1381 ann
= get_announce_by_netdev(dev
);
1390 mutex_unlock(&(neighbor_operation_lock
));
1393 int netdev_notify_func(struct notifier_block
*not, unsigned long event
,
1396 struct net_device
*dev
= (struct net_device
*) ptr
;
1401 rc
= create_queue(dev
);
1404 announce_send_adddev(dev
);
1408 announce_send_rmdev(dev
);
1412 case NETDEV_REGISTER
:
1413 case NETDEV_UNREGISTER
:
1414 case NETDEV_CHANGEMTU
:
1415 case NETDEV_CHANGEADDR
:
1416 case NETDEV_GOING_DOWN
:
1417 case NETDEV_CHANGENAME
:
1418 case NETDEV_FEAT_CHANGE
:
1419 case NETDEV_BONDING_FAILOVER
:
1428 static int set_announce(char *msg
, __u32 len
)
1430 struct announce
*ann
= kmalloc(sizeof(struct announce
), GFP_KERNEL
);
1432 if (unlikely(ann
== 0)) {
1437 memset(ann
, 0, sizeof(struct announce
));
1439 ann
->announce_msg
= msg
;
1440 ann
->announce_msg_len
= len
;
1442 kref_init(&(ann
->ref
));
1444 mutex_lock(&(neighbor_operation_lock
));
1446 if (last_announce
!= 0) {
1447 ann
->packet_version
= last_announce
->packet_version
+ 1;
1448 kref_put(&(last_announce
->ref
), announce_free
);
1451 last_announce
= ann
;
1453 mutex_unlock(&(neighbor_operation_lock
));
1458 static int generate_announce(void)
1460 __u32 addrtypelen
= strlen(addrtype
);
1463 __u32 cmd_hdr_len
= 8;
1464 __u32 cmd_len
= 2 + 2 + addrtypelen
+ addrlen
;
1466 __u32 len
= hdr_len
+ cmd_hdr_len
+ cmd_len
;
1469 char *msg
= kmalloc(len
, GFP_KERNEL
);
1470 if (unlikely(msg
== 0))
1473 put_u32(msg
+ offset
, 0, 1); /* min_announce_proto_version */
1475 put_u32(msg
+ offset
, 0, 1); /* max_announce_proto_version */
1477 put_u32(msg
+ offset
, 0, 1); /* min_cor_proto_version */
1479 put_u32(msg
+ offset
, 0, 1); /* max_cor_proto_version */
1483 put_u32(msg
+ offset
, NEIGHCMD_ADDADDR
, 1); /* command */
1485 put_u32(msg
+ offset
, cmd_len
, 1); /* command length */
1488 /* addrtypelen, addrlen */
1489 put_u16(msg
+ offset
, addrtypelen
, 1);
1491 put_u16(msg
+ offset
, addrlen
, 1);
1494 /* addrtype, addr */
1495 memcpy(msg
+ offset
, addrtype
, addrtypelen
);
1496 offset
+= addrtypelen
;
1497 memcpy(msg
+ offset
, addr
, addrlen
);
1500 BUG_ON(offset
!= len
);
1502 return set_announce(msg
, len
);
1505 int __init
cor_neighbor_init(void)
1509 addr
= kmalloc(addrlen
, GFP_KERNEL
);
1510 if (unlikely(addr
== 0))
1513 get_random_bytes(addr
, addrlen
);
1515 nb_slab
= kmem_cache_create("cor_neighbor", sizeof(struct neighbor
), 8,
1517 announce_in_slab
= kmem_cache_create("cor_announce_in",
1518 sizeof(struct announce_in
), 8, 0, 0);
1520 if (unlikely(generate_announce()))
1523 memset(&netdev_notify
, 0, sizeof(netdev_notify
));
1524 netdev_notify
.notifier_call
= netdev_notify_func
;
1525 register_netdevice_notifier(&netdev_notify
);
1536 MODULE_LICENSE("GPL");