1 /* net/sched/sch_teql.c "True" (or "trivial") link equalizer.
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of the GNU General Public License
5 * as published by the Free Software Foundation; either version
6 * 2 of the License, or (at your option) any later version.
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
11 #include <linux/module.h>
12 #include <linux/types.h>
13 #include <linux/kernel.h>
14 #include <linux/string.h>
15 #include <linux/errno.h>
16 #include <linux/if_arp.h>
17 #include <linux/netdevice.h>
18 #include <linux/init.h>
19 #include <linux/skbuff.h>
20 #include <linux/moduleparam.h>
22 #include <net/neighbour.h>
23 #include <net/pkt_sched.h>
29 After loading this module you will find a new device teqlN
30 and new qdisc with the same name. To join a slave to the equalizer
31 you should just set this qdisc on a device f.e.
33 # tc qdisc add dev eth0 root teql0
34 # tc qdisc add dev eth1 root teql0
36 That's all. Full PnP 8)
41 1. Slave devices MUST be active devices, i.e., they must raise the tbusy
42 signal and generate EOI events. If you want to equalize virtual devices
43 like tunnels, use a normal eql device.
44 2. This device puts no limitations on physical slave characteristics
45 f.e. it will equalize 9600baud line and 100Mb ethernet perfectly :-)
46 Certainly, large difference in link speeds will make the resulting
47 eqalized link unusable, because of huge packet reordering.
48 I estimate an upper useful difference as ~10 times.
49 3. If the slave requires address resolution, only protocols using
50 neighbour cache (IPv4/IPv6) will work over the equalized link.
51 Other protocols are still allowed to use the slave device directly,
52 which will not break load balancing, though native slave
53 traffic will have the highest priority. */
57 struct Qdisc_ops qops
;
58 struct net_device
*dev
;
60 struct list_head master_list
;
61 struct net_device_stats stats
;
64 struct teql_sched_data
67 struct teql_master
*m
;
68 struct neighbour
*ncache
;
69 struct sk_buff_head q
;
72 #define NEXT_SLAVE(q) (((struct teql_sched_data*)qdisc_priv(q))->next)
74 #define FMASK (IFF_BROADCAST|IFF_POINTOPOINT)
76 /* "teql*" qdisc routines */
79 teql_enqueue(struct sk_buff
*skb
, struct Qdisc
* sch
)
81 struct net_device
*dev
= sch
->dev
;
82 struct teql_sched_data
*q
= qdisc_priv(sch
);
84 if (q
->q
.qlen
< dev
->tx_queue_len
) {
85 __skb_queue_tail(&q
->q
, skb
);
86 sch
->bstats
.bytes
+= skb
->len
;
87 sch
->bstats
.packets
++;
97 teql_requeue(struct sk_buff
*skb
, struct Qdisc
* sch
)
99 struct teql_sched_data
*q
= qdisc_priv(sch
);
101 __skb_queue_head(&q
->q
, skb
);
102 sch
->qstats
.requeues
++;
106 static struct sk_buff
*
107 teql_dequeue(struct Qdisc
* sch
)
109 struct teql_sched_data
*dat
= qdisc_priv(sch
);
112 skb
= __skb_dequeue(&dat
->q
);
114 struct net_device
*m
= dat
->m
->dev
->qdisc
->dev
;
116 dat
->m
->slaves
= sch
;
120 sch
->q
.qlen
= dat
->q
.qlen
+ dat
->m
->dev
->qdisc
->q
.qlen
;
124 static __inline__
void
125 teql_neigh_release(struct neighbour
*n
)
132 teql_reset(struct Qdisc
* sch
)
134 struct teql_sched_data
*dat
= qdisc_priv(sch
);
136 skb_queue_purge(&dat
->q
);
138 teql_neigh_release(xchg(&dat
->ncache
, NULL
));
142 teql_destroy(struct Qdisc
* sch
)
144 struct Qdisc
*q
, *prev
;
145 struct teql_sched_data
*dat
= qdisc_priv(sch
);
146 struct teql_master
*master
= dat
->m
;
148 if ((prev
= master
->slaves
) != NULL
) {
150 q
= NEXT_SLAVE(prev
);
152 NEXT_SLAVE(prev
) = NEXT_SLAVE(q
);
153 if (q
== master
->slaves
) {
154 master
->slaves
= NEXT_SLAVE(q
);
155 if (q
== master
->slaves
) {
156 master
->slaves
= NULL
;
157 spin_lock_bh(&master
->dev
->queue_lock
);
158 qdisc_reset(master
->dev
->qdisc
);
159 spin_unlock_bh(&master
->dev
->queue_lock
);
162 skb_queue_purge(&dat
->q
);
163 teql_neigh_release(xchg(&dat
->ncache
, NULL
));
167 } while ((prev
= q
) != master
->slaves
);
171 static int teql_qdisc_init(struct Qdisc
*sch
, struct nlattr
*opt
)
173 struct net_device
*dev
= sch
->dev
;
174 struct teql_master
*m
= (struct teql_master
*)sch
->ops
;
175 struct teql_sched_data
*q
= qdisc_priv(sch
);
177 if (dev
->hard_header_len
> m
->dev
->hard_header_len
)
185 skb_queue_head_init(&q
->q
);
188 if (m
->dev
->flags
& IFF_UP
) {
189 if ((m
->dev
->flags
&IFF_POINTOPOINT
&& !(dev
->flags
&IFF_POINTOPOINT
))
190 || (m
->dev
->flags
&IFF_BROADCAST
&& !(dev
->flags
&IFF_BROADCAST
))
191 || (m
->dev
->flags
&IFF_MULTICAST
&& !(dev
->flags
&IFF_MULTICAST
))
192 || dev
->mtu
< m
->dev
->mtu
)
195 if (!(dev
->flags
&IFF_POINTOPOINT
))
196 m
->dev
->flags
&= ~IFF_POINTOPOINT
;
197 if (!(dev
->flags
&IFF_BROADCAST
))
198 m
->dev
->flags
&= ~IFF_BROADCAST
;
199 if (!(dev
->flags
&IFF_MULTICAST
))
200 m
->dev
->flags
&= ~IFF_MULTICAST
;
201 if (dev
->mtu
< m
->dev
->mtu
)
202 m
->dev
->mtu
= dev
->mtu
;
204 q
->next
= NEXT_SLAVE(m
->slaves
);
205 NEXT_SLAVE(m
->slaves
) = sch
;
209 m
->dev
->mtu
= dev
->mtu
;
210 m
->dev
->flags
= (m
->dev
->flags
&~FMASK
)|(dev
->flags
&FMASK
);
217 __teql_resolve(struct sk_buff
*skb
, struct sk_buff
*skb_res
, struct net_device
*dev
)
219 struct teql_sched_data
*q
= qdisc_priv(dev
->qdisc
);
220 struct neighbour
*mn
= skb
->dst
->neighbour
;
221 struct neighbour
*n
= q
->ncache
;
225 if (n
&& n
->tbl
== mn
->tbl
&&
226 memcmp(n
->primary_key
, mn
->primary_key
, mn
->tbl
->key_len
) == 0) {
227 atomic_inc(&n
->refcnt
);
229 n
= __neigh_lookup_errno(mn
->tbl
, mn
->primary_key
, dev
);
233 if (neigh_event_send(n
, skb_res
) == 0) {
237 err
= dev_hard_header(skb
, dev
, ntohs(skb
->protocol
),
238 n
->ha
, NULL
, skb
->len
);
239 read_unlock(&n
->lock
);
245 teql_neigh_release(xchg(&q
->ncache
, n
));
249 return (skb_res
== NULL
) ? -EAGAIN
: 1;
252 static inline int teql_resolve(struct sk_buff
*skb
,
253 struct sk_buff
*skb_res
, struct net_device
*dev
)
255 if (dev
->qdisc
== &noop_qdisc
)
258 if (dev
->header_ops
== NULL
||
260 skb
->dst
->neighbour
== NULL
)
262 return __teql_resolve(skb
, skb_res
, dev
);
265 static int teql_master_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
267 struct teql_master
*master
= netdev_priv(dev
);
268 struct Qdisc
*start
, *q
;
272 int subq
= skb_get_queue_mapping(skb
);
273 struct sk_buff
*skb_res
= NULL
;
275 start
= master
->slaves
;
281 if ((q
= start
) == NULL
)
285 struct net_device
*slave
= q
->dev
;
287 if (slave
->qdisc_sleeping
!= q
)
289 if (netif_queue_stopped(slave
) ||
290 __netif_subqueue_stopped(slave
, subq
) ||
291 !netif_running(slave
)) {
296 switch (teql_resolve(skb
, skb_res
, slave
)) {
298 if (netif_tx_trylock(slave
)) {
299 if (!netif_queue_stopped(slave
) &&
300 !__netif_subqueue_stopped(slave
, subq
) &&
301 slave
->hard_start_xmit(skb
, slave
) == 0) {
302 netif_tx_unlock(slave
);
303 master
->slaves
= NEXT_SLAVE(q
);
304 netif_wake_queue(dev
);
305 master
->stats
.tx_packets
++;
306 master
->stats
.tx_bytes
+= len
;
309 netif_tx_unlock(slave
);
311 if (netif_queue_stopped(dev
))
315 master
->slaves
= NEXT_SLAVE(q
);
321 __skb_pull(skb
, skb_network_offset(skb
));
322 } while ((q
= NEXT_SLAVE(q
)) != start
);
324 if (nores
&& skb_res
== NULL
) {
330 netif_stop_queue(dev
);
333 master
->stats
.tx_errors
++;
336 master
->stats
.tx_dropped
++;
341 static int teql_master_open(struct net_device
*dev
)
344 struct teql_master
*m
= netdev_priv(dev
);
346 unsigned flags
= IFF_NOARP
|IFF_MULTICAST
;
348 if (m
->slaves
== NULL
)
355 struct net_device
*slave
= q
->dev
;
360 if (slave
->mtu
< mtu
)
362 if (slave
->hard_header_len
> LL_MAX_HEADER
)
365 /* If all the slaves are BROADCAST, master is BROADCAST
366 If all the slaves are PtP, master is PtP
367 Otherwise, master is NBMA.
369 if (!(slave
->flags
&IFF_POINTOPOINT
))
370 flags
&= ~IFF_POINTOPOINT
;
371 if (!(slave
->flags
&IFF_BROADCAST
))
372 flags
&= ~IFF_BROADCAST
;
373 if (!(slave
->flags
&IFF_MULTICAST
))
374 flags
&= ~IFF_MULTICAST
;
375 } while ((q
= NEXT_SLAVE(q
)) != m
->slaves
);
378 m
->dev
->flags
= (m
->dev
->flags
&~FMASK
) | flags
;
379 netif_start_queue(m
->dev
);
383 static int teql_master_close(struct net_device
*dev
)
385 netif_stop_queue(dev
);
389 static struct net_device_stats
*teql_master_stats(struct net_device
*dev
)
391 struct teql_master
*m
= netdev_priv(dev
);
395 static int teql_master_mtu(struct net_device
*dev
, int new_mtu
)
397 struct teql_master
*m
= netdev_priv(dev
);
406 if (new_mtu
> q
->dev
->mtu
)
408 } while ((q
=NEXT_SLAVE(q
)) != m
->slaves
);
415 static __init
void teql_master_setup(struct net_device
*dev
)
417 struct teql_master
*master
= netdev_priv(dev
);
418 struct Qdisc_ops
*ops
= &master
->qops
;
421 ops
->priv_size
= sizeof(struct teql_sched_data
);
423 ops
->enqueue
= teql_enqueue
;
424 ops
->dequeue
= teql_dequeue
;
425 ops
->requeue
= teql_requeue
;
426 ops
->init
= teql_qdisc_init
;
427 ops
->reset
= teql_reset
;
428 ops
->destroy
= teql_destroy
;
429 ops
->owner
= THIS_MODULE
;
431 dev
->open
= teql_master_open
;
432 dev
->hard_start_xmit
= teql_master_xmit
;
433 dev
->stop
= teql_master_close
;
434 dev
->get_stats
= teql_master_stats
;
435 dev
->change_mtu
= teql_master_mtu
;
436 dev
->type
= ARPHRD_VOID
;
438 dev
->tx_queue_len
= 100;
439 dev
->flags
= IFF_NOARP
;
440 dev
->hard_header_len
= LL_MAX_HEADER
;
443 static LIST_HEAD(master_dev_list
);
444 static int max_equalizers
= 1;
445 module_param(max_equalizers
, int, 0);
446 MODULE_PARM_DESC(max_equalizers
, "Max number of link equalizers");
448 static int __init
teql_init(void)
453 for (i
= 0; i
< max_equalizers
; i
++) {
454 struct net_device
*dev
;
455 struct teql_master
*master
;
457 dev
= alloc_netdev(sizeof(struct teql_master
),
458 "teql%d", teql_master_setup
);
464 if ((err
= register_netdev(dev
))) {
469 master
= netdev_priv(dev
);
471 strlcpy(master
->qops
.id
, dev
->name
, IFNAMSIZ
);
472 err
= register_qdisc(&master
->qops
);
475 unregister_netdev(dev
);
480 list_add_tail(&master
->master_list
, &master_dev_list
);
485 static void __exit
teql_exit(void)
487 struct teql_master
*master
, *nxt
;
489 list_for_each_entry_safe(master
, nxt
, &master_dev_list
, master_list
) {
491 list_del(&master
->master_list
);
493 unregister_qdisc(&master
->qops
);
494 unregister_netdev(master
->dev
);
495 free_netdev(master
->dev
);
499 module_init(teql_init
);
500 module_exit(teql_exit
);
502 MODULE_LICENSE("GPL");