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
= qdisc_dev(sch
);
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
+= qdisc_pkt_len(skb
);
87 sch
->bstats
.packets
++;
96 static struct sk_buff
*
97 teql_dequeue(struct Qdisc
* sch
)
99 struct teql_sched_data
*dat
= qdisc_priv(sch
);
100 struct netdev_queue
*dat_queue
;
103 skb
= __skb_dequeue(&dat
->q
);
104 dat_queue
= netdev_get_tx_queue(dat
->m
->dev
, 0);
106 struct net_device
*m
= qdisc_dev(dat_queue
->qdisc
);
108 dat
->m
->slaves
= sch
;
112 sch
->q
.qlen
= dat
->q
.qlen
+ dat_queue
->qdisc
->q
.qlen
;
116 static struct sk_buff
*
117 teql_peek(struct Qdisc
* sch
)
119 /* teql is meant to be used as root qdisc */
123 static __inline__
void
124 teql_neigh_release(struct neighbour
*n
)
131 teql_reset(struct Qdisc
* sch
)
133 struct teql_sched_data
*dat
= qdisc_priv(sch
);
135 skb_queue_purge(&dat
->q
);
137 teql_neigh_release(xchg(&dat
->ncache
, NULL
));
141 teql_destroy(struct Qdisc
* sch
)
143 struct Qdisc
*q
, *prev
;
144 struct teql_sched_data
*dat
= qdisc_priv(sch
);
145 struct teql_master
*master
= dat
->m
;
147 if ((prev
= master
->slaves
) != NULL
) {
149 q
= NEXT_SLAVE(prev
);
151 NEXT_SLAVE(prev
) = NEXT_SLAVE(q
);
152 if (q
== master
->slaves
) {
153 master
->slaves
= NEXT_SLAVE(q
);
154 if (q
== master
->slaves
) {
155 struct netdev_queue
*txq
;
156 spinlock_t
*root_lock
;
158 txq
= netdev_get_tx_queue(master
->dev
, 0);
159 master
->slaves
= NULL
;
161 root_lock
= qdisc_root_sleeping_lock(txq
->qdisc
);
162 spin_lock_bh(root_lock
);
163 qdisc_reset(txq
->qdisc
);
164 spin_unlock_bh(root_lock
);
167 skb_queue_purge(&dat
->q
);
168 teql_neigh_release(xchg(&dat
->ncache
, NULL
));
172 } while ((prev
= q
) != master
->slaves
);
176 static int teql_qdisc_init(struct Qdisc
*sch
, struct nlattr
*opt
)
178 struct net_device
*dev
= qdisc_dev(sch
);
179 struct teql_master
*m
= (struct teql_master
*)sch
->ops
;
180 struct teql_sched_data
*q
= qdisc_priv(sch
);
182 if (dev
->hard_header_len
> m
->dev
->hard_header_len
)
190 skb_queue_head_init(&q
->q
);
193 if (m
->dev
->flags
& IFF_UP
) {
194 if ((m
->dev
->flags
&IFF_POINTOPOINT
&& !(dev
->flags
&IFF_POINTOPOINT
))
195 || (m
->dev
->flags
&IFF_BROADCAST
&& !(dev
->flags
&IFF_BROADCAST
))
196 || (m
->dev
->flags
&IFF_MULTICAST
&& !(dev
->flags
&IFF_MULTICAST
))
197 || dev
->mtu
< m
->dev
->mtu
)
200 if (!(dev
->flags
&IFF_POINTOPOINT
))
201 m
->dev
->flags
&= ~IFF_POINTOPOINT
;
202 if (!(dev
->flags
&IFF_BROADCAST
))
203 m
->dev
->flags
&= ~IFF_BROADCAST
;
204 if (!(dev
->flags
&IFF_MULTICAST
))
205 m
->dev
->flags
&= ~IFF_MULTICAST
;
206 if (dev
->mtu
< m
->dev
->mtu
)
207 m
->dev
->mtu
= dev
->mtu
;
209 q
->next
= NEXT_SLAVE(m
->slaves
);
210 NEXT_SLAVE(m
->slaves
) = sch
;
214 m
->dev
->mtu
= dev
->mtu
;
215 m
->dev
->flags
= (m
->dev
->flags
&~FMASK
)|(dev
->flags
&FMASK
);
222 __teql_resolve(struct sk_buff
*skb
, struct sk_buff
*skb_res
, struct net_device
*dev
)
224 struct netdev_queue
*dev_queue
= netdev_get_tx_queue(dev
, 0);
225 struct teql_sched_data
*q
= qdisc_priv(dev_queue
->qdisc
);
226 struct neighbour
*mn
= skb
->dst
->neighbour
;
227 struct neighbour
*n
= q
->ncache
;
231 if (n
&& n
->tbl
== mn
->tbl
&&
232 memcmp(n
->primary_key
, mn
->primary_key
, mn
->tbl
->key_len
) == 0) {
233 atomic_inc(&n
->refcnt
);
235 n
= __neigh_lookup_errno(mn
->tbl
, mn
->primary_key
, dev
);
239 if (neigh_event_send(n
, skb_res
) == 0) {
243 err
= dev_hard_header(skb
, dev
, ntohs(skb
->protocol
),
244 n
->ha
, NULL
, skb
->len
);
245 read_unlock(&n
->lock
);
251 teql_neigh_release(xchg(&q
->ncache
, n
));
255 return (skb_res
== NULL
) ? -EAGAIN
: 1;
258 static inline int teql_resolve(struct sk_buff
*skb
,
259 struct sk_buff
*skb_res
, struct net_device
*dev
)
261 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, 0);
262 if (txq
->qdisc
== &noop_qdisc
)
265 if (dev
->header_ops
== NULL
||
267 skb
->dst
->neighbour
== NULL
)
269 return __teql_resolve(skb
, skb_res
, dev
);
272 static int teql_master_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
274 struct teql_master
*master
= netdev_priv(dev
);
275 struct Qdisc
*start
, *q
;
278 int subq
= skb_get_queue_mapping(skb
);
279 struct sk_buff
*skb_res
= NULL
;
281 start
= master
->slaves
;
287 if ((q
= start
) == NULL
)
291 struct net_device
*slave
= qdisc_dev(q
);
292 struct netdev_queue
*slave_txq
;
294 slave_txq
= netdev_get_tx_queue(slave
, 0);
295 if (slave_txq
->qdisc_sleeping
!= q
)
297 if (__netif_subqueue_stopped(slave
, subq
) ||
298 !netif_running(slave
)) {
303 switch (teql_resolve(skb
, skb_res
, slave
)) {
305 if (__netif_tx_trylock(slave_txq
)) {
306 if (!netif_tx_queue_stopped(slave_txq
) &&
307 !netif_tx_queue_frozen(slave_txq
) &&
308 slave
->hard_start_xmit(skb
, slave
) == 0) {
309 __netif_tx_unlock(slave_txq
);
310 master
->slaves
= NEXT_SLAVE(q
);
311 netif_wake_queue(dev
);
312 master
->stats
.tx_packets
++;
313 master
->stats
.tx_bytes
+=
317 __netif_tx_unlock(slave_txq
);
319 if (netif_queue_stopped(dev
))
323 master
->slaves
= NEXT_SLAVE(q
);
329 __skb_pull(skb
, skb_network_offset(skb
));
330 } while ((q
= NEXT_SLAVE(q
)) != start
);
332 if (nores
&& skb_res
== NULL
) {
338 netif_stop_queue(dev
);
341 master
->stats
.tx_errors
++;
344 master
->stats
.tx_dropped
++;
349 static int teql_master_open(struct net_device
*dev
)
352 struct teql_master
*m
= netdev_priv(dev
);
354 unsigned flags
= IFF_NOARP
|IFF_MULTICAST
;
356 if (m
->slaves
== NULL
)
363 struct net_device
*slave
= qdisc_dev(q
);
368 if (slave
->mtu
< mtu
)
370 if (slave
->hard_header_len
> LL_MAX_HEADER
)
373 /* If all the slaves are BROADCAST, master is BROADCAST
374 If all the slaves are PtP, master is PtP
375 Otherwise, master is NBMA.
377 if (!(slave
->flags
&IFF_POINTOPOINT
))
378 flags
&= ~IFF_POINTOPOINT
;
379 if (!(slave
->flags
&IFF_BROADCAST
))
380 flags
&= ~IFF_BROADCAST
;
381 if (!(slave
->flags
&IFF_MULTICAST
))
382 flags
&= ~IFF_MULTICAST
;
383 } while ((q
= NEXT_SLAVE(q
)) != m
->slaves
);
386 m
->dev
->flags
= (m
->dev
->flags
&~FMASK
) | flags
;
387 netif_start_queue(m
->dev
);
391 static int teql_master_close(struct net_device
*dev
)
393 netif_stop_queue(dev
);
397 static struct net_device_stats
*teql_master_stats(struct net_device
*dev
)
399 struct teql_master
*m
= netdev_priv(dev
);
403 static int teql_master_mtu(struct net_device
*dev
, int new_mtu
)
405 struct teql_master
*m
= netdev_priv(dev
);
414 if (new_mtu
> qdisc_dev(q
)->mtu
)
416 } while ((q
=NEXT_SLAVE(q
)) != m
->slaves
);
423 static __init
void teql_master_setup(struct net_device
*dev
)
425 struct teql_master
*master
= netdev_priv(dev
);
426 struct Qdisc_ops
*ops
= &master
->qops
;
429 ops
->priv_size
= sizeof(struct teql_sched_data
);
431 ops
->enqueue
= teql_enqueue
;
432 ops
->dequeue
= teql_dequeue
;
433 ops
->peek
= teql_peek
;
434 ops
->init
= teql_qdisc_init
;
435 ops
->reset
= teql_reset
;
436 ops
->destroy
= teql_destroy
;
437 ops
->owner
= THIS_MODULE
;
439 dev
->open
= teql_master_open
;
440 dev
->hard_start_xmit
= teql_master_xmit
;
441 dev
->stop
= teql_master_close
;
442 dev
->get_stats
= teql_master_stats
;
443 dev
->change_mtu
= teql_master_mtu
;
444 dev
->type
= ARPHRD_VOID
;
446 dev
->tx_queue_len
= 100;
447 dev
->flags
= IFF_NOARP
;
448 dev
->hard_header_len
= LL_MAX_HEADER
;
451 static LIST_HEAD(master_dev_list
);
452 static int max_equalizers
= 1;
453 module_param(max_equalizers
, int, 0);
454 MODULE_PARM_DESC(max_equalizers
, "Max number of link equalizers");
456 static int __init
teql_init(void)
461 for (i
= 0; i
< max_equalizers
; i
++) {
462 struct net_device
*dev
;
463 struct teql_master
*master
;
465 dev
= alloc_netdev(sizeof(struct teql_master
),
466 "teql%d", teql_master_setup
);
472 if ((err
= register_netdev(dev
))) {
477 master
= netdev_priv(dev
);
479 strlcpy(master
->qops
.id
, dev
->name
, IFNAMSIZ
);
480 err
= register_qdisc(&master
->qops
);
483 unregister_netdev(dev
);
488 list_add_tail(&master
->master_list
, &master_dev_list
);
493 static void __exit
teql_exit(void)
495 struct teql_master
*master
, *nxt
;
497 list_for_each_entry_safe(master
, nxt
, &master_dev_list
, master_list
) {
499 list_del(&master
->master_list
);
501 unregister_qdisc(&master
->qops
);
502 unregister_netdev(master
->dev
);
503 free_netdev(master
->dev
);
507 module_init(teql_init
);
508 module_exit(teql_exit
);
510 MODULE_LICENSE("GPL");