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/slab.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/if_arp.h>
18 #include <linux/netdevice.h>
19 #include <linux/init.h>
20 #include <linux/skbuff.h>
21 #include <linux/moduleparam.h>
23 #include <net/neighbour.h>
24 #include <net/pkt_sched.h>
30 After loading this module you will find a new device teqlN
31 and new qdisc with the same name. To join a slave to the equalizer
32 you should just set this qdisc on a device f.e.
34 # tc qdisc add dev eth0 root teql0
35 # tc qdisc add dev eth1 root teql0
37 That's all. Full PnP 8)
42 1. Slave devices MUST be active devices, i.e., they must raise the tbusy
43 signal and generate EOI events. If you want to equalize virtual devices
44 like tunnels, use a normal eql device.
45 2. This device puts no limitations on physical slave characteristics
46 f.e. it will equalize 9600baud line and 100Mb ethernet perfectly :-)
47 Certainly, large difference in link speeds will make the resulting
48 eqalized link unusable, because of huge packet reordering.
49 I estimate an upper useful difference as ~10 times.
50 3. If the slave requires address resolution, only protocols using
51 neighbour cache (IPv4/IPv6) will work over the equalized link.
52 Other protocols are still allowed to use the slave device directly,
53 which will not break load balancing, though native slave
54 traffic will have the highest priority. */
57 struct Qdisc_ops qops
;
58 struct net_device
*dev
;
60 struct list_head master_list
;
61 unsigned long tx_bytes
;
62 unsigned long tx_packets
;
63 unsigned long tx_errors
;
64 unsigned long tx_dropped
;
67 struct teql_sched_data
{
69 struct teql_master
*m
;
70 struct sk_buff_head q
;
73 #define NEXT_SLAVE(q) (((struct teql_sched_data *)qdisc_priv(q))->next)
75 #define FMASK (IFF_BROADCAST | IFF_POINTOPOINT)
77 /* "teql*" qdisc routines */
80 teql_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
, struct sk_buff
**to_free
)
82 struct net_device
*dev
= qdisc_dev(sch
);
83 struct teql_sched_data
*q
= qdisc_priv(sch
);
85 if (q
->q
.qlen
< dev
->tx_queue_len
) {
86 __skb_queue_tail(&q
->q
, skb
);
87 return NET_XMIT_SUCCESS
;
90 return qdisc_drop(skb
, sch
, to_free
);
93 static struct sk_buff
*
94 teql_dequeue(struct Qdisc
*sch
)
96 struct teql_sched_data
*dat
= qdisc_priv(sch
);
97 struct netdev_queue
*dat_queue
;
101 skb
= __skb_dequeue(&dat
->q
);
102 dat_queue
= netdev_get_tx_queue(dat
->m
->dev
, 0);
103 q
= rcu_dereference_bh(dat_queue
->qdisc
);
106 struct net_device
*m
= qdisc_dev(q
);
108 dat
->m
->slaves
= sch
;
112 qdisc_bstats_update(sch
, skb
);
114 sch
->q
.qlen
= dat
->q
.qlen
+ q
->q
.qlen
;
118 static struct sk_buff
*
119 teql_peek(struct Qdisc
*sch
)
121 /* teql is meant to be used as root qdisc */
126 teql_reset(struct Qdisc
*sch
)
128 struct teql_sched_data
*dat
= qdisc_priv(sch
);
130 skb_queue_purge(&dat
->q
);
135 teql_destroy(struct Qdisc
*sch
)
137 struct Qdisc
*q
, *prev
;
138 struct teql_sched_data
*dat
= qdisc_priv(sch
);
139 struct teql_master
*master
= dat
->m
;
141 prev
= master
->slaves
;
144 q
= NEXT_SLAVE(prev
);
146 NEXT_SLAVE(prev
) = NEXT_SLAVE(q
);
147 if (q
== master
->slaves
) {
148 master
->slaves
= NEXT_SLAVE(q
);
149 if (q
== master
->slaves
) {
150 struct netdev_queue
*txq
;
151 spinlock_t
*root_lock
;
153 txq
= netdev_get_tx_queue(master
->dev
, 0);
154 master
->slaves
= NULL
;
156 root_lock
= qdisc_root_sleeping_lock(rtnl_dereference(txq
->qdisc
));
157 spin_lock_bh(root_lock
);
158 qdisc_reset(rtnl_dereference(txq
->qdisc
));
159 spin_unlock_bh(root_lock
);
162 skb_queue_purge(&dat
->q
);
166 } while ((prev
= q
) != master
->slaves
);
170 static int teql_qdisc_init(struct Qdisc
*sch
, struct nlattr
*opt
)
172 struct net_device
*dev
= qdisc_dev(sch
);
173 struct teql_master
*m
= (struct teql_master
*)sch
->ops
;
174 struct teql_sched_data
*q
= qdisc_priv(sch
);
176 if (dev
->hard_header_len
> m
->dev
->hard_header_len
)
184 skb_queue_head_init(&q
->q
);
187 if (m
->dev
->flags
& IFF_UP
) {
188 if ((m
->dev
->flags
& IFF_POINTOPOINT
&&
189 !(dev
->flags
& IFF_POINTOPOINT
)) ||
190 (m
->dev
->flags
& IFF_BROADCAST
&&
191 !(dev
->flags
& IFF_BROADCAST
)) ||
192 (m
->dev
->flags
& IFF_MULTICAST
&&
193 !(dev
->flags
& IFF_MULTICAST
)) ||
194 dev
->mtu
< m
->dev
->mtu
)
197 if (!(dev
->flags
&IFF_POINTOPOINT
))
198 m
->dev
->flags
&= ~IFF_POINTOPOINT
;
199 if (!(dev
->flags
&IFF_BROADCAST
))
200 m
->dev
->flags
&= ~IFF_BROADCAST
;
201 if (!(dev
->flags
&IFF_MULTICAST
))
202 m
->dev
->flags
&= ~IFF_MULTICAST
;
203 if (dev
->mtu
< m
->dev
->mtu
)
204 m
->dev
->mtu
= dev
->mtu
;
206 q
->next
= NEXT_SLAVE(m
->slaves
);
207 NEXT_SLAVE(m
->slaves
) = sch
;
211 m
->dev
->mtu
= dev
->mtu
;
212 m
->dev
->flags
= (m
->dev
->flags
&~FMASK
)|(dev
->flags
&FMASK
);
219 __teql_resolve(struct sk_buff
*skb
, struct sk_buff
*skb_res
,
220 struct net_device
*dev
, struct netdev_queue
*txq
,
221 struct dst_entry
*dst
)
226 n
= dst_neigh_lookup_skb(dst
, skb
);
230 if (dst
->dev
!= dev
) {
231 struct neighbour
*mn
;
233 mn
= __neigh_lookup_errno(n
->tbl
, n
->primary_key
, dev
);
240 if (neigh_event_send(n
, skb_res
) == 0) {
242 char haddr
[MAX_ADDR_LEN
];
244 neigh_ha_snapshot(haddr
, n
, dev
);
245 err
= dev_hard_header(skb
, dev
, ntohs(tc_skb_protocol(skb
)),
246 haddr
, NULL
, skb
->len
);
251 err
= (skb_res
== NULL
) ? -EAGAIN
: 1;
257 static inline int teql_resolve(struct sk_buff
*skb
,
258 struct sk_buff
*skb_res
,
259 struct net_device
*dev
,
260 struct netdev_queue
*txq
)
262 struct dst_entry
*dst
= skb_dst(skb
);
265 if (rcu_access_pointer(txq
->qdisc
) == &noop_qdisc
)
268 if (!dev
->header_ops
|| !dst
)
272 res
= __teql_resolve(skb
, skb_res
, dev
, txq
, dst
);
278 static netdev_tx_t
teql_master_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
280 struct teql_master
*master
= netdev_priv(dev
);
281 struct Qdisc
*start
, *q
;
284 int subq
= skb_get_queue_mapping(skb
);
285 struct sk_buff
*skb_res
= NULL
;
287 start
= master
->slaves
;
298 struct net_device
*slave
= qdisc_dev(q
);
299 struct netdev_queue
*slave_txq
= netdev_get_tx_queue(slave
, 0);
301 if (slave_txq
->qdisc_sleeping
!= q
)
303 if (netif_xmit_stopped(netdev_get_tx_queue(slave
, subq
)) ||
304 !netif_running(slave
)) {
309 switch (teql_resolve(skb
, skb_res
, slave
, slave_txq
)) {
311 if (__netif_tx_trylock(slave_txq
)) {
312 unsigned int length
= qdisc_pkt_len(skb
);
314 if (!netif_xmit_frozen_or_stopped(slave_txq
) &&
315 netdev_start_xmit(skb
, slave
, slave_txq
, false) ==
317 __netif_tx_unlock(slave_txq
);
318 master
->slaves
= NEXT_SLAVE(q
);
319 netif_wake_queue(dev
);
320 master
->tx_packets
++;
321 master
->tx_bytes
+= length
;
324 __netif_tx_unlock(slave_txq
);
326 if (netif_xmit_stopped(netdev_get_tx_queue(dev
, 0)))
330 master
->slaves
= NEXT_SLAVE(q
);
336 __skb_pull(skb
, skb_network_offset(skb
));
337 } while ((q
= NEXT_SLAVE(q
)) != start
);
339 if (nores
&& skb_res
== NULL
) {
345 netif_stop_queue(dev
);
346 return NETDEV_TX_BUSY
;
351 master
->tx_dropped
++;
356 static int teql_master_open(struct net_device
*dev
)
359 struct teql_master
*m
= netdev_priv(dev
);
361 unsigned int flags
= IFF_NOARP
| IFF_MULTICAST
;
363 if (m
->slaves
== NULL
)
370 struct net_device
*slave
= qdisc_dev(q
);
375 if (slave
->mtu
< mtu
)
377 if (slave
->hard_header_len
> LL_MAX_HEADER
)
380 /* If all the slaves are BROADCAST, master is BROADCAST
381 If all the slaves are PtP, master is PtP
382 Otherwise, master is NBMA.
384 if (!(slave
->flags
&IFF_POINTOPOINT
))
385 flags
&= ~IFF_POINTOPOINT
;
386 if (!(slave
->flags
&IFF_BROADCAST
))
387 flags
&= ~IFF_BROADCAST
;
388 if (!(slave
->flags
&IFF_MULTICAST
))
389 flags
&= ~IFF_MULTICAST
;
390 } while ((q
= NEXT_SLAVE(q
)) != m
->slaves
);
393 m
->dev
->flags
= (m
->dev
->flags
&~FMASK
) | flags
;
394 netif_start_queue(m
->dev
);
398 static int teql_master_close(struct net_device
*dev
)
400 netif_stop_queue(dev
);
404 static struct rtnl_link_stats64
*teql_master_stats64(struct net_device
*dev
,
405 struct rtnl_link_stats64
*stats
)
407 struct teql_master
*m
= netdev_priv(dev
);
409 stats
->tx_packets
= m
->tx_packets
;
410 stats
->tx_bytes
= m
->tx_bytes
;
411 stats
->tx_errors
= m
->tx_errors
;
412 stats
->tx_dropped
= m
->tx_dropped
;
416 static int teql_master_mtu(struct net_device
*dev
, int new_mtu
)
418 struct teql_master
*m
= netdev_priv(dev
);
427 if (new_mtu
> qdisc_dev(q
)->mtu
)
429 } while ((q
= NEXT_SLAVE(q
)) != m
->slaves
);
436 static const struct net_device_ops teql_netdev_ops
= {
437 .ndo_open
= teql_master_open
,
438 .ndo_stop
= teql_master_close
,
439 .ndo_start_xmit
= teql_master_xmit
,
440 .ndo_get_stats64
= teql_master_stats64
,
441 .ndo_change_mtu
= teql_master_mtu
,
444 static __init
void teql_master_setup(struct net_device
*dev
)
446 struct teql_master
*master
= netdev_priv(dev
);
447 struct Qdisc_ops
*ops
= &master
->qops
;
450 ops
->priv_size
= sizeof(struct teql_sched_data
);
452 ops
->enqueue
= teql_enqueue
;
453 ops
->dequeue
= teql_dequeue
;
454 ops
->peek
= teql_peek
;
455 ops
->init
= teql_qdisc_init
;
456 ops
->reset
= teql_reset
;
457 ops
->destroy
= teql_destroy
;
458 ops
->owner
= THIS_MODULE
;
460 dev
->netdev_ops
= &teql_netdev_ops
;
461 dev
->type
= ARPHRD_VOID
;
463 dev
->tx_queue_len
= 100;
464 dev
->flags
= IFF_NOARP
;
465 dev
->hard_header_len
= LL_MAX_HEADER
;
469 static LIST_HEAD(master_dev_list
);
470 static int max_equalizers
= 1;
471 module_param(max_equalizers
, int, 0);
472 MODULE_PARM_DESC(max_equalizers
, "Max number of link equalizers");
474 static int __init
teql_init(void)
479 for (i
= 0; i
< max_equalizers
; i
++) {
480 struct net_device
*dev
;
481 struct teql_master
*master
;
483 dev
= alloc_netdev(sizeof(struct teql_master
), "teql%d",
484 NET_NAME_UNKNOWN
, teql_master_setup
);
490 if ((err
= register_netdev(dev
))) {
495 master
= netdev_priv(dev
);
497 strlcpy(master
->qops
.id
, dev
->name
, IFNAMSIZ
);
498 err
= register_qdisc(&master
->qops
);
501 unregister_netdev(dev
);
506 list_add_tail(&master
->master_list
, &master_dev_list
);
511 static void __exit
teql_exit(void)
513 struct teql_master
*master
, *nxt
;
515 list_for_each_entry_safe(master
, nxt
, &master_dev_list
, master_list
) {
517 list_del(&master
->master_list
);
519 unregister_qdisc(&master
->qops
);
520 unregister_netdev(master
->dev
);
521 free_netdev(master
->dev
);
525 module_init(teql_init
);
526 module_exit(teql_exit
);
528 MODULE_LICENSE("GPL");