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;
144 prev = master->slaves;
147 q = NEXT_SLAVE(prev);
149 NEXT_SLAVE(prev) = NEXT_SLAVE(q);
150 if (q == master->slaves) {
151 master->slaves = NEXT_SLAVE(q);
152 if (q == master->slaves) {
153 struct netdev_queue *txq;
154 spinlock_t *root_lock;
156 txq = netdev_get_tx_queue(master->dev, 0);
157 master->slaves = NULL;
159 root_lock = qdisc_root_sleeping_lock(rtnl_dereference(txq->qdisc));
160 spin_lock_bh(root_lock);
161 qdisc_reset(rtnl_dereference(txq->qdisc));
162 spin_unlock_bh(root_lock);
165 skb_queue_purge(&dat->q);
169 } while ((prev = q) != master->slaves);
173 static int teql_qdisc_init(struct Qdisc *sch, struct nlattr *opt,
174 struct netlink_ext_ack *extack)
176 struct net_device *dev = qdisc_dev(sch);
177 struct teql_master *m = (struct teql_master *)sch->ops;
178 struct teql_sched_data *q = qdisc_priv(sch);
180 if (dev->hard_header_len > m->dev->hard_header_len)
188 skb_queue_head_init(&q->q);
191 if (m->dev->flags & IFF_UP) {
192 if ((m->dev->flags & IFF_POINTOPOINT &&
193 !(dev->flags & IFF_POINTOPOINT)) ||
194 (m->dev->flags & IFF_BROADCAST &&
195 !(dev->flags & IFF_BROADCAST)) ||
196 (m->dev->flags & IFF_MULTICAST &&
197 !(dev->flags & IFF_MULTICAST)) ||
198 dev->mtu < m->dev->mtu)
201 if (!(dev->flags&IFF_POINTOPOINT))
202 m->dev->flags &= ~IFF_POINTOPOINT;
203 if (!(dev->flags&IFF_BROADCAST))
204 m->dev->flags &= ~IFF_BROADCAST;
205 if (!(dev->flags&IFF_MULTICAST))
206 m->dev->flags &= ~IFF_MULTICAST;
207 if (dev->mtu < m->dev->mtu)
208 m->dev->mtu = dev->mtu;
210 q->next = NEXT_SLAVE(m->slaves);
211 NEXT_SLAVE(m->slaves) = sch;
215 m->dev->mtu = dev->mtu;
216 m->dev->flags = (m->dev->flags&~FMASK)|(dev->flags&FMASK);
223 __teql_resolve(struct sk_buff *skb, struct sk_buff *skb_res,
224 struct net_device *dev, struct netdev_queue *txq,
225 struct dst_entry *dst)
230 n = dst_neigh_lookup_skb(dst, skb);
234 if (dst->dev != dev) {
235 struct neighbour *mn;
237 mn = __neigh_lookup_errno(n->tbl, n->primary_key, dev);
244 if (neigh_event_send(n, skb_res) == 0) {
246 char haddr[MAX_ADDR_LEN];
248 neigh_ha_snapshot(haddr, n, dev);
249 err = dev_hard_header(skb, dev, ntohs(skb_protocol(skb, false)),
250 haddr, NULL, skb->len);
255 err = (skb_res == NULL) ? -EAGAIN : 1;
261 static inline int teql_resolve(struct sk_buff *skb,
262 struct sk_buff *skb_res,
263 struct net_device *dev,
264 struct netdev_queue *txq)
266 struct dst_entry *dst = skb_dst(skb);
269 if (rcu_access_pointer(txq->qdisc) == &noop_qdisc)
272 if (!dev->header_ops || !dst)
276 res = __teql_resolve(skb, skb_res, dev, txq, dst);
282 static netdev_tx_t teql_master_xmit(struct sk_buff *skb, struct net_device *dev)
284 struct teql_master *master = netdev_priv(dev);
285 struct Qdisc *start, *q;
288 int subq = skb_get_queue_mapping(skb);
289 struct sk_buff *skb_res = NULL;
291 start = master->slaves;
302 struct net_device *slave = qdisc_dev(q);
303 struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
305 if (slave_txq->qdisc_sleeping != q)
307 if (netif_xmit_stopped(netdev_get_tx_queue(slave, subq)) ||
308 !netif_running(slave)) {
313 switch (teql_resolve(skb, skb_res, slave, slave_txq)) {
315 if (__netif_tx_trylock(slave_txq)) {
316 unsigned int length = qdisc_pkt_len(skb);
318 if (!netif_xmit_frozen_or_stopped(slave_txq) &&
319 netdev_start_xmit(skb, slave, slave_txq, false) ==
321 __netif_tx_unlock(slave_txq);
322 master->slaves = NEXT_SLAVE(q);
323 netif_wake_queue(dev);
324 master->tx_packets++;
325 master->tx_bytes += length;
328 __netif_tx_unlock(slave_txq);
330 if (netif_xmit_stopped(netdev_get_tx_queue(dev, 0)))
334 master->slaves = NEXT_SLAVE(q);
340 __skb_pull(skb, skb_network_offset(skb));
341 } while ((q = NEXT_SLAVE(q)) != start);
343 if (nores && skb_res == NULL) {
349 netif_stop_queue(dev);
350 return NETDEV_TX_BUSY;
355 master->tx_dropped++;
360 static int teql_master_open(struct net_device *dev)
363 struct teql_master *m = netdev_priv(dev);
365 unsigned int flags = IFF_NOARP | IFF_MULTICAST;
367 if (m->slaves == NULL)
374 struct net_device *slave = qdisc_dev(q);
379 if (slave->mtu < mtu)
381 if (slave->hard_header_len > LL_MAX_HEADER)
384 /* If all the slaves are BROADCAST, master is BROADCAST
385 If all the slaves are PtP, master is PtP
386 Otherwise, master is NBMA.
388 if (!(slave->flags&IFF_POINTOPOINT))
389 flags &= ~IFF_POINTOPOINT;
390 if (!(slave->flags&IFF_BROADCAST))
391 flags &= ~IFF_BROADCAST;
392 if (!(slave->flags&IFF_MULTICAST))
393 flags &= ~IFF_MULTICAST;
394 } while ((q = NEXT_SLAVE(q)) != m->slaves);
397 m->dev->flags = (m->dev->flags&~FMASK) | flags;
398 netif_start_queue(m->dev);
402 static int teql_master_close(struct net_device *dev)
404 netif_stop_queue(dev);
408 static void teql_master_stats64(struct net_device *dev,
409 struct rtnl_link_stats64 *stats)
411 struct teql_master *m = netdev_priv(dev);
413 stats->tx_packets = m->tx_packets;
414 stats->tx_bytes = m->tx_bytes;
415 stats->tx_errors = m->tx_errors;
416 stats->tx_dropped = m->tx_dropped;
419 static int teql_master_mtu(struct net_device *dev, int new_mtu)
421 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;
464 dev->max_mtu = 65535;
465 dev->tx_queue_len = 100;
466 dev->flags = IFF_NOARP;
467 dev->hard_header_len = LL_MAX_HEADER;
471 static LIST_HEAD(master_dev_list);
472 static int max_equalizers = 1;
473 module_param(max_equalizers, int, 0);
474 MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
476 static int __init teql_init(void)
481 for (i = 0; i < max_equalizers; i++) {
482 struct net_device *dev;
483 struct teql_master *master;
485 dev = alloc_netdev(sizeof(struct teql_master), "teql%d",
486 NET_NAME_UNKNOWN, teql_master_setup);
492 if ((err = register_netdev(dev))) {
497 master = netdev_priv(dev);
499 strlcpy(master->qops.id, dev->name, IFNAMSIZ);
500 err = register_qdisc(&master->qops);
503 unregister_netdev(dev);
508 list_add_tail(&master->master_list, &master_dev_list);
513 static void __exit teql_exit(void)
515 struct teql_master *master, *nxt;
517 list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
519 list_del(&master->master_list);
521 unregister_qdisc(&master->qops);
522 unregister_netdev(master->dev);
523 free_netdev(master->dev);
527 module_init(teql_init);
528 module_exit(teql_exit);
530 MODULE_LICENSE("GPL");