Merge branch 'iommu/fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/joro...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / sched / sch_teql.c
CommitLineData
1da177e4
LT
1/* net/sched/sch_teql.c "True" (or "trivial") link equalizer.
2 *
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.
7 *
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 */
10
11#include <linux/module.h>
1da177e4
LT
12#include <linux/types.h>
13#include <linux/kernel.h>
5a0e3ad6 14#include <linux/slab.h>
1da177e4 15#include <linux/string.h>
1da177e4 16#include <linux/errno.h>
14c85021 17#include <linux/if_arp.h>
1da177e4 18#include <linux/netdevice.h>
1da177e4 19#include <linux/init.h>
1da177e4
LT
20#include <linux/skbuff.h>
21#include <linux/moduleparam.h>
0ba48053
PM
22#include <net/dst.h>
23#include <net/neighbour.h>
1da177e4
LT
24#include <net/pkt_sched.h>
25
26/*
27 How to setup it.
28 ----------------
29
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.
33
34 # tc qdisc add dev eth0 root teql0
35 # tc qdisc add dev eth1 root teql0
36
37 That's all. Full PnP 8)
38
39 Applicability.
40 --------------
41
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. */
55
cc7ec456 56struct teql_master {
1da177e4
LT
57 struct Qdisc_ops qops;
58 struct net_device *dev;
59 struct Qdisc *slaves;
60 struct list_head master_list;
1ac9ad13
ED
61 unsigned long tx_bytes;
62 unsigned long tx_packets;
63 unsigned long tx_errors;
64 unsigned long tx_dropped;
1da177e4
LT
65};
66
cc7ec456 67struct teql_sched_data {
1da177e4
LT
68 struct Qdisc *next;
69 struct teql_master *m;
70 struct neighbour *ncache;
71 struct sk_buff_head q;
72};
73
cc7ec456 74#define NEXT_SLAVE(q) (((struct teql_sched_data *)qdisc_priv(q))->next)
1da177e4 75
cc7ec456 76#define FMASK (IFF_BROADCAST | IFF_POINTOPOINT)
1da177e4
LT
77
78/* "teql*" qdisc routines */
79
80static int
cc7ec456 81teql_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1da177e4 82{
5ce2d488 83 struct net_device *dev = qdisc_dev(sch);
1da177e4
LT
84 struct teql_sched_data *q = qdisc_priv(sch);
85
4cd8c9e8
KK
86 if (q->q.qlen < dev->tx_queue_len) {
87 __skb_queue_tail(&q->q, skb);
9871e50e 88 return NET_XMIT_SUCCESS;
1da177e4
LT
89 }
90
1da177e4
LT
91 kfree_skb(skb);
92 sch->qstats.drops++;
93 return NET_XMIT_DROP;
94}
95
1da177e4 96static struct sk_buff *
cc7ec456 97teql_dequeue(struct Qdisc *sch)
1da177e4
LT
98{
99 struct teql_sched_data *dat = qdisc_priv(sch);
b0e1e646 100 struct netdev_queue *dat_queue;
1da177e4
LT
101 struct sk_buff *skb;
102
103 skb = __skb_dequeue(&dat->q);
e8a0464c 104 dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
1da177e4 105 if (skb == NULL) {
b0e1e646 106 struct net_device *m = qdisc_dev(dat_queue->qdisc);
1da177e4
LT
107 if (m) {
108 dat->m->slaves = sch;
109 netif_wake_queue(m);
110 }
9190b3b3
ED
111 } else {
112 qdisc_bstats_update(sch, skb);
1da177e4 113 }
b0e1e646 114 sch->q.qlen = dat->q.qlen + dat_queue->qdisc->q.qlen;
1da177e4
LT
115 return skb;
116}
117
8e3af978 118static struct sk_buff *
cc7ec456 119teql_peek(struct Qdisc *sch)
8e3af978
JP
120{
121 /* teql is meant to be used as root qdisc */
122 return NULL;
123}
124
cc7ec456 125static inline void
1da177e4
LT
126teql_neigh_release(struct neighbour *n)
127{
128 if (n)
129 neigh_release(n);
130}
131
132static void
cc7ec456 133teql_reset(struct Qdisc *sch)
1da177e4
LT
134{
135 struct teql_sched_data *dat = qdisc_priv(sch);
136
137 skb_queue_purge(&dat->q);
138 sch->q.qlen = 0;
139 teql_neigh_release(xchg(&dat->ncache, NULL));
140}
141
142static void
cc7ec456 143teql_destroy(struct Qdisc *sch)
1da177e4
LT
144{
145 struct Qdisc *q, *prev;
146 struct teql_sched_data *dat = qdisc_priv(sch);
147 struct teql_master *master = dat->m;
148
cc7ec456
ED
149 prev = master->slaves;
150 if (prev) {
1da177e4
LT
151 do {
152 q = NEXT_SLAVE(prev);
153 if (q == sch) {
154 NEXT_SLAVE(prev) = NEXT_SLAVE(q);
155 if (q == master->slaves) {
156 master->slaves = NEXT_SLAVE(q);
157 if (q == master->slaves) {
e8a0464c 158 struct netdev_queue *txq;
83874000 159 spinlock_t *root_lock;
e8a0464c
DM
160
161 txq = netdev_get_tx_queue(master->dev, 0);
1da177e4 162 master->slaves = NULL;
83874000 163
102396ae 164 root_lock = qdisc_root_sleeping_lock(txq->qdisc);
83874000 165 spin_lock_bh(root_lock);
e8a0464c 166 qdisc_reset(txq->qdisc);
83874000 167 spin_unlock_bh(root_lock);
1da177e4
LT
168 }
169 }
170 skb_queue_purge(&dat->q);
171 teql_neigh_release(xchg(&dat->ncache, NULL));
172 break;
173 }
10297b99 174
1da177e4
LT
175 } while ((prev = q) != master->slaves);
176 }
177}
178
1e90474c 179static int teql_qdisc_init(struct Qdisc *sch, struct nlattr *opt)
1da177e4 180{
5ce2d488 181 struct net_device *dev = qdisc_dev(sch);
cc7ec456 182 struct teql_master *m = (struct teql_master *)sch->ops;
1da177e4
LT
183 struct teql_sched_data *q = qdisc_priv(sch);
184
185 if (dev->hard_header_len > m->dev->hard_header_len)
186 return -EINVAL;
187
188 if (m->dev == dev)
189 return -ELOOP;
190
191 q->m = m;
192
193 skb_queue_head_init(&q->q);
194
195 if (m->slaves) {
196 if (m->dev->flags & IFF_UP) {
f64f9e71
JP
197 if ((m->dev->flags & IFF_POINTOPOINT &&
198 !(dev->flags & IFF_POINTOPOINT)) ||
199 (m->dev->flags & IFF_BROADCAST &&
200 !(dev->flags & IFF_BROADCAST)) ||
201 (m->dev->flags & IFF_MULTICAST &&
202 !(dev->flags & IFF_MULTICAST)) ||
203 dev->mtu < m->dev->mtu)
1da177e4
LT
204 return -EINVAL;
205 } else {
206 if (!(dev->flags&IFF_POINTOPOINT))
207 m->dev->flags &= ~IFF_POINTOPOINT;
208 if (!(dev->flags&IFF_BROADCAST))
209 m->dev->flags &= ~IFF_BROADCAST;
210 if (!(dev->flags&IFF_MULTICAST))
211 m->dev->flags &= ~IFF_MULTICAST;
212 if (dev->mtu < m->dev->mtu)
213 m->dev->mtu = dev->mtu;
214 }
215 q->next = NEXT_SLAVE(m->slaves);
216 NEXT_SLAVE(m->slaves) = sch;
217 } else {
218 q->next = sch;
219 m->slaves = sch;
220 m->dev->mtu = dev->mtu;
221 m->dev->flags = (m->dev->flags&~FMASK)|(dev->flags&FMASK);
222 }
223 return 0;
224}
225
1da177e4
LT
226
227static int
f7e57044
ED
228__teql_resolve(struct sk_buff *skb, struct sk_buff *skb_res,
229 struct net_device *dev, struct netdev_queue *txq,
230 struct neighbour *mn)
1da177e4 231{
f7e57044 232 struct teql_sched_data *q = qdisc_priv(txq->qdisc);
1da177e4
LT
233 struct neighbour *n = q->ncache;
234
235 if (mn->tbl == NULL)
236 return -EINVAL;
237 if (n && n->tbl == mn->tbl &&
238 memcmp(n->primary_key, mn->primary_key, mn->tbl->key_len) == 0) {
239 atomic_inc(&n->refcnt);
240 } else {
241 n = __neigh_lookup_errno(mn->tbl, mn->primary_key, dev);
242 if (IS_ERR(n))
243 return PTR_ERR(n);
244 }
245 if (neigh_event_send(n, skb_res) == 0) {
246 int err;
0ed8ddf4 247 char haddr[MAX_ADDR_LEN];
0c4e8581 248
0ed8ddf4
ED
249 neigh_ha_snapshot(haddr, n, dev);
250 err = dev_hard_header(skb, dev, ntohs(skb->protocol), haddr,
251 NULL, skb->len);
0c4e8581 252
1da177e4
LT
253 if (err < 0) {
254 neigh_release(n);
255 return -EINVAL;
256 }
257 teql_neigh_release(xchg(&q->ncache, n));
258 return 0;
259 }
260 neigh_release(n);
261 return (skb_res == NULL) ? -EAGAIN : 1;
262}
263
3b04ddde 264static inline int teql_resolve(struct sk_buff *skb,
f7e57044
ED
265 struct sk_buff *skb_res,
266 struct net_device *dev,
267 struct netdev_queue *txq)
1da177e4 268{
f7e57044
ED
269 struct dst_entry *dst = skb_dst(skb);
270 struct neighbour *mn;
271 int res;
272
e8a0464c 273 if (txq->qdisc == &noop_qdisc)
4f9f8311
EP
274 return -ENODEV;
275
f7e57044 276 if (!dev->header_ops || !dst)
1da177e4 277 return 0;
f7e57044
ED
278
279 rcu_read_lock();
280 mn = dst_get_neighbour(dst);
281 res = mn ? __teql_resolve(skb, skb_res, dev, txq, mn) : 0;
282 rcu_read_unlock();
283
284 return res;
1da177e4
LT
285}
286
6fef4c0c 287static netdev_tx_t teql_master_xmit(struct sk_buff *skb, struct net_device *dev)
1da177e4 288{
2941a486 289 struct teql_master *master = netdev_priv(dev);
1da177e4
LT
290 struct Qdisc *start, *q;
291 int busy;
292 int nores;
4e3ab47a 293 int subq = skb_get_queue_mapping(skb);
1da177e4
LT
294 struct sk_buff *skb_res = NULL;
295
296 start = master->slaves;
297
298restart:
299 nores = 0;
300 busy = 0;
301
cc7ec456
ED
302 q = start;
303 if (!q)
1da177e4
LT
304 goto drop;
305
306 do {
5ce2d488 307 struct net_device *slave = qdisc_dev(q);
61294e2e
SH
308 struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
309 const struct net_device_ops *slave_ops = slave->netdev_ops;
10297b99 310
e8a0464c 311 if (slave_txq->qdisc_sleeping != q)
1da177e4 312 continue;
fd2ea0a7 313 if (__netif_subqueue_stopped(slave, subq) ||
f25f4e44 314 !netif_running(slave)) {
1da177e4
LT
315 busy = 1;
316 continue;
317 }
318
f7e57044 319 switch (teql_resolve(skb, skb_res, slave, slave_txq)) {
1da177e4 320 case 0:
c3f26a26 321 if (__netif_tx_trylock(slave_txq)) {
c0f84d0d
ED
322 unsigned int length = qdisc_pkt_len(skb);
323
5a0d2268 324 if (!netif_tx_queue_frozen_or_stopped(slave_txq) &&
6fef4c0c 325 slave_ops->ndo_start_xmit(skb, slave) == NETDEV_TX_OK) {
08baf561 326 txq_trans_update(slave_txq);
c3f26a26 327 __netif_tx_unlock(slave_txq);
1da177e4
LT
328 master->slaves = NEXT_SLAVE(q);
329 netif_wake_queue(dev);
1ac9ad13
ED
330 master->tx_packets++;
331 master->tx_bytes += length;
6ed10654 332 return NETDEV_TX_OK;
1da177e4 333 }
c3f26a26 334 __netif_tx_unlock(slave_txq);
1da177e4
LT
335 }
336 if (netif_queue_stopped(dev))
337 busy = 1;
338 break;
339 case 1:
340 master->slaves = NEXT_SLAVE(q);
6ed10654 341 return NETDEV_TX_OK;
1da177e4
LT
342 default:
343 nores = 1;
344 break;
345 }
bbe735e4 346 __skb_pull(skb, skb_network_offset(skb));
1da177e4
LT
347 } while ((q = NEXT_SLAVE(q)) != start);
348
349 if (nores && skb_res == NULL) {
350 skb_res = skb;
351 goto restart;
352 }
353
354 if (busy) {
355 netif_stop_queue(dev);
5b548140 356 return NETDEV_TX_BUSY;
1da177e4 357 }
1ac9ad13 358 master->tx_errors++;
1da177e4
LT
359
360drop:
1ac9ad13 361 master->tx_dropped++;
1da177e4 362 dev_kfree_skb(skb);
6ed10654 363 return NETDEV_TX_OK;
1da177e4
LT
364}
365
366static int teql_master_open(struct net_device *dev)
367{
cc7ec456 368 struct Qdisc *q;
2941a486 369 struct teql_master *m = netdev_priv(dev);
1da177e4 370 int mtu = 0xFFFE;
cc7ec456 371 unsigned int flags = IFF_NOARP | IFF_MULTICAST;
1da177e4
LT
372
373 if (m->slaves == NULL)
374 return -EUNATCH;
375
376 flags = FMASK;
377
378 q = m->slaves;
379 do {
5ce2d488 380 struct net_device *slave = qdisc_dev(q);
1da177e4
LT
381
382 if (slave == NULL)
383 return -EUNATCH;
384
385 if (slave->mtu < mtu)
386 mtu = slave->mtu;
387 if (slave->hard_header_len > LL_MAX_HEADER)
388 return -EINVAL;
389
390 /* If all the slaves are BROADCAST, master is BROADCAST
391 If all the slaves are PtP, master is PtP
392 Otherwise, master is NBMA.
393 */
394 if (!(slave->flags&IFF_POINTOPOINT))
395 flags &= ~IFF_POINTOPOINT;
396 if (!(slave->flags&IFF_BROADCAST))
397 flags &= ~IFF_BROADCAST;
398 if (!(slave->flags&IFF_MULTICAST))
399 flags &= ~IFF_MULTICAST;
400 } while ((q = NEXT_SLAVE(q)) != m->slaves);
401
402 m->dev->mtu = mtu;
403 m->dev->flags = (m->dev->flags&~FMASK) | flags;
404 netif_start_queue(m->dev);
405 return 0;
406}
407
408static int teql_master_close(struct net_device *dev)
409{
410 netif_stop_queue(dev);
411 return 0;
412}
413
1ac9ad13
ED
414static struct rtnl_link_stats64 *teql_master_stats64(struct net_device *dev,
415 struct rtnl_link_stats64 *stats)
416{
417 struct teql_master *m = netdev_priv(dev);
418
419 stats->tx_packets = m->tx_packets;
420 stats->tx_bytes = m->tx_bytes;
421 stats->tx_errors = m->tx_errors;
422 stats->tx_dropped = m->tx_dropped;
423 return stats;
424}
425
1da177e4
LT
426static int teql_master_mtu(struct net_device *dev, int new_mtu)
427{
2941a486 428 struct teql_master *m = netdev_priv(dev);
1da177e4
LT
429 struct Qdisc *q;
430
431 if (new_mtu < 68)
432 return -EINVAL;
433
434 q = m->slaves;
435 if (q) {
436 do {
5ce2d488 437 if (new_mtu > qdisc_dev(q)->mtu)
1da177e4 438 return -EINVAL;
cc7ec456 439 } while ((q = NEXT_SLAVE(q)) != m->slaves);
1da177e4
LT
440 }
441
442 dev->mtu = new_mtu;
443 return 0;
444}
445
61294e2e
SH
446static const struct net_device_ops teql_netdev_ops = {
447 .ndo_open = teql_master_open,
448 .ndo_stop = teql_master_close,
449 .ndo_start_xmit = teql_master_xmit,
1ac9ad13 450 .ndo_get_stats64 = teql_master_stats64,
61294e2e
SH
451 .ndo_change_mtu = teql_master_mtu,
452};
453
1da177e4
LT
454static __init void teql_master_setup(struct net_device *dev)
455{
2941a486 456 struct teql_master *master = netdev_priv(dev);
1da177e4
LT
457 struct Qdisc_ops *ops = &master->qops;
458
459 master->dev = dev;
460 ops->priv_size = sizeof(struct teql_sched_data);
10297b99 461
1da177e4
LT
462 ops->enqueue = teql_enqueue;
463 ops->dequeue = teql_dequeue;
8e3af978 464 ops->peek = teql_peek;
1da177e4
LT
465 ops->init = teql_qdisc_init;
466 ops->reset = teql_reset;
467 ops->destroy = teql_destroy;
468 ops->owner = THIS_MODULE;
469
61294e2e 470 dev->netdev_ops = &teql_netdev_ops;
1da177e4
LT
471 dev->type = ARPHRD_VOID;
472 dev->mtu = 1500;
473 dev->tx_queue_len = 100;
474 dev->flags = IFF_NOARP;
475 dev->hard_header_len = LL_MAX_HEADER;
fa68a782 476 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1da177e4
LT
477}
478
479static LIST_HEAD(master_dev_list);
480static int max_equalizers = 1;
481module_param(max_equalizers, int, 0);
482MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
483
484static int __init teql_init(void)
485{
486 int i;
487 int err = -ENODEV;
488
489 for (i = 0; i < max_equalizers; i++) {
490 struct net_device *dev;
491 struct teql_master *master;
492
493 dev = alloc_netdev(sizeof(struct teql_master),
494 "teql%d", teql_master_setup);
495 if (!dev) {
496 err = -ENOMEM;
497 break;
498 }
499
500 if ((err = register_netdev(dev))) {
501 free_netdev(dev);
502 break;
503 }
504
2941a486 505 master = netdev_priv(dev);
1da177e4
LT
506
507 strlcpy(master->qops.id, dev->name, IFNAMSIZ);
508 err = register_qdisc(&master->qops);
509
510 if (err) {
511 unregister_netdev(dev);
512 free_netdev(dev);
513 break;
514 }
515
516 list_add_tail(&master->master_list, &master_dev_list);
517 }
518 return i ? 0 : err;
519}
520
10297b99 521static void __exit teql_exit(void)
1da177e4
LT
522{
523 struct teql_master *master, *nxt;
524
525 list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
526
527 list_del(&master->master_list);
528
529 unregister_qdisc(&master->qops);
530 unregister_netdev(master->dev);
531 free_netdev(master->dev);
532 }
533}
534
535module_init(teql_init);
536module_exit(teql_exit);
537
538MODULE_LICENSE("GPL");