Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* net/sched/sch_teql.c "True" (or "trivial") link equalizer.
3 *
4 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
5 */
6
7#include <linux/module.h>
8#include <linux/types.h>
9#include <linux/kernel.h>
10#include <linux/slab.h>
11#include <linux/string.h>
12#include <linux/errno.h>
13#include <linux/if_arp.h>
14#include <linux/netdevice.h>
15#include <linux/init.h>
16#include <linux/skbuff.h>
17#include <linux/moduleparam.h>
18#include <net/dst.h>
19#include <net/neighbour.h>
20#include <net/pkt_sched.h>
21
22/*
23 How to setup it.
24 ----------------
25
26 After loading this module you will find a new device teqlN
27 and new qdisc with the same name. To join a slave to the equalizer
28 you should just set this qdisc on a device f.e.
29
30 # tc qdisc add dev eth0 root teql0
31 # tc qdisc add dev eth1 root teql0
32
33 That's all. Full PnP 8)
34
35 Applicability.
36 --------------
37
38 1. Slave devices MUST be active devices, i.e., they must raise the tbusy
39 signal and generate EOI events. If you want to equalize virtual devices
40 like tunnels, use a normal eql device.
41 2. This device puts no limitations on physical slave characteristics
42 f.e. it will equalize 9600baud line and 100Mb ethernet perfectly :-)
43 Certainly, large difference in link speeds will make the resulting
44 eqalized link unusable, because of huge packet reordering.
45 I estimate an upper useful difference as ~10 times.
46 3. If the slave requires address resolution, only protocols using
47 neighbour cache (IPv4/IPv6) will work over the equalized link.
48 Other protocols are still allowed to use the slave device directly,
49 which will not break load balancing, though native slave
50 traffic will have the highest priority. */
51
52struct teql_master {
53 struct Qdisc_ops qops;
54 struct net_device *dev;
55 struct Qdisc *slaves;
56 struct list_head master_list;
57 unsigned long tx_bytes;
58 unsigned long tx_packets;
59 unsigned long tx_errors;
60 unsigned long tx_dropped;
61};
62
63struct teql_sched_data {
64 struct Qdisc *next;
65 struct teql_master *m;
66 struct sk_buff_head q;
67};
68
69#define NEXT_SLAVE(q) (((struct teql_sched_data *)qdisc_priv(q))->next)
70
71#define FMASK (IFF_BROADCAST | IFF_POINTOPOINT)
72
73/* "teql*" qdisc routines */
74
75static int
76teql_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free)
77{
78 struct net_device *dev = qdisc_dev(sch);
79 struct teql_sched_data *q = qdisc_priv(sch);
80
81 if (q->q.qlen < dev->tx_queue_len) {
82 __skb_queue_tail(&q->q, skb);
83 return NET_XMIT_SUCCESS;
84 }
85
86 return qdisc_drop(skb, sch, to_free);
87}
88
89static struct sk_buff *
90teql_dequeue(struct Qdisc *sch)
91{
92 struct teql_sched_data *dat = qdisc_priv(sch);
93 struct netdev_queue *dat_queue;
94 struct sk_buff *skb;
95 struct Qdisc *q;
96
97 skb = __skb_dequeue(&dat->q);
98 dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
99 q = rcu_dereference_bh(dat_queue->qdisc);
100
101 if (skb == NULL) {
102 struct net_device *m = qdisc_dev(q);
103 if (m) {
104 dat->m->slaves = sch;
105 netif_wake_queue(m);
106 }
107 } else {
108 qdisc_bstats_update(sch, skb);
109 }
110 sch->q.qlen = dat->q.qlen + q->q.qlen;
111 return skb;
112}
113
114static struct sk_buff *
115teql_peek(struct Qdisc *sch)
116{
117 /* teql is meant to be used as root qdisc */
118 return NULL;
119}
120
121static void
122teql_reset(struct Qdisc *sch)
123{
124 struct teql_sched_data *dat = qdisc_priv(sch);
125
126 skb_queue_purge(&dat->q);
127 sch->q.qlen = 0;
128}
129
130static void
131teql_destroy(struct Qdisc *sch)
132{
133 struct Qdisc *q, *prev;
134 struct teql_sched_data *dat = qdisc_priv(sch);
135 struct teql_master *master = dat->m;
136
137 prev = master->slaves;
138 if (prev) {
139 do {
140 q = NEXT_SLAVE(prev);
141 if (q == sch) {
142 NEXT_SLAVE(prev) = NEXT_SLAVE(q);
143 if (q == master->slaves) {
144 master->slaves = NEXT_SLAVE(q);
145 if (q == master->slaves) {
146 struct netdev_queue *txq;
147 spinlock_t *root_lock;
148
149 txq = netdev_get_tx_queue(master->dev, 0);
150 master->slaves = NULL;
151
152 root_lock = qdisc_root_sleeping_lock(rtnl_dereference(txq->qdisc));
153 spin_lock_bh(root_lock);
154 qdisc_reset(rtnl_dereference(txq->qdisc));
155 spin_unlock_bh(root_lock);
156 }
157 }
158 skb_queue_purge(&dat->q);
159 break;
160 }
161
162 } while ((prev = q) != master->slaves);
163 }
164}
165
166static int teql_qdisc_init(struct Qdisc *sch, struct nlattr *opt,
167 struct netlink_ext_ack *extack)
168{
169 struct net_device *dev = qdisc_dev(sch);
170 struct teql_master *m = (struct teql_master *)sch->ops;
171 struct teql_sched_data *q = qdisc_priv(sch);
172
173 if (dev->hard_header_len > m->dev->hard_header_len)
174 return -EINVAL;
175
176 if (m->dev == dev)
177 return -ELOOP;
178
179 q->m = m;
180
181 skb_queue_head_init(&q->q);
182
183 if (m->slaves) {
184 if (m->dev->flags & IFF_UP) {
185 if ((m->dev->flags & IFF_POINTOPOINT &&
186 !(dev->flags & IFF_POINTOPOINT)) ||
187 (m->dev->flags & IFF_BROADCAST &&
188 !(dev->flags & IFF_BROADCAST)) ||
189 (m->dev->flags & IFF_MULTICAST &&
190 !(dev->flags & IFF_MULTICAST)) ||
191 dev->mtu < m->dev->mtu)
192 return -EINVAL;
193 } else {
194 if (!(dev->flags&IFF_POINTOPOINT))
195 m->dev->flags &= ~IFF_POINTOPOINT;
196 if (!(dev->flags&IFF_BROADCAST))
197 m->dev->flags &= ~IFF_BROADCAST;
198 if (!(dev->flags&IFF_MULTICAST))
199 m->dev->flags &= ~IFF_MULTICAST;
200 if (dev->mtu < m->dev->mtu)
201 m->dev->mtu = dev->mtu;
202 }
203 q->next = NEXT_SLAVE(m->slaves);
204 NEXT_SLAVE(m->slaves) = sch;
205 } else {
206 q->next = sch;
207 m->slaves = sch;
208 m->dev->mtu = dev->mtu;
209 m->dev->flags = (m->dev->flags&~FMASK)|(dev->flags&FMASK);
210 }
211 return 0;
212}
213
214
215static int
216__teql_resolve(struct sk_buff *skb, struct sk_buff *skb_res,
217 struct net_device *dev, struct netdev_queue *txq,
218 struct dst_entry *dst)
219{
220 struct neighbour *n;
221 int err = 0;
222
223 n = dst_neigh_lookup_skb(dst, skb);
224 if (!n)
225 return -ENOENT;
226
227 if (dst->dev != dev) {
228 struct neighbour *mn;
229
230 mn = __neigh_lookup_errno(n->tbl, n->primary_key, dev);
231 neigh_release(n);
232 if (IS_ERR(mn))
233 return PTR_ERR(mn);
234 n = mn;
235 }
236
237 if (neigh_event_send(n, skb_res) == 0) {
238 int err;
239 char haddr[MAX_ADDR_LEN];
240
241 neigh_ha_snapshot(haddr, n, dev);
242 err = dev_hard_header(skb, dev, ntohs(tc_skb_protocol(skb)),
243 haddr, NULL, skb->len);
244
245 if (err < 0)
246 err = -EINVAL;
247 } else {
248 err = (skb_res == NULL) ? -EAGAIN : 1;
249 }
250 neigh_release(n);
251 return err;
252}
253
254static inline int teql_resolve(struct sk_buff *skb,
255 struct sk_buff *skb_res,
256 struct net_device *dev,
257 struct netdev_queue *txq)
258{
259 struct dst_entry *dst = skb_dst(skb);
260 int res;
261
262 if (rcu_access_pointer(txq->qdisc) == &noop_qdisc)
263 return -ENODEV;
264
265 if (!dev->header_ops || !dst)
266 return 0;
267
268 rcu_read_lock();
269 res = __teql_resolve(skb, skb_res, dev, txq, dst);
270 rcu_read_unlock();
271
272 return res;
273}
274
275static netdev_tx_t teql_master_xmit(struct sk_buff *skb, struct net_device *dev)
276{
277 struct teql_master *master = netdev_priv(dev);
278 struct Qdisc *start, *q;
279 int busy;
280 int nores;
281 int subq = skb_get_queue_mapping(skb);
282 struct sk_buff *skb_res = NULL;
283
284 start = master->slaves;
285
286restart:
287 nores = 0;
288 busy = 0;
289
290 q = start;
291 if (!q)
292 goto drop;
293
294 do {
295 struct net_device *slave = qdisc_dev(q);
296 struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
297
298 if (slave_txq->qdisc_sleeping != q)
299 continue;
300 if (netif_xmit_stopped(netdev_get_tx_queue(slave, subq)) ||
301 !netif_running(slave)) {
302 busy = 1;
303 continue;
304 }
305
306 switch (teql_resolve(skb, skb_res, slave, slave_txq)) {
307 case 0:
308 if (__netif_tx_trylock(slave_txq)) {
309 unsigned int length = qdisc_pkt_len(skb);
310
311 if (!netif_xmit_frozen_or_stopped(slave_txq) &&
312 netdev_start_xmit(skb, slave, slave_txq, false) ==
313 NETDEV_TX_OK) {
314 __netif_tx_unlock(slave_txq);
315 master->slaves = NEXT_SLAVE(q);
316 netif_wake_queue(dev);
317 master->tx_packets++;
318 master->tx_bytes += length;
319 return NETDEV_TX_OK;
320 }
321 __netif_tx_unlock(slave_txq);
322 }
323 if (netif_xmit_stopped(netdev_get_tx_queue(dev, 0)))
324 busy = 1;
325 break;
326 case 1:
327 master->slaves = NEXT_SLAVE(q);
328 return NETDEV_TX_OK;
329 default:
330 nores = 1;
331 break;
332 }
333 __skb_pull(skb, skb_network_offset(skb));
334 } while ((q = NEXT_SLAVE(q)) != start);
335
336 if (nores && skb_res == NULL) {
337 skb_res = skb;
338 goto restart;
339 }
340
341 if (busy) {
342 netif_stop_queue(dev);
343 return NETDEV_TX_BUSY;
344 }
345 master->tx_errors++;
346
347drop:
348 master->tx_dropped++;
349 dev_kfree_skb(skb);
350 return NETDEV_TX_OK;
351}
352
353static int teql_master_open(struct net_device *dev)
354{
355 struct Qdisc *q;
356 struct teql_master *m = netdev_priv(dev);
357 int mtu = 0xFFFE;
358 unsigned int flags = IFF_NOARP | IFF_MULTICAST;
359
360 if (m->slaves == NULL)
361 return -EUNATCH;
362
363 flags = FMASK;
364
365 q = m->slaves;
366 do {
367 struct net_device *slave = qdisc_dev(q);
368
369 if (slave == NULL)
370 return -EUNATCH;
371
372 if (slave->mtu < mtu)
373 mtu = slave->mtu;
374 if (slave->hard_header_len > LL_MAX_HEADER)
375 return -EINVAL;
376
377 /* If all the slaves are BROADCAST, master is BROADCAST
378 If all the slaves are PtP, master is PtP
379 Otherwise, master is NBMA.
380 */
381 if (!(slave->flags&IFF_POINTOPOINT))
382 flags &= ~IFF_POINTOPOINT;
383 if (!(slave->flags&IFF_BROADCAST))
384 flags &= ~IFF_BROADCAST;
385 if (!(slave->flags&IFF_MULTICAST))
386 flags &= ~IFF_MULTICAST;
387 } while ((q = NEXT_SLAVE(q)) != m->slaves);
388
389 m->dev->mtu = mtu;
390 m->dev->flags = (m->dev->flags&~FMASK) | flags;
391 netif_start_queue(m->dev);
392 return 0;
393}
394
395static int teql_master_close(struct net_device *dev)
396{
397 netif_stop_queue(dev);
398 return 0;
399}
400
401static void teql_master_stats64(struct net_device *dev,
402 struct rtnl_link_stats64 *stats)
403{
404 struct teql_master *m = netdev_priv(dev);
405
406 stats->tx_packets = m->tx_packets;
407 stats->tx_bytes = m->tx_bytes;
408 stats->tx_errors = m->tx_errors;
409 stats->tx_dropped = m->tx_dropped;
410}
411
412static int teql_master_mtu(struct net_device *dev, int new_mtu)
413{
414 struct teql_master *m = netdev_priv(dev);
415 struct Qdisc *q;
416
417 q = m->slaves;
418 if (q) {
419 do {
420 if (new_mtu > qdisc_dev(q)->mtu)
421 return -EINVAL;
422 } while ((q = NEXT_SLAVE(q)) != m->slaves);
423 }
424
425 dev->mtu = new_mtu;
426 return 0;
427}
428
429static const struct net_device_ops teql_netdev_ops = {
430 .ndo_open = teql_master_open,
431 .ndo_stop = teql_master_close,
432 .ndo_start_xmit = teql_master_xmit,
433 .ndo_get_stats64 = teql_master_stats64,
434 .ndo_change_mtu = teql_master_mtu,
435};
436
437static __init void teql_master_setup(struct net_device *dev)
438{
439 struct teql_master *master = netdev_priv(dev);
440 struct Qdisc_ops *ops = &master->qops;
441
442 master->dev = dev;
443 ops->priv_size = sizeof(struct teql_sched_data);
444
445 ops->enqueue = teql_enqueue;
446 ops->dequeue = teql_dequeue;
447 ops->peek = teql_peek;
448 ops->init = teql_qdisc_init;
449 ops->reset = teql_reset;
450 ops->destroy = teql_destroy;
451 ops->owner = THIS_MODULE;
452
453 dev->netdev_ops = &teql_netdev_ops;
454 dev->type = ARPHRD_VOID;
455 dev->mtu = 1500;
456 dev->min_mtu = 68;
457 dev->max_mtu = 65535;
458 dev->tx_queue_len = 100;
459 dev->flags = IFF_NOARP;
460 dev->hard_header_len = LL_MAX_HEADER;
461 netif_keep_dst(dev);
462}
463
464static LIST_HEAD(master_dev_list);
465static int max_equalizers = 1;
466module_param(max_equalizers, int, 0);
467MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
468
469static int __init teql_init(void)
470{
471 int i;
472 int err = -ENODEV;
473
474 for (i = 0; i < max_equalizers; i++) {
475 struct net_device *dev;
476 struct teql_master *master;
477
478 dev = alloc_netdev(sizeof(struct teql_master), "teql%d",
479 NET_NAME_UNKNOWN, teql_master_setup);
480 if (!dev) {
481 err = -ENOMEM;
482 break;
483 }
484
485 if ((err = register_netdev(dev))) {
486 free_netdev(dev);
487 break;
488 }
489
490 master = netdev_priv(dev);
491
492 strlcpy(master->qops.id, dev->name, IFNAMSIZ);
493 err = register_qdisc(&master->qops);
494
495 if (err) {
496 unregister_netdev(dev);
497 free_netdev(dev);
498 break;
499 }
500
501 list_add_tail(&master->master_list, &master_dev_list);
502 }
503 return i ? 0 : err;
504}
505
506static void __exit teql_exit(void)
507{
508 struct teql_master *master, *nxt;
509
510 list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
511
512 list_del(&master->master_list);
513
514 unregister_qdisc(&master->qops);
515 unregister_netdev(master->dev);
516 free_netdev(master->dev);
517 }
518}
519
520module_init(teql_init);
521module_exit(teql_exit);
522
523MODULE_LICENSE("GPL");
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>
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>
22#include <net/dst.h>
23#include <net/neighbour.h>
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
56struct teql_master {
57 struct Qdisc_ops qops;
58 struct net_device *dev;
59 struct Qdisc *slaves;
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;
65};
66
67struct teql_sched_data {
68 struct Qdisc *next;
69 struct teql_master *m;
70 struct neighbour *ncache;
71 struct sk_buff_head q;
72};
73
74#define NEXT_SLAVE(q) (((struct teql_sched_data *)qdisc_priv(q))->next)
75
76#define FMASK (IFF_BROADCAST | IFF_POINTOPOINT)
77
78/* "teql*" qdisc routines */
79
80static int
81teql_enqueue(struct sk_buff *skb, struct Qdisc *sch)
82{
83 struct net_device *dev = qdisc_dev(sch);
84 struct teql_sched_data *q = qdisc_priv(sch);
85
86 if (q->q.qlen < dev->tx_queue_len) {
87 __skb_queue_tail(&q->q, skb);
88 return NET_XMIT_SUCCESS;
89 }
90
91 return qdisc_drop(skb, sch);
92}
93
94static struct sk_buff *
95teql_dequeue(struct Qdisc *sch)
96{
97 struct teql_sched_data *dat = qdisc_priv(sch);
98 struct netdev_queue *dat_queue;
99 struct sk_buff *skb;
100
101 skb = __skb_dequeue(&dat->q);
102 dat_queue = netdev_get_tx_queue(dat->m->dev, 0);
103 if (skb == NULL) {
104 struct net_device *m = qdisc_dev(dat_queue->qdisc);
105 if (m) {
106 dat->m->slaves = sch;
107 netif_wake_queue(m);
108 }
109 } else {
110 qdisc_bstats_update(sch, skb);
111 }
112 sch->q.qlen = dat->q.qlen + dat_queue->qdisc->q.qlen;
113 return skb;
114}
115
116static struct sk_buff *
117teql_peek(struct Qdisc *sch)
118{
119 /* teql is meant to be used as root qdisc */
120 return NULL;
121}
122
123static inline void
124teql_neigh_release(struct neighbour *n)
125{
126 if (n)
127 neigh_release(n);
128}
129
130static void
131teql_reset(struct Qdisc *sch)
132{
133 struct teql_sched_data *dat = qdisc_priv(sch);
134
135 skb_queue_purge(&dat->q);
136 sch->q.qlen = 0;
137 teql_neigh_release(xchg(&dat->ncache, NULL));
138}
139
140static void
141teql_destroy(struct Qdisc *sch)
142{
143 struct Qdisc *q, *prev;
144 struct teql_sched_data *dat = qdisc_priv(sch);
145 struct teql_master *master = dat->m;
146
147 prev = master->slaves;
148 if (prev) {
149 do {
150 q = NEXT_SLAVE(prev);
151 if (q == sch) {
152 NEXT_SLAVE(prev) = NEXT_SLAVE(q);
153 if (q == master->slaves) {
154 master->slaves = NEXT_SLAVE(q);
155 if (q == master->slaves) {
156 struct netdev_queue *txq;
157 spinlock_t *root_lock;
158
159 txq = netdev_get_tx_queue(master->dev, 0);
160 master->slaves = NULL;
161
162 root_lock = qdisc_root_sleeping_lock(txq->qdisc);
163 spin_lock_bh(root_lock);
164 qdisc_reset(txq->qdisc);
165 spin_unlock_bh(root_lock);
166 }
167 }
168 skb_queue_purge(&dat->q);
169 teql_neigh_release(xchg(&dat->ncache, NULL));
170 break;
171 }
172
173 } while ((prev = q) != master->slaves);
174 }
175}
176
177static int teql_qdisc_init(struct Qdisc *sch, struct nlattr *opt)
178{
179 struct net_device *dev = qdisc_dev(sch);
180 struct teql_master *m = (struct teql_master *)sch->ops;
181 struct teql_sched_data *q = qdisc_priv(sch);
182
183 if (dev->hard_header_len > m->dev->hard_header_len)
184 return -EINVAL;
185
186 if (m->dev == dev)
187 return -ELOOP;
188
189 q->m = m;
190
191 skb_queue_head_init(&q->q);
192
193 if (m->slaves) {
194 if (m->dev->flags & IFF_UP) {
195 if ((m->dev->flags & IFF_POINTOPOINT &&
196 !(dev->flags & IFF_POINTOPOINT)) ||
197 (m->dev->flags & IFF_BROADCAST &&
198 !(dev->flags & IFF_BROADCAST)) ||
199 (m->dev->flags & IFF_MULTICAST &&
200 !(dev->flags & IFF_MULTICAST)) ||
201 dev->mtu < m->dev->mtu)
202 return -EINVAL;
203 } else {
204 if (!(dev->flags&IFF_POINTOPOINT))
205 m->dev->flags &= ~IFF_POINTOPOINT;
206 if (!(dev->flags&IFF_BROADCAST))
207 m->dev->flags &= ~IFF_BROADCAST;
208 if (!(dev->flags&IFF_MULTICAST))
209 m->dev->flags &= ~IFF_MULTICAST;
210 if (dev->mtu < m->dev->mtu)
211 m->dev->mtu = dev->mtu;
212 }
213 q->next = NEXT_SLAVE(m->slaves);
214 NEXT_SLAVE(m->slaves) = sch;
215 } else {
216 q->next = sch;
217 m->slaves = sch;
218 m->dev->mtu = dev->mtu;
219 m->dev->flags = (m->dev->flags&~FMASK)|(dev->flags&FMASK);
220 }
221 return 0;
222}
223
224
225static int
226__teql_resolve(struct sk_buff *skb, struct sk_buff *skb_res,
227 struct net_device *dev, struct netdev_queue *txq,
228 struct neighbour *mn)
229{
230 struct teql_sched_data *q = qdisc_priv(txq->qdisc);
231 struct neighbour *n = q->ncache;
232
233 if (mn->tbl == NULL)
234 return -EINVAL;
235 if (n && n->tbl == mn->tbl &&
236 memcmp(n->primary_key, mn->primary_key, mn->tbl->key_len) == 0) {
237 atomic_inc(&n->refcnt);
238 } else {
239 n = __neigh_lookup_errno(mn->tbl, mn->primary_key, dev);
240 if (IS_ERR(n))
241 return PTR_ERR(n);
242 }
243 if (neigh_event_send(n, skb_res) == 0) {
244 int err;
245 char haddr[MAX_ADDR_LEN];
246
247 neigh_ha_snapshot(haddr, n, dev);
248 err = dev_hard_header(skb, dev, ntohs(skb->protocol), haddr,
249 NULL, skb->len);
250
251 if (err < 0) {
252 neigh_release(n);
253 return -EINVAL;
254 }
255 teql_neigh_release(xchg(&q->ncache, n));
256 return 0;
257 }
258 neigh_release(n);
259 return (skb_res == NULL) ? -EAGAIN : 1;
260}
261
262static inline int teql_resolve(struct sk_buff *skb,
263 struct sk_buff *skb_res,
264 struct net_device *dev,
265 struct netdev_queue *txq)
266{
267 struct dst_entry *dst = skb_dst(skb);
268 struct neighbour *mn;
269 int res;
270
271 if (txq->qdisc == &noop_qdisc)
272 return -ENODEV;
273
274 if (!dev->header_ops || !dst)
275 return 0;
276
277 rcu_read_lock();
278 mn = dst_get_neighbour_noref(dst);
279 res = mn ? __teql_resolve(skb, skb_res, dev, txq, mn) : 0;
280 rcu_read_unlock();
281
282 return res;
283}
284
285static netdev_tx_t teql_master_xmit(struct sk_buff *skb, struct net_device *dev)
286{
287 struct teql_master *master = netdev_priv(dev);
288 struct Qdisc *start, *q;
289 int busy;
290 int nores;
291 int subq = skb_get_queue_mapping(skb);
292 struct sk_buff *skb_res = NULL;
293
294 start = master->slaves;
295
296restart:
297 nores = 0;
298 busy = 0;
299
300 q = start;
301 if (!q)
302 goto drop;
303
304 do {
305 struct net_device *slave = qdisc_dev(q);
306 struct netdev_queue *slave_txq = netdev_get_tx_queue(slave, 0);
307 const struct net_device_ops *slave_ops = slave->netdev_ops;
308
309 if (slave_txq->qdisc_sleeping != q)
310 continue;
311 if (netif_xmit_stopped(netdev_get_tx_queue(slave, subq)) ||
312 !netif_running(slave)) {
313 busy = 1;
314 continue;
315 }
316
317 switch (teql_resolve(skb, skb_res, slave, slave_txq)) {
318 case 0:
319 if (__netif_tx_trylock(slave_txq)) {
320 unsigned int length = qdisc_pkt_len(skb);
321
322 if (!netif_xmit_frozen_or_stopped(slave_txq) &&
323 slave_ops->ndo_start_xmit(skb, slave) == NETDEV_TX_OK) {
324 txq_trans_update(slave_txq);
325 __netif_tx_unlock(slave_txq);
326 master->slaves = NEXT_SLAVE(q);
327 netif_wake_queue(dev);
328 master->tx_packets++;
329 master->tx_bytes += length;
330 return NETDEV_TX_OK;
331 }
332 __netif_tx_unlock(slave_txq);
333 }
334 if (netif_xmit_stopped(netdev_get_tx_queue(dev, 0)))
335 busy = 1;
336 break;
337 case 1:
338 master->slaves = NEXT_SLAVE(q);
339 return NETDEV_TX_OK;
340 default:
341 nores = 1;
342 break;
343 }
344 __skb_pull(skb, skb_network_offset(skb));
345 } while ((q = NEXT_SLAVE(q)) != start);
346
347 if (nores && skb_res == NULL) {
348 skb_res = skb;
349 goto restart;
350 }
351
352 if (busy) {
353 netif_stop_queue(dev);
354 return NETDEV_TX_BUSY;
355 }
356 master->tx_errors++;
357
358drop:
359 master->tx_dropped++;
360 dev_kfree_skb(skb);
361 return NETDEV_TX_OK;
362}
363
364static int teql_master_open(struct net_device *dev)
365{
366 struct Qdisc *q;
367 struct teql_master *m = netdev_priv(dev);
368 int mtu = 0xFFFE;
369 unsigned int flags = IFF_NOARP | IFF_MULTICAST;
370
371 if (m->slaves == NULL)
372 return -EUNATCH;
373
374 flags = FMASK;
375
376 q = m->slaves;
377 do {
378 struct net_device *slave = qdisc_dev(q);
379
380 if (slave == NULL)
381 return -EUNATCH;
382
383 if (slave->mtu < mtu)
384 mtu = slave->mtu;
385 if (slave->hard_header_len > LL_MAX_HEADER)
386 return -EINVAL;
387
388 /* If all the slaves are BROADCAST, master is BROADCAST
389 If all the slaves are PtP, master is PtP
390 Otherwise, master is NBMA.
391 */
392 if (!(slave->flags&IFF_POINTOPOINT))
393 flags &= ~IFF_POINTOPOINT;
394 if (!(slave->flags&IFF_BROADCAST))
395 flags &= ~IFF_BROADCAST;
396 if (!(slave->flags&IFF_MULTICAST))
397 flags &= ~IFF_MULTICAST;
398 } while ((q = NEXT_SLAVE(q)) != m->slaves);
399
400 m->dev->mtu = mtu;
401 m->dev->flags = (m->dev->flags&~FMASK) | flags;
402 netif_start_queue(m->dev);
403 return 0;
404}
405
406static int teql_master_close(struct net_device *dev)
407{
408 netif_stop_queue(dev);
409 return 0;
410}
411
412static struct rtnl_link_stats64 *teql_master_stats64(struct net_device *dev,
413 struct rtnl_link_stats64 *stats)
414{
415 struct teql_master *m = netdev_priv(dev);
416
417 stats->tx_packets = m->tx_packets;
418 stats->tx_bytes = m->tx_bytes;
419 stats->tx_errors = m->tx_errors;
420 stats->tx_dropped = m->tx_dropped;
421 return stats;
422}
423
424static int teql_master_mtu(struct net_device *dev, int new_mtu)
425{
426 struct teql_master *m = netdev_priv(dev);
427 struct Qdisc *q;
428
429 if (new_mtu < 68)
430 return -EINVAL;
431
432 q = m->slaves;
433 if (q) {
434 do {
435 if (new_mtu > qdisc_dev(q)->mtu)
436 return -EINVAL;
437 } while ((q = NEXT_SLAVE(q)) != m->slaves);
438 }
439
440 dev->mtu = new_mtu;
441 return 0;
442}
443
444static const struct net_device_ops teql_netdev_ops = {
445 .ndo_open = teql_master_open,
446 .ndo_stop = teql_master_close,
447 .ndo_start_xmit = teql_master_xmit,
448 .ndo_get_stats64 = teql_master_stats64,
449 .ndo_change_mtu = teql_master_mtu,
450};
451
452static __init void teql_master_setup(struct net_device *dev)
453{
454 struct teql_master *master = netdev_priv(dev);
455 struct Qdisc_ops *ops = &master->qops;
456
457 master->dev = dev;
458 ops->priv_size = sizeof(struct teql_sched_data);
459
460 ops->enqueue = teql_enqueue;
461 ops->dequeue = teql_dequeue;
462 ops->peek = teql_peek;
463 ops->init = teql_qdisc_init;
464 ops->reset = teql_reset;
465 ops->destroy = teql_destroy;
466 ops->owner = THIS_MODULE;
467
468 dev->netdev_ops = &teql_netdev_ops;
469 dev->type = ARPHRD_VOID;
470 dev->mtu = 1500;
471 dev->tx_queue_len = 100;
472 dev->flags = IFF_NOARP;
473 dev->hard_header_len = LL_MAX_HEADER;
474 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
475}
476
477static LIST_HEAD(master_dev_list);
478static int max_equalizers = 1;
479module_param(max_equalizers, int, 0);
480MODULE_PARM_DESC(max_equalizers, "Max number of link equalizers");
481
482static int __init teql_init(void)
483{
484 int i;
485 int err = -ENODEV;
486
487 for (i = 0; i < max_equalizers; i++) {
488 struct net_device *dev;
489 struct teql_master *master;
490
491 dev = alloc_netdev(sizeof(struct teql_master),
492 "teql%d", teql_master_setup);
493 if (!dev) {
494 err = -ENOMEM;
495 break;
496 }
497
498 if ((err = register_netdev(dev))) {
499 free_netdev(dev);
500 break;
501 }
502
503 master = netdev_priv(dev);
504
505 strlcpy(master->qops.id, dev->name, IFNAMSIZ);
506 err = register_qdisc(&master->qops);
507
508 if (err) {
509 unregister_netdev(dev);
510 free_netdev(dev);
511 break;
512 }
513
514 list_add_tail(&master->master_list, &master_dev_list);
515 }
516 return i ? 0 : err;
517}
518
519static void __exit teql_exit(void)
520{
521 struct teql_master *master, *nxt;
522
523 list_for_each_entry_safe(master, nxt, &master_dev_list, master_list) {
524
525 list_del(&master->master_list);
526
527 unregister_qdisc(&master->qops);
528 unregister_netdev(master->dev);
529 free_netdev(master->dev);
530 }
531}
532
533module_init(teql_init);
534module_exit(teql_exit);
535
536MODULE_LICENSE("GPL");