]>
Commit | Line | Data |
---|---|---|
1 | /* drivers/net/ifb.c: | |
2 | ||
3 | The purpose of this driver is to provide a device that allows | |
4 | for sharing of resources: | |
5 | ||
6 | 1) qdiscs/policies that are per device as opposed to system wide. | |
7 | ifb allows for a device which can be redirected to thus providing | |
8 | an impression of sharing. | |
9 | ||
10 | 2) Allows for queueing incoming traffic for shaping instead of | |
11 | dropping. | |
12 | ||
13 | The original concept is based on what is known as the IMQ | |
14 | driver initially written by Martin Devera, later rewritten | |
15 | by Patrick McHardy and then maintained by Andre Correa. | |
16 | ||
17 | You need the tc action mirror or redirect to feed this device | |
18 | packets. | |
19 | ||
20 | This program is free software; you can redistribute it and/or | |
21 | modify it under the terms of the GNU General Public License | |
22 | as published by the Free Software Foundation; either version | |
23 | 2 of the License, or (at your option) any later version. | |
24 | ||
25 | Authors: Jamal Hadi Salim (2005) | |
26 | ||
27 | */ | |
28 | ||
29 | ||
30 | #include <linux/module.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/netdevice.h> | |
33 | #include <linux/etherdevice.h> | |
34 | #include <linux/init.h> | |
35 | #include <linux/interrupt.h> | |
36 | #include <linux/moduleparam.h> | |
37 | #include <net/pkt_sched.h> | |
38 | #include <net/net_namespace.h> | |
39 | ||
40 | #define TX_Q_LIMIT 32 | |
41 | struct ifb_private { | |
42 | struct tasklet_struct ifb_tasklet; | |
43 | int tasklet_pending; | |
44 | ||
45 | struct u64_stats_sync rsync; | |
46 | struct sk_buff_head rq; | |
47 | u64 rx_packets; | |
48 | u64 rx_bytes; | |
49 | ||
50 | struct u64_stats_sync tsync; | |
51 | struct sk_buff_head tq; | |
52 | u64 tx_packets; | |
53 | u64 tx_bytes; | |
54 | }; | |
55 | ||
56 | static int numifbs = 2; | |
57 | ||
58 | static void ri_tasklet(unsigned long dev); | |
59 | static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev); | |
60 | static int ifb_open(struct net_device *dev); | |
61 | static int ifb_close(struct net_device *dev); | |
62 | ||
63 | static void ri_tasklet(unsigned long dev) | |
64 | { | |
65 | struct net_device *_dev = (struct net_device *)dev; | |
66 | struct ifb_private *dp = netdev_priv(_dev); | |
67 | struct netdev_queue *txq; | |
68 | struct sk_buff *skb; | |
69 | ||
70 | txq = netdev_get_tx_queue(_dev, 0); | |
71 | if ((skb = skb_peek(&dp->tq)) == NULL) { | |
72 | if (__netif_tx_trylock(txq)) { | |
73 | skb_queue_splice_tail_init(&dp->rq, &dp->tq); | |
74 | __netif_tx_unlock(txq); | |
75 | } else { | |
76 | /* reschedule */ | |
77 | goto resched; | |
78 | } | |
79 | } | |
80 | ||
81 | while ((skb = __skb_dequeue(&dp->tq)) != NULL) { | |
82 | u32 from = G_TC_FROM(skb->tc_verd); | |
83 | ||
84 | skb->tc_verd = 0; | |
85 | skb->tc_verd = SET_TC_NCLS(skb->tc_verd); | |
86 | ||
87 | u64_stats_update_begin(&dp->tsync); | |
88 | dp->tx_packets++; | |
89 | dp->tx_bytes += skb->len; | |
90 | u64_stats_update_end(&dp->tsync); | |
91 | ||
92 | rcu_read_lock(); | |
93 | skb->dev = dev_get_by_index_rcu(&init_net, skb->skb_iif); | |
94 | if (!skb->dev) { | |
95 | rcu_read_unlock(); | |
96 | dev_kfree_skb(skb); | |
97 | _dev->stats.tx_dropped++; | |
98 | if (skb_queue_len(&dp->tq) != 0) | |
99 | goto resched; | |
100 | break; | |
101 | } | |
102 | rcu_read_unlock(); | |
103 | skb->skb_iif = _dev->ifindex; | |
104 | ||
105 | if (from & AT_EGRESS) { | |
106 | dev_queue_xmit(skb); | |
107 | } else if (from & AT_INGRESS) { | |
108 | skb_pull(skb, skb->dev->hard_header_len); | |
109 | netif_receive_skb(skb); | |
110 | } else | |
111 | BUG(); | |
112 | } | |
113 | ||
114 | if (__netif_tx_trylock(txq)) { | |
115 | if ((skb = skb_peek(&dp->rq)) == NULL) { | |
116 | dp->tasklet_pending = 0; | |
117 | if (netif_queue_stopped(_dev)) | |
118 | netif_wake_queue(_dev); | |
119 | } else { | |
120 | __netif_tx_unlock(txq); | |
121 | goto resched; | |
122 | } | |
123 | __netif_tx_unlock(txq); | |
124 | } else { | |
125 | resched: | |
126 | dp->tasklet_pending = 1; | |
127 | tasklet_schedule(&dp->ifb_tasklet); | |
128 | } | |
129 | ||
130 | } | |
131 | ||
132 | static struct rtnl_link_stats64 *ifb_stats64(struct net_device *dev, | |
133 | struct rtnl_link_stats64 *stats) | |
134 | { | |
135 | struct ifb_private *dp = netdev_priv(dev); | |
136 | unsigned int start; | |
137 | ||
138 | do { | |
139 | start = u64_stats_fetch_begin_bh(&dp->rsync); | |
140 | stats->rx_packets = dp->rx_packets; | |
141 | stats->rx_bytes = dp->rx_bytes; | |
142 | } while (u64_stats_fetch_retry_bh(&dp->rsync, start)); | |
143 | ||
144 | do { | |
145 | start = u64_stats_fetch_begin_bh(&dp->tsync); | |
146 | ||
147 | stats->tx_packets = dp->tx_packets; | |
148 | stats->tx_bytes = dp->tx_bytes; | |
149 | ||
150 | } while (u64_stats_fetch_retry_bh(&dp->tsync, start)); | |
151 | ||
152 | stats->rx_dropped = dev->stats.rx_dropped; | |
153 | stats->tx_dropped = dev->stats.tx_dropped; | |
154 | ||
155 | return stats; | |
156 | } | |
157 | ||
158 | ||
159 | static const struct net_device_ops ifb_netdev_ops = { | |
160 | .ndo_open = ifb_open, | |
161 | .ndo_stop = ifb_close, | |
162 | .ndo_get_stats64 = ifb_stats64, | |
163 | .ndo_start_xmit = ifb_xmit, | |
164 | .ndo_validate_addr = eth_validate_addr, | |
165 | }; | |
166 | ||
167 | #define IFB_FEATURES (NETIF_F_NO_CSUM | NETIF_F_SG | NETIF_F_FRAGLIST | \ | |
168 | NETIF_F_TSO_ECN | NETIF_F_TSO | NETIF_F_TSO6 | \ | |
169 | NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_TX) | |
170 | ||
171 | static void ifb_setup(struct net_device *dev) | |
172 | { | |
173 | /* Initialize the device structure. */ | |
174 | dev->destructor = free_netdev; | |
175 | dev->netdev_ops = &ifb_netdev_ops; | |
176 | ||
177 | /* Fill in device structure with ethernet-generic values. */ | |
178 | ether_setup(dev); | |
179 | dev->tx_queue_len = TX_Q_LIMIT; | |
180 | ||
181 | dev->features |= IFB_FEATURES; | |
182 | dev->vlan_features |= IFB_FEATURES; | |
183 | ||
184 | dev->flags |= IFF_NOARP; | |
185 | dev->flags &= ~IFF_MULTICAST; | |
186 | dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); | |
187 | random_ether_addr(dev->dev_addr); | |
188 | } | |
189 | ||
190 | static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev) | |
191 | { | |
192 | struct ifb_private *dp = netdev_priv(dev); | |
193 | u32 from = G_TC_FROM(skb->tc_verd); | |
194 | ||
195 | u64_stats_update_begin(&dp->rsync); | |
196 | dp->rx_packets++; | |
197 | dp->rx_bytes += skb->len; | |
198 | u64_stats_update_end(&dp->rsync); | |
199 | ||
200 | if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->skb_iif) { | |
201 | dev_kfree_skb(skb); | |
202 | dev->stats.rx_dropped++; | |
203 | return NETDEV_TX_OK; | |
204 | } | |
205 | ||
206 | if (skb_queue_len(&dp->rq) >= dev->tx_queue_len) { | |
207 | netif_stop_queue(dev); | |
208 | } | |
209 | ||
210 | __skb_queue_tail(&dp->rq, skb); | |
211 | if (!dp->tasklet_pending) { | |
212 | dp->tasklet_pending = 1; | |
213 | tasklet_schedule(&dp->ifb_tasklet); | |
214 | } | |
215 | ||
216 | return NETDEV_TX_OK; | |
217 | } | |
218 | ||
219 | static int ifb_close(struct net_device *dev) | |
220 | { | |
221 | struct ifb_private *dp = netdev_priv(dev); | |
222 | ||
223 | tasklet_kill(&dp->ifb_tasklet); | |
224 | netif_stop_queue(dev); | |
225 | __skb_queue_purge(&dp->rq); | |
226 | __skb_queue_purge(&dp->tq); | |
227 | return 0; | |
228 | } | |
229 | ||
230 | static int ifb_open(struct net_device *dev) | |
231 | { | |
232 | struct ifb_private *dp = netdev_priv(dev); | |
233 | ||
234 | tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev); | |
235 | __skb_queue_head_init(&dp->rq); | |
236 | __skb_queue_head_init(&dp->tq); | |
237 | netif_start_queue(dev); | |
238 | ||
239 | return 0; | |
240 | } | |
241 | ||
242 | static int ifb_validate(struct nlattr *tb[], struct nlattr *data[]) | |
243 | { | |
244 | if (tb[IFLA_ADDRESS]) { | |
245 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) | |
246 | return -EINVAL; | |
247 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) | |
248 | return -EADDRNOTAVAIL; | |
249 | } | |
250 | return 0; | |
251 | } | |
252 | ||
253 | static struct rtnl_link_ops ifb_link_ops __read_mostly = { | |
254 | .kind = "ifb", | |
255 | .priv_size = sizeof(struct ifb_private), | |
256 | .setup = ifb_setup, | |
257 | .validate = ifb_validate, | |
258 | }; | |
259 | ||
260 | /* Number of ifb devices to be set up by this module. */ | |
261 | module_param(numifbs, int, 0); | |
262 | MODULE_PARM_DESC(numifbs, "Number of ifb devices"); | |
263 | ||
264 | static int __init ifb_init_one(int index) | |
265 | { | |
266 | struct net_device *dev_ifb; | |
267 | int err; | |
268 | ||
269 | dev_ifb = alloc_netdev(sizeof(struct ifb_private), | |
270 | "ifb%d", ifb_setup); | |
271 | ||
272 | if (!dev_ifb) | |
273 | return -ENOMEM; | |
274 | ||
275 | dev_ifb->rtnl_link_ops = &ifb_link_ops; | |
276 | err = register_netdevice(dev_ifb); | |
277 | if (err < 0) | |
278 | goto err; | |
279 | ||
280 | return 0; | |
281 | ||
282 | err: | |
283 | free_netdev(dev_ifb); | |
284 | return err; | |
285 | } | |
286 | ||
287 | static int __init ifb_init_module(void) | |
288 | { | |
289 | int i, err; | |
290 | ||
291 | rtnl_lock(); | |
292 | err = __rtnl_link_register(&ifb_link_ops); | |
293 | ||
294 | for (i = 0; i < numifbs && !err; i++) | |
295 | err = ifb_init_one(i); | |
296 | if (err) | |
297 | __rtnl_link_unregister(&ifb_link_ops); | |
298 | rtnl_unlock(); | |
299 | ||
300 | return err; | |
301 | } | |
302 | ||
303 | static void __exit ifb_cleanup_module(void) | |
304 | { | |
305 | rtnl_link_unregister(&ifb_link_ops); | |
306 | } | |
307 | ||
308 | module_init(ifb_init_module); | |
309 | module_exit(ifb_cleanup_module); | |
310 | MODULE_LICENSE("GPL"); | |
311 | MODULE_AUTHOR("Jamal Hadi Salim"); | |
312 | MODULE_ALIAS_RTNL_LINK("ifb"); |