1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* drivers/net/ifb.c:
3
4 The purpose of this driver is to provide a device that allows
5 for sharing of resources:
6
7 1) qdiscs/policies that are per device as opposed to system wide.
8 ifb allows for a device which can be redirected to thus providing
9 an impression of sharing.
10
11 2) Allows for queueing incoming traffic for shaping instead of
12 dropping.
13
14 The original concept is based on what is known as the IMQ
15 driver initially written by Martin Devera, later rewritten
16 by Patrick McHardy and then maintained by Andre Correa.
17
18 You need the tc action mirror or redirect to feed this device
19 packets.
20
21
22 Authors: Jamal Hadi Salim (2005)
23
24 */
25
26
27 #include <linux/module.h>
28 #include <linux/kernel.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
31 #include <linux/init.h>
32 #include <linux/interrupt.h>
33 #include <linux/moduleparam.h>
34 #include <net/pkt_sched.h>
35 #include <net/net_namespace.h>
36
37 #define TX_Q_LIMIT 32
38 struct ifb_q_private {
39 struct net_device *dev;
40 struct tasklet_struct ifb_tasklet;
41 int tasklet_pending;
42 int txqnum;
43 struct sk_buff_head rq;
44 u64 rx_packets;
45 u64 rx_bytes;
46 struct u64_stats_sync rsync;
47
48 struct u64_stats_sync tsync;
49 u64 tx_packets;
50 u64 tx_bytes;
51 struct sk_buff_head tq;
52 } ____cacheline_aligned_in_smp;
53
54 struct ifb_dev_private {
55 struct ifb_q_private *tx_private;
56 };
57
58 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev);
59 static int ifb_open(struct net_device *dev);
60 static int ifb_close(struct net_device *dev);
61
ifb_ri_tasklet(unsigned long _txp)62 static void ifb_ri_tasklet(unsigned long _txp)
63 {
64 struct ifb_q_private *txp = (struct ifb_q_private *)_txp;
65 struct netdev_queue *txq;
66 struct sk_buff *skb;
67
68 txq = netdev_get_tx_queue(txp->dev, txp->txqnum);
69 skb = skb_peek(&txp->tq);
70 if (!skb) {
71 if (!__netif_tx_trylock(txq))
72 goto resched;
73 skb_queue_splice_tail_init(&txp->rq, &txp->tq);
74 __netif_tx_unlock(txq);
75 }
76
77 while ((skb = __skb_dequeue(&txp->tq)) != NULL) {
78 skb->redirected = 0;
79 #ifdef CONFIG_NET_CLS_ACT
80 skb->tc_skip_classify = 1;
81 #endif
82
83 u64_stats_update_begin(&txp->tsync);
84 txp->tx_packets++;
85 txp->tx_bytes += skb->len;
86 u64_stats_update_end(&txp->tsync);
87
88 rcu_read_lock();
89 skb->dev = dev_get_by_index_rcu(dev_net(txp->dev), skb->skb_iif);
90 if (!skb->dev) {
91 rcu_read_unlock();
92 dev_kfree_skb(skb);
93 txp->dev->stats.tx_dropped++;
94 if (skb_queue_len(&txp->tq) != 0)
95 goto resched;
96 break;
97 }
98 rcu_read_unlock();
99 skb->skb_iif = txp->dev->ifindex;
100
101 if (!skb->from_ingress) {
102 dev_queue_xmit(skb);
103 } else {
104 skb_pull_rcsum(skb, skb->mac_len);
105 netif_receive_skb(skb);
106 }
107 }
108
109 if (__netif_tx_trylock(txq)) {
110 skb = skb_peek(&txp->rq);
111 if (!skb) {
112 txp->tasklet_pending = 0;
113 if (netif_tx_queue_stopped(txq))
114 netif_tx_wake_queue(txq);
115 } else {
116 __netif_tx_unlock(txq);
117 goto resched;
118 }
119 __netif_tx_unlock(txq);
120 } else {
121 resched:
122 txp->tasklet_pending = 1;
123 tasklet_schedule(&txp->ifb_tasklet);
124 }
125
126 }
127
ifb_stats64(struct net_device * dev,struct rtnl_link_stats64 * stats)128 static void ifb_stats64(struct net_device *dev,
129 struct rtnl_link_stats64 *stats)
130 {
131 struct ifb_dev_private *dp = netdev_priv(dev);
132 struct ifb_q_private *txp = dp->tx_private;
133 unsigned int start;
134 u64 packets, bytes;
135 int i;
136
137 for (i = 0; i < dev->num_tx_queues; i++,txp++) {
138 do {
139 start = u64_stats_fetch_begin_irq(&txp->rsync);
140 packets = txp->rx_packets;
141 bytes = txp->rx_bytes;
142 } while (u64_stats_fetch_retry_irq(&txp->rsync, start));
143 stats->rx_packets += packets;
144 stats->rx_bytes += bytes;
145
146 do {
147 start = u64_stats_fetch_begin_irq(&txp->tsync);
148 packets = txp->tx_packets;
149 bytes = txp->tx_bytes;
150 } while (u64_stats_fetch_retry_irq(&txp->tsync, start));
151 stats->tx_packets += packets;
152 stats->tx_bytes += bytes;
153 }
154 stats->rx_dropped = dev->stats.rx_dropped;
155 stats->tx_dropped = dev->stats.tx_dropped;
156 }
157
ifb_dev_init(struct net_device * dev)158 static int ifb_dev_init(struct net_device *dev)
159 {
160 struct ifb_dev_private *dp = netdev_priv(dev);
161 struct ifb_q_private *txp;
162 int i;
163
164 txp = kcalloc(dev->num_tx_queues, sizeof(*txp), GFP_KERNEL);
165 if (!txp)
166 return -ENOMEM;
167 dp->tx_private = txp;
168 for (i = 0; i < dev->num_tx_queues; i++,txp++) {
169 txp->txqnum = i;
170 txp->dev = dev;
171 __skb_queue_head_init(&txp->rq);
172 __skb_queue_head_init(&txp->tq);
173 u64_stats_init(&txp->rsync);
174 u64_stats_init(&txp->tsync);
175 tasklet_init(&txp->ifb_tasklet, ifb_ri_tasklet,
176 (unsigned long)txp);
177 netif_tx_start_queue(netdev_get_tx_queue(dev, i));
178 }
179 return 0;
180 }
181
182 static const struct net_device_ops ifb_netdev_ops = {
183 .ndo_open = ifb_open,
184 .ndo_stop = ifb_close,
185 .ndo_get_stats64 = ifb_stats64,
186 .ndo_start_xmit = ifb_xmit,
187 .ndo_validate_addr = eth_validate_addr,
188 .ndo_init = ifb_dev_init,
189 };
190
191 #define IFB_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | NETIF_F_FRAGLIST | \
192 NETIF_F_TSO_ECN | NETIF_F_TSO | NETIF_F_TSO6 | \
193 NETIF_F_GSO_ENCAP_ALL | \
194 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX | \
195 NETIF_F_HW_VLAN_STAG_TX)
196
ifb_dev_free(struct net_device * dev)197 static void ifb_dev_free(struct net_device *dev)
198 {
199 struct ifb_dev_private *dp = netdev_priv(dev);
200 struct ifb_q_private *txp = dp->tx_private;
201 int i;
202
203 for (i = 0; i < dev->num_tx_queues; i++,txp++) {
204 tasklet_kill(&txp->ifb_tasklet);
205 __skb_queue_purge(&txp->rq);
206 __skb_queue_purge(&txp->tq);
207 }
208 kfree(dp->tx_private);
209 }
210
ifb_setup(struct net_device * dev)211 static void ifb_setup(struct net_device *dev)
212 {
213 /* Initialize the device structure. */
214 dev->netdev_ops = &ifb_netdev_ops;
215
216 /* Fill in device structure with ethernet-generic values. */
217 ether_setup(dev);
218 dev->tx_queue_len = TX_Q_LIMIT;
219
220 dev->features |= IFB_FEATURES;
221 dev->hw_features |= dev->features;
222 dev->hw_enc_features |= dev->features;
223 dev->vlan_features |= IFB_FEATURES & ~(NETIF_F_HW_VLAN_CTAG_TX |
224 NETIF_F_HW_VLAN_STAG_TX);
225
226 dev->flags |= IFF_NOARP;
227 dev->flags &= ~IFF_MULTICAST;
228 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
229 netif_keep_dst(dev);
230 eth_hw_addr_random(dev);
231 dev->needs_free_netdev = true;
232 dev->priv_destructor = ifb_dev_free;
233
234 dev->min_mtu = 0;
235 dev->max_mtu = 0;
236 }
237
ifb_xmit(struct sk_buff * skb,struct net_device * dev)238 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev)
239 {
240 struct ifb_dev_private *dp = netdev_priv(dev);
241 struct ifb_q_private *txp = dp->tx_private + skb_get_queue_mapping(skb);
242
243 u64_stats_update_begin(&txp->rsync);
244 txp->rx_packets++;
245 txp->rx_bytes += skb->len;
246 u64_stats_update_end(&txp->rsync);
247
248 if (!skb->redirected || !skb->skb_iif) {
249 dev_kfree_skb(skb);
250 dev->stats.rx_dropped++;
251 return NETDEV_TX_OK;
252 }
253
254 if (skb_queue_len(&txp->rq) >= dev->tx_queue_len)
255 netif_tx_stop_queue(netdev_get_tx_queue(dev, txp->txqnum));
256
257 __skb_queue_tail(&txp->rq, skb);
258 if (!txp->tasklet_pending) {
259 txp->tasklet_pending = 1;
260 tasklet_schedule(&txp->ifb_tasklet);
261 }
262
263 return NETDEV_TX_OK;
264 }
265
ifb_close(struct net_device * dev)266 static int ifb_close(struct net_device *dev)
267 {
268 netif_tx_stop_all_queues(dev);
269 return 0;
270 }
271
ifb_open(struct net_device * dev)272 static int ifb_open(struct net_device *dev)
273 {
274 netif_tx_start_all_queues(dev);
275 return 0;
276 }
277
ifb_validate(struct nlattr * tb[],struct nlattr * data[],struct netlink_ext_ack * extack)278 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[],
279 struct netlink_ext_ack *extack)
280 {
281 if (tb[IFLA_ADDRESS]) {
282 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
283 return -EINVAL;
284 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
285 return -EADDRNOTAVAIL;
286 }
287 return 0;
288 }
289
290 static struct rtnl_link_ops ifb_link_ops __read_mostly = {
291 .kind = "ifb",
292 .priv_size = sizeof(struct ifb_dev_private),
293 .setup = ifb_setup,
294 .validate = ifb_validate,
295 };
296
297 /* Number of ifb devices to be set up by this module.
298 * Note that these legacy devices have one queue.
299 * Prefer something like : ip link add ifb10 numtxqueues 8 type ifb
300 */
301 static int numifbs = 2;
302 module_param(numifbs, int, 0);
303 MODULE_PARM_DESC(numifbs, "Number of ifb devices");
304
ifb_init_one(int index)305 static int __init ifb_init_one(int index)
306 {
307 struct net_device *dev_ifb;
308 int err;
309
310 dev_ifb = alloc_netdev(sizeof(struct ifb_dev_private), "ifb%d",
311 NET_NAME_UNKNOWN, ifb_setup);
312
313 if (!dev_ifb)
314 return -ENOMEM;
315
316 dev_ifb->rtnl_link_ops = &ifb_link_ops;
317 err = register_netdevice(dev_ifb);
318 if (err < 0)
319 goto err;
320
321 return 0;
322
323 err:
324 free_netdev(dev_ifb);
325 return err;
326 }
327
ifb_init_module(void)328 static int __init ifb_init_module(void)
329 {
330 int i, err;
331
332 down_write(&pernet_ops_rwsem);
333 rtnl_lock();
334 err = __rtnl_link_register(&ifb_link_ops);
335 if (err < 0)
336 goto out;
337
338 for (i = 0; i < numifbs && !err; i++) {
339 err = ifb_init_one(i);
340 cond_resched();
341 }
342 if (err)
343 __rtnl_link_unregister(&ifb_link_ops);
344
345 out:
346 rtnl_unlock();
347 up_write(&pernet_ops_rwsem);
348
349 return err;
350 }
351
ifb_cleanup_module(void)352 static void __exit ifb_cleanup_module(void)
353 {
354 rtnl_link_unregister(&ifb_link_ops);
355 }
356
357 module_init(ifb_init_module);
358 module_exit(ifb_cleanup_module);
359 MODULE_LICENSE("GPL");
360 MODULE_AUTHOR("Jamal Hadi Salim");
361 MODULE_ALIAS_RTNL_LINK("ifb");
362