xref: /linux/drivers/net/ifb.c (revision 4317892db474ddcf0f9d4a9bca8e0d2ddb1d0ab9)
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 <linux/netfilter_netdev.h>
35 #include <net/pkt_sched.h>
36 #include <net/net_namespace.h>
37 
38 #define TX_Q_LIMIT    32
39 struct ifb_q_private {
40 	struct net_device	*dev;
41 	struct tasklet_struct   ifb_tasklet;
42 	int			tasklet_pending;
43 	int			txqnum;
44 	struct sk_buff_head     rq;
45 	u64			rx_packets;
46 	u64			rx_bytes;
47 	struct u64_stats_sync	rsync;
48 
49 	struct u64_stats_sync	tsync;
50 	u64			tx_packets;
51 	u64			tx_bytes;
52 	struct sk_buff_head     tq;
53 } ____cacheline_aligned_in_smp;
54 
55 struct ifb_dev_private {
56 	struct ifb_q_private *tx_private;
57 };
58 
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 ifb_ri_tasklet(struct tasklet_struct *t)
64 {
65 	struct ifb_q_private *txp = from_tasklet(txp, t, ifb_tasklet);
66 	struct netdev_queue *txq;
67 	struct sk_buff *skb;
68 
69 	txq = netdev_get_tx_queue(txp->dev, txp->txqnum);
70 	skb = skb_peek(&txp->tq);
71 	if (!skb) {
72 		if (!__netif_tx_trylock(txq))
73 			goto resched;
74 		skb_queue_splice_tail_init(&txp->rq, &txp->tq);
75 		__netif_tx_unlock(txq);
76 	}
77 
78 	while ((skb = __skb_dequeue(&txp->tq)) != NULL) {
79 		/* Skip tc and netfilter to prevent redirection loop. */
80 		skb->redirected = 0;
81 #ifdef CONFIG_NET_CLS_ACT
82 		skb->tc_skip_classify = 1;
83 #endif
84 		nf_skip_egress(skb, true);
85 
86 		u64_stats_update_begin(&txp->tsync);
87 		txp->tx_packets++;
88 		txp->tx_bytes += skb->len;
89 		u64_stats_update_end(&txp->tsync);
90 
91 		rcu_read_lock();
92 		skb->dev = dev_get_by_index_rcu(dev_net(txp->dev), skb->skb_iif);
93 		if (!skb->dev) {
94 			rcu_read_unlock();
95 			dev_kfree_skb(skb);
96 			txp->dev->stats.tx_dropped++;
97 			if (skb_queue_len(&txp->tq) != 0)
98 				goto resched;
99 			break;
100 		}
101 		rcu_read_unlock();
102 		skb->skb_iif = txp->dev->ifindex;
103 
104 		if (!skb->from_ingress) {
105 			dev_queue_xmit(skb);
106 		} else {
107 			skb_pull_rcsum(skb, skb->mac_len);
108 			netif_receive_skb(skb);
109 		}
110 	}
111 
112 	if (__netif_tx_trylock(txq)) {
113 		skb = skb_peek(&txp->rq);
114 		if (!skb) {
115 			txp->tasklet_pending = 0;
116 			if (netif_tx_queue_stopped(txq))
117 				netif_tx_wake_queue(txq);
118 		} else {
119 			__netif_tx_unlock(txq);
120 			goto resched;
121 		}
122 		__netif_tx_unlock(txq);
123 	} else {
124 resched:
125 		txp->tasklet_pending = 1;
126 		tasklet_schedule(&txp->ifb_tasklet);
127 	}
128 
129 }
130 
131 static void ifb_stats64(struct net_device *dev,
132 			struct rtnl_link_stats64 *stats)
133 {
134 	struct ifb_dev_private *dp = netdev_priv(dev);
135 	struct ifb_q_private *txp = dp->tx_private;
136 	unsigned int start;
137 	u64 packets, bytes;
138 	int i;
139 
140 	for (i = 0; i < dev->num_tx_queues; i++,txp++) {
141 		do {
142 			start = u64_stats_fetch_begin_irq(&txp->rsync);
143 			packets = txp->rx_packets;
144 			bytes = txp->rx_bytes;
145 		} while (u64_stats_fetch_retry_irq(&txp->rsync, start));
146 		stats->rx_packets += packets;
147 		stats->rx_bytes += bytes;
148 
149 		do {
150 			start = u64_stats_fetch_begin_irq(&txp->tsync);
151 			packets = txp->tx_packets;
152 			bytes = txp->tx_bytes;
153 		} while (u64_stats_fetch_retry_irq(&txp->tsync, start));
154 		stats->tx_packets += packets;
155 		stats->tx_bytes += bytes;
156 	}
157 	stats->rx_dropped = dev->stats.rx_dropped;
158 	stats->tx_dropped = dev->stats.tx_dropped;
159 }
160 
161 static int ifb_dev_init(struct net_device *dev)
162 {
163 	struct ifb_dev_private *dp = netdev_priv(dev);
164 	struct ifb_q_private *txp;
165 	int i;
166 
167 	txp = kcalloc(dev->num_tx_queues, sizeof(*txp), GFP_KERNEL);
168 	if (!txp)
169 		return -ENOMEM;
170 	dp->tx_private = txp;
171 	for (i = 0; i < dev->num_tx_queues; i++,txp++) {
172 		txp->txqnum = i;
173 		txp->dev = dev;
174 		__skb_queue_head_init(&txp->rq);
175 		__skb_queue_head_init(&txp->tq);
176 		u64_stats_init(&txp->rsync);
177 		u64_stats_init(&txp->tsync);
178 		tasklet_setup(&txp->ifb_tasklet, ifb_ri_tasklet);
179 		netif_tx_start_queue(netdev_get_tx_queue(dev, i));
180 	}
181 	return 0;
182 }
183 
184 static const struct net_device_ops ifb_netdev_ops = {
185 	.ndo_open	= ifb_open,
186 	.ndo_stop	= ifb_close,
187 	.ndo_get_stats64 = ifb_stats64,
188 	.ndo_start_xmit	= ifb_xmit,
189 	.ndo_validate_addr = eth_validate_addr,
190 	.ndo_init	= ifb_dev_init,
191 };
192 
193 #define IFB_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG  | NETIF_F_FRAGLIST	| \
194 		      NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL	| \
195 		      NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX		| \
196 		      NETIF_F_HW_VLAN_STAG_TX)
197 
198 static void ifb_dev_free(struct net_device *dev)
199 {
200 	struct ifb_dev_private *dp = netdev_priv(dev);
201 	struct ifb_q_private *txp = dp->tx_private;
202 	int i;
203 
204 	for (i = 0; i < dev->num_tx_queues; i++,txp++) {
205 		tasklet_kill(&txp->ifb_tasklet);
206 		__skb_queue_purge(&txp->rq);
207 		__skb_queue_purge(&txp->tq);
208 	}
209 	kfree(dp->tx_private);
210 }
211 
212 static void ifb_setup(struct net_device *dev)
213 {
214 	/* Initialize the device structure. */
215 	dev->netdev_ops = &ifb_netdev_ops;
216 
217 	/* Fill in device structure with ethernet-generic values. */
218 	ether_setup(dev);
219 	dev->tx_queue_len = TX_Q_LIMIT;
220 
221 	dev->features |= IFB_FEATURES;
222 	dev->hw_features |= dev->features;
223 	dev->hw_enc_features |= dev->features;
224 	dev->vlan_features |= IFB_FEATURES & ~(NETIF_F_HW_VLAN_CTAG_TX |
225 					       NETIF_F_HW_VLAN_STAG_TX);
226 
227 	dev->flags |= IFF_NOARP;
228 	dev->flags &= ~IFF_MULTICAST;
229 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
230 	netif_keep_dst(dev);
231 	eth_hw_addr_random(dev);
232 	dev->needs_free_netdev = true;
233 	dev->priv_destructor = ifb_dev_free;
234 
235 	dev->min_mtu = 0;
236 	dev->max_mtu = 0;
237 }
238 
239 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev)
240 {
241 	struct ifb_dev_private *dp = netdev_priv(dev);
242 	struct ifb_q_private *txp = dp->tx_private + skb_get_queue_mapping(skb);
243 
244 	u64_stats_update_begin(&txp->rsync);
245 	txp->rx_packets++;
246 	txp->rx_bytes += skb->len;
247 	u64_stats_update_end(&txp->rsync);
248 
249 	if (!skb->redirected || !skb->skb_iif) {
250 		dev_kfree_skb(skb);
251 		dev->stats.rx_dropped++;
252 		return NETDEV_TX_OK;
253 	}
254 
255 	if (skb_queue_len(&txp->rq) >= dev->tx_queue_len)
256 		netif_tx_stop_queue(netdev_get_tx_queue(dev, txp->txqnum));
257 
258 	__skb_queue_tail(&txp->rq, skb);
259 	if (!txp->tasklet_pending) {
260 		txp->tasklet_pending = 1;
261 		tasklet_schedule(&txp->ifb_tasklet);
262 	}
263 
264 	return NETDEV_TX_OK;
265 }
266 
267 static int ifb_close(struct net_device *dev)
268 {
269 	netif_tx_stop_all_queues(dev);
270 	return 0;
271 }
272 
273 static int ifb_open(struct net_device *dev)
274 {
275 	netif_tx_start_all_queues(dev);
276 	return 0;
277 }
278 
279 static int ifb_validate(struct nlattr *tb[], struct nlattr *data[],
280 			struct netlink_ext_ack *extack)
281 {
282 	if (tb[IFLA_ADDRESS]) {
283 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
284 			return -EINVAL;
285 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
286 			return -EADDRNOTAVAIL;
287 	}
288 	return 0;
289 }
290 
291 static struct rtnl_link_ops ifb_link_ops __read_mostly = {
292 	.kind		= "ifb",
293 	.priv_size	= sizeof(struct ifb_dev_private),
294 	.setup		= ifb_setup,
295 	.validate	= ifb_validate,
296 };
297 
298 /* Number of ifb devices to be set up by this module.
299  * Note that these legacy devices have one queue.
300  * Prefer something like : ip link add ifb10 numtxqueues 8 type ifb
301  */
302 static int numifbs = 2;
303 module_param(numifbs, int, 0);
304 MODULE_PARM_DESC(numifbs, "Number of ifb devices");
305 
306 static int __init ifb_init_one(int index)
307 {
308 	struct net_device *dev_ifb;
309 	int err;
310 
311 	dev_ifb = alloc_netdev(sizeof(struct ifb_dev_private), "ifb%d",
312 			       NET_NAME_UNKNOWN, ifb_setup);
313 
314 	if (!dev_ifb)
315 		return -ENOMEM;
316 
317 	dev_ifb->rtnl_link_ops = &ifb_link_ops;
318 	err = register_netdevice(dev_ifb);
319 	if (err < 0)
320 		goto err;
321 
322 	return 0;
323 
324 err:
325 	free_netdev(dev_ifb);
326 	return err;
327 }
328 
329 static int __init ifb_init_module(void)
330 {
331 	int i, err;
332 
333 	down_write(&pernet_ops_rwsem);
334 	rtnl_lock();
335 	err = __rtnl_link_register(&ifb_link_ops);
336 	if (err < 0)
337 		goto out;
338 
339 	for (i = 0; i < numifbs && !err; i++) {
340 		err = ifb_init_one(i);
341 		cond_resched();
342 	}
343 	if (err)
344 		__rtnl_link_unregister(&ifb_link_ops);
345 
346 out:
347 	rtnl_unlock();
348 	up_write(&pernet_ops_rwsem);
349 
350 	return err;
351 }
352 
353 static void __exit ifb_cleanup_module(void)
354 {
355 	rtnl_link_unregister(&ifb_link_ops);
356 }
357 
358 module_init(ifb_init_module);
359 module_exit(ifb_cleanup_module);
360 MODULE_LICENSE("GPL");
361 MODULE_AUTHOR("Jamal Hadi Salim");
362 MODULE_ALIAS_RTNL_LINK("ifb");
363