Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
6aa20a22 | 2 | /* drivers/net/ifb.c: |
253af423 JHS |
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 | |
6aa20a22 JG |
12 | dropping. |
13 | ||
253af423 JHS |
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 | ||
6aa20a22 | 21 | |
253af423 | 22 | Authors: Jamal Hadi Salim (2005) |
6aa20a22 | 23 | |
253af423 JHS |
24 | */ |
25 | ||
26 | ||
253af423 JHS |
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> | |
a6b7a407 | 32 | #include <linux/interrupt.h> |
253af423 | 33 | #include <linux/moduleparam.h> |
6aa20a22 | 34 | #include <net/pkt_sched.h> |
881d966b | 35 | #include <net/net_namespace.h> |
253af423 | 36 | |
253af423 | 37 | #define TX_Q_LIMIT 32 |
9e29e21a ED |
38 | struct ifb_q_private { |
39 | struct net_device *dev; | |
253af423 | 40 | struct tasklet_struct ifb_tasklet; |
9e29e21a ED |
41 | int tasklet_pending; |
42 | int txqnum; | |
253af423 | 43 | struct sk_buff_head rq; |
9e29e21a ED |
44 | u64 rx_packets; |
45 | u64 rx_bytes; | |
46 | struct u64_stats_sync rsync; | |
3b0c9cbb | 47 | |
48 | struct u64_stats_sync tsync; | |
9e29e21a ED |
49 | u64 tx_packets; |
50 | u64 tx_bytes; | |
253af423 | 51 | struct sk_buff_head tq; |
9e29e21a | 52 | } ____cacheline_aligned_in_smp; |
253af423 | 53 | |
9e29e21a ED |
54 | struct ifb_dev_private { |
55 | struct ifb_q_private *tx_private; | |
56 | }; | |
253af423 | 57 | |
424efe9c | 58 | static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev); |
253af423 JHS |
59 | static int ifb_open(struct net_device *dev); |
60 | static int ifb_close(struct net_device *dev); | |
61 | ||
9e29e21a | 62 | static void ifb_ri_tasklet(unsigned long _txp) |
253af423 | 63 | { |
9e29e21a | 64 | struct ifb_q_private *txp = (struct ifb_q_private *)_txp; |
c3f26a26 | 65 | struct netdev_queue *txq; |
253af423 JHS |
66 | struct sk_buff *skb; |
67 | ||
9e29e21a ED |
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)) | |
253af423 | 72 | goto resched; |
9e29e21a ED |
73 | skb_queue_splice_tail_init(&txp->rq, &txp->tq); |
74 | __netif_tx_unlock(txq); | |
253af423 JHS |
75 | } |
76 | ||
9e29e21a | 77 | while ((skb = __skb_dequeue(&txp->tq)) != NULL) { |
bc31c905 | 78 | skb->tc_redirected = 0; |
e7246e12 | 79 | skb->tc_skip_classify = 1; |
3b0c9cbb | 80 | |
9e29e21a ED |
81 | u64_stats_update_begin(&txp->tsync); |
82 | txp->tx_packets++; | |
83 | txp->tx_bytes += skb->len; | |
84 | u64_stats_update_end(&txp->tsync); | |
c01003c2 | 85 | |
05e8689c | 86 | rcu_read_lock(); |
9e29e21a | 87 | skb->dev = dev_get_by_index_rcu(dev_net(txp->dev), skb->skb_iif); |
c01003c2 | 88 | if (!skb->dev) { |
05e8689c | 89 | rcu_read_unlock(); |
c01003c2 | 90 | dev_kfree_skb(skb); |
9e29e21a ED |
91 | txp->dev->stats.tx_dropped++; |
92 | if (skb_queue_len(&txp->tq) != 0) | |
75c1c825 | 93 | goto resched; |
c01003c2 PM |
94 | break; |
95 | } | |
05e8689c | 96 | rcu_read_unlock(); |
9e29e21a | 97 | skb->skb_iif = txp->dev->ifindex; |
c01003c2 | 98 | |
bc31c905 | 99 | if (!skb->tc_from_ingress) { |
253af423 | 100 | dev_queue_xmit(skb); |
bc31c905 | 101 | } else { |
b1d2e4e0 | 102 | skb_pull_rcsum(skb, skb->mac_len); |
1a75972c | 103 | netif_receive_skb(skb); |
bc31c905 | 104 | } |
253af423 JHS |
105 | } |
106 | ||
c3f26a26 | 107 | if (__netif_tx_trylock(txq)) { |
9e29e21a ED |
108 | skb = skb_peek(&txp->rq); |
109 | if (!skb) { | |
110 | txp->tasklet_pending = 0; | |
111 | if (netif_tx_queue_stopped(txq)) | |
112 | netif_tx_wake_queue(txq); | |
253af423 | 113 | } else { |
c3f26a26 | 114 | __netif_tx_unlock(txq); |
253af423 JHS |
115 | goto resched; |
116 | } | |
c3f26a26 | 117 | __netif_tx_unlock(txq); |
253af423 JHS |
118 | } else { |
119 | resched: | |
9e29e21a ED |
120 | txp->tasklet_pending = 1; |
121 | tasklet_schedule(&txp->ifb_tasklet); | |
253af423 JHS |
122 | } |
123 | ||
124 | } | |
125 | ||
bc1f4470 | 126 | static void ifb_stats64(struct net_device *dev, |
127 | struct rtnl_link_stats64 *stats) | |
3b0c9cbb | 128 | { |
9e29e21a ED |
129 | struct ifb_dev_private *dp = netdev_priv(dev); |
130 | struct ifb_q_private *txp = dp->tx_private; | |
3b0c9cbb | 131 | unsigned int start; |
9e29e21a ED |
132 | u64 packets, bytes; |
133 | int i; | |
134 | ||
135 | for (i = 0; i < dev->num_tx_queues; i++,txp++) { | |
136 | do { | |
137 | start = u64_stats_fetch_begin_irq(&txp->rsync); | |
138 | packets = txp->rx_packets; | |
139 | bytes = txp->rx_bytes; | |
140 | } while (u64_stats_fetch_retry_irq(&txp->rsync, start)); | |
141 | stats->rx_packets += packets; | |
142 | stats->rx_bytes += bytes; | |
143 | ||
144 | do { | |
145 | start = u64_stats_fetch_begin_irq(&txp->tsync); | |
146 | packets = txp->tx_packets; | |
147 | bytes = txp->tx_bytes; | |
148 | } while (u64_stats_fetch_retry_irq(&txp->tsync, start)); | |
149 | stats->tx_packets += packets; | |
150 | stats->tx_bytes += bytes; | |
151 | } | |
3b0c9cbb | 152 | stats->rx_dropped = dev->stats.rx_dropped; |
153 | stats->tx_dropped = dev->stats.tx_dropped; | |
3b0c9cbb | 154 | } |
155 | ||
9e29e21a ED |
156 | static int ifb_dev_init(struct net_device *dev) |
157 | { | |
158 | struct ifb_dev_private *dp = netdev_priv(dev); | |
159 | struct ifb_q_private *txp; | |
160 | int i; | |
161 | ||
162 | txp = kcalloc(dev->num_tx_queues, sizeof(*txp), GFP_KERNEL); | |
163 | if (!txp) | |
164 | return -ENOMEM; | |
165 | dp->tx_private = txp; | |
166 | for (i = 0; i < dev->num_tx_queues; i++,txp++) { | |
167 | txp->txqnum = i; | |
168 | txp->dev = dev; | |
169 | __skb_queue_head_init(&txp->rq); | |
170 | __skb_queue_head_init(&txp->tq); | |
171 | u64_stats_init(&txp->rsync); | |
172 | u64_stats_init(&txp->tsync); | |
173 | tasklet_init(&txp->ifb_tasklet, ifb_ri_tasklet, | |
174 | (unsigned long)txp); | |
175 | netif_tx_start_queue(netdev_get_tx_queue(dev, i)); | |
176 | } | |
177 | return 0; | |
178 | } | |
3b0c9cbb | 179 | |
8dfcdf34 | 180 | static const struct net_device_ops ifb_netdev_ops = { |
8dfcdf34 SH |
181 | .ndo_open = ifb_open, |
182 | .ndo_stop = ifb_close, | |
3b0c9cbb | 183 | .ndo_get_stats64 = ifb_stats64, |
00829823 SH |
184 | .ndo_start_xmit = ifb_xmit, |
185 | .ndo_validate_addr = eth_validate_addr, | |
9e29e21a | 186 | .ndo_init = ifb_dev_init, |
8dfcdf34 SH |
187 | }; |
188 | ||
34324dc2 | 189 | #define IFB_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | NETIF_F_FRAGLIST | \ |
39980292 | 190 | NETIF_F_TSO_ECN | NETIF_F_TSO | NETIF_F_TSO6 | \ |
7d945796 | 191 | NETIF_F_GSO_ENCAP_ALL | \ |
28d2b136 PM |
192 | NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX | \ |
193 | NETIF_F_HW_VLAN_STAG_TX) | |
39980292 | 194 | |
9e29e21a ED |
195 | static void ifb_dev_free(struct net_device *dev) |
196 | { | |
197 | struct ifb_dev_private *dp = netdev_priv(dev); | |
198 | struct ifb_q_private *txp = dp->tx_private; | |
199 | int i; | |
200 | ||
201 | for (i = 0; i < dev->num_tx_queues; i++,txp++) { | |
202 | tasklet_kill(&txp->ifb_tasklet); | |
203 | __skb_queue_purge(&txp->rq); | |
204 | __skb_queue_purge(&txp->tq); | |
205 | } | |
206 | kfree(dp->tx_private); | |
9e29e21a ED |
207 | } |
208 | ||
9ba2cd65 | 209 | static void ifb_setup(struct net_device *dev) |
253af423 JHS |
210 | { |
211 | /* Initialize the device structure. */ | |
8dfcdf34 | 212 | dev->netdev_ops = &ifb_netdev_ops; |
253af423 JHS |
213 | |
214 | /* Fill in device structure with ethernet-generic values. */ | |
215 | ether_setup(dev); | |
216 | dev->tx_queue_len = TX_Q_LIMIT; | |
8dfcdf34 | 217 | |
39980292 | 218 | dev->features |= IFB_FEATURES; |
7d945796 ED |
219 | dev->hw_features |= dev->features; |
220 | dev->hw_enc_features |= dev->features; | |
8dd6e147 VY |
221 | dev->vlan_features |= IFB_FEATURES & ~(NETIF_F_HW_VLAN_CTAG_TX | |
222 | NETIF_F_HW_VLAN_STAG_TX); | |
39980292 | 223 | |
253af423 JHS |
224 | dev->flags |= IFF_NOARP; |
225 | dev->flags &= ~IFF_MULTICAST; | |
02875878 ED |
226 | dev->priv_flags &= ~IFF_TX_SKB_SHARING; |
227 | netif_keep_dst(dev); | |
f2cedb63 | 228 | eth_hw_addr_random(dev); |
cf124db5 DM |
229 | dev->needs_free_netdev = true; |
230 | dev->priv_destructor = ifb_dev_free; | |
e94cd811 ZS |
231 | |
232 | dev->min_mtu = 0; | |
233 | dev->max_mtu = 0; | |
253af423 JHS |
234 | } |
235 | ||
424efe9c | 236 | static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev) |
253af423 | 237 | { |
9e29e21a | 238 | struct ifb_dev_private *dp = netdev_priv(dev); |
9e29e21a | 239 | struct ifb_q_private *txp = dp->tx_private + skb_get_queue_mapping(skb); |
253af423 | 240 | |
9e29e21a ED |
241 | u64_stats_update_begin(&txp->rsync); |
242 | txp->rx_packets++; | |
243 | txp->rx_bytes += skb->len; | |
244 | u64_stats_update_end(&txp->rsync); | |
253af423 | 245 | |
bc31c905 | 246 | if (!skb->tc_redirected || !skb->skb_iif) { |
253af423 | 247 | dev_kfree_skb(skb); |
3b0c9cbb | 248 | dev->stats.rx_dropped++; |
424efe9c | 249 | return NETDEV_TX_OK; |
253af423 JHS |
250 | } |
251 | ||
9e29e21a ED |
252 | if (skb_queue_len(&txp->rq) >= dev->tx_queue_len) |
253 | netif_tx_stop_queue(netdev_get_tx_queue(dev, txp->txqnum)); | |
253af423 | 254 | |
9e29e21a ED |
255 | __skb_queue_tail(&txp->rq, skb); |
256 | if (!txp->tasklet_pending) { | |
257 | txp->tasklet_pending = 1; | |
258 | tasklet_schedule(&txp->ifb_tasklet); | |
253af423 JHS |
259 | } |
260 | ||
424efe9c | 261 | return NETDEV_TX_OK; |
253af423 JHS |
262 | } |
263 | ||
253af423 JHS |
264 | static int ifb_close(struct net_device *dev) |
265 | { | |
9e29e21a | 266 | netif_tx_stop_all_queues(dev); |
253af423 JHS |
267 | return 0; |
268 | } | |
269 | ||
270 | static int ifb_open(struct net_device *dev) | |
271 | { | |
9e29e21a | 272 | netif_tx_start_all_queues(dev); |
253af423 JHS |
273 | return 0; |
274 | } | |
275 | ||
a8b8a889 MS |
276 | static int ifb_validate(struct nlattr *tb[], struct nlattr *data[], |
277 | struct netlink_ext_ack *extack) | |
0e06877c PM |
278 | { |
279 | if (tb[IFLA_ADDRESS]) { | |
280 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) | |
281 | return -EINVAL; | |
282 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) | |
283 | return -EADDRNOTAVAIL; | |
284 | } | |
285 | return 0; | |
286 | } | |
287 | ||
9ba2cd65 PM |
288 | static struct rtnl_link_ops ifb_link_ops __read_mostly = { |
289 | .kind = "ifb", | |
9e29e21a | 290 | .priv_size = sizeof(struct ifb_dev_private), |
9ba2cd65 | 291 | .setup = ifb_setup, |
0e06877c | 292 | .validate = ifb_validate, |
9ba2cd65 PM |
293 | }; |
294 | ||
9e29e21a ED |
295 | /* Number of ifb devices to be set up by this module. |
296 | * Note that these legacy devices have one queue. | |
297 | * Prefer something like : ip link add ifb10 numtxqueues 8 type ifb | |
298 | */ | |
299 | static int numifbs = 2; | |
2d85cba2 PM |
300 | module_param(numifbs, int, 0); |
301 | MODULE_PARM_DESC(numifbs, "Number of ifb devices"); | |
302 | ||
253af423 JHS |
303 | static int __init ifb_init_one(int index) |
304 | { | |
305 | struct net_device *dev_ifb; | |
306 | int err; | |
307 | ||
9e29e21a | 308 | dev_ifb = alloc_netdev(sizeof(struct ifb_dev_private), "ifb%d", |
c835a677 | 309 | NET_NAME_UNKNOWN, ifb_setup); |
253af423 JHS |
310 | |
311 | if (!dev_ifb) | |
312 | return -ENOMEM; | |
313 | ||
9ba2cd65 PM |
314 | dev_ifb->rtnl_link_ops = &ifb_link_ops; |
315 | err = register_netdevice(dev_ifb); | |
316 | if (err < 0) | |
317 | goto err; | |
94833dfb | 318 | |
9ba2cd65 | 319 | return 0; |
62b7ffca | 320 | |
9ba2cd65 PM |
321 | err: |
322 | free_netdev(dev_ifb); | |
323 | return err; | |
6aa20a22 | 324 | } |
253af423 JHS |
325 | |
326 | static int __init ifb_init_module(void) | |
6aa20a22 | 327 | { |
9ba2cd65 PM |
328 | int i, err; |
329 | ||
554873e5 | 330 | down_write(&pernet_ops_rwsem); |
9ba2cd65 PM |
331 | rtnl_lock(); |
332 | err = __rtnl_link_register(&ifb_link_ops); | |
f2966cd5 | 333 | if (err < 0) |
334 | goto out; | |
62b7ffca | 335 | |
440d57bc | 336 | for (i = 0; i < numifbs && !err; i++) { |
6aa20a22 | 337 | err = ifb_init_one(i); |
440d57bc | 338 | cond_resched(); |
339 | } | |
2d85cba2 | 340 | if (err) |
9ba2cd65 | 341 | __rtnl_link_unregister(&ifb_link_ops); |
f2966cd5 | 342 | |
343 | out: | |
9ba2cd65 | 344 | rtnl_unlock(); |
554873e5 | 345 | up_write(&pernet_ops_rwsem); |
253af423 JHS |
346 | |
347 | return err; | |
6aa20a22 | 348 | } |
253af423 JHS |
349 | |
350 | static void __exit ifb_cleanup_module(void) | |
351 | { | |
2d85cba2 | 352 | rtnl_link_unregister(&ifb_link_ops); |
253af423 JHS |
353 | } |
354 | ||
355 | module_init(ifb_init_module); | |
356 | module_exit(ifb_cleanup_module); | |
357 | MODULE_LICENSE("GPL"); | |
358 | MODULE_AUTHOR("Jamal Hadi Salim"); | |
9ba2cd65 | 359 | MODULE_ALIAS_RTNL_LINK("ifb"); |