xref: /linux/drivers/net/netkit.c (revision e6a901a00822659181c93c86d8bbc2a17779fddc)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2023 Isovalent */
3 
4 #include <linux/netdevice.h>
5 #include <linux/ethtool.h>
6 #include <linux/etherdevice.h>
7 #include <linux/filter.h>
8 #include <linux/netfilter_netdev.h>
9 #include <linux/bpf_mprog.h>
10 #include <linux/indirect_call_wrapper.h>
11 
12 #include <net/netkit.h>
13 #include <net/dst.h>
14 #include <net/tcx.h>
15 
16 #define DRV_NAME "netkit"
17 
18 struct netkit {
19 	/* Needed in fast-path */
20 	struct net_device __rcu *peer;
21 	struct bpf_mprog_entry __rcu *active;
22 	enum netkit_action policy;
23 	struct bpf_mprog_bundle	bundle;
24 
25 	/* Needed in slow-path */
26 	enum netkit_mode mode;
27 	bool primary;
28 	u32 headroom;
29 };
30 
31 struct netkit_link {
32 	struct bpf_link link;
33 	struct net_device *dev;
34 	u32 location;
35 };
36 
37 static __always_inline int
38 netkit_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb,
39 	   enum netkit_action ret)
40 {
41 	const struct bpf_mprog_fp *fp;
42 	const struct bpf_prog *prog;
43 
44 	bpf_mprog_foreach_prog(entry, fp, prog) {
45 		bpf_compute_data_pointers(skb);
46 		ret = bpf_prog_run(prog, skb);
47 		if (ret != NETKIT_NEXT)
48 			break;
49 	}
50 	return ret;
51 }
52 
53 static void netkit_prep_forward(struct sk_buff *skb, bool xnet)
54 {
55 	skb_scrub_packet(skb, xnet);
56 	skb->priority = 0;
57 	nf_skip_egress(skb, true);
58 }
59 
60 static struct netkit *netkit_priv(const struct net_device *dev)
61 {
62 	return netdev_priv(dev);
63 }
64 
65 static netdev_tx_t netkit_xmit(struct sk_buff *skb, struct net_device *dev)
66 {
67 	struct netkit *nk = netkit_priv(dev);
68 	enum netkit_action ret = READ_ONCE(nk->policy);
69 	netdev_tx_t ret_dev = NET_XMIT_SUCCESS;
70 	const struct bpf_mprog_entry *entry;
71 	struct net_device *peer;
72 	int len = skb->len;
73 
74 	rcu_read_lock();
75 	peer = rcu_dereference(nk->peer);
76 	if (unlikely(!peer || !(peer->flags & IFF_UP) ||
77 		     !pskb_may_pull(skb, ETH_HLEN) ||
78 		     skb_orphan_frags(skb, GFP_ATOMIC)))
79 		goto drop;
80 	netkit_prep_forward(skb, !net_eq(dev_net(dev), dev_net(peer)));
81 	skb->dev = peer;
82 	entry = rcu_dereference(nk->active);
83 	if (entry)
84 		ret = netkit_run(entry, skb, ret);
85 	switch (ret) {
86 	case NETKIT_NEXT:
87 	case NETKIT_PASS:
88 		skb->protocol = eth_type_trans(skb, skb->dev);
89 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
90 		if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) {
91 			dev_sw_netstats_tx_add(dev, 1, len);
92 			dev_sw_netstats_rx_add(peer, len);
93 		} else {
94 			goto drop_stats;
95 		}
96 		break;
97 	case NETKIT_REDIRECT:
98 		dev_sw_netstats_tx_add(dev, 1, len);
99 		skb_do_redirect(skb);
100 		break;
101 	case NETKIT_DROP:
102 	default:
103 drop:
104 		kfree_skb(skb);
105 drop_stats:
106 		dev_core_stats_tx_dropped_inc(dev);
107 		ret_dev = NET_XMIT_DROP;
108 		break;
109 	}
110 	rcu_read_unlock();
111 	return ret_dev;
112 }
113 
114 static int netkit_open(struct net_device *dev)
115 {
116 	struct netkit *nk = netkit_priv(dev);
117 	struct net_device *peer = rtnl_dereference(nk->peer);
118 
119 	if (!peer)
120 		return -ENOTCONN;
121 	if (peer->flags & IFF_UP) {
122 		netif_carrier_on(dev);
123 		netif_carrier_on(peer);
124 	}
125 	return 0;
126 }
127 
128 static int netkit_close(struct net_device *dev)
129 {
130 	struct netkit *nk = netkit_priv(dev);
131 	struct net_device *peer = rtnl_dereference(nk->peer);
132 
133 	netif_carrier_off(dev);
134 	if (peer)
135 		netif_carrier_off(peer);
136 	return 0;
137 }
138 
139 static int netkit_get_iflink(const struct net_device *dev)
140 {
141 	struct netkit *nk = netkit_priv(dev);
142 	struct net_device *peer;
143 	int iflink = 0;
144 
145 	rcu_read_lock();
146 	peer = rcu_dereference(nk->peer);
147 	if (peer)
148 		iflink = READ_ONCE(peer->ifindex);
149 	rcu_read_unlock();
150 	return iflink;
151 }
152 
153 static void netkit_set_multicast(struct net_device *dev)
154 {
155 	/* Nothing to do, we receive whatever gets pushed to us! */
156 }
157 
158 static void netkit_set_headroom(struct net_device *dev, int headroom)
159 {
160 	struct netkit *nk = netkit_priv(dev), *nk2;
161 	struct net_device *peer;
162 
163 	if (headroom < 0)
164 		headroom = NET_SKB_PAD;
165 
166 	rcu_read_lock();
167 	peer = rcu_dereference(nk->peer);
168 	if (unlikely(!peer))
169 		goto out;
170 
171 	nk2 = netkit_priv(peer);
172 	nk->headroom = headroom;
173 	headroom = max(nk->headroom, nk2->headroom);
174 
175 	peer->needed_headroom = headroom;
176 	dev->needed_headroom = headroom;
177 out:
178 	rcu_read_unlock();
179 }
180 
181 INDIRECT_CALLABLE_SCOPE struct net_device *netkit_peer_dev(struct net_device *dev)
182 {
183 	return rcu_dereference(netkit_priv(dev)->peer);
184 }
185 
186 static void netkit_get_stats(struct net_device *dev,
187 			     struct rtnl_link_stats64 *stats)
188 {
189 	dev_fetch_sw_netstats(stats, dev->tstats);
190 	stats->tx_dropped = DEV_STATS_READ(dev, tx_dropped);
191 }
192 
193 static void netkit_uninit(struct net_device *dev);
194 
195 static const struct net_device_ops netkit_netdev_ops = {
196 	.ndo_open		= netkit_open,
197 	.ndo_stop		= netkit_close,
198 	.ndo_start_xmit		= netkit_xmit,
199 	.ndo_set_rx_mode	= netkit_set_multicast,
200 	.ndo_set_rx_headroom	= netkit_set_headroom,
201 	.ndo_get_iflink		= netkit_get_iflink,
202 	.ndo_get_peer_dev	= netkit_peer_dev,
203 	.ndo_get_stats64	= netkit_get_stats,
204 	.ndo_uninit		= netkit_uninit,
205 	.ndo_features_check	= passthru_features_check,
206 };
207 
208 static void netkit_get_drvinfo(struct net_device *dev,
209 			       struct ethtool_drvinfo *info)
210 {
211 	strscpy(info->driver, DRV_NAME, sizeof(info->driver));
212 }
213 
214 static const struct ethtool_ops netkit_ethtool_ops = {
215 	.get_drvinfo		= netkit_get_drvinfo,
216 };
217 
218 static void netkit_setup(struct net_device *dev)
219 {
220 	static const netdev_features_t netkit_features_hw_vlan =
221 		NETIF_F_HW_VLAN_CTAG_TX |
222 		NETIF_F_HW_VLAN_CTAG_RX |
223 		NETIF_F_HW_VLAN_STAG_TX |
224 		NETIF_F_HW_VLAN_STAG_RX;
225 	static const netdev_features_t netkit_features =
226 		netkit_features_hw_vlan |
227 		NETIF_F_SG |
228 		NETIF_F_FRAGLIST |
229 		NETIF_F_HW_CSUM |
230 		NETIF_F_RXCSUM |
231 		NETIF_F_SCTP_CRC |
232 		NETIF_F_HIGHDMA |
233 		NETIF_F_GSO_SOFTWARE |
234 		NETIF_F_GSO_ENCAP_ALL;
235 
236 	ether_setup(dev);
237 	dev->max_mtu = ETH_MAX_MTU;
238 	dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
239 
240 	dev->flags |= IFF_NOARP;
241 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
242 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
243 	dev->priv_flags |= IFF_PHONY_HEADROOM;
244 	dev->priv_flags |= IFF_NO_QUEUE;
245 
246 	dev->ethtool_ops = &netkit_ethtool_ops;
247 	dev->netdev_ops  = &netkit_netdev_ops;
248 
249 	dev->features |= netkit_features | NETIF_F_LLTX;
250 	dev->hw_features = netkit_features;
251 	dev->hw_enc_features = netkit_features;
252 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
253 	dev->vlan_features = dev->features & ~netkit_features_hw_vlan;
254 
255 	dev->needs_free_netdev = true;
256 
257 	netif_set_tso_max_size(dev, GSO_MAX_SIZE);
258 }
259 
260 static struct net *netkit_get_link_net(const struct net_device *dev)
261 {
262 	struct netkit *nk = netkit_priv(dev);
263 	struct net_device *peer = rtnl_dereference(nk->peer);
264 
265 	return peer ? dev_net(peer) : dev_net(dev);
266 }
267 
268 static int netkit_check_policy(int policy, struct nlattr *tb,
269 			       struct netlink_ext_ack *extack)
270 {
271 	switch (policy) {
272 	case NETKIT_PASS:
273 	case NETKIT_DROP:
274 		return 0;
275 	default:
276 		NL_SET_ERR_MSG_ATTR(extack, tb,
277 				    "Provided default xmit policy not supported");
278 		return -EINVAL;
279 	}
280 }
281 
282 static int netkit_check_mode(int mode, struct nlattr *tb,
283 			     struct netlink_ext_ack *extack)
284 {
285 	switch (mode) {
286 	case NETKIT_L2:
287 	case NETKIT_L3:
288 		return 0;
289 	default:
290 		NL_SET_ERR_MSG_ATTR(extack, tb,
291 				    "Provided device mode can only be L2 or L3");
292 		return -EINVAL;
293 	}
294 }
295 
296 static int netkit_validate(struct nlattr *tb[], struct nlattr *data[],
297 			   struct netlink_ext_ack *extack)
298 {
299 	struct nlattr *attr = tb[IFLA_ADDRESS];
300 
301 	if (!attr)
302 		return 0;
303 	NL_SET_ERR_MSG_ATTR(extack, attr,
304 			    "Setting Ethernet address is not supported");
305 	return -EOPNOTSUPP;
306 }
307 
308 static struct rtnl_link_ops netkit_link_ops;
309 
310 static int netkit_new_link(struct net *src_net, struct net_device *dev,
311 			   struct nlattr *tb[], struct nlattr *data[],
312 			   struct netlink_ext_ack *extack)
313 {
314 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp = tb, *attr;
315 	enum netkit_action default_prim = NETKIT_PASS;
316 	enum netkit_action default_peer = NETKIT_PASS;
317 	enum netkit_mode mode = NETKIT_L3;
318 	unsigned char ifname_assign_type;
319 	struct ifinfomsg *ifmp = NULL;
320 	struct net_device *peer;
321 	char ifname[IFNAMSIZ];
322 	struct netkit *nk;
323 	struct net *net;
324 	int err;
325 
326 	if (data) {
327 		if (data[IFLA_NETKIT_MODE]) {
328 			attr = data[IFLA_NETKIT_MODE];
329 			mode = nla_get_u32(attr);
330 			err = netkit_check_mode(mode, attr, extack);
331 			if (err < 0)
332 				return err;
333 		}
334 		if (data[IFLA_NETKIT_PEER_INFO]) {
335 			attr = data[IFLA_NETKIT_PEER_INFO];
336 			ifmp = nla_data(attr);
337 			err = rtnl_nla_parse_ifinfomsg(peer_tb, attr, extack);
338 			if (err < 0)
339 				return err;
340 			err = netkit_validate(peer_tb, NULL, extack);
341 			if (err < 0)
342 				return err;
343 			tbp = peer_tb;
344 		}
345 		if (data[IFLA_NETKIT_POLICY]) {
346 			attr = data[IFLA_NETKIT_POLICY];
347 			default_prim = nla_get_u32(attr);
348 			err = netkit_check_policy(default_prim, attr, extack);
349 			if (err < 0)
350 				return err;
351 		}
352 		if (data[IFLA_NETKIT_PEER_POLICY]) {
353 			attr = data[IFLA_NETKIT_PEER_POLICY];
354 			default_peer = nla_get_u32(attr);
355 			err = netkit_check_policy(default_peer, attr, extack);
356 			if (err < 0)
357 				return err;
358 		}
359 	}
360 
361 	if (ifmp && tbp[IFLA_IFNAME]) {
362 		nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
363 		ifname_assign_type = NET_NAME_USER;
364 	} else {
365 		strscpy(ifname, "nk%d", IFNAMSIZ);
366 		ifname_assign_type = NET_NAME_ENUM;
367 	}
368 
369 	net = rtnl_link_get_net(src_net, tbp);
370 	if (IS_ERR(net))
371 		return PTR_ERR(net);
372 
373 	peer = rtnl_create_link(net, ifname, ifname_assign_type,
374 				&netkit_link_ops, tbp, extack);
375 	if (IS_ERR(peer)) {
376 		put_net(net);
377 		return PTR_ERR(peer);
378 	}
379 
380 	netif_inherit_tso_max(peer, dev);
381 
382 	if (mode == NETKIT_L2)
383 		eth_hw_addr_random(peer);
384 	if (ifmp && dev->ifindex)
385 		peer->ifindex = ifmp->ifi_index;
386 
387 	nk = netkit_priv(peer);
388 	nk->primary = false;
389 	nk->policy = default_peer;
390 	nk->mode = mode;
391 	bpf_mprog_bundle_init(&nk->bundle);
392 
393 	err = register_netdevice(peer);
394 	put_net(net);
395 	if (err < 0)
396 		goto err_register_peer;
397 	netif_carrier_off(peer);
398 	if (mode == NETKIT_L2)
399 		dev_change_flags(peer, peer->flags & ~IFF_NOARP, NULL);
400 
401 	err = rtnl_configure_link(peer, NULL, 0, NULL);
402 	if (err < 0)
403 		goto err_configure_peer;
404 
405 	if (mode == NETKIT_L2)
406 		eth_hw_addr_random(dev);
407 	if (tb[IFLA_IFNAME])
408 		nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
409 	else
410 		strscpy(dev->name, "nk%d", IFNAMSIZ);
411 
412 	nk = netkit_priv(dev);
413 	nk->primary = true;
414 	nk->policy = default_prim;
415 	nk->mode = mode;
416 	bpf_mprog_bundle_init(&nk->bundle);
417 
418 	err = register_netdevice(dev);
419 	if (err < 0)
420 		goto err_configure_peer;
421 	netif_carrier_off(dev);
422 	if (mode == NETKIT_L2)
423 		dev_change_flags(dev, dev->flags & ~IFF_NOARP, NULL);
424 
425 	rcu_assign_pointer(netkit_priv(dev)->peer, peer);
426 	rcu_assign_pointer(netkit_priv(peer)->peer, dev);
427 	return 0;
428 err_configure_peer:
429 	unregister_netdevice(peer);
430 	return err;
431 err_register_peer:
432 	free_netdev(peer);
433 	return err;
434 }
435 
436 static struct bpf_mprog_entry *netkit_entry_fetch(struct net_device *dev,
437 						  bool bundle_fallback)
438 {
439 	struct netkit *nk = netkit_priv(dev);
440 	struct bpf_mprog_entry *entry;
441 
442 	ASSERT_RTNL();
443 	entry = rcu_dereference_rtnl(nk->active);
444 	if (entry)
445 		return entry;
446 	if (bundle_fallback)
447 		return &nk->bundle.a;
448 	return NULL;
449 }
450 
451 static void netkit_entry_update(struct net_device *dev,
452 				struct bpf_mprog_entry *entry)
453 {
454 	struct netkit *nk = netkit_priv(dev);
455 
456 	ASSERT_RTNL();
457 	rcu_assign_pointer(nk->active, entry);
458 }
459 
460 static void netkit_entry_sync(void)
461 {
462 	synchronize_rcu();
463 }
464 
465 static struct net_device *netkit_dev_fetch(struct net *net, u32 ifindex, u32 which)
466 {
467 	struct net_device *dev;
468 	struct netkit *nk;
469 
470 	ASSERT_RTNL();
471 
472 	switch (which) {
473 	case BPF_NETKIT_PRIMARY:
474 	case BPF_NETKIT_PEER:
475 		break;
476 	default:
477 		return ERR_PTR(-EINVAL);
478 	}
479 
480 	dev = __dev_get_by_index(net, ifindex);
481 	if (!dev)
482 		return ERR_PTR(-ENODEV);
483 	if (dev->netdev_ops != &netkit_netdev_ops)
484 		return ERR_PTR(-ENXIO);
485 
486 	nk = netkit_priv(dev);
487 	if (!nk->primary)
488 		return ERR_PTR(-EACCES);
489 	if (which == BPF_NETKIT_PEER) {
490 		dev = rcu_dereference_rtnl(nk->peer);
491 		if (!dev)
492 			return ERR_PTR(-ENODEV);
493 	}
494 	return dev;
495 }
496 
497 int netkit_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)
498 {
499 	struct bpf_mprog_entry *entry, *entry_new;
500 	struct bpf_prog *replace_prog = NULL;
501 	struct net_device *dev;
502 	int ret;
503 
504 	rtnl_lock();
505 	dev = netkit_dev_fetch(current->nsproxy->net_ns, attr->target_ifindex,
506 			       attr->attach_type);
507 	if (IS_ERR(dev)) {
508 		ret = PTR_ERR(dev);
509 		goto out;
510 	}
511 	entry = netkit_entry_fetch(dev, true);
512 	if (attr->attach_flags & BPF_F_REPLACE) {
513 		replace_prog = bpf_prog_get_type(attr->replace_bpf_fd,
514 						 prog->type);
515 		if (IS_ERR(replace_prog)) {
516 			ret = PTR_ERR(replace_prog);
517 			replace_prog = NULL;
518 			goto out;
519 		}
520 	}
521 	ret = bpf_mprog_attach(entry, &entry_new, prog, NULL, replace_prog,
522 			       attr->attach_flags, attr->relative_fd,
523 			       attr->expected_revision);
524 	if (!ret) {
525 		if (entry != entry_new) {
526 			netkit_entry_update(dev, entry_new);
527 			netkit_entry_sync();
528 		}
529 		bpf_mprog_commit(entry);
530 	}
531 out:
532 	if (replace_prog)
533 		bpf_prog_put(replace_prog);
534 	rtnl_unlock();
535 	return ret;
536 }
537 
538 int netkit_prog_detach(const union bpf_attr *attr, struct bpf_prog *prog)
539 {
540 	struct bpf_mprog_entry *entry, *entry_new;
541 	struct net_device *dev;
542 	int ret;
543 
544 	rtnl_lock();
545 	dev = netkit_dev_fetch(current->nsproxy->net_ns, attr->target_ifindex,
546 			       attr->attach_type);
547 	if (IS_ERR(dev)) {
548 		ret = PTR_ERR(dev);
549 		goto out;
550 	}
551 	entry = netkit_entry_fetch(dev, false);
552 	if (!entry) {
553 		ret = -ENOENT;
554 		goto out;
555 	}
556 	ret = bpf_mprog_detach(entry, &entry_new, prog, NULL, attr->attach_flags,
557 			       attr->relative_fd, attr->expected_revision);
558 	if (!ret) {
559 		if (!bpf_mprog_total(entry_new))
560 			entry_new = NULL;
561 		netkit_entry_update(dev, entry_new);
562 		netkit_entry_sync();
563 		bpf_mprog_commit(entry);
564 	}
565 out:
566 	rtnl_unlock();
567 	return ret;
568 }
569 
570 int netkit_prog_query(const union bpf_attr *attr, union bpf_attr __user *uattr)
571 {
572 	struct net_device *dev;
573 	int ret;
574 
575 	rtnl_lock();
576 	dev = netkit_dev_fetch(current->nsproxy->net_ns,
577 			       attr->query.target_ifindex,
578 			       attr->query.attach_type);
579 	if (IS_ERR(dev)) {
580 		ret = PTR_ERR(dev);
581 		goto out;
582 	}
583 	ret = bpf_mprog_query(attr, uattr, netkit_entry_fetch(dev, false));
584 out:
585 	rtnl_unlock();
586 	return ret;
587 }
588 
589 static struct netkit_link *netkit_link(const struct bpf_link *link)
590 {
591 	return container_of(link, struct netkit_link, link);
592 }
593 
594 static int netkit_link_prog_attach(struct bpf_link *link, u32 flags,
595 				   u32 id_or_fd, u64 revision)
596 {
597 	struct netkit_link *nkl = netkit_link(link);
598 	struct bpf_mprog_entry *entry, *entry_new;
599 	struct net_device *dev = nkl->dev;
600 	int ret;
601 
602 	ASSERT_RTNL();
603 	entry = netkit_entry_fetch(dev, true);
604 	ret = bpf_mprog_attach(entry, &entry_new, link->prog, link, NULL, flags,
605 			       id_or_fd, revision);
606 	if (!ret) {
607 		if (entry != entry_new) {
608 			netkit_entry_update(dev, entry_new);
609 			netkit_entry_sync();
610 		}
611 		bpf_mprog_commit(entry);
612 	}
613 	return ret;
614 }
615 
616 static void netkit_link_release(struct bpf_link *link)
617 {
618 	struct netkit_link *nkl = netkit_link(link);
619 	struct bpf_mprog_entry *entry, *entry_new;
620 	struct net_device *dev;
621 	int ret = 0;
622 
623 	rtnl_lock();
624 	dev = nkl->dev;
625 	if (!dev)
626 		goto out;
627 	entry = netkit_entry_fetch(dev, false);
628 	if (!entry) {
629 		ret = -ENOENT;
630 		goto out;
631 	}
632 	ret = bpf_mprog_detach(entry, &entry_new, link->prog, link, 0, 0, 0);
633 	if (!ret) {
634 		if (!bpf_mprog_total(entry_new))
635 			entry_new = NULL;
636 		netkit_entry_update(dev, entry_new);
637 		netkit_entry_sync();
638 		bpf_mprog_commit(entry);
639 		nkl->dev = NULL;
640 	}
641 out:
642 	WARN_ON_ONCE(ret);
643 	rtnl_unlock();
644 }
645 
646 static int netkit_link_update(struct bpf_link *link, struct bpf_prog *nprog,
647 			      struct bpf_prog *oprog)
648 {
649 	struct netkit_link *nkl = netkit_link(link);
650 	struct bpf_mprog_entry *entry, *entry_new;
651 	struct net_device *dev;
652 	int ret = 0;
653 
654 	rtnl_lock();
655 	dev = nkl->dev;
656 	if (!dev) {
657 		ret = -ENOLINK;
658 		goto out;
659 	}
660 	if (oprog && link->prog != oprog) {
661 		ret = -EPERM;
662 		goto out;
663 	}
664 	oprog = link->prog;
665 	if (oprog == nprog) {
666 		bpf_prog_put(nprog);
667 		goto out;
668 	}
669 	entry = netkit_entry_fetch(dev, false);
670 	if (!entry) {
671 		ret = -ENOENT;
672 		goto out;
673 	}
674 	ret = bpf_mprog_attach(entry, &entry_new, nprog, link, oprog,
675 			       BPF_F_REPLACE | BPF_F_ID,
676 			       link->prog->aux->id, 0);
677 	if (!ret) {
678 		WARN_ON_ONCE(entry != entry_new);
679 		oprog = xchg(&link->prog, nprog);
680 		bpf_prog_put(oprog);
681 		bpf_mprog_commit(entry);
682 	}
683 out:
684 	rtnl_unlock();
685 	return ret;
686 }
687 
688 static void netkit_link_dealloc(struct bpf_link *link)
689 {
690 	kfree(netkit_link(link));
691 }
692 
693 static void netkit_link_fdinfo(const struct bpf_link *link, struct seq_file *seq)
694 {
695 	const struct netkit_link *nkl = netkit_link(link);
696 	u32 ifindex = 0;
697 
698 	rtnl_lock();
699 	if (nkl->dev)
700 		ifindex = nkl->dev->ifindex;
701 	rtnl_unlock();
702 
703 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
704 	seq_printf(seq, "attach_type:\t%u (%s)\n",
705 		   nkl->location,
706 		   nkl->location == BPF_NETKIT_PRIMARY ? "primary" : "peer");
707 }
708 
709 static int netkit_link_fill_info(const struct bpf_link *link,
710 				 struct bpf_link_info *info)
711 {
712 	const struct netkit_link *nkl = netkit_link(link);
713 	u32 ifindex = 0;
714 
715 	rtnl_lock();
716 	if (nkl->dev)
717 		ifindex = nkl->dev->ifindex;
718 	rtnl_unlock();
719 
720 	info->netkit.ifindex = ifindex;
721 	info->netkit.attach_type = nkl->location;
722 	return 0;
723 }
724 
725 static int netkit_link_detach(struct bpf_link *link)
726 {
727 	netkit_link_release(link);
728 	return 0;
729 }
730 
731 static const struct bpf_link_ops netkit_link_lops = {
732 	.release	= netkit_link_release,
733 	.detach		= netkit_link_detach,
734 	.dealloc	= netkit_link_dealloc,
735 	.update_prog	= netkit_link_update,
736 	.show_fdinfo	= netkit_link_fdinfo,
737 	.fill_link_info	= netkit_link_fill_info,
738 };
739 
740 static int netkit_link_init(struct netkit_link *nkl,
741 			    struct bpf_link_primer *link_primer,
742 			    const union bpf_attr *attr,
743 			    struct net_device *dev,
744 			    struct bpf_prog *prog)
745 {
746 	bpf_link_init(&nkl->link, BPF_LINK_TYPE_NETKIT,
747 		      &netkit_link_lops, prog);
748 	nkl->location = attr->link_create.attach_type;
749 	nkl->dev = dev;
750 	return bpf_link_prime(&nkl->link, link_primer);
751 }
752 
753 int netkit_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
754 {
755 	struct bpf_link_primer link_primer;
756 	struct netkit_link *nkl;
757 	struct net_device *dev;
758 	int ret;
759 
760 	rtnl_lock();
761 	dev = netkit_dev_fetch(current->nsproxy->net_ns,
762 			       attr->link_create.target_ifindex,
763 			       attr->link_create.attach_type);
764 	if (IS_ERR(dev)) {
765 		ret = PTR_ERR(dev);
766 		goto out;
767 	}
768 	nkl = kzalloc(sizeof(*nkl), GFP_KERNEL_ACCOUNT);
769 	if (!nkl) {
770 		ret = -ENOMEM;
771 		goto out;
772 	}
773 	ret = netkit_link_init(nkl, &link_primer, attr, dev, prog);
774 	if (ret) {
775 		kfree(nkl);
776 		goto out;
777 	}
778 	ret = netkit_link_prog_attach(&nkl->link,
779 				      attr->link_create.flags,
780 				      attr->link_create.netkit.relative_fd,
781 				      attr->link_create.netkit.expected_revision);
782 	if (ret) {
783 		nkl->dev = NULL;
784 		bpf_link_cleanup(&link_primer);
785 		goto out;
786 	}
787 	ret = bpf_link_settle(&link_primer);
788 out:
789 	rtnl_unlock();
790 	return ret;
791 }
792 
793 static void netkit_release_all(struct net_device *dev)
794 {
795 	struct bpf_mprog_entry *entry;
796 	struct bpf_tuple tuple = {};
797 	struct bpf_mprog_fp *fp;
798 	struct bpf_mprog_cp *cp;
799 
800 	entry = netkit_entry_fetch(dev, false);
801 	if (!entry)
802 		return;
803 	netkit_entry_update(dev, NULL);
804 	netkit_entry_sync();
805 	bpf_mprog_foreach_tuple(entry, fp, cp, tuple) {
806 		if (tuple.link)
807 			netkit_link(tuple.link)->dev = NULL;
808 		else
809 			bpf_prog_put(tuple.prog);
810 	}
811 }
812 
813 static void netkit_uninit(struct net_device *dev)
814 {
815 	netkit_release_all(dev);
816 }
817 
818 static void netkit_del_link(struct net_device *dev, struct list_head *head)
819 {
820 	struct netkit *nk = netkit_priv(dev);
821 	struct net_device *peer = rtnl_dereference(nk->peer);
822 
823 	RCU_INIT_POINTER(nk->peer, NULL);
824 	unregister_netdevice_queue(dev, head);
825 	if (peer) {
826 		nk = netkit_priv(peer);
827 		RCU_INIT_POINTER(nk->peer, NULL);
828 		unregister_netdevice_queue(peer, head);
829 	}
830 }
831 
832 static int netkit_change_link(struct net_device *dev, struct nlattr *tb[],
833 			      struct nlattr *data[],
834 			      struct netlink_ext_ack *extack)
835 {
836 	struct netkit *nk = netkit_priv(dev);
837 	struct net_device *peer = rtnl_dereference(nk->peer);
838 	enum netkit_action policy;
839 	struct nlattr *attr;
840 	int err;
841 
842 	if (!nk->primary) {
843 		NL_SET_ERR_MSG(extack,
844 			       "netkit link settings can be changed only through the primary device");
845 		return -EACCES;
846 	}
847 
848 	if (data[IFLA_NETKIT_MODE]) {
849 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_NETKIT_MODE],
850 				    "netkit link operating mode cannot be changed after device creation");
851 		return -EACCES;
852 	}
853 
854 	if (data[IFLA_NETKIT_PEER_INFO]) {
855 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_NETKIT_PEER_INFO],
856 				    "netkit peer info cannot be changed after device creation");
857 		return -EINVAL;
858 	}
859 
860 	if (data[IFLA_NETKIT_POLICY]) {
861 		attr = data[IFLA_NETKIT_POLICY];
862 		policy = nla_get_u32(attr);
863 		err = netkit_check_policy(policy, attr, extack);
864 		if (err)
865 			return err;
866 		WRITE_ONCE(nk->policy, policy);
867 	}
868 
869 	if (data[IFLA_NETKIT_PEER_POLICY]) {
870 		err = -EOPNOTSUPP;
871 		attr = data[IFLA_NETKIT_PEER_POLICY];
872 		policy = nla_get_u32(attr);
873 		if (peer)
874 			err = netkit_check_policy(policy, attr, extack);
875 		if (err)
876 			return err;
877 		nk = netkit_priv(peer);
878 		WRITE_ONCE(nk->policy, policy);
879 	}
880 
881 	return 0;
882 }
883 
884 static size_t netkit_get_size(const struct net_device *dev)
885 {
886 	return nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_POLICY */
887 	       nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_PEER_POLICY */
888 	       nla_total_size(sizeof(u8))  + /* IFLA_NETKIT_PRIMARY */
889 	       nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_MODE */
890 	       0;
891 }
892 
893 static int netkit_fill_info(struct sk_buff *skb, const struct net_device *dev)
894 {
895 	struct netkit *nk = netkit_priv(dev);
896 	struct net_device *peer = rtnl_dereference(nk->peer);
897 
898 	if (nla_put_u8(skb, IFLA_NETKIT_PRIMARY, nk->primary))
899 		return -EMSGSIZE;
900 	if (nla_put_u32(skb, IFLA_NETKIT_POLICY, nk->policy))
901 		return -EMSGSIZE;
902 	if (nla_put_u32(skb, IFLA_NETKIT_MODE, nk->mode))
903 		return -EMSGSIZE;
904 
905 	if (peer) {
906 		nk = netkit_priv(peer);
907 		if (nla_put_u32(skb, IFLA_NETKIT_PEER_POLICY, nk->policy))
908 			return -EMSGSIZE;
909 	}
910 
911 	return 0;
912 }
913 
914 static const struct nla_policy netkit_policy[IFLA_NETKIT_MAX + 1] = {
915 	[IFLA_NETKIT_PEER_INFO]		= { .len = sizeof(struct ifinfomsg) },
916 	[IFLA_NETKIT_POLICY]		= { .type = NLA_U32 },
917 	[IFLA_NETKIT_MODE]		= { .type = NLA_U32 },
918 	[IFLA_NETKIT_PEER_POLICY]	= { .type = NLA_U32 },
919 	[IFLA_NETKIT_PRIMARY]		= { .type = NLA_REJECT,
920 					    .reject_message = "Primary attribute is read-only" },
921 };
922 
923 static struct rtnl_link_ops netkit_link_ops = {
924 	.kind		= DRV_NAME,
925 	.priv_size	= sizeof(struct netkit),
926 	.setup		= netkit_setup,
927 	.newlink	= netkit_new_link,
928 	.dellink	= netkit_del_link,
929 	.changelink	= netkit_change_link,
930 	.get_link_net	= netkit_get_link_net,
931 	.get_size	= netkit_get_size,
932 	.fill_info	= netkit_fill_info,
933 	.policy		= netkit_policy,
934 	.validate	= netkit_validate,
935 	.maxtype	= IFLA_NETKIT_MAX,
936 };
937 
938 static __init int netkit_init(void)
939 {
940 	BUILD_BUG_ON((int)NETKIT_NEXT != (int)TCX_NEXT ||
941 		     (int)NETKIT_PASS != (int)TCX_PASS ||
942 		     (int)NETKIT_DROP != (int)TCX_DROP ||
943 		     (int)NETKIT_REDIRECT != (int)TCX_REDIRECT);
944 
945 	return rtnl_link_register(&netkit_link_ops);
946 }
947 
948 static __exit void netkit_exit(void)
949 {
950 	rtnl_link_unregister(&netkit_link_ops);
951 }
952 
953 module_init(netkit_init);
954 module_exit(netkit_exit);
955 
956 MODULE_DESCRIPTION("BPF-programmable network device");
957 MODULE_AUTHOR("Daniel Borkmann <daniel@iogearbox.net>");
958 MODULE_AUTHOR("Nikolay Aleksandrov <razor@blackwall.org>");
959 MODULE_LICENSE("GPL");
960 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
961