xref: /linux/drivers/net/netkit.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 	skb_reset_mac_header(skb);
59 }
60 
61 static struct netkit *netkit_priv(const struct net_device *dev)
62 {
63 	return netdev_priv(dev);
64 }
65 
66 static netdev_tx_t netkit_xmit(struct sk_buff *skb, struct net_device *dev)
67 {
68 	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
69 	struct netkit *nk = netkit_priv(dev);
70 	enum netkit_action ret = READ_ONCE(nk->policy);
71 	netdev_tx_t ret_dev = NET_XMIT_SUCCESS;
72 	const struct bpf_mprog_entry *entry;
73 	struct net_device *peer;
74 	int len = skb->len;
75 
76 	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
77 	rcu_read_lock();
78 	peer = rcu_dereference(nk->peer);
79 	if (unlikely(!peer || !(peer->flags & IFF_UP) ||
80 		     !pskb_may_pull(skb, ETH_HLEN) ||
81 		     skb_orphan_frags(skb, GFP_ATOMIC)))
82 		goto drop;
83 	netkit_prep_forward(skb, !net_eq(dev_net(dev), dev_net(peer)));
84 	eth_skb_pkt_type(skb, peer);
85 	skb->dev = peer;
86 	entry = rcu_dereference(nk->active);
87 	if (entry)
88 		ret = netkit_run(entry, skb, ret);
89 	switch (ret) {
90 	case NETKIT_NEXT:
91 	case NETKIT_PASS:
92 		eth_skb_pull_mac(skb);
93 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
94 		if (likely(__netif_rx(skb) == NET_RX_SUCCESS)) {
95 			dev_sw_netstats_tx_add(dev, 1, len);
96 			dev_sw_netstats_rx_add(peer, len);
97 		} else {
98 			goto drop_stats;
99 		}
100 		break;
101 	case NETKIT_REDIRECT:
102 		dev_sw_netstats_tx_add(dev, 1, len);
103 		skb_do_redirect(skb);
104 		break;
105 	case NETKIT_DROP:
106 	default:
107 drop:
108 		kfree_skb(skb);
109 drop_stats:
110 		dev_core_stats_tx_dropped_inc(dev);
111 		ret_dev = NET_XMIT_DROP;
112 		break;
113 	}
114 	rcu_read_unlock();
115 	bpf_net_ctx_clear(bpf_net_ctx);
116 	return ret_dev;
117 }
118 
119 static int netkit_open(struct net_device *dev)
120 {
121 	struct netkit *nk = netkit_priv(dev);
122 	struct net_device *peer = rtnl_dereference(nk->peer);
123 
124 	if (!peer)
125 		return -ENOTCONN;
126 	if (peer->flags & IFF_UP) {
127 		netif_carrier_on(dev);
128 		netif_carrier_on(peer);
129 	}
130 	return 0;
131 }
132 
133 static int netkit_close(struct net_device *dev)
134 {
135 	struct netkit *nk = netkit_priv(dev);
136 	struct net_device *peer = rtnl_dereference(nk->peer);
137 
138 	netif_carrier_off(dev);
139 	if (peer)
140 		netif_carrier_off(peer);
141 	return 0;
142 }
143 
144 static int netkit_get_iflink(const struct net_device *dev)
145 {
146 	struct netkit *nk = netkit_priv(dev);
147 	struct net_device *peer;
148 	int iflink = 0;
149 
150 	rcu_read_lock();
151 	peer = rcu_dereference(nk->peer);
152 	if (peer)
153 		iflink = READ_ONCE(peer->ifindex);
154 	rcu_read_unlock();
155 	return iflink;
156 }
157 
158 static void netkit_set_multicast(struct net_device *dev)
159 {
160 	/* Nothing to do, we receive whatever gets pushed to us! */
161 }
162 
163 static int netkit_set_macaddr(struct net_device *dev, void *sa)
164 {
165 	struct netkit *nk = netkit_priv(dev);
166 
167 	if (nk->mode != NETKIT_L2)
168 		return -EOPNOTSUPP;
169 
170 	return eth_mac_addr(dev, sa);
171 }
172 
173 static void netkit_set_headroom(struct net_device *dev, int headroom)
174 {
175 	struct netkit *nk = netkit_priv(dev), *nk2;
176 	struct net_device *peer;
177 
178 	if (headroom < 0)
179 		headroom = NET_SKB_PAD;
180 
181 	rcu_read_lock();
182 	peer = rcu_dereference(nk->peer);
183 	if (unlikely(!peer))
184 		goto out;
185 
186 	nk2 = netkit_priv(peer);
187 	nk->headroom = headroom;
188 	headroom = max(nk->headroom, nk2->headroom);
189 
190 	peer->needed_headroom = headroom;
191 	dev->needed_headroom = headroom;
192 out:
193 	rcu_read_unlock();
194 }
195 
196 INDIRECT_CALLABLE_SCOPE struct net_device *netkit_peer_dev(struct net_device *dev)
197 {
198 	return rcu_dereference(netkit_priv(dev)->peer);
199 }
200 
201 static void netkit_get_stats(struct net_device *dev,
202 			     struct rtnl_link_stats64 *stats)
203 {
204 	dev_fetch_sw_netstats(stats, dev->tstats);
205 	stats->tx_dropped = DEV_STATS_READ(dev, tx_dropped);
206 }
207 
208 static void netkit_uninit(struct net_device *dev);
209 
210 static const struct net_device_ops netkit_netdev_ops = {
211 	.ndo_open		= netkit_open,
212 	.ndo_stop		= netkit_close,
213 	.ndo_start_xmit		= netkit_xmit,
214 	.ndo_set_rx_mode	= netkit_set_multicast,
215 	.ndo_set_rx_headroom	= netkit_set_headroom,
216 	.ndo_set_mac_address	= netkit_set_macaddr,
217 	.ndo_get_iflink		= netkit_get_iflink,
218 	.ndo_get_peer_dev	= netkit_peer_dev,
219 	.ndo_get_stats64	= netkit_get_stats,
220 	.ndo_uninit		= netkit_uninit,
221 	.ndo_features_check	= passthru_features_check,
222 };
223 
224 static void netkit_get_drvinfo(struct net_device *dev,
225 			       struct ethtool_drvinfo *info)
226 {
227 	strscpy(info->driver, DRV_NAME, sizeof(info->driver));
228 }
229 
230 static const struct ethtool_ops netkit_ethtool_ops = {
231 	.get_drvinfo		= netkit_get_drvinfo,
232 };
233 
234 static void netkit_setup(struct net_device *dev)
235 {
236 	static const netdev_features_t netkit_features_hw_vlan =
237 		NETIF_F_HW_VLAN_CTAG_TX |
238 		NETIF_F_HW_VLAN_CTAG_RX |
239 		NETIF_F_HW_VLAN_STAG_TX |
240 		NETIF_F_HW_VLAN_STAG_RX;
241 	static const netdev_features_t netkit_features =
242 		netkit_features_hw_vlan |
243 		NETIF_F_SG |
244 		NETIF_F_FRAGLIST |
245 		NETIF_F_HW_CSUM |
246 		NETIF_F_RXCSUM |
247 		NETIF_F_SCTP_CRC |
248 		NETIF_F_HIGHDMA |
249 		NETIF_F_GSO_SOFTWARE |
250 		NETIF_F_GSO_ENCAP_ALL;
251 
252 	ether_setup(dev);
253 	dev->max_mtu = ETH_MAX_MTU;
254 	dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
255 
256 	dev->flags |= IFF_NOARP;
257 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
258 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
259 	dev->priv_flags |= IFF_PHONY_HEADROOM;
260 	dev->priv_flags |= IFF_NO_QUEUE;
261 	dev->priv_flags |= IFF_DISABLE_NETPOLL;
262 	dev->lltx = true;
263 
264 	dev->ethtool_ops = &netkit_ethtool_ops;
265 	dev->netdev_ops  = &netkit_netdev_ops;
266 
267 	dev->features |= netkit_features;
268 	dev->hw_features = netkit_features;
269 	dev->hw_enc_features = netkit_features;
270 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
271 	dev->vlan_features = dev->features & ~netkit_features_hw_vlan;
272 
273 	dev->needs_free_netdev = true;
274 
275 	netif_set_tso_max_size(dev, GSO_MAX_SIZE);
276 }
277 
278 static struct net *netkit_get_link_net(const struct net_device *dev)
279 {
280 	struct netkit *nk = netkit_priv(dev);
281 	struct net_device *peer = rtnl_dereference(nk->peer);
282 
283 	return peer ? dev_net(peer) : dev_net(dev);
284 }
285 
286 static int netkit_check_policy(int policy, struct nlattr *tb,
287 			       struct netlink_ext_ack *extack)
288 {
289 	switch (policy) {
290 	case NETKIT_PASS:
291 	case NETKIT_DROP:
292 		return 0;
293 	default:
294 		NL_SET_ERR_MSG_ATTR(extack, tb,
295 				    "Provided default xmit policy not supported");
296 		return -EINVAL;
297 	}
298 }
299 
300 static int netkit_check_mode(int mode, struct nlattr *tb,
301 			     struct netlink_ext_ack *extack)
302 {
303 	switch (mode) {
304 	case NETKIT_L2:
305 	case NETKIT_L3:
306 		return 0;
307 	default:
308 		NL_SET_ERR_MSG_ATTR(extack, tb,
309 				    "Provided device mode can only be L2 or L3");
310 		return -EINVAL;
311 	}
312 }
313 
314 static int netkit_validate(struct nlattr *tb[], struct nlattr *data[],
315 			   struct netlink_ext_ack *extack)
316 {
317 	struct nlattr *attr = tb[IFLA_ADDRESS];
318 
319 	if (!attr)
320 		return 0;
321 	if (nla_len(attr) != ETH_ALEN)
322 		return -EINVAL;
323 	if (!is_valid_ether_addr(nla_data(attr)))
324 		return -EADDRNOTAVAIL;
325 	return 0;
326 }
327 
328 static struct rtnl_link_ops netkit_link_ops;
329 
330 static int netkit_new_link(struct net *src_net, struct net_device *dev,
331 			   struct nlattr *tb[], struct nlattr *data[],
332 			   struct netlink_ext_ack *extack)
333 {
334 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp = tb, *attr;
335 	enum netkit_action default_prim = NETKIT_PASS;
336 	enum netkit_action default_peer = NETKIT_PASS;
337 	enum netkit_mode mode = NETKIT_L3;
338 	unsigned char ifname_assign_type;
339 	struct ifinfomsg *ifmp = NULL;
340 	struct net_device *peer;
341 	char ifname[IFNAMSIZ];
342 	struct netkit *nk;
343 	struct net *net;
344 	int err;
345 
346 	if (data) {
347 		if (data[IFLA_NETKIT_MODE]) {
348 			attr = data[IFLA_NETKIT_MODE];
349 			mode = nla_get_u32(attr);
350 			err = netkit_check_mode(mode, attr, extack);
351 			if (err < 0)
352 				return err;
353 		}
354 		if (data[IFLA_NETKIT_PEER_INFO]) {
355 			attr = data[IFLA_NETKIT_PEER_INFO];
356 			ifmp = nla_data(attr);
357 			err = rtnl_nla_parse_ifinfomsg(peer_tb, attr, extack);
358 			if (err < 0)
359 				return err;
360 			err = netkit_validate(peer_tb, NULL, extack);
361 			if (err < 0)
362 				return err;
363 			tbp = peer_tb;
364 		}
365 		if (data[IFLA_NETKIT_POLICY]) {
366 			attr = data[IFLA_NETKIT_POLICY];
367 			default_prim = nla_get_u32(attr);
368 			err = netkit_check_policy(default_prim, attr, extack);
369 			if (err < 0)
370 				return err;
371 		}
372 		if (data[IFLA_NETKIT_PEER_POLICY]) {
373 			attr = data[IFLA_NETKIT_PEER_POLICY];
374 			default_peer = nla_get_u32(attr);
375 			err = netkit_check_policy(default_peer, attr, extack);
376 			if (err < 0)
377 				return err;
378 		}
379 	}
380 
381 	if (ifmp && tbp[IFLA_IFNAME]) {
382 		nla_strscpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
383 		ifname_assign_type = NET_NAME_USER;
384 	} else {
385 		strscpy(ifname, "nk%d", IFNAMSIZ);
386 		ifname_assign_type = NET_NAME_ENUM;
387 	}
388 	if (mode != NETKIT_L2 &&
389 	    (tb[IFLA_ADDRESS] || tbp[IFLA_ADDRESS]))
390 		return -EOPNOTSUPP;
391 
392 	net = rtnl_link_get_net(src_net, tbp);
393 	if (IS_ERR(net))
394 		return PTR_ERR(net);
395 
396 	peer = rtnl_create_link(net, ifname, ifname_assign_type,
397 				&netkit_link_ops, tbp, extack);
398 	if (IS_ERR(peer)) {
399 		put_net(net);
400 		return PTR_ERR(peer);
401 	}
402 
403 	netif_inherit_tso_max(peer, dev);
404 
405 	if (mode == NETKIT_L2 && !(ifmp && tbp[IFLA_ADDRESS]))
406 		eth_hw_addr_random(peer);
407 	if (ifmp && dev->ifindex)
408 		peer->ifindex = ifmp->ifi_index;
409 
410 	nk = netkit_priv(peer);
411 	nk->primary = false;
412 	nk->policy = default_peer;
413 	nk->mode = mode;
414 	bpf_mprog_bundle_init(&nk->bundle);
415 
416 	err = register_netdevice(peer);
417 	put_net(net);
418 	if (err < 0)
419 		goto err_register_peer;
420 	netif_carrier_off(peer);
421 	if (mode == NETKIT_L2)
422 		dev_change_flags(peer, peer->flags & ~IFF_NOARP, NULL);
423 
424 	err = rtnl_configure_link(peer, NULL, 0, NULL);
425 	if (err < 0)
426 		goto err_configure_peer;
427 
428 	if (mode == NETKIT_L2 && !tb[IFLA_ADDRESS])
429 		eth_hw_addr_random(dev);
430 	if (tb[IFLA_IFNAME])
431 		nla_strscpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
432 	else
433 		strscpy(dev->name, "nk%d", IFNAMSIZ);
434 
435 	nk = netkit_priv(dev);
436 	nk->primary = true;
437 	nk->policy = default_prim;
438 	nk->mode = mode;
439 	bpf_mprog_bundle_init(&nk->bundle);
440 
441 	err = register_netdevice(dev);
442 	if (err < 0)
443 		goto err_configure_peer;
444 	netif_carrier_off(dev);
445 	if (mode == NETKIT_L2)
446 		dev_change_flags(dev, dev->flags & ~IFF_NOARP, NULL);
447 
448 	rcu_assign_pointer(netkit_priv(dev)->peer, peer);
449 	rcu_assign_pointer(netkit_priv(peer)->peer, dev);
450 	return 0;
451 err_configure_peer:
452 	unregister_netdevice(peer);
453 	return err;
454 err_register_peer:
455 	free_netdev(peer);
456 	return err;
457 }
458 
459 static struct bpf_mprog_entry *netkit_entry_fetch(struct net_device *dev,
460 						  bool bundle_fallback)
461 {
462 	struct netkit *nk = netkit_priv(dev);
463 	struct bpf_mprog_entry *entry;
464 
465 	ASSERT_RTNL();
466 	entry = rcu_dereference_rtnl(nk->active);
467 	if (entry)
468 		return entry;
469 	if (bundle_fallback)
470 		return &nk->bundle.a;
471 	return NULL;
472 }
473 
474 static void netkit_entry_update(struct net_device *dev,
475 				struct bpf_mprog_entry *entry)
476 {
477 	struct netkit *nk = netkit_priv(dev);
478 
479 	ASSERT_RTNL();
480 	rcu_assign_pointer(nk->active, entry);
481 }
482 
483 static void netkit_entry_sync(void)
484 {
485 	synchronize_rcu();
486 }
487 
488 static struct net_device *netkit_dev_fetch(struct net *net, u32 ifindex, u32 which)
489 {
490 	struct net_device *dev;
491 	struct netkit *nk;
492 
493 	ASSERT_RTNL();
494 
495 	switch (which) {
496 	case BPF_NETKIT_PRIMARY:
497 	case BPF_NETKIT_PEER:
498 		break;
499 	default:
500 		return ERR_PTR(-EINVAL);
501 	}
502 
503 	dev = __dev_get_by_index(net, ifindex);
504 	if (!dev)
505 		return ERR_PTR(-ENODEV);
506 	if (dev->netdev_ops != &netkit_netdev_ops)
507 		return ERR_PTR(-ENXIO);
508 
509 	nk = netkit_priv(dev);
510 	if (!nk->primary)
511 		return ERR_PTR(-EACCES);
512 	if (which == BPF_NETKIT_PEER) {
513 		dev = rcu_dereference_rtnl(nk->peer);
514 		if (!dev)
515 			return ERR_PTR(-ENODEV);
516 	}
517 	return dev;
518 }
519 
520 int netkit_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)
521 {
522 	struct bpf_mprog_entry *entry, *entry_new;
523 	struct bpf_prog *replace_prog = NULL;
524 	struct net_device *dev;
525 	int ret;
526 
527 	rtnl_lock();
528 	dev = netkit_dev_fetch(current->nsproxy->net_ns, attr->target_ifindex,
529 			       attr->attach_type);
530 	if (IS_ERR(dev)) {
531 		ret = PTR_ERR(dev);
532 		goto out;
533 	}
534 	entry = netkit_entry_fetch(dev, true);
535 	if (attr->attach_flags & BPF_F_REPLACE) {
536 		replace_prog = bpf_prog_get_type(attr->replace_bpf_fd,
537 						 prog->type);
538 		if (IS_ERR(replace_prog)) {
539 			ret = PTR_ERR(replace_prog);
540 			replace_prog = NULL;
541 			goto out;
542 		}
543 	}
544 	ret = bpf_mprog_attach(entry, &entry_new, prog, NULL, replace_prog,
545 			       attr->attach_flags, attr->relative_fd,
546 			       attr->expected_revision);
547 	if (!ret) {
548 		if (entry != entry_new) {
549 			netkit_entry_update(dev, entry_new);
550 			netkit_entry_sync();
551 		}
552 		bpf_mprog_commit(entry);
553 	}
554 out:
555 	if (replace_prog)
556 		bpf_prog_put(replace_prog);
557 	rtnl_unlock();
558 	return ret;
559 }
560 
561 int netkit_prog_detach(const union bpf_attr *attr, struct bpf_prog *prog)
562 {
563 	struct bpf_mprog_entry *entry, *entry_new;
564 	struct net_device *dev;
565 	int ret;
566 
567 	rtnl_lock();
568 	dev = netkit_dev_fetch(current->nsproxy->net_ns, attr->target_ifindex,
569 			       attr->attach_type);
570 	if (IS_ERR(dev)) {
571 		ret = PTR_ERR(dev);
572 		goto out;
573 	}
574 	entry = netkit_entry_fetch(dev, false);
575 	if (!entry) {
576 		ret = -ENOENT;
577 		goto out;
578 	}
579 	ret = bpf_mprog_detach(entry, &entry_new, prog, NULL, attr->attach_flags,
580 			       attr->relative_fd, attr->expected_revision);
581 	if (!ret) {
582 		if (!bpf_mprog_total(entry_new))
583 			entry_new = NULL;
584 		netkit_entry_update(dev, entry_new);
585 		netkit_entry_sync();
586 		bpf_mprog_commit(entry);
587 	}
588 out:
589 	rtnl_unlock();
590 	return ret;
591 }
592 
593 int netkit_prog_query(const union bpf_attr *attr, union bpf_attr __user *uattr)
594 {
595 	struct net_device *dev;
596 	int ret;
597 
598 	rtnl_lock();
599 	dev = netkit_dev_fetch(current->nsproxy->net_ns,
600 			       attr->query.target_ifindex,
601 			       attr->query.attach_type);
602 	if (IS_ERR(dev)) {
603 		ret = PTR_ERR(dev);
604 		goto out;
605 	}
606 	ret = bpf_mprog_query(attr, uattr, netkit_entry_fetch(dev, false));
607 out:
608 	rtnl_unlock();
609 	return ret;
610 }
611 
612 static struct netkit_link *netkit_link(const struct bpf_link *link)
613 {
614 	return container_of(link, struct netkit_link, link);
615 }
616 
617 static int netkit_link_prog_attach(struct bpf_link *link, u32 flags,
618 				   u32 id_or_fd, u64 revision)
619 {
620 	struct netkit_link *nkl = netkit_link(link);
621 	struct bpf_mprog_entry *entry, *entry_new;
622 	struct net_device *dev = nkl->dev;
623 	int ret;
624 
625 	ASSERT_RTNL();
626 	entry = netkit_entry_fetch(dev, true);
627 	ret = bpf_mprog_attach(entry, &entry_new, link->prog, link, NULL, flags,
628 			       id_or_fd, revision);
629 	if (!ret) {
630 		if (entry != entry_new) {
631 			netkit_entry_update(dev, entry_new);
632 			netkit_entry_sync();
633 		}
634 		bpf_mprog_commit(entry);
635 	}
636 	return ret;
637 }
638 
639 static void netkit_link_release(struct bpf_link *link)
640 {
641 	struct netkit_link *nkl = netkit_link(link);
642 	struct bpf_mprog_entry *entry, *entry_new;
643 	struct net_device *dev;
644 	int ret = 0;
645 
646 	rtnl_lock();
647 	dev = nkl->dev;
648 	if (!dev)
649 		goto out;
650 	entry = netkit_entry_fetch(dev, false);
651 	if (!entry) {
652 		ret = -ENOENT;
653 		goto out;
654 	}
655 	ret = bpf_mprog_detach(entry, &entry_new, link->prog, link, 0, 0, 0);
656 	if (!ret) {
657 		if (!bpf_mprog_total(entry_new))
658 			entry_new = NULL;
659 		netkit_entry_update(dev, entry_new);
660 		netkit_entry_sync();
661 		bpf_mprog_commit(entry);
662 		nkl->dev = NULL;
663 	}
664 out:
665 	WARN_ON_ONCE(ret);
666 	rtnl_unlock();
667 }
668 
669 static int netkit_link_update(struct bpf_link *link, struct bpf_prog *nprog,
670 			      struct bpf_prog *oprog)
671 {
672 	struct netkit_link *nkl = netkit_link(link);
673 	struct bpf_mprog_entry *entry, *entry_new;
674 	struct net_device *dev;
675 	int ret = 0;
676 
677 	rtnl_lock();
678 	dev = nkl->dev;
679 	if (!dev) {
680 		ret = -ENOLINK;
681 		goto out;
682 	}
683 	if (oprog && link->prog != oprog) {
684 		ret = -EPERM;
685 		goto out;
686 	}
687 	oprog = link->prog;
688 	if (oprog == nprog) {
689 		bpf_prog_put(nprog);
690 		goto out;
691 	}
692 	entry = netkit_entry_fetch(dev, false);
693 	if (!entry) {
694 		ret = -ENOENT;
695 		goto out;
696 	}
697 	ret = bpf_mprog_attach(entry, &entry_new, nprog, link, oprog,
698 			       BPF_F_REPLACE | BPF_F_ID,
699 			       link->prog->aux->id, 0);
700 	if (!ret) {
701 		WARN_ON_ONCE(entry != entry_new);
702 		oprog = xchg(&link->prog, nprog);
703 		bpf_prog_put(oprog);
704 		bpf_mprog_commit(entry);
705 	}
706 out:
707 	rtnl_unlock();
708 	return ret;
709 }
710 
711 static void netkit_link_dealloc(struct bpf_link *link)
712 {
713 	kfree(netkit_link(link));
714 }
715 
716 static void netkit_link_fdinfo(const struct bpf_link *link, struct seq_file *seq)
717 {
718 	const struct netkit_link *nkl = netkit_link(link);
719 	u32 ifindex = 0;
720 
721 	rtnl_lock();
722 	if (nkl->dev)
723 		ifindex = nkl->dev->ifindex;
724 	rtnl_unlock();
725 
726 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
727 	seq_printf(seq, "attach_type:\t%u (%s)\n",
728 		   nkl->location,
729 		   nkl->location == BPF_NETKIT_PRIMARY ? "primary" : "peer");
730 }
731 
732 static int netkit_link_fill_info(const struct bpf_link *link,
733 				 struct bpf_link_info *info)
734 {
735 	const struct netkit_link *nkl = netkit_link(link);
736 	u32 ifindex = 0;
737 
738 	rtnl_lock();
739 	if (nkl->dev)
740 		ifindex = nkl->dev->ifindex;
741 	rtnl_unlock();
742 
743 	info->netkit.ifindex = ifindex;
744 	info->netkit.attach_type = nkl->location;
745 	return 0;
746 }
747 
748 static int netkit_link_detach(struct bpf_link *link)
749 {
750 	netkit_link_release(link);
751 	return 0;
752 }
753 
754 static const struct bpf_link_ops netkit_link_lops = {
755 	.release	= netkit_link_release,
756 	.detach		= netkit_link_detach,
757 	.dealloc	= netkit_link_dealloc,
758 	.update_prog	= netkit_link_update,
759 	.show_fdinfo	= netkit_link_fdinfo,
760 	.fill_link_info	= netkit_link_fill_info,
761 };
762 
763 static int netkit_link_init(struct netkit_link *nkl,
764 			    struct bpf_link_primer *link_primer,
765 			    const union bpf_attr *attr,
766 			    struct net_device *dev,
767 			    struct bpf_prog *prog)
768 {
769 	bpf_link_init(&nkl->link, BPF_LINK_TYPE_NETKIT,
770 		      &netkit_link_lops, prog);
771 	nkl->location = attr->link_create.attach_type;
772 	nkl->dev = dev;
773 	return bpf_link_prime(&nkl->link, link_primer);
774 }
775 
776 int netkit_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
777 {
778 	struct bpf_link_primer link_primer;
779 	struct netkit_link *nkl;
780 	struct net_device *dev;
781 	int ret;
782 
783 	rtnl_lock();
784 	dev = netkit_dev_fetch(current->nsproxy->net_ns,
785 			       attr->link_create.target_ifindex,
786 			       attr->link_create.attach_type);
787 	if (IS_ERR(dev)) {
788 		ret = PTR_ERR(dev);
789 		goto out;
790 	}
791 	nkl = kzalloc(sizeof(*nkl), GFP_KERNEL_ACCOUNT);
792 	if (!nkl) {
793 		ret = -ENOMEM;
794 		goto out;
795 	}
796 	ret = netkit_link_init(nkl, &link_primer, attr, dev, prog);
797 	if (ret) {
798 		kfree(nkl);
799 		goto out;
800 	}
801 	ret = netkit_link_prog_attach(&nkl->link,
802 				      attr->link_create.flags,
803 				      attr->link_create.netkit.relative_fd,
804 				      attr->link_create.netkit.expected_revision);
805 	if (ret) {
806 		nkl->dev = NULL;
807 		bpf_link_cleanup(&link_primer);
808 		goto out;
809 	}
810 	ret = bpf_link_settle(&link_primer);
811 out:
812 	rtnl_unlock();
813 	return ret;
814 }
815 
816 static void netkit_release_all(struct net_device *dev)
817 {
818 	struct bpf_mprog_entry *entry;
819 	struct bpf_tuple tuple = {};
820 	struct bpf_mprog_fp *fp;
821 	struct bpf_mprog_cp *cp;
822 
823 	entry = netkit_entry_fetch(dev, false);
824 	if (!entry)
825 		return;
826 	netkit_entry_update(dev, NULL);
827 	netkit_entry_sync();
828 	bpf_mprog_foreach_tuple(entry, fp, cp, tuple) {
829 		if (tuple.link)
830 			netkit_link(tuple.link)->dev = NULL;
831 		else
832 			bpf_prog_put(tuple.prog);
833 	}
834 }
835 
836 static void netkit_uninit(struct net_device *dev)
837 {
838 	netkit_release_all(dev);
839 }
840 
841 static void netkit_del_link(struct net_device *dev, struct list_head *head)
842 {
843 	struct netkit *nk = netkit_priv(dev);
844 	struct net_device *peer = rtnl_dereference(nk->peer);
845 
846 	RCU_INIT_POINTER(nk->peer, NULL);
847 	unregister_netdevice_queue(dev, head);
848 	if (peer) {
849 		nk = netkit_priv(peer);
850 		RCU_INIT_POINTER(nk->peer, NULL);
851 		unregister_netdevice_queue(peer, head);
852 	}
853 }
854 
855 static int netkit_change_link(struct net_device *dev, struct nlattr *tb[],
856 			      struct nlattr *data[],
857 			      struct netlink_ext_ack *extack)
858 {
859 	struct netkit *nk = netkit_priv(dev);
860 	struct net_device *peer = rtnl_dereference(nk->peer);
861 	enum netkit_action policy;
862 	struct nlattr *attr;
863 	int err;
864 
865 	if (!nk->primary) {
866 		NL_SET_ERR_MSG(extack,
867 			       "netkit link settings can be changed only through the primary device");
868 		return -EACCES;
869 	}
870 
871 	if (data[IFLA_NETKIT_MODE]) {
872 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_NETKIT_MODE],
873 				    "netkit link operating mode cannot be changed after device creation");
874 		return -EACCES;
875 	}
876 
877 	if (data[IFLA_NETKIT_PEER_INFO]) {
878 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_NETKIT_PEER_INFO],
879 				    "netkit peer info cannot be changed after device creation");
880 		return -EINVAL;
881 	}
882 
883 	if (data[IFLA_NETKIT_POLICY]) {
884 		attr = data[IFLA_NETKIT_POLICY];
885 		policy = nla_get_u32(attr);
886 		err = netkit_check_policy(policy, attr, extack);
887 		if (err)
888 			return err;
889 		WRITE_ONCE(nk->policy, policy);
890 	}
891 
892 	if (data[IFLA_NETKIT_PEER_POLICY]) {
893 		err = -EOPNOTSUPP;
894 		attr = data[IFLA_NETKIT_PEER_POLICY];
895 		policy = nla_get_u32(attr);
896 		if (peer)
897 			err = netkit_check_policy(policy, attr, extack);
898 		if (err)
899 			return err;
900 		nk = netkit_priv(peer);
901 		WRITE_ONCE(nk->policy, policy);
902 	}
903 
904 	return 0;
905 }
906 
907 static size_t netkit_get_size(const struct net_device *dev)
908 {
909 	return nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_POLICY */
910 	       nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_PEER_POLICY */
911 	       nla_total_size(sizeof(u8))  + /* IFLA_NETKIT_PRIMARY */
912 	       nla_total_size(sizeof(u32)) + /* IFLA_NETKIT_MODE */
913 	       0;
914 }
915 
916 static int netkit_fill_info(struct sk_buff *skb, const struct net_device *dev)
917 {
918 	struct netkit *nk = netkit_priv(dev);
919 	struct net_device *peer = rtnl_dereference(nk->peer);
920 
921 	if (nla_put_u8(skb, IFLA_NETKIT_PRIMARY, nk->primary))
922 		return -EMSGSIZE;
923 	if (nla_put_u32(skb, IFLA_NETKIT_POLICY, nk->policy))
924 		return -EMSGSIZE;
925 	if (nla_put_u32(skb, IFLA_NETKIT_MODE, nk->mode))
926 		return -EMSGSIZE;
927 
928 	if (peer) {
929 		nk = netkit_priv(peer);
930 		if (nla_put_u32(skb, IFLA_NETKIT_PEER_POLICY, nk->policy))
931 			return -EMSGSIZE;
932 	}
933 
934 	return 0;
935 }
936 
937 static const struct nla_policy netkit_policy[IFLA_NETKIT_MAX + 1] = {
938 	[IFLA_NETKIT_PEER_INFO]		= { .len = sizeof(struct ifinfomsg) },
939 	[IFLA_NETKIT_POLICY]		= { .type = NLA_U32 },
940 	[IFLA_NETKIT_MODE]		= { .type = NLA_U32 },
941 	[IFLA_NETKIT_PEER_POLICY]	= { .type = NLA_U32 },
942 	[IFLA_NETKIT_PRIMARY]		= { .type = NLA_REJECT,
943 					    .reject_message = "Primary attribute is read-only" },
944 };
945 
946 static struct rtnl_link_ops netkit_link_ops = {
947 	.kind		= DRV_NAME,
948 	.priv_size	= sizeof(struct netkit),
949 	.setup		= netkit_setup,
950 	.newlink	= netkit_new_link,
951 	.dellink	= netkit_del_link,
952 	.changelink	= netkit_change_link,
953 	.get_link_net	= netkit_get_link_net,
954 	.get_size	= netkit_get_size,
955 	.fill_info	= netkit_fill_info,
956 	.policy		= netkit_policy,
957 	.validate	= netkit_validate,
958 	.maxtype	= IFLA_NETKIT_MAX,
959 };
960 
961 static __init int netkit_init(void)
962 {
963 	BUILD_BUG_ON((int)NETKIT_NEXT != (int)TCX_NEXT ||
964 		     (int)NETKIT_PASS != (int)TCX_PASS ||
965 		     (int)NETKIT_DROP != (int)TCX_DROP ||
966 		     (int)NETKIT_REDIRECT != (int)TCX_REDIRECT);
967 
968 	return rtnl_link_register(&netkit_link_ops);
969 }
970 
971 static __exit void netkit_exit(void)
972 {
973 	rtnl_link_unregister(&netkit_link_ops);
974 }
975 
976 module_init(netkit_init);
977 module_exit(netkit_exit);
978 
979 MODULE_DESCRIPTION("BPF-programmable network device");
980 MODULE_AUTHOR("Daniel Borkmann <daniel@iogearbox.net>");
981 MODULE_AUTHOR("Nikolay Aleksandrov <razor@blackwall.org>");
982 MODULE_LICENSE("GPL");
983 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
984