xref: /openbmc/linux/net/openvswitch/actions.c (revision ee8a99bd) 
1 /*
2  * Copyright (c) 2007-2012 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/skbuff.h>
22 #include <linux/in.h>
23 #include <linux/ip.h>
24 #include <linux/openvswitch.h>
25 #include <linux/tcp.h>
26 #include <linux/udp.h>
27 #include <linux/in6.h>
28 #include <linux/if_arp.h>
29 #include <linux/if_vlan.h>
30 #include <net/ip.h>
31 #include <net/ipv6.h>
32 #include <net/checksum.h>
33 #include <net/dsfield.h>
34 
35 #include "datapath.h"
36 #include "vport.h"
37 
38 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
39 			const struct nlattr *attr, int len, bool keep_skb);
40 
41 static int make_writable(struct sk_buff *skb, int write_len)
42 {
43 	if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
44 		return 0;
45 
46 	return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
47 }
48 
49 /* remove VLAN header from packet and update csum accordingly. */
50 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
51 {
52 	struct vlan_hdr *vhdr;
53 	int err;
54 
55 	err = make_writable(skb, VLAN_ETH_HLEN);
56 	if (unlikely(err))
57 		return err;
58 
59 	if (skb->ip_summed == CHECKSUM_COMPLETE)
60 		skb->csum = csum_sub(skb->csum, csum_partial(skb->data
61 					+ (2 * ETH_ALEN), VLAN_HLEN, 0));
62 
63 	vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
64 	*current_tci = vhdr->h_vlan_TCI;
65 
66 	memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
67 	__skb_pull(skb, VLAN_HLEN);
68 
69 	vlan_set_encap_proto(skb, vhdr);
70 	skb->mac_header += VLAN_HLEN;
71 	skb_reset_mac_len(skb);
72 
73 	return 0;
74 }
75 
76 static int pop_vlan(struct sk_buff *skb)
77 {
78 	__be16 tci;
79 	int err;
80 
81 	if (likely(vlan_tx_tag_present(skb))) {
82 		skb->vlan_tci = 0;
83 	} else {
84 		if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
85 			     skb->len < VLAN_ETH_HLEN))
86 			return 0;
87 
88 		err = __pop_vlan_tci(skb, &tci);
89 		if (err)
90 			return err;
91 	}
92 	/* move next vlan tag to hw accel tag */
93 	if (likely(skb->protocol != htons(ETH_P_8021Q) ||
94 		   skb->len < VLAN_ETH_HLEN))
95 		return 0;
96 
97 	err = __pop_vlan_tci(skb, &tci);
98 	if (unlikely(err))
99 		return err;
100 
101 	__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
102 	return 0;
103 }
104 
105 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
106 {
107 	if (unlikely(vlan_tx_tag_present(skb))) {
108 		u16 current_tag;
109 
110 		/* push down current VLAN tag */
111 		current_tag = vlan_tx_tag_get(skb);
112 
113 		if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
114 			return -ENOMEM;
115 
116 		if (skb->ip_summed == CHECKSUM_COMPLETE)
117 			skb->csum = csum_add(skb->csum, csum_partial(skb->data
118 					+ (2 * ETH_ALEN), VLAN_HLEN, 0));
119 
120 	}
121 	__vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
122 	return 0;
123 }
124 
125 static int set_eth_addr(struct sk_buff *skb,
126 			const struct ovs_key_ethernet *eth_key)
127 {
128 	int err;
129 	err = make_writable(skb, ETH_HLEN);
130 	if (unlikely(err))
131 		return err;
132 
133 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
134 
135 	memcpy(eth_hdr(skb)->h_source, eth_key->eth_src, ETH_ALEN);
136 	memcpy(eth_hdr(skb)->h_dest, eth_key->eth_dst, ETH_ALEN);
137 
138 	ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
139 
140 	return 0;
141 }
142 
143 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
144 				__be32 *addr, __be32 new_addr)
145 {
146 	int transport_len = skb->len - skb_transport_offset(skb);
147 
148 	if (nh->protocol == IPPROTO_TCP) {
149 		if (likely(transport_len >= sizeof(struct tcphdr)))
150 			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
151 						 *addr, new_addr, 1);
152 	} else if (nh->protocol == IPPROTO_UDP) {
153 		if (likely(transport_len >= sizeof(struct udphdr))) {
154 			struct udphdr *uh = udp_hdr(skb);
155 
156 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
157 				inet_proto_csum_replace4(&uh->check, skb,
158 							 *addr, new_addr, 1);
159 				if (!uh->check)
160 					uh->check = CSUM_MANGLED_0;
161 			}
162 		}
163 	}
164 
165 	csum_replace4(&nh->check, *addr, new_addr);
166 	skb->rxhash = 0;
167 	*addr = new_addr;
168 }
169 
170 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
171 				 __be32 addr[4], const __be32 new_addr[4])
172 {
173 	int transport_len = skb->len - skb_transport_offset(skb);
174 
175 	if (l4_proto == IPPROTO_TCP) {
176 		if (likely(transport_len >= sizeof(struct tcphdr)))
177 			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
178 						  addr, new_addr, 1);
179 	} else if (l4_proto == IPPROTO_UDP) {
180 		if (likely(transport_len >= sizeof(struct udphdr))) {
181 			struct udphdr *uh = udp_hdr(skb);
182 
183 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
184 				inet_proto_csum_replace16(&uh->check, skb,
185 							  addr, new_addr, 1);
186 				if (!uh->check)
187 					uh->check = CSUM_MANGLED_0;
188 			}
189 		}
190 	}
191 }
192 
193 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
194 			  __be32 addr[4], const __be32 new_addr[4],
195 			  bool recalculate_csum)
196 {
197 	if (recalculate_csum)
198 		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
199 
200 	skb->rxhash = 0;
201 	memcpy(addr, new_addr, sizeof(__be32[4]));
202 }
203 
204 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
205 {
206 	nh->priority = tc >> 4;
207 	nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
208 }
209 
210 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
211 {
212 	nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
213 	nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
214 	nh->flow_lbl[2] = fl & 0x000000FF;
215 }
216 
217 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
218 {
219 	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
220 	nh->ttl = new_ttl;
221 }
222 
223 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
224 {
225 	struct iphdr *nh;
226 	int err;
227 
228 	err = make_writable(skb, skb_network_offset(skb) +
229 				 sizeof(struct iphdr));
230 	if (unlikely(err))
231 		return err;
232 
233 	nh = ip_hdr(skb);
234 
235 	if (ipv4_key->ipv4_src != nh->saddr)
236 		set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
237 
238 	if (ipv4_key->ipv4_dst != nh->daddr)
239 		set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
240 
241 	if (ipv4_key->ipv4_tos != nh->tos)
242 		ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
243 
244 	if (ipv4_key->ipv4_ttl != nh->ttl)
245 		set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
246 
247 	return 0;
248 }
249 
250 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
251 {
252 	struct ipv6hdr *nh;
253 	int err;
254 	__be32 *saddr;
255 	__be32 *daddr;
256 
257 	err = make_writable(skb, skb_network_offset(skb) +
258 			    sizeof(struct ipv6hdr));
259 	if (unlikely(err))
260 		return err;
261 
262 	nh = ipv6_hdr(skb);
263 	saddr = (__be32 *)&nh->saddr;
264 	daddr = (__be32 *)&nh->daddr;
265 
266 	if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
267 		set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
268 			      ipv6_key->ipv6_src, true);
269 
270 	if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
271 		unsigned int offset = 0;
272 		int flags = IP6_FH_F_SKIP_RH;
273 		bool recalc_csum = true;
274 
275 		if (ipv6_ext_hdr(nh->nexthdr))
276 			recalc_csum = ipv6_find_hdr(skb, &offset,
277 						    NEXTHDR_ROUTING, NULL,
278 						    &flags) != NEXTHDR_ROUTING;
279 
280 		set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
281 			      ipv6_key->ipv6_dst, recalc_csum);
282 	}
283 
284 	set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
285 	set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
286 	nh->hop_limit = ipv6_key->ipv6_hlimit;
287 
288 	return 0;
289 }
290 
291 /* Must follow make_writable() since that can move the skb data. */
292 static void set_tp_port(struct sk_buff *skb, __be16 *port,
293 			 __be16 new_port, __sum16 *check)
294 {
295 	inet_proto_csum_replace2(check, skb, *port, new_port, 0);
296 	*port = new_port;
297 	skb->rxhash = 0;
298 }
299 
300 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
301 {
302 	struct udphdr *uh = udp_hdr(skb);
303 
304 	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
305 		set_tp_port(skb, port, new_port, &uh->check);
306 
307 		if (!uh->check)
308 			uh->check = CSUM_MANGLED_0;
309 	} else {
310 		*port = new_port;
311 		skb->rxhash = 0;
312 	}
313 }
314 
315 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
316 {
317 	struct udphdr *uh;
318 	int err;
319 
320 	err = make_writable(skb, skb_transport_offset(skb) +
321 				 sizeof(struct udphdr));
322 	if (unlikely(err))
323 		return err;
324 
325 	uh = udp_hdr(skb);
326 	if (udp_port_key->udp_src != uh->source)
327 		set_udp_port(skb, &uh->source, udp_port_key->udp_src);
328 
329 	if (udp_port_key->udp_dst != uh->dest)
330 		set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
331 
332 	return 0;
333 }
334 
335 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
336 {
337 	struct tcphdr *th;
338 	int err;
339 
340 	err = make_writable(skb, skb_transport_offset(skb) +
341 				 sizeof(struct tcphdr));
342 	if (unlikely(err))
343 		return err;
344 
345 	th = tcp_hdr(skb);
346 	if (tcp_port_key->tcp_src != th->source)
347 		set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
348 
349 	if (tcp_port_key->tcp_dst != th->dest)
350 		set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
351 
352 	return 0;
353 }
354 
355 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
356 {
357 	struct vport *vport;
358 
359 	if (unlikely(!skb))
360 		return -ENOMEM;
361 
362 	vport = ovs_vport_rcu(dp, out_port);
363 	if (unlikely(!vport)) {
364 		kfree_skb(skb);
365 		return -ENODEV;
366 	}
367 
368 	ovs_vport_send(vport, skb);
369 	return 0;
370 }
371 
372 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
373 			    const struct nlattr *attr)
374 {
375 	struct dp_upcall_info upcall;
376 	const struct nlattr *a;
377 	int rem;
378 
379 	upcall.cmd = OVS_PACKET_CMD_ACTION;
380 	upcall.key = &OVS_CB(skb)->flow->key;
381 	upcall.userdata = NULL;
382 	upcall.portid = 0;
383 
384 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
385 		 a = nla_next(a, &rem)) {
386 		switch (nla_type(a)) {
387 		case OVS_USERSPACE_ATTR_USERDATA:
388 			upcall.userdata = a;
389 			break;
390 
391 		case OVS_USERSPACE_ATTR_PID:
392 			upcall.portid = nla_get_u32(a);
393 			break;
394 		}
395 	}
396 
397 	return ovs_dp_upcall(dp, skb, &upcall);
398 }
399 
400 static int sample(struct datapath *dp, struct sk_buff *skb,
401 		  const struct nlattr *attr)
402 {
403 	const struct nlattr *acts_list = NULL;
404 	const struct nlattr *a;
405 	int rem;
406 
407 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
408 		 a = nla_next(a, &rem)) {
409 		switch (nla_type(a)) {
410 		case OVS_SAMPLE_ATTR_PROBABILITY:
411 			if (net_random() >= nla_get_u32(a))
412 				return 0;
413 			break;
414 
415 		case OVS_SAMPLE_ATTR_ACTIONS:
416 			acts_list = a;
417 			break;
418 		}
419 	}
420 
421 	return do_execute_actions(dp, skb, nla_data(acts_list),
422 						 nla_len(acts_list), true);
423 }
424 
425 static int execute_set_action(struct sk_buff *skb,
426 				 const struct nlattr *nested_attr)
427 {
428 	int err = 0;
429 
430 	switch (nla_type(nested_attr)) {
431 	case OVS_KEY_ATTR_PRIORITY:
432 		skb->priority = nla_get_u32(nested_attr);
433 		break;
434 
435 	case OVS_KEY_ATTR_SKB_MARK:
436 		skb->mark = nla_get_u32(nested_attr);
437 		break;
438 
439 	case OVS_KEY_ATTR_IPV4_TUNNEL:
440 		OVS_CB(skb)->tun_key = nla_data(nested_attr);
441 		break;
442 
443 	case OVS_KEY_ATTR_ETHERNET:
444 		err = set_eth_addr(skb, nla_data(nested_attr));
445 		break;
446 
447 	case OVS_KEY_ATTR_IPV4:
448 		err = set_ipv4(skb, nla_data(nested_attr));
449 		break;
450 
451 	case OVS_KEY_ATTR_IPV6:
452 		err = set_ipv6(skb, nla_data(nested_attr));
453 		break;
454 
455 	case OVS_KEY_ATTR_TCP:
456 		err = set_tcp(skb, nla_data(nested_attr));
457 		break;
458 
459 	case OVS_KEY_ATTR_UDP:
460 		err = set_udp(skb, nla_data(nested_attr));
461 		break;
462 	}
463 
464 	return err;
465 }
466 
467 /* Execute a list of actions against 'skb'. */
468 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
469 			const struct nlattr *attr, int len, bool keep_skb)
470 {
471 	/* Every output action needs a separate clone of 'skb', but the common
472 	 * case is just a single output action, so that doing a clone and
473 	 * then freeing the original skbuff is wasteful.  So the following code
474 	 * is slightly obscure just to avoid that. */
475 	int prev_port = -1;
476 	const struct nlattr *a;
477 	int rem;
478 
479 	for (a = attr, rem = len; rem > 0;
480 	     a = nla_next(a, &rem)) {
481 		int err = 0;
482 
483 		if (prev_port != -1) {
484 			do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
485 			prev_port = -1;
486 		}
487 
488 		switch (nla_type(a)) {
489 		case OVS_ACTION_ATTR_OUTPUT:
490 			prev_port = nla_get_u32(a);
491 			break;
492 
493 		case OVS_ACTION_ATTR_USERSPACE:
494 			output_userspace(dp, skb, a);
495 			break;
496 
497 		case OVS_ACTION_ATTR_PUSH_VLAN:
498 			err = push_vlan(skb, nla_data(a));
499 			if (unlikely(err)) /* skb already freed. */
500 				return err;
501 			break;
502 
503 		case OVS_ACTION_ATTR_POP_VLAN:
504 			err = pop_vlan(skb);
505 			break;
506 
507 		case OVS_ACTION_ATTR_SET:
508 			err = execute_set_action(skb, nla_data(a));
509 			break;
510 
511 		case OVS_ACTION_ATTR_SAMPLE:
512 			err = sample(dp, skb, a);
513 			break;
514 		}
515 
516 		if (unlikely(err)) {
517 			kfree_skb(skb);
518 			return err;
519 		}
520 	}
521 
522 	if (prev_port != -1) {
523 		if (keep_skb)
524 			skb = skb_clone(skb, GFP_ATOMIC);
525 
526 		do_output(dp, skb, prev_port);
527 	} else if (!keep_skb)
528 		consume_skb(skb);
529 
530 	return 0;
531 }
532 
533 /* Execute a list of actions against 'skb'. */
534 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
535 {
536 	struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
537 
538 	OVS_CB(skb)->tun_key = NULL;
539 	return do_execute_actions(dp, skb, acts->actions,
540 					 acts->actions_len, false);
541 }
542