xref: /linux/net/openvswitch/actions.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 /*
2  * Copyright (c) 2007-2013 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/sctp.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/in6.h>
29 #include <linux/if_arp.h>
30 #include <linux/if_vlan.h>
31 #include <net/ip.h>
32 #include <net/ipv6.h>
33 #include <net/checksum.h>
34 #include <net/dsfield.h>
35 #include <net/sctp/checksum.h>
36 
37 #include "datapath.h"
38 #include "vport.h"
39 
40 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
41 			      const struct nlattr *attr, int len);
42 
43 static int make_writable(struct sk_buff *skb, int write_len)
44 {
45 	if (!pskb_may_pull(skb, write_len))
46 		return -ENOMEM;
47 
48 	if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
49 		return 0;
50 
51 	return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
52 }
53 
54 /* remove VLAN header from packet and update csum accordingly. */
55 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
56 {
57 	struct vlan_hdr *vhdr;
58 	int err;
59 
60 	err = make_writable(skb, VLAN_ETH_HLEN);
61 	if (unlikely(err))
62 		return err;
63 
64 	if (skb->ip_summed == CHECKSUM_COMPLETE)
65 		skb->csum = csum_sub(skb->csum, csum_partial(skb->data
66 					+ (2 * ETH_ALEN), VLAN_HLEN, 0));
67 
68 	vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
69 	*current_tci = vhdr->h_vlan_TCI;
70 
71 	memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
72 	__skb_pull(skb, VLAN_HLEN);
73 
74 	vlan_set_encap_proto(skb, vhdr);
75 	skb->mac_header += VLAN_HLEN;
76 	if (skb_network_offset(skb) < ETH_HLEN)
77 		skb_set_network_header(skb, ETH_HLEN);
78 	skb_reset_mac_len(skb);
79 
80 	return 0;
81 }
82 
83 static int pop_vlan(struct sk_buff *skb)
84 {
85 	__be16 tci;
86 	int err;
87 
88 	if (likely(vlan_tx_tag_present(skb))) {
89 		skb->vlan_tci = 0;
90 	} else {
91 		if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
92 			     skb->len < VLAN_ETH_HLEN))
93 			return 0;
94 
95 		err = __pop_vlan_tci(skb, &tci);
96 		if (err)
97 			return err;
98 	}
99 	/* move next vlan tag to hw accel tag */
100 	if (likely(skb->protocol != htons(ETH_P_8021Q) ||
101 		   skb->len < VLAN_ETH_HLEN))
102 		return 0;
103 
104 	err = __pop_vlan_tci(skb, &tci);
105 	if (unlikely(err))
106 		return err;
107 
108 	__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(tci));
109 	return 0;
110 }
111 
112 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
113 {
114 	if (unlikely(vlan_tx_tag_present(skb))) {
115 		u16 current_tag;
116 
117 		/* push down current VLAN tag */
118 		current_tag = vlan_tx_tag_get(skb);
119 
120 		if (!__vlan_put_tag(skb, skb->vlan_proto, current_tag))
121 			return -ENOMEM;
122 
123 		if (skb->ip_summed == CHECKSUM_COMPLETE)
124 			skb->csum = csum_add(skb->csum, csum_partial(skb->data
125 					+ (2 * ETH_ALEN), VLAN_HLEN, 0));
126 
127 	}
128 	__vlan_hwaccel_put_tag(skb, vlan->vlan_tpid, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
129 	return 0;
130 }
131 
132 static int set_eth_addr(struct sk_buff *skb,
133 			const struct ovs_key_ethernet *eth_key)
134 {
135 	int err;
136 	err = make_writable(skb, ETH_HLEN);
137 	if (unlikely(err))
138 		return err;
139 
140 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
141 
142 	ether_addr_copy(eth_hdr(skb)->h_source, eth_key->eth_src);
143 	ether_addr_copy(eth_hdr(skb)->h_dest, eth_key->eth_dst);
144 
145 	ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
146 
147 	return 0;
148 }
149 
150 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
151 				__be32 *addr, __be32 new_addr)
152 {
153 	int transport_len = skb->len - skb_transport_offset(skb);
154 
155 	if (nh->protocol == IPPROTO_TCP) {
156 		if (likely(transport_len >= sizeof(struct tcphdr)))
157 			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
158 						 *addr, new_addr, 1);
159 	} else if (nh->protocol == IPPROTO_UDP) {
160 		if (likely(transport_len >= sizeof(struct udphdr))) {
161 			struct udphdr *uh = udp_hdr(skb);
162 
163 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
164 				inet_proto_csum_replace4(&uh->check, skb,
165 							 *addr, new_addr, 1);
166 				if (!uh->check)
167 					uh->check = CSUM_MANGLED_0;
168 			}
169 		}
170 	}
171 
172 	csum_replace4(&nh->check, *addr, new_addr);
173 	skb_clear_hash(skb);
174 	*addr = new_addr;
175 }
176 
177 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
178 				 __be32 addr[4], const __be32 new_addr[4])
179 {
180 	int transport_len = skb->len - skb_transport_offset(skb);
181 
182 	if (l4_proto == IPPROTO_TCP) {
183 		if (likely(transport_len >= sizeof(struct tcphdr)))
184 			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
185 						  addr, new_addr, 1);
186 	} else if (l4_proto == IPPROTO_UDP) {
187 		if (likely(transport_len >= sizeof(struct udphdr))) {
188 			struct udphdr *uh = udp_hdr(skb);
189 
190 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
191 				inet_proto_csum_replace16(&uh->check, skb,
192 							  addr, new_addr, 1);
193 				if (!uh->check)
194 					uh->check = CSUM_MANGLED_0;
195 			}
196 		}
197 	}
198 }
199 
200 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
201 			  __be32 addr[4], const __be32 new_addr[4],
202 			  bool recalculate_csum)
203 {
204 	if (recalculate_csum)
205 		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
206 
207 	skb_clear_hash(skb);
208 	memcpy(addr, new_addr, sizeof(__be32[4]));
209 }
210 
211 static void set_ipv6_tc(struct ipv6hdr *nh, u8 tc)
212 {
213 	nh->priority = tc >> 4;
214 	nh->flow_lbl[0] = (nh->flow_lbl[0] & 0x0F) | ((tc & 0x0F) << 4);
215 }
216 
217 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl)
218 {
219 	nh->flow_lbl[0] = (nh->flow_lbl[0] & 0xF0) | (fl & 0x000F0000) >> 16;
220 	nh->flow_lbl[1] = (fl & 0x0000FF00) >> 8;
221 	nh->flow_lbl[2] = fl & 0x000000FF;
222 }
223 
224 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
225 {
226 	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
227 	nh->ttl = new_ttl;
228 }
229 
230 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
231 {
232 	struct iphdr *nh;
233 	int err;
234 
235 	err = make_writable(skb, skb_network_offset(skb) +
236 				 sizeof(struct iphdr));
237 	if (unlikely(err))
238 		return err;
239 
240 	nh = ip_hdr(skb);
241 
242 	if (ipv4_key->ipv4_src != nh->saddr)
243 		set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
244 
245 	if (ipv4_key->ipv4_dst != nh->daddr)
246 		set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
247 
248 	if (ipv4_key->ipv4_tos != nh->tos)
249 		ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
250 
251 	if (ipv4_key->ipv4_ttl != nh->ttl)
252 		set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
253 
254 	return 0;
255 }
256 
257 static int set_ipv6(struct sk_buff *skb, const struct ovs_key_ipv6 *ipv6_key)
258 {
259 	struct ipv6hdr *nh;
260 	int err;
261 	__be32 *saddr;
262 	__be32 *daddr;
263 
264 	err = make_writable(skb, skb_network_offset(skb) +
265 			    sizeof(struct ipv6hdr));
266 	if (unlikely(err))
267 		return err;
268 
269 	nh = ipv6_hdr(skb);
270 	saddr = (__be32 *)&nh->saddr;
271 	daddr = (__be32 *)&nh->daddr;
272 
273 	if (memcmp(ipv6_key->ipv6_src, saddr, sizeof(ipv6_key->ipv6_src)))
274 		set_ipv6_addr(skb, ipv6_key->ipv6_proto, saddr,
275 			      ipv6_key->ipv6_src, true);
276 
277 	if (memcmp(ipv6_key->ipv6_dst, daddr, sizeof(ipv6_key->ipv6_dst))) {
278 		unsigned int offset = 0;
279 		int flags = IP6_FH_F_SKIP_RH;
280 		bool recalc_csum = true;
281 
282 		if (ipv6_ext_hdr(nh->nexthdr))
283 			recalc_csum = ipv6_find_hdr(skb, &offset,
284 						    NEXTHDR_ROUTING, NULL,
285 						    &flags) != NEXTHDR_ROUTING;
286 
287 		set_ipv6_addr(skb, ipv6_key->ipv6_proto, daddr,
288 			      ipv6_key->ipv6_dst, recalc_csum);
289 	}
290 
291 	set_ipv6_tc(nh, ipv6_key->ipv6_tclass);
292 	set_ipv6_fl(nh, ntohl(ipv6_key->ipv6_label));
293 	nh->hop_limit = ipv6_key->ipv6_hlimit;
294 
295 	return 0;
296 }
297 
298 /* Must follow make_writable() since that can move the skb data. */
299 static void set_tp_port(struct sk_buff *skb, __be16 *port,
300 			 __be16 new_port, __sum16 *check)
301 {
302 	inet_proto_csum_replace2(check, skb, *port, new_port, 0);
303 	*port = new_port;
304 	skb_clear_hash(skb);
305 }
306 
307 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
308 {
309 	struct udphdr *uh = udp_hdr(skb);
310 
311 	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
312 		set_tp_port(skb, port, new_port, &uh->check);
313 
314 		if (!uh->check)
315 			uh->check = CSUM_MANGLED_0;
316 	} else {
317 		*port = new_port;
318 		skb_clear_hash(skb);
319 	}
320 }
321 
322 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
323 {
324 	struct udphdr *uh;
325 	int err;
326 
327 	err = make_writable(skb, skb_transport_offset(skb) +
328 				 sizeof(struct udphdr));
329 	if (unlikely(err))
330 		return err;
331 
332 	uh = udp_hdr(skb);
333 	if (udp_port_key->udp_src != uh->source)
334 		set_udp_port(skb, &uh->source, udp_port_key->udp_src);
335 
336 	if (udp_port_key->udp_dst != uh->dest)
337 		set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
338 
339 	return 0;
340 }
341 
342 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
343 {
344 	struct tcphdr *th;
345 	int err;
346 
347 	err = make_writable(skb, skb_transport_offset(skb) +
348 				 sizeof(struct tcphdr));
349 	if (unlikely(err))
350 		return err;
351 
352 	th = tcp_hdr(skb);
353 	if (tcp_port_key->tcp_src != th->source)
354 		set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
355 
356 	if (tcp_port_key->tcp_dst != th->dest)
357 		set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
358 
359 	return 0;
360 }
361 
362 static int set_sctp(struct sk_buff *skb,
363 		     const struct ovs_key_sctp *sctp_port_key)
364 {
365 	struct sctphdr *sh;
366 	int err;
367 	unsigned int sctphoff = skb_transport_offset(skb);
368 
369 	err = make_writable(skb, sctphoff + sizeof(struct sctphdr));
370 	if (unlikely(err))
371 		return err;
372 
373 	sh = sctp_hdr(skb);
374 	if (sctp_port_key->sctp_src != sh->source ||
375 	    sctp_port_key->sctp_dst != sh->dest) {
376 		__le32 old_correct_csum, new_csum, old_csum;
377 
378 		old_csum = sh->checksum;
379 		old_correct_csum = sctp_compute_cksum(skb, sctphoff);
380 
381 		sh->source = sctp_port_key->sctp_src;
382 		sh->dest = sctp_port_key->sctp_dst;
383 
384 		new_csum = sctp_compute_cksum(skb, sctphoff);
385 
386 		/* Carry any checksum errors through. */
387 		sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
388 
389 		skb_clear_hash(skb);
390 	}
391 
392 	return 0;
393 }
394 
395 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
396 {
397 	struct vport *vport;
398 
399 	if (unlikely(!skb))
400 		return -ENOMEM;
401 
402 	vport = ovs_vport_rcu(dp, out_port);
403 	if (unlikely(!vport)) {
404 		kfree_skb(skb);
405 		return -ENODEV;
406 	}
407 
408 	ovs_vport_send(vport, skb);
409 	return 0;
410 }
411 
412 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
413 			    const struct nlattr *attr)
414 {
415 	struct dp_upcall_info upcall;
416 	const struct nlattr *a;
417 	int rem;
418 
419 	BUG_ON(!OVS_CB(skb)->pkt_key);
420 
421 	upcall.cmd = OVS_PACKET_CMD_ACTION;
422 	upcall.key = OVS_CB(skb)->pkt_key;
423 	upcall.userdata = NULL;
424 	upcall.portid = 0;
425 
426 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
427 		 a = nla_next(a, &rem)) {
428 		switch (nla_type(a)) {
429 		case OVS_USERSPACE_ATTR_USERDATA:
430 			upcall.userdata = a;
431 			break;
432 
433 		case OVS_USERSPACE_ATTR_PID:
434 			upcall.portid = nla_get_u32(a);
435 			break;
436 		}
437 	}
438 
439 	return ovs_dp_upcall(dp, skb, &upcall);
440 }
441 
442 static bool last_action(const struct nlattr *a, int rem)
443 {
444 	return a->nla_len == rem;
445 }
446 
447 static int sample(struct datapath *dp, struct sk_buff *skb,
448 		  const struct nlattr *attr)
449 {
450 	const struct nlattr *acts_list = NULL;
451 	const struct nlattr *a;
452 	struct sk_buff *sample_skb;
453 	int rem;
454 
455 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
456 		 a = nla_next(a, &rem)) {
457 		switch (nla_type(a)) {
458 		case OVS_SAMPLE_ATTR_PROBABILITY:
459 			if (prandom_u32() >= nla_get_u32(a))
460 				return 0;
461 			break;
462 
463 		case OVS_SAMPLE_ATTR_ACTIONS:
464 			acts_list = a;
465 			break;
466 		}
467 	}
468 
469 	rem = nla_len(acts_list);
470 	a = nla_data(acts_list);
471 
472 	/* Actions list is either empty or only contains a single user-space
473 	 * action, the latter being a special case as it is the only known
474 	 * usage of the sample action.
475 	 * In these special cases don't clone the skb as there are no
476 	 * side-effects in the nested actions.
477 	 * Otherwise, clone in case the nested actions have side effects.
478 	 */
479 	if (likely(rem == 0 || (nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
480 				last_action(a, rem)))) {
481 		sample_skb = skb;
482 		skb_get(skb);
483 	} else {
484 		sample_skb = skb_clone(skb, GFP_ATOMIC);
485 		if (!sample_skb) /* Skip sample action when out of memory. */
486 			return 0;
487 	}
488 
489 	/* Note that do_execute_actions() never consumes skb.
490 	 * In the case where skb has been cloned above it is the clone that
491 	 * is consumed.  Otherwise the skb_get(skb) call prevents
492 	 * consumption by do_execute_actions(). Thus, it is safe to simply
493 	 * return the error code and let the caller (also
494 	 * do_execute_actions()) free skb on error.
495 	 */
496 	return do_execute_actions(dp, sample_skb, a, rem);
497 }
498 
499 static int execute_set_action(struct sk_buff *skb,
500 				 const struct nlattr *nested_attr)
501 {
502 	int err = 0;
503 
504 	switch (nla_type(nested_attr)) {
505 	case OVS_KEY_ATTR_PRIORITY:
506 		skb->priority = nla_get_u32(nested_attr);
507 		break;
508 
509 	case OVS_KEY_ATTR_SKB_MARK:
510 		skb->mark = nla_get_u32(nested_attr);
511 		break;
512 
513 	case OVS_KEY_ATTR_IPV4_TUNNEL:
514 		OVS_CB(skb)->tun_key = nla_data(nested_attr);
515 		break;
516 
517 	case OVS_KEY_ATTR_ETHERNET:
518 		err = set_eth_addr(skb, nla_data(nested_attr));
519 		break;
520 
521 	case OVS_KEY_ATTR_IPV4:
522 		err = set_ipv4(skb, nla_data(nested_attr));
523 		break;
524 
525 	case OVS_KEY_ATTR_IPV6:
526 		err = set_ipv6(skb, nla_data(nested_attr));
527 		break;
528 
529 	case OVS_KEY_ATTR_TCP:
530 		err = set_tcp(skb, nla_data(nested_attr));
531 		break;
532 
533 	case OVS_KEY_ATTR_UDP:
534 		err = set_udp(skb, nla_data(nested_attr));
535 		break;
536 
537 	case OVS_KEY_ATTR_SCTP:
538 		err = set_sctp(skb, nla_data(nested_attr));
539 		break;
540 	}
541 
542 	return err;
543 }
544 
545 /* Execute a list of actions against 'skb'. */
546 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
547 			      const struct nlattr *attr, int len)
548 {
549 	/* Every output action needs a separate clone of 'skb', but the common
550 	 * case is just a single output action, so that doing a clone and
551 	 * then freeing the original skbuff is wasteful.  So the following code
552 	 * is slightly obscure just to avoid that. */
553 	int prev_port = -1;
554 	const struct nlattr *a;
555 	int rem;
556 
557 	for (a = attr, rem = len; rem > 0;
558 	     a = nla_next(a, &rem)) {
559 		int err = 0;
560 
561 		if (prev_port != -1) {
562 			do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
563 			prev_port = -1;
564 		}
565 
566 		switch (nla_type(a)) {
567 		case OVS_ACTION_ATTR_OUTPUT:
568 			prev_port = nla_get_u32(a);
569 			break;
570 
571 		case OVS_ACTION_ATTR_USERSPACE:
572 			output_userspace(dp, skb, a);
573 			break;
574 
575 		case OVS_ACTION_ATTR_PUSH_VLAN:
576 			err = push_vlan(skb, nla_data(a));
577 			if (unlikely(err)) /* skb already freed. */
578 				return err;
579 			break;
580 
581 		case OVS_ACTION_ATTR_POP_VLAN:
582 			err = pop_vlan(skb);
583 			break;
584 
585 		case OVS_ACTION_ATTR_SET:
586 			err = execute_set_action(skb, nla_data(a));
587 			break;
588 
589 		case OVS_ACTION_ATTR_SAMPLE:
590 			err = sample(dp, skb, a);
591 			if (unlikely(err)) /* skb already freed. */
592 				return err;
593 			break;
594 		}
595 
596 		if (unlikely(err)) {
597 			kfree_skb(skb);
598 			return err;
599 		}
600 	}
601 
602 	if (prev_port != -1)
603 		do_output(dp, skb, prev_port);
604 	else
605 		consume_skb(skb);
606 
607 	return 0;
608 }
609 
610 /* Execute a list of actions against 'skb'. */
611 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
612 {
613 	struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
614 
615 	OVS_CB(skb)->tun_key = NULL;
616 	return do_execute_actions(dp, skb, acts->actions, acts->actions_len);
617 }
618