xref: /openbmc/linux/net/openvswitch/actions.c (revision 6dfcd296)
1 /*
2  * Copyright (c) 2007-2014 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/netfilter_ipv6.h>
26 #include <linux/sctp.h>
27 #include <linux/tcp.h>
28 #include <linux/udp.h>
29 #include <linux/in6.h>
30 #include <linux/if_arp.h>
31 #include <linux/if_vlan.h>
32 
33 #include <net/dst.h>
34 #include <net/ip.h>
35 #include <net/ipv6.h>
36 #include <net/ip6_fib.h>
37 #include <net/checksum.h>
38 #include <net/dsfield.h>
39 #include <net/mpls.h>
40 #include <net/sctp/checksum.h>
41 
42 #include "datapath.h"
43 #include "flow.h"
44 #include "conntrack.h"
45 #include "vport.h"
46 
47 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
48 			      struct sw_flow_key *key,
49 			      const struct nlattr *attr, int len);
50 
51 struct deferred_action {
52 	struct sk_buff *skb;
53 	const struct nlattr *actions;
54 
55 	/* Store pkt_key clone when creating deferred action. */
56 	struct sw_flow_key pkt_key;
57 };
58 
59 #define MAX_L2_LEN	(VLAN_ETH_HLEN + 3 * MPLS_HLEN)
60 struct ovs_frag_data {
61 	unsigned long dst;
62 	struct vport *vport;
63 	struct ovs_skb_cb cb;
64 	__be16 inner_protocol;
65 	__u16 vlan_tci;
66 	__be16 vlan_proto;
67 	unsigned int l2_len;
68 	u8 l2_data[MAX_L2_LEN];
69 };
70 
71 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
72 
73 #define DEFERRED_ACTION_FIFO_SIZE 10
74 #define OVS_RECURSION_LIMIT 5
75 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
76 struct action_fifo {
77 	int head;
78 	int tail;
79 	/* Deferred action fifo queue storage. */
80 	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
81 };
82 
83 struct recirc_keys {
84 	struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
85 };
86 
87 static struct action_fifo __percpu *action_fifos;
88 static struct recirc_keys __percpu *recirc_keys;
89 static DEFINE_PER_CPU(int, exec_actions_level);
90 
91 static void action_fifo_init(struct action_fifo *fifo)
92 {
93 	fifo->head = 0;
94 	fifo->tail = 0;
95 }
96 
97 static bool action_fifo_is_empty(const struct action_fifo *fifo)
98 {
99 	return (fifo->head == fifo->tail);
100 }
101 
102 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
103 {
104 	if (action_fifo_is_empty(fifo))
105 		return NULL;
106 
107 	return &fifo->fifo[fifo->tail++];
108 }
109 
110 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
111 {
112 	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
113 		return NULL;
114 
115 	return &fifo->fifo[fifo->head++];
116 }
117 
118 /* Return true if fifo is not full */
119 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
120 						    const struct sw_flow_key *key,
121 						    const struct nlattr *attr)
122 {
123 	struct action_fifo *fifo;
124 	struct deferred_action *da;
125 
126 	fifo = this_cpu_ptr(action_fifos);
127 	da = action_fifo_put(fifo);
128 	if (da) {
129 		da->skb = skb;
130 		da->actions = attr;
131 		da->pkt_key = *key;
132 	}
133 
134 	return da;
135 }
136 
137 static void invalidate_flow_key(struct sw_flow_key *key)
138 {
139 	key->eth.type = htons(0);
140 }
141 
142 static bool is_flow_key_valid(const struct sw_flow_key *key)
143 {
144 	return !!key->eth.type;
145 }
146 
147 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
148 			     __be16 ethertype)
149 {
150 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
151 		__be16 diff[] = { ~(hdr->h_proto), ethertype };
152 
153 		skb->csum = ~csum_partial((char *)diff, sizeof(diff),
154 					~skb->csum);
155 	}
156 
157 	hdr->h_proto = ethertype;
158 }
159 
160 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
161 		     const struct ovs_action_push_mpls *mpls)
162 {
163 	struct mpls_shim_hdr *new_mpls_lse;
164 
165 	/* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
166 	if (skb->encapsulation)
167 		return -ENOTSUPP;
168 
169 	if (skb_cow_head(skb, MPLS_HLEN) < 0)
170 		return -ENOMEM;
171 
172 	if (!skb->inner_protocol) {
173 		skb_set_inner_network_header(skb, skb->mac_len);
174 		skb_set_inner_protocol(skb, skb->protocol);
175 	}
176 
177 	skb_push(skb, MPLS_HLEN);
178 	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
179 		skb->mac_len);
180 	skb_reset_mac_header(skb);
181 	skb_set_network_header(skb, skb->mac_len);
182 
183 	new_mpls_lse = mpls_hdr(skb);
184 	new_mpls_lse->label_stack_entry = mpls->mpls_lse;
185 
186 	skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
187 
188 	update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
189 	skb->protocol = mpls->mpls_ethertype;
190 
191 	invalidate_flow_key(key);
192 	return 0;
193 }
194 
195 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
196 		    const __be16 ethertype)
197 {
198 	struct ethhdr *hdr;
199 	int err;
200 
201 	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
202 	if (unlikely(err))
203 		return err;
204 
205 	skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
206 
207 	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
208 		skb->mac_len);
209 
210 	__skb_pull(skb, MPLS_HLEN);
211 	skb_reset_mac_header(skb);
212 	skb_set_network_header(skb, skb->mac_len);
213 
214 	/* mpls_hdr() is used to locate the ethertype field correctly in the
215 	 * presence of VLAN tags.
216 	 */
217 	hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
218 	update_ethertype(skb, hdr, ethertype);
219 	if (eth_p_mpls(skb->protocol))
220 		skb->protocol = ethertype;
221 
222 	invalidate_flow_key(key);
223 	return 0;
224 }
225 
226 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
227 		    const __be32 *mpls_lse, const __be32 *mask)
228 {
229 	struct mpls_shim_hdr *stack;
230 	__be32 lse;
231 	int err;
232 
233 	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
234 	if (unlikely(err))
235 		return err;
236 
237 	stack = mpls_hdr(skb);
238 	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
239 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
240 		__be32 diff[] = { ~(stack->label_stack_entry), lse };
241 
242 		skb->csum = ~csum_partial((char *)diff, sizeof(diff),
243 					  ~skb->csum);
244 	}
245 
246 	stack->label_stack_entry = lse;
247 	flow_key->mpls.top_lse = lse;
248 	return 0;
249 }
250 
251 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
252 {
253 	int err;
254 
255 	err = skb_vlan_pop(skb);
256 	if (skb_vlan_tag_present(skb)) {
257 		invalidate_flow_key(key);
258 	} else {
259 		key->eth.vlan.tci = 0;
260 		key->eth.vlan.tpid = 0;
261 	}
262 	return err;
263 }
264 
265 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
266 		     const struct ovs_action_push_vlan *vlan)
267 {
268 	if (skb_vlan_tag_present(skb)) {
269 		invalidate_flow_key(key);
270 	} else {
271 		key->eth.vlan.tci = vlan->vlan_tci;
272 		key->eth.vlan.tpid = vlan->vlan_tpid;
273 	}
274 	return skb_vlan_push(skb, vlan->vlan_tpid,
275 			     ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
276 }
277 
278 /* 'src' is already properly masked. */
279 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
280 {
281 	u16 *dst = (u16 *)dst_;
282 	const u16 *src = (const u16 *)src_;
283 	const u16 *mask = (const u16 *)mask_;
284 
285 	OVS_SET_MASKED(dst[0], src[0], mask[0]);
286 	OVS_SET_MASKED(dst[1], src[1], mask[1]);
287 	OVS_SET_MASKED(dst[2], src[2], mask[2]);
288 }
289 
290 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
291 			const struct ovs_key_ethernet *key,
292 			const struct ovs_key_ethernet *mask)
293 {
294 	int err;
295 
296 	err = skb_ensure_writable(skb, ETH_HLEN);
297 	if (unlikely(err))
298 		return err;
299 
300 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
301 
302 	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
303 			       mask->eth_src);
304 	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
305 			       mask->eth_dst);
306 
307 	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
308 
309 	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
310 	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
311 	return 0;
312 }
313 
314 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
315 				  __be32 addr, __be32 new_addr)
316 {
317 	int transport_len = skb->len - skb_transport_offset(skb);
318 
319 	if (nh->frag_off & htons(IP_OFFSET))
320 		return;
321 
322 	if (nh->protocol == IPPROTO_TCP) {
323 		if (likely(transport_len >= sizeof(struct tcphdr)))
324 			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
325 						 addr, new_addr, true);
326 	} else if (nh->protocol == IPPROTO_UDP) {
327 		if (likely(transport_len >= sizeof(struct udphdr))) {
328 			struct udphdr *uh = udp_hdr(skb);
329 
330 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
331 				inet_proto_csum_replace4(&uh->check, skb,
332 							 addr, new_addr, true);
333 				if (!uh->check)
334 					uh->check = CSUM_MANGLED_0;
335 			}
336 		}
337 	}
338 }
339 
340 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
341 			__be32 *addr, __be32 new_addr)
342 {
343 	update_ip_l4_checksum(skb, nh, *addr, new_addr);
344 	csum_replace4(&nh->check, *addr, new_addr);
345 	skb_clear_hash(skb);
346 	*addr = new_addr;
347 }
348 
349 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
350 				 __be32 addr[4], const __be32 new_addr[4])
351 {
352 	int transport_len = skb->len - skb_transport_offset(skb);
353 
354 	if (l4_proto == NEXTHDR_TCP) {
355 		if (likely(transport_len >= sizeof(struct tcphdr)))
356 			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
357 						  addr, new_addr, true);
358 	} else if (l4_proto == NEXTHDR_UDP) {
359 		if (likely(transport_len >= sizeof(struct udphdr))) {
360 			struct udphdr *uh = udp_hdr(skb);
361 
362 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
363 				inet_proto_csum_replace16(&uh->check, skb,
364 							  addr, new_addr, true);
365 				if (!uh->check)
366 					uh->check = CSUM_MANGLED_0;
367 			}
368 		}
369 	} else if (l4_proto == NEXTHDR_ICMP) {
370 		if (likely(transport_len >= sizeof(struct icmp6hdr)))
371 			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
372 						  skb, addr, new_addr, true);
373 	}
374 }
375 
376 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
377 			   const __be32 mask[4], __be32 masked[4])
378 {
379 	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
380 	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
381 	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
382 	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
383 }
384 
385 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
386 			  __be32 addr[4], const __be32 new_addr[4],
387 			  bool recalculate_csum)
388 {
389 	if (recalculate_csum)
390 		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
391 
392 	skb_clear_hash(skb);
393 	memcpy(addr, new_addr, sizeof(__be32[4]));
394 }
395 
396 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
397 {
398 	/* Bits 21-24 are always unmasked, so this retains their values. */
399 	OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
400 	OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
401 	OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
402 }
403 
404 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
405 		       u8 mask)
406 {
407 	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
408 
409 	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
410 	nh->ttl = new_ttl;
411 }
412 
413 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
414 		    const struct ovs_key_ipv4 *key,
415 		    const struct ovs_key_ipv4 *mask)
416 {
417 	struct iphdr *nh;
418 	__be32 new_addr;
419 	int err;
420 
421 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
422 				  sizeof(struct iphdr));
423 	if (unlikely(err))
424 		return err;
425 
426 	nh = ip_hdr(skb);
427 
428 	/* Setting an IP addresses is typically only a side effect of
429 	 * matching on them in the current userspace implementation, so it
430 	 * makes sense to check if the value actually changed.
431 	 */
432 	if (mask->ipv4_src) {
433 		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
434 
435 		if (unlikely(new_addr != nh->saddr)) {
436 			set_ip_addr(skb, nh, &nh->saddr, new_addr);
437 			flow_key->ipv4.addr.src = new_addr;
438 		}
439 	}
440 	if (mask->ipv4_dst) {
441 		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
442 
443 		if (unlikely(new_addr != nh->daddr)) {
444 			set_ip_addr(skb, nh, &nh->daddr, new_addr);
445 			flow_key->ipv4.addr.dst = new_addr;
446 		}
447 	}
448 	if (mask->ipv4_tos) {
449 		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
450 		flow_key->ip.tos = nh->tos;
451 	}
452 	if (mask->ipv4_ttl) {
453 		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
454 		flow_key->ip.ttl = nh->ttl;
455 	}
456 
457 	return 0;
458 }
459 
460 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
461 {
462 	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
463 }
464 
465 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
466 		    const struct ovs_key_ipv6 *key,
467 		    const struct ovs_key_ipv6 *mask)
468 {
469 	struct ipv6hdr *nh;
470 	int err;
471 
472 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
473 				  sizeof(struct ipv6hdr));
474 	if (unlikely(err))
475 		return err;
476 
477 	nh = ipv6_hdr(skb);
478 
479 	/* Setting an IP addresses is typically only a side effect of
480 	 * matching on them in the current userspace implementation, so it
481 	 * makes sense to check if the value actually changed.
482 	 */
483 	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
484 		__be32 *saddr = (__be32 *)&nh->saddr;
485 		__be32 masked[4];
486 
487 		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
488 
489 		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
490 			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
491 				      true);
492 			memcpy(&flow_key->ipv6.addr.src, masked,
493 			       sizeof(flow_key->ipv6.addr.src));
494 		}
495 	}
496 	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
497 		unsigned int offset = 0;
498 		int flags = IP6_FH_F_SKIP_RH;
499 		bool recalc_csum = true;
500 		__be32 *daddr = (__be32 *)&nh->daddr;
501 		__be32 masked[4];
502 
503 		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
504 
505 		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
506 			if (ipv6_ext_hdr(nh->nexthdr))
507 				recalc_csum = (ipv6_find_hdr(skb, &offset,
508 							     NEXTHDR_ROUTING,
509 							     NULL, &flags)
510 					       != NEXTHDR_ROUTING);
511 
512 			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
513 				      recalc_csum);
514 			memcpy(&flow_key->ipv6.addr.dst, masked,
515 			       sizeof(flow_key->ipv6.addr.dst));
516 		}
517 	}
518 	if (mask->ipv6_tclass) {
519 		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
520 		flow_key->ip.tos = ipv6_get_dsfield(nh);
521 	}
522 	if (mask->ipv6_label) {
523 		set_ipv6_fl(nh, ntohl(key->ipv6_label),
524 			    ntohl(mask->ipv6_label));
525 		flow_key->ipv6.label =
526 		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
527 	}
528 	if (mask->ipv6_hlimit) {
529 		OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
530 			       mask->ipv6_hlimit);
531 		flow_key->ip.ttl = nh->hop_limit;
532 	}
533 	return 0;
534 }
535 
536 /* Must follow skb_ensure_writable() since that can move the skb data. */
537 static void set_tp_port(struct sk_buff *skb, __be16 *port,
538 			__be16 new_port, __sum16 *check)
539 {
540 	inet_proto_csum_replace2(check, skb, *port, new_port, false);
541 	*port = new_port;
542 }
543 
544 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
545 		   const struct ovs_key_udp *key,
546 		   const struct ovs_key_udp *mask)
547 {
548 	struct udphdr *uh;
549 	__be16 src, dst;
550 	int err;
551 
552 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
553 				  sizeof(struct udphdr));
554 	if (unlikely(err))
555 		return err;
556 
557 	uh = udp_hdr(skb);
558 	/* Either of the masks is non-zero, so do not bother checking them. */
559 	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
560 	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
561 
562 	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
563 		if (likely(src != uh->source)) {
564 			set_tp_port(skb, &uh->source, src, &uh->check);
565 			flow_key->tp.src = src;
566 		}
567 		if (likely(dst != uh->dest)) {
568 			set_tp_port(skb, &uh->dest, dst, &uh->check);
569 			flow_key->tp.dst = dst;
570 		}
571 
572 		if (unlikely(!uh->check))
573 			uh->check = CSUM_MANGLED_0;
574 	} else {
575 		uh->source = src;
576 		uh->dest = dst;
577 		flow_key->tp.src = src;
578 		flow_key->tp.dst = dst;
579 	}
580 
581 	skb_clear_hash(skb);
582 
583 	return 0;
584 }
585 
586 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
587 		   const struct ovs_key_tcp *key,
588 		   const struct ovs_key_tcp *mask)
589 {
590 	struct tcphdr *th;
591 	__be16 src, dst;
592 	int err;
593 
594 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
595 				  sizeof(struct tcphdr));
596 	if (unlikely(err))
597 		return err;
598 
599 	th = tcp_hdr(skb);
600 	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
601 	if (likely(src != th->source)) {
602 		set_tp_port(skb, &th->source, src, &th->check);
603 		flow_key->tp.src = src;
604 	}
605 	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
606 	if (likely(dst != th->dest)) {
607 		set_tp_port(skb, &th->dest, dst, &th->check);
608 		flow_key->tp.dst = dst;
609 	}
610 	skb_clear_hash(skb);
611 
612 	return 0;
613 }
614 
615 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
616 		    const struct ovs_key_sctp *key,
617 		    const struct ovs_key_sctp *mask)
618 {
619 	unsigned int sctphoff = skb_transport_offset(skb);
620 	struct sctphdr *sh;
621 	__le32 old_correct_csum, new_csum, old_csum;
622 	int err;
623 
624 	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
625 	if (unlikely(err))
626 		return err;
627 
628 	sh = sctp_hdr(skb);
629 	old_csum = sh->checksum;
630 	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
631 
632 	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
633 	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
634 
635 	new_csum = sctp_compute_cksum(skb, sctphoff);
636 
637 	/* Carry any checksum errors through. */
638 	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
639 
640 	skb_clear_hash(skb);
641 	flow_key->tp.src = sh->source;
642 	flow_key->tp.dst = sh->dest;
643 
644 	return 0;
645 }
646 
647 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
648 {
649 	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
650 	struct vport *vport = data->vport;
651 
652 	if (skb_cow_head(skb, data->l2_len) < 0) {
653 		kfree_skb(skb);
654 		return -ENOMEM;
655 	}
656 
657 	__skb_dst_copy(skb, data->dst);
658 	*OVS_CB(skb) = data->cb;
659 	skb->inner_protocol = data->inner_protocol;
660 	skb->vlan_tci = data->vlan_tci;
661 	skb->vlan_proto = data->vlan_proto;
662 
663 	/* Reconstruct the MAC header.  */
664 	skb_push(skb, data->l2_len);
665 	memcpy(skb->data, &data->l2_data, data->l2_len);
666 	skb_postpush_rcsum(skb, skb->data, data->l2_len);
667 	skb_reset_mac_header(skb);
668 
669 	ovs_vport_send(vport, skb);
670 	return 0;
671 }
672 
673 static unsigned int
674 ovs_dst_get_mtu(const struct dst_entry *dst)
675 {
676 	return dst->dev->mtu;
677 }
678 
679 static struct dst_ops ovs_dst_ops = {
680 	.family = AF_UNSPEC,
681 	.mtu = ovs_dst_get_mtu,
682 };
683 
684 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
685  * ovs_vport_output(), which is called once per fragmented packet.
686  */
687 static void prepare_frag(struct vport *vport, struct sk_buff *skb)
688 {
689 	unsigned int hlen = skb_network_offset(skb);
690 	struct ovs_frag_data *data;
691 
692 	data = this_cpu_ptr(&ovs_frag_data_storage);
693 	data->dst = skb->_skb_refdst;
694 	data->vport = vport;
695 	data->cb = *OVS_CB(skb);
696 	data->inner_protocol = skb->inner_protocol;
697 	data->vlan_tci = skb->vlan_tci;
698 	data->vlan_proto = skb->vlan_proto;
699 	data->l2_len = hlen;
700 	memcpy(&data->l2_data, skb->data, hlen);
701 
702 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
703 	skb_pull(skb, hlen);
704 }
705 
706 static void ovs_fragment(struct net *net, struct vport *vport,
707 			 struct sk_buff *skb, u16 mru, __be16 ethertype)
708 {
709 	if (skb_network_offset(skb) > MAX_L2_LEN) {
710 		OVS_NLERR(1, "L2 header too long to fragment");
711 		goto err;
712 	}
713 
714 	if (ethertype == htons(ETH_P_IP)) {
715 		struct dst_entry ovs_dst;
716 		unsigned long orig_dst;
717 
718 		prepare_frag(vport, skb);
719 		dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
720 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
721 		ovs_dst.dev = vport->dev;
722 
723 		orig_dst = skb->_skb_refdst;
724 		skb_dst_set_noref(skb, &ovs_dst);
725 		IPCB(skb)->frag_max_size = mru;
726 
727 		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
728 		refdst_drop(orig_dst);
729 	} else if (ethertype == htons(ETH_P_IPV6)) {
730 		const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
731 		unsigned long orig_dst;
732 		struct rt6_info ovs_rt;
733 
734 		if (!v6ops) {
735 			goto err;
736 		}
737 
738 		prepare_frag(vport, skb);
739 		memset(&ovs_rt, 0, sizeof(ovs_rt));
740 		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
741 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
742 		ovs_rt.dst.dev = vport->dev;
743 
744 		orig_dst = skb->_skb_refdst;
745 		skb_dst_set_noref(skb, &ovs_rt.dst);
746 		IP6CB(skb)->frag_max_size = mru;
747 
748 		v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
749 		refdst_drop(orig_dst);
750 	} else {
751 		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
752 			  ovs_vport_name(vport), ntohs(ethertype), mru,
753 			  vport->dev->mtu);
754 		goto err;
755 	}
756 
757 	return;
758 err:
759 	kfree_skb(skb);
760 }
761 
762 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
763 		      struct sw_flow_key *key)
764 {
765 	struct vport *vport = ovs_vport_rcu(dp, out_port);
766 
767 	if (likely(vport)) {
768 		u16 mru = OVS_CB(skb)->mru;
769 		u32 cutlen = OVS_CB(skb)->cutlen;
770 
771 		if (unlikely(cutlen > 0)) {
772 			if (skb->len - cutlen > ETH_HLEN)
773 				pskb_trim(skb, skb->len - cutlen);
774 			else
775 				pskb_trim(skb, ETH_HLEN);
776 		}
777 
778 		if (likely(!mru || (skb->len <= mru + ETH_HLEN))) {
779 			ovs_vport_send(vport, skb);
780 		} else if (mru <= vport->dev->mtu) {
781 			struct net *net = read_pnet(&dp->net);
782 			__be16 ethertype = key->eth.type;
783 
784 			if (!is_flow_key_valid(key)) {
785 				if (eth_p_mpls(skb->protocol))
786 					ethertype = skb->inner_protocol;
787 				else
788 					ethertype = vlan_get_protocol(skb);
789 			}
790 
791 			ovs_fragment(net, vport, skb, mru, ethertype);
792 		} else {
793 			kfree_skb(skb);
794 		}
795 	} else {
796 		kfree_skb(skb);
797 	}
798 }
799 
800 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
801 			    struct sw_flow_key *key, const struct nlattr *attr,
802 			    const struct nlattr *actions, int actions_len,
803 			    uint32_t cutlen)
804 {
805 	struct dp_upcall_info upcall;
806 	const struct nlattr *a;
807 	int rem;
808 
809 	memset(&upcall, 0, sizeof(upcall));
810 	upcall.cmd = OVS_PACKET_CMD_ACTION;
811 	upcall.mru = OVS_CB(skb)->mru;
812 
813 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
814 		 a = nla_next(a, &rem)) {
815 		switch (nla_type(a)) {
816 		case OVS_USERSPACE_ATTR_USERDATA:
817 			upcall.userdata = a;
818 			break;
819 
820 		case OVS_USERSPACE_ATTR_PID:
821 			upcall.portid = nla_get_u32(a);
822 			break;
823 
824 		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
825 			/* Get out tunnel info. */
826 			struct vport *vport;
827 
828 			vport = ovs_vport_rcu(dp, nla_get_u32(a));
829 			if (vport) {
830 				int err;
831 
832 				err = dev_fill_metadata_dst(vport->dev, skb);
833 				if (!err)
834 					upcall.egress_tun_info = skb_tunnel_info(skb);
835 			}
836 
837 			break;
838 		}
839 
840 		case OVS_USERSPACE_ATTR_ACTIONS: {
841 			/* Include actions. */
842 			upcall.actions = actions;
843 			upcall.actions_len = actions_len;
844 			break;
845 		}
846 
847 		} /* End of switch. */
848 	}
849 
850 	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
851 }
852 
853 static int sample(struct datapath *dp, struct sk_buff *skb,
854 		  struct sw_flow_key *key, const struct nlattr *attr,
855 		  const struct nlattr *actions, int actions_len)
856 {
857 	const struct nlattr *acts_list = NULL;
858 	const struct nlattr *a;
859 	int rem;
860 	u32 cutlen = 0;
861 
862 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
863 		 a = nla_next(a, &rem)) {
864 		u32 probability;
865 
866 		switch (nla_type(a)) {
867 		case OVS_SAMPLE_ATTR_PROBABILITY:
868 			probability = nla_get_u32(a);
869 			if (!probability || prandom_u32() > probability)
870 				return 0;
871 			break;
872 
873 		case OVS_SAMPLE_ATTR_ACTIONS:
874 			acts_list = a;
875 			break;
876 		}
877 	}
878 
879 	rem = nla_len(acts_list);
880 	a = nla_data(acts_list);
881 
882 	/* Actions list is empty, do nothing */
883 	if (unlikely(!rem))
884 		return 0;
885 
886 	/* The only known usage of sample action is having a single user-space
887 	 * action, or having a truncate action followed by a single user-space
888 	 * action. Treat this usage as a special case.
889 	 * The output_userspace() should clone the skb to be sent to the
890 	 * user space. This skb will be consumed by its caller.
891 	 */
892 	if (unlikely(nla_type(a) == OVS_ACTION_ATTR_TRUNC)) {
893 		struct ovs_action_trunc *trunc = nla_data(a);
894 
895 		if (skb->len > trunc->max_len)
896 			cutlen = skb->len - trunc->max_len;
897 
898 		a = nla_next(a, &rem);
899 	}
900 
901 	if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
902 		   nla_is_last(a, rem)))
903 		return output_userspace(dp, skb, key, a, actions,
904 					actions_len, cutlen);
905 
906 	skb = skb_clone(skb, GFP_ATOMIC);
907 	if (!skb)
908 		/* Skip the sample action when out of memory. */
909 		return 0;
910 
911 	if (!add_deferred_actions(skb, key, a)) {
912 		if (net_ratelimit())
913 			pr_warn("%s: deferred actions limit reached, dropping sample action\n",
914 				ovs_dp_name(dp));
915 
916 		kfree_skb(skb);
917 	}
918 	return 0;
919 }
920 
921 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
922 			 const struct nlattr *attr)
923 {
924 	struct ovs_action_hash *hash_act = nla_data(attr);
925 	u32 hash = 0;
926 
927 	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
928 	hash = skb_get_hash(skb);
929 	hash = jhash_1word(hash, hash_act->hash_basis);
930 	if (!hash)
931 		hash = 0x1;
932 
933 	key->ovs_flow_hash = hash;
934 }
935 
936 static int execute_set_action(struct sk_buff *skb,
937 			      struct sw_flow_key *flow_key,
938 			      const struct nlattr *a)
939 {
940 	/* Only tunnel set execution is supported without a mask. */
941 	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
942 		struct ovs_tunnel_info *tun = nla_data(a);
943 
944 		skb_dst_drop(skb);
945 		dst_hold((struct dst_entry *)tun->tun_dst);
946 		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
947 		return 0;
948 	}
949 
950 	return -EINVAL;
951 }
952 
953 /* Mask is at the midpoint of the data. */
954 #define get_mask(a, type) ((const type)nla_data(a) + 1)
955 
956 static int execute_masked_set_action(struct sk_buff *skb,
957 				     struct sw_flow_key *flow_key,
958 				     const struct nlattr *a)
959 {
960 	int err = 0;
961 
962 	switch (nla_type(a)) {
963 	case OVS_KEY_ATTR_PRIORITY:
964 		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
965 			       *get_mask(a, u32 *));
966 		flow_key->phy.priority = skb->priority;
967 		break;
968 
969 	case OVS_KEY_ATTR_SKB_MARK:
970 		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
971 		flow_key->phy.skb_mark = skb->mark;
972 		break;
973 
974 	case OVS_KEY_ATTR_TUNNEL_INFO:
975 		/* Masked data not supported for tunnel. */
976 		err = -EINVAL;
977 		break;
978 
979 	case OVS_KEY_ATTR_ETHERNET:
980 		err = set_eth_addr(skb, flow_key, nla_data(a),
981 				   get_mask(a, struct ovs_key_ethernet *));
982 		break;
983 
984 	case OVS_KEY_ATTR_IPV4:
985 		err = set_ipv4(skb, flow_key, nla_data(a),
986 			       get_mask(a, struct ovs_key_ipv4 *));
987 		break;
988 
989 	case OVS_KEY_ATTR_IPV6:
990 		err = set_ipv6(skb, flow_key, nla_data(a),
991 			       get_mask(a, struct ovs_key_ipv6 *));
992 		break;
993 
994 	case OVS_KEY_ATTR_TCP:
995 		err = set_tcp(skb, flow_key, nla_data(a),
996 			      get_mask(a, struct ovs_key_tcp *));
997 		break;
998 
999 	case OVS_KEY_ATTR_UDP:
1000 		err = set_udp(skb, flow_key, nla_data(a),
1001 			      get_mask(a, struct ovs_key_udp *));
1002 		break;
1003 
1004 	case OVS_KEY_ATTR_SCTP:
1005 		err = set_sctp(skb, flow_key, nla_data(a),
1006 			       get_mask(a, struct ovs_key_sctp *));
1007 		break;
1008 
1009 	case OVS_KEY_ATTR_MPLS:
1010 		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1011 								    __be32 *));
1012 		break;
1013 
1014 	case OVS_KEY_ATTR_CT_STATE:
1015 	case OVS_KEY_ATTR_CT_ZONE:
1016 	case OVS_KEY_ATTR_CT_MARK:
1017 	case OVS_KEY_ATTR_CT_LABELS:
1018 		err = -EINVAL;
1019 		break;
1020 	}
1021 
1022 	return err;
1023 }
1024 
1025 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1026 			  struct sw_flow_key *key,
1027 			  const struct nlattr *a, int rem)
1028 {
1029 	struct deferred_action *da;
1030 	int level;
1031 
1032 	if (!is_flow_key_valid(key)) {
1033 		int err;
1034 
1035 		err = ovs_flow_key_update(skb, key);
1036 		if (err)
1037 			return err;
1038 	}
1039 	BUG_ON(!is_flow_key_valid(key));
1040 
1041 	if (!nla_is_last(a, rem)) {
1042 		/* Recirc action is the not the last action
1043 		 * of the action list, need to clone the skb.
1044 		 */
1045 		skb = skb_clone(skb, GFP_ATOMIC);
1046 
1047 		/* Skip the recirc action when out of memory, but
1048 		 * continue on with the rest of the action list.
1049 		 */
1050 		if (!skb)
1051 			return 0;
1052 	}
1053 
1054 	level = this_cpu_read(exec_actions_level);
1055 	if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
1056 		struct recirc_keys *rks = this_cpu_ptr(recirc_keys);
1057 		struct sw_flow_key *recirc_key = &rks->key[level - 1];
1058 
1059 		*recirc_key = *key;
1060 		recirc_key->recirc_id = nla_get_u32(a);
1061 		ovs_dp_process_packet(skb, recirc_key);
1062 
1063 		return 0;
1064 	}
1065 
1066 	da = add_deferred_actions(skb, key, NULL);
1067 	if (da) {
1068 		da->pkt_key.recirc_id = nla_get_u32(a);
1069 	} else {
1070 		kfree_skb(skb);
1071 
1072 		if (net_ratelimit())
1073 			pr_warn("%s: deferred action limit reached, drop recirc action\n",
1074 				ovs_dp_name(dp));
1075 	}
1076 
1077 	return 0;
1078 }
1079 
1080 /* Execute a list of actions against 'skb'. */
1081 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1082 			      struct sw_flow_key *key,
1083 			      const struct nlattr *attr, int len)
1084 {
1085 	/* Every output action needs a separate clone of 'skb', but the common
1086 	 * case is just a single output action, so that doing a clone and
1087 	 * then freeing the original skbuff is wasteful.  So the following code
1088 	 * is slightly obscure just to avoid that.
1089 	 */
1090 	int prev_port = -1;
1091 	const struct nlattr *a;
1092 	int rem;
1093 
1094 	for (a = attr, rem = len; rem > 0;
1095 	     a = nla_next(a, &rem)) {
1096 		int err = 0;
1097 
1098 		if (unlikely(prev_port != -1)) {
1099 			struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1100 
1101 			if (out_skb)
1102 				do_output(dp, out_skb, prev_port, key);
1103 
1104 			OVS_CB(skb)->cutlen = 0;
1105 			prev_port = -1;
1106 		}
1107 
1108 		switch (nla_type(a)) {
1109 		case OVS_ACTION_ATTR_OUTPUT:
1110 			prev_port = nla_get_u32(a);
1111 			break;
1112 
1113 		case OVS_ACTION_ATTR_TRUNC: {
1114 			struct ovs_action_trunc *trunc = nla_data(a);
1115 
1116 			if (skb->len > trunc->max_len)
1117 				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1118 			break;
1119 		}
1120 
1121 		case OVS_ACTION_ATTR_USERSPACE:
1122 			output_userspace(dp, skb, key, a, attr,
1123 						     len, OVS_CB(skb)->cutlen);
1124 			OVS_CB(skb)->cutlen = 0;
1125 			break;
1126 
1127 		case OVS_ACTION_ATTR_HASH:
1128 			execute_hash(skb, key, a);
1129 			break;
1130 
1131 		case OVS_ACTION_ATTR_PUSH_MPLS:
1132 			err = push_mpls(skb, key, nla_data(a));
1133 			break;
1134 
1135 		case OVS_ACTION_ATTR_POP_MPLS:
1136 			err = pop_mpls(skb, key, nla_get_be16(a));
1137 			break;
1138 
1139 		case OVS_ACTION_ATTR_PUSH_VLAN:
1140 			err = push_vlan(skb, key, nla_data(a));
1141 			break;
1142 
1143 		case OVS_ACTION_ATTR_POP_VLAN:
1144 			err = pop_vlan(skb, key);
1145 			break;
1146 
1147 		case OVS_ACTION_ATTR_RECIRC:
1148 			err = execute_recirc(dp, skb, key, a, rem);
1149 			if (nla_is_last(a, rem)) {
1150 				/* If this is the last action, the skb has
1151 				 * been consumed or freed.
1152 				 * Return immediately.
1153 				 */
1154 				return err;
1155 			}
1156 			break;
1157 
1158 		case OVS_ACTION_ATTR_SET:
1159 			err = execute_set_action(skb, key, nla_data(a));
1160 			break;
1161 
1162 		case OVS_ACTION_ATTR_SET_MASKED:
1163 		case OVS_ACTION_ATTR_SET_TO_MASKED:
1164 			err = execute_masked_set_action(skb, key, nla_data(a));
1165 			break;
1166 
1167 		case OVS_ACTION_ATTR_SAMPLE:
1168 			err = sample(dp, skb, key, a, attr, len);
1169 			break;
1170 
1171 		case OVS_ACTION_ATTR_CT:
1172 			if (!is_flow_key_valid(key)) {
1173 				err = ovs_flow_key_update(skb, key);
1174 				if (err)
1175 					return err;
1176 			}
1177 
1178 			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1179 					     nla_data(a));
1180 
1181 			/* Hide stolen IP fragments from user space. */
1182 			if (err)
1183 				return err == -EINPROGRESS ? 0 : err;
1184 			break;
1185 		}
1186 
1187 		if (unlikely(err)) {
1188 			kfree_skb(skb);
1189 			return err;
1190 		}
1191 	}
1192 
1193 	if (prev_port != -1)
1194 		do_output(dp, skb, prev_port, key);
1195 	else
1196 		consume_skb(skb);
1197 
1198 	return 0;
1199 }
1200 
1201 static void process_deferred_actions(struct datapath *dp)
1202 {
1203 	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1204 
1205 	/* Do not touch the FIFO in case there is no deferred actions. */
1206 	if (action_fifo_is_empty(fifo))
1207 		return;
1208 
1209 	/* Finishing executing all deferred actions. */
1210 	do {
1211 		struct deferred_action *da = action_fifo_get(fifo);
1212 		struct sk_buff *skb = da->skb;
1213 		struct sw_flow_key *key = &da->pkt_key;
1214 		const struct nlattr *actions = da->actions;
1215 
1216 		if (actions)
1217 			do_execute_actions(dp, skb, key, actions,
1218 					   nla_len(actions));
1219 		else
1220 			ovs_dp_process_packet(skb, key);
1221 	} while (!action_fifo_is_empty(fifo));
1222 
1223 	/* Reset FIFO for the next packet.  */
1224 	action_fifo_init(fifo);
1225 }
1226 
1227 /* Execute a list of actions against 'skb'. */
1228 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1229 			const struct sw_flow_actions *acts,
1230 			struct sw_flow_key *key)
1231 {
1232 	int err, level;
1233 
1234 	level = __this_cpu_inc_return(exec_actions_level);
1235 	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1236 		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1237 				     ovs_dp_name(dp));
1238 		kfree_skb(skb);
1239 		err = -ENETDOWN;
1240 		goto out;
1241 	}
1242 
1243 	err = do_execute_actions(dp, skb, key,
1244 				 acts->actions, acts->actions_len);
1245 
1246 	if (level == 1)
1247 		process_deferred_actions(dp);
1248 
1249 out:
1250 	__this_cpu_dec(exec_actions_level);
1251 	return err;
1252 }
1253 
1254 int action_fifos_init(void)
1255 {
1256 	action_fifos = alloc_percpu(struct action_fifo);
1257 	if (!action_fifos)
1258 		return -ENOMEM;
1259 
1260 	recirc_keys = alloc_percpu(struct recirc_keys);
1261 	if (!recirc_keys) {
1262 		free_percpu(action_fifos);
1263 		return -ENOMEM;
1264 	}
1265 
1266 	return 0;
1267 }
1268 
1269 void action_fifos_exit(void)
1270 {
1271 	free_percpu(action_fifos);
1272 	free_percpu(recirc_keys);
1273 }
1274