xref: /openbmc/linux/net/openvswitch/actions.c (revision 835fd614)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2007-2017 Nicira, Inc.
4  */
5 
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #include <linux/skbuff.h>
9 #include <linux/in.h>
10 #include <linux/ip.h>
11 #include <linux/openvswitch.h>
12 #include <linux/sctp.h>
13 #include <linux/tcp.h>
14 #include <linux/udp.h>
15 #include <linux/in6.h>
16 #include <linux/if_arp.h>
17 #include <linux/if_vlan.h>
18 
19 #include <net/dst.h>
20 #include <net/ip.h>
21 #include <net/ipv6.h>
22 #include <net/ip6_fib.h>
23 #include <net/checksum.h>
24 #include <net/dsfield.h>
25 #include <net/mpls.h>
26 #include <net/sctp/checksum.h>
27 
28 #include "datapath.h"
29 #include "flow.h"
30 #include "conntrack.h"
31 #include "vport.h"
32 #include "flow_netlink.h"
33 
34 struct deferred_action {
35 	struct sk_buff *skb;
36 	const struct nlattr *actions;
37 	int actions_len;
38 
39 	/* Store pkt_key clone when creating deferred action. */
40 	struct sw_flow_key pkt_key;
41 };
42 
43 #define MAX_L2_LEN	(VLAN_ETH_HLEN + 3 * MPLS_HLEN)
44 struct ovs_frag_data {
45 	unsigned long dst;
46 	struct vport *vport;
47 	struct ovs_skb_cb cb;
48 	__be16 inner_protocol;
49 	u16 network_offset;	/* valid only for MPLS */
50 	u16 vlan_tci;
51 	__be16 vlan_proto;
52 	unsigned int l2_len;
53 	u8 mac_proto;
54 	u8 l2_data[MAX_L2_LEN];
55 };
56 
57 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
58 
59 #define DEFERRED_ACTION_FIFO_SIZE 10
60 #define OVS_RECURSION_LIMIT 5
61 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
62 struct action_fifo {
63 	int head;
64 	int tail;
65 	/* Deferred action fifo queue storage. */
66 	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
67 };
68 
69 struct action_flow_keys {
70 	struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
71 };
72 
73 static struct action_fifo __percpu *action_fifos;
74 static struct action_flow_keys __percpu *flow_keys;
75 static DEFINE_PER_CPU(int, exec_actions_level);
76 
77 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
78  * space. Return NULL if out of key spaces.
79  */
80 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
81 {
82 	struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
83 	int level = this_cpu_read(exec_actions_level);
84 	struct sw_flow_key *key = NULL;
85 
86 	if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
87 		key = &keys->key[level - 1];
88 		*key = *key_;
89 	}
90 
91 	return key;
92 }
93 
94 static void action_fifo_init(struct action_fifo *fifo)
95 {
96 	fifo->head = 0;
97 	fifo->tail = 0;
98 }
99 
100 static bool action_fifo_is_empty(const struct action_fifo *fifo)
101 {
102 	return (fifo->head == fifo->tail);
103 }
104 
105 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
106 {
107 	if (action_fifo_is_empty(fifo))
108 		return NULL;
109 
110 	return &fifo->fifo[fifo->tail++];
111 }
112 
113 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
114 {
115 	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
116 		return NULL;
117 
118 	return &fifo->fifo[fifo->head++];
119 }
120 
121 /* Return true if fifo is not full */
122 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
123 				    const struct sw_flow_key *key,
124 				    const struct nlattr *actions,
125 				    const int actions_len)
126 {
127 	struct action_fifo *fifo;
128 	struct deferred_action *da;
129 
130 	fifo = this_cpu_ptr(action_fifos);
131 	da = action_fifo_put(fifo);
132 	if (da) {
133 		da->skb = skb;
134 		da->actions = actions;
135 		da->actions_len = actions_len;
136 		da->pkt_key = *key;
137 	}
138 
139 	return da;
140 }
141 
142 static void invalidate_flow_key(struct sw_flow_key *key)
143 {
144 	key->mac_proto |= SW_FLOW_KEY_INVALID;
145 }
146 
147 static bool is_flow_key_valid(const struct sw_flow_key *key)
148 {
149 	return !(key->mac_proto & SW_FLOW_KEY_INVALID);
150 }
151 
152 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
153 			 struct sw_flow_key *key,
154 			 u32 recirc_id,
155 			 const struct nlattr *actions, int len,
156 			 bool last, bool clone_flow_key);
157 
158 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
159 			      struct sw_flow_key *key,
160 			      const struct nlattr *attr, int len);
161 
162 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
163 		     __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
164 {
165 	int err;
166 
167 	err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
168 	if (err)
169 		return err;
170 
171 	if (!mac_len)
172 		key->mac_proto = MAC_PROTO_NONE;
173 
174 	invalidate_flow_key(key);
175 	return 0;
176 }
177 
178 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
179 		    const __be16 ethertype)
180 {
181 	int err;
182 
183 	err = skb_mpls_pop(skb, ethertype, skb->mac_len,
184 			   ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
185 	if (err)
186 		return err;
187 
188 	if (ethertype == htons(ETH_P_TEB))
189 		key->mac_proto = MAC_PROTO_ETHERNET;
190 
191 	invalidate_flow_key(key);
192 	return 0;
193 }
194 
195 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
196 		    const __be32 *mpls_lse, const __be32 *mask)
197 {
198 	struct mpls_shim_hdr *stack;
199 	__be32 lse;
200 	int err;
201 
202 	stack = mpls_hdr(skb);
203 	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
204 	err = skb_mpls_update_lse(skb, lse);
205 	if (err)
206 		return err;
207 
208 	flow_key->mpls.lse[0] = lse;
209 	return 0;
210 }
211 
212 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
213 {
214 	int err;
215 
216 	err = skb_vlan_pop(skb);
217 	if (skb_vlan_tag_present(skb)) {
218 		invalidate_flow_key(key);
219 	} else {
220 		key->eth.vlan.tci = 0;
221 		key->eth.vlan.tpid = 0;
222 	}
223 	return err;
224 }
225 
226 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
227 		     const struct ovs_action_push_vlan *vlan)
228 {
229 	if (skb_vlan_tag_present(skb)) {
230 		invalidate_flow_key(key);
231 	} else {
232 		key->eth.vlan.tci = vlan->vlan_tci;
233 		key->eth.vlan.tpid = vlan->vlan_tpid;
234 	}
235 	return skb_vlan_push(skb, vlan->vlan_tpid,
236 			     ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
237 }
238 
239 /* 'src' is already properly masked. */
240 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
241 {
242 	u16 *dst = (u16 *)dst_;
243 	const u16 *src = (const u16 *)src_;
244 	const u16 *mask = (const u16 *)mask_;
245 
246 	OVS_SET_MASKED(dst[0], src[0], mask[0]);
247 	OVS_SET_MASKED(dst[1], src[1], mask[1]);
248 	OVS_SET_MASKED(dst[2], src[2], mask[2]);
249 }
250 
251 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
252 			const struct ovs_key_ethernet *key,
253 			const struct ovs_key_ethernet *mask)
254 {
255 	int err;
256 
257 	err = skb_ensure_writable(skb, ETH_HLEN);
258 	if (unlikely(err))
259 		return err;
260 
261 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
262 
263 	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
264 			       mask->eth_src);
265 	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
266 			       mask->eth_dst);
267 
268 	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
269 
270 	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
271 	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
272 	return 0;
273 }
274 
275 /* pop_eth does not support VLAN packets as this action is never called
276  * for them.
277  */
278 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
279 {
280 	int err;
281 
282 	err = skb_eth_pop(skb);
283 	if (err)
284 		return err;
285 
286 	/* safe right before invalidate_flow_key */
287 	key->mac_proto = MAC_PROTO_NONE;
288 	invalidate_flow_key(key);
289 	return 0;
290 }
291 
292 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
293 		    const struct ovs_action_push_eth *ethh)
294 {
295 	int err;
296 
297 	err = skb_eth_push(skb, ethh->addresses.eth_dst,
298 			   ethh->addresses.eth_src);
299 	if (err)
300 		return err;
301 
302 	/* safe right before invalidate_flow_key */
303 	key->mac_proto = MAC_PROTO_ETHERNET;
304 	invalidate_flow_key(key);
305 	return 0;
306 }
307 
308 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
309 		    const struct nshhdr *nh)
310 {
311 	int err;
312 
313 	err = nsh_push(skb, nh);
314 	if (err)
315 		return err;
316 
317 	/* safe right before invalidate_flow_key */
318 	key->mac_proto = MAC_PROTO_NONE;
319 	invalidate_flow_key(key);
320 	return 0;
321 }
322 
323 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
324 {
325 	int err;
326 
327 	err = nsh_pop(skb);
328 	if (err)
329 		return err;
330 
331 	/* safe right before invalidate_flow_key */
332 	if (skb->protocol == htons(ETH_P_TEB))
333 		key->mac_proto = MAC_PROTO_ETHERNET;
334 	else
335 		key->mac_proto = MAC_PROTO_NONE;
336 	invalidate_flow_key(key);
337 	return 0;
338 }
339 
340 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
341 				  __be32 addr, __be32 new_addr)
342 {
343 	int transport_len = skb->len - skb_transport_offset(skb);
344 
345 	if (nh->frag_off & htons(IP_OFFSET))
346 		return;
347 
348 	if (nh->protocol == IPPROTO_TCP) {
349 		if (likely(transport_len >= sizeof(struct tcphdr)))
350 			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
351 						 addr, new_addr, true);
352 	} else if (nh->protocol == IPPROTO_UDP) {
353 		if (likely(transport_len >= sizeof(struct udphdr))) {
354 			struct udphdr *uh = udp_hdr(skb);
355 
356 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
357 				inet_proto_csum_replace4(&uh->check, skb,
358 							 addr, new_addr, true);
359 				if (!uh->check)
360 					uh->check = CSUM_MANGLED_0;
361 			}
362 		}
363 	}
364 }
365 
366 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
367 			__be32 *addr, __be32 new_addr)
368 {
369 	update_ip_l4_checksum(skb, nh, *addr, new_addr);
370 	csum_replace4(&nh->check, *addr, new_addr);
371 	skb_clear_hash(skb);
372 	*addr = new_addr;
373 }
374 
375 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
376 				 __be32 addr[4], const __be32 new_addr[4])
377 {
378 	int transport_len = skb->len - skb_transport_offset(skb);
379 
380 	if (l4_proto == NEXTHDR_TCP) {
381 		if (likely(transport_len >= sizeof(struct tcphdr)))
382 			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
383 						  addr, new_addr, true);
384 	} else if (l4_proto == NEXTHDR_UDP) {
385 		if (likely(transport_len >= sizeof(struct udphdr))) {
386 			struct udphdr *uh = udp_hdr(skb);
387 
388 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
389 				inet_proto_csum_replace16(&uh->check, skb,
390 							  addr, new_addr, true);
391 				if (!uh->check)
392 					uh->check = CSUM_MANGLED_0;
393 			}
394 		}
395 	} else if (l4_proto == NEXTHDR_ICMP) {
396 		if (likely(transport_len >= sizeof(struct icmp6hdr)))
397 			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
398 						  skb, addr, new_addr, true);
399 	}
400 }
401 
402 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
403 			   const __be32 mask[4], __be32 masked[4])
404 {
405 	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
406 	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
407 	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
408 	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
409 }
410 
411 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
412 			  __be32 addr[4], const __be32 new_addr[4],
413 			  bool recalculate_csum)
414 {
415 	if (recalculate_csum)
416 		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
417 
418 	skb_clear_hash(skb);
419 	memcpy(addr, new_addr, sizeof(__be32[4]));
420 }
421 
422 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
423 {
424 	/* Bits 21-24 are always unmasked, so this retains their values. */
425 	OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
426 	OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
427 	OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
428 }
429 
430 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
431 		       u8 mask)
432 {
433 	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
434 
435 	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
436 	nh->ttl = new_ttl;
437 }
438 
439 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
440 		    const struct ovs_key_ipv4 *key,
441 		    const struct ovs_key_ipv4 *mask)
442 {
443 	struct iphdr *nh;
444 	__be32 new_addr;
445 	int err;
446 
447 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
448 				  sizeof(struct iphdr));
449 	if (unlikely(err))
450 		return err;
451 
452 	nh = ip_hdr(skb);
453 
454 	/* Setting an IP addresses is typically only a side effect of
455 	 * matching on them in the current userspace implementation, so it
456 	 * makes sense to check if the value actually changed.
457 	 */
458 	if (mask->ipv4_src) {
459 		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
460 
461 		if (unlikely(new_addr != nh->saddr)) {
462 			set_ip_addr(skb, nh, &nh->saddr, new_addr);
463 			flow_key->ipv4.addr.src = new_addr;
464 		}
465 	}
466 	if (mask->ipv4_dst) {
467 		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
468 
469 		if (unlikely(new_addr != nh->daddr)) {
470 			set_ip_addr(skb, nh, &nh->daddr, new_addr);
471 			flow_key->ipv4.addr.dst = new_addr;
472 		}
473 	}
474 	if (mask->ipv4_tos) {
475 		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
476 		flow_key->ip.tos = nh->tos;
477 	}
478 	if (mask->ipv4_ttl) {
479 		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
480 		flow_key->ip.ttl = nh->ttl;
481 	}
482 
483 	return 0;
484 }
485 
486 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
487 {
488 	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
489 }
490 
491 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
492 		    const struct ovs_key_ipv6 *key,
493 		    const struct ovs_key_ipv6 *mask)
494 {
495 	struct ipv6hdr *nh;
496 	int err;
497 
498 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
499 				  sizeof(struct ipv6hdr));
500 	if (unlikely(err))
501 		return err;
502 
503 	nh = ipv6_hdr(skb);
504 
505 	/* Setting an IP addresses is typically only a side effect of
506 	 * matching on them in the current userspace implementation, so it
507 	 * makes sense to check if the value actually changed.
508 	 */
509 	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
510 		__be32 *saddr = (__be32 *)&nh->saddr;
511 		__be32 masked[4];
512 
513 		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
514 
515 		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
516 			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
517 				      true);
518 			memcpy(&flow_key->ipv6.addr.src, masked,
519 			       sizeof(flow_key->ipv6.addr.src));
520 		}
521 	}
522 	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
523 		unsigned int offset = 0;
524 		int flags = IP6_FH_F_SKIP_RH;
525 		bool recalc_csum = true;
526 		__be32 *daddr = (__be32 *)&nh->daddr;
527 		__be32 masked[4];
528 
529 		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
530 
531 		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
532 			if (ipv6_ext_hdr(nh->nexthdr))
533 				recalc_csum = (ipv6_find_hdr(skb, &offset,
534 							     NEXTHDR_ROUTING,
535 							     NULL, &flags)
536 					       != NEXTHDR_ROUTING);
537 
538 			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
539 				      recalc_csum);
540 			memcpy(&flow_key->ipv6.addr.dst, masked,
541 			       sizeof(flow_key->ipv6.addr.dst));
542 		}
543 	}
544 	if (mask->ipv6_tclass) {
545 		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
546 		flow_key->ip.tos = ipv6_get_dsfield(nh);
547 	}
548 	if (mask->ipv6_label) {
549 		set_ipv6_fl(nh, ntohl(key->ipv6_label),
550 			    ntohl(mask->ipv6_label));
551 		flow_key->ipv6.label =
552 		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
553 	}
554 	if (mask->ipv6_hlimit) {
555 		OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
556 			       mask->ipv6_hlimit);
557 		flow_key->ip.ttl = nh->hop_limit;
558 	}
559 	return 0;
560 }
561 
562 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
563 		   const struct nlattr *a)
564 {
565 	struct nshhdr *nh;
566 	size_t length;
567 	int err;
568 	u8 flags;
569 	u8 ttl;
570 	int i;
571 
572 	struct ovs_key_nsh key;
573 	struct ovs_key_nsh mask;
574 
575 	err = nsh_key_from_nlattr(a, &key, &mask);
576 	if (err)
577 		return err;
578 
579 	/* Make sure the NSH base header is there */
580 	if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
581 		return -ENOMEM;
582 
583 	nh = nsh_hdr(skb);
584 	length = nsh_hdr_len(nh);
585 
586 	/* Make sure the whole NSH header is there */
587 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
588 				       length);
589 	if (unlikely(err))
590 		return err;
591 
592 	nh = nsh_hdr(skb);
593 	skb_postpull_rcsum(skb, nh, length);
594 	flags = nsh_get_flags(nh);
595 	flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
596 	flow_key->nsh.base.flags = flags;
597 	ttl = nsh_get_ttl(nh);
598 	ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
599 	flow_key->nsh.base.ttl = ttl;
600 	nsh_set_flags_and_ttl(nh, flags, ttl);
601 	nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
602 				  mask.base.path_hdr);
603 	flow_key->nsh.base.path_hdr = nh->path_hdr;
604 	switch (nh->mdtype) {
605 	case NSH_M_TYPE1:
606 		for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
607 			nh->md1.context[i] =
608 			    OVS_MASKED(nh->md1.context[i], key.context[i],
609 				       mask.context[i]);
610 		}
611 		memcpy(flow_key->nsh.context, nh->md1.context,
612 		       sizeof(nh->md1.context));
613 		break;
614 	case NSH_M_TYPE2:
615 		memset(flow_key->nsh.context, 0,
616 		       sizeof(flow_key->nsh.context));
617 		break;
618 	default:
619 		return -EINVAL;
620 	}
621 	skb_postpush_rcsum(skb, nh, length);
622 	return 0;
623 }
624 
625 /* Must follow skb_ensure_writable() since that can move the skb data. */
626 static void set_tp_port(struct sk_buff *skb, __be16 *port,
627 			__be16 new_port, __sum16 *check)
628 {
629 	inet_proto_csum_replace2(check, skb, *port, new_port, false);
630 	*port = new_port;
631 }
632 
633 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
634 		   const struct ovs_key_udp *key,
635 		   const struct ovs_key_udp *mask)
636 {
637 	struct udphdr *uh;
638 	__be16 src, dst;
639 	int err;
640 
641 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
642 				  sizeof(struct udphdr));
643 	if (unlikely(err))
644 		return err;
645 
646 	uh = udp_hdr(skb);
647 	/* Either of the masks is non-zero, so do not bother checking them. */
648 	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
649 	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
650 
651 	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
652 		if (likely(src != uh->source)) {
653 			set_tp_port(skb, &uh->source, src, &uh->check);
654 			flow_key->tp.src = src;
655 		}
656 		if (likely(dst != uh->dest)) {
657 			set_tp_port(skb, &uh->dest, dst, &uh->check);
658 			flow_key->tp.dst = dst;
659 		}
660 
661 		if (unlikely(!uh->check))
662 			uh->check = CSUM_MANGLED_0;
663 	} else {
664 		uh->source = src;
665 		uh->dest = dst;
666 		flow_key->tp.src = src;
667 		flow_key->tp.dst = dst;
668 	}
669 
670 	skb_clear_hash(skb);
671 
672 	return 0;
673 }
674 
675 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
676 		   const struct ovs_key_tcp *key,
677 		   const struct ovs_key_tcp *mask)
678 {
679 	struct tcphdr *th;
680 	__be16 src, dst;
681 	int err;
682 
683 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
684 				  sizeof(struct tcphdr));
685 	if (unlikely(err))
686 		return err;
687 
688 	th = tcp_hdr(skb);
689 	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
690 	if (likely(src != th->source)) {
691 		set_tp_port(skb, &th->source, src, &th->check);
692 		flow_key->tp.src = src;
693 	}
694 	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
695 	if (likely(dst != th->dest)) {
696 		set_tp_port(skb, &th->dest, dst, &th->check);
697 		flow_key->tp.dst = dst;
698 	}
699 	skb_clear_hash(skb);
700 
701 	return 0;
702 }
703 
704 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
705 		    const struct ovs_key_sctp *key,
706 		    const struct ovs_key_sctp *mask)
707 {
708 	unsigned int sctphoff = skb_transport_offset(skb);
709 	struct sctphdr *sh;
710 	__le32 old_correct_csum, new_csum, old_csum;
711 	int err;
712 
713 	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
714 	if (unlikely(err))
715 		return err;
716 
717 	sh = sctp_hdr(skb);
718 	old_csum = sh->checksum;
719 	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
720 
721 	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
722 	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
723 
724 	new_csum = sctp_compute_cksum(skb, sctphoff);
725 
726 	/* Carry any checksum errors through. */
727 	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
728 
729 	skb_clear_hash(skb);
730 	flow_key->tp.src = sh->source;
731 	flow_key->tp.dst = sh->dest;
732 
733 	return 0;
734 }
735 
736 static int ovs_vport_output(struct net *net, struct sock *sk,
737 			    struct sk_buff *skb)
738 {
739 	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
740 	struct vport *vport = data->vport;
741 
742 	if (skb_cow_head(skb, data->l2_len) < 0) {
743 		kfree_skb(skb);
744 		return -ENOMEM;
745 	}
746 
747 	__skb_dst_copy(skb, data->dst);
748 	*OVS_CB(skb) = data->cb;
749 	skb->inner_protocol = data->inner_protocol;
750 	if (data->vlan_tci & VLAN_CFI_MASK)
751 		__vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
752 	else
753 		__vlan_hwaccel_clear_tag(skb);
754 
755 	/* Reconstruct the MAC header.  */
756 	skb_push(skb, data->l2_len);
757 	memcpy(skb->data, &data->l2_data, data->l2_len);
758 	skb_postpush_rcsum(skb, skb->data, data->l2_len);
759 	skb_reset_mac_header(skb);
760 
761 	if (eth_p_mpls(skb->protocol)) {
762 		skb->inner_network_header = skb->network_header;
763 		skb_set_network_header(skb, data->network_offset);
764 		skb_reset_mac_len(skb);
765 	}
766 
767 	ovs_vport_send(vport, skb, data->mac_proto);
768 	return 0;
769 }
770 
771 static unsigned int
772 ovs_dst_get_mtu(const struct dst_entry *dst)
773 {
774 	return dst->dev->mtu;
775 }
776 
777 static struct dst_ops ovs_dst_ops = {
778 	.family = AF_UNSPEC,
779 	.mtu = ovs_dst_get_mtu,
780 };
781 
782 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
783  * ovs_vport_output(), which is called once per fragmented packet.
784  */
785 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
786 			 u16 orig_network_offset, u8 mac_proto)
787 {
788 	unsigned int hlen = skb_network_offset(skb);
789 	struct ovs_frag_data *data;
790 
791 	data = this_cpu_ptr(&ovs_frag_data_storage);
792 	data->dst = skb->_skb_refdst;
793 	data->vport = vport;
794 	data->cb = *OVS_CB(skb);
795 	data->inner_protocol = skb->inner_protocol;
796 	data->network_offset = orig_network_offset;
797 	if (skb_vlan_tag_present(skb))
798 		data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
799 	else
800 		data->vlan_tci = 0;
801 	data->vlan_proto = skb->vlan_proto;
802 	data->mac_proto = mac_proto;
803 	data->l2_len = hlen;
804 	memcpy(&data->l2_data, skb->data, hlen);
805 
806 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
807 	skb_pull(skb, hlen);
808 }
809 
810 static void ovs_fragment(struct net *net, struct vport *vport,
811 			 struct sk_buff *skb, u16 mru,
812 			 struct sw_flow_key *key)
813 {
814 	u16 orig_network_offset = 0;
815 
816 	if (eth_p_mpls(skb->protocol)) {
817 		orig_network_offset = skb_network_offset(skb);
818 		skb->network_header = skb->inner_network_header;
819 	}
820 
821 	if (skb_network_offset(skb) > MAX_L2_LEN) {
822 		OVS_NLERR(1, "L2 header too long to fragment");
823 		goto err;
824 	}
825 
826 	if (key->eth.type == htons(ETH_P_IP)) {
827 		struct dst_entry ovs_dst;
828 		unsigned long orig_dst;
829 
830 		prepare_frag(vport, skb, orig_network_offset,
831 			     ovs_key_mac_proto(key));
832 		dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
833 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
834 		ovs_dst.dev = vport->dev;
835 
836 		orig_dst = skb->_skb_refdst;
837 		skb_dst_set_noref(skb, &ovs_dst);
838 		IPCB(skb)->frag_max_size = mru;
839 
840 		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
841 		refdst_drop(orig_dst);
842 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
843 		unsigned long orig_dst;
844 		struct rt6_info ovs_rt;
845 
846 		prepare_frag(vport, skb, orig_network_offset,
847 			     ovs_key_mac_proto(key));
848 		memset(&ovs_rt, 0, sizeof(ovs_rt));
849 		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
850 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
851 		ovs_rt.dst.dev = vport->dev;
852 
853 		orig_dst = skb->_skb_refdst;
854 		skb_dst_set_noref(skb, &ovs_rt.dst);
855 		IP6CB(skb)->frag_max_size = mru;
856 
857 		ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
858 		refdst_drop(orig_dst);
859 	} else {
860 		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
861 			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
862 			  vport->dev->mtu);
863 		goto err;
864 	}
865 
866 	return;
867 err:
868 	kfree_skb(skb);
869 }
870 
871 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
872 		      struct sw_flow_key *key)
873 {
874 	struct vport *vport = ovs_vport_rcu(dp, out_port);
875 
876 	if (likely(vport)) {
877 		u16 mru = OVS_CB(skb)->mru;
878 		u32 cutlen = OVS_CB(skb)->cutlen;
879 
880 		if (unlikely(cutlen > 0)) {
881 			if (skb->len - cutlen > ovs_mac_header_len(key))
882 				pskb_trim(skb, skb->len - cutlen);
883 			else
884 				pskb_trim(skb, ovs_mac_header_len(key));
885 		}
886 
887 		if (likely(!mru ||
888 		           (skb->len <= mru + vport->dev->hard_header_len))) {
889 			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
890 		} else if (mru <= vport->dev->mtu) {
891 			struct net *net = read_pnet(&dp->net);
892 
893 			ovs_fragment(net, vport, skb, mru, key);
894 		} else {
895 			kfree_skb(skb);
896 		}
897 	} else {
898 		kfree_skb(skb);
899 	}
900 }
901 
902 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
903 			    struct sw_flow_key *key, const struct nlattr *attr,
904 			    const struct nlattr *actions, int actions_len,
905 			    uint32_t cutlen)
906 {
907 	struct dp_upcall_info upcall;
908 	const struct nlattr *a;
909 	int rem;
910 
911 	memset(&upcall, 0, sizeof(upcall));
912 	upcall.cmd = OVS_PACKET_CMD_ACTION;
913 	upcall.mru = OVS_CB(skb)->mru;
914 
915 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
916 	     a = nla_next(a, &rem)) {
917 		switch (nla_type(a)) {
918 		case OVS_USERSPACE_ATTR_USERDATA:
919 			upcall.userdata = a;
920 			break;
921 
922 		case OVS_USERSPACE_ATTR_PID:
923 			upcall.portid = nla_get_u32(a);
924 			break;
925 
926 		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
927 			/* Get out tunnel info. */
928 			struct vport *vport;
929 
930 			vport = ovs_vport_rcu(dp, nla_get_u32(a));
931 			if (vport) {
932 				int err;
933 
934 				err = dev_fill_metadata_dst(vport->dev, skb);
935 				if (!err)
936 					upcall.egress_tun_info = skb_tunnel_info(skb);
937 			}
938 
939 			break;
940 		}
941 
942 		case OVS_USERSPACE_ATTR_ACTIONS: {
943 			/* Include actions. */
944 			upcall.actions = actions;
945 			upcall.actions_len = actions_len;
946 			break;
947 		}
948 
949 		} /* End of switch. */
950 	}
951 
952 	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
953 }
954 
955 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
956 				     struct sw_flow_key *key,
957 				     const struct nlattr *attr, bool last)
958 {
959 	/* The first action is always 'OVS_DEC_TTL_ATTR_ARG'. */
960 	struct nlattr *dec_ttl_arg = nla_data(attr);
961 	int rem = nla_len(attr);
962 
963 	if (nla_len(dec_ttl_arg)) {
964 		struct nlattr *actions = nla_next(dec_ttl_arg, &rem);
965 
966 		if (actions)
967 			return clone_execute(dp, skb, key, 0, actions, rem,
968 					     last, false);
969 	}
970 	consume_skb(skb);
971 	return 0;
972 }
973 
974 /* When 'last' is true, sample() should always consume the 'skb'.
975  * Otherwise, sample() should keep 'skb' intact regardless what
976  * actions are executed within sample().
977  */
978 static int sample(struct datapath *dp, struct sk_buff *skb,
979 		  struct sw_flow_key *key, const struct nlattr *attr,
980 		  bool last)
981 {
982 	struct nlattr *actions;
983 	struct nlattr *sample_arg;
984 	int rem = nla_len(attr);
985 	const struct sample_arg *arg;
986 	bool clone_flow_key;
987 
988 	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
989 	sample_arg = nla_data(attr);
990 	arg = nla_data(sample_arg);
991 	actions = nla_next(sample_arg, &rem);
992 
993 	if ((arg->probability != U32_MAX) &&
994 	    (!arg->probability || prandom_u32() > arg->probability)) {
995 		if (last)
996 			consume_skb(skb);
997 		return 0;
998 	}
999 
1000 	clone_flow_key = !arg->exec;
1001 	return clone_execute(dp, skb, key, 0, actions, rem, last,
1002 			     clone_flow_key);
1003 }
1004 
1005 /* When 'last' is true, clone() should always consume the 'skb'.
1006  * Otherwise, clone() should keep 'skb' intact regardless what
1007  * actions are executed within clone().
1008  */
1009 static int clone(struct datapath *dp, struct sk_buff *skb,
1010 		 struct sw_flow_key *key, const struct nlattr *attr,
1011 		 bool last)
1012 {
1013 	struct nlattr *actions;
1014 	struct nlattr *clone_arg;
1015 	int rem = nla_len(attr);
1016 	bool dont_clone_flow_key;
1017 
1018 	/* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1019 	clone_arg = nla_data(attr);
1020 	dont_clone_flow_key = nla_get_u32(clone_arg);
1021 	actions = nla_next(clone_arg, &rem);
1022 
1023 	return clone_execute(dp, skb, key, 0, actions, rem, last,
1024 			     !dont_clone_flow_key);
1025 }
1026 
1027 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1028 			 const struct nlattr *attr)
1029 {
1030 	struct ovs_action_hash *hash_act = nla_data(attr);
1031 	u32 hash = 0;
1032 
1033 	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1034 	hash = skb_get_hash(skb);
1035 	hash = jhash_1word(hash, hash_act->hash_basis);
1036 	if (!hash)
1037 		hash = 0x1;
1038 
1039 	key->ovs_flow_hash = hash;
1040 }
1041 
1042 static int execute_set_action(struct sk_buff *skb,
1043 			      struct sw_flow_key *flow_key,
1044 			      const struct nlattr *a)
1045 {
1046 	/* Only tunnel set execution is supported without a mask. */
1047 	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1048 		struct ovs_tunnel_info *tun = nla_data(a);
1049 
1050 		skb_dst_drop(skb);
1051 		dst_hold((struct dst_entry *)tun->tun_dst);
1052 		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1053 		return 0;
1054 	}
1055 
1056 	return -EINVAL;
1057 }
1058 
1059 /* Mask is at the midpoint of the data. */
1060 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1061 
1062 static int execute_masked_set_action(struct sk_buff *skb,
1063 				     struct sw_flow_key *flow_key,
1064 				     const struct nlattr *a)
1065 {
1066 	int err = 0;
1067 
1068 	switch (nla_type(a)) {
1069 	case OVS_KEY_ATTR_PRIORITY:
1070 		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1071 			       *get_mask(a, u32 *));
1072 		flow_key->phy.priority = skb->priority;
1073 		break;
1074 
1075 	case OVS_KEY_ATTR_SKB_MARK:
1076 		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1077 		flow_key->phy.skb_mark = skb->mark;
1078 		break;
1079 
1080 	case OVS_KEY_ATTR_TUNNEL_INFO:
1081 		/* Masked data not supported for tunnel. */
1082 		err = -EINVAL;
1083 		break;
1084 
1085 	case OVS_KEY_ATTR_ETHERNET:
1086 		err = set_eth_addr(skb, flow_key, nla_data(a),
1087 				   get_mask(a, struct ovs_key_ethernet *));
1088 		break;
1089 
1090 	case OVS_KEY_ATTR_NSH:
1091 		err = set_nsh(skb, flow_key, a);
1092 		break;
1093 
1094 	case OVS_KEY_ATTR_IPV4:
1095 		err = set_ipv4(skb, flow_key, nla_data(a),
1096 			       get_mask(a, struct ovs_key_ipv4 *));
1097 		break;
1098 
1099 	case OVS_KEY_ATTR_IPV6:
1100 		err = set_ipv6(skb, flow_key, nla_data(a),
1101 			       get_mask(a, struct ovs_key_ipv6 *));
1102 		break;
1103 
1104 	case OVS_KEY_ATTR_TCP:
1105 		err = set_tcp(skb, flow_key, nla_data(a),
1106 			      get_mask(a, struct ovs_key_tcp *));
1107 		break;
1108 
1109 	case OVS_KEY_ATTR_UDP:
1110 		err = set_udp(skb, flow_key, nla_data(a),
1111 			      get_mask(a, struct ovs_key_udp *));
1112 		break;
1113 
1114 	case OVS_KEY_ATTR_SCTP:
1115 		err = set_sctp(skb, flow_key, nla_data(a),
1116 			       get_mask(a, struct ovs_key_sctp *));
1117 		break;
1118 
1119 	case OVS_KEY_ATTR_MPLS:
1120 		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1121 								    __be32 *));
1122 		break;
1123 
1124 	case OVS_KEY_ATTR_CT_STATE:
1125 	case OVS_KEY_ATTR_CT_ZONE:
1126 	case OVS_KEY_ATTR_CT_MARK:
1127 	case OVS_KEY_ATTR_CT_LABELS:
1128 	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1129 	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1130 		err = -EINVAL;
1131 		break;
1132 	}
1133 
1134 	return err;
1135 }
1136 
1137 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1138 			  struct sw_flow_key *key,
1139 			  const struct nlattr *a, bool last)
1140 {
1141 	u32 recirc_id;
1142 
1143 	if (!is_flow_key_valid(key)) {
1144 		int err;
1145 
1146 		err = ovs_flow_key_update(skb, key);
1147 		if (err)
1148 			return err;
1149 	}
1150 	BUG_ON(!is_flow_key_valid(key));
1151 
1152 	recirc_id = nla_get_u32(a);
1153 	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1154 }
1155 
1156 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1157 				 struct sw_flow_key *key,
1158 				 const struct nlattr *attr, bool last)
1159 {
1160 	struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1161 	const struct nlattr *actions, *cpl_arg;
1162 	int len, max_len, rem = nla_len(attr);
1163 	const struct check_pkt_len_arg *arg;
1164 	bool clone_flow_key;
1165 
1166 	/* The first netlink attribute in 'attr' is always
1167 	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1168 	 */
1169 	cpl_arg = nla_data(attr);
1170 	arg = nla_data(cpl_arg);
1171 
1172 	len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1173 	max_len = arg->pkt_len;
1174 
1175 	if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1176 	    len <= max_len) {
1177 		/* Second netlink attribute in 'attr' is always
1178 		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1179 		 */
1180 		actions = nla_next(cpl_arg, &rem);
1181 		clone_flow_key = !arg->exec_for_lesser_equal;
1182 	} else {
1183 		/* Third netlink attribute in 'attr' is always
1184 		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1185 		 */
1186 		actions = nla_next(cpl_arg, &rem);
1187 		actions = nla_next(actions, &rem);
1188 		clone_flow_key = !arg->exec_for_greater;
1189 	}
1190 
1191 	return clone_execute(dp, skb, key, 0, nla_data(actions),
1192 			     nla_len(actions), last, clone_flow_key);
1193 }
1194 
1195 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1196 {
1197 	int err;
1198 
1199 	if (skb->protocol == htons(ETH_P_IPV6)) {
1200 		struct ipv6hdr *nh;
1201 
1202 		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1203 					  sizeof(*nh));
1204 		if (unlikely(err))
1205 			return err;
1206 
1207 		nh = ipv6_hdr(skb);
1208 
1209 		if (nh->hop_limit <= 1)
1210 			return -EHOSTUNREACH;
1211 
1212 		key->ip.ttl = --nh->hop_limit;
1213 	} else {
1214 		struct iphdr *nh;
1215 		u8 old_ttl;
1216 
1217 		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1218 					  sizeof(*nh));
1219 		if (unlikely(err))
1220 			return err;
1221 
1222 		nh = ip_hdr(skb);
1223 		if (nh->ttl <= 1)
1224 			return -EHOSTUNREACH;
1225 
1226 		old_ttl = nh->ttl--;
1227 		csum_replace2(&nh->check, htons(old_ttl << 8),
1228 			      htons(nh->ttl << 8));
1229 		key->ip.ttl = nh->ttl;
1230 	}
1231 	return 0;
1232 }
1233 
1234 /* Execute a list of actions against 'skb'. */
1235 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1236 			      struct sw_flow_key *key,
1237 			      const struct nlattr *attr, int len)
1238 {
1239 	const struct nlattr *a;
1240 	int rem;
1241 
1242 	for (a = attr, rem = len; rem > 0;
1243 	     a = nla_next(a, &rem)) {
1244 		int err = 0;
1245 
1246 		switch (nla_type(a)) {
1247 		case OVS_ACTION_ATTR_OUTPUT: {
1248 			int port = nla_get_u32(a);
1249 			struct sk_buff *clone;
1250 
1251 			/* Every output action needs a separate clone
1252 			 * of 'skb', In case the output action is the
1253 			 * last action, cloning can be avoided.
1254 			 */
1255 			if (nla_is_last(a, rem)) {
1256 				do_output(dp, skb, port, key);
1257 				/* 'skb' has been used for output.
1258 				 */
1259 				return 0;
1260 			}
1261 
1262 			clone = skb_clone(skb, GFP_ATOMIC);
1263 			if (clone)
1264 				do_output(dp, clone, port, key);
1265 			OVS_CB(skb)->cutlen = 0;
1266 			break;
1267 		}
1268 
1269 		case OVS_ACTION_ATTR_TRUNC: {
1270 			struct ovs_action_trunc *trunc = nla_data(a);
1271 
1272 			if (skb->len > trunc->max_len)
1273 				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1274 			break;
1275 		}
1276 
1277 		case OVS_ACTION_ATTR_USERSPACE:
1278 			output_userspace(dp, skb, key, a, attr,
1279 						     len, OVS_CB(skb)->cutlen);
1280 			OVS_CB(skb)->cutlen = 0;
1281 			break;
1282 
1283 		case OVS_ACTION_ATTR_HASH:
1284 			execute_hash(skb, key, a);
1285 			break;
1286 
1287 		case OVS_ACTION_ATTR_PUSH_MPLS: {
1288 			struct ovs_action_push_mpls *mpls = nla_data(a);
1289 
1290 			err = push_mpls(skb, key, mpls->mpls_lse,
1291 					mpls->mpls_ethertype, skb->mac_len);
1292 			break;
1293 		}
1294 		case OVS_ACTION_ATTR_ADD_MPLS: {
1295 			struct ovs_action_add_mpls *mpls = nla_data(a);
1296 			__u16 mac_len = 0;
1297 
1298 			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1299 				mac_len = skb->mac_len;
1300 
1301 			err = push_mpls(skb, key, mpls->mpls_lse,
1302 					mpls->mpls_ethertype, mac_len);
1303 			break;
1304 		}
1305 		case OVS_ACTION_ATTR_POP_MPLS:
1306 			err = pop_mpls(skb, key, nla_get_be16(a));
1307 			break;
1308 
1309 		case OVS_ACTION_ATTR_PUSH_VLAN:
1310 			err = push_vlan(skb, key, nla_data(a));
1311 			break;
1312 
1313 		case OVS_ACTION_ATTR_POP_VLAN:
1314 			err = pop_vlan(skb, key);
1315 			break;
1316 
1317 		case OVS_ACTION_ATTR_RECIRC: {
1318 			bool last = nla_is_last(a, rem);
1319 
1320 			err = execute_recirc(dp, skb, key, a, last);
1321 			if (last) {
1322 				/* If this is the last action, the skb has
1323 				 * been consumed or freed.
1324 				 * Return immediately.
1325 				 */
1326 				return err;
1327 			}
1328 			break;
1329 		}
1330 
1331 		case OVS_ACTION_ATTR_SET:
1332 			err = execute_set_action(skb, key, nla_data(a));
1333 			break;
1334 
1335 		case OVS_ACTION_ATTR_SET_MASKED:
1336 		case OVS_ACTION_ATTR_SET_TO_MASKED:
1337 			err = execute_masked_set_action(skb, key, nla_data(a));
1338 			break;
1339 
1340 		case OVS_ACTION_ATTR_SAMPLE: {
1341 			bool last = nla_is_last(a, rem);
1342 
1343 			err = sample(dp, skb, key, a, last);
1344 			if (last)
1345 				return err;
1346 
1347 			break;
1348 		}
1349 
1350 		case OVS_ACTION_ATTR_CT:
1351 			if (!is_flow_key_valid(key)) {
1352 				err = ovs_flow_key_update(skb, key);
1353 				if (err)
1354 					return err;
1355 			}
1356 
1357 			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1358 					     nla_data(a));
1359 
1360 			/* Hide stolen IP fragments from user space. */
1361 			if (err)
1362 				return err == -EINPROGRESS ? 0 : err;
1363 			break;
1364 
1365 		case OVS_ACTION_ATTR_CT_CLEAR:
1366 			err = ovs_ct_clear(skb, key);
1367 			break;
1368 
1369 		case OVS_ACTION_ATTR_PUSH_ETH:
1370 			err = push_eth(skb, key, nla_data(a));
1371 			break;
1372 
1373 		case OVS_ACTION_ATTR_POP_ETH:
1374 			err = pop_eth(skb, key);
1375 			break;
1376 
1377 		case OVS_ACTION_ATTR_PUSH_NSH: {
1378 			u8 buffer[NSH_HDR_MAX_LEN];
1379 			struct nshhdr *nh = (struct nshhdr *)buffer;
1380 
1381 			err = nsh_hdr_from_nlattr(nla_data(a), nh,
1382 						  NSH_HDR_MAX_LEN);
1383 			if (unlikely(err))
1384 				break;
1385 			err = push_nsh(skb, key, nh);
1386 			break;
1387 		}
1388 
1389 		case OVS_ACTION_ATTR_POP_NSH:
1390 			err = pop_nsh(skb, key);
1391 			break;
1392 
1393 		case OVS_ACTION_ATTR_METER:
1394 			if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1395 				consume_skb(skb);
1396 				return 0;
1397 			}
1398 			break;
1399 
1400 		case OVS_ACTION_ATTR_CLONE: {
1401 			bool last = nla_is_last(a, rem);
1402 
1403 			err = clone(dp, skb, key, a, last);
1404 			if (last)
1405 				return err;
1406 
1407 			break;
1408 		}
1409 
1410 		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1411 			bool last = nla_is_last(a, rem);
1412 
1413 			err = execute_check_pkt_len(dp, skb, key, a, last);
1414 			if (last)
1415 				return err;
1416 
1417 			break;
1418 		}
1419 
1420 		case OVS_ACTION_ATTR_DEC_TTL:
1421 			err = execute_dec_ttl(skb, key);
1422 			if (err == -EHOSTUNREACH) {
1423 				err = dec_ttl_exception_handler(dp, skb, key,
1424 								a, true);
1425 				return err;
1426 			}
1427 			break;
1428 		}
1429 
1430 		if (unlikely(err)) {
1431 			kfree_skb(skb);
1432 			return err;
1433 		}
1434 	}
1435 
1436 	consume_skb(skb);
1437 	return 0;
1438 }
1439 
1440 /* Execute the actions on the clone of the packet. The effect of the
1441  * execution does not affect the original 'skb' nor the original 'key'.
1442  *
1443  * The execution may be deferred in case the actions can not be executed
1444  * immediately.
1445  */
1446 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1447 			 struct sw_flow_key *key, u32 recirc_id,
1448 			 const struct nlattr *actions, int len,
1449 			 bool last, bool clone_flow_key)
1450 {
1451 	struct deferred_action *da;
1452 	struct sw_flow_key *clone;
1453 
1454 	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1455 	if (!skb) {
1456 		/* Out of memory, skip this action.
1457 		 */
1458 		return 0;
1459 	}
1460 
1461 	/* When clone_flow_key is false, the 'key' will not be change
1462 	 * by the actions, then the 'key' can be used directly.
1463 	 * Otherwise, try to clone key from the next recursion level of
1464 	 * 'flow_keys'. If clone is successful, execute the actions
1465 	 * without deferring.
1466 	 */
1467 	clone = clone_flow_key ? clone_key(key) : key;
1468 	if (clone) {
1469 		int err = 0;
1470 
1471 		if (actions) { /* Sample action */
1472 			if (clone_flow_key)
1473 				__this_cpu_inc(exec_actions_level);
1474 
1475 			err = do_execute_actions(dp, skb, clone,
1476 						 actions, len);
1477 
1478 			if (clone_flow_key)
1479 				__this_cpu_dec(exec_actions_level);
1480 		} else { /* Recirc action */
1481 			clone->recirc_id = recirc_id;
1482 			ovs_dp_process_packet(skb, clone);
1483 		}
1484 		return err;
1485 	}
1486 
1487 	/* Out of 'flow_keys' space. Defer actions */
1488 	da = add_deferred_actions(skb, key, actions, len);
1489 	if (da) {
1490 		if (!actions) { /* Recirc action */
1491 			key = &da->pkt_key;
1492 			key->recirc_id = recirc_id;
1493 		}
1494 	} else {
1495 		/* Out of per CPU action FIFO space. Drop the 'skb' and
1496 		 * log an error.
1497 		 */
1498 		kfree_skb(skb);
1499 
1500 		if (net_ratelimit()) {
1501 			if (actions) { /* Sample action */
1502 				pr_warn("%s: deferred action limit reached, drop sample action\n",
1503 					ovs_dp_name(dp));
1504 			} else {  /* Recirc action */
1505 				pr_warn("%s: deferred action limit reached, drop recirc action\n",
1506 					ovs_dp_name(dp));
1507 			}
1508 		}
1509 	}
1510 	return 0;
1511 }
1512 
1513 static void process_deferred_actions(struct datapath *dp)
1514 {
1515 	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1516 
1517 	/* Do not touch the FIFO in case there is no deferred actions. */
1518 	if (action_fifo_is_empty(fifo))
1519 		return;
1520 
1521 	/* Finishing executing all deferred actions. */
1522 	do {
1523 		struct deferred_action *da = action_fifo_get(fifo);
1524 		struct sk_buff *skb = da->skb;
1525 		struct sw_flow_key *key = &da->pkt_key;
1526 		const struct nlattr *actions = da->actions;
1527 		int actions_len = da->actions_len;
1528 
1529 		if (actions)
1530 			do_execute_actions(dp, skb, key, actions, actions_len);
1531 		else
1532 			ovs_dp_process_packet(skb, key);
1533 	} while (!action_fifo_is_empty(fifo));
1534 
1535 	/* Reset FIFO for the next packet.  */
1536 	action_fifo_init(fifo);
1537 }
1538 
1539 /* Execute a list of actions against 'skb'. */
1540 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1541 			const struct sw_flow_actions *acts,
1542 			struct sw_flow_key *key)
1543 {
1544 	int err, level;
1545 
1546 	level = __this_cpu_inc_return(exec_actions_level);
1547 	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1548 		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1549 				     ovs_dp_name(dp));
1550 		kfree_skb(skb);
1551 		err = -ENETDOWN;
1552 		goto out;
1553 	}
1554 
1555 	OVS_CB(skb)->acts_origlen = acts->orig_len;
1556 	err = do_execute_actions(dp, skb, key,
1557 				 acts->actions, acts->actions_len);
1558 
1559 	if (level == 1)
1560 		process_deferred_actions(dp);
1561 
1562 out:
1563 	__this_cpu_dec(exec_actions_level);
1564 	return err;
1565 }
1566 
1567 int action_fifos_init(void)
1568 {
1569 	action_fifos = alloc_percpu(struct action_fifo);
1570 	if (!action_fifos)
1571 		return -ENOMEM;
1572 
1573 	flow_keys = alloc_percpu(struct action_flow_keys);
1574 	if (!flow_keys) {
1575 		free_percpu(action_fifos);
1576 		return -ENOMEM;
1577 	}
1578 
1579 	return 0;
1580 }
1581 
1582 void action_fifos_exit(void)
1583 {
1584 	free_percpu(action_fifos);
1585 	free_percpu(flow_keys);
1586 }
1587