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/gso.h>
21 #include <net/ip.h>
22 #include <net/ipv6.h>
23 #include <net/ip6_fib.h>
24 #include <net/checksum.h>
25 #include <net/dsfield.h>
26 #include <net/mpls.h>
27 
28 #if IS_ENABLED(CONFIG_PSAMPLE)
29 #include <net/psample.h>
30 #endif
31 
32 #include <net/sctp/checksum.h>
33 
34 #include "datapath.h"
35 #include "drop.h"
36 #include "flow.h"
37 #include "conntrack.h"
38 #include "vport.h"
39 #include "flow_netlink.h"
40 #include "openvswitch_trace.h"
41 
42 struct ovs_pcpu_storage __percpu *ovs_pcpu_storage;
43 
44 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
45  * space. Return NULL if out of key spaces.
46  */
47 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
48 {
49 	struct ovs_pcpu_storage *ovs_pcpu = this_cpu_ptr(ovs_pcpu_storage);
50 	struct action_flow_keys *keys = &ovs_pcpu->flow_keys;
51 	int level = ovs_pcpu->exec_level;
52 	struct sw_flow_key *key = NULL;
53 
54 	if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
55 		key = &keys->key[level - 1];
56 		*key = *key_;
57 	}
58 
59 	return key;
60 }
61 
62 static void action_fifo_init(struct action_fifo *fifo)
63 {
64 	fifo->head = 0;
65 	fifo->tail = 0;
66 }
67 
68 static bool action_fifo_is_empty(const struct action_fifo *fifo)
69 {
70 	return (fifo->head == fifo->tail);
71 }
72 
73 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
74 {
75 	if (action_fifo_is_empty(fifo))
76 		return NULL;
77 
78 	return &fifo->fifo[fifo->tail++];
79 }
80 
81 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
82 {
83 	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
84 		return NULL;
85 
86 	return &fifo->fifo[fifo->head++];
87 }
88 
89 /* Return true if fifo is not full */
90 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
91 				    const struct sw_flow_key *key,
92 				    const struct nlattr *actions,
93 				    const int actions_len)
94 {
95 	struct action_fifo *fifo = this_cpu_ptr(&ovs_pcpu_storage->action_fifos);
96 	struct deferred_action *da;
97 
98 	da = action_fifo_put(fifo);
99 	if (da) {
100 		da->skb = skb;
101 		da->actions = actions;
102 		da->actions_len = actions_len;
103 		da->pkt_key = *key;
104 	}
105 
106 	return da;
107 }
108 
109 static void invalidate_flow_key(struct sw_flow_key *key)
110 {
111 	key->mac_proto |= SW_FLOW_KEY_INVALID;
112 }
113 
114 static bool is_flow_key_valid(const struct sw_flow_key *key)
115 {
116 	return !(key->mac_proto & SW_FLOW_KEY_INVALID);
117 }
118 
119 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
120 			 struct sw_flow_key *key,
121 			 u32 recirc_id,
122 			 const struct nlattr *actions, int len,
123 			 bool last, bool clone_flow_key);
124 
125 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
126 			      struct sw_flow_key *key,
127 			      const struct nlattr *attr, int len);
128 
129 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
130 		     __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
131 {
132 	int err;
133 
134 	err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
135 	if (err)
136 		return err;
137 
138 	if (!mac_len)
139 		key->mac_proto = MAC_PROTO_NONE;
140 
141 	invalidate_flow_key(key);
142 	return 0;
143 }
144 
145 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
146 		    const __be16 ethertype)
147 {
148 	int err;
149 
150 	err = skb_mpls_pop(skb, ethertype, skb->mac_len,
151 			   ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
152 	if (err)
153 		return err;
154 
155 	if (ethertype == htons(ETH_P_TEB))
156 		key->mac_proto = MAC_PROTO_ETHERNET;
157 
158 	invalidate_flow_key(key);
159 	return 0;
160 }
161 
162 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
163 		    const __be32 *mpls_lse, const __be32 *mask)
164 {
165 	struct mpls_shim_hdr *stack;
166 	__be32 lse;
167 	int err;
168 
169 	if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
170 		return -ENOMEM;
171 
172 	stack = mpls_hdr(skb);
173 	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
174 	err = skb_mpls_update_lse(skb, lse);
175 	if (err)
176 		return err;
177 
178 	flow_key->mpls.lse[0] = lse;
179 	return 0;
180 }
181 
182 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
183 {
184 	int err;
185 
186 	err = skb_vlan_pop(skb);
187 	if (skb_vlan_tag_present(skb)) {
188 		invalidate_flow_key(key);
189 	} else {
190 		key->eth.vlan.tci = 0;
191 		key->eth.vlan.tpid = 0;
192 	}
193 	return err;
194 }
195 
196 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
197 		     const struct ovs_action_push_vlan *vlan)
198 {
199 	int err;
200 
201 	if (skb_vlan_tag_present(skb)) {
202 		invalidate_flow_key(key);
203 	} else {
204 		key->eth.vlan.tci = vlan->vlan_tci;
205 		key->eth.vlan.tpid = vlan->vlan_tpid;
206 	}
207 	err = skb_vlan_push(skb, vlan->vlan_tpid,
208 			    ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
209 	skb_reset_mac_len(skb);
210 	return err;
211 }
212 
213 /* 'src' is already properly masked. */
214 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
215 {
216 	u16 *dst = (u16 *)dst_;
217 	const u16 *src = (const u16 *)src_;
218 	const u16 *mask = (const u16 *)mask_;
219 
220 	OVS_SET_MASKED(dst[0], src[0], mask[0]);
221 	OVS_SET_MASKED(dst[1], src[1], mask[1]);
222 	OVS_SET_MASKED(dst[2], src[2], mask[2]);
223 }
224 
225 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
226 			const struct ovs_key_ethernet *key,
227 			const struct ovs_key_ethernet *mask)
228 {
229 	int err;
230 
231 	err = skb_ensure_writable(skb, ETH_HLEN);
232 	if (unlikely(err))
233 		return err;
234 
235 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
236 
237 	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
238 			       mask->eth_src);
239 	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
240 			       mask->eth_dst);
241 
242 	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
243 
244 	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
245 	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
246 	return 0;
247 }
248 
249 /* pop_eth does not support VLAN packets as this action is never called
250  * for them.
251  */
252 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
253 {
254 	int err;
255 
256 	err = skb_eth_pop(skb);
257 	if (err)
258 		return err;
259 
260 	/* safe right before invalidate_flow_key */
261 	key->mac_proto = MAC_PROTO_NONE;
262 	invalidate_flow_key(key);
263 	return 0;
264 }
265 
266 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
267 		    const struct ovs_action_push_eth *ethh)
268 {
269 	int err;
270 
271 	err = skb_eth_push(skb, ethh->addresses.eth_dst,
272 			   ethh->addresses.eth_src);
273 	if (err)
274 		return err;
275 
276 	/* safe right before invalidate_flow_key */
277 	key->mac_proto = MAC_PROTO_ETHERNET;
278 	invalidate_flow_key(key);
279 	return 0;
280 }
281 
282 static noinline_for_stack int push_nsh(struct sk_buff *skb,
283 				       struct sw_flow_key *key,
284 				       const struct nlattr *a)
285 {
286 	u8 buffer[NSH_HDR_MAX_LEN];
287 	struct nshhdr *nh = (struct nshhdr *)buffer;
288 	int err;
289 
290 	err = nsh_hdr_from_nlattr(a, nh, NSH_HDR_MAX_LEN);
291 	if (err)
292 		return err;
293 
294 	err = nsh_push(skb, nh);
295 	if (err)
296 		return err;
297 
298 	/* safe right before invalidate_flow_key */
299 	key->mac_proto = MAC_PROTO_NONE;
300 	invalidate_flow_key(key);
301 	return 0;
302 }
303 
304 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
305 {
306 	int err;
307 
308 	err = nsh_pop(skb);
309 	if (err)
310 		return err;
311 
312 	/* safe right before invalidate_flow_key */
313 	if (skb->protocol == htons(ETH_P_TEB))
314 		key->mac_proto = MAC_PROTO_ETHERNET;
315 	else
316 		key->mac_proto = MAC_PROTO_NONE;
317 	invalidate_flow_key(key);
318 	return 0;
319 }
320 
321 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
322 				  __be32 addr, __be32 new_addr)
323 {
324 	int transport_len = skb->len - skb_transport_offset(skb);
325 
326 	if (nh->frag_off & htons(IP_OFFSET))
327 		return;
328 
329 	if (nh->protocol == IPPROTO_TCP) {
330 		if (likely(transport_len >= sizeof(struct tcphdr)))
331 			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
332 						 addr, new_addr, true);
333 	} else if (nh->protocol == IPPROTO_UDP) {
334 		if (likely(transport_len >= sizeof(struct udphdr))) {
335 			struct udphdr *uh = udp_hdr(skb);
336 
337 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
338 				inet_proto_csum_replace4(&uh->check, skb,
339 							 addr, new_addr, true);
340 				if (!uh->check)
341 					uh->check = CSUM_MANGLED_0;
342 			}
343 		}
344 	}
345 }
346 
347 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
348 			__be32 *addr, __be32 new_addr)
349 {
350 	update_ip_l4_checksum(skb, nh, *addr, new_addr);
351 	csum_replace4(&nh->check, *addr, new_addr);
352 	skb_clear_hash(skb);
353 	ovs_ct_clear(skb, NULL);
354 	*addr = new_addr;
355 }
356 
357 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
358 				 __be32 addr[4], const __be32 new_addr[4])
359 {
360 	int transport_len = skb->len - skb_transport_offset(skb);
361 
362 	if (l4_proto == NEXTHDR_TCP) {
363 		if (likely(transport_len >= sizeof(struct tcphdr)))
364 			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
365 						  addr, new_addr, true);
366 	} else if (l4_proto == NEXTHDR_UDP) {
367 		if (likely(transport_len >= sizeof(struct udphdr))) {
368 			struct udphdr *uh = udp_hdr(skb);
369 
370 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
371 				inet_proto_csum_replace16(&uh->check, skb,
372 							  addr, new_addr, true);
373 				if (!uh->check)
374 					uh->check = CSUM_MANGLED_0;
375 			}
376 		}
377 	} else if (l4_proto == NEXTHDR_ICMP) {
378 		if (likely(transport_len >= sizeof(struct icmp6hdr)))
379 			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
380 						  skb, addr, new_addr, true);
381 	}
382 }
383 
384 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
385 			   const __be32 mask[4], __be32 masked[4])
386 {
387 	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
388 	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
389 	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
390 	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
391 }
392 
393 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
394 			  __be32 addr[4], const __be32 new_addr[4],
395 			  bool recalculate_csum)
396 {
397 	if (recalculate_csum)
398 		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
399 
400 	skb_clear_hash(skb);
401 	ovs_ct_clear(skb, NULL);
402 	memcpy(addr, new_addr, sizeof(__be32[4]));
403 }
404 
405 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
406 {
407 	u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
408 
409 	ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
410 
411 	if (skb->ip_summed == CHECKSUM_COMPLETE)
412 		csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
413 			     (__force __wsum)(ipv6_tclass << 12));
414 
415 	ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
416 }
417 
418 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
419 {
420 	u32 ofl;
421 
422 	ofl = nh->flow_lbl[0] << 16 |  nh->flow_lbl[1] << 8 |  nh->flow_lbl[2];
423 	fl = OVS_MASKED(ofl, fl, mask);
424 
425 	/* Bits 21-24 are always unmasked, so this retains their values. */
426 	nh->flow_lbl[0] = (u8)(fl >> 16);
427 	nh->flow_lbl[1] = (u8)(fl >> 8);
428 	nh->flow_lbl[2] = (u8)fl;
429 
430 	if (skb->ip_summed == CHECKSUM_COMPLETE)
431 		csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
432 }
433 
434 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
435 {
436 	new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
437 
438 	if (skb->ip_summed == CHECKSUM_COMPLETE)
439 		csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
440 			     (__force __wsum)(new_ttl << 8));
441 	nh->hop_limit = new_ttl;
442 }
443 
444 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
445 		       u8 mask)
446 {
447 	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
448 
449 	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
450 	nh->ttl = new_ttl;
451 }
452 
453 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
454 		    const struct ovs_key_ipv4 *key,
455 		    const struct ovs_key_ipv4 *mask)
456 {
457 	struct iphdr *nh;
458 	__be32 new_addr;
459 	int err;
460 
461 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
462 				  sizeof(struct iphdr));
463 	if (unlikely(err))
464 		return err;
465 
466 	nh = ip_hdr(skb);
467 
468 	/* Setting an IP addresses is typically only a side effect of
469 	 * matching on them in the current userspace implementation, so it
470 	 * makes sense to check if the value actually changed.
471 	 */
472 	if (mask->ipv4_src) {
473 		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
474 
475 		if (unlikely(new_addr != nh->saddr)) {
476 			set_ip_addr(skb, nh, &nh->saddr, new_addr);
477 			flow_key->ipv4.addr.src = new_addr;
478 		}
479 	}
480 	if (mask->ipv4_dst) {
481 		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
482 
483 		if (unlikely(new_addr != nh->daddr)) {
484 			set_ip_addr(skb, nh, &nh->daddr, new_addr);
485 			flow_key->ipv4.addr.dst = new_addr;
486 		}
487 	}
488 	if (mask->ipv4_tos) {
489 		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
490 		flow_key->ip.tos = nh->tos;
491 	}
492 	if (mask->ipv4_ttl) {
493 		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
494 		flow_key->ip.ttl = nh->ttl;
495 	}
496 
497 	return 0;
498 }
499 
500 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
501 {
502 	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
503 }
504 
505 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
506 		    const struct ovs_key_ipv6 *key,
507 		    const struct ovs_key_ipv6 *mask)
508 {
509 	struct ipv6hdr *nh;
510 	int err;
511 
512 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
513 				  sizeof(struct ipv6hdr));
514 	if (unlikely(err))
515 		return err;
516 
517 	nh = ipv6_hdr(skb);
518 
519 	/* Setting an IP addresses is typically only a side effect of
520 	 * matching on them in the current userspace implementation, so it
521 	 * makes sense to check if the value actually changed.
522 	 */
523 	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
524 		__be32 *saddr = (__be32 *)&nh->saddr;
525 		__be32 masked[4];
526 
527 		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
528 
529 		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
530 			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
531 				      true);
532 			memcpy(&flow_key->ipv6.addr.src, masked,
533 			       sizeof(flow_key->ipv6.addr.src));
534 		}
535 	}
536 	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
537 		unsigned int offset = 0;
538 		int flags = IP6_FH_F_SKIP_RH;
539 		bool recalc_csum = true;
540 		__be32 *daddr = (__be32 *)&nh->daddr;
541 		__be32 masked[4];
542 
543 		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
544 
545 		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
546 			if (ipv6_ext_hdr(nh->nexthdr))
547 				recalc_csum = (ipv6_find_hdr(skb, &offset,
548 							     NEXTHDR_ROUTING,
549 							     NULL, &flags)
550 					       != NEXTHDR_ROUTING);
551 
552 			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
553 				      recalc_csum);
554 			memcpy(&flow_key->ipv6.addr.dst, masked,
555 			       sizeof(flow_key->ipv6.addr.dst));
556 		}
557 	}
558 	if (mask->ipv6_tclass) {
559 		set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
560 		flow_key->ip.tos = ipv6_get_dsfield(nh);
561 	}
562 	if (mask->ipv6_label) {
563 		set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
564 			    ntohl(mask->ipv6_label));
565 		flow_key->ipv6.label =
566 		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
567 	}
568 	if (mask->ipv6_hlimit) {
569 		set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
570 		flow_key->ip.ttl = nh->hop_limit;
571 	}
572 	return 0;
573 }
574 
575 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
576 		   const struct nlattr *a)
577 {
578 	struct nshhdr *nh;
579 	size_t length;
580 	int err;
581 	u8 flags;
582 	u8 ttl;
583 	int i;
584 
585 	struct ovs_key_nsh key;
586 	struct ovs_key_nsh mask;
587 
588 	err = nsh_key_from_nlattr(a, &key, &mask);
589 	if (err)
590 		return err;
591 
592 	/* Make sure the NSH base header is there */
593 	if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
594 		return -ENOMEM;
595 
596 	nh = nsh_hdr(skb);
597 	length = nsh_hdr_len(nh);
598 
599 	/* Make sure the whole NSH header is there */
600 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
601 				       length);
602 	if (unlikely(err))
603 		return err;
604 
605 	nh = nsh_hdr(skb);
606 	skb_postpull_rcsum(skb, nh, length);
607 	flags = nsh_get_flags(nh);
608 	flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
609 	flow_key->nsh.base.flags = flags;
610 	ttl = nsh_get_ttl(nh);
611 	ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
612 	flow_key->nsh.base.ttl = ttl;
613 	nsh_set_flags_and_ttl(nh, flags, ttl);
614 	nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
615 				  mask.base.path_hdr);
616 	flow_key->nsh.base.path_hdr = nh->path_hdr;
617 	switch (nh->mdtype) {
618 	case NSH_M_TYPE1:
619 		for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
620 			nh->md1.context[i] =
621 			    OVS_MASKED(nh->md1.context[i], key.context[i],
622 				       mask.context[i]);
623 		}
624 		memcpy(flow_key->nsh.context, nh->md1.context,
625 		       sizeof(nh->md1.context));
626 		break;
627 	case NSH_M_TYPE2:
628 		memset(flow_key->nsh.context, 0,
629 		       sizeof(flow_key->nsh.context));
630 		break;
631 	default:
632 		return -EINVAL;
633 	}
634 	skb_postpush_rcsum(skb, nh, length);
635 	return 0;
636 }
637 
638 /* Must follow skb_ensure_writable() since that can move the skb data. */
639 static void set_tp_port(struct sk_buff *skb, __be16 *port,
640 			__be16 new_port, __sum16 *check)
641 {
642 	ovs_ct_clear(skb, NULL);
643 	inet_proto_csum_replace2(check, skb, *port, new_port, false);
644 	*port = new_port;
645 }
646 
647 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
648 		   const struct ovs_key_udp *key,
649 		   const struct ovs_key_udp *mask)
650 {
651 	struct udphdr *uh;
652 	__be16 src, dst;
653 	int err;
654 
655 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
656 				  sizeof(struct udphdr));
657 	if (unlikely(err))
658 		return err;
659 
660 	uh = udp_hdr(skb);
661 	/* Either of the masks is non-zero, so do not bother checking them. */
662 	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
663 	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
664 
665 	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
666 		if (likely(src != uh->source)) {
667 			set_tp_port(skb, &uh->source, src, &uh->check);
668 			flow_key->tp.src = src;
669 		}
670 		if (likely(dst != uh->dest)) {
671 			set_tp_port(skb, &uh->dest, dst, &uh->check);
672 			flow_key->tp.dst = dst;
673 		}
674 
675 		if (unlikely(!uh->check))
676 			uh->check = CSUM_MANGLED_0;
677 	} else {
678 		uh->source = src;
679 		uh->dest = dst;
680 		flow_key->tp.src = src;
681 		flow_key->tp.dst = dst;
682 		ovs_ct_clear(skb, NULL);
683 	}
684 
685 	skb_clear_hash(skb);
686 
687 	return 0;
688 }
689 
690 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
691 		   const struct ovs_key_tcp *key,
692 		   const struct ovs_key_tcp *mask)
693 {
694 	struct tcphdr *th;
695 	__be16 src, dst;
696 	int err;
697 
698 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
699 				  sizeof(struct tcphdr));
700 	if (unlikely(err))
701 		return err;
702 
703 	th = tcp_hdr(skb);
704 	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
705 	if (likely(src != th->source)) {
706 		set_tp_port(skb, &th->source, src, &th->check);
707 		flow_key->tp.src = src;
708 	}
709 	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
710 	if (likely(dst != th->dest)) {
711 		set_tp_port(skb, &th->dest, dst, &th->check);
712 		flow_key->tp.dst = dst;
713 	}
714 	skb_clear_hash(skb);
715 
716 	return 0;
717 }
718 
719 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
720 		    const struct ovs_key_sctp *key,
721 		    const struct ovs_key_sctp *mask)
722 {
723 	unsigned int sctphoff = skb_transport_offset(skb);
724 	struct sctphdr *sh;
725 	__le32 old_correct_csum, new_csum, old_csum;
726 	int err;
727 
728 	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
729 	if (unlikely(err))
730 		return err;
731 
732 	sh = sctp_hdr(skb);
733 	old_csum = sh->checksum;
734 	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
735 
736 	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
737 	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
738 
739 	new_csum = sctp_compute_cksum(skb, sctphoff);
740 
741 	/* Carry any checksum errors through. */
742 	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
743 
744 	skb_clear_hash(skb);
745 	ovs_ct_clear(skb, NULL);
746 
747 	flow_key->tp.src = sh->source;
748 	flow_key->tp.dst = sh->dest;
749 
750 	return 0;
751 }
752 
753 static int ovs_vport_output(struct net *net, struct sock *sk,
754 			    struct sk_buff *skb)
755 {
756 	struct ovs_frag_data *data = this_cpu_ptr(&ovs_pcpu_storage->frag_data);
757 	struct vport *vport = data->vport;
758 
759 	if (skb_cow_head(skb, data->l2_len) < 0) {
760 		kfree_skb_reason(skb, SKB_DROP_REASON_NOMEM);
761 		return -ENOMEM;
762 	}
763 
764 	__skb_dst_copy(skb, data->dst);
765 	*OVS_CB(skb) = data->cb;
766 	skb->inner_protocol = data->inner_protocol;
767 	if (data->vlan_tci & VLAN_CFI_MASK)
768 		__vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
769 	else
770 		__vlan_hwaccel_clear_tag(skb);
771 
772 	/* Reconstruct the MAC header.  */
773 	skb_push(skb, data->l2_len);
774 	memcpy(skb->data, &data->l2_data, data->l2_len);
775 	skb_postpush_rcsum(skb, skb->data, data->l2_len);
776 	skb_reset_mac_header(skb);
777 
778 	if (eth_p_mpls(skb->protocol)) {
779 		skb->inner_network_header = skb->network_header;
780 		skb_set_network_header(skb, data->network_offset);
781 		skb_reset_mac_len(skb);
782 	}
783 
784 	ovs_vport_send(vport, skb, data->mac_proto);
785 	return 0;
786 }
787 
788 static unsigned int
789 ovs_dst_get_mtu(const struct dst_entry *dst)
790 {
791 	return dst->dev->mtu;
792 }
793 
794 static struct dst_ops ovs_dst_ops = {
795 	.family = AF_UNSPEC,
796 	.mtu = ovs_dst_get_mtu,
797 };
798 
799 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
800  * ovs_vport_output(), which is called once per fragmented packet.
801  */
802 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
803 			 u16 orig_network_offset, u8 mac_proto)
804 {
805 	unsigned int hlen = skb_network_offset(skb);
806 	struct ovs_frag_data *data;
807 
808 	data = this_cpu_ptr(&ovs_pcpu_storage->frag_data);
809 	data->dst = skb->_skb_refdst;
810 	data->vport = vport;
811 	data->cb = *OVS_CB(skb);
812 	data->inner_protocol = skb->inner_protocol;
813 	data->network_offset = orig_network_offset;
814 	if (skb_vlan_tag_present(skb))
815 		data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
816 	else
817 		data->vlan_tci = 0;
818 	data->vlan_proto = skb->vlan_proto;
819 	data->mac_proto = mac_proto;
820 	data->l2_len = hlen;
821 	memcpy(&data->l2_data, skb->data, hlen);
822 
823 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
824 	skb_pull(skb, hlen);
825 }
826 
827 static void ovs_fragment(struct net *net, struct vport *vport,
828 			 struct sk_buff *skb, u16 mru,
829 			 struct sw_flow_key *key)
830 {
831 	enum ovs_drop_reason reason;
832 	u16 orig_network_offset = 0;
833 
834 	if (eth_p_mpls(skb->protocol)) {
835 		orig_network_offset = skb_network_offset(skb);
836 		skb->network_header = skb->inner_network_header;
837 	}
838 
839 	if (skb_network_offset(skb) > MAX_L2_LEN) {
840 		OVS_NLERR(1, "L2 header too long to fragment");
841 		reason = OVS_DROP_FRAG_L2_TOO_LONG;
842 		goto err;
843 	}
844 
845 	if (key->eth.type == htons(ETH_P_IP)) {
846 		struct rtable ovs_rt = { 0 };
847 		unsigned long orig_dst;
848 
849 		prepare_frag(vport, skb, orig_network_offset,
850 			     ovs_key_mac_proto(key));
851 		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL,
852 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
853 		ovs_rt.dst.dev = vport->dev;
854 
855 		orig_dst = skb->_skb_refdst;
856 		skb_dst_set_noref(skb, &ovs_rt.dst);
857 		IPCB(skb)->frag_max_size = mru;
858 
859 		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
860 		refdst_drop(orig_dst);
861 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
862 		unsigned long orig_dst;
863 		struct rt6_info ovs_rt;
864 
865 		prepare_frag(vport, skb, orig_network_offset,
866 			     ovs_key_mac_proto(key));
867 		memset(&ovs_rt, 0, sizeof(ovs_rt));
868 		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL,
869 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
870 		ovs_rt.dst.dev = vport->dev;
871 
872 		orig_dst = skb->_skb_refdst;
873 		skb_dst_set_noref(skb, &ovs_rt.dst);
874 		IP6CB(skb)->frag_max_size = mru;
875 
876 		ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
877 		refdst_drop(orig_dst);
878 	} else {
879 		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
880 			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
881 			  vport->dev->mtu);
882 		reason = OVS_DROP_FRAG_INVALID_PROTO;
883 		goto err;
884 	}
885 
886 	return;
887 err:
888 	ovs_kfree_skb_reason(skb, reason);
889 }
890 
891 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
892 		      struct sw_flow_key *key)
893 {
894 	struct vport *vport = ovs_vport_rcu(dp, out_port);
895 
896 	if (likely(vport &&
897 		   netif_running(vport->dev) &&
898 		   netif_carrier_ok(vport->dev))) {
899 		u16 mru = OVS_CB(skb)->mru;
900 		u32 cutlen = OVS_CB(skb)->cutlen;
901 
902 		if (unlikely(cutlen > 0)) {
903 			if (skb->len - cutlen > ovs_mac_header_len(key))
904 				pskb_trim(skb, skb->len - cutlen);
905 			else
906 				pskb_trim(skb, ovs_mac_header_len(key));
907 		}
908 
909 		if (likely(!mru ||
910 		           (skb->len <= mru + vport->dev->hard_header_len))) {
911 			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
912 		} else if (mru <= vport->dev->mtu) {
913 			struct net *net = read_pnet(&dp->net);
914 
915 			ovs_fragment(net, vport, skb, mru, key);
916 		} else {
917 			kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
918 		}
919 	} else {
920 		kfree_skb_reason(skb, SKB_DROP_REASON_DEV_READY);
921 	}
922 }
923 
924 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
925 			    struct sw_flow_key *key, const struct nlattr *attr,
926 			    const struct nlattr *actions, int actions_len,
927 			    uint32_t cutlen)
928 {
929 	struct dp_upcall_info upcall;
930 	const struct nlattr *a;
931 	int rem;
932 
933 	memset(&upcall, 0, sizeof(upcall));
934 	upcall.cmd = OVS_PACKET_CMD_ACTION;
935 	upcall.mru = OVS_CB(skb)->mru;
936 
937 	nla_for_each_nested(a, attr, rem) {
938 		switch (nla_type(a)) {
939 		case OVS_USERSPACE_ATTR_USERDATA:
940 			upcall.userdata = a;
941 			break;
942 
943 		case OVS_USERSPACE_ATTR_PID:
944 			if (dp->user_features &
945 			    OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
946 				upcall.portid =
947 				  ovs_dp_get_upcall_portid(dp,
948 							   smp_processor_id());
949 			else
950 				upcall.portid = nla_get_u32(a);
951 			break;
952 
953 		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
954 			/* Get out tunnel info. */
955 			struct vport *vport;
956 
957 			vport = ovs_vport_rcu(dp, nla_get_u32(a));
958 			if (vport) {
959 				int err;
960 
961 				err = dev_fill_metadata_dst(vport->dev, skb);
962 				if (!err)
963 					upcall.egress_tun_info = skb_tunnel_info(skb);
964 			}
965 
966 			break;
967 		}
968 
969 		case OVS_USERSPACE_ATTR_ACTIONS: {
970 			/* Include actions. */
971 			upcall.actions = actions;
972 			upcall.actions_len = actions_len;
973 			break;
974 		}
975 
976 		} /* End of switch. */
977 	}
978 
979 	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
980 }
981 
982 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
983 				     struct sw_flow_key *key,
984 				     const struct nlattr *attr)
985 {
986 	/* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
987 	struct nlattr *actions = nla_data(attr);
988 
989 	if (nla_len(actions))
990 		return clone_execute(dp, skb, key, 0, nla_data(actions),
991 				     nla_len(actions), true, false);
992 
993 	ovs_kfree_skb_reason(skb, OVS_DROP_IP_TTL);
994 	return 0;
995 }
996 
997 /* When 'last' is true, sample() should always consume the 'skb'.
998  * Otherwise, sample() should keep 'skb' intact regardless what
999  * actions are executed within sample().
1000  */
1001 static int sample(struct datapath *dp, struct sk_buff *skb,
1002 		  struct sw_flow_key *key, const struct nlattr *attr,
1003 		  bool last)
1004 {
1005 	struct nlattr *actions;
1006 	struct nlattr *sample_arg;
1007 	int rem = nla_len(attr);
1008 	const struct sample_arg *arg;
1009 	u32 init_probability;
1010 	bool clone_flow_key;
1011 	int err;
1012 
1013 	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1014 	sample_arg = nla_data(attr);
1015 	arg = nla_data(sample_arg);
1016 	actions = nla_next(sample_arg, &rem);
1017 	init_probability = OVS_CB(skb)->probability;
1018 
1019 	if ((arg->probability != U32_MAX) &&
1020 	    (!arg->probability || get_random_u32() > arg->probability)) {
1021 		if (last)
1022 			ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1023 		return 0;
1024 	}
1025 
1026 	OVS_CB(skb)->probability = arg->probability;
1027 
1028 	clone_flow_key = !arg->exec;
1029 	err = clone_execute(dp, skb, key, 0, actions, rem, last,
1030 			    clone_flow_key);
1031 
1032 	if (!last)
1033 		OVS_CB(skb)->probability = init_probability;
1034 
1035 	return err;
1036 }
1037 
1038 /* When 'last' is true, clone() should always consume the 'skb'.
1039  * Otherwise, clone() should keep 'skb' intact regardless what
1040  * actions are executed within clone().
1041  */
1042 static int clone(struct datapath *dp, struct sk_buff *skb,
1043 		 struct sw_flow_key *key, const struct nlattr *attr,
1044 		 bool last)
1045 {
1046 	struct nlattr *actions;
1047 	struct nlattr *clone_arg;
1048 	int rem = nla_len(attr);
1049 	bool dont_clone_flow_key;
1050 
1051 	/* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1052 	clone_arg = nla_data(attr);
1053 	dont_clone_flow_key = nla_get_u32(clone_arg);
1054 	actions = nla_next(clone_arg, &rem);
1055 
1056 	return clone_execute(dp, skb, key, 0, actions, rem, last,
1057 			     !dont_clone_flow_key);
1058 }
1059 
1060 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1061 			 const struct nlattr *attr)
1062 {
1063 	struct ovs_action_hash *hash_act = nla_data(attr);
1064 	u32 hash = 0;
1065 
1066 	if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
1067 		/* OVS_HASH_ALG_L4 hasing type. */
1068 		hash = skb_get_hash(skb);
1069 	} else if (hash_act->hash_alg == OVS_HASH_ALG_SYM_L4) {
1070 		/* OVS_HASH_ALG_SYM_L4 hashing type.  NOTE: this doesn't
1071 		 * extend past an encapsulated header.
1072 		 */
1073 		hash = __skb_get_hash_symmetric(skb);
1074 	}
1075 
1076 	hash = jhash_1word(hash, hash_act->hash_basis);
1077 	if (!hash)
1078 		hash = 0x1;
1079 
1080 	key->ovs_flow_hash = hash;
1081 }
1082 
1083 static int execute_set_action(struct sk_buff *skb,
1084 			      struct sw_flow_key *flow_key,
1085 			      const struct nlattr *a)
1086 {
1087 	/* Only tunnel set execution is supported without a mask. */
1088 	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1089 		struct ovs_tunnel_info *tun = nla_data(a);
1090 
1091 		skb_dst_drop(skb);
1092 		dst_hold((struct dst_entry *)tun->tun_dst);
1093 		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1094 		return 0;
1095 	}
1096 
1097 	return -EINVAL;
1098 }
1099 
1100 /* Mask is at the midpoint of the data. */
1101 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1102 
1103 static int execute_masked_set_action(struct sk_buff *skb,
1104 				     struct sw_flow_key *flow_key,
1105 				     const struct nlattr *a)
1106 {
1107 	int err = 0;
1108 
1109 	switch (nla_type(a)) {
1110 	case OVS_KEY_ATTR_PRIORITY:
1111 		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1112 			       *get_mask(a, u32 *));
1113 		flow_key->phy.priority = skb->priority;
1114 		break;
1115 
1116 	case OVS_KEY_ATTR_SKB_MARK:
1117 		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1118 		flow_key->phy.skb_mark = skb->mark;
1119 		break;
1120 
1121 	case OVS_KEY_ATTR_TUNNEL_INFO:
1122 		/* Masked data not supported for tunnel. */
1123 		err = -EINVAL;
1124 		break;
1125 
1126 	case OVS_KEY_ATTR_ETHERNET:
1127 		err = set_eth_addr(skb, flow_key, nla_data(a),
1128 				   get_mask(a, struct ovs_key_ethernet *));
1129 		break;
1130 
1131 	case OVS_KEY_ATTR_NSH:
1132 		err = set_nsh(skb, flow_key, a);
1133 		break;
1134 
1135 	case OVS_KEY_ATTR_IPV4:
1136 		err = set_ipv4(skb, flow_key, nla_data(a),
1137 			       get_mask(a, struct ovs_key_ipv4 *));
1138 		break;
1139 
1140 	case OVS_KEY_ATTR_IPV6:
1141 		err = set_ipv6(skb, flow_key, nla_data(a),
1142 			       get_mask(a, struct ovs_key_ipv6 *));
1143 		break;
1144 
1145 	case OVS_KEY_ATTR_TCP:
1146 		err = set_tcp(skb, flow_key, nla_data(a),
1147 			      get_mask(a, struct ovs_key_tcp *));
1148 		break;
1149 
1150 	case OVS_KEY_ATTR_UDP:
1151 		err = set_udp(skb, flow_key, nla_data(a),
1152 			      get_mask(a, struct ovs_key_udp *));
1153 		break;
1154 
1155 	case OVS_KEY_ATTR_SCTP:
1156 		err = set_sctp(skb, flow_key, nla_data(a),
1157 			       get_mask(a, struct ovs_key_sctp *));
1158 		break;
1159 
1160 	case OVS_KEY_ATTR_MPLS:
1161 		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1162 								    __be32 *));
1163 		break;
1164 
1165 	case OVS_KEY_ATTR_CT_STATE:
1166 	case OVS_KEY_ATTR_CT_ZONE:
1167 	case OVS_KEY_ATTR_CT_MARK:
1168 	case OVS_KEY_ATTR_CT_LABELS:
1169 	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1170 	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1171 		err = -EINVAL;
1172 		break;
1173 	}
1174 
1175 	return err;
1176 }
1177 
1178 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1179 			  struct sw_flow_key *key,
1180 			  const struct nlattr *a, bool last)
1181 {
1182 	u32 recirc_id;
1183 
1184 	if (!is_flow_key_valid(key)) {
1185 		int err;
1186 
1187 		err = ovs_flow_key_update(skb, key);
1188 		if (err)
1189 			return err;
1190 	}
1191 	BUG_ON(!is_flow_key_valid(key));
1192 
1193 	recirc_id = nla_get_u32(a);
1194 	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1195 }
1196 
1197 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1198 				 struct sw_flow_key *key,
1199 				 const struct nlattr *attr, bool last)
1200 {
1201 	struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1202 	const struct nlattr *actions, *cpl_arg;
1203 	int len, max_len, rem = nla_len(attr);
1204 	const struct check_pkt_len_arg *arg;
1205 	bool clone_flow_key;
1206 
1207 	/* The first netlink attribute in 'attr' is always
1208 	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1209 	 */
1210 	cpl_arg = nla_data(attr);
1211 	arg = nla_data(cpl_arg);
1212 
1213 	len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1214 	max_len = arg->pkt_len;
1215 
1216 	if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1217 	    len <= max_len) {
1218 		/* Second netlink attribute in 'attr' is always
1219 		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1220 		 */
1221 		actions = nla_next(cpl_arg, &rem);
1222 		clone_flow_key = !arg->exec_for_lesser_equal;
1223 	} else {
1224 		/* Third netlink attribute in 'attr' is always
1225 		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1226 		 */
1227 		actions = nla_next(cpl_arg, &rem);
1228 		actions = nla_next(actions, &rem);
1229 		clone_flow_key = !arg->exec_for_greater;
1230 	}
1231 
1232 	return clone_execute(dp, skb, key, 0, nla_data(actions),
1233 			     nla_len(actions), last, clone_flow_key);
1234 }
1235 
1236 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1237 {
1238 	int err;
1239 
1240 	if (skb->protocol == htons(ETH_P_IPV6)) {
1241 		struct ipv6hdr *nh;
1242 
1243 		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1244 					  sizeof(*nh));
1245 		if (unlikely(err))
1246 			return err;
1247 
1248 		nh = ipv6_hdr(skb);
1249 
1250 		if (nh->hop_limit <= 1)
1251 			return -EHOSTUNREACH;
1252 
1253 		key->ip.ttl = --nh->hop_limit;
1254 	} else if (skb->protocol == htons(ETH_P_IP)) {
1255 		struct iphdr *nh;
1256 		u8 old_ttl;
1257 
1258 		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1259 					  sizeof(*nh));
1260 		if (unlikely(err))
1261 			return err;
1262 
1263 		nh = ip_hdr(skb);
1264 		if (nh->ttl <= 1)
1265 			return -EHOSTUNREACH;
1266 
1267 		old_ttl = nh->ttl--;
1268 		csum_replace2(&nh->check, htons(old_ttl << 8),
1269 			      htons(nh->ttl << 8));
1270 		key->ip.ttl = nh->ttl;
1271 	}
1272 	return 0;
1273 }
1274 
1275 #if IS_ENABLED(CONFIG_PSAMPLE)
1276 static void execute_psample(struct datapath *dp, struct sk_buff *skb,
1277 			    const struct nlattr *attr)
1278 {
1279 	struct psample_group psample_group = {};
1280 	struct psample_metadata md = {};
1281 	const struct nlattr *a;
1282 	u32 rate;
1283 	int rem;
1284 
1285 	nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) {
1286 		switch (nla_type(a)) {
1287 		case OVS_PSAMPLE_ATTR_GROUP:
1288 			psample_group.group_num = nla_get_u32(a);
1289 			break;
1290 
1291 		case OVS_PSAMPLE_ATTR_COOKIE:
1292 			md.user_cookie = nla_data(a);
1293 			md.user_cookie_len = nla_len(a);
1294 			break;
1295 		}
1296 	}
1297 
1298 	psample_group.net = ovs_dp_get_net(dp);
1299 	md.in_ifindex = OVS_CB(skb)->input_vport->dev->ifindex;
1300 	md.trunc_size = skb->len - OVS_CB(skb)->cutlen;
1301 	md.rate_as_probability = 1;
1302 
1303 	rate = OVS_CB(skb)->probability ? OVS_CB(skb)->probability : U32_MAX;
1304 
1305 	psample_sample_packet(&psample_group, skb, rate, &md);
1306 }
1307 #else
1308 static void execute_psample(struct datapath *dp, struct sk_buff *skb,
1309 			    const struct nlattr *attr)
1310 {}
1311 #endif
1312 
1313 /* Execute a list of actions against 'skb'. */
1314 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1315 			      struct sw_flow_key *key,
1316 			      const struct nlattr *attr, int len)
1317 {
1318 	const struct nlattr *a;
1319 	int rem;
1320 
1321 	for (a = attr, rem = len; rem > 0;
1322 	     a = nla_next(a, &rem)) {
1323 		int err = 0;
1324 
1325 		if (trace_ovs_do_execute_action_enabled())
1326 			trace_ovs_do_execute_action(dp, skb, key, a, rem);
1327 
1328 		/* Actions that rightfully have to consume the skb should do it
1329 		 * and return directly.
1330 		 */
1331 		switch (nla_type(a)) {
1332 		case OVS_ACTION_ATTR_OUTPUT: {
1333 			int port = nla_get_u32(a);
1334 			struct sk_buff *clone;
1335 
1336 			/* Every output action needs a separate clone
1337 			 * of 'skb', In case the output action is the
1338 			 * last action, cloning can be avoided.
1339 			 */
1340 			if (nla_is_last(a, rem)) {
1341 				do_output(dp, skb, port, key);
1342 				/* 'skb' has been used for output.
1343 				 */
1344 				return 0;
1345 			}
1346 
1347 			clone = skb_clone(skb, GFP_ATOMIC);
1348 			if (clone)
1349 				do_output(dp, clone, port, key);
1350 			OVS_CB(skb)->cutlen = 0;
1351 			break;
1352 		}
1353 
1354 		case OVS_ACTION_ATTR_TRUNC: {
1355 			struct ovs_action_trunc *trunc = nla_data(a);
1356 
1357 			if (skb->len > trunc->max_len)
1358 				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1359 			break;
1360 		}
1361 
1362 		case OVS_ACTION_ATTR_USERSPACE:
1363 			output_userspace(dp, skb, key, a, attr,
1364 						     len, OVS_CB(skb)->cutlen);
1365 			OVS_CB(skb)->cutlen = 0;
1366 			if (nla_is_last(a, rem)) {
1367 				consume_skb(skb);
1368 				return 0;
1369 			}
1370 			break;
1371 
1372 		case OVS_ACTION_ATTR_HASH:
1373 			execute_hash(skb, key, a);
1374 			break;
1375 
1376 		case OVS_ACTION_ATTR_PUSH_MPLS: {
1377 			struct ovs_action_push_mpls *mpls = nla_data(a);
1378 
1379 			err = push_mpls(skb, key, mpls->mpls_lse,
1380 					mpls->mpls_ethertype, skb->mac_len);
1381 			break;
1382 		}
1383 		case OVS_ACTION_ATTR_ADD_MPLS: {
1384 			struct ovs_action_add_mpls *mpls = nla_data(a);
1385 			__u16 mac_len = 0;
1386 
1387 			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1388 				mac_len = skb->mac_len;
1389 
1390 			err = push_mpls(skb, key, mpls->mpls_lse,
1391 					mpls->mpls_ethertype, mac_len);
1392 			break;
1393 		}
1394 		case OVS_ACTION_ATTR_POP_MPLS:
1395 			err = pop_mpls(skb, key, nla_get_be16(a));
1396 			break;
1397 
1398 		case OVS_ACTION_ATTR_PUSH_VLAN:
1399 			err = push_vlan(skb, key, nla_data(a));
1400 			break;
1401 
1402 		case OVS_ACTION_ATTR_POP_VLAN:
1403 			err = pop_vlan(skb, key);
1404 			break;
1405 
1406 		case OVS_ACTION_ATTR_RECIRC: {
1407 			bool last = nla_is_last(a, rem);
1408 
1409 			err = execute_recirc(dp, skb, key, a, last);
1410 			if (last) {
1411 				/* If this is the last action, the skb has
1412 				 * been consumed or freed.
1413 				 * Return immediately.
1414 				 */
1415 				return err;
1416 			}
1417 			break;
1418 		}
1419 
1420 		case OVS_ACTION_ATTR_SET:
1421 			err = execute_set_action(skb, key, nla_data(a));
1422 			break;
1423 
1424 		case OVS_ACTION_ATTR_SET_MASKED:
1425 		case OVS_ACTION_ATTR_SET_TO_MASKED:
1426 			err = execute_masked_set_action(skb, key, nla_data(a));
1427 			break;
1428 
1429 		case OVS_ACTION_ATTR_SAMPLE: {
1430 			bool last = nla_is_last(a, rem);
1431 
1432 			err = sample(dp, skb, key, a, last);
1433 			if (last)
1434 				return err;
1435 
1436 			break;
1437 		}
1438 
1439 		case OVS_ACTION_ATTR_CT:
1440 			if (!is_flow_key_valid(key)) {
1441 				err = ovs_flow_key_update(skb, key);
1442 				if (err)
1443 					return err;
1444 			}
1445 
1446 			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1447 					     nla_data(a));
1448 
1449 			/* Hide stolen IP fragments from user space. */
1450 			if (err)
1451 				return err == -EINPROGRESS ? 0 : err;
1452 			break;
1453 
1454 		case OVS_ACTION_ATTR_CT_CLEAR:
1455 			err = ovs_ct_clear(skb, key);
1456 			break;
1457 
1458 		case OVS_ACTION_ATTR_PUSH_ETH:
1459 			err = push_eth(skb, key, nla_data(a));
1460 			break;
1461 
1462 		case OVS_ACTION_ATTR_POP_ETH:
1463 			err = pop_eth(skb, key);
1464 			break;
1465 
1466 		case OVS_ACTION_ATTR_PUSH_NSH:
1467 			err = push_nsh(skb, key, nla_data(a));
1468 			break;
1469 
1470 		case OVS_ACTION_ATTR_POP_NSH:
1471 			err = pop_nsh(skb, key);
1472 			break;
1473 
1474 		case OVS_ACTION_ATTR_METER:
1475 			if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1476 				ovs_kfree_skb_reason(skb, OVS_DROP_METER);
1477 				return 0;
1478 			}
1479 			break;
1480 
1481 		case OVS_ACTION_ATTR_CLONE: {
1482 			bool last = nla_is_last(a, rem);
1483 
1484 			err = clone(dp, skb, key, a, last);
1485 			if (last)
1486 				return err;
1487 
1488 			break;
1489 		}
1490 
1491 		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1492 			bool last = nla_is_last(a, rem);
1493 
1494 			err = execute_check_pkt_len(dp, skb, key, a, last);
1495 			if (last)
1496 				return err;
1497 
1498 			break;
1499 		}
1500 
1501 		case OVS_ACTION_ATTR_DEC_TTL:
1502 			err = execute_dec_ttl(skb, key);
1503 			if (err == -EHOSTUNREACH)
1504 				return dec_ttl_exception_handler(dp, skb,
1505 								 key, a);
1506 			break;
1507 
1508 		case OVS_ACTION_ATTR_DROP: {
1509 			enum ovs_drop_reason reason = nla_get_u32(a)
1510 				? OVS_DROP_EXPLICIT_WITH_ERROR
1511 				: OVS_DROP_EXPLICIT;
1512 
1513 			ovs_kfree_skb_reason(skb, reason);
1514 			return 0;
1515 		}
1516 
1517 		case OVS_ACTION_ATTR_PSAMPLE:
1518 			execute_psample(dp, skb, a);
1519 			OVS_CB(skb)->cutlen = 0;
1520 			if (nla_is_last(a, rem)) {
1521 				consume_skb(skb);
1522 				return 0;
1523 			}
1524 			break;
1525 		}
1526 
1527 		if (unlikely(err)) {
1528 			ovs_kfree_skb_reason(skb, OVS_DROP_ACTION_ERROR);
1529 			return err;
1530 		}
1531 	}
1532 
1533 	ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1534 	return 0;
1535 }
1536 
1537 /* Execute the actions on the clone of the packet. The effect of the
1538  * execution does not affect the original 'skb' nor the original 'key'.
1539  *
1540  * The execution may be deferred in case the actions can not be executed
1541  * immediately.
1542  */
1543 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1544 			 struct sw_flow_key *key, u32 recirc_id,
1545 			 const struct nlattr *actions, int len,
1546 			 bool last, bool clone_flow_key)
1547 {
1548 	struct deferred_action *da;
1549 	struct sw_flow_key *clone;
1550 
1551 	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1552 	if (!skb) {
1553 		/* Out of memory, skip this action.
1554 		 */
1555 		return 0;
1556 	}
1557 
1558 	/* When clone_flow_key is false, the 'key' will not be change
1559 	 * by the actions, then the 'key' can be used directly.
1560 	 * Otherwise, try to clone key from the next recursion level of
1561 	 * 'flow_keys'. If clone is successful, execute the actions
1562 	 * without deferring.
1563 	 */
1564 	clone = clone_flow_key ? clone_key(key) : key;
1565 	if (clone) {
1566 		int err = 0;
1567 		if (actions) { /* Sample action */
1568 			if (clone_flow_key)
1569 				__this_cpu_inc(ovs_pcpu_storage->exec_level);
1570 
1571 			err = do_execute_actions(dp, skb, clone,
1572 						 actions, len);
1573 
1574 			if (clone_flow_key)
1575 				__this_cpu_dec(ovs_pcpu_storage->exec_level);
1576 		} else { /* Recirc action */
1577 			clone->recirc_id = recirc_id;
1578 			ovs_dp_process_packet(skb, clone);
1579 		}
1580 		return err;
1581 	}
1582 
1583 	/* Out of 'flow_keys' space. Defer actions */
1584 	da = add_deferred_actions(skb, key, actions, len);
1585 	if (da) {
1586 		if (!actions) { /* Recirc action */
1587 			key = &da->pkt_key;
1588 			key->recirc_id = recirc_id;
1589 		}
1590 	} else {
1591 		/* Out of per CPU action FIFO space. Drop the 'skb' and
1592 		 * log an error.
1593 		 */
1594 		ovs_kfree_skb_reason(skb, OVS_DROP_DEFERRED_LIMIT);
1595 
1596 		if (net_ratelimit()) {
1597 			if (actions) { /* Sample action */
1598 				pr_warn("%s: deferred action limit reached, drop sample action\n",
1599 					ovs_dp_name(dp));
1600 			} else {  /* Recirc action */
1601 				pr_warn("%s: deferred action limit reached, drop recirc action (recirc_id=%#x)\n",
1602 					ovs_dp_name(dp), recirc_id);
1603 			}
1604 		}
1605 	}
1606 	return 0;
1607 }
1608 
1609 static void process_deferred_actions(struct datapath *dp)
1610 {
1611 	struct action_fifo *fifo = this_cpu_ptr(&ovs_pcpu_storage->action_fifos);
1612 
1613 	/* Do not touch the FIFO in case there is no deferred actions. */
1614 	if (action_fifo_is_empty(fifo))
1615 		return;
1616 
1617 	/* Finishing executing all deferred actions. */
1618 	do {
1619 		struct deferred_action *da = action_fifo_get(fifo);
1620 		struct sk_buff *skb = da->skb;
1621 		struct sw_flow_key *key = &da->pkt_key;
1622 		const struct nlattr *actions = da->actions;
1623 		int actions_len = da->actions_len;
1624 
1625 		if (actions)
1626 			do_execute_actions(dp, skb, key, actions, actions_len);
1627 		else
1628 			ovs_dp_process_packet(skb, key);
1629 	} while (!action_fifo_is_empty(fifo));
1630 
1631 	/* Reset FIFO for the next packet.  */
1632 	action_fifo_init(fifo);
1633 }
1634 
1635 /* Execute a list of actions against 'skb'. */
1636 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1637 			const struct sw_flow_actions *acts,
1638 			struct sw_flow_key *key)
1639 {
1640 	int err, level;
1641 
1642 	level = __this_cpu_inc_return(ovs_pcpu_storage->exec_level);
1643 	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1644 		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1645 				     ovs_dp_name(dp));
1646 		ovs_kfree_skb_reason(skb, OVS_DROP_RECURSION_LIMIT);
1647 		err = -ENETDOWN;
1648 		goto out;
1649 	}
1650 
1651 	OVS_CB(skb)->acts_origlen = acts->orig_len;
1652 	err = do_execute_actions(dp, skb, key,
1653 				 acts->actions, acts->actions_len);
1654 
1655 	if (level == 1)
1656 		process_deferred_actions(dp);
1657 
1658 out:
1659 	__this_cpu_dec(ovs_pcpu_storage->exec_level);
1660 	return err;
1661 }
1662