xref: /linux/net/openvswitch/actions.c (revision a634dda26186cf9a51567020fcce52bcba5e1e59)
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 deferred_action {
43 	struct sk_buff *skb;
44 	const struct nlattr *actions;
45 	int actions_len;
46 
47 	/* Store pkt_key clone when creating deferred action. */
48 	struct sw_flow_key pkt_key;
49 };
50 
51 #define MAX_L2_LEN	(VLAN_ETH_HLEN + 3 * MPLS_HLEN)
52 struct ovs_frag_data {
53 	unsigned long dst;
54 	struct vport *vport;
55 	struct ovs_skb_cb cb;
56 	__be16 inner_protocol;
57 	u16 network_offset;	/* valid only for MPLS */
58 	u16 vlan_tci;
59 	__be16 vlan_proto;
60 	unsigned int l2_len;
61 	u8 mac_proto;
62 	u8 l2_data[MAX_L2_LEN];
63 };
64 
65 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
66 
67 #define DEFERRED_ACTION_FIFO_SIZE 10
68 #define OVS_RECURSION_LIMIT 5
69 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
70 struct action_fifo {
71 	int head;
72 	int tail;
73 	/* Deferred action fifo queue storage. */
74 	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
75 };
76 
77 struct action_flow_keys {
78 	struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
79 };
80 
81 static struct action_fifo __percpu *action_fifos;
82 static struct action_flow_keys __percpu *flow_keys;
83 static DEFINE_PER_CPU(int, exec_actions_level);
84 
85 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
86  * space. Return NULL if out of key spaces.
87  */
88 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
89 {
90 	struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
91 	int level = this_cpu_read(exec_actions_level);
92 	struct sw_flow_key *key = NULL;
93 
94 	if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
95 		key = &keys->key[level - 1];
96 		*key = *key_;
97 	}
98 
99 	return key;
100 }
101 
102 static void action_fifo_init(struct action_fifo *fifo)
103 {
104 	fifo->head = 0;
105 	fifo->tail = 0;
106 }
107 
108 static bool action_fifo_is_empty(const struct action_fifo *fifo)
109 {
110 	return (fifo->head == fifo->tail);
111 }
112 
113 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
114 {
115 	if (action_fifo_is_empty(fifo))
116 		return NULL;
117 
118 	return &fifo->fifo[fifo->tail++];
119 }
120 
121 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
122 {
123 	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
124 		return NULL;
125 
126 	return &fifo->fifo[fifo->head++];
127 }
128 
129 /* Return true if fifo is not full */
130 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
131 				    const struct sw_flow_key *key,
132 				    const struct nlattr *actions,
133 				    const int actions_len)
134 {
135 	struct action_fifo *fifo;
136 	struct deferred_action *da;
137 
138 	fifo = this_cpu_ptr(action_fifos);
139 	da = action_fifo_put(fifo);
140 	if (da) {
141 		da->skb = skb;
142 		da->actions = actions;
143 		da->actions_len = actions_len;
144 		da->pkt_key = *key;
145 	}
146 
147 	return da;
148 }
149 
150 static void invalidate_flow_key(struct sw_flow_key *key)
151 {
152 	key->mac_proto |= SW_FLOW_KEY_INVALID;
153 }
154 
155 static bool is_flow_key_valid(const struct sw_flow_key *key)
156 {
157 	return !(key->mac_proto & SW_FLOW_KEY_INVALID);
158 }
159 
160 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
161 			 struct sw_flow_key *key,
162 			 u32 recirc_id,
163 			 const struct nlattr *actions, int len,
164 			 bool last, bool clone_flow_key);
165 
166 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
167 			      struct sw_flow_key *key,
168 			      const struct nlattr *attr, int len);
169 
170 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
171 		     __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
172 {
173 	int err;
174 
175 	err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
176 	if (err)
177 		return err;
178 
179 	if (!mac_len)
180 		key->mac_proto = MAC_PROTO_NONE;
181 
182 	invalidate_flow_key(key);
183 	return 0;
184 }
185 
186 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
187 		    const __be16 ethertype)
188 {
189 	int err;
190 
191 	err = skb_mpls_pop(skb, ethertype, skb->mac_len,
192 			   ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
193 	if (err)
194 		return err;
195 
196 	if (ethertype == htons(ETH_P_TEB))
197 		key->mac_proto = MAC_PROTO_ETHERNET;
198 
199 	invalidate_flow_key(key);
200 	return 0;
201 }
202 
203 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
204 		    const __be32 *mpls_lse, const __be32 *mask)
205 {
206 	struct mpls_shim_hdr *stack;
207 	__be32 lse;
208 	int err;
209 
210 	if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
211 		return -ENOMEM;
212 
213 	stack = mpls_hdr(skb);
214 	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
215 	err = skb_mpls_update_lse(skb, lse);
216 	if (err)
217 		return err;
218 
219 	flow_key->mpls.lse[0] = lse;
220 	return 0;
221 }
222 
223 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
224 {
225 	int err;
226 
227 	err = skb_vlan_pop(skb);
228 	if (skb_vlan_tag_present(skb)) {
229 		invalidate_flow_key(key);
230 	} else {
231 		key->eth.vlan.tci = 0;
232 		key->eth.vlan.tpid = 0;
233 	}
234 	return err;
235 }
236 
237 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
238 		     const struct ovs_action_push_vlan *vlan)
239 {
240 	int err;
241 
242 	if (skb_vlan_tag_present(skb)) {
243 		invalidate_flow_key(key);
244 	} else {
245 		key->eth.vlan.tci = vlan->vlan_tci;
246 		key->eth.vlan.tpid = vlan->vlan_tpid;
247 	}
248 	err = skb_vlan_push(skb, vlan->vlan_tpid,
249 			    ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
250 	skb_reset_mac_len(skb);
251 	return err;
252 }
253 
254 /* 'src' is already properly masked. */
255 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
256 {
257 	u16 *dst = (u16 *)dst_;
258 	const u16 *src = (const u16 *)src_;
259 	const u16 *mask = (const u16 *)mask_;
260 
261 	OVS_SET_MASKED(dst[0], src[0], mask[0]);
262 	OVS_SET_MASKED(dst[1], src[1], mask[1]);
263 	OVS_SET_MASKED(dst[2], src[2], mask[2]);
264 }
265 
266 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
267 			const struct ovs_key_ethernet *key,
268 			const struct ovs_key_ethernet *mask)
269 {
270 	int err;
271 
272 	err = skb_ensure_writable(skb, ETH_HLEN);
273 	if (unlikely(err))
274 		return err;
275 
276 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
277 
278 	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
279 			       mask->eth_src);
280 	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
281 			       mask->eth_dst);
282 
283 	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
284 
285 	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
286 	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
287 	return 0;
288 }
289 
290 /* pop_eth does not support VLAN packets as this action is never called
291  * for them.
292  */
293 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
294 {
295 	int err;
296 
297 	err = skb_eth_pop(skb);
298 	if (err)
299 		return err;
300 
301 	/* safe right before invalidate_flow_key */
302 	key->mac_proto = MAC_PROTO_NONE;
303 	invalidate_flow_key(key);
304 	return 0;
305 }
306 
307 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
308 		    const struct ovs_action_push_eth *ethh)
309 {
310 	int err;
311 
312 	err = skb_eth_push(skb, ethh->addresses.eth_dst,
313 			   ethh->addresses.eth_src);
314 	if (err)
315 		return err;
316 
317 	/* safe right before invalidate_flow_key */
318 	key->mac_proto = MAC_PROTO_ETHERNET;
319 	invalidate_flow_key(key);
320 	return 0;
321 }
322 
323 static noinline_for_stack int push_nsh(struct sk_buff *skb,
324 				       struct sw_flow_key *key,
325 				       const struct nlattr *a)
326 {
327 	u8 buffer[NSH_HDR_MAX_LEN];
328 	struct nshhdr *nh = (struct nshhdr *)buffer;
329 	int err;
330 
331 	err = nsh_hdr_from_nlattr(a, nh, NSH_HDR_MAX_LEN);
332 	if (err)
333 		return err;
334 
335 	err = nsh_push(skb, nh);
336 	if (err)
337 		return err;
338 
339 	/* safe right before invalidate_flow_key */
340 	key->mac_proto = MAC_PROTO_NONE;
341 	invalidate_flow_key(key);
342 	return 0;
343 }
344 
345 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
346 {
347 	int err;
348 
349 	err = nsh_pop(skb);
350 	if (err)
351 		return err;
352 
353 	/* safe right before invalidate_flow_key */
354 	if (skb->protocol == htons(ETH_P_TEB))
355 		key->mac_proto = MAC_PROTO_ETHERNET;
356 	else
357 		key->mac_proto = MAC_PROTO_NONE;
358 	invalidate_flow_key(key);
359 	return 0;
360 }
361 
362 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
363 				  __be32 addr, __be32 new_addr)
364 {
365 	int transport_len = skb->len - skb_transport_offset(skb);
366 
367 	if (nh->frag_off & htons(IP_OFFSET))
368 		return;
369 
370 	if (nh->protocol == IPPROTO_TCP) {
371 		if (likely(transport_len >= sizeof(struct tcphdr)))
372 			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
373 						 addr, new_addr, true);
374 	} else if (nh->protocol == IPPROTO_UDP) {
375 		if (likely(transport_len >= sizeof(struct udphdr))) {
376 			struct udphdr *uh = udp_hdr(skb);
377 
378 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
379 				inet_proto_csum_replace4(&uh->check, skb,
380 							 addr, new_addr, true);
381 				if (!uh->check)
382 					uh->check = CSUM_MANGLED_0;
383 			}
384 		}
385 	}
386 }
387 
388 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
389 			__be32 *addr, __be32 new_addr)
390 {
391 	update_ip_l4_checksum(skb, nh, *addr, new_addr);
392 	csum_replace4(&nh->check, *addr, new_addr);
393 	skb_clear_hash(skb);
394 	ovs_ct_clear(skb, NULL);
395 	*addr = new_addr;
396 }
397 
398 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
399 				 __be32 addr[4], const __be32 new_addr[4])
400 {
401 	int transport_len = skb->len - skb_transport_offset(skb);
402 
403 	if (l4_proto == NEXTHDR_TCP) {
404 		if (likely(transport_len >= sizeof(struct tcphdr)))
405 			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
406 						  addr, new_addr, true);
407 	} else if (l4_proto == NEXTHDR_UDP) {
408 		if (likely(transport_len >= sizeof(struct udphdr))) {
409 			struct udphdr *uh = udp_hdr(skb);
410 
411 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
412 				inet_proto_csum_replace16(&uh->check, skb,
413 							  addr, new_addr, true);
414 				if (!uh->check)
415 					uh->check = CSUM_MANGLED_0;
416 			}
417 		}
418 	} else if (l4_proto == NEXTHDR_ICMP) {
419 		if (likely(transport_len >= sizeof(struct icmp6hdr)))
420 			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
421 						  skb, addr, new_addr, true);
422 	}
423 }
424 
425 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
426 			   const __be32 mask[4], __be32 masked[4])
427 {
428 	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
429 	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
430 	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
431 	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
432 }
433 
434 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
435 			  __be32 addr[4], const __be32 new_addr[4],
436 			  bool recalculate_csum)
437 {
438 	if (recalculate_csum)
439 		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
440 
441 	skb_clear_hash(skb);
442 	ovs_ct_clear(skb, NULL);
443 	memcpy(addr, new_addr, sizeof(__be32[4]));
444 }
445 
446 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
447 {
448 	u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
449 
450 	ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
451 
452 	if (skb->ip_summed == CHECKSUM_COMPLETE)
453 		csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
454 			     (__force __wsum)(ipv6_tclass << 12));
455 
456 	ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
457 }
458 
459 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
460 {
461 	u32 ofl;
462 
463 	ofl = nh->flow_lbl[0] << 16 |  nh->flow_lbl[1] << 8 |  nh->flow_lbl[2];
464 	fl = OVS_MASKED(ofl, fl, mask);
465 
466 	/* Bits 21-24 are always unmasked, so this retains their values. */
467 	nh->flow_lbl[0] = (u8)(fl >> 16);
468 	nh->flow_lbl[1] = (u8)(fl >> 8);
469 	nh->flow_lbl[2] = (u8)fl;
470 
471 	if (skb->ip_summed == CHECKSUM_COMPLETE)
472 		csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
473 }
474 
475 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
476 {
477 	new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
478 
479 	if (skb->ip_summed == CHECKSUM_COMPLETE)
480 		csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
481 			     (__force __wsum)(new_ttl << 8));
482 	nh->hop_limit = new_ttl;
483 }
484 
485 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
486 		       u8 mask)
487 {
488 	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
489 
490 	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
491 	nh->ttl = new_ttl;
492 }
493 
494 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
495 		    const struct ovs_key_ipv4 *key,
496 		    const struct ovs_key_ipv4 *mask)
497 {
498 	struct iphdr *nh;
499 	__be32 new_addr;
500 	int err;
501 
502 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
503 				  sizeof(struct iphdr));
504 	if (unlikely(err))
505 		return err;
506 
507 	nh = ip_hdr(skb);
508 
509 	/* Setting an IP addresses is typically only a side effect of
510 	 * matching on them in the current userspace implementation, so it
511 	 * makes sense to check if the value actually changed.
512 	 */
513 	if (mask->ipv4_src) {
514 		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
515 
516 		if (unlikely(new_addr != nh->saddr)) {
517 			set_ip_addr(skb, nh, &nh->saddr, new_addr);
518 			flow_key->ipv4.addr.src = new_addr;
519 		}
520 	}
521 	if (mask->ipv4_dst) {
522 		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
523 
524 		if (unlikely(new_addr != nh->daddr)) {
525 			set_ip_addr(skb, nh, &nh->daddr, new_addr);
526 			flow_key->ipv4.addr.dst = new_addr;
527 		}
528 	}
529 	if (mask->ipv4_tos) {
530 		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
531 		flow_key->ip.tos = nh->tos;
532 	}
533 	if (mask->ipv4_ttl) {
534 		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
535 		flow_key->ip.ttl = nh->ttl;
536 	}
537 
538 	return 0;
539 }
540 
541 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
542 {
543 	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
544 }
545 
546 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
547 		    const struct ovs_key_ipv6 *key,
548 		    const struct ovs_key_ipv6 *mask)
549 {
550 	struct ipv6hdr *nh;
551 	int err;
552 
553 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
554 				  sizeof(struct ipv6hdr));
555 	if (unlikely(err))
556 		return err;
557 
558 	nh = ipv6_hdr(skb);
559 
560 	/* Setting an IP addresses is typically only a side effect of
561 	 * matching on them in the current userspace implementation, so it
562 	 * makes sense to check if the value actually changed.
563 	 */
564 	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
565 		__be32 *saddr = (__be32 *)&nh->saddr;
566 		__be32 masked[4];
567 
568 		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
569 
570 		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
571 			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
572 				      true);
573 			memcpy(&flow_key->ipv6.addr.src, masked,
574 			       sizeof(flow_key->ipv6.addr.src));
575 		}
576 	}
577 	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
578 		unsigned int offset = 0;
579 		int flags = IP6_FH_F_SKIP_RH;
580 		bool recalc_csum = true;
581 		__be32 *daddr = (__be32 *)&nh->daddr;
582 		__be32 masked[4];
583 
584 		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
585 
586 		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
587 			if (ipv6_ext_hdr(nh->nexthdr))
588 				recalc_csum = (ipv6_find_hdr(skb, &offset,
589 							     NEXTHDR_ROUTING,
590 							     NULL, &flags)
591 					       != NEXTHDR_ROUTING);
592 
593 			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
594 				      recalc_csum);
595 			memcpy(&flow_key->ipv6.addr.dst, masked,
596 			       sizeof(flow_key->ipv6.addr.dst));
597 		}
598 	}
599 	if (mask->ipv6_tclass) {
600 		set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
601 		flow_key->ip.tos = ipv6_get_dsfield(nh);
602 	}
603 	if (mask->ipv6_label) {
604 		set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
605 			    ntohl(mask->ipv6_label));
606 		flow_key->ipv6.label =
607 		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
608 	}
609 	if (mask->ipv6_hlimit) {
610 		set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
611 		flow_key->ip.ttl = nh->hop_limit;
612 	}
613 	return 0;
614 }
615 
616 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
617 		   const struct nlattr *a)
618 {
619 	struct nshhdr *nh;
620 	size_t length;
621 	int err;
622 	u8 flags;
623 	u8 ttl;
624 	int i;
625 
626 	struct ovs_key_nsh key;
627 	struct ovs_key_nsh mask;
628 
629 	err = nsh_key_from_nlattr(a, &key, &mask);
630 	if (err)
631 		return err;
632 
633 	/* Make sure the NSH base header is there */
634 	if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
635 		return -ENOMEM;
636 
637 	nh = nsh_hdr(skb);
638 	length = nsh_hdr_len(nh);
639 
640 	/* Make sure the whole NSH header is there */
641 	err = skb_ensure_writable(skb, skb_network_offset(skb) +
642 				       length);
643 	if (unlikely(err))
644 		return err;
645 
646 	nh = nsh_hdr(skb);
647 	skb_postpull_rcsum(skb, nh, length);
648 	flags = nsh_get_flags(nh);
649 	flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
650 	flow_key->nsh.base.flags = flags;
651 	ttl = nsh_get_ttl(nh);
652 	ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
653 	flow_key->nsh.base.ttl = ttl;
654 	nsh_set_flags_and_ttl(nh, flags, ttl);
655 	nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
656 				  mask.base.path_hdr);
657 	flow_key->nsh.base.path_hdr = nh->path_hdr;
658 	switch (nh->mdtype) {
659 	case NSH_M_TYPE1:
660 		for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
661 			nh->md1.context[i] =
662 			    OVS_MASKED(nh->md1.context[i], key.context[i],
663 				       mask.context[i]);
664 		}
665 		memcpy(flow_key->nsh.context, nh->md1.context,
666 		       sizeof(nh->md1.context));
667 		break;
668 	case NSH_M_TYPE2:
669 		memset(flow_key->nsh.context, 0,
670 		       sizeof(flow_key->nsh.context));
671 		break;
672 	default:
673 		return -EINVAL;
674 	}
675 	skb_postpush_rcsum(skb, nh, length);
676 	return 0;
677 }
678 
679 /* Must follow skb_ensure_writable() since that can move the skb data. */
680 static void set_tp_port(struct sk_buff *skb, __be16 *port,
681 			__be16 new_port, __sum16 *check)
682 {
683 	ovs_ct_clear(skb, NULL);
684 	inet_proto_csum_replace2(check, skb, *port, new_port, false);
685 	*port = new_port;
686 }
687 
688 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
689 		   const struct ovs_key_udp *key,
690 		   const struct ovs_key_udp *mask)
691 {
692 	struct udphdr *uh;
693 	__be16 src, dst;
694 	int err;
695 
696 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
697 				  sizeof(struct udphdr));
698 	if (unlikely(err))
699 		return err;
700 
701 	uh = udp_hdr(skb);
702 	/* Either of the masks is non-zero, so do not bother checking them. */
703 	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
704 	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
705 
706 	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
707 		if (likely(src != uh->source)) {
708 			set_tp_port(skb, &uh->source, src, &uh->check);
709 			flow_key->tp.src = src;
710 		}
711 		if (likely(dst != uh->dest)) {
712 			set_tp_port(skb, &uh->dest, dst, &uh->check);
713 			flow_key->tp.dst = dst;
714 		}
715 
716 		if (unlikely(!uh->check))
717 			uh->check = CSUM_MANGLED_0;
718 	} else {
719 		uh->source = src;
720 		uh->dest = dst;
721 		flow_key->tp.src = src;
722 		flow_key->tp.dst = dst;
723 		ovs_ct_clear(skb, NULL);
724 	}
725 
726 	skb_clear_hash(skb);
727 
728 	return 0;
729 }
730 
731 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
732 		   const struct ovs_key_tcp *key,
733 		   const struct ovs_key_tcp *mask)
734 {
735 	struct tcphdr *th;
736 	__be16 src, dst;
737 	int err;
738 
739 	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
740 				  sizeof(struct tcphdr));
741 	if (unlikely(err))
742 		return err;
743 
744 	th = tcp_hdr(skb);
745 	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
746 	if (likely(src != th->source)) {
747 		set_tp_port(skb, &th->source, src, &th->check);
748 		flow_key->tp.src = src;
749 	}
750 	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
751 	if (likely(dst != th->dest)) {
752 		set_tp_port(skb, &th->dest, dst, &th->check);
753 		flow_key->tp.dst = dst;
754 	}
755 	skb_clear_hash(skb);
756 
757 	return 0;
758 }
759 
760 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
761 		    const struct ovs_key_sctp *key,
762 		    const struct ovs_key_sctp *mask)
763 {
764 	unsigned int sctphoff = skb_transport_offset(skb);
765 	struct sctphdr *sh;
766 	__le32 old_correct_csum, new_csum, old_csum;
767 	int err;
768 
769 	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
770 	if (unlikely(err))
771 		return err;
772 
773 	sh = sctp_hdr(skb);
774 	old_csum = sh->checksum;
775 	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
776 
777 	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
778 	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
779 
780 	new_csum = sctp_compute_cksum(skb, sctphoff);
781 
782 	/* Carry any checksum errors through. */
783 	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
784 
785 	skb_clear_hash(skb);
786 	ovs_ct_clear(skb, NULL);
787 
788 	flow_key->tp.src = sh->source;
789 	flow_key->tp.dst = sh->dest;
790 
791 	return 0;
792 }
793 
794 static int ovs_vport_output(struct net *net, struct sock *sk,
795 			    struct sk_buff *skb)
796 {
797 	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
798 	struct vport *vport = data->vport;
799 
800 	if (skb_cow_head(skb, data->l2_len) < 0) {
801 		kfree_skb_reason(skb, SKB_DROP_REASON_NOMEM);
802 		return -ENOMEM;
803 	}
804 
805 	__skb_dst_copy(skb, data->dst);
806 	*OVS_CB(skb) = data->cb;
807 	skb->inner_protocol = data->inner_protocol;
808 	if (data->vlan_tci & VLAN_CFI_MASK)
809 		__vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
810 	else
811 		__vlan_hwaccel_clear_tag(skb);
812 
813 	/* Reconstruct the MAC header.  */
814 	skb_push(skb, data->l2_len);
815 	memcpy(skb->data, &data->l2_data, data->l2_len);
816 	skb_postpush_rcsum(skb, skb->data, data->l2_len);
817 	skb_reset_mac_header(skb);
818 
819 	if (eth_p_mpls(skb->protocol)) {
820 		skb->inner_network_header = skb->network_header;
821 		skb_set_network_header(skb, data->network_offset);
822 		skb_reset_mac_len(skb);
823 	}
824 
825 	ovs_vport_send(vport, skb, data->mac_proto);
826 	return 0;
827 }
828 
829 static unsigned int
830 ovs_dst_get_mtu(const struct dst_entry *dst)
831 {
832 	return dst->dev->mtu;
833 }
834 
835 static struct dst_ops ovs_dst_ops = {
836 	.family = AF_UNSPEC,
837 	.mtu = ovs_dst_get_mtu,
838 };
839 
840 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
841  * ovs_vport_output(), which is called once per fragmented packet.
842  */
843 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
844 			 u16 orig_network_offset, u8 mac_proto)
845 {
846 	unsigned int hlen = skb_network_offset(skb);
847 	struct ovs_frag_data *data;
848 
849 	data = this_cpu_ptr(&ovs_frag_data_storage);
850 	data->dst = skb->_skb_refdst;
851 	data->vport = vport;
852 	data->cb = *OVS_CB(skb);
853 	data->inner_protocol = skb->inner_protocol;
854 	data->network_offset = orig_network_offset;
855 	if (skb_vlan_tag_present(skb))
856 		data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
857 	else
858 		data->vlan_tci = 0;
859 	data->vlan_proto = skb->vlan_proto;
860 	data->mac_proto = mac_proto;
861 	data->l2_len = hlen;
862 	memcpy(&data->l2_data, skb->data, hlen);
863 
864 	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
865 	skb_pull(skb, hlen);
866 }
867 
868 static void ovs_fragment(struct net *net, struct vport *vport,
869 			 struct sk_buff *skb, u16 mru,
870 			 struct sw_flow_key *key)
871 {
872 	enum ovs_drop_reason reason;
873 	u16 orig_network_offset = 0;
874 
875 	if (eth_p_mpls(skb->protocol)) {
876 		orig_network_offset = skb_network_offset(skb);
877 		skb->network_header = skb->inner_network_header;
878 	}
879 
880 	if (skb_network_offset(skb) > MAX_L2_LEN) {
881 		OVS_NLERR(1, "L2 header too long to fragment");
882 		reason = OVS_DROP_FRAG_L2_TOO_LONG;
883 		goto err;
884 	}
885 
886 	if (key->eth.type == htons(ETH_P_IP)) {
887 		struct rtable ovs_rt = { 0 };
888 		unsigned long orig_dst;
889 
890 		prepare_frag(vport, skb, orig_network_offset,
891 			     ovs_key_mac_proto(key));
892 		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL,
893 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
894 		ovs_rt.dst.dev = vport->dev;
895 
896 		orig_dst = skb->_skb_refdst;
897 		skb_dst_set_noref(skb, &ovs_rt.dst);
898 		IPCB(skb)->frag_max_size = mru;
899 
900 		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
901 		refdst_drop(orig_dst);
902 	} else if (key->eth.type == htons(ETH_P_IPV6)) {
903 		unsigned long orig_dst;
904 		struct rt6_info ovs_rt;
905 
906 		prepare_frag(vport, skb, orig_network_offset,
907 			     ovs_key_mac_proto(key));
908 		memset(&ovs_rt, 0, sizeof(ovs_rt));
909 		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL,
910 			 DST_OBSOLETE_NONE, DST_NOCOUNT);
911 		ovs_rt.dst.dev = vport->dev;
912 
913 		orig_dst = skb->_skb_refdst;
914 		skb_dst_set_noref(skb, &ovs_rt.dst);
915 		IP6CB(skb)->frag_max_size = mru;
916 
917 		ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
918 		refdst_drop(orig_dst);
919 	} else {
920 		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
921 			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
922 			  vport->dev->mtu);
923 		reason = OVS_DROP_FRAG_INVALID_PROTO;
924 		goto err;
925 	}
926 
927 	return;
928 err:
929 	ovs_kfree_skb_reason(skb, reason);
930 }
931 
932 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
933 		      struct sw_flow_key *key)
934 {
935 	struct vport *vport = ovs_vport_rcu(dp, out_port);
936 
937 	if (likely(vport &&
938 		   netif_running(vport->dev) &&
939 		   netif_carrier_ok(vport->dev))) {
940 		u16 mru = OVS_CB(skb)->mru;
941 		u32 cutlen = OVS_CB(skb)->cutlen;
942 
943 		if (unlikely(cutlen > 0)) {
944 			if (skb->len - cutlen > ovs_mac_header_len(key))
945 				pskb_trim(skb, skb->len - cutlen);
946 			else
947 				pskb_trim(skb, ovs_mac_header_len(key));
948 		}
949 
950 		/* Need to set the pkt_type to involve the routing layer.  The
951 		 * packet movement through the OVS datapath doesn't generally
952 		 * use routing, but this is needed for tunnel cases.
953 		 */
954 		skb->pkt_type = PACKET_OUTGOING;
955 
956 		if (likely(!mru ||
957 		           (skb->len <= mru + vport->dev->hard_header_len))) {
958 			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
959 		} else if (mru <= vport->dev->mtu) {
960 			struct net *net = read_pnet(&dp->net);
961 
962 			ovs_fragment(net, vport, skb, mru, key);
963 		} else {
964 			kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
965 		}
966 	} else {
967 		kfree_skb_reason(skb, SKB_DROP_REASON_DEV_READY);
968 	}
969 }
970 
971 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
972 			    struct sw_flow_key *key, const struct nlattr *attr,
973 			    const struct nlattr *actions, int actions_len,
974 			    uint32_t cutlen)
975 {
976 	struct dp_upcall_info upcall;
977 	const struct nlattr *a;
978 	int rem;
979 
980 	memset(&upcall, 0, sizeof(upcall));
981 	upcall.cmd = OVS_PACKET_CMD_ACTION;
982 	upcall.mru = OVS_CB(skb)->mru;
983 
984 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
985 	     a = nla_next(a, &rem)) {
986 		switch (nla_type(a)) {
987 		case OVS_USERSPACE_ATTR_USERDATA:
988 			upcall.userdata = a;
989 			break;
990 
991 		case OVS_USERSPACE_ATTR_PID:
992 			if (dp->user_features &
993 			    OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
994 				upcall.portid =
995 				  ovs_dp_get_upcall_portid(dp,
996 							   smp_processor_id());
997 			else
998 				upcall.portid = nla_get_u32(a);
999 			break;
1000 
1001 		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
1002 			/* Get out tunnel info. */
1003 			struct vport *vport;
1004 
1005 			vport = ovs_vport_rcu(dp, nla_get_u32(a));
1006 			if (vport) {
1007 				int err;
1008 
1009 				err = dev_fill_metadata_dst(vport->dev, skb);
1010 				if (!err)
1011 					upcall.egress_tun_info = skb_tunnel_info(skb);
1012 			}
1013 
1014 			break;
1015 		}
1016 
1017 		case OVS_USERSPACE_ATTR_ACTIONS: {
1018 			/* Include actions. */
1019 			upcall.actions = actions;
1020 			upcall.actions_len = actions_len;
1021 			break;
1022 		}
1023 
1024 		} /* End of switch. */
1025 	}
1026 
1027 	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1028 }
1029 
1030 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
1031 				     struct sw_flow_key *key,
1032 				     const struct nlattr *attr)
1033 {
1034 	/* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
1035 	struct nlattr *actions = nla_data(attr);
1036 
1037 	if (nla_len(actions))
1038 		return clone_execute(dp, skb, key, 0, nla_data(actions),
1039 				     nla_len(actions), true, false);
1040 
1041 	ovs_kfree_skb_reason(skb, OVS_DROP_IP_TTL);
1042 	return 0;
1043 }
1044 
1045 /* When 'last' is true, sample() should always consume the 'skb'.
1046  * Otherwise, sample() should keep 'skb' intact regardless what
1047  * actions are executed within sample().
1048  */
1049 static int sample(struct datapath *dp, struct sk_buff *skb,
1050 		  struct sw_flow_key *key, const struct nlattr *attr,
1051 		  bool last)
1052 {
1053 	struct nlattr *actions;
1054 	struct nlattr *sample_arg;
1055 	int rem = nla_len(attr);
1056 	const struct sample_arg *arg;
1057 	u32 init_probability;
1058 	bool clone_flow_key;
1059 	int err;
1060 
1061 	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1062 	sample_arg = nla_data(attr);
1063 	arg = nla_data(sample_arg);
1064 	actions = nla_next(sample_arg, &rem);
1065 	init_probability = OVS_CB(skb)->probability;
1066 
1067 	if ((arg->probability != U32_MAX) &&
1068 	    (!arg->probability || get_random_u32() > arg->probability)) {
1069 		if (last)
1070 			ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1071 		return 0;
1072 	}
1073 
1074 	OVS_CB(skb)->probability = arg->probability;
1075 
1076 	clone_flow_key = !arg->exec;
1077 	err = clone_execute(dp, skb, key, 0, actions, rem, last,
1078 			    clone_flow_key);
1079 
1080 	if (!last)
1081 		OVS_CB(skb)->probability = init_probability;
1082 
1083 	return err;
1084 }
1085 
1086 /* When 'last' is true, clone() should always consume the 'skb'.
1087  * Otherwise, clone() should keep 'skb' intact regardless what
1088  * actions are executed within clone().
1089  */
1090 static int clone(struct datapath *dp, struct sk_buff *skb,
1091 		 struct sw_flow_key *key, const struct nlattr *attr,
1092 		 bool last)
1093 {
1094 	struct nlattr *actions;
1095 	struct nlattr *clone_arg;
1096 	int rem = nla_len(attr);
1097 	bool dont_clone_flow_key;
1098 
1099 	/* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1100 	clone_arg = nla_data(attr);
1101 	dont_clone_flow_key = nla_get_u32(clone_arg);
1102 	actions = nla_next(clone_arg, &rem);
1103 
1104 	return clone_execute(dp, skb, key, 0, actions, rem, last,
1105 			     !dont_clone_flow_key);
1106 }
1107 
1108 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1109 			 const struct nlattr *attr)
1110 {
1111 	struct ovs_action_hash *hash_act = nla_data(attr);
1112 	u32 hash = 0;
1113 
1114 	if (hash_act->hash_alg == OVS_HASH_ALG_L4) {
1115 		/* OVS_HASH_ALG_L4 hasing type. */
1116 		hash = skb_get_hash(skb);
1117 	} else if (hash_act->hash_alg == OVS_HASH_ALG_SYM_L4) {
1118 		/* OVS_HASH_ALG_SYM_L4 hashing type.  NOTE: this doesn't
1119 		 * extend past an encapsulated header.
1120 		 */
1121 		hash = __skb_get_hash_symmetric(skb);
1122 	}
1123 
1124 	hash = jhash_1word(hash, hash_act->hash_basis);
1125 	if (!hash)
1126 		hash = 0x1;
1127 
1128 	key->ovs_flow_hash = hash;
1129 }
1130 
1131 static int execute_set_action(struct sk_buff *skb,
1132 			      struct sw_flow_key *flow_key,
1133 			      const struct nlattr *a)
1134 {
1135 	/* Only tunnel set execution is supported without a mask. */
1136 	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1137 		struct ovs_tunnel_info *tun = nla_data(a);
1138 
1139 		skb_dst_drop(skb);
1140 		dst_hold((struct dst_entry *)tun->tun_dst);
1141 		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1142 		return 0;
1143 	}
1144 
1145 	return -EINVAL;
1146 }
1147 
1148 /* Mask is at the midpoint of the data. */
1149 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1150 
1151 static int execute_masked_set_action(struct sk_buff *skb,
1152 				     struct sw_flow_key *flow_key,
1153 				     const struct nlattr *a)
1154 {
1155 	int err = 0;
1156 
1157 	switch (nla_type(a)) {
1158 	case OVS_KEY_ATTR_PRIORITY:
1159 		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1160 			       *get_mask(a, u32 *));
1161 		flow_key->phy.priority = skb->priority;
1162 		break;
1163 
1164 	case OVS_KEY_ATTR_SKB_MARK:
1165 		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1166 		flow_key->phy.skb_mark = skb->mark;
1167 		break;
1168 
1169 	case OVS_KEY_ATTR_TUNNEL_INFO:
1170 		/* Masked data not supported for tunnel. */
1171 		err = -EINVAL;
1172 		break;
1173 
1174 	case OVS_KEY_ATTR_ETHERNET:
1175 		err = set_eth_addr(skb, flow_key, nla_data(a),
1176 				   get_mask(a, struct ovs_key_ethernet *));
1177 		break;
1178 
1179 	case OVS_KEY_ATTR_NSH:
1180 		err = set_nsh(skb, flow_key, a);
1181 		break;
1182 
1183 	case OVS_KEY_ATTR_IPV4:
1184 		err = set_ipv4(skb, flow_key, nla_data(a),
1185 			       get_mask(a, struct ovs_key_ipv4 *));
1186 		break;
1187 
1188 	case OVS_KEY_ATTR_IPV6:
1189 		err = set_ipv6(skb, flow_key, nla_data(a),
1190 			       get_mask(a, struct ovs_key_ipv6 *));
1191 		break;
1192 
1193 	case OVS_KEY_ATTR_TCP:
1194 		err = set_tcp(skb, flow_key, nla_data(a),
1195 			      get_mask(a, struct ovs_key_tcp *));
1196 		break;
1197 
1198 	case OVS_KEY_ATTR_UDP:
1199 		err = set_udp(skb, flow_key, nla_data(a),
1200 			      get_mask(a, struct ovs_key_udp *));
1201 		break;
1202 
1203 	case OVS_KEY_ATTR_SCTP:
1204 		err = set_sctp(skb, flow_key, nla_data(a),
1205 			       get_mask(a, struct ovs_key_sctp *));
1206 		break;
1207 
1208 	case OVS_KEY_ATTR_MPLS:
1209 		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1210 								    __be32 *));
1211 		break;
1212 
1213 	case OVS_KEY_ATTR_CT_STATE:
1214 	case OVS_KEY_ATTR_CT_ZONE:
1215 	case OVS_KEY_ATTR_CT_MARK:
1216 	case OVS_KEY_ATTR_CT_LABELS:
1217 	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1218 	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1219 		err = -EINVAL;
1220 		break;
1221 	}
1222 
1223 	return err;
1224 }
1225 
1226 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1227 			  struct sw_flow_key *key,
1228 			  const struct nlattr *a, bool last)
1229 {
1230 	u32 recirc_id;
1231 
1232 	if (!is_flow_key_valid(key)) {
1233 		int err;
1234 
1235 		err = ovs_flow_key_update(skb, key);
1236 		if (err)
1237 			return err;
1238 	}
1239 	BUG_ON(!is_flow_key_valid(key));
1240 
1241 	recirc_id = nla_get_u32(a);
1242 	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1243 }
1244 
1245 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1246 				 struct sw_flow_key *key,
1247 				 const struct nlattr *attr, bool last)
1248 {
1249 	struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1250 	const struct nlattr *actions, *cpl_arg;
1251 	int len, max_len, rem = nla_len(attr);
1252 	const struct check_pkt_len_arg *arg;
1253 	bool clone_flow_key;
1254 
1255 	/* The first netlink attribute in 'attr' is always
1256 	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1257 	 */
1258 	cpl_arg = nla_data(attr);
1259 	arg = nla_data(cpl_arg);
1260 
1261 	len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1262 	max_len = arg->pkt_len;
1263 
1264 	if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1265 	    len <= max_len) {
1266 		/* Second netlink attribute in 'attr' is always
1267 		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1268 		 */
1269 		actions = nla_next(cpl_arg, &rem);
1270 		clone_flow_key = !arg->exec_for_lesser_equal;
1271 	} else {
1272 		/* Third netlink attribute in 'attr' is always
1273 		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1274 		 */
1275 		actions = nla_next(cpl_arg, &rem);
1276 		actions = nla_next(actions, &rem);
1277 		clone_flow_key = !arg->exec_for_greater;
1278 	}
1279 
1280 	return clone_execute(dp, skb, key, 0, nla_data(actions),
1281 			     nla_len(actions), last, clone_flow_key);
1282 }
1283 
1284 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1285 {
1286 	int err;
1287 
1288 	if (skb->protocol == htons(ETH_P_IPV6)) {
1289 		struct ipv6hdr *nh;
1290 
1291 		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1292 					  sizeof(*nh));
1293 		if (unlikely(err))
1294 			return err;
1295 
1296 		nh = ipv6_hdr(skb);
1297 
1298 		if (nh->hop_limit <= 1)
1299 			return -EHOSTUNREACH;
1300 
1301 		key->ip.ttl = --nh->hop_limit;
1302 	} else if (skb->protocol == htons(ETH_P_IP)) {
1303 		struct iphdr *nh;
1304 		u8 old_ttl;
1305 
1306 		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1307 					  sizeof(*nh));
1308 		if (unlikely(err))
1309 			return err;
1310 
1311 		nh = ip_hdr(skb);
1312 		if (nh->ttl <= 1)
1313 			return -EHOSTUNREACH;
1314 
1315 		old_ttl = nh->ttl--;
1316 		csum_replace2(&nh->check, htons(old_ttl << 8),
1317 			      htons(nh->ttl << 8));
1318 		key->ip.ttl = nh->ttl;
1319 	}
1320 	return 0;
1321 }
1322 
1323 #if IS_ENABLED(CONFIG_PSAMPLE)
1324 static void execute_psample(struct datapath *dp, struct sk_buff *skb,
1325 			    const struct nlattr *attr)
1326 {
1327 	struct psample_group psample_group = {};
1328 	struct psample_metadata md = {};
1329 	const struct nlattr *a;
1330 	u32 rate;
1331 	int rem;
1332 
1333 	nla_for_each_attr(a, nla_data(attr), nla_len(attr), rem) {
1334 		switch (nla_type(a)) {
1335 		case OVS_PSAMPLE_ATTR_GROUP:
1336 			psample_group.group_num = nla_get_u32(a);
1337 			break;
1338 
1339 		case OVS_PSAMPLE_ATTR_COOKIE:
1340 			md.user_cookie = nla_data(a);
1341 			md.user_cookie_len = nla_len(a);
1342 			break;
1343 		}
1344 	}
1345 
1346 	psample_group.net = ovs_dp_get_net(dp);
1347 	md.in_ifindex = OVS_CB(skb)->input_vport->dev->ifindex;
1348 	md.trunc_size = skb->len - OVS_CB(skb)->cutlen;
1349 	md.rate_as_probability = 1;
1350 
1351 	rate = OVS_CB(skb)->probability ? OVS_CB(skb)->probability : U32_MAX;
1352 
1353 	psample_sample_packet(&psample_group, skb, rate, &md);
1354 }
1355 #else
1356 static void execute_psample(struct datapath *dp, struct sk_buff *skb,
1357 			    const struct nlattr *attr)
1358 {}
1359 #endif
1360 
1361 /* Execute a list of actions against 'skb'. */
1362 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1363 			      struct sw_flow_key *key,
1364 			      const struct nlattr *attr, int len)
1365 {
1366 	const struct nlattr *a;
1367 	int rem;
1368 
1369 	for (a = attr, rem = len; rem > 0;
1370 	     a = nla_next(a, &rem)) {
1371 		int err = 0;
1372 
1373 		if (trace_ovs_do_execute_action_enabled())
1374 			trace_ovs_do_execute_action(dp, skb, key, a, rem);
1375 
1376 		/* Actions that rightfully have to consume the skb should do it
1377 		 * and return directly.
1378 		 */
1379 		switch (nla_type(a)) {
1380 		case OVS_ACTION_ATTR_OUTPUT: {
1381 			int port = nla_get_u32(a);
1382 			struct sk_buff *clone;
1383 
1384 			/* Every output action needs a separate clone
1385 			 * of 'skb', In case the output action is the
1386 			 * last action, cloning can be avoided.
1387 			 */
1388 			if (nla_is_last(a, rem)) {
1389 				do_output(dp, skb, port, key);
1390 				/* 'skb' has been used for output.
1391 				 */
1392 				return 0;
1393 			}
1394 
1395 			clone = skb_clone(skb, GFP_ATOMIC);
1396 			if (clone)
1397 				do_output(dp, clone, port, key);
1398 			OVS_CB(skb)->cutlen = 0;
1399 			break;
1400 		}
1401 
1402 		case OVS_ACTION_ATTR_TRUNC: {
1403 			struct ovs_action_trunc *trunc = nla_data(a);
1404 
1405 			if (skb->len > trunc->max_len)
1406 				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1407 			break;
1408 		}
1409 
1410 		case OVS_ACTION_ATTR_USERSPACE:
1411 			output_userspace(dp, skb, key, a, attr,
1412 						     len, OVS_CB(skb)->cutlen);
1413 			OVS_CB(skb)->cutlen = 0;
1414 			if (nla_is_last(a, rem)) {
1415 				consume_skb(skb);
1416 				return 0;
1417 			}
1418 			break;
1419 
1420 		case OVS_ACTION_ATTR_HASH:
1421 			execute_hash(skb, key, a);
1422 			break;
1423 
1424 		case OVS_ACTION_ATTR_PUSH_MPLS: {
1425 			struct ovs_action_push_mpls *mpls = nla_data(a);
1426 
1427 			err = push_mpls(skb, key, mpls->mpls_lse,
1428 					mpls->mpls_ethertype, skb->mac_len);
1429 			break;
1430 		}
1431 		case OVS_ACTION_ATTR_ADD_MPLS: {
1432 			struct ovs_action_add_mpls *mpls = nla_data(a);
1433 			__u16 mac_len = 0;
1434 
1435 			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1436 				mac_len = skb->mac_len;
1437 
1438 			err = push_mpls(skb, key, mpls->mpls_lse,
1439 					mpls->mpls_ethertype, mac_len);
1440 			break;
1441 		}
1442 		case OVS_ACTION_ATTR_POP_MPLS:
1443 			err = pop_mpls(skb, key, nla_get_be16(a));
1444 			break;
1445 
1446 		case OVS_ACTION_ATTR_PUSH_VLAN:
1447 			err = push_vlan(skb, key, nla_data(a));
1448 			break;
1449 
1450 		case OVS_ACTION_ATTR_POP_VLAN:
1451 			err = pop_vlan(skb, key);
1452 			break;
1453 
1454 		case OVS_ACTION_ATTR_RECIRC: {
1455 			bool last = nla_is_last(a, rem);
1456 
1457 			err = execute_recirc(dp, skb, key, a, last);
1458 			if (last) {
1459 				/* If this is the last action, the skb has
1460 				 * been consumed or freed.
1461 				 * Return immediately.
1462 				 */
1463 				return err;
1464 			}
1465 			break;
1466 		}
1467 
1468 		case OVS_ACTION_ATTR_SET:
1469 			err = execute_set_action(skb, key, nla_data(a));
1470 			break;
1471 
1472 		case OVS_ACTION_ATTR_SET_MASKED:
1473 		case OVS_ACTION_ATTR_SET_TO_MASKED:
1474 			err = execute_masked_set_action(skb, key, nla_data(a));
1475 			break;
1476 
1477 		case OVS_ACTION_ATTR_SAMPLE: {
1478 			bool last = nla_is_last(a, rem);
1479 
1480 			err = sample(dp, skb, key, a, last);
1481 			if (last)
1482 				return err;
1483 
1484 			break;
1485 		}
1486 
1487 		case OVS_ACTION_ATTR_CT:
1488 			if (!is_flow_key_valid(key)) {
1489 				err = ovs_flow_key_update(skb, key);
1490 				if (err)
1491 					return err;
1492 			}
1493 
1494 			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1495 					     nla_data(a));
1496 
1497 			/* Hide stolen IP fragments from user space. */
1498 			if (err)
1499 				return err == -EINPROGRESS ? 0 : err;
1500 			break;
1501 
1502 		case OVS_ACTION_ATTR_CT_CLEAR:
1503 			err = ovs_ct_clear(skb, key);
1504 			break;
1505 
1506 		case OVS_ACTION_ATTR_PUSH_ETH:
1507 			err = push_eth(skb, key, nla_data(a));
1508 			break;
1509 
1510 		case OVS_ACTION_ATTR_POP_ETH:
1511 			err = pop_eth(skb, key);
1512 			break;
1513 
1514 		case OVS_ACTION_ATTR_PUSH_NSH:
1515 			err = push_nsh(skb, key, nla_data(a));
1516 			break;
1517 
1518 		case OVS_ACTION_ATTR_POP_NSH:
1519 			err = pop_nsh(skb, key);
1520 			break;
1521 
1522 		case OVS_ACTION_ATTR_METER:
1523 			if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1524 				ovs_kfree_skb_reason(skb, OVS_DROP_METER);
1525 				return 0;
1526 			}
1527 			break;
1528 
1529 		case OVS_ACTION_ATTR_CLONE: {
1530 			bool last = nla_is_last(a, rem);
1531 
1532 			err = clone(dp, skb, key, a, last);
1533 			if (last)
1534 				return err;
1535 
1536 			break;
1537 		}
1538 
1539 		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1540 			bool last = nla_is_last(a, rem);
1541 
1542 			err = execute_check_pkt_len(dp, skb, key, a, last);
1543 			if (last)
1544 				return err;
1545 
1546 			break;
1547 		}
1548 
1549 		case OVS_ACTION_ATTR_DEC_TTL:
1550 			err = execute_dec_ttl(skb, key);
1551 			if (err == -EHOSTUNREACH)
1552 				return dec_ttl_exception_handler(dp, skb,
1553 								 key, a);
1554 			break;
1555 
1556 		case OVS_ACTION_ATTR_DROP: {
1557 			enum ovs_drop_reason reason = nla_get_u32(a)
1558 				? OVS_DROP_EXPLICIT_WITH_ERROR
1559 				: OVS_DROP_EXPLICIT;
1560 
1561 			ovs_kfree_skb_reason(skb, reason);
1562 			return 0;
1563 		}
1564 
1565 		case OVS_ACTION_ATTR_PSAMPLE:
1566 			execute_psample(dp, skb, a);
1567 			OVS_CB(skb)->cutlen = 0;
1568 			if (nla_is_last(a, rem)) {
1569 				consume_skb(skb);
1570 				return 0;
1571 			}
1572 			break;
1573 		}
1574 
1575 		if (unlikely(err)) {
1576 			ovs_kfree_skb_reason(skb, OVS_DROP_ACTION_ERROR);
1577 			return err;
1578 		}
1579 	}
1580 
1581 	ovs_kfree_skb_reason(skb, OVS_DROP_LAST_ACTION);
1582 	return 0;
1583 }
1584 
1585 /* Execute the actions on the clone of the packet. The effect of the
1586  * execution does not affect the original 'skb' nor the original 'key'.
1587  *
1588  * The execution may be deferred in case the actions can not be executed
1589  * immediately.
1590  */
1591 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1592 			 struct sw_flow_key *key, u32 recirc_id,
1593 			 const struct nlattr *actions, int len,
1594 			 bool last, bool clone_flow_key)
1595 {
1596 	struct deferred_action *da;
1597 	struct sw_flow_key *clone;
1598 
1599 	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1600 	if (!skb) {
1601 		/* Out of memory, skip this action.
1602 		 */
1603 		return 0;
1604 	}
1605 
1606 	/* When clone_flow_key is false, the 'key' will not be change
1607 	 * by the actions, then the 'key' can be used directly.
1608 	 * Otherwise, try to clone key from the next recursion level of
1609 	 * 'flow_keys'. If clone is successful, execute the actions
1610 	 * without deferring.
1611 	 */
1612 	clone = clone_flow_key ? clone_key(key) : key;
1613 	if (clone) {
1614 		int err = 0;
1615 
1616 		if (actions) { /* Sample action */
1617 			if (clone_flow_key)
1618 				__this_cpu_inc(exec_actions_level);
1619 
1620 			err = do_execute_actions(dp, skb, clone,
1621 						 actions, len);
1622 
1623 			if (clone_flow_key)
1624 				__this_cpu_dec(exec_actions_level);
1625 		} else { /* Recirc action */
1626 			clone->recirc_id = recirc_id;
1627 			ovs_dp_process_packet(skb, clone);
1628 		}
1629 		return err;
1630 	}
1631 
1632 	/* Out of 'flow_keys' space. Defer actions */
1633 	da = add_deferred_actions(skb, key, actions, len);
1634 	if (da) {
1635 		if (!actions) { /* Recirc action */
1636 			key = &da->pkt_key;
1637 			key->recirc_id = recirc_id;
1638 		}
1639 	} else {
1640 		/* Out of per CPU action FIFO space. Drop the 'skb' and
1641 		 * log an error.
1642 		 */
1643 		ovs_kfree_skb_reason(skb, OVS_DROP_DEFERRED_LIMIT);
1644 
1645 		if (net_ratelimit()) {
1646 			if (actions) { /* Sample action */
1647 				pr_warn("%s: deferred action limit reached, drop sample action\n",
1648 					ovs_dp_name(dp));
1649 			} else {  /* Recirc action */
1650 				pr_warn("%s: deferred action limit reached, drop recirc action (recirc_id=%#x)\n",
1651 					ovs_dp_name(dp), recirc_id);
1652 			}
1653 		}
1654 	}
1655 	return 0;
1656 }
1657 
1658 static void process_deferred_actions(struct datapath *dp)
1659 {
1660 	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1661 
1662 	/* Do not touch the FIFO in case there is no deferred actions. */
1663 	if (action_fifo_is_empty(fifo))
1664 		return;
1665 
1666 	/* Finishing executing all deferred actions. */
1667 	do {
1668 		struct deferred_action *da = action_fifo_get(fifo);
1669 		struct sk_buff *skb = da->skb;
1670 		struct sw_flow_key *key = &da->pkt_key;
1671 		const struct nlattr *actions = da->actions;
1672 		int actions_len = da->actions_len;
1673 
1674 		if (actions)
1675 			do_execute_actions(dp, skb, key, actions, actions_len);
1676 		else
1677 			ovs_dp_process_packet(skb, key);
1678 	} while (!action_fifo_is_empty(fifo));
1679 
1680 	/* Reset FIFO for the next packet.  */
1681 	action_fifo_init(fifo);
1682 }
1683 
1684 /* Execute a list of actions against 'skb'. */
1685 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1686 			const struct sw_flow_actions *acts,
1687 			struct sw_flow_key *key)
1688 {
1689 	int err, level;
1690 
1691 	level = __this_cpu_inc_return(exec_actions_level);
1692 	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1693 		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1694 				     ovs_dp_name(dp));
1695 		ovs_kfree_skb_reason(skb, OVS_DROP_RECURSION_LIMIT);
1696 		err = -ENETDOWN;
1697 		goto out;
1698 	}
1699 
1700 	OVS_CB(skb)->acts_origlen = acts->orig_len;
1701 	err = do_execute_actions(dp, skb, key,
1702 				 acts->actions, acts->actions_len);
1703 
1704 	if (level == 1)
1705 		process_deferred_actions(dp);
1706 
1707 out:
1708 	__this_cpu_dec(exec_actions_level);
1709 	return err;
1710 }
1711 
1712 int action_fifos_init(void)
1713 {
1714 	action_fifos = alloc_percpu(struct action_fifo);
1715 	if (!action_fifos)
1716 		return -ENOMEM;
1717 
1718 	flow_keys = alloc_percpu(struct action_flow_keys);
1719 	if (!flow_keys) {
1720 		free_percpu(action_fifos);
1721 		return -ENOMEM;
1722 	}
1723 
1724 	return 0;
1725 }
1726 
1727 void action_fifos_exit(void)
1728 {
1729 	free_percpu(action_fifos);
1730 	free_percpu(flow_keys);
1731 }
1732