1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2007-2017 Nicira, Inc.
4 */
5
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8#include <linux/skbuff.h>
9#include <linux/in.h>
10#include <linux/ip.h>
11#include <linux/openvswitch.h>
12#include <linux/sctp.h>
13#include <linux/tcp.h>
14#include <linux/udp.h>
15#include <linux/in6.h>
16#include <linux/if_arp.h>
17#include <linux/if_vlan.h>
18
19#include <net/dst.h>
20#include <net/ip.h>
21#include <net/ipv6.h>
22#include <net/ip6_fib.h>
23#include <net/checksum.h>
24#include <net/dsfield.h>
25#include <net/mpls.h>
26#include <net/sctp/checksum.h>
27
28#include "datapath.h"
29#include "flow.h"
30#include "conntrack.h"
31#include "vport.h"
32#include "flow_netlink.h"
33
34struct deferred_action {
35	struct sk_buff *skb;
36	const struct nlattr *actions;
37	int actions_len;
38
39	/* Store pkt_key clone when creating deferred action. */
40	struct sw_flow_key pkt_key;
41};
42
43#define MAX_L2_LEN	(VLAN_ETH_HLEN + 3 * MPLS_HLEN)
44struct ovs_frag_data {
45	unsigned long dst;
46	struct vport *vport;
47	struct ovs_skb_cb cb;
48	__be16 inner_protocol;
49	u16 network_offset;	/* valid only for MPLS */
50	u16 vlan_tci;
51	__be16 vlan_proto;
52	unsigned int l2_len;
53	u8 mac_proto;
54	u8 l2_data[MAX_L2_LEN];
55};
56
57static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
58
59#define DEFERRED_ACTION_FIFO_SIZE 10
60#define OVS_RECURSION_LIMIT 5
61#define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
62struct action_fifo {
63	int head;
64	int tail;
65	/* Deferred action fifo queue storage. */
66	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
67};
68
69struct action_flow_keys {
70	struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
71};
72
73static struct action_fifo __percpu *action_fifos;
74static struct action_flow_keys __percpu *flow_keys;
75static DEFINE_PER_CPU(int, exec_actions_level);
76
77/* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
78 * space. Return NULL if out of key spaces.
79 */
80static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
81{
82	struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
83	int level = this_cpu_read(exec_actions_level);
84	struct sw_flow_key *key = NULL;
85
86	if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
87		key = &keys->key[level - 1];
88		*key = *key_;
89	}
90
91	return key;
92}
93
94static void action_fifo_init(struct action_fifo *fifo)
95{
96	fifo->head = 0;
97	fifo->tail = 0;
98}
99
100static bool action_fifo_is_empty(const struct action_fifo *fifo)
101{
102	return (fifo->head == fifo->tail);
103}
104
105static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
106{
107	if (action_fifo_is_empty(fifo))
108		return NULL;
109
110	return &fifo->fifo[fifo->tail++];
111}
112
113static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
114{
115	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
116		return NULL;
117
118	return &fifo->fifo[fifo->head++];
119}
120
121/* Return true if fifo is not full */
122static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
123				    const struct sw_flow_key *key,
124				    const struct nlattr *actions,
125				    const int actions_len)
126{
127	struct action_fifo *fifo;
128	struct deferred_action *da;
129
130	fifo = this_cpu_ptr(action_fifos);
131	da = action_fifo_put(fifo);
132	if (da) {
133		da->skb = skb;
134		da->actions = actions;
135		da->actions_len = actions_len;
136		da->pkt_key = *key;
137	}
138
139	return da;
140}
141
142static void invalidate_flow_key(struct sw_flow_key *key)
143{
144	key->mac_proto |= SW_FLOW_KEY_INVALID;
145}
146
147static bool is_flow_key_valid(const struct sw_flow_key *key)
148{
149	return !(key->mac_proto & SW_FLOW_KEY_INVALID);
150}
151
152static int clone_execute(struct datapath *dp, struct sk_buff *skb,
153			 struct sw_flow_key *key,
154			 u32 recirc_id,
155			 const struct nlattr *actions, int len,
156			 bool last, bool clone_flow_key);
157
158static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
159			      struct sw_flow_key *key,
160			      const struct nlattr *attr, int len);
161
162static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
163		     __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
164{
165	int err;
166
167	err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
168	if (err)
169		return err;
170
171	if (!mac_len)
172		key->mac_proto = MAC_PROTO_NONE;
173
174	invalidate_flow_key(key);
175	return 0;
176}
177
178static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
179		    const __be16 ethertype)
180{
181	int err;
182
183	err = skb_mpls_pop(skb, ethertype, skb->mac_len,
184			   ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
185	if (err)
186		return err;
187
188	if (ethertype == htons(ETH_P_TEB))
189		key->mac_proto = MAC_PROTO_ETHERNET;
190
191	invalidate_flow_key(key);
192	return 0;
193}
194
195static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
196		    const __be32 *mpls_lse, const __be32 *mask)
197{
198	struct mpls_shim_hdr *stack;
199	__be32 lse;
200	int err;
201
202	if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
203		return -ENOMEM;
204
205	stack = mpls_hdr(skb);
206	lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
207	err = skb_mpls_update_lse(skb, lse);
208	if (err)
209		return err;
210
211	flow_key->mpls.lse[0] = lse;
212	return 0;
213}
214
215static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
216{
217	int err;
218
219	err = skb_vlan_pop(skb);
220	if (skb_vlan_tag_present(skb)) {
221		invalidate_flow_key(key);
222	} else {
223		key->eth.vlan.tci = 0;
224		key->eth.vlan.tpid = 0;
225	}
226	return err;
227}
228
229static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
230		     const struct ovs_action_push_vlan *vlan)
231{
232	if (skb_vlan_tag_present(skb)) {
233		invalidate_flow_key(key);
234	} else {
235		key->eth.vlan.tci = vlan->vlan_tci;
236		key->eth.vlan.tpid = vlan->vlan_tpid;
237	}
238	return skb_vlan_push(skb, vlan->vlan_tpid,
239			     ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
240}
241
242/* 'src' is already properly masked. */
243static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
244{
245	u16 *dst = (u16 *)dst_;
246	const u16 *src = (const u16 *)src_;
247	const u16 *mask = (const u16 *)mask_;
248
249	OVS_SET_MASKED(dst[0], src[0], mask[0]);
250	OVS_SET_MASKED(dst[1], src[1], mask[1]);
251	OVS_SET_MASKED(dst[2], src[2], mask[2]);
252}
253
254static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
255			const struct ovs_key_ethernet *key,
256			const struct ovs_key_ethernet *mask)
257{
258	int err;
259
260	err = skb_ensure_writable(skb, ETH_HLEN);
261	if (unlikely(err))
262		return err;
263
264	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
265
266	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
267			       mask->eth_src);
268	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
269			       mask->eth_dst);
270
271	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
272
273	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
274	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
275	return 0;
276}
277
278/* pop_eth does not support VLAN packets as this action is never called
279 * for them.
280 */
281static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
282{
283	int err;
284
285	err = skb_eth_pop(skb);
286	if (err)
287		return err;
288
289	/* safe right before invalidate_flow_key */
290	key->mac_proto = MAC_PROTO_NONE;
291	invalidate_flow_key(key);
292	return 0;
293}
294
295static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
296		    const struct ovs_action_push_eth *ethh)
297{
298	int err;
299
300	err = skb_eth_push(skb, ethh->addresses.eth_dst,
301			   ethh->addresses.eth_src);
302	if (err)
303		return err;
304
305	/* safe right before invalidate_flow_key */
306	key->mac_proto = MAC_PROTO_ETHERNET;
307	invalidate_flow_key(key);
308	return 0;
309}
310
311static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
312		    const struct nshhdr *nh)
313{
314	int err;
315
316	err = nsh_push(skb, nh);
317	if (err)
318		return err;
319
320	/* safe right before invalidate_flow_key */
321	key->mac_proto = MAC_PROTO_NONE;
322	invalidate_flow_key(key);
323	return 0;
324}
325
326static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
327{
328	int err;
329
330	err = nsh_pop(skb);
331	if (err)
332		return err;
333
334	/* safe right before invalidate_flow_key */
335	if (skb->protocol == htons(ETH_P_TEB))
336		key->mac_proto = MAC_PROTO_ETHERNET;
337	else
338		key->mac_proto = MAC_PROTO_NONE;
339	invalidate_flow_key(key);
340	return 0;
341}
342
343static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
344				  __be32 addr, __be32 new_addr)
345{
346	int transport_len = skb->len - skb_transport_offset(skb);
347
348	if (nh->frag_off & htons(IP_OFFSET))
349		return;
350
351	if (nh->protocol == IPPROTO_TCP) {
352		if (likely(transport_len >= sizeof(struct tcphdr)))
353			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
354						 addr, new_addr, true);
355	} else if (nh->protocol == IPPROTO_UDP) {
356		if (likely(transport_len >= sizeof(struct udphdr))) {
357			struct udphdr *uh = udp_hdr(skb);
358
359			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
360				inet_proto_csum_replace4(&uh->check, skb,
361							 addr, new_addr, true);
362				if (!uh->check)
363					uh->check = CSUM_MANGLED_0;
364			}
365		}
366	}
367}
368
369static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
370			__be32 *addr, __be32 new_addr)
371{
372	update_ip_l4_checksum(skb, nh, *addr, new_addr);
373	csum_replace4(&nh->check, *addr, new_addr);
374	skb_clear_hash(skb);
375	ovs_ct_clear(skb, NULL);
376	*addr = new_addr;
377}
378
379static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
380				 __be32 addr[4], const __be32 new_addr[4])
381{
382	int transport_len = skb->len - skb_transport_offset(skb);
383
384	if (l4_proto == NEXTHDR_TCP) {
385		if (likely(transport_len >= sizeof(struct tcphdr)))
386			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
387						  addr, new_addr, true);
388	} else if (l4_proto == NEXTHDR_UDP) {
389		if (likely(transport_len >= sizeof(struct udphdr))) {
390			struct udphdr *uh = udp_hdr(skb);
391
392			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
393				inet_proto_csum_replace16(&uh->check, skb,
394							  addr, new_addr, true);
395				if (!uh->check)
396					uh->check = CSUM_MANGLED_0;
397			}
398		}
399	} else if (l4_proto == NEXTHDR_ICMP) {
400		if (likely(transport_len >= sizeof(struct icmp6hdr)))
401			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
402						  skb, addr, new_addr, true);
403	}
404}
405
406static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
407			   const __be32 mask[4], __be32 masked[4])
408{
409	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
410	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
411	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
412	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
413}
414
415static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
416			  __be32 addr[4], const __be32 new_addr[4],
417			  bool recalculate_csum)
418{
419	if (recalculate_csum)
420		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
421
422	skb_clear_hash(skb);
423	ovs_ct_clear(skb, NULL);
424	memcpy(addr, new_addr, sizeof(__be32[4]));
425}
426
427static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
428{
429	u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
430
431	ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
432
433	if (skb->ip_summed == CHECKSUM_COMPLETE)
434		csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
435			     (__force __wsum)(ipv6_tclass << 12));
436
437	ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
438}
439
440static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
441{
442	u32 ofl;
443
444	ofl = nh->flow_lbl[0] << 16 |  nh->flow_lbl[1] << 8 |  nh->flow_lbl[2];
445	fl = OVS_MASKED(ofl, fl, mask);
446
447	/* Bits 21-24 are always unmasked, so this retains their values. */
448	nh->flow_lbl[0] = (u8)(fl >> 16);
449	nh->flow_lbl[1] = (u8)(fl >> 8);
450	nh->flow_lbl[2] = (u8)fl;
451
452	if (skb->ip_summed == CHECKSUM_COMPLETE)
453		csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
454}
455
456static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
457{
458	new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
459
460	if (skb->ip_summed == CHECKSUM_COMPLETE)
461		csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
462			     (__force __wsum)(new_ttl << 8));
463	nh->hop_limit = new_ttl;
464}
465
466static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
467		       u8 mask)
468{
469	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
470
471	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
472	nh->ttl = new_ttl;
473}
474
475static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
476		    const struct ovs_key_ipv4 *key,
477		    const struct ovs_key_ipv4 *mask)
478{
479	struct iphdr *nh;
480	__be32 new_addr;
481	int err;
482
483	err = skb_ensure_writable(skb, skb_network_offset(skb) +
484				  sizeof(struct iphdr));
485	if (unlikely(err))
486		return err;
487
488	nh = ip_hdr(skb);
489
490	/* Setting an IP addresses is typically only a side effect of
491	 * matching on them in the current userspace implementation, so it
492	 * makes sense to check if the value actually changed.
493	 */
494	if (mask->ipv4_src) {
495		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
496
497		if (unlikely(new_addr != nh->saddr)) {
498			set_ip_addr(skb, nh, &nh->saddr, new_addr);
499			flow_key->ipv4.addr.src = new_addr;
500		}
501	}
502	if (mask->ipv4_dst) {
503		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
504
505		if (unlikely(new_addr != nh->daddr)) {
506			set_ip_addr(skb, nh, &nh->daddr, new_addr);
507			flow_key->ipv4.addr.dst = new_addr;
508		}
509	}
510	if (mask->ipv4_tos) {
511		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
512		flow_key->ip.tos = nh->tos;
513	}
514	if (mask->ipv4_ttl) {
515		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
516		flow_key->ip.ttl = nh->ttl;
517	}
518
519	return 0;
520}
521
522static bool is_ipv6_mask_nonzero(const __be32 addr[4])
523{
524	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
525}
526
527static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
528		    const struct ovs_key_ipv6 *key,
529		    const struct ovs_key_ipv6 *mask)
530{
531	struct ipv6hdr *nh;
532	int err;
533
534	err = skb_ensure_writable(skb, skb_network_offset(skb) +
535				  sizeof(struct ipv6hdr));
536	if (unlikely(err))
537		return err;
538
539	nh = ipv6_hdr(skb);
540
541	/* Setting an IP addresses is typically only a side effect of
542	 * matching on them in the current userspace implementation, so it
543	 * makes sense to check if the value actually changed.
544	 */
545	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
546		__be32 *saddr = (__be32 *)&nh->saddr;
547		__be32 masked[4];
548
549		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
550
551		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
552			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
553				      true);
554			memcpy(&flow_key->ipv6.addr.src, masked,
555			       sizeof(flow_key->ipv6.addr.src));
556		}
557	}
558	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
559		unsigned int offset = 0;
560		int flags = IP6_FH_F_SKIP_RH;
561		bool recalc_csum = true;
562		__be32 *daddr = (__be32 *)&nh->daddr;
563		__be32 masked[4];
564
565		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
566
567		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
568			if (ipv6_ext_hdr(nh->nexthdr))
569				recalc_csum = (ipv6_find_hdr(skb, &offset,
570							     NEXTHDR_ROUTING,
571							     NULL, &flags)
572					       != NEXTHDR_ROUTING);
573
574			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
575				      recalc_csum);
576			memcpy(&flow_key->ipv6.addr.dst, masked,
577			       sizeof(flow_key->ipv6.addr.dst));
578		}
579	}
580	if (mask->ipv6_tclass) {
581		set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
582		flow_key->ip.tos = ipv6_get_dsfield(nh);
583	}
584	if (mask->ipv6_label) {
585		set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
586			    ntohl(mask->ipv6_label));
587		flow_key->ipv6.label =
588		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
589	}
590	if (mask->ipv6_hlimit) {
591		set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
592		flow_key->ip.ttl = nh->hop_limit;
593	}
594	return 0;
595}
596
597static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
598		   const struct nlattr *a)
599{
600	struct nshhdr *nh;
601	size_t length;
602	int err;
603	u8 flags;
604	u8 ttl;
605	int i;
606
607	struct ovs_key_nsh key;
608	struct ovs_key_nsh mask;
609
610	err = nsh_key_from_nlattr(a, &key, &mask);
611	if (err)
612		return err;
613
614	/* Make sure the NSH base header is there */
615	if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
616		return -ENOMEM;
617
618	nh = nsh_hdr(skb);
619	length = nsh_hdr_len(nh);
620
621	/* Make sure the whole NSH header is there */
622	err = skb_ensure_writable(skb, skb_network_offset(skb) +
623				       length);
624	if (unlikely(err))
625		return err;
626
627	nh = nsh_hdr(skb);
628	skb_postpull_rcsum(skb, nh, length);
629	flags = nsh_get_flags(nh);
630	flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
631	flow_key->nsh.base.flags = flags;
632	ttl = nsh_get_ttl(nh);
633	ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
634	flow_key->nsh.base.ttl = ttl;
635	nsh_set_flags_and_ttl(nh, flags, ttl);
636	nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
637				  mask.base.path_hdr);
638	flow_key->nsh.base.path_hdr = nh->path_hdr;
639	switch (nh->mdtype) {
640	case NSH_M_TYPE1:
641		for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
642			nh->md1.context[i] =
643			    OVS_MASKED(nh->md1.context[i], key.context[i],
644				       mask.context[i]);
645		}
646		memcpy(flow_key->nsh.context, nh->md1.context,
647		       sizeof(nh->md1.context));
648		break;
649	case NSH_M_TYPE2:
650		memset(flow_key->nsh.context, 0,
651		       sizeof(flow_key->nsh.context));
652		break;
653	default:
654		return -EINVAL;
655	}
656	skb_postpush_rcsum(skb, nh, length);
657	return 0;
658}
659
660/* Must follow skb_ensure_writable() since that can move the skb data. */
661static void set_tp_port(struct sk_buff *skb, __be16 *port,
662			__be16 new_port, __sum16 *check)
663{
664	ovs_ct_clear(skb, NULL);
665	inet_proto_csum_replace2(check, skb, *port, new_port, false);
666	*port = new_port;
667}
668
669static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
670		   const struct ovs_key_udp *key,
671		   const struct ovs_key_udp *mask)
672{
673	struct udphdr *uh;
674	__be16 src, dst;
675	int err;
676
677	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
678				  sizeof(struct udphdr));
679	if (unlikely(err))
680		return err;
681
682	uh = udp_hdr(skb);
683	/* Either of the masks is non-zero, so do not bother checking them. */
684	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
685	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
686
687	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
688		if (likely(src != uh->source)) {
689			set_tp_port(skb, &uh->source, src, &uh->check);
690			flow_key->tp.src = src;
691		}
692		if (likely(dst != uh->dest)) {
693			set_tp_port(skb, &uh->dest, dst, &uh->check);
694			flow_key->tp.dst = dst;
695		}
696
697		if (unlikely(!uh->check))
698			uh->check = CSUM_MANGLED_0;
699	} else {
700		uh->source = src;
701		uh->dest = dst;
702		flow_key->tp.src = src;
703		flow_key->tp.dst = dst;
704		ovs_ct_clear(skb, NULL);
705	}
706
707	skb_clear_hash(skb);
708
709	return 0;
710}
711
712static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
713		   const struct ovs_key_tcp *key,
714		   const struct ovs_key_tcp *mask)
715{
716	struct tcphdr *th;
717	__be16 src, dst;
718	int err;
719
720	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
721				  sizeof(struct tcphdr));
722	if (unlikely(err))
723		return err;
724
725	th = tcp_hdr(skb);
726	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
727	if (likely(src != th->source)) {
728		set_tp_port(skb, &th->source, src, &th->check);
729		flow_key->tp.src = src;
730	}
731	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
732	if (likely(dst != th->dest)) {
733		set_tp_port(skb, &th->dest, dst, &th->check);
734		flow_key->tp.dst = dst;
735	}
736	skb_clear_hash(skb);
737
738	return 0;
739}
740
741static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
742		    const struct ovs_key_sctp *key,
743		    const struct ovs_key_sctp *mask)
744{
745	unsigned int sctphoff = skb_transport_offset(skb);
746	struct sctphdr *sh;
747	__le32 old_correct_csum, new_csum, old_csum;
748	int err;
749
750	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
751	if (unlikely(err))
752		return err;
753
754	sh = sctp_hdr(skb);
755	old_csum = sh->checksum;
756	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
757
758	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
759	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
760
761	new_csum = sctp_compute_cksum(skb, sctphoff);
762
763	/* Carry any checksum errors through. */
764	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
765
766	skb_clear_hash(skb);
767	ovs_ct_clear(skb, NULL);
768
769	flow_key->tp.src = sh->source;
770	flow_key->tp.dst = sh->dest;
771
772	return 0;
773}
774
775static int ovs_vport_output(struct net *net, struct sock *sk,
776			    struct sk_buff *skb)
777{
778	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
779	struct vport *vport = data->vport;
780
781	if (skb_cow_head(skb, data->l2_len) < 0) {
782		kfree_skb(skb);
783		return -ENOMEM;
784	}
785
786	__skb_dst_copy(skb, data->dst);
787	*OVS_CB(skb) = data->cb;
788	skb->inner_protocol = data->inner_protocol;
789	if (data->vlan_tci & VLAN_CFI_MASK)
790		__vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
791	else
792		__vlan_hwaccel_clear_tag(skb);
793
794	/* Reconstruct the MAC header.  */
795	skb_push(skb, data->l2_len);
796	memcpy(skb->data, &data->l2_data, data->l2_len);
797	skb_postpush_rcsum(skb, skb->data, data->l2_len);
798	skb_reset_mac_header(skb);
799
800	if (eth_p_mpls(skb->protocol)) {
801		skb->inner_network_header = skb->network_header;
802		skb_set_network_header(skb, data->network_offset);
803		skb_reset_mac_len(skb);
804	}
805
806	ovs_vport_send(vport, skb, data->mac_proto);
807	return 0;
808}
809
810static unsigned int
811ovs_dst_get_mtu(const struct dst_entry *dst)
812{
813	return dst->dev->mtu;
814}
815
816static struct dst_ops ovs_dst_ops = {
817	.family = AF_UNSPEC,
818	.mtu = ovs_dst_get_mtu,
819};
820
821/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
822 * ovs_vport_output(), which is called once per fragmented packet.
823 */
824static void prepare_frag(struct vport *vport, struct sk_buff *skb,
825			 u16 orig_network_offset, u8 mac_proto)
826{
827	unsigned int hlen = skb_network_offset(skb);
828	struct ovs_frag_data *data;
829
830	data = this_cpu_ptr(&ovs_frag_data_storage);
831	data->dst = skb->_skb_refdst;
832	data->vport = vport;
833	data->cb = *OVS_CB(skb);
834	data->inner_protocol = skb->inner_protocol;
835	data->network_offset = orig_network_offset;
836	if (skb_vlan_tag_present(skb))
837		data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
838	else
839		data->vlan_tci = 0;
840	data->vlan_proto = skb->vlan_proto;
841	data->mac_proto = mac_proto;
842	data->l2_len = hlen;
843	memcpy(&data->l2_data, skb->data, hlen);
844
845	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
846	skb_pull(skb, hlen);
847}
848
849static void ovs_fragment(struct net *net, struct vport *vport,
850			 struct sk_buff *skb, u16 mru,
851			 struct sw_flow_key *key)
852{
853	u16 orig_network_offset = 0;
854
855	if (eth_p_mpls(skb->protocol)) {
856		orig_network_offset = skb_network_offset(skb);
857		skb->network_header = skb->inner_network_header;
858	}
859
860	if (skb_network_offset(skb) > MAX_L2_LEN) {
861		OVS_NLERR(1, "L2 header too long to fragment");
862		goto err;
863	}
864
865	if (key->eth.type == htons(ETH_P_IP)) {
866		struct rtable ovs_rt = { 0 };
867		unsigned long orig_dst;
868
869		prepare_frag(vport, skb, orig_network_offset,
870			     ovs_key_mac_proto(key));
871		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
872			 DST_OBSOLETE_NONE, DST_NOCOUNT);
873		ovs_rt.dst.dev = vport->dev;
874
875		orig_dst = skb->_skb_refdst;
876		skb_dst_set_noref(skb, &ovs_rt.dst);
877		IPCB(skb)->frag_max_size = mru;
878
879		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
880		refdst_drop(orig_dst);
881	} else if (key->eth.type == htons(ETH_P_IPV6)) {
882		unsigned long orig_dst;
883		struct rt6_info ovs_rt;
884
885		prepare_frag(vport, skb, orig_network_offset,
886			     ovs_key_mac_proto(key));
887		memset(&ovs_rt, 0, sizeof(ovs_rt));
888		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
889			 DST_OBSOLETE_NONE, DST_NOCOUNT);
890		ovs_rt.dst.dev = vport->dev;
891
892		orig_dst = skb->_skb_refdst;
893		skb_dst_set_noref(skb, &ovs_rt.dst);
894		IP6CB(skb)->frag_max_size = mru;
895
896		ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
897		refdst_drop(orig_dst);
898	} else {
899		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
900			  ovs_vport_name(vport), ntohs(key->eth.type), mru,
901			  vport->dev->mtu);
902		goto err;
903	}
904
905	return;
906err:
907	kfree_skb(skb);
908}
909
910static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
911		      struct sw_flow_key *key)
912{
913	struct vport *vport = ovs_vport_rcu(dp, out_port);
914
915	if (likely(vport)) {
916		u16 mru = OVS_CB(skb)->mru;
917		u32 cutlen = OVS_CB(skb)->cutlen;
918
919		if (unlikely(cutlen > 0)) {
920			if (skb->len - cutlen > ovs_mac_header_len(key))
921				pskb_trim(skb, skb->len - cutlen);
922			else
923				pskb_trim(skb, ovs_mac_header_len(key));
924		}
925
926		if (likely(!mru ||
927		           (skb->len <= mru + vport->dev->hard_header_len))) {
928			ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
929		} else if (mru <= vport->dev->mtu) {
930			struct net *net = read_pnet(&dp->net);
931
932			ovs_fragment(net, vport, skb, mru, key);
933		} else {
934			kfree_skb(skb);
935		}
936	} else {
937		kfree_skb(skb);
938	}
939}
940
941static int output_userspace(struct datapath *dp, struct sk_buff *skb,
942			    struct sw_flow_key *key, const struct nlattr *attr,
943			    const struct nlattr *actions, int actions_len,
944			    uint32_t cutlen)
945{
946	struct dp_upcall_info upcall;
947	const struct nlattr *a;
948	int rem;
949
950	memset(&upcall, 0, sizeof(upcall));
951	upcall.cmd = OVS_PACKET_CMD_ACTION;
952	upcall.mru = OVS_CB(skb)->mru;
953
954	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
955	     a = nla_next(a, &rem)) {
956		switch (nla_type(a)) {
957		case OVS_USERSPACE_ATTR_USERDATA:
958			upcall.userdata = a;
959			break;
960
961		case OVS_USERSPACE_ATTR_PID:
962			upcall.portid = nla_get_u32(a);
963			break;
964
965		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
966			/* Get out tunnel info. */
967			struct vport *vport;
968
969			vport = ovs_vport_rcu(dp, nla_get_u32(a));
970			if (vport) {
971				int err;
972
973				err = dev_fill_metadata_dst(vport->dev, skb);
974				if (!err)
975					upcall.egress_tun_info = skb_tunnel_info(skb);
976			}
977
978			break;
979		}
980
981		case OVS_USERSPACE_ATTR_ACTIONS: {
982			/* Include actions. */
983			upcall.actions = actions;
984			upcall.actions_len = actions_len;
985			break;
986		}
987
988		} /* End of switch. */
989	}
990
991	return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
992}
993
994static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
995				     struct sw_flow_key *key,
996				     const struct nlattr *attr, bool last)
997{
998	/* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
999	struct nlattr *actions = nla_data(attr);
1000
1001	if (nla_len(actions))
1002		return clone_execute(dp, skb, key, 0, nla_data(actions),
1003				     nla_len(actions), last, false);
1004
1005	consume_skb(skb);
1006	return 0;
1007}
1008
1009/* When 'last' is true, sample() should always consume the 'skb'.
1010 * Otherwise, sample() should keep 'skb' intact regardless what
1011 * actions are executed within sample().
1012 */
1013static int sample(struct datapath *dp, struct sk_buff *skb,
1014		  struct sw_flow_key *key, const struct nlattr *attr,
1015		  bool last)
1016{
1017	struct nlattr *actions;
1018	struct nlattr *sample_arg;
1019	int rem = nla_len(attr);
1020	const struct sample_arg *arg;
1021	bool clone_flow_key;
1022
1023	/* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1024	sample_arg = nla_data(attr);
1025	arg = nla_data(sample_arg);
1026	actions = nla_next(sample_arg, &rem);
1027
1028	if ((arg->probability != U32_MAX) &&
1029	    (!arg->probability || prandom_u32() > arg->probability)) {
1030		if (last)
1031			consume_skb(skb);
1032		return 0;
1033	}
1034
1035	clone_flow_key = !arg->exec;
1036	return clone_execute(dp, skb, key, 0, actions, rem, last,
1037			     clone_flow_key);
1038}
1039
1040/* When 'last' is true, clone() should always consume the 'skb'.
1041 * Otherwise, clone() should keep 'skb' intact regardless what
1042 * actions are executed within clone().
1043 */
1044static int clone(struct datapath *dp, struct sk_buff *skb,
1045		 struct sw_flow_key *key, const struct nlattr *attr,
1046		 bool last)
1047{
1048	struct nlattr *actions;
1049	struct nlattr *clone_arg;
1050	int rem = nla_len(attr);
1051	bool dont_clone_flow_key;
1052
1053	/* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1054	clone_arg = nla_data(attr);
1055	dont_clone_flow_key = nla_get_u32(clone_arg);
1056	actions = nla_next(clone_arg, &rem);
1057
1058	return clone_execute(dp, skb, key, 0, actions, rem, last,
1059			     !dont_clone_flow_key);
1060}
1061
1062static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1063			 const struct nlattr *attr)
1064{
1065	struct ovs_action_hash *hash_act = nla_data(attr);
1066	u32 hash = 0;
1067
1068	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
1069	hash = skb_get_hash(skb);
1070	hash = jhash_1word(hash, hash_act->hash_basis);
1071	if (!hash)
1072		hash = 0x1;
1073
1074	key->ovs_flow_hash = hash;
1075}
1076
1077static int execute_set_action(struct sk_buff *skb,
1078			      struct sw_flow_key *flow_key,
1079			      const struct nlattr *a)
1080{
1081	/* Only tunnel set execution is supported without a mask. */
1082	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1083		struct ovs_tunnel_info *tun = nla_data(a);
1084
1085		skb_dst_drop(skb);
1086		dst_hold((struct dst_entry *)tun->tun_dst);
1087		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1088		return 0;
1089	}
1090
1091	return -EINVAL;
1092}
1093
1094/* Mask is at the midpoint of the data. */
1095#define get_mask(a, type) ((const type)nla_data(a) + 1)
1096
1097static int execute_masked_set_action(struct sk_buff *skb,
1098				     struct sw_flow_key *flow_key,
1099				     const struct nlattr *a)
1100{
1101	int err = 0;
1102
1103	switch (nla_type(a)) {
1104	case OVS_KEY_ATTR_PRIORITY:
1105		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1106			       *get_mask(a, u32 *));
1107		flow_key->phy.priority = skb->priority;
1108		break;
1109
1110	case OVS_KEY_ATTR_SKB_MARK:
1111		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1112		flow_key->phy.skb_mark = skb->mark;
1113		break;
1114
1115	case OVS_KEY_ATTR_TUNNEL_INFO:
1116		/* Masked data not supported for tunnel. */
1117		err = -EINVAL;
1118		break;
1119
1120	case OVS_KEY_ATTR_ETHERNET:
1121		err = set_eth_addr(skb, flow_key, nla_data(a),
1122				   get_mask(a, struct ovs_key_ethernet *));
1123		break;
1124
1125	case OVS_KEY_ATTR_NSH:
1126		err = set_nsh(skb, flow_key, a);
1127		break;
1128
1129	case OVS_KEY_ATTR_IPV4:
1130		err = set_ipv4(skb, flow_key, nla_data(a),
1131			       get_mask(a, struct ovs_key_ipv4 *));
1132		break;
1133
1134	case OVS_KEY_ATTR_IPV6:
1135		err = set_ipv6(skb, flow_key, nla_data(a),
1136			       get_mask(a, struct ovs_key_ipv6 *));
1137		break;
1138
1139	case OVS_KEY_ATTR_TCP:
1140		err = set_tcp(skb, flow_key, nla_data(a),
1141			      get_mask(a, struct ovs_key_tcp *));
1142		break;
1143
1144	case OVS_KEY_ATTR_UDP:
1145		err = set_udp(skb, flow_key, nla_data(a),
1146			      get_mask(a, struct ovs_key_udp *));
1147		break;
1148
1149	case OVS_KEY_ATTR_SCTP:
1150		err = set_sctp(skb, flow_key, nla_data(a),
1151			       get_mask(a, struct ovs_key_sctp *));
1152		break;
1153
1154	case OVS_KEY_ATTR_MPLS:
1155		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1156								    __be32 *));
1157		break;
1158
1159	case OVS_KEY_ATTR_CT_STATE:
1160	case OVS_KEY_ATTR_CT_ZONE:
1161	case OVS_KEY_ATTR_CT_MARK:
1162	case OVS_KEY_ATTR_CT_LABELS:
1163	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1164	case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1165		err = -EINVAL;
1166		break;
1167	}
1168
1169	return err;
1170}
1171
1172static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1173			  struct sw_flow_key *key,
1174			  const struct nlattr *a, bool last)
1175{
1176	u32 recirc_id;
1177
1178	if (!is_flow_key_valid(key)) {
1179		int err;
1180
1181		err = ovs_flow_key_update(skb, key);
1182		if (err)
1183			return err;
1184	}
1185	BUG_ON(!is_flow_key_valid(key));
1186
1187	recirc_id = nla_get_u32(a);
1188	return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1189}
1190
1191static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1192				 struct sw_flow_key *key,
1193				 const struct nlattr *attr, bool last)
1194{
1195	struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1196	const struct nlattr *actions, *cpl_arg;
1197	int len, max_len, rem = nla_len(attr);
1198	const struct check_pkt_len_arg *arg;
1199	bool clone_flow_key;
1200
1201	/* The first netlink attribute in 'attr' is always
1202	 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1203	 */
1204	cpl_arg = nla_data(attr);
1205	arg = nla_data(cpl_arg);
1206
1207	len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1208	max_len = arg->pkt_len;
1209
1210	if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1211	    len <= max_len) {
1212		/* Second netlink attribute in 'attr' is always
1213		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1214		 */
1215		actions = nla_next(cpl_arg, &rem);
1216		clone_flow_key = !arg->exec_for_lesser_equal;
1217	} else {
1218		/* Third netlink attribute in 'attr' is always
1219		 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1220		 */
1221		actions = nla_next(cpl_arg, &rem);
1222		actions = nla_next(actions, &rem);
1223		clone_flow_key = !arg->exec_for_greater;
1224	}
1225
1226	return clone_execute(dp, skb, key, 0, nla_data(actions),
1227			     nla_len(actions), last, clone_flow_key);
1228}
1229
1230static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1231{
1232	int err;
1233
1234	if (skb->protocol == htons(ETH_P_IPV6)) {
1235		struct ipv6hdr *nh;
1236
1237		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1238					  sizeof(*nh));
1239		if (unlikely(err))
1240			return err;
1241
1242		nh = ipv6_hdr(skb);
1243
1244		if (nh->hop_limit <= 1)
1245			return -EHOSTUNREACH;
1246
1247		key->ip.ttl = --nh->hop_limit;
1248	} else if (skb->protocol == htons(ETH_P_IP)) {
1249		struct iphdr *nh;
1250		u8 old_ttl;
1251
1252		err = skb_ensure_writable(skb, skb_network_offset(skb) +
1253					  sizeof(*nh));
1254		if (unlikely(err))
1255			return err;
1256
1257		nh = ip_hdr(skb);
1258		if (nh->ttl <= 1)
1259			return -EHOSTUNREACH;
1260
1261		old_ttl = nh->ttl--;
1262		csum_replace2(&nh->check, htons(old_ttl << 8),
1263			      htons(nh->ttl << 8));
1264		key->ip.ttl = nh->ttl;
1265	}
1266	return 0;
1267}
1268
1269/* Execute a list of actions against 'skb'. */
1270static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1271			      struct sw_flow_key *key,
1272			      const struct nlattr *attr, int len)
1273{
1274	const struct nlattr *a;
1275	int rem;
1276
1277	for (a = attr, rem = len; rem > 0;
1278	     a = nla_next(a, &rem)) {
1279		int err = 0;
1280
1281		switch (nla_type(a)) {
1282		case OVS_ACTION_ATTR_OUTPUT: {
1283			int port = nla_get_u32(a);
1284			struct sk_buff *clone;
1285
1286			/* Every output action needs a separate clone
1287			 * of 'skb', In case the output action is the
1288			 * last action, cloning can be avoided.
1289			 */
1290			if (nla_is_last(a, rem)) {
1291				do_output(dp, skb, port, key);
1292				/* 'skb' has been used for output.
1293				 */
1294				return 0;
1295			}
1296
1297			clone = skb_clone(skb, GFP_ATOMIC);
1298			if (clone)
1299				do_output(dp, clone, port, key);
1300			OVS_CB(skb)->cutlen = 0;
1301			break;
1302		}
1303
1304		case OVS_ACTION_ATTR_TRUNC: {
1305			struct ovs_action_trunc *trunc = nla_data(a);
1306
1307			if (skb->len > trunc->max_len)
1308				OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1309			break;
1310		}
1311
1312		case OVS_ACTION_ATTR_USERSPACE:
1313			output_userspace(dp, skb, key, a, attr,
1314						     len, OVS_CB(skb)->cutlen);
1315			OVS_CB(skb)->cutlen = 0;
1316			break;
1317
1318		case OVS_ACTION_ATTR_HASH:
1319			execute_hash(skb, key, a);
1320			break;
1321
1322		case OVS_ACTION_ATTR_PUSH_MPLS: {
1323			struct ovs_action_push_mpls *mpls = nla_data(a);
1324
1325			err = push_mpls(skb, key, mpls->mpls_lse,
1326					mpls->mpls_ethertype, skb->mac_len);
1327			break;
1328		}
1329		case OVS_ACTION_ATTR_ADD_MPLS: {
1330			struct ovs_action_add_mpls *mpls = nla_data(a);
1331			__u16 mac_len = 0;
1332
1333			if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1334				mac_len = skb->mac_len;
1335
1336			err = push_mpls(skb, key, mpls->mpls_lse,
1337					mpls->mpls_ethertype, mac_len);
1338			break;
1339		}
1340		case OVS_ACTION_ATTR_POP_MPLS:
1341			err = pop_mpls(skb, key, nla_get_be16(a));
1342			break;
1343
1344		case OVS_ACTION_ATTR_PUSH_VLAN:
1345			err = push_vlan(skb, key, nla_data(a));
1346			break;
1347
1348		case OVS_ACTION_ATTR_POP_VLAN:
1349			err = pop_vlan(skb, key);
1350			break;
1351
1352		case OVS_ACTION_ATTR_RECIRC: {
1353			bool last = nla_is_last(a, rem);
1354
1355			err = execute_recirc(dp, skb, key, a, last);
1356			if (last) {
1357				/* If this is the last action, the skb has
1358				 * been consumed or freed.
1359				 * Return immediately.
1360				 */
1361				return err;
1362			}
1363			break;
1364		}
1365
1366		case OVS_ACTION_ATTR_SET:
1367			err = execute_set_action(skb, key, nla_data(a));
1368			break;
1369
1370		case OVS_ACTION_ATTR_SET_MASKED:
1371		case OVS_ACTION_ATTR_SET_TO_MASKED:
1372			err = execute_masked_set_action(skb, key, nla_data(a));
1373			break;
1374
1375		case OVS_ACTION_ATTR_SAMPLE: {
1376			bool last = nla_is_last(a, rem);
1377
1378			err = sample(dp, skb, key, a, last);
1379			if (last)
1380				return err;
1381
1382			break;
1383		}
1384
1385		case OVS_ACTION_ATTR_CT:
1386			if (!is_flow_key_valid(key)) {
1387				err = ovs_flow_key_update(skb, key);
1388				if (err)
1389					return err;
1390			}
1391
1392			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1393					     nla_data(a));
1394
1395			/* Hide stolen IP fragments from user space. */
1396			if (err)
1397				return err == -EINPROGRESS ? 0 : err;
1398			break;
1399
1400		case OVS_ACTION_ATTR_CT_CLEAR:
1401			err = ovs_ct_clear(skb, key);
1402			break;
1403
1404		case OVS_ACTION_ATTR_PUSH_ETH:
1405			err = push_eth(skb, key, nla_data(a));
1406			break;
1407
1408		case OVS_ACTION_ATTR_POP_ETH:
1409			err = pop_eth(skb, key);
1410			break;
1411
1412		case OVS_ACTION_ATTR_PUSH_NSH: {
1413			u8 buffer[NSH_HDR_MAX_LEN];
1414			struct nshhdr *nh = (struct nshhdr *)buffer;
1415
1416			err = nsh_hdr_from_nlattr(nla_data(a), nh,
1417						  NSH_HDR_MAX_LEN);
1418			if (unlikely(err))
1419				break;
1420			err = push_nsh(skb, key, nh);
1421			break;
1422		}
1423
1424		case OVS_ACTION_ATTR_POP_NSH:
1425			err = pop_nsh(skb, key);
1426			break;
1427
1428		case OVS_ACTION_ATTR_METER:
1429			if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1430				consume_skb(skb);
1431				return 0;
1432			}
1433			break;
1434
1435		case OVS_ACTION_ATTR_CLONE: {
1436			bool last = nla_is_last(a, rem);
1437
1438			err = clone(dp, skb, key, a, last);
1439			if (last)
1440				return err;
1441
1442			break;
1443		}
1444
1445		case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1446			bool last = nla_is_last(a, rem);
1447
1448			err = execute_check_pkt_len(dp, skb, key, a, last);
1449			if (last)
1450				return err;
1451
1452			break;
1453		}
1454
1455		case OVS_ACTION_ATTR_DEC_TTL:
1456			err = execute_dec_ttl(skb, key);
1457			if (err == -EHOSTUNREACH) {
1458				err = dec_ttl_exception_handler(dp, skb, key,
1459								a, true);
1460				return err;
1461			}
1462			break;
1463		}
1464
1465		if (unlikely(err)) {
1466			kfree_skb(skb);
1467			return err;
1468		}
1469	}
1470
1471	consume_skb(skb);
1472	return 0;
1473}
1474
1475/* Execute the actions on the clone of the packet. The effect of the
1476 * execution does not affect the original 'skb' nor the original 'key'.
1477 *
1478 * The execution may be deferred in case the actions can not be executed
1479 * immediately.
1480 */
1481static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1482			 struct sw_flow_key *key, u32 recirc_id,
1483			 const struct nlattr *actions, int len,
1484			 bool last, bool clone_flow_key)
1485{
1486	struct deferred_action *da;
1487	struct sw_flow_key *clone;
1488
1489	skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1490	if (!skb) {
1491		/* Out of memory, skip this action.
1492		 */
1493		return 0;
1494	}
1495
1496	/* When clone_flow_key is false, the 'key' will not be change
1497	 * by the actions, then the 'key' can be used directly.
1498	 * Otherwise, try to clone key from the next recursion level of
1499	 * 'flow_keys'. If clone is successful, execute the actions
1500	 * without deferring.
1501	 */
1502	clone = clone_flow_key ? clone_key(key) : key;
1503	if (clone) {
1504		int err = 0;
1505
1506		if (actions) { /* Sample action */
1507			if (clone_flow_key)
1508				__this_cpu_inc(exec_actions_level);
1509
1510			err = do_execute_actions(dp, skb, clone,
1511						 actions, len);
1512
1513			if (clone_flow_key)
1514				__this_cpu_dec(exec_actions_level);
1515		} else { /* Recirc action */
1516			clone->recirc_id = recirc_id;
1517			ovs_dp_process_packet(skb, clone);
1518		}
1519		return err;
1520	}
1521
1522	/* Out of 'flow_keys' space. Defer actions */
1523	da = add_deferred_actions(skb, key, actions, len);
1524	if (da) {
1525		if (!actions) { /* Recirc action */
1526			key = &da->pkt_key;
1527			key->recirc_id = recirc_id;
1528		}
1529	} else {
1530		/* Out of per CPU action FIFO space. Drop the 'skb' and
1531		 * log an error.
1532		 */
1533		kfree_skb(skb);
1534
1535		if (net_ratelimit()) {
1536			if (actions) { /* Sample action */
1537				pr_warn("%s: deferred action limit reached, drop sample action\n",
1538					ovs_dp_name(dp));
1539			} else {  /* Recirc action */
1540				pr_warn("%s: deferred action limit reached, drop recirc action\n",
1541					ovs_dp_name(dp));
1542			}
1543		}
1544	}
1545	return 0;
1546}
1547
1548static void process_deferred_actions(struct datapath *dp)
1549{
1550	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1551
1552	/* Do not touch the FIFO in case there is no deferred actions. */
1553	if (action_fifo_is_empty(fifo))
1554		return;
1555
1556	/* Finishing executing all deferred actions. */
1557	do {
1558		struct deferred_action *da = action_fifo_get(fifo);
1559		struct sk_buff *skb = da->skb;
1560		struct sw_flow_key *key = &da->pkt_key;
1561		const struct nlattr *actions = da->actions;
1562		int actions_len = da->actions_len;
1563
1564		if (actions)
1565			do_execute_actions(dp, skb, key, actions, actions_len);
1566		else
1567			ovs_dp_process_packet(skb, key);
1568	} while (!action_fifo_is_empty(fifo));
1569
1570	/* Reset FIFO for the next packet.  */
1571	action_fifo_init(fifo);
1572}
1573
1574/* Execute a list of actions against 'skb'. */
1575int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1576			const struct sw_flow_actions *acts,
1577			struct sw_flow_key *key)
1578{
1579	int err, level;
1580
1581	level = __this_cpu_inc_return(exec_actions_level);
1582	if (unlikely(level > OVS_RECURSION_LIMIT)) {
1583		net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1584				     ovs_dp_name(dp));
1585		kfree_skb(skb);
1586		err = -ENETDOWN;
1587		goto out;
1588	}
1589
1590	OVS_CB(skb)->acts_origlen = acts->orig_len;
1591	err = do_execute_actions(dp, skb, key,
1592				 acts->actions, acts->actions_len);
1593
1594	if (level == 1)
1595		process_deferred_actions(dp);
1596
1597out:
1598	__this_cpu_dec(exec_actions_level);
1599	return err;
1600}
1601
1602int action_fifos_init(void)
1603{
1604	action_fifos = alloc_percpu(struct action_fifo);
1605	if (!action_fifos)
1606		return -ENOMEM;
1607
1608	flow_keys = alloc_percpu(struct action_flow_keys);
1609	if (!flow_keys) {
1610		free_percpu(action_fifos);
1611		return -ENOMEM;
1612	}
1613
1614	return 0;
1615}
1616
1617void action_fifos_exit(void)
1618{
1619	free_percpu(action_fifos);
1620	free_percpu(flow_keys);
1621}
1622