1/*
2 * Copyright (c) 2007-2014 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21#include <linux/skbuff.h>
22#include <linux/in.h>
23#include <linux/ip.h>
24#include <linux/openvswitch.h>
25#include <linux/sctp.h>
26#include <linux/tcp.h>
27#include <linux/udp.h>
28#include <linux/in6.h>
29#include <linux/if_arp.h>
30#include <linux/if_vlan.h>
31
32#include <net/ip.h>
33#include <net/ipv6.h>
34#include <net/checksum.h>
35#include <net/dsfield.h>
36#include <net/mpls.h>
37#include <net/sctp/checksum.h>
38
39#include "datapath.h"
40#include "flow.h"
41#include "vport.h"
42
43static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
44			      struct sw_flow_key *key,
45			      const struct nlattr *attr, int len);
46
47struct deferred_action {
48	struct sk_buff *skb;
49	const struct nlattr *actions;
50
51	/* Store pkt_key clone when creating deferred action. */
52	struct sw_flow_key pkt_key;
53};
54
55#define DEFERRED_ACTION_FIFO_SIZE 10
56struct action_fifo {
57	int head;
58	int tail;
59	/* Deferred action fifo queue storage. */
60	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
61};
62
63static struct action_fifo __percpu *action_fifos;
64static DEFINE_PER_CPU(int, exec_actions_level);
65
66static void action_fifo_init(struct action_fifo *fifo)
67{
68	fifo->head = 0;
69	fifo->tail = 0;
70}
71
72static bool action_fifo_is_empty(const struct action_fifo *fifo)
73{
74	return (fifo->head == fifo->tail);
75}
76
77static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
78{
79	if (action_fifo_is_empty(fifo))
80		return NULL;
81
82	return &fifo->fifo[fifo->tail++];
83}
84
85static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
86{
87	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
88		return NULL;
89
90	return &fifo->fifo[fifo->head++];
91}
92
93/* Return true if fifo is not full */
94static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
95						    const struct sw_flow_key *key,
96						    const struct nlattr *attr)
97{
98	struct action_fifo *fifo;
99	struct deferred_action *da;
100
101	fifo = this_cpu_ptr(action_fifos);
102	da = action_fifo_put(fifo);
103	if (da) {
104		da->skb = skb;
105		da->actions = attr;
106		da->pkt_key = *key;
107	}
108
109	return da;
110}
111
112static void invalidate_flow_key(struct sw_flow_key *key)
113{
114	key->eth.type = htons(0);
115}
116
117static bool is_flow_key_valid(const struct sw_flow_key *key)
118{
119	return !!key->eth.type;
120}
121
122static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
123		     const struct ovs_action_push_mpls *mpls)
124{
125	__be32 *new_mpls_lse;
126	struct ethhdr *hdr;
127
128	/* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
129	if (skb->encapsulation)
130		return -ENOTSUPP;
131
132	if (skb_cow_head(skb, MPLS_HLEN) < 0)
133		return -ENOMEM;
134
135	skb_push(skb, MPLS_HLEN);
136	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
137		skb->mac_len);
138	skb_reset_mac_header(skb);
139
140	new_mpls_lse = (__be32 *)skb_mpls_header(skb);
141	*new_mpls_lse = mpls->mpls_lse;
142
143	if (skb->ip_summed == CHECKSUM_COMPLETE)
144		skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
145							     MPLS_HLEN, 0));
146
147	hdr = eth_hdr(skb);
148	hdr->h_proto = mpls->mpls_ethertype;
149
150	if (!skb->inner_protocol)
151		skb_set_inner_protocol(skb, skb->protocol);
152	skb->protocol = mpls->mpls_ethertype;
153
154	invalidate_flow_key(key);
155	return 0;
156}
157
158static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
159		    const __be16 ethertype)
160{
161	struct ethhdr *hdr;
162	int err;
163
164	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
165	if (unlikely(err))
166		return err;
167
168	skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
169
170	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
171		skb->mac_len);
172
173	__skb_pull(skb, MPLS_HLEN);
174	skb_reset_mac_header(skb);
175
176	/* skb_mpls_header() is used to locate the ethertype
177	 * field correctly in the presence of VLAN tags.
178	 */
179	hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
180	hdr->h_proto = ethertype;
181	if (eth_p_mpls(skb->protocol))
182		skb->protocol = ethertype;
183
184	invalidate_flow_key(key);
185	return 0;
186}
187
188/* 'KEY' must not have any bits set outside of the 'MASK' */
189#define MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK)))
190#define SET_MASKED(OLD, KEY, MASK) ((OLD) = MASKED(OLD, KEY, MASK))
191
192static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
193		    const __be32 *mpls_lse, const __be32 *mask)
194{
195	__be32 *stack;
196	__be32 lse;
197	int err;
198
199	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
200	if (unlikely(err))
201		return err;
202
203	stack = (__be32 *)skb_mpls_header(skb);
204	lse = MASKED(*stack, *mpls_lse, *mask);
205	if (skb->ip_summed == CHECKSUM_COMPLETE) {
206		__be32 diff[] = { ~(*stack), lse };
207
208		skb->csum = ~csum_partial((char *)diff, sizeof(diff),
209					  ~skb->csum);
210	}
211
212	*stack = lse;
213	flow_key->mpls.top_lse = lse;
214	return 0;
215}
216
217static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
218{
219	int err;
220
221	err = skb_vlan_pop(skb);
222	if (skb_vlan_tag_present(skb))
223		invalidate_flow_key(key);
224	else
225		key->eth.tci = 0;
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.tci = vlan->vlan_tci;
236	return skb_vlan_push(skb, vlan->vlan_tpid,
237			     ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
238}
239
240/* 'src' is already properly masked. */
241static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
242{
243	u16 *dst = (u16 *)dst_;
244	const u16 *src = (const u16 *)src_;
245	const u16 *mask = (const u16 *)mask_;
246
247	SET_MASKED(dst[0], src[0], mask[0]);
248	SET_MASKED(dst[1], src[1], mask[1]);
249	SET_MASKED(dst[2], src[2], mask[2]);
250}
251
252static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
253			const struct ovs_key_ethernet *key,
254			const struct ovs_key_ethernet *mask)
255{
256	int err;
257
258	err = skb_ensure_writable(skb, ETH_HLEN);
259	if (unlikely(err))
260		return err;
261
262	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
263
264	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
265			       mask->eth_src);
266	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
267			       mask->eth_dst);
268
269	ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
270
271	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
272	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
273	return 0;
274}
275
276static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
277			__be32 *addr, __be32 new_addr)
278{
279	int transport_len = skb->len - skb_transport_offset(skb);
280
281	if (nh->protocol == IPPROTO_TCP) {
282		if (likely(transport_len >= sizeof(struct tcphdr)))
283			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
284						 *addr, new_addr, 1);
285	} else if (nh->protocol == IPPROTO_UDP) {
286		if (likely(transport_len >= sizeof(struct udphdr))) {
287			struct udphdr *uh = udp_hdr(skb);
288
289			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
290				inet_proto_csum_replace4(&uh->check, skb,
291							 *addr, new_addr, 1);
292				if (!uh->check)
293					uh->check = CSUM_MANGLED_0;
294			}
295		}
296	}
297
298	csum_replace4(&nh->check, *addr, new_addr);
299	skb_clear_hash(skb);
300	*addr = new_addr;
301}
302
303static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
304				 __be32 addr[4], const __be32 new_addr[4])
305{
306	int transport_len = skb->len - skb_transport_offset(skb);
307
308	if (l4_proto == NEXTHDR_TCP) {
309		if (likely(transport_len >= sizeof(struct tcphdr)))
310			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
311						  addr, new_addr, 1);
312	} else if (l4_proto == NEXTHDR_UDP) {
313		if (likely(transport_len >= sizeof(struct udphdr))) {
314			struct udphdr *uh = udp_hdr(skb);
315
316			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
317				inet_proto_csum_replace16(&uh->check, skb,
318							  addr, new_addr, 1);
319				if (!uh->check)
320					uh->check = CSUM_MANGLED_0;
321			}
322		}
323	} else if (l4_proto == NEXTHDR_ICMP) {
324		if (likely(transport_len >= sizeof(struct icmp6hdr)))
325			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
326						  skb, addr, new_addr, 1);
327	}
328}
329
330static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
331			   const __be32 mask[4], __be32 masked[4])
332{
333	masked[0] = MASKED(old[0], addr[0], mask[0]);
334	masked[1] = MASKED(old[1], addr[1], mask[1]);
335	masked[2] = MASKED(old[2], addr[2], mask[2]);
336	masked[3] = MASKED(old[3], addr[3], mask[3]);
337}
338
339static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
340			  __be32 addr[4], const __be32 new_addr[4],
341			  bool recalculate_csum)
342{
343	if (recalculate_csum)
344		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
345
346	skb_clear_hash(skb);
347	memcpy(addr, new_addr, sizeof(__be32[4]));
348}
349
350static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
351{
352	/* Bits 21-24 are always unmasked, so this retains their values. */
353	SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
354	SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
355	SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
356}
357
358static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
359		       u8 mask)
360{
361	new_ttl = MASKED(nh->ttl, new_ttl, mask);
362
363	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
364	nh->ttl = new_ttl;
365}
366
367static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
368		    const struct ovs_key_ipv4 *key,
369		    const struct ovs_key_ipv4 *mask)
370{
371	struct iphdr *nh;
372	__be32 new_addr;
373	int err;
374
375	err = skb_ensure_writable(skb, skb_network_offset(skb) +
376				  sizeof(struct iphdr));
377	if (unlikely(err))
378		return err;
379
380	nh = ip_hdr(skb);
381
382	/* Setting an IP addresses is typically only a side effect of
383	 * matching on them in the current userspace implementation, so it
384	 * makes sense to check if the value actually changed.
385	 */
386	if (mask->ipv4_src) {
387		new_addr = MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
388
389		if (unlikely(new_addr != nh->saddr)) {
390			set_ip_addr(skb, nh, &nh->saddr, new_addr);
391			flow_key->ipv4.addr.src = new_addr;
392		}
393	}
394	if (mask->ipv4_dst) {
395		new_addr = MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
396
397		if (unlikely(new_addr != nh->daddr)) {
398			set_ip_addr(skb, nh, &nh->daddr, new_addr);
399			flow_key->ipv4.addr.dst = new_addr;
400		}
401	}
402	if (mask->ipv4_tos) {
403		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
404		flow_key->ip.tos = nh->tos;
405	}
406	if (mask->ipv4_ttl) {
407		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
408		flow_key->ip.ttl = nh->ttl;
409	}
410
411	return 0;
412}
413
414static bool is_ipv6_mask_nonzero(const __be32 addr[4])
415{
416	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
417}
418
419static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
420		    const struct ovs_key_ipv6 *key,
421		    const struct ovs_key_ipv6 *mask)
422{
423	struct ipv6hdr *nh;
424	int err;
425
426	err = skb_ensure_writable(skb, skb_network_offset(skb) +
427				  sizeof(struct ipv6hdr));
428	if (unlikely(err))
429		return err;
430
431	nh = ipv6_hdr(skb);
432
433	/* Setting an IP addresses is typically only a side effect of
434	 * matching on them in the current userspace implementation, so it
435	 * makes sense to check if the value actually changed.
436	 */
437	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
438		__be32 *saddr = (__be32 *)&nh->saddr;
439		__be32 masked[4];
440
441		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
442
443		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
444			set_ipv6_addr(skb, key->ipv6_proto, saddr, masked,
445				      true);
446			memcpy(&flow_key->ipv6.addr.src, masked,
447			       sizeof(flow_key->ipv6.addr.src));
448		}
449	}
450	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
451		unsigned int offset = 0;
452		int flags = IP6_FH_F_SKIP_RH;
453		bool recalc_csum = true;
454		__be32 *daddr = (__be32 *)&nh->daddr;
455		__be32 masked[4];
456
457		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
458
459		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
460			if (ipv6_ext_hdr(nh->nexthdr))
461				recalc_csum = (ipv6_find_hdr(skb, &offset,
462							     NEXTHDR_ROUTING,
463							     NULL, &flags)
464					       != NEXTHDR_ROUTING);
465
466			set_ipv6_addr(skb, key->ipv6_proto, daddr, masked,
467				      recalc_csum);
468			memcpy(&flow_key->ipv6.addr.dst, masked,
469			       sizeof(flow_key->ipv6.addr.dst));
470		}
471	}
472	if (mask->ipv6_tclass) {
473		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
474		flow_key->ip.tos = ipv6_get_dsfield(nh);
475	}
476	if (mask->ipv6_label) {
477		set_ipv6_fl(nh, ntohl(key->ipv6_label),
478			    ntohl(mask->ipv6_label));
479		flow_key->ipv6.label =
480		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
481	}
482	if (mask->ipv6_hlimit) {
483		SET_MASKED(nh->hop_limit, key->ipv6_hlimit, mask->ipv6_hlimit);
484		flow_key->ip.ttl = nh->hop_limit;
485	}
486	return 0;
487}
488
489/* Must follow skb_ensure_writable() since that can move the skb data. */
490static void set_tp_port(struct sk_buff *skb, __be16 *port,
491			__be16 new_port, __sum16 *check)
492{
493	inet_proto_csum_replace2(check, skb, *port, new_port, 0);
494	*port = new_port;
495}
496
497static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
498		   const struct ovs_key_udp *key,
499		   const struct ovs_key_udp *mask)
500{
501	struct udphdr *uh;
502	__be16 src, dst;
503	int err;
504
505	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
506				  sizeof(struct udphdr));
507	if (unlikely(err))
508		return err;
509
510	uh = udp_hdr(skb);
511	/* Either of the masks is non-zero, so do not bother checking them. */
512	src = MASKED(uh->source, key->udp_src, mask->udp_src);
513	dst = MASKED(uh->dest, key->udp_dst, mask->udp_dst);
514
515	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
516		if (likely(src != uh->source)) {
517			set_tp_port(skb, &uh->source, src, &uh->check);
518			flow_key->tp.src = src;
519		}
520		if (likely(dst != uh->dest)) {
521			set_tp_port(skb, &uh->dest, dst, &uh->check);
522			flow_key->tp.dst = dst;
523		}
524
525		if (unlikely(!uh->check))
526			uh->check = CSUM_MANGLED_0;
527	} else {
528		uh->source = src;
529		uh->dest = dst;
530		flow_key->tp.src = src;
531		flow_key->tp.dst = dst;
532	}
533
534	skb_clear_hash(skb);
535
536	return 0;
537}
538
539static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
540		   const struct ovs_key_tcp *key,
541		   const struct ovs_key_tcp *mask)
542{
543	struct tcphdr *th;
544	__be16 src, dst;
545	int err;
546
547	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
548				  sizeof(struct tcphdr));
549	if (unlikely(err))
550		return err;
551
552	th = tcp_hdr(skb);
553	src = MASKED(th->source, key->tcp_src, mask->tcp_src);
554	if (likely(src != th->source)) {
555		set_tp_port(skb, &th->source, src, &th->check);
556		flow_key->tp.src = src;
557	}
558	dst = MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
559	if (likely(dst != th->dest)) {
560		set_tp_port(skb, &th->dest, dst, &th->check);
561		flow_key->tp.dst = dst;
562	}
563	skb_clear_hash(skb);
564
565	return 0;
566}
567
568static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
569		    const struct ovs_key_sctp *key,
570		    const struct ovs_key_sctp *mask)
571{
572	unsigned int sctphoff = skb_transport_offset(skb);
573	struct sctphdr *sh;
574	__le32 old_correct_csum, new_csum, old_csum;
575	int err;
576
577	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
578	if (unlikely(err))
579		return err;
580
581	sh = sctp_hdr(skb);
582	old_csum = sh->checksum;
583	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
584
585	sh->source = MASKED(sh->source, key->sctp_src, mask->sctp_src);
586	sh->dest = MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
587
588	new_csum = sctp_compute_cksum(skb, sctphoff);
589
590	/* Carry any checksum errors through. */
591	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
592
593	skb_clear_hash(skb);
594	flow_key->tp.src = sh->source;
595	flow_key->tp.dst = sh->dest;
596
597	return 0;
598}
599
600static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
601{
602	struct vport *vport = ovs_vport_rcu(dp, out_port);
603
604	if (likely(vport))
605		ovs_vport_send(vport, skb);
606	else
607		kfree_skb(skb);
608}
609
610static int output_userspace(struct datapath *dp, struct sk_buff *skb,
611			    struct sw_flow_key *key, const struct nlattr *attr)
612{
613	struct ovs_tunnel_info info;
614	struct dp_upcall_info upcall;
615	const struct nlattr *a;
616	int rem;
617
618	upcall.cmd = OVS_PACKET_CMD_ACTION;
619	upcall.userdata = NULL;
620	upcall.portid = 0;
621	upcall.egress_tun_info = NULL;
622
623	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
624		 a = nla_next(a, &rem)) {
625		switch (nla_type(a)) {
626		case OVS_USERSPACE_ATTR_USERDATA:
627			upcall.userdata = a;
628			break;
629
630		case OVS_USERSPACE_ATTR_PID:
631			upcall.portid = nla_get_u32(a);
632			break;
633
634		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
635			/* Get out tunnel info. */
636			struct vport *vport;
637
638			vport = ovs_vport_rcu(dp, nla_get_u32(a));
639			if (vport) {
640				int err;
641
642				err = ovs_vport_get_egress_tun_info(vport, skb,
643								    &info);
644				if (!err)
645					upcall.egress_tun_info = &info;
646			}
647			break;
648		}
649
650		} /* End of switch. */
651	}
652
653	return ovs_dp_upcall(dp, skb, key, &upcall);
654}
655
656static int sample(struct datapath *dp, struct sk_buff *skb,
657		  struct sw_flow_key *key, const struct nlattr *attr)
658{
659	const struct nlattr *acts_list = NULL;
660	const struct nlattr *a;
661	int rem;
662
663	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
664		 a = nla_next(a, &rem)) {
665		switch (nla_type(a)) {
666		case OVS_SAMPLE_ATTR_PROBABILITY:
667			if (prandom_u32() >= nla_get_u32(a))
668				return 0;
669			break;
670
671		case OVS_SAMPLE_ATTR_ACTIONS:
672			acts_list = a;
673			break;
674		}
675	}
676
677	rem = nla_len(acts_list);
678	a = nla_data(acts_list);
679
680	/* Actions list is empty, do nothing */
681	if (unlikely(!rem))
682		return 0;
683
684	/* The only known usage of sample action is having a single user-space
685	 * action. Treat this usage as a special case.
686	 * The output_userspace() should clone the skb to be sent to the
687	 * user space. This skb will be consumed by its caller.
688	 */
689	if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
690		   nla_is_last(a, rem)))
691		return output_userspace(dp, skb, key, a);
692
693	skb = skb_clone(skb, GFP_ATOMIC);
694	if (!skb)
695		/* Skip the sample action when out of memory. */
696		return 0;
697
698	if (!add_deferred_actions(skb, key, a)) {
699		if (net_ratelimit())
700			pr_warn("%s: deferred actions limit reached, dropping sample action\n",
701				ovs_dp_name(dp));
702
703		kfree_skb(skb);
704	}
705	return 0;
706}
707
708static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
709			 const struct nlattr *attr)
710{
711	struct ovs_action_hash *hash_act = nla_data(attr);
712	u32 hash = 0;
713
714	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
715	hash = skb_get_hash(skb);
716	hash = jhash_1word(hash, hash_act->hash_basis);
717	if (!hash)
718		hash = 0x1;
719
720	key->ovs_flow_hash = hash;
721}
722
723static int execute_set_action(struct sk_buff *skb,
724			      struct sw_flow_key *flow_key,
725			      const struct nlattr *a)
726{
727	/* Only tunnel set execution is supported without a mask. */
728	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
729		OVS_CB(skb)->egress_tun_info = nla_data(a);
730		return 0;
731	}
732
733	return -EINVAL;
734}
735
736/* Mask is at the midpoint of the data. */
737#define get_mask(a, type) ((const type)nla_data(a) + 1)
738
739static int execute_masked_set_action(struct sk_buff *skb,
740				     struct sw_flow_key *flow_key,
741				     const struct nlattr *a)
742{
743	int err = 0;
744
745	switch (nla_type(a)) {
746	case OVS_KEY_ATTR_PRIORITY:
747		SET_MASKED(skb->priority, nla_get_u32(a), *get_mask(a, u32 *));
748		flow_key->phy.priority = skb->priority;
749		break;
750
751	case OVS_KEY_ATTR_SKB_MARK:
752		SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
753		flow_key->phy.skb_mark = skb->mark;
754		break;
755
756	case OVS_KEY_ATTR_TUNNEL_INFO:
757		/* Masked data not supported for tunnel. */
758		err = -EINVAL;
759		break;
760
761	case OVS_KEY_ATTR_ETHERNET:
762		err = set_eth_addr(skb, flow_key, nla_data(a),
763				   get_mask(a, struct ovs_key_ethernet *));
764		break;
765
766	case OVS_KEY_ATTR_IPV4:
767		err = set_ipv4(skb, flow_key, nla_data(a),
768			       get_mask(a, struct ovs_key_ipv4 *));
769		break;
770
771	case OVS_KEY_ATTR_IPV6:
772		err = set_ipv6(skb, flow_key, nla_data(a),
773			       get_mask(a, struct ovs_key_ipv6 *));
774		break;
775
776	case OVS_KEY_ATTR_TCP:
777		err = set_tcp(skb, flow_key, nla_data(a),
778			      get_mask(a, struct ovs_key_tcp *));
779		break;
780
781	case OVS_KEY_ATTR_UDP:
782		err = set_udp(skb, flow_key, nla_data(a),
783			      get_mask(a, struct ovs_key_udp *));
784		break;
785
786	case OVS_KEY_ATTR_SCTP:
787		err = set_sctp(skb, flow_key, nla_data(a),
788			       get_mask(a, struct ovs_key_sctp *));
789		break;
790
791	case OVS_KEY_ATTR_MPLS:
792		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
793								    __be32 *));
794		break;
795	}
796
797	return err;
798}
799
800static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
801			  struct sw_flow_key *key,
802			  const struct nlattr *a, int rem)
803{
804	struct deferred_action *da;
805
806	if (!is_flow_key_valid(key)) {
807		int err;
808
809		err = ovs_flow_key_update(skb, key);
810		if (err)
811			return err;
812	}
813	BUG_ON(!is_flow_key_valid(key));
814
815	if (!nla_is_last(a, rem)) {
816		/* Recirc action is the not the last action
817		 * of the action list, need to clone the skb.
818		 */
819		skb = skb_clone(skb, GFP_ATOMIC);
820
821		/* Skip the recirc action when out of memory, but
822		 * continue on with the rest of the action list.
823		 */
824		if (!skb)
825			return 0;
826	}
827
828	da = add_deferred_actions(skb, key, NULL);
829	if (da) {
830		da->pkt_key.recirc_id = nla_get_u32(a);
831	} else {
832		kfree_skb(skb);
833
834		if (net_ratelimit())
835			pr_warn("%s: deferred action limit reached, drop recirc action\n",
836				ovs_dp_name(dp));
837	}
838
839	return 0;
840}
841
842/* Execute a list of actions against 'skb'. */
843static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
844			      struct sw_flow_key *key,
845			      const struct nlattr *attr, int len)
846{
847	/* Every output action needs a separate clone of 'skb', but the common
848	 * case is just a single output action, so that doing a clone and
849	 * then freeing the original skbuff is wasteful.  So the following code
850	 * is slightly obscure just to avoid that.
851	 */
852	int prev_port = -1;
853	const struct nlattr *a;
854	int rem;
855
856	for (a = attr, rem = len; rem > 0;
857	     a = nla_next(a, &rem)) {
858		int err = 0;
859
860		if (unlikely(prev_port != -1)) {
861			struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
862
863			if (out_skb)
864				do_output(dp, out_skb, prev_port);
865
866			prev_port = -1;
867		}
868
869		switch (nla_type(a)) {
870		case OVS_ACTION_ATTR_OUTPUT:
871			prev_port = nla_get_u32(a);
872			break;
873
874		case OVS_ACTION_ATTR_USERSPACE:
875			output_userspace(dp, skb, key, a);
876			break;
877
878		case OVS_ACTION_ATTR_HASH:
879			execute_hash(skb, key, a);
880			break;
881
882		case OVS_ACTION_ATTR_PUSH_MPLS:
883			err = push_mpls(skb, key, nla_data(a));
884			break;
885
886		case OVS_ACTION_ATTR_POP_MPLS:
887			err = pop_mpls(skb, key, nla_get_be16(a));
888			break;
889
890		case OVS_ACTION_ATTR_PUSH_VLAN:
891			err = push_vlan(skb, key, nla_data(a));
892			break;
893
894		case OVS_ACTION_ATTR_POP_VLAN:
895			err = pop_vlan(skb, key);
896			break;
897
898		case OVS_ACTION_ATTR_RECIRC:
899			err = execute_recirc(dp, skb, key, a, rem);
900			if (nla_is_last(a, rem)) {
901				/* If this is the last action, the skb has
902				 * been consumed or freed.
903				 * Return immediately.
904				 */
905				return err;
906			}
907			break;
908
909		case OVS_ACTION_ATTR_SET:
910			err = execute_set_action(skb, key, nla_data(a));
911			break;
912
913		case OVS_ACTION_ATTR_SET_MASKED:
914		case OVS_ACTION_ATTR_SET_TO_MASKED:
915			err = execute_masked_set_action(skb, key, nla_data(a));
916			break;
917
918		case OVS_ACTION_ATTR_SAMPLE:
919			err = sample(dp, skb, key, a);
920			break;
921		}
922
923		if (unlikely(err)) {
924			kfree_skb(skb);
925			return err;
926		}
927	}
928
929	if (prev_port != -1)
930		do_output(dp, skb, prev_port);
931	else
932		consume_skb(skb);
933
934	return 0;
935}
936
937static void process_deferred_actions(struct datapath *dp)
938{
939	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
940
941	/* Do not touch the FIFO in case there is no deferred actions. */
942	if (action_fifo_is_empty(fifo))
943		return;
944
945	/* Finishing executing all deferred actions. */
946	do {
947		struct deferred_action *da = action_fifo_get(fifo);
948		struct sk_buff *skb = da->skb;
949		struct sw_flow_key *key = &da->pkt_key;
950		const struct nlattr *actions = da->actions;
951
952		if (actions)
953			do_execute_actions(dp, skb, key, actions,
954					   nla_len(actions));
955		else
956			ovs_dp_process_packet(skb, key);
957	} while (!action_fifo_is_empty(fifo));
958
959	/* Reset FIFO for the next packet.  */
960	action_fifo_init(fifo);
961}
962
963/* Execute a list of actions against 'skb'. */
964int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
965			const struct sw_flow_actions *acts,
966			struct sw_flow_key *key)
967{
968	int level = this_cpu_read(exec_actions_level);
969	int err;
970
971	this_cpu_inc(exec_actions_level);
972	OVS_CB(skb)->egress_tun_info = NULL;
973	err = do_execute_actions(dp, skb, key,
974				 acts->actions, acts->actions_len);
975
976	if (!level)
977		process_deferred_actions(dp);
978
979	this_cpu_dec(exec_actions_level);
980	return err;
981}
982
983int action_fifos_init(void)
984{
985	action_fifos = alloc_percpu(struct action_fifo);
986	if (!action_fifos)
987		return -ENOMEM;
988
989	return 0;
990}
991
992void action_fifos_exit(void)
993{
994	free_percpu(action_fifos);
995}
996