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/netfilter_ipv6.h>
26#include <linux/sctp.h>
27#include <linux/tcp.h>
28#include <linux/udp.h>
29#include <linux/in6.h>
30#include <linux/if_arp.h>
31#include <linux/if_vlan.h>
32
33#include <net/dst.h>
34#include <net/ip.h>
35#include <net/ipv6.h>
36#include <net/ip6_fib.h>
37#include <net/checksum.h>
38#include <net/dsfield.h>
39#include <net/mpls.h>
40#include <net/sctp/checksum.h>
41
42#include "datapath.h"
43#include "flow.h"
44#include "conntrack.h"
45#include "vport.h"
46
47static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
48			      struct sw_flow_key *key,
49			      const struct nlattr *attr, int len);
50
51struct deferred_action {
52	struct sk_buff *skb;
53	const struct nlattr *actions;
54
55	/* Store pkt_key clone when creating deferred action. */
56	struct sw_flow_key pkt_key;
57};
58
59#define MAX_L2_LEN	(VLAN_ETH_HLEN + 3 * MPLS_HLEN)
60struct ovs_frag_data {
61	unsigned long dst;
62	struct vport *vport;
63	struct ovs_skb_cb cb;
64	__be16 inner_protocol;
65	__u16 vlan_tci;
66	__be16 vlan_proto;
67	unsigned int l2_len;
68	u8 l2_data[MAX_L2_LEN];
69};
70
71static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
72
73#define DEFERRED_ACTION_FIFO_SIZE 10
74struct action_fifo {
75	int head;
76	int tail;
77	/* Deferred action fifo queue storage. */
78	struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
79};
80
81static struct action_fifo __percpu *action_fifos;
82static DEFINE_PER_CPU(int, exec_actions_level);
83
84static void action_fifo_init(struct action_fifo *fifo)
85{
86	fifo->head = 0;
87	fifo->tail = 0;
88}
89
90static bool action_fifo_is_empty(const struct action_fifo *fifo)
91{
92	return (fifo->head == fifo->tail);
93}
94
95static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
96{
97	if (action_fifo_is_empty(fifo))
98		return NULL;
99
100	return &fifo->fifo[fifo->tail++];
101}
102
103static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
104{
105	if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
106		return NULL;
107
108	return &fifo->fifo[fifo->head++];
109}
110
111/* Return true if fifo is not full */
112static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
113						    const struct sw_flow_key *key,
114						    const struct nlattr *attr)
115{
116	struct action_fifo *fifo;
117	struct deferred_action *da;
118
119	fifo = this_cpu_ptr(action_fifos);
120	da = action_fifo_put(fifo);
121	if (da) {
122		da->skb = skb;
123		da->actions = attr;
124		da->pkt_key = *key;
125	}
126
127	return da;
128}
129
130static void invalidate_flow_key(struct sw_flow_key *key)
131{
132	key->eth.type = htons(0);
133}
134
135static bool is_flow_key_valid(const struct sw_flow_key *key)
136{
137	return !!key->eth.type;
138}
139
140static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
141		     const struct ovs_action_push_mpls *mpls)
142{
143	__be32 *new_mpls_lse;
144	struct ethhdr *hdr;
145
146	/* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
147	if (skb->encapsulation)
148		return -ENOTSUPP;
149
150	if (skb_cow_head(skb, MPLS_HLEN) < 0)
151		return -ENOMEM;
152
153	skb_push(skb, MPLS_HLEN);
154	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
155		skb->mac_len);
156	skb_reset_mac_header(skb);
157
158	new_mpls_lse = (__be32 *)skb_mpls_header(skb);
159	*new_mpls_lse = mpls->mpls_lse;
160
161	skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
162
163	hdr = eth_hdr(skb);
164	hdr->h_proto = mpls->mpls_ethertype;
165
166	if (!skb->inner_protocol)
167		skb_set_inner_protocol(skb, skb->protocol);
168	skb->protocol = mpls->mpls_ethertype;
169
170	invalidate_flow_key(key);
171	return 0;
172}
173
174static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
175		    const __be16 ethertype)
176{
177	struct ethhdr *hdr;
178	int err;
179
180	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
181	if (unlikely(err))
182		return err;
183
184	skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
185
186	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
187		skb->mac_len);
188
189	__skb_pull(skb, MPLS_HLEN);
190	skb_reset_mac_header(skb);
191
192	/* skb_mpls_header() is used to locate the ethertype
193	 * field correctly in the presence of VLAN tags.
194	 */
195	hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
196	hdr->h_proto = ethertype;
197	if (eth_p_mpls(skb->protocol))
198		skb->protocol = ethertype;
199
200	invalidate_flow_key(key);
201	return 0;
202}
203
204static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
205		    const __be32 *mpls_lse, const __be32 *mask)
206{
207	__be32 *stack;
208	__be32 lse;
209	int err;
210
211	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
212	if (unlikely(err))
213		return err;
214
215	stack = (__be32 *)skb_mpls_header(skb);
216	lse = OVS_MASKED(*stack, *mpls_lse, *mask);
217	if (skb->ip_summed == CHECKSUM_COMPLETE) {
218		__be32 diff[] = { ~(*stack), lse };
219
220		skb->csum = ~csum_partial((char *)diff, sizeof(diff),
221					  ~skb->csum);
222	}
223
224	*stack = lse;
225	flow_key->mpls.top_lse = lse;
226	return 0;
227}
228
229static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
230{
231	int err;
232
233	err = skb_vlan_pop(skb);
234	if (skb_vlan_tag_present(skb))
235		invalidate_flow_key(key);
236	else
237		key->eth.tci = 0;
238	return err;
239}
240
241static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
242		     const struct ovs_action_push_vlan *vlan)
243{
244	if (skb_vlan_tag_present(skb))
245		invalidate_flow_key(key);
246	else
247		key->eth.tci = vlan->vlan_tci;
248	return skb_vlan_push(skb, vlan->vlan_tpid,
249			     ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
250}
251
252/* 'src' is already properly masked. */
253static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
254{
255	u16 *dst = (u16 *)dst_;
256	const u16 *src = (const u16 *)src_;
257	const u16 *mask = (const u16 *)mask_;
258
259	OVS_SET_MASKED(dst[0], src[0], mask[0]);
260	OVS_SET_MASKED(dst[1], src[1], mask[1]);
261	OVS_SET_MASKED(dst[2], src[2], mask[2]);
262}
263
264static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
265			const struct ovs_key_ethernet *key,
266			const struct ovs_key_ethernet *mask)
267{
268	int err;
269
270	err = skb_ensure_writable(skb, ETH_HLEN);
271	if (unlikely(err))
272		return err;
273
274	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
275
276	ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
277			       mask->eth_src);
278	ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
279			       mask->eth_dst);
280
281	skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
282
283	ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
284	ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
285	return 0;
286}
287
288static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
289				  __be32 addr, __be32 new_addr)
290{
291	int transport_len = skb->len - skb_transport_offset(skb);
292
293	if (nh->frag_off & htons(IP_OFFSET))
294		return;
295
296	if (nh->protocol == IPPROTO_TCP) {
297		if (likely(transport_len >= sizeof(struct tcphdr)))
298			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
299						 addr, new_addr, true);
300	} else if (nh->protocol == IPPROTO_UDP) {
301		if (likely(transport_len >= sizeof(struct udphdr))) {
302			struct udphdr *uh = udp_hdr(skb);
303
304			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
305				inet_proto_csum_replace4(&uh->check, skb,
306							 addr, new_addr, true);
307				if (!uh->check)
308					uh->check = CSUM_MANGLED_0;
309			}
310		}
311	}
312}
313
314static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
315			__be32 *addr, __be32 new_addr)
316{
317	update_ip_l4_checksum(skb, nh, *addr, new_addr);
318	csum_replace4(&nh->check, *addr, new_addr);
319	skb_clear_hash(skb);
320	*addr = new_addr;
321}
322
323static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
324				 __be32 addr[4], const __be32 new_addr[4])
325{
326	int transport_len = skb->len - skb_transport_offset(skb);
327
328	if (l4_proto == NEXTHDR_TCP) {
329		if (likely(transport_len >= sizeof(struct tcphdr)))
330			inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
331						  addr, new_addr, true);
332	} else if (l4_proto == NEXTHDR_UDP) {
333		if (likely(transport_len >= sizeof(struct udphdr))) {
334			struct udphdr *uh = udp_hdr(skb);
335
336			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
337				inet_proto_csum_replace16(&uh->check, skb,
338							  addr, new_addr, true);
339				if (!uh->check)
340					uh->check = CSUM_MANGLED_0;
341			}
342		}
343	} else if (l4_proto == NEXTHDR_ICMP) {
344		if (likely(transport_len >= sizeof(struct icmp6hdr)))
345			inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
346						  skb, addr, new_addr, true);
347	}
348}
349
350static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
351			   const __be32 mask[4], __be32 masked[4])
352{
353	masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
354	masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
355	masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
356	masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
357}
358
359static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
360			  __be32 addr[4], const __be32 new_addr[4],
361			  bool recalculate_csum)
362{
363	if (recalculate_csum)
364		update_ipv6_checksum(skb, l4_proto, addr, new_addr);
365
366	skb_clear_hash(skb);
367	memcpy(addr, new_addr, sizeof(__be32[4]));
368}
369
370static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
371{
372	/* Bits 21-24 are always unmasked, so this retains their values. */
373	OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
374	OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
375	OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
376}
377
378static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
379		       u8 mask)
380{
381	new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
382
383	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
384	nh->ttl = new_ttl;
385}
386
387static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
388		    const struct ovs_key_ipv4 *key,
389		    const struct ovs_key_ipv4 *mask)
390{
391	struct iphdr *nh;
392	__be32 new_addr;
393	int err;
394
395	err = skb_ensure_writable(skb, skb_network_offset(skb) +
396				  sizeof(struct iphdr));
397	if (unlikely(err))
398		return err;
399
400	nh = ip_hdr(skb);
401
402	/* Setting an IP addresses is typically only a side effect of
403	 * matching on them in the current userspace implementation, so it
404	 * makes sense to check if the value actually changed.
405	 */
406	if (mask->ipv4_src) {
407		new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
408
409		if (unlikely(new_addr != nh->saddr)) {
410			set_ip_addr(skb, nh, &nh->saddr, new_addr);
411			flow_key->ipv4.addr.src = new_addr;
412		}
413	}
414	if (mask->ipv4_dst) {
415		new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
416
417		if (unlikely(new_addr != nh->daddr)) {
418			set_ip_addr(skb, nh, &nh->daddr, new_addr);
419			flow_key->ipv4.addr.dst = new_addr;
420		}
421	}
422	if (mask->ipv4_tos) {
423		ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
424		flow_key->ip.tos = nh->tos;
425	}
426	if (mask->ipv4_ttl) {
427		set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
428		flow_key->ip.ttl = nh->ttl;
429	}
430
431	return 0;
432}
433
434static bool is_ipv6_mask_nonzero(const __be32 addr[4])
435{
436	return !!(addr[0] | addr[1] | addr[2] | addr[3]);
437}
438
439static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
440		    const struct ovs_key_ipv6 *key,
441		    const struct ovs_key_ipv6 *mask)
442{
443	struct ipv6hdr *nh;
444	int err;
445
446	err = skb_ensure_writable(skb, skb_network_offset(skb) +
447				  sizeof(struct ipv6hdr));
448	if (unlikely(err))
449		return err;
450
451	nh = ipv6_hdr(skb);
452
453	/* Setting an IP addresses is typically only a side effect of
454	 * matching on them in the current userspace implementation, so it
455	 * makes sense to check if the value actually changed.
456	 */
457	if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
458		__be32 *saddr = (__be32 *)&nh->saddr;
459		__be32 masked[4];
460
461		mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
462
463		if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
464			set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
465				      true);
466			memcpy(&flow_key->ipv6.addr.src, masked,
467			       sizeof(flow_key->ipv6.addr.src));
468		}
469	}
470	if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
471		unsigned int offset = 0;
472		int flags = IP6_FH_F_SKIP_RH;
473		bool recalc_csum = true;
474		__be32 *daddr = (__be32 *)&nh->daddr;
475		__be32 masked[4];
476
477		mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
478
479		if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
480			if (ipv6_ext_hdr(nh->nexthdr))
481				recalc_csum = (ipv6_find_hdr(skb, &offset,
482							     NEXTHDR_ROUTING,
483							     NULL, &flags)
484					       != NEXTHDR_ROUTING);
485
486			set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
487				      recalc_csum);
488			memcpy(&flow_key->ipv6.addr.dst, masked,
489			       sizeof(flow_key->ipv6.addr.dst));
490		}
491	}
492	if (mask->ipv6_tclass) {
493		ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
494		flow_key->ip.tos = ipv6_get_dsfield(nh);
495	}
496	if (mask->ipv6_label) {
497		set_ipv6_fl(nh, ntohl(key->ipv6_label),
498			    ntohl(mask->ipv6_label));
499		flow_key->ipv6.label =
500		    *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
501	}
502	if (mask->ipv6_hlimit) {
503		OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
504			       mask->ipv6_hlimit);
505		flow_key->ip.ttl = nh->hop_limit;
506	}
507	return 0;
508}
509
510/* Must follow skb_ensure_writable() since that can move the skb data. */
511static void set_tp_port(struct sk_buff *skb, __be16 *port,
512			__be16 new_port, __sum16 *check)
513{
514	inet_proto_csum_replace2(check, skb, *port, new_port, false);
515	*port = new_port;
516}
517
518static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
519		   const struct ovs_key_udp *key,
520		   const struct ovs_key_udp *mask)
521{
522	struct udphdr *uh;
523	__be16 src, dst;
524	int err;
525
526	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
527				  sizeof(struct udphdr));
528	if (unlikely(err))
529		return err;
530
531	uh = udp_hdr(skb);
532	/* Either of the masks is non-zero, so do not bother checking them. */
533	src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
534	dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
535
536	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
537		if (likely(src != uh->source)) {
538			set_tp_port(skb, &uh->source, src, &uh->check);
539			flow_key->tp.src = src;
540		}
541		if (likely(dst != uh->dest)) {
542			set_tp_port(skb, &uh->dest, dst, &uh->check);
543			flow_key->tp.dst = dst;
544		}
545
546		if (unlikely(!uh->check))
547			uh->check = CSUM_MANGLED_0;
548	} else {
549		uh->source = src;
550		uh->dest = dst;
551		flow_key->tp.src = src;
552		flow_key->tp.dst = dst;
553	}
554
555	skb_clear_hash(skb);
556
557	return 0;
558}
559
560static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
561		   const struct ovs_key_tcp *key,
562		   const struct ovs_key_tcp *mask)
563{
564	struct tcphdr *th;
565	__be16 src, dst;
566	int err;
567
568	err = skb_ensure_writable(skb, skb_transport_offset(skb) +
569				  sizeof(struct tcphdr));
570	if (unlikely(err))
571		return err;
572
573	th = tcp_hdr(skb);
574	src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
575	if (likely(src != th->source)) {
576		set_tp_port(skb, &th->source, src, &th->check);
577		flow_key->tp.src = src;
578	}
579	dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
580	if (likely(dst != th->dest)) {
581		set_tp_port(skb, &th->dest, dst, &th->check);
582		flow_key->tp.dst = dst;
583	}
584	skb_clear_hash(skb);
585
586	return 0;
587}
588
589static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
590		    const struct ovs_key_sctp *key,
591		    const struct ovs_key_sctp *mask)
592{
593	unsigned int sctphoff = skb_transport_offset(skb);
594	struct sctphdr *sh;
595	__le32 old_correct_csum, new_csum, old_csum;
596	int err;
597
598	err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
599	if (unlikely(err))
600		return err;
601
602	sh = sctp_hdr(skb);
603	old_csum = sh->checksum;
604	old_correct_csum = sctp_compute_cksum(skb, sctphoff);
605
606	sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
607	sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
608
609	new_csum = sctp_compute_cksum(skb, sctphoff);
610
611	/* Carry any checksum errors through. */
612	sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
613
614	skb_clear_hash(skb);
615	flow_key->tp.src = sh->source;
616	flow_key->tp.dst = sh->dest;
617
618	return 0;
619}
620
621static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
622{
623	struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
624	struct vport *vport = data->vport;
625
626	if (skb_cow_head(skb, data->l2_len) < 0) {
627		kfree_skb(skb);
628		return -ENOMEM;
629	}
630
631	__skb_dst_copy(skb, data->dst);
632	*OVS_CB(skb) = data->cb;
633	skb->inner_protocol = data->inner_protocol;
634	skb->vlan_tci = data->vlan_tci;
635	skb->vlan_proto = data->vlan_proto;
636
637	/* Reconstruct the MAC header.  */
638	skb_push(skb, data->l2_len);
639	memcpy(skb->data, &data->l2_data, data->l2_len);
640	skb_postpush_rcsum(skb, skb->data, data->l2_len);
641	skb_reset_mac_header(skb);
642
643	ovs_vport_send(vport, skb);
644	return 0;
645}
646
647static unsigned int
648ovs_dst_get_mtu(const struct dst_entry *dst)
649{
650	return dst->dev->mtu;
651}
652
653static struct dst_ops ovs_dst_ops = {
654	.family = AF_UNSPEC,
655	.mtu = ovs_dst_get_mtu,
656};
657
658/* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
659 * ovs_vport_output(), which is called once per fragmented packet.
660 */
661static void prepare_frag(struct vport *vport, struct sk_buff *skb)
662{
663	unsigned int hlen = skb_network_offset(skb);
664	struct ovs_frag_data *data;
665
666	data = this_cpu_ptr(&ovs_frag_data_storage);
667	data->dst = skb->_skb_refdst;
668	data->vport = vport;
669	data->cb = *OVS_CB(skb);
670	data->inner_protocol = skb->inner_protocol;
671	data->vlan_tci = skb->vlan_tci;
672	data->vlan_proto = skb->vlan_proto;
673	data->l2_len = hlen;
674	memcpy(&data->l2_data, skb->data, hlen);
675
676	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
677	skb_pull(skb, hlen);
678}
679
680static void ovs_fragment(struct net *net, struct vport *vport,
681			 struct sk_buff *skb, u16 mru, __be16 ethertype)
682{
683	if (skb_network_offset(skb) > MAX_L2_LEN) {
684		OVS_NLERR(1, "L2 header too long to fragment");
685		goto err;
686	}
687
688	if (ethertype == htons(ETH_P_IP)) {
689		struct dst_entry ovs_dst;
690		unsigned long orig_dst;
691
692		prepare_frag(vport, skb);
693		dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
694			 DST_OBSOLETE_NONE, DST_NOCOUNT);
695		ovs_dst.dev = vport->dev;
696
697		orig_dst = skb->_skb_refdst;
698		skb_dst_set_noref(skb, &ovs_dst);
699		IPCB(skb)->frag_max_size = mru;
700
701		ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
702		refdst_drop(orig_dst);
703	} else if (ethertype == htons(ETH_P_IPV6)) {
704		const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
705		unsigned long orig_dst;
706		struct rt6_info ovs_rt;
707
708		if (!v6ops) {
709			goto err;
710		}
711
712		prepare_frag(vport, skb);
713		memset(&ovs_rt, 0, sizeof(ovs_rt));
714		dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
715			 DST_OBSOLETE_NONE, DST_NOCOUNT);
716		ovs_rt.dst.dev = vport->dev;
717
718		orig_dst = skb->_skb_refdst;
719		skb_dst_set_noref(skb, &ovs_rt.dst);
720		IP6CB(skb)->frag_max_size = mru;
721
722		v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
723		refdst_drop(orig_dst);
724	} else {
725		WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
726			  ovs_vport_name(vport), ntohs(ethertype), mru,
727			  vport->dev->mtu);
728		goto err;
729	}
730
731	return;
732err:
733	kfree_skb(skb);
734}
735
736static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
737		      struct sw_flow_key *key)
738{
739	struct vport *vport = ovs_vport_rcu(dp, out_port);
740
741	if (likely(vport)) {
742		u16 mru = OVS_CB(skb)->mru;
743
744		if (likely(!mru || (skb->len <= mru + ETH_HLEN))) {
745			ovs_vport_send(vport, skb);
746		} else if (mru <= vport->dev->mtu) {
747			struct net *net = read_pnet(&dp->net);
748			__be16 ethertype = key->eth.type;
749
750			if (!is_flow_key_valid(key)) {
751				if (eth_p_mpls(skb->protocol))
752					ethertype = skb->inner_protocol;
753				else
754					ethertype = vlan_get_protocol(skb);
755			}
756
757			ovs_fragment(net, vport, skb, mru, ethertype);
758		} else {
759			kfree_skb(skb);
760		}
761	} else {
762		kfree_skb(skb);
763	}
764}
765
766static int output_userspace(struct datapath *dp, struct sk_buff *skb,
767			    struct sw_flow_key *key, const struct nlattr *attr,
768			    const struct nlattr *actions, int actions_len)
769{
770	struct dp_upcall_info upcall;
771	const struct nlattr *a;
772	int rem;
773
774	memset(&upcall, 0, sizeof(upcall));
775	upcall.cmd = OVS_PACKET_CMD_ACTION;
776	upcall.mru = OVS_CB(skb)->mru;
777
778	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
779		 a = nla_next(a, &rem)) {
780		switch (nla_type(a)) {
781		case OVS_USERSPACE_ATTR_USERDATA:
782			upcall.userdata = a;
783			break;
784
785		case OVS_USERSPACE_ATTR_PID:
786			upcall.portid = nla_get_u32(a);
787			break;
788
789		case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
790			/* Get out tunnel info. */
791			struct vport *vport;
792
793			vport = ovs_vport_rcu(dp, nla_get_u32(a));
794			if (vport) {
795				int err;
796
797				err = dev_fill_metadata_dst(vport->dev, skb);
798				if (!err)
799					upcall.egress_tun_info = skb_tunnel_info(skb);
800			}
801
802			break;
803		}
804
805		case OVS_USERSPACE_ATTR_ACTIONS: {
806			/* Include actions. */
807			upcall.actions = actions;
808			upcall.actions_len = actions_len;
809			break;
810		}
811
812		} /* End of switch. */
813	}
814
815	return ovs_dp_upcall(dp, skb, key, &upcall);
816}
817
818static int sample(struct datapath *dp, struct sk_buff *skb,
819		  struct sw_flow_key *key, const struct nlattr *attr,
820		  const struct nlattr *actions, int actions_len)
821{
822	const struct nlattr *acts_list = NULL;
823	const struct nlattr *a;
824	int rem;
825
826	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
827		 a = nla_next(a, &rem)) {
828		u32 probability;
829
830		switch (nla_type(a)) {
831		case OVS_SAMPLE_ATTR_PROBABILITY:
832			probability = nla_get_u32(a);
833			if (!probability || prandom_u32() > probability)
834				return 0;
835			break;
836
837		case OVS_SAMPLE_ATTR_ACTIONS:
838			acts_list = a;
839			break;
840		}
841	}
842
843	rem = nla_len(acts_list);
844	a = nla_data(acts_list);
845
846	/* Actions list is empty, do nothing */
847	if (unlikely(!rem))
848		return 0;
849
850	/* The only known usage of sample action is having a single user-space
851	 * action. Treat this usage as a special case.
852	 * The output_userspace() should clone the skb to be sent to the
853	 * user space. This skb will be consumed by its caller.
854	 */
855	if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
856		   nla_is_last(a, rem)))
857		return output_userspace(dp, skb, key, a, actions, actions_len);
858
859	skb = skb_clone(skb, GFP_ATOMIC);
860	if (!skb)
861		/* Skip the sample action when out of memory. */
862		return 0;
863
864	if (!add_deferred_actions(skb, key, a)) {
865		if (net_ratelimit())
866			pr_warn("%s: deferred actions limit reached, dropping sample action\n",
867				ovs_dp_name(dp));
868
869		kfree_skb(skb);
870	}
871	return 0;
872}
873
874static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
875			 const struct nlattr *attr)
876{
877	struct ovs_action_hash *hash_act = nla_data(attr);
878	u32 hash = 0;
879
880	/* OVS_HASH_ALG_L4 is the only possible hash algorithm.  */
881	hash = skb_get_hash(skb);
882	hash = jhash_1word(hash, hash_act->hash_basis);
883	if (!hash)
884		hash = 0x1;
885
886	key->ovs_flow_hash = hash;
887}
888
889static int execute_set_action(struct sk_buff *skb,
890			      struct sw_flow_key *flow_key,
891			      const struct nlattr *a)
892{
893	/* Only tunnel set execution is supported without a mask. */
894	if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
895		struct ovs_tunnel_info *tun = nla_data(a);
896
897		skb_dst_drop(skb);
898		dst_hold((struct dst_entry *)tun->tun_dst);
899		skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
900		return 0;
901	}
902
903	return -EINVAL;
904}
905
906/* Mask is at the midpoint of the data. */
907#define get_mask(a, type) ((const type)nla_data(a) + 1)
908
909static int execute_masked_set_action(struct sk_buff *skb,
910				     struct sw_flow_key *flow_key,
911				     const struct nlattr *a)
912{
913	int err = 0;
914
915	switch (nla_type(a)) {
916	case OVS_KEY_ATTR_PRIORITY:
917		OVS_SET_MASKED(skb->priority, nla_get_u32(a),
918			       *get_mask(a, u32 *));
919		flow_key->phy.priority = skb->priority;
920		break;
921
922	case OVS_KEY_ATTR_SKB_MARK:
923		OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
924		flow_key->phy.skb_mark = skb->mark;
925		break;
926
927	case OVS_KEY_ATTR_TUNNEL_INFO:
928		/* Masked data not supported for tunnel. */
929		err = -EINVAL;
930		break;
931
932	case OVS_KEY_ATTR_ETHERNET:
933		err = set_eth_addr(skb, flow_key, nla_data(a),
934				   get_mask(a, struct ovs_key_ethernet *));
935		break;
936
937	case OVS_KEY_ATTR_IPV4:
938		err = set_ipv4(skb, flow_key, nla_data(a),
939			       get_mask(a, struct ovs_key_ipv4 *));
940		break;
941
942	case OVS_KEY_ATTR_IPV6:
943		err = set_ipv6(skb, flow_key, nla_data(a),
944			       get_mask(a, struct ovs_key_ipv6 *));
945		break;
946
947	case OVS_KEY_ATTR_TCP:
948		err = set_tcp(skb, flow_key, nla_data(a),
949			      get_mask(a, struct ovs_key_tcp *));
950		break;
951
952	case OVS_KEY_ATTR_UDP:
953		err = set_udp(skb, flow_key, nla_data(a),
954			      get_mask(a, struct ovs_key_udp *));
955		break;
956
957	case OVS_KEY_ATTR_SCTP:
958		err = set_sctp(skb, flow_key, nla_data(a),
959			       get_mask(a, struct ovs_key_sctp *));
960		break;
961
962	case OVS_KEY_ATTR_MPLS:
963		err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
964								    __be32 *));
965		break;
966
967	case OVS_KEY_ATTR_CT_STATE:
968	case OVS_KEY_ATTR_CT_ZONE:
969	case OVS_KEY_ATTR_CT_MARK:
970	case OVS_KEY_ATTR_CT_LABELS:
971		err = -EINVAL;
972		break;
973	}
974
975	return err;
976}
977
978static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
979			  struct sw_flow_key *key,
980			  const struct nlattr *a, int rem)
981{
982	struct deferred_action *da;
983
984	if (!is_flow_key_valid(key)) {
985		int err;
986
987		err = ovs_flow_key_update(skb, key);
988		if (err)
989			return err;
990	}
991	BUG_ON(!is_flow_key_valid(key));
992
993	if (!nla_is_last(a, rem)) {
994		/* Recirc action is the not the last action
995		 * of the action list, need to clone the skb.
996		 */
997		skb = skb_clone(skb, GFP_ATOMIC);
998
999		/* Skip the recirc action when out of memory, but
1000		 * continue on with the rest of the action list.
1001		 */
1002		if (!skb)
1003			return 0;
1004	}
1005
1006	da = add_deferred_actions(skb, key, NULL);
1007	if (da) {
1008		da->pkt_key.recirc_id = nla_get_u32(a);
1009	} else {
1010		kfree_skb(skb);
1011
1012		if (net_ratelimit())
1013			pr_warn("%s: deferred action limit reached, drop recirc action\n",
1014				ovs_dp_name(dp));
1015	}
1016
1017	return 0;
1018}
1019
1020/* Execute a list of actions against 'skb'. */
1021static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1022			      struct sw_flow_key *key,
1023			      const struct nlattr *attr, int len)
1024{
1025	/* Every output action needs a separate clone of 'skb', but the common
1026	 * case is just a single output action, so that doing a clone and
1027	 * then freeing the original skbuff is wasteful.  So the following code
1028	 * is slightly obscure just to avoid that.
1029	 */
1030	int prev_port = -1;
1031	const struct nlattr *a;
1032	int rem;
1033
1034	for (a = attr, rem = len; rem > 0;
1035	     a = nla_next(a, &rem)) {
1036		int err = 0;
1037
1038		if (unlikely(prev_port != -1)) {
1039			struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1040
1041			if (out_skb)
1042				do_output(dp, out_skb, prev_port, key);
1043
1044			prev_port = -1;
1045		}
1046
1047		switch (nla_type(a)) {
1048		case OVS_ACTION_ATTR_OUTPUT:
1049			prev_port = nla_get_u32(a);
1050			break;
1051
1052		case OVS_ACTION_ATTR_USERSPACE:
1053			output_userspace(dp, skb, key, a, attr, len);
1054			break;
1055
1056		case OVS_ACTION_ATTR_HASH:
1057			execute_hash(skb, key, a);
1058			break;
1059
1060		case OVS_ACTION_ATTR_PUSH_MPLS:
1061			err = push_mpls(skb, key, nla_data(a));
1062			break;
1063
1064		case OVS_ACTION_ATTR_POP_MPLS:
1065			err = pop_mpls(skb, key, nla_get_be16(a));
1066			break;
1067
1068		case OVS_ACTION_ATTR_PUSH_VLAN:
1069			err = push_vlan(skb, key, nla_data(a));
1070			break;
1071
1072		case OVS_ACTION_ATTR_POP_VLAN:
1073			err = pop_vlan(skb, key);
1074			break;
1075
1076		case OVS_ACTION_ATTR_RECIRC:
1077			err = execute_recirc(dp, skb, key, a, rem);
1078			if (nla_is_last(a, rem)) {
1079				/* If this is the last action, the skb has
1080				 * been consumed or freed.
1081				 * Return immediately.
1082				 */
1083				return err;
1084			}
1085			break;
1086
1087		case OVS_ACTION_ATTR_SET:
1088			err = execute_set_action(skb, key, nla_data(a));
1089			break;
1090
1091		case OVS_ACTION_ATTR_SET_MASKED:
1092		case OVS_ACTION_ATTR_SET_TO_MASKED:
1093			err = execute_masked_set_action(skb, key, nla_data(a));
1094			break;
1095
1096		case OVS_ACTION_ATTR_SAMPLE:
1097			err = sample(dp, skb, key, a, attr, len);
1098			break;
1099
1100		case OVS_ACTION_ATTR_CT:
1101			if (!is_flow_key_valid(key)) {
1102				err = ovs_flow_key_update(skb, key);
1103				if (err)
1104					return err;
1105			}
1106
1107			err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1108					     nla_data(a));
1109
1110			/* Hide stolen IP fragments from user space. */
1111			if (err)
1112				return err == -EINPROGRESS ? 0 : err;
1113			break;
1114		}
1115
1116		if (unlikely(err)) {
1117			kfree_skb(skb);
1118			return err;
1119		}
1120	}
1121
1122	if (prev_port != -1)
1123		do_output(dp, skb, prev_port, key);
1124	else
1125		consume_skb(skb);
1126
1127	return 0;
1128}
1129
1130static void process_deferred_actions(struct datapath *dp)
1131{
1132	struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1133
1134	/* Do not touch the FIFO in case there is no deferred actions. */
1135	if (action_fifo_is_empty(fifo))
1136		return;
1137
1138	/* Finishing executing all deferred actions. */
1139	do {
1140		struct deferred_action *da = action_fifo_get(fifo);
1141		struct sk_buff *skb = da->skb;
1142		struct sw_flow_key *key = &da->pkt_key;
1143		const struct nlattr *actions = da->actions;
1144
1145		if (actions)
1146			do_execute_actions(dp, skb, key, actions,
1147					   nla_len(actions));
1148		else
1149			ovs_dp_process_packet(skb, key);
1150	} while (!action_fifo_is_empty(fifo));
1151
1152	/* Reset FIFO for the next packet.  */
1153	action_fifo_init(fifo);
1154}
1155
1156/* Execute a list of actions against 'skb'. */
1157int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1158			const struct sw_flow_actions *acts,
1159			struct sw_flow_key *key)
1160{
1161	int level = this_cpu_read(exec_actions_level);
1162	int err;
1163
1164	this_cpu_inc(exec_actions_level);
1165	err = do_execute_actions(dp, skb, key,
1166				 acts->actions, acts->actions_len);
1167
1168	if (!level)
1169		process_deferred_actions(dp);
1170
1171	this_cpu_dec(exec_actions_level);
1172	return err;
1173}
1174
1175int action_fifos_init(void)
1176{
1177	action_fifos = alloc_percpu(struct action_fifo);
1178	if (!action_fifos)
1179		return -ENOMEM;
1180
1181	return 0;
1182}
1183
1184void action_fifos_exit(void)
1185{
1186	free_percpu(action_fifos);
1187}
1188