2 * Copyright (c) 2007-2015 Nicira, Inc.
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.
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.
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
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/skbuff.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>
36 #include <net/checksum.h>
37 #include <net/dsfield.h>
39 #include <net/sctp/checksum.h>
42 #include "conntrack.h"
46 static int do_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
47 struct sw_flow_key
*key
,
48 const struct nlattr
*attr
, int len
);
50 struct deferred_action
{
52 const struct nlattr
*actions
;
54 /* Store pkt_key clone when creating deferred action. */
55 struct sw_flow_key pkt_key
;
58 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
59 struct ovs_frag_data
{
63 __be16 inner_protocol
;
67 u8 l2_data
[MAX_L2_LEN
];
70 static DEFINE_PER_CPU(struct ovs_frag_data
, ovs_frag_data_storage
);
72 #define DEFERRED_ACTION_FIFO_SIZE 10
73 #define OVS_RECURSION_LIMIT 4
74 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
78 /* Deferred action fifo queue storage. */
79 struct deferred_action fifo
[DEFERRED_ACTION_FIFO_SIZE
];
83 struct sw_flow_key key
[OVS_DEFERRED_ACTION_THRESHOLD
];
86 static struct action_fifo __percpu
*action_fifos
;
87 static struct recirc_keys __percpu
*recirc_keys
;
88 static DEFINE_PER_CPU(int, exec_actions_level
);
90 static void action_fifo_init(struct action_fifo
*fifo
)
96 static bool action_fifo_is_empty(const struct action_fifo
*fifo
)
98 return (fifo
->head
== fifo
->tail
);
101 static struct deferred_action
*action_fifo_get(struct action_fifo
*fifo
)
103 if (action_fifo_is_empty(fifo
))
106 return &fifo
->fifo
[fifo
->tail
++];
109 static struct deferred_action
*action_fifo_put(struct action_fifo
*fifo
)
111 if (fifo
->head
>= DEFERRED_ACTION_FIFO_SIZE
- 1)
114 return &fifo
->fifo
[fifo
->head
++];
117 /* Return queue entry if fifo is not full */
118 static struct deferred_action
*add_deferred_actions(struct sk_buff
*skb
,
119 const struct sw_flow_key
*key
,
120 const struct nlattr
*attr
)
122 struct action_fifo
*fifo
;
123 struct deferred_action
*da
;
125 fifo
= this_cpu_ptr(action_fifos
);
126 da
= action_fifo_put(fifo
);
136 static void invalidate_flow_key(struct sw_flow_key
*key
)
138 key
->eth
.type
= htons(0);
141 static bool is_flow_key_valid(const struct sw_flow_key
*key
)
143 return !!key
->eth
.type
;
146 static void update_ethertype(struct sk_buff
*skb
, struct ethhdr
*hdr
,
149 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
150 __be16 diff
[] = { ~(hdr
->h_proto
), ethertype
};
152 skb
->csum
= ~csum_partial((char *)diff
, sizeof(diff
),
156 hdr
->h_proto
= ethertype
;
159 static int push_mpls(struct sk_buff
*skb
, struct sw_flow_key
*key
,
160 const struct ovs_action_push_mpls
*mpls
)
162 __be32
*new_mpls_lse
;
164 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
165 if (skb
->encapsulation
)
168 if (skb_cow_head(skb
, MPLS_HLEN
) < 0)
171 skb_push(skb
, MPLS_HLEN
);
172 memmove(skb_mac_header(skb
) - MPLS_HLEN
, skb_mac_header(skb
),
174 skb_reset_mac_header(skb
);
176 new_mpls_lse
= (__be32
*)skb_mpls_header(skb
);
177 *new_mpls_lse
= mpls
->mpls_lse
;
179 skb_postpush_rcsum(skb
, new_mpls_lse
, MPLS_HLEN
);
181 update_ethertype(skb
, eth_hdr(skb
), mpls
->mpls_ethertype
);
182 if (!ovs_skb_get_inner_protocol(skb
))
183 ovs_skb_set_inner_protocol(skb
, skb
->protocol
);
184 skb
->protocol
= mpls
->mpls_ethertype
;
186 invalidate_flow_key(key
);
190 static int pop_mpls(struct sk_buff
*skb
, struct sw_flow_key
*key
,
191 const __be16 ethertype
)
196 err
= skb_ensure_writable(skb
, skb
->mac_len
+ MPLS_HLEN
);
200 skb_postpull_rcsum(skb
, skb_mpls_header(skb
), MPLS_HLEN
);
202 memmove(skb_mac_header(skb
) + MPLS_HLEN
, skb_mac_header(skb
),
205 __skb_pull(skb
, MPLS_HLEN
);
206 skb_reset_mac_header(skb
);
208 /* skb_mpls_header() is used to locate the ethertype
209 * field correctly in the presence of VLAN tags.
211 hdr
= (struct ethhdr
*)(skb_mpls_header(skb
) - ETH_HLEN
);
212 update_ethertype(skb
, hdr
, ethertype
);
213 if (eth_p_mpls(skb
->protocol
))
214 skb
->protocol
= ethertype
;
216 invalidate_flow_key(key
);
220 static int set_mpls(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
221 const __be32
*mpls_lse
, const __be32
*mask
)
227 err
= skb_ensure_writable(skb
, skb
->mac_len
+ MPLS_HLEN
);
231 stack
= (__be32
*)skb_mpls_header(skb
);
232 lse
= OVS_MASKED(*stack
, *mpls_lse
, *mask
);
233 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
234 __be32 diff
[] = { ~(*stack
), lse
};
236 skb
->csum
= ~csum_partial((char *)diff
, sizeof(diff
),
241 flow_key
->mpls
.top_lse
= lse
;
245 static int pop_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
)
249 err
= skb_vlan_pop(skb
);
250 if (skb_vlan_tag_present(skb
))
251 invalidate_flow_key(key
);
257 static int push_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
,
258 const struct ovs_action_push_vlan
*vlan
)
260 if (skb_vlan_tag_present(skb
))
261 invalidate_flow_key(key
);
263 key
->eth
.tci
= vlan
->vlan_tci
;
264 return skb_vlan_push(skb
, vlan
->vlan_tpid
,
265 ntohs(vlan
->vlan_tci
) & ~VLAN_TAG_PRESENT
);
268 /* 'src' is already properly masked. */
269 static void ether_addr_copy_masked(u8
*dst_
, const u8
*src_
, const u8
*mask_
)
271 u16
*dst
= (u16
*)dst_
;
272 const u16
*src
= (const u16
*)src_
;
273 const u16
*mask
= (const u16
*)mask_
;
275 OVS_SET_MASKED(dst
[0], src
[0], mask
[0]);
276 OVS_SET_MASKED(dst
[1], src
[1], mask
[1]);
277 OVS_SET_MASKED(dst
[2], src
[2], mask
[2]);
280 static int set_eth_addr(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
281 const struct ovs_key_ethernet
*key
,
282 const struct ovs_key_ethernet
*mask
)
286 err
= skb_ensure_writable(skb
, ETH_HLEN
);
290 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_ALEN
* 2);
292 ether_addr_copy_masked(eth_hdr(skb
)->h_source
, key
->eth_src
,
294 ether_addr_copy_masked(eth_hdr(skb
)->h_dest
, key
->eth_dst
,
297 skb_postpush_rcsum(skb
, eth_hdr(skb
), ETH_ALEN
* 2);
299 ether_addr_copy(flow_key
->eth
.src
, eth_hdr(skb
)->h_source
);
300 ether_addr_copy(flow_key
->eth
.dst
, eth_hdr(skb
)->h_dest
);
304 static void update_ip_l4_checksum(struct sk_buff
*skb
, struct iphdr
*nh
,
305 __be32 addr
, __be32 new_addr
)
307 int transport_len
= skb
->len
- skb_transport_offset(skb
);
309 if (nh
->frag_off
& htons(IP_OFFSET
))
312 if (nh
->protocol
== IPPROTO_TCP
) {
313 if (likely(transport_len
>= sizeof(struct tcphdr
)))
314 inet_proto_csum_replace4(&tcp_hdr(skb
)->check
, skb
,
315 addr
, new_addr
, true);
316 } else if (nh
->protocol
== IPPROTO_UDP
) {
317 if (likely(transport_len
>= sizeof(struct udphdr
))) {
318 struct udphdr
*uh
= udp_hdr(skb
);
320 if (uh
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
321 inet_proto_csum_replace4(&uh
->check
, skb
,
322 addr
, new_addr
, true);
324 uh
->check
= CSUM_MANGLED_0
;
331 static void set_ip_addr(struct sk_buff
*skb
, struct iphdr
*nh
,
332 __be32
*addr
, __be32 new_addr
)
334 update_ip_l4_checksum(skb
, nh
, *addr
, new_addr
);
335 csum_replace4(&nh
->check
, *addr
, new_addr
);
340 static void update_ipv6_checksum(struct sk_buff
*skb
, u8 l4_proto
,
341 __be32 addr
[4], const __be32 new_addr
[4])
343 int transport_len
= skb
->len
- skb_transport_offset(skb
);
345 if (l4_proto
== NEXTHDR_TCP
) {
346 if (likely(transport_len
>= sizeof(struct tcphdr
)))
347 inet_proto_csum_replace16(&tcp_hdr(skb
)->check
, skb
,
348 addr
, new_addr
, true);
349 } else if (l4_proto
== NEXTHDR_UDP
) {
350 if (likely(transport_len
>= sizeof(struct udphdr
))) {
351 struct udphdr
*uh
= udp_hdr(skb
);
353 if (uh
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
354 inet_proto_csum_replace16(&uh
->check
, skb
,
355 addr
, new_addr
, true);
357 uh
->check
= CSUM_MANGLED_0
;
360 } else if (l4_proto
== NEXTHDR_ICMP
) {
361 if (likely(transport_len
>= sizeof(struct icmp6hdr
)))
362 inet_proto_csum_replace16(&icmp6_hdr(skb
)->icmp6_cksum
,
363 skb
, addr
, new_addr
, true);
367 static void mask_ipv6_addr(const __be32 old
[4], const __be32 addr
[4],
368 const __be32 mask
[4], __be32 masked
[4])
370 masked
[0] = OVS_MASKED(old
[0], addr
[0], mask
[0]);
371 masked
[1] = OVS_MASKED(old
[1], addr
[1], mask
[1]);
372 masked
[2] = OVS_MASKED(old
[2], addr
[2], mask
[2]);
373 masked
[3] = OVS_MASKED(old
[3], addr
[3], mask
[3]);
376 static void set_ipv6_addr(struct sk_buff
*skb
, u8 l4_proto
,
377 __be32 addr
[4], const __be32 new_addr
[4],
378 bool recalculate_csum
)
380 if (likely(recalculate_csum
))
381 update_ipv6_checksum(skb
, l4_proto
, addr
, new_addr
);
384 memcpy(addr
, new_addr
, sizeof(__be32
[4]));
387 static void set_ipv6_fl(struct ipv6hdr
*nh
, u32 fl
, u32 mask
)
389 /* Bits 21-24 are always unmasked, so this retains their values. */
390 OVS_SET_MASKED(nh
->flow_lbl
[0], (u8
)(fl
>> 16), (u8
)(mask
>> 16));
391 OVS_SET_MASKED(nh
->flow_lbl
[1], (u8
)(fl
>> 8), (u8
)(mask
>> 8));
392 OVS_SET_MASKED(nh
->flow_lbl
[2], (u8
)fl
, (u8
)mask
);
395 static void set_ip_ttl(struct sk_buff
*skb
, struct iphdr
*nh
, u8 new_ttl
,
398 new_ttl
= OVS_MASKED(nh
->ttl
, new_ttl
, mask
);
400 csum_replace2(&nh
->check
, htons(nh
->ttl
<< 8), htons(new_ttl
<< 8));
404 static int set_ipv4(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
405 const struct ovs_key_ipv4
*key
,
406 const struct ovs_key_ipv4
*mask
)
412 err
= skb_ensure_writable(skb
, skb_network_offset(skb
) +
413 sizeof(struct iphdr
));
419 /* Setting an IP addresses is typically only a side effect of
420 * matching on them in the current userspace implementation, so it
421 * makes sense to check if the value actually changed.
423 if (mask
->ipv4_src
) {
424 new_addr
= OVS_MASKED(nh
->saddr
, key
->ipv4_src
, mask
->ipv4_src
);
426 if (unlikely(new_addr
!= nh
->saddr
)) {
427 set_ip_addr(skb
, nh
, &nh
->saddr
, new_addr
);
428 flow_key
->ipv4
.addr
.src
= new_addr
;
431 if (mask
->ipv4_dst
) {
432 new_addr
= OVS_MASKED(nh
->daddr
, key
->ipv4_dst
, mask
->ipv4_dst
);
434 if (unlikely(new_addr
!= nh
->daddr
)) {
435 set_ip_addr(skb
, nh
, &nh
->daddr
, new_addr
);
436 flow_key
->ipv4
.addr
.dst
= new_addr
;
439 if (mask
->ipv4_tos
) {
440 ipv4_change_dsfield(nh
, ~mask
->ipv4_tos
, key
->ipv4_tos
);
441 flow_key
->ip
.tos
= nh
->tos
;
443 if (mask
->ipv4_ttl
) {
444 set_ip_ttl(skb
, nh
, key
->ipv4_ttl
, mask
->ipv4_ttl
);
445 flow_key
->ip
.ttl
= nh
->ttl
;
451 static bool is_ipv6_mask_nonzero(const __be32 addr
[4])
453 return !!(addr
[0] | addr
[1] | addr
[2] | addr
[3]);
456 static int set_ipv6(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
457 const struct ovs_key_ipv6
*key
,
458 const struct ovs_key_ipv6
*mask
)
463 err
= skb_ensure_writable(skb
, skb_network_offset(skb
) +
464 sizeof(struct ipv6hdr
));
470 /* Setting an IP addresses is typically only a side effect of
471 * matching on them in the current userspace implementation, so it
472 * makes sense to check if the value actually changed.
474 if (is_ipv6_mask_nonzero(mask
->ipv6_src
)) {
475 __be32
*saddr
= (__be32
*)&nh
->saddr
;
478 mask_ipv6_addr(saddr
, key
->ipv6_src
, mask
->ipv6_src
, masked
);
480 if (unlikely(memcmp(saddr
, masked
, sizeof(masked
)))) {
481 set_ipv6_addr(skb
, flow_key
->ip
.proto
, saddr
, masked
,
483 memcpy(&flow_key
->ipv6
.addr
.src
, masked
,
484 sizeof(flow_key
->ipv6
.addr
.src
));
487 if (is_ipv6_mask_nonzero(mask
->ipv6_dst
)) {
488 unsigned int offset
= 0;
489 int flags
= IP6_FH_F_SKIP_RH
;
490 bool recalc_csum
= true;
491 __be32
*daddr
= (__be32
*)&nh
->daddr
;
494 mask_ipv6_addr(daddr
, key
->ipv6_dst
, mask
->ipv6_dst
, masked
);
496 if (unlikely(memcmp(daddr
, masked
, sizeof(masked
)))) {
497 if (ipv6_ext_hdr(nh
->nexthdr
))
498 recalc_csum
= (ipv6_find_hdr(skb
, &offset
,
503 set_ipv6_addr(skb
, flow_key
->ip
.proto
, daddr
, masked
,
505 memcpy(&flow_key
->ipv6
.addr
.dst
, masked
,
506 sizeof(flow_key
->ipv6
.addr
.dst
));
509 if (mask
->ipv6_tclass
) {
510 ipv6_change_dsfield(nh
, ~mask
->ipv6_tclass
, key
->ipv6_tclass
);
511 flow_key
->ip
.tos
= ipv6_get_dsfield(nh
);
513 if (mask
->ipv6_label
) {
514 set_ipv6_fl(nh
, ntohl(key
->ipv6_label
),
515 ntohl(mask
->ipv6_label
));
516 flow_key
->ipv6
.label
=
517 *(__be32
*)nh
& htonl(IPV6_FLOWINFO_FLOWLABEL
);
519 if (mask
->ipv6_hlimit
) {
520 OVS_SET_MASKED(nh
->hop_limit
, key
->ipv6_hlimit
,
522 flow_key
->ip
.ttl
= nh
->hop_limit
;
527 /* Must follow skb_ensure_writable() since that can move the skb data. */
528 static void set_tp_port(struct sk_buff
*skb
, __be16
*port
,
529 __be16 new_port
, __sum16
*check
)
531 inet_proto_csum_replace2(check
, skb
, *port
, new_port
, false);
535 static int set_udp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
536 const struct ovs_key_udp
*key
,
537 const struct ovs_key_udp
*mask
)
543 err
= skb_ensure_writable(skb
, skb_transport_offset(skb
) +
544 sizeof(struct udphdr
));
549 /* Either of the masks is non-zero, so do not bother checking them. */
550 src
= OVS_MASKED(uh
->source
, key
->udp_src
, mask
->udp_src
);
551 dst
= OVS_MASKED(uh
->dest
, key
->udp_dst
, mask
->udp_dst
);
553 if (uh
->check
&& skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
554 if (likely(src
!= uh
->source
)) {
555 set_tp_port(skb
, &uh
->source
, src
, &uh
->check
);
556 flow_key
->tp
.src
= src
;
558 if (likely(dst
!= uh
->dest
)) {
559 set_tp_port(skb
, &uh
->dest
, dst
, &uh
->check
);
560 flow_key
->tp
.dst
= dst
;
563 if (unlikely(!uh
->check
))
564 uh
->check
= CSUM_MANGLED_0
;
568 flow_key
->tp
.src
= src
;
569 flow_key
->tp
.dst
= dst
;
577 static int set_tcp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
578 const struct ovs_key_tcp
*key
,
579 const struct ovs_key_tcp
*mask
)
585 err
= skb_ensure_writable(skb
, skb_transport_offset(skb
) +
586 sizeof(struct tcphdr
));
591 src
= OVS_MASKED(th
->source
, key
->tcp_src
, mask
->tcp_src
);
592 if (likely(src
!= th
->source
)) {
593 set_tp_port(skb
, &th
->source
, src
, &th
->check
);
594 flow_key
->tp
.src
= src
;
596 dst
= OVS_MASKED(th
->dest
, key
->tcp_dst
, mask
->tcp_dst
);
597 if (likely(dst
!= th
->dest
)) {
598 set_tp_port(skb
, &th
->dest
, dst
, &th
->check
);
599 flow_key
->tp
.dst
= dst
;
606 static int set_sctp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
607 const struct ovs_key_sctp
*key
,
608 const struct ovs_key_sctp
*mask
)
610 unsigned int sctphoff
= skb_transport_offset(skb
);
612 __le32 old_correct_csum
, new_csum
, old_csum
;
615 err
= skb_ensure_writable(skb
, sctphoff
+ sizeof(struct sctphdr
));
620 old_csum
= sh
->checksum
;
621 old_correct_csum
= sctp_compute_cksum(skb
, sctphoff
);
623 sh
->source
= OVS_MASKED(sh
->source
, key
->sctp_src
, mask
->sctp_src
);
624 sh
->dest
= OVS_MASKED(sh
->dest
, key
->sctp_dst
, mask
->sctp_dst
);
626 new_csum
= sctp_compute_cksum(skb
, sctphoff
);
628 /* Carry any checksum errors through. */
629 sh
->checksum
= old_csum
^ old_correct_csum
^ new_csum
;
632 flow_key
->tp
.src
= sh
->source
;
633 flow_key
->tp
.dst
= sh
->dest
;
638 static int ovs_vport_output(OVS_VPORT_OUTPUT_PARAMS
)
640 struct ovs_frag_data
*data
= this_cpu_ptr(&ovs_frag_data_storage
);
641 struct vport
*vport
= data
->vport
;
643 if (skb_cow_head(skb
, data
->l2_len
) < 0) {
648 __skb_dst_copy(skb
, data
->dst
);
649 *OVS_GSO_CB(skb
) = data
->cb
;
650 ovs_skb_set_inner_protocol(skb
, data
->inner_protocol
);
651 skb
->vlan_tci
= data
->vlan_tci
;
652 skb
->vlan_proto
= data
->vlan_proto
;
654 /* Reconstruct the MAC header. */
655 skb_push(skb
, data
->l2_len
);
656 memcpy(skb
->data
, &data
->l2_data
, data
->l2_len
);
657 skb_postpush_rcsum(skb
, skb
->data
, data
->l2_len
);
658 skb_reset_mac_header(skb
);
660 ovs_vport_send(vport
, skb
);
665 ovs_dst_get_mtu(const struct dst_entry
*dst
)
667 return dst
->dev
->mtu
;
670 static struct dst_ops ovs_dst_ops
= {
672 .mtu
= ovs_dst_get_mtu
,
675 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
676 * ovs_vport_output(), which is called once per fragmented packet.
678 static void prepare_frag(struct vport
*vport
, struct sk_buff
*skb
)
680 unsigned int hlen
= skb_network_offset(skb
);
681 struct ovs_frag_data
*data
;
683 data
= this_cpu_ptr(&ovs_frag_data_storage
);
684 data
->dst
= (unsigned long) skb_dst(skb
);
686 data
->cb
= *OVS_GSO_CB(skb
);
687 data
->inner_protocol
= ovs_skb_get_inner_protocol(skb
);
688 data
->vlan_tci
= skb
->vlan_tci
;
689 data
->vlan_proto
= skb
->vlan_proto
;
691 memcpy(&data
->l2_data
, skb
->data
, hlen
);
693 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
697 static void ovs_fragment(struct net
*net
, struct vport
*vport
,
698 struct sk_buff
*skb
, u16 mru
, __be16 ethertype
)
700 if (skb_network_offset(skb
) > MAX_L2_LEN
) {
701 OVS_NLERR(1, "L2 header too long to fragment");
705 if (ethertype
== htons(ETH_P_IP
)) {
706 struct dst_entry ovs_dst
;
707 unsigned long orig_dst
;
709 prepare_frag(vport
, skb
);
710 dst_init(&ovs_dst
, &ovs_dst_ops
, NULL
, 1,
711 DST_OBSOLETE_NONE
, DST_NOCOUNT
);
712 ovs_dst
.dev
= vport
->dev
;
714 orig_dst
= (unsigned long) skb_dst(skb
);
715 skb_dst_set_noref(skb
, &ovs_dst
);
716 IPCB(skb
)->frag_max_size
= mru
;
718 ip_do_fragment(net
, skb
->sk
, skb
, ovs_vport_output
);
719 refdst_drop(orig_dst
);
720 } else if (ethertype
== htons(ETH_P_IPV6
)) {
721 const struct nf_ipv6_ops
*v6ops
= nf_get_ipv6_ops();
722 unsigned long orig_dst
;
723 struct rt6_info ovs_rt
;
729 prepare_frag(vport
, skb
);
730 memset(&ovs_rt
, 0, sizeof(ovs_rt
));
731 dst_init(&ovs_rt
.dst
, &ovs_dst_ops
, NULL
, 1,
732 DST_OBSOLETE_NONE
, DST_NOCOUNT
);
733 ovs_rt
.dst
.dev
= vport
->dev
;
735 orig_dst
= (unsigned long) skb_dst(skb
);
736 skb_dst_set_noref(skb
, &ovs_rt
.dst
);
737 IP6CB(skb
)->frag_max_size
= mru
;
738 #ifdef HAVE_IP_LOCAL_OUT_TAKES_NET
739 v6ops
->fragment(net
, skb
->sk
, skb
, ovs_vport_output
);
741 v6ops
->fragment(skb
->sk
, skb
, ovs_vport_output
);
743 refdst_drop(orig_dst
);
745 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
746 ovs_vport_name(vport
), ntohs(ethertype
), mru
,
756 static void do_output(struct datapath
*dp
, struct sk_buff
*skb
, int out_port
,
757 struct sw_flow_key
*key
)
759 struct vport
*vport
= ovs_vport_rcu(dp
, out_port
);
762 u16 mru
= OVS_CB(skb
)->mru
;
763 u32 cutlen
= OVS_CB(skb
)->cutlen
;
765 if (unlikely(cutlen
> 0)) {
766 if (skb
->len
- cutlen
> ETH_HLEN
)
767 pskb_trim(skb
, skb
->len
- cutlen
);
769 pskb_trim(skb
, ETH_HLEN
);
772 if (likely(!mru
|| (skb
->len
<= mru
+ ETH_HLEN
))) {
773 ovs_vport_send(vport
, skb
);
774 } else if (mru
<= vport
->dev
->mtu
) {
775 struct net
*net
= ovs_dp_get_net(dp
);
776 __be16 ethertype
= key
->eth
.type
;
778 if (!is_flow_key_valid(key
)) {
779 if (eth_p_mpls(skb
->protocol
))
780 ethertype
= ovs_skb_get_inner_protocol(skb
);
782 ethertype
= vlan_get_protocol(skb
);
785 ovs_fragment(net
, vport
, skb
, mru
, ethertype
);
787 OVS_NLERR(true, "Cannot fragment IP frames");
795 static int output_userspace(struct datapath
*dp
, struct sk_buff
*skb
,
796 struct sw_flow_key
*key
, const struct nlattr
*attr
,
797 const struct nlattr
*actions
, int actions_len
,
800 struct dp_upcall_info upcall
;
801 const struct nlattr
*a
;
804 memset(&upcall
, 0, sizeof(upcall
));
805 upcall
.cmd
= OVS_PACKET_CMD_ACTION
;
806 upcall
.mru
= OVS_CB(skb
)->mru
;
808 SKB_INIT_FILL_METADATA_DST(skb
);
809 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
810 a
= nla_next(a
, &rem
)) {
811 switch (nla_type(a
)) {
812 case OVS_USERSPACE_ATTR_USERDATA
:
816 case OVS_USERSPACE_ATTR_PID
:
817 upcall
.portid
= nla_get_u32(a
);
820 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT
: {
821 /* Get out tunnel info. */
824 vport
= ovs_vport_rcu(dp
, nla_get_u32(a
));
826 err
= dev_fill_metadata_dst(vport
->dev
, skb
);
828 upcall
.egress_tun_info
= skb_tunnel_info(skb
);
834 case OVS_USERSPACE_ATTR_ACTIONS
: {
835 /* Include actions. */
836 upcall
.actions
= actions
;
837 upcall
.actions_len
= actions_len
;
841 } /* End of switch. */
844 err
= ovs_dp_upcall(dp
, skb
, key
, &upcall
, cutlen
);
845 SKB_RESTORE_FILL_METADATA_DST(skb
);
849 static int sample(struct datapath
*dp
, struct sk_buff
*skb
,
850 struct sw_flow_key
*key
, const struct nlattr
*attr
,
851 const struct nlattr
*actions
, int actions_len
)
853 const struct nlattr
*acts_list
= NULL
;
854 const struct nlattr
*a
;
858 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
859 a
= nla_next(a
, &rem
)) {
862 switch (nla_type(a
)) {
863 case OVS_SAMPLE_ATTR_PROBABILITY
:
864 probability
= nla_get_u32(a
);
865 if (!probability
|| prandom_u32() > probability
)
869 case OVS_SAMPLE_ATTR_ACTIONS
:
875 rem
= nla_len(acts_list
);
876 a
= nla_data(acts_list
);
878 /* Actions list is empty, do nothing */
882 /* The only known usage of sample action is having a single user-space
883 * action, or having a truncate action followed by a single user-space
884 * action. Treat this usage as a special case.
885 * The output_userspace() should clone the skb to be sent to the
886 * user space. This skb will be consumed by its caller.
888 if (unlikely(nla_type(a
) == OVS_ACTION_ATTR_TRUNC
)) {
889 struct ovs_action_trunc
*trunc
= nla_data(a
);
891 if (skb
->len
> trunc
->max_len
)
892 cutlen
= skb
->len
- trunc
->max_len
;
894 a
= nla_next(a
, &rem
);
897 if (likely(nla_type(a
) == OVS_ACTION_ATTR_USERSPACE
&&
898 nla_is_last(a
, rem
)))
899 return output_userspace(dp
, skb
, key
, a
, actions
,
900 actions_len
, cutlen
);
902 skb
= skb_clone(skb
, GFP_ATOMIC
);
904 /* Skip the sample action when out of memory. */
907 if (!add_deferred_actions(skb
, key
, a
)) {
909 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
917 static void execute_hash(struct sk_buff
*skb
, struct sw_flow_key
*key
,
918 const struct nlattr
*attr
)
920 struct ovs_action_hash
*hash_act
= nla_data(attr
);
923 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
924 hash
= skb_get_hash(skb
);
925 hash
= jhash_1word(hash
, hash_act
->hash_basis
);
929 key
->ovs_flow_hash
= hash
;
932 static int execute_set_action(struct sk_buff
*skb
,
933 struct sw_flow_key
*flow_key
,
934 const struct nlattr
*a
)
936 /* Only tunnel set execution is supported without a mask. */
937 if (nla_type(a
) == OVS_KEY_ATTR_TUNNEL_INFO
) {
938 struct ovs_tunnel_info
*tun
= nla_data(a
);
940 ovs_skb_dst_drop(skb
);
941 ovs_dst_hold((struct dst_entry
*)tun
->tun_dst
);
942 ovs_skb_dst_set(skb
, (struct dst_entry
*)tun
->tun_dst
);
949 /* Mask is at the midpoint of the data. */
950 #define get_mask(a, type) ((const type)nla_data(a) + 1)
952 static int execute_masked_set_action(struct sk_buff
*skb
,
953 struct sw_flow_key
*flow_key
,
954 const struct nlattr
*a
)
958 switch (nla_type(a
)) {
959 case OVS_KEY_ATTR_PRIORITY
:
960 OVS_SET_MASKED(skb
->priority
, nla_get_u32(a
),
961 *get_mask(a
, u32
*));
962 flow_key
->phy
.priority
= skb
->priority
;
965 case OVS_KEY_ATTR_SKB_MARK
:
966 OVS_SET_MASKED(skb
->mark
, nla_get_u32(a
), *get_mask(a
, u32
*));
967 flow_key
->phy
.skb_mark
= skb
->mark
;
970 case OVS_KEY_ATTR_TUNNEL_INFO
:
971 /* Masked data not supported for tunnel. */
975 case OVS_KEY_ATTR_ETHERNET
:
976 err
= set_eth_addr(skb
, flow_key
, nla_data(a
),
977 get_mask(a
, struct ovs_key_ethernet
*));
980 case OVS_KEY_ATTR_IPV4
:
981 err
= set_ipv4(skb
, flow_key
, nla_data(a
),
982 get_mask(a
, struct ovs_key_ipv4
*));
985 case OVS_KEY_ATTR_IPV6
:
986 err
= set_ipv6(skb
, flow_key
, nla_data(a
),
987 get_mask(a
, struct ovs_key_ipv6
*));
990 case OVS_KEY_ATTR_TCP
:
991 err
= set_tcp(skb
, flow_key
, nla_data(a
),
992 get_mask(a
, struct ovs_key_tcp
*));
995 case OVS_KEY_ATTR_UDP
:
996 err
= set_udp(skb
, flow_key
, nla_data(a
),
997 get_mask(a
, struct ovs_key_udp
*));
1000 case OVS_KEY_ATTR_SCTP
:
1001 err
= set_sctp(skb
, flow_key
, nla_data(a
),
1002 get_mask(a
, struct ovs_key_sctp
*));
1005 case OVS_KEY_ATTR_MPLS
:
1006 err
= set_mpls(skb
, flow_key
, nla_data(a
), get_mask(a
,
1010 case OVS_KEY_ATTR_CT_STATE
:
1011 case OVS_KEY_ATTR_CT_ZONE
:
1012 case OVS_KEY_ATTR_CT_MARK
:
1013 case OVS_KEY_ATTR_CT_LABELS
:
1021 static int execute_recirc(struct datapath
*dp
, struct sk_buff
*skb
,
1022 struct sw_flow_key
*key
,
1023 const struct nlattr
*a
, int rem
)
1025 struct deferred_action
*da
;
1028 if (!is_flow_key_valid(key
)) {
1031 err
= ovs_flow_key_update(skb
, key
);
1035 BUG_ON(!is_flow_key_valid(key
));
1037 if (!nla_is_last(a
, rem
)) {
1038 /* Recirc action is the not the last action
1039 * of the action list, need to clone the skb.
1041 skb
= skb_clone(skb
, GFP_ATOMIC
);
1043 /* Skip the recirc action when out of memory, but
1044 * continue on with the rest of the action list.
1050 level
= this_cpu_read(exec_actions_level
);
1051 if (level
<= OVS_DEFERRED_ACTION_THRESHOLD
) {
1052 struct recirc_keys
*rks
= this_cpu_ptr(recirc_keys
);
1053 struct sw_flow_key
*recirc_key
= &rks
->key
[level
- 1];
1056 recirc_key
->recirc_id
= nla_get_u32(a
);
1057 ovs_dp_process_packet(skb
, recirc_key
);
1062 da
= add_deferred_actions(skb
, key
, NULL
);
1064 da
->pkt_key
.recirc_id
= nla_get_u32(a
);
1068 if (net_ratelimit())
1069 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1076 /* Execute a list of actions against 'skb'. */
1077 static int do_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
1078 struct sw_flow_key
*key
,
1079 const struct nlattr
*attr
, int len
)
1081 /* Every output action needs a separate clone of 'skb', but the common
1082 * case is just a single output action, so that doing a clone and
1083 * then freeing the original skbuff is wasteful. So the following code
1084 * is slightly obscure just to avoid that.
1087 const struct nlattr
*a
;
1090 for (a
= attr
, rem
= len
; rem
> 0;
1091 a
= nla_next(a
, &rem
)) {
1094 if (unlikely(prev_port
!= -1)) {
1095 struct sk_buff
*out_skb
= skb_clone(skb
, GFP_ATOMIC
);
1098 do_output(dp
, out_skb
, prev_port
, key
);
1100 OVS_CB(skb
)->cutlen
= 0;
1104 switch (nla_type(a
)) {
1105 case OVS_ACTION_ATTR_OUTPUT
:
1106 prev_port
= nla_get_u32(a
);
1109 case OVS_ACTION_ATTR_TRUNC
: {
1110 struct ovs_action_trunc
*trunc
= nla_data(a
);
1112 if (skb
->len
> trunc
->max_len
)
1113 OVS_CB(skb
)->cutlen
= skb
->len
- trunc
->max_len
;
1117 case OVS_ACTION_ATTR_USERSPACE
:
1118 output_userspace(dp
, skb
, key
, a
, attr
,
1119 len
, OVS_CB(skb
)->cutlen
);
1120 OVS_CB(skb
)->cutlen
= 0;
1123 case OVS_ACTION_ATTR_HASH
:
1124 execute_hash(skb
, key
, a
);
1127 case OVS_ACTION_ATTR_PUSH_MPLS
:
1128 err
= push_mpls(skb
, key
, nla_data(a
));
1131 case OVS_ACTION_ATTR_POP_MPLS
:
1132 err
= pop_mpls(skb
, key
, nla_get_be16(a
));
1135 case OVS_ACTION_ATTR_PUSH_VLAN
:
1136 err
= push_vlan(skb
, key
, nla_data(a
));
1139 case OVS_ACTION_ATTR_POP_VLAN
:
1140 err
= pop_vlan(skb
, key
);
1143 case OVS_ACTION_ATTR_RECIRC
:
1144 err
= execute_recirc(dp
, skb
, key
, a
, rem
);
1145 if (nla_is_last(a
, rem
)) {
1146 /* If this is the last action, the skb has
1147 * been consumed or freed.
1148 * Return immediately.
1154 case OVS_ACTION_ATTR_SET
:
1155 err
= execute_set_action(skb
, key
, nla_data(a
));
1158 case OVS_ACTION_ATTR_SET_MASKED
:
1159 case OVS_ACTION_ATTR_SET_TO_MASKED
:
1160 err
= execute_masked_set_action(skb
, key
, nla_data(a
));
1163 case OVS_ACTION_ATTR_SAMPLE
:
1164 err
= sample(dp
, skb
, key
, a
, attr
, len
);
1167 case OVS_ACTION_ATTR_CT
:
1168 if (!is_flow_key_valid(key
)) {
1169 err
= ovs_flow_key_update(skb
, key
);
1174 err
= ovs_ct_execute(ovs_dp_get_net(dp
), skb
, key
,
1177 /* Hide stolen IP fragments from user space. */
1179 return err
== -EINPROGRESS
? 0 : err
;
1183 if (unlikely(err
)) {
1189 if (prev_port
!= -1)
1190 do_output(dp
, skb
, prev_port
, key
);
1197 static void process_deferred_actions(struct datapath
*dp
)
1199 struct action_fifo
*fifo
= this_cpu_ptr(action_fifos
);
1201 /* Do not touch the FIFO in case there is no deferred actions. */
1202 if (action_fifo_is_empty(fifo
))
1205 /* Finishing executing all deferred actions. */
1207 struct deferred_action
*da
= action_fifo_get(fifo
);
1208 struct sk_buff
*skb
= da
->skb
;
1209 struct sw_flow_key
*key
= &da
->pkt_key
;
1210 const struct nlattr
*actions
= da
->actions
;
1213 do_execute_actions(dp
, skb
, key
, actions
,
1216 ovs_dp_process_packet(skb
, key
);
1217 } while (!action_fifo_is_empty(fifo
));
1219 /* Reset FIFO for the next packet. */
1220 action_fifo_init(fifo
);
1223 /* Execute a list of actions against 'skb'. */
1224 int ovs_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
1225 const struct sw_flow_actions
*acts
,
1226 struct sw_flow_key
*key
)
1230 level
= __this_cpu_inc_return(exec_actions_level
);
1231 if (unlikely(level
> OVS_RECURSION_LIMIT
)) {
1232 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1239 err
= do_execute_actions(dp
, skb
, key
,
1240 acts
->actions
, acts
->actions_len
);
1243 process_deferred_actions(dp
);
1246 __this_cpu_dec(exec_actions_level
);
1250 int action_fifos_init(void)
1252 action_fifos
= alloc_percpu(struct action_fifo
);
1256 recirc_keys
= alloc_percpu(struct recirc_keys
);
1258 free_percpu(action_fifos
);
1265 void action_fifos_exit(void)
1267 free_percpu(action_fifos
);
1268 free_percpu(recirc_keys
);