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
;
68 u8 l2_data
[MAX_L2_LEN
];
71 static DEFINE_PER_CPU(struct ovs_frag_data
, ovs_frag_data_storage
);
73 #define DEFERRED_ACTION_FIFO_SIZE 10
74 #define OVS_RECURSION_LIMIT 4
75 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
79 /* Deferred action fifo queue storage. */
80 struct deferred_action fifo
[DEFERRED_ACTION_FIFO_SIZE
];
84 struct sw_flow_key key
[OVS_DEFERRED_ACTION_THRESHOLD
];
87 static struct action_fifo __percpu
*action_fifos
;
88 static struct recirc_keys __percpu
*recirc_keys
;
89 static DEFINE_PER_CPU(int, exec_actions_level
);
91 static void action_fifo_init(struct action_fifo
*fifo
)
97 static bool action_fifo_is_empty(const struct action_fifo
*fifo
)
99 return (fifo
->head
== fifo
->tail
);
102 static struct deferred_action
*action_fifo_get(struct action_fifo
*fifo
)
104 if (action_fifo_is_empty(fifo
))
107 return &fifo
->fifo
[fifo
->tail
++];
110 static struct deferred_action
*action_fifo_put(struct action_fifo
*fifo
)
112 if (fifo
->head
>= DEFERRED_ACTION_FIFO_SIZE
- 1)
115 return &fifo
->fifo
[fifo
->head
++];
118 /* Return queue entry if fifo is not full */
119 static struct deferred_action
*add_deferred_actions(struct sk_buff
*skb
,
120 const struct sw_flow_key
*key
,
121 const struct nlattr
*attr
)
123 struct action_fifo
*fifo
;
124 struct deferred_action
*da
;
126 fifo
= this_cpu_ptr(action_fifos
);
127 da
= action_fifo_put(fifo
);
137 static void invalidate_flow_key(struct sw_flow_key
*key
)
139 key
->mac_proto
|= SW_FLOW_KEY_INVALID
;
142 static bool is_flow_key_valid(const struct sw_flow_key
*key
)
144 return !(key
->mac_proto
& SW_FLOW_KEY_INVALID
);
147 static void update_ethertype(struct sk_buff
*skb
, struct ethhdr
*hdr
,
150 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
151 __be16 diff
[] = { ~(hdr
->h_proto
), ethertype
};
153 skb
->csum
= ~csum_partial((char *)diff
, sizeof(diff
),
157 hdr
->h_proto
= ethertype
;
160 static int push_mpls(struct sk_buff
*skb
, struct sw_flow_key
*key
,
161 const struct ovs_action_push_mpls
*mpls
)
163 __be32
*new_mpls_lse
;
165 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
166 if (skb
->encapsulation
)
169 if (skb_cow_head(skb
, MPLS_HLEN
) < 0)
172 skb_push(skb
, MPLS_HLEN
);
173 memmove(skb_mac_header(skb
) - MPLS_HLEN
, skb_mac_header(skb
),
175 skb_reset_mac_header(skb
);
177 new_mpls_lse
= (__be32
*)skb_mpls_header(skb
);
178 *new_mpls_lse
= mpls
->mpls_lse
;
180 skb_postpush_rcsum(skb
, new_mpls_lse
, MPLS_HLEN
);
182 update_ethertype(skb
, eth_hdr(skb
), mpls
->mpls_ethertype
);
183 if (!ovs_skb_get_inner_protocol(skb
))
184 ovs_skb_set_inner_protocol(skb
, skb
->protocol
);
185 skb
->protocol
= mpls
->mpls_ethertype
;
187 invalidate_flow_key(key
);
191 static int pop_mpls(struct sk_buff
*skb
, struct sw_flow_key
*key
,
192 const __be16 ethertype
)
197 err
= skb_ensure_writable(skb
, skb
->mac_len
+ MPLS_HLEN
);
201 skb_postpull_rcsum(skb
, skb_mpls_header(skb
), MPLS_HLEN
);
203 memmove(skb_mac_header(skb
) + MPLS_HLEN
, skb_mac_header(skb
),
206 __skb_pull(skb
, MPLS_HLEN
);
207 skb_reset_mac_header(skb
);
209 /* skb_mpls_header() is used to locate the ethertype
210 * field correctly in the presence of VLAN tags.
212 hdr
= (struct ethhdr
*)(skb_mpls_header(skb
) - ETH_HLEN
);
213 update_ethertype(skb
, hdr
, ethertype
);
214 if (eth_p_mpls(skb
->protocol
))
215 skb
->protocol
= ethertype
;
217 invalidate_flow_key(key
);
221 static int set_mpls(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
222 const __be32
*mpls_lse
, const __be32
*mask
)
228 err
= skb_ensure_writable(skb
, skb
->mac_len
+ MPLS_HLEN
);
232 stack
= (__be32
*)skb_mpls_header(skb
);
233 lse
= OVS_MASKED(*stack
, *mpls_lse
, *mask
);
234 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
235 __be32 diff
[] = { ~(*stack
), lse
};
237 skb
->csum
= ~csum_partial((char *)diff
, sizeof(diff
),
242 flow_key
->mpls
.top_lse
= lse
;
246 static int pop_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
)
250 err
= skb_vlan_pop(skb
);
251 if (skb_vlan_tag_present(skb
)) {
252 invalidate_flow_key(key
);
254 key
->eth
.vlan
.tci
= 0;
255 key
->eth
.vlan
.tpid
= 0;
260 static int push_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
,
261 const struct ovs_action_push_vlan
*vlan
)
263 if (skb_vlan_tag_present(skb
)) {
264 invalidate_flow_key(key
);
266 key
->eth
.vlan
.tci
= vlan
->vlan_tci
;
267 key
->eth
.vlan
.tpid
= vlan
->vlan_tpid
;
269 return skb_vlan_push(skb
, vlan
->vlan_tpid
,
270 ntohs(vlan
->vlan_tci
) & ~VLAN_TAG_PRESENT
);
273 /* 'src' is already properly masked. */
274 static void ether_addr_copy_masked(u8
*dst_
, const u8
*src_
, const u8
*mask_
)
276 u16
*dst
= (u16
*)dst_
;
277 const u16
*src
= (const u16
*)src_
;
278 const u16
*mask
= (const u16
*)mask_
;
280 OVS_SET_MASKED(dst
[0], src
[0], mask
[0]);
281 OVS_SET_MASKED(dst
[1], src
[1], mask
[1]);
282 OVS_SET_MASKED(dst
[2], src
[2], mask
[2]);
285 static int set_eth_addr(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
286 const struct ovs_key_ethernet
*key
,
287 const struct ovs_key_ethernet
*mask
)
291 err
= skb_ensure_writable(skb
, ETH_HLEN
);
295 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_ALEN
* 2);
297 ether_addr_copy_masked(eth_hdr(skb
)->h_source
, key
->eth_src
,
299 ether_addr_copy_masked(eth_hdr(skb
)->h_dest
, key
->eth_dst
,
302 skb_postpush_rcsum(skb
, eth_hdr(skb
), ETH_ALEN
* 2);
304 ether_addr_copy(flow_key
->eth
.src
, eth_hdr(skb
)->h_source
);
305 ether_addr_copy(flow_key
->eth
.dst
, eth_hdr(skb
)->h_dest
);
309 static void update_ip_l4_checksum(struct sk_buff
*skb
, struct iphdr
*nh
,
310 __be32 addr
, __be32 new_addr
)
312 int transport_len
= skb
->len
- skb_transport_offset(skb
);
314 if (nh
->frag_off
& htons(IP_OFFSET
))
317 if (nh
->protocol
== IPPROTO_TCP
) {
318 if (likely(transport_len
>= sizeof(struct tcphdr
)))
319 inet_proto_csum_replace4(&tcp_hdr(skb
)->check
, skb
,
320 addr
, new_addr
, true);
321 } else if (nh
->protocol
== IPPROTO_UDP
) {
322 if (likely(transport_len
>= sizeof(struct udphdr
))) {
323 struct udphdr
*uh
= udp_hdr(skb
);
325 if (uh
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
326 inet_proto_csum_replace4(&uh
->check
, skb
,
327 addr
, new_addr
, true);
329 uh
->check
= CSUM_MANGLED_0
;
336 static void set_ip_addr(struct sk_buff
*skb
, struct iphdr
*nh
,
337 __be32
*addr
, __be32 new_addr
)
339 update_ip_l4_checksum(skb
, nh
, *addr
, new_addr
);
340 csum_replace4(&nh
->check
, *addr
, new_addr
);
345 static void update_ipv6_checksum(struct sk_buff
*skb
, u8 l4_proto
,
346 __be32 addr
[4], const __be32 new_addr
[4])
348 int transport_len
= skb
->len
- skb_transport_offset(skb
);
350 if (l4_proto
== NEXTHDR_TCP
) {
351 if (likely(transport_len
>= sizeof(struct tcphdr
)))
352 inet_proto_csum_replace16(&tcp_hdr(skb
)->check
, skb
,
353 addr
, new_addr
, true);
354 } else if (l4_proto
== NEXTHDR_UDP
) {
355 if (likely(transport_len
>= sizeof(struct udphdr
))) {
356 struct udphdr
*uh
= udp_hdr(skb
);
358 if (uh
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
359 inet_proto_csum_replace16(&uh
->check
, skb
,
360 addr
, new_addr
, true);
362 uh
->check
= CSUM_MANGLED_0
;
365 } else if (l4_proto
== NEXTHDR_ICMP
) {
366 if (likely(transport_len
>= sizeof(struct icmp6hdr
)))
367 inet_proto_csum_replace16(&icmp6_hdr(skb
)->icmp6_cksum
,
368 skb
, addr
, new_addr
, true);
372 static void mask_ipv6_addr(const __be32 old
[4], const __be32 addr
[4],
373 const __be32 mask
[4], __be32 masked
[4])
375 masked
[0] = OVS_MASKED(old
[0], addr
[0], mask
[0]);
376 masked
[1] = OVS_MASKED(old
[1], addr
[1], mask
[1]);
377 masked
[2] = OVS_MASKED(old
[2], addr
[2], mask
[2]);
378 masked
[3] = OVS_MASKED(old
[3], addr
[3], mask
[3]);
381 static void set_ipv6_addr(struct sk_buff
*skb
, u8 l4_proto
,
382 __be32 addr
[4], const __be32 new_addr
[4],
383 bool recalculate_csum
)
385 if (likely(recalculate_csum
))
386 update_ipv6_checksum(skb
, l4_proto
, addr
, new_addr
);
389 memcpy(addr
, new_addr
, sizeof(__be32
[4]));
392 static void set_ipv6_fl(struct ipv6hdr
*nh
, u32 fl
, u32 mask
)
394 /* Bits 21-24 are always unmasked, so this retains their values. */
395 OVS_SET_MASKED(nh
->flow_lbl
[0], (u8
)(fl
>> 16), (u8
)(mask
>> 16));
396 OVS_SET_MASKED(nh
->flow_lbl
[1], (u8
)(fl
>> 8), (u8
)(mask
>> 8));
397 OVS_SET_MASKED(nh
->flow_lbl
[2], (u8
)fl
, (u8
)mask
);
400 static void set_ip_ttl(struct sk_buff
*skb
, struct iphdr
*nh
, u8 new_ttl
,
403 new_ttl
= OVS_MASKED(nh
->ttl
, new_ttl
, mask
);
405 csum_replace2(&nh
->check
, htons(nh
->ttl
<< 8), htons(new_ttl
<< 8));
409 static int set_ipv4(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
410 const struct ovs_key_ipv4
*key
,
411 const struct ovs_key_ipv4
*mask
)
417 err
= skb_ensure_writable(skb
, skb_network_offset(skb
) +
418 sizeof(struct iphdr
));
424 /* Setting an IP addresses is typically only a side effect of
425 * matching on them in the current userspace implementation, so it
426 * makes sense to check if the value actually changed.
428 if (mask
->ipv4_src
) {
429 new_addr
= OVS_MASKED(nh
->saddr
, key
->ipv4_src
, mask
->ipv4_src
);
431 if (unlikely(new_addr
!= nh
->saddr
)) {
432 set_ip_addr(skb
, nh
, &nh
->saddr
, new_addr
);
433 flow_key
->ipv4
.addr
.src
= new_addr
;
436 if (mask
->ipv4_dst
) {
437 new_addr
= OVS_MASKED(nh
->daddr
, key
->ipv4_dst
, mask
->ipv4_dst
);
439 if (unlikely(new_addr
!= nh
->daddr
)) {
440 set_ip_addr(skb
, nh
, &nh
->daddr
, new_addr
);
441 flow_key
->ipv4
.addr
.dst
= new_addr
;
444 if (mask
->ipv4_tos
) {
445 ipv4_change_dsfield(nh
, ~mask
->ipv4_tos
, key
->ipv4_tos
);
446 flow_key
->ip
.tos
= nh
->tos
;
448 if (mask
->ipv4_ttl
) {
449 set_ip_ttl(skb
, nh
, key
->ipv4_ttl
, mask
->ipv4_ttl
);
450 flow_key
->ip
.ttl
= nh
->ttl
;
456 static bool is_ipv6_mask_nonzero(const __be32 addr
[4])
458 return !!(addr
[0] | addr
[1] | addr
[2] | addr
[3]);
461 static int set_ipv6(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
462 const struct ovs_key_ipv6
*key
,
463 const struct ovs_key_ipv6
*mask
)
468 err
= skb_ensure_writable(skb
, skb_network_offset(skb
) +
469 sizeof(struct ipv6hdr
));
475 /* Setting an IP addresses is typically only a side effect of
476 * matching on them in the current userspace implementation, so it
477 * makes sense to check if the value actually changed.
479 if (is_ipv6_mask_nonzero(mask
->ipv6_src
)) {
480 __be32
*saddr
= (__be32
*)&nh
->saddr
;
483 mask_ipv6_addr(saddr
, key
->ipv6_src
, mask
->ipv6_src
, masked
);
485 if (unlikely(memcmp(saddr
, masked
, sizeof(masked
)))) {
486 set_ipv6_addr(skb
, flow_key
->ip
.proto
, saddr
, masked
,
488 memcpy(&flow_key
->ipv6
.addr
.src
, masked
,
489 sizeof(flow_key
->ipv6
.addr
.src
));
492 if (is_ipv6_mask_nonzero(mask
->ipv6_dst
)) {
493 unsigned int offset
= 0;
494 int flags
= IP6_FH_F_SKIP_RH
;
495 bool recalc_csum
= true;
496 __be32
*daddr
= (__be32
*)&nh
->daddr
;
499 mask_ipv6_addr(daddr
, key
->ipv6_dst
, mask
->ipv6_dst
, masked
);
501 if (unlikely(memcmp(daddr
, masked
, sizeof(masked
)))) {
502 if (ipv6_ext_hdr(nh
->nexthdr
))
503 recalc_csum
= (ipv6_find_hdr(skb
, &offset
,
508 set_ipv6_addr(skb
, flow_key
->ip
.proto
, daddr
, masked
,
510 memcpy(&flow_key
->ipv6
.addr
.dst
, masked
,
511 sizeof(flow_key
->ipv6
.addr
.dst
));
514 if (mask
->ipv6_tclass
) {
515 ipv6_change_dsfield(nh
, ~mask
->ipv6_tclass
, key
->ipv6_tclass
);
516 flow_key
->ip
.tos
= ipv6_get_dsfield(nh
);
518 if (mask
->ipv6_label
) {
519 set_ipv6_fl(nh
, ntohl(key
->ipv6_label
),
520 ntohl(mask
->ipv6_label
));
521 flow_key
->ipv6
.label
=
522 *(__be32
*)nh
& htonl(IPV6_FLOWINFO_FLOWLABEL
);
524 if (mask
->ipv6_hlimit
) {
525 OVS_SET_MASKED(nh
->hop_limit
, key
->ipv6_hlimit
,
527 flow_key
->ip
.ttl
= nh
->hop_limit
;
532 /* Must follow skb_ensure_writable() since that can move the skb data. */
533 static void set_tp_port(struct sk_buff
*skb
, __be16
*port
,
534 __be16 new_port
, __sum16
*check
)
536 inet_proto_csum_replace2(check
, skb
, *port
, new_port
, false);
540 static int set_udp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
541 const struct ovs_key_udp
*key
,
542 const struct ovs_key_udp
*mask
)
548 err
= skb_ensure_writable(skb
, skb_transport_offset(skb
) +
549 sizeof(struct udphdr
));
554 /* Either of the masks is non-zero, so do not bother checking them. */
555 src
= OVS_MASKED(uh
->source
, key
->udp_src
, mask
->udp_src
);
556 dst
= OVS_MASKED(uh
->dest
, key
->udp_dst
, mask
->udp_dst
);
558 if (uh
->check
&& skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
559 if (likely(src
!= uh
->source
)) {
560 set_tp_port(skb
, &uh
->source
, src
, &uh
->check
);
561 flow_key
->tp
.src
= src
;
563 if (likely(dst
!= uh
->dest
)) {
564 set_tp_port(skb
, &uh
->dest
, dst
, &uh
->check
);
565 flow_key
->tp
.dst
= dst
;
568 if (unlikely(!uh
->check
))
569 uh
->check
= CSUM_MANGLED_0
;
573 flow_key
->tp
.src
= src
;
574 flow_key
->tp
.dst
= dst
;
582 static int set_tcp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
583 const struct ovs_key_tcp
*key
,
584 const struct ovs_key_tcp
*mask
)
590 err
= skb_ensure_writable(skb
, skb_transport_offset(skb
) +
591 sizeof(struct tcphdr
));
596 src
= OVS_MASKED(th
->source
, key
->tcp_src
, mask
->tcp_src
);
597 if (likely(src
!= th
->source
)) {
598 set_tp_port(skb
, &th
->source
, src
, &th
->check
);
599 flow_key
->tp
.src
= src
;
601 dst
= OVS_MASKED(th
->dest
, key
->tcp_dst
, mask
->tcp_dst
);
602 if (likely(dst
!= th
->dest
)) {
603 set_tp_port(skb
, &th
->dest
, dst
, &th
->check
);
604 flow_key
->tp
.dst
= dst
;
611 static int set_sctp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
612 const struct ovs_key_sctp
*key
,
613 const struct ovs_key_sctp
*mask
)
615 unsigned int sctphoff
= skb_transport_offset(skb
);
617 __le32 old_correct_csum
, new_csum
, old_csum
;
620 err
= skb_ensure_writable(skb
, sctphoff
+ sizeof(struct sctphdr
));
625 old_csum
= sh
->checksum
;
626 old_correct_csum
= sctp_compute_cksum(skb
, sctphoff
);
628 sh
->source
= OVS_MASKED(sh
->source
, key
->sctp_src
, mask
->sctp_src
);
629 sh
->dest
= OVS_MASKED(sh
->dest
, key
->sctp_dst
, mask
->sctp_dst
);
631 new_csum
= sctp_compute_cksum(skb
, sctphoff
);
633 /* Carry any checksum errors through. */
634 sh
->checksum
= old_csum
^ old_correct_csum
^ new_csum
;
637 flow_key
->tp
.src
= sh
->source
;
638 flow_key
->tp
.dst
= sh
->dest
;
643 static int ovs_vport_output(OVS_VPORT_OUTPUT_PARAMS
)
645 struct ovs_frag_data
*data
= this_cpu_ptr(&ovs_frag_data_storage
);
646 struct vport
*vport
= data
->vport
;
648 if (skb_cow_head(skb
, data
->l2_len
) < 0) {
653 __skb_dst_copy(skb
, data
->dst
);
654 *OVS_GSO_CB(skb
) = data
->cb
;
655 ovs_skb_set_inner_protocol(skb
, data
->inner_protocol
);
656 skb
->vlan_tci
= data
->vlan_tci
;
657 skb
->vlan_proto
= data
->vlan_proto
;
659 /* Reconstruct the MAC header. */
660 skb_push(skb
, data
->l2_len
);
661 memcpy(skb
->data
, &data
->l2_data
, data
->l2_len
);
662 skb_postpush_rcsum(skb
, skb
->data
, data
->l2_len
);
663 skb_reset_mac_header(skb
);
665 ovs_vport_send(vport
, skb
, data
->mac_proto
);
670 ovs_dst_get_mtu(const struct dst_entry
*dst
)
672 return dst
->dev
->mtu
;
675 static struct dst_ops ovs_dst_ops
= {
677 .mtu
= ovs_dst_get_mtu
,
680 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
681 * ovs_vport_output(), which is called once per fragmented packet.
683 static void prepare_frag(struct vport
*vport
, struct sk_buff
*skb
,
686 unsigned int hlen
= skb_network_offset(skb
);
687 struct ovs_frag_data
*data
;
689 data
= this_cpu_ptr(&ovs_frag_data_storage
);
690 data
->dst
= (unsigned long) skb_dst(skb
);
692 data
->cb
= *OVS_GSO_CB(skb
);
693 data
->inner_protocol
= ovs_skb_get_inner_protocol(skb
);
694 data
->vlan_tci
= skb
->vlan_tci
;
695 data
->vlan_proto
= skb
->vlan_proto
;
696 data
->mac_proto
= mac_proto
;
698 memcpy(&data
->l2_data
, skb
->data
, hlen
);
700 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
704 static void ovs_fragment(struct net
*net
, struct vport
*vport
,
705 struct sk_buff
*skb
, u16 mru
,
706 struct sw_flow_key
*key
)
708 if (skb_network_offset(skb
) > MAX_L2_LEN
) {
709 OVS_NLERR(1, "L2 header too long to fragment");
713 if (key
->eth
.type
== htons(ETH_P_IP
)) {
714 struct dst_entry ovs_dst
;
715 unsigned long orig_dst
;
717 prepare_frag(vport
, skb
, ovs_key_mac_proto(key
));
718 dst_init(&ovs_dst
, &ovs_dst_ops
, NULL
, 1,
719 DST_OBSOLETE_NONE
, DST_NOCOUNT
);
720 ovs_dst
.dev
= vport
->dev
;
722 orig_dst
= (unsigned long) skb_dst(skb
);
723 skb_dst_set_noref(skb
, &ovs_dst
);
724 IPCB(skb
)->frag_max_size
= mru
;
726 ip_do_fragment(net
, skb
->sk
, skb
, ovs_vport_output
);
727 refdst_drop(orig_dst
);
728 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
729 const struct nf_ipv6_ops
*v6ops
= nf_get_ipv6_ops();
730 unsigned long orig_dst
;
731 struct rt6_info ovs_rt
;
737 prepare_frag(vport
, skb
,
738 ovs_key_mac_proto(key
));
739 memset(&ovs_rt
, 0, sizeof(ovs_rt
));
740 dst_init(&ovs_rt
.dst
, &ovs_dst_ops
, NULL
, 1,
741 DST_OBSOLETE_NONE
, DST_NOCOUNT
);
742 ovs_rt
.dst
.dev
= vport
->dev
;
744 orig_dst
= (unsigned long) skb_dst(skb
);
745 skb_dst_set_noref(skb
, &ovs_rt
.dst
);
746 IP6CB(skb
)->frag_max_size
= mru
;
747 #ifdef HAVE_IP_LOCAL_OUT_TAKES_NET
748 v6ops
->fragment(net
, skb
->sk
, skb
, ovs_vport_output
);
750 v6ops
->fragment(skb
->sk
, skb
, ovs_vport_output
);
752 refdst_drop(orig_dst
);
754 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
755 ovs_vport_name(vport
), ntohs(key
->eth
.type
), mru
,
765 static void do_output(struct datapath
*dp
, struct sk_buff
*skb
, int out_port
,
766 struct sw_flow_key
*key
)
768 struct vport
*vport
= ovs_vport_rcu(dp
, out_port
);
771 u16 mru
= OVS_CB(skb
)->mru
;
772 u32 cutlen
= OVS_CB(skb
)->cutlen
;
774 if (unlikely(cutlen
> 0)) {
775 if (skb
->len
- cutlen
> ovs_mac_header_len(key
))
776 pskb_trim(skb
, skb
->len
- cutlen
);
778 pskb_trim(skb
, ovs_mac_header_len(key
));
782 (skb
->len
<= mru
+ vport
->dev
->hard_header_len
))) {
783 ovs_vport_send(vport
, skb
, ovs_key_mac_proto(key
));
784 } else if (mru
<= vport
->dev
->mtu
) {
785 struct net
*net
= ovs_dp_get_net(dp
);
787 ovs_fragment(net
, vport
, skb
, mru
, key
);
789 OVS_NLERR(true, "Cannot fragment IP frames");
797 static int output_userspace(struct datapath
*dp
, struct sk_buff
*skb
,
798 struct sw_flow_key
*key
, const struct nlattr
*attr
,
799 const struct nlattr
*actions
, int actions_len
,
802 struct dp_upcall_info upcall
;
803 const struct nlattr
*a
;
806 memset(&upcall
, 0, sizeof(upcall
));
807 upcall
.cmd
= OVS_PACKET_CMD_ACTION
;
808 upcall
.mru
= OVS_CB(skb
)->mru
;
810 SKB_INIT_FILL_METADATA_DST(skb
);
811 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
812 a
= nla_next(a
, &rem
)) {
813 switch (nla_type(a
)) {
814 case OVS_USERSPACE_ATTR_USERDATA
:
818 case OVS_USERSPACE_ATTR_PID
:
819 upcall
.portid
= nla_get_u32(a
);
822 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT
: {
823 /* Get out tunnel info. */
826 vport
= ovs_vport_rcu(dp
, nla_get_u32(a
));
828 err
= dev_fill_metadata_dst(vport
->dev
, skb
);
830 upcall
.egress_tun_info
= skb_tunnel_info(skb
);
836 case OVS_USERSPACE_ATTR_ACTIONS
: {
837 /* Include actions. */
838 upcall
.actions
= actions
;
839 upcall
.actions_len
= actions_len
;
843 } /* End of switch. */
846 err
= ovs_dp_upcall(dp
, skb
, key
, &upcall
, cutlen
);
847 SKB_RESTORE_FILL_METADATA_DST(skb
);
851 static int sample(struct datapath
*dp
, struct sk_buff
*skb
,
852 struct sw_flow_key
*key
, const struct nlattr
*attr
,
853 const struct nlattr
*actions
, int actions_len
)
855 const struct nlattr
*acts_list
= NULL
;
856 const struct nlattr
*a
;
860 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
861 a
= nla_next(a
, &rem
)) {
864 switch (nla_type(a
)) {
865 case OVS_SAMPLE_ATTR_PROBABILITY
:
866 probability
= nla_get_u32(a
);
867 if (!probability
|| prandom_u32() > probability
)
871 case OVS_SAMPLE_ATTR_ACTIONS
:
877 rem
= nla_len(acts_list
);
878 a
= nla_data(acts_list
);
880 /* Actions list is empty, do nothing */
884 /* The only known usage of sample action is having a single user-space
885 * action, or having a truncate action followed by a single user-space
886 * action. Treat this usage as a special case.
887 * The output_userspace() should clone the skb to be sent to the
888 * user space. This skb will be consumed by its caller.
890 if (unlikely(nla_type(a
) == OVS_ACTION_ATTR_TRUNC
)) {
891 struct ovs_action_trunc
*trunc
= nla_data(a
);
893 if (skb
->len
> trunc
->max_len
)
894 cutlen
= skb
->len
- trunc
->max_len
;
896 a
= nla_next(a
, &rem
);
899 if (likely(nla_type(a
) == OVS_ACTION_ATTR_USERSPACE
&&
900 nla_is_last(a
, rem
)))
901 return output_userspace(dp
, skb
, key
, a
, actions
,
902 actions_len
, cutlen
);
904 skb
= skb_clone(skb
, GFP_ATOMIC
);
906 /* Skip the sample action when out of memory. */
909 if (!add_deferred_actions(skb
, key
, a
)) {
911 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
919 static void execute_hash(struct sk_buff
*skb
, struct sw_flow_key
*key
,
920 const struct nlattr
*attr
)
922 struct ovs_action_hash
*hash_act
= nla_data(attr
);
925 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
926 hash
= skb_get_hash(skb
);
927 hash
= jhash_1word(hash
, hash_act
->hash_basis
);
931 key
->ovs_flow_hash
= hash
;
934 static int execute_set_action(struct sk_buff
*skb
,
935 struct sw_flow_key
*flow_key
,
936 const struct nlattr
*a
)
938 /* Only tunnel set execution is supported without a mask. */
939 if (nla_type(a
) == OVS_KEY_ATTR_TUNNEL_INFO
) {
940 struct ovs_tunnel_info
*tun
= nla_data(a
);
942 ovs_skb_dst_drop(skb
);
943 ovs_dst_hold((struct dst_entry
*)tun
->tun_dst
);
944 ovs_skb_dst_set(skb
, (struct dst_entry
*)tun
->tun_dst
);
951 /* Mask is at the midpoint of the data. */
952 #define get_mask(a, type) ((const type)nla_data(a) + 1)
954 static int execute_masked_set_action(struct sk_buff
*skb
,
955 struct sw_flow_key
*flow_key
,
956 const struct nlattr
*a
)
960 switch (nla_type(a
)) {
961 case OVS_KEY_ATTR_PRIORITY
:
962 OVS_SET_MASKED(skb
->priority
, nla_get_u32(a
),
963 *get_mask(a
, u32
*));
964 flow_key
->phy
.priority
= skb
->priority
;
967 case OVS_KEY_ATTR_SKB_MARK
:
968 OVS_SET_MASKED(skb
->mark
, nla_get_u32(a
), *get_mask(a
, u32
*));
969 flow_key
->phy
.skb_mark
= skb
->mark
;
972 case OVS_KEY_ATTR_TUNNEL_INFO
:
973 /* Masked data not supported for tunnel. */
977 case OVS_KEY_ATTR_ETHERNET
:
978 err
= set_eth_addr(skb
, flow_key
, nla_data(a
),
979 get_mask(a
, struct ovs_key_ethernet
*));
982 case OVS_KEY_ATTR_IPV4
:
983 err
= set_ipv4(skb
, flow_key
, nla_data(a
),
984 get_mask(a
, struct ovs_key_ipv4
*));
987 case OVS_KEY_ATTR_IPV6
:
988 err
= set_ipv6(skb
, flow_key
, nla_data(a
),
989 get_mask(a
, struct ovs_key_ipv6
*));
992 case OVS_KEY_ATTR_TCP
:
993 err
= set_tcp(skb
, flow_key
, nla_data(a
),
994 get_mask(a
, struct ovs_key_tcp
*));
997 case OVS_KEY_ATTR_UDP
:
998 err
= set_udp(skb
, flow_key
, nla_data(a
),
999 get_mask(a
, struct ovs_key_udp
*));
1002 case OVS_KEY_ATTR_SCTP
:
1003 err
= set_sctp(skb
, flow_key
, nla_data(a
),
1004 get_mask(a
, struct ovs_key_sctp
*));
1007 case OVS_KEY_ATTR_MPLS
:
1008 err
= set_mpls(skb
, flow_key
, nla_data(a
), get_mask(a
,
1012 case OVS_KEY_ATTR_CT_STATE
:
1013 case OVS_KEY_ATTR_CT_ZONE
:
1014 case OVS_KEY_ATTR_CT_MARK
:
1015 case OVS_KEY_ATTR_CT_LABELS
:
1023 static int execute_recirc(struct datapath
*dp
, struct sk_buff
*skb
,
1024 struct sw_flow_key
*key
,
1025 const struct nlattr
*a
, int rem
)
1027 struct deferred_action
*da
;
1030 if (!is_flow_key_valid(key
)) {
1033 err
= ovs_flow_key_update(skb
, key
);
1037 BUG_ON(!is_flow_key_valid(key
));
1039 if (!nla_is_last(a
, rem
)) {
1040 /* Recirc action is the not the last action
1041 * of the action list, need to clone the skb.
1043 skb
= skb_clone(skb
, GFP_ATOMIC
);
1045 /* Skip the recirc action when out of memory, but
1046 * continue on with the rest of the action list.
1052 level
= this_cpu_read(exec_actions_level
);
1053 if (level
<= OVS_DEFERRED_ACTION_THRESHOLD
) {
1054 struct recirc_keys
*rks
= this_cpu_ptr(recirc_keys
);
1055 struct sw_flow_key
*recirc_key
= &rks
->key
[level
- 1];
1058 recirc_key
->recirc_id
= nla_get_u32(a
);
1059 ovs_dp_process_packet(skb
, recirc_key
);
1064 da
= add_deferred_actions(skb
, key
, NULL
);
1066 da
->pkt_key
.recirc_id
= nla_get_u32(a
);
1070 if (net_ratelimit())
1071 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1078 /* Execute a list of actions against 'skb'. */
1079 static int do_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
1080 struct sw_flow_key
*key
,
1081 const struct nlattr
*attr
, int len
)
1083 /* Every output action needs a separate clone of 'skb', but the common
1084 * case is just a single output action, so that doing a clone and
1085 * then freeing the original skbuff is wasteful. So the following code
1086 * is slightly obscure just to avoid that.
1089 const struct nlattr
*a
;
1092 for (a
= attr
, rem
= len
; rem
> 0;
1093 a
= nla_next(a
, &rem
)) {
1096 if (unlikely(prev_port
!= -1)) {
1097 struct sk_buff
*out_skb
= skb_clone(skb
, GFP_ATOMIC
);
1100 do_output(dp
, out_skb
, prev_port
, key
);
1102 OVS_CB(skb
)->cutlen
= 0;
1106 switch (nla_type(a
)) {
1107 case OVS_ACTION_ATTR_OUTPUT
:
1108 prev_port
= nla_get_u32(a
);
1111 case OVS_ACTION_ATTR_TRUNC
: {
1112 struct ovs_action_trunc
*trunc
= nla_data(a
);
1114 if (skb
->len
> trunc
->max_len
)
1115 OVS_CB(skb
)->cutlen
= skb
->len
- trunc
->max_len
;
1119 case OVS_ACTION_ATTR_USERSPACE
:
1120 output_userspace(dp
, skb
, key
, a
, attr
,
1121 len
, OVS_CB(skb
)->cutlen
);
1122 OVS_CB(skb
)->cutlen
= 0;
1125 case OVS_ACTION_ATTR_HASH
:
1126 execute_hash(skb
, key
, a
);
1129 case OVS_ACTION_ATTR_PUSH_MPLS
:
1130 err
= push_mpls(skb
, key
, nla_data(a
));
1133 case OVS_ACTION_ATTR_POP_MPLS
:
1134 err
= pop_mpls(skb
, key
, nla_get_be16(a
));
1137 case OVS_ACTION_ATTR_PUSH_VLAN
:
1138 err
= push_vlan(skb
, key
, nla_data(a
));
1141 case OVS_ACTION_ATTR_POP_VLAN
:
1142 err
= pop_vlan(skb
, key
);
1145 case OVS_ACTION_ATTR_RECIRC
:
1146 err
= execute_recirc(dp
, skb
, key
, a
, rem
);
1147 if (nla_is_last(a
, rem
)) {
1148 /* If this is the last action, the skb has
1149 * been consumed or freed.
1150 * Return immediately.
1156 case OVS_ACTION_ATTR_SET
:
1157 err
= execute_set_action(skb
, key
, nla_data(a
));
1160 case OVS_ACTION_ATTR_SET_MASKED
:
1161 case OVS_ACTION_ATTR_SET_TO_MASKED
:
1162 err
= execute_masked_set_action(skb
, key
, nla_data(a
));
1165 case OVS_ACTION_ATTR_SAMPLE
:
1166 err
= sample(dp
, skb
, key
, a
, attr
, len
);
1169 case OVS_ACTION_ATTR_CT
:
1170 if (!is_flow_key_valid(key
)) {
1171 err
= ovs_flow_key_update(skb
, key
);
1176 err
= ovs_ct_execute(ovs_dp_get_net(dp
), skb
, key
,
1179 /* Hide stolen IP fragments from user space. */
1181 return err
== -EINPROGRESS
? 0 : err
;
1185 if (unlikely(err
)) {
1191 if (prev_port
!= -1)
1192 do_output(dp
, skb
, prev_port
, key
);
1199 static void process_deferred_actions(struct datapath
*dp
)
1201 struct action_fifo
*fifo
= this_cpu_ptr(action_fifos
);
1203 /* Do not touch the FIFO in case there is no deferred actions. */
1204 if (action_fifo_is_empty(fifo
))
1207 /* Finishing executing all deferred actions. */
1209 struct deferred_action
*da
= action_fifo_get(fifo
);
1210 struct sk_buff
*skb
= da
->skb
;
1211 struct sw_flow_key
*key
= &da
->pkt_key
;
1212 const struct nlattr
*actions
= da
->actions
;
1215 do_execute_actions(dp
, skb
, key
, actions
,
1218 ovs_dp_process_packet(skb
, key
);
1219 } while (!action_fifo_is_empty(fifo
));
1221 /* Reset FIFO for the next packet. */
1222 action_fifo_init(fifo
);
1225 /* Execute a list of actions against 'skb'. */
1226 int ovs_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
1227 const struct sw_flow_actions
*acts
,
1228 struct sw_flow_key
*key
)
1232 level
= __this_cpu_inc_return(exec_actions_level
);
1233 if (unlikely(level
> OVS_RECURSION_LIMIT
)) {
1234 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1241 err
= do_execute_actions(dp
, skb
, key
,
1242 acts
->actions
, acts
->actions_len
);
1245 process_deferred_actions(dp
);
1248 __this_cpu_dec(exec_actions_level
);
1252 int action_fifos_init(void)
1254 action_fifos
= alloc_percpu(struct action_fifo
);
1258 recirc_keys
= alloc_percpu(struct recirc_keys
);
1260 free_percpu(action_fifos
);
1267 void action_fifos_exit(void)
1269 free_percpu(action_fifos
);
1270 free_percpu(recirc_keys
);