2 * Copyright (c) 2007-2017 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/ip6_fib.h>
37 #include <net/checksum.h>
38 #include <net/dsfield.h>
40 #include <net/sctp/checksum.h>
44 #include "conntrack.h"
47 struct deferred_action
{
49 const struct nlattr
*actions
;
52 /* Store pkt_key clone when creating deferred action. */
53 struct sw_flow_key pkt_key
;
56 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
57 struct ovs_frag_data
{
61 __be16 inner_protocol
;
62 u16 network_offset
; /* valid only for MPLS */
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 5
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
];
82 struct action_flow_keys
{
83 struct sw_flow_key key
[OVS_DEFERRED_ACTION_THRESHOLD
];
86 static struct action_fifo __percpu
*action_fifos
;
87 static struct action_flow_keys __percpu
*flow_keys
;
88 static DEFINE_PER_CPU(int, exec_actions_level
);
90 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
91 * space. Return NULL if out of key spaces.
93 static struct sw_flow_key
*clone_key(const struct sw_flow_key
*key_
)
95 struct action_flow_keys
*keys
= this_cpu_ptr(flow_keys
);
96 int level
= this_cpu_read(exec_actions_level
);
97 struct sw_flow_key
*key
= NULL
;
99 if (level
<= OVS_DEFERRED_ACTION_THRESHOLD
) {
100 key
= &keys
->key
[level
- 1];
107 static void action_fifo_init(struct action_fifo
*fifo
)
113 static bool action_fifo_is_empty(const struct action_fifo
*fifo
)
115 return (fifo
->head
== fifo
->tail
);
118 static struct deferred_action
*action_fifo_get(struct action_fifo
*fifo
)
120 if (action_fifo_is_empty(fifo
))
123 return &fifo
->fifo
[fifo
->tail
++];
126 static struct deferred_action
*action_fifo_put(struct action_fifo
*fifo
)
128 if (fifo
->head
>= DEFERRED_ACTION_FIFO_SIZE
- 1)
131 return &fifo
->fifo
[fifo
->head
++];
134 /* Return true if fifo is not full */
135 static struct deferred_action
*add_deferred_actions(struct sk_buff
*skb
,
136 const struct sw_flow_key
*key
,
137 const struct nlattr
*actions
,
138 const int actions_len
)
140 struct action_fifo
*fifo
;
141 struct deferred_action
*da
;
143 fifo
= this_cpu_ptr(action_fifos
);
144 da
= action_fifo_put(fifo
);
147 da
->actions
= actions
;
148 da
->actions_len
= actions_len
;
155 static void invalidate_flow_key(struct sw_flow_key
*key
)
157 key
->mac_proto
|= SW_FLOW_KEY_INVALID
;
160 static bool is_flow_key_valid(const struct sw_flow_key
*key
)
162 return !(key
->mac_proto
& SW_FLOW_KEY_INVALID
);
165 static int clone_execute(struct datapath
*dp
, struct sk_buff
*skb
,
166 struct sw_flow_key
*key
,
168 const struct nlattr
*actions
, int len
,
169 bool last
, bool clone_flow_key
);
171 static void update_ethertype(struct sk_buff
*skb
, struct ethhdr
*hdr
,
174 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
175 __be16 diff
[] = { ~(hdr
->h_proto
), ethertype
};
177 skb
->csum
= ~csum_partial((char *)diff
, sizeof(diff
),
181 hdr
->h_proto
= ethertype
;
184 static int push_mpls(struct sk_buff
*skb
, struct sw_flow_key
*key
,
185 const struct ovs_action_push_mpls
*mpls
)
187 struct mpls_shim_hdr
*new_mpls_lse
;
189 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
190 if (skb
->encapsulation
)
193 if (skb_cow_head(skb
, MPLS_HLEN
) < 0)
196 if (!skb
->inner_protocol
) {
197 skb_set_inner_network_header(skb
, skb
->mac_len
);
198 skb_set_inner_protocol(skb
, skb
->protocol
);
201 skb_push(skb
, MPLS_HLEN
);
202 memmove(skb_mac_header(skb
) - MPLS_HLEN
, skb_mac_header(skb
),
204 skb_reset_mac_header(skb
);
205 skb_set_network_header(skb
, skb
->mac_len
);
207 new_mpls_lse
= mpls_hdr(skb
);
208 new_mpls_lse
->label_stack_entry
= mpls
->mpls_lse
;
210 skb_postpush_rcsum(skb
, new_mpls_lse
, MPLS_HLEN
);
212 if (ovs_key_mac_proto(key
) == MAC_PROTO_ETHERNET
)
213 update_ethertype(skb
, eth_hdr(skb
), mpls
->mpls_ethertype
);
214 skb
->protocol
= mpls
->mpls_ethertype
;
216 invalidate_flow_key(key
);
220 static int pop_mpls(struct sk_buff
*skb
, struct sw_flow_key
*key
,
221 const __be16 ethertype
)
225 err
= skb_ensure_writable(skb
, skb
->mac_len
+ MPLS_HLEN
);
229 skb_postpull_rcsum(skb
, mpls_hdr(skb
), MPLS_HLEN
);
231 memmove(skb_mac_header(skb
) + MPLS_HLEN
, skb_mac_header(skb
),
234 __skb_pull(skb
, MPLS_HLEN
);
235 skb_reset_mac_header(skb
);
236 skb_set_network_header(skb
, skb
->mac_len
);
238 if (ovs_key_mac_proto(key
) == MAC_PROTO_ETHERNET
) {
241 /* mpls_hdr() is used to locate the ethertype field correctly in the
242 * presence of VLAN tags.
244 hdr
= (struct ethhdr
*)((void *)mpls_hdr(skb
) - ETH_HLEN
);
245 update_ethertype(skb
, hdr
, ethertype
);
247 if (eth_p_mpls(skb
->protocol
))
248 skb
->protocol
= ethertype
;
250 invalidate_flow_key(key
);
254 static int set_mpls(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
255 const __be32
*mpls_lse
, const __be32
*mask
)
257 struct mpls_shim_hdr
*stack
;
261 err
= skb_ensure_writable(skb
, skb
->mac_len
+ MPLS_HLEN
);
265 stack
= mpls_hdr(skb
);
266 lse
= OVS_MASKED(stack
->label_stack_entry
, *mpls_lse
, *mask
);
267 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
268 __be32 diff
[] = { ~(stack
->label_stack_entry
), lse
};
270 skb
->csum
= ~csum_partial((char *)diff
, sizeof(diff
),
274 stack
->label_stack_entry
= lse
;
275 flow_key
->mpls
.top_lse
= lse
;
279 static int pop_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
)
283 err
= skb_vlan_pop(skb
);
284 if (skb_vlan_tag_present(skb
)) {
285 invalidate_flow_key(key
);
287 key
->eth
.vlan
.tci
= 0;
288 key
->eth
.vlan
.tpid
= 0;
293 static int push_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
,
294 const struct ovs_action_push_vlan
*vlan
)
296 if (skb_vlan_tag_present(skb
)) {
297 invalidate_flow_key(key
);
299 key
->eth
.vlan
.tci
= vlan
->vlan_tci
;
300 key
->eth
.vlan
.tpid
= vlan
->vlan_tpid
;
302 return skb_vlan_push(skb
, vlan
->vlan_tpid
,
303 ntohs(vlan
->vlan_tci
) & ~VLAN_TAG_PRESENT
);
306 /* 'src' is already properly masked. */
307 static void ether_addr_copy_masked(u8
*dst_
, const u8
*src_
, const u8
*mask_
)
309 u16
*dst
= (u16
*)dst_
;
310 const u16
*src
= (const u16
*)src_
;
311 const u16
*mask
= (const u16
*)mask_
;
313 OVS_SET_MASKED(dst
[0], src
[0], mask
[0]);
314 OVS_SET_MASKED(dst
[1], src
[1], mask
[1]);
315 OVS_SET_MASKED(dst
[2], src
[2], mask
[2]);
318 static int set_eth_addr(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
319 const struct ovs_key_ethernet
*key
,
320 const struct ovs_key_ethernet
*mask
)
324 err
= skb_ensure_writable(skb
, ETH_HLEN
);
328 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_ALEN
* 2);
330 ether_addr_copy_masked(eth_hdr(skb
)->h_source
, key
->eth_src
,
332 ether_addr_copy_masked(eth_hdr(skb
)->h_dest
, key
->eth_dst
,
335 skb_postpush_rcsum(skb
, eth_hdr(skb
), ETH_ALEN
* 2);
337 ether_addr_copy(flow_key
->eth
.src
, eth_hdr(skb
)->h_source
);
338 ether_addr_copy(flow_key
->eth
.dst
, eth_hdr(skb
)->h_dest
);
342 /* pop_eth does not support VLAN packets as this action is never called
345 static int pop_eth(struct sk_buff
*skb
, struct sw_flow_key
*key
)
347 skb_pull_rcsum(skb
, ETH_HLEN
);
348 skb_reset_mac_header(skb
);
349 skb_reset_mac_len(skb
);
351 /* safe right before invalidate_flow_key */
352 key
->mac_proto
= MAC_PROTO_NONE
;
353 invalidate_flow_key(key
);
357 static int push_eth(struct sk_buff
*skb
, struct sw_flow_key
*key
,
358 const struct ovs_action_push_eth
*ethh
)
362 /* Add the new Ethernet header */
363 if (skb_cow_head(skb
, ETH_HLEN
) < 0)
366 skb_push(skb
, ETH_HLEN
);
367 skb_reset_mac_header(skb
);
368 skb_reset_mac_len(skb
);
371 ether_addr_copy(hdr
->h_source
, ethh
->addresses
.eth_src
);
372 ether_addr_copy(hdr
->h_dest
, ethh
->addresses
.eth_dst
);
373 hdr
->h_proto
= skb
->protocol
;
375 skb_postpush_rcsum(skb
, hdr
, ETH_HLEN
);
377 /* safe right before invalidate_flow_key */
378 key
->mac_proto
= MAC_PROTO_ETHERNET
;
379 invalidate_flow_key(key
);
383 static void update_ip_l4_checksum(struct sk_buff
*skb
, struct iphdr
*nh
,
384 __be32 addr
, __be32 new_addr
)
386 int transport_len
= skb
->len
- skb_transport_offset(skb
);
388 if (nh
->frag_off
& htons(IP_OFFSET
))
391 if (nh
->protocol
== IPPROTO_TCP
) {
392 if (likely(transport_len
>= sizeof(struct tcphdr
)))
393 inet_proto_csum_replace4(&tcp_hdr(skb
)->check
, skb
,
394 addr
, new_addr
, true);
395 } else if (nh
->protocol
== IPPROTO_UDP
) {
396 if (likely(transport_len
>= sizeof(struct udphdr
))) {
397 struct udphdr
*uh
= udp_hdr(skb
);
399 if (uh
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
400 inet_proto_csum_replace4(&uh
->check
, skb
,
401 addr
, new_addr
, true);
403 uh
->check
= CSUM_MANGLED_0
;
409 static void set_ip_addr(struct sk_buff
*skb
, struct iphdr
*nh
,
410 __be32
*addr
, __be32 new_addr
)
412 update_ip_l4_checksum(skb
, nh
, *addr
, new_addr
);
413 csum_replace4(&nh
->check
, *addr
, new_addr
);
418 static void update_ipv6_checksum(struct sk_buff
*skb
, u8 l4_proto
,
419 __be32 addr
[4], const __be32 new_addr
[4])
421 int transport_len
= skb
->len
- skb_transport_offset(skb
);
423 if (l4_proto
== NEXTHDR_TCP
) {
424 if (likely(transport_len
>= sizeof(struct tcphdr
)))
425 inet_proto_csum_replace16(&tcp_hdr(skb
)->check
, skb
,
426 addr
, new_addr
, true);
427 } else if (l4_proto
== NEXTHDR_UDP
) {
428 if (likely(transport_len
>= sizeof(struct udphdr
))) {
429 struct udphdr
*uh
= udp_hdr(skb
);
431 if (uh
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
432 inet_proto_csum_replace16(&uh
->check
, skb
,
433 addr
, new_addr
, true);
435 uh
->check
= CSUM_MANGLED_0
;
438 } else if (l4_proto
== NEXTHDR_ICMP
) {
439 if (likely(transport_len
>= sizeof(struct icmp6hdr
)))
440 inet_proto_csum_replace16(&icmp6_hdr(skb
)->icmp6_cksum
,
441 skb
, addr
, new_addr
, true);
445 static void mask_ipv6_addr(const __be32 old
[4], const __be32 addr
[4],
446 const __be32 mask
[4], __be32 masked
[4])
448 masked
[0] = OVS_MASKED(old
[0], addr
[0], mask
[0]);
449 masked
[1] = OVS_MASKED(old
[1], addr
[1], mask
[1]);
450 masked
[2] = OVS_MASKED(old
[2], addr
[2], mask
[2]);
451 masked
[3] = OVS_MASKED(old
[3], addr
[3], mask
[3]);
454 static void set_ipv6_addr(struct sk_buff
*skb
, u8 l4_proto
,
455 __be32 addr
[4], const __be32 new_addr
[4],
456 bool recalculate_csum
)
458 if (recalculate_csum
)
459 update_ipv6_checksum(skb
, l4_proto
, addr
, new_addr
);
462 memcpy(addr
, new_addr
, sizeof(__be32
[4]));
465 static void set_ipv6_fl(struct ipv6hdr
*nh
, u32 fl
, u32 mask
)
467 /* Bits 21-24 are always unmasked, so this retains their values. */
468 OVS_SET_MASKED(nh
->flow_lbl
[0], (u8
)(fl
>> 16), (u8
)(mask
>> 16));
469 OVS_SET_MASKED(nh
->flow_lbl
[1], (u8
)(fl
>> 8), (u8
)(mask
>> 8));
470 OVS_SET_MASKED(nh
->flow_lbl
[2], (u8
)fl
, (u8
)mask
);
473 static void set_ip_ttl(struct sk_buff
*skb
, struct iphdr
*nh
, u8 new_ttl
,
476 new_ttl
= OVS_MASKED(nh
->ttl
, new_ttl
, mask
);
478 csum_replace2(&nh
->check
, htons(nh
->ttl
<< 8), htons(new_ttl
<< 8));
482 static int set_ipv4(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
483 const struct ovs_key_ipv4
*key
,
484 const struct ovs_key_ipv4
*mask
)
490 err
= skb_ensure_writable(skb
, skb_network_offset(skb
) +
491 sizeof(struct iphdr
));
497 /* Setting an IP addresses is typically only a side effect of
498 * matching on them in the current userspace implementation, so it
499 * makes sense to check if the value actually changed.
501 if (mask
->ipv4_src
) {
502 new_addr
= OVS_MASKED(nh
->saddr
, key
->ipv4_src
, mask
->ipv4_src
);
504 if (unlikely(new_addr
!= nh
->saddr
)) {
505 set_ip_addr(skb
, nh
, &nh
->saddr
, new_addr
);
506 flow_key
->ipv4
.addr
.src
= new_addr
;
509 if (mask
->ipv4_dst
) {
510 new_addr
= OVS_MASKED(nh
->daddr
, key
->ipv4_dst
, mask
->ipv4_dst
);
512 if (unlikely(new_addr
!= nh
->daddr
)) {
513 set_ip_addr(skb
, nh
, &nh
->daddr
, new_addr
);
514 flow_key
->ipv4
.addr
.dst
= new_addr
;
517 if (mask
->ipv4_tos
) {
518 ipv4_change_dsfield(nh
, ~mask
->ipv4_tos
, key
->ipv4_tos
);
519 flow_key
->ip
.tos
= nh
->tos
;
521 if (mask
->ipv4_ttl
) {
522 set_ip_ttl(skb
, nh
, key
->ipv4_ttl
, mask
->ipv4_ttl
);
523 flow_key
->ip
.ttl
= nh
->ttl
;
529 static bool is_ipv6_mask_nonzero(const __be32 addr
[4])
531 return !!(addr
[0] | addr
[1] | addr
[2] | addr
[3]);
534 static int set_ipv6(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
535 const struct ovs_key_ipv6
*key
,
536 const struct ovs_key_ipv6
*mask
)
541 err
= skb_ensure_writable(skb
, skb_network_offset(skb
) +
542 sizeof(struct ipv6hdr
));
548 /* Setting an IP addresses is typically only a side effect of
549 * matching on them in the current userspace implementation, so it
550 * makes sense to check if the value actually changed.
552 if (is_ipv6_mask_nonzero(mask
->ipv6_src
)) {
553 __be32
*saddr
= (__be32
*)&nh
->saddr
;
556 mask_ipv6_addr(saddr
, key
->ipv6_src
, mask
->ipv6_src
, masked
);
558 if (unlikely(memcmp(saddr
, masked
, sizeof(masked
)))) {
559 set_ipv6_addr(skb
, flow_key
->ip
.proto
, saddr
, masked
,
561 memcpy(&flow_key
->ipv6
.addr
.src
, masked
,
562 sizeof(flow_key
->ipv6
.addr
.src
));
565 if (is_ipv6_mask_nonzero(mask
->ipv6_dst
)) {
566 unsigned int offset
= 0;
567 int flags
= IP6_FH_F_SKIP_RH
;
568 bool recalc_csum
= true;
569 __be32
*daddr
= (__be32
*)&nh
->daddr
;
572 mask_ipv6_addr(daddr
, key
->ipv6_dst
, mask
->ipv6_dst
, masked
);
574 if (unlikely(memcmp(daddr
, masked
, sizeof(masked
)))) {
575 if (ipv6_ext_hdr(nh
->nexthdr
))
576 recalc_csum
= (ipv6_find_hdr(skb
, &offset
,
581 set_ipv6_addr(skb
, flow_key
->ip
.proto
, daddr
, masked
,
583 memcpy(&flow_key
->ipv6
.addr
.dst
, masked
,
584 sizeof(flow_key
->ipv6
.addr
.dst
));
587 if (mask
->ipv6_tclass
) {
588 ipv6_change_dsfield(nh
, ~mask
->ipv6_tclass
, key
->ipv6_tclass
);
589 flow_key
->ip
.tos
= ipv6_get_dsfield(nh
);
591 if (mask
->ipv6_label
) {
592 set_ipv6_fl(nh
, ntohl(key
->ipv6_label
),
593 ntohl(mask
->ipv6_label
));
594 flow_key
->ipv6
.label
=
595 *(__be32
*)nh
& htonl(IPV6_FLOWINFO_FLOWLABEL
);
597 if (mask
->ipv6_hlimit
) {
598 OVS_SET_MASKED(nh
->hop_limit
, key
->ipv6_hlimit
,
600 flow_key
->ip
.ttl
= nh
->hop_limit
;
605 /* Must follow skb_ensure_writable() since that can move the skb data. */
606 static void set_tp_port(struct sk_buff
*skb
, __be16
*port
,
607 __be16 new_port
, __sum16
*check
)
609 inet_proto_csum_replace2(check
, skb
, *port
, new_port
, false);
613 static int set_udp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
614 const struct ovs_key_udp
*key
,
615 const struct ovs_key_udp
*mask
)
621 err
= skb_ensure_writable(skb
, skb_transport_offset(skb
) +
622 sizeof(struct udphdr
));
627 /* Either of the masks is non-zero, so do not bother checking them. */
628 src
= OVS_MASKED(uh
->source
, key
->udp_src
, mask
->udp_src
);
629 dst
= OVS_MASKED(uh
->dest
, key
->udp_dst
, mask
->udp_dst
);
631 if (uh
->check
&& skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
632 if (likely(src
!= uh
->source
)) {
633 set_tp_port(skb
, &uh
->source
, src
, &uh
->check
);
634 flow_key
->tp
.src
= src
;
636 if (likely(dst
!= uh
->dest
)) {
637 set_tp_port(skb
, &uh
->dest
, dst
, &uh
->check
);
638 flow_key
->tp
.dst
= dst
;
641 if (unlikely(!uh
->check
))
642 uh
->check
= CSUM_MANGLED_0
;
646 flow_key
->tp
.src
= src
;
647 flow_key
->tp
.dst
= dst
;
655 static int set_tcp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
656 const struct ovs_key_tcp
*key
,
657 const struct ovs_key_tcp
*mask
)
663 err
= skb_ensure_writable(skb
, skb_transport_offset(skb
) +
664 sizeof(struct tcphdr
));
669 src
= OVS_MASKED(th
->source
, key
->tcp_src
, mask
->tcp_src
);
670 if (likely(src
!= th
->source
)) {
671 set_tp_port(skb
, &th
->source
, src
, &th
->check
);
672 flow_key
->tp
.src
= src
;
674 dst
= OVS_MASKED(th
->dest
, key
->tcp_dst
, mask
->tcp_dst
);
675 if (likely(dst
!= th
->dest
)) {
676 set_tp_port(skb
, &th
->dest
, dst
, &th
->check
);
677 flow_key
->tp
.dst
= dst
;
684 static int set_sctp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
685 const struct ovs_key_sctp
*key
,
686 const struct ovs_key_sctp
*mask
)
688 unsigned int sctphoff
= skb_transport_offset(skb
);
690 __le32 old_correct_csum
, new_csum
, old_csum
;
693 err
= skb_ensure_writable(skb
, sctphoff
+ sizeof(struct sctphdr
));
698 old_csum
= sh
->checksum
;
699 old_correct_csum
= sctp_compute_cksum(skb
, sctphoff
);
701 sh
->source
= OVS_MASKED(sh
->source
, key
->sctp_src
, mask
->sctp_src
);
702 sh
->dest
= OVS_MASKED(sh
->dest
, key
->sctp_dst
, mask
->sctp_dst
);
704 new_csum
= sctp_compute_cksum(skb
, sctphoff
);
706 /* Carry any checksum errors through. */
707 sh
->checksum
= old_csum
^ old_correct_csum
^ new_csum
;
710 flow_key
->tp
.src
= sh
->source
;
711 flow_key
->tp
.dst
= sh
->dest
;
716 static int ovs_vport_output(struct net
*net
, struct sock
*sk
, struct sk_buff
*skb
)
718 struct ovs_frag_data
*data
= this_cpu_ptr(&ovs_frag_data_storage
);
719 struct vport
*vport
= data
->vport
;
721 if (skb_cow_head(skb
, data
->l2_len
) < 0) {
726 __skb_dst_copy(skb
, data
->dst
);
727 *OVS_CB(skb
) = data
->cb
;
728 skb
->inner_protocol
= data
->inner_protocol
;
729 skb
->vlan_tci
= data
->vlan_tci
;
730 skb
->vlan_proto
= data
->vlan_proto
;
732 /* Reconstruct the MAC header. */
733 skb_push(skb
, data
->l2_len
);
734 memcpy(skb
->data
, &data
->l2_data
, data
->l2_len
);
735 skb_postpush_rcsum(skb
, skb
->data
, data
->l2_len
);
736 skb_reset_mac_header(skb
);
738 if (eth_p_mpls(skb
->protocol
)) {
739 skb
->inner_network_header
= skb
->network_header
;
740 skb_set_network_header(skb
, data
->network_offset
);
741 skb_reset_mac_len(skb
);
744 ovs_vport_send(vport
, skb
, data
->mac_proto
);
749 ovs_dst_get_mtu(const struct dst_entry
*dst
)
751 return dst
->dev
->mtu
;
754 static struct dst_ops ovs_dst_ops
= {
756 .mtu
= ovs_dst_get_mtu
,
759 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
760 * ovs_vport_output(), which is called once per fragmented packet.
762 static void prepare_frag(struct vport
*vport
, struct sk_buff
*skb
,
763 u16 orig_network_offset
, u8 mac_proto
)
765 unsigned int hlen
= skb_network_offset(skb
);
766 struct ovs_frag_data
*data
;
768 data
= this_cpu_ptr(&ovs_frag_data_storage
);
769 data
->dst
= skb
->_skb_refdst
;
771 data
->cb
= *OVS_CB(skb
);
772 data
->inner_protocol
= skb
->inner_protocol
;
773 data
->network_offset
= orig_network_offset
;
774 data
->vlan_tci
= skb
->vlan_tci
;
775 data
->vlan_proto
= skb
->vlan_proto
;
776 data
->mac_proto
= mac_proto
;
778 memcpy(&data
->l2_data
, skb
->data
, hlen
);
780 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
784 static void ovs_fragment(struct net
*net
, struct vport
*vport
,
785 struct sk_buff
*skb
, u16 mru
,
786 struct sw_flow_key
*key
)
788 u16 orig_network_offset
= 0;
790 if (eth_p_mpls(skb
->protocol
)) {
791 orig_network_offset
= skb_network_offset(skb
);
792 skb
->network_header
= skb
->inner_network_header
;
795 if (skb_network_offset(skb
) > MAX_L2_LEN
) {
796 OVS_NLERR(1, "L2 header too long to fragment");
800 if (key
->eth
.type
== htons(ETH_P_IP
)) {
801 struct dst_entry ovs_dst
;
802 unsigned long orig_dst
;
804 prepare_frag(vport
, skb
, orig_network_offset
,
805 ovs_key_mac_proto(key
));
806 dst_init(&ovs_dst
, &ovs_dst_ops
, NULL
, 1,
807 DST_OBSOLETE_NONE
, DST_NOCOUNT
);
808 ovs_dst
.dev
= vport
->dev
;
810 orig_dst
= skb
->_skb_refdst
;
811 skb_dst_set_noref(skb
, &ovs_dst
);
812 IPCB(skb
)->frag_max_size
= mru
;
814 ip_do_fragment(net
, skb
->sk
, skb
, ovs_vport_output
);
815 refdst_drop(orig_dst
);
816 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
817 const struct nf_ipv6_ops
*v6ops
= nf_get_ipv6_ops();
818 unsigned long orig_dst
;
819 struct rt6_info ovs_rt
;
824 prepare_frag(vport
, skb
, orig_network_offset
,
825 ovs_key_mac_proto(key
));
826 memset(&ovs_rt
, 0, sizeof(ovs_rt
));
827 dst_init(&ovs_rt
.dst
, &ovs_dst_ops
, NULL
, 1,
828 DST_OBSOLETE_NONE
, DST_NOCOUNT
);
829 ovs_rt
.dst
.dev
= vport
->dev
;
831 orig_dst
= skb
->_skb_refdst
;
832 skb_dst_set_noref(skb
, &ovs_rt
.dst
);
833 IP6CB(skb
)->frag_max_size
= mru
;
835 v6ops
->fragment(net
, skb
->sk
, skb
, ovs_vport_output
);
836 refdst_drop(orig_dst
);
838 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
839 ovs_vport_name(vport
), ntohs(key
->eth
.type
), mru
,
849 static void do_output(struct datapath
*dp
, struct sk_buff
*skb
, int out_port
,
850 struct sw_flow_key
*key
)
852 struct vport
*vport
= ovs_vport_rcu(dp
, out_port
);
855 u16 mru
= OVS_CB(skb
)->mru
;
856 u32 cutlen
= OVS_CB(skb
)->cutlen
;
858 if (unlikely(cutlen
> 0)) {
859 if (skb
->len
- cutlen
> ovs_mac_header_len(key
))
860 pskb_trim(skb
, skb
->len
- cutlen
);
862 pskb_trim(skb
, ovs_mac_header_len(key
));
866 (skb
->len
<= mru
+ vport
->dev
->hard_header_len
))) {
867 ovs_vport_send(vport
, skb
, ovs_key_mac_proto(key
));
868 } else if (mru
<= vport
->dev
->mtu
) {
869 struct net
*net
= read_pnet(&dp
->net
);
871 ovs_fragment(net
, vport
, skb
, mru
, key
);
880 static int output_userspace(struct datapath
*dp
, struct sk_buff
*skb
,
881 struct sw_flow_key
*key
, const struct nlattr
*attr
,
882 const struct nlattr
*actions
, int actions_len
,
885 struct dp_upcall_info upcall
;
886 const struct nlattr
*a
;
889 memset(&upcall
, 0, sizeof(upcall
));
890 upcall
.cmd
= OVS_PACKET_CMD_ACTION
;
891 upcall
.mru
= OVS_CB(skb
)->mru
;
893 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
894 a
= nla_next(a
, &rem
)) {
895 switch (nla_type(a
)) {
896 case OVS_USERSPACE_ATTR_USERDATA
:
900 case OVS_USERSPACE_ATTR_PID
:
901 upcall
.portid
= nla_get_u32(a
);
904 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT
: {
905 /* Get out tunnel info. */
908 vport
= ovs_vport_rcu(dp
, nla_get_u32(a
));
912 err
= dev_fill_metadata_dst(vport
->dev
, skb
);
914 upcall
.egress_tun_info
= skb_tunnel_info(skb
);
920 case OVS_USERSPACE_ATTR_ACTIONS
: {
921 /* Include actions. */
922 upcall
.actions
= actions
;
923 upcall
.actions_len
= actions_len
;
927 } /* End of switch. */
930 return ovs_dp_upcall(dp
, skb
, key
, &upcall
, cutlen
);
933 /* When 'last' is true, sample() should always consume the 'skb'.
934 * Otherwise, sample() should keep 'skb' intact regardless what
935 * actions are executed within sample().
937 static int sample(struct datapath
*dp
, struct sk_buff
*skb
,
938 struct sw_flow_key
*key
, const struct nlattr
*attr
,
941 struct nlattr
*actions
;
942 struct nlattr
*sample_arg
;
943 int rem
= nla_len(attr
);
944 const struct sample_arg
*arg
;
947 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
948 sample_arg
= nla_data(attr
);
949 arg
= nla_data(sample_arg
);
950 actions
= nla_next(sample_arg
, &rem
);
952 if ((arg
->probability
!= U32_MAX
) &&
953 (!arg
->probability
|| prandom_u32() > arg
->probability
)) {
959 clone_flow_key
= !arg
->exec
;
960 return clone_execute(dp
, skb
, key
, 0, actions
, rem
, last
,
964 static void execute_hash(struct sk_buff
*skb
, struct sw_flow_key
*key
,
965 const struct nlattr
*attr
)
967 struct ovs_action_hash
*hash_act
= nla_data(attr
);
970 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
971 hash
= skb_get_hash(skb
);
972 hash
= jhash_1word(hash
, hash_act
->hash_basis
);
976 key
->ovs_flow_hash
= hash
;
979 static int execute_set_action(struct sk_buff
*skb
,
980 struct sw_flow_key
*flow_key
,
981 const struct nlattr
*a
)
983 /* Only tunnel set execution is supported without a mask. */
984 if (nla_type(a
) == OVS_KEY_ATTR_TUNNEL_INFO
) {
985 struct ovs_tunnel_info
*tun
= nla_data(a
);
988 dst_hold((struct dst_entry
*)tun
->tun_dst
);
989 skb_dst_set(skb
, (struct dst_entry
*)tun
->tun_dst
);
996 /* Mask is at the midpoint of the data. */
997 #define get_mask(a, type) ((const type)nla_data(a) + 1)
999 static int execute_masked_set_action(struct sk_buff
*skb
,
1000 struct sw_flow_key
*flow_key
,
1001 const struct nlattr
*a
)
1005 switch (nla_type(a
)) {
1006 case OVS_KEY_ATTR_PRIORITY
:
1007 OVS_SET_MASKED(skb
->priority
, nla_get_u32(a
),
1008 *get_mask(a
, u32
*));
1009 flow_key
->phy
.priority
= skb
->priority
;
1012 case OVS_KEY_ATTR_SKB_MARK
:
1013 OVS_SET_MASKED(skb
->mark
, nla_get_u32(a
), *get_mask(a
, u32
*));
1014 flow_key
->phy
.skb_mark
= skb
->mark
;
1017 case OVS_KEY_ATTR_TUNNEL_INFO
:
1018 /* Masked data not supported for tunnel. */
1022 case OVS_KEY_ATTR_ETHERNET
:
1023 err
= set_eth_addr(skb
, flow_key
, nla_data(a
),
1024 get_mask(a
, struct ovs_key_ethernet
*));
1027 case OVS_KEY_ATTR_IPV4
:
1028 err
= set_ipv4(skb
, flow_key
, nla_data(a
),
1029 get_mask(a
, struct ovs_key_ipv4
*));
1032 case OVS_KEY_ATTR_IPV6
:
1033 err
= set_ipv6(skb
, flow_key
, nla_data(a
),
1034 get_mask(a
, struct ovs_key_ipv6
*));
1037 case OVS_KEY_ATTR_TCP
:
1038 err
= set_tcp(skb
, flow_key
, nla_data(a
),
1039 get_mask(a
, struct ovs_key_tcp
*));
1042 case OVS_KEY_ATTR_UDP
:
1043 err
= set_udp(skb
, flow_key
, nla_data(a
),
1044 get_mask(a
, struct ovs_key_udp
*));
1047 case OVS_KEY_ATTR_SCTP
:
1048 err
= set_sctp(skb
, flow_key
, nla_data(a
),
1049 get_mask(a
, struct ovs_key_sctp
*));
1052 case OVS_KEY_ATTR_MPLS
:
1053 err
= set_mpls(skb
, flow_key
, nla_data(a
), get_mask(a
,
1057 case OVS_KEY_ATTR_CT_STATE
:
1058 case OVS_KEY_ATTR_CT_ZONE
:
1059 case OVS_KEY_ATTR_CT_MARK
:
1060 case OVS_KEY_ATTR_CT_LABELS
:
1061 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4
:
1062 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6
:
1070 static int execute_recirc(struct datapath
*dp
, struct sk_buff
*skb
,
1071 struct sw_flow_key
*key
,
1072 const struct nlattr
*a
, bool last
)
1076 if (!is_flow_key_valid(key
)) {
1079 err
= ovs_flow_key_update(skb
, key
);
1083 BUG_ON(!is_flow_key_valid(key
));
1085 recirc_id
= nla_get_u32(a
);
1086 return clone_execute(dp
, skb
, key
, recirc_id
, NULL
, 0, last
, true);
1089 /* Execute a list of actions against 'skb'. */
1090 static int do_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
1091 struct sw_flow_key
*key
,
1092 const struct nlattr
*attr
, int len
)
1094 const struct nlattr
*a
;
1097 for (a
= attr
, rem
= len
; rem
> 0;
1098 a
= nla_next(a
, &rem
)) {
1101 switch (nla_type(a
)) {
1102 case OVS_ACTION_ATTR_OUTPUT
: {
1103 int port
= nla_get_u32(a
);
1104 struct sk_buff
*clone
;
1106 /* Every output action needs a separate clone
1107 * of 'skb', In case the output action is the
1108 * last action, cloning can be avoided.
1110 if (nla_is_last(a
, rem
)) {
1111 do_output(dp
, skb
, port
, key
);
1112 /* 'skb' has been used for output.
1117 clone
= skb_clone(skb
, GFP_ATOMIC
);
1119 do_output(dp
, clone
, port
, key
);
1120 OVS_CB(skb
)->cutlen
= 0;
1124 case OVS_ACTION_ATTR_TRUNC
: {
1125 struct ovs_action_trunc
*trunc
= nla_data(a
);
1127 if (skb
->len
> trunc
->max_len
)
1128 OVS_CB(skb
)->cutlen
= skb
->len
- trunc
->max_len
;
1132 case OVS_ACTION_ATTR_USERSPACE
:
1133 output_userspace(dp
, skb
, key
, a
, attr
,
1134 len
, OVS_CB(skb
)->cutlen
);
1135 OVS_CB(skb
)->cutlen
= 0;
1138 case OVS_ACTION_ATTR_HASH
:
1139 execute_hash(skb
, key
, a
);
1142 case OVS_ACTION_ATTR_PUSH_MPLS
:
1143 err
= push_mpls(skb
, key
, nla_data(a
));
1146 case OVS_ACTION_ATTR_POP_MPLS
:
1147 err
= pop_mpls(skb
, key
, nla_get_be16(a
));
1150 case OVS_ACTION_ATTR_PUSH_VLAN
:
1151 err
= push_vlan(skb
, key
, nla_data(a
));
1154 case OVS_ACTION_ATTR_POP_VLAN
:
1155 err
= pop_vlan(skb
, key
);
1158 case OVS_ACTION_ATTR_RECIRC
: {
1159 bool last
= nla_is_last(a
, rem
);
1161 err
= execute_recirc(dp
, skb
, key
, a
, last
);
1163 /* If this is the last action, the skb has
1164 * been consumed or freed.
1165 * Return immediately.
1172 case OVS_ACTION_ATTR_SET
:
1173 err
= execute_set_action(skb
, key
, nla_data(a
));
1176 case OVS_ACTION_ATTR_SET_MASKED
:
1177 case OVS_ACTION_ATTR_SET_TO_MASKED
:
1178 err
= execute_masked_set_action(skb
, key
, nla_data(a
));
1181 case OVS_ACTION_ATTR_SAMPLE
: {
1182 bool last
= nla_is_last(a
, rem
);
1184 err
= sample(dp
, skb
, key
, a
, last
);
1191 case OVS_ACTION_ATTR_CT
:
1192 if (!is_flow_key_valid(key
)) {
1193 err
= ovs_flow_key_update(skb
, key
);
1198 err
= ovs_ct_execute(ovs_dp_get_net(dp
), skb
, key
,
1201 /* Hide stolen IP fragments from user space. */
1203 return err
== -EINPROGRESS
? 0 : err
;
1206 case OVS_ACTION_ATTR_PUSH_ETH
:
1207 err
= push_eth(skb
, key
, nla_data(a
));
1210 case OVS_ACTION_ATTR_POP_ETH
:
1211 err
= pop_eth(skb
, key
);
1215 if (unlikely(err
)) {
1225 /* Execute the actions on the clone of the packet. The effect of the
1226 * execution does not affect the original 'skb' nor the original 'key'.
1228 * The execution may be deferred in case the actions can not be executed
1231 static int clone_execute(struct datapath
*dp
, struct sk_buff
*skb
,
1232 struct sw_flow_key
*key
, u32 recirc_id
,
1233 const struct nlattr
*actions
, int len
,
1234 bool last
, bool clone_flow_key
)
1236 struct deferred_action
*da
;
1237 struct sw_flow_key
*clone
;
1239 skb
= last
? skb
: skb_clone(skb
, GFP_ATOMIC
);
1241 /* Out of memory, skip this action.
1246 /* When clone_flow_key is false, the 'key' will not be change
1247 * by the actions, then the 'key' can be used directly.
1248 * Otherwise, try to clone key from the next recursion level of
1249 * 'flow_keys'. If clone is successful, execute the actions
1250 * without deferring.
1252 clone
= clone_flow_key
? clone_key(key
) : key
;
1256 if (actions
) { /* Sample action */
1258 __this_cpu_inc(exec_actions_level
);
1260 err
= do_execute_actions(dp
, skb
, clone
,
1264 __this_cpu_dec(exec_actions_level
);
1265 } else { /* Recirc action */
1266 clone
->recirc_id
= recirc_id
;
1267 ovs_dp_process_packet(skb
, clone
);
1272 /* Out of 'flow_keys' space. Defer actions */
1273 da
= add_deferred_actions(skb
, key
, actions
, len
);
1275 if (!actions
) { /* Recirc action */
1277 key
->recirc_id
= recirc_id
;
1280 /* Out of per CPU action FIFO space. Drop the 'skb' and
1285 if (net_ratelimit()) {
1286 if (actions
) { /* Sample action */
1287 pr_warn("%s: deferred action limit reached, drop sample action\n",
1289 } else { /* Recirc action */
1290 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1298 static void process_deferred_actions(struct datapath
*dp
)
1300 struct action_fifo
*fifo
= this_cpu_ptr(action_fifos
);
1302 /* Do not touch the FIFO in case there is no deferred actions. */
1303 if (action_fifo_is_empty(fifo
))
1306 /* Finishing executing all deferred actions. */
1308 struct deferred_action
*da
= action_fifo_get(fifo
);
1309 struct sk_buff
*skb
= da
->skb
;
1310 struct sw_flow_key
*key
= &da
->pkt_key
;
1311 const struct nlattr
*actions
= da
->actions
;
1312 int actions_len
= da
->actions_len
;
1315 do_execute_actions(dp
, skb
, key
, actions
, actions_len
);
1317 ovs_dp_process_packet(skb
, key
);
1318 } while (!action_fifo_is_empty(fifo
));
1320 /* Reset FIFO for the next packet. */
1321 action_fifo_init(fifo
);
1324 /* Execute a list of actions against 'skb'. */
1325 int ovs_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
1326 const struct sw_flow_actions
*acts
,
1327 struct sw_flow_key
*key
)
1331 level
= __this_cpu_inc_return(exec_actions_level
);
1332 if (unlikely(level
> OVS_RECURSION_LIMIT
)) {
1333 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1340 OVS_CB(skb
)->acts_origlen
= acts
->orig_len
;
1341 err
= do_execute_actions(dp
, skb
, key
,
1342 acts
->actions
, acts
->actions_len
);
1345 process_deferred_actions(dp
);
1348 __this_cpu_dec(exec_actions_level
);
1352 int action_fifos_init(void)
1354 action_fifos
= alloc_percpu(struct action_fifo
);
1358 flow_keys
= alloc_percpu(struct action_flow_keys
);
1360 free_percpu(action_fifos
);
1367 void action_fifos_exit(void)
1369 free_percpu(action_fifos
);
1370 free_percpu(flow_keys
);