2 * Copyright (c) 2007-2014 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/sctp.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/in6.h>
29 #include <linux/if_arp.h>
30 #include <linux/if_vlan.h>
33 #include <net/checksum.h>
34 #include <net/dsfield.h>
36 #include <net/sctp/checksum.h>
43 static int do_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
44 struct sw_flow_key
*key
,
45 const struct nlattr
*attr
, int len
);
47 struct deferred_action
{
49 const struct nlattr
*actions
;
51 /* Store pkt_key clone when creating deferred action. */
52 struct sw_flow_key pkt_key
;
55 #define DEFERRED_ACTION_FIFO_SIZE 10
59 /* Deferred action fifo queue storage. */
60 struct deferred_action fifo
[DEFERRED_ACTION_FIFO_SIZE
];
63 static struct action_fifo __percpu
*action_fifos
;
64 #define EXEC_ACTIONS_LEVEL_LIMIT 4 /* limit used to detect packet
65 * looping by the network stack
67 static DEFINE_PER_CPU(int, exec_actions_level
);
69 static void action_fifo_init(struct action_fifo
*fifo
)
75 static bool action_fifo_is_empty(const struct action_fifo
*fifo
)
77 return (fifo
->head
== fifo
->tail
);
80 static struct deferred_action
*action_fifo_get(struct action_fifo
*fifo
)
82 if (action_fifo_is_empty(fifo
))
85 return &fifo
->fifo
[fifo
->tail
++];
88 static struct deferred_action
*action_fifo_put(struct action_fifo
*fifo
)
90 if (fifo
->head
>= DEFERRED_ACTION_FIFO_SIZE
- 1)
93 return &fifo
->fifo
[fifo
->head
++];
96 /* Return queue entry if fifo is not full */
97 static struct deferred_action
*add_deferred_actions(struct sk_buff
*skb
,
98 const struct sw_flow_key
*key
,
99 const struct nlattr
*attr
)
101 struct action_fifo
*fifo
;
102 struct deferred_action
*da
;
104 fifo
= this_cpu_ptr(action_fifos
);
105 da
= action_fifo_put(fifo
);
115 static void invalidate_flow_key(struct sw_flow_key
*key
)
117 key
->eth
.type
= htons(0);
120 static bool is_flow_key_valid(const struct sw_flow_key
*key
)
122 return !!key
->eth
.type
;
125 static int push_mpls(struct sk_buff
*skb
, struct sw_flow_key
*key
,
126 const struct ovs_action_push_mpls
*mpls
)
128 __be32
*new_mpls_lse
;
131 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
132 if (skb_encapsulation(skb
))
135 if (skb_cow_head(skb
, MPLS_HLEN
) < 0)
138 skb_push(skb
, MPLS_HLEN
);
139 memmove(skb_mac_header(skb
) - MPLS_HLEN
, skb_mac_header(skb
),
141 skb_reset_mac_header(skb
);
143 new_mpls_lse
= (__be32
*)skb_mpls_header(skb
);
144 *new_mpls_lse
= mpls
->mpls_lse
;
146 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
147 skb
->csum
= csum_add(skb
->csum
, csum_partial(new_mpls_lse
,
151 hdr
->h_proto
= mpls
->mpls_ethertype
;
152 if (!ovs_skb_get_inner_protocol(skb
))
153 ovs_skb_set_inner_protocol(skb
, skb
->protocol
);
154 skb
->protocol
= mpls
->mpls_ethertype
;
156 invalidate_flow_key(key
);
160 static int pop_mpls(struct sk_buff
*skb
, struct sw_flow_key
*key
,
161 const __be16 ethertype
)
166 err
= skb_ensure_writable(skb
, skb
->mac_len
+ MPLS_HLEN
);
170 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
171 skb
->csum
= csum_sub(skb
->csum
,
172 csum_partial(skb_mpls_header(skb
),
175 memmove(skb_mac_header(skb
) + MPLS_HLEN
, skb_mac_header(skb
),
178 __skb_pull(skb
, MPLS_HLEN
);
179 skb_reset_mac_header(skb
);
181 /* skb_mpls_header() is used to locate the ethertype
182 * field correctly in the presence of VLAN tags.
184 hdr
= (struct ethhdr
*)(skb_mpls_header(skb
) - ETH_HLEN
);
185 hdr
->h_proto
= ethertype
;
186 if (eth_p_mpls(skb
->protocol
))
187 skb
->protocol
= ethertype
;
189 invalidate_flow_key(key
);
193 /* 'KEY' must not have any bits set outside of the 'MASK' */
194 #define MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK)))
195 #define SET_MASKED(OLD, KEY, MASK) ((OLD) = MASKED(OLD, KEY, MASK))
197 static int set_mpls(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
198 const __be32
*mpls_lse
, const __be32
*mask
)
204 err
= skb_ensure_writable(skb
, skb
->mac_len
+ MPLS_HLEN
);
208 stack
= (__be32
*)skb_mpls_header(skb
);
209 lse
= MASKED(*stack
, *mpls_lse
, *mask
);
210 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
211 __be32 diff
[] = { ~(*stack
), lse
};
213 skb
->csum
= ~csum_partial((char *)diff
, sizeof(diff
),
218 flow_key
->mpls
.top_lse
= lse
;
222 static int pop_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
)
226 err
= skb_vlan_pop(skb
);
227 if (skb_vlan_tag_present(skb
))
228 invalidate_flow_key(key
);
235 static int push_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
,
236 const struct ovs_action_push_vlan
*vlan
)
238 if (skb_vlan_tag_present(skb
))
239 invalidate_flow_key(key
);
241 key
->eth
.tci
= vlan
->vlan_tci
;
243 return skb_vlan_push(skb
, vlan
->vlan_tpid
,
244 ntohs(vlan
->vlan_tci
) & ~VLAN_TAG_PRESENT
);
247 /* 'src' is already properly masked. */
248 static void ether_addr_copy_masked(u8
*dst_
, const u8
*src_
, const u8
*mask_
)
250 u16
*dst
= (u16
*)dst_
;
251 const u16
*src
= (const u16
*)src_
;
252 const u16
*mask
= (const u16
*)mask_
;
254 SET_MASKED(dst
[0], src
[0], mask
[0]);
255 SET_MASKED(dst
[1], src
[1], mask
[1]);
256 SET_MASKED(dst
[2], src
[2], mask
[2]);
259 static int set_eth_addr(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
260 const struct ovs_key_ethernet
*key
,
261 const struct ovs_key_ethernet
*mask
)
265 err
= skb_ensure_writable(skb
, ETH_HLEN
);
269 skb_postpull_rcsum(skb
, eth_hdr(skb
), ETH_ALEN
* 2);
271 ether_addr_copy_masked(eth_hdr(skb
)->h_source
, key
->eth_src
,
273 ether_addr_copy_masked(eth_hdr(skb
)->h_dest
, key
->eth_dst
,
276 ovs_skb_postpush_rcsum(skb
, eth_hdr(skb
), ETH_ALEN
* 2);
278 ether_addr_copy(flow_key
->eth
.src
, eth_hdr(skb
)->h_source
);
279 ether_addr_copy(flow_key
->eth
.dst
, eth_hdr(skb
)->h_dest
);
283 static void set_ip_addr(struct sk_buff
*skb
, struct iphdr
*nh
,
284 __be32
*addr
, __be32 new_addr
)
286 int transport_len
= skb
->len
- skb_transport_offset(skb
);
288 if (nh
->protocol
== IPPROTO_TCP
) {
289 if (likely(transport_len
>= sizeof(struct tcphdr
)))
290 inet_proto_csum_replace4(&tcp_hdr(skb
)->check
, skb
,
292 } else if (nh
->protocol
== IPPROTO_UDP
) {
293 if (likely(transport_len
>= sizeof(struct udphdr
))) {
294 struct udphdr
*uh
= udp_hdr(skb
);
296 if (uh
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
297 inet_proto_csum_replace4(&uh
->check
, skb
,
300 uh
->check
= CSUM_MANGLED_0
;
305 csum_replace4(&nh
->check
, *addr
, new_addr
);
310 static void update_ipv6_checksum(struct sk_buff
*skb
, u8 l4_proto
,
311 __be32 addr
[4], const __be32 new_addr
[4])
313 int transport_len
= skb
->len
- skb_transport_offset(skb
);
315 if (l4_proto
== NEXTHDR_TCP
) {
316 if (likely(transport_len
>= sizeof(struct tcphdr
)))
317 inet_proto_csum_replace16(&tcp_hdr(skb
)->check
, skb
,
319 } else if (l4_proto
== NEXTHDR_UDP
) {
320 if (likely(transport_len
>= sizeof(struct udphdr
))) {
321 struct udphdr
*uh
= udp_hdr(skb
);
323 if (uh
->check
|| skb
->ip_summed
== CHECKSUM_PARTIAL
) {
324 inet_proto_csum_replace16(&uh
->check
, skb
,
327 uh
->check
= CSUM_MANGLED_0
;
330 } else if (l4_proto
== NEXTHDR_ICMP
) {
331 if (likely(transport_len
>= sizeof(struct icmp6hdr
)))
332 inet_proto_csum_replace16(&icmp6_hdr(skb
)->icmp6_cksum
,
333 skb
, addr
, new_addr
, 1);
337 static void mask_ipv6_addr(const __be32 old
[4], const __be32 addr
[4],
338 const __be32 mask
[4], __be32 masked
[4])
340 masked
[0] = MASKED(old
[0], addr
[0], mask
[0]);
341 masked
[1] = MASKED(old
[1], addr
[1], mask
[1]);
342 masked
[2] = MASKED(old
[2], addr
[2], mask
[2]);
343 masked
[3] = MASKED(old
[3], addr
[3], mask
[3]);
346 static void set_ipv6_addr(struct sk_buff
*skb
, u8 l4_proto
,
347 __be32 addr
[4], const __be32 new_addr
[4],
348 bool recalculate_csum
)
350 if (likely(recalculate_csum
))
351 update_ipv6_checksum(skb
, l4_proto
, addr
, new_addr
);
354 memcpy(addr
, new_addr
, sizeof(__be32
[4]));
357 static void set_ipv6_fl(struct ipv6hdr
*nh
, u32 fl
, u32 mask
)
359 /* Bits 21-24 are always unmasked, so this retains their values. */
360 SET_MASKED(nh
->flow_lbl
[0], (u8
)(fl
>> 16), (u8
)(mask
>> 16));
361 SET_MASKED(nh
->flow_lbl
[1], (u8
)(fl
>> 8), (u8
)(mask
>> 8));
362 SET_MASKED(nh
->flow_lbl
[2], (u8
)fl
, (u8
)mask
);
365 static void set_ip_ttl(struct sk_buff
*skb
, struct iphdr
*nh
, u8 new_ttl
,
368 new_ttl
= MASKED(nh
->ttl
, new_ttl
, mask
);
370 csum_replace2(&nh
->check
, htons(nh
->ttl
<< 8), htons(new_ttl
<< 8));
374 static int set_ipv4(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
375 const struct ovs_key_ipv4
*key
,
376 const struct ovs_key_ipv4
*mask
)
382 err
= skb_ensure_writable(skb
, skb_network_offset(skb
) +
383 sizeof(struct iphdr
));
389 /* Setting an IP addresses is typically only a side effect of
390 * matching on them in the current userspace implementation, so it
391 * makes sense to check if the value actually changed.
393 if (mask
->ipv4_src
) {
394 new_addr
= MASKED(nh
->saddr
, key
->ipv4_src
, mask
->ipv4_src
);
396 if (unlikely(new_addr
!= nh
->saddr
)) {
397 set_ip_addr(skb
, nh
, &nh
->saddr
, new_addr
);
398 flow_key
->ipv4
.addr
.src
= new_addr
;
401 if (mask
->ipv4_dst
) {
402 new_addr
= MASKED(nh
->daddr
, key
->ipv4_dst
, mask
->ipv4_dst
);
404 if (unlikely(new_addr
!= nh
->daddr
)) {
405 set_ip_addr(skb
, nh
, &nh
->daddr
, new_addr
);
406 flow_key
->ipv4
.addr
.dst
= new_addr
;
409 if (mask
->ipv4_tos
) {
410 ipv4_change_dsfield(nh
, ~mask
->ipv4_tos
, key
->ipv4_tos
);
411 flow_key
->ip
.tos
= nh
->tos
;
413 if (mask
->ipv4_ttl
) {
414 set_ip_ttl(skb
, nh
, key
->ipv4_ttl
, mask
->ipv4_ttl
);
415 flow_key
->ip
.ttl
= nh
->ttl
;
421 static bool is_ipv6_mask_nonzero(const __be32 addr
[4])
423 return !!(addr
[0] | addr
[1] | addr
[2] | addr
[3]);
426 static int set_ipv6(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
427 const struct ovs_key_ipv6
*key
,
428 const struct ovs_key_ipv6
*mask
)
433 err
= skb_ensure_writable(skb
, skb_network_offset(skb
) +
434 sizeof(struct ipv6hdr
));
440 /* Setting an IP addresses is typically only a side effect of
441 * matching on them in the current userspace implementation, so it
442 * makes sense to check if the value actually changed.
444 if (is_ipv6_mask_nonzero(mask
->ipv6_src
)) {
445 __be32
*saddr
= (__be32
*)&nh
->saddr
;
448 mask_ipv6_addr(saddr
, key
->ipv6_src
, mask
->ipv6_src
, masked
);
450 if (unlikely(memcmp(saddr
, masked
, sizeof(masked
)))) {
451 set_ipv6_addr(skb
, key
->ipv6_proto
, saddr
, masked
,
453 memcpy(&flow_key
->ipv6
.addr
.src
, masked
,
454 sizeof(flow_key
->ipv6
.addr
.src
));
457 if (is_ipv6_mask_nonzero(mask
->ipv6_dst
)) {
458 unsigned int offset
= 0;
459 int flags
= IP6_FH_F_SKIP_RH
;
460 bool recalc_csum
= true;
461 __be32
*daddr
= (__be32
*)&nh
->daddr
;
464 mask_ipv6_addr(daddr
, key
->ipv6_dst
, mask
->ipv6_dst
, masked
);
466 if (unlikely(memcmp(daddr
, masked
, sizeof(masked
)))) {
467 if (ipv6_ext_hdr(nh
->nexthdr
))
468 recalc_csum
= (ipv6_find_hdr(skb
, &offset
,
473 set_ipv6_addr(skb
, key
->ipv6_proto
, daddr
, masked
,
475 memcpy(&flow_key
->ipv6
.addr
.dst
, masked
,
476 sizeof(flow_key
->ipv6
.addr
.dst
));
479 if (mask
->ipv6_tclass
) {
480 ipv6_change_dsfield(nh
, ~mask
->ipv6_tclass
, key
->ipv6_tclass
);
481 flow_key
->ip
.tos
= ipv6_get_dsfield(nh
);
483 if (mask
->ipv6_label
) {
484 set_ipv6_fl(nh
, ntohl(key
->ipv6_label
),
485 ntohl(mask
->ipv6_label
));
486 flow_key
->ipv6
.label
=
487 *(__be32
*)nh
& htonl(IPV6_FLOWINFO_FLOWLABEL
);
489 if (mask
->ipv6_hlimit
) {
490 SET_MASKED(nh
->hop_limit
, key
->ipv6_hlimit
, mask
->ipv6_hlimit
);
491 flow_key
->ip
.ttl
= nh
->hop_limit
;
496 /* Must follow skb_ensure_writable() since that can move the skb data. */
497 static void set_tp_port(struct sk_buff
*skb
, __be16
*port
,
498 __be16 new_port
, __sum16
*check
)
500 inet_proto_csum_replace2(check
, skb
, *port
, new_port
, 0);
504 static int set_udp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
505 const struct ovs_key_udp
*key
,
506 const struct ovs_key_udp
*mask
)
512 err
= skb_ensure_writable(skb
, skb_transport_offset(skb
) +
513 sizeof(struct udphdr
));
518 /* Either of the masks is non-zero, so do not bother checking them. */
519 src
= MASKED(uh
->source
, key
->udp_src
, mask
->udp_src
);
520 dst
= MASKED(uh
->dest
, key
->udp_dst
, mask
->udp_dst
);
522 if (uh
->check
&& skb
->ip_summed
!= CHECKSUM_PARTIAL
) {
523 if (likely(src
!= uh
->source
)) {
524 set_tp_port(skb
, &uh
->source
, src
, &uh
->check
);
525 flow_key
->tp
.src
= src
;
527 if (likely(dst
!= uh
->dest
)) {
528 set_tp_port(skb
, &uh
->dest
, dst
, &uh
->check
);
529 flow_key
->tp
.dst
= dst
;
532 if (unlikely(!uh
->check
))
533 uh
->check
= CSUM_MANGLED_0
;
537 flow_key
->tp
.src
= src
;
538 flow_key
->tp
.dst
= dst
;
546 static int set_tcp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
547 const struct ovs_key_tcp
*key
,
548 const struct ovs_key_tcp
*mask
)
554 err
= skb_ensure_writable(skb
, skb_transport_offset(skb
) +
555 sizeof(struct tcphdr
));
561 src
= MASKED(th
->source
, key
->tcp_src
, mask
->tcp_src
);
562 if (likely(src
!= th
->source
)) {
563 set_tp_port(skb
, &th
->source
, src
, &th
->check
);
564 flow_key
->tp
.src
= src
;
566 dst
= MASKED(th
->dest
, key
->tcp_dst
, mask
->tcp_dst
);
567 if (likely(dst
!= th
->dest
)) {
568 set_tp_port(skb
, &th
->dest
, dst
, &th
->check
);
569 flow_key
->tp
.dst
= dst
;
576 static int set_sctp(struct sk_buff
*skb
, struct sw_flow_key
*flow_key
,
577 const struct ovs_key_sctp
*key
,
578 const struct ovs_key_sctp
*mask
)
580 unsigned int sctphoff
= skb_transport_offset(skb
);
582 __le32 old_correct_csum
, new_csum
, old_csum
;
585 err
= skb_ensure_writable(skb
, sctphoff
+ sizeof(struct sctphdr
));
591 old_csum
= sh
->checksum
;
592 old_correct_csum
= sctp_compute_cksum(skb
, sctphoff
);
594 sh
->source
= MASKED(sh
->source
, key
->sctp_src
, mask
->sctp_src
);
595 sh
->dest
= MASKED(sh
->dest
, key
->sctp_dst
, mask
->sctp_dst
);
597 new_csum
= sctp_compute_cksum(skb
, sctphoff
);
599 /* Carry any checksum errors through. */
600 sh
->checksum
= old_csum
^ old_correct_csum
^ new_csum
;
603 flow_key
->tp
.src
= sh
->source
;
604 flow_key
->tp
.dst
= sh
->dest
;
609 static void do_output(struct datapath
*dp
, struct sk_buff
*skb
, int out_port
)
611 struct vport
*vport
= ovs_vport_rcu(dp
, out_port
);
614 ovs_vport_send(vport
, skb
);
619 static int output_userspace(struct datapath
*dp
, struct sk_buff
*skb
,
620 struct sw_flow_key
*key
, const struct nlattr
*attr
,
621 const struct nlattr
*actions
, int actions_len
)
623 struct ovs_tunnel_info info
;
624 struct dp_upcall_info upcall
;
625 const struct nlattr
*a
;
628 memset(&upcall
, 0, sizeof(upcall
));
629 upcall
.cmd
= OVS_PACKET_CMD_ACTION
;
631 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
632 a
= nla_next(a
, &rem
)) {
633 switch (nla_type(a
)) {
634 case OVS_USERSPACE_ATTR_USERDATA
:
638 case OVS_USERSPACE_ATTR_PID
:
639 upcall
.portid
= nla_get_u32(a
);
642 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT
: {
643 /* Get out tunnel info. */
646 vport
= ovs_vport_rcu(dp
, nla_get_u32(a
));
650 err
= ovs_vport_get_egress_tun_info(vport
, skb
,
653 upcall
.egress_tun_info
= &info
;
658 case OVS_USERSPACE_ATTR_ACTIONS
: {
659 /* Include actions. */
660 upcall
.actions
= actions
;
661 upcall
.actions_len
= actions_len
;
665 } /* End of switch. */
668 return ovs_dp_upcall(dp
, skb
, key
, &upcall
);
671 static int sample(struct datapath
*dp
, struct sk_buff
*skb
,
672 struct sw_flow_key
*key
, const struct nlattr
*attr
,
673 const struct nlattr
*actions
, int actions_len
)
675 const struct nlattr
*acts_list
= NULL
;
676 const struct nlattr
*a
;
679 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
680 a
= nla_next(a
, &rem
)) {
681 switch (nla_type(a
)) {
682 case OVS_SAMPLE_ATTR_PROBABILITY
:
683 if (prandom_u32() >= nla_get_u32(a
))
687 case OVS_SAMPLE_ATTR_ACTIONS
:
693 rem
= nla_len(acts_list
);
694 a
= nla_data(acts_list
);
696 /* Actions list is empty, do nothing */
700 /* The only known usage of sample action is having a single user-space
701 * action. Treat this usage as a special case.
702 * The output_userspace() should clone the skb to be sent to the
703 * user space. This skb will be consumed by its caller.
705 if (likely(nla_type(a
) == OVS_ACTION_ATTR_USERSPACE
&&
706 nla_is_last(a
, rem
)))
707 return output_userspace(dp
, skb
, key
, a
, actions
, actions_len
);
709 skb
= skb_clone(skb
, GFP_ATOMIC
);
711 /* Skip the sample action when out of memory. */
714 if (!add_deferred_actions(skb
, key
, a
)) {
716 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
724 static void execute_hash(struct sk_buff
*skb
, struct sw_flow_key
*key
,
725 const struct nlattr
*attr
)
727 struct ovs_action_hash
*hash_act
= nla_data(attr
);
730 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
731 hash
= skb_get_hash(skb
);
732 hash
= jhash_1word(hash
, hash_act
->hash_basis
);
736 key
->ovs_flow_hash
= hash
;
739 static int execute_set_action(struct sk_buff
*skb
,
740 struct sw_flow_key
*flow_key
,
741 const struct nlattr
*a
)
743 /* Only tunnel set execution is supported without a mask. */
744 if (nla_type(a
) == OVS_KEY_ATTR_TUNNEL_INFO
) {
745 OVS_CB(skb
)->egress_tun_info
= nla_data(a
);
753 /* Mask is at the midpoint of the data. */
754 #define get_mask(a, type) ((const type)nla_data(a) + 1)
756 static int execute_masked_set_action(struct sk_buff
*skb
,
757 struct sw_flow_key
*flow_key
,
758 const struct nlattr
*a
)
762 switch (nla_type(a
)) {
763 case OVS_KEY_ATTR_PRIORITY
:
764 SET_MASKED(skb
->priority
, nla_get_u32(a
), *get_mask(a
, u32
*));
765 flow_key
->phy
.priority
= skb
->priority
;
768 case OVS_KEY_ATTR_SKB_MARK
:
769 SET_MASKED(skb
->mark
, nla_get_u32(a
), *get_mask(a
, u32
*));
770 flow_key
->phy
.skb_mark
= skb
->mark
;
773 case OVS_KEY_ATTR_TUNNEL_INFO
:
774 /* Masked data not supported for tunnel. */
778 case OVS_KEY_ATTR_ETHERNET
:
779 err
= set_eth_addr(skb
, flow_key
, nla_data(a
),
780 get_mask(a
, struct ovs_key_ethernet
*));
783 case OVS_KEY_ATTR_IPV4
:
784 err
= set_ipv4(skb
, flow_key
, nla_data(a
),
785 get_mask(a
, struct ovs_key_ipv4
*));
788 case OVS_KEY_ATTR_IPV6
:
789 err
= set_ipv6(skb
, flow_key
, nla_data(a
),
790 get_mask(a
, struct ovs_key_ipv6
*));
793 case OVS_KEY_ATTR_TCP
:
794 err
= set_tcp(skb
, flow_key
, nla_data(a
),
795 get_mask(a
, struct ovs_key_tcp
*));
798 case OVS_KEY_ATTR_UDP
:
799 err
= set_udp(skb
, flow_key
, nla_data(a
),
800 get_mask(a
, struct ovs_key_udp
*));
803 case OVS_KEY_ATTR_SCTP
:
804 err
= set_sctp(skb
, flow_key
, nla_data(a
),
805 get_mask(a
, struct ovs_key_sctp
*));
808 case OVS_KEY_ATTR_MPLS
:
809 err
= set_mpls(skb
, flow_key
, nla_data(a
), get_mask(a
,
817 static int execute_recirc(struct datapath
*dp
, struct sk_buff
*skb
,
818 struct sw_flow_key
*key
,
819 const struct nlattr
*a
, int rem
)
821 struct deferred_action
*da
;
823 if (!is_flow_key_valid(key
)) {
826 err
= ovs_flow_key_update(skb
, key
);
830 BUG_ON(!is_flow_key_valid(key
));
832 if (!nla_is_last(a
, rem
)) {
833 /* Recirc action is the not the last action
834 * of the action list, need to clone the skb.
836 skb
= skb_clone(skb
, GFP_ATOMIC
);
838 /* Skip the recirc action when out of memory, but
839 * continue on with the rest of the action list.
845 da
= add_deferred_actions(skb
, key
, NULL
);
847 da
->pkt_key
.recirc_id
= nla_get_u32(a
);
852 pr_warn("%s: deferred action limit reached, drop recirc action\n",
859 /* Execute a list of actions against 'skb'. */
860 static int do_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
861 struct sw_flow_key
*key
,
862 const struct nlattr
*attr
, int len
)
864 /* Every output action needs a separate clone of 'skb', but the common
865 * case is just a single output action, so that doing a clone and
866 * then freeing the original skbuff is wasteful. So the following code
867 * is slightly obscure just to avoid that.
870 const struct nlattr
*a
;
873 for (a
= attr
, rem
= len
; rem
> 0;
874 a
= nla_next(a
, &rem
)) {
877 if (unlikely(prev_port
!= -1)) {
878 struct sk_buff
*out_skb
= skb_clone(skb
, GFP_ATOMIC
);
881 do_output(dp
, out_skb
, prev_port
);
886 switch (nla_type(a
)) {
887 case OVS_ACTION_ATTR_OUTPUT
:
888 prev_port
= nla_get_u32(a
);
891 case OVS_ACTION_ATTR_USERSPACE
:
892 output_userspace(dp
, skb
, key
, a
, attr
, len
);
895 case OVS_ACTION_ATTR_HASH
:
896 execute_hash(skb
, key
, a
);
899 case OVS_ACTION_ATTR_PUSH_MPLS
:
900 err
= push_mpls(skb
, key
, nla_data(a
));
903 case OVS_ACTION_ATTR_POP_MPLS
:
904 err
= pop_mpls(skb
, key
, nla_get_be16(a
));
907 case OVS_ACTION_ATTR_PUSH_VLAN
:
908 err
= push_vlan(skb
, key
, nla_data(a
));
911 case OVS_ACTION_ATTR_POP_VLAN
:
912 err
= pop_vlan(skb
, key
);
915 case OVS_ACTION_ATTR_RECIRC
:
916 err
= execute_recirc(dp
, skb
, key
, a
, rem
);
917 if (nla_is_last(a
, rem
)) {
918 /* If this is the last action, the skb has
919 * been consumed or freed.
920 * Return immediately.
926 case OVS_ACTION_ATTR_SET
:
927 err
= execute_set_action(skb
, key
, nla_data(a
));
930 case OVS_ACTION_ATTR_SET_MASKED
:
931 case OVS_ACTION_ATTR_SET_TO_MASKED
:
932 err
= execute_masked_set_action(skb
, key
, nla_data(a
));
935 case OVS_ACTION_ATTR_SAMPLE
:
936 err
= sample(dp
, skb
, key
, a
, attr
, len
);
947 do_output(dp
, skb
, prev_port
);
954 static void process_deferred_actions(struct datapath
*dp
)
956 struct action_fifo
*fifo
= this_cpu_ptr(action_fifos
);
958 /* Do not touch the FIFO in case there is no deferred actions. */
959 if (action_fifo_is_empty(fifo
))
962 /* Finishing executing all deferred actions. */
964 struct deferred_action
*da
= action_fifo_get(fifo
);
965 struct sk_buff
*skb
= da
->skb
;
966 struct sw_flow_key
*key
= &da
->pkt_key
;
967 const struct nlattr
*actions
= da
->actions
;
970 do_execute_actions(dp
, skb
, key
, actions
,
973 ovs_dp_process_packet(skb
, key
);
974 } while (!action_fifo_is_empty(fifo
));
976 /* Reset FIFO for the next packet. */
977 action_fifo_init(fifo
);
980 /* Execute a list of actions against 'skb'. */
981 int ovs_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
982 const struct sw_flow_actions
*acts
,
983 struct sw_flow_key
*key
)
985 int level
= this_cpu_read(exec_actions_level
);
988 if (unlikely(level
>= EXEC_ACTIONS_LEVEL_LIMIT
)) {
990 pr_warn("%s: packet loop detected, dropping.\n",
997 this_cpu_inc(exec_actions_level
);
998 err
= do_execute_actions(dp
, skb
, key
,
999 acts
->actions
, acts
->actions_len
);
1002 process_deferred_actions(dp
);
1004 this_cpu_dec(exec_actions_level
);
1006 /* This return status currently does not reflect the errors
1007 * encounted during deferred actions execution. Probably needs to
1008 * be fixed in the future.
1013 int action_fifos_init(void)
1015 action_fifos
= alloc_percpu(struct action_fifo
);
1022 void action_fifos_exit(void)
1024 free_percpu(action_fifos
);