2 * Copyright (c) 2007-2012 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/tcp.h>
26 #include <linux/udp.h>
27 #include <linux/in6.h>
28 #include <linux/if_arp.h>
29 #include <linux/if_vlan.h>
32 #include <net/checksum.h>
33 #include <net/dsfield.h>
40 static int do_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
41 const struct nlattr
*attr
, int len
,
42 struct ovs_key_ipv4_tunnel
*tun_key
, bool keep_skb
);
44 static int make_writable(struct sk_buff
*skb
, int write_len
)
46 if (!skb_cloned(skb
) || skb_clone_writable(skb
, write_len
))
49 return pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
52 /* remove VLAN header from packet and update csum accordingly. */
53 static int __pop_vlan_tci(struct sk_buff
*skb
, __be16
*current_tci
)
55 struct vlan_hdr
*vhdr
;
58 err
= make_writable(skb
, VLAN_ETH_HLEN
);
62 if (get_ip_summed(skb
) == OVS_CSUM_COMPLETE
)
63 skb
->csum
= csum_sub(skb
->csum
, csum_partial(skb
->data
64 + ETH_HLEN
, VLAN_HLEN
, 0));
66 vhdr
= (struct vlan_hdr
*)(skb
->data
+ ETH_HLEN
);
67 *current_tci
= vhdr
->h_vlan_TCI
;
69 memmove(skb
->data
+ VLAN_HLEN
, skb
->data
, 2 * ETH_ALEN
);
70 __skb_pull(skb
, VLAN_HLEN
);
72 vlan_set_encap_proto(skb
, vhdr
);
73 skb
->mac_header
+= VLAN_HLEN
;
74 skb_reset_mac_len(skb
);
79 static int pop_vlan(struct sk_buff
*skb
)
84 if (likely(vlan_tx_tag_present(skb
))) {
87 if (unlikely(skb
->protocol
!= htons(ETH_P_8021Q
) ||
88 skb
->len
< VLAN_ETH_HLEN
))
91 err
= __pop_vlan_tci(skb
, &tci
);
95 /* move next vlan tag to hw accel tag */
96 if (likely(skb
->protocol
!= htons(ETH_P_8021Q
) ||
97 skb
->len
< VLAN_ETH_HLEN
))
100 err
= __pop_vlan_tci(skb
, &tci
);
104 __vlan_hwaccel_put_tag(skb
, ntohs(tci
));
108 static int push_vlan(struct sk_buff
*skb
, const struct ovs_action_push_vlan
*vlan
)
110 if (unlikely(vlan_tx_tag_present(skb
))) {
113 /* push down current VLAN tag */
114 current_tag
= vlan_tx_tag_get(skb
);
116 if (!__vlan_put_tag(skb
, current_tag
))
119 if (get_ip_summed(skb
) == OVS_CSUM_COMPLETE
)
120 skb
->csum
= csum_add(skb
->csum
, csum_partial(skb
->data
121 + ETH_HLEN
, VLAN_HLEN
, 0));
124 __vlan_hwaccel_put_tag(skb
, ntohs(vlan
->vlan_tci
) & ~VLAN_TAG_PRESENT
);
128 static int set_eth_addr(struct sk_buff
*skb
,
129 const struct ovs_key_ethernet
*eth_key
)
132 err
= make_writable(skb
, ETH_HLEN
);
136 memcpy(eth_hdr(skb
)->h_source
, eth_key
->eth_src
, ETH_ALEN
);
137 memcpy(eth_hdr(skb
)->h_dest
, eth_key
->eth_dst
, ETH_ALEN
);
142 static void set_ip_addr(struct sk_buff
*skb
, struct iphdr
*nh
,
143 __be32
*addr
, __be32 new_addr
)
145 int transport_len
= skb
->len
- skb_transport_offset(skb
);
147 if (nh
->protocol
== IPPROTO_TCP
) {
148 if (likely(transport_len
>= sizeof(struct tcphdr
)))
149 inet_proto_csum_replace4(&tcp_hdr(skb
)->check
, skb
,
151 } else if (nh
->protocol
== IPPROTO_UDP
) {
152 if (likely(transport_len
>= sizeof(struct udphdr
))) {
153 struct udphdr
*uh
= udp_hdr(skb
);
156 get_ip_summed(skb
) == OVS_CSUM_PARTIAL
) {
157 inet_proto_csum_replace4(&uh
->check
, skb
,
160 uh
->check
= CSUM_MANGLED_0
;
165 csum_replace4(&nh
->check
, *addr
, new_addr
);
166 skb_clear_rxhash(skb
);
170 static void update_ipv6_checksum(struct sk_buff
*skb
, u8 l4_proto
,
171 __be32 addr
[4], const __be32 new_addr
[4])
173 int transport_len
= skb
->len
- skb_transport_offset(skb
);
175 if (l4_proto
== IPPROTO_TCP
) {
176 if (likely(transport_len
>= sizeof(struct tcphdr
)))
177 inet_proto_csum_replace16(&tcp_hdr(skb
)->check
, skb
,
179 } else if (l4_proto
== IPPROTO_UDP
) {
180 if (likely(transport_len
>= sizeof(struct udphdr
))) {
181 struct udphdr
*uh
= udp_hdr(skb
);
184 get_ip_summed(skb
) == OVS_CSUM_PARTIAL
) {
185 inet_proto_csum_replace16(&uh
->check
, skb
,
188 uh
->check
= CSUM_MANGLED_0
;
194 static void set_ipv6_addr(struct sk_buff
*skb
, u8 l4_proto
,
195 __be32 addr
[4], const __be32 new_addr
[4],
196 bool recalculate_csum
)
198 if (recalculate_csum
)
199 update_ipv6_checksum(skb
, l4_proto
, addr
, new_addr
);
201 skb_clear_rxhash(skb
);
202 memcpy(addr
, new_addr
, sizeof(__be32
[4]));
205 static void set_ipv6_tc(struct ipv6hdr
*nh
, u8 tc
)
207 nh
->priority
= tc
>> 4;
208 nh
->flow_lbl
[0] = (nh
->flow_lbl
[0] & 0x0F) | ((tc
& 0x0F) << 4);
211 static void set_ipv6_fl(struct ipv6hdr
*nh
, u32 fl
)
213 nh
->flow_lbl
[0] = (nh
->flow_lbl
[0] & 0xF0) | (fl
& 0x000F0000) >> 16;
214 nh
->flow_lbl
[1] = (fl
& 0x0000FF00) >> 8;
215 nh
->flow_lbl
[2] = fl
& 0x000000FF;
218 static void set_ip_ttl(struct sk_buff
*skb
, struct iphdr
*nh
, u8 new_ttl
)
220 csum_replace2(&nh
->check
, htons(nh
->ttl
<< 8), htons(new_ttl
<< 8));
224 static int set_ipv4(struct sk_buff
*skb
, const struct ovs_key_ipv4
*ipv4_key
)
229 err
= make_writable(skb
, skb_network_offset(skb
) +
230 sizeof(struct iphdr
));
236 if (ipv4_key
->ipv4_src
!= nh
->saddr
)
237 set_ip_addr(skb
, nh
, &nh
->saddr
, ipv4_key
->ipv4_src
);
239 if (ipv4_key
->ipv4_dst
!= nh
->daddr
)
240 set_ip_addr(skb
, nh
, &nh
->daddr
, ipv4_key
->ipv4_dst
);
242 if (ipv4_key
->ipv4_tos
!= nh
->tos
)
243 ipv4_change_dsfield(nh
, 0, ipv4_key
->ipv4_tos
);
245 if (ipv4_key
->ipv4_ttl
!= nh
->ttl
)
246 set_ip_ttl(skb
, nh
, ipv4_key
->ipv4_ttl
);
251 static int set_ipv6(struct sk_buff
*skb
, const struct ovs_key_ipv6
*ipv6_key
)
258 err
= make_writable(skb
, skb_network_offset(skb
) +
259 sizeof(struct ipv6hdr
));
264 saddr
= (__be32
*)&nh
->saddr
;
265 daddr
= (__be32
*)&nh
->daddr
;
267 if (memcmp(ipv6_key
->ipv6_src
, saddr
, sizeof(ipv6_key
->ipv6_src
)))
268 set_ipv6_addr(skb
, ipv6_key
->ipv6_proto
, saddr
,
269 ipv6_key
->ipv6_src
, true);
271 if (memcmp(ipv6_key
->ipv6_dst
, daddr
, sizeof(ipv6_key
->ipv6_dst
))) {
272 unsigned int offset
= 0;
273 int flags
= OVS_IP6T_FH_F_SKIP_RH
;
274 bool recalc_csum
= true;
276 if (ipv6_ext_hdr(nh
->nexthdr
))
277 recalc_csum
= ipv6_find_hdr(skb
, &offset
,
278 NEXTHDR_ROUTING
, NULL
,
279 &flags
) != NEXTHDR_ROUTING
;
281 set_ipv6_addr(skb
, ipv6_key
->ipv6_proto
, daddr
,
282 ipv6_key
->ipv6_dst
, recalc_csum
);
285 set_ipv6_tc(nh
, ipv6_key
->ipv6_tclass
);
286 set_ipv6_fl(nh
, ntohl(ipv6_key
->ipv6_label
));
287 nh
->hop_limit
= ipv6_key
->ipv6_hlimit
;
292 /* Must follow make_writable() since that can move the skb data. */
293 static void set_tp_port(struct sk_buff
*skb
, __be16
*port
,
294 __be16 new_port
, __sum16
*check
)
296 inet_proto_csum_replace2(check
, skb
, *port
, new_port
, 0);
298 skb_clear_rxhash(skb
);
301 static void set_udp_port(struct sk_buff
*skb
, __be16
*port
, __be16 new_port
)
303 struct udphdr
*uh
= udp_hdr(skb
);
305 if (uh
->check
&& get_ip_summed(skb
) != OVS_CSUM_PARTIAL
) {
306 set_tp_port(skb
, port
, new_port
, &uh
->check
);
309 uh
->check
= CSUM_MANGLED_0
;
312 skb_clear_rxhash(skb
);
316 static int set_udp(struct sk_buff
*skb
, const struct ovs_key_udp
*udp_port_key
)
321 err
= make_writable(skb
, skb_transport_offset(skb
) +
322 sizeof(struct udphdr
));
327 if (udp_port_key
->udp_src
!= uh
->source
)
328 set_udp_port(skb
, &uh
->source
, udp_port_key
->udp_src
);
330 if (udp_port_key
->udp_dst
!= uh
->dest
)
331 set_udp_port(skb
, &uh
->dest
, udp_port_key
->udp_dst
);
336 static int set_tcp(struct sk_buff
*skb
, const struct ovs_key_tcp
*tcp_port_key
)
341 err
= make_writable(skb
, skb_transport_offset(skb
) +
342 sizeof(struct tcphdr
));
347 if (tcp_port_key
->tcp_src
!= th
->source
)
348 set_tp_port(skb
, &th
->source
, tcp_port_key
->tcp_src
, &th
->check
);
350 if (tcp_port_key
->tcp_dst
!= th
->dest
)
351 set_tp_port(skb
, &th
->dest
, tcp_port_key
->tcp_dst
, &th
->check
);
356 static int do_output(struct datapath
*dp
, struct sk_buff
*skb
, int out_port
)
363 vport
= ovs_vport_rcu(dp
, out_port
);
364 if (unlikely(!vport
)) {
369 ovs_vport_send(vport
, skb
);
373 static int output_userspace(struct datapath
*dp
, struct sk_buff
*skb
,
374 const struct nlattr
*attr
)
376 struct dp_upcall_info upcall
;
377 const struct nlattr
*a
;
380 upcall
.cmd
= OVS_PACKET_CMD_ACTION
;
381 upcall
.key
= &OVS_CB(skb
)->flow
->key
;
382 upcall
.userdata
= NULL
;
385 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
386 a
= nla_next(a
, &rem
)) {
387 switch (nla_type(a
)) {
388 case OVS_USERSPACE_ATTR_USERDATA
:
392 case OVS_USERSPACE_ATTR_PID
:
393 upcall
.pid
= nla_get_u32(a
);
398 return ovs_dp_upcall(dp
, skb
, &upcall
);
401 static int sample(struct datapath
*dp
, struct sk_buff
*skb
,
402 const struct nlattr
*attr
,
403 struct ovs_key_ipv4_tunnel
*tun_key
)
405 const struct nlattr
*acts_list
= NULL
;
406 const struct nlattr
*a
;
409 for (a
= nla_data(attr
), rem
= nla_len(attr
); rem
> 0;
410 a
= nla_next(a
, &rem
)) {
411 switch (nla_type(a
)) {
412 case OVS_SAMPLE_ATTR_PROBABILITY
:
413 if (net_random() >= nla_get_u32(a
))
417 case OVS_SAMPLE_ATTR_ACTIONS
:
423 return do_execute_actions(dp
, skb
, nla_data(acts_list
),
424 nla_len(acts_list
), tun_key
, true);
427 static int execute_set_action(struct sk_buff
*skb
,
428 const struct nlattr
*nested_attr
,
429 struct ovs_key_ipv4_tunnel
*tun_key
)
433 switch (nla_type(nested_attr
)) {
434 case OVS_KEY_ATTR_PRIORITY
:
435 skb
->priority
= nla_get_u32(nested_attr
);
438 case OVS_KEY_ATTR_TUN_ID
:
439 /* If we're only using the TUN_ID action, store the value in a
440 * temporary instance of struct ovs_key_ipv4_tunnel on the stack.
441 * If both IPV4_TUNNEL and TUN_ID are being used together we
442 * can't write into the IPV4_TUNNEL action, so make a copy and
443 * write into that version.
445 if (!OVS_CB(skb
)->tun_key
)
446 memset(tun_key
, 0, sizeof(*tun_key
));
447 else if (OVS_CB(skb
)->tun_key
!= tun_key
)
448 memcpy(tun_key
, OVS_CB(skb
)->tun_key
, sizeof(*tun_key
));
449 OVS_CB(skb
)->tun_key
= tun_key
;
451 OVS_CB(skb
)->tun_key
->tun_id
= nla_get_be64(nested_attr
);
454 case OVS_KEY_ATTR_IPV4_TUNNEL
:
455 OVS_CB(skb
)->tun_key
= nla_data(nested_attr
);
458 case OVS_KEY_ATTR_ETHERNET
:
459 err
= set_eth_addr(skb
, nla_data(nested_attr
));
462 case OVS_KEY_ATTR_IPV4
:
463 err
= set_ipv4(skb
, nla_data(nested_attr
));
466 case OVS_KEY_ATTR_IPV6
:
467 err
= set_ipv6(skb
, nla_data(nested_attr
));
470 case OVS_KEY_ATTR_TCP
:
471 err
= set_tcp(skb
, nla_data(nested_attr
));
474 case OVS_KEY_ATTR_UDP
:
475 err
= set_udp(skb
, nla_data(nested_attr
));
482 /* Execute a list of actions against 'skb'. */
483 static int do_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
,
484 const struct nlattr
*attr
, int len
,
485 struct ovs_key_ipv4_tunnel
*tun_key
, bool keep_skb
)
487 /* Every output action needs a separate clone of 'skb', but the common
488 * case is just a single output action, so that doing a clone and
489 * then freeing the original skbuff is wasteful. So the following code
490 * is slightly obscure just to avoid that. */
492 const struct nlattr
*a
;
495 for (a
= attr
, rem
= len
; rem
> 0;
496 a
= nla_next(a
, &rem
)) {
499 if (prev_port
!= -1) {
500 do_output(dp
, skb_clone(skb
, GFP_ATOMIC
), prev_port
);
504 switch (nla_type(a
)) {
505 case OVS_ACTION_ATTR_OUTPUT
:
506 prev_port
= nla_get_u32(a
);
509 case OVS_ACTION_ATTR_USERSPACE
:
510 output_userspace(dp
, skb
, a
);
513 case OVS_ACTION_ATTR_PUSH_VLAN
:
514 err
= push_vlan(skb
, nla_data(a
));
515 if (unlikely(err
)) /* skb already freed. */
519 case OVS_ACTION_ATTR_POP_VLAN
:
523 case OVS_ACTION_ATTR_SET
:
524 err
= execute_set_action(skb
, nla_data(a
), tun_key
);
527 case OVS_ACTION_ATTR_SAMPLE
:
528 err
= sample(dp
, skb
, a
, tun_key
);
538 if (prev_port
!= -1) {
540 skb
= skb_clone(skb
, GFP_ATOMIC
);
542 do_output(dp
, skb
, prev_port
);
543 } else if (!keep_skb
)
549 /* We limit the number of times that we pass into execute_actions()
550 * to avoid blowing out the stack in the event that we have a loop. */
553 struct loop_counter
{
554 u8 count
; /* Count. */
555 bool looping
; /* Loop detected? */
558 static DEFINE_PER_CPU(struct loop_counter
, loop_counters
);
560 static int loop_suppress(struct datapath
*dp
, struct sw_flow_actions
*actions
)
563 pr_warn("%s: flow looped %d times, dropping\n",
564 ovs_dp_name(dp
), MAX_LOOPS
);
565 actions
->actions_len
= 0;
569 /* Execute a list of actions against 'skb'. */
570 int ovs_execute_actions(struct datapath
*dp
, struct sk_buff
*skb
)
572 struct sw_flow_actions
*acts
= rcu_dereference(OVS_CB(skb
)->flow
->sf_acts
);
573 struct loop_counter
*loop
;
575 struct ovs_key_ipv4_tunnel tun_key
;
577 /* Check whether we've looped too much. */
578 loop
= &__get_cpu_var(loop_counters
);
579 if (unlikely(++loop
->count
> MAX_LOOPS
))
580 loop
->looping
= true;
581 if (unlikely(loop
->looping
)) {
582 error
= loop_suppress(dp
, acts
);
587 OVS_CB(skb
)->tun_key
= NULL
;
588 error
= do_execute_actions(dp
, skb
, acts
->actions
,
589 acts
->actions_len
, &tun_key
, false);
591 /* Check whether sub-actions looped too much. */
592 if (unlikely(loop
->looping
))
593 error
= loop_suppress(dp
, acts
);
596 /* Decrement loop counter. */
598 loop
->looping
= false;