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 #include <linux/uaccess.h>
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/if_ether.h>
23 #include <linux/if_vlan.h>
24 #include <net/llc_pdu.h>
25 #include <linux/kernel.h>
26 #include <linux/jhash.h>
27 #include <linux/jiffies.h>
28 #include <linux/llc.h>
29 #include <linux/module.h>
31 #include <linux/rcupdate.h>
32 #include <linux/cpumask.h>
33 #include <linux/if_arp.h>
35 #include <linux/ipv6.h>
36 #include <linux/mpls.h>
37 #include <linux/sctp.h>
38 #include <linux/smp.h>
39 #include <linux/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/icmp.h>
42 #include <linux/icmpv6.h>
43 #include <linux/rculist.h>
44 #include <linux/timekeeping.h>
48 #include <net/ndisc.h>
52 #include "conntrack.h"
54 #include "flow_netlink.h"
57 u64
ovs_flow_used_time(unsigned long flow_jiffies
)
59 struct timespec64 cur_ts
;
62 ktime_get_ts64(&cur_ts
);
63 idle_ms
= jiffies_to_msecs(jiffies
- flow_jiffies
);
64 cur_ms
= (u64
)(u32
)cur_ts
.tv_sec
* MSEC_PER_SEC
+
65 cur_ts
.tv_nsec
/ NSEC_PER_MSEC
;
67 return cur_ms
- idle_ms
;
70 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
72 void ovs_flow_stats_update(struct sw_flow
*flow
, __be16 tcp_flags
,
73 const struct sk_buff
*skb
)
75 struct flow_stats
*stats
;
76 unsigned int cpu
= smp_processor_id();
77 int len
= skb
->len
+ (skb_vlan_tag_present(skb
) ? VLAN_HLEN
: 0);
79 stats
= rcu_dereference(flow
->stats
[cpu
]);
81 /* Check if already have CPU-specific stats. */
83 spin_lock(&stats
->lock
);
84 /* Mark if we write on the pre-allocated stats. */
85 if (cpu
== 0 && unlikely(flow
->stats_last_writer
!= cpu
))
86 flow
->stats_last_writer
= cpu
;
88 stats
= rcu_dereference(flow
->stats
[0]); /* Pre-allocated. */
89 spin_lock(&stats
->lock
);
91 /* If the current CPU is the only writer on the
92 * pre-allocated stats keep using them.
94 if (unlikely(flow
->stats_last_writer
!= cpu
)) {
95 /* A previous locker may have already allocated the
96 * stats, so we need to check again. If CPU-specific
97 * stats were already allocated, we update the pre-
98 * allocated stats as we have already locked them.
100 if (likely(flow
->stats_last_writer
!= -1) &&
101 likely(!rcu_access_pointer(flow
->stats
[cpu
]))) {
102 /* Try to allocate CPU-specific stats. */
103 struct flow_stats
*new_stats
;
106 kmem_cache_alloc_node(flow_stats_cache
,
112 if (likely(new_stats
)) {
113 new_stats
->used
= jiffies
;
114 new_stats
->packet_count
= 1;
115 new_stats
->byte_count
= len
;
116 new_stats
->tcp_flags
= tcp_flags
;
117 spin_lock_init(&new_stats
->lock
);
119 rcu_assign_pointer(flow
->stats
[cpu
],
121 cpumask_set_cpu(cpu
, &flow
->cpu_used_mask
);
125 flow
->stats_last_writer
= cpu
;
129 stats
->used
= jiffies
;
130 stats
->packet_count
++;
131 stats
->byte_count
+= len
;
132 stats
->tcp_flags
|= tcp_flags
;
134 spin_unlock(&stats
->lock
);
137 /* Must be called with rcu_read_lock or ovs_mutex. */
138 void ovs_flow_stats_get(const struct sw_flow
*flow
,
139 struct ovs_flow_stats
*ovs_stats
,
140 unsigned long *used
, __be16
*tcp_flags
)
146 memset(ovs_stats
, 0, sizeof(*ovs_stats
));
148 /* We open code this to make sure cpu 0 is always considered */
149 for (cpu
= 0; cpu
< nr_cpu_ids
; cpu
= cpumask_next(cpu
, &flow
->cpu_used_mask
)) {
150 struct flow_stats
*stats
= rcu_dereference_ovsl(flow
->stats
[cpu
]);
153 /* Local CPU may write on non-local stats, so we must
154 * block bottom-halves here.
156 spin_lock_bh(&stats
->lock
);
157 if (!*used
|| time_after(stats
->used
, *used
))
159 *tcp_flags
|= stats
->tcp_flags
;
160 ovs_stats
->n_packets
+= stats
->packet_count
;
161 ovs_stats
->n_bytes
+= stats
->byte_count
;
162 spin_unlock_bh(&stats
->lock
);
167 /* Called with ovs_mutex. */
168 void ovs_flow_stats_clear(struct sw_flow
*flow
)
172 /* We open code this to make sure cpu 0 is always considered */
173 for (cpu
= 0; cpu
< nr_cpu_ids
; cpu
= cpumask_next(cpu
, &flow
->cpu_used_mask
)) {
174 struct flow_stats
*stats
= ovsl_dereference(flow
->stats
[cpu
]);
177 spin_lock_bh(&stats
->lock
);
179 stats
->packet_count
= 0;
180 stats
->byte_count
= 0;
181 stats
->tcp_flags
= 0;
182 spin_unlock_bh(&stats
->lock
);
187 static int check_header(struct sk_buff
*skb
, int len
)
189 if (unlikely(skb
->len
< len
))
191 if (unlikely(!pskb_may_pull(skb
, len
)))
196 static bool arphdr_ok(struct sk_buff
*skb
)
198 return pskb_may_pull(skb
, skb_network_offset(skb
) +
199 sizeof(struct arp_eth_header
));
202 static int check_iphdr(struct sk_buff
*skb
)
204 unsigned int nh_ofs
= skb_network_offset(skb
);
208 err
= check_header(skb
, nh_ofs
+ sizeof(struct iphdr
));
212 ip_len
= ip_hdrlen(skb
);
213 if (unlikely(ip_len
< sizeof(struct iphdr
) ||
214 skb
->len
< nh_ofs
+ ip_len
))
217 skb_set_transport_header(skb
, nh_ofs
+ ip_len
);
221 static bool tcphdr_ok(struct sk_buff
*skb
)
223 int th_ofs
= skb_transport_offset(skb
);
226 if (unlikely(!pskb_may_pull(skb
, th_ofs
+ sizeof(struct tcphdr
))))
229 tcp_len
= tcp_hdrlen(skb
);
230 if (unlikely(tcp_len
< sizeof(struct tcphdr
) ||
231 skb
->len
< th_ofs
+ tcp_len
))
237 static bool udphdr_ok(struct sk_buff
*skb
)
239 return pskb_may_pull(skb
, skb_transport_offset(skb
) +
240 sizeof(struct udphdr
));
243 static bool sctphdr_ok(struct sk_buff
*skb
)
245 return pskb_may_pull(skb
, skb_transport_offset(skb
) +
246 sizeof(struct sctphdr
));
249 static bool icmphdr_ok(struct sk_buff
*skb
)
251 return pskb_may_pull(skb
, skb_transport_offset(skb
) +
252 sizeof(struct icmphdr
));
255 static int parse_ipv6hdr(struct sk_buff
*skb
, struct sw_flow_key
*key
)
257 unsigned short frag_off
;
258 unsigned int payload_ofs
= 0;
259 unsigned int nh_ofs
= skb_network_offset(skb
);
262 int err
, nexthdr
, flags
= 0;
264 err
= check_header(skb
, nh_ofs
+ sizeof(*nh
));
270 key
->ip
.proto
= NEXTHDR_NONE
;
271 key
->ip
.tos
= ipv6_get_dsfield(nh
);
272 key
->ip
.ttl
= nh
->hop_limit
;
273 key
->ipv6
.label
= *(__be32
*)nh
& htonl(IPV6_FLOWINFO_FLOWLABEL
);
274 key
->ipv6
.addr
.src
= nh
->saddr
;
275 key
->ipv6
.addr
.dst
= nh
->daddr
;
277 nexthdr
= ipv6_find_hdr(skb
, &payload_ofs
, -1, &frag_off
, &flags
);
278 if (flags
& IP6_FH_F_FRAG
) {
280 key
->ip
.frag
= OVS_FRAG_TYPE_LATER
;
281 key
->ip
.proto
= nexthdr
;
284 key
->ip
.frag
= OVS_FRAG_TYPE_FIRST
;
286 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
289 /* Delayed handling of error in ipv6_find_hdr() as it
290 * always sets flags and frag_off to a valid value which may be
291 * used to set key->ip.frag above.
293 if (unlikely(nexthdr
< 0))
296 nh_len
= payload_ofs
- nh_ofs
;
297 skb_set_transport_header(skb
, nh_ofs
+ nh_len
);
298 key
->ip
.proto
= nexthdr
;
302 static bool icmp6hdr_ok(struct sk_buff
*skb
)
304 return pskb_may_pull(skb
, skb_transport_offset(skb
) +
305 sizeof(struct icmp6hdr
));
309 * Parse vlan tag from vlan header.
310 * Returns ERROR on memory error.
311 * Returns 0 if it encounters a non-vlan or incomplete packet.
312 * Returns 1 after successfully parsing vlan tag.
314 static int parse_vlan_tag(struct sk_buff
*skb
, struct vlan_head
*key_vh
,
317 struct vlan_head
*vh
= (struct vlan_head
*)skb
->data
;
319 if (likely(!eth_type_vlan(vh
->tpid
)))
322 if (unlikely(skb
->len
< sizeof(struct vlan_head
) + sizeof(__be16
)))
325 if (unlikely(!pskb_may_pull(skb
, sizeof(struct vlan_head
) +
329 vh
= (struct vlan_head
*)skb
->data
;
330 key_vh
->tci
= vh
->tci
| htons(VLAN_CFI_MASK
);
331 key_vh
->tpid
= vh
->tpid
;
333 if (unlikely(untag_vlan
)) {
334 int offset
= skb
->data
- skb_mac_header(skb
);
338 __skb_push(skb
, offset
);
339 err
= __skb_vlan_pop(skb
, &tci
);
340 __skb_pull(skb
, offset
);
343 __vlan_hwaccel_put_tag(skb
, key_vh
->tpid
, tci
);
345 __skb_pull(skb
, sizeof(struct vlan_head
));
350 static void clear_vlan(struct sw_flow_key
*key
)
352 key
->eth
.vlan
.tci
= 0;
353 key
->eth
.vlan
.tpid
= 0;
354 key
->eth
.cvlan
.tci
= 0;
355 key
->eth
.cvlan
.tpid
= 0;
358 static int parse_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
)
362 key
->eth
.vlan
.tci
= 0;
363 key
->eth
.vlan
.tpid
= 0;
364 key
->eth
.cvlan
.tci
= 0;
365 key
->eth
.cvlan
.tpid
= 0;
367 if (skb_vlan_tag_present(skb
)) {
368 key
->eth
.vlan
.tci
= htons(skb
->vlan_tci
) | htons(VLAN_CFI_MASK
);
369 key
->eth
.vlan
.tpid
= skb
->vlan_proto
;
371 /* Parse outer vlan tag in the non-accelerated case. */
372 res
= parse_vlan_tag(skb
, &key
->eth
.vlan
, true);
377 /* Parse inner vlan tag. */
378 res
= parse_vlan_tag(skb
, &key
->eth
.cvlan
, false);
385 static __be16
parse_ethertype(struct sk_buff
*skb
)
387 struct llc_snap_hdr
{
388 u8 dsap
; /* Always 0xAA */
389 u8 ssap
; /* Always 0xAA */
394 struct llc_snap_hdr
*llc
;
397 proto
= *(__be16
*) skb
->data
;
398 __skb_pull(skb
, sizeof(__be16
));
400 if (eth_proto_is_802_3(proto
))
403 if (skb
->len
< sizeof(struct llc_snap_hdr
))
404 return htons(ETH_P_802_2
);
406 if (unlikely(!pskb_may_pull(skb
, sizeof(struct llc_snap_hdr
))))
409 llc
= (struct llc_snap_hdr
*) skb
->data
;
410 if (llc
->dsap
!= LLC_SAP_SNAP
||
411 llc
->ssap
!= LLC_SAP_SNAP
||
412 (llc
->oui
[0] | llc
->oui
[1] | llc
->oui
[2]) != 0)
413 return htons(ETH_P_802_2
);
415 __skb_pull(skb
, sizeof(struct llc_snap_hdr
));
417 if (eth_proto_is_802_3(llc
->ethertype
))
418 return llc
->ethertype
;
420 return htons(ETH_P_802_2
);
423 static int parse_icmpv6(struct sk_buff
*skb
, struct sw_flow_key
*key
,
426 struct icmp6hdr
*icmp
= icmp6_hdr(skb
);
428 /* The ICMPv6 type and code fields use the 16-bit transport port
429 * fields, so we need to store them in 16-bit network byte order.
431 key
->tp
.src
= htons(icmp
->icmp6_type
);
432 key
->tp
.dst
= htons(icmp
->icmp6_code
);
433 memset(&key
->ipv6
.nd
, 0, sizeof(key
->ipv6
.nd
));
435 if (icmp
->icmp6_code
== 0 &&
436 (icmp
->icmp6_type
== NDISC_NEIGHBOUR_SOLICITATION
||
437 icmp
->icmp6_type
== NDISC_NEIGHBOUR_ADVERTISEMENT
)) {
438 int icmp_len
= skb
->len
- skb_transport_offset(skb
);
442 /* In order to process neighbor discovery options, we need the
445 if (unlikely(icmp_len
< sizeof(*nd
)))
448 if (unlikely(skb_linearize(skb
)))
451 nd
= (struct nd_msg
*)skb_transport_header(skb
);
452 key
->ipv6
.nd
.target
= nd
->target
;
454 icmp_len
-= sizeof(*nd
);
456 while (icmp_len
>= 8) {
457 struct nd_opt_hdr
*nd_opt
=
458 (struct nd_opt_hdr
*)(nd
->opt
+ offset
);
459 int opt_len
= nd_opt
->nd_opt_len
* 8;
461 if (unlikely(!opt_len
|| opt_len
> icmp_len
))
464 /* Store the link layer address if the appropriate
465 * option is provided. It is considered an error if
466 * the same link layer option is specified twice.
468 if (nd_opt
->nd_opt_type
== ND_OPT_SOURCE_LL_ADDR
470 if (unlikely(!is_zero_ether_addr(key
->ipv6
.nd
.sll
)))
472 ether_addr_copy(key
->ipv6
.nd
.sll
,
473 &nd
->opt
[offset
+sizeof(*nd_opt
)]);
474 } else if (nd_opt
->nd_opt_type
== ND_OPT_TARGET_LL_ADDR
476 if (unlikely(!is_zero_ether_addr(key
->ipv6
.nd
.tll
)))
478 ether_addr_copy(key
->ipv6
.nd
.tll
,
479 &nd
->opt
[offset
+sizeof(*nd_opt
)]);
490 memset(&key
->ipv6
.nd
.target
, 0, sizeof(key
->ipv6
.nd
.target
));
491 memset(key
->ipv6
.nd
.sll
, 0, sizeof(key
->ipv6
.nd
.sll
));
492 memset(key
->ipv6
.nd
.tll
, 0, sizeof(key
->ipv6
.nd
.tll
));
497 static int parse_nsh(struct sk_buff
*skb
, struct sw_flow_key
*key
)
500 unsigned int nh_ofs
= skb_network_offset(skb
);
504 err
= check_header(skb
, nh_ofs
+ NSH_BASE_HDR_LEN
);
509 version
= nsh_get_ver(nh
);
510 length
= nsh_hdr_len(nh
);
515 err
= check_header(skb
, nh_ofs
+ length
);
520 key
->nsh
.base
.flags
= nsh_get_flags(nh
);
521 key
->nsh
.base
.ttl
= nsh_get_ttl(nh
);
522 key
->nsh
.base
.mdtype
= nh
->mdtype
;
523 key
->nsh
.base
.np
= nh
->np
;
524 key
->nsh
.base
.path_hdr
= nh
->path_hdr
;
525 switch (key
->nsh
.base
.mdtype
) {
527 if (length
!= NSH_M_TYPE1_LEN
)
529 memcpy(key
->nsh
.context
, nh
->md1
.context
,
533 memset(key
->nsh
.context
, 0,
544 * key_extract_l3l4 - extracts L3/L4 header information.
545 * @skb: sk_buff that contains the frame, with skb->data pointing to the
547 * @key: output flow key
549 static int key_extract_l3l4(struct sk_buff
*skb
, struct sw_flow_key
*key
)
554 if (key
->eth
.type
== htons(ETH_P_IP
)) {
558 error
= check_iphdr(skb
);
559 if (unlikely(error
)) {
560 memset(&key
->ip
, 0, sizeof(key
->ip
));
561 memset(&key
->ipv4
, 0, sizeof(key
->ipv4
));
562 if (error
== -EINVAL
) {
563 skb
->transport_header
= skb
->network_header
;
570 key
->ipv4
.addr
.src
= nh
->saddr
;
571 key
->ipv4
.addr
.dst
= nh
->daddr
;
573 key
->ip
.proto
= nh
->protocol
;
574 key
->ip
.tos
= nh
->tos
;
575 key
->ip
.ttl
= nh
->ttl
;
577 offset
= nh
->frag_off
& htons(IP_OFFSET
);
579 key
->ip
.frag
= OVS_FRAG_TYPE_LATER
;
580 memset(&key
->tp
, 0, sizeof(key
->tp
));
583 #ifdef HAVE_SKB_GSO_UDP
584 if (nh
->frag_off
& htons(IP_MF
) ||
585 skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
587 if (nh
->frag_off
& htons(IP_MF
))
589 key
->ip
.frag
= OVS_FRAG_TYPE_FIRST
;
591 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
593 /* Transport layer. */
594 if (key
->ip
.proto
== IPPROTO_TCP
) {
595 if (tcphdr_ok(skb
)) {
596 struct tcphdr
*tcp
= tcp_hdr(skb
);
597 key
->tp
.src
= tcp
->source
;
598 key
->tp
.dst
= tcp
->dest
;
599 key
->tp
.flags
= TCP_FLAGS_BE16(tcp
);
601 memset(&key
->tp
, 0, sizeof(key
->tp
));
604 } else if (key
->ip
.proto
== IPPROTO_UDP
) {
605 if (udphdr_ok(skb
)) {
606 struct udphdr
*udp
= udp_hdr(skb
);
607 key
->tp
.src
= udp
->source
;
608 key
->tp
.dst
= udp
->dest
;
610 memset(&key
->tp
, 0, sizeof(key
->tp
));
612 } else if (key
->ip
.proto
== IPPROTO_SCTP
) {
613 if (sctphdr_ok(skb
)) {
614 struct sctphdr
*sctp
= sctp_hdr(skb
);
615 key
->tp
.src
= sctp
->source
;
616 key
->tp
.dst
= sctp
->dest
;
618 memset(&key
->tp
, 0, sizeof(key
->tp
));
620 } else if (key
->ip
.proto
== IPPROTO_ICMP
) {
621 if (icmphdr_ok(skb
)) {
622 struct icmphdr
*icmp
= icmp_hdr(skb
);
623 /* The ICMP type and code fields use the 16-bit
624 * transport port fields, so we need to store
625 * them in 16-bit network byte order.
627 key
->tp
.src
= htons(icmp
->type
);
628 key
->tp
.dst
= htons(icmp
->code
);
630 memset(&key
->tp
, 0, sizeof(key
->tp
));
634 } else if (key
->eth
.type
== htons(ETH_P_ARP
) ||
635 key
->eth
.type
== htons(ETH_P_RARP
)) {
636 struct arp_eth_header
*arp
;
637 bool arp_available
= arphdr_ok(skb
);
639 arp
= (struct arp_eth_header
*)skb_network_header(skb
);
642 arp
->ar_hrd
== htons(ARPHRD_ETHER
) &&
643 arp
->ar_pro
== htons(ETH_P_IP
) &&
644 arp
->ar_hln
== ETH_ALEN
&&
647 /* We only match on the lower 8 bits of the opcode. */
648 if (ntohs(arp
->ar_op
) <= 0xff)
649 key
->ip
.proto
= ntohs(arp
->ar_op
);
653 memcpy(&key
->ipv4
.addr
.src
, arp
->ar_sip
, sizeof(key
->ipv4
.addr
.src
));
654 memcpy(&key
->ipv4
.addr
.dst
, arp
->ar_tip
, sizeof(key
->ipv4
.addr
.dst
));
655 ether_addr_copy(key
->ipv4
.arp
.sha
, arp
->ar_sha
);
656 ether_addr_copy(key
->ipv4
.arp
.tha
, arp
->ar_tha
);
658 memset(&key
->ip
, 0, sizeof(key
->ip
));
659 memset(&key
->ipv4
, 0, sizeof(key
->ipv4
));
661 } else if (eth_p_mpls(key
->eth
.type
)) {
662 size_t stack_len
= MPLS_HLEN
;
664 skb_set_inner_network_header(skb
, skb
->mac_len
);
668 error
= check_header(skb
, skb
->mac_len
+ stack_len
);
672 memcpy(&lse
, skb_inner_network_header(skb
), MPLS_HLEN
);
674 if (stack_len
== MPLS_HLEN
)
675 memcpy(&key
->mpls
.top_lse
, &lse
, MPLS_HLEN
);
677 skb_set_inner_network_header(skb
, skb
->mac_len
+ stack_len
);
678 if (lse
& htonl(MPLS_LS_S_MASK
))
681 stack_len
+= MPLS_HLEN
;
683 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
684 int nh_len
; /* IPv6 Header + Extensions */
686 nh_len
= parse_ipv6hdr(skb
, key
);
687 if (unlikely(nh_len
< 0)) {
690 memset(&key
->ip
, 0, sizeof(key
->ip
));
691 memset(&key
->ipv6
.addr
, 0, sizeof(key
->ipv6
.addr
));
694 skb
->transport_header
= skb
->network_header
;
703 if (key
->ip
.frag
== OVS_FRAG_TYPE_LATER
) {
704 memset(&key
->tp
, 0, sizeof(key
->tp
));
707 #ifdef HAVE_SKB_GSO_UDP
708 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
709 key
->ip
.frag
= OVS_FRAG_TYPE_FIRST
;
712 /* Transport layer. */
713 if (key
->ip
.proto
== NEXTHDR_TCP
) {
714 if (tcphdr_ok(skb
)) {
715 struct tcphdr
*tcp
= tcp_hdr(skb
);
716 key
->tp
.src
= tcp
->source
;
717 key
->tp
.dst
= tcp
->dest
;
718 key
->tp
.flags
= TCP_FLAGS_BE16(tcp
);
720 memset(&key
->tp
, 0, sizeof(key
->tp
));
722 } else if (key
->ip
.proto
== NEXTHDR_UDP
) {
723 if (udphdr_ok(skb
)) {
724 struct udphdr
*udp
= udp_hdr(skb
);
725 key
->tp
.src
= udp
->source
;
726 key
->tp
.dst
= udp
->dest
;
728 memset(&key
->tp
, 0, sizeof(key
->tp
));
730 } else if (key
->ip
.proto
== NEXTHDR_SCTP
) {
731 if (sctphdr_ok(skb
)) {
732 struct sctphdr
*sctp
= sctp_hdr(skb
);
733 key
->tp
.src
= sctp
->source
;
734 key
->tp
.dst
= sctp
->dest
;
736 memset(&key
->tp
, 0, sizeof(key
->tp
));
738 } else if (key
->ip
.proto
== NEXTHDR_ICMP
) {
739 if (icmp6hdr_ok(skb
)) {
740 error
= parse_icmpv6(skb
, key
, nh_len
);
744 memset(&key
->tp
, 0, sizeof(key
->tp
));
747 } else if (key
->eth
.type
== htons(ETH_P_NSH
)) {
748 error
= parse_nsh(skb
, key
);
756 * key_extract - extracts a flow key from an Ethernet frame.
757 * @skb: sk_buff that contains the frame, with skb->data pointing to the
759 * @key: output flow key
761 * The caller must ensure that skb->len >= ETH_HLEN.
763 * Returns 0 if successful, otherwise a negative errno value.
765 * Initializes @skb header fields as follows:
767 * - skb->mac_header: the L2 header.
769 * - skb->network_header: just past the L2 header, or just past the
770 * VLAN header, to the first byte of the L2 payload.
772 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
773 * on output, then just past the IP header, if one is present and
774 * of a correct length, otherwise the same as skb->network_header.
775 * For other key->eth.type values it is left untouched.
777 * - skb->protocol: the type of the data starting at skb->network_header.
778 * Equals to key->eth.type.
780 static int key_extract(struct sk_buff
*skb
, struct sw_flow_key
*key
)
784 /* Flags are always used as part of stats */
787 skb_reset_mac_header(skb
);
791 if (ovs_key_mac_proto(key
) == MAC_PROTO_NONE
) {
792 if (unlikely(eth_type_vlan(skb
->protocol
)))
795 skb_reset_network_header(skb
);
796 key
->eth
.type
= skb
->protocol
;
799 ether_addr_copy(key
->eth
.src
, eth
->h_source
);
800 ether_addr_copy(key
->eth
.dst
, eth
->h_dest
);
802 __skb_pull(skb
, 2 * ETH_ALEN
);
803 /* We are going to push all headers that we pull, so no need to
804 * update skb->csum here.
807 if (unlikely(parse_vlan(skb
, key
)))
810 key
->eth
.type
= parse_ethertype(skb
);
811 if (unlikely(key
->eth
.type
== htons(0)))
814 /* Multiple tagged packets need to retain TPID to satisfy
815 * skb_vlan_pop(), which will later shift the ethertype into
818 if (key
->eth
.cvlan
.tci
& htons(VLAN_CFI_MASK
))
819 skb
->protocol
= key
->eth
.cvlan
.tpid
;
821 skb
->protocol
= key
->eth
.type
;
823 skb_reset_network_header(skb
);
824 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
827 skb_reset_mac_len(skb
);
829 /* Fill out L3/L4 key info, if any */
830 return key_extract_l3l4(skb
, key
);
833 /* In the case of conntrack fragment handling it expects L3 headers,
836 int ovs_flow_key_update_l3l4(struct sk_buff
*skb
, struct sw_flow_key
*key
)
838 return key_extract_l3l4(skb
, key
);
841 int ovs_flow_key_update(struct sk_buff
*skb
, struct sw_flow_key
*key
)
845 res
= key_extract(skb
, key
);
847 key
->mac_proto
&= ~SW_FLOW_KEY_INVALID
;
852 static int key_extract_mac_proto(struct sk_buff
*skb
)
854 switch (skb
->dev
->type
) {
856 return MAC_PROTO_ETHERNET
;
858 if (skb
->protocol
== htons(ETH_P_TEB
))
859 return MAC_PROTO_ETHERNET
;
860 return MAC_PROTO_NONE
;
866 int ovs_flow_key_extract(const struct ip_tunnel_info
*tun_info
,
867 struct sk_buff
*skb
, struct sw_flow_key
*key
)
871 /* Extract metadata from packet. */
873 key
->tun_proto
= ip_tunnel_info_af(tun_info
);
874 memcpy(&key
->tun_key
, &tun_info
->key
, sizeof(key
->tun_key
));
875 BUILD_BUG_ON(((1 << (sizeof(tun_info
->options_len
) * 8)) - 1) >
876 sizeof(key
->tun_opts
));
878 if (tun_info
->options_len
) {
879 ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key
, tun_info
->options_len
),
881 key
->tun_opts_len
= tun_info
->options_len
;
883 key
->tun_opts_len
= 0;
887 key
->tun_opts_len
= 0;
888 memset(&key
->tun_key
, 0, sizeof(key
->tun_key
));
891 key
->phy
.priority
= skb
->priority
;
892 key
->phy
.in_port
= OVS_CB(skb
)->input_vport
->port_no
;
893 key
->phy
.skb_mark
= skb
->mark
;
894 key
->ovs_flow_hash
= 0;
895 res
= key_extract_mac_proto(skb
);
898 key
->mac_proto
= res
;
901 err
= key_extract(skb
, key
);
903 ovs_ct_fill_key(skb
, key
); /* Must be after key_extract(). */
907 int ovs_flow_key_extract_userspace(struct net
*net
, const struct nlattr
*attr
,
909 struct sw_flow_key
*key
, bool log
)
911 const struct nlattr
*a
[OVS_KEY_ATTR_MAX
+ 1];
915 err
= parse_flow_nlattrs(attr
, a
, &attrs
, log
);
919 /* Extract metadata from netlink attributes. */
920 err
= ovs_nla_get_flow_metadata(net
, a
, attrs
, key
, log
);
924 /* key_extract assumes that skb->protocol is set-up for
925 * layer 3 packets which is the case for other callers,
926 * in particular packets received from the network stack.
927 * Here the correct value can be set from the metadata
929 * For L2 packet key eth type would be zero. skb protocol
930 * would be set to correct value later during key-extact.
933 skb
->protocol
= key
->eth
.type
;
934 err
= key_extract(skb
, key
);
938 /* Check that we have conntrack original direction tuple metadata only
939 * for packets for which it makes sense. Otherwise the key may be
940 * corrupted due to overlapping key fields.
942 if (attrs
& (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4
) &&
943 key
->eth
.type
!= htons(ETH_P_IP
))
945 if (attrs
& (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6
) &&
946 (key
->eth
.type
!= htons(ETH_P_IPV6
) ||
947 sw_flow_key_is_nd(key
)))