1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2007-2014 Nicira, Inc.
6 #include <linux/uaccess.h>
7 #include <linux/netdevice.h>
8 #include <linux/etherdevice.h>
9 #include <linux/if_ether.h>
10 #include <linux/if_vlan.h>
11 #include <net/llc_pdu.h>
12 #include <linux/kernel.h>
13 #include <linux/jhash.h>
14 #include <linux/jiffies.h>
15 #include <linux/llc.h>
16 #include <linux/module.h>
18 #include <linux/rcupdate.h>
19 #include <linux/cpumask.h>
20 #include <linux/if_arp.h>
22 #include <linux/ipv6.h>
23 #include <linux/mpls.h>
24 #include <linux/sctp.h>
25 #include <linux/smp.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/icmp.h>
29 #include <linux/icmpv6.h>
30 #include <linux/rculist.h>
32 #include <net/ip_tunnels.h>
35 #include <net/ndisc.h>
38 #include "conntrack.h"
41 #include "flow_netlink.h"
44 u64
ovs_flow_used_time(unsigned long flow_jiffies
)
46 struct timespec64 cur_ts
;
49 ktime_get_ts64(&cur_ts
);
50 idle_ms
= jiffies_to_msecs(jiffies
- flow_jiffies
);
51 cur_ms
= (u64
)(u32
)cur_ts
.tv_sec
* MSEC_PER_SEC
+
52 cur_ts
.tv_nsec
/ NSEC_PER_MSEC
;
54 return cur_ms
- idle_ms
;
57 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
59 void ovs_flow_stats_update(struct sw_flow
*flow
, __be16 tcp_flags
,
60 const struct sk_buff
*skb
)
62 struct sw_flow_stats
*stats
;
63 unsigned int cpu
= smp_processor_id();
64 int len
= skb
->len
+ (skb_vlan_tag_present(skb
) ? VLAN_HLEN
: 0);
66 stats
= rcu_dereference(flow
->stats
[cpu
]);
68 /* Check if already have CPU-specific stats. */
70 spin_lock(&stats
->lock
);
71 /* Mark if we write on the pre-allocated stats. */
72 if (cpu
== 0 && unlikely(flow
->stats_last_writer
!= cpu
))
73 flow
->stats_last_writer
= cpu
;
75 stats
= rcu_dereference(flow
->stats
[0]); /* Pre-allocated. */
76 spin_lock(&stats
->lock
);
78 /* If the current CPU is the only writer on the
79 * pre-allocated stats keep using them.
81 if (unlikely(flow
->stats_last_writer
!= cpu
)) {
82 /* A previous locker may have already allocated the
83 * stats, so we need to check again. If CPU-specific
84 * stats were already allocated, we update the pre-
85 * allocated stats as we have already locked them.
87 if (likely(flow
->stats_last_writer
!= -1) &&
88 likely(!rcu_access_pointer(flow
->stats
[cpu
]))) {
89 /* Try to allocate CPU-specific stats. */
90 struct sw_flow_stats
*new_stats
;
93 kmem_cache_alloc_node(flow_stats_cache
,
99 if (likely(new_stats
)) {
100 new_stats
->used
= jiffies
;
101 new_stats
->packet_count
= 1;
102 new_stats
->byte_count
= len
;
103 new_stats
->tcp_flags
= tcp_flags
;
104 spin_lock_init(&new_stats
->lock
);
106 rcu_assign_pointer(flow
->stats
[cpu
],
108 cpumask_set_cpu(cpu
, &flow
->cpu_used_mask
);
112 flow
->stats_last_writer
= cpu
;
116 stats
->used
= jiffies
;
117 stats
->packet_count
++;
118 stats
->byte_count
+= len
;
119 stats
->tcp_flags
|= tcp_flags
;
121 spin_unlock(&stats
->lock
);
124 /* Must be called with rcu_read_lock or ovs_mutex. */
125 void ovs_flow_stats_get(const struct sw_flow
*flow
,
126 struct ovs_flow_stats
*ovs_stats
,
127 unsigned long *used
, __be16
*tcp_flags
)
133 memset(ovs_stats
, 0, sizeof(*ovs_stats
));
135 /* We open code this to make sure cpu 0 is always considered */
136 for (cpu
= 0; cpu
< nr_cpu_ids
; cpu
= cpumask_next(cpu
, &flow
->cpu_used_mask
)) {
137 struct sw_flow_stats
*stats
= rcu_dereference_ovsl(flow
->stats
[cpu
]);
140 /* Local CPU may write on non-local stats, so we must
141 * block bottom-halves here.
143 spin_lock_bh(&stats
->lock
);
144 if (!*used
|| time_after(stats
->used
, *used
))
146 *tcp_flags
|= stats
->tcp_flags
;
147 ovs_stats
->n_packets
+= stats
->packet_count
;
148 ovs_stats
->n_bytes
+= stats
->byte_count
;
149 spin_unlock_bh(&stats
->lock
);
154 /* Called with ovs_mutex. */
155 void ovs_flow_stats_clear(struct sw_flow
*flow
)
159 /* We open code this to make sure cpu 0 is always considered */
160 for (cpu
= 0; cpu
< nr_cpu_ids
; cpu
= cpumask_next(cpu
, &flow
->cpu_used_mask
)) {
161 struct sw_flow_stats
*stats
= ovsl_dereference(flow
->stats
[cpu
]);
164 spin_lock_bh(&stats
->lock
);
166 stats
->packet_count
= 0;
167 stats
->byte_count
= 0;
168 stats
->tcp_flags
= 0;
169 spin_unlock_bh(&stats
->lock
);
174 static int check_header(struct sk_buff
*skb
, int len
)
176 if (unlikely(skb
->len
< len
))
178 if (unlikely(!pskb_may_pull(skb
, len
)))
183 static bool arphdr_ok(struct sk_buff
*skb
)
185 return pskb_may_pull(skb
, skb_network_offset(skb
) +
186 sizeof(struct arp_eth_header
));
189 static int check_iphdr(struct sk_buff
*skb
)
191 unsigned int nh_ofs
= skb_network_offset(skb
);
195 err
= check_header(skb
, nh_ofs
+ sizeof(struct iphdr
));
199 ip_len
= ip_hdrlen(skb
);
200 if (unlikely(ip_len
< sizeof(struct iphdr
) ||
201 skb
->len
< nh_ofs
+ ip_len
))
204 skb_set_transport_header(skb
, nh_ofs
+ ip_len
);
208 static bool tcphdr_ok(struct sk_buff
*skb
)
210 int th_ofs
= skb_transport_offset(skb
);
213 if (unlikely(!pskb_may_pull(skb
, th_ofs
+ sizeof(struct tcphdr
))))
216 tcp_len
= tcp_hdrlen(skb
);
217 if (unlikely(tcp_len
< sizeof(struct tcphdr
) ||
218 skb
->len
< th_ofs
+ tcp_len
))
224 static bool udphdr_ok(struct sk_buff
*skb
)
226 return pskb_may_pull(skb
, skb_transport_offset(skb
) +
227 sizeof(struct udphdr
));
230 static bool sctphdr_ok(struct sk_buff
*skb
)
232 return pskb_may_pull(skb
, skb_transport_offset(skb
) +
233 sizeof(struct sctphdr
));
236 static bool icmphdr_ok(struct sk_buff
*skb
)
238 return pskb_may_pull(skb
, skb_transport_offset(skb
) +
239 sizeof(struct icmphdr
));
242 static int parse_ipv6hdr(struct sk_buff
*skb
, struct sw_flow_key
*key
)
244 unsigned short frag_off
;
245 unsigned int payload_ofs
= 0;
246 unsigned int nh_ofs
= skb_network_offset(skb
);
249 int err
, nexthdr
, flags
= 0;
251 err
= check_header(skb
, nh_ofs
+ sizeof(*nh
));
257 key
->ip
.proto
= NEXTHDR_NONE
;
258 key
->ip
.tos
= ipv6_get_dsfield(nh
);
259 key
->ip
.ttl
= nh
->hop_limit
;
260 key
->ipv6
.label
= *(__be32
*)nh
& htonl(IPV6_FLOWINFO_FLOWLABEL
);
261 key
->ipv6
.addr
.src
= nh
->saddr
;
262 key
->ipv6
.addr
.dst
= nh
->daddr
;
264 nexthdr
= ipv6_find_hdr(skb
, &payload_ofs
, -1, &frag_off
, &flags
);
265 if (flags
& IP6_FH_F_FRAG
) {
267 key
->ip
.frag
= OVS_FRAG_TYPE_LATER
;
268 key
->ip
.proto
= nexthdr
;
271 key
->ip
.frag
= OVS_FRAG_TYPE_FIRST
;
273 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
276 /* Delayed handling of error in ipv6_find_hdr() as it
277 * always sets flags and frag_off to a valid value which may be
278 * used to set key->ip.frag above.
280 if (unlikely(nexthdr
< 0))
283 nh_len
= payload_ofs
- nh_ofs
;
284 skb_set_transport_header(skb
, nh_ofs
+ nh_len
);
285 key
->ip
.proto
= nexthdr
;
289 static bool icmp6hdr_ok(struct sk_buff
*skb
)
291 return pskb_may_pull(skb
, skb_transport_offset(skb
) +
292 sizeof(struct icmp6hdr
));
296 * parse_vlan_tag - Parse vlan tag from vlan header.
297 * @skb: skb containing frame to parse
298 * @key_vh: pointer to parsed vlan tag
299 * @untag_vlan: should the vlan header be removed from the frame
301 * Return: ERROR on memory error.
302 * %0 if it encounters a non-vlan or incomplete packet.
303 * %1 after successfully parsing vlan tag.
305 static int parse_vlan_tag(struct sk_buff
*skb
, struct vlan_head
*key_vh
,
308 struct vlan_head
*vh
= (struct vlan_head
*)skb
->data
;
310 if (likely(!eth_type_vlan(vh
->tpid
)))
313 if (unlikely(skb
->len
< sizeof(struct vlan_head
) + sizeof(__be16
)))
316 if (unlikely(!pskb_may_pull(skb
, sizeof(struct vlan_head
) +
320 vh
= (struct vlan_head
*)skb
->data
;
321 key_vh
->tci
= vh
->tci
| htons(VLAN_CFI_MASK
);
322 key_vh
->tpid
= vh
->tpid
;
324 if (unlikely(untag_vlan
)) {
325 int offset
= skb
->data
- skb_mac_header(skb
);
329 __skb_push(skb
, offset
);
330 err
= __skb_vlan_pop(skb
, &tci
);
331 __skb_pull(skb
, offset
);
334 __vlan_hwaccel_put_tag(skb
, key_vh
->tpid
, tci
);
336 __skb_pull(skb
, sizeof(struct vlan_head
));
341 static void clear_vlan(struct sw_flow_key
*key
)
343 key
->eth
.vlan
.tci
= 0;
344 key
->eth
.vlan
.tpid
= 0;
345 key
->eth
.cvlan
.tci
= 0;
346 key
->eth
.cvlan
.tpid
= 0;
349 static int parse_vlan(struct sk_buff
*skb
, struct sw_flow_key
*key
)
353 if (skb_vlan_tag_present(skb
)) {
354 key
->eth
.vlan
.tci
= htons(skb
->vlan_tci
) | htons(VLAN_CFI_MASK
);
355 key
->eth
.vlan
.tpid
= skb
->vlan_proto
;
357 /* Parse outer vlan tag in the non-accelerated case. */
358 res
= parse_vlan_tag(skb
, &key
->eth
.vlan
, true);
363 /* Parse inner vlan tag. */
364 res
= parse_vlan_tag(skb
, &key
->eth
.cvlan
, false);
371 static __be16
parse_ethertype(struct sk_buff
*skb
)
373 struct llc_snap_hdr
{
374 u8 dsap
; /* Always 0xAA */
375 u8 ssap
; /* Always 0xAA */
380 struct llc_snap_hdr
*llc
;
383 proto
= *(__be16
*) skb
->data
;
384 __skb_pull(skb
, sizeof(__be16
));
386 if (eth_proto_is_802_3(proto
))
389 if (skb
->len
< sizeof(struct llc_snap_hdr
))
390 return htons(ETH_P_802_2
);
392 if (unlikely(!pskb_may_pull(skb
, sizeof(struct llc_snap_hdr
))))
395 llc
= (struct llc_snap_hdr
*) skb
->data
;
396 if (llc
->dsap
!= LLC_SAP_SNAP
||
397 llc
->ssap
!= LLC_SAP_SNAP
||
398 (llc
->oui
[0] | llc
->oui
[1] | llc
->oui
[2]) != 0)
399 return htons(ETH_P_802_2
);
401 __skb_pull(skb
, sizeof(struct llc_snap_hdr
));
403 if (eth_proto_is_802_3(llc
->ethertype
))
404 return llc
->ethertype
;
406 return htons(ETH_P_802_2
);
409 static int parse_icmpv6(struct sk_buff
*skb
, struct sw_flow_key
*key
,
412 struct icmp6hdr
*icmp
= icmp6_hdr(skb
);
414 /* The ICMPv6 type and code fields use the 16-bit transport port
415 * fields, so we need to store them in 16-bit network byte order.
417 key
->tp
.src
= htons(icmp
->icmp6_type
);
418 key
->tp
.dst
= htons(icmp
->icmp6_code
);
419 memset(&key
->ipv6
.nd
, 0, sizeof(key
->ipv6
.nd
));
421 if (icmp
->icmp6_code
== 0 &&
422 (icmp
->icmp6_type
== NDISC_NEIGHBOUR_SOLICITATION
||
423 icmp
->icmp6_type
== NDISC_NEIGHBOUR_ADVERTISEMENT
)) {
424 int icmp_len
= skb
->len
- skb_transport_offset(skb
);
428 /* In order to process neighbor discovery options, we need the
431 if (unlikely(icmp_len
< sizeof(*nd
)))
434 if (unlikely(skb_linearize(skb
)))
437 nd
= (struct nd_msg
*)skb_transport_header(skb
);
438 key
->ipv6
.nd
.target
= nd
->target
;
440 icmp_len
-= sizeof(*nd
);
442 while (icmp_len
>= 8) {
443 struct nd_opt_hdr
*nd_opt
=
444 (struct nd_opt_hdr
*)(nd
->opt
+ offset
);
445 int opt_len
= nd_opt
->nd_opt_len
* 8;
447 if (unlikely(!opt_len
|| opt_len
> icmp_len
))
450 /* Store the link layer address if the appropriate
451 * option is provided. It is considered an error if
452 * the same link layer option is specified twice.
454 if (nd_opt
->nd_opt_type
== ND_OPT_SOURCE_LL_ADDR
456 if (unlikely(!is_zero_ether_addr(key
->ipv6
.nd
.sll
)))
458 ether_addr_copy(key
->ipv6
.nd
.sll
,
459 &nd
->opt
[offset
+sizeof(*nd_opt
)]);
460 } else if (nd_opt
->nd_opt_type
== ND_OPT_TARGET_LL_ADDR
462 if (unlikely(!is_zero_ether_addr(key
->ipv6
.nd
.tll
)))
464 ether_addr_copy(key
->ipv6
.nd
.tll
,
465 &nd
->opt
[offset
+sizeof(*nd_opt
)]);
476 memset(&key
->ipv6
.nd
.target
, 0, sizeof(key
->ipv6
.nd
.target
));
477 memset(key
->ipv6
.nd
.sll
, 0, sizeof(key
->ipv6
.nd
.sll
));
478 memset(key
->ipv6
.nd
.tll
, 0, sizeof(key
->ipv6
.nd
.tll
));
483 static int parse_nsh(struct sk_buff
*skb
, struct sw_flow_key
*key
)
486 unsigned int nh_ofs
= skb_network_offset(skb
);
490 err
= check_header(skb
, nh_ofs
+ NSH_BASE_HDR_LEN
);
495 version
= nsh_get_ver(nh
);
496 length
= nsh_hdr_len(nh
);
501 err
= check_header(skb
, nh_ofs
+ length
);
506 key
->nsh
.base
.flags
= nsh_get_flags(nh
);
507 key
->nsh
.base
.ttl
= nsh_get_ttl(nh
);
508 key
->nsh
.base
.mdtype
= nh
->mdtype
;
509 key
->nsh
.base
.np
= nh
->np
;
510 key
->nsh
.base
.path_hdr
= nh
->path_hdr
;
511 switch (key
->nsh
.base
.mdtype
) {
513 if (length
!= NSH_M_TYPE1_LEN
)
515 memcpy(key
->nsh
.context
, nh
->md1
.context
,
519 memset(key
->nsh
.context
, 0,
530 * key_extract_l3l4 - extracts L3/L4 header information.
531 * @skb: sk_buff that contains the frame, with skb->data pointing to the
533 * @key: output flow key
535 * Return: %0 if successful, otherwise a negative errno value.
537 static int key_extract_l3l4(struct sk_buff
*skb
, struct sw_flow_key
*key
)
542 if (key
->eth
.type
== htons(ETH_P_IP
)) {
546 error
= check_iphdr(skb
);
547 if (unlikely(error
)) {
548 memset(&key
->ip
, 0, sizeof(key
->ip
));
549 memset(&key
->ipv4
, 0, sizeof(key
->ipv4
));
550 if (error
== -EINVAL
) {
551 skb
->transport_header
= skb
->network_header
;
558 key
->ipv4
.addr
.src
= nh
->saddr
;
559 key
->ipv4
.addr
.dst
= nh
->daddr
;
561 key
->ip
.proto
= nh
->protocol
;
562 key
->ip
.tos
= nh
->tos
;
563 key
->ip
.ttl
= nh
->ttl
;
565 offset
= nh
->frag_off
& htons(IP_OFFSET
);
567 key
->ip
.frag
= OVS_FRAG_TYPE_LATER
;
568 memset(&key
->tp
, 0, sizeof(key
->tp
));
571 if (nh
->frag_off
& htons(IP_MF
) ||
572 skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
573 key
->ip
.frag
= OVS_FRAG_TYPE_FIRST
;
575 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
577 /* Transport layer. */
578 if (key
->ip
.proto
== IPPROTO_TCP
) {
579 if (tcphdr_ok(skb
)) {
580 struct tcphdr
*tcp
= tcp_hdr(skb
);
581 key
->tp
.src
= tcp
->source
;
582 key
->tp
.dst
= tcp
->dest
;
583 key
->tp
.flags
= TCP_FLAGS_BE16(tcp
);
585 memset(&key
->tp
, 0, sizeof(key
->tp
));
588 } else if (key
->ip
.proto
== IPPROTO_UDP
) {
589 if (udphdr_ok(skb
)) {
590 struct udphdr
*udp
= udp_hdr(skb
);
591 key
->tp
.src
= udp
->source
;
592 key
->tp
.dst
= udp
->dest
;
594 memset(&key
->tp
, 0, sizeof(key
->tp
));
596 } else if (key
->ip
.proto
== IPPROTO_SCTP
) {
597 if (sctphdr_ok(skb
)) {
598 struct sctphdr
*sctp
= sctp_hdr(skb
);
599 key
->tp
.src
= sctp
->source
;
600 key
->tp
.dst
= sctp
->dest
;
602 memset(&key
->tp
, 0, sizeof(key
->tp
));
604 } else if (key
->ip
.proto
== IPPROTO_ICMP
) {
605 if (icmphdr_ok(skb
)) {
606 struct icmphdr
*icmp
= icmp_hdr(skb
);
607 /* The ICMP type and code fields use the 16-bit
608 * transport port fields, so we need to store
609 * them in 16-bit network byte order. */
610 key
->tp
.src
= htons(icmp
->type
);
611 key
->tp
.dst
= htons(icmp
->code
);
613 memset(&key
->tp
, 0, sizeof(key
->tp
));
617 } else if (key
->eth
.type
== htons(ETH_P_ARP
) ||
618 key
->eth
.type
== htons(ETH_P_RARP
)) {
619 struct arp_eth_header
*arp
;
620 bool arp_available
= arphdr_ok(skb
);
622 arp
= (struct arp_eth_header
*)skb_network_header(skb
);
625 arp
->ar_hrd
== htons(ARPHRD_ETHER
) &&
626 arp
->ar_pro
== htons(ETH_P_IP
) &&
627 arp
->ar_hln
== ETH_ALEN
&&
630 /* We only match on the lower 8 bits of the opcode. */
631 if (ntohs(arp
->ar_op
) <= 0xff)
632 key
->ip
.proto
= ntohs(arp
->ar_op
);
636 memcpy(&key
->ipv4
.addr
.src
, arp
->ar_sip
, sizeof(key
->ipv4
.addr
.src
));
637 memcpy(&key
->ipv4
.addr
.dst
, arp
->ar_tip
, sizeof(key
->ipv4
.addr
.dst
));
638 ether_addr_copy(key
->ipv4
.arp
.sha
, arp
->ar_sha
);
639 ether_addr_copy(key
->ipv4
.arp
.tha
, arp
->ar_tha
);
641 memset(&key
->ip
, 0, sizeof(key
->ip
));
642 memset(&key
->ipv4
, 0, sizeof(key
->ipv4
));
644 } else if (eth_p_mpls(key
->eth
.type
)) {
647 memset(&key
->mpls
, 0, sizeof(key
->mpls
));
648 skb_set_inner_network_header(skb
, skb
->mac_len
);
652 error
= check_header(skb
, skb
->mac_len
+
653 label_count
* MPLS_HLEN
);
657 memcpy(&lse
, skb_inner_network_header(skb
), MPLS_HLEN
);
659 if (label_count
<= MPLS_LABEL_DEPTH
)
660 memcpy(&key
->mpls
.lse
[label_count
- 1], &lse
,
663 skb_set_inner_network_header(skb
, skb
->mac_len
+
664 label_count
* MPLS_HLEN
);
665 if (lse
& htonl(MPLS_LS_S_MASK
))
670 if (label_count
> MPLS_LABEL_DEPTH
)
671 label_count
= MPLS_LABEL_DEPTH
;
673 key
->mpls
.num_labels_mask
= GENMASK(label_count
- 1, 0);
674 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
675 int nh_len
; /* IPv6 Header + Extensions */
677 nh_len
= parse_ipv6hdr(skb
, key
);
678 if (unlikely(nh_len
< 0)) {
681 memset(&key
->ip
, 0, sizeof(key
->ip
));
682 memset(&key
->ipv6
.addr
, 0, sizeof(key
->ipv6
.addr
));
685 skb
->transport_header
= skb
->network_header
;
694 if (key
->ip
.frag
== OVS_FRAG_TYPE_LATER
) {
695 memset(&key
->tp
, 0, sizeof(key
->tp
));
698 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
699 key
->ip
.frag
= OVS_FRAG_TYPE_FIRST
;
701 /* Transport layer. */
702 if (key
->ip
.proto
== NEXTHDR_TCP
) {
703 if (tcphdr_ok(skb
)) {
704 struct tcphdr
*tcp
= tcp_hdr(skb
);
705 key
->tp
.src
= tcp
->source
;
706 key
->tp
.dst
= tcp
->dest
;
707 key
->tp
.flags
= TCP_FLAGS_BE16(tcp
);
709 memset(&key
->tp
, 0, sizeof(key
->tp
));
711 } else if (key
->ip
.proto
== NEXTHDR_UDP
) {
712 if (udphdr_ok(skb
)) {
713 struct udphdr
*udp
= udp_hdr(skb
);
714 key
->tp
.src
= udp
->source
;
715 key
->tp
.dst
= udp
->dest
;
717 memset(&key
->tp
, 0, sizeof(key
->tp
));
719 } else if (key
->ip
.proto
== NEXTHDR_SCTP
) {
720 if (sctphdr_ok(skb
)) {
721 struct sctphdr
*sctp
= sctp_hdr(skb
);
722 key
->tp
.src
= sctp
->source
;
723 key
->tp
.dst
= sctp
->dest
;
725 memset(&key
->tp
, 0, sizeof(key
->tp
));
727 } else if (key
->ip
.proto
== NEXTHDR_ICMP
) {
728 if (icmp6hdr_ok(skb
)) {
729 error
= parse_icmpv6(skb
, key
, nh_len
);
733 memset(&key
->tp
, 0, sizeof(key
->tp
));
736 } else if (key
->eth
.type
== htons(ETH_P_NSH
)) {
737 error
= parse_nsh(skb
, key
);
745 * key_extract - extracts a flow key from an Ethernet frame.
746 * @skb: sk_buff that contains the frame, with skb->data pointing to the
748 * @key: output flow key
750 * The caller must ensure that skb->len >= ETH_HLEN.
752 * Initializes @skb header fields as follows:
754 * - skb->mac_header: the L2 header.
756 * - skb->network_header: just past the L2 header, or just past the
757 * VLAN header, to the first byte of the L2 payload.
759 * - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
760 * on output, then just past the IP header, if one is present and
761 * of a correct length, otherwise the same as skb->network_header.
762 * For other key->eth.type values it is left untouched.
764 * - skb->protocol: the type of the data starting at skb->network_header.
765 * Equals to key->eth.type.
767 * Return: %0 if successful, otherwise a negative errno value.
769 static int key_extract(struct sk_buff
*skb
, struct sw_flow_key
*key
)
773 /* Flags are always used as part of stats */
776 skb_reset_mac_header(skb
);
780 if (ovs_key_mac_proto(key
) == MAC_PROTO_NONE
) {
781 if (unlikely(eth_type_vlan(skb
->protocol
)))
784 skb_reset_network_header(skb
);
785 key
->eth
.type
= skb
->protocol
;
788 ether_addr_copy(key
->eth
.src
, eth
->h_source
);
789 ether_addr_copy(key
->eth
.dst
, eth
->h_dest
);
791 __skb_pull(skb
, 2 * ETH_ALEN
);
792 /* We are going to push all headers that we pull, so no need to
793 * update skb->csum here.
796 if (unlikely(parse_vlan(skb
, key
)))
799 key
->eth
.type
= parse_ethertype(skb
);
800 if (unlikely(key
->eth
.type
== htons(0)))
803 /* Multiple tagged packets need to retain TPID to satisfy
804 * skb_vlan_pop(), which will later shift the ethertype into
807 if (key
->eth
.cvlan
.tci
& htons(VLAN_CFI_MASK
))
808 skb
->protocol
= key
->eth
.cvlan
.tpid
;
810 skb
->protocol
= key
->eth
.type
;
812 skb_reset_network_header(skb
);
813 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
816 skb_reset_mac_len(skb
);
818 /* Fill out L3/L4 key info, if any */
819 return key_extract_l3l4(skb
, key
);
822 /* In the case of conntrack fragment handling it expects L3 headers,
825 int ovs_flow_key_update_l3l4(struct sk_buff
*skb
, struct sw_flow_key
*key
)
827 return key_extract_l3l4(skb
, key
);
830 int ovs_flow_key_update(struct sk_buff
*skb
, struct sw_flow_key
*key
)
834 res
= key_extract(skb
, key
);
836 key
->mac_proto
&= ~SW_FLOW_KEY_INVALID
;
841 static int key_extract_mac_proto(struct sk_buff
*skb
)
843 switch (skb
->dev
->type
) {
845 return MAC_PROTO_ETHERNET
;
847 if (skb
->protocol
== htons(ETH_P_TEB
))
848 return MAC_PROTO_ETHERNET
;
849 return MAC_PROTO_NONE
;
855 int ovs_flow_key_extract(const struct ip_tunnel_info
*tun_info
,
856 struct sk_buff
*skb
, struct sw_flow_key
*key
)
858 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
859 struct tc_skb_ext
*tc_ext
;
861 bool post_ct
= false;
864 /* Extract metadata from packet. */
866 key
->tun_proto
= ip_tunnel_info_af(tun_info
);
867 memcpy(&key
->tun_key
, &tun_info
->key
, sizeof(key
->tun_key
));
869 if (tun_info
->options_len
) {
870 BUILD_BUG_ON((1 << (sizeof(tun_info
->options_len
) *
872 > sizeof(key
->tun_opts
));
874 ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key
, tun_info
->options_len
),
876 key
->tun_opts_len
= tun_info
->options_len
;
878 key
->tun_opts_len
= 0;
882 key
->tun_opts_len
= 0;
883 memset(&key
->tun_key
, 0, sizeof(key
->tun_key
));
886 key
->phy
.priority
= skb
->priority
;
887 key
->phy
.in_port
= OVS_CB(skb
)->input_vport
->port_no
;
888 key
->phy
.skb_mark
= skb
->mark
;
889 key
->ovs_flow_hash
= 0;
890 res
= key_extract_mac_proto(skb
);
893 key
->mac_proto
= res
;
895 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
896 if (static_branch_unlikely(&tc_recirc_sharing_support
)) {
897 tc_ext
= skb_ext_find(skb
, TC_SKB_EXT
);
898 key
->recirc_id
= tc_ext
? tc_ext
->chain
: 0;
899 OVS_CB(skb
)->mru
= tc_ext
? tc_ext
->mru
: 0;
900 post_ct
= tc_ext
? tc_ext
->post_ct
: false;
908 err
= key_extract(skb
, key
);
910 ovs_ct_fill_key(skb
, key
, post_ct
); /* Must be after key_extract(). */
914 int ovs_flow_key_extract_userspace(struct net
*net
, const struct nlattr
*attr
,
916 struct sw_flow_key
*key
, bool log
)
918 const struct nlattr
*a
[OVS_KEY_ATTR_MAX
+ 1];
922 err
= parse_flow_nlattrs(attr
, a
, &attrs
, log
);
926 /* Extract metadata from netlink attributes. */
927 err
= ovs_nla_get_flow_metadata(net
, a
, attrs
, key
, log
);
931 /* key_extract assumes that skb->protocol is set-up for
932 * layer 3 packets which is the case for other callers,
933 * in particular packets received from the network stack.
934 * Here the correct value can be set from the metadata
936 * For L2 packet key eth type would be zero. skb protocol
937 * would be set to correct value later during key-extact.
940 skb
->protocol
= key
->eth
.type
;
941 err
= key_extract(skb
, key
);
945 /* Check that we have conntrack original direction tuple metadata only
946 * for packets for which it makes sense. Otherwise the key may be
947 * corrupted due to overlapping key fields.
949 if (attrs
& (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4
) &&
950 key
->eth
.type
!= htons(ETH_P_IP
))
952 if (attrs
& (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6
) &&
953 (key
->eth
.type
!= htons(ETH_P_IPV6
) ||
954 sw_flow_key_is_nd(key
)))