1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2007-2017 Nicira, Inc.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #include <linux/skbuff.h>
11 #include <linux/openvswitch.h>
12 #include <linux/netfilter_ipv6.h>
13 #include <linux/sctp.h>
14 #include <linux/tcp.h>
15 #include <linux/udp.h>
16 #include <linux/in6.h>
17 #include <linux/if_arp.h>
18 #include <linux/if_vlan.h>
23 #include <net/ip6_fib.h>
24 #include <net/checksum.h>
25 #include <net/dsfield.h>
27 #include <net/sctp/checksum.h>
31 #include "conntrack.h"
33 #include "flow_netlink.h"
35 struct deferred_action {
37 const struct nlattr *actions;
40 /* Store pkt_key clone when creating deferred action. */
41 struct sw_flow_key pkt_key;
44 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
45 struct ovs_frag_data {
49 __be16 inner_protocol;
50 u16 network_offset; /* valid only for MPLS */
55 u8 l2_data[MAX_L2_LEN];
58 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
60 #define DEFERRED_ACTION_FIFO_SIZE 10
61 #define OVS_RECURSION_LIMIT 5
62 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
66 /* Deferred action fifo queue storage. */
67 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
70 struct action_flow_keys {
71 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
74 static struct action_fifo __percpu *action_fifos;
75 static struct action_flow_keys __percpu *flow_keys;
76 static DEFINE_PER_CPU(int, exec_actions_level);
78 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
79 * space. Return NULL if out of key spaces.
81 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
83 struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
84 int level = this_cpu_read(exec_actions_level);
85 struct sw_flow_key *key = NULL;
87 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
88 key = &keys->key[level - 1];
95 static void action_fifo_init(struct action_fifo *fifo)
101 static bool action_fifo_is_empty(const struct action_fifo *fifo)
103 return (fifo->head == fifo->tail);
106 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
108 if (action_fifo_is_empty(fifo))
111 return &fifo->fifo[fifo->tail++];
114 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
116 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
119 return &fifo->fifo[fifo->head++];
122 /* Return true if fifo is not full */
123 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
124 const struct sw_flow_key *key,
125 const struct nlattr *actions,
126 const int actions_len)
128 struct action_fifo *fifo;
129 struct deferred_action *da;
131 fifo = this_cpu_ptr(action_fifos);
132 da = action_fifo_put(fifo);
135 da->actions = actions;
136 da->actions_len = actions_len;
143 static void invalidate_flow_key(struct sw_flow_key *key)
145 key->mac_proto |= SW_FLOW_KEY_INVALID;
148 static bool is_flow_key_valid(const struct sw_flow_key *key)
150 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
153 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
154 struct sw_flow_key *key,
156 const struct nlattr *actions, int len,
157 bool last, bool clone_flow_key);
159 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
160 struct sw_flow_key *key,
161 const struct nlattr *attr, int len);
163 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
164 const struct ovs_action_push_mpls *mpls)
168 err = skb_mpls_push(skb, mpls->mpls_lse, mpls->mpls_ethertype,
170 ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
174 invalidate_flow_key(key);
178 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
179 const __be16 ethertype)
183 err = skb_mpls_pop(skb, ethertype, skb->mac_len,
184 ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
188 invalidate_flow_key(key);
192 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
193 const __be32 *mpls_lse, const __be32 *mask)
195 struct mpls_shim_hdr *stack;
199 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
202 stack = mpls_hdr(skb);
203 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
204 err = skb_mpls_update_lse(skb, lse);
208 flow_key->mpls.top_lse = lse;
212 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
216 err = skb_vlan_pop(skb);
217 if (skb_vlan_tag_present(skb)) {
218 invalidate_flow_key(key);
220 key->eth.vlan.tci = 0;
221 key->eth.vlan.tpid = 0;
226 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
227 const struct ovs_action_push_vlan *vlan)
229 if (skb_vlan_tag_present(skb)) {
230 invalidate_flow_key(key);
232 key->eth.vlan.tci = vlan->vlan_tci;
233 key->eth.vlan.tpid = vlan->vlan_tpid;
235 return skb_vlan_push(skb, vlan->vlan_tpid,
236 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
239 /* 'src' is already properly masked. */
240 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
242 u16 *dst = (u16 *)dst_;
243 const u16 *src = (const u16 *)src_;
244 const u16 *mask = (const u16 *)mask_;
246 OVS_SET_MASKED(dst[0], src[0], mask[0]);
247 OVS_SET_MASKED(dst[1], src[1], mask[1]);
248 OVS_SET_MASKED(dst[2], src[2], mask[2]);
251 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
252 const struct ovs_key_ethernet *key,
253 const struct ovs_key_ethernet *mask)
257 err = skb_ensure_writable(skb, ETH_HLEN);
261 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
263 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
265 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
268 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
270 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
271 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
275 /* pop_eth does not support VLAN packets as this action is never called
278 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
280 skb_pull_rcsum(skb, ETH_HLEN);
281 skb_reset_mac_header(skb);
282 skb_reset_mac_len(skb);
284 /* safe right before invalidate_flow_key */
285 key->mac_proto = MAC_PROTO_NONE;
286 invalidate_flow_key(key);
290 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
291 const struct ovs_action_push_eth *ethh)
295 /* Add the new Ethernet header */
296 if (skb_cow_head(skb, ETH_HLEN) < 0)
299 skb_push(skb, ETH_HLEN);
300 skb_reset_mac_header(skb);
301 skb_reset_mac_len(skb);
304 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
305 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
306 hdr->h_proto = skb->protocol;
308 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
310 /* safe right before invalidate_flow_key */
311 key->mac_proto = MAC_PROTO_ETHERNET;
312 invalidate_flow_key(key);
316 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
317 const struct nshhdr *nh)
321 err = nsh_push(skb, nh);
325 /* safe right before invalidate_flow_key */
326 key->mac_proto = MAC_PROTO_NONE;
327 invalidate_flow_key(key);
331 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
339 /* safe right before invalidate_flow_key */
340 if (skb->protocol == htons(ETH_P_TEB))
341 key->mac_proto = MAC_PROTO_ETHERNET;
343 key->mac_proto = MAC_PROTO_NONE;
344 invalidate_flow_key(key);
348 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
349 __be32 addr, __be32 new_addr)
351 int transport_len = skb->len - skb_transport_offset(skb);
353 if (nh->frag_off & htons(IP_OFFSET))
356 if (nh->protocol == IPPROTO_TCP) {
357 if (likely(transport_len >= sizeof(struct tcphdr)))
358 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
359 addr, new_addr, true);
360 } else if (nh->protocol == IPPROTO_UDP) {
361 if (likely(transport_len >= sizeof(struct udphdr))) {
362 struct udphdr *uh = udp_hdr(skb);
364 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
365 inet_proto_csum_replace4(&uh->check, skb,
366 addr, new_addr, true);
368 uh->check = CSUM_MANGLED_0;
374 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
375 __be32 *addr, __be32 new_addr)
377 update_ip_l4_checksum(skb, nh, *addr, new_addr);
378 csum_replace4(&nh->check, *addr, new_addr);
380 ovs_ct_clear(skb, NULL);
384 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
385 __be32 addr[4], const __be32 new_addr[4])
387 int transport_len = skb->len - skb_transport_offset(skb);
389 if (l4_proto == NEXTHDR_TCP) {
390 if (likely(transport_len >= sizeof(struct tcphdr)))
391 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
392 addr, new_addr, true);
393 } else if (l4_proto == NEXTHDR_UDP) {
394 if (likely(transport_len >= sizeof(struct udphdr))) {
395 struct udphdr *uh = udp_hdr(skb);
397 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
398 inet_proto_csum_replace16(&uh->check, skb,
399 addr, new_addr, true);
401 uh->check = CSUM_MANGLED_0;
404 } else if (l4_proto == NEXTHDR_ICMP) {
405 if (likely(transport_len >= sizeof(struct icmp6hdr)))
406 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
407 skb, addr, new_addr, true);
411 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
412 const __be32 mask[4], __be32 masked[4])
414 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
415 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
416 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
417 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
420 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
421 __be32 addr[4], const __be32 new_addr[4],
422 bool recalculate_csum)
424 if (recalculate_csum)
425 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
428 ovs_ct_clear(skb, NULL);
429 memcpy(addr, new_addr, sizeof(__be32[4]));
432 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
434 u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
436 ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
438 if (skb->ip_summed == CHECKSUM_COMPLETE)
439 csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
440 (__force __wsum)(ipv6_tclass << 12));
442 ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
445 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
449 ofl = nh->flow_lbl[0] << 16 | nh->flow_lbl[1] << 8 | nh->flow_lbl[2];
450 fl = OVS_MASKED(ofl, fl, mask);
452 /* Bits 21-24 are always unmasked, so this retains their values. */
453 nh->flow_lbl[0] = (u8)(fl >> 16);
454 nh->flow_lbl[1] = (u8)(fl >> 8);
455 nh->flow_lbl[2] = (u8)fl;
457 if (skb->ip_summed == CHECKSUM_COMPLETE)
458 csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
461 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
463 new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
465 if (skb->ip_summed == CHECKSUM_COMPLETE)
466 csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
467 (__force __wsum)(new_ttl << 8));
468 nh->hop_limit = new_ttl;
471 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
474 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
476 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
480 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
481 const struct ovs_key_ipv4 *key,
482 const struct ovs_key_ipv4 *mask)
488 err = skb_ensure_writable(skb, skb_network_offset(skb) +
489 sizeof(struct iphdr));
495 /* Setting an IP addresses is typically only a side effect of
496 * matching on them in the current userspace implementation, so it
497 * makes sense to check if the value actually changed.
499 if (mask->ipv4_src) {
500 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
502 if (unlikely(new_addr != nh->saddr)) {
503 set_ip_addr(skb, nh, &nh->saddr, new_addr);
504 flow_key->ipv4.addr.src = new_addr;
507 if (mask->ipv4_dst) {
508 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
510 if (unlikely(new_addr != nh->daddr)) {
511 set_ip_addr(skb, nh, &nh->daddr, new_addr);
512 flow_key->ipv4.addr.dst = new_addr;
515 if (mask->ipv4_tos) {
516 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
517 flow_key->ip.tos = nh->tos;
519 if (mask->ipv4_ttl) {
520 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
521 flow_key->ip.ttl = nh->ttl;
527 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
529 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
532 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
533 const struct ovs_key_ipv6 *key,
534 const struct ovs_key_ipv6 *mask)
539 err = skb_ensure_writable(skb, skb_network_offset(skb) +
540 sizeof(struct ipv6hdr));
546 /* Setting an IP addresses is typically only a side effect of
547 * matching on them in the current userspace implementation, so it
548 * makes sense to check if the value actually changed.
550 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
551 __be32 *saddr = (__be32 *)&nh->saddr;
554 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
556 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
557 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
559 memcpy(&flow_key->ipv6.addr.src, masked,
560 sizeof(flow_key->ipv6.addr.src));
563 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
564 unsigned int offset = 0;
565 int flags = IP6_FH_F_SKIP_RH;
566 bool recalc_csum = true;
567 __be32 *daddr = (__be32 *)&nh->daddr;
570 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
572 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
573 if (ipv6_ext_hdr(nh->nexthdr))
574 recalc_csum = (ipv6_find_hdr(skb, &offset,
579 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
581 memcpy(&flow_key->ipv6.addr.dst, masked,
582 sizeof(flow_key->ipv6.addr.dst));
585 if (mask->ipv6_tclass) {
586 set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
587 flow_key->ip.tos = ipv6_get_dsfield(nh);
589 if (mask->ipv6_label) {
590 set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
591 ntohl(mask->ipv6_label));
592 flow_key->ipv6.label =
593 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
595 if (mask->ipv6_hlimit) {
596 set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
597 flow_key->ip.ttl = nh->hop_limit;
602 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
603 const struct nlattr *a)
612 struct ovs_key_nsh key;
613 struct ovs_key_nsh mask;
615 err = nsh_key_from_nlattr(a, &key, &mask);
619 /* Make sure the NSH base header is there */
620 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
624 length = nsh_hdr_len(nh);
626 /* Make sure the whole NSH header is there */
627 err = skb_ensure_writable(skb, skb_network_offset(skb) +
633 skb_postpull_rcsum(skb, nh, length);
634 flags = nsh_get_flags(nh);
635 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
636 flow_key->nsh.base.flags = flags;
637 ttl = nsh_get_ttl(nh);
638 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
639 flow_key->nsh.base.ttl = ttl;
640 nsh_set_flags_and_ttl(nh, flags, ttl);
641 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
643 flow_key->nsh.base.path_hdr = nh->path_hdr;
644 switch (nh->mdtype) {
646 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
648 OVS_MASKED(nh->md1.context[i], key.context[i],
651 memcpy(flow_key->nsh.context, nh->md1.context,
652 sizeof(nh->md1.context));
655 memset(flow_key->nsh.context, 0,
656 sizeof(flow_key->nsh.context));
661 skb_postpush_rcsum(skb, nh, length);
665 /* Must follow skb_ensure_writable() since that can move the skb data. */
666 static void set_tp_port(struct sk_buff *skb, __be16 *port,
667 __be16 new_port, __sum16 *check)
669 ovs_ct_clear(skb, NULL);
670 inet_proto_csum_replace2(check, skb, *port, new_port, false);
674 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
675 const struct ovs_key_udp *key,
676 const struct ovs_key_udp *mask)
682 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
683 sizeof(struct udphdr));
688 /* Either of the masks is non-zero, so do not bother checking them. */
689 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
690 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
692 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
693 if (likely(src != uh->source)) {
694 set_tp_port(skb, &uh->source, src, &uh->check);
695 flow_key->tp.src = src;
697 if (likely(dst != uh->dest)) {
698 set_tp_port(skb, &uh->dest, dst, &uh->check);
699 flow_key->tp.dst = dst;
702 if (unlikely(!uh->check))
703 uh->check = CSUM_MANGLED_0;
707 flow_key->tp.src = src;
708 flow_key->tp.dst = dst;
709 ovs_ct_clear(skb, NULL);
717 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
718 const struct ovs_key_tcp *key,
719 const struct ovs_key_tcp *mask)
725 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
726 sizeof(struct tcphdr));
731 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
732 if (likely(src != th->source)) {
733 set_tp_port(skb, &th->source, src, &th->check);
734 flow_key->tp.src = src;
736 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
737 if (likely(dst != th->dest)) {
738 set_tp_port(skb, &th->dest, dst, &th->check);
739 flow_key->tp.dst = dst;
746 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
747 const struct ovs_key_sctp *key,
748 const struct ovs_key_sctp *mask)
750 unsigned int sctphoff = skb_transport_offset(skb);
752 __le32 old_correct_csum, new_csum, old_csum;
755 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
760 old_csum = sh->checksum;
761 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
763 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
764 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
766 new_csum = sctp_compute_cksum(skb, sctphoff);
768 /* Carry any checksum errors through. */
769 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
772 ovs_ct_clear(skb, NULL);
774 flow_key->tp.src = sh->source;
775 flow_key->tp.dst = sh->dest;
780 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
782 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
783 struct vport *vport = data->vport;
785 if (skb_cow_head(skb, data->l2_len) < 0) {
790 __skb_dst_copy(skb, data->dst);
791 *OVS_CB(skb) = data->cb;
792 skb->inner_protocol = data->inner_protocol;
793 if (data->vlan_tci & VLAN_CFI_MASK)
794 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
796 __vlan_hwaccel_clear_tag(skb);
798 /* Reconstruct the MAC header. */
799 skb_push(skb, data->l2_len);
800 memcpy(skb->data, &data->l2_data, data->l2_len);
801 skb_postpush_rcsum(skb, skb->data, data->l2_len);
802 skb_reset_mac_header(skb);
804 if (eth_p_mpls(skb->protocol)) {
805 skb->inner_network_header = skb->network_header;
806 skb_set_network_header(skb, data->network_offset);
807 skb_reset_mac_len(skb);
810 ovs_vport_send(vport, skb, data->mac_proto);
815 ovs_dst_get_mtu(const struct dst_entry *dst)
817 return dst->dev->mtu;
820 static struct dst_ops ovs_dst_ops = {
822 .mtu = ovs_dst_get_mtu,
825 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
826 * ovs_vport_output(), which is called once per fragmented packet.
828 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
829 u16 orig_network_offset, u8 mac_proto)
831 unsigned int hlen = skb_network_offset(skb);
832 struct ovs_frag_data *data;
834 data = this_cpu_ptr(&ovs_frag_data_storage);
835 data->dst = skb->_skb_refdst;
837 data->cb = *OVS_CB(skb);
838 data->inner_protocol = skb->inner_protocol;
839 data->network_offset = orig_network_offset;
840 if (skb_vlan_tag_present(skb))
841 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
844 data->vlan_proto = skb->vlan_proto;
845 data->mac_proto = mac_proto;
847 memcpy(&data->l2_data, skb->data, hlen);
849 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
853 static void ovs_fragment(struct net *net, struct vport *vport,
854 struct sk_buff *skb, u16 mru,
855 struct sw_flow_key *key)
857 u16 orig_network_offset = 0;
859 if (eth_p_mpls(skb->protocol)) {
860 orig_network_offset = skb_network_offset(skb);
861 skb->network_header = skb->inner_network_header;
864 if (skb_network_offset(skb) > MAX_L2_LEN) {
865 OVS_NLERR(1, "L2 header too long to fragment");
869 if (key->eth.type == htons(ETH_P_IP)) {
870 struct rtable ovs_rt = { 0 };
871 unsigned long orig_dst;
873 prepare_frag(vport, skb, orig_network_offset,
874 ovs_key_mac_proto(key));
875 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
876 DST_OBSOLETE_NONE, DST_NOCOUNT);
877 ovs_rt.dst.dev = vport->dev;
879 orig_dst = skb->_skb_refdst;
880 skb_dst_set_noref(skb, &ovs_rt.dst);
881 IPCB(skb)->frag_max_size = mru;
883 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
884 refdst_drop(orig_dst);
885 } else if (key->eth.type == htons(ETH_P_IPV6)) {
886 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
887 unsigned long orig_dst;
888 struct rt6_info ovs_rt;
893 prepare_frag(vport, skb, orig_network_offset,
894 ovs_key_mac_proto(key));
895 memset(&ovs_rt, 0, sizeof(ovs_rt));
896 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
897 DST_OBSOLETE_NONE, DST_NOCOUNT);
898 ovs_rt.dst.dev = vport->dev;
900 orig_dst = skb->_skb_refdst;
901 skb_dst_set_noref(skb, &ovs_rt.dst);
902 IP6CB(skb)->frag_max_size = mru;
904 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
905 refdst_drop(orig_dst);
907 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
908 ovs_vport_name(vport), ntohs(key->eth.type), mru,
918 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
919 struct sw_flow_key *key)
921 struct vport *vport = ovs_vport_rcu(dp, out_port);
924 u16 mru = OVS_CB(skb)->mru;
925 u32 cutlen = OVS_CB(skb)->cutlen;
927 if (unlikely(cutlen > 0)) {
928 if (skb->len - cutlen > ovs_mac_header_len(key))
929 pskb_trim(skb, skb->len - cutlen);
931 pskb_trim(skb, ovs_mac_header_len(key));
935 (skb->len <= mru + vport->dev->hard_header_len))) {
936 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
937 } else if (mru <= vport->dev->mtu) {
938 struct net *net = read_pnet(&dp->net);
940 ovs_fragment(net, vport, skb, mru, key);
949 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
950 struct sw_flow_key *key, const struct nlattr *attr,
951 const struct nlattr *actions, int actions_len,
954 struct dp_upcall_info upcall;
955 const struct nlattr *a;
958 memset(&upcall, 0, sizeof(upcall));
959 upcall.cmd = OVS_PACKET_CMD_ACTION;
960 upcall.mru = OVS_CB(skb)->mru;
962 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
963 a = nla_next(a, &rem)) {
964 switch (nla_type(a)) {
965 case OVS_USERSPACE_ATTR_USERDATA:
969 case OVS_USERSPACE_ATTR_PID:
970 upcall.portid = nla_get_u32(a);
973 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
974 /* Get out tunnel info. */
977 vport = ovs_vport_rcu(dp, nla_get_u32(a));
981 err = dev_fill_metadata_dst(vport->dev, skb);
983 upcall.egress_tun_info = skb_tunnel_info(skb);
989 case OVS_USERSPACE_ATTR_ACTIONS: {
990 /* Include actions. */
991 upcall.actions = actions;
992 upcall.actions_len = actions_len;
996 } /* End of switch. */
999 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1002 /* When 'last' is true, sample() should always consume the 'skb'.
1003 * Otherwise, sample() should keep 'skb' intact regardless what
1004 * actions are executed within sample().
1006 static int sample(struct datapath *dp, struct sk_buff *skb,
1007 struct sw_flow_key *key, const struct nlattr *attr,
1010 struct nlattr *actions;
1011 struct nlattr *sample_arg;
1012 int rem = nla_len(attr);
1013 const struct sample_arg *arg;
1014 bool clone_flow_key;
1016 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1017 sample_arg = nla_data(attr);
1018 arg = nla_data(sample_arg);
1019 actions = nla_next(sample_arg, &rem);
1021 if ((arg->probability != U32_MAX) &&
1022 (!arg->probability || prandom_u32() > arg->probability)) {
1028 clone_flow_key = !arg->exec;
1029 return clone_execute(dp, skb, key, 0, actions, rem, last,
1033 /* When 'last' is true, clone() should always consume the 'skb'.
1034 * Otherwise, clone() should keep 'skb' intact regardless what
1035 * actions are executed within clone().
1037 static int clone(struct datapath *dp, struct sk_buff *skb,
1038 struct sw_flow_key *key, const struct nlattr *attr,
1041 struct nlattr *actions;
1042 struct nlattr *clone_arg;
1043 int rem = nla_len(attr);
1044 bool dont_clone_flow_key;
1046 /* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1047 clone_arg = nla_data(attr);
1048 dont_clone_flow_key = nla_get_u32(clone_arg);
1049 actions = nla_next(clone_arg, &rem);
1051 return clone_execute(dp, skb, key, 0, actions, rem, last,
1052 !dont_clone_flow_key);
1055 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1056 const struct nlattr *attr)
1058 struct ovs_action_hash *hash_act = nla_data(attr);
1061 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1062 hash = skb_get_hash(skb);
1063 hash = jhash_1word(hash, hash_act->hash_basis);
1067 key->ovs_flow_hash = hash;
1070 static int execute_set_action(struct sk_buff *skb,
1071 struct sw_flow_key *flow_key,
1072 const struct nlattr *a)
1074 /* Only tunnel set execution is supported without a mask. */
1075 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1076 struct ovs_tunnel_info *tun = nla_data(a);
1079 dst_hold((struct dst_entry *)tun->tun_dst);
1080 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1087 /* Mask is at the midpoint of the data. */
1088 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1090 static int execute_masked_set_action(struct sk_buff *skb,
1091 struct sw_flow_key *flow_key,
1092 const struct nlattr *a)
1096 switch (nla_type(a)) {
1097 case OVS_KEY_ATTR_PRIORITY:
1098 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1099 *get_mask(a, u32 *));
1100 flow_key->phy.priority = skb->priority;
1103 case OVS_KEY_ATTR_SKB_MARK:
1104 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1105 flow_key->phy.skb_mark = skb->mark;
1108 case OVS_KEY_ATTR_TUNNEL_INFO:
1109 /* Masked data not supported for tunnel. */
1113 case OVS_KEY_ATTR_ETHERNET:
1114 err = set_eth_addr(skb, flow_key, nla_data(a),
1115 get_mask(a, struct ovs_key_ethernet *));
1118 case OVS_KEY_ATTR_NSH:
1119 err = set_nsh(skb, flow_key, a);
1122 case OVS_KEY_ATTR_IPV4:
1123 err = set_ipv4(skb, flow_key, nla_data(a),
1124 get_mask(a, struct ovs_key_ipv4 *));
1127 case OVS_KEY_ATTR_IPV6:
1128 err = set_ipv6(skb, flow_key, nla_data(a),
1129 get_mask(a, struct ovs_key_ipv6 *));
1132 case OVS_KEY_ATTR_TCP:
1133 err = set_tcp(skb, flow_key, nla_data(a),
1134 get_mask(a, struct ovs_key_tcp *));
1137 case OVS_KEY_ATTR_UDP:
1138 err = set_udp(skb, flow_key, nla_data(a),
1139 get_mask(a, struct ovs_key_udp *));
1142 case OVS_KEY_ATTR_SCTP:
1143 err = set_sctp(skb, flow_key, nla_data(a),
1144 get_mask(a, struct ovs_key_sctp *));
1147 case OVS_KEY_ATTR_MPLS:
1148 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1152 case OVS_KEY_ATTR_CT_STATE:
1153 case OVS_KEY_ATTR_CT_ZONE:
1154 case OVS_KEY_ATTR_CT_MARK:
1155 case OVS_KEY_ATTR_CT_LABELS:
1156 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1157 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1165 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1166 struct sw_flow_key *key,
1167 const struct nlattr *a, bool last)
1171 if (!is_flow_key_valid(key)) {
1174 err = ovs_flow_key_update(skb, key);
1178 BUG_ON(!is_flow_key_valid(key));
1180 recirc_id = nla_get_u32(a);
1181 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1184 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1185 struct sw_flow_key *key,
1186 const struct nlattr *attr, bool last)
1188 struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1189 const struct nlattr *actions, *cpl_arg;
1190 int len, max_len, rem = nla_len(attr);
1191 const struct check_pkt_len_arg *arg;
1192 bool clone_flow_key;
1194 /* The first netlink attribute in 'attr' is always
1195 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1197 cpl_arg = nla_data(attr);
1198 arg = nla_data(cpl_arg);
1200 len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1201 max_len = arg->pkt_len;
1203 if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1205 /* Second netlink attribute in 'attr' is always
1206 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1208 actions = nla_next(cpl_arg, &rem);
1209 clone_flow_key = !arg->exec_for_lesser_equal;
1211 /* Third netlink attribute in 'attr' is always
1212 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1214 actions = nla_next(cpl_arg, &rem);
1215 actions = nla_next(actions, &rem);
1216 clone_flow_key = !arg->exec_for_greater;
1219 return clone_execute(dp, skb, key, 0, nla_data(actions),
1220 nla_len(actions), last, clone_flow_key);
1223 /* Execute a list of actions against 'skb'. */
1224 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1225 struct sw_flow_key *key,
1226 const struct nlattr *attr, int len)
1228 const struct nlattr *a;
1231 for (a = attr, rem = len; rem > 0;
1232 a = nla_next(a, &rem)) {
1235 switch (nla_type(a)) {
1236 case OVS_ACTION_ATTR_OUTPUT: {
1237 int port = nla_get_u32(a);
1238 struct sk_buff *clone;
1240 /* Every output action needs a separate clone
1241 * of 'skb', In case the output action is the
1242 * last action, cloning can be avoided.
1244 if (nla_is_last(a, rem)) {
1245 do_output(dp, skb, port, key);
1246 /* 'skb' has been used for output.
1251 clone = skb_clone(skb, GFP_ATOMIC);
1253 do_output(dp, clone, port, key);
1254 OVS_CB(skb)->cutlen = 0;
1258 case OVS_ACTION_ATTR_TRUNC: {
1259 struct ovs_action_trunc *trunc = nla_data(a);
1261 if (skb->len > trunc->max_len)
1262 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1266 case OVS_ACTION_ATTR_USERSPACE:
1267 output_userspace(dp, skb, key, a, attr,
1268 len, OVS_CB(skb)->cutlen);
1269 OVS_CB(skb)->cutlen = 0;
1272 case OVS_ACTION_ATTR_HASH:
1273 execute_hash(skb, key, a);
1276 case OVS_ACTION_ATTR_PUSH_MPLS:
1277 err = push_mpls(skb, key, nla_data(a));
1280 case OVS_ACTION_ATTR_POP_MPLS:
1281 err = pop_mpls(skb, key, nla_get_be16(a));
1284 case OVS_ACTION_ATTR_PUSH_VLAN:
1285 err = push_vlan(skb, key, nla_data(a));
1288 case OVS_ACTION_ATTR_POP_VLAN:
1289 err = pop_vlan(skb, key);
1292 case OVS_ACTION_ATTR_RECIRC: {
1293 bool last = nla_is_last(a, rem);
1295 err = execute_recirc(dp, skb, key, a, last);
1297 /* If this is the last action, the skb has
1298 * been consumed or freed.
1299 * Return immediately.
1306 case OVS_ACTION_ATTR_SET:
1307 err = execute_set_action(skb, key, nla_data(a));
1310 case OVS_ACTION_ATTR_SET_MASKED:
1311 case OVS_ACTION_ATTR_SET_TO_MASKED:
1312 err = execute_masked_set_action(skb, key, nla_data(a));
1315 case OVS_ACTION_ATTR_SAMPLE: {
1316 bool last = nla_is_last(a, rem);
1318 err = sample(dp, skb, key, a, last);
1325 case OVS_ACTION_ATTR_CT:
1326 if (!is_flow_key_valid(key)) {
1327 err = ovs_flow_key_update(skb, key);
1332 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1335 /* Hide stolen IP fragments from user space. */
1337 return err == -EINPROGRESS ? 0 : err;
1340 case OVS_ACTION_ATTR_CT_CLEAR:
1341 err = ovs_ct_clear(skb, key);
1344 case OVS_ACTION_ATTR_PUSH_ETH:
1345 err = push_eth(skb, key, nla_data(a));
1348 case OVS_ACTION_ATTR_POP_ETH:
1349 err = pop_eth(skb, key);
1352 case OVS_ACTION_ATTR_PUSH_NSH: {
1353 u8 buffer[NSH_HDR_MAX_LEN];
1354 struct nshhdr *nh = (struct nshhdr *)buffer;
1356 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1360 err = push_nsh(skb, key, nh);
1364 case OVS_ACTION_ATTR_POP_NSH:
1365 err = pop_nsh(skb, key);
1368 case OVS_ACTION_ATTR_METER:
1369 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1375 case OVS_ACTION_ATTR_CLONE: {
1376 bool last = nla_is_last(a, rem);
1378 err = clone(dp, skb, key, a, last);
1385 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1386 bool last = nla_is_last(a, rem);
1388 err = execute_check_pkt_len(dp, skb, key, a, last);
1396 if (unlikely(err)) {
1406 /* Execute the actions on the clone of the packet. The effect of the
1407 * execution does not affect the original 'skb' nor the original 'key'.
1409 * The execution may be deferred in case the actions can not be executed
1412 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1413 struct sw_flow_key *key, u32 recirc_id,
1414 const struct nlattr *actions, int len,
1415 bool last, bool clone_flow_key)
1417 struct deferred_action *da;
1418 struct sw_flow_key *clone;
1420 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1422 /* Out of memory, skip this action.
1427 /* When clone_flow_key is false, the 'key' will not be change
1428 * by the actions, then the 'key' can be used directly.
1429 * Otherwise, try to clone key from the next recursion level of
1430 * 'flow_keys'. If clone is successful, execute the actions
1431 * without deferring.
1433 clone = clone_flow_key ? clone_key(key) : key;
1437 if (actions) { /* Sample action */
1439 __this_cpu_inc(exec_actions_level);
1441 err = do_execute_actions(dp, skb, clone,
1445 __this_cpu_dec(exec_actions_level);
1446 } else { /* Recirc action */
1447 clone->recirc_id = recirc_id;
1448 ovs_dp_process_packet(skb, clone);
1453 /* Out of 'flow_keys' space. Defer actions */
1454 da = add_deferred_actions(skb, key, actions, len);
1456 if (!actions) { /* Recirc action */
1458 key->recirc_id = recirc_id;
1461 /* Out of per CPU action FIFO space. Drop the 'skb' and
1466 if (net_ratelimit()) {
1467 if (actions) { /* Sample action */
1468 pr_warn("%s: deferred action limit reached, drop sample action\n",
1470 } else { /* Recirc action */
1471 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1479 static void process_deferred_actions(struct datapath *dp)
1481 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1483 /* Do not touch the FIFO in case there is no deferred actions. */
1484 if (action_fifo_is_empty(fifo))
1487 /* Finishing executing all deferred actions. */
1489 struct deferred_action *da = action_fifo_get(fifo);
1490 struct sk_buff *skb = da->skb;
1491 struct sw_flow_key *key = &da->pkt_key;
1492 const struct nlattr *actions = da->actions;
1493 int actions_len = da->actions_len;
1496 do_execute_actions(dp, skb, key, actions, actions_len);
1498 ovs_dp_process_packet(skb, key);
1499 } while (!action_fifo_is_empty(fifo));
1501 /* Reset FIFO for the next packet. */
1502 action_fifo_init(fifo);
1505 /* Execute a list of actions against 'skb'. */
1506 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1507 const struct sw_flow_actions *acts,
1508 struct sw_flow_key *key)
1512 level = __this_cpu_inc_return(exec_actions_level);
1513 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1514 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1521 OVS_CB(skb)->acts_origlen = acts->orig_len;
1522 err = do_execute_actions(dp, skb, key,
1523 acts->actions, acts->actions_len);
1526 process_deferred_actions(dp);
1529 __this_cpu_dec(exec_actions_level);
1533 int action_fifos_init(void)
1535 action_fifos = alloc_percpu(struct action_fifo);
1539 flow_keys = alloc_percpu(struct action_flow_keys);
1541 free_percpu(action_fifos);
1548 void action_fifos_exit(void)
1550 free_percpu(action_fifos);
1551 free_percpu(flow_keys);