2 * Copyright (c) 2007-2017 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/netfilter_ipv6.h>
26 #include <linux/sctp.h>
27 #include <linux/tcp.h>
28 #include <linux/udp.h>
29 #include <linux/in6.h>
30 #include <linux/if_arp.h>
31 #include <linux/if_vlan.h>
36 #include <net/ip6_fib.h>
37 #include <net/checksum.h>
38 #include <net/dsfield.h>
40 #include <net/sctp/checksum.h>
44 #include "conntrack.h"
46 #include "flow_netlink.h"
48 struct deferred_action {
50 const struct nlattr *actions;
53 /* Store pkt_key clone when creating deferred action. */
54 struct sw_flow_key pkt_key;
57 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
58 struct ovs_frag_data {
62 __be16 inner_protocol;
63 u16 network_offset; /* valid only for MPLS */
68 u8 l2_data[MAX_L2_LEN];
71 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
73 #define DEFERRED_ACTION_FIFO_SIZE 10
74 #define OVS_RECURSION_LIMIT 5
75 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
79 /* Deferred action fifo queue storage. */
80 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
83 struct action_flow_keys {
84 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
87 static struct action_fifo __percpu *action_fifos;
88 static struct action_flow_keys __percpu *flow_keys;
89 static DEFINE_PER_CPU(int, exec_actions_level);
91 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
92 * space. Return NULL if out of key spaces.
94 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
96 struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
97 int level = this_cpu_read(exec_actions_level);
98 struct sw_flow_key *key = NULL;
100 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
101 key = &keys->key[level - 1];
108 static void action_fifo_init(struct action_fifo *fifo)
114 static bool action_fifo_is_empty(const struct action_fifo *fifo)
116 return (fifo->head == fifo->tail);
119 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
121 if (action_fifo_is_empty(fifo))
124 return &fifo->fifo[fifo->tail++];
127 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
129 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
132 return &fifo->fifo[fifo->head++];
135 /* Return true if fifo is not full */
136 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
137 const struct sw_flow_key *key,
138 const struct nlattr *actions,
139 const int actions_len)
141 struct action_fifo *fifo;
142 struct deferred_action *da;
144 fifo = this_cpu_ptr(action_fifos);
145 da = action_fifo_put(fifo);
148 da->actions = actions;
149 da->actions_len = actions_len;
156 static void invalidate_flow_key(struct sw_flow_key *key)
158 key->mac_proto |= SW_FLOW_KEY_INVALID;
161 static bool is_flow_key_valid(const struct sw_flow_key *key)
163 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
166 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
167 struct sw_flow_key *key,
169 const struct nlattr *actions, int len,
170 bool last, bool clone_flow_key);
172 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
173 struct sw_flow_key *key,
174 const struct nlattr *attr, int len);
176 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
179 if (skb->ip_summed == CHECKSUM_COMPLETE) {
180 __be16 diff[] = { ~(hdr->h_proto), ethertype };
182 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
186 hdr->h_proto = ethertype;
189 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
190 const struct ovs_action_push_mpls *mpls)
192 struct mpls_shim_hdr *new_mpls_lse;
194 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
195 if (skb->encapsulation)
198 if (skb_cow_head(skb, MPLS_HLEN) < 0)
201 if (!skb->inner_protocol) {
202 skb_set_inner_network_header(skb, skb->mac_len);
203 skb_set_inner_protocol(skb, skb->protocol);
206 skb_push(skb, MPLS_HLEN);
207 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
209 skb_reset_mac_header(skb);
210 skb_set_network_header(skb, skb->mac_len);
212 new_mpls_lse = mpls_hdr(skb);
213 new_mpls_lse->label_stack_entry = mpls->mpls_lse;
215 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
217 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
218 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
219 skb->protocol = mpls->mpls_ethertype;
221 invalidate_flow_key(key);
225 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
226 const __be16 ethertype)
230 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
234 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
236 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
239 __skb_pull(skb, MPLS_HLEN);
240 skb_reset_mac_header(skb);
241 skb_set_network_header(skb, skb->mac_len);
243 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
246 /* mpls_hdr() is used to locate the ethertype field correctly in the
247 * presence of VLAN tags.
249 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
250 update_ethertype(skb, hdr, ethertype);
252 if (eth_p_mpls(skb->protocol))
253 skb->protocol = ethertype;
255 invalidate_flow_key(key);
259 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
260 const __be32 *mpls_lse, const __be32 *mask)
262 struct mpls_shim_hdr *stack;
266 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
270 stack = mpls_hdr(skb);
271 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
272 if (skb->ip_summed == CHECKSUM_COMPLETE) {
273 __be32 diff[] = { ~(stack->label_stack_entry), lse };
275 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
279 stack->label_stack_entry = lse;
280 flow_key->mpls.top_lse = lse;
284 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
288 err = skb_vlan_pop(skb);
289 if (skb_vlan_tag_present(skb)) {
290 invalidate_flow_key(key);
292 key->eth.vlan.tci = 0;
293 key->eth.vlan.tpid = 0;
298 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
299 const struct ovs_action_push_vlan *vlan)
301 if (skb_vlan_tag_present(skb)) {
302 invalidate_flow_key(key);
304 key->eth.vlan.tci = vlan->vlan_tci;
305 key->eth.vlan.tpid = vlan->vlan_tpid;
307 return skb_vlan_push(skb, vlan->vlan_tpid,
308 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
311 /* 'src' is already properly masked. */
312 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
314 u16 *dst = (u16 *)dst_;
315 const u16 *src = (const u16 *)src_;
316 const u16 *mask = (const u16 *)mask_;
318 OVS_SET_MASKED(dst[0], src[0], mask[0]);
319 OVS_SET_MASKED(dst[1], src[1], mask[1]);
320 OVS_SET_MASKED(dst[2], src[2], mask[2]);
323 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
324 const struct ovs_key_ethernet *key,
325 const struct ovs_key_ethernet *mask)
329 err = skb_ensure_writable(skb, ETH_HLEN);
333 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
335 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
337 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
340 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
342 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
343 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
347 /* pop_eth does not support VLAN packets as this action is never called
350 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
352 skb_pull_rcsum(skb, ETH_HLEN);
353 skb_reset_mac_header(skb);
354 skb_reset_mac_len(skb);
356 /* safe right before invalidate_flow_key */
357 key->mac_proto = MAC_PROTO_NONE;
358 invalidate_flow_key(key);
362 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
363 const struct ovs_action_push_eth *ethh)
367 /* Add the new Ethernet header */
368 if (skb_cow_head(skb, ETH_HLEN) < 0)
371 skb_push(skb, ETH_HLEN);
372 skb_reset_mac_header(skb);
373 skb_reset_mac_len(skb);
376 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
377 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
378 hdr->h_proto = skb->protocol;
380 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
382 /* safe right before invalidate_flow_key */
383 key->mac_proto = MAC_PROTO_ETHERNET;
384 invalidate_flow_key(key);
388 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
389 const struct nshhdr *nh)
393 err = nsh_push(skb, nh);
397 /* safe right before invalidate_flow_key */
398 key->mac_proto = MAC_PROTO_NONE;
399 invalidate_flow_key(key);
403 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
411 /* safe right before invalidate_flow_key */
412 if (skb->protocol == htons(ETH_P_TEB))
413 key->mac_proto = MAC_PROTO_ETHERNET;
415 key->mac_proto = MAC_PROTO_NONE;
416 invalidate_flow_key(key);
420 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
421 __be32 addr, __be32 new_addr)
423 int transport_len = skb->len - skb_transport_offset(skb);
425 if (nh->frag_off & htons(IP_OFFSET))
428 if (nh->protocol == IPPROTO_TCP) {
429 if (likely(transport_len >= sizeof(struct tcphdr)))
430 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
431 addr, new_addr, true);
432 } else if (nh->protocol == IPPROTO_UDP) {
433 if (likely(transport_len >= sizeof(struct udphdr))) {
434 struct udphdr *uh = udp_hdr(skb);
436 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
437 inet_proto_csum_replace4(&uh->check, skb,
438 addr, new_addr, true);
440 uh->check = CSUM_MANGLED_0;
446 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
447 __be32 *addr, __be32 new_addr)
449 update_ip_l4_checksum(skb, nh, *addr, new_addr);
450 csum_replace4(&nh->check, *addr, new_addr);
455 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
456 __be32 addr[4], const __be32 new_addr[4])
458 int transport_len = skb->len - skb_transport_offset(skb);
460 if (l4_proto == NEXTHDR_TCP) {
461 if (likely(transport_len >= sizeof(struct tcphdr)))
462 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
463 addr, new_addr, true);
464 } else if (l4_proto == NEXTHDR_UDP) {
465 if (likely(transport_len >= sizeof(struct udphdr))) {
466 struct udphdr *uh = udp_hdr(skb);
468 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
469 inet_proto_csum_replace16(&uh->check, skb,
470 addr, new_addr, true);
472 uh->check = CSUM_MANGLED_0;
475 } else if (l4_proto == NEXTHDR_ICMP) {
476 if (likely(transport_len >= sizeof(struct icmp6hdr)))
477 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
478 skb, addr, new_addr, true);
482 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
483 const __be32 mask[4], __be32 masked[4])
485 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
486 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
487 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
488 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
491 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
492 __be32 addr[4], const __be32 new_addr[4],
493 bool recalculate_csum)
495 if (recalculate_csum)
496 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
499 memcpy(addr, new_addr, sizeof(__be32[4]));
502 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
504 /* Bits 21-24 are always unmasked, so this retains their values. */
505 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
506 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
507 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
510 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
513 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
515 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
519 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
520 const struct ovs_key_ipv4 *key,
521 const struct ovs_key_ipv4 *mask)
527 err = skb_ensure_writable(skb, skb_network_offset(skb) +
528 sizeof(struct iphdr));
534 /* Setting an IP addresses is typically only a side effect of
535 * matching on them in the current userspace implementation, so it
536 * makes sense to check if the value actually changed.
538 if (mask->ipv4_src) {
539 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
541 if (unlikely(new_addr != nh->saddr)) {
542 set_ip_addr(skb, nh, &nh->saddr, new_addr);
543 flow_key->ipv4.addr.src = new_addr;
546 if (mask->ipv4_dst) {
547 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
549 if (unlikely(new_addr != nh->daddr)) {
550 set_ip_addr(skb, nh, &nh->daddr, new_addr);
551 flow_key->ipv4.addr.dst = new_addr;
554 if (mask->ipv4_tos) {
555 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
556 flow_key->ip.tos = nh->tos;
558 if (mask->ipv4_ttl) {
559 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
560 flow_key->ip.ttl = nh->ttl;
566 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
568 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
571 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
572 const struct ovs_key_ipv6 *key,
573 const struct ovs_key_ipv6 *mask)
578 err = skb_ensure_writable(skb, skb_network_offset(skb) +
579 sizeof(struct ipv6hdr));
585 /* Setting an IP addresses is typically only a side effect of
586 * matching on them in the current userspace implementation, so it
587 * makes sense to check if the value actually changed.
589 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
590 __be32 *saddr = (__be32 *)&nh->saddr;
593 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
595 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
596 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
598 memcpy(&flow_key->ipv6.addr.src, masked,
599 sizeof(flow_key->ipv6.addr.src));
602 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
603 unsigned int offset = 0;
604 int flags = IP6_FH_F_SKIP_RH;
605 bool recalc_csum = true;
606 __be32 *daddr = (__be32 *)&nh->daddr;
609 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
611 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
612 if (ipv6_ext_hdr(nh->nexthdr))
613 recalc_csum = (ipv6_find_hdr(skb, &offset,
618 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
620 memcpy(&flow_key->ipv6.addr.dst, masked,
621 sizeof(flow_key->ipv6.addr.dst));
624 if (mask->ipv6_tclass) {
625 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
626 flow_key->ip.tos = ipv6_get_dsfield(nh);
628 if (mask->ipv6_label) {
629 set_ipv6_fl(nh, ntohl(key->ipv6_label),
630 ntohl(mask->ipv6_label));
631 flow_key->ipv6.label =
632 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
634 if (mask->ipv6_hlimit) {
635 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
637 flow_key->ip.ttl = nh->hop_limit;
642 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
643 const struct nlattr *a)
652 struct ovs_key_nsh key;
653 struct ovs_key_nsh mask;
655 err = nsh_key_from_nlattr(a, &key, &mask);
659 /* Make sure the NSH base header is there */
660 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
664 length = nsh_hdr_len(nh);
666 /* Make sure the whole NSH header is there */
667 err = skb_ensure_writable(skb, skb_network_offset(skb) +
673 skb_postpull_rcsum(skb, nh, length);
674 flags = nsh_get_flags(nh);
675 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
676 flow_key->nsh.base.flags = flags;
677 ttl = nsh_get_ttl(nh);
678 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
679 flow_key->nsh.base.ttl = ttl;
680 nsh_set_flags_and_ttl(nh, flags, ttl);
681 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
683 flow_key->nsh.base.path_hdr = nh->path_hdr;
684 switch (nh->mdtype) {
686 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
688 OVS_MASKED(nh->md1.context[i], key.context[i],
691 memcpy(flow_key->nsh.context, nh->md1.context,
692 sizeof(nh->md1.context));
695 memset(flow_key->nsh.context, 0,
696 sizeof(flow_key->nsh.context));
701 skb_postpush_rcsum(skb, nh, length);
705 /* Must follow skb_ensure_writable() since that can move the skb data. */
706 static void set_tp_port(struct sk_buff *skb, __be16 *port,
707 __be16 new_port, __sum16 *check)
709 inet_proto_csum_replace2(check, skb, *port, new_port, false);
713 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
714 const struct ovs_key_udp *key,
715 const struct ovs_key_udp *mask)
721 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
722 sizeof(struct udphdr));
727 /* Either of the masks is non-zero, so do not bother checking them. */
728 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
729 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
731 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
732 if (likely(src != uh->source)) {
733 set_tp_port(skb, &uh->source, src, &uh->check);
734 flow_key->tp.src = src;
736 if (likely(dst != uh->dest)) {
737 set_tp_port(skb, &uh->dest, dst, &uh->check);
738 flow_key->tp.dst = dst;
741 if (unlikely(!uh->check))
742 uh->check = CSUM_MANGLED_0;
746 flow_key->tp.src = src;
747 flow_key->tp.dst = dst;
755 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
756 const struct ovs_key_tcp *key,
757 const struct ovs_key_tcp *mask)
763 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
764 sizeof(struct tcphdr));
769 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
770 if (likely(src != th->source)) {
771 set_tp_port(skb, &th->source, src, &th->check);
772 flow_key->tp.src = src;
774 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
775 if (likely(dst != th->dest)) {
776 set_tp_port(skb, &th->dest, dst, &th->check);
777 flow_key->tp.dst = dst;
784 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
785 const struct ovs_key_sctp *key,
786 const struct ovs_key_sctp *mask)
788 unsigned int sctphoff = skb_transport_offset(skb);
790 __le32 old_correct_csum, new_csum, old_csum;
793 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
798 old_csum = sh->checksum;
799 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
801 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
802 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
804 new_csum = sctp_compute_cksum(skb, sctphoff);
806 /* Carry any checksum errors through. */
807 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
810 flow_key->tp.src = sh->source;
811 flow_key->tp.dst = sh->dest;
816 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
818 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
819 struct vport *vport = data->vport;
821 if (skb_cow_head(skb, data->l2_len) < 0) {
826 __skb_dst_copy(skb, data->dst);
827 *OVS_CB(skb) = data->cb;
828 skb->inner_protocol = data->inner_protocol;
829 if (data->vlan_tci & VLAN_CFI_MASK)
830 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
832 __vlan_hwaccel_clear_tag(skb);
834 /* Reconstruct the MAC header. */
835 skb_push(skb, data->l2_len);
836 memcpy(skb->data, &data->l2_data, data->l2_len);
837 skb_postpush_rcsum(skb, skb->data, data->l2_len);
838 skb_reset_mac_header(skb);
840 if (eth_p_mpls(skb->protocol)) {
841 skb->inner_network_header = skb->network_header;
842 skb_set_network_header(skb, data->network_offset);
843 skb_reset_mac_len(skb);
846 ovs_vport_send(vport, skb, data->mac_proto);
851 ovs_dst_get_mtu(const struct dst_entry *dst)
853 return dst->dev->mtu;
856 static struct dst_ops ovs_dst_ops = {
858 .mtu = ovs_dst_get_mtu,
861 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
862 * ovs_vport_output(), which is called once per fragmented packet.
864 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
865 u16 orig_network_offset, u8 mac_proto)
867 unsigned int hlen = skb_network_offset(skb);
868 struct ovs_frag_data *data;
870 data = this_cpu_ptr(&ovs_frag_data_storage);
871 data->dst = skb->_skb_refdst;
873 data->cb = *OVS_CB(skb);
874 data->inner_protocol = skb->inner_protocol;
875 data->network_offset = orig_network_offset;
876 if (skb_vlan_tag_present(skb))
877 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
880 data->vlan_proto = skb->vlan_proto;
881 data->mac_proto = mac_proto;
883 memcpy(&data->l2_data, skb->data, hlen);
885 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
889 static void ovs_fragment(struct net *net, struct vport *vport,
890 struct sk_buff *skb, u16 mru,
891 struct sw_flow_key *key)
893 u16 orig_network_offset = 0;
895 if (eth_p_mpls(skb->protocol)) {
896 orig_network_offset = skb_network_offset(skb);
897 skb->network_header = skb->inner_network_header;
900 if (skb_network_offset(skb) > MAX_L2_LEN) {
901 OVS_NLERR(1, "L2 header too long to fragment");
905 if (key->eth.type == htons(ETH_P_IP)) {
906 struct dst_entry ovs_dst;
907 unsigned long orig_dst;
909 prepare_frag(vport, skb, orig_network_offset,
910 ovs_key_mac_proto(key));
911 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
912 DST_OBSOLETE_NONE, DST_NOCOUNT);
913 ovs_dst.dev = vport->dev;
915 orig_dst = skb->_skb_refdst;
916 skb_dst_set_noref(skb, &ovs_dst);
917 IPCB(skb)->frag_max_size = mru;
919 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
920 refdst_drop(orig_dst);
921 } else if (key->eth.type == htons(ETH_P_IPV6)) {
922 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
923 unsigned long orig_dst;
924 struct rt6_info ovs_rt;
929 prepare_frag(vport, skb, orig_network_offset,
930 ovs_key_mac_proto(key));
931 memset(&ovs_rt, 0, sizeof(ovs_rt));
932 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
933 DST_OBSOLETE_NONE, DST_NOCOUNT);
934 ovs_rt.dst.dev = vport->dev;
936 orig_dst = skb->_skb_refdst;
937 skb_dst_set_noref(skb, &ovs_rt.dst);
938 IP6CB(skb)->frag_max_size = mru;
940 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
941 refdst_drop(orig_dst);
943 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
944 ovs_vport_name(vport), ntohs(key->eth.type), mru,
954 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
955 struct sw_flow_key *key)
957 struct vport *vport = ovs_vport_rcu(dp, out_port);
960 u16 mru = OVS_CB(skb)->mru;
961 u32 cutlen = OVS_CB(skb)->cutlen;
963 if (unlikely(cutlen > 0)) {
964 if (skb->len - cutlen > ovs_mac_header_len(key))
965 pskb_trim(skb, skb->len - cutlen);
967 pskb_trim(skb, ovs_mac_header_len(key));
971 (skb->len <= mru + vport->dev->hard_header_len))) {
972 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
973 } else if (mru <= vport->dev->mtu) {
974 struct net *net = read_pnet(&dp->net);
976 ovs_fragment(net, vport, skb, mru, key);
985 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
986 struct sw_flow_key *key, const struct nlattr *attr,
987 const struct nlattr *actions, int actions_len,
990 struct dp_upcall_info upcall;
991 const struct nlattr *a;
994 memset(&upcall, 0, sizeof(upcall));
995 upcall.cmd = OVS_PACKET_CMD_ACTION;
996 upcall.mru = OVS_CB(skb)->mru;
998 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
999 a = nla_next(a, &rem)) {
1000 switch (nla_type(a)) {
1001 case OVS_USERSPACE_ATTR_USERDATA:
1002 upcall.userdata = a;
1005 case OVS_USERSPACE_ATTR_PID:
1006 upcall.portid = nla_get_u32(a);
1009 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
1010 /* Get out tunnel info. */
1011 struct vport *vport;
1013 vport = ovs_vport_rcu(dp, nla_get_u32(a));
1017 err = dev_fill_metadata_dst(vport->dev, skb);
1019 upcall.egress_tun_info = skb_tunnel_info(skb);
1025 case OVS_USERSPACE_ATTR_ACTIONS: {
1026 /* Include actions. */
1027 upcall.actions = actions;
1028 upcall.actions_len = actions_len;
1032 } /* End of switch. */
1035 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1038 /* When 'last' is true, sample() should always consume the 'skb'.
1039 * Otherwise, sample() should keep 'skb' intact regardless what
1040 * actions are executed within sample().
1042 static int sample(struct datapath *dp, struct sk_buff *skb,
1043 struct sw_flow_key *key, const struct nlattr *attr,
1046 struct nlattr *actions;
1047 struct nlattr *sample_arg;
1048 int rem = nla_len(attr);
1049 const struct sample_arg *arg;
1050 bool clone_flow_key;
1052 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1053 sample_arg = nla_data(attr);
1054 arg = nla_data(sample_arg);
1055 actions = nla_next(sample_arg, &rem);
1057 if ((arg->probability != U32_MAX) &&
1058 (!arg->probability || prandom_u32() > arg->probability)) {
1064 clone_flow_key = !arg->exec;
1065 return clone_execute(dp, skb, key, 0, actions, rem, last,
1069 /* When 'last' is true, clone() should always consume the 'skb'.
1070 * Otherwise, clone() should keep 'skb' intact regardless what
1071 * actions are executed within clone().
1073 static int clone(struct datapath *dp, struct sk_buff *skb,
1074 struct sw_flow_key *key, const struct nlattr *attr,
1077 struct nlattr *actions;
1078 struct nlattr *clone_arg;
1079 int rem = nla_len(attr);
1080 bool dont_clone_flow_key;
1082 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1083 clone_arg = nla_data(attr);
1084 dont_clone_flow_key = nla_get_u32(clone_arg);
1085 actions = nla_next(clone_arg, &rem);
1087 return clone_execute(dp, skb, key, 0, actions, rem, last,
1088 !dont_clone_flow_key);
1091 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1092 const struct nlattr *attr)
1094 struct ovs_action_hash *hash_act = nla_data(attr);
1097 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1098 hash = skb_get_hash(skb);
1099 hash = jhash_1word(hash, hash_act->hash_basis);
1103 key->ovs_flow_hash = hash;
1106 static int execute_set_action(struct sk_buff *skb,
1107 struct sw_flow_key *flow_key,
1108 const struct nlattr *a)
1110 /* Only tunnel set execution is supported without a mask. */
1111 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1112 struct ovs_tunnel_info *tun = nla_data(a);
1115 dst_hold((struct dst_entry *)tun->tun_dst);
1116 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1123 /* Mask is at the midpoint of the data. */
1124 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1126 static int execute_masked_set_action(struct sk_buff *skb,
1127 struct sw_flow_key *flow_key,
1128 const struct nlattr *a)
1132 switch (nla_type(a)) {
1133 case OVS_KEY_ATTR_PRIORITY:
1134 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1135 *get_mask(a, u32 *));
1136 flow_key->phy.priority = skb->priority;
1139 case OVS_KEY_ATTR_SKB_MARK:
1140 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1141 flow_key->phy.skb_mark = skb->mark;
1144 case OVS_KEY_ATTR_TUNNEL_INFO:
1145 /* Masked data not supported for tunnel. */
1149 case OVS_KEY_ATTR_ETHERNET:
1150 err = set_eth_addr(skb, flow_key, nla_data(a),
1151 get_mask(a, struct ovs_key_ethernet *));
1154 case OVS_KEY_ATTR_NSH:
1155 err = set_nsh(skb, flow_key, a);
1158 case OVS_KEY_ATTR_IPV4:
1159 err = set_ipv4(skb, flow_key, nla_data(a),
1160 get_mask(a, struct ovs_key_ipv4 *));
1163 case OVS_KEY_ATTR_IPV6:
1164 err = set_ipv6(skb, flow_key, nla_data(a),
1165 get_mask(a, struct ovs_key_ipv6 *));
1168 case OVS_KEY_ATTR_TCP:
1169 err = set_tcp(skb, flow_key, nla_data(a),
1170 get_mask(a, struct ovs_key_tcp *));
1173 case OVS_KEY_ATTR_UDP:
1174 err = set_udp(skb, flow_key, nla_data(a),
1175 get_mask(a, struct ovs_key_udp *));
1178 case OVS_KEY_ATTR_SCTP:
1179 err = set_sctp(skb, flow_key, nla_data(a),
1180 get_mask(a, struct ovs_key_sctp *));
1183 case OVS_KEY_ATTR_MPLS:
1184 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1188 case OVS_KEY_ATTR_CT_STATE:
1189 case OVS_KEY_ATTR_CT_ZONE:
1190 case OVS_KEY_ATTR_CT_MARK:
1191 case OVS_KEY_ATTR_CT_LABELS:
1192 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1193 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1201 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1202 struct sw_flow_key *key,
1203 const struct nlattr *a, bool last)
1207 if (!is_flow_key_valid(key)) {
1210 err = ovs_flow_key_update(skb, key);
1214 BUG_ON(!is_flow_key_valid(key));
1216 recirc_id = nla_get_u32(a);
1217 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1220 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1221 struct sw_flow_key *key,
1222 const struct nlattr *attr, bool last)
1224 const struct nlattr *actions, *cpl_arg;
1225 const struct check_pkt_len_arg *arg;
1226 int rem = nla_len(attr);
1227 bool clone_flow_key;
1229 /* The first netlink attribute in 'attr' is always
1230 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1232 cpl_arg = nla_data(attr);
1233 arg = nla_data(cpl_arg);
1235 if (skb->len <= arg->pkt_len) {
1236 /* Second netlink attribute in 'attr' is always
1237 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1239 actions = nla_next(cpl_arg, &rem);
1240 clone_flow_key = !arg->exec_for_lesser_equal;
1242 /* Third netlink attribute in 'attr' is always
1243 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1245 actions = nla_next(cpl_arg, &rem);
1246 actions = nla_next(actions, &rem);
1247 clone_flow_key = !arg->exec_for_greater;
1250 return clone_execute(dp, skb, key, 0, nla_data(actions),
1251 nla_len(actions), last, clone_flow_key);
1254 /* Execute a list of actions against 'skb'. */
1255 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1256 struct sw_flow_key *key,
1257 const struct nlattr *attr, int len)
1259 const struct nlattr *a;
1262 for (a = attr, rem = len; rem > 0;
1263 a = nla_next(a, &rem)) {
1266 switch (nla_type(a)) {
1267 case OVS_ACTION_ATTR_OUTPUT: {
1268 int port = nla_get_u32(a);
1269 struct sk_buff *clone;
1271 /* Every output action needs a separate clone
1272 * of 'skb', In case the output action is the
1273 * last action, cloning can be avoided.
1275 if (nla_is_last(a, rem)) {
1276 do_output(dp, skb, port, key);
1277 /* 'skb' has been used for output.
1282 clone = skb_clone(skb, GFP_ATOMIC);
1284 do_output(dp, clone, port, key);
1285 OVS_CB(skb)->cutlen = 0;
1289 case OVS_ACTION_ATTR_TRUNC: {
1290 struct ovs_action_trunc *trunc = nla_data(a);
1292 if (skb->len > trunc->max_len)
1293 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1297 case OVS_ACTION_ATTR_USERSPACE:
1298 output_userspace(dp, skb, key, a, attr,
1299 len, OVS_CB(skb)->cutlen);
1300 OVS_CB(skb)->cutlen = 0;
1303 case OVS_ACTION_ATTR_HASH:
1304 execute_hash(skb, key, a);
1307 case OVS_ACTION_ATTR_PUSH_MPLS:
1308 err = push_mpls(skb, key, nla_data(a));
1311 case OVS_ACTION_ATTR_POP_MPLS:
1312 err = pop_mpls(skb, key, nla_get_be16(a));
1315 case OVS_ACTION_ATTR_PUSH_VLAN:
1316 err = push_vlan(skb, key, nla_data(a));
1319 case OVS_ACTION_ATTR_POP_VLAN:
1320 err = pop_vlan(skb, key);
1323 case OVS_ACTION_ATTR_RECIRC: {
1324 bool last = nla_is_last(a, rem);
1326 err = execute_recirc(dp, skb, key, a, last);
1328 /* If this is the last action, the skb has
1329 * been consumed or freed.
1330 * Return immediately.
1337 case OVS_ACTION_ATTR_SET:
1338 err = execute_set_action(skb, key, nla_data(a));
1341 case OVS_ACTION_ATTR_SET_MASKED:
1342 case OVS_ACTION_ATTR_SET_TO_MASKED:
1343 err = execute_masked_set_action(skb, key, nla_data(a));
1346 case OVS_ACTION_ATTR_SAMPLE: {
1347 bool last = nla_is_last(a, rem);
1349 err = sample(dp, skb, key, a, last);
1356 case OVS_ACTION_ATTR_CT:
1357 if (!is_flow_key_valid(key)) {
1358 err = ovs_flow_key_update(skb, key);
1363 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1366 /* Hide stolen IP fragments from user space. */
1368 return err == -EINPROGRESS ? 0 : err;
1371 case OVS_ACTION_ATTR_CT_CLEAR:
1372 err = ovs_ct_clear(skb, key);
1375 case OVS_ACTION_ATTR_PUSH_ETH:
1376 err = push_eth(skb, key, nla_data(a));
1379 case OVS_ACTION_ATTR_POP_ETH:
1380 err = pop_eth(skb, key);
1383 case OVS_ACTION_ATTR_PUSH_NSH: {
1384 u8 buffer[NSH_HDR_MAX_LEN];
1385 struct nshhdr *nh = (struct nshhdr *)buffer;
1387 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1391 err = push_nsh(skb, key, nh);
1395 case OVS_ACTION_ATTR_POP_NSH:
1396 err = pop_nsh(skb, key);
1399 case OVS_ACTION_ATTR_METER:
1400 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1406 case OVS_ACTION_ATTR_CLONE: {
1407 bool last = nla_is_last(a, rem);
1409 err = clone(dp, skb, key, a, last);
1416 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1417 bool last = nla_is_last(a, rem);
1419 err = execute_check_pkt_len(dp, skb, key, a, last);
1427 if (unlikely(err)) {
1437 /* Execute the actions on the clone of the packet. The effect of the
1438 * execution does not affect the original 'skb' nor the original 'key'.
1440 * The execution may be deferred in case the actions can not be executed
1443 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1444 struct sw_flow_key *key, u32 recirc_id,
1445 const struct nlattr *actions, int len,
1446 bool last, bool clone_flow_key)
1448 struct deferred_action *da;
1449 struct sw_flow_key *clone;
1451 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1453 /* Out of memory, skip this action.
1458 /* When clone_flow_key is false, the 'key' will not be change
1459 * by the actions, then the 'key' can be used directly.
1460 * Otherwise, try to clone key from the next recursion level of
1461 * 'flow_keys'. If clone is successful, execute the actions
1462 * without deferring.
1464 clone = clone_flow_key ? clone_key(key) : key;
1468 if (actions) { /* Sample action */
1470 __this_cpu_inc(exec_actions_level);
1472 err = do_execute_actions(dp, skb, clone,
1476 __this_cpu_dec(exec_actions_level);
1477 } else { /* Recirc action */
1478 clone->recirc_id = recirc_id;
1479 ovs_dp_process_packet(skb, clone);
1484 /* Out of 'flow_keys' space. Defer actions */
1485 da = add_deferred_actions(skb, key, actions, len);
1487 if (!actions) { /* Recirc action */
1489 key->recirc_id = recirc_id;
1492 /* Out of per CPU action FIFO space. Drop the 'skb' and
1497 if (net_ratelimit()) {
1498 if (actions) { /* Sample action */
1499 pr_warn("%s: deferred action limit reached, drop sample action\n",
1501 } else { /* Recirc action */
1502 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1510 static void process_deferred_actions(struct datapath *dp)
1512 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1514 /* Do not touch the FIFO in case there is no deferred actions. */
1515 if (action_fifo_is_empty(fifo))
1518 /* Finishing executing all deferred actions. */
1520 struct deferred_action *da = action_fifo_get(fifo);
1521 struct sk_buff *skb = da->skb;
1522 struct sw_flow_key *key = &da->pkt_key;
1523 const struct nlattr *actions = da->actions;
1524 int actions_len = da->actions_len;
1527 do_execute_actions(dp, skb, key, actions, actions_len);
1529 ovs_dp_process_packet(skb, key);
1530 } while (!action_fifo_is_empty(fifo));
1532 /* Reset FIFO for the next packet. */
1533 action_fifo_init(fifo);
1536 /* Execute a list of actions against 'skb'. */
1537 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1538 const struct sw_flow_actions *acts,
1539 struct sw_flow_key *key)
1543 level = __this_cpu_inc_return(exec_actions_level);
1544 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1545 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1552 OVS_CB(skb)->acts_origlen = acts->orig_len;
1553 err = do_execute_actions(dp, skb, key,
1554 acts->actions, acts->actions_len);
1557 process_deferred_actions(dp);
1560 __this_cpu_dec(exec_actions_level);
1564 int action_fifos_init(void)
1566 action_fifos = alloc_percpu(struct action_fifo);
1570 flow_keys = alloc_percpu(struct action_flow_keys);
1572 free_percpu(action_fifos);
1579 void action_fifos_exit(void)
1581 free_percpu(action_fifos);
1582 free_percpu(flow_keys);