2 * Copyright (c) 2007-2014 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"
47 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
48 struct sw_flow_key *key,
49 const struct nlattr *attr, int len);
51 struct deferred_action {
53 const struct nlattr *actions;
55 /* Store pkt_key clone when creating deferred action. */
56 struct sw_flow_key pkt_key;
59 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
60 struct ovs_frag_data {
64 __be16 inner_protocol;
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];
84 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
87 static struct action_fifo __percpu *action_fifos;
88 static struct recirc_keys __percpu *recirc_keys;
89 static DEFINE_PER_CPU(int, exec_actions_level);
91 static void action_fifo_init(struct action_fifo *fifo)
97 static bool action_fifo_is_empty(const struct action_fifo *fifo)
99 return (fifo->head == fifo->tail);
102 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
104 if (action_fifo_is_empty(fifo))
107 return &fifo->fifo[fifo->tail++];
110 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
112 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
115 return &fifo->fifo[fifo->head++];
118 /* Return true if fifo is not full */
119 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
120 const struct sw_flow_key *key,
121 const struct nlattr *attr)
123 struct action_fifo *fifo;
124 struct deferred_action *da;
126 fifo = this_cpu_ptr(action_fifos);
127 da = action_fifo_put(fifo);
137 static void invalidate_flow_key(struct sw_flow_key *key)
139 key->eth.type = htons(0);
142 static bool is_flow_key_valid(const struct sw_flow_key *key)
144 return !!key->eth.type;
147 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
150 if (skb->ip_summed == CHECKSUM_COMPLETE) {
151 __be16 diff[] = { ~(hdr->h_proto), ethertype };
153 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
156 hdr->h_proto = ethertype;
159 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
160 const struct ovs_action_push_mpls *mpls)
162 struct mpls_shim_hdr *new_mpls_lse;
164 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
165 if (skb->encapsulation)
168 if (skb_cow_head(skb, MPLS_HLEN) < 0)
171 if (!skb->inner_protocol) {
172 skb_set_inner_network_header(skb, skb->mac_len);
173 skb_set_inner_protocol(skb, skb->protocol);
176 skb_push(skb, MPLS_HLEN);
177 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
179 skb_reset_mac_header(skb);
180 skb_set_network_header(skb, skb->mac_len);
182 new_mpls_lse = mpls_hdr(skb);
183 new_mpls_lse->label_stack_entry = mpls->mpls_lse;
185 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
187 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
188 skb->protocol = mpls->mpls_ethertype;
190 invalidate_flow_key(key);
194 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
195 const __be16 ethertype)
200 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
204 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
206 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
209 __skb_pull(skb, MPLS_HLEN);
210 skb_reset_mac_header(skb);
211 skb_set_network_header(skb, skb->mac_len);
213 /* mpls_hdr() is used to locate the ethertype field correctly in the
214 * presence of VLAN tags.
216 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
217 update_ethertype(skb, hdr, ethertype);
218 if (eth_p_mpls(skb->protocol))
219 skb->protocol = ethertype;
221 invalidate_flow_key(key);
225 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
226 const __be32 *mpls_lse, const __be32 *mask)
228 struct mpls_shim_hdr *stack;
232 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
236 stack = mpls_hdr(skb);
237 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
238 if (skb->ip_summed == CHECKSUM_COMPLETE) {
239 __be32 diff[] = { ~(stack->label_stack_entry), lse };
241 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
244 stack->label_stack_entry = lse;
245 flow_key->mpls.top_lse = lse;
249 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
253 err = skb_vlan_pop(skb);
254 if (skb_vlan_tag_present(skb)) {
255 invalidate_flow_key(key);
257 key->eth.vlan.tci = 0;
258 key->eth.vlan.tpid = 0;
263 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
264 const struct ovs_action_push_vlan *vlan)
266 if (skb_vlan_tag_present(skb)) {
267 invalidate_flow_key(key);
269 key->eth.vlan.tci = vlan->vlan_tci;
270 key->eth.vlan.tpid = vlan->vlan_tpid;
272 return skb_vlan_push(skb, vlan->vlan_tpid,
273 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
276 /* 'src' is already properly masked. */
277 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
279 u16 *dst = (u16 *)dst_;
280 const u16 *src = (const u16 *)src_;
281 const u16 *mask = (const u16 *)mask_;
283 OVS_SET_MASKED(dst[0], src[0], mask[0]);
284 OVS_SET_MASKED(dst[1], src[1], mask[1]);
285 OVS_SET_MASKED(dst[2], src[2], mask[2]);
288 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
289 const struct ovs_key_ethernet *key,
290 const struct ovs_key_ethernet *mask)
294 err = skb_ensure_writable(skb, ETH_HLEN);
298 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
300 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
302 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
305 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
307 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
308 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
312 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
313 __be32 addr, __be32 new_addr)
315 int transport_len = skb->len - skb_transport_offset(skb);
317 if (nh->frag_off & htons(IP_OFFSET))
320 if (nh->protocol == IPPROTO_TCP) {
321 if (likely(transport_len >= sizeof(struct tcphdr)))
322 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
323 addr, new_addr, true);
324 } else if (nh->protocol == IPPROTO_UDP) {
325 if (likely(transport_len >= sizeof(struct udphdr))) {
326 struct udphdr *uh = udp_hdr(skb);
328 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
329 inet_proto_csum_replace4(&uh->check, skb,
330 addr, new_addr, true);
332 uh->check = CSUM_MANGLED_0;
338 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
339 __be32 *addr, __be32 new_addr)
341 update_ip_l4_checksum(skb, nh, *addr, new_addr);
342 csum_replace4(&nh->check, *addr, new_addr);
347 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
348 __be32 addr[4], const __be32 new_addr[4])
350 int transport_len = skb->len - skb_transport_offset(skb);
352 if (l4_proto == NEXTHDR_TCP) {
353 if (likely(transport_len >= sizeof(struct tcphdr)))
354 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
355 addr, new_addr, true);
356 } else if (l4_proto == NEXTHDR_UDP) {
357 if (likely(transport_len >= sizeof(struct udphdr))) {
358 struct udphdr *uh = udp_hdr(skb);
360 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
361 inet_proto_csum_replace16(&uh->check, skb,
362 addr, new_addr, true);
364 uh->check = CSUM_MANGLED_0;
367 } else if (l4_proto == NEXTHDR_ICMP) {
368 if (likely(transport_len >= sizeof(struct icmp6hdr)))
369 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
370 skb, addr, new_addr, true);
374 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
375 const __be32 mask[4], __be32 masked[4])
377 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
378 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
379 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
380 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
383 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
384 __be32 addr[4], const __be32 new_addr[4],
385 bool recalculate_csum)
387 if (recalculate_csum)
388 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
391 memcpy(addr, new_addr, sizeof(__be32[4]));
394 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
396 /* Bits 21-24 are always unmasked, so this retains their values. */
397 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
398 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
399 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
402 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
405 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
407 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
411 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
412 const struct ovs_key_ipv4 *key,
413 const struct ovs_key_ipv4 *mask)
419 err = skb_ensure_writable(skb, skb_network_offset(skb) +
420 sizeof(struct iphdr));
426 /* Setting an IP addresses is typically only a side effect of
427 * matching on them in the current userspace implementation, so it
428 * makes sense to check if the value actually changed.
430 if (mask->ipv4_src) {
431 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
433 if (unlikely(new_addr != nh->saddr)) {
434 set_ip_addr(skb, nh, &nh->saddr, new_addr);
435 flow_key->ipv4.addr.src = new_addr;
438 if (mask->ipv4_dst) {
439 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
441 if (unlikely(new_addr != nh->daddr)) {
442 set_ip_addr(skb, nh, &nh->daddr, new_addr);
443 flow_key->ipv4.addr.dst = new_addr;
446 if (mask->ipv4_tos) {
447 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
448 flow_key->ip.tos = nh->tos;
450 if (mask->ipv4_ttl) {
451 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
452 flow_key->ip.ttl = nh->ttl;
458 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
460 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
463 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
464 const struct ovs_key_ipv6 *key,
465 const struct ovs_key_ipv6 *mask)
470 err = skb_ensure_writable(skb, skb_network_offset(skb) +
471 sizeof(struct ipv6hdr));
477 /* Setting an IP addresses is typically only a side effect of
478 * matching on them in the current userspace implementation, so it
479 * makes sense to check if the value actually changed.
481 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
482 __be32 *saddr = (__be32 *)&nh->saddr;
485 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
487 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
488 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
490 memcpy(&flow_key->ipv6.addr.src, masked,
491 sizeof(flow_key->ipv6.addr.src));
494 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
495 unsigned int offset = 0;
496 int flags = IP6_FH_F_SKIP_RH;
497 bool recalc_csum = true;
498 __be32 *daddr = (__be32 *)&nh->daddr;
501 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
503 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
504 if (ipv6_ext_hdr(nh->nexthdr))
505 recalc_csum = (ipv6_find_hdr(skb, &offset,
510 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
512 memcpy(&flow_key->ipv6.addr.dst, masked,
513 sizeof(flow_key->ipv6.addr.dst));
516 if (mask->ipv6_tclass) {
517 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
518 flow_key->ip.tos = ipv6_get_dsfield(nh);
520 if (mask->ipv6_label) {
521 set_ipv6_fl(nh, ntohl(key->ipv6_label),
522 ntohl(mask->ipv6_label));
523 flow_key->ipv6.label =
524 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
526 if (mask->ipv6_hlimit) {
527 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
529 flow_key->ip.ttl = nh->hop_limit;
534 /* Must follow skb_ensure_writable() since that can move the skb data. */
535 static void set_tp_port(struct sk_buff *skb, __be16 *port,
536 __be16 new_port, __sum16 *check)
538 inet_proto_csum_replace2(check, skb, *port, new_port, false);
542 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
543 const struct ovs_key_udp *key,
544 const struct ovs_key_udp *mask)
550 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
551 sizeof(struct udphdr));
556 /* Either of the masks is non-zero, so do not bother checking them. */
557 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
558 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
560 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
561 if (likely(src != uh->source)) {
562 set_tp_port(skb, &uh->source, src, &uh->check);
563 flow_key->tp.src = src;
565 if (likely(dst != uh->dest)) {
566 set_tp_port(skb, &uh->dest, dst, &uh->check);
567 flow_key->tp.dst = dst;
570 if (unlikely(!uh->check))
571 uh->check = CSUM_MANGLED_0;
575 flow_key->tp.src = src;
576 flow_key->tp.dst = dst;
584 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
585 const struct ovs_key_tcp *key,
586 const struct ovs_key_tcp *mask)
592 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
593 sizeof(struct tcphdr));
598 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
599 if (likely(src != th->source)) {
600 set_tp_port(skb, &th->source, src, &th->check);
601 flow_key->tp.src = src;
603 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
604 if (likely(dst != th->dest)) {
605 set_tp_port(skb, &th->dest, dst, &th->check);
606 flow_key->tp.dst = dst;
613 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
614 const struct ovs_key_sctp *key,
615 const struct ovs_key_sctp *mask)
617 unsigned int sctphoff = skb_transport_offset(skb);
619 __le32 old_correct_csum, new_csum, old_csum;
622 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
627 old_csum = sh->checksum;
628 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
630 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
631 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
633 new_csum = sctp_compute_cksum(skb, sctphoff);
635 /* Carry any checksum errors through. */
636 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
639 flow_key->tp.src = sh->source;
640 flow_key->tp.dst = sh->dest;
645 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
647 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
648 struct vport *vport = data->vport;
650 if (skb_cow_head(skb, data->l2_len) < 0) {
655 __skb_dst_copy(skb, data->dst);
656 *OVS_CB(skb) = data->cb;
657 skb->inner_protocol = data->inner_protocol;
658 skb->vlan_tci = data->vlan_tci;
659 skb->vlan_proto = data->vlan_proto;
661 /* Reconstruct the MAC header. */
662 skb_push(skb, data->l2_len);
663 memcpy(skb->data, &data->l2_data, data->l2_len);
664 skb_postpush_rcsum(skb, skb->data, data->l2_len);
665 skb_reset_mac_header(skb);
667 ovs_vport_send(vport, skb);
672 ovs_dst_get_mtu(const struct dst_entry *dst)
674 return dst->dev->mtu;
677 static struct dst_ops ovs_dst_ops = {
679 .mtu = ovs_dst_get_mtu,
682 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
683 * ovs_vport_output(), which is called once per fragmented packet.
685 static void prepare_frag(struct vport *vport, struct sk_buff *skb)
687 unsigned int hlen = skb_network_offset(skb);
688 struct ovs_frag_data *data;
690 data = this_cpu_ptr(&ovs_frag_data_storage);
691 data->dst = skb->_skb_refdst;
693 data->cb = *OVS_CB(skb);
694 data->inner_protocol = skb->inner_protocol;
695 data->vlan_tci = skb->vlan_tci;
696 data->vlan_proto = skb->vlan_proto;
698 memcpy(&data->l2_data, skb->data, hlen);
700 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
704 static void ovs_fragment(struct net *net, struct vport *vport,
705 struct sk_buff *skb, u16 mru, __be16 ethertype)
707 if (skb_network_offset(skb) > MAX_L2_LEN) {
708 OVS_NLERR(1, "L2 header too long to fragment");
712 if (ethertype == htons(ETH_P_IP)) {
713 struct rtable ovs_rt = { 0 };
714 unsigned long orig_dst;
716 prepare_frag(vport, skb);
717 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
718 DST_OBSOLETE_NONE, DST_NOCOUNT);
719 ovs_rt.dst.dev = vport->dev;
721 orig_dst = skb->_skb_refdst;
722 skb_dst_set_noref(skb, &ovs_rt.dst);
723 IPCB(skb)->frag_max_size = mru;
725 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
726 refdst_drop(orig_dst);
727 } else if (ethertype == htons(ETH_P_IPV6)) {
728 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
729 unsigned long orig_dst;
730 struct rt6_info ovs_rt;
736 prepare_frag(vport, skb);
737 memset(&ovs_rt, 0, sizeof(ovs_rt));
738 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
739 DST_OBSOLETE_NONE, DST_NOCOUNT);
740 ovs_rt.dst.dev = vport->dev;
742 orig_dst = skb->_skb_refdst;
743 skb_dst_set_noref(skb, &ovs_rt.dst);
744 IP6CB(skb)->frag_max_size = mru;
746 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
747 refdst_drop(orig_dst);
749 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
750 ovs_vport_name(vport), ntohs(ethertype), mru,
760 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
761 struct sw_flow_key *key)
763 struct vport *vport = ovs_vport_rcu(dp, out_port);
766 u16 mru = OVS_CB(skb)->mru;
767 u32 cutlen = OVS_CB(skb)->cutlen;
769 if (unlikely(cutlen > 0)) {
770 if (skb->len - cutlen > ETH_HLEN)
771 pskb_trim(skb, skb->len - cutlen);
773 pskb_trim(skb, ETH_HLEN);
776 if (likely(!mru || (skb->len <= mru + ETH_HLEN))) {
777 ovs_vport_send(vport, skb);
778 } else if (mru <= vport->dev->mtu) {
779 struct net *net = read_pnet(&dp->net);
780 __be16 ethertype = key->eth.type;
782 if (!is_flow_key_valid(key)) {
783 if (eth_p_mpls(skb->protocol))
784 ethertype = skb->inner_protocol;
786 ethertype = vlan_get_protocol(skb);
789 ovs_fragment(net, vport, skb, mru, ethertype);
798 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
799 struct sw_flow_key *key, const struct nlattr *attr,
800 const struct nlattr *actions, int actions_len,
803 struct dp_upcall_info upcall;
804 const struct nlattr *a;
807 memset(&upcall, 0, sizeof(upcall));
808 upcall.cmd = OVS_PACKET_CMD_ACTION;
809 upcall.mru = OVS_CB(skb)->mru;
811 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
812 a = nla_next(a, &rem)) {
813 switch (nla_type(a)) {
814 case OVS_USERSPACE_ATTR_USERDATA:
818 case OVS_USERSPACE_ATTR_PID:
819 upcall.portid = nla_get_u32(a);
822 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
823 /* Get out tunnel info. */
826 vport = ovs_vport_rcu(dp, nla_get_u32(a));
830 err = dev_fill_metadata_dst(vport->dev, skb);
832 upcall.egress_tun_info = skb_tunnel_info(skb);
838 case OVS_USERSPACE_ATTR_ACTIONS: {
839 /* Include actions. */
840 upcall.actions = actions;
841 upcall.actions_len = actions_len;
845 } /* End of switch. */
848 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
851 static int sample(struct datapath *dp, struct sk_buff *skb,
852 struct sw_flow_key *key, const struct nlattr *attr,
853 const struct nlattr *actions, int actions_len)
855 const struct nlattr *acts_list = NULL;
856 const struct nlattr *a;
860 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
861 a = nla_next(a, &rem)) {
864 switch (nla_type(a)) {
865 case OVS_SAMPLE_ATTR_PROBABILITY:
866 probability = nla_get_u32(a);
867 if (!probability || prandom_u32() > probability)
871 case OVS_SAMPLE_ATTR_ACTIONS:
877 rem = nla_len(acts_list);
878 a = nla_data(acts_list);
880 /* Actions list is empty, do nothing */
884 /* The only known usage of sample action is having a single user-space
885 * action, or having a truncate action followed by a single user-space
886 * action. Treat this usage as a special case.
887 * The output_userspace() should clone the skb to be sent to the
888 * user space. This skb will be consumed by its caller.
890 if (unlikely(nla_type(a) == OVS_ACTION_ATTR_TRUNC)) {
891 struct ovs_action_trunc *trunc = nla_data(a);
893 if (skb->len > trunc->max_len)
894 cutlen = skb->len - trunc->max_len;
896 a = nla_next(a, &rem);
899 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
900 nla_is_last(a, rem)))
901 return output_userspace(dp, skb, key, a, actions,
902 actions_len, cutlen);
904 skb = skb_clone(skb, GFP_ATOMIC);
906 /* Skip the sample action when out of memory. */
909 if (!add_deferred_actions(skb, key, a)) {
911 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
919 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
920 const struct nlattr *attr)
922 struct ovs_action_hash *hash_act = nla_data(attr);
925 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
926 hash = skb_get_hash(skb);
927 hash = jhash_1word(hash, hash_act->hash_basis);
931 key->ovs_flow_hash = hash;
934 static int execute_set_action(struct sk_buff *skb,
935 struct sw_flow_key *flow_key,
936 const struct nlattr *a)
938 /* Only tunnel set execution is supported without a mask. */
939 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
940 struct ovs_tunnel_info *tun = nla_data(a);
943 dst_hold((struct dst_entry *)tun->tun_dst);
944 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
951 /* Mask is at the midpoint of the data. */
952 #define get_mask(a, type) ((const type)nla_data(a) + 1)
954 static int execute_masked_set_action(struct sk_buff *skb,
955 struct sw_flow_key *flow_key,
956 const struct nlattr *a)
960 switch (nla_type(a)) {
961 case OVS_KEY_ATTR_PRIORITY:
962 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
963 *get_mask(a, u32 *));
964 flow_key->phy.priority = skb->priority;
967 case OVS_KEY_ATTR_SKB_MARK:
968 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
969 flow_key->phy.skb_mark = skb->mark;
972 case OVS_KEY_ATTR_TUNNEL_INFO:
973 /* Masked data not supported for tunnel. */
977 case OVS_KEY_ATTR_ETHERNET:
978 err = set_eth_addr(skb, flow_key, nla_data(a),
979 get_mask(a, struct ovs_key_ethernet *));
982 case OVS_KEY_ATTR_IPV4:
983 err = set_ipv4(skb, flow_key, nla_data(a),
984 get_mask(a, struct ovs_key_ipv4 *));
987 case OVS_KEY_ATTR_IPV6:
988 err = set_ipv6(skb, flow_key, nla_data(a),
989 get_mask(a, struct ovs_key_ipv6 *));
992 case OVS_KEY_ATTR_TCP:
993 err = set_tcp(skb, flow_key, nla_data(a),
994 get_mask(a, struct ovs_key_tcp *));
997 case OVS_KEY_ATTR_UDP:
998 err = set_udp(skb, flow_key, nla_data(a),
999 get_mask(a, struct ovs_key_udp *));
1002 case OVS_KEY_ATTR_SCTP:
1003 err = set_sctp(skb, flow_key, nla_data(a),
1004 get_mask(a, struct ovs_key_sctp *));
1007 case OVS_KEY_ATTR_MPLS:
1008 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1012 case OVS_KEY_ATTR_CT_STATE:
1013 case OVS_KEY_ATTR_CT_ZONE:
1014 case OVS_KEY_ATTR_CT_MARK:
1015 case OVS_KEY_ATTR_CT_LABELS:
1023 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1024 struct sw_flow_key *key,
1025 const struct nlattr *a, int rem)
1027 struct deferred_action *da;
1030 if (!is_flow_key_valid(key)) {
1033 err = ovs_flow_key_update(skb, key);
1037 BUG_ON(!is_flow_key_valid(key));
1039 if (!nla_is_last(a, rem)) {
1040 /* Recirc action is the not the last action
1041 * of the action list, need to clone the skb.
1043 skb = skb_clone(skb, GFP_ATOMIC);
1045 /* Skip the recirc action when out of memory, but
1046 * continue on with the rest of the action list.
1052 level = this_cpu_read(exec_actions_level);
1053 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
1054 struct recirc_keys *rks = this_cpu_ptr(recirc_keys);
1055 struct sw_flow_key *recirc_key = &rks->key[level - 1];
1058 recirc_key->recirc_id = nla_get_u32(a);
1059 ovs_dp_process_packet(skb, recirc_key);
1064 da = add_deferred_actions(skb, key, NULL);
1066 da->pkt_key.recirc_id = nla_get_u32(a);
1070 if (net_ratelimit())
1071 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1078 /* Execute a list of actions against 'skb'. */
1079 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1080 struct sw_flow_key *key,
1081 const struct nlattr *attr, int len)
1083 /* Every output action needs a separate clone of 'skb', but the common
1084 * case is just a single output action, so that doing a clone and
1085 * then freeing the original skbuff is wasteful. So the following code
1086 * is slightly obscure just to avoid that.
1089 const struct nlattr *a;
1092 for (a = attr, rem = len; rem > 0;
1093 a = nla_next(a, &rem)) {
1096 if (unlikely(prev_port != -1)) {
1097 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
1100 do_output(dp, out_skb, prev_port, key);
1102 OVS_CB(skb)->cutlen = 0;
1106 switch (nla_type(a)) {
1107 case OVS_ACTION_ATTR_OUTPUT:
1108 prev_port = nla_get_u32(a);
1111 case OVS_ACTION_ATTR_TRUNC: {
1112 struct ovs_action_trunc *trunc = nla_data(a);
1114 if (skb->len > trunc->max_len)
1115 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1119 case OVS_ACTION_ATTR_USERSPACE:
1120 output_userspace(dp, skb, key, a, attr,
1121 len, OVS_CB(skb)->cutlen);
1122 OVS_CB(skb)->cutlen = 0;
1125 case OVS_ACTION_ATTR_HASH:
1126 execute_hash(skb, key, a);
1129 case OVS_ACTION_ATTR_PUSH_MPLS:
1130 err = push_mpls(skb, key, nla_data(a));
1133 case OVS_ACTION_ATTR_POP_MPLS:
1134 err = pop_mpls(skb, key, nla_get_be16(a));
1137 case OVS_ACTION_ATTR_PUSH_VLAN:
1138 err = push_vlan(skb, key, nla_data(a));
1141 case OVS_ACTION_ATTR_POP_VLAN:
1142 err = pop_vlan(skb, key);
1145 case OVS_ACTION_ATTR_RECIRC:
1146 err = execute_recirc(dp, skb, key, a, rem);
1147 if (nla_is_last(a, rem)) {
1148 /* If this is the last action, the skb has
1149 * been consumed or freed.
1150 * Return immediately.
1156 case OVS_ACTION_ATTR_SET:
1157 err = execute_set_action(skb, key, nla_data(a));
1160 case OVS_ACTION_ATTR_SET_MASKED:
1161 case OVS_ACTION_ATTR_SET_TO_MASKED:
1162 err = execute_masked_set_action(skb, key, nla_data(a));
1165 case OVS_ACTION_ATTR_SAMPLE:
1166 err = sample(dp, skb, key, a, attr, len);
1169 case OVS_ACTION_ATTR_CT:
1170 if (!is_flow_key_valid(key)) {
1171 err = ovs_flow_key_update(skb, key);
1176 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1179 /* Hide stolen IP fragments from user space. */
1181 return err == -EINPROGRESS ? 0 : err;
1185 if (unlikely(err)) {
1191 if (prev_port != -1)
1192 do_output(dp, skb, prev_port, key);
1199 static void process_deferred_actions(struct datapath *dp)
1201 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1203 /* Do not touch the FIFO in case there is no deferred actions. */
1204 if (action_fifo_is_empty(fifo))
1207 /* Finishing executing all deferred actions. */
1209 struct deferred_action *da = action_fifo_get(fifo);
1210 struct sk_buff *skb = da->skb;
1211 struct sw_flow_key *key = &da->pkt_key;
1212 const struct nlattr *actions = da->actions;
1215 do_execute_actions(dp, skb, key, actions,
1218 ovs_dp_process_packet(skb, key);
1219 } while (!action_fifo_is_empty(fifo));
1221 /* Reset FIFO for the next packet. */
1222 action_fifo_init(fifo);
1225 /* Execute a list of actions against 'skb'. */
1226 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1227 const struct sw_flow_actions *acts,
1228 struct sw_flow_key *key)
1232 level = __this_cpu_inc_return(exec_actions_level);
1233 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1234 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1241 OVS_CB(skb)->acts_origlen = acts->orig_len;
1242 err = do_execute_actions(dp, skb, key,
1243 acts->actions, acts->actions_len);
1246 process_deferred_actions(dp);
1249 __this_cpu_dec(exec_actions_level);
1253 int action_fifos_init(void)
1255 action_fifos = alloc_percpu(struct action_fifo);
1259 recirc_keys = alloc_percpu(struct recirc_keys);
1261 free_percpu(action_fifos);
1268 void action_fifos_exit(void)
1270 free_percpu(action_fifos);
1271 free_percpu(recirc_keys);