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/sctp.h>
26 #include <linux/tcp.h>
27 #include <linux/udp.h>
28 #include <linux/in6.h>
29 #include <linux/if_arp.h>
30 #include <linux/if_vlan.h>
34 #include <net/checksum.h>
35 #include <net/dsfield.h>
37 #include <net/sctp/checksum.h>
43 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
44 struct sw_flow_key *key,
45 const struct nlattr *attr, int len);
47 struct deferred_action {
49 const struct nlattr *actions;
51 /* Store pkt_key clone when creating deferred action. */
52 struct sw_flow_key pkt_key;
55 #define DEFERRED_ACTION_FIFO_SIZE 10
59 /* Deferred action fifo queue storage. */
60 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
63 static struct action_fifo __percpu *action_fifos;
64 static DEFINE_PER_CPU(int, exec_actions_level);
66 static void action_fifo_init(struct action_fifo *fifo)
72 static bool action_fifo_is_empty(const struct action_fifo *fifo)
74 return (fifo->head == fifo->tail);
77 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
79 if (action_fifo_is_empty(fifo))
82 return &fifo->fifo[fifo->tail++];
85 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
87 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
90 return &fifo->fifo[fifo->head++];
93 /* Return true if fifo is not full */
94 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
95 const struct sw_flow_key *key,
96 const struct nlattr *attr)
98 struct action_fifo *fifo;
99 struct deferred_action *da;
101 fifo = this_cpu_ptr(action_fifos);
102 da = action_fifo_put(fifo);
112 static void invalidate_flow_key(struct sw_flow_key *key)
114 key->eth.type = htons(0);
117 static bool is_flow_key_valid(const struct sw_flow_key *key)
119 return !!key->eth.type;
122 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
123 const struct ovs_action_push_mpls *mpls)
125 __be32 *new_mpls_lse;
128 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
129 if (skb->encapsulation)
132 if (skb_cow_head(skb, MPLS_HLEN) < 0)
135 skb_push(skb, MPLS_HLEN);
136 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
138 skb_reset_mac_header(skb);
140 new_mpls_lse = (__be32 *)skb_mpls_header(skb);
141 *new_mpls_lse = mpls->mpls_lse;
143 if (skb->ip_summed == CHECKSUM_COMPLETE)
144 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
148 hdr->h_proto = mpls->mpls_ethertype;
150 if (!skb->inner_protocol)
151 skb_set_inner_protocol(skb, skb->protocol);
152 skb->protocol = mpls->mpls_ethertype;
154 invalidate_flow_key(key);
158 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
159 const __be16 ethertype)
164 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
168 skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
170 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
173 __skb_pull(skb, MPLS_HLEN);
174 skb_reset_mac_header(skb);
176 /* skb_mpls_header() is used to locate the ethertype
177 * field correctly in the presence of VLAN tags.
179 hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
180 hdr->h_proto = ethertype;
181 if (eth_p_mpls(skb->protocol))
182 skb->protocol = ethertype;
184 invalidate_flow_key(key);
188 /* 'KEY' must not have any bits set outside of the 'MASK' */
189 #define MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK)))
190 #define SET_MASKED(OLD, KEY, MASK) ((OLD) = MASKED(OLD, KEY, MASK))
192 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
193 const __be32 *mpls_lse, const __be32 *mask)
199 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
203 stack = (__be32 *)skb_mpls_header(skb);
204 lse = MASKED(*stack, *mpls_lse, *mask);
205 if (skb->ip_summed == CHECKSUM_COMPLETE) {
206 __be32 diff[] = { ~(*stack), lse };
208 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
213 flow_key->mpls.top_lse = lse;
217 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
221 err = skb_vlan_pop(skb);
222 if (skb_vlan_tag_present(skb))
223 invalidate_flow_key(key);
229 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
230 const struct ovs_action_push_vlan *vlan)
232 if (skb_vlan_tag_present(skb))
233 invalidate_flow_key(key);
235 key->eth.tci = vlan->vlan_tci;
236 return skb_vlan_push(skb, vlan->vlan_tpid,
237 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
240 /* 'src' is already properly masked. */
241 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
243 u16 *dst = (u16 *)dst_;
244 const u16 *src = (const u16 *)src_;
245 const u16 *mask = (const u16 *)mask_;
247 SET_MASKED(dst[0], src[0], mask[0]);
248 SET_MASKED(dst[1], src[1], mask[1]);
249 SET_MASKED(dst[2], src[2], mask[2]);
252 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
253 const struct ovs_key_ethernet *key,
254 const struct ovs_key_ethernet *mask)
258 err = skb_ensure_writable(skb, ETH_HLEN);
262 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
264 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
266 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
269 ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
271 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
272 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
276 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
277 __be32 addr, __be32 new_addr)
279 int transport_len = skb->len - skb_transport_offset(skb);
281 if (nh->frag_off & htons(IP_OFFSET))
284 if (nh->protocol == IPPROTO_TCP) {
285 if (likely(transport_len >= sizeof(struct tcphdr)))
286 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
288 } else if (nh->protocol == IPPROTO_UDP) {
289 if (likely(transport_len >= sizeof(struct udphdr))) {
290 struct udphdr *uh = udp_hdr(skb);
292 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
293 inet_proto_csum_replace4(&uh->check, skb,
296 uh->check = CSUM_MANGLED_0;
302 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
303 __be32 *addr, __be32 new_addr)
305 update_ip_l4_checksum(skb, nh, *addr, new_addr);
306 csum_replace4(&nh->check, *addr, new_addr);
311 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
312 __be32 addr[4], const __be32 new_addr[4])
314 int transport_len = skb->len - skb_transport_offset(skb);
316 if (l4_proto == NEXTHDR_TCP) {
317 if (likely(transport_len >= sizeof(struct tcphdr)))
318 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
320 } else if (l4_proto == NEXTHDR_UDP) {
321 if (likely(transport_len >= sizeof(struct udphdr))) {
322 struct udphdr *uh = udp_hdr(skb);
324 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
325 inet_proto_csum_replace16(&uh->check, skb,
328 uh->check = CSUM_MANGLED_0;
331 } else if (l4_proto == NEXTHDR_ICMP) {
332 if (likely(transport_len >= sizeof(struct icmp6hdr)))
333 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
334 skb, addr, new_addr, 1);
338 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
339 const __be32 mask[4], __be32 masked[4])
341 masked[0] = MASKED(old[0], addr[0], mask[0]);
342 masked[1] = MASKED(old[1], addr[1], mask[1]);
343 masked[2] = MASKED(old[2], addr[2], mask[2]);
344 masked[3] = MASKED(old[3], addr[3], mask[3]);
347 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
348 __be32 addr[4], const __be32 new_addr[4],
349 bool recalculate_csum)
351 if (recalculate_csum)
352 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
355 memcpy(addr, new_addr, sizeof(__be32[4]));
358 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
360 /* Bits 21-24 are always unmasked, so this retains their values. */
361 SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
362 SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
363 SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
366 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
369 new_ttl = MASKED(nh->ttl, new_ttl, mask);
371 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
375 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
376 const struct ovs_key_ipv4 *key,
377 const struct ovs_key_ipv4 *mask)
383 err = skb_ensure_writable(skb, skb_network_offset(skb) +
384 sizeof(struct iphdr));
390 /* Setting an IP addresses is typically only a side effect of
391 * matching on them in the current userspace implementation, so it
392 * makes sense to check if the value actually changed.
394 if (mask->ipv4_src) {
395 new_addr = MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
397 if (unlikely(new_addr != nh->saddr)) {
398 set_ip_addr(skb, nh, &nh->saddr, new_addr);
399 flow_key->ipv4.addr.src = new_addr;
402 if (mask->ipv4_dst) {
403 new_addr = MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
405 if (unlikely(new_addr != nh->daddr)) {
406 set_ip_addr(skb, nh, &nh->daddr, new_addr);
407 flow_key->ipv4.addr.dst = new_addr;
410 if (mask->ipv4_tos) {
411 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
412 flow_key->ip.tos = nh->tos;
414 if (mask->ipv4_ttl) {
415 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
416 flow_key->ip.ttl = nh->ttl;
422 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
424 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
427 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
428 const struct ovs_key_ipv6 *key,
429 const struct ovs_key_ipv6 *mask)
434 err = skb_ensure_writable(skb, skb_network_offset(skb) +
435 sizeof(struct ipv6hdr));
441 /* Setting an IP addresses is typically only a side effect of
442 * matching on them in the current userspace implementation, so it
443 * makes sense to check if the value actually changed.
445 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
446 __be32 *saddr = (__be32 *)&nh->saddr;
449 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
451 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
452 set_ipv6_addr(skb, key->ipv6_proto, saddr, masked,
454 memcpy(&flow_key->ipv6.addr.src, masked,
455 sizeof(flow_key->ipv6.addr.src));
458 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
459 unsigned int offset = 0;
460 int flags = IP6_FH_F_SKIP_RH;
461 bool recalc_csum = true;
462 __be32 *daddr = (__be32 *)&nh->daddr;
465 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
467 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
468 if (ipv6_ext_hdr(nh->nexthdr))
469 recalc_csum = (ipv6_find_hdr(skb, &offset,
474 set_ipv6_addr(skb, key->ipv6_proto, daddr, masked,
476 memcpy(&flow_key->ipv6.addr.dst, masked,
477 sizeof(flow_key->ipv6.addr.dst));
480 if (mask->ipv6_tclass) {
481 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
482 flow_key->ip.tos = ipv6_get_dsfield(nh);
484 if (mask->ipv6_label) {
485 set_ipv6_fl(nh, ntohl(key->ipv6_label),
486 ntohl(mask->ipv6_label));
487 flow_key->ipv6.label =
488 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
490 if (mask->ipv6_hlimit) {
491 SET_MASKED(nh->hop_limit, key->ipv6_hlimit, mask->ipv6_hlimit);
492 flow_key->ip.ttl = nh->hop_limit;
497 /* Must follow skb_ensure_writable() since that can move the skb data. */
498 static void set_tp_port(struct sk_buff *skb, __be16 *port,
499 __be16 new_port, __sum16 *check)
501 inet_proto_csum_replace2(check, skb, *port, new_port, 0);
505 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
506 const struct ovs_key_udp *key,
507 const struct ovs_key_udp *mask)
513 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
514 sizeof(struct udphdr));
519 /* Either of the masks is non-zero, so do not bother checking them. */
520 src = MASKED(uh->source, key->udp_src, mask->udp_src);
521 dst = MASKED(uh->dest, key->udp_dst, mask->udp_dst);
523 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
524 if (likely(src != uh->source)) {
525 set_tp_port(skb, &uh->source, src, &uh->check);
526 flow_key->tp.src = src;
528 if (likely(dst != uh->dest)) {
529 set_tp_port(skb, &uh->dest, dst, &uh->check);
530 flow_key->tp.dst = dst;
533 if (unlikely(!uh->check))
534 uh->check = CSUM_MANGLED_0;
538 flow_key->tp.src = src;
539 flow_key->tp.dst = dst;
547 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
548 const struct ovs_key_tcp *key,
549 const struct ovs_key_tcp *mask)
555 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
556 sizeof(struct tcphdr));
561 src = MASKED(th->source, key->tcp_src, mask->tcp_src);
562 if (likely(src != th->source)) {
563 set_tp_port(skb, &th->source, src, &th->check);
564 flow_key->tp.src = src;
566 dst = MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
567 if (likely(dst != th->dest)) {
568 set_tp_port(skb, &th->dest, dst, &th->check);
569 flow_key->tp.dst = dst;
576 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
577 const struct ovs_key_sctp *key,
578 const struct ovs_key_sctp *mask)
580 unsigned int sctphoff = skb_transport_offset(skb);
582 __le32 old_correct_csum, new_csum, old_csum;
585 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
590 old_csum = sh->checksum;
591 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
593 sh->source = MASKED(sh->source, key->sctp_src, mask->sctp_src);
594 sh->dest = MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
596 new_csum = sctp_compute_cksum(skb, sctphoff);
598 /* Carry any checksum errors through. */
599 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
602 flow_key->tp.src = sh->source;
603 flow_key->tp.dst = sh->dest;
608 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
610 struct vport *vport = ovs_vport_rcu(dp, out_port);
613 ovs_vport_send(vport, skb);
618 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
619 struct sw_flow_key *key, const struct nlattr *attr,
620 const struct nlattr *actions, int actions_len)
622 struct ip_tunnel_info info;
623 struct dp_upcall_info upcall;
624 const struct nlattr *a;
627 memset(&upcall, 0, sizeof(upcall));
628 upcall.cmd = OVS_PACKET_CMD_ACTION;
630 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
631 a = nla_next(a, &rem)) {
632 switch (nla_type(a)) {
633 case OVS_USERSPACE_ATTR_USERDATA:
637 case OVS_USERSPACE_ATTR_PID:
638 upcall.portid = nla_get_u32(a);
641 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
642 /* Get out tunnel info. */
645 vport = ovs_vport_rcu(dp, nla_get_u32(a));
649 err = ovs_vport_get_egress_tun_info(vport, skb,
652 upcall.egress_tun_info = &info;
657 case OVS_USERSPACE_ATTR_ACTIONS: {
658 /* Include actions. */
659 upcall.actions = actions;
660 upcall.actions_len = actions_len;
664 } /* End of switch. */
667 return ovs_dp_upcall(dp, skb, key, &upcall);
670 static int sample(struct datapath *dp, struct sk_buff *skb,
671 struct sw_flow_key *key, const struct nlattr *attr,
672 const struct nlattr *actions, int actions_len)
674 const struct nlattr *acts_list = NULL;
675 const struct nlattr *a;
678 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
679 a = nla_next(a, &rem)) {
682 switch (nla_type(a)) {
683 case OVS_SAMPLE_ATTR_PROBABILITY:
684 probability = nla_get_u32(a);
685 if (!probability || prandom_u32() > probability)
689 case OVS_SAMPLE_ATTR_ACTIONS:
695 rem = nla_len(acts_list);
696 a = nla_data(acts_list);
698 /* Actions list is empty, do nothing */
702 /* The only known usage of sample action is having a single user-space
703 * action. Treat this usage as a special case.
704 * The output_userspace() should clone the skb to be sent to the
705 * user space. This skb will be consumed by its caller.
707 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
708 nla_is_last(a, rem)))
709 return output_userspace(dp, skb, key, a, actions, actions_len);
711 skb = skb_clone(skb, GFP_ATOMIC);
713 /* Skip the sample action when out of memory. */
716 if (!add_deferred_actions(skb, key, a)) {
718 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
726 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
727 const struct nlattr *attr)
729 struct ovs_action_hash *hash_act = nla_data(attr);
732 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
733 hash = skb_get_hash(skb);
734 hash = jhash_1word(hash, hash_act->hash_basis);
738 key->ovs_flow_hash = hash;
741 static int execute_set_action(struct sk_buff *skb,
742 struct sw_flow_key *flow_key,
743 const struct nlattr *a)
745 /* Only tunnel set execution is supported without a mask. */
746 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
747 struct ovs_tunnel_info *tun = nla_data(a);
750 dst_hold((struct dst_entry *)tun->tun_dst);
751 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
753 /* FIXME: Remove when all vports have been converted */
754 OVS_CB(skb)->egress_tun_info = &tun->tun_dst->u.tun_info;
762 /* Mask is at the midpoint of the data. */
763 #define get_mask(a, type) ((const type)nla_data(a) + 1)
765 static int execute_masked_set_action(struct sk_buff *skb,
766 struct sw_flow_key *flow_key,
767 const struct nlattr *a)
771 switch (nla_type(a)) {
772 case OVS_KEY_ATTR_PRIORITY:
773 SET_MASKED(skb->priority, nla_get_u32(a), *get_mask(a, u32 *));
774 flow_key->phy.priority = skb->priority;
777 case OVS_KEY_ATTR_SKB_MARK:
778 SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
779 flow_key->phy.skb_mark = skb->mark;
782 case OVS_KEY_ATTR_TUNNEL_INFO:
783 /* Masked data not supported for tunnel. */
787 case OVS_KEY_ATTR_ETHERNET:
788 err = set_eth_addr(skb, flow_key, nla_data(a),
789 get_mask(a, struct ovs_key_ethernet *));
792 case OVS_KEY_ATTR_IPV4:
793 err = set_ipv4(skb, flow_key, nla_data(a),
794 get_mask(a, struct ovs_key_ipv4 *));
797 case OVS_KEY_ATTR_IPV6:
798 err = set_ipv6(skb, flow_key, nla_data(a),
799 get_mask(a, struct ovs_key_ipv6 *));
802 case OVS_KEY_ATTR_TCP:
803 err = set_tcp(skb, flow_key, nla_data(a),
804 get_mask(a, struct ovs_key_tcp *));
807 case OVS_KEY_ATTR_UDP:
808 err = set_udp(skb, flow_key, nla_data(a),
809 get_mask(a, struct ovs_key_udp *));
812 case OVS_KEY_ATTR_SCTP:
813 err = set_sctp(skb, flow_key, nla_data(a),
814 get_mask(a, struct ovs_key_sctp *));
817 case OVS_KEY_ATTR_MPLS:
818 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
826 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
827 struct sw_flow_key *key,
828 const struct nlattr *a, int rem)
830 struct deferred_action *da;
832 if (!is_flow_key_valid(key)) {
835 err = ovs_flow_key_update(skb, key);
839 BUG_ON(!is_flow_key_valid(key));
841 if (!nla_is_last(a, rem)) {
842 /* Recirc action is the not the last action
843 * of the action list, need to clone the skb.
845 skb = skb_clone(skb, GFP_ATOMIC);
847 /* Skip the recirc action when out of memory, but
848 * continue on with the rest of the action list.
854 da = add_deferred_actions(skb, key, NULL);
856 da->pkt_key.recirc_id = nla_get_u32(a);
861 pr_warn("%s: deferred action limit reached, drop recirc action\n",
868 /* Execute a list of actions against 'skb'. */
869 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
870 struct sw_flow_key *key,
871 const struct nlattr *attr, int len)
873 /* Every output action needs a separate clone of 'skb', but the common
874 * case is just a single output action, so that doing a clone and
875 * then freeing the original skbuff is wasteful. So the following code
876 * is slightly obscure just to avoid that.
879 const struct nlattr *a;
882 for (a = attr, rem = len; rem > 0;
883 a = nla_next(a, &rem)) {
886 if (unlikely(prev_port != -1)) {
887 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
890 do_output(dp, out_skb, prev_port);
895 switch (nla_type(a)) {
896 case OVS_ACTION_ATTR_OUTPUT:
897 prev_port = nla_get_u32(a);
900 case OVS_ACTION_ATTR_USERSPACE:
901 output_userspace(dp, skb, key, a, attr, len);
904 case OVS_ACTION_ATTR_HASH:
905 execute_hash(skb, key, a);
908 case OVS_ACTION_ATTR_PUSH_MPLS:
909 err = push_mpls(skb, key, nla_data(a));
912 case OVS_ACTION_ATTR_POP_MPLS:
913 err = pop_mpls(skb, key, nla_get_be16(a));
916 case OVS_ACTION_ATTR_PUSH_VLAN:
917 err = push_vlan(skb, key, nla_data(a));
920 case OVS_ACTION_ATTR_POP_VLAN:
921 err = pop_vlan(skb, key);
924 case OVS_ACTION_ATTR_RECIRC:
925 err = execute_recirc(dp, skb, key, a, rem);
926 if (nla_is_last(a, rem)) {
927 /* If this is the last action, the skb has
928 * been consumed or freed.
929 * Return immediately.
935 case OVS_ACTION_ATTR_SET:
936 err = execute_set_action(skb, key, nla_data(a));
939 case OVS_ACTION_ATTR_SET_MASKED:
940 case OVS_ACTION_ATTR_SET_TO_MASKED:
941 err = execute_masked_set_action(skb, key, nla_data(a));
944 case OVS_ACTION_ATTR_SAMPLE:
945 err = sample(dp, skb, key, a, attr, len);
956 do_output(dp, skb, prev_port);
963 static void process_deferred_actions(struct datapath *dp)
965 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
967 /* Do not touch the FIFO in case there is no deferred actions. */
968 if (action_fifo_is_empty(fifo))
971 /* Finishing executing all deferred actions. */
973 struct deferred_action *da = action_fifo_get(fifo);
974 struct sk_buff *skb = da->skb;
975 struct sw_flow_key *key = &da->pkt_key;
976 const struct nlattr *actions = da->actions;
979 do_execute_actions(dp, skb, key, actions,
982 ovs_dp_process_packet(skb, key);
983 } while (!action_fifo_is_empty(fifo));
985 /* Reset FIFO for the next packet. */
986 action_fifo_init(fifo);
989 /* Execute a list of actions against 'skb'. */
990 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
991 const struct sw_flow_actions *acts,
992 struct sw_flow_key *key)
994 int level = this_cpu_read(exec_actions_level);
997 this_cpu_inc(exec_actions_level);
998 OVS_CB(skb)->egress_tun_info = NULL;
999 err = do_execute_actions(dp, skb, key,
1000 acts->actions, acts->actions_len);
1003 process_deferred_actions(dp);
1005 this_cpu_dec(exec_actions_level);
1009 int action_fifos_init(void)
1011 action_fifos = alloc_percpu(struct action_fifo);
1018 void action_fifos_exit(void)
1020 free_percpu(action_fifos);