2 * Copyright (c) 2007-2015 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>
44 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
45 struct sw_flow_key *key,
46 const struct nlattr *attr, int len);
48 struct deferred_action {
50 const struct nlattr *actions;
52 /* Store pkt_key clone when creating deferred action. */
53 struct sw_flow_key pkt_key;
56 #define DEFERRED_ACTION_FIFO_SIZE 10
60 /* Deferred action fifo queue storage. */
61 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
64 static struct action_fifo __percpu *action_fifos;
65 #define EXEC_ACTIONS_LEVEL_LIMIT 4 /* limit used to detect packet
66 * looping by the network stack
68 static DEFINE_PER_CPU(int, exec_actions_level);
70 static void action_fifo_init(struct action_fifo *fifo)
76 static bool action_fifo_is_empty(const struct action_fifo *fifo)
78 return (fifo->head == fifo->tail);
81 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
83 if (action_fifo_is_empty(fifo))
86 return &fifo->fifo[fifo->tail++];
89 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
91 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
94 return &fifo->fifo[fifo->head++];
97 /* Return queue entry if fifo is not full */
98 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
99 const struct sw_flow_key *key,
100 const struct nlattr *attr)
102 struct action_fifo *fifo;
103 struct deferred_action *da;
105 fifo = this_cpu_ptr(action_fifos);
106 da = action_fifo_put(fifo);
116 static void invalidate_flow_key(struct sw_flow_key *key)
118 key->eth.type = htons(0);
121 static bool is_flow_key_valid(const struct sw_flow_key *key)
123 return !!key->eth.type;
126 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
127 const struct ovs_action_push_mpls *mpls)
129 __be32 *new_mpls_lse;
132 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
133 if (skb_encapsulation(skb))
136 if (skb_cow_head(skb, MPLS_HLEN) < 0)
139 skb_push(skb, MPLS_HLEN);
140 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
142 skb_reset_mac_header(skb);
144 new_mpls_lse = (__be32 *)skb_mpls_header(skb);
145 *new_mpls_lse = mpls->mpls_lse;
147 if (skb->ip_summed == CHECKSUM_COMPLETE)
148 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
152 hdr->h_proto = mpls->mpls_ethertype;
153 if (!ovs_skb_get_inner_protocol(skb))
154 ovs_skb_set_inner_protocol(skb, skb->protocol);
155 skb->protocol = mpls->mpls_ethertype;
157 invalidate_flow_key(key);
161 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
162 const __be16 ethertype)
167 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
171 skb_postpull_rcsum(skb, skb_mpls_header(skb), MPLS_HLEN);
173 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
176 __skb_pull(skb, MPLS_HLEN);
177 skb_reset_mac_header(skb);
179 /* skb_mpls_header() is used to locate the ethertype
180 * field correctly in the presence of VLAN tags.
182 hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
183 hdr->h_proto = ethertype;
184 if (eth_p_mpls(skb->protocol))
185 skb->protocol = ethertype;
187 invalidate_flow_key(key);
191 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
192 const __be32 *mpls_lse, const __be32 *mask)
198 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
202 stack = (__be32 *)skb_mpls_header(skb);
203 lse = OVS_MASKED(*stack, *mpls_lse, *mask);
204 if (skb->ip_summed == CHECKSUM_COMPLETE) {
205 __be32 diff[] = { ~(*stack), lse };
207 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
212 flow_key->mpls.top_lse = lse;
216 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
220 err = skb_vlan_pop(skb);
221 if (skb_vlan_tag_present(skb))
222 invalidate_flow_key(key);
228 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
229 const struct ovs_action_push_vlan *vlan)
231 if (skb_vlan_tag_present(skb))
232 invalidate_flow_key(key);
234 key->eth.tci = vlan->vlan_tci;
235 return skb_vlan_push(skb, vlan->vlan_tpid,
236 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
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 ovs_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 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
276 __be32 addr, __be32 new_addr)
278 int transport_len = skb->len - skb_transport_offset(skb);
280 if (nh->frag_off & htons(IP_OFFSET))
283 if (nh->protocol == IPPROTO_TCP) {
284 if (likely(transport_len >= sizeof(struct tcphdr)))
285 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
287 } else if (nh->protocol == IPPROTO_UDP) {
288 if (likely(transport_len >= sizeof(struct udphdr))) {
289 struct udphdr *uh = udp_hdr(skb);
291 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
292 inet_proto_csum_replace4(&uh->check, skb,
295 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] = OVS_MASKED(old[0], addr[0], mask[0]);
342 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
343 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
344 masked[3] = OVS_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 (likely(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 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
362 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
363 OVS_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 = OVS_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 = OVS_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 = OVS_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 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
493 flow_key->ip.ttl = nh->hop_limit;
498 /* Must follow skb_ensure_writable() since that can move the skb data. */
499 static void set_tp_port(struct sk_buff *skb, __be16 *port,
500 __be16 new_port, __sum16 *check)
502 inet_proto_csum_replace2(check, skb, *port, new_port, 0);
506 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
507 const struct ovs_key_udp *key,
508 const struct ovs_key_udp *mask)
514 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
515 sizeof(struct udphdr));
520 /* Either of the masks is non-zero, so do not bother checking them. */
521 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
522 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
524 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
525 if (likely(src != uh->source)) {
526 set_tp_port(skb, &uh->source, src, &uh->check);
527 flow_key->tp.src = src;
529 if (likely(dst != uh->dest)) {
530 set_tp_port(skb, &uh->dest, dst, &uh->check);
531 flow_key->tp.dst = dst;
534 if (unlikely(!uh->check))
535 uh->check = CSUM_MANGLED_0;
539 flow_key->tp.src = src;
540 flow_key->tp.dst = dst;
548 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
549 const struct ovs_key_tcp *key,
550 const struct ovs_key_tcp *mask)
556 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
557 sizeof(struct tcphdr));
562 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
563 if (likely(src != th->source)) {
564 set_tp_port(skb, &th->source, src, &th->check);
565 flow_key->tp.src = src;
567 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
568 if (likely(dst != th->dest)) {
569 set_tp_port(skb, &th->dest, dst, &th->check);
570 flow_key->tp.dst = dst;
577 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
578 const struct ovs_key_sctp *key,
579 const struct ovs_key_sctp *mask)
581 unsigned int sctphoff = skb_transport_offset(skb);
583 __le32 old_correct_csum, new_csum, old_csum;
586 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
591 old_csum = sh->checksum;
592 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
594 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
595 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
597 new_csum = sctp_compute_cksum(skb, sctphoff);
599 /* Carry any checksum errors through. */
600 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
603 flow_key->tp.src = sh->source;
604 flow_key->tp.dst = sh->dest;
609 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
611 struct vport *vport = ovs_vport_rcu(dp, out_port);
614 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 upcall.egress_tun_info = &info;
650 err = ovs_vport_get_egress_tun_info(vport, skb,
653 upcall.egress_tun_info = NULL;
659 case OVS_USERSPACE_ATTR_ACTIONS: {
660 /* Include actions. */
661 upcall.actions = actions;
662 upcall.actions_len = actions_len;
666 } /* End of switch. */
669 return ovs_dp_upcall(dp, skb, key, &upcall);
672 static int sample(struct datapath *dp, struct sk_buff *skb,
673 struct sw_flow_key *key, const struct nlattr *attr,
674 const struct nlattr *actions, int actions_len)
676 const struct nlattr *acts_list = NULL;
677 const struct nlattr *a;
680 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
681 a = nla_next(a, &rem)) {
684 switch (nla_type(a)) {
685 case OVS_SAMPLE_ATTR_PROBABILITY:
686 probability = nla_get_u32(a);
687 if (!probability || prandom_u32() > probability)
691 case OVS_SAMPLE_ATTR_ACTIONS:
697 rem = nla_len(acts_list);
698 a = nla_data(acts_list);
700 /* Actions list is empty, do nothing */
704 /* The only known usage of sample action is having a single user-space
705 * action. Treat this usage as a special case.
706 * The output_userspace() should clone the skb to be sent to the
707 * user space. This skb will be consumed by its caller.
709 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
710 nla_is_last(a, rem)))
711 return output_userspace(dp, skb, key, a, actions, actions_len);
713 skb = skb_clone(skb, GFP_ATOMIC);
715 /* Skip the sample action when out of memory. */
718 if (!add_deferred_actions(skb, key, a)) {
720 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
728 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
729 const struct nlattr *attr)
731 struct ovs_action_hash *hash_act = nla_data(attr);
734 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
735 hash = skb_get_hash(skb);
736 hash = jhash_1word(hash, hash_act->hash_basis);
740 key->ovs_flow_hash = hash;
743 static int execute_set_action(struct sk_buff *skb,
744 struct sw_flow_key *flow_key,
745 const struct nlattr *a)
747 /* Only tunnel set execution is supported without a mask. */
748 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
749 struct ovs_tunnel_info *tun = nla_data(a);
751 ovs_skb_dst_drop(skb);
752 ovs_dst_hold((struct dst_entry *)tun->tun_dst);
753 ovs_skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
760 /* Mask is at the midpoint of the data. */
761 #define get_mask(a, type) ((const type)nla_data(a) + 1)
763 static int execute_masked_set_action(struct sk_buff *skb,
764 struct sw_flow_key *flow_key,
765 const struct nlattr *a)
769 switch (nla_type(a)) {
770 case OVS_KEY_ATTR_PRIORITY:
771 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
772 *get_mask(a, u32 *));
773 flow_key->phy.priority = skb->priority;
776 case OVS_KEY_ATTR_SKB_MARK:
777 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
778 flow_key->phy.skb_mark = skb->mark;
781 case OVS_KEY_ATTR_TUNNEL_INFO:
782 /* Masked data not supported for tunnel. */
786 case OVS_KEY_ATTR_ETHERNET:
787 err = set_eth_addr(skb, flow_key, nla_data(a),
788 get_mask(a, struct ovs_key_ethernet *));
791 case OVS_KEY_ATTR_IPV4:
792 err = set_ipv4(skb, flow_key, nla_data(a),
793 get_mask(a, struct ovs_key_ipv4 *));
796 case OVS_KEY_ATTR_IPV6:
797 err = set_ipv6(skb, flow_key, nla_data(a),
798 get_mask(a, struct ovs_key_ipv6 *));
801 case OVS_KEY_ATTR_TCP:
802 err = set_tcp(skb, flow_key, nla_data(a),
803 get_mask(a, struct ovs_key_tcp *));
806 case OVS_KEY_ATTR_UDP:
807 err = set_udp(skb, flow_key, nla_data(a),
808 get_mask(a, struct ovs_key_udp *));
811 case OVS_KEY_ATTR_SCTP:
812 err = set_sctp(skb, flow_key, nla_data(a),
813 get_mask(a, struct ovs_key_sctp *));
816 case OVS_KEY_ATTR_MPLS:
817 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
825 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
826 struct sw_flow_key *key,
827 const struct nlattr *a, int rem)
829 struct deferred_action *da;
831 if (!is_flow_key_valid(key)) {
834 err = ovs_flow_key_update(skb, key);
838 BUG_ON(!is_flow_key_valid(key));
840 if (!nla_is_last(a, rem)) {
841 /* Recirc action is the not the last action
842 * of the action list, need to clone the skb.
844 skb = skb_clone(skb, GFP_ATOMIC);
846 /* Skip the recirc action when out of memory, but
847 * continue on with the rest of the action list.
853 da = add_deferred_actions(skb, key, NULL);
855 da->pkt_key.recirc_id = nla_get_u32(a);
860 pr_warn("%s: deferred action limit reached, drop recirc action\n",
867 /* Execute a list of actions against 'skb'. */
868 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
869 struct sw_flow_key *key,
870 const struct nlattr *attr, int len)
872 /* Every output action needs a separate clone of 'skb', but the common
873 * case is just a single output action, so that doing a clone and
874 * then freeing the original skbuff is wasteful. So the following code
875 * is slightly obscure just to avoid that.
878 const struct nlattr *a;
881 for (a = attr, rem = len; rem > 0;
882 a = nla_next(a, &rem)) {
885 if (unlikely(prev_port != -1)) {
886 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
889 do_output(dp, out_skb, prev_port);
894 switch (nla_type(a)) {
895 case OVS_ACTION_ATTR_OUTPUT:
896 prev_port = nla_get_u32(a);
899 case OVS_ACTION_ATTR_USERSPACE:
900 output_userspace(dp, skb, key, a, attr, len);
903 case OVS_ACTION_ATTR_HASH:
904 execute_hash(skb, key, a);
907 case OVS_ACTION_ATTR_PUSH_MPLS:
908 err = push_mpls(skb, key, nla_data(a));
911 case OVS_ACTION_ATTR_POP_MPLS:
912 err = pop_mpls(skb, key, nla_get_be16(a));
915 case OVS_ACTION_ATTR_PUSH_VLAN:
916 err = push_vlan(skb, key, nla_data(a));
919 case OVS_ACTION_ATTR_POP_VLAN:
920 err = pop_vlan(skb, key);
923 case OVS_ACTION_ATTR_RECIRC:
924 err = execute_recirc(dp, skb, key, a, rem);
925 if (nla_is_last(a, rem)) {
926 /* If this is the last action, the skb has
927 * been consumed or freed.
928 * Return immediately.
934 case OVS_ACTION_ATTR_SET:
935 err = execute_set_action(skb, key, nla_data(a));
938 case OVS_ACTION_ATTR_SET_MASKED:
939 case OVS_ACTION_ATTR_SET_TO_MASKED:
940 err = execute_masked_set_action(skb, key, nla_data(a));
943 case OVS_ACTION_ATTR_SAMPLE:
944 err = sample(dp, skb, key, a, attr, len);
955 do_output(dp, skb, prev_port);
962 static void process_deferred_actions(struct datapath *dp)
964 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
966 /* Do not touch the FIFO in case there is no deferred actions. */
967 if (action_fifo_is_empty(fifo))
970 /* Finishing executing all deferred actions. */
972 struct deferred_action *da = action_fifo_get(fifo);
973 struct sk_buff *skb = da->skb;
974 struct sw_flow_key *key = &da->pkt_key;
975 const struct nlattr *actions = da->actions;
978 do_execute_actions(dp, skb, key, actions,
981 ovs_dp_process_packet(skb, key);
982 } while (!action_fifo_is_empty(fifo));
984 /* Reset FIFO for the next packet. */
985 action_fifo_init(fifo);
988 /* Execute a list of actions against 'skb'. */
989 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
990 const struct sw_flow_actions *acts,
991 struct sw_flow_key *key)
993 int level = this_cpu_read(exec_actions_level);
996 if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
998 pr_warn("%s: packet loop detected, dropping.\n",
1005 this_cpu_inc(exec_actions_level);
1006 err = do_execute_actions(dp, skb, key,
1007 acts->actions, acts->actions_len);
1010 process_deferred_actions(dp);
1012 this_cpu_dec(exec_actions_level);
1014 /* This return status currently does not reflect the errors
1015 * encounted during deferred actions execution. Probably needs to
1016 * be fixed in the future.
1021 int action_fifos_init(void)
1023 action_fifos = alloc_percpu(struct action_fifo);
1030 void action_fifos_exit(void)
1032 free_percpu(action_fifos);