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>
33 #include <net/checksum.h>
34 #include <net/dsfield.h>
36 #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 #define EXEC_ACTIONS_LEVEL_LIMIT 4 /* limit used to detect packet
65 * looping by the network stack
67 static DEFINE_PER_CPU(int, exec_actions_level);
69 static void action_fifo_init(struct action_fifo *fifo)
75 static bool action_fifo_is_empty(const struct action_fifo *fifo)
77 return (fifo->head == fifo->tail);
80 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
82 if (action_fifo_is_empty(fifo))
85 return &fifo->fifo[fifo->tail++];
88 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
90 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
93 return &fifo->fifo[fifo->head++];
96 /* Return queue entry if fifo is not full */
97 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
98 const struct sw_flow_key *key,
99 const struct nlattr *attr)
101 struct action_fifo *fifo;
102 struct deferred_action *da;
104 fifo = this_cpu_ptr(action_fifos);
105 da = action_fifo_put(fifo);
115 static void invalidate_flow_key(struct sw_flow_key *key)
117 key->eth.type = htons(0);
120 static bool is_flow_key_valid(const struct sw_flow_key *key)
122 return !!key->eth.type;
125 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
126 const struct ovs_action_push_mpls *mpls)
128 __be32 *new_mpls_lse;
131 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
132 if (skb_encapsulation(skb))
135 if (skb_cow_head(skb, MPLS_HLEN) < 0)
138 skb_push(skb, MPLS_HLEN);
139 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
141 skb_reset_mac_header(skb);
143 new_mpls_lse = (__be32 *)skb_mpls_header(skb);
144 *new_mpls_lse = mpls->mpls_lse;
146 if (skb->ip_summed == CHECKSUM_COMPLETE)
147 skb->csum = csum_add(skb->csum, csum_partial(new_mpls_lse,
151 hdr->h_proto = mpls->mpls_ethertype;
152 if (!ovs_skb_get_inner_protocol(skb))
153 ovs_skb_set_inner_protocol(skb, skb->protocol);
154 skb->protocol = mpls->mpls_ethertype;
156 invalidate_flow_key(key);
160 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
161 const __be16 ethertype)
166 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
170 if (skb->ip_summed == CHECKSUM_COMPLETE)
171 skb->csum = csum_sub(skb->csum,
172 csum_partial(skb_mpls_header(skb),
175 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
178 __skb_pull(skb, MPLS_HLEN);
179 skb_reset_mac_header(skb);
181 /* skb_mpls_header() is used to locate the ethertype
182 * field correctly in the presence of VLAN tags.
184 hdr = (struct ethhdr *)(skb_mpls_header(skb) - ETH_HLEN);
185 hdr->h_proto = ethertype;
186 if (eth_p_mpls(skb->protocol))
187 skb->protocol = ethertype;
189 invalidate_flow_key(key);
193 /* 'KEY' must not have any bits set outside of the 'MASK' */
194 #define MASKED(OLD, KEY, MASK) ((KEY) | ((OLD) & ~(MASK)))
195 #define SET_MASKED(OLD, KEY, MASK) ((OLD) = MASKED(OLD, KEY, MASK))
197 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
198 const __be32 *mpls_lse, const __be32 *mask)
204 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
208 stack = (__be32 *)skb_mpls_header(skb);
209 lse = MASKED(*stack, *mpls_lse, *mask);
210 if (skb->ip_summed == CHECKSUM_COMPLETE) {
211 __be32 diff[] = { ~(*stack), lse };
213 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
218 flow_key->mpls.top_lse = lse;
222 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
226 err = skb_vlan_pop(skb);
227 if (skb_vlan_tag_present(skb))
228 invalidate_flow_key(key);
235 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
236 const struct ovs_action_push_vlan *vlan)
238 if (skb_vlan_tag_present(skb))
239 invalidate_flow_key(key);
241 key->eth.tci = vlan->vlan_tci;
243 return skb_vlan_push(skb, vlan->vlan_tpid,
244 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
247 /* 'src' is already properly masked. */
248 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
250 u16 *dst = (u16 *)dst_;
251 const u16 *src = (const u16 *)src_;
252 const u16 *mask = (const u16 *)mask_;
254 SET_MASKED(dst[0], src[0], mask[0]);
255 SET_MASKED(dst[1], src[1], mask[1]);
256 SET_MASKED(dst[2], src[2], mask[2]);
259 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
260 const struct ovs_key_ethernet *key,
261 const struct ovs_key_ethernet *mask)
265 err = skb_ensure_writable(skb, ETH_HLEN);
269 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
271 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
273 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
276 ovs_skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
278 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
279 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
283 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
284 __be32 addr, __be32 new_addr)
286 int transport_len = skb->len - skb_transport_offset(skb);
288 if (nh->frag_off & htons(IP_OFFSET))
291 if (nh->protocol == IPPROTO_TCP) {
292 if (likely(transport_len >= sizeof(struct tcphdr)))
293 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
295 } else if (nh->protocol == IPPROTO_UDP) {
296 if (likely(transport_len >= sizeof(struct udphdr))) {
297 struct udphdr *uh = udp_hdr(skb);
299 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
300 inet_proto_csum_replace4(&uh->check, skb,
303 uh->check = CSUM_MANGLED_0;
310 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
311 __be32 *addr, __be32 new_addr)
313 update_ip_l4_checksum(skb, nh, *addr, new_addr);
314 csum_replace4(&nh->check, *addr, new_addr);
319 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
320 __be32 addr[4], const __be32 new_addr[4])
322 int transport_len = skb->len - skb_transport_offset(skb);
324 if (l4_proto == NEXTHDR_TCP) {
325 if (likely(transport_len >= sizeof(struct tcphdr)))
326 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
328 } else if (l4_proto == NEXTHDR_UDP) {
329 if (likely(transport_len >= sizeof(struct udphdr))) {
330 struct udphdr *uh = udp_hdr(skb);
332 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
333 inet_proto_csum_replace16(&uh->check, skb,
336 uh->check = CSUM_MANGLED_0;
339 } else if (l4_proto == NEXTHDR_ICMP) {
340 if (likely(transport_len >= sizeof(struct icmp6hdr)))
341 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
342 skb, addr, new_addr, 1);
346 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
347 const __be32 mask[4], __be32 masked[4])
349 masked[0] = MASKED(old[0], addr[0], mask[0]);
350 masked[1] = MASKED(old[1], addr[1], mask[1]);
351 masked[2] = MASKED(old[2], addr[2], mask[2]);
352 masked[3] = MASKED(old[3], addr[3], mask[3]);
355 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
356 __be32 addr[4], const __be32 new_addr[4],
357 bool recalculate_csum)
359 if (likely(recalculate_csum))
360 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
363 memcpy(addr, new_addr, sizeof(__be32[4]));
366 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
368 /* Bits 21-24 are always unmasked, so this retains their values. */
369 SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
370 SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
371 SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
374 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
377 new_ttl = MASKED(nh->ttl, new_ttl, mask);
379 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
383 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
384 const struct ovs_key_ipv4 *key,
385 const struct ovs_key_ipv4 *mask)
391 err = skb_ensure_writable(skb, skb_network_offset(skb) +
392 sizeof(struct iphdr));
398 /* Setting an IP addresses is typically only a side effect of
399 * matching on them in the current userspace implementation, so it
400 * makes sense to check if the value actually changed.
402 if (mask->ipv4_src) {
403 new_addr = MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
405 if (unlikely(new_addr != nh->saddr)) {
406 set_ip_addr(skb, nh, &nh->saddr, new_addr);
407 flow_key->ipv4.addr.src = new_addr;
410 if (mask->ipv4_dst) {
411 new_addr = MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
413 if (unlikely(new_addr != nh->daddr)) {
414 set_ip_addr(skb, nh, &nh->daddr, new_addr);
415 flow_key->ipv4.addr.dst = new_addr;
418 if (mask->ipv4_tos) {
419 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
420 flow_key->ip.tos = nh->tos;
422 if (mask->ipv4_ttl) {
423 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
424 flow_key->ip.ttl = nh->ttl;
430 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
432 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
435 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
436 const struct ovs_key_ipv6 *key,
437 const struct ovs_key_ipv6 *mask)
442 err = skb_ensure_writable(skb, skb_network_offset(skb) +
443 sizeof(struct ipv6hdr));
449 /* Setting an IP addresses is typically only a side effect of
450 * matching on them in the current userspace implementation, so it
451 * makes sense to check if the value actually changed.
453 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
454 __be32 *saddr = (__be32 *)&nh->saddr;
457 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
459 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
460 set_ipv6_addr(skb, key->ipv6_proto, saddr, masked,
462 memcpy(&flow_key->ipv6.addr.src, masked,
463 sizeof(flow_key->ipv6.addr.src));
466 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
467 unsigned int offset = 0;
468 int flags = IP6_FH_F_SKIP_RH;
469 bool recalc_csum = true;
470 __be32 *daddr = (__be32 *)&nh->daddr;
473 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
475 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
476 if (ipv6_ext_hdr(nh->nexthdr))
477 recalc_csum = (ipv6_find_hdr(skb, &offset,
482 set_ipv6_addr(skb, key->ipv6_proto, daddr, masked,
484 memcpy(&flow_key->ipv6.addr.dst, masked,
485 sizeof(flow_key->ipv6.addr.dst));
488 if (mask->ipv6_tclass) {
489 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
490 flow_key->ip.tos = ipv6_get_dsfield(nh);
492 if (mask->ipv6_label) {
493 set_ipv6_fl(nh, ntohl(key->ipv6_label),
494 ntohl(mask->ipv6_label));
495 flow_key->ipv6.label =
496 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
498 if (mask->ipv6_hlimit) {
499 SET_MASKED(nh->hop_limit, key->ipv6_hlimit, mask->ipv6_hlimit);
500 flow_key->ip.ttl = nh->hop_limit;
505 /* Must follow skb_ensure_writable() since that can move the skb data. */
506 static void set_tp_port(struct sk_buff *skb, __be16 *port,
507 __be16 new_port, __sum16 *check)
509 inet_proto_csum_replace2(check, skb, *port, new_port, 0);
513 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
514 const struct ovs_key_udp *key,
515 const struct ovs_key_udp *mask)
521 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
522 sizeof(struct udphdr));
527 /* Either of the masks is non-zero, so do not bother checking them. */
528 src = MASKED(uh->source, key->udp_src, mask->udp_src);
529 dst = MASKED(uh->dest, key->udp_dst, mask->udp_dst);
531 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
532 if (likely(src != uh->source)) {
533 set_tp_port(skb, &uh->source, src, &uh->check);
534 flow_key->tp.src = src;
536 if (likely(dst != uh->dest)) {
537 set_tp_port(skb, &uh->dest, dst, &uh->check);
538 flow_key->tp.dst = dst;
541 if (unlikely(!uh->check))
542 uh->check = CSUM_MANGLED_0;
546 flow_key->tp.src = src;
547 flow_key->tp.dst = dst;
555 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
556 const struct ovs_key_tcp *key,
557 const struct ovs_key_tcp *mask)
563 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
564 sizeof(struct tcphdr));
570 src = MASKED(th->source, key->tcp_src, mask->tcp_src);
571 if (likely(src != th->source)) {
572 set_tp_port(skb, &th->source, src, &th->check);
573 flow_key->tp.src = src;
575 dst = MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
576 if (likely(dst != th->dest)) {
577 set_tp_port(skb, &th->dest, dst, &th->check);
578 flow_key->tp.dst = dst;
585 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
586 const struct ovs_key_sctp *key,
587 const struct ovs_key_sctp *mask)
589 unsigned int sctphoff = skb_transport_offset(skb);
591 __le32 old_correct_csum, new_csum, old_csum;
594 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
600 old_csum = sh->checksum;
601 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
603 sh->source = MASKED(sh->source, key->sctp_src, mask->sctp_src);
604 sh->dest = MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
606 new_csum = sctp_compute_cksum(skb, sctphoff);
608 /* Carry any checksum errors through. */
609 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
612 flow_key->tp.src = sh->source;
613 flow_key->tp.dst = sh->dest;
618 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
620 struct vport *vport = ovs_vport_rcu(dp, out_port);
623 ovs_vport_send(vport, skb);
628 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
629 struct sw_flow_key *key, const struct nlattr *attr)
631 struct ovs_tunnel_info info;
632 struct dp_upcall_info upcall;
633 const struct nlattr *a;
636 upcall.cmd = OVS_PACKET_CMD_ACTION;
637 upcall.userdata = NULL;
639 upcall.egress_tun_info = NULL;
641 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
642 a = nla_next(a, &rem)) {
643 switch (nla_type(a)) {
644 case OVS_USERSPACE_ATTR_USERDATA:
648 case OVS_USERSPACE_ATTR_PID:
649 upcall.portid = nla_get_u32(a);
652 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
653 /* Get out tunnel info. */
656 vport = ovs_vport_rcu(dp, nla_get_u32(a));
660 err = ovs_vport_get_egress_tun_info(vport, skb,
663 upcall.egress_tun_info = &info;
668 } /* End of switch. */
671 return ovs_dp_upcall(dp, skb, key, &upcall);
674 static int sample(struct datapath *dp, struct sk_buff *skb,
675 struct sw_flow_key *key, const struct nlattr *attr)
677 const struct nlattr *acts_list = NULL;
678 const struct nlattr *a;
681 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
682 a = nla_next(a, &rem)) {
683 switch (nla_type(a)) {
684 case OVS_SAMPLE_ATTR_PROBABILITY:
685 if (prandom_u32() >= nla_get_u32(a))
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);
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 OVS_CB(skb)->egress_tun_info = nla_data(a);
755 /* Mask is at the midpoint of the data. */
756 #define get_mask(a, type) ((const type)nla_data(a) + 1)
758 static int execute_masked_set_action(struct sk_buff *skb,
759 struct sw_flow_key *flow_key,
760 const struct nlattr *a)
764 switch (nla_type(a)) {
765 case OVS_KEY_ATTR_PRIORITY:
766 SET_MASKED(skb->priority, nla_get_u32(a), *get_mask(a, u32 *));
767 flow_key->phy.priority = skb->priority;
770 case OVS_KEY_ATTR_SKB_MARK:
771 SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
772 flow_key->phy.skb_mark = skb->mark;
775 case OVS_KEY_ATTR_TUNNEL_INFO:
776 /* Masked data not supported for tunnel. */
780 case OVS_KEY_ATTR_ETHERNET:
781 err = set_eth_addr(skb, flow_key, nla_data(a),
782 get_mask(a, struct ovs_key_ethernet *));
785 case OVS_KEY_ATTR_IPV4:
786 err = set_ipv4(skb, flow_key, nla_data(a),
787 get_mask(a, struct ovs_key_ipv4 *));
790 case OVS_KEY_ATTR_IPV6:
791 err = set_ipv6(skb, flow_key, nla_data(a),
792 get_mask(a, struct ovs_key_ipv6 *));
795 case OVS_KEY_ATTR_TCP:
796 err = set_tcp(skb, flow_key, nla_data(a),
797 get_mask(a, struct ovs_key_tcp *));
800 case OVS_KEY_ATTR_UDP:
801 err = set_udp(skb, flow_key, nla_data(a),
802 get_mask(a, struct ovs_key_udp *));
805 case OVS_KEY_ATTR_SCTP:
806 err = set_sctp(skb, flow_key, nla_data(a),
807 get_mask(a, struct ovs_key_sctp *));
810 case OVS_KEY_ATTR_MPLS:
811 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
819 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
820 struct sw_flow_key *key,
821 const struct nlattr *a, int rem)
823 struct deferred_action *da;
825 if (!is_flow_key_valid(key)) {
828 err = ovs_flow_key_update(skb, key);
832 BUG_ON(!is_flow_key_valid(key));
834 if (!nla_is_last(a, rem)) {
835 /* Recirc action is the not the last action
836 * of the action list, need to clone the skb.
838 skb = skb_clone(skb, GFP_ATOMIC);
840 /* Skip the recirc action when out of memory, but
841 * continue on with the rest of the action list.
847 da = add_deferred_actions(skb, key, NULL);
849 da->pkt_key.recirc_id = nla_get_u32(a);
854 pr_warn("%s: deferred action limit reached, drop recirc action\n",
861 /* Execute a list of actions against 'skb'. */
862 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
863 struct sw_flow_key *key,
864 const struct nlattr *attr, int len)
866 /* Every output action needs a separate clone of 'skb', but the common
867 * case is just a single output action, so that doing a clone and
868 * then freeing the original skbuff is wasteful. So the following code
869 * is slightly obscure just to avoid that.
872 const struct nlattr *a;
875 for (a = attr, rem = len; rem > 0;
876 a = nla_next(a, &rem)) {
879 if (unlikely(prev_port != -1)) {
880 struct sk_buff *out_skb = skb_clone(skb, GFP_ATOMIC);
883 do_output(dp, out_skb, prev_port);
888 switch (nla_type(a)) {
889 case OVS_ACTION_ATTR_OUTPUT:
890 prev_port = nla_get_u32(a);
893 case OVS_ACTION_ATTR_USERSPACE:
894 output_userspace(dp, skb, key, a);
897 case OVS_ACTION_ATTR_HASH:
898 execute_hash(skb, key, a);
901 case OVS_ACTION_ATTR_PUSH_MPLS:
902 err = push_mpls(skb, key, nla_data(a));
905 case OVS_ACTION_ATTR_POP_MPLS:
906 err = pop_mpls(skb, key, nla_get_be16(a));
909 case OVS_ACTION_ATTR_PUSH_VLAN:
910 err = push_vlan(skb, key, nla_data(a));
913 case OVS_ACTION_ATTR_POP_VLAN:
914 err = pop_vlan(skb, key);
917 case OVS_ACTION_ATTR_RECIRC:
918 err = execute_recirc(dp, skb, key, a, rem);
919 if (nla_is_last(a, rem)) {
920 /* If this is the last action, the skb has
921 * been consumed or freed.
922 * Return immediately.
928 case OVS_ACTION_ATTR_SET:
929 err = execute_set_action(skb, key, nla_data(a));
932 case OVS_ACTION_ATTR_SET_MASKED:
933 case OVS_ACTION_ATTR_SET_TO_MASKED:
934 err = execute_masked_set_action(skb, key, nla_data(a));
937 case OVS_ACTION_ATTR_SAMPLE:
938 err = sample(dp, skb, key, a);
949 do_output(dp, skb, prev_port);
956 static void process_deferred_actions(struct datapath *dp)
958 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
960 /* Do not touch the FIFO in case there is no deferred actions. */
961 if (action_fifo_is_empty(fifo))
964 /* Finishing executing all deferred actions. */
966 struct deferred_action *da = action_fifo_get(fifo);
967 struct sk_buff *skb = da->skb;
968 struct sw_flow_key *key = &da->pkt_key;
969 const struct nlattr *actions = da->actions;
972 do_execute_actions(dp, skb, key, actions,
975 ovs_dp_process_packet(skb, key);
976 } while (!action_fifo_is_empty(fifo));
978 /* Reset FIFO for the next packet. */
979 action_fifo_init(fifo);
982 /* Execute a list of actions against 'skb'. */
983 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
984 const struct sw_flow_actions *acts,
985 struct sw_flow_key *key)
987 int level = this_cpu_read(exec_actions_level);
990 if (unlikely(level >= EXEC_ACTIONS_LEVEL_LIMIT)) {
992 pr_warn("%s: packet loop detected, dropping.\n",
999 this_cpu_inc(exec_actions_level);
1000 err = do_execute_actions(dp, skb, key,
1001 acts->actions, acts->actions_len);
1004 process_deferred_actions(dp);
1006 this_cpu_dec(exec_actions_level);
1008 /* This return status currently does not reflect the errors
1009 * encounted during deferred actions execution. Probably needs to
1010 * be fixed in the future.
1015 int action_fifos_init(void)
1017 action_fifos = alloc_percpu(struct action_fifo);
1024 void action_fifos_exit(void)
1026 free_percpu(action_fifos);