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;
65 u16 network_offset; /* valid only for MPLS */
70 u8 l2_data[MAX_L2_LEN];
73 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
75 #define DEFERRED_ACTION_FIFO_SIZE 10
76 #define OVS_RECURSION_LIMIT 5
77 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
81 /* Deferred action fifo queue storage. */
82 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
86 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
89 static struct action_fifo __percpu *action_fifos;
90 static struct recirc_keys __percpu *recirc_keys;
91 static DEFINE_PER_CPU(int, exec_actions_level);
93 static void action_fifo_init(struct action_fifo *fifo)
99 static bool action_fifo_is_empty(const struct action_fifo *fifo)
101 return (fifo->head == fifo->tail);
104 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
106 if (action_fifo_is_empty(fifo))
109 return &fifo->fifo[fifo->tail++];
112 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
114 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
117 return &fifo->fifo[fifo->head++];
120 /* Return true if fifo is not full */
121 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
122 const struct sw_flow_key *key,
123 const struct nlattr *attr)
125 struct action_fifo *fifo;
126 struct deferred_action *da;
128 fifo = this_cpu_ptr(action_fifos);
129 da = action_fifo_put(fifo);
139 static void invalidate_flow_key(struct sw_flow_key *key)
141 key->mac_proto |= SW_FLOW_KEY_INVALID;
144 static bool is_flow_key_valid(const struct sw_flow_key *key)
146 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
149 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
152 if (skb->ip_summed == CHECKSUM_COMPLETE) {
153 __be16 diff[] = { ~(hdr->h_proto), ethertype };
155 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
159 hdr->h_proto = ethertype;
162 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
163 const struct ovs_action_push_mpls *mpls)
165 struct mpls_shim_hdr *new_mpls_lse;
167 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
168 if (skb->encapsulation)
171 if (skb_cow_head(skb, MPLS_HLEN) < 0)
174 if (!skb->inner_protocol) {
175 skb_set_inner_network_header(skb, skb->mac_len);
176 skb_set_inner_protocol(skb, skb->protocol);
179 skb_push(skb, MPLS_HLEN);
180 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
182 skb_reset_mac_header(skb);
183 skb_set_network_header(skb, skb->mac_len);
185 new_mpls_lse = mpls_hdr(skb);
186 new_mpls_lse->label_stack_entry = mpls->mpls_lse;
188 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
190 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
191 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
192 skb->protocol = mpls->mpls_ethertype;
194 invalidate_flow_key(key);
198 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
199 const __be16 ethertype)
203 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
207 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
209 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
212 __skb_pull(skb, MPLS_HLEN);
213 skb_reset_mac_header(skb);
214 skb_set_network_header(skb, skb->mac_len);
216 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
219 /* mpls_hdr() is used to locate the ethertype field correctly in the
220 * presence of VLAN tags.
222 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
223 update_ethertype(skb, hdr, ethertype);
225 if (eth_p_mpls(skb->protocol))
226 skb->protocol = ethertype;
228 invalidate_flow_key(key);
232 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
233 const __be32 *mpls_lse, const __be32 *mask)
235 struct mpls_shim_hdr *stack;
239 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
243 stack = mpls_hdr(skb);
244 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
245 if (skb->ip_summed == CHECKSUM_COMPLETE) {
246 __be32 diff[] = { ~(stack->label_stack_entry), lse };
248 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
252 stack->label_stack_entry = lse;
253 flow_key->mpls.top_lse = lse;
257 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
261 err = skb_vlan_pop(skb);
262 if (skb_vlan_tag_present(skb)) {
263 invalidate_flow_key(key);
265 key->eth.vlan.tci = 0;
266 key->eth.vlan.tpid = 0;
271 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
272 const struct ovs_action_push_vlan *vlan)
274 if (skb_vlan_tag_present(skb)) {
275 invalidate_flow_key(key);
277 key->eth.vlan.tci = vlan->vlan_tci;
278 key->eth.vlan.tpid = vlan->vlan_tpid;
280 return skb_vlan_push(skb, vlan->vlan_tpid,
281 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
284 /* 'src' is already properly masked. */
285 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
287 u16 *dst = (u16 *)dst_;
288 const u16 *src = (const u16 *)src_;
289 const u16 *mask = (const u16 *)mask_;
291 OVS_SET_MASKED(dst[0], src[0], mask[0]);
292 OVS_SET_MASKED(dst[1], src[1], mask[1]);
293 OVS_SET_MASKED(dst[2], src[2], mask[2]);
296 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
297 const struct ovs_key_ethernet *key,
298 const struct ovs_key_ethernet *mask)
302 err = skb_ensure_writable(skb, ETH_HLEN);
306 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
308 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
310 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
313 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
315 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
316 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
320 /* pop_eth does not support VLAN packets as this action is never called
323 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
325 skb_pull_rcsum(skb, ETH_HLEN);
326 skb_reset_mac_header(skb);
327 skb_reset_mac_len(skb);
329 /* safe right before invalidate_flow_key */
330 key->mac_proto = MAC_PROTO_NONE;
331 invalidate_flow_key(key);
335 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
336 const struct ovs_action_push_eth *ethh)
340 /* Add the new Ethernet header */
341 if (skb_cow_head(skb, ETH_HLEN) < 0)
344 skb_push(skb, ETH_HLEN);
345 skb_reset_mac_header(skb);
346 skb_reset_mac_len(skb);
349 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
350 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
351 hdr->h_proto = skb->protocol;
353 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
355 /* safe right before invalidate_flow_key */
356 key->mac_proto = MAC_PROTO_ETHERNET;
357 invalidate_flow_key(key);
361 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
362 __be32 addr, __be32 new_addr)
364 int transport_len = skb->len - skb_transport_offset(skb);
366 if (nh->frag_off & htons(IP_OFFSET))
369 if (nh->protocol == IPPROTO_TCP) {
370 if (likely(transport_len >= sizeof(struct tcphdr)))
371 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
372 addr, new_addr, true);
373 } else if (nh->protocol == IPPROTO_UDP) {
374 if (likely(transport_len >= sizeof(struct udphdr))) {
375 struct udphdr *uh = udp_hdr(skb);
377 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
378 inet_proto_csum_replace4(&uh->check, skb,
379 addr, new_addr, true);
381 uh->check = CSUM_MANGLED_0;
387 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
388 __be32 *addr, __be32 new_addr)
390 update_ip_l4_checksum(skb, nh, *addr, new_addr);
391 csum_replace4(&nh->check, *addr, new_addr);
396 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
397 __be32 addr[4], const __be32 new_addr[4])
399 int transport_len = skb->len - skb_transport_offset(skb);
401 if (l4_proto == NEXTHDR_TCP) {
402 if (likely(transport_len >= sizeof(struct tcphdr)))
403 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
404 addr, new_addr, true);
405 } else if (l4_proto == NEXTHDR_UDP) {
406 if (likely(transport_len >= sizeof(struct udphdr))) {
407 struct udphdr *uh = udp_hdr(skb);
409 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
410 inet_proto_csum_replace16(&uh->check, skb,
411 addr, new_addr, true);
413 uh->check = CSUM_MANGLED_0;
416 } else if (l4_proto == NEXTHDR_ICMP) {
417 if (likely(transport_len >= sizeof(struct icmp6hdr)))
418 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
419 skb, addr, new_addr, true);
423 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
424 const __be32 mask[4], __be32 masked[4])
426 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
427 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
428 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
429 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
432 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
433 __be32 addr[4], const __be32 new_addr[4],
434 bool recalculate_csum)
436 if (recalculate_csum)
437 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
440 memcpy(addr, new_addr, sizeof(__be32[4]));
443 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
445 /* Bits 21-24 are always unmasked, so this retains their values. */
446 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
447 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
448 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
451 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
454 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
456 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
460 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
461 const struct ovs_key_ipv4 *key,
462 const struct ovs_key_ipv4 *mask)
468 err = skb_ensure_writable(skb, skb_network_offset(skb) +
469 sizeof(struct iphdr));
475 /* Setting an IP addresses is typically only a side effect of
476 * matching on them in the current userspace implementation, so it
477 * makes sense to check if the value actually changed.
479 if (mask->ipv4_src) {
480 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
482 if (unlikely(new_addr != nh->saddr)) {
483 set_ip_addr(skb, nh, &nh->saddr, new_addr);
484 flow_key->ipv4.addr.src = new_addr;
487 if (mask->ipv4_dst) {
488 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
490 if (unlikely(new_addr != nh->daddr)) {
491 set_ip_addr(skb, nh, &nh->daddr, new_addr);
492 flow_key->ipv4.addr.dst = new_addr;
495 if (mask->ipv4_tos) {
496 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
497 flow_key->ip.tos = nh->tos;
499 if (mask->ipv4_ttl) {
500 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
501 flow_key->ip.ttl = nh->ttl;
507 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
509 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
512 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
513 const struct ovs_key_ipv6 *key,
514 const struct ovs_key_ipv6 *mask)
519 err = skb_ensure_writable(skb, skb_network_offset(skb) +
520 sizeof(struct ipv6hdr));
526 /* Setting an IP addresses is typically only a side effect of
527 * matching on them in the current userspace implementation, so it
528 * makes sense to check if the value actually changed.
530 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
531 __be32 *saddr = (__be32 *)&nh->saddr;
534 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
536 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
537 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
539 memcpy(&flow_key->ipv6.addr.src, masked,
540 sizeof(flow_key->ipv6.addr.src));
543 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
544 unsigned int offset = 0;
545 int flags = IP6_FH_F_SKIP_RH;
546 bool recalc_csum = true;
547 __be32 *daddr = (__be32 *)&nh->daddr;
550 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
552 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
553 if (ipv6_ext_hdr(nh->nexthdr))
554 recalc_csum = (ipv6_find_hdr(skb, &offset,
559 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
561 memcpy(&flow_key->ipv6.addr.dst, masked,
562 sizeof(flow_key->ipv6.addr.dst));
565 if (mask->ipv6_tclass) {
566 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
567 flow_key->ip.tos = ipv6_get_dsfield(nh);
569 if (mask->ipv6_label) {
570 set_ipv6_fl(nh, ntohl(key->ipv6_label),
571 ntohl(mask->ipv6_label));
572 flow_key->ipv6.label =
573 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
575 if (mask->ipv6_hlimit) {
576 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
578 flow_key->ip.ttl = nh->hop_limit;
583 /* Must follow skb_ensure_writable() since that can move the skb data. */
584 static void set_tp_port(struct sk_buff *skb, __be16 *port,
585 __be16 new_port, __sum16 *check)
587 inet_proto_csum_replace2(check, skb, *port, new_port, false);
591 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
592 const struct ovs_key_udp *key,
593 const struct ovs_key_udp *mask)
599 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
600 sizeof(struct udphdr));
605 /* Either of the masks is non-zero, so do not bother checking them. */
606 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
607 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
609 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
610 if (likely(src != uh->source)) {
611 set_tp_port(skb, &uh->source, src, &uh->check);
612 flow_key->tp.src = src;
614 if (likely(dst != uh->dest)) {
615 set_tp_port(skb, &uh->dest, dst, &uh->check);
616 flow_key->tp.dst = dst;
619 if (unlikely(!uh->check))
620 uh->check = CSUM_MANGLED_0;
624 flow_key->tp.src = src;
625 flow_key->tp.dst = dst;
633 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
634 const struct ovs_key_tcp *key,
635 const struct ovs_key_tcp *mask)
641 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
642 sizeof(struct tcphdr));
647 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
648 if (likely(src != th->source)) {
649 set_tp_port(skb, &th->source, src, &th->check);
650 flow_key->tp.src = src;
652 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
653 if (likely(dst != th->dest)) {
654 set_tp_port(skb, &th->dest, dst, &th->check);
655 flow_key->tp.dst = dst;
662 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
663 const struct ovs_key_sctp *key,
664 const struct ovs_key_sctp *mask)
666 unsigned int sctphoff = skb_transport_offset(skb);
668 __le32 old_correct_csum, new_csum, old_csum;
671 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
676 old_csum = sh->checksum;
677 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
679 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
680 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
682 new_csum = sctp_compute_cksum(skb, sctphoff);
684 /* Carry any checksum errors through. */
685 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
688 flow_key->tp.src = sh->source;
689 flow_key->tp.dst = sh->dest;
694 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
696 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
697 struct vport *vport = data->vport;
699 if (skb_cow_head(skb, data->l2_len) < 0) {
704 __skb_dst_copy(skb, data->dst);
705 *OVS_CB(skb) = data->cb;
706 skb->inner_protocol = data->inner_protocol;
707 skb->vlan_tci = data->vlan_tci;
708 skb->vlan_proto = data->vlan_proto;
710 /* Reconstruct the MAC header. */
711 skb_push(skb, data->l2_len);
712 memcpy(skb->data, &data->l2_data, data->l2_len);
713 skb_postpush_rcsum(skb, skb->data, data->l2_len);
714 skb_reset_mac_header(skb);
716 if (eth_p_mpls(skb->protocol)) {
717 skb->inner_network_header = skb->network_header;
718 skb_set_network_header(skb, data->network_offset);
719 skb_reset_mac_len(skb);
722 ovs_vport_send(vport, skb, data->mac_proto);
727 ovs_dst_get_mtu(const struct dst_entry *dst)
729 return dst->dev->mtu;
732 static struct dst_ops ovs_dst_ops = {
734 .mtu = ovs_dst_get_mtu,
737 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
738 * ovs_vport_output(), which is called once per fragmented packet.
740 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
741 u16 orig_network_offset, u8 mac_proto)
743 unsigned int hlen = skb_network_offset(skb);
744 struct ovs_frag_data *data;
746 data = this_cpu_ptr(&ovs_frag_data_storage);
747 data->dst = skb->_skb_refdst;
749 data->cb = *OVS_CB(skb);
750 data->inner_protocol = skb->inner_protocol;
751 data->network_offset = orig_network_offset;
752 data->vlan_tci = skb->vlan_tci;
753 data->vlan_proto = skb->vlan_proto;
754 data->mac_proto = mac_proto;
756 memcpy(&data->l2_data, skb->data, hlen);
758 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
762 static void ovs_fragment(struct net *net, struct vport *vport,
763 struct sk_buff *skb, u16 mru,
764 struct sw_flow_key *key)
766 u16 orig_network_offset = 0;
768 if (eth_p_mpls(skb->protocol)) {
769 orig_network_offset = skb_network_offset(skb);
770 skb->network_header = skb->inner_network_header;
773 if (skb_network_offset(skb) > MAX_L2_LEN) {
774 OVS_NLERR(1, "L2 header too long to fragment");
778 if (key->eth.type == htons(ETH_P_IP)) {
779 struct dst_entry ovs_dst;
780 unsigned long orig_dst;
782 prepare_frag(vport, skb, orig_network_offset,
783 ovs_key_mac_proto(key));
784 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
785 DST_OBSOLETE_NONE, DST_NOCOUNT);
786 ovs_dst.dev = vport->dev;
788 orig_dst = skb->_skb_refdst;
789 skb_dst_set_noref(skb, &ovs_dst);
790 IPCB(skb)->frag_max_size = mru;
792 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
793 refdst_drop(orig_dst);
794 } else if (key->eth.type == htons(ETH_P_IPV6)) {
795 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
796 unsigned long orig_dst;
797 struct rt6_info ovs_rt;
803 prepare_frag(vport, skb, orig_network_offset,
804 ovs_key_mac_proto(key));
805 memset(&ovs_rt, 0, sizeof(ovs_rt));
806 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
807 DST_OBSOLETE_NONE, DST_NOCOUNT);
808 ovs_rt.dst.dev = vport->dev;
810 orig_dst = skb->_skb_refdst;
811 skb_dst_set_noref(skb, &ovs_rt.dst);
812 IP6CB(skb)->frag_max_size = mru;
814 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
815 refdst_drop(orig_dst);
817 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
818 ovs_vport_name(vport), ntohs(key->eth.type), mru,
828 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
829 struct sw_flow_key *key)
831 struct vport *vport = ovs_vport_rcu(dp, out_port);
834 u16 mru = OVS_CB(skb)->mru;
835 u32 cutlen = OVS_CB(skb)->cutlen;
837 if (unlikely(cutlen > 0)) {
838 if (skb->len - cutlen > ovs_mac_header_len(key))
839 pskb_trim(skb, skb->len - cutlen);
841 pskb_trim(skb, ovs_mac_header_len(key));
845 (skb->len <= mru + vport->dev->hard_header_len))) {
846 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
847 } else if (mru <= vport->dev->mtu) {
848 struct net *net = read_pnet(&dp->net);
850 ovs_fragment(net, vport, skb, mru, key);
859 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
860 struct sw_flow_key *key, const struct nlattr *attr,
861 const struct nlattr *actions, int actions_len,
864 struct dp_upcall_info upcall;
865 const struct nlattr *a;
868 memset(&upcall, 0, sizeof(upcall));
869 upcall.cmd = OVS_PACKET_CMD_ACTION;
870 upcall.mru = OVS_CB(skb)->mru;
872 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
873 a = nla_next(a, &rem)) {
874 switch (nla_type(a)) {
875 case OVS_USERSPACE_ATTR_USERDATA:
879 case OVS_USERSPACE_ATTR_PID:
880 upcall.portid = nla_get_u32(a);
883 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
884 /* Get out tunnel info. */
887 vport = ovs_vport_rcu(dp, nla_get_u32(a));
891 err = dev_fill_metadata_dst(vport->dev, skb);
893 upcall.egress_tun_info = skb_tunnel_info(skb);
899 case OVS_USERSPACE_ATTR_ACTIONS: {
900 /* Include actions. */
901 upcall.actions = actions;
902 upcall.actions_len = actions_len;
906 } /* End of switch. */
909 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
912 static int sample(struct datapath *dp, struct sk_buff *skb,
913 struct sw_flow_key *key, const struct nlattr *attr,
914 const struct nlattr *actions, int actions_len)
916 const struct nlattr *acts_list = NULL;
917 const struct nlattr *a;
921 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
922 a = nla_next(a, &rem)) {
925 switch (nla_type(a)) {
926 case OVS_SAMPLE_ATTR_PROBABILITY:
927 probability = nla_get_u32(a);
928 if (!probability || prandom_u32() > probability)
932 case OVS_SAMPLE_ATTR_ACTIONS:
938 rem = nla_len(acts_list);
939 a = nla_data(acts_list);
941 /* Actions list is empty, do nothing */
945 /* The only known usage of sample action is having a single user-space
946 * action, or having a truncate action followed by a single user-space
947 * action. Treat this usage as a special case.
948 * The output_userspace() should clone the skb to be sent to the
949 * user space. This skb will be consumed by its caller.
951 if (unlikely(nla_type(a) == OVS_ACTION_ATTR_TRUNC)) {
952 struct ovs_action_trunc *trunc = nla_data(a);
954 if (skb->len > trunc->max_len)
955 cutlen = skb->len - trunc->max_len;
957 a = nla_next(a, &rem);
960 if (likely(nla_type(a) == OVS_ACTION_ATTR_USERSPACE &&
961 nla_is_last(a, rem)))
962 return output_userspace(dp, skb, key, a, actions,
963 actions_len, cutlen);
965 skb = skb_clone(skb, GFP_ATOMIC);
967 /* Skip the sample action when out of memory. */
970 if (!add_deferred_actions(skb, key, a)) {
972 pr_warn("%s: deferred actions limit reached, dropping sample action\n",
980 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
981 const struct nlattr *attr)
983 struct ovs_action_hash *hash_act = nla_data(attr);
986 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
987 hash = skb_get_hash(skb);
988 hash = jhash_1word(hash, hash_act->hash_basis);
992 key->ovs_flow_hash = hash;
995 static int execute_set_action(struct sk_buff *skb,
996 struct sw_flow_key *flow_key,
997 const struct nlattr *a)
999 /* Only tunnel set execution is supported without a mask. */
1000 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1001 struct ovs_tunnel_info *tun = nla_data(a);
1004 dst_hold((struct dst_entry *)tun->tun_dst);
1005 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1012 /* Mask is at the midpoint of the data. */
1013 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1015 static int execute_masked_set_action(struct sk_buff *skb,
1016 struct sw_flow_key *flow_key,
1017 const struct nlattr *a)
1021 switch (nla_type(a)) {
1022 case OVS_KEY_ATTR_PRIORITY:
1023 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1024 *get_mask(a, u32 *));
1025 flow_key->phy.priority = skb->priority;
1028 case OVS_KEY_ATTR_SKB_MARK:
1029 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1030 flow_key->phy.skb_mark = skb->mark;
1033 case OVS_KEY_ATTR_TUNNEL_INFO:
1034 /* Masked data not supported for tunnel. */
1038 case OVS_KEY_ATTR_ETHERNET:
1039 err = set_eth_addr(skb, flow_key, nla_data(a),
1040 get_mask(a, struct ovs_key_ethernet *));
1043 case OVS_KEY_ATTR_IPV4:
1044 err = set_ipv4(skb, flow_key, nla_data(a),
1045 get_mask(a, struct ovs_key_ipv4 *));
1048 case OVS_KEY_ATTR_IPV6:
1049 err = set_ipv6(skb, flow_key, nla_data(a),
1050 get_mask(a, struct ovs_key_ipv6 *));
1053 case OVS_KEY_ATTR_TCP:
1054 err = set_tcp(skb, flow_key, nla_data(a),
1055 get_mask(a, struct ovs_key_tcp *));
1058 case OVS_KEY_ATTR_UDP:
1059 err = set_udp(skb, flow_key, nla_data(a),
1060 get_mask(a, struct ovs_key_udp *));
1063 case OVS_KEY_ATTR_SCTP:
1064 err = set_sctp(skb, flow_key, nla_data(a),
1065 get_mask(a, struct ovs_key_sctp *));
1068 case OVS_KEY_ATTR_MPLS:
1069 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1073 case OVS_KEY_ATTR_CT_STATE:
1074 case OVS_KEY_ATTR_CT_ZONE:
1075 case OVS_KEY_ATTR_CT_MARK:
1076 case OVS_KEY_ATTR_CT_LABELS:
1077 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1078 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1086 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1087 struct sw_flow_key *key,
1088 const struct nlattr *a, int rem)
1090 struct deferred_action *da;
1093 if (!is_flow_key_valid(key)) {
1096 err = ovs_flow_key_update(skb, key);
1100 BUG_ON(!is_flow_key_valid(key));
1102 if (!nla_is_last(a, rem)) {
1103 /* Recirc action is the not the last action
1104 * of the action list, need to clone the skb.
1106 skb = skb_clone(skb, GFP_ATOMIC);
1108 /* Skip the recirc action when out of memory, but
1109 * continue on with the rest of the action list.
1115 level = this_cpu_read(exec_actions_level);
1116 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
1117 struct recirc_keys *rks = this_cpu_ptr(recirc_keys);
1118 struct sw_flow_key *recirc_key = &rks->key[level - 1];
1121 recirc_key->recirc_id = nla_get_u32(a);
1122 ovs_dp_process_packet(skb, recirc_key);
1127 da = add_deferred_actions(skb, key, NULL);
1129 da->pkt_key.recirc_id = nla_get_u32(a);
1133 if (net_ratelimit())
1134 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1141 /* Execute a list of actions against 'skb'. */
1142 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1143 struct sw_flow_key *key,
1144 const struct nlattr *attr, int len)
1146 const struct nlattr *a;
1149 for (a = attr, rem = len; rem > 0;
1150 a = nla_next(a, &rem)) {
1153 switch (nla_type(a)) {
1154 case OVS_ACTION_ATTR_OUTPUT: {
1155 int port = nla_get_u32(a);
1156 struct sk_buff *clone;
1158 /* Every output action needs a separate clone
1159 * of 'skb', In case the output action is the
1160 * last action, cloning can be avoided.
1162 if (nla_is_last(a, rem)) {
1163 do_output(dp, skb, port, key);
1164 /* 'skb' has been used for output.
1169 clone = skb_clone(skb, GFP_ATOMIC);
1171 do_output(dp, clone, port, key);
1172 OVS_CB(skb)->cutlen = 0;
1176 case OVS_ACTION_ATTR_TRUNC: {
1177 struct ovs_action_trunc *trunc = nla_data(a);
1179 if (skb->len > trunc->max_len)
1180 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1184 case OVS_ACTION_ATTR_USERSPACE:
1185 output_userspace(dp, skb, key, a, attr,
1186 len, OVS_CB(skb)->cutlen);
1187 OVS_CB(skb)->cutlen = 0;
1190 case OVS_ACTION_ATTR_HASH:
1191 execute_hash(skb, key, a);
1194 case OVS_ACTION_ATTR_PUSH_MPLS:
1195 err = push_mpls(skb, key, nla_data(a));
1198 case OVS_ACTION_ATTR_POP_MPLS:
1199 err = pop_mpls(skb, key, nla_get_be16(a));
1202 case OVS_ACTION_ATTR_PUSH_VLAN:
1203 err = push_vlan(skb, key, nla_data(a));
1206 case OVS_ACTION_ATTR_POP_VLAN:
1207 err = pop_vlan(skb, key);
1210 case OVS_ACTION_ATTR_RECIRC:
1211 err = execute_recirc(dp, skb, key, a, rem);
1212 if (nla_is_last(a, rem)) {
1213 /* If this is the last action, the skb has
1214 * been consumed or freed.
1215 * Return immediately.
1221 case OVS_ACTION_ATTR_SET:
1222 err = execute_set_action(skb, key, nla_data(a));
1225 case OVS_ACTION_ATTR_SET_MASKED:
1226 case OVS_ACTION_ATTR_SET_TO_MASKED:
1227 err = execute_masked_set_action(skb, key, nla_data(a));
1230 case OVS_ACTION_ATTR_SAMPLE:
1231 err = sample(dp, skb, key, a, attr, len);
1234 case OVS_ACTION_ATTR_CT:
1235 if (!is_flow_key_valid(key)) {
1236 err = ovs_flow_key_update(skb, key);
1241 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1244 /* Hide stolen IP fragments from user space. */
1246 return err == -EINPROGRESS ? 0 : err;
1249 case OVS_ACTION_ATTR_PUSH_ETH:
1250 err = push_eth(skb, key, nla_data(a));
1253 case OVS_ACTION_ATTR_POP_ETH:
1254 err = pop_eth(skb, key);
1258 if (unlikely(err)) {
1268 static void process_deferred_actions(struct datapath *dp)
1270 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1272 /* Do not touch the FIFO in case there is no deferred actions. */
1273 if (action_fifo_is_empty(fifo))
1276 /* Finishing executing all deferred actions. */
1278 struct deferred_action *da = action_fifo_get(fifo);
1279 struct sk_buff *skb = da->skb;
1280 struct sw_flow_key *key = &da->pkt_key;
1281 const struct nlattr *actions = da->actions;
1284 do_execute_actions(dp, skb, key, actions,
1287 ovs_dp_process_packet(skb, key);
1288 } while (!action_fifo_is_empty(fifo));
1290 /* Reset FIFO for the next packet. */
1291 action_fifo_init(fifo);
1294 /* Execute a list of actions against 'skb'. */
1295 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1296 const struct sw_flow_actions *acts,
1297 struct sw_flow_key *key)
1301 level = __this_cpu_inc_return(exec_actions_level);
1302 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1303 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1310 err = do_execute_actions(dp, skb, key,
1311 acts->actions, acts->actions_len);
1314 process_deferred_actions(dp);
1317 __this_cpu_dec(exec_actions_level);
1321 int action_fifos_init(void)
1323 action_fifos = alloc_percpu(struct action_fifo);
1327 recirc_keys = alloc_percpu(struct recirc_keys);
1329 free_percpu(action_fifos);
1336 void action_fifos_exit(void)
1338 free_percpu(action_fifos);
1339 free_percpu(recirc_keys);