2 * Copyright (c) 2007-2017 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"
46 #include "flow_netlink.h"
48 struct deferred_action {
50 const struct nlattr *actions;
53 /* Store pkt_key clone when creating deferred action. */
54 struct sw_flow_key pkt_key;
57 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
58 struct ovs_frag_data {
62 __be16 inner_protocol;
63 u16 network_offset; /* valid only for MPLS */
68 u8 l2_data[MAX_L2_LEN];
71 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
73 #define DEFERRED_ACTION_FIFO_SIZE 10
74 #define OVS_RECURSION_LIMIT 5
75 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
79 /* Deferred action fifo queue storage. */
80 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
83 struct action_flow_keys {
84 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
87 static struct action_fifo __percpu *action_fifos;
88 static struct action_flow_keys __percpu *flow_keys;
89 static DEFINE_PER_CPU(int, exec_actions_level);
91 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
92 * space. Return NULL if out of key spaces.
94 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
96 struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
97 int level = this_cpu_read(exec_actions_level);
98 struct sw_flow_key *key = NULL;
100 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
101 key = &keys->key[level - 1];
108 static void action_fifo_init(struct action_fifo *fifo)
114 static bool action_fifo_is_empty(const struct action_fifo *fifo)
116 return (fifo->head == fifo->tail);
119 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
121 if (action_fifo_is_empty(fifo))
124 return &fifo->fifo[fifo->tail++];
127 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
129 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
132 return &fifo->fifo[fifo->head++];
135 /* Return true if fifo is not full */
136 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
137 const struct sw_flow_key *key,
138 const struct nlattr *actions,
139 const int actions_len)
141 struct action_fifo *fifo;
142 struct deferred_action *da;
144 fifo = this_cpu_ptr(action_fifos);
145 da = action_fifo_put(fifo);
148 da->actions = actions;
149 da->actions_len = actions_len;
156 static void invalidate_flow_key(struct sw_flow_key *key)
158 key->mac_proto |= SW_FLOW_KEY_INVALID;
161 static bool is_flow_key_valid(const struct sw_flow_key *key)
163 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
166 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
167 struct sw_flow_key *key,
169 const struct nlattr *actions, int len,
170 bool last, bool clone_flow_key);
172 static void update_ethertype(struct sk_buff *skb, struct ethhdr *hdr,
175 if (skb->ip_summed == CHECKSUM_COMPLETE) {
176 __be16 diff[] = { ~(hdr->h_proto), ethertype };
178 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
182 hdr->h_proto = ethertype;
185 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
186 const struct ovs_action_push_mpls *mpls)
188 struct mpls_shim_hdr *new_mpls_lse;
190 /* Networking stack do not allow simultaneous Tunnel and MPLS GSO. */
191 if (skb->encapsulation)
194 if (skb_cow_head(skb, MPLS_HLEN) < 0)
197 if (!skb->inner_protocol) {
198 skb_set_inner_network_header(skb, skb->mac_len);
199 skb_set_inner_protocol(skb, skb->protocol);
202 skb_push(skb, MPLS_HLEN);
203 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
205 skb_reset_mac_header(skb);
206 skb_set_network_header(skb, skb->mac_len);
208 new_mpls_lse = mpls_hdr(skb);
209 new_mpls_lse->label_stack_entry = mpls->mpls_lse;
211 skb_postpush_rcsum(skb, new_mpls_lse, MPLS_HLEN);
213 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET)
214 update_ethertype(skb, eth_hdr(skb), mpls->mpls_ethertype);
215 skb->protocol = mpls->mpls_ethertype;
217 invalidate_flow_key(key);
221 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
222 const __be16 ethertype)
226 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
230 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
232 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
235 __skb_pull(skb, MPLS_HLEN);
236 skb_reset_mac_header(skb);
237 skb_set_network_header(skb, skb->mac_len);
239 if (ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET) {
242 /* mpls_hdr() is used to locate the ethertype field correctly in the
243 * presence of VLAN tags.
245 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
246 update_ethertype(skb, hdr, ethertype);
248 if (eth_p_mpls(skb->protocol))
249 skb->protocol = ethertype;
251 invalidate_flow_key(key);
255 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
256 const __be32 *mpls_lse, const __be32 *mask)
258 struct mpls_shim_hdr *stack;
262 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
266 stack = mpls_hdr(skb);
267 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
268 if (skb->ip_summed == CHECKSUM_COMPLETE) {
269 __be32 diff[] = { ~(stack->label_stack_entry), lse };
271 skb->csum = ~csum_partial((char *)diff, sizeof(diff),
275 stack->label_stack_entry = lse;
276 flow_key->mpls.top_lse = lse;
280 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
284 err = skb_vlan_pop(skb);
285 if (skb_vlan_tag_present(skb)) {
286 invalidate_flow_key(key);
288 key->eth.vlan.tci = 0;
289 key->eth.vlan.tpid = 0;
294 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
295 const struct ovs_action_push_vlan *vlan)
297 if (skb_vlan_tag_present(skb)) {
298 invalidate_flow_key(key);
300 key->eth.vlan.tci = vlan->vlan_tci;
301 key->eth.vlan.tpid = vlan->vlan_tpid;
303 return skb_vlan_push(skb, vlan->vlan_tpid,
304 ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
307 /* 'src' is already properly masked. */
308 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
310 u16 *dst = (u16 *)dst_;
311 const u16 *src = (const u16 *)src_;
312 const u16 *mask = (const u16 *)mask_;
314 OVS_SET_MASKED(dst[0], src[0], mask[0]);
315 OVS_SET_MASKED(dst[1], src[1], mask[1]);
316 OVS_SET_MASKED(dst[2], src[2], mask[2]);
319 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
320 const struct ovs_key_ethernet *key,
321 const struct ovs_key_ethernet *mask)
325 err = skb_ensure_writable(skb, ETH_HLEN);
329 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
331 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
333 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
336 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
338 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
339 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
343 /* pop_eth does not support VLAN packets as this action is never called
346 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
348 skb_pull_rcsum(skb, ETH_HLEN);
349 skb_reset_mac_header(skb);
350 skb_reset_mac_len(skb);
352 /* safe right before invalidate_flow_key */
353 key->mac_proto = MAC_PROTO_NONE;
354 invalidate_flow_key(key);
358 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
359 const struct ovs_action_push_eth *ethh)
363 /* Add the new Ethernet header */
364 if (skb_cow_head(skb, ETH_HLEN) < 0)
367 skb_push(skb, ETH_HLEN);
368 skb_reset_mac_header(skb);
369 skb_reset_mac_len(skb);
372 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
373 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
374 hdr->h_proto = skb->protocol;
376 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
378 /* safe right before invalidate_flow_key */
379 key->mac_proto = MAC_PROTO_ETHERNET;
380 invalidate_flow_key(key);
384 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
385 const struct nshhdr *nh)
389 err = nsh_push(skb, nh);
393 /* safe right before invalidate_flow_key */
394 key->mac_proto = MAC_PROTO_NONE;
395 invalidate_flow_key(key);
399 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
407 /* safe right before invalidate_flow_key */
408 if (skb->protocol == htons(ETH_P_TEB))
409 key->mac_proto = MAC_PROTO_ETHERNET;
411 key->mac_proto = MAC_PROTO_NONE;
412 invalidate_flow_key(key);
416 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
417 __be32 addr, __be32 new_addr)
419 int transport_len = skb->len - skb_transport_offset(skb);
421 if (nh->frag_off & htons(IP_OFFSET))
424 if (nh->protocol == IPPROTO_TCP) {
425 if (likely(transport_len >= sizeof(struct tcphdr)))
426 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
427 addr, new_addr, true);
428 } else if (nh->protocol == IPPROTO_UDP) {
429 if (likely(transport_len >= sizeof(struct udphdr))) {
430 struct udphdr *uh = udp_hdr(skb);
432 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
433 inet_proto_csum_replace4(&uh->check, skb,
434 addr, new_addr, true);
436 uh->check = CSUM_MANGLED_0;
442 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
443 __be32 *addr, __be32 new_addr)
445 update_ip_l4_checksum(skb, nh, *addr, new_addr);
446 csum_replace4(&nh->check, *addr, new_addr);
451 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
452 __be32 addr[4], const __be32 new_addr[4])
454 int transport_len = skb->len - skb_transport_offset(skb);
456 if (l4_proto == NEXTHDR_TCP) {
457 if (likely(transport_len >= sizeof(struct tcphdr)))
458 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
459 addr, new_addr, true);
460 } else if (l4_proto == NEXTHDR_UDP) {
461 if (likely(transport_len >= sizeof(struct udphdr))) {
462 struct udphdr *uh = udp_hdr(skb);
464 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
465 inet_proto_csum_replace16(&uh->check, skb,
466 addr, new_addr, true);
468 uh->check = CSUM_MANGLED_0;
471 } else if (l4_proto == NEXTHDR_ICMP) {
472 if (likely(transport_len >= sizeof(struct icmp6hdr)))
473 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
474 skb, addr, new_addr, true);
478 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
479 const __be32 mask[4], __be32 masked[4])
481 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
482 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
483 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
484 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
487 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
488 __be32 addr[4], const __be32 new_addr[4],
489 bool recalculate_csum)
491 if (recalculate_csum)
492 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
495 memcpy(addr, new_addr, sizeof(__be32[4]));
498 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
500 /* Bits 21-24 are always unmasked, so this retains their values. */
501 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
502 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
503 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
506 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
509 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
511 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
515 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
516 const struct ovs_key_ipv4 *key,
517 const struct ovs_key_ipv4 *mask)
523 err = skb_ensure_writable(skb, skb_network_offset(skb) +
524 sizeof(struct iphdr));
530 /* Setting an IP addresses is typically only a side effect of
531 * matching on them in the current userspace implementation, so it
532 * makes sense to check if the value actually changed.
534 if (mask->ipv4_src) {
535 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
537 if (unlikely(new_addr != nh->saddr)) {
538 set_ip_addr(skb, nh, &nh->saddr, new_addr);
539 flow_key->ipv4.addr.src = new_addr;
542 if (mask->ipv4_dst) {
543 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
545 if (unlikely(new_addr != nh->daddr)) {
546 set_ip_addr(skb, nh, &nh->daddr, new_addr);
547 flow_key->ipv4.addr.dst = new_addr;
550 if (mask->ipv4_tos) {
551 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
552 flow_key->ip.tos = nh->tos;
554 if (mask->ipv4_ttl) {
555 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
556 flow_key->ip.ttl = nh->ttl;
562 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
564 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
567 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
568 const struct ovs_key_ipv6 *key,
569 const struct ovs_key_ipv6 *mask)
574 err = skb_ensure_writable(skb, skb_network_offset(skb) +
575 sizeof(struct ipv6hdr));
581 /* Setting an IP addresses is typically only a side effect of
582 * matching on them in the current userspace implementation, so it
583 * makes sense to check if the value actually changed.
585 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
586 __be32 *saddr = (__be32 *)&nh->saddr;
589 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
591 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
592 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
594 memcpy(&flow_key->ipv6.addr.src, masked,
595 sizeof(flow_key->ipv6.addr.src));
598 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
599 unsigned int offset = 0;
600 int flags = IP6_FH_F_SKIP_RH;
601 bool recalc_csum = true;
602 __be32 *daddr = (__be32 *)&nh->daddr;
605 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
607 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
608 if (ipv6_ext_hdr(nh->nexthdr))
609 recalc_csum = (ipv6_find_hdr(skb, &offset,
614 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
616 memcpy(&flow_key->ipv6.addr.dst, masked,
617 sizeof(flow_key->ipv6.addr.dst));
620 if (mask->ipv6_tclass) {
621 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
622 flow_key->ip.tos = ipv6_get_dsfield(nh);
624 if (mask->ipv6_label) {
625 set_ipv6_fl(nh, ntohl(key->ipv6_label),
626 ntohl(mask->ipv6_label));
627 flow_key->ipv6.label =
628 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
630 if (mask->ipv6_hlimit) {
631 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
633 flow_key->ip.ttl = nh->hop_limit;
638 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
639 const struct nlattr *a)
648 struct ovs_key_nsh key;
649 struct ovs_key_nsh mask;
651 err = nsh_key_from_nlattr(a, &key, &mask);
655 /* Make sure the NSH base header is there */
656 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
660 length = nsh_hdr_len(nh);
662 /* Make sure the whole NSH header is there */
663 err = skb_ensure_writable(skb, skb_network_offset(skb) +
669 skb_postpull_rcsum(skb, nh, length);
670 flags = nsh_get_flags(nh);
671 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
672 flow_key->nsh.base.flags = flags;
673 ttl = nsh_get_ttl(nh);
674 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
675 flow_key->nsh.base.ttl = ttl;
676 nsh_set_flags_and_ttl(nh, flags, ttl);
677 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
679 flow_key->nsh.base.path_hdr = nh->path_hdr;
680 switch (nh->mdtype) {
682 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
684 OVS_MASKED(nh->md1.context[i], key.context[i],
687 memcpy(flow_key->nsh.context, nh->md1.context,
688 sizeof(nh->md1.context));
691 memset(flow_key->nsh.context, 0,
692 sizeof(flow_key->nsh.context));
697 skb_postpush_rcsum(skb, nh, length);
701 /* Must follow skb_ensure_writable() since that can move the skb data. */
702 static void set_tp_port(struct sk_buff *skb, __be16 *port,
703 __be16 new_port, __sum16 *check)
705 inet_proto_csum_replace2(check, skb, *port, new_port, false);
709 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
710 const struct ovs_key_udp *key,
711 const struct ovs_key_udp *mask)
717 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
718 sizeof(struct udphdr));
723 /* Either of the masks is non-zero, so do not bother checking them. */
724 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
725 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
727 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
728 if (likely(src != uh->source)) {
729 set_tp_port(skb, &uh->source, src, &uh->check);
730 flow_key->tp.src = src;
732 if (likely(dst != uh->dest)) {
733 set_tp_port(skb, &uh->dest, dst, &uh->check);
734 flow_key->tp.dst = dst;
737 if (unlikely(!uh->check))
738 uh->check = CSUM_MANGLED_0;
742 flow_key->tp.src = src;
743 flow_key->tp.dst = dst;
751 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
752 const struct ovs_key_tcp *key,
753 const struct ovs_key_tcp *mask)
759 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
760 sizeof(struct tcphdr));
765 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
766 if (likely(src != th->source)) {
767 set_tp_port(skb, &th->source, src, &th->check);
768 flow_key->tp.src = src;
770 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
771 if (likely(dst != th->dest)) {
772 set_tp_port(skb, &th->dest, dst, &th->check);
773 flow_key->tp.dst = dst;
780 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
781 const struct ovs_key_sctp *key,
782 const struct ovs_key_sctp *mask)
784 unsigned int sctphoff = skb_transport_offset(skb);
786 __le32 old_correct_csum, new_csum, old_csum;
789 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
794 old_csum = sh->checksum;
795 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
797 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
798 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
800 new_csum = sctp_compute_cksum(skb, sctphoff);
802 /* Carry any checksum errors through. */
803 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
806 flow_key->tp.src = sh->source;
807 flow_key->tp.dst = sh->dest;
812 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
814 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
815 struct vport *vport = data->vport;
817 if (skb_cow_head(skb, data->l2_len) < 0) {
822 __skb_dst_copy(skb, data->dst);
823 *OVS_CB(skb) = data->cb;
824 skb->inner_protocol = data->inner_protocol;
825 skb->vlan_tci = data->vlan_tci;
826 skb->vlan_proto = data->vlan_proto;
828 /* Reconstruct the MAC header. */
829 skb_push(skb, data->l2_len);
830 memcpy(skb->data, &data->l2_data, data->l2_len);
831 skb_postpush_rcsum(skb, skb->data, data->l2_len);
832 skb_reset_mac_header(skb);
834 if (eth_p_mpls(skb->protocol)) {
835 skb->inner_network_header = skb->network_header;
836 skb_set_network_header(skb, data->network_offset);
837 skb_reset_mac_len(skb);
840 ovs_vport_send(vport, skb, data->mac_proto);
845 ovs_dst_get_mtu(const struct dst_entry *dst)
847 return dst->dev->mtu;
850 static struct dst_ops ovs_dst_ops = {
852 .mtu = ovs_dst_get_mtu,
855 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
856 * ovs_vport_output(), which is called once per fragmented packet.
858 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
859 u16 orig_network_offset, u8 mac_proto)
861 unsigned int hlen = skb_network_offset(skb);
862 struct ovs_frag_data *data;
864 data = this_cpu_ptr(&ovs_frag_data_storage);
865 data->dst = skb->_skb_refdst;
867 data->cb = *OVS_CB(skb);
868 data->inner_protocol = skb->inner_protocol;
869 data->network_offset = orig_network_offset;
870 data->vlan_tci = skb->vlan_tci;
871 data->vlan_proto = skb->vlan_proto;
872 data->mac_proto = mac_proto;
874 memcpy(&data->l2_data, skb->data, hlen);
876 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
880 static void ovs_fragment(struct net *net, struct vport *vport,
881 struct sk_buff *skb, u16 mru,
882 struct sw_flow_key *key)
884 u16 orig_network_offset = 0;
886 if (eth_p_mpls(skb->protocol)) {
887 orig_network_offset = skb_network_offset(skb);
888 skb->network_header = skb->inner_network_header;
891 if (skb_network_offset(skb) > MAX_L2_LEN) {
892 OVS_NLERR(1, "L2 header too long to fragment");
896 if (key->eth.type == htons(ETH_P_IP)) {
897 struct dst_entry ovs_dst;
898 unsigned long orig_dst;
900 prepare_frag(vport, skb, orig_network_offset,
901 ovs_key_mac_proto(key));
902 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
903 DST_OBSOLETE_NONE, DST_NOCOUNT);
904 ovs_dst.dev = vport->dev;
906 orig_dst = skb->_skb_refdst;
907 skb_dst_set_noref(skb, &ovs_dst);
908 IPCB(skb)->frag_max_size = mru;
910 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
911 refdst_drop(orig_dst);
912 } else if (key->eth.type == htons(ETH_P_IPV6)) {
913 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
914 unsigned long orig_dst;
915 struct rt6_info ovs_rt;
920 prepare_frag(vport, skb, orig_network_offset,
921 ovs_key_mac_proto(key));
922 memset(&ovs_rt, 0, sizeof(ovs_rt));
923 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
924 DST_OBSOLETE_NONE, DST_NOCOUNT);
925 ovs_rt.dst.dev = vport->dev;
927 orig_dst = skb->_skb_refdst;
928 skb_dst_set_noref(skb, &ovs_rt.dst);
929 IP6CB(skb)->frag_max_size = mru;
931 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
932 refdst_drop(orig_dst);
934 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
935 ovs_vport_name(vport), ntohs(key->eth.type), mru,
945 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
946 struct sw_flow_key *key)
948 struct vport *vport = ovs_vport_rcu(dp, out_port);
951 u16 mru = OVS_CB(skb)->mru;
952 u32 cutlen = OVS_CB(skb)->cutlen;
954 if (unlikely(cutlen > 0)) {
955 if (skb->len - cutlen > ovs_mac_header_len(key))
956 pskb_trim(skb, skb->len - cutlen);
958 pskb_trim(skb, ovs_mac_header_len(key));
962 (skb->len <= mru + vport->dev->hard_header_len))) {
963 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
964 } else if (mru <= vport->dev->mtu) {
965 struct net *net = read_pnet(&dp->net);
967 ovs_fragment(net, vport, skb, mru, key);
976 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
977 struct sw_flow_key *key, const struct nlattr *attr,
978 const struct nlattr *actions, int actions_len,
981 struct dp_upcall_info upcall;
982 const struct nlattr *a;
985 memset(&upcall, 0, sizeof(upcall));
986 upcall.cmd = OVS_PACKET_CMD_ACTION;
987 upcall.mru = OVS_CB(skb)->mru;
989 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
990 a = nla_next(a, &rem)) {
991 switch (nla_type(a)) {
992 case OVS_USERSPACE_ATTR_USERDATA:
996 case OVS_USERSPACE_ATTR_PID:
997 upcall.portid = nla_get_u32(a);
1000 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
1001 /* Get out tunnel info. */
1002 struct vport *vport;
1004 vport = ovs_vport_rcu(dp, nla_get_u32(a));
1008 err = dev_fill_metadata_dst(vport->dev, skb);
1010 upcall.egress_tun_info = skb_tunnel_info(skb);
1016 case OVS_USERSPACE_ATTR_ACTIONS: {
1017 /* Include actions. */
1018 upcall.actions = actions;
1019 upcall.actions_len = actions_len;
1023 } /* End of switch. */
1026 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
1029 /* When 'last' is true, sample() should always consume the 'skb'.
1030 * Otherwise, sample() should keep 'skb' intact regardless what
1031 * actions are executed within sample().
1033 static int sample(struct datapath *dp, struct sk_buff *skb,
1034 struct sw_flow_key *key, const struct nlattr *attr,
1037 struct nlattr *actions;
1038 struct nlattr *sample_arg;
1039 int rem = nla_len(attr);
1040 const struct sample_arg *arg;
1041 bool clone_flow_key;
1043 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1044 sample_arg = nla_data(attr);
1045 arg = nla_data(sample_arg);
1046 actions = nla_next(sample_arg, &rem);
1048 if ((arg->probability != U32_MAX) &&
1049 (!arg->probability || prandom_u32() > arg->probability)) {
1055 clone_flow_key = !arg->exec;
1056 return clone_execute(dp, skb, key, 0, actions, rem, last,
1060 /* When 'last' is true, clone() should always consume the 'skb'.
1061 * Otherwise, clone() should keep 'skb' intact regardless what
1062 * actions are executed within clone().
1064 static int clone(struct datapath *dp, struct sk_buff *skb,
1065 struct sw_flow_key *key, const struct nlattr *attr,
1068 struct nlattr *actions;
1069 struct nlattr *clone_arg;
1070 int rem = nla_len(attr);
1071 bool dont_clone_flow_key;
1073 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1074 clone_arg = nla_data(attr);
1075 dont_clone_flow_key = nla_get_u32(clone_arg);
1076 actions = nla_next(clone_arg, &rem);
1078 return clone_execute(dp, skb, key, 0, actions, rem, last,
1079 !dont_clone_flow_key);
1082 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1083 const struct nlattr *attr)
1085 struct ovs_action_hash *hash_act = nla_data(attr);
1088 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1089 hash = skb_get_hash(skb);
1090 hash = jhash_1word(hash, hash_act->hash_basis);
1094 key->ovs_flow_hash = hash;
1097 static int execute_set_action(struct sk_buff *skb,
1098 struct sw_flow_key *flow_key,
1099 const struct nlattr *a)
1101 /* Only tunnel set execution is supported without a mask. */
1102 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1103 struct ovs_tunnel_info *tun = nla_data(a);
1106 dst_hold((struct dst_entry *)tun->tun_dst);
1107 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1114 /* Mask is at the midpoint of the data. */
1115 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1117 static int execute_masked_set_action(struct sk_buff *skb,
1118 struct sw_flow_key *flow_key,
1119 const struct nlattr *a)
1123 switch (nla_type(a)) {
1124 case OVS_KEY_ATTR_PRIORITY:
1125 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1126 *get_mask(a, u32 *));
1127 flow_key->phy.priority = skb->priority;
1130 case OVS_KEY_ATTR_SKB_MARK:
1131 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1132 flow_key->phy.skb_mark = skb->mark;
1135 case OVS_KEY_ATTR_TUNNEL_INFO:
1136 /* Masked data not supported for tunnel. */
1140 case OVS_KEY_ATTR_ETHERNET:
1141 err = set_eth_addr(skb, flow_key, nla_data(a),
1142 get_mask(a, struct ovs_key_ethernet *));
1145 case OVS_KEY_ATTR_NSH:
1146 err = set_nsh(skb, flow_key, a);
1149 case OVS_KEY_ATTR_IPV4:
1150 err = set_ipv4(skb, flow_key, nla_data(a),
1151 get_mask(a, struct ovs_key_ipv4 *));
1154 case OVS_KEY_ATTR_IPV6:
1155 err = set_ipv6(skb, flow_key, nla_data(a),
1156 get_mask(a, struct ovs_key_ipv6 *));
1159 case OVS_KEY_ATTR_TCP:
1160 err = set_tcp(skb, flow_key, nla_data(a),
1161 get_mask(a, struct ovs_key_tcp *));
1164 case OVS_KEY_ATTR_UDP:
1165 err = set_udp(skb, flow_key, nla_data(a),
1166 get_mask(a, struct ovs_key_udp *));
1169 case OVS_KEY_ATTR_SCTP:
1170 err = set_sctp(skb, flow_key, nla_data(a),
1171 get_mask(a, struct ovs_key_sctp *));
1174 case OVS_KEY_ATTR_MPLS:
1175 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1179 case OVS_KEY_ATTR_CT_STATE:
1180 case OVS_KEY_ATTR_CT_ZONE:
1181 case OVS_KEY_ATTR_CT_MARK:
1182 case OVS_KEY_ATTR_CT_LABELS:
1183 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1184 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1192 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1193 struct sw_flow_key *key,
1194 const struct nlattr *a, bool last)
1198 if (!is_flow_key_valid(key)) {
1201 err = ovs_flow_key_update(skb, key);
1205 BUG_ON(!is_flow_key_valid(key));
1207 recirc_id = nla_get_u32(a);
1208 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1211 /* Execute a list of actions against 'skb'. */
1212 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1213 struct sw_flow_key *key,
1214 const struct nlattr *attr, int len)
1216 const struct nlattr *a;
1219 for (a = attr, rem = len; rem > 0;
1220 a = nla_next(a, &rem)) {
1223 switch (nla_type(a)) {
1224 case OVS_ACTION_ATTR_OUTPUT: {
1225 int port = nla_get_u32(a);
1226 struct sk_buff *clone;
1228 /* Every output action needs a separate clone
1229 * of 'skb', In case the output action is the
1230 * last action, cloning can be avoided.
1232 if (nla_is_last(a, rem)) {
1233 do_output(dp, skb, port, key);
1234 /* 'skb' has been used for output.
1239 clone = skb_clone(skb, GFP_ATOMIC);
1241 do_output(dp, clone, port, key);
1242 OVS_CB(skb)->cutlen = 0;
1246 case OVS_ACTION_ATTR_TRUNC: {
1247 struct ovs_action_trunc *trunc = nla_data(a);
1249 if (skb->len > trunc->max_len)
1250 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1254 case OVS_ACTION_ATTR_USERSPACE:
1255 output_userspace(dp, skb, key, a, attr,
1256 len, OVS_CB(skb)->cutlen);
1257 OVS_CB(skb)->cutlen = 0;
1260 case OVS_ACTION_ATTR_HASH:
1261 execute_hash(skb, key, a);
1264 case OVS_ACTION_ATTR_PUSH_MPLS:
1265 err = push_mpls(skb, key, nla_data(a));
1268 case OVS_ACTION_ATTR_POP_MPLS:
1269 err = pop_mpls(skb, key, nla_get_be16(a));
1272 case OVS_ACTION_ATTR_PUSH_VLAN:
1273 err = push_vlan(skb, key, nla_data(a));
1276 case OVS_ACTION_ATTR_POP_VLAN:
1277 err = pop_vlan(skb, key);
1280 case OVS_ACTION_ATTR_RECIRC: {
1281 bool last = nla_is_last(a, rem);
1283 err = execute_recirc(dp, skb, key, a, last);
1285 /* If this is the last action, the skb has
1286 * been consumed or freed.
1287 * Return immediately.
1294 case OVS_ACTION_ATTR_SET:
1295 err = execute_set_action(skb, key, nla_data(a));
1298 case OVS_ACTION_ATTR_SET_MASKED:
1299 case OVS_ACTION_ATTR_SET_TO_MASKED:
1300 err = execute_masked_set_action(skb, key, nla_data(a));
1303 case OVS_ACTION_ATTR_SAMPLE: {
1304 bool last = nla_is_last(a, rem);
1306 err = sample(dp, skb, key, a, last);
1313 case OVS_ACTION_ATTR_CT:
1314 if (!is_flow_key_valid(key)) {
1315 err = ovs_flow_key_update(skb, key);
1320 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1323 /* Hide stolen IP fragments from user space. */
1325 return err == -EINPROGRESS ? 0 : err;
1328 case OVS_ACTION_ATTR_CT_CLEAR:
1329 err = ovs_ct_clear(skb, key);
1332 case OVS_ACTION_ATTR_PUSH_ETH:
1333 err = push_eth(skb, key, nla_data(a));
1336 case OVS_ACTION_ATTR_POP_ETH:
1337 err = pop_eth(skb, key);
1340 case OVS_ACTION_ATTR_PUSH_NSH: {
1341 u8 buffer[NSH_HDR_MAX_LEN];
1342 struct nshhdr *nh = (struct nshhdr *)buffer;
1344 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1348 err = push_nsh(skb, key, nh);
1352 case OVS_ACTION_ATTR_POP_NSH:
1353 err = pop_nsh(skb, key);
1356 case OVS_ACTION_ATTR_METER:
1357 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1363 case OVS_ACTION_ATTR_CLONE: {
1364 bool last = nla_is_last(a, rem);
1366 err = clone(dp, skb, key, a, last);
1374 if (unlikely(err)) {
1384 /* Execute the actions on the clone of the packet. The effect of the
1385 * execution does not affect the original 'skb' nor the original 'key'.
1387 * The execution may be deferred in case the actions can not be executed
1390 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1391 struct sw_flow_key *key, u32 recirc_id,
1392 const struct nlattr *actions, int len,
1393 bool last, bool clone_flow_key)
1395 struct deferred_action *da;
1396 struct sw_flow_key *clone;
1398 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1400 /* Out of memory, skip this action.
1405 /* When clone_flow_key is false, the 'key' will not be change
1406 * by the actions, then the 'key' can be used directly.
1407 * Otherwise, try to clone key from the next recursion level of
1408 * 'flow_keys'. If clone is successful, execute the actions
1409 * without deferring.
1411 clone = clone_flow_key ? clone_key(key) : key;
1415 if (actions) { /* Sample action */
1417 __this_cpu_inc(exec_actions_level);
1419 err = do_execute_actions(dp, skb, clone,
1423 __this_cpu_dec(exec_actions_level);
1424 } else { /* Recirc action */
1425 clone->recirc_id = recirc_id;
1426 ovs_dp_process_packet(skb, clone);
1431 /* Out of 'flow_keys' space. Defer actions */
1432 da = add_deferred_actions(skb, key, actions, len);
1434 if (!actions) { /* Recirc action */
1436 key->recirc_id = recirc_id;
1439 /* Out of per CPU action FIFO space. Drop the 'skb' and
1444 if (net_ratelimit()) {
1445 if (actions) { /* Sample action */
1446 pr_warn("%s: deferred action limit reached, drop sample action\n",
1448 } else { /* Recirc action */
1449 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1457 static void process_deferred_actions(struct datapath *dp)
1459 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1461 /* Do not touch the FIFO in case there is no deferred actions. */
1462 if (action_fifo_is_empty(fifo))
1465 /* Finishing executing all deferred actions. */
1467 struct deferred_action *da = action_fifo_get(fifo);
1468 struct sk_buff *skb = da->skb;
1469 struct sw_flow_key *key = &da->pkt_key;
1470 const struct nlattr *actions = da->actions;
1471 int actions_len = da->actions_len;
1474 do_execute_actions(dp, skb, key, actions, actions_len);
1476 ovs_dp_process_packet(skb, key);
1477 } while (!action_fifo_is_empty(fifo));
1479 /* Reset FIFO for the next packet. */
1480 action_fifo_init(fifo);
1483 /* Execute a list of actions against 'skb'. */
1484 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1485 const struct sw_flow_actions *acts,
1486 struct sw_flow_key *key)
1490 level = __this_cpu_inc_return(exec_actions_level);
1491 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1492 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1499 OVS_CB(skb)->acts_origlen = acts->orig_len;
1500 err = do_execute_actions(dp, skb, key,
1501 acts->actions, acts->actions_len);
1504 process_deferred_actions(dp);
1507 __this_cpu_dec(exec_actions_level);
1511 int action_fifos_init(void)
1513 action_fifos = alloc_percpu(struct action_fifo);
1517 flow_keys = alloc_percpu(struct action_flow_keys);
1519 free_percpu(action_fifos);
1526 void action_fifos_exit(void)
1528 free_percpu(action_fifos);
1529 free_percpu(flow_keys);