1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2007-2017 Nicira, Inc.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #include <linux/skbuff.h>
11 #include <linux/openvswitch.h>
12 #include <linux/sctp.h>
13 #include <linux/tcp.h>
14 #include <linux/udp.h>
15 #include <linux/in6.h>
16 #include <linux/if_arp.h>
17 #include <linux/if_vlan.h>
22 #include <net/ip6_fib.h>
23 #include <net/checksum.h>
24 #include <net/dsfield.h>
26 #include <net/sctp/checksum.h>
30 #include "conntrack.h"
32 #include "flow_netlink.h"
34 struct deferred_action {
36 const struct nlattr *actions;
39 /* Store pkt_key clone when creating deferred action. */
40 struct sw_flow_key pkt_key;
43 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
44 struct ovs_frag_data {
48 __be16 inner_protocol;
49 u16 network_offset; /* valid only for MPLS */
54 u8 l2_data[MAX_L2_LEN];
57 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
59 #define DEFERRED_ACTION_FIFO_SIZE 10
60 #define OVS_RECURSION_LIMIT 5
61 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
65 /* Deferred action fifo queue storage. */
66 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
69 struct action_flow_keys {
70 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
73 static struct action_fifo __percpu *action_fifos;
74 static struct action_flow_keys __percpu *flow_keys;
75 static DEFINE_PER_CPU(int, exec_actions_level);
77 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
78 * space. Return NULL if out of key spaces.
80 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
82 struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
83 int level = this_cpu_read(exec_actions_level);
84 struct sw_flow_key *key = NULL;
86 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
87 key = &keys->key[level - 1];
94 static void action_fifo_init(struct action_fifo *fifo)
100 static bool action_fifo_is_empty(const struct action_fifo *fifo)
102 return (fifo->head == fifo->tail);
105 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
107 if (action_fifo_is_empty(fifo))
110 return &fifo->fifo[fifo->tail++];
113 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
115 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
118 return &fifo->fifo[fifo->head++];
121 /* Return true if fifo is not full */
122 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
123 const struct sw_flow_key *key,
124 const struct nlattr *actions,
125 const int actions_len)
127 struct action_fifo *fifo;
128 struct deferred_action *da;
130 fifo = this_cpu_ptr(action_fifos);
131 da = action_fifo_put(fifo);
134 da->actions = actions;
135 da->actions_len = actions_len;
142 static void invalidate_flow_key(struct sw_flow_key *key)
144 key->mac_proto |= SW_FLOW_KEY_INVALID;
147 static bool is_flow_key_valid(const struct sw_flow_key *key)
149 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
152 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
153 struct sw_flow_key *key,
155 const struct nlattr *actions, int len,
156 bool last, bool clone_flow_key);
158 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
159 struct sw_flow_key *key,
160 const struct nlattr *attr, int len);
162 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
163 __be32 mpls_lse, __be16 mpls_ethertype, __u16 mac_len)
167 err = skb_mpls_push(skb, mpls_lse, mpls_ethertype, mac_len, !!mac_len);
172 key->mac_proto = MAC_PROTO_NONE;
174 invalidate_flow_key(key);
178 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
179 const __be16 ethertype)
183 err = skb_mpls_pop(skb, ethertype, skb->mac_len,
184 ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
188 if (ethertype == htons(ETH_P_TEB))
189 key->mac_proto = MAC_PROTO_ETHERNET;
191 invalidate_flow_key(key);
195 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
196 const __be32 *mpls_lse, const __be32 *mask)
198 struct mpls_shim_hdr *stack;
202 stack = mpls_hdr(skb);
203 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
204 err = skb_mpls_update_lse(skb, lse);
208 flow_key->mpls.lse[0] = lse;
212 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
216 err = skb_vlan_pop(skb);
217 if (skb_vlan_tag_present(skb)) {
218 invalidate_flow_key(key);
220 key->eth.vlan.tci = 0;
221 key->eth.vlan.tpid = 0;
226 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
227 const struct ovs_action_push_vlan *vlan)
229 if (skb_vlan_tag_present(skb)) {
230 invalidate_flow_key(key);
232 key->eth.vlan.tci = vlan->vlan_tci;
233 key->eth.vlan.tpid = vlan->vlan_tpid;
235 return skb_vlan_push(skb, vlan->vlan_tpid,
236 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
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 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 /* pop_eth does not support VLAN packets as this action is never called
278 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
282 err = skb_eth_pop(skb);
286 /* safe right before invalidate_flow_key */
287 key->mac_proto = MAC_PROTO_NONE;
288 invalidate_flow_key(key);
292 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
293 const struct ovs_action_push_eth *ethh)
297 err = skb_eth_push(skb, ethh->addresses.eth_dst,
298 ethh->addresses.eth_src);
302 /* safe right before invalidate_flow_key */
303 key->mac_proto = MAC_PROTO_ETHERNET;
304 invalidate_flow_key(key);
308 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
309 const struct nshhdr *nh)
313 err = nsh_push(skb, nh);
317 /* safe right before invalidate_flow_key */
318 key->mac_proto = MAC_PROTO_NONE;
319 invalidate_flow_key(key);
323 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
331 /* safe right before invalidate_flow_key */
332 if (skb->protocol == htons(ETH_P_TEB))
333 key->mac_proto = MAC_PROTO_ETHERNET;
335 key->mac_proto = MAC_PROTO_NONE;
336 invalidate_flow_key(key);
340 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
341 __be32 addr, __be32 new_addr)
343 int transport_len = skb->len - skb_transport_offset(skb);
345 if (nh->frag_off & htons(IP_OFFSET))
348 if (nh->protocol == IPPROTO_TCP) {
349 if (likely(transport_len >= sizeof(struct tcphdr)))
350 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
351 addr, new_addr, true);
352 } else if (nh->protocol == IPPROTO_UDP) {
353 if (likely(transport_len >= sizeof(struct udphdr))) {
354 struct udphdr *uh = udp_hdr(skb);
356 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
357 inet_proto_csum_replace4(&uh->check, skb,
358 addr, new_addr, true);
360 uh->check = CSUM_MANGLED_0;
366 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
367 __be32 *addr, __be32 new_addr)
369 update_ip_l4_checksum(skb, nh, *addr, new_addr);
370 csum_replace4(&nh->check, *addr, new_addr);
375 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
376 __be32 addr[4], const __be32 new_addr[4])
378 int transport_len = skb->len - skb_transport_offset(skb);
380 if (l4_proto == NEXTHDR_TCP) {
381 if (likely(transport_len >= sizeof(struct tcphdr)))
382 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
383 addr, new_addr, true);
384 } else if (l4_proto == NEXTHDR_UDP) {
385 if (likely(transport_len >= sizeof(struct udphdr))) {
386 struct udphdr *uh = udp_hdr(skb);
388 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
389 inet_proto_csum_replace16(&uh->check, skb,
390 addr, new_addr, true);
392 uh->check = CSUM_MANGLED_0;
395 } else if (l4_proto == NEXTHDR_ICMP) {
396 if (likely(transport_len >= sizeof(struct icmp6hdr)))
397 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
398 skb, addr, new_addr, true);
402 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
403 const __be32 mask[4], __be32 masked[4])
405 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
406 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
407 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
408 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
411 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
412 __be32 addr[4], const __be32 new_addr[4],
413 bool recalculate_csum)
415 if (recalculate_csum)
416 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
419 memcpy(addr, new_addr, sizeof(__be32[4]));
422 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
424 /* Bits 21-24 are always unmasked, so this retains their values. */
425 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
426 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
427 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
430 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
433 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
435 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
439 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
440 const struct ovs_key_ipv4 *key,
441 const struct ovs_key_ipv4 *mask)
447 err = skb_ensure_writable(skb, skb_network_offset(skb) +
448 sizeof(struct iphdr));
454 /* Setting an IP addresses is typically only a side effect of
455 * matching on them in the current userspace implementation, so it
456 * makes sense to check if the value actually changed.
458 if (mask->ipv4_src) {
459 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
461 if (unlikely(new_addr != nh->saddr)) {
462 set_ip_addr(skb, nh, &nh->saddr, new_addr);
463 flow_key->ipv4.addr.src = new_addr;
466 if (mask->ipv4_dst) {
467 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
469 if (unlikely(new_addr != nh->daddr)) {
470 set_ip_addr(skb, nh, &nh->daddr, new_addr);
471 flow_key->ipv4.addr.dst = new_addr;
474 if (mask->ipv4_tos) {
475 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
476 flow_key->ip.tos = nh->tos;
478 if (mask->ipv4_ttl) {
479 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
480 flow_key->ip.ttl = nh->ttl;
486 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
488 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
491 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
492 const struct ovs_key_ipv6 *key,
493 const struct ovs_key_ipv6 *mask)
498 err = skb_ensure_writable(skb, skb_network_offset(skb) +
499 sizeof(struct ipv6hdr));
505 /* Setting an IP addresses is typically only a side effect of
506 * matching on them in the current userspace implementation, so it
507 * makes sense to check if the value actually changed.
509 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
510 __be32 *saddr = (__be32 *)&nh->saddr;
513 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
515 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
516 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
518 memcpy(&flow_key->ipv6.addr.src, masked,
519 sizeof(flow_key->ipv6.addr.src));
522 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
523 unsigned int offset = 0;
524 int flags = IP6_FH_F_SKIP_RH;
525 bool recalc_csum = true;
526 __be32 *daddr = (__be32 *)&nh->daddr;
529 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
531 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
532 if (ipv6_ext_hdr(nh->nexthdr))
533 recalc_csum = (ipv6_find_hdr(skb, &offset,
538 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
540 memcpy(&flow_key->ipv6.addr.dst, masked,
541 sizeof(flow_key->ipv6.addr.dst));
544 if (mask->ipv6_tclass) {
545 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
546 flow_key->ip.tos = ipv6_get_dsfield(nh);
548 if (mask->ipv6_label) {
549 set_ipv6_fl(nh, ntohl(key->ipv6_label),
550 ntohl(mask->ipv6_label));
551 flow_key->ipv6.label =
552 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
554 if (mask->ipv6_hlimit) {
555 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
557 flow_key->ip.ttl = nh->hop_limit;
562 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
563 const struct nlattr *a)
572 struct ovs_key_nsh key;
573 struct ovs_key_nsh mask;
575 err = nsh_key_from_nlattr(a, &key, &mask);
579 /* Make sure the NSH base header is there */
580 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
584 length = nsh_hdr_len(nh);
586 /* Make sure the whole NSH header is there */
587 err = skb_ensure_writable(skb, skb_network_offset(skb) +
593 skb_postpull_rcsum(skb, nh, length);
594 flags = nsh_get_flags(nh);
595 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
596 flow_key->nsh.base.flags = flags;
597 ttl = nsh_get_ttl(nh);
598 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
599 flow_key->nsh.base.ttl = ttl;
600 nsh_set_flags_and_ttl(nh, flags, ttl);
601 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
603 flow_key->nsh.base.path_hdr = nh->path_hdr;
604 switch (nh->mdtype) {
606 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
608 OVS_MASKED(nh->md1.context[i], key.context[i],
611 memcpy(flow_key->nsh.context, nh->md1.context,
612 sizeof(nh->md1.context));
615 memset(flow_key->nsh.context, 0,
616 sizeof(flow_key->nsh.context));
621 skb_postpush_rcsum(skb, nh, length);
625 /* Must follow skb_ensure_writable() since that can move the skb data. */
626 static void set_tp_port(struct sk_buff *skb, __be16 *port,
627 __be16 new_port, __sum16 *check)
629 inet_proto_csum_replace2(check, skb, *port, new_port, false);
633 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
634 const struct ovs_key_udp *key,
635 const struct ovs_key_udp *mask)
641 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
642 sizeof(struct udphdr));
647 /* Either of the masks is non-zero, so do not bother checking them. */
648 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
649 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
651 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
652 if (likely(src != uh->source)) {
653 set_tp_port(skb, &uh->source, src, &uh->check);
654 flow_key->tp.src = src;
656 if (likely(dst != uh->dest)) {
657 set_tp_port(skb, &uh->dest, dst, &uh->check);
658 flow_key->tp.dst = dst;
661 if (unlikely(!uh->check))
662 uh->check = CSUM_MANGLED_0;
666 flow_key->tp.src = src;
667 flow_key->tp.dst = dst;
675 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
676 const struct ovs_key_tcp *key,
677 const struct ovs_key_tcp *mask)
683 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
684 sizeof(struct tcphdr));
689 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
690 if (likely(src != th->source)) {
691 set_tp_port(skb, &th->source, src, &th->check);
692 flow_key->tp.src = src;
694 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
695 if (likely(dst != th->dest)) {
696 set_tp_port(skb, &th->dest, dst, &th->check);
697 flow_key->tp.dst = dst;
704 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
705 const struct ovs_key_sctp *key,
706 const struct ovs_key_sctp *mask)
708 unsigned int sctphoff = skb_transport_offset(skb);
710 __le32 old_correct_csum, new_csum, old_csum;
713 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
718 old_csum = sh->checksum;
719 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
721 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
722 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
724 new_csum = sctp_compute_cksum(skb, sctphoff);
726 /* Carry any checksum errors through. */
727 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
730 flow_key->tp.src = sh->source;
731 flow_key->tp.dst = sh->dest;
736 static int ovs_vport_output(struct net *net, struct sock *sk,
739 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
740 struct vport *vport = data->vport;
742 if (skb_cow_head(skb, data->l2_len) < 0) {
747 __skb_dst_copy(skb, data->dst);
748 *OVS_CB(skb) = data->cb;
749 skb->inner_protocol = data->inner_protocol;
750 if (data->vlan_tci & VLAN_CFI_MASK)
751 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
753 __vlan_hwaccel_clear_tag(skb);
755 /* Reconstruct the MAC header. */
756 skb_push(skb, data->l2_len);
757 memcpy(skb->data, &data->l2_data, data->l2_len);
758 skb_postpush_rcsum(skb, skb->data, data->l2_len);
759 skb_reset_mac_header(skb);
761 if (eth_p_mpls(skb->protocol)) {
762 skb->inner_network_header = skb->network_header;
763 skb_set_network_header(skb, data->network_offset);
764 skb_reset_mac_len(skb);
767 ovs_vport_send(vport, skb, data->mac_proto);
772 ovs_dst_get_mtu(const struct dst_entry *dst)
774 return dst->dev->mtu;
777 static struct dst_ops ovs_dst_ops = {
779 .mtu = ovs_dst_get_mtu,
782 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
783 * ovs_vport_output(), which is called once per fragmented packet.
785 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
786 u16 orig_network_offset, u8 mac_proto)
788 unsigned int hlen = skb_network_offset(skb);
789 struct ovs_frag_data *data;
791 data = this_cpu_ptr(&ovs_frag_data_storage);
792 data->dst = skb->_skb_refdst;
794 data->cb = *OVS_CB(skb);
795 data->inner_protocol = skb->inner_protocol;
796 data->network_offset = orig_network_offset;
797 if (skb_vlan_tag_present(skb))
798 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
801 data->vlan_proto = skb->vlan_proto;
802 data->mac_proto = mac_proto;
804 memcpy(&data->l2_data, skb->data, hlen);
806 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
810 static void ovs_fragment(struct net *net, struct vport *vport,
811 struct sk_buff *skb, u16 mru,
812 struct sw_flow_key *key)
814 u16 orig_network_offset = 0;
816 if (eth_p_mpls(skb->protocol)) {
817 orig_network_offset = skb_network_offset(skb);
818 skb->network_header = skb->inner_network_header;
821 if (skb_network_offset(skb) > MAX_L2_LEN) {
822 OVS_NLERR(1, "L2 header too long to fragment");
826 if (key->eth.type == htons(ETH_P_IP)) {
827 struct dst_entry ovs_dst;
828 unsigned long orig_dst;
830 prepare_frag(vport, skb, orig_network_offset,
831 ovs_key_mac_proto(key));
832 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
833 DST_OBSOLETE_NONE, DST_NOCOUNT);
834 ovs_dst.dev = vport->dev;
836 orig_dst = skb->_skb_refdst;
837 skb_dst_set_noref(skb, &ovs_dst);
838 IPCB(skb)->frag_max_size = mru;
840 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
841 refdst_drop(orig_dst);
842 } else if (key->eth.type == htons(ETH_P_IPV6)) {
843 unsigned long orig_dst;
844 struct rt6_info ovs_rt;
846 prepare_frag(vport, skb, orig_network_offset,
847 ovs_key_mac_proto(key));
848 memset(&ovs_rt, 0, sizeof(ovs_rt));
849 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
850 DST_OBSOLETE_NONE, DST_NOCOUNT);
851 ovs_rt.dst.dev = vport->dev;
853 orig_dst = skb->_skb_refdst;
854 skb_dst_set_noref(skb, &ovs_rt.dst);
855 IP6CB(skb)->frag_max_size = mru;
857 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
858 refdst_drop(orig_dst);
860 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
861 ovs_vport_name(vport), ntohs(key->eth.type), mru,
871 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
872 struct sw_flow_key *key)
874 struct vport *vport = ovs_vport_rcu(dp, out_port);
877 u16 mru = OVS_CB(skb)->mru;
878 u32 cutlen = OVS_CB(skb)->cutlen;
880 if (unlikely(cutlen > 0)) {
881 if (skb->len - cutlen > ovs_mac_header_len(key))
882 pskb_trim(skb, skb->len - cutlen);
884 pskb_trim(skb, ovs_mac_header_len(key));
888 (skb->len <= mru + vport->dev->hard_header_len))) {
889 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
890 } else if (mru <= vport->dev->mtu) {
891 struct net *net = read_pnet(&dp->net);
893 ovs_fragment(net, vport, skb, mru, key);
902 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
903 struct sw_flow_key *key, const struct nlattr *attr,
904 const struct nlattr *actions, int actions_len,
907 struct dp_upcall_info upcall;
908 const struct nlattr *a;
911 memset(&upcall, 0, sizeof(upcall));
912 upcall.cmd = OVS_PACKET_CMD_ACTION;
913 upcall.mru = OVS_CB(skb)->mru;
915 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
916 a = nla_next(a, &rem)) {
917 switch (nla_type(a)) {
918 case OVS_USERSPACE_ATTR_USERDATA:
922 case OVS_USERSPACE_ATTR_PID:
923 upcall.portid = nla_get_u32(a);
926 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
927 /* Get out tunnel info. */
930 vport = ovs_vport_rcu(dp, nla_get_u32(a));
934 err = dev_fill_metadata_dst(vport->dev, skb);
936 upcall.egress_tun_info = skb_tunnel_info(skb);
942 case OVS_USERSPACE_ATTR_ACTIONS: {
943 /* Include actions. */
944 upcall.actions = actions;
945 upcall.actions_len = actions_len;
949 } /* End of switch. */
952 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
955 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
956 struct sw_flow_key *key,
957 const struct nlattr *attr, bool last)
959 /* The first action is always 'OVS_DEC_TTL_ATTR_ARG'. */
960 struct nlattr *dec_ttl_arg = nla_data(attr);
961 int rem = nla_len(attr);
963 if (nla_len(dec_ttl_arg)) {
964 struct nlattr *actions = nla_next(dec_ttl_arg, &rem);
967 return clone_execute(dp, skb, key, 0, actions, rem,
974 /* When 'last' is true, sample() should always consume the 'skb'.
975 * Otherwise, sample() should keep 'skb' intact regardless what
976 * actions are executed within sample().
978 static int sample(struct datapath *dp, struct sk_buff *skb,
979 struct sw_flow_key *key, const struct nlattr *attr,
982 struct nlattr *actions;
983 struct nlattr *sample_arg;
984 int rem = nla_len(attr);
985 const struct sample_arg *arg;
988 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
989 sample_arg = nla_data(attr);
990 arg = nla_data(sample_arg);
991 actions = nla_next(sample_arg, &rem);
993 if ((arg->probability != U32_MAX) &&
994 (!arg->probability || prandom_u32() > arg->probability)) {
1000 clone_flow_key = !arg->exec;
1001 return clone_execute(dp, skb, key, 0, actions, rem, last,
1005 /* When 'last' is true, clone() should always consume the 'skb'.
1006 * Otherwise, clone() should keep 'skb' intact regardless what
1007 * actions are executed within clone().
1009 static int clone(struct datapath *dp, struct sk_buff *skb,
1010 struct sw_flow_key *key, const struct nlattr *attr,
1013 struct nlattr *actions;
1014 struct nlattr *clone_arg;
1015 int rem = nla_len(attr);
1016 bool dont_clone_flow_key;
1018 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1019 clone_arg = nla_data(attr);
1020 dont_clone_flow_key = nla_get_u32(clone_arg);
1021 actions = nla_next(clone_arg, &rem);
1023 return clone_execute(dp, skb, key, 0, actions, rem, last,
1024 !dont_clone_flow_key);
1027 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1028 const struct nlattr *attr)
1030 struct ovs_action_hash *hash_act = nla_data(attr);
1033 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1034 hash = skb_get_hash(skb);
1035 hash = jhash_1word(hash, hash_act->hash_basis);
1039 key->ovs_flow_hash = hash;
1042 static int execute_set_action(struct sk_buff *skb,
1043 struct sw_flow_key *flow_key,
1044 const struct nlattr *a)
1046 /* Only tunnel set execution is supported without a mask. */
1047 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1048 struct ovs_tunnel_info *tun = nla_data(a);
1051 dst_hold((struct dst_entry *)tun->tun_dst);
1052 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1059 /* Mask is at the midpoint of the data. */
1060 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1062 static int execute_masked_set_action(struct sk_buff *skb,
1063 struct sw_flow_key *flow_key,
1064 const struct nlattr *a)
1068 switch (nla_type(a)) {
1069 case OVS_KEY_ATTR_PRIORITY:
1070 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1071 *get_mask(a, u32 *));
1072 flow_key->phy.priority = skb->priority;
1075 case OVS_KEY_ATTR_SKB_MARK:
1076 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1077 flow_key->phy.skb_mark = skb->mark;
1080 case OVS_KEY_ATTR_TUNNEL_INFO:
1081 /* Masked data not supported for tunnel. */
1085 case OVS_KEY_ATTR_ETHERNET:
1086 err = set_eth_addr(skb, flow_key, nla_data(a),
1087 get_mask(a, struct ovs_key_ethernet *));
1090 case OVS_KEY_ATTR_NSH:
1091 err = set_nsh(skb, flow_key, a);
1094 case OVS_KEY_ATTR_IPV4:
1095 err = set_ipv4(skb, flow_key, nla_data(a),
1096 get_mask(a, struct ovs_key_ipv4 *));
1099 case OVS_KEY_ATTR_IPV6:
1100 err = set_ipv6(skb, flow_key, nla_data(a),
1101 get_mask(a, struct ovs_key_ipv6 *));
1104 case OVS_KEY_ATTR_TCP:
1105 err = set_tcp(skb, flow_key, nla_data(a),
1106 get_mask(a, struct ovs_key_tcp *));
1109 case OVS_KEY_ATTR_UDP:
1110 err = set_udp(skb, flow_key, nla_data(a),
1111 get_mask(a, struct ovs_key_udp *));
1114 case OVS_KEY_ATTR_SCTP:
1115 err = set_sctp(skb, flow_key, nla_data(a),
1116 get_mask(a, struct ovs_key_sctp *));
1119 case OVS_KEY_ATTR_MPLS:
1120 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1124 case OVS_KEY_ATTR_CT_STATE:
1125 case OVS_KEY_ATTR_CT_ZONE:
1126 case OVS_KEY_ATTR_CT_MARK:
1127 case OVS_KEY_ATTR_CT_LABELS:
1128 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1129 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1137 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1138 struct sw_flow_key *key,
1139 const struct nlattr *a, bool last)
1143 if (!is_flow_key_valid(key)) {
1146 err = ovs_flow_key_update(skb, key);
1150 BUG_ON(!is_flow_key_valid(key));
1152 recirc_id = nla_get_u32(a);
1153 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1156 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1157 struct sw_flow_key *key,
1158 const struct nlattr *attr, bool last)
1160 struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1161 const struct nlattr *actions, *cpl_arg;
1162 int len, max_len, rem = nla_len(attr);
1163 const struct check_pkt_len_arg *arg;
1164 bool clone_flow_key;
1166 /* The first netlink attribute in 'attr' is always
1167 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1169 cpl_arg = nla_data(attr);
1170 arg = nla_data(cpl_arg);
1172 len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1173 max_len = arg->pkt_len;
1175 if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1177 /* Second netlink attribute in 'attr' is always
1178 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1180 actions = nla_next(cpl_arg, &rem);
1181 clone_flow_key = !arg->exec_for_lesser_equal;
1183 /* Third netlink attribute in 'attr' is always
1184 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1186 actions = nla_next(cpl_arg, &rem);
1187 actions = nla_next(actions, &rem);
1188 clone_flow_key = !arg->exec_for_greater;
1191 return clone_execute(dp, skb, key, 0, nla_data(actions),
1192 nla_len(actions), last, clone_flow_key);
1195 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1199 if (skb->protocol == htons(ETH_P_IPV6)) {
1202 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1209 if (nh->hop_limit <= 1)
1210 return -EHOSTUNREACH;
1212 key->ip.ttl = --nh->hop_limit;
1217 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1224 return -EHOSTUNREACH;
1226 old_ttl = nh->ttl--;
1227 csum_replace2(&nh->check, htons(old_ttl << 8),
1228 htons(nh->ttl << 8));
1229 key->ip.ttl = nh->ttl;
1234 /* Execute a list of actions against 'skb'. */
1235 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1236 struct sw_flow_key *key,
1237 const struct nlattr *attr, int len)
1239 const struct nlattr *a;
1242 for (a = attr, rem = len; rem > 0;
1243 a = nla_next(a, &rem)) {
1246 switch (nla_type(a)) {
1247 case OVS_ACTION_ATTR_OUTPUT: {
1248 int port = nla_get_u32(a);
1249 struct sk_buff *clone;
1251 /* Every output action needs a separate clone
1252 * of 'skb', In case the output action is the
1253 * last action, cloning can be avoided.
1255 if (nla_is_last(a, rem)) {
1256 do_output(dp, skb, port, key);
1257 /* 'skb' has been used for output.
1262 clone = skb_clone(skb, GFP_ATOMIC);
1264 do_output(dp, clone, port, key);
1265 OVS_CB(skb)->cutlen = 0;
1269 case OVS_ACTION_ATTR_TRUNC: {
1270 struct ovs_action_trunc *trunc = nla_data(a);
1272 if (skb->len > trunc->max_len)
1273 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1277 case OVS_ACTION_ATTR_USERSPACE:
1278 output_userspace(dp, skb, key, a, attr,
1279 len, OVS_CB(skb)->cutlen);
1280 OVS_CB(skb)->cutlen = 0;
1283 case OVS_ACTION_ATTR_HASH:
1284 execute_hash(skb, key, a);
1287 case OVS_ACTION_ATTR_PUSH_MPLS: {
1288 struct ovs_action_push_mpls *mpls = nla_data(a);
1290 err = push_mpls(skb, key, mpls->mpls_lse,
1291 mpls->mpls_ethertype, skb->mac_len);
1294 case OVS_ACTION_ATTR_ADD_MPLS: {
1295 struct ovs_action_add_mpls *mpls = nla_data(a);
1298 if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1299 mac_len = skb->mac_len;
1301 err = push_mpls(skb, key, mpls->mpls_lse,
1302 mpls->mpls_ethertype, mac_len);
1305 case OVS_ACTION_ATTR_POP_MPLS:
1306 err = pop_mpls(skb, key, nla_get_be16(a));
1309 case OVS_ACTION_ATTR_PUSH_VLAN:
1310 err = push_vlan(skb, key, nla_data(a));
1313 case OVS_ACTION_ATTR_POP_VLAN:
1314 err = pop_vlan(skb, key);
1317 case OVS_ACTION_ATTR_RECIRC: {
1318 bool last = nla_is_last(a, rem);
1320 err = execute_recirc(dp, skb, key, a, last);
1322 /* If this is the last action, the skb has
1323 * been consumed or freed.
1324 * Return immediately.
1331 case OVS_ACTION_ATTR_SET:
1332 err = execute_set_action(skb, key, nla_data(a));
1335 case OVS_ACTION_ATTR_SET_MASKED:
1336 case OVS_ACTION_ATTR_SET_TO_MASKED:
1337 err = execute_masked_set_action(skb, key, nla_data(a));
1340 case OVS_ACTION_ATTR_SAMPLE: {
1341 bool last = nla_is_last(a, rem);
1343 err = sample(dp, skb, key, a, last);
1350 case OVS_ACTION_ATTR_CT:
1351 if (!is_flow_key_valid(key)) {
1352 err = ovs_flow_key_update(skb, key);
1357 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1360 /* Hide stolen IP fragments from user space. */
1362 return err == -EINPROGRESS ? 0 : err;
1365 case OVS_ACTION_ATTR_CT_CLEAR:
1366 err = ovs_ct_clear(skb, key);
1369 case OVS_ACTION_ATTR_PUSH_ETH:
1370 err = push_eth(skb, key, nla_data(a));
1373 case OVS_ACTION_ATTR_POP_ETH:
1374 err = pop_eth(skb, key);
1377 case OVS_ACTION_ATTR_PUSH_NSH: {
1378 u8 buffer[NSH_HDR_MAX_LEN];
1379 struct nshhdr *nh = (struct nshhdr *)buffer;
1381 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1385 err = push_nsh(skb, key, nh);
1389 case OVS_ACTION_ATTR_POP_NSH:
1390 err = pop_nsh(skb, key);
1393 case OVS_ACTION_ATTR_METER:
1394 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1400 case OVS_ACTION_ATTR_CLONE: {
1401 bool last = nla_is_last(a, rem);
1403 err = clone(dp, skb, key, a, last);
1410 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1411 bool last = nla_is_last(a, rem);
1413 err = execute_check_pkt_len(dp, skb, key, a, last);
1420 case OVS_ACTION_ATTR_DEC_TTL:
1421 err = execute_dec_ttl(skb, key);
1422 if (err == -EHOSTUNREACH) {
1423 err = dec_ttl_exception_handler(dp, skb, key,
1430 if (unlikely(err)) {
1440 /* Execute the actions on the clone of the packet. The effect of the
1441 * execution does not affect the original 'skb' nor the original 'key'.
1443 * The execution may be deferred in case the actions can not be executed
1446 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1447 struct sw_flow_key *key, u32 recirc_id,
1448 const struct nlattr *actions, int len,
1449 bool last, bool clone_flow_key)
1451 struct deferred_action *da;
1452 struct sw_flow_key *clone;
1454 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1456 /* Out of memory, skip this action.
1461 /* When clone_flow_key is false, the 'key' will not be change
1462 * by the actions, then the 'key' can be used directly.
1463 * Otherwise, try to clone key from the next recursion level of
1464 * 'flow_keys'. If clone is successful, execute the actions
1465 * without deferring.
1467 clone = clone_flow_key ? clone_key(key) : key;
1471 if (actions) { /* Sample action */
1473 __this_cpu_inc(exec_actions_level);
1475 err = do_execute_actions(dp, skb, clone,
1479 __this_cpu_dec(exec_actions_level);
1480 } else { /* Recirc action */
1481 clone->recirc_id = recirc_id;
1482 ovs_dp_process_packet(skb, clone);
1487 /* Out of 'flow_keys' space. Defer actions */
1488 da = add_deferred_actions(skb, key, actions, len);
1490 if (!actions) { /* Recirc action */
1492 key->recirc_id = recirc_id;
1495 /* Out of per CPU action FIFO space. Drop the 'skb' and
1500 if (net_ratelimit()) {
1501 if (actions) { /* Sample action */
1502 pr_warn("%s: deferred action limit reached, drop sample action\n",
1504 } else { /* Recirc action */
1505 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1513 static void process_deferred_actions(struct datapath *dp)
1515 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1517 /* Do not touch the FIFO in case there is no deferred actions. */
1518 if (action_fifo_is_empty(fifo))
1521 /* Finishing executing all deferred actions. */
1523 struct deferred_action *da = action_fifo_get(fifo);
1524 struct sk_buff *skb = da->skb;
1525 struct sw_flow_key *key = &da->pkt_key;
1526 const struct nlattr *actions = da->actions;
1527 int actions_len = da->actions_len;
1530 do_execute_actions(dp, skb, key, actions, actions_len);
1532 ovs_dp_process_packet(skb, key);
1533 } while (!action_fifo_is_empty(fifo));
1535 /* Reset FIFO for the next packet. */
1536 action_fifo_init(fifo);
1539 /* Execute a list of actions against 'skb'. */
1540 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1541 const struct sw_flow_actions *acts,
1542 struct sw_flow_key *key)
1546 level = __this_cpu_inc_return(exec_actions_level);
1547 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1548 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1555 OVS_CB(skb)->acts_origlen = acts->orig_len;
1556 err = do_execute_actions(dp, skb, key,
1557 acts->actions, acts->actions_len);
1560 process_deferred_actions(dp);
1563 __this_cpu_dec(exec_actions_level);
1567 int action_fifos_init(void)
1569 action_fifos = alloc_percpu(struct action_fifo);
1573 flow_keys = alloc_percpu(struct action_flow_keys);
1575 free_percpu(action_fifos);
1582 void action_fifos_exit(void)
1584 free_percpu(action_fifos);
1585 free_percpu(flow_keys);