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/netfilter_ipv6.h>
13 #include <linux/sctp.h>
14 #include <linux/tcp.h>
15 #include <linux/udp.h>
16 #include <linux/in6.h>
17 #include <linux/if_arp.h>
18 #include <linux/if_vlan.h>
23 #include <net/ip6_fib.h>
24 #include <net/checksum.h>
25 #include <net/dsfield.h>
27 #include <net/sctp/checksum.h>
31 #include "conntrack.h"
33 #include "flow_netlink.h"
35 struct deferred_action {
37 const struct nlattr *actions;
40 /* Store pkt_key clone when creating deferred action. */
41 struct sw_flow_key pkt_key;
44 #define MAX_L2_LEN (VLAN_ETH_HLEN + 3 * MPLS_HLEN)
45 struct ovs_frag_data {
49 __be16 inner_protocol;
50 u16 network_offset; /* valid only for MPLS */
55 u8 l2_data[MAX_L2_LEN];
58 static DEFINE_PER_CPU(struct ovs_frag_data, ovs_frag_data_storage);
60 #define DEFERRED_ACTION_FIFO_SIZE 10
61 #define OVS_RECURSION_LIMIT 5
62 #define OVS_DEFERRED_ACTION_THRESHOLD (OVS_RECURSION_LIMIT - 2)
66 /* Deferred action fifo queue storage. */
67 struct deferred_action fifo[DEFERRED_ACTION_FIFO_SIZE];
70 struct action_flow_keys {
71 struct sw_flow_key key[OVS_DEFERRED_ACTION_THRESHOLD];
74 static struct action_fifo __percpu *action_fifos;
75 static struct action_flow_keys __percpu *flow_keys;
76 static DEFINE_PER_CPU(int, exec_actions_level);
78 /* Make a clone of the 'key', using the pre-allocated percpu 'flow_keys'
79 * space. Return NULL if out of key spaces.
81 static struct sw_flow_key *clone_key(const struct sw_flow_key *key_)
83 struct action_flow_keys *keys = this_cpu_ptr(flow_keys);
84 int level = this_cpu_read(exec_actions_level);
85 struct sw_flow_key *key = NULL;
87 if (level <= OVS_DEFERRED_ACTION_THRESHOLD) {
88 key = &keys->key[level - 1];
95 static void action_fifo_init(struct action_fifo *fifo)
101 static bool action_fifo_is_empty(const struct action_fifo *fifo)
103 return (fifo->head == fifo->tail);
106 static struct deferred_action *action_fifo_get(struct action_fifo *fifo)
108 if (action_fifo_is_empty(fifo))
111 return &fifo->fifo[fifo->tail++];
114 static struct deferred_action *action_fifo_put(struct action_fifo *fifo)
116 if (fifo->head >= DEFERRED_ACTION_FIFO_SIZE - 1)
119 return &fifo->fifo[fifo->head++];
122 /* Return true if fifo is not full */
123 static struct deferred_action *add_deferred_actions(struct sk_buff *skb,
124 const struct sw_flow_key *key,
125 const struct nlattr *actions,
126 const int actions_len)
128 struct action_fifo *fifo;
129 struct deferred_action *da;
131 fifo = this_cpu_ptr(action_fifos);
132 da = action_fifo_put(fifo);
135 da->actions = actions;
136 da->actions_len = actions_len;
143 static void invalidate_flow_key(struct sw_flow_key *key)
145 key->mac_proto |= SW_FLOW_KEY_INVALID;
148 static bool is_flow_key_valid(const struct sw_flow_key *key)
150 return !(key->mac_proto & SW_FLOW_KEY_INVALID);
153 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
154 struct sw_flow_key *key,
156 const struct nlattr *actions, int len,
157 bool last, bool clone_flow_key);
159 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
160 struct sw_flow_key *key,
161 const struct nlattr *attr, int len);
163 static int push_mpls(struct sk_buff *skb, struct sw_flow_key *key,
164 const struct ovs_action_push_mpls *mpls)
168 err = skb_mpls_push(skb, mpls->mpls_lse, mpls->mpls_ethertype,
173 invalidate_flow_key(key);
177 static int pop_mpls(struct sk_buff *skb, struct sw_flow_key *key,
178 const __be16 ethertype)
182 err = skb_mpls_pop(skb, ethertype, skb->mac_len);
186 invalidate_flow_key(key);
190 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
191 const __be32 *mpls_lse, const __be32 *mask)
193 struct mpls_shim_hdr *stack;
197 stack = mpls_hdr(skb);
198 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
199 err = skb_mpls_update_lse(skb, lse);
203 flow_key->mpls.lse[0] = lse;
207 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
211 err = skb_vlan_pop(skb);
212 if (skb_vlan_tag_present(skb)) {
213 invalidate_flow_key(key);
215 key->eth.vlan.tci = 0;
216 key->eth.vlan.tpid = 0;
221 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
222 const struct ovs_action_push_vlan *vlan)
224 if (skb_vlan_tag_present(skb)) {
225 invalidate_flow_key(key);
227 key->eth.vlan.tci = vlan->vlan_tci;
228 key->eth.vlan.tpid = vlan->vlan_tpid;
230 return skb_vlan_push(skb, vlan->vlan_tpid,
231 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
234 /* 'src' is already properly masked. */
235 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
237 u16 *dst = (u16 *)dst_;
238 const u16 *src = (const u16 *)src_;
239 const u16 *mask = (const u16 *)mask_;
241 OVS_SET_MASKED(dst[0], src[0], mask[0]);
242 OVS_SET_MASKED(dst[1], src[1], mask[1]);
243 OVS_SET_MASKED(dst[2], src[2], mask[2]);
246 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
247 const struct ovs_key_ethernet *key,
248 const struct ovs_key_ethernet *mask)
252 err = skb_ensure_writable(skb, ETH_HLEN);
256 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
258 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
260 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
263 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
265 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
266 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
270 /* pop_eth does not support VLAN packets as this action is never called
273 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
275 skb_pull_rcsum(skb, ETH_HLEN);
276 skb_reset_mac_header(skb);
277 skb_reset_mac_len(skb);
279 /* safe right before invalidate_flow_key */
280 key->mac_proto = MAC_PROTO_NONE;
281 invalidate_flow_key(key);
285 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
286 const struct ovs_action_push_eth *ethh)
290 /* Add the new Ethernet header */
291 if (skb_cow_head(skb, ETH_HLEN) < 0)
294 skb_push(skb, ETH_HLEN);
295 skb_reset_mac_header(skb);
296 skb_reset_mac_len(skb);
299 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
300 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
301 hdr->h_proto = skb->protocol;
303 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
305 /* safe right before invalidate_flow_key */
306 key->mac_proto = MAC_PROTO_ETHERNET;
307 invalidate_flow_key(key);
311 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
312 const struct nshhdr *nh)
316 err = nsh_push(skb, nh);
320 /* safe right before invalidate_flow_key */
321 key->mac_proto = MAC_PROTO_NONE;
322 invalidate_flow_key(key);
326 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
334 /* safe right before invalidate_flow_key */
335 if (skb->protocol == htons(ETH_P_TEB))
336 key->mac_proto = MAC_PROTO_ETHERNET;
338 key->mac_proto = MAC_PROTO_NONE;
339 invalidate_flow_key(key);
343 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
344 __be32 addr, __be32 new_addr)
346 int transport_len = skb->len - skb_transport_offset(skb);
348 if (nh->frag_off & htons(IP_OFFSET))
351 if (nh->protocol == IPPROTO_TCP) {
352 if (likely(transport_len >= sizeof(struct tcphdr)))
353 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
354 addr, new_addr, true);
355 } else if (nh->protocol == IPPROTO_UDP) {
356 if (likely(transport_len >= sizeof(struct udphdr))) {
357 struct udphdr *uh = udp_hdr(skb);
359 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
360 inet_proto_csum_replace4(&uh->check, skb,
361 addr, new_addr, true);
363 uh->check = CSUM_MANGLED_0;
369 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
370 __be32 *addr, __be32 new_addr)
372 update_ip_l4_checksum(skb, nh, *addr, new_addr);
373 csum_replace4(&nh->check, *addr, new_addr);
378 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
379 __be32 addr[4], const __be32 new_addr[4])
381 int transport_len = skb->len - skb_transport_offset(skb);
383 if (l4_proto == NEXTHDR_TCP) {
384 if (likely(transport_len >= sizeof(struct tcphdr)))
385 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
386 addr, new_addr, true);
387 } else if (l4_proto == NEXTHDR_UDP) {
388 if (likely(transport_len >= sizeof(struct udphdr))) {
389 struct udphdr *uh = udp_hdr(skb);
391 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
392 inet_proto_csum_replace16(&uh->check, skb,
393 addr, new_addr, true);
395 uh->check = CSUM_MANGLED_0;
398 } else if (l4_proto == NEXTHDR_ICMP) {
399 if (likely(transport_len >= sizeof(struct icmp6hdr)))
400 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
401 skb, addr, new_addr, true);
405 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
406 const __be32 mask[4], __be32 masked[4])
408 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
409 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
410 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
411 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
414 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
415 __be32 addr[4], const __be32 new_addr[4],
416 bool recalculate_csum)
418 if (recalculate_csum)
419 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
422 memcpy(addr, new_addr, sizeof(__be32[4]));
425 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
427 /* Bits 21-24 are always unmasked, so this retains their values. */
428 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
429 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
430 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
433 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
436 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
438 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
442 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
443 const struct ovs_key_ipv4 *key,
444 const struct ovs_key_ipv4 *mask)
450 err = skb_ensure_writable(skb, skb_network_offset(skb) +
451 sizeof(struct iphdr));
457 /* Setting an IP addresses is typically only a side effect of
458 * matching on them in the current userspace implementation, so it
459 * makes sense to check if the value actually changed.
461 if (mask->ipv4_src) {
462 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
464 if (unlikely(new_addr != nh->saddr)) {
465 set_ip_addr(skb, nh, &nh->saddr, new_addr);
466 flow_key->ipv4.addr.src = new_addr;
469 if (mask->ipv4_dst) {
470 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
472 if (unlikely(new_addr != nh->daddr)) {
473 set_ip_addr(skb, nh, &nh->daddr, new_addr);
474 flow_key->ipv4.addr.dst = new_addr;
477 if (mask->ipv4_tos) {
478 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
479 flow_key->ip.tos = nh->tos;
481 if (mask->ipv4_ttl) {
482 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
483 flow_key->ip.ttl = nh->ttl;
489 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
491 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
494 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
495 const struct ovs_key_ipv6 *key,
496 const struct ovs_key_ipv6 *mask)
501 err = skb_ensure_writable(skb, skb_network_offset(skb) +
502 sizeof(struct ipv6hdr));
508 /* Setting an IP addresses is typically only a side effect of
509 * matching on them in the current userspace implementation, so it
510 * makes sense to check if the value actually changed.
512 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
513 __be32 *saddr = (__be32 *)&nh->saddr;
516 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
518 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
519 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
521 memcpy(&flow_key->ipv6.addr.src, masked,
522 sizeof(flow_key->ipv6.addr.src));
525 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
526 unsigned int offset = 0;
527 int flags = IP6_FH_F_SKIP_RH;
528 bool recalc_csum = true;
529 __be32 *daddr = (__be32 *)&nh->daddr;
532 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
534 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
535 if (ipv6_ext_hdr(nh->nexthdr))
536 recalc_csum = (ipv6_find_hdr(skb, &offset,
541 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
543 memcpy(&flow_key->ipv6.addr.dst, masked,
544 sizeof(flow_key->ipv6.addr.dst));
547 if (mask->ipv6_tclass) {
548 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
549 flow_key->ip.tos = ipv6_get_dsfield(nh);
551 if (mask->ipv6_label) {
552 set_ipv6_fl(nh, ntohl(key->ipv6_label),
553 ntohl(mask->ipv6_label));
554 flow_key->ipv6.label =
555 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
557 if (mask->ipv6_hlimit) {
558 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
560 flow_key->ip.ttl = nh->hop_limit;
565 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
566 const struct nlattr *a)
575 struct ovs_key_nsh key;
576 struct ovs_key_nsh mask;
578 err = nsh_key_from_nlattr(a, &key, &mask);
582 /* Make sure the NSH base header is there */
583 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
587 length = nsh_hdr_len(nh);
589 /* Make sure the whole NSH header is there */
590 err = skb_ensure_writable(skb, skb_network_offset(skb) +
596 skb_postpull_rcsum(skb, nh, length);
597 flags = nsh_get_flags(nh);
598 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
599 flow_key->nsh.base.flags = flags;
600 ttl = nsh_get_ttl(nh);
601 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
602 flow_key->nsh.base.ttl = ttl;
603 nsh_set_flags_and_ttl(nh, flags, ttl);
604 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
606 flow_key->nsh.base.path_hdr = nh->path_hdr;
607 switch (nh->mdtype) {
609 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
611 OVS_MASKED(nh->md1.context[i], key.context[i],
614 memcpy(flow_key->nsh.context, nh->md1.context,
615 sizeof(nh->md1.context));
618 memset(flow_key->nsh.context, 0,
619 sizeof(flow_key->nsh.context));
624 skb_postpush_rcsum(skb, nh, length);
628 /* Must follow skb_ensure_writable() since that can move the skb data. */
629 static void set_tp_port(struct sk_buff *skb, __be16 *port,
630 __be16 new_port, __sum16 *check)
632 inet_proto_csum_replace2(check, skb, *port, new_port, false);
636 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
637 const struct ovs_key_udp *key,
638 const struct ovs_key_udp *mask)
644 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
645 sizeof(struct udphdr));
650 /* Either of the masks is non-zero, so do not bother checking them. */
651 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
652 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
654 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
655 if (likely(src != uh->source)) {
656 set_tp_port(skb, &uh->source, src, &uh->check);
657 flow_key->tp.src = src;
659 if (likely(dst != uh->dest)) {
660 set_tp_port(skb, &uh->dest, dst, &uh->check);
661 flow_key->tp.dst = dst;
664 if (unlikely(!uh->check))
665 uh->check = CSUM_MANGLED_0;
669 flow_key->tp.src = src;
670 flow_key->tp.dst = dst;
678 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
679 const struct ovs_key_tcp *key,
680 const struct ovs_key_tcp *mask)
686 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
687 sizeof(struct tcphdr));
692 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
693 if (likely(src != th->source)) {
694 set_tp_port(skb, &th->source, src, &th->check);
695 flow_key->tp.src = src;
697 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
698 if (likely(dst != th->dest)) {
699 set_tp_port(skb, &th->dest, dst, &th->check);
700 flow_key->tp.dst = dst;
707 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
708 const struct ovs_key_sctp *key,
709 const struct ovs_key_sctp *mask)
711 unsigned int sctphoff = skb_transport_offset(skb);
713 __le32 old_correct_csum, new_csum, old_csum;
716 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
721 old_csum = sh->checksum;
722 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
724 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
725 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
727 new_csum = sctp_compute_cksum(skb, sctphoff);
729 /* Carry any checksum errors through. */
730 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
733 flow_key->tp.src = sh->source;
734 flow_key->tp.dst = sh->dest;
739 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
741 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
742 struct vport *vport = data->vport;
744 if (skb_cow_head(skb, data->l2_len) < 0) {
749 __skb_dst_copy(skb, data->dst);
750 *OVS_CB(skb) = data->cb;
751 skb->inner_protocol = data->inner_protocol;
752 if (data->vlan_tci & VLAN_CFI_MASK)
753 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
755 __vlan_hwaccel_clear_tag(skb);
757 /* Reconstruct the MAC header. */
758 skb_push(skb, data->l2_len);
759 memcpy(skb->data, &data->l2_data, data->l2_len);
760 skb_postpush_rcsum(skb, skb->data, data->l2_len);
761 skb_reset_mac_header(skb);
763 if (eth_p_mpls(skb->protocol)) {
764 skb->inner_network_header = skb->network_header;
765 skb_set_network_header(skb, data->network_offset);
766 skb_reset_mac_len(skb);
769 ovs_vport_send(vport, skb, data->mac_proto);
774 ovs_dst_get_mtu(const struct dst_entry *dst)
776 return dst->dev->mtu;
779 static struct dst_ops ovs_dst_ops = {
781 .mtu = ovs_dst_get_mtu,
784 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
785 * ovs_vport_output(), which is called once per fragmented packet.
787 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
788 u16 orig_network_offset, u8 mac_proto)
790 unsigned int hlen = skb_network_offset(skb);
791 struct ovs_frag_data *data;
793 data = this_cpu_ptr(&ovs_frag_data_storage);
794 data->dst = skb->_skb_refdst;
796 data->cb = *OVS_CB(skb);
797 data->inner_protocol = skb->inner_protocol;
798 data->network_offset = orig_network_offset;
799 if (skb_vlan_tag_present(skb))
800 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
803 data->vlan_proto = skb->vlan_proto;
804 data->mac_proto = mac_proto;
806 memcpy(&data->l2_data, skb->data, hlen);
808 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
812 static void ovs_fragment(struct net *net, struct vport *vport,
813 struct sk_buff *skb, u16 mru,
814 struct sw_flow_key *key)
816 u16 orig_network_offset = 0;
818 if (eth_p_mpls(skb->protocol)) {
819 orig_network_offset = skb_network_offset(skb);
820 skb->network_header = skb->inner_network_header;
823 if (skb_network_offset(skb) > MAX_L2_LEN) {
824 OVS_NLERR(1, "L2 header too long to fragment");
828 if (key->eth.type == htons(ETH_P_IP)) {
829 struct dst_entry ovs_dst;
830 unsigned long orig_dst;
832 prepare_frag(vport, skb, orig_network_offset,
833 ovs_key_mac_proto(key));
834 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
835 DST_OBSOLETE_NONE, DST_NOCOUNT);
836 ovs_dst.dev = vport->dev;
838 orig_dst = skb->_skb_refdst;
839 skb_dst_set_noref(skb, &ovs_dst);
840 IPCB(skb)->frag_max_size = mru;
842 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
843 refdst_drop(orig_dst);
844 } else if (key->eth.type == htons(ETH_P_IPV6)) {
845 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
846 unsigned long orig_dst;
847 struct rt6_info ovs_rt;
852 prepare_frag(vport, skb, orig_network_offset,
853 ovs_key_mac_proto(key));
854 memset(&ovs_rt, 0, sizeof(ovs_rt));
855 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
856 DST_OBSOLETE_NONE, DST_NOCOUNT);
857 ovs_rt.dst.dev = vport->dev;
859 orig_dst = skb->_skb_refdst;
860 skb_dst_set_noref(skb, &ovs_rt.dst);
861 IP6CB(skb)->frag_max_size = mru;
863 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
864 refdst_drop(orig_dst);
866 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
867 ovs_vport_name(vport), ntohs(key->eth.type), mru,
877 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
878 struct sw_flow_key *key)
880 struct vport *vport = ovs_vport_rcu(dp, out_port);
883 u16 mru = OVS_CB(skb)->mru;
884 u32 cutlen = OVS_CB(skb)->cutlen;
886 if (unlikely(cutlen > 0)) {
887 if (skb->len - cutlen > ovs_mac_header_len(key))
888 pskb_trim(skb, skb->len - cutlen);
890 pskb_trim(skb, ovs_mac_header_len(key));
894 (skb->len <= mru + vport->dev->hard_header_len))) {
895 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
896 } else if (mru <= vport->dev->mtu) {
897 struct net *net = read_pnet(&dp->net);
899 ovs_fragment(net, vport, skb, mru, key);
908 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
909 struct sw_flow_key *key, const struct nlattr *attr,
910 const struct nlattr *actions, int actions_len,
913 struct dp_upcall_info upcall;
914 const struct nlattr *a;
917 memset(&upcall, 0, sizeof(upcall));
918 upcall.cmd = OVS_PACKET_CMD_ACTION;
919 upcall.mru = OVS_CB(skb)->mru;
921 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
922 a = nla_next(a, &rem)) {
923 switch (nla_type(a)) {
924 case OVS_USERSPACE_ATTR_USERDATA:
928 case OVS_USERSPACE_ATTR_PID:
929 upcall.portid = nla_get_u32(a);
932 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
933 /* Get out tunnel info. */
936 vport = ovs_vport_rcu(dp, nla_get_u32(a));
940 err = dev_fill_metadata_dst(vport->dev, skb);
942 upcall.egress_tun_info = skb_tunnel_info(skb);
948 case OVS_USERSPACE_ATTR_ACTIONS: {
949 /* Include actions. */
950 upcall.actions = actions;
951 upcall.actions_len = actions_len;
955 } /* End of switch. */
958 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
961 /* When 'last' is true, sample() should always consume the 'skb'.
962 * Otherwise, sample() should keep 'skb' intact regardless what
963 * actions are executed within sample().
965 static int sample(struct datapath *dp, struct sk_buff *skb,
966 struct sw_flow_key *key, const struct nlattr *attr,
969 struct nlattr *actions;
970 struct nlattr *sample_arg;
971 int rem = nla_len(attr);
972 const struct sample_arg *arg;
975 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
976 sample_arg = nla_data(attr);
977 arg = nla_data(sample_arg);
978 actions = nla_next(sample_arg, &rem);
980 if ((arg->probability != U32_MAX) &&
981 (!arg->probability || prandom_u32() > arg->probability)) {
987 clone_flow_key = !arg->exec;
988 return clone_execute(dp, skb, key, 0, actions, rem, last,
992 /* When 'last' is true, clone() should always consume the 'skb'.
993 * Otherwise, clone() should keep 'skb' intact regardless what
994 * actions are executed within clone().
996 static int clone(struct datapath *dp, struct sk_buff *skb,
997 struct sw_flow_key *key, const struct nlattr *attr,
1000 struct nlattr *actions;
1001 struct nlattr *clone_arg;
1002 int rem = nla_len(attr);
1003 bool dont_clone_flow_key;
1005 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1006 clone_arg = nla_data(attr);
1007 dont_clone_flow_key = nla_get_u32(clone_arg);
1008 actions = nla_next(clone_arg, &rem);
1010 return clone_execute(dp, skb, key, 0, actions, rem, last,
1011 !dont_clone_flow_key);
1014 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1015 const struct nlattr *attr)
1017 struct ovs_action_hash *hash_act = nla_data(attr);
1020 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1021 hash = skb_get_hash(skb);
1022 hash = jhash_1word(hash, hash_act->hash_basis);
1026 key->ovs_flow_hash = hash;
1029 static int execute_set_action(struct sk_buff *skb,
1030 struct sw_flow_key *flow_key,
1031 const struct nlattr *a)
1033 /* Only tunnel set execution is supported without a mask. */
1034 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1035 struct ovs_tunnel_info *tun = nla_data(a);
1038 dst_hold((struct dst_entry *)tun->tun_dst);
1039 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1046 /* Mask is at the midpoint of the data. */
1047 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1049 static int execute_masked_set_action(struct sk_buff *skb,
1050 struct sw_flow_key *flow_key,
1051 const struct nlattr *a)
1055 switch (nla_type(a)) {
1056 case OVS_KEY_ATTR_PRIORITY:
1057 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1058 *get_mask(a, u32 *));
1059 flow_key->phy.priority = skb->priority;
1062 case OVS_KEY_ATTR_SKB_MARK:
1063 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1064 flow_key->phy.skb_mark = skb->mark;
1067 case OVS_KEY_ATTR_TUNNEL_INFO:
1068 /* Masked data not supported for tunnel. */
1072 case OVS_KEY_ATTR_ETHERNET:
1073 err = set_eth_addr(skb, flow_key, nla_data(a),
1074 get_mask(a, struct ovs_key_ethernet *));
1077 case OVS_KEY_ATTR_NSH:
1078 err = set_nsh(skb, flow_key, a);
1081 case OVS_KEY_ATTR_IPV4:
1082 err = set_ipv4(skb, flow_key, nla_data(a),
1083 get_mask(a, struct ovs_key_ipv4 *));
1086 case OVS_KEY_ATTR_IPV6:
1087 err = set_ipv6(skb, flow_key, nla_data(a),
1088 get_mask(a, struct ovs_key_ipv6 *));
1091 case OVS_KEY_ATTR_TCP:
1092 err = set_tcp(skb, flow_key, nla_data(a),
1093 get_mask(a, struct ovs_key_tcp *));
1096 case OVS_KEY_ATTR_UDP:
1097 err = set_udp(skb, flow_key, nla_data(a),
1098 get_mask(a, struct ovs_key_udp *));
1101 case OVS_KEY_ATTR_SCTP:
1102 err = set_sctp(skb, flow_key, nla_data(a),
1103 get_mask(a, struct ovs_key_sctp *));
1106 case OVS_KEY_ATTR_MPLS:
1107 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1111 case OVS_KEY_ATTR_CT_STATE:
1112 case OVS_KEY_ATTR_CT_ZONE:
1113 case OVS_KEY_ATTR_CT_MARK:
1114 case OVS_KEY_ATTR_CT_LABELS:
1115 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1116 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1124 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1125 struct sw_flow_key *key,
1126 const struct nlattr *a, bool last)
1130 if (!is_flow_key_valid(key)) {
1133 err = ovs_flow_key_update(skb, key);
1137 BUG_ON(!is_flow_key_valid(key));
1139 recirc_id = nla_get_u32(a);
1140 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1143 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1144 struct sw_flow_key *key,
1145 const struct nlattr *attr, bool last)
1147 const struct nlattr *actions, *cpl_arg;
1148 const struct check_pkt_len_arg *arg;
1149 int rem = nla_len(attr);
1150 bool clone_flow_key;
1152 /* The first netlink attribute in 'attr' is always
1153 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1155 cpl_arg = nla_data(attr);
1156 arg = nla_data(cpl_arg);
1158 if (skb->len <= arg->pkt_len) {
1159 /* Second netlink attribute in 'attr' is always
1160 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1162 actions = nla_next(cpl_arg, &rem);
1163 clone_flow_key = !arg->exec_for_lesser_equal;
1165 /* Third netlink attribute in 'attr' is always
1166 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1168 actions = nla_next(cpl_arg, &rem);
1169 actions = nla_next(actions, &rem);
1170 clone_flow_key = !arg->exec_for_greater;
1173 return clone_execute(dp, skb, key, 0, nla_data(actions),
1174 nla_len(actions), last, clone_flow_key);
1177 /* Execute a list of actions against 'skb'. */
1178 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1179 struct sw_flow_key *key,
1180 const struct nlattr *attr, int len)
1182 const struct nlattr *a;
1185 for (a = attr, rem = len; rem > 0;
1186 a = nla_next(a, &rem)) {
1189 switch (nla_type(a)) {
1190 case OVS_ACTION_ATTR_OUTPUT: {
1191 int port = nla_get_u32(a);
1192 struct sk_buff *clone;
1194 /* Every output action needs a separate clone
1195 * of 'skb', In case the output action is the
1196 * last action, cloning can be avoided.
1198 if (nla_is_last(a, rem)) {
1199 do_output(dp, skb, port, key);
1200 /* 'skb' has been used for output.
1205 clone = skb_clone(skb, GFP_ATOMIC);
1207 do_output(dp, clone, port, key);
1208 OVS_CB(skb)->cutlen = 0;
1212 case OVS_ACTION_ATTR_TRUNC: {
1213 struct ovs_action_trunc *trunc = nla_data(a);
1215 if (skb->len > trunc->max_len)
1216 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1220 case OVS_ACTION_ATTR_USERSPACE:
1221 output_userspace(dp, skb, key, a, attr,
1222 len, OVS_CB(skb)->cutlen);
1223 OVS_CB(skb)->cutlen = 0;
1226 case OVS_ACTION_ATTR_HASH:
1227 execute_hash(skb, key, a);
1230 case OVS_ACTION_ATTR_PUSH_MPLS:
1231 err = push_mpls(skb, key, nla_data(a));
1234 case OVS_ACTION_ATTR_POP_MPLS:
1235 err = pop_mpls(skb, key, nla_get_be16(a));
1238 case OVS_ACTION_ATTR_PUSH_VLAN:
1239 err = push_vlan(skb, key, nla_data(a));
1242 case OVS_ACTION_ATTR_POP_VLAN:
1243 err = pop_vlan(skb, key);
1246 case OVS_ACTION_ATTR_RECIRC: {
1247 bool last = nla_is_last(a, rem);
1249 err = execute_recirc(dp, skb, key, a, last);
1251 /* If this is the last action, the skb has
1252 * been consumed or freed.
1253 * Return immediately.
1260 case OVS_ACTION_ATTR_SET:
1261 err = execute_set_action(skb, key, nla_data(a));
1264 case OVS_ACTION_ATTR_SET_MASKED:
1265 case OVS_ACTION_ATTR_SET_TO_MASKED:
1266 err = execute_masked_set_action(skb, key, nla_data(a));
1269 case OVS_ACTION_ATTR_SAMPLE: {
1270 bool last = nla_is_last(a, rem);
1272 err = sample(dp, skb, key, a, last);
1279 case OVS_ACTION_ATTR_CT:
1280 if (!is_flow_key_valid(key)) {
1281 err = ovs_flow_key_update(skb, key);
1286 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1289 /* Hide stolen IP fragments from user space. */
1291 return err == -EINPROGRESS ? 0 : err;
1294 case OVS_ACTION_ATTR_CT_CLEAR:
1295 err = ovs_ct_clear(skb, key);
1298 case OVS_ACTION_ATTR_PUSH_ETH:
1299 err = push_eth(skb, key, nla_data(a));
1302 case OVS_ACTION_ATTR_POP_ETH:
1303 err = pop_eth(skb, key);
1306 case OVS_ACTION_ATTR_PUSH_NSH: {
1307 u8 buffer[NSH_HDR_MAX_LEN];
1308 struct nshhdr *nh = (struct nshhdr *)buffer;
1310 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1314 err = push_nsh(skb, key, nh);
1318 case OVS_ACTION_ATTR_POP_NSH:
1319 err = pop_nsh(skb, key);
1322 case OVS_ACTION_ATTR_METER:
1323 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1329 case OVS_ACTION_ATTR_CLONE: {
1330 bool last = nla_is_last(a, rem);
1332 err = clone(dp, skb, key, a, last);
1339 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1340 bool last = nla_is_last(a, rem);
1342 err = execute_check_pkt_len(dp, skb, key, a, last);
1350 if (unlikely(err)) {
1360 /* Execute the actions on the clone of the packet. The effect of the
1361 * execution does not affect the original 'skb' nor the original 'key'.
1363 * The execution may be deferred in case the actions can not be executed
1366 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1367 struct sw_flow_key *key, u32 recirc_id,
1368 const struct nlattr *actions, int len,
1369 bool last, bool clone_flow_key)
1371 struct deferred_action *da;
1372 struct sw_flow_key *clone;
1374 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1376 /* Out of memory, skip this action.
1381 /* When clone_flow_key is false, the 'key' will not be change
1382 * by the actions, then the 'key' can be used directly.
1383 * Otherwise, try to clone key from the next recursion level of
1384 * 'flow_keys'. If clone is successful, execute the actions
1385 * without deferring.
1387 clone = clone_flow_key ? clone_key(key) : key;
1391 if (actions) { /* Sample action */
1393 __this_cpu_inc(exec_actions_level);
1395 err = do_execute_actions(dp, skb, clone,
1399 __this_cpu_dec(exec_actions_level);
1400 } else { /* Recirc action */
1401 clone->recirc_id = recirc_id;
1402 ovs_dp_process_packet(skb, clone);
1407 /* Out of 'flow_keys' space. Defer actions */
1408 da = add_deferred_actions(skb, key, actions, len);
1410 if (!actions) { /* Recirc action */
1412 key->recirc_id = recirc_id;
1415 /* Out of per CPU action FIFO space. Drop the 'skb' and
1420 if (net_ratelimit()) {
1421 if (actions) { /* Sample action */
1422 pr_warn("%s: deferred action limit reached, drop sample action\n",
1424 } else { /* Recirc action */
1425 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1433 static void process_deferred_actions(struct datapath *dp)
1435 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1437 /* Do not touch the FIFO in case there is no deferred actions. */
1438 if (action_fifo_is_empty(fifo))
1441 /* Finishing executing all deferred actions. */
1443 struct deferred_action *da = action_fifo_get(fifo);
1444 struct sk_buff *skb = da->skb;
1445 struct sw_flow_key *key = &da->pkt_key;
1446 const struct nlattr *actions = da->actions;
1447 int actions_len = da->actions_len;
1450 do_execute_actions(dp, skb, key, actions, actions_len);
1452 ovs_dp_process_packet(skb, key);
1453 } while (!action_fifo_is_empty(fifo));
1455 /* Reset FIFO for the next packet. */
1456 action_fifo_init(fifo);
1459 /* Execute a list of actions against 'skb'. */
1460 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1461 const struct sw_flow_actions *acts,
1462 struct sw_flow_key *key)
1466 level = __this_cpu_inc_return(exec_actions_level);
1467 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1468 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1475 OVS_CB(skb)->acts_origlen = acts->orig_len;
1476 err = do_execute_actions(dp, skb, key,
1477 acts->actions, acts->actions_len);
1480 process_deferred_actions(dp);
1483 __this_cpu_dec(exec_actions_level);
1487 int action_fifos_init(void)
1489 action_fifos = alloc_percpu(struct action_fifo);
1493 flow_keys = alloc_percpu(struct action_flow_keys);
1495 free_percpu(action_fifos);
1502 void action_fifos_exit(void)
1504 free_percpu(action_fifos);
1505 free_percpu(flow_keys);