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,
170 ovs_key_mac_proto(key) == MAC_PROTO_ETHERNET);
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 invalidate_flow_key(key);
192 static int set_mpls(struct sk_buff *skb, struct sw_flow_key *flow_key,
193 const __be32 *mpls_lse, const __be32 *mask)
195 struct mpls_shim_hdr *stack;
199 stack = mpls_hdr(skb);
200 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
201 err = skb_mpls_update_lse(skb, lse);
205 flow_key->mpls.lse[0] = lse;
209 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
213 err = skb_vlan_pop(skb);
214 if (skb_vlan_tag_present(skb)) {
215 invalidate_flow_key(key);
217 key->eth.vlan.tci = 0;
218 key->eth.vlan.tpid = 0;
223 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
224 const struct ovs_action_push_vlan *vlan)
226 if (skb_vlan_tag_present(skb)) {
227 invalidate_flow_key(key);
229 key->eth.vlan.tci = vlan->vlan_tci;
230 key->eth.vlan.tpid = vlan->vlan_tpid;
232 return skb_vlan_push(skb, vlan->vlan_tpid,
233 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
236 /* 'src' is already properly masked. */
237 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
239 u16 *dst = (u16 *)dst_;
240 const u16 *src = (const u16 *)src_;
241 const u16 *mask = (const u16 *)mask_;
243 OVS_SET_MASKED(dst[0], src[0], mask[0]);
244 OVS_SET_MASKED(dst[1], src[1], mask[1]);
245 OVS_SET_MASKED(dst[2], src[2], mask[2]);
248 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
249 const struct ovs_key_ethernet *key,
250 const struct ovs_key_ethernet *mask)
254 err = skb_ensure_writable(skb, ETH_HLEN);
258 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
260 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
262 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
265 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
267 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
268 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
272 /* pop_eth does not support VLAN packets as this action is never called
275 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
277 skb_pull_rcsum(skb, ETH_HLEN);
278 skb_reset_mac_header(skb);
279 skb_reset_mac_len(skb);
281 /* safe right before invalidate_flow_key */
282 key->mac_proto = MAC_PROTO_NONE;
283 invalidate_flow_key(key);
287 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
288 const struct ovs_action_push_eth *ethh)
292 /* Add the new Ethernet header */
293 if (skb_cow_head(skb, ETH_HLEN) < 0)
296 skb_push(skb, ETH_HLEN);
297 skb_reset_mac_header(skb);
298 skb_reset_mac_len(skb);
301 ether_addr_copy(hdr->h_source, ethh->addresses.eth_src);
302 ether_addr_copy(hdr->h_dest, ethh->addresses.eth_dst);
303 hdr->h_proto = skb->protocol;
305 skb_postpush_rcsum(skb, hdr, ETH_HLEN);
307 /* safe right before invalidate_flow_key */
308 key->mac_proto = MAC_PROTO_ETHERNET;
309 invalidate_flow_key(key);
313 static int push_nsh(struct sk_buff *skb, struct sw_flow_key *key,
314 const struct nshhdr *nh)
318 err = nsh_push(skb, nh);
322 /* safe right before invalidate_flow_key */
323 key->mac_proto = MAC_PROTO_NONE;
324 invalidate_flow_key(key);
328 static int pop_nsh(struct sk_buff *skb, struct sw_flow_key *key)
336 /* safe right before invalidate_flow_key */
337 if (skb->protocol == htons(ETH_P_TEB))
338 key->mac_proto = MAC_PROTO_ETHERNET;
340 key->mac_proto = MAC_PROTO_NONE;
341 invalidate_flow_key(key);
345 static void update_ip_l4_checksum(struct sk_buff *skb, struct iphdr *nh,
346 __be32 addr, __be32 new_addr)
348 int transport_len = skb->len - skb_transport_offset(skb);
350 if (nh->frag_off & htons(IP_OFFSET))
353 if (nh->protocol == IPPROTO_TCP) {
354 if (likely(transport_len >= sizeof(struct tcphdr)))
355 inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
356 addr, new_addr, true);
357 } else if (nh->protocol == IPPROTO_UDP) {
358 if (likely(transport_len >= sizeof(struct udphdr))) {
359 struct udphdr *uh = udp_hdr(skb);
361 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
362 inet_proto_csum_replace4(&uh->check, skb,
363 addr, new_addr, true);
365 uh->check = CSUM_MANGLED_0;
371 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
372 __be32 *addr, __be32 new_addr)
374 update_ip_l4_checksum(skb, nh, *addr, new_addr);
375 csum_replace4(&nh->check, *addr, new_addr);
380 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
381 __be32 addr[4], const __be32 new_addr[4])
383 int transport_len = skb->len - skb_transport_offset(skb);
385 if (l4_proto == NEXTHDR_TCP) {
386 if (likely(transport_len >= sizeof(struct tcphdr)))
387 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
388 addr, new_addr, true);
389 } else if (l4_proto == NEXTHDR_UDP) {
390 if (likely(transport_len >= sizeof(struct udphdr))) {
391 struct udphdr *uh = udp_hdr(skb);
393 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
394 inet_proto_csum_replace16(&uh->check, skb,
395 addr, new_addr, true);
397 uh->check = CSUM_MANGLED_0;
400 } else if (l4_proto == NEXTHDR_ICMP) {
401 if (likely(transport_len >= sizeof(struct icmp6hdr)))
402 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
403 skb, addr, new_addr, true);
407 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
408 const __be32 mask[4], __be32 masked[4])
410 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
411 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
412 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
413 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
416 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
417 __be32 addr[4], const __be32 new_addr[4],
418 bool recalculate_csum)
420 if (recalculate_csum)
421 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
424 memcpy(addr, new_addr, sizeof(__be32[4]));
427 static void set_ipv6_fl(struct ipv6hdr *nh, u32 fl, u32 mask)
429 /* Bits 21-24 are always unmasked, so this retains their values. */
430 OVS_SET_MASKED(nh->flow_lbl[0], (u8)(fl >> 16), (u8)(mask >> 16));
431 OVS_SET_MASKED(nh->flow_lbl[1], (u8)(fl >> 8), (u8)(mask >> 8));
432 OVS_SET_MASKED(nh->flow_lbl[2], (u8)fl, (u8)mask);
435 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
438 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
440 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
444 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
445 const struct ovs_key_ipv4 *key,
446 const struct ovs_key_ipv4 *mask)
452 err = skb_ensure_writable(skb, skb_network_offset(skb) +
453 sizeof(struct iphdr));
459 /* Setting an IP addresses is typically only a side effect of
460 * matching on them in the current userspace implementation, so it
461 * makes sense to check if the value actually changed.
463 if (mask->ipv4_src) {
464 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
466 if (unlikely(new_addr != nh->saddr)) {
467 set_ip_addr(skb, nh, &nh->saddr, new_addr);
468 flow_key->ipv4.addr.src = new_addr;
471 if (mask->ipv4_dst) {
472 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
474 if (unlikely(new_addr != nh->daddr)) {
475 set_ip_addr(skb, nh, &nh->daddr, new_addr);
476 flow_key->ipv4.addr.dst = new_addr;
479 if (mask->ipv4_tos) {
480 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
481 flow_key->ip.tos = nh->tos;
483 if (mask->ipv4_ttl) {
484 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
485 flow_key->ip.ttl = nh->ttl;
491 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
493 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
496 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
497 const struct ovs_key_ipv6 *key,
498 const struct ovs_key_ipv6 *mask)
503 err = skb_ensure_writable(skb, skb_network_offset(skb) +
504 sizeof(struct ipv6hdr));
510 /* Setting an IP addresses is typically only a side effect of
511 * matching on them in the current userspace implementation, so it
512 * makes sense to check if the value actually changed.
514 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
515 __be32 *saddr = (__be32 *)&nh->saddr;
518 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
520 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
521 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
523 memcpy(&flow_key->ipv6.addr.src, masked,
524 sizeof(flow_key->ipv6.addr.src));
527 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
528 unsigned int offset = 0;
529 int flags = IP6_FH_F_SKIP_RH;
530 bool recalc_csum = true;
531 __be32 *daddr = (__be32 *)&nh->daddr;
534 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
536 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
537 if (ipv6_ext_hdr(nh->nexthdr))
538 recalc_csum = (ipv6_find_hdr(skb, &offset,
543 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
545 memcpy(&flow_key->ipv6.addr.dst, masked,
546 sizeof(flow_key->ipv6.addr.dst));
549 if (mask->ipv6_tclass) {
550 ipv6_change_dsfield(nh, ~mask->ipv6_tclass, key->ipv6_tclass);
551 flow_key->ip.tos = ipv6_get_dsfield(nh);
553 if (mask->ipv6_label) {
554 set_ipv6_fl(nh, ntohl(key->ipv6_label),
555 ntohl(mask->ipv6_label));
556 flow_key->ipv6.label =
557 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
559 if (mask->ipv6_hlimit) {
560 OVS_SET_MASKED(nh->hop_limit, key->ipv6_hlimit,
562 flow_key->ip.ttl = nh->hop_limit;
567 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
568 const struct nlattr *a)
577 struct ovs_key_nsh key;
578 struct ovs_key_nsh mask;
580 err = nsh_key_from_nlattr(a, &key, &mask);
584 /* Make sure the NSH base header is there */
585 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
589 length = nsh_hdr_len(nh);
591 /* Make sure the whole NSH header is there */
592 err = skb_ensure_writable(skb, skb_network_offset(skb) +
598 skb_postpull_rcsum(skb, nh, length);
599 flags = nsh_get_flags(nh);
600 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
601 flow_key->nsh.base.flags = flags;
602 ttl = nsh_get_ttl(nh);
603 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
604 flow_key->nsh.base.ttl = ttl;
605 nsh_set_flags_and_ttl(nh, flags, ttl);
606 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
608 flow_key->nsh.base.path_hdr = nh->path_hdr;
609 switch (nh->mdtype) {
611 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
613 OVS_MASKED(nh->md1.context[i], key.context[i],
616 memcpy(flow_key->nsh.context, nh->md1.context,
617 sizeof(nh->md1.context));
620 memset(flow_key->nsh.context, 0,
621 sizeof(flow_key->nsh.context));
626 skb_postpush_rcsum(skb, nh, length);
630 /* Must follow skb_ensure_writable() since that can move the skb data. */
631 static void set_tp_port(struct sk_buff *skb, __be16 *port,
632 __be16 new_port, __sum16 *check)
634 inet_proto_csum_replace2(check, skb, *port, new_port, false);
638 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
639 const struct ovs_key_udp *key,
640 const struct ovs_key_udp *mask)
646 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
647 sizeof(struct udphdr));
652 /* Either of the masks is non-zero, so do not bother checking them. */
653 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
654 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
656 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
657 if (likely(src != uh->source)) {
658 set_tp_port(skb, &uh->source, src, &uh->check);
659 flow_key->tp.src = src;
661 if (likely(dst != uh->dest)) {
662 set_tp_port(skb, &uh->dest, dst, &uh->check);
663 flow_key->tp.dst = dst;
666 if (unlikely(!uh->check))
667 uh->check = CSUM_MANGLED_0;
671 flow_key->tp.src = src;
672 flow_key->tp.dst = dst;
680 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
681 const struct ovs_key_tcp *key,
682 const struct ovs_key_tcp *mask)
688 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
689 sizeof(struct tcphdr));
694 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
695 if (likely(src != th->source)) {
696 set_tp_port(skb, &th->source, src, &th->check);
697 flow_key->tp.src = src;
699 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
700 if (likely(dst != th->dest)) {
701 set_tp_port(skb, &th->dest, dst, &th->check);
702 flow_key->tp.dst = dst;
709 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
710 const struct ovs_key_sctp *key,
711 const struct ovs_key_sctp *mask)
713 unsigned int sctphoff = skb_transport_offset(skb);
715 __le32 old_correct_csum, new_csum, old_csum;
718 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
723 old_csum = sh->checksum;
724 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
726 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
727 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
729 new_csum = sctp_compute_cksum(skb, sctphoff);
731 /* Carry any checksum errors through. */
732 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
735 flow_key->tp.src = sh->source;
736 flow_key->tp.dst = sh->dest;
741 static int ovs_vport_output(struct net *net, struct sock *sk, struct sk_buff *skb)
743 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
744 struct vport *vport = data->vport;
746 if (skb_cow_head(skb, data->l2_len) < 0) {
751 __skb_dst_copy(skb, data->dst);
752 *OVS_CB(skb) = data->cb;
753 skb->inner_protocol = data->inner_protocol;
754 if (data->vlan_tci & VLAN_CFI_MASK)
755 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
757 __vlan_hwaccel_clear_tag(skb);
759 /* Reconstruct the MAC header. */
760 skb_push(skb, data->l2_len);
761 memcpy(skb->data, &data->l2_data, data->l2_len);
762 skb_postpush_rcsum(skb, skb->data, data->l2_len);
763 skb_reset_mac_header(skb);
765 if (eth_p_mpls(skb->protocol)) {
766 skb->inner_network_header = skb->network_header;
767 skb_set_network_header(skb, data->network_offset);
768 skb_reset_mac_len(skb);
771 ovs_vport_send(vport, skb, data->mac_proto);
776 ovs_dst_get_mtu(const struct dst_entry *dst)
778 return dst->dev->mtu;
781 static struct dst_ops ovs_dst_ops = {
783 .mtu = ovs_dst_get_mtu,
786 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
787 * ovs_vport_output(), which is called once per fragmented packet.
789 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
790 u16 orig_network_offset, u8 mac_proto)
792 unsigned int hlen = skb_network_offset(skb);
793 struct ovs_frag_data *data;
795 data = this_cpu_ptr(&ovs_frag_data_storage);
796 data->dst = skb->_skb_refdst;
798 data->cb = *OVS_CB(skb);
799 data->inner_protocol = skb->inner_protocol;
800 data->network_offset = orig_network_offset;
801 if (skb_vlan_tag_present(skb))
802 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
805 data->vlan_proto = skb->vlan_proto;
806 data->mac_proto = mac_proto;
808 memcpy(&data->l2_data, skb->data, hlen);
810 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
814 static void ovs_fragment(struct net *net, struct vport *vport,
815 struct sk_buff *skb, u16 mru,
816 struct sw_flow_key *key)
818 u16 orig_network_offset = 0;
820 if (eth_p_mpls(skb->protocol)) {
821 orig_network_offset = skb_network_offset(skb);
822 skb->network_header = skb->inner_network_header;
825 if (skb_network_offset(skb) > MAX_L2_LEN) {
826 OVS_NLERR(1, "L2 header too long to fragment");
830 if (key->eth.type == htons(ETH_P_IP)) {
831 struct dst_entry ovs_dst;
832 unsigned long orig_dst;
834 prepare_frag(vport, skb, orig_network_offset,
835 ovs_key_mac_proto(key));
836 dst_init(&ovs_dst, &ovs_dst_ops, NULL, 1,
837 DST_OBSOLETE_NONE, DST_NOCOUNT);
838 ovs_dst.dev = vport->dev;
840 orig_dst = skb->_skb_refdst;
841 skb_dst_set_noref(skb, &ovs_dst);
842 IPCB(skb)->frag_max_size = mru;
844 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
845 refdst_drop(orig_dst);
846 } else if (key->eth.type == htons(ETH_P_IPV6)) {
847 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
848 unsigned long orig_dst;
849 struct rt6_info ovs_rt;
854 prepare_frag(vport, skb, orig_network_offset,
855 ovs_key_mac_proto(key));
856 memset(&ovs_rt, 0, sizeof(ovs_rt));
857 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
858 DST_OBSOLETE_NONE, DST_NOCOUNT);
859 ovs_rt.dst.dev = vport->dev;
861 orig_dst = skb->_skb_refdst;
862 skb_dst_set_noref(skb, &ovs_rt.dst);
863 IP6CB(skb)->frag_max_size = mru;
865 v6ops->fragment(net, skb->sk, skb, ovs_vport_output);
866 refdst_drop(orig_dst);
868 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
869 ovs_vport_name(vport), ntohs(key->eth.type), mru,
879 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
880 struct sw_flow_key *key)
882 struct vport *vport = ovs_vport_rcu(dp, out_port);
885 u16 mru = OVS_CB(skb)->mru;
886 u32 cutlen = OVS_CB(skb)->cutlen;
888 if (unlikely(cutlen > 0)) {
889 if (skb->len - cutlen > ovs_mac_header_len(key))
890 pskb_trim(skb, skb->len - cutlen);
892 pskb_trim(skb, ovs_mac_header_len(key));
896 (skb->len <= mru + vport->dev->hard_header_len))) {
897 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
898 } else if (mru <= vport->dev->mtu) {
899 struct net *net = read_pnet(&dp->net);
901 ovs_fragment(net, vport, skb, mru, key);
910 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
911 struct sw_flow_key *key, const struct nlattr *attr,
912 const struct nlattr *actions, int actions_len,
915 struct dp_upcall_info upcall;
916 const struct nlattr *a;
919 memset(&upcall, 0, sizeof(upcall));
920 upcall.cmd = OVS_PACKET_CMD_ACTION;
921 upcall.mru = OVS_CB(skb)->mru;
923 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
924 a = nla_next(a, &rem)) {
925 switch (nla_type(a)) {
926 case OVS_USERSPACE_ATTR_USERDATA:
930 case OVS_USERSPACE_ATTR_PID:
931 upcall.portid = nla_get_u32(a);
934 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
935 /* Get out tunnel info. */
938 vport = ovs_vport_rcu(dp, nla_get_u32(a));
942 err = dev_fill_metadata_dst(vport->dev, skb);
944 upcall.egress_tun_info = skb_tunnel_info(skb);
950 case OVS_USERSPACE_ATTR_ACTIONS: {
951 /* Include actions. */
952 upcall.actions = actions;
953 upcall.actions_len = actions_len;
957 } /* End of switch. */
960 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
963 /* When 'last' is true, sample() should always consume the 'skb'.
964 * Otherwise, sample() should keep 'skb' intact regardless what
965 * actions are executed within sample().
967 static int sample(struct datapath *dp, struct sk_buff *skb,
968 struct sw_flow_key *key, const struct nlattr *attr,
971 struct nlattr *actions;
972 struct nlattr *sample_arg;
973 int rem = nla_len(attr);
974 const struct sample_arg *arg;
977 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
978 sample_arg = nla_data(attr);
979 arg = nla_data(sample_arg);
980 actions = nla_next(sample_arg, &rem);
982 if ((arg->probability != U32_MAX) &&
983 (!arg->probability || prandom_u32() > arg->probability)) {
989 clone_flow_key = !arg->exec;
990 return clone_execute(dp, skb, key, 0, actions, rem, last,
994 /* When 'last' is true, clone() should always consume the 'skb'.
995 * Otherwise, clone() should keep 'skb' intact regardless what
996 * actions are executed within clone().
998 static int clone(struct datapath *dp, struct sk_buff *skb,
999 struct sw_flow_key *key, const struct nlattr *attr,
1002 struct nlattr *actions;
1003 struct nlattr *clone_arg;
1004 int rem = nla_len(attr);
1005 bool dont_clone_flow_key;
1007 /* The first action is always 'OVS_CLONE_ATTR_ARG'. */
1008 clone_arg = nla_data(attr);
1009 dont_clone_flow_key = nla_get_u32(clone_arg);
1010 actions = nla_next(clone_arg, &rem);
1012 return clone_execute(dp, skb, key, 0, actions, rem, last,
1013 !dont_clone_flow_key);
1016 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1017 const struct nlattr *attr)
1019 struct ovs_action_hash *hash_act = nla_data(attr);
1022 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1023 hash = skb_get_hash(skb);
1024 hash = jhash_1word(hash, hash_act->hash_basis);
1028 key->ovs_flow_hash = hash;
1031 static int execute_set_action(struct sk_buff *skb,
1032 struct sw_flow_key *flow_key,
1033 const struct nlattr *a)
1035 /* Only tunnel set execution is supported without a mask. */
1036 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1037 struct ovs_tunnel_info *tun = nla_data(a);
1040 dst_hold((struct dst_entry *)tun->tun_dst);
1041 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1048 /* Mask is at the midpoint of the data. */
1049 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1051 static int execute_masked_set_action(struct sk_buff *skb,
1052 struct sw_flow_key *flow_key,
1053 const struct nlattr *a)
1057 switch (nla_type(a)) {
1058 case OVS_KEY_ATTR_PRIORITY:
1059 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1060 *get_mask(a, u32 *));
1061 flow_key->phy.priority = skb->priority;
1064 case OVS_KEY_ATTR_SKB_MARK:
1065 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1066 flow_key->phy.skb_mark = skb->mark;
1069 case OVS_KEY_ATTR_TUNNEL_INFO:
1070 /* Masked data not supported for tunnel. */
1074 case OVS_KEY_ATTR_ETHERNET:
1075 err = set_eth_addr(skb, flow_key, nla_data(a),
1076 get_mask(a, struct ovs_key_ethernet *));
1079 case OVS_KEY_ATTR_NSH:
1080 err = set_nsh(skb, flow_key, a);
1083 case OVS_KEY_ATTR_IPV4:
1084 err = set_ipv4(skb, flow_key, nla_data(a),
1085 get_mask(a, struct ovs_key_ipv4 *));
1088 case OVS_KEY_ATTR_IPV6:
1089 err = set_ipv6(skb, flow_key, nla_data(a),
1090 get_mask(a, struct ovs_key_ipv6 *));
1093 case OVS_KEY_ATTR_TCP:
1094 err = set_tcp(skb, flow_key, nla_data(a),
1095 get_mask(a, struct ovs_key_tcp *));
1098 case OVS_KEY_ATTR_UDP:
1099 err = set_udp(skb, flow_key, nla_data(a),
1100 get_mask(a, struct ovs_key_udp *));
1103 case OVS_KEY_ATTR_SCTP:
1104 err = set_sctp(skb, flow_key, nla_data(a),
1105 get_mask(a, struct ovs_key_sctp *));
1108 case OVS_KEY_ATTR_MPLS:
1109 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1113 case OVS_KEY_ATTR_CT_STATE:
1114 case OVS_KEY_ATTR_CT_ZONE:
1115 case OVS_KEY_ATTR_CT_MARK:
1116 case OVS_KEY_ATTR_CT_LABELS:
1117 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1118 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1126 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1127 struct sw_flow_key *key,
1128 const struct nlattr *a, bool last)
1132 if (!is_flow_key_valid(key)) {
1135 err = ovs_flow_key_update(skb, key);
1139 BUG_ON(!is_flow_key_valid(key));
1141 recirc_id = nla_get_u32(a);
1142 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1145 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1146 struct sw_flow_key *key,
1147 const struct nlattr *attr, bool last)
1149 const struct nlattr *actions, *cpl_arg;
1150 const struct check_pkt_len_arg *arg;
1151 int rem = nla_len(attr);
1152 bool clone_flow_key;
1154 /* The first netlink attribute in 'attr' is always
1155 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1157 cpl_arg = nla_data(attr);
1158 arg = nla_data(cpl_arg);
1160 if (skb->len <= arg->pkt_len) {
1161 /* Second netlink attribute in 'attr' is always
1162 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1164 actions = nla_next(cpl_arg, &rem);
1165 clone_flow_key = !arg->exec_for_lesser_equal;
1167 /* Third netlink attribute in 'attr' is always
1168 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1170 actions = nla_next(cpl_arg, &rem);
1171 actions = nla_next(actions, &rem);
1172 clone_flow_key = !arg->exec_for_greater;
1175 return clone_execute(dp, skb, key, 0, nla_data(actions),
1176 nla_len(actions), last, clone_flow_key);
1179 /* Execute a list of actions against 'skb'. */
1180 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1181 struct sw_flow_key *key,
1182 const struct nlattr *attr, int len)
1184 const struct nlattr *a;
1187 for (a = attr, rem = len; rem > 0;
1188 a = nla_next(a, &rem)) {
1191 switch (nla_type(a)) {
1192 case OVS_ACTION_ATTR_OUTPUT: {
1193 int port = nla_get_u32(a);
1194 struct sk_buff *clone;
1196 /* Every output action needs a separate clone
1197 * of 'skb', In case the output action is the
1198 * last action, cloning can be avoided.
1200 if (nla_is_last(a, rem)) {
1201 do_output(dp, skb, port, key);
1202 /* 'skb' has been used for output.
1207 clone = skb_clone(skb, GFP_ATOMIC);
1209 do_output(dp, clone, port, key);
1210 OVS_CB(skb)->cutlen = 0;
1214 case OVS_ACTION_ATTR_TRUNC: {
1215 struct ovs_action_trunc *trunc = nla_data(a);
1217 if (skb->len > trunc->max_len)
1218 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1222 case OVS_ACTION_ATTR_USERSPACE:
1223 output_userspace(dp, skb, key, a, attr,
1224 len, OVS_CB(skb)->cutlen);
1225 OVS_CB(skb)->cutlen = 0;
1228 case OVS_ACTION_ATTR_HASH:
1229 execute_hash(skb, key, a);
1232 case OVS_ACTION_ATTR_PUSH_MPLS:
1233 err = push_mpls(skb, key, nla_data(a));
1236 case OVS_ACTION_ATTR_POP_MPLS:
1237 err = pop_mpls(skb, key, nla_get_be16(a));
1240 case OVS_ACTION_ATTR_PUSH_VLAN:
1241 err = push_vlan(skb, key, nla_data(a));
1244 case OVS_ACTION_ATTR_POP_VLAN:
1245 err = pop_vlan(skb, key);
1248 case OVS_ACTION_ATTR_RECIRC: {
1249 bool last = nla_is_last(a, rem);
1251 err = execute_recirc(dp, skb, key, a, last);
1253 /* If this is the last action, the skb has
1254 * been consumed or freed.
1255 * Return immediately.
1262 case OVS_ACTION_ATTR_SET:
1263 err = execute_set_action(skb, key, nla_data(a));
1266 case OVS_ACTION_ATTR_SET_MASKED:
1267 case OVS_ACTION_ATTR_SET_TO_MASKED:
1268 err = execute_masked_set_action(skb, key, nla_data(a));
1271 case OVS_ACTION_ATTR_SAMPLE: {
1272 bool last = nla_is_last(a, rem);
1274 err = sample(dp, skb, key, a, last);
1281 case OVS_ACTION_ATTR_CT:
1282 if (!is_flow_key_valid(key)) {
1283 err = ovs_flow_key_update(skb, key);
1288 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1291 /* Hide stolen IP fragments from user space. */
1293 return err == -EINPROGRESS ? 0 : err;
1296 case OVS_ACTION_ATTR_CT_CLEAR:
1297 err = ovs_ct_clear(skb, key);
1300 case OVS_ACTION_ATTR_PUSH_ETH:
1301 err = push_eth(skb, key, nla_data(a));
1304 case OVS_ACTION_ATTR_POP_ETH:
1305 err = pop_eth(skb, key);
1308 case OVS_ACTION_ATTR_PUSH_NSH: {
1309 u8 buffer[NSH_HDR_MAX_LEN];
1310 struct nshhdr *nh = (struct nshhdr *)buffer;
1312 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1316 err = push_nsh(skb, key, nh);
1320 case OVS_ACTION_ATTR_POP_NSH:
1321 err = pop_nsh(skb, key);
1324 case OVS_ACTION_ATTR_METER:
1325 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1331 case OVS_ACTION_ATTR_CLONE: {
1332 bool last = nla_is_last(a, rem);
1334 err = clone(dp, skb, key, a, last);
1341 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1342 bool last = nla_is_last(a, rem);
1344 err = execute_check_pkt_len(dp, skb, key, a, last);
1352 if (unlikely(err)) {
1362 /* Execute the actions on the clone of the packet. The effect of the
1363 * execution does not affect the original 'skb' nor the original 'key'.
1365 * The execution may be deferred in case the actions can not be executed
1368 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1369 struct sw_flow_key *key, u32 recirc_id,
1370 const struct nlattr *actions, int len,
1371 bool last, bool clone_flow_key)
1373 struct deferred_action *da;
1374 struct sw_flow_key *clone;
1376 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1378 /* Out of memory, skip this action.
1383 /* When clone_flow_key is false, the 'key' will not be change
1384 * by the actions, then the 'key' can be used directly.
1385 * Otherwise, try to clone key from the next recursion level of
1386 * 'flow_keys'. If clone is successful, execute the actions
1387 * without deferring.
1389 clone = clone_flow_key ? clone_key(key) : key;
1393 if (actions) { /* Sample action */
1395 __this_cpu_inc(exec_actions_level);
1397 err = do_execute_actions(dp, skb, clone,
1401 __this_cpu_dec(exec_actions_level);
1402 } else { /* Recirc action */
1403 clone->recirc_id = recirc_id;
1404 ovs_dp_process_packet(skb, clone);
1409 /* Out of 'flow_keys' space. Defer actions */
1410 da = add_deferred_actions(skb, key, actions, len);
1412 if (!actions) { /* Recirc action */
1414 key->recirc_id = recirc_id;
1417 /* Out of per CPU action FIFO space. Drop the 'skb' and
1422 if (net_ratelimit()) {
1423 if (actions) { /* Sample action */
1424 pr_warn("%s: deferred action limit reached, drop sample action\n",
1426 } else { /* Recirc action */
1427 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1435 static void process_deferred_actions(struct datapath *dp)
1437 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1439 /* Do not touch the FIFO in case there is no deferred actions. */
1440 if (action_fifo_is_empty(fifo))
1443 /* Finishing executing all deferred actions. */
1445 struct deferred_action *da = action_fifo_get(fifo);
1446 struct sk_buff *skb = da->skb;
1447 struct sw_flow_key *key = &da->pkt_key;
1448 const struct nlattr *actions = da->actions;
1449 int actions_len = da->actions_len;
1452 do_execute_actions(dp, skb, key, actions, actions_len);
1454 ovs_dp_process_packet(skb, key);
1455 } while (!action_fifo_is_empty(fifo));
1457 /* Reset FIFO for the next packet. */
1458 action_fifo_init(fifo);
1461 /* Execute a list of actions against 'skb'. */
1462 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1463 const struct sw_flow_actions *acts,
1464 struct sw_flow_key *key)
1468 level = __this_cpu_inc_return(exec_actions_level);
1469 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1470 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1477 OVS_CB(skb)->acts_origlen = acts->orig_len;
1478 err = do_execute_actions(dp, skb, key,
1479 acts->actions, acts->actions_len);
1482 process_deferred_actions(dp);
1485 __this_cpu_dec(exec_actions_level);
1489 int action_fifos_init(void)
1491 action_fifos = alloc_percpu(struct action_fifo);
1495 flow_keys = alloc_percpu(struct action_flow_keys);
1497 free_percpu(action_fifos);
1504 void action_fifos_exit(void)
1506 free_percpu(action_fifos);
1507 free_percpu(flow_keys);