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 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
205 stack = mpls_hdr(skb);
206 lse = OVS_MASKED(stack->label_stack_entry, *mpls_lse, *mask);
207 err = skb_mpls_update_lse(skb, lse);
211 flow_key->mpls.lse[0] = lse;
215 static int pop_vlan(struct sk_buff *skb, struct sw_flow_key *key)
219 err = skb_vlan_pop(skb);
220 if (skb_vlan_tag_present(skb)) {
221 invalidate_flow_key(key);
223 key->eth.vlan.tci = 0;
224 key->eth.vlan.tpid = 0;
229 static int push_vlan(struct sk_buff *skb, struct sw_flow_key *key,
230 const struct ovs_action_push_vlan *vlan)
232 if (skb_vlan_tag_present(skb)) {
233 invalidate_flow_key(key);
235 key->eth.vlan.tci = vlan->vlan_tci;
236 key->eth.vlan.tpid = vlan->vlan_tpid;
238 return skb_vlan_push(skb, vlan->vlan_tpid,
239 ntohs(vlan->vlan_tci) & ~VLAN_CFI_MASK);
242 /* 'src' is already properly masked. */
243 static void ether_addr_copy_masked(u8 *dst_, const u8 *src_, const u8 *mask_)
245 u16 *dst = (u16 *)dst_;
246 const u16 *src = (const u16 *)src_;
247 const u16 *mask = (const u16 *)mask_;
249 OVS_SET_MASKED(dst[0], src[0], mask[0]);
250 OVS_SET_MASKED(dst[1], src[1], mask[1]);
251 OVS_SET_MASKED(dst[2], src[2], mask[2]);
254 static int set_eth_addr(struct sk_buff *skb, struct sw_flow_key *flow_key,
255 const struct ovs_key_ethernet *key,
256 const struct ovs_key_ethernet *mask)
260 err = skb_ensure_writable(skb, ETH_HLEN);
264 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
266 ether_addr_copy_masked(eth_hdr(skb)->h_source, key->eth_src,
268 ether_addr_copy_masked(eth_hdr(skb)->h_dest, key->eth_dst,
271 skb_postpush_rcsum(skb, eth_hdr(skb), ETH_ALEN * 2);
273 ether_addr_copy(flow_key->eth.src, eth_hdr(skb)->h_source);
274 ether_addr_copy(flow_key->eth.dst, eth_hdr(skb)->h_dest);
278 /* pop_eth does not support VLAN packets as this action is never called
281 static int pop_eth(struct sk_buff *skb, struct sw_flow_key *key)
285 err = skb_eth_pop(skb);
289 /* safe right before invalidate_flow_key */
290 key->mac_proto = MAC_PROTO_NONE;
291 invalidate_flow_key(key);
295 static int push_eth(struct sk_buff *skb, struct sw_flow_key *key,
296 const struct ovs_action_push_eth *ethh)
300 err = skb_eth_push(skb, ethh->addresses.eth_dst,
301 ethh->addresses.eth_src);
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);
375 ovs_ct_clear(skb, NULL);
379 static void update_ipv6_checksum(struct sk_buff *skb, u8 l4_proto,
380 __be32 addr[4], const __be32 new_addr[4])
382 int transport_len = skb->len - skb_transport_offset(skb);
384 if (l4_proto == NEXTHDR_TCP) {
385 if (likely(transport_len >= sizeof(struct tcphdr)))
386 inet_proto_csum_replace16(&tcp_hdr(skb)->check, skb,
387 addr, new_addr, true);
388 } else if (l4_proto == NEXTHDR_UDP) {
389 if (likely(transport_len >= sizeof(struct udphdr))) {
390 struct udphdr *uh = udp_hdr(skb);
392 if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
393 inet_proto_csum_replace16(&uh->check, skb,
394 addr, new_addr, true);
396 uh->check = CSUM_MANGLED_0;
399 } else if (l4_proto == NEXTHDR_ICMP) {
400 if (likely(transport_len >= sizeof(struct icmp6hdr)))
401 inet_proto_csum_replace16(&icmp6_hdr(skb)->icmp6_cksum,
402 skb, addr, new_addr, true);
406 static void mask_ipv6_addr(const __be32 old[4], const __be32 addr[4],
407 const __be32 mask[4], __be32 masked[4])
409 masked[0] = OVS_MASKED(old[0], addr[0], mask[0]);
410 masked[1] = OVS_MASKED(old[1], addr[1], mask[1]);
411 masked[2] = OVS_MASKED(old[2], addr[2], mask[2]);
412 masked[3] = OVS_MASKED(old[3], addr[3], mask[3]);
415 static void set_ipv6_addr(struct sk_buff *skb, u8 l4_proto,
416 __be32 addr[4], const __be32 new_addr[4],
417 bool recalculate_csum)
419 if (recalculate_csum)
420 update_ipv6_checksum(skb, l4_proto, addr, new_addr);
423 ovs_ct_clear(skb, NULL);
424 memcpy(addr, new_addr, sizeof(__be32[4]));
427 static void set_ipv6_dsfield(struct sk_buff *skb, struct ipv6hdr *nh, u8 ipv6_tclass, u8 mask)
429 u8 old_ipv6_tclass = ipv6_get_dsfield(nh);
431 ipv6_tclass = OVS_MASKED(old_ipv6_tclass, ipv6_tclass, mask);
433 if (skb->ip_summed == CHECKSUM_COMPLETE)
434 csum_replace(&skb->csum, (__force __wsum)(old_ipv6_tclass << 12),
435 (__force __wsum)(ipv6_tclass << 12));
437 ipv6_change_dsfield(nh, ~mask, ipv6_tclass);
440 static void set_ipv6_fl(struct sk_buff *skb, struct ipv6hdr *nh, u32 fl, u32 mask)
444 ofl = nh->flow_lbl[0] << 16 | nh->flow_lbl[1] << 8 | nh->flow_lbl[2];
445 fl = OVS_MASKED(ofl, fl, mask);
447 /* Bits 21-24 are always unmasked, so this retains their values. */
448 nh->flow_lbl[0] = (u8)(fl >> 16);
449 nh->flow_lbl[1] = (u8)(fl >> 8);
450 nh->flow_lbl[2] = (u8)fl;
452 if (skb->ip_summed == CHECKSUM_COMPLETE)
453 csum_replace(&skb->csum, (__force __wsum)htonl(ofl), (__force __wsum)htonl(fl));
456 static void set_ipv6_ttl(struct sk_buff *skb, struct ipv6hdr *nh, u8 new_ttl, u8 mask)
458 new_ttl = OVS_MASKED(nh->hop_limit, new_ttl, mask);
460 if (skb->ip_summed == CHECKSUM_COMPLETE)
461 csum_replace(&skb->csum, (__force __wsum)(nh->hop_limit << 8),
462 (__force __wsum)(new_ttl << 8));
463 nh->hop_limit = new_ttl;
466 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl,
469 new_ttl = OVS_MASKED(nh->ttl, new_ttl, mask);
471 csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
475 static int set_ipv4(struct sk_buff *skb, struct sw_flow_key *flow_key,
476 const struct ovs_key_ipv4 *key,
477 const struct ovs_key_ipv4 *mask)
483 err = skb_ensure_writable(skb, skb_network_offset(skb) +
484 sizeof(struct iphdr));
490 /* Setting an IP addresses is typically only a side effect of
491 * matching on them in the current userspace implementation, so it
492 * makes sense to check if the value actually changed.
494 if (mask->ipv4_src) {
495 new_addr = OVS_MASKED(nh->saddr, key->ipv4_src, mask->ipv4_src);
497 if (unlikely(new_addr != nh->saddr)) {
498 set_ip_addr(skb, nh, &nh->saddr, new_addr);
499 flow_key->ipv4.addr.src = new_addr;
502 if (mask->ipv4_dst) {
503 new_addr = OVS_MASKED(nh->daddr, key->ipv4_dst, mask->ipv4_dst);
505 if (unlikely(new_addr != nh->daddr)) {
506 set_ip_addr(skb, nh, &nh->daddr, new_addr);
507 flow_key->ipv4.addr.dst = new_addr;
510 if (mask->ipv4_tos) {
511 ipv4_change_dsfield(nh, ~mask->ipv4_tos, key->ipv4_tos);
512 flow_key->ip.tos = nh->tos;
514 if (mask->ipv4_ttl) {
515 set_ip_ttl(skb, nh, key->ipv4_ttl, mask->ipv4_ttl);
516 flow_key->ip.ttl = nh->ttl;
522 static bool is_ipv6_mask_nonzero(const __be32 addr[4])
524 return !!(addr[0] | addr[1] | addr[2] | addr[3]);
527 static int set_ipv6(struct sk_buff *skb, struct sw_flow_key *flow_key,
528 const struct ovs_key_ipv6 *key,
529 const struct ovs_key_ipv6 *mask)
534 err = skb_ensure_writable(skb, skb_network_offset(skb) +
535 sizeof(struct ipv6hdr));
541 /* Setting an IP addresses is typically only a side effect of
542 * matching on them in the current userspace implementation, so it
543 * makes sense to check if the value actually changed.
545 if (is_ipv6_mask_nonzero(mask->ipv6_src)) {
546 __be32 *saddr = (__be32 *)&nh->saddr;
549 mask_ipv6_addr(saddr, key->ipv6_src, mask->ipv6_src, masked);
551 if (unlikely(memcmp(saddr, masked, sizeof(masked)))) {
552 set_ipv6_addr(skb, flow_key->ip.proto, saddr, masked,
554 memcpy(&flow_key->ipv6.addr.src, masked,
555 sizeof(flow_key->ipv6.addr.src));
558 if (is_ipv6_mask_nonzero(mask->ipv6_dst)) {
559 unsigned int offset = 0;
560 int flags = IP6_FH_F_SKIP_RH;
561 bool recalc_csum = true;
562 __be32 *daddr = (__be32 *)&nh->daddr;
565 mask_ipv6_addr(daddr, key->ipv6_dst, mask->ipv6_dst, masked);
567 if (unlikely(memcmp(daddr, masked, sizeof(masked)))) {
568 if (ipv6_ext_hdr(nh->nexthdr))
569 recalc_csum = (ipv6_find_hdr(skb, &offset,
574 set_ipv6_addr(skb, flow_key->ip.proto, daddr, masked,
576 memcpy(&flow_key->ipv6.addr.dst, masked,
577 sizeof(flow_key->ipv6.addr.dst));
580 if (mask->ipv6_tclass) {
581 set_ipv6_dsfield(skb, nh, key->ipv6_tclass, mask->ipv6_tclass);
582 flow_key->ip.tos = ipv6_get_dsfield(nh);
584 if (mask->ipv6_label) {
585 set_ipv6_fl(skb, nh, ntohl(key->ipv6_label),
586 ntohl(mask->ipv6_label));
587 flow_key->ipv6.label =
588 *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
590 if (mask->ipv6_hlimit) {
591 set_ipv6_ttl(skb, nh, key->ipv6_hlimit, mask->ipv6_hlimit);
592 flow_key->ip.ttl = nh->hop_limit;
597 static int set_nsh(struct sk_buff *skb, struct sw_flow_key *flow_key,
598 const struct nlattr *a)
607 struct ovs_key_nsh key;
608 struct ovs_key_nsh mask;
610 err = nsh_key_from_nlattr(a, &key, &mask);
614 /* Make sure the NSH base header is there */
615 if (!pskb_may_pull(skb, skb_network_offset(skb) + NSH_BASE_HDR_LEN))
619 length = nsh_hdr_len(nh);
621 /* Make sure the whole NSH header is there */
622 err = skb_ensure_writable(skb, skb_network_offset(skb) +
628 skb_postpull_rcsum(skb, nh, length);
629 flags = nsh_get_flags(nh);
630 flags = OVS_MASKED(flags, key.base.flags, mask.base.flags);
631 flow_key->nsh.base.flags = flags;
632 ttl = nsh_get_ttl(nh);
633 ttl = OVS_MASKED(ttl, key.base.ttl, mask.base.ttl);
634 flow_key->nsh.base.ttl = ttl;
635 nsh_set_flags_and_ttl(nh, flags, ttl);
636 nh->path_hdr = OVS_MASKED(nh->path_hdr, key.base.path_hdr,
638 flow_key->nsh.base.path_hdr = nh->path_hdr;
639 switch (nh->mdtype) {
641 for (i = 0; i < NSH_MD1_CONTEXT_SIZE; i++) {
643 OVS_MASKED(nh->md1.context[i], key.context[i],
646 memcpy(flow_key->nsh.context, nh->md1.context,
647 sizeof(nh->md1.context));
650 memset(flow_key->nsh.context, 0,
651 sizeof(flow_key->nsh.context));
656 skb_postpush_rcsum(skb, nh, length);
660 /* Must follow skb_ensure_writable() since that can move the skb data. */
661 static void set_tp_port(struct sk_buff *skb, __be16 *port,
662 __be16 new_port, __sum16 *check)
664 ovs_ct_clear(skb, NULL);
665 inet_proto_csum_replace2(check, skb, *port, new_port, false);
669 static int set_udp(struct sk_buff *skb, struct sw_flow_key *flow_key,
670 const struct ovs_key_udp *key,
671 const struct ovs_key_udp *mask)
677 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
678 sizeof(struct udphdr));
683 /* Either of the masks is non-zero, so do not bother checking them. */
684 src = OVS_MASKED(uh->source, key->udp_src, mask->udp_src);
685 dst = OVS_MASKED(uh->dest, key->udp_dst, mask->udp_dst);
687 if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
688 if (likely(src != uh->source)) {
689 set_tp_port(skb, &uh->source, src, &uh->check);
690 flow_key->tp.src = src;
692 if (likely(dst != uh->dest)) {
693 set_tp_port(skb, &uh->dest, dst, &uh->check);
694 flow_key->tp.dst = dst;
697 if (unlikely(!uh->check))
698 uh->check = CSUM_MANGLED_0;
702 flow_key->tp.src = src;
703 flow_key->tp.dst = dst;
704 ovs_ct_clear(skb, NULL);
712 static int set_tcp(struct sk_buff *skb, struct sw_flow_key *flow_key,
713 const struct ovs_key_tcp *key,
714 const struct ovs_key_tcp *mask)
720 err = skb_ensure_writable(skb, skb_transport_offset(skb) +
721 sizeof(struct tcphdr));
726 src = OVS_MASKED(th->source, key->tcp_src, mask->tcp_src);
727 if (likely(src != th->source)) {
728 set_tp_port(skb, &th->source, src, &th->check);
729 flow_key->tp.src = src;
731 dst = OVS_MASKED(th->dest, key->tcp_dst, mask->tcp_dst);
732 if (likely(dst != th->dest)) {
733 set_tp_port(skb, &th->dest, dst, &th->check);
734 flow_key->tp.dst = dst;
741 static int set_sctp(struct sk_buff *skb, struct sw_flow_key *flow_key,
742 const struct ovs_key_sctp *key,
743 const struct ovs_key_sctp *mask)
745 unsigned int sctphoff = skb_transport_offset(skb);
747 __le32 old_correct_csum, new_csum, old_csum;
750 err = skb_ensure_writable(skb, sctphoff + sizeof(struct sctphdr));
755 old_csum = sh->checksum;
756 old_correct_csum = sctp_compute_cksum(skb, sctphoff);
758 sh->source = OVS_MASKED(sh->source, key->sctp_src, mask->sctp_src);
759 sh->dest = OVS_MASKED(sh->dest, key->sctp_dst, mask->sctp_dst);
761 new_csum = sctp_compute_cksum(skb, sctphoff);
763 /* Carry any checksum errors through. */
764 sh->checksum = old_csum ^ old_correct_csum ^ new_csum;
767 ovs_ct_clear(skb, NULL);
769 flow_key->tp.src = sh->source;
770 flow_key->tp.dst = sh->dest;
775 static int ovs_vport_output(struct net *net, struct sock *sk,
778 struct ovs_frag_data *data = this_cpu_ptr(&ovs_frag_data_storage);
779 struct vport *vport = data->vport;
781 if (skb_cow_head(skb, data->l2_len) < 0) {
786 __skb_dst_copy(skb, data->dst);
787 *OVS_CB(skb) = data->cb;
788 skb->inner_protocol = data->inner_protocol;
789 if (data->vlan_tci & VLAN_CFI_MASK)
790 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci & ~VLAN_CFI_MASK);
792 __vlan_hwaccel_clear_tag(skb);
794 /* Reconstruct the MAC header. */
795 skb_push(skb, data->l2_len);
796 memcpy(skb->data, &data->l2_data, data->l2_len);
797 skb_postpush_rcsum(skb, skb->data, data->l2_len);
798 skb_reset_mac_header(skb);
800 if (eth_p_mpls(skb->protocol)) {
801 skb->inner_network_header = skb->network_header;
802 skb_set_network_header(skb, data->network_offset);
803 skb_reset_mac_len(skb);
806 ovs_vport_send(vport, skb, data->mac_proto);
811 ovs_dst_get_mtu(const struct dst_entry *dst)
813 return dst->dev->mtu;
816 static struct dst_ops ovs_dst_ops = {
818 .mtu = ovs_dst_get_mtu,
821 /* prepare_frag() is called once per (larger-than-MTU) frame; its inverse is
822 * ovs_vport_output(), which is called once per fragmented packet.
824 static void prepare_frag(struct vport *vport, struct sk_buff *skb,
825 u16 orig_network_offset, u8 mac_proto)
827 unsigned int hlen = skb_network_offset(skb);
828 struct ovs_frag_data *data;
830 data = this_cpu_ptr(&ovs_frag_data_storage);
831 data->dst = skb->_skb_refdst;
833 data->cb = *OVS_CB(skb);
834 data->inner_protocol = skb->inner_protocol;
835 data->network_offset = orig_network_offset;
836 if (skb_vlan_tag_present(skb))
837 data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
840 data->vlan_proto = skb->vlan_proto;
841 data->mac_proto = mac_proto;
843 memcpy(&data->l2_data, skb->data, hlen);
845 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
849 static void ovs_fragment(struct net *net, struct vport *vport,
850 struct sk_buff *skb, u16 mru,
851 struct sw_flow_key *key)
853 u16 orig_network_offset = 0;
855 if (eth_p_mpls(skb->protocol)) {
856 orig_network_offset = skb_network_offset(skb);
857 skb->network_header = skb->inner_network_header;
860 if (skb_network_offset(skb) > MAX_L2_LEN) {
861 OVS_NLERR(1, "L2 header too long to fragment");
865 if (key->eth.type == htons(ETH_P_IP)) {
866 struct rtable ovs_rt = { 0 };
867 unsigned long orig_dst;
869 prepare_frag(vport, skb, orig_network_offset,
870 ovs_key_mac_proto(key));
871 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
872 DST_OBSOLETE_NONE, DST_NOCOUNT);
873 ovs_rt.dst.dev = vport->dev;
875 orig_dst = skb->_skb_refdst;
876 skb_dst_set_noref(skb, &ovs_rt.dst);
877 IPCB(skb)->frag_max_size = mru;
879 ip_do_fragment(net, skb->sk, skb, ovs_vport_output);
880 refdst_drop(orig_dst);
881 } else if (key->eth.type == htons(ETH_P_IPV6)) {
882 unsigned long orig_dst;
883 struct rt6_info ovs_rt;
885 prepare_frag(vport, skb, orig_network_offset,
886 ovs_key_mac_proto(key));
887 memset(&ovs_rt, 0, sizeof(ovs_rt));
888 dst_init(&ovs_rt.dst, &ovs_dst_ops, NULL, 1,
889 DST_OBSOLETE_NONE, DST_NOCOUNT);
890 ovs_rt.dst.dev = vport->dev;
892 orig_dst = skb->_skb_refdst;
893 skb_dst_set_noref(skb, &ovs_rt.dst);
894 IP6CB(skb)->frag_max_size = mru;
896 ipv6_stub->ipv6_fragment(net, skb->sk, skb, ovs_vport_output);
897 refdst_drop(orig_dst);
899 WARN_ONCE(1, "Failed fragment ->%s: eth=%04x, MRU=%d, MTU=%d.",
900 ovs_vport_name(vport), ntohs(key->eth.type), mru,
910 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port,
911 struct sw_flow_key *key)
913 struct vport *vport = ovs_vport_rcu(dp, out_port);
916 u16 mru = OVS_CB(skb)->mru;
917 u32 cutlen = OVS_CB(skb)->cutlen;
919 if (unlikely(cutlen > 0)) {
920 if (skb->len - cutlen > ovs_mac_header_len(key))
921 pskb_trim(skb, skb->len - cutlen);
923 pskb_trim(skb, ovs_mac_header_len(key));
927 (skb->len <= mru + vport->dev->hard_header_len))) {
928 ovs_vport_send(vport, skb, ovs_key_mac_proto(key));
929 } else if (mru <= vport->dev->mtu) {
930 struct net *net = read_pnet(&dp->net);
932 ovs_fragment(net, vport, skb, mru, key);
941 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
942 struct sw_flow_key *key, const struct nlattr *attr,
943 const struct nlattr *actions, int actions_len,
946 struct dp_upcall_info upcall;
947 const struct nlattr *a;
950 memset(&upcall, 0, sizeof(upcall));
951 upcall.cmd = OVS_PACKET_CMD_ACTION;
952 upcall.mru = OVS_CB(skb)->mru;
954 for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
955 a = nla_next(a, &rem)) {
956 switch (nla_type(a)) {
957 case OVS_USERSPACE_ATTR_USERDATA:
961 case OVS_USERSPACE_ATTR_PID:
962 upcall.portid = nla_get_u32(a);
965 case OVS_USERSPACE_ATTR_EGRESS_TUN_PORT: {
966 /* Get out tunnel info. */
969 vport = ovs_vport_rcu(dp, nla_get_u32(a));
973 err = dev_fill_metadata_dst(vport->dev, skb);
975 upcall.egress_tun_info = skb_tunnel_info(skb);
981 case OVS_USERSPACE_ATTR_ACTIONS: {
982 /* Include actions. */
983 upcall.actions = actions;
984 upcall.actions_len = actions_len;
988 } /* End of switch. */
991 return ovs_dp_upcall(dp, skb, key, &upcall, cutlen);
994 static int dec_ttl_exception_handler(struct datapath *dp, struct sk_buff *skb,
995 struct sw_flow_key *key,
996 const struct nlattr *attr, bool last)
998 /* The first attribute is always 'OVS_DEC_TTL_ATTR_ACTION'. */
999 struct nlattr *actions = nla_data(attr);
1001 if (nla_len(actions))
1002 return clone_execute(dp, skb, key, 0, nla_data(actions),
1003 nla_len(actions), last, false);
1009 /* When 'last' is true, sample() should always consume the 'skb'.
1010 * Otherwise, sample() should keep 'skb' intact regardless what
1011 * actions are executed within sample().
1013 static int sample(struct datapath *dp, struct sk_buff *skb,
1014 struct sw_flow_key *key, const struct nlattr *attr,
1017 struct nlattr *actions;
1018 struct nlattr *sample_arg;
1019 int rem = nla_len(attr);
1020 const struct sample_arg *arg;
1021 bool clone_flow_key;
1023 /* The first action is always 'OVS_SAMPLE_ATTR_ARG'. */
1024 sample_arg = nla_data(attr);
1025 arg = nla_data(sample_arg);
1026 actions = nla_next(sample_arg, &rem);
1028 if ((arg->probability != U32_MAX) &&
1029 (!arg->probability || prandom_u32() > arg->probability)) {
1035 clone_flow_key = !arg->exec;
1036 return clone_execute(dp, skb, key, 0, actions, rem, last,
1040 /* When 'last' is true, clone() should always consume the 'skb'.
1041 * Otherwise, clone() should keep 'skb' intact regardless what
1042 * actions are executed within clone().
1044 static int clone(struct datapath *dp, struct sk_buff *skb,
1045 struct sw_flow_key *key, const struct nlattr *attr,
1048 struct nlattr *actions;
1049 struct nlattr *clone_arg;
1050 int rem = nla_len(attr);
1051 bool dont_clone_flow_key;
1053 /* The first action is always 'OVS_CLONE_ATTR_EXEC'. */
1054 clone_arg = nla_data(attr);
1055 dont_clone_flow_key = nla_get_u32(clone_arg);
1056 actions = nla_next(clone_arg, &rem);
1058 return clone_execute(dp, skb, key, 0, actions, rem, last,
1059 !dont_clone_flow_key);
1062 static void execute_hash(struct sk_buff *skb, struct sw_flow_key *key,
1063 const struct nlattr *attr)
1065 struct ovs_action_hash *hash_act = nla_data(attr);
1068 /* OVS_HASH_ALG_L4 is the only possible hash algorithm. */
1069 hash = skb_get_hash(skb);
1070 hash = jhash_1word(hash, hash_act->hash_basis);
1074 key->ovs_flow_hash = hash;
1077 static int execute_set_action(struct sk_buff *skb,
1078 struct sw_flow_key *flow_key,
1079 const struct nlattr *a)
1081 /* Only tunnel set execution is supported without a mask. */
1082 if (nla_type(a) == OVS_KEY_ATTR_TUNNEL_INFO) {
1083 struct ovs_tunnel_info *tun = nla_data(a);
1086 dst_hold((struct dst_entry *)tun->tun_dst);
1087 skb_dst_set(skb, (struct dst_entry *)tun->tun_dst);
1094 /* Mask is at the midpoint of the data. */
1095 #define get_mask(a, type) ((const type)nla_data(a) + 1)
1097 static int execute_masked_set_action(struct sk_buff *skb,
1098 struct sw_flow_key *flow_key,
1099 const struct nlattr *a)
1103 switch (nla_type(a)) {
1104 case OVS_KEY_ATTR_PRIORITY:
1105 OVS_SET_MASKED(skb->priority, nla_get_u32(a),
1106 *get_mask(a, u32 *));
1107 flow_key->phy.priority = skb->priority;
1110 case OVS_KEY_ATTR_SKB_MARK:
1111 OVS_SET_MASKED(skb->mark, nla_get_u32(a), *get_mask(a, u32 *));
1112 flow_key->phy.skb_mark = skb->mark;
1115 case OVS_KEY_ATTR_TUNNEL_INFO:
1116 /* Masked data not supported for tunnel. */
1120 case OVS_KEY_ATTR_ETHERNET:
1121 err = set_eth_addr(skb, flow_key, nla_data(a),
1122 get_mask(a, struct ovs_key_ethernet *));
1125 case OVS_KEY_ATTR_NSH:
1126 err = set_nsh(skb, flow_key, a);
1129 case OVS_KEY_ATTR_IPV4:
1130 err = set_ipv4(skb, flow_key, nla_data(a),
1131 get_mask(a, struct ovs_key_ipv4 *));
1134 case OVS_KEY_ATTR_IPV6:
1135 err = set_ipv6(skb, flow_key, nla_data(a),
1136 get_mask(a, struct ovs_key_ipv6 *));
1139 case OVS_KEY_ATTR_TCP:
1140 err = set_tcp(skb, flow_key, nla_data(a),
1141 get_mask(a, struct ovs_key_tcp *));
1144 case OVS_KEY_ATTR_UDP:
1145 err = set_udp(skb, flow_key, nla_data(a),
1146 get_mask(a, struct ovs_key_udp *));
1149 case OVS_KEY_ATTR_SCTP:
1150 err = set_sctp(skb, flow_key, nla_data(a),
1151 get_mask(a, struct ovs_key_sctp *));
1154 case OVS_KEY_ATTR_MPLS:
1155 err = set_mpls(skb, flow_key, nla_data(a), get_mask(a,
1159 case OVS_KEY_ATTR_CT_STATE:
1160 case OVS_KEY_ATTR_CT_ZONE:
1161 case OVS_KEY_ATTR_CT_MARK:
1162 case OVS_KEY_ATTR_CT_LABELS:
1163 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4:
1164 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6:
1172 static int execute_recirc(struct datapath *dp, struct sk_buff *skb,
1173 struct sw_flow_key *key,
1174 const struct nlattr *a, bool last)
1178 if (!is_flow_key_valid(key)) {
1181 err = ovs_flow_key_update(skb, key);
1185 BUG_ON(!is_flow_key_valid(key));
1187 recirc_id = nla_get_u32(a);
1188 return clone_execute(dp, skb, key, recirc_id, NULL, 0, last, true);
1191 static int execute_check_pkt_len(struct datapath *dp, struct sk_buff *skb,
1192 struct sw_flow_key *key,
1193 const struct nlattr *attr, bool last)
1195 struct ovs_skb_cb *ovs_cb = OVS_CB(skb);
1196 const struct nlattr *actions, *cpl_arg;
1197 int len, max_len, rem = nla_len(attr);
1198 const struct check_pkt_len_arg *arg;
1199 bool clone_flow_key;
1201 /* The first netlink attribute in 'attr' is always
1202 * 'OVS_CHECK_PKT_LEN_ATTR_ARG'.
1204 cpl_arg = nla_data(attr);
1205 arg = nla_data(cpl_arg);
1207 len = ovs_cb->mru ? ovs_cb->mru + skb->mac_len : skb->len;
1208 max_len = arg->pkt_len;
1210 if ((skb_is_gso(skb) && skb_gso_validate_mac_len(skb, max_len)) ||
1212 /* Second netlink attribute in 'attr' is always
1213 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL'.
1215 actions = nla_next(cpl_arg, &rem);
1216 clone_flow_key = !arg->exec_for_lesser_equal;
1218 /* Third netlink attribute in 'attr' is always
1219 * 'OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER'.
1221 actions = nla_next(cpl_arg, &rem);
1222 actions = nla_next(actions, &rem);
1223 clone_flow_key = !arg->exec_for_greater;
1226 return clone_execute(dp, skb, key, 0, nla_data(actions),
1227 nla_len(actions), last, clone_flow_key);
1230 static int execute_dec_ttl(struct sk_buff *skb, struct sw_flow_key *key)
1234 if (skb->protocol == htons(ETH_P_IPV6)) {
1237 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1244 if (nh->hop_limit <= 1)
1245 return -EHOSTUNREACH;
1247 key->ip.ttl = --nh->hop_limit;
1248 } else if (skb->protocol == htons(ETH_P_IP)) {
1252 err = skb_ensure_writable(skb, skb_network_offset(skb) +
1259 return -EHOSTUNREACH;
1261 old_ttl = nh->ttl--;
1262 csum_replace2(&nh->check, htons(old_ttl << 8),
1263 htons(nh->ttl << 8));
1264 key->ip.ttl = nh->ttl;
1269 /* Execute a list of actions against 'skb'. */
1270 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
1271 struct sw_flow_key *key,
1272 const struct nlattr *attr, int len)
1274 const struct nlattr *a;
1277 for (a = attr, rem = len; rem > 0;
1278 a = nla_next(a, &rem)) {
1281 switch (nla_type(a)) {
1282 case OVS_ACTION_ATTR_OUTPUT: {
1283 int port = nla_get_u32(a);
1284 struct sk_buff *clone;
1286 /* Every output action needs a separate clone
1287 * of 'skb', In case the output action is the
1288 * last action, cloning can be avoided.
1290 if (nla_is_last(a, rem)) {
1291 do_output(dp, skb, port, key);
1292 /* 'skb' has been used for output.
1297 clone = skb_clone(skb, GFP_ATOMIC);
1299 do_output(dp, clone, port, key);
1300 OVS_CB(skb)->cutlen = 0;
1304 case OVS_ACTION_ATTR_TRUNC: {
1305 struct ovs_action_trunc *trunc = nla_data(a);
1307 if (skb->len > trunc->max_len)
1308 OVS_CB(skb)->cutlen = skb->len - trunc->max_len;
1312 case OVS_ACTION_ATTR_USERSPACE:
1313 output_userspace(dp, skb, key, a, attr,
1314 len, OVS_CB(skb)->cutlen);
1315 OVS_CB(skb)->cutlen = 0;
1318 case OVS_ACTION_ATTR_HASH:
1319 execute_hash(skb, key, a);
1322 case OVS_ACTION_ATTR_PUSH_MPLS: {
1323 struct ovs_action_push_mpls *mpls = nla_data(a);
1325 err = push_mpls(skb, key, mpls->mpls_lse,
1326 mpls->mpls_ethertype, skb->mac_len);
1329 case OVS_ACTION_ATTR_ADD_MPLS: {
1330 struct ovs_action_add_mpls *mpls = nla_data(a);
1333 if (mpls->tun_flags & OVS_MPLS_L3_TUNNEL_FLAG_MASK)
1334 mac_len = skb->mac_len;
1336 err = push_mpls(skb, key, mpls->mpls_lse,
1337 mpls->mpls_ethertype, mac_len);
1340 case OVS_ACTION_ATTR_POP_MPLS:
1341 err = pop_mpls(skb, key, nla_get_be16(a));
1344 case OVS_ACTION_ATTR_PUSH_VLAN:
1345 err = push_vlan(skb, key, nla_data(a));
1348 case OVS_ACTION_ATTR_POP_VLAN:
1349 err = pop_vlan(skb, key);
1352 case OVS_ACTION_ATTR_RECIRC: {
1353 bool last = nla_is_last(a, rem);
1355 err = execute_recirc(dp, skb, key, a, last);
1357 /* If this is the last action, the skb has
1358 * been consumed or freed.
1359 * Return immediately.
1366 case OVS_ACTION_ATTR_SET:
1367 err = execute_set_action(skb, key, nla_data(a));
1370 case OVS_ACTION_ATTR_SET_MASKED:
1371 case OVS_ACTION_ATTR_SET_TO_MASKED:
1372 err = execute_masked_set_action(skb, key, nla_data(a));
1375 case OVS_ACTION_ATTR_SAMPLE: {
1376 bool last = nla_is_last(a, rem);
1378 err = sample(dp, skb, key, a, last);
1385 case OVS_ACTION_ATTR_CT:
1386 if (!is_flow_key_valid(key)) {
1387 err = ovs_flow_key_update(skb, key);
1392 err = ovs_ct_execute(ovs_dp_get_net(dp), skb, key,
1395 /* Hide stolen IP fragments from user space. */
1397 return err == -EINPROGRESS ? 0 : err;
1400 case OVS_ACTION_ATTR_CT_CLEAR:
1401 err = ovs_ct_clear(skb, key);
1404 case OVS_ACTION_ATTR_PUSH_ETH:
1405 err = push_eth(skb, key, nla_data(a));
1408 case OVS_ACTION_ATTR_POP_ETH:
1409 err = pop_eth(skb, key);
1412 case OVS_ACTION_ATTR_PUSH_NSH: {
1413 u8 buffer[NSH_HDR_MAX_LEN];
1414 struct nshhdr *nh = (struct nshhdr *)buffer;
1416 err = nsh_hdr_from_nlattr(nla_data(a), nh,
1420 err = push_nsh(skb, key, nh);
1424 case OVS_ACTION_ATTR_POP_NSH:
1425 err = pop_nsh(skb, key);
1428 case OVS_ACTION_ATTR_METER:
1429 if (ovs_meter_execute(dp, skb, key, nla_get_u32(a))) {
1435 case OVS_ACTION_ATTR_CLONE: {
1436 bool last = nla_is_last(a, rem);
1438 err = clone(dp, skb, key, a, last);
1445 case OVS_ACTION_ATTR_CHECK_PKT_LEN: {
1446 bool last = nla_is_last(a, rem);
1448 err = execute_check_pkt_len(dp, skb, key, a, last);
1455 case OVS_ACTION_ATTR_DEC_TTL:
1456 err = execute_dec_ttl(skb, key);
1457 if (err == -EHOSTUNREACH) {
1458 err = dec_ttl_exception_handler(dp, skb, key,
1465 if (unlikely(err)) {
1475 /* Execute the actions on the clone of the packet. The effect of the
1476 * execution does not affect the original 'skb' nor the original 'key'.
1478 * The execution may be deferred in case the actions can not be executed
1481 static int clone_execute(struct datapath *dp, struct sk_buff *skb,
1482 struct sw_flow_key *key, u32 recirc_id,
1483 const struct nlattr *actions, int len,
1484 bool last, bool clone_flow_key)
1486 struct deferred_action *da;
1487 struct sw_flow_key *clone;
1489 skb = last ? skb : skb_clone(skb, GFP_ATOMIC);
1491 /* Out of memory, skip this action.
1496 /* When clone_flow_key is false, the 'key' will not be change
1497 * by the actions, then the 'key' can be used directly.
1498 * Otherwise, try to clone key from the next recursion level of
1499 * 'flow_keys'. If clone is successful, execute the actions
1500 * without deferring.
1502 clone = clone_flow_key ? clone_key(key) : key;
1506 if (actions) { /* Sample action */
1508 __this_cpu_inc(exec_actions_level);
1510 err = do_execute_actions(dp, skb, clone,
1514 __this_cpu_dec(exec_actions_level);
1515 } else { /* Recirc action */
1516 clone->recirc_id = recirc_id;
1517 ovs_dp_process_packet(skb, clone);
1522 /* Out of 'flow_keys' space. Defer actions */
1523 da = add_deferred_actions(skb, key, actions, len);
1525 if (!actions) { /* Recirc action */
1527 key->recirc_id = recirc_id;
1530 /* Out of per CPU action FIFO space. Drop the 'skb' and
1535 if (net_ratelimit()) {
1536 if (actions) { /* Sample action */
1537 pr_warn("%s: deferred action limit reached, drop sample action\n",
1539 } else { /* Recirc action */
1540 pr_warn("%s: deferred action limit reached, drop recirc action\n",
1548 static void process_deferred_actions(struct datapath *dp)
1550 struct action_fifo *fifo = this_cpu_ptr(action_fifos);
1552 /* Do not touch the FIFO in case there is no deferred actions. */
1553 if (action_fifo_is_empty(fifo))
1556 /* Finishing executing all deferred actions. */
1558 struct deferred_action *da = action_fifo_get(fifo);
1559 struct sk_buff *skb = da->skb;
1560 struct sw_flow_key *key = &da->pkt_key;
1561 const struct nlattr *actions = da->actions;
1562 int actions_len = da->actions_len;
1565 do_execute_actions(dp, skb, key, actions, actions_len);
1567 ovs_dp_process_packet(skb, key);
1568 } while (!action_fifo_is_empty(fifo));
1570 /* Reset FIFO for the next packet. */
1571 action_fifo_init(fifo);
1574 /* Execute a list of actions against 'skb'. */
1575 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb,
1576 const struct sw_flow_actions *acts,
1577 struct sw_flow_key *key)
1581 level = __this_cpu_inc_return(exec_actions_level);
1582 if (unlikely(level > OVS_RECURSION_LIMIT)) {
1583 net_crit_ratelimited("ovs: recursion limit reached on datapath %s, probable configuration error\n",
1590 OVS_CB(skb)->acts_origlen = acts->orig_len;
1591 err = do_execute_actions(dp, skb, key,
1592 acts->actions, acts->actions_len);
1595 process_deferred_actions(dp);
1598 __this_cpu_dec(exec_actions_level);
1602 int action_fifos_init(void)
1604 action_fifos = alloc_percpu(struct action_fifo);
1608 flow_keys = alloc_percpu(struct action_flow_keys);
1610 free_percpu(action_fifos);
1617 void action_fifos_exit(void)
1619 free_percpu(action_fifos);
1620 free_percpu(flow_keys);