2 * Copyright (c) 2007-2012 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/if_arp.h>
22 #include <linux/if_ether.h>
24 #include <linux/if_vlan.h>
25 #include <linux/igmp.h>
27 #include <linux/in_route.h>
28 #include <linux/inetdevice.h>
29 #include <linux/jhash.h>
30 #include <linux/list.h>
31 #include <linux/kernel.h>
32 #include <linux/version.h>
33 #include <linux/workqueue.h>
34 #include <linux/rculist.h>
36 #include <net/dsfield.h>
39 #include <net/inet_ecn.h>
41 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
44 #include <net/route.h>
52 #include "vport-generic.h"
53 #include "vport-internal_dev.h"
55 #define PORT_TABLE_SIZE 1024
57 static struct hlist_head *port_table __read_mostly;
60 * These are just used as an optimization: they don't require any kind of
61 * synchronization because we could have just as easily read the value before
62 * the port change happened.
64 static unsigned int key_local_remote_ports __read_mostly;
65 static unsigned int key_remote_ports __read_mostly;
66 static unsigned int key_multicast_ports __read_mostly;
67 static unsigned int local_remote_ports __read_mostly;
68 static unsigned int remote_ports __read_mostly;
69 static unsigned int null_ports __read_mostly;
70 static unsigned int multicast_ports __read_mostly;
72 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
73 #define rt_dst(rt) (rt->dst)
75 #define rt_dst(rt) (rt->u.dst)
78 static struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
80 return vport_from_priv(tnl_vport);
83 static void free_config_rcu(struct rcu_head *rcu)
85 struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
89 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
90 * within an RCU callback. Fortunately this part doesn't require waiting for
91 * an RCU grace period.
93 static void free_mutable_rtnl(struct tnl_mutable_config *mutable)
96 if (ipv4_is_multicast(mutable->key.daddr) && mutable->mlink) {
97 struct in_device *in_dev;
98 in_dev = inetdev_by_index(port_key_get_net(&mutable->key), mutable->mlink);
100 ip_mc_dec_group(in_dev, mutable->key.daddr);
104 static void assign_config_rcu(struct vport *vport,
105 struct tnl_mutable_config *new_config)
107 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
108 struct tnl_mutable_config *old_config;
110 old_config = rtnl_dereference(tnl_vport->mutable);
111 rcu_assign_pointer(tnl_vport->mutable, new_config);
113 free_mutable_rtnl(old_config);
114 call_rcu(&old_config->rcu, free_config_rcu);
117 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
119 bool is_multicast = ipv4_is_multicast(mutable->key.daddr);
121 if (mutable->flags & TNL_F_IN_KEY_MATCH) {
122 if (mutable->key.saddr)
123 return &local_remote_ports;
124 else if (is_multicast)
125 return &multicast_ports;
127 return &remote_ports;
129 if (mutable->key.saddr)
130 return &key_local_remote_ports;
131 else if (is_multicast)
132 return &key_multicast_ports;
133 else if (mutable->key.daddr)
134 return &key_remote_ports;
140 static u32 port_hash(const struct port_lookup_key *key)
142 return jhash2((u32 *)key, (PORT_KEY_LEN / sizeof(u32)), 0);
145 static struct hlist_head *find_bucket(u32 hash)
147 return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
150 static void port_table_add_port(struct vport *vport)
152 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
153 const struct tnl_mutable_config *mutable;
156 mutable = rtnl_dereference(tnl_vport->mutable);
157 hash = port_hash(&mutable->key);
158 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
160 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
163 static void port_table_move_port(struct vport *vport,
164 struct tnl_mutable_config *new_mutable)
166 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
169 hash = port_hash(&new_mutable->key);
170 hlist_del_init_rcu(&tnl_vport->hash_node);
171 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
173 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
174 assign_config_rcu(vport, new_mutable);
175 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
178 static void port_table_remove_port(struct vport *vport)
180 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
182 hlist_del_init_rcu(&tnl_vport->hash_node);
184 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
187 static struct vport *port_table_lookup(struct port_lookup_key *key,
188 const struct tnl_mutable_config **pmutable)
190 struct hlist_node *n;
191 struct hlist_head *bucket;
192 u32 hash = port_hash(key);
193 struct tnl_vport *tnl_vport;
195 bucket = find_bucket(hash);
197 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
198 struct tnl_mutable_config *mutable;
200 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
201 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
203 return tnl_vport_to_vport(tnl_vport);
210 struct vport *ovs_tnl_find_port(struct net *net, __be32 saddr, __be32 daddr,
211 __be64 key, int tunnel_type,
212 const struct tnl_mutable_config **mutable)
214 struct port_lookup_key lookup;
216 bool is_multicast = ipv4_is_multicast(saddr);
218 port_key_set_net(&lookup, net);
219 lookup.saddr = saddr;
220 lookup.daddr = daddr;
222 /* First try for exact match on in_key. */
224 lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
225 if (!is_multicast && key_local_remote_ports) {
226 vport = port_table_lookup(&lookup, mutable);
230 if (key_remote_ports) {
232 vport = port_table_lookup(&lookup, mutable);
236 lookup.saddr = saddr;
239 /* Then try matches that wildcard in_key. */
241 lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
242 if (!is_multicast && local_remote_ports) {
243 vport = port_table_lookup(&lookup, mutable);
249 vport = port_table_lookup(&lookup, mutable);
256 lookup.daddr = saddr;
257 if (key_multicast_ports) {
258 lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
260 vport = port_table_lookup(&lookup, mutable);
264 if (multicast_ports) {
265 lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
267 vport = port_table_lookup(&lookup, mutable);
277 lookup.tunnel_type = tunnel_type;
278 vport = port_table_lookup(&lookup, mutable);
285 static void ecn_decapsulate(struct sk_buff *skb)
287 if (unlikely(INET_ECN_is_ce(OVS_CB(skb)->tun_key->ipv4_tos))) {
288 __be16 protocol = skb->protocol;
290 skb_set_network_header(skb, ETH_HLEN);
292 if (protocol == htons(ETH_P_8021Q)) {
293 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
296 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
297 skb_set_network_header(skb, VLAN_ETH_HLEN);
300 if (protocol == htons(ETH_P_IP)) {
301 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
302 + sizeof(struct iphdr))))
305 IP_ECN_set_ce(ip_hdr(skb));
307 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
308 else if (protocol == htons(ETH_P_IPV6)) {
309 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
310 + sizeof(struct ipv6hdr))))
313 IP6_ECN_set_ce(ipv6_hdr(skb));
320 * ovs_tnl_rcv - ingress point for generic tunnel code
322 * @vport: port this packet was received on
323 * @skb: received packet
324 * @tos: ToS from encapsulating IP packet, used to copy ECN bits
326 * Must be called with rcu_read_lock.
328 * Packets received by this function are in the following state:
329 * - skb->data points to the inner Ethernet header.
330 * - The inner Ethernet header is in the linear data area.
331 * - skb->csum does not include the inner Ethernet header.
332 * - The layer pointers are undefined.
334 void ovs_tnl_rcv(struct vport *vport, struct sk_buff *skb)
338 skb_reset_mac_header(skb);
341 if (likely(ntohs(eh->h_proto) >= 1536))
342 skb->protocol = eh->h_proto;
344 skb->protocol = htons(ETH_P_802_2);
348 skb_clear_rxhash(skb);
351 ecn_decapsulate(skb);
352 vlan_set_tci(skb, 0);
354 if (unlikely(compute_ip_summed(skb, false))) {
359 ovs_vport_receive(vport, skb);
362 static bool check_ipv4_address(__be32 addr)
364 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
365 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
371 static bool ipv4_should_icmp(struct sk_buff *skb)
373 struct iphdr *old_iph = ip_hdr(skb);
375 /* Don't respond to L2 broadcast. */
376 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
379 /* Don't respond to L3 broadcast or invalid addresses. */
380 if (!check_ipv4_address(old_iph->daddr) ||
381 !check_ipv4_address(old_iph->saddr))
384 /* Only respond to the first fragment. */
385 if (old_iph->frag_off & htons(IP_OFFSET))
388 /* Don't respond to ICMP error messages. */
389 if (old_iph->protocol == IPPROTO_ICMP) {
390 u8 icmp_type, *icmp_typep;
392 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
393 (old_iph->ihl << 2) +
394 offsetof(struct icmphdr, type) -
395 skb->data, sizeof(icmp_type),
401 if (*icmp_typep > NR_ICMP_TYPES
402 || (*icmp_typep <= ICMP_PARAMETERPROB
403 && *icmp_typep != ICMP_ECHOREPLY
404 && *icmp_typep != ICMP_ECHO))
411 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
412 unsigned int mtu, unsigned int payload_length)
414 struct iphdr *iph, *old_iph = ip_hdr(skb);
415 struct icmphdr *icmph;
418 iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
419 icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
420 payload = skb_put(nskb, payload_length);
424 iph->ihl = sizeof(struct iphdr) >> 2;
425 iph->tos = (old_iph->tos & IPTOS_TOS_MASK) |
426 IPTOS_PREC_INTERNETCONTROL;
427 iph->tot_len = htons(sizeof(struct iphdr)
428 + sizeof(struct icmphdr)
430 get_random_bytes(&iph->id, sizeof(iph->id));
433 iph->protocol = IPPROTO_ICMP;
434 iph->daddr = old_iph->saddr;
435 iph->saddr = old_iph->daddr;
440 icmph->type = ICMP_DEST_UNREACH;
441 icmph->code = ICMP_FRAG_NEEDED;
442 icmph->un.gateway = htonl(mtu);
445 nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
446 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
447 payload, payload_length,
449 icmph->checksum = csum_fold(nskb->csum);
452 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
453 static bool ipv6_should_icmp(struct sk_buff *skb)
455 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
457 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
458 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
461 /* Check source address is valid. */
462 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
463 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
466 /* Don't reply to unspecified addresses. */
467 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
470 /* Don't respond to ICMP error messages. */
471 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr, &frag_off);
475 if (nexthdr == NEXTHDR_ICMP) {
476 u8 icmp_type, *icmp_typep;
478 icmp_typep = skb_header_pointer(skb, payload_off +
479 offsetof(struct icmp6hdr,
481 sizeof(icmp_type), &icmp_type);
483 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
490 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
491 unsigned int mtu, unsigned int payload_length)
493 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
494 struct icmp6hdr *icmp6h;
497 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
498 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
499 payload = skb_put(nskb, payload_length);
504 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
505 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
507 ipv6h->nexthdr = NEXTHDR_ICMP;
508 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
509 ipv6h->daddr = old_ipv6h->saddr;
510 ipv6h->saddr = old_ipv6h->daddr;
513 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
514 icmp6h->icmp6_code = 0;
515 icmp6h->icmp6_cksum = 0;
516 icmp6h->icmp6_mtu = htonl(mtu);
518 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
519 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
520 payload, payload_length,
522 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
523 sizeof(struct icmp6hdr)
525 ipv6h->nexthdr, nskb->csum);
529 bool ovs_tnl_frag_needed(struct vport *vport,
530 const struct tnl_mutable_config *mutable,
531 struct sk_buff *skb, unsigned int mtu)
533 unsigned int eth_hdr_len = ETH_HLEN;
534 unsigned int total_length = 0, header_length = 0, payload_length;
535 struct ethhdr *eh, *old_eh = eth_hdr(skb);
536 struct sk_buff *nskb;
539 if (skb->protocol == htons(ETH_P_IP)) {
540 if (mtu < IP_MIN_MTU)
543 if (!ipv4_should_icmp(skb))
546 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
547 else if (skb->protocol == htons(ETH_P_IPV6)) {
548 if (mtu < IPV6_MIN_MTU)
552 * In theory we should do PMTUD on IPv6 multicast messages but
553 * we don't have an address to send from so just fragment.
555 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
558 if (!ipv6_should_icmp(skb))
566 if (old_eh->h_proto == htons(ETH_P_8021Q))
567 eth_hdr_len = VLAN_ETH_HLEN;
569 payload_length = skb->len - eth_hdr_len;
570 if (skb->protocol == htons(ETH_P_IP)) {
571 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
572 total_length = min_t(unsigned int, header_length +
573 payload_length, 576);
575 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
577 header_length = sizeof(struct ipv6hdr) +
578 sizeof(struct icmp6hdr);
579 total_length = min_t(unsigned int, header_length +
580 payload_length, IPV6_MIN_MTU);
584 payload_length = total_length - header_length;
586 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
591 skb_reserve(nskb, NET_IP_ALIGN);
593 /* Ethernet / VLAN */
594 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
595 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
596 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
597 nskb->protocol = eh->h_proto = old_eh->h_proto;
598 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
599 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
601 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
602 vh->h_vlan_encapsulated_proto = skb->protocol;
604 vlan_set_tci(nskb, vlan_get_tci(skb));
605 skb_reset_mac_header(nskb);
608 if (skb->protocol == htons(ETH_P_IP))
609 ipv4_build_icmp(skb, nskb, mtu, payload_length);
610 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
612 ipv6_build_icmp(skb, nskb, mtu, payload_length);
615 if (unlikely(compute_ip_summed(nskb, false))) {
620 ovs_vport_receive(vport, nskb);
625 static bool check_mtu(struct sk_buff *skb,
627 const struct tnl_mutable_config *mutable,
628 const struct rtable *rt, __be16 *frag_offp,
635 unsigned int packet_length = skb->len - ETH_HLEN;
637 if (OVS_CB(skb)->tun_key->ipv4_dst) {
640 frag_off = OVS_CB(skb)->tun_key->tun_flags & OVS_TNL_F_DONT_FRAGMENT ?
643 df_inherit = mutable->flags & TNL_F_DF_INHERIT;
644 pmtud = mutable->flags & TNL_F_PMTUD;
645 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
648 /* Allow for one level of tagging in the packet length. */
649 if (!vlan_tx_tag_present(skb) &&
650 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
651 packet_length -= VLAN_HLEN;
656 /* The tag needs to go in packet regardless of where it
657 * currently is, so subtract it from the MTU.
659 if (vlan_tx_tag_present(skb) ||
660 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
661 vlan_header = VLAN_HLEN;
663 mtu = dst_mtu(&rt_dst(rt))
669 if (skb->protocol == htons(ETH_P_IP)) {
670 struct iphdr *iph = ip_hdr(skb);
673 frag_off = iph->frag_off & htons(IP_DF);
675 if (pmtud && iph->frag_off & htons(IP_DF)) {
676 mtu = max(mtu, IP_MIN_MTU);
678 if (packet_length > mtu &&
679 ovs_tnl_frag_needed(vport, mutable, skb, mtu))
683 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
684 else if (skb->protocol == htons(ETH_P_IPV6)) {
685 /* IPv6 requires end hosts to do fragmentation
686 * if the packet is above the minimum MTU.
688 if (df_inherit && packet_length > IPV6_MIN_MTU)
689 frag_off = htons(IP_DF);
692 mtu = max(mtu, IPV6_MIN_MTU);
694 if (packet_length > mtu &&
695 ovs_tnl_frag_needed(vport, mutable, skb, mtu))
701 *frag_offp = frag_off;
705 static struct rtable *find_route(struct net *net,
706 __be32 *saddr, __be32 daddr, u8 ipproto,
710 /* Tunnel configuration keeps DSCP part of TOS bits, But Linux
711 * router expect RT_TOS bits only. */
713 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
714 struct flowi fl = { .nl_u = { .ip4_u = {
717 .tos = RT_TOS(tos) } },
720 if (unlikely(ip_route_output_key(net, &rt, &fl)))
721 return ERR_PTR(-EADDRNOTAVAIL);
722 *saddr = fl.nl_u.ip4_u.saddr;
725 struct flowi4 fl = { .daddr = daddr,
727 .flowi4_tos = RT_TOS(tos),
728 .flowi4_proto = ipproto };
730 rt = ip_route_output_key(net, &fl);
736 static bool need_linearize(const struct sk_buff *skb)
740 if (unlikely(skb_shinfo(skb)->frag_list))
744 * Generally speaking we should linearize if there are paged frags.
745 * However, if all of the refcounts are 1 we know nobody else can
746 * change them from underneath us and we can skip the linearization.
748 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
749 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
755 static struct sk_buff *handle_offloads(struct sk_buff *skb,
756 const struct tnl_mutable_config *mutable,
757 const struct rtable *rt,
763 min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
765 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
767 if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
768 int head_delta = SKB_DATA_ALIGN(min_headroom -
771 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
777 forward_ip_summed(skb, true);
779 if (skb_is_gso(skb)) {
780 struct sk_buff *nskb;
782 nskb = skb_gso_segment(skb, 0);
791 } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
792 /* Pages aren't locked and could change at any time.
793 * If this happens after we compute the checksum, the
794 * checksum will be wrong. We linearize now to avoid
797 if (unlikely(need_linearize(skb))) {
798 err = __skb_linearize(skb);
803 err = skb_checksum_help(skb);
808 set_ip_summed(skb, OVS_CSUM_NONE);
818 static int send_frags(struct sk_buff *skb,
825 struct sk_buff *next = skb->next;
826 int frag_len = skb->len - tunnel_hlen;
830 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
832 err = ip_local_out(skb);
834 if (unlikely(net_xmit_eval(err)))
836 sent_len += frag_len;
843 * There's no point in continuing to send fragments once one has been
844 * dropped so just free the rest. This may help improve the congestion
845 * that caused the first packet to be dropped.
847 ovs_tnl_free_linked_skbs(skb);
851 int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
853 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
854 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
855 enum vport_err_type err = VPORT_E_TX_ERROR;
857 struct ovs_key_ipv4_tunnel tun_key;
866 /* Validate the protocol headers before we try to use them. */
867 if (skb->protocol == htons(ETH_P_8021Q) &&
868 !vlan_tx_tag_present(skb)) {
869 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
872 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
873 skb_set_network_header(skb, VLAN_ETH_HLEN);
876 if (skb->protocol == htons(ETH_P_IP)) {
877 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
878 + sizeof(struct iphdr))))
881 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
882 else if (skb->protocol == htons(ETH_P_IPV6)) {
883 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
884 + sizeof(struct ipv6hdr))))
889 /* If OVS_CB(skb)->tun_key is NULL, point it at the local tun_key here,
892 if (!OVS_CB(skb)->tun_key) {
893 memset(&tun_key, 0, sizeof(tun_key));
894 OVS_CB(skb)->tun_key = &tun_key;
897 tunnel_hlen = tnl_vport->tnl_ops->hdr_len(mutable, OVS_CB(skb)->tun_key);
898 if (unlikely(tunnel_hlen < 0)) {
899 err = VPORT_E_TX_DROPPED;
902 tunnel_hlen += sizeof(struct iphdr);
904 if (OVS_CB(skb)->tun_key->ipv4_dst) {
905 daddr = OVS_CB(skb)->tun_key->ipv4_dst;
906 saddr = OVS_CB(skb)->tun_key->ipv4_src;
907 tos = OVS_CB(skb)->tun_key->ipv4_tos;
908 ttl = OVS_CB(skb)->tun_key->ipv4_ttl;
911 daddr = mutable->key.daddr;
912 saddr = mutable->key.saddr;
914 if (unlikely(!daddr)) {
915 /* Trying to sent packet from Null-port without
916 * tunnel info? Drop this packet. */
917 err = VPORT_E_TX_DROPPED;
922 if (skb->protocol == htons(ETH_P_IP))
923 inner_tos = ip_hdr(skb)->tos;
924 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
925 else if (skb->protocol == htons(ETH_P_IPV6))
926 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
931 if (mutable->flags & TNL_F_TOS_INHERIT)
936 tos = INET_ECN_encapsulate(tos, inner_tos);
940 if (mutable->flags & TNL_F_TTL_INHERIT) {
941 if (skb->protocol == htons(ETH_P_IP))
942 ttl = ip_hdr(skb)->ttl;
943 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
944 else if (skb->protocol == htons(ETH_P_IPV6))
945 ttl = ipv6_hdr(skb)->hop_limit;
952 rt = find_route(port_key_get_net(&mutable->key), &saddr, daddr,
953 tnl_vport->tnl_ops->ipproto, tos);
961 skb_clear_rxhash(skb);
964 skb = handle_offloads(skb, mutable, rt, tunnel_hlen);
971 if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off, tunnel_hlen))) {
972 err = VPORT_E_TX_DROPPED;
977 if (!OVS_CB(skb)->tun_key->ipv4_dst) {
978 if (!(mutable->flags & TNL_F_TTL_INHERIT)) {
980 ttl = ip4_dst_hoplimit(&rt_dst(rt));
986 struct sk_buff *next_skb = skb->next;
989 if (unlikely(vlan_deaccel_tag(skb)))
992 skb_push(skb, tunnel_hlen);
993 skb_reset_network_header(skb);
994 skb_set_transport_header(skb, sizeof(struct iphdr));
997 skb_dst_set(skb, dst_clone(&rt_dst(rt)));
999 skb_dst_set(skb, &rt_dst(rt));
1001 /* Push IP header. */
1004 iph->ihl = sizeof(struct iphdr) >> 2;
1005 iph->protocol = tnl_vport->tnl_ops->ipproto;
1010 iph->frag_off = frag_off;
1011 ip_select_ident(iph, &rt_dst(rt), NULL);
1013 /* Push Tunnel header. */
1014 skb = tnl_vport->tnl_ops->build_header(vport, mutable,
1015 &rt_dst(rt), skb, tunnel_hlen);
1019 sent_len += send_frags(skb, tunnel_hlen);
1025 if (unlikely(sent_len == 0))
1026 ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
1033 ovs_tnl_free_linked_skbs(skb);
1034 ovs_vport_record_error(vport, err);
1038 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
1039 [OVS_TUNNEL_ATTR_FLAGS] = { .type = NLA_U32 },
1040 [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1041 [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1042 [OVS_TUNNEL_ATTR_OUT_KEY] = { .type = NLA_U64 },
1043 [OVS_TUNNEL_ATTR_IN_KEY] = { .type = NLA_U64 },
1044 [OVS_TUNNEL_ATTR_TOS] = { .type = NLA_U8 },
1045 [OVS_TUNNEL_ATTR_TTL] = { .type = NLA_U8 },
1046 [OVS_TUNNEL_ATTR_DST_PORT] = { .type = NLA_U16 },
1049 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be
1051 static int tnl_set_config(struct net *net, struct nlattr *options,
1052 const struct tnl_ops *tnl_ops,
1053 const struct vport *cur_vport,
1054 struct tnl_mutable_config *mutable)
1056 const struct vport *old_vport;
1057 const struct tnl_mutable_config *old_mutable;
1058 struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
1061 port_key_set_net(&mutable->key, net);
1062 mutable->key.tunnel_type = tnl_ops->tunnel_type;
1066 err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
1070 if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4])
1073 mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1074 mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
1076 if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
1077 if (ipv4_is_multicast(mutable->key.daddr))
1079 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
1082 if (a[OVS_TUNNEL_ATTR_TOS]) {
1083 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
1084 /* Reject ToS config with ECN bits set. */
1085 if (mutable->tos & INET_ECN_MASK)
1089 if (a[OVS_TUNNEL_ATTR_TTL])
1090 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
1092 if (a[OVS_TUNNEL_ATTR_DST_PORT])
1094 htons(nla_get_u16(a[OVS_TUNNEL_ATTR_DST_PORT]));
1096 if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
1097 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
1098 mutable->flags |= TNL_F_IN_KEY_MATCH;
1100 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
1101 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
1104 if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
1105 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1107 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
1110 if (ipv4_is_multicast(mutable->key.daddr)) {
1111 struct net_device *dev;
1113 __be32 saddr = mutable->key.saddr;
1115 rt = find_route(port_key_get_net(&mutable->key),
1116 &saddr, mutable->key.daddr,
1117 tnl_ops->ipproto, mutable->tos);
1119 return -EADDRNOTAVAIL;
1120 dev = rt_dst(rt).dev;
1122 if (__in_dev_get_rtnl(dev) == NULL)
1123 return -EADDRNOTAVAIL;
1124 mutable->mlink = dev->ifindex;
1125 ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
1129 old_vport = port_table_lookup(&mutable->key, &old_mutable);
1130 if (old_vport && old_vport != cur_vport)
1136 struct vport *ovs_tnl_create(const struct vport_parms *parms,
1137 const struct vport_ops *vport_ops,
1138 const struct tnl_ops *tnl_ops)
1140 struct vport *vport;
1141 struct tnl_vport *tnl_vport;
1142 struct tnl_mutable_config *mutable;
1143 int initial_frag_id;
1146 vport = ovs_vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1147 if (IS_ERR(vport)) {
1148 err = PTR_ERR(vport);
1152 tnl_vport = tnl_vport_priv(vport);
1154 strcpy(tnl_vport->name, parms->name);
1155 tnl_vport->tnl_ops = tnl_ops;
1157 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1160 goto error_free_vport;
1163 random_ether_addr(mutable->eth_addr);
1165 get_random_bytes(&initial_frag_id, sizeof(int));
1166 atomic_set(&tnl_vport->frag_id, initial_frag_id);
1168 err = tnl_set_config(ovs_dp_get_net(parms->dp), parms->options, tnl_ops,
1171 goto error_free_mutable;
1173 rcu_assign_pointer(tnl_vport->mutable, mutable);
1175 port_table_add_port(vport);
1179 free_mutable_rtnl(mutable);
1182 ovs_vport_free(vport);
1184 return ERR_PTR(err);
1187 int ovs_tnl_set_options(struct vport *vport, struct nlattr *options)
1189 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1190 const struct tnl_mutable_config *old_mutable;
1191 struct tnl_mutable_config *mutable;
1194 old_mutable = rtnl_dereference(tnl_vport->mutable);
1195 if (!old_mutable->key.daddr)
1198 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1204 /* Copy fields whose values should be retained. */
1205 mutable->seq = old_mutable->seq + 1;
1206 memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1208 /* Parse the others configured by userspace. */
1209 err = tnl_set_config(ovs_dp_get_net(vport->dp), options, tnl_vport->tnl_ops,
1214 if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
1215 port_table_move_port(vport, mutable);
1217 assign_config_rcu(vport, mutable);
1222 free_mutable_rtnl(mutable);
1228 int ovs_tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1230 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1231 const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1233 if (nla_put_u32(skb, OVS_TUNNEL_ATTR_FLAGS,
1234 mutable->flags & TNL_F_PUBLIC) ||
1235 nla_put_be32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr))
1236 goto nla_put_failure;
1238 if (!(mutable->flags & TNL_F_IN_KEY_MATCH) &&
1239 nla_put_be64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key))
1240 goto nla_put_failure;
1241 if (!(mutable->flags & TNL_F_OUT_KEY_ACTION) &&
1242 nla_put_be64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key))
1243 goto nla_put_failure;
1244 if (mutable->key.saddr &&
1245 nla_put_be32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr))
1246 goto nla_put_failure;
1247 if (mutable->tos && nla_put_u8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos))
1248 goto nla_put_failure;
1249 if (mutable->ttl && nla_put_u8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl))
1250 goto nla_put_failure;
1251 if (mutable->dst_port && nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT,
1252 ntohs(mutable->dst_port)))
1253 goto nla_put_failure;
1261 static void free_port_rcu(struct rcu_head *rcu)
1263 struct tnl_vport *tnl_vport = container_of(rcu,
1264 struct tnl_vport, rcu);
1266 kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1267 ovs_vport_free(tnl_vport_to_vport(tnl_vport));
1270 void ovs_tnl_destroy(struct vport *vport)
1272 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1273 struct tnl_mutable_config *mutable;
1275 mutable = rtnl_dereference(tnl_vport->mutable);
1276 port_table_remove_port(vport);
1277 free_mutable_rtnl(mutable);
1278 call_rcu(&tnl_vport->rcu, free_port_rcu);
1281 int ovs_tnl_set_addr(struct vport *vport, const unsigned char *addr)
1283 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1284 struct tnl_mutable_config *old_mutable, *mutable;
1286 old_mutable = rtnl_dereference(tnl_vport->mutable);
1287 mutable = kmemdup(old_mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1291 old_mutable->mlink = 0;
1293 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1294 assign_config_rcu(vport, mutable);
1299 const char *ovs_tnl_get_name(const struct vport *vport)
1301 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1302 return tnl_vport->name;
1305 const unsigned char *ovs_tnl_get_addr(const struct vport *vport)
1307 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1308 return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1311 void ovs_tnl_free_linked_skbs(struct sk_buff *skb)
1314 struct sk_buff *next = skb->next;
1320 int ovs_tnl_init(void)
1324 port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
1329 for (i = 0; i < PORT_TABLE_SIZE; i++)
1330 INIT_HLIST_HEAD(&port_table[i]);
1335 void ovs_tnl_exit(void)