2 * Distributed under the terms of the GNU GPL version 2.
3 * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
11 #include <asm/uaccess.h>
12 #include <linux/netdevice.h>
13 #include <linux/etherdevice.h>
14 #include <linux/if_ether.h>
15 #include <linux/if_vlan.h>
16 #include <net/llc_pdu.h>
17 #include <linux/kernel.h>
18 #include <linux/jhash.h>
19 #include <linux/jiffies.h>
20 #include <linux/llc.h>
21 #include <linux/module.h>
23 #include <linux/rcupdate.h>
24 #include <linux/if_arp.h>
25 #include <linux/if_ether.h>
27 #include <linux/tcp.h>
28 #include <linux/udp.h>
29 #include <linux/icmp.h>
30 #include <net/inet_ecn.h>
33 static struct kmem_cache *flow_cache;
34 static unsigned int hash_seed __read_mostly;
36 static inline bool arphdr_ok(struct sk_buff *skb)
38 return skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
41 static inline int check_iphdr(struct sk_buff *skb)
43 unsigned int nh_ofs = skb_network_offset(skb);
46 if (skb->len < nh_ofs + sizeof(struct iphdr))
49 ip_len = ip_hdrlen(skb);
50 if (ip_len < sizeof(struct iphdr) || skb->len < nh_ofs + ip_len)
54 * Pull enough header bytes to account for the IP header plus the
55 * longest transport header that we parse, currently 20 bytes for TCP.
57 if (!pskb_may_pull(skb, min(nh_ofs + ip_len + 20, skb->len)))
60 skb_set_transport_header(skb, nh_ofs + ip_len);
64 static inline bool tcphdr_ok(struct sk_buff *skb)
66 int th_ofs = skb_transport_offset(skb);
67 if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
68 int tcp_len = tcp_hdrlen(skb);
69 return (tcp_len >= sizeof(struct tcphdr)
70 && skb->len >= th_ofs + tcp_len);
75 static inline bool udphdr_ok(struct sk_buff *skb)
77 return skb->len >= skb_transport_offset(skb) + sizeof(struct udphdr);
80 static inline bool icmphdr_ok(struct sk_buff *skb)
82 return skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
85 #define TCP_FLAGS_OFFSET 13
86 #define TCP_FLAG_MASK 0x3f
88 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
92 if (flow->key.dl_type == htons(ETH_P_IP) &&
93 flow->key.nw_proto == IPPROTO_TCP) {
94 u8 *tcp = (u8 *)tcp_hdr(skb);
95 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
98 spin_lock_bh(&flow->lock);
100 flow->packet_count++;
101 flow->byte_count += skb->len;
102 flow->tcp_flags |= tcp_flags;
103 spin_unlock_bh(&flow->lock);
106 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
108 int actions_len = nla_len(actions);
109 struct sw_flow_actions *sfa;
111 /* At least DP_MAX_PORTS actions are required to be able to flood a
112 * packet to every port. Factor of 2 allows for setting VLAN tags,
114 if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
115 return ERR_PTR(-EINVAL);
117 sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
119 return ERR_PTR(-ENOMEM);
121 sfa->actions_len = actions_len;
122 memcpy(sfa->actions, nla_data(actions), actions_len);
126 struct sw_flow *flow_alloc(void)
128 struct sw_flow *flow;
130 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
132 return ERR_PTR(-ENOMEM);
134 spin_lock_init(&flow->lock);
135 atomic_set(&flow->refcnt, 1);
141 void flow_free_tbl(struct tbl_node *node)
143 struct sw_flow *flow = flow_cast(node);
149 /* RCU callback used by flow_deferred_free. */
150 static void rcu_free_flow_callback(struct rcu_head *rcu)
152 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
158 /* Schedules 'flow' to be freed after the next RCU grace period.
159 * The caller must hold rcu_read_lock for this to be sensible. */
160 void flow_deferred_free(struct sw_flow *flow)
162 call_rcu(&flow->rcu, rcu_free_flow_callback);
165 void flow_hold(struct sw_flow *flow)
167 atomic_inc(&flow->refcnt);
170 void flow_put(struct sw_flow *flow)
175 if (atomic_dec_and_test(&flow->refcnt)) {
176 kfree((struct sf_flow_acts __force *)flow->sf_acts);
177 kmem_cache_free(flow_cache, flow);
181 /* RCU callback used by flow_deferred_free_acts. */
182 static void rcu_free_acts_callback(struct rcu_head *rcu)
184 struct sw_flow_actions *sf_acts = container_of(rcu,
185 struct sw_flow_actions, rcu);
189 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
190 * The caller must hold rcu_read_lock for this to be sensible. */
191 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
193 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
196 static void parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
199 __be16 eth_type; /* ETH_P_8021Q */
202 struct qtag_prefix *qp;
204 if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
207 qp = (struct qtag_prefix *) skb->data;
208 key->dl_tci = qp->tci | htons(VLAN_TAG_PRESENT);
209 __skb_pull(skb, sizeof(struct qtag_prefix));
212 static __be16 parse_ethertype(struct sk_buff *skb)
214 struct llc_snap_hdr {
215 u8 dsap; /* Always 0xAA */
216 u8 ssap; /* Always 0xAA */
221 struct llc_snap_hdr *llc;
224 proto = *(__be16 *) skb->data;
225 __skb_pull(skb, sizeof(__be16));
227 if (ntohs(proto) >= 1536)
230 if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
231 return htons(ETH_P_802_2);
233 llc = (struct llc_snap_hdr *) skb->data;
234 if (llc->dsap != LLC_SAP_SNAP ||
235 llc->ssap != LLC_SAP_SNAP ||
236 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
237 return htons(ETH_P_802_2);
239 __skb_pull(skb, sizeof(struct llc_snap_hdr));
240 return llc->ethertype;
244 * flow_extract - extracts a flow key from an Ethernet frame.
245 * @skb: sk_buff that contains the frame, with skb->data pointing to the
247 * @in_port: port number on which @skb was received.
248 * @key: output flow key
249 * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
250 * not contain an IPv4 packet or if it is not a fragment.
252 * The caller must ensure that skb->len >= ETH_HLEN.
254 * Returns 0 if successful, otherwise a negative errno value.
256 * Initializes @skb header pointers as follows:
258 * - skb->mac_header: the Ethernet header.
260 * - skb->network_header: just past the Ethernet header, or just past the
261 * VLAN header, to the first byte of the Ethernet payload.
263 * - skb->transport_header: If key->dl_type is ETH_P_IP on output, then just
264 * past the IPv4 header, if one is present and of a correct length,
265 * otherwise the same as skb->network_header. For other key->dl_type
266 * values it is left untouched.
268 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
273 memset(key, 0, sizeof(*key));
274 key->tun_id = OVS_CB(skb)->tun_id;
275 key->in_port = in_port;
279 * We would really like to pull as many bytes as we could possibly
280 * want to parse into the linear data area. Currently that is:
284 * 60 max IP header with options
285 * 20 max TCP/UDP/ICMP header (don't care about options)
289 * But Xen only allocates 64 or 72 bytes for the linear data area in
290 * netback, which means that we would reallocate and copy the skb's
291 * linear data on every packet if we did that. So instead just pull 64
292 * bytes, which is always sufficient without IP options, and then check
293 * whether we need to pull more later when we look at the IP header.
295 if (!pskb_may_pull(skb, min(skb->len, 64u)))
298 skb_reset_mac_header(skb);
302 memcpy(key->dl_src, eth->h_source, ETH_ALEN);
303 memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
305 /* dl_type, dl_vlan, dl_vlan_pcp. */
306 __skb_pull(skb, 2 * ETH_ALEN);
307 if (eth->h_proto == htons(ETH_P_8021Q))
308 parse_vlan(skb, key);
309 key->dl_type = parse_ethertype(skb);
310 skb_reset_network_header(skb);
311 __skb_push(skb, skb->data - (unsigned char *)eth);
314 if (key->dl_type == htons(ETH_P_IP)) {
318 error = check_iphdr(skb);
319 if (unlikely(error)) {
320 if (error == -EINVAL) {
321 skb->transport_header = skb->network_header;
328 key->nw_src = nh->saddr;
329 key->nw_dst = nh->daddr;
330 key->nw_tos = nh->tos & ~INET_ECN_MASK;
331 key->nw_proto = nh->protocol;
333 /* Transport layer. */
334 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
335 !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
336 if (key->nw_proto == IPPROTO_TCP) {
337 if (tcphdr_ok(skb)) {
338 struct tcphdr *tcp = tcp_hdr(skb);
339 key->tp_src = tcp->source;
340 key->tp_dst = tcp->dest;
342 } else if (key->nw_proto == IPPROTO_UDP) {
343 if (udphdr_ok(skb)) {
344 struct udphdr *udp = udp_hdr(skb);
345 key->tp_src = udp->source;
346 key->tp_dst = udp->dest;
348 } else if (key->nw_proto == IPPROTO_ICMP) {
349 if (icmphdr_ok(skb)) {
350 struct icmphdr *icmp = icmp_hdr(skb);
351 /* The ICMP type and code fields use the 16-bit
352 * transport port fields, so we need to store them
353 * in 16-bit network byte order. */
354 key->tp_src = htons(icmp->type);
355 key->tp_dst = htons(icmp->code);
361 } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
362 struct arp_eth_header *arp;
364 arp = (struct arp_eth_header *)skb_network_header(skb);
366 if (arp->ar_hrd == htons(ARPHRD_ETHER)
367 && arp->ar_pro == htons(ETH_P_IP)
368 && arp->ar_hln == ETH_ALEN
369 && arp->ar_pln == 4) {
371 /* We only match on the lower 8 bits of the opcode. */
372 if (ntohs(arp->ar_op) <= 0xff)
373 key->nw_proto = ntohs(arp->ar_op);
375 if (key->nw_proto == ARPOP_REQUEST
376 || key->nw_proto == ARPOP_REPLY) {
377 memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
378 memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
385 u32 flow_hash(const struct sw_flow_key *key)
387 return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed);
390 int flow_cmp(const struct tbl_node *node, void *key2_)
392 const struct sw_flow_key *key1 = &flow_cast(node)->key;
393 const struct sw_flow_key *key2 = key2_;
395 return !memcmp(key1, key2, sizeof(struct sw_flow_key));
399 * flow_from_nlattrs - parses Netlink attributes into a flow key.
400 * @swkey: receives the extracted flow key.
401 * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
404 * This state machine accepts the following forms, with [] for optional
405 * elements and | for alternatives:
407 * [tun_id] in_port ethernet [8021q] [ethertype [IP [TCP|UDP|ICMP] | ARP]
409 int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
411 const struct nlattr *nla;
415 memset(swkey, 0, sizeof(*swkey));
416 swkey->dl_type = htons(ETH_P_802_2);
418 prev_type = ODP_KEY_ATTR_UNSPEC;
419 nla_for_each_nested(nla, attr, rem) {
420 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
421 [ODP_KEY_ATTR_TUN_ID] = 8,
422 [ODP_KEY_ATTR_IN_PORT] = 4,
423 [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
424 [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
425 [ODP_KEY_ATTR_ETHERTYPE] = 2,
426 [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
427 [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
428 [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
429 [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
430 [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
433 const struct odp_key_ethernet *eth_key;
434 const struct odp_key_8021q *q_key;
435 const struct odp_key_ipv4 *ipv4_key;
436 const struct odp_key_tcp *tcp_key;
437 const struct odp_key_udp *udp_key;
438 const struct odp_key_icmp *icmp_key;
439 const struct odp_key_arp *arp_key;
441 int type = nla_type(nla);
443 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
446 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
447 switch (TRANSITION(prev_type, type)) {
448 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
449 swkey->tun_id = nla_get_be64(nla);
452 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
453 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
454 if (nla_get_u32(nla) >= DP_MAX_PORTS)
456 swkey->in_port = nla_get_u32(nla);
459 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
460 eth_key = nla_data(nla);
461 memcpy(swkey->dl_src, eth_key->eth_src, ETH_ALEN);
462 memcpy(swkey->dl_dst, eth_key->eth_dst, ETH_ALEN);
465 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
466 q_key = nla_data(nla);
467 /* Only standard 0x8100 VLANs currently supported. */
468 if (q_key->q_tpid != htons(ETH_P_8021Q))
470 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
472 swkey->dl_tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
475 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
476 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
477 swkey->dl_type = nla_get_be16(nla);
478 if (ntohs(swkey->dl_type) < 1536)
482 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
483 if (swkey->dl_type != htons(ETH_P_IP))
485 ipv4_key = nla_data(nla);
486 swkey->nw_src = ipv4_key->ipv4_src;
487 swkey->nw_dst = ipv4_key->ipv4_dst;
488 swkey->nw_proto = ipv4_key->ipv4_proto;
489 swkey->nw_tos = ipv4_key->ipv4_tos;
490 if (swkey->nw_tos & INET_ECN_MASK)
494 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
495 if (swkey->nw_proto != IPPROTO_TCP)
497 tcp_key = nla_data(nla);
498 swkey->tp_src = tcp_key->tcp_src;
499 swkey->tp_dst = tcp_key->tcp_dst;
502 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
503 if (swkey->nw_proto != IPPROTO_UDP)
505 udp_key = nla_data(nla);
506 swkey->tp_src = udp_key->udp_src;
507 swkey->tp_dst = udp_key->udp_dst;
510 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
511 if (swkey->nw_proto != IPPROTO_ICMP)
513 icmp_key = nla_data(nla);
514 swkey->tp_src = htons(icmp_key->icmp_type);
515 swkey->tp_dst = htons(icmp_key->icmp_code);
518 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
519 if (swkey->dl_type != htons(ETH_P_ARP))
521 arp_key = nla_data(nla);
522 swkey->nw_src = arp_key->arp_sip;
523 swkey->nw_dst = arp_key->arp_tip;
524 if (arp_key->arp_op & htons(0xff00))
526 swkey->nw_proto = ntohs(arp_key->arp_op);
539 case ODP_KEY_ATTR_UNSPEC:
542 case ODP_KEY_ATTR_TUN_ID:
543 case ODP_KEY_ATTR_IN_PORT:
546 case ODP_KEY_ATTR_ETHERNET:
547 case ODP_KEY_ATTR_8021Q:
550 case ODP_KEY_ATTR_ETHERTYPE:
551 if (swkey->dl_type == htons(ETH_P_IP) ||
552 swkey->dl_type == htons(ETH_P_ARP))
556 case ODP_KEY_ATTR_IPV4:
557 if (swkey->nw_proto == IPPROTO_TCP ||
558 swkey->nw_proto == IPPROTO_UDP ||
559 swkey->nw_proto == IPPROTO_ICMP)
563 case ODP_KEY_ATTR_TCP:
564 case ODP_KEY_ATTR_UDP:
565 case ODP_KEY_ATTR_ICMP:
566 case ODP_KEY_ATTR_ARP:
574 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
576 struct odp_key_ethernet *eth_key;
579 if (swkey->tun_id != cpu_to_be64(0))
580 NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id);
582 NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port);
584 nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
586 goto nla_put_failure;
587 eth_key = nla_data(nla);
588 memcpy(eth_key->eth_src, swkey->dl_src, ETH_ALEN);
589 memcpy(eth_key->eth_dst, swkey->dl_dst, ETH_ALEN);
591 if (swkey->dl_tci != htons(0)) {
592 struct odp_key_8021q q_key;
594 q_key.q_tpid = htons(ETH_P_8021Q);
595 q_key.q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT);
596 NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
599 if (swkey->dl_type == htons(ETH_P_802_2))
602 NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type);
604 if (swkey->dl_type == htons(ETH_P_IP)) {
605 struct odp_key_ipv4 *ipv4_key;
607 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
609 goto nla_put_failure;
610 ipv4_key = nla_data(nla);
611 ipv4_key->ipv4_src = swkey->nw_src;
612 ipv4_key->ipv4_dst = swkey->nw_dst;
613 ipv4_key->ipv4_proto = swkey->nw_proto;
614 ipv4_key->ipv4_tos = swkey->nw_tos;
616 if (swkey->nw_proto == IPPROTO_TCP) {
617 struct odp_key_tcp *tcp_key;
619 nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
621 goto nla_put_failure;
622 tcp_key = nla_data(nla);
623 tcp_key->tcp_src = swkey->tp_src;
624 tcp_key->tcp_dst = swkey->tp_dst;
625 } else if (swkey->nw_proto == IPPROTO_UDP) {
626 struct odp_key_udp *udp_key;
628 nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
630 goto nla_put_failure;
631 udp_key = nla_data(nla);
632 udp_key->udp_src = swkey->tp_src;
633 udp_key->udp_dst = swkey->tp_dst;
634 } else if (swkey->nw_proto == IPPROTO_ICMP) {
635 struct odp_key_icmp *icmp_key;
637 nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
639 goto nla_put_failure;
640 icmp_key = nla_data(nla);
641 icmp_key->icmp_type = ntohs(swkey->tp_src);
642 icmp_key->icmp_code = ntohs(swkey->tp_dst);
644 } else if (swkey->dl_type == htons(ETH_P_ARP)) {
645 struct odp_key_arp *arp_key;
647 nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
649 goto nla_put_failure;
650 arp_key = nla_data(nla);
651 arp_key->arp_sip = swkey->nw_src;
652 arp_key->arp_tip = swkey->nw_dst;
653 arp_key->arp_op = htons(swkey->nw_proto);
662 /* Initializes the flow module.
663 * Returns zero if successful or a negative error code. */
666 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
668 if (flow_cache == NULL)
671 get_random_bytes(&hash_seed, sizeof(hash_seed));
676 /* Uninitializes the flow module. */
679 kmem_cache_destroy(flow_cache);