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 u64 flow_used_time(unsigned long flow_jiffies)
87 struct timespec cur_ts;
90 ktime_get_ts(&cur_ts);
91 idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
92 cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
93 cur_ts.tv_nsec / NSEC_PER_MSEC;
95 return cur_ms - idle_ms;
99 #define TCP_FLAGS_OFFSET 13
100 #define TCP_FLAG_MASK 0x3f
102 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
106 if (flow->key.dl_type == htons(ETH_P_IP) &&
107 flow->key.nw_proto == IPPROTO_TCP) {
108 u8 *tcp = (u8 *)tcp_hdr(skb);
109 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
112 spin_lock_bh(&flow->lock);
113 flow->used = jiffies;
114 flow->packet_count++;
115 flow->byte_count += skb->len;
116 flow->tcp_flags |= tcp_flags;
117 spin_unlock_bh(&flow->lock);
120 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
122 int actions_len = nla_len(actions);
123 struct sw_flow_actions *sfa;
125 /* At least DP_MAX_PORTS actions are required to be able to flood a
126 * packet to every port. Factor of 2 allows for setting VLAN tags,
128 if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
129 return ERR_PTR(-EINVAL);
131 sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
133 return ERR_PTR(-ENOMEM);
135 sfa->actions_len = actions_len;
136 memcpy(sfa->actions, nla_data(actions), actions_len);
140 struct sw_flow *flow_alloc(void)
142 struct sw_flow *flow;
144 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
146 return ERR_PTR(-ENOMEM);
148 spin_lock_init(&flow->lock);
149 atomic_set(&flow->refcnt, 1);
155 void flow_free_tbl(struct tbl_node *node)
157 struct sw_flow *flow = flow_cast(node);
163 /* RCU callback used by flow_deferred_free. */
164 static void rcu_free_flow_callback(struct rcu_head *rcu)
166 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
172 /* Schedules 'flow' to be freed after the next RCU grace period.
173 * The caller must hold rcu_read_lock for this to be sensible. */
174 void flow_deferred_free(struct sw_flow *flow)
176 call_rcu(&flow->rcu, rcu_free_flow_callback);
179 void flow_hold(struct sw_flow *flow)
181 atomic_inc(&flow->refcnt);
184 void flow_put(struct sw_flow *flow)
189 if (atomic_dec_and_test(&flow->refcnt)) {
190 kfree((struct sf_flow_acts __force *)flow->sf_acts);
191 kmem_cache_free(flow_cache, flow);
195 /* RCU callback used by flow_deferred_free_acts. */
196 static void rcu_free_acts_callback(struct rcu_head *rcu)
198 struct sw_flow_actions *sf_acts = container_of(rcu,
199 struct sw_flow_actions, rcu);
203 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
204 * The caller must hold rcu_read_lock for this to be sensible. */
205 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
207 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
210 static void parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
213 __be16 eth_type; /* ETH_P_8021Q */
216 struct qtag_prefix *qp;
218 if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
221 qp = (struct qtag_prefix *) skb->data;
222 key->dl_tci = qp->tci | htons(VLAN_TAG_PRESENT);
223 __skb_pull(skb, sizeof(struct qtag_prefix));
226 static __be16 parse_ethertype(struct sk_buff *skb)
228 struct llc_snap_hdr {
229 u8 dsap; /* Always 0xAA */
230 u8 ssap; /* Always 0xAA */
235 struct llc_snap_hdr *llc;
238 proto = *(__be16 *) skb->data;
239 __skb_pull(skb, sizeof(__be16));
241 if (ntohs(proto) >= 1536)
244 if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
245 return htons(ETH_P_802_2);
247 llc = (struct llc_snap_hdr *) skb->data;
248 if (llc->dsap != LLC_SAP_SNAP ||
249 llc->ssap != LLC_SAP_SNAP ||
250 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
251 return htons(ETH_P_802_2);
253 __skb_pull(skb, sizeof(struct llc_snap_hdr));
254 return llc->ethertype;
258 * flow_extract - extracts a flow key from an Ethernet frame.
259 * @skb: sk_buff that contains the frame, with skb->data pointing to the
261 * @in_port: port number on which @skb was received.
262 * @key: output flow key
263 * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
264 * not contain an IPv4 packet or if it is not a fragment.
266 * The caller must ensure that skb->len >= ETH_HLEN.
268 * Returns 0 if successful, otherwise a negative errno value.
270 * Initializes @skb header pointers as follows:
272 * - skb->mac_header: the Ethernet header.
274 * - skb->network_header: just past the Ethernet header, or just past the
275 * VLAN header, to the first byte of the Ethernet payload.
277 * - skb->transport_header: If key->dl_type is ETH_P_IP on output, then just
278 * past the IPv4 header, if one is present and of a correct length,
279 * otherwise the same as skb->network_header. For other key->dl_type
280 * values it is left untouched.
282 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
287 memset(key, 0, sizeof(*key));
288 key->tun_id = OVS_CB(skb)->tun_id;
289 key->in_port = in_port;
293 * We would really like to pull as many bytes as we could possibly
294 * want to parse into the linear data area. Currently that is:
298 * 60 max IP header with options
299 * 20 max TCP/UDP/ICMP header (don't care about options)
303 * But Xen only allocates 64 or 72 bytes for the linear data area in
304 * netback, which means that we would reallocate and copy the skb's
305 * linear data on every packet if we did that. So instead just pull 64
306 * bytes, which is always sufficient without IP options, and then check
307 * whether we need to pull more later when we look at the IP header.
309 if (!pskb_may_pull(skb, min(skb->len, 64u)))
312 skb_reset_mac_header(skb);
316 memcpy(key->dl_src, eth->h_source, ETH_ALEN);
317 memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
319 /* dl_type, dl_vlan, dl_vlan_pcp. */
320 __skb_pull(skb, 2 * ETH_ALEN);
321 if (eth->h_proto == htons(ETH_P_8021Q))
322 parse_vlan(skb, key);
323 key->dl_type = parse_ethertype(skb);
324 skb_reset_network_header(skb);
325 __skb_push(skb, skb->data - (unsigned char *)eth);
328 if (key->dl_type == htons(ETH_P_IP)) {
332 error = check_iphdr(skb);
333 if (unlikely(error)) {
334 if (error == -EINVAL) {
335 skb->transport_header = skb->network_header;
342 key->nw_src = nh->saddr;
343 key->nw_dst = nh->daddr;
344 key->nw_tos = nh->tos & ~INET_ECN_MASK;
345 key->nw_proto = nh->protocol;
347 /* Transport layer. */
348 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
349 !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
350 if (key->nw_proto == IPPROTO_TCP) {
351 if (tcphdr_ok(skb)) {
352 struct tcphdr *tcp = tcp_hdr(skb);
353 key->tp_src = tcp->source;
354 key->tp_dst = tcp->dest;
356 } else if (key->nw_proto == IPPROTO_UDP) {
357 if (udphdr_ok(skb)) {
358 struct udphdr *udp = udp_hdr(skb);
359 key->tp_src = udp->source;
360 key->tp_dst = udp->dest;
362 } else if (key->nw_proto == IPPROTO_ICMP) {
363 if (icmphdr_ok(skb)) {
364 struct icmphdr *icmp = icmp_hdr(skb);
365 /* The ICMP type and code fields use the 16-bit
366 * transport port fields, so we need to store them
367 * in 16-bit network byte order. */
368 key->tp_src = htons(icmp->type);
369 key->tp_dst = htons(icmp->code);
375 } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
376 struct arp_eth_header *arp;
378 arp = (struct arp_eth_header *)skb_network_header(skb);
380 if (arp->ar_hrd == htons(ARPHRD_ETHER)
381 && arp->ar_pro == htons(ETH_P_IP)
382 && arp->ar_hln == ETH_ALEN
383 && arp->ar_pln == 4) {
385 /* We only match on the lower 8 bits of the opcode. */
386 if (ntohs(arp->ar_op) <= 0xff)
387 key->nw_proto = ntohs(arp->ar_op);
389 if (key->nw_proto == ARPOP_REQUEST
390 || key->nw_proto == ARPOP_REPLY) {
391 memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
392 memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
393 memcpy(key->arp_sha, arp->ar_sha, ETH_ALEN);
394 memcpy(key->arp_tha, arp->ar_tha, ETH_ALEN);
401 u32 flow_hash(const struct sw_flow_key *key)
403 return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed);
406 int flow_cmp(const struct tbl_node *node, void *key2_)
408 const struct sw_flow_key *key1 = &flow_cast(node)->key;
409 const struct sw_flow_key *key2 = key2_;
411 return !memcmp(key1, key2, sizeof(struct sw_flow_key));
415 * flow_from_nlattrs - parses Netlink attributes into a flow key.
416 * @swkey: receives the extracted flow key.
417 * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
420 * This state machine accepts the following forms, with [] for optional
421 * elements and | for alternatives:
423 * [tun_id] in_port ethernet [8021q] [ethertype [IP [TCP|UDP|ICMP] | ARP]
425 int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr)
427 const struct nlattr *nla;
431 memset(swkey, 0, sizeof(*swkey));
432 swkey->dl_type = htons(ETH_P_802_2);
434 prev_type = ODP_KEY_ATTR_UNSPEC;
435 nla_for_each_nested(nla, attr, rem) {
436 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
437 [ODP_KEY_ATTR_TUN_ID] = 8,
438 [ODP_KEY_ATTR_IN_PORT] = 4,
439 [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
440 [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
441 [ODP_KEY_ATTR_ETHERTYPE] = 2,
442 [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
443 [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
444 [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
445 [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
446 [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
449 const struct odp_key_ethernet *eth_key;
450 const struct odp_key_8021q *q_key;
451 const struct odp_key_ipv4 *ipv4_key;
452 const struct odp_key_tcp *tcp_key;
453 const struct odp_key_udp *udp_key;
454 const struct odp_key_icmp *icmp_key;
455 const struct odp_key_arp *arp_key;
457 int type = nla_type(nla);
459 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
462 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
463 switch (TRANSITION(prev_type, type)) {
464 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
465 swkey->tun_id = nla_get_be64(nla);
468 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
469 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
470 if (nla_get_u32(nla) >= DP_MAX_PORTS)
472 swkey->in_port = nla_get_u32(nla);
475 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
476 eth_key = nla_data(nla);
477 memcpy(swkey->dl_src, eth_key->eth_src, ETH_ALEN);
478 memcpy(swkey->dl_dst, eth_key->eth_dst, ETH_ALEN);
481 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
482 q_key = nla_data(nla);
483 /* Only standard 0x8100 VLANs currently supported. */
484 if (q_key->q_tpid != htons(ETH_P_8021Q))
486 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
488 swkey->dl_tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
491 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
492 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
493 swkey->dl_type = nla_get_be16(nla);
494 if (ntohs(swkey->dl_type) < 1536)
498 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
499 if (swkey->dl_type != htons(ETH_P_IP))
501 ipv4_key = nla_data(nla);
502 swkey->nw_src = ipv4_key->ipv4_src;
503 swkey->nw_dst = ipv4_key->ipv4_dst;
504 swkey->nw_proto = ipv4_key->ipv4_proto;
505 swkey->nw_tos = ipv4_key->ipv4_tos;
506 if (swkey->nw_tos & INET_ECN_MASK)
510 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
511 if (swkey->nw_proto != IPPROTO_TCP)
513 tcp_key = nla_data(nla);
514 swkey->tp_src = tcp_key->tcp_src;
515 swkey->tp_dst = tcp_key->tcp_dst;
518 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
519 if (swkey->nw_proto != IPPROTO_UDP)
521 udp_key = nla_data(nla);
522 swkey->tp_src = udp_key->udp_src;
523 swkey->tp_dst = udp_key->udp_dst;
526 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
527 if (swkey->nw_proto != IPPROTO_ICMP)
529 icmp_key = nla_data(nla);
530 swkey->tp_src = htons(icmp_key->icmp_type);
531 swkey->tp_dst = htons(icmp_key->icmp_code);
534 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
535 if (swkey->dl_type != htons(ETH_P_ARP))
537 arp_key = nla_data(nla);
538 swkey->nw_src = arp_key->arp_sip;
539 swkey->nw_dst = arp_key->arp_tip;
540 if (arp_key->arp_op & htons(0xff00))
542 swkey->nw_proto = ntohs(arp_key->arp_op);
543 memcpy(swkey->arp_sha, arp_key->arp_sha, ETH_ALEN);
544 memcpy(swkey->arp_tha, arp_key->arp_tha, ETH_ALEN);
557 case ODP_KEY_ATTR_UNSPEC:
560 case ODP_KEY_ATTR_TUN_ID:
561 case ODP_KEY_ATTR_IN_PORT:
564 case ODP_KEY_ATTR_ETHERNET:
565 case ODP_KEY_ATTR_8021Q:
568 case ODP_KEY_ATTR_ETHERTYPE:
569 if (swkey->dl_type == htons(ETH_P_IP) ||
570 swkey->dl_type == htons(ETH_P_ARP))
574 case ODP_KEY_ATTR_IPV4:
575 if (swkey->nw_proto == IPPROTO_TCP ||
576 swkey->nw_proto == IPPROTO_UDP ||
577 swkey->nw_proto == IPPROTO_ICMP)
581 case ODP_KEY_ATTR_TCP:
582 case ODP_KEY_ATTR_UDP:
583 case ODP_KEY_ATTR_ICMP:
584 case ODP_KEY_ATTR_ARP:
592 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
594 struct odp_key_ethernet *eth_key;
597 if (swkey->tun_id != cpu_to_be64(0))
598 NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id);
600 NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port);
602 nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
604 goto nla_put_failure;
605 eth_key = nla_data(nla);
606 memcpy(eth_key->eth_src, swkey->dl_src, ETH_ALEN);
607 memcpy(eth_key->eth_dst, swkey->dl_dst, ETH_ALEN);
609 if (swkey->dl_tci != htons(0)) {
610 struct odp_key_8021q q_key;
612 q_key.q_tpid = htons(ETH_P_8021Q);
613 q_key.q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT);
614 NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
617 if (swkey->dl_type == htons(ETH_P_802_2))
620 NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type);
622 if (swkey->dl_type == htons(ETH_P_IP)) {
623 struct odp_key_ipv4 *ipv4_key;
625 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
627 goto nla_put_failure;
628 ipv4_key = nla_data(nla);
629 ipv4_key->ipv4_src = swkey->nw_src;
630 ipv4_key->ipv4_dst = swkey->nw_dst;
631 ipv4_key->ipv4_proto = swkey->nw_proto;
632 ipv4_key->ipv4_tos = swkey->nw_tos;
634 if (swkey->nw_proto == IPPROTO_TCP) {
635 struct odp_key_tcp *tcp_key;
637 nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
639 goto nla_put_failure;
640 tcp_key = nla_data(nla);
641 tcp_key->tcp_src = swkey->tp_src;
642 tcp_key->tcp_dst = swkey->tp_dst;
643 } else if (swkey->nw_proto == IPPROTO_UDP) {
644 struct odp_key_udp *udp_key;
646 nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
648 goto nla_put_failure;
649 udp_key = nla_data(nla);
650 udp_key->udp_src = swkey->tp_src;
651 udp_key->udp_dst = swkey->tp_dst;
652 } else if (swkey->nw_proto == IPPROTO_ICMP) {
653 struct odp_key_icmp *icmp_key;
655 nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
657 goto nla_put_failure;
658 icmp_key = nla_data(nla);
659 icmp_key->icmp_type = ntohs(swkey->tp_src);
660 icmp_key->icmp_code = ntohs(swkey->tp_dst);
662 } else if (swkey->dl_type == htons(ETH_P_ARP)) {
663 struct odp_key_arp *arp_key;
665 nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
667 goto nla_put_failure;
668 arp_key = nla_data(nla);
669 arp_key->arp_sip = swkey->nw_src;
670 arp_key->arp_tip = swkey->nw_dst;
671 arp_key->arp_op = htons(swkey->nw_proto);
672 memcpy(arp_key->arp_sha, swkey->arp_sha, ETH_ALEN);
673 memcpy(arp_key->arp_tha, swkey->arp_tha, ETH_ALEN);
682 /* Initializes the flow module.
683 * Returns zero if successful or a negative error code. */
686 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
688 if (flow_cache == NULL)
691 get_random_bytes(&hash_seed, sizeof(hash_seed));
696 /* Uninitializes the flow module. */
699 kmem_cache_destroy(flow_cache);