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(u32 actions_len)
108 struct sw_flow_actions *sfa;
110 if (actions_len % NLA_ALIGNTO)
111 return ERR_PTR(-EINVAL);
113 /* At least DP_MAX_PORTS actions are required to be able to flood a
114 * packet to every port. Factor of 2 allows for setting VLAN tags,
116 if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
117 return ERR_PTR(-EINVAL);
119 sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
121 return ERR_PTR(-ENOMEM);
123 sfa->actions_len = actions_len;
127 struct sw_flow *flow_alloc(void)
129 struct sw_flow *flow;
131 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
133 return ERR_PTR(-ENOMEM);
135 spin_lock_init(&flow->lock);
136 atomic_set(&flow->refcnt, 1);
142 void flow_free_tbl(struct tbl_node *node)
144 struct sw_flow *flow = flow_cast(node);
150 /* RCU callback used by flow_deferred_free. */
151 static void rcu_free_flow_callback(struct rcu_head *rcu)
153 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
159 /* Schedules 'flow' to be freed after the next RCU grace period.
160 * The caller must hold rcu_read_lock for this to be sensible. */
161 void flow_deferred_free(struct sw_flow *flow)
163 call_rcu(&flow->rcu, rcu_free_flow_callback);
166 void flow_hold(struct sw_flow *flow)
168 atomic_inc(&flow->refcnt);
171 void flow_put(struct sw_flow *flow)
176 if (atomic_dec_and_test(&flow->refcnt)) {
177 kfree((struct sf_flow_acts __force *)flow->sf_acts);
178 kmem_cache_free(flow_cache, flow);
182 /* RCU callback used by flow_deferred_free_acts. */
183 static void rcu_free_acts_callback(struct rcu_head *rcu)
185 struct sw_flow_actions *sf_acts = container_of(rcu,
186 struct sw_flow_actions, rcu);
190 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
191 * The caller must hold rcu_read_lock for this to be sensible. */
192 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
194 call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
197 static void parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
200 __be16 eth_type; /* ETH_P_8021Q */
203 struct qtag_prefix *qp;
205 if (skb->len < sizeof(struct qtag_prefix) + sizeof(__be16))
208 qp = (struct qtag_prefix *) skb->data;
209 key->dl_tci = qp->tci | htons(VLAN_TAG_PRESENT);
210 __skb_pull(skb, sizeof(struct qtag_prefix));
213 static __be16 parse_ethertype(struct sk_buff *skb)
215 struct llc_snap_hdr {
216 u8 dsap; /* Always 0xAA */
217 u8 ssap; /* Always 0xAA */
222 struct llc_snap_hdr *llc;
225 proto = *(__be16 *) skb->data;
226 __skb_pull(skb, sizeof(__be16));
228 if (ntohs(proto) >= 1536)
231 if (unlikely(skb->len < sizeof(struct llc_snap_hdr)))
232 return htons(ETH_P_802_2);
234 llc = (struct llc_snap_hdr *) skb->data;
235 if (llc->dsap != LLC_SAP_SNAP ||
236 llc->ssap != LLC_SAP_SNAP ||
237 (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
238 return htons(ETH_P_802_2);
240 __skb_pull(skb, sizeof(struct llc_snap_hdr));
241 return llc->ethertype;
245 * flow_extract - extracts a flow key from an Ethernet frame.
246 * @skb: sk_buff that contains the frame, with skb->data pointing to the
248 * @in_port: port number on which @skb was received.
249 * @key: output flow key
250 * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
251 * not contain an IPv4 packet or if it is not a fragment.
253 * The caller must ensure that skb->len >= ETH_HLEN.
255 * Returns 0 if successful, otherwise a negative errno value.
257 * Initializes @skb header pointers as follows:
259 * - skb->mac_header: the Ethernet header.
261 * - skb->network_header: just past the Ethernet header, or just past the
262 * VLAN header, to the first byte of the Ethernet payload.
264 * - skb->transport_header: If key->dl_type is ETH_P_IP on output, then just
265 * past the IPv4 header, if one is present and of a correct length,
266 * otherwise the same as skb->network_header. For other key->dl_type
267 * values it is left untouched.
269 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
274 memset(key, 0, sizeof(*key));
275 key->tun_id = OVS_CB(skb)->tun_id;
276 key->in_port = in_port;
280 * We would really like to pull as many bytes as we could possibly
281 * want to parse into the linear data area. Currently that is:
285 * 60 max IP header with options
286 * 20 max TCP/UDP/ICMP header (don't care about options)
290 * But Xen only allocates 64 or 72 bytes for the linear data area in
291 * netback, which means that we would reallocate and copy the skb's
292 * linear data on every packet if we did that. So instead just pull 64
293 * bytes, which is always sufficient without IP options, and then check
294 * whether we need to pull more later when we look at the IP header.
296 if (!pskb_may_pull(skb, min(skb->len, 64u)))
299 skb_reset_mac_header(skb);
303 memcpy(key->dl_src, eth->h_source, ETH_ALEN);
304 memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
306 /* dl_type, dl_vlan, dl_vlan_pcp. */
307 __skb_pull(skb, 2 * ETH_ALEN);
308 if (eth->h_proto == htons(ETH_P_8021Q))
309 parse_vlan(skb, key);
310 key->dl_type = parse_ethertype(skb);
311 skb_reset_network_header(skb);
312 __skb_push(skb, skb->data - (unsigned char *)eth);
315 if (key->dl_type == htons(ETH_P_IP)) {
319 error = check_iphdr(skb);
320 if (unlikely(error)) {
321 if (error == -EINVAL) {
322 skb->transport_header = skb->network_header;
329 key->nw_src = nh->saddr;
330 key->nw_dst = nh->daddr;
331 key->nw_tos = nh->tos & ~INET_ECN_MASK;
332 key->nw_proto = nh->protocol;
334 /* Transport layer. */
335 if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
336 !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
337 if (key->nw_proto == IPPROTO_TCP) {
338 if (tcphdr_ok(skb)) {
339 struct tcphdr *tcp = tcp_hdr(skb);
340 key->tp_src = tcp->source;
341 key->tp_dst = tcp->dest;
343 } else if (key->nw_proto == IPPROTO_UDP) {
344 if (udphdr_ok(skb)) {
345 struct udphdr *udp = udp_hdr(skb);
346 key->tp_src = udp->source;
347 key->tp_dst = udp->dest;
349 } else if (key->nw_proto == IPPROTO_ICMP) {
350 if (icmphdr_ok(skb)) {
351 struct icmphdr *icmp = icmp_hdr(skb);
352 /* The ICMP type and code fields use the 16-bit
353 * transport port fields, so we need to store them
354 * in 16-bit network byte order. */
355 key->tp_src = htons(icmp->type);
356 key->tp_dst = htons(icmp->code);
362 } else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
363 struct arp_eth_header *arp;
365 arp = (struct arp_eth_header *)skb_network_header(skb);
367 if (arp->ar_hrd == htons(ARPHRD_ETHER)
368 && arp->ar_pro == htons(ETH_P_IP)
369 && arp->ar_hln == ETH_ALEN
370 && arp->ar_pln == 4) {
372 /* We only match on the lower 8 bits of the opcode. */
373 if (ntohs(arp->ar_op) <= 0xff)
374 key->nw_proto = ntohs(arp->ar_op);
376 if (key->nw_proto == ARPOP_REQUEST
377 || key->nw_proto == ARPOP_REPLY) {
378 memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
379 memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
386 u32 flow_hash(const struct sw_flow_key *key)
388 return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed);
391 int flow_cmp(const struct tbl_node *node, void *key2_)
393 const struct sw_flow_key *key1 = &flow_cast(node)->key;
394 const struct sw_flow_key *key2 = key2_;
396 return !memcmp(key1, key2, sizeof(struct sw_flow_key));
400 * flow_from_nlattrs - parses Netlink attributes into a flow key.
401 * @swkey: receives the extracted flow key.
402 * @key: start of %ODP_KEY_ATTR_* Netlink attribute sequence.
403 * @key_len: number of bytes in @key.
405 * This state machine accepts the following forms, with [] for optional
406 * elements and | for alternatives:
408 * [tun_id] in_port ethernet [8021q] [ethertype [IP [TCP|UDP|ICMP] | ARP]
410 static int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *key, u32 key_len)
412 const struct nlattr *nla;
416 memset(swkey, 0, sizeof(*swkey));
417 swkey->dl_type = htons(ETH_P_802_2);
419 prev_type = ODP_KEY_ATTR_UNSPEC;
420 nla_for_each_attr(nla, key, key_len, rem) {
421 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
422 [ODP_KEY_ATTR_TUN_ID] = 8,
423 [ODP_KEY_ATTR_IN_PORT] = 4,
424 [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
425 [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
426 [ODP_KEY_ATTR_ETHERTYPE] = 2,
427 [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
428 [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
429 [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
430 [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
431 [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
434 const struct odp_key_ethernet *eth_key;
435 const struct odp_key_8021q *q_key;
436 const struct odp_key_ipv4 *ipv4_key;
437 const struct odp_key_tcp *tcp_key;
438 const struct odp_key_udp *udp_key;
439 const struct odp_key_icmp *icmp_key;
440 const struct odp_key_arp *arp_key;
442 int type = nla_type(nla);
444 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
447 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
448 switch (TRANSITION(prev_type, type)) {
449 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
450 swkey->tun_id = nla_get_be64(nla);
453 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
454 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
455 if (nla_get_u32(nla) >= DP_MAX_PORTS)
457 swkey->in_port = nla_get_u32(nla);
460 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
461 eth_key = nla_data(nla);
462 memcpy(swkey->dl_src, eth_key->eth_src, ETH_ALEN);
463 memcpy(swkey->dl_dst, eth_key->eth_dst, ETH_ALEN);
466 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
467 q_key = nla_data(nla);
468 /* Only standard 0x8100 VLANs currently supported. */
469 if (q_key->q_tpid != htons(ETH_P_8021Q))
471 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
473 swkey->dl_tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
476 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
477 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
478 swkey->dl_type = nla_get_be16(nla);
479 if (ntohs(swkey->dl_type) < 1536)
483 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
484 if (swkey->dl_type != htons(ETH_P_IP))
486 ipv4_key = nla_data(nla);
487 swkey->nw_src = ipv4_key->ipv4_src;
488 swkey->nw_dst = ipv4_key->ipv4_dst;
489 swkey->nw_proto = ipv4_key->ipv4_proto;
490 swkey->nw_tos = ipv4_key->ipv4_tos;
491 if (swkey->nw_tos & INET_ECN_MASK)
495 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
496 if (swkey->nw_proto != IPPROTO_TCP)
498 tcp_key = nla_data(nla);
499 swkey->tp_src = tcp_key->tcp_src;
500 swkey->tp_dst = tcp_key->tcp_dst;
503 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
504 if (swkey->nw_proto != IPPROTO_UDP)
506 udp_key = nla_data(nla);
507 swkey->tp_src = udp_key->udp_src;
508 swkey->tp_dst = udp_key->udp_dst;
511 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
512 if (swkey->nw_proto != IPPROTO_ICMP)
514 icmp_key = nla_data(nla);
515 swkey->tp_src = htons(icmp_key->icmp_type);
516 swkey->tp_dst = htons(icmp_key->icmp_code);
519 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
520 if (swkey->dl_type != htons(ETH_P_ARP))
522 arp_key = nla_data(nla);
523 swkey->nw_src = arp_key->arp_sip;
524 swkey->nw_dst = arp_key->arp_tip;
525 if (arp_key->arp_op & htons(0xff00))
527 swkey->nw_proto = ntohs(arp_key->arp_op);
540 case ODP_KEY_ATTR_UNSPEC:
543 case ODP_KEY_ATTR_TUN_ID:
544 case ODP_KEY_ATTR_IN_PORT:
547 case ODP_KEY_ATTR_ETHERNET:
548 case ODP_KEY_ATTR_8021Q:
551 case ODP_KEY_ATTR_ETHERTYPE:
552 if (swkey->dl_type == htons(ETH_P_IP) ||
553 swkey->dl_type == htons(ETH_P_ARP))
557 case ODP_KEY_ATTR_IPV4:
558 if (swkey->nw_proto == IPPROTO_TCP ||
559 swkey->nw_proto == IPPROTO_UDP ||
560 swkey->nw_proto == IPPROTO_ICMP)
564 case ODP_KEY_ATTR_TCP:
565 case ODP_KEY_ATTR_UDP:
566 case ODP_KEY_ATTR_ICMP:
567 case ODP_KEY_ATTR_ARP:
575 u32 flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
577 struct odp_key_ethernet *eth_key;
579 if (skb_tailroom(skb) < FLOW_BUFSIZE)
582 if (swkey->tun_id != cpu_to_be64(0))
583 nla_put_be64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id);
585 nla_put_u32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port);
587 eth_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key)));
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 = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_8021Q, sizeof(*q_key)));
595 q_key->q_tpid = htons(ETH_P_8021Q);
596 q_key->q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT);
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 ipv4_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key)));
608 ipv4_key->ipv4_src = swkey->nw_src;
609 ipv4_key->ipv4_dst = swkey->nw_dst;
610 ipv4_key->ipv4_proto = swkey->nw_proto;
611 ipv4_key->ipv4_tos = swkey->nw_tos;
613 if (swkey->nw_proto == IPPROTO_TCP) {
614 struct odp_key_tcp *tcp_key;
616 tcp_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key)));
617 tcp_key->tcp_src = swkey->tp_src;
618 tcp_key->tcp_dst = swkey->tp_dst;
619 } else if (swkey->nw_proto == IPPROTO_UDP) {
620 struct odp_key_udp *udp_key;
622 udp_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key)));
623 udp_key->udp_src = swkey->tp_src;
624 udp_key->udp_dst = swkey->tp_dst;
625 } else if (swkey->nw_proto == IPPROTO_ICMP) {
626 struct odp_key_icmp *icmp_key;
628 icmp_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key)));
629 icmp_key->icmp_type = ntohs(swkey->tp_src);
630 icmp_key->icmp_code = ntohs(swkey->tp_dst);
632 } else if (swkey->dl_type == htons(ETH_P_ARP)) {
633 struct odp_key_arp *arp_key;
635 arp_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key)));
636 arp_key->arp_sip = swkey->nw_src;
637 arp_key->arp_tip = swkey->nw_dst;
638 arp_key->arp_op = htons(swkey->nw_proto);
645 int flow_copy_from_user(struct sw_flow_key *swkey, const struct nlattr __user *ukey, u32 ukey_len)
647 char key[FLOW_BUFSIZE] __aligned(NLA_ALIGNTO);
649 if (ukey_len > FLOW_BUFSIZE || ukey_len % NLA_ALIGNTO)
652 if (copy_from_user(key, ukey, ukey_len))
655 return flow_from_nlattrs(swkey, (const struct nlattr *)key, ukey_len);
658 int flow_copy_to_user(struct nlattr __user *ukey, const struct sw_flow_key *swkey, u32 ukey_len)
663 skb = alloc_skb(FLOW_BUFSIZE, GFP_KERNEL);
667 retval = flow_to_nlattrs(swkey, skb);
668 if (copy_to_user(ukey, skb->data, min(skb->len, ukey_len)))
675 /* Initializes the flow module.
676 * Returns zero if successful or a negative error code. */
679 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
681 if (flow_cache == NULL)
684 get_random_bytes(&hash_seed, sizeof(hash_seed));
689 /* Uninitializes the flow module. */
692 kmem_cache_destroy(flow_cache);