Linux Audio

Check our new training course

Loading...
v4.10.11
  1/*
  2 * Copyright (c) 2007-2014 Nicira, Inc.
  3 *
  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.
  7 *
  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.
 12 *
 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
 16 * 02110-1301, USA
 17 */
 18
 
 
 19#include <linux/uaccess.h>
 20#include <linux/netdevice.h>
 21#include <linux/etherdevice.h>
 22#include <linux/if_ether.h>
 23#include <linux/if_vlan.h>
 24#include <net/llc_pdu.h>
 25#include <linux/kernel.h>
 26#include <linux/jhash.h>
 27#include <linux/jiffies.h>
 28#include <linux/llc.h>
 29#include <linux/module.h>
 30#include <linux/in.h>
 31#include <linux/rcupdate.h>
 32#include <linux/cpumask.h>
 33#include <linux/if_arp.h>
 34#include <linux/ip.h>
 35#include <linux/ipv6.h>
 36#include <linux/mpls.h>
 37#include <linux/sctp.h>
 38#include <linux/smp.h>
 39#include <linux/tcp.h>
 40#include <linux/udp.h>
 41#include <linux/icmp.h>
 42#include <linux/icmpv6.h>
 43#include <linux/rculist.h>
 44#include <net/ip.h>
 45#include <net/ip_tunnels.h>
 46#include <net/ipv6.h>
 47#include <net/mpls.h>
 48#include <net/ndisc.h>
 49
 50#include "conntrack.h"
 51#include "datapath.h"
 52#include "flow.h"
 53#include "flow_netlink.h"
 54#include "vport.h"
 55
 56u64 ovs_flow_used_time(unsigned long flow_jiffies)
 57{
 58	struct timespec cur_ts;
 59	u64 cur_ms, idle_ms;
 60
 61	ktime_get_ts(&cur_ts);
 62	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 63	cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 64		 cur_ts.tv_nsec / NSEC_PER_MSEC;
 65
 66	return cur_ms - idle_ms;
 67}
 68
 69#define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
 70
 71void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
 72			   const struct sk_buff *skb)
 73{
 74	struct flow_stats *stats;
 75	int node = numa_node_id();
 76	int cpu = smp_processor_id();
 77	int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
 78
 79	stats = rcu_dereference(flow->stats[cpu]);
 80
 81	/* Check if already have CPU-specific stats. */
 82	if (likely(stats)) {
 83		spin_lock(&stats->lock);
 84		/* Mark if we write on the pre-allocated stats. */
 85		if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
 86			flow->stats_last_writer = cpu;
 87	} else {
 88		stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
 89		spin_lock(&stats->lock);
 90
 91		/* If the current CPU is the only writer on the
 92		 * pre-allocated stats keep using them.
 93		 */
 94		if (unlikely(flow->stats_last_writer != cpu)) {
 95			/* A previous locker may have already allocated the
 96			 * stats, so we need to check again.  If CPU-specific
 97			 * stats were already allocated, we update the pre-
 98			 * allocated stats as we have already locked them.
 99			 */
100			if (likely(flow->stats_last_writer != -1) &&
101			    likely(!rcu_access_pointer(flow->stats[cpu]))) {
102				/* Try to allocate CPU-specific stats. */
103				struct flow_stats *new_stats;
104
105				new_stats =
106					kmem_cache_alloc_node(flow_stats_cache,
107							      GFP_NOWAIT |
108							      __GFP_THISNODE |
109							      __GFP_NOWARN |
110							      __GFP_NOMEMALLOC,
111							      node);
112				if (likely(new_stats)) {
113					new_stats->used = jiffies;
114					new_stats->packet_count = 1;
115					new_stats->byte_count = len;
116					new_stats->tcp_flags = tcp_flags;
117					spin_lock_init(&new_stats->lock);
118
119					rcu_assign_pointer(flow->stats[cpu],
120							   new_stats);
121					goto unlock;
122				}
123			}
124			flow->stats_last_writer = cpu;
125		}
126	}
127
 
128	stats->used = jiffies;
129	stats->packet_count++;
130	stats->byte_count += len;
131	stats->tcp_flags |= tcp_flags;
132unlock:
133	spin_unlock(&stats->lock);
134}
135
136/* Must be called with rcu_read_lock or ovs_mutex. */
137void ovs_flow_stats_get(const struct sw_flow *flow,
138			struct ovs_flow_stats *ovs_stats,
 
 
 
 
 
 
 
 
 
 
 
139			unsigned long *used, __be16 *tcp_flags)
140{
141	int cpu;
142
143	*used = 0;
144	*tcp_flags = 0;
145	memset(ovs_stats, 0, sizeof(*ovs_stats));
146
147	/* We open code this to make sure cpu 0 is always considered */
148	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, cpu_possible_mask)) {
149		struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
150
151		if (stats) {
152			/* Local CPU may write on non-local stats, so we must
153			 * block bottom-halves here.
154			 */
155			spin_lock_bh(&stats->lock);
156			if (!*used || time_after(stats->used, *used))
157				*used = stats->used;
158			*tcp_flags |= stats->tcp_flags;
159			ovs_stats->n_packets += stats->packet_count;
160			ovs_stats->n_bytes += stats->byte_count;
161			spin_unlock_bh(&stats->lock);
162		}
163	}
 
 
 
 
 
 
 
 
 
 
 
164}
165
166/* Called with ovs_mutex. */
167void ovs_flow_stats_clear(struct sw_flow *flow)
168{
169	int cpu;
170
171	/* We open code this to make sure cpu 0 is always considered */
172	for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, cpu_possible_mask)) {
173		struct flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
174
175		if (stats) {
176			spin_lock_bh(&stats->lock);
177			stats->used = 0;
178			stats->packet_count = 0;
179			stats->byte_count = 0;
180			stats->tcp_flags = 0;
181			spin_unlock_bh(&stats->lock);
182		}
183	}
 
184}
185
186static int check_header(struct sk_buff *skb, int len)
187{
188	if (unlikely(skb->len < len))
189		return -EINVAL;
190	if (unlikely(!pskb_may_pull(skb, len)))
191		return -ENOMEM;
192	return 0;
193}
194
195static bool arphdr_ok(struct sk_buff *skb)
196{
197	return pskb_may_pull(skb, skb_network_offset(skb) +
198				  sizeof(struct arp_eth_header));
199}
200
201static int check_iphdr(struct sk_buff *skb)
202{
203	unsigned int nh_ofs = skb_network_offset(skb);
204	unsigned int ip_len;
205	int err;
206
207	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
208	if (unlikely(err))
209		return err;
210
211	ip_len = ip_hdrlen(skb);
212	if (unlikely(ip_len < sizeof(struct iphdr) ||
213		     skb->len < nh_ofs + ip_len))
214		return -EINVAL;
215
216	skb_set_transport_header(skb, nh_ofs + ip_len);
217	return 0;
218}
219
220static bool tcphdr_ok(struct sk_buff *skb)
221{
222	int th_ofs = skb_transport_offset(skb);
223	int tcp_len;
224
225	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
226		return false;
227
228	tcp_len = tcp_hdrlen(skb);
229	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
230		     skb->len < th_ofs + tcp_len))
231		return false;
232
233	return true;
234}
235
236static bool udphdr_ok(struct sk_buff *skb)
237{
238	return pskb_may_pull(skb, skb_transport_offset(skb) +
239				  sizeof(struct udphdr));
240}
241
242static bool sctphdr_ok(struct sk_buff *skb)
243{
244	return pskb_may_pull(skb, skb_transport_offset(skb) +
245				  sizeof(struct sctphdr));
246}
247
248static bool icmphdr_ok(struct sk_buff *skb)
249{
250	return pskb_may_pull(skb, skb_transport_offset(skb) +
251				  sizeof(struct icmphdr));
252}
253
254static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
255{
256	unsigned int nh_ofs = skb_network_offset(skb);
257	unsigned int nh_len;
258	int payload_ofs;
259	struct ipv6hdr *nh;
260	uint8_t nexthdr;
261	__be16 frag_off;
262	int err;
263
264	err = check_header(skb, nh_ofs + sizeof(*nh));
265	if (unlikely(err))
266		return err;
267
268	nh = ipv6_hdr(skb);
269	nexthdr = nh->nexthdr;
270	payload_ofs = (u8 *)(nh + 1) - skb->data;
271
272	key->ip.proto = NEXTHDR_NONE;
273	key->ip.tos = ipv6_get_dsfield(nh);
274	key->ip.ttl = nh->hop_limit;
275	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
276	key->ipv6.addr.src = nh->saddr;
277	key->ipv6.addr.dst = nh->daddr;
278
279	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 
 
280
281	if (frag_off) {
282		if (frag_off & htons(~0x7))
283			key->ip.frag = OVS_FRAG_TYPE_LATER;
284		else
285			key->ip.frag = OVS_FRAG_TYPE_FIRST;
286	} else {
287		key->ip.frag = OVS_FRAG_TYPE_NONE;
288	}
289
290	/* Delayed handling of error in ipv6_skip_exthdr() as it
291	 * always sets frag_off to a valid value which may be
292	 * used to set key->ip.frag above.
293	 */
294	if (unlikely(payload_ofs < 0))
295		return -EPROTO;
296
297	nh_len = payload_ofs - nh_ofs;
298	skb_set_transport_header(skb, nh_ofs + nh_len);
299	key->ip.proto = nexthdr;
300	return nh_len;
301}
302
303static bool icmp6hdr_ok(struct sk_buff *skb)
304{
305	return pskb_may_pull(skb, skb_transport_offset(skb) +
306				  sizeof(struct icmp6hdr));
307}
308
309/**
310 * Parse vlan tag from vlan header.
311 * Returns ERROR on memory error.
312 * Returns 0 if it encounters a non-vlan or incomplete packet.
313 * Returns 1 after successfully parsing vlan tag.
314 */
315static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
316			  bool untag_vlan)
317{
318	struct vlan_head *vh = (struct vlan_head *)skb->data;
319
320	if (likely(!eth_type_vlan(vh->tpid)))
321		return 0;
 
322
323	if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
324		return 0;
325
326	if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
327				 sizeof(__be16))))
328		return -ENOMEM;
329
330	vh = (struct vlan_head *)skb->data;
331	key_vh->tci = vh->tci | htons(VLAN_TAG_PRESENT);
332	key_vh->tpid = vh->tpid;
333
334	if (unlikely(untag_vlan)) {
335		int offset = skb->data - skb_mac_header(skb);
336		u16 tci;
337		int err;
338
339		__skb_push(skb, offset);
340		err = __skb_vlan_pop(skb, &tci);
341		__skb_pull(skb, offset);
342		if (err)
343			return err;
344		__vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
345	} else {
346		__skb_pull(skb, sizeof(struct vlan_head));
347	}
348	return 1;
349}
350
351static void clear_vlan(struct sw_flow_key *key)
352{
353	key->eth.vlan.tci = 0;
354	key->eth.vlan.tpid = 0;
355	key->eth.cvlan.tci = 0;
356	key->eth.cvlan.tpid = 0;
357}
358
359static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
360{
361	int res;
362
363	if (skb_vlan_tag_present(skb)) {
364		key->eth.vlan.tci = htons(skb->vlan_tci);
365		key->eth.vlan.tpid = skb->vlan_proto;
366	} else {
367		/* Parse outer vlan tag in the non-accelerated case. */
368		res = parse_vlan_tag(skb, &key->eth.vlan, true);
369		if (res <= 0)
370			return res;
371	}
372
373	/* Parse inner vlan tag. */
374	res = parse_vlan_tag(skb, &key->eth.cvlan, false);
375	if (res <= 0)
376		return res;
377
378	return 0;
379}
380
381static __be16 parse_ethertype(struct sk_buff *skb)
382{
383	struct llc_snap_hdr {
384		u8  dsap;  /* Always 0xAA */
385		u8  ssap;  /* Always 0xAA */
386		u8  ctrl;
387		u8  oui[3];
388		__be16 ethertype;
389	};
390	struct llc_snap_hdr *llc;
391	__be16 proto;
392
393	proto = *(__be16 *) skb->data;
394	__skb_pull(skb, sizeof(__be16));
395
396	if (eth_proto_is_802_3(proto))
397		return proto;
398
399	if (skb->len < sizeof(struct llc_snap_hdr))
400		return htons(ETH_P_802_2);
401
402	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
403		return htons(0);
404
405	llc = (struct llc_snap_hdr *) skb->data;
406	if (llc->dsap != LLC_SAP_SNAP ||
407	    llc->ssap != LLC_SAP_SNAP ||
408	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
409		return htons(ETH_P_802_2);
410
411	__skb_pull(skb, sizeof(struct llc_snap_hdr));
412
413	if (eth_proto_is_802_3(llc->ethertype))
414		return llc->ethertype;
415
416	return htons(ETH_P_802_2);
417}
418
419static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
420			int nh_len)
421{
422	struct icmp6hdr *icmp = icmp6_hdr(skb);
423
424	/* The ICMPv6 type and code fields use the 16-bit transport port
425	 * fields, so we need to store them in 16-bit network byte order.
426	 */
427	key->tp.src = htons(icmp->icmp6_type);
428	key->tp.dst = htons(icmp->icmp6_code);
429	memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
430
431	if (icmp->icmp6_code == 0 &&
432	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
433	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
434		int icmp_len = skb->len - skb_transport_offset(skb);
435		struct nd_msg *nd;
436		int offset;
437
438		/* In order to process neighbor discovery options, we need the
439		 * entire packet.
440		 */
441		if (unlikely(icmp_len < sizeof(*nd)))
442			return 0;
443
444		if (unlikely(skb_linearize(skb)))
445			return -ENOMEM;
446
447		nd = (struct nd_msg *)skb_transport_header(skb);
448		key->ipv6.nd.target = nd->target;
449
450		icmp_len -= sizeof(*nd);
451		offset = 0;
452		while (icmp_len >= 8) {
453			struct nd_opt_hdr *nd_opt =
454				 (struct nd_opt_hdr *)(nd->opt + offset);
455			int opt_len = nd_opt->nd_opt_len * 8;
456
457			if (unlikely(!opt_len || opt_len > icmp_len))
458				return 0;
459
460			/* Store the link layer address if the appropriate
461			 * option is provided.  It is considered an error if
462			 * the same link layer option is specified twice.
463			 */
464			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
465			    && opt_len == 8) {
466				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
467					goto invalid;
468				ether_addr_copy(key->ipv6.nd.sll,
469						&nd->opt[offset+sizeof(*nd_opt)]);
470			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
471				   && opt_len == 8) {
472				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
473					goto invalid;
474				ether_addr_copy(key->ipv6.nd.tll,
475						&nd->opt[offset+sizeof(*nd_opt)]);
476			}
477
478			icmp_len -= opt_len;
479			offset += opt_len;
480		}
481	}
482
483	return 0;
484
485invalid:
486	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
487	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
488	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
489
490	return 0;
491}
492
493/**
494 * key_extract - extracts a flow key from an Ethernet frame.
495 * @skb: sk_buff that contains the frame, with skb->data pointing to the
496 * Ethernet header
 
497 * @key: output flow key
498 *
499 * The caller must ensure that skb->len >= ETH_HLEN.
500 *
501 * Returns 0 if successful, otherwise a negative errno value.
502 *
503 * Initializes @skb header fields as follows:
504 *
505 *    - skb->mac_header: the L2 header.
506 *
507 *    - skb->network_header: just past the L2 header, or just past the
508 *      VLAN header, to the first byte of the L2 payload.
509 *
510 *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
511 *      on output, then just past the IP header, if one is present and
512 *      of a correct length, otherwise the same as skb->network_header.
513 *      For other key->eth.type values it is left untouched.
514 *
515 *    - skb->protocol: the type of the data starting at skb->network_header.
516 *      Equals to key->eth.type.
517 */
518static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
519{
520	int error;
521	struct ethhdr *eth;
522
523	/* Flags are always used as part of stats */
524	key->tp.flags = 0;
525
526	skb_reset_mac_header(skb);
 
 
 
 
527
528	/* Link layer. */
529	clear_vlan(key);
530	if (key->mac_proto == MAC_PROTO_NONE) {
531		if (unlikely(eth_type_vlan(skb->protocol)))
532			return -EINVAL;
533
534		skb_reset_network_header(skb);
535	} else {
536		eth = eth_hdr(skb);
537		ether_addr_copy(key->eth.src, eth->h_source);
538		ether_addr_copy(key->eth.dst, eth->h_dest);
539
540		__skb_pull(skb, 2 * ETH_ALEN);
541		/* We are going to push all headers that we pull, so no need to
542		* update skb->csum here.
543		*/
 
544
 
 
 
545		if (unlikely(parse_vlan(skb, key)))
546			return -ENOMEM;
547
548		skb->protocol = parse_ethertype(skb);
549		if (unlikely(skb->protocol == htons(0)))
550			return -ENOMEM;
551
552		skb_reset_network_header(skb);
553		__skb_push(skb, skb->data - skb_mac_header(skb));
554	}
555	skb_reset_mac_len(skb);
556	key->eth.type = skb->protocol;
557
558	/* Network layer. */
559	if (key->eth.type == htons(ETH_P_IP)) {
560		struct iphdr *nh;
561		__be16 offset;
562
563		error = check_iphdr(skb);
564		if (unlikely(error)) {
565			memset(&key->ip, 0, sizeof(key->ip));
566			memset(&key->ipv4, 0, sizeof(key->ipv4));
567			if (error == -EINVAL) {
568				skb->transport_header = skb->network_header;
569				error = 0;
570			}
571			return error;
572		}
573
574		nh = ip_hdr(skb);
575		key->ipv4.addr.src = nh->saddr;
576		key->ipv4.addr.dst = nh->daddr;
577
578		key->ip.proto = nh->protocol;
579		key->ip.tos = nh->tos;
580		key->ip.ttl = nh->ttl;
581
582		offset = nh->frag_off & htons(IP_OFFSET);
583		if (offset) {
584			key->ip.frag = OVS_FRAG_TYPE_LATER;
585			return 0;
586		}
587		if (nh->frag_off & htons(IP_MF) ||
588			skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
589			key->ip.frag = OVS_FRAG_TYPE_FIRST;
590		else
591			key->ip.frag = OVS_FRAG_TYPE_NONE;
592
593		/* Transport layer. */
594		if (key->ip.proto == IPPROTO_TCP) {
595			if (tcphdr_ok(skb)) {
596				struct tcphdr *tcp = tcp_hdr(skb);
597				key->tp.src = tcp->source;
598				key->tp.dst = tcp->dest;
599				key->tp.flags = TCP_FLAGS_BE16(tcp);
600			} else {
601				memset(&key->tp, 0, sizeof(key->tp));
602			}
603
604		} else if (key->ip.proto == IPPROTO_UDP) {
605			if (udphdr_ok(skb)) {
606				struct udphdr *udp = udp_hdr(skb);
607				key->tp.src = udp->source;
608				key->tp.dst = udp->dest;
609			} else {
610				memset(&key->tp, 0, sizeof(key->tp));
611			}
612		} else if (key->ip.proto == IPPROTO_SCTP) {
613			if (sctphdr_ok(skb)) {
614				struct sctphdr *sctp = sctp_hdr(skb);
615				key->tp.src = sctp->source;
616				key->tp.dst = sctp->dest;
617			} else {
618				memset(&key->tp, 0, sizeof(key->tp));
619			}
620		} else if (key->ip.proto == IPPROTO_ICMP) {
621			if (icmphdr_ok(skb)) {
622				struct icmphdr *icmp = icmp_hdr(skb);
623				/* The ICMP type and code fields use the 16-bit
624				 * transport port fields, so we need to store
625				 * them in 16-bit network byte order. */
626				key->tp.src = htons(icmp->type);
627				key->tp.dst = htons(icmp->code);
628			} else {
629				memset(&key->tp, 0, sizeof(key->tp));
630			}
631		}
632
633	} else if (key->eth.type == htons(ETH_P_ARP) ||
634		   key->eth.type == htons(ETH_P_RARP)) {
635		struct arp_eth_header *arp;
636		bool arp_available = arphdr_ok(skb);
637
638		arp = (struct arp_eth_header *)skb_network_header(skb);
639
640		if (arp_available &&
641		    arp->ar_hrd == htons(ARPHRD_ETHER) &&
642		    arp->ar_pro == htons(ETH_P_IP) &&
643		    arp->ar_hln == ETH_ALEN &&
644		    arp->ar_pln == 4) {
645
646			/* We only match on the lower 8 bits of the opcode. */
647			if (ntohs(arp->ar_op) <= 0xff)
648				key->ip.proto = ntohs(arp->ar_op);
649			else
650				key->ip.proto = 0;
651
652			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
653			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
654			ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
655			ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
656		} else {
657			memset(&key->ip, 0, sizeof(key->ip));
658			memset(&key->ipv4, 0, sizeof(key->ipv4));
659		}
660	} else if (eth_p_mpls(key->eth.type)) {
661		size_t stack_len = MPLS_HLEN;
662
663		skb_set_inner_network_header(skb, skb->mac_len);
664		while (1) {
665			__be32 lse;
666
667			error = check_header(skb, skb->mac_len + stack_len);
668			if (unlikely(error))
669				return 0;
670
671			memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
672
673			if (stack_len == MPLS_HLEN)
674				memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
675
676			skb_set_inner_network_header(skb, skb->mac_len + stack_len);
677			if (lse & htonl(MPLS_LS_S_MASK))
678				break;
679
680			stack_len += MPLS_HLEN;
681		}
682	} else if (key->eth.type == htons(ETH_P_IPV6)) {
683		int nh_len;             /* IPv6 Header + Extensions */
684
685		nh_len = parse_ipv6hdr(skb, key);
686		if (unlikely(nh_len < 0)) {
687			switch (nh_len) {
688			case -EINVAL:
689				memset(&key->ip, 0, sizeof(key->ip));
690				memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
691				/* fall-through */
692			case -EPROTO:
693				skb->transport_header = skb->network_header;
694				error = 0;
695				break;
696			default:
697				error = nh_len;
698			}
699			return error;
700		}
701
702		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
703			return 0;
704		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
705			key->ip.frag = OVS_FRAG_TYPE_FIRST;
706
707		/* Transport layer. */
708		if (key->ip.proto == NEXTHDR_TCP) {
709			if (tcphdr_ok(skb)) {
710				struct tcphdr *tcp = tcp_hdr(skb);
711				key->tp.src = tcp->source;
712				key->tp.dst = tcp->dest;
713				key->tp.flags = TCP_FLAGS_BE16(tcp);
714			} else {
715				memset(&key->tp, 0, sizeof(key->tp));
716			}
717		} else if (key->ip.proto == NEXTHDR_UDP) {
718			if (udphdr_ok(skb)) {
719				struct udphdr *udp = udp_hdr(skb);
720				key->tp.src = udp->source;
721				key->tp.dst = udp->dest;
722			} else {
723				memset(&key->tp, 0, sizeof(key->tp));
724			}
725		} else if (key->ip.proto == NEXTHDR_SCTP) {
726			if (sctphdr_ok(skb)) {
727				struct sctphdr *sctp = sctp_hdr(skb);
728				key->tp.src = sctp->source;
729				key->tp.dst = sctp->dest;
730			} else {
731				memset(&key->tp, 0, sizeof(key->tp));
732			}
733		} else if (key->ip.proto == NEXTHDR_ICMP) {
734			if (icmp6hdr_ok(skb)) {
735				error = parse_icmpv6(skb, key, nh_len);
736				if (error)
737					return error;
738			} else {
739				memset(&key->tp, 0, sizeof(key->tp));
740			}
741		}
742	}
743	return 0;
744}
745
746int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
747{
748	return key_extract(skb, key);
749}
750
751static int key_extract_mac_proto(struct sk_buff *skb)
752{
753	switch (skb->dev->type) {
754	case ARPHRD_ETHER:
755		return MAC_PROTO_ETHERNET;
756	case ARPHRD_NONE:
757		if (skb->protocol == htons(ETH_P_TEB))
758			return MAC_PROTO_ETHERNET;
759		return MAC_PROTO_NONE;
760	}
761	WARN_ON_ONCE(1);
762	return -EINVAL;
763}
764
765int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
766			 struct sk_buff *skb, struct sw_flow_key *key)
767{
768	int res;
769
770	/* Extract metadata from packet. */
771	if (tun_info) {
772		key->tun_proto = ip_tunnel_info_af(tun_info);
773		memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
774
775		if (tun_info->options_len) {
776			BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
777						   8)) - 1
778					> sizeof(key->tun_opts));
779
780			ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
781						tun_info);
782			key->tun_opts_len = tun_info->options_len;
783		} else {
784			key->tun_opts_len = 0;
785		}
786	} else  {
787		key->tun_proto = 0;
788		key->tun_opts_len = 0;
789		memset(&key->tun_key, 0, sizeof(key->tun_key));
790	}
791
792	key->phy.priority = skb->priority;
793	key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
794	key->phy.skb_mark = skb->mark;
795	ovs_ct_fill_key(skb, key);
796	key->ovs_flow_hash = 0;
797	res = key_extract_mac_proto(skb);
798	if (res < 0)
799		return res;
800	key->mac_proto = res;
801	key->recirc_id = 0;
802
803	return key_extract(skb, key);
804}
805
806int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
807				   struct sk_buff *skb,
808				   struct sw_flow_key *key, bool log)
809{
810	int err;
811
812	/* Extract metadata from netlink attributes. */
813	err = ovs_nla_get_flow_metadata(net, attr, key, log);
814	if (err)
815		return err;
816
817	/* key_extract assumes that skb->protocol is set-up for
818	 * layer 3 packets which is the case for other callers,
819	 * in particular packets received from the network stack.
820	 * Here the correct value can be set from the metadata
821	 * extracted above.
822	 * For L2 packet key eth type would be zero. skb protocol
823	 * would be set to correct value later during key-extact.
824	 */
825
826	skb->protocol = key->eth.type;
827	return key_extract(skb, key);
828}
v3.15
  1/*
  2 * Copyright (c) 2007-2013 Nicira, Inc.
  3 *
  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.
  7 *
  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.
 12 *
 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
 16 * 02110-1301, USA
 17 */
 18
 19#include "flow.h"
 20#include "datapath.h"
 21#include <linux/uaccess.h>
 22#include <linux/netdevice.h>
 23#include <linux/etherdevice.h>
 24#include <linux/if_ether.h>
 25#include <linux/if_vlan.h>
 26#include <net/llc_pdu.h>
 27#include <linux/kernel.h>
 28#include <linux/jhash.h>
 29#include <linux/jiffies.h>
 30#include <linux/llc.h>
 31#include <linux/module.h>
 32#include <linux/in.h>
 33#include <linux/rcupdate.h>
 
 34#include <linux/if_arp.h>
 35#include <linux/ip.h>
 36#include <linux/ipv6.h>
 
 37#include <linux/sctp.h>
 38#include <linux/smp.h>
 39#include <linux/tcp.h>
 40#include <linux/udp.h>
 41#include <linux/icmp.h>
 42#include <linux/icmpv6.h>
 43#include <linux/rculist.h>
 44#include <net/ip.h>
 45#include <net/ip_tunnels.h>
 46#include <net/ipv6.h>
 
 47#include <net/ndisc.h>
 48
 
 
 
 
 
 
 49u64 ovs_flow_used_time(unsigned long flow_jiffies)
 50{
 51	struct timespec cur_ts;
 52	u64 cur_ms, idle_ms;
 53
 54	ktime_get_ts(&cur_ts);
 55	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 56	cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 57		 cur_ts.tv_nsec / NSEC_PER_MSEC;
 58
 59	return cur_ms - idle_ms;
 60}
 61
 62#define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
 63
 64void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
 
 65{
 66	struct flow_stats *stats;
 67	__be16 tcp_flags = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 68
 69	if (!flow->stats.is_percpu)
 70		stats = flow->stats.stat;
 71	else
 72		stats = this_cpu_ptr(flow->stats.cpu_stats);
 73
 74	if ((flow->key.eth.type == htons(ETH_P_IP) ||
 75	     flow->key.eth.type == htons(ETH_P_IPV6)) &&
 76	    flow->key.ip.frag != OVS_FRAG_TYPE_LATER &&
 77	    flow->key.ip.proto == IPPROTO_TCP &&
 78	    likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 79		tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 80	}
 81
 82	spin_lock(&stats->lock);
 83	stats->used = jiffies;
 84	stats->packet_count++;
 85	stats->byte_count += skb->len;
 86	stats->tcp_flags |= tcp_flags;
 
 87	spin_unlock(&stats->lock);
 88}
 89
 90static void stats_read(struct flow_stats *stats,
 91		       struct ovs_flow_stats *ovs_stats,
 92		       unsigned long *used, __be16 *tcp_flags)
 93{
 94	spin_lock(&stats->lock);
 95	if (!*used || time_after(stats->used, *used))
 96		*used = stats->used;
 97	*tcp_flags |= stats->tcp_flags;
 98	ovs_stats->n_packets += stats->packet_count;
 99	ovs_stats->n_bytes += stats->byte_count;
100	spin_unlock(&stats->lock);
101}
102
103void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
104			unsigned long *used, __be16 *tcp_flags)
105{
106	int cpu;
107
108	*used = 0;
109	*tcp_flags = 0;
110	memset(ovs_stats, 0, sizeof(*ovs_stats));
111
112	local_bh_disable();
113	if (!flow->stats.is_percpu) {
114		stats_read(flow->stats.stat, ovs_stats, used, tcp_flags);
115	} else {
116		for_each_possible_cpu(cpu) {
117			struct flow_stats *stats;
118
119			stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
120			stats_read(stats, ovs_stats, used, tcp_flags);
 
 
 
 
 
 
121		}
122	}
123	local_bh_enable();
124}
125
126static void stats_reset(struct flow_stats *stats)
127{
128	spin_lock(&stats->lock);
129	stats->used = 0;
130	stats->packet_count = 0;
131	stats->byte_count = 0;
132	stats->tcp_flags = 0;
133	spin_unlock(&stats->lock);
134}
135
 
136void ovs_flow_stats_clear(struct sw_flow *flow)
137{
138	int cpu;
139
140	local_bh_disable();
141	if (!flow->stats.is_percpu) {
142		stats_reset(flow->stats.stat);
143	} else {
144		for_each_possible_cpu(cpu) {
145			stats_reset(per_cpu_ptr(flow->stats.cpu_stats, cpu));
 
 
 
 
 
146		}
147	}
148	local_bh_enable();
149}
150
151static int check_header(struct sk_buff *skb, int len)
152{
153	if (unlikely(skb->len < len))
154		return -EINVAL;
155	if (unlikely(!pskb_may_pull(skb, len)))
156		return -ENOMEM;
157	return 0;
158}
159
160static bool arphdr_ok(struct sk_buff *skb)
161{
162	return pskb_may_pull(skb, skb_network_offset(skb) +
163				  sizeof(struct arp_eth_header));
164}
165
166static int check_iphdr(struct sk_buff *skb)
167{
168	unsigned int nh_ofs = skb_network_offset(skb);
169	unsigned int ip_len;
170	int err;
171
172	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
173	if (unlikely(err))
174		return err;
175
176	ip_len = ip_hdrlen(skb);
177	if (unlikely(ip_len < sizeof(struct iphdr) ||
178		     skb->len < nh_ofs + ip_len))
179		return -EINVAL;
180
181	skb_set_transport_header(skb, nh_ofs + ip_len);
182	return 0;
183}
184
185static bool tcphdr_ok(struct sk_buff *skb)
186{
187	int th_ofs = skb_transport_offset(skb);
188	int tcp_len;
189
190	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
191		return false;
192
193	tcp_len = tcp_hdrlen(skb);
194	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
195		     skb->len < th_ofs + tcp_len))
196		return false;
197
198	return true;
199}
200
201static bool udphdr_ok(struct sk_buff *skb)
202{
203	return pskb_may_pull(skb, skb_transport_offset(skb) +
204				  sizeof(struct udphdr));
205}
206
207static bool sctphdr_ok(struct sk_buff *skb)
208{
209	return pskb_may_pull(skb, skb_transport_offset(skb) +
210				  sizeof(struct sctphdr));
211}
212
213static bool icmphdr_ok(struct sk_buff *skb)
214{
215	return pskb_may_pull(skb, skb_transport_offset(skb) +
216				  sizeof(struct icmphdr));
217}
218
219static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
220{
221	unsigned int nh_ofs = skb_network_offset(skb);
222	unsigned int nh_len;
223	int payload_ofs;
224	struct ipv6hdr *nh;
225	uint8_t nexthdr;
226	__be16 frag_off;
227	int err;
228
229	err = check_header(skb, nh_ofs + sizeof(*nh));
230	if (unlikely(err))
231		return err;
232
233	nh = ipv6_hdr(skb);
234	nexthdr = nh->nexthdr;
235	payload_ofs = (u8 *)(nh + 1) - skb->data;
236
237	key->ip.proto = NEXTHDR_NONE;
238	key->ip.tos = ipv6_get_dsfield(nh);
239	key->ip.ttl = nh->hop_limit;
240	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
241	key->ipv6.addr.src = nh->saddr;
242	key->ipv6.addr.dst = nh->daddr;
243
244	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
245	if (unlikely(payload_ofs < 0))
246		return -EINVAL;
247
248	if (frag_off) {
249		if (frag_off & htons(~0x7))
250			key->ip.frag = OVS_FRAG_TYPE_LATER;
251		else
252			key->ip.frag = OVS_FRAG_TYPE_FIRST;
 
 
253	}
254
 
 
 
 
 
 
 
255	nh_len = payload_ofs - nh_ofs;
256	skb_set_transport_header(skb, nh_ofs + nh_len);
257	key->ip.proto = nexthdr;
258	return nh_len;
259}
260
261static bool icmp6hdr_ok(struct sk_buff *skb)
262{
263	return pskb_may_pull(skb, skb_transport_offset(skb) +
264				  sizeof(struct icmp6hdr));
265}
266
267static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 
 
 
 
 
 
 
268{
269	struct qtag_prefix {
270		__be16 eth_type; /* ETH_P_8021Q */
271		__be16 tci;
272	};
273	struct qtag_prefix *qp;
274
275	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
276		return 0;
277
278	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
279					 sizeof(__be16))))
280		return -ENOMEM;
281
282	qp = (struct qtag_prefix *) skb->data;
283	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
284	__skb_pull(skb, sizeof(struct qtag_prefix));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
285
286	return 0;
287}
288
289static __be16 parse_ethertype(struct sk_buff *skb)
290{
291	struct llc_snap_hdr {
292		u8  dsap;  /* Always 0xAA */
293		u8  ssap;  /* Always 0xAA */
294		u8  ctrl;
295		u8  oui[3];
296		__be16 ethertype;
297	};
298	struct llc_snap_hdr *llc;
299	__be16 proto;
300
301	proto = *(__be16 *) skb->data;
302	__skb_pull(skb, sizeof(__be16));
303
304	if (ntohs(proto) >= ETH_P_802_3_MIN)
305		return proto;
306
307	if (skb->len < sizeof(struct llc_snap_hdr))
308		return htons(ETH_P_802_2);
309
310	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
311		return htons(0);
312
313	llc = (struct llc_snap_hdr *) skb->data;
314	if (llc->dsap != LLC_SAP_SNAP ||
315	    llc->ssap != LLC_SAP_SNAP ||
316	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
317		return htons(ETH_P_802_2);
318
319	__skb_pull(skb, sizeof(struct llc_snap_hdr));
320
321	if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
322		return llc->ethertype;
323
324	return htons(ETH_P_802_2);
325}
326
327static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
328			int nh_len)
329{
330	struct icmp6hdr *icmp = icmp6_hdr(skb);
331
332	/* The ICMPv6 type and code fields use the 16-bit transport port
333	 * fields, so we need to store them in 16-bit network byte order.
334	 */
335	key->ipv6.tp.src = htons(icmp->icmp6_type);
336	key->ipv6.tp.dst = htons(icmp->icmp6_code);
 
337
338	if (icmp->icmp6_code == 0 &&
339	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
340	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
341		int icmp_len = skb->len - skb_transport_offset(skb);
342		struct nd_msg *nd;
343		int offset;
344
345		/* In order to process neighbor discovery options, we need the
346		 * entire packet.
347		 */
348		if (unlikely(icmp_len < sizeof(*nd)))
349			return 0;
350
351		if (unlikely(skb_linearize(skb)))
352			return -ENOMEM;
353
354		nd = (struct nd_msg *)skb_transport_header(skb);
355		key->ipv6.nd.target = nd->target;
356
357		icmp_len -= sizeof(*nd);
358		offset = 0;
359		while (icmp_len >= 8) {
360			struct nd_opt_hdr *nd_opt =
361				 (struct nd_opt_hdr *)(nd->opt + offset);
362			int opt_len = nd_opt->nd_opt_len * 8;
363
364			if (unlikely(!opt_len || opt_len > icmp_len))
365				return 0;
366
367			/* Store the link layer address if the appropriate
368			 * option is provided.  It is considered an error if
369			 * the same link layer option is specified twice.
370			 */
371			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
372			    && opt_len == 8) {
373				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
374					goto invalid;
375				memcpy(key->ipv6.nd.sll,
376				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
377			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
378				   && opt_len == 8) {
379				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
380					goto invalid;
381				memcpy(key->ipv6.nd.tll,
382				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
383			}
384
385			icmp_len -= opt_len;
386			offset += opt_len;
387		}
388	}
389
390	return 0;
391
392invalid:
393	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
394	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
395	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
396
397	return 0;
398}
399
400/**
401 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
402 * @skb: sk_buff that contains the frame, with skb->data pointing to the
403 * Ethernet header
404 * @in_port: port number on which @skb was received.
405 * @key: output flow key
406 *
407 * The caller must ensure that skb->len >= ETH_HLEN.
408 *
409 * Returns 0 if successful, otherwise a negative errno value.
410 *
411 * Initializes @skb header pointers as follows:
412 *
413 *    - skb->mac_header: the Ethernet header.
414 *
415 *    - skb->network_header: just past the Ethernet header, or just past the
416 *      VLAN header, to the first byte of the Ethernet payload.
417 *
418 *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
419 *      on output, then just past the IP header, if one is present and
420 *      of a correct length, otherwise the same as skb->network_header.
421 *      For other key->eth.type values it is left untouched.
 
 
 
422 */
423int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
424{
425	int error;
426	struct ethhdr *eth;
427
428	memset(key, 0, sizeof(*key));
 
429
430	key->phy.priority = skb->priority;
431	if (OVS_CB(skb)->tun_key)
432		memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
433	key->phy.in_port = in_port;
434	key->phy.skb_mark = skb->mark;
435
436	skb_reset_mac_header(skb);
 
 
 
 
437
438	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
439	 * header in the linear data area.
440	 */
441	eth = eth_hdr(skb);
442	memcpy(key->eth.src, eth->h_source, ETH_ALEN);
443	memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
444
445	__skb_pull(skb, 2 * ETH_ALEN);
446	/* We are going to push all headers that we pull, so no need to
447	 * update skb->csum here.
448	 */
449
450	if (vlan_tx_tag_present(skb))
451		key->eth.tci = htons(skb->vlan_tci);
452	else if (eth->h_proto == htons(ETH_P_8021Q))
453		if (unlikely(parse_vlan(skb, key)))
454			return -ENOMEM;
455
456	key->eth.type = parse_ethertype(skb);
457	if (unlikely(key->eth.type == htons(0)))
458		return -ENOMEM;
459
460	skb_reset_network_header(skb);
461	__skb_push(skb, skb->data - skb_mac_header(skb));
 
 
 
462
463	/* Network layer. */
464	if (key->eth.type == htons(ETH_P_IP)) {
465		struct iphdr *nh;
466		__be16 offset;
467
468		error = check_iphdr(skb);
469		if (unlikely(error)) {
 
 
470			if (error == -EINVAL) {
471				skb->transport_header = skb->network_header;
472				error = 0;
473			}
474			return error;
475		}
476
477		nh = ip_hdr(skb);
478		key->ipv4.addr.src = nh->saddr;
479		key->ipv4.addr.dst = nh->daddr;
480
481		key->ip.proto = nh->protocol;
482		key->ip.tos = nh->tos;
483		key->ip.ttl = nh->ttl;
484
485		offset = nh->frag_off & htons(IP_OFFSET);
486		if (offset) {
487			key->ip.frag = OVS_FRAG_TYPE_LATER;
488			return 0;
489		}
490		if (nh->frag_off & htons(IP_MF) ||
491			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
492			key->ip.frag = OVS_FRAG_TYPE_FIRST;
 
 
493
494		/* Transport layer. */
495		if (key->ip.proto == IPPROTO_TCP) {
496			if (tcphdr_ok(skb)) {
497				struct tcphdr *tcp = tcp_hdr(skb);
498				key->ipv4.tp.src = tcp->source;
499				key->ipv4.tp.dst = tcp->dest;
500				key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
 
 
501			}
 
502		} else if (key->ip.proto == IPPROTO_UDP) {
503			if (udphdr_ok(skb)) {
504				struct udphdr *udp = udp_hdr(skb);
505				key->ipv4.tp.src = udp->source;
506				key->ipv4.tp.dst = udp->dest;
 
 
507			}
508		} else if (key->ip.proto == IPPROTO_SCTP) {
509			if (sctphdr_ok(skb)) {
510				struct sctphdr *sctp = sctp_hdr(skb);
511				key->ipv4.tp.src = sctp->source;
512				key->ipv4.tp.dst = sctp->dest;
 
 
513			}
514		} else if (key->ip.proto == IPPROTO_ICMP) {
515			if (icmphdr_ok(skb)) {
516				struct icmphdr *icmp = icmp_hdr(skb);
517				/* The ICMP type and code fields use the 16-bit
518				 * transport port fields, so we need to store
519				 * them in 16-bit network byte order. */
520				key->ipv4.tp.src = htons(icmp->type);
521				key->ipv4.tp.dst = htons(icmp->code);
 
 
522			}
523		}
524
525	} else if ((key->eth.type == htons(ETH_P_ARP) ||
526		   key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
527		struct arp_eth_header *arp;
 
528
529		arp = (struct arp_eth_header *)skb_network_header(skb);
530
531		if (arp->ar_hrd == htons(ARPHRD_ETHER)
532				&& arp->ar_pro == htons(ETH_P_IP)
533				&& arp->ar_hln == ETH_ALEN
534				&& arp->ar_pln == 4) {
 
535
536			/* We only match on the lower 8 bits of the opcode. */
537			if (ntohs(arp->ar_op) <= 0xff)
538				key->ip.proto = ntohs(arp->ar_op);
 
 
 
539			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
540			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
541			memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
542			memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
543		}
544	} else if (key->eth.type == htons(ETH_P_IPV6)) {
545		int nh_len;             /* IPv6 Header + Extensions */
546
547		nh_len = parse_ipv6hdr(skb, key);
548		if (unlikely(nh_len < 0)) {
549			if (nh_len == -EINVAL) {
 
 
 
 
 
550				skb->transport_header = skb->network_header;
551				error = 0;
552			} else {
 
553				error = nh_len;
554			}
555			return error;
556		}
557
558		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
559			return 0;
560		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
561			key->ip.frag = OVS_FRAG_TYPE_FIRST;
562
563		/* Transport layer. */
564		if (key->ip.proto == NEXTHDR_TCP) {
565			if (tcphdr_ok(skb)) {
566				struct tcphdr *tcp = tcp_hdr(skb);
567				key->ipv6.tp.src = tcp->source;
568				key->ipv6.tp.dst = tcp->dest;
569				key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
 
 
570			}
571		} else if (key->ip.proto == NEXTHDR_UDP) {
572			if (udphdr_ok(skb)) {
573				struct udphdr *udp = udp_hdr(skb);
574				key->ipv6.tp.src = udp->source;
575				key->ipv6.tp.dst = udp->dest;
 
 
576			}
577		} else if (key->ip.proto == NEXTHDR_SCTP) {
578			if (sctphdr_ok(skb)) {
579				struct sctphdr *sctp = sctp_hdr(skb);
580				key->ipv6.tp.src = sctp->source;
581				key->ipv6.tp.dst = sctp->dest;
 
 
582			}
583		} else if (key->ip.proto == NEXTHDR_ICMP) {
584			if (icmp6hdr_ok(skb)) {
585				error = parse_icmpv6(skb, key, nh_len);
586				if (error)
587					return error;
 
 
588			}
589		}
590	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
591
592	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
593}