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}
v3.5.6
   1/*
   2 * Copyright (c) 2007-2011 Nicira Networks.
   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/tcp.h>
  38#include <linux/udp.h>
  39#include <linux/icmp.h>
  40#include <linux/icmpv6.h>
  41#include <linux/rculist.h>
  42#include <net/ip.h>
 
  43#include <net/ipv6.h>
 
  44#include <net/ndisc.h>
  45
  46static struct kmem_cache *flow_cache;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  47
  48static int check_header(struct sk_buff *skb, int len)
  49{
  50	if (unlikely(skb->len < len))
  51		return -EINVAL;
  52	if (unlikely(!pskb_may_pull(skb, len)))
  53		return -ENOMEM;
  54	return 0;
  55}
  56
  57static bool arphdr_ok(struct sk_buff *skb)
  58{
  59	return pskb_may_pull(skb, skb_network_offset(skb) +
  60				  sizeof(struct arp_eth_header));
  61}
  62
  63static int check_iphdr(struct sk_buff *skb)
  64{
  65	unsigned int nh_ofs = skb_network_offset(skb);
  66	unsigned int ip_len;
  67	int err;
  68
  69	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  70	if (unlikely(err))
  71		return err;
  72
  73	ip_len = ip_hdrlen(skb);
  74	if (unlikely(ip_len < sizeof(struct iphdr) ||
  75		     skb->len < nh_ofs + ip_len))
  76		return -EINVAL;
  77
  78	skb_set_transport_header(skb, nh_ofs + ip_len);
  79	return 0;
  80}
  81
  82static bool tcphdr_ok(struct sk_buff *skb)
  83{
  84	int th_ofs = skb_transport_offset(skb);
  85	int tcp_len;
  86
  87	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  88		return false;
  89
  90	tcp_len = tcp_hdrlen(skb);
  91	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  92		     skb->len < th_ofs + tcp_len))
  93		return false;
  94
  95	return true;
  96}
  97
  98static bool udphdr_ok(struct sk_buff *skb)
  99{
 100	return pskb_may_pull(skb, skb_transport_offset(skb) +
 101				  sizeof(struct udphdr));
 102}
 103
 104static bool icmphdr_ok(struct sk_buff *skb)
 105{
 106	return pskb_may_pull(skb, skb_transport_offset(skb) +
 107				  sizeof(struct icmphdr));
 108}
 109
 110u64 ovs_flow_used_time(unsigned long flow_jiffies)
 111{
 112	struct timespec cur_ts;
 113	u64 cur_ms, idle_ms;
 114
 115	ktime_get_ts(&cur_ts);
 116	idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 117	cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 118		 cur_ts.tv_nsec / NSEC_PER_MSEC;
 119
 120	return cur_ms - idle_ms;
 121}
 122
 123#define SW_FLOW_KEY_OFFSET(field)		\
 124	(offsetof(struct sw_flow_key, field) +	\
 125	 FIELD_SIZEOF(struct sw_flow_key, field))
 126
 127static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 128			 int *key_lenp)
 129{
 130	unsigned int nh_ofs = skb_network_offset(skb);
 131	unsigned int nh_len;
 132	int payload_ofs;
 133	struct ipv6hdr *nh;
 134	uint8_t nexthdr;
 135	__be16 frag_off;
 136	int err;
 137
 138	*key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 139
 140	err = check_header(skb, nh_ofs + sizeof(*nh));
 141	if (unlikely(err))
 142		return err;
 143
 144	nh = ipv6_hdr(skb);
 145	nexthdr = nh->nexthdr;
 146	payload_ofs = (u8 *)(nh + 1) - skb->data;
 147
 148	key->ip.proto = NEXTHDR_NONE;
 149	key->ip.tos = ipv6_get_dsfield(nh);
 150	key->ip.ttl = nh->hop_limit;
 151	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 152	key->ipv6.addr.src = nh->saddr;
 153	key->ipv6.addr.dst = nh->daddr;
 154
 155	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 156	if (unlikely(payload_ofs < 0))
 157		return -EINVAL;
 158
 159	if (frag_off) {
 160		if (frag_off & htons(~0x7))
 161			key->ip.frag = OVS_FRAG_TYPE_LATER;
 162		else
 163			key->ip.frag = OVS_FRAG_TYPE_FIRST;
 
 
 164	}
 165
 
 
 
 
 
 
 
 166	nh_len = payload_ofs - nh_ofs;
 167	skb_set_transport_header(skb, nh_ofs + nh_len);
 168	key->ip.proto = nexthdr;
 169	return nh_len;
 170}
 171
 172static bool icmp6hdr_ok(struct sk_buff *skb)
 173{
 174	return pskb_may_pull(skb, skb_transport_offset(skb) +
 175				  sizeof(struct icmp6hdr));
 176}
 177
 178#define TCP_FLAGS_OFFSET 13
 179#define TCP_FLAG_MASK 0x3f
 180
 181void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 
 
 
 
 182{
 183	u8 tcp_flags = 0;
 184
 185	if (flow->key.eth.type == htons(ETH_P_IP) &&
 186	    flow->key.ip.proto == IPPROTO_TCP &&
 187	    likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 188		u8 *tcp = (u8 *)tcp_hdr(skb);
 189		tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 190	}
 191
 192	spin_lock(&flow->lock);
 193	flow->used = jiffies;
 194	flow->packet_count++;
 195	flow->byte_count += skb->len;
 196	flow->tcp_flags |= tcp_flags;
 197	spin_unlock(&flow->lock);
 198}
 199
 200struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
 201{
 202	int actions_len = nla_len(actions);
 203	struct sw_flow_actions *sfa;
 204
 205	/* At least DP_MAX_PORTS actions are required to be able to flood a
 206	 * packet to every port.  Factor of 2 allows for setting VLAN tags,
 207	 * etc. */
 208	if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 209		return ERR_PTR(-EINVAL);
 210
 211	sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 212	if (!sfa)
 213		return ERR_PTR(-ENOMEM);
 214
 215	sfa->actions_len = actions_len;
 216	memcpy(sfa->actions, nla_data(actions), actions_len);
 217	return sfa;
 218}
 219
 220struct sw_flow *ovs_flow_alloc(void)
 221{
 222	struct sw_flow *flow;
 223
 224	flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 225	if (!flow)
 226		return ERR_PTR(-ENOMEM);
 227
 228	spin_lock_init(&flow->lock);
 229	flow->sf_acts = NULL;
 230
 231	return flow;
 232}
 233
 234static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 235{
 236	hash = jhash_1word(hash, table->hash_seed);
 237	return flex_array_get(table->buckets,
 238				(hash & (table->n_buckets - 1)));
 239}
 240
 241static struct flex_array *alloc_buckets(unsigned int n_buckets)
 242{
 243	struct flex_array *buckets;
 244	int i, err;
 245
 246	buckets = flex_array_alloc(sizeof(struct hlist_head *),
 247				   n_buckets, GFP_KERNEL);
 248	if (!buckets)
 249		return NULL;
 250
 251	err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 252	if (err) {
 253		flex_array_free(buckets);
 254		return NULL;
 255	}
 256
 257	for (i = 0; i < n_buckets; i++)
 258		INIT_HLIST_HEAD((struct hlist_head *)
 259					flex_array_get(buckets, i));
 260
 261	return buckets;
 262}
 263
 264static void free_buckets(struct flex_array *buckets)
 265{
 266	flex_array_free(buckets);
 
 
 
 267}
 268
 269struct flow_table *ovs_flow_tbl_alloc(int new_size)
 270{
 271	struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 272
 273	if (!table)
 274		return NULL;
 275
 276	table->buckets = alloc_buckets(new_size);
 277
 278	if (!table->buckets) {
 279		kfree(table);
 280		return NULL;
 281	}
 282	table->n_buckets = new_size;
 283	table->count = 0;
 284	table->node_ver = 0;
 285	table->keep_flows = false;
 286	get_random_bytes(&table->hash_seed, sizeof(u32));
 287
 288	return table;
 289}
 290
 291void ovs_flow_tbl_destroy(struct flow_table *table)
 292{
 293	int i;
 294
 295	if (!table)
 296		return;
 297
 298	if (table->keep_flows)
 299		goto skip_flows;
 300
 301	for (i = 0; i < table->n_buckets; i++) {
 302		struct sw_flow *flow;
 303		struct hlist_head *head = flex_array_get(table->buckets, i);
 304		struct hlist_node *node, *n;
 305		int ver = table->node_ver;
 306
 307		hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 308			hlist_del_rcu(&flow->hash_node[ver]);
 309			ovs_flow_free(flow);
 310		}
 311	}
 312
 313skip_flows:
 314	free_buckets(table->buckets);
 315	kfree(table);
 316}
 317
 318static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 319{
 320	struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 321
 322	ovs_flow_tbl_destroy(table);
 323}
 324
 325void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 326{
 327	if (!table)
 328		return;
 329
 330	call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 331}
 332
 333struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 334{
 335	struct sw_flow *flow;
 336	struct hlist_head *head;
 337	struct hlist_node *n;
 338	int ver;
 339	int i;
 340
 341	ver = table->node_ver;
 342	while (*bucket < table->n_buckets) {
 343		i = 0;
 344		head = flex_array_get(table->buckets, *bucket);
 345		hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 346			if (i < *last) {
 347				i++;
 348				continue;
 349			}
 350			*last = i + 1;
 351			return flow;
 352		}
 353		(*bucket)++;
 354		*last = 0;
 355	}
 356
 357	return NULL;
 358}
 359
 360static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 361{
 362	int old_ver;
 363	int i;
 364
 365	old_ver = old->node_ver;
 366	new->node_ver = !old_ver;
 367
 368	/* Insert in new table. */
 369	for (i = 0; i < old->n_buckets; i++) {
 370		struct sw_flow *flow;
 371		struct hlist_head *head;
 372		struct hlist_node *n;
 373
 374		head = flex_array_get(old->buckets, i);
 375
 376		hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 377			ovs_flow_tbl_insert(new, flow);
 378	}
 379	old->keep_flows = true;
 380}
 381
 382static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 383{
 384	struct flow_table *new_table;
 385
 386	new_table = ovs_flow_tbl_alloc(n_buckets);
 387	if (!new_table)
 388		return ERR_PTR(-ENOMEM);
 389
 390	flow_table_copy_flows(table, new_table);
 391
 392	return new_table;
 393}
 394
 395struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 396{
 397	return __flow_tbl_rehash(table, table->n_buckets);
 398}
 399
 400struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 401{
 402	return __flow_tbl_rehash(table, table->n_buckets * 2);
 403}
 404
 405void ovs_flow_free(struct sw_flow *flow)
 406{
 407	if (unlikely(!flow))
 408		return;
 409
 410	kfree((struct sf_flow_acts __force *)flow->sf_acts);
 411	kmem_cache_free(flow_cache, flow);
 412}
 413
 414/* RCU callback used by ovs_flow_deferred_free. */
 415static void rcu_free_flow_callback(struct rcu_head *rcu)
 416{
 417	struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 418
 419	ovs_flow_free(flow);
 420}
 421
 422/* Schedules 'flow' to be freed after the next RCU grace period.
 423 * The caller must hold rcu_read_lock for this to be sensible. */
 424void ovs_flow_deferred_free(struct sw_flow *flow)
 425{
 426	call_rcu(&flow->rcu, rcu_free_flow_callback);
 427}
 428
 429/* RCU callback used by ovs_flow_deferred_free_acts. */
 430static void rcu_free_acts_callback(struct rcu_head *rcu)
 431{
 432	struct sw_flow_actions *sf_acts = container_of(rcu,
 433			struct sw_flow_actions, rcu);
 434	kfree(sf_acts);
 435}
 436
 437/* Schedules 'sf_acts' to be freed after the next RCU grace period.
 438 * The caller must hold rcu_read_lock for this to be sensible. */
 439void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 440{
 441	call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 442}
 443
 444static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 445{
 446	struct qtag_prefix {
 447		__be16 eth_type; /* ETH_P_8021Q */
 448		__be16 tci;
 449	};
 450	struct qtag_prefix *qp;
 451
 452	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 453		return 0;
 454
 455	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 456					 sizeof(__be16))))
 457		return -ENOMEM;
 458
 459	qp = (struct qtag_prefix *) skb->data;
 460	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 461	__skb_pull(skb, sizeof(struct qtag_prefix));
 462
 463	return 0;
 464}
 465
 466static __be16 parse_ethertype(struct sk_buff *skb)
 467{
 468	struct llc_snap_hdr {
 469		u8  dsap;  /* Always 0xAA */
 470		u8  ssap;  /* Always 0xAA */
 471		u8  ctrl;
 472		u8  oui[3];
 473		__be16 ethertype;
 474	};
 475	struct llc_snap_hdr *llc;
 476	__be16 proto;
 477
 478	proto = *(__be16 *) skb->data;
 479	__skb_pull(skb, sizeof(__be16));
 480
 481	if (ntohs(proto) >= 1536)
 482		return proto;
 483
 484	if (skb->len < sizeof(struct llc_snap_hdr))
 485		return htons(ETH_P_802_2);
 486
 487	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 488		return htons(0);
 489
 490	llc = (struct llc_snap_hdr *) skb->data;
 491	if (llc->dsap != LLC_SAP_SNAP ||
 492	    llc->ssap != LLC_SAP_SNAP ||
 493	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 494		return htons(ETH_P_802_2);
 495
 496	__skb_pull(skb, sizeof(struct llc_snap_hdr));
 497	return llc->ethertype;
 
 
 
 
 498}
 499
 500static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 501			int *key_lenp, int nh_len)
 502{
 503	struct icmp6hdr *icmp = icmp6_hdr(skb);
 504	int error = 0;
 505	int key_len;
 506
 507	/* The ICMPv6 type and code fields use the 16-bit transport port
 508	 * fields, so we need to store them in 16-bit network byte order.
 509	 */
 510	key->ipv6.tp.src = htons(icmp->icmp6_type);
 511	key->ipv6.tp.dst = htons(icmp->icmp6_code);
 512	key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 513
 514	if (icmp->icmp6_code == 0 &&
 515	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 516	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 517		int icmp_len = skb->len - skb_transport_offset(skb);
 518		struct nd_msg *nd;
 519		int offset;
 520
 521		key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 522
 523		/* In order to process neighbor discovery options, we need the
 524		 * entire packet.
 525		 */
 526		if (unlikely(icmp_len < sizeof(*nd)))
 527			goto out;
 528		if (unlikely(skb_linearize(skb))) {
 529			error = -ENOMEM;
 530			goto out;
 531		}
 532
 533		nd = (struct nd_msg *)skb_transport_header(skb);
 534		key->ipv6.nd.target = nd->target;
 535		key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 536
 537		icmp_len -= sizeof(*nd);
 538		offset = 0;
 539		while (icmp_len >= 8) {
 540			struct nd_opt_hdr *nd_opt =
 541				 (struct nd_opt_hdr *)(nd->opt + offset);
 542			int opt_len = nd_opt->nd_opt_len * 8;
 543
 544			if (unlikely(!opt_len || opt_len > icmp_len))
 545				goto invalid;
 546
 547			/* Store the link layer address if the appropriate
 548			 * option is provided.  It is considered an error if
 549			 * the same link layer option is specified twice.
 550			 */
 551			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 552			    && opt_len == 8) {
 553				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 554					goto invalid;
 555				memcpy(key->ipv6.nd.sll,
 556				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 557			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 558				   && opt_len == 8) {
 559				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 560					goto invalid;
 561				memcpy(key->ipv6.nd.tll,
 562				    &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 563			}
 564
 565			icmp_len -= opt_len;
 566			offset += opt_len;
 567		}
 568	}
 569
 570	goto out;
 571
 572invalid:
 573	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 574	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 575	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 576
 577out:
 578	*key_lenp = key_len;
 579	return error;
 580}
 581
 582/**
 583 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 584 * @skb: sk_buff that contains the frame, with skb->data pointing to the
 585 * Ethernet header
 586 * @in_port: port number on which @skb was received.
 587 * @key: output flow key
 588 * @key_lenp: length of output flow key
 589 *
 590 * The caller must ensure that skb->len >= ETH_HLEN.
 591 *
 592 * Returns 0 if successful, otherwise a negative errno value.
 593 *
 594 * Initializes @skb header pointers as follows:
 595 *
 596 *    - skb->mac_header: the Ethernet header.
 597 *
 598 *    - skb->network_header: just past the Ethernet header, or just past the
 599 *      VLAN header, to the first byte of the Ethernet payload.
 600 *
 601 *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 602 *      on output, then just past the IP header, if one is present and
 603 *      of a correct length, otherwise the same as skb->network_header.
 604 *      For other key->dl_type values it is left untouched.
 
 
 
 605 */
 606int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 607		 int *key_lenp)
 608{
 609	int error = 0;
 610	int key_len = SW_FLOW_KEY_OFFSET(eth);
 611	struct ethhdr *eth;
 612
 613	memset(key, 0, sizeof(*key));
 
 
 
 614
 615	key->phy.priority = skb->priority;
 616	key->phy.in_port = in_port;
 
 
 
 617
 618	skb_reset_mac_header(skb);
 
 
 
 
 
 
 
 
 
 619
 620	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 621	 * header in the linear data area.
 622	 */
 623	eth = eth_hdr(skb);
 624	memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 625	memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 626
 627	__skb_pull(skb, 2 * ETH_ALEN);
 628
 629	if (vlan_tx_tag_present(skb))
 630		key->eth.tci = htons(skb->vlan_tci);
 631	else if (eth->h_proto == htons(ETH_P_8021Q))
 632		if (unlikely(parse_vlan(skb, key)))
 633			return -ENOMEM;
 634
 635	key->eth.type = parse_ethertype(skb);
 636	if (unlikely(key->eth.type == htons(0)))
 637		return -ENOMEM;
 638
 639	skb_reset_network_header(skb);
 640	__skb_push(skb, skb->data - skb_mac_header(skb));
 
 
 
 641
 642	/* Network layer. */
 643	if (key->eth.type == htons(ETH_P_IP)) {
 644		struct iphdr *nh;
 645		__be16 offset;
 646
 647		key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 648
 649		error = check_iphdr(skb);
 650		if (unlikely(error)) {
 
 
 651			if (error == -EINVAL) {
 652				skb->transport_header = skb->network_header;
 653				error = 0;
 654			}
 655			goto out;
 656		}
 657
 658		nh = ip_hdr(skb);
 659		key->ipv4.addr.src = nh->saddr;
 660		key->ipv4.addr.dst = nh->daddr;
 661
 662		key->ip.proto = nh->protocol;
 663		key->ip.tos = nh->tos;
 664		key->ip.ttl = nh->ttl;
 665
 666		offset = nh->frag_off & htons(IP_OFFSET);
 667		if (offset) {
 668			key->ip.frag = OVS_FRAG_TYPE_LATER;
 669			goto out;
 670		}
 671		if (nh->frag_off & htons(IP_MF) ||
 672			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 673			key->ip.frag = OVS_FRAG_TYPE_FIRST;
 
 
 674
 675		/* Transport layer. */
 676		if (key->ip.proto == IPPROTO_TCP) {
 677			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 678			if (tcphdr_ok(skb)) {
 679				struct tcphdr *tcp = tcp_hdr(skb);
 680				key->ipv4.tp.src = tcp->source;
 681				key->ipv4.tp.dst = tcp->dest;
 
 
 
 682			}
 
 683		} else if (key->ip.proto == IPPROTO_UDP) {
 684			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 685			if (udphdr_ok(skb)) {
 686				struct udphdr *udp = udp_hdr(skb);
 687				key->ipv4.tp.src = udp->source;
 688				key->ipv4.tp.dst = udp->dest;
 
 
 
 
 
 
 
 
 
 
 689			}
 690		} else if (key->ip.proto == IPPROTO_ICMP) {
 691			key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 692			if (icmphdr_ok(skb)) {
 693				struct icmphdr *icmp = icmp_hdr(skb);
 694				/* The ICMP type and code fields use the 16-bit
 695				 * transport port fields, so we need to store
 696				 * them in 16-bit network byte order. */
 697				key->ipv4.tp.src = htons(icmp->type);
 698				key->ipv4.tp.dst = htons(icmp->code);
 
 
 699			}
 700		}
 701
 702	} else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 
 703		struct arp_eth_header *arp;
 
 704
 705		arp = (struct arp_eth_header *)skb_network_header(skb);
 706
 707		if (arp->ar_hrd == htons(ARPHRD_ETHER)
 708				&& arp->ar_pro == htons(ETH_P_IP)
 709				&& arp->ar_hln == ETH_ALEN
 710				&& arp->ar_pln == 4) {
 
 711
 712			/* We only match on the lower 8 bits of the opcode. */
 713			if (ntohs(arp->ar_op) <= 0xff)
 714				key->ip.proto = ntohs(arp->ar_op);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 715
 716			if (key->ip.proto == ARPOP_REQUEST
 717					|| key->ip.proto == ARPOP_REPLY) {
 718				memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 719				memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 720				memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 721				memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 722				key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 723			}
 
 
 724		}
 725	} else if (key->eth.type == htons(ETH_P_IPV6)) {
 726		int nh_len;             /* IPv6 Header + Extensions */
 727
 728		nh_len = parse_ipv6hdr(skb, key, &key_len);
 729		if (unlikely(nh_len < 0)) {
 730			if (nh_len == -EINVAL)
 
 
 
 
 
 731				skb->transport_header = skb->network_header;
 732			else
 
 
 733				error = nh_len;
 734			goto out;
 
 735		}
 736
 737		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 738			goto out;
 739		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 740			key->ip.frag = OVS_FRAG_TYPE_FIRST;
 741
 742		/* Transport layer. */
 743		if (key->ip.proto == NEXTHDR_TCP) {
 744			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 745			if (tcphdr_ok(skb)) {
 746				struct tcphdr *tcp = tcp_hdr(skb);
 747				key->ipv6.tp.src = tcp->source;
 748				key->ipv6.tp.dst = tcp->dest;
 
 
 
 749			}
 750		} else if (key->ip.proto == NEXTHDR_UDP) {
 751			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 752			if (udphdr_ok(skb)) {
 753				struct udphdr *udp = udp_hdr(skb);
 754				key->ipv6.tp.src = udp->source;
 755				key->ipv6.tp.dst = udp->dest;
 
 
 
 
 
 
 
 
 
 
 756			}
 757		} else if (key->ip.proto == NEXTHDR_ICMP) {
 758			key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 759			if (icmp6hdr_ok(skb)) {
 760				error = parse_icmpv6(skb, key, &key_len, nh_len);
 761				if (error < 0)
 762					goto out;
 
 
 763			}
 764		}
 765	}
 766
 767out:
 768	*key_lenp = key_len;
 769	return error;
 770}
 771
 772u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 773{
 774	return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 775}
 776
 777struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 778				struct sw_flow_key *key, int key_len)
 779{
 780	struct sw_flow *flow;
 781	struct hlist_node *n;
 782	struct hlist_head *head;
 783	u32 hash;
 784
 785	hash = ovs_flow_hash(key, key_len);
 786
 787	head = find_bucket(table, hash);
 788	hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 789
 790		if (flow->hash == hash &&
 791		    !memcmp(&flow->key, key, key_len)) {
 792			return flow;
 793		}
 794	}
 795	return NULL;
 
 796}
 797
 798void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 
 799{
 800	struct hlist_head *head;
 801
 802	head = find_bucket(table, flow->hash);
 803	hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 804	table->count++;
 805}
 806
 807void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 808{
 809	hlist_del_rcu(&flow->hash_node[table->node_ver]);
 810	table->count--;
 811	BUG_ON(table->count < 0);
 812}
 813
 814/* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 815const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 816	[OVS_KEY_ATTR_ENCAP] = -1,
 817	[OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 818	[OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 819	[OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 820	[OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 821	[OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 822	[OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 823	[OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 824	[OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 825	[OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 826	[OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 827	[OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 828	[OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 829	[OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 830};
 831
 832static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 833				  const struct nlattr *a[], u32 *attrs)
 834{
 835	const struct ovs_key_icmp *icmp_key;
 836	const struct ovs_key_tcp *tcp_key;
 837	const struct ovs_key_udp *udp_key;
 838
 839	switch (swkey->ip.proto) {
 840	case IPPROTO_TCP:
 841		if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 842			return -EINVAL;
 843		*attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 844
 845		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 846		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 847		swkey->ipv4.tp.src = tcp_key->tcp_src;
 848		swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 849		break;
 850
 851	case IPPROTO_UDP:
 852		if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 853			return -EINVAL;
 854		*attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 855
 856		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 857		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 858		swkey->ipv4.tp.src = udp_key->udp_src;
 859		swkey->ipv4.tp.dst = udp_key->udp_dst;
 860		break;
 861
 862	case IPPROTO_ICMP:
 863		if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 864			return -EINVAL;
 865		*attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 866
 867		*key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 868		icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 869		swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 870		swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 871		break;
 872	}
 873
 874	return 0;
 875}
 876
 877static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 878				  const struct nlattr *a[], u32 *attrs)
 879{
 880	const struct ovs_key_icmpv6 *icmpv6_key;
 881	const struct ovs_key_tcp *tcp_key;
 882	const struct ovs_key_udp *udp_key;
 883
 884	switch (swkey->ip.proto) {
 885	case IPPROTO_TCP:
 886		if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 887			return -EINVAL;
 888		*attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 889
 890		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 891		tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 892		swkey->ipv6.tp.src = tcp_key->tcp_src;
 893		swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 894		break;
 895
 896	case IPPROTO_UDP:
 897		if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 898			return -EINVAL;
 899		*attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 900
 901		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 902		udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 903		swkey->ipv6.tp.src = udp_key->udp_src;
 904		swkey->ipv6.tp.dst = udp_key->udp_dst;
 905		break;
 906
 907	case IPPROTO_ICMPV6:
 908		if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 909			return -EINVAL;
 910		*attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 911
 912		*key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 913		icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 914		swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 915		swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 916
 917		if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 918		    swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 919			const struct ovs_key_nd *nd_key;
 920
 921			if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 922				return -EINVAL;
 923			*attrs &= ~(1 << OVS_KEY_ATTR_ND);
 924
 925			*key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 926			nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 927			memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 928			       sizeof(swkey->ipv6.nd.target));
 929			memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 930			memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 931		}
 932		break;
 
 
 
 933	}
 934
 935	return 0;
 936}
 937
 938static int parse_flow_nlattrs(const struct nlattr *attr,
 939			      const struct nlattr *a[], u32 *attrsp)
 
 
 
 
 
 
 
 
 
 
 
 
 940{
 941	const struct nlattr *nla;
 942	u32 attrs;
 943	int rem;
 944
 945	attrs = 0;
 946	nla_for_each_nested(nla, attr, rem) {
 947		u16 type = nla_type(nla);
 948		int expected_len;
 949
 950		if (type > OVS_KEY_ATTR_MAX || attrs & (1 << type))
 951			return -EINVAL;
 952
 953		expected_len = ovs_key_lens[type];
 954		if (nla_len(nla) != expected_len && expected_len != -1)
 955			return -EINVAL;
 956
 957		attrs |= 1 << type;
 958		a[type] = nla;
 959	}
 960	if (rem)
 961		return -EINVAL;
 962
 963	*attrsp = attrs;
 964	return 0;
 965}
 966
 967/**
 968 * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 969 * @swkey: receives the extracted flow key.
 970 * @key_lenp: number of bytes used in @swkey.
 971 * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 972 * sequence.
 973 */
 974int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 975		      const struct nlattr *attr)
 976{
 977	const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 978	const struct ovs_key_ethernet *eth_key;
 979	int key_len;
 980	u32 attrs;
 981	int err;
 982
 983	memset(swkey, 0, sizeof(struct sw_flow_key));
 984	key_len = SW_FLOW_KEY_OFFSET(eth);
 985
 986	err = parse_flow_nlattrs(attr, a, &attrs);
 987	if (err)
 988		return err;
 989
 990	/* Metadata attributes. */
 991	if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 992		swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
 993		attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 994	}
 995	if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 996		u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 997		if (in_port >= DP_MAX_PORTS)
 998			return -EINVAL;
 999		swkey->phy.in_port = in_port;
1000		attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1001	} else {
1002		swkey->phy.in_port = USHRT_MAX;
1003	}
1004
1005	/* Data attributes. */
1006	if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1007		return -EINVAL;
1008	attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1009
1010	eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1011	memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1012	memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1013
1014	if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1015	    nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1016		const struct nlattr *encap;
1017		__be16 tci;
1018
1019		if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1020			      (1 << OVS_KEY_ATTR_ETHERTYPE) |
1021			      (1 << OVS_KEY_ATTR_ENCAP)))
1022			return -EINVAL;
1023
1024		encap = a[OVS_KEY_ATTR_ENCAP];
1025		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1026		if (tci & htons(VLAN_TAG_PRESENT)) {
1027			swkey->eth.tci = tci;
1028
1029			err = parse_flow_nlattrs(encap, a, &attrs);
1030			if (err)
1031				return err;
1032		} else if (!tci) {
1033			/* Corner case for truncated 802.1Q header. */
1034			if (nla_len(encap))
1035				return -EINVAL;
1036
1037			swkey->eth.type = htons(ETH_P_8021Q);
1038			*key_lenp = key_len;
1039			return 0;
1040		} else {
1041			return -EINVAL;
1042		}
1043	}
1044
1045	if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1046		swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1047		if (ntohs(swkey->eth.type) < 1536)
1048			return -EINVAL;
1049		attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1050	} else {
1051		swkey->eth.type = htons(ETH_P_802_2);
1052	}
1053
1054	if (swkey->eth.type == htons(ETH_P_IP)) {
1055		const struct ovs_key_ipv4 *ipv4_key;
1056
1057		if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1058			return -EINVAL;
1059		attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1060
1061		key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1062		ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1063		if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1064			return -EINVAL;
1065		swkey->ip.proto = ipv4_key->ipv4_proto;
1066		swkey->ip.tos = ipv4_key->ipv4_tos;
1067		swkey->ip.ttl = ipv4_key->ipv4_ttl;
1068		swkey->ip.frag = ipv4_key->ipv4_frag;
1069		swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1070		swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1071
1072		if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1073			err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1074			if (err)
1075				return err;
1076		}
1077	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1078		const struct ovs_key_ipv6 *ipv6_key;
1079
1080		if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1081			return -EINVAL;
1082		attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1083
1084		key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1085		ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1086		if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1087			return -EINVAL;
1088		swkey->ipv6.label = ipv6_key->ipv6_label;
1089		swkey->ip.proto = ipv6_key->ipv6_proto;
1090		swkey->ip.tos = ipv6_key->ipv6_tclass;
1091		swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1092		swkey->ip.frag = ipv6_key->ipv6_frag;
1093		memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1094		       sizeof(swkey->ipv6.addr.src));
1095		memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1096		       sizeof(swkey->ipv6.addr.dst));
1097
1098		if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1099			err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1100			if (err)
1101				return err;
1102		}
1103	} else if (swkey->eth.type == htons(ETH_P_ARP)) {
1104		const struct ovs_key_arp *arp_key;
1105
1106		if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1107			return -EINVAL;
1108		attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1109
1110		key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1111		arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1112		swkey->ipv4.addr.src = arp_key->arp_sip;
1113		swkey->ipv4.addr.dst = arp_key->arp_tip;
1114		if (arp_key->arp_op & htons(0xff00))
1115			return -EINVAL;
1116		swkey->ip.proto = ntohs(arp_key->arp_op);
1117		memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1118		memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1119	}
1120
1121	if (attrs)
1122		return -EINVAL;
1123	*key_lenp = key_len;
1124
1125	return 0;
1126}
1127
1128/**
1129 * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1130 * @in_port: receives the extracted input port.
1131 * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1132 * sequence.
1133 *
1134 * This parses a series of Netlink attributes that form a flow key, which must
1135 * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1136 * get the metadata, that is, the parts of the flow key that cannot be
1137 * extracted from the packet itself.
1138 */
1139int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
1140			       const struct nlattr *attr)
1141{
1142	const struct nlattr *nla;
1143	int rem;
1144
1145	*in_port = USHRT_MAX;
1146	*priority = 0;
1147
1148	nla_for_each_nested(nla, attr, rem) {
1149		int type = nla_type(nla);
1150
1151		if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1152			if (nla_len(nla) != ovs_key_lens[type])
1153				return -EINVAL;
1154
1155			switch (type) {
1156			case OVS_KEY_ATTR_PRIORITY:
1157				*priority = nla_get_u32(nla);
1158				break;
1159
1160			case OVS_KEY_ATTR_IN_PORT:
1161				if (nla_get_u32(nla) >= DP_MAX_PORTS)
1162					return -EINVAL;
1163				*in_port = nla_get_u32(nla);
1164				break;
1165			}
1166		}
1167	}
1168	if (rem)
1169		return -EINVAL;
1170	return 0;
1171}
1172
1173int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1174{
1175	struct ovs_key_ethernet *eth_key;
1176	struct nlattr *nla, *encap;
1177
1178	if (swkey->phy.priority &&
1179	    nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1180		goto nla_put_failure;
1181
1182	if (swkey->phy.in_port != USHRT_MAX &&
1183	    nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1184		goto nla_put_failure;
1185
1186	nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1187	if (!nla)
1188		goto nla_put_failure;
1189	eth_key = nla_data(nla);
1190	memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1191	memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1192
1193	if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1194		if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1195		    nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1196			goto nla_put_failure;
1197		encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1198		if (!swkey->eth.tci)
1199			goto unencap;
1200	} else {
1201		encap = NULL;
1202	}
1203
1204	if (swkey->eth.type == htons(ETH_P_802_2))
1205		goto unencap;
1206
1207	if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1208		goto nla_put_failure;
1209
1210	if (swkey->eth.type == htons(ETH_P_IP)) {
1211		struct ovs_key_ipv4 *ipv4_key;
1212
1213		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1214		if (!nla)
1215			goto nla_put_failure;
1216		ipv4_key = nla_data(nla);
1217		ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1218		ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1219		ipv4_key->ipv4_proto = swkey->ip.proto;
1220		ipv4_key->ipv4_tos = swkey->ip.tos;
1221		ipv4_key->ipv4_ttl = swkey->ip.ttl;
1222		ipv4_key->ipv4_frag = swkey->ip.frag;
1223	} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1224		struct ovs_key_ipv6 *ipv6_key;
1225
1226		nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1227		if (!nla)
1228			goto nla_put_failure;
1229		ipv6_key = nla_data(nla);
1230		memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1231				sizeof(ipv6_key->ipv6_src));
1232		memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1233				sizeof(ipv6_key->ipv6_dst));
1234		ipv6_key->ipv6_label = swkey->ipv6.label;
1235		ipv6_key->ipv6_proto = swkey->ip.proto;
1236		ipv6_key->ipv6_tclass = swkey->ip.tos;
1237		ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1238		ipv6_key->ipv6_frag = swkey->ip.frag;
1239	} else if (swkey->eth.type == htons(ETH_P_ARP)) {
1240		struct ovs_key_arp *arp_key;
1241
1242		nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1243		if (!nla)
1244			goto nla_put_failure;
1245		arp_key = nla_data(nla);
1246		memset(arp_key, 0, sizeof(struct ovs_key_arp));
1247		arp_key->arp_sip = swkey->ipv4.addr.src;
1248		arp_key->arp_tip = swkey->ipv4.addr.dst;
1249		arp_key->arp_op = htons(swkey->ip.proto);
1250		memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1251		memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1252	}
1253
1254	if ((swkey->eth.type == htons(ETH_P_IP) ||
1255	     swkey->eth.type == htons(ETH_P_IPV6)) &&
1256	     swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1257
1258		if (swkey->ip.proto == IPPROTO_TCP) {
1259			struct ovs_key_tcp *tcp_key;
1260
1261			nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1262			if (!nla)
1263				goto nla_put_failure;
1264			tcp_key = nla_data(nla);
1265			if (swkey->eth.type == htons(ETH_P_IP)) {
1266				tcp_key->tcp_src = swkey->ipv4.tp.src;
1267				tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1268			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1269				tcp_key->tcp_src = swkey->ipv6.tp.src;
1270				tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1271			}
1272		} else if (swkey->ip.proto == IPPROTO_UDP) {
1273			struct ovs_key_udp *udp_key;
1274
1275			nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1276			if (!nla)
1277				goto nla_put_failure;
1278			udp_key = nla_data(nla);
1279			if (swkey->eth.type == htons(ETH_P_IP)) {
1280				udp_key->udp_src = swkey->ipv4.tp.src;
1281				udp_key->udp_dst = swkey->ipv4.tp.dst;
1282			} else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1283				udp_key->udp_src = swkey->ipv6.tp.src;
1284				udp_key->udp_dst = swkey->ipv6.tp.dst;
1285			}
1286		} else if (swkey->eth.type == htons(ETH_P_IP) &&
1287			   swkey->ip.proto == IPPROTO_ICMP) {
1288			struct ovs_key_icmp *icmp_key;
1289
1290			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1291			if (!nla)
1292				goto nla_put_failure;
1293			icmp_key = nla_data(nla);
1294			icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1295			icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1296		} else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1297			   swkey->ip.proto == IPPROTO_ICMPV6) {
1298			struct ovs_key_icmpv6 *icmpv6_key;
1299
1300			nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1301						sizeof(*icmpv6_key));
1302			if (!nla)
1303				goto nla_put_failure;
1304			icmpv6_key = nla_data(nla);
1305			icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1306			icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1307
1308			if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1309			    icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1310				struct ovs_key_nd *nd_key;
1311
1312				nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1313				if (!nla)
1314					goto nla_put_failure;
1315				nd_key = nla_data(nla);
1316				memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1317							sizeof(nd_key->nd_target));
1318				memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1319				memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1320			}
1321		}
1322	}
1323
1324unencap:
1325	if (encap)
1326		nla_nest_end(skb, encap);
1327
1328	return 0;
1329
1330nla_put_failure:
1331	return -EMSGSIZE;
1332}
1333
1334/* Initializes the flow module.
1335 * Returns zero if successful or a negative error code. */
1336int ovs_flow_init(void)
1337{
1338	flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1339					0, NULL);
1340	if (flow_cache == NULL)
1341		return -ENOMEM;
1342
1343	return 0;
1344}
1345
1346/* Uninitializes the flow module. */
1347void ovs_flow_exit(void)
1348{
1349	kmem_cache_destroy(flow_cache);
1350}