flow.c revision bb6f9a708d4067713afae2e9eb2637f6b4c01ecb 
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 = flow->key.tp.flags;
68	int node = numa_node_id();
69
70	stats = rcu_dereference(flow->stats[node]);
71
72	/* Check if already have node-specific stats. */
73	if (likely(stats)) {
74		spin_lock(&stats->lock);
75		/* Mark if we write on the pre-allocated stats. */
76		if (node == 0 && unlikely(flow->stats_last_writer != node))
77			flow->stats_last_writer = node;
78	} else {
79		stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
80		spin_lock(&stats->lock);
81
82		/* If the current NUMA-node is the only writer on the
83		 * pre-allocated stats keep using them.
84		 */
85		if (unlikely(flow->stats_last_writer != node)) {
86			/* A previous locker may have already allocated the
87			 * stats, so we need to check again.  If node-specific
88			 * stats were already allocated, we update the pre-
89			 * allocated stats as we have already locked them.
90			 */
91			if (likely(flow->stats_last_writer != NUMA_NO_NODE)
92			    && likely(!rcu_dereference(flow->stats[node]))) {
93				/* Try to allocate node-specific stats. */
94				struct flow_stats *new_stats;
95
96				new_stats =
97					kmem_cache_alloc_node(flow_stats_cache,
98							      GFP_THISNODE |
99							      __GFP_NOMEMALLOC,
100							      node);
101				if (likely(new_stats)) {
102					new_stats->used = jiffies;
103					new_stats->packet_count = 1;
104					new_stats->byte_count = skb->len;
105					new_stats->tcp_flags = tcp_flags;
106					spin_lock_init(&new_stats->lock);
107
108					rcu_assign_pointer(flow->stats[node],
109							   new_stats);
110					goto unlock;
111				}
112			}
113			flow->stats_last_writer = node;
114		}
115	}
116
117	stats->used = jiffies;
118	stats->packet_count++;
119	stats->byte_count += skb->len;
120	stats->tcp_flags |= tcp_flags;
121unlock:
122	spin_unlock(&stats->lock);
123}
124
125/* Called with ovs_mutex. */
126void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
127			unsigned long *used, __be16 *tcp_flags)
128{
129	int node;
130
131	*used = 0;
132	*tcp_flags = 0;
133	memset(ovs_stats, 0, sizeof(*ovs_stats));
134
135	for_each_node(node) {
136		struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
137
138		if (stats) {
139			/* Local CPU may write on non-local stats, so we must
140			 * block bottom-halves here.
141			 */
142			spin_lock_bh(&stats->lock);
143			if (!*used || time_after(stats->used, *used))
144				*used = stats->used;
145			*tcp_flags |= stats->tcp_flags;
146			ovs_stats->n_packets += stats->packet_count;
147			ovs_stats->n_bytes += stats->byte_count;
148			spin_unlock_bh(&stats->lock);
149		}
150	}
151}
152
153void ovs_flow_stats_clear(struct sw_flow *flow)
154{
155	int node;
156
157	for_each_node(node) {
158		struct flow_stats *stats = rcu_dereference(flow->stats[node]);
159
160		if (stats) {
161			spin_lock_bh(&stats->lock);
162			stats->used = 0;
163			stats->packet_count = 0;
164			stats->byte_count = 0;
165			stats->tcp_flags = 0;
166			spin_unlock_bh(&stats->lock);
167		}
168	}
169}
170
171static int check_header(struct sk_buff *skb, int len)
172{
173	if (unlikely(skb->len < len))
174		return -EINVAL;
175	if (unlikely(!pskb_may_pull(skb, len)))
176		return -ENOMEM;
177	return 0;
178}
179
180static bool arphdr_ok(struct sk_buff *skb)
181{
182	return pskb_may_pull(skb, skb_network_offset(skb) +
183				  sizeof(struct arp_eth_header));
184}
185
186static int check_iphdr(struct sk_buff *skb)
187{
188	unsigned int nh_ofs = skb_network_offset(skb);
189	unsigned int ip_len;
190	int err;
191
192	err = check_header(skb, nh_ofs + sizeof(struct iphdr));
193	if (unlikely(err))
194		return err;
195
196	ip_len = ip_hdrlen(skb);
197	if (unlikely(ip_len < sizeof(struct iphdr) ||
198		     skb->len < nh_ofs + ip_len))
199		return -EINVAL;
200
201	skb_set_transport_header(skb, nh_ofs + ip_len);
202	return 0;
203}
204
205static bool tcphdr_ok(struct sk_buff *skb)
206{
207	int th_ofs = skb_transport_offset(skb);
208	int tcp_len;
209
210	if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
211		return false;
212
213	tcp_len = tcp_hdrlen(skb);
214	if (unlikely(tcp_len < sizeof(struct tcphdr) ||
215		     skb->len < th_ofs + tcp_len))
216		return false;
217
218	return true;
219}
220
221static bool udphdr_ok(struct sk_buff *skb)
222{
223	return pskb_may_pull(skb, skb_transport_offset(skb) +
224				  sizeof(struct udphdr));
225}
226
227static bool sctphdr_ok(struct sk_buff *skb)
228{
229	return pskb_may_pull(skb, skb_transport_offset(skb) +
230				  sizeof(struct sctphdr));
231}
232
233static bool icmphdr_ok(struct sk_buff *skb)
234{
235	return pskb_may_pull(skb, skb_transport_offset(skb) +
236				  sizeof(struct icmphdr));
237}
238
239static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
240{
241	unsigned int nh_ofs = skb_network_offset(skb);
242	unsigned int nh_len;
243	int payload_ofs;
244	struct ipv6hdr *nh;
245	uint8_t nexthdr;
246	__be16 frag_off;
247	int err;
248
249	err = check_header(skb, nh_ofs + sizeof(*nh));
250	if (unlikely(err))
251		return err;
252
253	nh = ipv6_hdr(skb);
254	nexthdr = nh->nexthdr;
255	payload_ofs = (u8 *)(nh + 1) - skb->data;
256
257	key->ip.proto = NEXTHDR_NONE;
258	key->ip.tos = ipv6_get_dsfield(nh);
259	key->ip.ttl = nh->hop_limit;
260	key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
261	key->ipv6.addr.src = nh->saddr;
262	key->ipv6.addr.dst = nh->daddr;
263
264	payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
265	if (unlikely(payload_ofs < 0))
266		return -EINVAL;
267
268	if (frag_off) {
269		if (frag_off & htons(~0x7))
270			key->ip.frag = OVS_FRAG_TYPE_LATER;
271		else
272			key->ip.frag = OVS_FRAG_TYPE_FIRST;
273	}
274
275	nh_len = payload_ofs - nh_ofs;
276	skb_set_transport_header(skb, nh_ofs + nh_len);
277	key->ip.proto = nexthdr;
278	return nh_len;
279}
280
281static bool icmp6hdr_ok(struct sk_buff *skb)
282{
283	return pskb_may_pull(skb, skb_transport_offset(skb) +
284				  sizeof(struct icmp6hdr));
285}
286
287static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
288{
289	struct qtag_prefix {
290		__be16 eth_type; /* ETH_P_8021Q */
291		__be16 tci;
292	};
293	struct qtag_prefix *qp;
294
295	if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
296		return 0;
297
298	if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
299					 sizeof(__be16))))
300		return -ENOMEM;
301
302	qp = (struct qtag_prefix *) skb->data;
303	key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
304	__skb_pull(skb, sizeof(struct qtag_prefix));
305
306	return 0;
307}
308
309static __be16 parse_ethertype(struct sk_buff *skb)
310{
311	struct llc_snap_hdr {
312		u8  dsap;  /* Always 0xAA */
313		u8  ssap;  /* Always 0xAA */
314		u8  ctrl;
315		u8  oui[3];
316		__be16 ethertype;
317	};
318	struct llc_snap_hdr *llc;
319	__be16 proto;
320
321	proto = *(__be16 *) skb->data;
322	__skb_pull(skb, sizeof(__be16));
323
324	if (ntohs(proto) >= ETH_P_802_3_MIN)
325		return proto;
326
327	if (skb->len < sizeof(struct llc_snap_hdr))
328		return htons(ETH_P_802_2);
329
330	if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
331		return htons(0);
332
333	llc = (struct llc_snap_hdr *) skb->data;
334	if (llc->dsap != LLC_SAP_SNAP ||
335	    llc->ssap != LLC_SAP_SNAP ||
336	    (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
337		return htons(ETH_P_802_2);
338
339	__skb_pull(skb, sizeof(struct llc_snap_hdr));
340
341	if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
342		return llc->ethertype;
343
344	return htons(ETH_P_802_2);
345}
346
347static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
348			int nh_len)
349{
350	struct icmp6hdr *icmp = icmp6_hdr(skb);
351
352	/* The ICMPv6 type and code fields use the 16-bit transport port
353	 * fields, so we need to store them in 16-bit network byte order.
354	 */
355	key->tp.src = htons(icmp->icmp6_type);
356	key->tp.dst = htons(icmp->icmp6_code);
357
358	if (icmp->icmp6_code == 0 &&
359	    (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
360	     icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
361		int icmp_len = skb->len - skb_transport_offset(skb);
362		struct nd_msg *nd;
363		int offset;
364
365		/* In order to process neighbor discovery options, we need the
366		 * entire packet.
367		 */
368		if (unlikely(icmp_len < sizeof(*nd)))
369			return 0;
370
371		if (unlikely(skb_linearize(skb)))
372			return -ENOMEM;
373
374		nd = (struct nd_msg *)skb_transport_header(skb);
375		key->ipv6.nd.target = nd->target;
376
377		icmp_len -= sizeof(*nd);
378		offset = 0;
379		while (icmp_len >= 8) {
380			struct nd_opt_hdr *nd_opt =
381				 (struct nd_opt_hdr *)(nd->opt + offset);
382			int opt_len = nd_opt->nd_opt_len * 8;
383
384			if (unlikely(!opt_len || opt_len > icmp_len))
385				return 0;
386
387			/* Store the link layer address if the appropriate
388			 * option is provided.  It is considered an error if
389			 * the same link layer option is specified twice.
390			 */
391			if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
392			    && opt_len == 8) {
393				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
394					goto invalid;
395				ether_addr_copy(key->ipv6.nd.sll,
396						&nd->opt[offset+sizeof(*nd_opt)]);
397			} else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
398				   && opt_len == 8) {
399				if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
400					goto invalid;
401				ether_addr_copy(key->ipv6.nd.tll,
402						&nd->opt[offset+sizeof(*nd_opt)]);
403			}
404
405			icmp_len -= opt_len;
406			offset += opt_len;
407		}
408	}
409
410	return 0;
411
412invalid:
413	memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
414	memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
415	memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
416
417	return 0;
418}
419
420/**
421 * ovs_flow_extract - extracts a flow key from an Ethernet frame.
422 * @skb: sk_buff that contains the frame, with skb->data pointing to the
423 * Ethernet header
424 * @in_port: port number on which @skb was received.
425 * @key: output flow key
426 *
427 * The caller must ensure that skb->len >= ETH_HLEN.
428 *
429 * Returns 0 if successful, otherwise a negative errno value.
430 *
431 * Initializes @skb header pointers as follows:
432 *
433 *    - skb->mac_header: the Ethernet header.
434 *
435 *    - skb->network_header: just past the Ethernet header, or just past the
436 *      VLAN header, to the first byte of the Ethernet payload.
437 *
438 *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
439 *      on output, then just past the IP header, if one is present and
440 *      of a correct length, otherwise the same as skb->network_header.
441 *      For other key->eth.type values it is left untouched.
442 */
443int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
444{
445	int error;
446	struct ethhdr *eth;
447
448	memset(key, 0, sizeof(*key));
449
450	key->phy.priority = skb->priority;
451	if (OVS_CB(skb)->tun_key)
452		memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
453	key->phy.in_port = in_port;
454	key->phy.skb_mark = skb->mark;
455
456	skb_reset_mac_header(skb);
457
458	/* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
459	 * header in the linear data area.
460	 */
461	eth = eth_hdr(skb);
462	ether_addr_copy(key->eth.src, eth->h_source);
463	ether_addr_copy(key->eth.dst, eth->h_dest);
464
465	__skb_pull(skb, 2 * ETH_ALEN);
466	/* We are going to push all headers that we pull, so no need to
467	 * update skb->csum here.
468	 */
469
470	if (vlan_tx_tag_present(skb))
471		key->eth.tci = htons(skb->vlan_tci);
472	else if (eth->h_proto == htons(ETH_P_8021Q))
473		if (unlikely(parse_vlan(skb, key)))
474			return -ENOMEM;
475
476	key->eth.type = parse_ethertype(skb);
477	if (unlikely(key->eth.type == htons(0)))
478		return -ENOMEM;
479
480	skb_reset_network_header(skb);
481	__skb_push(skb, skb->data - skb_mac_header(skb));
482
483	/* Network layer. */
484	if (key->eth.type == htons(ETH_P_IP)) {
485		struct iphdr *nh;
486		__be16 offset;
487
488		error = check_iphdr(skb);
489		if (unlikely(error)) {
490			if (error == -EINVAL) {
491				skb->transport_header = skb->network_header;
492				error = 0;
493			}
494			return error;
495		}
496
497		nh = ip_hdr(skb);
498		key->ipv4.addr.src = nh->saddr;
499		key->ipv4.addr.dst = nh->daddr;
500
501		key->ip.proto = nh->protocol;
502		key->ip.tos = nh->tos;
503		key->ip.ttl = nh->ttl;
504
505		offset = nh->frag_off & htons(IP_OFFSET);
506		if (offset) {
507			key->ip.frag = OVS_FRAG_TYPE_LATER;
508			return 0;
509		}
510		if (nh->frag_off & htons(IP_MF) ||
511			 skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
512			key->ip.frag = OVS_FRAG_TYPE_FIRST;
513
514		/* Transport layer. */
515		if (key->ip.proto == IPPROTO_TCP) {
516			if (tcphdr_ok(skb)) {
517				struct tcphdr *tcp = tcp_hdr(skb);
518				key->tp.src = tcp->source;
519				key->tp.dst = tcp->dest;
520				key->tp.flags = TCP_FLAGS_BE16(tcp);
521			}
522		} else if (key->ip.proto == IPPROTO_UDP) {
523			if (udphdr_ok(skb)) {
524				struct udphdr *udp = udp_hdr(skb);
525				key->tp.src = udp->source;
526				key->tp.dst = udp->dest;
527			}
528		} else if (key->ip.proto == IPPROTO_SCTP) {
529			if (sctphdr_ok(skb)) {
530				struct sctphdr *sctp = sctp_hdr(skb);
531				key->tp.src = sctp->source;
532				key->tp.dst = sctp->dest;
533			}
534		} else if (key->ip.proto == IPPROTO_ICMP) {
535			if (icmphdr_ok(skb)) {
536				struct icmphdr *icmp = icmp_hdr(skb);
537				/* The ICMP type and code fields use the 16-bit
538				 * transport port fields, so we need to store
539				 * them in 16-bit network byte order. */
540				key->tp.src = htons(icmp->type);
541				key->tp.dst = htons(icmp->code);
542			}
543		}
544
545	} else if ((key->eth.type == htons(ETH_P_ARP) ||
546		   key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
547		struct arp_eth_header *arp;
548
549		arp = (struct arp_eth_header *)skb_network_header(skb);
550
551		if (arp->ar_hrd == htons(ARPHRD_ETHER)
552				&& arp->ar_pro == htons(ETH_P_IP)
553				&& arp->ar_hln == ETH_ALEN
554				&& arp->ar_pln == 4) {
555
556			/* We only match on the lower 8 bits of the opcode. */
557			if (ntohs(arp->ar_op) <= 0xff)
558				key->ip.proto = ntohs(arp->ar_op);
559			memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
560			memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
561			ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
562			ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
563		}
564	} else if (key->eth.type == htons(ETH_P_IPV6)) {
565		int nh_len;             /* IPv6 Header + Extensions */
566
567		nh_len = parse_ipv6hdr(skb, key);
568		if (unlikely(nh_len < 0)) {
569			if (nh_len == -EINVAL) {
570				skb->transport_header = skb->network_header;
571				error = 0;
572			} else {
573				error = nh_len;
574			}
575			return error;
576		}
577
578		if (key->ip.frag == OVS_FRAG_TYPE_LATER)
579			return 0;
580		if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
581			key->ip.frag = OVS_FRAG_TYPE_FIRST;
582
583		/* Transport layer. */
584		if (key->ip.proto == NEXTHDR_TCP) {
585			if (tcphdr_ok(skb)) {
586				struct tcphdr *tcp = tcp_hdr(skb);
587				key->tp.src = tcp->source;
588				key->tp.dst = tcp->dest;
589				key->tp.flags = TCP_FLAGS_BE16(tcp);
590			}
591		} else if (key->ip.proto == NEXTHDR_UDP) {
592			if (udphdr_ok(skb)) {
593				struct udphdr *udp = udp_hdr(skb);
594				key->tp.src = udp->source;
595				key->tp.dst = udp->dest;
596			}
597		} else if (key->ip.proto == NEXTHDR_SCTP) {
598			if (sctphdr_ok(skb)) {
599				struct sctphdr *sctp = sctp_hdr(skb);
600				key->tp.src = sctp->source;
601				key->tp.dst = sctp->dest;
602			}
603		} else if (key->ip.proto == NEXTHDR_ICMP) {
604			if (icmp6hdr_ok(skb)) {
605				error = parse_icmpv6(skb, key, nh_len);
606				if (error)
607					return error;
608			}
609		}
610	}
611
612	return 0;
613}
614