datapath: Remove 5-tuple optimization.
[sliver-openvswitch.git] / datapath / flow.c
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/ipv6.h>
46 #include <net/ndisc.h>
47
48 #include "vlan.h"
49
50 u64 ovs_flow_used_time(unsigned long flow_jiffies)
51 {
52         struct timespec cur_ts;
53         u64 cur_ms, idle_ms;
54
55         ktime_get_ts(&cur_ts);
56         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
57         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
58                  cur_ts.tv_nsec / NSEC_PER_MSEC;
59
60         return cur_ms - idle_ms;
61 }
62
63 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
64
65 void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
66 {
67         struct flow_stats *stats;
68         __be16 tcp_flags = 0;
69
70         stats = this_cpu_ptr(flow->stats);
71
72         if ((flow->key.eth.type == htons(ETH_P_IP) ||
73              flow->key.eth.type == htons(ETH_P_IPV6)) &&
74             flow->key.ip.frag != OVS_FRAG_TYPE_LATER &&
75             flow->key.ip.proto == IPPROTO_TCP &&
76             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
77                 tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
78         }
79
80         spin_lock(&stats->lock);
81         stats->used = jiffies;
82         stats->packet_count++;
83         stats->byte_count += skb->len;
84         stats->tcp_flags |= tcp_flags;
85         spin_unlock(&stats->lock);
86 }
87
88 static void stats_read(struct flow_stats *stats, bool lock_bh,
89                        struct ovs_flow_stats *ovs_stats,
90                        unsigned long *used, __be16 *tcp_flags)
91 {
92         if (lock_bh)
93                 spin_lock_bh(&stats->lock);
94         else
95                 spin_lock(&stats->lock);
96
97         if (time_after(stats->used, *used))
98                 *used = stats->used;
99         *tcp_flags |= stats->tcp_flags;
100         ovs_stats->n_packets += stats->packet_count;
101         ovs_stats->n_bytes += stats->byte_count;
102
103         if (lock_bh)
104                 spin_unlock_bh(&stats->lock);
105         else
106                 spin_unlock(&stats->lock);
107 }
108
109 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
110                         unsigned long *used, __be16 *tcp_flags)
111 {
112         int cpu, cur_cpu;
113
114         *used = 0;
115         *tcp_flags = 0;
116         memset(ovs_stats, 0, sizeof(*ovs_stats));
117
118         cur_cpu = get_cpu();
119
120         for_each_possible_cpu(cpu) {
121                 struct flow_stats *stats;
122                 bool lock_bh;
123
124                 stats = per_cpu_ptr(flow->stats, cpu);
125                 lock_bh = (cpu == cur_cpu);
126                 stats_read(stats, lock_bh, ovs_stats, used, tcp_flags);
127         }
128         put_cpu();
129 }
130
131 static void stats_reset(struct flow_stats *stats, bool lock_bh)
132 {
133         if (lock_bh)
134                 spin_lock_bh(&stats->lock);
135         else
136                 spin_lock(&stats->lock);
137
138         stats->used = 0;
139         stats->packet_count = 0;
140         stats->byte_count = 0;
141         stats->tcp_flags = 0;
142
143         if (lock_bh)
144                 spin_unlock_bh(&stats->lock);
145         else
146                 spin_unlock(&stats->lock);
147 }
148
149 void ovs_flow_stats_clear(struct sw_flow *flow)
150 {
151         int cpu, cur_cpu;
152
153         cur_cpu = get_cpu();
154
155         for_each_possible_cpu(cpu) {
156                 bool lock_bh;
157
158                 lock_bh = (cpu == cur_cpu);
159                 stats_reset(per_cpu_ptr(flow->stats, cpu), lock_bh);
160         }
161         put_cpu();
162 }
163
164 static int check_header(struct sk_buff *skb, int len)
165 {
166         if (unlikely(skb->len < len))
167                 return -EINVAL;
168         if (unlikely(!pskb_may_pull(skb, len)))
169                 return -ENOMEM;
170         return 0;
171 }
172
173 static bool arphdr_ok(struct sk_buff *skb)
174 {
175         return pskb_may_pull(skb, skb_network_offset(skb) +
176                                   sizeof(struct arp_eth_header));
177 }
178
179 static int check_iphdr(struct sk_buff *skb)
180 {
181         unsigned int nh_ofs = skb_network_offset(skb);
182         unsigned int ip_len;
183         int err;
184
185         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
186         if (unlikely(err))
187                 return err;
188
189         ip_len = ip_hdrlen(skb);
190         if (unlikely(ip_len < sizeof(struct iphdr) ||
191                      skb->len < nh_ofs + ip_len))
192                 return -EINVAL;
193
194         skb_set_transport_header(skb, nh_ofs + ip_len);
195         return 0;
196 }
197
198 static bool tcphdr_ok(struct sk_buff *skb)
199 {
200         int th_ofs = skb_transport_offset(skb);
201         int tcp_len;
202
203         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
204                 return false;
205
206         tcp_len = tcp_hdrlen(skb);
207         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
208                      skb->len < th_ofs + tcp_len))
209                 return false;
210
211         return true;
212 }
213
214 static bool udphdr_ok(struct sk_buff *skb)
215 {
216         return pskb_may_pull(skb, skb_transport_offset(skb) +
217                                   sizeof(struct udphdr));
218 }
219
220 static bool sctphdr_ok(struct sk_buff *skb)
221 {
222         return pskb_may_pull(skb, skb_transport_offset(skb) +
223                                   sizeof(struct sctphdr));
224 }
225
226 static bool icmphdr_ok(struct sk_buff *skb)
227 {
228         return pskb_may_pull(skb, skb_transport_offset(skb) +
229                                   sizeof(struct icmphdr));
230 }
231
232 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
233 {
234         unsigned int nh_ofs = skb_network_offset(skb);
235         unsigned int nh_len;
236         int payload_ofs;
237         struct ipv6hdr *nh;
238         uint8_t nexthdr;
239         __be16 frag_off;
240         int err;
241
242         err = check_header(skb, nh_ofs + sizeof(*nh));
243         if (unlikely(err))
244                 return err;
245
246         nh = ipv6_hdr(skb);
247         nexthdr = nh->nexthdr;
248         payload_ofs = (u8 *)(nh + 1) - skb->data;
249
250         key->ip.proto = NEXTHDR_NONE;
251         key->ip.tos = ipv6_get_dsfield(nh);
252         key->ip.ttl = nh->hop_limit;
253         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
254         key->ipv6.addr.src = nh->saddr;
255         key->ipv6.addr.dst = nh->daddr;
256
257         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
258         if (unlikely(payload_ofs < 0))
259                 return -EINVAL;
260
261         if (frag_off) {
262                 if (frag_off & htons(~0x7))
263                         key->ip.frag = OVS_FRAG_TYPE_LATER;
264                 else
265                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
266         }
267
268         nh_len = payload_ofs - nh_ofs;
269         skb_set_transport_header(skb, nh_ofs + nh_len);
270         key->ip.proto = nexthdr;
271         return nh_len;
272 }
273
274 static bool icmp6hdr_ok(struct sk_buff *skb)
275 {
276         return pskb_may_pull(skb, skb_transport_offset(skb) +
277                                   sizeof(struct icmp6hdr));
278 }
279
280 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
281 {
282         struct qtag_prefix {
283                 __be16 eth_type; /* ETH_P_8021Q */
284                 __be16 tci;
285         };
286         struct qtag_prefix *qp;
287
288         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
289                 return 0;
290
291         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
292                                          sizeof(__be16))))
293                 return -ENOMEM;
294
295         qp = (struct qtag_prefix *) skb->data;
296         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
297         __skb_pull(skb, sizeof(struct qtag_prefix));
298
299         return 0;
300 }
301
302 static __be16 parse_ethertype(struct sk_buff *skb)
303 {
304         struct llc_snap_hdr {
305                 u8  dsap;  /* Always 0xAA */
306                 u8  ssap;  /* Always 0xAA */
307                 u8  ctrl;
308                 u8  oui[3];
309                 __be16 ethertype;
310         };
311         struct llc_snap_hdr *llc;
312         __be16 proto;
313
314         proto = *(__be16 *) skb->data;
315         __skb_pull(skb, sizeof(__be16));
316
317         if (ntohs(proto) >= ETH_P_802_3_MIN)
318                 return proto;
319
320         if (skb->len < sizeof(struct llc_snap_hdr))
321                 return htons(ETH_P_802_2);
322
323         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
324                 return htons(0);
325
326         llc = (struct llc_snap_hdr *) skb->data;
327         if (llc->dsap != LLC_SAP_SNAP ||
328             llc->ssap != LLC_SAP_SNAP ||
329             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
330                 return htons(ETH_P_802_2);
331
332         __skb_pull(skb, sizeof(struct llc_snap_hdr));
333
334         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
335                 return llc->ethertype;
336
337         return htons(ETH_P_802_2);
338 }
339
340 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
341                         int nh_len)
342 {
343         struct icmp6hdr *icmp = icmp6_hdr(skb);
344
345         /* The ICMPv6 type and code fields use the 16-bit transport port
346          * fields, so we need to store them in 16-bit network byte order.
347          */
348         key->ipv6.tp.src = htons(icmp->icmp6_type);
349         key->ipv6.tp.dst = htons(icmp->icmp6_code);
350
351         if (icmp->icmp6_code == 0 &&
352             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
353              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
354                 int icmp_len = skb->len - skb_transport_offset(skb);
355                 struct nd_msg *nd;
356                 int offset;
357
358                 /* In order to process neighbor discovery options, we need the
359                  * entire packet.
360                  */
361                 if (unlikely(icmp_len < sizeof(*nd)))
362                         return 0;
363
364                 if (unlikely(skb_linearize(skb)))
365                         return -ENOMEM;
366
367                 nd = (struct nd_msg *)skb_transport_header(skb);
368                 key->ipv6.nd.target = nd->target;
369
370                 icmp_len -= sizeof(*nd);
371                 offset = 0;
372                 while (icmp_len >= 8) {
373                         struct nd_opt_hdr *nd_opt =
374                                  (struct nd_opt_hdr *)(nd->opt + offset);
375                         int opt_len = nd_opt->nd_opt_len * 8;
376
377                         if (unlikely(!opt_len || opt_len > icmp_len))
378                                 return 0;
379
380                         /* Store the link layer address if the appropriate
381                          * option is provided.  It is considered an error if
382                          * the same link layer option is specified twice.
383                          */
384                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
385                             && opt_len == 8) {
386                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
387                                         goto invalid;
388                                 memcpy(key->ipv6.nd.sll,
389                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
390                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
391                                    && opt_len == 8) {
392                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
393                                         goto invalid;
394                                 memcpy(key->ipv6.nd.tll,
395                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
396                         }
397
398                         icmp_len -= opt_len;
399                         offset += opt_len;
400                 }
401         }
402
403         return 0;
404
405 invalid:
406         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
407         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
408         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
409
410         return 0;
411 }
412
413 /**
414  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
415  * @skb: sk_buff that contains the frame, with skb->data pointing to the
416  * Ethernet header
417  * @in_port: port number on which @skb was received.
418  * @key: output flow key
419  *
420  * The caller must ensure that skb->len >= ETH_HLEN.
421  *
422  * Returns 0 if successful, otherwise a negative errno value.
423  *
424  * Initializes @skb header pointers as follows:
425  *
426  *    - skb->mac_header: the Ethernet header.
427  *
428  *    - skb->network_header: just past the Ethernet header, or just past the
429  *      VLAN header, to the first byte of the Ethernet payload.
430  *
431  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
432  *      on output, then just past the IP header, if one is present and
433  *      of a correct length, otherwise the same as skb->network_header.
434  *      For other key->eth.type values it is left untouched.
435  */
436 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
437 {
438         int error;
439         struct ethhdr *eth;
440
441         memset(key, 0, sizeof(*key));
442
443         key->phy.priority = skb->priority;
444         if (OVS_CB(skb)->tun_key)
445                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
446         key->phy.in_port = in_port;
447         key->phy.skb_mark = skb->mark;
448
449         skb_reset_mac_header(skb);
450
451         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
452          * header in the linear data area.
453          */
454         eth = eth_hdr(skb);
455         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
456         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
457
458         __skb_pull(skb, 2 * ETH_ALEN);
459         /* We are going to push all headers that we pull, so no need to
460          * update skb->csum here. */
461
462         if (vlan_tx_tag_present(skb))
463                 key->eth.tci = htons(vlan_get_tci(skb));
464         else if (eth->h_proto == htons(ETH_P_8021Q))
465                 if (unlikely(parse_vlan(skb, key)))
466                         return -ENOMEM;
467
468         key->eth.type = parse_ethertype(skb);
469         if (unlikely(key->eth.type == htons(0)))
470                 return -ENOMEM;
471
472         skb_reset_network_header(skb);
473         __skb_push(skb, skb->data - skb_mac_header(skb));
474
475         /* Network layer. */
476         if (key->eth.type == htons(ETH_P_IP)) {
477                 struct iphdr *nh;
478                 __be16 offset;
479
480                 error = check_iphdr(skb);
481                 if (unlikely(error)) {
482                         if (error == -EINVAL) {
483                                 skb->transport_header = skb->network_header;
484                                 error = 0;
485                         }
486                         return error;
487                 }
488
489                 nh = ip_hdr(skb);
490                 key->ipv4.addr.src = nh->saddr;
491                 key->ipv4.addr.dst = nh->daddr;
492
493                 key->ip.proto = nh->protocol;
494                 key->ip.tos = nh->tos;
495                 key->ip.ttl = nh->ttl;
496
497                 offset = nh->frag_off & htons(IP_OFFSET);
498                 if (offset) {
499                         key->ip.frag = OVS_FRAG_TYPE_LATER;
500                         return 0;
501                 }
502                 if (nh->frag_off & htons(IP_MF) ||
503                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
504                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
505
506                 /* Transport layer. */
507                 if (key->ip.proto == IPPROTO_TCP) {
508                         if (tcphdr_ok(skb)) {
509                                 struct tcphdr *tcp = tcp_hdr(skb);
510                                 key->ipv4.tp.src = tcp->source;
511                                 key->ipv4.tp.dst = tcp->dest;
512                                 key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
513                         }
514                 } else if (key->ip.proto == IPPROTO_UDP) {
515                         if (udphdr_ok(skb)) {
516                                 struct udphdr *udp = udp_hdr(skb);
517                                 key->ipv4.tp.src = udp->source;
518                                 key->ipv4.tp.dst = udp->dest;
519                         }
520                 } else if (key->ip.proto == IPPROTO_SCTP) {
521                         if (sctphdr_ok(skb)) {
522                                 struct sctphdr *sctp = sctp_hdr(skb);
523                                 key->ipv4.tp.src = sctp->source;
524                                 key->ipv4.tp.dst = sctp->dest;
525                         }
526                 } else if (key->ip.proto == IPPROTO_ICMP) {
527                         if (icmphdr_ok(skb)) {
528                                 struct icmphdr *icmp = icmp_hdr(skb);
529                                 /* The ICMP type and code fields use the 16-bit
530                                  * transport port fields, so we need to store
531                                  * them in 16-bit network byte order. */
532                                 key->ipv4.tp.src = htons(icmp->type);
533                                 key->ipv4.tp.dst = htons(icmp->code);
534                         }
535                 }
536
537         } else if ((key->eth.type == htons(ETH_P_ARP) ||
538                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
539                 struct arp_eth_header *arp;
540
541                 arp = (struct arp_eth_header *)skb_network_header(skb);
542
543                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
544                                 && arp->ar_pro == htons(ETH_P_IP)
545                                 && arp->ar_hln == ETH_ALEN
546                                 && arp->ar_pln == 4) {
547
548                         /* We only match on the lower 8 bits of the opcode. */
549                         if (ntohs(arp->ar_op) <= 0xff)
550                                 key->ip.proto = ntohs(arp->ar_op);
551                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
552                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
553                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
554                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
555                 }
556         } else if (key->eth.type == htons(ETH_P_IPV6)) {
557                 int nh_len;             /* IPv6 Header + Extensions */
558
559                 nh_len = parse_ipv6hdr(skb, key);
560                 if (unlikely(nh_len < 0)) {
561                         if (nh_len == -EINVAL) {
562                                 skb->transport_header = skb->network_header;
563                                 error = 0;
564                         } else {
565                                 error = nh_len;
566                         }
567                         return error;
568                 }
569
570                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
571                         return 0;
572                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
573                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
574
575                 /* Transport layer. */
576                 if (key->ip.proto == NEXTHDR_TCP) {
577                         if (tcphdr_ok(skb)) {
578                                 struct tcphdr *tcp = tcp_hdr(skb);
579                                 key->ipv6.tp.src = tcp->source;
580                                 key->ipv6.tp.dst = tcp->dest;
581                                 key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
582                         }
583                 } else if (key->ip.proto == NEXTHDR_UDP) {
584                         if (udphdr_ok(skb)) {
585                                 struct udphdr *udp = udp_hdr(skb);
586                                 key->ipv6.tp.src = udp->source;
587                                 key->ipv6.tp.dst = udp->dest;
588                         }
589                 } else if (key->ip.proto == NEXTHDR_SCTP) {
590                         if (sctphdr_ok(skb)) {
591                                 struct sctphdr *sctp = sctp_hdr(skb);
592                                 key->ipv6.tp.src = sctp->source;
593                                 key->ipv6.tp.dst = sctp->dest;
594                         }
595                 } else if (key->ip.proto == NEXTHDR_ICMP) {
596                         if (icmp6hdr_ok(skb)) {
597                                 error = parse_icmpv6(skb, key, nh_len);
598                                 if (error)
599                                         return error;
600                         }
601                 }
602         }
603
604         return 0;
605 }