datapath: Read tcp flags only then the tranport header is present.
[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         if (!flow->stats.is_percpu)
71                 stats = flow->stats.stat;
72         else
73                 stats = this_cpu_ptr(flow->stats.cpu_stats);
74
75         if ((flow->key.eth.type == htons(ETH_P_IP) ||
76              flow->key.eth.type == htons(ETH_P_IPV6)) &&
77             flow->key.ip.frag != OVS_FRAG_TYPE_LATER &&
78             flow->key.ip.proto == IPPROTO_TCP &&
79             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
80                 tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
81         }
82
83         spin_lock(&stats->lock);
84         stats->used = jiffies;
85         stats->packet_count++;
86         stats->byte_count += skb->len;
87         stats->tcp_flags |= tcp_flags;
88         spin_unlock(&stats->lock);
89 }
90
91 static void stats_read(struct flow_stats *stats, bool lock_bh,
92                        struct ovs_flow_stats *ovs_stats,
93                        unsigned long *used, __be16 *tcp_flags)
94 {
95         if (lock_bh)
96                 spin_lock_bh(&stats->lock);
97         else
98                 spin_lock(&stats->lock);
99
100         if (time_after(stats->used, *used))
101                 *used = stats->used;
102         *tcp_flags |= stats->tcp_flags;
103         ovs_stats->n_packets += stats->packet_count;
104         ovs_stats->n_bytes += stats->byte_count;
105
106         if (lock_bh)
107                 spin_unlock_bh(&stats->lock);
108         else
109                 spin_unlock(&stats->lock);
110 }
111
112 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
113                         unsigned long *used, __be16 *tcp_flags)
114 {
115         int cpu, cur_cpu;
116
117         *used = 0;
118         *tcp_flags = 0;
119         memset(ovs_stats, 0, sizeof(*ovs_stats));
120
121         if (!flow->stats.is_percpu) {
122                 stats_read(flow->stats.stat, true, ovs_stats, used, tcp_flags);
123         } else {
124                 cur_cpu = get_cpu();
125
126                 for_each_possible_cpu(cpu) {
127                         struct flow_stats *stats;
128                         bool lock_bh;
129
130                         stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
131                         lock_bh = (cpu == cur_cpu);
132                         stats_read(stats, lock_bh, ovs_stats, used, tcp_flags);
133                 }
134                 put_cpu();
135         }
136 }
137
138 static void stats_reset(struct flow_stats *stats, bool lock_bh)
139 {
140         if (lock_bh)
141                 spin_lock_bh(&stats->lock);
142         else
143                 spin_lock(&stats->lock);
144
145         stats->used = 0;
146         stats->packet_count = 0;
147         stats->byte_count = 0;
148         stats->tcp_flags = 0;
149
150         if (lock_bh)
151                 spin_unlock_bh(&stats->lock);
152         else
153                 spin_unlock(&stats->lock);
154 }
155
156 void ovs_flow_stats_clear(struct sw_flow *flow)
157 {
158         int cpu, cur_cpu;
159
160         if (!flow->stats.is_percpu) {
161                 stats_reset(flow->stats.stat, true);
162         } else {
163                 cur_cpu = get_cpu();
164
165                 for_each_possible_cpu(cpu) {
166                         bool lock_bh;
167
168                         lock_bh = (cpu == cur_cpu);
169                         stats_reset(per_cpu_ptr(flow->stats.cpu_stats, cpu), lock_bh);
170                 }
171                 put_cpu();
172         }
173 }
174
175 static int check_header(struct sk_buff *skb, int len)
176 {
177         if (unlikely(skb->len < len))
178                 return -EINVAL;
179         if (unlikely(!pskb_may_pull(skb, len)))
180                 return -ENOMEM;
181         return 0;
182 }
183
184 static bool arphdr_ok(struct sk_buff *skb)
185 {
186         return pskb_may_pull(skb, skb_network_offset(skb) +
187                                   sizeof(struct arp_eth_header));
188 }
189
190 static int check_iphdr(struct sk_buff *skb)
191 {
192         unsigned int nh_ofs = skb_network_offset(skb);
193         unsigned int ip_len;
194         int err;
195
196         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
197         if (unlikely(err))
198                 return err;
199
200         ip_len = ip_hdrlen(skb);
201         if (unlikely(ip_len < sizeof(struct iphdr) ||
202                      skb->len < nh_ofs + ip_len))
203                 return -EINVAL;
204
205         skb_set_transport_header(skb, nh_ofs + ip_len);
206         return 0;
207 }
208
209 static bool tcphdr_ok(struct sk_buff *skb)
210 {
211         int th_ofs = skb_transport_offset(skb);
212         int tcp_len;
213
214         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
215                 return false;
216
217         tcp_len = tcp_hdrlen(skb);
218         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
219                      skb->len < th_ofs + tcp_len))
220                 return false;
221
222         return true;
223 }
224
225 static bool udphdr_ok(struct sk_buff *skb)
226 {
227         return pskb_may_pull(skb, skb_transport_offset(skb) +
228                                   sizeof(struct udphdr));
229 }
230
231 static bool sctphdr_ok(struct sk_buff *skb)
232 {
233         return pskb_may_pull(skb, skb_transport_offset(skb) +
234                                   sizeof(struct sctphdr));
235 }
236
237 static bool icmphdr_ok(struct sk_buff *skb)
238 {
239         return pskb_may_pull(skb, skb_transport_offset(skb) +
240                                   sizeof(struct icmphdr));
241 }
242
243 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
244 {
245         unsigned int nh_ofs = skb_network_offset(skb);
246         unsigned int nh_len;
247         int payload_ofs;
248         struct ipv6hdr *nh;
249         uint8_t nexthdr;
250         __be16 frag_off;
251         int err;
252
253         err = check_header(skb, nh_ofs + sizeof(*nh));
254         if (unlikely(err))
255                 return err;
256
257         nh = ipv6_hdr(skb);
258         nexthdr = nh->nexthdr;
259         payload_ofs = (u8 *)(nh + 1) - skb->data;
260
261         key->ip.proto = NEXTHDR_NONE;
262         key->ip.tos = ipv6_get_dsfield(nh);
263         key->ip.ttl = nh->hop_limit;
264         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
265         key->ipv6.addr.src = nh->saddr;
266         key->ipv6.addr.dst = nh->daddr;
267
268         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
269         if (unlikely(payload_ofs < 0))
270                 return -EINVAL;
271
272         if (frag_off) {
273                 if (frag_off & htons(~0x7))
274                         key->ip.frag = OVS_FRAG_TYPE_LATER;
275                 else
276                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
277         }
278
279         nh_len = payload_ofs - nh_ofs;
280         skb_set_transport_header(skb, nh_ofs + nh_len);
281         key->ip.proto = nexthdr;
282         return nh_len;
283 }
284
285 static bool icmp6hdr_ok(struct sk_buff *skb)
286 {
287         return pskb_may_pull(skb, skb_transport_offset(skb) +
288                                   sizeof(struct icmp6hdr));
289 }
290
291 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
292 {
293         struct qtag_prefix {
294                 __be16 eth_type; /* ETH_P_8021Q */
295                 __be16 tci;
296         };
297         struct qtag_prefix *qp;
298
299         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
300                 return 0;
301
302         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
303                                          sizeof(__be16))))
304                 return -ENOMEM;
305
306         qp = (struct qtag_prefix *) skb->data;
307         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
308         __skb_pull(skb, sizeof(struct qtag_prefix));
309
310         return 0;
311 }
312
313 static __be16 parse_ethertype(struct sk_buff *skb)
314 {
315         struct llc_snap_hdr {
316                 u8  dsap;  /* Always 0xAA */
317                 u8  ssap;  /* Always 0xAA */
318                 u8  ctrl;
319                 u8  oui[3];
320                 __be16 ethertype;
321         };
322         struct llc_snap_hdr *llc;
323         __be16 proto;
324
325         proto = *(__be16 *) skb->data;
326         __skb_pull(skb, sizeof(__be16));
327
328         if (ntohs(proto) >= ETH_P_802_3_MIN)
329                 return proto;
330
331         if (skb->len < sizeof(struct llc_snap_hdr))
332                 return htons(ETH_P_802_2);
333
334         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
335                 return htons(0);
336
337         llc = (struct llc_snap_hdr *) skb->data;
338         if (llc->dsap != LLC_SAP_SNAP ||
339             llc->ssap != LLC_SAP_SNAP ||
340             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
341                 return htons(ETH_P_802_2);
342
343         __skb_pull(skb, sizeof(struct llc_snap_hdr));
344
345         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
346                 return llc->ethertype;
347
348         return htons(ETH_P_802_2);
349 }
350
351 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
352                         int nh_len)
353 {
354         struct icmp6hdr *icmp = icmp6_hdr(skb);
355
356         /* The ICMPv6 type and code fields use the 16-bit transport port
357          * fields, so we need to store them in 16-bit network byte order.
358          */
359         key->ipv6.tp.src = htons(icmp->icmp6_type);
360         key->ipv6.tp.dst = htons(icmp->icmp6_code);
361
362         if (icmp->icmp6_code == 0 &&
363             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
364              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
365                 int icmp_len = skb->len - skb_transport_offset(skb);
366                 struct nd_msg *nd;
367                 int offset;
368
369                 /* In order to process neighbor discovery options, we need the
370                  * entire packet.
371                  */
372                 if (unlikely(icmp_len < sizeof(*nd)))
373                         return 0;
374
375                 if (unlikely(skb_linearize(skb)))
376                         return -ENOMEM;
377
378                 nd = (struct nd_msg *)skb_transport_header(skb);
379                 key->ipv6.nd.target = nd->target;
380
381                 icmp_len -= sizeof(*nd);
382                 offset = 0;
383                 while (icmp_len >= 8) {
384                         struct nd_opt_hdr *nd_opt =
385                                  (struct nd_opt_hdr *)(nd->opt + offset);
386                         int opt_len = nd_opt->nd_opt_len * 8;
387
388                         if (unlikely(!opt_len || opt_len > icmp_len))
389                                 return 0;
390
391                         /* Store the link layer address if the appropriate
392                          * option is provided.  It is considered an error if
393                          * the same link layer option is specified twice.
394                          */
395                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
396                             && opt_len == 8) {
397                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
398                                         goto invalid;
399                                 memcpy(key->ipv6.nd.sll,
400                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
401                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
402                                    && opt_len == 8) {
403                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
404                                         goto invalid;
405                                 memcpy(key->ipv6.nd.tll,
406                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
407                         }
408
409                         icmp_len -= opt_len;
410                         offset += opt_len;
411                 }
412         }
413
414         return 0;
415
416 invalid:
417         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
418         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
419         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
420
421         return 0;
422 }
423
424 /**
425  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
426  * @skb: sk_buff that contains the frame, with skb->data pointing to the
427  * Ethernet header
428  * @in_port: port number on which @skb was received.
429  * @key: output flow key
430  *
431  * The caller must ensure that skb->len >= ETH_HLEN.
432  *
433  * Returns 0 if successful, otherwise a negative errno value.
434  *
435  * Initializes @skb header pointers as follows:
436  *
437  *    - skb->mac_header: the Ethernet header.
438  *
439  *    - skb->network_header: just past the Ethernet header, or just past the
440  *      VLAN header, to the first byte of the Ethernet payload.
441  *
442  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
443  *      on output, then just past the IP header, if one is present and
444  *      of a correct length, otherwise the same as skb->network_header.
445  *      For other key->eth.type values it is left untouched.
446  */
447 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
448 {
449         int error;
450         struct ethhdr *eth;
451
452         memset(key, 0, sizeof(*key));
453
454         key->phy.priority = skb->priority;
455         if (OVS_CB(skb)->tun_key)
456                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
457         key->phy.in_port = in_port;
458         key->phy.skb_mark = skb->mark;
459
460         skb_reset_mac_header(skb);
461
462         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
463          * header in the linear data area.
464          */
465         eth = eth_hdr(skb);
466         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
467         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
468
469         __skb_pull(skb, 2 * ETH_ALEN);
470         /* We are going to push all headers that we pull, so no need to
471          * update skb->csum here. */
472
473         if (vlan_tx_tag_present(skb))
474                 key->eth.tci = htons(vlan_get_tci(skb));
475         else if (eth->h_proto == htons(ETH_P_8021Q))
476                 if (unlikely(parse_vlan(skb, key)))
477                         return -ENOMEM;
478
479         key->eth.type = parse_ethertype(skb);
480         if (unlikely(key->eth.type == htons(0)))
481                 return -ENOMEM;
482
483         skb_reset_network_header(skb);
484         __skb_push(skb, skb->data - skb_mac_header(skb));
485
486         /* Network layer. */
487         if (key->eth.type == htons(ETH_P_IP)) {
488                 struct iphdr *nh;
489                 __be16 offset;
490
491                 error = check_iphdr(skb);
492                 if (unlikely(error)) {
493                         if (error == -EINVAL) {
494                                 skb->transport_header = skb->network_header;
495                                 error = 0;
496                         }
497                         return error;
498                 }
499
500                 nh = ip_hdr(skb);
501                 key->ipv4.addr.src = nh->saddr;
502                 key->ipv4.addr.dst = nh->daddr;
503
504                 key->ip.proto = nh->protocol;
505                 key->ip.tos = nh->tos;
506                 key->ip.ttl = nh->ttl;
507
508                 offset = nh->frag_off & htons(IP_OFFSET);
509                 if (offset) {
510                         key->ip.frag = OVS_FRAG_TYPE_LATER;
511                         return 0;
512                 }
513                 if (nh->frag_off & htons(IP_MF) ||
514                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
515                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
516
517                 /* Transport layer. */
518                 if (key->ip.proto == IPPROTO_TCP) {
519                         if (tcphdr_ok(skb)) {
520                                 struct tcphdr *tcp = tcp_hdr(skb);
521                                 key->ipv4.tp.src = tcp->source;
522                                 key->ipv4.tp.dst = tcp->dest;
523                                 key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
524                         }
525                 } else if (key->ip.proto == IPPROTO_UDP) {
526                         if (udphdr_ok(skb)) {
527                                 struct udphdr *udp = udp_hdr(skb);
528                                 key->ipv4.tp.src = udp->source;
529                                 key->ipv4.tp.dst = udp->dest;
530                         }
531                 } else if (key->ip.proto == IPPROTO_SCTP) {
532                         if (sctphdr_ok(skb)) {
533                                 struct sctphdr *sctp = sctp_hdr(skb);
534                                 key->ipv4.tp.src = sctp->source;
535                                 key->ipv4.tp.dst = sctp->dest;
536                         }
537                 } else if (key->ip.proto == IPPROTO_ICMP) {
538                         if (icmphdr_ok(skb)) {
539                                 struct icmphdr *icmp = icmp_hdr(skb);
540                                 /* The ICMP type and code fields use the 16-bit
541                                  * transport port fields, so we need to store
542                                  * them in 16-bit network byte order. */
543                                 key->ipv4.tp.src = htons(icmp->type);
544                                 key->ipv4.tp.dst = htons(icmp->code);
545                         }
546                 }
547
548         } else if ((key->eth.type == htons(ETH_P_ARP) ||
549                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
550                 struct arp_eth_header *arp;
551
552                 arp = (struct arp_eth_header *)skb_network_header(skb);
553
554                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
555                                 && arp->ar_pro == htons(ETH_P_IP)
556                                 && arp->ar_hln == ETH_ALEN
557                                 && arp->ar_pln == 4) {
558
559                         /* We only match on the lower 8 bits of the opcode. */
560                         if (ntohs(arp->ar_op) <= 0xff)
561                                 key->ip.proto = ntohs(arp->ar_op);
562                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
563                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
564                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
565                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
566                 }
567         } else if (key->eth.type == htons(ETH_P_IPV6)) {
568                 int nh_len;             /* IPv6 Header + Extensions */
569
570                 nh_len = parse_ipv6hdr(skb, key);
571                 if (unlikely(nh_len < 0)) {
572                         if (nh_len == -EINVAL) {
573                                 skb->transport_header = skb->network_header;
574                                 error = 0;
575                         } else {
576                                 error = nh_len;
577                         }
578                         return error;
579                 }
580
581                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
582                         return 0;
583                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
584                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
585
586                 /* Transport layer. */
587                 if (key->ip.proto == NEXTHDR_TCP) {
588                         if (tcphdr_ok(skb)) {
589                                 struct tcphdr *tcp = tcp_hdr(skb);
590                                 key->ipv6.tp.src = tcp->source;
591                                 key->ipv6.tp.dst = tcp->dest;
592                                 key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
593                         }
594                 } else if (key->ip.proto == NEXTHDR_UDP) {
595                         if (udphdr_ok(skb)) {
596                                 struct udphdr *udp = udp_hdr(skb);
597                                 key->ipv6.tp.src = udp->source;
598                                 key->ipv6.tp.dst = udp->dest;
599                         }
600                 } else if (key->ip.proto == NEXTHDR_SCTP) {
601                         if (sctphdr_ok(skb)) {
602                                 struct sctphdr *sctp = sctp_hdr(skb);
603                                 key->ipv6.tp.src = sctp->source;
604                                 key->ipv6.tp.dst = sctp->dest;
605                         }
606                 } else if (key->ip.proto == NEXTHDR_ICMP) {
607                         if (icmp6hdr_ok(skb)) {
608                                 error = parse_icmpv6(skb, key, nh_len);
609                                 if (error)
610                                         return error;
611                         }
612                 }
613         }
614
615         return 0;
616 }