Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jesse/openvswitch
[sfrench/cifs-2.6.git] / net / openvswitch / 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/ip_tunnels.h>
46 #include <net/ipv6.h>
47 #include <net/ndisc.h>
48
49 u64 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
64 void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
65 {
66         struct flow_stats *stats;
67         __be16 tcp_flags = 0;
68
69         if (!flow->stats.is_percpu)
70                 stats = flow->stats.stat;
71         else
72                 stats = this_cpu_ptr(flow->stats.cpu_stats);
73
74         if ((flow->key.eth.type == htons(ETH_P_IP) ||
75              flow->key.eth.type == htons(ETH_P_IPV6)) &&
76             flow->key.ip.frag != OVS_FRAG_TYPE_LATER &&
77             flow->key.ip.proto == IPPROTO_TCP &&
78             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
79                 tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
80         }
81
82         spin_lock(&stats->lock);
83         stats->used = jiffies;
84         stats->packet_count++;
85         stats->byte_count += skb->len;
86         stats->tcp_flags |= tcp_flags;
87         spin_unlock(&stats->lock);
88 }
89
90 static void stats_read(struct flow_stats *stats,
91                        struct ovs_flow_stats *ovs_stats,
92                        unsigned long *used, __be16 *tcp_flags)
93 {
94         spin_lock(&stats->lock);
95         if (!*used || time_after(stats->used, *used))
96                 *used = stats->used;
97         *tcp_flags |= stats->tcp_flags;
98         ovs_stats->n_packets += stats->packet_count;
99         ovs_stats->n_bytes += stats->byte_count;
100         spin_unlock(&stats->lock);
101 }
102
103 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
104                         unsigned long *used, __be16 *tcp_flags)
105 {
106         int cpu, cur_cpu;
107
108         *used = 0;
109         *tcp_flags = 0;
110         memset(ovs_stats, 0, sizeof(*ovs_stats));
111
112         if (!flow->stats.is_percpu) {
113                 stats_read(flow->stats.stat, ovs_stats, used, tcp_flags);
114         } else {
115                 cur_cpu = get_cpu();
116                 for_each_possible_cpu(cpu) {
117                         struct flow_stats *stats;
118
119                         if (cpu == cur_cpu)
120                                 local_bh_disable();
121
122                         stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
123                         stats_read(stats, ovs_stats, used, tcp_flags);
124
125                         if (cpu == cur_cpu)
126                                 local_bh_enable();
127                 }
128                 put_cpu();
129         }
130 }
131
132 static void stats_reset(struct flow_stats *stats)
133 {
134         spin_lock(&stats->lock);
135         stats->used = 0;
136         stats->packet_count = 0;
137         stats->byte_count = 0;
138         stats->tcp_flags = 0;
139         spin_unlock(&stats->lock);
140 }
141
142 void ovs_flow_stats_clear(struct sw_flow *flow)
143 {
144         int cpu, cur_cpu;
145
146         if (!flow->stats.is_percpu) {
147                 stats_reset(flow->stats.stat);
148         } else {
149                 cur_cpu = get_cpu();
150
151                 for_each_possible_cpu(cpu) {
152
153                         if (cpu == cur_cpu)
154                                 local_bh_disable();
155
156                         stats_reset(per_cpu_ptr(flow->stats.cpu_stats, cpu));
157
158                         if (cpu == cur_cpu)
159                                 local_bh_enable();
160                 }
161                 put_cpu();
162         }
163 }
164
165 static int check_header(struct sk_buff *skb, int len)
166 {
167         if (unlikely(skb->len < len))
168                 return -EINVAL;
169         if (unlikely(!pskb_may_pull(skb, len)))
170                 return -ENOMEM;
171         return 0;
172 }
173
174 static bool arphdr_ok(struct sk_buff *skb)
175 {
176         return pskb_may_pull(skb, skb_network_offset(skb) +
177                                   sizeof(struct arp_eth_header));
178 }
179
180 static int check_iphdr(struct sk_buff *skb)
181 {
182         unsigned int nh_ofs = skb_network_offset(skb);
183         unsigned int ip_len;
184         int err;
185
186         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
187         if (unlikely(err))
188                 return err;
189
190         ip_len = ip_hdrlen(skb);
191         if (unlikely(ip_len < sizeof(struct iphdr) ||
192                      skb->len < nh_ofs + ip_len))
193                 return -EINVAL;
194
195         skb_set_transport_header(skb, nh_ofs + ip_len);
196         return 0;
197 }
198
199 static bool tcphdr_ok(struct sk_buff *skb)
200 {
201         int th_ofs = skb_transport_offset(skb);
202         int tcp_len;
203
204         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
205                 return false;
206
207         tcp_len = tcp_hdrlen(skb);
208         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
209                      skb->len < th_ofs + tcp_len))
210                 return false;
211
212         return true;
213 }
214
215 static bool udphdr_ok(struct sk_buff *skb)
216 {
217         return pskb_may_pull(skb, skb_transport_offset(skb) +
218                                   sizeof(struct udphdr));
219 }
220
221 static bool sctphdr_ok(struct sk_buff *skb)
222 {
223         return pskb_may_pull(skb, skb_transport_offset(skb) +
224                                   sizeof(struct sctphdr));
225 }
226
227 static bool icmphdr_ok(struct sk_buff *skb)
228 {
229         return pskb_may_pull(skb, skb_transport_offset(skb) +
230                                   sizeof(struct icmphdr));
231 }
232
233 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
234 {
235         unsigned int nh_ofs = skb_network_offset(skb);
236         unsigned int nh_len;
237         int payload_ofs;
238         struct ipv6hdr *nh;
239         uint8_t nexthdr;
240         __be16 frag_off;
241         int err;
242
243         err = check_header(skb, nh_ofs + sizeof(*nh));
244         if (unlikely(err))
245                 return err;
246
247         nh = ipv6_hdr(skb);
248         nexthdr = nh->nexthdr;
249         payload_ofs = (u8 *)(nh + 1) - skb->data;
250
251         key->ip.proto = NEXTHDR_NONE;
252         key->ip.tos = ipv6_get_dsfield(nh);
253         key->ip.ttl = nh->hop_limit;
254         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
255         key->ipv6.addr.src = nh->saddr;
256         key->ipv6.addr.dst = nh->daddr;
257
258         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
259         if (unlikely(payload_ofs < 0))
260                 return -EINVAL;
261
262         if (frag_off) {
263                 if (frag_off & htons(~0x7))
264                         key->ip.frag = OVS_FRAG_TYPE_LATER;
265                 else
266                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
267         }
268
269         nh_len = payload_ofs - nh_ofs;
270         skb_set_transport_header(skb, nh_ofs + nh_len);
271         key->ip.proto = nexthdr;
272         return nh_len;
273 }
274
275 static bool icmp6hdr_ok(struct sk_buff *skb)
276 {
277         return pskb_may_pull(skb, skb_transport_offset(skb) +
278                                   sizeof(struct icmp6hdr));
279 }
280
281 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
282 {
283         struct qtag_prefix {
284                 __be16 eth_type; /* ETH_P_8021Q */
285                 __be16 tci;
286         };
287         struct qtag_prefix *qp;
288
289         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
290                 return 0;
291
292         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
293                                          sizeof(__be16))))
294                 return -ENOMEM;
295
296         qp = (struct qtag_prefix *) skb->data;
297         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
298         __skb_pull(skb, sizeof(struct qtag_prefix));
299
300         return 0;
301 }
302
303 static __be16 parse_ethertype(struct sk_buff *skb)
304 {
305         struct llc_snap_hdr {
306                 u8  dsap;  /* Always 0xAA */
307                 u8  ssap;  /* Always 0xAA */
308                 u8  ctrl;
309                 u8  oui[3];
310                 __be16 ethertype;
311         };
312         struct llc_snap_hdr *llc;
313         __be16 proto;
314
315         proto = *(__be16 *) skb->data;
316         __skb_pull(skb, sizeof(__be16));
317
318         if (ntohs(proto) >= ETH_P_802_3_MIN)
319                 return proto;
320
321         if (skb->len < sizeof(struct llc_snap_hdr))
322                 return htons(ETH_P_802_2);
323
324         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
325                 return htons(0);
326
327         llc = (struct llc_snap_hdr *) skb->data;
328         if (llc->dsap != LLC_SAP_SNAP ||
329             llc->ssap != LLC_SAP_SNAP ||
330             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
331                 return htons(ETH_P_802_2);
332
333         __skb_pull(skb, sizeof(struct llc_snap_hdr));
334
335         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
336                 return llc->ethertype;
337
338         return htons(ETH_P_802_2);
339 }
340
341 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
342                         int nh_len)
343 {
344         struct icmp6hdr *icmp = icmp6_hdr(skb);
345
346         /* The ICMPv6 type and code fields use the 16-bit transport port
347          * fields, so we need to store them in 16-bit network byte order.
348          */
349         key->ipv6.tp.src = htons(icmp->icmp6_type);
350         key->ipv6.tp.dst = htons(icmp->icmp6_code);
351
352         if (icmp->icmp6_code == 0 &&
353             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
354              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
355                 int icmp_len = skb->len - skb_transport_offset(skb);
356                 struct nd_msg *nd;
357                 int offset;
358
359                 /* In order to process neighbor discovery options, we need the
360                  * entire packet.
361                  */
362                 if (unlikely(icmp_len < sizeof(*nd)))
363                         return 0;
364
365                 if (unlikely(skb_linearize(skb)))
366                         return -ENOMEM;
367
368                 nd = (struct nd_msg *)skb_transport_header(skb);
369                 key->ipv6.nd.target = nd->target;
370
371                 icmp_len -= sizeof(*nd);
372                 offset = 0;
373                 while (icmp_len >= 8) {
374                         struct nd_opt_hdr *nd_opt =
375                                  (struct nd_opt_hdr *)(nd->opt + offset);
376                         int opt_len = nd_opt->nd_opt_len * 8;
377
378                         if (unlikely(!opt_len || opt_len > icmp_len))
379                                 return 0;
380
381                         /* Store the link layer address if the appropriate
382                          * option is provided.  It is considered an error if
383                          * the same link layer option is specified twice.
384                          */
385                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
386                             && opt_len == 8) {
387                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
388                                         goto invalid;
389                                 memcpy(key->ipv6.nd.sll,
390                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
391                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
392                                    && opt_len == 8) {
393                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
394                                         goto invalid;
395                                 memcpy(key->ipv6.nd.tll,
396                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
397                         }
398
399                         icmp_len -= opt_len;
400                         offset += opt_len;
401                 }
402         }
403
404         return 0;
405
406 invalid:
407         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
408         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
409         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
410
411         return 0;
412 }
413
414 /**
415  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
416  * @skb: sk_buff that contains the frame, with skb->data pointing to the
417  * Ethernet header
418  * @in_port: port number on which @skb was received.
419  * @key: output flow key
420  *
421  * The caller must ensure that skb->len >= ETH_HLEN.
422  *
423  * Returns 0 if successful, otherwise a negative errno value.
424  *
425  * Initializes @skb header pointers as follows:
426  *
427  *    - skb->mac_header: the Ethernet header.
428  *
429  *    - skb->network_header: just past the Ethernet header, or just past the
430  *      VLAN header, to the first byte of the Ethernet payload.
431  *
432  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
433  *      on output, then just past the IP header, if one is present and
434  *      of a correct length, otherwise the same as skb->network_header.
435  *      For other key->eth.type values it is left untouched.
436  */
437 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
438 {
439         int error;
440         struct ethhdr *eth;
441
442         memset(key, 0, sizeof(*key));
443
444         key->phy.priority = skb->priority;
445         if (OVS_CB(skb)->tun_key)
446                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
447         key->phy.in_port = in_port;
448         key->phy.skb_mark = skb->mark;
449
450         skb_reset_mac_header(skb);
451
452         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
453          * header in the linear data area.
454          */
455         eth = eth_hdr(skb);
456         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
457         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
458
459         __skb_pull(skb, 2 * ETH_ALEN);
460         /* We are going to push all headers that we pull, so no need to
461          * update skb->csum here.
462          */
463
464         if (vlan_tx_tag_present(skb))
465                 key->eth.tci = htons(skb->vlan_tci);
466         else if (eth->h_proto == htons(ETH_P_8021Q))
467                 if (unlikely(parse_vlan(skb, key)))
468                         return -ENOMEM;
469
470         key->eth.type = parse_ethertype(skb);
471         if (unlikely(key->eth.type == htons(0)))
472                 return -ENOMEM;
473
474         skb_reset_network_header(skb);
475         __skb_push(skb, skb->data - skb_mac_header(skb));
476
477         /* Network layer. */
478         if (key->eth.type == htons(ETH_P_IP)) {
479                 struct iphdr *nh;
480                 __be16 offset;
481
482                 error = check_iphdr(skb);
483                 if (unlikely(error)) {
484                         if (error == -EINVAL) {
485                                 skb->transport_header = skb->network_header;
486                                 error = 0;
487                         }
488                         return error;
489                 }
490
491                 nh = ip_hdr(skb);
492                 key->ipv4.addr.src = nh->saddr;
493                 key->ipv4.addr.dst = nh->daddr;
494
495                 key->ip.proto = nh->protocol;
496                 key->ip.tos = nh->tos;
497                 key->ip.ttl = nh->ttl;
498
499                 offset = nh->frag_off & htons(IP_OFFSET);
500                 if (offset) {
501                         key->ip.frag = OVS_FRAG_TYPE_LATER;
502                         return 0;
503                 }
504                 if (nh->frag_off & htons(IP_MF) ||
505                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
506                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
507
508                 /* Transport layer. */
509                 if (key->ip.proto == IPPROTO_TCP) {
510                         if (tcphdr_ok(skb)) {
511                                 struct tcphdr *tcp = tcp_hdr(skb);
512                                 key->ipv4.tp.src = tcp->source;
513                                 key->ipv4.tp.dst = tcp->dest;
514                                 key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
515                         }
516                 } else if (key->ip.proto == IPPROTO_UDP) {
517                         if (udphdr_ok(skb)) {
518                                 struct udphdr *udp = udp_hdr(skb);
519                                 key->ipv4.tp.src = udp->source;
520                                 key->ipv4.tp.dst = udp->dest;
521                         }
522                 } else if (key->ip.proto == IPPROTO_SCTP) {
523                         if (sctphdr_ok(skb)) {
524                                 struct sctphdr *sctp = sctp_hdr(skb);
525                                 key->ipv4.tp.src = sctp->source;
526                                 key->ipv4.tp.dst = sctp->dest;
527                         }
528                 } else if (key->ip.proto == IPPROTO_ICMP) {
529                         if (icmphdr_ok(skb)) {
530                                 struct icmphdr *icmp = icmp_hdr(skb);
531                                 /* The ICMP type and code fields use the 16-bit
532                                  * transport port fields, so we need to store
533                                  * them in 16-bit network byte order. */
534                                 key->ipv4.tp.src = htons(icmp->type);
535                                 key->ipv4.tp.dst = htons(icmp->code);
536                         }
537                 }
538
539         } else if ((key->eth.type == htons(ETH_P_ARP) ||
540                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
541                 struct arp_eth_header *arp;
542
543                 arp = (struct arp_eth_header *)skb_network_header(skb);
544
545                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
546                                 && arp->ar_pro == htons(ETH_P_IP)
547                                 && arp->ar_hln == ETH_ALEN
548                                 && arp->ar_pln == 4) {
549
550                         /* We only match on the lower 8 bits of the opcode. */
551                         if (ntohs(arp->ar_op) <= 0xff)
552                                 key->ip.proto = ntohs(arp->ar_op);
553                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
554                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
555                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
556                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
557                 }
558         } else if (key->eth.type == htons(ETH_P_IPV6)) {
559                 int nh_len;             /* IPv6 Header + Extensions */
560
561                 nh_len = parse_ipv6hdr(skb, key);
562                 if (unlikely(nh_len < 0)) {
563                         if (nh_len == -EINVAL) {
564                                 skb->transport_header = skb->network_header;
565                                 error = 0;
566                         } else {
567                                 error = nh_len;
568                         }
569                         return error;
570                 }
571
572                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
573                         return 0;
574                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
575                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
576
577                 /* Transport layer. */
578                 if (key->ip.proto == NEXTHDR_TCP) {
579                         if (tcphdr_ok(skb)) {
580                                 struct tcphdr *tcp = tcp_hdr(skb);
581                                 key->ipv6.tp.src = tcp->source;
582                                 key->ipv6.tp.dst = tcp->dest;
583                                 key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
584                         }
585                 } else if (key->ip.proto == NEXTHDR_UDP) {
586                         if (udphdr_ok(skb)) {
587                                 struct udphdr *udp = udp_hdr(skb);
588                                 key->ipv6.tp.src = udp->source;
589                                 key->ipv6.tp.dst = udp->dest;
590                         }
591                 } else if (key->ip.proto == NEXTHDR_SCTP) {
592                         if (sctphdr_ok(skb)) {
593                                 struct sctphdr *sctp = sctp_hdr(skb);
594                                 key->ipv6.tp.src = sctp->source;
595                                 key->ipv6.tp.dst = sctp->dest;
596                         }
597                 } else if (key->ip.proto == NEXTHDR_ICMP) {
598                         if (icmp6hdr_ok(skb)) {
599                                 error = parse_icmpv6(skb, key, nh_len);
600                                 if (error)
601                                         return error;
602                         }
603                 }
604         }
605
606         return 0;
607 }