26af257299ecf830fa37d4083355545b0f2620ac
[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         int node = numa_node_id();
70
71         stats = rcu_dereference(flow->stats[node]);
72
73         if ((flow->key.eth.type == htons(ETH_P_IP) ||
74              flow->key.eth.type == htons(ETH_P_IPV6)) &&
75             flow->key.ip.frag != OVS_FRAG_TYPE_LATER &&
76             flow->key.ip.proto == IPPROTO_TCP &&
77             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
78                 tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
79         }
80
81         /* Check if already have node-specific stats. */
82         if (likely(stats)) {
83                 spin_lock(&stats->lock);
84                 /* Mark if we write on the pre-allocated stats. */
85                 if (node == 0 && unlikely(flow->stats_last_writer != node))
86                         flow->stats_last_writer = node;
87         } else {
88                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
89                 spin_lock(&stats->lock);
90
91                 /* If the current NUMA-node is the only writer on the
92                  * pre-allocated stats keep using them.
93                  */
94                 if (unlikely(flow->stats_last_writer != node)) {
95                         /* A previous locker may have already allocated the
96                          * stats, so we need to check again.  If node-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 != NUMA_NO_NODE)
101                             && likely(!rcu_dereference(flow->stats[node]))) {
102                                 /* Try to allocate node-specific stats. */
103                                 struct flow_stats *new_stats;
104
105                                 new_stats =
106                                         kmem_cache_alloc_node(flow_stats_cache,
107                                                               GFP_THISNODE |
108                                                               __GFP_NOMEMALLOC,
109                                                               node);
110                                 if (likely(new_stats)) {
111                                         new_stats->used = jiffies;
112                                         new_stats->packet_count = 1;
113                                         new_stats->byte_count = skb->len;
114                                         new_stats->tcp_flags = tcp_flags;
115                                         spin_lock_init(&new_stats->lock);
116
117                                         rcu_assign_pointer(flow->stats[node],
118                                                            new_stats);
119                                         goto unlock;
120                                 }
121                         }
122                         flow->stats_last_writer = node;
123                 }
124         }
125
126         stats->used = jiffies;
127         stats->packet_count++;
128         stats->byte_count += skb->len;
129         stats->tcp_flags |= tcp_flags;
130 unlock:
131         spin_unlock(&stats->lock);
132 }
133
134 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
135                         unsigned long *used, __be16 *tcp_flags)
136 {
137         int node;
138
139         *used = 0;
140         *tcp_flags = 0;
141         memset(ovs_stats, 0, sizeof(*ovs_stats));
142
143         for_each_node(node) {
144                 struct flow_stats *stats = rcu_dereference(flow->stats[node]);
145
146                 if (stats) {
147                         /* Local CPU may write on non-local stats, so we must
148                          * block bottom-halves here.
149                          */
150                         spin_lock_bh(&stats->lock);
151                         if (time_after(stats->used, *used))
152                                 *used = stats->used;
153                         *tcp_flags |= stats->tcp_flags;
154                         ovs_stats->n_packets += stats->packet_count;
155                         ovs_stats->n_bytes += stats->byte_count;
156                         spin_unlock_bh(&stats->lock);
157                 }
158         }
159 }
160
161 void ovs_flow_stats_clear(struct sw_flow *flow)
162 {
163         int node;
164
165         for_each_node(node) {
166                 struct flow_stats *stats = rcu_dereference(flow->stats[node]);
167
168                 if (stats) {
169                         spin_lock_bh(&stats->lock);
170                         stats->used = 0;
171                         stats->packet_count = 0;
172                         stats->byte_count = 0;
173                         stats->tcp_flags = 0;
174                         spin_unlock_bh(&stats->lock);
175                 }
176         }
177 }
178
179 static int check_header(struct sk_buff *skb, int len)
180 {
181         if (unlikely(skb->len < len))
182                 return -EINVAL;
183         if (unlikely(!pskb_may_pull(skb, len)))
184                 return -ENOMEM;
185         return 0;
186 }
187
188 static bool arphdr_ok(struct sk_buff *skb)
189 {
190         return pskb_may_pull(skb, skb_network_offset(skb) +
191                                   sizeof(struct arp_eth_header));
192 }
193
194 static int check_iphdr(struct sk_buff *skb)
195 {
196         unsigned int nh_ofs = skb_network_offset(skb);
197         unsigned int ip_len;
198         int err;
199
200         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
201         if (unlikely(err))
202                 return err;
203
204         ip_len = ip_hdrlen(skb);
205         if (unlikely(ip_len < sizeof(struct iphdr) ||
206                      skb->len < nh_ofs + ip_len))
207                 return -EINVAL;
208
209         skb_set_transport_header(skb, nh_ofs + ip_len);
210         return 0;
211 }
212
213 static bool tcphdr_ok(struct sk_buff *skb)
214 {
215         int th_ofs = skb_transport_offset(skb);
216         int tcp_len;
217
218         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
219                 return false;
220
221         tcp_len = tcp_hdrlen(skb);
222         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
223                      skb->len < th_ofs + tcp_len))
224                 return false;
225
226         return true;
227 }
228
229 static bool udphdr_ok(struct sk_buff *skb)
230 {
231         return pskb_may_pull(skb, skb_transport_offset(skb) +
232                                   sizeof(struct udphdr));
233 }
234
235 static bool sctphdr_ok(struct sk_buff *skb)
236 {
237         return pskb_may_pull(skb, skb_transport_offset(skb) +
238                                   sizeof(struct sctphdr));
239 }
240
241 static bool icmphdr_ok(struct sk_buff *skb)
242 {
243         return pskb_may_pull(skb, skb_transport_offset(skb) +
244                                   sizeof(struct icmphdr));
245 }
246
247 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
248 {
249         unsigned int nh_ofs = skb_network_offset(skb);
250         unsigned int nh_len;
251         int payload_ofs;
252         struct ipv6hdr *nh;
253         uint8_t nexthdr;
254         __be16 frag_off;
255         int err;
256
257         err = check_header(skb, nh_ofs + sizeof(*nh));
258         if (unlikely(err))
259                 return err;
260
261         nh = ipv6_hdr(skb);
262         nexthdr = nh->nexthdr;
263         payload_ofs = (u8 *)(nh + 1) - skb->data;
264
265         key->ip.proto = NEXTHDR_NONE;
266         key->ip.tos = ipv6_get_dsfield(nh);
267         key->ip.ttl = nh->hop_limit;
268         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
269         key->ipv6.addr.src = nh->saddr;
270         key->ipv6.addr.dst = nh->daddr;
271
272         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
273         if (unlikely(payload_ofs < 0))
274                 return -EINVAL;
275
276         if (frag_off) {
277                 if (frag_off & htons(~0x7))
278                         key->ip.frag = OVS_FRAG_TYPE_LATER;
279                 else
280                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
281         }
282
283         nh_len = payload_ofs - nh_ofs;
284         skb_set_transport_header(skb, nh_ofs + nh_len);
285         key->ip.proto = nexthdr;
286         return nh_len;
287 }
288
289 static bool icmp6hdr_ok(struct sk_buff *skb)
290 {
291         return pskb_may_pull(skb, skb_transport_offset(skb) +
292                                   sizeof(struct icmp6hdr));
293 }
294
295 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
296 {
297         struct qtag_prefix {
298                 __be16 eth_type; /* ETH_P_8021Q */
299                 __be16 tci;
300         };
301         struct qtag_prefix *qp;
302
303         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
304                 return 0;
305
306         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
307                                          sizeof(__be16))))
308                 return -ENOMEM;
309
310         qp = (struct qtag_prefix *) skb->data;
311         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
312         __skb_pull(skb, sizeof(struct qtag_prefix));
313
314         return 0;
315 }
316
317 static __be16 parse_ethertype(struct sk_buff *skb)
318 {
319         struct llc_snap_hdr {
320                 u8  dsap;  /* Always 0xAA */
321                 u8  ssap;  /* Always 0xAA */
322                 u8  ctrl;
323                 u8  oui[3];
324                 __be16 ethertype;
325         };
326         struct llc_snap_hdr *llc;
327         __be16 proto;
328
329         proto = *(__be16 *) skb->data;
330         __skb_pull(skb, sizeof(__be16));
331
332         if (ntohs(proto) >= ETH_P_802_3_MIN)
333                 return proto;
334
335         if (skb->len < sizeof(struct llc_snap_hdr))
336                 return htons(ETH_P_802_2);
337
338         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
339                 return htons(0);
340
341         llc = (struct llc_snap_hdr *) skb->data;
342         if (llc->dsap != LLC_SAP_SNAP ||
343             llc->ssap != LLC_SAP_SNAP ||
344             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
345                 return htons(ETH_P_802_2);
346
347         __skb_pull(skb, sizeof(struct llc_snap_hdr));
348
349         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
350                 return llc->ethertype;
351
352         return htons(ETH_P_802_2);
353 }
354
355 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
356                         int nh_len)
357 {
358         struct icmp6hdr *icmp = icmp6_hdr(skb);
359
360         /* The ICMPv6 type and code fields use the 16-bit transport port
361          * fields, so we need to store them in 16-bit network byte order.
362          */
363         key->ipv6.tp.src = htons(icmp->icmp6_type);
364         key->ipv6.tp.dst = htons(icmp->icmp6_code);
365
366         if (icmp->icmp6_code == 0 &&
367             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
368              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
369                 int icmp_len = skb->len - skb_transport_offset(skb);
370                 struct nd_msg *nd;
371                 int offset;
372
373                 /* In order to process neighbor discovery options, we need the
374                  * entire packet.
375                  */
376                 if (unlikely(icmp_len < sizeof(*nd)))
377                         return 0;
378
379                 if (unlikely(skb_linearize(skb)))
380                         return -ENOMEM;
381
382                 nd = (struct nd_msg *)skb_transport_header(skb);
383                 key->ipv6.nd.target = nd->target;
384
385                 icmp_len -= sizeof(*nd);
386                 offset = 0;
387                 while (icmp_len >= 8) {
388                         struct nd_opt_hdr *nd_opt =
389                                  (struct nd_opt_hdr *)(nd->opt + offset);
390                         int opt_len = nd_opt->nd_opt_len * 8;
391
392                         if (unlikely(!opt_len || opt_len > icmp_len))
393                                 return 0;
394
395                         /* Store the link layer address if the appropriate
396                          * option is provided.  It is considered an error if
397                          * the same link layer option is specified twice.
398                          */
399                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
400                             && opt_len == 8) {
401                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
402                                         goto invalid;
403                                 memcpy(key->ipv6.nd.sll,
404                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
405                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
406                                    && opt_len == 8) {
407                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
408                                         goto invalid;
409                                 memcpy(key->ipv6.nd.tll,
410                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
411                         }
412
413                         icmp_len -= opt_len;
414                         offset += opt_len;
415                 }
416         }
417
418         return 0;
419
420 invalid:
421         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
422         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
423         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
424
425         return 0;
426 }
427
428 /**
429  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
430  * @skb: sk_buff that contains the frame, with skb->data pointing to the
431  * Ethernet header
432  * @in_port: port number on which @skb was received.
433  * @key: output flow key
434  *
435  * The caller must ensure that skb->len >= ETH_HLEN.
436  *
437  * Returns 0 if successful, otherwise a negative errno value.
438  *
439  * Initializes @skb header pointers as follows:
440  *
441  *    - skb->mac_header: the Ethernet header.
442  *
443  *    - skb->network_header: just past the Ethernet header, or just past the
444  *      VLAN header, to the first byte of the Ethernet payload.
445  *
446  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
447  *      on output, then just past the IP header, if one is present and
448  *      of a correct length, otherwise the same as skb->network_header.
449  *      For other key->eth.type values it is left untouched.
450  */
451 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
452 {
453         int error;
454         struct ethhdr *eth;
455
456         memset(key, 0, sizeof(*key));
457
458         key->phy.priority = skb->priority;
459         if (OVS_CB(skb)->tun_key)
460                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
461         key->phy.in_port = in_port;
462         key->phy.skb_mark = skb->mark;
463
464         skb_reset_mac_header(skb);
465
466         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
467          * header in the linear data area.
468          */
469         eth = eth_hdr(skb);
470         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
471         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
472
473         __skb_pull(skb, 2 * ETH_ALEN);
474         /* We are going to push all headers that we pull, so no need to
475          * update skb->csum here. */
476
477         if (vlan_tx_tag_present(skb))
478                 key->eth.tci = htons(vlan_get_tci(skb));
479         else if (eth->h_proto == htons(ETH_P_8021Q))
480                 if (unlikely(parse_vlan(skb, key)))
481                         return -ENOMEM;
482
483         key->eth.type = parse_ethertype(skb);
484         if (unlikely(key->eth.type == htons(0)))
485                 return -ENOMEM;
486
487         skb_reset_network_header(skb);
488         __skb_push(skb, skb->data - skb_mac_header(skb));
489
490         /* Network layer. */
491         if (key->eth.type == htons(ETH_P_IP)) {
492                 struct iphdr *nh;
493                 __be16 offset;
494
495                 error = check_iphdr(skb);
496                 if (unlikely(error)) {
497                         if (error == -EINVAL) {
498                                 skb->transport_header = skb->network_header;
499                                 error = 0;
500                         }
501                         return error;
502                 }
503
504                 nh = ip_hdr(skb);
505                 key->ipv4.addr.src = nh->saddr;
506                 key->ipv4.addr.dst = nh->daddr;
507
508                 key->ip.proto = nh->protocol;
509                 key->ip.tos = nh->tos;
510                 key->ip.ttl = nh->ttl;
511
512                 offset = nh->frag_off & htons(IP_OFFSET);
513                 if (offset) {
514                         key->ip.frag = OVS_FRAG_TYPE_LATER;
515                         return 0;
516                 }
517                 if (nh->frag_off & htons(IP_MF) ||
518                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
519                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
520
521                 /* Transport layer. */
522                 if (key->ip.proto == IPPROTO_TCP) {
523                         if (tcphdr_ok(skb)) {
524                                 struct tcphdr *tcp = tcp_hdr(skb);
525                                 key->ipv4.tp.src = tcp->source;
526                                 key->ipv4.tp.dst = tcp->dest;
527                                 key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp);
528                         }
529                 } else if (key->ip.proto == IPPROTO_UDP) {
530                         if (udphdr_ok(skb)) {
531                                 struct udphdr *udp = udp_hdr(skb);
532                                 key->ipv4.tp.src = udp->source;
533                                 key->ipv4.tp.dst = udp->dest;
534                         }
535                 } else if (key->ip.proto == IPPROTO_SCTP) {
536                         if (sctphdr_ok(skb)) {
537                                 struct sctphdr *sctp = sctp_hdr(skb);
538                                 key->ipv4.tp.src = sctp->source;
539                                 key->ipv4.tp.dst = sctp->dest;
540                         }
541                 } else if (key->ip.proto == IPPROTO_ICMP) {
542                         if (icmphdr_ok(skb)) {
543                                 struct icmphdr *icmp = icmp_hdr(skb);
544                                 /* The ICMP type and code fields use the 16-bit
545                                  * transport port fields, so we need to store
546                                  * them in 16-bit network byte order. */
547                                 key->ipv4.tp.src = htons(icmp->type);
548                                 key->ipv4.tp.dst = htons(icmp->code);
549                         }
550                 }
551
552         } else if ((key->eth.type == htons(ETH_P_ARP) ||
553                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
554                 struct arp_eth_header *arp;
555
556                 arp = (struct arp_eth_header *)skb_network_header(skb);
557
558                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
559                                 && arp->ar_pro == htons(ETH_P_IP)
560                                 && arp->ar_hln == ETH_ALEN
561                                 && arp->ar_pln == 4) {
562
563                         /* We only match on the lower 8 bits of the opcode. */
564                         if (ntohs(arp->ar_op) <= 0xff)
565                                 key->ip.proto = ntohs(arp->ar_op);
566                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
567                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
568                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
569                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
570                 }
571         } else if (key->eth.type == htons(ETH_P_IPV6)) {
572                 int nh_len;             /* IPv6 Header + Extensions */
573
574                 nh_len = parse_ipv6hdr(skb, key);
575                 if (unlikely(nh_len < 0)) {
576                         if (nh_len == -EINVAL) {
577                                 skb->transport_header = skb->network_header;
578                                 error = 0;
579                         } else {
580                                 error = nh_len;
581                         }
582                         return error;
583                 }
584
585                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
586                         return 0;
587                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
588                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
589
590                 /* Transport layer. */
591                 if (key->ip.proto == NEXTHDR_TCP) {
592                         if (tcphdr_ok(skb)) {
593                                 struct tcphdr *tcp = tcp_hdr(skb);
594                                 key->ipv6.tp.src = tcp->source;
595                                 key->ipv6.tp.dst = tcp->dest;
596                                 key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp);
597                         }
598                 } else if (key->ip.proto == NEXTHDR_UDP) {
599                         if (udphdr_ok(skb)) {
600                                 struct udphdr *udp = udp_hdr(skb);
601                                 key->ipv6.tp.src = udp->source;
602                                 key->ipv6.tp.dst = udp->dest;
603                         }
604                 } else if (key->ip.proto == NEXTHDR_SCTP) {
605                         if (sctphdr_ok(skb)) {
606                                 struct sctphdr *sctp = sctp_hdr(skb);
607                                 key->ipv6.tp.src = sctp->source;
608                                 key->ipv6.tp.dst = sctp->dest;
609                         }
610                 } else if (key->ip.proto == NEXTHDR_ICMP) {
611                         if (icmp6hdr_ok(skb)) {
612                                 error = parse_icmpv6(skb, key, nh_len);
613                                 if (error)
614                                         return error;
615                         }
616                 }
617         }
618
619         return 0;
620 }