Setting tag sliver-openvswitch-2.2.90-1
[sliver-openvswitch.git] / datapath / flow.c
1 /*
2  * Copyright (c) 2007-2014 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, __be16 tcp_flags,
66                            struct sk_buff *skb)
67 {
68         struct flow_stats *stats;
69         int node = numa_node_id();
70
71         stats = rcu_dereference(flow->stats[node]);
72
73         /* Check if already have node-specific stats. */
74         if (likely(stats)) {
75                 spin_lock(&stats->lock);
76                 /* Mark if we write on the pre-allocated stats. */
77                 if (node == 0 && unlikely(flow->stats_last_writer != node))
78                         flow->stats_last_writer = node;
79         } else {
80                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
81                 spin_lock(&stats->lock);
82
83                 /* If the current NUMA-node is the only writer on the
84                  * pre-allocated stats keep using them.
85                  */
86                 if (unlikely(flow->stats_last_writer != node)) {
87                         /* A previous locker may have already allocated the
88                          * stats, so we need to check again.  If node-specific
89                          * stats were already allocated, we update the pre-
90                          * allocated stats as we have already locked them.
91                          */
92                         if (likely(flow->stats_last_writer != NUMA_NO_NODE)
93                             && likely(!rcu_dereference(flow->stats[node]))) {
94                                 /* Try to allocate node-specific stats. */
95                                 struct flow_stats *new_stats;
96
97                                 new_stats =
98                                         kmem_cache_alloc_node(flow_stats_cache,
99                                                               GFP_THISNODE |
100                                                               __GFP_NOMEMALLOC,
101                                                               node);
102                                 if (likely(new_stats)) {
103                                         new_stats->used = jiffies;
104                                         new_stats->packet_count = 1;
105                                         new_stats->byte_count = skb->len;
106                                         new_stats->tcp_flags = tcp_flags;
107                                         spin_lock_init(&new_stats->lock);
108
109                                         rcu_assign_pointer(flow->stats[node],
110                                                            new_stats);
111                                         goto unlock;
112                                 }
113                         }
114                         flow->stats_last_writer = node;
115                 }
116         }
117
118         stats->used = jiffies;
119         stats->packet_count++;
120         stats->byte_count += skb->len;
121         stats->tcp_flags |= tcp_flags;
122 unlock:
123         spin_unlock(&stats->lock);
124 }
125
126 /* Must be called with rcu_read_lock or ovs_mutex. */
127 void ovs_flow_stats_get(const struct sw_flow *flow,
128                         struct ovs_flow_stats *ovs_stats,
129                         unsigned long *used, __be16 *tcp_flags)
130 {
131         int node;
132
133         *used = 0;
134         *tcp_flags = 0;
135         memset(ovs_stats, 0, sizeof(*ovs_stats));
136
137         for_each_node(node) {
138                 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
139
140                 if (stats) {
141                         /* Local CPU may write on non-local stats, so we must
142                          * block bottom-halves here.
143                          */
144                         spin_lock_bh(&stats->lock);
145                         if (!*used || time_after(stats->used, *used))
146                                 *used = stats->used;
147                         *tcp_flags |= stats->tcp_flags;
148                         ovs_stats->n_packets += stats->packet_count;
149                         ovs_stats->n_bytes += stats->byte_count;
150                         spin_unlock_bh(&stats->lock);
151                 }
152         }
153 }
154
155 /* Called with ovs_mutex. */
156 void ovs_flow_stats_clear(struct sw_flow *flow)
157 {
158         int node;
159
160         for_each_node(node) {
161                 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
162
163                 if (stats) {
164                         spin_lock_bh(&stats->lock);
165                         stats->used = 0;
166                         stats->packet_count = 0;
167                         stats->byte_count = 0;
168                         stats->tcp_flags = 0;
169                         spin_unlock_bh(&stats->lock);
170                 }
171         }
172 }
173
174 static int check_header(struct sk_buff *skb, int len)
175 {
176         if (unlikely(skb->len < len))
177                 return -EINVAL;
178         if (unlikely(!pskb_may_pull(skb, len)))
179                 return -ENOMEM;
180         return 0;
181 }
182
183 static bool arphdr_ok(struct sk_buff *skb)
184 {
185         return pskb_may_pull(skb, skb_network_offset(skb) +
186                                   sizeof(struct arp_eth_header));
187 }
188
189 static int check_iphdr(struct sk_buff *skb)
190 {
191         unsigned int nh_ofs = skb_network_offset(skb);
192         unsigned int ip_len;
193         int err;
194
195         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
196         if (unlikely(err))
197                 return err;
198
199         ip_len = ip_hdrlen(skb);
200         if (unlikely(ip_len < sizeof(struct iphdr) ||
201                      skb->len < nh_ofs + ip_len))
202                 return -EINVAL;
203
204         skb_set_transport_header(skb, nh_ofs + ip_len);
205         return 0;
206 }
207
208 static bool tcphdr_ok(struct sk_buff *skb)
209 {
210         int th_ofs = skb_transport_offset(skb);
211         int tcp_len;
212
213         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
214                 return false;
215
216         tcp_len = tcp_hdrlen(skb);
217         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
218                      skb->len < th_ofs + tcp_len))
219                 return false;
220
221         return true;
222 }
223
224 static bool udphdr_ok(struct sk_buff *skb)
225 {
226         return pskb_may_pull(skb, skb_transport_offset(skb) +
227                                   sizeof(struct udphdr));
228 }
229
230 static bool sctphdr_ok(struct sk_buff *skb)
231 {
232         return pskb_may_pull(skb, skb_transport_offset(skb) +
233                                   sizeof(struct sctphdr));
234 }
235
236 static bool icmphdr_ok(struct sk_buff *skb)
237 {
238         return pskb_may_pull(skb, skb_transport_offset(skb) +
239                                   sizeof(struct icmphdr));
240 }
241
242 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
243 {
244         unsigned int nh_ofs = skb_network_offset(skb);
245         unsigned int nh_len;
246         int payload_ofs;
247         struct ipv6hdr *nh;
248         uint8_t nexthdr;
249         __be16 frag_off;
250         int err;
251
252         err = check_header(skb, nh_ofs + sizeof(*nh));
253         if (unlikely(err))
254                 return err;
255
256         nh = ipv6_hdr(skb);
257         nexthdr = nh->nexthdr;
258         payload_ofs = (u8 *)(nh + 1) - skb->data;
259
260         key->ip.proto = NEXTHDR_NONE;
261         key->ip.tos = ipv6_get_dsfield(nh);
262         key->ip.ttl = nh->hop_limit;
263         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
264         key->ipv6.addr.src = nh->saddr;
265         key->ipv6.addr.dst = nh->daddr;
266
267         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
268         if (unlikely(payload_ofs < 0))
269                 return -EINVAL;
270
271         if (frag_off) {
272                 if (frag_off & htons(~0x7))
273                         key->ip.frag = OVS_FRAG_TYPE_LATER;
274                 else
275                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
276         }
277
278         nh_len = payload_ofs - nh_ofs;
279         skb_set_transport_header(skb, nh_ofs + nh_len);
280         key->ip.proto = nexthdr;
281         return nh_len;
282 }
283
284 static bool icmp6hdr_ok(struct sk_buff *skb)
285 {
286         return pskb_may_pull(skb, skb_transport_offset(skb) +
287                                   sizeof(struct icmp6hdr));
288 }
289
290 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
291 {
292         struct qtag_prefix {
293                 __be16 eth_type; /* ETH_P_8021Q */
294                 __be16 tci;
295         };
296         struct qtag_prefix *qp;
297
298         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
299                 return 0;
300
301         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
302                                          sizeof(__be16))))
303                 return -ENOMEM;
304
305         qp = (struct qtag_prefix *) skb->data;
306         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
307         __skb_pull(skb, sizeof(struct qtag_prefix));
308
309         return 0;
310 }
311
312 static __be16 parse_ethertype(struct sk_buff *skb)
313 {
314         struct llc_snap_hdr {
315                 u8  dsap;  /* Always 0xAA */
316                 u8  ssap;  /* Always 0xAA */
317                 u8  ctrl;
318                 u8  oui[3];
319                 __be16 ethertype;
320         };
321         struct llc_snap_hdr *llc;
322         __be16 proto;
323
324         proto = *(__be16 *) skb->data;
325         __skb_pull(skb, sizeof(__be16));
326
327         if (ntohs(proto) >= ETH_P_802_3_MIN)
328                 return proto;
329
330         if (skb->len < sizeof(struct llc_snap_hdr))
331                 return htons(ETH_P_802_2);
332
333         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
334                 return htons(0);
335
336         llc = (struct llc_snap_hdr *) skb->data;
337         if (llc->dsap != LLC_SAP_SNAP ||
338             llc->ssap != LLC_SAP_SNAP ||
339             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
340                 return htons(ETH_P_802_2);
341
342         __skb_pull(skb, sizeof(struct llc_snap_hdr));
343
344         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
345                 return llc->ethertype;
346
347         return htons(ETH_P_802_2);
348 }
349
350 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
351                         int nh_len)
352 {
353         struct icmp6hdr *icmp = icmp6_hdr(skb);
354
355         /* The ICMPv6 type and code fields use the 16-bit transport port
356          * fields, so we need to store them in 16-bit network byte order.
357          */
358         key->tp.src = htons(icmp->icmp6_type);
359         key->tp.dst = htons(icmp->icmp6_code);
360
361         if (icmp->icmp6_code == 0 &&
362             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
363              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
364                 int icmp_len = skb->len - skb_transport_offset(skb);
365                 struct nd_msg *nd;
366                 int offset;
367
368                 /* In order to process neighbor discovery options, we need the
369                  * entire packet.
370                  */
371                 if (unlikely(icmp_len < sizeof(*nd)))
372                         return 0;
373
374                 if (unlikely(skb_linearize(skb)))
375                         return -ENOMEM;
376
377                 nd = (struct nd_msg *)skb_transport_header(skb);
378                 key->ipv6.nd.target = nd->target;
379
380                 icmp_len -= sizeof(*nd);
381                 offset = 0;
382                 while (icmp_len >= 8) {
383                         struct nd_opt_hdr *nd_opt =
384                                  (struct nd_opt_hdr *)(nd->opt + offset);
385                         int opt_len = nd_opt->nd_opt_len * 8;
386
387                         if (unlikely(!opt_len || opt_len > icmp_len))
388                                 return 0;
389
390                         /* Store the link layer address if the appropriate
391                          * option is provided.  It is considered an error if
392                          * the same link layer option is specified twice.
393                          */
394                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
395                             && opt_len == 8) {
396                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
397                                         goto invalid;
398                                 ether_addr_copy(key->ipv6.nd.sll,
399                                     &nd->opt[offset+sizeof(*nd_opt)]);
400                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
401                                    && opt_len == 8) {
402                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
403                                         goto invalid;
404                                 ether_addr_copy(key->ipv6.nd.tll,
405                                     &nd->opt[offset+sizeof(*nd_opt)]);
406                         }
407
408                         icmp_len -= opt_len;
409                         offset += opt_len;
410                 }
411         }
412
413         return 0;
414
415 invalid:
416         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
417         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
418         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
419
420         return 0;
421 }
422
423 /**
424  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
425  * @skb: sk_buff that contains the frame, with skb->data pointing to the
426  * Ethernet header
427  * @in_port: port number on which @skb was received.
428  * @key: output flow key
429  *
430  * The caller must ensure that skb->len >= ETH_HLEN.
431  *
432  * Returns 0 if successful, otherwise a negative errno value.
433  *
434  * Initializes @skb header pointers as follows:
435  *
436  *    - skb->mac_header: the Ethernet header.
437  *
438  *    - skb->network_header: just past the Ethernet header, or just past the
439  *      VLAN header, to the first byte of the Ethernet payload.
440  *
441  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
442  *      on output, then just past the IP header, if one is present and
443  *      of a correct length, otherwise the same as skb->network_header.
444  *      For other key->eth.type values it is left untouched.
445  */
446 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
447 {
448         int error;
449         struct ethhdr *eth;
450
451         memset(key, 0, sizeof(*key));
452
453         key->phy.priority = skb->priority;
454         if (OVS_CB(skb)->tun_key)
455                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
456         key->phy.in_port = in_port;
457         key->phy.skb_mark = skb->mark;
458
459         skb_reset_mac_header(skb);
460
461         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
462          * header in the linear data area.
463          */
464         eth = eth_hdr(skb);
465         ether_addr_copy(key->eth.src, eth->h_source);
466         ether_addr_copy(key->eth.dst, eth->h_dest);
467
468         __skb_pull(skb, 2 * ETH_ALEN);
469         /* We are going to push all headers that we pull, so no need to
470          * update skb->csum here. */
471
472         if (vlan_tx_tag_present(skb))
473                 key->eth.tci = htons(vlan_get_tci(skb));
474         else if (eth->h_proto == htons(ETH_P_8021Q))
475                 if (unlikely(parse_vlan(skb, key)))
476                         return -ENOMEM;
477
478         key->eth.type = parse_ethertype(skb);
479         if (unlikely(key->eth.type == htons(0)))
480                 return -ENOMEM;
481
482         skb_reset_network_header(skb);
483         __skb_push(skb, skb->data - skb_mac_header(skb));
484
485         /* Network layer. */
486         if (key->eth.type == htons(ETH_P_IP)) {
487                 struct iphdr *nh;
488                 __be16 offset;
489
490                 error = check_iphdr(skb);
491                 if (unlikely(error)) {
492                         if (error == -EINVAL) {
493                                 skb->transport_header = skb->network_header;
494                                 error = 0;
495                         }
496                         return error;
497                 }
498
499                 nh = ip_hdr(skb);
500                 key->ipv4.addr.src = nh->saddr;
501                 key->ipv4.addr.dst = nh->daddr;
502
503                 key->ip.proto = nh->protocol;
504                 key->ip.tos = nh->tos;
505                 key->ip.ttl = nh->ttl;
506
507                 offset = nh->frag_off & htons(IP_OFFSET);
508                 if (offset) {
509                         key->ip.frag = OVS_FRAG_TYPE_LATER;
510                         return 0;
511                 }
512                 if (nh->frag_off & htons(IP_MF) ||
513                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
514                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
515
516                 /* Transport layer. */
517                 if (key->ip.proto == IPPROTO_TCP) {
518                         if (tcphdr_ok(skb)) {
519                                 struct tcphdr *tcp = tcp_hdr(skb);
520                                 key->tp.src = tcp->source;
521                                 key->tp.dst = tcp->dest;
522                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
523                         }
524                 } else if (key->ip.proto == IPPROTO_UDP) {
525                         if (udphdr_ok(skb)) {
526                                 struct udphdr *udp = udp_hdr(skb);
527                                 key->tp.src = udp->source;
528                                 key->tp.dst = udp->dest;
529                         }
530                 } else if (key->ip.proto == IPPROTO_SCTP) {
531                         if (sctphdr_ok(skb)) {
532                                 struct sctphdr *sctp = sctp_hdr(skb);
533                                 key->tp.src = sctp->source;
534                                 key->tp.dst = sctp->dest;
535                         }
536                 } else if (key->ip.proto == IPPROTO_ICMP) {
537                         if (icmphdr_ok(skb)) {
538                                 struct icmphdr *icmp = icmp_hdr(skb);
539                                 /* The ICMP type and code fields use the 16-bit
540                                  * transport port fields, so we need to store
541                                  * them in 16-bit network byte order. */
542                                 key->tp.src = htons(icmp->type);
543                                 key->tp.dst = htons(icmp->code);
544                         }
545                 }
546
547         } else if ((key->eth.type == htons(ETH_P_ARP) ||
548                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
549                 struct arp_eth_header *arp;
550
551                 arp = (struct arp_eth_header *)skb_network_header(skb);
552
553                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
554                                 && arp->ar_pro == htons(ETH_P_IP)
555                                 && arp->ar_hln == ETH_ALEN
556                                 && arp->ar_pln == 4) {
557
558                         /* We only match on the lower 8 bits of the opcode. */
559                         if (ntohs(arp->ar_op) <= 0xff)
560                                 key->ip.proto = ntohs(arp->ar_op);
561                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
562                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
563                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
564                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
565                 }
566         } else if (key->eth.type == htons(ETH_P_IPV6)) {
567                 int nh_len;             /* IPv6 Header + Extensions */
568
569                 nh_len = parse_ipv6hdr(skb, key);
570                 if (unlikely(nh_len < 0)) {
571                         if (nh_len == -EINVAL) {
572                                 skb->transport_header = skb->network_header;
573                                 error = 0;
574                         } else {
575                                 error = nh_len;
576                         }
577                         return error;
578                 }
579
580                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
581                         return 0;
582                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
583                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
584
585                 /* Transport layer. */
586                 if (key->ip.proto == NEXTHDR_TCP) {
587                         if (tcphdr_ok(skb)) {
588                                 struct tcphdr *tcp = tcp_hdr(skb);
589                                 key->tp.src = tcp->source;
590                                 key->tp.dst = tcp->dest;
591                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
592                         }
593                 } else if (key->ip.proto == NEXTHDR_UDP) {
594                         if (udphdr_ok(skb)) {
595                                 struct udphdr *udp = udp_hdr(skb);
596                                 key->tp.src = udp->source;
597                                 key->tp.dst = udp->dest;
598                         }
599                 } else if (key->ip.proto == NEXTHDR_SCTP) {
600                         if (sctphdr_ok(skb)) {
601                                 struct sctphdr *sctp = sctp_hdr(skb);
602                                 key->tp.src = sctp->source;
603                                 key->tp.dst = sctp->dest;
604                         }
605                 } else if (key->ip.proto == NEXTHDR_ICMP) {
606                         if (icmp6hdr_ok(skb)) {
607                                 error = parse_icmpv6(skb, key, nh_len);
608                                 if (error)
609                                         return error;
610                         }
611                 }
612         }
613
614         return 0;
615 }