openvswitch: Use tun_key only for egress tunnel path.
[firefly-linux-kernel-4.4.55.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 <linux/uaccess.h>
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/if_ether.h>
23 #include <linux/if_vlan.h>
24 #include <net/llc_pdu.h>
25 #include <linux/kernel.h>
26 #include <linux/jhash.h>
27 #include <linux/jiffies.h>
28 #include <linux/llc.h>
29 #include <linux/module.h>
30 #include <linux/in.h>
31 #include <linux/rcupdate.h>
32 #include <linux/if_arp.h>
33 #include <linux/ip.h>
34 #include <linux/ipv6.h>
35 #include <linux/sctp.h>
36 #include <linux/smp.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/icmp.h>
40 #include <linux/icmpv6.h>
41 #include <linux/rculist.h>
42 #include <net/ip.h>
43 #include <net/ip_tunnels.h>
44 #include <net/ipv6.h>
45 #include <net/ndisc.h>
46
47 #include "datapath.h"
48 #include "flow.h"
49 #include "flow_netlink.h"
50
51 u64 ovs_flow_used_time(unsigned long flow_jiffies)
52 {
53         struct timespec cur_ts;
54         u64 cur_ms, idle_ms;
55
56         ktime_get_ts(&cur_ts);
57         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
58         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
59                  cur_ts.tv_nsec / NSEC_PER_MSEC;
60
61         return cur_ms - idle_ms;
62 }
63
64 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
65
66 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
67                            struct sk_buff *skb)
68 {
69         struct flow_stats *stats;
70         int node = numa_node_id();
71
72         stats = rcu_dereference(flow->stats[node]);
73
74         /* Check if already have node-specific stats. */
75         if (likely(stats)) {
76                 spin_lock(&stats->lock);
77                 /* Mark if we write on the pre-allocated stats. */
78                 if (node == 0 && unlikely(flow->stats_last_writer != node))
79                         flow->stats_last_writer = node;
80         } else {
81                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
82                 spin_lock(&stats->lock);
83
84                 /* If the current NUMA-node is the only writer on the
85                  * pre-allocated stats keep using them.
86                  */
87                 if (unlikely(flow->stats_last_writer != node)) {
88                         /* A previous locker may have already allocated the
89                          * stats, so we need to check again.  If node-specific
90                          * stats were already allocated, we update the pre-
91                          * allocated stats as we have already locked them.
92                          */
93                         if (likely(flow->stats_last_writer != NUMA_NO_NODE)
94                             && likely(!rcu_access_pointer(flow->stats[node]))) {
95                                 /* Try to allocate node-specific stats. */
96                                 struct flow_stats *new_stats;
97
98                                 new_stats =
99                                         kmem_cache_alloc_node(flow_stats_cache,
100                                                               GFP_THISNODE |
101                                                               __GFP_NOMEMALLOC,
102                                                               node);
103                                 if (likely(new_stats)) {
104                                         new_stats->used = jiffies;
105                                         new_stats->packet_count = 1;
106                                         new_stats->byte_count = skb->len;
107                                         new_stats->tcp_flags = tcp_flags;
108                                         spin_lock_init(&new_stats->lock);
109
110                                         rcu_assign_pointer(flow->stats[node],
111                                                            new_stats);
112                                         goto unlock;
113                                 }
114                         }
115                         flow->stats_last_writer = node;
116                 }
117         }
118
119         stats->used = jiffies;
120         stats->packet_count++;
121         stats->byte_count += skb->len;
122         stats->tcp_flags |= tcp_flags;
123 unlock:
124         spin_unlock(&stats->lock);
125 }
126
127 /* Must be called with rcu_read_lock or ovs_mutex. */
128 void ovs_flow_stats_get(const struct sw_flow *flow,
129                         struct ovs_flow_stats *ovs_stats,
130                         unsigned long *used, __be16 *tcp_flags)
131 {
132         int node;
133
134         *used = 0;
135         *tcp_flags = 0;
136         memset(ovs_stats, 0, sizeof(*ovs_stats));
137
138         for_each_node(node) {
139                 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
140
141                 if (stats) {
142                         /* Local CPU may write on non-local stats, so we must
143                          * block bottom-halves here.
144                          */
145                         spin_lock_bh(&stats->lock);
146                         if (!*used || time_after(stats->used, *used))
147                                 *used = stats->used;
148                         *tcp_flags |= stats->tcp_flags;
149                         ovs_stats->n_packets += stats->packet_count;
150                         ovs_stats->n_bytes += stats->byte_count;
151                         spin_unlock_bh(&stats->lock);
152                 }
153         }
154 }
155
156 /* Called with ovs_mutex. */
157 void ovs_flow_stats_clear(struct sw_flow *flow)
158 {
159         int node;
160
161         for_each_node(node) {
162                 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
163
164                 if (stats) {
165                         spin_lock_bh(&stats->lock);
166                         stats->used = 0;
167                         stats->packet_count = 0;
168                         stats->byte_count = 0;
169                         stats->tcp_flags = 0;
170                         spin_unlock_bh(&stats->lock);
171                 }
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->tp.src = htons(icmp->icmp6_type);
360         key->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                                 ether_addr_copy(key->ipv6.nd.sll,
400                                                 &nd->opt[offset+sizeof(*nd_opt)]);
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                                 ether_addr_copy(key->ipv6.nd.tll,
406                                                 &nd->opt[offset+sizeof(*nd_opt)]);
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  * key_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  * @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 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
447 {
448         int error;
449         struct ethhdr *eth;
450
451         skb_reset_mac_header(skb);
452
453         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
454          * header in the linear data area.
455          */
456         eth = eth_hdr(skb);
457         ether_addr_copy(key->eth.src, eth->h_source);
458         ether_addr_copy(key->eth.dst, eth->h_dest);
459
460         __skb_pull(skb, 2 * ETH_ALEN);
461         /* We are going to push all headers that we pull, so no need to
462          * update skb->csum here.
463          */
464
465         if (vlan_tx_tag_present(skb))
466                 key->eth.tci = htons(skb->vlan_tci);
467         else if (eth->h_proto == htons(ETH_P_8021Q))
468                 if (unlikely(parse_vlan(skb, key)))
469                         return -ENOMEM;
470
471         key->eth.type = parse_ethertype(skb);
472         if (unlikely(key->eth.type == htons(0)))
473                 return -ENOMEM;
474
475         skb_reset_network_header(skb);
476         __skb_push(skb, skb->data - skb_mac_header(skb));
477
478         /* Network layer. */
479         if (key->eth.type == htons(ETH_P_IP)) {
480                 struct iphdr *nh;
481                 __be16 offset;
482
483                 error = check_iphdr(skb);
484                 if (unlikely(error)) {
485                         if (error == -EINVAL) {
486                                 skb->transport_header = skb->network_header;
487                                 error = 0;
488                         }
489                         return error;
490                 }
491
492                 nh = ip_hdr(skb);
493                 key->ipv4.addr.src = nh->saddr;
494                 key->ipv4.addr.dst = nh->daddr;
495
496                 key->ip.proto = nh->protocol;
497                 key->ip.tos = nh->tos;
498                 key->ip.ttl = nh->ttl;
499
500                 offset = nh->frag_off & htons(IP_OFFSET);
501                 if (offset) {
502                         key->ip.frag = OVS_FRAG_TYPE_LATER;
503                         return 0;
504                 }
505                 if (nh->frag_off & htons(IP_MF) ||
506                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
507                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
508
509                 /* Transport layer. */
510                 if (key->ip.proto == IPPROTO_TCP) {
511                         if (tcphdr_ok(skb)) {
512                                 struct tcphdr *tcp = tcp_hdr(skb);
513                                 key->tp.src = tcp->source;
514                                 key->tp.dst = tcp->dest;
515                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
516                         }
517                 } else if (key->ip.proto == IPPROTO_UDP) {
518                         if (udphdr_ok(skb)) {
519                                 struct udphdr *udp = udp_hdr(skb);
520                                 key->tp.src = udp->source;
521                                 key->tp.dst = udp->dest;
522                         }
523                 } else if (key->ip.proto == IPPROTO_SCTP) {
524                         if (sctphdr_ok(skb)) {
525                                 struct sctphdr *sctp = sctp_hdr(skb);
526                                 key->tp.src = sctp->source;
527                                 key->tp.dst = sctp->dest;
528                         }
529                 } else if (key->ip.proto == IPPROTO_ICMP) {
530                         if (icmphdr_ok(skb)) {
531                                 struct icmphdr *icmp = icmp_hdr(skb);
532                                 /* The ICMP type and code fields use the 16-bit
533                                  * transport port fields, so we need to store
534                                  * them in 16-bit network byte order. */
535                                 key->tp.src = htons(icmp->type);
536                                 key->tp.dst = htons(icmp->code);
537                         }
538                 }
539
540         } else if ((key->eth.type == htons(ETH_P_ARP) ||
541                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
542                 struct arp_eth_header *arp;
543
544                 arp = (struct arp_eth_header *)skb_network_header(skb);
545
546                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
547                                 && arp->ar_pro == htons(ETH_P_IP)
548                                 && arp->ar_hln == ETH_ALEN
549                                 && arp->ar_pln == 4) {
550
551                         /* We only match on the lower 8 bits of the opcode. */
552                         if (ntohs(arp->ar_op) <= 0xff)
553                                 key->ip.proto = ntohs(arp->ar_op);
554                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
555                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
556                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
557                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
558                 }
559         } else if (key->eth.type == htons(ETH_P_IPV6)) {
560                 int nh_len;             /* IPv6 Header + Extensions */
561
562                 nh_len = parse_ipv6hdr(skb, key);
563                 if (unlikely(nh_len < 0)) {
564                         if (nh_len == -EINVAL) {
565                                 skb->transport_header = skb->network_header;
566                                 error = 0;
567                         } else {
568                                 error = nh_len;
569                         }
570                         return error;
571                 }
572
573                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
574                         return 0;
575                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
576                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
577
578                 /* Transport layer. */
579                 if (key->ip.proto == NEXTHDR_TCP) {
580                         if (tcphdr_ok(skb)) {
581                                 struct tcphdr *tcp = tcp_hdr(skb);
582                                 key->tp.src = tcp->source;
583                                 key->tp.dst = tcp->dest;
584                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
585                         }
586                 } else if (key->ip.proto == NEXTHDR_UDP) {
587                         if (udphdr_ok(skb)) {
588                                 struct udphdr *udp = udp_hdr(skb);
589                                 key->tp.src = udp->source;
590                                 key->tp.dst = udp->dest;
591                         }
592                 } else if (key->ip.proto == NEXTHDR_SCTP) {
593                         if (sctphdr_ok(skb)) {
594                                 struct sctphdr *sctp = sctp_hdr(skb);
595                                 key->tp.src = sctp->source;
596                                 key->tp.dst = sctp->dest;
597                         }
598                 } else if (key->ip.proto == NEXTHDR_ICMP) {
599                         if (icmp6hdr_ok(skb)) {
600                                 error = parse_icmpv6(skb, key, nh_len);
601                                 if (error)
602                                         return error;
603                         }
604                 }
605         }
606         return 0;
607 }
608
609 int ovs_flow_key_extract(struct ovs_key_ipv4_tunnel *tun_key,
610                          struct sk_buff *skb, struct sw_flow_key *key)
611 {
612         /* Extract metadata from packet. */
613         memset(key, 0, sizeof(*key));
614         if (tun_key)
615                 memcpy(&key->tun_key, tun_key, sizeof(key->tun_key));
616
617         key->phy.priority = skb->priority;
618         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
619         key->phy.skb_mark = skb->mark;
620
621         return key_extract(skb, key);
622 }
623
624 int ovs_flow_key_extract_userspace(const struct nlattr *attr,
625                                    struct sk_buff *skb,
626                                    struct sw_flow_key *key)
627 {
628         int err;
629
630         memset(key, 0, sizeof(*key));
631         /* Extract metadata from netlink attributes. */
632         err = ovs_nla_get_flow_metadata(attr, key);
633         if (err)
634                 return err;
635
636         return key_extract(skb, key);
637 }