Merge tag 'pci-v5.0-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
[linux-2.6-microblaze.git] / net / openvswitch / 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 <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/cpumask.h>
33 #include <linux/if_arp.h>
34 #include <linux/ip.h>
35 #include <linux/ipv6.h>
36 #include <linux/mpls.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/mpls.h>
48 #include <net/ndisc.h>
49 #include <net/nsh.h>
50
51 #include "conntrack.h"
52 #include "datapath.h"
53 #include "flow.h"
54 #include "flow_netlink.h"
55 #include "vport.h"
56
57 u64 ovs_flow_used_time(unsigned long flow_jiffies)
58 {
59         struct timespec64 cur_ts;
60         u64 cur_ms, idle_ms;
61
62         ktime_get_ts64(&cur_ts);
63         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
64         cur_ms = (u64)(u32)cur_ts.tv_sec * MSEC_PER_SEC +
65                  cur_ts.tv_nsec / NSEC_PER_MSEC;
66
67         return cur_ms - idle_ms;
68 }
69
70 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
71
72 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
73                            const struct sk_buff *skb)
74 {
75         struct flow_stats *stats;
76         unsigned int cpu = smp_processor_id();
77         int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
78
79         stats = rcu_dereference(flow->stats[cpu]);
80
81         /* Check if already have CPU-specific stats. */
82         if (likely(stats)) {
83                 spin_lock(&stats->lock);
84                 /* Mark if we write on the pre-allocated stats. */
85                 if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
86                         flow->stats_last_writer = cpu;
87         } else {
88                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
89                 spin_lock(&stats->lock);
90
91                 /* If the current CPU is the only writer on the
92                  * pre-allocated stats keep using them.
93                  */
94                 if (unlikely(flow->stats_last_writer != cpu)) {
95                         /* A previous locker may have already allocated the
96                          * stats, so we need to check again.  If CPU-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 != -1) &&
101                             likely(!rcu_access_pointer(flow->stats[cpu]))) {
102                                 /* Try to allocate CPU-specific stats. */
103                                 struct flow_stats *new_stats;
104
105                                 new_stats =
106                                         kmem_cache_alloc_node(flow_stats_cache,
107                                                               GFP_NOWAIT |
108                                                               __GFP_THISNODE |
109                                                               __GFP_NOWARN |
110                                                               __GFP_NOMEMALLOC,
111                                                               numa_node_id());
112                                 if (likely(new_stats)) {
113                                         new_stats->used = jiffies;
114                                         new_stats->packet_count = 1;
115                                         new_stats->byte_count = len;
116                                         new_stats->tcp_flags = tcp_flags;
117                                         spin_lock_init(&new_stats->lock);
118
119                                         rcu_assign_pointer(flow->stats[cpu],
120                                                            new_stats);
121                                         cpumask_set_cpu(cpu, &flow->cpu_used_mask);
122                                         goto unlock;
123                                 }
124                         }
125                         flow->stats_last_writer = cpu;
126                 }
127         }
128
129         stats->used = jiffies;
130         stats->packet_count++;
131         stats->byte_count += len;
132         stats->tcp_flags |= tcp_flags;
133 unlock:
134         spin_unlock(&stats->lock);
135 }
136
137 /* Must be called with rcu_read_lock or ovs_mutex. */
138 void ovs_flow_stats_get(const struct sw_flow *flow,
139                         struct ovs_flow_stats *ovs_stats,
140                         unsigned long *used, __be16 *tcp_flags)
141 {
142         int cpu;
143
144         *used = 0;
145         *tcp_flags = 0;
146         memset(ovs_stats, 0, sizeof(*ovs_stats));
147
148         /* We open code this to make sure cpu 0 is always considered */
149         for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
150                 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
151
152                 if (stats) {
153                         /* Local CPU may write on non-local stats, so we must
154                          * block bottom-halves here.
155                          */
156                         spin_lock_bh(&stats->lock);
157                         if (!*used || time_after(stats->used, *used))
158                                 *used = stats->used;
159                         *tcp_flags |= stats->tcp_flags;
160                         ovs_stats->n_packets += stats->packet_count;
161                         ovs_stats->n_bytes += stats->byte_count;
162                         spin_unlock_bh(&stats->lock);
163                 }
164         }
165 }
166
167 /* Called with ovs_mutex. */
168 void ovs_flow_stats_clear(struct sw_flow *flow)
169 {
170         int cpu;
171
172         /* We open code this to make sure cpu 0 is always considered */
173         for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
174                 struct flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
175
176                 if (stats) {
177                         spin_lock_bh(&stats->lock);
178                         stats->used = 0;
179                         stats->packet_count = 0;
180                         stats->byte_count = 0;
181                         stats->tcp_flags = 0;
182                         spin_unlock_bh(&stats->lock);
183                 }
184         }
185 }
186
187 static int check_header(struct sk_buff *skb, int len)
188 {
189         if (unlikely(skb->len < len))
190                 return -EINVAL;
191         if (unlikely(!pskb_may_pull(skb, len)))
192                 return -ENOMEM;
193         return 0;
194 }
195
196 static bool arphdr_ok(struct sk_buff *skb)
197 {
198         return pskb_may_pull(skb, skb_network_offset(skb) +
199                                   sizeof(struct arp_eth_header));
200 }
201
202 static int check_iphdr(struct sk_buff *skb)
203 {
204         unsigned int nh_ofs = skb_network_offset(skb);
205         unsigned int ip_len;
206         int err;
207
208         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
209         if (unlikely(err))
210                 return err;
211
212         ip_len = ip_hdrlen(skb);
213         if (unlikely(ip_len < sizeof(struct iphdr) ||
214                      skb->len < nh_ofs + ip_len))
215                 return -EINVAL;
216
217         skb_set_transport_header(skb, nh_ofs + ip_len);
218         return 0;
219 }
220
221 static bool tcphdr_ok(struct sk_buff *skb)
222 {
223         int th_ofs = skb_transport_offset(skb);
224         int tcp_len;
225
226         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
227                 return false;
228
229         tcp_len = tcp_hdrlen(skb);
230         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
231                      skb->len < th_ofs + tcp_len))
232                 return false;
233
234         return true;
235 }
236
237 static bool udphdr_ok(struct sk_buff *skb)
238 {
239         return pskb_may_pull(skb, skb_transport_offset(skb) +
240                                   sizeof(struct udphdr));
241 }
242
243 static bool sctphdr_ok(struct sk_buff *skb)
244 {
245         return pskb_may_pull(skb, skb_transport_offset(skb) +
246                                   sizeof(struct sctphdr));
247 }
248
249 static bool icmphdr_ok(struct sk_buff *skb)
250 {
251         return pskb_may_pull(skb, skb_transport_offset(skb) +
252                                   sizeof(struct icmphdr));
253 }
254
255 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
256 {
257         unsigned short frag_off;
258         unsigned int payload_ofs = 0;
259         unsigned int nh_ofs = skb_network_offset(skb);
260         unsigned int nh_len;
261         struct ipv6hdr *nh;
262         int err, nexthdr, flags = 0;
263
264         err = check_header(skb, nh_ofs + sizeof(*nh));
265         if (unlikely(err))
266                 return err;
267
268         nh = ipv6_hdr(skb);
269
270         key->ip.proto = NEXTHDR_NONE;
271         key->ip.tos = ipv6_get_dsfield(nh);
272         key->ip.ttl = nh->hop_limit;
273         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
274         key->ipv6.addr.src = nh->saddr;
275         key->ipv6.addr.dst = nh->daddr;
276
277         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
278         if (flags & IP6_FH_F_FRAG) {
279                 if (frag_off) {
280                         key->ip.frag = OVS_FRAG_TYPE_LATER;
281                         key->ip.proto = nexthdr;
282                         return 0;
283                 }
284                 key->ip.frag = OVS_FRAG_TYPE_FIRST;
285         } else {
286                 key->ip.frag = OVS_FRAG_TYPE_NONE;
287         }
288
289         /* Delayed handling of error in ipv6_find_hdr() as it
290          * always sets flags and frag_off to a valid value which may be
291          * used to set key->ip.frag above.
292          */
293         if (unlikely(nexthdr < 0))
294                 return -EPROTO;
295
296         nh_len = payload_ofs - nh_ofs;
297         skb_set_transport_header(skb, nh_ofs + nh_len);
298         key->ip.proto = nexthdr;
299         return nh_len;
300 }
301
302 static bool icmp6hdr_ok(struct sk_buff *skb)
303 {
304         return pskb_may_pull(skb, skb_transport_offset(skb) +
305                                   sizeof(struct icmp6hdr));
306 }
307
308 /**
309  * Parse vlan tag from vlan header.
310  * Returns ERROR on memory error.
311  * Returns 0 if it encounters a non-vlan or incomplete packet.
312  * Returns 1 after successfully parsing vlan tag.
313  */
314 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
315                           bool untag_vlan)
316 {
317         struct vlan_head *vh = (struct vlan_head *)skb->data;
318
319         if (likely(!eth_type_vlan(vh->tpid)))
320                 return 0;
321
322         if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
323                 return 0;
324
325         if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
326                                  sizeof(__be16))))
327                 return -ENOMEM;
328
329         vh = (struct vlan_head *)skb->data;
330         key_vh->tci = vh->tci | htons(VLAN_CFI_MASK);
331         key_vh->tpid = vh->tpid;
332
333         if (unlikely(untag_vlan)) {
334                 int offset = skb->data - skb_mac_header(skb);
335                 u16 tci;
336                 int err;
337
338                 __skb_push(skb, offset);
339                 err = __skb_vlan_pop(skb, &tci);
340                 __skb_pull(skb, offset);
341                 if (err)
342                         return err;
343                 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
344         } else {
345                 __skb_pull(skb, sizeof(struct vlan_head));
346         }
347         return 1;
348 }
349
350 static void clear_vlan(struct sw_flow_key *key)
351 {
352         key->eth.vlan.tci = 0;
353         key->eth.vlan.tpid = 0;
354         key->eth.cvlan.tci = 0;
355         key->eth.cvlan.tpid = 0;
356 }
357
358 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
359 {
360         int res;
361
362         if (skb_vlan_tag_present(skb)) {
363                 key->eth.vlan.tci = htons(skb->vlan_tci) | htons(VLAN_CFI_MASK);
364                 key->eth.vlan.tpid = skb->vlan_proto;
365         } else {
366                 /* Parse outer vlan tag in the non-accelerated case. */
367                 res = parse_vlan_tag(skb, &key->eth.vlan, true);
368                 if (res <= 0)
369                         return res;
370         }
371
372         /* Parse inner vlan tag. */
373         res = parse_vlan_tag(skb, &key->eth.cvlan, false);
374         if (res <= 0)
375                 return res;
376
377         return 0;
378 }
379
380 static __be16 parse_ethertype(struct sk_buff *skb)
381 {
382         struct llc_snap_hdr {
383                 u8  dsap;  /* Always 0xAA */
384                 u8  ssap;  /* Always 0xAA */
385                 u8  ctrl;
386                 u8  oui[3];
387                 __be16 ethertype;
388         };
389         struct llc_snap_hdr *llc;
390         __be16 proto;
391
392         proto = *(__be16 *) skb->data;
393         __skb_pull(skb, sizeof(__be16));
394
395         if (eth_proto_is_802_3(proto))
396                 return proto;
397
398         if (skb->len < sizeof(struct llc_snap_hdr))
399                 return htons(ETH_P_802_2);
400
401         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
402                 return htons(0);
403
404         llc = (struct llc_snap_hdr *) skb->data;
405         if (llc->dsap != LLC_SAP_SNAP ||
406             llc->ssap != LLC_SAP_SNAP ||
407             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
408                 return htons(ETH_P_802_2);
409
410         __skb_pull(skb, sizeof(struct llc_snap_hdr));
411
412         if (eth_proto_is_802_3(llc->ethertype))
413                 return llc->ethertype;
414
415         return htons(ETH_P_802_2);
416 }
417
418 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
419                         int nh_len)
420 {
421         struct icmp6hdr *icmp = icmp6_hdr(skb);
422
423         /* The ICMPv6 type and code fields use the 16-bit transport port
424          * fields, so we need to store them in 16-bit network byte order.
425          */
426         key->tp.src = htons(icmp->icmp6_type);
427         key->tp.dst = htons(icmp->icmp6_code);
428         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
429
430         if (icmp->icmp6_code == 0 &&
431             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
432              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
433                 int icmp_len = skb->len - skb_transport_offset(skb);
434                 struct nd_msg *nd;
435                 int offset;
436
437                 /* In order to process neighbor discovery options, we need the
438                  * entire packet.
439                  */
440                 if (unlikely(icmp_len < sizeof(*nd)))
441                         return 0;
442
443                 if (unlikely(skb_linearize(skb)))
444                         return -ENOMEM;
445
446                 nd = (struct nd_msg *)skb_transport_header(skb);
447                 key->ipv6.nd.target = nd->target;
448
449                 icmp_len -= sizeof(*nd);
450                 offset = 0;
451                 while (icmp_len >= 8) {
452                         struct nd_opt_hdr *nd_opt =
453                                  (struct nd_opt_hdr *)(nd->opt + offset);
454                         int opt_len = nd_opt->nd_opt_len * 8;
455
456                         if (unlikely(!opt_len || opt_len > icmp_len))
457                                 return 0;
458
459                         /* Store the link layer address if the appropriate
460                          * option is provided.  It is considered an error if
461                          * the same link layer option is specified twice.
462                          */
463                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
464                             && opt_len == 8) {
465                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
466                                         goto invalid;
467                                 ether_addr_copy(key->ipv6.nd.sll,
468                                                 &nd->opt[offset+sizeof(*nd_opt)]);
469                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
470                                    && opt_len == 8) {
471                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
472                                         goto invalid;
473                                 ether_addr_copy(key->ipv6.nd.tll,
474                                                 &nd->opt[offset+sizeof(*nd_opt)]);
475                         }
476
477                         icmp_len -= opt_len;
478                         offset += opt_len;
479                 }
480         }
481
482         return 0;
483
484 invalid:
485         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
486         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
487         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
488
489         return 0;
490 }
491
492 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
493 {
494         struct nshhdr *nh;
495         unsigned int nh_ofs = skb_network_offset(skb);
496         u8 version, length;
497         int err;
498
499         err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
500         if (unlikely(err))
501                 return err;
502
503         nh = nsh_hdr(skb);
504         version = nsh_get_ver(nh);
505         length = nsh_hdr_len(nh);
506
507         if (version != 0)
508                 return -EINVAL;
509
510         err = check_header(skb, nh_ofs + length);
511         if (unlikely(err))
512                 return err;
513
514         nh = nsh_hdr(skb);
515         key->nsh.base.flags = nsh_get_flags(nh);
516         key->nsh.base.ttl = nsh_get_ttl(nh);
517         key->nsh.base.mdtype = nh->mdtype;
518         key->nsh.base.np = nh->np;
519         key->nsh.base.path_hdr = nh->path_hdr;
520         switch (key->nsh.base.mdtype) {
521         case NSH_M_TYPE1:
522                 if (length != NSH_M_TYPE1_LEN)
523                         return -EINVAL;
524                 memcpy(key->nsh.context, nh->md1.context,
525                        sizeof(nh->md1));
526                 break;
527         case NSH_M_TYPE2:
528                 memset(key->nsh.context, 0,
529                        sizeof(nh->md1));
530                 break;
531         default:
532                 return -EINVAL;
533         }
534
535         return 0;
536 }
537
538 /**
539  * key_extract - extracts a flow key from an Ethernet frame.
540  * @skb: sk_buff that contains the frame, with skb->data pointing to the
541  * Ethernet header
542  * @key: output flow key
543  *
544  * The caller must ensure that skb->len >= ETH_HLEN.
545  *
546  * Returns 0 if successful, otherwise a negative errno value.
547  *
548  * Initializes @skb header fields as follows:
549  *
550  *    - skb->mac_header: the L2 header.
551  *
552  *    - skb->network_header: just past the L2 header, or just past the
553  *      VLAN header, to the first byte of the L2 payload.
554  *
555  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
556  *      on output, then just past the IP header, if one is present and
557  *      of a correct length, otherwise the same as skb->network_header.
558  *      For other key->eth.type values it is left untouched.
559  *
560  *    - skb->protocol: the type of the data starting at skb->network_header.
561  *      Equals to key->eth.type.
562  */
563 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
564 {
565         int error;
566         struct ethhdr *eth;
567
568         /* Flags are always used as part of stats */
569         key->tp.flags = 0;
570
571         skb_reset_mac_header(skb);
572
573         /* Link layer. */
574         clear_vlan(key);
575         if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
576                 if (unlikely(eth_type_vlan(skb->protocol)))
577                         return -EINVAL;
578
579                 skb_reset_network_header(skb);
580                 key->eth.type = skb->protocol;
581         } else {
582                 eth = eth_hdr(skb);
583                 ether_addr_copy(key->eth.src, eth->h_source);
584                 ether_addr_copy(key->eth.dst, eth->h_dest);
585
586                 __skb_pull(skb, 2 * ETH_ALEN);
587                 /* We are going to push all headers that we pull, so no need to
588                 * update skb->csum here.
589                 */
590
591                 if (unlikely(parse_vlan(skb, key)))
592                         return -ENOMEM;
593
594                 key->eth.type = parse_ethertype(skb);
595                 if (unlikely(key->eth.type == htons(0)))
596                         return -ENOMEM;
597
598                 /* Multiple tagged packets need to retain TPID to satisfy
599                  * skb_vlan_pop(), which will later shift the ethertype into
600                  * skb->protocol.
601                  */
602                 if (key->eth.cvlan.tci & htons(VLAN_CFI_MASK))
603                         skb->protocol = key->eth.cvlan.tpid;
604                 else
605                         skb->protocol = key->eth.type;
606
607                 skb_reset_network_header(skb);
608                 __skb_push(skb, skb->data - skb_mac_header(skb));
609         }
610         skb_reset_mac_len(skb);
611
612         /* Network layer. */
613         if (key->eth.type == htons(ETH_P_IP)) {
614                 struct iphdr *nh;
615                 __be16 offset;
616
617                 error = check_iphdr(skb);
618                 if (unlikely(error)) {
619                         memset(&key->ip, 0, sizeof(key->ip));
620                         memset(&key->ipv4, 0, sizeof(key->ipv4));
621                         if (error == -EINVAL) {
622                                 skb->transport_header = skb->network_header;
623                                 error = 0;
624                         }
625                         return error;
626                 }
627
628                 nh = ip_hdr(skb);
629                 key->ipv4.addr.src = nh->saddr;
630                 key->ipv4.addr.dst = nh->daddr;
631
632                 key->ip.proto = nh->protocol;
633                 key->ip.tos = nh->tos;
634                 key->ip.ttl = nh->ttl;
635
636                 offset = nh->frag_off & htons(IP_OFFSET);
637                 if (offset) {
638                         key->ip.frag = OVS_FRAG_TYPE_LATER;
639                         return 0;
640                 }
641                 if (nh->frag_off & htons(IP_MF) ||
642                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
643                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
644                 else
645                         key->ip.frag = OVS_FRAG_TYPE_NONE;
646
647                 /* Transport layer. */
648                 if (key->ip.proto == IPPROTO_TCP) {
649                         if (tcphdr_ok(skb)) {
650                                 struct tcphdr *tcp = tcp_hdr(skb);
651                                 key->tp.src = tcp->source;
652                                 key->tp.dst = tcp->dest;
653                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
654                         } else {
655                                 memset(&key->tp, 0, sizeof(key->tp));
656                         }
657
658                 } else if (key->ip.proto == IPPROTO_UDP) {
659                         if (udphdr_ok(skb)) {
660                                 struct udphdr *udp = udp_hdr(skb);
661                                 key->tp.src = udp->source;
662                                 key->tp.dst = udp->dest;
663                         } else {
664                                 memset(&key->tp, 0, sizeof(key->tp));
665                         }
666                 } else if (key->ip.proto == IPPROTO_SCTP) {
667                         if (sctphdr_ok(skb)) {
668                                 struct sctphdr *sctp = sctp_hdr(skb);
669                                 key->tp.src = sctp->source;
670                                 key->tp.dst = sctp->dest;
671                         } else {
672                                 memset(&key->tp, 0, sizeof(key->tp));
673                         }
674                 } else if (key->ip.proto == IPPROTO_ICMP) {
675                         if (icmphdr_ok(skb)) {
676                                 struct icmphdr *icmp = icmp_hdr(skb);
677                                 /* The ICMP type and code fields use the 16-bit
678                                  * transport port fields, so we need to store
679                                  * them in 16-bit network byte order. */
680                                 key->tp.src = htons(icmp->type);
681                                 key->tp.dst = htons(icmp->code);
682                         } else {
683                                 memset(&key->tp, 0, sizeof(key->tp));
684                         }
685                 }
686
687         } else if (key->eth.type == htons(ETH_P_ARP) ||
688                    key->eth.type == htons(ETH_P_RARP)) {
689                 struct arp_eth_header *arp;
690                 bool arp_available = arphdr_ok(skb);
691
692                 arp = (struct arp_eth_header *)skb_network_header(skb);
693
694                 if (arp_available &&
695                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
696                     arp->ar_pro == htons(ETH_P_IP) &&
697                     arp->ar_hln == ETH_ALEN &&
698                     arp->ar_pln == 4) {
699
700                         /* We only match on the lower 8 bits of the opcode. */
701                         if (ntohs(arp->ar_op) <= 0xff)
702                                 key->ip.proto = ntohs(arp->ar_op);
703                         else
704                                 key->ip.proto = 0;
705
706                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
707                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
708                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
709                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
710                 } else {
711                         memset(&key->ip, 0, sizeof(key->ip));
712                         memset(&key->ipv4, 0, sizeof(key->ipv4));
713                 }
714         } else if (eth_p_mpls(key->eth.type)) {
715                 size_t stack_len = MPLS_HLEN;
716
717                 skb_set_inner_network_header(skb, skb->mac_len);
718                 while (1) {
719                         __be32 lse;
720
721                         error = check_header(skb, skb->mac_len + stack_len);
722                         if (unlikely(error))
723                                 return 0;
724
725                         memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
726
727                         if (stack_len == MPLS_HLEN)
728                                 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
729
730                         skb_set_inner_network_header(skb, skb->mac_len + stack_len);
731                         if (lse & htonl(MPLS_LS_S_MASK))
732                                 break;
733
734                         stack_len += MPLS_HLEN;
735                 }
736         } else if (key->eth.type == htons(ETH_P_IPV6)) {
737                 int nh_len;             /* IPv6 Header + Extensions */
738
739                 nh_len = parse_ipv6hdr(skb, key);
740                 if (unlikely(nh_len < 0)) {
741                         switch (nh_len) {
742                         case -EINVAL:
743                                 memset(&key->ip, 0, sizeof(key->ip));
744                                 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
745                                 /* fall-through */
746                         case -EPROTO:
747                                 skb->transport_header = skb->network_header;
748                                 error = 0;
749                                 break;
750                         default:
751                                 error = nh_len;
752                         }
753                         return error;
754                 }
755
756                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
757                         return 0;
758                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
759                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
760
761                 /* Transport layer. */
762                 if (key->ip.proto == NEXTHDR_TCP) {
763                         if (tcphdr_ok(skb)) {
764                                 struct tcphdr *tcp = tcp_hdr(skb);
765                                 key->tp.src = tcp->source;
766                                 key->tp.dst = tcp->dest;
767                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
768                         } else {
769                                 memset(&key->tp, 0, sizeof(key->tp));
770                         }
771                 } else if (key->ip.proto == NEXTHDR_UDP) {
772                         if (udphdr_ok(skb)) {
773                                 struct udphdr *udp = udp_hdr(skb);
774                                 key->tp.src = udp->source;
775                                 key->tp.dst = udp->dest;
776                         } else {
777                                 memset(&key->tp, 0, sizeof(key->tp));
778                         }
779                 } else if (key->ip.proto == NEXTHDR_SCTP) {
780                         if (sctphdr_ok(skb)) {
781                                 struct sctphdr *sctp = sctp_hdr(skb);
782                                 key->tp.src = sctp->source;
783                                 key->tp.dst = sctp->dest;
784                         } else {
785                                 memset(&key->tp, 0, sizeof(key->tp));
786                         }
787                 } else if (key->ip.proto == NEXTHDR_ICMP) {
788                         if (icmp6hdr_ok(skb)) {
789                                 error = parse_icmpv6(skb, key, nh_len);
790                                 if (error)
791                                         return error;
792                         } else {
793                                 memset(&key->tp, 0, sizeof(key->tp));
794                         }
795                 }
796         } else if (key->eth.type == htons(ETH_P_NSH)) {
797                 error = parse_nsh(skb, key);
798                 if (error)
799                         return error;
800         }
801         return 0;
802 }
803
804 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
805 {
806         int res;
807
808         res = key_extract(skb, key);
809         if (!res)
810                 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
811
812         return res;
813 }
814
815 static int key_extract_mac_proto(struct sk_buff *skb)
816 {
817         switch (skb->dev->type) {
818         case ARPHRD_ETHER:
819                 return MAC_PROTO_ETHERNET;
820         case ARPHRD_NONE:
821                 if (skb->protocol == htons(ETH_P_TEB))
822                         return MAC_PROTO_ETHERNET;
823                 return MAC_PROTO_NONE;
824         }
825         WARN_ON_ONCE(1);
826         return -EINVAL;
827 }
828
829 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
830                          struct sk_buff *skb, struct sw_flow_key *key)
831 {
832         int res, err;
833
834         /* Extract metadata from packet. */
835         if (tun_info) {
836                 key->tun_proto = ip_tunnel_info_af(tun_info);
837                 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
838
839                 if (tun_info->options_len) {
840                         BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
841                                                    8)) - 1
842                                         > sizeof(key->tun_opts));
843
844                         ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
845                                                 tun_info);
846                         key->tun_opts_len = tun_info->options_len;
847                 } else {
848                         key->tun_opts_len = 0;
849                 }
850         } else  {
851                 key->tun_proto = 0;
852                 key->tun_opts_len = 0;
853                 memset(&key->tun_key, 0, sizeof(key->tun_key));
854         }
855
856         key->phy.priority = skb->priority;
857         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
858         key->phy.skb_mark = skb->mark;
859         key->ovs_flow_hash = 0;
860         res = key_extract_mac_proto(skb);
861         if (res < 0)
862                 return res;
863         key->mac_proto = res;
864         key->recirc_id = 0;
865
866         err = key_extract(skb, key);
867         if (!err)
868                 ovs_ct_fill_key(skb, key);   /* Must be after key_extract(). */
869         return err;
870 }
871
872 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
873                                    struct sk_buff *skb,
874                                    struct sw_flow_key *key, bool log)
875 {
876         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
877         u64 attrs = 0;
878         int err;
879
880         err = parse_flow_nlattrs(attr, a, &attrs, log);
881         if (err)
882                 return -EINVAL;
883
884         /* Extract metadata from netlink attributes. */
885         err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
886         if (err)
887                 return err;
888
889         /* key_extract assumes that skb->protocol is set-up for
890          * layer 3 packets which is the case for other callers,
891          * in particular packets received from the network stack.
892          * Here the correct value can be set from the metadata
893          * extracted above.
894          * For L2 packet key eth type would be zero. skb protocol
895          * would be set to correct value later during key-extact.
896          */
897
898         skb->protocol = key->eth.type;
899         err = key_extract(skb, key);
900         if (err)
901                 return err;
902
903         /* Check that we have conntrack original direction tuple metadata only
904          * for packets for which it makes sense.  Otherwise the key may be
905          * corrupted due to overlapping key fields.
906          */
907         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
908             key->eth.type != htons(ETH_P_IP))
909                 return -EINVAL;
910         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
911             (key->eth.type != htons(ETH_P_IPV6) ||
912              sw_flow_key_is_nd(key)))
913                 return -EINVAL;
914
915         return 0;
916 }