datapath: IP fragments should include L4 header in flow length.
[openvswitch] / datapath / flow.c
1 /*
2  * Distributed under the terms of the GNU GPL version 2.
3  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
4  *
5  * Significant portions of this file may be copied from parts of the Linux
6  * kernel, by Linus Torvalds and others.
7  */
8
9 #include "flow.h"
10 #include "datapath.h"
11 #include <asm/uaccess.h>
12 #include <linux/netdevice.h>
13 #include <linux/etherdevice.h>
14 #include <linux/if_ether.h>
15 #include <linux/if_vlan.h>
16 #include <net/llc_pdu.h>
17 #include <linux/kernel.h>
18 #include <linux/jhash.h>
19 #include <linux/jiffies.h>
20 #include <linux/llc.h>
21 #include <linux/module.h>
22 #include <linux/in.h>
23 #include <linux/rcupdate.h>
24 #include <linux/if_arp.h>
25 #include <linux/if_ether.h>
26 #include <linux/ip.h>
27 #include <linux/ipv6.h>
28 #include <linux/tcp.h>
29 #include <linux/udp.h>
30 #include <linux/icmp.h>
31 #include <linux/icmpv6.h>
32 #include <net/inet_ecn.h>
33 #include <net/ip.h>
34 #include <net/ipv6.h>
35 #include <net/ndisc.h>
36
37 #include "vlan.h"
38
39 static struct kmem_cache *flow_cache;
40 static unsigned int hash_seed __read_mostly;
41
42 static int check_header(struct sk_buff *skb, int len)
43 {
44         if (unlikely(skb->len < len))
45                 return -EINVAL;
46         if (unlikely(!pskb_may_pull(skb, len)))
47                 return -ENOMEM;
48         return 0;
49 }
50
51 static inline bool arphdr_ok(struct sk_buff *skb)
52 {
53         return pskb_may_pull(skb, skb_network_offset(skb) +
54                                   sizeof(struct arp_eth_header));
55 }
56
57 static inline int check_iphdr(struct sk_buff *skb)
58 {
59         unsigned int nh_ofs = skb_network_offset(skb);
60         unsigned int ip_len;
61         int err;
62
63         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
64         if (unlikely(err))
65                 return err;
66
67         ip_len = ip_hdrlen(skb);
68         if (unlikely(ip_len < sizeof(struct iphdr) ||
69                      skb->len < nh_ofs + ip_len))
70                 return -EINVAL;
71
72         skb_set_transport_header(skb, nh_ofs + ip_len);
73         return 0;
74 }
75
76 static inline bool tcphdr_ok(struct sk_buff *skb)
77 {
78         int th_ofs = skb_transport_offset(skb);
79         int tcp_len;
80
81         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
82                 return false;
83
84         tcp_len = tcp_hdrlen(skb);
85         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
86                      skb->len < th_ofs + tcp_len))
87                 return false;
88
89         return true;
90 }
91
92 static inline bool udphdr_ok(struct sk_buff *skb)
93 {
94         return pskb_may_pull(skb, skb_transport_offset(skb) +
95                                   sizeof(struct udphdr));
96 }
97
98 static inline bool icmphdr_ok(struct sk_buff *skb)
99 {
100         return pskb_may_pull(skb, skb_transport_offset(skb) +
101                                   sizeof(struct icmphdr));
102 }
103
104 u64 flow_used_time(unsigned long flow_jiffies)
105 {
106         struct timespec cur_ts;
107         u64 cur_ms, idle_ms;
108
109         ktime_get_ts(&cur_ts);
110         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
111         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
112                  cur_ts.tv_nsec / NSEC_PER_MSEC;
113
114         return cur_ms - idle_ms;
115 }
116
117 #define SW_FLOW_KEY_OFFSET(field)               \
118         offsetof(struct sw_flow_key, field) +   \
119         FIELD_SIZEOF(struct sw_flow_key, field)
120
121 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
122                          int *key_lenp)
123 {
124         unsigned int nh_ofs = skb_network_offset(skb);
125         unsigned int nh_len;
126         int payload_ofs;
127         struct ipv6hdr *nh;
128         uint8_t nexthdr;
129         int err;
130
131         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.addr);
132
133         err = check_header(skb, nh_ofs + sizeof(*nh));
134         if (unlikely(err))
135                 return err;
136
137         nh = ipv6_hdr(skb);
138         nexthdr = nh->nexthdr;
139         payload_ofs = (u8 *)(nh + 1) - skb->data;
140
141         key->ip.nw_proto = NEXTHDR_NONE;
142         key->ip.nw_tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
143         ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
144         ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
145
146         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr);
147         if (unlikely(payload_ofs < 0))
148                 return -EINVAL;
149
150         nh_len = payload_ofs - nh_ofs;
151         skb_set_transport_header(skb, nh_ofs + nh_len);
152         key->ip.nw_proto = nexthdr;
153         return nh_len;
154 }
155
156 static bool icmp6hdr_ok(struct sk_buff *skb)
157 {
158         return pskb_may_pull(skb, skb_transport_offset(skb) +
159                                   sizeof(struct icmp6hdr));
160 }
161
162 #define TCP_FLAGS_OFFSET 13
163 #define TCP_FLAG_MASK 0x3f
164
165 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
166 {
167         u8 tcp_flags = 0;
168
169         if (flow->key.eth.type == htons(ETH_P_IP) &&
170             flow->key.ip.nw_proto == IPPROTO_TCP) {
171                 u8 *tcp = (u8 *)tcp_hdr(skb);
172                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
173         }
174
175         spin_lock_bh(&flow->lock);
176         flow->used = jiffies;
177         flow->packet_count++;
178         flow->byte_count += skb->len;
179         flow->tcp_flags |= tcp_flags;
180         spin_unlock_bh(&flow->lock);
181 }
182
183 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
184 {
185         int actions_len = nla_len(actions);
186         struct sw_flow_actions *sfa;
187
188         /* At least DP_MAX_PORTS actions are required to be able to flood a
189          * packet to every port.  Factor of 2 allows for setting VLAN tags,
190          * etc. */
191         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
192                 return ERR_PTR(-EINVAL);
193
194         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
195         if (!sfa)
196                 return ERR_PTR(-ENOMEM);
197
198         sfa->actions_len = actions_len;
199         memcpy(sfa->actions, nla_data(actions), actions_len);
200         return sfa;
201 }
202
203 struct sw_flow *flow_alloc(void)
204 {
205         struct sw_flow *flow;
206
207         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
208         if (!flow)
209                 return ERR_PTR(-ENOMEM);
210
211         spin_lock_init(&flow->lock);
212         atomic_set(&flow->refcnt, 1);
213         flow->sf_acts = NULL;
214         flow->dead = false;
215
216         return flow;
217 }
218
219 void flow_free_tbl(struct tbl_node *node)
220 {
221         struct sw_flow *flow = flow_cast(node);
222
223         flow->dead = true;
224         flow_put(flow);
225 }
226
227 /* RCU callback used by flow_deferred_free. */
228 static void rcu_free_flow_callback(struct rcu_head *rcu)
229 {
230         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
231
232         flow->dead = true;
233         flow_put(flow);
234 }
235
236 /* Schedules 'flow' to be freed after the next RCU grace period.
237  * The caller must hold rcu_read_lock for this to be sensible. */
238 void flow_deferred_free(struct sw_flow *flow)
239 {
240         call_rcu(&flow->rcu, rcu_free_flow_callback);
241 }
242
243 void flow_hold(struct sw_flow *flow)
244 {
245         atomic_inc(&flow->refcnt);
246 }
247
248 void flow_put(struct sw_flow *flow)
249 {
250         if (unlikely(!flow))
251                 return;
252
253         if (atomic_dec_and_test(&flow->refcnt)) {
254                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
255                 kmem_cache_free(flow_cache, flow);
256         }
257 }
258
259 /* RCU callback used by flow_deferred_free_acts. */
260 static void rcu_free_acts_callback(struct rcu_head *rcu)
261 {
262         struct sw_flow_actions *sf_acts = container_of(rcu,
263                         struct sw_flow_actions, rcu);
264         kfree(sf_acts);
265 }
266
267 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
268  * The caller must hold rcu_read_lock for this to be sensible. */
269 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
270 {
271         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
272 }
273
274 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
275 {
276         struct qtag_prefix {
277                 __be16 eth_type; /* ETH_P_8021Q */
278                 __be16 tci;
279         };
280         struct qtag_prefix *qp;
281
282         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
283                                          sizeof(__be16))))
284                 return -ENOMEM;
285
286         qp = (struct qtag_prefix *) skb->data;
287         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
288         __skb_pull(skb, sizeof(struct qtag_prefix));
289
290         return 0;
291 }
292
293 static __be16 parse_ethertype(struct sk_buff *skb)
294 {
295         struct llc_snap_hdr {
296                 u8  dsap;  /* Always 0xAA */
297                 u8  ssap;  /* Always 0xAA */
298                 u8  ctrl;
299                 u8  oui[3];
300                 __be16 ethertype;
301         };
302         struct llc_snap_hdr *llc;
303         __be16 proto;
304
305         proto = *(__be16 *) skb->data;
306         __skb_pull(skb, sizeof(__be16));
307
308         if (ntohs(proto) >= 1536)
309                 return proto;
310
311         if (skb->len < sizeof(struct llc_snap_hdr))
312                 return htons(ETH_P_802_2);
313
314         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
315                 return htons(0);
316
317         llc = (struct llc_snap_hdr *) skb->data;
318         if (llc->dsap != LLC_SAP_SNAP ||
319             llc->ssap != LLC_SAP_SNAP ||
320             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
321                 return htons(ETH_P_802_2);
322
323         __skb_pull(skb, sizeof(struct llc_snap_hdr));
324         return llc->ethertype;
325 }
326
327 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
328                         int *key_lenp, int nh_len)
329 {
330         struct icmp6hdr *icmp = icmp6_hdr(skb);
331         int error = 0;
332         int key_len;
333
334         /* The ICMPv6 type and code fields use the 16-bit transport port
335          * fields, so we need to store them in 16-bit network byte order.
336          */
337         key->ipv6.tp.src = htons(icmp->icmp6_type);
338         key->ipv6.tp.dst = htons(icmp->icmp6_code);
339         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
340
341         if (icmp->icmp6_code == 0 &&
342             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
343              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
344                 int icmp_len = skb->len - skb_transport_offset(skb);
345                 struct nd_msg *nd;
346                 int offset;
347
348                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
349
350                 /* In order to process neighbor discovery options, we need the
351                  * entire packet.
352                  */
353                 if (unlikely(icmp_len < sizeof(*nd)))
354                         goto out;
355                 if (unlikely(skb_linearize(skb))) {
356                         error = -ENOMEM;
357                         goto out;
358                 }
359
360                 nd = (struct nd_msg *)skb_transport_header(skb);
361                 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
362                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
363
364                 icmp_len -= sizeof(*nd);
365                 offset = 0;
366                 while (icmp_len >= 8) {
367                         struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
368                         int opt_len = nd_opt->nd_opt_len * 8;
369
370                         if (unlikely(!opt_len || opt_len > icmp_len))
371                                 goto invalid;
372
373                         /* Store the link layer address if the appropriate
374                          * option is provided.  It is considered an error if
375                          * the same link layer option is specified twice.
376                          */
377                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
378                             && opt_len == 8) {
379                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
380                                         goto invalid;
381                                 memcpy(key->ipv6.nd.sll,
382                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
383                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
384                                    && opt_len == 8) {
385                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
386                                         goto invalid;
387                                 memcpy(key->ipv6.nd.tll,
388                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
389                         }
390
391                         icmp_len -= opt_len;
392                         offset += opt_len;
393                 }
394         }
395
396         goto out;
397
398 invalid:
399         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
400         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
401         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
402
403 out:
404         *key_lenp = key_len;
405         return error;
406 }
407
408 /**
409  * flow_extract - extracts a flow key from an Ethernet frame.
410  * @skb: sk_buff that contains the frame, with skb->data pointing to the
411  * Ethernet header
412  * @in_port: port number on which @skb was received.
413  * @key: output flow key
414  * @key_lenp: length of output flow key
415  * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
416  * not contain an IPv4 packet or if it is not a fragment.
417  *
418  * The caller must ensure that skb->len >= ETH_HLEN.
419  *
420  * Returns 0 if successful, otherwise a negative errno value.
421  *
422  * Initializes @skb header pointers as follows:
423  *
424  *    - skb->mac_header: the Ethernet header.
425  *
426  *    - skb->network_header: just past the Ethernet header, or just past the
427  *      VLAN header, to the first byte of the Ethernet payload.
428  *
429  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
430  *      on output, then just past the IP header, if one is present and
431  *      of a correct length, otherwise the same as skb->network_header.
432  *      For other key->dl_type values it is left untouched.
433  */
434 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
435                  int *key_lenp, bool *is_frag)
436 {
437         int error = 0;
438         int key_len = SW_FLOW_KEY_OFFSET(eth);
439         struct ethhdr *eth;
440
441         memset(key, 0, sizeof(*key));
442         key->eth.tun_id = OVS_CB(skb)->tun_id;
443         key->eth.in_port = in_port;
444         *is_frag = false;
445
446         skb_reset_mac_header(skb);
447
448         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
449          * header in the linear data area.
450          */
451         eth = eth_hdr(skb);
452         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
453         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
454
455         __skb_pull(skb, 2 * ETH_ALEN);
456
457         if (vlan_tx_tag_present(skb))
458                 key->eth.tci = htons(vlan_get_tci(skb));
459         else if (eth->h_proto == htons(ETH_P_8021Q))
460                 if (unlikely(parse_vlan(skb, key)))
461                         return -ENOMEM;
462
463         key->eth.type = parse_ethertype(skb);
464         if (unlikely(key->eth.type == htons(0)))
465                 return -ENOMEM;
466
467         skb_reset_network_header(skb);
468         __skb_push(skb, skb->data - skb_mac_header(skb));
469
470         /* Network layer. */
471         if (key->eth.type == htons(ETH_P_IP)) {
472                 struct iphdr *nh;
473
474                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
475
476                 error = check_iphdr(skb);
477                 if (unlikely(error)) {
478                         if (error == -EINVAL) {
479                                 skb->transport_header = skb->network_header;
480                                 error = 0;
481                         }
482                         goto out;
483                 }
484
485                 nh = ip_hdr(skb);
486                 key->ipv4.addr.src = nh->saddr;
487                 key->ipv4.addr.dst = nh->daddr;
488                 key->ip.nw_tos = nh->tos & ~INET_ECN_MASK;
489                 key->ip.nw_proto = nh->protocol;
490
491                 /* Transport layer. */
492                 if ((nh->frag_off & htons(IP_MF | IP_OFFSET)) ||
493                     (skb_shinfo(skb)->gso_type & SKB_GSO_UDP))
494                         *is_frag = true;
495
496                 if (key->ip.nw_proto == IPPROTO_TCP) {
497                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
498                         if (!*is_frag && tcphdr_ok(skb)) {
499                                 struct tcphdr *tcp = tcp_hdr(skb);
500                                 key->ipv4.tp.src = tcp->source;
501                                 key->ipv4.tp.dst = tcp->dest;
502                         }
503                 } else if (key->ip.nw_proto == IPPROTO_UDP) {
504                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
505                         if (!*is_frag && udphdr_ok(skb)) {
506                                 struct udphdr *udp = udp_hdr(skb);
507                                 key->ipv4.tp.src = udp->source;
508                                 key->ipv4.tp.dst = udp->dest;
509                         }
510                 } else if (key->ip.nw_proto == IPPROTO_ICMP) {
511                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
512                         if (!*is_frag && icmphdr_ok(skb)) {
513                                 struct icmphdr *icmp = icmp_hdr(skb);
514                                 /* The ICMP type and code fields use the 16-bit
515                                  * transport port fields, so we need to store them
516                                  * in 16-bit network byte order. */
517                                 key->ipv4.tp.src = htons(icmp->type);
518                                 key->ipv4.tp.dst = htons(icmp->code);
519                         }
520                 }
521
522         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
523                 struct arp_eth_header *arp;
524
525                 arp = (struct arp_eth_header *)skb_network_header(skb);
526
527                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
528                                 && arp->ar_pro == htons(ETH_P_IP)
529                                 && arp->ar_hln == ETH_ALEN
530                                 && arp->ar_pln == 4) {
531
532                         /* We only match on the lower 8 bits of the opcode. */
533                         if (ntohs(arp->ar_op) <= 0xff)
534                                 key->ip.nw_proto = ntohs(arp->ar_op);
535
536                         if (key->ip.nw_proto == ARPOP_REQUEST
537                                         || key->ip.nw_proto == ARPOP_REPLY) {
538                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
539                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
540                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
541                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
542                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
543                         }
544                 }
545         } else if (key->eth.type == htons(ETH_P_IPV6)) {
546                 int nh_len;             /* IPv6 Header + Extensions */
547
548                 nh_len = parse_ipv6hdr(skb, key, &key_len);
549                 if (unlikely(nh_len < 0)) {
550                         if (nh_len == -EINVAL)
551                                 skb->transport_header = skb->network_header;
552                         else
553                                 error = nh_len;
554                         goto out;
555                 }
556
557                 /* Transport layer. */
558                 if (key->ip.nw_proto == NEXTHDR_TCP) {
559                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
560                         if (tcphdr_ok(skb)) {
561                                 struct tcphdr *tcp = tcp_hdr(skb);
562                                 key->ipv6.tp.src = tcp->source;
563                                 key->ipv6.tp.dst = tcp->dest;
564                         }
565                 } else if (key->ip.nw_proto == NEXTHDR_UDP) {
566                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
567                         if (udphdr_ok(skb)) {
568                                 struct udphdr *udp = udp_hdr(skb);
569                                 key->ipv6.tp.src = udp->source;
570                                 key->ipv6.tp.dst = udp->dest;
571                         }
572                 } else if (key->ip.nw_proto == NEXTHDR_ICMP) {
573                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
574                         if (icmp6hdr_ok(skb)) {
575                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
576                                 if (error < 0)
577                                         goto out;
578                         }
579                 }
580         }
581
582 out:
583         *key_lenp = key_len;
584         return error;
585 }
586
587 u32 flow_hash(const struct sw_flow_key *key, int key_len)
588 {
589         return jhash2((u32*)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
590 }
591
592 int flow_cmp(const struct tbl_node *node, void *key2_, int len)
593 {
594         const struct sw_flow_key *key1 = &flow_cast(node)->key;
595         const struct sw_flow_key *key2 = key2_;
596
597         return !memcmp(key1, key2, len);
598 }
599
600 /* The size of the argument for each %ODP_KEY_ATTR_* Netlink attribute.  */
601 static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = {
602         [ODP_KEY_ATTR_TUN_ID] = 8,
603         [ODP_KEY_ATTR_IN_PORT] = 4,
604         [ODP_KEY_ATTR_ETHERNET] = sizeof(struct odp_key_ethernet),
605         [ODP_KEY_ATTR_8021Q] = sizeof(struct odp_key_8021q),
606         [ODP_KEY_ATTR_ETHERTYPE] = 2,
607         [ODP_KEY_ATTR_IPV4] = sizeof(struct odp_key_ipv4),
608         [ODP_KEY_ATTR_IPV6] = sizeof(struct odp_key_ipv6),
609         [ODP_KEY_ATTR_TCP] = sizeof(struct odp_key_tcp),
610         [ODP_KEY_ATTR_UDP] = sizeof(struct odp_key_udp),
611         [ODP_KEY_ATTR_ICMP] = sizeof(struct odp_key_icmp),
612         [ODP_KEY_ATTR_ICMPV6] = sizeof(struct odp_key_icmpv6),
613         [ODP_KEY_ATTR_ARP] = sizeof(struct odp_key_arp),
614         [ODP_KEY_ATTR_ND] = sizeof(struct odp_key_nd),
615 };
616
617 /**
618  * flow_from_nlattrs - parses Netlink attributes into a flow key.
619  * @swkey: receives the extracted flow key.
620  * @key_lenp: number of bytes used in @swkey.
621  * @attr: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
622  * sequence.
623  *
624  * This state machine accepts the following forms, with [] for optional
625  * elements and | for alternatives:
626  *
627  * [tun_id] in_port ethernet [8021q] [ethertype \
628  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
629  */
630 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
631                       const struct nlattr *attr)
632 {
633         int error = 0;
634         const struct nlattr *nla;
635         u16 prev_type;
636         int rem;
637         int key_len;
638
639         memset(swkey, 0, sizeof(*swkey));
640         swkey->eth.type = htons(ETH_P_802_2);
641         key_len = SW_FLOW_KEY_OFFSET(eth);
642
643         prev_type = ODP_KEY_ATTR_UNSPEC;
644         nla_for_each_nested(nla, attr, rem) {
645                 const struct odp_key_ethernet *eth_key;
646                 const struct odp_key_8021q *q_key;
647                 const struct odp_key_ipv4 *ipv4_key;
648                 const struct odp_key_ipv6 *ipv6_key;
649                 const struct odp_key_tcp *tcp_key;
650                 const struct odp_key_udp *udp_key;
651                 const struct odp_key_icmp *icmp_key;
652                 const struct odp_key_icmpv6 *icmpv6_key;
653                 const struct odp_key_arp *arp_key;
654                 const struct odp_key_nd *nd_key;
655
656                 int type = nla_type(nla);
657
658                 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
659                         goto invalid;
660
661 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
662                 switch (TRANSITION(prev_type, type)) {
663                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
664                         swkey->eth.tun_id = nla_get_be64(nla);
665                         break;
666
667                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
668                 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
669                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
670                                 goto invalid;
671                         swkey->eth.in_port = nla_get_u32(nla);
672                         break;
673
674                 case TRANSITION(ODP_KEY_ATTR_IN_PORT, ODP_KEY_ATTR_ETHERNET):
675                         eth_key = nla_data(nla);
676                         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
677                         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
678                         break;
679
680                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_8021Q):
681                         q_key = nla_data(nla);
682                         /* Only standard 0x8100 VLANs currently supported. */
683                         if (q_key->q_tpid != htons(ETH_P_8021Q))
684                                 goto invalid;
685                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
686                                 goto invalid;
687                         swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
688                         break;
689
690                 case TRANSITION(ODP_KEY_ATTR_8021Q, ODP_KEY_ATTR_ETHERTYPE):
691                 case TRANSITION(ODP_KEY_ATTR_ETHERNET, ODP_KEY_ATTR_ETHERTYPE):
692                         swkey->eth.type = nla_get_be16(nla);
693                         if (ntohs(swkey->eth.type) < 1536)
694                                 goto invalid;
695                         break;
696
697                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV4):
698                         key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
699                         if (swkey->eth.type != htons(ETH_P_IP))
700                                 goto invalid;
701                         ipv4_key = nla_data(nla);
702                         swkey->ip.nw_proto = ipv4_key->ipv4_proto;
703                         swkey->ip.nw_tos = ipv4_key->ipv4_tos;
704                         swkey->ipv4.addr.src = ipv4_key->ipv4_src;
705                         swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
706                         if (swkey->ip.nw_tos & INET_ECN_MASK)
707                                 goto invalid;
708                         break;
709
710                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_IPV6):
711                         key_len = SW_FLOW_KEY_OFFSET(ipv6.addr);
712                         if (swkey->eth.type != htons(ETH_P_IPV6))
713                                 goto invalid;
714                         ipv6_key = nla_data(nla);
715                         swkey->ip.nw_proto = ipv6_key->ipv6_proto;
716                         swkey->ip.nw_tos = ipv6_key->ipv6_tos;
717                         memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
718                                         sizeof(swkey->ipv6.addr.src));
719                         memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
720                                         sizeof(swkey->ipv6.addr.dst));
721                         if (swkey->ip.nw_tos & INET_ECN_MASK)
722                                 goto invalid;
723                         break;
724
725                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_TCP):
726                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
727                         if (swkey->ip.nw_proto != IPPROTO_TCP)
728                                 goto invalid;
729                         tcp_key = nla_data(nla);
730                         swkey->ipv4.tp.src = tcp_key->tcp_src;
731                         swkey->ipv4.tp.dst = tcp_key->tcp_dst;
732                         break;
733
734                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_TCP):
735                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
736                         if (swkey->ip.nw_proto != IPPROTO_TCP)
737                                 goto invalid;
738                         tcp_key = nla_data(nla);
739                         swkey->ipv6.tp.src = tcp_key->tcp_src;
740                         swkey->ipv6.tp.dst = tcp_key->tcp_dst;
741                         break;
742
743                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_UDP):
744                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
745                         if (swkey->ip.nw_proto != IPPROTO_UDP)
746                                 goto invalid;
747                         udp_key = nla_data(nla);
748                         swkey->ipv4.tp.src = udp_key->udp_src;
749                         swkey->ipv4.tp.dst = udp_key->udp_dst;
750                         break;
751
752                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_UDP):
753                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
754                         if (swkey->ip.nw_proto != IPPROTO_UDP)
755                                 goto invalid;
756                         udp_key = nla_data(nla);
757                         swkey->ipv6.tp.src = udp_key->udp_src;
758                         swkey->ipv6.tp.dst = udp_key->udp_dst;
759                         break;
760
761                 case TRANSITION(ODP_KEY_ATTR_IPV4, ODP_KEY_ATTR_ICMP):
762                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
763                         if (swkey->ip.nw_proto != IPPROTO_ICMP)
764                                 goto invalid;
765                         icmp_key = nla_data(nla);
766                         swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
767                         swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
768                         break;
769
770                 case TRANSITION(ODP_KEY_ATTR_IPV6, ODP_KEY_ATTR_ICMPV6):
771                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
772                         if (swkey->ip.nw_proto != IPPROTO_ICMPV6)
773                                 goto invalid;
774                         icmpv6_key = nla_data(nla);
775                         swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
776                         swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
777                         break;
778
779                 case TRANSITION(ODP_KEY_ATTR_ETHERTYPE, ODP_KEY_ATTR_ARP):
780                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
781                         if (swkey->eth.type != htons(ETH_P_ARP))
782                                 goto invalid;
783                         arp_key = nla_data(nla);
784                         swkey->ipv4.addr.src = arp_key->arp_sip;
785                         swkey->ipv4.addr.dst = arp_key->arp_tip;
786                         if (arp_key->arp_op & htons(0xff00))
787                                 goto invalid;
788                         swkey->ip.nw_proto = ntohs(arp_key->arp_op);
789                         memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
790                         memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
791                         break;
792
793                 case TRANSITION(ODP_KEY_ATTR_ICMPV6, ODP_KEY_ATTR_ND):
794                         key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
795                         if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
796                             && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
797                                 goto invalid;
798                         nd_key = nla_data(nla);
799                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
800                                         sizeof(swkey->ipv6.nd.target));
801                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
802                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
803                         break;
804
805                 default:
806                         goto invalid;
807                 }
808
809                 prev_type = type;
810         }
811         if (rem)
812                 goto invalid;
813
814         switch (prev_type) {
815         case ODP_KEY_ATTR_UNSPEC:
816                 goto invalid;
817
818         case ODP_KEY_ATTR_TUN_ID:
819         case ODP_KEY_ATTR_IN_PORT:
820                 goto invalid;
821
822         case ODP_KEY_ATTR_ETHERNET:
823         case ODP_KEY_ATTR_8021Q:
824                 goto ok;
825
826         case ODP_KEY_ATTR_ETHERTYPE:
827                 if (swkey->eth.type == htons(ETH_P_IP) ||
828                     swkey->eth.type == htons(ETH_P_ARP))
829                         goto invalid;
830                 goto ok;
831
832         case ODP_KEY_ATTR_IPV4:
833                 if (swkey->ip.nw_proto == IPPROTO_TCP ||
834                     swkey->ip.nw_proto == IPPROTO_UDP ||
835                     swkey->ip.nw_proto == IPPROTO_ICMP)
836                         goto invalid;
837                 goto ok;
838
839         case ODP_KEY_ATTR_IPV6:
840                 if (swkey->ip.nw_proto == IPPROTO_TCP ||
841                     swkey->ip.nw_proto == IPPROTO_UDP ||
842                     swkey->ip.nw_proto == IPPROTO_ICMPV6)
843                         goto invalid;
844                 goto ok;
845
846         case ODP_KEY_ATTR_ICMPV6:
847                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
848                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
849                         goto invalid;
850                 goto ok;
851
852         case ODP_KEY_ATTR_TCP:
853         case ODP_KEY_ATTR_UDP:
854         case ODP_KEY_ATTR_ICMP:
855         case ODP_KEY_ATTR_ARP:
856         case ODP_KEY_ATTR_ND:
857                 goto ok;
858
859         default:
860                 WARN_ON_ONCE(1);
861         }
862
863 invalid:
864         error = -EINVAL;
865
866 ok:
867         WARN_ON_ONCE(!key_len && !error);
868         *key_lenp = key_len;
869         return error;
870 }
871
872 /**
873  * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
874  * @in_port: receives the extracted input port.
875  * @tun_id: receives the extracted tunnel ID.
876  * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute
877  * sequence.
878  *
879  * This parses a series of Netlink attributes that form a flow key, which must
880  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
881  * get the metadata, that is, the parts of the flow key that cannot be
882  * extracted from the packet itself.
883  */
884 int flow_metadata_from_nlattrs(u16 *in_port, __be64 *tun_id,
885                                const struct nlattr *attr)
886 {
887         const struct nlattr *nla;
888         u16 prev_type;
889         int rem;
890
891         *tun_id = 0;
892
893         prev_type = ODP_KEY_ATTR_UNSPEC;
894         nla_for_each_nested(nla, attr, rem) {
895                 int type = nla_type(nla);
896
897                 if (type > ODP_KEY_ATTR_MAX || nla_len(nla) != key_lens[type])
898                         return -EINVAL;
899
900                 switch (TRANSITION(prev_type, type)) {
901                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_TUN_ID):
902                         *tun_id = nla_get_be64(nla);
903                         break;
904
905                 case TRANSITION(ODP_KEY_ATTR_UNSPEC, ODP_KEY_ATTR_IN_PORT):
906                 case TRANSITION(ODP_KEY_ATTR_TUN_ID, ODP_KEY_ATTR_IN_PORT):
907                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
908                                 return -EINVAL;
909                         *in_port = nla_get_u32(nla);
910                         break;
911
912                 default:
913                         goto done;
914                 }
915
916                 prev_type = type;
917         }
918         if (rem)
919                 return -EINVAL;
920
921 done:
922         if (prev_type == ODP_KEY_ATTR_UNSPEC ||
923             prev_type == ODP_KEY_ATTR_TUN_ID)
924                 return -EINVAL;
925         return 0;
926 }
927
928 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
929 {
930         struct odp_key_ethernet *eth_key;
931         struct nlattr *nla;
932
933         /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
934          * to be updated, but will at least raise awareness when new ODP key
935          * types are added. */
936         BUILD_BUG_ON(__ODP_KEY_ATTR_MAX != 14);
937
938         if (swkey->eth.tun_id != cpu_to_be64(0))
939                 NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->eth.tun_id);
940
941         NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->eth.in_port);
942
943         nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key));
944         if (!nla)
945                 goto nla_put_failure;
946         eth_key = nla_data(nla);
947         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
948         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
949
950         if (swkey->eth.tci != htons(0)) {
951                 struct odp_key_8021q q_key;
952
953                 q_key.q_tpid = htons(ETH_P_8021Q);
954                 q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
955                 NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
956         }
957
958         if (swkey->eth.type == htons(ETH_P_802_2))
959                 return 0;
960
961         NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->eth.type);
962
963         if (swkey->eth.type == htons(ETH_P_IP)) {
964                 struct odp_key_ipv4 *ipv4_key;
965
966                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key));
967                 if (!nla)
968                         goto nla_put_failure;
969                 ipv4_key = nla_data(nla);
970                 memset(ipv4_key, 0, sizeof(struct odp_key_ipv4));
971                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
972                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
973                 ipv4_key->ipv4_proto = swkey->ip.nw_proto;
974                 ipv4_key->ipv4_tos = swkey->ip.nw_tos;
975         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
976                 struct odp_key_ipv6 *ipv6_key;
977
978                 nla = nla_reserve(skb, ODP_KEY_ATTR_IPV6, sizeof(*ipv6_key));
979                 if (!nla)
980                         goto nla_put_failure;
981                 ipv6_key = nla_data(nla);
982                 memset(ipv6_key, 0, sizeof(struct odp_key_ipv6));
983                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
984                                 sizeof(ipv6_key->ipv6_src));
985                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
986                                 sizeof(ipv6_key->ipv6_dst));
987                 ipv6_key->ipv6_proto = swkey->ip.nw_proto;
988                 ipv6_key->ipv6_tos = swkey->ip.nw_tos;
989         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
990                 struct odp_key_arp *arp_key;
991
992                 nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key));
993                 if (!nla)
994                         goto nla_put_failure;
995                 arp_key = nla_data(nla);
996                 memset(arp_key, 0, sizeof(struct odp_key_arp));
997                 arp_key->arp_sip = swkey->ipv4.addr.src;
998                 arp_key->arp_tip = swkey->ipv4.addr.dst;
999                 arp_key->arp_op = htons(swkey->ip.nw_proto);
1000                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1001                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1002         }
1003
1004         if (swkey->eth.type == htons(ETH_P_IP) ||
1005             swkey->eth.type == htons(ETH_P_IPV6)) {
1006
1007                 if (swkey->ip.nw_proto == IPPROTO_TCP) {
1008                         struct odp_key_tcp *tcp_key;
1009
1010                         nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key));
1011                         if (!nla)
1012                                 goto nla_put_failure;
1013                         tcp_key = nla_data(nla);
1014                         if (swkey->eth.type == htons(ETH_P_IP)) {
1015                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1016                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1017                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1018                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1019                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1020                         }
1021                 } else if (swkey->ip.nw_proto == IPPROTO_UDP) {
1022                         struct odp_key_udp *udp_key;
1023
1024                         nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key));
1025                         if (!nla)
1026                                 goto nla_put_failure;
1027                         udp_key = nla_data(nla);
1028                         if (swkey->eth.type == htons(ETH_P_IP)) {
1029                                 udp_key->udp_src = swkey->ipv4.tp.src;
1030                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1031                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1032                                 udp_key->udp_src = swkey->ipv6.tp.src;
1033                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1034                         }
1035                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1036                            swkey->ip.nw_proto == IPPROTO_ICMP) {
1037                         struct odp_key_icmp *icmp_key;
1038
1039                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key));
1040                         if (!nla)
1041                                 goto nla_put_failure;
1042                         icmp_key = nla_data(nla);
1043                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1044                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1045                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1046                            swkey->ip.nw_proto == IPPROTO_ICMPV6) {
1047                         struct odp_key_icmpv6 *icmpv6_key;
1048
1049                         nla = nla_reserve(skb, ODP_KEY_ATTR_ICMPV6,
1050                                                 sizeof(*icmpv6_key));
1051                         if (!nla)
1052                                 goto nla_put_failure;
1053                         icmpv6_key = nla_data(nla);
1054                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1055                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1056
1057                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1058                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1059                                 struct odp_key_nd *nd_key;
1060
1061                                 nla = nla_reserve(skb, ODP_KEY_ATTR_ND, sizeof(*nd_key));
1062                                 if (!nla)
1063                                         goto nla_put_failure;
1064                                 nd_key = nla_data(nla);
1065                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1066                                                         sizeof(nd_key->nd_target));
1067                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1068                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1069                         }
1070                 }
1071         }
1072
1073         return 0;
1074
1075 nla_put_failure:
1076         return -EMSGSIZE;
1077 }
1078
1079 /* Initializes the flow module.
1080  * Returns zero if successful or a negative error code. */
1081 int flow_init(void)
1082 {
1083         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1084                                         0, NULL);
1085         if (flow_cache == NULL)
1086                 return -ENOMEM;
1087
1088         get_random_bytes(&hash_seed, sizeof(hash_seed));
1089
1090         return 0;
1091 }
1092
1093 /* Uninitializes the flow module. */
1094 void flow_exit(void)
1095 {
1096         kmem_cache_destroy(flow_cache);
1097 }