datapath: Change userspace vport interface to use Netlink attributes.
[sliver-openvswitch.git] / datapath / tunnel.c
1 /*
2  * Copyright (c) 2010, 2011 Nicira Networks.
3  * Distributed under the terms of the GNU GPL version 2.
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 <linux/if_arp.h>
10 #include <linux/if_ether.h>
11 #include <linux/ip.h>
12 #include <linux/if_vlan.h>
13 #include <linux/in.h>
14 #include <linux/in_route.h>
15 #include <linux/jhash.h>
16 #include <linux/kernel.h>
17 #include <linux/version.h>
18 #include <linux/workqueue.h>
19
20 #include <net/dsfield.h>
21 #include <net/dst.h>
22 #include <net/icmp.h>
23 #include <net/inet_ecn.h>
24 #include <net/ip.h>
25 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
26 #include <net/ipv6.h>
27 #endif
28 #include <net/route.h>
29 #include <net/xfrm.h>
30
31 #include "actions.h"
32 #include "checksum.h"
33 #include "datapath.h"
34 #include "table.h"
35 #include "tunnel.h"
36 #include "vport.h"
37 #include "vport-generic.h"
38 #include "vport-internal_dev.h"
39
40 #ifdef NEED_CACHE_TIMEOUT
41 /*
42  * On kernels where we can't quickly detect changes in the rest of the system
43  * we use an expiration time to invalidate the cache.  A shorter expiration
44  * reduces the length of time that we may potentially blackhole packets while
45  * a longer time increases performance by reducing the frequency that the
46  * cache needs to be rebuilt.  A variety of factors may cause the cache to be
47  * invalidated before the expiration time but this is the maximum.  The time
48  * is expressed in jiffies.
49  */
50 #define MAX_CACHE_EXP HZ
51 #endif
52
53 /*
54  * Interval to check for and remove caches that are no longer valid.  Caches
55  * are checked for validity before they are used for packet encapsulation and
56  * old caches are removed at that time.  However, if no packets are sent through
57  * the tunnel then the cache will never be destroyed.  Since it holds
58  * references to a number of system objects, the cache will continue to use
59  * system resources by not allowing those objects to be destroyed.  The cache
60  * cleaner is periodically run to free invalid caches.  It does not
61  * significantly affect system performance.  A lower interval will release
62  * resources faster but will itself consume resources by requiring more frequent
63  * checks.  A longer interval may result in messages being printed to the kernel
64  * message buffer about unreleased resources.  The interval is expressed in
65  * jiffies.
66  */
67 #define CACHE_CLEANER_INTERVAL (5 * HZ)
68
69 #define CACHE_DATA_ALIGN 16
70
71 static struct tbl __rcu *port_table __read_mostly;
72
73 static void cache_cleaner(struct work_struct *work);
74 static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
75
76 /*
77  * These are just used as an optimization: they don't require any kind of
78  * synchronization because we could have just as easily read the value before
79  * the port change happened.
80  */
81 static unsigned int key_local_remote_ports __read_mostly;
82 static unsigned int key_remote_ports __read_mostly;
83 static unsigned int local_remote_ports __read_mostly;
84 static unsigned int remote_ports __read_mostly;
85
86 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
87 #define rt_dst(rt) (rt->dst)
88 #else
89 #define rt_dst(rt) (rt->u.dst)
90 #endif
91
92 static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
93 {
94         return vport_from_priv(tnl_vport);
95 }
96
97 static inline struct tnl_vport *tnl_vport_table_cast(const struct tbl_node *node)
98 {
99         return container_of(node, struct tnl_vport, tbl_node);
100 }
101
102 /* This is analogous to rtnl_dereference for the tunnel cache.  It checks that
103  * cache_lock is held, so it is only for update side code.
104  */
105 static inline struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
106 {
107         return rcu_dereference_protected(tnl_vport->cache,
108                                          lockdep_is_held(&tnl_vport->cache_lock));
109 }
110
111 static inline void schedule_cache_cleaner(void)
112 {
113         schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
114 }
115
116 static void free_cache(struct tnl_cache *cache)
117 {
118         if (!cache)
119                 return;
120
121         flow_put(cache->flow);
122         ip_rt_put(cache->rt);
123         kfree(cache);
124 }
125
126 static void free_config_rcu(struct rcu_head *rcu)
127 {
128         struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
129         kfree(c);
130 }
131
132 static void free_cache_rcu(struct rcu_head *rcu)
133 {
134         struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
135         free_cache(c);
136 }
137
138 static void assign_config_rcu(struct vport *vport,
139                               struct tnl_mutable_config *new_config)
140 {
141         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
142         struct tnl_mutable_config *old_config;
143
144         old_config = rtnl_dereference(tnl_vport->mutable);
145         rcu_assign_pointer(tnl_vport->mutable, new_config);
146         call_rcu(&old_config->rcu, free_config_rcu);
147 }
148
149 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
150 {
151         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
152         struct tnl_cache *old_cache;
153
154         old_cache = cache_dereference(tnl_vport);
155         rcu_assign_pointer(tnl_vport->cache, new_cache);
156
157         if (old_cache)
158                 call_rcu(&old_cache->rcu, free_cache_rcu);
159 }
160
161 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
162 {
163         if (mutable->flags & TNL_F_IN_KEY_MATCH) {
164                 if (mutable->saddr)
165                         return &local_remote_ports;
166                 else
167                         return &remote_ports;
168         } else {
169                 if (mutable->saddr)
170                         return &key_local_remote_ports;
171                 else
172                         return &key_remote_ports;
173         }
174 }
175
176 struct port_lookup_key {
177         const struct tnl_mutable_config *mutable;
178         __be64 key;
179         u32 tunnel_type;
180         __be32 saddr;
181         __be32 daddr;
182 };
183
184 /*
185  * Modifies 'target' to store the rcu_dereferenced pointer that was used to do
186  * the comparision.
187  */
188 static int port_cmp(const struct tbl_node *node, void *target)
189 {
190         const struct tnl_vport *tnl_vport = tnl_vport_table_cast(node);
191         struct port_lookup_key *lookup = target;
192
193         lookup->mutable = rcu_dereference_rtnl(tnl_vport->mutable);
194
195         return (lookup->mutable->tunnel_type == lookup->tunnel_type &&
196                 lookup->mutable->daddr == lookup->daddr &&
197                 lookup->mutable->in_key == lookup->key &&
198                 lookup->mutable->saddr == lookup->saddr);
199 }
200
201 static u32 port_hash(struct port_lookup_key *k)
202 {
203         u32 x = jhash_3words((__force u32)k->saddr, (__force u32)k->daddr,
204                              k->tunnel_type, 0);
205         return jhash_2words((__force u64)k->key >> 32, (__force u32)k->key, x);
206 }
207
208 static u32 mutable_hash(const struct tnl_mutable_config *mutable)
209 {
210         struct port_lookup_key lookup;
211
212         lookup.saddr = mutable->saddr;
213         lookup.daddr = mutable->daddr;
214         lookup.key = mutable->in_key;
215         lookup.tunnel_type = mutable->tunnel_type;
216
217         return port_hash(&lookup);
218 }
219
220 static void check_table_empty(void)
221 {
222         struct tbl *old_table = rtnl_dereference(port_table);
223
224         if (tbl_count(old_table) == 0) {
225                 cancel_delayed_work_sync(&cache_cleaner_wq);
226                 rcu_assign_pointer(port_table, NULL);
227                 tbl_deferred_destroy(old_table, NULL);
228         }
229 }
230
231 static int add_port(struct vport *vport)
232 {
233         struct tbl *cur_table = rtnl_dereference(port_table);
234         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
235         int err;
236
237         if (!port_table) {
238                 struct tbl *new_table;
239
240                 new_table = tbl_create(TBL_MIN_BUCKETS);
241                 if (!new_table)
242                         return -ENOMEM;
243
244                 rcu_assign_pointer(port_table, new_table);
245                 schedule_cache_cleaner();
246
247         } else if (tbl_count(cur_table) > tbl_n_buckets(cur_table)) {
248                 struct tbl *new_table;
249
250                 new_table = tbl_expand(cur_table);
251                 if (IS_ERR(new_table))
252                         return PTR_ERR(new_table);
253
254                 rcu_assign_pointer(port_table, new_table);
255                 tbl_deferred_destroy(cur_table, NULL);
256         }
257
258         err = tbl_insert(rtnl_dereference(port_table), &tnl_vport->tbl_node,
259                          mutable_hash(rtnl_dereference(tnl_vport->mutable)));
260         if (err) {
261                 check_table_empty();
262                 return err;
263         }
264
265         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
266
267         return 0;
268 }
269
270 static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable)
271 {
272         int err;
273         struct tbl *cur_table = rtnl_dereference(port_table);
274         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
275         u32 hash;
276
277         hash = mutable_hash(new_mutable);
278         if (hash == tnl_vport->tbl_node.hash)
279                 goto table_updated;
280
281         /*
282          * Ideally we should make this move atomic to avoid having gaps in
283          * finding tunnels or the possibility of failure.  However, if we do
284          * find a tunnel it will always be consistent.
285          */
286         err = tbl_remove(cur_table, &tnl_vport->tbl_node);
287         if (err)
288                 return err;
289
290         err = tbl_insert(cur_table, &tnl_vport->tbl_node, hash);
291         if (err) {
292                 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
293                 check_table_empty();
294                 return err;
295         }
296
297 table_updated:
298         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
299         assign_config_rcu(vport, new_mutable);
300         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
301
302         return 0;
303 }
304
305 static int del_port(struct vport *vport)
306 {
307         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
308         int err;
309
310         err = tbl_remove(rtnl_dereference(port_table), &tnl_vport->tbl_node);
311         if (err)
312                 return err;
313
314         check_table_empty();
315         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
316
317         return 0;
318 }
319
320 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key,
321                             int tunnel_type,
322                             const struct tnl_mutable_config **mutable)
323 {
324         struct port_lookup_key lookup;
325         struct tbl *table = rcu_dereference_rtnl(port_table);
326         struct tbl_node *tbl_node;
327
328         if (unlikely(!table))
329                 return NULL;
330
331         lookup.saddr = saddr;
332         lookup.daddr = daddr;
333
334         if (tunnel_type & TNL_T_KEY_EXACT) {
335                 lookup.key = key;
336                 lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_MATCH;
337
338                 if (key_local_remote_ports) {
339                         tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
340                         if (tbl_node)
341                                 goto found;
342                 }
343
344                 if (key_remote_ports) {
345                         lookup.saddr = 0;
346
347                         tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
348                         if (tbl_node)
349                                 goto found;
350
351                         lookup.saddr = saddr;
352                 }
353         }
354
355         if (tunnel_type & TNL_T_KEY_MATCH) {
356                 lookup.key = 0;
357                 lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_EXACT;
358
359                 if (local_remote_ports) {
360                         tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
361                         if (tbl_node)
362                                 goto found;
363                 }
364
365                 if (remote_ports) {
366                         lookup.saddr = 0;
367
368                         tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
369                         if (tbl_node)
370                                 goto found;
371                 }
372         }
373
374         return NULL;
375
376 found:
377         *mutable = lookup.mutable;
378         return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node));
379 }
380
381 static inline void ecn_decapsulate(struct sk_buff *skb)
382 {
383         /* This is accessing the outer IP header of the tunnel, which we've
384          * already validated to be OK.  skb->data is currently set to the start
385          * of the inner Ethernet header, and we've validated ETH_HLEN.
386          */
387         if (unlikely(INET_ECN_is_ce(ip_hdr(skb)->tos))) {
388                 __be16 protocol = skb->protocol;
389
390                 skb_set_network_header(skb, ETH_HLEN);
391
392                 if (skb->protocol == htons(ETH_P_8021Q)) {
393                         if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
394                                 return;
395
396                         protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
397                         skb_set_network_header(skb, VLAN_ETH_HLEN);
398                 }
399
400                 if (protocol == htons(ETH_P_IP)) {
401                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
402                             + sizeof(struct iphdr))))
403                                 return;
404
405                         IP_ECN_set_ce(ip_hdr(skb));
406                 }
407 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
408                 else if (protocol == htons(ETH_P_IPV6)) {
409                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
410                             + sizeof(struct ipv6hdr))))
411                                 return;
412
413                         IP6_ECN_set_ce(ipv6_hdr(skb));
414                 }
415 #endif
416         }
417 }
418
419 /* Called with rcu_read_lock. */
420 void tnl_rcv(struct vport *vport, struct sk_buff *skb)
421 {
422         /* Packets received by this function are in the following state:
423          * - skb->data points to the inner Ethernet header.
424          * - The inner Ethernet header is in the linear data area.
425          * - skb->csum does not include the inner Ethernet header.
426          * - The layer pointers point at the outer headers.
427          */
428
429         struct ethhdr *eh = (struct ethhdr *)skb->data;
430
431         if (likely(ntohs(eh->h_proto) >= 1536))
432                 skb->protocol = eh->h_proto;
433         else
434                 skb->protocol = htons(ETH_P_802_2);
435
436         skb_dst_drop(skb);
437         nf_reset(skb);
438         secpath_reset(skb);
439
440         ecn_decapsulate(skb);
441         compute_ip_summed(skb, false);
442
443         vport_receive(vport, skb);
444 }
445
446 static bool check_ipv4_address(__be32 addr)
447 {
448         if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
449             || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
450                 return false;
451
452         return true;
453 }
454
455 static bool ipv4_should_icmp(struct sk_buff *skb)
456 {
457         struct iphdr *old_iph = ip_hdr(skb);
458
459         /* Don't respond to L2 broadcast. */
460         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
461                 return false;
462
463         /* Don't respond to L3 broadcast or invalid addresses. */
464         if (!check_ipv4_address(old_iph->daddr) ||
465             !check_ipv4_address(old_iph->saddr))
466                 return false;
467
468         /* Only respond to the first fragment. */
469         if (old_iph->frag_off & htons(IP_OFFSET))
470                 return false;
471
472         /* Don't respond to ICMP error messages. */
473         if (old_iph->protocol == IPPROTO_ICMP) {
474                 u8 icmp_type, *icmp_typep;
475
476                 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
477                                                 (old_iph->ihl << 2) +
478                                                 offsetof(struct icmphdr, type) -
479                                                 skb->data, sizeof(icmp_type),
480                                                 &icmp_type);
481
482                 if (!icmp_typep)
483                         return false;
484
485                 if (*icmp_typep > NR_ICMP_TYPES
486                         || (*icmp_typep <= ICMP_PARAMETERPROB
487                                 && *icmp_typep != ICMP_ECHOREPLY
488                                 && *icmp_typep != ICMP_ECHO))
489                         return false;
490         }
491
492         return true;
493 }
494
495 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
496                             unsigned int mtu, unsigned int payload_length)
497 {
498         struct iphdr *iph, *old_iph = ip_hdr(skb);
499         struct icmphdr *icmph;
500         u8 *payload;
501
502         iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
503         icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
504         payload = skb_put(nskb, payload_length);
505
506         /* IP */
507         iph->version            =       4;
508         iph->ihl                =       sizeof(struct iphdr) >> 2;
509         iph->tos                =       (old_iph->tos & IPTOS_TOS_MASK) |
510                                         IPTOS_PREC_INTERNETCONTROL;
511         iph->tot_len            =       htons(sizeof(struct iphdr)
512                                               + sizeof(struct icmphdr)
513                                               + payload_length);
514         get_random_bytes(&iph->id, sizeof(iph->id));
515         iph->frag_off           =       0;
516         iph->ttl                =       IPDEFTTL;
517         iph->protocol           =       IPPROTO_ICMP;
518         iph->daddr              =       old_iph->saddr;
519         iph->saddr              =       old_iph->daddr;
520
521         ip_send_check(iph);
522
523         /* ICMP */
524         icmph->type             =       ICMP_DEST_UNREACH;
525         icmph->code             =       ICMP_FRAG_NEEDED;
526         icmph->un.gateway       =       htonl(mtu);
527         icmph->checksum         =       0;
528
529         nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
530         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
531                                             payload, payload_length,
532                                             nskb->csum);
533         icmph->checksum = csum_fold(nskb->csum);
534 }
535
536 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
537 static bool ipv6_should_icmp(struct sk_buff *skb)
538 {
539         struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
540         int addr_type;
541         int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
542         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
543
544         /* Check source address is valid. */
545         addr_type = ipv6_addr_type(&old_ipv6h->saddr);
546         if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
547                 return false;
548
549         /* Don't reply to unspecified addresses. */
550         if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
551                 return false;
552
553         /* Don't respond to ICMP error messages. */
554         payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
555         if (payload_off < 0)
556                 return false;
557
558         if (nexthdr == NEXTHDR_ICMP) {
559                 u8 icmp_type, *icmp_typep;
560
561                 icmp_typep = skb_header_pointer(skb, payload_off +
562                                                 offsetof(struct icmp6hdr,
563                                                         icmp6_type),
564                                                 sizeof(icmp_type), &icmp_type);
565
566                 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
567                         return false;
568         }
569
570         return true;
571 }
572
573 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
574                             unsigned int mtu, unsigned int payload_length)
575 {
576         struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
577         struct icmp6hdr *icmp6h;
578         u8 *payload;
579
580         ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
581         icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
582         payload = skb_put(nskb, payload_length);
583
584         /* IPv6 */
585         ipv6h->version          =       6;
586         ipv6h->priority         =       0;
587         memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
588         ipv6h->payload_len      =       htons(sizeof(struct icmp6hdr)
589                                               + payload_length);
590         ipv6h->nexthdr          =       NEXTHDR_ICMP;
591         ipv6h->hop_limit        =       IPV6_DEFAULT_HOPLIMIT;
592         ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
593         ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
594
595         /* ICMPv6 */
596         icmp6h->icmp6_type      =       ICMPV6_PKT_TOOBIG;
597         icmp6h->icmp6_code      =       0;
598         icmp6h->icmp6_cksum     =       0;
599         icmp6h->icmp6_mtu       =       htonl(mtu);
600
601         nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
602         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
603                                             payload, payload_length,
604                                             nskb->csum);
605         icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
606                                                 sizeof(struct icmp6hdr)
607                                                 + payload_length,
608                                                 ipv6h->nexthdr, nskb->csum);
609 }
610 #endif /* IPv6 */
611
612 bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable,
613                      struct sk_buff *skb, unsigned int mtu, __be64 flow_key)
614 {
615         unsigned int eth_hdr_len = ETH_HLEN;
616         unsigned int total_length = 0, header_length = 0, payload_length;
617         struct ethhdr *eh, *old_eh = eth_hdr(skb);
618         struct sk_buff *nskb;
619
620         /* Sanity check */
621         if (skb->protocol == htons(ETH_P_IP)) {
622                 if (mtu < IP_MIN_MTU)
623                         return false;
624
625                 if (!ipv4_should_icmp(skb))
626                         return true;
627         }
628 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
629         else if (skb->protocol == htons(ETH_P_IPV6)) {
630                 if (mtu < IPV6_MIN_MTU)
631                         return false;
632
633                 /*
634                  * In theory we should do PMTUD on IPv6 multicast messages but
635                  * we don't have an address to send from so just fragment.
636                  */
637                 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
638                         return false;
639
640                 if (!ipv6_should_icmp(skb))
641                         return true;
642         }
643 #endif
644         else
645                 return false;
646
647         /* Allocate */
648         if (old_eh->h_proto == htons(ETH_P_8021Q))
649                 eth_hdr_len = VLAN_ETH_HLEN;
650
651         payload_length = skb->len - eth_hdr_len;
652         if (skb->protocol == htons(ETH_P_IP)) {
653                 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
654                 total_length = min_t(unsigned int, header_length +
655                                                    payload_length, 576);
656         }
657 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
658         else {
659                 header_length = sizeof(struct ipv6hdr) +
660                                 sizeof(struct icmp6hdr);
661                 total_length = min_t(unsigned int, header_length +
662                                                   payload_length, IPV6_MIN_MTU);
663         }
664 #endif
665
666         total_length = min(total_length, mutable->mtu);
667         payload_length = total_length - header_length;
668
669         nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
670                              payload_length);
671         if (!nskb)
672                 return false;
673
674         skb_reserve(nskb, NET_IP_ALIGN);
675
676         /* Ethernet / VLAN */
677         eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
678         memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
679         memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
680         nskb->protocol = eh->h_proto = old_eh->h_proto;
681         if (old_eh->h_proto == htons(ETH_P_8021Q)) {
682                 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
683
684                 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
685                 vh->h_vlan_encapsulated_proto = skb->protocol;
686         }
687         skb_reset_mac_header(nskb);
688
689         /* Protocol */
690         if (skb->protocol == htons(ETH_P_IP))
691                 ipv4_build_icmp(skb, nskb, mtu, payload_length);
692 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
693         else
694                 ipv6_build_icmp(skb, nskb, mtu, payload_length);
695 #endif
696
697         /*
698          * Assume that flow based keys are symmetric with respect to input
699          * and output and use the key that we were going to put on the
700          * outgoing packet for the fake received packet.  If the keys are
701          * not symmetric then PMTUD needs to be disabled since we won't have
702          * any way of synthesizing packets.
703          */
704         if ((mutable->flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
705             (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
706                 OVS_CB(nskb)->tun_id = flow_key;
707
708         compute_ip_summed(nskb, false);
709         vport_receive(vport, nskb);
710
711         return true;
712 }
713
714 static bool check_mtu(struct sk_buff *skb,
715                       struct vport *vport,
716                       const struct tnl_mutable_config *mutable,
717                       const struct rtable *rt, __be16 *frag_offp)
718 {
719         int mtu;
720         __be16 frag_off;
721
722         frag_off = (mutable->flags & TNL_F_PMTUD) ? htons(IP_DF) : 0;
723         if (frag_off)
724                 mtu = dst_mtu(&rt_dst(rt))
725                         - ETH_HLEN
726                         - mutable->tunnel_hlen
727                         - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
728         else
729                 mtu = mutable->mtu;
730
731         if (skb->protocol == htons(ETH_P_IP)) {
732                 struct iphdr *old_iph = ip_hdr(skb);
733
734                 frag_off |= old_iph->frag_off & htons(IP_DF);
735                 mtu = max(mtu, IP_MIN_MTU);
736
737                 if ((old_iph->frag_off & htons(IP_DF)) &&
738                     mtu < ntohs(old_iph->tot_len)) {
739                         if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
740                                 goto drop;
741                 }
742         }
743 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
744         else if (skb->protocol == htons(ETH_P_IPV6)) {
745                 unsigned int packet_length = skb->len - ETH_HLEN
746                         - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
747
748                 mtu = max(mtu, IPV6_MIN_MTU);
749
750                 /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
751                 if (packet_length > IPV6_MIN_MTU)
752                         frag_off = htons(IP_DF);
753
754                 if (mtu < packet_length) {
755                         if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
756                                 goto drop;
757                 }
758         }
759 #endif
760
761         *frag_offp = frag_off;
762         return true;
763
764 drop:
765         *frag_offp = 0;
766         return false;
767 }
768
769 static void create_tunnel_header(const struct vport *vport,
770                                  const struct tnl_mutable_config *mutable,
771                                  const struct rtable *rt, void *header)
772 {
773         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
774         struct iphdr *iph = header;
775
776         iph->version    = 4;
777         iph->ihl        = sizeof(struct iphdr) >> 2;
778         iph->frag_off   = htons(IP_DF);
779         iph->protocol   = tnl_vport->tnl_ops->ipproto;
780         iph->tos        = mutable->tos;
781         iph->daddr      = rt->rt_dst;
782         iph->saddr      = rt->rt_src;
783         iph->ttl        = mutable->ttl;
784         if (!iph->ttl)
785                 iph->ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
786
787         tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
788 }
789
790 static inline void *get_cached_header(const struct tnl_cache *cache)
791 {
792         return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
793 }
794
795 static inline bool check_cache_valid(const struct tnl_cache *cache,
796                                      const struct tnl_mutable_config *mutable)
797 {
798         return cache &&
799 #ifdef NEED_CACHE_TIMEOUT
800                 time_before(jiffies, cache->expiration) &&
801 #endif
802 #ifdef HAVE_RT_GENID
803                 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
804 #endif
805 #ifdef HAVE_HH_SEQ
806                 rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq &&
807 #endif
808                 mutable->seq == cache->mutable_seq &&
809                 (!is_internal_dev(rt_dst(cache->rt).dev) ||
810                 (cache->flow && !cache->flow->dead));
811 }
812
813 static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux)
814 {
815         struct tnl_vport *tnl_vport = tnl_vport_table_cast(tbl_node);
816         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
817         const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
818
819         if (cache && !check_cache_valid(cache, mutable) &&
820             spin_trylock_bh(&tnl_vport->cache_lock)) {
821                 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
822                 spin_unlock_bh(&tnl_vport->cache_lock);
823         }
824
825         return 0;
826 }
827
828 static void cache_cleaner(struct work_struct *work)
829 {
830         schedule_cache_cleaner();
831
832         rcu_read_lock();
833         tbl_foreach(rcu_dereference(port_table), cache_cleaner_cb, NULL);
834         rcu_read_unlock();
835 }
836
837 static inline void create_eth_hdr(struct tnl_cache *cache,
838                                   const struct rtable *rt)
839 {
840         void *cache_data = get_cached_header(cache);
841         int hh_len = rt_dst(rt).hh->hh_len;
842         int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len;
843
844 #ifdef HAVE_HH_SEQ
845         unsigned hh_seq;
846
847         do {
848                 hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock);
849                 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
850         } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq));
851
852         cache->hh_seq = hh_seq;
853 #else
854         read_lock_bh(&rt_dst(rt).hh->hh_lock);
855         memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
856         read_unlock_bh(&rt_dst(rt).hh->hh_lock);
857 #endif
858 }
859
860 static struct tnl_cache *build_cache(struct vport *vport,
861                                      const struct tnl_mutable_config *mutable,
862                                      struct rtable *rt)
863 {
864         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
865         struct tnl_cache *cache;
866         void *cache_data;
867         int cache_len;
868
869         if (!(mutable->flags & TNL_F_HDR_CACHE))
870                 return NULL;
871
872         /*
873          * If there is no entry in the ARP cache or if this device does not
874          * support hard header caching just fall back to the IP stack.
875          */
876         if (!rt_dst(rt).hh)
877                 return NULL;
878
879         /*
880          * If lock is contended fall back to directly building the header.
881          * We're not going to help performance by sitting here spinning.
882          */
883         if (!spin_trylock_bh(&tnl_vport->cache_lock))
884                 return NULL;
885
886         cache = cache_dereference(tnl_vport);
887         if (check_cache_valid(cache, mutable))
888                 goto unlock;
889         else
890                 cache = NULL;
891
892         cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen;
893
894         cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
895                         cache_len, GFP_ATOMIC);
896         if (!cache)
897                 goto unlock;
898
899         cache->len = cache_len;
900
901         create_eth_hdr(cache, rt);
902         cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len;
903
904         create_tunnel_header(vport, mutable, rt, cache_data);
905
906         cache->mutable_seq = mutable->seq;
907         cache->rt = rt;
908 #ifdef NEED_CACHE_TIMEOUT
909         cache->expiration = jiffies + tnl_vport->cache_exp_interval;
910 #endif
911
912         if (is_internal_dev(rt_dst(rt).dev)) {
913                 struct sw_flow_key flow_key;
914                 struct tbl_node *flow_node;
915                 struct vport *dst_vport;
916                 struct sk_buff *skb;
917                 bool is_frag;
918                 int err;
919
920                 dst_vport = internal_dev_get_vport(rt_dst(rt).dev);
921                 if (!dst_vport)
922                         goto done;
923
924                 skb = alloc_skb(cache->len, GFP_ATOMIC);
925                 if (!skb)
926                         goto done;
927
928                 __skb_put(skb, cache->len);
929                 memcpy(skb->data, get_cached_header(cache), cache->len);
930
931                 err = flow_extract(skb, dst_vport->port_no, &flow_key, &is_frag);
932
933                 kfree_skb(skb);
934                 if (err || is_frag)
935                         goto done;
936
937                 flow_node = tbl_lookup(rcu_dereference(dst_vport->dp->table),
938                                        &flow_key, flow_hash(&flow_key),
939                                        flow_cmp);
940                 if (flow_node) {
941                         struct sw_flow *flow = flow_cast(flow_node);
942
943                         cache->flow = flow;
944                         flow_hold(flow);
945                 }
946         }
947
948 done:
949         assign_cache_rcu(vport, cache);
950
951 unlock:
952         spin_unlock_bh(&tnl_vport->cache_lock);
953
954         return cache;
955 }
956
957 static struct rtable *find_route(struct vport *vport,
958                                  const struct tnl_mutable_config *mutable,
959                                  u8 tos, struct tnl_cache **cache)
960 {
961         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
962         struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
963
964         *cache = NULL;
965         tos = RT_TOS(tos);
966
967         if (likely(tos == mutable->tos && check_cache_valid(cur_cache, mutable))) {
968                 *cache = cur_cache;
969                 return cur_cache->rt;
970         } else {
971                 struct rtable *rt;
972                 struct flowi fl = { .nl_u = { .ip4_u =
973                                               { .daddr = mutable->daddr,
974                                                 .saddr = mutable->saddr,
975                                                 .tos = tos } },
976                                     .proto = tnl_vport->tnl_ops->ipproto };
977
978                 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
979                         return NULL;
980
981                 if (likely(tos == mutable->tos))
982                         *cache = build_cache(vport, mutable, rt);
983
984                 return rt;
985         }
986 }
987
988 static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom)
989 {
990         if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) {
991                 struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16);
992                 if (unlikely(!nskb)) {
993                         kfree_skb(skb);
994                         return ERR_PTR(-ENOMEM);
995                 }
996
997                 set_skb_csum_bits(skb, nskb);
998
999                 if (skb->sk)
1000                         skb_set_owner_w(nskb, skb->sk);
1001
1002                 kfree_skb(skb);
1003                 return nskb;
1004         }
1005
1006         return skb;
1007 }
1008
1009 static inline bool need_linearize(const struct sk_buff *skb)
1010 {
1011         int i;
1012
1013         if (unlikely(skb_shinfo(skb)->frag_list))
1014                 return true;
1015
1016         /*
1017          * Generally speaking we should linearize if there are paged frags.
1018          * However, if all of the refcounts are 1 we know nobody else can
1019          * change them from underneath us and we can skip the linearization.
1020          */
1021         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1022                 if (unlikely(page_count(skb_shinfo(skb)->frags[0].page) > 1))
1023                         return true;
1024
1025         return false;
1026 }
1027
1028 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1029                                        const struct tnl_mutable_config *mutable,
1030                                        const struct rtable *rt)
1031 {
1032         int min_headroom;
1033         int err;
1034
1035         forward_ip_summed(skb);
1036
1037         err = vswitch_skb_checksum_setup(skb);
1038         if (unlikely(err))
1039                 goto error_free;
1040
1041         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1042                         + mutable->tunnel_hlen;
1043
1044         if (skb_is_gso(skb)) {
1045                 struct sk_buff *nskb;
1046
1047                 /*
1048                  * If we are doing GSO on a pskb it is better to make sure that
1049                  * the headroom is correct now.  We will only have to copy the
1050                  * portion in the linear data area and GSO will preserve
1051                  * headroom when it creates the segments.  This is particularly
1052                  * beneficial on Xen where we get a lot of GSO pskbs.
1053                  * Conversely, we avoid copying if it is just to get our own
1054                  * writable clone because GSO will do the copy for us.
1055                  */
1056                 if (skb_headroom(skb) < min_headroom) {
1057                         skb = check_headroom(skb, min_headroom);
1058                         if (IS_ERR(skb)) {
1059                                 err = PTR_ERR(skb);
1060                                 goto error;
1061                         }
1062                 }
1063
1064                 nskb = skb_gso_segment(skb, 0);
1065                 kfree_skb(skb);
1066                 if (IS_ERR(nskb)) {
1067                         err = PTR_ERR(nskb);
1068                         goto error;
1069                 }
1070
1071                 skb = nskb;
1072         } else {
1073                 skb = check_headroom(skb, min_headroom);
1074                 if (IS_ERR(skb)) {
1075                         err = PTR_ERR(skb);
1076                         goto error;
1077                 }
1078
1079                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1080                         /*
1081                          * Pages aren't locked and could change at any time.
1082                          * If this happens after we compute the checksum, the
1083                          * checksum will be wrong.  We linearize now to avoid
1084                          * this problem.
1085                          */
1086                         if (unlikely(need_linearize(skb))) {
1087                                 err = __skb_linearize(skb);
1088                                 if (unlikely(err))
1089                                         goto error_free;
1090                         }
1091
1092                         err = skb_checksum_help(skb);
1093                         if (unlikely(err))
1094                                 goto error_free;
1095                 } else if (skb->ip_summed == CHECKSUM_COMPLETE)
1096                         skb->ip_summed = CHECKSUM_NONE;
1097         }
1098
1099         return skb;
1100
1101 error_free:
1102         kfree_skb(skb);
1103 error:
1104         return ERR_PTR(err);
1105 }
1106
1107 static int send_frags(struct sk_buff *skb,
1108                       const struct tnl_mutable_config *mutable)
1109 {
1110         int sent_len;
1111         int err;
1112
1113         sent_len = 0;
1114         while (skb) {
1115                 struct sk_buff *next = skb->next;
1116                 int frag_len = skb->len - mutable->tunnel_hlen;
1117
1118                 skb->next = NULL;
1119                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1120
1121                 err = ip_local_out(skb);
1122                 if (likely(net_xmit_eval(err) == 0))
1123                         sent_len += frag_len;
1124                 else {
1125                         skb = next;
1126                         goto free_frags;
1127                 }
1128
1129                 skb = next;
1130         }
1131
1132         return sent_len;
1133
1134 free_frags:
1135         /*
1136          * There's no point in continuing to send fragments once one has been
1137          * dropped so just free the rest.  This may help improve the congestion
1138          * that caused the first packet to be dropped.
1139          */
1140         tnl_free_linked_skbs(skb);
1141         return sent_len;
1142 }
1143
1144 int tnl_send(struct vport *vport, struct sk_buff *skb)
1145 {
1146         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1147         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1148
1149         enum vport_err_type err = VPORT_E_TX_ERROR;
1150         struct rtable *rt;
1151         struct dst_entry *unattached_dst = NULL;
1152         struct tnl_cache *cache;
1153         int sent_len = 0;
1154         __be16 frag_off;
1155         u8 ttl;
1156         u8 inner_tos;
1157         u8 tos;
1158
1159         /* Validate the protocol headers before we try to use them. */
1160         if (skb->protocol == htons(ETH_P_8021Q)) {
1161                 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1162                         goto error_free;
1163
1164                 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1165                 skb_set_network_header(skb, VLAN_ETH_HLEN);
1166         }
1167
1168         if (skb->protocol == htons(ETH_P_IP)) {
1169                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1170                     + sizeof(struct iphdr))))
1171                         skb->protocol = 0;
1172         }
1173 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1174         else if (skb->protocol == htons(ETH_P_IPV6)) {
1175                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1176                     + sizeof(struct ipv6hdr))))
1177                         skb->protocol = 0;
1178         }
1179 #endif
1180
1181         /* ToS */
1182         if (skb->protocol == htons(ETH_P_IP))
1183                 inner_tos = ip_hdr(skb)->tos;
1184 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1185         else if (skb->protocol == htons(ETH_P_IPV6))
1186                 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1187 #endif
1188         else
1189                 inner_tos = 0;
1190
1191         if (mutable->flags & TNL_F_TOS_INHERIT)
1192                 tos = inner_tos;
1193         else
1194                 tos = mutable->tos;
1195
1196         tos = INET_ECN_encapsulate(tos, inner_tos);
1197
1198         /* Route lookup */
1199         rt = find_route(vport, mutable, tos, &cache);
1200         if (unlikely(!rt))
1201                 goto error_free;
1202         if (unlikely(!cache))
1203                 unattached_dst = &rt_dst(rt);
1204
1205         /* Reset SKB */
1206         nf_reset(skb);
1207         secpath_reset(skb);
1208         skb_dst_drop(skb);
1209
1210         /* Offloading */
1211         skb = handle_offloads(skb, mutable, rt);
1212         if (IS_ERR(skb))
1213                 goto error;
1214
1215         /* MTU */
1216         if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1217                 err = VPORT_E_TX_DROPPED;
1218                 goto error_free;
1219         }
1220
1221         /*
1222          * If we are over the MTU, allow the IP stack to handle fragmentation.
1223          * Fragmentation is a slow path anyways.
1224          */
1225         if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1226                      cache)) {
1227                 unattached_dst = &rt_dst(rt);
1228                 dst_hold(unattached_dst);
1229                 cache = NULL;
1230         }
1231
1232         /* TTL */
1233         ttl = mutable->ttl;
1234         if (!ttl)
1235                 ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
1236
1237         if (mutable->flags & TNL_F_TTL_INHERIT) {
1238                 if (skb->protocol == htons(ETH_P_IP))
1239                         ttl = ip_hdr(skb)->ttl;
1240 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1241                 else if (skb->protocol == htons(ETH_P_IPV6))
1242                         ttl = ipv6_hdr(skb)->hop_limit;
1243 #endif
1244         }
1245
1246         while (skb) {
1247                 struct iphdr *iph;
1248                 struct sk_buff *next_skb = skb->next;
1249                 skb->next = NULL;
1250
1251                 if (likely(cache)) {
1252                         skb_push(skb, cache->len);
1253                         memcpy(skb->data, get_cached_header(cache), cache->len);
1254                         skb_reset_mac_header(skb);
1255                         skb_set_network_header(skb, rt_dst(rt).hh->hh_len);
1256
1257                 } else {
1258                         skb_push(skb, mutable->tunnel_hlen);
1259                         create_tunnel_header(vport, mutable, rt, skb->data);
1260                         skb_reset_network_header(skb);
1261
1262                         if (next_skb)
1263                                 skb_dst_set(skb, dst_clone(unattached_dst));
1264                         else {
1265                                 skb_dst_set(skb, unattached_dst);
1266                                 unattached_dst = NULL;
1267                         }
1268                 }
1269                 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1270
1271                 iph = ip_hdr(skb);
1272                 iph->tos = tos;
1273                 iph->ttl = ttl;
1274                 iph->frag_off = frag_off;
1275                 ip_select_ident(iph, &rt_dst(rt), NULL);
1276
1277                 skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
1278                 if (unlikely(!skb))
1279                         goto next;
1280
1281                 if (likely(cache)) {
1282                         int orig_len = skb->len - cache->len;
1283                         struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1284
1285                         skb->protocol = htons(ETH_P_IP);
1286                         iph = ip_hdr(skb);
1287                         iph->tot_len = htons(skb->len - skb_network_offset(skb));
1288                         ip_send_check(iph);
1289
1290                         if (cache_vport) {
1291                                 OVS_CB(skb)->flow = cache->flow;
1292                                 compute_ip_summed(skb, true);
1293                                 vport_receive(cache_vport, skb);
1294                                 sent_len += orig_len;
1295                         } else {
1296                                 int xmit_err;
1297
1298                                 skb->dev = rt_dst(rt).dev;
1299                                 xmit_err = dev_queue_xmit(skb);
1300
1301                                 if (likely(net_xmit_eval(xmit_err) == 0))
1302                                         sent_len += orig_len;
1303                         }
1304                 } else
1305                         sent_len += send_frags(skb, mutable);
1306
1307 next:
1308                 skb = next_skb;
1309         }
1310
1311         if (unlikely(sent_len == 0))
1312                 vport_record_error(vport, VPORT_E_TX_DROPPED);
1313
1314         goto out;
1315
1316 error_free:
1317         tnl_free_linked_skbs(skb);
1318 error:
1319         dst_release(unattached_dst);
1320         vport_record_error(vport, err);
1321 out:
1322         return sent_len;
1323 }
1324
1325 static const struct nla_policy tnl_policy[ODP_TUNNEL_ATTR_MAX + 1] = {
1326         [ODP_TUNNEL_ATTR_FLAGS]    = { .type = NLA_U32 },
1327         [ODP_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1328         [ODP_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1329         [ODP_TUNNEL_ATTR_OUT_KEY]  = { .type = NLA_U64 },
1330         [ODP_TUNNEL_ATTR_IN_KEY]   = { .type = NLA_U64 },
1331         [ODP_TUNNEL_ATTR_TOS]      = { .type = NLA_U8 },
1332         [ODP_TUNNEL_ATTR_TTL]      = { .type = NLA_U8 },
1333 };
1334
1335 /* Sets ODP_TUNNEL_ATTR_* fields in 'mutable', which must initially be zeroed. */
1336 static int tnl_set_config(struct nlattr *options, const struct tnl_ops *tnl_ops,
1337                           const struct vport *cur_vport,
1338                           struct tnl_mutable_config *mutable)
1339 {
1340         const struct vport *old_vport;
1341         const struct tnl_mutable_config *old_mutable;
1342         struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1];
1343         int err;
1344
1345         if (!options)
1346                 return -EINVAL;
1347
1348         err = nla_parse_nested(a, ODP_TUNNEL_ATTR_MAX, options, tnl_policy);
1349         if (err)
1350                 return err;
1351
1352         if (!a[ODP_TUNNEL_ATTR_FLAGS] || !a[ODP_TUNNEL_ATTR_DST_IPV4])
1353                 return -EINVAL;
1354
1355         mutable->flags = nla_get_u32(a[ODP_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1356
1357         if (a[ODP_TUNNEL_ATTR_SRC_IPV4])
1358                 mutable->saddr = nla_get_be32(a[ODP_TUNNEL_ATTR_SRC_IPV4]);
1359         mutable->daddr = nla_get_be32(a[ODP_TUNNEL_ATTR_DST_IPV4]);
1360
1361         if (a[ODP_TUNNEL_ATTR_TOS]) {
1362                 mutable->tos = nla_get_u8(a[ODP_TUNNEL_ATTR_TOS]);
1363                 if (mutable->tos != RT_TOS(mutable->tos))
1364                         return -EINVAL;
1365         }
1366
1367         if (a[ODP_TUNNEL_ATTR_TTL])
1368                 mutable->ttl = nla_get_u8(a[ODP_TUNNEL_ATTR_TTL]);
1369
1370         mutable->tunnel_hlen = tnl_ops->hdr_len(mutable);
1371         if (mutable->tunnel_hlen < 0)
1372                 return mutable->tunnel_hlen;
1373
1374         mutable->tunnel_hlen += sizeof(struct iphdr);
1375
1376         mutable->tunnel_type = tnl_ops->tunnel_type;
1377         if (!a[ODP_TUNNEL_ATTR_IN_KEY]) {
1378                 mutable->tunnel_type |= TNL_T_KEY_MATCH;
1379                 mutable->flags |= TNL_F_IN_KEY_MATCH;
1380         } else {
1381                 mutable->tunnel_type |= TNL_T_KEY_EXACT;
1382                 mutable->in_key = nla_get_be64(a[ODP_TUNNEL_ATTR_IN_KEY]);
1383         }
1384
1385         if (!a[ODP_TUNNEL_ATTR_OUT_KEY])
1386                 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1387         else
1388                 mutable->out_key = nla_get_be64(a[ODP_TUNNEL_ATTR_OUT_KEY]);
1389
1390         old_vport = tnl_find_port(mutable->saddr, mutable->daddr,
1391                                   mutable->in_key, mutable->tunnel_type,
1392                                   &old_mutable);
1393
1394         if (old_vport && old_vport != cur_vport)
1395                 return -EEXIST;
1396
1397         return 0;
1398 }
1399
1400 struct vport *tnl_create(const struct vport_parms *parms,
1401                          const struct vport_ops *vport_ops,
1402                          const struct tnl_ops *tnl_ops)
1403 {
1404         struct vport *vport;
1405         struct tnl_vport *tnl_vport;
1406         struct tnl_mutable_config *mutable;
1407         int initial_frag_id;
1408         int err;
1409
1410         vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1411         if (IS_ERR(vport)) {
1412                 err = PTR_ERR(vport);
1413                 goto error;
1414         }
1415
1416         tnl_vport = tnl_vport_priv(vport);
1417
1418         strcpy(tnl_vport->name, parms->name);
1419         tnl_vport->tnl_ops = tnl_ops;
1420
1421         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1422         if (!mutable) {
1423                 err = -ENOMEM;
1424                 goto error_free_vport;
1425         }
1426
1427         vport_gen_rand_ether_addr(mutable->eth_addr);
1428         mutable->mtu = ETH_DATA_LEN;
1429
1430         get_random_bytes(&initial_frag_id, sizeof(int));
1431         atomic_set(&tnl_vport->frag_id, initial_frag_id);
1432
1433         err = tnl_set_config(parms->options, tnl_ops, NULL, mutable);
1434         if (err)
1435                 goto error_free_mutable;
1436
1437         spin_lock_init(&tnl_vport->cache_lock);
1438
1439 #ifdef NEED_CACHE_TIMEOUT
1440         tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1441                                        (net_random() % (MAX_CACHE_EXP / 2));
1442 #endif
1443
1444         rcu_assign_pointer(tnl_vport->mutable, mutable);
1445
1446         err = add_port(vport);
1447         if (err)
1448                 goto error_free_mutable;
1449
1450         return vport;
1451
1452 error_free_mutable:
1453         kfree(mutable);
1454 error_free_vport:
1455         vport_free(vport);
1456 error:
1457         return ERR_PTR(err);
1458 }
1459
1460 int tnl_set_options(struct vport *vport, struct nlattr *options)
1461 {
1462         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1463         const struct tnl_mutable_config *old_mutable;
1464         struct tnl_mutable_config *mutable;
1465         int err;
1466
1467         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1468         if (!mutable) {
1469                 err = -ENOMEM;
1470                 goto error;
1471         }
1472
1473         /* Copy fields whose values should be retained. */
1474         old_mutable = rtnl_dereference(tnl_vport->mutable);
1475         mutable->seq = old_mutable->seq + 1;
1476         memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1477         mutable->mtu = old_mutable->mtu;
1478
1479         /* Parse the others configured by userspace. */
1480         err = tnl_set_config(options, tnl_vport->tnl_ops, vport, mutable);
1481         if (err)
1482                 goto error_free;
1483
1484         err = move_port(vport, mutable);
1485         if (err)
1486                 goto error_free;
1487
1488         return 0;
1489
1490 error_free:
1491         kfree(mutable);
1492 error:
1493         return err;
1494 }
1495
1496 int tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1497 {
1498         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1499         const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1500
1501         NLA_PUT_U32(skb, ODP_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC);
1502         NLA_PUT_BE32(skb, ODP_TUNNEL_ATTR_DST_IPV4, mutable->daddr);
1503
1504         if (!(mutable->flags & TNL_F_IN_KEY_MATCH))
1505                 NLA_PUT_BE64(skb, ODP_TUNNEL_ATTR_IN_KEY, mutable->in_key);
1506         if (!(mutable->flags & TNL_F_OUT_KEY_ACTION))
1507                 NLA_PUT_BE64(skb, ODP_TUNNEL_ATTR_OUT_KEY, mutable->out_key);
1508         if (mutable->saddr)
1509                 NLA_PUT_BE32(skb, ODP_TUNNEL_ATTR_SRC_IPV4, mutable->saddr);
1510         if (mutable->tos)
1511                 NLA_PUT_U8(skb, ODP_TUNNEL_ATTR_TOS, mutable->tos);
1512         if (mutable->ttl)
1513                 NLA_PUT_U8(skb, ODP_TUNNEL_ATTR_TTL, mutable->ttl);
1514
1515         return 0;
1516
1517 nla_put_failure:
1518         return -EMSGSIZE;
1519 }
1520
1521 static void free_port_rcu(struct rcu_head *rcu)
1522 {
1523         struct tnl_vport *tnl_vport = container_of(rcu,
1524                                                    struct tnl_vport, rcu);
1525
1526         free_cache((struct tnl_cache __force *)tnl_vport->cache);
1527         kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1528         vport_free(tnl_vport_to_vport(tnl_vport));
1529 }
1530
1531 int tnl_destroy(struct vport *vport)
1532 {
1533         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1534         const struct tnl_mutable_config *mutable, *old_mutable;
1535
1536         mutable = rtnl_dereference(tnl_vport->mutable);
1537
1538         if (vport == tnl_find_port(mutable->saddr, mutable->daddr,
1539                                    mutable->in_key, mutable->tunnel_type,
1540                                    &old_mutable))
1541                 del_port(vport);
1542
1543         call_rcu(&tnl_vport->rcu, free_port_rcu);
1544
1545         return 0;
1546 }
1547
1548 int tnl_set_mtu(struct vport *vport, int mtu)
1549 {
1550         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1551         struct tnl_mutable_config *mutable;
1552
1553         mutable = kmemdup(rtnl_dereference(tnl_vport->mutable),
1554                           sizeof(struct tnl_mutable_config), GFP_KERNEL);
1555         if (!mutable)
1556                 return -ENOMEM;
1557
1558         mutable->mtu = mtu;
1559         assign_config_rcu(vport, mutable);
1560
1561         return 0;
1562 }
1563
1564 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1565 {
1566         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1567         struct tnl_mutable_config *mutable;
1568
1569         mutable = kmemdup(rtnl_dereference(tnl_vport->mutable),
1570                           sizeof(struct tnl_mutable_config), GFP_KERNEL);
1571         if (!mutable)
1572                 return -ENOMEM;
1573
1574         memcpy(mutable->eth_addr, addr, ETH_ALEN);
1575         assign_config_rcu(vport, mutable);
1576
1577         return 0;
1578 }
1579
1580 const char *tnl_get_name(const struct vport *vport)
1581 {
1582         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1583         return tnl_vport->name;
1584 }
1585
1586 const unsigned char *tnl_get_addr(const struct vport *vport)
1587 {
1588         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1589         return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1590 }
1591
1592 int tnl_get_mtu(const struct vport *vport)
1593 {
1594         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1595         return rcu_dereference_rtnl(tnl_vport->mutable)->mtu;
1596 }
1597
1598 void tnl_free_linked_skbs(struct sk_buff *skb)
1599 {
1600         if (unlikely(!skb))
1601                 return;
1602
1603         while (skb) {
1604                 struct sk_buff *next = skb->next;
1605                 kfree_skb(skb);
1606                 skb = next;
1607         }
1608 }