]>
Commit | Line | Data |
---|---|---|
d1eb60cc JG |
1 | /* |
2 | * Copyright (c) 2010 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> | |
842cf6f4 | 18 | #include <linux/workqueue.h> |
d1eb60cc JG |
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" | |
dd8d6b8c | 32 | #include "checksum.h" |
d1eb60cc JG |
33 | #include "datapath.h" |
34 | #include "table.h" | |
35 | #include "tunnel.h" | |
36 | #include "vport.h" | |
37 | #include "vport-generic.h" | |
842cf6f4 JG |
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54 | * Interval to check for and remove caches that are no longer valid.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 | |
d1eb60cc JG |
70 | |
71 | /* Protected by RCU. */ | |
83e3e75b | 72 | static struct tbl *port_table __read_mostly; |
d1eb60cc | 73 | |
842cf6f4 JG |
74 | static void cache_cleaner(struct work_struct *work); |
75 | DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner); | |
76 | ||
d1eb60cc JG |
77 | /* |
78 | * These are just used as an optimization: they don't require any kind of | |
79 | * synchronization because we could have just as easily read the value before | |
80 | * the port change happened. | |
81 | */ | |
83e3e75b JG |
82 | static unsigned int key_local_remote_ports __read_mostly; |
83 | static unsigned int key_remote_ports __read_mostly; | |
84 | static unsigned int local_remote_ports __read_mostly; | |
85 | static unsigned int remote_ports __read_mostly; | |
d1eb60cc JG |
86 | |
87 | #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) | |
88 | #define rt_dst(rt) (rt->dst) | |
89 | #else | |
90 | #define rt_dst(rt) (rt->u.dst) | |
91 | #endif | |
92 | ||
93 | static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport) | |
94 | { | |
95 | return vport_from_priv(tnl_vport); | |
96 | } | |
97 | ||
98 | static inline struct tnl_vport *tnl_vport_table_cast(const struct tbl_node *node) | |
99 | { | |
100 | return container_of(node, struct tnl_vport, tbl_node); | |
101 | } | |
102 | ||
842cf6f4 JG |
103 | static inline void schedule_cache_cleaner(void) |
104 | { | |
105 | schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL); | |
106 | } | |
107 | ||
108 | static void free_cache(struct tnl_cache *cache) | |
109 | { | |
110 | if (!cache) | |
111 | return; | |
112 | ||
113 | flow_put(cache->flow); | |
114 | ip_rt_put(cache->rt); | |
115 | kfree(cache); | |
116 | } | |
117 | ||
118 | static void free_config_rcu(struct rcu_head *rcu) | |
d1eb60cc JG |
119 | { |
120 | struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu); | |
121 | kfree(c); | |
122 | } | |
123 | ||
842cf6f4 JG |
124 | static void free_cache_rcu(struct rcu_head *rcu) |
125 | { | |
126 | struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu); | |
127 | free_cache(c); | |
128 | } | |
129 | ||
d1eb60cc JG |
130 | static void assign_config_rcu(struct vport *vport, |
131 | struct tnl_mutable_config *new_config) | |
132 | { | |
133 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
134 | struct tnl_mutable_config *old_config; | |
135 | ||
842cf6f4 | 136 | old_config = tnl_vport->mutable; |
d1eb60cc | 137 | rcu_assign_pointer(tnl_vport->mutable, new_config); |
842cf6f4 JG |
138 | call_rcu(&old_config->rcu, free_config_rcu); |
139 | } | |
140 | ||
141 | static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache) | |
142 | { | |
143 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
144 | struct tnl_cache *old_cache; | |
145 | ||
146 | old_cache = tnl_vport->cache; | |
147 | rcu_assign_pointer(tnl_vport->cache, new_cache); | |
148 | ||
149 | if (old_cache) | |
150 | call_rcu(&old_cache->rcu, free_cache_rcu); | |
d1eb60cc JG |
151 | } |
152 | ||
153 | static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable) | |
154 | { | |
155 | if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) { | |
156 | if (mutable->port_config.saddr) | |
157 | return &local_remote_ports; | |
158 | else | |
159 | return &remote_ports; | |
160 | } else { | |
161 | if (mutable->port_config.saddr) | |
162 | return &key_local_remote_ports; | |
163 | else | |
164 | return &key_remote_ports; | |
165 | } | |
166 | } | |
167 | ||
d1eb60cc | 168 | struct port_lookup_key { |
4029c21a BP |
169 | u32 tunnel_type; |
170 | __be32 saddr; | |
171 | __be32 daddr; | |
172 | __be32 key; | |
d1eb60cc JG |
173 | const struct tnl_mutable_config *mutable; |
174 | }; | |
175 | ||
176 | /* | |
177 | * Modifies 'target' to store the rcu_dereferenced pointer that was used to do | |
178 | * the comparision. | |
179 | */ | |
180 | static int port_cmp(const struct tbl_node *node, void *target) | |
181 | { | |
182 | const struct tnl_vport *tnl_vport = tnl_vport_table_cast(node); | |
183 | struct port_lookup_key *lookup = target; | |
184 | ||
185 | lookup->mutable = rcu_dereference(tnl_vport->mutable); | |
186 | ||
4029c21a BP |
187 | return (lookup->mutable->tunnel_type == lookup->tunnel_type && |
188 | lookup->mutable->port_config.daddr == lookup->daddr && | |
189 | lookup->mutable->port_config.in_key == lookup->key && | |
190 | lookup->mutable->port_config.saddr == lookup->saddr); | |
d1eb60cc JG |
191 | } |
192 | ||
4029c21a | 193 | static u32 port_hash(struct port_lookup_key *k) |
d1eb60cc | 194 | { |
4029c21a | 195 | return jhash_3words(k->key, k->saddr, k->daddr, k->tunnel_type); |
d1eb60cc JG |
196 | } |
197 | ||
842cf6f4 JG |
198 | static u32 mutable_hash(const struct tnl_mutable_config *mutable) |
199 | { | |
200 | struct port_lookup_key lookup; | |
201 | ||
4029c21a BP |
202 | lookup.saddr = mutable->port_config.saddr; |
203 | lookup.daddr = mutable->port_config.daddr; | |
204 | lookup.key = mutable->port_config.in_key; | |
205 | lookup.tunnel_type = mutable->tunnel_type; | |
842cf6f4 JG |
206 | |
207 | return port_hash(&lookup); | |
208 | } | |
209 | ||
210 | static void check_table_empty(void) | |
211 | { | |
212 | if (tbl_count(port_table) == 0) { | |
213 | struct tbl *old_table = port_table; | |
214 | ||
215 | cancel_delayed_work_sync(&cache_cleaner_wq); | |
216 | rcu_assign_pointer(port_table, NULL); | |
217 | tbl_deferred_destroy(old_table, NULL); | |
218 | } | |
219 | } | |
220 | ||
d1eb60cc JG |
221 | static int add_port(struct vport *vport) |
222 | { | |
223 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
d1eb60cc JG |
224 | int err; |
225 | ||
226 | if (!port_table) { | |
227 | struct tbl *new_table; | |
228 | ||
229 | new_table = tbl_create(0); | |
230 | if (!new_table) | |
231 | return -ENOMEM; | |
232 | ||
233 | rcu_assign_pointer(port_table, new_table); | |
842cf6f4 | 234 | schedule_cache_cleaner(); |
d1eb60cc JG |
235 | |
236 | } else if (tbl_count(port_table) > tbl_n_buckets(port_table)) { | |
237 | struct tbl *old_table = port_table; | |
238 | struct tbl *new_table; | |
239 | ||
240 | new_table = tbl_expand(old_table); | |
241 | if (IS_ERR(new_table)) | |
242 | return PTR_ERR(new_table); | |
243 | ||
244 | rcu_assign_pointer(port_table, new_table); | |
245 | tbl_deferred_destroy(old_table, NULL); | |
246 | } | |
247 | ||
842cf6f4 JG |
248 | err = tbl_insert(port_table, &tnl_vport->tbl_node, mutable_hash(tnl_vport->mutable)); |
249 | if (err) { | |
6569d3f4 | 250 | (*find_port_pool(tnl_vport->mutable))--; |
842cf6f4 JG |
251 | check_table_empty(); |
252 | return err; | |
253 | } | |
254 | ||
255 | (*find_port_pool(tnl_vport->mutable))++; | |
256 | ||
257 | return 0; | |
258 | } | |
259 | ||
260 | static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable) | |
261 | { | |
262 | int err; | |
263 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
264 | u32 hash; | |
265 | ||
266 | hash = mutable_hash(new_mutable); | |
267 | if (hash == tnl_vport->tbl_node.hash) | |
268 | goto table_updated; | |
d1eb60cc | 269 | |
842cf6f4 JG |
270 | /* |
271 | * Ideally we should make this move atomic to avoid having gaps in | |
272 | * finding tunnels or the possibility of failure. However, if we do | |
273 | * find a tunnel it will always be consistent. | |
274 | */ | |
275 | err = tbl_remove(port_table, &tnl_vport->tbl_node); | |
d1eb60cc JG |
276 | if (err) |
277 | return err; | |
278 | ||
842cf6f4 JG |
279 | err = tbl_insert(port_table, &tnl_vport->tbl_node, hash); |
280 | if (err) { | |
281 | check_table_empty(); | |
282 | return err; | |
283 | } | |
284 | ||
285 | table_updated: | |
6569d3f4 | 286 | (*find_port_pool(tnl_vport->mutable))--; |
842cf6f4 | 287 | assign_config_rcu(vport, new_mutable); |
6569d3f4 | 288 | (*find_port_pool(tnl_vport->mutable))++; |
d1eb60cc JG |
289 | |
290 | return 0; | |
291 | } | |
292 | ||
293 | static int del_port(struct vport *vport) | |
294 | { | |
295 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
296 | int err; | |
297 | ||
298 | err = tbl_remove(port_table, &tnl_vport->tbl_node); | |
299 | if (err) | |
300 | return err; | |
301 | ||
842cf6f4 | 302 | check_table_empty(); |
d1eb60cc JG |
303 | (*find_port_pool(tnl_vport->mutable))--; |
304 | ||
305 | return 0; | |
306 | } | |
307 | ||
308 | struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be32 key, | |
309 | int tunnel_type, | |
310 | const struct tnl_mutable_config **mutable) | |
311 | { | |
312 | struct port_lookup_key lookup; | |
313 | struct tbl *table = rcu_dereference(port_table); | |
314 | struct tbl_node *tbl_node; | |
315 | ||
842cf6f4 | 316 | if (unlikely(!table)) |
d1eb60cc JG |
317 | return NULL; |
318 | ||
4029c21a BP |
319 | lookup.saddr = saddr; |
320 | lookup.daddr = daddr; | |
d1eb60cc JG |
321 | |
322 | if (tunnel_type & TNL_T_KEY_EXACT) { | |
4029c21a BP |
323 | lookup.key = key; |
324 | lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_MATCH; | |
d1eb60cc JG |
325 | |
326 | if (key_local_remote_ports) { | |
327 | tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp); | |
328 | if (tbl_node) | |
329 | goto found; | |
330 | } | |
331 | ||
332 | if (key_remote_ports) { | |
4029c21a | 333 | lookup.saddr = 0; |
d1eb60cc JG |
334 | |
335 | tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp); | |
336 | if (tbl_node) | |
337 | goto found; | |
338 | ||
4029c21a | 339 | lookup.saddr = saddr; |
d1eb60cc JG |
340 | } |
341 | } | |
342 | ||
343 | if (tunnel_type & TNL_T_KEY_MATCH) { | |
4029c21a BP |
344 | lookup.key = 0; |
345 | lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_EXACT; | |
d1eb60cc JG |
346 | |
347 | if (local_remote_ports) { | |
348 | tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp); | |
349 | if (tbl_node) | |
350 | goto found; | |
351 | } | |
352 | ||
353 | if (remote_ports) { | |
4029c21a | 354 | lookup.saddr = 0; |
d1eb60cc JG |
355 | |
356 | tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp); | |
357 | if (tbl_node) | |
358 | goto found; | |
359 | } | |
360 | } | |
361 | ||
362 | return NULL; | |
363 | ||
364 | found: | |
365 | *mutable = lookup.mutable; | |
366 | return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node)); | |
367 | } | |
368 | ||
842cf6f4 JG |
369 | static inline void ecn_decapsulate(struct sk_buff *skb) |
370 | { | |
a2a96c04 JG |
371 | /* This is accessing the outer IP header of the tunnel, which we've |
372 | * already validated to be OK. skb->data is currently set to the start | |
373 | * of the inner Ethernet header, and we've validated ETH_HLEN. | |
374 | */ | |
375 | if (unlikely(INET_ECN_is_ce(ip_hdr(skb)->tos))) { | |
842cf6f4 | 376 | __be16 protocol = skb->protocol; |
a2a96c04 JG |
377 | |
378 | skb_set_network_header(skb, ETH_HLEN); | |
842cf6f4 JG |
379 | |
380 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
381 | if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN))) | |
382 | return; | |
383 | ||
384 | protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto; | |
a2a96c04 | 385 | skb_set_network_header(skb, VLAN_ETH_HLEN); |
842cf6f4 JG |
386 | } |
387 | ||
388 | if (protocol == htons(ETH_P_IP)) { | |
a2a96c04 | 389 | if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb) |
842cf6f4 JG |
390 | + sizeof(struct iphdr)))) |
391 | return; | |
392 | ||
a2a96c04 | 393 | IP_ECN_set_ce(ip_hdr(skb)); |
842cf6f4 JG |
394 | } |
395 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
396 | else if (protocol == htons(ETH_P_IPV6)) { | |
a2a96c04 | 397 | if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb) |
842cf6f4 JG |
398 | + sizeof(struct ipv6hdr)))) |
399 | return; | |
400 | ||
a2a96c04 | 401 | IP6_ECN_set_ce(ipv6_hdr(skb)); |
842cf6f4 JG |
402 | } |
403 | #endif | |
404 | } | |
405 | } | |
406 | ||
407 | /* Called with rcu_read_lock. */ | |
408 | void tnl_rcv(struct vport *vport, struct sk_buff *skb) | |
409 | { | |
9851dd67 JG |
410 | /* Packets received by this function are in the following state: |
411 | * - skb->data points to the inner Ethernet header. | |
412 | * - The inner Ethernet header is in the linear data area. | |
413 | * - skb->csum does not include the inner Ethernet header. | |
414 | * - The layer pointers point at the outer headers. | |
415 | */ | |
416 | ||
417 | struct ethhdr *eh = (struct ethhdr *)skb->data; | |
418 | ||
419 | if (likely(ntohs(eh->h_proto) >= 1536)) | |
420 | skb->protocol = eh->h_proto; | |
421 | else | |
422 | skb->protocol = htons(ETH_P_802_2); | |
842cf6f4 JG |
423 | |
424 | skb_dst_drop(skb); | |
425 | nf_reset(skb); | |
426 | secpath_reset(skb); | |
842cf6f4 JG |
427 | |
428 | ecn_decapsulate(skb); | |
842cf6f4 JG |
429 | compute_ip_summed(skb, false); |
430 | ||
431 | vport_receive(vport, skb); | |
432 | } | |
433 | ||
d1eb60cc JG |
434 | static bool check_ipv4_address(__be32 addr) |
435 | { | |
436 | if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr) | |
437 | || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr)) | |
438 | return false; | |
439 | ||
440 | return true; | |
441 | } | |
442 | ||
443 | static bool ipv4_should_icmp(struct sk_buff *skb) | |
444 | { | |
445 | struct iphdr *old_iph = ip_hdr(skb); | |
446 | ||
447 | /* Don't respond to L2 broadcast. */ | |
448 | if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) | |
449 | return false; | |
450 | ||
451 | /* Don't respond to L3 broadcast or invalid addresses. */ | |
452 | if (!check_ipv4_address(old_iph->daddr) || | |
453 | !check_ipv4_address(old_iph->saddr)) | |
454 | return false; | |
455 | ||
456 | /* Only respond to the first fragment. */ | |
457 | if (old_iph->frag_off & htons(IP_OFFSET)) | |
458 | return false; | |
459 | ||
460 | /* Don't respond to ICMP error messages. */ | |
461 | if (old_iph->protocol == IPPROTO_ICMP) { | |
462 | u8 icmp_type, *icmp_typep; | |
463 | ||
464 | icmp_typep = skb_header_pointer(skb, (u8 *)old_iph + | |
465 | (old_iph->ihl << 2) + | |
466 | offsetof(struct icmphdr, type) - | |
467 | skb->data, sizeof(icmp_type), | |
468 | &icmp_type); | |
469 | ||
470 | if (!icmp_typep) | |
471 | return false; | |
472 | ||
473 | if (*icmp_typep > NR_ICMP_TYPES | |
474 | || (*icmp_typep <= ICMP_PARAMETERPROB | |
475 | && *icmp_typep != ICMP_ECHOREPLY | |
476 | && *icmp_typep != ICMP_ECHO)) | |
477 | return false; | |
478 | } | |
479 | ||
480 | return true; | |
481 | } | |
482 | ||
483 | static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb, | |
484 | unsigned int mtu, unsigned int payload_length) | |
485 | { | |
486 | struct iphdr *iph, *old_iph = ip_hdr(skb); | |
487 | struct icmphdr *icmph; | |
488 | u8 *payload; | |
489 | ||
490 | iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr)); | |
491 | icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr)); | |
492 | payload = skb_put(nskb, payload_length); | |
493 | ||
494 | /* IP */ | |
495 | iph->version = 4; | |
496 | iph->ihl = sizeof(struct iphdr) >> 2; | |
497 | iph->tos = (old_iph->tos & IPTOS_TOS_MASK) | | |
498 | IPTOS_PREC_INTERNETCONTROL; | |
499 | iph->tot_len = htons(sizeof(struct iphdr) | |
500 | + sizeof(struct icmphdr) | |
501 | + payload_length); | |
502 | get_random_bytes(&iph->id, sizeof(iph->id)); | |
503 | iph->frag_off = 0; | |
504 | iph->ttl = IPDEFTTL; | |
505 | iph->protocol = IPPROTO_ICMP; | |
506 | iph->daddr = old_iph->saddr; | |
507 | iph->saddr = old_iph->daddr; | |
508 | ||
509 | ip_send_check(iph); | |
510 | ||
511 | /* ICMP */ | |
512 | icmph->type = ICMP_DEST_UNREACH; | |
513 | icmph->code = ICMP_FRAG_NEEDED; | |
514 | icmph->un.gateway = htonl(mtu); | |
515 | icmph->checksum = 0; | |
516 | ||
517 | nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0); | |
518 | nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data, | |
519 | payload, payload_length, | |
520 | nskb->csum); | |
521 | icmph->checksum = csum_fold(nskb->csum); | |
522 | } | |
523 | ||
524 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
525 | static bool ipv6_should_icmp(struct sk_buff *skb) | |
526 | { | |
527 | struct ipv6hdr *old_ipv6h = ipv6_hdr(skb); | |
528 | int addr_type; | |
529 | int payload_off = (u8 *)(old_ipv6h + 1) - skb->data; | |
530 | u8 nexthdr = ipv6_hdr(skb)->nexthdr; | |
531 | ||
532 | /* Check source address is valid. */ | |
533 | addr_type = ipv6_addr_type(&old_ipv6h->saddr); | |
534 | if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY) | |
535 | return false; | |
536 | ||
537 | /* Don't reply to unspecified addresses. */ | |
538 | if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY) | |
539 | return false; | |
540 | ||
541 | /* Don't respond to ICMP error messages. */ | |
542 | payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr); | |
543 | if (payload_off < 0) | |
544 | return false; | |
545 | ||
546 | if (nexthdr == NEXTHDR_ICMP) { | |
547 | u8 icmp_type, *icmp_typep; | |
548 | ||
549 | icmp_typep = skb_header_pointer(skb, payload_off + | |
550 | offsetof(struct icmp6hdr, | |
551 | icmp6_type), | |
552 | sizeof(icmp_type), &icmp_type); | |
553 | ||
554 | if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK)) | |
555 | return false; | |
556 | } | |
557 | ||
558 | return true; | |
559 | } | |
560 | ||
561 | static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb, | |
562 | unsigned int mtu, unsigned int payload_length) | |
563 | { | |
564 | struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb); | |
565 | struct icmp6hdr *icmp6h; | |
566 | u8 *payload; | |
567 | ||
568 | ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr)); | |
569 | icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr)); | |
570 | payload = skb_put(nskb, payload_length); | |
571 | ||
572 | /* IPv6 */ | |
573 | ipv6h->version = 6; | |
574 | ipv6h->priority = 0; | |
575 | memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl)); | |
576 | ipv6h->payload_len = htons(sizeof(struct icmp6hdr) | |
577 | + payload_length); | |
578 | ipv6h->nexthdr = NEXTHDR_ICMP; | |
579 | ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT; | |
580 | ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr); | |
581 | ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr); | |
582 | ||
583 | /* ICMPv6 */ | |
584 | icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG; | |
585 | icmp6h->icmp6_code = 0; | |
586 | icmp6h->icmp6_cksum = 0; | |
587 | icmp6h->icmp6_mtu = htonl(mtu); | |
588 | ||
589 | nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0); | |
590 | nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data, | |
591 | payload, payload_length, | |
592 | nskb->csum); | |
593 | icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, | |
594 | sizeof(struct icmp6hdr) | |
595 | + payload_length, | |
596 | ipv6h->nexthdr, nskb->csum); | |
597 | } | |
598 | #endif /* IPv6 */ | |
599 | ||
600 | bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable, | |
601 | struct sk_buff *skb, unsigned int mtu, __be32 flow_key) | |
602 | { | |
603 | unsigned int eth_hdr_len = ETH_HLEN; | |
604 | unsigned int total_length = 0, header_length = 0, payload_length; | |
605 | struct ethhdr *eh, *old_eh = eth_hdr(skb); | |
606 | struct sk_buff *nskb; | |
607 | ||
608 | /* Sanity check */ | |
609 | if (skb->protocol == htons(ETH_P_IP)) { | |
610 | if (mtu < IP_MIN_MTU) | |
611 | return false; | |
612 | ||
613 | if (!ipv4_should_icmp(skb)) | |
614 | return true; | |
615 | } | |
616 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
617 | else if (skb->protocol == htons(ETH_P_IPV6)) { | |
618 | if (mtu < IPV6_MIN_MTU) | |
619 | return false; | |
620 | ||
621 | /* | |
622 | * In theory we should do PMTUD on IPv6 multicast messages but | |
623 | * we don't have an address to send from so just fragment. | |
624 | */ | |
625 | if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST) | |
626 | return false; | |
627 | ||
628 | if (!ipv6_should_icmp(skb)) | |
629 | return true; | |
630 | } | |
631 | #endif | |
632 | else | |
633 | return false; | |
634 | ||
635 | /* Allocate */ | |
636 | if (old_eh->h_proto == htons(ETH_P_8021Q)) | |
637 | eth_hdr_len = VLAN_ETH_HLEN; | |
638 | ||
639 | payload_length = skb->len - eth_hdr_len; | |
640 | if (skb->protocol == htons(ETH_P_IP)) { | |
641 | header_length = sizeof(struct iphdr) + sizeof(struct icmphdr); | |
642 | total_length = min_t(unsigned int, header_length + | |
643 | payload_length, 576); | |
644 | } | |
645 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
646 | else { | |
647 | header_length = sizeof(struct ipv6hdr) + | |
648 | sizeof(struct icmp6hdr); | |
649 | total_length = min_t(unsigned int, header_length + | |
650 | payload_length, IPV6_MIN_MTU); | |
651 | } | |
652 | #endif | |
653 | ||
654 | total_length = min(total_length, mutable->mtu); | |
655 | payload_length = total_length - header_length; | |
656 | ||
657 | nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length + | |
658 | payload_length); | |
659 | if (!nskb) | |
660 | return false; | |
661 | ||
662 | skb_reserve(nskb, NET_IP_ALIGN); | |
663 | ||
664 | /* Ethernet / VLAN */ | |
665 | eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len); | |
666 | memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN); | |
667 | memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN); | |
668 | nskb->protocol = eh->h_proto = old_eh->h_proto; | |
669 | if (old_eh->h_proto == htons(ETH_P_8021Q)) { | |
670 | struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh; | |
671 | ||
672 | vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI; | |
673 | vh->h_vlan_encapsulated_proto = skb->protocol; | |
674 | } | |
675 | skb_reset_mac_header(nskb); | |
676 | ||
677 | /* Protocol */ | |
678 | if (skb->protocol == htons(ETH_P_IP)) | |
679 | ipv4_build_icmp(skb, nskb, mtu, payload_length); | |
680 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
681 | else | |
682 | ipv6_build_icmp(skb, nskb, mtu, payload_length); | |
683 | #endif | |
684 | ||
685 | /* | |
686 | * Assume that flow based keys are symmetric with respect to input | |
687 | * and output and use the key that we were going to put on the | |
688 | * outgoing packet for the fake received packet. If the keys are | |
689 | * not symmetric then PMTUD needs to be disabled since we won't have | |
690 | * any way of synthesizing packets. | |
691 | */ | |
692 | if ((mutable->port_config.flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) == | |
693 | (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) | |
694 | OVS_CB(nskb)->tun_id = flow_key; | |
695 | ||
696 | compute_ip_summed(nskb, false); | |
697 | vport_receive(vport, nskb); | |
698 | ||
699 | return true; | |
700 | } | |
701 | ||
842cf6f4 JG |
702 | static bool check_mtu(struct sk_buff *skb, |
703 | struct vport *vport, | |
704 | const struct tnl_mutable_config *mutable, | |
705 | const struct rtable *rt, __be16 *frag_offp) | |
d1eb60cc | 706 | { |
842cf6f4 JG |
707 | int mtu; |
708 | __be16 frag_off; | |
709 | ||
710 | frag_off = (mutable->port_config.flags & TNL_F_PMTUD) ? htons(IP_DF) : 0; | |
711 | if (frag_off) | |
712 | mtu = dst_mtu(&rt_dst(rt)) | |
713 | - ETH_HLEN | |
714 | - mutable->tunnel_hlen | |
715 | - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0); | |
716 | else | |
717 | mtu = mutable->mtu; | |
718 | ||
719 | if (skb->protocol == htons(ETH_P_IP)) { | |
720 | struct iphdr *old_iph = ip_hdr(skb); | |
721 | ||
722 | frag_off |= old_iph->frag_off & htons(IP_DF); | |
723 | mtu = max(mtu, IP_MIN_MTU); | |
724 | ||
725 | if ((old_iph->frag_off & htons(IP_DF)) && | |
726 | mtu < ntohs(old_iph->tot_len)) { | |
727 | if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id)) | |
728 | goto drop; | |
d1eb60cc | 729 | } |
842cf6f4 JG |
730 | } |
731 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
732 | else if (skb->protocol == htons(ETH_P_IPV6)) { | |
733 | unsigned int packet_length = skb->len - ETH_HLEN | |
734 | - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0); | |
d1eb60cc | 735 | |
842cf6f4 | 736 | mtu = max(mtu, IPV6_MIN_MTU); |
d1eb60cc | 737 | |
842cf6f4 JG |
738 | /* IPv6 requires PMTUD if the packet is above the minimum MTU. */ |
739 | if (packet_length > IPV6_MIN_MTU) | |
740 | frag_off = htons(IP_DF); | |
d1eb60cc | 741 | |
842cf6f4 JG |
742 | if (mtu < packet_length) { |
743 | if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id)) | |
744 | goto drop; | |
745 | } | |
d1eb60cc | 746 | } |
842cf6f4 | 747 | #endif |
d1eb60cc | 748 | |
842cf6f4 JG |
749 | *frag_offp = frag_off; |
750 | return true; | |
751 | ||
752 | drop: | |
753 | *frag_offp = 0; | |
754 | return false; | |
d1eb60cc JG |
755 | } |
756 | ||
842cf6f4 JG |
757 | static void create_tunnel_header(const struct vport *vport, |
758 | const struct tnl_mutable_config *mutable, | |
759 | const struct rtable *rt, void *header) | |
d1eb60cc | 760 | { |
842cf6f4 JG |
761 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); |
762 | struct iphdr *iph = header; | |
763 | ||
764 | iph->version = 4; | |
765 | iph->ihl = sizeof(struct iphdr) >> 2; | |
766 | iph->frag_off = htons(IP_DF); | |
767 | iph->protocol = tnl_vport->tnl_ops->ipproto; | |
768 | iph->tos = mutable->port_config.tos; | |
769 | iph->daddr = rt->rt_dst; | |
770 | iph->saddr = rt->rt_src; | |
771 | iph->ttl = mutable->port_config.ttl; | |
772 | if (!iph->ttl) | |
773 | iph->ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT); | |
774 | ||
775 | tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1); | |
776 | } | |
d1eb60cc | 777 | |
842cf6f4 JG |
778 | static inline void *get_cached_header(const struct tnl_cache *cache) |
779 | { | |
780 | return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN); | |
781 | } | |
d1eb60cc | 782 | |
842cf6f4 JG |
783 | static inline bool check_cache_valid(const struct tnl_cache *cache, |
784 | const struct tnl_mutable_config *mutable) | |
785 | { | |
786 | return cache && | |
787 | #ifdef NEED_CACHE_TIMEOUT | |
788 | time_before(jiffies, cache->expiration) && | |
789 | #endif | |
790 | #ifdef HAVE_RT_GENID | |
791 | atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid && | |
792 | #endif | |
793 | #ifdef HAVE_HH_SEQ | |
794 | rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq && | |
795 | #endif | |
796 | mutable->seq == cache->mutable_seq && | |
797 | (!is_internal_dev(rt_dst(cache->rt).dev) || | |
798 | (cache->flow && !cache->flow->dead)); | |
d1eb60cc JG |
799 | } |
800 | ||
842cf6f4 | 801 | static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux) |
d1eb60cc | 802 | { |
842cf6f4 JG |
803 | struct tnl_vport *tnl_vport = tnl_vport_table_cast(tbl_node); |
804 | const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable); | |
805 | const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache); | |
d1eb60cc | 806 | |
842cf6f4 JG |
807 | if (cache && !check_cache_valid(cache, mutable) && |
808 | spin_trylock_bh(&tnl_vport->cache_lock)) { | |
809 | assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL); | |
810 | spin_unlock_bh(&tnl_vport->cache_lock); | |
811 | } | |
d1eb60cc | 812 | |
842cf6f4 JG |
813 | return 0; |
814 | } | |
d1eb60cc | 815 | |
842cf6f4 JG |
816 | static void cache_cleaner(struct work_struct *work) |
817 | { | |
818 | schedule_cache_cleaner(); | |
d1eb60cc | 819 | |
842cf6f4 JG |
820 | rcu_read_lock(); |
821 | tbl_foreach(port_table, cache_cleaner_cb, NULL); | |
822 | rcu_read_unlock(); | |
823 | } | |
d1eb60cc | 824 | |
842cf6f4 JG |
825 | static inline void create_eth_hdr(struct tnl_cache *cache, |
826 | const struct rtable *rt) | |
827 | { | |
828 | void *cache_data = get_cached_header(cache); | |
829 | int hh_len = rt_dst(rt).hh->hh_len; | |
830 | int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len; | |
d1eb60cc | 831 | |
842cf6f4 JG |
832 | #ifdef HAVE_HH_SEQ |
833 | unsigned hh_seq; | |
834 | ||
835 | do { | |
836 | hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock); | |
837 | memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len); | |
838 | } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq)); | |
839 | ||
840 | cache->hh_seq = hh_seq; | |
841 | #else | |
842 | read_lock_bh(&rt_dst(rt).hh->hh_lock); | |
843 | memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len); | |
844 | read_unlock_bh(&rt_dst(rt).hh->hh_lock); | |
d1eb60cc | 845 | #endif |
d1eb60cc JG |
846 | } |
847 | ||
842cf6f4 JG |
848 | static struct tnl_cache *build_cache(struct vport *vport, |
849 | const struct tnl_mutable_config *mutable, | |
850 | struct rtable *rt) | |
d1eb60cc | 851 | { |
842cf6f4 JG |
852 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); |
853 | struct tnl_cache *cache; | |
854 | void *cache_data; | |
855 | int cache_len; | |
d1eb60cc | 856 | |
842cf6f4 JG |
857 | if (!(mutable->port_config.flags & TNL_F_HDR_CACHE)) |
858 | return NULL; | |
859 | ||
860 | /* | |
861 | * If there is no entry in the ARP cache or if this device does not | |
862 | * support hard header caching just fall back to the IP stack. | |
863 | */ | |
864 | if (!rt_dst(rt).hh) | |
865 | return NULL; | |
866 | ||
867 | /* | |
868 | * If lock is contended fall back to directly building the header. | |
869 | * We're not going to help performance by sitting here spinning. | |
870 | */ | |
871 | if (!spin_trylock_bh(&tnl_vport->cache_lock)) | |
872 | return NULL; | |
873 | ||
874 | cache = tnl_vport->cache; | |
875 | if (check_cache_valid(cache, mutable)) | |
876 | goto unlock; | |
877 | else | |
878 | cache = NULL; | |
879 | ||
880 | cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen; | |
881 | ||
882 | cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) + | |
883 | cache_len, GFP_ATOMIC); | |
884 | if (!cache) | |
885 | goto unlock; | |
886 | ||
887 | cache->len = cache_len; | |
888 | ||
889 | create_eth_hdr(cache, rt); | |
890 | cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len; | |
891 | ||
892 | create_tunnel_header(vport, mutable, rt, cache_data); | |
893 | ||
894 | cache->mutable_seq = mutable->seq; | |
895 | cache->rt = rt; | |
896 | #ifdef NEED_CACHE_TIMEOUT | |
897 | cache->expiration = jiffies + tnl_vport->cache_exp_interval; | |
898 | #endif | |
899 | ||
900 | if (is_internal_dev(rt_dst(rt).dev)) { | |
e779d8d9 BP |
901 | struct odp_flow_key flow_key; |
902 | struct tbl_node *flow_node; | |
842cf6f4 | 903 | struct vport *vport; |
842cf6f4 JG |
904 | struct sk_buff *skb; |
905 | bool is_frag; | |
e779d8d9 | 906 | int err; |
842cf6f4 JG |
907 | |
908 | vport = internal_dev_get_vport(rt_dst(rt).dev); | |
909 | if (!vport) | |
910 | goto done; | |
911 | ||
842cf6f4 JG |
912 | skb = alloc_skb(cache->len, GFP_ATOMIC); |
913 | if (!skb) | |
914 | goto done; | |
915 | ||
916 | __skb_put(skb, cache->len); | |
917 | memcpy(skb->data, get_cached_header(cache), cache->len); | |
918 | ||
e779d8d9 | 919 | err = flow_extract(skb, vport->port_no, &flow_key, &is_frag); |
842cf6f4 JG |
920 | |
921 | kfree_skb(skb); | |
922 | if (err || is_frag) | |
923 | goto done; | |
924 | ||
e779d8d9 | 925 | flow_node = tbl_lookup(rcu_dereference(vport->dp->table), |
842cf6f4 JG |
926 | &flow_key, flow_hash(&flow_key), |
927 | flow_cmp); | |
928 | if (flow_node) { | |
929 | struct sw_flow *flow = flow_cast(flow_node); | |
930 | ||
931 | cache->flow = flow; | |
932 | flow_hold(flow); | |
933 | } | |
d1eb60cc JG |
934 | } |
935 | ||
842cf6f4 JG |
936 | done: |
937 | assign_cache_rcu(vport, cache); | |
938 | ||
939 | unlock: | |
940 | spin_unlock_bh(&tnl_vport->cache_lock); | |
941 | ||
942 | return cache; | |
d1eb60cc JG |
943 | } |
944 | ||
842cf6f4 JG |
945 | static struct rtable *find_route(struct vport *vport, |
946 | const struct tnl_mutable_config *mutable, | |
947 | u8 tos, struct tnl_cache **cache) | |
d1eb60cc | 948 | { |
842cf6f4 JG |
949 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); |
950 | struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache); | |
951 | ||
952 | *cache = NULL; | |
953 | tos = RT_TOS(tos); | |
954 | ||
955 | if (likely(tos == mutable->port_config.tos && | |
956 | check_cache_valid(cur_cache, mutable))) { | |
957 | *cache = cur_cache; | |
958 | return cur_cache->rt; | |
959 | } else { | |
960 | struct rtable *rt; | |
961 | struct flowi fl = { .nl_u = { .ip4_u = | |
962 | { .daddr = mutable->port_config.daddr, | |
963 | .saddr = mutable->port_config.saddr, | |
964 | .tos = tos } }, | |
965 | .proto = tnl_vport->tnl_ops->ipproto }; | |
966 | ||
967 | if (unlikely(ip_route_output_key(&init_net, &rt, &fl))) | |
968 | return NULL; | |
969 | ||
970 | if (likely(tos == mutable->port_config.tos)) | |
971 | *cache = build_cache(vport, mutable, rt); | |
972 | ||
973 | return rt; | |
d1eb60cc JG |
974 | } |
975 | } | |
976 | ||
842cf6f4 | 977 | static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom) |
d1eb60cc | 978 | { |
842cf6f4 JG |
979 | if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) { |
980 | struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16); | |
981 | if (unlikely(!nskb)) { | |
982 | kfree_skb(skb); | |
983 | return ERR_PTR(-ENOMEM); | |
984 | } | |
d1eb60cc | 985 | |
842cf6f4 | 986 | set_skb_csum_bits(skb, nskb); |
d1eb60cc | 987 | |
842cf6f4 JG |
988 | if (skb->sk) |
989 | skb_set_owner_w(nskb, skb->sk); | |
d1eb60cc | 990 | |
842cf6f4 JG |
991 | kfree_skb(skb); |
992 | return nskb; | |
993 | } | |
d1eb60cc | 994 | |
842cf6f4 | 995 | return skb; |
d1eb60cc JG |
996 | } |
997 | ||
842cf6f4 | 998 | static inline bool need_linearize(const struct sk_buff *skb) |
d1eb60cc | 999 | { |
842cf6f4 JG |
1000 | int i; |
1001 | ||
1002 | if (unlikely(skb_shinfo(skb)->frag_list)) | |
1003 | return true; | |
1004 | ||
1005 | /* | |
1006 | * Generally speaking we should linearize if there are paged frags. | |
1007 | * However, if all of the refcounts are 1 we know nobody else can | |
1008 | * change them from underneath us and we can skip the linearization. | |
1009 | */ | |
1010 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) | |
1011 | if (unlikely(page_count(skb_shinfo(skb)->frags[0].page) > 1)) | |
1012 | return true; | |
1013 | ||
1014 | return false; | |
1015 | } | |
1016 | ||
1017 | static struct sk_buff *handle_offloads(struct sk_buff *skb, | |
1018 | const struct tnl_mutable_config *mutable, | |
1019 | const struct rtable *rt) | |
1020 | { | |
1021 | int min_headroom; | |
d1eb60cc | 1022 | int err; |
d1eb60cc | 1023 | |
842cf6f4 | 1024 | forward_ip_summed(skb); |
d1eb60cc | 1025 | |
842cf6f4 | 1026 | err = vswitch_skb_checksum_setup(skb); |
d1eb60cc JG |
1027 | if (unlikely(err)) |
1028 | goto error_free; | |
1029 | ||
842cf6f4 JG |
1030 | min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len |
1031 | + mutable->tunnel_hlen; | |
d1eb60cc | 1032 | |
842cf6f4 JG |
1033 | if (skb_is_gso(skb)) { |
1034 | struct sk_buff *nskb; | |
1035 | ||
1036 | /* | |
1037 | * If we are doing GSO on a pskb it is better to make sure that | |
1038 | * the headroom is correct now. We will only have to copy the | |
1039 | * portion in the linear data area and GSO will preserve | |
1040 | * headroom when it creates the segments. This is particularly | |
1041 | * beneficial on Xen where we get a lot of GSO pskbs. | |
1042 | * Conversely, we avoid copying if it is just to get our own | |
1043 | * writable clone because GSO will do the copy for us. | |
1044 | */ | |
1045 | if (skb_headroom(skb) < min_headroom) { | |
1046 | skb = check_headroom(skb, min_headroom); | |
1047 | if (unlikely(IS_ERR(skb))) { | |
1048 | err = PTR_ERR(skb); | |
1049 | goto error; | |
1050 | } | |
d1eb60cc JG |
1051 | } |
1052 | ||
842cf6f4 JG |
1053 | nskb = skb_gso_segment(skb, 0); |
1054 | kfree_skb(skb); | |
1055 | if (unlikely(IS_ERR(nskb))) { | |
1056 | err = PTR_ERR(nskb); | |
1057 | goto error; | |
1058 | } | |
d1eb60cc | 1059 | |
842cf6f4 JG |
1060 | skb = nskb; |
1061 | } else { | |
1062 | skb = check_headroom(skb, min_headroom); | |
1063 | if (unlikely(IS_ERR(skb))) { | |
1064 | err = PTR_ERR(skb); | |
1065 | goto error; | |
1066 | } | |
d1eb60cc | 1067 | |
842cf6f4 JG |
1068 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1069 | /* | |
1070 | * Pages aren't locked and could change at any time. | |
1071 | * If this happens after we compute the checksum, the | |
1072 | * checksum will be wrong. We linearize now to avoid | |
1073 | * this problem. | |
1074 | */ | |
1075 | if (unlikely(need_linearize(skb))) { | |
1076 | err = __skb_linearize(skb); | |
1077 | if (unlikely(err)) | |
1078 | goto error_free; | |
1079 | } | |
1080 | ||
1081 | err = skb_checksum_help(skb); | |
1082 | if (unlikely(err)) | |
d1eb60cc | 1083 | goto error_free; |
842cf6f4 JG |
1084 | } else if (skb->ip_summed == CHECKSUM_COMPLETE) |
1085 | skb->ip_summed = CHECKSUM_NONE; | |
d1eb60cc | 1086 | } |
d1eb60cc | 1087 | |
842cf6f4 | 1088 | return skb; |
d1eb60cc | 1089 | |
842cf6f4 JG |
1090 | error_free: |
1091 | kfree_skb(skb); | |
1092 | error: | |
1093 | return ERR_PTR(err); | |
1094 | } | |
d1eb60cc | 1095 | |
842cf6f4 JG |
1096 | static int send_frags(struct sk_buff *skb, |
1097 | const struct tnl_mutable_config *mutable) | |
1098 | { | |
1099 | int sent_len; | |
1100 | int err; | |
d1eb60cc | 1101 | |
842cf6f4 | 1102 | sent_len = 0; |
5214f5c4 JG |
1103 | while (skb) { |
1104 | struct sk_buff *next = skb->next; | |
1105 | int frag_len = skb->len - mutable->tunnel_hlen; | |
d1eb60cc | 1106 | |
5214f5c4 | 1107 | skb->next = NULL; |
b1195d37 | 1108 | memset(IPCB(skb), 0, sizeof(*IPCB(skb))); |
7da5c939 | 1109 | |
5214f5c4 | 1110 | err = ip_local_out(skb); |
842cf6f4 JG |
1111 | if (likely(net_xmit_eval(err) == 0)) |
1112 | sent_len += frag_len; | |
1113 | else { | |
5214f5c4 JG |
1114 | skb = next; |
1115 | goto free_frags; | |
1116 | } | |
1117 | ||
1118 | skb = next; | |
842cf6f4 | 1119 | } |
5214f5c4 | 1120 | |
842cf6f4 | 1121 | return sent_len; |
d1eb60cc | 1122 | |
5214f5c4 JG |
1123 | free_frags: |
1124 | /* | |
1125 | * There's no point in continuing to send fragments once one has been | |
1126 | * dropped so just free the rest. This may help improve the congestion | |
1127 | * that caused the first packet to be dropped. | |
1128 | */ | |
842cf6f4 JG |
1129 | tnl_free_linked_skbs(skb); |
1130 | return sent_len; | |
d1eb60cc JG |
1131 | } |
1132 | ||
1133 | int tnl_send(struct vport *vport, struct sk_buff *skb) | |
1134 | { | |
1135 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
1136 | const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable); | |
1137 | ||
842cf6f4 | 1138 | enum vport_err_type err = VPORT_E_TX_ERROR; |
d1eb60cc | 1139 | struct rtable *rt; |
842cf6f4 JG |
1140 | struct dst_entry *unattached_dst = NULL; |
1141 | struct tnl_cache *cache; | |
1142 | int sent_len = 0; | |
1143 | __be16 frag_off; | |
1144 | u8 ttl; | |
1145 | u8 inner_tos; | |
1146 | u8 tos; | |
d1eb60cc JG |
1147 | |
1148 | /* Validate the protocol headers before we try to use them. */ | |
1149 | if (skb->protocol == htons(ETH_P_8021Q)) { | |
1150 | if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN))) | |
1151 | goto error_free; | |
1152 | ||
1153 | skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto; | |
1154 | skb_set_network_header(skb, VLAN_ETH_HLEN); | |
1155 | } | |
1156 | ||
1157 | if (skb->protocol == htons(ETH_P_IP)) { | |
842cf6f4 JG |
1158 | if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb) |
1159 | + sizeof(struct iphdr)))) | |
d1eb60cc JG |
1160 | skb->protocol = 0; |
1161 | } | |
1162 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
1163 | else if (skb->protocol == htons(ETH_P_IPV6)) { | |
842cf6f4 JG |
1164 | if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb) |
1165 | + sizeof(struct ipv6hdr)))) | |
d1eb60cc JG |
1166 | skb->protocol = 0; |
1167 | } | |
1168 | #endif | |
d1eb60cc | 1169 | |
842cf6f4 JG |
1170 | /* ToS */ |
1171 | if (skb->protocol == htons(ETH_P_IP)) | |
1172 | inner_tos = ip_hdr(skb)->tos; | |
d1eb60cc | 1173 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
842cf6f4 JG |
1174 | else if (skb->protocol == htons(ETH_P_IPV6)) |
1175 | inner_tos = ipv6_get_dsfield(ipv6_hdr(skb)); | |
d1eb60cc | 1176 | #endif |
842cf6f4 JG |
1177 | else |
1178 | inner_tos = 0; | |
d1eb60cc | 1179 | |
842cf6f4 JG |
1180 | if (mutable->port_config.flags & TNL_F_TOS_INHERIT) |
1181 | tos = inner_tos; | |
1182 | else | |
1183 | tos = mutable->port_config.tos; | |
d1eb60cc | 1184 | |
842cf6f4 JG |
1185 | tos = INET_ECN_encapsulate(tos, inner_tos); |
1186 | ||
1187 | /* Route lookup */ | |
1188 | rt = find_route(vport, mutable, tos, &cache); | |
1189 | if (unlikely(!rt)) | |
1190 | goto error_free; | |
1191 | if (unlikely(!cache)) | |
1192 | unattached_dst = &rt_dst(rt); | |
1193 | ||
1194 | /* Reset SKB */ | |
1195 | nf_reset(skb); | |
1196 | secpath_reset(skb); | |
1197 | skb_dst_drop(skb); | |
1198 | ||
1199 | /* Offloading */ | |
1200 | skb = handle_offloads(skb, mutable, rt); | |
1201 | if (unlikely(IS_ERR(skb))) | |
1202 | goto error; | |
1203 | ||
1204 | /* MTU */ | |
1205 | if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) { | |
1206 | err = VPORT_E_TX_DROPPED; | |
1207 | goto error_free; | |
d1eb60cc JG |
1208 | } |
1209 | ||
842cf6f4 JG |
1210 | /* |
1211 | * If we are over the MTU, allow the IP stack to handle fragmentation. | |
1212 | * Fragmentation is a slow path anyways. | |
1213 | */ | |
1214 | if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) && | |
1215 | cache)) { | |
1216 | unattached_dst = &rt_dst(rt); | |
1217 | dst_hold(unattached_dst); | |
1218 | cache = NULL; | |
1219 | } | |
1220 | ||
1221 | /* TTL */ | |
1222 | ttl = mutable->port_config.ttl; | |
1223 | if (!ttl) | |
1224 | ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT); | |
1225 | ||
d1eb60cc JG |
1226 | if (mutable->port_config.flags & TNL_F_TTL_INHERIT) { |
1227 | if (skb->protocol == htons(ETH_P_IP)) | |
842cf6f4 | 1228 | ttl = ip_hdr(skb)->ttl; |
d1eb60cc JG |
1229 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
1230 | else if (skb->protocol == htons(ETH_P_IPV6)) | |
842cf6f4 | 1231 | ttl = ipv6_hdr(skb)->hop_limit; |
d1eb60cc JG |
1232 | #endif |
1233 | } | |
d1eb60cc | 1234 | |
842cf6f4 JG |
1235 | while (skb) { |
1236 | struct iphdr *iph; | |
1237 | struct sk_buff *next_skb = skb->next; | |
1238 | skb->next = NULL; | |
d1eb60cc | 1239 | |
842cf6f4 JG |
1240 | if (likely(cache)) { |
1241 | skb_push(skb, cache->len); | |
1242 | memcpy(skb->data, get_cached_header(cache), cache->len); | |
1243 | skb_reset_mac_header(skb); | |
1244 | skb_set_network_header(skb, rt_dst(rt).hh->hh_len); | |
d1eb60cc | 1245 | |
842cf6f4 JG |
1246 | } else { |
1247 | skb_push(skb, mutable->tunnel_hlen); | |
1248 | create_tunnel_header(vport, mutable, rt, skb->data); | |
1249 | skb_reset_network_header(skb); | |
d1eb60cc | 1250 | |
842cf6f4 JG |
1251 | if (next_skb) |
1252 | skb_dst_set(skb, dst_clone(unattached_dst)); | |
1253 | else { | |
1254 | skb_dst_set(skb, unattached_dst); | |
1255 | unattached_dst = NULL; | |
1256 | } | |
d1eb60cc | 1257 | } |
842cf6f4 | 1258 | skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr)); |
d1eb60cc | 1259 | |
842cf6f4 JG |
1260 | iph = ip_hdr(skb); |
1261 | iph->tos = tos; | |
1262 | iph->ttl = ttl; | |
1263 | iph->frag_off = frag_off; | |
1264 | ip_select_ident(iph, &rt_dst(rt), NULL); | |
d1eb60cc | 1265 | |
842cf6f4 JG |
1266 | skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb); |
1267 | if (unlikely(!skb)) | |
1268 | goto next; | |
d1eb60cc | 1269 | |
842cf6f4 JG |
1270 | if (likely(cache)) { |
1271 | int orig_len = skb->len - cache->len; | |
17a07f9f | 1272 | struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev); |
d1eb60cc | 1273 | |
842cf6f4 | 1274 | skb->protocol = htons(ETH_P_IP); |
842cf6f4 JG |
1275 | iph->tot_len = htons(skb->len - skb_network_offset(skb)); |
1276 | ip_send_check(iph); | |
d1eb60cc | 1277 | |
17a07f9f | 1278 | if (cache_vport) { |
842cf6f4 JG |
1279 | OVS_CB(skb)->flow = cache->flow; |
1280 | compute_ip_summed(skb, true); | |
17a07f9f | 1281 | vport_receive(cache_vport, skb); |
842cf6f4 JG |
1282 | sent_len += orig_len; |
1283 | } else { | |
1284 | int err; | |
d1eb60cc | 1285 | |
842cf6f4 JG |
1286 | skb->dev = rt_dst(rt).dev; |
1287 | err = dev_queue_xmit(skb); | |
1288 | ||
1289 | if (likely(net_xmit_eval(err) == 0)) | |
1290 | sent_len += orig_len; | |
1291 | } | |
1292 | } else | |
1293 | sent_len += send_frags(skb, mutable); | |
1294 | ||
1295 | next: | |
d1eb60cc | 1296 | skb = next_skb; |
842cf6f4 | 1297 | } |
d1eb60cc | 1298 | |
842cf6f4 | 1299 | if (unlikely(sent_len == 0)) |
5214f5c4 JG |
1300 | vport_record_error(vport, VPORT_E_TX_DROPPED); |
1301 | ||
842cf6f4 | 1302 | goto out; |
d1eb60cc JG |
1303 | |
1304 | error_free: | |
842cf6f4 | 1305 | tnl_free_linked_skbs(skb); |
d1eb60cc | 1306 | error: |
842cf6f4 JG |
1307 | dst_release(unattached_dst); |
1308 | vport_record_error(vport, err); | |
1309 | out: | |
1310 | return sent_len; | |
d1eb60cc JG |
1311 | } |
1312 | ||
c3827f61 | 1313 | static int set_config(const void *config, const struct tnl_ops *tnl_ops, |
d1eb60cc JG |
1314 | const struct vport *cur_vport, |
1315 | struct tnl_mutable_config *mutable) | |
1316 | { | |
1317 | const struct vport *old_vport; | |
1318 | const struct tnl_mutable_config *old_mutable; | |
1319 | ||
c3827f61 | 1320 | mutable->port_config = *(struct tnl_port_config *)config; |
d1eb60cc | 1321 | |
842cf6f4 JG |
1322 | if (mutable->port_config.daddr == 0) |
1323 | return -EINVAL; | |
1324 | ||
1325 | if (mutable->port_config.tos != RT_TOS(mutable->port_config.tos)) | |
1326 | return -EINVAL; | |
1327 | ||
d1eb60cc JG |
1328 | mutable->tunnel_hlen = tnl_ops->hdr_len(&mutable->port_config); |
1329 | if (mutable->tunnel_hlen < 0) | |
1330 | return mutable->tunnel_hlen; | |
1331 | ||
1332 | mutable->tunnel_hlen += sizeof(struct iphdr); | |
1333 | ||
d1eb60cc JG |
1334 | mutable->tunnel_type = tnl_ops->tunnel_type; |
1335 | if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) { | |
1336 | mutable->tunnel_type |= TNL_T_KEY_MATCH; | |
1337 | mutable->port_config.in_key = 0; | |
1338 | } else | |
1339 | mutable->tunnel_type |= TNL_T_KEY_EXACT; | |
1340 | ||
1341 | old_vport = tnl_find_port(mutable->port_config.saddr, | |
1342 | mutable->port_config.daddr, | |
1343 | mutable->port_config.in_key, | |
1344 | mutable->tunnel_type, | |
1345 | &old_mutable); | |
1346 | ||
1347 | if (old_vport && old_vport != cur_vport) | |
1348 | return -EEXIST; | |
1349 | ||
1350 | if (mutable->port_config.flags & TNL_F_OUT_KEY_ACTION) | |
1351 | mutable->port_config.out_key = 0; | |
1352 | ||
1353 | return 0; | |
1354 | } | |
1355 | ||
94903c98 | 1356 | struct vport *tnl_create(const struct vport_parms *parms, |
d1eb60cc JG |
1357 | const struct vport_ops *vport_ops, |
1358 | const struct tnl_ops *tnl_ops) | |
1359 | { | |
1360 | struct vport *vport; | |
1361 | struct tnl_vport *tnl_vport; | |
5214f5c4 | 1362 | int initial_frag_id; |
d1eb60cc JG |
1363 | int err; |
1364 | ||
e779d8d9 | 1365 | vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms); |
d1eb60cc JG |
1366 | if (IS_ERR(vport)) { |
1367 | err = PTR_ERR(vport); | |
1368 | goto error; | |
1369 | } | |
1370 | ||
1371 | tnl_vport = tnl_vport_priv(vport); | |
1372 | ||
94903c98 | 1373 | strcpy(tnl_vport->name, parms->name); |
d1eb60cc JG |
1374 | tnl_vport->tnl_ops = tnl_ops; |
1375 | ||
842cf6f4 | 1376 | tnl_vport->mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL); |
d1eb60cc JG |
1377 | if (!tnl_vport->mutable) { |
1378 | err = -ENOMEM; | |
1379 | goto error_free_vport; | |
1380 | } | |
1381 | ||
1382 | vport_gen_rand_ether_addr(tnl_vport->mutable->eth_addr); | |
1383 | tnl_vport->mutable->mtu = ETH_DATA_LEN; | |
1384 | ||
5214f5c4 JG |
1385 | get_random_bytes(&initial_frag_id, sizeof(int)); |
1386 | atomic_set(&tnl_vport->frag_id, initial_frag_id); | |
1387 | ||
94903c98 | 1388 | err = set_config(parms->config, tnl_ops, NULL, tnl_vport->mutable); |
d1eb60cc JG |
1389 | if (err) |
1390 | goto error_free_mutable; | |
1391 | ||
842cf6f4 JG |
1392 | spin_lock_init(&tnl_vport->cache_lock); |
1393 | ||
1394 | #ifdef NEED_CACHE_TIMEOUT | |
1395 | tnl_vport->cache_exp_interval = MAX_CACHE_EXP - | |
1396 | (net_random() % (MAX_CACHE_EXP / 2)); | |
1397 | #endif | |
1398 | ||
d1eb60cc JG |
1399 | err = add_port(vport); |
1400 | if (err) | |
1401 | goto error_free_mutable; | |
1402 | ||
1403 | return vport; | |
1404 | ||
1405 | error_free_mutable: | |
1406 | kfree(tnl_vport->mutable); | |
1407 | error_free_vport: | |
1408 | vport_free(vport); | |
1409 | error: | |
1410 | return ERR_PTR(err); | |
1411 | } | |
1412 | ||
c3827f61 | 1413 | int tnl_modify(struct vport *vport, struct odp_port *port) |
d1eb60cc JG |
1414 | { |
1415 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
1416 | struct tnl_mutable_config *mutable; | |
1417 | int err; | |
d1eb60cc JG |
1418 | |
1419 | mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL); | |
1420 | if (!mutable) { | |
1421 | err = -ENOMEM; | |
1422 | goto error; | |
1423 | } | |
1424 | ||
c3827f61 | 1425 | err = set_config(port->config, tnl_vport->tnl_ops, vport, mutable); |
d1eb60cc JG |
1426 | if (err) |
1427 | goto error_free; | |
1428 | ||
842cf6f4 | 1429 | mutable->seq++; |
d1eb60cc | 1430 | |
842cf6f4 JG |
1431 | err = move_port(vport, mutable); |
1432 | if (err) | |
1433 | goto error_free; | |
d1eb60cc JG |
1434 | |
1435 | return 0; | |
1436 | ||
1437 | error_free: | |
1438 | kfree(mutable); | |
1439 | error: | |
1440 | return err; | |
1441 | } | |
1442 | ||
842cf6f4 | 1443 | static void free_port_rcu(struct rcu_head *rcu) |
d1eb60cc JG |
1444 | { |
1445 | struct tnl_vport *tnl_vport = container_of(rcu, struct tnl_vport, rcu); | |
1446 | ||
842cf6f4 JG |
1447 | spin_lock_bh(&tnl_vport->cache_lock); |
1448 | free_cache(tnl_vport->cache); | |
1449 | spin_unlock_bh(&tnl_vport->cache_lock); | |
1450 | ||
d1eb60cc JG |
1451 | kfree(tnl_vport->mutable); |
1452 | vport_free(tnl_vport_to_vport(tnl_vport)); | |
1453 | } | |
1454 | ||
1455 | int tnl_destroy(struct vport *vport) | |
1456 | { | |
1457 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
1458 | const struct tnl_mutable_config *old_mutable; | |
1459 | ||
1460 | if (vport == tnl_find_port(tnl_vport->mutable->port_config.saddr, | |
1461 | tnl_vport->mutable->port_config.daddr, | |
1462 | tnl_vport->mutable->port_config.in_key, | |
1463 | tnl_vport->mutable->tunnel_type, | |
1464 | &old_mutable)) | |
1465 | del_port(vport); | |
1466 | ||
842cf6f4 | 1467 | call_rcu(&tnl_vport->rcu, free_port_rcu); |
d1eb60cc JG |
1468 | |
1469 | return 0; | |
1470 | } | |
1471 | ||
1472 | int tnl_set_mtu(struct vport *vport, int mtu) | |
1473 | { | |
1474 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
1475 | struct tnl_mutable_config *mutable; | |
1476 | ||
1477 | mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL); | |
1478 | if (!mutable) | |
1479 | return -ENOMEM; | |
1480 | ||
1481 | mutable->mtu = mtu; | |
1482 | assign_config_rcu(vport, mutable); | |
1483 | ||
1484 | return 0; | |
1485 | } | |
1486 | ||
1487 | int tnl_set_addr(struct vport *vport, const unsigned char *addr) | |
1488 | { | |
1489 | struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
1490 | struct tnl_mutable_config *mutable; | |
1491 | ||
1492 | mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL); | |
1493 | if (!mutable) | |
1494 | return -ENOMEM; | |
1495 | ||
1496 | memcpy(mutable->eth_addr, addr, ETH_ALEN); | |
1497 | assign_config_rcu(vport, mutable); | |
1498 | ||
1499 | return 0; | |
1500 | } | |
1501 | ||
d1eb60cc JG |
1502 | const char *tnl_get_name(const struct vport *vport) |
1503 | { | |
1504 | const struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
1505 | return tnl_vport->name; | |
1506 | } | |
1507 | ||
1508 | const unsigned char *tnl_get_addr(const struct vport *vport) | |
1509 | { | |
1510 | const struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
1511 | return rcu_dereference(tnl_vport->mutable)->eth_addr; | |
1512 | } | |
1513 | ||
1514 | int tnl_get_mtu(const struct vport *vport) | |
1515 | { | |
1516 | const struct tnl_vport *tnl_vport = tnl_vport_priv(vport); | |
1517 | return rcu_dereference(tnl_vport->mutable)->mtu; | |
1518 | } | |
842cf6f4 JG |
1519 | |
1520 | void tnl_free_linked_skbs(struct sk_buff *skb) | |
1521 | { | |
1522 | if (unlikely(!skb)) | |
1523 | return; | |
1524 | ||
1525 | while (skb) { | |
1526 | struct sk_buff *next = skb->next; | |
1527 | kfree_skb(skb); | |
1528 | skb = next; | |
1529 | } | |
1530 | } |