]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPv6 fragment reassembly | |
1ab1457c | 3 | * Linux INET6 implementation |
1da177e4 LT |
4 | * |
5 | * Authors: | |
1ab1457c | 6 | * Pedro Roque <roque@di.fc.ul.pt> |
1da177e4 LT |
7 | * |
8 | * $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $ | |
9 | * | |
10 | * Based on: net/ipv4/ip_fragment.c | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or | |
13 | * modify it under the terms of the GNU General Public License | |
14 | * as published by the Free Software Foundation; either version | |
15 | * 2 of the License, or (at your option) any later version. | |
16 | */ | |
17 | ||
1ab1457c YH |
18 | /* |
19 | * Fixes: | |
1da177e4 LT |
20 | * Andi Kleen Make it work with multiple hosts. |
21 | * More RFC compliance. | |
22 | * | |
23 | * Horst von Brand Add missing #include <linux/string.h> | |
24 | * Alexey Kuznetsov SMP races, threading, cleanup. | |
25 | * Patrick McHardy LRU queue of frag heads for evictor. | |
26 | * Mitsuru KANDA @USAGI Register inet6_protocol{}. | |
27 | * David Stevens and | |
28 | * YOSHIFUJI,H. @USAGI Always remove fragment header to | |
29 | * calculate ICV correctly. | |
30 | */ | |
1da177e4 LT |
31 | #include <linux/errno.h> |
32 | #include <linux/types.h> | |
33 | #include <linux/string.h> | |
34 | #include <linux/socket.h> | |
35 | #include <linux/sockios.h> | |
36 | #include <linux/jiffies.h> | |
37 | #include <linux/net.h> | |
38 | #include <linux/list.h> | |
39 | #include <linux/netdevice.h> | |
40 | #include <linux/in6.h> | |
41 | #include <linux/ipv6.h> | |
42 | #include <linux/icmpv6.h> | |
43 | #include <linux/random.h> | |
44 | #include <linux/jhash.h> | |
f61944ef | 45 | #include <linux/skbuff.h> |
1da177e4 LT |
46 | |
47 | #include <net/sock.h> | |
48 | #include <net/snmp.h> | |
49 | ||
50 | #include <net/ipv6.h> | |
a11d206d | 51 | #include <net/ip6_route.h> |
1da177e4 LT |
52 | #include <net/protocol.h> |
53 | #include <net/transp_v6.h> | |
54 | #include <net/rawv6.h> | |
55 | #include <net/ndisc.h> | |
56 | #include <net/addrconf.h> | |
5ab11c98 | 57 | #include <net/inet_frag.h> |
1da177e4 | 58 | |
1da177e4 LT |
59 | struct ip6frag_skb_cb |
60 | { | |
61 | struct inet6_skb_parm h; | |
62 | int offset; | |
63 | }; | |
64 | ||
65 | #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb)) | |
66 | ||
67 | ||
68 | /* | |
69 | * Equivalent of ipv4 struct ipq | |
70 | */ | |
71 | ||
72 | struct frag_queue | |
73 | { | |
5ab11c98 | 74 | struct inet_frag_queue q; |
1da177e4 | 75 | |
e69a4adc | 76 | __be32 id; /* fragment id */ |
1da177e4 LT |
77 | struct in6_addr saddr; |
78 | struct in6_addr daddr; | |
79 | ||
1da177e4 | 80 | int iif; |
1da177e4 | 81 | unsigned int csum; |
1da177e4 | 82 | __u16 nhoffset; |
1da177e4 LT |
83 | }; |
84 | ||
04128f23 PE |
85 | struct inet_frags_ctl ip6_frags_ctl __read_mostly = { |
86 | .high_thresh = 256 * 1024, | |
87 | .low_thresh = 192 * 1024, | |
88 | .timeout = IPV6_FRAG_TIMEOUT, | |
89 | .secret_interval = 10 * 60 * HZ, | |
90 | }; | |
91 | ||
7eb95156 | 92 | static struct inet_frags ip6_frags; |
1da177e4 | 93 | |
7eb95156 PE |
94 | int ip6_frag_nqueues(void) |
95 | { | |
96 | return ip6_frags.nqueues; | |
97 | } | |
1da177e4 | 98 | |
7eb95156 PE |
99 | int ip6_frag_mem(void) |
100 | { | |
101 | return atomic_read(&ip6_frags.mem); | |
102 | } | |
1da177e4 | 103 | |
f61944ef HX |
104 | static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev, |
105 | struct net_device *dev); | |
106 | ||
f6596f9d ZB |
107 | /* |
108 | * callers should be careful not to use the hash value outside the ipfrag_lock | |
109 | * as doing so could race with ipfrag_hash_rnd being recalculated. | |
110 | */ | |
e69a4adc | 111 | static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr, |
1da177e4 LT |
112 | struct in6_addr *daddr) |
113 | { | |
114 | u32 a, b, c; | |
115 | ||
e69a4adc AV |
116 | a = (__force u32)saddr->s6_addr32[0]; |
117 | b = (__force u32)saddr->s6_addr32[1]; | |
118 | c = (__force u32)saddr->s6_addr32[2]; | |
1da177e4 LT |
119 | |
120 | a += JHASH_GOLDEN_RATIO; | |
121 | b += JHASH_GOLDEN_RATIO; | |
7eb95156 | 122 | c += ip6_frags.rnd; |
1da177e4 LT |
123 | __jhash_mix(a, b, c); |
124 | ||
e69a4adc AV |
125 | a += (__force u32)saddr->s6_addr32[3]; |
126 | b += (__force u32)daddr->s6_addr32[0]; | |
127 | c += (__force u32)daddr->s6_addr32[1]; | |
1da177e4 LT |
128 | __jhash_mix(a, b, c); |
129 | ||
e69a4adc AV |
130 | a += (__force u32)daddr->s6_addr32[2]; |
131 | b += (__force u32)daddr->s6_addr32[3]; | |
132 | c += (__force u32)id; | |
1da177e4 LT |
133 | __jhash_mix(a, b, c); |
134 | ||
7eb95156 | 135 | return c & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
136 | } |
137 | ||
321a3a99 | 138 | static unsigned int ip6_hashfn(struct inet_frag_queue *q) |
1da177e4 | 139 | { |
321a3a99 | 140 | struct frag_queue *fq; |
1da177e4 | 141 | |
321a3a99 PE |
142 | fq = container_of(q, struct frag_queue, q); |
143 | return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr); | |
1da177e4 LT |
144 | } |
145 | ||
1da177e4 LT |
146 | /* Memory Tracking Functions. */ |
147 | static inline void frag_kfree_skb(struct sk_buff *skb, int *work) | |
148 | { | |
149 | if (work) | |
150 | *work -= skb->truesize; | |
7eb95156 | 151 | atomic_sub(skb->truesize, &ip6_frags.mem); |
1da177e4 LT |
152 | kfree_skb(skb); |
153 | } | |
154 | ||
1e4b8287 | 155 | static void ip6_frag_free(struct inet_frag_queue *fq) |
1da177e4 | 156 | { |
1e4b8287 | 157 | kfree(container_of(fq, struct frag_queue, q)); |
1da177e4 LT |
158 | } |
159 | ||
160 | static inline struct frag_queue *frag_alloc_queue(void) | |
161 | { | |
78c784c4 | 162 | struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC); |
1da177e4 LT |
163 | |
164 | if(!fq) | |
165 | return NULL; | |
7eb95156 | 166 | atomic_add(sizeof(struct frag_queue), &ip6_frags.mem); |
1da177e4 LT |
167 | return fq; |
168 | } | |
169 | ||
170 | /* Destruction primitives. */ | |
171 | ||
4b6cb5d8 | 172 | static __inline__ void fq_put(struct frag_queue *fq) |
1da177e4 | 173 | { |
762cc408 | 174 | inet_frag_put(&fq->q, &ip6_frags); |
1da177e4 LT |
175 | } |
176 | ||
177 | /* Kill fq entry. It is not destroyed immediately, | |
178 | * because caller (and someone more) holds reference count. | |
179 | */ | |
180 | static __inline__ void fq_kill(struct frag_queue *fq) | |
181 | { | |
277e650d | 182 | inet_frag_kill(&fq->q, &ip6_frags); |
1da177e4 LT |
183 | } |
184 | ||
a11d206d | 185 | static void ip6_evictor(struct inet6_dev *idev) |
1da177e4 | 186 | { |
8e7999c4 PE |
187 | int evicted; |
188 | ||
189 | evicted = inet_frag_evictor(&ip6_frags); | |
190 | if (evicted) | |
191 | IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted); | |
1da177e4 LT |
192 | } |
193 | ||
194 | static void ip6_frag_expire(unsigned long data) | |
195 | { | |
196 | struct frag_queue *fq = (struct frag_queue *) data; | |
a11d206d | 197 | struct net_device *dev = NULL; |
1da177e4 | 198 | |
5ab11c98 | 199 | spin_lock(&fq->q.lock); |
1da177e4 | 200 | |
5ab11c98 | 201 | if (fq->q.last_in & COMPLETE) |
1da177e4 LT |
202 | goto out; |
203 | ||
204 | fq_kill(fq); | |
205 | ||
881d966b | 206 | dev = dev_get_by_index(&init_net, fq->iif); |
a11d206d YH |
207 | if (!dev) |
208 | goto out; | |
209 | ||
210 | rcu_read_lock(); | |
211 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT); | |
212 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
213 | rcu_read_unlock(); | |
1da177e4 | 214 | |
78c784c4 | 215 | /* Don't send error if the first segment did not arrive. */ |
5ab11c98 | 216 | if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments) |
78c784c4 IO |
217 | goto out; |
218 | ||
78c784c4 IO |
219 | /* |
220 | But use as source device on which LAST ARRIVED | |
221 | segment was received. And do not use fq->dev | |
222 | pointer directly, device might already disappeared. | |
223 | */ | |
5ab11c98 PE |
224 | fq->q.fragments->dev = dev; |
225 | icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev); | |
1da177e4 | 226 | out: |
a11d206d YH |
227 | if (dev) |
228 | dev_put(dev); | |
5ab11c98 | 229 | spin_unlock(&fq->q.lock); |
4b6cb5d8 | 230 | fq_put(fq); |
1da177e4 LT |
231 | } |
232 | ||
233 | /* Creation primitives. */ | |
234 | ||
235 | ||
f6596f9d | 236 | static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in) |
1da177e4 LT |
237 | { |
238 | struct frag_queue *fq; | |
f6596f9d | 239 | unsigned int hash; |
e7c8a41e YK |
240 | #ifdef CONFIG_SMP |
241 | struct hlist_node *n; | |
242 | #endif | |
1da177e4 | 243 | |
7eb95156 | 244 | write_lock(&ip6_frags.lock); |
f6596f9d | 245 | hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr); |
1da177e4 | 246 | #ifdef CONFIG_SMP |
7eb95156 | 247 | hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) { |
1ab1457c | 248 | if (fq->id == fq_in->id && |
1da177e4 LT |
249 | ipv6_addr_equal(&fq_in->saddr, &fq->saddr) && |
250 | ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) { | |
5ab11c98 | 251 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 252 | write_unlock(&ip6_frags.lock); |
5ab11c98 | 253 | fq_in->q.last_in |= COMPLETE; |
4b6cb5d8 | 254 | fq_put(fq_in); |
1da177e4 LT |
255 | return fq; |
256 | } | |
257 | } | |
258 | #endif | |
259 | fq = fq_in; | |
260 | ||
04128f23 | 261 | if (!mod_timer(&fq->q.timer, jiffies + ip6_frags_ctl.timeout)) |
5ab11c98 | 262 | atomic_inc(&fq->q.refcnt); |
1da177e4 | 263 | |
5ab11c98 | 264 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 265 | hlist_add_head(&fq->q.list, &ip6_frags.hash[hash]); |
5ab11c98 | 266 | INIT_LIST_HEAD(&fq->q.lru_list); |
7eb95156 PE |
267 | list_add_tail(&fq->q.lru_list, &ip6_frags.lru_list); |
268 | ip6_frags.nqueues++; | |
269 | write_unlock(&ip6_frags.lock); | |
1da177e4 LT |
270 | return fq; |
271 | } | |
272 | ||
273 | ||
274 | static struct frag_queue * | |
e69a4adc | 275 | ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst, |
a11d206d | 276 | struct inet6_dev *idev) |
1da177e4 LT |
277 | { |
278 | struct frag_queue *fq; | |
279 | ||
280 | if ((fq = frag_alloc_queue()) == NULL) | |
281 | goto oom; | |
282 | ||
1da177e4 LT |
283 | fq->id = id; |
284 | ipv6_addr_copy(&fq->saddr, src); | |
285 | ipv6_addr_copy(&fq->daddr, dst); | |
286 | ||
5ab11c98 PE |
287 | init_timer(&fq->q.timer); |
288 | fq->q.timer.function = ip6_frag_expire; | |
289 | fq->q.timer.data = (long) fq; | |
290 | spin_lock_init(&fq->q.lock); | |
291 | atomic_set(&fq->q.refcnt, 1); | |
1da177e4 | 292 | |
f6596f9d | 293 | return ip6_frag_intern(fq); |
1da177e4 LT |
294 | |
295 | oom: | |
a11d206d | 296 | IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
297 | return NULL; |
298 | } | |
299 | ||
300 | static __inline__ struct frag_queue * | |
e69a4adc | 301 | fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst, |
a11d206d | 302 | struct inet6_dev *idev) |
1da177e4 LT |
303 | { |
304 | struct frag_queue *fq; | |
e7c8a41e | 305 | struct hlist_node *n; |
f6596f9d | 306 | unsigned int hash; |
1da177e4 | 307 | |
7eb95156 | 308 | read_lock(&ip6_frags.lock); |
f6596f9d | 309 | hash = ip6qhashfn(id, src, dst); |
7eb95156 | 310 | hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) { |
1ab1457c | 311 | if (fq->id == id && |
1da177e4 LT |
312 | ipv6_addr_equal(src, &fq->saddr) && |
313 | ipv6_addr_equal(dst, &fq->daddr)) { | |
5ab11c98 | 314 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 315 | read_unlock(&ip6_frags.lock); |
1da177e4 LT |
316 | return fq; |
317 | } | |
318 | } | |
7eb95156 | 319 | read_unlock(&ip6_frags.lock); |
1da177e4 | 320 | |
a11d206d | 321 | return ip6_frag_create(id, src, dst, idev); |
1da177e4 LT |
322 | } |
323 | ||
324 | ||
f61944ef | 325 | static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb, |
1da177e4 LT |
326 | struct frag_hdr *fhdr, int nhoff) |
327 | { | |
328 | struct sk_buff *prev, *next; | |
f61944ef | 329 | struct net_device *dev; |
1da177e4 LT |
330 | int offset, end; |
331 | ||
5ab11c98 | 332 | if (fq->q.last_in & COMPLETE) |
1da177e4 LT |
333 | goto err; |
334 | ||
335 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
0660e03f ACM |
336 | end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - |
337 | ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); | |
1da177e4 LT |
338 | |
339 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
a11d206d YH |
340 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
341 | IPSTATS_MIB_INHDRERRORS); | |
d56f90a7 ACM |
342 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
343 | ((u8 *)&fhdr->frag_off - | |
344 | skb_network_header(skb))); | |
f61944ef | 345 | return -1; |
1da177e4 LT |
346 | } |
347 | ||
d56f90a7 ACM |
348 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
349 | const unsigned char *nh = skb_network_header(skb); | |
1ab1457c | 350 | skb->csum = csum_sub(skb->csum, |
d56f90a7 ACM |
351 | csum_partial(nh, (u8 *)(fhdr + 1) - nh, |
352 | 0)); | |
353 | } | |
1da177e4 LT |
354 | |
355 | /* Is this the final fragment? */ | |
356 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
357 | /* If we already have some bits beyond end | |
358 | * or have different end, the segment is corrupted. | |
359 | */ | |
5ab11c98 PE |
360 | if (end < fq->q.len || |
361 | ((fq->q.last_in & LAST_IN) && end != fq->q.len)) | |
1da177e4 | 362 | goto err; |
5ab11c98 PE |
363 | fq->q.last_in |= LAST_IN; |
364 | fq->q.len = end; | |
1da177e4 LT |
365 | } else { |
366 | /* Check if the fragment is rounded to 8 bytes. | |
367 | * Required by the RFC. | |
368 | */ | |
369 | if (end & 0x7) { | |
370 | /* RFC2460 says always send parameter problem in | |
371 | * this case. -DaveM | |
372 | */ | |
a11d206d YH |
373 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
374 | IPSTATS_MIB_INHDRERRORS); | |
1ab1457c | 375 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
1da177e4 | 376 | offsetof(struct ipv6hdr, payload_len)); |
f61944ef | 377 | return -1; |
1da177e4 | 378 | } |
5ab11c98 | 379 | if (end > fq->q.len) { |
1da177e4 | 380 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 381 | if (fq->q.last_in & LAST_IN) |
1da177e4 | 382 | goto err; |
5ab11c98 | 383 | fq->q.len = end; |
1da177e4 LT |
384 | } |
385 | } | |
386 | ||
387 | if (end == offset) | |
388 | goto err; | |
389 | ||
390 | /* Point into the IP datagram 'data' part. */ | |
391 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) | |
392 | goto err; | |
1ab1457c | 393 | |
42ca89c1 SH |
394 | if (pskb_trim_rcsum(skb, end - offset)) |
395 | goto err; | |
1da177e4 LT |
396 | |
397 | /* Find out which fragments are in front and at the back of us | |
398 | * in the chain of fragments so far. We must know where to put | |
399 | * this fragment, right? | |
400 | */ | |
401 | prev = NULL; | |
5ab11c98 | 402 | for(next = fq->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
403 | if (FRAG6_CB(next)->offset >= offset) |
404 | break; /* bingo! */ | |
405 | prev = next; | |
406 | } | |
407 | ||
408 | /* We found where to put this one. Check for overlap with | |
409 | * preceding fragment, and, if needed, align things so that | |
410 | * any overlaps are eliminated. | |
411 | */ | |
412 | if (prev) { | |
413 | int i = (FRAG6_CB(prev)->offset + prev->len) - offset; | |
414 | ||
415 | if (i > 0) { | |
416 | offset += i; | |
417 | if (end <= offset) | |
418 | goto err; | |
419 | if (!pskb_pull(skb, i)) | |
420 | goto err; | |
421 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
422 | skb->ip_summed = CHECKSUM_NONE; | |
423 | } | |
424 | } | |
425 | ||
426 | /* Look for overlap with succeeding segments. | |
427 | * If we can merge fragments, do it. | |
428 | */ | |
429 | while (next && FRAG6_CB(next)->offset < end) { | |
430 | int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */ | |
431 | ||
432 | if (i < next->len) { | |
433 | /* Eat head of the next overlapped fragment | |
434 | * and leave the loop. The next ones cannot overlap. | |
435 | */ | |
436 | if (!pskb_pull(next, i)) | |
437 | goto err; | |
438 | FRAG6_CB(next)->offset += i; /* next fragment */ | |
5ab11c98 | 439 | fq->q.meat -= i; |
1da177e4 LT |
440 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
441 | next->ip_summed = CHECKSUM_NONE; | |
442 | break; | |
443 | } else { | |
444 | struct sk_buff *free_it = next; | |
445 | ||
446 | /* Old fragment is completely overridden with | |
447 | * new one drop it. | |
448 | */ | |
449 | next = next->next; | |
450 | ||
451 | if (prev) | |
452 | prev->next = next; | |
453 | else | |
5ab11c98 | 454 | fq->q.fragments = next; |
1da177e4 | 455 | |
5ab11c98 | 456 | fq->q.meat -= free_it->len; |
1da177e4 LT |
457 | frag_kfree_skb(free_it, NULL); |
458 | } | |
459 | } | |
460 | ||
461 | FRAG6_CB(skb)->offset = offset; | |
462 | ||
463 | /* Insert this fragment in the chain of fragments. */ | |
464 | skb->next = next; | |
465 | if (prev) | |
466 | prev->next = skb; | |
467 | else | |
5ab11c98 | 468 | fq->q.fragments = skb; |
1da177e4 | 469 | |
f61944ef HX |
470 | dev = skb->dev; |
471 | if (dev) { | |
472 | fq->iif = dev->ifindex; | |
473 | skb->dev = NULL; | |
474 | } | |
5ab11c98 PE |
475 | fq->q.stamp = skb->tstamp; |
476 | fq->q.meat += skb->len; | |
7eb95156 | 477 | atomic_add(skb->truesize, &ip6_frags.mem); |
1da177e4 LT |
478 | |
479 | /* The first fragment. | |
480 | * nhoffset is obtained from the first fragment, of course. | |
481 | */ | |
482 | if (offset == 0) { | |
483 | fq->nhoffset = nhoff; | |
5ab11c98 | 484 | fq->q.last_in |= FIRST_IN; |
1da177e4 | 485 | } |
f61944ef | 486 | |
5ab11c98 | 487 | if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len) |
f61944ef HX |
488 | return ip6_frag_reasm(fq, prev, dev); |
489 | ||
7eb95156 PE |
490 | write_lock(&ip6_frags.lock); |
491 | list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list); | |
492 | write_unlock(&ip6_frags.lock); | |
f61944ef | 493 | return -1; |
1da177e4 LT |
494 | |
495 | err: | |
a11d206d | 496 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 | 497 | kfree_skb(skb); |
f61944ef | 498 | return -1; |
1da177e4 LT |
499 | } |
500 | ||
501 | /* | |
502 | * Check if this packet is complete. | |
503 | * Returns NULL on failure by any reason, and pointer | |
504 | * to current nexthdr field in reassembled frame. | |
505 | * | |
506 | * It is called with locked fq, and caller must check that | |
507 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
508 | * the last and the first frames arrived and all the bits are here. | |
509 | */ | |
f61944ef | 510 | static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev, |
1da177e4 LT |
511 | struct net_device *dev) |
512 | { | |
5ab11c98 | 513 | struct sk_buff *fp, *head = fq->q.fragments; |
1da177e4 LT |
514 | int payload_len; |
515 | unsigned int nhoff; | |
516 | ||
517 | fq_kill(fq); | |
518 | ||
f61944ef HX |
519 | /* Make the one we just received the head. */ |
520 | if (prev) { | |
521 | head = prev->next; | |
522 | fp = skb_clone(head, GFP_ATOMIC); | |
523 | ||
524 | if (!fp) | |
525 | goto out_oom; | |
526 | ||
527 | fp->next = head->next; | |
528 | prev->next = fp; | |
529 | ||
5ab11c98 PE |
530 | skb_morph(head, fq->q.fragments); |
531 | head->next = fq->q.fragments->next; | |
f61944ef | 532 | |
5ab11c98 PE |
533 | kfree_skb(fq->q.fragments); |
534 | fq->q.fragments = head; | |
f61944ef HX |
535 | } |
536 | ||
1da177e4 LT |
537 | BUG_TRAP(head != NULL); |
538 | BUG_TRAP(FRAG6_CB(head)->offset == 0); | |
539 | ||
540 | /* Unfragmented part is taken from the first segment. */ | |
d56f90a7 | 541 | payload_len = ((head->data - skb_network_header(head)) - |
5ab11c98 | 542 | sizeof(struct ipv6hdr) + fq->q.len - |
d56f90a7 | 543 | sizeof(struct frag_hdr)); |
1da177e4 LT |
544 | if (payload_len > IPV6_MAXPLEN) |
545 | goto out_oversize; | |
546 | ||
547 | /* Head of list must not be cloned. */ | |
548 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
549 | goto out_oom; | |
550 | ||
551 | /* If the first fragment is fragmented itself, we split | |
552 | * it to two chunks: the first with data and paged part | |
553 | * and the second, holding only fragments. */ | |
554 | if (skb_shinfo(head)->frag_list) { | |
555 | struct sk_buff *clone; | |
556 | int i, plen = 0; | |
557 | ||
558 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
559 | goto out_oom; | |
560 | clone->next = head->next; | |
561 | head->next = clone; | |
562 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
563 | skb_shinfo(head)->frag_list = NULL; | |
564 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
565 | plen += skb_shinfo(head)->frags[i].size; | |
566 | clone->len = clone->data_len = head->data_len - plen; | |
567 | head->data_len -= clone->len; | |
568 | head->len -= clone->len; | |
569 | clone->csum = 0; | |
570 | clone->ip_summed = head->ip_summed; | |
7eb95156 | 571 | atomic_add(clone->truesize, &ip6_frags.mem); |
1da177e4 LT |
572 | } |
573 | ||
574 | /* We have to remove fragment header from datagram and to relocate | |
575 | * header in order to calculate ICV correctly. */ | |
576 | nhoff = fq->nhoffset; | |
b0e380b1 | 577 | skb_network_header(head)[nhoff] = skb_transport_header(head)[0]; |
1ab1457c | 578 | memmove(head->head + sizeof(struct frag_hdr), head->head, |
1da177e4 | 579 | (head->data - head->head) - sizeof(struct frag_hdr)); |
b0e380b1 ACM |
580 | head->mac_header += sizeof(struct frag_hdr); |
581 | head->network_header += sizeof(struct frag_hdr); | |
1da177e4 LT |
582 | |
583 | skb_shinfo(head)->frag_list = head->next; | |
badff6d0 | 584 | skb_reset_transport_header(head); |
d56f90a7 | 585 | skb_push(head, head->data - skb_network_header(head)); |
7eb95156 | 586 | atomic_sub(head->truesize, &ip6_frags.mem); |
1da177e4 LT |
587 | |
588 | for (fp=head->next; fp; fp = fp->next) { | |
589 | head->data_len += fp->len; | |
590 | head->len += fp->len; | |
591 | if (head->ip_summed != fp->ip_summed) | |
592 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 593 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
594 | head->csum = csum_add(head->csum, fp->csum); |
595 | head->truesize += fp->truesize; | |
7eb95156 | 596 | atomic_sub(fp->truesize, &ip6_frags.mem); |
1da177e4 LT |
597 | } |
598 | ||
599 | head->next = NULL; | |
600 | head->dev = dev; | |
5ab11c98 | 601 | head->tstamp = fq->q.stamp; |
0660e03f | 602 | ipv6_hdr(head)->payload_len = htons(payload_len); |
951dbc8a | 603 | IP6CB(head)->nhoff = nhoff; |
1da177e4 | 604 | |
1da177e4 | 605 | /* Yes, and fold redundant checksum back. 8) */ |
84fa7933 | 606 | if (head->ip_summed == CHECKSUM_COMPLETE) |
d56f90a7 | 607 | head->csum = csum_partial(skb_network_header(head), |
cfe1fc77 | 608 | skb_network_header_len(head), |
d56f90a7 | 609 | head->csum); |
1da177e4 | 610 | |
a11d206d YH |
611 | rcu_read_lock(); |
612 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS); | |
613 | rcu_read_unlock(); | |
5ab11c98 | 614 | fq->q.fragments = NULL; |
1da177e4 LT |
615 | return 1; |
616 | ||
617 | out_oversize: | |
618 | if (net_ratelimit()) | |
619 | printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len); | |
620 | goto out_fail; | |
621 | out_oom: | |
622 | if (net_ratelimit()) | |
623 | printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n"); | |
624 | out_fail: | |
a11d206d YH |
625 | rcu_read_lock(); |
626 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
627 | rcu_read_unlock(); | |
1da177e4 LT |
628 | return -1; |
629 | } | |
630 | ||
951dbc8a | 631 | static int ipv6_frag_rcv(struct sk_buff **skbp) |
1da177e4 | 632 | { |
1ab1457c | 633 | struct sk_buff *skb = *skbp; |
1da177e4 LT |
634 | struct frag_hdr *fhdr; |
635 | struct frag_queue *fq; | |
0660e03f | 636 | struct ipv6hdr *hdr = ipv6_hdr(skb); |
1da177e4 | 637 | |
a11d206d | 638 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
639 | |
640 | /* Jumbo payload inhibits frag. header */ | |
641 | if (hdr->payload_len==0) { | |
a11d206d | 642 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
cfe1fc77 ACM |
643 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
644 | skb_network_header_len(skb)); | |
1da177e4 LT |
645 | return -1; |
646 | } | |
ea2ae17d ACM |
647 | if (!pskb_may_pull(skb, (skb_transport_offset(skb) + |
648 | sizeof(struct frag_hdr)))) { | |
a11d206d | 649 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
cfe1fc77 ACM |
650 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
651 | skb_network_header_len(skb)); | |
1da177e4 LT |
652 | return -1; |
653 | } | |
654 | ||
0660e03f | 655 | hdr = ipv6_hdr(skb); |
9c70220b | 656 | fhdr = (struct frag_hdr *)skb_transport_header(skb); |
1da177e4 LT |
657 | |
658 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
659 | /* It is not a fragmented frame */ | |
b0e380b1 | 660 | skb->transport_header += sizeof(struct frag_hdr); |
a11d206d | 661 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS); |
1da177e4 | 662 | |
d56f90a7 | 663 | IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb); |
1da177e4 LT |
664 | return 1; |
665 | } | |
666 | ||
04128f23 | 667 | if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh) |
a11d206d | 668 | ip6_evictor(ip6_dst_idev(skb->dst)); |
1da177e4 | 669 | |
a11d206d YH |
670 | if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr, |
671 | ip6_dst_idev(skb->dst))) != NULL) { | |
f61944ef | 672 | int ret; |
1da177e4 | 673 | |
5ab11c98 | 674 | spin_lock(&fq->q.lock); |
1da177e4 | 675 | |
f61944ef | 676 | ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff); |
1da177e4 | 677 | |
5ab11c98 | 678 | spin_unlock(&fq->q.lock); |
4b6cb5d8 | 679 | fq_put(fq); |
1da177e4 LT |
680 | return ret; |
681 | } | |
682 | ||
a11d206d | 683 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
684 | kfree_skb(skb); |
685 | return -1; | |
686 | } | |
687 | ||
688 | static struct inet6_protocol frag_protocol = | |
689 | { | |
690 | .handler = ipv6_frag_rcv, | |
691 | .flags = INET6_PROTO_NOPOLICY, | |
692 | }; | |
693 | ||
694 | void __init ipv6_frag_init(void) | |
695 | { | |
696 | if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0) | |
697 | printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n"); | |
698 | ||
04128f23 | 699 | ip6_frags.ctl = &ip6_frags_ctl; |
321a3a99 | 700 | ip6_frags.hashfn = ip6_hashfn; |
1e4b8287 PE |
701 | ip6_frags.destructor = ip6_frag_free; |
702 | ip6_frags.skb_free = NULL; | |
703 | ip6_frags.qsize = sizeof(struct frag_queue); | |
7eb95156 | 704 | inet_frags_init(&ip6_frags); |
1da177e4 | 705 | } |