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 | ||
2588fe1d PE |
146 | int ip6_frag_equal(struct inet_frag_queue *q1, struct inet_frag_queue *q2) |
147 | { | |
148 | struct frag_queue *fq1, *fq2; | |
149 | ||
150 | fq1 = container_of(q1, struct frag_queue, q); | |
151 | fq2 = container_of(q2, struct frag_queue, q); | |
152 | return (fq1->id == fq2->id && | |
153 | ipv6_addr_equal(&fq2->saddr, &fq1->saddr) && | |
154 | ipv6_addr_equal(&fq2->daddr, &fq1->daddr)); | |
155 | } | |
156 | EXPORT_SYMBOL(ip6_frag_equal); | |
157 | ||
abd6523d PE |
158 | int ip6_frag_match(struct inet_frag_queue *q, void *a) |
159 | { | |
160 | struct frag_queue *fq; | |
161 | struct ip6_create_arg *arg = a; | |
162 | ||
163 | fq = container_of(q, struct frag_queue, q); | |
164 | return (fq->id == arg->id && | |
165 | ipv6_addr_equal(&fq->saddr, arg->src) && | |
166 | ipv6_addr_equal(&fq->daddr, arg->dst)); | |
167 | } | |
168 | EXPORT_SYMBOL(ip6_frag_match); | |
169 | ||
1da177e4 LT |
170 | /* Memory Tracking Functions. */ |
171 | static inline void frag_kfree_skb(struct sk_buff *skb, int *work) | |
172 | { | |
173 | if (work) | |
174 | *work -= skb->truesize; | |
7eb95156 | 175 | atomic_sub(skb->truesize, &ip6_frags.mem); |
1da177e4 LT |
176 | kfree_skb(skb); |
177 | } | |
178 | ||
c6fda282 | 179 | void ip6_frag_init(struct inet_frag_queue *q, void *a) |
1da177e4 | 180 | { |
c6fda282 PE |
181 | struct frag_queue *fq = container_of(q, struct frag_queue, q); |
182 | struct ip6_create_arg *arg = a; | |
183 | ||
184 | fq->id = arg->id; | |
185 | ipv6_addr_copy(&fq->saddr, arg->src); | |
186 | ipv6_addr_copy(&fq->daddr, arg->dst); | |
1da177e4 | 187 | } |
c6fda282 | 188 | EXPORT_SYMBOL(ip6_frag_init); |
1da177e4 | 189 | |
c6fda282 | 190 | static void ip6_frag_free(struct inet_frag_queue *fq) |
1da177e4 | 191 | { |
c6fda282 | 192 | kfree(container_of(fq, struct frag_queue, q)); |
1da177e4 LT |
193 | } |
194 | ||
195 | /* Destruction primitives. */ | |
196 | ||
4b6cb5d8 | 197 | static __inline__ void fq_put(struct frag_queue *fq) |
1da177e4 | 198 | { |
762cc408 | 199 | inet_frag_put(&fq->q, &ip6_frags); |
1da177e4 LT |
200 | } |
201 | ||
202 | /* Kill fq entry. It is not destroyed immediately, | |
203 | * because caller (and someone more) holds reference count. | |
204 | */ | |
205 | static __inline__ void fq_kill(struct frag_queue *fq) | |
206 | { | |
277e650d | 207 | inet_frag_kill(&fq->q, &ip6_frags); |
1da177e4 LT |
208 | } |
209 | ||
a11d206d | 210 | static void ip6_evictor(struct inet6_dev *idev) |
1da177e4 | 211 | { |
8e7999c4 PE |
212 | int evicted; |
213 | ||
214 | evicted = inet_frag_evictor(&ip6_frags); | |
215 | if (evicted) | |
216 | IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted); | |
1da177e4 LT |
217 | } |
218 | ||
219 | static void ip6_frag_expire(unsigned long data) | |
220 | { | |
e521db9d | 221 | struct frag_queue *fq; |
a11d206d | 222 | struct net_device *dev = NULL; |
1da177e4 | 223 | |
e521db9d PE |
224 | fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q); |
225 | ||
5ab11c98 | 226 | spin_lock(&fq->q.lock); |
1da177e4 | 227 | |
5ab11c98 | 228 | if (fq->q.last_in & COMPLETE) |
1da177e4 LT |
229 | goto out; |
230 | ||
231 | fq_kill(fq); | |
232 | ||
881d966b | 233 | dev = dev_get_by_index(&init_net, fq->iif); |
a11d206d YH |
234 | if (!dev) |
235 | goto out; | |
236 | ||
237 | rcu_read_lock(); | |
238 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT); | |
239 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
240 | rcu_read_unlock(); | |
1da177e4 | 241 | |
78c784c4 | 242 | /* Don't send error if the first segment did not arrive. */ |
5ab11c98 | 243 | if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments) |
78c784c4 IO |
244 | goto out; |
245 | ||
78c784c4 IO |
246 | /* |
247 | But use as source device on which LAST ARRIVED | |
248 | segment was received. And do not use fq->dev | |
249 | pointer directly, device might already disappeared. | |
250 | */ | |
5ab11c98 PE |
251 | fq->q.fragments->dev = dev; |
252 | icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev); | |
1da177e4 | 253 | out: |
a11d206d YH |
254 | if (dev) |
255 | dev_put(dev); | |
5ab11c98 | 256 | spin_unlock(&fq->q.lock); |
4b6cb5d8 | 257 | fq_put(fq); |
1da177e4 LT |
258 | } |
259 | ||
abd6523d PE |
260 | static __inline__ struct frag_queue * |
261 | fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst, | |
262 | struct inet6_dev *idev) | |
1da177e4 | 263 | { |
c6fda282 PE |
264 | struct inet_frag_queue *q; |
265 | struct ip6_create_arg arg; | |
abd6523d | 266 | unsigned int hash; |
1da177e4 | 267 | |
c6fda282 PE |
268 | arg.id = id; |
269 | arg.src = src; | |
270 | arg.dst = dst; | |
abd6523d | 271 | hash = ip6qhashfn(id, src, dst); |
1da177e4 | 272 | |
abd6523d | 273 | q = inet_frag_find(&ip6_frags, &arg, hash); |
c6fda282 PE |
274 | if (q == NULL) |
275 | goto oom; | |
1da177e4 | 276 | |
c6fda282 | 277 | return container_of(q, struct frag_queue, q); |
1da177e4 LT |
278 | |
279 | oom: | |
a11d206d | 280 | IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
281 | return NULL; |
282 | } | |
283 | ||
f61944ef | 284 | static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb, |
1da177e4 LT |
285 | struct frag_hdr *fhdr, int nhoff) |
286 | { | |
287 | struct sk_buff *prev, *next; | |
f61944ef | 288 | struct net_device *dev; |
1da177e4 LT |
289 | int offset, end; |
290 | ||
5ab11c98 | 291 | if (fq->q.last_in & COMPLETE) |
1da177e4 LT |
292 | goto err; |
293 | ||
294 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
0660e03f ACM |
295 | end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - |
296 | ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); | |
1da177e4 LT |
297 | |
298 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
a11d206d YH |
299 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
300 | IPSTATS_MIB_INHDRERRORS); | |
d56f90a7 ACM |
301 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
302 | ((u8 *)&fhdr->frag_off - | |
303 | skb_network_header(skb))); | |
f61944ef | 304 | return -1; |
1da177e4 LT |
305 | } |
306 | ||
d56f90a7 ACM |
307 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
308 | const unsigned char *nh = skb_network_header(skb); | |
1ab1457c | 309 | skb->csum = csum_sub(skb->csum, |
d56f90a7 ACM |
310 | csum_partial(nh, (u8 *)(fhdr + 1) - nh, |
311 | 0)); | |
312 | } | |
1da177e4 LT |
313 | |
314 | /* Is this the final fragment? */ | |
315 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
316 | /* If we already have some bits beyond end | |
317 | * or have different end, the segment is corrupted. | |
318 | */ | |
5ab11c98 PE |
319 | if (end < fq->q.len || |
320 | ((fq->q.last_in & LAST_IN) && end != fq->q.len)) | |
1da177e4 | 321 | goto err; |
5ab11c98 PE |
322 | fq->q.last_in |= LAST_IN; |
323 | fq->q.len = end; | |
1da177e4 LT |
324 | } else { |
325 | /* Check if the fragment is rounded to 8 bytes. | |
326 | * Required by the RFC. | |
327 | */ | |
328 | if (end & 0x7) { | |
329 | /* RFC2460 says always send parameter problem in | |
330 | * this case. -DaveM | |
331 | */ | |
a11d206d YH |
332 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
333 | IPSTATS_MIB_INHDRERRORS); | |
1ab1457c | 334 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
1da177e4 | 335 | offsetof(struct ipv6hdr, payload_len)); |
f61944ef | 336 | return -1; |
1da177e4 | 337 | } |
5ab11c98 | 338 | if (end > fq->q.len) { |
1da177e4 | 339 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 340 | if (fq->q.last_in & LAST_IN) |
1da177e4 | 341 | goto err; |
5ab11c98 | 342 | fq->q.len = end; |
1da177e4 LT |
343 | } |
344 | } | |
345 | ||
346 | if (end == offset) | |
347 | goto err; | |
348 | ||
349 | /* Point into the IP datagram 'data' part. */ | |
350 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) | |
351 | goto err; | |
1ab1457c | 352 | |
42ca89c1 SH |
353 | if (pskb_trim_rcsum(skb, end - offset)) |
354 | goto err; | |
1da177e4 LT |
355 | |
356 | /* Find out which fragments are in front and at the back of us | |
357 | * in the chain of fragments so far. We must know where to put | |
358 | * this fragment, right? | |
359 | */ | |
360 | prev = NULL; | |
5ab11c98 | 361 | for(next = fq->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
362 | if (FRAG6_CB(next)->offset >= offset) |
363 | break; /* bingo! */ | |
364 | prev = next; | |
365 | } | |
366 | ||
367 | /* We found where to put this one. Check for overlap with | |
368 | * preceding fragment, and, if needed, align things so that | |
369 | * any overlaps are eliminated. | |
370 | */ | |
371 | if (prev) { | |
372 | int i = (FRAG6_CB(prev)->offset + prev->len) - offset; | |
373 | ||
374 | if (i > 0) { | |
375 | offset += i; | |
376 | if (end <= offset) | |
377 | goto err; | |
378 | if (!pskb_pull(skb, i)) | |
379 | goto err; | |
380 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
381 | skb->ip_summed = CHECKSUM_NONE; | |
382 | } | |
383 | } | |
384 | ||
385 | /* Look for overlap with succeeding segments. | |
386 | * If we can merge fragments, do it. | |
387 | */ | |
388 | while (next && FRAG6_CB(next)->offset < end) { | |
389 | int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */ | |
390 | ||
391 | if (i < next->len) { | |
392 | /* Eat head of the next overlapped fragment | |
393 | * and leave the loop. The next ones cannot overlap. | |
394 | */ | |
395 | if (!pskb_pull(next, i)) | |
396 | goto err; | |
397 | FRAG6_CB(next)->offset += i; /* next fragment */ | |
5ab11c98 | 398 | fq->q.meat -= i; |
1da177e4 LT |
399 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
400 | next->ip_summed = CHECKSUM_NONE; | |
401 | break; | |
402 | } else { | |
403 | struct sk_buff *free_it = next; | |
404 | ||
405 | /* Old fragment is completely overridden with | |
406 | * new one drop it. | |
407 | */ | |
408 | next = next->next; | |
409 | ||
410 | if (prev) | |
411 | prev->next = next; | |
412 | else | |
5ab11c98 | 413 | fq->q.fragments = next; |
1da177e4 | 414 | |
5ab11c98 | 415 | fq->q.meat -= free_it->len; |
1da177e4 LT |
416 | frag_kfree_skb(free_it, NULL); |
417 | } | |
418 | } | |
419 | ||
420 | FRAG6_CB(skb)->offset = offset; | |
421 | ||
422 | /* Insert this fragment in the chain of fragments. */ | |
423 | skb->next = next; | |
424 | if (prev) | |
425 | prev->next = skb; | |
426 | else | |
5ab11c98 | 427 | fq->q.fragments = skb; |
1da177e4 | 428 | |
f61944ef HX |
429 | dev = skb->dev; |
430 | if (dev) { | |
431 | fq->iif = dev->ifindex; | |
432 | skb->dev = NULL; | |
433 | } | |
5ab11c98 PE |
434 | fq->q.stamp = skb->tstamp; |
435 | fq->q.meat += skb->len; | |
7eb95156 | 436 | atomic_add(skb->truesize, &ip6_frags.mem); |
1da177e4 LT |
437 | |
438 | /* The first fragment. | |
439 | * nhoffset is obtained from the first fragment, of course. | |
440 | */ | |
441 | if (offset == 0) { | |
442 | fq->nhoffset = nhoff; | |
5ab11c98 | 443 | fq->q.last_in |= FIRST_IN; |
1da177e4 | 444 | } |
f61944ef | 445 | |
5ab11c98 | 446 | if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len) |
f61944ef HX |
447 | return ip6_frag_reasm(fq, prev, dev); |
448 | ||
7eb95156 PE |
449 | write_lock(&ip6_frags.lock); |
450 | list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list); | |
451 | write_unlock(&ip6_frags.lock); | |
f61944ef | 452 | return -1; |
1da177e4 LT |
453 | |
454 | err: | |
a11d206d | 455 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 | 456 | kfree_skb(skb); |
f61944ef | 457 | return -1; |
1da177e4 LT |
458 | } |
459 | ||
460 | /* | |
461 | * Check if this packet is complete. | |
462 | * Returns NULL on failure by any reason, and pointer | |
463 | * to current nexthdr field in reassembled frame. | |
464 | * | |
465 | * It is called with locked fq, and caller must check that | |
466 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
467 | * the last and the first frames arrived and all the bits are here. | |
468 | */ | |
f61944ef | 469 | static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev, |
1da177e4 LT |
470 | struct net_device *dev) |
471 | { | |
5ab11c98 | 472 | struct sk_buff *fp, *head = fq->q.fragments; |
1da177e4 LT |
473 | int payload_len; |
474 | unsigned int nhoff; | |
475 | ||
476 | fq_kill(fq); | |
477 | ||
f61944ef HX |
478 | /* Make the one we just received the head. */ |
479 | if (prev) { | |
480 | head = prev->next; | |
481 | fp = skb_clone(head, GFP_ATOMIC); | |
482 | ||
483 | if (!fp) | |
484 | goto out_oom; | |
485 | ||
486 | fp->next = head->next; | |
487 | prev->next = fp; | |
488 | ||
5ab11c98 PE |
489 | skb_morph(head, fq->q.fragments); |
490 | head->next = fq->q.fragments->next; | |
f61944ef | 491 | |
5ab11c98 PE |
492 | kfree_skb(fq->q.fragments); |
493 | fq->q.fragments = head; | |
f61944ef HX |
494 | } |
495 | ||
1da177e4 LT |
496 | BUG_TRAP(head != NULL); |
497 | BUG_TRAP(FRAG6_CB(head)->offset == 0); | |
498 | ||
499 | /* Unfragmented part is taken from the first segment. */ | |
d56f90a7 | 500 | payload_len = ((head->data - skb_network_header(head)) - |
5ab11c98 | 501 | sizeof(struct ipv6hdr) + fq->q.len - |
d56f90a7 | 502 | sizeof(struct frag_hdr)); |
1da177e4 LT |
503 | if (payload_len > IPV6_MAXPLEN) |
504 | goto out_oversize; | |
505 | ||
506 | /* Head of list must not be cloned. */ | |
507 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
508 | goto out_oom; | |
509 | ||
510 | /* If the first fragment is fragmented itself, we split | |
511 | * it to two chunks: the first with data and paged part | |
512 | * and the second, holding only fragments. */ | |
513 | if (skb_shinfo(head)->frag_list) { | |
514 | struct sk_buff *clone; | |
515 | int i, plen = 0; | |
516 | ||
517 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
518 | goto out_oom; | |
519 | clone->next = head->next; | |
520 | head->next = clone; | |
521 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
522 | skb_shinfo(head)->frag_list = NULL; | |
523 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
524 | plen += skb_shinfo(head)->frags[i].size; | |
525 | clone->len = clone->data_len = head->data_len - plen; | |
526 | head->data_len -= clone->len; | |
527 | head->len -= clone->len; | |
528 | clone->csum = 0; | |
529 | clone->ip_summed = head->ip_summed; | |
7eb95156 | 530 | atomic_add(clone->truesize, &ip6_frags.mem); |
1da177e4 LT |
531 | } |
532 | ||
533 | /* We have to remove fragment header from datagram and to relocate | |
534 | * header in order to calculate ICV correctly. */ | |
535 | nhoff = fq->nhoffset; | |
b0e380b1 | 536 | skb_network_header(head)[nhoff] = skb_transport_header(head)[0]; |
1ab1457c | 537 | memmove(head->head + sizeof(struct frag_hdr), head->head, |
1da177e4 | 538 | (head->data - head->head) - sizeof(struct frag_hdr)); |
b0e380b1 ACM |
539 | head->mac_header += sizeof(struct frag_hdr); |
540 | head->network_header += sizeof(struct frag_hdr); | |
1da177e4 LT |
541 | |
542 | skb_shinfo(head)->frag_list = head->next; | |
badff6d0 | 543 | skb_reset_transport_header(head); |
d56f90a7 | 544 | skb_push(head, head->data - skb_network_header(head)); |
7eb95156 | 545 | atomic_sub(head->truesize, &ip6_frags.mem); |
1da177e4 LT |
546 | |
547 | for (fp=head->next; fp; fp = fp->next) { | |
548 | head->data_len += fp->len; | |
549 | head->len += fp->len; | |
550 | if (head->ip_summed != fp->ip_summed) | |
551 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 552 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
553 | head->csum = csum_add(head->csum, fp->csum); |
554 | head->truesize += fp->truesize; | |
7eb95156 | 555 | atomic_sub(fp->truesize, &ip6_frags.mem); |
1da177e4 LT |
556 | } |
557 | ||
558 | head->next = NULL; | |
559 | head->dev = dev; | |
5ab11c98 | 560 | head->tstamp = fq->q.stamp; |
0660e03f | 561 | ipv6_hdr(head)->payload_len = htons(payload_len); |
951dbc8a | 562 | IP6CB(head)->nhoff = nhoff; |
1da177e4 | 563 | |
1da177e4 | 564 | /* Yes, and fold redundant checksum back. 8) */ |
84fa7933 | 565 | if (head->ip_summed == CHECKSUM_COMPLETE) |
d56f90a7 | 566 | head->csum = csum_partial(skb_network_header(head), |
cfe1fc77 | 567 | skb_network_header_len(head), |
d56f90a7 | 568 | head->csum); |
1da177e4 | 569 | |
a11d206d YH |
570 | rcu_read_lock(); |
571 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS); | |
572 | rcu_read_unlock(); | |
5ab11c98 | 573 | fq->q.fragments = NULL; |
1da177e4 LT |
574 | return 1; |
575 | ||
576 | out_oversize: | |
577 | if (net_ratelimit()) | |
578 | printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len); | |
579 | goto out_fail; | |
580 | out_oom: | |
581 | if (net_ratelimit()) | |
582 | printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n"); | |
583 | out_fail: | |
a11d206d YH |
584 | rcu_read_lock(); |
585 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
586 | rcu_read_unlock(); | |
1da177e4 LT |
587 | return -1; |
588 | } | |
589 | ||
e5bbef20 | 590 | static int ipv6_frag_rcv(struct sk_buff *skb) |
1da177e4 | 591 | { |
1da177e4 LT |
592 | struct frag_hdr *fhdr; |
593 | struct frag_queue *fq; | |
0660e03f | 594 | struct ipv6hdr *hdr = ipv6_hdr(skb); |
1da177e4 | 595 | |
a11d206d | 596 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
597 | |
598 | /* Jumbo payload inhibits frag. header */ | |
599 | if (hdr->payload_len==0) { | |
a11d206d | 600 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
cfe1fc77 ACM |
601 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
602 | skb_network_header_len(skb)); | |
1da177e4 LT |
603 | return -1; |
604 | } | |
ea2ae17d ACM |
605 | if (!pskb_may_pull(skb, (skb_transport_offset(skb) + |
606 | sizeof(struct frag_hdr)))) { | |
a11d206d | 607 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
cfe1fc77 ACM |
608 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
609 | skb_network_header_len(skb)); | |
1da177e4 LT |
610 | return -1; |
611 | } | |
612 | ||
0660e03f | 613 | hdr = ipv6_hdr(skb); |
9c70220b | 614 | fhdr = (struct frag_hdr *)skb_transport_header(skb); |
1da177e4 LT |
615 | |
616 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
617 | /* It is not a fragmented frame */ | |
b0e380b1 | 618 | skb->transport_header += sizeof(struct frag_hdr); |
a11d206d | 619 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS); |
1da177e4 | 620 | |
d56f90a7 | 621 | IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb); |
1da177e4 LT |
622 | return 1; |
623 | } | |
624 | ||
04128f23 | 625 | if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh) |
a11d206d | 626 | ip6_evictor(ip6_dst_idev(skb->dst)); |
1da177e4 | 627 | |
a11d206d YH |
628 | if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr, |
629 | ip6_dst_idev(skb->dst))) != NULL) { | |
f61944ef | 630 | int ret; |
1da177e4 | 631 | |
5ab11c98 | 632 | spin_lock(&fq->q.lock); |
1da177e4 | 633 | |
f61944ef | 634 | ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff); |
1da177e4 | 635 | |
5ab11c98 | 636 | spin_unlock(&fq->q.lock); |
4b6cb5d8 | 637 | fq_put(fq); |
1da177e4 LT |
638 | return ret; |
639 | } | |
640 | ||
a11d206d | 641 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
642 | kfree_skb(skb); |
643 | return -1; | |
644 | } | |
645 | ||
646 | static struct inet6_protocol frag_protocol = | |
647 | { | |
648 | .handler = ipv6_frag_rcv, | |
649 | .flags = INET6_PROTO_NOPOLICY, | |
650 | }; | |
651 | ||
652 | void __init ipv6_frag_init(void) | |
653 | { | |
654 | if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0) | |
655 | printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n"); | |
656 | ||
04128f23 | 657 | ip6_frags.ctl = &ip6_frags_ctl; |
321a3a99 | 658 | ip6_frags.hashfn = ip6_hashfn; |
c6fda282 | 659 | ip6_frags.constructor = ip6_frag_init; |
1e4b8287 PE |
660 | ip6_frags.destructor = ip6_frag_free; |
661 | ip6_frags.skb_free = NULL; | |
662 | ip6_frags.qsize = sizeof(struct frag_queue); | |
abd6523d | 663 | ip6_frags.match = ip6_frag_match; |
2588fe1d | 664 | ip6_frags.equal = ip6_frag_equal; |
e521db9d | 665 | ip6_frags.frag_expire = ip6_frag_expire; |
7eb95156 | 666 | inet_frags_init(&ip6_frags); |
1da177e4 | 667 | } |