]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/ipv6/reassembly.c
[NET] NETNS: Omit net_device->nd_net without CONFIG_NET_NS.
[net-next-2.6.git] / net / ipv6 / reassembly.c
CommitLineData
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
59struct 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
72struct 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
7eb95156 85static struct inet_frags ip6_frags;
1da177e4 86
e5a2bb84 87int ip6_frag_nqueues(struct net *net)
7eb95156 88{
e5a2bb84 89 return net->ipv6.frags.nqueues;
7eb95156 90}
1da177e4 91
6ddc0822 92int ip6_frag_mem(struct net *net)
7eb95156 93{
6ddc0822 94 return atomic_read(&net->ipv6.frags.mem);
7eb95156 95}
1da177e4 96
f61944ef
HX
97static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
98 struct net_device *dev);
99
f6596f9d
ZB
100/*
101 * callers should be careful not to use the hash value outside the ipfrag_lock
102 * as doing so could race with ipfrag_hash_rnd being recalculated.
103 */
e69a4adc 104static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
1da177e4
LT
105 struct in6_addr *daddr)
106{
107 u32 a, b, c;
108
e69a4adc
AV
109 a = (__force u32)saddr->s6_addr32[0];
110 b = (__force u32)saddr->s6_addr32[1];
111 c = (__force u32)saddr->s6_addr32[2];
1da177e4
LT
112
113 a += JHASH_GOLDEN_RATIO;
114 b += JHASH_GOLDEN_RATIO;
7eb95156 115 c += ip6_frags.rnd;
1da177e4
LT
116 __jhash_mix(a, b, c);
117
e69a4adc
AV
118 a += (__force u32)saddr->s6_addr32[3];
119 b += (__force u32)daddr->s6_addr32[0];
120 c += (__force u32)daddr->s6_addr32[1];
1da177e4
LT
121 __jhash_mix(a, b, c);
122
e69a4adc
AV
123 a += (__force u32)daddr->s6_addr32[2];
124 b += (__force u32)daddr->s6_addr32[3];
125 c += (__force u32)id;
1da177e4
LT
126 __jhash_mix(a, b, c);
127
7eb95156 128 return c & (INETFRAGS_HASHSZ - 1);
1da177e4
LT
129}
130
321a3a99 131static unsigned int ip6_hashfn(struct inet_frag_queue *q)
1da177e4 132{
321a3a99 133 struct frag_queue *fq;
1da177e4 134
321a3a99
PE
135 fq = container_of(q, struct frag_queue, q);
136 return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
1da177e4
LT
137}
138
abd6523d
PE
139int ip6_frag_match(struct inet_frag_queue *q, void *a)
140{
141 struct frag_queue *fq;
142 struct ip6_create_arg *arg = a;
143
144 fq = container_of(q, struct frag_queue, q);
145 return (fq->id == arg->id &&
146 ipv6_addr_equal(&fq->saddr, arg->src) &&
147 ipv6_addr_equal(&fq->daddr, arg->dst));
148}
149EXPORT_SYMBOL(ip6_frag_match);
150
1da177e4 151/* Memory Tracking Functions. */
6ddc0822
PE
152static inline void frag_kfree_skb(struct netns_frags *nf,
153 struct sk_buff *skb, int *work)
1da177e4
LT
154{
155 if (work)
156 *work -= skb->truesize;
6ddc0822 157 atomic_sub(skb->truesize, &nf->mem);
1da177e4
LT
158 kfree_skb(skb);
159}
160
c6fda282 161void ip6_frag_init(struct inet_frag_queue *q, void *a)
1da177e4 162{
c6fda282
PE
163 struct frag_queue *fq = container_of(q, struct frag_queue, q);
164 struct ip6_create_arg *arg = a;
165
166 fq->id = arg->id;
167 ipv6_addr_copy(&fq->saddr, arg->src);
168 ipv6_addr_copy(&fq->daddr, arg->dst);
1da177e4 169}
c6fda282 170EXPORT_SYMBOL(ip6_frag_init);
1da177e4 171
1da177e4
LT
172/* Destruction primitives. */
173
4b6cb5d8 174static __inline__ void fq_put(struct frag_queue *fq)
1da177e4 175{
762cc408 176 inet_frag_put(&fq->q, &ip6_frags);
1da177e4
LT
177}
178
179/* Kill fq entry. It is not destroyed immediately,
180 * because caller (and someone more) holds reference count.
181 */
182static __inline__ void fq_kill(struct frag_queue *fq)
183{
277e650d 184 inet_frag_kill(&fq->q, &ip6_frags);
1da177e4
LT
185}
186
6ddc0822 187static void ip6_evictor(struct net *net, struct inet6_dev *idev)
1da177e4 188{
8e7999c4
PE
189 int evicted;
190
6ddc0822 191 evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
8e7999c4
PE
192 if (evicted)
193 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
1da177e4
LT
194}
195
196static void ip6_frag_expire(unsigned long data)
197{
e521db9d 198 struct frag_queue *fq;
a11d206d 199 struct net_device *dev = NULL;
1da177e4 200
e521db9d
PE
201 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
202
5ab11c98 203 spin_lock(&fq->q.lock);
1da177e4 204
5ab11c98 205 if (fq->q.last_in & COMPLETE)
1da177e4
LT
206 goto out;
207
208 fq_kill(fq);
209
881d966b 210 dev = dev_get_by_index(&init_net, fq->iif);
a11d206d
YH
211 if (!dev)
212 goto out;
213
214 rcu_read_lock();
215 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
216 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
217 rcu_read_unlock();
1da177e4 218
78c784c4 219 /* Don't send error if the first segment did not arrive. */
5ab11c98 220 if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
78c784c4
IO
221 goto out;
222
78c784c4
IO
223 /*
224 But use as source device on which LAST ARRIVED
225 segment was received. And do not use fq->dev
226 pointer directly, device might already disappeared.
227 */
5ab11c98
PE
228 fq->q.fragments->dev = dev;
229 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
1da177e4 230out:
a11d206d
YH
231 if (dev)
232 dev_put(dev);
5ab11c98 233 spin_unlock(&fq->q.lock);
4b6cb5d8 234 fq_put(fq);
1da177e4
LT
235}
236
abd6523d 237static __inline__ struct frag_queue *
ac18e750 238fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
abd6523d 239 struct inet6_dev *idev)
1da177e4 240{
c6fda282
PE
241 struct inet_frag_queue *q;
242 struct ip6_create_arg arg;
abd6523d 243 unsigned int hash;
1da177e4 244
c6fda282
PE
245 arg.id = id;
246 arg.src = src;
247 arg.dst = dst;
abd6523d 248 hash = ip6qhashfn(id, src, dst);
1da177e4 249
ac18e750 250 q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
c6fda282
PE
251 if (q == NULL)
252 goto oom;
1da177e4 253
c6fda282 254 return container_of(q, struct frag_queue, q);
1da177e4
LT
255
256oom:
a11d206d 257 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
1da177e4
LT
258 return NULL;
259}
260
f61944ef 261static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
1da177e4
LT
262 struct frag_hdr *fhdr, int nhoff)
263{
264 struct sk_buff *prev, *next;
f61944ef 265 struct net_device *dev;
1da177e4
LT
266 int offset, end;
267
5ab11c98 268 if (fq->q.last_in & COMPLETE)
1da177e4
LT
269 goto err;
270
271 offset = ntohs(fhdr->frag_off) & ~0x7;
0660e03f
ACM
272 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
273 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
1da177e4
LT
274
275 if ((unsigned int)end > IPV6_MAXPLEN) {
a11d206d
YH
276 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
277 IPSTATS_MIB_INHDRERRORS);
d56f90a7
ACM
278 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
279 ((u8 *)&fhdr->frag_off -
280 skb_network_header(skb)));
f61944ef 281 return -1;
1da177e4
LT
282 }
283
d56f90a7
ACM
284 if (skb->ip_summed == CHECKSUM_COMPLETE) {
285 const unsigned char *nh = skb_network_header(skb);
1ab1457c 286 skb->csum = csum_sub(skb->csum,
d56f90a7
ACM
287 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
288 0));
289 }
1da177e4
LT
290
291 /* Is this the final fragment? */
292 if (!(fhdr->frag_off & htons(IP6_MF))) {
293 /* If we already have some bits beyond end
294 * or have different end, the segment is corrupted.
295 */
5ab11c98
PE
296 if (end < fq->q.len ||
297 ((fq->q.last_in & LAST_IN) && end != fq->q.len))
1da177e4 298 goto err;
5ab11c98
PE
299 fq->q.last_in |= LAST_IN;
300 fq->q.len = end;
1da177e4
LT
301 } else {
302 /* Check if the fragment is rounded to 8 bytes.
303 * Required by the RFC.
304 */
305 if (end & 0x7) {
306 /* RFC2460 says always send parameter problem in
307 * this case. -DaveM
308 */
a11d206d
YH
309 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
310 IPSTATS_MIB_INHDRERRORS);
1ab1457c 311 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
1da177e4 312 offsetof(struct ipv6hdr, payload_len));
f61944ef 313 return -1;
1da177e4 314 }
5ab11c98 315 if (end > fq->q.len) {
1da177e4 316 /* Some bits beyond end -> corruption. */
5ab11c98 317 if (fq->q.last_in & LAST_IN)
1da177e4 318 goto err;
5ab11c98 319 fq->q.len = end;
1da177e4
LT
320 }
321 }
322
323 if (end == offset)
324 goto err;
325
326 /* Point into the IP datagram 'data' part. */
327 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
328 goto err;
1ab1457c 329
42ca89c1
SH
330 if (pskb_trim_rcsum(skb, end - offset))
331 goto err;
1da177e4
LT
332
333 /* Find out which fragments are in front and at the back of us
334 * in the chain of fragments so far. We must know where to put
335 * this fragment, right?
336 */
337 prev = NULL;
5ab11c98 338 for(next = fq->q.fragments; next != NULL; next = next->next) {
1da177e4
LT
339 if (FRAG6_CB(next)->offset >= offset)
340 break; /* bingo! */
341 prev = next;
342 }
343
344 /* We found where to put this one. Check for overlap with
345 * preceding fragment, and, if needed, align things so that
346 * any overlaps are eliminated.
347 */
348 if (prev) {
349 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
350
351 if (i > 0) {
352 offset += i;
353 if (end <= offset)
354 goto err;
355 if (!pskb_pull(skb, i))
356 goto err;
357 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
358 skb->ip_summed = CHECKSUM_NONE;
359 }
360 }
361
362 /* Look for overlap with succeeding segments.
363 * If we can merge fragments, do it.
364 */
365 while (next && FRAG6_CB(next)->offset < end) {
366 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
367
368 if (i < next->len) {
369 /* Eat head of the next overlapped fragment
370 * and leave the loop. The next ones cannot overlap.
371 */
372 if (!pskb_pull(next, i))
373 goto err;
374 FRAG6_CB(next)->offset += i; /* next fragment */
5ab11c98 375 fq->q.meat -= i;
1da177e4
LT
376 if (next->ip_summed != CHECKSUM_UNNECESSARY)
377 next->ip_summed = CHECKSUM_NONE;
378 break;
379 } else {
380 struct sk_buff *free_it = next;
381
382 /* Old fragment is completely overridden with
383 * new one drop it.
384 */
385 next = next->next;
386
387 if (prev)
388 prev->next = next;
389 else
5ab11c98 390 fq->q.fragments = next;
1da177e4 391
5ab11c98 392 fq->q.meat -= free_it->len;
6ddc0822 393 frag_kfree_skb(fq->q.net, free_it, NULL);
1da177e4
LT
394 }
395 }
396
397 FRAG6_CB(skb)->offset = offset;
398
399 /* Insert this fragment in the chain of fragments. */
400 skb->next = next;
401 if (prev)
402 prev->next = skb;
403 else
5ab11c98 404 fq->q.fragments = skb;
1da177e4 405
f61944ef
HX
406 dev = skb->dev;
407 if (dev) {
408 fq->iif = dev->ifindex;
409 skb->dev = NULL;
410 }
5ab11c98
PE
411 fq->q.stamp = skb->tstamp;
412 fq->q.meat += skb->len;
6ddc0822 413 atomic_add(skb->truesize, &fq->q.net->mem);
1da177e4
LT
414
415 /* The first fragment.
416 * nhoffset is obtained from the first fragment, of course.
417 */
418 if (offset == 0) {
419 fq->nhoffset = nhoff;
5ab11c98 420 fq->q.last_in |= FIRST_IN;
1da177e4 421 }
f61944ef 422
5ab11c98 423 if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
f61944ef
HX
424 return ip6_frag_reasm(fq, prev, dev);
425
7eb95156 426 write_lock(&ip6_frags.lock);
3140c25c 427 list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
7eb95156 428 write_unlock(&ip6_frags.lock);
f61944ef 429 return -1;
1da177e4
LT
430
431err:
a11d206d 432 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
1da177e4 433 kfree_skb(skb);
f61944ef 434 return -1;
1da177e4
LT
435}
436
437/*
438 * Check if this packet is complete.
439 * Returns NULL on failure by any reason, and pointer
440 * to current nexthdr field in reassembled frame.
441 *
442 * It is called with locked fq, and caller must check that
443 * queue is eligible for reassembly i.e. it is not COMPLETE,
444 * the last and the first frames arrived and all the bits are here.
445 */
f61944ef 446static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
1da177e4
LT
447 struct net_device *dev)
448{
5ab11c98 449 struct sk_buff *fp, *head = fq->q.fragments;
1da177e4
LT
450 int payload_len;
451 unsigned int nhoff;
452
453 fq_kill(fq);
454
f61944ef
HX
455 /* Make the one we just received the head. */
456 if (prev) {
457 head = prev->next;
458 fp = skb_clone(head, GFP_ATOMIC);
459
460 if (!fp)
461 goto out_oom;
462
463 fp->next = head->next;
464 prev->next = fp;
465
5ab11c98
PE
466 skb_morph(head, fq->q.fragments);
467 head->next = fq->q.fragments->next;
f61944ef 468
5ab11c98
PE
469 kfree_skb(fq->q.fragments);
470 fq->q.fragments = head;
f61944ef
HX
471 }
472
1da177e4
LT
473 BUG_TRAP(head != NULL);
474 BUG_TRAP(FRAG6_CB(head)->offset == 0);
475
476 /* Unfragmented part is taken from the first segment. */
d56f90a7 477 payload_len = ((head->data - skb_network_header(head)) -
5ab11c98 478 sizeof(struct ipv6hdr) + fq->q.len -
d56f90a7 479 sizeof(struct frag_hdr));
1da177e4
LT
480 if (payload_len > IPV6_MAXPLEN)
481 goto out_oversize;
482
483 /* Head of list must not be cloned. */
484 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
485 goto out_oom;
486
487 /* If the first fragment is fragmented itself, we split
488 * it to two chunks: the first with data and paged part
489 * and the second, holding only fragments. */
490 if (skb_shinfo(head)->frag_list) {
491 struct sk_buff *clone;
492 int i, plen = 0;
493
494 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
495 goto out_oom;
496 clone->next = head->next;
497 head->next = clone;
498 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
499 skb_shinfo(head)->frag_list = NULL;
500 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
501 plen += skb_shinfo(head)->frags[i].size;
502 clone->len = clone->data_len = head->data_len - plen;
503 head->data_len -= clone->len;
504 head->len -= clone->len;
505 clone->csum = 0;
506 clone->ip_summed = head->ip_summed;
6ddc0822 507 atomic_add(clone->truesize, &fq->q.net->mem);
1da177e4
LT
508 }
509
510 /* We have to remove fragment header from datagram and to relocate
511 * header in order to calculate ICV correctly. */
512 nhoff = fq->nhoffset;
b0e380b1 513 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
1ab1457c 514 memmove(head->head + sizeof(struct frag_hdr), head->head,
1da177e4 515 (head->data - head->head) - sizeof(struct frag_hdr));
b0e380b1
ACM
516 head->mac_header += sizeof(struct frag_hdr);
517 head->network_header += sizeof(struct frag_hdr);
1da177e4
LT
518
519 skb_shinfo(head)->frag_list = head->next;
badff6d0 520 skb_reset_transport_header(head);
d56f90a7 521 skb_push(head, head->data - skb_network_header(head));
6ddc0822 522 atomic_sub(head->truesize, &fq->q.net->mem);
1da177e4
LT
523
524 for (fp=head->next; fp; fp = fp->next) {
525 head->data_len += fp->len;
526 head->len += fp->len;
527 if (head->ip_summed != fp->ip_summed)
528 head->ip_summed = CHECKSUM_NONE;
84fa7933 529 else if (head->ip_summed == CHECKSUM_COMPLETE)
1da177e4
LT
530 head->csum = csum_add(head->csum, fp->csum);
531 head->truesize += fp->truesize;
6ddc0822 532 atomic_sub(fp->truesize, &fq->q.net->mem);
1da177e4
LT
533 }
534
535 head->next = NULL;
536 head->dev = dev;
5ab11c98 537 head->tstamp = fq->q.stamp;
0660e03f 538 ipv6_hdr(head)->payload_len = htons(payload_len);
951dbc8a 539 IP6CB(head)->nhoff = nhoff;
1da177e4 540
1da177e4 541 /* Yes, and fold redundant checksum back. 8) */
84fa7933 542 if (head->ip_summed == CHECKSUM_COMPLETE)
d56f90a7 543 head->csum = csum_partial(skb_network_header(head),
cfe1fc77 544 skb_network_header_len(head),
d56f90a7 545 head->csum);
1da177e4 546
a11d206d
YH
547 rcu_read_lock();
548 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
549 rcu_read_unlock();
5ab11c98 550 fq->q.fragments = NULL;
1da177e4
LT
551 return 1;
552
553out_oversize:
554 if (net_ratelimit())
555 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
556 goto out_fail;
557out_oom:
558 if (net_ratelimit())
559 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
560out_fail:
a11d206d
YH
561 rcu_read_lock();
562 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
563 rcu_read_unlock();
1da177e4
LT
564 return -1;
565}
566
e5bbef20 567static int ipv6_frag_rcv(struct sk_buff *skb)
1da177e4 568{
1da177e4
LT
569 struct frag_hdr *fhdr;
570 struct frag_queue *fq;
0660e03f 571 struct ipv6hdr *hdr = ipv6_hdr(skb);
ac18e750 572 struct net *net;
1da177e4 573
a11d206d 574 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
1da177e4
LT
575
576 /* Jumbo payload inhibits frag. header */
577 if (hdr->payload_len==0) {
a11d206d 578 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
cfe1fc77
ACM
579 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
580 skb_network_header_len(skb));
1da177e4
LT
581 return -1;
582 }
ea2ae17d
ACM
583 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
584 sizeof(struct frag_hdr)))) {
a11d206d 585 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
cfe1fc77
ACM
586 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
587 skb_network_header_len(skb));
1da177e4
LT
588 return -1;
589 }
590
0660e03f 591 hdr = ipv6_hdr(skb);
9c70220b 592 fhdr = (struct frag_hdr *)skb_transport_header(skb);
1da177e4
LT
593
594 if (!(fhdr->frag_off & htons(0xFFF9))) {
595 /* It is not a fragmented frame */
b0e380b1 596 skb->transport_header += sizeof(struct frag_hdr);
a11d206d 597 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
1da177e4 598
d56f90a7 599 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
1da177e4
LT
600 return 1;
601 }
602
c346dca1 603 net = dev_net(skb->dev);
e31e0bdc 604 if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
6ddc0822 605 ip6_evictor(net, ip6_dst_idev(skb->dst));
1da177e4 606
ac18e750 607 if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
a11d206d 608 ip6_dst_idev(skb->dst))) != NULL) {
f61944ef 609 int ret;
1da177e4 610
5ab11c98 611 spin_lock(&fq->q.lock);
1da177e4 612
f61944ef 613 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
1da177e4 614
5ab11c98 615 spin_unlock(&fq->q.lock);
4b6cb5d8 616 fq_put(fq);
1da177e4
LT
617 return ret;
618 }
619
a11d206d 620 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
1da177e4
LT
621 kfree_skb(skb);
622 return -1;
623}
624
625static struct inet6_protocol frag_protocol =
626{
627 .handler = ipv6_frag_rcv,
628 .flags = INET6_PROTO_NOPOLICY,
629};
630
8d8354d2
PE
631#ifdef CONFIG_SYSCTL
632static struct ctl_table ip6_frags_ctl_table[] = {
633 {
634 .ctl_name = NET_IPV6_IP6FRAG_HIGH_THRESH,
635 .procname = "ip6frag_high_thresh",
e31e0bdc 636 .data = &init_net.ipv6.frags.high_thresh,
8d8354d2
PE
637 .maxlen = sizeof(int),
638 .mode = 0644,
639 .proc_handler = &proc_dointvec
640 },
641 {
642 .ctl_name = NET_IPV6_IP6FRAG_LOW_THRESH,
643 .procname = "ip6frag_low_thresh",
e31e0bdc 644 .data = &init_net.ipv6.frags.low_thresh,
8d8354d2
PE
645 .maxlen = sizeof(int),
646 .mode = 0644,
647 .proc_handler = &proc_dointvec
648 },
649 {
650 .ctl_name = NET_IPV6_IP6FRAG_TIME,
651 .procname = "ip6frag_time",
b2fd5321 652 .data = &init_net.ipv6.frags.timeout,
8d8354d2
PE
653 .maxlen = sizeof(int),
654 .mode = 0644,
655 .proc_handler = &proc_dointvec_jiffies,
656 .strategy = &sysctl_jiffies,
657 },
658 {
659 .ctl_name = NET_IPV6_IP6FRAG_SECRET_INTERVAL,
660 .procname = "ip6frag_secret_interval",
3b4bc4a2 661 .data = &ip6_frags.secret_interval,
8d8354d2
PE
662 .maxlen = sizeof(int),
663 .mode = 0644,
664 .proc_handler = &proc_dointvec_jiffies,
665 .strategy = &sysctl_jiffies
666 },
667 { }
668};
669
670static int ip6_frags_sysctl_register(struct net *net)
671{
e4a2d5c2 672 struct ctl_table *table;
8d8354d2
PE
673 struct ctl_table_header *hdr;
674
e4a2d5c2
PE
675 table = ip6_frags_ctl_table;
676 if (net != &init_net) {
677 table = kmemdup(table, sizeof(ip6_frags_ctl_table), GFP_KERNEL);
678 if (table == NULL)
679 goto err_alloc;
680
e31e0bdc
PE
681 table[0].data = &net->ipv6.frags.high_thresh;
682 table[1].data = &net->ipv6.frags.low_thresh;
b2fd5321 683 table[2].data = &net->ipv6.frags.timeout;
e4a2d5c2
PE
684 table[3].mode &= ~0222;
685 }
686
687 hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
688 if (hdr == NULL)
689 goto err_reg;
690
691 net->ipv6.sysctl.frags_hdr = hdr;
692 return 0;
693
694err_reg:
695 if (net != &init_net)
696 kfree(table);
697err_alloc:
698 return -ENOMEM;
699}
700
701static void ip6_frags_sysctl_unregister(struct net *net)
702{
703 struct ctl_table *table;
704
705 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
706 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
707 kfree(table);
8d8354d2
PE
708}
709#else
710static inline int ip6_frags_sysctl_register(struct net *net)
e71e0349 711{
8d8354d2
PE
712 return 0;
713}
e4a2d5c2
PE
714
715static inline void ip6_frags_sysctl_unregister(struct net *net)
716{
717}
8d8354d2 718#endif
7d460db9 719
8d8354d2
PE
720static int ipv6_frags_init_net(struct net *net)
721{
e31e0bdc
PE
722 net->ipv6.frags.high_thresh = 256 * 1024;
723 net->ipv6.frags.low_thresh = 192 * 1024;
b2fd5321 724 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
8d8354d2 725
e5a2bb84
PE
726 inet_frags_init_net(&net->ipv6.frags);
727
8d8354d2 728 return ip6_frags_sysctl_register(net);
e71e0349
DL
729}
730
81566e83
PE
731static void ipv6_frags_exit_net(struct net *net)
732{
733 ip6_frags_sysctl_unregister(net);
734 inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
735}
736
737static struct pernet_operations ip6_frags_ops = {
738 .init = ipv6_frags_init_net,
739 .exit = ipv6_frags_exit_net,
740};
741
853cbbaa 742int __init ipv6_frag_init(void)
1da177e4 743{
853cbbaa 744 int ret;
1da177e4 745
853cbbaa
DL
746 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
747 if (ret)
748 goto out;
e71e0349 749
81566e83 750 register_pernet_subsys(&ip6_frags_ops);
8d8354d2 751
321a3a99 752 ip6_frags.hashfn = ip6_hashfn;
c6fda282 753 ip6_frags.constructor = ip6_frag_init;
c9547709 754 ip6_frags.destructor = NULL;
1e4b8287
PE
755 ip6_frags.skb_free = NULL;
756 ip6_frags.qsize = sizeof(struct frag_queue);
abd6523d 757 ip6_frags.match = ip6_frag_match;
e521db9d 758 ip6_frags.frag_expire = ip6_frag_expire;
3b4bc4a2 759 ip6_frags.secret_interval = 10 * 60 * HZ;
7eb95156 760 inet_frags_init(&ip6_frags);
853cbbaa
DL
761out:
762 return ret;
763}
764
765void ipv6_frag_exit(void)
766{
767 inet_frags_fini(&ip6_frags);
81566e83 768 unregister_pernet_subsys(&ip6_frags_ops);
853cbbaa 769 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
1da177e4 770}