[INET]: Small cleanup for xxx_put after evictor consolidation
[linux-2.6-block.git] / net / ipv6 / netfilter / nf_conntrack_reasm.c
CommitLineData
9fb9cbb1
YK
1/*
2 * IPv6 fragment reassembly for connection tracking
3 *
4 * Copyright (C)2004 USAGI/WIDE Project
5 *
6 * Author:
7 * Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
8 *
9 * Based on: net/ipv6/reassembly.c
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
15 */
16
9fb9cbb1
YK
17#include <linux/errno.h>
18#include <linux/types.h>
19#include <linux/string.h>
20#include <linux/socket.h>
21#include <linux/sockios.h>
22#include <linux/jiffies.h>
23#include <linux/net.h>
24#include <linux/list.h>
25#include <linux/netdevice.h>
26#include <linux/in6.h>
27#include <linux/ipv6.h>
28#include <linux/icmpv6.h>
29#include <linux/random.h>
30#include <linux/jhash.h>
31
32#include <net/sock.h>
33#include <net/snmp.h>
5ab11c98 34#include <net/inet_frag.h>
9fb9cbb1
YK
35
36#include <net/ipv6.h>
37#include <net/protocol.h>
38#include <net/transp_v6.h>
39#include <net/rawv6.h>
40#include <net/ndisc.h>
41#include <net/addrconf.h>
42#include <linux/sysctl.h>
43#include <linux/netfilter.h>
44#include <linux/netfilter_ipv6.h>
45#include <linux/kernel.h>
46#include <linux/module.h>
47
9fb9cbb1
YK
48#define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */
49#define NF_CT_FRAG6_LOW_THRESH 196608 /* == 192*1024 */
50#define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT
51
9fb9cbb1
YK
52struct nf_ct_frag6_skb_cb
53{
54 struct inet6_skb_parm h;
55 int offset;
56 struct sk_buff *orig;
57};
58
59#define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
60
61struct nf_ct_frag6_queue
62{
5ab11c98 63 struct inet_frag_queue q;
9fb9cbb1 64
bff9a89b 65 __be32 id; /* fragment id */
9fb9cbb1
YK
66 struct in6_addr saddr;
67 struct in6_addr daddr;
68
9fb9cbb1 69 unsigned int csum;
9fb9cbb1 70 __u16 nhoffset;
9fb9cbb1
YK
71};
72
04128f23
PE
73struct inet_frags_ctl nf_frags_ctl __read_mostly = {
74 .high_thresh = 256 * 1024,
75 .low_thresh = 192 * 1024,
76 .timeout = IPV6_FRAG_TIMEOUT,
77 .secret_interval = 10 * 60 * HZ,
78};
79
7eb95156 80static struct inet_frags nf_frags;
9fb9cbb1 81
bff9a89b 82static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
9fb9cbb1
YK
83 struct in6_addr *daddr)
84{
85 u32 a, b, c;
86
bff9a89b
PM
87 a = (__force u32)saddr->s6_addr32[0];
88 b = (__force u32)saddr->s6_addr32[1];
89 c = (__force u32)saddr->s6_addr32[2];
9fb9cbb1
YK
90
91 a += JHASH_GOLDEN_RATIO;
92 b += JHASH_GOLDEN_RATIO;
7eb95156 93 c += nf_frags.rnd;
9fb9cbb1
YK
94 __jhash_mix(a, b, c);
95
bff9a89b
PM
96 a += (__force u32)saddr->s6_addr32[3];
97 b += (__force u32)daddr->s6_addr32[0];
98 c += (__force u32)daddr->s6_addr32[1];
9fb9cbb1
YK
99 __jhash_mix(a, b, c);
100
bff9a89b
PM
101 a += (__force u32)daddr->s6_addr32[2];
102 b += (__force u32)daddr->s6_addr32[3];
103 c += (__force u32)id;
9fb9cbb1
YK
104 __jhash_mix(a, b, c);
105
7eb95156 106 return c & (INETFRAGS_HASHSZ - 1);
9fb9cbb1
YK
107}
108
321a3a99 109static unsigned int nf_hashfn(struct inet_frag_queue *q)
9fb9cbb1 110{
321a3a99 111 struct nf_ct_frag6_queue *nq;
9fb9cbb1 112
321a3a99
PE
113 nq = container_of(q, struct nf_ct_frag6_queue, q);
114 return ip6qhashfn(nq->id, &nq->saddr, &nq->daddr);
9fb9cbb1
YK
115}
116
1e4b8287
PE
117static void nf_skb_free(struct sk_buff *skb)
118{
119 if (NFCT_FRAG6_CB(skb)->orig)
120 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
121}
122
9fb9cbb1 123/* Memory Tracking Functions. */
1ba430bc 124static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
9fb9cbb1 125{
1ba430bc
YK
126 if (work)
127 *work -= skb->truesize;
7eb95156 128 atomic_sub(skb->truesize, &nf_frags.mem);
1e4b8287 129 nf_skb_free(skb);
9fb9cbb1
YK
130 kfree_skb(skb);
131}
132
1e4b8287 133static void nf_frag_free(struct inet_frag_queue *q)
9fb9cbb1 134{
1e4b8287 135 kfree(container_of(q, struct nf_ct_frag6_queue, q));
9fb9cbb1
YK
136}
137
138static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
139{
140 struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
141
142 if (!fq)
143 return NULL;
7eb95156 144 atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_frags.mem);
9fb9cbb1
YK
145 return fq;
146}
147
148/* Destruction primitives. */
149
4b6cb5d8 150static __inline__ void fq_put(struct nf_ct_frag6_queue *fq)
9fb9cbb1 151{
5ab11c98 152 if (atomic_dec_and_test(&fq->q.refcnt))
4b6cb5d8 153 inet_frag_destroy(&fq->q, &nf_frags, NULL);
9fb9cbb1
YK
154}
155
156/* Kill fq entry. It is not destroyed immediately,
157 * because caller (and someone more) holds reference count.
158 */
159static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
160{
277e650d 161 inet_frag_kill(&fq->q, &nf_frags);
9fb9cbb1
YK
162}
163
164static void nf_ct_frag6_evictor(void)
165{
8e7999c4 166 inet_frag_evictor(&nf_frags);
9fb9cbb1
YK
167}
168
169static void nf_ct_frag6_expire(unsigned long data)
170{
171 struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
172
5ab11c98 173 spin_lock(&fq->q.lock);
9fb9cbb1 174
5ab11c98 175 if (fq->q.last_in & COMPLETE)
9fb9cbb1
YK
176 goto out;
177
178 fq_kill(fq);
179
180out:
5ab11c98 181 spin_unlock(&fq->q.lock);
4b6cb5d8 182 fq_put(fq);
9fb9cbb1
YK
183}
184
185/* Creation primitives. */
186
9fb9cbb1
YK
187static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
188 struct nf_ct_frag6_queue *fq_in)
189{
190 struct nf_ct_frag6_queue *fq;
2e4e6a17
HW
191#ifdef CONFIG_SMP
192 struct hlist_node *n;
193#endif
9fb9cbb1 194
7eb95156 195 write_lock(&nf_frags.lock);
9fb9cbb1 196#ifdef CONFIG_SMP
7eb95156 197 hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
1ab1457c 198 if (fq->id == fq_in->id &&
6ea46c9c
YK
199 ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
200 ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
5ab11c98 201 atomic_inc(&fq->q.refcnt);
7eb95156 202 write_unlock(&nf_frags.lock);
5ab11c98 203 fq_in->q.last_in |= COMPLETE;
4b6cb5d8 204 fq_put(fq_in);
9fb9cbb1
YK
205 return fq;
206 }
207 }
208#endif
209 fq = fq_in;
210
04128f23 211 if (!mod_timer(&fq->q.timer, jiffies + nf_frags_ctl.timeout))
5ab11c98 212 atomic_inc(&fq->q.refcnt);
9fb9cbb1 213
5ab11c98 214 atomic_inc(&fq->q.refcnt);
7eb95156 215 hlist_add_head(&fq->q.list, &nf_frags.hash[hash]);
5ab11c98 216 INIT_LIST_HEAD(&fq->q.lru_list);
7eb95156
PE
217 list_add_tail(&fq->q.lru_list, &nf_frags.lru_list);
218 nf_frags.nqueues++;
219 write_unlock(&nf_frags.lock);
9fb9cbb1
YK
220 return fq;
221}
222
223
224static struct nf_ct_frag6_queue *
bff9a89b 225nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src, struct in6_addr *dst)
9fb9cbb1
YK
226{
227 struct nf_ct_frag6_queue *fq;
228
229 if ((fq = frag_alloc_queue()) == NULL) {
0d53778e 230 pr_debug("Can't alloc new queue\n");
9fb9cbb1
YK
231 goto oom;
232 }
233
234 memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
235
236 fq->id = id;
237 ipv6_addr_copy(&fq->saddr, src);
238 ipv6_addr_copy(&fq->daddr, dst);
239
5ab11c98
PE
240 setup_timer(&fq->q.timer, nf_ct_frag6_expire, (unsigned long)fq);
241 spin_lock_init(&fq->q.lock);
242 atomic_set(&fq->q.refcnt, 1);
9fb9cbb1
YK
243
244 return nf_ct_frag6_intern(hash, fq);
245
246oom:
247 return NULL;
248}
249
250static __inline__ struct nf_ct_frag6_queue *
bff9a89b 251fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
9fb9cbb1
YK
252{
253 struct nf_ct_frag6_queue *fq;
2e4e6a17 254 struct hlist_node *n;
9fb9cbb1
YK
255 unsigned int hash = ip6qhashfn(id, src, dst);
256
7eb95156
PE
257 read_lock(&nf_frags.lock);
258 hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) {
1ab1457c 259 if (fq->id == id &&
6ea46c9c
YK
260 ipv6_addr_equal(src, &fq->saddr) &&
261 ipv6_addr_equal(dst, &fq->daddr)) {
5ab11c98 262 atomic_inc(&fq->q.refcnt);
7eb95156 263 read_unlock(&nf_frags.lock);
9fb9cbb1
YK
264 return fq;
265 }
266 }
7eb95156 267 read_unlock(&nf_frags.lock);
9fb9cbb1
YK
268
269 return nf_ct_frag6_create(hash, id, src, dst);
270}
271
272
1ab1457c 273static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
9fb9cbb1
YK
274 struct frag_hdr *fhdr, int nhoff)
275{
276 struct sk_buff *prev, *next;
277 int offset, end;
278
5ab11c98 279 if (fq->q.last_in & COMPLETE) {
0d53778e 280 pr_debug("Allready completed\n");
9fb9cbb1
YK
281 goto err;
282 }
283
284 offset = ntohs(fhdr->frag_off) & ~0x7;
0660e03f
ACM
285 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
286 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
9fb9cbb1
YK
287
288 if ((unsigned int)end > IPV6_MAXPLEN) {
0d53778e 289 pr_debug("offset is too large.\n");
1ab1457c 290 return -1;
9fb9cbb1
YK
291 }
292
d56f90a7
ACM
293 if (skb->ip_summed == CHECKSUM_COMPLETE) {
294 const unsigned char *nh = skb_network_header(skb);
1ab1457c 295 skb->csum = csum_sub(skb->csum,
d56f90a7 296 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
9fb9cbb1 297 0));
d56f90a7 298 }
9fb9cbb1
YK
299
300 /* Is this the final fragment? */
301 if (!(fhdr->frag_off & htons(IP6_MF))) {
302 /* If we already have some bits beyond end
303 * or have different end, the segment is corrupted.
304 */
5ab11c98
PE
305 if (end < fq->q.len ||
306 ((fq->q.last_in & LAST_IN) && end != fq->q.len)) {
0d53778e 307 pr_debug("already received last fragment\n");
9fb9cbb1
YK
308 goto err;
309 }
5ab11c98
PE
310 fq->q.last_in |= LAST_IN;
311 fq->q.len = end;
9fb9cbb1
YK
312 } else {
313 /* Check if the fragment is rounded to 8 bytes.
314 * Required by the RFC.
315 */
316 if (end & 0x7) {
317 /* RFC2460 says always send parameter problem in
318 * this case. -DaveM
319 */
0d53778e 320 pr_debug("end of fragment not rounded to 8 bytes.\n");
9fb9cbb1
YK
321 return -1;
322 }
5ab11c98 323 if (end > fq->q.len) {
9fb9cbb1 324 /* Some bits beyond end -> corruption. */
5ab11c98 325 if (fq->q.last_in & LAST_IN) {
0d53778e 326 pr_debug("last packet already reached.\n");
9fb9cbb1
YK
327 goto err;
328 }
5ab11c98 329 fq->q.len = end;
9fb9cbb1
YK
330 }
331 }
332
333 if (end == offset)
334 goto err;
335
336 /* Point into the IP datagram 'data' part. */
337 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
0d53778e 338 pr_debug("queue: message is too short.\n");
9fb9cbb1
YK
339 goto err;
340 }
b38dfee3 341 if (pskb_trim_rcsum(skb, end - offset)) {
0d53778e 342 pr_debug("Can't trim\n");
b38dfee3 343 goto err;
9fb9cbb1
YK
344 }
345
346 /* Find out which fragments are in front and at the back of us
347 * in the chain of fragments so far. We must know where to put
348 * this fragment, right?
349 */
350 prev = NULL;
5ab11c98 351 for (next = fq->q.fragments; next != NULL; next = next->next) {
9fb9cbb1
YK
352 if (NFCT_FRAG6_CB(next)->offset >= offset)
353 break; /* bingo! */
354 prev = next;
355 }
356
357 /* We found where to put this one. Check for overlap with
358 * preceding fragment, and, if needed, align things so that
359 * any overlaps are eliminated.
360 */
361 if (prev) {
362 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
363
364 if (i > 0) {
365 offset += i;
366 if (end <= offset) {
0d53778e 367 pr_debug("overlap\n");
9fb9cbb1
YK
368 goto err;
369 }
370 if (!pskb_pull(skb, i)) {
0d53778e 371 pr_debug("Can't pull\n");
9fb9cbb1
YK
372 goto err;
373 }
374 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
375 skb->ip_summed = CHECKSUM_NONE;
376 }
377 }
378
379 /* Look for overlap with succeeding segments.
380 * If we can merge fragments, do it.
381 */
382 while (next && NFCT_FRAG6_CB(next)->offset < end) {
383 /* overlap is 'i' bytes */
384 int i = end - NFCT_FRAG6_CB(next)->offset;
385
386 if (i < next->len) {
387 /* Eat head of the next overlapped fragment
388 * and leave the loop. The next ones cannot overlap.
389 */
0d53778e 390 pr_debug("Eat head of the overlapped parts.: %d", i);
9fb9cbb1
YK
391 if (!pskb_pull(next, i))
392 goto err;
393
394 /* next fragment */
395 NFCT_FRAG6_CB(next)->offset += i;
5ab11c98 396 fq->q.meat -= i;
9fb9cbb1
YK
397 if (next->ip_summed != CHECKSUM_UNNECESSARY)
398 next->ip_summed = CHECKSUM_NONE;
399 break;
400 } else {
401 struct sk_buff *free_it = next;
402
403 /* Old fragmnet is completely overridden with
404 * new one drop it.
405 */
406 next = next->next;
407
408 if (prev)
409 prev->next = next;
410 else
5ab11c98 411 fq->q.fragments = next;
9fb9cbb1 412
5ab11c98 413 fq->q.meat -= free_it->len;
1ba430bc 414 frag_kfree_skb(free_it, NULL);
9fb9cbb1
YK
415 }
416 }
417
418 NFCT_FRAG6_CB(skb)->offset = offset;
419
420 /* Insert this fragment in the chain of fragments. */
421 skb->next = next;
422 if (prev)
423 prev->next = skb;
424 else
5ab11c98 425 fq->q.fragments = skb;
9fb9cbb1
YK
426
427 skb->dev = NULL;
5ab11c98
PE
428 fq->q.stamp = skb->tstamp;
429 fq->q.meat += skb->len;
7eb95156 430 atomic_add(skb->truesize, &nf_frags.mem);
9fb9cbb1
YK
431
432 /* The first fragment.
433 * nhoffset is obtained from the first fragment, of course.
434 */
435 if (offset == 0) {
436 fq->nhoffset = nhoff;
5ab11c98 437 fq->q.last_in |= FIRST_IN;
9fb9cbb1 438 }
7eb95156
PE
439 write_lock(&nf_frags.lock);
440 list_move_tail(&fq->q.lru_list, &nf_frags.lru_list);
441 write_unlock(&nf_frags.lock);
9fb9cbb1
YK
442 return 0;
443
444err:
445 return -1;
446}
447
448/*
449 * Check if this packet is complete.
450 * Returns NULL on failure by any reason, and pointer
451 * to current nexthdr field in reassembled frame.
452 *
453 * It is called with locked fq, and caller must check that
454 * queue is eligible for reassembly i.e. it is not COMPLETE,
455 * the last and the first frames arrived and all the bits are here.
456 */
457static struct sk_buff *
458nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
459{
5ab11c98 460 struct sk_buff *fp, *op, *head = fq->q.fragments;
9fb9cbb1
YK
461 int payload_len;
462
463 fq_kill(fq);
464
465 BUG_TRAP(head != NULL);
466 BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
467
468 /* Unfragmented part is taken from the first segment. */
d56f90a7 469 payload_len = ((head->data - skb_network_header(head)) -
5ab11c98 470 sizeof(struct ipv6hdr) + fq->q.len -
d56f90a7 471 sizeof(struct frag_hdr));
9fb9cbb1 472 if (payload_len > IPV6_MAXPLEN) {
0d53778e 473 pr_debug("payload len is too large.\n");
9fb9cbb1
YK
474 goto out_oversize;
475 }
476
477 /* Head of list must not be cloned. */
478 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
0d53778e 479 pr_debug("skb is cloned but can't expand head");
9fb9cbb1
YK
480 goto out_oom;
481 }
482
483 /* If the first fragment is fragmented itself, we split
484 * it to two chunks: the first with data and paged part
485 * and the second, holding only fragments. */
486 if (skb_shinfo(head)->frag_list) {
487 struct sk_buff *clone;
488 int i, plen = 0;
489
490 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
0d53778e 491 pr_debug("Can't alloc skb\n");
9fb9cbb1
YK
492 goto out_oom;
493 }
494 clone->next = head->next;
495 head->next = clone;
496 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
497 skb_shinfo(head)->frag_list = NULL;
498 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
499 plen += skb_shinfo(head)->frags[i].size;
500 clone->len = clone->data_len = head->data_len - plen;
501 head->data_len -= clone->len;
502 head->len -= clone->len;
503 clone->csum = 0;
504 clone->ip_summed = head->ip_summed;
505
506 NFCT_FRAG6_CB(clone)->orig = NULL;
7eb95156 507 atomic_add(clone->truesize, &nf_frags.mem);
9fb9cbb1
YK
508 }
509
510 /* We have to remove fragment header from datagram and to relocate
511 * header in order to calculate ICV correctly. */
bff9b61c 512 skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0];
1ab1457c 513 memmove(head->head + sizeof(struct frag_hdr), head->head,
9fb9cbb1 514 (head->data - head->head) - sizeof(struct frag_hdr));
b0e380b1
ACM
515 head->mac_header += sizeof(struct frag_hdr);
516 head->network_header += sizeof(struct frag_hdr);
9fb9cbb1
YK
517
518 skb_shinfo(head)->frag_list = head->next;
badff6d0 519 skb_reset_transport_header(head);
d56f90a7 520 skb_push(head, head->data - skb_network_header(head));
7eb95156 521 atomic_sub(head->truesize, &nf_frags.mem);
9fb9cbb1
YK
522
523 for (fp=head->next; fp; fp = fp->next) {
524 head->data_len += fp->len;
525 head->len += fp->len;
526 if (head->ip_summed != fp->ip_summed)
527 head->ip_summed = CHECKSUM_NONE;
84fa7933 528 else if (head->ip_summed == CHECKSUM_COMPLETE)
9fb9cbb1
YK
529 head->csum = csum_add(head->csum, fp->csum);
530 head->truesize += fp->truesize;
7eb95156 531 atomic_sub(fp->truesize, &nf_frags.mem);
9fb9cbb1
YK
532 }
533
534 head->next = NULL;
535 head->dev = dev;
5ab11c98 536 head->tstamp = fq->q.stamp;
0660e03f 537 ipv6_hdr(head)->payload_len = htons(payload_len);
9fb9cbb1
YK
538
539 /* Yes, and fold redundant checksum back. 8) */
84fa7933 540 if (head->ip_summed == CHECKSUM_COMPLETE)
d56f90a7 541 head->csum = csum_partial(skb_network_header(head),
cfe1fc77 542 skb_network_header_len(head),
d56f90a7 543 head->csum);
9fb9cbb1 544
5ab11c98 545 fq->q.fragments = NULL;
9fb9cbb1
YK
546
547 /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
548 fp = skb_shinfo(head)->frag_list;
549 if (NFCT_FRAG6_CB(fp)->orig == NULL)
550 /* at above code, head skb is divided into two skbs. */
551 fp = fp->next;
552
553 op = NFCT_FRAG6_CB(head)->orig;
554 for (; fp; fp = fp->next) {
555 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
556
557 op->next = orig;
558 op = orig;
559 NFCT_FRAG6_CB(fp)->orig = NULL;
560 }
561
562 return head;
563
564out_oversize:
565 if (net_ratelimit())
566 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
567 goto out_fail;
568out_oom:
569 if (net_ratelimit())
570 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
571out_fail:
572 return NULL;
573}
574
575/*
576 * find the header just before Fragment Header.
577 *
578 * if success return 0 and set ...
579 * (*prevhdrp): the value of "Next Header Field" in the header
580 * just before Fragment Header.
581 * (*prevhoff): the offset of "Next Header Field" in the header
582 * just before Fragment Header.
583 * (*fhoff) : the offset of Fragment Header.
584 *
585 * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
586 *
587 */
588static int
589find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
590{
0660e03f 591 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
6b88dd96
ACM
592 const int netoff = skb_network_offset(skb);
593 u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr);
594 int start = netoff + sizeof(struct ipv6hdr);
9fb9cbb1
YK
595 int len = skb->len - start;
596 u8 prevhdr = NEXTHDR_IPV6;
597
1ab1457c
YH
598 while (nexthdr != NEXTHDR_FRAGMENT) {
599 struct ipv6_opt_hdr hdr;
600 int hdrlen;
9fb9cbb1
YK
601
602 if (!ipv6_ext_hdr(nexthdr)) {
603 return -1;
604 }
1ab1457c 605 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
0d53778e 606 pr_debug("too short\n");
9fb9cbb1
YK
607 return -1;
608 }
1ab1457c 609 if (nexthdr == NEXTHDR_NONE) {
0d53778e 610 pr_debug("next header is none\n");
9fb9cbb1
YK
611 return -1;
612 }
1ab1457c
YH
613 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
614 BUG();
615 if (nexthdr == NEXTHDR_AUTH)
616 hdrlen = (hdr.hdrlen+2)<<2;
617 else
618 hdrlen = ipv6_optlen(&hdr);
9fb9cbb1
YK
619
620 prevhdr = nexthdr;
621 prev_nhoff = start;
622
1ab1457c
YH
623 nexthdr = hdr.nexthdr;
624 len -= hdrlen;
625 start += hdrlen;
626 }
9fb9cbb1
YK
627
628 if (len < 0)
629 return -1;
630
631 *prevhdrp = prevhdr;
632 *prevhoff = prev_nhoff;
633 *fhoff = start;
634
635 return 0;
636}
637
638struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
639{
1ab1457c 640 struct sk_buff *clone;
9fb9cbb1
YK
641 struct net_device *dev = skb->dev;
642 struct frag_hdr *fhdr;
643 struct nf_ct_frag6_queue *fq;
644 struct ipv6hdr *hdr;
645 int fhoff, nhoff;
646 u8 prevhdr;
647 struct sk_buff *ret_skb = NULL;
648
649 /* Jumbo payload inhibits frag. header */
0660e03f 650 if (ipv6_hdr(skb)->payload_len == 0) {
0d53778e 651 pr_debug("payload len = 0\n");
9fb9cbb1
YK
652 return skb;
653 }
654
655 if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
656 return skb;
657
658 clone = skb_clone(skb, GFP_ATOMIC);
659 if (clone == NULL) {
0d53778e 660 pr_debug("Can't clone skb\n");
9fb9cbb1
YK
661 return skb;
662 }
663
664 NFCT_FRAG6_CB(clone)->orig = skb;
665
666 if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
0d53778e 667 pr_debug("message is too short.\n");
9fb9cbb1
YK
668 goto ret_orig;
669 }
670
967b05f6 671 skb_set_transport_header(clone, fhoff);
0660e03f 672 hdr = ipv6_hdr(clone);
bff9b61c 673 fhdr = (struct frag_hdr *)skb_transport_header(clone);
9fb9cbb1
YK
674
675 if (!(fhdr->frag_off & htons(0xFFF9))) {
0d53778e 676 pr_debug("Invalid fragment offset\n");
9fb9cbb1
YK
677 /* It is not a fragmented frame */
678 goto ret_orig;
679 }
680
04128f23 681 if (atomic_read(&nf_frags.mem) > nf_frags_ctl.high_thresh)
9fb9cbb1
YK
682 nf_ct_frag6_evictor();
683
684 fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
685 if (fq == NULL) {
0d53778e 686 pr_debug("Can't find and can't create new queue\n");
9fb9cbb1
YK
687 goto ret_orig;
688 }
689
5ab11c98 690 spin_lock(&fq->q.lock);
9fb9cbb1
YK
691
692 if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
5ab11c98 693 spin_unlock(&fq->q.lock);
0d53778e 694 pr_debug("Can't insert skb to queue\n");
4b6cb5d8 695 fq_put(fq);
9fb9cbb1
YK
696 goto ret_orig;
697 }
698
5ab11c98 699 if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) {
9fb9cbb1
YK
700 ret_skb = nf_ct_frag6_reasm(fq, dev);
701 if (ret_skb == NULL)
0d53778e 702 pr_debug("Can't reassemble fragmented packets\n");
9fb9cbb1 703 }
5ab11c98 704 spin_unlock(&fq->q.lock);
9fb9cbb1 705
4b6cb5d8 706 fq_put(fq);
9fb9cbb1
YK
707 return ret_skb;
708
709ret_orig:
710 kfree_skb(clone);
711 return skb;
712}
713
714void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
715 struct net_device *in, struct net_device *out,
716 int (*okfn)(struct sk_buff *))
717{
718 struct sk_buff *s, *s2;
719
720 for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
721 nf_conntrack_put_reasm(s->nfct_reasm);
722 nf_conntrack_get_reasm(skb);
723 s->nfct_reasm = skb;
724
725 s2 = s->next;
f9f02cca
PM
726 s->next = NULL;
727
9fb9cbb1
YK
728 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
729 NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
730 s = s2;
731 }
732 nf_conntrack_put_reasm(skb);
733}
734
735int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
736{
737 struct sk_buff *s, *s2;
738
739 for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
740
741 s2 = s->next;
742 kfree_skb(s);
743 }
744
745 kfree_skb(skb);
746
747 return 0;
748}
749
750int nf_ct_frag6_init(void)
751{
04128f23 752 nf_frags.ctl = &nf_frags_ctl;
321a3a99 753 nf_frags.hashfn = nf_hashfn;
1e4b8287
PE
754 nf_frags.destructor = nf_frag_free;
755 nf_frags.skb_free = nf_skb_free;
756 nf_frags.qsize = sizeof(struct nf_ct_frag6_queue);
7eb95156 757 inet_frags_init(&nf_frags);
9fb9cbb1
YK
758
759 return 0;
760}
761
762void nf_ct_frag6_cleanup(void)
763{
7eb95156
PE
764 inet_frags_fini(&nf_frags);
765
04128f23 766 nf_frags_ctl.low_thresh = 0;
9fb9cbb1
YK
767 nf_ct_frag6_evictor();
768}