]>
Commit | Line | Data |
---|---|---|
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 |
52 | struct 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 | ||
61 | struct 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 | ||
8d8354d2 | 73 | static struct inet_frags_ctl nf_frags_ctl __read_mostly = { |
04128f23 PE |
74 | .high_thresh = 256 * 1024, |
75 | .low_thresh = 192 * 1024, | |
76 | .timeout = IPV6_FRAG_TIMEOUT, | |
77 | .secret_interval = 10 * 60 * HZ, | |
78 | }; | |
79 | ||
7eb95156 | 80 | static struct inet_frags nf_frags; |
ac18e750 | 81 | static struct netns_frags nf_init_frags; |
9fb9cbb1 | 82 | |
8d8354d2 PE |
83 | #ifdef CONFIG_SYSCTL |
84 | struct ctl_table nf_ct_ipv6_sysctl_table[] = { | |
85 | { | |
86 | .procname = "nf_conntrack_frag6_timeout", | |
87 | .data = &nf_frags_ctl.timeout, | |
88 | .maxlen = sizeof(unsigned int), | |
89 | .mode = 0644, | |
90 | .proc_handler = &proc_dointvec_jiffies, | |
91 | }, | |
92 | { | |
93 | .ctl_name = NET_NF_CONNTRACK_FRAG6_LOW_THRESH, | |
94 | .procname = "nf_conntrack_frag6_low_thresh", | |
95 | .data = &nf_frags_ctl.low_thresh, | |
96 | .maxlen = sizeof(unsigned int), | |
97 | .mode = 0644, | |
98 | .proc_handler = &proc_dointvec, | |
99 | }, | |
100 | { | |
101 | .ctl_name = NET_NF_CONNTRACK_FRAG6_HIGH_THRESH, | |
102 | .procname = "nf_conntrack_frag6_high_thresh", | |
103 | .data = &nf_frags_ctl.high_thresh, | |
104 | .maxlen = sizeof(unsigned int), | |
105 | .mode = 0644, | |
106 | .proc_handler = &proc_dointvec, | |
107 | }, | |
108 | { .ctl_name = 0 } | |
109 | }; | |
110 | #endif | |
111 | ||
bff9a89b | 112 | static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr, |
9fb9cbb1 YK |
113 | struct in6_addr *daddr) |
114 | { | |
115 | u32 a, b, c; | |
116 | ||
bff9a89b PM |
117 | a = (__force u32)saddr->s6_addr32[0]; |
118 | b = (__force u32)saddr->s6_addr32[1]; | |
119 | c = (__force u32)saddr->s6_addr32[2]; | |
9fb9cbb1 YK |
120 | |
121 | a += JHASH_GOLDEN_RATIO; | |
122 | b += JHASH_GOLDEN_RATIO; | |
7eb95156 | 123 | c += nf_frags.rnd; |
9fb9cbb1 YK |
124 | __jhash_mix(a, b, c); |
125 | ||
bff9a89b PM |
126 | a += (__force u32)saddr->s6_addr32[3]; |
127 | b += (__force u32)daddr->s6_addr32[0]; | |
128 | c += (__force u32)daddr->s6_addr32[1]; | |
9fb9cbb1 YK |
129 | __jhash_mix(a, b, c); |
130 | ||
bff9a89b PM |
131 | a += (__force u32)daddr->s6_addr32[2]; |
132 | b += (__force u32)daddr->s6_addr32[3]; | |
133 | c += (__force u32)id; | |
9fb9cbb1 YK |
134 | __jhash_mix(a, b, c); |
135 | ||
7eb95156 | 136 | return c & (INETFRAGS_HASHSZ - 1); |
9fb9cbb1 YK |
137 | } |
138 | ||
321a3a99 | 139 | static unsigned int nf_hashfn(struct inet_frag_queue *q) |
9fb9cbb1 | 140 | { |
321a3a99 | 141 | struct nf_ct_frag6_queue *nq; |
9fb9cbb1 | 142 | |
321a3a99 PE |
143 | nq = container_of(q, struct nf_ct_frag6_queue, q); |
144 | return ip6qhashfn(nq->id, &nq->saddr, &nq->daddr); | |
9fb9cbb1 YK |
145 | } |
146 | ||
1e4b8287 PE |
147 | static void nf_skb_free(struct sk_buff *skb) |
148 | { | |
149 | if (NFCT_FRAG6_CB(skb)->orig) | |
150 | kfree_skb(NFCT_FRAG6_CB(skb)->orig); | |
151 | } | |
152 | ||
9fb9cbb1 | 153 | /* Memory Tracking Functions. */ |
1ba430bc | 154 | static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work) |
9fb9cbb1 | 155 | { |
1ba430bc YK |
156 | if (work) |
157 | *work -= skb->truesize; | |
6ddc0822 | 158 | atomic_sub(skb->truesize, &nf_init_frags.mem); |
1e4b8287 | 159 | nf_skb_free(skb); |
9fb9cbb1 YK |
160 | kfree_skb(skb); |
161 | } | |
162 | ||
9fb9cbb1 YK |
163 | /* Destruction primitives. */ |
164 | ||
4b6cb5d8 | 165 | static __inline__ void fq_put(struct nf_ct_frag6_queue *fq) |
9fb9cbb1 | 166 | { |
762cc408 | 167 | inet_frag_put(&fq->q, &nf_frags); |
9fb9cbb1 YK |
168 | } |
169 | ||
170 | /* Kill fq entry. It is not destroyed immediately, | |
171 | * because caller (and someone more) holds reference count. | |
172 | */ | |
173 | static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq) | |
174 | { | |
277e650d | 175 | inet_frag_kill(&fq->q, &nf_frags); |
9fb9cbb1 YK |
176 | } |
177 | ||
178 | static void nf_ct_frag6_evictor(void) | |
179 | { | |
6ddc0822 | 180 | inet_frag_evictor(&nf_init_frags, &nf_frags); |
9fb9cbb1 YK |
181 | } |
182 | ||
183 | static void nf_ct_frag6_expire(unsigned long data) | |
184 | { | |
e521db9d PE |
185 | struct nf_ct_frag6_queue *fq; |
186 | ||
187 | fq = container_of((struct inet_frag_queue *)data, | |
188 | struct nf_ct_frag6_queue, q); | |
9fb9cbb1 | 189 | |
5ab11c98 | 190 | spin_lock(&fq->q.lock); |
9fb9cbb1 | 191 | |
5ab11c98 | 192 | if (fq->q.last_in & COMPLETE) |
9fb9cbb1 YK |
193 | goto out; |
194 | ||
195 | fq_kill(fq); | |
196 | ||
197 | out: | |
5ab11c98 | 198 | spin_unlock(&fq->q.lock); |
4b6cb5d8 | 199 | fq_put(fq); |
9fb9cbb1 YK |
200 | } |
201 | ||
202 | /* Creation primitives. */ | |
203 | ||
abd6523d PE |
204 | static __inline__ struct nf_ct_frag6_queue * |
205 | fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst) | |
9fb9cbb1 | 206 | { |
2588fe1d | 207 | struct inet_frag_queue *q; |
c6fda282 | 208 | struct ip6_create_arg arg; |
abd6523d | 209 | unsigned int hash; |
9fb9cbb1 | 210 | |
c6fda282 PE |
211 | arg.id = id; |
212 | arg.src = src; | |
213 | arg.dst = dst; | |
abd6523d | 214 | hash = ip6qhashfn(id, src, dst); |
9fb9cbb1 | 215 | |
ac18e750 | 216 | q = inet_frag_find(&nf_init_frags, &nf_frags, &arg, hash); |
c6fda282 | 217 | if (q == NULL) |
9fb9cbb1 | 218 | goto oom; |
9fb9cbb1 | 219 | |
c6fda282 | 220 | return container_of(q, struct nf_ct_frag6_queue, q); |
9fb9cbb1 YK |
221 | |
222 | oom: | |
c6fda282 | 223 | pr_debug("Can't alloc new queue\n"); |
9fb9cbb1 YK |
224 | return NULL; |
225 | } | |
226 | ||
9fb9cbb1 | 227 | |
1ab1457c | 228 | static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb, |
9fb9cbb1 YK |
229 | struct frag_hdr *fhdr, int nhoff) |
230 | { | |
231 | struct sk_buff *prev, *next; | |
232 | int offset, end; | |
233 | ||
5ab11c98 | 234 | if (fq->q.last_in & COMPLETE) { |
0d53778e | 235 | pr_debug("Allready completed\n"); |
9fb9cbb1 YK |
236 | goto err; |
237 | } | |
238 | ||
239 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
0660e03f ACM |
240 | end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - |
241 | ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); | |
9fb9cbb1 YK |
242 | |
243 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
0d53778e | 244 | pr_debug("offset is too large.\n"); |
1ab1457c | 245 | return -1; |
9fb9cbb1 YK |
246 | } |
247 | ||
d56f90a7 ACM |
248 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
249 | const unsigned char *nh = skb_network_header(skb); | |
1ab1457c | 250 | skb->csum = csum_sub(skb->csum, |
d56f90a7 | 251 | csum_partial(nh, (u8 *)(fhdr + 1) - nh, |
9fb9cbb1 | 252 | 0)); |
d56f90a7 | 253 | } |
9fb9cbb1 YK |
254 | |
255 | /* Is this the final fragment? */ | |
256 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
257 | /* If we already have some bits beyond end | |
258 | * or have different end, the segment is corrupted. | |
259 | */ | |
5ab11c98 PE |
260 | if (end < fq->q.len || |
261 | ((fq->q.last_in & LAST_IN) && end != fq->q.len)) { | |
0d53778e | 262 | pr_debug("already received last fragment\n"); |
9fb9cbb1 YK |
263 | goto err; |
264 | } | |
5ab11c98 PE |
265 | fq->q.last_in |= LAST_IN; |
266 | fq->q.len = end; | |
9fb9cbb1 YK |
267 | } else { |
268 | /* Check if the fragment is rounded to 8 bytes. | |
269 | * Required by the RFC. | |
270 | */ | |
271 | if (end & 0x7) { | |
272 | /* RFC2460 says always send parameter problem in | |
273 | * this case. -DaveM | |
274 | */ | |
0d53778e | 275 | pr_debug("end of fragment not rounded to 8 bytes.\n"); |
9fb9cbb1 YK |
276 | return -1; |
277 | } | |
5ab11c98 | 278 | if (end > fq->q.len) { |
9fb9cbb1 | 279 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 280 | if (fq->q.last_in & LAST_IN) { |
0d53778e | 281 | pr_debug("last packet already reached.\n"); |
9fb9cbb1 YK |
282 | goto err; |
283 | } | |
5ab11c98 | 284 | fq->q.len = end; |
9fb9cbb1 YK |
285 | } |
286 | } | |
287 | ||
288 | if (end == offset) | |
289 | goto err; | |
290 | ||
291 | /* Point into the IP datagram 'data' part. */ | |
292 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) { | |
0d53778e | 293 | pr_debug("queue: message is too short.\n"); |
9fb9cbb1 YK |
294 | goto err; |
295 | } | |
b38dfee3 | 296 | if (pskb_trim_rcsum(skb, end - offset)) { |
0d53778e | 297 | pr_debug("Can't trim\n"); |
b38dfee3 | 298 | goto err; |
9fb9cbb1 YK |
299 | } |
300 | ||
301 | /* Find out which fragments are in front and at the back of us | |
302 | * in the chain of fragments so far. We must know where to put | |
303 | * this fragment, right? | |
304 | */ | |
305 | prev = NULL; | |
5ab11c98 | 306 | for (next = fq->q.fragments; next != NULL; next = next->next) { |
9fb9cbb1 YK |
307 | if (NFCT_FRAG6_CB(next)->offset >= offset) |
308 | break; /* bingo! */ | |
309 | prev = next; | |
310 | } | |
311 | ||
312 | /* We found where to put this one. Check for overlap with | |
313 | * preceding fragment, and, if needed, align things so that | |
314 | * any overlaps are eliminated. | |
315 | */ | |
316 | if (prev) { | |
317 | int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset; | |
318 | ||
319 | if (i > 0) { | |
320 | offset += i; | |
321 | if (end <= offset) { | |
0d53778e | 322 | pr_debug("overlap\n"); |
9fb9cbb1 YK |
323 | goto err; |
324 | } | |
325 | if (!pskb_pull(skb, i)) { | |
0d53778e | 326 | pr_debug("Can't pull\n"); |
9fb9cbb1 YK |
327 | goto err; |
328 | } | |
329 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
330 | skb->ip_summed = CHECKSUM_NONE; | |
331 | } | |
332 | } | |
333 | ||
334 | /* Look for overlap with succeeding segments. | |
335 | * If we can merge fragments, do it. | |
336 | */ | |
337 | while (next && NFCT_FRAG6_CB(next)->offset < end) { | |
338 | /* overlap is 'i' bytes */ | |
339 | int i = end - NFCT_FRAG6_CB(next)->offset; | |
340 | ||
341 | if (i < next->len) { | |
342 | /* Eat head of the next overlapped fragment | |
343 | * and leave the loop. The next ones cannot overlap. | |
344 | */ | |
0d53778e | 345 | pr_debug("Eat head of the overlapped parts.: %d", i); |
9fb9cbb1 YK |
346 | if (!pskb_pull(next, i)) |
347 | goto err; | |
348 | ||
349 | /* next fragment */ | |
350 | NFCT_FRAG6_CB(next)->offset += i; | |
5ab11c98 | 351 | fq->q.meat -= i; |
9fb9cbb1 YK |
352 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
353 | next->ip_summed = CHECKSUM_NONE; | |
354 | break; | |
355 | } else { | |
356 | struct sk_buff *free_it = next; | |
357 | ||
358 | /* Old fragmnet is completely overridden with | |
359 | * new one drop it. | |
360 | */ | |
361 | next = next->next; | |
362 | ||
363 | if (prev) | |
364 | prev->next = next; | |
365 | else | |
5ab11c98 | 366 | fq->q.fragments = next; |
9fb9cbb1 | 367 | |
5ab11c98 | 368 | fq->q.meat -= free_it->len; |
1ba430bc | 369 | frag_kfree_skb(free_it, NULL); |
9fb9cbb1 YK |
370 | } |
371 | } | |
372 | ||
373 | NFCT_FRAG6_CB(skb)->offset = offset; | |
374 | ||
375 | /* Insert this fragment in the chain of fragments. */ | |
376 | skb->next = next; | |
377 | if (prev) | |
378 | prev->next = skb; | |
379 | else | |
5ab11c98 | 380 | fq->q.fragments = skb; |
9fb9cbb1 YK |
381 | |
382 | skb->dev = NULL; | |
5ab11c98 PE |
383 | fq->q.stamp = skb->tstamp; |
384 | fq->q.meat += skb->len; | |
6ddc0822 | 385 | atomic_add(skb->truesize, &nf_init_frags.mem); |
9fb9cbb1 YK |
386 | |
387 | /* The first fragment. | |
388 | * nhoffset is obtained from the first fragment, of course. | |
389 | */ | |
390 | if (offset == 0) { | |
391 | fq->nhoffset = nhoff; | |
5ab11c98 | 392 | fq->q.last_in |= FIRST_IN; |
9fb9cbb1 | 393 | } |
7eb95156 PE |
394 | write_lock(&nf_frags.lock); |
395 | list_move_tail(&fq->q.lru_list, &nf_frags.lru_list); | |
396 | write_unlock(&nf_frags.lock); | |
9fb9cbb1 YK |
397 | return 0; |
398 | ||
399 | err: | |
400 | return -1; | |
401 | } | |
402 | ||
403 | /* | |
404 | * Check if this packet is complete. | |
405 | * Returns NULL on failure by any reason, and pointer | |
406 | * to current nexthdr field in reassembled frame. | |
407 | * | |
408 | * It is called with locked fq, and caller must check that | |
409 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
410 | * the last and the first frames arrived and all the bits are here. | |
411 | */ | |
412 | static struct sk_buff * | |
413 | nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev) | |
414 | { | |
5ab11c98 | 415 | struct sk_buff *fp, *op, *head = fq->q.fragments; |
9fb9cbb1 YK |
416 | int payload_len; |
417 | ||
418 | fq_kill(fq); | |
419 | ||
420 | BUG_TRAP(head != NULL); | |
421 | BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0); | |
422 | ||
423 | /* Unfragmented part is taken from the first segment. */ | |
d56f90a7 | 424 | payload_len = ((head->data - skb_network_header(head)) - |
5ab11c98 | 425 | sizeof(struct ipv6hdr) + fq->q.len - |
d56f90a7 | 426 | sizeof(struct frag_hdr)); |
9fb9cbb1 | 427 | if (payload_len > IPV6_MAXPLEN) { |
0d53778e | 428 | pr_debug("payload len is too large.\n"); |
9fb9cbb1 YK |
429 | goto out_oversize; |
430 | } | |
431 | ||
432 | /* Head of list must not be cloned. */ | |
433 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) { | |
0d53778e | 434 | pr_debug("skb is cloned but can't expand head"); |
9fb9cbb1 YK |
435 | goto out_oom; |
436 | } | |
437 | ||
438 | /* If the first fragment is fragmented itself, we split | |
439 | * it to two chunks: the first with data and paged part | |
440 | * and the second, holding only fragments. */ | |
441 | if (skb_shinfo(head)->frag_list) { | |
442 | struct sk_buff *clone; | |
443 | int i, plen = 0; | |
444 | ||
445 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) { | |
0d53778e | 446 | pr_debug("Can't alloc skb\n"); |
9fb9cbb1 YK |
447 | goto out_oom; |
448 | } | |
449 | clone->next = head->next; | |
450 | head->next = clone; | |
451 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
452 | skb_shinfo(head)->frag_list = NULL; | |
453 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
454 | plen += skb_shinfo(head)->frags[i].size; | |
455 | clone->len = clone->data_len = head->data_len - plen; | |
456 | head->data_len -= clone->len; | |
457 | head->len -= clone->len; | |
458 | clone->csum = 0; | |
459 | clone->ip_summed = head->ip_summed; | |
460 | ||
461 | NFCT_FRAG6_CB(clone)->orig = NULL; | |
6ddc0822 | 462 | atomic_add(clone->truesize, &nf_init_frags.mem); |
9fb9cbb1 YK |
463 | } |
464 | ||
465 | /* We have to remove fragment header from datagram and to relocate | |
466 | * header in order to calculate ICV correctly. */ | |
bff9b61c | 467 | skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0]; |
1ab1457c | 468 | memmove(head->head + sizeof(struct frag_hdr), head->head, |
9fb9cbb1 | 469 | (head->data - head->head) - sizeof(struct frag_hdr)); |
b0e380b1 ACM |
470 | head->mac_header += sizeof(struct frag_hdr); |
471 | head->network_header += sizeof(struct frag_hdr); | |
9fb9cbb1 YK |
472 | |
473 | skb_shinfo(head)->frag_list = head->next; | |
badff6d0 | 474 | skb_reset_transport_header(head); |
d56f90a7 | 475 | skb_push(head, head->data - skb_network_header(head)); |
6ddc0822 | 476 | atomic_sub(head->truesize, &nf_init_frags.mem); |
9fb9cbb1 YK |
477 | |
478 | for (fp=head->next; fp; fp = fp->next) { | |
479 | head->data_len += fp->len; | |
480 | head->len += fp->len; | |
481 | if (head->ip_summed != fp->ip_summed) | |
482 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 483 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
9fb9cbb1 YK |
484 | head->csum = csum_add(head->csum, fp->csum); |
485 | head->truesize += fp->truesize; | |
6ddc0822 | 486 | atomic_sub(fp->truesize, &nf_init_frags.mem); |
9fb9cbb1 YK |
487 | } |
488 | ||
489 | head->next = NULL; | |
490 | head->dev = dev; | |
5ab11c98 | 491 | head->tstamp = fq->q.stamp; |
0660e03f | 492 | ipv6_hdr(head)->payload_len = htons(payload_len); |
9fb9cbb1 YK |
493 | |
494 | /* Yes, and fold redundant checksum back. 8) */ | |
84fa7933 | 495 | if (head->ip_summed == CHECKSUM_COMPLETE) |
d56f90a7 | 496 | head->csum = csum_partial(skb_network_header(head), |
cfe1fc77 | 497 | skb_network_header_len(head), |
d56f90a7 | 498 | head->csum); |
9fb9cbb1 | 499 | |
5ab11c98 | 500 | fq->q.fragments = NULL; |
9fb9cbb1 YK |
501 | |
502 | /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */ | |
503 | fp = skb_shinfo(head)->frag_list; | |
504 | if (NFCT_FRAG6_CB(fp)->orig == NULL) | |
505 | /* at above code, head skb is divided into two skbs. */ | |
506 | fp = fp->next; | |
507 | ||
508 | op = NFCT_FRAG6_CB(head)->orig; | |
509 | for (; fp; fp = fp->next) { | |
510 | struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig; | |
511 | ||
512 | op->next = orig; | |
513 | op = orig; | |
514 | NFCT_FRAG6_CB(fp)->orig = NULL; | |
515 | } | |
516 | ||
517 | return head; | |
518 | ||
519 | out_oversize: | |
520 | if (net_ratelimit()) | |
521 | printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len); | |
522 | goto out_fail; | |
523 | out_oom: | |
524 | if (net_ratelimit()) | |
525 | printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n"); | |
526 | out_fail: | |
527 | return NULL; | |
528 | } | |
529 | ||
530 | /* | |
531 | * find the header just before Fragment Header. | |
532 | * | |
533 | * if success return 0 and set ... | |
534 | * (*prevhdrp): the value of "Next Header Field" in the header | |
535 | * just before Fragment Header. | |
536 | * (*prevhoff): the offset of "Next Header Field" in the header | |
537 | * just before Fragment Header. | |
538 | * (*fhoff) : the offset of Fragment Header. | |
539 | * | |
540 | * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c | |
541 | * | |
542 | */ | |
543 | static int | |
544 | find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff) | |
545 | { | |
0660e03f | 546 | u8 nexthdr = ipv6_hdr(skb)->nexthdr; |
6b88dd96 ACM |
547 | const int netoff = skb_network_offset(skb); |
548 | u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr); | |
549 | int start = netoff + sizeof(struct ipv6hdr); | |
9fb9cbb1 YK |
550 | int len = skb->len - start; |
551 | u8 prevhdr = NEXTHDR_IPV6; | |
552 | ||
1ab1457c YH |
553 | while (nexthdr != NEXTHDR_FRAGMENT) { |
554 | struct ipv6_opt_hdr hdr; | |
555 | int hdrlen; | |
9fb9cbb1 YK |
556 | |
557 | if (!ipv6_ext_hdr(nexthdr)) { | |
558 | return -1; | |
559 | } | |
1ab1457c | 560 | if (len < (int)sizeof(struct ipv6_opt_hdr)) { |
0d53778e | 561 | pr_debug("too short\n"); |
9fb9cbb1 YK |
562 | return -1; |
563 | } | |
1ab1457c | 564 | if (nexthdr == NEXTHDR_NONE) { |
0d53778e | 565 | pr_debug("next header is none\n"); |
9fb9cbb1 YK |
566 | return -1; |
567 | } | |
1ab1457c YH |
568 | if (skb_copy_bits(skb, start, &hdr, sizeof(hdr))) |
569 | BUG(); | |
570 | if (nexthdr == NEXTHDR_AUTH) | |
571 | hdrlen = (hdr.hdrlen+2)<<2; | |
572 | else | |
573 | hdrlen = ipv6_optlen(&hdr); | |
9fb9cbb1 YK |
574 | |
575 | prevhdr = nexthdr; | |
576 | prev_nhoff = start; | |
577 | ||
1ab1457c YH |
578 | nexthdr = hdr.nexthdr; |
579 | len -= hdrlen; | |
580 | start += hdrlen; | |
581 | } | |
9fb9cbb1 YK |
582 | |
583 | if (len < 0) | |
584 | return -1; | |
585 | ||
586 | *prevhdrp = prevhdr; | |
587 | *prevhoff = prev_nhoff; | |
588 | *fhoff = start; | |
589 | ||
590 | return 0; | |
591 | } | |
592 | ||
593 | struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb) | |
594 | { | |
1ab1457c | 595 | struct sk_buff *clone; |
9fb9cbb1 YK |
596 | struct net_device *dev = skb->dev; |
597 | struct frag_hdr *fhdr; | |
598 | struct nf_ct_frag6_queue *fq; | |
599 | struct ipv6hdr *hdr; | |
600 | int fhoff, nhoff; | |
601 | u8 prevhdr; | |
602 | struct sk_buff *ret_skb = NULL; | |
603 | ||
604 | /* Jumbo payload inhibits frag. header */ | |
0660e03f | 605 | if (ipv6_hdr(skb)->payload_len == 0) { |
0d53778e | 606 | pr_debug("payload len = 0\n"); |
9fb9cbb1 YK |
607 | return skb; |
608 | } | |
609 | ||
610 | if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0) | |
611 | return skb; | |
612 | ||
613 | clone = skb_clone(skb, GFP_ATOMIC); | |
614 | if (clone == NULL) { | |
0d53778e | 615 | pr_debug("Can't clone skb\n"); |
9fb9cbb1 YK |
616 | return skb; |
617 | } | |
618 | ||
619 | NFCT_FRAG6_CB(clone)->orig = skb; | |
620 | ||
621 | if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) { | |
0d53778e | 622 | pr_debug("message is too short.\n"); |
9fb9cbb1 YK |
623 | goto ret_orig; |
624 | } | |
625 | ||
967b05f6 | 626 | skb_set_transport_header(clone, fhoff); |
0660e03f | 627 | hdr = ipv6_hdr(clone); |
bff9b61c | 628 | fhdr = (struct frag_hdr *)skb_transport_header(clone); |
9fb9cbb1 YK |
629 | |
630 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
0d53778e | 631 | pr_debug("Invalid fragment offset\n"); |
9fb9cbb1 YK |
632 | /* It is not a fragmented frame */ |
633 | goto ret_orig; | |
634 | } | |
635 | ||
6ddc0822 | 636 | if (atomic_read(&nf_init_frags.mem) > nf_frags_ctl.high_thresh) |
9fb9cbb1 YK |
637 | nf_ct_frag6_evictor(); |
638 | ||
639 | fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr); | |
640 | if (fq == NULL) { | |
0d53778e | 641 | pr_debug("Can't find and can't create new queue\n"); |
9fb9cbb1 YK |
642 | goto ret_orig; |
643 | } | |
644 | ||
5ab11c98 | 645 | spin_lock(&fq->q.lock); |
9fb9cbb1 YK |
646 | |
647 | if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) { | |
5ab11c98 | 648 | spin_unlock(&fq->q.lock); |
0d53778e | 649 | pr_debug("Can't insert skb to queue\n"); |
4b6cb5d8 | 650 | fq_put(fq); |
9fb9cbb1 YK |
651 | goto ret_orig; |
652 | } | |
653 | ||
5ab11c98 | 654 | if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) { |
9fb9cbb1 YK |
655 | ret_skb = nf_ct_frag6_reasm(fq, dev); |
656 | if (ret_skb == NULL) | |
0d53778e | 657 | pr_debug("Can't reassemble fragmented packets\n"); |
9fb9cbb1 | 658 | } |
5ab11c98 | 659 | spin_unlock(&fq->q.lock); |
9fb9cbb1 | 660 | |
4b6cb5d8 | 661 | fq_put(fq); |
9fb9cbb1 YK |
662 | return ret_skb; |
663 | ||
664 | ret_orig: | |
665 | kfree_skb(clone); | |
666 | return skb; | |
667 | } | |
668 | ||
669 | void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb, | |
670 | struct net_device *in, struct net_device *out, | |
671 | int (*okfn)(struct sk_buff *)) | |
672 | { | |
673 | struct sk_buff *s, *s2; | |
674 | ||
675 | for (s = NFCT_FRAG6_CB(skb)->orig; s;) { | |
676 | nf_conntrack_put_reasm(s->nfct_reasm); | |
677 | nf_conntrack_get_reasm(skb); | |
678 | s->nfct_reasm = skb; | |
679 | ||
680 | s2 = s->next; | |
f9f02cca PM |
681 | s->next = NULL; |
682 | ||
9fb9cbb1 YK |
683 | NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn, |
684 | NF_IP6_PRI_CONNTRACK_DEFRAG + 1); | |
685 | s = s2; | |
686 | } | |
687 | nf_conntrack_put_reasm(skb); | |
688 | } | |
689 | ||
690 | int nf_ct_frag6_kfree_frags(struct sk_buff *skb) | |
691 | { | |
692 | struct sk_buff *s, *s2; | |
693 | ||
694 | for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) { | |
695 | ||
696 | s2 = s->next; | |
697 | kfree_skb(s); | |
698 | } | |
699 | ||
700 | kfree_skb(skb); | |
701 | ||
702 | return 0; | |
703 | } | |
704 | ||
705 | int nf_ct_frag6_init(void) | |
706 | { | |
04128f23 | 707 | nf_frags.ctl = &nf_frags_ctl; |
321a3a99 | 708 | nf_frags.hashfn = nf_hashfn; |
c6fda282 | 709 | nf_frags.constructor = ip6_frag_init; |
c9547709 | 710 | nf_frags.destructor = NULL; |
1e4b8287 PE |
711 | nf_frags.skb_free = nf_skb_free; |
712 | nf_frags.qsize = sizeof(struct nf_ct_frag6_queue); | |
abd6523d | 713 | nf_frags.match = ip6_frag_match; |
e521db9d | 714 | nf_frags.frag_expire = nf_ct_frag6_expire; |
e5a2bb84 | 715 | inet_frags_init_net(&nf_init_frags); |
7eb95156 | 716 | inet_frags_init(&nf_frags); |
9fb9cbb1 YK |
717 | |
718 | return 0; | |
719 | } | |
720 | ||
721 | void nf_ct_frag6_cleanup(void) | |
722 | { | |
7eb95156 PE |
723 | inet_frags_fini(&nf_frags); |
724 | ||
04128f23 | 725 | nf_frags_ctl.low_thresh = 0; |
9fb9cbb1 YK |
726 | nf_ct_frag6_evictor(); |
727 | } |