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 | ||
17 | #include <linux/config.h> | |
18 | #include <linux/errno.h> | |
19 | #include <linux/types.h> | |
20 | #include <linux/string.h> | |
21 | #include <linux/socket.h> | |
22 | #include <linux/sockios.h> | |
23 | #include <linux/jiffies.h> | |
24 | #include <linux/net.h> | |
25 | #include <linux/list.h> | |
26 | #include <linux/netdevice.h> | |
27 | #include <linux/in6.h> | |
28 | #include <linux/ipv6.h> | |
29 | #include <linux/icmpv6.h> | |
30 | #include <linux/random.h> | |
31 | #include <linux/jhash.h> | |
32 | ||
33 | #include <net/sock.h> | |
34 | #include <net/snmp.h> | |
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 | ||
48 | #if 0 | |
49 | #define DEBUGP printk | |
50 | #else | |
51 | #define DEBUGP(format, args...) | |
52 | #endif | |
53 | ||
54 | #define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */ | |
55 | #define NF_CT_FRAG6_LOW_THRESH 196608 /* == 192*1024 */ | |
56 | #define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT | |
57 | ||
58 | int nf_ct_frag6_high_thresh = 256*1024; | |
59 | int nf_ct_frag6_low_thresh = 192*1024; | |
60 | int nf_ct_frag6_timeout = IPV6_FRAG_TIMEOUT; | |
61 | ||
62 | struct nf_ct_frag6_skb_cb | |
63 | { | |
64 | struct inet6_skb_parm h; | |
65 | int offset; | |
66 | struct sk_buff *orig; | |
67 | }; | |
68 | ||
69 | #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb)) | |
70 | ||
71 | struct nf_ct_frag6_queue | |
72 | { | |
73 | struct nf_ct_frag6_queue *next; | |
74 | struct list_head lru_list; /* lru list member */ | |
75 | ||
76 | __u32 id; /* fragment id */ | |
77 | struct in6_addr saddr; | |
78 | struct in6_addr daddr; | |
79 | ||
80 | spinlock_t lock; | |
81 | atomic_t refcnt; | |
82 | struct timer_list timer; /* expire timer */ | |
83 | struct sk_buff *fragments; | |
84 | int len; | |
85 | int meat; | |
86 | struct timeval stamp; | |
87 | unsigned int csum; | |
88 | __u8 last_in; /* has first/last segment arrived? */ | |
89 | #define COMPLETE 4 | |
90 | #define FIRST_IN 2 | |
91 | #define LAST_IN 1 | |
92 | __u16 nhoffset; | |
93 | struct nf_ct_frag6_queue **pprev; | |
94 | }; | |
95 | ||
96 | /* Hash table. */ | |
97 | ||
98 | #define FRAG6Q_HASHSZ 64 | |
99 | ||
100 | static struct nf_ct_frag6_queue *nf_ct_frag6_hash[FRAG6Q_HASHSZ]; | |
101 | static rwlock_t nf_ct_frag6_lock = RW_LOCK_UNLOCKED; | |
102 | static u32 nf_ct_frag6_hash_rnd; | |
103 | static LIST_HEAD(nf_ct_frag6_lru_list); | |
104 | int nf_ct_frag6_nqueues = 0; | |
105 | ||
106 | static __inline__ void __fq_unlink(struct nf_ct_frag6_queue *fq) | |
107 | { | |
108 | if (fq->next) | |
109 | fq->next->pprev = fq->pprev; | |
110 | *fq->pprev = fq->next; | |
111 | list_del(&fq->lru_list); | |
112 | nf_ct_frag6_nqueues--; | |
113 | } | |
114 | ||
115 | static __inline__ void fq_unlink(struct nf_ct_frag6_queue *fq) | |
116 | { | |
117 | write_lock(&nf_ct_frag6_lock); | |
118 | __fq_unlink(fq); | |
119 | write_unlock(&nf_ct_frag6_lock); | |
120 | } | |
121 | ||
122 | static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr, | |
123 | struct in6_addr *daddr) | |
124 | { | |
125 | u32 a, b, c; | |
126 | ||
127 | a = saddr->s6_addr32[0]; | |
128 | b = saddr->s6_addr32[1]; | |
129 | c = saddr->s6_addr32[2]; | |
130 | ||
131 | a += JHASH_GOLDEN_RATIO; | |
132 | b += JHASH_GOLDEN_RATIO; | |
133 | c += nf_ct_frag6_hash_rnd; | |
134 | __jhash_mix(a, b, c); | |
135 | ||
136 | a += saddr->s6_addr32[3]; | |
137 | b += daddr->s6_addr32[0]; | |
138 | c += daddr->s6_addr32[1]; | |
139 | __jhash_mix(a, b, c); | |
140 | ||
141 | a += daddr->s6_addr32[2]; | |
142 | b += daddr->s6_addr32[3]; | |
143 | c += id; | |
144 | __jhash_mix(a, b, c); | |
145 | ||
146 | return c & (FRAG6Q_HASHSZ - 1); | |
147 | } | |
148 | ||
149 | static struct timer_list nf_ct_frag6_secret_timer; | |
150 | int nf_ct_frag6_secret_interval = 10 * 60 * HZ; | |
151 | ||
152 | static void nf_ct_frag6_secret_rebuild(unsigned long dummy) | |
153 | { | |
154 | unsigned long now = jiffies; | |
155 | int i; | |
156 | ||
157 | write_lock(&nf_ct_frag6_lock); | |
158 | get_random_bytes(&nf_ct_frag6_hash_rnd, sizeof(u32)); | |
159 | for (i = 0; i < FRAG6Q_HASHSZ; i++) { | |
160 | struct nf_ct_frag6_queue *q; | |
161 | ||
162 | q = nf_ct_frag6_hash[i]; | |
163 | while (q) { | |
164 | struct nf_ct_frag6_queue *next = q->next; | |
165 | unsigned int hval = ip6qhashfn(q->id, | |
166 | &q->saddr, | |
167 | &q->daddr); | |
168 | ||
169 | if (hval != i) { | |
170 | /* Unlink. */ | |
171 | if (q->next) | |
172 | q->next->pprev = q->pprev; | |
173 | *q->pprev = q->next; | |
174 | ||
175 | /* Relink to new hash chain. */ | |
176 | if ((q->next = nf_ct_frag6_hash[hval]) != NULL) | |
177 | q->next->pprev = &q->next; | |
178 | nf_ct_frag6_hash[hval] = q; | |
179 | q->pprev = &nf_ct_frag6_hash[hval]; | |
180 | } | |
181 | ||
182 | q = next; | |
183 | } | |
184 | } | |
185 | write_unlock(&nf_ct_frag6_lock); | |
186 | ||
187 | mod_timer(&nf_ct_frag6_secret_timer, now + nf_ct_frag6_secret_interval); | |
188 | } | |
189 | ||
190 | atomic_t nf_ct_frag6_mem = ATOMIC_INIT(0); | |
191 | ||
192 | /* Memory Tracking Functions. */ | |
193 | static inline void frag_kfree_skb(struct sk_buff *skb) | |
194 | { | |
195 | atomic_sub(skb->truesize, &nf_ct_frag6_mem); | |
196 | if (NFCT_FRAG6_CB(skb)->orig) | |
197 | kfree_skb(NFCT_FRAG6_CB(skb)->orig); | |
198 | ||
199 | kfree_skb(skb); | |
200 | } | |
201 | ||
202 | static inline void frag_free_queue(struct nf_ct_frag6_queue *fq) | |
203 | { | |
204 | atomic_sub(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem); | |
205 | kfree(fq); | |
206 | } | |
207 | ||
208 | static inline struct nf_ct_frag6_queue *frag_alloc_queue(void) | |
209 | { | |
210 | struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC); | |
211 | ||
212 | if (!fq) | |
213 | return NULL; | |
214 | atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem); | |
215 | return fq; | |
216 | } | |
217 | ||
218 | /* Destruction primitives. */ | |
219 | ||
220 | /* Complete destruction of fq. */ | |
221 | static void nf_ct_frag6_destroy(struct nf_ct_frag6_queue *fq) | |
222 | { | |
223 | struct sk_buff *fp; | |
224 | ||
225 | BUG_TRAP(fq->last_in&COMPLETE); | |
226 | BUG_TRAP(del_timer(&fq->timer) == 0); | |
227 | ||
228 | /* Release all fragment data. */ | |
229 | fp = fq->fragments; | |
230 | while (fp) { | |
231 | struct sk_buff *xp = fp->next; | |
232 | ||
233 | frag_kfree_skb(fp); | |
234 | fp = xp; | |
235 | } | |
236 | ||
237 | frag_free_queue(fq); | |
238 | } | |
239 | ||
240 | static __inline__ void fq_put(struct nf_ct_frag6_queue *fq) | |
241 | { | |
242 | if (atomic_dec_and_test(&fq->refcnt)) | |
243 | nf_ct_frag6_destroy(fq); | |
244 | } | |
245 | ||
246 | /* Kill fq entry. It is not destroyed immediately, | |
247 | * because caller (and someone more) holds reference count. | |
248 | */ | |
249 | static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq) | |
250 | { | |
251 | if (del_timer(&fq->timer)) | |
252 | atomic_dec(&fq->refcnt); | |
253 | ||
254 | if (!(fq->last_in & COMPLETE)) { | |
255 | fq_unlink(fq); | |
256 | atomic_dec(&fq->refcnt); | |
257 | fq->last_in |= COMPLETE; | |
258 | } | |
259 | } | |
260 | ||
261 | static void nf_ct_frag6_evictor(void) | |
262 | { | |
263 | struct nf_ct_frag6_queue *fq; | |
264 | struct list_head *tmp; | |
265 | ||
266 | for (;;) { | |
267 | if (atomic_read(&nf_ct_frag6_mem) <= nf_ct_frag6_low_thresh) | |
268 | return; | |
269 | read_lock(&nf_ct_frag6_lock); | |
270 | if (list_empty(&nf_ct_frag6_lru_list)) { | |
271 | read_unlock(&nf_ct_frag6_lock); | |
272 | return; | |
273 | } | |
274 | tmp = nf_ct_frag6_lru_list.next; | |
275 | fq = list_entry(tmp, struct nf_ct_frag6_queue, lru_list); | |
276 | atomic_inc(&fq->refcnt); | |
277 | read_unlock(&nf_ct_frag6_lock); | |
278 | ||
279 | spin_lock(&fq->lock); | |
280 | if (!(fq->last_in&COMPLETE)) | |
281 | fq_kill(fq); | |
282 | spin_unlock(&fq->lock); | |
283 | ||
284 | fq_put(fq); | |
285 | } | |
286 | } | |
287 | ||
288 | static void nf_ct_frag6_expire(unsigned long data) | |
289 | { | |
290 | struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data; | |
291 | ||
292 | spin_lock(&fq->lock); | |
293 | ||
294 | if (fq->last_in & COMPLETE) | |
295 | goto out; | |
296 | ||
297 | fq_kill(fq); | |
298 | ||
299 | out: | |
300 | spin_unlock(&fq->lock); | |
301 | fq_put(fq); | |
302 | } | |
303 | ||
304 | /* Creation primitives. */ | |
305 | ||
306 | ||
307 | static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash, | |
308 | struct nf_ct_frag6_queue *fq_in) | |
309 | { | |
310 | struct nf_ct_frag6_queue *fq; | |
311 | ||
312 | write_lock(&nf_ct_frag6_lock); | |
313 | #ifdef CONFIG_SMP | |
314 | for (fq = nf_ct_frag6_hash[hash]; fq; fq = fq->next) { | |
315 | if (fq->id == fq_in->id && | |
316 | !ipv6_addr_cmp(&fq_in->saddr, &fq->saddr) && | |
317 | !ipv6_addr_cmp(&fq_in->daddr, &fq->daddr)) { | |
318 | atomic_inc(&fq->refcnt); | |
319 | write_unlock(&nf_ct_frag6_lock); | |
320 | fq_in->last_in |= COMPLETE; | |
321 | fq_put(fq_in); | |
322 | return fq; | |
323 | } | |
324 | } | |
325 | #endif | |
326 | fq = fq_in; | |
327 | ||
328 | if (!mod_timer(&fq->timer, jiffies + nf_ct_frag6_timeout)) | |
329 | atomic_inc(&fq->refcnt); | |
330 | ||
331 | atomic_inc(&fq->refcnt); | |
332 | if ((fq->next = nf_ct_frag6_hash[hash]) != NULL) | |
333 | fq->next->pprev = &fq->next; | |
334 | nf_ct_frag6_hash[hash] = fq; | |
335 | fq->pprev = &nf_ct_frag6_hash[hash]; | |
336 | INIT_LIST_HEAD(&fq->lru_list); | |
337 | list_add_tail(&fq->lru_list, &nf_ct_frag6_lru_list); | |
338 | nf_ct_frag6_nqueues++; | |
339 | write_unlock(&nf_ct_frag6_lock); | |
340 | return fq; | |
341 | } | |
342 | ||
343 | ||
344 | static struct nf_ct_frag6_queue * | |
345 | nf_ct_frag6_create(unsigned int hash, u32 id, struct in6_addr *src, struct in6_addr *dst) | |
346 | { | |
347 | struct nf_ct_frag6_queue *fq; | |
348 | ||
349 | if ((fq = frag_alloc_queue()) == NULL) { | |
350 | DEBUGP("Can't alloc new queue\n"); | |
351 | goto oom; | |
352 | } | |
353 | ||
354 | memset(fq, 0, sizeof(struct nf_ct_frag6_queue)); | |
355 | ||
356 | fq->id = id; | |
357 | ipv6_addr_copy(&fq->saddr, src); | |
358 | ipv6_addr_copy(&fq->daddr, dst); | |
359 | ||
360 | init_timer(&fq->timer); | |
361 | fq->timer.function = nf_ct_frag6_expire; | |
362 | fq->timer.data = (long) fq; | |
363 | fq->lock = SPIN_LOCK_UNLOCKED; | |
364 | atomic_set(&fq->refcnt, 1); | |
365 | ||
366 | return nf_ct_frag6_intern(hash, fq); | |
367 | ||
368 | oom: | |
369 | return NULL; | |
370 | } | |
371 | ||
372 | static __inline__ struct nf_ct_frag6_queue * | |
373 | fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst) | |
374 | { | |
375 | struct nf_ct_frag6_queue *fq; | |
376 | unsigned int hash = ip6qhashfn(id, src, dst); | |
377 | ||
378 | read_lock(&nf_ct_frag6_lock); | |
379 | for (fq = nf_ct_frag6_hash[hash]; fq; fq = fq->next) { | |
380 | if (fq->id == id && | |
381 | !ipv6_addr_cmp(src, &fq->saddr) && | |
382 | !ipv6_addr_cmp(dst, &fq->daddr)) { | |
383 | atomic_inc(&fq->refcnt); | |
384 | read_unlock(&nf_ct_frag6_lock); | |
385 | return fq; | |
386 | } | |
387 | } | |
388 | read_unlock(&nf_ct_frag6_lock); | |
389 | ||
390 | return nf_ct_frag6_create(hash, id, src, dst); | |
391 | } | |
392 | ||
393 | ||
394 | static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb, | |
395 | struct frag_hdr *fhdr, int nhoff) | |
396 | { | |
397 | struct sk_buff *prev, *next; | |
398 | int offset, end; | |
399 | ||
400 | if (fq->last_in & COMPLETE) { | |
401 | DEBUGP("Allready completed\n"); | |
402 | goto err; | |
403 | } | |
404 | ||
405 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
406 | end = offset + (ntohs(skb->nh.ipv6h->payload_len) - | |
407 | ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1))); | |
408 | ||
409 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
410 | DEBUGP("offset is too large.\n"); | |
411 | return -1; | |
412 | } | |
413 | ||
414 | if (skb->ip_summed == CHECKSUM_HW) | |
415 | skb->csum = csum_sub(skb->csum, | |
416 | csum_partial(skb->nh.raw, | |
417 | (u8*)(fhdr + 1) - skb->nh.raw, | |
418 | 0)); | |
419 | ||
420 | /* Is this the final fragment? */ | |
421 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
422 | /* If we already have some bits beyond end | |
423 | * or have different end, the segment is corrupted. | |
424 | */ | |
425 | if (end < fq->len || | |
426 | ((fq->last_in & LAST_IN) && end != fq->len)) { | |
427 | DEBUGP("already received last fragment\n"); | |
428 | goto err; | |
429 | } | |
430 | fq->last_in |= LAST_IN; | |
431 | fq->len = end; | |
432 | } else { | |
433 | /* Check if the fragment is rounded to 8 bytes. | |
434 | * Required by the RFC. | |
435 | */ | |
436 | if (end & 0x7) { | |
437 | /* RFC2460 says always send parameter problem in | |
438 | * this case. -DaveM | |
439 | */ | |
440 | DEBUGP("the end of this fragment is not rounded to 8 bytes.\n"); | |
441 | return -1; | |
442 | } | |
443 | if (end > fq->len) { | |
444 | /* Some bits beyond end -> corruption. */ | |
445 | if (fq->last_in & LAST_IN) { | |
446 | DEBUGP("last packet already reached.\n"); | |
447 | goto err; | |
448 | } | |
449 | fq->len = end; | |
450 | } | |
451 | } | |
452 | ||
453 | if (end == offset) | |
454 | goto err; | |
455 | ||
456 | /* Point into the IP datagram 'data' part. */ | |
457 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) { | |
458 | DEBUGP("queue: message is too short.\n"); | |
459 | goto err; | |
460 | } | |
461 | if (end-offset < skb->len) { | |
462 | if (pskb_trim(skb, end - offset)) { | |
463 | DEBUGP("Can't trim\n"); | |
464 | goto err; | |
465 | } | |
466 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
467 | skb->ip_summed = CHECKSUM_NONE; | |
468 | } | |
469 | ||
470 | /* Find out which fragments are in front and at the back of us | |
471 | * in the chain of fragments so far. We must know where to put | |
472 | * this fragment, right? | |
473 | */ | |
474 | prev = NULL; | |
475 | for (next = fq->fragments; next != NULL; next = next->next) { | |
476 | if (NFCT_FRAG6_CB(next)->offset >= offset) | |
477 | break; /* bingo! */ | |
478 | prev = next; | |
479 | } | |
480 | ||
481 | /* We found where to put this one. Check for overlap with | |
482 | * preceding fragment, and, if needed, align things so that | |
483 | * any overlaps are eliminated. | |
484 | */ | |
485 | if (prev) { | |
486 | int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset; | |
487 | ||
488 | if (i > 0) { | |
489 | offset += i; | |
490 | if (end <= offset) { | |
491 | DEBUGP("overlap\n"); | |
492 | goto err; | |
493 | } | |
494 | if (!pskb_pull(skb, i)) { | |
495 | DEBUGP("Can't pull\n"); | |
496 | goto err; | |
497 | } | |
498 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
499 | skb->ip_summed = CHECKSUM_NONE; | |
500 | } | |
501 | } | |
502 | ||
503 | /* Look for overlap with succeeding segments. | |
504 | * If we can merge fragments, do it. | |
505 | */ | |
506 | while (next && NFCT_FRAG6_CB(next)->offset < end) { | |
507 | /* overlap is 'i' bytes */ | |
508 | int i = end - NFCT_FRAG6_CB(next)->offset; | |
509 | ||
510 | if (i < next->len) { | |
511 | /* Eat head of the next overlapped fragment | |
512 | * and leave the loop. The next ones cannot overlap. | |
513 | */ | |
514 | DEBUGP("Eat head of the overlapped parts.: %d", i); | |
515 | if (!pskb_pull(next, i)) | |
516 | goto err; | |
517 | ||
518 | /* next fragment */ | |
519 | NFCT_FRAG6_CB(next)->offset += i; | |
520 | fq->meat -= i; | |
521 | if (next->ip_summed != CHECKSUM_UNNECESSARY) | |
522 | next->ip_summed = CHECKSUM_NONE; | |
523 | break; | |
524 | } else { | |
525 | struct sk_buff *free_it = next; | |
526 | ||
527 | /* Old fragmnet is completely overridden with | |
528 | * new one drop it. | |
529 | */ | |
530 | next = next->next; | |
531 | ||
532 | if (prev) | |
533 | prev->next = next; | |
534 | else | |
535 | fq->fragments = next; | |
536 | ||
537 | fq->meat -= free_it->len; | |
538 | frag_kfree_skb(free_it); | |
539 | } | |
540 | } | |
541 | ||
542 | NFCT_FRAG6_CB(skb)->offset = offset; | |
543 | ||
544 | /* Insert this fragment in the chain of fragments. */ | |
545 | skb->next = next; | |
546 | if (prev) | |
547 | prev->next = skb; | |
548 | else | |
549 | fq->fragments = skb; | |
550 | ||
551 | skb->dev = NULL; | |
552 | skb_get_timestamp(skb, &fq->stamp); | |
553 | fq->meat += skb->len; | |
554 | atomic_add(skb->truesize, &nf_ct_frag6_mem); | |
555 | ||
556 | /* The first fragment. | |
557 | * nhoffset is obtained from the first fragment, of course. | |
558 | */ | |
559 | if (offset == 0) { | |
560 | fq->nhoffset = nhoff; | |
561 | fq->last_in |= FIRST_IN; | |
562 | } | |
563 | write_lock(&nf_ct_frag6_lock); | |
564 | list_move_tail(&fq->lru_list, &nf_ct_frag6_lru_list); | |
565 | write_unlock(&nf_ct_frag6_lock); | |
566 | return 0; | |
567 | ||
568 | err: | |
569 | return -1; | |
570 | } | |
571 | ||
572 | /* | |
573 | * Check if this packet is complete. | |
574 | * Returns NULL on failure by any reason, and pointer | |
575 | * to current nexthdr field in reassembled frame. | |
576 | * | |
577 | * It is called with locked fq, and caller must check that | |
578 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
579 | * the last and the first frames arrived and all the bits are here. | |
580 | */ | |
581 | static struct sk_buff * | |
582 | nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev) | |
583 | { | |
584 | struct sk_buff *fp, *op, *head = fq->fragments; | |
585 | int payload_len; | |
586 | ||
587 | fq_kill(fq); | |
588 | ||
589 | BUG_TRAP(head != NULL); | |
590 | BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0); | |
591 | ||
592 | /* Unfragmented part is taken from the first segment. */ | |
593 | payload_len = (head->data - head->nh.raw) - sizeof(struct ipv6hdr) + fq->len - sizeof(struct frag_hdr); | |
594 | if (payload_len > IPV6_MAXPLEN) { | |
595 | DEBUGP("payload len is too large.\n"); | |
596 | goto out_oversize; | |
597 | } | |
598 | ||
599 | /* Head of list must not be cloned. */ | |
600 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) { | |
601 | DEBUGP("skb is cloned but can't expand head"); | |
602 | goto out_oom; | |
603 | } | |
604 | ||
605 | /* If the first fragment is fragmented itself, we split | |
606 | * it to two chunks: the first with data and paged part | |
607 | * and the second, holding only fragments. */ | |
608 | if (skb_shinfo(head)->frag_list) { | |
609 | struct sk_buff *clone; | |
610 | int i, plen = 0; | |
611 | ||
612 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) { | |
613 | DEBUGP("Can't alloc skb\n"); | |
614 | goto out_oom; | |
615 | } | |
616 | clone->next = head->next; | |
617 | head->next = clone; | |
618 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
619 | skb_shinfo(head)->frag_list = NULL; | |
620 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
621 | plen += skb_shinfo(head)->frags[i].size; | |
622 | clone->len = clone->data_len = head->data_len - plen; | |
623 | head->data_len -= clone->len; | |
624 | head->len -= clone->len; | |
625 | clone->csum = 0; | |
626 | clone->ip_summed = head->ip_summed; | |
627 | ||
628 | NFCT_FRAG6_CB(clone)->orig = NULL; | |
629 | atomic_add(clone->truesize, &nf_ct_frag6_mem); | |
630 | } | |
631 | ||
632 | /* We have to remove fragment header from datagram and to relocate | |
633 | * header in order to calculate ICV correctly. */ | |
634 | head->nh.raw[fq->nhoffset] = head->h.raw[0]; | |
635 | memmove(head->head + sizeof(struct frag_hdr), head->head, | |
636 | (head->data - head->head) - sizeof(struct frag_hdr)); | |
637 | head->mac.raw += sizeof(struct frag_hdr); | |
638 | head->nh.raw += sizeof(struct frag_hdr); | |
639 | ||
640 | skb_shinfo(head)->frag_list = head->next; | |
641 | head->h.raw = head->data; | |
642 | skb_push(head, head->data - head->nh.raw); | |
643 | atomic_sub(head->truesize, &nf_ct_frag6_mem); | |
644 | ||
645 | for (fp=head->next; fp; fp = fp->next) { | |
646 | head->data_len += fp->len; | |
647 | head->len += fp->len; | |
648 | if (head->ip_summed != fp->ip_summed) | |
649 | head->ip_summed = CHECKSUM_NONE; | |
650 | else if (head->ip_summed == CHECKSUM_HW) | |
651 | head->csum = csum_add(head->csum, fp->csum); | |
652 | head->truesize += fp->truesize; | |
653 | atomic_sub(fp->truesize, &nf_ct_frag6_mem); | |
654 | } | |
655 | ||
656 | head->next = NULL; | |
657 | head->dev = dev; | |
658 | skb_set_timestamp(head, &fq->stamp); | |
659 | head->nh.ipv6h->payload_len = htons(payload_len); | |
660 | ||
661 | /* Yes, and fold redundant checksum back. 8) */ | |
662 | if (head->ip_summed == CHECKSUM_HW) | |
663 | head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum); | |
664 | ||
665 | fq->fragments = NULL; | |
666 | ||
667 | /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */ | |
668 | fp = skb_shinfo(head)->frag_list; | |
669 | if (NFCT_FRAG6_CB(fp)->orig == NULL) | |
670 | /* at above code, head skb is divided into two skbs. */ | |
671 | fp = fp->next; | |
672 | ||
673 | op = NFCT_FRAG6_CB(head)->orig; | |
674 | for (; fp; fp = fp->next) { | |
675 | struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig; | |
676 | ||
677 | op->next = orig; | |
678 | op = orig; | |
679 | NFCT_FRAG6_CB(fp)->orig = NULL; | |
680 | } | |
681 | ||
682 | return head; | |
683 | ||
684 | out_oversize: | |
685 | if (net_ratelimit()) | |
686 | printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len); | |
687 | goto out_fail; | |
688 | out_oom: | |
689 | if (net_ratelimit()) | |
690 | printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n"); | |
691 | out_fail: | |
692 | return NULL; | |
693 | } | |
694 | ||
695 | /* | |
696 | * find the header just before Fragment Header. | |
697 | * | |
698 | * if success return 0 and set ... | |
699 | * (*prevhdrp): the value of "Next Header Field" in the header | |
700 | * just before Fragment Header. | |
701 | * (*prevhoff): the offset of "Next Header Field" in the header | |
702 | * just before Fragment Header. | |
703 | * (*fhoff) : the offset of Fragment Header. | |
704 | * | |
705 | * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c | |
706 | * | |
707 | */ | |
708 | static int | |
709 | find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff) | |
710 | { | |
711 | u8 nexthdr = skb->nh.ipv6h->nexthdr; | |
712 | u8 prev_nhoff = (u8 *)&skb->nh.ipv6h->nexthdr - skb->data; | |
713 | int start = (u8 *)(skb->nh.ipv6h+1) - skb->data; | |
714 | int len = skb->len - start; | |
715 | u8 prevhdr = NEXTHDR_IPV6; | |
716 | ||
717 | while (nexthdr != NEXTHDR_FRAGMENT) { | |
718 | struct ipv6_opt_hdr hdr; | |
719 | int hdrlen; | |
720 | ||
721 | if (!ipv6_ext_hdr(nexthdr)) { | |
722 | return -1; | |
723 | } | |
724 | if (len < (int)sizeof(struct ipv6_opt_hdr)) { | |
725 | DEBUGP("too short\n"); | |
726 | return -1; | |
727 | } | |
728 | if (nexthdr == NEXTHDR_NONE) { | |
729 | DEBUGP("next header is none\n"); | |
730 | return -1; | |
731 | } | |
732 | if (skb_copy_bits(skb, start, &hdr, sizeof(hdr))) | |
733 | BUG(); | |
734 | if (nexthdr == NEXTHDR_AUTH) | |
735 | hdrlen = (hdr.hdrlen+2)<<2; | |
736 | else | |
737 | hdrlen = ipv6_optlen(&hdr); | |
738 | ||
739 | prevhdr = nexthdr; | |
740 | prev_nhoff = start; | |
741 | ||
742 | nexthdr = hdr.nexthdr; | |
743 | len -= hdrlen; | |
744 | start += hdrlen; | |
745 | } | |
746 | ||
747 | if (len < 0) | |
748 | return -1; | |
749 | ||
750 | *prevhdrp = prevhdr; | |
751 | *prevhoff = prev_nhoff; | |
752 | *fhoff = start; | |
753 | ||
754 | return 0; | |
755 | } | |
756 | ||
757 | struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb) | |
758 | { | |
759 | struct sk_buff *clone; | |
760 | struct net_device *dev = skb->dev; | |
761 | struct frag_hdr *fhdr; | |
762 | struct nf_ct_frag6_queue *fq; | |
763 | struct ipv6hdr *hdr; | |
764 | int fhoff, nhoff; | |
765 | u8 prevhdr; | |
766 | struct sk_buff *ret_skb = NULL; | |
767 | ||
768 | /* Jumbo payload inhibits frag. header */ | |
769 | if (skb->nh.ipv6h->payload_len == 0) { | |
770 | DEBUGP("payload len = 0\n"); | |
771 | return skb; | |
772 | } | |
773 | ||
774 | if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0) | |
775 | return skb; | |
776 | ||
777 | clone = skb_clone(skb, GFP_ATOMIC); | |
778 | if (clone == NULL) { | |
779 | DEBUGP("Can't clone skb\n"); | |
780 | return skb; | |
781 | } | |
782 | ||
783 | NFCT_FRAG6_CB(clone)->orig = skb; | |
784 | ||
785 | if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) { | |
786 | DEBUGP("message is too short.\n"); | |
787 | goto ret_orig; | |
788 | } | |
789 | ||
790 | clone->h.raw = clone->data + fhoff; | |
791 | hdr = clone->nh.ipv6h; | |
792 | fhdr = (struct frag_hdr *)clone->h.raw; | |
793 | ||
794 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
795 | DEBUGP("Invalid fragment offset\n"); | |
796 | /* It is not a fragmented frame */ | |
797 | goto ret_orig; | |
798 | } | |
799 | ||
800 | if (atomic_read(&nf_ct_frag6_mem) > nf_ct_frag6_high_thresh) | |
801 | nf_ct_frag6_evictor(); | |
802 | ||
803 | fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr); | |
804 | if (fq == NULL) { | |
805 | DEBUGP("Can't find and can't create new queue\n"); | |
806 | goto ret_orig; | |
807 | } | |
808 | ||
809 | spin_lock(&fq->lock); | |
810 | ||
811 | if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) { | |
812 | spin_unlock(&fq->lock); | |
813 | DEBUGP("Can't insert skb to queue\n"); | |
814 | fq_put(fq); | |
815 | goto ret_orig; | |
816 | } | |
817 | ||
818 | if (fq->last_in == (FIRST_IN|LAST_IN) && fq->meat == fq->len) { | |
819 | ret_skb = nf_ct_frag6_reasm(fq, dev); | |
820 | if (ret_skb == NULL) | |
821 | DEBUGP("Can't reassemble fragmented packets\n"); | |
822 | } | |
823 | spin_unlock(&fq->lock); | |
824 | ||
825 | fq_put(fq); | |
826 | return ret_skb; | |
827 | ||
828 | ret_orig: | |
829 | kfree_skb(clone); | |
830 | return skb; | |
831 | } | |
832 | ||
833 | void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb, | |
834 | struct net_device *in, struct net_device *out, | |
835 | int (*okfn)(struct sk_buff *)) | |
836 | { | |
837 | struct sk_buff *s, *s2; | |
838 | ||
839 | for (s = NFCT_FRAG6_CB(skb)->orig; s;) { | |
840 | nf_conntrack_put_reasm(s->nfct_reasm); | |
841 | nf_conntrack_get_reasm(skb); | |
842 | s->nfct_reasm = skb; | |
843 | ||
844 | s2 = s->next; | |
845 | NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn, | |
846 | NF_IP6_PRI_CONNTRACK_DEFRAG + 1); | |
847 | s = s2; | |
848 | } | |
849 | nf_conntrack_put_reasm(skb); | |
850 | } | |
851 | ||
852 | int nf_ct_frag6_kfree_frags(struct sk_buff *skb) | |
853 | { | |
854 | struct sk_buff *s, *s2; | |
855 | ||
856 | for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) { | |
857 | ||
858 | s2 = s->next; | |
859 | kfree_skb(s); | |
860 | } | |
861 | ||
862 | kfree_skb(skb); | |
863 | ||
864 | return 0; | |
865 | } | |
866 | ||
867 | int nf_ct_frag6_init(void) | |
868 | { | |
869 | nf_ct_frag6_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ | |
870 | (jiffies ^ (jiffies >> 6))); | |
871 | ||
872 | init_timer(&nf_ct_frag6_secret_timer); | |
873 | nf_ct_frag6_secret_timer.function = nf_ct_frag6_secret_rebuild; | |
874 | nf_ct_frag6_secret_timer.expires = jiffies | |
875 | + nf_ct_frag6_secret_interval; | |
876 | add_timer(&nf_ct_frag6_secret_timer); | |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
881 | void nf_ct_frag6_cleanup(void) | |
882 | { | |
883 | del_timer(&nf_ct_frag6_secret_timer); | |
884 | nf_ct_frag6_evictor(); | |
885 | } |