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> | |
34 | ||
35 | #include <net/ipv6.h> | |
36 | #include <net/protocol.h> | |
37 | #include <net/transp_v6.h> | |
38 | #include <net/rawv6.h> | |
39 | #include <net/ndisc.h> | |
40 | #include <net/addrconf.h> | |
41 | #include <linux/sysctl.h> | |
42 | #include <linux/netfilter.h> | |
43 | #include <linux/netfilter_ipv6.h> | |
44 | #include <linux/kernel.h> | |
45 | #include <linux/module.h> | |
46 | ||
9fb9cbb1 YK |
47 | #define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */ |
48 | #define NF_CT_FRAG6_LOW_THRESH 196608 /* == 192*1024 */ | |
49 | #define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT | |
50 | ||
94aec08e BH |
51 | unsigned int nf_ct_frag6_high_thresh __read_mostly = 256*1024; |
52 | unsigned int nf_ct_frag6_low_thresh __read_mostly = 192*1024; | |
53 | unsigned long nf_ct_frag6_timeout __read_mostly = IPV6_FRAG_TIMEOUT; | |
9fb9cbb1 YK |
54 | |
55 | struct nf_ct_frag6_skb_cb | |
56 | { | |
57 | struct inet6_skb_parm h; | |
58 | int offset; | |
59 | struct sk_buff *orig; | |
60 | }; | |
61 | ||
62 | #define NFCT_FRAG6_CB(skb) ((struct nf_ct_frag6_skb_cb*)((skb)->cb)) | |
63 | ||
64 | struct nf_ct_frag6_queue | |
65 | { | |
2e4e6a17 HW |
66 | struct hlist_node list; |
67 | struct list_head lru_list; /* lru list member */ | |
9fb9cbb1 | 68 | |
bff9a89b | 69 | __be32 id; /* fragment id */ |
9fb9cbb1 YK |
70 | struct in6_addr saddr; |
71 | struct in6_addr daddr; | |
72 | ||
73 | spinlock_t lock; | |
74 | atomic_t refcnt; | |
75 | struct timer_list timer; /* expire timer */ | |
76 | struct sk_buff *fragments; | |
77 | int len; | |
78 | int meat; | |
b7aa0bf7 | 79 | ktime_t stamp; |
9fb9cbb1 YK |
80 | unsigned int csum; |
81 | __u8 last_in; /* has first/last segment arrived? */ | |
82 | #define COMPLETE 4 | |
83 | #define FIRST_IN 2 | |
84 | #define LAST_IN 1 | |
85 | __u16 nhoffset; | |
9fb9cbb1 YK |
86 | }; |
87 | ||
88 | /* Hash table. */ | |
89 | ||
90 | #define FRAG6Q_HASHSZ 64 | |
91 | ||
2e4e6a17 | 92 | static struct hlist_head nf_ct_frag6_hash[FRAG6Q_HASHSZ]; |
181a46a5 | 93 | static DEFINE_RWLOCK(nf_ct_frag6_lock); |
9fb9cbb1 YK |
94 | static u32 nf_ct_frag6_hash_rnd; |
95 | static LIST_HEAD(nf_ct_frag6_lru_list); | |
96 | int nf_ct_frag6_nqueues = 0; | |
97 | ||
98 | static __inline__ void __fq_unlink(struct nf_ct_frag6_queue *fq) | |
99 | { | |
2e4e6a17 | 100 | hlist_del(&fq->list); |
9fb9cbb1 YK |
101 | list_del(&fq->lru_list); |
102 | nf_ct_frag6_nqueues--; | |
103 | } | |
104 | ||
105 | static __inline__ void fq_unlink(struct nf_ct_frag6_queue *fq) | |
106 | { | |
107 | write_lock(&nf_ct_frag6_lock); | |
108 | __fq_unlink(fq); | |
109 | write_unlock(&nf_ct_frag6_lock); | |
110 | } | |
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; | |
123 | c += nf_ct_frag6_hash_rnd; | |
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 | ||
136 | return c & (FRAG6Q_HASHSZ - 1); | |
137 | } | |
138 | ||
139 | static struct timer_list nf_ct_frag6_secret_timer; | |
140 | int nf_ct_frag6_secret_interval = 10 * 60 * HZ; | |
141 | ||
142 | static void nf_ct_frag6_secret_rebuild(unsigned long dummy) | |
143 | { | |
144 | unsigned long now = jiffies; | |
145 | int i; | |
146 | ||
147 | write_lock(&nf_ct_frag6_lock); | |
148 | get_random_bytes(&nf_ct_frag6_hash_rnd, sizeof(u32)); | |
149 | for (i = 0; i < FRAG6Q_HASHSZ; i++) { | |
150 | struct nf_ct_frag6_queue *q; | |
2e4e6a17 | 151 | struct hlist_node *p, *n; |
9fb9cbb1 | 152 | |
2e4e6a17 | 153 | hlist_for_each_entry_safe(q, p, n, &nf_ct_frag6_hash[i], list) { |
9fb9cbb1 YK |
154 | unsigned int hval = ip6qhashfn(q->id, |
155 | &q->saddr, | |
156 | &q->daddr); | |
9fb9cbb1 | 157 | if (hval != i) { |
2e4e6a17 | 158 | hlist_del(&q->list); |
9fb9cbb1 | 159 | /* Relink to new hash chain. */ |
2e4e6a17 HW |
160 | hlist_add_head(&q->list, |
161 | &nf_ct_frag6_hash[hval]); | |
9fb9cbb1 | 162 | } |
9fb9cbb1 YK |
163 | } |
164 | } | |
165 | write_unlock(&nf_ct_frag6_lock); | |
166 | ||
167 | mod_timer(&nf_ct_frag6_secret_timer, now + nf_ct_frag6_secret_interval); | |
168 | } | |
169 | ||
170 | atomic_t nf_ct_frag6_mem = ATOMIC_INIT(0); | |
171 | ||
172 | /* Memory Tracking Functions. */ | |
1ba430bc | 173 | static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work) |
9fb9cbb1 | 174 | { |
1ba430bc YK |
175 | if (work) |
176 | *work -= skb->truesize; | |
9fb9cbb1 YK |
177 | atomic_sub(skb->truesize, &nf_ct_frag6_mem); |
178 | if (NFCT_FRAG6_CB(skb)->orig) | |
179 | kfree_skb(NFCT_FRAG6_CB(skb)->orig); | |
180 | ||
181 | kfree_skb(skb); | |
182 | } | |
183 | ||
1ba430bc YK |
184 | static inline void frag_free_queue(struct nf_ct_frag6_queue *fq, |
185 | unsigned int *work) | |
9fb9cbb1 | 186 | { |
1ba430bc YK |
187 | if (work) |
188 | *work -= sizeof(struct nf_ct_frag6_queue); | |
9fb9cbb1 YK |
189 | atomic_sub(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem); |
190 | kfree(fq); | |
191 | } | |
192 | ||
193 | static inline struct nf_ct_frag6_queue *frag_alloc_queue(void) | |
194 | { | |
195 | struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC); | |
196 | ||
197 | if (!fq) | |
198 | return NULL; | |
199 | atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem); | |
200 | return fq; | |
201 | } | |
202 | ||
203 | /* Destruction primitives. */ | |
204 | ||
205 | /* Complete destruction of fq. */ | |
1ba430bc YK |
206 | static void nf_ct_frag6_destroy(struct nf_ct_frag6_queue *fq, |
207 | unsigned int *work) | |
9fb9cbb1 YK |
208 | { |
209 | struct sk_buff *fp; | |
210 | ||
211 | BUG_TRAP(fq->last_in&COMPLETE); | |
212 | BUG_TRAP(del_timer(&fq->timer) == 0); | |
213 | ||
214 | /* Release all fragment data. */ | |
215 | fp = fq->fragments; | |
216 | while (fp) { | |
217 | struct sk_buff *xp = fp->next; | |
218 | ||
1ba430bc | 219 | frag_kfree_skb(fp, work); |
9fb9cbb1 YK |
220 | fp = xp; |
221 | } | |
222 | ||
1ba430bc | 223 | frag_free_queue(fq, work); |
9fb9cbb1 YK |
224 | } |
225 | ||
1ba430bc | 226 | static __inline__ void fq_put(struct nf_ct_frag6_queue *fq, unsigned int *work) |
9fb9cbb1 YK |
227 | { |
228 | if (atomic_dec_and_test(&fq->refcnt)) | |
1ba430bc | 229 | nf_ct_frag6_destroy(fq, work); |
9fb9cbb1 YK |
230 | } |
231 | ||
232 | /* Kill fq entry. It is not destroyed immediately, | |
233 | * because caller (and someone more) holds reference count. | |
234 | */ | |
235 | static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq) | |
236 | { | |
237 | if (del_timer(&fq->timer)) | |
238 | atomic_dec(&fq->refcnt); | |
239 | ||
240 | if (!(fq->last_in & COMPLETE)) { | |
241 | fq_unlink(fq); | |
242 | atomic_dec(&fq->refcnt); | |
243 | fq->last_in |= COMPLETE; | |
244 | } | |
245 | } | |
246 | ||
247 | static void nf_ct_frag6_evictor(void) | |
248 | { | |
249 | struct nf_ct_frag6_queue *fq; | |
250 | struct list_head *tmp; | |
1ba430bc | 251 | unsigned int work; |
9fb9cbb1 | 252 | |
1ba430bc YK |
253 | work = atomic_read(&nf_ct_frag6_mem); |
254 | if (work <= nf_ct_frag6_low_thresh) | |
255 | return; | |
256 | ||
257 | work -= nf_ct_frag6_low_thresh; | |
258 | while (work > 0) { | |
9fb9cbb1 YK |
259 | read_lock(&nf_ct_frag6_lock); |
260 | if (list_empty(&nf_ct_frag6_lru_list)) { | |
261 | read_unlock(&nf_ct_frag6_lock); | |
262 | return; | |
263 | } | |
264 | tmp = nf_ct_frag6_lru_list.next; | |
302fe175 | 265 | BUG_ON(tmp == NULL); |
9fb9cbb1 YK |
266 | fq = list_entry(tmp, struct nf_ct_frag6_queue, lru_list); |
267 | atomic_inc(&fq->refcnt); | |
268 | read_unlock(&nf_ct_frag6_lock); | |
269 | ||
270 | spin_lock(&fq->lock); | |
271 | if (!(fq->last_in&COMPLETE)) | |
272 | fq_kill(fq); | |
273 | spin_unlock(&fq->lock); | |
274 | ||
1ba430bc | 275 | fq_put(fq, &work); |
9fb9cbb1 YK |
276 | } |
277 | } | |
278 | ||
279 | static void nf_ct_frag6_expire(unsigned long data) | |
280 | { | |
281 | struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data; | |
282 | ||
283 | spin_lock(&fq->lock); | |
284 | ||
285 | if (fq->last_in & COMPLETE) | |
286 | goto out; | |
287 | ||
288 | fq_kill(fq); | |
289 | ||
290 | out: | |
291 | spin_unlock(&fq->lock); | |
1ba430bc | 292 | fq_put(fq, NULL); |
9fb9cbb1 YK |
293 | } |
294 | ||
295 | /* Creation primitives. */ | |
296 | ||
9fb9cbb1 YK |
297 | static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash, |
298 | struct nf_ct_frag6_queue *fq_in) | |
299 | { | |
300 | struct nf_ct_frag6_queue *fq; | |
2e4e6a17 HW |
301 | #ifdef CONFIG_SMP |
302 | struct hlist_node *n; | |
303 | #endif | |
9fb9cbb1 YK |
304 | |
305 | write_lock(&nf_ct_frag6_lock); | |
306 | #ifdef CONFIG_SMP | |
2e4e6a17 | 307 | hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], list) { |
1ab1457c | 308 | if (fq->id == fq_in->id && |
6ea46c9c YK |
309 | ipv6_addr_equal(&fq_in->saddr, &fq->saddr) && |
310 | ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) { | |
9fb9cbb1 YK |
311 | atomic_inc(&fq->refcnt); |
312 | write_unlock(&nf_ct_frag6_lock); | |
313 | fq_in->last_in |= COMPLETE; | |
1ba430bc | 314 | fq_put(fq_in, NULL); |
9fb9cbb1 YK |
315 | return fq; |
316 | } | |
317 | } | |
318 | #endif | |
319 | fq = fq_in; | |
320 | ||
321 | if (!mod_timer(&fq->timer, jiffies + nf_ct_frag6_timeout)) | |
322 | atomic_inc(&fq->refcnt); | |
323 | ||
324 | atomic_inc(&fq->refcnt); | |
2e4e6a17 | 325 | hlist_add_head(&fq->list, &nf_ct_frag6_hash[hash]); |
9fb9cbb1 YK |
326 | INIT_LIST_HEAD(&fq->lru_list); |
327 | list_add_tail(&fq->lru_list, &nf_ct_frag6_lru_list); | |
328 | nf_ct_frag6_nqueues++; | |
329 | write_unlock(&nf_ct_frag6_lock); | |
330 | return fq; | |
331 | } | |
332 | ||
333 | ||
334 | static struct nf_ct_frag6_queue * | |
bff9a89b | 335 | nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src, struct in6_addr *dst) |
9fb9cbb1 YK |
336 | { |
337 | struct nf_ct_frag6_queue *fq; | |
338 | ||
339 | if ((fq = frag_alloc_queue()) == NULL) { | |
0d53778e | 340 | pr_debug("Can't alloc new queue\n"); |
9fb9cbb1 YK |
341 | goto oom; |
342 | } | |
343 | ||
344 | memset(fq, 0, sizeof(struct nf_ct_frag6_queue)); | |
345 | ||
346 | fq->id = id; | |
347 | ipv6_addr_copy(&fq->saddr, src); | |
348 | ipv6_addr_copy(&fq->daddr, dst); | |
349 | ||
e6f689db | 350 | setup_timer(&fq->timer, nf_ct_frag6_expire, (unsigned long)fq); |
181a46a5 | 351 | spin_lock_init(&fq->lock); |
9fb9cbb1 YK |
352 | atomic_set(&fq->refcnt, 1); |
353 | ||
354 | return nf_ct_frag6_intern(hash, fq); | |
355 | ||
356 | oom: | |
357 | return NULL; | |
358 | } | |
359 | ||
360 | static __inline__ struct nf_ct_frag6_queue * | |
bff9a89b | 361 | fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst) |
9fb9cbb1 YK |
362 | { |
363 | struct nf_ct_frag6_queue *fq; | |
2e4e6a17 | 364 | struct hlist_node *n; |
9fb9cbb1 YK |
365 | unsigned int hash = ip6qhashfn(id, src, dst); |
366 | ||
367 | read_lock(&nf_ct_frag6_lock); | |
2e4e6a17 | 368 | hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], list) { |
1ab1457c | 369 | if (fq->id == id && |
6ea46c9c YK |
370 | ipv6_addr_equal(src, &fq->saddr) && |
371 | ipv6_addr_equal(dst, &fq->daddr)) { | |
9fb9cbb1 YK |
372 | atomic_inc(&fq->refcnt); |
373 | read_unlock(&nf_ct_frag6_lock); | |
374 | return fq; | |
375 | } | |
376 | } | |
377 | read_unlock(&nf_ct_frag6_lock); | |
378 | ||
379 | return nf_ct_frag6_create(hash, id, src, dst); | |
380 | } | |
381 | ||
382 | ||
1ab1457c | 383 | static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb, |
9fb9cbb1 YK |
384 | struct frag_hdr *fhdr, int nhoff) |
385 | { | |
386 | struct sk_buff *prev, *next; | |
387 | int offset, end; | |
388 | ||
389 | if (fq->last_in & COMPLETE) { | |
0d53778e | 390 | pr_debug("Allready completed\n"); |
9fb9cbb1 YK |
391 | goto err; |
392 | } | |
393 | ||
394 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
0660e03f ACM |
395 | end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - |
396 | ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); | |
9fb9cbb1 YK |
397 | |
398 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
0d53778e | 399 | pr_debug("offset is too large.\n"); |
1ab1457c | 400 | return -1; |
9fb9cbb1 YK |
401 | } |
402 | ||
d56f90a7 ACM |
403 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
404 | const unsigned char *nh = skb_network_header(skb); | |
1ab1457c | 405 | skb->csum = csum_sub(skb->csum, |
d56f90a7 | 406 | csum_partial(nh, (u8 *)(fhdr + 1) - nh, |
9fb9cbb1 | 407 | 0)); |
d56f90a7 | 408 | } |
9fb9cbb1 YK |
409 | |
410 | /* Is this the final fragment? */ | |
411 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
412 | /* If we already have some bits beyond end | |
413 | * or have different end, the segment is corrupted. | |
414 | */ | |
415 | if (end < fq->len || | |
416 | ((fq->last_in & LAST_IN) && end != fq->len)) { | |
0d53778e | 417 | pr_debug("already received last fragment\n"); |
9fb9cbb1 YK |
418 | goto err; |
419 | } | |
420 | fq->last_in |= LAST_IN; | |
421 | fq->len = end; | |
422 | } else { | |
423 | /* Check if the fragment is rounded to 8 bytes. | |
424 | * Required by the RFC. | |
425 | */ | |
426 | if (end & 0x7) { | |
427 | /* RFC2460 says always send parameter problem in | |
428 | * this case. -DaveM | |
429 | */ | |
0d53778e | 430 | pr_debug("end of fragment not rounded to 8 bytes.\n"); |
9fb9cbb1 YK |
431 | return -1; |
432 | } | |
433 | if (end > fq->len) { | |
434 | /* Some bits beyond end -> corruption. */ | |
435 | if (fq->last_in & LAST_IN) { | |
0d53778e | 436 | pr_debug("last packet already reached.\n"); |
9fb9cbb1 YK |
437 | goto err; |
438 | } | |
439 | fq->len = end; | |
440 | } | |
441 | } | |
442 | ||
443 | if (end == offset) | |
444 | goto err; | |
445 | ||
446 | /* Point into the IP datagram 'data' part. */ | |
447 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) { | |
0d53778e | 448 | pr_debug("queue: message is too short.\n"); |
9fb9cbb1 YK |
449 | goto err; |
450 | } | |
b38dfee3 | 451 | if (pskb_trim_rcsum(skb, end - offset)) { |
0d53778e | 452 | pr_debug("Can't trim\n"); |
b38dfee3 | 453 | goto err; |
9fb9cbb1 YK |
454 | } |
455 | ||
456 | /* Find out which fragments are in front and at the back of us | |
457 | * in the chain of fragments so far. We must know where to put | |
458 | * this fragment, right? | |
459 | */ | |
460 | prev = NULL; | |
461 | for (next = fq->fragments; next != NULL; next = next->next) { | |
462 | if (NFCT_FRAG6_CB(next)->offset >= offset) | |
463 | break; /* bingo! */ | |
464 | prev = next; | |
465 | } | |
466 | ||
467 | /* We found where to put this one. Check for overlap with | |
468 | * preceding fragment, and, if needed, align things so that | |
469 | * any overlaps are eliminated. | |
470 | */ | |
471 | if (prev) { | |
472 | int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset; | |
473 | ||
474 | if (i > 0) { | |
475 | offset += i; | |
476 | if (end <= offset) { | |
0d53778e | 477 | pr_debug("overlap\n"); |
9fb9cbb1 YK |
478 | goto err; |
479 | } | |
480 | if (!pskb_pull(skb, i)) { | |
0d53778e | 481 | pr_debug("Can't pull\n"); |
9fb9cbb1 YK |
482 | goto err; |
483 | } | |
484 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
485 | skb->ip_summed = CHECKSUM_NONE; | |
486 | } | |
487 | } | |
488 | ||
489 | /* Look for overlap with succeeding segments. | |
490 | * If we can merge fragments, do it. | |
491 | */ | |
492 | while (next && NFCT_FRAG6_CB(next)->offset < end) { | |
493 | /* overlap is 'i' bytes */ | |
494 | int i = end - NFCT_FRAG6_CB(next)->offset; | |
495 | ||
496 | if (i < next->len) { | |
497 | /* Eat head of the next overlapped fragment | |
498 | * and leave the loop. The next ones cannot overlap. | |
499 | */ | |
0d53778e | 500 | pr_debug("Eat head of the overlapped parts.: %d", i); |
9fb9cbb1 YK |
501 | if (!pskb_pull(next, i)) |
502 | goto err; | |
503 | ||
504 | /* next fragment */ | |
505 | NFCT_FRAG6_CB(next)->offset += i; | |
506 | fq->meat -= i; | |
507 | if (next->ip_summed != CHECKSUM_UNNECESSARY) | |
508 | next->ip_summed = CHECKSUM_NONE; | |
509 | break; | |
510 | } else { | |
511 | struct sk_buff *free_it = next; | |
512 | ||
513 | /* Old fragmnet is completely overridden with | |
514 | * new one drop it. | |
515 | */ | |
516 | next = next->next; | |
517 | ||
518 | if (prev) | |
519 | prev->next = next; | |
520 | else | |
521 | fq->fragments = next; | |
522 | ||
523 | fq->meat -= free_it->len; | |
1ba430bc | 524 | frag_kfree_skb(free_it, NULL); |
9fb9cbb1 YK |
525 | } |
526 | } | |
527 | ||
528 | NFCT_FRAG6_CB(skb)->offset = offset; | |
529 | ||
530 | /* Insert this fragment in the chain of fragments. */ | |
531 | skb->next = next; | |
532 | if (prev) | |
533 | prev->next = skb; | |
534 | else | |
535 | fq->fragments = skb; | |
536 | ||
537 | skb->dev = NULL; | |
b7aa0bf7 | 538 | fq->stamp = skb->tstamp; |
9fb9cbb1 YK |
539 | fq->meat += skb->len; |
540 | atomic_add(skb->truesize, &nf_ct_frag6_mem); | |
541 | ||
542 | /* The first fragment. | |
543 | * nhoffset is obtained from the first fragment, of course. | |
544 | */ | |
545 | if (offset == 0) { | |
546 | fq->nhoffset = nhoff; | |
547 | fq->last_in |= FIRST_IN; | |
548 | } | |
549 | write_lock(&nf_ct_frag6_lock); | |
550 | list_move_tail(&fq->lru_list, &nf_ct_frag6_lru_list); | |
551 | write_unlock(&nf_ct_frag6_lock); | |
552 | return 0; | |
553 | ||
554 | err: | |
555 | return -1; | |
556 | } | |
557 | ||
558 | /* | |
559 | * Check if this packet is complete. | |
560 | * Returns NULL on failure by any reason, and pointer | |
561 | * to current nexthdr field in reassembled frame. | |
562 | * | |
563 | * It is called with locked fq, and caller must check that | |
564 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
565 | * the last and the first frames arrived and all the bits are here. | |
566 | */ | |
567 | static struct sk_buff * | |
568 | nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev) | |
569 | { | |
570 | struct sk_buff *fp, *op, *head = fq->fragments; | |
571 | int payload_len; | |
572 | ||
573 | fq_kill(fq); | |
574 | ||
575 | BUG_TRAP(head != NULL); | |
576 | BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0); | |
577 | ||
578 | /* Unfragmented part is taken from the first segment. */ | |
d56f90a7 ACM |
579 | payload_len = ((head->data - skb_network_header(head)) - |
580 | sizeof(struct ipv6hdr) + fq->len - | |
581 | sizeof(struct frag_hdr)); | |
9fb9cbb1 | 582 | if (payload_len > IPV6_MAXPLEN) { |
0d53778e | 583 | pr_debug("payload len is too large.\n"); |
9fb9cbb1 YK |
584 | goto out_oversize; |
585 | } | |
586 | ||
587 | /* Head of list must not be cloned. */ | |
588 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) { | |
0d53778e | 589 | pr_debug("skb is cloned but can't expand head"); |
9fb9cbb1 YK |
590 | goto out_oom; |
591 | } | |
592 | ||
593 | /* If the first fragment is fragmented itself, we split | |
594 | * it to two chunks: the first with data and paged part | |
595 | * and the second, holding only fragments. */ | |
596 | if (skb_shinfo(head)->frag_list) { | |
597 | struct sk_buff *clone; | |
598 | int i, plen = 0; | |
599 | ||
600 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) { | |
0d53778e | 601 | pr_debug("Can't alloc skb\n"); |
9fb9cbb1 YK |
602 | goto out_oom; |
603 | } | |
604 | clone->next = head->next; | |
605 | head->next = clone; | |
606 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
607 | skb_shinfo(head)->frag_list = NULL; | |
608 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
609 | plen += skb_shinfo(head)->frags[i].size; | |
610 | clone->len = clone->data_len = head->data_len - plen; | |
611 | head->data_len -= clone->len; | |
612 | head->len -= clone->len; | |
613 | clone->csum = 0; | |
614 | clone->ip_summed = head->ip_summed; | |
615 | ||
616 | NFCT_FRAG6_CB(clone)->orig = NULL; | |
617 | atomic_add(clone->truesize, &nf_ct_frag6_mem); | |
618 | } | |
619 | ||
620 | /* We have to remove fragment header from datagram and to relocate | |
621 | * header in order to calculate ICV correctly. */ | |
bff9b61c | 622 | skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0]; |
1ab1457c | 623 | memmove(head->head + sizeof(struct frag_hdr), head->head, |
9fb9cbb1 | 624 | (head->data - head->head) - sizeof(struct frag_hdr)); |
b0e380b1 ACM |
625 | head->mac_header += sizeof(struct frag_hdr); |
626 | head->network_header += sizeof(struct frag_hdr); | |
9fb9cbb1 YK |
627 | |
628 | skb_shinfo(head)->frag_list = head->next; | |
badff6d0 | 629 | skb_reset_transport_header(head); |
d56f90a7 | 630 | skb_push(head, head->data - skb_network_header(head)); |
9fb9cbb1 YK |
631 | atomic_sub(head->truesize, &nf_ct_frag6_mem); |
632 | ||
633 | for (fp=head->next; fp; fp = fp->next) { | |
634 | head->data_len += fp->len; | |
635 | head->len += fp->len; | |
636 | if (head->ip_summed != fp->ip_summed) | |
637 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 638 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
9fb9cbb1 YK |
639 | head->csum = csum_add(head->csum, fp->csum); |
640 | head->truesize += fp->truesize; | |
641 | atomic_sub(fp->truesize, &nf_ct_frag6_mem); | |
642 | } | |
643 | ||
644 | head->next = NULL; | |
645 | head->dev = dev; | |
b7aa0bf7 | 646 | head->tstamp = fq->stamp; |
0660e03f | 647 | ipv6_hdr(head)->payload_len = htons(payload_len); |
9fb9cbb1 YK |
648 | |
649 | /* Yes, and fold redundant checksum back. 8) */ | |
84fa7933 | 650 | if (head->ip_summed == CHECKSUM_COMPLETE) |
d56f90a7 | 651 | head->csum = csum_partial(skb_network_header(head), |
cfe1fc77 | 652 | skb_network_header_len(head), |
d56f90a7 | 653 | head->csum); |
9fb9cbb1 YK |
654 | |
655 | fq->fragments = NULL; | |
656 | ||
657 | /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */ | |
658 | fp = skb_shinfo(head)->frag_list; | |
659 | if (NFCT_FRAG6_CB(fp)->orig == NULL) | |
660 | /* at above code, head skb is divided into two skbs. */ | |
661 | fp = fp->next; | |
662 | ||
663 | op = NFCT_FRAG6_CB(head)->orig; | |
664 | for (; fp; fp = fp->next) { | |
665 | struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig; | |
666 | ||
667 | op->next = orig; | |
668 | op = orig; | |
669 | NFCT_FRAG6_CB(fp)->orig = NULL; | |
670 | } | |
671 | ||
672 | return head; | |
673 | ||
674 | out_oversize: | |
675 | if (net_ratelimit()) | |
676 | printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len); | |
677 | goto out_fail; | |
678 | out_oom: | |
679 | if (net_ratelimit()) | |
680 | printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n"); | |
681 | out_fail: | |
682 | return NULL; | |
683 | } | |
684 | ||
685 | /* | |
686 | * find the header just before Fragment Header. | |
687 | * | |
688 | * if success return 0 and set ... | |
689 | * (*prevhdrp): the value of "Next Header Field" in the header | |
690 | * just before Fragment Header. | |
691 | * (*prevhoff): the offset of "Next Header Field" in the header | |
692 | * just before Fragment Header. | |
693 | * (*fhoff) : the offset of Fragment Header. | |
694 | * | |
695 | * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c | |
696 | * | |
697 | */ | |
698 | static int | |
699 | find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff) | |
700 | { | |
0660e03f | 701 | u8 nexthdr = ipv6_hdr(skb)->nexthdr; |
6b88dd96 ACM |
702 | const int netoff = skb_network_offset(skb); |
703 | u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr); | |
704 | int start = netoff + sizeof(struct ipv6hdr); | |
9fb9cbb1 YK |
705 | int len = skb->len - start; |
706 | u8 prevhdr = NEXTHDR_IPV6; | |
707 | ||
1ab1457c YH |
708 | while (nexthdr != NEXTHDR_FRAGMENT) { |
709 | struct ipv6_opt_hdr hdr; | |
710 | int hdrlen; | |
9fb9cbb1 YK |
711 | |
712 | if (!ipv6_ext_hdr(nexthdr)) { | |
713 | return -1; | |
714 | } | |
1ab1457c | 715 | if (len < (int)sizeof(struct ipv6_opt_hdr)) { |
0d53778e | 716 | pr_debug("too short\n"); |
9fb9cbb1 YK |
717 | return -1; |
718 | } | |
1ab1457c | 719 | if (nexthdr == NEXTHDR_NONE) { |
0d53778e | 720 | pr_debug("next header is none\n"); |
9fb9cbb1 YK |
721 | return -1; |
722 | } | |
1ab1457c YH |
723 | if (skb_copy_bits(skb, start, &hdr, sizeof(hdr))) |
724 | BUG(); | |
725 | if (nexthdr == NEXTHDR_AUTH) | |
726 | hdrlen = (hdr.hdrlen+2)<<2; | |
727 | else | |
728 | hdrlen = ipv6_optlen(&hdr); | |
9fb9cbb1 YK |
729 | |
730 | prevhdr = nexthdr; | |
731 | prev_nhoff = start; | |
732 | ||
1ab1457c YH |
733 | nexthdr = hdr.nexthdr; |
734 | len -= hdrlen; | |
735 | start += hdrlen; | |
736 | } | |
9fb9cbb1 YK |
737 | |
738 | if (len < 0) | |
739 | return -1; | |
740 | ||
741 | *prevhdrp = prevhdr; | |
742 | *prevhoff = prev_nhoff; | |
743 | *fhoff = start; | |
744 | ||
745 | return 0; | |
746 | } | |
747 | ||
748 | struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb) | |
749 | { | |
1ab1457c | 750 | struct sk_buff *clone; |
9fb9cbb1 YK |
751 | struct net_device *dev = skb->dev; |
752 | struct frag_hdr *fhdr; | |
753 | struct nf_ct_frag6_queue *fq; | |
754 | struct ipv6hdr *hdr; | |
755 | int fhoff, nhoff; | |
756 | u8 prevhdr; | |
757 | struct sk_buff *ret_skb = NULL; | |
758 | ||
759 | /* Jumbo payload inhibits frag. header */ | |
0660e03f | 760 | if (ipv6_hdr(skb)->payload_len == 0) { |
0d53778e | 761 | pr_debug("payload len = 0\n"); |
9fb9cbb1 YK |
762 | return skb; |
763 | } | |
764 | ||
765 | if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0) | |
766 | return skb; | |
767 | ||
768 | clone = skb_clone(skb, GFP_ATOMIC); | |
769 | if (clone == NULL) { | |
0d53778e | 770 | pr_debug("Can't clone skb\n"); |
9fb9cbb1 YK |
771 | return skb; |
772 | } | |
773 | ||
774 | NFCT_FRAG6_CB(clone)->orig = skb; | |
775 | ||
776 | if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) { | |
0d53778e | 777 | pr_debug("message is too short.\n"); |
9fb9cbb1 YK |
778 | goto ret_orig; |
779 | } | |
780 | ||
967b05f6 | 781 | skb_set_transport_header(clone, fhoff); |
0660e03f | 782 | hdr = ipv6_hdr(clone); |
bff9b61c | 783 | fhdr = (struct frag_hdr *)skb_transport_header(clone); |
9fb9cbb1 YK |
784 | |
785 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
0d53778e | 786 | pr_debug("Invalid fragment offset\n"); |
9fb9cbb1 YK |
787 | /* It is not a fragmented frame */ |
788 | goto ret_orig; | |
789 | } | |
790 | ||
791 | if (atomic_read(&nf_ct_frag6_mem) > nf_ct_frag6_high_thresh) | |
792 | nf_ct_frag6_evictor(); | |
793 | ||
794 | fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr); | |
795 | if (fq == NULL) { | |
0d53778e | 796 | pr_debug("Can't find and can't create new queue\n"); |
9fb9cbb1 YK |
797 | goto ret_orig; |
798 | } | |
799 | ||
800 | spin_lock(&fq->lock); | |
801 | ||
802 | if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) { | |
803 | spin_unlock(&fq->lock); | |
0d53778e | 804 | pr_debug("Can't insert skb to queue\n"); |
1ba430bc | 805 | fq_put(fq, NULL); |
9fb9cbb1 YK |
806 | goto ret_orig; |
807 | } | |
808 | ||
809 | if (fq->last_in == (FIRST_IN|LAST_IN) && fq->meat == fq->len) { | |
810 | ret_skb = nf_ct_frag6_reasm(fq, dev); | |
811 | if (ret_skb == NULL) | |
0d53778e | 812 | pr_debug("Can't reassemble fragmented packets\n"); |
9fb9cbb1 YK |
813 | } |
814 | spin_unlock(&fq->lock); | |
815 | ||
1ba430bc | 816 | fq_put(fq, NULL); |
9fb9cbb1 YK |
817 | return ret_skb; |
818 | ||
819 | ret_orig: | |
820 | kfree_skb(clone); | |
821 | return skb; | |
822 | } | |
823 | ||
824 | void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb, | |
825 | struct net_device *in, struct net_device *out, | |
826 | int (*okfn)(struct sk_buff *)) | |
827 | { | |
828 | struct sk_buff *s, *s2; | |
829 | ||
830 | for (s = NFCT_FRAG6_CB(skb)->orig; s;) { | |
831 | nf_conntrack_put_reasm(s->nfct_reasm); | |
832 | nf_conntrack_get_reasm(skb); | |
833 | s->nfct_reasm = skb; | |
834 | ||
835 | s2 = s->next; | |
f9f02cca PM |
836 | s->next = NULL; |
837 | ||
9fb9cbb1 YK |
838 | NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn, |
839 | NF_IP6_PRI_CONNTRACK_DEFRAG + 1); | |
840 | s = s2; | |
841 | } | |
842 | nf_conntrack_put_reasm(skb); | |
843 | } | |
844 | ||
845 | int nf_ct_frag6_kfree_frags(struct sk_buff *skb) | |
846 | { | |
847 | struct sk_buff *s, *s2; | |
848 | ||
849 | for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) { | |
850 | ||
851 | s2 = s->next; | |
852 | kfree_skb(s); | |
853 | } | |
854 | ||
855 | kfree_skb(skb); | |
856 | ||
857 | return 0; | |
858 | } | |
859 | ||
860 | int nf_ct_frag6_init(void) | |
861 | { | |
862 | nf_ct_frag6_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ | |
863 | (jiffies ^ (jiffies >> 6))); | |
864 | ||
e6f689db | 865 | setup_timer(&nf_ct_frag6_secret_timer, nf_ct_frag6_secret_rebuild, 0); |
9fb9cbb1 YK |
866 | nf_ct_frag6_secret_timer.expires = jiffies |
867 | + nf_ct_frag6_secret_interval; | |
868 | add_timer(&nf_ct_frag6_secret_timer); | |
869 | ||
870 | return 0; | |
871 | } | |
872 | ||
873 | void nf_ct_frag6_cleanup(void) | |
874 | { | |
875 | del_timer(&nf_ct_frag6_secret_timer); | |
302fe175 | 876 | nf_ct_frag6_low_thresh = 0; |
9fb9cbb1 YK |
877 | nf_ct_frag6_evictor(); |
878 | } |