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 | ||
04128f23 PE |
73 | struct 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 | 80 | static struct inet_frags nf_frags; |
9fb9cbb1 | 81 | |
bff9a89b | 82 | static 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 | 109 | static 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 | ||
9fb9cbb1 | 117 | /* Memory Tracking Functions. */ |
1ba430bc | 118 | static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work) |
9fb9cbb1 | 119 | { |
1ba430bc YK |
120 | if (work) |
121 | *work -= skb->truesize; | |
7eb95156 | 122 | atomic_sub(skb->truesize, &nf_frags.mem); |
9fb9cbb1 YK |
123 | if (NFCT_FRAG6_CB(skb)->orig) |
124 | kfree_skb(NFCT_FRAG6_CB(skb)->orig); | |
125 | ||
126 | kfree_skb(skb); | |
127 | } | |
128 | ||
1ba430bc YK |
129 | static inline void frag_free_queue(struct nf_ct_frag6_queue *fq, |
130 | unsigned int *work) | |
9fb9cbb1 | 131 | { |
1ba430bc YK |
132 | if (work) |
133 | *work -= sizeof(struct nf_ct_frag6_queue); | |
7eb95156 | 134 | atomic_sub(sizeof(struct nf_ct_frag6_queue), &nf_frags.mem); |
9fb9cbb1 YK |
135 | kfree(fq); |
136 | } | |
137 | ||
138 | static 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 | ||
150 | /* Complete destruction of fq. */ | |
1ba430bc YK |
151 | static void nf_ct_frag6_destroy(struct nf_ct_frag6_queue *fq, |
152 | unsigned int *work) | |
9fb9cbb1 YK |
153 | { |
154 | struct sk_buff *fp; | |
155 | ||
5ab11c98 PE |
156 | BUG_TRAP(fq->q.last_in&COMPLETE); |
157 | BUG_TRAP(del_timer(&fq->q.timer) == 0); | |
9fb9cbb1 YK |
158 | |
159 | /* Release all fragment data. */ | |
5ab11c98 | 160 | fp = fq->q.fragments; |
9fb9cbb1 YK |
161 | while (fp) { |
162 | struct sk_buff *xp = fp->next; | |
163 | ||
1ba430bc | 164 | frag_kfree_skb(fp, work); |
9fb9cbb1 YK |
165 | fp = xp; |
166 | } | |
167 | ||
1ba430bc | 168 | frag_free_queue(fq, work); |
9fb9cbb1 YK |
169 | } |
170 | ||
1ba430bc | 171 | static __inline__ void fq_put(struct nf_ct_frag6_queue *fq, unsigned int *work) |
9fb9cbb1 | 172 | { |
5ab11c98 | 173 | if (atomic_dec_and_test(&fq->q.refcnt)) |
1ba430bc | 174 | nf_ct_frag6_destroy(fq, work); |
9fb9cbb1 YK |
175 | } |
176 | ||
177 | /* Kill fq entry. It is not destroyed immediately, | |
178 | * because caller (and someone more) holds reference count. | |
179 | */ | |
180 | static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq) | |
181 | { | |
277e650d | 182 | inet_frag_kill(&fq->q, &nf_frags); |
9fb9cbb1 YK |
183 | } |
184 | ||
185 | static void nf_ct_frag6_evictor(void) | |
186 | { | |
187 | struct nf_ct_frag6_queue *fq; | |
188 | struct list_head *tmp; | |
1ba430bc | 189 | unsigned int work; |
9fb9cbb1 | 190 | |
7eb95156 | 191 | work = atomic_read(&nf_frags.mem); |
04128f23 | 192 | if (work <= nf_frags_ctl.low_thresh) |
1ba430bc YK |
193 | return; |
194 | ||
04128f23 | 195 | work -= nf_frags_ctl.low_thresh; |
1ba430bc | 196 | while (work > 0) { |
7eb95156 PE |
197 | read_lock(&nf_frags.lock); |
198 | if (list_empty(&nf_frags.lru_list)) { | |
199 | read_unlock(&nf_frags.lock); | |
9fb9cbb1 YK |
200 | return; |
201 | } | |
7eb95156 | 202 | tmp = nf_frags.lru_list.next; |
302fe175 | 203 | BUG_ON(tmp == NULL); |
5ab11c98 PE |
204 | fq = list_entry(tmp, struct nf_ct_frag6_queue, q.lru_list); |
205 | atomic_inc(&fq->q.refcnt); | |
7eb95156 | 206 | read_unlock(&nf_frags.lock); |
9fb9cbb1 | 207 | |
5ab11c98 PE |
208 | spin_lock(&fq->q.lock); |
209 | if (!(fq->q.last_in&COMPLETE)) | |
9fb9cbb1 | 210 | fq_kill(fq); |
5ab11c98 | 211 | spin_unlock(&fq->q.lock); |
9fb9cbb1 | 212 | |
1ba430bc | 213 | fq_put(fq, &work); |
9fb9cbb1 YK |
214 | } |
215 | } | |
216 | ||
217 | static void nf_ct_frag6_expire(unsigned long data) | |
218 | { | |
219 | struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data; | |
220 | ||
5ab11c98 | 221 | spin_lock(&fq->q.lock); |
9fb9cbb1 | 222 | |
5ab11c98 | 223 | if (fq->q.last_in & COMPLETE) |
9fb9cbb1 YK |
224 | goto out; |
225 | ||
226 | fq_kill(fq); | |
227 | ||
228 | out: | |
5ab11c98 | 229 | spin_unlock(&fq->q.lock); |
1ba430bc | 230 | fq_put(fq, NULL); |
9fb9cbb1 YK |
231 | } |
232 | ||
233 | /* Creation primitives. */ | |
234 | ||
9fb9cbb1 YK |
235 | static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash, |
236 | struct nf_ct_frag6_queue *fq_in) | |
237 | { | |
238 | struct nf_ct_frag6_queue *fq; | |
2e4e6a17 HW |
239 | #ifdef CONFIG_SMP |
240 | struct hlist_node *n; | |
241 | #endif | |
9fb9cbb1 | 242 | |
7eb95156 | 243 | write_lock(&nf_frags.lock); |
9fb9cbb1 | 244 | #ifdef CONFIG_SMP |
7eb95156 | 245 | hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) { |
1ab1457c | 246 | if (fq->id == fq_in->id && |
6ea46c9c YK |
247 | ipv6_addr_equal(&fq_in->saddr, &fq->saddr) && |
248 | ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) { | |
5ab11c98 | 249 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 250 | write_unlock(&nf_frags.lock); |
5ab11c98 | 251 | fq_in->q.last_in |= COMPLETE; |
1ba430bc | 252 | fq_put(fq_in, NULL); |
9fb9cbb1 YK |
253 | return fq; |
254 | } | |
255 | } | |
256 | #endif | |
257 | fq = fq_in; | |
258 | ||
04128f23 | 259 | if (!mod_timer(&fq->q.timer, jiffies + nf_frags_ctl.timeout)) |
5ab11c98 | 260 | atomic_inc(&fq->q.refcnt); |
9fb9cbb1 | 261 | |
5ab11c98 | 262 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 263 | hlist_add_head(&fq->q.list, &nf_frags.hash[hash]); |
5ab11c98 | 264 | INIT_LIST_HEAD(&fq->q.lru_list); |
7eb95156 PE |
265 | list_add_tail(&fq->q.lru_list, &nf_frags.lru_list); |
266 | nf_frags.nqueues++; | |
267 | write_unlock(&nf_frags.lock); | |
9fb9cbb1 YK |
268 | return fq; |
269 | } | |
270 | ||
271 | ||
272 | static struct nf_ct_frag6_queue * | |
bff9a89b | 273 | nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src, struct in6_addr *dst) |
9fb9cbb1 YK |
274 | { |
275 | struct nf_ct_frag6_queue *fq; | |
276 | ||
277 | if ((fq = frag_alloc_queue()) == NULL) { | |
0d53778e | 278 | pr_debug("Can't alloc new queue\n"); |
9fb9cbb1 YK |
279 | goto oom; |
280 | } | |
281 | ||
282 | memset(fq, 0, sizeof(struct nf_ct_frag6_queue)); | |
283 | ||
284 | fq->id = id; | |
285 | ipv6_addr_copy(&fq->saddr, src); | |
286 | ipv6_addr_copy(&fq->daddr, dst); | |
287 | ||
5ab11c98 PE |
288 | setup_timer(&fq->q.timer, nf_ct_frag6_expire, (unsigned long)fq); |
289 | spin_lock_init(&fq->q.lock); | |
290 | atomic_set(&fq->q.refcnt, 1); | |
9fb9cbb1 YK |
291 | |
292 | return nf_ct_frag6_intern(hash, fq); | |
293 | ||
294 | oom: | |
295 | return NULL; | |
296 | } | |
297 | ||
298 | static __inline__ struct nf_ct_frag6_queue * | |
bff9a89b | 299 | fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst) |
9fb9cbb1 YK |
300 | { |
301 | struct nf_ct_frag6_queue *fq; | |
2e4e6a17 | 302 | struct hlist_node *n; |
9fb9cbb1 YK |
303 | unsigned int hash = ip6qhashfn(id, src, dst); |
304 | ||
7eb95156 PE |
305 | read_lock(&nf_frags.lock); |
306 | hlist_for_each_entry(fq, n, &nf_frags.hash[hash], q.list) { | |
1ab1457c | 307 | if (fq->id == id && |
6ea46c9c YK |
308 | ipv6_addr_equal(src, &fq->saddr) && |
309 | ipv6_addr_equal(dst, &fq->daddr)) { | |
5ab11c98 | 310 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 311 | read_unlock(&nf_frags.lock); |
9fb9cbb1 YK |
312 | return fq; |
313 | } | |
314 | } | |
7eb95156 | 315 | read_unlock(&nf_frags.lock); |
9fb9cbb1 YK |
316 | |
317 | return nf_ct_frag6_create(hash, id, src, dst); | |
318 | } | |
319 | ||
320 | ||
1ab1457c | 321 | static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb, |
9fb9cbb1 YK |
322 | struct frag_hdr *fhdr, int nhoff) |
323 | { | |
324 | struct sk_buff *prev, *next; | |
325 | int offset, end; | |
326 | ||
5ab11c98 | 327 | if (fq->q.last_in & COMPLETE) { |
0d53778e | 328 | pr_debug("Allready completed\n"); |
9fb9cbb1 YK |
329 | goto err; |
330 | } | |
331 | ||
332 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
0660e03f ACM |
333 | end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - |
334 | ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); | |
9fb9cbb1 YK |
335 | |
336 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
0d53778e | 337 | pr_debug("offset is too large.\n"); |
1ab1457c | 338 | return -1; |
9fb9cbb1 YK |
339 | } |
340 | ||
d56f90a7 ACM |
341 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
342 | const unsigned char *nh = skb_network_header(skb); | |
1ab1457c | 343 | skb->csum = csum_sub(skb->csum, |
d56f90a7 | 344 | csum_partial(nh, (u8 *)(fhdr + 1) - nh, |
9fb9cbb1 | 345 | 0)); |
d56f90a7 | 346 | } |
9fb9cbb1 YK |
347 | |
348 | /* Is this the final fragment? */ | |
349 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
350 | /* If we already have some bits beyond end | |
351 | * or have different end, the segment is corrupted. | |
352 | */ | |
5ab11c98 PE |
353 | if (end < fq->q.len || |
354 | ((fq->q.last_in & LAST_IN) && end != fq->q.len)) { | |
0d53778e | 355 | pr_debug("already received last fragment\n"); |
9fb9cbb1 YK |
356 | goto err; |
357 | } | |
5ab11c98 PE |
358 | fq->q.last_in |= LAST_IN; |
359 | fq->q.len = end; | |
9fb9cbb1 YK |
360 | } else { |
361 | /* Check if the fragment is rounded to 8 bytes. | |
362 | * Required by the RFC. | |
363 | */ | |
364 | if (end & 0x7) { | |
365 | /* RFC2460 says always send parameter problem in | |
366 | * this case. -DaveM | |
367 | */ | |
0d53778e | 368 | pr_debug("end of fragment not rounded to 8 bytes.\n"); |
9fb9cbb1 YK |
369 | return -1; |
370 | } | |
5ab11c98 | 371 | if (end > fq->q.len) { |
9fb9cbb1 | 372 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 373 | if (fq->q.last_in & LAST_IN) { |
0d53778e | 374 | pr_debug("last packet already reached.\n"); |
9fb9cbb1 YK |
375 | goto err; |
376 | } | |
5ab11c98 | 377 | fq->q.len = end; |
9fb9cbb1 YK |
378 | } |
379 | } | |
380 | ||
381 | if (end == offset) | |
382 | goto err; | |
383 | ||
384 | /* Point into the IP datagram 'data' part. */ | |
385 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) { | |
0d53778e | 386 | pr_debug("queue: message is too short.\n"); |
9fb9cbb1 YK |
387 | goto err; |
388 | } | |
b38dfee3 | 389 | if (pskb_trim_rcsum(skb, end - offset)) { |
0d53778e | 390 | pr_debug("Can't trim\n"); |
b38dfee3 | 391 | goto err; |
9fb9cbb1 YK |
392 | } |
393 | ||
394 | /* Find out which fragments are in front and at the back of us | |
395 | * in the chain of fragments so far. We must know where to put | |
396 | * this fragment, right? | |
397 | */ | |
398 | prev = NULL; | |
5ab11c98 | 399 | for (next = fq->q.fragments; next != NULL; next = next->next) { |
9fb9cbb1 YK |
400 | if (NFCT_FRAG6_CB(next)->offset >= offset) |
401 | break; /* bingo! */ | |
402 | prev = next; | |
403 | } | |
404 | ||
405 | /* We found where to put this one. Check for overlap with | |
406 | * preceding fragment, and, if needed, align things so that | |
407 | * any overlaps are eliminated. | |
408 | */ | |
409 | if (prev) { | |
410 | int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset; | |
411 | ||
412 | if (i > 0) { | |
413 | offset += i; | |
414 | if (end <= offset) { | |
0d53778e | 415 | pr_debug("overlap\n"); |
9fb9cbb1 YK |
416 | goto err; |
417 | } | |
418 | if (!pskb_pull(skb, i)) { | |
0d53778e | 419 | pr_debug("Can't pull\n"); |
9fb9cbb1 YK |
420 | goto err; |
421 | } | |
422 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
423 | skb->ip_summed = CHECKSUM_NONE; | |
424 | } | |
425 | } | |
426 | ||
427 | /* Look for overlap with succeeding segments. | |
428 | * If we can merge fragments, do it. | |
429 | */ | |
430 | while (next && NFCT_FRAG6_CB(next)->offset < end) { | |
431 | /* overlap is 'i' bytes */ | |
432 | int i = end - NFCT_FRAG6_CB(next)->offset; | |
433 | ||
434 | if (i < next->len) { | |
435 | /* Eat head of the next overlapped fragment | |
436 | * and leave the loop. The next ones cannot overlap. | |
437 | */ | |
0d53778e | 438 | pr_debug("Eat head of the overlapped parts.: %d", i); |
9fb9cbb1 YK |
439 | if (!pskb_pull(next, i)) |
440 | goto err; | |
441 | ||
442 | /* next fragment */ | |
443 | NFCT_FRAG6_CB(next)->offset += i; | |
5ab11c98 | 444 | fq->q.meat -= i; |
9fb9cbb1 YK |
445 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
446 | next->ip_summed = CHECKSUM_NONE; | |
447 | break; | |
448 | } else { | |
449 | struct sk_buff *free_it = next; | |
450 | ||
451 | /* Old fragmnet is completely overridden with | |
452 | * new one drop it. | |
453 | */ | |
454 | next = next->next; | |
455 | ||
456 | if (prev) | |
457 | prev->next = next; | |
458 | else | |
5ab11c98 | 459 | fq->q.fragments = next; |
9fb9cbb1 | 460 | |
5ab11c98 | 461 | fq->q.meat -= free_it->len; |
1ba430bc | 462 | frag_kfree_skb(free_it, NULL); |
9fb9cbb1 YK |
463 | } |
464 | } | |
465 | ||
466 | NFCT_FRAG6_CB(skb)->offset = offset; | |
467 | ||
468 | /* Insert this fragment in the chain of fragments. */ | |
469 | skb->next = next; | |
470 | if (prev) | |
471 | prev->next = skb; | |
472 | else | |
5ab11c98 | 473 | fq->q.fragments = skb; |
9fb9cbb1 YK |
474 | |
475 | skb->dev = NULL; | |
5ab11c98 PE |
476 | fq->q.stamp = skb->tstamp; |
477 | fq->q.meat += skb->len; | |
7eb95156 | 478 | atomic_add(skb->truesize, &nf_frags.mem); |
9fb9cbb1 YK |
479 | |
480 | /* The first fragment. | |
481 | * nhoffset is obtained from the first fragment, of course. | |
482 | */ | |
483 | if (offset == 0) { | |
484 | fq->nhoffset = nhoff; | |
5ab11c98 | 485 | fq->q.last_in |= FIRST_IN; |
9fb9cbb1 | 486 | } |
7eb95156 PE |
487 | write_lock(&nf_frags.lock); |
488 | list_move_tail(&fq->q.lru_list, &nf_frags.lru_list); | |
489 | write_unlock(&nf_frags.lock); | |
9fb9cbb1 YK |
490 | return 0; |
491 | ||
492 | err: | |
493 | return -1; | |
494 | } | |
495 | ||
496 | /* | |
497 | * Check if this packet is complete. | |
498 | * Returns NULL on failure by any reason, and pointer | |
499 | * to current nexthdr field in reassembled frame. | |
500 | * | |
501 | * It is called with locked fq, and caller must check that | |
502 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
503 | * the last and the first frames arrived and all the bits are here. | |
504 | */ | |
505 | static struct sk_buff * | |
506 | nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev) | |
507 | { | |
5ab11c98 | 508 | struct sk_buff *fp, *op, *head = fq->q.fragments; |
9fb9cbb1 YK |
509 | int payload_len; |
510 | ||
511 | fq_kill(fq); | |
512 | ||
513 | BUG_TRAP(head != NULL); | |
514 | BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0); | |
515 | ||
516 | /* Unfragmented part is taken from the first segment. */ | |
d56f90a7 | 517 | payload_len = ((head->data - skb_network_header(head)) - |
5ab11c98 | 518 | sizeof(struct ipv6hdr) + fq->q.len - |
d56f90a7 | 519 | sizeof(struct frag_hdr)); |
9fb9cbb1 | 520 | if (payload_len > IPV6_MAXPLEN) { |
0d53778e | 521 | pr_debug("payload len is too large.\n"); |
9fb9cbb1 YK |
522 | goto out_oversize; |
523 | } | |
524 | ||
525 | /* Head of list must not be cloned. */ | |
526 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) { | |
0d53778e | 527 | pr_debug("skb is cloned but can't expand head"); |
9fb9cbb1 YK |
528 | goto out_oom; |
529 | } | |
530 | ||
531 | /* If the first fragment is fragmented itself, we split | |
532 | * it to two chunks: the first with data and paged part | |
533 | * and the second, holding only fragments. */ | |
534 | if (skb_shinfo(head)->frag_list) { | |
535 | struct sk_buff *clone; | |
536 | int i, plen = 0; | |
537 | ||
538 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) { | |
0d53778e | 539 | pr_debug("Can't alloc skb\n"); |
9fb9cbb1 YK |
540 | goto out_oom; |
541 | } | |
542 | clone->next = head->next; | |
543 | head->next = clone; | |
544 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
545 | skb_shinfo(head)->frag_list = NULL; | |
546 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
547 | plen += skb_shinfo(head)->frags[i].size; | |
548 | clone->len = clone->data_len = head->data_len - plen; | |
549 | head->data_len -= clone->len; | |
550 | head->len -= clone->len; | |
551 | clone->csum = 0; | |
552 | clone->ip_summed = head->ip_summed; | |
553 | ||
554 | NFCT_FRAG6_CB(clone)->orig = NULL; | |
7eb95156 | 555 | atomic_add(clone->truesize, &nf_frags.mem); |
9fb9cbb1 YK |
556 | } |
557 | ||
558 | /* We have to remove fragment header from datagram and to relocate | |
559 | * header in order to calculate ICV correctly. */ | |
bff9b61c | 560 | skb_network_header(head)[fq->nhoffset] = skb_transport_header(head)[0]; |
1ab1457c | 561 | memmove(head->head + sizeof(struct frag_hdr), head->head, |
9fb9cbb1 | 562 | (head->data - head->head) - sizeof(struct frag_hdr)); |
b0e380b1 ACM |
563 | head->mac_header += sizeof(struct frag_hdr); |
564 | head->network_header += sizeof(struct frag_hdr); | |
9fb9cbb1 YK |
565 | |
566 | skb_shinfo(head)->frag_list = head->next; | |
badff6d0 | 567 | skb_reset_transport_header(head); |
d56f90a7 | 568 | skb_push(head, head->data - skb_network_header(head)); |
7eb95156 | 569 | atomic_sub(head->truesize, &nf_frags.mem); |
9fb9cbb1 YK |
570 | |
571 | for (fp=head->next; fp; fp = fp->next) { | |
572 | head->data_len += fp->len; | |
573 | head->len += fp->len; | |
574 | if (head->ip_summed != fp->ip_summed) | |
575 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 576 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
9fb9cbb1 YK |
577 | head->csum = csum_add(head->csum, fp->csum); |
578 | head->truesize += fp->truesize; | |
7eb95156 | 579 | atomic_sub(fp->truesize, &nf_frags.mem); |
9fb9cbb1 YK |
580 | } |
581 | ||
582 | head->next = NULL; | |
583 | head->dev = dev; | |
5ab11c98 | 584 | head->tstamp = fq->q.stamp; |
0660e03f | 585 | ipv6_hdr(head)->payload_len = htons(payload_len); |
9fb9cbb1 YK |
586 | |
587 | /* Yes, and fold redundant checksum back. 8) */ | |
84fa7933 | 588 | if (head->ip_summed == CHECKSUM_COMPLETE) |
d56f90a7 | 589 | head->csum = csum_partial(skb_network_header(head), |
cfe1fc77 | 590 | skb_network_header_len(head), |
d56f90a7 | 591 | head->csum); |
9fb9cbb1 | 592 | |
5ab11c98 | 593 | fq->q.fragments = NULL; |
9fb9cbb1 YK |
594 | |
595 | /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */ | |
596 | fp = skb_shinfo(head)->frag_list; | |
597 | if (NFCT_FRAG6_CB(fp)->orig == NULL) | |
598 | /* at above code, head skb is divided into two skbs. */ | |
599 | fp = fp->next; | |
600 | ||
601 | op = NFCT_FRAG6_CB(head)->orig; | |
602 | for (; fp; fp = fp->next) { | |
603 | struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig; | |
604 | ||
605 | op->next = orig; | |
606 | op = orig; | |
607 | NFCT_FRAG6_CB(fp)->orig = NULL; | |
608 | } | |
609 | ||
610 | return head; | |
611 | ||
612 | out_oversize: | |
613 | if (net_ratelimit()) | |
614 | printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len); | |
615 | goto out_fail; | |
616 | out_oom: | |
617 | if (net_ratelimit()) | |
618 | printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n"); | |
619 | out_fail: | |
620 | return NULL; | |
621 | } | |
622 | ||
623 | /* | |
624 | * find the header just before Fragment Header. | |
625 | * | |
626 | * if success return 0 and set ... | |
627 | * (*prevhdrp): the value of "Next Header Field" in the header | |
628 | * just before Fragment Header. | |
629 | * (*prevhoff): the offset of "Next Header Field" in the header | |
630 | * just before Fragment Header. | |
631 | * (*fhoff) : the offset of Fragment Header. | |
632 | * | |
633 | * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c | |
634 | * | |
635 | */ | |
636 | static int | |
637 | find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff) | |
638 | { | |
0660e03f | 639 | u8 nexthdr = ipv6_hdr(skb)->nexthdr; |
6b88dd96 ACM |
640 | const int netoff = skb_network_offset(skb); |
641 | u8 prev_nhoff = netoff + offsetof(struct ipv6hdr, nexthdr); | |
642 | int start = netoff + sizeof(struct ipv6hdr); | |
9fb9cbb1 YK |
643 | int len = skb->len - start; |
644 | u8 prevhdr = NEXTHDR_IPV6; | |
645 | ||
1ab1457c YH |
646 | while (nexthdr != NEXTHDR_FRAGMENT) { |
647 | struct ipv6_opt_hdr hdr; | |
648 | int hdrlen; | |
9fb9cbb1 YK |
649 | |
650 | if (!ipv6_ext_hdr(nexthdr)) { | |
651 | return -1; | |
652 | } | |
1ab1457c | 653 | if (len < (int)sizeof(struct ipv6_opt_hdr)) { |
0d53778e | 654 | pr_debug("too short\n"); |
9fb9cbb1 YK |
655 | return -1; |
656 | } | |
1ab1457c | 657 | if (nexthdr == NEXTHDR_NONE) { |
0d53778e | 658 | pr_debug("next header is none\n"); |
9fb9cbb1 YK |
659 | return -1; |
660 | } | |
1ab1457c YH |
661 | if (skb_copy_bits(skb, start, &hdr, sizeof(hdr))) |
662 | BUG(); | |
663 | if (nexthdr == NEXTHDR_AUTH) | |
664 | hdrlen = (hdr.hdrlen+2)<<2; | |
665 | else | |
666 | hdrlen = ipv6_optlen(&hdr); | |
9fb9cbb1 YK |
667 | |
668 | prevhdr = nexthdr; | |
669 | prev_nhoff = start; | |
670 | ||
1ab1457c YH |
671 | nexthdr = hdr.nexthdr; |
672 | len -= hdrlen; | |
673 | start += hdrlen; | |
674 | } | |
9fb9cbb1 YK |
675 | |
676 | if (len < 0) | |
677 | return -1; | |
678 | ||
679 | *prevhdrp = prevhdr; | |
680 | *prevhoff = prev_nhoff; | |
681 | *fhoff = start; | |
682 | ||
683 | return 0; | |
684 | } | |
685 | ||
686 | struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb) | |
687 | { | |
1ab1457c | 688 | struct sk_buff *clone; |
9fb9cbb1 YK |
689 | struct net_device *dev = skb->dev; |
690 | struct frag_hdr *fhdr; | |
691 | struct nf_ct_frag6_queue *fq; | |
692 | struct ipv6hdr *hdr; | |
693 | int fhoff, nhoff; | |
694 | u8 prevhdr; | |
695 | struct sk_buff *ret_skb = NULL; | |
696 | ||
697 | /* Jumbo payload inhibits frag. header */ | |
0660e03f | 698 | if (ipv6_hdr(skb)->payload_len == 0) { |
0d53778e | 699 | pr_debug("payload len = 0\n"); |
9fb9cbb1 YK |
700 | return skb; |
701 | } | |
702 | ||
703 | if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0) | |
704 | return skb; | |
705 | ||
706 | clone = skb_clone(skb, GFP_ATOMIC); | |
707 | if (clone == NULL) { | |
0d53778e | 708 | pr_debug("Can't clone skb\n"); |
9fb9cbb1 YK |
709 | return skb; |
710 | } | |
711 | ||
712 | NFCT_FRAG6_CB(clone)->orig = skb; | |
713 | ||
714 | if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) { | |
0d53778e | 715 | pr_debug("message is too short.\n"); |
9fb9cbb1 YK |
716 | goto ret_orig; |
717 | } | |
718 | ||
967b05f6 | 719 | skb_set_transport_header(clone, fhoff); |
0660e03f | 720 | hdr = ipv6_hdr(clone); |
bff9b61c | 721 | fhdr = (struct frag_hdr *)skb_transport_header(clone); |
9fb9cbb1 YK |
722 | |
723 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
0d53778e | 724 | pr_debug("Invalid fragment offset\n"); |
9fb9cbb1 YK |
725 | /* It is not a fragmented frame */ |
726 | goto ret_orig; | |
727 | } | |
728 | ||
04128f23 | 729 | if (atomic_read(&nf_frags.mem) > nf_frags_ctl.high_thresh) |
9fb9cbb1 YK |
730 | nf_ct_frag6_evictor(); |
731 | ||
732 | fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr); | |
733 | if (fq == NULL) { | |
0d53778e | 734 | pr_debug("Can't find and can't create new queue\n"); |
9fb9cbb1 YK |
735 | goto ret_orig; |
736 | } | |
737 | ||
5ab11c98 | 738 | spin_lock(&fq->q.lock); |
9fb9cbb1 YK |
739 | |
740 | if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) { | |
5ab11c98 | 741 | spin_unlock(&fq->q.lock); |
0d53778e | 742 | pr_debug("Can't insert skb to queue\n"); |
1ba430bc | 743 | fq_put(fq, NULL); |
9fb9cbb1 YK |
744 | goto ret_orig; |
745 | } | |
746 | ||
5ab11c98 | 747 | if (fq->q.last_in == (FIRST_IN|LAST_IN) && fq->q.meat == fq->q.len) { |
9fb9cbb1 YK |
748 | ret_skb = nf_ct_frag6_reasm(fq, dev); |
749 | if (ret_skb == NULL) | |
0d53778e | 750 | pr_debug("Can't reassemble fragmented packets\n"); |
9fb9cbb1 | 751 | } |
5ab11c98 | 752 | spin_unlock(&fq->q.lock); |
9fb9cbb1 | 753 | |
1ba430bc | 754 | fq_put(fq, NULL); |
9fb9cbb1 YK |
755 | return ret_skb; |
756 | ||
757 | ret_orig: | |
758 | kfree_skb(clone); | |
759 | return skb; | |
760 | } | |
761 | ||
762 | void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb, | |
763 | struct net_device *in, struct net_device *out, | |
764 | int (*okfn)(struct sk_buff *)) | |
765 | { | |
766 | struct sk_buff *s, *s2; | |
767 | ||
768 | for (s = NFCT_FRAG6_CB(skb)->orig; s;) { | |
769 | nf_conntrack_put_reasm(s->nfct_reasm); | |
770 | nf_conntrack_get_reasm(skb); | |
771 | s->nfct_reasm = skb; | |
772 | ||
773 | s2 = s->next; | |
f9f02cca PM |
774 | s->next = NULL; |
775 | ||
9fb9cbb1 YK |
776 | NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn, |
777 | NF_IP6_PRI_CONNTRACK_DEFRAG + 1); | |
778 | s = s2; | |
779 | } | |
780 | nf_conntrack_put_reasm(skb); | |
781 | } | |
782 | ||
783 | int nf_ct_frag6_kfree_frags(struct sk_buff *skb) | |
784 | { | |
785 | struct sk_buff *s, *s2; | |
786 | ||
787 | for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) { | |
788 | ||
789 | s2 = s->next; | |
790 | kfree_skb(s); | |
791 | } | |
792 | ||
793 | kfree_skb(skb); | |
794 | ||
795 | return 0; | |
796 | } | |
797 | ||
798 | int nf_ct_frag6_init(void) | |
799 | { | |
04128f23 | 800 | nf_frags.ctl = &nf_frags_ctl; |
321a3a99 | 801 | nf_frags.hashfn = nf_hashfn; |
7eb95156 | 802 | inet_frags_init(&nf_frags); |
9fb9cbb1 YK |
803 | |
804 | return 0; | |
805 | } | |
806 | ||
807 | void nf_ct_frag6_cleanup(void) | |
808 | { | |
7eb95156 PE |
809 | inet_frags_fini(&nf_frags); |
810 | ||
04128f23 | 811 | nf_frags_ctl.low_thresh = 0; |
9fb9cbb1 YK |
812 | nf_ct_frag6_evictor(); |
813 | } |