]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPv6 fragment reassembly | |
1ab1457c | 3 | * Linux INET6 implementation |
1da177e4 LT |
4 | * |
5 | * Authors: | |
1ab1457c | 6 | * Pedro Roque <roque@di.fc.ul.pt> |
1da177e4 LT |
7 | * |
8 | * $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $ | |
9 | * | |
10 | * Based on: net/ipv4/ip_fragment.c | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or | |
13 | * modify it under the terms of the GNU General Public License | |
14 | * as published by the Free Software Foundation; either version | |
15 | * 2 of the License, or (at your option) any later version. | |
16 | */ | |
17 | ||
1ab1457c YH |
18 | /* |
19 | * Fixes: | |
1da177e4 LT |
20 | * Andi Kleen Make it work with multiple hosts. |
21 | * More RFC compliance. | |
22 | * | |
23 | * Horst von Brand Add missing #include <linux/string.h> | |
24 | * Alexey Kuznetsov SMP races, threading, cleanup. | |
25 | * Patrick McHardy LRU queue of frag heads for evictor. | |
26 | * Mitsuru KANDA @USAGI Register inet6_protocol{}. | |
27 | * David Stevens and | |
28 | * YOSHIFUJI,H. @USAGI Always remove fragment header to | |
29 | * calculate ICV correctly. | |
30 | */ | |
1da177e4 LT |
31 | #include <linux/errno.h> |
32 | #include <linux/types.h> | |
33 | #include <linux/string.h> | |
34 | #include <linux/socket.h> | |
35 | #include <linux/sockios.h> | |
36 | #include <linux/jiffies.h> | |
37 | #include <linux/net.h> | |
38 | #include <linux/list.h> | |
39 | #include <linux/netdevice.h> | |
40 | #include <linux/in6.h> | |
41 | #include <linux/ipv6.h> | |
42 | #include <linux/icmpv6.h> | |
43 | #include <linux/random.h> | |
44 | #include <linux/jhash.h> | |
f61944ef | 45 | #include <linux/skbuff.h> |
1da177e4 LT |
46 | |
47 | #include <net/sock.h> | |
48 | #include <net/snmp.h> | |
49 | ||
50 | #include <net/ipv6.h> | |
a11d206d | 51 | #include <net/ip6_route.h> |
1da177e4 LT |
52 | #include <net/protocol.h> |
53 | #include <net/transp_v6.h> | |
54 | #include <net/rawv6.h> | |
55 | #include <net/ndisc.h> | |
56 | #include <net/addrconf.h> | |
5ab11c98 | 57 | #include <net/inet_frag.h> |
1da177e4 | 58 | |
1da177e4 LT |
59 | struct ip6frag_skb_cb |
60 | { | |
61 | struct inet6_skb_parm h; | |
62 | int offset; | |
63 | }; | |
64 | ||
65 | #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb)) | |
66 | ||
67 | ||
68 | /* | |
69 | * Equivalent of ipv4 struct ipq | |
70 | */ | |
71 | ||
72 | struct frag_queue | |
73 | { | |
5ab11c98 | 74 | struct inet_frag_queue q; |
1da177e4 | 75 | |
e69a4adc | 76 | __be32 id; /* fragment id */ |
1da177e4 LT |
77 | struct in6_addr saddr; |
78 | struct in6_addr daddr; | |
79 | ||
1da177e4 | 80 | int iif; |
1da177e4 | 81 | unsigned int csum; |
1da177e4 | 82 | __u16 nhoffset; |
1da177e4 LT |
83 | }; |
84 | ||
04128f23 PE |
85 | struct inet_frags_ctl ip6_frags_ctl __read_mostly = { |
86 | .high_thresh = 256 * 1024, | |
87 | .low_thresh = 192 * 1024, | |
88 | .timeout = IPV6_FRAG_TIMEOUT, | |
89 | .secret_interval = 10 * 60 * HZ, | |
90 | }; | |
91 | ||
7eb95156 | 92 | static struct inet_frags ip6_frags; |
1da177e4 | 93 | |
7eb95156 PE |
94 | int ip6_frag_nqueues(void) |
95 | { | |
96 | return ip6_frags.nqueues; | |
97 | } | |
1da177e4 | 98 | |
7eb95156 PE |
99 | int ip6_frag_mem(void) |
100 | { | |
101 | return atomic_read(&ip6_frags.mem); | |
102 | } | |
1da177e4 | 103 | |
f61944ef HX |
104 | static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev, |
105 | struct net_device *dev); | |
106 | ||
f6596f9d ZB |
107 | /* |
108 | * callers should be careful not to use the hash value outside the ipfrag_lock | |
109 | * as doing so could race with ipfrag_hash_rnd being recalculated. | |
110 | */ | |
e69a4adc | 111 | static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr, |
1da177e4 LT |
112 | struct in6_addr *daddr) |
113 | { | |
114 | u32 a, b, c; | |
115 | ||
e69a4adc AV |
116 | a = (__force u32)saddr->s6_addr32[0]; |
117 | b = (__force u32)saddr->s6_addr32[1]; | |
118 | c = (__force u32)saddr->s6_addr32[2]; | |
1da177e4 LT |
119 | |
120 | a += JHASH_GOLDEN_RATIO; | |
121 | b += JHASH_GOLDEN_RATIO; | |
7eb95156 | 122 | c += ip6_frags.rnd; |
1da177e4 LT |
123 | __jhash_mix(a, b, c); |
124 | ||
e69a4adc AV |
125 | a += (__force u32)saddr->s6_addr32[3]; |
126 | b += (__force u32)daddr->s6_addr32[0]; | |
127 | c += (__force u32)daddr->s6_addr32[1]; | |
1da177e4 LT |
128 | __jhash_mix(a, b, c); |
129 | ||
e69a4adc AV |
130 | a += (__force u32)daddr->s6_addr32[2]; |
131 | b += (__force u32)daddr->s6_addr32[3]; | |
132 | c += (__force u32)id; | |
1da177e4 LT |
133 | __jhash_mix(a, b, c); |
134 | ||
7eb95156 | 135 | return c & (INETFRAGS_HASHSZ - 1); |
1da177e4 LT |
136 | } |
137 | ||
1da177e4 LT |
138 | static void ip6_frag_secret_rebuild(unsigned long dummy) |
139 | { | |
140 | unsigned long now = jiffies; | |
141 | int i; | |
142 | ||
7eb95156 PE |
143 | write_lock(&ip6_frags.lock); |
144 | get_random_bytes(&ip6_frags.rnd, sizeof(u32)); | |
145 | for (i = 0; i < INETFRAGS_HASHSZ; i++) { | |
1da177e4 | 146 | struct frag_queue *q; |
e7c8a41e | 147 | struct hlist_node *p, *n; |
1da177e4 | 148 | |
7eb95156 | 149 | hlist_for_each_entry_safe(q, p, n, &ip6_frags.hash[i], q.list) { |
1da177e4 LT |
150 | unsigned int hval = ip6qhashfn(q->id, |
151 | &q->saddr, | |
152 | &q->daddr); | |
153 | ||
154 | if (hval != i) { | |
5ab11c98 | 155 | hlist_del(&q->q.list); |
1da177e4 LT |
156 | |
157 | /* Relink to new hash chain. */ | |
5ab11c98 | 158 | hlist_add_head(&q->q.list, |
7eb95156 | 159 | &ip6_frags.hash[hval]); |
1da177e4 | 160 | |
e7c8a41e | 161 | } |
1da177e4 LT |
162 | } |
163 | } | |
7eb95156 | 164 | write_unlock(&ip6_frags.lock); |
1da177e4 | 165 | |
04128f23 | 166 | mod_timer(&ip6_frags.secret_timer, now + ip6_frags_ctl.secret_interval); |
1da177e4 LT |
167 | } |
168 | ||
1da177e4 LT |
169 | /* Memory Tracking Functions. */ |
170 | static inline void frag_kfree_skb(struct sk_buff *skb, int *work) | |
171 | { | |
172 | if (work) | |
173 | *work -= skb->truesize; | |
7eb95156 | 174 | atomic_sub(skb->truesize, &ip6_frags.mem); |
1da177e4 LT |
175 | kfree_skb(skb); |
176 | } | |
177 | ||
178 | static inline void frag_free_queue(struct frag_queue *fq, int *work) | |
179 | { | |
180 | if (work) | |
181 | *work -= sizeof(struct frag_queue); | |
7eb95156 | 182 | atomic_sub(sizeof(struct frag_queue), &ip6_frags.mem); |
1da177e4 LT |
183 | kfree(fq); |
184 | } | |
185 | ||
186 | static inline struct frag_queue *frag_alloc_queue(void) | |
187 | { | |
78c784c4 | 188 | struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC); |
1da177e4 LT |
189 | |
190 | if(!fq) | |
191 | return NULL; | |
7eb95156 | 192 | atomic_add(sizeof(struct frag_queue), &ip6_frags.mem); |
1da177e4 LT |
193 | return fq; |
194 | } | |
195 | ||
196 | /* Destruction primitives. */ | |
197 | ||
198 | /* Complete destruction of fq. */ | |
199 | static void ip6_frag_destroy(struct frag_queue *fq, int *work) | |
200 | { | |
201 | struct sk_buff *fp; | |
202 | ||
5ab11c98 PE |
203 | BUG_TRAP(fq->q.last_in&COMPLETE); |
204 | BUG_TRAP(del_timer(&fq->q.timer) == 0); | |
1da177e4 LT |
205 | |
206 | /* Release all fragment data. */ | |
5ab11c98 | 207 | fp = fq->q.fragments; |
1da177e4 LT |
208 | while (fp) { |
209 | struct sk_buff *xp = fp->next; | |
210 | ||
211 | frag_kfree_skb(fp, work); | |
212 | fp = xp; | |
213 | } | |
214 | ||
215 | frag_free_queue(fq, work); | |
216 | } | |
217 | ||
218 | static __inline__ void fq_put(struct frag_queue *fq, int *work) | |
219 | { | |
5ab11c98 | 220 | if (atomic_dec_and_test(&fq->q.refcnt)) |
1da177e4 LT |
221 | ip6_frag_destroy(fq, work); |
222 | } | |
223 | ||
224 | /* Kill fq entry. It is not destroyed immediately, | |
225 | * because caller (and someone more) holds reference count. | |
226 | */ | |
227 | static __inline__ void fq_kill(struct frag_queue *fq) | |
228 | { | |
277e650d | 229 | inet_frag_kill(&fq->q, &ip6_frags); |
1da177e4 LT |
230 | } |
231 | ||
a11d206d | 232 | static void ip6_evictor(struct inet6_dev *idev) |
1da177e4 LT |
233 | { |
234 | struct frag_queue *fq; | |
235 | struct list_head *tmp; | |
236 | int work; | |
237 | ||
04128f23 | 238 | work = atomic_read(&ip6_frags.mem) - ip6_frags_ctl.low_thresh; |
1da177e4 LT |
239 | if (work <= 0) |
240 | return; | |
241 | ||
242 | while(work > 0) { | |
7eb95156 PE |
243 | read_lock(&ip6_frags.lock); |
244 | if (list_empty(&ip6_frags.lru_list)) { | |
245 | read_unlock(&ip6_frags.lock); | |
1da177e4 LT |
246 | return; |
247 | } | |
7eb95156 | 248 | tmp = ip6_frags.lru_list.next; |
5ab11c98 PE |
249 | fq = list_entry(tmp, struct frag_queue, q.lru_list); |
250 | atomic_inc(&fq->q.refcnt); | |
7eb95156 | 251 | read_unlock(&ip6_frags.lock); |
1da177e4 | 252 | |
5ab11c98 PE |
253 | spin_lock(&fq->q.lock); |
254 | if (!(fq->q.last_in&COMPLETE)) | |
1da177e4 | 255 | fq_kill(fq); |
5ab11c98 | 256 | spin_unlock(&fq->q.lock); |
1da177e4 LT |
257 | |
258 | fq_put(fq, &work); | |
a11d206d | 259 | IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
260 | } |
261 | } | |
262 | ||
263 | static void ip6_frag_expire(unsigned long data) | |
264 | { | |
265 | struct frag_queue *fq = (struct frag_queue *) data; | |
a11d206d | 266 | struct net_device *dev = NULL; |
1da177e4 | 267 | |
5ab11c98 | 268 | spin_lock(&fq->q.lock); |
1da177e4 | 269 | |
5ab11c98 | 270 | if (fq->q.last_in & COMPLETE) |
1da177e4 LT |
271 | goto out; |
272 | ||
273 | fq_kill(fq); | |
274 | ||
881d966b | 275 | dev = dev_get_by_index(&init_net, fq->iif); |
a11d206d YH |
276 | if (!dev) |
277 | goto out; | |
278 | ||
279 | rcu_read_lock(); | |
280 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT); | |
281 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
282 | rcu_read_unlock(); | |
1da177e4 | 283 | |
78c784c4 | 284 | /* Don't send error if the first segment did not arrive. */ |
5ab11c98 | 285 | if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments) |
78c784c4 IO |
286 | goto out; |
287 | ||
78c784c4 IO |
288 | /* |
289 | But use as source device on which LAST ARRIVED | |
290 | segment was received. And do not use fq->dev | |
291 | pointer directly, device might already disappeared. | |
292 | */ | |
5ab11c98 PE |
293 | fq->q.fragments->dev = dev; |
294 | icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev); | |
1da177e4 | 295 | out: |
a11d206d YH |
296 | if (dev) |
297 | dev_put(dev); | |
5ab11c98 | 298 | spin_unlock(&fq->q.lock); |
1da177e4 LT |
299 | fq_put(fq, NULL); |
300 | } | |
301 | ||
302 | /* Creation primitives. */ | |
303 | ||
304 | ||
f6596f9d | 305 | static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in) |
1da177e4 LT |
306 | { |
307 | struct frag_queue *fq; | |
f6596f9d | 308 | unsigned int hash; |
e7c8a41e YK |
309 | #ifdef CONFIG_SMP |
310 | struct hlist_node *n; | |
311 | #endif | |
1da177e4 | 312 | |
7eb95156 | 313 | write_lock(&ip6_frags.lock); |
f6596f9d | 314 | hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr); |
1da177e4 | 315 | #ifdef CONFIG_SMP |
7eb95156 | 316 | hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) { |
1ab1457c | 317 | if (fq->id == fq_in->id && |
1da177e4 LT |
318 | ipv6_addr_equal(&fq_in->saddr, &fq->saddr) && |
319 | ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) { | |
5ab11c98 | 320 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 321 | write_unlock(&ip6_frags.lock); |
5ab11c98 | 322 | fq_in->q.last_in |= COMPLETE; |
1da177e4 LT |
323 | fq_put(fq_in, NULL); |
324 | return fq; | |
325 | } | |
326 | } | |
327 | #endif | |
328 | fq = fq_in; | |
329 | ||
04128f23 | 330 | if (!mod_timer(&fq->q.timer, jiffies + ip6_frags_ctl.timeout)) |
5ab11c98 | 331 | atomic_inc(&fq->q.refcnt); |
1da177e4 | 332 | |
5ab11c98 | 333 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 334 | hlist_add_head(&fq->q.list, &ip6_frags.hash[hash]); |
5ab11c98 | 335 | INIT_LIST_HEAD(&fq->q.lru_list); |
7eb95156 PE |
336 | list_add_tail(&fq->q.lru_list, &ip6_frags.lru_list); |
337 | ip6_frags.nqueues++; | |
338 | write_unlock(&ip6_frags.lock); | |
1da177e4 LT |
339 | return fq; |
340 | } | |
341 | ||
342 | ||
343 | static struct frag_queue * | |
e69a4adc | 344 | ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst, |
a11d206d | 345 | struct inet6_dev *idev) |
1da177e4 LT |
346 | { |
347 | struct frag_queue *fq; | |
348 | ||
349 | if ((fq = frag_alloc_queue()) == NULL) | |
350 | goto oom; | |
351 | ||
1da177e4 LT |
352 | fq->id = id; |
353 | ipv6_addr_copy(&fq->saddr, src); | |
354 | ipv6_addr_copy(&fq->daddr, dst); | |
355 | ||
5ab11c98 PE |
356 | init_timer(&fq->q.timer); |
357 | fq->q.timer.function = ip6_frag_expire; | |
358 | fq->q.timer.data = (long) fq; | |
359 | spin_lock_init(&fq->q.lock); | |
360 | atomic_set(&fq->q.refcnt, 1); | |
1da177e4 | 361 | |
f6596f9d | 362 | return ip6_frag_intern(fq); |
1da177e4 LT |
363 | |
364 | oom: | |
a11d206d | 365 | IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
366 | return NULL; |
367 | } | |
368 | ||
369 | static __inline__ struct frag_queue * | |
e69a4adc | 370 | fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst, |
a11d206d | 371 | struct inet6_dev *idev) |
1da177e4 LT |
372 | { |
373 | struct frag_queue *fq; | |
e7c8a41e | 374 | struct hlist_node *n; |
f6596f9d | 375 | unsigned int hash; |
1da177e4 | 376 | |
7eb95156 | 377 | read_lock(&ip6_frags.lock); |
f6596f9d | 378 | hash = ip6qhashfn(id, src, dst); |
7eb95156 | 379 | hlist_for_each_entry(fq, n, &ip6_frags.hash[hash], q.list) { |
1ab1457c | 380 | if (fq->id == id && |
1da177e4 LT |
381 | ipv6_addr_equal(src, &fq->saddr) && |
382 | ipv6_addr_equal(dst, &fq->daddr)) { | |
5ab11c98 | 383 | atomic_inc(&fq->q.refcnt); |
7eb95156 | 384 | read_unlock(&ip6_frags.lock); |
1da177e4 LT |
385 | return fq; |
386 | } | |
387 | } | |
7eb95156 | 388 | read_unlock(&ip6_frags.lock); |
1da177e4 | 389 | |
a11d206d | 390 | return ip6_frag_create(id, src, dst, idev); |
1da177e4 LT |
391 | } |
392 | ||
393 | ||
f61944ef | 394 | static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb, |
1da177e4 LT |
395 | struct frag_hdr *fhdr, int nhoff) |
396 | { | |
397 | struct sk_buff *prev, *next; | |
f61944ef | 398 | struct net_device *dev; |
1da177e4 LT |
399 | int offset, end; |
400 | ||
5ab11c98 | 401 | if (fq->q.last_in & COMPLETE) |
1da177e4 LT |
402 | goto err; |
403 | ||
404 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
0660e03f ACM |
405 | end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - |
406 | ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); | |
1da177e4 LT |
407 | |
408 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
a11d206d YH |
409 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
410 | IPSTATS_MIB_INHDRERRORS); | |
d56f90a7 ACM |
411 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
412 | ((u8 *)&fhdr->frag_off - | |
413 | skb_network_header(skb))); | |
f61944ef | 414 | return -1; |
1da177e4 LT |
415 | } |
416 | ||
d56f90a7 ACM |
417 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
418 | const unsigned char *nh = skb_network_header(skb); | |
1ab1457c | 419 | skb->csum = csum_sub(skb->csum, |
d56f90a7 ACM |
420 | csum_partial(nh, (u8 *)(fhdr + 1) - nh, |
421 | 0)); | |
422 | } | |
1da177e4 LT |
423 | |
424 | /* Is this the final fragment? */ | |
425 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
426 | /* If we already have some bits beyond end | |
427 | * or have different end, the segment is corrupted. | |
428 | */ | |
5ab11c98 PE |
429 | if (end < fq->q.len || |
430 | ((fq->q.last_in & LAST_IN) && end != fq->q.len)) | |
1da177e4 | 431 | goto err; |
5ab11c98 PE |
432 | fq->q.last_in |= LAST_IN; |
433 | fq->q.len = end; | |
1da177e4 LT |
434 | } else { |
435 | /* Check if the fragment is rounded to 8 bytes. | |
436 | * Required by the RFC. | |
437 | */ | |
438 | if (end & 0x7) { | |
439 | /* RFC2460 says always send parameter problem in | |
440 | * this case. -DaveM | |
441 | */ | |
a11d206d YH |
442 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
443 | IPSTATS_MIB_INHDRERRORS); | |
1ab1457c | 444 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
1da177e4 | 445 | offsetof(struct ipv6hdr, payload_len)); |
f61944ef | 446 | return -1; |
1da177e4 | 447 | } |
5ab11c98 | 448 | if (end > fq->q.len) { |
1da177e4 | 449 | /* Some bits beyond end -> corruption. */ |
5ab11c98 | 450 | if (fq->q.last_in & LAST_IN) |
1da177e4 | 451 | goto err; |
5ab11c98 | 452 | fq->q.len = end; |
1da177e4 LT |
453 | } |
454 | } | |
455 | ||
456 | if (end == offset) | |
457 | goto err; | |
458 | ||
459 | /* Point into the IP datagram 'data' part. */ | |
460 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) | |
461 | goto err; | |
1ab1457c | 462 | |
42ca89c1 SH |
463 | if (pskb_trim_rcsum(skb, end - offset)) |
464 | goto err; | |
1da177e4 LT |
465 | |
466 | /* Find out which fragments are in front and at the back of us | |
467 | * in the chain of fragments so far. We must know where to put | |
468 | * this fragment, right? | |
469 | */ | |
470 | prev = NULL; | |
5ab11c98 | 471 | for(next = fq->q.fragments; next != NULL; next = next->next) { |
1da177e4 LT |
472 | if (FRAG6_CB(next)->offset >= offset) |
473 | break; /* bingo! */ | |
474 | prev = next; | |
475 | } | |
476 | ||
477 | /* We found where to put this one. Check for overlap with | |
478 | * preceding fragment, and, if needed, align things so that | |
479 | * any overlaps are eliminated. | |
480 | */ | |
481 | if (prev) { | |
482 | int i = (FRAG6_CB(prev)->offset + prev->len) - offset; | |
483 | ||
484 | if (i > 0) { | |
485 | offset += i; | |
486 | if (end <= offset) | |
487 | goto err; | |
488 | if (!pskb_pull(skb, i)) | |
489 | goto err; | |
490 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
491 | skb->ip_summed = CHECKSUM_NONE; | |
492 | } | |
493 | } | |
494 | ||
495 | /* Look for overlap with succeeding segments. | |
496 | * If we can merge fragments, do it. | |
497 | */ | |
498 | while (next && FRAG6_CB(next)->offset < end) { | |
499 | int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */ | |
500 | ||
501 | if (i < next->len) { | |
502 | /* Eat head of the next overlapped fragment | |
503 | * and leave the loop. The next ones cannot overlap. | |
504 | */ | |
505 | if (!pskb_pull(next, i)) | |
506 | goto err; | |
507 | FRAG6_CB(next)->offset += i; /* next fragment */ | |
5ab11c98 | 508 | fq->q.meat -= i; |
1da177e4 LT |
509 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
510 | next->ip_summed = CHECKSUM_NONE; | |
511 | break; | |
512 | } else { | |
513 | struct sk_buff *free_it = next; | |
514 | ||
515 | /* Old fragment is completely overridden with | |
516 | * new one drop it. | |
517 | */ | |
518 | next = next->next; | |
519 | ||
520 | if (prev) | |
521 | prev->next = next; | |
522 | else | |
5ab11c98 | 523 | fq->q.fragments = next; |
1da177e4 | 524 | |
5ab11c98 | 525 | fq->q.meat -= free_it->len; |
1da177e4 LT |
526 | frag_kfree_skb(free_it, NULL); |
527 | } | |
528 | } | |
529 | ||
530 | FRAG6_CB(skb)->offset = offset; | |
531 | ||
532 | /* Insert this fragment in the chain of fragments. */ | |
533 | skb->next = next; | |
534 | if (prev) | |
535 | prev->next = skb; | |
536 | else | |
5ab11c98 | 537 | fq->q.fragments = skb; |
1da177e4 | 538 | |
f61944ef HX |
539 | dev = skb->dev; |
540 | if (dev) { | |
541 | fq->iif = dev->ifindex; | |
542 | skb->dev = NULL; | |
543 | } | |
5ab11c98 PE |
544 | fq->q.stamp = skb->tstamp; |
545 | fq->q.meat += skb->len; | |
7eb95156 | 546 | atomic_add(skb->truesize, &ip6_frags.mem); |
1da177e4 LT |
547 | |
548 | /* The first fragment. | |
549 | * nhoffset is obtained from the first fragment, of course. | |
550 | */ | |
551 | if (offset == 0) { | |
552 | fq->nhoffset = nhoff; | |
5ab11c98 | 553 | fq->q.last_in |= FIRST_IN; |
1da177e4 | 554 | } |
f61944ef | 555 | |
5ab11c98 | 556 | if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len) |
f61944ef HX |
557 | return ip6_frag_reasm(fq, prev, dev); |
558 | ||
7eb95156 PE |
559 | write_lock(&ip6_frags.lock); |
560 | list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list); | |
561 | write_unlock(&ip6_frags.lock); | |
f61944ef | 562 | return -1; |
1da177e4 LT |
563 | |
564 | err: | |
a11d206d | 565 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 | 566 | kfree_skb(skb); |
f61944ef | 567 | return -1; |
1da177e4 LT |
568 | } |
569 | ||
570 | /* | |
571 | * Check if this packet is complete. | |
572 | * Returns NULL on failure by any reason, and pointer | |
573 | * to current nexthdr field in reassembled frame. | |
574 | * | |
575 | * It is called with locked fq, and caller must check that | |
576 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
577 | * the last and the first frames arrived and all the bits are here. | |
578 | */ | |
f61944ef | 579 | static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev, |
1da177e4 LT |
580 | struct net_device *dev) |
581 | { | |
5ab11c98 | 582 | struct sk_buff *fp, *head = fq->q.fragments; |
1da177e4 LT |
583 | int payload_len; |
584 | unsigned int nhoff; | |
585 | ||
586 | fq_kill(fq); | |
587 | ||
f61944ef HX |
588 | /* Make the one we just received the head. */ |
589 | if (prev) { | |
590 | head = prev->next; | |
591 | fp = skb_clone(head, GFP_ATOMIC); | |
592 | ||
593 | if (!fp) | |
594 | goto out_oom; | |
595 | ||
596 | fp->next = head->next; | |
597 | prev->next = fp; | |
598 | ||
5ab11c98 PE |
599 | skb_morph(head, fq->q.fragments); |
600 | head->next = fq->q.fragments->next; | |
f61944ef | 601 | |
5ab11c98 PE |
602 | kfree_skb(fq->q.fragments); |
603 | fq->q.fragments = head; | |
f61944ef HX |
604 | } |
605 | ||
1da177e4 LT |
606 | BUG_TRAP(head != NULL); |
607 | BUG_TRAP(FRAG6_CB(head)->offset == 0); | |
608 | ||
609 | /* Unfragmented part is taken from the first segment. */ | |
d56f90a7 | 610 | payload_len = ((head->data - skb_network_header(head)) - |
5ab11c98 | 611 | sizeof(struct ipv6hdr) + fq->q.len - |
d56f90a7 | 612 | sizeof(struct frag_hdr)); |
1da177e4 LT |
613 | if (payload_len > IPV6_MAXPLEN) |
614 | goto out_oversize; | |
615 | ||
616 | /* Head of list must not be cloned. */ | |
617 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
618 | goto out_oom; | |
619 | ||
620 | /* If the first fragment is fragmented itself, we split | |
621 | * it to two chunks: the first with data and paged part | |
622 | * and the second, holding only fragments. */ | |
623 | if (skb_shinfo(head)->frag_list) { | |
624 | struct sk_buff *clone; | |
625 | int i, plen = 0; | |
626 | ||
627 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
628 | goto out_oom; | |
629 | clone->next = head->next; | |
630 | head->next = clone; | |
631 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
632 | skb_shinfo(head)->frag_list = NULL; | |
633 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
634 | plen += skb_shinfo(head)->frags[i].size; | |
635 | clone->len = clone->data_len = head->data_len - plen; | |
636 | head->data_len -= clone->len; | |
637 | head->len -= clone->len; | |
638 | clone->csum = 0; | |
639 | clone->ip_summed = head->ip_summed; | |
7eb95156 | 640 | atomic_add(clone->truesize, &ip6_frags.mem); |
1da177e4 LT |
641 | } |
642 | ||
643 | /* We have to remove fragment header from datagram and to relocate | |
644 | * header in order to calculate ICV correctly. */ | |
645 | nhoff = fq->nhoffset; | |
b0e380b1 | 646 | skb_network_header(head)[nhoff] = skb_transport_header(head)[0]; |
1ab1457c | 647 | memmove(head->head + sizeof(struct frag_hdr), head->head, |
1da177e4 | 648 | (head->data - head->head) - sizeof(struct frag_hdr)); |
b0e380b1 ACM |
649 | head->mac_header += sizeof(struct frag_hdr); |
650 | head->network_header += sizeof(struct frag_hdr); | |
1da177e4 LT |
651 | |
652 | skb_shinfo(head)->frag_list = head->next; | |
badff6d0 | 653 | skb_reset_transport_header(head); |
d56f90a7 | 654 | skb_push(head, head->data - skb_network_header(head)); |
7eb95156 | 655 | atomic_sub(head->truesize, &ip6_frags.mem); |
1da177e4 LT |
656 | |
657 | for (fp=head->next; fp; fp = fp->next) { | |
658 | head->data_len += fp->len; | |
659 | head->len += fp->len; | |
660 | if (head->ip_summed != fp->ip_summed) | |
661 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 662 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
663 | head->csum = csum_add(head->csum, fp->csum); |
664 | head->truesize += fp->truesize; | |
7eb95156 | 665 | atomic_sub(fp->truesize, &ip6_frags.mem); |
1da177e4 LT |
666 | } |
667 | ||
668 | head->next = NULL; | |
669 | head->dev = dev; | |
5ab11c98 | 670 | head->tstamp = fq->q.stamp; |
0660e03f | 671 | ipv6_hdr(head)->payload_len = htons(payload_len); |
951dbc8a | 672 | IP6CB(head)->nhoff = nhoff; |
1da177e4 | 673 | |
1da177e4 | 674 | /* Yes, and fold redundant checksum back. 8) */ |
84fa7933 | 675 | if (head->ip_summed == CHECKSUM_COMPLETE) |
d56f90a7 | 676 | head->csum = csum_partial(skb_network_header(head), |
cfe1fc77 | 677 | skb_network_header_len(head), |
d56f90a7 | 678 | head->csum); |
1da177e4 | 679 | |
a11d206d YH |
680 | rcu_read_lock(); |
681 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS); | |
682 | rcu_read_unlock(); | |
5ab11c98 | 683 | fq->q.fragments = NULL; |
1da177e4 LT |
684 | return 1; |
685 | ||
686 | out_oversize: | |
687 | if (net_ratelimit()) | |
688 | printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len); | |
689 | goto out_fail; | |
690 | out_oom: | |
691 | if (net_ratelimit()) | |
692 | printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n"); | |
693 | out_fail: | |
a11d206d YH |
694 | rcu_read_lock(); |
695 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
696 | rcu_read_unlock(); | |
1da177e4 LT |
697 | return -1; |
698 | } | |
699 | ||
951dbc8a | 700 | static int ipv6_frag_rcv(struct sk_buff **skbp) |
1da177e4 | 701 | { |
1ab1457c | 702 | struct sk_buff *skb = *skbp; |
1da177e4 LT |
703 | struct frag_hdr *fhdr; |
704 | struct frag_queue *fq; | |
0660e03f | 705 | struct ipv6hdr *hdr = ipv6_hdr(skb); |
1da177e4 | 706 | |
a11d206d | 707 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
708 | |
709 | /* Jumbo payload inhibits frag. header */ | |
710 | if (hdr->payload_len==0) { | |
a11d206d | 711 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
cfe1fc77 ACM |
712 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
713 | skb_network_header_len(skb)); | |
1da177e4 LT |
714 | return -1; |
715 | } | |
ea2ae17d ACM |
716 | if (!pskb_may_pull(skb, (skb_transport_offset(skb) + |
717 | sizeof(struct frag_hdr)))) { | |
a11d206d | 718 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
cfe1fc77 ACM |
719 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
720 | skb_network_header_len(skb)); | |
1da177e4 LT |
721 | return -1; |
722 | } | |
723 | ||
0660e03f | 724 | hdr = ipv6_hdr(skb); |
9c70220b | 725 | fhdr = (struct frag_hdr *)skb_transport_header(skb); |
1da177e4 LT |
726 | |
727 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
728 | /* It is not a fragmented frame */ | |
b0e380b1 | 729 | skb->transport_header += sizeof(struct frag_hdr); |
a11d206d | 730 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS); |
1da177e4 | 731 | |
d56f90a7 | 732 | IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb); |
1da177e4 LT |
733 | return 1; |
734 | } | |
735 | ||
04128f23 | 736 | if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh) |
a11d206d | 737 | ip6_evictor(ip6_dst_idev(skb->dst)); |
1da177e4 | 738 | |
a11d206d YH |
739 | if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr, |
740 | ip6_dst_idev(skb->dst))) != NULL) { | |
f61944ef | 741 | int ret; |
1da177e4 | 742 | |
5ab11c98 | 743 | spin_lock(&fq->q.lock); |
1da177e4 | 744 | |
f61944ef | 745 | ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff); |
1da177e4 | 746 | |
5ab11c98 | 747 | spin_unlock(&fq->q.lock); |
1da177e4 LT |
748 | fq_put(fq, NULL); |
749 | return ret; | |
750 | } | |
751 | ||
a11d206d | 752 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
753 | kfree_skb(skb); |
754 | return -1; | |
755 | } | |
756 | ||
757 | static struct inet6_protocol frag_protocol = | |
758 | { | |
759 | .handler = ipv6_frag_rcv, | |
760 | .flags = INET6_PROTO_NOPOLICY, | |
761 | }; | |
762 | ||
763 | void __init ipv6_frag_init(void) | |
764 | { | |
765 | if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0) | |
766 | printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n"); | |
767 | ||
7eb95156 PE |
768 | init_timer(&ip6_frags.secret_timer); |
769 | ip6_frags.secret_timer.function = ip6_frag_secret_rebuild; | |
04128f23 | 770 | ip6_frags.secret_timer.expires = jiffies + ip6_frags_ctl.secret_interval; |
7eb95156 | 771 | add_timer(&ip6_frags.secret_timer); |
1da177e4 | 772 | |
04128f23 | 773 | ip6_frags.ctl = &ip6_frags_ctl; |
7eb95156 | 774 | inet_frags_init(&ip6_frags); |
1da177e4 | 775 | } |