]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPv6 fragment reassembly | |
3 | * Linux INET6 implementation | |
4 | * | |
5 | * Authors: | |
6 | * Pedro Roque <roque@di.fc.ul.pt> | |
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 | ||
18 | /* | |
19 | * Fixes: | |
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> | |
45 | ||
46 | #include <net/sock.h> | |
47 | #include <net/snmp.h> | |
48 | ||
49 | #include <net/ipv6.h> | |
a11d206d | 50 | #include <net/ip6_route.h> |
1da177e4 LT |
51 | #include <net/protocol.h> |
52 | #include <net/transp_v6.h> | |
53 | #include <net/rawv6.h> | |
54 | #include <net/ndisc.h> | |
55 | #include <net/addrconf.h> | |
56 | ||
ab32ea5d BH |
57 | int sysctl_ip6frag_high_thresh __read_mostly = 256*1024; |
58 | int sysctl_ip6frag_low_thresh __read_mostly = 192*1024; | |
1da177e4 | 59 | |
ab32ea5d | 60 | int sysctl_ip6frag_time __read_mostly = IPV6_FRAG_TIMEOUT; |
1da177e4 LT |
61 | |
62 | struct ip6frag_skb_cb | |
63 | { | |
64 | struct inet6_skb_parm h; | |
65 | int offset; | |
66 | }; | |
67 | ||
68 | #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb)) | |
69 | ||
70 | ||
71 | /* | |
72 | * Equivalent of ipv4 struct ipq | |
73 | */ | |
74 | ||
75 | struct frag_queue | |
76 | { | |
e7c8a41e | 77 | struct hlist_node list; |
1da177e4 LT |
78 | struct list_head lru_list; /* lru list member */ |
79 | ||
80 | __u32 id; /* fragment id */ | |
81 | struct in6_addr saddr; | |
82 | struct in6_addr daddr; | |
83 | ||
84 | spinlock_t lock; | |
85 | atomic_t refcnt; | |
86 | struct timer_list timer; /* expire timer */ | |
87 | struct sk_buff *fragments; | |
88 | int len; | |
89 | int meat; | |
90 | int iif; | |
91 | struct timeval stamp; | |
92 | unsigned int csum; | |
93 | __u8 last_in; /* has first/last segment arrived? */ | |
94 | #define COMPLETE 4 | |
95 | #define FIRST_IN 2 | |
96 | #define LAST_IN 1 | |
97 | __u16 nhoffset; | |
1da177e4 LT |
98 | }; |
99 | ||
100 | /* Hash table. */ | |
101 | ||
102 | #define IP6Q_HASHSZ 64 | |
103 | ||
e7c8a41e | 104 | static struct hlist_head ip6_frag_hash[IP6Q_HASHSZ]; |
1da177e4 LT |
105 | static DEFINE_RWLOCK(ip6_frag_lock); |
106 | static u32 ip6_frag_hash_rnd; | |
107 | static LIST_HEAD(ip6_frag_lru_list); | |
108 | int ip6_frag_nqueues = 0; | |
109 | ||
110 | static __inline__ void __fq_unlink(struct frag_queue *fq) | |
111 | { | |
e7c8a41e | 112 | hlist_del(&fq->list); |
1da177e4 LT |
113 | list_del(&fq->lru_list); |
114 | ip6_frag_nqueues--; | |
115 | } | |
116 | ||
117 | static __inline__ void fq_unlink(struct frag_queue *fq) | |
118 | { | |
119 | write_lock(&ip6_frag_lock); | |
120 | __fq_unlink(fq); | |
121 | write_unlock(&ip6_frag_lock); | |
122 | } | |
123 | ||
f6596f9d ZB |
124 | /* |
125 | * callers should be careful not to use the hash value outside the ipfrag_lock | |
126 | * as doing so could race with ipfrag_hash_rnd being recalculated. | |
127 | */ | |
1da177e4 LT |
128 | static unsigned int ip6qhashfn(u32 id, struct in6_addr *saddr, |
129 | struct in6_addr *daddr) | |
130 | { | |
131 | u32 a, b, c; | |
132 | ||
133 | a = saddr->s6_addr32[0]; | |
134 | b = saddr->s6_addr32[1]; | |
135 | c = saddr->s6_addr32[2]; | |
136 | ||
137 | a += JHASH_GOLDEN_RATIO; | |
138 | b += JHASH_GOLDEN_RATIO; | |
139 | c += ip6_frag_hash_rnd; | |
140 | __jhash_mix(a, b, c); | |
141 | ||
142 | a += saddr->s6_addr32[3]; | |
143 | b += daddr->s6_addr32[0]; | |
144 | c += daddr->s6_addr32[1]; | |
145 | __jhash_mix(a, b, c); | |
146 | ||
147 | a += daddr->s6_addr32[2]; | |
148 | b += daddr->s6_addr32[3]; | |
149 | c += id; | |
150 | __jhash_mix(a, b, c); | |
151 | ||
152 | return c & (IP6Q_HASHSZ - 1); | |
153 | } | |
154 | ||
155 | static struct timer_list ip6_frag_secret_timer; | |
ab32ea5d | 156 | int sysctl_ip6frag_secret_interval __read_mostly = 10 * 60 * HZ; |
1da177e4 LT |
157 | |
158 | static void ip6_frag_secret_rebuild(unsigned long dummy) | |
159 | { | |
160 | unsigned long now = jiffies; | |
161 | int i; | |
162 | ||
163 | write_lock(&ip6_frag_lock); | |
164 | get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32)); | |
165 | for (i = 0; i < IP6Q_HASHSZ; i++) { | |
166 | struct frag_queue *q; | |
e7c8a41e | 167 | struct hlist_node *p, *n; |
1da177e4 | 168 | |
e7c8a41e | 169 | hlist_for_each_entry_safe(q, p, n, &ip6_frag_hash[i], list) { |
1da177e4 LT |
170 | unsigned int hval = ip6qhashfn(q->id, |
171 | &q->saddr, | |
172 | &q->daddr); | |
173 | ||
174 | if (hval != i) { | |
e7c8a41e | 175 | hlist_del(&q->list); |
1da177e4 LT |
176 | |
177 | /* Relink to new hash chain. */ | |
e7c8a41e YK |
178 | hlist_add_head(&q->list, |
179 | &ip6_frag_hash[hval]); | |
1da177e4 | 180 | |
e7c8a41e | 181 | } |
1da177e4 LT |
182 | } |
183 | } | |
184 | write_unlock(&ip6_frag_lock); | |
185 | ||
186 | mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval); | |
187 | } | |
188 | ||
189 | atomic_t ip6_frag_mem = ATOMIC_INIT(0); | |
190 | ||
191 | /* Memory Tracking Functions. */ | |
192 | static inline void frag_kfree_skb(struct sk_buff *skb, int *work) | |
193 | { | |
194 | if (work) | |
195 | *work -= skb->truesize; | |
196 | atomic_sub(skb->truesize, &ip6_frag_mem); | |
197 | kfree_skb(skb); | |
198 | } | |
199 | ||
200 | static inline void frag_free_queue(struct frag_queue *fq, int *work) | |
201 | { | |
202 | if (work) | |
203 | *work -= sizeof(struct frag_queue); | |
204 | atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem); | |
205 | kfree(fq); | |
206 | } | |
207 | ||
208 | static inline struct frag_queue *frag_alloc_queue(void) | |
209 | { | |
78c784c4 | 210 | struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC); |
1da177e4 LT |
211 | |
212 | if(!fq) | |
213 | return NULL; | |
214 | atomic_add(sizeof(struct frag_queue), &ip6_frag_mem); | |
215 | return fq; | |
216 | } | |
217 | ||
218 | /* Destruction primitives. */ | |
219 | ||
220 | /* Complete destruction of fq. */ | |
221 | static void ip6_frag_destroy(struct frag_queue *fq, int *work) | |
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, work); | |
234 | fp = xp; | |
235 | } | |
236 | ||
237 | frag_free_queue(fq, work); | |
238 | } | |
239 | ||
240 | static __inline__ void fq_put(struct frag_queue *fq, int *work) | |
241 | { | |
242 | if (atomic_dec_and_test(&fq->refcnt)) | |
243 | ip6_frag_destroy(fq, work); | |
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 frag_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 | ||
a11d206d | 261 | static void ip6_evictor(struct inet6_dev *idev) |
1da177e4 LT |
262 | { |
263 | struct frag_queue *fq; | |
264 | struct list_head *tmp; | |
265 | int work; | |
266 | ||
267 | work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh; | |
268 | if (work <= 0) | |
269 | return; | |
270 | ||
271 | while(work > 0) { | |
272 | read_lock(&ip6_frag_lock); | |
273 | if (list_empty(&ip6_frag_lru_list)) { | |
274 | read_unlock(&ip6_frag_lock); | |
275 | return; | |
276 | } | |
277 | tmp = ip6_frag_lru_list.next; | |
278 | fq = list_entry(tmp, struct frag_queue, lru_list); | |
279 | atomic_inc(&fq->refcnt); | |
280 | read_unlock(&ip6_frag_lock); | |
281 | ||
282 | spin_lock(&fq->lock); | |
283 | if (!(fq->last_in&COMPLETE)) | |
284 | fq_kill(fq); | |
285 | spin_unlock(&fq->lock); | |
286 | ||
287 | fq_put(fq, &work); | |
a11d206d | 288 | IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
289 | } |
290 | } | |
291 | ||
292 | static void ip6_frag_expire(unsigned long data) | |
293 | { | |
294 | struct frag_queue *fq = (struct frag_queue *) data; | |
a11d206d | 295 | struct net_device *dev = NULL; |
1da177e4 LT |
296 | |
297 | spin_lock(&fq->lock); | |
298 | ||
299 | if (fq->last_in & COMPLETE) | |
300 | goto out; | |
301 | ||
302 | fq_kill(fq); | |
303 | ||
a11d206d YH |
304 | dev = dev_get_by_index(fq->iif); |
305 | if (!dev) | |
306 | goto out; | |
307 | ||
308 | rcu_read_lock(); | |
309 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT); | |
310 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
311 | rcu_read_unlock(); | |
1da177e4 | 312 | |
78c784c4 IO |
313 | /* Don't send error if the first segment did not arrive. */ |
314 | if (!(fq->last_in&FIRST_IN) || !fq->fragments) | |
315 | goto out; | |
316 | ||
78c784c4 IO |
317 | /* |
318 | But use as source device on which LAST ARRIVED | |
319 | segment was received. And do not use fq->dev | |
320 | pointer directly, device might already disappeared. | |
321 | */ | |
322 | fq->fragments->dev = dev; | |
323 | icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev); | |
1da177e4 | 324 | out: |
a11d206d YH |
325 | if (dev) |
326 | dev_put(dev); | |
1da177e4 LT |
327 | spin_unlock(&fq->lock); |
328 | fq_put(fq, NULL); | |
329 | } | |
330 | ||
331 | /* Creation primitives. */ | |
332 | ||
333 | ||
f6596f9d | 334 | static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in) |
1da177e4 LT |
335 | { |
336 | struct frag_queue *fq; | |
f6596f9d | 337 | unsigned int hash; |
e7c8a41e YK |
338 | #ifdef CONFIG_SMP |
339 | struct hlist_node *n; | |
340 | #endif | |
1da177e4 LT |
341 | |
342 | write_lock(&ip6_frag_lock); | |
f6596f9d | 343 | hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr); |
1da177e4 | 344 | #ifdef CONFIG_SMP |
e7c8a41e | 345 | hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) { |
1da177e4 LT |
346 | if (fq->id == fq_in->id && |
347 | ipv6_addr_equal(&fq_in->saddr, &fq->saddr) && | |
348 | ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) { | |
349 | atomic_inc(&fq->refcnt); | |
350 | write_unlock(&ip6_frag_lock); | |
351 | fq_in->last_in |= COMPLETE; | |
352 | fq_put(fq_in, NULL); | |
353 | return fq; | |
354 | } | |
355 | } | |
356 | #endif | |
357 | fq = fq_in; | |
358 | ||
359 | if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time)) | |
360 | atomic_inc(&fq->refcnt); | |
361 | ||
362 | atomic_inc(&fq->refcnt); | |
e7c8a41e | 363 | hlist_add_head(&fq->list, &ip6_frag_hash[hash]); |
1da177e4 LT |
364 | INIT_LIST_HEAD(&fq->lru_list); |
365 | list_add_tail(&fq->lru_list, &ip6_frag_lru_list); | |
366 | ip6_frag_nqueues++; | |
367 | write_unlock(&ip6_frag_lock); | |
368 | return fq; | |
369 | } | |
370 | ||
371 | ||
372 | static struct frag_queue * | |
a11d206d YH |
373 | ip6_frag_create(u32 id, struct in6_addr *src, struct in6_addr *dst, |
374 | struct inet6_dev *idev) | |
1da177e4 LT |
375 | { |
376 | struct frag_queue *fq; | |
377 | ||
378 | if ((fq = frag_alloc_queue()) == NULL) | |
379 | goto oom; | |
380 | ||
1da177e4 LT |
381 | fq->id = id; |
382 | ipv6_addr_copy(&fq->saddr, src); | |
383 | ipv6_addr_copy(&fq->daddr, dst); | |
384 | ||
385 | init_timer(&fq->timer); | |
386 | fq->timer.function = ip6_frag_expire; | |
387 | fq->timer.data = (long) fq; | |
388 | spin_lock_init(&fq->lock); | |
389 | atomic_set(&fq->refcnt, 1); | |
390 | ||
f6596f9d | 391 | return ip6_frag_intern(fq); |
1da177e4 LT |
392 | |
393 | oom: | |
a11d206d | 394 | IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
395 | return NULL; |
396 | } | |
397 | ||
398 | static __inline__ struct frag_queue * | |
a11d206d YH |
399 | fq_find(u32 id, struct in6_addr *src, struct in6_addr *dst, |
400 | struct inet6_dev *idev) | |
1da177e4 LT |
401 | { |
402 | struct frag_queue *fq; | |
e7c8a41e | 403 | struct hlist_node *n; |
f6596f9d | 404 | unsigned int hash; |
1da177e4 LT |
405 | |
406 | read_lock(&ip6_frag_lock); | |
f6596f9d | 407 | hash = ip6qhashfn(id, src, dst); |
e7c8a41e | 408 | hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) { |
1da177e4 LT |
409 | if (fq->id == id && |
410 | ipv6_addr_equal(src, &fq->saddr) && | |
411 | ipv6_addr_equal(dst, &fq->daddr)) { | |
412 | atomic_inc(&fq->refcnt); | |
413 | read_unlock(&ip6_frag_lock); | |
414 | return fq; | |
415 | } | |
416 | } | |
417 | read_unlock(&ip6_frag_lock); | |
418 | ||
a11d206d | 419 | return ip6_frag_create(id, src, dst, idev); |
1da177e4 LT |
420 | } |
421 | ||
422 | ||
423 | static void ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb, | |
424 | struct frag_hdr *fhdr, int nhoff) | |
425 | { | |
426 | struct sk_buff *prev, *next; | |
427 | int offset, end; | |
428 | ||
429 | if (fq->last_in & COMPLETE) | |
430 | goto err; | |
431 | ||
432 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
433 | end = offset + (ntohs(skb->nh.ipv6h->payload_len) - | |
434 | ((u8 *) (fhdr + 1) - (u8 *) (skb->nh.ipv6h + 1))); | |
435 | ||
436 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
a11d206d YH |
437 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
438 | IPSTATS_MIB_INHDRERRORS); | |
1da177e4 LT |
439 | icmpv6_param_prob(skb,ICMPV6_HDR_FIELD, (u8*)&fhdr->frag_off - skb->nh.raw); |
440 | return; | |
441 | } | |
442 | ||
84fa7933 | 443 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
444 | skb->csum = csum_sub(skb->csum, |
445 | csum_partial(skb->nh.raw, (u8*)(fhdr+1)-skb->nh.raw, 0)); | |
446 | ||
447 | /* Is this the final fragment? */ | |
448 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
449 | /* If we already have some bits beyond end | |
450 | * or have different end, the segment is corrupted. | |
451 | */ | |
452 | if (end < fq->len || | |
453 | ((fq->last_in & LAST_IN) && end != fq->len)) | |
454 | goto err; | |
455 | fq->last_in |= LAST_IN; | |
456 | fq->len = end; | |
457 | } else { | |
458 | /* Check if the fragment is rounded to 8 bytes. | |
459 | * Required by the RFC. | |
460 | */ | |
461 | if (end & 0x7) { | |
462 | /* RFC2460 says always send parameter problem in | |
463 | * this case. -DaveM | |
464 | */ | |
a11d206d YH |
465 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
466 | IPSTATS_MIB_INHDRERRORS); | |
1da177e4 LT |
467 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
468 | offsetof(struct ipv6hdr, payload_len)); | |
469 | return; | |
470 | } | |
471 | if (end > fq->len) { | |
472 | /* Some bits beyond end -> corruption. */ | |
473 | if (fq->last_in & LAST_IN) | |
474 | goto err; | |
475 | fq->len = end; | |
476 | } | |
477 | } | |
478 | ||
479 | if (end == offset) | |
480 | goto err; | |
481 | ||
482 | /* Point into the IP datagram 'data' part. */ | |
483 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) | |
484 | goto err; | |
42ca89c1 SH |
485 | |
486 | if (pskb_trim_rcsum(skb, end - offset)) | |
487 | goto err; | |
1da177e4 LT |
488 | |
489 | /* Find out which fragments are in front and at the back of us | |
490 | * in the chain of fragments so far. We must know where to put | |
491 | * this fragment, right? | |
492 | */ | |
493 | prev = NULL; | |
494 | for(next = fq->fragments; next != NULL; next = next->next) { | |
495 | if (FRAG6_CB(next)->offset >= offset) | |
496 | break; /* bingo! */ | |
497 | prev = next; | |
498 | } | |
499 | ||
500 | /* We found where to put this one. Check for overlap with | |
501 | * preceding fragment, and, if needed, align things so that | |
502 | * any overlaps are eliminated. | |
503 | */ | |
504 | if (prev) { | |
505 | int i = (FRAG6_CB(prev)->offset + prev->len) - offset; | |
506 | ||
507 | if (i > 0) { | |
508 | offset += i; | |
509 | if (end <= offset) | |
510 | goto err; | |
511 | if (!pskb_pull(skb, i)) | |
512 | goto err; | |
513 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
514 | skb->ip_summed = CHECKSUM_NONE; | |
515 | } | |
516 | } | |
517 | ||
518 | /* Look for overlap with succeeding segments. | |
519 | * If we can merge fragments, do it. | |
520 | */ | |
521 | while (next && FRAG6_CB(next)->offset < end) { | |
522 | int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */ | |
523 | ||
524 | if (i < next->len) { | |
525 | /* Eat head of the next overlapped fragment | |
526 | * and leave the loop. The next ones cannot overlap. | |
527 | */ | |
528 | if (!pskb_pull(next, i)) | |
529 | goto err; | |
530 | FRAG6_CB(next)->offset += i; /* next fragment */ | |
531 | fq->meat -= i; | |
532 | if (next->ip_summed != CHECKSUM_UNNECESSARY) | |
533 | next->ip_summed = CHECKSUM_NONE; | |
534 | break; | |
535 | } else { | |
536 | struct sk_buff *free_it = next; | |
537 | ||
538 | /* Old fragment is completely overridden with | |
539 | * new one drop it. | |
540 | */ | |
541 | next = next->next; | |
542 | ||
543 | if (prev) | |
544 | prev->next = next; | |
545 | else | |
546 | fq->fragments = next; | |
547 | ||
548 | fq->meat -= free_it->len; | |
549 | frag_kfree_skb(free_it, NULL); | |
550 | } | |
551 | } | |
552 | ||
553 | FRAG6_CB(skb)->offset = offset; | |
554 | ||
555 | /* Insert this fragment in the chain of fragments. */ | |
556 | skb->next = next; | |
557 | if (prev) | |
558 | prev->next = skb; | |
559 | else | |
560 | fq->fragments = skb; | |
561 | ||
562 | if (skb->dev) | |
563 | fq->iif = skb->dev->ifindex; | |
564 | skb->dev = NULL; | |
a61bbcf2 | 565 | skb_get_timestamp(skb, &fq->stamp); |
1da177e4 LT |
566 | fq->meat += skb->len; |
567 | atomic_add(skb->truesize, &ip6_frag_mem); | |
568 | ||
569 | /* The first fragment. | |
570 | * nhoffset is obtained from the first fragment, of course. | |
571 | */ | |
572 | if (offset == 0) { | |
573 | fq->nhoffset = nhoff; | |
574 | fq->last_in |= FIRST_IN; | |
575 | } | |
576 | write_lock(&ip6_frag_lock); | |
577 | list_move_tail(&fq->lru_list, &ip6_frag_lru_list); | |
578 | write_unlock(&ip6_frag_lock); | |
579 | return; | |
580 | ||
581 | err: | |
a11d206d | 582 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
583 | kfree_skb(skb); |
584 | } | |
585 | ||
586 | /* | |
587 | * Check if this packet is complete. | |
588 | * Returns NULL on failure by any reason, and pointer | |
589 | * to current nexthdr field in reassembled frame. | |
590 | * | |
591 | * It is called with locked fq, and caller must check that | |
592 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
593 | * the last and the first frames arrived and all the bits are here. | |
594 | */ | |
595 | static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff **skb_in, | |
1da177e4 LT |
596 | struct net_device *dev) |
597 | { | |
598 | struct sk_buff *fp, *head = fq->fragments; | |
599 | int payload_len; | |
600 | unsigned int nhoff; | |
601 | ||
602 | fq_kill(fq); | |
603 | ||
604 | BUG_TRAP(head != NULL); | |
605 | BUG_TRAP(FRAG6_CB(head)->offset == 0); | |
606 | ||
607 | /* Unfragmented part is taken from the first segment. */ | |
608 | payload_len = (head->data - head->nh.raw) - sizeof(struct ipv6hdr) + fq->len - sizeof(struct frag_hdr); | |
609 | if (payload_len > IPV6_MAXPLEN) | |
610 | goto out_oversize; | |
611 | ||
612 | /* Head of list must not be cloned. */ | |
613 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
614 | goto out_oom; | |
615 | ||
616 | /* If the first fragment is fragmented itself, we split | |
617 | * it to two chunks: the first with data and paged part | |
618 | * and the second, holding only fragments. */ | |
619 | if (skb_shinfo(head)->frag_list) { | |
620 | struct sk_buff *clone; | |
621 | int i, plen = 0; | |
622 | ||
623 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
624 | goto out_oom; | |
625 | clone->next = head->next; | |
626 | head->next = clone; | |
627 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
628 | skb_shinfo(head)->frag_list = NULL; | |
629 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
630 | plen += skb_shinfo(head)->frags[i].size; | |
631 | clone->len = clone->data_len = head->data_len - plen; | |
632 | head->data_len -= clone->len; | |
633 | head->len -= clone->len; | |
634 | clone->csum = 0; | |
635 | clone->ip_summed = head->ip_summed; | |
636 | atomic_add(clone->truesize, &ip6_frag_mem); | |
637 | } | |
638 | ||
639 | /* We have to remove fragment header from datagram and to relocate | |
640 | * header in order to calculate ICV correctly. */ | |
641 | nhoff = fq->nhoffset; | |
642 | head->nh.raw[nhoff] = head->h.raw[0]; | |
643 | memmove(head->head + sizeof(struct frag_hdr), head->head, | |
644 | (head->data - head->head) - sizeof(struct frag_hdr)); | |
645 | head->mac.raw += sizeof(struct frag_hdr); | |
646 | head->nh.raw += sizeof(struct frag_hdr); | |
647 | ||
648 | skb_shinfo(head)->frag_list = head->next; | |
649 | head->h.raw = head->data; | |
650 | skb_push(head, head->data - head->nh.raw); | |
651 | atomic_sub(head->truesize, &ip6_frag_mem); | |
652 | ||
653 | for (fp=head->next; fp; fp = fp->next) { | |
654 | head->data_len += fp->len; | |
655 | head->len += fp->len; | |
656 | if (head->ip_summed != fp->ip_summed) | |
657 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 658 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
659 | head->csum = csum_add(head->csum, fp->csum); |
660 | head->truesize += fp->truesize; | |
661 | atomic_sub(fp->truesize, &ip6_frag_mem); | |
662 | } | |
663 | ||
664 | head->next = NULL; | |
665 | head->dev = dev; | |
a61bbcf2 | 666 | skb_set_timestamp(head, &fq->stamp); |
1da177e4 | 667 | head->nh.ipv6h->payload_len = htons(payload_len); |
951dbc8a | 668 | IP6CB(head)->nhoff = nhoff; |
1da177e4 LT |
669 | |
670 | *skb_in = head; | |
671 | ||
672 | /* Yes, and fold redundant checksum back. 8) */ | |
84fa7933 | 673 | if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
674 | head->csum = csum_partial(head->nh.raw, head->h.raw-head->nh.raw, head->csum); |
675 | ||
a11d206d YH |
676 | rcu_read_lock(); |
677 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS); | |
678 | rcu_read_unlock(); | |
1da177e4 | 679 | fq->fragments = NULL; |
1da177e4 LT |
680 | return 1; |
681 | ||
682 | out_oversize: | |
683 | if (net_ratelimit()) | |
684 | printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len); | |
685 | goto out_fail; | |
686 | out_oom: | |
687 | if (net_ratelimit()) | |
688 | printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n"); | |
689 | out_fail: | |
a11d206d YH |
690 | rcu_read_lock(); |
691 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
692 | rcu_read_unlock(); | |
1da177e4 LT |
693 | return -1; |
694 | } | |
695 | ||
951dbc8a | 696 | static int ipv6_frag_rcv(struct sk_buff **skbp) |
1da177e4 LT |
697 | { |
698 | struct sk_buff *skb = *skbp; | |
699 | struct net_device *dev = skb->dev; | |
700 | struct frag_hdr *fhdr; | |
701 | struct frag_queue *fq; | |
702 | struct ipv6hdr *hdr; | |
703 | ||
704 | hdr = skb->nh.ipv6h; | |
705 | ||
a11d206d | 706 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
707 | |
708 | /* Jumbo payload inhibits frag. header */ | |
709 | if (hdr->payload_len==0) { | |
a11d206d | 710 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
1da177e4 LT |
711 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw); |
712 | return -1; | |
713 | } | |
714 | if (!pskb_may_pull(skb, (skb->h.raw-skb->data)+sizeof(struct frag_hdr))) { | |
a11d206d | 715 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
1da177e4 LT |
716 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb->h.raw-skb->nh.raw); |
717 | return -1; | |
718 | } | |
719 | ||
720 | hdr = skb->nh.ipv6h; | |
721 | fhdr = (struct frag_hdr *)skb->h.raw; | |
722 | ||
723 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
724 | /* It is not a fragmented frame */ | |
725 | skb->h.raw += sizeof(struct frag_hdr); | |
a11d206d | 726 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS); |
1da177e4 | 727 | |
951dbc8a | 728 | IP6CB(skb)->nhoff = (u8*)fhdr - skb->nh.raw; |
1da177e4 LT |
729 | return 1; |
730 | } | |
731 | ||
732 | if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh) | |
a11d206d | 733 | ip6_evictor(ip6_dst_idev(skb->dst)); |
1da177e4 | 734 | |
a11d206d YH |
735 | if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr, |
736 | ip6_dst_idev(skb->dst))) != NULL) { | |
1da177e4 LT |
737 | int ret = -1; |
738 | ||
739 | spin_lock(&fq->lock); | |
740 | ||
951dbc8a | 741 | ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff); |
1da177e4 LT |
742 | |
743 | if (fq->last_in == (FIRST_IN|LAST_IN) && | |
744 | fq->meat == fq->len) | |
951dbc8a | 745 | ret = ip6_frag_reasm(fq, skbp, dev); |
1da177e4 LT |
746 | |
747 | spin_unlock(&fq->lock); | |
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 | ||
768 | ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ | |
769 | (jiffies ^ (jiffies >> 6))); | |
770 | ||
771 | init_timer(&ip6_frag_secret_timer); | |
772 | ip6_frag_secret_timer.function = ip6_frag_secret_rebuild; | |
773 | ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval; | |
774 | add_timer(&ip6_frag_secret_timer); | |
775 | } |