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> | |
57 | ||
ab32ea5d BH |
58 | int sysctl_ip6frag_high_thresh __read_mostly = 256*1024; |
59 | int sysctl_ip6frag_low_thresh __read_mostly = 192*1024; | |
1da177e4 | 60 | |
ab32ea5d | 61 | int sysctl_ip6frag_time __read_mostly = IPV6_FRAG_TIMEOUT; |
1da177e4 LT |
62 | |
63 | struct ip6frag_skb_cb | |
64 | { | |
65 | struct inet6_skb_parm h; | |
66 | int offset; | |
67 | }; | |
68 | ||
69 | #define FRAG6_CB(skb) ((struct ip6frag_skb_cb*)((skb)->cb)) | |
70 | ||
71 | ||
72 | /* | |
73 | * Equivalent of ipv4 struct ipq | |
74 | */ | |
75 | ||
76 | struct frag_queue | |
77 | { | |
e7c8a41e | 78 | struct hlist_node list; |
1da177e4 LT |
79 | struct list_head lru_list; /* lru list member */ |
80 | ||
e69a4adc | 81 | __be32 id; /* fragment id */ |
1da177e4 LT |
82 | struct in6_addr saddr; |
83 | struct in6_addr daddr; | |
84 | ||
85 | spinlock_t lock; | |
86 | atomic_t refcnt; | |
87 | struct timer_list timer; /* expire timer */ | |
88 | struct sk_buff *fragments; | |
89 | int len; | |
90 | int meat; | |
91 | int iif; | |
b7aa0bf7 | 92 | ktime_t stamp; |
1da177e4 LT |
93 | unsigned int csum; |
94 | __u8 last_in; /* has first/last segment arrived? */ | |
95 | #define COMPLETE 4 | |
96 | #define FIRST_IN 2 | |
97 | #define LAST_IN 1 | |
98 | __u16 nhoffset; | |
1da177e4 LT |
99 | }; |
100 | ||
101 | /* Hash table. */ | |
102 | ||
103 | #define IP6Q_HASHSZ 64 | |
104 | ||
e7c8a41e | 105 | static struct hlist_head ip6_frag_hash[IP6Q_HASHSZ]; |
1da177e4 LT |
106 | static DEFINE_RWLOCK(ip6_frag_lock); |
107 | static u32 ip6_frag_hash_rnd; | |
108 | static LIST_HEAD(ip6_frag_lru_list); | |
109 | int ip6_frag_nqueues = 0; | |
110 | ||
f61944ef HX |
111 | static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev, |
112 | struct net_device *dev); | |
113 | ||
1da177e4 LT |
114 | static __inline__ void __fq_unlink(struct frag_queue *fq) |
115 | { | |
e7c8a41e | 116 | hlist_del(&fq->list); |
1da177e4 LT |
117 | list_del(&fq->lru_list); |
118 | ip6_frag_nqueues--; | |
119 | } | |
120 | ||
121 | static __inline__ void fq_unlink(struct frag_queue *fq) | |
122 | { | |
123 | write_lock(&ip6_frag_lock); | |
124 | __fq_unlink(fq); | |
125 | write_unlock(&ip6_frag_lock); | |
126 | } | |
127 | ||
f6596f9d ZB |
128 | /* |
129 | * callers should be careful not to use the hash value outside the ipfrag_lock | |
130 | * as doing so could race with ipfrag_hash_rnd being recalculated. | |
131 | */ | |
e69a4adc | 132 | static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr, |
1da177e4 LT |
133 | struct in6_addr *daddr) |
134 | { | |
135 | u32 a, b, c; | |
136 | ||
e69a4adc AV |
137 | a = (__force u32)saddr->s6_addr32[0]; |
138 | b = (__force u32)saddr->s6_addr32[1]; | |
139 | c = (__force u32)saddr->s6_addr32[2]; | |
1da177e4 LT |
140 | |
141 | a += JHASH_GOLDEN_RATIO; | |
142 | b += JHASH_GOLDEN_RATIO; | |
143 | c += ip6_frag_hash_rnd; | |
144 | __jhash_mix(a, b, c); | |
145 | ||
e69a4adc AV |
146 | a += (__force u32)saddr->s6_addr32[3]; |
147 | b += (__force u32)daddr->s6_addr32[0]; | |
148 | c += (__force u32)daddr->s6_addr32[1]; | |
1da177e4 LT |
149 | __jhash_mix(a, b, c); |
150 | ||
e69a4adc AV |
151 | a += (__force u32)daddr->s6_addr32[2]; |
152 | b += (__force u32)daddr->s6_addr32[3]; | |
153 | c += (__force u32)id; | |
1da177e4 LT |
154 | __jhash_mix(a, b, c); |
155 | ||
156 | return c & (IP6Q_HASHSZ - 1); | |
157 | } | |
158 | ||
159 | static struct timer_list ip6_frag_secret_timer; | |
ab32ea5d | 160 | int sysctl_ip6frag_secret_interval __read_mostly = 10 * 60 * HZ; |
1da177e4 LT |
161 | |
162 | static void ip6_frag_secret_rebuild(unsigned long dummy) | |
163 | { | |
164 | unsigned long now = jiffies; | |
165 | int i; | |
166 | ||
167 | write_lock(&ip6_frag_lock); | |
168 | get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32)); | |
169 | for (i = 0; i < IP6Q_HASHSZ; i++) { | |
170 | struct frag_queue *q; | |
e7c8a41e | 171 | struct hlist_node *p, *n; |
1da177e4 | 172 | |
e7c8a41e | 173 | hlist_for_each_entry_safe(q, p, n, &ip6_frag_hash[i], list) { |
1da177e4 LT |
174 | unsigned int hval = ip6qhashfn(q->id, |
175 | &q->saddr, | |
176 | &q->daddr); | |
177 | ||
178 | if (hval != i) { | |
e7c8a41e | 179 | hlist_del(&q->list); |
1da177e4 LT |
180 | |
181 | /* Relink to new hash chain. */ | |
e7c8a41e YK |
182 | hlist_add_head(&q->list, |
183 | &ip6_frag_hash[hval]); | |
1da177e4 | 184 | |
e7c8a41e | 185 | } |
1da177e4 LT |
186 | } |
187 | } | |
188 | write_unlock(&ip6_frag_lock); | |
189 | ||
190 | mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval); | |
191 | } | |
192 | ||
193 | atomic_t ip6_frag_mem = ATOMIC_INIT(0); | |
194 | ||
195 | /* Memory Tracking Functions. */ | |
196 | static inline void frag_kfree_skb(struct sk_buff *skb, int *work) | |
197 | { | |
198 | if (work) | |
199 | *work -= skb->truesize; | |
200 | atomic_sub(skb->truesize, &ip6_frag_mem); | |
201 | kfree_skb(skb); | |
202 | } | |
203 | ||
204 | static inline void frag_free_queue(struct frag_queue *fq, int *work) | |
205 | { | |
206 | if (work) | |
207 | *work -= sizeof(struct frag_queue); | |
208 | atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem); | |
209 | kfree(fq); | |
210 | } | |
211 | ||
212 | static inline struct frag_queue *frag_alloc_queue(void) | |
213 | { | |
78c784c4 | 214 | struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC); |
1da177e4 LT |
215 | |
216 | if(!fq) | |
217 | return NULL; | |
218 | atomic_add(sizeof(struct frag_queue), &ip6_frag_mem); | |
219 | return fq; | |
220 | } | |
221 | ||
222 | /* Destruction primitives. */ | |
223 | ||
224 | /* Complete destruction of fq. */ | |
225 | static void ip6_frag_destroy(struct frag_queue *fq, int *work) | |
226 | { | |
227 | struct sk_buff *fp; | |
228 | ||
229 | BUG_TRAP(fq->last_in&COMPLETE); | |
230 | BUG_TRAP(del_timer(&fq->timer) == 0); | |
231 | ||
232 | /* Release all fragment data. */ | |
233 | fp = fq->fragments; | |
234 | while (fp) { | |
235 | struct sk_buff *xp = fp->next; | |
236 | ||
237 | frag_kfree_skb(fp, work); | |
238 | fp = xp; | |
239 | } | |
240 | ||
241 | frag_free_queue(fq, work); | |
242 | } | |
243 | ||
244 | static __inline__ void fq_put(struct frag_queue *fq, int *work) | |
245 | { | |
246 | if (atomic_dec_and_test(&fq->refcnt)) | |
247 | ip6_frag_destroy(fq, work); | |
248 | } | |
249 | ||
250 | /* Kill fq entry. It is not destroyed immediately, | |
251 | * because caller (and someone more) holds reference count. | |
252 | */ | |
253 | static __inline__ void fq_kill(struct frag_queue *fq) | |
254 | { | |
255 | if (del_timer(&fq->timer)) | |
256 | atomic_dec(&fq->refcnt); | |
257 | ||
258 | if (!(fq->last_in & COMPLETE)) { | |
259 | fq_unlink(fq); | |
260 | atomic_dec(&fq->refcnt); | |
261 | fq->last_in |= COMPLETE; | |
262 | } | |
263 | } | |
264 | ||
a11d206d | 265 | static void ip6_evictor(struct inet6_dev *idev) |
1da177e4 LT |
266 | { |
267 | struct frag_queue *fq; | |
268 | struct list_head *tmp; | |
269 | int work; | |
270 | ||
271 | work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh; | |
272 | if (work <= 0) | |
273 | return; | |
274 | ||
275 | while(work > 0) { | |
276 | read_lock(&ip6_frag_lock); | |
277 | if (list_empty(&ip6_frag_lru_list)) { | |
278 | read_unlock(&ip6_frag_lock); | |
279 | return; | |
280 | } | |
281 | tmp = ip6_frag_lru_list.next; | |
282 | fq = list_entry(tmp, struct frag_queue, lru_list); | |
283 | atomic_inc(&fq->refcnt); | |
284 | read_unlock(&ip6_frag_lock); | |
285 | ||
286 | spin_lock(&fq->lock); | |
287 | if (!(fq->last_in&COMPLETE)) | |
288 | fq_kill(fq); | |
289 | spin_unlock(&fq->lock); | |
290 | ||
291 | fq_put(fq, &work); | |
a11d206d | 292 | IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
293 | } |
294 | } | |
295 | ||
296 | static void ip6_frag_expire(unsigned long data) | |
297 | { | |
298 | struct frag_queue *fq = (struct frag_queue *) data; | |
a11d206d | 299 | struct net_device *dev = NULL; |
1da177e4 LT |
300 | |
301 | spin_lock(&fq->lock); | |
302 | ||
303 | if (fq->last_in & COMPLETE) | |
304 | goto out; | |
305 | ||
306 | fq_kill(fq); | |
307 | ||
881d966b | 308 | dev = dev_get_by_index(&init_net, fq->iif); |
a11d206d YH |
309 | if (!dev) |
310 | goto out; | |
311 | ||
312 | rcu_read_lock(); | |
313 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT); | |
314 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
315 | rcu_read_unlock(); | |
1da177e4 | 316 | |
78c784c4 IO |
317 | /* Don't send error if the first segment did not arrive. */ |
318 | if (!(fq->last_in&FIRST_IN) || !fq->fragments) | |
319 | goto out; | |
320 | ||
78c784c4 IO |
321 | /* |
322 | But use as source device on which LAST ARRIVED | |
323 | segment was received. And do not use fq->dev | |
324 | pointer directly, device might already disappeared. | |
325 | */ | |
326 | fq->fragments->dev = dev; | |
327 | icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev); | |
1da177e4 | 328 | out: |
a11d206d YH |
329 | if (dev) |
330 | dev_put(dev); | |
1da177e4 LT |
331 | spin_unlock(&fq->lock); |
332 | fq_put(fq, NULL); | |
333 | } | |
334 | ||
335 | /* Creation primitives. */ | |
336 | ||
337 | ||
f6596f9d | 338 | static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in) |
1da177e4 LT |
339 | { |
340 | struct frag_queue *fq; | |
f6596f9d | 341 | unsigned int hash; |
e7c8a41e YK |
342 | #ifdef CONFIG_SMP |
343 | struct hlist_node *n; | |
344 | #endif | |
1da177e4 LT |
345 | |
346 | write_lock(&ip6_frag_lock); | |
f6596f9d | 347 | hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr); |
1da177e4 | 348 | #ifdef CONFIG_SMP |
e7c8a41e | 349 | hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) { |
1ab1457c | 350 | if (fq->id == fq_in->id && |
1da177e4 LT |
351 | ipv6_addr_equal(&fq_in->saddr, &fq->saddr) && |
352 | ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) { | |
353 | atomic_inc(&fq->refcnt); | |
354 | write_unlock(&ip6_frag_lock); | |
355 | fq_in->last_in |= COMPLETE; | |
356 | fq_put(fq_in, NULL); | |
357 | return fq; | |
358 | } | |
359 | } | |
360 | #endif | |
361 | fq = fq_in; | |
362 | ||
363 | if (!mod_timer(&fq->timer, jiffies + sysctl_ip6frag_time)) | |
364 | atomic_inc(&fq->refcnt); | |
365 | ||
366 | atomic_inc(&fq->refcnt); | |
e7c8a41e | 367 | hlist_add_head(&fq->list, &ip6_frag_hash[hash]); |
1da177e4 LT |
368 | INIT_LIST_HEAD(&fq->lru_list); |
369 | list_add_tail(&fq->lru_list, &ip6_frag_lru_list); | |
370 | ip6_frag_nqueues++; | |
371 | write_unlock(&ip6_frag_lock); | |
372 | return fq; | |
373 | } | |
374 | ||
375 | ||
376 | static struct frag_queue * | |
e69a4adc | 377 | ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst, |
a11d206d | 378 | struct inet6_dev *idev) |
1da177e4 LT |
379 | { |
380 | struct frag_queue *fq; | |
381 | ||
382 | if ((fq = frag_alloc_queue()) == NULL) | |
383 | goto oom; | |
384 | ||
1da177e4 LT |
385 | fq->id = id; |
386 | ipv6_addr_copy(&fq->saddr, src); | |
387 | ipv6_addr_copy(&fq->daddr, dst); | |
388 | ||
389 | init_timer(&fq->timer); | |
390 | fq->timer.function = ip6_frag_expire; | |
391 | fq->timer.data = (long) fq; | |
392 | spin_lock_init(&fq->lock); | |
393 | atomic_set(&fq->refcnt, 1); | |
394 | ||
f6596f9d | 395 | return ip6_frag_intern(fq); |
1da177e4 LT |
396 | |
397 | oom: | |
a11d206d | 398 | IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
399 | return NULL; |
400 | } | |
401 | ||
402 | static __inline__ struct frag_queue * | |
e69a4adc | 403 | fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst, |
a11d206d | 404 | struct inet6_dev *idev) |
1da177e4 LT |
405 | { |
406 | struct frag_queue *fq; | |
e7c8a41e | 407 | struct hlist_node *n; |
f6596f9d | 408 | unsigned int hash; |
1da177e4 LT |
409 | |
410 | read_lock(&ip6_frag_lock); | |
f6596f9d | 411 | hash = ip6qhashfn(id, src, dst); |
e7c8a41e | 412 | hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], list) { |
1ab1457c | 413 | if (fq->id == id && |
1da177e4 LT |
414 | ipv6_addr_equal(src, &fq->saddr) && |
415 | ipv6_addr_equal(dst, &fq->daddr)) { | |
416 | atomic_inc(&fq->refcnt); | |
417 | read_unlock(&ip6_frag_lock); | |
418 | return fq; | |
419 | } | |
420 | } | |
421 | read_unlock(&ip6_frag_lock); | |
422 | ||
a11d206d | 423 | return ip6_frag_create(id, src, dst, idev); |
1da177e4 LT |
424 | } |
425 | ||
426 | ||
f61944ef | 427 | static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb, |
1da177e4 LT |
428 | struct frag_hdr *fhdr, int nhoff) |
429 | { | |
430 | struct sk_buff *prev, *next; | |
f61944ef | 431 | struct net_device *dev; |
1da177e4 LT |
432 | int offset, end; |
433 | ||
434 | if (fq->last_in & COMPLETE) | |
435 | goto err; | |
436 | ||
437 | offset = ntohs(fhdr->frag_off) & ~0x7; | |
0660e03f ACM |
438 | end = offset + (ntohs(ipv6_hdr(skb)->payload_len) - |
439 | ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1))); | |
1da177e4 LT |
440 | |
441 | if ((unsigned int)end > IPV6_MAXPLEN) { | |
a11d206d YH |
442 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
443 | IPSTATS_MIB_INHDRERRORS); | |
d56f90a7 ACM |
444 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
445 | ((u8 *)&fhdr->frag_off - | |
446 | skb_network_header(skb))); | |
f61944ef | 447 | return -1; |
1da177e4 LT |
448 | } |
449 | ||
d56f90a7 ACM |
450 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
451 | const unsigned char *nh = skb_network_header(skb); | |
1ab1457c | 452 | skb->csum = csum_sub(skb->csum, |
d56f90a7 ACM |
453 | csum_partial(nh, (u8 *)(fhdr + 1) - nh, |
454 | 0)); | |
455 | } | |
1da177e4 LT |
456 | |
457 | /* Is this the final fragment? */ | |
458 | if (!(fhdr->frag_off & htons(IP6_MF))) { | |
459 | /* If we already have some bits beyond end | |
460 | * or have different end, the segment is corrupted. | |
461 | */ | |
462 | if (end < fq->len || | |
463 | ((fq->last_in & LAST_IN) && end != fq->len)) | |
464 | goto err; | |
465 | fq->last_in |= LAST_IN; | |
466 | fq->len = end; | |
467 | } else { | |
468 | /* Check if the fragment is rounded to 8 bytes. | |
469 | * Required by the RFC. | |
470 | */ | |
471 | if (end & 0x7) { | |
472 | /* RFC2460 says always send parameter problem in | |
473 | * this case. -DaveM | |
474 | */ | |
a11d206d YH |
475 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), |
476 | IPSTATS_MIB_INHDRERRORS); | |
1ab1457c | 477 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
1da177e4 | 478 | offsetof(struct ipv6hdr, payload_len)); |
f61944ef | 479 | return -1; |
1da177e4 LT |
480 | } |
481 | if (end > fq->len) { | |
482 | /* Some bits beyond end -> corruption. */ | |
483 | if (fq->last_in & LAST_IN) | |
484 | goto err; | |
485 | fq->len = end; | |
486 | } | |
487 | } | |
488 | ||
489 | if (end == offset) | |
490 | goto err; | |
491 | ||
492 | /* Point into the IP datagram 'data' part. */ | |
493 | if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) | |
494 | goto err; | |
1ab1457c | 495 | |
42ca89c1 SH |
496 | if (pskb_trim_rcsum(skb, end - offset)) |
497 | goto err; | |
1da177e4 LT |
498 | |
499 | /* Find out which fragments are in front and at the back of us | |
500 | * in the chain of fragments so far. We must know where to put | |
501 | * this fragment, right? | |
502 | */ | |
503 | prev = NULL; | |
504 | for(next = fq->fragments; next != NULL; next = next->next) { | |
505 | if (FRAG6_CB(next)->offset >= offset) | |
506 | break; /* bingo! */ | |
507 | prev = next; | |
508 | } | |
509 | ||
510 | /* We found where to put this one. Check for overlap with | |
511 | * preceding fragment, and, if needed, align things so that | |
512 | * any overlaps are eliminated. | |
513 | */ | |
514 | if (prev) { | |
515 | int i = (FRAG6_CB(prev)->offset + prev->len) - offset; | |
516 | ||
517 | if (i > 0) { | |
518 | offset += i; | |
519 | if (end <= offset) | |
520 | goto err; | |
521 | if (!pskb_pull(skb, i)) | |
522 | goto err; | |
523 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) | |
524 | skb->ip_summed = CHECKSUM_NONE; | |
525 | } | |
526 | } | |
527 | ||
528 | /* Look for overlap with succeeding segments. | |
529 | * If we can merge fragments, do it. | |
530 | */ | |
531 | while (next && FRAG6_CB(next)->offset < end) { | |
532 | int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */ | |
533 | ||
534 | if (i < next->len) { | |
535 | /* Eat head of the next overlapped fragment | |
536 | * and leave the loop. The next ones cannot overlap. | |
537 | */ | |
538 | if (!pskb_pull(next, i)) | |
539 | goto err; | |
540 | FRAG6_CB(next)->offset += i; /* next fragment */ | |
541 | fq->meat -= i; | |
542 | if (next->ip_summed != CHECKSUM_UNNECESSARY) | |
543 | next->ip_summed = CHECKSUM_NONE; | |
544 | break; | |
545 | } else { | |
546 | struct sk_buff *free_it = next; | |
547 | ||
548 | /* Old fragment is completely overridden with | |
549 | * new one drop it. | |
550 | */ | |
551 | next = next->next; | |
552 | ||
553 | if (prev) | |
554 | prev->next = next; | |
555 | else | |
556 | fq->fragments = next; | |
557 | ||
558 | fq->meat -= free_it->len; | |
559 | frag_kfree_skb(free_it, NULL); | |
560 | } | |
561 | } | |
562 | ||
563 | FRAG6_CB(skb)->offset = offset; | |
564 | ||
565 | /* Insert this fragment in the chain of fragments. */ | |
566 | skb->next = next; | |
567 | if (prev) | |
568 | prev->next = skb; | |
569 | else | |
570 | fq->fragments = skb; | |
571 | ||
f61944ef HX |
572 | dev = skb->dev; |
573 | if (dev) { | |
574 | fq->iif = dev->ifindex; | |
575 | skb->dev = NULL; | |
576 | } | |
b7aa0bf7 | 577 | fq->stamp = skb->tstamp; |
1da177e4 LT |
578 | fq->meat += skb->len; |
579 | atomic_add(skb->truesize, &ip6_frag_mem); | |
580 | ||
581 | /* The first fragment. | |
582 | * nhoffset is obtained from the first fragment, of course. | |
583 | */ | |
584 | if (offset == 0) { | |
585 | fq->nhoffset = nhoff; | |
586 | fq->last_in |= FIRST_IN; | |
587 | } | |
f61944ef HX |
588 | |
589 | if (fq->last_in == (FIRST_IN | LAST_IN) && fq->meat == fq->len) | |
590 | return ip6_frag_reasm(fq, prev, dev); | |
591 | ||
1da177e4 LT |
592 | write_lock(&ip6_frag_lock); |
593 | list_move_tail(&fq->lru_list, &ip6_frag_lru_list); | |
594 | write_unlock(&ip6_frag_lock); | |
f61944ef | 595 | return -1; |
1da177e4 LT |
596 | |
597 | err: | |
a11d206d | 598 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 | 599 | kfree_skb(skb); |
f61944ef | 600 | return -1; |
1da177e4 LT |
601 | } |
602 | ||
603 | /* | |
604 | * Check if this packet is complete. | |
605 | * Returns NULL on failure by any reason, and pointer | |
606 | * to current nexthdr field in reassembled frame. | |
607 | * | |
608 | * It is called with locked fq, and caller must check that | |
609 | * queue is eligible for reassembly i.e. it is not COMPLETE, | |
610 | * the last and the first frames arrived and all the bits are here. | |
611 | */ | |
f61944ef | 612 | static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev, |
1da177e4 LT |
613 | struct net_device *dev) |
614 | { | |
615 | struct sk_buff *fp, *head = fq->fragments; | |
616 | int payload_len; | |
617 | unsigned int nhoff; | |
618 | ||
619 | fq_kill(fq); | |
620 | ||
f61944ef HX |
621 | /* Make the one we just received the head. */ |
622 | if (prev) { | |
623 | head = prev->next; | |
624 | fp = skb_clone(head, GFP_ATOMIC); | |
625 | ||
626 | if (!fp) | |
627 | goto out_oom; | |
628 | ||
629 | fp->next = head->next; | |
630 | prev->next = fp; | |
631 | ||
632 | skb_morph(head, fq->fragments); | |
633 | head->next = fq->fragments->next; | |
634 | ||
635 | kfree_skb(fq->fragments); | |
636 | fq->fragments = head; | |
637 | } | |
638 | ||
1da177e4 LT |
639 | BUG_TRAP(head != NULL); |
640 | BUG_TRAP(FRAG6_CB(head)->offset == 0); | |
641 | ||
642 | /* Unfragmented part is taken from the first segment. */ | |
d56f90a7 ACM |
643 | payload_len = ((head->data - skb_network_header(head)) - |
644 | sizeof(struct ipv6hdr) + fq->len - | |
645 | sizeof(struct frag_hdr)); | |
1da177e4 LT |
646 | if (payload_len > IPV6_MAXPLEN) |
647 | goto out_oversize; | |
648 | ||
649 | /* Head of list must not be cloned. */ | |
650 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) | |
651 | goto out_oom; | |
652 | ||
653 | /* If the first fragment is fragmented itself, we split | |
654 | * it to two chunks: the first with data and paged part | |
655 | * and the second, holding only fragments. */ | |
656 | if (skb_shinfo(head)->frag_list) { | |
657 | struct sk_buff *clone; | |
658 | int i, plen = 0; | |
659 | ||
660 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) | |
661 | goto out_oom; | |
662 | clone->next = head->next; | |
663 | head->next = clone; | |
664 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; | |
665 | skb_shinfo(head)->frag_list = NULL; | |
666 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) | |
667 | plen += skb_shinfo(head)->frags[i].size; | |
668 | clone->len = clone->data_len = head->data_len - plen; | |
669 | head->data_len -= clone->len; | |
670 | head->len -= clone->len; | |
671 | clone->csum = 0; | |
672 | clone->ip_summed = head->ip_summed; | |
673 | atomic_add(clone->truesize, &ip6_frag_mem); | |
674 | } | |
675 | ||
676 | /* We have to remove fragment header from datagram and to relocate | |
677 | * header in order to calculate ICV correctly. */ | |
678 | nhoff = fq->nhoffset; | |
b0e380b1 | 679 | skb_network_header(head)[nhoff] = skb_transport_header(head)[0]; |
1ab1457c | 680 | memmove(head->head + sizeof(struct frag_hdr), head->head, |
1da177e4 | 681 | (head->data - head->head) - sizeof(struct frag_hdr)); |
b0e380b1 ACM |
682 | head->mac_header += sizeof(struct frag_hdr); |
683 | head->network_header += sizeof(struct frag_hdr); | |
1da177e4 LT |
684 | |
685 | skb_shinfo(head)->frag_list = head->next; | |
badff6d0 | 686 | skb_reset_transport_header(head); |
d56f90a7 | 687 | skb_push(head, head->data - skb_network_header(head)); |
1da177e4 LT |
688 | atomic_sub(head->truesize, &ip6_frag_mem); |
689 | ||
690 | for (fp=head->next; fp; fp = fp->next) { | |
691 | head->data_len += fp->len; | |
692 | head->len += fp->len; | |
693 | if (head->ip_summed != fp->ip_summed) | |
694 | head->ip_summed = CHECKSUM_NONE; | |
84fa7933 | 695 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
1da177e4 LT |
696 | head->csum = csum_add(head->csum, fp->csum); |
697 | head->truesize += fp->truesize; | |
698 | atomic_sub(fp->truesize, &ip6_frag_mem); | |
699 | } | |
700 | ||
701 | head->next = NULL; | |
702 | head->dev = dev; | |
b7aa0bf7 | 703 | head->tstamp = fq->stamp; |
0660e03f | 704 | ipv6_hdr(head)->payload_len = htons(payload_len); |
951dbc8a | 705 | IP6CB(head)->nhoff = nhoff; |
1da177e4 | 706 | |
1da177e4 | 707 | /* Yes, and fold redundant checksum back. 8) */ |
84fa7933 | 708 | if (head->ip_summed == CHECKSUM_COMPLETE) |
d56f90a7 | 709 | head->csum = csum_partial(skb_network_header(head), |
cfe1fc77 | 710 | skb_network_header_len(head), |
d56f90a7 | 711 | head->csum); |
1da177e4 | 712 | |
a11d206d YH |
713 | rcu_read_lock(); |
714 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS); | |
715 | rcu_read_unlock(); | |
1da177e4 | 716 | fq->fragments = NULL; |
1da177e4 LT |
717 | return 1; |
718 | ||
719 | out_oversize: | |
720 | if (net_ratelimit()) | |
721 | printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len); | |
722 | goto out_fail; | |
723 | out_oom: | |
724 | if (net_ratelimit()) | |
725 | printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n"); | |
726 | out_fail: | |
a11d206d YH |
727 | rcu_read_lock(); |
728 | IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS); | |
729 | rcu_read_unlock(); | |
1da177e4 LT |
730 | return -1; |
731 | } | |
732 | ||
951dbc8a | 733 | static int ipv6_frag_rcv(struct sk_buff **skbp) |
1da177e4 | 734 | { |
1ab1457c | 735 | struct sk_buff *skb = *skbp; |
1da177e4 LT |
736 | struct frag_hdr *fhdr; |
737 | struct frag_queue *fq; | |
0660e03f | 738 | struct ipv6hdr *hdr = ipv6_hdr(skb); |
1da177e4 | 739 | |
a11d206d | 740 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS); |
1da177e4 LT |
741 | |
742 | /* Jumbo payload inhibits frag. header */ | |
743 | if (hdr->payload_len==0) { | |
a11d206d | 744 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
cfe1fc77 ACM |
745 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
746 | skb_network_header_len(skb)); | |
1da177e4 LT |
747 | return -1; |
748 | } | |
ea2ae17d ACM |
749 | if (!pskb_may_pull(skb, (skb_transport_offset(skb) + |
750 | sizeof(struct frag_hdr)))) { | |
a11d206d | 751 | IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS); |
cfe1fc77 ACM |
752 | icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, |
753 | skb_network_header_len(skb)); | |
1da177e4 LT |
754 | return -1; |
755 | } | |
756 | ||
0660e03f | 757 | hdr = ipv6_hdr(skb); |
9c70220b | 758 | fhdr = (struct frag_hdr *)skb_transport_header(skb); |
1da177e4 LT |
759 | |
760 | if (!(fhdr->frag_off & htons(0xFFF9))) { | |
761 | /* It is not a fragmented frame */ | |
b0e380b1 | 762 | skb->transport_header += sizeof(struct frag_hdr); |
a11d206d | 763 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS); |
1da177e4 | 764 | |
d56f90a7 | 765 | IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb); |
1da177e4 LT |
766 | return 1; |
767 | } | |
768 | ||
769 | if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh) | |
a11d206d | 770 | ip6_evictor(ip6_dst_idev(skb->dst)); |
1da177e4 | 771 | |
a11d206d YH |
772 | if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr, |
773 | ip6_dst_idev(skb->dst))) != NULL) { | |
f61944ef | 774 | int ret; |
1da177e4 LT |
775 | |
776 | spin_lock(&fq->lock); | |
777 | ||
f61944ef | 778 | ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff); |
1da177e4 LT |
779 | |
780 | spin_unlock(&fq->lock); | |
781 | fq_put(fq, NULL); | |
782 | return ret; | |
783 | } | |
784 | ||
a11d206d | 785 | IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS); |
1da177e4 LT |
786 | kfree_skb(skb); |
787 | return -1; | |
788 | } | |
789 | ||
790 | static struct inet6_protocol frag_protocol = | |
791 | { | |
792 | .handler = ipv6_frag_rcv, | |
793 | .flags = INET6_PROTO_NOPOLICY, | |
794 | }; | |
795 | ||
796 | void __init ipv6_frag_init(void) | |
797 | { | |
798 | if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0) | |
799 | printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n"); | |
800 | ||
801 | ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ | |
802 | (jiffies ^ (jiffies >> 6))); | |
803 | ||
804 | init_timer(&ip6_frag_secret_timer); | |
805 | ip6_frag_secret_timer.function = ip6_frag_secret_rebuild; | |
806 | ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval; | |
807 | add_timer(&ip6_frag_secret_timer); | |
808 | } |