blob: bf4173daecbb21aea838be355fa53a069ecec0c0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * IPv6 fragment reassembly
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09003 * Linux INET6 implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * Authors:
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09006 * Pedro Roque <roque@di.fc.ul.pt>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
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
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +090018/*
19 * Fixes:
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 * 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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#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>
Herbert Xuf61944e2007-10-15 01:28:47 -070045#include <linux/skbuff.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47#include <net/sock.h>
48#include <net/snmp.h>
49
50#include <net/ipv6.h>
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +090051#include <net/ip6_route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#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>
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070057#include <net/inet_frag.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059struct 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
72struct frag_queue
73{
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070074 struct inet_frag_queue q;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Al Viroe69a4ad2006-11-14 20:56:00 -080076 __be32 id; /* fragment id */
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 struct in6_addr saddr;
78 struct in6_addr daddr;
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 int iif;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 unsigned int csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 __u16 nhoffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083};
84
Pavel Emelyanov04128f22007-10-15 02:33:45 -070085struct 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
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070092static struct inet_frags ip6_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070094int ip6_frag_nqueues(void)
95{
96 return ip6_frags.nqueues;
97}
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070099int ip6_frag_mem(void)
100{
101 return atomic_read(&ip6_frags.mem);
102}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Herbert Xuf61944e2007-10-15 01:28:47 -0700104static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
105 struct net_device *dev);
106
Zach Brownf6596f92006-04-10 16:05:34 -0700107/*
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 */
Al Viroe69a4ad2006-11-14 20:56:00 -0800111static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 struct in6_addr *daddr)
113{
114 u32 a, b, c;
115
Al Viroe69a4ad2006-11-14 20:56:00 -0800116 a = (__force u32)saddr->s6_addr32[0];
117 b = (__force u32)saddr->s6_addr32[1];
118 c = (__force u32)saddr->s6_addr32[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
120 a += JHASH_GOLDEN_RATIO;
121 b += JHASH_GOLDEN_RATIO;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700122 c += ip6_frags.rnd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 __jhash_mix(a, b, c);
124
Al Viroe69a4ad2006-11-14 20:56:00 -0800125 a += (__force u32)saddr->s6_addr32[3];
126 b += (__force u32)daddr->s6_addr32[0];
127 c += (__force u32)daddr->s6_addr32[1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 __jhash_mix(a, b, c);
129
Al Viroe69a4ad2006-11-14 20:56:00 -0800130 a += (__force u32)daddr->s6_addr32[2];
131 b += (__force u32)daddr->s6_addr32[3];
132 c += (__force u32)id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 __jhash_mix(a, b, c);
134
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700135 return c & (INETFRAGS_HASHSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
137
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700138static unsigned int ip6_hashfn(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700140 struct frag_queue *fq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700142 fq = container_of(q, struct frag_queue, q);
143 return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700146int ip6_frag_match(struct inet_frag_queue *q, void *a)
147{
148 struct frag_queue *fq;
149 struct ip6_create_arg *arg = a;
150
151 fq = container_of(q, struct frag_queue, q);
152 return (fq->id == arg->id &&
153 ipv6_addr_equal(&fq->saddr, arg->src) &&
154 ipv6_addr_equal(&fq->daddr, arg->dst));
155}
156EXPORT_SYMBOL(ip6_frag_match);
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158/* Memory Tracking Functions. */
159static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
160{
161 if (work)
162 *work -= skb->truesize;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700163 atomic_sub(skb->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 kfree_skb(skb);
165}
166
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700167void ip6_frag_init(struct inet_frag_queue *q, void *a)
168{
169 struct frag_queue *fq = container_of(q, struct frag_queue, q);
170 struct ip6_create_arg *arg = a;
171
172 fq->id = arg->id;
173 ipv6_addr_copy(&fq->saddr, arg->src);
174 ipv6_addr_copy(&fq->daddr, arg->dst);
175}
176EXPORT_SYMBOL(ip6_frag_init);
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178/* Destruction primitives. */
179
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700180static __inline__ void fq_put(struct frag_queue *fq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700182 inet_frag_put(&fq->q, &ip6_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185/* Kill fq entry. It is not destroyed immediately,
186 * because caller (and someone more) holds reference count.
187 */
188static __inline__ void fq_kill(struct frag_queue *fq)
189{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700190 inet_frag_kill(&fq->q, &ip6_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191}
192
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900193static void ip6_evictor(struct inet6_dev *idev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700195 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700197 evicted = inet_frag_evictor(&ip6_frags);
198 if (evicted)
199 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200}
201
202static void ip6_frag_expire(unsigned long data)
203{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700204 struct frag_queue *fq;
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900205 struct net_device *dev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Pavel Emelyanove521db92007-10-17 19:45:23 -0700207 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
208
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700209 spin_lock(&fq->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700211 if (fq->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 goto out;
213
214 fq_kill(fq);
215
Eric W. Biederman881d9662007-09-17 11:56:21 -0700216 dev = dev_get_by_index(&init_net, fq->iif);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900217 if (!dev)
218 goto out;
219
220 rcu_read_lock();
221 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
222 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
223 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
Ingo Oeser78c784c2006-03-20 23:01:17 -0800225 /* Don't send error if the first segment did not arrive. */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700226 if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
Ingo Oeser78c784c2006-03-20 23:01:17 -0800227 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Ingo Oeser78c784c2006-03-20 23:01:17 -0800229 /*
230 But use as source device on which LAST ARRIVED
231 segment was received. And do not use fq->dev
232 pointer directly, device might already disappeared.
233 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700234 fq->q.fragments->dev = dev;
235 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236out:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900237 if (dev)
238 dev_put(dev);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700239 spin_unlock(&fq->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700240 fq_put(fq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700243static __inline__ struct frag_queue *
244fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
245 struct inet6_dev *idev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700247 struct inet_frag_queue *q;
248 struct ip6_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700249 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700251 arg.id = id;
252 arg.src = src;
253 arg.dst = dst;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700254 hash = ip6qhashfn(id, src, dst);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700255
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700256 q = inet_frag_find(&ip6_frags, &arg, hash);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700257 if (q == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 goto oom;
259
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700260 return container_of(q, struct frag_queue, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262oom:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900263 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 return NULL;
265}
266
Herbert Xuf61944e2007-10-15 01:28:47 -0700267static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 struct frag_hdr *fhdr, int nhoff)
269{
270 struct sk_buff *prev, *next;
Herbert Xuf61944e2007-10-15 01:28:47 -0700271 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 int offset, end;
273
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700274 if (fq->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 goto err;
276
277 offset = ntohs(fhdr->frag_off) & ~0x7;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700278 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
279 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281 if ((unsigned int)end > IPV6_MAXPLEN) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900282 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
283 IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700284 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
285 ((u8 *)&fhdr->frag_off -
286 skb_network_header(skb)));
Herbert Xuf61944e2007-10-15 01:28:47 -0700287 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 }
289
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700290 if (skb->ip_summed == CHECKSUM_COMPLETE) {
291 const unsigned char *nh = skb_network_header(skb);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900292 skb->csum = csum_sub(skb->csum,
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700293 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
294 0));
295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296
297 /* Is this the final fragment? */
298 if (!(fhdr->frag_off & htons(IP6_MF))) {
299 /* If we already have some bits beyond end
300 * or have different end, the segment is corrupted.
301 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700302 if (end < fq->q.len ||
303 ((fq->q.last_in & LAST_IN) && end != fq->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700305 fq->q.last_in |= LAST_IN;
306 fq->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 } else {
308 /* Check if the fragment is rounded to 8 bytes.
309 * Required by the RFC.
310 */
311 if (end & 0x7) {
312 /* RFC2460 says always send parameter problem in
313 * this case. -DaveM
314 */
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900315 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
316 IPSTATS_MIB_INHDRERRORS);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900317 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 offsetof(struct ipv6hdr, payload_len));
Herbert Xuf61944e2007-10-15 01:28:47 -0700319 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700321 if (end > fq->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 /* Some bits beyond end -> corruption. */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700323 if (fq->q.last_in & LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700325 fq->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 }
327 }
328
329 if (end == offset)
330 goto err;
331
332 /* Point into the IP datagram 'data' part. */
333 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
334 goto err;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900335
Stephen Hemminger42ca89c2005-09-08 12:57:43 -0700336 if (pskb_trim_rcsum(skb, end - offset))
337 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
339 /* Find out which fragments are in front and at the back of us
340 * in the chain of fragments so far. We must know where to put
341 * this fragment, right?
342 */
343 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700344 for(next = fq->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (FRAG6_CB(next)->offset >= offset)
346 break; /* bingo! */
347 prev = next;
348 }
349
350 /* We found where to put this one. Check for overlap with
351 * preceding fragment, and, if needed, align things so that
352 * any overlaps are eliminated.
353 */
354 if (prev) {
355 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
356
357 if (i > 0) {
358 offset += i;
359 if (end <= offset)
360 goto err;
361 if (!pskb_pull(skb, i))
362 goto err;
363 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
364 skb->ip_summed = CHECKSUM_NONE;
365 }
366 }
367
368 /* Look for overlap with succeeding segments.
369 * If we can merge fragments, do it.
370 */
371 while (next && FRAG6_CB(next)->offset < end) {
372 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
373
374 if (i < next->len) {
375 /* Eat head of the next overlapped fragment
376 * and leave the loop. The next ones cannot overlap.
377 */
378 if (!pskb_pull(next, i))
379 goto err;
380 FRAG6_CB(next)->offset += i; /* next fragment */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700381 fq->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (next->ip_summed != CHECKSUM_UNNECESSARY)
383 next->ip_summed = CHECKSUM_NONE;
384 break;
385 } else {
386 struct sk_buff *free_it = next;
387
388 /* Old fragment is completely overridden with
389 * new one drop it.
390 */
391 next = next->next;
392
393 if (prev)
394 prev->next = next;
395 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700396 fq->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700398 fq->q.meat -= free_it->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 frag_kfree_skb(free_it, NULL);
400 }
401 }
402
403 FRAG6_CB(skb)->offset = offset;
404
405 /* Insert this fragment in the chain of fragments. */
406 skb->next = next;
407 if (prev)
408 prev->next = skb;
409 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700410 fq->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Herbert Xuf61944e2007-10-15 01:28:47 -0700412 dev = skb->dev;
413 if (dev) {
414 fq->iif = dev->ifindex;
415 skb->dev = NULL;
416 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700417 fq->q.stamp = skb->tstamp;
418 fq->q.meat += skb->len;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700419 atomic_add(skb->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421 /* The first fragment.
422 * nhoffset is obtained from the first fragment, of course.
423 */
424 if (offset == 0) {
425 fq->nhoffset = nhoff;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700426 fq->q.last_in |= FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 }
Herbert Xuf61944e2007-10-15 01:28:47 -0700428
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700429 if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
Herbert Xuf61944e2007-10-15 01:28:47 -0700430 return ip6_frag_reasm(fq, prev, dev);
431
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700432 write_lock(&ip6_frags.lock);
433 list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
434 write_unlock(&ip6_frags.lock);
Herbert Xuf61944e2007-10-15 01:28:47 -0700435 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437err:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900438 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 kfree_skb(skb);
Herbert Xuf61944e2007-10-15 01:28:47 -0700440 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441}
442
443/*
444 * Check if this packet is complete.
445 * Returns NULL on failure by any reason, and pointer
446 * to current nexthdr field in reassembled frame.
447 *
448 * It is called with locked fq, and caller must check that
449 * queue is eligible for reassembly i.e. it is not COMPLETE,
450 * the last and the first frames arrived and all the bits are here.
451 */
Herbert Xuf61944e2007-10-15 01:28:47 -0700452static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 struct net_device *dev)
454{
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700455 struct sk_buff *fp, *head = fq->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 int payload_len;
457 unsigned int nhoff;
458
459 fq_kill(fq);
460
Herbert Xuf61944e2007-10-15 01:28:47 -0700461 /* Make the one we just received the head. */
462 if (prev) {
463 head = prev->next;
464 fp = skb_clone(head, GFP_ATOMIC);
465
466 if (!fp)
467 goto out_oom;
468
469 fp->next = head->next;
470 prev->next = fp;
471
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700472 skb_morph(head, fq->q.fragments);
473 head->next = fq->q.fragments->next;
Herbert Xuf61944e2007-10-15 01:28:47 -0700474
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700475 kfree_skb(fq->q.fragments);
476 fq->q.fragments = head;
Herbert Xuf61944e2007-10-15 01:28:47 -0700477 }
478
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 BUG_TRAP(head != NULL);
480 BUG_TRAP(FRAG6_CB(head)->offset == 0);
481
482 /* Unfragmented part is taken from the first segment. */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700483 payload_len = ((head->data - skb_network_header(head)) -
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700484 sizeof(struct ipv6hdr) + fq->q.len -
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700485 sizeof(struct frag_hdr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 if (payload_len > IPV6_MAXPLEN)
487 goto out_oversize;
488
489 /* Head of list must not be cloned. */
490 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
491 goto out_oom;
492
493 /* If the first fragment is fragmented itself, we split
494 * it to two chunks: the first with data and paged part
495 * and the second, holding only fragments. */
496 if (skb_shinfo(head)->frag_list) {
497 struct sk_buff *clone;
498 int i, plen = 0;
499
500 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
501 goto out_oom;
502 clone->next = head->next;
503 head->next = clone;
504 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
505 skb_shinfo(head)->frag_list = NULL;
506 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
507 plen += skb_shinfo(head)->frags[i].size;
508 clone->len = clone->data_len = head->data_len - plen;
509 head->data_len -= clone->len;
510 head->len -= clone->len;
511 clone->csum = 0;
512 clone->ip_summed = head->ip_summed;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700513 atomic_add(clone->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 }
515
516 /* We have to remove fragment header from datagram and to relocate
517 * header in order to calculate ICV correctly. */
518 nhoff = fq->nhoffset;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700519 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900520 memmove(head->head + sizeof(struct frag_hdr), head->head,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 (head->data - head->head) - sizeof(struct frag_hdr));
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700522 head->mac_header += sizeof(struct frag_hdr);
523 head->network_header += sizeof(struct frag_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300526 skb_reset_transport_header(head);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700527 skb_push(head, head->data - skb_network_header(head));
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700528 atomic_sub(head->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
530 for (fp=head->next; fp; fp = fp->next) {
531 head->data_len += fp->len;
532 head->len += fp->len;
533 if (head->ip_summed != fp->ip_summed)
534 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700535 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 head->csum = csum_add(head->csum, fp->csum);
537 head->truesize += fp->truesize;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700538 atomic_sub(fp->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
540
541 head->next = NULL;
542 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700543 head->tstamp = fq->q.stamp;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700544 ipv6_hdr(head)->payload_len = htons(payload_len);
Patrick McHardy951dbc82006-01-06 23:02:34 -0800545 IP6CB(head)->nhoff = nhoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 /* Yes, and fold redundant checksum back. 8) */
Patrick McHardy84fa7932006-08-29 16:44:56 -0700548 if (head->ip_summed == CHECKSUM_COMPLETE)
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700549 head->csum = csum_partial(skb_network_header(head),
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300550 skb_network_header_len(head),
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700551 head->csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900553 rcu_read_lock();
554 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
555 rcu_read_unlock();
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700556 fq->q.fragments = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 return 1;
558
559out_oversize:
560 if (net_ratelimit())
561 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
562 goto out_fail;
563out_oom:
564 if (net_ratelimit())
565 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
566out_fail:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900567 rcu_read_lock();
568 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
569 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 return -1;
571}
572
Herbert Xue5bbef22007-10-15 12:50:28 -0700573static int ipv6_frag_rcv(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 struct frag_hdr *fhdr;
576 struct frag_queue *fq;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700577 struct ipv6hdr *hdr = ipv6_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900579 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 /* Jumbo payload inhibits frag. header */
582 if (hdr->payload_len==0) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900583 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300584 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
585 skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return -1;
587 }
Arnaldo Carvalho de Meloea2ae172007-04-25 17:55:53 -0700588 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
589 sizeof(struct frag_hdr)))) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900590 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300591 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
592 skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 return -1;
594 }
595
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700596 hdr = ipv6_hdr(skb);
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -0700597 fhdr = (struct frag_hdr *)skb_transport_header(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599 if (!(fhdr->frag_off & htons(0xFFF9))) {
600 /* It is not a fragmented frame */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700601 skb->transport_header += sizeof(struct frag_hdr);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900602 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700604 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return 1;
606 }
607
Pavel Emelyanov04128f22007-10-15 02:33:45 -0700608 if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh)
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900609 ip6_evictor(ip6_dst_idev(skb->dst));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900611 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
612 ip6_dst_idev(skb->dst))) != NULL) {
Herbert Xuf61944e2007-10-15 01:28:47 -0700613 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700615 spin_lock(&fq->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Herbert Xuf61944e2007-10-15 01:28:47 -0700617 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700619 spin_unlock(&fq->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700620 fq_put(fq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 return ret;
622 }
623
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900624 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 kfree_skb(skb);
626 return -1;
627}
628
629static struct inet6_protocol frag_protocol =
630{
631 .handler = ipv6_frag_rcv,
632 .flags = INET6_PROTO_NOPOLICY,
633};
634
Daniel Lezcano853cbba2007-12-11 02:24:29 -0800635int __init ipv6_frag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Daniel Lezcano853cbba2007-12-11 02:24:29 -0800637 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
Daniel Lezcano853cbba2007-12-11 02:24:29 -0800639 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
640 if (ret)
641 goto out;
Pavel Emelyanov04128f22007-10-15 02:33:45 -0700642 ip6_frags.ctl = &ip6_frags_ctl;
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700643 ip6_frags.hashfn = ip6_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700644 ip6_frags.constructor = ip6_frag_init;
Pavel Emelyanovc9547702007-10-17 19:48:26 -0700645 ip6_frags.destructor = NULL;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700646 ip6_frags.skb_free = NULL;
647 ip6_frags.qsize = sizeof(struct frag_queue);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700648 ip6_frags.match = ip6_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700649 ip6_frags.frag_expire = ip6_frag_expire;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700650 inet_frags_init(&ip6_frags);
Daniel Lezcano853cbba2007-12-11 02:24:29 -0800651out:
652 return ret;
653}
654
655void ipv6_frag_exit(void)
656{
657 inet_frags_fini(&ip6_frags);
658 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}