blob: 11fffe791fc4f966853cbaeed6d40cde9b1fa0de [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 Viroe69a4adc2006-11-14 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 Viroe69a4adc2006-11-14 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 Viroe69a4adc2006-11-14 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 Viroe69a4adc2006-11-14 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 Viroe69a4adc2006-11-14 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 Emelyanov2588fe12007-10-17 19:44:34 -0700146int ip6_frag_equal(struct inet_frag_queue *q1, struct inet_frag_queue *q2)
147{
148 struct frag_queue *fq1, *fq2;
149
150 fq1 = container_of(q1, struct frag_queue, q);
151 fq2 = container_of(q2, struct frag_queue, q);
152 return (fq1->id == fq2->id &&
153 ipv6_addr_equal(&fq2->saddr, &fq1->saddr) &&
154 ipv6_addr_equal(&fq2->daddr, &fq1->daddr));
155}
156EXPORT_SYMBOL(ip6_frag_equal);
157
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700158int ip6_frag_match(struct inet_frag_queue *q, void *a)
159{
160 struct frag_queue *fq;
161 struct ip6_create_arg *arg = a;
162
163 fq = container_of(q, struct frag_queue, q);
164 return (fq->id == arg->id &&
165 ipv6_addr_equal(&fq->saddr, arg->src) &&
166 ipv6_addr_equal(&fq->daddr, arg->dst));
167}
168EXPORT_SYMBOL(ip6_frag_match);
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170/* Memory Tracking Functions. */
171static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
172{
173 if (work)
174 *work -= skb->truesize;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700175 atomic_sub(skb->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kfree_skb(skb);
177}
178
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700179void ip6_frag_init(struct inet_frag_queue *q, void *a)
180{
181 struct frag_queue *fq = container_of(q, struct frag_queue, q);
182 struct ip6_create_arg *arg = a;
183
184 fq->id = arg->id;
185 ipv6_addr_copy(&fq->saddr, arg->src);
186 ipv6_addr_copy(&fq->daddr, arg->dst);
187}
188EXPORT_SYMBOL(ip6_frag_init);
189
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700190static void ip6_frag_free(struct inet_frag_queue *fq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700192 kfree(container_of(fq, struct frag_queue, q));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
194
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195/* Destruction primitives. */
196
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700197static __inline__ void fq_put(struct frag_queue *fq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700199 inet_frag_put(&fq->q, &ip6_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200}
201
202/* Kill fq entry. It is not destroyed immediately,
203 * because caller (and someone more) holds reference count.
204 */
205static __inline__ void fq_kill(struct frag_queue *fq)
206{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700207 inet_frag_kill(&fq->q, &ip6_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900210static void ip6_evictor(struct inet6_dev *idev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700212 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700214 evicted = inet_frag_evictor(&ip6_frags);
215 if (evicted)
216 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
219static void ip6_frag_expire(unsigned long data)
220{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700221 struct frag_queue *fq;
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900222 struct net_device *dev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Pavel Emelyanove521db92007-10-17 19:45:23 -0700224 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
225
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700226 spin_lock(&fq->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700228 if (fq->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 goto out;
230
231 fq_kill(fq);
232
Eric W. Biederman881d9662007-09-17 11:56:21 -0700233 dev = dev_get_by_index(&init_net, fq->iif);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900234 if (!dev)
235 goto out;
236
237 rcu_read_lock();
238 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
239 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
240 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Ingo Oeser78c784c2006-03-20 23:01:17 -0800242 /* Don't send error if the first segment did not arrive. */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700243 if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
Ingo Oeser78c784c2006-03-20 23:01:17 -0800244 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Ingo Oeser78c784c2006-03-20 23:01:17 -0800246 /*
247 But use as source device on which LAST ARRIVED
248 segment was received. And do not use fq->dev
249 pointer directly, device might already disappeared.
250 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700251 fq->q.fragments->dev = dev;
252 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253out:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900254 if (dev)
255 dev_put(dev);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700256 spin_unlock(&fq->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700257 fq_put(fq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700260static __inline__ struct frag_queue *
261fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
262 struct inet6_dev *idev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700264 struct inet_frag_queue *q;
265 struct ip6_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700266 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700268 arg.id = id;
269 arg.src = src;
270 arg.dst = dst;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700271 hash = ip6qhashfn(id, src, dst);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700272
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700273 q = inet_frag_find(&ip6_frags, &arg, hash);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700274 if (q == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 goto oom;
276
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700277 return container_of(q, struct frag_queue, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279oom:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900280 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return NULL;
282}
283
Herbert Xuf61944e2007-10-15 01:28:47 -0700284static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 struct frag_hdr *fhdr, int nhoff)
286{
287 struct sk_buff *prev, *next;
Herbert Xuf61944e2007-10-15 01:28:47 -0700288 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 int offset, end;
290
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700291 if (fq->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 goto err;
293
294 offset = ntohs(fhdr->frag_off) & ~0x7;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700295 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
296 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
298 if ((unsigned int)end > IPV6_MAXPLEN) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900299 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
300 IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700301 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
302 ((u8 *)&fhdr->frag_off -
303 skb_network_header(skb)));
Herbert Xuf61944e2007-10-15 01:28:47 -0700304 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 }
306
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700307 if (skb->ip_summed == CHECKSUM_COMPLETE) {
308 const unsigned char *nh = skb_network_header(skb);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900309 skb->csum = csum_sub(skb->csum,
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700310 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
311 0));
312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
314 /* Is this the final fragment? */
315 if (!(fhdr->frag_off & htons(IP6_MF))) {
316 /* If we already have some bits beyond end
317 * or have different end, the segment is corrupted.
318 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700319 if (end < fq->q.len ||
320 ((fq->q.last_in & LAST_IN) && end != fq->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700322 fq->q.last_in |= LAST_IN;
323 fq->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 } else {
325 /* Check if the fragment is rounded to 8 bytes.
326 * Required by the RFC.
327 */
328 if (end & 0x7) {
329 /* RFC2460 says always send parameter problem in
330 * this case. -DaveM
331 */
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900332 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
333 IPSTATS_MIB_INHDRERRORS);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900334 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 offsetof(struct ipv6hdr, payload_len));
Herbert Xuf61944e2007-10-15 01:28:47 -0700336 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700338 if (end > fq->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 /* Some bits beyond end -> corruption. */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700340 if (fq->q.last_in & LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700342 fq->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 }
344 }
345
346 if (end == offset)
347 goto err;
348
349 /* Point into the IP datagram 'data' part. */
350 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
351 goto err;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900352
Stephen Hemminger42ca89c2005-09-08 12:57:43 -0700353 if (pskb_trim_rcsum(skb, end - offset))
354 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
356 /* Find out which fragments are in front and at the back of us
357 * in the chain of fragments so far. We must know where to put
358 * this fragment, right?
359 */
360 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700361 for(next = fq->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 if (FRAG6_CB(next)->offset >= offset)
363 break; /* bingo! */
364 prev = next;
365 }
366
367 /* We found where to put this one. Check for overlap with
368 * preceding fragment, and, if needed, align things so that
369 * any overlaps are eliminated.
370 */
371 if (prev) {
372 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
373
374 if (i > 0) {
375 offset += i;
376 if (end <= offset)
377 goto err;
378 if (!pskb_pull(skb, i))
379 goto err;
380 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
381 skb->ip_summed = CHECKSUM_NONE;
382 }
383 }
384
385 /* Look for overlap with succeeding segments.
386 * If we can merge fragments, do it.
387 */
388 while (next && FRAG6_CB(next)->offset < end) {
389 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
390
391 if (i < next->len) {
392 /* Eat head of the next overlapped fragment
393 * and leave the loop. The next ones cannot overlap.
394 */
395 if (!pskb_pull(next, i))
396 goto err;
397 FRAG6_CB(next)->offset += i; /* next fragment */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700398 fq->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 if (next->ip_summed != CHECKSUM_UNNECESSARY)
400 next->ip_summed = CHECKSUM_NONE;
401 break;
402 } else {
403 struct sk_buff *free_it = next;
404
405 /* Old fragment is completely overridden with
406 * new one drop it.
407 */
408 next = next->next;
409
410 if (prev)
411 prev->next = next;
412 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700413 fq->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700415 fq->q.meat -= free_it->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 frag_kfree_skb(free_it, NULL);
417 }
418 }
419
420 FRAG6_CB(skb)->offset = offset;
421
422 /* Insert this fragment in the chain of fragments. */
423 skb->next = next;
424 if (prev)
425 prev->next = skb;
426 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700427 fq->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Herbert Xuf61944e2007-10-15 01:28:47 -0700429 dev = skb->dev;
430 if (dev) {
431 fq->iif = dev->ifindex;
432 skb->dev = NULL;
433 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700434 fq->q.stamp = skb->tstamp;
435 fq->q.meat += skb->len;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700436 atomic_add(skb->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
438 /* The first fragment.
439 * nhoffset is obtained from the first fragment, of course.
440 */
441 if (offset == 0) {
442 fq->nhoffset = nhoff;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700443 fq->q.last_in |= FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 }
Herbert Xuf61944e2007-10-15 01:28:47 -0700445
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700446 if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
Herbert Xuf61944e2007-10-15 01:28:47 -0700447 return ip6_frag_reasm(fq, prev, dev);
448
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700449 write_lock(&ip6_frags.lock);
450 list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
451 write_unlock(&ip6_frags.lock);
Herbert Xuf61944e2007-10-15 01:28:47 -0700452 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454err:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900455 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 kfree_skb(skb);
Herbert Xuf61944e2007-10-15 01:28:47 -0700457 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
460/*
461 * Check if this packet is complete.
462 * Returns NULL on failure by any reason, and pointer
463 * to current nexthdr field in reassembled frame.
464 *
465 * It is called with locked fq, and caller must check that
466 * queue is eligible for reassembly i.e. it is not COMPLETE,
467 * the last and the first frames arrived and all the bits are here.
468 */
Herbert Xuf61944e2007-10-15 01:28:47 -0700469static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 struct net_device *dev)
471{
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700472 struct sk_buff *fp, *head = fq->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 int payload_len;
474 unsigned int nhoff;
475
476 fq_kill(fq);
477
Herbert Xuf61944e2007-10-15 01:28:47 -0700478 /* Make the one we just received the head. */
479 if (prev) {
480 head = prev->next;
481 fp = skb_clone(head, GFP_ATOMIC);
482
483 if (!fp)
484 goto out_oom;
485
486 fp->next = head->next;
487 prev->next = fp;
488
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700489 skb_morph(head, fq->q.fragments);
490 head->next = fq->q.fragments->next;
Herbert Xuf61944e2007-10-15 01:28:47 -0700491
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700492 kfree_skb(fq->q.fragments);
493 fq->q.fragments = head;
Herbert Xuf61944e2007-10-15 01:28:47 -0700494 }
495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 BUG_TRAP(head != NULL);
497 BUG_TRAP(FRAG6_CB(head)->offset == 0);
498
499 /* Unfragmented part is taken from the first segment. */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700500 payload_len = ((head->data - skb_network_header(head)) -
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700501 sizeof(struct ipv6hdr) + fq->q.len -
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700502 sizeof(struct frag_hdr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 if (payload_len > IPV6_MAXPLEN)
504 goto out_oversize;
505
506 /* Head of list must not be cloned. */
507 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
508 goto out_oom;
509
510 /* If the first fragment is fragmented itself, we split
511 * it to two chunks: the first with data and paged part
512 * and the second, holding only fragments. */
513 if (skb_shinfo(head)->frag_list) {
514 struct sk_buff *clone;
515 int i, plen = 0;
516
517 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
518 goto out_oom;
519 clone->next = head->next;
520 head->next = clone;
521 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
522 skb_shinfo(head)->frag_list = NULL;
523 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
524 plen += skb_shinfo(head)->frags[i].size;
525 clone->len = clone->data_len = head->data_len - plen;
526 head->data_len -= clone->len;
527 head->len -= clone->len;
528 clone->csum = 0;
529 clone->ip_summed = head->ip_summed;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700530 atomic_add(clone->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 }
532
533 /* We have to remove fragment header from datagram and to relocate
534 * header in order to calculate ICV correctly. */
535 nhoff = fq->nhoffset;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700536 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900537 memmove(head->head + sizeof(struct frag_hdr), head->head,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 (head->data - head->head) - sizeof(struct frag_hdr));
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700539 head->mac_header += sizeof(struct frag_hdr);
540 head->network_header += sizeof(struct frag_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300543 skb_reset_transport_header(head);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700544 skb_push(head, head->data - skb_network_header(head));
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700545 atomic_sub(head->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
547 for (fp=head->next; fp; fp = fp->next) {
548 head->data_len += fp->len;
549 head->len += fp->len;
550 if (head->ip_summed != fp->ip_summed)
551 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700552 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 head->csum = csum_add(head->csum, fp->csum);
554 head->truesize += fp->truesize;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700555 atomic_sub(fp->truesize, &ip6_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 }
557
558 head->next = NULL;
559 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700560 head->tstamp = fq->q.stamp;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700561 ipv6_hdr(head)->payload_len = htons(payload_len);
Patrick McHardy951dbc82006-01-06 23:02:34 -0800562 IP6CB(head)->nhoff = nhoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 /* Yes, and fold redundant checksum back. 8) */
Patrick McHardy84fa7932006-08-29 16:44:56 -0700565 if (head->ip_summed == CHECKSUM_COMPLETE)
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700566 head->csum = csum_partial(skb_network_header(head),
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300567 skb_network_header_len(head),
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700568 head->csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900570 rcu_read_lock();
571 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
572 rcu_read_unlock();
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700573 fq->q.fragments = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 return 1;
575
576out_oversize:
577 if (net_ratelimit())
578 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
579 goto out_fail;
580out_oom:
581 if (net_ratelimit())
582 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
583out_fail:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900584 rcu_read_lock();
585 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
586 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 return -1;
588}
589
Herbert Xue5bbef22007-10-15 12:50:28 -0700590static int ipv6_frag_rcv(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 struct frag_hdr *fhdr;
593 struct frag_queue *fq;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700594 struct ipv6hdr *hdr = ipv6_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900596 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
598 /* Jumbo payload inhibits frag. header */
599 if (hdr->payload_len==0) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900600 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300601 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
602 skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 return -1;
604 }
Arnaldo Carvalho de Meloea2ae172007-04-25 17:55:53 -0700605 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
606 sizeof(struct frag_hdr)))) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900607 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300608 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
609 skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 return -1;
611 }
612
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700613 hdr = ipv6_hdr(skb);
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -0700614 fhdr = (struct frag_hdr *)skb_transport_header(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
616 if (!(fhdr->frag_off & htons(0xFFF9))) {
617 /* It is not a fragmented frame */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700618 skb->transport_header += sizeof(struct frag_hdr);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900619 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700621 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 return 1;
623 }
624
Pavel Emelyanov04128f22007-10-15 02:33:45 -0700625 if (atomic_read(&ip6_frags.mem) > ip6_frags_ctl.high_thresh)
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900626 ip6_evictor(ip6_dst_idev(skb->dst));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900628 if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
629 ip6_dst_idev(skb->dst))) != NULL) {
Herbert Xuf61944e2007-10-15 01:28:47 -0700630 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700632 spin_lock(&fq->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Herbert Xuf61944e2007-10-15 01:28:47 -0700634 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700636 spin_unlock(&fq->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700637 fq_put(fq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 return ret;
639 }
640
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900641 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 kfree_skb(skb);
643 return -1;
644}
645
646static struct inet6_protocol frag_protocol =
647{
648 .handler = ipv6_frag_rcv,
649 .flags = INET6_PROTO_NOPOLICY,
650};
651
652void __init ipv6_frag_init(void)
653{
654 if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
655 printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
656
Pavel Emelyanov04128f22007-10-15 02:33:45 -0700657 ip6_frags.ctl = &ip6_frags_ctl;
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700658 ip6_frags.hashfn = ip6_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700659 ip6_frags.constructor = ip6_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700660 ip6_frags.destructor = ip6_frag_free;
661 ip6_frags.skb_free = NULL;
662 ip6_frags.qsize = sizeof(struct frag_queue);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700663 ip6_frags.match = ip6_frag_match;
Pavel Emelyanov2588fe12007-10-17 19:44:34 -0700664 ip6_frags.equal = ip6_frag_equal;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700665 ip6_frags.frag_expire = ip6_frag_expire;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700666 inet_frags_init(&ip6_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}