blob: 85f3fa3822308fd6e26e9372c153c90e4cfb2657 [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 Emelyanov7eb95152007-10-15 02:31:52 -070085static struct inet_frags ip6_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -080087int ip6_frag_nqueues(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070088{
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -080089 return net->ipv6.frags.nqueues;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070090}
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080092int ip6_frag_mem(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070093{
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080094 return atomic_read(&net->ipv6.frags.mem);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070095}
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Herbert Xuf61944e2007-10-15 01:28:47 -070097static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
98 struct net_device *dev);
99
Zach Brownf6596f92006-04-10 16:05:34 -0700100/*
101 * callers should be careful not to use the hash value outside the ipfrag_lock
102 * as doing so could race with ipfrag_hash_rnd being recalculated.
103 */
Al Viroe69a4adc2006-11-14 20:56:00 -0800104static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 struct in6_addr *daddr)
106{
107 u32 a, b, c;
108
Al Viroe69a4adc2006-11-14 20:56:00 -0800109 a = (__force u32)saddr->s6_addr32[0];
110 b = (__force u32)saddr->s6_addr32[1];
111 c = (__force u32)saddr->s6_addr32[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
113 a += JHASH_GOLDEN_RATIO;
114 b += JHASH_GOLDEN_RATIO;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700115 c += ip6_frags.rnd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 __jhash_mix(a, b, c);
117
Al Viroe69a4adc2006-11-14 20:56:00 -0800118 a += (__force u32)saddr->s6_addr32[3];
119 b += (__force u32)daddr->s6_addr32[0];
120 c += (__force u32)daddr->s6_addr32[1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 __jhash_mix(a, b, c);
122
Al Viroe69a4adc2006-11-14 20:56:00 -0800123 a += (__force u32)daddr->s6_addr32[2];
124 b += (__force u32)daddr->s6_addr32[3];
125 c += (__force u32)id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 __jhash_mix(a, b, c);
127
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700128 return c & (INETFRAGS_HASHSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129}
130
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700131static unsigned int ip6_hashfn(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700133 struct frag_queue *fq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700135 fq = container_of(q, struct frag_queue, q);
136 return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700139int ip6_frag_match(struct inet_frag_queue *q, void *a)
140{
141 struct frag_queue *fq;
142 struct ip6_create_arg *arg = a;
143
144 fq = container_of(q, struct frag_queue, q);
145 return (fq->id == arg->id &&
146 ipv6_addr_equal(&fq->saddr, arg->src) &&
147 ipv6_addr_equal(&fq->daddr, arg->dst));
148}
149EXPORT_SYMBOL(ip6_frag_match);
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151/* Memory Tracking Functions. */
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800152static inline void frag_kfree_skb(struct netns_frags *nf,
153 struct sk_buff *skb, int *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
155 if (work)
156 *work -= skb->truesize;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800157 atomic_sub(skb->truesize, &nf->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 kfree_skb(skb);
159}
160
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700161void ip6_frag_init(struct inet_frag_queue *q, void *a)
162{
163 struct frag_queue *fq = container_of(q, struct frag_queue, q);
164 struct ip6_create_arg *arg = a;
165
166 fq->id = arg->id;
167 ipv6_addr_copy(&fq->saddr, arg->src);
168 ipv6_addr_copy(&fq->daddr, arg->dst);
169}
170EXPORT_SYMBOL(ip6_frag_init);
171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172/* Destruction primitives. */
173
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700174static __inline__ void fq_put(struct frag_queue *fq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700176 inet_frag_put(&fq->q, &ip6_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
179/* Kill fq entry. It is not destroyed immediately,
180 * because caller (and someone more) holds reference count.
181 */
182static __inline__ void fq_kill(struct frag_queue *fq)
183{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700184 inet_frag_kill(&fq->q, &ip6_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185}
186
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800187static void ip6_evictor(struct net *net, struct inet6_dev *idev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700189 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800191 evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700192 if (evicted)
193 IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194}
195
196static void ip6_frag_expire(unsigned long data)
197{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700198 struct frag_queue *fq;
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900199 struct net_device *dev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Pavel Emelyanove521db92007-10-17 19:45:23 -0700201 fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
202
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700203 spin_lock(&fq->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700205 if (fq->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 goto out;
207
208 fq_kill(fq);
209
Eric W. Biederman881d9662007-09-17 11:56:21 -0700210 dev = dev_get_by_index(&init_net, fq->iif);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900211 if (!dev)
212 goto out;
213
214 rcu_read_lock();
215 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
216 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
217 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Ingo Oeser78c784c2006-03-20 23:01:17 -0800219 /* Don't send error if the first segment did not arrive. */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700220 if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
Ingo Oeser78c784c2006-03-20 23:01:17 -0800221 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Ingo Oeser78c784c2006-03-20 23:01:17 -0800223 /*
224 But use as source device on which LAST ARRIVED
225 segment was received. And do not use fq->dev
226 pointer directly, device might already disappeared.
227 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700228 fq->q.fragments->dev = dev;
229 icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230out:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900231 if (dev)
232 dev_put(dev);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700233 spin_unlock(&fq->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700234 fq_put(fq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
236
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700237static __inline__ struct frag_queue *
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800238fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700239 struct inet6_dev *idev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700241 struct inet_frag_queue *q;
242 struct ip6_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700243 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700245 arg.id = id;
246 arg.src = src;
247 arg.dst = dst;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700248 hash = ip6qhashfn(id, src, dst);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700249
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800250 q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700251 if (q == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 goto oom;
253
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700254 return container_of(q, struct frag_queue, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256oom:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900257 IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 return NULL;
259}
260
Herbert Xuf61944e2007-10-15 01:28:47 -0700261static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 struct frag_hdr *fhdr, int nhoff)
263{
264 struct sk_buff *prev, *next;
Herbert Xuf61944e2007-10-15 01:28:47 -0700265 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 int offset, end;
267
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700268 if (fq->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 goto err;
270
271 offset = ntohs(fhdr->frag_off) & ~0x7;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700272 end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
273 ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 if ((unsigned int)end > IPV6_MAXPLEN) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900276 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
277 IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700278 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
279 ((u8 *)&fhdr->frag_off -
280 skb_network_header(skb)));
Herbert Xuf61944e2007-10-15 01:28:47 -0700281 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 }
283
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700284 if (skb->ip_summed == CHECKSUM_COMPLETE) {
285 const unsigned char *nh = skb_network_header(skb);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900286 skb->csum = csum_sub(skb->csum,
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700287 csum_partial(nh, (u8 *)(fhdr + 1) - nh,
288 0));
289 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
291 /* Is this the final fragment? */
292 if (!(fhdr->frag_off & htons(IP6_MF))) {
293 /* If we already have some bits beyond end
294 * or have different end, the segment is corrupted.
295 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700296 if (end < fq->q.len ||
297 ((fq->q.last_in & LAST_IN) && end != fq->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700299 fq->q.last_in |= LAST_IN;
300 fq->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 } else {
302 /* Check if the fragment is rounded to 8 bytes.
303 * Required by the RFC.
304 */
305 if (end & 0x7) {
306 /* RFC2460 says always send parameter problem in
307 * this case. -DaveM
308 */
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900309 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
310 IPSTATS_MIB_INHDRERRORS);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900311 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 offsetof(struct ipv6hdr, payload_len));
Herbert Xuf61944e2007-10-15 01:28:47 -0700313 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700315 if (end > fq->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 /* Some bits beyond end -> corruption. */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700317 if (fq->q.last_in & LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700319 fq->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 }
321 }
322
323 if (end == offset)
324 goto err;
325
326 /* Point into the IP datagram 'data' part. */
327 if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data))
328 goto err;
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900329
Stephen Hemminger42ca89c2005-09-08 12:57:43 -0700330 if (pskb_trim_rcsum(skb, end - offset))
331 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
333 /* Find out which fragments are in front and at the back of us
334 * in the chain of fragments so far. We must know where to put
335 * this fragment, right?
336 */
337 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700338 for(next = fq->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 if (FRAG6_CB(next)->offset >= offset)
340 break; /* bingo! */
341 prev = next;
342 }
343
344 /* We found where to put this one. Check for overlap with
345 * preceding fragment, and, if needed, align things so that
346 * any overlaps are eliminated.
347 */
348 if (prev) {
349 int i = (FRAG6_CB(prev)->offset + prev->len) - offset;
350
351 if (i > 0) {
352 offset += i;
353 if (end <= offset)
354 goto err;
355 if (!pskb_pull(skb, i))
356 goto err;
357 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
358 skb->ip_summed = CHECKSUM_NONE;
359 }
360 }
361
362 /* Look for overlap with succeeding segments.
363 * If we can merge fragments, do it.
364 */
365 while (next && FRAG6_CB(next)->offset < end) {
366 int i = end - FRAG6_CB(next)->offset; /* overlap is 'i' bytes */
367
368 if (i < next->len) {
369 /* Eat head of the next overlapped fragment
370 * and leave the loop. The next ones cannot overlap.
371 */
372 if (!pskb_pull(next, i))
373 goto err;
374 FRAG6_CB(next)->offset += i; /* next fragment */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700375 fq->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 if (next->ip_summed != CHECKSUM_UNNECESSARY)
377 next->ip_summed = CHECKSUM_NONE;
378 break;
379 } else {
380 struct sk_buff *free_it = next;
381
382 /* Old fragment is completely overridden with
383 * new one drop it.
384 */
385 next = next->next;
386
387 if (prev)
388 prev->next = next;
389 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700390 fq->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700392 fq->q.meat -= free_it->len;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800393 frag_kfree_skb(fq->q.net, free_it, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 }
395 }
396
397 FRAG6_CB(skb)->offset = offset;
398
399 /* Insert this fragment in the chain of fragments. */
400 skb->next = next;
401 if (prev)
402 prev->next = skb;
403 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700404 fq->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Herbert Xuf61944e2007-10-15 01:28:47 -0700406 dev = skb->dev;
407 if (dev) {
408 fq->iif = dev->ifindex;
409 skb->dev = NULL;
410 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700411 fq->q.stamp = skb->tstamp;
412 fq->q.meat += skb->len;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800413 atomic_add(skb->truesize, &fq->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
415 /* The first fragment.
416 * nhoffset is obtained from the first fragment, of course.
417 */
418 if (offset == 0) {
419 fq->nhoffset = nhoff;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700420 fq->q.last_in |= FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
Herbert Xuf61944e2007-10-15 01:28:47 -0700422
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700423 if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
Herbert Xuf61944e2007-10-15 01:28:47 -0700424 return ip6_frag_reasm(fq, prev, dev);
425
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700426 write_lock(&ip6_frags.lock);
427 list_move_tail(&fq->q.lru_list, &ip6_frags.lru_list);
428 write_unlock(&ip6_frags.lock);
Herbert Xuf61944e2007-10-15 01:28:47 -0700429 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431err:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900432 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 kfree_skb(skb);
Herbert Xuf61944e2007-10-15 01:28:47 -0700434 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435}
436
437/*
438 * Check if this packet is complete.
439 * Returns NULL on failure by any reason, and pointer
440 * to current nexthdr field in reassembled frame.
441 *
442 * It is called with locked fq, and caller must check that
443 * queue is eligible for reassembly i.e. it is not COMPLETE,
444 * the last and the first frames arrived and all the bits are here.
445 */
Herbert Xuf61944e2007-10-15 01:28:47 -0700446static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 struct net_device *dev)
448{
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700449 struct sk_buff *fp, *head = fq->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 int payload_len;
451 unsigned int nhoff;
452
453 fq_kill(fq);
454
Herbert Xuf61944e2007-10-15 01:28:47 -0700455 /* Make the one we just received the head. */
456 if (prev) {
457 head = prev->next;
458 fp = skb_clone(head, GFP_ATOMIC);
459
460 if (!fp)
461 goto out_oom;
462
463 fp->next = head->next;
464 prev->next = fp;
465
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700466 skb_morph(head, fq->q.fragments);
467 head->next = fq->q.fragments->next;
Herbert Xuf61944e2007-10-15 01:28:47 -0700468
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700469 kfree_skb(fq->q.fragments);
470 fq->q.fragments = head;
Herbert Xuf61944e2007-10-15 01:28:47 -0700471 }
472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 BUG_TRAP(head != NULL);
474 BUG_TRAP(FRAG6_CB(head)->offset == 0);
475
476 /* Unfragmented part is taken from the first segment. */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700477 payload_len = ((head->data - skb_network_header(head)) -
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700478 sizeof(struct ipv6hdr) + fq->q.len -
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700479 sizeof(struct frag_hdr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 if (payload_len > IPV6_MAXPLEN)
481 goto out_oversize;
482
483 /* Head of list must not be cloned. */
484 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
485 goto out_oom;
486
487 /* If the first fragment is fragmented itself, we split
488 * it to two chunks: the first with data and paged part
489 * and the second, holding only fragments. */
490 if (skb_shinfo(head)->frag_list) {
491 struct sk_buff *clone;
492 int i, plen = 0;
493
494 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
495 goto out_oom;
496 clone->next = head->next;
497 head->next = clone;
498 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
499 skb_shinfo(head)->frag_list = NULL;
500 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
501 plen += skb_shinfo(head)->frags[i].size;
502 clone->len = clone->data_len = head->data_len - plen;
503 head->data_len -= clone->len;
504 head->len -= clone->len;
505 clone->csum = 0;
506 clone->ip_summed = head->ip_summed;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800507 atomic_add(clone->truesize, &fq->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 }
509
510 /* We have to remove fragment header from datagram and to relocate
511 * header in order to calculate ICV correctly. */
512 nhoff = fq->nhoffset;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700513 skb_network_header(head)[nhoff] = skb_transport_header(head)[0];
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900514 memmove(head->head + sizeof(struct frag_hdr), head->head,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 (head->data - head->head) - sizeof(struct frag_hdr));
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700516 head->mac_header += sizeof(struct frag_hdr);
517 head->network_header += sizeof(struct frag_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
519 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300520 skb_reset_transport_header(head);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700521 skb_push(head, head->data - skb_network_header(head));
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800522 atomic_sub(head->truesize, &fq->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
524 for (fp=head->next; fp; fp = fp->next) {
525 head->data_len += fp->len;
526 head->len += fp->len;
527 if (head->ip_summed != fp->ip_summed)
528 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700529 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 head->csum = csum_add(head->csum, fp->csum);
531 head->truesize += fp->truesize;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800532 atomic_sub(fp->truesize, &fq->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
534
535 head->next = NULL;
536 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700537 head->tstamp = fq->q.stamp;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700538 ipv6_hdr(head)->payload_len = htons(payload_len);
Patrick McHardy951dbc82006-01-06 23:02:34 -0800539 IP6CB(head)->nhoff = nhoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 /* Yes, and fold redundant checksum back. 8) */
Patrick McHardy84fa7932006-08-29 16:44:56 -0700542 if (head->ip_summed == CHECKSUM_COMPLETE)
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700543 head->csum = csum_partial(skb_network_header(head),
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300544 skb_network_header_len(head),
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700545 head->csum);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900547 rcu_read_lock();
548 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
549 rcu_read_unlock();
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700550 fq->q.fragments = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 return 1;
552
553out_oversize:
554 if (net_ratelimit())
555 printk(KERN_DEBUG "ip6_frag_reasm: payload len = %d\n", payload_len);
556 goto out_fail;
557out_oom:
558 if (net_ratelimit())
559 printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
560out_fail:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900561 rcu_read_lock();
562 IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
563 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return -1;
565}
566
Herbert Xue5bbef22007-10-15 12:50:28 -0700567static int ipv6_frag_rcv(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 struct frag_hdr *fhdr;
570 struct frag_queue *fq;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700571 struct ipv6hdr *hdr = ipv6_hdr(skb);
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800572 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900574 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
576 /* Jumbo payload inhibits frag. header */
577 if (hdr->payload_len==0) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900578 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300579 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
580 skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 return -1;
582 }
Arnaldo Carvalho de Meloea2ae172007-04-25 17:55:53 -0700583 if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
584 sizeof(struct frag_hdr)))) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900585 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -0300586 icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
587 skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 return -1;
589 }
590
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700591 hdr = ipv6_hdr(skb);
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -0700592 fhdr = (struct frag_hdr *)skb_transport_header(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594 if (!(fhdr->frag_off & htons(0xFFF9))) {
595 /* It is not a fragmented frame */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700596 skb->transport_header += sizeof(struct frag_hdr);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900597 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700599 IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 return 1;
601 }
602
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800603 net = skb->dev->nd_net;
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800604 if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800605 ip6_evictor(net, ip6_dst_idev(skb->dst));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800607 if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900608 ip6_dst_idev(skb->dst))) != NULL) {
Herbert Xuf61944e2007-10-15 01:28:47 -0700609 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700611 spin_lock(&fq->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Herbert Xuf61944e2007-10-15 01:28:47 -0700613 ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700615 spin_unlock(&fq->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700616 fq_put(fq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 return ret;
618 }
619
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900620 IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 kfree_skb(skb);
622 return -1;
623}
624
625static struct inet6_protocol frag_protocol =
626{
627 .handler = ipv6_frag_rcv,
628 .flags = INET6_PROTO_NOPOLICY,
629};
630
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800631#ifdef CONFIG_SYSCTL
632static struct ctl_table ip6_frags_ctl_table[] = {
633 {
634 .ctl_name = NET_IPV6_IP6FRAG_HIGH_THRESH,
635 .procname = "ip6frag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800636 .data = &init_net.ipv6.frags.high_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800637 .maxlen = sizeof(int),
638 .mode = 0644,
639 .proc_handler = &proc_dointvec
640 },
641 {
642 .ctl_name = NET_IPV6_IP6FRAG_LOW_THRESH,
643 .procname = "ip6frag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800644 .data = &init_net.ipv6.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800645 .maxlen = sizeof(int),
646 .mode = 0644,
647 .proc_handler = &proc_dointvec
648 },
649 {
650 .ctl_name = NET_IPV6_IP6FRAG_TIME,
651 .procname = "ip6frag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800652 .data = &init_net.ipv6.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800653 .maxlen = sizeof(int),
654 .mode = 0644,
655 .proc_handler = &proc_dointvec_jiffies,
656 .strategy = &sysctl_jiffies,
657 },
658 {
659 .ctl_name = NET_IPV6_IP6FRAG_SECRET_INTERVAL,
660 .procname = "ip6frag_secret_interval",
661 .data = &init_net.ipv6.sysctl.frags.secret_interval,
662 .maxlen = sizeof(int),
663 .mode = 0644,
664 .proc_handler = &proc_dointvec_jiffies,
665 .strategy = &sysctl_jiffies
666 },
667 { }
668};
Daniel Lezcano7d460db2008-01-18 23:52:35 -0800669
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800670static int ip6_frags_sysctl_register(struct net *net)
671{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800672 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800673 struct ctl_table_header *hdr;
674
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800675 table = ip6_frags_ctl_table;
676 if (net != &init_net) {
677 table = kmemdup(table, sizeof(ip6_frags_ctl_table), GFP_KERNEL);
678 if (table == NULL)
679 goto err_alloc;
680
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800681 table[0].data = &net->ipv6.frags.high_thresh;
682 table[1].data = &net->ipv6.frags.low_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800683 table[2].data = &net->ipv6.frags.timeout;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800684 table[3].mode &= ~0222;
685 }
686
687 hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
688 if (hdr == NULL)
689 goto err_reg;
690
691 net->ipv6.sysctl.frags_hdr = hdr;
692 return 0;
693
694err_reg:
695 if (net != &init_net)
696 kfree(table);
697err_alloc:
698 return -ENOMEM;
699}
700
701static void ip6_frags_sysctl_unregister(struct net *net)
702{
703 struct ctl_table *table;
704
705 table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
706 unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
707 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800708}
709#else
710static inline int ip6_frags_sysctl_register(struct net *net)
711{
712 return 0;
713}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800714
715static inline void ip6_frags_sysctl_unregister(struct net *net)
716{
717}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800718#endif
719
720static int ipv6_frags_init_net(struct net *net)
721{
Daniel Lezcanoe71e0342008-01-10 02:56:03 -0800722 ip6_frags.ctl = &net->ipv6.sysctl.frags;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800723
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800724 net->ipv6.frags.high_thresh = 256 * 1024;
725 net->ipv6.frags.low_thresh = 192 * 1024;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800726 net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800727 net->ipv6.sysctl.frags.secret_interval = 10 * 60 * HZ;
728
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800729 inet_frags_init_net(&net->ipv6.frags);
730
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800731 return ip6_frags_sysctl_register(net);
Daniel Lezcanoe71e0342008-01-10 02:56:03 -0800732}
733
Daniel Lezcano853cbba2007-12-11 02:24:29 -0800734int __init ipv6_frag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Daniel Lezcano853cbba2007-12-11 02:24:29 -0800736 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
Daniel Lezcano853cbba2007-12-11 02:24:29 -0800738 ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
739 if (ret)
740 goto out;
Daniel Lezcanoe71e0342008-01-10 02:56:03 -0800741
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800742 ipv6_frags_init_net(&init_net);
743
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700744 ip6_frags.hashfn = ip6_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700745 ip6_frags.constructor = ip6_frag_init;
Pavel Emelyanovc9547702007-10-17 19:48:26 -0700746 ip6_frags.destructor = NULL;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700747 ip6_frags.skb_free = NULL;
748 ip6_frags.qsize = sizeof(struct frag_queue);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700749 ip6_frags.match = ip6_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700750 ip6_frags.frag_expire = ip6_frag_expire;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700751 inet_frags_init(&ip6_frags);
Daniel Lezcano853cbba2007-12-11 02:24:29 -0800752out:
753 return ret;
754}
755
756void ipv6_frag_exit(void)
757{
758 inet_frags_fini(&ip6_frags);
759 inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760}