blob: cd6ce6ac635843e4598e5332c32256bb0ff93926 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The IP fragmentation functionality.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * Version: $Id: ip_fragment.c,v 1.59 2002/01/12 07:54:56 davem Exp $
9 *
10 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox <Alan.Cox@linux.org>
12 *
13 * Fixes:
14 * Alan Cox : Split from ip.c , see ip_input.c for history.
15 * David S. Miller : Begin massive cleanup...
16 * Andi Kleen : Add sysctls.
17 * xxxx : Overlapfrag bug.
18 * Ultima : ip_expire() kernel panic.
19 * Bill Hawes : Frag accounting and evictor fixes.
20 * John McDonald : 0 length frag bug.
21 * Alexey Kuznetsov: SMP races, threading, cleanup.
22 * Patrick McHardy : LRU queue of frag heads for evictor.
23 */
24
Herbert Xu89cee8b2005-12-13 23:14:27 -080025#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/types.h>
28#include <linux/mm.h>
29#include <linux/jiffies.h>
30#include <linux/skbuff.h>
31#include <linux/list.h>
32#include <linux/ip.h>
33#include <linux/icmp.h>
34#include <linux/netdevice.h>
35#include <linux/jhash.h>
36#include <linux/random.h>
37#include <net/sock.h>
38#include <net/ip.h>
39#include <net/icmp.h>
40#include <net/checksum.h>
Herbert Xu89cee8b2005-12-13 23:14:27 -080041#include <net/inetpeer.h>
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070042#include <net/inet_frag.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/tcp.h>
44#include <linux/udp.h>
45#include <linux/inet.h>
46#include <linux/netfilter_ipv4.h>
47
48/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
49 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
50 * as well. Or notify me, at least. --ANK
51 */
52
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -080053static int sysctl_ipfrag_max_dist __read_mostly = 64;
Herbert Xu89cee8b2005-12-13 23:14:27 -080054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055struct ipfrag_skb_cb
56{
57 struct inet_skb_parm h;
58 int offset;
59};
60
61#define FRAG_CB(skb) ((struct ipfrag_skb_cb*)((skb)->cb))
62
63/* Describe an entry in the "incomplete datagrams" queue. */
64struct ipq {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070065 struct inet_frag_queue q;
66
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 u32 user;
Al Viro18277772006-09-26 22:19:02 -070068 __be32 saddr;
69 __be32 daddr;
70 __be16 id;
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 u8 protocol;
Herbert Xu89cee8b2005-12-13 23:14:27 -080072 int iif;
73 unsigned int rid;
74 struct inet_peer *peer;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075};
76
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070077static struct inet_frags ip4_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -080079int ip_frag_nqueues(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070080{
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -080081 return net->ipv4.frags.nqueues;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070082}
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080084int ip_frag_mem(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070085{
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080086 return atomic_read(&net->ipv4.frags.mem);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070087}
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Herbert Xu1706d582007-10-14 00:38:15 -070089static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
90 struct net_device *dev);
91
Pavel Emelyanovc6fda282007-10-17 19:46:47 -070092struct ip4_create_arg {
93 struct iphdr *iph;
94 u32 user;
95};
96
Al Viro18277772006-09-26 22:19:02 -070097static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
Al Viro18277772006-09-26 22:19:02 -070099 return jhash_3words((__force u32)id << 16 | prot,
100 (__force u32)saddr, (__force u32)daddr,
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700101 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102}
103
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700104static unsigned int ip4_hashfn(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105{
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700106 struct ipq *ipq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700108 ipq = container_of(q, struct ipq, q);
109 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110}
111
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700112static int ip4_frag_match(struct inet_frag_queue *q, void *a)
113{
114 struct ipq *qp;
115 struct ip4_create_arg *arg = a;
116
117 qp = container_of(q, struct ipq, q);
118 return (qp->id == arg->iph->id &&
119 qp->saddr == arg->iph->saddr &&
120 qp->daddr == arg->iph->daddr &&
121 qp->protocol == arg->iph->protocol &&
122 qp->user == arg->user);
123}
124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125/* Memory Tracking Functions. */
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800126static __inline__ void frag_kfree_skb(struct netns_frags *nf,
127 struct sk_buff *skb, int *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128{
129 if (work)
130 *work -= skb->truesize;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800131 atomic_sub(skb->truesize, &nf->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 kfree_skb(skb);
133}
134
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700135static void ip4_frag_init(struct inet_frag_queue *q, void *a)
136{
137 struct ipq *qp = container_of(q, struct ipq, q);
138 struct ip4_create_arg *arg = a;
139
140 qp->protocol = arg->iph->protocol;
141 qp->id = arg->iph->id;
142 qp->saddr = arg->iph->saddr;
143 qp->daddr = arg->iph->daddr;
144 qp->user = arg->user;
145 qp->peer = sysctl_ipfrag_max_dist ?
146 inet_getpeer(arg->iph->saddr, 1) : NULL;
147}
148
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700149static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700151 struct ipq *qp;
152
153 qp = container_of(q, struct ipq, q);
154 if (qp->peer)
155 inet_putpeer(qp->peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159/* Destruction primitives. */
160
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700161static __inline__ void ipq_put(struct ipq *ipq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700163 inet_frag_put(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
165
166/* Kill ipq entry. It is not destroyed immediately,
167 * because caller (and someone more) holds reference count.
168 */
169static void ipq_kill(struct ipq *ipq)
170{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700171 inet_frag_kill(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172}
173
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900174/* Memory limiting on fragments. Evictor trashes the oldest
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 * fragment queue until we are back under the threshold.
176 */
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800177static void ip_evictor(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700179 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800181 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700182 if (evicted)
183 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184}
185
186/*
187 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
188 */
189static void ip_expire(unsigned long arg)
190{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700191 struct ipq *qp;
192
193 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700195 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Joe Perchesbc578a52008-03-28 16:35:27 -0700197 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 goto out;
199
200 ipq_kill(qp);
201
202 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
203 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
204
Joe Perchesbc578a52008-03-28 16:35:27 -0700205 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700206 struct sk_buff *head = qp->q.fragments;
Denis V. Lunevcb846632008-03-24 15:31:00 -0700207 struct net *net;
208
209 net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 /* Send an ICMP "Fragment Reassembly Timeout" message. */
Denis V. Lunevcb846632008-03-24 15:31:00 -0700211 if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
213 dev_put(head->dev);
214 }
215 }
216out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700217 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700218 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
220
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700221/* Find the correct entry in the "incomplete datagrams" queue for
222 * this IP datagram, and create new one, if nothing is found.
223 */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800224static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700226 struct inet_frag_queue *q;
227 struct ip4_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700228 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700230 arg.iph = iph;
231 arg.user = user;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700232 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700233
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800234 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700235 if (q == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 goto out_nomem;
237
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700238 return container_of(q, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240out_nomem:
Patrick McHardy64ce2072005-08-09 20:50:53 -0700241 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 return NULL;
243}
244
Herbert Xu89cee8b2005-12-13 23:14:27 -0800245/* Is the fragment too far ahead to be part of ipq? */
246static inline int ip_frag_too_far(struct ipq *qp)
247{
248 struct inet_peer *peer = qp->peer;
249 unsigned int max = sysctl_ipfrag_max_dist;
250 unsigned int start, end;
251
252 int rc;
253
254 if (!peer || !max)
255 return 0;
256
257 start = qp->rid;
258 end = atomic_inc_return(&peer->rid);
259 qp->rid = end;
260
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700261 rc = qp->q.fragments && (end - start) > max;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800262
263 if (rc) {
264 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
265 }
266
267 return rc;
268}
269
270static int ip_frag_reinit(struct ipq *qp)
271{
272 struct sk_buff *fp;
273
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800274 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700275 atomic_inc(&qp->q.refcnt);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800276 return -ETIMEDOUT;
277 }
278
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700279 fp = qp->q.fragments;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800280 do {
281 struct sk_buff *xp = fp->next;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800282 frag_kfree_skb(qp->q.net, fp, NULL);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800283 fp = xp;
284 } while (fp);
285
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700286 qp->q.last_in = 0;
287 qp->q.len = 0;
288 qp->q.meat = 0;
289 qp->q.fragments = NULL;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800290 qp->iif = 0;
291
292 return 0;
293}
294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295/* Add new segment to existing queue. */
Herbert Xu1706d582007-10-14 00:38:15 -0700296static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
298 struct sk_buff *prev, *next;
Herbert Xu1706d582007-10-14 00:38:15 -0700299 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 int flags, offset;
301 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700302 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Joe Perchesbc578a52008-03-28 16:35:27 -0700304 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 goto err;
306
Herbert Xu89cee8b2005-12-13 23:14:27 -0800307 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700308 unlikely(ip_frag_too_far(qp)) &&
309 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800310 ipq_kill(qp);
311 goto err;
312 }
313
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700314 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 flags = offset & ~IP_OFFSET;
316 offset &= IP_OFFSET;
317 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300318 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320 /* Determine the position of this fragment. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900321 end = offset + skb->len - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700322 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
324 /* Is this the final fragment? */
325 if ((flags & IP_MF) == 0) {
326 /* If we already have some bits beyond end
327 * or have different end, the segment is corrrupted.
328 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700329 if (end < qp->q.len ||
Joe Perchesbc578a52008-03-28 16:35:27 -0700330 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 goto err;
Joe Perchesbc578a52008-03-28 16:35:27 -0700332 qp->q.last_in |= INET_FRAG_LAST_IN;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700333 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 } else {
335 if (end&7) {
336 end &= ~7;
337 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
338 skb->ip_summed = CHECKSUM_NONE;
339 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700340 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 /* Some bits beyond end -> corruption. */
Joe Perchesbc578a52008-03-28 16:35:27 -0700342 if (qp->q.last_in & INET_FRAG_LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700344 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 }
346 }
347 if (end == offset)
348 goto err;
349
Herbert Xu1706d582007-10-14 00:38:15 -0700350 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (pskb_pull(skb, ihl) == NULL)
352 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700353
354 err = pskb_trim_rcsum(skb, end - offset);
355 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 goto err;
357
358 /* Find out which fragments are in front and at the back of us
359 * in the chain of fragments so far. We must know where to put
360 * this fragment, right?
361 */
362 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700363 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 if (FRAG_CB(next)->offset >= offset)
365 break; /* bingo! */
366 prev = next;
367 }
368
369 /* We found where to put this one. Check for overlap with
370 * preceding fragment, and, if needed, align things so that
371 * any overlaps are eliminated.
372 */
373 if (prev) {
374 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
375
376 if (i > 0) {
377 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700378 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 if (end <= offset)
380 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700381 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (!pskb_pull(skb, i))
383 goto err;
384 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
385 skb->ip_summed = CHECKSUM_NONE;
386 }
387 }
388
Herbert Xu1706d582007-10-14 00:38:15 -0700389 err = -ENOMEM;
390
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 while (next && FRAG_CB(next)->offset < end) {
392 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
393
394 if (i < next->len) {
395 /* Eat head of the next overlapped fragment
396 * and leave the loop. The next ones cannot overlap.
397 */
398 if (!pskb_pull(next, i))
399 goto err;
400 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700401 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 if (next->ip_summed != CHECKSUM_UNNECESSARY)
403 next->ip_summed = CHECKSUM_NONE;
404 break;
405 } else {
406 struct sk_buff *free_it = next;
407
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100408 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 * new one drop it.
410 */
411 next = next->next;
412
413 if (prev)
414 prev->next = next;
415 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700416 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700418 qp->q.meat -= free_it->len;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800419 frag_kfree_skb(qp->q.net, free_it, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
421 }
422
423 FRAG_CB(skb)->offset = offset;
424
425 /* Insert this fragment in the chain of fragments. */
426 skb->next = next;
427 if (prev)
428 prev->next = skb;
429 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700430 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
Herbert Xu1706d582007-10-14 00:38:15 -0700432 dev = skb->dev;
433 if (dev) {
434 qp->iif = dev->ifindex;
435 skb->dev = NULL;
436 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700437 qp->q.stamp = skb->tstamp;
438 qp->q.meat += skb->len;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800439 atomic_add(skb->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 if (offset == 0)
Joe Perchesbc578a52008-03-28 16:35:27 -0700441 qp->q.last_in |= INET_FRAG_FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Joe Perchesbc578a52008-03-28 16:35:27 -0700443 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
444 qp->q.meat == qp->q.len)
Herbert Xu1706d582007-10-14 00:38:15 -0700445 return ip_frag_reasm(qp, prev, dev);
446
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700447 write_lock(&ip4_frags.lock);
Pavel Emelyanov3140c252008-01-22 06:11:48 -0800448 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700449 write_unlock(&ip4_frags.lock);
Herbert Xu1706d582007-10-14 00:38:15 -0700450 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452err:
453 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700454 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456
457
458/* Build a new IP datagram from all its fragments. */
459
Herbert Xu1706d582007-10-14 00:38:15 -0700460static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
461 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
463 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700464 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 int len;
466 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700467 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 ipq_kill(qp);
470
Herbert Xu1706d582007-10-14 00:38:15 -0700471 /* Make the one we just received the head. */
472 if (prev) {
473 head = prev->next;
474 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700475 if (!fp)
476 goto out_nomem;
477
478 fp->next = head->next;
479 prev->next = fp;
480
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700481 skb_morph(head, qp->q.fragments);
482 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700483
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700484 kfree_skb(qp->q.fragments);
485 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700486 }
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 BUG_TRAP(head != NULL);
489 BUG_TRAP(FRAG_CB(head)->offset == 0);
490
491 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300492 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700493 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Herbert Xu1706d582007-10-14 00:38:15 -0700495 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800496 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 goto out_oversize;
498
499 /* Head of list must not be cloned. */
500 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
501 goto out_nomem;
502
503 /* If the first fragment is fragmented itself, we split
504 * it to two chunks: the first with data and paged part
505 * and the second, holding only fragments. */
506 if (skb_shinfo(head)->frag_list) {
507 struct sk_buff *clone;
508 int i, plen = 0;
509
510 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
511 goto out_nomem;
512 clone->next = head->next;
513 head->next = clone;
514 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
515 skb_shinfo(head)->frag_list = NULL;
516 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
517 plen += skb_shinfo(head)->frags[i].size;
518 clone->len = clone->data_len = head->data_len - plen;
519 head->data_len -= clone->len;
520 head->len -= clone->len;
521 clone->csum = 0;
522 clone->ip_summed = head->ip_summed;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800523 atomic_add(clone->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 }
525
526 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700527 skb_push(head, head->data - skb_network_header(head));
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800528 atomic_sub(head->truesize, &qp->q.net->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 Emelyanov6ddc0822008-01-22 06:07:25 -0800538 atomic_sub(fp->truesize, &qp->q.net->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 = qp->q.stamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700545 iph = ip_hdr(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 iph->frag_off = 0;
547 iph->tot_len = htons(len);
548 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700549 qp->q.fragments = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700550 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552out_nomem:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900553 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
Patrick McHardy64ce2072005-08-09 20:50:53 -0700554 "queue %p\n", qp);
David Howells45542472007-10-17 21:37:22 -0700555 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 goto out_fail;
557out_oversize:
558 if (net_ratelimit())
559 printk(KERN_INFO
YOSHIFUJI Hideakia7d632b2008-04-14 04:09:00 -0700560 "Oversized IP packet from " NIPQUAD_FMT ".\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 NIPQUAD(qp->saddr));
562out_fail:
563 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700564 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
567/* Process an incoming IP datagram fragment. */
Herbert Xu776c7292007-10-14 00:38:32 -0700568int ip_defrag(struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 struct ipq *qp;
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800571 struct net *net;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
574
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +0900575 net = skb->dev ? dev_net(skb->dev) : dev_net(skb->dst->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 /* Start by cleaning up the memory. */
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800577 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800578 ip_evictor(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 /* Lookup (or create) queue header */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800581 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
Herbert Xu1706d582007-10-14 00:38:15 -0700582 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700584 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Herbert Xu1706d582007-10-14 00:38:15 -0700586 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700588 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700589 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700590 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 }
592
593 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
594 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700595 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800598#ifdef CONFIG_SYSCTL
599static int zero;
600
601static struct ctl_table ip4_frags_ctl_table[] = {
602 {
603 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
604 .procname = "ipfrag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800605 .data = &init_net.ipv4.frags.high_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800606 .maxlen = sizeof(int),
607 .mode = 0644,
608 .proc_handler = &proc_dointvec
609 },
610 {
611 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
612 .procname = "ipfrag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800613 .data = &init_net.ipv4.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800614 .maxlen = sizeof(int),
615 .mode = 0644,
616 .proc_handler = &proc_dointvec
617 },
618 {
619 .ctl_name = NET_IPV4_IPFRAG_TIME,
620 .procname = "ipfrag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800621 .data = &init_net.ipv4.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800622 .maxlen = sizeof(int),
623 .mode = 0644,
624 .proc_handler = &proc_dointvec_jiffies,
625 .strategy = &sysctl_jiffies
626 },
627 {
628 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
629 .procname = "ipfrag_secret_interval",
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800630 .data = &ip4_frags.secret_interval,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800631 .maxlen = sizeof(int),
632 .mode = 0644,
633 .proc_handler = &proc_dointvec_jiffies,
634 .strategy = &sysctl_jiffies
635 },
636 {
637 .procname = "ipfrag_max_dist",
638 .data = &sysctl_ipfrag_max_dist,
639 .maxlen = sizeof(int),
640 .mode = 0644,
641 .proc_handler = &proc_dointvec_minmax,
642 .extra1 = &zero
643 },
644 { }
645};
646
647static int ip4_frags_ctl_register(struct net *net)
648{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800649 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800650 struct ctl_table_header *hdr;
651
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800652 table = ip4_frags_ctl_table;
653 if (net != &init_net) {
654 table = kmemdup(table, sizeof(ip4_frags_ctl_table), GFP_KERNEL);
655 if (table == NULL)
656 goto err_alloc;
657
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800658 table[0].data = &net->ipv4.frags.high_thresh;
659 table[1].data = &net->ipv4.frags.low_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800660 table[2].data = &net->ipv4.frags.timeout;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800661 table[3].mode &= ~0222;
662 table[4].mode &= ~0222;
663 }
664
665 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
666 if (hdr == NULL)
667 goto err_reg;
668
669 net->ipv4.frags_hdr = hdr;
670 return 0;
671
672err_reg:
673 if (net != &init_net)
674 kfree(table);
675err_alloc:
676 return -ENOMEM;
677}
678
679static void ip4_frags_ctl_unregister(struct net *net)
680{
681 struct ctl_table *table;
682
683 table = net->ipv4.frags_hdr->ctl_table_arg;
684 unregister_net_sysctl_table(net->ipv4.frags_hdr);
685 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800686}
687#else
688static inline int ip4_frags_ctl_register(struct net *net)
689{
690 return 0;
691}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800692
693static inline void ip4_frags_ctl_unregister(struct net *net)
694{
695}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800696#endif
697
698static int ipv4_frags_init_net(struct net *net)
699{
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800700 /*
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800701 * Fragment cache limits. We will commit 256K at one time. Should we
702 * cross that limit we will prune down to 192K. This should cope with
703 * even the most extreme cases without allowing an attacker to
704 * measurably harm machine performance.
705 */
706 net->ipv4.frags.high_thresh = 256 * 1024;
707 net->ipv4.frags.low_thresh = 192 * 1024;
708 /*
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800709 * Important NOTE! Fragment queue must be destroyed before MSL expires.
710 * RFC791 is wrong proposing to prolongate timer each fragment arrival
711 * by TTL.
712 */
713 net->ipv4.frags.timeout = IP_FRAG_TIME;
714
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800715 inet_frags_init_net(&net->ipv4.frags);
716
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800717 return ip4_frags_ctl_register(net);
718}
719
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800720static void ipv4_frags_exit_net(struct net *net)
721{
722 ip4_frags_ctl_unregister(net);
723 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
724}
725
726static struct pernet_operations ip4_frags_ops = {
727 .init = ipv4_frags_init_net,
728 .exit = ipv4_frags_exit_net,
729};
730
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700731void __init ipfrag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800733 register_pernet_subsys(&ip4_frags_ops);
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700734 ip4_frags.hashfn = ip4_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700735 ip4_frags.constructor = ip4_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700736 ip4_frags.destructor = ip4_frag_free;
737 ip4_frags.skb_free = NULL;
738 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700739 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700740 ip4_frags.frag_expire = ip_expire;
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800741 ip4_frags.secret_interval = 10 * 60 * HZ;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700742 inet_frags_init(&ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
745EXPORT_SYMBOL(ip_defrag);