blob: cd8c83025b48f2b7ca99d54e19b2f2c61a054ad2 [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 Emelyanov8d8354d2008-01-22 05:58:31 -080077static struct inet_frags_ctl ip4_frags_ctl __read_mostly = {
Pavel Emelyanov04128f22007-10-15 02:33:45 -070078 /*
79 * Fragment cache limits. We will commit 256K at one time. Should we
80 * cross that limit we will prune down to 192K. This should cope with
81 * even the most extreme cases without allowing an attacker to
82 * measurably harm machine performance.
83 */
84 .high_thresh = 256 * 1024,
85 .low_thresh = 192 * 1024,
86
87 /*
88 * Important NOTE! Fragment queue must be destroyed before MSL expires.
89 * RFC791 is wrong proposing to prolongate timer each fragment arrival
90 * by TTL.
91 */
92 .timeout = IP_FRAG_TIME,
93 .secret_interval = 10 * 60 * HZ,
94};
95
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070096static struct inet_frags ip4_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Pavel Emelyanove5a2bb8422008-01-22 06:06:23 -080098int ip_frag_nqueues(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070099{
Pavel Emelyanove5a2bb8422008-01-22 06:06:23 -0800100 return net->ipv4.frags.nqueues;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700101}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700103int ip_frag_mem(void)
104{
105 return atomic_read(&ip4_frags.mem);
106}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Herbert Xu1706d582007-10-14 00:38:15 -0700108static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
109 struct net_device *dev);
110
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700111struct ip4_create_arg {
112 struct iphdr *iph;
113 u32 user;
114};
115
Al Viro18277772006-09-26 22:19:02 -0700116static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
Al Viro18277772006-09-26 22:19:02 -0700118 return jhash_3words((__force u32)id << 16 | prot,
119 (__force u32)saddr, (__force u32)daddr,
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700120 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121}
122
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700123static unsigned int ip4_hashfn(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700125 struct ipq *ipq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700127 ipq = container_of(q, struct ipq, q);
128 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129}
130
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700131static int ip4_frag_match(struct inet_frag_queue *q, void *a)
132{
133 struct ipq *qp;
134 struct ip4_create_arg *arg = a;
135
136 qp = container_of(q, struct ipq, q);
137 return (qp->id == arg->iph->id &&
138 qp->saddr == arg->iph->saddr &&
139 qp->daddr == arg->iph->daddr &&
140 qp->protocol == arg->iph->protocol &&
141 qp->user == arg->user);
142}
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144/* Memory Tracking Functions. */
145static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work)
146{
147 if (work)
148 *work -= skb->truesize;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700149 atomic_sub(skb->truesize, &ip4_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 kfree_skb(skb);
151}
152
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700153static void ip4_frag_init(struct inet_frag_queue *q, void *a)
154{
155 struct ipq *qp = container_of(q, struct ipq, q);
156 struct ip4_create_arg *arg = a;
157
158 qp->protocol = arg->iph->protocol;
159 qp->id = arg->iph->id;
160 qp->saddr = arg->iph->saddr;
161 qp->daddr = arg->iph->daddr;
162 qp->user = arg->user;
163 qp->peer = sysctl_ipfrag_max_dist ?
164 inet_getpeer(arg->iph->saddr, 1) : NULL;
165}
166
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700167static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700169 struct ipq *qp;
170
171 qp = container_of(q, struct ipq, q);
172 if (qp->peer)
173 inet_putpeer(qp->peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174}
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
177/* Destruction primitives. */
178
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700179static __inline__ void ipq_put(struct ipq *ipq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700181 inet_frag_put(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
184/* Kill ipq entry. It is not destroyed immediately,
185 * because caller (and someone more) holds reference count.
186 */
187static void ipq_kill(struct ipq *ipq)
188{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700189 inet_frag_kill(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190}
191
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900192/* Memory limiting on fragments. Evictor trashes the oldest
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 * fragment queue until we are back under the threshold.
194 */
195static void ip_evictor(void)
196{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700197 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700199 evicted = inet_frag_evictor(&ip4_frags);
200 if (evicted)
201 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202}
203
204/*
205 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
206 */
207static void ip_expire(unsigned long arg)
208{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700209 struct ipq *qp;
210
211 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700213 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700215 if (qp->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 goto out;
217
218 ipq_kill(qp);
219
220 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
221 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
222
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700223 if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
224 struct sk_buff *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 /* Send an ICMP "Fragment Reassembly Timeout" message. */
Eric W. Biederman881d9662007-09-17 11:56:21 -0700226 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
228 dev_put(head->dev);
229 }
230 }
231out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700232 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700233 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700236/* Find the correct entry in the "incomplete datagrams" queue for
237 * this IP datagram, and create new one, if nothing is found.
238 */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800239static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700241 struct inet_frag_queue *q;
242 struct ip4_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.iph = iph;
246 arg.user = user;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700247 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700248
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800249 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700250 if (q == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 goto out_nomem;
252
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700253 return container_of(q, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
255out_nomem:
Patrick McHardy64ce2072005-08-09 20:50:53 -0700256 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 return NULL;
258}
259
Herbert Xu89cee8b2005-12-13 23:14:27 -0800260/* Is the fragment too far ahead to be part of ipq? */
261static inline int ip_frag_too_far(struct ipq *qp)
262{
263 struct inet_peer *peer = qp->peer;
264 unsigned int max = sysctl_ipfrag_max_dist;
265 unsigned int start, end;
266
267 int rc;
268
269 if (!peer || !max)
270 return 0;
271
272 start = qp->rid;
273 end = atomic_inc_return(&peer->rid);
274 qp->rid = end;
275
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700276 rc = qp->q.fragments && (end - start) > max;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800277
278 if (rc) {
279 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
280 }
281
282 return rc;
283}
284
285static int ip_frag_reinit(struct ipq *qp)
286{
287 struct sk_buff *fp;
288
Pavel Emelyanov04128f22007-10-15 02:33:45 -0700289 if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700290 atomic_inc(&qp->q.refcnt);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800291 return -ETIMEDOUT;
292 }
293
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700294 fp = qp->q.fragments;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800295 do {
296 struct sk_buff *xp = fp->next;
297 frag_kfree_skb(fp, NULL);
298 fp = xp;
299 } while (fp);
300
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700301 qp->q.last_in = 0;
302 qp->q.len = 0;
303 qp->q.meat = 0;
304 qp->q.fragments = NULL;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800305 qp->iif = 0;
306
307 return 0;
308}
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310/* Add new segment to existing queue. */
Herbert Xu1706d582007-10-14 00:38:15 -0700311static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
313 struct sk_buff *prev, *next;
Herbert Xu1706d582007-10-14 00:38:15 -0700314 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 int flags, offset;
316 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700317 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700319 if (qp->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 goto err;
321
Herbert Xu89cee8b2005-12-13 23:14:27 -0800322 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700323 unlikely(ip_frag_too_far(qp)) &&
324 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800325 ipq_kill(qp);
326 goto err;
327 }
328
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700329 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 flags = offset & ~IP_OFFSET;
331 offset &= IP_OFFSET;
332 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300333 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 /* Determine the position of this fragment. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900336 end = offset + skb->len - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700337 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
339 /* Is this the final fragment? */
340 if ((flags & IP_MF) == 0) {
341 /* If we already have some bits beyond end
342 * or have different end, the segment is corrrupted.
343 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700344 if (end < qp->q.len ||
345 ((qp->q.last_in & LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700347 qp->q.last_in |= LAST_IN;
348 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 } else {
350 if (end&7) {
351 end &= ~7;
352 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
353 skb->ip_summed = CHECKSUM_NONE;
354 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700355 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 /* Some bits beyond end -> corruption. */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700357 if (qp->q.last_in & LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700359 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 }
361 }
362 if (end == offset)
363 goto err;
364
Herbert Xu1706d582007-10-14 00:38:15 -0700365 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 if (pskb_pull(skb, ihl) == NULL)
367 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700368
369 err = pskb_trim_rcsum(skb, end - offset);
370 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 goto err;
372
373 /* Find out which fragments are in front and at the back of us
374 * in the chain of fragments so far. We must know where to put
375 * this fragment, right?
376 */
377 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700378 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 if (FRAG_CB(next)->offset >= offset)
380 break; /* bingo! */
381 prev = next;
382 }
383
384 /* We found where to put this one. Check for overlap with
385 * preceding fragment, and, if needed, align things so that
386 * any overlaps are eliminated.
387 */
388 if (prev) {
389 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
390
391 if (i > 0) {
392 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700393 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 if (end <= offset)
395 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700396 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 if (!pskb_pull(skb, i))
398 goto err;
399 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
400 skb->ip_summed = CHECKSUM_NONE;
401 }
402 }
403
Herbert Xu1706d582007-10-14 00:38:15 -0700404 err = -ENOMEM;
405
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 while (next && FRAG_CB(next)->offset < end) {
407 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
408
409 if (i < next->len) {
410 /* Eat head of the next overlapped fragment
411 * and leave the loop. The next ones cannot overlap.
412 */
413 if (!pskb_pull(next, i))
414 goto err;
415 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700416 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 if (next->ip_summed != CHECKSUM_UNNECESSARY)
418 next->ip_summed = CHECKSUM_NONE;
419 break;
420 } else {
421 struct sk_buff *free_it = next;
422
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100423 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 * new one drop it.
425 */
426 next = next->next;
427
428 if (prev)
429 prev->next = next;
430 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700431 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700433 qp->q.meat -= free_it->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 frag_kfree_skb(free_it, NULL);
435 }
436 }
437
438 FRAG_CB(skb)->offset = offset;
439
440 /* Insert this fragment in the chain of fragments. */
441 skb->next = next;
442 if (prev)
443 prev->next = skb;
444 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700445 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Herbert Xu1706d582007-10-14 00:38:15 -0700447 dev = skb->dev;
448 if (dev) {
449 qp->iif = dev->ifindex;
450 skb->dev = NULL;
451 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700452 qp->q.stamp = skb->tstamp;
453 qp->q.meat += skb->len;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700454 atomic_add(skb->truesize, &ip4_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 if (offset == 0)
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700456 qp->q.last_in |= FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700458 if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
Herbert Xu1706d582007-10-14 00:38:15 -0700459 return ip_frag_reasm(qp, prev, dev);
460
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700461 write_lock(&ip4_frags.lock);
462 list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list);
463 write_unlock(&ip4_frags.lock);
Herbert Xu1706d582007-10-14 00:38:15 -0700464 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466err:
467 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700468 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470
471
472/* Build a new IP datagram from all its fragments. */
473
Herbert Xu1706d582007-10-14 00:38:15 -0700474static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
475 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700478 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 int len;
480 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700481 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483 ipq_kill(qp);
484
Herbert Xu1706d582007-10-14 00:38:15 -0700485 /* Make the one we just received the head. */
486 if (prev) {
487 head = prev->next;
488 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700489 if (!fp)
490 goto out_nomem;
491
492 fp->next = head->next;
493 prev->next = fp;
494
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700495 skb_morph(head, qp->q.fragments);
496 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700497
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700498 kfree_skb(qp->q.fragments);
499 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700500 }
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 BUG_TRAP(head != NULL);
503 BUG_TRAP(FRAG_CB(head)->offset == 0);
504
505 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300506 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700507 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
Herbert Xu1706d582007-10-14 00:38:15 -0700509 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800510 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 goto out_oversize;
512
513 /* Head of list must not be cloned. */
514 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
515 goto out_nomem;
516
517 /* If the first fragment is fragmented itself, we split
518 * it to two chunks: the first with data and paged part
519 * and the second, holding only fragments. */
520 if (skb_shinfo(head)->frag_list) {
521 struct sk_buff *clone;
522 int i, plen = 0;
523
524 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
525 goto out_nomem;
526 clone->next = head->next;
527 head->next = clone;
528 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
529 skb_shinfo(head)->frag_list = NULL;
530 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
531 plen += skb_shinfo(head)->frags[i].size;
532 clone->len = clone->data_len = head->data_len - plen;
533 head->data_len -= clone->len;
534 head->len -= clone->len;
535 clone->csum = 0;
536 clone->ip_summed = head->ip_summed;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700537 atomic_add(clone->truesize, &ip4_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 }
539
540 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700541 skb_push(head, head->data - skb_network_header(head));
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700542 atomic_sub(head->truesize, &ip4_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544 for (fp=head->next; fp; fp = fp->next) {
545 head->data_len += fp->len;
546 head->len += fp->len;
547 if (head->ip_summed != fp->ip_summed)
548 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700549 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 head->csum = csum_add(head->csum, fp->csum);
551 head->truesize += fp->truesize;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700552 atomic_sub(fp->truesize, &ip4_frags.mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
554
555 head->next = NULL;
556 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700557 head->tstamp = qp->q.stamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700559 iph = ip_hdr(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 iph->frag_off = 0;
561 iph->tot_len = htons(len);
562 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700563 qp->q.fragments = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700564 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
566out_nomem:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900567 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
Patrick McHardy64ce2072005-08-09 20:50:53 -0700568 "queue %p\n", qp);
David Howells45542472007-10-17 21:37:22 -0700569 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 goto out_fail;
571out_oversize:
572 if (net_ratelimit())
573 printk(KERN_INFO
574 "Oversized IP packet from %d.%d.%d.%d.\n",
575 NIPQUAD(qp->saddr));
576out_fail:
577 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700578 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579}
580
581/* Process an incoming IP datagram fragment. */
Herbert Xu776c7292007-10-14 00:38:32 -0700582int ip_defrag(struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 struct ipq *qp;
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800585 struct net *net;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
588
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800589 net = skb->dev->nd_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 /* Start by cleaning up the memory. */
Pavel Emelyanov04128f22007-10-15 02:33:45 -0700591 if (atomic_read(&ip4_frags.mem) > ip4_frags_ctl.high_thresh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 ip_evictor();
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 /* Lookup (or create) queue header */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800595 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
Herbert Xu1706d582007-10-14 00:38:15 -0700596 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700598 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
Herbert Xu1706d582007-10-14 00:38:15 -0700600 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700602 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700603 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700604 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 }
606
607 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
608 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700609 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}
611
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800612#ifdef CONFIG_SYSCTL
613static int zero;
614
615static struct ctl_table ip4_frags_ctl_table[] = {
616 {
617 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
618 .procname = "ipfrag_high_thresh",
619 .data = &ip4_frags_ctl.high_thresh,
620 .maxlen = sizeof(int),
621 .mode = 0644,
622 .proc_handler = &proc_dointvec
623 },
624 {
625 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
626 .procname = "ipfrag_low_thresh",
627 .data = &ip4_frags_ctl.low_thresh,
628 .maxlen = sizeof(int),
629 .mode = 0644,
630 .proc_handler = &proc_dointvec
631 },
632 {
633 .ctl_name = NET_IPV4_IPFRAG_TIME,
634 .procname = "ipfrag_time",
635 .data = &ip4_frags_ctl.timeout,
636 .maxlen = sizeof(int),
637 .mode = 0644,
638 .proc_handler = &proc_dointvec_jiffies,
639 .strategy = &sysctl_jiffies
640 },
641 {
642 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
643 .procname = "ipfrag_secret_interval",
644 .data = &ip4_frags_ctl.secret_interval,
645 .maxlen = sizeof(int),
646 .mode = 0644,
647 .proc_handler = &proc_dointvec_jiffies,
648 .strategy = &sysctl_jiffies
649 },
650 {
651 .procname = "ipfrag_max_dist",
652 .data = &sysctl_ipfrag_max_dist,
653 .maxlen = sizeof(int),
654 .mode = 0644,
655 .proc_handler = &proc_dointvec_minmax,
656 .extra1 = &zero
657 },
658 { }
659};
660
661static int ip4_frags_ctl_register(struct net *net)
662{
663 struct ctl_table_header *hdr;
664
665 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path,
666 ip4_frags_ctl_table);
667 return hdr == NULL ? -ENOMEM : 0;
668}
669#else
670static inline int ip4_frags_ctl_register(struct net *net)
671{
672 return 0;
673}
674#endif
675
676static int ipv4_frags_init_net(struct net *net)
677{
Pavel Emelyanove5a2bb8422008-01-22 06:06:23 -0800678 inet_frags_init_net(&net->ipv4.frags);
679
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800680 return ip4_frags_ctl_register(net);
681}
682
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700683void __init ipfrag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800685 ipv4_frags_init_net(&init_net);
Pavel Emelyanov04128f22007-10-15 02:33:45 -0700686 ip4_frags.ctl = &ip4_frags_ctl;
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700687 ip4_frags.hashfn = ip4_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700688 ip4_frags.constructor = ip4_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700689 ip4_frags.destructor = ip4_frag_free;
690 ip4_frags.skb_free = NULL;
691 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700692 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700693 ip4_frags.frag_expire = ip_expire;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700694 inet_frags_init(&ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
697EXPORT_SYMBOL(ip_defrag);