blob: 29b4b0972e44d4556369ea52988266ae0cb31c07 [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 Emelyanove5a2bb8422008-01-22 06:06:23 -080079int ip_frag_nqueues(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070080{
Pavel Emelyanove5a2bb8422008-01-22 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
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700197 if (qp->q.last_in & 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
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700205 if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
206 struct sk_buff *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 /* Send an ICMP "Fragment Reassembly Timeout" message. */
Eric W. Biederman881d9662007-09-17 11:56:21 -0700208 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
210 dev_put(head->dev);
211 }
212 }
213out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700214 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700215 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700218/* Find the correct entry in the "incomplete datagrams" queue for
219 * this IP datagram, and create new one, if nothing is found.
220 */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800221static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700223 struct inet_frag_queue *q;
224 struct ip4_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700225 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700227 arg.iph = iph;
228 arg.user = user;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700229 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700230
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800231 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700232 if (q == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 goto out_nomem;
234
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700235 return container_of(q, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
237out_nomem:
Patrick McHardy64ce2072005-08-09 20:50:53 -0700238 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return NULL;
240}
241
Herbert Xu89cee8b2005-12-13 23:14:27 -0800242/* Is the fragment too far ahead to be part of ipq? */
243static inline int ip_frag_too_far(struct ipq *qp)
244{
245 struct inet_peer *peer = qp->peer;
246 unsigned int max = sysctl_ipfrag_max_dist;
247 unsigned int start, end;
248
249 int rc;
250
251 if (!peer || !max)
252 return 0;
253
254 start = qp->rid;
255 end = atomic_inc_return(&peer->rid);
256 qp->rid = end;
257
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700258 rc = qp->q.fragments && (end - start) > max;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800259
260 if (rc) {
261 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
262 }
263
264 return rc;
265}
266
267static int ip_frag_reinit(struct ipq *qp)
268{
269 struct sk_buff *fp;
270
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800271 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700272 atomic_inc(&qp->q.refcnt);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800273 return -ETIMEDOUT;
274 }
275
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700276 fp = qp->q.fragments;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800277 do {
278 struct sk_buff *xp = fp->next;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800279 frag_kfree_skb(qp->q.net, fp, NULL);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800280 fp = xp;
281 } while (fp);
282
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700283 qp->q.last_in = 0;
284 qp->q.len = 0;
285 qp->q.meat = 0;
286 qp->q.fragments = NULL;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800287 qp->iif = 0;
288
289 return 0;
290}
291
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292/* Add new segment to existing queue. */
Herbert Xu1706d582007-10-14 00:38:15 -0700293static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
295 struct sk_buff *prev, *next;
Herbert Xu1706d582007-10-14 00:38:15 -0700296 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 int flags, offset;
298 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700299 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700301 if (qp->q.last_in & COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 goto err;
303
Herbert Xu89cee8b2005-12-13 23:14:27 -0800304 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700305 unlikely(ip_frag_too_far(qp)) &&
306 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800307 ipq_kill(qp);
308 goto err;
309 }
310
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700311 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 flags = offset & ~IP_OFFSET;
313 offset &= IP_OFFSET;
314 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300315 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 /* Determine the position of this fragment. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900318 end = offset + skb->len - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700319 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
321 /* Is this the final fragment? */
322 if ((flags & IP_MF) == 0) {
323 /* If we already have some bits beyond end
324 * or have different end, the segment is corrrupted.
325 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700326 if (end < qp->q.len ||
327 ((qp->q.last_in & LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700329 qp->q.last_in |= LAST_IN;
330 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 } else {
332 if (end&7) {
333 end &= ~7;
334 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
335 skb->ip_summed = CHECKSUM_NONE;
336 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700337 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 /* Some bits beyond end -> corruption. */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700339 if (qp->q.last_in & LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700341 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 }
343 }
344 if (end == offset)
345 goto err;
346
Herbert Xu1706d582007-10-14 00:38:15 -0700347 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 if (pskb_pull(skb, ihl) == NULL)
349 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700350
351 err = pskb_trim_rcsum(skb, end - offset);
352 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 goto err;
354
355 /* Find out which fragments are in front and at the back of us
356 * in the chain of fragments so far. We must know where to put
357 * this fragment, right?
358 */
359 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700360 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 if (FRAG_CB(next)->offset >= offset)
362 break; /* bingo! */
363 prev = next;
364 }
365
366 /* We found where to put this one. Check for overlap with
367 * preceding fragment, and, if needed, align things so that
368 * any overlaps are eliminated.
369 */
370 if (prev) {
371 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
372
373 if (i > 0) {
374 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700375 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 if (end <= offset)
377 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700378 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 if (!pskb_pull(skb, i))
380 goto err;
381 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
382 skb->ip_summed = CHECKSUM_NONE;
383 }
384 }
385
Herbert Xu1706d582007-10-14 00:38:15 -0700386 err = -ENOMEM;
387
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 while (next && FRAG_CB(next)->offset < end) {
389 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
390
391 if (i < next->len) {
392 /* Eat head of the next overlapped fragment
393 * and leave the loop. The next ones cannot overlap.
394 */
395 if (!pskb_pull(next, i))
396 goto err;
397 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700398 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 if (next->ip_summed != CHECKSUM_UNNECESSARY)
400 next->ip_summed = CHECKSUM_NONE;
401 break;
402 } else {
403 struct sk_buff *free_it = next;
404
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100405 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 * new one drop it.
407 */
408 next = next->next;
409
410 if (prev)
411 prev->next = next;
412 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700413 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700415 qp->q.meat -= free_it->len;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800416 frag_kfree_skb(qp->q.net, free_it, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 }
418 }
419
420 FRAG_CB(skb)->offset = offset;
421
422 /* Insert this fragment in the chain of fragments. */
423 skb->next = next;
424 if (prev)
425 prev->next = skb;
426 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700427 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Herbert Xu1706d582007-10-14 00:38:15 -0700429 dev = skb->dev;
430 if (dev) {
431 qp->iif = dev->ifindex;
432 skb->dev = NULL;
433 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700434 qp->q.stamp = skb->tstamp;
435 qp->q.meat += skb->len;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800436 atomic_add(skb->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 if (offset == 0)
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700438 qp->q.last_in |= FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700440 if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
Herbert Xu1706d582007-10-14 00:38:15 -0700441 return ip_frag_reasm(qp, prev, dev);
442
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700443 write_lock(&ip4_frags.lock);
Pavel Emelyanov3140c252008-01-22 06:11:48 -0800444 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700445 write_unlock(&ip4_frags.lock);
Herbert Xu1706d582007-10-14 00:38:15 -0700446 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
448err:
449 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700450 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451}
452
453
454/* Build a new IP datagram from all its fragments. */
455
Herbert Xu1706d582007-10-14 00:38:15 -0700456static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
457 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
459 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700460 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 int len;
462 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700463 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 ipq_kill(qp);
466
Herbert Xu1706d582007-10-14 00:38:15 -0700467 /* Make the one we just received the head. */
468 if (prev) {
469 head = prev->next;
470 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700471 if (!fp)
472 goto out_nomem;
473
474 fp->next = head->next;
475 prev->next = fp;
476
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700477 skb_morph(head, qp->q.fragments);
478 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700479
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700480 kfree_skb(qp->q.fragments);
481 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700482 }
483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 BUG_TRAP(head != NULL);
485 BUG_TRAP(FRAG_CB(head)->offset == 0);
486
487 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300488 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700489 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
Herbert Xu1706d582007-10-14 00:38:15 -0700491 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800492 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 goto out_oversize;
494
495 /* Head of list must not be cloned. */
496 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
497 goto out_nomem;
498
499 /* If the first fragment is fragmented itself, we split
500 * it to two chunks: the first with data and paged part
501 * and the second, holding only fragments. */
502 if (skb_shinfo(head)->frag_list) {
503 struct sk_buff *clone;
504 int i, plen = 0;
505
506 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
507 goto out_nomem;
508 clone->next = head->next;
509 head->next = clone;
510 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
511 skb_shinfo(head)->frag_list = NULL;
512 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
513 plen += skb_shinfo(head)->frags[i].size;
514 clone->len = clone->data_len = head->data_len - plen;
515 head->data_len -= clone->len;
516 head->len -= clone->len;
517 clone->csum = 0;
518 clone->ip_summed = head->ip_summed;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800519 atomic_add(clone->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 }
521
522 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700523 skb_push(head, head->data - skb_network_header(head));
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800524 atomic_sub(head->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
526 for (fp=head->next; fp; fp = fp->next) {
527 head->data_len += fp->len;
528 head->len += fp->len;
529 if (head->ip_summed != fp->ip_summed)
530 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700531 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 head->csum = csum_add(head->csum, fp->csum);
533 head->truesize += fp->truesize;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800534 atomic_sub(fp->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
536
537 head->next = NULL;
538 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700539 head->tstamp = qp->q.stamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700541 iph = ip_hdr(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 iph->frag_off = 0;
543 iph->tot_len = htons(len);
544 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700545 qp->q.fragments = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700546 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548out_nomem:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900549 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
Patrick McHardy64ce2072005-08-09 20:50:53 -0700550 "queue %p\n", qp);
David Howells45542472007-10-17 21:37:22 -0700551 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 goto out_fail;
553out_oversize:
554 if (net_ratelimit())
555 printk(KERN_INFO
556 "Oversized IP packet from %d.%d.%d.%d.\n",
557 NIPQUAD(qp->saddr));
558out_fail:
559 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700560 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
562
563/* Process an incoming IP datagram fragment. */
Herbert Xu776c7292007-10-14 00:38:32 -0700564int ip_defrag(struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 struct ipq *qp;
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800567 struct net *net;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
570
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800571 net = skb->dev->nd_net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 /* Start by cleaning up the memory. */
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800573 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800574 ip_evictor(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 /* Lookup (or create) queue header */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800577 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
Herbert Xu1706d582007-10-14 00:38:15 -0700578 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700580 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Herbert Xu1706d582007-10-14 00:38:15 -0700582 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700584 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700585 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700586 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
588
589 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
590 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700591 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592}
593
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800594#ifdef CONFIG_SYSCTL
595static int zero;
596
597static struct ctl_table ip4_frags_ctl_table[] = {
598 {
599 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
600 .procname = "ipfrag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800601 .data = &init_net.ipv4.frags.high_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800602 .maxlen = sizeof(int),
603 .mode = 0644,
604 .proc_handler = &proc_dointvec
605 },
606 {
607 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
608 .procname = "ipfrag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800609 .data = &init_net.ipv4.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800610 .maxlen = sizeof(int),
611 .mode = 0644,
612 .proc_handler = &proc_dointvec
613 },
614 {
615 .ctl_name = NET_IPV4_IPFRAG_TIME,
616 .procname = "ipfrag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800617 .data = &init_net.ipv4.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800618 .maxlen = sizeof(int),
619 .mode = 0644,
620 .proc_handler = &proc_dointvec_jiffies,
621 .strategy = &sysctl_jiffies
622 },
623 {
624 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
625 .procname = "ipfrag_secret_interval",
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800626 .data = &ip4_frags.secret_interval,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800627 .maxlen = sizeof(int),
628 .mode = 0644,
629 .proc_handler = &proc_dointvec_jiffies,
630 .strategy = &sysctl_jiffies
631 },
632 {
633 .procname = "ipfrag_max_dist",
634 .data = &sysctl_ipfrag_max_dist,
635 .maxlen = sizeof(int),
636 .mode = 0644,
637 .proc_handler = &proc_dointvec_minmax,
638 .extra1 = &zero
639 },
640 { }
641};
642
643static int ip4_frags_ctl_register(struct net *net)
644{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800645 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800646 struct ctl_table_header *hdr;
647
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800648 table = ip4_frags_ctl_table;
649 if (net != &init_net) {
650 table = kmemdup(table, sizeof(ip4_frags_ctl_table), GFP_KERNEL);
651 if (table == NULL)
652 goto err_alloc;
653
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800654 table[0].data = &net->ipv4.frags.high_thresh;
655 table[1].data = &net->ipv4.frags.low_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800656 table[2].data = &net->ipv4.frags.timeout;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800657 table[3].mode &= ~0222;
658 table[4].mode &= ~0222;
659 }
660
661 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
662 if (hdr == NULL)
663 goto err_reg;
664
665 net->ipv4.frags_hdr = hdr;
666 return 0;
667
668err_reg:
669 if (net != &init_net)
670 kfree(table);
671err_alloc:
672 return -ENOMEM;
673}
674
675static void ip4_frags_ctl_unregister(struct net *net)
676{
677 struct ctl_table *table;
678
679 table = net->ipv4.frags_hdr->ctl_table_arg;
680 unregister_net_sysctl_table(net->ipv4.frags_hdr);
681 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800682}
683#else
684static inline int ip4_frags_ctl_register(struct net *net)
685{
686 return 0;
687}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800688
689static inline void ip4_frags_ctl_unregister(struct net *net)
690{
691}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800692#endif
693
694static int ipv4_frags_init_net(struct net *net)
695{
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800696 /*
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800697 * Fragment cache limits. We will commit 256K at one time. Should we
698 * cross that limit we will prune down to 192K. This should cope with
699 * even the most extreme cases without allowing an attacker to
700 * measurably harm machine performance.
701 */
702 net->ipv4.frags.high_thresh = 256 * 1024;
703 net->ipv4.frags.low_thresh = 192 * 1024;
704 /*
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800705 * Important NOTE! Fragment queue must be destroyed before MSL expires.
706 * RFC791 is wrong proposing to prolongate timer each fragment arrival
707 * by TTL.
708 */
709 net->ipv4.frags.timeout = IP_FRAG_TIME;
710
Pavel Emelyanove5a2bb8422008-01-22 06:06:23 -0800711 inet_frags_init_net(&net->ipv4.frags);
712
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800713 return ip4_frags_ctl_register(net);
714}
715
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700716void __init ipfrag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800718 ipv4_frags_init_net(&init_net);
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700719 ip4_frags.hashfn = ip4_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700720 ip4_frags.constructor = ip4_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700721 ip4_frags.destructor = ip4_frag_free;
722 ip4_frags.skb_free = NULL;
723 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700724 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700725 ip4_frags.frag_expire = ip_expire;
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800726 ip4_frags.secret_interval = 10 * 60 * HZ;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700727 inet_frags_init(&ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728}
729
730EXPORT_SYMBOL(ip_defrag);