blob: e4f81f54befee04ad398f6e5a6e3877d388d65bf [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 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
Alan Cox113aa832008-10-13 19:01:08 -07009 * Alan Cox <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 *
11 * Fixes:
12 * Alan Cox : Split from ip.c , see ip_input.c for history.
13 * David S. Miller : Begin massive cleanup...
14 * Andi Kleen : Add sysctls.
15 * xxxx : Overlapfrag bug.
16 * Ultima : ip_expire() kernel panic.
17 * Bill Hawes : Frag accounting and evictor fixes.
18 * John McDonald : 0 length frag bug.
19 * Alexey Kuznetsov: SMP races, threading, cleanup.
20 * Patrick McHardy : LRU queue of frag heads for evictor.
21 */
22
Herbert Xu89cee8b2005-12-13 23:14:27 -080023#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/module.h>
25#include <linux/types.h>
26#include <linux/mm.h>
27#include <linux/jiffies.h>
28#include <linux/skbuff.h>
29#include <linux/list.h>
30#include <linux/ip.h>
31#include <linux/icmp.h>
32#include <linux/netdevice.h>
33#include <linux/jhash.h>
34#include <linux/random.h>
35#include <net/sock.h>
36#include <net/ip.h>
37#include <net/icmp.h>
38#include <net/checksum.h>
Herbert Xu89cee8b2005-12-13 23:14:27 -080039#include <net/inetpeer.h>
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070040#include <net/inet_frag.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/tcp.h>
42#include <linux/udp.h>
43#include <linux/inet.h>
44#include <linux/netfilter_ipv4.h>
45
46/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
47 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
48 * as well. Or notify me, at least. --ANK
49 */
50
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -080051static int sysctl_ipfrag_max_dist __read_mostly = 64;
Herbert Xu89cee8b2005-12-13 23:14:27 -080052
Linus Torvalds1da177e2005-04-16 15:20:36 -070053struct ipfrag_skb_cb
54{
55 struct inet_skb_parm h;
56 int offset;
57};
58
59#define FRAG_CB(skb) ((struct ipfrag_skb_cb*)((skb)->cb))
60
61/* Describe an entry in the "incomplete datagrams" queue. */
62struct ipq {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070063 struct inet_frag_queue q;
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 u32 user;
Al Viro18277772006-09-26 22:19:02 -070066 __be32 saddr;
67 __be32 daddr;
68 __be16 id;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 u8 protocol;
Herbert Xu89cee8b2005-12-13 23:14:27 -080070 int iif;
71 unsigned int rid;
72 struct inet_peer *peer;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073};
74
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070075static struct inet_frags ip4_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -080077int ip_frag_nqueues(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070078{
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -080079 return net->ipv4.frags.nqueues;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070080}
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080082int ip_frag_mem(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070083{
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080084 return atomic_read(&net->ipv4.frags.mem);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070085}
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Herbert Xu1706d582007-10-14 00:38:15 -070087static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
88 struct net_device *dev);
89
Pavel Emelyanovc6fda282007-10-17 19:46:47 -070090struct ip4_create_arg {
91 struct iphdr *iph;
92 u32 user;
93};
94
Al Viro18277772006-09-26 22:19:02 -070095static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
Al Viro18277772006-09-26 22:19:02 -070097 return jhash_3words((__force u32)id << 16 | prot,
98 (__force u32)saddr, (__force u32)daddr,
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070099 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100}
101
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700102static unsigned int ip4_hashfn(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700104 struct ipq *ipq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700106 ipq = container_of(q, struct ipq, q);
107 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108}
109
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700110static int ip4_frag_match(struct inet_frag_queue *q, void *a)
111{
112 struct ipq *qp;
113 struct ip4_create_arg *arg = a;
114
115 qp = container_of(q, struct ipq, q);
116 return (qp->id == arg->iph->id &&
117 qp->saddr == arg->iph->saddr &&
118 qp->daddr == arg->iph->daddr &&
119 qp->protocol == arg->iph->protocol &&
120 qp->user == arg->user);
121}
122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123/* Memory Tracking Functions. */
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800124static __inline__ void frag_kfree_skb(struct netns_frags *nf,
125 struct sk_buff *skb, int *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
127 if (work)
128 *work -= skb->truesize;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800129 atomic_sub(skb->truesize, &nf->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 kfree_skb(skb);
131}
132
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700133static void ip4_frag_init(struct inet_frag_queue *q, void *a)
134{
135 struct ipq *qp = container_of(q, struct ipq, q);
136 struct ip4_create_arg *arg = a;
137
138 qp->protocol = arg->iph->protocol;
139 qp->id = arg->iph->id;
140 qp->saddr = arg->iph->saddr;
141 qp->daddr = arg->iph->daddr;
142 qp->user = arg->user;
143 qp->peer = sysctl_ipfrag_max_dist ?
144 inet_getpeer(arg->iph->saddr, 1) : NULL;
145}
146
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700147static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700149 struct ipq *qp;
150
151 qp = container_of(q, struct ipq, q);
152 if (qp->peer)
153 inet_putpeer(qp->peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154}
155
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157/* Destruction primitives. */
158
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700159static __inline__ void ipq_put(struct ipq *ipq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700161 inet_frag_put(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
164/* Kill ipq entry. It is not destroyed immediately,
165 * because caller (and someone more) holds reference count.
166 */
167static void ipq_kill(struct ipq *ipq)
168{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700169 inet_frag_kill(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170}
171
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900172/* Memory limiting on fragments. Evictor trashes the oldest
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 * fragment queue until we are back under the threshold.
174 */
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800175static void ip_evictor(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700177 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800179 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700180 if (evicted)
Pavel Emelyanovc5346fe2008-07-16 20:20:33 -0700181 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
184/*
185 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
186 */
187static void ip_expire(unsigned long arg)
188{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700189 struct ipq *qp;
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700190 struct net *net;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700191
192 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700193 net = container_of(qp->q.net, struct net, ipv4.frags);
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
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700202 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
203 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 /* Send an ICMP "Fragment Reassembly Timeout" message. */
Denis V. Lunevcb846632008-03-24 15:31:00 -0700209 if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
211 dev_put(head->dev);
212 }
213 }
214out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700215 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700216 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700219/* Find the correct entry in the "incomplete datagrams" queue for
220 * this IP datagram, and create new one, if nothing is found.
221 */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800222static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700224 struct inet_frag_queue *q;
225 struct ip4_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700226 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700228 arg.iph = iph;
229 arg.user = user;
Pavel Emelyanov9a375802008-06-27 20:06:08 -0700230
231 read_lock(&ip4_frags.lock);
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) {
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700264 struct net *net;
265
266 net = container_of(qp->q.net, struct net, ipv4.frags);
267 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800268 }
269
270 return rc;
271}
272
273static int ip_frag_reinit(struct ipq *qp)
274{
275 struct sk_buff *fp;
276
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800277 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700278 atomic_inc(&qp->q.refcnt);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800279 return -ETIMEDOUT;
280 }
281
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700282 fp = qp->q.fragments;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800283 do {
284 struct sk_buff *xp = fp->next;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800285 frag_kfree_skb(qp->q.net, fp, NULL);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800286 fp = xp;
287 } while (fp);
288
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700289 qp->q.last_in = 0;
290 qp->q.len = 0;
291 qp->q.meat = 0;
292 qp->q.fragments = NULL;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800293 qp->iif = 0;
294
295 return 0;
296}
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298/* Add new segment to existing queue. */
Herbert Xu1706d582007-10-14 00:38:15 -0700299static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
301 struct sk_buff *prev, *next;
Herbert Xu1706d582007-10-14 00:38:15 -0700302 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 int flags, offset;
304 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700305 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Joe Perchesbc578a52008-03-28 16:35:27 -0700307 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 goto err;
309
Herbert Xu89cee8b2005-12-13 23:14:27 -0800310 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700311 unlikely(ip_frag_too_far(qp)) &&
312 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800313 ipq_kill(qp);
314 goto err;
315 }
316
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700317 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 flags = offset & ~IP_OFFSET;
319 offset &= IP_OFFSET;
320 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300321 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323 /* Determine the position of this fragment. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900324 end = offset + skb->len - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700325 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 /* Is this the final fragment? */
328 if ((flags & IP_MF) == 0) {
329 /* If we already have some bits beyond end
330 * or have different end, the segment is corrrupted.
331 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700332 if (end < qp->q.len ||
Joe Perchesbc578a52008-03-28 16:35:27 -0700333 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 goto err;
Joe Perchesbc578a52008-03-28 16:35:27 -0700335 qp->q.last_in |= INET_FRAG_LAST_IN;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700336 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 } else {
338 if (end&7) {
339 end &= ~7;
340 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
341 skb->ip_summed = CHECKSUM_NONE;
342 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700343 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 /* Some bits beyond end -> corruption. */
Joe Perchesbc578a52008-03-28 16:35:27 -0700345 if (qp->q.last_in & INET_FRAG_LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700347 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 }
349 }
350 if (end == offset)
351 goto err;
352
Herbert Xu1706d582007-10-14 00:38:15 -0700353 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 if (pskb_pull(skb, ihl) == NULL)
355 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700356
357 err = pskb_trim_rcsum(skb, end - offset);
358 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 goto err;
360
361 /* Find out which fragments are in front and at the back of us
362 * in the chain of fragments so far. We must know where to put
363 * this fragment, right?
364 */
365 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700366 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 if (FRAG_CB(next)->offset >= offset)
368 break; /* bingo! */
369 prev = next;
370 }
371
372 /* We found where to put this one. Check for overlap with
373 * preceding fragment, and, if needed, align things so that
374 * any overlaps are eliminated.
375 */
376 if (prev) {
377 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
378
379 if (i > 0) {
380 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700381 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (end <= offset)
383 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700384 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 if (!pskb_pull(skb, i))
386 goto err;
387 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
388 skb->ip_summed = CHECKSUM_NONE;
389 }
390 }
391
Herbert Xu1706d582007-10-14 00:38:15 -0700392 err = -ENOMEM;
393
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 while (next && FRAG_CB(next)->offset < end) {
395 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
396
397 if (i < next->len) {
398 /* Eat head of the next overlapped fragment
399 * and leave the loop. The next ones cannot overlap.
400 */
401 if (!pskb_pull(next, i))
402 goto err;
403 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700404 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 if (next->ip_summed != CHECKSUM_UNNECESSARY)
406 next->ip_summed = CHECKSUM_NONE;
407 break;
408 } else {
409 struct sk_buff *free_it = next;
410
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100411 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 * new one drop it.
413 */
414 next = next->next;
415
416 if (prev)
417 prev->next = next;
418 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700419 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700421 qp->q.meat -= free_it->len;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800422 frag_kfree_skb(qp->q.net, free_it, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
424 }
425
426 FRAG_CB(skb)->offset = offset;
427
428 /* Insert this fragment in the chain of fragments. */
429 skb->next = next;
430 if (prev)
431 prev->next = skb;
432 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700433 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Herbert Xu1706d582007-10-14 00:38:15 -0700435 dev = skb->dev;
436 if (dev) {
437 qp->iif = dev->ifindex;
438 skb->dev = NULL;
439 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700440 qp->q.stamp = skb->tstamp;
441 qp->q.meat += skb->len;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800442 atomic_add(skb->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (offset == 0)
Joe Perchesbc578a52008-03-28 16:35:27 -0700444 qp->q.last_in |= INET_FRAG_FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Joe Perchesbc578a52008-03-28 16:35:27 -0700446 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
447 qp->q.meat == qp->q.len)
Herbert Xu1706d582007-10-14 00:38:15 -0700448 return ip_frag_reasm(qp, prev, dev);
449
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700450 write_lock(&ip4_frags.lock);
Pavel Emelyanov3140c252008-01-22 06:11:48 -0800451 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700452 write_unlock(&ip4_frags.lock);
Herbert Xu1706d582007-10-14 00:38:15 -0700453 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455err:
456 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700457 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
460
461/* Build a new IP datagram from all its fragments. */
462
Herbert Xu1706d582007-10-14 00:38:15 -0700463static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
464 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
466 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700467 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 int len;
469 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700470 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 ipq_kill(qp);
473
Herbert Xu1706d582007-10-14 00:38:15 -0700474 /* Make the one we just received the head. */
475 if (prev) {
476 head = prev->next;
477 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700478 if (!fp)
479 goto out_nomem;
480
481 fp->next = head->next;
482 prev->next = fp;
483
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700484 skb_morph(head, qp->q.fragments);
485 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700486
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700487 kfree_skb(qp->q.fragments);
488 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700489 }
490
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700491 WARN_ON(head == NULL);
492 WARN_ON(FRAG_CB(head)->offset != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300495 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700496 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Herbert Xu1706d582007-10-14 00:38:15 -0700498 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800499 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 goto out_oversize;
501
502 /* Head of list must not be cloned. */
503 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
504 goto out_nomem;
505
506 /* If the first fragment is fragmented itself, we split
507 * it to two chunks: the first with data and paged part
508 * and the second, holding only fragments. */
509 if (skb_shinfo(head)->frag_list) {
510 struct sk_buff *clone;
511 int i, plen = 0;
512
513 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
514 goto out_nomem;
515 clone->next = head->next;
516 head->next = clone;
517 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
518 skb_shinfo(head)->frag_list = NULL;
519 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
520 plen += skb_shinfo(head)->frags[i].size;
521 clone->len = clone->data_len = head->data_len - plen;
522 head->data_len -= clone->len;
523 head->len -= clone->len;
524 clone->csum = 0;
525 clone->ip_summed = head->ip_summed;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800526 atomic_add(clone->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 }
528
529 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700530 skb_push(head, head->data - skb_network_header(head));
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800531 atomic_sub(head->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533 for (fp=head->next; fp; fp = fp->next) {
534 head->data_len += fp->len;
535 head->len += fp->len;
536 if (head->ip_summed != fp->ip_summed)
537 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700538 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 head->csum = csum_add(head->csum, fp->csum);
540 head->truesize += fp->truesize;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800541 atomic_sub(fp->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
543
544 head->next = NULL;
545 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700546 head->tstamp = qp->q.stamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700548 iph = ip_hdr(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 iph->frag_off = 0;
550 iph->tot_len = htons(len);
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700551 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700552 qp->q.fragments = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700553 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555out_nomem:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900556 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
Patrick McHardy64ce2072005-08-09 20:50:53 -0700557 "queue %p\n", qp);
David Howells45542472007-10-17 21:37:22 -0700558 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 goto out_fail;
560out_oversize:
561 if (net_ratelimit())
562 printk(KERN_INFO
YOSHIFUJI Hideakia7d632b2008-04-14 04:09:00 -0700563 "Oversized IP packet from " NIPQUAD_FMT ".\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 NIPQUAD(qp->saddr));
565out_fail:
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700566 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700567 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
570/* Process an incoming IP datagram fragment. */
Herbert Xu776c7292007-10-14 00:38:32 -0700571int ip_defrag(struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 struct ipq *qp;
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800574 struct net *net;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900575
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700576 net = skb->dev ? dev_net(skb->dev) : dev_net(skb->dst->dev);
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700577 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579 /* Start by cleaning up the memory. */
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800580 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800581 ip_evictor(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 /* Lookup (or create) queue header */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800584 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
Herbert Xu1706d582007-10-14 00:38:15 -0700585 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700587 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Herbert Xu1706d582007-10-14 00:38:15 -0700589 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700591 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700592 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700593 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700596 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700598 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800601#ifdef CONFIG_SYSCTL
602static int zero;
603
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700604static struct ctl_table ip4_frags_ns_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800605 {
606 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
607 .procname = "ipfrag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800608 .data = &init_net.ipv4.frags.high_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800609 .maxlen = sizeof(int),
610 .mode = 0644,
611 .proc_handler = &proc_dointvec
612 },
613 {
614 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
615 .procname = "ipfrag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800616 .data = &init_net.ipv4.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800617 .maxlen = sizeof(int),
618 .mode = 0644,
619 .proc_handler = &proc_dointvec
620 },
621 {
622 .ctl_name = NET_IPV4_IPFRAG_TIME,
623 .procname = "ipfrag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800624 .data = &init_net.ipv4.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800625 .maxlen = sizeof(int),
626 .mode = 0644,
627 .proc_handler = &proc_dointvec_jiffies,
628 .strategy = &sysctl_jiffies
629 },
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700630 { }
631};
632
633static struct ctl_table ip4_frags_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800634 {
635 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
636 .procname = "ipfrag_secret_interval",
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800637 .data = &ip4_frags.secret_interval,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800638 .maxlen = sizeof(int),
639 .mode = 0644,
640 .proc_handler = &proc_dointvec_jiffies,
641 .strategy = &sysctl_jiffies
642 },
643 {
644 .procname = "ipfrag_max_dist",
645 .data = &sysctl_ipfrag_max_dist,
646 .maxlen = sizeof(int),
647 .mode = 0644,
648 .proc_handler = &proc_dointvec_minmax,
649 .extra1 = &zero
650 },
651 { }
652};
653
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700654static int ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800655{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800656 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800657 struct ctl_table_header *hdr;
658
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700659 table = ip4_frags_ns_ctl_table;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800660 if (net != &init_net) {
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700661 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800662 if (table == NULL)
663 goto err_alloc;
664
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800665 table[0].data = &net->ipv4.frags.high_thresh;
666 table[1].data = &net->ipv4.frags.low_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800667 table[2].data = &net->ipv4.frags.timeout;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800668 }
669
670 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
671 if (hdr == NULL)
672 goto err_reg;
673
674 net->ipv4.frags_hdr = hdr;
675 return 0;
676
677err_reg:
678 if (net != &init_net)
679 kfree(table);
680err_alloc:
681 return -ENOMEM;
682}
683
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700684static void ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800685{
686 struct ctl_table *table;
687
688 table = net->ipv4.frags_hdr->ctl_table_arg;
689 unregister_net_sysctl_table(net->ipv4.frags_hdr);
690 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800691}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700692
693static void ip4_frags_ctl_register(void)
694{
695 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
696}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800697#else
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700698static inline int ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800699{
700 return 0;
701}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800702
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700703static inline void ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800704{
705}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700706
707static inline void ip4_frags_ctl_register(void)
708{
709}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800710#endif
711
712static int ipv4_frags_init_net(struct net *net)
713{
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800714 /*
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800715 * Fragment cache limits. We will commit 256K at one time. Should we
716 * cross that limit we will prune down to 192K. This should cope with
717 * even the most extreme cases without allowing an attacker to
718 * measurably harm machine performance.
719 */
720 net->ipv4.frags.high_thresh = 256 * 1024;
721 net->ipv4.frags.low_thresh = 192 * 1024;
722 /*
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800723 * Important NOTE! Fragment queue must be destroyed before MSL expires.
724 * RFC791 is wrong proposing to prolongate timer each fragment arrival
725 * by TTL.
726 */
727 net->ipv4.frags.timeout = IP_FRAG_TIME;
728
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800729 inet_frags_init_net(&net->ipv4.frags);
730
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700731 return ip4_frags_ns_ctl_register(net);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800732}
733
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800734static void ipv4_frags_exit_net(struct net *net)
735{
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700736 ip4_frags_ns_ctl_unregister(net);
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800737 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
738}
739
740static struct pernet_operations ip4_frags_ops = {
741 .init = ipv4_frags_init_net,
742 .exit = ipv4_frags_exit_net,
743};
744
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700745void __init ipfrag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746{
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700747 ip4_frags_ctl_register();
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800748 register_pernet_subsys(&ip4_frags_ops);
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700749 ip4_frags.hashfn = ip4_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700750 ip4_frags.constructor = ip4_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700751 ip4_frags.destructor = ip4_frag_free;
752 ip4_frags.skb_free = NULL;
753 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700754 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700755 ip4_frags.frag_expire = ip_expire;
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800756 ip4_frags.secret_interval = 10 * 60 * HZ;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700757 inet_frags_init(&ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758}
759
760EXPORT_SYMBOL(ip_defrag);